Root/Documentation/scsi/scsi_eh.txt

1
2SCSI EH
3======================================
4
5 This document describes SCSI midlayer error handling infrastructure.
6Please refer to Documentation/scsi/scsi_mid_low_api.txt for more
7information regarding SCSI midlayer.
8
9TABLE OF CONTENTS
10
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
17[2] How SCSI EH works
18    [2-1] EH through fine-grained callbacks
19    [2-1-1] Overview
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
26
27
28[1] How SCSI commands travel through the midlayer and to EH
29
30[1-1] struct scsi_cmnd
31
32 Each SCSI command is represented with struct scsi_cmnd (== scmd). A
33scmd has two list_head's to link itself into lists. The two are
34scmd->list and scmd->eh_entry. The former is used for free list or
35per-device allocated scmd list and not of much interest to this EH
36discussion. The latter is used for completion and EH lists and unless
37otherwise stated scmds are always linked using scmd->eh_entry in this
38discussion.
39
40
41[1-2] How do scmd's get completed?
42
43 Once LLDD gets hold of a scmd, either the LLDD will complete the
44command by calling scsi_done callback passed from midlayer when
45invoking hostt->queuecommand() or SCSI midlayer will time it out.
46
47
48[1-2-1] Completing a scmd w/ scsi_done
49
50 For all non-EH commands, scsi_done() is the completion callback. It
51does the following.
52
53 1. Delete timeout timer. If it fails, it means that timeout timer
54    has expired and is going to finish the command. Just return.
55
56 2. Link scmd to per-cpu scsi_done_q using scmd->en_entry
57
58 3. Raise SCSI_SOFTIRQ
59
60 SCSI_SOFTIRQ handler scsi_softirq calls scsi_decide_disposition() to
61determine what to do with the command. scsi_decide_disposition()
62looks at the scmd->result value and sense data to determine what to do
63with the command.
64
65 - SUCCESS
66    scsi_finish_command() is invoked for the command. The
67    function does some maintenance choirs and notify completion by
68    calling scmd->done() callback, which, for fs requests, would
69    be HLD completion callback - sd:sd_rw_intr, sr:rw_intr,
70    st:st_intr.
71
72 - NEEDS_RETRY
73 - ADD_TO_MLQUEUE
74    scmd is requeued to blk queue.
75
76 - otherwise
77    scsi_eh_scmd_add(scmd, 0) is invoked for the command. See
78    [1-3] for details of this function.
79
80
81[1-2-2] Completing a scmd w/ timeout
82
83 The timeout handler is scsi_times_out(). When a timeout occurs, this
84function
85
86 1. invokes optional hostt->eh_timed_out() callback. Return value can
87    be one of
88
89    - EH_HANDLED
90    This indicates that eh_timed_out() dealt with the timeout. The
91    scmd is passed to __scsi_done() and thus linked into per-cpu
92    scsi_done_q. Normal command completion described in [1-2-1]
93    follows.
94
95    - EH_RESET_TIMER
96    This indicates that more time is required to finish the
97    command. Timer is restarted. This action is counted as a
98    retry and only allowed scmd->allowed + 1(!) times. Once the
99    limit is reached, action for EH_NOT_HANDLED is taken instead.
100
101    *NOTE* This action is racy as the LLDD could finish the scmd
102    after the timeout has expired but before it's added back. In
103    such cases, scsi_done() would think that timeout has occurred
104    and return without doing anything. We lose completion and the
105    command will time out again.
106
107    - EH_NOT_HANDLED
108    This is the same as when eh_timed_out() callback doesn't exist.
109    Step #2 is taken.
110
111 2. scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD) is invoked for the
112    command. See [1-3] for more information.
113
114
115[1-3] How EH takes over
116
117 scmds enter EH via scsi_eh_scmd_add(), which does the following.
118
119 1. Turns on scmd->eh_eflags as requested. It's 0 for error
120    completions and SCSI_EH_CANCEL_CMD for timeouts.
121
122 2. Links scmd->eh_entry to shost->eh_cmd_q
123
124 3. Sets SHOST_RECOVERY bit in shost->shost_state
125
126 4. Increments shost->host_failed
127
128 5. Wakes up SCSI EH thread if shost->host_busy == shost->host_failed
129
130 As can be seen above, once any scmd is added to shost->eh_cmd_q,
131SHOST_RECOVERY shost_state bit is turned on. This prevents any new
132scmd to be issued from blk queue to the host; eventually, all scmds on
133the host either complete normally, fail and get added to eh_cmd_q, or
134time out and get added to shost->eh_cmd_q.
135
136 If all scmds either complete or fail, the number of in-flight scmds
137becomes equal to the number of failed scmds - i.e. shost->host_busy ==
138shost->host_failed. This wakes up SCSI EH thread. So, once woken up,
139SCSI EH thread can expect that all in-flight commands have failed and
140are linked on shost->eh_cmd_q.
141
142 Note that this does not mean lower layers are quiescent. If a LLDD
143completed a scmd with error status, the LLDD and lower layers are
144assumed to forget about the scmd at that point. However, if a scmd
145has timed out, unless hostt->eh_timed_out() made lower layers forget
146about the scmd, which currently no LLDD does, the command is still
147active as long as lower layers are concerned and completion could
148occur at any time. Of course, all such completions are ignored as the
149timer has already expired.
150
151 We'll talk about how SCSI EH takes actions to abort - make LLDD
152forget about - timed out scmds later.
153
154
155[2] How SCSI EH works
156
157 LLDD's can implement SCSI EH actions in one of the following two
158ways.
159
160 - Fine-grained EH callbacks
161    LLDD can implement fine-grained EH callbacks and let SCSI
162    midlayer drive error handling and call appropriate callbacks.
163    This will be discussed further in [2-1].
164
165 - eh_strategy_handler() callback
166    This is one big callback which should perform whole error
167    handling. As such, it should do all choirs SCSI midlayer
168    performs during recovery. This will be discussed in [2-2].
169
170 Once recovery is complete, SCSI EH resumes normal operation by
171calling scsi_restart_operations(), which
172
173 1. Checks if door locking is needed and locks door.
174
175 2. Clears SHOST_RECOVERY shost_state bit
176
177 3. Wakes up waiters on shost->host_wait. This occurs if someone
178    calls scsi_block_when_processing_errors() on the host.
179    (*QUESTION* why is it needed? All operations will be blocked
180    anyway after it reaches blk queue.)
181
182 4. Kicks queues in all devices on the host in the asses
183
184
185[2-1] EH through fine-grained callbacks
186
187[2-1-1] Overview
188
189 If eh_strategy_handler() is not present, SCSI midlayer takes charge
190of driving error handling. EH's goals are two - make LLDD, host and
191device forget about timed out scmds and make them ready for new
192commands. A scmd is said to be recovered if the scmd is forgotten by
193lower layers and lower layers are ready to process or fail the scmd
194again.
195
196 To achieve these goals, EH performs recovery actions with increasing
197severity. Some actions are performed by issuing SCSI commands and
198others are performed by invoking one of the following fine-grained
199hostt EH callbacks. Callbacks may be omitted and omitted ones are
200considered to fail always.
201
202int (* eh_abort_handler)(struct scsi_cmnd *);
203int (* eh_device_reset_handler)(struct scsi_cmnd *);
204int (* eh_bus_reset_handler)(struct scsi_cmnd *);
205int (* eh_host_reset_handler)(struct scsi_cmnd *);
206
207 Higher-severity actions are taken only when lower-severity actions
208cannot recover some of failed scmds. Also, note that failure of the
209highest-severity action means EH failure and results in offlining of
210all unrecovered devices.
211
212 During recovery, the following rules are followed
213
214 - Recovery actions are performed on failed scmds on the to do list,
215   eh_work_q. If a recovery action succeeds for a scmd, recovered
216   scmds are removed from eh_work_q.
217
218   Note that single recovery action on a scmd can recover multiple
219   scmds. e.g. resetting a device recovers all failed scmds on the
220   device.
221
222 - Higher severity actions are taken iff eh_work_q is not empty after
223   lower severity actions are complete.
224
225 - EH reuses failed scmds to issue commands for recovery. For
226   timed-out scmds, SCSI EH ensures that LLDD forgets about a scmd
227   before reusing it for EH commands.
228
229 When a scmd is recovered, the scmd is moved from eh_work_q to EH
230local eh_done_q using scsi_eh_finish_cmd(). After all scmds are
231recovered (eh_work_q is empty), scsi_eh_flush_done_q() is invoked to
232either retry or error-finish (notify upper layer of failure) recovered
233scmds.
234
235 scmds are retried iff its sdev is still online (not offlined during
236EH), REQ_FAILFAST is not set and ++scmd->retries is less than
237scmd->allowed.
238
239
240[2-1-2] Flow of scmds through EH
241
242 1. Error completion / time out
243    ACTION: scsi_eh_scmd_add() is invoked for scmd
244    - set scmd->eh_eflags
245    - add scmd to shost->eh_cmd_q
246    - set SHOST_RECOVERY
247    - shost->host_failed++
248    LOCKING: shost->host_lock
249
250 2. EH starts
251    ACTION: move all scmds to EH's local eh_work_q. shost->eh_cmd_q
252        is cleared.
253    LOCKING: shost->host_lock (not strictly necessary, just for
254             consistency)
255
256 3. scmd recovered
257    ACTION: scsi_eh_finish_cmd() is invoked to EH-finish scmd
258    - shost->host_failed--
259    - clear scmd->eh_eflags
260    - scsi_setup_cmd_retry()
261    - move from local eh_work_q to local eh_done_q
262    LOCKING: none
263
264 4. EH completes
265    ACTION: scsi_eh_flush_done_q() retries scmds or notifies upper
266        layer of failure.
267    - scmd is removed from eh_done_q and scmd->eh_entry is cleared
268    - if retry is necessary, scmd is requeued using
269          scsi_queue_insert()
270    - otherwise, scsi_finish_command() is invoked for scmd
271    LOCKING: queue or finish function performs appropriate locking
272
273
274[2-1-3] Flow of control
275
276 EH through fine-grained callbacks start from scsi_unjam_host().
277
278<<scsi_unjam_host>>
279
280    1. Lock shost->host_lock, splice_init shost->eh_cmd_q into local
281       eh_work_q and unlock host_lock. Note that shost->eh_cmd_q is
282       cleared by this action.
283
284    2. Invoke scsi_eh_get_sense.
285
286    <<scsi_eh_get_sense>>
287
288    This action is taken for each error-completed
289    (!SCSI_EH_CANCEL_CMD) commands without valid sense data. Most
290    SCSI transports/LLDDs automatically acquire sense data on
291    command failures (autosense). Autosense is recommended for
292    performance reasons and as sense information could get out of
293    sync between occurrence of CHECK CONDITION and this action.
294
295    Note that if autosense is not supported, scmd->sense_buffer
296    contains invalid sense data when error-completing the scmd
297    with scsi_done(). scsi_decide_disposition() always returns
298    FAILED in such cases thus invoking SCSI EH. When the scmd
299    reaches here, sense data is acquired and
300    scsi_decide_disposition() is called again.
301
302    1. Invoke scsi_request_sense() which issues REQUEST_SENSE
303           command. If fails, no action. Note that taking no action
304           causes higher-severity recovery to be taken for the scmd.
305
306    2. Invoke scsi_decide_disposition() on the scmd
307
308       - SUCCESS
309        scmd->retries is set to scmd->allowed preventing
310        scsi_eh_flush_done_q() from retrying the scmd and
311        scsi_eh_finish_cmd() is invoked.
312
313       - NEEDS_RETRY
314        scsi_eh_finish_cmd() invoked
315
316       - otherwise
317        No action.
318
319    3. If !list_empty(&eh_work_q), invoke scsi_eh_abort_cmds().
320
321    <<scsi_eh_abort_cmds>>
322
323    This action is taken for each timed out command.
324    hostt->eh_abort_handler() is invoked for each scmd. The
325    handler returns SUCCESS if it has succeeded to make LLDD and
326    all related hardware forget about the scmd.
327
328    If a timedout scmd is successfully aborted and the sdev is
329    either offline or ready, scsi_eh_finish_cmd() is invoked for
330    the scmd. Otherwise, the scmd is left in eh_work_q for
331    higher-severity actions.
332
333    Note that both offline and ready status mean that the sdev is
334    ready to process new scmds, where processing also implies
335    immediate failing; thus, if a sdev is in one of the two
336    states, no further recovery action is needed.
337
338    Device readiness is tested using scsi_eh_tur() which issues
339    TEST_UNIT_READY command. Note that the scmd must have been
340    aborted successfully before reusing it for TEST_UNIT_READY.
341
342    4. If !list_empty(&eh_work_q), invoke scsi_eh_ready_devs()
343
344    <<scsi_eh_ready_devs>>
345
346    This function takes four increasingly more severe measures to
347    make failed sdevs ready for new commands.
348
349    1. Invoke scsi_eh_stu()
350
351    <<scsi_eh_stu>>
352
353        For each sdev which has failed scmds with valid sense data
354        of which scsi_check_sense()'s verdict is FAILED,
355        START_STOP_UNIT command is issued w/ start=1. Note that
356        as we explicitly choose error-completed scmds, it is known
357        that lower layers have forgotten about the scmd and we can
358        reuse it for STU.
359
360        If STU succeeds and the sdev is either offline or ready,
361        all failed scmds on the sdev are EH-finished with
362        scsi_eh_finish_cmd().
363
364        *NOTE* If hostt->eh_abort_handler() isn't implemented or
365        failed, we may still have timed out scmds at this point
366        and STU doesn't make lower layers forget about those
367        scmds. Yet, this function EH-finish all scmds on the sdev
368        if STU succeeds leaving lower layers in an inconsistent
369        state. It seems that STU action should be taken only when
370        a sdev has no timed out scmd.
371
372    2. If !list_empty(&eh_work_q), invoke scsi_eh_bus_device_reset().
373
374    <<scsi_eh_bus_device_reset>>
375
376        This action is very similar to scsi_eh_stu() except that,
377        instead of issuing STU, hostt->eh_device_reset_handler()
378        is used. Also, as we're not issuing SCSI commands and
379        resetting clears all scmds on the sdev, there is no need
380        to choose error-completed scmds.
381
382    3. If !list_empty(&eh_work_q), invoke scsi_eh_bus_reset()
383
384    <<scsi_eh_bus_reset>>
385
386        hostt->eh_bus_reset_handler() is invoked for each channel
387        with failed scmds. If bus reset succeeds, all failed
388        scmds on all ready or offline sdevs on the channel are
389        EH-finished.
390
391    4. If !list_empty(&eh_work_q), invoke scsi_eh_host_reset()
392
393    <<scsi_eh_host_reset>>
394
395        This is the last resort. hostt->eh_host_reset_handler()
396        is invoked. If host reset succeeds, all failed scmds on
397        all ready or offline sdevs on the host are EH-finished.
398
399    5. If !list_empty(&eh_work_q), invoke scsi_eh_offline_sdevs()
400
401    <<scsi_eh_offline_sdevs>>
402
403        Take all sdevs which still have unrecovered scmds offline
404        and EH-finish the scmds.
405
406    5. Invoke scsi_eh_flush_done_q().
407
408    <<scsi_eh_flush_done_q>>
409
410        At this point all scmds are recovered (or given up) and
411        put on eh_done_q by scsi_eh_finish_cmd(). This function
412        flushes eh_done_q by either retrying or notifying upper
413        layer of failure of the scmds.
414
415
416[2-2] EH through transportt->eh_strategy_handler()
417
418 transportt->eh_strategy_handler() is invoked in the place of
419scsi_unjam_host() and it is responsible for whole recovery process.
420On completion, the handler should have made lower layers forget about
421all failed scmds and either ready for new commands or offline. Also,
422it should perform SCSI EH maintenance choirs to maintain integrity of
423SCSI midlayer. IOW, of the steps described in [2-1-2], all steps
424except for #1 must be implemented by eh_strategy_handler().
425
426
427[2-2-1] Pre transportt->eh_strategy_handler() SCSI midlayer conditions
428
429 The following conditions are true on entry to the handler.
430
431 - Each failed scmd's eh_flags field is set appropriately.
432
433 - Each failed scmd is linked on scmd->eh_cmd_q by scmd->eh_entry.
434
435 - SHOST_RECOVERY is set.
436
437 - shost->host_failed == shost->host_busy
438
439
440[2-2-2] Post transportt->eh_strategy_handler() SCSI midlayer conditions
441
442 The following conditions must be true on exit from the handler.
443
444 - shost->host_failed is zero.
445
446 - Each scmd's eh_eflags field is cleared.
447
448 - Each scmd is in such a state that scsi_setup_cmd_retry() on the
449   scmd doesn't make any difference.
450
451 - shost->eh_cmd_q is cleared.
452
453 - Each scmd->eh_entry is cleared.
454
455 - Either scsi_queue_insert() or scsi_finish_command() is called on
456   each scmd. Note that the handler is free to use scmd->retries and
457   ->allowed to limit the number of retries.
458
459
460[2-2-3] Things to consider
461
462 - Know that timed out scmds are still active on lower layers. Make
463   lower layers forget about them before doing anything else with
464   those scmds.
465
466 - For consistency, when accessing/modifying shost data structure,
467   grab shost->host_lock.
468
469 - On completion, each failed sdev must have forgotten about all
470   active scmds.
471
472 - On completion, each failed sdev must be ready for new commands or
473   offline.
474
475
476--
477Tejun Heo
478htejun@gmail.com
47911th September 2005
480

Archive Download this file



interactive