Use menuconfig objects - CONFIG_ISDN_I4L
[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/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 #include <linux/scatterlist.h>
28
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_transport.h>
35 #include <scsi/scsi_host.h>
36 #include <scsi/scsi_ioctl.h>
37
38 #include "scsi_priv.h"
39 #include "scsi_logging.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_send_eh_cmnd  - submit a scsi command as part of error recory
593  * @scmd:       SCSI command structure to hijack
594  * @cmnd:       CDB to send
595  * @cmnd_size:  size in bytes of @cmnd
596  * @timeout:    timeout for this request
597  * @copy_sense: request sense data if set to 1
598  *
599  * This function is used to send a scsi command down to a target device
600  * as part of the error recovery process.  If @copy_sense is 0 the command
601  * sent must be one that does not transfer any data.  If @copy_sense is 1
602  * the command must be REQUEST_SENSE and this functions copies out the
603  * sense buffer it got into @scmd->sense_buffer.
604  *
605  * Return value:
606  *    SUCCESS or FAILED or NEEDS_RETRY
607  **/
608 static int scsi_send_eh_cmnd(struct scsi_cmnd *scmd, unsigned char *cmnd,
609                              int cmnd_size, int timeout, int copy_sense)
610 {
611         struct scsi_device *sdev = scmd->device;
612         struct Scsi_Host *shost = sdev->host;
613         int old_result = scmd->result;
614         DECLARE_COMPLETION_ONSTACK(done);
615         unsigned long timeleft;
616         unsigned long flags;
617         struct scatterlist sgl;
618         unsigned char old_cmnd[MAX_COMMAND_SIZE];
619         enum dma_data_direction old_data_direction;
620         unsigned short old_use_sg;
621         unsigned char old_cmd_len;
622         unsigned old_bufflen;
623         void *old_buffer;
624         int rtn;
625
626         /*
627          * We need saved copies of a number of fields - this is because
628          * error handling may need to overwrite these with different values
629          * to run different commands, and once error handling is complete,
630          * we will need to restore these values prior to running the actual
631          * command.
632          */
633         old_buffer = scmd->request_buffer;
634         old_bufflen = scmd->request_bufflen;
635         memcpy(old_cmnd, scmd->cmnd, sizeof(scmd->cmnd));
636         old_data_direction = scmd->sc_data_direction;
637         old_cmd_len = scmd->cmd_len;
638         old_use_sg = scmd->use_sg;
639
640         memset(scmd->cmnd, 0, sizeof(scmd->cmnd));
641         memcpy(scmd->cmnd, cmnd, cmnd_size);
642
643         if (copy_sense) {
644                 sg_init_one(&sgl, scmd->sense_buffer,
645                             sizeof(scmd->sense_buffer));
646
647                 scmd->sc_data_direction = DMA_FROM_DEVICE;
648                 scmd->request_bufflen = sgl.length;
649                 scmd->request_buffer = &sgl;
650                 scmd->use_sg = 1;
651         } else {
652                 scmd->request_buffer = NULL;
653                 scmd->request_bufflen = 0;
654                 scmd->sc_data_direction = DMA_NONE;
655                 scmd->use_sg = 0;
656         }
657
658         scmd->underflow = 0;
659         scmd->cmd_len = COMMAND_SIZE(scmd->cmnd[0]);
660
661         if (sdev->scsi_level <= SCSI_2)
662                 scmd->cmnd[1] = (scmd->cmnd[1] & 0x1f) |
663                         (sdev->lun << 5 & 0xe0);
664
665         /*
666          * Zero the sense buffer.  The scsi spec mandates that any
667          * untransferred sense data should be interpreted as being zero.
668          */
669         memset(scmd->sense_buffer, 0, sizeof(scmd->sense_buffer));
670
671         shost->eh_action = &done;
672
673         spin_lock_irqsave(shost->host_lock, flags);
674         scsi_log_send(scmd);
675         shost->hostt->queuecommand(scmd, scsi_eh_done);
676         spin_unlock_irqrestore(shost->host_lock, flags);
677
678         timeleft = wait_for_completion_timeout(&done, timeout);
679
680         shost->eh_action = NULL;
681
682         scsi_log_completion(scmd, SUCCESS);
683
684         SCSI_LOG_ERROR_RECOVERY(3,
685                 printk("%s: scmd: %p, timeleft: %ld\n",
686                         __FUNCTION__, scmd, timeleft));
687
688         /*
689          * If there is time left scsi_eh_done got called, and we will
690          * examine the actual status codes to see whether the command
691          * actually did complete normally, else tell the host to forget
692          * about this command.
693          */
694         if (timeleft) {
695                 rtn = scsi_eh_completed_normally(scmd);
696                 SCSI_LOG_ERROR_RECOVERY(3,
697                         printk("%s: scsi_eh_completed_normally %x\n",
698                                __FUNCTION__, rtn));
699
700                 switch (rtn) {
701                 case SUCCESS:
702                 case NEEDS_RETRY:
703                 case FAILED:
704                         break;
705                 default:
706                         rtn = FAILED;
707                         break;
708                 }
709         } else {
710                 scsi_abort_eh_cmnd(scmd);
711                 rtn = FAILED;
712         }
713
714
715         /*
716          * Restore original data
717          */
718         scmd->request_buffer = old_buffer;
719         scmd->request_bufflen = old_bufflen;
720         memcpy(scmd->cmnd, old_cmnd, sizeof(scmd->cmnd));
721         scmd->sc_data_direction = old_data_direction;
722         scmd->cmd_len = old_cmd_len;
723         scmd->use_sg = old_use_sg;
724         scmd->result = old_result;
725         return rtn;
726 }
727
728 /**
729  * scsi_request_sense - Request sense data from a particular target.
730  * @scmd:       SCSI cmd for request sense.
731  *
732  * Notes:
733  *    Some hosts automatically obtain this information, others require
734  *    that we obtain it on our own. This function will *not* return until
735  *    the command either times out, or it completes.
736  **/
737 static int scsi_request_sense(struct scsi_cmnd *scmd)
738 {
739         static unsigned char generic_sense[6] =
740                 {REQUEST_SENSE, 0, 0, 0, 252, 0};
741
742         return scsi_send_eh_cmnd(scmd, generic_sense, 6, SENSE_TIMEOUT, 1);
743 }
744
745 /**
746  * scsi_eh_finish_cmd - Handle a cmd that eh is finished with.
747  * @scmd:       Original SCSI cmd that eh has finished.
748  * @done_q:     Queue for processed commands.
749  *
750  * Notes:
751  *    We don't want to use the normal command completion while we are are
752  *    still handling errors - it may cause other commands to be queued,
753  *    and that would disturb what we are doing.  thus we really want to
754  *    keep a list of pending commands for final completion, and once we
755  *    are ready to leave error handling we handle completion for real.
756  **/
757 void scsi_eh_finish_cmd(struct scsi_cmnd *scmd, struct list_head *done_q)
758 {
759         scmd->device->host->host_failed--;
760         scmd->eh_eflags = 0;
761         list_move_tail(&scmd->eh_entry, done_q);
762 }
763 EXPORT_SYMBOL(scsi_eh_finish_cmd);
764
765 /**
766  * scsi_eh_get_sense - Get device sense data.
767  * @work_q:     Queue of commands to process.
768  * @done_q:     Queue of proccessed commands..
769  *
770  * Description:
771  *    See if we need to request sense information.  if so, then get it
772  *    now, so we have a better idea of what to do.  
773  *
774  * Notes:
775  *    This has the unfortunate side effect that if a shost adapter does
776  *    not automatically request sense information, that we end up shutting
777  *    it down before we request it.
778  *
779  *    All drivers should request sense information internally these days,
780  *    so for now all I have to say is tough noogies if you end up in here.
781  *
782  *    XXX: Long term this code should go away, but that needs an audit of
783  *         all LLDDs first.
784  **/
785 int scsi_eh_get_sense(struct list_head *work_q,
786                       struct list_head *done_q)
787 {
788         struct scsi_cmnd *scmd, *next;
789         int rtn;
790
791         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
792                 if ((scmd->eh_eflags & SCSI_EH_CANCEL_CMD) ||
793                     SCSI_SENSE_VALID(scmd))
794                         continue;
795
796                 SCSI_LOG_ERROR_RECOVERY(2, scmd_printk(KERN_INFO, scmd,
797                                                   "%s: requesting sense\n",
798                                                   current->comm));
799                 rtn = scsi_request_sense(scmd);
800                 if (rtn != SUCCESS)
801                         continue;
802
803                 SCSI_LOG_ERROR_RECOVERY(3, printk("sense requested for %p"
804                                                   " result %x\n", scmd,
805                                                   scmd->result));
806                 SCSI_LOG_ERROR_RECOVERY(3, scsi_print_sense("bh", scmd));
807
808                 rtn = scsi_decide_disposition(scmd);
809
810                 /*
811                  * if the result was normal, then just pass it along to the
812                  * upper level.
813                  */
814                 if (rtn == SUCCESS)
815                         /* we don't want this command reissued, just
816                          * finished with the sense data, so set
817                          * retries to the max allowed to ensure it
818                          * won't get reissued */
819                         scmd->retries = scmd->allowed;
820                 else if (rtn != NEEDS_RETRY)
821                         continue;
822
823                 scsi_eh_finish_cmd(scmd, done_q);
824         }
825
826         return list_empty(work_q);
827 }
828 EXPORT_SYMBOL_GPL(scsi_eh_get_sense);
829
830 /**
831  * scsi_eh_tur - Send TUR to device.
832  * @scmd:       Scsi cmd to send TUR
833  *
834  * Return value:
835  *    0 - Device is ready. 1 - Device NOT ready.
836  **/
837 static int scsi_eh_tur(struct scsi_cmnd *scmd)
838 {
839         static unsigned char tur_command[6] = {TEST_UNIT_READY, 0, 0, 0, 0, 0};
840         int retry_cnt = 1, rtn;
841
842 retry_tur:
843         rtn = scsi_send_eh_cmnd(scmd, tur_command, 6, SENSE_TIMEOUT, 0);
844
845         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: scmd %p rtn %x\n",
846                 __FUNCTION__, scmd, rtn));
847
848         switch (rtn) {
849         case NEEDS_RETRY:
850                 if (retry_cnt--)
851                         goto retry_tur;
852                 /*FALLTHRU*/
853         case SUCCESS:
854                 return 0;
855         default:
856                 return 1;
857         }
858 }
859
860 /**
861  * scsi_eh_abort_cmds - abort canceled commands.
862  * @shost:      scsi host being recovered.
863  * @eh_done_q:  list_head for processed commands.
864  *
865  * Decription:
866  *    Try and see whether or not it makes sense to try and abort the
867  *    running command.  this only works out to be the case if we have one
868  *    command that has timed out.  if the command simply failed, it makes
869  *    no sense to try and abort the command, since as far as the shost
870  *    adapter is concerned, it isn't running.
871  **/
872 static int scsi_eh_abort_cmds(struct list_head *work_q,
873                               struct list_head *done_q)
874 {
875         struct scsi_cmnd *scmd, *next;
876         int rtn;
877
878         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
879                 if (!(scmd->eh_eflags & SCSI_EH_CANCEL_CMD))
880                         continue;
881                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting cmd:"
882                                                   "0x%p\n", current->comm,
883                                                   scmd));
884                 rtn = scsi_try_to_abort_cmd(scmd);
885                 if (rtn == SUCCESS) {
886                         scmd->eh_eflags &= ~SCSI_EH_CANCEL_CMD;
887                         if (!scsi_device_online(scmd->device) ||
888                             !scsi_eh_tur(scmd)) {
889                                 scsi_eh_finish_cmd(scmd, done_q);
890                         }
891                                 
892                 } else
893                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: aborting"
894                                                           " cmd failed:"
895                                                           "0x%p\n",
896                                                           current->comm,
897                                                           scmd));
898         }
899
900         return list_empty(work_q);
901 }
902
903 /**
904  * scsi_eh_try_stu - Send START_UNIT to device.
905  * @scmd:       Scsi cmd to send START_UNIT
906  *
907  * Return value:
908  *    0 - Device is ready. 1 - Device NOT ready.
909  **/
910 static int scsi_eh_try_stu(struct scsi_cmnd *scmd)
911 {
912         static unsigned char stu_command[6] = {START_STOP, 0, 0, 0, 1, 0};
913
914         if (scmd->device->allow_restart) {
915                 int i, rtn = NEEDS_RETRY;
916
917                 for (i = 0; rtn == NEEDS_RETRY && i < 2; i++)
918                         rtn = scsi_send_eh_cmnd(scmd, stu_command, 6,
919                                                 scmd->device->timeout, 0);
920
921                 if (rtn == SUCCESS)
922                         return 0;
923         }
924
925         return 1;
926 }
927
928  /**
929  * scsi_eh_stu - send START_UNIT if needed
930  * @shost:      scsi host being recovered.
931  * @eh_done_q:  list_head for processed commands.
932  *
933  * Notes:
934  *    If commands are failing due to not ready, initializing command required,
935  *      try revalidating the device, which will end up sending a start unit. 
936  **/
937 static int scsi_eh_stu(struct Scsi_Host *shost,
938                               struct list_head *work_q,
939                               struct list_head *done_q)
940 {
941         struct scsi_cmnd *scmd, *stu_scmd, *next;
942         struct scsi_device *sdev;
943
944         shost_for_each_device(sdev, shost) {
945                 stu_scmd = NULL;
946                 list_for_each_entry(scmd, work_q, eh_entry)
947                         if (scmd->device == sdev && SCSI_SENSE_VALID(scmd) &&
948                             scsi_check_sense(scmd) == FAILED ) {
949                                 stu_scmd = scmd;
950                                 break;
951                         }
952
953                 if (!stu_scmd)
954                         continue;
955
956                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending START_UNIT to sdev:"
957                                                   " 0x%p\n", current->comm, sdev));
958
959                 if (!scsi_eh_try_stu(stu_scmd)) {
960                         if (!scsi_device_online(sdev) ||
961                             !scsi_eh_tur(stu_scmd)) {
962                                 list_for_each_entry_safe(scmd, next,
963                                                           work_q, eh_entry) {
964                                         if (scmd->device == sdev)
965                                                 scsi_eh_finish_cmd(scmd, done_q);
966                                 }
967                         }
968                 } else {
969                         SCSI_LOG_ERROR_RECOVERY(3,
970                                                 printk("%s: START_UNIT failed to sdev:"
971                                                        " 0x%p\n", current->comm, sdev));
972                 }
973         }
974
975         return list_empty(work_q);
976 }
977
978
979 /**
980  * scsi_eh_bus_device_reset - send bdr if needed
981  * @shost:      scsi host being recovered.
982  * @eh_done_q:  list_head for processed commands.
983  *
984  * Notes:
985  *    Try a bus device reset.  still, look to see whether we have multiple
986  *    devices that are jammed or not - if we have multiple devices, it
987  *    makes no sense to try bus_device_reset - we really would need to try
988  *    a bus_reset instead. 
989  **/
990 static int scsi_eh_bus_device_reset(struct Scsi_Host *shost,
991                                     struct list_head *work_q,
992                                     struct list_head *done_q)
993 {
994         struct scsi_cmnd *scmd, *bdr_scmd, *next;
995         struct scsi_device *sdev;
996         int rtn;
997
998         shost_for_each_device(sdev, shost) {
999                 bdr_scmd = NULL;
1000                 list_for_each_entry(scmd, work_q, eh_entry)
1001                         if (scmd->device == sdev) {
1002                                 bdr_scmd = scmd;
1003                                 break;
1004                         }
1005
1006                 if (!bdr_scmd)
1007                         continue;
1008
1009                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BDR sdev:"
1010                                                   " 0x%p\n", current->comm,
1011                                                   sdev));
1012                 rtn = scsi_try_bus_device_reset(bdr_scmd);
1013                 if (rtn == SUCCESS) {
1014                         if (!scsi_device_online(sdev) ||
1015                             !scsi_eh_tur(bdr_scmd)) {
1016                                 list_for_each_entry_safe(scmd, next,
1017                                                          work_q, eh_entry) {
1018                                         if (scmd->device == sdev)
1019                                                 scsi_eh_finish_cmd(scmd,
1020                                                                    done_q);
1021                                 }
1022                         }
1023                 } else {
1024                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BDR"
1025                                                           " failed sdev:"
1026                                                           "0x%p\n",
1027                                                           current->comm,
1028                                                            sdev));
1029                 }
1030         }
1031
1032         return list_empty(work_q);
1033 }
1034
1035 /**
1036  * scsi_eh_bus_reset - send a bus reset 
1037  * @shost:      scsi host being recovered.
1038  * @eh_done_q:  list_head for processed commands.
1039  **/
1040 static int scsi_eh_bus_reset(struct Scsi_Host *shost,
1041                              struct list_head *work_q,
1042                              struct list_head *done_q)
1043 {
1044         struct scsi_cmnd *scmd, *chan_scmd, *next;
1045         unsigned int channel;
1046         int rtn;
1047
1048         /*
1049          * we really want to loop over the various channels, and do this on
1050          * a channel by channel basis.  we should also check to see if any
1051          * of the failed commands are on soft_reset devices, and if so, skip
1052          * the reset.  
1053          */
1054
1055         for (channel = 0; channel <= shost->max_channel; channel++) {
1056                 chan_scmd = NULL;
1057                 list_for_each_entry(scmd, work_q, eh_entry) {
1058                         if (channel == scmd_channel(scmd)) {
1059                                 chan_scmd = scmd;
1060                                 break;
1061                                 /*
1062                                  * FIXME add back in some support for
1063                                  * soft_reset devices.
1064                                  */
1065                         }
1066                 }
1067
1068                 if (!chan_scmd)
1069                         continue;
1070                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending BRST chan:"
1071                                                   " %d\n", current->comm,
1072                                                   channel));
1073                 rtn = scsi_try_bus_reset(chan_scmd);
1074                 if (rtn == SUCCESS) {
1075                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1076                                 if (channel == scmd_channel(scmd))
1077                                         if (!scsi_device_online(scmd->device) ||
1078                                             !scsi_eh_tur(scmd))
1079                                                 scsi_eh_finish_cmd(scmd,
1080                                                                    done_q);
1081                         }
1082                 } else {
1083                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: BRST"
1084                                                           " failed chan: %d\n",
1085                                                           current->comm,
1086                                                           channel));
1087                 }
1088         }
1089         return list_empty(work_q);
1090 }
1091
1092 /**
1093  * scsi_eh_host_reset - send a host reset 
1094  * @work_q:     list_head for processed commands.
1095  * @done_q:     list_head for processed commands.
1096  **/
1097 static int scsi_eh_host_reset(struct list_head *work_q,
1098                               struct list_head *done_q)
1099 {
1100         struct scsi_cmnd *scmd, *next;
1101         int rtn;
1102
1103         if (!list_empty(work_q)) {
1104                 scmd = list_entry(work_q->next,
1105                                   struct scsi_cmnd, eh_entry);
1106
1107                 SCSI_LOG_ERROR_RECOVERY(3, printk("%s: Sending HRST\n"
1108                                                   , current->comm));
1109
1110                 rtn = scsi_try_host_reset(scmd);
1111                 if (rtn == SUCCESS) {
1112                         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1113                                 if (!scsi_device_online(scmd->device) ||
1114                                     (!scsi_eh_try_stu(scmd) && !scsi_eh_tur(scmd)) ||
1115                                     !scsi_eh_tur(scmd))
1116                                         scsi_eh_finish_cmd(scmd, done_q);
1117                         }
1118                 } else {
1119                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: HRST"
1120                                                           " failed\n",
1121                                                           current->comm));
1122                 }
1123         }
1124         return list_empty(work_q);
1125 }
1126
1127 /**
1128  * scsi_eh_offline_sdevs - offline scsi devices that fail to recover
1129  * @work_q:     list_head for processed commands.
1130  * @done_q:     list_head for processed commands.
1131  *
1132  **/
1133 static void scsi_eh_offline_sdevs(struct list_head *work_q,
1134                                   struct list_head *done_q)
1135 {
1136         struct scsi_cmnd *scmd, *next;
1137
1138         list_for_each_entry_safe(scmd, next, work_q, eh_entry) {
1139                 sdev_printk(KERN_INFO, scmd->device,
1140                             "scsi: Device offlined - not"
1141                             " ready after error recovery\n");
1142                 scsi_device_set_state(scmd->device, SDEV_OFFLINE);
1143                 if (scmd->eh_eflags & SCSI_EH_CANCEL_CMD) {
1144                         /*
1145                          * FIXME: Handle lost cmds.
1146                          */
1147                 }
1148                 scsi_eh_finish_cmd(scmd, done_q);
1149         }
1150         return;
1151 }
1152
1153 /**
1154  * scsi_decide_disposition - Disposition a cmd on return from LLD.
1155  * @scmd:       SCSI cmd to examine.
1156  *
1157  * Notes:
1158  *    This is *only* called when we are examining the status after sending
1159  *    out the actual data command.  any commands that are queued for error
1160  *    recovery (e.g. test_unit_ready) do *not* come through here.
1161  *
1162  *    When this routine returns failed, it means the error handler thread
1163  *    is woken.  In cases where the error code indicates an error that
1164  *    doesn't require the error handler read (i.e. we don't need to
1165  *    abort/reset), this function should return SUCCESS.
1166  **/
1167 int scsi_decide_disposition(struct scsi_cmnd *scmd)
1168 {
1169         int rtn;
1170
1171         /*
1172          * if the device is offline, then we clearly just pass the result back
1173          * up to the top level.
1174          */
1175         if (!scsi_device_online(scmd->device)) {
1176                 SCSI_LOG_ERROR_RECOVERY(5, printk("%s: device offline - report"
1177                                                   " as SUCCESS\n",
1178                                                   __FUNCTION__));
1179                 return SUCCESS;
1180         }
1181
1182         /*
1183          * first check the host byte, to see if there is anything in there
1184          * that would indicate what we need to do.
1185          */
1186         switch (host_byte(scmd->result)) {
1187         case DID_PASSTHROUGH:
1188                 /*
1189                  * no matter what, pass this through to the upper layer.
1190                  * nuke this special code so that it looks like we are saying
1191                  * did_ok.
1192                  */
1193                 scmd->result &= 0xff00ffff;
1194                 return SUCCESS;
1195         case DID_OK:
1196                 /*
1197                  * looks good.  drop through, and check the next byte.
1198                  */
1199                 break;
1200         case DID_NO_CONNECT:
1201         case DID_BAD_TARGET:
1202         case DID_ABORT:
1203                 /*
1204                  * note - this means that we just report the status back
1205                  * to the top level driver, not that we actually think
1206                  * that it indicates SUCCESS.
1207                  */
1208                 return SUCCESS;
1209                 /*
1210                  * when the low level driver returns did_soft_error,
1211                  * it is responsible for keeping an internal retry counter 
1212                  * in order to avoid endless loops (db)
1213                  *
1214                  * actually this is a bug in this function here.  we should
1215                  * be mindful of the maximum number of retries specified
1216                  * and not get stuck in a loop.
1217                  */
1218         case DID_SOFT_ERROR:
1219                 goto maybe_retry;
1220         case DID_IMM_RETRY:
1221                 return NEEDS_RETRY;
1222
1223         case DID_REQUEUE:
1224                 return ADD_TO_MLQUEUE;
1225
1226         case DID_ERROR:
1227                 if (msg_byte(scmd->result) == COMMAND_COMPLETE &&
1228                     status_byte(scmd->result) == RESERVATION_CONFLICT)
1229                         /*
1230                          * execute reservation conflict processing code
1231                          * lower down
1232                          */
1233                         break;
1234                 /* fallthrough */
1235
1236         case DID_BUS_BUSY:
1237         case DID_PARITY:
1238                 goto maybe_retry;
1239         case DID_TIME_OUT:
1240                 /*
1241                  * when we scan the bus, we get timeout messages for
1242                  * these commands if there is no device available.
1243                  * other hosts report did_no_connect for the same thing.
1244                  */
1245                 if ((scmd->cmnd[0] == TEST_UNIT_READY ||
1246                      scmd->cmnd[0] == INQUIRY)) {
1247                         return SUCCESS;
1248                 } else {
1249                         return FAILED;
1250                 }
1251         case DID_RESET:
1252                 return SUCCESS;
1253         default:
1254                 return FAILED;
1255         }
1256
1257         /*
1258          * next, check the message byte.
1259          */
1260         if (msg_byte(scmd->result) != COMMAND_COMPLETE)
1261                 return FAILED;
1262
1263         /*
1264          * check the status byte to see if this indicates anything special.
1265          */
1266         switch (status_byte(scmd->result)) {
1267         case QUEUE_FULL:
1268                 /*
1269                  * the case of trying to send too many commands to a
1270                  * tagged queueing device.
1271                  */
1272         case BUSY:
1273                 /*
1274                  * device can't talk to us at the moment.  Should only
1275                  * occur (SAM-3) when the task queue is empty, so will cause
1276                  * the empty queue handling to trigger a stall in the
1277                  * device.
1278                  */
1279                 return ADD_TO_MLQUEUE;
1280         case GOOD:
1281         case COMMAND_TERMINATED:
1282         case TASK_ABORTED:
1283                 return SUCCESS;
1284         case CHECK_CONDITION:
1285                 rtn = scsi_check_sense(scmd);
1286                 if (rtn == NEEDS_RETRY)
1287                         goto maybe_retry;
1288                 /* if rtn == FAILED, we have no sense information;
1289                  * returning FAILED will wake the error handler thread
1290                  * to collect the sense and redo the decide
1291                  * disposition */
1292                 return rtn;
1293         case CONDITION_GOOD:
1294         case INTERMEDIATE_GOOD:
1295         case INTERMEDIATE_C_GOOD:
1296         case ACA_ACTIVE:
1297                 /*
1298                  * who knows?  FIXME(eric)
1299                  */
1300                 return SUCCESS;
1301
1302         case RESERVATION_CONFLICT:
1303                 sdev_printk(KERN_INFO, scmd->device,
1304                             "reservation conflict\n");
1305                 return SUCCESS; /* causes immediate i/o error */
1306         default:
1307                 return FAILED;
1308         }
1309         return FAILED;
1310
1311       maybe_retry:
1312
1313         /* we requeue for retry because the error was retryable, and
1314          * the request was not marked fast fail.  Note that above,
1315          * even if the request is marked fast fail, we still requeue
1316          * for queue congestion conditions (QUEUE_FULL or BUSY) */
1317         if ((++scmd->retries) <= scmd->allowed
1318             && !blk_noretry_request(scmd->request)) {
1319                 return NEEDS_RETRY;
1320         } else {
1321                 /*
1322                  * no more retries - report this one back to upper level.
1323                  */
1324                 return SUCCESS;
1325         }
1326 }
1327
1328 /**
1329  * scsi_eh_lock_door - Prevent medium removal for the specified device
1330  * @sdev:       SCSI device to prevent medium removal
1331  *
1332  * Locking:
1333  *      We must be called from process context; scsi_allocate_request()
1334  *      may sleep.
1335  *
1336  * Notes:
1337  *      We queue up an asynchronous "ALLOW MEDIUM REMOVAL" request on the
1338  *      head of the devices request queue, and continue.
1339  *
1340  * Bugs:
1341  *      scsi_allocate_request() may sleep waiting for existing requests to
1342  *      be processed.  However, since we haven't kicked off any request
1343  *      processing for this host, this may deadlock.
1344  *
1345  *      If scsi_allocate_request() fails for what ever reason, we
1346  *      completely forget to lock the door.
1347  **/
1348 static void scsi_eh_lock_door(struct scsi_device *sdev)
1349 {
1350         unsigned char cmnd[MAX_COMMAND_SIZE];
1351
1352         cmnd[0] = ALLOW_MEDIUM_REMOVAL;
1353         cmnd[1] = 0;
1354         cmnd[2] = 0;
1355         cmnd[3] = 0;
1356         cmnd[4] = SCSI_REMOVAL_PREVENT;
1357         cmnd[5] = 0;
1358
1359         scsi_execute_async(sdev, cmnd, 6, DMA_NONE, NULL, 0, 0, 10 * HZ,
1360                            5, NULL, NULL, GFP_KERNEL);
1361 }
1362
1363
1364 /**
1365  * scsi_restart_operations - restart io operations to the specified host.
1366  * @shost:      Host we are restarting.
1367  *
1368  * Notes:
1369  *    When we entered the error handler, we blocked all further i/o to
1370  *    this device.  we need to 'reverse' this process.
1371  **/
1372 static void scsi_restart_operations(struct Scsi_Host *shost)
1373 {
1374         struct scsi_device *sdev;
1375         unsigned long flags;
1376
1377         /*
1378          * If the door was locked, we need to insert a door lock request
1379          * onto the head of the SCSI request queue for the device.  There
1380          * is no point trying to lock the door of an off-line device.
1381          */
1382         shost_for_each_device(sdev, shost) {
1383                 if (scsi_device_online(sdev) && sdev->locked)
1384                         scsi_eh_lock_door(sdev);
1385         }
1386
1387         /*
1388          * next free up anything directly waiting upon the host.  this
1389          * will be requests for character device operations, and also for
1390          * ioctls to queued block devices.
1391          */
1392         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: waking up host to restart\n",
1393                                           __FUNCTION__));
1394
1395         spin_lock_irqsave(shost->host_lock, flags);
1396         if (scsi_host_set_state(shost, SHOST_RUNNING))
1397                 if (scsi_host_set_state(shost, SHOST_CANCEL))
1398                         BUG_ON(scsi_host_set_state(shost, SHOST_DEL));
1399         spin_unlock_irqrestore(shost->host_lock, flags);
1400
1401         wake_up(&shost->host_wait);
1402
1403         /*
1404          * finally we need to re-initiate requests that may be pending.  we will
1405          * have had everything blocked while error handling is taking place, and
1406          * now that error recovery is done, we will need to ensure that these
1407          * requests are started.
1408          */
1409         scsi_run_host_queues(shost);
1410 }
1411
1412 /**
1413  * scsi_eh_ready_devs - check device ready state and recover if not.
1414  * @shost:      host to be recovered.
1415  * @eh_done_q:  list_head for processed commands.
1416  *
1417  **/
1418 void scsi_eh_ready_devs(struct Scsi_Host *shost,
1419                         struct list_head *work_q,
1420                         struct list_head *done_q)
1421 {
1422         if (!scsi_eh_stu(shost, work_q, done_q))
1423                 if (!scsi_eh_bus_device_reset(shost, work_q, done_q))
1424                         if (!scsi_eh_bus_reset(shost, work_q, done_q))
1425                                 if (!scsi_eh_host_reset(work_q, done_q))
1426                                         scsi_eh_offline_sdevs(work_q, done_q);
1427 }
1428 EXPORT_SYMBOL_GPL(scsi_eh_ready_devs);
1429
1430 /**
1431  * scsi_eh_flush_done_q - finish processed commands or retry them.
1432  * @done_q:     list_head of processed commands.
1433  *
1434  **/
1435 void scsi_eh_flush_done_q(struct list_head *done_q)
1436 {
1437         struct scsi_cmnd *scmd, *next;
1438
1439         list_for_each_entry_safe(scmd, next, done_q, eh_entry) {
1440                 list_del_init(&scmd->eh_entry);
1441                 if (scsi_device_online(scmd->device) &&
1442                     !blk_noretry_request(scmd->request) &&
1443                     (++scmd->retries <= scmd->allowed)) {
1444                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush"
1445                                                           " retry cmd: %p\n",
1446                                                           current->comm,
1447                                                           scmd));
1448                                 scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY);
1449                 } else {
1450                         /*
1451                          * If just we got sense for the device (called
1452                          * scsi_eh_get_sense), scmd->result is already
1453                          * set, do not set DRIVER_TIMEOUT.
1454                          */
1455                         if (!scmd->result)
1456                                 scmd->result |= (DRIVER_TIMEOUT << 24);
1457                         SCSI_LOG_ERROR_RECOVERY(3, printk("%s: flush finish"
1458                                                         " cmd: %p\n",
1459                                                         current->comm, scmd));
1460                         scsi_finish_command(scmd);
1461                 }
1462         }
1463 }
1464 EXPORT_SYMBOL(scsi_eh_flush_done_q);
1465
1466 /**
1467  * scsi_unjam_host - Attempt to fix a host which has a cmd that failed.
1468  * @shost:      Host to unjam.
1469  *
1470  * Notes:
1471  *    When we come in here, we *know* that all commands on the bus have
1472  *    either completed, failed or timed out.  we also know that no further
1473  *    commands are being sent to the host, so things are relatively quiet
1474  *    and we have freedom to fiddle with things as we wish.
1475  *
1476  *    This is only the *default* implementation.  it is possible for
1477  *    individual drivers to supply their own version of this function, and
1478  *    if the maintainer wishes to do this, it is strongly suggested that
1479  *    this function be taken as a template and modified.  this function
1480  *    was designed to correctly handle problems for about 95% of the
1481  *    different cases out there, and it should always provide at least a
1482  *    reasonable amount of error recovery.
1483  *
1484  *    Any command marked 'failed' or 'timeout' must eventually have
1485  *    scsi_finish_cmd() called for it.  we do all of the retry stuff
1486  *    here, so when we restart the host after we return it should have an
1487  *    empty queue.
1488  **/
1489 static void scsi_unjam_host(struct Scsi_Host *shost)
1490 {
1491         unsigned long flags;
1492         LIST_HEAD(eh_work_q);
1493         LIST_HEAD(eh_done_q);
1494
1495         spin_lock_irqsave(shost->host_lock, flags);
1496         list_splice_init(&shost->eh_cmd_q, &eh_work_q);
1497         spin_unlock_irqrestore(shost->host_lock, flags);
1498
1499         SCSI_LOG_ERROR_RECOVERY(1, scsi_eh_prt_fail_stats(shost, &eh_work_q));
1500
1501         if (!scsi_eh_get_sense(&eh_work_q, &eh_done_q))
1502                 if (!scsi_eh_abort_cmds(&eh_work_q, &eh_done_q))
1503                         scsi_eh_ready_devs(shost, &eh_work_q, &eh_done_q);
1504
1505         scsi_eh_flush_done_q(&eh_done_q);
1506 }
1507
1508 /**
1509  * scsi_error_handler - SCSI error handler thread
1510  * @data:       Host for which we are running.
1511  *
1512  * Notes:
1513  *    This is the main error handling loop.  This is run as a kernel thread
1514  *    for every SCSI host and handles all error handling activity.
1515  **/
1516 int scsi_error_handler(void *data)
1517 {
1518         struct Scsi_Host *shost = data;
1519
1520         /*
1521          * We use TASK_INTERRUPTIBLE so that the thread is not
1522          * counted against the load average as a running process.
1523          * We never actually get interrupted because kthread_run
1524          * disables singal delivery for the created thread.
1525          */
1526         set_current_state(TASK_INTERRUPTIBLE);
1527         while (!kthread_should_stop()) {
1528                 if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
1529                     shost->host_failed != shost->host_busy) {
1530                         SCSI_LOG_ERROR_RECOVERY(1,
1531                                 printk("Error handler scsi_eh_%d sleeping\n",
1532                                         shost->host_no));
1533                         schedule();
1534                         set_current_state(TASK_INTERRUPTIBLE);
1535                         continue;
1536                 }
1537
1538                 __set_current_state(TASK_RUNNING);
1539                 SCSI_LOG_ERROR_RECOVERY(1,
1540                         printk("Error handler scsi_eh_%d waking up\n",
1541                                 shost->host_no));
1542
1543                 /*
1544                  * We have a host that is failing for some reason.  Figure out
1545                  * what we need to do to get it up and online again (if we can).
1546                  * If we fail, we end up taking the thing offline.
1547                  */
1548                 if (shost->transportt->eh_strategy_handler)
1549                         shost->transportt->eh_strategy_handler(shost);
1550                 else
1551                         scsi_unjam_host(shost);
1552
1553                 /*
1554                  * Note - if the above fails completely, the action is to take
1555                  * individual devices offline and flush the queue of any
1556                  * outstanding requests that may have been pending.  When we
1557                  * restart, we restart any I/O to any other devices on the bus
1558                  * which are still online.
1559                  */
1560                 scsi_restart_operations(shost);
1561                 set_current_state(TASK_INTERRUPTIBLE);
1562         }
1563         __set_current_state(TASK_RUNNING);
1564
1565         SCSI_LOG_ERROR_RECOVERY(1,
1566                 printk("Error handler scsi_eh_%d exiting\n", shost->host_no));
1567         shost->ehandler = NULL;
1568         return 0;
1569 }
1570
1571 /*
1572  * Function:    scsi_report_bus_reset()
1573  *
1574  * Purpose:     Utility function used by low-level drivers to report that
1575  *              they have observed a bus reset on the bus being handled.
1576  *
1577  * Arguments:   shost       - Host in question
1578  *              channel     - channel on which reset was observed.
1579  *
1580  * Returns:     Nothing
1581  *
1582  * Lock status: Host lock must be held.
1583  *
1584  * Notes:       This only needs to be called if the reset is one which
1585  *              originates from an unknown location.  Resets originated
1586  *              by the mid-level itself don't need to call this, but there
1587  *              should be no harm.
1588  *
1589  *              The main purpose of this is to make sure that a CHECK_CONDITION
1590  *              is properly treated.
1591  */
1592 void scsi_report_bus_reset(struct Scsi_Host *shost, int channel)
1593 {
1594         struct scsi_device *sdev;
1595
1596         __shost_for_each_device(sdev, shost) {
1597                 if (channel == sdev_channel(sdev)) {
1598                         sdev->was_reset = 1;
1599                         sdev->expecting_cc_ua = 1;
1600                 }
1601         }
1602 }
1603 EXPORT_SYMBOL(scsi_report_bus_reset);
1604
1605 /*
1606  * Function:    scsi_report_device_reset()
1607  *
1608  * Purpose:     Utility function used by low-level drivers to report that
1609  *              they have observed a device reset on the device being handled.
1610  *
1611  * Arguments:   shost       - Host in question
1612  *              channel     - channel on which reset was observed
1613  *              target      - target on which reset was observed
1614  *
1615  * Returns:     Nothing
1616  *
1617  * Lock status: Host lock must be held
1618  *
1619  * Notes:       This only needs to be called if the reset is one which
1620  *              originates from an unknown location.  Resets originated
1621  *              by the mid-level itself don't need to call this, but there
1622  *              should be no harm.
1623  *
1624  *              The main purpose of this is to make sure that a CHECK_CONDITION
1625  *              is properly treated.
1626  */
1627 void scsi_report_device_reset(struct Scsi_Host *shost, int channel, int target)
1628 {
1629         struct scsi_device *sdev;
1630
1631         __shost_for_each_device(sdev, shost) {
1632                 if (channel == sdev_channel(sdev) &&
1633                     target == sdev_id(sdev)) {
1634                         sdev->was_reset = 1;
1635                         sdev->expecting_cc_ua = 1;
1636                 }
1637         }
1638 }
1639 EXPORT_SYMBOL(scsi_report_device_reset);
1640
1641 static void
1642 scsi_reset_provider_done_command(struct scsi_cmnd *scmd)
1643 {
1644 }
1645
1646 /*
1647  * Function:    scsi_reset_provider
1648  *
1649  * Purpose:     Send requested reset to a bus or device at any phase.
1650  *
1651  * Arguments:   device  - device to send reset to
1652  *              flag - reset type (see scsi.h)
1653  *
1654  * Returns:     SUCCESS/FAILURE.
1655  *
1656  * Notes:       This is used by the SCSI Generic driver to provide
1657  *              Bus/Device reset capability.
1658  */
1659 int
1660 scsi_reset_provider(struct scsi_device *dev, int flag)
1661 {
1662         struct scsi_cmnd *scmd = scsi_get_command(dev, GFP_KERNEL);
1663         struct Scsi_Host *shost = dev->host;
1664         struct request req;
1665         unsigned long flags;
1666         int rtn;
1667
1668         scmd->request = &req;
1669         memset(&scmd->eh_timeout, 0, sizeof(scmd->eh_timeout));
1670
1671         memset(&scmd->cmnd, '\0', sizeof(scmd->cmnd));
1672     
1673         scmd->scsi_done         = scsi_reset_provider_done_command;
1674         scmd->done                      = NULL;
1675         scmd->request_buffer            = NULL;
1676         scmd->request_bufflen           = 0;
1677
1678         scmd->cmd_len                   = 0;
1679
1680         scmd->sc_data_direction         = DMA_BIDIRECTIONAL;
1681
1682         init_timer(&scmd->eh_timeout);
1683
1684         /*
1685          * Sometimes the command can get back into the timer chain,
1686          * so use the pid as an identifier.
1687          */
1688         scmd->pid                       = 0;
1689
1690         spin_lock_irqsave(shost->host_lock, flags);
1691         shost->tmf_in_progress = 1;
1692         spin_unlock_irqrestore(shost->host_lock, flags);
1693
1694         switch (flag) {
1695         case SCSI_TRY_RESET_DEVICE:
1696                 rtn = scsi_try_bus_device_reset(scmd);
1697                 if (rtn == SUCCESS)
1698                         break;
1699                 /* FALLTHROUGH */
1700         case SCSI_TRY_RESET_BUS:
1701                 rtn = scsi_try_bus_reset(scmd);
1702                 if (rtn == SUCCESS)
1703                         break;
1704                 /* FALLTHROUGH */
1705         case SCSI_TRY_RESET_HOST:
1706                 rtn = scsi_try_host_reset(scmd);
1707                 break;
1708         default:
1709                 rtn = FAILED;
1710         }
1711
1712         spin_lock_irqsave(shost->host_lock, flags);
1713         shost->tmf_in_progress = 0;
1714         spin_unlock_irqrestore(shost->host_lock, flags);
1715
1716         /*
1717          * be sure to wake up anyone who was sleeping or had their queue
1718          * suspended while we performed the TMF.
1719          */
1720         SCSI_LOG_ERROR_RECOVERY(3,
1721                 printk("%s: waking up host to restart after TMF\n",
1722                 __FUNCTION__));
1723
1724         wake_up(&shost->host_wait);
1725
1726         scsi_run_host_queues(shost);
1727
1728         scsi_next_command(scmd);
1729         return rtn;
1730 }
1731 EXPORT_SYMBOL(scsi_reset_provider);
1732
1733 /**
1734  * scsi_normalize_sense - normalize main elements from either fixed or
1735  *                      descriptor sense data format into a common format.
1736  *
1737  * @sense_buffer:       byte array containing sense data returned by device
1738  * @sb_len:             number of valid bytes in sense_buffer
1739  * @sshdr:              pointer to instance of structure that common
1740  *                      elements are written to.
1741  *
1742  * Notes:
1743  *      The "main elements" from sense data are: response_code, sense_key,
1744  *      asc, ascq and additional_length (only for descriptor format).
1745  *
1746  *      Typically this function can be called after a device has
1747  *      responded to a SCSI command with the CHECK_CONDITION status.
1748  *
1749  * Return value:
1750  *      1 if valid sense data information found, else 0;
1751  **/
1752 int scsi_normalize_sense(const u8 *sense_buffer, int sb_len,
1753                          struct scsi_sense_hdr *sshdr)
1754 {
1755         if (!sense_buffer || !sb_len)
1756                 return 0;
1757
1758         memset(sshdr, 0, sizeof(struct scsi_sense_hdr));
1759
1760         sshdr->response_code = (sense_buffer[0] & 0x7f);
1761
1762         if (!scsi_sense_valid(sshdr))
1763                 return 0;
1764
1765         if (sshdr->response_code >= 0x72) {
1766                 /*
1767                  * descriptor format
1768                  */
1769                 if (sb_len > 1)
1770                         sshdr->sense_key = (sense_buffer[1] & 0xf);
1771                 if (sb_len > 2)
1772                         sshdr->asc = sense_buffer[2];
1773                 if (sb_len > 3)
1774                         sshdr->ascq = sense_buffer[3];
1775                 if (sb_len > 7)
1776                         sshdr->additional_length = sense_buffer[7];
1777         } else {
1778                 /* 
1779                  * fixed format
1780                  */
1781                 if (sb_len > 2)
1782                         sshdr->sense_key = (sense_buffer[2] & 0xf);
1783                 if (sb_len > 7) {
1784                         sb_len = (sb_len < (sense_buffer[7] + 8)) ?
1785                                          sb_len : (sense_buffer[7] + 8);
1786                         if (sb_len > 12)
1787                                 sshdr->asc = sense_buffer[12];
1788                         if (sb_len > 13)
1789                                 sshdr->ascq = sense_buffer[13];
1790                 }
1791         }
1792
1793         return 1;
1794 }
1795 EXPORT_SYMBOL(scsi_normalize_sense);
1796
1797 int scsi_command_normalize_sense(struct scsi_cmnd *cmd,
1798                                  struct scsi_sense_hdr *sshdr)
1799 {
1800         return scsi_normalize_sense(cmd->sense_buffer,
1801                         sizeof(cmd->sense_buffer), sshdr);
1802 }
1803 EXPORT_SYMBOL(scsi_command_normalize_sense);
1804
1805 /**
1806  * scsi_sense_desc_find - search for a given descriptor type in
1807  *                      descriptor sense data format.
1808  *
1809  * @sense_buffer:       byte array of descriptor format sense data
1810  * @sb_len:             number of valid bytes in sense_buffer
1811  * @desc_type:          value of descriptor type to find
1812  *                      (e.g. 0 -> information)
1813  *
1814  * Notes:
1815  *      only valid when sense data is in descriptor format
1816  *
1817  * Return value:
1818  *      pointer to start of (first) descriptor if found else NULL
1819  **/
1820 const u8 * scsi_sense_desc_find(const u8 * sense_buffer, int sb_len,
1821                                 int desc_type)
1822 {
1823         int add_sen_len, add_len, desc_len, k;
1824         const u8 * descp;
1825
1826         if ((sb_len < 8) || (0 == (add_sen_len = sense_buffer[7])))
1827                 return NULL;
1828         if ((sense_buffer[0] < 0x72) || (sense_buffer[0] > 0x73))
1829                 return NULL;
1830         add_sen_len = (add_sen_len < (sb_len - 8)) ?
1831                         add_sen_len : (sb_len - 8);
1832         descp = &sense_buffer[8];
1833         for (desc_len = 0, k = 0; k < add_sen_len; k += desc_len) {
1834                 descp += desc_len;
1835                 add_len = (k < (add_sen_len - 1)) ? descp[1]: -1;
1836                 desc_len = add_len + 2;
1837                 if (descp[0] == desc_type)
1838                         return descp;
1839                 if (add_len < 0) // short descriptor ??
1840                         break;
1841         }
1842         return NULL;
1843 }
1844 EXPORT_SYMBOL(scsi_sense_desc_find);
1845
1846 /**
1847  * scsi_get_sense_info_fld - attempts to get information field from
1848  *                      sense data (either fixed or descriptor format)
1849  *
1850  * @sense_buffer:       byte array of sense data
1851  * @sb_len:             number of valid bytes in sense_buffer
1852  * @info_out:           pointer to 64 integer where 8 or 4 byte information
1853  *                      field will be placed if found.
1854  *
1855  * Return value:
1856  *      1 if information field found, 0 if not found.
1857  **/
1858 int scsi_get_sense_info_fld(const u8 * sense_buffer, int sb_len,
1859                             u64 * info_out)
1860 {
1861         int j;
1862         const u8 * ucp;
1863         u64 ull;
1864
1865         if (sb_len < 7)
1866                 return 0;
1867         switch (sense_buffer[0] & 0x7f) {
1868         case 0x70:
1869         case 0x71:
1870                 if (sense_buffer[0] & 0x80) {
1871                         *info_out = (sense_buffer[3] << 24) +
1872                                     (sense_buffer[4] << 16) +
1873                                     (sense_buffer[5] << 8) + sense_buffer[6];
1874                         return 1;
1875                 } else
1876                         return 0;
1877         case 0x72:
1878         case 0x73:
1879                 ucp = scsi_sense_desc_find(sense_buffer, sb_len,
1880                                            0 /* info desc */);
1881                 if (ucp && (0xa == ucp[1])) {
1882                         ull = 0;
1883                         for (j = 0; j < 8; ++j) {
1884                                 if (j > 0)
1885                                         ull <<= 8;
1886                                 ull |= ucp[4 + j];
1887                         }
1888                         *info_out = ull;
1889                         return 1;
1890                 } else
1891                         return 0;
1892         default:
1893                 return 0;
1894         }
1895 }
1896 EXPORT_SYMBOL(scsi_get_sense_info_fld);