3 ======================================
5 This document describes SCSI midlayer error handling infrastructure.
6 Please refer to Documentation/scsi/scsi_mid_low_api.txt for more
7 information regarding SCSI midlayer.
11 [1] How SCSI commands travel through the midlayer and to EH
12 [1-1] struct scsi_cmnd
13 [1-2] How do scmd's get completed?
14 [1-2-1] Completing a scmd w/ scsi_done
15 [1-2-2] Completing a scmd w/ timeout
16 [1-3] How EH takes over
18 [2-1] EH through fine-grained callbacks
20 [2-1-2] Flow of scmds through EH
21 [2-1-3] Flow of control
22 [2-2] EH through transportt->eh_strategy_handler()
23 [2-2-1] Pre transportt->eh_strategy_handler() SCSI midlayer conditions
24 [2-2-2] Post transportt->eh_strategy_handler() SCSI midlayer conditions
25 [2-2-3] Things to consider
28 [1] How SCSI commands travel through the midlayer and to EH
30 [1-1] struct scsi_cmnd
32 Each SCSI command is represented with struct scsi_cmnd (== scmd). A
33 scmd has two list_head's to link itself into lists. The two are
34 scmd->list and scmd->eh_entry. The former is used for free list or
35 per-device allocated scmd list and not of much interest to this EH
36 discussion. The latter is used for completion and EH lists and unless
37 otherwise stated scmds are always linked using scmd->eh_entry in this
41 [1-2] How do scmd's get completed?
43 Once LLDD gets hold of a scmd, either the LLDD will complete the
44 command by calling scsi_done callback passed from midlayer when
45 invoking hostt->queuecommand() or the block layer will time it out.
48 [1-2-1] Completing a scmd w/ scsi_done
50 For all non-EH commands, scsi_done() is the completion callback. It
51 just calls blk_complete_request() to delete the block layer timer and
54 SCSI_SOFTIRQ handler scsi_softirq calls scsi_decide_disposition() to
55 determine what to do with the command. scsi_decide_disposition()
56 looks at the scmd->result value and sense data to determine what to do
60 scsi_finish_command() is invoked for the command. The
61 function does some maintenance chores and then calls
62 scsi_io_completion() to finish the I/O.
63 scsi_io_completion() then notifies the block layer on
64 the completed request by calling blk_end_request and
65 friends or figures out what to do with the remainder
66 of the data in case of an error.
70 scmd is requeued to blk queue.
73 scsi_eh_scmd_add(scmd) is invoked for the command. See
74 [1-3] for details of this function.
77 [1-2-2] Completing a scmd w/ timeout
79 The timeout handler is scsi_times_out(). When a timeout occurs, this
82 1. invokes optional hostt->eh_timed_out() callback. Return value can
86 This indicates that more time is required to finish the
87 command. Timer is restarted. This action is counted as a
88 retry and only allowed scmd->allowed + 1(!) times. Once the
89 limit is reached, action for BLK_EH_DONE is taken instead.
92 eh_timed_out() callback did not handle the command.
95 2. scsi_abort_command() is invoked to schedule an asynchrous abort.
96 Asynchronous abort are not invoked for commands which the
97 SCSI_EH_ABORT_SCHEDULED flag is set (this indicates that the command
98 already had been aborted once, and this is a retry which failed),
99 or when the EH deadline is expired. In these case Step #3 is taken.
101 3. scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD) is invoked for the
102 command. See [1-4] for more information.
104 [1-3] Asynchronous command aborts
106 After a timeout occurs a command abort is scheduled from
107 scsi_abort_command(). If the abort is successful the command
108 will either be retried (if the number of retries is not exhausted)
109 or terminated with DID_TIME_OUT.
110 Otherwise scsi_eh_scmd_add() is invoked for the command.
111 See [1-4] for more information.
113 [1-4] How EH takes over
115 scmds enter EH via scsi_eh_scmd_add(), which does the following.
117 1. Links scmd->eh_entry to shost->eh_cmd_q
119 2. Sets SHOST_RECOVERY bit in shost->shost_state
121 3. Increments shost->host_failed
123 4. Wakes up SCSI EH thread if shost->host_busy == shost->host_failed
125 As can be seen above, once any scmd is added to shost->eh_cmd_q,
126 SHOST_RECOVERY shost_state bit is turned on. This prevents any new
127 scmd to be issued from blk queue to the host; eventually, all scmds on
128 the host either complete normally, fail and get added to eh_cmd_q, or
129 time out and get added to shost->eh_cmd_q.
131 If all scmds either complete or fail, the number of in-flight scmds
132 becomes equal to the number of failed scmds - i.e. shost->host_busy ==
133 shost->host_failed. This wakes up SCSI EH thread. So, once woken up,
134 SCSI EH thread can expect that all in-flight commands have failed and
135 are linked on shost->eh_cmd_q.
137 Note that this does not mean lower layers are quiescent. If a LLDD
138 completed a scmd with error status, the LLDD and lower layers are
139 assumed to forget about the scmd at that point. However, if a scmd
140 has timed out, unless hostt->eh_timed_out() made lower layers forget
141 about the scmd, which currently no LLDD does, the command is still
142 active as long as lower layers are concerned and completion could
143 occur at any time. Of course, all such completions are ignored as the
144 timer has already expired.
146 We'll talk about how SCSI EH takes actions to abort - make LLDD
147 forget about - timed out scmds later.
150 [2] How SCSI EH works
152 LLDD's can implement SCSI EH actions in one of the following two
155 - Fine-grained EH callbacks
156 LLDD can implement fine-grained EH callbacks and let SCSI
157 midlayer drive error handling and call appropriate callbacks.
158 This will be discussed further in [2-1].
160 - eh_strategy_handler() callback
161 This is one big callback which should perform whole error
162 handling. As such, it should do all chores the SCSI midlayer
163 performs during recovery. This will be discussed in [2-2].
165 Once recovery is complete, SCSI EH resumes normal operation by
166 calling scsi_restart_operations(), which
168 1. Checks if door locking is needed and locks door.
170 2. Clears SHOST_RECOVERY shost_state bit
172 3. Wakes up waiters on shost->host_wait. This occurs if someone
173 calls scsi_block_when_processing_errors() on the host.
174 (*QUESTION* why is it needed? All operations will be blocked
175 anyway after it reaches blk queue.)
177 4. Kicks queues in all devices on the host in the asses
180 [2-1] EH through fine-grained callbacks
184 If eh_strategy_handler() is not present, SCSI midlayer takes charge
185 of driving error handling. EH's goals are two - make LLDD, host and
186 device forget about timed out scmds and make them ready for new
187 commands. A scmd is said to be recovered if the scmd is forgotten by
188 lower layers and lower layers are ready to process or fail the scmd
191 To achieve these goals, EH performs recovery actions with increasing
192 severity. Some actions are performed by issuing SCSI commands and
193 others are performed by invoking one of the following fine-grained
194 hostt EH callbacks. Callbacks may be omitted and omitted ones are
195 considered to fail always.
197 int (* eh_abort_handler)(struct scsi_cmnd *);
198 int (* eh_device_reset_handler)(struct scsi_cmnd *);
199 int (* eh_bus_reset_handler)(struct scsi_cmnd *);
200 int (* eh_host_reset_handler)(struct scsi_cmnd *);
202 Higher-severity actions are taken only when lower-severity actions
203 cannot recover some of failed scmds. Also, note that failure of the
204 highest-severity action means EH failure and results in offlining of
205 all unrecovered devices.
207 During recovery, the following rules are followed
209 - Recovery actions are performed on failed scmds on the to do list,
210 eh_work_q. If a recovery action succeeds for a scmd, recovered
211 scmds are removed from eh_work_q.
213 Note that single recovery action on a scmd can recover multiple
214 scmds. e.g. resetting a device recovers all failed scmds on the
217 - Higher severity actions are taken iff eh_work_q is not empty after
218 lower severity actions are complete.
220 - EH reuses failed scmds to issue commands for recovery. For
221 timed-out scmds, SCSI EH ensures that LLDD forgets about a scmd
222 before reusing it for EH commands.
224 When a scmd is recovered, the scmd is moved from eh_work_q to EH
225 local eh_done_q using scsi_eh_finish_cmd(). After all scmds are
226 recovered (eh_work_q is empty), scsi_eh_flush_done_q() is invoked to
227 either retry or error-finish (notify upper layer of failure) recovered
230 scmds are retried iff its sdev is still online (not offlined during
231 EH), REQ_FAILFAST is not set and ++scmd->retries is less than
235 [2-1-2] Flow of scmds through EH
237 1. Error completion / time out
238 ACTION: scsi_eh_scmd_add() is invoked for scmd
239 - add scmd to shost->eh_cmd_q
241 - shost->host_failed++
242 LOCKING: shost->host_lock
245 ACTION: move all scmds to EH's local eh_work_q. shost->eh_cmd_q
247 LOCKING: shost->host_lock (not strictly necessary, just for
251 ACTION: scsi_eh_finish_cmd() is invoked to EH-finish scmd
252 - scsi_setup_cmd_retry()
253 - move from local eh_work_q to local eh_done_q
255 CONCURRENCY: at most one thread per separate eh_work_q to
256 keep queue manipulation lockless
259 ACTION: scsi_eh_flush_done_q() retries scmds or notifies upper
260 layer of failure. May be called concurrently but must have
261 a no more than one thread per separate eh_work_q to
262 manipulate the queue locklessly
263 - scmd is removed from eh_done_q and scmd->eh_entry is cleared
264 - if retry is necessary, scmd is requeued using
266 - otherwise, scsi_finish_command() is invoked for scmd
267 - zero shost->host_failed
268 LOCKING: queue or finish function performs appropriate locking
271 [2-1-3] Flow of control
273 EH through fine-grained callbacks start from scsi_unjam_host().
277 1. Lock shost->host_lock, splice_init shost->eh_cmd_q into local
278 eh_work_q and unlock host_lock. Note that shost->eh_cmd_q is
279 cleared by this action.
281 2. Invoke scsi_eh_get_sense.
283 <<scsi_eh_get_sense>>
285 This action is taken for each error-completed
286 (!SCSI_EH_CANCEL_CMD) commands without valid sense data. Most
287 SCSI transports/LLDDs automatically acquire sense data on
288 command failures (autosense). Autosense is recommended for
289 performance reasons and as sense information could get out of
290 sync between occurrence of CHECK CONDITION and this action.
292 Note that if autosense is not supported, scmd->sense_buffer
293 contains invalid sense data when error-completing the scmd
294 with scsi_done(). scsi_decide_disposition() always returns
295 FAILED in such cases thus invoking SCSI EH. When the scmd
296 reaches here, sense data is acquired and
297 scsi_decide_disposition() is called again.
299 1. Invoke scsi_request_sense() which issues REQUEST_SENSE
300 command. If fails, no action. Note that taking no action
301 causes higher-severity recovery to be taken for the scmd.
303 2. Invoke scsi_decide_disposition() on the scmd
306 scmd->retries is set to scmd->allowed preventing
307 scsi_eh_flush_done_q() from retrying the scmd and
308 scsi_eh_finish_cmd() is invoked.
311 scsi_eh_finish_cmd() invoked
316 3. If !list_empty(&eh_work_q), invoke scsi_eh_abort_cmds().
318 <<scsi_eh_abort_cmds>>
320 This action is taken for each timed out command when
321 no_async_abort is enabled in the host template.
322 hostt->eh_abort_handler() is invoked for each scmd. The
323 handler returns SUCCESS if it has succeeded to make LLDD and
324 all related hardware forget about the scmd.
326 If a timedout scmd is successfully aborted and the sdev is
327 either offline or ready, scsi_eh_finish_cmd() is invoked for
328 the scmd. Otherwise, the scmd is left in eh_work_q for
329 higher-severity actions.
331 Note that both offline and ready status mean that the sdev is
332 ready to process new scmds, where processing also implies
333 immediate failing; thus, if a sdev is in one of the two
334 states, no further recovery action is needed.
336 Device readiness is tested using scsi_eh_tur() which issues
337 TEST_UNIT_READY command. Note that the scmd must have been
338 aborted successfully before reusing it for TEST_UNIT_READY.
340 4. If !list_empty(&eh_work_q), invoke scsi_eh_ready_devs()
342 <<scsi_eh_ready_devs>>
344 This function takes four increasingly more severe measures to
345 make failed sdevs ready for new commands.
347 1. Invoke scsi_eh_stu()
351 For each sdev which has failed scmds with valid sense data
352 of which scsi_check_sense()'s verdict is FAILED,
353 START_STOP_UNIT command is issued w/ start=1. Note that
354 as we explicitly choose error-completed scmds, it is known
355 that lower layers have forgotten about the scmd and we can
358 If STU succeeds and the sdev is either offline or ready,
359 all failed scmds on the sdev are EH-finished with
360 scsi_eh_finish_cmd().
362 *NOTE* If hostt->eh_abort_handler() isn't implemented or
363 failed, we may still have timed out scmds at this point
364 and STU doesn't make lower layers forget about those
365 scmds. Yet, this function EH-finish all scmds on the sdev
366 if STU succeeds leaving lower layers in an inconsistent
367 state. It seems that STU action should be taken only when
368 a sdev has no timed out scmd.
370 2. If !list_empty(&eh_work_q), invoke scsi_eh_bus_device_reset().
372 <<scsi_eh_bus_device_reset>>
374 This action is very similar to scsi_eh_stu() except that,
375 instead of issuing STU, hostt->eh_device_reset_handler()
376 is used. Also, as we're not issuing SCSI commands and
377 resetting clears all scmds on the sdev, there is no need
378 to choose error-completed scmds.
380 3. If !list_empty(&eh_work_q), invoke scsi_eh_bus_reset()
382 <<scsi_eh_bus_reset>>
384 hostt->eh_bus_reset_handler() is invoked for each channel
385 with failed scmds. If bus reset succeeds, all failed
386 scmds on all ready or offline sdevs on the channel are
389 4. If !list_empty(&eh_work_q), invoke scsi_eh_host_reset()
391 <<scsi_eh_host_reset>>
393 This is the last resort. hostt->eh_host_reset_handler()
394 is invoked. If host reset succeeds, all failed scmds on
395 all ready or offline sdevs on the host are EH-finished.
397 5. If !list_empty(&eh_work_q), invoke scsi_eh_offline_sdevs()
399 <<scsi_eh_offline_sdevs>>
401 Take all sdevs which still have unrecovered scmds offline
402 and EH-finish the scmds.
404 5. Invoke scsi_eh_flush_done_q().
406 <<scsi_eh_flush_done_q>>
408 At this point all scmds are recovered (or given up) and
409 put on eh_done_q by scsi_eh_finish_cmd(). This function
410 flushes eh_done_q by either retrying or notifying upper
411 layer of failure of the scmds.
414 [2-2] EH through transportt->eh_strategy_handler()
416 transportt->eh_strategy_handler() is invoked in the place of
417 scsi_unjam_host() and it is responsible for whole recovery process.
418 On completion, the handler should have made lower layers forget about
419 all failed scmds and either ready for new commands or offline. Also,
420 it should perform SCSI EH maintenance chores to maintain integrity of
421 SCSI midlayer. IOW, of the steps described in [2-1-2], all steps
422 except for #1 must be implemented by eh_strategy_handler().
425 [2-2-1] Pre transportt->eh_strategy_handler() SCSI midlayer conditions
427 The following conditions are true on entry to the handler.
429 - Each failed scmd's eh_flags field is set appropriately.
431 - Each failed scmd is linked on scmd->eh_cmd_q by scmd->eh_entry.
433 - SHOST_RECOVERY is set.
435 - shost->host_failed == shost->host_busy
438 [2-2-2] Post transportt->eh_strategy_handler() SCSI midlayer conditions
440 The following conditions must be true on exit from the handler.
442 - shost->host_failed is zero.
444 - Each scmd is in such a state that scsi_setup_cmd_retry() on the
445 scmd doesn't make any difference.
447 - shost->eh_cmd_q is cleared.
449 - Each scmd->eh_entry is cleared.
451 - Either scsi_queue_insert() or scsi_finish_command() is called on
452 each scmd. Note that the handler is free to use scmd->retries and
453 ->allowed to limit the number of retries.
456 [2-2-3] Things to consider
458 - Know that timed out scmds are still active on lower layers. Make
459 lower layers forget about them before doing anything else with
462 - For consistency, when accessing/modifying shost data structure,
463 grab shost->host_lock.
465 - On completion, each failed sdev must have forgotten about all
468 - On completion, each failed sdev must be ready for new commands or