blob: af624619d547243f00f970ee9fc6d68d7f3ac391 [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>
James Smartc829c392006-03-13 08:28:57 -050036#include <scsi/scsi_transport.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <scsi/scsi_host.h>
38#include <scsi/scsi_ioctl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
40#include "scsi_priv.h"
41#include "scsi_logging.h"
Adrian Bunk79ee8302007-08-10 14:50:42 -070042#include "scsi_transport_api.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Kei Tokunagabf816232010-04-01 20:41:40 +090044#include <trace/events/scsi.h>
45
James Bottomley14216562012-07-25 23:55:55 +040046static void scsi_eh_done(struct scsi_cmnd *scmd);
47
Linus Torvalds1da177e2005-04-16 15:20:36 -070048/*
49 * These should *probably* be handled by the host itself.
50 * Since it is allowed to sleep, it probably should.
51 */
52#define BUS_RESET_SETTLE_TIME (10)
53#define HOST_RESET_SETTLE_TIME (10)
54
David Jeffery3eef6252011-05-19 14:41:12 -040055static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
Hannes Reineckee494f6a2013-11-11 13:44:54 +010056static int scsi_try_to_abort_cmd(struct scsi_host_template *,
57 struct scsi_cmnd *);
David Jeffery3eef6252011-05-19 14:41:12 -040058
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/* called with shost->host_lock held */
60void scsi_eh_wakeup(struct Scsi_Host *shost)
61{
62 if (shost->host_busy == shost->host_failed) {
Kei Tokunagabf816232010-04-01 20:41:40 +090063 trace_scsi_eh_wakeup(shost);
James Bottomley3ed7a472005-09-19 09:50:04 -050064 wake_up_process(shost->ehandler);
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 SCSI_LOG_ERROR_RECOVERY(5,
66 printk("Waking error handler thread\n"));
67 }
68}
Tejun Heof8bbfc22006-05-19 21:07:05 +090069
70/**
71 * scsi_schedule_eh - schedule EH for SCSI host
72 * @shost: SCSI host to invoke error handling on.
73 *
74 * Schedule SCSI EH without scmd.
Randy Dunlapdc8875e2007-11-15 15:42:30 -080075 */
Tejun Heof8bbfc22006-05-19 21:07:05 +090076void scsi_schedule_eh(struct Scsi_Host *shost)
77{
78 unsigned long flags;
79
80 spin_lock_irqsave(shost->host_lock, flags);
81
82 if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
83 scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
84 shost->host_eh_scheduled++;
85 scsi_eh_wakeup(shost);
86 }
87
88 spin_unlock_irqrestore(shost->host_lock, flags);
89}
90EXPORT_SYMBOL_GPL(scsi_schedule_eh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Hannes Reineckeb4562022013-10-23 10:51:21 +020092static int scsi_host_eh_past_deadline(struct Scsi_Host *shost)
93{
Ren Mingxinbb3b6212013-11-11 13:44:56 +010094 if (!shost->last_reset || shost->eh_deadline == -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +020095 return 0;
96
Hannes Reinecke76ad3e52013-11-11 13:44:55 +010097 /*
98 * 32bit accesses are guaranteed to be atomic
99 * (on all supported architectures), so instead
100 * of using a spinlock we can as well double check
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100101 * if eh_deadline has been set to 'off' during the
Hannes Reinecke76ad3e52013-11-11 13:44:55 +0100102 * time_before call.
103 */
104 if (time_before(jiffies, shost->last_reset + shost->eh_deadline) &&
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100105 shost->eh_deadline > -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200106 return 0;
107
108 return 1;
109}
110
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100112 * scmd_eh_abort_handler - Handle command aborts
113 * @work: command to be aborted.
114 */
115void
116scmd_eh_abort_handler(struct work_struct *work)
117{
118 struct scsi_cmnd *scmd =
119 container_of(work, struct scsi_cmnd, abort_work.work);
120 struct scsi_device *sdev = scmd->device;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100121 int rtn;
122
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100123 if (scsi_host_eh_past_deadline(sdev->host)) {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100124 SCSI_LOG_ERROR_RECOVERY(3,
125 scmd_printk(KERN_INFO, scmd,
126 "scmd %p eh timeout, not aborting\n",
127 scmd));
128 } else {
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100129 SCSI_LOG_ERROR_RECOVERY(3,
130 scmd_printk(KERN_INFO, scmd,
131 "aborting command %p\n", scmd));
132 rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd);
133 if (rtn == SUCCESS) {
134 scmd->result |= DID_TIME_OUT << 16;
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100135 if (scsi_host_eh_past_deadline(sdev->host)) {
136 SCSI_LOG_ERROR_RECOVERY(3,
137 scmd_printk(KERN_INFO, scmd,
138 "scmd %p eh timeout, "
139 "not retrying aborted "
140 "command\n", scmd));
141 } else if (!scsi_noretry_cmd(scmd) &&
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100142 (++scmd->retries <= scmd->allowed)) {
143 SCSI_LOG_ERROR_RECOVERY(3,
144 scmd_printk(KERN_WARNING, scmd,
145 "scmd %p retry "
146 "aborted command\n", scmd));
147 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,
152 "scmd %p finish "
153 "aborted command\n", scmd));
154 scsi_finish_command(scmd);
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100155 return;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100156 }
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100157 } else {
158 SCSI_LOG_ERROR_RECOVERY(3,
159 scmd_printk(KERN_INFO, scmd,
160 "scmd %p abort failed, rtn %d\n",
161 scmd, rtn));
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,
168 "scmd %p terminate "
169 "aborted command\n", scmd));
170 scmd->result |= DID_TIME_OUT << 16;
171 scsi_finish_command(scmd);
172 }
173}
174
175/**
176 * scsi_abort_command - schedule a command abort
177 * @scmd: scmd to abort.
178 *
179 * We only need to abort commands after a command timeout
180 */
181static int
182scsi_abort_command(struct scsi_cmnd *scmd)
183{
184 struct scsi_device *sdev = scmd->device;
185 struct Scsi_Host *shost = sdev->host;
186 unsigned long flags;
187
188 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
189 /*
190 * Retry after abort failed, escalate to next level.
191 */
Hannes Reinecke7daf4802014-03-31 16:37:34 +0200192 scmd->eh_eflags &= ~SCSI_EH_ABORT_SCHEDULED;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100193 SCSI_LOG_ERROR_RECOVERY(3,
194 scmd_printk(KERN_INFO, scmd,
195 "scmd %p previous abort failed\n", scmd));
196 cancel_delayed_work(&scmd->abort_work);
197 return FAILED;
198 }
199
200 /*
201 * Do not try a command abort if
202 * SCSI EH has already started.
203 */
204 spin_lock_irqsave(shost->host_lock, flags);
205 if (scsi_host_in_recovery(shost)) {
206 spin_unlock_irqrestore(shost->host_lock, flags);
207 SCSI_LOG_ERROR_RECOVERY(3,
208 scmd_printk(KERN_INFO, scmd,
209 "scmd %p not aborting, host in recovery\n",
210 scmd));
211 return FAILED;
212 }
213
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100214 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100215 shost->last_reset = jiffies;
216 spin_unlock_irqrestore(shost->host_lock, flags);
217
218 scmd->eh_eflags |= SCSI_EH_ABORT_SCHEDULED;
219 SCSI_LOG_ERROR_RECOVERY(3,
220 scmd_printk(KERN_INFO, scmd,
221 "scmd %p abort scheduled\n", scmd));
222 queue_delayed_work(shost->tmf_work_q, &scmd->abort_work, HZ / 100);
223 return SUCCESS;
224}
225
226/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 * scsi_eh_scmd_add - add scsi cmd to error handling.
228 * @scmd: scmd to run eh on.
229 * @eh_flag: optional SCSI_EH flag.
230 *
231 * Return value:
232 * 0 on failure.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800233 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
235{
236 struct Scsi_Host *shost = scmd->device->host;
237 unsigned long flags;
James Bottomley939647e2005-09-18 15:05:20 -0500238 int ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
James Bottomley3ed7a472005-09-19 09:50:04 -0500240 if (!shost->ehandler)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 return 0;
242
243 spin_lock_irqsave(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500244 if (scsi_host_set_state(shost, SHOST_RECOVERY))
245 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
246 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100248 if (shost->eh_deadline != -1 && !shost->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200249 shost->last_reset = jiffies;
250
James Bottomley939647e2005-09-18 15:05:20 -0500251 ret = 1;
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100252 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED)
253 eh_flag &= ~SCSI_EH_CANCEL_CMD;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200254 scmd->eh_eflags |= eh_flag;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 shost->host_failed++;
257 scsi_eh_wakeup(shost);
James Bottomley939647e2005-09-18 15:05:20 -0500258 out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 spin_unlock_irqrestore(shost->host_lock, flags);
James Bottomley939647e2005-09-18 15:05:20 -0500260 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261}
262
263/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 * scsi_times_out - Timeout function for normal scsi commands.
Jens Axboe242f9dc2008-09-14 05:55:09 -0700265 * @req: request that is timing out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 *
267 * Notes:
268 * We do not need to lock this. There is the potential for a race
269 * only in that the normal completion handling might run, but if the
270 * normal completion function determines that the timer has already
271 * fired, then it mustn't do anything.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800272 */
Jens Axboe242f9dc2008-09-14 05:55:09 -0700273enum blk_eh_timer_return scsi_times_out(struct request *req)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274{
Jens Axboe242f9dc2008-09-14 05:55:09 -0700275 struct scsi_cmnd *scmd = req->special;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700276 enum blk_eh_timer_return rtn = BLK_EH_NOT_HANDLED;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100277 struct Scsi_Host *host = scmd->device->host;
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500278
Kei Tokunagabf816232010-04-01 20:41:40 +0900279 trace_scsi_dispatch_cmd_timeout(scmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 scsi_log_completion(scmd, TIMEOUT_ERROR);
281
Ren Mingxinbb3b6212013-11-11 13:44:56 +0100282 if (host->eh_deadline != -1 && !host->last_reset)
Hannes Reineckeb4562022013-10-23 10:51:21 +0200283 host->last_reset = jiffies;
284
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100285 if (host->transportt->eh_timed_out)
286 rtn = host->transportt->eh_timed_out(scmd);
287 else if (host->hostt->eh_timed_out)
288 rtn = host->hostt->eh_timed_out(scmd);
James Bottomley6c5f8ce2007-03-16 17:44:41 -0500289
Hannes Reineckee494f6a2013-11-11 13:44:54 +0100290 if (rtn == BLK_EH_NOT_HANDLED && !host->hostt->no_async_abort)
291 if (scsi_abort_command(scmd) == SUCCESS)
292 return BLK_EH_NOT_HANDLED;
293
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500294 scmd->result |= DID_TIME_OUT << 16;
295
Christoph Hellwigfa990782008-11-05 15:03:15 +0100296 if (unlikely(rtn == BLK_EH_NOT_HANDLED &&
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -0500297 !scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD)))
Christoph Hellwigfa990782008-11-05 15:03:15 +0100298 rtn = BLK_EH_HANDLED;
Jens Axboe242f9dc2008-09-14 05:55:09 -0700299
Christoph Hellwigfa990782008-11-05 15:03:15 +0100300 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301}
302
303/**
304 * scsi_block_when_processing_errors - Prevent cmds from being queued.
305 * @sdev: Device on which we are performing recovery.
306 *
307 * Description:
308 * We block until the host is out of error recovery, and then check to
309 * see whether the host or the device is offline.
310 *
311 * Return value:
312 * 0 when dev was taken offline by error recovery. 1 OK to proceed.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800313 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314int scsi_block_when_processing_errors(struct scsi_device *sdev)
315{
316 int online;
317
James Bottomley939647e2005-09-18 15:05:20 -0500318 wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320 online = scsi_device_online(sdev);
321
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700322 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 online));
324
325 return online;
326}
327EXPORT_SYMBOL(scsi_block_when_processing_errors);
328
329#ifdef CONFIG_SCSI_LOGGING
330/**
331 * scsi_eh_prt_fail_stats - Log info on failures.
332 * @shost: scsi host being recovered.
333 * @work_q: Queue of scsi cmds to process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800334 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
336 struct list_head *work_q)
337{
338 struct scsi_cmnd *scmd;
339 struct scsi_device *sdev;
340 int total_failures = 0;
341 int cmd_failed = 0;
342 int cmd_cancel = 0;
343 int devices_failed = 0;
344
345 shost_for_each_device(sdev, shost) {
346 list_for_each_entry(scmd, work_q, eh_entry) {
347 if (scmd->device == sdev) {
348 ++total_failures;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +0200349 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 ++cmd_cancel;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100351 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 ++cmd_failed;
353 }
354 }
355
356 if (cmd_cancel || cmd_failed) {
357 SCSI_LOG_ERROR_RECOVERY(3,
James Bottomley9ccfc752005-10-02 11:45:08 -0500358 sdev_printk(KERN_INFO, sdev,
359 "%s: cmds failed: %d, cancel: %d\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700360 __func__, cmd_failed,
James Bottomley9ccfc752005-10-02 11:45:08 -0500361 cmd_cancel));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 cmd_cancel = 0;
363 cmd_failed = 0;
364 ++devices_failed;
365 }
366 }
367
368 SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
369 " devices require eh work\n",
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100370 total_failures, devices_failed));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372#endif
373
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400374 /**
375 * scsi_report_lun_change - Set flag on all *other* devices on the same target
376 * to indicate that a UNIT ATTENTION is expected.
377 * @sdev: Device reporting the UNIT ATTENTION
378 */
379static void scsi_report_lun_change(struct scsi_device *sdev)
380{
381 sdev->sdev_target->expecting_lun_change = 1;
382}
383
384/**
385 * scsi_report_sense - Examine scsi sense information and log messages for
386 * certain conditions, also issue uevents for some of them.
387 * @sdev: Device reporting the sense code
388 * @sshdr: sshdr to be examined
389 */
390static void scsi_report_sense(struct scsi_device *sdev,
391 struct scsi_sense_hdr *sshdr)
392{
393 enum scsi_device_event evt_type = SDEV_EVT_MAXBITS; /* i.e. none */
394
395 if (sshdr->sense_key == UNIT_ATTENTION) {
396 if (sshdr->asc == 0x3f && sshdr->ascq == 0x03) {
397 evt_type = SDEV_EVT_INQUIRY_CHANGE_REPORTED;
398 sdev_printk(KERN_WARNING, sdev,
399 "Inquiry data has changed");
400 } else if (sshdr->asc == 0x3f && sshdr->ascq == 0x0e) {
401 evt_type = SDEV_EVT_LUN_CHANGE_REPORTED;
402 scsi_report_lun_change(sdev);
403 sdev_printk(KERN_WARNING, sdev,
404 "Warning! Received an indication that the "
405 "LUN assignments on this target have "
406 "changed. The Linux SCSI layer does not "
407 "automatically remap LUN assignments.\n");
408 } else if (sshdr->asc == 0x3f)
409 sdev_printk(KERN_WARNING, sdev,
410 "Warning! Received an indication that the "
411 "operating parameters on this target have "
412 "changed. The Linux SCSI layer does not "
413 "automatically adjust these parameters.\n");
414
415 if (sshdr->asc == 0x38 && sshdr->ascq == 0x07) {
416 evt_type = SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED;
417 sdev_printk(KERN_WARNING, sdev,
418 "Warning! Received an indication that the "
419 "LUN reached a thin provisioning soft "
420 "threshold.\n");
421 }
422
423 if (sshdr->asc == 0x2a && sshdr->ascq == 0x01) {
424 evt_type = SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED;
425 sdev_printk(KERN_WARNING, sdev,
426 "Mode parameters changed");
427 } else if (sshdr->asc == 0x2a && sshdr->ascq == 0x09) {
428 evt_type = SDEV_EVT_CAPACITY_CHANGE_REPORTED;
429 sdev_printk(KERN_WARNING, sdev,
430 "Capacity data has changed");
431 } else if (sshdr->asc == 0x2a)
432 sdev_printk(KERN_WARNING, sdev,
433 "Parameters changed");
434 }
435
436 if (evt_type != SDEV_EVT_MAXBITS) {
437 set_bit(evt_type, sdev->pending_events);
438 schedule_work(&sdev->event_work);
439 }
440}
441
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442/**
443 * scsi_check_sense - Examine scsi cmd sense
444 * @scmd: Cmd to have sense checked.
445 *
446 * Return value:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200447 * SUCCESS or FAILED or NEEDS_RETRY or ADD_TO_MLQUEUE
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 *
449 * Notes:
450 * When a deferred error is detected the current command has
451 * not been executed and needs retrying.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800452 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453static int scsi_check_sense(struct scsi_cmnd *scmd)
454{
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700455 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 struct scsi_sense_hdr sshdr;
457
458 if (! scsi_command_normalize_sense(scmd, &sshdr))
459 return FAILED; /* no valid sense data */
460
James Bottomley14216562012-07-25 23:55:55 +0400461 if (scmd->cmnd[0] == TEST_UNIT_READY && scmd->scsi_done != scsi_eh_done)
462 /*
463 * nasty: for mid-layer issued TURs, we need to return the
464 * actual sense data without any recovery attempt. For eh
465 * issued ones, we need to try to recover and interpret
466 */
467 return SUCCESS;
468
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400469 scsi_report_sense(sdev, &sshdr);
470
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 if (scsi_sense_is_deferred(&sshdr))
472 return NEEDS_RETRY;
473
Chandra Seetharamana6a8d9f2008-05-01 14:49:46 -0700474 if (sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh &&
475 sdev->scsi_dh_data->scsi_dh->check_sense) {
476 int rc;
477
478 rc = sdev->scsi_dh_data->scsi_dh->check_sense(sdev, &sshdr);
479 if (rc != SCSI_RETURN_NOT_HANDLED)
480 return rc;
481 /* handler does not care. Drop down to default handling */
482 }
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 /*
485 * Previous logic looked for FILEMARK, EOM or ILI which are
486 * mainly associated with tapes and returned SUCCESS.
487 */
488 if (sshdr.response_code == 0x70) {
489 /* fixed format */
490 if (scmd->sense_buffer[2] & 0xe0)
491 return SUCCESS;
492 } else {
493 /*
494 * descriptor format: look for "stream commands sense data
495 * descriptor" (see SSC-3). Assume single sense data
496 * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
497 */
498 if ((sshdr.additional_length > 3) &&
499 (scmd->sense_buffer[8] == 0x4) &&
500 (scmd->sense_buffer[11] & 0xe0))
501 return SUCCESS;
502 }
503
504 switch (sshdr.sense_key) {
505 case NO_SENSE:
506 return SUCCESS;
507 case RECOVERED_ERROR:
508 return /* soft_error */ SUCCESS;
509
510 case ABORTED_COMMAND:
Martin K. Petersen511e44f2008-07-17 04:28:33 -0400511 if (sshdr.asc == 0x10) /* DIF */
512 return SUCCESS;
513
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 return NEEDS_RETRY;
515 case NOT_READY:
516 case UNIT_ATTENTION:
517 /*
518 * if we are expecting a cc/ua because of a bus reset that we
519 * performed, treat this just as a retry. otherwise this is
520 * information that we should pass up to the upper-level driver
521 * so that we can deal with it there.
522 */
523 if (scmd->device->expecting_cc_ua) {
TARUISI Hiroakidfcf7772011-08-11 20:25:20 +0900524 /*
525 * Because some device does not queue unit
526 * attentions correctly, we carefully check
527 * additional sense code and qualifier so as
528 * not to squash media change unit attention.
529 */
530 if (sshdr.asc != 0x28 || sshdr.ascq != 0x00) {
531 scmd->device->expecting_cc_ua = 0;
532 return NEEDS_RETRY;
533 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 }
535 /*
Ewan D. Milne279afdf2013-08-08 15:07:48 -0400536 * we might also expect a cc/ua if another LUN on the target
537 * reported a UA with an ASC/ASCQ of 3F 0E -
538 * REPORTED LUNS DATA HAS CHANGED.
539 */
540 if (scmd->device->sdev_target->expecting_lun_change &&
541 sshdr.asc == 0x3f && sshdr.ascq == 0x0e)
542 return NEEDS_RETRY;
543 /*
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100544 * if the device is in the process of becoming ready, we
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 * should retry.
546 */
547 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
548 return NEEDS_RETRY;
549 /*
550 * if the device is not started, we need to wake
551 * the error handler to start the motor
552 */
553 if (scmd->device->allow_restart &&
554 (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
555 return FAILED;
Christoph Hellwig02e031c2010-11-10 14:54:09 +0100556 /*
557 * Pass the UA upwards for a determination in the completion
558 * functions.
559 */
560 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100562 /* these are not supported */
Hannes Reineckea9d6ceb82013-07-01 15:16:25 +0200563 case DATA_PROTECT:
564 if (sshdr.asc == 0x27 && sshdr.ascq == 0x07) {
565 /* Thin provisioning hard threshold reached */
566 set_host_byte(scmd, DID_ALLOC_FAILURE);
567 return SUCCESS;
568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 case COPY_ABORTED:
570 case VOLUME_OVERFLOW:
571 case MISCOMPARE:
Hannes Reinecke63583cc2011-01-18 10:13:11 +0100572 case BLANK_CHECK:
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200573 set_host_byte(scmd, DID_TARGET_FAILURE);
574 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575
576 case MEDIUM_ERROR:
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800577 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
578 sshdr.asc == 0x13 || /* AMNF DATA FIELD */
579 sshdr.asc == 0x14) { /* RECORD NOT FOUND */
Hannes Reinecke7e782af2013-07-01 15:16:26 +0200580 set_host_byte(scmd, DID_MEDIUM_ERROR);
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200581 return SUCCESS;
Luben Tuikovfd1b4942006-11-29 19:45:23 -0800582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 return NEEDS_RETRY;
584
585 case HARDWARE_ERROR:
586 if (scmd->device->retry_hwerror)
Mike Andersonbb0003c2008-08-12 12:11:58 -0700587 return ADD_TO_MLQUEUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 else
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200589 set_host_byte(scmd, DID_TARGET_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 case ILLEGAL_REQUEST:
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500592 if (sshdr.asc == 0x20 || /* Invalid command operation code */
593 sshdr.asc == 0x21 || /* Logical block address out of range */
594 sshdr.asc == 0x24 || /* Invalid field in cdb */
595 sshdr.asc == 0x26) { /* Parameter value invalid */
Hannes Reinecke87f14e62013-07-01 15:16:24 +0200596 set_host_byte(scmd, DID_TARGET_FAILURE);
Mike Snitzer47ac56d2012-02-13 18:35:11 -0500597 }
598 return SUCCESS;
599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 default:
601 return SUCCESS;
602 }
603}
604
Vasu Dev4a840672009-10-22 15:46:33 -0700605static void scsi_handle_queue_ramp_up(struct scsi_device *sdev)
606{
607 struct scsi_host_template *sht = sdev->host->hostt;
608 struct scsi_device *tmp_sdev;
609
610 if (!sht->change_queue_depth ||
611 sdev->queue_depth >= sdev->max_queue_depth)
612 return;
613
614 if (time_before(jiffies,
615 sdev->last_queue_ramp_up + sdev->queue_ramp_up_period))
616 return;
617
618 if (time_before(jiffies,
619 sdev->last_queue_full_time + sdev->queue_ramp_up_period))
620 return;
621
622 /*
623 * Walk all devices of a target and do
624 * ramp up on them.
625 */
626 shost_for_each_device(tmp_sdev, sdev->host) {
627 if (tmp_sdev->channel != sdev->channel ||
628 tmp_sdev->id != sdev->id ||
629 tmp_sdev->queue_depth == sdev->max_queue_depth)
630 continue;
631 /*
632 * call back into LLD to increase queue_depth by one
633 * with ramp up reason code.
634 */
635 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth + 1,
636 SCSI_QDEPTH_RAMP_UP);
637 sdev->last_queue_ramp_up = jiffies;
638 }
639}
640
Mike Christie42a6a912009-10-15 17:46:44 -0700641static void scsi_handle_queue_full(struct scsi_device *sdev)
642{
643 struct scsi_host_template *sht = sdev->host->hostt;
644 struct scsi_device *tmp_sdev;
645
646 if (!sht->change_queue_depth)
647 return;
648
649 shost_for_each_device(tmp_sdev, sdev->host) {
650 if (tmp_sdev->channel != sdev->channel ||
651 tmp_sdev->id != sdev->id)
652 continue;
653 /*
654 * We do not know the number of commands that were at
655 * the device when we got the queue full so we start
656 * from the highest possible value and work our way down.
657 */
658 sht->change_queue_depth(tmp_sdev, tmp_sdev->queue_depth - 1,
659 SCSI_QDEPTH_QFULL);
660 }
661}
662
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663/**
664 * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
665 * @scmd: SCSI cmd to examine.
666 *
667 * Notes:
668 * This is *only* called when we are examining the status of commands
669 * queued during error recovery. the main difference here is that we
670 * don't allow for the possibility of retries here, and we are a lot
671 * more restrictive about what we consider acceptable.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800672 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
674{
675 /*
676 * first check the host byte, to see if there is anything in there
677 * that would indicate what we need to do.
678 */
679 if (host_byte(scmd->result) == DID_RESET) {
680 /*
681 * rats. we are already in the error handler, so we now
682 * get to try and figure out what to do next. if the sense
683 * is valid, we have a pretty good idea of what to do.
684 * if not, we mark it as FAILED.
685 */
686 return scsi_check_sense(scmd);
687 }
688 if (host_byte(scmd->result) != DID_OK)
689 return FAILED;
690
691 /*
692 * next, check the message byte.
693 */
694 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
695 return FAILED;
696
697 /*
698 * now, check the status byte to see if this indicates
699 * anything special.
700 */
701 switch (status_byte(scmd->result)) {
702 case GOOD:
Vasu Dev4a840672009-10-22 15:46:33 -0700703 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 case COMMAND_TERMINATED:
705 return SUCCESS;
706 case CHECK_CONDITION:
707 return scsi_check_sense(scmd);
708 case CONDITION_GOOD:
709 case INTERMEDIATE_GOOD:
710 case INTERMEDIATE_C_GOOD:
711 /*
712 * who knows? FIXME(eric)
713 */
714 return SUCCESS;
Michael Reed5f91bb02009-08-10 11:59:28 -0500715 case RESERVATION_CONFLICT:
James Bottomley67110df2010-08-06 15:17:24 -0500716 if (scmd->cmnd[0] == TEST_UNIT_READY)
717 /* it is a success, we probed the device and
718 * found it */
719 return SUCCESS;
720 /* otherwise, we failed to send the command */
721 return FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -0700723 scsi_handle_queue_full(scmd->device);
724 /* fall through */
725 case BUSY:
Hannes Reinecke3eb3a922010-07-29 10:10:16 +0200726 return NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 default:
728 return FAILED;
729 }
730 return FAILED;
731}
732
733/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 * scsi_eh_done - Completion function for error handling.
735 * @scmd: Cmd that is done.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800736 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737static void scsi_eh_done(struct scsi_cmnd *scmd)
738{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100739 struct completion *eh_action;
Michael Reed85631672005-12-07 21:46:27 -0600740
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +0100741 SCSI_LOG_ERROR_RECOVERY(3,
742 printk("%s scmd: %p result: %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700743 __func__, scmd, scmd->result));
Michael Reed85631672005-12-07 21:46:27 -0600744
745 eh_action = scmd->device->host->eh_action;
746 if (eh_action)
747 complete(eh_action);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748}
749
750/**
Brian King292148f2007-01-30 17:51:17 -0600751 * scsi_try_host_reset - ask host adapter to reset itself
Geert Uytterhoevenc2b3ebd2013-05-17 13:22:29 +0200752 * @scmd: SCSI cmd to send host reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800753 */
Brian King292148f2007-01-30 17:51:17 -0600754static int scsi_try_host_reset(struct scsi_cmnd *scmd)
755{
756 unsigned long flags;
757 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100758 struct Scsi_Host *host = scmd->device->host;
759 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600760
761 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700762 __func__));
Brian King292148f2007-01-30 17:51:17 -0600763
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100764 if (!hostt->eh_host_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600765 return FAILED;
766
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100767 rtn = hostt->eh_host_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600768
769 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100770 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600771 ssleep(HOST_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100772 spin_lock_irqsave(host->host_lock, flags);
773 scsi_report_bus_reset(host, scmd_channel(scmd));
774 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600775 }
776
777 return rtn;
778}
779
780/**
781 * scsi_try_bus_reset - ask host to perform a bus reset
782 * @scmd: SCSI cmd to send bus reset.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800783 */
Brian King292148f2007-01-30 17:51:17 -0600784static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
785{
786 unsigned long flags;
787 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100788 struct Scsi_Host *host = scmd->device->host;
789 struct scsi_host_template *hostt = host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600790
791 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -0700792 __func__));
Brian King292148f2007-01-30 17:51:17 -0600793
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100794 if (!hostt->eh_bus_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600795 return FAILED;
796
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100797 rtn = hostt->eh_bus_reset_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600798
799 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100800 if (!hostt->skip_settle_delay)
Brian King292148f2007-01-30 17:51:17 -0600801 ssleep(BUS_RESET_SETTLE_TIME);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100802 spin_lock_irqsave(host->host_lock, flags);
803 scsi_report_bus_reset(host, scmd_channel(scmd));
804 spin_unlock_irqrestore(host->host_lock, flags);
Brian King292148f2007-01-30 17:51:17 -0600805 }
806
807 return rtn;
808}
809
Mike Christie30bd7df2008-02-29 18:25:19 -0600810static void __scsi_report_device_reset(struct scsi_device *sdev, void *data)
811{
812 sdev->was_reset = 1;
813 sdev->expecting_cc_ua = 1;
814}
815
816/**
817 * scsi_try_target_reset - Ask host to perform a target reset
818 * @scmd: SCSI cmd used to send a target reset
819 *
820 * Notes:
821 * There is no timeout for this operation. if this operation is
822 * unreliable for a given host, then the host itself needs to put a
823 * timer on it, and set the host back to a consistent state prior to
824 * returning.
825 */
826static int scsi_try_target_reset(struct scsi_cmnd *scmd)
827{
828 unsigned long flags;
829 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100830 struct Scsi_Host *host = scmd->device->host;
831 struct scsi_host_template *hostt = host->hostt;
Mike Christie30bd7df2008-02-29 18:25:19 -0600832
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100833 if (!hostt->eh_target_reset_handler)
Mike Christie30bd7df2008-02-29 18:25:19 -0600834 return FAILED;
835
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100836 rtn = hostt->eh_target_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600837 if (rtn == SUCCESS) {
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100838 spin_lock_irqsave(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600839 __starget_for_each_device(scsi_target(scmd->device), NULL,
840 __scsi_report_device_reset);
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100841 spin_unlock_irqrestore(host->host_lock, flags);
Mike Christie30bd7df2008-02-29 18:25:19 -0600842 }
843
844 return rtn;
845}
846
Brian King292148f2007-01-30 17:51:17 -0600847/**
848 * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
849 * @scmd: SCSI cmd used to send BDR
850 *
851 * Notes:
852 * There is no timeout for this operation. if this operation is
853 * unreliable for a given host, then the host itself needs to put a
854 * timer on it, and set the host back to a consistent state prior to
855 * returning.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800856 */
Brian King292148f2007-01-30 17:51:17 -0600857static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
858{
859 int rtn;
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100860 struct scsi_host_template *hostt = scmd->device->host->hostt;
Brian King292148f2007-01-30 17:51:17 -0600861
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100862 if (!hostt->eh_device_reset_handler)
Brian King292148f2007-01-30 17:51:17 -0600863 return FAILED;
864
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100865 rtn = hostt->eh_device_reset_handler(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -0600866 if (rtn == SUCCESS)
867 __scsi_report_device_reset(scmd->device, NULL);
Brian King292148f2007-01-30 17:51:17 -0600868 return rtn;
869}
870
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100871static int scsi_try_to_abort_cmd(struct scsi_host_template *hostt, struct scsi_cmnd *scmd)
Brian King292148f2007-01-30 17:51:17 -0600872{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100873 if (!hostt->eh_abort_handler)
Brian King292148f2007-01-30 17:51:17 -0600874 return FAILED;
875
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100876 return hostt->eh_abort_handler(scmd);
Brian King292148f2007-01-30 17:51:17 -0600877}
878
Brian King292148f2007-01-30 17:51:17 -0600879static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
880{
Jesper Juhl0bf8c862011-03-21 20:47:31 +0100881 if (scsi_try_to_abort_cmd(scmd->device->host->hostt, scmd) != SUCCESS)
Brian King292148f2007-01-30 17:51:17 -0600882 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
Mike Christie30bd7df2008-02-29 18:25:19 -0600883 if (scsi_try_target_reset(scmd) != SUCCESS)
884 if (scsi_try_bus_reset(scmd) != SUCCESS)
885 scsi_try_host_reset(scmd);
Brian King292148f2007-01-30 17:51:17 -0600886}
887
888/**
Santosh Y3b729f72012-04-08 18:47:09 +0530889 * scsi_eh_prep_cmnd - Save a scsi command info as part of error recovery
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100890 * @scmd: SCSI command structure to hijack
Boaz Harroshe1c23462007-10-08 16:36:45 +0200891 * @ses: structure to save restore information
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200892 * @cmnd: CDB to send. Can be NULL if no new cmnd is needed
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300893 * @cmnd_size: size in bytes of @cmnd (must be <= BLK_MAX_CDB)
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200894 * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
Christoph Hellwig2dc611d2006-11-04 20:04:21 +0100895 *
Boaz Harroshe1c23462007-10-08 16:36:45 +0200896 * This function is used to save a scsi command information before re-execution
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200897 * as part of the error recovery process. If @sense_bytes is 0 the command
898 * sent must be one that does not transfer any data. If @sense_bytes != 0
899 * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
900 * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800901 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200902void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
903 unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904{
f59114b2005-04-17 15:00:23 -0500905 struct scsi_device *sdev = scmd->device;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Christoph Hellwig631c2282006-07-08 20:42:15 +0200907 /*
908 * We need saved copies of a number of fields - this is because
909 * error handling may need to overwrite these with different values
910 * to run different commands, and once error handling is complete,
911 * we will need to restore these values prior to running the actual
912 * command.
913 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200914 ses->cmd_len = scmd->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300915 ses->cmnd = scmd->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200916 ses->data_direction = scmd->sc_data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200917 ses->sdb = scmd->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200918 ses->next_rq = scmd->request->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200919 ses->result = scmd->result;
Alan Stern12265702008-07-21 10:25:52 -0400920 ses->underflow = scmd->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400921 ses->prot_op = scmd->prot_op;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200922
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400923 scmd->prot_op = SCSI_PROT_NORMAL;
James Bottomleyc69e6f82014-04-10 13:36:11 -0700924 scmd->eh_eflags = 0;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300925 scmd->cmnd = ses->eh_cmnd;
926 memset(scmd->cmnd, 0, BLK_MAX_CDB);
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200927 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200928 scmd->request->next_rq = NULL;
Alan Stern644373a2014-03-28 10:51:15 -0700929 scmd->result = 0;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200930
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200931 if (sense_bytes) {
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200932 scmd->sdb.length = min_t(unsigned, SCSI_SENSE_BUFFERSIZE,
933 sense_bytes);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200934 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200935 scmd->sdb.length);
936 scmd->sdb.table.sgl = &ses->sense_sgl;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200937 scmd->sc_data_direction = DMA_FROM_DEVICE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200938 scmd->sdb.table.nents = 1;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200939 scmd->cmnd[0] = REQUEST_SENSE;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200940 scmd->cmnd[4] = scmd->sdb.length;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200941 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200942 } else {
Christoph Hellwig631c2282006-07-08 20:42:15 +0200943 scmd->sc_data_direction = DMA_NONE;
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200944 if (cmnd) {
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300945 BUG_ON(cmnd_size > BLK_MAX_CDB);
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200946 memcpy(scmd->cmnd, cmnd, cmnd_size);
947 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
948 }
Christoph Hellwig631c2282006-07-08 20:42:15 +0200949 }
950
951 scmd->underflow = 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +0200952
Boaz Harrosh55db6c12007-10-08 16:35:19 +0200953 if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
f59114b2005-04-17 15:00:23 -0500955 (sdev->lun << 5 & 0xe0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Christoph Hellwig631c2282006-07-08 20:42:15 +0200957 /*
958 * Zero the sense buffer. The scsi spec mandates that any
959 * untransferred sense data should be interpreted as being zero.
960 */
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +0900961 memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
Boaz Harroshe1c23462007-10-08 16:36:45 +0200962}
963EXPORT_SYMBOL(scsi_eh_prep_cmnd);
Christoph Hellwig631c2282006-07-08 20:42:15 +0200964
Boaz Harroshe1c23462007-10-08 16:36:45 +0200965/**
Santosh Y3b729f72012-04-08 18:47:09 +0530966 * scsi_eh_restore_cmnd - Restore a scsi command info as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +0200967 * @scmd: SCSI command structure to restore
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200968 * @ses: saved information from a coresponding call to scsi_eh_prep_cmnd
Boaz Harroshe1c23462007-10-08 16:36:45 +0200969 *
Bartlomiej Zolnierkiewicz477e6082009-04-27 20:54:22 +0200970 * Undo any damage done by above scsi_eh_prep_cmnd().
Randy Dunlapdc8875e2007-11-15 15:42:30 -0800971 */
Boaz Harroshe1c23462007-10-08 16:36:45 +0200972void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
973{
974 /*
975 * Restore original data
976 */
977 scmd->cmd_len = ses->cmd_len;
Boaz Harrosh64a87b22008-04-30 11:19:47 +0300978 scmd->cmnd = ses->cmnd;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200979 scmd->sc_data_direction = ses->data_direction;
Boaz Harrosh30b0c372007-12-13 13:47:40 +0200980 scmd->sdb = ses->sdb;
Boaz Harrosh6f9a35e2007-12-13 13:50:53 +0200981 scmd->request->next_rq = ses->next_rq;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200982 scmd->result = ses->result;
Alan Stern12265702008-07-21 10:25:52 -0400983 scmd->underflow = ses->underflow;
Martin K. Petersendb007fc2008-07-17 04:28:31 -0400984 scmd->prot_op = ses->prot_op;
Boaz Harroshe1c23462007-10-08 16:36:45 +0200985}
986EXPORT_SYMBOL(scsi_eh_restore_cmnd);
987
988/**
Santosh Y3b729f72012-04-08 18:47:09 +0530989 * scsi_send_eh_cmnd - submit a scsi command as part of error recovery
Boaz Harroshe1c23462007-10-08 16:36:45 +0200990 * @scmd: SCSI command structure to hijack
991 * @cmnd: CDB to send
992 * @cmnd_size: size in bytes of @cmnd
993 * @timeout: timeout for this request
994 * @sense_bytes: size of sense data to copy or 0
995 *
996 * This function is used to send a scsi command down to a target device
997 * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
998 *
999 * Return value:
1000 * SUCCESS or FAILED or NEEDS_RETRY
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001001 */
Boaz Harroshe1c23462007-10-08 16:36:45 +02001002static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
1003 int cmnd_size, int timeout, unsigned sense_bytes)
1004{
1005 struct scsi_device *sdev = scmd->device;
1006 struct Scsi_Host *shost = sdev->host;
1007 DECLARE_COMPLETION_ONSTACK(done);
Hannes Reineckefc736482013-04-25 08:10:00 +02001008 unsigned long timeleft = timeout;
Boaz Harroshe1c23462007-10-08 16:36:45 +02001009 struct scsi_eh_save ses;
Hannes Reineckefc736482013-04-25 08:10:00 +02001010 const unsigned long stall_for = msecs_to_jiffies(100);
Boaz Harroshe1c23462007-10-08 16:36:45 +02001011 int rtn;
1012
Hannes Reineckefc736482013-04-25 08:10:00 +02001013retry:
Boaz Harroshe1c23462007-10-08 16:36:45 +02001014 scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001015 shost->eh_action = &done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 scsi_log_send(scmd);
Jeff Garzikf2812332010-11-16 02:10:29 -05001018 scmd->scsi_done = scsi_eh_done;
Hannes Reineckefc736482013-04-25 08:10:00 +02001019 rtn = shost->hostt->queuecommand(shost, scmd);
1020 if (rtn) {
1021 if (timeleft > stall_for) {
1022 scsi_eh_restore_cmnd(scmd, &ses);
1023 timeleft -= stall_for;
1024 msleep(jiffies_to_msecs(stall_for));
1025 goto retry;
1026 }
1027 /* signal not to enter either branch of the if () below */
1028 timeleft = 0;
1029 rtn = NEEDS_RETRY;
1030 } else {
1031 timeleft = wait_for_completion_timeout(&done, timeout);
1032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
f59114b2005-04-17 15:00:23 -05001034 shost->eh_action = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
Hannes Reineckefc736482013-04-25 08:10:00 +02001036 scsi_log_completion(scmd, rtn);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001038 SCSI_LOG_ERROR_RECOVERY(3,
1039 printk("%s: scmd: %p, timeleft: %ld\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001040 __func__, scmd, timeleft));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 /*
Hannes Reineckefc736482013-04-25 08:10:00 +02001043 * If there is time left scsi_eh_done got called, and we will examine
1044 * the actual status codes to see whether the command actually did
1045 * complete normally, else if we have a zero return and no time left,
1046 * the command must still be pending, so abort it and return FAILED.
1047 * If we never actually managed to issue the command, because
1048 * ->queuecommand() kept returning non zero, use the rtn = FAILED
1049 * value above (so don't execute either branch of the if)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 */
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001051 if (timeleft) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 rtn = scsi_eh_completed_normally(scmd);
1053 SCSI_LOG_ERROR_RECOVERY(3,
1054 printk("%s: scsi_eh_completed_normally %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001055 __func__, rtn));
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 switch (rtn) {
1058 case SUCCESS:
1059 case NEEDS_RETRY:
1060 case FAILED:
1061 break;
Hannes Reinecke6e883b02009-09-17 17:00:26 +02001062 case ADD_TO_MLQUEUE:
1063 rtn = NEEDS_RETRY;
1064 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 default:
1066 rtn = FAILED;
1067 break;
1068 }
Hannes Reineckefc736482013-04-25 08:10:00 +02001069 } else if (!rtn) {
Brian King292148f2007-01-30 17:51:17 -06001070 scsi_abort_eh_cmnd(scmd);
Christoph Hellwig7dfdc9a2005-10-31 18:49:52 +01001071 rtn = FAILED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 }
1073
Boaz Harroshe1c23462007-10-08 16:36:45 +02001074 scsi_eh_restore_cmnd(scmd, &ses);
Martin K. Petersen18a4d0a2012-02-09 13:48:53 -05001075
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 return rtn;
1077}
1078
1079/**
1080 * scsi_request_sense - Request sense data from a particular target.
1081 * @scmd: SCSI cmd for request sense.
1082 *
1083 * Notes:
1084 * Some hosts automatically obtain this information, others require
1085 * that we obtain it on our own. This function will *not* return until
1086 * the command either times out, or it completes.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001087 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088static int scsi_request_sense(struct scsi_cmnd *scmd)
1089{
Martin K. Petersen0816c922013-05-10 10:36:04 -04001090 return scsi_send_eh_cmnd(scmd, NULL, 0, scmd->device->eh_timeout, ~0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091}
1092
James Bottomley24510792013-11-11 13:44:53 +01001093static int scsi_eh_action(struct scsi_cmnd *scmd, int rtn)
1094{
1095 if (scmd->request->cmd_type != REQ_TYPE_BLOCK_PC) {
1096 struct scsi_driver *sdrv = scsi_cmd_to_driver(scmd);
1097 if (sdrv->eh_action)
1098 rtn = sdrv->eh_action(scmd, rtn);
1099 }
1100 return rtn;
1101}
1102
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103/**
1104 * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
1105 * @scmd: Original SCSI cmd that eh has finished.
1106 * @done_q: Queue for processed commands.
1107 *
1108 * Notes:
1109 * We don't want to use the normal command completion while we are are
1110 * still handling errors - it may cause other commands to be queued,
Rob Landleyeb448202007-11-03 13:30:39 -05001111 * and that would disturb what we are doing. Thus we really want to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 * keep a list of pending commands for final completion, and once we
1113 * are ready to leave error handling we handle completion for real.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001114 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09001115void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116{
1117 scmd->device->host->host_failed--;
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001118 scmd->eh_eflags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 list_move_tail(&scmd->eh_entry, done_q);
1120}
Tejun Heo041c5fc2006-01-23 13:09:36 +09001121EXPORT_SYMBOL(scsi_eh_finish_cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
1123/**
1124 * scsi_eh_get_sense - Get device sense data.
1125 * @work_q: Queue of commands to process.
Rob Landleyeb448202007-11-03 13:30:39 -05001126 * @done_q: Queue of processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 *
1128 * Description:
1129 * See if we need to request sense information. if so, then get it
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001130 * now, so we have a better idea of what to do.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 *
1132 * Notes:
1133 * This has the unfortunate side effect that if a shost adapter does
Rob Landleyeb448202007-11-03 13:30:39 -05001134 * not automatically request sense information, we end up shutting
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 * it down before we request it.
1136 *
1137 * All drivers should request sense information internally these days,
1138 * so for now all I have to say is tough noogies if you end up in here.
1139 *
1140 * XXX: Long term this code should go away, but that needs an audit of
1141 * all LLDDs first.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001142 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001143int scsi_eh_get_sense(struct list_head *work_q,
1144 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001146 struct scsi_cmnd *scmd, *next;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001147 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 int rtn;
1149
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001150 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001151 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 SCSI_SENSE_VALID(scmd))
1153 continue;
1154
Hannes Reineckeb4562022013-10-23 10:51:21 +02001155 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001156 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001157 SCSI_LOG_ERROR_RECOVERY(3,
1158 shost_printk(KERN_INFO, shost,
1159 "skip %s, past eh deadline\n",
1160 __func__));
1161 break;
1162 }
James Bottomleyd555a2a2014-03-28 10:50:17 -07001163 if (status_byte(scmd->result) != CHECK_CONDITION)
1164 /*
1165 * don't request sense if there's no check condition
1166 * status because the error we're processing isn't one
1167 * that has a sense code (and some devices get
1168 * confused by sense requests out of the blue)
1169 */
1170 continue;
1171
Jeff Garzik3bf743e2005-10-24 18:04:06 -04001172 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
1173 "%s: requesting sense\n",
1174 current->comm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 rtn = scsi_request_sense(scmd);
1176 if (rtn != SUCCESS)
1177 continue;
1178
1179 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
1180 " result %x\n", scmd,
1181 scmd->result));
1182 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
1183
1184 rtn = scsi_decide_disposition(scmd);
1185
1186 /*
1187 * if the result was normal, then just pass it along to the
1188 * upper level.
1189 */
1190 if (rtn == SUCCESS)
1191 /* we don't want this command reissued, just
1192 * finished with the sense data, so set
1193 * retries to the max allowed to ensure it
1194 * won't get reissued */
1195 scmd->retries = scmd->allowed;
1196 else if (rtn != NEEDS_RETRY)
1197 continue;
1198
1199 scsi_eh_finish_cmd(scmd, done_q);
1200 }
1201
1202 return list_empty(work_q);
1203}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08001204EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
1206/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 * scsi_eh_tur - Send TUR to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001208 * @scmd: &scsi_cmnd to send TUR
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 *
1210 * Return value:
1211 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001212 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213static int scsi_eh_tur(struct scsi_cmnd *scmd)
1214{
1215 static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
1216 int retry_cnt = 1, rtn;
1217
1218retry_tur:
Martin K. Petersen0816c922013-05-10 10:36:04 -04001219 rtn = scsi_send_eh_cmnd(scmd, tur_command, 6,
1220 scmd->device->eh_timeout, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001223 __func__, scmd, rtn));
Christoph Hellwig631c2282006-07-08 20:42:15 +02001224
1225 switch (rtn) {
1226 case NEEDS_RETRY:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (retry_cnt--)
1228 goto retry_tur;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001229 /*FALLTHRU*/
1230 case SUCCESS:
Alan Sterne47373e2005-03-30 15:05:45 -05001231 return 0;
Christoph Hellwig631c2282006-07-08 20:42:15 +02001232 default:
1233 return 1;
Alan Sterne47373e2005-03-30 15:05:45 -05001234 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235}
1236
1237/**
David Jeffery3eef6252011-05-19 14:41:12 -04001238 * scsi_eh_test_devices - check if devices are responding from error recovery.
1239 * @cmd_list: scsi commands in error recovery.
1240 * @work_q: queue for commands which still need more error recovery
1241 * @done_q: queue for commands which are finished
1242 * @try_stu: boolean on if a STU command should be tried in addition to TUR.
1243 *
1244 * Decription:
1245 * Tests if devices are in a working state. Commands to devices now in
1246 * a working state are sent to the done_q while commands to devices which
1247 * are still failing to respond are returned to the work_q for more
1248 * processing.
1249 **/
1250static int scsi_eh_test_devices(struct list_head *cmd_list,
1251 struct list_head *work_q,
1252 struct list_head *done_q, int try_stu)
1253{
1254 struct scsi_cmnd *scmd, *next;
1255 struct scsi_device *sdev;
1256 int finish_cmds;
1257
1258 while (!list_empty(cmd_list)) {
1259 scmd = list_entry(cmd_list->next, struct scsi_cmnd, eh_entry);
1260 sdev = scmd->device;
1261
Hannes Reineckeb4562022013-10-23 10:51:21 +02001262 if (!try_stu) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001263 if (scsi_host_eh_past_deadline(sdev->host)) {
1264 /* Push items back onto work_q */
1265 list_splice_init(cmd_list, work_q);
Hannes Reineckeb4562022013-10-23 10:51:21 +02001266 SCSI_LOG_ERROR_RECOVERY(3,
1267 shost_printk(KERN_INFO, sdev->host,
1268 "skip %s, past eh deadline",
1269 __func__));
1270 break;
1271 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001272 }
1273
David Jeffery3eef6252011-05-19 14:41:12 -04001274 finish_cmds = !scsi_device_online(scmd->device) ||
1275 (try_stu && !scsi_eh_try_stu(scmd) &&
1276 !scsi_eh_tur(scmd)) ||
1277 !scsi_eh_tur(scmd);
1278
1279 list_for_each_entry_safe(scmd, next, cmd_list, eh_entry)
1280 if (scmd->device == sdev) {
James Bottomley24510792013-11-11 13:44:53 +01001281 if (finish_cmds &&
1282 (try_stu ||
1283 scsi_eh_action(scmd, SUCCESS) == SUCCESS))
David Jeffery3eef6252011-05-19 14:41:12 -04001284 scsi_eh_finish_cmd(scmd, done_q);
1285 else
1286 list_move_tail(&scmd->eh_entry, work_q);
1287 }
1288 }
1289 return list_empty(work_q);
1290}
1291
1292
1293/**
Rob Landleyeb448202007-11-03 13:30:39 -05001294 * scsi_eh_abort_cmds - abort pending commands.
1295 * @work_q: &list_head for pending commands.
1296 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 *
1298 * Decription:
1299 * Try and see whether or not it makes sense to try and abort the
Rob Landleyeb448202007-11-03 13:30:39 -05001300 * running command. This only works out to be the case if we have one
1301 * command that has timed out. If the command simply failed, it makes
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 * no sense to try and abort the command, since as far as the shost
1303 * adapter is concerned, it isn't running.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001304 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305static int scsi_eh_abort_cmds(struct list_head *work_q,
1306 struct list_head *done_q)
1307{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001308 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001309 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 int rtn;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001311 struct Scsi_Host *shost;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001313 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001314 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 continue;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001316 shost = scmd->device->host;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001317 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001318 list_splice_init(&check_list, work_q);
1319 SCSI_LOG_ERROR_RECOVERY(3,
1320 shost_printk(KERN_INFO, shost,
1321 "skip %s, past eh deadline\n",
1322 __func__));
1323 return list_empty(work_q);
1324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
1326 "0x%p\n", current->comm,
1327 scmd));
Hannes Reineckeb4562022013-10-23 10:51:21 +02001328 rtn = scsi_try_to_abort_cmd(shost->hostt, scmd);
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001329 if (rtn == FAILED) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
1331 " cmd failed:"
1332 "0x%p\n",
1333 current->comm,
1334 scmd));
Hannes Reinecke6fd046f2013-10-23 10:51:22 +02001335 list_splice_init(&check_list, work_q);
1336 return list_empty(work_q);
1337 }
1338 scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
1339 if (rtn == FAST_IO_FAIL)
1340 scsi_eh_finish_cmd(scmd, done_q);
1341 else
1342 list_move_tail(&scmd->eh_entry, &check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 }
1344
David Jeffery3eef6252011-05-19 14:41:12 -04001345 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346}
1347
1348/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * scsi_eh_try_stu - Send START_UNIT to device.
Rob Landleyeb448202007-11-03 13:30:39 -05001350 * @scmd: &scsi_cmnd to send START_UNIT
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 *
1352 * Return value:
1353 * 0 - Device is ready. 1 - Device NOT ready.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001354 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
1356{
1357 static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
Christoph Hellwig631c2282006-07-08 20:42:15 +02001359 if (scmd->device->allow_restart) {
Brian Kinged773e62007-03-29 15:25:52 -05001360 int i, rtn = NEEDS_RETRY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Brian Kinged773e62007-03-29 15:25:52 -05001362 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
James Bottomley9728c082008-11-30 10:32:26 -06001363 rtn = scsi_send_eh_cmnd(scmd, stu_command, 6, scmd->device->request_queue->rq_timeout, 0);
Brian Kinged773e62007-03-29 15:25:52 -05001364
Christoph Hellwig631c2282006-07-08 20:42:15 +02001365 if (rtn == SUCCESS)
1366 return 0;
1367 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 return 1;
1370}
1371
1372 /**
1373 * scsi_eh_stu - send START_UNIT if needed
Rob Landleyeb448202007-11-03 13:30:39 -05001374 * @shost: &scsi host being recovered.
1375 * @work_q: &list_head for pending commands.
1376 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 *
1378 * Notes:
1379 * If commands are failing due to not ready, initializing command required,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001380 * try revalidating the device, which will end up sending a start unit.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001381 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382static int scsi_eh_stu(struct Scsi_Host *shost,
1383 struct list_head *work_q,
1384 struct list_head *done_q)
1385{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001386 struct scsi_cmnd *scmd, *stu_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 struct scsi_device *sdev;
1388
1389 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001390 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001391 SCSI_LOG_ERROR_RECOVERY(3,
1392 shost_printk(KERN_INFO, shost,
1393 "skip %s, past eh deadline\n",
1394 __func__));
1395 break;
1396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 stu_scmd = NULL;
1398 list_for_each_entry(scmd, work_q, eh_entry)
1399 if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
1400 scsi_check_sense(scmd) == FAILED ) {
1401 stu_scmd = scmd;
1402 break;
1403 }
1404
1405 if (!stu_scmd)
1406 continue;
1407
1408 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
1409 " 0x%p\n", current->comm, sdev));
1410
1411 if (!scsi_eh_try_stu(stu_scmd)) {
1412 if (!scsi_device_online(sdev) ||
1413 !scsi_eh_tur(stu_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001414 list_for_each_entry_safe(scmd, next,
1415 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001416 if (scmd->device == sdev &&
1417 scsi_eh_action(scmd, SUCCESS) == SUCCESS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 scsi_eh_finish_cmd(scmd, done_q);
1419 }
1420 }
1421 } else {
1422 SCSI_LOG_ERROR_RECOVERY(3,
1423 printk("%s: START_UNIT failed to sdev:"
1424 " 0x%p\n", current->comm, sdev));
1425 }
1426 }
1427
1428 return list_empty(work_q);
1429}
1430
1431
1432/**
1433 * scsi_eh_bus_device_reset - send bdr if needed
1434 * @shost: scsi host being recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05001435 * @work_q: &list_head for pending commands.
1436 * @done_q: &list_head for processed commands.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 *
1438 * Notes:
Rob Landleyeb448202007-11-03 13:30:39 -05001439 * Try a bus device reset. Still, look to see whether we have multiple
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 * devices that are jammed or not - if we have multiple devices, it
1441 * makes no sense to try bus_device_reset - we really would need to try
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001442 * a bus_reset instead.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001443 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1445 struct list_head *work_q,
1446 struct list_head *done_q)
1447{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001448 struct scsi_cmnd *scmd, *bdr_scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 struct scsi_device *sdev;
1450 int rtn;
1451
1452 shost_for_each_device(sdev, shost) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001453 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001454 SCSI_LOG_ERROR_RECOVERY(3,
1455 shost_printk(KERN_INFO, shost,
1456 "skip %s, past eh deadline\n",
1457 __func__));
1458 break;
1459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 bdr_scmd = NULL;
1461 list_for_each_entry(scmd, work_q, eh_entry)
1462 if (scmd->device == sdev) {
1463 bdr_scmd = scmd;
1464 break;
1465 }
1466
1467 if (!bdr_scmd)
1468 continue;
1469
1470 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1471 " 0x%p\n", current->comm,
1472 sdev));
1473 rtn = scsi_try_bus_device_reset(bdr_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001474 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 if (!scsi_device_online(sdev) ||
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001476 rtn == FAST_IO_FAIL ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 !scsi_eh_tur(bdr_scmd)) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001478 list_for_each_entry_safe(scmd, next,
1479 work_q, eh_entry) {
James Bottomley24510792013-11-11 13:44:53 +01001480 if (scmd->device == sdev &&
1481 scsi_eh_action(scmd, rtn) != FAILED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 scsi_eh_finish_cmd(scmd,
1483 done_q);
1484 }
1485 }
1486 } else {
1487 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1488 " failed sdev:"
1489 "0x%p\n",
1490 current->comm,
1491 sdev));
1492 }
1493 }
1494
1495 return list_empty(work_q);
1496}
1497
1498/**
Mike Christie30bd7df2008-02-29 18:25:19 -06001499 * scsi_eh_target_reset - send target reset if needed
1500 * @shost: scsi host being recovered.
1501 * @work_q: &list_head for pending commands.
1502 * @done_q: &list_head for processed commands.
1503 *
1504 * Notes:
1505 * Try a target reset.
1506 */
1507static int scsi_eh_target_reset(struct Scsi_Host *shost,
1508 struct list_head *work_q,
1509 struct list_head *done_q)
1510{
James Bottomley98db5192010-10-25 15:53:41 -05001511 LIST_HEAD(tmp_list);
David Jeffery3eef6252011-05-19 14:41:12 -04001512 LIST_HEAD(check_list);
Mike Christie30bd7df2008-02-29 18:25:19 -06001513
James Bottomley98db5192010-10-25 15:53:41 -05001514 list_splice_init(work_q, &tmp_list);
1515
1516 while (!list_empty(&tmp_list)) {
1517 struct scsi_cmnd *next, *scmd;
1518 int rtn;
1519 unsigned int id;
Hannes Reineckeb4562022013-10-23 10:51:21 +02001520
Hannes Reineckeb4562022013-10-23 10:51:21 +02001521 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001522 /* push back on work queue for further processing */
1523 list_splice_init(&check_list, work_q);
1524 list_splice_init(&tmp_list, work_q);
1525 SCSI_LOG_ERROR_RECOVERY(3,
1526 shost_printk(KERN_INFO, shost,
1527 "skip %s, past eh deadline\n",
1528 __func__));
1529 return list_empty(work_q);
1530 }
James Bottomley98db5192010-10-25 15:53:41 -05001531
1532 scmd = list_entry(tmp_list.next, struct scsi_cmnd, eh_entry);
1533 id = scmd_id(scmd);
Mike Christie30bd7df2008-02-29 18:25:19 -06001534
1535 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending target reset "
1536 "to target %d\n",
1537 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001538 rtn = scsi_try_target_reset(scmd);
1539 if (rtn != SUCCESS && rtn != FAST_IO_FAIL)
Mike Christie30bd7df2008-02-29 18:25:19 -06001540 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Target reset"
1541 " failed target: "
1542 "%d\n",
1543 current->comm, id));
James Bottomley98db5192010-10-25 15:53:41 -05001544 list_for_each_entry_safe(scmd, next, &tmp_list, eh_entry) {
1545 if (scmd_id(scmd) != id)
1546 continue;
1547
David Jeffery3eef6252011-05-19 14:41:12 -04001548 if (rtn == SUCCESS)
1549 list_move_tail(&scmd->eh_entry, &check_list);
1550 else if (rtn == FAST_IO_FAIL)
James Bottomley98db5192010-10-25 15:53:41 -05001551 scsi_eh_finish_cmd(scmd, done_q);
1552 else
1553 /* push back on work queue for further processing */
1554 list_move(&scmd->eh_entry, work_q);
1555 }
1556 }
Mike Christie30bd7df2008-02-29 18:25:19 -06001557
David Jeffery3eef6252011-05-19 14:41:12 -04001558 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Mike Christie30bd7df2008-02-29 18:25:19 -06001559}
1560
1561/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001562 * scsi_eh_bus_reset - send a bus reset
Rob Landleyeb448202007-11-03 13:30:39 -05001563 * @shost: &scsi host being recovered.
1564 * @work_q: &list_head for pending commands.
1565 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001566 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1568 struct list_head *work_q,
1569 struct list_head *done_q)
1570{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001571 struct scsi_cmnd *scmd, *chan_scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001572 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 unsigned int channel;
1574 int rtn;
1575
1576 /*
1577 * we really want to loop over the various channels, and do this on
1578 * a channel by channel basis. we should also check to see if any
1579 * of the failed commands are on soft_reset devices, and if so, skip
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001580 * the reset.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 */
1582
1583 for (channel = 0; channel <= shost->max_channel; channel++) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001584 if (scsi_host_eh_past_deadline(shost)) {
Hannes Reineckeb4562022013-10-23 10:51:21 +02001585 list_splice_init(&check_list, work_q);
1586 SCSI_LOG_ERROR_RECOVERY(3,
1587 shost_printk(KERN_INFO, shost,
1588 "skip %s, past eh deadline\n",
1589 __func__));
1590 return list_empty(work_q);
1591 }
Hannes Reineckeb4562022013-10-23 10:51:21 +02001592
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 chan_scmd = NULL;
1594 list_for_each_entry(scmd, work_q, eh_entry) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04001595 if (channel == scmd_channel(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 chan_scmd = scmd;
1597 break;
1598 /*
1599 * FIXME add back in some support for
1600 * soft_reset devices.
1601 */
1602 }
1603 }
1604
1605 if (!chan_scmd)
1606 continue;
1607 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1608 " %d\n", current->comm,
1609 channel));
1610 rtn = scsi_try_bus_reset(chan_scmd);
Christof Schmitt2f2eb582010-03-24 16:50:30 +01001611 if (rtn == SUCCESS || rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001612 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
David Jeffery3eef6252011-05-19 14:41:12 -04001613 if (channel == scmd_channel(scmd)) {
1614 if (rtn == FAST_IO_FAIL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 scsi_eh_finish_cmd(scmd,
1616 done_q);
David Jeffery3eef6252011-05-19 14:41:12 -04001617 else
1618 list_move_tail(&scmd->eh_entry,
1619 &check_list);
1620 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 }
1622 } else {
1623 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1624 " failed chan: %d\n",
1625 current->comm,
1626 channel));
1627 }
1628 }
David Jeffery3eef6252011-05-19 14:41:12 -04001629 return scsi_eh_test_devices(&check_list, work_q, done_q, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630}
1631
1632/**
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001633 * scsi_eh_host_reset - send a host reset
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 * @work_q: list_head for processed commands.
1635 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001636 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637static int scsi_eh_host_reset(struct list_head *work_q,
1638 struct list_head *done_q)
1639{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001640 struct scsi_cmnd *scmd, *next;
David Jeffery3eef6252011-05-19 14:41:12 -04001641 LIST_HEAD(check_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 int rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644 if (!list_empty(work_q)) {
1645 scmd = list_entry(work_q->next,
1646 struct scsi_cmnd, eh_entry);
1647
1648 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1649 , current->comm));
1650
1651 rtn = scsi_try_host_reset(scmd);
David Jeffery3eef6252011-05-19 14:41:12 -04001652 if (rtn == SUCCESS) {
1653 list_splice_init(work_q, &check_list);
1654 } else if (rtn == FAST_IO_FAIL) {
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001655 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 scsi_eh_finish_cmd(scmd, done_q);
1657 }
1658 } else {
1659 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1660 " failed\n",
1661 current->comm));
1662 }
1663 }
David Jeffery3eef6252011-05-19 14:41:12 -04001664 return scsi_eh_test_devices(&check_list, work_q, done_q, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665}
1666
1667/**
1668 * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1669 * @work_q: list_head for processed commands.
1670 * @done_q: list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001671 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672static void scsi_eh_offline_sdevs(struct list_head *work_q,
1673 struct list_head *done_q)
1674{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001675 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02001677 list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
Matthew Wilcox31765d72007-08-17 11:02:10 -06001678 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1679 "not ready after error recovery\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
Christoph Hellwig3111b0d2005-06-19 13:43:26 +02001681 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 /*
1683 * FIXME: Handle lost cmds.
1684 */
1685 }
1686 scsi_eh_finish_cmd(scmd, done_q);
1687 }
1688 return;
1689}
1690
1691/**
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001692 * scsi_noretry_cmd - determine if command should be failed fast
Mike Christie4a274462008-08-19 18:45:31 -05001693 * @scmd: SCSI cmd to examine.
1694 */
1695int scsi_noretry_cmd(struct scsi_cmnd *scmd)
1696{
1697 switch (host_byte(scmd->result)) {
1698 case DID_OK:
1699 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001700 case DID_TIME_OUT:
1701 goto check_type;
Mike Christie4a274462008-08-19 18:45:31 -05001702 case DID_BUS_BUSY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001703 return (scmd->request->cmd_flags & REQ_FAILFAST_TRANSPORT);
Mike Christie4a274462008-08-19 18:45:31 -05001704 case DID_PARITY:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001705 return (scmd->request->cmd_flags & REQ_FAILFAST_DEV);
Mike Christie4a274462008-08-19 18:45:31 -05001706 case DID_ERROR:
1707 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1708 status_byte(scmd->result) == RESERVATION_CONFLICT)
1709 return 0;
1710 /* fall through */
1711 case DID_SOFT_ERROR:
Christoph Hellwig33659eb2010-08-07 18:17:56 +02001712 return (scmd->request->cmd_flags & REQ_FAILFAST_DRIVER);
Mike Christie4a274462008-08-19 18:45:31 -05001713 }
1714
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001715 if (status_byte(scmd->result) != CHECK_CONDITION)
1716 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001717
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001718check_type:
1719 /*
1720 * assume caller has checked sense and determined
1721 * the check condition was retryable.
1722 */
1723 if (scmd->request->cmd_flags & REQ_FAILFAST_DEV ||
1724 scmd->request->cmd_type == REQ_TYPE_BLOCK_PC)
1725 return 1;
1726 else
1727 return 0;
Mike Christie4a274462008-08-19 18:45:31 -05001728}
1729
1730/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 * scsi_decide_disposition - Disposition a cmd on return from LLD.
1732 * @scmd: SCSI cmd to examine.
1733 *
1734 * Notes:
1735 * This is *only* called when we are examining the status after sending
1736 * out the actual data command. any commands that are queued for error
1737 * recovery (e.g. test_unit_ready) do *not* come through here.
1738 *
1739 * When this routine returns failed, it means the error handler thread
1740 * is woken. In cases where the error code indicates an error that
1741 * doesn't require the error handler read (i.e. we don't need to
1742 * abort/reset), this function should return SUCCESS.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001743 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744int scsi_decide_disposition(struct scsi_cmnd *scmd)
1745{
1746 int rtn;
1747
1748 /*
1749 * if the device is offline, then we clearly just pass the result back
1750 * up to the top level.
1751 */
1752 if (!scsi_device_online(scmd->device)) {
1753 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1754 " as SUCCESS\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07001755 __func__));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 return SUCCESS;
1757 }
1758
1759 /*
1760 * first check the host byte, to see if there is anything in there
1761 * that would indicate what we need to do.
1762 */
1763 switch (host_byte(scmd->result)) {
1764 case DID_PASSTHROUGH:
1765 /*
1766 * no matter what, pass this through to the upper layer.
1767 * nuke this special code so that it looks like we are saying
1768 * did_ok.
1769 */
1770 scmd->result &= 0xff00ffff;
1771 return SUCCESS;
1772 case DID_OK:
1773 /*
1774 * looks good. drop through, and check the next byte.
1775 */
1776 break;
Hannes Reineckee494f6a2013-11-11 13:44:54 +01001777 case DID_ABORT:
1778 if (scmd->eh_eflags & SCSI_EH_ABORT_SCHEDULED) {
1779 scmd->result |= DID_TIME_OUT << 16;
1780 return SUCCESS;
1781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 case DID_NO_CONNECT:
1783 case DID_BAD_TARGET:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 /*
1785 * note - this means that we just report the status back
1786 * to the top level driver, not that we actually think
1787 * that it indicates SUCCESS.
1788 */
1789 return SUCCESS;
1790 /*
1791 * when the low level driver returns did_soft_error,
Jesper Juhl0bf8c862011-03-21 20:47:31 +01001792 * it is responsible for keeping an internal retry counter
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 * in order to avoid endless loops (db)
1794 *
1795 * actually this is a bug in this function here. we should
1796 * be mindful of the maximum number of retries specified
1797 * and not get stuck in a loop.
1798 */
1799 case DID_SOFT_ERROR:
1800 goto maybe_retry;
1801 case DID_IMM_RETRY:
1802 return NEEDS_RETRY;
1803
bf341912005-04-12 17:49:09 -05001804 case DID_REQUEUE:
1805 return ADD_TO_MLQUEUE;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001806 case DID_TRANSPORT_DISRUPTED:
1807 /*
1808 * LLD/transport was disrupted during processing of the IO.
1809 * The transport class is now blocked/blocking,
1810 * and the transport will decide what to do with the IO
Mike Christie939c22882008-11-04 19:47:19 -06001811 * based on its timers and recovery capablilities if
1812 * there are enough retries.
Mike Christiea4dfaa62008-08-19 18:45:25 -05001813 */
Mike Christie939c22882008-11-04 19:47:19 -06001814 goto maybe_retry;
Mike Christiea4dfaa62008-08-19 18:45:25 -05001815 case DID_TRANSPORT_FAILFAST:
1816 /*
1817 * The transport decided to failfast the IO (most likely
1818 * the fast io fail tmo fired), so send IO directly upwards.
1819 */
1820 return SUCCESS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 case DID_ERROR:
1822 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1823 status_byte(scmd->result) == RESERVATION_CONFLICT)
1824 /*
1825 * execute reservation conflict processing code
1826 * lower down
1827 */
1828 break;
1829 /* fallthrough */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 case DID_BUS_BUSY:
1831 case DID_PARITY:
1832 goto maybe_retry;
1833 case DID_TIME_OUT:
1834 /*
1835 * when we scan the bus, we get timeout messages for
1836 * these commands if there is no device available.
1837 * other hosts report did_no_connect for the same thing.
1838 */
1839 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1840 scmd->cmnd[0] == INQUIRY)) {
1841 return SUCCESS;
1842 } else {
1843 return FAILED;
1844 }
1845 case DID_RESET:
1846 return SUCCESS;
1847 default:
1848 return FAILED;
1849 }
1850
1851 /*
1852 * next, check the message byte.
1853 */
1854 if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1855 return FAILED;
1856
1857 /*
1858 * check the status byte to see if this indicates anything special.
1859 */
1860 switch (status_byte(scmd->result)) {
1861 case QUEUE_FULL:
Mike Christie42a6a912009-10-15 17:46:44 -07001862 scsi_handle_queue_full(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 /*
1864 * the case of trying to send too many commands to a
1865 * tagged queueing device.
1866 */
1867 case BUSY:
1868 /*
1869 * device can't talk to us at the moment. Should only
1870 * occur (SAM-3) when the task queue is empty, so will cause
1871 * the empty queue handling to trigger a stall in the
1872 * device.
1873 */
1874 return ADD_TO_MLQUEUE;
1875 case GOOD:
Ewan D. Milne279afdf2013-08-08 15:07:48 -04001876 if (scmd->cmnd[0] == REPORT_LUNS)
1877 scmd->device->sdev_target->expecting_lun_change = 0;
Vasu Dev4a840672009-10-22 15:46:33 -07001878 scsi_handle_queue_ramp_up(scmd->device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 case COMMAND_TERMINATED:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 return SUCCESS;
Vladislav Bolkhovitina9b589d2008-11-06 13:57:52 +03001881 case TASK_ABORTED:
1882 goto maybe_retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 case CHECK_CONDITION:
1884 rtn = scsi_check_sense(scmd);
1885 if (rtn == NEEDS_RETRY)
1886 goto maybe_retry;
1887 /* if rtn == FAILED, we have no sense information;
1888 * returning FAILED will wake the error handler thread
1889 * to collect the sense and redo the decide
1890 * disposition */
1891 return rtn;
1892 case CONDITION_GOOD:
1893 case INTERMEDIATE_GOOD:
1894 case INTERMEDIATE_C_GOOD:
1895 case ACA_ACTIVE:
1896 /*
1897 * who knows? FIXME(eric)
1898 */
1899 return SUCCESS;
1900
1901 case RESERVATION_CONFLICT:
James Bottomley9ccfc752005-10-02 11:45:08 -05001902 sdev_printk(KERN_INFO, scmd->device,
1903 "reservation conflict\n");
Moger, Babu2082ebc2012-01-24 20:38:46 +00001904 set_host_byte(scmd, DID_NEXUS_FAILURE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 return SUCCESS; /* causes immediate i/o error */
1906 default:
1907 return FAILED;
1908 }
1909 return FAILED;
1910
1911 maybe_retry:
1912
1913 /* we requeue for retry because the error was retryable, and
1914 * the request was not marked fast fail. Note that above,
1915 * even if the request is marked fast fail, we still requeue
1916 * for queue congestion conditions (QUEUE_FULL or BUSY) */
Brian King8884efa2006-02-24 17:10:04 -06001917 if ((++scmd->retries) <= scmd->allowed
Mike Christie4a274462008-08-19 18:45:31 -05001918 && !scsi_noretry_cmd(scmd)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 return NEEDS_RETRY;
1920 } else {
1921 /*
1922 * no more retries - report this one back to upper level.
1923 */
1924 return SUCCESS;
1925 }
1926}
1927
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001928static void eh_lock_door_done(struct request *req, int uptodate)
1929{
1930 __blk_put_request(req->q, req);
1931}
1932
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 * scsi_eh_lock_door - Prevent medium removal for the specified device
1935 * @sdev: SCSI device to prevent medium removal
1936 *
1937 * Locking:
James Bottomley91bc31f2009-05-17 09:30:48 -05001938 * We must be called from process context.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 *
1940 * Notes:
1941 * We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1942 * head of the devices request queue, and continue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001943 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944static void scsi_eh_lock_door(struct scsi_device *sdev)
1945{
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001946 struct request *req;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947
James Bottomley91bc31f2009-05-17 09:30:48 -05001948 /*
1949 * blk_get_request with GFP_KERNEL (__GFP_WAIT) sleeps until a
1950 * request becomes available
1951 */
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001952 req = blk_get_request(sdev->request_queue, READ, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
Jens Axboef27b0872014-06-06 07:57:37 -06001954 blk_rq_set_block_pc(req);
1955
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001956 req->cmd[0] = ALLOW_MEDIUM_REMOVAL;
1957 req->cmd[1] = 0;
1958 req->cmd[2] = 0;
1959 req->cmd[3] = 0;
1960 req->cmd[4] = SCSI_REMOVAL_PREVENT;
1961 req->cmd[5] = 0;
1962
1963 req->cmd_len = COMMAND_SIZE(req->cmd[0]);
1964
FUJITA Tomonorif0787272008-12-14 01:23:45 +09001965 req->cmd_flags |= REQ_QUIET;
1966 req->timeout = 10 * HZ;
1967 req->retries = 5;
1968
1969 blk_execute_rq_nowait(req->q, NULL, req, 1, eh_lock_door_done);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970}
1971
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972/**
1973 * scsi_restart_operations - restart io operations to the specified host.
1974 * @shost: Host we are restarting.
1975 *
1976 * Notes:
1977 * When we entered the error handler, we blocked all further i/o to
1978 * this device. we need to 'reverse' this process.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08001979 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980static void scsi_restart_operations(struct Scsi_Host *shost)
1981{
1982 struct scsi_device *sdev;
James Bottomley939647e2005-09-18 15:05:20 -05001983 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
1985 /*
1986 * If the door was locked, we need to insert a door lock request
1987 * onto the head of the SCSI request queue for the device. There
1988 * is no point trying to lock the door of an off-line device.
1989 */
1990 shost_for_each_device(sdev, shost) {
1991 if (scsi_device_online(sdev) && sdev->locked)
1992 scsi_eh_lock_door(sdev);
1993 }
1994
1995 /*
1996 * next free up anything directly waiting upon the host. this
1997 * will be requests for character device operations, and also for
1998 * ioctls to queued block devices.
1999 */
Hannes Reineckeb4562022013-10-23 10:51:21 +02002000 SCSI_LOG_ERROR_RECOVERY(3,
2001 printk("scsi_eh_%d waking up host to restart\n",
2002 shost->host_no));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003
James Bottomley939647e2005-09-18 15:05:20 -05002004 spin_lock_irqsave(shost->host_lock, flags);
2005 if (scsi_host_set_state(shost, SHOST_RUNNING))
2006 if (scsi_host_set_state(shost, SHOST_CANCEL))
2007 BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
2008 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009
2010 wake_up(&shost->host_wait);
2011
2012 /*
2013 * finally we need to re-initiate requests that may be pending. we will
2014 * have had everything blocked while error handling is taking place, and
2015 * now that error recovery is done, we will need to ensure that these
2016 * requests are started.
2017 */
2018 scsi_run_host_queues(shost);
Dan Williams57fc2e32012-06-21 23:25:32 -07002019
2020 /*
2021 * if eh is active and host_eh_scheduled is pending we need to re-run
2022 * recovery. we do this check after scsi_run_host_queues() to allow
2023 * everything pent up since the last eh run a chance to make forward
2024 * progress before we sync again. Either we'll immediately re-run
2025 * recovery or scsi_device_unbusy() will wake us again when these
2026 * pending commands complete.
2027 */
2028 spin_lock_irqsave(shost->host_lock, flags);
2029 if (shost->host_eh_scheduled)
2030 if (scsi_host_set_state(shost, SHOST_RECOVERY))
2031 WARN_ON(scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY));
2032 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033}
2034
2035/**
2036 * scsi_eh_ready_devs - check device ready state and recover if not.
2037 * @shost: host to be recovered.
Rob Landleyeb448202007-11-03 13:30:39 -05002038 * @work_q: &list_head for pending commands.
2039 * @done_q: &list_head for processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002040 */
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002041void scsi_eh_ready_devs(struct Scsi_Host *shost,
2042 struct list_head *work_q,
2043 struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044{
2045 if (!scsi_eh_stu(shost, work_q, done_q))
2046 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
Mike Christie30bd7df2008-02-29 18:25:19 -06002047 if (!scsi_eh_target_reset(shost, work_q, done_q))
2048 if (!scsi_eh_bus_reset(shost, work_q, done_q))
2049 if (!scsi_eh_host_reset(work_q, done_q))
2050 scsi_eh_offline_sdevs(work_q,
2051 done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052}
Darrick J. Wongdca84e42007-01-26 14:08:49 -08002053EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
2055/**
2056 * scsi_eh_flush_done_q - finish processed commands or retry them.
2057 * @done_q: list_head of processed commands.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002058 */
Tejun Heo041c5fc2006-01-23 13:09:36 +09002059void scsi_eh_flush_done_q(struct list_head *done_q)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060{
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002061 struct scsi_cmnd *scmd, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062
Christoph Hellwig937abeaa2005-06-19 13:43:56 +02002063 list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
2064 list_del_init(&scmd->eh_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 if (scsi_device_online(scmd->device) &&
Mike Christie4a274462008-08-19 18:45:31 -05002066 !scsi_noretry_cmd(scmd) &&
Brian King8884efa2006-02-24 17:10:04 -06002067 (++scmd->retries <= scmd->allowed)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
2069 " retry cmd: %p\n",
2070 current->comm,
2071 scmd));
2072 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
2073 } else {
Patrick Mansfield 793698c2005-05-16 17:42:15 -07002074 /*
2075 * If just we got sense for the device (called
2076 * scsi_eh_get_sense), scmd->result is already
2077 * set, do not set DRIVER_TIMEOUT.
2078 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (!scmd->result)
2080 scmd->result |= (DRIVER_TIMEOUT << 24);
2081 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
2082 " cmd: %p\n",
2083 current->comm, scmd));
2084 scsi_finish_command(scmd);
2085 }
2086 }
2087}
Tejun Heo041c5fc2006-01-23 13:09:36 +09002088EXPORT_SYMBOL(scsi_eh_flush_done_q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089
2090/**
2091 * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
2092 * @shost: Host to unjam.
2093 *
2094 * Notes:
2095 * When we come in here, we *know* that all commands on the bus have
2096 * either completed, failed or timed out. we also know that no further
2097 * commands are being sent to the host, so things are relatively quiet
2098 * and we have freedom to fiddle with things as we wish.
2099 *
2100 * This is only the *default* implementation. it is possible for
2101 * individual drivers to supply their own version of this function, and
2102 * if the maintainer wishes to do this, it is strongly suggested that
2103 * this function be taken as a template and modified. this function
2104 * was designed to correctly handle problems for about 95% of the
2105 * different cases out there, and it should always provide at least a
2106 * reasonable amount of error recovery.
2107 *
2108 * Any command marked 'failed' or 'timeout' must eventually have
2109 * scsi_finish_cmd() called for it. we do all of the retry stuff
2110 * here, so when we restart the host after we return it should have an
2111 * empty queue.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002112 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113static void scsi_unjam_host(struct Scsi_Host *shost)
2114{
2115 unsigned long flags;
2116 LIST_HEAD(eh_work_q);
2117 LIST_HEAD(eh_done_q);
2118
2119 spin_lock_irqsave(shost->host_lock, flags);
2120 list_splice_init(&shost->eh_cmd_q, &eh_work_q);
2121 spin_unlock_irqrestore(shost->host_lock, flags);
2122
2123 SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
2124
2125 if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
2126 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
2127 scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
2128
Hannes Reineckeb4562022013-10-23 10:51:21 +02002129 spin_lock_irqsave(shost->host_lock, flags);
Ren Mingxinbb3b6212013-11-11 13:44:56 +01002130 if (shost->eh_deadline != -1)
Hannes Reineckeb4562022013-10-23 10:51:21 +02002131 shost->last_reset = 0;
2132 spin_unlock_irqrestore(shost->host_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 scsi_eh_flush_done_q(&eh_done_q);
2134}
2135
2136/**
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002137 * scsi_error_handler - SCSI error handler thread
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 * @data: Host for which we are running.
2139 *
2140 * Notes:
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002141 * This is the main error handling loop. This is run as a kernel thread
2142 * for every SCSI host and handles all error handling activity.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002143 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144int scsi_error_handler(void *data)
2145{
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002146 struct Scsi_Host *shost = data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 /*
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002149 * We use TASK_INTERRUPTIBLE so that the thread is not
2150 * counted against the load average as a running process.
2151 * We never actually get interrupted because kthread_run
Frederik Schwarzerc03264a2008-12-23 00:54:00 +01002152 * disables signal delivery for the created thread.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 */
James Bottomley3ed7a472005-09-19 09:50:04 -05002154 while (!kthread_should_stop()) {
Dan Williamsb9d5c6b2012-06-21 23:25:42 -07002155 set_current_state(TASK_INTERRUPTIBLE);
Tejun Heoee7863b2006-05-15 20:57:20 +09002156 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
James Bottomley3ed7a472005-09-19 09:50:04 -05002157 shost->host_failed != shost->host_busy) {
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002158 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reineckeb4562022013-10-23 10:51:21 +02002159 printk("scsi_eh_%d: sleeping\n",
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002160 shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002161 schedule();
James Bottomley3ed7a472005-09-19 09:50:04 -05002162 continue;
2163 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
James Bottomley3ed7a472005-09-19 09:50:04 -05002165 __set_current_state(TASK_RUNNING);
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002166 SCSI_LOG_ERROR_RECOVERY(1,
Hannes Reineckeb4562022013-10-23 10:51:21 +02002167 printk("scsi_eh_%d: waking up %d/%d/%d\n",
2168 shost->host_no, shost->host_eh_scheduled,
2169 shost->host_failed, shost->host_busy));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 /*
2172 * We have a host that is failing for some reason. Figure out
2173 * what we need to do to get it up and online again (if we can).
2174 * If we fail, we end up taking the thing offline.
2175 */
Lin Mingae0751f2011-12-05 09:20:24 +08002176 if (!shost->eh_noresume && scsi_autopm_get_host(shost) != 0) {
Alan Sternbc4f2402010-06-17 10:41:42 -04002177 SCSI_LOG_ERROR_RECOVERY(1,
2178 printk(KERN_ERR "Error handler scsi_eh_%d "
2179 "unable to autoresume\n",
2180 shost->host_no));
2181 continue;
2182 }
2183
Christoph Hellwig9227c332006-04-01 19:21:04 +02002184 if (shost->transportt->eh_strategy_handler)
2185 shost->transportt->eh_strategy_handler(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 else
2187 scsi_unjam_host(shost);
2188
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 /*
2190 * Note - if the above fails completely, the action is to take
2191 * individual devices offline and flush the queue of any
2192 * outstanding requests that may have been pending. When we
2193 * restart, we restart any I/O to any other devices on the bus
2194 * which are still online.
2195 */
2196 scsi_restart_operations(shost);
Lin Mingae0751f2011-12-05 09:20:24 +08002197 if (!shost->eh_noresume)
2198 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 }
Steven Rostedt461a0ff2005-10-19 08:22:13 -04002200 __set_current_state(TASK_RUNNING);
2201
Christoph Hellwigad42eb12005-10-29 01:01:55 +02002202 SCSI_LOG_ERROR_RECOVERY(1,
2203 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
James Bottomley3ed7a472005-09-19 09:50:04 -05002204 shost->ehandler = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 return 0;
2206}
2207
2208/*
2209 * Function: scsi_report_bus_reset()
2210 *
2211 * Purpose: Utility function used by low-level drivers to report that
2212 * they have observed a bus reset on the bus being handled.
2213 *
2214 * Arguments: shost - Host in question
2215 * channel - channel on which reset was observed.
2216 *
2217 * Returns: Nothing
2218 *
2219 * Lock status: Host lock must be held.
2220 *
2221 * Notes: This only needs to be called if the reset is one which
2222 * originates from an unknown location. Resets originated
2223 * by the mid-level itself don't need to call this, but there
2224 * should be no harm.
2225 *
2226 * The main purpose of this is to make sure that a CHECK_CONDITION
2227 * is properly treated.
2228 */
2229void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
2230{
2231 struct scsi_device *sdev;
2232
2233 __shost_for_each_device(sdev, shost) {
Mike Christie30bd7df2008-02-29 18:25:19 -06002234 if (channel == sdev_channel(sdev))
2235 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 }
2237}
2238EXPORT_SYMBOL(scsi_report_bus_reset);
2239
2240/*
2241 * Function: scsi_report_device_reset()
2242 *
2243 * Purpose: Utility function used by low-level drivers to report that
2244 * they have observed a device reset on the device being handled.
2245 *
2246 * Arguments: shost - Host in question
2247 * channel - channel on which reset was observed
2248 * target - target on which reset was observed
2249 *
2250 * Returns: Nothing
2251 *
2252 * Lock status: Host lock must be held
2253 *
2254 * Notes: This only needs to be called if the reset is one which
2255 * originates from an unknown location. Resets originated
2256 * by the mid-level itself don't need to call this, but there
2257 * should be no harm.
2258 *
2259 * The main purpose of this is to make sure that a CHECK_CONDITION
2260 * is properly treated.
2261 */
2262void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
2263{
2264 struct scsi_device *sdev;
2265
2266 __shost_for_each_device(sdev, shost) {
Jeff Garzik422c0d62005-10-24 18:05:09 -04002267 if (channel == sdev_channel(sdev) &&
Mike Christie30bd7df2008-02-29 18:25:19 -06002268 target == sdev_id(sdev))
2269 __scsi_report_device_reset(sdev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 }
2271}
2272EXPORT_SYMBOL(scsi_report_device_reset);
2273
2274static void
2275scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
2276{
2277}
2278
2279/*
2280 * Function: scsi_reset_provider
2281 *
2282 * Purpose: Send requested reset to a bus or device at any phase.
2283 *
2284 * Arguments: device - device to send reset to
2285 * flag - reset type (see scsi.h)
2286 *
2287 * Returns: SUCCESS/FAILURE.
2288 *
2289 * Notes: This is used by the SCSI Generic driver to provide
2290 * Bus/Device reset capability.
2291 */
2292int
2293scsi_reset_provider(struct scsi_device *dev, int flag)
2294{
Alan Sternbc4f2402010-06-17 10:41:42 -04002295 struct scsi_cmnd *scmd;
James Smartd7a1bb02006-03-08 14:50:12 -05002296 struct Scsi_Host *shost = dev->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 struct request req;
James Smartd7a1bb02006-03-08 14:50:12 -05002298 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 int rtn;
2300
Alan Sternbc4f2402010-06-17 10:41:42 -04002301 if (scsi_autopm_get_host(shost) < 0)
2302 return FAILED;
2303
Christoph Hellwig04796332014-02-20 14:20:55 -08002304 if (!get_device(&dev->sdev_gendev)) {
2305 rtn = FAILED;
2306 goto out_put_autopm_host;
2307 }
2308
Alan Sternbc4f2402010-06-17 10:41:42 -04002309 scmd = scsi_get_command(dev, GFP_KERNEL);
FUJITA Tomonori4f54eec2008-04-29 09:54:37 +02002310 blk_rq_init(NULL, &req);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 scmd->request = &req;
Christoph Hellwigb4edcbc2005-06-19 13:40:52 +02002312
Boaz Harrosh64a87b22008-04-30 11:19:47 +03002313 scmd->cmnd = req.cmd;
2314
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 scmd->scsi_done = scsi_reset_provider_done_command;
Boaz Harrosh30b0c372007-12-13 13:47:40 +02002316 memset(&scmd->sdb, 0, sizeof(scmd->sdb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
2318 scmd->cmd_len = 0;
2319
2320 scmd->sc_data_direction = DMA_BIDIRECTIONAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
James Smartd7a1bb02006-03-08 14:50:12 -05002322 spin_lock_irqsave(shost->host_lock, flags);
2323 shost->tmf_in_progress = 1;
2324 spin_unlock_irqrestore(shost->host_lock, flags);
2325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 switch (flag) {
2327 case SCSI_TRY_RESET_DEVICE:
2328 rtn = scsi_try_bus_device_reset(scmd);
2329 if (rtn == SUCCESS)
2330 break;
2331 /* FALLTHROUGH */
Mike Christie30bd7df2008-02-29 18:25:19 -06002332 case SCSI_TRY_RESET_TARGET:
2333 rtn = scsi_try_target_reset(scmd);
2334 if (rtn == SUCCESS)
2335 break;
2336 /* FALLTHROUGH */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 case SCSI_TRY_RESET_BUS:
2338 rtn = scsi_try_bus_reset(scmd);
2339 if (rtn == SUCCESS)
2340 break;
2341 /* FALLTHROUGH */
2342 case SCSI_TRY_RESET_HOST:
2343 rtn = scsi_try_host_reset(scmd);
2344 break;
2345 default:
2346 rtn = FAILED;
2347 }
2348
James Smartd7a1bb02006-03-08 14:50:12 -05002349 spin_lock_irqsave(shost->host_lock, flags);
2350 shost->tmf_in_progress = 0;
2351 spin_unlock_irqrestore(shost->host_lock, flags);
2352
2353 /*
2354 * be sure to wake up anyone who was sleeping or had their queue
2355 * suspended while we performed the TMF.
2356 */
2357 SCSI_LOG_ERROR_RECOVERY(3,
2358 printk("%s: waking up host to restart after TMF\n",
Harvey Harrisoncadbd4a2008-07-03 23:47:27 -07002359 __func__));
James Smartd7a1bb02006-03-08 14:50:12 -05002360
2361 wake_up(&shost->host_wait);
2362
2363 scsi_run_host_queues(shost);
2364
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365 scsi_next_command(scmd);
Christoph Hellwig04796332014-02-20 14:20:55 -08002366out_put_autopm_host:
Alan Sternbc4f2402010-06-17 10:41:42 -04002367 scsi_autopm_put_host(shost);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 return rtn;
2369}
2370EXPORT_SYMBOL(scsi_reset_provider);
2371
2372/**
2373 * scsi_normalize_sense - normalize main elements from either fixed or
2374 * descriptor sense data format into a common format.
2375 *
2376 * @sense_buffer: byte array containing sense data returned by device
2377 * @sb_len: number of valid bytes in sense_buffer
2378 * @sshdr: pointer to instance of structure that common
2379 * elements are written to.
2380 *
2381 * Notes:
2382 * The "main elements" from sense data are: response_code, sense_key,
2383 * asc, ascq and additional_length (only for descriptor format).
2384 *
2385 * Typically this function can be called after a device has
2386 * responded to a SCSI command with the CHECK_CONDITION status.
2387 *
2388 * Return value:
2389 * 1 if valid sense data information found, else 0;
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002390 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
2392 struct scsi_sense_hdr *sshdr)
2393{
James Bottomley33aa6872005-08-28 11:31:14 -05002394 if (!sense_buffer || !sb_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395 return 0;
2396
2397 memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
2398
2399 sshdr->response_code = (sense_buffer[0] & 0x7f);
James Bottomley33aa6872005-08-28 11:31:14 -05002400
2401 if (!scsi_sense_valid(sshdr))
2402 return 0;
2403
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 if (sshdr->response_code >= 0x72) {
2405 /*
2406 * descriptor format
2407 */
2408 if (sb_len > 1)
2409 sshdr->sense_key = (sense_buffer[1] & 0xf);
2410 if (sb_len > 2)
2411 sshdr->asc = sense_buffer[2];
2412 if (sb_len > 3)
2413 sshdr->ascq = sense_buffer[3];
2414 if (sb_len > 7)
2415 sshdr->additional_length = sense_buffer[7];
2416 } else {
Jesper Juhl0bf8c862011-03-21 20:47:31 +01002417 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 * fixed format
2419 */
2420 if (sb_len > 2)
2421 sshdr->sense_key = (sense_buffer[2] & 0xf);
2422 if (sb_len > 7) {
2423 sb_len = (sb_len < (sense_buffer[7] + 8)) ?
2424 sb_len : (sense_buffer[7] + 8);
2425 if (sb_len > 12)
2426 sshdr->asc = sense_buffer[12];
2427 if (sb_len > 13)
2428 sshdr->ascq = sense_buffer[13];
2429 }
2430 }
2431
2432 return 1;
2433}
2434EXPORT_SYMBOL(scsi_normalize_sense);
2435
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
2437 struct scsi_sense_hdr *sshdr)
2438{
2439 return scsi_normalize_sense(cmd->sense_buffer,
FUJITA Tomonorib80ca4f2008-01-13 15:46:13 +09002440 SCSI_SENSE_BUFFERSIZE, sshdr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441}
2442EXPORT_SYMBOL(scsi_command_normalize_sense);
2443
2444/**
Rob Landleyeb448202007-11-03 13:30:39 -05002445 * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 * @sense_buffer: byte array of descriptor format sense data
2447 * @sb_len: number of valid bytes in sense_buffer
2448 * @desc_type: value of descriptor type to find
2449 * (e.g. 0 -> information)
2450 *
2451 * Notes:
2452 * only valid when sense data is in descriptor format
2453 *
2454 * Return value:
2455 * pointer to start of (first) descriptor if found else NULL
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002456 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
2458 int desc_type)
2459{
2460 int add_sen_len, add_len, desc_len, k;
2461 const u8 * descp;
2462
2463 if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
2464 return NULL;
2465 if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
2466 return NULL;
2467 add_sen_len = (add_sen_len < (sb_len - 8)) ?
2468 add_sen_len : (sb_len - 8);
2469 descp = &sense_buffer[8];
2470 for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
2471 descp += desc_len;
2472 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
2473 desc_len = add_len + 2;
2474 if (descp[0] == desc_type)
2475 return descp;
2476 if (add_len < 0) // short descriptor ??
2477 break;
2478 }
2479 return NULL;
2480}
2481EXPORT_SYMBOL(scsi_sense_desc_find);
2482
2483/**
Rob Landleyeb448202007-11-03 13:30:39 -05002484 * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 * @sense_buffer: byte array of sense data
2486 * @sb_len: number of valid bytes in sense_buffer
2487 * @info_out: pointer to 64 integer where 8 or 4 byte information
2488 * field will be placed if found.
2489 *
2490 * Return value:
2491 * 1 if information field found, 0 if not found.
Randy Dunlapdc8875e2007-11-15 15:42:30 -08002492 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
2494 u64 * info_out)
2495{
2496 int j;
2497 const u8 * ucp;
2498 u64 ull;
2499
2500 if (sb_len < 7)
2501 return 0;
2502 switch (sense_buffer[0] & 0x7f) {
2503 case 0x70:
2504 case 0x71:
2505 if (sense_buffer[0] & 0x80) {
2506 *info_out = (sense_buffer[3] << 24) +
2507 (sense_buffer[4] << 16) +
2508 (sense_buffer[5] << 8) + sense_buffer[6];
2509 return 1;
2510 } else
2511 return 0;
2512 case 0x72:
2513 case 0x73:
2514 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
2515 0 /* info desc */);
2516 if (ucp && (0xa == ucp[1])) {
2517 ull = 0;
2518 for (j = 0; j < 8; ++j) {
2519 if (j > 0)
2520 ull <<= 8;
2521 ull |= ucp[4 + j];
2522 }
2523 *info_out = ull;
2524 return 1;
2525 } else
2526 return 0;
2527 default:
2528 return 0;
2529 }
2530}
2531EXPORT_SYMBOL(scsi_get_sense_info_fld);
FUJITA Tomonori3bc6a262008-03-25 09:26:49 +09002532
2533/**
2534 * scsi_build_sense_buffer - build sense data in a buffer
2535 * @desc: Sense format (non zero == descriptor format,
2536 * 0 == fixed format)
2537 * @buf: Where to build sense data
2538 * @key: Sense key
2539 * @asc: Additional sense code
2540 * @ascq: Additional sense code qualifier
2541 *
2542 **/
2543void scsi_build_sense_buffer(int desc, u8 *buf, u8 key, u8 asc, u8 ascq)
2544{
2545 if (desc) {
2546 buf[0] = 0x72; /* descriptor, current */
2547 buf[1] = key;
2548 buf[2] = asc;
2549 buf[3] = ascq;
2550 buf[7] = 0;
2551 } else {
2552 buf[0] = 0x70; /* fixed, current */
2553 buf[2] = key;
2554 buf[7] = 0xa;
2555 buf[12] = asc;
2556 buf[13] = ascq;
2557 }
2558}
2559EXPORT_SYMBOL(scsi_build_sense_buffer);