blob: 1b9c049bd5c550269083f6435e1547e13b78b5a2 [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>
Hannes Reineckefc736482013-04-25 08:10:00 +020028#include <linux/jiffies.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
30#include <scsi/scsi.h>
Christoph Hellwigbeb40482006-06-10 18:01:03 +020031#include <scsi/scsi_cmnd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <scsi/scsi_dbg.h>
33#include <scsi/scsi_device.h>
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -050034#include <scsi/scsi_driver.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <scsi/scsi_eh.h>
Sagi Grimberg7708c162015-07-08 17:58:52 +030036#include <scsi/scsi_common.h>
James Smartc829c392006-03-13 08:28:57 -050037#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <scsi/scsi_host.h>
39#include <scsi/scsi_ioctl.h>
Christoph Hellwigee14c672015-08-27 14:16:59 +020040#include <scsi/scsi_dh.h>
Christoph Hellwig176aa9d2014-10-11 12:06:47 +020041#include <scsi/sg.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
43#include "scsi_priv.h"
44#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070045#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Kei Tokunagabf816232010-04-01 20:41:40 +090047#include <trace/events/scsi.h>
48
James Bottomley14216562012-07-25 23:55:55 +040049static void scsi_eh_done(struct scsi_cmnd *scmd);
50
Linus Torvalds1da177e2005-04-16 15:20:36 -070051/*
52 * These should *probably* be handled by the host itself.
53 * Since it is allowed to sleep, it probably should.
54 */
55#define BUS_RESET_SETTLE_TIME (10)
56#define HOST_RESET_SETTLE_TIME (10)
57
David Jeffery3eef6252011-05-19 14:41:12 -040058static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
Hannes Reineckee494f6a2013-11-11 13:44:54 +010059static int scsi_try_to_abort_cmd(struct scsi_host_template *,
60 struct scsi_cmnd *);
David Jeffery3eef6252011-05-19 14:41:12 -040061
Linus Torvalds1da177e2005-04-16 15:20:36 -070062/* called with shost->host_lock held */
63void scsi_eh_wakeup(struct Scsi_Host *shost)
64{
Christoph Hellwig74665012014-01-22 15:29:29 +010065 if (atomic_read(&shost->host_busy) == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090066 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050067 wake_up_process(shost->ehandler);
Hannes Reinecke91921e02014-06-25 16:39:59 +020068 SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
69 "Waking error handler thread\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 }
71}
Tejun Heof8bbfc22006-05-19 21:07:05 +090072
73/**
74 * scsi_schedule_eh - schedule EH for SCSI host
75 * @shost: SCSI host to invoke error handling on.
76 *
77 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080078 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090079void scsi_schedule_eh(struct Scsi_Host *shost)
80{
81 unsigned long flags;
82
83 spin_lock_irqsave(shost->host_lock, flags);
84
85 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
86 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
87 shost->host_eh_scheduled++;
88 scsi_eh_wakeup(shost);
89 }
90
91 spin_unlock_irqrestore(shost->host_lock, flags);
92}
93EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Hannes Reineckeb4562022013-10-23 10:51:21 +020095static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
96{
Ren Mingxinbb3b6212013-11-11 13:44:56 +010097 if (!shost->last_reset || shost->eh_deadline == -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +020098 return 0;
99
Hannes Reinecke76ad3e52013-11-11 13:44:55 +0100100 /*
101 * 32bit accesses are guaranteed to be atomic
102 * (on all supported architectures), so instead
103 * of using a spinlock we can as well double check
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100104 * if eh_deadline has been set to 'off' during the
Hannes Reinecke76ad3e52013-11-11 13:44:55 +0100105 * time_before call.
106 */
107 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100108 shost->eh_deadline > -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200109 return 0;
110
111 return 1;
112}
113
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100115 * scmd_eh_abort_handler - Handle command aborts
116 * @work: command to be aborted.
117 */
118void
119scmd_eh_abort_handler(struct work_struct *work)
120{
121 struct scsi_cmnd *scmd =
122 container_of(work, struct scsi_cmnd, abort_work.work);
123 struct scsi_device *sdev = scmd->device;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100124 int rtn;
125
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100126 if (scsi_host_eh_past_deadline(sdev->host)) {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100127 SCSI_LOG_ERROR_RECOVERY(3,
128 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100129 "eh timeout, not aborting\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100130 } else {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100131 SCSI_LOG_ERROR_RECOVERY(3,
132 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100133 "aborting command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100134 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
135 if (rtn == SUCCESS) {
Ulrich Obergfell8922a902014-06-04 13:34:57 +0200136 set_host_byte(scmd, DID_TIME_OUT);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100137 if (scsi_host_eh_past_deadline(sdev->host)) {
138 SCSI_LOG_ERROR_RECOVERY(3,
139 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100140 "eh timeout, not retrying "
141 "aborted command\n"));
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100142 } else if (!scsi_noretry_cmd(scmd) &&
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100143 (++scmd->retries <= scmd->allowed)) {
144 SCSI_LOG_ERROR_RECOVERY(3,
145 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100146 "retry aborted command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100147 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100148 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100149 } else {
150 SCSI_LOG_ERROR_RECOVERY(3,
151 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100152 "finish aborted command\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100153 scsi_finish_command(scmd);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100154 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100155 }
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100156 } else {
157 SCSI_LOG_ERROR_RECOVERY(3,
158 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100159 "cmd abort %s\n",
Hannes Reinecke883a0302014-10-24 14:27:04 +0200160 (rtn == FAST_IO_FAIL) ?
161 "not send" : "failed"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100162 }
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100163 }
164
165 if (!scsi_eh_scmd_add(scmd, 0)) {
166 SCSI_LOG_ERROR_RECOVERY(3,
167 scmd_printk(KERN_WARNING, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100168 "terminate aborted command\n"));
Ulrich Obergfell8922a902014-06-04 13:34:57 +0200169 set_host_byte(scmd, DID_TIME_OUT);
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100170 scsi_finish_command(scmd);
171 }
172}
173
174/**
175 * scsi_abort_command - schedule a command abort
176 * @scmd: scmd to abort.
177 *
178 * We only need to abort commands after a command timeout
179 */
180static int
181scsi_abort_command(struct scsi_cmnd *scmd)
182{
183 struct scsi_device *sdev = scmd->device;
184 struct Scsi_Host *shost = sdev->host;
185 unsigned long flags;
186
187 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
188 /*
189 * Retry after abort failed, escalate to next level.
190 */
Hannes Reinecke7daf4802014-03-31 16:37:34 +0200191 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100192 SCSI_LOG_ERROR_RECOVERY(3,
193 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100194 "previous abort failed\n"));
Bart Van Asschefcc95a72014-06-02 11:50:52 +0200195 BUG_ON(delayed_work_pending(&scmd->abort_work));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100196 return FAILED;
197 }
198
199 /*
200 * Do not try a command abort if
201 * SCSI EH has already started.
202 */
203 spin_lock_irqsave(shost->host_lock, flags);
204 if (scsi_host_in_recovery(shost)) {
205 spin_unlock_irqrestore(shost->host_lock, flags);
206 SCSI_LOG_ERROR_RECOVERY(3,
207 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100208 "not aborting, host in recovery\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100209 return FAILED;
210 }
211
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100212 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100213 shost->last_reset = jiffies;
214 spin_unlock_irqrestore(shost->host_lock, flags);
215
216 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
217 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100218 scmd_printk(KERN_INFO, scmd, "abort scheduled\n"));
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100219 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
220 return SUCCESS;
221}
222
223/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 * scsi_eh_scmd_add - add scsi cmd to error handling.
225 * @scmd: scmd to run eh on.
226 * @eh_flag: optional SCSI_EH flag.
227 *
228 * Return value:
229 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800230 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
232{
233 struct Scsi_Host *shost = scmd->device->host;
234 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -0500235 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236
James Bottomley3ed7a472005-09-19 09:50:04 -0500237 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 return 0;
239
240 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500241 if (scsi_host_set_state(shost, SHOST_RECOVERY))
242 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
243 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100245 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200246 shost->last_reset = jiffies;
247
James Bottomley939647e2005-09-18 15:05:20 -0500248 ret = 1;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100249 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
250 eh_flag &= ~SCSI_EH_CANCEL_CMD;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200251 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 shost->host_failed++;
254 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500255 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500257 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258}
259
260/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700262 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 *
264 * Notes:
265 * We do not need to lock this. There is the potential for a race
266 * only in that the normal completion handling might run, but if the
267 * normal completion function determines that the timer has already
268 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800269 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700270enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700272 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700273 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100274 struct Scsi_Host *host = scmd->device->host;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500275
Kei Tokunagabf816232010-04-01 20:41:40 +0900276 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 scsi_log_completion(scmd, TIMEOUT_ERROR);
278
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100279 if (host->eh_deadline != -1 && !host->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200280 host->last_reset = jiffies;
281
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100282 if (host->transportt->eh_timed_out)
283 rtn = host->transportt->eh_timed_out(scmd);
284 else if (host->hostt->eh_timed_out)
285 rtn = host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500286
Hannes Reineckea33c0702014-06-13 14:01:45 +0200287 if (rtn == BLK_EH_NOT_HANDLED) {
288 if (!host->hostt->no_async_abort &&
289 scsi_abort_command(scmd) == SUCCESS)
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100290 return BLK_EH_NOT_HANDLED;
291
Hannes Reineckea33c0702014-06-13 14:01:45 +0200292 set_host_byte(scmd, DID_TIME_OUT);
293 if (!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))
294 rtn = BLK_EH_HANDLED;
295 }
Jens Axboe242f9dc2008-09-14 05:55:09 -0700296
Christoph Hellwigfa990782008-11-05 15:03:15 +0100297 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
299
300/**
301 * scsi_block_when_processing_errors - Prevent cmds from being queued.
302 * @sdev: Device on which we are performing recovery.
303 *
304 * Description:
305 * We block until the host is out of error recovery, and then check to
306 * see whether the host or the device is offline.
307 *
308 * Return value:
309 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800310 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311int scsi_block_when_processing_errors(struct scsi_device *sdev)
312{
313 int online;
314
James Bottomley939647e2005-09-18 15:05:20 -0500315 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 online = scsi_device_online(sdev);
318
Hannes Reinecke91921e02014-06-25 16:39:59 +0200319 SCSI_LOG_ERROR_RECOVERY(5, sdev_printk(KERN_INFO, sdev,
320 "%s: rtn: %d\n", __func__, online));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322 return online;
323}
324EXPORT_SYMBOL(scsi_block_when_processing_errors);
325
326#ifdef CONFIG_SCSI_LOGGING
327/**
328 * scsi_eh_prt_fail_stats - Log info on failures.
329 * @shost: scsi host being recovered.
330 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800331 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
333 struct list_head *work_q)
334{
335 struct scsi_cmnd *scmd;
336 struct scsi_device *sdev;
337 int total_failures = 0;
338 int cmd_failed = 0;
339 int cmd_cancel = 0;
340 int devices_failed = 0;
341
342 shost_for_each_device(sdev, shost) {
343 list_for_each_entry(scmd, work_q, eh_entry) {
344 if (scmd->device == sdev) {
345 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200346 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 ++cmd_cancel;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100348 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 ++cmd_failed;
350 }
351 }
352
353 if (cmd_cancel || cmd_failed) {
354 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea3a790d2014-10-24 14:27:03 +0200355 shost_printk(KERN_INFO, shost,
James Bottomley9ccfc752005-10-02 11:45:08 -0500356 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700357 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500358 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 cmd_cancel = 0;
360 cmd_failed = 0;
361 ++devices_failed;
362 }
363 }
364
Hannes Reinecke91921e02014-06-25 16:39:59 +0200365 SCSI_LOG_ERROR_RECOVERY(2, shost_printk(KERN_INFO, shost,
366 "Total of %d commands on %d"
367 " devices require eh work\n",
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100368 total_failures, devices_failed));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
370#endif
371
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400372 /**
373 * scsi_report_lun_change - Set flag on all *other* devices on the same target
374 * to indicate that a UNIT ATTENTION is expected.
375 * @sdev: Device reporting the UNIT ATTENTION
376 */
377static void scsi_report_lun_change(struct scsi_device *sdev)
378{
379 sdev->sdev_target->expecting_lun_change = 1;
380}
381
382/**
383 * scsi_report_sense - Examine scsi sense information and log messages for
384 * certain conditions, also issue uevents for some of them.
385 * @sdev: Device reporting the sense code
386 * @sshdr: sshdr to be examined
387 */
388static void scsi_report_sense(struct scsi_device *sdev,
389 struct scsi_sense_hdr *sshdr)
390{
391 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
392
393 if (sshdr->sense_key == UNIT_ATTENTION) {
394 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
395 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
396 sdev_printk(KERN_WARNING, sdev,
397 "Inquiry data has changed");
398 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
399 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
400 scsi_report_lun_change(sdev);
401 sdev_printk(KERN_WARNING, sdev,
402 "Warning! Received an indication that the "
403 "LUN assignments on this target have "
404 "changed. The Linux SCSI layer does not "
405 "automatically remap LUN assignments.\n");
406 } else if (sshdr->asc == 0x3f)
407 sdev_printk(KERN_WARNING, sdev,
408 "Warning! Received an indication that the "
409 "operating parameters on this target have "
410 "changed. The Linux SCSI layer does not "
411 "automatically adjust these parameters.\n");
412
413 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
414 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
415 sdev_printk(KERN_WARNING, sdev,
416 "Warning! Received an indication that the "
417 "LUN reached a thin provisioning soft "
418 "threshold.\n");
419 }
420
421 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
422 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
423 sdev_printk(KERN_WARNING, sdev,
424 "Mode parameters changed");
Hannes Reinecke14c3e672015-07-06 13:41:53 +0200425 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x06) {
426 evt_type = SDEV_EVT_ALUA_STATE_CHANGE_REPORTED;
427 sdev_printk(KERN_WARNING, sdev,
428 "Asymmetric access state changed");
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400429 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
430 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
431 sdev_printk(KERN_WARNING, sdev,
432 "Capacity data has changed");
433 } else if (sshdr->asc == 0x2a)
434 sdev_printk(KERN_WARNING, sdev,
435 "Parameters changed");
436 }
437
438 if (evt_type != SDEV_EVT_MAXBITS) {
439 set_bit(evt_type, sdev->pending_events);
440 schedule_work(&sdev->event_work);
441 }
442}
443
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444/**
445 * scsi_check_sense - Examine scsi cmd sense
446 * @scmd: Cmd to have sense checked.
447 *
448 * Return value:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200449 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 *
451 * Notes:
452 * When a deferred error is detected the current command has
453 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800454 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455static int scsi_check_sense(struct scsi_cmnd *scmd)
456{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700457 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 struct scsi_sense_hdr sshdr;
459
460 if (! scsi_command_normalize_sense(scmd, &sshdr))
461 return FAILED; /* no valid sense data */
462
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400463 scsi_report_sense(sdev, &sshdr);
464
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 if (scsi_sense_is_deferred(&sshdr))
466 return NEEDS_RETRY;
467
Christoph Hellwigee14c672015-08-27 14:16:59 +0200468 if (sdev->handler && sdev->handler->check_sense) {
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700469 int rc;
470
Christoph Hellwigee14c672015-08-27 14:16:59 +0200471 rc = sdev->handler->check_sense(sdev, &sshdr);
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700472 if (rc != SCSI_RETURN_NOT_HANDLED)
473 return rc;
474 /* handler does not care. Drop down to default handling */
475 }
476
Christoph Hellwige925cc42014-11-06 15:11:22 -0600477 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
478 /*
479 * nasty: for mid-layer issued TURs, we need to return the
480 * actual sense data without any recovery attempt. For eh
481 * issued ones, we need to try to recover and interpret
482 */
483 return SUCCESS;
484
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 /*
486 * Previous logic looked for FILEMARK, EOM or ILI which are
487 * mainly associated with tapes and returned SUCCESS.
488 */
489 if (sshdr.response_code == 0x70) {
490 /* fixed format */
491 if (scmd->sense_buffer[2] & 0xe0)
492 return SUCCESS;
493 } else {
494 /*
495 * descriptor format: look for "stream commands sense data
496 * descriptor" (see SSC-3). Assume single sense data
497 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
498 */
499 if ((sshdr.additional_length > 3) &&
500 (scmd->sense_buffer[8] == 0x4) &&
501 (scmd->sense_buffer[11] & 0xe0))
502 return SUCCESS;
503 }
504
505 switch (sshdr.sense_key) {
506 case NO_SENSE:
507 return SUCCESS;
508 case RECOVERED_ERROR:
509 return /* soft_error */ SUCCESS;
510
511 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400512 if (sshdr.asc == 0x10) /* DIF */
513 return SUCCESS;
514
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 return NEEDS_RETRY;
516 case NOT_READY:
517 case UNIT_ATTENTION:
518 /*
519 * if we are expecting a cc/ua because of a bus reset that we
520 * performed, treat this just as a retry. otherwise this is
521 * information that we should pass up to the upper-level driver
522 * so that we can deal with it there.
523 */
524 if (scmd->device->expecting_cc_ua) {
TARUISI Hiroakidfcf7772011-08-11 20:25:20 +0900525 /*
526 * Because some device does not queue unit
527 * attentions correctly, we carefully check
528 * additional sense code and qualifier so as
529 * not to squash media change unit attention.
530 */
531 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
532 scmd->device->expecting_cc_ua = 0;
533 return NEEDS_RETRY;
534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 }
536 /*
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400537 * we might also expect a cc/ua if another LUN on the target
538 * reported a UA with an ASC/ASCQ of 3F 0E -
539 * REPORTED LUNS DATA HAS CHANGED.
540 */
541 if (scmd->device->sdev_target->expecting_lun_change &&
542 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
543 return NEEDS_RETRY;
544 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100545 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 * should retry.
547 */
548 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
549 return NEEDS_RETRY;
550 /*
551 * if the device is not started, we need to wake
552 * the error handler to start the motor
553 */
554 if (scmd->device->allow_restart &&
555 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
556 return FAILED;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100557 /*
558 * Pass the UA upwards for a determination in the completion
559 * functions.
560 */
561 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100563 /* these are not supported */
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +0200564 case DATA_PROTECT:
565 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
566 /* Thin provisioning hard threshold reached */
567 set_host_byte(scmd, DID_ALLOC_FAILURE);
568 return SUCCESS;
569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 case COPY_ABORTED:
571 case VOLUME_OVERFLOW:
572 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100573 case BLANK_CHECK:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200574 set_host_byte(scmd, DID_TARGET_FAILURE);
575 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800578 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
579 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
580 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke7e782af2013-07-01 15:16:26 +0200581 set_host_byte(scmd, DID_MEDIUM_ERROR);
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200582 return SUCCESS;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800583 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 return NEEDS_RETRY;
585
586 case HARDWARE_ERROR:
587 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700588 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 else
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200590 set_host_byte(scmd, DID_TARGET_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591
592 case ILLEGAL_REQUEST:
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500593 if (sshdr.asc == 0x20 || /* Invalid command operation code */
594 sshdr.asc == 0x21 || /* Logical block address out of range */
595 sshdr.asc == 0x24 || /* Invalid field in cdb */
596 sshdr.asc == 0x26) { /* Parameter value invalid */
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200597 set_host_byte(scmd, DID_TARGET_FAILURE);
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500598 }
599 return SUCCESS;
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 default:
602 return SUCCESS;
603 }
604}
605
Vasu Dev4a840672009-10-22 15:46:33 -0700606static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
607{
608 struct scsi_host_template *sht = sdev->host->hostt;
609 struct scsi_device *tmp_sdev;
610
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100611 if (!sht->track_queue_depth ||
Vasu Dev4a840672009-10-22 15:46:33 -0700612 sdev->queue_depth >= sdev->max_queue_depth)
613 return;
614
615 if (time_before(jiffies,
616 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
617 return;
618
619 if (time_before(jiffies,
620 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
621 return;
622
623 /*
624 * Walk all devices of a target and do
625 * ramp up on them.
626 */
627 shost_for_each_device(tmp_sdev, sdev->host) {
628 if (tmp_sdev->channel != sdev->channel ||
629 tmp_sdev->id != sdev->id ||
630 tmp_sdev->queue_depth == sdev->max_queue_depth)
631 continue;
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100632
Christoph Hellwigdb5ed4d2014-11-13 15:08:42 +0100633 scsi_change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1);
Vasu Dev4a840672009-10-22 15:46:33 -0700634 sdev->last_queue_ramp_up = jiffies;
635 }
636}
637
Mike Christie42a6a912009-10-15 17:46:44 -0700638static void scsi_handle_queue_full(struct scsi_device *sdev)
639{
640 struct scsi_host_template *sht = sdev->host->hostt;
641 struct scsi_device *tmp_sdev;
642
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100643 if (!sht->track_queue_depth)
Mike Christie42a6a912009-10-15 17:46:44 -0700644 return;
645
646 shost_for_each_device(tmp_sdev, sdev->host) {
647 if (tmp_sdev->channel != sdev->channel ||
648 tmp_sdev->id != sdev->id)
649 continue;
650 /*
651 * We do not know the number of commands that were at
652 * the device when we got the queue full so we start
653 * from the highest possible value and work our way down.
654 */
Christoph Hellwigc40ecc12014-11-13 14:25:11 +0100655 scsi_track_queue_full(tmp_sdev, tmp_sdev->queue_depth - 1);
Mike Christie42a6a912009-10-15 17:46:44 -0700656 }
657}
658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659/**
660 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
661 * @scmd: SCSI cmd to examine.
662 *
663 * Notes:
664 * This is *only* called when we are examining the status of commands
665 * queued during error recovery. the main difference here is that we
666 * don't allow for the possibility of retries here, and we are a lot
667 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800668 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
670{
671 /*
672 * first check the host byte, to see if there is anything in there
673 * that would indicate what we need to do.
674 */
675 if (host_byte(scmd->result) == DID_RESET) {
676 /*
677 * rats. we are already in the error handler, so we now
678 * get to try and figure out what to do next. if the sense
679 * is valid, we have a pretty good idea of what to do.
680 * if not, we mark it as FAILED.
681 */
682 return scsi_check_sense(scmd);
683 }
684 if (host_byte(scmd->result) != DID_OK)
685 return FAILED;
686
687 /*
688 * next, check the message byte.
689 */
690 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
691 return FAILED;
692
693 /*
694 * now, check the status byte to see if this indicates
695 * anything special.
696 */
697 switch (status_byte(scmd->result)) {
698 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700699 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 case COMMAND_TERMINATED:
701 return SUCCESS;
702 case CHECK_CONDITION:
703 return scsi_check_sense(scmd);
704 case CONDITION_GOOD:
705 case INTERMEDIATE_GOOD:
706 case INTERMEDIATE_C_GOOD:
707 /*
708 * who knows? FIXME(eric)
709 */
710 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500711 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500712 if (scmd->cmnd[0] == TEST_UNIT_READY)
713 /* it is a success, we probed the device and
714 * found it */
715 return SUCCESS;
716 /* otherwise, we failed to send the command */
717 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700719 scsi_handle_queue_full(scmd->device);
720 /* fall through */
721 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200722 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 default:
724 return FAILED;
725 }
726 return FAILED;
727}
728
729/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 * scsi_eh_done - Completion function for error handling.
731 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800732 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733static void scsi_eh_done(struct scsi_cmnd *scmd)
734{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100735 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600736
Hannes Reinecke91921e02014-06-25 16:39:59 +0200737 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +0100738 "%s result: %x\n", __func__, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600739
740 eh_action = scmd->device->host->eh_action;
741 if (eh_action)
742 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743}
744
745/**
Brian King292148f2007-01-30 17:51:17 -0600746 * scsi_try_host_reset - ask host adapter to reset itself
Geert Uytterhoevenc2b3ebd2013-05-17 13:22:29 +0200747 * @scmd: SCSI cmd to send host reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800748 */
Brian King292148f2007-01-30 17:51:17 -0600749static int scsi_try_host_reset(struct scsi_cmnd *scmd)
750{
751 unsigned long flags;
752 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100753 struct Scsi_Host *host = scmd->device->host;
754 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600755
Hannes Reinecke91921e02014-06-25 16:39:59 +0200756 SCSI_LOG_ERROR_RECOVERY(3,
757 shost_printk(KERN_INFO, host, "Snd Host RST\n"));
Brian King292148f2007-01-30 17:51:17 -0600758
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100759 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600760 return FAILED;
761
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100762 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600763
764 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100765 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600766 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100767 spin_lock_irqsave(host->host_lock, flags);
768 scsi_report_bus_reset(host, scmd_channel(scmd));
769 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600770 }
771
772 return rtn;
773}
774
775/**
776 * scsi_try_bus_reset - ask host to perform a bus reset
777 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800778 */
Brian King292148f2007-01-30 17:51:17 -0600779static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
780{
781 unsigned long flags;
782 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100783 struct Scsi_Host *host = scmd->device->host;
784 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600785
Hannes Reinecke91921e02014-06-25 16:39:59 +0200786 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
787 "%s: Snd Bus RST\n", __func__));
Brian King292148f2007-01-30 17:51:17 -0600788
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100789 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600790 return FAILED;
791
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100792 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600793
794 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100795 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600796 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100797 spin_lock_irqsave(host->host_lock, flags);
798 scsi_report_bus_reset(host, scmd_channel(scmd));
799 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600800 }
801
802 return rtn;
803}
804
Mike Christie30bd7df2008-02-29 18:25:19 -0600805static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
806{
807 sdev->was_reset = 1;
808 sdev->expecting_cc_ua = 1;
809}
810
811/**
812 * scsi_try_target_reset - Ask host to perform a target reset
813 * @scmd: SCSI cmd used to send a target reset
814 *
815 * Notes:
816 * There is no timeout for this operation. if this operation is
817 * unreliable for a given host, then the host itself needs to put a
818 * timer on it, and set the host back to a consistent state prior to
819 * returning.
820 */
821static int scsi_try_target_reset(struct scsi_cmnd *scmd)
822{
823 unsigned long flags;
824 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100825 struct Scsi_Host *host = scmd->device->host;
826 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600827
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100828 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600829 return FAILED;
830
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100831 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600832 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100833 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600834 __starget_for_each_device(scsi_target(scmd->device), NULL,
835 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100836 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600837 }
838
839 return rtn;
840}
841
Brian King292148f2007-01-30 17:51:17 -0600842/**
843 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
844 * @scmd: SCSI cmd used to send BDR
845 *
846 * Notes:
847 * There is no timeout for this operation. if this operation is
848 * unreliable for a given host, then the host itself needs to put a
849 * timer on it, and set the host back to a consistent state prior to
850 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800851 */
Brian King292148f2007-01-30 17:51:17 -0600852static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
853{
854 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100855 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600856
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100857 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600858 return FAILED;
859
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100860 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600861 if (rtn == SUCCESS)
862 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600863 return rtn;
864}
865
Hannes Reinecke883a0302014-10-24 14:27:04 +0200866/**
867 * scsi_try_to_abort_cmd - Ask host to abort a SCSI command
Randy Dunlap6583f6f2014-12-29 09:40:56 -0800868 * @hostt: SCSI driver host template
Hannes Reinecke883a0302014-10-24 14:27:04 +0200869 * @scmd: SCSI cmd used to send a target reset
870 *
871 * Return value:
872 * SUCCESS, FAILED, or FAST_IO_FAIL
873 *
874 * Notes:
875 * SUCCESS does not necessarily indicate that the command
876 * has been aborted; it only indicates that the LLDDs
877 * has cleared all references to that command.
878 * LLDDs should return FAILED only if an abort was required
879 * but could not be executed. LLDDs should return FAST_IO_FAIL
880 * if the device is temporarily unavailable (eg due to a
881 * link down on FibreChannel)
882 */
883static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt,
884 struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600885{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100886 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600887 return FAILED;
888
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100889 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600890}
891
Brian King292148f2007-01-30 17:51:17 -0600892static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
893{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100894 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600895 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600896 if (scsi_try_target_reset(scmd) != SUCCESS)
897 if (scsi_try_bus_reset(scmd) != SUCCESS)
898 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600899}
900
901/**
Santosh Y3b729f72012-04-08 18:47:09 +0530902 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100903 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200904 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200905 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300906 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200907 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100908 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200909 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200910 * as part of the error recovery process. If @sense_bytes is 0 the command
911 * sent must be one that does not transfer any data. If @sense_bytes != 0
912 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
913 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800914 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200915void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
916 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917{
f59114b2005-04-17 15:00:23 -0500918 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Christoph Hellwig631c2282006-07-08 20:42:15 +0200920 /*
921 * We need saved copies of a number of fields - this is because
922 * error handling may need to overwrite these with different values
923 * to run different commands, and once error handling is complete,
924 * we will need to restore these values prior to running the actual
925 * command.
926 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200927 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300928 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200929 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200930 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200931 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200932 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400933 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400934 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200935
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400936 scmd->prot_op = SCSI_PROT_NORMAL;
James Bottomleyc69e6f82014-04-10 13:36:11 -0700937 scmd->eh_eflags = 0;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300938 scmd->cmnd = ses->eh_cmnd;
939 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200940 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200941 scmd->request->next_rq = NULL;
Alan Stern644373a2014-03-28 10:51:15 -0700942 scmd->result = 0;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200943
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200944 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200945 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
946 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200947 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200948 scmd->sdb.length);
949 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200950 scmd->sc_data_direction = DMA_FROM_DEVICE;
Tony Battersby0c958ec2015-07-16 11:40:41 -0400951 scmd->sdb.table.nents = scmd->sdb.table.orig_nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200952 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200953 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200954 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200955 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200956 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200957 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300958 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200959 memcpy(scmd->cmnd, cmnd, cmnd_size);
960 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
961 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200962 }
963
964 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200965
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200966 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500968 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Christoph Hellwig631c2282006-07-08 20:42:15 +0200970 /*
971 * Zero the sense buffer. The scsi spec mandates that any
972 * untransferred sense data should be interpreted as being zero.
973 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900974 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200975}
976EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200977
Boaz Harroshe1c23462007-10-08 16:36:45 +0200978/**
Santosh Y3b729f72012-04-08 18:47:09 +0530979 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +0200980 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200981 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200982 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200983 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800984 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200985void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
986{
987 /*
988 * Restore original data
989 */
990 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300991 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200992 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200993 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200994 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200995 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400996 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400997 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200998}
999EXPORT_SYMBOL(scsi_eh_restore_cmnd);
1000
1001/**
Santosh Y3b729f72012-04-08 18:47:09 +05301002 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +02001003 * @scmd: SCSI command structure to hijack
1004 * @cmnd: CDB to send
1005 * @cmnd_size: size in bytes of @cmnd
1006 * @timeout: timeout for this request
1007 * @sense_bytes: size of sense data to copy or 0
1008 *
1009 * This function is used to send a scsi command down to a target device
1010 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
1011 *
1012 * Return value:
1013 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001014 */
Boaz Harroshe1c23462007-10-08 16:36:45 +02001015static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1016 int cmnd_size, int timeout, unsigned sense_bytes)
1017{
1018 struct scsi_device *sdev = scmd->device;
1019 struct Scsi_Host *shost = sdev->host;
1020 DECLARE_COMPLETION_ONSTACK(done);
Hannes Reineckefc736482013-04-25 08:10:00 +02001021 unsigned long timeleft = timeout;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001022 struct scsi_eh_save ses;
Hannes Reineckefc736482013-04-25 08:10:00 +02001023 const unsigned long stall_for = msecs_to_jiffies(100);
Boaz Harroshe1c23462007-10-08 16:36:45 +02001024 int rtn;
1025
Hannes Reineckefc736482013-04-25 08:10:00 +02001026retry:
Boaz Harroshe1c23462007-10-08 16:36:45 +02001027 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001028 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -05001031 scmd->scsi_done = scsi_eh_done;
Hannes Reineckefc736482013-04-25 08:10:00 +02001032 rtn = shost->hostt->queuecommand(shost, scmd);
1033 if (rtn) {
1034 if (timeleft > stall_for) {
1035 scsi_eh_restore_cmnd(scmd, &ses);
1036 timeleft -= stall_for;
1037 msleep(jiffies_to_msecs(stall_for));
1038 goto retry;
1039 }
1040 /* signal not to enter either branch of the if () below */
1041 timeleft = 0;
Alan Stern511833a2014-11-21 10:44:49 -05001042 rtn = FAILED;
Hannes Reineckefc736482013-04-25 08:10:00 +02001043 } else {
1044 timeleft = wait_for_completion_timeout(&done, timeout);
Hannes Reineckeac61d192014-05-08 08:09:56 +02001045 rtn = SUCCESS;
Hannes Reineckefc736482013-04-25 08:10:00 +02001046 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
f59114b2005-04-17 15:00:23 -05001048 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Hannes Reineckefc736482013-04-25 08:10:00 +02001050 scsi_log_completion(scmd, rtn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
Hannes Reinecke91921e02014-06-25 16:39:59 +02001052 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001053 "%s timeleft: %ld\n",
1054 __func__, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
1056 /*
Hannes Reineckefc736482013-04-25 08:10:00 +02001057 * If there is time left scsi_eh_done got called, and we will examine
1058 * the actual status codes to see whether the command actually did
1059 * complete normally, else if we have a zero return and no time left,
1060 * the command must still be pending, so abort it and return FAILED.
1061 * If we never actually managed to issue the command, because
1062 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1063 * value above (so don't execute either branch of the if)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001065 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 rtn = scsi_eh_completed_normally(scmd);
Hannes Reinecke91921e02014-06-25 16:39:59 +02001067 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
1068 "%s: scsi_eh_completed_normally %x\n", __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001069
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 switch (rtn) {
1071 case SUCCESS:
1072 case NEEDS_RETRY:
1073 case FAILED:
1074 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +02001075 case ADD_TO_MLQUEUE:
1076 rtn = NEEDS_RETRY;
1077 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 default:
1079 rtn = FAILED;
1080 break;
1081 }
Alan Stern511833a2014-11-21 10:44:49 -05001082 } else if (rtn != FAILED) {
Brian King292148f2007-01-30 17:51:17 -06001083 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001084 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
1086
Boaz Harroshe1c23462007-10-08 16:36:45 +02001087 scsi_eh_restore_cmnd(scmd, &ses);
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -05001088
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 return rtn;
1090}
1091
1092/**
1093 * scsi_request_sense - Request sense data from a particular target.
1094 * @scmd: SCSI cmd for request sense.
1095 *
1096 * Notes:
1097 * Some hosts automatically obtain this information, others require
1098 * that we obtain it on our own. This function will *not* return until
1099 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001100 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101static int scsi_request_sense(struct scsi_cmnd *scmd)
1102{
Martin K. Petersen0816c922013-05-10 10:36:04 -04001103 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104}
1105
James Bottomley24510792013-11-11 13:44:53 +01001106static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1107{
1108 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1109 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1110 if (sdrv->eh_action)
1111 rtn = sdrv->eh_action(scmd, rtn);
1112 }
1113 return rtn;
1114}
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116/**
1117 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1118 * @scmd: Original SCSI cmd that eh has finished.
1119 * @done_q: Queue for processed commands.
1120 *
1121 * Notes:
1122 * We don't want to use the normal command completion while we are are
1123 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -05001124 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 * keep a list of pending commands for final completion, and once we
1126 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001127 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001128void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129{
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001130 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 list_move_tail(&scmd->eh_entry, done_q);
1132}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001133EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
1135/**
1136 * scsi_eh_get_sense - Get device sense data.
1137 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -05001138 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 *
1140 * Description:
1141 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001142 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 *
1144 * Notes:
1145 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -05001146 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 * it down before we request it.
1148 *
1149 * All drivers should request sense information internally these days,
1150 * so for now all I have to say is tough noogies if you end up in here.
1151 *
1152 * XXX: Long term this code should go away, but that needs an audit of
1153 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001154 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001155int scsi_eh_get_sense(struct list_head *work_q,
1156 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001158 struct scsi_cmnd *scmd, *next;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001159 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 int rtn;
1161
jiang.biao2@zte.com.cn709c75b2015-07-31 17:52:10 +08001162 /*
1163 * If SCSI_EH_ABORT_SCHEDULED has been set, it is timeout IO,
1164 * should not get sense.
1165 */
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001166 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001167 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
jiang.biao2@zte.com.cn709c75b2015-07-31 17:52:10 +08001168 (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 SCSI_SENSE_VALID(scmd))
1170 continue;
1171
Hannes Reineckeb4562022013-10-23 10:51:21 +02001172 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001173 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001174 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001175 scmd_printk(KERN_INFO, scmd,
1176 "%s: skip request sense, past eh deadline\n",
1177 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001178 break;
1179 }
James Bottomleyd555a2a2014-03-28 10:50:17 -07001180 if (status_byte(scmd->result) != CHECK_CONDITION)
1181 /*
1182 * don't request sense if there's no check condition
1183 * status because the error we're processing isn't one
1184 * that has a sense code (and some devices get
1185 * confused by sense requests out of the blue)
1186 */
1187 continue;
1188
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001189 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1190 "%s: requesting sense\n",
1191 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 rtn = scsi_request_sense(scmd);
1193 if (rtn != SUCCESS)
1194 continue;
1195
Hannes Reinecke91921e02014-06-25 16:39:59 +02001196 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001197 "sense requested, result %x\n", scmd->result));
Hannes Reinecked811b842014-10-24 14:26:45 +02001198 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense(scmd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
1200 rtn = scsi_decide_disposition(scmd);
1201
1202 /*
1203 * if the result was normal, then just pass it along to the
1204 * upper level.
1205 */
1206 if (rtn == SUCCESS)
1207 /* we don't want this command reissued, just
1208 * finished with the sense data, so set
1209 * retries to the max allowed to ensure it
1210 * won't get reissued */
1211 scmd->retries = scmd->allowed;
1212 else if (rtn != NEEDS_RETRY)
1213 continue;
1214
1215 scsi_eh_finish_cmd(scmd, done_q);
1216 }
1217
1218 return list_empty(work_q);
1219}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001220EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
1222/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001224 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 *
1226 * Return value:
1227 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001228 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229static int scsi_eh_tur(struct scsi_cmnd *scmd)
1230{
1231 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1232 int retry_cnt = 1, rtn;
1233
1234retry_tur:
Martin K. Petersen0816c922013-05-10 10:36:04 -04001235 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1236 scmd->device->eh_timeout, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237
Hannes Reinecke91921e02014-06-25 16:39:59 +02001238 SCSI_LOG_ERROR_RECOVERY(3, scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01001239 "%s return: %x\n", __func__, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +02001240
1241 switch (rtn) {
1242 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 if (retry_cnt--)
1244 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001245 /*FALLTHRU*/
1246 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -05001247 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001248 default:
1249 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -05001250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251}
1252
1253/**
David Jeffery3eef6252011-05-19 14:41:12 -04001254 * scsi_eh_test_devices - check if devices are responding from error recovery.
1255 * @cmd_list: scsi commands in error recovery.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001256 * @work_q: queue for commands which still need more error recovery
1257 * @done_q: queue for commands which are finished
1258 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
David Jeffery3eef6252011-05-19 14:41:12 -04001259 *
1260 * Decription:
1261 * Tests if devices are in a working state. Commands to devices now in
1262 * a working state are sent to the done_q while commands to devices which
1263 * are still failing to respond are returned to the work_q for more
1264 * processing.
1265 **/
1266static int scsi_eh_test_devices(struct list_head *cmd_list,
1267 struct list_head *work_q,
1268 struct list_head *done_q, int try_stu)
1269{
1270 struct scsi_cmnd *scmd, *next;
1271 struct scsi_device *sdev;
1272 int finish_cmds;
1273
1274 while (!list_empty(cmd_list)) {
1275 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1276 sdev = scmd->device;
1277
Hannes Reineckeb4562022013-10-23 10:51:21 +02001278 if (!try_stu) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001279 if (scsi_host_eh_past_deadline(sdev->host)) {
1280 /* Push items back onto work_q */
1281 list_splice_init(cmd_list, work_q);
Hannes Reineckeb4562022013-10-23 10:51:21 +02001282 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001283 sdev_printk(KERN_INFO, sdev,
1284 "%s: skip test device, past eh deadline",
1285 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001286 break;
1287 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001288 }
1289
David Jeffery3eef6252011-05-19 14:41:12 -04001290 finish_cmds = !scsi_device_online(scmd->device) ||
1291 (try_stu && !scsi_eh_try_stu(scmd) &&
1292 !scsi_eh_tur(scmd)) ||
1293 !scsi_eh_tur(scmd);
1294
1295 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1296 if (scmd->device == sdev) {
James Bottomley24510792013-11-11 13:44:53 +01001297 if (finish_cmds &&
1298 (try_stu ||
1299 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
David Jeffery3eef6252011-05-19 14:41:12 -04001300 scsi_eh_finish_cmd(scmd, done_q);
1301 else
1302 list_move_tail(&scmd->eh_entry, work_q);
1303 }
1304 }
1305 return list_empty(work_q);
1306}
1307
1308
1309/**
Rob Landleyeb448202007-11-03 13:30:39 -05001310 * scsi_eh_abort_cmds - abort pending commands.
1311 * @work_q: &list_head for pending commands.
1312 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 *
1314 * Decription:
1315 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -05001316 * running command. This only works out to be the case if we have one
1317 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 * no sense to try and abort the command, since as far as the shost
1319 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001320 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321static int scsi_eh_abort_cmds(struct list_head *work_q,
1322 struct list_head *done_q)
1323{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001324 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001325 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 int rtn;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001327 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001329 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001330 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 continue;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001332 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001333 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001334 list_splice_init(&check_list, work_q);
1335 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001336 scmd_printk(KERN_INFO, scmd,
1337 "%s: skip aborting cmd, past eh deadline\n",
1338 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001339 return list_empty(work_q);
1340 }
Hannes Reinecke91921e02014-06-25 16:39:59 +02001341 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001342 scmd_printk(KERN_INFO, scmd,
1343 "%s: aborting cmd\n", current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001344 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001345 if (rtn == FAILED) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001346 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001347 scmd_printk(KERN_INFO, scmd,
1348 "%s: aborting cmd failed\n",
1349 current->comm));
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001350 list_splice_init(&check_list, work_q);
1351 return list_empty(work_q);
1352 }
1353 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1354 if (rtn == FAST_IO_FAIL)
1355 scsi_eh_finish_cmd(scmd, done_q);
1356 else
1357 list_move_tail(&scmd->eh_entry, &check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 }
1359
David Jeffery3eef6252011-05-19 14:41:12 -04001360 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361}
1362
1363/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001365 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 *
1367 * Return value:
1368 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001369 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1371{
1372 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
Christoph Hellwig631c2282006-07-08 20:42:15 +02001374 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001375 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
Brian Kinged773e62007-03-29 15:25:52 -05001377 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001378 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001379
Christoph Hellwig631c2282006-07-08 20:42:15 +02001380 if (rtn == SUCCESS)
1381 return 0;
1382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 return 1;
1385}
1386
1387 /**
1388 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001389 * @shost: &scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001390 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001391 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 *
1393 * Notes:
1394 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001395 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001396 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397static int scsi_eh_stu(struct Scsi_Host *shost,
1398 struct list_head *work_q,
1399 struct list_head *done_q)
1400{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001401 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 struct scsi_device *sdev;
1403
1404 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001405 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001406 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001407 sdev_printk(KERN_INFO, sdev,
1408 "%s: skip START_UNIT, past eh deadline\n",
1409 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001410 break;
1411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 stu_scmd = NULL;
1413 list_for_each_entry(scmd, work_q, eh_entry)
1414 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1415 scsi_check_sense(scmd) == FAILED ) {
1416 stu_scmd = scmd;
1417 break;
1418 }
1419
1420 if (!stu_scmd)
1421 continue;
1422
Hannes Reinecke91921e02014-06-25 16:39:59 +02001423 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001424 sdev_printk(KERN_INFO, sdev,
1425 "%s: Sending START_UNIT\n",
1426 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 if (!scsi_eh_try_stu(stu_scmd)) {
1429 if (!scsi_device_online(sdev) ||
1430 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001431 list_for_each_entry_safe(scmd, next,
1432 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001433 if (scmd->device == sdev &&
1434 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 scsi_eh_finish_cmd(scmd, done_q);
1436 }
1437 }
1438 } else {
1439 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001440 sdev_printk(KERN_INFO, sdev,
1441 "%s: START_UNIT failed\n",
1442 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
1444 }
1445
1446 return list_empty(work_q);
1447}
1448
1449
1450/**
1451 * scsi_eh_bus_device_reset - send bdr if needed
1452 * @shost: scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001453 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001454 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 *
1456 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001457 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 * devices that are jammed or not - if we have multiple devices, it
1459 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001460 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001461 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1463 struct list_head *work_q,
1464 struct list_head *done_q)
1465{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001466 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 struct scsi_device *sdev;
1468 int rtn;
1469
1470 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001471 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001472 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001473 sdev_printk(KERN_INFO, sdev,
1474 "%s: skip BDR, past eh deadline\n",
1475 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001476 break;
1477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 bdr_scmd = NULL;
1479 list_for_each_entry(scmd, work_q, eh_entry)
1480 if (scmd->device == sdev) {
1481 bdr_scmd = scmd;
1482 break;
1483 }
1484
1485 if (!bdr_scmd)
1486 continue;
1487
Hannes Reinecke91921e02014-06-25 16:39:59 +02001488 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001489 sdev_printk(KERN_INFO, sdev,
1490 "%s: Sending BDR\n", current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001492 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001494 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001496 list_for_each_entry_safe(scmd, next,
1497 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001498 if (scmd->device == sdev &&
1499 scsi_eh_action(scmd, rtn) != FAILED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 scsi_eh_finish_cmd(scmd,
1501 done_q);
1502 }
1503 }
1504 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001505 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001506 sdev_printk(KERN_INFO, sdev,
1507 "%s: BDR failed\n", current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 }
1509 }
1510
1511 return list_empty(work_q);
1512}
1513
1514/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001515 * scsi_eh_target_reset - send target reset if needed
1516 * @shost: scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001517 * @work_q: &list_head for pending commands.
Mike Christie30bd7df2008-02-29 18:25:19 -06001518 * @done_q: &list_head for processed commands.
1519 *
1520 * Notes:
1521 * Try a target reset.
1522 */
1523static int scsi_eh_target_reset(struct Scsi_Host *shost,
1524 struct list_head *work_q,
1525 struct list_head *done_q)
1526{
James Bottomley98db5192010-10-25 15:53:41 -05001527 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001528 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001529
James Bottomley98db5192010-10-25 15:53:41 -05001530 list_splice_init(work_q, &tmp_list);
1531
1532 while (!list_empty(&tmp_list)) {
1533 struct scsi_cmnd *next, *scmd;
1534 int rtn;
1535 unsigned int id;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001536
Hannes Reineckeb4562022013-10-23 10:51:21 +02001537 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001538 /* push back on work queue for further processing */
1539 list_splice_init(&check_list, work_q);
1540 list_splice_init(&tmp_list, work_q);
1541 SCSI_LOG_ERROR_RECOVERY(3,
1542 shost_printk(KERN_INFO, shost,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001543 "%s: Skip target reset, past eh deadline\n",
1544 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001545 return list_empty(work_q);
1546 }
James Bottomley98db5192010-10-25 15:53:41 -05001547
1548 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1549 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001550
Hannes Reinecke91921e02014-06-25 16:39:59 +02001551 SCSI_LOG_ERROR_RECOVERY(3,
1552 shost_printk(KERN_INFO, shost,
1553 "%s: Sending target reset to target %d\n",
1554 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001555 rtn = scsi_try_target_reset(scmd);
1556 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Hannes Reinecke91921e02014-06-25 16:39:59 +02001557 SCSI_LOG_ERROR_RECOVERY(3,
1558 shost_printk(KERN_INFO, shost,
1559 "%s: Target reset failed"
1560 " target: %d\n",
1561 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001562 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1563 if (scmd_id(scmd) != id)
1564 continue;
1565
David Jeffery3eef6252011-05-19 14:41:12 -04001566 if (rtn == SUCCESS)
1567 list_move_tail(&scmd->eh_entry, &check_list);
1568 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001569 scsi_eh_finish_cmd(scmd, done_q);
1570 else
1571 /* push back on work queue for further processing */
1572 list_move(&scmd->eh_entry, work_q);
1573 }
1574 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001575
David Jeffery3eef6252011-05-19 14:41:12 -04001576 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001577}
1578
1579/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001580 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001581 * @shost: &scsi host being recovered.
Randy Dunlap74cf2982014-08-16 14:15:11 -07001582 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05001583 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001584 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1586 struct list_head *work_q,
1587 struct list_head *done_q)
1588{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001589 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001590 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 unsigned int channel;
1592 int rtn;
1593
1594 /*
1595 * we really want to loop over the various channels, and do this on
1596 * a channel by channel basis. we should also check to see if any
1597 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001598 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 */
1600
1601 for (channel = 0; channel <= shost->max_channel; channel++) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001602 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001603 list_splice_init(&check_list, work_q);
1604 SCSI_LOG_ERROR_RECOVERY(3,
1605 shost_printk(KERN_INFO, shost,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02001606 "%s: skip BRST, past eh deadline\n",
1607 current->comm));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001608 return list_empty(work_q);
1609 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001610
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 chan_scmd = NULL;
1612 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001613 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 chan_scmd = scmd;
1615 break;
1616 /*
1617 * FIXME add back in some support for
1618 * soft_reset devices.
1619 */
1620 }
1621 }
1622
1623 if (!chan_scmd)
1624 continue;
Hannes Reinecke91921e02014-06-25 16:39:59 +02001625 SCSI_LOG_ERROR_RECOVERY(3,
1626 shost_printk(KERN_INFO, shost,
1627 "%s: Sending BRST chan: %d\n",
1628 current->comm, channel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001630 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001631 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001632 if (channel == scmd_channel(scmd)) {
1633 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 scsi_eh_finish_cmd(scmd,
1635 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001636 else
1637 list_move_tail(&scmd->eh_entry,
1638 &check_list);
1639 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 }
1641 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001642 SCSI_LOG_ERROR_RECOVERY(3,
1643 shost_printk(KERN_INFO, shost,
1644 "%s: BRST failed chan: %d\n",
1645 current->comm, channel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 }
1647 }
David Jeffery3eef6252011-05-19 14:41:12 -04001648 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649}
1650
1651/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001652 * scsi_eh_host_reset - send a host reset
Randy Dunlap74cf2982014-08-16 14:15:11 -07001653 * @shost: host to be reset.
1654 * @work_q: &list_head for pending commands.
1655 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001656 */
Hannes Reinecke91921e02014-06-25 16:39:59 +02001657static int scsi_eh_host_reset(struct Scsi_Host *shost,
1658 struct list_head *work_q,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 struct list_head *done_q)
1660{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001661 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001662 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665 if (!list_empty(work_q)) {
1666 scmd = list_entry(work_q->next,
1667 struct scsi_cmnd, eh_entry);
1668
Hannes Reinecke91921e02014-06-25 16:39:59 +02001669 SCSI_LOG_ERROR_RECOVERY(3,
1670 shost_printk(KERN_INFO, shost,
1671 "%s: Sending HRST\n",
1672 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673
1674 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001675 if (rtn == SUCCESS) {
1676 list_splice_init(work_q, &check_list);
1677 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001678 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 scsi_eh_finish_cmd(scmd, done_q);
1680 }
1681 } else {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001682 SCSI_LOG_ERROR_RECOVERY(3,
1683 shost_printk(KERN_INFO, shost,
1684 "%s: HRST failed\n",
1685 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 }
1687 }
David Jeffery3eef6252011-05-19 14:41:12 -04001688 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689}
1690
1691/**
1692 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
Randy Dunlap74cf2982014-08-16 14:15:11 -07001693 * @work_q: &list_head for pending commands.
1694 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001695 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696static void scsi_eh_offline_sdevs(struct list_head *work_q,
1697 struct list_head *done_q)
1698{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001699 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001701 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001702 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1703 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001705 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 /*
1707 * FIXME: Handle lost cmds.
1708 */
1709 }
1710 scsi_eh_finish_cmd(scmd, done_q);
1711 }
1712 return;
1713}
1714
1715/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001716 * scsi_noretry_cmd - determine if command should be failed fast
Mike Christie4a274462008-08-19 18:45:31 -05001717 * @scmd: SCSI cmd to examine.
1718 */
1719int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1720{
1721 switch (host_byte(scmd->result)) {
1722 case DID_OK:
1723 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001724 case DID_TIME_OUT:
1725 goto check_type;
Mike Christie4a274462008-08-19 18:45:31 -05001726 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001727 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001728 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001729 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001730 case DID_ERROR:
1731 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1732 status_byte(scmd->result) == RESERVATION_CONFLICT)
1733 return 0;
1734 /* fall through */
1735 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001736 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001737 }
1738
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001739 if (status_byte(scmd->result) != CHECK_CONDITION)
1740 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001741
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001742check_type:
1743 /*
1744 * assume caller has checked sense and determined
1745 * the check condition was retryable.
1746 */
1747 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1748 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1749 return 1;
1750 else
1751 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001752}
1753
1754/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1756 * @scmd: SCSI cmd to examine.
1757 *
1758 * Notes:
1759 * This is *only* called when we are examining the status after sending
1760 * out the actual data command. any commands that are queued for error
1761 * recovery (e.g. test_unit_ready) do *not* come through here.
1762 *
1763 * When this routine returns failed, it means the error handler thread
1764 * is woken. In cases where the error code indicates an error that
1765 * doesn't require the error handler read (i.e. we don't need to
1766 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001767 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768int scsi_decide_disposition(struct scsi_cmnd *scmd)
1769{
1770 int rtn;
1771
1772 /*
1773 * if the device is offline, then we clearly just pass the result back
1774 * up to the top level.
1775 */
1776 if (!scsi_device_online(scmd->device)) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02001777 SCSI_LOG_ERROR_RECOVERY(5, scmd_printk(KERN_INFO, scmd,
1778 "%s: device offline - report as SUCCESS\n", __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 return SUCCESS;
1780 }
1781
1782 /*
1783 * first check the host byte, to see if there is anything in there
1784 * that would indicate what we need to do.
1785 */
1786 switch (host_byte(scmd->result)) {
1787 case DID_PASSTHROUGH:
1788 /*
1789 * no matter what, pass this through to the upper layer.
1790 * nuke this special code so that it looks like we are saying
1791 * did_ok.
1792 */
1793 scmd->result &= 0xff00ffff;
1794 return SUCCESS;
1795 case DID_OK:
1796 /*
1797 * looks good. drop through, and check the next byte.
1798 */
1799 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001800 case DID_ABORT:
1801 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
Ulrich Obergfell8922a902014-06-04 13:34:57 +02001802 set_host_byte(scmd, DID_TIME_OUT);
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001803 return SUCCESS;
1804 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 case DID_NO_CONNECT:
1806 case DID_BAD_TARGET:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 /*
1808 * note - this means that we just report the status back
1809 * to the top level driver, not that we actually think
1810 * that it indicates SUCCESS.
1811 */
1812 return SUCCESS;
1813 /*
1814 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001815 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 * in order to avoid endless loops (db)
1817 *
1818 * actually this is a bug in this function here. we should
1819 * be mindful of the maximum number of retries specified
1820 * and not get stuck in a loop.
1821 */
1822 case DID_SOFT_ERROR:
1823 goto maybe_retry;
1824 case DID_IMM_RETRY:
1825 return NEEDS_RETRY;
1826
bf341912005-04-12 17:49:09 -05001827 case DID_REQUEUE:
1828 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001829 case DID_TRANSPORT_DISRUPTED:
1830 /*
1831 * LLD/transport was disrupted during processing of the IO.
1832 * The transport class is now blocked/blocking,
1833 * and the transport will decide what to do with the IO
Mike Christie939c22882008-11-04 19:47:19 -06001834 * based on its timers and recovery capablilities if
1835 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001836 */
Mike Christie939c22882008-11-04 19:47:19 -06001837 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001838 case DID_TRANSPORT_FAILFAST:
1839 /*
1840 * The transport decided to failfast the IO (most likely
1841 * the fast io fail tmo fired), so send IO directly upwards.
1842 */
1843 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 case DID_ERROR:
1845 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1846 status_byte(scmd->result) == RESERVATION_CONFLICT)
1847 /*
1848 * execute reservation conflict processing code
1849 * lower down
1850 */
1851 break;
1852 /* fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 case DID_BUS_BUSY:
1854 case DID_PARITY:
1855 goto maybe_retry;
1856 case DID_TIME_OUT:
1857 /*
1858 * when we scan the bus, we get timeout messages for
1859 * these commands if there is no device available.
1860 * other hosts report did_no_connect for the same thing.
1861 */
1862 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1863 scmd->cmnd[0] == INQUIRY)) {
1864 return SUCCESS;
1865 } else {
1866 return FAILED;
1867 }
1868 case DID_RESET:
1869 return SUCCESS;
1870 default:
1871 return FAILED;
1872 }
1873
1874 /*
1875 * next, check the message byte.
1876 */
1877 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1878 return FAILED;
1879
1880 /*
1881 * check the status byte to see if this indicates anything special.
1882 */
1883 switch (status_byte(scmd->result)) {
1884 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001885 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 /*
1887 * the case of trying to send too many commands to a
1888 * tagged queueing device.
1889 */
1890 case BUSY:
1891 /*
1892 * device can't talk to us at the moment. Should only
1893 * occur (SAM-3) when the task queue is empty, so will cause
1894 * the empty queue handling to trigger a stall in the
1895 * device.
1896 */
1897 return ADD_TO_MLQUEUE;
1898 case GOOD:
Ewan D. Milne279afdf2013-08-08 15:07:48 -04001899 if (scmd->cmnd[0] == REPORT_LUNS)
1900 scmd->device->sdev_target->expecting_lun_change = 0;
Vasu Dev4a840672009-10-22 15:46:33 -07001901 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001904 case TASK_ABORTED:
1905 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 case CHECK_CONDITION:
1907 rtn = scsi_check_sense(scmd);
1908 if (rtn == NEEDS_RETRY)
1909 goto maybe_retry;
1910 /* if rtn == FAILED, we have no sense information;
1911 * returning FAILED will wake the error handler thread
1912 * to collect the sense and redo the decide
1913 * disposition */
1914 return rtn;
1915 case CONDITION_GOOD:
1916 case INTERMEDIATE_GOOD:
1917 case INTERMEDIATE_C_GOOD:
1918 case ACA_ACTIVE:
1919 /*
1920 * who knows? FIXME(eric)
1921 */
1922 return SUCCESS;
1923
1924 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001925 sdev_printk(KERN_INFO, scmd->device,
1926 "reservation conflict\n");
Moger, Babu2082ebc2012-01-24 20:38:46 +00001927 set_host_byte(scmd, DID_NEXUS_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 return SUCCESS; /* causes immediate i/o error */
1929 default:
1930 return FAILED;
1931 }
1932 return FAILED;
1933
1934 maybe_retry:
1935
1936 /* we requeue for retry because the error was retryable, and
1937 * the request was not marked fast fail. Note that above,
1938 * even if the request is marked fast fail, we still requeue
1939 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001940 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001941 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 return NEEDS_RETRY;
1943 } else {
1944 /*
1945 * no more retries - report this one back to upper level.
1946 */
1947 return SUCCESS;
1948 }
1949}
1950
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001951static void eh_lock_door_done(struct request *req, int uptodate)
1952{
1953 __blk_put_request(req->q, req);
1954}
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 * scsi_eh_lock_door - Prevent medium removal for the specified device
1958 * @sdev: SCSI device to prevent medium removal
1959 *
1960 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001961 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 *
1963 * Notes:
1964 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1965 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001966 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967static void scsi_eh_lock_door(struct scsi_device *sdev)
1968{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001969 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
James Bottomley91bc31f2009-05-17 09:30:48 -05001971 /*
Mel Gorman71baba42015-11-06 16:28:28 -08001972 * blk_get_request with GFP_KERNEL (__GFP_RECLAIM) sleeps until a
James Bottomley91bc31f2009-05-17 09:30:48 -05001973 * request becomes available
1974 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001975 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Joe Lawrencea492f072014-08-28 08:15:21 -06001976 if (IS_ERR(req))
Joe Lawrenceeb571ee2014-07-02 15:35:16 -04001977 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
Jens Axboef27b0872014-06-06 07:57:37 -06001979 blk_rq_set_block_pc(req);
1980
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001981 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1982 req->cmd[1] = 0;
1983 req->cmd[2] = 0;
1984 req->cmd[3] = 0;
1985 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1986 req->cmd[5] = 0;
1987
1988 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1989
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001990 req->cmd_flags |= REQ_QUIET;
1991 req->timeout = 10 * HZ;
1992 req->retries = 5;
1993
1994 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995}
1996
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997/**
1998 * scsi_restart_operations - restart io operations to the specified host.
1999 * @shost: Host we are restarting.
2000 *
2001 * Notes:
2002 * When we entered the error handler, we blocked all further i/o to
2003 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005static void scsi_restart_operations(struct Scsi_Host *shost)
2006{
2007 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05002008 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
2010 /*
2011 * If the door was locked, we need to insert a door lock request
2012 * onto the head of the SCSI request queue for the device. There
2013 * is no point trying to lock the door of an off-line device.
2014 */
2015 shost_for_each_device(sdev, shost) {
Christoph Hellwig48379272014-11-03 19:36:40 +01002016 if (scsi_device_online(sdev) && sdev->was_reset && sdev->locked) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 scsi_eh_lock_door(sdev);
Christoph Hellwig48379272014-11-03 19:36:40 +01002018 sdev->was_reset = 0;
2019 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 }
2021
2022 /*
2023 * next free up anything directly waiting upon the host. this
2024 * will be requests for character device operations, and also for
2025 * ioctls to queued block devices.
2026 */
Hannes Reineckeb4562022013-10-23 10:51:21 +02002027 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002028 shost_printk(KERN_INFO, shost, "waking up host to restart\n"));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
James Bottomley939647e2005-09-18 15:05:20 -05002030 spin_lock_irqsave(shost->host_lock, flags);
2031 if (scsi_host_set_state(shost, SHOST_RUNNING))
2032 if (scsi_host_set_state(shost, SHOST_CANCEL))
2033 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2034 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
2036 wake_up(&shost->host_wait);
2037
2038 /*
2039 * finally we need to re-initiate requests that may be pending. we will
2040 * have had everything blocked while error handling is taking place, and
2041 * now that error recovery is done, we will need to ensure that these
2042 * requests are started.
2043 */
2044 scsi_run_host_queues(shost);
Dan Williams57fc2e32012-06-21 23:25:32 -07002045
2046 /*
2047 * if eh is active and host_eh_scheduled is pending we need to re-run
2048 * recovery. we do this check after scsi_run_host_queues() to allow
2049 * everything pent up since the last eh run a chance to make forward
2050 * progress before we sync again. Either we'll immediately re-run
2051 * recovery or scsi_device_unbusy() will wake us again when these
2052 * pending commands complete.
2053 */
2054 spin_lock_irqsave(shost->host_lock, flags);
2055 if (shost->host_eh_scheduled)
2056 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2057 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2058 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059}
2060
2061/**
2062 * scsi_eh_ready_devs - check device ready state and recover if not.
Randy Dunlap74cf2982014-08-16 14:15:11 -07002063 * @shost: host to be recovered.
2064 * @work_q: &list_head for pending commands.
Rob Landleyeb448202007-11-03 13:30:39 -05002065 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002066 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002067void scsi_eh_ready_devs(struct Scsi_Host *shost,
2068 struct list_head *work_q,
2069 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070{
2071 if (!scsi_eh_stu(shost, work_q, done_q))
2072 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002073 if (!scsi_eh_target_reset(shost, work_q, done_q))
2074 if (!scsi_eh_bus_reset(shost, work_q, done_q))
Hannes Reinecke91921e02014-06-25 16:39:59 +02002075 if (!scsi_eh_host_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002076 scsi_eh_offline_sdevs(work_q,
2077 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002079EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080
2081/**
2082 * scsi_eh_flush_done_q - finish processed commands or retry them.
2083 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002084 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09002085void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002087 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002089 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2090 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05002092 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06002093 (++scmd->retries <= scmd->allowed)) {
Hannes Reinecke91921e02014-06-25 16:39:59 +02002094 SCSI_LOG_ERROR_RECOVERY(3,
2095 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01002096 "%s: flush retry cmd\n",
2097 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2099 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07002100 /*
2101 * If just we got sense for the device (called
2102 * scsi_eh_get_sense), scmd->result is already
2103 * set, do not set DRIVER_TIMEOUT.
2104 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if (!scmd->result)
2106 scmd->result |= (DRIVER_TIMEOUT << 24);
Hannes Reinecke91921e02014-06-25 16:39:59 +02002107 SCSI_LOG_ERROR_RECOVERY(3,
2108 scmd_printk(KERN_INFO, scmd,
Hannes Reinecke470613b2015-01-08 07:43:50 +01002109 "%s: flush finish cmd\n",
2110 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 scsi_finish_command(scmd);
2112 }
2113 }
2114}
Tejun Heo041c5fc2006-01-23 13:09:36 +09002115EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
2117/**
2118 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2119 * @shost: Host to unjam.
2120 *
2121 * Notes:
2122 * When we come in here, we *know* that all commands on the bus have
2123 * either completed, failed or timed out. we also know that no further
2124 * commands are being sent to the host, so things are relatively quiet
2125 * and we have freedom to fiddle with things as we wish.
2126 *
2127 * This is only the *default* implementation. it is possible for
2128 * individual drivers to supply their own version of this function, and
2129 * if the maintainer wishes to do this, it is strongly suggested that
2130 * this function be taken as a template and modified. this function
2131 * was designed to correctly handle problems for about 95% of the
2132 * different cases out there, and it should always provide at least a
2133 * reasonable amount of error recovery.
2134 *
2135 * Any command marked 'failed' or 'timeout' must eventually have
2136 * scsi_finish_cmd() called for it. we do all of the retry stuff
2137 * here, so when we restart the host after we return it should have an
2138 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002139 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140static void scsi_unjam_host(struct Scsi_Host *shost)
2141{
2142 unsigned long flags;
2143 LIST_HEAD(eh_work_q);
2144 LIST_HEAD(eh_done_q);
2145
2146 spin_lock_irqsave(shost->host_lock, flags);
2147 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2148 spin_unlock_irqrestore(shost->host_lock, flags);
2149
2150 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2151
2152 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2153 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2154 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2155
Hannes Reineckeb4562022013-10-23 10:51:21 +02002156 spin_lock_irqsave(shost->host_lock, flags);
Ren Mingxinbb3b6212013-11-11 13:44:56 +01002157 if (shost->eh_deadline != -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +02002158 shost->last_reset = 0;
2159 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 scsi_eh_flush_done_q(&eh_done_q);
2161}
2162
2163/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002164 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 * @data: Host for which we are running.
2166 *
2167 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002168 * This is the main error handling loop. This is run as a kernel thread
2169 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002170 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171int scsi_error_handler(void *data)
2172{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002173 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002176 * We use TASK_INTERRUPTIBLE so that the thread is not
2177 * counted against the load average as a running process.
2178 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01002179 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 */
Michal Hocko537b6042015-08-27 20:16:37 +02002181 while (true) {
2182 /*
2183 * The sequence in kthread_stop() sets the stop flag first
2184 * then wakes the process. To avoid missed wakeups, the task
2185 * should always be in a non running state before the stop
2186 * flag is checked
2187 */
Dan Williamsb9d5c6b2012-06-21 23:25:42 -07002188 set_current_state(TASK_INTERRUPTIBLE);
Michal Hocko537b6042015-08-27 20:16:37 +02002189 if (kthread_should_stop())
2190 break;
2191
Tejun Heoee7863b2006-05-15 20:57:20 +09002192 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
Christoph Hellwig74665012014-01-22 15:29:29 +01002193 shost->host_failed != atomic_read(&shost->host_busy)) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002194 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002195 shost_printk(KERN_INFO, shost,
2196 "scsi_eh_%d: sleeping\n",
2197 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002198 schedule();
James Bottomley3ed7a472005-09-19 09:50:04 -05002199 continue;
2200 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201
James Bottomley3ed7a472005-09-19 09:50:04 -05002202 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002203 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002204 shost_printk(KERN_INFO, shost,
2205 "scsi_eh_%d: waking up %d/%d/%d\n",
2206 shost->host_no, shost->host_eh_scheduled,
Christoph Hellwig74665012014-01-22 15:29:29 +01002207 shost->host_failed,
2208 atomic_read(&shost->host_busy)));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 /*
2211 * We have a host that is failing for some reason. Figure out
2212 * what we need to do to get it up and online again (if we can).
2213 * If we fail, we end up taking the thing offline.
2214 */
Lin Mingae0751f2011-12-05 09:20:24 +08002215 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
Alan Sternbc4f2402010-06-17 10:41:42 -04002216 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reineckea222b1e2014-10-24 14:27:02 +02002217 shost_printk(KERN_ERR, shost,
2218 "scsi_eh_%d: unable to autoresume\n",
2219 shost->host_no));
Alan Sternbc4f2402010-06-17 10:41:42 -04002220 continue;
2221 }
2222
Christoph Hellwig9227c332006-04-01 19:21:04 +02002223 if (shost->transportt->eh_strategy_handler)
2224 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 else
2226 scsi_unjam_host(shost);
2227
Wei Fang72d8c362016-06-07 14:53:56 +08002228 /* All scmds have been handled */
2229 shost->host_failed = 0;
2230
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 /*
2232 * Note - if the above fails completely, the action is to take
2233 * individual devices offline and flush the queue of any
2234 * outstanding requests that may have been pending. When we
2235 * restart, we restart any I/O to any other devices on the bus
2236 * which are still online.
2237 */
2238 scsi_restart_operations(shost);
Lin Mingae0751f2011-12-05 09:20:24 +08002239 if (!shost->eh_noresume)
2240 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04002242 __set_current_state(TASK_RUNNING);
2243
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002244 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002245 shost_printk(KERN_INFO, shost,
2246 "Error handler scsi_eh_%d exiting\n",
2247 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002248 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 return 0;
2250}
2251
2252/*
2253 * Function: scsi_report_bus_reset()
2254 *
2255 * Purpose: Utility function used by low-level drivers to report that
2256 * they have observed a bus reset on the bus being handled.
2257 *
2258 * Arguments: shost - Host in question
2259 * channel - channel on which reset was observed.
2260 *
2261 * Returns: Nothing
2262 *
2263 * Lock status: Host lock must be held.
2264 *
2265 * Notes: This only needs to be called if the reset is one which
2266 * originates from an unknown location. Resets originated
2267 * by the mid-level itself don't need to call this, but there
2268 * should be no harm.
2269 *
2270 * The main purpose of this is to make sure that a CHECK_CONDITION
2271 * is properly treated.
2272 */
2273void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2274{
2275 struct scsi_device *sdev;
2276
2277 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06002278 if (channel == sdev_channel(sdev))
2279 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 }
2281}
2282EXPORT_SYMBOL(scsi_report_bus_reset);
2283
2284/*
2285 * Function: scsi_report_device_reset()
2286 *
2287 * Purpose: Utility function used by low-level drivers to report that
2288 * they have observed a device reset on the device being handled.
2289 *
2290 * Arguments: shost - Host in question
2291 * channel - channel on which reset was observed
2292 * target - target on which reset was observed
2293 *
2294 * Returns: Nothing
2295 *
2296 * Lock status: Host lock must be held
2297 *
2298 * Notes: This only needs to be called if the reset is one which
2299 * originates from an unknown location. Resets originated
2300 * by the mid-level itself don't need to call this, but there
2301 * should be no harm.
2302 *
2303 * The main purpose of this is to make sure that a CHECK_CONDITION
2304 * is properly treated.
2305 */
2306void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2307{
2308 struct scsi_device *sdev;
2309
2310 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04002311 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06002312 target == sdev_id(sdev))
2313 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 }
2315}
2316EXPORT_SYMBOL(scsi_report_device_reset);
2317
2318static void
2319scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2320{
2321}
2322
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002323/**
2324 * scsi_ioctl_reset: explicitly reset a host/bus/target/device
2325 * @dev: scsi_device to operate on
2326 * @arg: reset type (see sg.h)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 */
2328int
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002329scsi_ioctl_reset(struct scsi_device *dev, int __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330{
Alan Sternbc4f2402010-06-17 10:41:42 -04002331 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05002332 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05002334 unsigned long flags;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002335 int error = 0, rtn, val;
2336
2337 if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_RAWIO))
2338 return -EACCES;
2339
2340 error = get_user(val, arg);
2341 if (error)
2342 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343
Alan Sternbc4f2402010-06-17 10:41:42 -04002344 if (scsi_autopm_get_host(shost) < 0)
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002345 return -EIO;
Alan Sternbc4f2402010-06-17 10:41:42 -04002346
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002347 error = -EIO;
Alan Sternbc4f2402010-06-17 10:41:42 -04002348 scmd = scsi_get_command(dev, GFP_KERNEL);
Christoph Hellwig0f121dd2014-09-05 18:00:05 -07002349 if (!scmd)
Christoph Hellwig95eeb5f2014-05-01 16:51:05 +02002350 goto out_put_autopm_host;
Christoph Hellwig95eeb5f2014-05-01 16:51:05 +02002351
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02002352 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02002354
Boaz Harrosh64a87b22008-04-30 11:19:47 +03002355 scmd->cmnd = req.cmd;
2356
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02002358 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
2360 scmd->cmd_len = 0;
2361
2362 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
James Smartd7a1bb02006-03-08 14:50:12 -05002364 spin_lock_irqsave(shost->host_lock, flags);
2365 shost->tmf_in_progress = 1;
2366 spin_unlock_irqrestore(shost->host_lock, flags);
2367
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002368 switch (val & ~SG_SCSI_RESET_NO_ESCALATE) {
2369 case SG_SCSI_RESET_NOTHING:
2370 rtn = SUCCESS;
2371 break;
2372 case SG_SCSI_RESET_DEVICE:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 rtn = scsi_try_bus_device_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002374 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 break;
2376 /* FALLTHROUGH */
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002377 case SG_SCSI_RESET_TARGET:
Mike Christie30bd7df2008-02-29 18:25:19 -06002378 rtn = scsi_try_target_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002379 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Mike Christie30bd7df2008-02-29 18:25:19 -06002380 break;
2381 /* FALLTHROUGH */
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002382 case SG_SCSI_RESET_BUS:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383 rtn = scsi_try_bus_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002384 if (rtn == SUCCESS || (val & SG_SCSI_RESET_NO_ESCALATE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 break;
2386 /* FALLTHROUGH */
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002387 case SG_SCSI_RESET_HOST:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 rtn = scsi_try_host_reset(scmd);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002389 if (rtn == SUCCESS)
2390 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 default:
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002392 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 rtn = FAILED;
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002394 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 }
2396
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002397 error = (rtn == SUCCESS) ? 0 : -EIO;
2398
James Smartd7a1bb02006-03-08 14:50:12 -05002399 spin_lock_irqsave(shost->host_lock, flags);
2400 shost->tmf_in_progress = 0;
2401 spin_unlock_irqrestore(shost->host_lock, flags);
2402
2403 /*
2404 * be sure to wake up anyone who was sleeping or had their queue
2405 * suspended while we performed the TMF.
2406 */
2407 SCSI_LOG_ERROR_RECOVERY(3,
Hannes Reinecke91921e02014-06-25 16:39:59 +02002408 shost_printk(KERN_INFO, shost,
2409 "waking up host to restart after TMF\n"));
James Smartd7a1bb02006-03-08 14:50:12 -05002410
2411 wake_up(&shost->host_wait);
James Smartd7a1bb02006-03-08 14:50:12 -05002412 scsi_run_host_queues(shost);
2413
Christoph Hellwig0f121dd2014-09-05 18:00:05 -07002414 scsi_put_command(scmd);
2415
Christoph Hellwig04796332014-02-20 14:20:55 -08002416out_put_autopm_host:
Alan Sternbc4f2402010-06-17 10:41:42 -04002417 scsi_autopm_put_host(shost);
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002418 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419}
Christoph Hellwig176aa9d2014-10-11 12:06:47 +02002420EXPORT_SYMBOL(scsi_ioctl_reset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
Hannes Reinecke4753cbc2014-10-24 14:26:52 +02002422bool scsi_command_normalize_sense(const struct scsi_cmnd *cmd,
2423 struct scsi_sense_hdr *sshdr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424{
2425 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002426 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427}
2428EXPORT_SYMBOL(scsi_command_normalize_sense);
2429
2430/**
Rob Landleyeb448202007-11-03 13:30:39 -05002431 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 * @sense_buffer: byte array of sense data
2433 * @sb_len: number of valid bytes in sense_buffer
2434 * @info_out: pointer to 64 integer where 8 or 4 byte information
2435 * field will be placed if found.
2436 *
2437 * Return value:
2438 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002439 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2441 u64 * info_out)
2442{
2443 int j;
2444 const u8 * ucp;
2445 u64 ull;
2446
2447 if (sb_len < 7)
2448 return 0;
2449 switch (sense_buffer[0] & 0x7f) {
2450 case 0x70:
2451 case 0x71:
2452 if (sense_buffer[0] & 0x80) {
2453 *info_out = (sense_buffer[3] << 24) +
2454 (sense_buffer[4] << 16) +
2455 (sense_buffer[5] << 8) + sense_buffer[6];
2456 return 1;
2457 } else
2458 return 0;
2459 case 0x72:
2460 case 0x73:
2461 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2462 0 /* info desc */);
2463 if (ucp && (0xa == ucp[1])) {
2464 ull = 0;
2465 for (j = 0; j < 8; ++j) {
2466 if (j > 0)
2467 ull <<= 8;
2468 ull |= ucp[4 + j];
2469 }
2470 *info_out = ull;
2471 return 1;
2472 } else
2473 return 0;
2474 default:
2475 return 0;
2476 }
2477}
2478EXPORT_SYMBOL(scsi_get_sense_info_fld);