547e85aa414f2b06a27a5b298e917c456e472f60
[linux-2.6.git] / drivers / scsi / scsi_error.c
1 /*
2  *  scsi_error.c Copyright (C) 1997 Eric Youngdale
3  *
4  *  SCSI error/timeout handling
5  *      Initial versions: Eric Youngdale.  Based upon conversations with
6  *                        Leonard Zubkoff and David Miller at Linux Expo, 
7  *                        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
13  *      minor  cleanups.
14  *      September 30, 2002 Mike Anderson (andmike@us.ibm.com)
15  */
16
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/timer.h>
20 #include <linux/string.h>
21 #include <linux/kernel.h>
22 #include <linux/freezer.h>
23 #include <linux/kthread.h>
24 #include <linux/interrupt.h>
25 #include <linux/blkdev.h>
26 #include <linux/delay.h>
27
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_eh.h>
33 #include <scsi/scsi_transport.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_ioctl.h>
36
37 #include "scsi_priv.h"
38 #include "scsi_logging.h"
39 #include "scsi_transport_api.h"
40
41 #define SENSE_TIMEOUT           (10*HZ)
42
43 /*
44  * These should *probably* be handled by the host itself.
45  * Since it is allowed to sleep, it probably should.
46  */
47 #define BUS_RESET_SETTLE_TIME   (10)
48 #define HOST_RESET_SETTLE_TIME  (10)
49
50 /* called with shost->host_lock held */
51 void scsi_eh_wakeup(struct Scsi_Host *shost)
52 {
53         if (shost->host_busy == shost->host_failed) {
54                 wake_up_process(shost->ehandler);
55                 SCSI_LOG_ERROR_RECOVERY(5,
56                                 printk("Waking error handler thread\n"));
57         }
58 }
59
60 /**
61  * scsi_schedule_eh - schedule EH for SCSI host
62  * @shost:      SCSI host to invoke error handling on.
63  *
64  * Schedule SCSI EH without scmd.
65  */
66 void scsi_schedule_eh(struct Scsi_Host *shost)
67 {
68         unsigned long flags;
69
70         spin_lock_irqsave(shost->host_lock, flags);
71
72         if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
73             scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
74                 shost->host_eh_scheduled++;
75                 scsi_eh_wakeup(shost);
76         }
77
78         spin_unlock_irqrestore(shost->host_lock, flags);
79 }
80 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
81
82 /**
83  * scsi_eh_scmd_add - add scsi cmd to error handling.
84  * @scmd:       scmd to run eh on.
85  * @eh_flag:    optional SCSI_EH flag.
86  *
87  * Return value:
88  *      0 on failure.
89  */
90 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
91 {
92         struct Scsi_Host *shost = scmd->device->host;
93         unsigned long flags;
94         int ret = 0;
95
96         if (!shost->ehandler)
97                 return 0;
98
99         spin_lock_irqsave(shost->host_lock, flags);
100         if (scsi_host_set_state(shost, SHOST_RECOVERY))
101                 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
102                         goto out_unlock;
103
104         ret = 1;
105         scmd->eh_eflags |= eh_flag;
106         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
107         shost->host_failed++;
108         scsi_eh_wakeup(shost);
109  out_unlock:
110         spin_unlock_irqrestore(shost->host_lock, flags);
111         return ret;
112 }
113
114 /**
115  * scsi_add_timer - Start timeout timer for a single scsi command.
116  * @scmd:       scsi command that is about to start running.
117  * @timeout:    amount of time to allow this command to run.
118  * @complete:   timeout function to call if timer isn't canceled.
119  *
120  * Notes:
121  *    This should be turned into an inline function.  Each scsi command
122  *    has its own timer, and as it is added to the queue, we set up the
123  *    timer.  When the command completes, we cancel the timer.
124  */
125 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
126                     void (*complete)(struct scsi_cmnd *))
127 {
128
129         /*
130          * If the clock was already running for this command, then
131          * first delete the timer.  The timer handling code gets rather
132          * confused if we don't do this.
133          */
134         if (scmd->eh_timeout.function)
135                 del_timer(&scmd->eh_timeout);
136
137         scmd->eh_timeout.data = (unsigned long)scmd;
138         scmd->eh_timeout.expires = jiffies + timeout;
139         scmd->eh_timeout.function = (void (*)(unsigned long)) complete;
140
141         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
142                                           " %d, (%p)\n", __FUNCTION__,
143                                           scmd, timeout, complete));
144
145         add_timer(&scmd->eh_timeout);
146 }
147
148 /**
149  * scsi_delete_timer - Delete/cancel timer for a given function.
150  * @scmd:       Cmd that we are canceling timer for
151  *
152  * Notes:
153  *     This should be turned into an inline function.
154  *
155  * Return value:
156  *     1 if we were able to detach the timer.  0 if we blew it, and the
157  *     timer function has already started to run.
158  */
159 int scsi_delete_timer(struct scsi_cmnd *scmd)
160 {
161         int rtn;
162
163         rtn = del_timer(&scmd->eh_timeout);
164
165         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
166                                          " rtn: %d\n", __FUNCTION__,
167                                          scmd, rtn));
168
169         scmd->eh_timeout.data = (unsigned long)NULL;
170         scmd->eh_timeout.function = NULL;
171
172         return rtn;
173 }
174
175 /**
176  * scsi_times_out - Timeout function for normal scsi commands.
177  * @scmd:       Cmd that is timing out.
178  *
179  * Notes:
180  *     We do not need to lock this.  There is the potential for a race
181  *     only in that the normal completion handling might run, but if the
182  *     normal completion function determines that the timer has already
183  *     fired, then it mustn't do anything.
184  */
185 void scsi_times_out(struct scsi_cmnd *scmd)
186 {
187         enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
188
189         scsi_log_completion(scmd, TIMEOUT_ERROR);
190
191         if (scmd->device->host->transportt->eh_timed_out)
192                 eh_timed_out = scmd->device->host->transportt->eh_timed_out;
193         else if (scmd->device->host->hostt->eh_timed_out)
194                 eh_timed_out = scmd->device->host->hostt->eh_timed_out;
195         else
196                 eh_timed_out = NULL;
197
198         if (eh_timed_out)
199                 switch (eh_timed_out(scmd)) {
200                 case EH_HANDLED:
201                         __scsi_done(scmd);
202                         return;
203                 case EH_RESET_TIMER:
204                         scsi_add_timer(scmd, scmd->timeout_per_command,
205                                        scsi_times_out);
206                         return;
207                 case EH_NOT_HANDLED:
208                         break;
209                 }
210
211         if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
212                 scmd->result |= DID_TIME_OUT << 16;
213                 __scsi_done(scmd);
214         }
215 }
216
217 /**
218  * scsi_block_when_processing_errors - Prevent cmds from being queued.
219  * @sdev:       Device on which we are performing recovery.
220  *
221  * Description:
222  *     We block until the host is out of error recovery, and then check to
223  *     see whether the host or the device is offline.
224  *
225  * Return value:
226  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
227  */
228 int scsi_block_when_processing_errors(struct scsi_device *sdev)
229 {
230         int online;
231
232         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
233
234         online = scsi_device_online(sdev);
235
236         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__,
237                                           online));
238
239         return online;
240 }
241 EXPORT_SYMBOL(scsi_block_when_processing_errors);
242
243 #ifdef CONFIG_SCSI_LOGGING
244 /**
245  * scsi_eh_prt_fail_stats - Log info on failures.
246  * @shost:      scsi host being recovered.
247  * @work_q:     Queue of scsi cmds to process.
248  */
249 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
250                                           struct list_head *work_q)
251 {
252         struct scsi_cmnd *scmd;
253         struct scsi_device *sdev;
254         int total_failures = 0;
255         int cmd_failed = 0;
256         int cmd_cancel = 0;
257         int devices_failed = 0;
258
259         shost_for_each_device(sdev, shost) {
260                 list_for_each_entry(scmd, work_q, eh_entry) {
261                         if (scmd->device == sdev) {
262                                 ++total_failures;
263                                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
264                                         ++cmd_cancel;
265                                 else 
266                                         ++cmd_failed;
267                         }
268                 }
269
270                 if (cmd_cancel || cmd_failed) {
271                         SCSI_LOG_ERROR_RECOVERY(3,
272                                 sdev_printk(KERN_INFO, sdev,
273                                             "%s: cmds failed: %d, cancel: %d\n",
274                                             __FUNCTION__, cmd_failed,
275                                             cmd_cancel));
276                         cmd_cancel = 0;
277                         cmd_failed = 0;
278                         ++devices_failed;
279                 }
280         }
281
282         SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
283                                           " devices require eh work\n",
284                                   total_failures, devices_failed));
285 }
286 #endif
287
288 /**
289  * scsi_check_sense - Examine scsi cmd sense
290  * @scmd:       Cmd to have sense checked.
291  *
292  * Return value:
293  *      SUCCESS or FAILED or NEEDS_RETRY
294  *
295  * Notes:
296  *      When a deferred error is detected the current command has
297  *      not been executed and needs retrying.
298  */
299 static int scsi_check_sense(struct scsi_cmnd *scmd)
300 {
301         struct scsi_sense_hdr sshdr;
302
303         if (! scsi_command_normalize_sense(scmd, &sshdr))
304                 return FAILED;  /* no valid sense data */
305
306         if (scsi_sense_is_deferred(&sshdr))
307                 return NEEDS_RETRY;
308
309         /*
310          * Previous logic looked for FILEMARK, EOM or ILI which are
311          * mainly associated with tapes and returned SUCCESS.
312          */
313         if (sshdr.response_code == 0x70) {
314                 /* fixed format */
315                 if (scmd->sense_buffer[2] & 0xe0)
316                         return SUCCESS;
317         } else {
318                 /*
319                  * descriptor format: look for "stream commands sense data
320                  * descriptor" (see SSC-3). Assume single sense data
321                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
322                  */
323                 if ((sshdr.additional_length > 3) &&
324                     (scmd->sense_buffer[8] == 0x4) &&
325                     (scmd->sense_buffer[11] & 0xe0))
326                         return SUCCESS;
327         }
328
329         switch (sshdr.sense_key) {
330         case NO_SENSE:
331                 return SUCCESS;
332         case RECOVERED_ERROR:
333                 return /* soft_error */ SUCCESS;
334
335         case ABORTED_COMMAND:
336                 return NEEDS_RETRY;
337         case NOT_READY:
338         case UNIT_ATTENTION:
339                 /*
340                  * if we are expecting a cc/ua because of a bus reset that we
341                  * performed, treat this just as a retry.  otherwise this is
342                  * information that we should pass up to the upper-level driver
343                  * so that we can deal with it there.
344                  */
345                 if (scmd->device->expecting_cc_ua) {
346                         scmd->device->expecting_cc_ua = 0;
347                         return NEEDS_RETRY;
348                 }
349                 /*
350                  * if the device is in the process of becoming ready, we 
351                  * should retry.
352                  */
353                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
354                         return NEEDS_RETRY;
355                 /*
356                  * if the device is not started, we need to wake
357                  * the error handler to start the motor
358                  */
359                 if (scmd->device->allow_restart &&
360                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
361                         return FAILED;
362                 return SUCCESS;
363
364                 /* these three are not supported */
365         case COPY_ABORTED:
366         case VOLUME_OVERFLOW:
367         case MISCOMPARE:
368                 return SUCCESS;
369
370         case MEDIUM_ERROR:
371                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
372                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
373                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
374                         return SUCCESS;
375                 }
376                 return NEEDS_RETRY;
377
378         case HARDWARE_ERROR:
379                 if (scmd->device->retry_hwerror)
380                         return NEEDS_RETRY;
381                 else
382                         return SUCCESS;
383
384         case ILLEGAL_REQUEST:
385         case BLANK_CHECK:
386         case DATA_PROTECT:
387         default:
388                 return SUCCESS;
389         }
390 }
391
392 /**
393  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
394  * @scmd:       SCSI cmd to examine.
395  *
396  * Notes:
397  *    This is *only* called when we are examining the status of commands
398  *    queued during error recovery.  the main difference here is that we
399  *    don't allow for the possibility of retries here, and we are a lot
400  *    more restrictive about what we consider acceptable.
401  */
402 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
403 {
404         /*
405          * first check the host byte, to see if there is anything in there
406          * that would indicate what we need to do.
407          */
408         if (host_byte(scmd->result) == DID_RESET) {
409                 /*
410                  * rats.  we are already in the error handler, so we now
411                  * get to try and figure out what to do next.  if the sense
412                  * is valid, we have a pretty good idea of what to do.
413                  * if not, we mark it as FAILED.
414                  */
415                 return scsi_check_sense(scmd);
416         }
417         if (host_byte(scmd->result) != DID_OK)
418                 return FAILED;
419
420         /*
421          * next, check the message byte.
422          */
423         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
424                 return FAILED;
425
426         /*
427          * now, check the status byte to see if this indicates
428          * anything special.
429          */
430         switch (status_byte(scmd->result)) {
431         case GOOD:
432         case COMMAND_TERMINATED:
433                 return SUCCESS;
434         case CHECK_CONDITION:
435                 return scsi_check_sense(scmd);
436         case CONDITION_GOOD:
437         case INTERMEDIATE_GOOD:
438         case INTERMEDIATE_C_GOOD:
439                 /*
440                  * who knows?  FIXME(eric)
441                  */
442                 return SUCCESS;
443         case BUSY:
444         case QUEUE_FULL:
445         case RESERVATION_CONFLICT:
446         default:
447                 return FAILED;
448         }
449         return FAILED;
450 }
451
452 /**
453  * scsi_eh_done - Completion function for error handling.
454  * @scmd:       Cmd that is done.
455  */
456 static void scsi_eh_done(struct scsi_cmnd *scmd)
457 {
458         struct completion     *eh_action;
459
460         SCSI_LOG_ERROR_RECOVERY(3,
461                 printk("%s scmd: %p result: %x\n",
462                         __FUNCTION__, scmd, scmd->result));
463
464         eh_action = scmd->device->host->eh_action;
465         if (eh_action)
466                 complete(eh_action);
467 }
468
469 /**
470  * scsi_try_host_reset - ask host adapter to reset itself
471  * @scmd:       SCSI cmd to send hsot reset.
472  */
473 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
474 {
475         unsigned long flags;
476         int rtn;
477
478         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
479                                           __FUNCTION__));
480
481         if (!scmd->device->host->hostt->eh_host_reset_handler)
482                 return FAILED;
483
484         rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
485
486         if (rtn == SUCCESS) {
487                 if (!scmd->device->host->hostt->skip_settle_delay)
488                         ssleep(HOST_RESET_SETTLE_TIME);
489                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
490                 scsi_report_bus_reset(scmd->device->host,
491                                       scmd_channel(scmd));
492                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
493         }
494
495         return rtn;
496 }
497
498 /**
499  * scsi_try_bus_reset - ask host to perform a bus reset
500  * @scmd:       SCSI cmd to send bus reset.
501  */
502 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
503 {
504         unsigned long flags;
505         int rtn;
506
507         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
508                                           __FUNCTION__));
509
510         if (!scmd->device->host->hostt->eh_bus_reset_handler)
511                 return FAILED;
512
513         rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
514
515         if (rtn == SUCCESS) {
516                 if (!scmd->device->host->hostt->skip_settle_delay)
517                         ssleep(BUS_RESET_SETTLE_TIME);
518                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
519                 scsi_report_bus_reset(scmd->device->host,
520                                       scmd_channel(scmd));
521                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
522         }
523
524         return rtn;
525 }
526
527 /**
528  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
529  * @scmd:       SCSI cmd used to send BDR
530  *
531  * Notes:
532  *    There is no timeout for this operation.  if this operation is
533  *    unreliable for a given host, then the host itself needs to put a
534  *    timer on it, and set the host back to a consistent state prior to
535  *    returning.
536  */
537 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
538 {
539         int rtn;
540
541         if (!scmd->device->host->hostt->eh_device_reset_handler)
542                 return FAILED;
543
544         rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
545         if (rtn == SUCCESS) {
546                 scmd->device->was_reset = 1;
547                 scmd->device->expecting_cc_ua = 1;
548         }
549
550         return rtn;
551 }
552
553 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
554 {
555         if (!scmd->device->host->hostt->eh_abort_handler)
556                 return FAILED;
557
558         return scmd->device->host->hostt->eh_abort_handler(scmd);
559 }
560
561 /**
562  * scsi_try_to_abort_cmd - Ask host to abort a running command.
563  * @scmd:       SCSI cmd to abort from Lower Level.
564  *
565  * Notes:
566  *    This function will not return until the user's completion function
567  *    has been called.  there is no timeout on this operation.  if the
568  *    author of the low-level driver wishes this operation to be timed,
569  *    they can provide this facility themselves.  helper functions in
570  *    scsi_error.c can be supplied to make this easier to do.
571  */
572 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
573 {
574         /*
575          * scsi_done was called just after the command timed out and before
576          * we had a chance to process it. (db)
577          */
578         if (scmd->serial_number == 0)
579                 return SUCCESS;
580         return __scsi_try_to_abort_cmd(scmd);
581 }
582
583 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
584 {
585         if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
586                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
587                         if (scsi_try_bus_reset(scmd) != SUCCESS)
588                                 scsi_try_host_reset(scmd);
589 }
590
591 /**
592  * scsi_eh_prep_cmnd  - Save a scsi command info as part of error recory
593  * @scmd:       SCSI command structure to hijack
594  * @ses:        structure to save restore information
595  * @cmnd:       CDB to send. Can be NULL if no new cmnd is needed
596  * @cmnd_size:  size in bytes of @cmnd
597  * @sense_bytes: size of sense data to copy. or 0 (if != 0 @cmnd is ignored)
598  *
599  * This function is used to save a scsi command information before re-execution
600  * as part of the error recovery process.  If @sense_bytes is 0 the command
601  * sent must be one that does not transfer any data.  If @sense_bytes != 0
602  * @cmnd is ignored and this functions sets up a REQUEST_SENSE command
603  * and cmnd buffers to read @sense_bytes into @scmd->sense_buffer.
604  */
605 void scsi_eh_prep_cmnd(struct scsi_cmnd *scmd, struct scsi_eh_save *ses,
606                         unsigned char *cmnd, int cmnd_size, unsigned sense_bytes)
607 {
608         struct scsi_device *sdev = scmd->device;
609
610         /*
611          * We need saved copies of a number of fields - this is because
612          * error handling may need to overwrite these with different values
613          * to run different commands, and once error handling is complete,
614          * we will need to restore these values prior to running the actual
615          * command.
616          */
617         ses->cmd_len = scmd->cmd_len;
618         memcpy(ses->cmnd, scmd->cmnd, sizeof(scmd->cmnd));
619         ses->data_direction = scmd->sc_data_direction;
620         ses->bufflen = scmd->request_bufflen;
621         ses->buffer = scmd->request_buffer;
622         ses->use_sg = scmd->use_sg;
623         ses->resid = scmd->resid;
624         ses->result = scmd->result;
625
626         if (sense_bytes) {
627                 scmd->request_bufflen = min_t(unsigned,
628                                        SCSI_SENSE_BUFFERSIZE, sense_bytes);
629                 sg_init_one(&ses->sense_sgl, scmd->sense_buffer,
630                                                        scmd->request_bufflen);
631                 scmd->request_buffer = &ses->sense_sgl;
632                 scmd->sc_data_direction = DMA_FROM_DEVICE;
633                 scmd->use_sg = 1;
634                 memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
635                 scmd->cmnd[0] = REQUEST_SENSE;
636                 scmd->cmnd[4] = scmd->request_bufflen;
637                 scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
638         } else {
639                 scmd->request_buffer = NULL;
640                 scmd->request_bufflen = 0;
641                 scmd->sc_data_direction = DMA_NONE;
642                 scmd->use_sg = 0;
643                 if (cmnd) {
644                         memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
645                         memcpy(scmd->cmnd, cmnd, cmnd_size);
646                         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
647                 }
648         }
649
650         scmd->underflow = 0;
651
652         if (sdev->scsi_level <= SCSI_2 && sdev->scsi_level != SCSI_UNKNOWN)
653                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
654                         (sdev->lun << 5 & 0xe0);
655
656         /*
657          * Zero the sense buffer.  The scsi spec mandates that any
658          * untransferred sense data should be interpreted as being zero.
659          */
660         memset(scmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
661 }
662 EXPORT_SYMBOL(scsi_eh_prep_cmnd);
663
664 /**
665  * scsi_eh_restore_cmnd  - Restore a scsi command info as part of error recory
666  * @scmd:       SCSI command structure to restore
667  * @ses:        saved information from a coresponding call to scsi_prep_eh_cmnd
668  *
669  * Undo any damage done by above scsi_prep_eh_cmnd().
670  */
671 void scsi_eh_restore_cmnd(struct scsi_cmnd* scmd, struct scsi_eh_save *ses)
672 {
673         /*
674          * Restore original data
675          */
676         scmd->cmd_len = ses->cmd_len;
677         memcpy(scmd->cmnd, ses->cmnd, sizeof(scmd->cmnd));
678         scmd->sc_data_direction = ses->data_direction;
679         scmd->request_bufflen = ses->bufflen;
680         scmd->request_buffer = ses->buffer;
681         scmd->use_sg = ses->use_sg;
682         scmd->resid = ses->resid;
683         scmd->result = ses->result;
684 }
685 EXPORT_SYMBOL(scsi_eh_restore_cmnd);
686
687 /**
688  * scsi_send_eh_cmnd  - submit a scsi command as part of error recory
689  * @scmd:       SCSI command structure to hijack
690  * @cmnd:       CDB to send
691  * @cmnd_size:  size in bytes of @cmnd
692  * @timeout:    timeout for this request
693  * @sense_bytes: size of sense data to copy or 0
694  *
695  * This function is used to send a scsi command down to a target device
696  * as part of the error recovery process. See also scsi_eh_prep_cmnd() above.
697  *
698  * Return value:
699  *    SUCCESS or FAILED or NEEDS_RETRY
700  */
701 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
702                              int cmnd_size, int timeout, unsigned sense_bytes)
703 {
704         struct scsi_device *sdev = scmd->device;
705         struct Scsi_Host *shost = sdev->host;
706         DECLARE_COMPLETION_ONSTACK(done);
707         unsigned long timeleft;
708         unsigned long flags;
709         struct scsi_eh_save ses;
710         int rtn;
711
712         scsi_eh_prep_cmnd(scmd, &ses, cmnd, cmnd_size, sense_bytes);
713         shost->eh_action = &done;
714
715         spin_lock_irqsave(shost->host_lock, flags);
716         scsi_log_send(scmd);
717         shost->hostt->queuecommand(scmd, scsi_eh_done);
718         spin_unlock_irqrestore(shost->host_lock, flags);
719
720         timeleft = wait_for_completion_timeout(&done, timeout);
721
722         shost->eh_action = NULL;
723
724         scsi_log_completion(scmd, SUCCESS);
725
726         SCSI_LOG_ERROR_RECOVERY(3,
727                 printk("%s: scmd: %p, timeleft: %ld\n",
728                         __FUNCTION__, scmd, timeleft));
729
730         /*
731          * If there is time left scsi_eh_done got called, and we will
732          * examine the actual status codes to see whether the command
733          * actually did complete normally, else tell the host to forget
734          * about this command.
735          */
736         if (timeleft) {
737                 rtn = scsi_eh_completed_normally(scmd);
738                 SCSI_LOG_ERROR_RECOVERY(3,
739                         printk("%s: scsi_eh_completed_normally %x\n",
740                                __FUNCTION__, rtn));
741
742                 switch (rtn) {
743                 case SUCCESS:
744                 case NEEDS_RETRY:
745                 case FAILED:
746                         break;
747                 default:
748                         rtn = FAILED;
749                         break;
750                 }
751         } else {
752                 scsi_abort_eh_cmnd(scmd);
753                 rtn = FAILED;
754         }
755
756         scsi_eh_restore_cmnd(scmd, &ses);
757         return rtn;
758 }
759
760 /**
761  * scsi_request_sense - Request sense data from a particular target.
762  * @scmd:       SCSI cmd for request sense.
763  *
764  * Notes:
765  *    Some hosts automatically obtain this information, others require
766  *    that we obtain it on our own. This function will *not* return until
767  *    the command either times out, or it completes.
768  */
769 static int scsi_request_sense(struct scsi_cmnd *scmd)
770 {
771         return scsi_send_eh_cmnd(scmd, NULL, 0, SENSE_TIMEOUT, ~0);
772 }
773
774 /**
775  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
776  * @scmd:       Original SCSI cmd that eh has finished.
777  * @done_q:     Queue for processed commands.
778  *
779  * Notes:
780  *    We don't want to use the normal command completion while we are are
781  *    still handling errors - it may cause other commands to be queued,
782  *    and that would disturb what we are doing.  Thus we really want to
783  *    keep a list of pending commands for final completion, and once we
784  *    are ready to leave error handling we handle completion for real.
785  */
786 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
787 {
788         scmd->device->host->host_failed--;
789         scmd->eh_eflags = 0;
790         list_move_tail(&scmd->eh_entry, done_q);
791 }
792 EXPORT_SYMBOL(scsi_eh_finish_cmd);
793
794 /**
795  * scsi_eh_get_sense - Get device sense data.
796  * @work_q:     Queue of commands to process.
797  * @done_q:     Queue of processed commands.
798  *
799  * Description:
800  *    See if we need to request sense information.  if so, then get it
801  *    now, so we have a better idea of what to do.  
802  *
803  * Notes:
804  *    This has the unfortunate side effect that if a shost adapter does
805  *    not automatically request sense information, we end up shutting
806  *    it down before we request it.
807  *
808  *    All drivers should request sense information internally these days,
809  *    so for now all I have to say is tough noogies if you end up in here.
810  *
811  *    XXX: Long term this code should go away, but that needs an audit of
812  *         all LLDDs first.
813  */
814 int scsi_eh_get_sense(struct list_head *work_q,
815                       struct list_head *done_q)
816 {
817         struct scsi_cmnd *scmd, *next;
818         int rtn;
819
820         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
821                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
822                     SCSI_SENSE_VALID(scmd))
823                         continue;
824
825                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
826                                                   "%s: requesting sense\n",
827                                                   current->comm));
828                 rtn = scsi_request_sense(scmd);
829                 if (rtn != SUCCESS)
830                         continue;
831
832                 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
833                                                   " result %x\n", scmd,
834                                                   scmd->result));
835                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
836
837                 rtn = scsi_decide_disposition(scmd);
838
839                 /*
840                  * if the result was normal, then just pass it along to the
841                  * upper level.
842                  */
843                 if (rtn == SUCCESS)
844                         /* we don't want this command reissued, just
845                          * finished with the sense data, so set
846                          * retries to the max allowed to ensure it
847                          * won't get reissued */
848                         scmd->retries = scmd->allowed;
849                 else if (rtn != NEEDS_RETRY)
850                         continue;
851
852                 scsi_eh_finish_cmd(scmd, done_q);
853         }
854
855         return list_empty(work_q);
856 }
857 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
858
859 /**
860  * scsi_eh_tur - Send TUR to device.
861  * @scmd:       &scsi_cmnd to send TUR
862  *
863  * Return value:
864  *    0 - Device is ready. 1 - Device NOT ready.
865  */
866 static int scsi_eh_tur(struct scsi_cmnd *scmd)
867 {
868         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
869         int retry_cnt = 1, rtn;
870
871 retry_tur:
872         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
873
874         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
875                 __FUNCTION__, scmd, rtn));
876
877         switch (rtn) {
878         case NEEDS_RETRY:
879                 if (retry_cnt--)
880                         goto retry_tur;
881                 /*FALLTHRU*/
882         case SUCCESS:
883                 return 0;
884         default:
885                 return 1;
886         }
887 }
888
889 /**
890  * scsi_eh_abort_cmds - abort pending commands.
891  * @work_q:     &list_head for pending commands.
892  * @done_q:     &list_head for processed commands.
893  *
894  * Decription:
895  *    Try and see whether or not it makes sense to try and abort the
896  *    running command.  This only works out to be the case if we have one
897  *    command that has timed out.  If the command simply failed, it makes
898  *    no sense to try and abort the command, since as far as the shost
899  *    adapter is concerned, it isn't running.
900  */
901 static int scsi_eh_abort_cmds(struct list_head *work_q,
902                               struct list_head *done_q)
903 {
904         struct scsi_cmnd *scmd, *next;
905         int rtn;
906
907         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
908                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
909                         continue;
910                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
911                                                   "0x%p\n", current->comm,
912                                                   scmd));
913                 rtn = scsi_try_to_abort_cmd(scmd);
914                 if (rtn == SUCCESS) {
915                         scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
916                         if (!scsi_device_online(scmd->device) ||
917                             !scsi_eh_tur(scmd)) {
918                                 scsi_eh_finish_cmd(scmd, done_q);
919                         }
920                                 
921                 } else
922                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
923                                                           " cmd failed:"
924                                                           "0x%p\n",
925                                                           current->comm,
926                                                           scmd));
927         }
928
929         return list_empty(work_q);
930 }
931
932 /**
933  * scsi_eh_try_stu - Send START_UNIT to device.
934  * @scmd:       &scsi_cmnd to send START_UNIT
935  *
936  * Return value:
937  *    0 - Device is ready. 1 - Device NOT ready.
938  */
939 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
940 {
941         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
942
943         if (scmd->device->allow_restart) {
944                 int i, rtn = NEEDS_RETRY;
945
946                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
947                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
948                                                 scmd->device->timeout, 0);
949
950                 if (rtn == SUCCESS)
951                         return 0;
952         }
953
954         return 1;
955 }
956
957  /**
958  * scsi_eh_stu - send START_UNIT if needed
959  * @shost:      &scsi host being recovered.
960  * @work_q:     &list_head for pending commands.
961  * @done_q:     &list_head for processed commands.
962  *
963  * Notes:
964  *    If commands are failing due to not ready, initializing command required,
965  *      try revalidating the device, which will end up sending a start unit. 
966  */
967 static int scsi_eh_stu(struct Scsi_Host *shost,
968                               struct list_head *work_q,
969                               struct list_head *done_q)
970 {
971         struct scsi_cmnd *scmd, *stu_scmd, *next;
972         struct scsi_device *sdev;
973
974         shost_for_each_device(sdev, shost) {
975                 stu_scmd = NULL;
976                 list_for_each_entry(scmd, work_q, eh_entry)
977                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
978                             scsi_check_sense(scmd) == FAILED ) {
979                                 stu_scmd = scmd;
980                                 break;
981                         }
982
983                 if (!stu_scmd)
984                         continue;
985
986                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
987                                                   " 0x%p\n", current->comm, sdev));
988
989                 if (!scsi_eh_try_stu(stu_scmd)) {
990                         if (!scsi_device_online(sdev) ||
991                             !scsi_eh_tur(stu_scmd)) {
992                                 list_for_each_entry_safe(scmd, next,
993                                                           work_q, eh_entry) {
994                                         if (scmd->device == sdev)
995                                                 scsi_eh_finish_cmd(scmd, done_q);
996                                 }
997                         }
998                 } else {
999                         SCSI_LOG_ERROR_RECOVERY(3,
1000                                                 printk("%s: START_UNIT failed to sdev:"
1001                                                        " 0x%p\n", current->comm, sdev));
1002                 }
1003         }
1004
1005         return list_empty(work_q);
1006 }
1007
1008
1009 /**
1010  * scsi_eh_bus_device_reset - send bdr if needed
1011  * @shost:      scsi host being recovered.
1012  * @work_q:     &list_head for pending commands.
1013  * @done_q:     &list_head for processed commands.
1014  *
1015  * Notes:
1016  *    Try a bus device reset.  Still, look to see whether we have multiple
1017  *    devices that are jammed or not - if we have multiple devices, it
1018  *    makes no sense to try bus_device_reset - we really would need to try
1019  *    a bus_reset instead. 
1020  */
1021 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1022                                     struct list_head *work_q,
1023                                     struct list_head *done_q)
1024 {
1025         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1026         struct scsi_device *sdev;
1027         int rtn;
1028
1029         shost_for_each_device(sdev, shost) {
1030                 bdr_scmd = NULL;
1031                 list_for_each_entry(scmd, work_q, eh_entry)
1032                         if (scmd->device == sdev) {
1033                                 bdr_scmd = scmd;
1034                                 break;
1035                         }
1036
1037                 if (!bdr_scmd)
1038                         continue;
1039
1040                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1041                                                   " 0x%p\n", current->comm,
1042                                                   sdev));
1043                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1044                 if (rtn == SUCCESS) {
1045                         if (!scsi_device_online(sdev) ||
1046                             !scsi_eh_tur(bdr_scmd)) {
1047                                 list_for_each_entry_safe(scmd, next,
1048                                                          work_q, eh_entry) {
1049                                         if (scmd->device == sdev)
1050                                                 scsi_eh_finish_cmd(scmd,
1051                                                                    done_q);
1052                                 }
1053                         }
1054                 } else {
1055                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1056                                                           " failed sdev:"
1057                                                           "0x%p\n",
1058                                                           current->comm,
1059                                                            sdev));
1060                 }
1061         }
1062
1063         return list_empty(work_q);
1064 }
1065
1066 /**
1067  * scsi_eh_bus_reset - send a bus reset 
1068  * @shost:      &scsi host being recovered.
1069  * @work_q:     &list_head for pending commands.
1070  * @done_q:     &list_head for processed commands.
1071  */
1072 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1073                              struct list_head *work_q,
1074                              struct list_head *done_q)
1075 {
1076         struct scsi_cmnd *scmd, *chan_scmd, *next;
1077         unsigned int channel;
1078         int rtn;
1079
1080         /*
1081          * we really want to loop over the various channels, and do this on
1082          * a channel by channel basis.  we should also check to see if any
1083          * of the failed commands are on soft_reset devices, and if so, skip
1084          * the reset.  
1085          */
1086
1087         for (channel = 0; channel <= shost->max_channel; channel++) {
1088                 chan_scmd = NULL;
1089                 list_for_each_entry(scmd, work_q, eh_entry) {
1090                         if (channel == scmd_channel(scmd)) {
1091                                 chan_scmd = scmd;
1092                                 break;
1093                                 /*
1094                                  * FIXME add back in some support for
1095                                  * soft_reset devices.
1096                                  */
1097                         }
1098                 }
1099
1100                 if (!chan_scmd)
1101                         continue;
1102                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1103                                                   " %d\n", current->comm,
1104                                                   channel));
1105                 rtn = scsi_try_bus_reset(chan_scmd);
1106                 if (rtn == SUCCESS) {
1107                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1108                                 if (channel == scmd_channel(scmd))
1109                                         if (!scsi_device_online(scmd->device) ||
1110                                             !scsi_eh_tur(scmd))
1111                                                 scsi_eh_finish_cmd(scmd,
1112                                                                    done_q);
1113                         }
1114                 } else {
1115                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1116                                                           " failed chan: %d\n",
1117                                                           current->comm,
1118                                                           channel));
1119                 }
1120         }
1121         return list_empty(work_q);
1122 }
1123
1124 /**
1125  * scsi_eh_host_reset - send a host reset 
1126  * @work_q:     list_head for processed commands.
1127  * @done_q:     list_head for processed commands.
1128  */
1129 static int scsi_eh_host_reset(struct list_head *work_q,
1130                               struct list_head *done_q)
1131 {
1132         struct scsi_cmnd *scmd, *next;
1133         int rtn;
1134
1135         if (!list_empty(work_q)) {
1136                 scmd = list_entry(work_q->next,
1137                                   struct scsi_cmnd, eh_entry);
1138
1139                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1140                                                   , current->comm));
1141
1142                 rtn = scsi_try_host_reset(scmd);
1143                 if (rtn == SUCCESS) {
1144                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1145                                 if (!scsi_device_online(scmd->device) ||
1146                                     (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1147                                     !scsi_eh_tur(scmd))
1148                                         scsi_eh_finish_cmd(scmd, done_q);
1149                         }
1150                 } else {
1151                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1152                                                           " failed\n",
1153                                                           current->comm));
1154                 }
1155         }
1156         return list_empty(work_q);
1157 }
1158
1159 /**
1160  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1161  * @work_q:     list_head for processed commands.
1162  * @done_q:     list_head for processed commands.
1163  */
1164 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1165                                   struct list_head *done_q)
1166 {
1167         struct scsi_cmnd *scmd, *next;
1168
1169         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1170                 sdev_printk(KERN_INFO, scmd->device, "Device offlined - "
1171                             "not ready after error recovery\n");
1172                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1173                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1174                         /*
1175                          * FIXME: Handle lost cmds.
1176                          */
1177                 }
1178                 scsi_eh_finish_cmd(scmd, done_q);
1179         }
1180         return;
1181 }
1182
1183 /**
1184  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1185  * @scmd:       SCSI cmd to examine.
1186  *
1187  * Notes:
1188  *    This is *only* called when we are examining the status after sending
1189  *    out the actual data command.  any commands that are queued for error
1190  *    recovery (e.g. test_unit_ready) do *not* come through here.
1191  *
1192  *    When this routine returns failed, it means the error handler thread
1193  *    is woken.  In cases where the error code indicates an error that
1194  *    doesn't require the error handler read (i.e. we don't need to
1195  *    abort/reset), this function should return SUCCESS.
1196  */
1197 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1198 {
1199         int rtn;
1200
1201         /*
1202          * if the device is offline, then we clearly just pass the result back
1203          * up to the top level.
1204          */
1205         if (!scsi_device_online(scmd->device)) {
1206                 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1207                                                   " as SUCCESS\n",
1208                                                   __FUNCTION__));
1209                 return SUCCESS;
1210         }
1211
1212         /*
1213          * first check the host byte, to see if there is anything in there
1214          * that would indicate what we need to do.
1215          */
1216         switch (host_byte(scmd->result)) {
1217         case DID_PASSTHROUGH:
1218                 /*
1219                  * no matter what, pass this through to the upper layer.
1220                  * nuke this special code so that it looks like we are saying
1221                  * did_ok.
1222                  */
1223                 scmd->result &= 0xff00ffff;
1224                 return SUCCESS;
1225         case DID_OK:
1226                 /*
1227                  * looks good.  drop through, and check the next byte.
1228                  */
1229                 break;
1230         case DID_NO_CONNECT:
1231         case DID_BAD_TARGET:
1232         case DID_ABORT:
1233                 /*
1234                  * note - this means that we just report the status back
1235                  * to the top level driver, not that we actually think
1236                  * that it indicates SUCCESS.
1237                  */
1238                 return SUCCESS;
1239                 /*
1240                  * when the low level driver returns did_soft_error,
1241                  * it is responsible for keeping an internal retry counter 
1242                  * in order to avoid endless loops (db)
1243                  *
1244                  * actually this is a bug in this function here.  we should
1245                  * be mindful of the maximum number of retries specified
1246                  * and not get stuck in a loop.
1247                  */
1248         case DID_SOFT_ERROR:
1249                 goto maybe_retry;
1250         case DID_IMM_RETRY:
1251                 return NEEDS_RETRY;
1252
1253         case DID_REQUEUE:
1254                 return ADD_TO_MLQUEUE;
1255
1256         case DID_ERROR:
1257                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1258                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1259                         /*
1260                          * execute reservation conflict processing code
1261                          * lower down
1262                          */
1263                         break;
1264                 /* fallthrough */
1265
1266         case DID_BUS_BUSY:
1267         case DID_PARITY:
1268                 goto maybe_retry;
1269         case DID_TIME_OUT:
1270                 /*
1271                  * when we scan the bus, we get timeout messages for
1272                  * these commands if there is no device available.
1273                  * other hosts report did_no_connect for the same thing.
1274                  */
1275                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1276                      scmd->cmnd[0] == INQUIRY)) {
1277                         return SUCCESS;
1278                 } else {
1279                         return FAILED;
1280                 }
1281         case DID_RESET:
1282                 return SUCCESS;
1283         default:
1284                 return FAILED;
1285         }
1286
1287         /*
1288          * next, check the message byte.
1289          */
1290         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1291                 return FAILED;
1292
1293         /*
1294          * check the status byte to see if this indicates anything special.
1295          */
1296         switch (status_byte(scmd->result)) {
1297         case QUEUE_FULL:
1298                 /*
1299                  * the case of trying to send too many commands to a
1300                  * tagged queueing device.
1301                  */
1302         case BUSY:
1303                 /*
1304                  * device can't talk to us at the moment.  Should only
1305                  * occur (SAM-3) when the task queue is empty, so will cause
1306                  * the empty queue handling to trigger a stall in the
1307                  * device.
1308                  */
1309                 return ADD_TO_MLQUEUE;
1310         case GOOD:
1311         case COMMAND_TERMINATED:
1312         case TASK_ABORTED:
1313                 return SUCCESS;
1314         case CHECK_CONDITION:
1315                 rtn = scsi_check_sense(scmd);
1316                 if (rtn == NEEDS_RETRY)
1317                         goto maybe_retry;
1318                 /* if rtn == FAILED, we have no sense information;
1319                  * returning FAILED will wake the error handler thread
1320                  * to collect the sense and redo the decide
1321                  * disposition */
1322                 return rtn;
1323         case CONDITION_GOOD:
1324         case INTERMEDIATE_GOOD:
1325         case INTERMEDIATE_C_GOOD:
1326         case ACA_ACTIVE:
1327                 /*
1328                  * who knows?  FIXME(eric)
1329                  */
1330                 return SUCCESS;
1331
1332         case RESERVATION_CONFLICT:
1333                 sdev_printk(KERN_INFO, scmd->device,
1334                             "reservation conflict\n");
1335                 return SUCCESS; /* causes immediate i/o error */
1336         default:
1337                 return FAILED;
1338         }
1339         return FAILED;
1340
1341       maybe_retry:
1342
1343         /* we requeue for retry because the error was retryable, and
1344          * the request was not marked fast fail.  Note that above,
1345          * even if the request is marked fast fail, we still requeue
1346          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1347         if ((++scmd->retries) <= scmd->allowed
1348             && !blk_noretry_request(scmd->request)) {
1349                 return NEEDS_RETRY;
1350         } else {
1351                 /*
1352                  * no more retries - report this one back to upper level.
1353                  */
1354                 return SUCCESS;
1355         }
1356 }
1357
1358 /**
1359  * scsi_eh_lock_door - Prevent medium removal for the specified device
1360  * @sdev:       SCSI device to prevent medium removal
1361  *
1362  * Locking:
1363  *      We must be called from process context; scsi_allocate_request()
1364  *      may sleep.
1365  *
1366  * Notes:
1367  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1368  *      head of the devices request queue, and continue.
1369  *
1370  * Bugs:
1371  *      scsi_allocate_request() may sleep waiting for existing requests to
1372  *      be processed.  However, since we haven't kicked off any request
1373  *      processing for this host, this may deadlock.
1374  *
1375  *      If scsi_allocate_request() fails for what ever reason, we
1376  *      completely forget to lock the door.
1377  */
1378 static void scsi_eh_lock_door(struct scsi_device *sdev)
1379 {
1380         unsigned char cmnd[MAX_COMMAND_SIZE];
1381
1382         cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1383         cmnd[1] = 0;
1384         cmnd[2] = 0;
1385         cmnd[3] = 0;
1386         cmnd[4] = SCSI_REMOVAL_PREVENT;
1387         cmnd[5] = 0;
1388
1389         scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
1390                            5, NULL, NULL, GFP_KERNEL);
1391 }
1392
1393
1394 /**
1395  * scsi_restart_operations - restart io operations to the specified host.
1396  * @shost:      Host we are restarting.
1397  *
1398  * Notes:
1399  *    When we entered the error handler, we blocked all further i/o to
1400  *    this device.  we need to 'reverse' this process.
1401  */
1402 static void scsi_restart_operations(struct Scsi_Host *shost)
1403 {
1404         struct scsi_device *sdev;
1405         unsigned long flags;
1406
1407         /*
1408          * If the door was locked, we need to insert a door lock request
1409          * onto the head of the SCSI request queue for the device.  There
1410          * is no point trying to lock the door of an off-line device.
1411          */
1412         shost_for_each_device(sdev, shost) {
1413                 if (scsi_device_online(sdev) && sdev->locked)
1414                         scsi_eh_lock_door(sdev);
1415         }
1416
1417         /*
1418          * next free up anything directly waiting upon the host.  this
1419          * will be requests for character device operations, and also for
1420          * ioctls to queued block devices.
1421          */
1422         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1423                                           __FUNCTION__));
1424
1425         spin_lock_irqsave(shost->host_lock, flags);
1426         if (scsi_host_set_state(shost, SHOST_RUNNING))
1427                 if (scsi_host_set_state(shost, SHOST_CANCEL))
1428                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1429         spin_unlock_irqrestore(shost->host_lock, flags);
1430
1431         wake_up(&shost->host_wait);
1432
1433         /*
1434          * finally we need to re-initiate requests that may be pending.  we will
1435          * have had everything blocked while error handling is taking place, and
1436          * now that error recovery is done, we will need to ensure that these
1437          * requests are started.
1438          */
1439         scsi_run_host_queues(shost);
1440 }
1441
1442 /**
1443  * scsi_eh_ready_devs - check device ready state and recover if not.
1444  * @shost:      host to be recovered.
1445  * @work_q:     &list_head for pending commands.
1446  * @done_q:     &list_head for processed commands.
1447  */
1448 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1449                         struct list_head *work_q,
1450                         struct list_head *done_q)
1451 {
1452         if (!scsi_eh_stu(shost, work_q, done_q))
1453                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1454                         if (!scsi_eh_bus_reset(shost, work_q, done_q))
1455                                 if (!scsi_eh_host_reset(work_q, done_q))
1456                                         scsi_eh_offline_sdevs(work_q, done_q);
1457 }
1458 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1459
1460 /**
1461  * scsi_eh_flush_done_q - finish processed commands or retry them.
1462  * @done_q:     list_head of processed commands.
1463  */
1464 void scsi_eh_flush_done_q(struct list_head *done_q)
1465 {
1466         struct scsi_cmnd *scmd, *next;
1467
1468         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1469                 list_del_init(&scmd->eh_entry);
1470                 if (scsi_device_online(scmd->device) &&
1471                     !blk_noretry_request(scmd->request) &&
1472                     (++scmd->retries <= scmd->allowed)) {
1473                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1474                                                           " retry cmd: %p\n",
1475                                                           current->comm,
1476                                                           scmd));
1477                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1478                 } else {
1479                         /*
1480                          * If just we got sense for the device (called
1481                          * scsi_eh_get_sense), scmd->result is already
1482                          * set, do not set DRIVER_TIMEOUT.
1483                          */
1484                         if (!scmd->result)
1485                                 scmd->result |= (DRIVER_TIMEOUT << 24);
1486                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1487                                                         " cmd: %p\n",
1488                                                         current->comm, scmd));
1489                         scsi_finish_command(scmd);
1490                 }
1491         }
1492 }
1493 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1494
1495 /**
1496  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1497  * @shost:      Host to unjam.
1498  *
1499  * Notes:
1500  *    When we come in here, we *know* that all commands on the bus have
1501  *    either completed, failed or timed out.  we also know that no further
1502  *    commands are being sent to the host, so things are relatively quiet
1503  *    and we have freedom to fiddle with things as we wish.
1504  *
1505  *    This is only the *default* implementation.  it is possible for
1506  *    individual drivers to supply their own version of this function, and
1507  *    if the maintainer wishes to do this, it is strongly suggested that
1508  *    this function be taken as a template and modified.  this function
1509  *    was designed to correctly handle problems for about 95% of the
1510  *    different cases out there, and it should always provide at least a
1511  *    reasonable amount of error recovery.
1512  *
1513  *    Any command marked 'failed' or 'timeout' must eventually have
1514  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1515  *    here, so when we restart the host after we return it should have an
1516  *    empty queue.
1517  */
1518 static void scsi_unjam_host(struct Scsi_Host *shost)
1519 {
1520         unsigned long flags;
1521         LIST_HEAD(eh_work_q);
1522         LIST_HEAD(eh_done_q);
1523
1524         spin_lock_irqsave(shost->host_lock, flags);
1525         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1526         spin_unlock_irqrestore(shost->host_lock, flags);
1527
1528         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1529
1530         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1531                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1532                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1533
1534         scsi_eh_flush_done_q(&eh_done_q);
1535 }
1536
1537 /**
1538  * scsi_error_handler - SCSI error handler thread
1539  * @data:       Host for which we are running.
1540  *
1541  * Notes:
1542  *    This is the main error handling loop.  This is run as a kernel thread
1543  *    for every SCSI host and handles all error handling activity.
1544  */
1545 int scsi_error_handler(void *data)
1546 {
1547         struct Scsi_Host *shost = data;
1548
1549         /*
1550          * We use TASK_INTERRUPTIBLE so that the thread is not
1551          * counted against the load average as a running process.
1552          * We never actually get interrupted because kthread_run
1553          * disables singal delivery for the created thread.
1554          */
1555         set_current_state(TASK_INTERRUPTIBLE);
1556         while (!kthread_should_stop()) {
1557                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1558                     shost->host_failed != shost->host_busy) {
1559                         SCSI_LOG_ERROR_RECOVERY(1,
1560                                 printk("Error handler scsi_eh_%d sleeping\n",
1561                                         shost->host_no));
1562                         schedule();
1563                         set_current_state(TASK_INTERRUPTIBLE);
1564                         continue;
1565                 }
1566
1567                 __set_current_state(TASK_RUNNING);
1568                 SCSI_LOG_ERROR_RECOVERY(1,
1569                         printk("Error handler scsi_eh_%d waking up\n",
1570                                 shost->host_no));
1571
1572                 /*
1573                  * We have a host that is failing for some reason.  Figure out
1574                  * what we need to do to get it up and online again (if we can).
1575                  * If we fail, we end up taking the thing offline.
1576                  */
1577                 if (shost->transportt->eh_strategy_handler)
1578                         shost->transportt->eh_strategy_handler(shost);
1579                 else
1580                         scsi_unjam_host(shost);
1581
1582                 /*
1583                  * Note - if the above fails completely, the action is to take
1584                  * individual devices offline and flush the queue of any
1585                  * outstanding requests that may have been pending.  When we
1586                  * restart, we restart any I/O to any other devices on the bus
1587                  * which are still online.
1588                  */
1589                 scsi_restart_operations(shost);
1590                 set_current_state(TASK_INTERRUPTIBLE);
1591         }
1592         __set_current_state(TASK_RUNNING);
1593
1594         SCSI_LOG_ERROR_RECOVERY(1,
1595                 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1596         shost->ehandler = NULL;
1597         return 0;
1598 }
1599
1600 /*
1601  * Function:    scsi_report_bus_reset()
1602  *
1603  * Purpose:     Utility function used by low-level drivers to report that
1604  *              they have observed a bus reset on the bus being handled.
1605  *
1606  * Arguments:   shost       - Host in question
1607  *              channel     - channel on which reset was observed.
1608  *
1609  * Returns:     Nothing
1610  *
1611  * Lock status: Host lock must be held.
1612  *
1613  * Notes:       This only needs to be called if the reset is one which
1614  *              originates from an unknown location.  Resets originated
1615  *              by the mid-level itself don't need to call this, but there
1616  *              should be no harm.
1617  *
1618  *              The main purpose of this is to make sure that a CHECK_CONDITION
1619  *              is properly treated.
1620  */
1621 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1622 {
1623         struct scsi_device *sdev;
1624
1625         __shost_for_each_device(sdev, shost) {
1626                 if (channel == sdev_channel(sdev)) {
1627                         sdev->was_reset = 1;
1628                         sdev->expecting_cc_ua = 1;
1629                 }
1630         }
1631 }
1632 EXPORT_SYMBOL(scsi_report_bus_reset);
1633
1634 /*
1635  * Function:    scsi_report_device_reset()
1636  *
1637  * Purpose:     Utility function used by low-level drivers to report that
1638  *              they have observed a device reset on the device being handled.
1639  *
1640  * Arguments:   shost       - Host in question
1641  *              channel     - channel on which reset was observed
1642  *              target      - target on which reset was observed
1643  *
1644  * Returns:     Nothing
1645  *
1646  * Lock status: Host lock must be held
1647  *
1648  * Notes:       This only needs to be called if the reset is one which
1649  *              originates from an unknown location.  Resets originated
1650  *              by the mid-level itself don't need to call this, but there
1651  *              should be no harm.
1652  *
1653  *              The main purpose of this is to make sure that a CHECK_CONDITION
1654  *              is properly treated.
1655  */
1656 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1657 {
1658         struct scsi_device *sdev;
1659
1660         __shost_for_each_device(sdev, shost) {
1661                 if (channel == sdev_channel(sdev) &&
1662                     target == sdev_id(sdev)) {
1663                         sdev->was_reset = 1;
1664                         sdev->expecting_cc_ua = 1;
1665                 }
1666         }
1667 }
1668 EXPORT_SYMBOL(scsi_report_device_reset);
1669
1670 static void
1671 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1672 {
1673 }
1674
1675 /*
1676  * Function:    scsi_reset_provider
1677  *
1678  * Purpose:     Send requested reset to a bus or device at any phase.
1679  *
1680  * Arguments:   device  - device to send reset to
1681  *              flag - reset type (see scsi.h)
1682  *
1683  * Returns:     SUCCESS/FAILURE.
1684  *
1685  * Notes:       This is used by the SCSI Generic driver to provide
1686  *              Bus/Device reset capability.
1687  */
1688 int
1689 scsi_reset_provider(struct scsi_device *dev, int flag)
1690 {
1691         struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1692         struct Scsi_Host *shost = dev->host;
1693         struct request req;
1694         unsigned long flags;
1695         int rtn;
1696
1697         scmd->request = &req;
1698         memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1699
1700         memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1701     
1702         scmd->scsi_done         = scsi_reset_provider_done_command;
1703         scmd->request_buffer            = NULL;
1704         scmd->request_bufflen           = 0;
1705
1706         scmd->cmd_len                   = 0;
1707
1708         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
1709
1710         init_timer(&scmd->eh_timeout);
1711
1712         spin_lock_irqsave(shost->host_lock, flags);
1713         shost->tmf_in_progress = 1;
1714         spin_unlock_irqrestore(shost->host_lock, flags);
1715
1716         switch (flag) {
1717         case SCSI_TRY_RESET_DEVICE:
1718                 rtn = scsi_try_bus_device_reset(scmd);
1719                 if (rtn == SUCCESS)
1720                         break;
1721                 /* FALLTHROUGH */
1722         case SCSI_TRY_RESET_BUS:
1723                 rtn = scsi_try_bus_reset(scmd);
1724                 if (rtn == SUCCESS)
1725                         break;
1726                 /* FALLTHROUGH */
1727         case SCSI_TRY_RESET_HOST:
1728                 rtn = scsi_try_host_reset(scmd);
1729                 break;
1730         default:
1731                 rtn = FAILED;
1732         }
1733
1734         spin_lock_irqsave(shost->host_lock, flags);
1735         shost->tmf_in_progress = 0;
1736         spin_unlock_irqrestore(shost->host_lock, flags);
1737
1738         /*
1739          * be sure to wake up anyone who was sleeping or had their queue
1740          * suspended while we performed the TMF.
1741          */
1742         SCSI_LOG_ERROR_RECOVERY(3,
1743                 printk("%s: waking up host to restart after TMF\n",
1744                 __FUNCTION__));
1745
1746         wake_up(&shost->host_wait);
1747
1748         scsi_run_host_queues(shost);
1749
1750         scsi_next_command(scmd);
1751         return rtn;
1752 }
1753 EXPORT_SYMBOL(scsi_reset_provider);
1754
1755 /**
1756  * scsi_normalize_sense - normalize main elements from either fixed or
1757  *                      descriptor sense data format into a common format.
1758  *
1759  * @sense_buffer:       byte array containing sense data returned by device
1760  * @sb_len:             number of valid bytes in sense_buffer
1761  * @sshdr:              pointer to instance of structure that common
1762  *                      elements are written to.
1763  *
1764  * Notes:
1765  *      The "main elements" from sense data are: response_code, sense_key,
1766  *      asc, ascq and additional_length (only for descriptor format).
1767  *
1768  *      Typically this function can be called after a device has
1769  *      responded to a SCSI command with the CHECK_CONDITION status.
1770  *
1771  * Return value:
1772  *      1 if valid sense data information found, else 0;
1773  */
1774 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1775                          struct scsi_sense_hdr *sshdr)
1776 {
1777         if (!sense_buffer || !sb_len)
1778                 return 0;
1779
1780         memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1781
1782         sshdr->response_code = (sense_buffer[0] & 0x7f);
1783
1784         if (!scsi_sense_valid(sshdr))
1785                 return 0;
1786
1787         if (sshdr->response_code >= 0x72) {
1788                 /*
1789                  * descriptor format
1790                  */
1791                 if (sb_len > 1)
1792                         sshdr->sense_key = (sense_buffer[1] & 0xf);
1793                 if (sb_len > 2)
1794                         sshdr->asc = sense_buffer[2];
1795                 if (sb_len > 3)
1796                         sshdr->ascq = sense_buffer[3];
1797                 if (sb_len > 7)
1798                         sshdr->additional_length = sense_buffer[7];
1799         } else {
1800                 /* 
1801                  * fixed format
1802                  */
1803                 if (sb_len > 2)
1804                         sshdr->sense_key = (sense_buffer[2] & 0xf);
1805                 if (sb_len > 7) {
1806                         sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1807                                          sb_len : (sense_buffer[7] + 8);
1808                         if (sb_len > 12)
1809                                 sshdr->asc = sense_buffer[12];
1810                         if (sb_len > 13)
1811                                 sshdr->ascq = sense_buffer[13];
1812                 }
1813         }
1814
1815         return 1;
1816 }
1817 EXPORT_SYMBOL(scsi_normalize_sense);
1818
1819 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1820                                  struct scsi_sense_hdr *sshdr)
1821 {
1822         return scsi_normalize_sense(cmd->sense_buffer,
1823                         SCSI_SENSE_BUFFERSIZE, sshdr);
1824 }
1825 EXPORT_SYMBOL(scsi_command_normalize_sense);
1826
1827 /**
1828  * scsi_sense_desc_find - search for a given descriptor type in descriptor sense data format.
1829  * @sense_buffer:       byte array of descriptor format sense data
1830  * @sb_len:             number of valid bytes in sense_buffer
1831  * @desc_type:          value of descriptor type to find
1832  *                      (e.g. 0 -> information)
1833  *
1834  * Notes:
1835  *      only valid when sense data is in descriptor format
1836  *
1837  * Return value:
1838  *      pointer to start of (first) descriptor if found else NULL
1839  */
1840 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1841                                 int desc_type)
1842 {
1843         int add_sen_len, add_len, desc_len, k;
1844         const u8 * descp;
1845
1846         if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1847                 return NULL;
1848         if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1849                 return NULL;
1850         add_sen_len = (add_sen_len < (sb_len - 8)) ?
1851                         add_sen_len : (sb_len - 8);
1852         descp = &sense_buffer[8];
1853         for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1854                 descp += desc_len;
1855                 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1856                 desc_len = add_len + 2;
1857                 if (descp[0] == desc_type)
1858                         return descp;
1859                 if (add_len < 0) // short descriptor ??
1860                         break;
1861         }
1862         return NULL;
1863 }
1864 EXPORT_SYMBOL(scsi_sense_desc_find);
1865
1866 /**
1867  * scsi_get_sense_info_fld - get information field from sense data (either fixed or descriptor format)
1868  * @sense_buffer:       byte array of sense data
1869  * @sb_len:             number of valid bytes in sense_buffer
1870  * @info_out:           pointer to 64 integer where 8 or 4 byte information
1871  *                      field will be placed if found.
1872  *
1873  * Return value:
1874  *      1 if information field found, 0 if not found.
1875  */
1876 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1877                             u64 * info_out)
1878 {
1879         int j;
1880         const u8 * ucp;
1881         u64 ull;
1882
1883         if (sb_len < 7)
1884                 return 0;
1885         switch (sense_buffer[0] & 0x7f) {
1886         case 0x70:
1887         case 0x71:
1888                 if (sense_buffer[0] & 0x80) {
1889                         *info_out = (sense_buffer[3] << 24) +
1890                                     (sense_buffer[4] << 16) +
1891                                     (sense_buffer[5] << 8) + sense_buffer[6];
1892                         return 1;
1893                 } else
1894                         return 0;
1895         case 0x72:
1896         case 0x73:
1897                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1898                                            0 /* info desc */);
1899                 if (ucp && (0xa == ucp[1])) {
1900                         ull = 0;
1901                         for (j = 0; j < 8; ++j) {
1902                                 if (j > 0)
1903                                         ull <<= 8;
1904                                 ull |= ucp[4 + j];
1905                         }
1906                         *info_out = ull;
1907                         return 1;
1908                 } else
1909                         return 0;
1910         default:
1911                 return 0;
1912         }
1913 }
1914 EXPORT_SYMBOL(scsi_get_sense_info_fld);