/drivers/scsi/scsi_lib.c
C | 2560 lines | 1523 code | 310 blank | 727 comment | 289 complexity | a2587118637162b49f349d2c4f4766d6 MD5 | raw file
Possible License(s): CC-BY-SA-3.0, GPL-2.0, LGPL-2.0, AGPL-1.0
Large files files are truncated, but you can click here to view the full file
1/*
2 * scsi_lib.c Copyright (C) 1999 Eric Youngdale
3 *
4 * SCSI queueing library.
5 * Initial versions: Eric Youngdale (eric@andante.org).
6 * Based upon conversations with large numbers
7 * of people at Linux Expo.
8 */
9
10#include <linux/bio.h>
11#include <linux/bitops.h>
12#include <linux/blkdev.h>
13#include <linux/completion.h>
14#include <linux/kernel.h>
15#include <linux/mempool.h>
16#include <linux/slab.h>
17#include <linux/init.h>
18#include <linux/pci.h>
19#include <linux/delay.h>
20#include <linux/hardirq.h>
21#include <linux/scatterlist.h>
22
23#include <scsi/scsi.h>
24#include <scsi/scsi_cmnd.h>
25#include <scsi/scsi_dbg.h>
26#include <scsi/scsi_device.h>
27#include <scsi/scsi_driver.h>
28#include <scsi/scsi_eh.h>
29#include <scsi/scsi_host.h>
30
31#include "scsi_priv.h"
32#include "scsi_logging.h"
33
34
35#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
36#define SG_MEMPOOL_SIZE 2
37
38struct scsi_host_sg_pool {
39 size_t size;
40 char *name;
41 struct kmem_cache *slab;
42 mempool_t *pool;
43};
44
45#define SP(x) { x, "sgpool-" __stringify(x) }
46#if (SCSI_MAX_SG_SEGMENTS < 32)
47#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
48#endif
49static struct scsi_host_sg_pool scsi_sg_pools[] = {
50 SP(8),
51 SP(16),
52#if (SCSI_MAX_SG_SEGMENTS > 32)
53 SP(32),
54#if (SCSI_MAX_SG_SEGMENTS > 64)
55 SP(64),
56#if (SCSI_MAX_SG_SEGMENTS > 128)
57 SP(128),
58#if (SCSI_MAX_SG_SEGMENTS > 256)
59#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
60#endif
61#endif
62#endif
63#endif
64 SP(SCSI_MAX_SG_SEGMENTS)
65};
66#undef SP
67
68struct kmem_cache *scsi_sdb_cache;
69
70static void scsi_run_queue(struct request_queue *q);
71
72/*
73 * Function: scsi_unprep_request()
74 *
75 * Purpose: Remove all preparation done for a request, including its
76 * associated scsi_cmnd, so that it can be requeued.
77 *
78 * Arguments: req - request to unprepare
79 *
80 * Lock status: Assumed that no locks are held upon entry.
81 *
82 * Returns: Nothing.
83 */
84static void scsi_unprep_request(struct request *req)
85{
86 struct scsi_cmnd *cmd = req->special;
87
88 req->cmd_flags &= ~REQ_DONTPREP;
89 req->special = NULL;
90
91 scsi_put_command(cmd);
92}
93
94/**
95 * __scsi_queue_insert - private queue insertion
96 * @cmd: The SCSI command being requeued
97 * @reason: The reason for the requeue
98 * @unbusy: Whether the queue should be unbusied
99 *
100 * This is a private queue insertion. The public interface
101 * scsi_queue_insert() always assumes the queue should be unbusied
102 * because it's always called before the completion. This function is
103 * for a requeue after completion, which should only occur in this
104 * file.
105 */
106static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
107{
108 struct Scsi_Host *host = cmd->device->host;
109 struct scsi_device *device = cmd->device;
110 struct scsi_target *starget = scsi_target(device);
111 struct request_queue *q = device->request_queue;
112 unsigned long flags;
113
114 SCSI_LOG_MLQUEUE(1,
115 printk("Inserting command %p into mlqueue\n", cmd));
116
117 /*
118 * Set the appropriate busy bit for the device/host.
119 *
120 * If the host/device isn't busy, assume that something actually
121 * completed, and that we should be able to queue a command now.
122 *
123 * Note that the prior mid-layer assumption that any host could
124 * always queue at least one command is now broken. The mid-layer
125 * will implement a user specifiable stall (see
126 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
127 * if a command is requeued with no other commands outstanding
128 * either for the device or for the host.
129 */
130 switch (reason) {
131 case SCSI_MLQUEUE_HOST_BUSY:
132 host->host_blocked = host->max_host_blocked;
133 break;
134 case SCSI_MLQUEUE_DEVICE_BUSY:
135 device->device_blocked = device->max_device_blocked;
136 break;
137 case SCSI_MLQUEUE_TARGET_BUSY:
138 starget->target_blocked = starget->max_target_blocked;
139 break;
140 }
141
142 /*
143 * Decrement the counters, since these commands are no longer
144 * active on the host/device.
145 */
146 if (unbusy)
147 scsi_device_unbusy(device);
148
149 /*
150 * Requeue this command. It will go before all other commands
151 * that are already in the queue.
152 *
153 * NOTE: there is magic here about the way the queue is plugged if
154 * we have no outstanding commands.
155 *
156 * Although we *don't* plug the queue, we call the request
157 * function. The SCSI request function detects the blocked condition
158 * and plugs the queue appropriately.
159 */
160 spin_lock_irqsave(q->queue_lock, flags);
161 blk_requeue_request(q, cmd->request);
162 spin_unlock_irqrestore(q->queue_lock, flags);
163
164 scsi_run_queue(q);
165
166 return 0;
167}
168
169/*
170 * Function: scsi_queue_insert()
171 *
172 * Purpose: Insert a command in the midlevel queue.
173 *
174 * Arguments: cmd - command that we are adding to queue.
175 * reason - why we are inserting command to queue.
176 *
177 * Lock status: Assumed that lock is not held upon entry.
178 *
179 * Returns: Nothing.
180 *
181 * Notes: We do this for one of two cases. Either the host is busy
182 * and it cannot accept any more commands for the time being,
183 * or the device returned QUEUE_FULL and can accept no more
184 * commands.
185 * Notes: This could be called either from an interrupt context or a
186 * normal process context.
187 */
188int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
189{
190 return __scsi_queue_insert(cmd, reason, 1);
191}
192/**
193 * scsi_execute - insert request and wait for the result
194 * @sdev: scsi device
195 * @cmd: scsi command
196 * @data_direction: data direction
197 * @buffer: data buffer
198 * @bufflen: len of buffer
199 * @sense: optional sense buffer
200 * @timeout: request timeout in seconds
201 * @retries: number of times to retry request
202 * @flags: or into request flags;
203 * @resid: optional residual length
204 *
205 * returns the req->errors value which is the scsi_cmnd result
206 * field.
207 */
208int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
209 int data_direction, void *buffer, unsigned bufflen,
210 unsigned char *sense, int timeout, int retries, int flags,
211 int *resid)
212{
213 struct request *req;
214 int write = (data_direction == DMA_TO_DEVICE);
215 int ret = DRIVER_ERROR << 24;
216
217 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
218
219 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
220 buffer, bufflen, __GFP_WAIT))
221 goto out;
222
223 req->cmd_len = COMMAND_SIZE(cmd[0]);
224 memcpy(req->cmd, cmd, req->cmd_len);
225 req->sense = sense;
226 req->sense_len = 0;
227 req->retries = retries;
228 req->timeout = timeout;
229 req->cmd_type = REQ_TYPE_BLOCK_PC;
230 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
231
232 /*
233 * head injection *required* here otherwise quiesce won't work
234 */
235 blk_execute_rq(req->q, NULL, req, 1);
236
237 /*
238 * Some devices (USB mass-storage in particular) may transfer
239 * garbage data together with a residue indicating that the data
240 * is invalid. Prevent the garbage from being misinterpreted
241 * and prevent security leaks by zeroing out the excess data.
242 */
243 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
244 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
245
246 if (resid)
247 *resid = req->resid_len;
248 ret = req->errors;
249 out:
250 blk_put_request(req);
251
252 return ret;
253}
254EXPORT_SYMBOL(scsi_execute);
255
256
257int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
258 int data_direction, void *buffer, unsigned bufflen,
259 struct scsi_sense_hdr *sshdr, int timeout, int retries,
260 int *resid)
261{
262 char *sense = NULL;
263 int result;
264
265 if (sshdr) {
266 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
267 if (!sense)
268 return DRIVER_ERROR << 24;
269 }
270 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
271 sense, timeout, retries, 0, resid);
272 if (sshdr)
273 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
274
275 kfree(sense);
276 return result;
277}
278EXPORT_SYMBOL(scsi_execute_req);
279
280/*
281 * Function: scsi_init_cmd_errh()
282 *
283 * Purpose: Initialize cmd fields related to error handling.
284 *
285 * Arguments: cmd - command that is ready to be queued.
286 *
287 * Notes: This function has the job of initializing a number of
288 * fields related to error handling. Typically this will
289 * be called once for each command, as required.
290 */
291static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
292{
293 cmd->serial_number = 0;
294 scsi_set_resid(cmd, 0);
295 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
296 if (cmd->cmd_len == 0)
297 cmd->cmd_len = scsi_command_size(cmd->cmnd);
298}
299
300void scsi_device_unbusy(struct scsi_device *sdev)
301{
302 struct Scsi_Host *shost = sdev->host;
303 struct scsi_target *starget = scsi_target(sdev);
304 unsigned long flags;
305
306 spin_lock_irqsave(shost->host_lock, flags);
307 shost->host_busy--;
308 starget->target_busy--;
309 if (unlikely(scsi_host_in_recovery(shost) &&
310 (shost->host_failed || shost->host_eh_scheduled)))
311 scsi_eh_wakeup(shost);
312 spin_unlock(shost->host_lock);
313 spin_lock(sdev->request_queue->queue_lock);
314 sdev->device_busy--;
315 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
316}
317
318/*
319 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
320 * and call blk_run_queue for all the scsi_devices on the target -
321 * including current_sdev first.
322 *
323 * Called with *no* scsi locks held.
324 */
325static void scsi_single_lun_run(struct scsi_device *current_sdev)
326{
327 struct Scsi_Host *shost = current_sdev->host;
328 struct scsi_device *sdev, *tmp;
329 struct scsi_target *starget = scsi_target(current_sdev);
330 unsigned long flags;
331
332 spin_lock_irqsave(shost->host_lock, flags);
333 starget->starget_sdev_user = NULL;
334 spin_unlock_irqrestore(shost->host_lock, flags);
335
336 /*
337 * Call blk_run_queue for all LUNs on the target, starting with
338 * current_sdev. We race with others (to set starget_sdev_user),
339 * but in most cases, we will be first. Ideally, each LU on the
340 * target would get some limited time or requests on the target.
341 */
342 blk_run_queue(current_sdev->request_queue);
343
344 spin_lock_irqsave(shost->host_lock, flags);
345 if (starget->starget_sdev_user)
346 goto out;
347 list_for_each_entry_safe(sdev, tmp, &starget->devices,
348 same_target_siblings) {
349 if (sdev == current_sdev)
350 continue;
351 if (scsi_device_get(sdev))
352 continue;
353
354 spin_unlock_irqrestore(shost->host_lock, flags);
355 blk_run_queue(sdev->request_queue);
356 spin_lock_irqsave(shost->host_lock, flags);
357
358 scsi_device_put(sdev);
359 }
360 out:
361 spin_unlock_irqrestore(shost->host_lock, flags);
362}
363
364static inline int scsi_device_is_busy(struct scsi_device *sdev)
365{
366 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
367 return 1;
368
369 return 0;
370}
371
372static inline int scsi_target_is_busy(struct scsi_target *starget)
373{
374 return ((starget->can_queue > 0 &&
375 starget->target_busy >= starget->can_queue) ||
376 starget->target_blocked);
377}
378
379static inline int scsi_host_is_busy(struct Scsi_Host *shost)
380{
381 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
382 shost->host_blocked || shost->host_self_blocked)
383 return 1;
384
385 return 0;
386}
387
388/*
389 * Function: scsi_run_queue()
390 *
391 * Purpose: Select a proper request queue to serve next
392 *
393 * Arguments: q - last request's queue
394 *
395 * Returns: Nothing
396 *
397 * Notes: The previous command was completely finished, start
398 * a new one if possible.
399 */
400static void scsi_run_queue(struct request_queue *q)
401{
402 struct scsi_device *sdev = q->queuedata;
403 struct Scsi_Host *shost = sdev->host;
404 LIST_HEAD(starved_list);
405 unsigned long flags;
406
407 if (scsi_target(sdev)->single_lun)
408 scsi_single_lun_run(sdev);
409
410 spin_lock_irqsave(shost->host_lock, flags);
411 list_splice_init(&shost->starved_list, &starved_list);
412
413 while (!list_empty(&starved_list)) {
414 int flagset;
415
416 /*
417 * As long as shost is accepting commands and we have
418 * starved queues, call blk_run_queue. scsi_request_fn
419 * drops the queue_lock and can add us back to the
420 * starved_list.
421 *
422 * host_lock protects the starved_list and starved_entry.
423 * scsi_request_fn must get the host_lock before checking
424 * or modifying starved_list or starved_entry.
425 */
426 if (scsi_host_is_busy(shost))
427 break;
428
429 sdev = list_entry(starved_list.next,
430 struct scsi_device, starved_entry);
431 list_del_init(&sdev->starved_entry);
432 if (scsi_target_is_busy(scsi_target(sdev))) {
433 list_move_tail(&sdev->starved_entry,
434 &shost->starved_list);
435 continue;
436 }
437
438 spin_unlock(shost->host_lock);
439
440 spin_lock(sdev->request_queue->queue_lock);
441 flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
442 !test_bit(QUEUE_FLAG_REENTER,
443 &sdev->request_queue->queue_flags);
444 if (flagset)
445 queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue);
446 __blk_run_queue(sdev->request_queue);
447 if (flagset)
448 queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue);
449 spin_unlock(sdev->request_queue->queue_lock);
450
451 spin_lock(shost->host_lock);
452 }
453 /* put any unprocessed entries back */
454 list_splice(&starved_list, &shost->starved_list);
455 spin_unlock_irqrestore(shost->host_lock, flags);
456
457 blk_run_queue(q);
458}
459
460/*
461 * Function: scsi_requeue_command()
462 *
463 * Purpose: Handle post-processing of completed commands.
464 *
465 * Arguments: q - queue to operate on
466 * cmd - command that may need to be requeued.
467 *
468 * Returns: Nothing
469 *
470 * Notes: After command completion, there may be blocks left
471 * over which weren't finished by the previous command
472 * this can be for a number of reasons - the main one is
473 * I/O errors in the middle of the request, in which case
474 * we need to request the blocks that come after the bad
475 * sector.
476 * Notes: Upon return, cmd is a stale pointer.
477 */
478static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
479{
480 struct request *req = cmd->request;
481 unsigned long flags;
482
483 spin_lock_irqsave(q->queue_lock, flags);
484 scsi_unprep_request(req);
485 blk_requeue_request(q, req);
486 spin_unlock_irqrestore(q->queue_lock, flags);
487
488 scsi_run_queue(q);
489}
490
491void scsi_next_command(struct scsi_cmnd *cmd)
492{
493 struct scsi_device *sdev = cmd->device;
494 struct request_queue *q = sdev->request_queue;
495
496 /* need to hold a reference on the device before we let go of the cmd */
497 get_device(&sdev->sdev_gendev);
498
499 scsi_put_command(cmd);
500 scsi_run_queue(q);
501
502 /* ok to remove device now */
503 put_device(&sdev->sdev_gendev);
504}
505
506void scsi_run_host_queues(struct Scsi_Host *shost)
507{
508 struct scsi_device *sdev;
509
510 shost_for_each_device(sdev, shost)
511 scsi_run_queue(sdev->request_queue);
512}
513
514static void __scsi_release_buffers(struct scsi_cmnd *, int);
515
516/*
517 * Function: scsi_end_request()
518 *
519 * Purpose: Post-processing of completed commands (usually invoked at end
520 * of upper level post-processing and scsi_io_completion).
521 *
522 * Arguments: cmd - command that is complete.
523 * error - 0 if I/O indicates success, < 0 for I/O error.
524 * bytes - number of bytes of completed I/O
525 * requeue - indicates whether we should requeue leftovers.
526 *
527 * Lock status: Assumed that lock is not held upon entry.
528 *
529 * Returns: cmd if requeue required, NULL otherwise.
530 *
531 * Notes: This is called for block device requests in order to
532 * mark some number of sectors as complete.
533 *
534 * We are guaranteeing that the request queue will be goosed
535 * at some point during this call.
536 * Notes: If cmd was requeued, upon return it will be a stale pointer.
537 */
538static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
539 int bytes, int requeue)
540{
541 struct request_queue *q = cmd->device->request_queue;
542 struct request *req = cmd->request;
543
544 /*
545 * If there are blocks left over at the end, set up the command
546 * to queue the remainder of them.
547 */
548 if (blk_end_request(req, error, bytes)) {
549 /* kill remainder if no retrys */
550 if (error && scsi_noretry_cmd(cmd))
551 blk_end_request_all(req, error);
552 else {
553 if (requeue) {
554 /*
555 * Bleah. Leftovers again. Stick the
556 * leftovers in the front of the
557 * queue, and goose the queue again.
558 */
559 scsi_release_buffers(cmd);
560 scsi_requeue_command(q, cmd);
561 cmd = NULL;
562 }
563 return cmd;
564 }
565 }
566
567 /*
568 * This will goose the queue request function at the end, so we don't
569 * need to worry about launching another command.
570 */
571 __scsi_release_buffers(cmd, 0);
572 scsi_next_command(cmd);
573 return NULL;
574}
575
576static inline unsigned int scsi_sgtable_index(unsigned short nents)
577{
578 unsigned int index;
579
580 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
581
582 if (nents <= 8)
583 index = 0;
584 else
585 index = get_count_order(nents) - 3;
586
587 return index;
588}
589
590static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
591{
592 struct scsi_host_sg_pool *sgp;
593
594 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
595 mempool_free(sgl, sgp->pool);
596}
597
598static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
599{
600 struct scsi_host_sg_pool *sgp;
601
602 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
603 return mempool_alloc(sgp->pool, gfp_mask);
604}
605
606static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
607 gfp_t gfp_mask)
608{
609 int ret;
610
611 BUG_ON(!nents);
612
613 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
614 gfp_mask, scsi_sg_alloc);
615 if (unlikely(ret))
616 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
617 scsi_sg_free);
618
619 return ret;
620}
621
622static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
623{
624 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
625}
626
627static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
628{
629
630 if (cmd->sdb.table.nents)
631 scsi_free_sgtable(&cmd->sdb);
632
633 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
634
635 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
636 struct scsi_data_buffer *bidi_sdb =
637 cmd->request->next_rq->special;
638 scsi_free_sgtable(bidi_sdb);
639 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
640 cmd->request->next_rq->special = NULL;
641 }
642
643 if (scsi_prot_sg_count(cmd))
644 scsi_free_sgtable(cmd->prot_sdb);
645}
646
647/*
648 * Function: scsi_release_buffers()
649 *
650 * Purpose: Completion processing for block device I/O requests.
651 *
652 * Arguments: cmd - command that we are bailing.
653 *
654 * Lock status: Assumed that no lock is held upon entry.
655 *
656 * Returns: Nothing
657 *
658 * Notes: In the event that an upper level driver rejects a
659 * command, we must release resources allocated during
660 * the __init_io() function. Primarily this would involve
661 * the scatter-gather table, and potentially any bounce
662 * buffers.
663 */
664void scsi_release_buffers(struct scsi_cmnd *cmd)
665{
666 __scsi_release_buffers(cmd, 1);
667}
668EXPORT_SYMBOL(scsi_release_buffers);
669
670/*
671 * Function: scsi_io_completion()
672 *
673 * Purpose: Completion processing for block device I/O requests.
674 *
675 * Arguments: cmd - command that is finished.
676 *
677 * Lock status: Assumed that no lock is held upon entry.
678 *
679 * Returns: Nothing
680 *
681 * Notes: This function is matched in terms of capabilities to
682 * the function that created the scatter-gather list.
683 * In other words, if there are no bounce buffers
684 * (the normal case for most drivers), we don't need
685 * the logic to deal with cleaning up afterwards.
686 *
687 * We must call scsi_end_request(). This will finish off
688 * the specified number of sectors. If we are done, the
689 * command block will be released and the queue function
690 * will be goosed. If we are not done then we have to
691 * figure out what to do next:
692 *
693 * a) We can call scsi_requeue_command(). The request
694 * will be unprepared and put back on the queue. Then
695 * a new command will be created for it. This should
696 * be used if we made forward progress, or if we want
697 * to switch from READ(10) to READ(6) for example.
698 *
699 * b) We can call scsi_queue_insert(). The request will
700 * be put back on the queue and retried using the same
701 * command as before, possibly after a delay.
702 *
703 * c) We can call blk_end_request() with -EIO to fail
704 * the remainder of the request.
705 */
706void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
707{
708 int result = cmd->result;
709 struct request_queue *q = cmd->device->request_queue;
710 struct request *req = cmd->request;
711 int error = 0;
712 struct scsi_sense_hdr sshdr;
713 int sense_valid = 0;
714 int sense_deferred = 0;
715 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
716 ACTION_DELAYED_RETRY} action;
717 char *description = NULL;
718
719 if (result) {
720 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
721 if (sense_valid)
722 sense_deferred = scsi_sense_is_deferred(&sshdr);
723 }
724
725 if (blk_pc_request(req)) { /* SG_IO ioctl from block level */
726 req->errors = result;
727 if (result) {
728 if (sense_valid && req->sense) {
729 /*
730 * SG_IO wants current and deferred errors
731 */
732 int len = 8 + cmd->sense_buffer[7];
733
734 if (len > SCSI_SENSE_BUFFERSIZE)
735 len = SCSI_SENSE_BUFFERSIZE;
736 memcpy(req->sense, cmd->sense_buffer, len);
737 req->sense_len = len;
738 }
739 if (!sense_deferred)
740 error = -EIO;
741 }
742
743 req->resid_len = scsi_get_resid(cmd);
744
745 if (scsi_bidi_cmnd(cmd)) {
746 /*
747 * Bidi commands Must be complete as a whole,
748 * both sides at once.
749 */
750 req->next_rq->resid_len = scsi_in(cmd)->resid;
751
752 scsi_release_buffers(cmd);
753 blk_end_request_all(req, 0);
754
755 scsi_next_command(cmd);
756 return;
757 }
758 }
759
760 BUG_ON(blk_bidi_rq(req)); /* bidi not support for !blk_pc_request yet */
761
762 /*
763 * Next deal with any sectors which we were able to correctly
764 * handle.
765 */
766 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
767 "%d bytes done.\n",
768 blk_rq_sectors(req), good_bytes));
769
770 /*
771 * Recovered errors need reporting, but they're always treated
772 * as success, so fiddle the result code here. For BLOCK_PC
773 * we already took a copy of the original into rq->errors which
774 * is what gets returned to the user
775 */
776 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
777 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
778 * print since caller wants ATA registers. Only occurs on
779 * SCSI ATA PASS_THROUGH commands when CK_COND=1
780 */
781 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
782 ;
783 else if (!(req->cmd_flags & REQ_QUIET))
784 scsi_print_sense("", cmd);
785 result = 0;
786 /* BLOCK_PC may have set error */
787 error = 0;
788 }
789
790 /*
791 * A number of bytes were successfully read. If there
792 * are leftovers and there is some kind of error
793 * (result != 0), retry the rest.
794 */
795 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
796 return;
797
798 error = -EIO;
799
800 if (host_byte(result) == DID_RESET) {
801 /* Third party bus reset or reset for error recovery
802 * reasons. Just retry the command and see what
803 * happens.
804 */
805 action = ACTION_RETRY;
806 } else if (sense_valid && !sense_deferred) {
807 switch (sshdr.sense_key) {
808 case UNIT_ATTENTION:
809 if (cmd->device->removable) {
810 /* Detected disc change. Set a bit
811 * and quietly refuse further access.
812 */
813 cmd->device->changed = 1;
814 description = "Media Changed";
815 action = ACTION_FAIL;
816 } else {
817 /* Must have been a power glitch, or a
818 * bus reset. Could not have been a
819 * media change, so we just retry the
820 * command and see what happens.
821 */
822 action = ACTION_RETRY;
823 }
824 break;
825 case ILLEGAL_REQUEST:
826 /* If we had an ILLEGAL REQUEST returned, then
827 * we may have performed an unsupported
828 * command. The only thing this should be
829 * would be a ten byte read where only a six
830 * byte read was supported. Also, on a system
831 * where READ CAPACITY failed, we may have
832 * read past the end of the disk.
833 */
834 if ((cmd->device->use_10_for_rw &&
835 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
836 (cmd->cmnd[0] == READ_10 ||
837 cmd->cmnd[0] == WRITE_10)) {
838 /* This will issue a new 6-byte command. */
839 cmd->device->use_10_for_rw = 0;
840 action = ACTION_REPREP;
841 } else if (sshdr.asc == 0x10) /* DIX */ {
842 description = "Host Data Integrity Failure";
843 action = ACTION_FAIL;
844 error = -EILSEQ;
845 } else
846 action = ACTION_FAIL;
847 break;
848 case ABORTED_COMMAND:
849 action = ACTION_FAIL;
850 if (sshdr.asc == 0x10) { /* DIF */
851 description = "Target Data Integrity Failure";
852 error = -EILSEQ;
853 }
854 break;
855 case NOT_READY:
856 /* If the device is in the process of becoming
857 * ready, or has a temporary blockage, retry.
858 */
859 if (sshdr.asc == 0x04) {
860 switch (sshdr.ascq) {
861 case 0x01: /* becoming ready */
862 case 0x04: /* format in progress */
863 case 0x05: /* rebuild in progress */
864 case 0x06: /* recalculation in progress */
865 case 0x07: /* operation in progress */
866 case 0x08: /* Long write in progress */
867 case 0x09: /* self test in progress */
868 case 0x14: /* space allocation in progress */
869 action = ACTION_DELAYED_RETRY;
870 break;
871 default:
872 description = "Device not ready";
873 action = ACTION_FAIL;
874 break;
875 }
876 } else {
877 description = "Device not ready";
878 action = ACTION_FAIL;
879 }
880 break;
881 case VOLUME_OVERFLOW:
882 /* See SSC3rXX or current. */
883 action = ACTION_FAIL;
884 break;
885 default:
886 description = "Unhandled sense code";
887 action = ACTION_FAIL;
888 break;
889 }
890 } else {
891 description = "Unhandled error code";
892 action = ACTION_FAIL;
893 }
894
895 switch (action) {
896 case ACTION_FAIL:
897 /* Give up and fail the remainder of the request */
898 scsi_release_buffers(cmd);
899 if (!(req->cmd_flags & REQ_QUIET)) {
900 if (description)
901 scmd_printk(KERN_INFO, cmd, "%s\n",
902 description);
903 scsi_print_result(cmd);
904 if (driver_byte(result) & DRIVER_SENSE)
905 scsi_print_sense("", cmd);
906 scsi_print_command(cmd);
907 }
908 if (blk_end_request_err(req, error))
909 scsi_requeue_command(q, cmd);
910 else
911 scsi_next_command(cmd);
912 break;
913 case ACTION_REPREP:
914 /* Unprep the request and put it back at the head of the queue.
915 * A new command will be prepared and issued.
916 */
917 scsi_release_buffers(cmd);
918 scsi_requeue_command(q, cmd);
919 break;
920 case ACTION_RETRY:
921 /* Retry the same command immediately */
922 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
923 break;
924 case ACTION_DELAYED_RETRY:
925 /* Retry the same command after a delay */
926 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
927 break;
928 }
929}
930
931static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
932 gfp_t gfp_mask)
933{
934 int count;
935
936 /*
937 * If sg table allocation fails, requeue request later.
938 */
939 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
940 gfp_mask))) {
941 return BLKPREP_DEFER;
942 }
943
944 req->buffer = NULL;
945
946 /*
947 * Next, walk the list, and fill in the addresses and sizes of
948 * each segment.
949 */
950 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
951 BUG_ON(count > sdb->table.nents);
952 sdb->table.nents = count;
953 sdb->length = blk_rq_bytes(req);
954 return BLKPREP_OK;
955}
956
957/*
958 * Function: scsi_init_io()
959 *
960 * Purpose: SCSI I/O initialize function.
961 *
962 * Arguments: cmd - Command descriptor we wish to initialize
963 *
964 * Returns: 0 on success
965 * BLKPREP_DEFER if the failure is retryable
966 * BLKPREP_KILL if the failure is fatal
967 */
968int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
969{
970 int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
971 if (error)
972 goto err_exit;
973
974 if (blk_bidi_rq(cmd->request)) {
975 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
976 scsi_sdb_cache, GFP_ATOMIC);
977 if (!bidi_sdb) {
978 error = BLKPREP_DEFER;
979 goto err_exit;
980 }
981
982 cmd->request->next_rq->special = bidi_sdb;
983 error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
984 GFP_ATOMIC);
985 if (error)
986 goto err_exit;
987 }
988
989 if (blk_integrity_rq(cmd->request)) {
990 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
991 int ivecs, count;
992
993 BUG_ON(prot_sdb == NULL);
994 ivecs = blk_rq_count_integrity_sg(cmd->request);
995
996 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
997 error = BLKPREP_DEFER;
998 goto err_exit;
999 }
1000
1001 count = blk_rq_map_integrity_sg(cmd->request,
1002 prot_sdb->table.sgl);
1003 BUG_ON(unlikely(count > ivecs));
1004
1005 cmd->prot_sdb = prot_sdb;
1006 cmd->prot_sdb->table.nents = count;
1007 }
1008
1009 return BLKPREP_OK ;
1010
1011err_exit:
1012 scsi_release_buffers(cmd);
1013 if (error == BLKPREP_KILL)
1014 scsi_put_command(cmd);
1015 else /* BLKPREP_DEFER */
1016 scsi_unprep_request(cmd->request);
1017
1018 return error;
1019}
1020EXPORT_SYMBOL(scsi_init_io);
1021
1022static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1023 struct request *req)
1024{
1025 struct scsi_cmnd *cmd;
1026
1027 if (!req->special) {
1028 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1029 if (unlikely(!cmd))
1030 return NULL;
1031 req->special = cmd;
1032 } else {
1033 cmd = req->special;
1034 }
1035
1036 /* pull a tag out of the request if we have one */
1037 cmd->tag = req->tag;
1038 cmd->request = req;
1039
1040 cmd->cmnd = req->cmd;
1041
1042 return cmd;
1043}
1044
1045int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1046{
1047 struct scsi_cmnd *cmd;
1048 int ret = scsi_prep_state_check(sdev, req);
1049
1050 if (ret != BLKPREP_OK)
1051 return ret;
1052
1053 cmd = scsi_get_cmd_from_req(sdev, req);
1054 if (unlikely(!cmd))
1055 return BLKPREP_DEFER;
1056
1057 /*
1058 * BLOCK_PC requests may transfer data, in which case they must
1059 * a bio attached to them. Or they might contain a SCSI command
1060 * that does not transfer data, in which case they may optionally
1061 * submit a request without an attached bio.
1062 */
1063 if (req->bio) {
1064 int ret;
1065
1066 BUG_ON(!req->nr_phys_segments);
1067
1068 ret = scsi_init_io(cmd, GFP_ATOMIC);
1069 if (unlikely(ret))
1070 return ret;
1071 } else {
1072 BUG_ON(blk_rq_bytes(req));
1073
1074 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1075 req->buffer = NULL;
1076 }
1077
1078 cmd->cmd_len = req->cmd_len;
1079 if (!blk_rq_bytes(req))
1080 cmd->sc_data_direction = DMA_NONE;
1081 else if (rq_data_dir(req) == WRITE)
1082 cmd->sc_data_direction = DMA_TO_DEVICE;
1083 else
1084 cmd->sc_data_direction = DMA_FROM_DEVICE;
1085
1086 cmd->transfersize = blk_rq_bytes(req);
1087 cmd->allowed = req->retries;
1088 return BLKPREP_OK;
1089}
1090EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1091
1092/*
1093 * Setup a REQ_TYPE_FS command. These are simple read/write request
1094 * from filesystems that still need to be translated to SCSI CDBs from
1095 * the ULD.
1096 */
1097int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1098{
1099 struct scsi_cmnd *cmd;
1100 int ret = scsi_prep_state_check(sdev, req);
1101
1102 if (ret != BLKPREP_OK)
1103 return ret;
1104
1105 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1106 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1107 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1108 if (ret != BLKPREP_OK)
1109 return ret;
1110 }
1111
1112 /*
1113 * Filesystem requests must transfer data.
1114 */
1115 BUG_ON(!req->nr_phys_segments);
1116
1117 cmd = scsi_get_cmd_from_req(sdev, req);
1118 if (unlikely(!cmd))
1119 return BLKPREP_DEFER;
1120
1121 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1122 return scsi_init_io(cmd, GFP_ATOMIC);
1123}
1124EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1125
1126int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1127{
1128 int ret = BLKPREP_OK;
1129
1130 /*
1131 * If the device is not in running state we will reject some
1132 * or all commands.
1133 */
1134 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1135 switch (sdev->sdev_state) {
1136 case SDEV_OFFLINE:
1137 /*
1138 * If the device is offline we refuse to process any
1139 * commands. The device must be brought online
1140 * before trying any recovery commands.
1141 */
1142 sdev_printk(KERN_ERR, sdev,
1143 "rejecting I/O to offline device\n");
1144 ret = BLKPREP_KILL;
1145 break;
1146 case SDEV_DEL:
1147 /*
1148 * If the device is fully deleted, we refuse to
1149 * process any commands as well.
1150 */
1151 sdev_printk(KERN_ERR, sdev,
1152 "rejecting I/O to dead device\n");
1153 ret = BLKPREP_KILL;
1154 break;
1155 case SDEV_QUIESCE:
1156 case SDEV_BLOCK:
1157 case SDEV_CREATED_BLOCK:
1158 /*
1159 * If the devices is blocked we defer normal commands.
1160 */
1161 if (!(req->cmd_flags & REQ_PREEMPT))
1162 ret = BLKPREP_DEFER;
1163 break;
1164 default:
1165 /*
1166 * For any other not fully online state we only allow
1167 * special commands. In particular any user initiated
1168 * command is not allowed.
1169 */
1170 if (!(req->cmd_flags & REQ_PREEMPT))
1171 ret = BLKPREP_KILL;
1172 break;
1173 }
1174 }
1175 return ret;
1176}
1177EXPORT_SYMBOL(scsi_prep_state_check);
1178
1179int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1180{
1181 struct scsi_device *sdev = q->queuedata;
1182
1183 switch (ret) {
1184 case BLKPREP_KILL:
1185 req->errors = DID_NO_CONNECT << 16;
1186 /* release the command and kill it */
1187 if (req->special) {
1188 struct scsi_cmnd *cmd = req->special;
1189 scsi_release_buffers(cmd);
1190 scsi_put_command(cmd);
1191 req->special = NULL;
1192 }
1193 break;
1194 case BLKPREP_DEFER:
1195 /*
1196 * If we defer, the blk_peek_request() returns NULL, but the
1197 * queue must be restarted, so we plug here if no returning
1198 * command will automatically do that.
1199 */
1200 if (sdev->device_busy == 0)
1201 blk_plug_device(q);
1202 break;
1203 default:
1204 req->cmd_flags |= REQ_DONTPREP;
1205 }
1206
1207 return ret;
1208}
1209EXPORT_SYMBOL(scsi_prep_return);
1210
1211int scsi_prep_fn(struct request_queue *q, struct request *req)
1212{
1213 struct scsi_device *sdev = q->queuedata;
1214 int ret = BLKPREP_KILL;
1215
1216 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1217 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1218 return scsi_prep_return(q, req, ret);
1219}
1220EXPORT_SYMBOL(scsi_prep_fn);
1221
1222/*
1223 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1224 * return 0.
1225 *
1226 * Called with the queue_lock held.
1227 */
1228static inline int scsi_dev_queue_ready(struct request_queue *q,
1229 struct scsi_device *sdev)
1230{
1231 if (sdev->device_busy == 0 && sdev->device_blocked) {
1232 /*
1233 * unblock after device_blocked iterates to zero
1234 */
1235 if (--sdev->device_blocked == 0) {
1236 SCSI_LOG_MLQUEUE(3,
1237 sdev_printk(KERN_INFO, sdev,
1238 "unblocking device at zero depth\n"));
1239 } else {
1240 blk_plug_device(q);
1241 return 0;
1242 }
1243 }
1244 if (scsi_device_is_busy(sdev))
1245 return 0;
1246
1247 return 1;
1248}
1249
1250
1251/*
1252 * scsi_target_queue_ready: checks if there we can send commands to target
1253 * @sdev: scsi device on starget to check.
1254 *
1255 * Called with the host lock held.
1256 */
1257static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1258 struct scsi_device *sdev)
1259{
1260 struct scsi_target *starget = scsi_target(sdev);
1261
1262 if (starget->single_lun) {
1263 if (starget->starget_sdev_user &&
1264 starget->starget_sdev_user != sdev)
1265 return 0;
1266 starget->starget_sdev_user = sdev;
1267 }
1268
1269 if (starget->target_busy == 0 && starget->target_blocked) {
1270 /*
1271 * unblock after target_blocked iterates to zero
1272 */
1273 if (--starget->target_blocked == 0) {
1274 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1275 "unblocking target at zero depth\n"));
1276 } else
1277 return 0;
1278 }
1279
1280 if (scsi_target_is_busy(starget)) {
1281 if (list_empty(&sdev->starved_entry)) {
1282 list_add_tail(&sdev->starved_entry,
1283 &shost->starved_list);
1284 return 0;
1285 }
1286 }
1287
1288 /* We're OK to process the command, so we can't be starved */
1289 if (!list_empty(&sdev->starved_entry))
1290 list_del_init(&sdev->starved_entry);
1291 return 1;
1292}
1293
1294/*
1295 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1296 * return 0. We must end up running the queue again whenever 0 is
1297 * returned, else IO can hang.
1298 *
1299 * Called with host_lock held.
1300 */
1301static inline int scsi_host_queue_ready(struct request_queue *q,
1302 struct Scsi_Host *shost,
1303 struct scsi_device *sdev)
1304{
1305 if (scsi_host_in_recovery(shost))
1306 return 0;
1307 if (shost->host_busy == 0 && shost->host_blocked) {
1308 /*
1309 * unblock after host_blocked iterates to zero
1310 */
1311 if (--shost->host_blocked == 0) {
1312 SCSI_LOG_MLQUEUE(3,
1313 printk("scsi%d unblocking host at zero depth\n",
1314 shost->host_no));
1315 } else {
1316 return 0;
1317 }
1318 }
1319 if (scsi_host_is_busy(shost)) {
1320 if (list_empty(&sdev->starved_entry))
1321 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1322 return 0;
1323 }
1324
1325 /* We're OK to process the command, so we can't be starved */
1326 if (!list_empty(&sdev->starved_entry))
1327 list_del_init(&sdev->starved_entry);
1328
1329 return 1;
1330}
1331
1332/*
1333 * Busy state exporting function for request stacking drivers.
1334 *
1335 * For efficiency, no lock is taken to check the busy state of
1336 * shost/starget/sdev, since the returned value is not guaranteed and
1337 * may be changed after request stacking drivers call the function,
1338 * regardless of taking lock or not.
1339 *
1340 * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1341 * (e.g. !sdev), scsi needs to return 'not busy'.
1342 * Otherwise, request stacking drivers may hold requests forever.
1343 */
1344static int scsi_lld_busy(struct request_queue *q)
1345{
1346 struct scsi_device *sdev = q->queuedata;
1347 struct Scsi_Host *shost;
1348 struct scsi_target *starget;
1349
1350 if (!sdev)
1351 return 0;
1352
1353 shost = sdev->host;
1354 starget = scsi_target(sdev);
1355
1356 if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) ||
1357 scsi_target_is_busy(starget) || scsi_device_is_busy(sdev))
1358 return 1;
1359
1360 return 0;
1361}
1362
1363/*
1364 * Kill a request for a dead device
1365 */
1366static void scsi_kill_request(struct request *req, struct request_queue *q)
1367{
1368 struct scsi_cmnd *cmd = req->special;
1369 struct scsi_device *sdev;
1370 struct scsi_target *starget;
1371 struct Scsi_Host *shost;
1372
1373 blk_start_request(req);
1374
1375 if (unlikely(cmd == NULL)) {
1376 printk(KERN_CRIT "impossible request in %s.\n",
1377 __func__);
1378 BUG();
1379 }
1380
1381 sdev = cmd->device;
1382 starget = scsi_target(sdev);
1383 shost = sdev->host;
1384 scsi_init_cmd_errh(cmd);
1385 cmd->result = DID_NO_CONNECT << 16;
1386 atomic_inc(&cmd->device->iorequest_cnt);
1387
1388 /*
1389 * SCSI request completion path will do scsi_device_unbusy(),
1390 * bump busy counts. To bump the counters, we need to dance
1391 * with the locks as normal issue path does.
1392 */
1393 sdev->device_busy++;
1394 spin_unlock(sdev->request_queue->queue_lock);
1395 spin_lock(shost->host_lock);
1396 shost->host_busy++;
1397 starget->target_busy++;
1398 spin_unlock(shost->host_lock);
1399 spin_lock(sdev->request_queue->queue_lock);
1400
1401 blk_complete_request(req);
1402}
1403
1404static void scsi_softirq_done(struct request *rq)
1405{
1406 struct scsi_cmnd *cmd = rq->special;
1407 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1408 int disposition;
1409
1410 INIT_LIST_HEAD(&cmd->eh_entry);
1411
1412 /*
1413 * Set the serial numbers back to zero
1414 */
1415 cmd->serial_number = 0;
1416
1417 atomic_inc(&cmd->device->iodone_cnt);
1418 if (cmd->result)
1419 atomic_inc(&cmd->device->ioerr_cnt);
1420
1421 disposition = scsi_decide_disposition(cmd);
1422 if (disposition != SUCCESS &&
1423 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1424 sdev_printk(KERN_ERR, cmd->device,
1425 "timing out command, waited %lus\n",
1426 wait_for/HZ);
1427 disposition = SUCCESS;
1428 }
1429
1430 scsi_log_completion(cmd, disposition);
1431
1432 switch (disposition) {
1433 case SUCCESS:
1434 scsi_finish_command(cmd);
1435 break;
1436 case NEEDS_RETRY:
1437 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1438 break;
1439 case ADD_TO_MLQUEUE:
1440 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1441 break;
1442 default:
1443 if (!scsi_eh_scmd_add(cmd, 0))
1444 scsi_finish_command(cmd);
1445 }
1446}
1447
1448/*
1449 * Function: scsi_request_fn()
1450 *
1451 * Purpose: Main strategy routine for SCSI.
1452 *
1453 * Arguments: q - Pointer to actual queue.
1454 *
1455 * Returns: Nothing
1456 *
1457 * Lock status: IO request lock assumed to be held when called.
1458 */
1459static void scsi_request_fn(struct request_queue *q)
1460{
1461 struct scsi_device *sdev = q->queuedata;
1462 struct Scsi_Host *shost;
1463 struct scsi_cmnd *cmd;
1464 struct request *req;
1465
1466 if (!sdev) {
1467 printk("scsi: killing requests for dead queue\n");
1468 while ((req = blk_peek_request(q)) != NULL)
1469 scsi_kill_request(req, q);
1470 return;
1471 }
1472
1473 if(!get_device(&sdev->sdev_gendev))
1474 /* We must be tearing the block queue down already */
1475 return;
1476
1477 /*
1478 * To start with, we keep looping until the queue is empty, or until
1479 * the host is no longer able to accept any more requests.
1480 */
1481 shost = sdev->host;
1482 while (!blk_queue_plugged(q)) {
1483 int rtn;
1484 /*
1485 * get next queueable request. We do this early to make sure
1486 * that the request is fully prepared even if we cannot
1487 * accept it.
1488 */
1489 req = blk_peek_request(q);
1490 if (!req || !scsi_dev_queue_ready(q, sdev))
1491 break;
1492
1493 if (unlikely(!scsi_device_online(sdev))) {
1494 sdev_printk(KERN_ERR, sdev,
1495 "rejecting I/O to offline device\n");
1496 scsi_kill_request(req, q);
1497 continue;
1498 }
1499
1500
1501 /*
1502 * Remove the request from the request list.
1503 */
1504 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1505 blk_start_request(req);
1506 sdev->device_busy++;
1507
1508 spin_unlock(q->queue_lock);
1509 cmd = req->special;
1510 if (unlikely(cmd == NULL)) {
1511 printk(KERN_CRIT "impossible request in %s.\n"
1512 "please mail a stack trace to "
1513 "linux-scsi@vger.kernel.org\n",
1514 __func__);
1515 blk_dump_rq_flags(req, "foo");
1516 BUG();
1517 }
1518 spin_lock(shost->host_lock);
1519
1520 /*
1521 * We hit this when the driver is using a host wide
1522 * tag map. For device level tag maps the queue_depth check
1523 * in the device ready fn would prevent us from trying
1524 * to allocate a tag. Since the map is a shared host resource
1525 * we add the dev to the starved list so it eventually gets
1526 * a run when a tag is freed.
1527 */
1528 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1529 if (list_empty(&sdev->starved_entry))
1530 list_add_tail(&sdev->starved_entry,
1531 &shost->starved_list);
1532 goto not_ready;
1533 }
1534
1535 if (!scsi_target_queue_ready(shost, sdev))
1536 goto not_ready;
1537
1538 if (!scsi_host_queue_ready(q, shost, sdev))
1539 goto not_ready;
1540
1541 scsi_target(sdev)->target_busy++;
1542 shost->host_busy++;
1543
1544 /*
1545 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1546 * take the lock again.
1547 */
1548 spin_unlock_irq(shost->host_lock);
1549
1550 /*
1551 * Finally, initialize any error handling parameters, and set up
1552 * the timers for timeouts.
1553 */
1554 scsi_init_cmd_errh(cmd);
1555
1556 /*
1557 * Dispatch the command to the low-level driver.
1558 */
1559 rtn = scsi_dispatch_cmd(cmd);
1560 spin_lock_irq(q->queue_lock);
1561 if(rtn) {
1562 /* we're refusing the command; because of
1563 * the way locks get dropped, we need to
1564 * check here if plugging is required */
1565 if(sdev->device_busy == 0)
1566 blk_plug_device(q);
1567
1568 break;
1569 }
1570 }
1571
1572 goto out;
1573
1574 not_ready:
1575 spin_unlock_irq(shost->host_lock);
1576
1577 /*
1578 * lock q, handle tag, requeue req, and decrement device_busy. We
1579 * must return with queue_lock held.
1580 *
1581 * Decrementing device_busy without checking it is OK, as all such
1582 * cases (host limits or settings) should run the queue at some
1583 * later time.
1584 */
1585 spin_lock_irq(q->queue_lock);
1586 blk_requeue_request(q, req);
1587 sdev->device_busy--;
1588 if(sdev->device_busy == 0)
1589 blk_plug_device(q);
1590 out:
1591 /* must be careful here...if we trigger the ->remove() function
1592 * we cannot be holding the q lock */
1593 spin_unlock_irq(q->queue_lock);
1594 put_device(&sdev->sdev_gendev);
1595 spin_lock_irq(q->queue_lock);
1596}
1597
1598u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1599{
1600 struct device *host_dev;
1601 u64 bounce_limit = 0xffffffff;
1602
1603 if (shost->unchecked_isa_dma)
1604 return BLK_BOUNCE_ISA;
1605 /*
1606 * Platforms with virtual-DMA translation
1607 * hardware have no practical limit.
1608 */
1609 if (!PCI_DMA_BUS_IS_PHYS)
1610 return BLK_BOUNCE_ANY;
1611
1612 host_dev = scsi_get_device(shost);
1613 if (host_dev && host_dev->dma_mask)
1614 bounce_limit = *host_dev->dma_mask;
1615
1616 return bounce_limit;
1617}
1618EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1619
1620struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1621 request_fn_proc *request_fn)
1622{
1623 struct request_queue *q;
1624 struct device *dev = shost->shost_gendev.parent;
1625
1626 q = blk_init_queue(request_fn, NULL);
1627 if (!q)
1628 return NULL;
1629
1630 /*
1631 * this limit is imposed by hardware restrictions
1632 */
1633 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1634 SCSI_MAX_SG_CHAIN_SEGMENTS));
1635
1636 blk_queue_max_hw_sectors(q, shost->max_sectors);
1637 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1638 blk_queue_segment_boundary(q, shost->dma_boundary);
1639 dma_set_seg_boundary(dev, shost->dma_boundary);
1640
1641 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1642
1643 /* New queue, no concurrency on queue_flags */
1644 if (!shost->use_clustering)
1645 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
1646
1647 /*
1648 * set a reasonable default alignment on word boundaries: the
1649 * host and device may alter it using
1650 * blk_queue_update_dma_alignment() later.
1651 */
1652 blk_queue_dma_alignment(q, 0x03);
1653
1654 return q;
1655}
1656EXPORT_SYMBOL(__scsi_alloc_queue);
1657
1658struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1659{
1660 struct request_queue *q;
1661
1662 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1663 if (!q)
1664 return NULL;
1665
1666 blk_queue_prep_rq(q, scsi_prep_fn);
1667 blk_queue_softirq_done(q, scsi_softirq_done);
1668 blk_queue_rq_timed_out(q, scsi_times_out);
1669 blk_queue_lld_busy(q, scsi_lld_busy);
1670 return q;
1671}
1672
1673void scsi_free_queue(struct request_queue *q)
1674{
1675 blk_cleanup_queue(q);
1676}
1677
1678/*
1679 * Function: scsi_block_requests()
1680 *
1681 * Purpose: Utility function used by low-level drivers to prevent further
1682 * commands from being queued to the device.
1683 *
1684 * Arguments: shost - Host in question
1685 *
1686 * Returns: Nothing
1687 *
1688 * Lock status: No locks are assumed held.
1689 *
1690 * Notes: There is no timer nor any other means by which the requests
1691 * get unblocked other than the low-level driver calling
1692 * scsi_unblock_requests().
1693 */
1694void scsi_block_requests(struct Scsi_Host *shost)
1695{
1696 shost->host_self_blocked = 1;
1697}
1698EXPORT_SYMBOL(scsi_block_requests);
1699
1700/*
1701 * Function: scsi_unblock_requests()
1702 *
1703 * Purpose: Utility function used by low-level drivers to allow further
1704 * commands from being queued to the device.
1705 *
1706 * Arguments: shost - Host in question
1707 *
1708 * Returns: Nothing
1709 *
1710 * Lock status: No locks are assumed held.
1711 *
1712 * Notes: There is no timer nor any other means by which the requests
1713 * get unblocked other than the low-level driver calling
1714 * scsi_unblock_requests().
1715 *
1716 * This is done as an API function so that changes to the
1717 * internals of the scsi mid-layer won't require wholesale
1718 * changes to drivers that use this feature.
1719 */
1720void scsi_unblock_requests(struct Scsi_Host *shost)
1721{
1722 shost->host_self_blocked = 0;
1723 scsi_run_host_queues(shost);
1724}
1725EXPORT_SYMBOL(scsi_unblock_requests);
1726
1727int __init scsi_init_queue(void)
1728{
1729 int i;
1730
1731 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1732 sizeof(struct scsi_data_buffer),
1733 0, 0, NULL);
1734 if (!scsi_sdb_cache) {
1735 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1736 return -ENOMEM;
1737 }
1738
1739 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1740 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1741 int size = sgp->size * sizeof(struct scatterlist);
1742
1743 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1744 SLAB_HWCACHE_ALIGN, NULL);
1745 if (!sgp->slab) {
1746 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1747 sgp->name);
1748 goto cleanup_sdb;
1749 }
1750
1751 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1752 sgp->slab);
1753 if (!sgp->pool) {
1754 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1755 sgp->name);
1756 goto cleanup_sdb;
1757 }
1758 }
1759
1760 return 0;
1761
1762cleanup_sdb:
1763 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1764 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1765 if (sgp->pool)
1766 mempool_destroy(sgp->pool);
1767 if (sgp->slab)
1768 kmem_cache_destroy(sgp->slab);
1769 }
1770 kmem_cache_destroy(scsi_sdb_cache);
1771
1772 return -ENOMEM;
1773}
1774
1775void scsi_exit_queue(void)
1776{
1777 int i;
1778
1779 kmem_cache_destroy(scsi_sdb_cache);
1780
1781 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1782 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1783 mempool_destroy(sgp->pool);
1784 kmem_cache_destroy(sgp->slab);
1785 }
1786}
1787
1788/**
1789 * scsi_mode_select - issue a mode select
1790 * @sdev: SCSI device to be queried
1791 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1792 * @sp: Save page bit (0 == don't save, 1 == save)
1793 * @modepage: mode page being requested
1794 * @buffer: request buffer (may not be smaller than eight bytes)
1795 * @len: length of request buffer.
1796 * @timeout: command timeout
1797 * @retries: number of retries before failing
1798 * @data: returns a structure abstracting the mode header data
1799 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1800 * must be SCSI_SENSE_BUFFERSIZE big.
1801 *
1802 * Returns zero if successful; negative error number or scsi
1803 * status on error
1804 *
1805 */
1806int
1807scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1808 unsigned char *buffer, int len, int timeout, int retries,
1809 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1810{
1811 unsigned char cmd[10];
1812 unsigned char *real_buffer;
1813 int ret;
1814
1815 memset(cmd, 0, sizeof(cmd));
1816 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1817
1818 if (sdev->use_10_for_ms) {
1819 if (len > 65535)
1820 return -EINVAL;
1821 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1822 if (!real_buffer)
1823 return -ENOMEM;
1824 memcpy(real_buffer + 8, buffer, len);
1825 len += 8;
1826 real_buffer[0] = 0;
1827 real_buffer[1] = 0;
1828 real_buffer[2] = data->medium_type;
1829 real_buffer[3] = data->device_specific;
1830 real_buffer[4] = data->longlba ? 0x01 : 0;
1831 real_buffer[5] = 0;
1832 real_buffer[6] = data->block_descriptor_length >> 8;
1833 real_buffer[7] = data->block_descriptor_length;
1834
1835 cmd[0] = MODE_SELECT_10;
1836 cmd[7] = len >> 8;
1837 cmd[8] = len;
1838 } else {
1839 if (len > 255 || data->block_descriptor_length > 255 ||
1840 data->longlba)
1841 return -EINVAL;
1842
1843 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1844 if (!real_buffer)
1845 return -ENOMEM;
1846 memcpy(real_buffer + 4, buffer, len);
1847 len += 4;
1848 real_buffer[0] = 0;
1849 real_buffer[1] = data->medium_type;
1850 real_buffer[2] = data->device_specific;
1851 real_buffer[3] = data->block_descriptor_length;
1852
1853
1854 cmd[0] = MODE_SELECT;
1855 cmd[4] = len;
1856 }
1857
1858 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1859 sshdr, timeout, retries, NULL);
1860 kfree(real_buffer);
1861 return ret;
1862}
1863EXPORT_SYMBOL_GPL(scsi_mode_select);
1864
1865/**
1866 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1867 * @sdev: SCSI device to be queried
1868 * @dbd: set if mode sense will allow block descriptors to be returned
1869 * @modepage: mode page being requested
1870 * @buffer: request buffer (may not be smaller than eight bytes)
1871 * @len: length of request buffer.
1872 * @timeout: command timeout
1873 * @retries: number of retries before failing
1874 * @data: returns a structure abstracting the mode header data
1875 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1876 * must be SCSI_SENSE_BUFFERSIZE big.
1877 *
1878 * Returns zero if unsuccessful, or the header offset (either 4
1879 * or 8 depending on whether a six or ten byte command was
1880 * issued) if successful.
1881 */
1882int
1883scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1884 unsigned char *buffer, int len, int timeout, int retries,
1885 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1886{
1887 unsigned char cmd[12];
1888 int use_10_for_ms;
1889 int header_length;
1890 int result;
1891 struct scsi_sense_hdr my_sshdr;
1892
1893 memset(da…Large files files are truncated, but you can click here to view the full file