blob: 5f84a148eb142417d9e296191329a6126e89ef4a [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) {
TARUISI Hiroakidfcf7772011-08-11 20:25:20 +0900296 /*
297 * Because some device does not queue unit
298 * attentions correctly, we carefully check
299 * additional sense code and qualifier so as
300 * not to squash media change unit attention.
301 */
302 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
303 scmd->device->expecting_cc_ua = 0;
304 return NEEDS_RETRY;
305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 }
307 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100308 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 * should retry.
310 */
311 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
312 return NEEDS_RETRY;
313 /*
314 * if the device is not started, we need to wake
315 * the error handler to start the motor
316 */
317 if (scmd->device->allow_restart &&
318 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
319 return FAILED;
James Bottomley77a42292010-05-04 16:51:40 -0400320
Mike Christie6e499492010-04-25 07:03:57 -0500321 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
322 scmd_printk(KERN_WARNING, scmd,
323 "Warning! Received an indication that the "
324 "LUN assignments on this target have "
325 "changed. The Linux SCSI layer does not "
326 "automatically remap LUN assignments.\n");
327 else if (sshdr.asc == 0x3f)
328 scmd_printk(KERN_WARNING, scmd,
329 "Warning! Received an indication that the "
330 "operating parameters on this target have "
331 "changed. The Linux SCSI layer does not "
332 "automatically adjust these parameters.\n");
333
Shyam Iyerdeb1cb62011-02-26 01:59:44 -0500334 if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
335 scmd_printk(KERN_WARNING, scmd,
336 "Warning! Received an indication that the "
337 "LUN reached a thin provisioning soft "
338 "threshold.\n");
339
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100340 /*
341 * Pass the UA upwards for a determination in the completion
342 * functions.
343 */
344 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100346 /* these are not supported */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 case COPY_ABORTED:
348 case VOLUME_OVERFLOW:
349 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100350 case BLANK_CHECK:
351 case DATA_PROTECT:
352 return TARGET_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800355 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
356 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
357 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100358 return TARGET_ERROR;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800359 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 return NEEDS_RETRY;
361
362 case HARDWARE_ERROR:
363 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700364 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 else
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100366 return TARGET_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367
368 case ILLEGAL_REQUEST:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 default:
370 return SUCCESS;
371 }
372}
373
Vasu Dev4a840672009-10-22 15:46:33 -0700374static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
375{
376 struct scsi_host_template *sht = sdev->host->hostt;
377 struct scsi_device *tmp_sdev;
378
379 if (!sht->change_queue_depth ||
380 sdev->queue_depth >= sdev->max_queue_depth)
381 return;
382
383 if (time_before(jiffies,
384 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
385 return;
386
387 if (time_before(jiffies,
388 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
389 return;
390
391 /*
392 * Walk all devices of a target and do
393 * ramp up on them.
394 */
395 shost_for_each_device(tmp_sdev, sdev->host) {
396 if (tmp_sdev->channel != sdev->channel ||
397 tmp_sdev->id != sdev->id ||
398 tmp_sdev->queue_depth == sdev->max_queue_depth)
399 continue;
400 /*
401 * call back into LLD to increase queue_depth by one
402 * with ramp up reason code.
403 */
404 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
405 SCSI_QDEPTH_RAMP_UP);
406 sdev->last_queue_ramp_up = jiffies;
407 }
408}
409
Mike Christie42a6a912009-10-15 17:46:44 -0700410static void scsi_handle_queue_full(struct scsi_device *sdev)
411{
412 struct scsi_host_template *sht = sdev->host->hostt;
413 struct scsi_device *tmp_sdev;
414
415 if (!sht->change_queue_depth)
416 return;
417
418 shost_for_each_device(tmp_sdev, sdev->host) {
419 if (tmp_sdev->channel != sdev->channel ||
420 tmp_sdev->id != sdev->id)
421 continue;
422 /*
423 * We do not know the number of commands that were at
424 * the device when we got the queue full so we start
425 * from the highest possible value and work our way down.
426 */
427 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
428 SCSI_QDEPTH_QFULL);
429 }
430}
431
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432/**
433 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
434 * @scmd: SCSI cmd to examine.
435 *
436 * Notes:
437 * This is *only* called when we are examining the status of commands
438 * queued during error recovery. the main difference here is that we
439 * don't allow for the possibility of retries here, and we are a lot
440 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800441 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
443{
444 /*
445 * first check the host byte, to see if there is anything in there
446 * that would indicate what we need to do.
447 */
448 if (host_byte(scmd->result) == DID_RESET) {
449 /*
450 * rats. we are already in the error handler, so we now
451 * get to try and figure out what to do next. if the sense
452 * is valid, we have a pretty good idea of what to do.
453 * if not, we mark it as FAILED.
454 */
455 return scsi_check_sense(scmd);
456 }
457 if (host_byte(scmd->result) != DID_OK)
458 return FAILED;
459
460 /*
461 * next, check the message byte.
462 */
463 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
464 return FAILED;
465
466 /*
467 * now, check the status byte to see if this indicates
468 * anything special.
469 */
470 switch (status_byte(scmd->result)) {
471 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700472 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 case COMMAND_TERMINATED:
474 return SUCCESS;
475 case CHECK_CONDITION:
476 return scsi_check_sense(scmd);
477 case CONDITION_GOOD:
478 case INTERMEDIATE_GOOD:
479 case INTERMEDIATE_C_GOOD:
480 /*
481 * who knows? FIXME(eric)
482 */
483 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500484 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500485 if (scmd->cmnd[0] == TEST_UNIT_READY)
486 /* it is a success, we probed the device and
487 * found it */
488 return SUCCESS;
489 /* otherwise, we failed to send the command */
490 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700492 scsi_handle_queue_full(scmd->device);
493 /* fall through */
494 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200495 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 default:
497 return FAILED;
498 }
499 return FAILED;
500}
501
502/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 * scsi_eh_done - Completion function for error handling.
504 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800505 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506static void scsi_eh_done(struct scsi_cmnd *scmd)
507{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100508 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600509
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100510 SCSI_LOG_ERROR_RECOVERY(3,
511 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700512 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600513
514 eh_action = scmd->device->host->eh_action;
515 if (eh_action)
516 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517}
518
519/**
Brian King292148f2007-01-30 17:51:17 -0600520 * scsi_try_host_reset - ask host adapter to reset itself
521 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800522 */
Brian King292148f2007-01-30 17:51:17 -0600523static int scsi_try_host_reset(struct scsi_cmnd *scmd)
524{
525 unsigned long flags;
526 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100527 struct Scsi_Host *host = scmd->device->host;
528 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600529
530 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700531 __func__));
Brian King292148f2007-01-30 17:51:17 -0600532
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100533 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600534 return FAILED;
535
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100536 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600537
538 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100539 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600540 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100541 spin_lock_irqsave(host->host_lock, flags);
542 scsi_report_bus_reset(host, scmd_channel(scmd));
543 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600544 }
545
546 return rtn;
547}
548
549/**
550 * scsi_try_bus_reset - ask host to perform a bus reset
551 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800552 */
Brian King292148f2007-01-30 17:51:17 -0600553static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
554{
555 unsigned long flags;
556 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100557 struct Scsi_Host *host = scmd->device->host;
558 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600559
560 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700561 __func__));
Brian King292148f2007-01-30 17:51:17 -0600562
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100563 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600564 return FAILED;
565
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100566 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600567
568 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100569 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600570 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100571 spin_lock_irqsave(host->host_lock, flags);
572 scsi_report_bus_reset(host, scmd_channel(scmd));
573 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600574 }
575
576 return rtn;
577}
578
Mike Christie30bd7df2008-02-29 18:25:19 -0600579static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
580{
581 sdev->was_reset = 1;
582 sdev->expecting_cc_ua = 1;
583}
584
585/**
586 * scsi_try_target_reset - Ask host to perform a target reset
587 * @scmd: SCSI cmd used to send a target reset
588 *
589 * Notes:
590 * There is no timeout for this operation. if this operation is
591 * unreliable for a given host, then the host itself needs to put a
592 * timer on it, and set the host back to a consistent state prior to
593 * returning.
594 */
595static int scsi_try_target_reset(struct scsi_cmnd *scmd)
596{
597 unsigned long flags;
598 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100599 struct Scsi_Host *host = scmd->device->host;
600 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600601
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100602 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600603 return FAILED;
604
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100605 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600606 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100607 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600608 __starget_for_each_device(scsi_target(scmd->device), NULL,
609 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100610 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600611 }
612
613 return rtn;
614}
615
Brian King292148f2007-01-30 17:51:17 -0600616/**
617 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
618 * @scmd: SCSI cmd used to send BDR
619 *
620 * Notes:
621 * There is no timeout for this operation. if this operation is
622 * unreliable for a given host, then the host itself needs to put a
623 * timer on it, and set the host back to a consistent state prior to
624 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800625 */
Brian King292148f2007-01-30 17:51:17 -0600626static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
627{
628 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100629 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600630
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100631 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600632 return FAILED;
633
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100634 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600635 if (rtn == SUCCESS)
636 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600637 return rtn;
638}
639
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100640static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600641{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100642 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600643 return FAILED;
644
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100645 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600646}
647
Brian King292148f2007-01-30 17:51:17 -0600648static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
649{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100650 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600651 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600652 if (scsi_try_target_reset(scmd) != SUCCESS)
653 if (scsi_try_bus_reset(scmd) != SUCCESS)
654 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600655}
656
657/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200658 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100659 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200660 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200661 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300662 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200663 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100664 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200665 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200666 * as part of the error recovery process. If @sense_bytes is 0 the command
667 * sent must be one that does not transfer any data. If @sense_bytes != 0
668 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
669 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800670 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200671void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
672 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673{
f59114b2005-04-17 15:00:23 -0500674 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
Christoph Hellwig631c2282006-07-08 20:42:15 +0200676 /*
677 * We need saved copies of a number of fields - this is because
678 * error handling may need to overwrite these with different values
679 * to run different commands, and once error handling is complete,
680 * we will need to restore these values prior to running the actual
681 * command.
682 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200683 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300684 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200685 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200686 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200687 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200688 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400689 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400690 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200691
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400692 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300693 scmd->cmnd = ses->eh_cmnd;
694 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200695 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200696 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200697
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200698 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200699 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
700 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200701 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200702 scmd->sdb.length);
703 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200704 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200705 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200706 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200707 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200708 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200709 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200710 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200711 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300712 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200713 memcpy(scmd->cmnd, cmnd, cmnd_size);
714 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
715 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200716 }
717
718 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200719
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200720 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500722 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
Christoph Hellwig631c2282006-07-08 20:42:15 +0200724 /*
725 * Zero the sense buffer. The scsi spec mandates that any
726 * untransferred sense data should be interpreted as being zero.
727 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900728 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200729}
730EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200731
Boaz Harroshe1c23462007-10-08 16:36:45 +0200732/**
733 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
734 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200735 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200736 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200737 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800738 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200739void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
740{
741 /*
742 * Restore original data
743 */
744 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300745 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200746 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200747 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200748 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200749 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400750 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400751 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200752}
753EXPORT_SYMBOL(scsi_eh_restore_cmnd);
754
755/**
756 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
757 * @scmd: SCSI command structure to hijack
758 * @cmnd: CDB to send
759 * @cmnd_size: size in bytes of @cmnd
760 * @timeout: timeout for this request
761 * @sense_bytes: size of sense data to copy or 0
762 *
763 * This function is used to send a scsi command down to a target device
764 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
765 *
766 * Return value:
767 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800768 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200769static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
770 int cmnd_size, int timeout, unsigned sense_bytes)
771{
772 struct scsi_device *sdev = scmd->device;
773 struct Scsi_Host *shost = sdev->host;
774 DECLARE_COMPLETION_ONSTACK(done);
775 unsigned long timeleft;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200776 struct scsi_eh_save ses;
777 int rtn;
778
779 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100780 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -0500783 scmd->scsi_done = scsi_eh_done;
784 shost->hostt->queuecommand(shost, scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100786 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
f59114b2005-04-17 15:00:23 -0500788 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100790 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100792 SCSI_LOG_ERROR_RECOVERY(3,
793 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700794 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
796 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100797 * If there is time left scsi_eh_done got called, and we will
798 * examine the actual status codes to see whether the command
799 * actually did complete normally, else tell the host to forget
800 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100802 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 rtn = scsi_eh_completed_normally(scmd);
804 SCSI_LOG_ERROR_RECOVERY(3,
805 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700806 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100807
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 switch (rtn) {
809 case SUCCESS:
810 case NEEDS_RETRY:
811 case FAILED:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100812 case TARGET_ERROR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +0200814 case ADD_TO_MLQUEUE:
815 rtn = NEEDS_RETRY;
816 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 default:
818 rtn = FAILED;
819 break;
820 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100821 } else {
Brian King292148f2007-01-30 17:51:17 -0600822 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100823 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 }
825
Boaz Harroshe1c23462007-10-08 16:36:45 +0200826 scsi_eh_restore_cmnd(scmd, &ses);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 return rtn;
828}
829
830/**
831 * scsi_request_sense - Request sense data from a particular target.
832 * @scmd: SCSI cmd for request sense.
833 *
834 * Notes:
835 * Some hosts automatically obtain this information, others require
836 * that we obtain it on our own. This function will *not* return until
837 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800838 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839static int scsi_request_sense(struct scsi_cmnd *scmd)
840{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200841 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842}
843
844/**
845 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
846 * @scmd: Original SCSI cmd that eh has finished.
847 * @done_q: Queue for processed commands.
848 *
849 * Notes:
850 * We don't want to use the normal command completion while we are are
851 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500852 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 * keep a list of pending commands for final completion, and once we
854 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800855 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900856void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857{
858 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200859 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 list_move_tail(&scmd->eh_entry, done_q);
861}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900862EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864/**
865 * scsi_eh_get_sense - Get device sense data.
866 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500867 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 *
869 * Description:
870 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100871 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 *
873 * Notes:
874 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500875 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 * it down before we request it.
877 *
878 * All drivers should request sense information internally these days,
879 * so for now all I have to say is tough noogies if you end up in here.
880 *
881 * XXX: Long term this code should go away, but that needs an audit of
882 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800883 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800884int scsi_eh_get_sense(struct list_head *work_q,
885 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200887 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 int rtn;
889
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200890 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200891 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 SCSI_SENSE_VALID(scmd))
893 continue;
894
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400895 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
896 "%s: requesting sense\n",
897 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 rtn = scsi_request_sense(scmd);
899 if (rtn != SUCCESS)
900 continue;
901
902 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
903 " result %x\n", scmd,
904 scmd->result));
905 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
906
907 rtn = scsi_decide_disposition(scmd);
908
909 /*
910 * if the result was normal, then just pass it along to the
911 * upper level.
912 */
913 if (rtn == SUCCESS)
914 /* we don't want this command reissued, just
915 * finished with the sense data, so set
916 * retries to the max allowed to ensure it
917 * won't get reissued */
918 scmd->retries = scmd->allowed;
919 else if (rtn != NEEDS_RETRY)
920 continue;
921
922 scsi_eh_finish_cmd(scmd, done_q);
923 }
924
925 return list_empty(work_q);
926}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800927EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500931 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 *
933 * Return value:
934 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800935 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936static int scsi_eh_tur(struct scsi_cmnd *scmd)
937{
938 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
939 int retry_cnt = 1, rtn;
940
941retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400942 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700945 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200946
947 switch (rtn) {
948 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (retry_cnt--)
950 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200951 /*FALLTHRU*/
952 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500953 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200954 default:
955 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500956 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
959/**
David Jeffery3eef6252011-05-19 14:41:12 -0400960 * scsi_eh_test_devices - check if devices are responding from error recovery.
961 * @cmd_list: scsi commands in error recovery.
962 * @work_q: queue for commands which still need more error recovery
963 * @done_q: queue for commands which are finished
964 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
965 *
966 * Decription:
967 * Tests if devices are in a working state. Commands to devices now in
968 * a working state are sent to the done_q while commands to devices which
969 * are still failing to respond are returned to the work_q for more
970 * processing.
971 **/
972static int scsi_eh_test_devices(struct list_head *cmd_list,
973 struct list_head *work_q,
974 struct list_head *done_q, int try_stu)
975{
976 struct scsi_cmnd *scmd, *next;
977 struct scsi_device *sdev;
978 int finish_cmds;
979
980 while (!list_empty(cmd_list)) {
981 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
982 sdev = scmd->device;
983
984 finish_cmds = !scsi_device_online(scmd->device) ||
985 (try_stu && !scsi_eh_try_stu(scmd) &&
986 !scsi_eh_tur(scmd)) ||
987 !scsi_eh_tur(scmd);
988
989 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
990 if (scmd->device == sdev) {
991 if (finish_cmds)
992 scsi_eh_finish_cmd(scmd, done_q);
993 else
994 list_move_tail(&scmd->eh_entry, work_q);
995 }
996 }
997 return list_empty(work_q);
998}
999
1000
1001/**
Rob Landleyeb448202007-11-03 13:30:39 -05001002 * scsi_eh_abort_cmds - abort pending commands.
1003 * @work_q: &list_head for pending commands.
1004 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 *
1006 * Decription:
1007 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -05001008 * running command. This only works out to be the case if we have one
1009 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 * no sense to try and abort the command, since as far as the shost
1011 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001012 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013static int scsi_eh_abort_cmds(struct list_head *work_q,
1014 struct list_head *done_q)
1015{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001016 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001017 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 int rtn;
1019
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001020 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001021 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 continue;
1023 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1024 "0x%p\n", current->comm,
1025 scmd));
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001026 rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001027 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001028 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
David Jeffery3eef6252011-05-19 14:41:12 -04001029 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 scsi_eh_finish_cmd(scmd, done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001031 else
1032 list_move_tail(&scmd->eh_entry, &check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 } else
1034 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1035 " cmd failed:"
1036 "0x%p\n",
1037 current->comm,
1038 scmd));
1039 }
1040
David Jeffery3eef6252011-05-19 14:41:12 -04001041 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042}
1043
1044/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001046 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 *
1048 * Return value:
1049 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001050 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1052{
1053 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054
Christoph Hellwig631c2282006-07-08 20:42:15 +02001055 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001056 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Brian Kinged773e62007-03-29 15:25:52 -05001058 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001059 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001060
Christoph Hellwig631c2282006-07-08 20:42:15 +02001061 if (rtn == SUCCESS)
1062 return 0;
1063 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 return 1;
1066}
1067
1068 /**
1069 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001070 * @shost: &scsi host being recovered.
1071 * @work_q: &list_head for pending commands.
1072 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 *
1074 * Notes:
1075 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001076 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001077 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078static int scsi_eh_stu(struct Scsi_Host *shost,
1079 struct list_head *work_q,
1080 struct list_head *done_q)
1081{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001082 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 struct scsi_device *sdev;
1084
1085 shost_for_each_device(sdev, shost) {
1086 stu_scmd = NULL;
1087 list_for_each_entry(scmd, work_q, eh_entry)
1088 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1089 scsi_check_sense(scmd) == FAILED ) {
1090 stu_scmd = scmd;
1091 break;
1092 }
1093
1094 if (!stu_scmd)
1095 continue;
1096
1097 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1098 " 0x%p\n", current->comm, sdev));
1099
1100 if (!scsi_eh_try_stu(stu_scmd)) {
1101 if (!scsi_device_online(sdev) ||
1102 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001103 list_for_each_entry_safe(scmd, next,
1104 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 if (scmd->device == sdev)
1106 scsi_eh_finish_cmd(scmd, done_q);
1107 }
1108 }
1109 } else {
1110 SCSI_LOG_ERROR_RECOVERY(3,
1111 printk("%s: START_UNIT failed to sdev:"
1112 " 0x%p\n", current->comm, sdev));
1113 }
1114 }
1115
1116 return list_empty(work_q);
1117}
1118
1119
1120/**
1121 * scsi_eh_bus_device_reset - send bdr if needed
1122 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001123 * @work_q: &list_head for pending commands.
1124 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 *
1126 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001127 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 * devices that are jammed or not - if we have multiple devices, it
1129 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001130 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001131 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1133 struct list_head *work_q,
1134 struct list_head *done_q)
1135{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001136 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 struct scsi_device *sdev;
1138 int rtn;
1139
1140 shost_for_each_device(sdev, shost) {
1141 bdr_scmd = NULL;
1142 list_for_each_entry(scmd, work_q, eh_entry)
1143 if (scmd->device == sdev) {
1144 bdr_scmd = scmd;
1145 break;
1146 }
1147
1148 if (!bdr_scmd)
1149 continue;
1150
1151 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1152 " 0x%p\n", current->comm,
1153 sdev));
1154 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001155 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001157 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001159 list_for_each_entry_safe(scmd, next,
1160 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (scmd->device == sdev)
1162 scsi_eh_finish_cmd(scmd,
1163 done_q);
1164 }
1165 }
1166 } else {
1167 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1168 " failed sdev:"
1169 "0x%p\n",
1170 current->comm,
1171 sdev));
1172 }
1173 }
1174
1175 return list_empty(work_q);
1176}
1177
1178/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001179 * scsi_eh_target_reset - send target reset if needed
1180 * @shost: scsi host being recovered.
1181 * @work_q: &list_head for pending commands.
1182 * @done_q: &list_head for processed commands.
1183 *
1184 * Notes:
1185 * Try a target reset.
1186 */
1187static int scsi_eh_target_reset(struct Scsi_Host *shost,
1188 struct list_head *work_q,
1189 struct list_head *done_q)
1190{
James Bottomley98db5192010-10-25 15:53:41 -05001191 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001192 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001193
James Bottomley98db5192010-10-25 15:53:41 -05001194 list_splice_init(work_q, &tmp_list);
1195
1196 while (!list_empty(&tmp_list)) {
1197 struct scsi_cmnd *next, *scmd;
1198 int rtn;
1199 unsigned int id;
1200
1201 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1202 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001203
1204 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1205 "to target %d\n",
1206 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001207 rtn = scsi_try_target_reset(scmd);
1208 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Mike Christie30bd7df2008-02-29 18:25:19 -06001209 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1210 " failed target: "
1211 "%d\n",
1212 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001213 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1214 if (scmd_id(scmd) != id)
1215 continue;
1216
David Jeffery3eef6252011-05-19 14:41:12 -04001217 if (rtn == SUCCESS)
1218 list_move_tail(&scmd->eh_entry, &check_list);
1219 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001220 scsi_eh_finish_cmd(scmd, done_q);
1221 else
1222 /* push back on work queue for further processing */
1223 list_move(&scmd->eh_entry, work_q);
1224 }
1225 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001226
David Jeffery3eef6252011-05-19 14:41:12 -04001227 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001228}
1229
1230/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001231 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001232 * @shost: &scsi host being recovered.
1233 * @work_q: &list_head for pending commands.
1234 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001235 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1237 struct list_head *work_q,
1238 struct list_head *done_q)
1239{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001240 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001241 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 unsigned int channel;
1243 int rtn;
1244
1245 /*
1246 * we really want to loop over the various channels, and do this on
1247 * a channel by channel basis. we should also check to see if any
1248 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001249 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 */
1251
1252 for (channel = 0; channel <= shost->max_channel; channel++) {
1253 chan_scmd = NULL;
1254 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001255 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 chan_scmd = scmd;
1257 break;
1258 /*
1259 * FIXME add back in some support for
1260 * soft_reset devices.
1261 */
1262 }
1263 }
1264
1265 if (!chan_scmd)
1266 continue;
1267 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1268 " %d\n", current->comm,
1269 channel));
1270 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001271 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001272 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001273 if (channel == scmd_channel(scmd)) {
1274 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 scsi_eh_finish_cmd(scmd,
1276 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001277 else
1278 list_move_tail(&scmd->eh_entry,
1279 &check_list);
1280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 }
1282 } else {
1283 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1284 " failed chan: %d\n",
1285 current->comm,
1286 channel));
1287 }
1288 }
David Jeffery3eef6252011-05-19 14:41:12 -04001289 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290}
1291
1292/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001293 * scsi_eh_host_reset - send a host reset
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 * @work_q: list_head for processed commands.
1295 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001296 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297static int scsi_eh_host_reset(struct list_head *work_q,
1298 struct list_head *done_q)
1299{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001300 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001301 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
1304 if (!list_empty(work_q)) {
1305 scmd = list_entry(work_q->next,
1306 struct scsi_cmnd, eh_entry);
1307
1308 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1309 , current->comm));
1310
1311 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001312 if (rtn == SUCCESS) {
1313 list_splice_init(work_q, &check_list);
1314 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001315 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 scsi_eh_finish_cmd(scmd, done_q);
1317 }
1318 } else {
1319 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1320 " failed\n",
1321 current->comm));
1322 }
1323 }
David Jeffery3eef6252011-05-19 14:41:12 -04001324 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325}
1326
1327/**
1328 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1329 * @work_q: list_head for processed commands.
1330 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001331 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332static void scsi_eh_offline_sdevs(struct list_head *work_q,
1333 struct list_head *done_q)
1334{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001335 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001337 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001338 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1339 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001341 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 /*
1343 * FIXME: Handle lost cmds.
1344 */
1345 }
1346 scsi_eh_finish_cmd(scmd, done_q);
1347 }
1348 return;
1349}
1350
1351/**
Mike Christie4a274462008-08-19 18:45:31 -05001352 * scsi_noretry_cmd - determinte if command should be failed fast
1353 * @scmd: SCSI cmd to examine.
1354 */
1355int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1356{
1357 switch (host_byte(scmd->result)) {
1358 case DID_OK:
1359 break;
1360 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001361 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001362 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001363 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001364 case DID_ERROR:
1365 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1366 status_byte(scmd->result) == RESERVATION_CONFLICT)
1367 return 0;
1368 /* fall through */
1369 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001370 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001371 }
1372
1373 switch (status_byte(scmd->result)) {
1374 case CHECK_CONDITION:
1375 /*
1376 * assume caller has checked sense and determinted
1377 * the check condition was retryable.
1378 */
FUJITA Tomonorie96f6ab2010-07-09 09:38:26 +09001379 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1380 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1381 return 1;
Mike Christie4a274462008-08-19 18:45:31 -05001382 }
1383
1384 return 0;
1385}
1386
1387/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1389 * @scmd: SCSI cmd to examine.
1390 *
1391 * Notes:
1392 * This is *only* called when we are examining the status after sending
1393 * out the actual data command. any commands that are queued for error
1394 * recovery (e.g. test_unit_ready) do *not* come through here.
1395 *
1396 * When this routine returns failed, it means the error handler thread
1397 * is woken. In cases where the error code indicates an error that
1398 * doesn't require the error handler read (i.e. we don't need to
1399 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001400 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401int scsi_decide_disposition(struct scsi_cmnd *scmd)
1402{
1403 int rtn;
1404
1405 /*
1406 * if the device is offline, then we clearly just pass the result back
1407 * up to the top level.
1408 */
1409 if (!scsi_device_online(scmd->device)) {
1410 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1411 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001412 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 return SUCCESS;
1414 }
1415
1416 /*
1417 * first check the host byte, to see if there is anything in there
1418 * that would indicate what we need to do.
1419 */
1420 switch (host_byte(scmd->result)) {
1421 case DID_PASSTHROUGH:
1422 /*
1423 * no matter what, pass this through to the upper layer.
1424 * nuke this special code so that it looks like we are saying
1425 * did_ok.
1426 */
1427 scmd->result &= 0xff00ffff;
1428 return SUCCESS;
1429 case DID_OK:
1430 /*
1431 * looks good. drop through, and check the next byte.
1432 */
1433 break;
1434 case DID_NO_CONNECT:
1435 case DID_BAD_TARGET:
1436 case DID_ABORT:
1437 /*
1438 * note - this means that we just report the status back
1439 * to the top level driver, not that we actually think
1440 * that it indicates SUCCESS.
1441 */
1442 return SUCCESS;
1443 /*
1444 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001445 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 * in order to avoid endless loops (db)
1447 *
1448 * actually this is a bug in this function here. we should
1449 * be mindful of the maximum number of retries specified
1450 * and not get stuck in a loop.
1451 */
1452 case DID_SOFT_ERROR:
1453 goto maybe_retry;
1454 case DID_IMM_RETRY:
1455 return NEEDS_RETRY;
1456
bf341912005-04-12 17:49:09 -05001457 case DID_REQUEUE:
1458 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001459 case DID_TRANSPORT_DISRUPTED:
1460 /*
1461 * LLD/transport was disrupted during processing of the IO.
1462 * The transport class is now blocked/blocking,
1463 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001464 * based on its timers and recovery capablilities if
1465 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001466 */
Mike Christie939c2282008-11-04 19:47:19 -06001467 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001468 case DID_TRANSPORT_FAILFAST:
1469 /*
1470 * The transport decided to failfast the IO (most likely
1471 * the fast io fail tmo fired), so send IO directly upwards.
1472 */
1473 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 case DID_ERROR:
1475 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1476 status_byte(scmd->result) == RESERVATION_CONFLICT)
1477 /*
1478 * execute reservation conflict processing code
1479 * lower down
1480 */
1481 break;
1482 /* fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 case DID_BUS_BUSY:
1484 case DID_PARITY:
1485 goto maybe_retry;
1486 case DID_TIME_OUT:
1487 /*
1488 * when we scan the bus, we get timeout messages for
1489 * these commands if there is no device available.
1490 * other hosts report did_no_connect for the same thing.
1491 */
1492 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1493 scmd->cmnd[0] == INQUIRY)) {
1494 return SUCCESS;
1495 } else {
1496 return FAILED;
1497 }
1498 case DID_RESET:
1499 return SUCCESS;
1500 default:
1501 return FAILED;
1502 }
1503
1504 /*
1505 * next, check the message byte.
1506 */
1507 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1508 return FAILED;
1509
1510 /*
1511 * check the status byte to see if this indicates anything special.
1512 */
1513 switch (status_byte(scmd->result)) {
1514 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001515 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 /*
1517 * the case of trying to send too many commands to a
1518 * tagged queueing device.
1519 */
1520 case BUSY:
1521 /*
1522 * device can't talk to us at the moment. Should only
1523 * occur (SAM-3) when the task queue is empty, so will cause
1524 * the empty queue handling to trigger a stall in the
1525 * device.
1526 */
1527 return ADD_TO_MLQUEUE;
1528 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -07001529 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001532 case TASK_ABORTED:
1533 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 case CHECK_CONDITION:
1535 rtn = scsi_check_sense(scmd);
1536 if (rtn == NEEDS_RETRY)
1537 goto maybe_retry;
Hannes Reinecke63583cc2011-01-18 10:13:11 +01001538 else if (rtn == TARGET_ERROR) {
1539 /*
1540 * Need to modify host byte to signal a
1541 * permanent target failure
1542 */
1543 scmd->result |= (DID_TARGET_FAILURE << 16);
1544 rtn = SUCCESS;
1545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 /* if rtn == FAILED, we have no sense information;
1547 * returning FAILED will wake the error handler thread
1548 * to collect the sense and redo the decide
1549 * disposition */
1550 return rtn;
1551 case CONDITION_GOOD:
1552 case INTERMEDIATE_GOOD:
1553 case INTERMEDIATE_C_GOOD:
1554 case ACA_ACTIVE:
1555 /*
1556 * who knows? FIXME(eric)
1557 */
1558 return SUCCESS;
1559
1560 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001561 sdev_printk(KERN_INFO, scmd->device,
1562 "reservation conflict\n");
Hannes Reinecke63583cc2011-01-18 10:13:11 +01001563 scmd->result |= (DID_NEXUS_FAILURE << 16);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 return SUCCESS; /* causes immediate i/o error */
1565 default:
1566 return FAILED;
1567 }
1568 return FAILED;
1569
1570 maybe_retry:
1571
1572 /* we requeue for retry because the error was retryable, and
1573 * the request was not marked fast fail. Note that above,
1574 * even if the request is marked fast fail, we still requeue
1575 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001576 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001577 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return NEEDS_RETRY;
1579 } else {
1580 /*
1581 * no more retries - report this one back to upper level.
1582 */
1583 return SUCCESS;
1584 }
1585}
1586
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001587static void eh_lock_door_done(struct request *req, int uptodate)
1588{
1589 __blk_put_request(req->q, req);
1590}
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 * scsi_eh_lock_door - Prevent medium removal for the specified device
1594 * @sdev: SCSI device to prevent medium removal
1595 *
1596 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001597 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 *
1599 * Notes:
1600 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1601 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001602 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603static void scsi_eh_lock_door(struct scsi_device *sdev)
1604{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001605 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
James Bottomley91bc31f2009-05-17 09:30:48 -05001607 /*
1608 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1609 * request becomes available
1610 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001611 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001613 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1614 req->cmd[1] = 0;
1615 req->cmd[2] = 0;
1616 req->cmd[3] = 0;
1617 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1618 req->cmd[5] = 0;
1619
1620 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1621
1622 req->cmd_type = REQ_TYPE_BLOCK_PC;
1623 req->cmd_flags |= REQ_QUIET;
1624 req->timeout = 10 * HZ;
1625 req->retries = 5;
1626
1627 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628}
1629
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630/**
1631 * scsi_restart_operations - restart io operations to the specified host.
1632 * @shost: Host we are restarting.
1633 *
1634 * Notes:
1635 * When we entered the error handler, we blocked all further i/o to
1636 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001637 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638static void scsi_restart_operations(struct Scsi_Host *shost)
1639{
1640 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001641 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642
1643 /*
1644 * If the door was locked, we need to insert a door lock request
1645 * onto the head of the SCSI request queue for the device. There
1646 * is no point trying to lock the door of an off-line device.
1647 */
1648 shost_for_each_device(sdev, shost) {
1649 if (scsi_device_online(sdev) && sdev->locked)
1650 scsi_eh_lock_door(sdev);
1651 }
1652
1653 /*
1654 * next free up anything directly waiting upon the host. this
1655 * will be requests for character device operations, and also for
1656 * ioctls to queued block devices.
1657 */
1658 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001659 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
James Bottomley939647e2005-09-18 15:05:20 -05001661 spin_lock_irqsave(shost->host_lock, flags);
1662 if (scsi_host_set_state(shost, SHOST_RUNNING))
1663 if (scsi_host_set_state(shost, SHOST_CANCEL))
1664 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1665 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667 wake_up(&shost->host_wait);
1668
1669 /*
1670 * finally we need to re-initiate requests that may be pending. we will
1671 * have had everything blocked while error handling is taking place, and
1672 * now that error recovery is done, we will need to ensure that these
1673 * requests are started.
1674 */
1675 scsi_run_host_queues(shost);
1676}
1677
1678/**
1679 * scsi_eh_ready_devs - check device ready state and recover if not.
1680 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001681 * @work_q: &list_head for pending commands.
1682 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001683 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001684void scsi_eh_ready_devs(struct Scsi_Host *shost,
1685 struct list_head *work_q,
1686 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687{
1688 if (!scsi_eh_stu(shost, work_q, done_q))
1689 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001690 if (!scsi_eh_target_reset(shost, work_q, done_q))
1691 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1692 if (!scsi_eh_host_reset(work_q, done_q))
1693 scsi_eh_offline_sdevs(work_q,
1694 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001696EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697
1698/**
1699 * scsi_eh_flush_done_q - finish processed commands or retry them.
1700 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001701 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001702void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001704 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001706 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1707 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05001709 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06001710 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1712 " retry cmd: %p\n",
1713 current->comm,
1714 scmd));
1715 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1716 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001717 /*
1718 * If just we got sense for the device (called
1719 * scsi_eh_get_sense), scmd->result is already
1720 * set, do not set DRIVER_TIMEOUT.
1721 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 if (!scmd->result)
1723 scmd->result |= (DRIVER_TIMEOUT << 24);
1724 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1725 " cmd: %p\n",
1726 current->comm, scmd));
1727 scsi_finish_command(scmd);
1728 }
1729 }
1730}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001731EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733/**
1734 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1735 * @shost: Host to unjam.
1736 *
1737 * Notes:
1738 * When we come in here, we *know* that all commands on the bus have
1739 * either completed, failed or timed out. we also know that no further
1740 * commands are being sent to the host, so things are relatively quiet
1741 * and we have freedom to fiddle with things as we wish.
1742 *
1743 * This is only the *default* implementation. it is possible for
1744 * individual drivers to supply their own version of this function, and
1745 * if the maintainer wishes to do this, it is strongly suggested that
1746 * this function be taken as a template and modified. this function
1747 * was designed to correctly handle problems for about 95% of the
1748 * different cases out there, and it should always provide at least a
1749 * reasonable amount of error recovery.
1750 *
1751 * Any command marked 'failed' or 'timeout' must eventually have
1752 * scsi_finish_cmd() called for it. we do all of the retry stuff
1753 * here, so when we restart the host after we return it should have an
1754 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001755 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756static void scsi_unjam_host(struct Scsi_Host *shost)
1757{
1758 unsigned long flags;
1759 LIST_HEAD(eh_work_q);
1760 LIST_HEAD(eh_done_q);
1761
1762 spin_lock_irqsave(shost->host_lock, flags);
1763 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1764 spin_unlock_irqrestore(shost->host_lock, flags);
1765
1766 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1767
1768 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1769 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1770 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1771
1772 scsi_eh_flush_done_q(&eh_done_q);
1773}
1774
1775/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001776 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 * @data: Host for which we are running.
1778 *
1779 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001780 * This is the main error handling loop. This is run as a kernel thread
1781 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001782 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783int scsi_error_handler(void *data)
1784{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001785 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001788 * We use TASK_INTERRUPTIBLE so that the thread is not
1789 * counted against the load average as a running process.
1790 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01001791 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001793 set_current_state(TASK_INTERRUPTIBLE);
1794 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001795 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001796 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001797 SCSI_LOG_ERROR_RECOVERY(1,
1798 printk("Error handler scsi_eh_%d sleeping\n",
1799 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001800 schedule();
1801 set_current_state(TASK_INTERRUPTIBLE);
1802 continue;
1803 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
James Bottomley3ed7a472005-09-19 09:50:04 -05001805 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001806 SCSI_LOG_ERROR_RECOVERY(1,
1807 printk("Error handler scsi_eh_%d waking up\n",
1808 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 /*
1811 * We have a host that is failing for some reason. Figure out
1812 * what we need to do to get it up and online again (if we can).
1813 * If we fail, we end up taking the thing offline.
1814 */
Lin Mingae0751f2011-12-05 09:20:24 +08001815 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
Alan Sternbc4f2402010-06-17 10:41:42 -04001816 SCSI_LOG_ERROR_RECOVERY(1,
1817 printk(KERN_ERR "Error handler scsi_eh_%d "
1818 "unable to autoresume\n",
1819 shost->host_no));
1820 continue;
1821 }
1822
Christoph Hellwig9227c332006-04-01 19:21:04 +02001823 if (shost->transportt->eh_strategy_handler)
1824 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 else
1826 scsi_unjam_host(shost);
1827
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 /*
1829 * Note - if the above fails completely, the action is to take
1830 * individual devices offline and flush the queue of any
1831 * outstanding requests that may have been pending. When we
1832 * restart, we restart any I/O to any other devices on the bus
1833 * which are still online.
1834 */
1835 scsi_restart_operations(shost);
Lin Mingae0751f2011-12-05 09:20:24 +08001836 if (!shost->eh_noresume)
1837 scsi_autopm_put_host(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001838 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001840 __set_current_state(TASK_RUNNING);
1841
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001842 SCSI_LOG_ERROR_RECOVERY(1,
1843 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001844 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 return 0;
1846}
1847
1848/*
1849 * Function: scsi_report_bus_reset()
1850 *
1851 * Purpose: Utility function used by low-level drivers to report that
1852 * they have observed a bus reset on the bus being handled.
1853 *
1854 * Arguments: shost - Host in question
1855 * channel - channel on which reset was observed.
1856 *
1857 * Returns: Nothing
1858 *
1859 * Lock status: Host lock must be held.
1860 *
1861 * Notes: This only needs to be called if the reset is one which
1862 * originates from an unknown location. Resets originated
1863 * by the mid-level itself don't need to call this, but there
1864 * should be no harm.
1865 *
1866 * The main purpose of this is to make sure that a CHECK_CONDITION
1867 * is properly treated.
1868 */
1869void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1870{
1871 struct scsi_device *sdev;
1872
1873 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001874 if (channel == sdev_channel(sdev))
1875 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 }
1877}
1878EXPORT_SYMBOL(scsi_report_bus_reset);
1879
1880/*
1881 * Function: scsi_report_device_reset()
1882 *
1883 * Purpose: Utility function used by low-level drivers to report that
1884 * they have observed a device reset on the device being handled.
1885 *
1886 * Arguments: shost - Host in question
1887 * channel - channel on which reset was observed
1888 * target - target on which reset was observed
1889 *
1890 * Returns: Nothing
1891 *
1892 * Lock status: Host lock must be held
1893 *
1894 * Notes: This only needs to be called if the reset is one which
1895 * originates from an unknown location. Resets originated
1896 * by the mid-level itself don't need to call this, but there
1897 * should be no harm.
1898 *
1899 * The main purpose of this is to make sure that a CHECK_CONDITION
1900 * is properly treated.
1901 */
1902void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1903{
1904 struct scsi_device *sdev;
1905
1906 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001907 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001908 target == sdev_id(sdev))
1909 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 }
1911}
1912EXPORT_SYMBOL(scsi_report_device_reset);
1913
1914static void
1915scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1916{
1917}
1918
1919/*
1920 * Function: scsi_reset_provider
1921 *
1922 * Purpose: Send requested reset to a bus or device at any phase.
1923 *
1924 * Arguments: device - device to send reset to
1925 * flag - reset type (see scsi.h)
1926 *
1927 * Returns: SUCCESS/FAILURE.
1928 *
1929 * Notes: This is used by the SCSI Generic driver to provide
1930 * Bus/Device reset capability.
1931 */
1932int
1933scsi_reset_provider(struct scsi_device *dev, int flag)
1934{
Alan Sternbc4f2402010-06-17 10:41:42 -04001935 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05001936 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001938 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 int rtn;
1940
Alan Sternbc4f2402010-06-17 10:41:42 -04001941 if (scsi_autopm_get_host(shost) < 0)
1942 return FAILED;
1943
1944 scmd = scsi_get_command(dev, GFP_KERNEL);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001945 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001947
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001948 scmd->cmnd = req.cmd;
1949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001951 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 scmd->cmd_len = 0;
1954
1955 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
James Smartd7a1bb02006-03-08 14:50:12 -05001957 spin_lock_irqsave(shost->host_lock, flags);
1958 shost->tmf_in_progress = 1;
1959 spin_unlock_irqrestore(shost->host_lock, flags);
1960
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 switch (flag) {
1962 case SCSI_TRY_RESET_DEVICE:
1963 rtn = scsi_try_bus_device_reset(scmd);
1964 if (rtn == SUCCESS)
1965 break;
1966 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001967 case SCSI_TRY_RESET_TARGET:
1968 rtn = scsi_try_target_reset(scmd);
1969 if (rtn == SUCCESS)
1970 break;
1971 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 case SCSI_TRY_RESET_BUS:
1973 rtn = scsi_try_bus_reset(scmd);
1974 if (rtn == SUCCESS)
1975 break;
1976 /* FALLTHROUGH */
1977 case SCSI_TRY_RESET_HOST:
1978 rtn = scsi_try_host_reset(scmd);
1979 break;
1980 default:
1981 rtn = FAILED;
1982 }
1983
James Smartd7a1bb02006-03-08 14:50:12 -05001984 spin_lock_irqsave(shost->host_lock, flags);
1985 shost->tmf_in_progress = 0;
1986 spin_unlock_irqrestore(shost->host_lock, flags);
1987
1988 /*
1989 * be sure to wake up anyone who was sleeping or had their queue
1990 * suspended while we performed the TMF.
1991 */
1992 SCSI_LOG_ERROR_RECOVERY(3,
1993 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001994 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05001995
1996 wake_up(&shost->host_wait);
1997
1998 scsi_run_host_queues(shost);
1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 scsi_next_command(scmd);
Alan Sternbc4f2402010-06-17 10:41:42 -04002001 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 return rtn;
2003}
2004EXPORT_SYMBOL(scsi_reset_provider);
2005
2006/**
2007 * scsi_normalize_sense - normalize main elements from either fixed or
2008 * descriptor sense data format into a common format.
2009 *
2010 * @sense_buffer: byte array containing sense data returned by device
2011 * @sb_len: number of valid bytes in sense_buffer
2012 * @sshdr: pointer to instance of structure that common
2013 * elements are written to.
2014 *
2015 * Notes:
2016 * The "main elements" from sense data are: response_code, sense_key,
2017 * asc, ascq and additional_length (only for descriptor format).
2018 *
2019 * Typically this function can be called after a device has
2020 * responded to a SCSI command with the CHECK_CONDITION status.
2021 *
2022 * Return value:
2023 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002024 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2026 struct scsi_sense_hdr *sshdr)
2027{
James Bottomley33aa6872005-08-28 11:31:14 -05002028 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 return 0;
2030
2031 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2032
2033 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05002034
2035 if (!scsi_sense_valid(sshdr))
2036 return 0;
2037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 if (sshdr->response_code >= 0x72) {
2039 /*
2040 * descriptor format
2041 */
2042 if (sb_len > 1)
2043 sshdr->sense_key = (sense_buffer[1] & 0xf);
2044 if (sb_len > 2)
2045 sshdr->asc = sense_buffer[2];
2046 if (sb_len > 3)
2047 sshdr->ascq = sense_buffer[3];
2048 if (sb_len > 7)
2049 sshdr->additional_length = sense_buffer[7];
2050 } else {
Jesper Juhl0bf8c862011-03-21 20:47:31 +01002051 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 * fixed format
2053 */
2054 if (sb_len > 2)
2055 sshdr->sense_key = (sense_buffer[2] & 0xf);
2056 if (sb_len > 7) {
2057 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2058 sb_len : (sense_buffer[7] + 8);
2059 if (sb_len > 12)
2060 sshdr->asc = sense_buffer[12];
2061 if (sb_len > 13)
2062 sshdr->ascq = sense_buffer[13];
2063 }
2064 }
2065
2066 return 1;
2067}
2068EXPORT_SYMBOL(scsi_normalize_sense);
2069
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2071 struct scsi_sense_hdr *sshdr)
2072{
2073 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002074 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075}
2076EXPORT_SYMBOL(scsi_command_normalize_sense);
2077
2078/**
Rob Landleyeb448202007-11-03 13:30:39 -05002079 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080 * @sense_buffer: byte array of descriptor format sense data
2081 * @sb_len: number of valid bytes in sense_buffer
2082 * @desc_type: value of descriptor type to find
2083 * (e.g. 0 -> information)
2084 *
2085 * Notes:
2086 * only valid when sense data is in descriptor format
2087 *
2088 * Return value:
2089 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002090 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2092 int desc_type)
2093{
2094 int add_sen_len, add_len, desc_len, k;
2095 const u8 * descp;
2096
2097 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2098 return NULL;
2099 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2100 return NULL;
2101 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2102 add_sen_len : (sb_len - 8);
2103 descp = &sense_buffer[8];
2104 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2105 descp += desc_len;
2106 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2107 desc_len = add_len + 2;
2108 if (descp[0] == desc_type)
2109 return descp;
2110 if (add_len < 0) // short descriptor ??
2111 break;
2112 }
2113 return NULL;
2114}
2115EXPORT_SYMBOL(scsi_sense_desc_find);
2116
2117/**
Rob Landleyeb448202007-11-03 13:30:39 -05002118 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 * @sense_buffer: byte array of sense data
2120 * @sb_len: number of valid bytes in sense_buffer
2121 * @info_out: pointer to 64 integer where 8 or 4 byte information
2122 * field will be placed if found.
2123 *
2124 * Return value:
2125 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002126 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2128 u64 * info_out)
2129{
2130 int j;
2131 const u8 * ucp;
2132 u64 ull;
2133
2134 if (sb_len < 7)
2135 return 0;
2136 switch (sense_buffer[0] & 0x7f) {
2137 case 0x70:
2138 case 0x71:
2139 if (sense_buffer[0] & 0x80) {
2140 *info_out = (sense_buffer[3] << 24) +
2141 (sense_buffer[4] << 16) +
2142 (sense_buffer[5] << 8) + sense_buffer[6];
2143 return 1;
2144 } else
2145 return 0;
2146 case 0x72:
2147 case 0x73:
2148 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2149 0 /* info desc */);
2150 if (ucp && (0xa == ucp[1])) {
2151 ull = 0;
2152 for (j = 0; j < 8; ++j) {
2153 if (j > 0)
2154 ull <<= 8;
2155 ull |= ucp[4 + j];
2156 }
2157 *info_out = ull;
2158 return 1;
2159 } else
2160 return 0;
2161 default:
2162 return 0;
2163 }
2164}
2165EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002166
2167/**
2168 * scsi_build_sense_buffer - build sense data in a buffer
2169 * @desc: Sense format (non zero == descriptor format,
2170 * 0 == fixed format)
2171 * @buf: Where to build sense data
2172 * @key: Sense key
2173 * @asc: Additional sense code
2174 * @ascq: Additional sense code qualifier
2175 *
2176 **/
2177void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2178{
2179 if (desc) {
2180 buf[0] = 0x72; /* descriptor, current */
2181 buf[1] = key;
2182 buf[2] = asc;
2183 buf[3] = ascq;
2184 buf[7] = 0;
2185 } else {
2186 buf[0] = 0x70; /* fixed, current */
2187 buf[2] = key;
2188 buf[7] = 0xa;
2189 buf[12] = asc;
2190 buf[13] = ascq;
2191 }
2192}
2193EXPORT_SYMBOL(scsi_build_sense_buffer);