Merge git://git.kernel.org/pub/scm/linux/kernel/git/sam/kbuild-fix
[linux-2.6] / 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/slab.h>
22 #include <linux/kernel.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
40 #define SENSE_TIMEOUT           (10*HZ)
41
42 /*
43  * These should *probably* be handled by the host itself.
44  * Since it is allowed to sleep, it probably should.
45  */
46 #define BUS_RESET_SETTLE_TIME   (10)
47 #define HOST_RESET_SETTLE_TIME  (10)
48
49 /* called with shost->host_lock held */
50 void scsi_eh_wakeup(struct Scsi_Host *shost)
51 {
52         if (shost->host_busy == shost->host_failed) {
53                 wake_up_process(shost->ehandler);
54                 SCSI_LOG_ERROR_RECOVERY(5,
55                                 printk("Waking error handler thread\n"));
56         }
57 }
58
59 /**
60  * scsi_schedule_eh - schedule EH for SCSI host
61  * @shost:      SCSI host to invoke error handling on.
62  *
63  * Schedule SCSI EH without scmd.
64  **/
65 void scsi_schedule_eh(struct Scsi_Host *shost)
66 {
67         unsigned long flags;
68
69         spin_lock_irqsave(shost->host_lock, flags);
70
71         if (scsi_host_set_state(shost, SHOST_RECOVERY) == 0 ||
72             scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY) == 0) {
73                 shost->host_eh_scheduled++;
74                 scsi_eh_wakeup(shost);
75         }
76
77         spin_unlock_irqrestore(shost->host_lock, flags);
78 }
79 EXPORT_SYMBOL_GPL(scsi_schedule_eh);
80
81 /**
82  * scsi_eh_scmd_add - add scsi cmd to error handling.
83  * @scmd:       scmd to run eh on.
84  * @eh_flag:    optional SCSI_EH flag.
85  *
86  * Return value:
87  *      0 on failure.
88  **/
89 int scsi_eh_scmd_add(struct scsi_cmnd *scmd, int eh_flag)
90 {
91         struct Scsi_Host *shost = scmd->device->host;
92         unsigned long flags;
93         int ret = 0;
94
95         if (!shost->ehandler)
96                 return 0;
97
98         spin_lock_irqsave(shost->host_lock, flags);
99         if (scsi_host_set_state(shost, SHOST_RECOVERY))
100                 if (scsi_host_set_state(shost, SHOST_CANCEL_RECOVERY))
101                         goto out_unlock;
102
103         ret = 1;
104         scmd->eh_eflags |= eh_flag;
105         list_add_tail(&scmd->eh_entry, &shost->eh_cmd_q);
106         shost->host_failed++;
107         scsi_eh_wakeup(shost);
108  out_unlock:
109         spin_unlock_irqrestore(shost->host_lock, flags);
110         return ret;
111 }
112
113 /**
114  * scsi_add_timer - Start timeout timer for a single scsi command.
115  * @scmd:       scsi command that is about to start running.
116  * @timeout:    amount of time to allow this command to run.
117  * @complete:   timeout function to call if timer isn't canceled.
118  *
119  * Notes:
120  *    This should be turned into an inline function.  Each scsi command
121  *    has its own timer, and as it is added to the queue, we set up the
122  *    timer.  When the command completes, we cancel the timer.
123  **/
124 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
125                     void (*complete)(struct scsi_cmnd *))
126 {
127
128         /*
129          * If the clock was already running for this command, then
130          * first delete the timer.  The timer handling code gets rather
131          * confused if we don't do this.
132          */
133         if (scmd->eh_timeout.function)
134                 del_timer(&scmd->eh_timeout);
135
136         scmd->eh_timeout.data = (unsigned long)scmd;
137         scmd->eh_timeout.expires = jiffies + timeout;
138         scmd->eh_timeout.function = (void (*)(unsigned long)) complete;
139
140         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p, time:"
141                                           " %d, (%p)\n", __FUNCTION__,
142                                           scmd, timeout, complete));
143
144         add_timer(&scmd->eh_timeout);
145 }
146
147 /**
148  * scsi_delete_timer - Delete/cancel timer for a given function.
149  * @scmd:       Cmd that we are canceling timer for
150  *
151  * Notes:
152  *     This should be turned into an inline function.
153  *
154  * Return value:
155  *     1 if we were able to detach the timer.  0 if we blew it, and the
156  *     timer function has already started to run.
157  **/
158 int scsi_delete_timer(struct scsi_cmnd *scmd)
159 {
160         int rtn;
161
162         rtn = del_timer(&scmd->eh_timeout);
163
164         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: scmd: %p,"
165                                          " rtn: %d\n", __FUNCTION__,
166                                          scmd, rtn));
167
168         scmd->eh_timeout.data = (unsigned long)NULL;
169         scmd->eh_timeout.function = NULL;
170
171         return rtn;
172 }
173
174 /**
175  * scsi_times_out - Timeout function for normal scsi commands.
176  * @scmd:       Cmd that is timing out.
177  *
178  * Notes:
179  *     We do not need to lock this.  There is the potential for a race
180  *     only in that the normal completion handling might run, but if the
181  *     normal completion function determines that the timer has already
182  *     fired, then it mustn't do anything.
183  **/
184 void scsi_times_out(struct scsi_cmnd *scmd)
185 {
186         enum scsi_eh_timer_return (* eh_timed_out)(struct scsi_cmnd *);
187
188         scsi_log_completion(scmd, TIMEOUT_ERROR);
189
190         if (scmd->device->host->transportt->eh_timed_out)
191                 eh_timed_out = scmd->device->host->transportt->eh_timed_out;
192         else if (scmd->device->host->hostt->eh_timed_out)
193                 eh_timed_out = scmd->device->host->hostt->eh_timed_out;
194         else
195                 eh_timed_out = NULL;
196
197         if (eh_timed_out)
198                 switch (eh_timed_out(scmd)) {
199                 case EH_HANDLED:
200                         __scsi_done(scmd);
201                         return;
202                 case EH_RESET_TIMER:
203                         scsi_add_timer(scmd, scmd->timeout_per_command,
204                                        scsi_times_out);
205                         return;
206                 case EH_NOT_HANDLED:
207                         break;
208                 }
209
210         if (unlikely(!scsi_eh_scmd_add(scmd, SCSI_EH_CANCEL_CMD))) {
211                 scmd->result |= DID_TIME_OUT << 16;
212                 __scsi_done(scmd);
213         }
214 }
215
216 /**
217  * scsi_block_when_processing_errors - Prevent cmds from being queued.
218  * @sdev:       Device on which we are performing recovery.
219  *
220  * Description:
221  *     We block until the host is out of error recovery, and then check to
222  *     see whether the host or the device is offline.
223  *
224  * Return value:
225  *     0 when dev was taken offline by error recovery. 1 OK to proceed.
226  **/
227 int scsi_block_when_processing_errors(struct scsi_device *sdev)
228 {
229         int online;
230
231         wait_event(sdev->host->host_wait, !scsi_host_in_recovery(sdev->host));
232
233         online = scsi_device_online(sdev);
234
235         SCSI_LOG_ERROR_RECOVERY(5, printk("%s: rtn: %d\n", __FUNCTION__,
236                                           online));
237
238         return online;
239 }
240 EXPORT_SYMBOL(scsi_block_when_processing_errors);
241
242 #ifdef CONFIG_SCSI_LOGGING
243 /**
244  * scsi_eh_prt_fail_stats - Log info on failures.
245  * @shost:      scsi host being recovered.
246  * @work_q:     Queue of scsi cmds to process.
247  **/
248 static inline void scsi_eh_prt_fail_stats(struct Scsi_Host *shost,
249                                           struct list_head *work_q)
250 {
251         struct scsi_cmnd *scmd;
252         struct scsi_device *sdev;
253         int total_failures = 0;
254         int cmd_failed = 0;
255         int cmd_cancel = 0;
256         int devices_failed = 0;
257
258         shost_for_each_device(sdev, shost) {
259                 list_for_each_entry(scmd, work_q, eh_entry) {
260                         if (scmd->device == sdev) {
261                                 ++total_failures;
262                                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD)
263                                         ++cmd_cancel;
264                                 else 
265                                         ++cmd_failed;
266                         }
267                 }
268
269                 if (cmd_cancel || cmd_failed) {
270                         SCSI_LOG_ERROR_RECOVERY(3,
271                                 sdev_printk(KERN_INFO, sdev,
272                                             "%s: cmds failed: %d, cancel: %d\n",
273                                             __FUNCTION__, cmd_failed,
274                                             cmd_cancel));
275                         cmd_cancel = 0;
276                         cmd_failed = 0;
277                         ++devices_failed;
278                 }
279         }
280
281         SCSI_LOG_ERROR_RECOVERY(2, printk("Total of %d commands on %d"
282                                           " devices require eh work\n",
283                                   total_failures, devices_failed));
284 }
285 #endif
286
287 /**
288  * scsi_check_sense - Examine scsi cmd sense
289  * @scmd:       Cmd to have sense checked.
290  *
291  * Return value:
292  *      SUCCESS or FAILED or NEEDS_RETRY
293  *
294  * Notes:
295  *      When a deferred error is detected the current command has
296  *      not been executed and needs retrying.
297  **/
298 static int scsi_check_sense(struct scsi_cmnd *scmd)
299 {
300         struct scsi_sense_hdr sshdr;
301
302         if (! scsi_command_normalize_sense(scmd, &sshdr))
303                 return FAILED;  /* no valid sense data */
304
305         if (scsi_sense_is_deferred(&sshdr))
306                 return NEEDS_RETRY;
307
308         /*
309          * Previous logic looked for FILEMARK, EOM or ILI which are
310          * mainly associated with tapes and returned SUCCESS.
311          */
312         if (sshdr.response_code == 0x70) {
313                 /* fixed format */
314                 if (scmd->sense_buffer[2] & 0xe0)
315                         return SUCCESS;
316         } else {
317                 /*
318                  * descriptor format: look for "stream commands sense data
319                  * descriptor" (see SSC-3). Assume single sense data
320                  * descriptor. Ignore ILI from SBC-2 READ LONG and WRITE LONG.
321                  */
322                 if ((sshdr.additional_length > 3) &&
323                     (scmd->sense_buffer[8] == 0x4) &&
324                     (scmd->sense_buffer[11] & 0xe0))
325                         return SUCCESS;
326         }
327
328         switch (sshdr.sense_key) {
329         case NO_SENSE:
330                 return SUCCESS;
331         case RECOVERED_ERROR:
332                 return /* soft_error */ SUCCESS;
333
334         case ABORTED_COMMAND:
335                 return NEEDS_RETRY;
336         case NOT_READY:
337         case UNIT_ATTENTION:
338                 /*
339                  * if we are expecting a cc/ua because of a bus reset that we
340                  * performed, treat this just as a retry.  otherwise this is
341                  * information that we should pass up to the upper-level driver
342                  * so that we can deal with it there.
343                  */
344                 if (scmd->device->expecting_cc_ua) {
345                         scmd->device->expecting_cc_ua = 0;
346                         return NEEDS_RETRY;
347                 }
348                 /*
349                  * if the device is in the process of becoming ready, we 
350                  * should retry.
351                  */
352                 if ((sshdr.asc == 0x04) && (sshdr.ascq == 0x01))
353                         return NEEDS_RETRY;
354                 /*
355                  * if the device is not started, we need to wake
356                  * the error handler to start the motor
357                  */
358                 if (scmd->device->allow_restart &&
359                     (sshdr.asc == 0x04) && (sshdr.ascq == 0x02))
360                         return FAILED;
361                 return SUCCESS;
362
363                 /* these three are not supported */
364         case COPY_ABORTED:
365         case VOLUME_OVERFLOW:
366         case MISCOMPARE:
367                 return SUCCESS;
368
369         case MEDIUM_ERROR:
370                 if (sshdr.asc == 0x11 || /* UNRECOVERED READ ERR */
371                     sshdr.asc == 0x13 || /* AMNF DATA FIELD */
372                     sshdr.asc == 0x14) { /* RECORD NOT FOUND */
373                         return SUCCESS;
374                 }
375                 return NEEDS_RETRY;
376
377         case HARDWARE_ERROR:
378                 if (scmd->device->retry_hwerror)
379                         return NEEDS_RETRY;
380                 else
381                         return SUCCESS;
382
383         case ILLEGAL_REQUEST:
384         case BLANK_CHECK:
385         case DATA_PROTECT:
386         default:
387                 return SUCCESS;
388         }
389 }
390
391 /**
392  * scsi_eh_completed_normally - Disposition a eh cmd on return from LLD.
393  * @scmd:       SCSI cmd to examine.
394  *
395  * Notes:
396  *    This is *only* called when we are examining the status of commands
397  *    queued during error recovery.  the main difference here is that we
398  *    don't allow for the possibility of retries here, and we are a lot
399  *    more restrictive about what we consider acceptable.
400  **/
401 static int scsi_eh_completed_normally(struct scsi_cmnd *scmd)
402 {
403         /*
404          * first check the host byte, to see if there is anything in there
405          * that would indicate what we need to do.
406          */
407         if (host_byte(scmd->result) == DID_RESET) {
408                 /*
409                  * rats.  we are already in the error handler, so we now
410                  * get to try and figure out what to do next.  if the sense
411                  * is valid, we have a pretty good idea of what to do.
412                  * if not, we mark it as FAILED.
413                  */
414                 return scsi_check_sense(scmd);
415         }
416         if (host_byte(scmd->result) != DID_OK)
417                 return FAILED;
418
419         /*
420          * next, check the message byte.
421          */
422         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
423                 return FAILED;
424
425         /*
426          * now, check the status byte to see if this indicates
427          * anything special.
428          */
429         switch (status_byte(scmd->result)) {
430         case GOOD:
431         case COMMAND_TERMINATED:
432                 return SUCCESS;
433         case CHECK_CONDITION:
434                 return scsi_check_sense(scmd);
435         case CONDITION_GOOD:
436         case INTERMEDIATE_GOOD:
437         case INTERMEDIATE_C_GOOD:
438                 /*
439                  * who knows?  FIXME(eric)
440                  */
441                 return SUCCESS;
442         case BUSY:
443         case QUEUE_FULL:
444         case RESERVATION_CONFLICT:
445         default:
446                 return FAILED;
447         }
448         return FAILED;
449 }
450
451 /**
452  * scsi_eh_done - Completion function for error handling.
453  * @scmd:       Cmd that is done.
454  **/
455 static void scsi_eh_done(struct scsi_cmnd *scmd)
456 {
457         struct completion     *eh_action;
458
459         SCSI_LOG_ERROR_RECOVERY(3,
460                 printk("%s scmd: %p result: %x\n",
461                         __FUNCTION__, scmd, scmd->result));
462
463         eh_action = scmd->device->host->eh_action;
464         if (eh_action)
465                 complete(eh_action);
466 }
467
468 /**
469  * scsi_try_host_reset - ask host adapter to reset itself
470  * @scmd:       SCSI cmd to send hsot reset.
471  **/
472 static int scsi_try_host_reset(struct scsi_cmnd *scmd)
473 {
474         unsigned long flags;
475         int rtn;
476
477         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Host RST\n",
478                                           __FUNCTION__));
479
480         if (!scmd->device->host->hostt->eh_host_reset_handler)
481                 return FAILED;
482
483         rtn = scmd->device->host->hostt->eh_host_reset_handler(scmd);
484
485         if (rtn == SUCCESS) {
486                 if (!scmd->device->host->hostt->skip_settle_delay)
487                         ssleep(HOST_RESET_SETTLE_TIME);
488                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
489                 scsi_report_bus_reset(scmd->device->host,
490                                       scmd_channel(scmd));
491                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
492         }
493
494         return rtn;
495 }
496
497 /**
498  * scsi_try_bus_reset - ask host to perform a bus reset
499  * @scmd:       SCSI cmd to send bus reset.
500  **/
501 static int scsi_try_bus_reset(struct scsi_cmnd *scmd)
502 {
503         unsigned long flags;
504         int rtn;
505
506         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Snd Bus RST\n",
507                                           __FUNCTION__));
508
509         if (!scmd->device->host->hostt->eh_bus_reset_handler)
510                 return FAILED;
511
512         rtn = scmd->device->host->hostt->eh_bus_reset_handler(scmd);
513
514         if (rtn == SUCCESS) {
515                 if (!scmd->device->host->hostt->skip_settle_delay)
516                         ssleep(BUS_RESET_SETTLE_TIME);
517                 spin_lock_irqsave(scmd->device->host->host_lock, flags);
518                 scsi_report_bus_reset(scmd->device->host,
519                                       scmd_channel(scmd));
520                 spin_unlock_irqrestore(scmd->device->host->host_lock, flags);
521         }
522
523         return rtn;
524 }
525
526 /**
527  * scsi_try_bus_device_reset - Ask host to perform a BDR on a dev
528  * @scmd:       SCSI cmd used to send BDR
529  *
530  * Notes:
531  *    There is no timeout for this operation.  if this operation is
532  *    unreliable for a given host, then the host itself needs to put a
533  *    timer on it, and set the host back to a consistent state prior to
534  *    returning.
535  **/
536 static int scsi_try_bus_device_reset(struct scsi_cmnd *scmd)
537 {
538         int rtn;
539
540         if (!scmd->device->host->hostt->eh_device_reset_handler)
541                 return FAILED;
542
543         rtn = scmd->device->host->hostt->eh_device_reset_handler(scmd);
544         if (rtn == SUCCESS) {
545                 scmd->device->was_reset = 1;
546                 scmd->device->expecting_cc_ua = 1;
547         }
548
549         return rtn;
550 }
551
552 static int __scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
553 {
554         if (!scmd->device->host->hostt->eh_abort_handler)
555                 return FAILED;
556
557         return scmd->device->host->hostt->eh_abort_handler(scmd);
558 }
559
560 /**
561  * scsi_try_to_abort_cmd - Ask host to abort a running command.
562  * @scmd:       SCSI cmd to abort from Lower Level.
563  *
564  * Notes:
565  *    This function will not return until the user's completion function
566  *    has been called.  there is no timeout on this operation.  if the
567  *    author of the low-level driver wishes this operation to be timed,
568  *    they can provide this facility themselves.  helper functions in
569  *    scsi_error.c can be supplied to make this easier to do.
570  **/
571 static int scsi_try_to_abort_cmd(struct scsi_cmnd *scmd)
572 {
573         /*
574          * scsi_done was called just after the command timed out and before
575          * we had a chance to process it. (db)
576          */
577         if (scmd->serial_number == 0)
578                 return SUCCESS;
579         return __scsi_try_to_abort_cmd(scmd);
580 }
581
582 static void scsi_abort_eh_cmnd(struct scsi_cmnd *scmd)
583 {
584         if (__scsi_try_to_abort_cmd(scmd) != SUCCESS)
585                 if (scsi_try_bus_device_reset(scmd) != SUCCESS)
586                         if (scsi_try_bus_reset(scmd) != SUCCESS)
587                                 scsi_try_host_reset(scmd);
588 }
589
590 /**
591  * scsi_send_eh_cmnd  - submit a scsi command as part of error recory
592  * @scmd:       SCSI command structure to hijack
593  * @cmnd:       CDB to send
594  * @cmnd_size:  size in bytes of @cmnd
595  * @timeout:    timeout for this request
596  * @copy_sense: request sense data if set to 1
597  *
598  * This function is used to send a scsi command down to a target device
599  * as part of the error recovery process.  If @copy_sense is 0 the command
600  * sent must be one that does not transfer any data.  If @copy_sense is 1
601  * the command must be REQUEST_SENSE and this functions copies out the
602  * sense buffer it got into @scmd->sense_buffer.
603  *
604  * Return value:
605  *    SUCCESS or FAILED or NEEDS_RETRY
606  **/
607 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
608                              int cmnd_size, int timeout, int copy_sense)
609 {
610         struct scsi_device *sdev = scmd->device;
611         struct Scsi_Host *shost = sdev->host;
612         int old_result = scmd->result;
613         DECLARE_COMPLETION_ONSTACK(done);
614         unsigned long timeleft;
615         unsigned long flags;
616         struct scatterlist sgl;
617         unsigned char old_cmnd[MAX_COMMAND_SIZE];
618         enum dma_data_direction old_data_direction;
619         unsigned short old_use_sg;
620         unsigned char old_cmd_len;
621         unsigned old_bufflen;
622         void *old_buffer;
623         int rtn;
624
625         /*
626          * We need saved copies of a number of fields - this is because
627          * error handling may need to overwrite these with different values
628          * to run different commands, and once error handling is complete,
629          * we will need to restore these values prior to running the actual
630          * command.
631          */
632         old_buffer = scmd->request_buffer;
633         old_bufflen = scmd->request_bufflen;
634         memcpy(old_cmnd, scmd->cmnd, sizeof(scmd->cmnd));
635         old_data_direction = scmd->sc_data_direction;
636         old_cmd_len = scmd->cmd_len;
637         old_use_sg = scmd->use_sg;
638
639         memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
640         memcpy(scmd->cmnd, cmnd, cmnd_size);
641
642         if (copy_sense) {
643                 gfp_t gfp_mask = GFP_ATOMIC;
644
645                 if (shost->hostt->unchecked_isa_dma)
646                         gfp_mask |= __GFP_DMA;
647
648                 sgl.page = alloc_page(gfp_mask);
649                 if (!sgl.page)
650                         return FAILED;
651                 sgl.offset = 0;
652                 sgl.length = 252;
653
654                 scmd->sc_data_direction = DMA_FROM_DEVICE;
655                 scmd->request_bufflen = sgl.length;
656                 scmd->request_buffer = &sgl;
657                 scmd->use_sg = 1;
658         } else {
659                 scmd->request_buffer = NULL;
660                 scmd->request_bufflen = 0;
661                 scmd->sc_data_direction = DMA_NONE;
662                 scmd->use_sg = 0;
663         }
664
665         scmd->underflow = 0;
666         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
667
668         if (sdev->scsi_level <= SCSI_2)
669                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
670                         (sdev->lun << 5 & 0xe0);
671
672         /*
673          * Zero the sense buffer.  The scsi spec mandates that any
674          * untransferred sense data should be interpreted as being zero.
675          */
676         memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
677
678         shost->eh_action = &done;
679
680         spin_lock_irqsave(shost->host_lock, flags);
681         scsi_log_send(scmd);
682         shost->hostt->queuecommand(scmd, scsi_eh_done);
683         spin_unlock_irqrestore(shost->host_lock, flags);
684
685         timeleft = wait_for_completion_timeout(&done, timeout);
686
687         shost->eh_action = NULL;
688
689         scsi_log_completion(scmd, SUCCESS);
690
691         SCSI_LOG_ERROR_RECOVERY(3,
692                 printk("%s: scmd: %p, timeleft: %ld\n",
693                         __FUNCTION__, scmd, timeleft));
694
695         /*
696          * If there is time left scsi_eh_done got called, and we will
697          * examine the actual status codes to see whether the command
698          * actually did complete normally, else tell the host to forget
699          * about this command.
700          */
701         if (timeleft) {
702                 rtn = scsi_eh_completed_normally(scmd);
703                 SCSI_LOG_ERROR_RECOVERY(3,
704                         printk("%s: scsi_eh_completed_normally %x\n",
705                                __FUNCTION__, rtn));
706
707                 switch (rtn) {
708                 case SUCCESS:
709                 case NEEDS_RETRY:
710                 case FAILED:
711                         break;
712                 default:
713                         rtn = FAILED;
714                         break;
715                 }
716         } else {
717                 scsi_abort_eh_cmnd(scmd);
718                 rtn = FAILED;
719         }
720
721
722         /*
723          * Last chance to have valid sense data.
724          */
725         if (copy_sense) {
726                 if (!SCSI_SENSE_VALID(scmd)) {
727                         memcpy(scmd->sense_buffer, page_address(sgl.page),
728                                sizeof(scmd->sense_buffer));
729                 }
730                 __free_page(sgl.page);
731         }
732
733
734         /*
735          * Restore original data
736          */
737         scmd->request_buffer = old_buffer;
738         scmd->request_bufflen = old_bufflen;
739         memcpy(scmd->cmnd, old_cmnd, sizeof(scmd->cmnd));
740         scmd->sc_data_direction = old_data_direction;
741         scmd->cmd_len = old_cmd_len;
742         scmd->use_sg = old_use_sg;
743         scmd->result = old_result;
744         return rtn;
745 }
746
747 /**
748  * scsi_request_sense - Request sense data from a particular target.
749  * @scmd:       SCSI cmd for request sense.
750  *
751  * Notes:
752  *    Some hosts automatically obtain this information, others require
753  *    that we obtain it on our own. This function will *not* return until
754  *    the command either times out, or it completes.
755  **/
756 static int scsi_request_sense(struct scsi_cmnd *scmd)
757 {
758         static unsigned char generic_sense[6] =
759                 {REQUEST_SENSE, 0, 0, 0, 252, 0};
760
761         return scsi_send_eh_cmnd(scmd, generic_sense, 6, SENSE_TIMEOUT, 1);
762 }
763
764 /**
765  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
766  * @scmd:       Original SCSI cmd that eh has finished.
767  * @done_q:     Queue for processed commands.
768  *
769  * Notes:
770  *    We don't want to use the normal command completion while we are are
771  *    still handling errors - it may cause other commands to be queued,
772  *    and that would disturb what we are doing.  thus we really want to
773  *    keep a list of pending commands for final completion, and once we
774  *    are ready to leave error handling we handle completion for real.
775  **/
776 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
777 {
778         scmd->device->host->host_failed--;
779         scmd->eh_eflags = 0;
780         list_move_tail(&scmd->eh_entry, done_q);
781 }
782 EXPORT_SYMBOL(scsi_eh_finish_cmd);
783
784 /**
785  * scsi_eh_get_sense - Get device sense data.
786  * @work_q:     Queue of commands to process.
787  * @done_q:     Queue of proccessed commands..
788  *
789  * Description:
790  *    See if we need to request sense information.  if so, then get it
791  *    now, so we have a better idea of what to do.  
792  *
793  * Notes:
794  *    This has the unfortunate side effect that if a shost adapter does
795  *    not automatically request sense information, that we end up shutting
796  *    it down before we request it.
797  *
798  *    All drivers should request sense information internally these days,
799  *    so for now all I have to say is tough noogies if you end up in here.
800  *
801  *    XXX: Long term this code should go away, but that needs an audit of
802  *         all LLDDs first.
803  **/
804 int scsi_eh_get_sense(struct list_head *work_q,
805                       struct list_head *done_q)
806 {
807         struct scsi_cmnd *scmd, *next;
808         int rtn;
809
810         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
811                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
812                     SCSI_SENSE_VALID(scmd))
813                         continue;
814
815                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
816                                                   "%s: requesting sense\n",
817                                                   current->comm));
818                 rtn = scsi_request_sense(scmd);
819                 if (rtn != SUCCESS)
820                         continue;
821
822                 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
823                                                   " result %x\n", scmd,
824                                                   scmd->result));
825                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
826
827                 rtn = scsi_decide_disposition(scmd);
828
829                 /*
830                  * if the result was normal, then just pass it along to the
831                  * upper level.
832                  */
833                 if (rtn == SUCCESS)
834                         /* we don't want this command reissued, just
835                          * finished with the sense data, so set
836                          * retries to the max allowed to ensure it
837                          * won't get reissued */
838                         scmd->retries = scmd->allowed;
839                 else if (rtn != NEEDS_RETRY)
840                         continue;
841
842                 scsi_eh_finish_cmd(scmd, done_q);
843         }
844
845         return list_empty(work_q);
846 }
847 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
848
849 /**
850  * scsi_eh_tur - Send TUR to device.
851  * @scmd:       Scsi cmd to send TUR
852  *
853  * Return value:
854  *    0 - Device is ready. 1 - Device NOT ready.
855  **/
856 static int scsi_eh_tur(struct scsi_cmnd *scmd)
857 {
858         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
859         int retry_cnt = 1, rtn;
860
861 retry_tur:
862         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
863
864         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
865                 __FUNCTION__, scmd, rtn));
866
867         switch (rtn) {
868         case NEEDS_RETRY:
869                 if (retry_cnt--)
870                         goto retry_tur;
871                 /*FALLTHRU*/
872         case SUCCESS:
873                 return 0;
874         default:
875                 return 1;
876         }
877 }
878
879 /**
880  * scsi_eh_abort_cmds - abort canceled commands.
881  * @shost:      scsi host being recovered.
882  * @eh_done_q:  list_head for processed commands.
883  *
884  * Decription:
885  *    Try and see whether or not it makes sense to try and abort the
886  *    running command.  this only works out to be the case if we have one
887  *    command that has timed out.  if the command simply failed, it makes
888  *    no sense to try and abort the command, since as far as the shost
889  *    adapter is concerned, it isn't running.
890  **/
891 static int scsi_eh_abort_cmds(struct list_head *work_q,
892                               struct list_head *done_q)
893 {
894         struct scsi_cmnd *scmd, *next;
895         int rtn;
896
897         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
898                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
899                         continue;
900                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
901                                                   "0x%p\n", current->comm,
902                                                   scmd));
903                 rtn = scsi_try_to_abort_cmd(scmd);
904                 if (rtn == SUCCESS) {
905                         scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
906                         if (!scsi_device_online(scmd->device) ||
907                             !scsi_eh_tur(scmd)) {
908                                 scsi_eh_finish_cmd(scmd, done_q);
909                         }
910                                 
911                 } else
912                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
913                                                           " cmd failed:"
914                                                           "0x%p\n",
915                                                           current->comm,
916                                                           scmd));
917         }
918
919         return list_empty(work_q);
920 }
921
922 /**
923  * scsi_eh_try_stu - Send START_UNIT to device.
924  * @scmd:       Scsi cmd to send START_UNIT
925  *
926  * Return value:
927  *    0 - Device is ready. 1 - Device NOT ready.
928  **/
929 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
930 {
931         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
932
933         if (scmd->device->allow_restart) {
934                 int i, rtn = NEEDS_RETRY;
935
936                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
937                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
938                                                 scmd->device->timeout, 0);
939
940                 if (rtn == SUCCESS)
941                         return 0;
942         }
943
944         return 1;
945 }
946
947  /**
948  * scsi_eh_stu - send START_UNIT if needed
949  * @shost:      scsi host being recovered.
950  * @eh_done_q:  list_head for processed commands.
951  *
952  * Notes:
953  *    If commands are failing due to not ready, initializing command required,
954  *      try revalidating the device, which will end up sending a start unit. 
955  **/
956 static int scsi_eh_stu(struct Scsi_Host *shost,
957                               struct list_head *work_q,
958                               struct list_head *done_q)
959 {
960         struct scsi_cmnd *scmd, *stu_scmd, *next;
961         struct scsi_device *sdev;
962
963         shost_for_each_device(sdev, shost) {
964                 stu_scmd = NULL;
965                 list_for_each_entry(scmd, work_q, eh_entry)
966                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
967                             scsi_check_sense(scmd) == FAILED ) {
968                                 stu_scmd = scmd;
969                                 break;
970                         }
971
972                 if (!stu_scmd)
973                         continue;
974
975                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
976                                                   " 0x%p\n", current->comm, sdev));
977
978                 if (!scsi_eh_try_stu(stu_scmd)) {
979                         if (!scsi_device_online(sdev) ||
980                             !scsi_eh_tur(stu_scmd)) {
981                                 list_for_each_entry_safe(scmd, next,
982                                                           work_q, eh_entry) {
983                                         if (scmd->device == sdev)
984                                                 scsi_eh_finish_cmd(scmd, done_q);
985                                 }
986                         }
987                 } else {
988                         SCSI_LOG_ERROR_RECOVERY(3,
989                                                 printk("%s: START_UNIT failed to sdev:"
990                                                        " 0x%p\n", current->comm, sdev));
991                 }
992         }
993
994         return list_empty(work_q);
995 }
996
997
998 /**
999  * scsi_eh_bus_device_reset - send bdr if needed
1000  * @shost:      scsi host being recovered.
1001  * @eh_done_q:  list_head for processed commands.
1002  *
1003  * Notes:
1004  *    Try a bus device reset.  still, look to see whether we have multiple
1005  *    devices that are jammed or not - if we have multiple devices, it
1006  *    makes no sense to try bus_device_reset - we really would need to try
1007  *    a bus_reset instead. 
1008  **/
1009 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
1010                                     struct list_head *work_q,
1011                                     struct list_head *done_q)
1012 {
1013         struct scsi_cmnd *scmd, *bdr_scmd, *next;
1014         struct scsi_device *sdev;
1015         int rtn;
1016
1017         shost_for_each_device(sdev, shost) {
1018                 bdr_scmd = NULL;
1019                 list_for_each_entry(scmd, work_q, eh_entry)
1020                         if (scmd->device == sdev) {
1021                                 bdr_scmd = scmd;
1022                                 break;
1023                         }
1024
1025                 if (!bdr_scmd)
1026                         continue;
1027
1028                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1029                                                   " 0x%p\n", current->comm,
1030                                                   sdev));
1031                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1032                 if (rtn == SUCCESS) {
1033                         if (!scsi_device_online(sdev) ||
1034                             !scsi_eh_tur(bdr_scmd)) {
1035                                 list_for_each_entry_safe(scmd, next,
1036                                                          work_q, eh_entry) {
1037                                         if (scmd->device == sdev)
1038                                                 scsi_eh_finish_cmd(scmd,
1039                                                                    done_q);
1040                                 }
1041                         }
1042                 } else {
1043                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1044                                                           " failed sdev:"
1045                                                           "0x%p\n",
1046                                                           current->comm,
1047                                                            sdev));
1048                 }
1049         }
1050
1051         return list_empty(work_q);
1052 }
1053
1054 /**
1055  * scsi_eh_bus_reset - send a bus reset 
1056  * @shost:      scsi host being recovered.
1057  * @eh_done_q:  list_head for processed commands.
1058  **/
1059 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1060                              struct list_head *work_q,
1061                              struct list_head *done_q)
1062 {
1063         struct scsi_cmnd *scmd, *chan_scmd, *next;
1064         unsigned int channel;
1065         int rtn;
1066
1067         /*
1068          * we really want to loop over the various channels, and do this on
1069          * a channel by channel basis.  we should also check to see if any
1070          * of the failed commands are on soft_reset devices, and if so, skip
1071          * the reset.  
1072          */
1073
1074         for (channel = 0; channel <= shost->max_channel; channel++) {
1075                 chan_scmd = NULL;
1076                 list_for_each_entry(scmd, work_q, eh_entry) {
1077                         if (channel == scmd_channel(scmd)) {
1078                                 chan_scmd = scmd;
1079                                 break;
1080                                 /*
1081                                  * FIXME add back in some support for
1082                                  * soft_reset devices.
1083                                  */
1084                         }
1085                 }
1086
1087                 if (!chan_scmd)
1088                         continue;
1089                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1090                                                   " %d\n", current->comm,
1091                                                   channel));
1092                 rtn = scsi_try_bus_reset(chan_scmd);
1093                 if (rtn == SUCCESS) {
1094                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1095                                 if (channel == scmd_channel(scmd))
1096                                         if (!scsi_device_online(scmd->device) ||
1097                                             !scsi_eh_tur(scmd))
1098                                                 scsi_eh_finish_cmd(scmd,
1099                                                                    done_q);
1100                         }
1101                 } else {
1102                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1103                                                           " failed chan: %d\n",
1104                                                           current->comm,
1105                                                           channel));
1106                 }
1107         }
1108         return list_empty(work_q);
1109 }
1110
1111 /**
1112  * scsi_eh_host_reset - send a host reset 
1113  * @work_q:     list_head for processed commands.
1114  * @done_q:     list_head for processed commands.
1115  **/
1116 static int scsi_eh_host_reset(struct list_head *work_q,
1117                               struct list_head *done_q)
1118 {
1119         struct scsi_cmnd *scmd, *next;
1120         int rtn;
1121
1122         if (!list_empty(work_q)) {
1123                 scmd = list_entry(work_q->next,
1124                                   struct scsi_cmnd, eh_entry);
1125
1126                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1127                                                   , current->comm));
1128
1129                 rtn = scsi_try_host_reset(scmd);
1130                 if (rtn == SUCCESS) {
1131                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1132                                 if (!scsi_device_online(scmd->device) ||
1133                                     (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1134                                     !scsi_eh_tur(scmd))
1135                                         scsi_eh_finish_cmd(scmd, done_q);
1136                         }
1137                 } else {
1138                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1139                                                           " failed\n",
1140                                                           current->comm));
1141                 }
1142         }
1143         return list_empty(work_q);
1144 }
1145
1146 /**
1147  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1148  * @work_q:     list_head for processed commands.
1149  * @done_q:     list_head for processed commands.
1150  *
1151  **/
1152 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1153                                   struct list_head *done_q)
1154 {
1155         struct scsi_cmnd *scmd, *next;
1156
1157         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1158                 sdev_printk(KERN_INFO, scmd->device,
1159                             "scsi: Device offlined - not"
1160                             " ready after error recovery\n");
1161                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1162                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1163                         /*
1164                          * FIXME: Handle lost cmds.
1165                          */
1166                 }
1167                 scsi_eh_finish_cmd(scmd, done_q);
1168         }
1169         return;
1170 }
1171
1172 /**
1173  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1174  * @scmd:       SCSI cmd to examine.
1175  *
1176  * Notes:
1177  *    This is *only* called when we are examining the status after sending
1178  *    out the actual data command.  any commands that are queued for error
1179  *    recovery (e.g. test_unit_ready) do *not* come through here.
1180  *
1181  *    When this routine returns failed, it means the error handler thread
1182  *    is woken.  In cases where the error code indicates an error that
1183  *    doesn't require the error handler read (i.e. we don't need to
1184  *    abort/reset), this function should return SUCCESS.
1185  **/
1186 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1187 {
1188         int rtn;
1189
1190         /*
1191          * if the device is offline, then we clearly just pass the result back
1192          * up to the top level.
1193          */
1194         if (!scsi_device_online(scmd->device)) {
1195                 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1196                                                   " as SUCCESS\n",
1197                                                   __FUNCTION__));
1198                 return SUCCESS;
1199         }
1200
1201         /*
1202          * first check the host byte, to see if there is anything in there
1203          * that would indicate what we need to do.
1204          */
1205         switch (host_byte(scmd->result)) {
1206         case DID_PASSTHROUGH:
1207                 /*
1208                  * no matter what, pass this through to the upper layer.
1209                  * nuke this special code so that it looks like we are saying
1210                  * did_ok.
1211                  */
1212                 scmd->result &= 0xff00ffff;
1213                 return SUCCESS;
1214         case DID_OK:
1215                 /*
1216                  * looks good.  drop through, and check the next byte.
1217                  */
1218                 break;
1219         case DID_NO_CONNECT:
1220         case DID_BAD_TARGET:
1221         case DID_ABORT:
1222                 /*
1223                  * note - this means that we just report the status back
1224                  * to the top level driver, not that we actually think
1225                  * that it indicates SUCCESS.
1226                  */
1227                 return SUCCESS;
1228                 /*
1229                  * when the low level driver returns did_soft_error,
1230                  * it is responsible for keeping an internal retry counter 
1231                  * in order to avoid endless loops (db)
1232                  *
1233                  * actually this is a bug in this function here.  we should
1234                  * be mindful of the maximum number of retries specified
1235                  * and not get stuck in a loop.
1236                  */
1237         case DID_SOFT_ERROR:
1238                 goto maybe_retry;
1239         case DID_IMM_RETRY:
1240                 return NEEDS_RETRY;
1241
1242         case DID_REQUEUE:
1243                 return ADD_TO_MLQUEUE;
1244
1245         case DID_ERROR:
1246                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1247                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1248                         /*
1249                          * execute reservation conflict processing code
1250                          * lower down
1251                          */
1252                         break;
1253                 /* fallthrough */
1254
1255         case DID_BUS_BUSY:
1256         case DID_PARITY:
1257                 goto maybe_retry;
1258         case DID_TIME_OUT:
1259                 /*
1260                  * when we scan the bus, we get timeout messages for
1261                  * these commands if there is no device available.
1262                  * other hosts report did_no_connect for the same thing.
1263                  */
1264                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1265                      scmd->cmnd[0] == INQUIRY)) {
1266                         return SUCCESS;
1267                 } else {
1268                         return FAILED;
1269                 }
1270         case DID_RESET:
1271                 return SUCCESS;
1272         default:
1273                 return FAILED;
1274         }
1275
1276         /*
1277          * next, check the message byte.
1278          */
1279         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1280                 return FAILED;
1281
1282         /*
1283          * check the status byte to see if this indicates anything special.
1284          */
1285         switch (status_byte(scmd->result)) {
1286         case QUEUE_FULL:
1287                 /*
1288                  * the case of trying to send too many commands to a
1289                  * tagged queueing device.
1290                  */
1291         case BUSY:
1292                 /*
1293                  * device can't talk to us at the moment.  Should only
1294                  * occur (SAM-3) when the task queue is empty, so will cause
1295                  * the empty queue handling to trigger a stall in the
1296                  * device.
1297                  */
1298                 return ADD_TO_MLQUEUE;
1299         case GOOD:
1300         case COMMAND_TERMINATED:
1301         case TASK_ABORTED:
1302                 return SUCCESS;
1303         case CHECK_CONDITION:
1304                 rtn = scsi_check_sense(scmd);
1305                 if (rtn == NEEDS_RETRY)
1306                         goto maybe_retry;
1307                 /* if rtn == FAILED, we have no sense information;
1308                  * returning FAILED will wake the error handler thread
1309                  * to collect the sense and redo the decide
1310                  * disposition */
1311                 return rtn;
1312         case CONDITION_GOOD:
1313         case INTERMEDIATE_GOOD:
1314         case INTERMEDIATE_C_GOOD:
1315         case ACA_ACTIVE:
1316                 /*
1317                  * who knows?  FIXME(eric)
1318                  */
1319                 return SUCCESS;
1320
1321         case RESERVATION_CONFLICT:
1322                 sdev_printk(KERN_INFO, scmd->device,
1323                             "reservation conflict\n");
1324                 return SUCCESS; /* causes immediate i/o error */
1325         default:
1326                 return FAILED;
1327         }
1328         return FAILED;
1329
1330       maybe_retry:
1331
1332         /* we requeue for retry because the error was retryable, and
1333          * the request was not marked fast fail.  Note that above,
1334          * even if the request is marked fast fail, we still requeue
1335          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1336         if ((++scmd->retries) <= scmd->allowed
1337             && !blk_noretry_request(scmd->request)) {
1338                 return NEEDS_RETRY;
1339         } else {
1340                 /*
1341                  * no more retries - report this one back to upper level.
1342                  */
1343                 return SUCCESS;
1344         }
1345 }
1346
1347 /**
1348  * scsi_eh_lock_door - Prevent medium removal for the specified device
1349  * @sdev:       SCSI device to prevent medium removal
1350  *
1351  * Locking:
1352  *      We must be called from process context; scsi_allocate_request()
1353  *      may sleep.
1354  *
1355  * Notes:
1356  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1357  *      head of the devices request queue, and continue.
1358  *
1359  * Bugs:
1360  *      scsi_allocate_request() may sleep waiting for existing requests to
1361  *      be processed.  However, since we haven't kicked off any request
1362  *      processing for this host, this may deadlock.
1363  *
1364  *      If scsi_allocate_request() fails for what ever reason, we
1365  *      completely forget to lock the door.
1366  **/
1367 static void scsi_eh_lock_door(struct scsi_device *sdev)
1368 {
1369         unsigned char cmnd[MAX_COMMAND_SIZE];
1370
1371         cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1372         cmnd[1] = 0;
1373         cmnd[2] = 0;
1374         cmnd[3] = 0;
1375         cmnd[4] = SCSI_REMOVAL_PREVENT;
1376         cmnd[5] = 0;
1377
1378         scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
1379                            5, NULL, NULL, GFP_KERNEL);
1380 }
1381
1382
1383 /**
1384  * scsi_restart_operations - restart io operations to the specified host.
1385  * @shost:      Host we are restarting.
1386  *
1387  * Notes:
1388  *    When we entered the error handler, we blocked all further i/o to
1389  *    this device.  we need to 'reverse' this process.
1390  **/
1391 static void scsi_restart_operations(struct Scsi_Host *shost)
1392 {
1393         struct scsi_device *sdev;
1394         unsigned long flags;
1395
1396         /*
1397          * If the door was locked, we need to insert a door lock request
1398          * onto the head of the SCSI request queue for the device.  There
1399          * is no point trying to lock the door of an off-line device.
1400          */
1401         shost_for_each_device(sdev, shost) {
1402                 if (scsi_device_online(sdev) && sdev->locked)
1403                         scsi_eh_lock_door(sdev);
1404         }
1405
1406         /*
1407          * next free up anything directly waiting upon the host.  this
1408          * will be requests for character device operations, and also for
1409          * ioctls to queued block devices.
1410          */
1411         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1412                                           __FUNCTION__));
1413
1414         spin_lock_irqsave(shost->host_lock, flags);
1415         if (scsi_host_set_state(shost, SHOST_RUNNING))
1416                 if (scsi_host_set_state(shost, SHOST_CANCEL))
1417                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1418         spin_unlock_irqrestore(shost->host_lock, flags);
1419
1420         wake_up(&shost->host_wait);
1421
1422         /*
1423          * finally we need to re-initiate requests that may be pending.  we will
1424          * have had everything blocked while error handling is taking place, and
1425          * now that error recovery is done, we will need to ensure that these
1426          * requests are started.
1427          */
1428         scsi_run_host_queues(shost);
1429 }
1430
1431 /**
1432  * scsi_eh_ready_devs - check device ready state and recover if not.
1433  * @shost:      host to be recovered.
1434  * @eh_done_q:  list_head for processed commands.
1435  *
1436  **/
1437 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1438                         struct list_head *work_q,
1439                         struct list_head *done_q)
1440 {
1441         if (!scsi_eh_stu(shost, work_q, done_q))
1442                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1443                         if (!scsi_eh_bus_reset(shost, work_q, done_q))
1444                                 if (!scsi_eh_host_reset(work_q, done_q))
1445                                         scsi_eh_offline_sdevs(work_q, done_q);
1446 }
1447 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1448
1449 /**
1450  * scsi_eh_flush_done_q - finish processed commands or retry them.
1451  * @done_q:     list_head of processed commands.
1452  *
1453  **/
1454 void scsi_eh_flush_done_q(struct list_head *done_q)
1455 {
1456         struct scsi_cmnd *scmd, *next;
1457
1458         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1459                 list_del_init(&scmd->eh_entry);
1460                 if (scsi_device_online(scmd->device) &&
1461                     !blk_noretry_request(scmd->request) &&
1462                     (++scmd->retries <= scmd->allowed)) {
1463                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1464                                                           " retry cmd: %p\n",
1465                                                           current->comm,
1466                                                           scmd));
1467                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1468                 } else {
1469                         /*
1470                          * If just we got sense for the device (called
1471                          * scsi_eh_get_sense), scmd->result is already
1472                          * set, do not set DRIVER_TIMEOUT.
1473                          */
1474                         if (!scmd->result)
1475                                 scmd->result |= (DRIVER_TIMEOUT << 24);
1476                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1477                                                         " cmd: %p\n",
1478                                                         current->comm, scmd));
1479                         scsi_finish_command(scmd);
1480                 }
1481         }
1482 }
1483 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1484
1485 /**
1486  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1487  * @shost:      Host to unjam.
1488  *
1489  * Notes:
1490  *    When we come in here, we *know* that all commands on the bus have
1491  *    either completed, failed or timed out.  we also know that no further
1492  *    commands are being sent to the host, so things are relatively quiet
1493  *    and we have freedom to fiddle with things as we wish.
1494  *
1495  *    This is only the *default* implementation.  it is possible for
1496  *    individual drivers to supply their own version of this function, and
1497  *    if the maintainer wishes to do this, it is strongly suggested that
1498  *    this function be taken as a template and modified.  this function
1499  *    was designed to correctly handle problems for about 95% of the
1500  *    different cases out there, and it should always provide at least a
1501  *    reasonable amount of error recovery.
1502  *
1503  *    Any command marked 'failed' or 'timeout' must eventually have
1504  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1505  *    here, so when we restart the host after we return it should have an
1506  *    empty queue.
1507  **/
1508 static void scsi_unjam_host(struct Scsi_Host *shost)
1509 {
1510         unsigned long flags;
1511         LIST_HEAD(eh_work_q);
1512         LIST_HEAD(eh_done_q);
1513
1514         spin_lock_irqsave(shost->host_lock, flags);
1515         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1516         spin_unlock_irqrestore(shost->host_lock, flags);
1517
1518         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1519
1520         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1521                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1522                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1523
1524         scsi_eh_flush_done_q(&eh_done_q);
1525 }
1526
1527 /**
1528  * scsi_error_handler - SCSI error handler thread
1529  * @data:       Host for which we are running.
1530  *
1531  * Notes:
1532  *    This is the main error handling loop.  This is run as a kernel thread
1533  *    for every SCSI host and handles all error handling activity.
1534  **/
1535 int scsi_error_handler(void *data)
1536 {
1537         struct Scsi_Host *shost = data;
1538
1539         current->flags |= PF_NOFREEZE;
1540
1541         /*
1542          * We use TASK_INTERRUPTIBLE so that the thread is not
1543          * counted against the load average as a running process.
1544          * We never actually get interrupted because kthread_run
1545          * disables singal delivery for the created thread.
1546          */
1547         set_current_state(TASK_INTERRUPTIBLE);
1548         while (!kthread_should_stop()) {
1549                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1550                     shost->host_failed != shost->host_busy) {
1551                         SCSI_LOG_ERROR_RECOVERY(1,
1552                                 printk("Error handler scsi_eh_%d sleeping\n",
1553                                         shost->host_no));
1554                         schedule();
1555                         set_current_state(TASK_INTERRUPTIBLE);
1556                         continue;
1557                 }
1558
1559                 __set_current_state(TASK_RUNNING);
1560                 SCSI_LOG_ERROR_RECOVERY(1,
1561                         printk("Error handler scsi_eh_%d waking up\n",
1562                                 shost->host_no));
1563
1564                 /*
1565                  * We have a host that is failing for some reason.  Figure out
1566                  * what we need to do to get it up and online again (if we can).
1567                  * If we fail, we end up taking the thing offline.
1568                  */
1569                 if (shost->transportt->eh_strategy_handler)
1570                         shost->transportt->eh_strategy_handler(shost);
1571                 else
1572                         scsi_unjam_host(shost);
1573
1574                 /*
1575                  * Note - if the above fails completely, the action is to take
1576                  * individual devices offline and flush the queue of any
1577                  * outstanding requests that may have been pending.  When we
1578                  * restart, we restart any I/O to any other devices on the bus
1579                  * which are still online.
1580                  */
1581                 scsi_restart_operations(shost);
1582                 set_current_state(TASK_INTERRUPTIBLE);
1583         }
1584         __set_current_state(TASK_RUNNING);
1585
1586         SCSI_LOG_ERROR_RECOVERY(1,
1587                 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1588         shost->ehandler = NULL;
1589         return 0;
1590 }
1591
1592 /*
1593  * Function:    scsi_report_bus_reset()
1594  *
1595  * Purpose:     Utility function used by low-level drivers to report that
1596  *              they have observed a bus reset on the bus being handled.
1597  *
1598  * Arguments:   shost       - Host in question
1599  *              channel     - channel on which reset was observed.
1600  *
1601  * Returns:     Nothing
1602  *
1603  * Lock status: Host lock must be held.
1604  *
1605  * Notes:       This only needs to be called if the reset is one which
1606  *              originates from an unknown location.  Resets originated
1607  *              by the mid-level itself don't need to call this, but there
1608  *              should be no harm.
1609  *
1610  *              The main purpose of this is to make sure that a CHECK_CONDITION
1611  *              is properly treated.
1612  */
1613 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1614 {
1615         struct scsi_device *sdev;
1616
1617         __shost_for_each_device(sdev, shost) {
1618                 if (channel == sdev_channel(sdev)) {
1619                         sdev->was_reset = 1;
1620                         sdev->expecting_cc_ua = 1;
1621                 }
1622         }
1623 }
1624 EXPORT_SYMBOL(scsi_report_bus_reset);
1625
1626 /*
1627  * Function:    scsi_report_device_reset()
1628  *
1629  * Purpose:     Utility function used by low-level drivers to report that
1630  *              they have observed a device reset on the device being handled.
1631  *
1632  * Arguments:   shost       - Host in question
1633  *              channel     - channel on which reset was observed
1634  *              target      - target on which reset was observed
1635  *
1636  * Returns:     Nothing
1637  *
1638  * Lock status: Host lock must be held
1639  *
1640  * Notes:       This only needs to be called if the reset is one which
1641  *              originates from an unknown location.  Resets originated
1642  *              by the mid-level itself don't need to call this, but there
1643  *              should be no harm.
1644  *
1645  *              The main purpose of this is to make sure that a CHECK_CONDITION
1646  *              is properly treated.
1647  */
1648 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1649 {
1650         struct scsi_device *sdev;
1651
1652         __shost_for_each_device(sdev, shost) {
1653                 if (channel == sdev_channel(sdev) &&
1654                     target == sdev_id(sdev)) {
1655                         sdev->was_reset = 1;
1656                         sdev->expecting_cc_ua = 1;
1657                 }
1658         }
1659 }
1660 EXPORT_SYMBOL(scsi_report_device_reset);
1661
1662 static void
1663 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1664 {
1665 }
1666
1667 /*
1668  * Function:    scsi_reset_provider
1669  *
1670  * Purpose:     Send requested reset to a bus or device at any phase.
1671  *
1672  * Arguments:   device  - device to send reset to
1673  *              flag - reset type (see scsi.h)
1674  *
1675  * Returns:     SUCCESS/FAILURE.
1676  *
1677  * Notes:       This is used by the SCSI Generic driver to provide
1678  *              Bus/Device reset capability.
1679  */
1680 int
1681 scsi_reset_provider(struct scsi_device *dev, int flag)
1682 {
1683         struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1684         struct Scsi_Host *shost = dev->host;
1685         struct request req;
1686         unsigned long flags;
1687         int rtn;
1688
1689         scmd->request = &req;
1690         memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1691
1692         memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1693     
1694         scmd->scsi_done         = scsi_reset_provider_done_command;
1695         scmd->done                      = NULL;
1696         scmd->request_buffer            = NULL;
1697         scmd->request_bufflen           = 0;
1698
1699         scmd->cmd_len                   = 0;
1700
1701         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
1702
1703         init_timer(&scmd->eh_timeout);
1704
1705         /*
1706          * Sometimes the command can get back into the timer chain,
1707          * so use the pid as an identifier.
1708          */
1709         scmd->pid                       = 0;
1710
1711         spin_lock_irqsave(shost->host_lock, flags);
1712         shost->tmf_in_progress = 1;
1713         spin_unlock_irqrestore(shost->host_lock, flags);
1714
1715         switch (flag) {
1716         case SCSI_TRY_RESET_DEVICE:
1717                 rtn = scsi_try_bus_device_reset(scmd);
1718                 if (rtn == SUCCESS)
1719                         break;
1720                 /* FALLTHROUGH */
1721         case SCSI_TRY_RESET_BUS:
1722                 rtn = scsi_try_bus_reset(scmd);
1723                 if (rtn == SUCCESS)
1724                         break;
1725                 /* FALLTHROUGH */
1726         case SCSI_TRY_RESET_HOST:
1727                 rtn = scsi_try_host_reset(scmd);
1728                 break;
1729         default:
1730                 rtn = FAILED;
1731         }
1732
1733         spin_lock_irqsave(shost->host_lock, flags);
1734         shost->tmf_in_progress = 0;
1735         spin_unlock_irqrestore(shost->host_lock, flags);
1736
1737         /*
1738          * be sure to wake up anyone who was sleeping or had their queue
1739          * suspended while we performed the TMF.
1740          */
1741         SCSI_LOG_ERROR_RECOVERY(3,
1742                 printk("%s: waking up host to restart after TMF\n",
1743                 __FUNCTION__));
1744
1745         wake_up(&shost->host_wait);
1746
1747         scsi_run_host_queues(shost);
1748
1749         scsi_next_command(scmd);
1750         return rtn;
1751 }
1752 EXPORT_SYMBOL(scsi_reset_provider);
1753
1754 /**
1755  * scsi_normalize_sense - normalize main elements from either fixed or
1756  *                      descriptor sense data format into a common format.
1757  *
1758  * @sense_buffer:       byte array containing sense data returned by device
1759  * @sb_len:             number of valid bytes in sense_buffer
1760  * @sshdr:              pointer to instance of structure that common
1761  *                      elements are written to.
1762  *
1763  * Notes:
1764  *      The "main elements" from sense data are: response_code, sense_key,
1765  *      asc, ascq and additional_length (only for descriptor format).
1766  *
1767  *      Typically this function can be called after a device has
1768  *      responded to a SCSI command with the CHECK_CONDITION status.
1769  *
1770  * Return value:
1771  *      1 if valid sense data information found, else 0;
1772  **/
1773 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1774                          struct scsi_sense_hdr *sshdr)
1775 {
1776         if (!sense_buffer || !sb_len)
1777                 return 0;
1778
1779         memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1780
1781         sshdr->response_code = (sense_buffer[0] & 0x7f);
1782
1783         if (!scsi_sense_valid(sshdr))
1784                 return 0;
1785
1786         if (sshdr->response_code >= 0x72) {
1787                 /*
1788                  * descriptor format
1789                  */
1790                 if (sb_len > 1)
1791                         sshdr->sense_key = (sense_buffer[1] & 0xf);
1792                 if (sb_len > 2)
1793                         sshdr->asc = sense_buffer[2];
1794                 if (sb_len > 3)
1795                         sshdr->ascq = sense_buffer[3];
1796                 if (sb_len > 7)
1797                         sshdr->additional_length = sense_buffer[7];
1798         } else {
1799                 /* 
1800                  * fixed format
1801                  */
1802                 if (sb_len > 2)
1803                         sshdr->sense_key = (sense_buffer[2] & 0xf);
1804                 if (sb_len > 7) {
1805                         sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1806                                          sb_len : (sense_buffer[7] + 8);
1807                         if (sb_len > 12)
1808                                 sshdr->asc = sense_buffer[12];
1809                         if (sb_len > 13)
1810                                 sshdr->ascq = sense_buffer[13];
1811                 }
1812         }
1813
1814         return 1;
1815 }
1816 EXPORT_SYMBOL(scsi_normalize_sense);
1817
1818 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1819                                  struct scsi_sense_hdr *sshdr)
1820 {
1821         return scsi_normalize_sense(cmd->sense_buffer,
1822                         sizeof(cmd->sense_buffer), sshdr);
1823 }
1824 EXPORT_SYMBOL(scsi_command_normalize_sense);
1825
1826 /**
1827  * scsi_sense_desc_find - search for a given descriptor type in
1828  *                      descriptor sense data format.
1829  *
1830  * @sense_buffer:       byte array of descriptor format sense data
1831  * @sb_len:             number of valid bytes in sense_buffer
1832  * @desc_type:          value of descriptor type to find
1833  *                      (e.g. 0 -> information)
1834  *
1835  * Notes:
1836  *      only valid when sense data is in descriptor format
1837  *
1838  * Return value:
1839  *      pointer to start of (first) descriptor if found else NULL
1840  **/
1841 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1842                                 int desc_type)
1843 {
1844         int add_sen_len, add_len, desc_len, k;
1845         const u8 * descp;
1846
1847         if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1848                 return NULL;
1849         if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1850                 return NULL;
1851         add_sen_len = (add_sen_len < (sb_len - 8)) ?
1852                         add_sen_len : (sb_len - 8);
1853         descp = &sense_buffer[8];
1854         for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1855                 descp += desc_len;
1856                 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1857                 desc_len = add_len + 2;
1858                 if (descp[0] == desc_type)
1859                         return descp;
1860                 if (add_len < 0) // short descriptor ??
1861                         break;
1862         }
1863         return NULL;
1864 }
1865 EXPORT_SYMBOL(scsi_sense_desc_find);
1866
1867 /**
1868  * scsi_get_sense_info_fld - attempts to get information field from
1869  *                      sense data (either fixed or descriptor format)
1870  *
1871  * @sense_buffer:       byte array of sense data
1872  * @sb_len:             number of valid bytes in sense_buffer
1873  * @info_out:           pointer to 64 integer where 8 or 4 byte information
1874  *                      field will be placed if found.
1875  *
1876  * Return value:
1877  *      1 if information field found, 0 if not found.
1878  **/
1879 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1880                             u64 * info_out)
1881 {
1882         int j;
1883         const u8 * ucp;
1884         u64 ull;
1885
1886         if (sb_len < 7)
1887                 return 0;
1888         switch (sense_buffer[0] & 0x7f) {
1889         case 0x70:
1890         case 0x71:
1891                 if (sense_buffer[0] & 0x80) {
1892                         *info_out = (sense_buffer[3] << 24) +
1893                                     (sense_buffer[4] << 16) +
1894                                     (sense_buffer[5] << 8) + sense_buffer[6];
1895                         return 1;
1896                 } else
1897                         return 0;
1898         case 0x72:
1899         case 0x73:
1900                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1901                                            0 /* info desc */);
1902                 if (ucp && (0xa == ucp[1])) {
1903                         ull = 0;
1904                         for (j = 0; j < 8; ++j) {
1905                                 if (j > 0)
1906                                         ull <<= 8;
1907                                 ull |= ucp[4 + j];
1908                         }
1909                         *info_out = ull;
1910                         return 1;
1911                 } else
1912                         return 0;
1913         default:
1914                 return 0;
1915         }
1916 }
1917 EXPORT_SYMBOL(scsi_get_sense_info_fld);