blob: 386f0c53bea7e17cf2b74fe29af5d1d7176c25b9 [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>
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -050033#include <scsi/scsi_driver.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <scsi/scsi_eh.h>
James Smartc829c392006-03-13 08:28:57 -050035#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <scsi/scsi_host.h>
37#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include "scsi_priv.h"
40#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070041#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Kei Tokunagabf816232010-04-01 20:41:40 +090043#include <trace/events/scsi.h>
44
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#define SENSE_TIMEOUT (10*HZ)
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47/*
48 * These should *probably* be handled by the host itself.
49 * Since it is allowed to sleep, it probably should.
50 */
51#define BUS_RESET_SETTLE_TIME (10)
52#define HOST_RESET_SETTLE_TIME (10)
53
David Jeffery3eef6252011-05-19 14:41:12 -040054static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
55
Linus Torvalds1da177e2005-04-16 15:20:36 -070056/* called with shost->host_lock held */
57void scsi_eh_wakeup(struct Scsi_Host *shost)
58{
59 if (shost->host_busy == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090060 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050061 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 SCSI_LOG_ERROR_RECOVERY(5,
63 printk("Waking error handler thread\n"));
64 }
65}
Tejun Heof8bbfc22006-05-19 21:07:05 +090066
67/**
68 * scsi_schedule_eh - schedule EH for SCSI host
69 * @shost: SCSI host to invoke error handling on.
70 *
71 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080072 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090073void scsi_schedule_eh(struct Scsi_Host *shost)
74{
75 unsigned long flags;
76
77 spin_lock_irqsave(shost->host_lock, flags);
78
79 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
80 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
81 shost->host_eh_scheduled++;
82 scsi_eh_wakeup(shost);
83 }
84
85 spin_unlock_irqrestore(shost->host_lock, flags);
86}
87EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
89/**
90 * scsi_eh_scmd_add - add scsi cmd to error handling.
91 * @scmd: scmd to run eh on.
92 * @eh_flag: optional SCSI_EH flag.
93 *
94 * Return value:
95 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080096 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070097int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
98{
99 struct Scsi_Host *shost = scmd->device->host;
100 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -0500101 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
James Bottomley3ed7a472005-09-19 09:50:04 -0500103 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 return 0;
105
106 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500107 if (scsi_host_set_state(shost, SHOST_RECOVERY))
108 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
109 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110
James Bottomley939647e2005-09-18 15:05:20 -0500111 ret = 1;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200112 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 shost->host_failed++;
115 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500116 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500118 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119}
120
121/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700123 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 *
125 * Notes:
126 * We do not need to lock this. There is the potential for a race
127 * only in that the normal completion handling might run, but if the
128 * normal completion function determines that the timer has already
129 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800130 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700131enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700133 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700134 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100135 struct Scsi_Host *host = scmd->device->host;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500136
Kei Tokunagabf816232010-04-01 20:41:40 +0900137 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 scsi_log_completion(scmd, TIMEOUT_ERROR);
139
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100140 if (host->transportt->eh_timed_out)
141 rtn = host->transportt->eh_timed_out(scmd);
142 else if (host->hostt->eh_timed_out)
143 rtn = host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500144
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500145 scmd->result |= DID_TIME_OUT << 16;
146
Christoph Hellwigfa990782008-11-05 15:03:15 +0100147 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500148 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
Christoph Hellwigfa990782008-11-05 15:03:15 +0100149 rtn = BLK_EH_HANDLED;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700150
Christoph Hellwigfa990782008-11-05 15:03:15 +0100151 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152}
153
154/**
155 * scsi_block_when_processing_errors - Prevent cmds from being queued.
156 * @sdev: Device on which we are performing recovery.
157 *
158 * Description:
159 * We block until the host is out of error recovery, and then check to
160 * see whether the host or the device is offline.
161 *
162 * Return value:
163 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800164 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165int scsi_block_when_processing_errors(struct scsi_device *sdev)
166{
167 int online;
168
James Bottomley939647e2005-09-18 15:05:20 -0500169 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
171 online = scsi_device_online(sdev);
172
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700173 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 online));
175
176 return online;
177}
178EXPORT_SYMBOL(scsi_block_when_processing_errors);
179
180#ifdef CONFIG_SCSI_LOGGING
181/**
182 * scsi_eh_prt_fail_stats - Log info on failures.
183 * @shost: scsi host being recovered.
184 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800185 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
187 struct list_head *work_q)
188{
189 struct scsi_cmnd *scmd;
190 struct scsi_device *sdev;
191 int total_failures = 0;
192 int cmd_failed = 0;
193 int cmd_cancel = 0;
194 int devices_failed = 0;
195
196 shost_for_each_device(sdev, shost) {
197 list_for_each_entry(scmd, work_q, eh_entry) {
198 if (scmd->device == sdev) {
199 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200200 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 ++cmd_cancel;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100202 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 ++cmd_failed;
204 }
205 }
206
207 if (cmd_cancel || cmd_failed) {
208 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500209 sdev_printk(KERN_INFO, sdev,
210 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700211 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500212 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 cmd_cancel = 0;
214 cmd_failed = 0;
215 ++devices_failed;
216 }
217 }
218
219 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
220 " devices require eh work\n",
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100221 total_failures, devices_failed));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223#endif
224
225/**
226 * scsi_check_sense - Examine scsi cmd sense
227 * @scmd: Cmd to have sense checked.
228 *
229 * Return value:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100230 * SUCCESS or FAILED or NEEDS_RETRY or TARGET_ERROR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 *
232 * Notes:
233 * When a deferred error is detected the current command has
234 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800235 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236static int scsi_check_sense(struct scsi_cmnd *scmd)
237{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700238 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 struct scsi_sense_hdr sshdr;
240
241 if (! scsi_command_normalize_sense(scmd, &sshdr))
242 return FAILED; /* no valid sense data */
243
244 if (scsi_sense_is_deferred(&sshdr))
245 return NEEDS_RETRY;
246
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700247 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
248 sdev->scsi_dh_data->scsi_dh->check_sense) {
249 int rc;
250
251 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
252 if (rc != SCSI_RETURN_NOT_HANDLED)
253 return rc;
254 /* handler does not care. Drop down to default handling */
255 }
256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 /*
258 * Previous logic looked for FILEMARK, EOM or ILI which are
259 * mainly associated with tapes and returned SUCCESS.
260 */
261 if (sshdr.response_code == 0x70) {
262 /* fixed format */
263 if (scmd->sense_buffer[2] & 0xe0)
264 return SUCCESS;
265 } else {
266 /*
267 * descriptor format: look for "stream commands sense data
268 * descriptor" (see SSC-3). Assume single sense data
269 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
270 */
271 if ((sshdr.additional_length > 3) &&
272 (scmd->sense_buffer[8] == 0x4) &&
273 (scmd->sense_buffer[11] & 0xe0))
274 return SUCCESS;
275 }
276
277 switch (sshdr.sense_key) {
278 case NO_SENSE:
279 return SUCCESS;
280 case RECOVERED_ERROR:
281 return /* soft_error */ SUCCESS;
282
283 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400284 if (sshdr.asc == 0x10) /* DIF */
285 return SUCCESS;
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return NEEDS_RETRY;
288 case NOT_READY:
289 case UNIT_ATTENTION:
290 /*
291 * if we are expecting a cc/ua because of a bus reset that we
292 * performed, treat this just as a retry. otherwise this is
293 * information that we should pass up to the upper-level driver
294 * so that we can deal with it there.
295 */
296 if (scmd->device->expecting_cc_ua) {
TARUISI Hiroakidfcf7772011-08-11 20:25:20 +0900297 /*
298 * Because some device does not queue unit
299 * attentions correctly, we carefully check
300 * additional sense code and qualifier so as
301 * not to squash media change unit attention.
302 */
303 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
304 scmd->device->expecting_cc_ua = 0;
305 return NEEDS_RETRY;
306 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 }
308 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100309 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * should retry.
311 */
312 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
313 return NEEDS_RETRY;
314 /*
315 * if the device is not started, we need to wake
316 * the error handler to start the motor
317 */
318 if (scmd->device->allow_restart &&
319 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
320 return FAILED;
James Bottomley77a42292010-05-04 16:51:40 -0400321
Mike Christie6e499492010-04-25 07:03:57 -0500322 if (sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
323 scmd_printk(KERN_WARNING, scmd,
324 "Warning! Received an indication that the "
325 "LUN assignments on this target have "
326 "changed. The Linux SCSI layer does not "
327 "automatically remap LUN assignments.\n");
328 else if (sshdr.asc == 0x3f)
329 scmd_printk(KERN_WARNING, scmd,
330 "Warning! Received an indication that the "
331 "operating parameters on this target have "
332 "changed. The Linux SCSI layer does not "
333 "automatically adjust these parameters.\n");
334
Shyam Iyerdeb1cb62011-02-26 01:59:44 -0500335 if (sshdr.asc == 0x38 && sshdr.ascq == 0x07)
336 scmd_printk(KERN_WARNING, scmd,
337 "Warning! Received an indication that the "
338 "LUN reached a thin provisioning soft "
339 "threshold.\n");
340
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100341 /*
342 * Pass the UA upwards for a determination in the completion
343 * functions.
344 */
345 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100347 /* these are not supported */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 case COPY_ABORTED:
349 case VOLUME_OVERFLOW:
350 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100351 case BLANK_CHECK:
352 case DATA_PROTECT:
353 return TARGET_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800356 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
357 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
358 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100359 return TARGET_ERROR;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800360 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 return NEEDS_RETRY;
362
363 case HARDWARE_ERROR:
364 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700365 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 else
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100367 return TARGET_ERROR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
369 case ILLEGAL_REQUEST:
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500370 if (sshdr.asc == 0x20 || /* Invalid command operation code */
371 sshdr.asc == 0x21 || /* Logical block address out of range */
372 sshdr.asc == 0x24 || /* Invalid field in cdb */
373 sshdr.asc == 0x26) { /* Parameter value invalid */
374 return TARGET_ERROR;
375 }
376 return SUCCESS;
377
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 default:
379 return SUCCESS;
380 }
381}
382
Vasu Dev4a840672009-10-22 15:46:33 -0700383static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
384{
385 struct scsi_host_template *sht = sdev->host->hostt;
386 struct scsi_device *tmp_sdev;
387
388 if (!sht->change_queue_depth ||
389 sdev->queue_depth >= sdev->max_queue_depth)
390 return;
391
392 if (time_before(jiffies,
393 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
394 return;
395
396 if (time_before(jiffies,
397 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
398 return;
399
400 /*
401 * Walk all devices of a target and do
402 * ramp up on them.
403 */
404 shost_for_each_device(tmp_sdev, sdev->host) {
405 if (tmp_sdev->channel != sdev->channel ||
406 tmp_sdev->id != sdev->id ||
407 tmp_sdev->queue_depth == sdev->max_queue_depth)
408 continue;
409 /*
410 * call back into LLD to increase queue_depth by one
411 * with ramp up reason code.
412 */
413 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
414 SCSI_QDEPTH_RAMP_UP);
415 sdev->last_queue_ramp_up = jiffies;
416 }
417}
418
Mike Christie42a6a912009-10-15 17:46:44 -0700419static void scsi_handle_queue_full(struct scsi_device *sdev)
420{
421 struct scsi_host_template *sht = sdev->host->hostt;
422 struct scsi_device *tmp_sdev;
423
424 if (!sht->change_queue_depth)
425 return;
426
427 shost_for_each_device(tmp_sdev, sdev->host) {
428 if (tmp_sdev->channel != sdev->channel ||
429 tmp_sdev->id != sdev->id)
430 continue;
431 /*
432 * We do not know the number of commands that were at
433 * the device when we got the queue full so we start
434 * from the highest possible value and work our way down.
435 */
436 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
437 SCSI_QDEPTH_QFULL);
438 }
439}
440
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441/**
442 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
443 * @scmd: SCSI cmd to examine.
444 *
445 * Notes:
446 * This is *only* called when we are examining the status of commands
447 * queued during error recovery. the main difference here is that we
448 * don't allow for the possibility of retries here, and we are a lot
449 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800450 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
452{
453 /*
454 * first check the host byte, to see if there is anything in there
455 * that would indicate what we need to do.
456 */
457 if (host_byte(scmd->result) == DID_RESET) {
458 /*
459 * rats. we are already in the error handler, so we now
460 * get to try and figure out what to do next. if the sense
461 * is valid, we have a pretty good idea of what to do.
462 * if not, we mark it as FAILED.
463 */
464 return scsi_check_sense(scmd);
465 }
466 if (host_byte(scmd->result) != DID_OK)
467 return FAILED;
468
469 /*
470 * next, check the message byte.
471 */
472 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
473 return FAILED;
474
475 /*
476 * now, check the status byte to see if this indicates
477 * anything special.
478 */
479 switch (status_byte(scmd->result)) {
480 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700481 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 case COMMAND_TERMINATED:
483 return SUCCESS;
484 case CHECK_CONDITION:
485 return scsi_check_sense(scmd);
486 case CONDITION_GOOD:
487 case INTERMEDIATE_GOOD:
488 case INTERMEDIATE_C_GOOD:
489 /*
490 * who knows? FIXME(eric)
491 */
492 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500493 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500494 if (scmd->cmnd[0] == TEST_UNIT_READY)
495 /* it is a success, we probed the device and
496 * found it */
497 return SUCCESS;
498 /* otherwise, we failed to send the command */
499 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700501 scsi_handle_queue_full(scmd->device);
502 /* fall through */
503 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200504 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 default:
506 return FAILED;
507 }
508 return FAILED;
509}
510
511/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 * scsi_eh_done - Completion function for error handling.
513 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800514 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515static void scsi_eh_done(struct scsi_cmnd *scmd)
516{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100517 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600518
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100519 SCSI_LOG_ERROR_RECOVERY(3,
520 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700521 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600522
523 eh_action = scmd->device->host->eh_action;
524 if (eh_action)
525 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
528/**
Brian King292148f2007-01-30 17:51:17 -0600529 * scsi_try_host_reset - ask host adapter to reset itself
530 * @scmd: SCSI cmd to send hsot reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800531 */
Brian King292148f2007-01-30 17:51:17 -0600532static int scsi_try_host_reset(struct scsi_cmnd *scmd)
533{
534 unsigned long flags;
535 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100536 struct Scsi_Host *host = scmd->device->host;
537 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600538
539 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700540 __func__));
Brian King292148f2007-01-30 17:51:17 -0600541
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100542 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600543 return FAILED;
544
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100545 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600546
547 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100548 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600549 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100550 spin_lock_irqsave(host->host_lock, flags);
551 scsi_report_bus_reset(host, scmd_channel(scmd));
552 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600553 }
554
555 return rtn;
556}
557
558/**
559 * scsi_try_bus_reset - ask host to perform a bus reset
560 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800561 */
Brian King292148f2007-01-30 17:51:17 -0600562static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
563{
564 unsigned long flags;
565 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100566 struct Scsi_Host *host = scmd->device->host;
567 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600568
569 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700570 __func__));
Brian King292148f2007-01-30 17:51:17 -0600571
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100572 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600573 return FAILED;
574
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100575 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600576
577 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100578 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600579 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100580 spin_lock_irqsave(host->host_lock, flags);
581 scsi_report_bus_reset(host, scmd_channel(scmd));
582 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600583 }
584
585 return rtn;
586}
587
Mike Christie30bd7df2008-02-29 18:25:19 -0600588static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
589{
590 sdev->was_reset = 1;
591 sdev->expecting_cc_ua = 1;
592}
593
594/**
595 * scsi_try_target_reset - Ask host to perform a target reset
596 * @scmd: SCSI cmd used to send a target reset
597 *
598 * Notes:
599 * There is no timeout for this operation. if this operation is
600 * unreliable for a given host, then the host itself needs to put a
601 * timer on it, and set the host back to a consistent state prior to
602 * returning.
603 */
604static int scsi_try_target_reset(struct scsi_cmnd *scmd)
605{
606 unsigned long flags;
607 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100608 struct Scsi_Host *host = scmd->device->host;
609 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600610
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100611 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600612 return FAILED;
613
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100614 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600615 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100616 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600617 __starget_for_each_device(scsi_target(scmd->device), NULL,
618 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100619 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600620 }
621
622 return rtn;
623}
624
Brian King292148f2007-01-30 17:51:17 -0600625/**
626 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
627 * @scmd: SCSI cmd used to send BDR
628 *
629 * Notes:
630 * There is no timeout for this operation. if this operation is
631 * unreliable for a given host, then the host itself needs to put a
632 * timer on it, and set the host back to a consistent state prior to
633 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800634 */
Brian King292148f2007-01-30 17:51:17 -0600635static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
636{
637 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100638 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600639
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100640 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600641 return FAILED;
642
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100643 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600644 if (rtn == SUCCESS)
645 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600646 return rtn;
647}
648
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100649static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600650{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100651 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600652 return FAILED;
653
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100654 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600655}
656
Brian King292148f2007-01-30 17:51:17 -0600657static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
658{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100659 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600660 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600661 if (scsi_try_target_reset(scmd) != SUCCESS)
662 if (scsi_try_bus_reset(scmd) != SUCCESS)
663 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600664}
665
666/**
Boaz Harroshe1c23462007-10-08 16:36:45 +0200667 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recory
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100668 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200669 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200670 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300671 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200672 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100673 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200674 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200675 * as part of the error recovery process. If @sense_bytes is 0 the command
676 * sent must be one that does not transfer any data. If @sense_bytes != 0
677 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
678 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800679 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200680void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
681 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
f59114b2005-04-17 15:00:23 -0500683 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Christoph Hellwig631c2282006-07-08 20:42:15 +0200685 /*
686 * We need saved copies of a number of fields - this is because
687 * error handling may need to overwrite these with different values
688 * to run different commands, and once error handling is complete,
689 * we will need to restore these values prior to running the actual
690 * command.
691 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200692 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300693 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200694 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200695 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200696 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200697 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400698 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400699 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200700
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400701 scmd->prot_op = SCSI_PROT_NORMAL;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300702 scmd->cmnd = ses->eh_cmnd;
703 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200704 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200705 scmd->request->next_rq = NULL;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200706
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200707 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200708 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
709 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200710 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200711 scmd->sdb.length);
712 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200713 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200714 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200715 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200716 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200717 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200718 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200719 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200720 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300721 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200722 memcpy(scmd->cmnd, cmnd, cmnd_size);
723 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
724 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200725 }
726
727 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200728
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200729 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500731 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Christoph Hellwig631c2282006-07-08 20:42:15 +0200733 /*
734 * Zero the sense buffer. The scsi spec mandates that any
735 * untransferred sense data should be interpreted as being zero.
736 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900737 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200738}
739EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200740
Boaz Harroshe1c23462007-10-08 16:36:45 +0200741/**
742 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recory
743 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200744 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200745 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200746 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800747 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200748void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
749{
750 /*
751 * Restore original data
752 */
753 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300754 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200755 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200756 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200757 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200758 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400759 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400760 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200761}
762EXPORT_SYMBOL(scsi_eh_restore_cmnd);
763
764/**
765 * scsi_send_eh_cmnd - submit a scsi command as part of error recory
766 * @scmd: SCSI command structure to hijack
767 * @cmnd: CDB to send
768 * @cmnd_size: size in bytes of @cmnd
769 * @timeout: timeout for this request
770 * @sense_bytes: size of sense data to copy or 0
771 *
772 * This function is used to send a scsi command down to a target device
773 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
774 *
775 * Return value:
776 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800777 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200778static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
779 int cmnd_size, int timeout, unsigned sense_bytes)
780{
781 struct scsi_device *sdev = scmd->device;
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500782 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200783 struct Scsi_Host *shost = sdev->host;
784 DECLARE_COMPLETION_ONSTACK(done);
785 unsigned long timeleft;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200786 struct scsi_eh_save ses;
787 int rtn;
788
789 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100790 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -0500793 scmd->scsi_done = scsi_eh_done;
794 shost->hostt->queuecommand(shost, scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100796 timeleft = wait_for_completion_timeout(&done, timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
f59114b2005-04-17 15:00:23 -0500798 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100800 scsi_log_completion(scmd, SUCCESS);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100802 SCSI_LOG_ERROR_RECOVERY(3,
803 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700804 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
806 /*
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100807 * If there is time left scsi_eh_done got called, and we will
808 * examine the actual status codes to see whether the command
809 * actually did complete normally, else tell the host to forget
810 * about this command.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100812 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 rtn = scsi_eh_completed_normally(scmd);
814 SCSI_LOG_ERROR_RECOVERY(3,
815 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700816 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100817
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 switch (rtn) {
819 case SUCCESS:
820 case NEEDS_RETRY:
821 case FAILED:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100822 case TARGET_ERROR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +0200824 case ADD_TO_MLQUEUE:
825 rtn = NEEDS_RETRY;
826 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 default:
828 rtn = FAILED;
829 break;
830 }
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100831 } else {
Brian King292148f2007-01-30 17:51:17 -0600832 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100833 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 }
835
Boaz Harroshe1c23462007-10-08 16:36:45 +0200836 scsi_eh_restore_cmnd(scmd, &ses);
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500837
Martin K. Petersen919f7972012-04-14 23:01:28 -0400838 if (sdrv && sdrv->eh_action)
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500839 rtn = sdrv->eh_action(scmd, cmnd, cmnd_size, rtn);
840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 return rtn;
842}
843
844/**
845 * scsi_request_sense - Request sense data from a particular target.
846 * @scmd: SCSI cmd for request sense.
847 *
848 * Notes:
849 * Some hosts automatically obtain this information, others require
850 * that we obtain it on our own. This function will *not* return until
851 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800852 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853static int scsi_request_sense(struct scsi_cmnd *scmd)
854{
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200855 return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856}
857
858/**
859 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
860 * @scmd: Original SCSI cmd that eh has finished.
861 * @done_q: Queue for processed commands.
862 *
863 * Notes:
864 * We don't want to use the normal command completion while we are are
865 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -0500866 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 * keep a list of pending commands for final completion, and once we
868 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800869 */
Tejun Heo041c5fc2006-01-23 13:09:36 +0900870void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871{
872 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200873 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 list_move_tail(&scmd->eh_entry, done_q);
875}
Tejun Heo041c5fc2006-01-23 13:09:36 +0900876EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
878/**
879 * scsi_eh_get_sense - Get device sense data.
880 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -0500881 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 *
883 * Description:
884 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100885 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 *
887 * Notes:
888 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -0500889 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 * it down before we request it.
891 *
892 * All drivers should request sense information internally these days,
893 * so for now all I have to say is tough noogies if you end up in here.
894 *
895 * XXX: Long term this code should go away, but that needs an audit of
896 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800897 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800898int scsi_eh_get_sense(struct list_head *work_q,
899 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200901 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 int rtn;
903
Christoph Hellwig937abeaa2005-06-19 13:43:56 +0200904 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200905 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 SCSI_SENSE_VALID(scmd))
907 continue;
908
Jeff Garzik3bf743e2005-10-24 18:04:06 -0400909 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
910 "%s: requesting sense\n",
911 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 rtn = scsi_request_sense(scmd);
913 if (rtn != SUCCESS)
914 continue;
915
916 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
917 " result %x\n", scmd,
918 scmd->result));
919 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
920
921 rtn = scsi_decide_disposition(scmd);
922
923 /*
924 * if the result was normal, then just pass it along to the
925 * upper level.
926 */
927 if (rtn == SUCCESS)
928 /* we don't want this command reissued, just
929 * finished with the sense data, so set
930 * retries to the max allowed to ensure it
931 * won't get reissued */
932 scmd->retries = scmd->allowed;
933 else if (rtn != NEEDS_RETRY)
934 continue;
935
936 scsi_eh_finish_cmd(scmd, done_q);
937 }
938
939 return list_empty(work_q);
940}
Darrick J. Wongdca84e42007-01-26 14:08:49 -0800941EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
943/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -0500945 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 *
947 * Return value:
948 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800949 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950static int scsi_eh_tur(struct scsi_cmnd *scmd)
951{
952 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
953 int retry_cnt = 1, rtn;
954
955retry_tur:
Mike Christie0db99e32006-08-26 03:00:22 -0400956 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700959 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +0200960
961 switch (rtn) {
962 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 if (retry_cnt--)
964 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200965 /*FALLTHRU*/
966 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -0500967 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200968 default:
969 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -0500970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
972
973/**
David Jeffery3eef6252011-05-19 14:41:12 -0400974 * scsi_eh_test_devices - check if devices are responding from error recovery.
975 * @cmd_list: scsi commands in error recovery.
976 * @work_q: queue for commands which still need more error recovery
977 * @done_q: queue for commands which are finished
978 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
979 *
980 * Decription:
981 * Tests if devices are in a working state. Commands to devices now in
982 * a working state are sent to the done_q while commands to devices which
983 * are still failing to respond are returned to the work_q for more
984 * processing.
985 **/
986static int scsi_eh_test_devices(struct list_head *cmd_list,
987 struct list_head *work_q,
988 struct list_head *done_q, int try_stu)
989{
990 struct scsi_cmnd *scmd, *next;
991 struct scsi_device *sdev;
992 int finish_cmds;
993
994 while (!list_empty(cmd_list)) {
995 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
996 sdev = scmd->device;
997
998 finish_cmds = !scsi_device_online(scmd->device) ||
999 (try_stu && !scsi_eh_try_stu(scmd) &&
1000 !scsi_eh_tur(scmd)) ||
1001 !scsi_eh_tur(scmd);
1002
1003 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1004 if (scmd->device == sdev) {
1005 if (finish_cmds)
1006 scsi_eh_finish_cmd(scmd, done_q);
1007 else
1008 list_move_tail(&scmd->eh_entry, work_q);
1009 }
1010 }
1011 return list_empty(work_q);
1012}
1013
1014
1015/**
Rob Landleyeb448202007-11-03 13:30:39 -05001016 * scsi_eh_abort_cmds - abort pending commands.
1017 * @work_q: &list_head for pending commands.
1018 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 *
1020 * Decription:
1021 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -05001022 * running command. This only works out to be the case if we have one
1023 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 * no sense to try and abort the command, since as far as the shost
1025 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001026 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027static int scsi_eh_abort_cmds(struct list_head *work_q,
1028 struct list_head *done_q)
1029{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001030 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001031 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 int rtn;
1033
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001034 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001035 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 continue;
1037 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1038 "0x%p\n", current->comm,
1039 scmd));
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001040 rtn = scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001041 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001042 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
David Jeffery3eef6252011-05-19 14:41:12 -04001043 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 scsi_eh_finish_cmd(scmd, done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001045 else
1046 list_move_tail(&scmd->eh_entry, &check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 } else
1048 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1049 " cmd failed:"
1050 "0x%p\n",
1051 current->comm,
1052 scmd));
1053 }
1054
David Jeffery3eef6252011-05-19 14:41:12 -04001055 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056}
1057
1058/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001060 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 *
1062 * Return value:
1063 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001064 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1066{
1067 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
Christoph Hellwig631c2282006-07-08 20:42:15 +02001069 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001070 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
Brian Kinged773e62007-03-29 15:25:52 -05001072 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001073 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001074
Christoph Hellwig631c2282006-07-08 20:42:15 +02001075 if (rtn == SUCCESS)
1076 return 0;
1077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return 1;
1080}
1081
1082 /**
1083 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001084 * @shost: &scsi host being recovered.
1085 * @work_q: &list_head for pending commands.
1086 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 *
1088 * Notes:
1089 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001090 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001091 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092static int scsi_eh_stu(struct Scsi_Host *shost,
1093 struct list_head *work_q,
1094 struct list_head *done_q)
1095{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001096 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 struct scsi_device *sdev;
1098
1099 shost_for_each_device(sdev, shost) {
1100 stu_scmd = NULL;
1101 list_for_each_entry(scmd, work_q, eh_entry)
1102 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1103 scsi_check_sense(scmd) == FAILED ) {
1104 stu_scmd = scmd;
1105 break;
1106 }
1107
1108 if (!stu_scmd)
1109 continue;
1110
1111 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1112 " 0x%p\n", current->comm, sdev));
1113
1114 if (!scsi_eh_try_stu(stu_scmd)) {
1115 if (!scsi_device_online(sdev) ||
1116 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001117 list_for_each_entry_safe(scmd, next,
1118 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (scmd->device == sdev)
1120 scsi_eh_finish_cmd(scmd, done_q);
1121 }
1122 }
1123 } else {
1124 SCSI_LOG_ERROR_RECOVERY(3,
1125 printk("%s: START_UNIT failed to sdev:"
1126 " 0x%p\n", current->comm, sdev));
1127 }
1128 }
1129
1130 return list_empty(work_q);
1131}
1132
1133
1134/**
1135 * scsi_eh_bus_device_reset - send bdr if needed
1136 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001137 * @work_q: &list_head for pending commands.
1138 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 *
1140 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001141 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 * devices that are jammed or not - if we have multiple devices, it
1143 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001144 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001145 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1147 struct list_head *work_q,
1148 struct list_head *done_q)
1149{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001150 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 struct scsi_device *sdev;
1152 int rtn;
1153
1154 shost_for_each_device(sdev, shost) {
1155 bdr_scmd = NULL;
1156 list_for_each_entry(scmd, work_q, eh_entry)
1157 if (scmd->device == sdev) {
1158 bdr_scmd = scmd;
1159 break;
1160 }
1161
1162 if (!bdr_scmd)
1163 continue;
1164
1165 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1166 " 0x%p\n", current->comm,
1167 sdev));
1168 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001169 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001171 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001173 list_for_each_entry_safe(scmd, next,
1174 work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 if (scmd->device == sdev)
1176 scsi_eh_finish_cmd(scmd,
1177 done_q);
1178 }
1179 }
1180 } else {
1181 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1182 " failed sdev:"
1183 "0x%p\n",
1184 current->comm,
1185 sdev));
1186 }
1187 }
1188
1189 return list_empty(work_q);
1190}
1191
1192/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001193 * scsi_eh_target_reset - send target reset if needed
1194 * @shost: scsi host being recovered.
1195 * @work_q: &list_head for pending commands.
1196 * @done_q: &list_head for processed commands.
1197 *
1198 * Notes:
1199 * Try a target reset.
1200 */
1201static int scsi_eh_target_reset(struct Scsi_Host *shost,
1202 struct list_head *work_q,
1203 struct list_head *done_q)
1204{
James Bottomley98db5192010-10-25 15:53:41 -05001205 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001206 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001207
James Bottomley98db5192010-10-25 15:53:41 -05001208 list_splice_init(work_q, &tmp_list);
1209
1210 while (!list_empty(&tmp_list)) {
1211 struct scsi_cmnd *next, *scmd;
1212 int rtn;
1213 unsigned int id;
1214
1215 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1216 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001217
1218 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1219 "to target %d\n",
1220 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001221 rtn = scsi_try_target_reset(scmd);
1222 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Mike Christie30bd7df2008-02-29 18:25:19 -06001223 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1224 " failed target: "
1225 "%d\n",
1226 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001227 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1228 if (scmd_id(scmd) != id)
1229 continue;
1230
David Jeffery3eef6252011-05-19 14:41:12 -04001231 if (rtn == SUCCESS)
1232 list_move_tail(&scmd->eh_entry, &check_list);
1233 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001234 scsi_eh_finish_cmd(scmd, done_q);
1235 else
1236 /* push back on work queue for further processing */
1237 list_move(&scmd->eh_entry, work_q);
1238 }
1239 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001240
David Jeffery3eef6252011-05-19 14:41:12 -04001241 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001242}
1243
1244/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001245 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001246 * @shost: &scsi host being recovered.
1247 * @work_q: &list_head for pending commands.
1248 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001249 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1251 struct list_head *work_q,
1252 struct list_head *done_q)
1253{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001254 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001255 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 unsigned int channel;
1257 int rtn;
1258
1259 /*
1260 * we really want to loop over the various channels, and do this on
1261 * a channel by channel basis. we should also check to see if any
1262 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001263 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 */
1265
1266 for (channel = 0; channel <= shost->max_channel; channel++) {
1267 chan_scmd = NULL;
1268 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001269 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 chan_scmd = scmd;
1271 break;
1272 /*
1273 * FIXME add back in some support for
1274 * soft_reset devices.
1275 */
1276 }
1277 }
1278
1279 if (!chan_scmd)
1280 continue;
1281 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1282 " %d\n", current->comm,
1283 channel));
1284 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001285 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001286 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001287 if (channel == scmd_channel(scmd)) {
1288 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 scsi_eh_finish_cmd(scmd,
1290 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001291 else
1292 list_move_tail(&scmd->eh_entry,
1293 &check_list);
1294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 }
1296 } else {
1297 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1298 " failed chan: %d\n",
1299 current->comm,
1300 channel));
1301 }
1302 }
David Jeffery3eef6252011-05-19 14:41:12 -04001303 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304}
1305
1306/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001307 * scsi_eh_host_reset - send a host reset
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 * @work_q: list_head for processed commands.
1309 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001310 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311static int scsi_eh_host_reset(struct list_head *work_q,
1312 struct list_head *done_q)
1313{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001314 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001315 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 if (!list_empty(work_q)) {
1319 scmd = list_entry(work_q->next,
1320 struct scsi_cmnd, eh_entry);
1321
1322 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1323 , current->comm));
1324
1325 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001326 if (rtn == SUCCESS) {
1327 list_splice_init(work_q, &check_list);
1328 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001329 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 scsi_eh_finish_cmd(scmd, done_q);
1331 }
1332 } else {
1333 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1334 " failed\n",
1335 current->comm));
1336 }
1337 }
David Jeffery3eef6252011-05-19 14:41:12 -04001338 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339}
1340
1341/**
1342 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1343 * @work_q: list_head for processed commands.
1344 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001345 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346static void scsi_eh_offline_sdevs(struct list_head *work_q,
1347 struct list_head *done_q)
1348{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001349 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001351 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001352 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1353 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001355 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 /*
1357 * FIXME: Handle lost cmds.
1358 */
1359 }
1360 scsi_eh_finish_cmd(scmd, done_q);
1361 }
1362 return;
1363}
1364
1365/**
Mike Christie4a274462008-08-19 18:45:31 -05001366 * scsi_noretry_cmd - determinte if command should be failed fast
1367 * @scmd: SCSI cmd to examine.
1368 */
1369int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1370{
1371 switch (host_byte(scmd->result)) {
1372 case DID_OK:
1373 break;
1374 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001375 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001376 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001377 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001378 case DID_ERROR:
1379 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1380 status_byte(scmd->result) == RESERVATION_CONFLICT)
1381 return 0;
1382 /* fall through */
1383 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001384 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001385 }
1386
1387 switch (status_byte(scmd->result)) {
1388 case CHECK_CONDITION:
1389 /*
1390 * assume caller has checked sense and determinted
1391 * the check condition was retryable.
1392 */
FUJITA Tomonorie96f6ab2010-07-09 09:38:26 +09001393 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1394 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1395 return 1;
Mike Christie4a274462008-08-19 18:45:31 -05001396 }
1397
1398 return 0;
1399}
1400
1401/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1403 * @scmd: SCSI cmd to examine.
1404 *
1405 * Notes:
1406 * This is *only* called when we are examining the status after sending
1407 * out the actual data command. any commands that are queued for error
1408 * recovery (e.g. test_unit_ready) do *not* come through here.
1409 *
1410 * When this routine returns failed, it means the error handler thread
1411 * is woken. In cases where the error code indicates an error that
1412 * doesn't require the error handler read (i.e. we don't need to
1413 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001414 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415int scsi_decide_disposition(struct scsi_cmnd *scmd)
1416{
1417 int rtn;
1418
1419 /*
1420 * if the device is offline, then we clearly just pass the result back
1421 * up to the top level.
1422 */
1423 if (!scsi_device_online(scmd->device)) {
1424 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1425 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001426 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 return SUCCESS;
1428 }
1429
1430 /*
1431 * first check the host byte, to see if there is anything in there
1432 * that would indicate what we need to do.
1433 */
1434 switch (host_byte(scmd->result)) {
1435 case DID_PASSTHROUGH:
1436 /*
1437 * no matter what, pass this through to the upper layer.
1438 * nuke this special code so that it looks like we are saying
1439 * did_ok.
1440 */
1441 scmd->result &= 0xff00ffff;
1442 return SUCCESS;
1443 case DID_OK:
1444 /*
1445 * looks good. drop through, and check the next byte.
1446 */
1447 break;
1448 case DID_NO_CONNECT:
1449 case DID_BAD_TARGET:
1450 case DID_ABORT:
1451 /*
1452 * note - this means that we just report the status back
1453 * to the top level driver, not that we actually think
1454 * that it indicates SUCCESS.
1455 */
1456 return SUCCESS;
1457 /*
1458 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001459 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 * in order to avoid endless loops (db)
1461 *
1462 * actually this is a bug in this function here. we should
1463 * be mindful of the maximum number of retries specified
1464 * and not get stuck in a loop.
1465 */
1466 case DID_SOFT_ERROR:
1467 goto maybe_retry;
1468 case DID_IMM_RETRY:
1469 return NEEDS_RETRY;
1470
bf341912005-04-12 17:49:09 -05001471 case DID_REQUEUE:
1472 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001473 case DID_TRANSPORT_DISRUPTED:
1474 /*
1475 * LLD/transport was disrupted during processing of the IO.
1476 * The transport class is now blocked/blocking,
1477 * and the transport will decide what to do with the IO
Mike Christie939c2282008-11-04 19:47:19 -06001478 * based on its timers and recovery capablilities if
1479 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001480 */
Mike Christie939c2282008-11-04 19:47:19 -06001481 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001482 case DID_TRANSPORT_FAILFAST:
1483 /*
1484 * The transport decided to failfast the IO (most likely
1485 * the fast io fail tmo fired), so send IO directly upwards.
1486 */
1487 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 case DID_ERROR:
1489 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1490 status_byte(scmd->result) == RESERVATION_CONFLICT)
1491 /*
1492 * execute reservation conflict processing code
1493 * lower down
1494 */
1495 break;
1496 /* fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 case DID_BUS_BUSY:
1498 case DID_PARITY:
1499 goto maybe_retry;
1500 case DID_TIME_OUT:
1501 /*
1502 * when we scan the bus, we get timeout messages for
1503 * these commands if there is no device available.
1504 * other hosts report did_no_connect for the same thing.
1505 */
1506 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1507 scmd->cmnd[0] == INQUIRY)) {
1508 return SUCCESS;
1509 } else {
1510 return FAILED;
1511 }
1512 case DID_RESET:
1513 return SUCCESS;
1514 default:
1515 return FAILED;
1516 }
1517
1518 /*
1519 * next, check the message byte.
1520 */
1521 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1522 return FAILED;
1523
1524 /*
1525 * check the status byte to see if this indicates anything special.
1526 */
1527 switch (status_byte(scmd->result)) {
1528 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001529 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530 /*
1531 * the case of trying to send too many commands to a
1532 * tagged queueing device.
1533 */
1534 case BUSY:
1535 /*
1536 * device can't talk to us at the moment. Should only
1537 * occur (SAM-3) when the task queue is empty, so will cause
1538 * the empty queue handling to trigger a stall in the
1539 * device.
1540 */
1541 return ADD_TO_MLQUEUE;
1542 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -07001543 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001546 case TASK_ABORTED:
1547 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 case CHECK_CONDITION:
1549 rtn = scsi_check_sense(scmd);
1550 if (rtn == NEEDS_RETRY)
1551 goto maybe_retry;
Hannes Reinecke63583cc2011-01-18 10:13:11 +01001552 else if (rtn == TARGET_ERROR) {
1553 /*
1554 * Need to modify host byte to signal a
1555 * permanent target failure
1556 */
Moger, Babu2082ebc2012-01-24 20:38:46 +00001557 set_host_byte(scmd, DID_TARGET_FAILURE);
Hannes Reinecke63583cc2011-01-18 10:13:11 +01001558 rtn = SUCCESS;
1559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 /* if rtn == FAILED, we have no sense information;
1561 * returning FAILED will wake the error handler thread
1562 * to collect the sense and redo the decide
1563 * disposition */
1564 return rtn;
1565 case CONDITION_GOOD:
1566 case INTERMEDIATE_GOOD:
1567 case INTERMEDIATE_C_GOOD:
1568 case ACA_ACTIVE:
1569 /*
1570 * who knows? FIXME(eric)
1571 */
1572 return SUCCESS;
1573
1574 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001575 sdev_printk(KERN_INFO, scmd->device,
1576 "reservation conflict\n");
Moger, Babu2082ebc2012-01-24 20:38:46 +00001577 set_host_byte(scmd, DID_NEXUS_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 return SUCCESS; /* causes immediate i/o error */
1579 default:
1580 return FAILED;
1581 }
1582 return FAILED;
1583
1584 maybe_retry:
1585
1586 /* we requeue for retry because the error was retryable, and
1587 * the request was not marked fast fail. Note that above,
1588 * even if the request is marked fast fail, we still requeue
1589 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001590 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001591 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 return NEEDS_RETRY;
1593 } else {
1594 /*
1595 * no more retries - report this one back to upper level.
1596 */
1597 return SUCCESS;
1598 }
1599}
1600
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001601static void eh_lock_door_done(struct request *req, int uptodate)
1602{
1603 __blk_put_request(req->q, req);
1604}
1605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 * scsi_eh_lock_door - Prevent medium removal for the specified device
1608 * @sdev: SCSI device to prevent medium removal
1609 *
1610 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001611 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 *
1613 * Notes:
1614 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1615 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001616 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617static void scsi_eh_lock_door(struct scsi_device *sdev)
1618{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001619 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
James Bottomley91bc31f2009-05-17 09:30:48 -05001621 /*
1622 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1623 * request becomes available
1624 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001625 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001627 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1628 req->cmd[1] = 0;
1629 req->cmd[2] = 0;
1630 req->cmd[3] = 0;
1631 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1632 req->cmd[5] = 0;
1633
1634 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1635
1636 req->cmd_type = REQ_TYPE_BLOCK_PC;
1637 req->cmd_flags |= REQ_QUIET;
1638 req->timeout = 10 * HZ;
1639 req->retries = 5;
1640
1641 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642}
1643
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644/**
1645 * scsi_restart_operations - restart io operations to the specified host.
1646 * @shost: Host we are restarting.
1647 *
1648 * Notes:
1649 * When we entered the error handler, we blocked all further i/o to
1650 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001651 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652static void scsi_restart_operations(struct Scsi_Host *shost)
1653{
1654 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001655 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657 /*
1658 * If the door was locked, we need to insert a door lock request
1659 * onto the head of the SCSI request queue for the device. There
1660 * is no point trying to lock the door of an off-line device.
1661 */
1662 shost_for_each_device(sdev, shost) {
1663 if (scsi_device_online(sdev) && sdev->locked)
1664 scsi_eh_lock_door(sdev);
1665 }
1666
1667 /*
1668 * next free up anything directly waiting upon the host. this
1669 * will be requests for character device operations, and also for
1670 * ioctls to queued block devices.
1671 */
1672 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001673 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
James Bottomley939647e2005-09-18 15:05:20 -05001675 spin_lock_irqsave(shost->host_lock, flags);
1676 if (scsi_host_set_state(shost, SHOST_RUNNING))
1677 if (scsi_host_set_state(shost, SHOST_CANCEL))
1678 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1679 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
1681 wake_up(&shost->host_wait);
1682
1683 /*
1684 * finally we need to re-initiate requests that may be pending. we will
1685 * have had everything blocked while error handling is taking place, and
1686 * now that error recovery is done, we will need to ensure that these
1687 * requests are started.
1688 */
1689 scsi_run_host_queues(shost);
1690}
1691
1692/**
1693 * scsi_eh_ready_devs - check device ready state and recover if not.
1694 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001695 * @work_q: &list_head for pending commands.
1696 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001697 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001698void scsi_eh_ready_devs(struct Scsi_Host *shost,
1699 struct list_head *work_q,
1700 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701{
1702 if (!scsi_eh_stu(shost, work_q, done_q))
1703 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06001704 if (!scsi_eh_target_reset(shost, work_q, done_q))
1705 if (!scsi_eh_bus_reset(shost, work_q, done_q))
1706 if (!scsi_eh_host_reset(work_q, done_q))
1707 scsi_eh_offline_sdevs(work_q,
1708 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001710EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712/**
1713 * scsi_eh_flush_done_q - finish processed commands or retry them.
1714 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001715 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001716void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001718 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001720 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1721 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05001723 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06001724 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1726 " retry cmd: %p\n",
1727 current->comm,
1728 scmd));
1729 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1730 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07001731 /*
1732 * If just we got sense for the device (called
1733 * scsi_eh_get_sense), scmd->result is already
1734 * set, do not set DRIVER_TIMEOUT.
1735 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 if (!scmd->result)
1737 scmd->result |= (DRIVER_TIMEOUT << 24);
1738 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1739 " cmd: %p\n",
1740 current->comm, scmd));
1741 scsi_finish_command(scmd);
1742 }
1743 }
1744}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001745EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
1747/**
1748 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1749 * @shost: Host to unjam.
1750 *
1751 * Notes:
1752 * When we come in here, we *know* that all commands on the bus have
1753 * either completed, failed or timed out. we also know that no further
1754 * commands are being sent to the host, so things are relatively quiet
1755 * and we have freedom to fiddle with things as we wish.
1756 *
1757 * This is only the *default* implementation. it is possible for
1758 * individual drivers to supply their own version of this function, and
1759 * if the maintainer wishes to do this, it is strongly suggested that
1760 * this function be taken as a template and modified. this function
1761 * was designed to correctly handle problems for about 95% of the
1762 * different cases out there, and it should always provide at least a
1763 * reasonable amount of error recovery.
1764 *
1765 * Any command marked 'failed' or 'timeout' must eventually have
1766 * scsi_finish_cmd() called for it. we do all of the retry stuff
1767 * here, so when we restart the host after we return it should have an
1768 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001769 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770static void scsi_unjam_host(struct Scsi_Host *shost)
1771{
1772 unsigned long flags;
1773 LIST_HEAD(eh_work_q);
1774 LIST_HEAD(eh_done_q);
1775
1776 spin_lock_irqsave(shost->host_lock, flags);
1777 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1778 spin_unlock_irqrestore(shost->host_lock, flags);
1779
1780 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1781
1782 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1783 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1784 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1785
1786 scsi_eh_flush_done_q(&eh_done_q);
1787}
1788
1789/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001790 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 * @data: Host for which we are running.
1792 *
1793 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001794 * This is the main error handling loop. This is run as a kernel thread
1795 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001796 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797int scsi_error_handler(void *data)
1798{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001799 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001802 * We use TASK_INTERRUPTIBLE so that the thread is not
1803 * counted against the load average as a running process.
1804 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01001805 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 */
James Bottomley3ed7a472005-09-19 09:50:04 -05001807 set_current_state(TASK_INTERRUPTIBLE);
1808 while (!kthread_should_stop()) {
Tejun Heoee7863b2006-05-15 20:57:20 +09001809 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05001810 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001811 SCSI_LOG_ERROR_RECOVERY(1,
1812 printk("Error handler scsi_eh_%d sleeping\n",
1813 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001814 schedule();
1815 set_current_state(TASK_INTERRUPTIBLE);
1816 continue;
1817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
James Bottomley3ed7a472005-09-19 09:50:04 -05001819 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001820 SCSI_LOG_ERROR_RECOVERY(1,
1821 printk("Error handler scsi_eh_%d waking up\n",
1822 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 /*
1825 * We have a host that is failing for some reason. Figure out
1826 * what we need to do to get it up and online again (if we can).
1827 * If we fail, we end up taking the thing offline.
1828 */
Lin Mingae0751f2011-12-05 09:20:24 +08001829 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
Alan Sternbc4f2402010-06-17 10:41:42 -04001830 SCSI_LOG_ERROR_RECOVERY(1,
1831 printk(KERN_ERR "Error handler scsi_eh_%d "
1832 "unable to autoresume\n",
1833 shost->host_no));
1834 continue;
1835 }
1836
Christoph Hellwig9227c332006-04-01 19:21:04 +02001837 if (shost->transportt->eh_strategy_handler)
1838 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 else
1840 scsi_unjam_host(shost);
1841
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 /*
1843 * Note - if the above fails completely, the action is to take
1844 * individual devices offline and flush the queue of any
1845 * outstanding requests that may have been pending. When we
1846 * restart, we restart any I/O to any other devices on the bus
1847 * which are still online.
1848 */
1849 scsi_restart_operations(shost);
Lin Mingae0751f2011-12-05 09:20:24 +08001850 if (!shost->eh_noresume)
1851 scsi_autopm_put_host(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -05001852 set_current_state(TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04001854 __set_current_state(TASK_RUNNING);
1855
Christoph Hellwigad42eb12005-10-29 01:01:55 +02001856 SCSI_LOG_ERROR_RECOVERY(1,
1857 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05001858 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 return 0;
1860}
1861
1862/*
1863 * Function: scsi_report_bus_reset()
1864 *
1865 * Purpose: Utility function used by low-level drivers to report that
1866 * they have observed a bus reset on the bus being handled.
1867 *
1868 * Arguments: shost - Host in question
1869 * channel - channel on which reset was observed.
1870 *
1871 * Returns: Nothing
1872 *
1873 * Lock status: Host lock must be held.
1874 *
1875 * Notes: This only needs to be called if the reset is one which
1876 * originates from an unknown location. Resets originated
1877 * by the mid-level itself don't need to call this, but there
1878 * should be no harm.
1879 *
1880 * The main purpose of this is to make sure that a CHECK_CONDITION
1881 * is properly treated.
1882 */
1883void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1884{
1885 struct scsi_device *sdev;
1886
1887 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06001888 if (channel == sdev_channel(sdev))
1889 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 }
1891}
1892EXPORT_SYMBOL(scsi_report_bus_reset);
1893
1894/*
1895 * Function: scsi_report_device_reset()
1896 *
1897 * Purpose: Utility function used by low-level drivers to report that
1898 * they have observed a device reset on the device being handled.
1899 *
1900 * Arguments: shost - Host in question
1901 * channel - channel on which reset was observed
1902 * target - target on which reset was observed
1903 *
1904 * Returns: Nothing
1905 *
1906 * Lock status: Host lock must be held
1907 *
1908 * Notes: This only needs to be called if the reset is one which
1909 * originates from an unknown location. Resets originated
1910 * by the mid-level itself don't need to call this, but there
1911 * should be no harm.
1912 *
1913 * The main purpose of this is to make sure that a CHECK_CONDITION
1914 * is properly treated.
1915 */
1916void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1917{
1918 struct scsi_device *sdev;
1919
1920 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001921 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06001922 target == sdev_id(sdev))
1923 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 }
1925}
1926EXPORT_SYMBOL(scsi_report_device_reset);
1927
1928static void
1929scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1930{
1931}
1932
1933/*
1934 * Function: scsi_reset_provider
1935 *
1936 * Purpose: Send requested reset to a bus or device at any phase.
1937 *
1938 * Arguments: device - device to send reset to
1939 * flag - reset type (see scsi.h)
1940 *
1941 * Returns: SUCCESS/FAILURE.
1942 *
1943 * Notes: This is used by the SCSI Generic driver to provide
1944 * Bus/Device reset capability.
1945 */
1946int
1947scsi_reset_provider(struct scsi_device *dev, int flag)
1948{
Alan Sternbc4f2402010-06-17 10:41:42 -04001949 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05001950 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05001952 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 int rtn;
1954
Alan Sternbc4f2402010-06-17 10:41:42 -04001955 if (scsi_autopm_get_host(shost) < 0)
1956 return FAILED;
1957
1958 scmd = scsi_get_command(dev, GFP_KERNEL);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02001959 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02001961
Boaz Harrosh64a87b22008-04-30 11:19:47 +03001962 scmd->cmnd = req.cmd;
1963
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02001965 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966
1967 scmd->cmd_len = 0;
1968
1969 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
James Smartd7a1bb02006-03-08 14:50:12 -05001971 spin_lock_irqsave(shost->host_lock, flags);
1972 shost->tmf_in_progress = 1;
1973 spin_unlock_irqrestore(shost->host_lock, flags);
1974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 switch (flag) {
1976 case SCSI_TRY_RESET_DEVICE:
1977 rtn = scsi_try_bus_device_reset(scmd);
1978 if (rtn == SUCCESS)
1979 break;
1980 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06001981 case SCSI_TRY_RESET_TARGET:
1982 rtn = scsi_try_target_reset(scmd);
1983 if (rtn == SUCCESS)
1984 break;
1985 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 case SCSI_TRY_RESET_BUS:
1987 rtn = scsi_try_bus_reset(scmd);
1988 if (rtn == SUCCESS)
1989 break;
1990 /* FALLTHROUGH */
1991 case SCSI_TRY_RESET_HOST:
1992 rtn = scsi_try_host_reset(scmd);
1993 break;
1994 default:
1995 rtn = FAILED;
1996 }
1997
James Smartd7a1bb02006-03-08 14:50:12 -05001998 spin_lock_irqsave(shost->host_lock, flags);
1999 shost->tmf_in_progress = 0;
2000 spin_unlock_irqrestore(shost->host_lock, flags);
2001
2002 /*
2003 * be sure to wake up anyone who was sleeping or had their queue
2004 * suspended while we performed the TMF.
2005 */
2006 SCSI_LOG_ERROR_RECOVERY(3,
2007 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07002008 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05002009
2010 wake_up(&shost->host_wait);
2011
2012 scsi_run_host_queues(shost);
2013
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 scsi_next_command(scmd);
Alan Sternbc4f2402010-06-17 10:41:42 -04002015 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 return rtn;
2017}
2018EXPORT_SYMBOL(scsi_reset_provider);
2019
2020/**
2021 * scsi_normalize_sense - normalize main elements from either fixed or
2022 * descriptor sense data format into a common format.
2023 *
2024 * @sense_buffer: byte array containing sense data returned by device
2025 * @sb_len: number of valid bytes in sense_buffer
2026 * @sshdr: pointer to instance of structure that common
2027 * elements are written to.
2028 *
2029 * Notes:
2030 * The "main elements" from sense data are: response_code, sense_key,
2031 * asc, ascq and additional_length (only for descriptor format).
2032 *
2033 * Typically this function can be called after a device has
2034 * responded to a SCSI command with the CHECK_CONDITION status.
2035 *
2036 * Return value:
2037 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002038 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2040 struct scsi_sense_hdr *sshdr)
2041{
James Bottomley33aa6872005-08-28 11:31:14 -05002042 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 return 0;
2044
2045 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2046
2047 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05002048
2049 if (!scsi_sense_valid(sshdr))
2050 return 0;
2051
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 if (sshdr->response_code >= 0x72) {
2053 /*
2054 * descriptor format
2055 */
2056 if (sb_len > 1)
2057 sshdr->sense_key = (sense_buffer[1] & 0xf);
2058 if (sb_len > 2)
2059 sshdr->asc = sense_buffer[2];
2060 if (sb_len > 3)
2061 sshdr->ascq = sense_buffer[3];
2062 if (sb_len > 7)
2063 sshdr->additional_length = sense_buffer[7];
2064 } else {
Jesper Juhl0bf8c862011-03-21 20:47:31 +01002065 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 * fixed format
2067 */
2068 if (sb_len > 2)
2069 sshdr->sense_key = (sense_buffer[2] & 0xf);
2070 if (sb_len > 7) {
2071 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2072 sb_len : (sense_buffer[7] + 8);
2073 if (sb_len > 12)
2074 sshdr->asc = sense_buffer[12];
2075 if (sb_len > 13)
2076 sshdr->ascq = sense_buffer[13];
2077 }
2078 }
2079
2080 return 1;
2081}
2082EXPORT_SYMBOL(scsi_normalize_sense);
2083
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2085 struct scsi_sense_hdr *sshdr)
2086{
2087 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002088 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089}
2090EXPORT_SYMBOL(scsi_command_normalize_sense);
2091
2092/**
Rob Landleyeb448202007-11-03 13:30:39 -05002093 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 * @sense_buffer: byte array of descriptor format sense data
2095 * @sb_len: number of valid bytes in sense_buffer
2096 * @desc_type: value of descriptor type to find
2097 * (e.g. 0 -> information)
2098 *
2099 * Notes:
2100 * only valid when sense data is in descriptor format
2101 *
2102 * Return value:
2103 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002104 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2106 int desc_type)
2107{
2108 int add_sen_len, add_len, desc_len, k;
2109 const u8 * descp;
2110
2111 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2112 return NULL;
2113 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2114 return NULL;
2115 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2116 add_sen_len : (sb_len - 8);
2117 descp = &sense_buffer[8];
2118 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2119 descp += desc_len;
2120 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2121 desc_len = add_len + 2;
2122 if (descp[0] == desc_type)
2123 return descp;
2124 if (add_len < 0) // short descriptor ??
2125 break;
2126 }
2127 return NULL;
2128}
2129EXPORT_SYMBOL(scsi_sense_desc_find);
2130
2131/**
Rob Landleyeb448202007-11-03 13:30:39 -05002132 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 * @sense_buffer: byte array of sense data
2134 * @sb_len: number of valid bytes in sense_buffer
2135 * @info_out: pointer to 64 integer where 8 or 4 byte information
2136 * field will be placed if found.
2137 *
2138 * Return value:
2139 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002140 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2142 u64 * info_out)
2143{
2144 int j;
2145 const u8 * ucp;
2146 u64 ull;
2147
2148 if (sb_len < 7)
2149 return 0;
2150 switch (sense_buffer[0] & 0x7f) {
2151 case 0x70:
2152 case 0x71:
2153 if (sense_buffer[0] & 0x80) {
2154 *info_out = (sense_buffer[3] << 24) +
2155 (sense_buffer[4] << 16) +
2156 (sense_buffer[5] << 8) + sense_buffer[6];
2157 return 1;
2158 } else
2159 return 0;
2160 case 0x72:
2161 case 0x73:
2162 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2163 0 /* info desc */);
2164 if (ucp && (0xa == ucp[1])) {
2165 ull = 0;
2166 for (j = 0; j < 8; ++j) {
2167 if (j > 0)
2168 ull <<= 8;
2169 ull |= ucp[4 + j];
2170 }
2171 *info_out = ull;
2172 return 1;
2173 } else
2174 return 0;
2175 default:
2176 return 0;
2177 }
2178}
2179EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002180
2181/**
2182 * scsi_build_sense_buffer - build sense data in a buffer
2183 * @desc: Sense format (non zero == descriptor format,
2184 * 0 == fixed format)
2185 * @buf: Where to build sense data
2186 * @key: Sense key
2187 * @asc: Additional sense code
2188 * @ascq: Additional sense code qualifier
2189 *
2190 **/
2191void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2192{
2193 if (desc) {
2194 buf[0] = 0x72; /* descriptor, current */
2195 buf[1] = key;
2196 buf[2] = asc;
2197 buf[3] = ascq;
2198 buf[7] = 0;
2199 } else {
2200 buf[0] = 0x70; /* fixed, current */
2201 buf[2] = key;
2202 buf[7] = 0xa;
2203 buf[12] = asc;
2204 buf[13] = ascq;
2205 }
2206}
2207EXPORT_SYMBOL(scsi_build_sense_buffer);