blob: a4b9cdbaaa0b99d321a37e2fa1db8bf4063e2c86 [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
Jesper Juhl0bf8c862011-03-21 20:47:31 +01006 * Leonard Zubkoff and David Miller at Linux Expo,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * 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
Jesper Juhl0bf8c862011-03-21 20:47:31 +010013 * minor cleanups.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 * September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15 */
16
17#include <linux/module.h>
18#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090019#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/timer.h>
21#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/kernel.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070023#include <linux/freezer.h>
Christoph Hellwigc5478de2005-09-06 14:04:26 +020024#include <linux/kthread.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/interrupt.h>
26#include <linux/blkdev.h>
27#include <linux/delay.h>
28
29#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020030#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <scsi/scsi_dbg.h>
32#include <scsi/scsi_device.h>
33#include <scsi/scsi_eh.h>
James Smartc829c392006-03-13 08:28:57 -050034#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <scsi/scsi_host.h>
36#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38#include "scsi_priv.h"
39#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070040#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Kei Tokunagabf816232010-04-01 20:41:40 +090042#include <trace/events/scsi.h>
43
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#define SENSE_TIMEOUT (10*HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
46/*
47 * These should *probably* be handled by the host itself.
48 * Since it is allowed to sleep, it probably should.
49 */
50#define BUS_RESET_SETTLE_TIME (10)
51#define HOST_RESET_SETTLE_TIME (10)
52
David Jeffery3eef6252011-05-19 14:41:12 -040053static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055/* called with shost->host_lock held */
56void scsi_eh_wakeup(struct Scsi_Host *shost)
57{
58 if (shost->host_busy == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090059 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050060 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 SCSI_LOG_ERROR_RECOVERY(5,
62 printk("Waking error handler thread\n"));
63 }
64}
Tejun Heof8bbfc22006-05-19 21:07:05 +090065
66/**
67 * scsi_schedule_eh - schedule EH for SCSI host
68 * @shost: SCSI host to invoke error handling on.
69 *
70 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080071 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090072void scsi_schedule_eh(struct Scsi_Host *shost)
73{
74 unsigned long flags;
75
76 spin_lock_irqsave(shost->host_lock, flags);
77
78 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
79 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
80 shost->host_eh_scheduled++;
81 scsi_eh_wakeup(shost);
82 }
83
84 spin_unlock_irqrestore(shost->host_lock, flags);
85}
86EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88/**
89 * scsi_eh_scmd_add - add scsi cmd to error handling.
90 * @scmd: scmd to run eh on.
91 * @eh_flag: optional SCSI_EH flag.
92 *
93 * Return value:
94 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080095 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070096int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
97{
98 struct Scsi_Host *shost = scmd->device->host;
99 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -0500100 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
James Bottomley3ed7a472005-09-19 09:50:04 -0500102 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 return 0;
104
105 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500106 if (scsi_host_set_state(shost, SHOST_RECOVERY))
107 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
108 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
James Bottomley939647e2005-09-18 15:05:20 -0500110 ret = 1;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200111 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 shost->host_failed++;
114 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500115 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500117 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118}
119
120/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700122 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 *
124 * Notes:
125 * We do not need to lock this. There is the potential for a race
126 * only in that the normal completion handling might run, but if the
127 * normal completion function determines that the timer has already
128 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800129 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700130enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700132 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700133 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100134 struct Scsi_Host *host = scmd->device->host;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500135
Kei Tokunagabf816232010-04-01 20:41:40 +0900136 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 scsi_log_completion(scmd, TIMEOUT_ERROR);
138
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100139 if (host->transportt->eh_timed_out)
140 rtn = host->transportt->eh_timed_out(scmd);
141 else if (host->hostt->eh_timed_out)
142 rtn = host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500143
Christoph Hellwigfa990782008-11-05 15:03:15 +0100144 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
145 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
James Bottomley939647e2005-09-18 15:05:20 -0500146 scmd->result |= DID_TIME_OUT << 16;
Christoph Hellwigfa990782008-11-05 15:03:15 +0100147 rtn = BLK_EH_HANDLED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 }
Jens Axboe242f9dc2008-09-14 05:55:09 -0700149
Christoph Hellwigfa990782008-11-05 15:03:15 +0100150 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
153/**
154 * scsi_block_when_processing_errors - Prevent cmds from being queued.
155 * @sdev: Device on which we are performing recovery.
156 *
157 * Description:
158 * We block until the host is out of error recovery, and then check to
159 * see whether the host or the device is offline.
160 *
161 * Return value:
162 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800163 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164int scsi_block_when_processing_errors(struct scsi_device *sdev)
165{
166 int online;
167
James Bottomley939647e2005-09-18 15:05:20 -0500168 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
170 online = scsi_device_online(sdev);
171
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700172 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 online));
174
175 return online;
176}
177EXPORT_SYMBOL(scsi_block_when_processing_errors);
178
179#ifdef CONFIG_SCSI_LOGGING
180/**
181 * scsi_eh_prt_fail_stats - Log info on failures.
182 * @shost: scsi host being recovered.
183 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800184 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
186 struct list_head *work_q)
187{
188 struct scsi_cmnd *scmd;
189 struct scsi_device *sdev;
190 int total_failures = 0;
191 int cmd_failed = 0;
192 int cmd_cancel = 0;
193 int devices_failed = 0;
194
195 shost_for_each_device(sdev, shost) {
196 list_for_each_entry(scmd, work_q, eh_entry) {
197 if (scmd->device == sdev) {
198 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200199 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 ++cmd_cancel;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100201 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 ++cmd_failed;
203 }
204 }
205
206 if (cmd_cancel || cmd_failed) {
207 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500208 sdev_printk(KERN_INFO, sdev,
209 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700210 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500211 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 cmd_cancel = 0;
213 cmd_failed = 0;
214 ++devices_failed;
215 }
216 }
217
218 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
219 " devices require eh work\n",
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100220 total_failures, devices_failed));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221}
222#endif
223
224/**
225 * scsi_check_sense - Examine scsi cmd sense
226 * @scmd: Cmd to have sense checked.
227 *
228 * Return value:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100229 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 *
231 * Notes:
232 * When a deferred error is detected the current command has
233 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800234 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235static int scsi_check_sense(struct scsi_cmnd *scmd)
236{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700237 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 struct scsi_sense_hdr sshdr;
239
240 if (! scsi_command_normalize_sense(scmd, &sshdr))
241 return FAILED; /* no valid sense data */
242
243 if (scsi_sense_is_deferred(&sshdr))
244 return NEEDS_RETRY;
245
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700246 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
247 sdev->scsi_dh_data->scsi_dh->check_sense) {
248 int rc;
249
250 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
251 if (rc != SCSI_RETURN_NOT_HANDLED)
252 return rc;
253 /* handler does not care. Drop down to default handling */
254 }
255
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 /*
257 * Previous logic looked for FILEMARK, EOM or ILI which are
258 * mainly associated with tapes and returned SUCCESS.
259 */
260 if (sshdr.response_code == 0x70) {
261 /* fixed format */
262 if (scmd->sense_buffer[2] & 0xe0)
263 return SUCCESS;
264 } else {
265 /*
266 * descriptor format: look for "stream commands sense data
267 * descriptor" (see SSC-3). Assume single sense data
268 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
269 */
270 if ((sshdr.additional_length > 3) &&
271 (scmd->sense_buffer[8] == 0x4) &&
272 (scmd->sense_buffer[11] & 0xe0))
273 return SUCCESS;
274 }
275
276 switch (sshdr.sense_key) {
277 case NO_SENSE:
278 return SUCCESS;
279 case RECOVERED_ERROR:
280 return /* soft_error */ SUCCESS;
281
282 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400283 if (sshdr.asc == 0x10) /* DIF */
284 return SUCCESS;
285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 return NEEDS_RETRY;
287 case NOT_READY:
288 case UNIT_ATTENTION:
289 /*
290 * if we are expecting a cc/ua because of a bus reset that we
291 * performed, treat this just as a retry. otherwise this is
292 * information that we should pass up to the upper-level driver
293 * so that we can deal with it there.
294 */
295 if (scmd->device->expecting_cc_ua) {
296 scmd->device->expecting_cc_ua = 0;
297 return NEEDS_RETRY;
298 }
299 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100300 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 * should retry.
302 */
303 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
304 return NEEDS_RETRY;
305 /*
306 * if the device is not started, we need to wake
307 * the error handler to start the motor
308 */
309 if (scmd->device->allow_restart &&
310 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
311 return FAILED;
James Bottomley77a42292010-05-04 16:51:40 -0400312
Mike Christie6e499492010-04-25 07:03:57 -0500313 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
314 scmd_printk(KERN_WARNING, scmd,
315 "Warning! Received an indication that the "
316 "LUN assignments on this target have "
317 "changed. The Linux SCSI layer does not "
318 "automatically remap LUN assignments.\n");
319 else if (sshdr.asc == 0x3f)
320 scmd_printk(KERN_WARNING, scmd,
321 "Warning! Received an indication that the "
322 "operating parameters on this target have "
323 "changed. The Linux SCSI layer does not "
324 "automatically adjust these parameters.\n");
325
Shyam Iyerdeb1cb62011-02-26 01:59:44 -0500326 if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
327 scmd_printk(KERN_WARNING, scmd,
328 "Warning! Received an indication that the "
329 "LUN reached a thin provisioning soft "
330 "threshold.\n");
331
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100332 /*
333 * Pass the UA upwards for a determination in the completion
334 * functions.
335 */
336 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100338 /* these are not supported */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 case COPY_ABORTED:
340 case VOLUME_OVERFLOW:
341 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100342 case BLANK_CHECK:
343 case DATA_PROTECT:
344 return TARGET_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800347 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
348 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
349 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100350 return TARGET_ERROR;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800351 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 return NEEDS_RETRY;
353
354 case HARDWARE_ERROR:
355 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700356 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 else
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100358 return TARGET_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
360 case ILLEGAL_REQUEST:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 default:
362 return SUCCESS;
363 }
364}
365
Vasu Dev4a840672009-10-22 15:46:33 -0700366static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
367{
368 struct scsi_host_template *sht = sdev->host->hostt;
369 struct scsi_device *tmp_sdev;
370
371 if (!sht->change_queue_depth ||
372 sdev->queue_depth >= sdev->max_queue_depth)
373 return;
374
375 if (time_before(jiffies,
376 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
377 return;
378
379 if (time_before(jiffies,
380 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
381 return;
382
383 /*
384 * Walk all devices of a target and do
385 * ramp up on them.
386 */
387 shost_for_each_device(tmp_sdev, sdev->host) {
388 if (tmp_sdev->channel != sdev->channel ||
389 tmp_sdev->id != sdev->id ||
390 tmp_sdev->queue_depth == sdev->max_queue_depth)
391 continue;
392 /*
393 * call back into LLD to increase queue_depth by one
394 * with ramp up reason code.
395 */
396 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
397 SCSI_QDEPTH_RAMP_UP);
398 sdev->last_queue_ramp_up = jiffies;
399 }
400}
401
Mike Christie42a6a912009-10-15 17:46:44 -0700402static void scsi_handle_queue_full(struct scsi_device *sdev)
403{
404 struct scsi_host_template *sht = sdev->host->hostt;
405 struct scsi_device *tmp_sdev;
406
407 if (!sht->change_queue_depth)
408 return;
409
410 shost_for_each_device(tmp_sdev, sdev->host) {
411 if (tmp_sdev->channel != sdev->channel ||
412 tmp_sdev->id != sdev->id)
413 continue;
414 /*
415 * We do not know the number of commands that were at
416 * the device when we got the queue full so we start
417 * from the highest possible value and work our way down.
418 */
419 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
420 SCSI_QDEPTH_QFULL);
421 }
422}
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424/**
425 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
426 * @scmd: SCSI cmd to examine.
427 *
428 * Notes:
429 * This is *only* called when we are examining the status of commands
430 * queued during error recovery. the main difference here is that we
431 * don't allow for the possibility of retries here, and we are a lot
432 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800433 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
435{
436 /*
437 * first check the host byte, to see if there is anything in there
438 * that would indicate what we need to do.
439 */
440 if (host_byte(scmd->result) == DID_RESET) {
441 /*
442 * rats. we are already in the error handler, so we now
443 * get to try and figure out what to do next. if the sense
444 * is valid, we have a pretty good idea of what to do.
445 * if not, we mark it as FAILED.
446 */
447 return scsi_check_sense(scmd);
448 }
449 if (host_byte(scmd->result) != DID_OK)
450 return FAILED;
451
452 /*
453 * next, check the message byte.
454 */
455 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
456 return FAILED;
457
458 /*
459 * now, check the status byte to see if this indicates
460 * anything special.
461 */
462 switch (status_byte(scmd->result)) {
463 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700464 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 case COMMAND_TERMINATED:
466 return SUCCESS;
467 case CHECK_CONDITION:
468 return scsi_check_sense(scmd);
469 case CONDITION_GOOD:
470 case INTERMEDIATE_GOOD:
471 case INTERMEDIATE_C_GOOD:
472 /*
473 * who knows? FIXME(eric)
474 */
475 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500476 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500477 if (scmd->cmnd[0] == TEST_UNIT_READY)
478 /* it is a success, we probed the device and
479 * found it */
480 return SUCCESS;
481 /* otherwise, we failed to send the command */
482 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700484 scsi_handle_queue_full(scmd->device);
485 /* fall through */
486 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200487 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 default:
489 return FAILED;
490 }
491 return FAILED;
492}
493
494/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 * scsi_eh_done - Completion function for error handling.
496 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800497 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498static void scsi_eh_done(struct scsi_cmnd *scmd)
499{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100500 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600501
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100502 SCSI_LOG_ERROR_RECOVERY(3,
503 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700504 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600505
506 eh_action = scmd->device->host->eh_action;
507 if (eh_action)
508 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509}
510
511/**
Brian King292148f2007-01-30 17:51:17 -0600512 * scsi_try_host_reset - ask host adapter to reset itself
513 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800514 */
Brian King292148f2007-01-30 17:51:17 -0600515static int scsi_try_host_reset(struct scsi_cmnd *scmd)
516{
517 unsigned long flags;
518 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100519 struct Scsi_Host *host = scmd->device->host;
520 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600521
522 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700523 __func__));
Brian King292148f2007-01-30 17:51:17 -0600524
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100525 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600526 return FAILED;
527
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100528 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600529
530 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100531 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600532 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100533 spin_lock_irqsave(host->host_lock, flags);
534 scsi_report_bus_reset(host, scmd_channel(scmd));
535 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600536 }
537
538 return rtn;
539}
540
541/**
542 * scsi_try_bus_reset - ask host to perform a bus reset
543 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800544 */
Brian King292148f2007-01-30 17:51:17 -0600545static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
546{
547 unsigned long flags;
548 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100549 struct Scsi_Host *host = scmd->device->host;
550 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600551
552 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700553 __func__));
Brian King292148f2007-01-30 17:51:17 -0600554
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100555 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600556 return FAILED;
557
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100558 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600559
560 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100561 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600562 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100563 spin_lock_irqsave(host->host_lock, flags);
564 scsi_report_bus_reset(host, scmd_channel(scmd));
565 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600566 }
567
568 return rtn;
569}
570
Mike Christie30bd7df2008-02-29 18:25:19 -0600571static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
572{
573 sdev->was_reset = 1;
574 sdev->expecting_cc_ua = 1;
575}
576
577/**
578 * scsi_try_target_reset - Ask host to perform a target reset
579 * @scmd: SCSI cmd used to send a target reset
580 *
581 * Notes:
582 * There is no timeout for this operation. if this operation is
583 * unreliable for a given host, then the host itself needs to put a
584 * timer on it, and set the host back to a consistent state prior to
585 * returning.
586 */
587static int scsi_try_target_reset(struct scsi_cmnd *scmd)
588{
589 unsigned long flags;
590 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100591 struct Scsi_Host *host = scmd->device->host;
592 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600593
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100594 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600595 return FAILED;
596
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100597 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600598 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100599 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600600 __starget_for_each_device(scsi_target(scmd->device), NULL,
601 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100602 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600603 }
604
605 return rtn;
606}
607
Brian King292148f2007-01-30 17:51:17 -0600608/**
609 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
610 * @scmd: SCSI cmd used to send BDR
611 *
612 * Notes:
613 * There is no timeout for this operation. if this operation is
614 * unreliable for a given host, then the host itself needs to put a
615 * timer on it, and set the host back to a consistent state prior to
616 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800617 */
Brian King292148f2007-01-30 17:51:17 -0600618static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
619{
620 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100621 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600622
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100623 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600624 return FAILED;
625
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100626 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600627 if (rtn == SUCCESS)
628 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600629 return rtn;
630}
631
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100632static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600633{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100634 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600635 return FAILED;
636
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100637 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600638}
639
Brian King292148f2007-01-30 17:51:17 -0600640static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
641{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100642 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600643 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600644 if (scsi_try_target_reset(scmd) != SUCCESS)
645 if (scsi_try_bus_reset(scmd) != SUCCESS)
646 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600647}
648
649/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200650 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100651 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200652 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200653 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300654 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200655 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100656 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200657 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200658 * as part of the error recovery process. If @sense_bytes is 0 the command
659 * sent must be one that does not transfer any data. If @sense_bytes != 0
660 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
661 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800662 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200663void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
664 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665{
f59114b2005-04-17 15:00:23 -0500666 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
Christoph Hellwig631c2282006-07-08 20:42:15 +0200668 /*
669 * We need saved copies of a number of fields - this is because
670 * error handling may need to overwrite these with different values
671 * to run different commands, and once error handling is complete,
672 * we will need to restore these values prior to running the actual
673 * command.
674 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200675 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300676 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200677 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200678 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200679 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200680 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400681 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400682 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200683
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400684 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300685 scmd->cmnd = ses->eh_cmnd;
686 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200687 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200688 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200689
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200690 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200691 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
692 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200693 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200694 scmd->sdb.length);
695 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200696 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200697 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200698 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200699 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200700 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200701 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200702 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200703 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300704 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200705 memcpy(scmd->cmnd, cmnd, cmnd_size);
706 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
707 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200708 }
709
710 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200711
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200712 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500714 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715
Christoph Hellwig631c2282006-07-08 20:42:15 +0200716 /*
717 * Zero the sense buffer. The scsi spec mandates that any
718 * untransferred sense data should be interpreted as being zero.
719 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900720 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200721}
722EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200723
Boaz Harroshe1c23462007-10-08 16:36:45 +0200724/**
725 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
726 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200727 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200728 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200729 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800730 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200731void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
732{
733 /*
734 * Restore original data
735 */
736 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300737 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200738 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200739 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200740 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200741 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400742 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400743 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200744}
745EXPORT_SYMBOL(scsi_eh_restore_cmnd);
746
747/**
748 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
749 * @scmd: SCSI command structure to hijack
750 * @cmnd: CDB to send
751 * @cmnd_size: size in bytes of @cmnd
752 * @timeout: timeout for this request
753 * @sense_bytes: size of sense data to copy or 0
754 *
755 * This function is used to send a scsi command down to a target device
756 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
757 *
758 * Return value:
759 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800760 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200761static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
762 int cmnd_size, int timeout, unsigned sense_bytes)
763{
764 struct scsi_device *sdev = scmd->device;
765 struct Scsi_Host *shost = sdev->host;
766 DECLARE_COMPLETION_ONSTACK(done);
767 unsigned long timeleft;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200768 struct scsi_eh_save ses;
769 int rtn;
770
771 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100772 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -0500775 scmd->scsi_done = scsi_eh_done;
776 shost->hostt->queuecommand(shost, scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100778 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
f59114b2005-04-17 15:00:23 -0500780 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100782 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100784 SCSI_LOG_ERROR_RECOVERY(3,
785 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700786 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
788 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100789 * If there is time left scsi_eh_done got called, and we will
790 * examine the actual status codes to see whether the command
791 * actually did complete normally, else tell the host to forget
792 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100794 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 rtn = scsi_eh_completed_normally(scmd);
796 SCSI_LOG_ERROR_RECOVERY(3,
797 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700798 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100799
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 switch (rtn) {
801 case SUCCESS:
802 case NEEDS_RETRY:
803 case FAILED:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100804 case TARGET_ERROR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +0200806 case ADD_TO_MLQUEUE:
807 rtn = NEEDS_RETRY;
808 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 default:
810 rtn = FAILED;
811 break;
812 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100813 } else {
Brian King292148f2007-01-30 17:51:17 -0600814 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100815 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 }
817
Boaz Harroshe1c23462007-10-08 16:36:45 +0200818 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return rtn;
820}
821
822/**
823 * scsi_request_sense - Request sense data from a particular target.
824 * @scmd: SCSI cmd for request sense.
825 *
826 * Notes:
827 * Some hosts automatically obtain this information, others require
828 * that we obtain it on our own. This function will *not* return until
829 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800830 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831static int scsi_request_sense(struct scsi_cmnd *scmd)
832{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200833 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834}
835
836/**
837 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
838 * @scmd: Original SCSI cmd that eh has finished.
839 * @done_q: Queue for processed commands.
840 *
841 * Notes:
842 * We don't want to use the normal command completion while we are are
843 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500844 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 * keep a list of pending commands for final completion, and once we
846 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800847 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900848void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
850 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200851 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 list_move_tail(&scmd->eh_entry, done_q);
853}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900854EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
856/**
857 * scsi_eh_get_sense - Get device sense data.
858 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500859 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 *
861 * Description:
862 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100863 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 *
865 * Notes:
866 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500867 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 * it down before we request it.
869 *
870 * All drivers should request sense information internally these days,
871 * so for now all I have to say is tough noogies if you end up in here.
872 *
873 * XXX: Long term this code should go away, but that needs an audit of
874 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800875 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800876int scsi_eh_get_sense(struct list_head *work_q,
877 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200879 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 int rtn;
881
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200882 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200883 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 SCSI_SENSE_VALID(scmd))
885 continue;
886
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400887 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
888 "%s: requesting sense\n",
889 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 rtn = scsi_request_sense(scmd);
891 if (rtn != SUCCESS)
892 continue;
893
894 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
895 " result %x\n", scmd,
896 scmd->result));
897 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
898
899 rtn = scsi_decide_disposition(scmd);
900
901 /*
902 * if the result was normal, then just pass it along to the
903 * upper level.
904 */
905 if (rtn == SUCCESS)
906 /* we don't want this command reissued, just
907 * finished with the sense data, so set
908 * retries to the max allowed to ensure it
909 * won't get reissued */
910 scmd->retries = scmd->allowed;
911 else if (rtn != NEEDS_RETRY)
912 continue;
913
914 scsi_eh_finish_cmd(scmd, done_q);
915 }
916
917 return list_empty(work_q);
918}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800919EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920
921/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500923 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 *
925 * Return value:
926 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800927 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928static int scsi_eh_tur(struct scsi_cmnd *scmd)
929{
930 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
931 int retry_cnt = 1, rtn;
932
933retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400934 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700937 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200938
939 switch (rtn) {
940 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 if (retry_cnt--)
942 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200943 /*FALLTHRU*/
944 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500945 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200946 default:
947 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949}
950
951/**
David Jeffery3eef6252011-05-19 14:41:12 -0400952 * scsi_eh_test_devices - check if devices are responding from error recovery.
953 * @cmd_list: scsi commands in error recovery.
954 * @work_q: queue for commands which still need more error recovery
955 * @done_q: queue for commands which are finished
956 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
957 *
958 * Decription:
959 * Tests if devices are in a working state. Commands to devices now in
960 * a working state are sent to the done_q while commands to devices which
961 * are still failing to respond are returned to the work_q for more
962 * processing.
963 **/
964static int scsi_eh_test_devices(struct list_head *cmd_list,
965 struct list_head *work_q,
966 struct list_head *done_q, int try_stu)
967{
968 struct scsi_cmnd *scmd, *next;
969 struct scsi_device *sdev;
970 int finish_cmds;
971
972 while (!list_empty(cmd_list)) {
973 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
974 sdev = scmd->device;
975
976 finish_cmds = !scsi_device_online(scmd->device) ||
977 (try_stu && !scsi_eh_try_stu(scmd) &&
978 !scsi_eh_tur(scmd)) ||
979 !scsi_eh_tur(scmd);
980
981 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
982 if (scmd->device == sdev) {
983 if (finish_cmds)
984 scsi_eh_finish_cmd(scmd, done_q);
985 else
986 list_move_tail(&scmd->eh_entry, work_q);
987 }
988 }
989 return list_empty(work_q);
990}
991
992
993/**
Rob Landleyeb448202007-11-03 13:30:39 -0500994 * scsi_eh_abort_cmds - abort pending commands.
995 * @work_q: &list_head for pending commands.
996 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 *
998 * Decription:
999 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -05001000 * running command. This only works out to be the case if we have one
1001 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 * no sense to try and abort the command, since as far as the shost
1003 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005static int scsi_eh_abort_cmds(struct list_head *work_q,
1006 struct list_head *done_q)
1007{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001008 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001009 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 int rtn;
1011
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001012 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001013 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 continue;
1015 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1016 "0x%p\n", current->comm,
1017 scmd));
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001018 rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001019 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001020 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
David Jeffery3eef6252011-05-19 14:41:12 -04001021 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 scsi_eh_finish_cmd(scmd, done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001023 else
1024 list_move_tail(&scmd->eh_entry, &check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 } else
1026 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1027 " cmd failed:"
1028 "0x%p\n",
1029 current->comm,
1030 scmd));
1031 }
1032
David Jeffery3eef6252011-05-19 14:41:12 -04001033 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034}
1035
1036/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001038 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 *
1040 * Return value:
1041 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001042 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1044{
1045 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Christoph Hellwig631c2282006-07-08 20:42:15 +02001047 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001048 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Brian Kinged773e62007-03-29 15:25:52 -05001050 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001051 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001052
Christoph Hellwig631c2282006-07-08 20:42:15 +02001053 if (rtn == SUCCESS)
1054 return 0;
1055 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 return 1;
1058}
1059
1060 /**
1061 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001062 * @shost: &scsi host being recovered.
1063 * @work_q: &list_head for pending commands.
1064 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 *
1066 * Notes:
1067 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001068 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001069 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070static int scsi_eh_stu(struct Scsi_Host *shost,
1071 struct list_head *work_q,
1072 struct list_head *done_q)
1073{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001074 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 struct scsi_device *sdev;
1076
1077 shost_for_each_device(sdev, shost) {
1078 stu_scmd = NULL;
1079 list_for_each_entry(scmd, work_q, eh_entry)
1080 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1081 scsi_check_sense(scmd) == FAILED ) {
1082 stu_scmd = scmd;
1083 break;
1084 }
1085
1086 if (!stu_scmd)
1087 continue;
1088
1089 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1090 " 0x%p\n", current->comm, sdev));
1091
1092 if (!scsi_eh_try_stu(stu_scmd)) {
1093 if (!scsi_device_online(sdev) ||
1094 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001095 list_for_each_entry_safe(scmd, next,
1096 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (scmd->device == sdev)
1098 scsi_eh_finish_cmd(scmd, done_q);
1099 }
1100 }
1101 } else {
1102 SCSI_LOG_ERROR_RECOVERY(3,
1103 printk("%s: START_UNIT failed to sdev:"
1104 " 0x%p\n", current->comm, sdev));
1105 }
1106 }
1107
1108 return list_empty(work_q);
1109}
1110
1111
1112/**
1113 * scsi_eh_bus_device_reset - send bdr if needed
1114 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001115 * @work_q: &list_head for pending commands.
1116 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 *
1118 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001119 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 * devices that are jammed or not - if we have multiple devices, it
1121 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001122 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001123 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1125 struct list_head *work_q,
1126 struct list_head *done_q)
1127{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001128 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 struct scsi_device *sdev;
1130 int rtn;
1131
1132 shost_for_each_device(sdev, shost) {
1133 bdr_scmd = NULL;
1134 list_for_each_entry(scmd, work_q, eh_entry)
1135 if (scmd->device == sdev) {
1136 bdr_scmd = scmd;
1137 break;
1138 }
1139
1140 if (!bdr_scmd)
1141 continue;
1142
1143 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1144 " 0x%p\n", current->comm,
1145 sdev));
1146 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001147 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001149 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001151 list_for_each_entry_safe(scmd, next,
1152 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 if (scmd->device == sdev)
1154 scsi_eh_finish_cmd(scmd,
1155 done_q);
1156 }
1157 }
1158 } else {
1159 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1160 " failed sdev:"
1161 "0x%p\n",
1162 current->comm,
1163 sdev));
1164 }
1165 }
1166
1167 return list_empty(work_q);
1168}
1169
1170/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001171 * scsi_eh_target_reset - send target reset if needed
1172 * @shost: scsi host being recovered.
1173 * @work_q: &list_head for pending commands.
1174 * @done_q: &list_head for processed commands.
1175 *
1176 * Notes:
1177 * Try a target reset.
1178 */
1179static int scsi_eh_target_reset(struct Scsi_Host *shost,
1180 struct list_head *work_q,
1181 struct list_head *done_q)
1182{
James Bottomley98db5192010-10-25 15:53:41 -05001183 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001184 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001185
James Bottomley98db5192010-10-25 15:53:41 -05001186 list_splice_init(work_q, &tmp_list);
1187
1188 while (!list_empty(&tmp_list)) {
1189 struct scsi_cmnd *next, *scmd;
1190 int rtn;
1191 unsigned int id;
1192
1193 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1194 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001195
1196 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1197 "to target %d\n",
1198 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001199 rtn = scsi_try_target_reset(scmd);
1200 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Mike Christie30bd7df2008-02-29 18:25:19 -06001201 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1202 " failed target: "
1203 "%d\n",
1204 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001205 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1206 if (scmd_id(scmd) != id)
1207 continue;
1208
David Jeffery3eef6252011-05-19 14:41:12 -04001209 if (rtn == SUCCESS)
1210 list_move_tail(&scmd->eh_entry, &check_list);
1211 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001212 scsi_eh_finish_cmd(scmd, done_q);
1213 else
1214 /* push back on work queue for further processing */
1215 list_move(&scmd->eh_entry, work_q);
1216 }
1217 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001218
David Jeffery3eef6252011-05-19 14:41:12 -04001219 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001220}
1221
1222/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001223 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001224 * @shost: &scsi host being recovered.
1225 * @work_q: &list_head for pending 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_bus_reset(struct Scsi_Host *shost,
1229 struct list_head *work_q,
1230 struct list_head *done_q)
1231{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001232 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001233 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 unsigned int channel;
1235 int rtn;
1236
1237 /*
1238 * we really want to loop over the various channels, and do this on
1239 * a channel by channel basis. we should also check to see if any
1240 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001241 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 */
1243
1244 for (channel = 0; channel <= shost->max_channel; channel++) {
1245 chan_scmd = NULL;
1246 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001247 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 chan_scmd = scmd;
1249 break;
1250 /*
1251 * FIXME add back in some support for
1252 * soft_reset devices.
1253 */
1254 }
1255 }
1256
1257 if (!chan_scmd)
1258 continue;
1259 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1260 " %d\n", current->comm,
1261 channel));
1262 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001263 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001264 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001265 if (channel == scmd_channel(scmd)) {
1266 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 scsi_eh_finish_cmd(scmd,
1268 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001269 else
1270 list_move_tail(&scmd->eh_entry,
1271 &check_list);
1272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 }
1274 } else {
1275 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1276 " failed chan: %d\n",
1277 current->comm,
1278 channel));
1279 }
1280 }
David Jeffery3eef6252011-05-19 14:41:12 -04001281 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282}
1283
1284/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001285 * scsi_eh_host_reset - send a host reset
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 * @work_q: list_head for processed commands.
1287 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001288 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289static int scsi_eh_host_reset(struct list_head *work_q,
1290 struct list_head *done_q)
1291{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001292 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001293 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295
1296 if (!list_empty(work_q)) {
1297 scmd = list_entry(work_q->next,
1298 struct scsi_cmnd, eh_entry);
1299
1300 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1301 , current->comm));
1302
1303 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001304 if (rtn == SUCCESS) {
1305 list_splice_init(work_q, &check_list);
1306 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001307 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 scsi_eh_finish_cmd(scmd, done_q);
1309 }
1310 } else {
1311 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1312 " failed\n",
1313 current->comm));
1314 }
1315 }
David Jeffery3eef6252011-05-19 14:41:12 -04001316 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317}
1318
1319/**
1320 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1321 * @work_q: list_head for processed commands.
1322 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001323 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324static void scsi_eh_offline_sdevs(struct list_head *work_q,
1325 struct list_head *done_q)
1326{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001327 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001329 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001330 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1331 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001333 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 /*
1335 * FIXME: Handle lost cmds.
1336 */
1337 }
1338 scsi_eh_finish_cmd(scmd, done_q);
1339 }
1340 return;
1341}
1342
1343/**
Mike Christie4a274462008-08-19 18:45:31 -05001344 * scsi_noretry_cmd - determinte if command should be failed fast
1345 * @scmd: SCSI cmd to examine.
1346 */
1347int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1348{
1349 switch (host_byte(scmd->result)) {
1350 case DID_OK:
1351 break;
1352 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001353 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001354 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001355 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001356 case DID_ERROR:
1357 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1358 status_byte(scmd->result) == RESERVATION_CONFLICT)
1359 return 0;
1360 /* fall through */
1361 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001362 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001363 }
1364
1365 switch (status_byte(scmd->result)) {
1366 case CHECK_CONDITION:
1367 /*
1368 * assume caller has checked sense and determinted
1369 * the check condition was retryable.
1370 */
FUJITA Tomonorie96f6ab2010-07-09 09:38:26 +09001371 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1372 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1373 return 1;
Mike Christie4a274462008-08-19 18:45:31 -05001374 }
1375
1376 return 0;
1377}
1378
1379/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1381 * @scmd: SCSI cmd to examine.
1382 *
1383 * Notes:
1384 * This is *only* called when we are examining the status after sending
1385 * out the actual data command. any commands that are queued for error
1386 * recovery (e.g. test_unit_ready) do *not* come through here.
1387 *
1388 * When this routine returns failed, it means the error handler thread
1389 * is woken. In cases where the error code indicates an error that
1390 * doesn't require the error handler read (i.e. we don't need to
1391 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001392 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393int scsi_decide_disposition(struct scsi_cmnd *scmd)
1394{
1395 int rtn;
1396
1397 /*
1398 * if the device is offline, then we clearly just pass the result back
1399 * up to the top level.
1400 */
1401 if (!scsi_device_online(scmd->device)) {
1402 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1403 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001404 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 return SUCCESS;
1406 }
1407
1408 /*
1409 * first check the host byte, to see if there is anything in there
1410 * that would indicate what we need to do.
1411 */
1412 switch (host_byte(scmd->result)) {
1413 case DID_PASSTHROUGH:
1414 /*
1415 * no matter what, pass this through to the upper layer.
1416 * nuke this special code so that it looks like we are saying
1417 * did_ok.
1418 */
1419 scmd->result &= 0xff00ffff;
1420 return SUCCESS;
1421 case DID_OK:
1422 /*
1423 * looks good. drop through, and check the next byte.
1424 */
1425 break;
1426 case DID_NO_CONNECT:
1427 case DID_BAD_TARGET:
1428 case DID_ABORT:
1429 /*
1430 * note - this means that we just report the status back
1431 * to the top level driver, not that we actually think
1432 * that it indicates SUCCESS.
1433 */
1434 return SUCCESS;
1435 /*
1436 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001437 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 * in order to avoid endless loops (db)
1439 *
1440 * actually this is a bug in this function here. we should
1441 * be mindful of the maximum number of retries specified
1442 * and not get stuck in a loop.
1443 */
1444 case DID_SOFT_ERROR:
1445 goto maybe_retry;
1446 case DID_IMM_RETRY:
1447 return NEEDS_RETRY;
1448
bf341912005-04-12 17:49:09 -05001449 case DID_REQUEUE:
1450 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001451 case DID_TRANSPORT_DISRUPTED:
1452 /*
1453 * LLD/transport was disrupted during processing of the IO.
1454 * The transport class is now blocked/blocking,
1455 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001456 * based on its timers and recovery capablilities if
1457 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001458 */
Mike Christie939c2282008-11-04 19:47:19 -06001459 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001460 case DID_TRANSPORT_FAILFAST:
1461 /*
1462 * The transport decided to failfast the IO (most likely
1463 * the fast io fail tmo fired), so send IO directly upwards.
1464 */
1465 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 case DID_ERROR:
1467 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1468 status_byte(scmd->result) == RESERVATION_CONFLICT)
1469 /*
1470 * execute reservation conflict processing code
1471 * lower down
1472 */
1473 break;
1474 /* fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 case DID_BUS_BUSY:
1476 case DID_PARITY:
1477 goto maybe_retry;
1478 case DID_TIME_OUT:
1479 /*
1480 * when we scan the bus, we get timeout messages for
1481 * these commands if there is no device available.
1482 * other hosts report did_no_connect for the same thing.
1483 */
1484 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1485 scmd->cmnd[0] == INQUIRY)) {
1486 return SUCCESS;
1487 } else {
1488 return FAILED;
1489 }
1490 case DID_RESET:
1491 return SUCCESS;
1492 default:
1493 return FAILED;
1494 }
1495
1496 /*
1497 * next, check the message byte.
1498 */
1499 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1500 return FAILED;
1501
1502 /*
1503 * check the status byte to see if this indicates anything special.
1504 */
1505 switch (status_byte(scmd->result)) {
1506 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001507 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 /*
1509 * the case of trying to send too many commands to a
1510 * tagged queueing device.
1511 */
1512 case BUSY:
1513 /*
1514 * device can't talk to us at the moment. Should only
1515 * occur (SAM-3) when the task queue is empty, so will cause
1516 * the empty queue handling to trigger a stall in the
1517 * device.
1518 */
1519 return ADD_TO_MLQUEUE;
1520 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -07001521 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001524 case TASK_ABORTED:
1525 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 case CHECK_CONDITION:
1527 rtn = scsi_check_sense(scmd);
1528 if (rtn == NEEDS_RETRY)
1529 goto maybe_retry;
Hannes Reinecke63583cc2011-01-18 10:13:11 +01001530 else if (rtn == TARGET_ERROR) {
1531 /*
1532 * Need to modify host byte to signal a
1533 * permanent target failure
1534 */
1535 scmd->result |= (DID_TARGET_FAILURE << 16);
1536 rtn = SUCCESS;
1537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 /* if rtn == FAILED, we have no sense information;
1539 * returning FAILED will wake the error handler thread
1540 * to collect the sense and redo the decide
1541 * disposition */
1542 return rtn;
1543 case CONDITION_GOOD:
1544 case INTERMEDIATE_GOOD:
1545 case INTERMEDIATE_C_GOOD:
1546 case ACA_ACTIVE:
1547 /*
1548 * who knows? FIXME(eric)
1549 */
1550 return SUCCESS;
1551
1552 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001553 sdev_printk(KERN_INFO, scmd->device,
1554 "reservation conflict\n");
Hannes Reinecke63583cc2011-01-18 10:13:11 +01001555 scmd->result |= (DID_NEXUS_FAILURE << 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 return SUCCESS; /* causes immediate i/o error */
1557 default:
1558 return FAILED;
1559 }
1560 return FAILED;
1561
1562 maybe_retry:
1563
1564 /* we requeue for retry because the error was retryable, and
1565 * the request was not marked fast fail. Note that above,
1566 * even if the request is marked fast fail, we still requeue
1567 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001568 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001569 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return NEEDS_RETRY;
1571 } else {
1572 /*
1573 * no more retries - report this one back to upper level.
1574 */
1575 return SUCCESS;
1576 }
1577}
1578
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001579static void eh_lock_door_done(struct request *req, int uptodate)
1580{
1581 __blk_put_request(req->q, req);
1582}
1583
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 * scsi_eh_lock_door - Prevent medium removal for the specified device
1586 * @sdev: SCSI device to prevent medium removal
1587 *
1588 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001589 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 *
1591 * Notes:
1592 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1593 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001594 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595static void scsi_eh_lock_door(struct scsi_device *sdev)
1596{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001597 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598
James Bottomley91bc31f2009-05-17 09:30:48 -05001599 /*
1600 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1601 * request becomes available
1602 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001603 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001605 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1606 req->cmd[1] = 0;
1607 req->cmd[2] = 0;
1608 req->cmd[3] = 0;
1609 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1610 req->cmd[5] = 0;
1611
1612 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1613
1614 req->cmd_type = REQ_TYPE_BLOCK_PC;
1615 req->cmd_flags |= REQ_QUIET;
1616 req->timeout = 10 * HZ;
1617 req->retries = 5;
1618
1619 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620}
1621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622/**
1623 * scsi_restart_operations - restart io operations to the specified host.
1624 * @shost: Host we are restarting.
1625 *
1626 * Notes:
1627 * When we entered the error handler, we blocked all further i/o to
1628 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001629 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630static void scsi_restart_operations(struct Scsi_Host *shost)
1631{
1632 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001633 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634
1635 /*
1636 * If the door was locked, we need to insert a door lock request
1637 * onto the head of the SCSI request queue for the device. There
1638 * is no point trying to lock the door of an off-line device.
1639 */
1640 shost_for_each_device(sdev, shost) {
1641 if (scsi_device_online(sdev) && sdev->locked)
1642 scsi_eh_lock_door(sdev);
1643 }
1644
1645 /*
1646 * next free up anything directly waiting upon the host. this
1647 * will be requests for character device operations, and also for
1648 * ioctls to queued block devices.
1649 */
1650 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001651 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652
James Bottomley939647e2005-09-18 15:05:20 -05001653 spin_lock_irqsave(shost->host_lock, flags);
1654 if (scsi_host_set_state(shost, SHOST_RUNNING))
1655 if (scsi_host_set_state(shost, SHOST_CANCEL))
1656 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1657 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 wake_up(&shost->host_wait);
1660
1661 /*
1662 * finally we need to re-initiate requests that may be pending. we will
1663 * have had everything blocked while error handling is taking place, and
1664 * now that error recovery is done, we will need to ensure that these
1665 * requests are started.
1666 */
1667 scsi_run_host_queues(shost);
1668}
1669
1670/**
1671 * scsi_eh_ready_devs - check device ready state and recover if not.
1672 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001673 * @work_q: &list_head for pending commands.
1674 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001675 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001676void scsi_eh_ready_devs(struct Scsi_Host *shost,
1677 struct list_head *work_q,
1678 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 if (!scsi_eh_stu(shost, work_q, done_q))
1681 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001682 if (!scsi_eh_target_reset(shost, work_q, done_q))
1683 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1684 if (!scsi_eh_host_reset(work_q, done_q))
1685 scsi_eh_offline_sdevs(work_q,
1686 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001688EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689
1690/**
1691 * scsi_eh_flush_done_q - finish processed commands or retry them.
1692 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001693 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001694void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001696 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001698 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1699 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05001701 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06001702 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1704 " retry cmd: %p\n",
1705 current->comm,
1706 scmd));
1707 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1708 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001709 /*
1710 * If just we got sense for the device (called
1711 * scsi_eh_get_sense), scmd->result is already
1712 * set, do not set DRIVER_TIMEOUT.
1713 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 if (!scmd->result)
1715 scmd->result |= (DRIVER_TIMEOUT << 24);
1716 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1717 " cmd: %p\n",
1718 current->comm, scmd));
1719 scsi_finish_command(scmd);
1720 }
1721 }
1722}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001723EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725/**
1726 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1727 * @shost: Host to unjam.
1728 *
1729 * Notes:
1730 * When we come in here, we *know* that all commands on the bus have
1731 * either completed, failed or timed out. we also know that no further
1732 * commands are being sent to the host, so things are relatively quiet
1733 * and we have freedom to fiddle with things as we wish.
1734 *
1735 * This is only the *default* implementation. it is possible for
1736 * individual drivers to supply their own version of this function, and
1737 * if the maintainer wishes to do this, it is strongly suggested that
1738 * this function be taken as a template and modified. this function
1739 * was designed to correctly handle problems for about 95% of the
1740 * different cases out there, and it should always provide at least a
1741 * reasonable amount of error recovery.
1742 *
1743 * Any command marked 'failed' or 'timeout' must eventually have
1744 * scsi_finish_cmd() called for it. we do all of the retry stuff
1745 * here, so when we restart the host after we return it should have an
1746 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001747 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748static void scsi_unjam_host(struct Scsi_Host *shost)
1749{
1750 unsigned long flags;
1751 LIST_HEAD(eh_work_q);
1752 LIST_HEAD(eh_done_q);
1753
1754 spin_lock_irqsave(shost->host_lock, flags);
1755 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1756 spin_unlock_irqrestore(shost->host_lock, flags);
1757
1758 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1759
1760 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1761 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1762 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1763
1764 scsi_eh_flush_done_q(&eh_done_q);
1765}
1766
1767/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001768 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 * @data: Host for which we are running.
1770 *
1771 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001772 * This is the main error handling loop. This is run as a kernel thread
1773 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001774 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775int scsi_error_handler(void *data)
1776{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001777 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001780 * We use TASK_INTERRUPTIBLE so that the thread is not
1781 * counted against the load average as a running process.
1782 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01001783 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001785 set_current_state(TASK_INTERRUPTIBLE);
1786 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001787 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001788 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001789 SCSI_LOG_ERROR_RECOVERY(1,
1790 printk("Error handler scsi_eh_%d sleeping\n",
1791 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001792 schedule();
1793 set_current_state(TASK_INTERRUPTIBLE);
1794 continue;
1795 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
James Bottomley3ed7a472005-09-19 09:50:04 -05001797 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001798 SCSI_LOG_ERROR_RECOVERY(1,
1799 printk("Error handler scsi_eh_%d waking up\n",
1800 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 /*
1803 * We have a host that is failing for some reason. Figure out
1804 * what we need to do to get it up and online again (if we can).
1805 * If we fail, we end up taking the thing offline.
1806 */
Alan Sternbc4f2402010-06-17 10:41:42 -04001807 if (scsi_autopm_get_host(shost) != 0) {
1808 SCSI_LOG_ERROR_RECOVERY(1,
1809 printk(KERN_ERR "Error handler scsi_eh_%d "
1810 "unable to autoresume\n",
1811 shost->host_no));
1812 continue;
1813 }
1814
Christoph Hellwig9227c332006-04-01 19:21:04 +02001815 if (shost->transportt->eh_strategy_handler)
1816 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 else
1818 scsi_unjam_host(shost);
1819
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 /*
1821 * Note - if the above fails completely, the action is to take
1822 * individual devices offline and flush the queue of any
1823 * outstanding requests that may have been pending. When we
1824 * restart, we restart any I/O to any other devices on the bus
1825 * which are still online.
1826 */
1827 scsi_restart_operations(shost);
Alan Sternbc4f2402010-06-17 10:41:42 -04001828 scsi_autopm_put_host(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001829 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001831 __set_current_state(TASK_RUNNING);
1832
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001833 SCSI_LOG_ERROR_RECOVERY(1,
1834 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001835 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 return 0;
1837}
1838
1839/*
1840 * Function: scsi_report_bus_reset()
1841 *
1842 * Purpose: Utility function used by low-level drivers to report that
1843 * they have observed a bus reset on the bus being handled.
1844 *
1845 * Arguments: shost - Host in question
1846 * channel - channel on which reset was observed.
1847 *
1848 * Returns: Nothing
1849 *
1850 * Lock status: Host lock must be held.
1851 *
1852 * Notes: This only needs to be called if the reset is one which
1853 * originates from an unknown location. Resets originated
1854 * by the mid-level itself don't need to call this, but there
1855 * should be no harm.
1856 *
1857 * The main purpose of this is to make sure that a CHECK_CONDITION
1858 * is properly treated.
1859 */
1860void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1861{
1862 struct scsi_device *sdev;
1863
1864 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001865 if (channel == sdev_channel(sdev))
1866 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 }
1868}
1869EXPORT_SYMBOL(scsi_report_bus_reset);
1870
1871/*
1872 * Function: scsi_report_device_reset()
1873 *
1874 * Purpose: Utility function used by low-level drivers to report that
1875 * they have observed a device reset on the device being handled.
1876 *
1877 * Arguments: shost - Host in question
1878 * channel - channel on which reset was observed
1879 * target - target on which reset was observed
1880 *
1881 * Returns: Nothing
1882 *
1883 * Lock status: Host lock must be held
1884 *
1885 * Notes: This only needs to be called if the reset is one which
1886 * originates from an unknown location. Resets originated
1887 * by the mid-level itself don't need to call this, but there
1888 * should be no harm.
1889 *
1890 * The main purpose of this is to make sure that a CHECK_CONDITION
1891 * is properly treated.
1892 */
1893void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1894{
1895 struct scsi_device *sdev;
1896
1897 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001898 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001899 target == sdev_id(sdev))
1900 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 }
1902}
1903EXPORT_SYMBOL(scsi_report_device_reset);
1904
1905static void
1906scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1907{
1908}
1909
1910/*
1911 * Function: scsi_reset_provider
1912 *
1913 * Purpose: Send requested reset to a bus or device at any phase.
1914 *
1915 * Arguments: device - device to send reset to
1916 * flag - reset type (see scsi.h)
1917 *
1918 * Returns: SUCCESS/FAILURE.
1919 *
1920 * Notes: This is used by the SCSI Generic driver to provide
1921 * Bus/Device reset capability.
1922 */
1923int
1924scsi_reset_provider(struct scsi_device *dev, int flag)
1925{
Alan Sternbc4f2402010-06-17 10:41:42 -04001926 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05001927 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001929 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 int rtn;
1931
Alan Sternbc4f2402010-06-17 10:41:42 -04001932 if (scsi_autopm_get_host(shost) < 0)
1933 return FAILED;
1934
1935 scmd = scsi_get_command(dev, GFP_KERNEL);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001936 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001938
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001939 scmd->cmnd = req.cmd;
1940
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001942 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943
1944 scmd->cmd_len = 0;
1945
1946 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
James Smartd7a1bb02006-03-08 14:50:12 -05001948 spin_lock_irqsave(shost->host_lock, flags);
1949 shost->tmf_in_progress = 1;
1950 spin_unlock_irqrestore(shost->host_lock, flags);
1951
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 switch (flag) {
1953 case SCSI_TRY_RESET_DEVICE:
1954 rtn = scsi_try_bus_device_reset(scmd);
1955 if (rtn == SUCCESS)
1956 break;
1957 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001958 case SCSI_TRY_RESET_TARGET:
1959 rtn = scsi_try_target_reset(scmd);
1960 if (rtn == SUCCESS)
1961 break;
1962 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 case SCSI_TRY_RESET_BUS:
1964 rtn = scsi_try_bus_reset(scmd);
1965 if (rtn == SUCCESS)
1966 break;
1967 /* FALLTHROUGH */
1968 case SCSI_TRY_RESET_HOST:
1969 rtn = scsi_try_host_reset(scmd);
1970 break;
1971 default:
1972 rtn = FAILED;
1973 }
1974
James Smartd7a1bb02006-03-08 14:50:12 -05001975 spin_lock_irqsave(shost->host_lock, flags);
1976 shost->tmf_in_progress = 0;
1977 spin_unlock_irqrestore(shost->host_lock, flags);
1978
1979 /*
1980 * be sure to wake up anyone who was sleeping or had their queue
1981 * suspended while we performed the TMF.
1982 */
1983 SCSI_LOG_ERROR_RECOVERY(3,
1984 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001985 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05001986
1987 wake_up(&shost->host_wait);
1988
1989 scsi_run_host_queues(shost);
1990
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 scsi_next_command(scmd);
Alan Sternbc4f2402010-06-17 10:41:42 -04001992 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 return rtn;
1994}
1995EXPORT_SYMBOL(scsi_reset_provider);
1996
1997/**
1998 * scsi_normalize_sense - normalize main elements from either fixed or
1999 * descriptor sense data format into a common format.
2000 *
2001 * @sense_buffer: byte array containing sense data returned by device
2002 * @sb_len: number of valid bytes in sense_buffer
2003 * @sshdr: pointer to instance of structure that common
2004 * elements are written to.
2005 *
2006 * Notes:
2007 * The "main elements" from sense data are: response_code, sense_key,
2008 * asc, ascq and additional_length (only for descriptor format).
2009 *
2010 * Typically this function can be called after a device has
2011 * responded to a SCSI command with the CHECK_CONDITION status.
2012 *
2013 * Return value:
2014 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002015 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2017 struct scsi_sense_hdr *sshdr)
2018{
James Bottomley33aa6872005-08-28 11:31:14 -05002019 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 return 0;
2021
2022 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2023
2024 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05002025
2026 if (!scsi_sense_valid(sshdr))
2027 return 0;
2028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 if (sshdr->response_code >= 0x72) {
2030 /*
2031 * descriptor format
2032 */
2033 if (sb_len > 1)
2034 sshdr->sense_key = (sense_buffer[1] & 0xf);
2035 if (sb_len > 2)
2036 sshdr->asc = sense_buffer[2];
2037 if (sb_len > 3)
2038 sshdr->ascq = sense_buffer[3];
2039 if (sb_len > 7)
2040 sshdr->additional_length = sense_buffer[7];
2041 } else {
Jesper Juhl0bf8c862011-03-21 20:47:31 +01002042 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 * fixed format
2044 */
2045 if (sb_len > 2)
2046 sshdr->sense_key = (sense_buffer[2] & 0xf);
2047 if (sb_len > 7) {
2048 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2049 sb_len : (sense_buffer[7] + 8);
2050 if (sb_len > 12)
2051 sshdr->asc = sense_buffer[12];
2052 if (sb_len > 13)
2053 sshdr->ascq = sense_buffer[13];
2054 }
2055 }
2056
2057 return 1;
2058}
2059EXPORT_SYMBOL(scsi_normalize_sense);
2060
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2062 struct scsi_sense_hdr *sshdr)
2063{
2064 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002065 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066}
2067EXPORT_SYMBOL(scsi_command_normalize_sense);
2068
2069/**
Rob Landleyeb448202007-11-03 13:30:39 -05002070 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 * @sense_buffer: byte array of descriptor format sense data
2072 * @sb_len: number of valid bytes in sense_buffer
2073 * @desc_type: value of descriptor type to find
2074 * (e.g. 0 -> information)
2075 *
2076 * Notes:
2077 * only valid when sense data is in descriptor format
2078 *
2079 * Return value:
2080 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002081 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2083 int desc_type)
2084{
2085 int add_sen_len, add_len, desc_len, k;
2086 const u8 * descp;
2087
2088 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2089 return NULL;
2090 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2091 return NULL;
2092 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2093 add_sen_len : (sb_len - 8);
2094 descp = &sense_buffer[8];
2095 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2096 descp += desc_len;
2097 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2098 desc_len = add_len + 2;
2099 if (descp[0] == desc_type)
2100 return descp;
2101 if (add_len < 0) // short descriptor ??
2102 break;
2103 }
2104 return NULL;
2105}
2106EXPORT_SYMBOL(scsi_sense_desc_find);
2107
2108/**
Rob Landleyeb448202007-11-03 13:30:39 -05002109 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 * @sense_buffer: byte array of sense data
2111 * @sb_len: number of valid bytes in sense_buffer
2112 * @info_out: pointer to 64 integer where 8 or 4 byte information
2113 * field will be placed if found.
2114 *
2115 * Return value:
2116 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002117 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2119 u64 * info_out)
2120{
2121 int j;
2122 const u8 * ucp;
2123 u64 ull;
2124
2125 if (sb_len < 7)
2126 return 0;
2127 switch (sense_buffer[0] & 0x7f) {
2128 case 0x70:
2129 case 0x71:
2130 if (sense_buffer[0] & 0x80) {
2131 *info_out = (sense_buffer[3] << 24) +
2132 (sense_buffer[4] << 16) +
2133 (sense_buffer[5] << 8) + sense_buffer[6];
2134 return 1;
2135 } else
2136 return 0;
2137 case 0x72:
2138 case 0x73:
2139 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2140 0 /* info desc */);
2141 if (ucp && (0xa == ucp[1])) {
2142 ull = 0;
2143 for (j = 0; j < 8; ++j) {
2144 if (j > 0)
2145 ull <<= 8;
2146 ull |= ucp[4 + j];
2147 }
2148 *info_out = ull;
2149 return 1;
2150 } else
2151 return 0;
2152 default:
2153 return 0;
2154 }
2155}
2156EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002157
2158/**
2159 * scsi_build_sense_buffer - build sense data in a buffer
2160 * @desc: Sense format (non zero == descriptor format,
2161 * 0 == fixed format)
2162 * @buf: Where to build sense data
2163 * @key: Sense key
2164 * @asc: Additional sense code
2165 * @ascq: Additional sense code qualifier
2166 *
2167 **/
2168void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2169{
2170 if (desc) {
2171 buf[0] = 0x72; /* descriptor, current */
2172 buf[1] = key;
2173 buf[2] = asc;
2174 buf[3] = ascq;
2175 buf[7] = 0;
2176 } else {
2177 buf[0] = 0x70; /* fixed, current */
2178 buf[2] = key;
2179 buf[7] = 0xa;
2180 buf[12] = asc;
2181 buf[13] = ascq;
2182 }
2183}
2184EXPORT_SYMBOL(scsi_build_sense_buffer);