Loading...
1/*
2 * Copyright (C) 1999 Eric Youngdale
3 * Copyright (C) 2014 Christoph Hellwig
4 *
5 * SCSI queueing library.
6 * Initial versions: Eric Youngdale (eric@andante.org).
7 * Based upon conversations with large numbers
8 * of people at Linux Expo.
9 */
10
11#include <linux/bio.h>
12#include <linux/bitops.h>
13#include <linux/blkdev.h>
14#include <linux/completion.h>
15#include <linux/kernel.h>
16#include <linux/export.h>
17#include <linux/mempool.h>
18#include <linux/slab.h>
19#include <linux/init.h>
20#include <linux/pci.h>
21#include <linux/delay.h>
22#include <linux/hardirq.h>
23#include <linux/scatterlist.h>
24#include <linux/blk-mq.h>
25#include <linux/ratelimit.h>
26#include <asm/unaligned.h>
27
28#include <scsi/scsi.h>
29#include <scsi/scsi_cmnd.h>
30#include <scsi/scsi_dbg.h>
31#include <scsi/scsi_device.h>
32#include <scsi/scsi_driver.h>
33#include <scsi/scsi_eh.h>
34#include <scsi/scsi_host.h>
35#include <scsi/scsi_dh.h>
36
37#include <trace/events/scsi.h>
38
39#include "scsi_priv.h"
40#include "scsi_logging.h"
41
42
43#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
44#define SG_MEMPOOL_SIZE 2
45
46struct scsi_host_sg_pool {
47 size_t size;
48 char *name;
49 struct kmem_cache *slab;
50 mempool_t *pool;
51};
52
53#define SP(x) { .size = x, "sgpool-" __stringify(x) }
54#if (SCSI_MAX_SG_SEGMENTS < 32)
55#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
56#endif
57static struct scsi_host_sg_pool scsi_sg_pools[] = {
58 SP(8),
59 SP(16),
60#if (SCSI_MAX_SG_SEGMENTS > 32)
61 SP(32),
62#if (SCSI_MAX_SG_SEGMENTS > 64)
63 SP(64),
64#if (SCSI_MAX_SG_SEGMENTS > 128)
65 SP(128),
66#if (SCSI_MAX_SG_SEGMENTS > 256)
67#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
68#endif
69#endif
70#endif
71#endif
72 SP(SCSI_MAX_SG_SEGMENTS)
73};
74#undef SP
75
76struct kmem_cache *scsi_sdb_cache;
77
78/*
79 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
80 * not change behaviour from the previous unplug mechanism, experimentation
81 * may prove this needs changing.
82 */
83#define SCSI_QUEUE_DELAY 3
84
85static void
86scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
87{
88 struct Scsi_Host *host = cmd->device->host;
89 struct scsi_device *device = cmd->device;
90 struct scsi_target *starget = scsi_target(device);
91
92 /*
93 * Set the appropriate busy bit for the device/host.
94 *
95 * If the host/device isn't busy, assume that something actually
96 * completed, and that we should be able to queue a command now.
97 *
98 * Note that the prior mid-layer assumption that any host could
99 * always queue at least one command is now broken. The mid-layer
100 * will implement a user specifiable stall (see
101 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
102 * if a command is requeued with no other commands outstanding
103 * either for the device or for the host.
104 */
105 switch (reason) {
106 case SCSI_MLQUEUE_HOST_BUSY:
107 atomic_set(&host->host_blocked, host->max_host_blocked);
108 break;
109 case SCSI_MLQUEUE_DEVICE_BUSY:
110 case SCSI_MLQUEUE_EH_RETRY:
111 atomic_set(&device->device_blocked,
112 device->max_device_blocked);
113 break;
114 case SCSI_MLQUEUE_TARGET_BUSY:
115 atomic_set(&starget->target_blocked,
116 starget->max_target_blocked);
117 break;
118 }
119}
120
121static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
122{
123 struct scsi_device *sdev = cmd->device;
124 struct request_queue *q = cmd->request->q;
125
126 blk_mq_requeue_request(cmd->request);
127 blk_mq_kick_requeue_list(q);
128 put_device(&sdev->sdev_gendev);
129}
130
131/**
132 * __scsi_queue_insert - private queue insertion
133 * @cmd: The SCSI command being requeued
134 * @reason: The reason for the requeue
135 * @unbusy: Whether the queue should be unbusied
136 *
137 * This is a private queue insertion. The public interface
138 * scsi_queue_insert() always assumes the queue should be unbusied
139 * because it's always called before the completion. This function is
140 * for a requeue after completion, which should only occur in this
141 * file.
142 */
143static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
144{
145 struct scsi_device *device = cmd->device;
146 struct request_queue *q = device->request_queue;
147 unsigned long flags;
148
149 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
150 "Inserting command %p into mlqueue\n", cmd));
151
152 scsi_set_blocked(cmd, reason);
153
154 /*
155 * Decrement the counters, since these commands are no longer
156 * active on the host/device.
157 */
158 if (unbusy)
159 scsi_device_unbusy(device);
160
161 /*
162 * Requeue this command. It will go before all other commands
163 * that are already in the queue. Schedule requeue work under
164 * lock such that the kblockd_schedule_work() call happens
165 * before blk_cleanup_queue() finishes.
166 */
167 cmd->result = 0;
168 if (q->mq_ops) {
169 scsi_mq_requeue_cmd(cmd);
170 return;
171 }
172 spin_lock_irqsave(q->queue_lock, flags);
173 blk_requeue_request(q, cmd->request);
174 kblockd_schedule_work(&device->requeue_work);
175 spin_unlock_irqrestore(q->queue_lock, flags);
176}
177
178/*
179 * Function: scsi_queue_insert()
180 *
181 * Purpose: Insert a command in the midlevel queue.
182 *
183 * Arguments: cmd - command that we are adding to queue.
184 * reason - why we are inserting command to queue.
185 *
186 * Lock status: Assumed that lock is not held upon entry.
187 *
188 * Returns: Nothing.
189 *
190 * Notes: We do this for one of two cases. Either the host is busy
191 * and it cannot accept any more commands for the time being,
192 * or the device returned QUEUE_FULL and can accept no more
193 * commands.
194 * Notes: This could be called either from an interrupt context or a
195 * normal process context.
196 */
197void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
198{
199 __scsi_queue_insert(cmd, reason, 1);
200}
201/**
202 * scsi_execute - insert request and wait for the result
203 * @sdev: scsi device
204 * @cmd: scsi command
205 * @data_direction: data direction
206 * @buffer: data buffer
207 * @bufflen: len of buffer
208 * @sense: optional sense buffer
209 * @timeout: request timeout in seconds
210 * @retries: number of times to retry request
211 * @flags: or into request flags;
212 * @resid: optional residual length
213 *
214 * returns the req->errors value which is the scsi_cmnd result
215 * field.
216 */
217int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
218 int data_direction, void *buffer, unsigned bufflen,
219 unsigned char *sense, int timeout, int retries, u64 flags,
220 int *resid)
221{
222 struct request *req;
223 int write = (data_direction == DMA_TO_DEVICE);
224 int ret = DRIVER_ERROR << 24;
225
226 req = blk_get_request(sdev->request_queue, write, __GFP_RECLAIM);
227 if (IS_ERR(req))
228 return ret;
229 blk_rq_set_block_pc(req);
230
231 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
232 buffer, bufflen, __GFP_RECLAIM))
233 goto out;
234
235 req->cmd_len = COMMAND_SIZE(cmd[0]);
236 memcpy(req->cmd, cmd, req->cmd_len);
237 req->sense = sense;
238 req->sense_len = 0;
239 req->retries = retries;
240 req->timeout = timeout;
241 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
242
243 /*
244 * head injection *required* here otherwise quiesce won't work
245 */
246 blk_execute_rq(req->q, NULL, req, 1);
247
248 /*
249 * Some devices (USB mass-storage in particular) may transfer
250 * garbage data together with a residue indicating that the data
251 * is invalid. Prevent the garbage from being misinterpreted
252 * and prevent security leaks by zeroing out the excess data.
253 */
254 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
255 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
256
257 if (resid)
258 *resid = req->resid_len;
259 ret = req->errors;
260 out:
261 blk_put_request(req);
262
263 return ret;
264}
265EXPORT_SYMBOL(scsi_execute);
266
267int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
268 int data_direction, void *buffer, unsigned bufflen,
269 struct scsi_sense_hdr *sshdr, int timeout, int retries,
270 int *resid, u64 flags)
271{
272 char *sense = NULL;
273 int result;
274
275 if (sshdr) {
276 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
277 if (!sense)
278 return DRIVER_ERROR << 24;
279 }
280 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
281 sense, timeout, retries, flags, resid);
282 if (sshdr)
283 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
284
285 kfree(sense);
286 return result;
287}
288EXPORT_SYMBOL(scsi_execute_req_flags);
289
290/*
291 * Function: scsi_init_cmd_errh()
292 *
293 * Purpose: Initialize cmd fields related to error handling.
294 *
295 * Arguments: cmd - command that is ready to be queued.
296 *
297 * Notes: This function has the job of initializing a number of
298 * fields related to error handling. Typically this will
299 * be called once for each command, as required.
300 */
301static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
302{
303 cmd->serial_number = 0;
304 scsi_set_resid(cmd, 0);
305 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
306 if (cmd->cmd_len == 0)
307 cmd->cmd_len = scsi_command_size(cmd->cmnd);
308}
309
310void scsi_device_unbusy(struct scsi_device *sdev)
311{
312 struct Scsi_Host *shost = sdev->host;
313 struct scsi_target *starget = scsi_target(sdev);
314 unsigned long flags;
315
316 atomic_dec(&shost->host_busy);
317 if (starget->can_queue > 0)
318 atomic_dec(&starget->target_busy);
319
320 if (unlikely(scsi_host_in_recovery(shost) &&
321 (shost->host_failed || shost->host_eh_scheduled))) {
322 spin_lock_irqsave(shost->host_lock, flags);
323 scsi_eh_wakeup(shost);
324 spin_unlock_irqrestore(shost->host_lock, flags);
325 }
326
327 atomic_dec(&sdev->device_busy);
328}
329
330static void scsi_kick_queue(struct request_queue *q)
331{
332 if (q->mq_ops)
333 blk_mq_start_hw_queues(q);
334 else
335 blk_run_queue(q);
336}
337
338/*
339 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
340 * and call blk_run_queue for all the scsi_devices on the target -
341 * including current_sdev first.
342 *
343 * Called with *no* scsi locks held.
344 */
345static void scsi_single_lun_run(struct scsi_device *current_sdev)
346{
347 struct Scsi_Host *shost = current_sdev->host;
348 struct scsi_device *sdev, *tmp;
349 struct scsi_target *starget = scsi_target(current_sdev);
350 unsigned long flags;
351
352 spin_lock_irqsave(shost->host_lock, flags);
353 starget->starget_sdev_user = NULL;
354 spin_unlock_irqrestore(shost->host_lock, flags);
355
356 /*
357 * Call blk_run_queue for all LUNs on the target, starting with
358 * current_sdev. We race with others (to set starget_sdev_user),
359 * but in most cases, we will be first. Ideally, each LU on the
360 * target would get some limited time or requests on the target.
361 */
362 scsi_kick_queue(current_sdev->request_queue);
363
364 spin_lock_irqsave(shost->host_lock, flags);
365 if (starget->starget_sdev_user)
366 goto out;
367 list_for_each_entry_safe(sdev, tmp, &starget->devices,
368 same_target_siblings) {
369 if (sdev == current_sdev)
370 continue;
371 if (scsi_device_get(sdev))
372 continue;
373
374 spin_unlock_irqrestore(shost->host_lock, flags);
375 scsi_kick_queue(sdev->request_queue);
376 spin_lock_irqsave(shost->host_lock, flags);
377
378 scsi_device_put(sdev);
379 }
380 out:
381 spin_unlock_irqrestore(shost->host_lock, flags);
382}
383
384static inline bool scsi_device_is_busy(struct scsi_device *sdev)
385{
386 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
387 return true;
388 if (atomic_read(&sdev->device_blocked) > 0)
389 return true;
390 return false;
391}
392
393static inline bool scsi_target_is_busy(struct scsi_target *starget)
394{
395 if (starget->can_queue > 0) {
396 if (atomic_read(&starget->target_busy) >= starget->can_queue)
397 return true;
398 if (atomic_read(&starget->target_blocked) > 0)
399 return true;
400 }
401 return false;
402}
403
404static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
405{
406 if (shost->can_queue > 0 &&
407 atomic_read(&shost->host_busy) >= shost->can_queue)
408 return true;
409 if (atomic_read(&shost->host_blocked) > 0)
410 return true;
411 if (shost->host_self_blocked)
412 return true;
413 return false;
414}
415
416static void scsi_starved_list_run(struct Scsi_Host *shost)
417{
418 LIST_HEAD(starved_list);
419 struct scsi_device *sdev;
420 unsigned long flags;
421
422 spin_lock_irqsave(shost->host_lock, flags);
423 list_splice_init(&shost->starved_list, &starved_list);
424
425 while (!list_empty(&starved_list)) {
426 struct request_queue *slq;
427
428 /*
429 * As long as shost is accepting commands and we have
430 * starved queues, call blk_run_queue. scsi_request_fn
431 * drops the queue_lock and can add us back to the
432 * starved_list.
433 *
434 * host_lock protects the starved_list and starved_entry.
435 * scsi_request_fn must get the host_lock before checking
436 * or modifying starved_list or starved_entry.
437 */
438 if (scsi_host_is_busy(shost))
439 break;
440
441 sdev = list_entry(starved_list.next,
442 struct scsi_device, starved_entry);
443 list_del_init(&sdev->starved_entry);
444 if (scsi_target_is_busy(scsi_target(sdev))) {
445 list_move_tail(&sdev->starved_entry,
446 &shost->starved_list);
447 continue;
448 }
449
450 /*
451 * Once we drop the host lock, a racing scsi_remove_device()
452 * call may remove the sdev from the starved list and destroy
453 * it and the queue. Mitigate by taking a reference to the
454 * queue and never touching the sdev again after we drop the
455 * host lock. Note: if __scsi_remove_device() invokes
456 * blk_cleanup_queue() before the queue is run from this
457 * function then blk_run_queue() will return immediately since
458 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
459 */
460 slq = sdev->request_queue;
461 if (!blk_get_queue(slq))
462 continue;
463 spin_unlock_irqrestore(shost->host_lock, flags);
464
465 scsi_kick_queue(slq);
466 blk_put_queue(slq);
467
468 spin_lock_irqsave(shost->host_lock, flags);
469 }
470 /* put any unprocessed entries back */
471 list_splice(&starved_list, &shost->starved_list);
472 spin_unlock_irqrestore(shost->host_lock, flags);
473}
474
475/*
476 * Function: scsi_run_queue()
477 *
478 * Purpose: Select a proper request queue to serve next
479 *
480 * Arguments: q - last request's queue
481 *
482 * Returns: Nothing
483 *
484 * Notes: The previous command was completely finished, start
485 * a new one if possible.
486 */
487static void scsi_run_queue(struct request_queue *q)
488{
489 struct scsi_device *sdev = q->queuedata;
490
491 if (scsi_target(sdev)->single_lun)
492 scsi_single_lun_run(sdev);
493 if (!list_empty(&sdev->host->starved_list))
494 scsi_starved_list_run(sdev->host);
495
496 if (q->mq_ops)
497 blk_mq_start_stopped_hw_queues(q, false);
498 else
499 blk_run_queue(q);
500}
501
502void scsi_requeue_run_queue(struct work_struct *work)
503{
504 struct scsi_device *sdev;
505 struct request_queue *q;
506
507 sdev = container_of(work, struct scsi_device, requeue_work);
508 q = sdev->request_queue;
509 scsi_run_queue(q);
510}
511
512/*
513 * Function: scsi_requeue_command()
514 *
515 * Purpose: Handle post-processing of completed commands.
516 *
517 * Arguments: q - queue to operate on
518 * cmd - command that may need to be requeued.
519 *
520 * Returns: Nothing
521 *
522 * Notes: After command completion, there may be blocks left
523 * over which weren't finished by the previous command
524 * this can be for a number of reasons - the main one is
525 * I/O errors in the middle of the request, in which case
526 * we need to request the blocks that come after the bad
527 * sector.
528 * Notes: Upon return, cmd is a stale pointer.
529 */
530static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
531{
532 struct scsi_device *sdev = cmd->device;
533 struct request *req = cmd->request;
534 unsigned long flags;
535
536 spin_lock_irqsave(q->queue_lock, flags);
537 blk_unprep_request(req);
538 req->special = NULL;
539 scsi_put_command(cmd);
540 blk_requeue_request(q, req);
541 spin_unlock_irqrestore(q->queue_lock, flags);
542
543 scsi_run_queue(q);
544
545 put_device(&sdev->sdev_gendev);
546}
547
548void scsi_run_host_queues(struct Scsi_Host *shost)
549{
550 struct scsi_device *sdev;
551
552 shost_for_each_device(sdev, shost)
553 scsi_run_queue(sdev->request_queue);
554}
555
556static inline unsigned int scsi_sgtable_index(unsigned short nents)
557{
558 unsigned int index;
559
560 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
561
562 if (nents <= 8)
563 index = 0;
564 else
565 index = get_count_order(nents) - 3;
566
567 return index;
568}
569
570static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
571{
572 struct scsi_host_sg_pool *sgp;
573
574 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
575 mempool_free(sgl, sgp->pool);
576}
577
578static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
579{
580 struct scsi_host_sg_pool *sgp;
581
582 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
583 return mempool_alloc(sgp->pool, gfp_mask);
584}
585
586static void scsi_free_sgtable(struct scsi_data_buffer *sdb, bool mq)
587{
588 if (mq && sdb->table.orig_nents <= SCSI_MAX_SG_SEGMENTS)
589 return;
590 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, mq, scsi_sg_free);
591}
592
593static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents, bool mq)
594{
595 struct scatterlist *first_chunk = NULL;
596 int ret;
597
598 BUG_ON(!nents);
599
600 if (mq) {
601 if (nents <= SCSI_MAX_SG_SEGMENTS) {
602 sdb->table.nents = sdb->table.orig_nents = nents;
603 sg_init_table(sdb->table.sgl, nents);
604 return 0;
605 }
606 first_chunk = sdb->table.sgl;
607 }
608
609 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
610 first_chunk, GFP_ATOMIC, scsi_sg_alloc);
611 if (unlikely(ret))
612 scsi_free_sgtable(sdb, mq);
613 return ret;
614}
615
616static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
617{
618 if (cmd->request->cmd_type == REQ_TYPE_FS) {
619 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
620
621 if (drv->uninit_command)
622 drv->uninit_command(cmd);
623 }
624}
625
626static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
627{
628 if (cmd->sdb.table.nents)
629 scsi_free_sgtable(&cmd->sdb, true);
630 if (cmd->request->next_rq && cmd->request->next_rq->special)
631 scsi_free_sgtable(cmd->request->next_rq->special, true);
632 if (scsi_prot_sg_count(cmd))
633 scsi_free_sgtable(cmd->prot_sdb, true);
634}
635
636static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
637{
638 struct scsi_device *sdev = cmd->device;
639 struct Scsi_Host *shost = sdev->host;
640 unsigned long flags;
641
642 scsi_mq_free_sgtables(cmd);
643 scsi_uninit_cmd(cmd);
644
645 if (shost->use_cmd_list) {
646 BUG_ON(list_empty(&cmd->list));
647 spin_lock_irqsave(&sdev->list_lock, flags);
648 list_del_init(&cmd->list);
649 spin_unlock_irqrestore(&sdev->list_lock, flags);
650 }
651}
652
653/*
654 * Function: scsi_release_buffers()
655 *
656 * Purpose: Free resources allocate for a scsi_command.
657 *
658 * Arguments: cmd - command that we are bailing.
659 *
660 * Lock status: Assumed that no lock is held upon entry.
661 *
662 * Returns: Nothing
663 *
664 * Notes: In the event that an upper level driver rejects a
665 * command, we must release resources allocated during
666 * the __init_io() function. Primarily this would involve
667 * the scatter-gather table.
668 */
669static void scsi_release_buffers(struct scsi_cmnd *cmd)
670{
671 if (cmd->sdb.table.nents)
672 scsi_free_sgtable(&cmd->sdb, false);
673
674 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
675
676 if (scsi_prot_sg_count(cmd))
677 scsi_free_sgtable(cmd->prot_sdb, false);
678}
679
680static void scsi_release_bidi_buffers(struct scsi_cmnd *cmd)
681{
682 struct scsi_data_buffer *bidi_sdb = cmd->request->next_rq->special;
683
684 scsi_free_sgtable(bidi_sdb, false);
685 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
686 cmd->request->next_rq->special = NULL;
687}
688
689static bool scsi_end_request(struct request *req, int error,
690 unsigned int bytes, unsigned int bidi_bytes)
691{
692 struct scsi_cmnd *cmd = req->special;
693 struct scsi_device *sdev = cmd->device;
694 struct request_queue *q = sdev->request_queue;
695
696 if (blk_update_request(req, error, bytes))
697 return true;
698
699 /* Bidi request must be completed as a whole */
700 if (unlikely(bidi_bytes) &&
701 blk_update_request(req->next_rq, error, bidi_bytes))
702 return true;
703
704 if (blk_queue_add_random(q))
705 add_disk_randomness(req->rq_disk);
706
707 if (req->mq_ctx) {
708 /*
709 * In the MQ case the command gets freed by __blk_mq_end_request,
710 * so we have to do all cleanup that depends on it earlier.
711 *
712 * We also can't kick the queues from irq context, so we
713 * will have to defer it to a workqueue.
714 */
715 scsi_mq_uninit_cmd(cmd);
716
717 __blk_mq_end_request(req, error);
718
719 if (scsi_target(sdev)->single_lun ||
720 !list_empty(&sdev->host->starved_list))
721 kblockd_schedule_work(&sdev->requeue_work);
722 else
723 blk_mq_start_stopped_hw_queues(q, true);
724 } else {
725 unsigned long flags;
726
727 if (bidi_bytes)
728 scsi_release_bidi_buffers(cmd);
729
730 spin_lock_irqsave(q->queue_lock, flags);
731 blk_finish_request(req, error);
732 spin_unlock_irqrestore(q->queue_lock, flags);
733
734 scsi_release_buffers(cmd);
735
736 scsi_put_command(cmd);
737 scsi_run_queue(q);
738 }
739
740 put_device(&sdev->sdev_gendev);
741 return false;
742}
743
744/**
745 * __scsi_error_from_host_byte - translate SCSI error code into errno
746 * @cmd: SCSI command (unused)
747 * @result: scsi error code
748 *
749 * Translate SCSI error code into standard UNIX errno.
750 * Return values:
751 * -ENOLINK temporary transport failure
752 * -EREMOTEIO permanent target failure, do not retry
753 * -EBADE permanent nexus failure, retry on other path
754 * -ENOSPC No write space available
755 * -ENODATA Medium error
756 * -EIO unspecified I/O error
757 */
758static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
759{
760 int error = 0;
761
762 switch(host_byte(result)) {
763 case DID_TRANSPORT_FAILFAST:
764 error = -ENOLINK;
765 break;
766 case DID_TARGET_FAILURE:
767 set_host_byte(cmd, DID_OK);
768 error = -EREMOTEIO;
769 break;
770 case DID_NEXUS_FAILURE:
771 set_host_byte(cmd, DID_OK);
772 error = -EBADE;
773 break;
774 case DID_ALLOC_FAILURE:
775 set_host_byte(cmd, DID_OK);
776 error = -ENOSPC;
777 break;
778 case DID_MEDIUM_ERROR:
779 set_host_byte(cmd, DID_OK);
780 error = -ENODATA;
781 break;
782 default:
783 error = -EIO;
784 break;
785 }
786
787 return error;
788}
789
790/*
791 * Function: scsi_io_completion()
792 *
793 * Purpose: Completion processing for block device I/O requests.
794 *
795 * Arguments: cmd - command that is finished.
796 *
797 * Lock status: Assumed that no lock is held upon entry.
798 *
799 * Returns: Nothing
800 *
801 * Notes: We will finish off the specified number of sectors. If we
802 * are done, the command block will be released and the queue
803 * function will be goosed. If we are not done then we have to
804 * figure out what to do next:
805 *
806 * a) We can call scsi_requeue_command(). The request
807 * will be unprepared and put back on the queue. Then
808 * a new command will be created for it. This should
809 * be used if we made forward progress, or if we want
810 * to switch from READ(10) to READ(6) for example.
811 *
812 * b) We can call __scsi_queue_insert(). The request will
813 * be put back on the queue and retried using the same
814 * command as before, possibly after a delay.
815 *
816 * c) We can call scsi_end_request() with -EIO to fail
817 * the remainder of the request.
818 */
819void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
820{
821 int result = cmd->result;
822 struct request_queue *q = cmd->device->request_queue;
823 struct request *req = cmd->request;
824 int error = 0;
825 struct scsi_sense_hdr sshdr;
826 bool sense_valid = false;
827 int sense_deferred = 0, level = 0;
828 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
829 ACTION_DELAYED_RETRY} action;
830 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
831
832 if (result) {
833 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
834 if (sense_valid)
835 sense_deferred = scsi_sense_is_deferred(&sshdr);
836 }
837
838 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
839 if (result) {
840 if (sense_valid && req->sense) {
841 /*
842 * SG_IO wants current and deferred errors
843 */
844 int len = 8 + cmd->sense_buffer[7];
845
846 if (len > SCSI_SENSE_BUFFERSIZE)
847 len = SCSI_SENSE_BUFFERSIZE;
848 memcpy(req->sense, cmd->sense_buffer, len);
849 req->sense_len = len;
850 }
851 if (!sense_deferred)
852 error = __scsi_error_from_host_byte(cmd, result);
853 }
854 /*
855 * __scsi_error_from_host_byte may have reset the host_byte
856 */
857 req->errors = cmd->result;
858
859 req->resid_len = scsi_get_resid(cmd);
860
861 if (scsi_bidi_cmnd(cmd)) {
862 /*
863 * Bidi commands Must be complete as a whole,
864 * both sides at once.
865 */
866 req->next_rq->resid_len = scsi_in(cmd)->resid;
867 if (scsi_end_request(req, 0, blk_rq_bytes(req),
868 blk_rq_bytes(req->next_rq)))
869 BUG();
870 return;
871 }
872 } else if (blk_rq_bytes(req) == 0 && result && !sense_deferred) {
873 /*
874 * Certain non BLOCK_PC requests are commands that don't
875 * actually transfer anything (FLUSH), so cannot use
876 * good_bytes != blk_rq_bytes(req) as the signal for an error.
877 * This sets the error explicitly for the problem case.
878 */
879 error = __scsi_error_from_host_byte(cmd, result);
880 }
881
882 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
883 BUG_ON(blk_bidi_rq(req));
884
885 /*
886 * Next deal with any sectors which we were able to correctly
887 * handle.
888 */
889 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
890 "%u sectors total, %d bytes done.\n",
891 blk_rq_sectors(req), good_bytes));
892
893 /*
894 * Recovered errors need reporting, but they're always treated
895 * as success, so fiddle the result code here. For BLOCK_PC
896 * we already took a copy of the original into rq->errors which
897 * is what gets returned to the user
898 */
899 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
900 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
901 * print since caller wants ATA registers. Only occurs on
902 * SCSI ATA PASS_THROUGH commands when CK_COND=1
903 */
904 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
905 ;
906 else if (!(req->cmd_flags & REQ_QUIET))
907 scsi_print_sense(cmd);
908 result = 0;
909 /* BLOCK_PC may have set error */
910 error = 0;
911 }
912
913 /*
914 * If we finished all bytes in the request we are done now.
915 */
916 if (!scsi_end_request(req, error, good_bytes, 0))
917 return;
918
919 /*
920 * Kill remainder if no retrys.
921 */
922 if (error && scsi_noretry_cmd(cmd)) {
923 if (scsi_end_request(req, error, blk_rq_bytes(req), 0))
924 BUG();
925 return;
926 }
927
928 /*
929 * If there had been no error, but we have leftover bytes in the
930 * requeues just queue the command up again.
931 */
932 if (result == 0)
933 goto requeue;
934
935 error = __scsi_error_from_host_byte(cmd, result);
936
937 if (host_byte(result) == DID_RESET) {
938 /* Third party bus reset or reset for error recovery
939 * reasons. Just retry the command and see what
940 * happens.
941 */
942 action = ACTION_RETRY;
943 } else if (sense_valid && !sense_deferred) {
944 switch (sshdr.sense_key) {
945 case UNIT_ATTENTION:
946 if (cmd->device->removable) {
947 /* Detected disc change. Set a bit
948 * and quietly refuse further access.
949 */
950 cmd->device->changed = 1;
951 action = ACTION_FAIL;
952 } else {
953 /* Must have been a power glitch, or a
954 * bus reset. Could not have been a
955 * media change, so we just retry the
956 * command and see what happens.
957 */
958 action = ACTION_RETRY;
959 }
960 break;
961 case ILLEGAL_REQUEST:
962 /* If we had an ILLEGAL REQUEST returned, then
963 * we may have performed an unsupported
964 * command. The only thing this should be
965 * would be a ten byte read where only a six
966 * byte read was supported. Also, on a system
967 * where READ CAPACITY failed, we may have
968 * read past the end of the disk.
969 */
970 if ((cmd->device->use_10_for_rw &&
971 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
972 (cmd->cmnd[0] == READ_10 ||
973 cmd->cmnd[0] == WRITE_10)) {
974 /* This will issue a new 6-byte command. */
975 cmd->device->use_10_for_rw = 0;
976 action = ACTION_REPREP;
977 } else if (sshdr.asc == 0x10) /* DIX */ {
978 action = ACTION_FAIL;
979 error = -EILSEQ;
980 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
981 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
982 action = ACTION_FAIL;
983 error = -EREMOTEIO;
984 } else
985 action = ACTION_FAIL;
986 break;
987 case ABORTED_COMMAND:
988 action = ACTION_FAIL;
989 if (sshdr.asc == 0x10) /* DIF */
990 error = -EILSEQ;
991 break;
992 case NOT_READY:
993 /* If the device is in the process of becoming
994 * ready, or has a temporary blockage, retry.
995 */
996 if (sshdr.asc == 0x04) {
997 switch (sshdr.ascq) {
998 case 0x01: /* becoming ready */
999 case 0x04: /* format in progress */
1000 case 0x05: /* rebuild in progress */
1001 case 0x06: /* recalculation in progress */
1002 case 0x07: /* operation in progress */
1003 case 0x08: /* Long write in progress */
1004 case 0x09: /* self test in progress */
1005 case 0x14: /* space allocation in progress */
1006 action = ACTION_DELAYED_RETRY;
1007 break;
1008 default:
1009 action = ACTION_FAIL;
1010 break;
1011 }
1012 } else
1013 action = ACTION_FAIL;
1014 break;
1015 case VOLUME_OVERFLOW:
1016 /* See SSC3rXX or current. */
1017 action = ACTION_FAIL;
1018 break;
1019 default:
1020 action = ACTION_FAIL;
1021 break;
1022 }
1023 } else
1024 action = ACTION_FAIL;
1025
1026 if (action != ACTION_FAIL &&
1027 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
1028 action = ACTION_FAIL;
1029
1030 switch (action) {
1031 case ACTION_FAIL:
1032 /* Give up and fail the remainder of the request */
1033 if (!(req->cmd_flags & REQ_QUIET)) {
1034 static DEFINE_RATELIMIT_STATE(_rs,
1035 DEFAULT_RATELIMIT_INTERVAL,
1036 DEFAULT_RATELIMIT_BURST);
1037
1038 if (unlikely(scsi_logging_level))
1039 level = SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
1040 SCSI_LOG_MLCOMPLETE_BITS);
1041
1042 /*
1043 * if logging is enabled the failure will be printed
1044 * in scsi_log_completion(), so avoid duplicate messages
1045 */
1046 if (!level && __ratelimit(&_rs)) {
1047 scsi_print_result(cmd, NULL, FAILED);
1048 if (driver_byte(result) & DRIVER_SENSE)
1049 scsi_print_sense(cmd);
1050 scsi_print_command(cmd);
1051 }
1052 }
1053 if (!scsi_end_request(req, error, blk_rq_err_bytes(req), 0))
1054 return;
1055 /*FALLTHRU*/
1056 case ACTION_REPREP:
1057 requeue:
1058 /* Unprep the request and put it back at the head of the queue.
1059 * A new command will be prepared and issued.
1060 */
1061 if (q->mq_ops) {
1062 cmd->request->cmd_flags &= ~REQ_DONTPREP;
1063 scsi_mq_uninit_cmd(cmd);
1064 scsi_mq_requeue_cmd(cmd);
1065 } else {
1066 scsi_release_buffers(cmd);
1067 scsi_requeue_command(q, cmd);
1068 }
1069 break;
1070 case ACTION_RETRY:
1071 /* Retry the same command immediately */
1072 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1073 break;
1074 case ACTION_DELAYED_RETRY:
1075 /* Retry the same command after a delay */
1076 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1077 break;
1078 }
1079}
1080
1081static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb)
1082{
1083 int count;
1084
1085 /*
1086 * If sg table allocation fails, requeue request later.
1087 */
1088 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1089 req->mq_ctx != NULL)))
1090 return BLKPREP_DEFER;
1091
1092 /*
1093 * Next, walk the list, and fill in the addresses and sizes of
1094 * each segment.
1095 */
1096 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1097 BUG_ON(count > sdb->table.nents);
1098 sdb->table.nents = count;
1099 sdb->length = blk_rq_bytes(req);
1100 return BLKPREP_OK;
1101}
1102
1103/*
1104 * Function: scsi_init_io()
1105 *
1106 * Purpose: SCSI I/O initialize function.
1107 *
1108 * Arguments: cmd - Command descriptor we wish to initialize
1109 *
1110 * Returns: 0 on success
1111 * BLKPREP_DEFER if the failure is retryable
1112 * BLKPREP_KILL if the failure is fatal
1113 */
1114int scsi_init_io(struct scsi_cmnd *cmd)
1115{
1116 struct scsi_device *sdev = cmd->device;
1117 struct request *rq = cmd->request;
1118 bool is_mq = (rq->mq_ctx != NULL);
1119 int error;
1120
1121 BUG_ON(!rq->nr_phys_segments);
1122
1123 error = scsi_init_sgtable(rq, &cmd->sdb);
1124 if (error)
1125 goto err_exit;
1126
1127 if (blk_bidi_rq(rq)) {
1128 if (!rq->q->mq_ops) {
1129 struct scsi_data_buffer *bidi_sdb =
1130 kmem_cache_zalloc(scsi_sdb_cache, GFP_ATOMIC);
1131 if (!bidi_sdb) {
1132 error = BLKPREP_DEFER;
1133 goto err_exit;
1134 }
1135
1136 rq->next_rq->special = bidi_sdb;
1137 }
1138
1139 error = scsi_init_sgtable(rq->next_rq, rq->next_rq->special);
1140 if (error)
1141 goto err_exit;
1142 }
1143
1144 if (blk_integrity_rq(rq)) {
1145 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1146 int ivecs, count;
1147
1148 if (prot_sdb == NULL) {
1149 /*
1150 * This can happen if someone (e.g. multipath)
1151 * queues a command to a device on an adapter
1152 * that does not support DIX.
1153 */
1154 WARN_ON_ONCE(1);
1155 error = BLKPREP_KILL;
1156 goto err_exit;
1157 }
1158
1159 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1160
1161 if (scsi_alloc_sgtable(prot_sdb, ivecs, is_mq)) {
1162 error = BLKPREP_DEFER;
1163 goto err_exit;
1164 }
1165
1166 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1167 prot_sdb->table.sgl);
1168 BUG_ON(unlikely(count > ivecs));
1169 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1170
1171 cmd->prot_sdb = prot_sdb;
1172 cmd->prot_sdb->table.nents = count;
1173 }
1174
1175 return BLKPREP_OK;
1176err_exit:
1177 if (is_mq) {
1178 scsi_mq_free_sgtables(cmd);
1179 } else {
1180 scsi_release_buffers(cmd);
1181 cmd->request->special = NULL;
1182 scsi_put_command(cmd);
1183 put_device(&sdev->sdev_gendev);
1184 }
1185 return error;
1186}
1187EXPORT_SYMBOL(scsi_init_io);
1188
1189static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1190 struct request *req)
1191{
1192 struct scsi_cmnd *cmd;
1193
1194 if (!req->special) {
1195 /* Bail if we can't get a reference to the device */
1196 if (!get_device(&sdev->sdev_gendev))
1197 return NULL;
1198
1199 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1200 if (unlikely(!cmd)) {
1201 put_device(&sdev->sdev_gendev);
1202 return NULL;
1203 }
1204 req->special = cmd;
1205 } else {
1206 cmd = req->special;
1207 }
1208
1209 /* pull a tag out of the request if we have one */
1210 cmd->tag = req->tag;
1211 cmd->request = req;
1212
1213 cmd->cmnd = req->cmd;
1214 cmd->prot_op = SCSI_PROT_NORMAL;
1215
1216 return cmd;
1217}
1218
1219static int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1220{
1221 struct scsi_cmnd *cmd = req->special;
1222
1223 /*
1224 * BLOCK_PC requests may transfer data, in which case they must
1225 * a bio attached to them. Or they might contain a SCSI command
1226 * that does not transfer data, in which case they may optionally
1227 * submit a request without an attached bio.
1228 */
1229 if (req->bio) {
1230 int ret = scsi_init_io(cmd);
1231 if (unlikely(ret))
1232 return ret;
1233 } else {
1234 BUG_ON(blk_rq_bytes(req));
1235
1236 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1237 }
1238
1239 cmd->cmd_len = req->cmd_len;
1240 cmd->transfersize = blk_rq_bytes(req);
1241 cmd->allowed = req->retries;
1242 return BLKPREP_OK;
1243}
1244
1245/*
1246 * Setup a REQ_TYPE_FS command. These are simple request from filesystems
1247 * that still need to be translated to SCSI CDBs from the ULD.
1248 */
1249static int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1250{
1251 struct scsi_cmnd *cmd = req->special;
1252
1253 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1254 int ret = sdev->handler->prep_fn(sdev, req);
1255 if (ret != BLKPREP_OK)
1256 return ret;
1257 }
1258
1259 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1260 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1261}
1262
1263static int scsi_setup_cmnd(struct scsi_device *sdev, struct request *req)
1264{
1265 struct scsi_cmnd *cmd = req->special;
1266
1267 if (!blk_rq_bytes(req))
1268 cmd->sc_data_direction = DMA_NONE;
1269 else if (rq_data_dir(req) == WRITE)
1270 cmd->sc_data_direction = DMA_TO_DEVICE;
1271 else
1272 cmd->sc_data_direction = DMA_FROM_DEVICE;
1273
1274 switch (req->cmd_type) {
1275 case REQ_TYPE_FS:
1276 return scsi_setup_fs_cmnd(sdev, req);
1277 case REQ_TYPE_BLOCK_PC:
1278 return scsi_setup_blk_pc_cmnd(sdev, req);
1279 default:
1280 return BLKPREP_KILL;
1281 }
1282}
1283
1284static int
1285scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1286{
1287 int ret = BLKPREP_OK;
1288
1289 /*
1290 * If the device is not in running state we will reject some
1291 * or all commands.
1292 */
1293 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1294 switch (sdev->sdev_state) {
1295 case SDEV_OFFLINE:
1296 case SDEV_TRANSPORT_OFFLINE:
1297 /*
1298 * If the device is offline we refuse to process any
1299 * commands. The device must be brought online
1300 * before trying any recovery commands.
1301 */
1302 sdev_printk(KERN_ERR, sdev,
1303 "rejecting I/O to offline device\n");
1304 ret = BLKPREP_KILL;
1305 break;
1306 case SDEV_DEL:
1307 /*
1308 * If the device is fully deleted, we refuse to
1309 * process any commands as well.
1310 */
1311 sdev_printk(KERN_ERR, sdev,
1312 "rejecting I/O to dead device\n");
1313 ret = BLKPREP_KILL;
1314 break;
1315 case SDEV_BLOCK:
1316 case SDEV_CREATED_BLOCK:
1317 ret = BLKPREP_DEFER;
1318 break;
1319 case SDEV_QUIESCE:
1320 /*
1321 * If the devices is blocked we defer normal commands.
1322 */
1323 if (!(req->cmd_flags & REQ_PREEMPT))
1324 ret = BLKPREP_DEFER;
1325 break;
1326 default:
1327 /*
1328 * For any other not fully online state we only allow
1329 * special commands. In particular any user initiated
1330 * command is not allowed.
1331 */
1332 if (!(req->cmd_flags & REQ_PREEMPT))
1333 ret = BLKPREP_KILL;
1334 break;
1335 }
1336 }
1337 return ret;
1338}
1339
1340static int
1341scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1342{
1343 struct scsi_device *sdev = q->queuedata;
1344
1345 switch (ret) {
1346 case BLKPREP_KILL:
1347 case BLKPREP_INVALID:
1348 req->errors = DID_NO_CONNECT << 16;
1349 /* release the command and kill it */
1350 if (req->special) {
1351 struct scsi_cmnd *cmd = req->special;
1352 scsi_release_buffers(cmd);
1353 scsi_put_command(cmd);
1354 put_device(&sdev->sdev_gendev);
1355 req->special = NULL;
1356 }
1357 break;
1358 case BLKPREP_DEFER:
1359 /*
1360 * If we defer, the blk_peek_request() returns NULL, but the
1361 * queue must be restarted, so we schedule a callback to happen
1362 * shortly.
1363 */
1364 if (atomic_read(&sdev->device_busy) == 0)
1365 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1366 break;
1367 default:
1368 req->cmd_flags |= REQ_DONTPREP;
1369 }
1370
1371 return ret;
1372}
1373
1374static int scsi_prep_fn(struct request_queue *q, struct request *req)
1375{
1376 struct scsi_device *sdev = q->queuedata;
1377 struct scsi_cmnd *cmd;
1378 int ret;
1379
1380 ret = scsi_prep_state_check(sdev, req);
1381 if (ret != BLKPREP_OK)
1382 goto out;
1383
1384 cmd = scsi_get_cmd_from_req(sdev, req);
1385 if (unlikely(!cmd)) {
1386 ret = BLKPREP_DEFER;
1387 goto out;
1388 }
1389
1390 ret = scsi_setup_cmnd(sdev, req);
1391out:
1392 return scsi_prep_return(q, req, ret);
1393}
1394
1395static void scsi_unprep_fn(struct request_queue *q, struct request *req)
1396{
1397 scsi_uninit_cmd(req->special);
1398}
1399
1400/*
1401 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1402 * return 0.
1403 *
1404 * Called with the queue_lock held.
1405 */
1406static inline int scsi_dev_queue_ready(struct request_queue *q,
1407 struct scsi_device *sdev)
1408{
1409 unsigned int busy;
1410
1411 busy = atomic_inc_return(&sdev->device_busy) - 1;
1412 if (atomic_read(&sdev->device_blocked)) {
1413 if (busy)
1414 goto out_dec;
1415
1416 /*
1417 * unblock after device_blocked iterates to zero
1418 */
1419 if (atomic_dec_return(&sdev->device_blocked) > 0) {
1420 /*
1421 * For the MQ case we take care of this in the caller.
1422 */
1423 if (!q->mq_ops)
1424 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1425 goto out_dec;
1426 }
1427 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1428 "unblocking device at zero depth\n"));
1429 }
1430
1431 if (busy >= sdev->queue_depth)
1432 goto out_dec;
1433
1434 return 1;
1435out_dec:
1436 atomic_dec(&sdev->device_busy);
1437 return 0;
1438}
1439
1440/*
1441 * scsi_target_queue_ready: checks if there we can send commands to target
1442 * @sdev: scsi device on starget to check.
1443 */
1444static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1445 struct scsi_device *sdev)
1446{
1447 struct scsi_target *starget = scsi_target(sdev);
1448 unsigned int busy;
1449
1450 if (starget->single_lun) {
1451 spin_lock_irq(shost->host_lock);
1452 if (starget->starget_sdev_user &&
1453 starget->starget_sdev_user != sdev) {
1454 spin_unlock_irq(shost->host_lock);
1455 return 0;
1456 }
1457 starget->starget_sdev_user = sdev;
1458 spin_unlock_irq(shost->host_lock);
1459 }
1460
1461 if (starget->can_queue <= 0)
1462 return 1;
1463
1464 busy = atomic_inc_return(&starget->target_busy) - 1;
1465 if (atomic_read(&starget->target_blocked) > 0) {
1466 if (busy)
1467 goto starved;
1468
1469 /*
1470 * unblock after target_blocked iterates to zero
1471 */
1472 if (atomic_dec_return(&starget->target_blocked) > 0)
1473 goto out_dec;
1474
1475 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1476 "unblocking target at zero depth\n"));
1477 }
1478
1479 if (busy >= starget->can_queue)
1480 goto starved;
1481
1482 return 1;
1483
1484starved:
1485 spin_lock_irq(shost->host_lock);
1486 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1487 spin_unlock_irq(shost->host_lock);
1488out_dec:
1489 if (starget->can_queue > 0)
1490 atomic_dec(&starget->target_busy);
1491 return 0;
1492}
1493
1494/*
1495 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1496 * return 0. We must end up running the queue again whenever 0 is
1497 * returned, else IO can hang.
1498 */
1499static inline int scsi_host_queue_ready(struct request_queue *q,
1500 struct Scsi_Host *shost,
1501 struct scsi_device *sdev)
1502{
1503 unsigned int busy;
1504
1505 if (scsi_host_in_recovery(shost))
1506 return 0;
1507
1508 busy = atomic_inc_return(&shost->host_busy) - 1;
1509 if (atomic_read(&shost->host_blocked) > 0) {
1510 if (busy)
1511 goto starved;
1512
1513 /*
1514 * unblock after host_blocked iterates to zero
1515 */
1516 if (atomic_dec_return(&shost->host_blocked) > 0)
1517 goto out_dec;
1518
1519 SCSI_LOG_MLQUEUE(3,
1520 shost_printk(KERN_INFO, shost,
1521 "unblocking host at zero depth\n"));
1522 }
1523
1524 if (shost->can_queue > 0 && busy >= shost->can_queue)
1525 goto starved;
1526 if (shost->host_self_blocked)
1527 goto starved;
1528
1529 /* We're OK to process the command, so we can't be starved */
1530 if (!list_empty(&sdev->starved_entry)) {
1531 spin_lock_irq(shost->host_lock);
1532 if (!list_empty(&sdev->starved_entry))
1533 list_del_init(&sdev->starved_entry);
1534 spin_unlock_irq(shost->host_lock);
1535 }
1536
1537 return 1;
1538
1539starved:
1540 spin_lock_irq(shost->host_lock);
1541 if (list_empty(&sdev->starved_entry))
1542 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1543 spin_unlock_irq(shost->host_lock);
1544out_dec:
1545 atomic_dec(&shost->host_busy);
1546 return 0;
1547}
1548
1549/*
1550 * Busy state exporting function for request stacking drivers.
1551 *
1552 * For efficiency, no lock is taken to check the busy state of
1553 * shost/starget/sdev, since the returned value is not guaranteed and
1554 * may be changed after request stacking drivers call the function,
1555 * regardless of taking lock or not.
1556 *
1557 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1558 * needs to return 'not busy'. Otherwise, request stacking drivers
1559 * may hold requests forever.
1560 */
1561static int scsi_lld_busy(struct request_queue *q)
1562{
1563 struct scsi_device *sdev = q->queuedata;
1564 struct Scsi_Host *shost;
1565
1566 if (blk_queue_dying(q))
1567 return 0;
1568
1569 shost = sdev->host;
1570
1571 /*
1572 * Ignore host/starget busy state.
1573 * Since block layer does not have a concept of fairness across
1574 * multiple queues, congestion of host/starget needs to be handled
1575 * in SCSI layer.
1576 */
1577 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1578 return 1;
1579
1580 return 0;
1581}
1582
1583/*
1584 * Kill a request for a dead device
1585 */
1586static void scsi_kill_request(struct request *req, struct request_queue *q)
1587{
1588 struct scsi_cmnd *cmd = req->special;
1589 struct scsi_device *sdev;
1590 struct scsi_target *starget;
1591 struct Scsi_Host *shost;
1592
1593 blk_start_request(req);
1594
1595 scmd_printk(KERN_INFO, cmd, "killing request\n");
1596
1597 sdev = cmd->device;
1598 starget = scsi_target(sdev);
1599 shost = sdev->host;
1600 scsi_init_cmd_errh(cmd);
1601 cmd->result = DID_NO_CONNECT << 16;
1602 atomic_inc(&cmd->device->iorequest_cnt);
1603
1604 /*
1605 * SCSI request completion path will do scsi_device_unbusy(),
1606 * bump busy counts. To bump the counters, we need to dance
1607 * with the locks as normal issue path does.
1608 */
1609 atomic_inc(&sdev->device_busy);
1610 atomic_inc(&shost->host_busy);
1611 if (starget->can_queue > 0)
1612 atomic_inc(&starget->target_busy);
1613
1614 blk_complete_request(req);
1615}
1616
1617static void scsi_softirq_done(struct request *rq)
1618{
1619 struct scsi_cmnd *cmd = rq->special;
1620 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1621 int disposition;
1622
1623 INIT_LIST_HEAD(&cmd->eh_entry);
1624
1625 atomic_inc(&cmd->device->iodone_cnt);
1626 if (cmd->result)
1627 atomic_inc(&cmd->device->ioerr_cnt);
1628
1629 disposition = scsi_decide_disposition(cmd);
1630 if (disposition != SUCCESS &&
1631 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1632 sdev_printk(KERN_ERR, cmd->device,
1633 "timing out command, waited %lus\n",
1634 wait_for/HZ);
1635 disposition = SUCCESS;
1636 }
1637
1638 scsi_log_completion(cmd, disposition);
1639
1640 switch (disposition) {
1641 case SUCCESS:
1642 scsi_finish_command(cmd);
1643 break;
1644 case NEEDS_RETRY:
1645 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1646 break;
1647 case ADD_TO_MLQUEUE:
1648 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1649 break;
1650 default:
1651 if (!scsi_eh_scmd_add(cmd, 0))
1652 scsi_finish_command(cmd);
1653 }
1654}
1655
1656/**
1657 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1658 * @cmd: command block we are dispatching.
1659 *
1660 * Return: nonzero return request was rejected and device's queue needs to be
1661 * plugged.
1662 */
1663static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1664{
1665 struct Scsi_Host *host = cmd->device->host;
1666 int rtn = 0;
1667
1668 atomic_inc(&cmd->device->iorequest_cnt);
1669
1670 /* check if the device is still usable */
1671 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1672 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1673 * returns an immediate error upwards, and signals
1674 * that the device is no longer present */
1675 cmd->result = DID_NO_CONNECT << 16;
1676 goto done;
1677 }
1678
1679 /* Check to see if the scsi lld made this device blocked. */
1680 if (unlikely(scsi_device_blocked(cmd->device))) {
1681 /*
1682 * in blocked state, the command is just put back on
1683 * the device queue. The suspend state has already
1684 * blocked the queue so future requests should not
1685 * occur until the device transitions out of the
1686 * suspend state.
1687 */
1688 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1689 "queuecommand : device blocked\n"));
1690 return SCSI_MLQUEUE_DEVICE_BUSY;
1691 }
1692
1693 /* Store the LUN value in cmnd, if needed. */
1694 if (cmd->device->lun_in_cdb)
1695 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1696 (cmd->device->lun << 5 & 0xe0);
1697
1698 scsi_log_send(cmd);
1699
1700 /*
1701 * Before we queue this command, check if the command
1702 * length exceeds what the host adapter can handle.
1703 */
1704 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1705 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1706 "queuecommand : command too long. "
1707 "cdb_size=%d host->max_cmd_len=%d\n",
1708 cmd->cmd_len, cmd->device->host->max_cmd_len));
1709 cmd->result = (DID_ABORT << 16);
1710 goto done;
1711 }
1712
1713 if (unlikely(host->shost_state == SHOST_DEL)) {
1714 cmd->result = (DID_NO_CONNECT << 16);
1715 goto done;
1716
1717 }
1718
1719 trace_scsi_dispatch_cmd_start(cmd);
1720 rtn = host->hostt->queuecommand(host, cmd);
1721 if (rtn) {
1722 trace_scsi_dispatch_cmd_error(cmd, rtn);
1723 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1724 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1725 rtn = SCSI_MLQUEUE_HOST_BUSY;
1726
1727 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1728 "queuecommand : request rejected\n"));
1729 }
1730
1731 return rtn;
1732 done:
1733 cmd->scsi_done(cmd);
1734 return 0;
1735}
1736
1737/**
1738 * scsi_done - Invoke completion on finished SCSI command.
1739 * @cmd: The SCSI Command for which a low-level device driver (LLDD) gives
1740 * ownership back to SCSI Core -- i.e. the LLDD has finished with it.
1741 *
1742 * Description: This function is the mid-level's (SCSI Core) interrupt routine,
1743 * which regains ownership of the SCSI command (de facto) from a LLDD, and
1744 * calls blk_complete_request() for further processing.
1745 *
1746 * This function is interrupt context safe.
1747 */
1748static void scsi_done(struct scsi_cmnd *cmd)
1749{
1750 trace_scsi_dispatch_cmd_done(cmd);
1751 blk_complete_request(cmd->request);
1752}
1753
1754/*
1755 * Function: scsi_request_fn()
1756 *
1757 * Purpose: Main strategy routine for SCSI.
1758 *
1759 * Arguments: q - Pointer to actual queue.
1760 *
1761 * Returns: Nothing
1762 *
1763 * Lock status: IO request lock assumed to be held when called.
1764 */
1765static void scsi_request_fn(struct request_queue *q)
1766 __releases(q->queue_lock)
1767 __acquires(q->queue_lock)
1768{
1769 struct scsi_device *sdev = q->queuedata;
1770 struct Scsi_Host *shost;
1771 struct scsi_cmnd *cmd;
1772 struct request *req;
1773
1774 /*
1775 * To start with, we keep looping until the queue is empty, or until
1776 * the host is no longer able to accept any more requests.
1777 */
1778 shost = sdev->host;
1779 for (;;) {
1780 int rtn;
1781 /*
1782 * get next queueable request. We do this early to make sure
1783 * that the request is fully prepared even if we cannot
1784 * accept it.
1785 */
1786 req = blk_peek_request(q);
1787 if (!req)
1788 break;
1789
1790 if (unlikely(!scsi_device_online(sdev))) {
1791 sdev_printk(KERN_ERR, sdev,
1792 "rejecting I/O to offline device\n");
1793 scsi_kill_request(req, q);
1794 continue;
1795 }
1796
1797 if (!scsi_dev_queue_ready(q, sdev))
1798 break;
1799
1800 /*
1801 * Remove the request from the request list.
1802 */
1803 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1804 blk_start_request(req);
1805
1806 spin_unlock_irq(q->queue_lock);
1807 cmd = req->special;
1808 if (unlikely(cmd == NULL)) {
1809 printk(KERN_CRIT "impossible request in %s.\n"
1810 "please mail a stack trace to "
1811 "linux-scsi@vger.kernel.org\n",
1812 __func__);
1813 blk_dump_rq_flags(req, "foo");
1814 BUG();
1815 }
1816
1817 /*
1818 * We hit this when the driver is using a host wide
1819 * tag map. For device level tag maps the queue_depth check
1820 * in the device ready fn would prevent us from trying
1821 * to allocate a tag. Since the map is a shared host resource
1822 * we add the dev to the starved list so it eventually gets
1823 * a run when a tag is freed.
1824 */
1825 if (blk_queue_tagged(q) && !(req->cmd_flags & REQ_QUEUED)) {
1826 spin_lock_irq(shost->host_lock);
1827 if (list_empty(&sdev->starved_entry))
1828 list_add_tail(&sdev->starved_entry,
1829 &shost->starved_list);
1830 spin_unlock_irq(shost->host_lock);
1831 goto not_ready;
1832 }
1833
1834 if (!scsi_target_queue_ready(shost, sdev))
1835 goto not_ready;
1836
1837 if (!scsi_host_queue_ready(q, shost, sdev))
1838 goto host_not_ready;
1839
1840 if (sdev->simple_tags)
1841 cmd->flags |= SCMD_TAGGED;
1842 else
1843 cmd->flags &= ~SCMD_TAGGED;
1844
1845 /*
1846 * Finally, initialize any error handling parameters, and set up
1847 * the timers for timeouts.
1848 */
1849 scsi_init_cmd_errh(cmd);
1850
1851 /*
1852 * Dispatch the command to the low-level driver.
1853 */
1854 cmd->scsi_done = scsi_done;
1855 rtn = scsi_dispatch_cmd(cmd);
1856 if (rtn) {
1857 scsi_queue_insert(cmd, rtn);
1858 spin_lock_irq(q->queue_lock);
1859 goto out_delay;
1860 }
1861 spin_lock_irq(q->queue_lock);
1862 }
1863
1864 return;
1865
1866 host_not_ready:
1867 if (scsi_target(sdev)->can_queue > 0)
1868 atomic_dec(&scsi_target(sdev)->target_busy);
1869 not_ready:
1870 /*
1871 * lock q, handle tag, requeue req, and decrement device_busy. We
1872 * must return with queue_lock held.
1873 *
1874 * Decrementing device_busy without checking it is OK, as all such
1875 * cases (host limits or settings) should run the queue at some
1876 * later time.
1877 */
1878 spin_lock_irq(q->queue_lock);
1879 blk_requeue_request(q, req);
1880 atomic_dec(&sdev->device_busy);
1881out_delay:
1882 if (!atomic_read(&sdev->device_busy) && !scsi_device_blocked(sdev))
1883 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1884}
1885
1886static inline int prep_to_mq(int ret)
1887{
1888 switch (ret) {
1889 case BLKPREP_OK:
1890 return 0;
1891 case BLKPREP_DEFER:
1892 return BLK_MQ_RQ_QUEUE_BUSY;
1893 default:
1894 return BLK_MQ_RQ_QUEUE_ERROR;
1895 }
1896}
1897
1898static int scsi_mq_prep_fn(struct request *req)
1899{
1900 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1901 struct scsi_device *sdev = req->q->queuedata;
1902 struct Scsi_Host *shost = sdev->host;
1903 unsigned char *sense_buf = cmd->sense_buffer;
1904 struct scatterlist *sg;
1905
1906 memset(cmd, 0, sizeof(struct scsi_cmnd));
1907
1908 req->special = cmd;
1909
1910 cmd->request = req;
1911 cmd->device = sdev;
1912 cmd->sense_buffer = sense_buf;
1913
1914 cmd->tag = req->tag;
1915
1916 cmd->cmnd = req->cmd;
1917 cmd->prot_op = SCSI_PROT_NORMAL;
1918
1919 INIT_LIST_HEAD(&cmd->list);
1920 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1921 cmd->jiffies_at_alloc = jiffies;
1922
1923 if (shost->use_cmd_list) {
1924 spin_lock_irq(&sdev->list_lock);
1925 list_add_tail(&cmd->list, &sdev->cmd_list);
1926 spin_unlock_irq(&sdev->list_lock);
1927 }
1928
1929 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1930 cmd->sdb.table.sgl = sg;
1931
1932 if (scsi_host_get_prot(shost)) {
1933 cmd->prot_sdb = (void *)sg +
1934 min_t(unsigned int,
1935 shost->sg_tablesize, SCSI_MAX_SG_SEGMENTS) *
1936 sizeof(struct scatterlist);
1937 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1938
1939 cmd->prot_sdb->table.sgl =
1940 (struct scatterlist *)(cmd->prot_sdb + 1);
1941 }
1942
1943 if (blk_bidi_rq(req)) {
1944 struct request *next_rq = req->next_rq;
1945 struct scsi_data_buffer *bidi_sdb = blk_mq_rq_to_pdu(next_rq);
1946
1947 memset(bidi_sdb, 0, sizeof(struct scsi_data_buffer));
1948 bidi_sdb->table.sgl =
1949 (struct scatterlist *)(bidi_sdb + 1);
1950
1951 next_rq->special = bidi_sdb;
1952 }
1953
1954 blk_mq_start_request(req);
1955
1956 return scsi_setup_cmnd(sdev, req);
1957}
1958
1959static void scsi_mq_done(struct scsi_cmnd *cmd)
1960{
1961 trace_scsi_dispatch_cmd_done(cmd);
1962 blk_mq_complete_request(cmd->request, cmd->request->errors);
1963}
1964
1965static int scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1966 const struct blk_mq_queue_data *bd)
1967{
1968 struct request *req = bd->rq;
1969 struct request_queue *q = req->q;
1970 struct scsi_device *sdev = q->queuedata;
1971 struct Scsi_Host *shost = sdev->host;
1972 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1973 int ret;
1974 int reason;
1975
1976 ret = prep_to_mq(scsi_prep_state_check(sdev, req));
1977 if (ret)
1978 goto out;
1979
1980 ret = BLK_MQ_RQ_QUEUE_BUSY;
1981 if (!get_device(&sdev->sdev_gendev))
1982 goto out;
1983
1984 if (!scsi_dev_queue_ready(q, sdev))
1985 goto out_put_device;
1986 if (!scsi_target_queue_ready(shost, sdev))
1987 goto out_dec_device_busy;
1988 if (!scsi_host_queue_ready(q, shost, sdev))
1989 goto out_dec_target_busy;
1990
1991
1992 if (!(req->cmd_flags & REQ_DONTPREP)) {
1993 ret = prep_to_mq(scsi_mq_prep_fn(req));
1994 if (ret)
1995 goto out_dec_host_busy;
1996 req->cmd_flags |= REQ_DONTPREP;
1997 } else {
1998 blk_mq_start_request(req);
1999 }
2000
2001 if (sdev->simple_tags)
2002 cmd->flags |= SCMD_TAGGED;
2003 else
2004 cmd->flags &= ~SCMD_TAGGED;
2005
2006 scsi_init_cmd_errh(cmd);
2007 cmd->scsi_done = scsi_mq_done;
2008
2009 reason = scsi_dispatch_cmd(cmd);
2010 if (reason) {
2011 scsi_set_blocked(cmd, reason);
2012 ret = BLK_MQ_RQ_QUEUE_BUSY;
2013 goto out_dec_host_busy;
2014 }
2015
2016 return BLK_MQ_RQ_QUEUE_OK;
2017
2018out_dec_host_busy:
2019 atomic_dec(&shost->host_busy);
2020out_dec_target_busy:
2021 if (scsi_target(sdev)->can_queue > 0)
2022 atomic_dec(&scsi_target(sdev)->target_busy);
2023out_dec_device_busy:
2024 atomic_dec(&sdev->device_busy);
2025out_put_device:
2026 put_device(&sdev->sdev_gendev);
2027out:
2028 switch (ret) {
2029 case BLK_MQ_RQ_QUEUE_BUSY:
2030 blk_mq_stop_hw_queue(hctx);
2031 if (atomic_read(&sdev->device_busy) == 0 &&
2032 !scsi_device_blocked(sdev))
2033 blk_mq_delay_queue(hctx, SCSI_QUEUE_DELAY);
2034 break;
2035 case BLK_MQ_RQ_QUEUE_ERROR:
2036 /*
2037 * Make sure to release all allocated ressources when
2038 * we hit an error, as we will never see this command
2039 * again.
2040 */
2041 if (req->cmd_flags & REQ_DONTPREP)
2042 scsi_mq_uninit_cmd(cmd);
2043 break;
2044 default:
2045 break;
2046 }
2047 return ret;
2048}
2049
2050static enum blk_eh_timer_return scsi_timeout(struct request *req,
2051 bool reserved)
2052{
2053 if (reserved)
2054 return BLK_EH_RESET_TIMER;
2055 return scsi_times_out(req);
2056}
2057
2058static int scsi_init_request(void *data, struct request *rq,
2059 unsigned int hctx_idx, unsigned int request_idx,
2060 unsigned int numa_node)
2061{
2062 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2063
2064 cmd->sense_buffer = kzalloc_node(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL,
2065 numa_node);
2066 if (!cmd->sense_buffer)
2067 return -ENOMEM;
2068 return 0;
2069}
2070
2071static void scsi_exit_request(void *data, struct request *rq,
2072 unsigned int hctx_idx, unsigned int request_idx)
2073{
2074 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
2075
2076 kfree(cmd->sense_buffer);
2077}
2078
2079static u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
2080{
2081 struct device *host_dev;
2082 u64 bounce_limit = 0xffffffff;
2083
2084 if (shost->unchecked_isa_dma)
2085 return BLK_BOUNCE_ISA;
2086 /*
2087 * Platforms with virtual-DMA translation
2088 * hardware have no practical limit.
2089 */
2090 if (!PCI_DMA_BUS_IS_PHYS)
2091 return BLK_BOUNCE_ANY;
2092
2093 host_dev = scsi_get_device(shost);
2094 if (host_dev && host_dev->dma_mask)
2095 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
2096
2097 return bounce_limit;
2098}
2099
2100static void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
2101{
2102 struct device *dev = shost->dma_dev;
2103
2104 /*
2105 * this limit is imposed by hardware restrictions
2106 */
2107 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
2108 SCSI_MAX_SG_CHAIN_SEGMENTS));
2109
2110 if (scsi_host_prot_dma(shost)) {
2111 shost->sg_prot_tablesize =
2112 min_not_zero(shost->sg_prot_tablesize,
2113 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
2114 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
2115 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
2116 }
2117
2118 blk_queue_max_hw_sectors(q, shost->max_sectors);
2119 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
2120 blk_queue_segment_boundary(q, shost->dma_boundary);
2121 dma_set_seg_boundary(dev, shost->dma_boundary);
2122
2123 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
2124
2125 if (!shost->use_clustering)
2126 q->limits.cluster = 0;
2127
2128 /*
2129 * set a reasonable default alignment on word boundaries: the
2130 * host and device may alter it using
2131 * blk_queue_update_dma_alignment() later.
2132 */
2133 blk_queue_dma_alignment(q, 0x03);
2134}
2135
2136struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
2137 request_fn_proc *request_fn)
2138{
2139 struct request_queue *q;
2140
2141 q = blk_init_queue(request_fn, NULL);
2142 if (!q)
2143 return NULL;
2144 __scsi_init_queue(shost, q);
2145 return q;
2146}
2147EXPORT_SYMBOL(__scsi_alloc_queue);
2148
2149struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
2150{
2151 struct request_queue *q;
2152
2153 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
2154 if (!q)
2155 return NULL;
2156
2157 blk_queue_prep_rq(q, scsi_prep_fn);
2158 blk_queue_unprep_rq(q, scsi_unprep_fn);
2159 blk_queue_softirq_done(q, scsi_softirq_done);
2160 blk_queue_rq_timed_out(q, scsi_times_out);
2161 blk_queue_lld_busy(q, scsi_lld_busy);
2162 return q;
2163}
2164
2165static struct blk_mq_ops scsi_mq_ops = {
2166 .map_queue = blk_mq_map_queue,
2167 .queue_rq = scsi_queue_rq,
2168 .complete = scsi_softirq_done,
2169 .timeout = scsi_timeout,
2170 .init_request = scsi_init_request,
2171 .exit_request = scsi_exit_request,
2172};
2173
2174struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
2175{
2176 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
2177 if (IS_ERR(sdev->request_queue))
2178 return NULL;
2179
2180 sdev->request_queue->queuedata = sdev;
2181 __scsi_init_queue(sdev->host, sdev->request_queue);
2182 return sdev->request_queue;
2183}
2184
2185int scsi_mq_setup_tags(struct Scsi_Host *shost)
2186{
2187 unsigned int cmd_size, sgl_size, tbl_size;
2188
2189 tbl_size = shost->sg_tablesize;
2190 if (tbl_size > SCSI_MAX_SG_SEGMENTS)
2191 tbl_size = SCSI_MAX_SG_SEGMENTS;
2192 sgl_size = tbl_size * sizeof(struct scatterlist);
2193 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2194 if (scsi_host_get_prot(shost))
2195 cmd_size += sizeof(struct scsi_data_buffer) + sgl_size;
2196
2197 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
2198 shost->tag_set.ops = &scsi_mq_ops;
2199 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
2200 shost->tag_set.queue_depth = shost->can_queue;
2201 shost->tag_set.cmd_size = cmd_size;
2202 shost->tag_set.numa_node = NUMA_NO_NODE;
2203 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE;
2204 shost->tag_set.flags |=
2205 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2206 shost->tag_set.driver_data = shost;
2207
2208 return blk_mq_alloc_tag_set(&shost->tag_set);
2209}
2210
2211void scsi_mq_destroy_tags(struct Scsi_Host *shost)
2212{
2213 blk_mq_free_tag_set(&shost->tag_set);
2214}
2215
2216/*
2217 * Function: scsi_block_requests()
2218 *
2219 * Purpose: Utility function used by low-level drivers to prevent further
2220 * commands from being queued to the device.
2221 *
2222 * Arguments: shost - Host in question
2223 *
2224 * Returns: Nothing
2225 *
2226 * Lock status: No locks are assumed held.
2227 *
2228 * Notes: There is no timer nor any other means by which the requests
2229 * get unblocked other than the low-level driver calling
2230 * scsi_unblock_requests().
2231 */
2232void scsi_block_requests(struct Scsi_Host *shost)
2233{
2234 shost->host_self_blocked = 1;
2235}
2236EXPORT_SYMBOL(scsi_block_requests);
2237
2238/*
2239 * Function: scsi_unblock_requests()
2240 *
2241 * Purpose: Utility function used by low-level drivers to allow further
2242 * commands from being queued to the device.
2243 *
2244 * Arguments: shost - Host in question
2245 *
2246 * Returns: Nothing
2247 *
2248 * Lock status: No locks are assumed held.
2249 *
2250 * Notes: There is no timer nor any other means by which the requests
2251 * get unblocked other than the low-level driver calling
2252 * scsi_unblock_requests().
2253 *
2254 * This is done as an API function so that changes to the
2255 * internals of the scsi mid-layer won't require wholesale
2256 * changes to drivers that use this feature.
2257 */
2258void scsi_unblock_requests(struct Scsi_Host *shost)
2259{
2260 shost->host_self_blocked = 0;
2261 scsi_run_host_queues(shost);
2262}
2263EXPORT_SYMBOL(scsi_unblock_requests);
2264
2265int __init scsi_init_queue(void)
2266{
2267 int i;
2268
2269 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
2270 sizeof(struct scsi_data_buffer),
2271 0, 0, NULL);
2272 if (!scsi_sdb_cache) {
2273 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
2274 return -ENOMEM;
2275 }
2276
2277 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2278 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2279 int size = sgp->size * sizeof(struct scatterlist);
2280
2281 sgp->slab = kmem_cache_create(sgp->name, size, 0,
2282 SLAB_HWCACHE_ALIGN, NULL);
2283 if (!sgp->slab) {
2284 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
2285 sgp->name);
2286 goto cleanup_sdb;
2287 }
2288
2289 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
2290 sgp->slab);
2291 if (!sgp->pool) {
2292 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
2293 sgp->name);
2294 goto cleanup_sdb;
2295 }
2296 }
2297
2298 return 0;
2299
2300cleanup_sdb:
2301 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2302 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2303 if (sgp->pool)
2304 mempool_destroy(sgp->pool);
2305 if (sgp->slab)
2306 kmem_cache_destroy(sgp->slab);
2307 }
2308 kmem_cache_destroy(scsi_sdb_cache);
2309
2310 return -ENOMEM;
2311}
2312
2313void scsi_exit_queue(void)
2314{
2315 int i;
2316
2317 kmem_cache_destroy(scsi_sdb_cache);
2318
2319 for (i = 0; i < SG_MEMPOOL_NR; i++) {
2320 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
2321 mempool_destroy(sgp->pool);
2322 kmem_cache_destroy(sgp->slab);
2323 }
2324}
2325
2326/**
2327 * scsi_mode_select - issue a mode select
2328 * @sdev: SCSI device to be queried
2329 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2330 * @sp: Save page bit (0 == don't save, 1 == save)
2331 * @modepage: mode page being requested
2332 * @buffer: request buffer (may not be smaller than eight bytes)
2333 * @len: length of request buffer.
2334 * @timeout: command timeout
2335 * @retries: number of retries before failing
2336 * @data: returns a structure abstracting the mode header data
2337 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2338 * must be SCSI_SENSE_BUFFERSIZE big.
2339 *
2340 * Returns zero if successful; negative error number or scsi
2341 * status on error
2342 *
2343 */
2344int
2345scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2346 unsigned char *buffer, int len, int timeout, int retries,
2347 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2348{
2349 unsigned char cmd[10];
2350 unsigned char *real_buffer;
2351 int ret;
2352
2353 memset(cmd, 0, sizeof(cmd));
2354 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2355
2356 if (sdev->use_10_for_ms) {
2357 if (len > 65535)
2358 return -EINVAL;
2359 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2360 if (!real_buffer)
2361 return -ENOMEM;
2362 memcpy(real_buffer + 8, buffer, len);
2363 len += 8;
2364 real_buffer[0] = 0;
2365 real_buffer[1] = 0;
2366 real_buffer[2] = data->medium_type;
2367 real_buffer[3] = data->device_specific;
2368 real_buffer[4] = data->longlba ? 0x01 : 0;
2369 real_buffer[5] = 0;
2370 real_buffer[6] = data->block_descriptor_length >> 8;
2371 real_buffer[7] = data->block_descriptor_length;
2372
2373 cmd[0] = MODE_SELECT_10;
2374 cmd[7] = len >> 8;
2375 cmd[8] = len;
2376 } else {
2377 if (len > 255 || data->block_descriptor_length > 255 ||
2378 data->longlba)
2379 return -EINVAL;
2380
2381 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2382 if (!real_buffer)
2383 return -ENOMEM;
2384 memcpy(real_buffer + 4, buffer, len);
2385 len += 4;
2386 real_buffer[0] = 0;
2387 real_buffer[1] = data->medium_type;
2388 real_buffer[2] = data->device_specific;
2389 real_buffer[3] = data->block_descriptor_length;
2390
2391
2392 cmd[0] = MODE_SELECT;
2393 cmd[4] = len;
2394 }
2395
2396 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2397 sshdr, timeout, retries, NULL);
2398 kfree(real_buffer);
2399 return ret;
2400}
2401EXPORT_SYMBOL_GPL(scsi_mode_select);
2402
2403/**
2404 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2405 * @sdev: SCSI device to be queried
2406 * @dbd: set if mode sense will allow block descriptors to be returned
2407 * @modepage: mode page being requested
2408 * @buffer: request buffer (may not be smaller than eight bytes)
2409 * @len: length of request buffer.
2410 * @timeout: command timeout
2411 * @retries: number of retries before failing
2412 * @data: returns a structure abstracting the mode header data
2413 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2414 * must be SCSI_SENSE_BUFFERSIZE big.
2415 *
2416 * Returns zero if unsuccessful, or the header offset (either 4
2417 * or 8 depending on whether a six or ten byte command was
2418 * issued) if successful.
2419 */
2420int
2421scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2422 unsigned char *buffer, int len, int timeout, int retries,
2423 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2424{
2425 unsigned char cmd[12];
2426 int use_10_for_ms;
2427 int header_length;
2428 int result, retry_count = retries;
2429 struct scsi_sense_hdr my_sshdr;
2430
2431 memset(data, 0, sizeof(*data));
2432 memset(&cmd[0], 0, 12);
2433 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2434 cmd[2] = modepage;
2435
2436 /* caller might not be interested in sense, but we need it */
2437 if (!sshdr)
2438 sshdr = &my_sshdr;
2439
2440 retry:
2441 use_10_for_ms = sdev->use_10_for_ms;
2442
2443 if (use_10_for_ms) {
2444 if (len < 8)
2445 len = 8;
2446
2447 cmd[0] = MODE_SENSE_10;
2448 cmd[8] = len;
2449 header_length = 8;
2450 } else {
2451 if (len < 4)
2452 len = 4;
2453
2454 cmd[0] = MODE_SENSE;
2455 cmd[4] = len;
2456 header_length = 4;
2457 }
2458
2459 memset(buffer, 0, len);
2460
2461 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2462 sshdr, timeout, retries, NULL);
2463
2464 /* This code looks awful: what it's doing is making sure an
2465 * ILLEGAL REQUEST sense return identifies the actual command
2466 * byte as the problem. MODE_SENSE commands can return
2467 * ILLEGAL REQUEST if the code page isn't supported */
2468
2469 if (use_10_for_ms && !scsi_status_is_good(result) &&
2470 (driver_byte(result) & DRIVER_SENSE)) {
2471 if (scsi_sense_valid(sshdr)) {
2472 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2473 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2474 /*
2475 * Invalid command operation code
2476 */
2477 sdev->use_10_for_ms = 0;
2478 goto retry;
2479 }
2480 }
2481 }
2482
2483 if(scsi_status_is_good(result)) {
2484 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2485 (modepage == 6 || modepage == 8))) {
2486 /* Initio breakage? */
2487 header_length = 0;
2488 data->length = 13;
2489 data->medium_type = 0;
2490 data->device_specific = 0;
2491 data->longlba = 0;
2492 data->block_descriptor_length = 0;
2493 } else if(use_10_for_ms) {
2494 data->length = buffer[0]*256 + buffer[1] + 2;
2495 data->medium_type = buffer[2];
2496 data->device_specific = buffer[3];
2497 data->longlba = buffer[4] & 0x01;
2498 data->block_descriptor_length = buffer[6]*256
2499 + buffer[7];
2500 } else {
2501 data->length = buffer[0] + 1;
2502 data->medium_type = buffer[1];
2503 data->device_specific = buffer[2];
2504 data->block_descriptor_length = buffer[3];
2505 }
2506 data->header_length = header_length;
2507 } else if ((status_byte(result) == CHECK_CONDITION) &&
2508 scsi_sense_valid(sshdr) &&
2509 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2510 retry_count--;
2511 goto retry;
2512 }
2513
2514 return result;
2515}
2516EXPORT_SYMBOL(scsi_mode_sense);
2517
2518/**
2519 * scsi_test_unit_ready - test if unit is ready
2520 * @sdev: scsi device to change the state of.
2521 * @timeout: command timeout
2522 * @retries: number of retries before failing
2523 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2524 * returning sense. Make sure that this is cleared before passing
2525 * in.
2526 *
2527 * Returns zero if unsuccessful or an error if TUR failed. For
2528 * removable media, UNIT_ATTENTION sets ->changed flag.
2529 **/
2530int
2531scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2532 struct scsi_sense_hdr *sshdr_external)
2533{
2534 char cmd[] = {
2535 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2536 };
2537 struct scsi_sense_hdr *sshdr;
2538 int result;
2539
2540 if (!sshdr_external)
2541 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2542 else
2543 sshdr = sshdr_external;
2544
2545 /* try to eat the UNIT_ATTENTION if there are enough retries */
2546 do {
2547 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2548 timeout, retries, NULL);
2549 if (sdev->removable && scsi_sense_valid(sshdr) &&
2550 sshdr->sense_key == UNIT_ATTENTION)
2551 sdev->changed = 1;
2552 } while (scsi_sense_valid(sshdr) &&
2553 sshdr->sense_key == UNIT_ATTENTION && --retries);
2554
2555 if (!sshdr_external)
2556 kfree(sshdr);
2557 return result;
2558}
2559EXPORT_SYMBOL(scsi_test_unit_ready);
2560
2561/**
2562 * scsi_device_set_state - Take the given device through the device state model.
2563 * @sdev: scsi device to change the state of.
2564 * @state: state to change to.
2565 *
2566 * Returns zero if unsuccessful or an error if the requested
2567 * transition is illegal.
2568 */
2569int
2570scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2571{
2572 enum scsi_device_state oldstate = sdev->sdev_state;
2573
2574 if (state == oldstate)
2575 return 0;
2576
2577 switch (state) {
2578 case SDEV_CREATED:
2579 switch (oldstate) {
2580 case SDEV_CREATED_BLOCK:
2581 break;
2582 default:
2583 goto illegal;
2584 }
2585 break;
2586
2587 case SDEV_RUNNING:
2588 switch (oldstate) {
2589 case SDEV_CREATED:
2590 case SDEV_OFFLINE:
2591 case SDEV_TRANSPORT_OFFLINE:
2592 case SDEV_QUIESCE:
2593 case SDEV_BLOCK:
2594 break;
2595 default:
2596 goto illegal;
2597 }
2598 break;
2599
2600 case SDEV_QUIESCE:
2601 switch (oldstate) {
2602 case SDEV_RUNNING:
2603 case SDEV_OFFLINE:
2604 case SDEV_TRANSPORT_OFFLINE:
2605 break;
2606 default:
2607 goto illegal;
2608 }
2609 break;
2610
2611 case SDEV_OFFLINE:
2612 case SDEV_TRANSPORT_OFFLINE:
2613 switch (oldstate) {
2614 case SDEV_CREATED:
2615 case SDEV_RUNNING:
2616 case SDEV_QUIESCE:
2617 case SDEV_BLOCK:
2618 break;
2619 default:
2620 goto illegal;
2621 }
2622 break;
2623
2624 case SDEV_BLOCK:
2625 switch (oldstate) {
2626 case SDEV_RUNNING:
2627 case SDEV_CREATED_BLOCK:
2628 break;
2629 default:
2630 goto illegal;
2631 }
2632 break;
2633
2634 case SDEV_CREATED_BLOCK:
2635 switch (oldstate) {
2636 case SDEV_CREATED:
2637 break;
2638 default:
2639 goto illegal;
2640 }
2641 break;
2642
2643 case SDEV_CANCEL:
2644 switch (oldstate) {
2645 case SDEV_CREATED:
2646 case SDEV_RUNNING:
2647 case SDEV_QUIESCE:
2648 case SDEV_OFFLINE:
2649 case SDEV_TRANSPORT_OFFLINE:
2650 case SDEV_BLOCK:
2651 break;
2652 default:
2653 goto illegal;
2654 }
2655 break;
2656
2657 case SDEV_DEL:
2658 switch (oldstate) {
2659 case SDEV_CREATED:
2660 case SDEV_RUNNING:
2661 case SDEV_OFFLINE:
2662 case SDEV_TRANSPORT_OFFLINE:
2663 case SDEV_CANCEL:
2664 case SDEV_CREATED_BLOCK:
2665 break;
2666 default:
2667 goto illegal;
2668 }
2669 break;
2670
2671 }
2672 sdev->sdev_state = state;
2673 return 0;
2674
2675 illegal:
2676 SCSI_LOG_ERROR_RECOVERY(1,
2677 sdev_printk(KERN_ERR, sdev,
2678 "Illegal state transition %s->%s",
2679 scsi_device_state_name(oldstate),
2680 scsi_device_state_name(state))
2681 );
2682 return -EINVAL;
2683}
2684EXPORT_SYMBOL(scsi_device_set_state);
2685
2686/**
2687 * sdev_evt_emit - emit a single SCSI device uevent
2688 * @sdev: associated SCSI device
2689 * @evt: event to emit
2690 *
2691 * Send a single uevent (scsi_event) to the associated scsi_device.
2692 */
2693static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2694{
2695 int idx = 0;
2696 char *envp[3];
2697
2698 switch (evt->evt_type) {
2699 case SDEV_EVT_MEDIA_CHANGE:
2700 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2701 break;
2702 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2703 scsi_rescan_device(&sdev->sdev_gendev);
2704 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2705 break;
2706 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2707 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2708 break;
2709 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2710 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2711 break;
2712 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2713 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2714 break;
2715 case SDEV_EVT_LUN_CHANGE_REPORTED:
2716 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2717 break;
2718 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2719 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2720 break;
2721 default:
2722 /* do nothing */
2723 break;
2724 }
2725
2726 envp[idx++] = NULL;
2727
2728 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2729}
2730
2731/**
2732 * sdev_evt_thread - send a uevent for each scsi event
2733 * @work: work struct for scsi_device
2734 *
2735 * Dispatch queued events to their associated scsi_device kobjects
2736 * as uevents.
2737 */
2738void scsi_evt_thread(struct work_struct *work)
2739{
2740 struct scsi_device *sdev;
2741 enum scsi_device_event evt_type;
2742 LIST_HEAD(event_list);
2743
2744 sdev = container_of(work, struct scsi_device, event_work);
2745
2746 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2747 if (test_and_clear_bit(evt_type, sdev->pending_events))
2748 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2749
2750 while (1) {
2751 struct scsi_event *evt;
2752 struct list_head *this, *tmp;
2753 unsigned long flags;
2754
2755 spin_lock_irqsave(&sdev->list_lock, flags);
2756 list_splice_init(&sdev->event_list, &event_list);
2757 spin_unlock_irqrestore(&sdev->list_lock, flags);
2758
2759 if (list_empty(&event_list))
2760 break;
2761
2762 list_for_each_safe(this, tmp, &event_list) {
2763 evt = list_entry(this, struct scsi_event, node);
2764 list_del(&evt->node);
2765 scsi_evt_emit(sdev, evt);
2766 kfree(evt);
2767 }
2768 }
2769}
2770
2771/**
2772 * sdev_evt_send - send asserted event to uevent thread
2773 * @sdev: scsi_device event occurred on
2774 * @evt: event to send
2775 *
2776 * Assert scsi device event asynchronously.
2777 */
2778void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2779{
2780 unsigned long flags;
2781
2782#if 0
2783 /* FIXME: currently this check eliminates all media change events
2784 * for polled devices. Need to update to discriminate between AN
2785 * and polled events */
2786 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2787 kfree(evt);
2788 return;
2789 }
2790#endif
2791
2792 spin_lock_irqsave(&sdev->list_lock, flags);
2793 list_add_tail(&evt->node, &sdev->event_list);
2794 schedule_work(&sdev->event_work);
2795 spin_unlock_irqrestore(&sdev->list_lock, flags);
2796}
2797EXPORT_SYMBOL_GPL(sdev_evt_send);
2798
2799/**
2800 * sdev_evt_alloc - allocate a new scsi event
2801 * @evt_type: type of event to allocate
2802 * @gfpflags: GFP flags for allocation
2803 *
2804 * Allocates and returns a new scsi_event.
2805 */
2806struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2807 gfp_t gfpflags)
2808{
2809 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2810 if (!evt)
2811 return NULL;
2812
2813 evt->evt_type = evt_type;
2814 INIT_LIST_HEAD(&evt->node);
2815
2816 /* evt_type-specific initialization, if any */
2817 switch (evt_type) {
2818 case SDEV_EVT_MEDIA_CHANGE:
2819 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2820 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2821 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2822 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2823 case SDEV_EVT_LUN_CHANGE_REPORTED:
2824 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2825 default:
2826 /* do nothing */
2827 break;
2828 }
2829
2830 return evt;
2831}
2832EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2833
2834/**
2835 * sdev_evt_send_simple - send asserted event to uevent thread
2836 * @sdev: scsi_device event occurred on
2837 * @evt_type: type of event to send
2838 * @gfpflags: GFP flags for allocation
2839 *
2840 * Assert scsi device event asynchronously, given an event type.
2841 */
2842void sdev_evt_send_simple(struct scsi_device *sdev,
2843 enum scsi_device_event evt_type, gfp_t gfpflags)
2844{
2845 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2846 if (!evt) {
2847 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2848 evt_type);
2849 return;
2850 }
2851
2852 sdev_evt_send(sdev, evt);
2853}
2854EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2855
2856/**
2857 * scsi_device_quiesce - Block user issued commands.
2858 * @sdev: scsi device to quiesce.
2859 *
2860 * This works by trying to transition to the SDEV_QUIESCE state
2861 * (which must be a legal transition). When the device is in this
2862 * state, only special requests will be accepted, all others will
2863 * be deferred. Since special requests may also be requeued requests,
2864 * a successful return doesn't guarantee the device will be
2865 * totally quiescent.
2866 *
2867 * Must be called with user context, may sleep.
2868 *
2869 * Returns zero if unsuccessful or an error if not.
2870 */
2871int
2872scsi_device_quiesce(struct scsi_device *sdev)
2873{
2874 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2875 if (err)
2876 return err;
2877
2878 scsi_run_queue(sdev->request_queue);
2879 while (atomic_read(&sdev->device_busy)) {
2880 msleep_interruptible(200);
2881 scsi_run_queue(sdev->request_queue);
2882 }
2883 return 0;
2884}
2885EXPORT_SYMBOL(scsi_device_quiesce);
2886
2887/**
2888 * scsi_device_resume - Restart user issued commands to a quiesced device.
2889 * @sdev: scsi device to resume.
2890 *
2891 * Moves the device from quiesced back to running and restarts the
2892 * queues.
2893 *
2894 * Must be called with user context, may sleep.
2895 */
2896void scsi_device_resume(struct scsi_device *sdev)
2897{
2898 /* check if the device state was mutated prior to resume, and if
2899 * so assume the state is being managed elsewhere (for example
2900 * device deleted during suspend)
2901 */
2902 if (sdev->sdev_state != SDEV_QUIESCE ||
2903 scsi_device_set_state(sdev, SDEV_RUNNING))
2904 return;
2905 scsi_run_queue(sdev->request_queue);
2906}
2907EXPORT_SYMBOL(scsi_device_resume);
2908
2909static void
2910device_quiesce_fn(struct scsi_device *sdev, void *data)
2911{
2912 scsi_device_quiesce(sdev);
2913}
2914
2915void
2916scsi_target_quiesce(struct scsi_target *starget)
2917{
2918 starget_for_each_device(starget, NULL, device_quiesce_fn);
2919}
2920EXPORT_SYMBOL(scsi_target_quiesce);
2921
2922static void
2923device_resume_fn(struct scsi_device *sdev, void *data)
2924{
2925 scsi_device_resume(sdev);
2926}
2927
2928void
2929scsi_target_resume(struct scsi_target *starget)
2930{
2931 starget_for_each_device(starget, NULL, device_resume_fn);
2932}
2933EXPORT_SYMBOL(scsi_target_resume);
2934
2935/**
2936 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2937 * @sdev: device to block
2938 *
2939 * Block request made by scsi lld's to temporarily stop all
2940 * scsi commands on the specified device. Called from interrupt
2941 * or normal process context.
2942 *
2943 * Returns zero if successful or error if not
2944 *
2945 * Notes:
2946 * This routine transitions the device to the SDEV_BLOCK state
2947 * (which must be a legal transition). When the device is in this
2948 * state, all commands are deferred until the scsi lld reenables
2949 * the device with scsi_device_unblock or device_block_tmo fires.
2950 */
2951int
2952scsi_internal_device_block(struct scsi_device *sdev)
2953{
2954 struct request_queue *q = sdev->request_queue;
2955 unsigned long flags;
2956 int err = 0;
2957
2958 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2959 if (err) {
2960 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2961
2962 if (err)
2963 return err;
2964 }
2965
2966 /*
2967 * The device has transitioned to SDEV_BLOCK. Stop the
2968 * block layer from calling the midlayer with this device's
2969 * request queue.
2970 */
2971 if (q->mq_ops) {
2972 blk_mq_stop_hw_queues(q);
2973 } else {
2974 spin_lock_irqsave(q->queue_lock, flags);
2975 blk_stop_queue(q);
2976 spin_unlock_irqrestore(q->queue_lock, flags);
2977 }
2978
2979 return 0;
2980}
2981EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2982
2983/**
2984 * scsi_internal_device_unblock - resume a device after a block request
2985 * @sdev: device to resume
2986 * @new_state: state to set devices to after unblocking
2987 *
2988 * Called by scsi lld's or the midlayer to restart the device queue
2989 * for the previously suspended scsi device. Called from interrupt or
2990 * normal process context.
2991 *
2992 * Returns zero if successful or error if not.
2993 *
2994 * Notes:
2995 * This routine transitions the device to the SDEV_RUNNING state
2996 * or to one of the offline states (which must be a legal transition)
2997 * allowing the midlayer to goose the queue for this device.
2998 */
2999int
3000scsi_internal_device_unblock(struct scsi_device *sdev,
3001 enum scsi_device_state new_state)
3002{
3003 struct request_queue *q = sdev->request_queue;
3004 unsigned long flags;
3005
3006 /*
3007 * Try to transition the scsi device to SDEV_RUNNING or one of the
3008 * offlined states and goose the device queue if successful.
3009 */
3010 if ((sdev->sdev_state == SDEV_BLOCK) ||
3011 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
3012 sdev->sdev_state = new_state;
3013 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
3014 if (new_state == SDEV_TRANSPORT_OFFLINE ||
3015 new_state == SDEV_OFFLINE)
3016 sdev->sdev_state = new_state;
3017 else
3018 sdev->sdev_state = SDEV_CREATED;
3019 } else if (sdev->sdev_state != SDEV_CANCEL &&
3020 sdev->sdev_state != SDEV_OFFLINE)
3021 return -EINVAL;
3022
3023 if (q->mq_ops) {
3024 blk_mq_start_stopped_hw_queues(q, false);
3025 } else {
3026 spin_lock_irqsave(q->queue_lock, flags);
3027 blk_start_queue(q);
3028 spin_unlock_irqrestore(q->queue_lock, flags);
3029 }
3030
3031 return 0;
3032}
3033EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
3034
3035static void
3036device_block(struct scsi_device *sdev, void *data)
3037{
3038 scsi_internal_device_block(sdev);
3039}
3040
3041static int
3042target_block(struct device *dev, void *data)
3043{
3044 if (scsi_is_target_device(dev))
3045 starget_for_each_device(to_scsi_target(dev), NULL,
3046 device_block);
3047 return 0;
3048}
3049
3050void
3051scsi_target_block(struct device *dev)
3052{
3053 if (scsi_is_target_device(dev))
3054 starget_for_each_device(to_scsi_target(dev), NULL,
3055 device_block);
3056 else
3057 device_for_each_child(dev, NULL, target_block);
3058}
3059EXPORT_SYMBOL_GPL(scsi_target_block);
3060
3061static void
3062device_unblock(struct scsi_device *sdev, void *data)
3063{
3064 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3065}
3066
3067static int
3068target_unblock(struct device *dev, void *data)
3069{
3070 if (scsi_is_target_device(dev))
3071 starget_for_each_device(to_scsi_target(dev), data,
3072 device_unblock);
3073 return 0;
3074}
3075
3076void
3077scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3078{
3079 if (scsi_is_target_device(dev))
3080 starget_for_each_device(to_scsi_target(dev), &new_state,
3081 device_unblock);
3082 else
3083 device_for_each_child(dev, &new_state, target_unblock);
3084}
3085EXPORT_SYMBOL_GPL(scsi_target_unblock);
3086
3087/**
3088 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3089 * @sgl: scatter-gather list
3090 * @sg_count: number of segments in sg
3091 * @offset: offset in bytes into sg, on return offset into the mapped area
3092 * @len: bytes to map, on return number of bytes mapped
3093 *
3094 * Returns virtual address of the start of the mapped page
3095 */
3096void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3097 size_t *offset, size_t *len)
3098{
3099 int i;
3100 size_t sg_len = 0, len_complete = 0;
3101 struct scatterlist *sg;
3102 struct page *page;
3103
3104 WARN_ON(!irqs_disabled());
3105
3106 for_each_sg(sgl, sg, sg_count, i) {
3107 len_complete = sg_len; /* Complete sg-entries */
3108 sg_len += sg->length;
3109 if (sg_len > *offset)
3110 break;
3111 }
3112
3113 if (unlikely(i == sg_count)) {
3114 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3115 "elements %d\n",
3116 __func__, sg_len, *offset, sg_count);
3117 WARN_ON(1);
3118 return NULL;
3119 }
3120
3121 /* Offset starting from the beginning of first page in this sg-entry */
3122 *offset = *offset - len_complete + sg->offset;
3123
3124 /* Assumption: contiguous pages can be accessed as "page + i" */
3125 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3126 *offset &= ~PAGE_MASK;
3127
3128 /* Bytes in this sg-entry from *offset to the end of the page */
3129 sg_len = PAGE_SIZE - *offset;
3130 if (*len > sg_len)
3131 *len = sg_len;
3132
3133 return kmap_atomic(page);
3134}
3135EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3136
3137/**
3138 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3139 * @virt: virtual address to be unmapped
3140 */
3141void scsi_kunmap_atomic_sg(void *virt)
3142{
3143 kunmap_atomic(virt);
3144}
3145EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3146
3147void sdev_disable_disk_events(struct scsi_device *sdev)
3148{
3149 atomic_inc(&sdev->disk_events_disable_depth);
3150}
3151EXPORT_SYMBOL(sdev_disable_disk_events);
3152
3153void sdev_enable_disk_events(struct scsi_device *sdev)
3154{
3155 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3156 return;
3157 atomic_dec(&sdev->disk_events_disable_depth);
3158}
3159EXPORT_SYMBOL(sdev_enable_disk_events);
3160
3161/**
3162 * scsi_vpd_lun_id - return a unique device identification
3163 * @sdev: SCSI device
3164 * @id: buffer for the identification
3165 * @id_len: length of the buffer
3166 *
3167 * Copies a unique device identification into @id based
3168 * on the information in the VPD page 0x83 of the device.
3169 * The string will be formatted as a SCSI name string.
3170 *
3171 * Returns the length of the identification or error on failure.
3172 * If the identifier is longer than the supplied buffer the actual
3173 * identifier length is returned and the buffer is not zero-padded.
3174 */
3175int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3176{
3177 u8 cur_id_type = 0xff;
3178 u8 cur_id_size = 0;
3179 unsigned char *d, *cur_id_str;
3180 unsigned char __rcu *vpd_pg83;
3181 int id_size = -EINVAL;
3182
3183 rcu_read_lock();
3184 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3185 if (!vpd_pg83) {
3186 rcu_read_unlock();
3187 return -ENXIO;
3188 }
3189
3190 /*
3191 * Look for the correct descriptor.
3192 * Order of preference for lun descriptor:
3193 * - SCSI name string
3194 * - NAA IEEE Registered Extended
3195 * - EUI-64 based 16-byte
3196 * - EUI-64 based 12-byte
3197 * - NAA IEEE Registered
3198 * - NAA IEEE Extended
3199 * as longer descriptors reduce the likelyhood
3200 * of identification clashes.
3201 */
3202
3203 /* The id string must be at least 20 bytes + terminating NULL byte */
3204 if (id_len < 21) {
3205 rcu_read_unlock();
3206 return -EINVAL;
3207 }
3208
3209 memset(id, 0, id_len);
3210 d = vpd_pg83 + 4;
3211 while (d < vpd_pg83 + sdev->vpd_pg83_len) {
3212 /* Skip designators not referring to the LUN */
3213 if ((d[1] & 0x30) != 0x00)
3214 goto next_desig;
3215
3216 switch (d[1] & 0xf) {
3217 case 0x2:
3218 /* EUI-64 */
3219 if (cur_id_size > d[3])
3220 break;
3221 /* Prefer NAA IEEE Registered Extended */
3222 if (cur_id_type == 0x3 &&
3223 cur_id_size == d[3])
3224 break;
3225 cur_id_size = d[3];
3226 cur_id_str = d + 4;
3227 cur_id_type = d[1] & 0xf;
3228 switch (cur_id_size) {
3229 case 8:
3230 id_size = snprintf(id, id_len,
3231 "eui.%8phN",
3232 cur_id_str);
3233 break;
3234 case 12:
3235 id_size = snprintf(id, id_len,
3236 "eui.%12phN",
3237 cur_id_str);
3238 break;
3239 case 16:
3240 id_size = snprintf(id, id_len,
3241 "eui.%16phN",
3242 cur_id_str);
3243 break;
3244 default:
3245 cur_id_size = 0;
3246 break;
3247 }
3248 break;
3249 case 0x3:
3250 /* NAA */
3251 if (cur_id_size > d[3])
3252 break;
3253 cur_id_size = d[3];
3254 cur_id_str = d + 4;
3255 cur_id_type = d[1] & 0xf;
3256 switch (cur_id_size) {
3257 case 8:
3258 id_size = snprintf(id, id_len,
3259 "naa.%8phN",
3260 cur_id_str);
3261 break;
3262 case 16:
3263 id_size = snprintf(id, id_len,
3264 "naa.%16phN",
3265 cur_id_str);
3266 break;
3267 default:
3268 cur_id_size = 0;
3269 break;
3270 }
3271 break;
3272 case 0x8:
3273 /* SCSI name string */
3274 if (cur_id_size + 4 > d[3])
3275 break;
3276 /* Prefer others for truncated descriptor */
3277 if (cur_id_size && d[3] > id_len)
3278 break;
3279 cur_id_size = id_size = d[3];
3280 cur_id_str = d + 4;
3281 cur_id_type = d[1] & 0xf;
3282 if (cur_id_size >= id_len)
3283 cur_id_size = id_len - 1;
3284 memcpy(id, cur_id_str, cur_id_size);
3285 /* Decrease priority for truncated descriptor */
3286 if (cur_id_size != id_size)
3287 cur_id_size = 6;
3288 break;
3289 default:
3290 break;
3291 }
3292next_desig:
3293 d += d[3] + 4;
3294 }
3295 rcu_read_unlock();
3296
3297 return id_size;
3298}
3299EXPORT_SYMBOL(scsi_vpd_lun_id);
3300
3301/*
3302 * scsi_vpd_tpg_id - return a target port group identifier
3303 * @sdev: SCSI device
3304 *
3305 * Returns the Target Port Group identifier from the information
3306 * froom VPD page 0x83 of the device.
3307 *
3308 * Returns the identifier or error on failure.
3309 */
3310int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3311{
3312 unsigned char *d;
3313 unsigned char __rcu *vpd_pg83;
3314 int group_id = -EAGAIN, rel_port = -1;
3315
3316 rcu_read_lock();
3317 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3318 if (!vpd_pg83) {
3319 rcu_read_unlock();
3320 return -ENXIO;
3321 }
3322
3323 d = sdev->vpd_pg83 + 4;
3324 while (d < sdev->vpd_pg83 + sdev->vpd_pg83_len) {
3325 switch (d[1] & 0xf) {
3326 case 0x4:
3327 /* Relative target port */
3328 rel_port = get_unaligned_be16(&d[6]);
3329 break;
3330 case 0x5:
3331 /* Target port group */
3332 group_id = get_unaligned_be16(&d[6]);
3333 break;
3334 default:
3335 break;
3336 }
3337 d += d[3] + 4;
3338 }
3339 rcu_read_unlock();
3340
3341 if (group_id >= 0 && rel_id && rel_port != -1)
3342 *rel_id = rel_port;
3343
3344 return group_id;
3345}
3346EXPORT_SYMBOL(scsi_vpd_tpg_id);
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/export.h>
16#include <linux/mempool.h>
17#include <linux/slab.h>
18#include <linux/init.h>
19#include <linux/pci.h>
20#include <linux/delay.h>
21#include <linux/hardirq.h>
22#include <linux/scatterlist.h>
23
24#include <scsi/scsi.h>
25#include <scsi/scsi_cmnd.h>
26#include <scsi/scsi_dbg.h>
27#include <scsi/scsi_device.h>
28#include <scsi/scsi_driver.h>
29#include <scsi/scsi_eh.h>
30#include <scsi/scsi_host.h>
31
32#include "scsi_priv.h"
33#include "scsi_logging.h"
34
35
36#define SG_MEMPOOL_NR ARRAY_SIZE(scsi_sg_pools)
37#define SG_MEMPOOL_SIZE 2
38
39struct scsi_host_sg_pool {
40 size_t size;
41 char *name;
42 struct kmem_cache *slab;
43 mempool_t *pool;
44};
45
46#define SP(x) { x, "sgpool-" __stringify(x) }
47#if (SCSI_MAX_SG_SEGMENTS < 32)
48#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
49#endif
50static struct scsi_host_sg_pool scsi_sg_pools[] = {
51 SP(8),
52 SP(16),
53#if (SCSI_MAX_SG_SEGMENTS > 32)
54 SP(32),
55#if (SCSI_MAX_SG_SEGMENTS > 64)
56 SP(64),
57#if (SCSI_MAX_SG_SEGMENTS > 128)
58 SP(128),
59#if (SCSI_MAX_SG_SEGMENTS > 256)
60#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
61#endif
62#endif
63#endif
64#endif
65 SP(SCSI_MAX_SG_SEGMENTS)
66};
67#undef SP
68
69struct kmem_cache *scsi_sdb_cache;
70
71/*
72 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
73 * not change behaviour from the previous unplug mechanism, experimentation
74 * may prove this needs changing.
75 */
76#define SCSI_QUEUE_DELAY 3
77
78/**
79 * __scsi_queue_insert - private queue insertion
80 * @cmd: The SCSI command being requeued
81 * @reason: The reason for the requeue
82 * @unbusy: Whether the queue should be unbusied
83 *
84 * This is a private queue insertion. The public interface
85 * scsi_queue_insert() always assumes the queue should be unbusied
86 * because it's always called before the completion. This function is
87 * for a requeue after completion, which should only occur in this
88 * file.
89 */
90static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
91{
92 struct Scsi_Host *host = cmd->device->host;
93 struct scsi_device *device = cmd->device;
94 struct scsi_target *starget = scsi_target(device);
95 struct request_queue *q = device->request_queue;
96 unsigned long flags;
97
98 SCSI_LOG_MLQUEUE(1,
99 printk("Inserting command %p into mlqueue\n", cmd));
100
101 /*
102 * Set the appropriate busy bit for the device/host.
103 *
104 * If the host/device isn't busy, assume that something actually
105 * completed, and that we should be able to queue a command now.
106 *
107 * Note that the prior mid-layer assumption that any host could
108 * always queue at least one command is now broken. The mid-layer
109 * will implement a user specifiable stall (see
110 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
111 * if a command is requeued with no other commands outstanding
112 * either for the device or for the host.
113 */
114 switch (reason) {
115 case SCSI_MLQUEUE_HOST_BUSY:
116 host->host_blocked = host->max_host_blocked;
117 break;
118 case SCSI_MLQUEUE_DEVICE_BUSY:
119 case SCSI_MLQUEUE_EH_RETRY:
120 device->device_blocked = device->max_device_blocked;
121 break;
122 case SCSI_MLQUEUE_TARGET_BUSY:
123 starget->target_blocked = starget->max_target_blocked;
124 break;
125 }
126
127 /*
128 * Decrement the counters, since these commands are no longer
129 * active on the host/device.
130 */
131 if (unbusy)
132 scsi_device_unbusy(device);
133
134 /*
135 * Requeue this command. It will go before all other commands
136 * that are already in the queue. Schedule requeue work under
137 * lock such that the kblockd_schedule_work() call happens
138 * before blk_cleanup_queue() finishes.
139 */
140 cmd->result = 0;
141 spin_lock_irqsave(q->queue_lock, flags);
142 blk_requeue_request(q, cmd->request);
143 kblockd_schedule_work(q, &device->requeue_work);
144 spin_unlock_irqrestore(q->queue_lock, flags);
145}
146
147/*
148 * Function: scsi_queue_insert()
149 *
150 * Purpose: Insert a command in the midlevel queue.
151 *
152 * Arguments: cmd - command that we are adding to queue.
153 * reason - why we are inserting command to queue.
154 *
155 * Lock status: Assumed that lock is not held upon entry.
156 *
157 * Returns: Nothing.
158 *
159 * Notes: We do this for one of two cases. Either the host is busy
160 * and it cannot accept any more commands for the time being,
161 * or the device returned QUEUE_FULL and can accept no more
162 * commands.
163 * Notes: This could be called either from an interrupt context or a
164 * normal process context.
165 */
166void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
167{
168 __scsi_queue_insert(cmd, reason, 1);
169}
170/**
171 * scsi_execute - insert request and wait for the result
172 * @sdev: scsi device
173 * @cmd: scsi command
174 * @data_direction: data direction
175 * @buffer: data buffer
176 * @bufflen: len of buffer
177 * @sense: optional sense buffer
178 * @timeout: request timeout in seconds
179 * @retries: number of times to retry request
180 * @flags: or into request flags;
181 * @resid: optional residual length
182 *
183 * returns the req->errors value which is the scsi_cmnd result
184 * field.
185 */
186int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
187 int data_direction, void *buffer, unsigned bufflen,
188 unsigned char *sense, int timeout, int retries, u64 flags,
189 int *resid)
190{
191 struct request *req;
192 int write = (data_direction == DMA_TO_DEVICE);
193 int ret = DRIVER_ERROR << 24;
194
195 req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
196 if (!req)
197 return ret;
198
199 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
200 buffer, bufflen, __GFP_WAIT))
201 goto out;
202
203 req->cmd_len = COMMAND_SIZE(cmd[0]);
204 memcpy(req->cmd, cmd, req->cmd_len);
205 req->sense = sense;
206 req->sense_len = 0;
207 req->retries = retries;
208 req->timeout = timeout;
209 req->cmd_type = REQ_TYPE_BLOCK_PC;
210 req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
211
212 /*
213 * head injection *required* here otherwise quiesce won't work
214 */
215 blk_execute_rq(req->q, NULL, req, 1);
216
217 /*
218 * Some devices (USB mass-storage in particular) may transfer
219 * garbage data together with a residue indicating that the data
220 * is invalid. Prevent the garbage from being misinterpreted
221 * and prevent security leaks by zeroing out the excess data.
222 */
223 if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
224 memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
225
226 if (resid)
227 *resid = req->resid_len;
228 ret = req->errors;
229 out:
230 blk_put_request(req);
231
232 return ret;
233}
234EXPORT_SYMBOL(scsi_execute);
235
236int scsi_execute_req_flags(struct scsi_device *sdev, const unsigned char *cmd,
237 int data_direction, void *buffer, unsigned bufflen,
238 struct scsi_sense_hdr *sshdr, int timeout, int retries,
239 int *resid, u64 flags)
240{
241 char *sense = NULL;
242 int result;
243
244 if (sshdr) {
245 sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
246 if (!sense)
247 return DRIVER_ERROR << 24;
248 }
249 result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
250 sense, timeout, retries, flags, resid);
251 if (sshdr)
252 scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
253
254 kfree(sense);
255 return result;
256}
257EXPORT_SYMBOL(scsi_execute_req_flags);
258
259/*
260 * Function: scsi_init_cmd_errh()
261 *
262 * Purpose: Initialize cmd fields related to error handling.
263 *
264 * Arguments: cmd - command that is ready to be queued.
265 *
266 * Notes: This function has the job of initializing a number of
267 * fields related to error handling. Typically this will
268 * be called once for each command, as required.
269 */
270static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
271{
272 cmd->serial_number = 0;
273 scsi_set_resid(cmd, 0);
274 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
275 if (cmd->cmd_len == 0)
276 cmd->cmd_len = scsi_command_size(cmd->cmnd);
277}
278
279void scsi_device_unbusy(struct scsi_device *sdev)
280{
281 struct Scsi_Host *shost = sdev->host;
282 struct scsi_target *starget = scsi_target(sdev);
283 unsigned long flags;
284
285 spin_lock_irqsave(shost->host_lock, flags);
286 shost->host_busy--;
287 starget->target_busy--;
288 if (unlikely(scsi_host_in_recovery(shost) &&
289 (shost->host_failed || shost->host_eh_scheduled)))
290 scsi_eh_wakeup(shost);
291 spin_unlock(shost->host_lock);
292 spin_lock(sdev->request_queue->queue_lock);
293 sdev->device_busy--;
294 spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
295}
296
297/*
298 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
299 * and call blk_run_queue for all the scsi_devices on the target -
300 * including current_sdev first.
301 *
302 * Called with *no* scsi locks held.
303 */
304static void scsi_single_lun_run(struct scsi_device *current_sdev)
305{
306 struct Scsi_Host *shost = current_sdev->host;
307 struct scsi_device *sdev, *tmp;
308 struct scsi_target *starget = scsi_target(current_sdev);
309 unsigned long flags;
310
311 spin_lock_irqsave(shost->host_lock, flags);
312 starget->starget_sdev_user = NULL;
313 spin_unlock_irqrestore(shost->host_lock, flags);
314
315 /*
316 * Call blk_run_queue for all LUNs on the target, starting with
317 * current_sdev. We race with others (to set starget_sdev_user),
318 * but in most cases, we will be first. Ideally, each LU on the
319 * target would get some limited time or requests on the target.
320 */
321 blk_run_queue(current_sdev->request_queue);
322
323 spin_lock_irqsave(shost->host_lock, flags);
324 if (starget->starget_sdev_user)
325 goto out;
326 list_for_each_entry_safe(sdev, tmp, &starget->devices,
327 same_target_siblings) {
328 if (sdev == current_sdev)
329 continue;
330 if (scsi_device_get(sdev))
331 continue;
332
333 spin_unlock_irqrestore(shost->host_lock, flags);
334 blk_run_queue(sdev->request_queue);
335 spin_lock_irqsave(shost->host_lock, flags);
336
337 scsi_device_put(sdev);
338 }
339 out:
340 spin_unlock_irqrestore(shost->host_lock, flags);
341}
342
343static inline int scsi_device_is_busy(struct scsi_device *sdev)
344{
345 if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
346 return 1;
347
348 return 0;
349}
350
351static inline int scsi_target_is_busy(struct scsi_target *starget)
352{
353 return ((starget->can_queue > 0 &&
354 starget->target_busy >= starget->can_queue) ||
355 starget->target_blocked);
356}
357
358static inline int scsi_host_is_busy(struct Scsi_Host *shost)
359{
360 if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
361 shost->host_blocked || shost->host_self_blocked)
362 return 1;
363
364 return 0;
365}
366
367static void scsi_starved_list_run(struct Scsi_Host *shost)
368{
369 LIST_HEAD(starved_list);
370 struct scsi_device *sdev;
371 unsigned long flags;
372
373 spin_lock_irqsave(shost->host_lock, flags);
374 list_splice_init(&shost->starved_list, &starved_list);
375
376 while (!list_empty(&starved_list)) {
377 struct request_queue *slq;
378
379 /*
380 * As long as shost is accepting commands and we have
381 * starved queues, call blk_run_queue. scsi_request_fn
382 * drops the queue_lock and can add us back to the
383 * starved_list.
384 *
385 * host_lock protects the starved_list and starved_entry.
386 * scsi_request_fn must get the host_lock before checking
387 * or modifying starved_list or starved_entry.
388 */
389 if (scsi_host_is_busy(shost))
390 break;
391
392 sdev = list_entry(starved_list.next,
393 struct scsi_device, starved_entry);
394 list_del_init(&sdev->starved_entry);
395 if (scsi_target_is_busy(scsi_target(sdev))) {
396 list_move_tail(&sdev->starved_entry,
397 &shost->starved_list);
398 continue;
399 }
400
401 /*
402 * Once we drop the host lock, a racing scsi_remove_device()
403 * call may remove the sdev from the starved list and destroy
404 * it and the queue. Mitigate by taking a reference to the
405 * queue and never touching the sdev again after we drop the
406 * host lock. Note: if __scsi_remove_device() invokes
407 * blk_cleanup_queue() before the queue is run from this
408 * function then blk_run_queue() will return immediately since
409 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
410 */
411 slq = sdev->request_queue;
412 if (!blk_get_queue(slq))
413 continue;
414 spin_unlock_irqrestore(shost->host_lock, flags);
415
416 blk_run_queue(slq);
417 blk_put_queue(slq);
418
419 spin_lock_irqsave(shost->host_lock, flags);
420 }
421 /* put any unprocessed entries back */
422 list_splice(&starved_list, &shost->starved_list);
423 spin_unlock_irqrestore(shost->host_lock, flags);
424}
425
426/*
427 * Function: scsi_run_queue()
428 *
429 * Purpose: Select a proper request queue to serve next
430 *
431 * Arguments: q - last request's queue
432 *
433 * Returns: Nothing
434 *
435 * Notes: The previous command was completely finished, start
436 * a new one if possible.
437 */
438static void scsi_run_queue(struct request_queue *q)
439{
440 struct scsi_device *sdev = q->queuedata;
441
442 if (scsi_target(sdev)->single_lun)
443 scsi_single_lun_run(sdev);
444 if (!list_empty(&sdev->host->starved_list))
445 scsi_starved_list_run(sdev->host);
446
447 blk_run_queue(q);
448}
449
450void scsi_requeue_run_queue(struct work_struct *work)
451{
452 struct scsi_device *sdev;
453 struct request_queue *q;
454
455 sdev = container_of(work, struct scsi_device, requeue_work);
456 q = sdev->request_queue;
457 scsi_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 scsi_device *sdev = cmd->device;
481 struct request *req = cmd->request;
482 unsigned long flags;
483
484 spin_lock_irqsave(q->queue_lock, flags);
485 blk_unprep_request(req);
486 req->special = NULL;
487 scsi_put_command(cmd);
488 blk_requeue_request(q, req);
489 spin_unlock_irqrestore(q->queue_lock, flags);
490
491 scsi_run_queue(q);
492
493 put_device(&sdev->sdev_gendev);
494}
495
496void scsi_next_command(struct scsi_cmnd *cmd)
497{
498 struct scsi_device *sdev = cmd->device;
499 struct request_queue *q = sdev->request_queue;
500
501 scsi_put_command(cmd);
502 scsi_run_queue(q);
503
504 put_device(&sdev->sdev_gendev);
505}
506
507void scsi_run_host_queues(struct Scsi_Host *shost)
508{
509 struct scsi_device *sdev;
510
511 shost_for_each_device(sdev, shost)
512 scsi_run_queue(sdev->request_queue);
513}
514
515static void __scsi_release_buffers(struct scsi_cmnd *, int);
516
517/*
518 * Function: scsi_end_request()
519 *
520 * Purpose: Post-processing of completed commands (usually invoked at end
521 * of upper level post-processing and scsi_io_completion).
522 *
523 * Arguments: cmd - command that is complete.
524 * error - 0 if I/O indicates success, < 0 for I/O error.
525 * bytes - number of bytes of completed I/O
526 * requeue - indicates whether we should requeue leftovers.
527 *
528 * Lock status: Assumed that lock is not held upon entry.
529 *
530 * Returns: cmd if requeue required, NULL otherwise.
531 *
532 * Notes: This is called for block device requests in order to
533 * mark some number of sectors as complete.
534 *
535 * We are guaranteeing that the request queue will be goosed
536 * at some point during this call.
537 * Notes: If cmd was requeued, upon return it will be a stale pointer.
538 */
539static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
540 int bytes, int requeue)
541{
542 struct request_queue *q = cmd->device->request_queue;
543 struct request *req = cmd->request;
544
545 /*
546 * If there are blocks left over at the end, set up the command
547 * to queue the remainder of them.
548 */
549 if (blk_end_request(req, error, bytes)) {
550 /* kill remainder if no retrys */
551 if (error && scsi_noretry_cmd(cmd))
552 blk_end_request_all(req, error);
553 else {
554 if (requeue) {
555 /*
556 * Bleah. Leftovers again. Stick the
557 * leftovers in the front of the
558 * queue, and goose the queue again.
559 */
560 scsi_release_buffers(cmd);
561 scsi_requeue_command(q, cmd);
562 cmd = NULL;
563 }
564 return cmd;
565 }
566 }
567
568 /*
569 * This will goose the queue request function at the end, so we don't
570 * need to worry about launching another command.
571 */
572 __scsi_release_buffers(cmd, 0);
573 scsi_next_command(cmd);
574 return NULL;
575}
576
577static inline unsigned int scsi_sgtable_index(unsigned short nents)
578{
579 unsigned int index;
580
581 BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
582
583 if (nents <= 8)
584 index = 0;
585 else
586 index = get_count_order(nents) - 3;
587
588 return index;
589}
590
591static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
592{
593 struct scsi_host_sg_pool *sgp;
594
595 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
596 mempool_free(sgl, sgp->pool);
597}
598
599static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
600{
601 struct scsi_host_sg_pool *sgp;
602
603 sgp = scsi_sg_pools + scsi_sgtable_index(nents);
604 return mempool_alloc(sgp->pool, gfp_mask);
605}
606
607static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
608 gfp_t gfp_mask)
609{
610 int ret;
611
612 BUG_ON(!nents);
613
614 ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
615 gfp_mask, scsi_sg_alloc);
616 if (unlikely(ret))
617 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
618 scsi_sg_free);
619
620 return ret;
621}
622
623static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
624{
625 __sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
626}
627
628static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
629{
630
631 if (cmd->sdb.table.nents)
632 scsi_free_sgtable(&cmd->sdb);
633
634 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
635
636 if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
637 struct scsi_data_buffer *bidi_sdb =
638 cmd->request->next_rq->special;
639 scsi_free_sgtable(bidi_sdb);
640 kmem_cache_free(scsi_sdb_cache, bidi_sdb);
641 cmd->request->next_rq->special = NULL;
642 }
643
644 if (scsi_prot_sg_count(cmd))
645 scsi_free_sgtable(cmd->prot_sdb);
646}
647
648/*
649 * Function: scsi_release_buffers()
650 *
651 * Purpose: Completion processing for block device I/O requests.
652 *
653 * Arguments: cmd - command that we are bailing.
654 *
655 * Lock status: Assumed that no lock is held upon entry.
656 *
657 * Returns: Nothing
658 *
659 * Notes: In the event that an upper level driver rejects a
660 * command, we must release resources allocated during
661 * the __init_io() function. Primarily this would involve
662 * the scatter-gather table, and potentially any bounce
663 * buffers.
664 */
665void scsi_release_buffers(struct scsi_cmnd *cmd)
666{
667 __scsi_release_buffers(cmd, 1);
668}
669EXPORT_SYMBOL(scsi_release_buffers);
670
671/**
672 * __scsi_error_from_host_byte - translate SCSI error code into errno
673 * @cmd: SCSI command (unused)
674 * @result: scsi error code
675 *
676 * Translate SCSI error code into standard UNIX errno.
677 * Return values:
678 * -ENOLINK temporary transport failure
679 * -EREMOTEIO permanent target failure, do not retry
680 * -EBADE permanent nexus failure, retry on other path
681 * -ENOSPC No write space available
682 * -ENODATA Medium error
683 * -EIO unspecified I/O error
684 */
685static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
686{
687 int error = 0;
688
689 switch(host_byte(result)) {
690 case DID_TRANSPORT_FAILFAST:
691 error = -ENOLINK;
692 break;
693 case DID_TARGET_FAILURE:
694 set_host_byte(cmd, DID_OK);
695 error = -EREMOTEIO;
696 break;
697 case DID_NEXUS_FAILURE:
698 set_host_byte(cmd, DID_OK);
699 error = -EBADE;
700 break;
701 case DID_ALLOC_FAILURE:
702 set_host_byte(cmd, DID_OK);
703 error = -ENOSPC;
704 break;
705 case DID_MEDIUM_ERROR:
706 set_host_byte(cmd, DID_OK);
707 error = -ENODATA;
708 break;
709 default:
710 error = -EIO;
711 break;
712 }
713
714 return error;
715}
716
717/*
718 * Function: scsi_io_completion()
719 *
720 * Purpose: Completion processing for block device I/O requests.
721 *
722 * Arguments: cmd - command that is finished.
723 *
724 * Lock status: Assumed that no lock is held upon entry.
725 *
726 * Returns: Nothing
727 *
728 * Notes: This function is matched in terms of capabilities to
729 * the function that created the scatter-gather list.
730 * In other words, if there are no bounce buffers
731 * (the normal case for most drivers), we don't need
732 * the logic to deal with cleaning up afterwards.
733 *
734 * We must call scsi_end_request(). This will finish off
735 * the specified number of sectors. If we are done, the
736 * command block will be released and the queue function
737 * will be goosed. If we are not done then we have to
738 * figure out what to do next:
739 *
740 * a) We can call scsi_requeue_command(). The request
741 * will be unprepared and put back on the queue. Then
742 * a new command will be created for it. This should
743 * be used if we made forward progress, or if we want
744 * to switch from READ(10) to READ(6) for example.
745 *
746 * b) We can call scsi_queue_insert(). The request will
747 * be put back on the queue and retried using the same
748 * command as before, possibly after a delay.
749 *
750 * c) We can call blk_end_request() with -EIO to fail
751 * the remainder of the request.
752 */
753void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
754{
755 int result = cmd->result;
756 struct request_queue *q = cmd->device->request_queue;
757 struct request *req = cmd->request;
758 int error = 0;
759 struct scsi_sense_hdr sshdr;
760 int sense_valid = 0;
761 int sense_deferred = 0;
762 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
763 ACTION_DELAYED_RETRY} action;
764 char *description = NULL;
765 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
766
767 if (result) {
768 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
769 if (sense_valid)
770 sense_deferred = scsi_sense_is_deferred(&sshdr);
771 }
772
773 if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
774 if (result) {
775 if (sense_valid && req->sense) {
776 /*
777 * SG_IO wants current and deferred errors
778 */
779 int len = 8 + cmd->sense_buffer[7];
780
781 if (len > SCSI_SENSE_BUFFERSIZE)
782 len = SCSI_SENSE_BUFFERSIZE;
783 memcpy(req->sense, cmd->sense_buffer, len);
784 req->sense_len = len;
785 }
786 if (!sense_deferred)
787 error = __scsi_error_from_host_byte(cmd, result);
788 }
789 /*
790 * __scsi_error_from_host_byte may have reset the host_byte
791 */
792 req->errors = cmd->result;
793
794 req->resid_len = scsi_get_resid(cmd);
795
796 if (scsi_bidi_cmnd(cmd)) {
797 /*
798 * Bidi commands Must be complete as a whole,
799 * both sides at once.
800 */
801 req->next_rq->resid_len = scsi_in(cmd)->resid;
802
803 scsi_release_buffers(cmd);
804 blk_end_request_all(req, 0);
805
806 scsi_next_command(cmd);
807 return;
808 }
809 }
810
811 /* no bidi support for !REQ_TYPE_BLOCK_PC yet */
812 BUG_ON(blk_bidi_rq(req));
813
814 /*
815 * Next deal with any sectors which we were able to correctly
816 * handle.
817 */
818 SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
819 "%d bytes done.\n",
820 blk_rq_sectors(req), good_bytes));
821
822 /*
823 * Recovered errors need reporting, but they're always treated
824 * as success, so fiddle the result code here. For BLOCK_PC
825 * we already took a copy of the original into rq->errors which
826 * is what gets returned to the user
827 */
828 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
829 /* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
830 * print since caller wants ATA registers. Only occurs on
831 * SCSI ATA PASS_THROUGH commands when CK_COND=1
832 */
833 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
834 ;
835 else if (!(req->cmd_flags & REQ_QUIET))
836 scsi_print_sense("", cmd);
837 result = 0;
838 /* BLOCK_PC may have set error */
839 error = 0;
840 }
841
842 /*
843 * A number of bytes were successfully read. If there
844 * are leftovers and there is some kind of error
845 * (result != 0), retry the rest.
846 */
847 if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
848 return;
849
850 error = __scsi_error_from_host_byte(cmd, result);
851
852 if (host_byte(result) == DID_RESET) {
853 /* Third party bus reset or reset for error recovery
854 * reasons. Just retry the command and see what
855 * happens.
856 */
857 action = ACTION_RETRY;
858 } else if (sense_valid && !sense_deferred) {
859 switch (sshdr.sense_key) {
860 case UNIT_ATTENTION:
861 if (cmd->device->removable) {
862 /* Detected disc change. Set a bit
863 * and quietly refuse further access.
864 */
865 cmd->device->changed = 1;
866 description = "Media Changed";
867 action = ACTION_FAIL;
868 } else {
869 /* Must have been a power glitch, or a
870 * bus reset. Could not have been a
871 * media change, so we just retry the
872 * command and see what happens.
873 */
874 action = ACTION_RETRY;
875 }
876 break;
877 case ILLEGAL_REQUEST:
878 /* If we had an ILLEGAL REQUEST returned, then
879 * we may have performed an unsupported
880 * command. The only thing this should be
881 * would be a ten byte read where only a six
882 * byte read was supported. Also, on a system
883 * where READ CAPACITY failed, we may have
884 * read past the end of the disk.
885 */
886 if ((cmd->device->use_10_for_rw &&
887 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
888 (cmd->cmnd[0] == READ_10 ||
889 cmd->cmnd[0] == WRITE_10)) {
890 /* This will issue a new 6-byte command. */
891 cmd->device->use_10_for_rw = 0;
892 action = ACTION_REPREP;
893 } else if (sshdr.asc == 0x10) /* DIX */ {
894 description = "Host Data Integrity Failure";
895 action = ACTION_FAIL;
896 error = -EILSEQ;
897 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
898 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
899 switch (cmd->cmnd[0]) {
900 case UNMAP:
901 description = "Discard failure";
902 break;
903 case WRITE_SAME:
904 case WRITE_SAME_16:
905 if (cmd->cmnd[1] & 0x8)
906 description = "Discard failure";
907 else
908 description =
909 "Write same failure";
910 break;
911 default:
912 description = "Invalid command failure";
913 break;
914 }
915 action = ACTION_FAIL;
916 error = -EREMOTEIO;
917 } else
918 action = ACTION_FAIL;
919 break;
920 case ABORTED_COMMAND:
921 action = ACTION_FAIL;
922 if (sshdr.asc == 0x10) { /* DIF */
923 description = "Target Data Integrity Failure";
924 error = -EILSEQ;
925 }
926 break;
927 case NOT_READY:
928 /* If the device is in the process of becoming
929 * ready, or has a temporary blockage, retry.
930 */
931 if (sshdr.asc == 0x04) {
932 switch (sshdr.ascq) {
933 case 0x01: /* becoming ready */
934 case 0x04: /* format in progress */
935 case 0x05: /* rebuild in progress */
936 case 0x06: /* recalculation in progress */
937 case 0x07: /* operation in progress */
938 case 0x08: /* Long write in progress */
939 case 0x09: /* self test in progress */
940 case 0x14: /* space allocation in progress */
941 action = ACTION_DELAYED_RETRY;
942 break;
943 default:
944 description = "Device not ready";
945 action = ACTION_FAIL;
946 break;
947 }
948 } else {
949 description = "Device not ready";
950 action = ACTION_FAIL;
951 }
952 break;
953 case VOLUME_OVERFLOW:
954 /* See SSC3rXX or current. */
955 action = ACTION_FAIL;
956 break;
957 default:
958 description = "Unhandled sense code";
959 action = ACTION_FAIL;
960 break;
961 }
962 } else {
963 description = "Unhandled error code";
964 action = ACTION_FAIL;
965 }
966
967 if (action != ACTION_FAIL &&
968 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
969 action = ACTION_FAIL;
970 description = "Command timed out";
971 }
972
973 switch (action) {
974 case ACTION_FAIL:
975 /* Give up and fail the remainder of the request */
976 scsi_release_buffers(cmd);
977 if (!(req->cmd_flags & REQ_QUIET)) {
978 if (description)
979 scmd_printk(KERN_INFO, cmd, "%s\n",
980 description);
981 scsi_print_result(cmd);
982 if (driver_byte(result) & DRIVER_SENSE)
983 scsi_print_sense("", cmd);
984 scsi_print_command(cmd);
985 }
986 if (blk_end_request_err(req, error))
987 scsi_requeue_command(q, cmd);
988 else
989 scsi_next_command(cmd);
990 break;
991 case ACTION_REPREP:
992 /* Unprep the request and put it back at the head of the queue.
993 * A new command will be prepared and issued.
994 */
995 scsi_release_buffers(cmd);
996 scsi_requeue_command(q, cmd);
997 break;
998 case ACTION_RETRY:
999 /* Retry the same command immediately */
1000 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
1001 break;
1002 case ACTION_DELAYED_RETRY:
1003 /* Retry the same command after a delay */
1004 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
1005 break;
1006 }
1007}
1008
1009static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
1010 gfp_t gfp_mask)
1011{
1012 int count;
1013
1014 /*
1015 * If sg table allocation fails, requeue request later.
1016 */
1017 if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
1018 gfp_mask))) {
1019 return BLKPREP_DEFER;
1020 }
1021
1022 req->buffer = NULL;
1023
1024 /*
1025 * Next, walk the list, and fill in the addresses and sizes of
1026 * each segment.
1027 */
1028 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1029 BUG_ON(count > sdb->table.nents);
1030 sdb->table.nents = count;
1031 sdb->length = blk_rq_bytes(req);
1032 return BLKPREP_OK;
1033}
1034
1035/*
1036 * Function: scsi_init_io()
1037 *
1038 * Purpose: SCSI I/O initialize function.
1039 *
1040 * Arguments: cmd - Command descriptor we wish to initialize
1041 *
1042 * Returns: 0 on success
1043 * BLKPREP_DEFER if the failure is retryable
1044 * BLKPREP_KILL if the failure is fatal
1045 */
1046int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1047{
1048 struct scsi_device *sdev = cmd->device;
1049 struct request *rq = cmd->request;
1050
1051 int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1052 if (error)
1053 goto err_exit;
1054
1055 if (blk_bidi_rq(rq)) {
1056 struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1057 scsi_sdb_cache, GFP_ATOMIC);
1058 if (!bidi_sdb) {
1059 error = BLKPREP_DEFER;
1060 goto err_exit;
1061 }
1062
1063 rq->next_rq->special = bidi_sdb;
1064 error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1065 if (error)
1066 goto err_exit;
1067 }
1068
1069 if (blk_integrity_rq(rq)) {
1070 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1071 int ivecs, count;
1072
1073 BUG_ON(prot_sdb == NULL);
1074 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1075
1076 if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1077 error = BLKPREP_DEFER;
1078 goto err_exit;
1079 }
1080
1081 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1082 prot_sdb->table.sgl);
1083 BUG_ON(unlikely(count > ivecs));
1084 BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1085
1086 cmd->prot_sdb = prot_sdb;
1087 cmd->prot_sdb->table.nents = count;
1088 }
1089
1090 return BLKPREP_OK ;
1091
1092err_exit:
1093 scsi_release_buffers(cmd);
1094 cmd->request->special = NULL;
1095 scsi_put_command(cmd);
1096 put_device(&sdev->sdev_gendev);
1097 return error;
1098}
1099EXPORT_SYMBOL(scsi_init_io);
1100
1101static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1102 struct request *req)
1103{
1104 struct scsi_cmnd *cmd;
1105
1106 if (!req->special) {
1107 /* Bail if we can't get a reference to the device */
1108 if (!get_device(&sdev->sdev_gendev))
1109 return NULL;
1110
1111 cmd = scsi_get_command(sdev, GFP_ATOMIC);
1112 if (unlikely(!cmd)) {
1113 put_device(&sdev->sdev_gendev);
1114 return NULL;
1115 }
1116 req->special = cmd;
1117 } else {
1118 cmd = req->special;
1119 }
1120
1121 /* pull a tag out of the request if we have one */
1122 cmd->tag = req->tag;
1123 cmd->request = req;
1124
1125 cmd->cmnd = req->cmd;
1126 cmd->prot_op = SCSI_PROT_NORMAL;
1127
1128 return cmd;
1129}
1130
1131int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1132{
1133 struct scsi_cmnd *cmd;
1134 int ret = scsi_prep_state_check(sdev, req);
1135
1136 if (ret != BLKPREP_OK)
1137 return ret;
1138
1139 cmd = scsi_get_cmd_from_req(sdev, req);
1140 if (unlikely(!cmd))
1141 return BLKPREP_DEFER;
1142
1143 /*
1144 * BLOCK_PC requests may transfer data, in which case they must
1145 * a bio attached to them. Or they might contain a SCSI command
1146 * that does not transfer data, in which case they may optionally
1147 * submit a request without an attached bio.
1148 */
1149 if (req->bio) {
1150 int ret;
1151
1152 BUG_ON(!req->nr_phys_segments);
1153
1154 ret = scsi_init_io(cmd, GFP_ATOMIC);
1155 if (unlikely(ret))
1156 return ret;
1157 } else {
1158 BUG_ON(blk_rq_bytes(req));
1159
1160 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1161 req->buffer = NULL;
1162 }
1163
1164 cmd->cmd_len = req->cmd_len;
1165 if (!blk_rq_bytes(req))
1166 cmd->sc_data_direction = DMA_NONE;
1167 else if (rq_data_dir(req) == WRITE)
1168 cmd->sc_data_direction = DMA_TO_DEVICE;
1169 else
1170 cmd->sc_data_direction = DMA_FROM_DEVICE;
1171
1172 cmd->transfersize = blk_rq_bytes(req);
1173 cmd->allowed = req->retries;
1174 return BLKPREP_OK;
1175}
1176EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1177
1178/*
1179 * Setup a REQ_TYPE_FS command. These are simple read/write request
1180 * from filesystems that still need to be translated to SCSI CDBs from
1181 * the ULD.
1182 */
1183int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1184{
1185 struct scsi_cmnd *cmd;
1186 int ret = scsi_prep_state_check(sdev, req);
1187
1188 if (ret != BLKPREP_OK)
1189 return ret;
1190
1191 if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1192 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1193 ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1194 if (ret != BLKPREP_OK)
1195 return ret;
1196 }
1197
1198 /*
1199 * Filesystem requests must transfer data.
1200 */
1201 BUG_ON(!req->nr_phys_segments);
1202
1203 cmd = scsi_get_cmd_from_req(sdev, req);
1204 if (unlikely(!cmd))
1205 return BLKPREP_DEFER;
1206
1207 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1208 return scsi_init_io(cmd, GFP_ATOMIC);
1209}
1210EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1211
1212int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1213{
1214 int ret = BLKPREP_OK;
1215
1216 /*
1217 * If the device is not in running state we will reject some
1218 * or all commands.
1219 */
1220 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1221 switch (sdev->sdev_state) {
1222 case SDEV_OFFLINE:
1223 case SDEV_TRANSPORT_OFFLINE:
1224 /*
1225 * If the device is offline we refuse to process any
1226 * commands. The device must be brought online
1227 * before trying any recovery commands.
1228 */
1229 sdev_printk(KERN_ERR, sdev,
1230 "rejecting I/O to offline device\n");
1231 ret = BLKPREP_KILL;
1232 break;
1233 case SDEV_DEL:
1234 /*
1235 * If the device is fully deleted, we refuse to
1236 * process any commands as well.
1237 */
1238 sdev_printk(KERN_ERR, sdev,
1239 "rejecting I/O to dead device\n");
1240 ret = BLKPREP_KILL;
1241 break;
1242 case SDEV_QUIESCE:
1243 case SDEV_BLOCK:
1244 case SDEV_CREATED_BLOCK:
1245 /*
1246 * If the devices is blocked we defer normal commands.
1247 */
1248 if (!(req->cmd_flags & REQ_PREEMPT))
1249 ret = BLKPREP_DEFER;
1250 break;
1251 default:
1252 /*
1253 * For any other not fully online state we only allow
1254 * special commands. In particular any user initiated
1255 * command is not allowed.
1256 */
1257 if (!(req->cmd_flags & REQ_PREEMPT))
1258 ret = BLKPREP_KILL;
1259 break;
1260 }
1261 }
1262 return ret;
1263}
1264EXPORT_SYMBOL(scsi_prep_state_check);
1265
1266int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1267{
1268 struct scsi_device *sdev = q->queuedata;
1269
1270 switch (ret) {
1271 case BLKPREP_KILL:
1272 req->errors = DID_NO_CONNECT << 16;
1273 /* release the command and kill it */
1274 if (req->special) {
1275 struct scsi_cmnd *cmd = req->special;
1276 scsi_release_buffers(cmd);
1277 scsi_put_command(cmd);
1278 put_device(&sdev->sdev_gendev);
1279 req->special = NULL;
1280 }
1281 break;
1282 case BLKPREP_DEFER:
1283 /*
1284 * If we defer, the blk_peek_request() returns NULL, but the
1285 * queue must be restarted, so we schedule a callback to happen
1286 * shortly.
1287 */
1288 if (sdev->device_busy == 0)
1289 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1290 break;
1291 default:
1292 req->cmd_flags |= REQ_DONTPREP;
1293 }
1294
1295 return ret;
1296}
1297EXPORT_SYMBOL(scsi_prep_return);
1298
1299int scsi_prep_fn(struct request_queue *q, struct request *req)
1300{
1301 struct scsi_device *sdev = q->queuedata;
1302 int ret = BLKPREP_KILL;
1303
1304 if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1305 ret = scsi_setup_blk_pc_cmnd(sdev, req);
1306 return scsi_prep_return(q, req, ret);
1307}
1308EXPORT_SYMBOL(scsi_prep_fn);
1309
1310/*
1311 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1312 * return 0.
1313 *
1314 * Called with the queue_lock held.
1315 */
1316static inline int scsi_dev_queue_ready(struct request_queue *q,
1317 struct scsi_device *sdev)
1318{
1319 if (sdev->device_busy == 0 && sdev->device_blocked) {
1320 /*
1321 * unblock after device_blocked iterates to zero
1322 */
1323 if (--sdev->device_blocked == 0) {
1324 SCSI_LOG_MLQUEUE(3,
1325 sdev_printk(KERN_INFO, sdev,
1326 "unblocking device at zero depth\n"));
1327 } else {
1328 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1329 return 0;
1330 }
1331 }
1332 if (scsi_device_is_busy(sdev))
1333 return 0;
1334
1335 return 1;
1336}
1337
1338
1339/*
1340 * scsi_target_queue_ready: checks if there we can send commands to target
1341 * @sdev: scsi device on starget to check.
1342 *
1343 * Called with the host lock held.
1344 */
1345static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1346 struct scsi_device *sdev)
1347{
1348 struct scsi_target *starget = scsi_target(sdev);
1349
1350 if (starget->single_lun) {
1351 if (starget->starget_sdev_user &&
1352 starget->starget_sdev_user != sdev)
1353 return 0;
1354 starget->starget_sdev_user = sdev;
1355 }
1356
1357 if (starget->target_busy == 0 && starget->target_blocked) {
1358 /*
1359 * unblock after target_blocked iterates to zero
1360 */
1361 if (--starget->target_blocked == 0) {
1362 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1363 "unblocking target at zero depth\n"));
1364 } else
1365 return 0;
1366 }
1367
1368 if (scsi_target_is_busy(starget)) {
1369 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1370 return 0;
1371 }
1372
1373 return 1;
1374}
1375
1376/*
1377 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1378 * return 0. We must end up running the queue again whenever 0 is
1379 * returned, else IO can hang.
1380 *
1381 * Called with host_lock held.
1382 */
1383static inline int scsi_host_queue_ready(struct request_queue *q,
1384 struct Scsi_Host *shost,
1385 struct scsi_device *sdev)
1386{
1387 if (scsi_host_in_recovery(shost))
1388 return 0;
1389 if (shost->host_busy == 0 && shost->host_blocked) {
1390 /*
1391 * unblock after host_blocked iterates to zero
1392 */
1393 if (--shost->host_blocked == 0) {
1394 SCSI_LOG_MLQUEUE(3,
1395 printk("scsi%d unblocking host at zero depth\n",
1396 shost->host_no));
1397 } else {
1398 return 0;
1399 }
1400 }
1401 if (scsi_host_is_busy(shost)) {
1402 if (list_empty(&sdev->starved_entry))
1403 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1404 return 0;
1405 }
1406
1407 /* We're OK to process the command, so we can't be starved */
1408 if (!list_empty(&sdev->starved_entry))
1409 list_del_init(&sdev->starved_entry);
1410
1411 return 1;
1412}
1413
1414/*
1415 * Busy state exporting function for request stacking drivers.
1416 *
1417 * For efficiency, no lock is taken to check the busy state of
1418 * shost/starget/sdev, since the returned value is not guaranteed and
1419 * may be changed after request stacking drivers call the function,
1420 * regardless of taking lock or not.
1421 *
1422 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1423 * needs to return 'not busy'. Otherwise, request stacking drivers
1424 * may hold requests forever.
1425 */
1426static int scsi_lld_busy(struct request_queue *q)
1427{
1428 struct scsi_device *sdev = q->queuedata;
1429 struct Scsi_Host *shost;
1430
1431 if (blk_queue_dying(q))
1432 return 0;
1433
1434 shost = sdev->host;
1435
1436 /*
1437 * Ignore host/starget busy state.
1438 * Since block layer does not have a concept of fairness across
1439 * multiple queues, congestion of host/starget needs to be handled
1440 * in SCSI layer.
1441 */
1442 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1443 return 1;
1444
1445 return 0;
1446}
1447
1448/*
1449 * Kill a request for a dead device
1450 */
1451static void scsi_kill_request(struct request *req, struct request_queue *q)
1452{
1453 struct scsi_cmnd *cmd = req->special;
1454 struct scsi_device *sdev;
1455 struct scsi_target *starget;
1456 struct Scsi_Host *shost;
1457
1458 blk_start_request(req);
1459
1460 scmd_printk(KERN_INFO, cmd, "killing request\n");
1461
1462 sdev = cmd->device;
1463 starget = scsi_target(sdev);
1464 shost = sdev->host;
1465 scsi_init_cmd_errh(cmd);
1466 cmd->result = DID_NO_CONNECT << 16;
1467 atomic_inc(&cmd->device->iorequest_cnt);
1468
1469 /*
1470 * SCSI request completion path will do scsi_device_unbusy(),
1471 * bump busy counts. To bump the counters, we need to dance
1472 * with the locks as normal issue path does.
1473 */
1474 sdev->device_busy++;
1475 spin_unlock(sdev->request_queue->queue_lock);
1476 spin_lock(shost->host_lock);
1477 shost->host_busy++;
1478 starget->target_busy++;
1479 spin_unlock(shost->host_lock);
1480 spin_lock(sdev->request_queue->queue_lock);
1481
1482 blk_complete_request(req);
1483}
1484
1485static void scsi_softirq_done(struct request *rq)
1486{
1487 struct scsi_cmnd *cmd = rq->special;
1488 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1489 int disposition;
1490
1491 INIT_LIST_HEAD(&cmd->eh_entry);
1492
1493 atomic_inc(&cmd->device->iodone_cnt);
1494 if (cmd->result)
1495 atomic_inc(&cmd->device->ioerr_cnt);
1496
1497 disposition = scsi_decide_disposition(cmd);
1498 if (disposition != SUCCESS &&
1499 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1500 sdev_printk(KERN_ERR, cmd->device,
1501 "timing out command, waited %lus\n",
1502 wait_for/HZ);
1503 disposition = SUCCESS;
1504 }
1505
1506 scsi_log_completion(cmd, disposition);
1507
1508 switch (disposition) {
1509 case SUCCESS:
1510 scsi_finish_command(cmd);
1511 break;
1512 case NEEDS_RETRY:
1513 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1514 break;
1515 case ADD_TO_MLQUEUE:
1516 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1517 break;
1518 default:
1519 if (!scsi_eh_scmd_add(cmd, 0))
1520 scsi_finish_command(cmd);
1521 }
1522}
1523
1524/*
1525 * Function: scsi_request_fn()
1526 *
1527 * Purpose: Main strategy routine for SCSI.
1528 *
1529 * Arguments: q - Pointer to actual queue.
1530 *
1531 * Returns: Nothing
1532 *
1533 * Lock status: IO request lock assumed to be held when called.
1534 */
1535static void scsi_request_fn(struct request_queue *q)
1536 __releases(q->queue_lock)
1537 __acquires(q->queue_lock)
1538{
1539 struct scsi_device *sdev = q->queuedata;
1540 struct Scsi_Host *shost;
1541 struct scsi_cmnd *cmd;
1542 struct request *req;
1543
1544 /*
1545 * To start with, we keep looping until the queue is empty, or until
1546 * the host is no longer able to accept any more requests.
1547 */
1548 shost = sdev->host;
1549 for (;;) {
1550 int rtn;
1551 /*
1552 * get next queueable request. We do this early to make sure
1553 * that the request is fully prepared even if we cannot
1554 * accept it.
1555 */
1556 req = blk_peek_request(q);
1557 if (!req || !scsi_dev_queue_ready(q, sdev))
1558 break;
1559
1560 if (unlikely(!scsi_device_online(sdev))) {
1561 sdev_printk(KERN_ERR, sdev,
1562 "rejecting I/O to offline device\n");
1563 scsi_kill_request(req, q);
1564 continue;
1565 }
1566
1567
1568 /*
1569 * Remove the request from the request list.
1570 */
1571 if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1572 blk_start_request(req);
1573 sdev->device_busy++;
1574
1575 spin_unlock(q->queue_lock);
1576 cmd = req->special;
1577 if (unlikely(cmd == NULL)) {
1578 printk(KERN_CRIT "impossible request in %s.\n"
1579 "please mail a stack trace to "
1580 "linux-scsi@vger.kernel.org\n",
1581 __func__);
1582 blk_dump_rq_flags(req, "foo");
1583 BUG();
1584 }
1585 spin_lock(shost->host_lock);
1586
1587 /*
1588 * We hit this when the driver is using a host wide
1589 * tag map. For device level tag maps the queue_depth check
1590 * in the device ready fn would prevent us from trying
1591 * to allocate a tag. Since the map is a shared host resource
1592 * we add the dev to the starved list so it eventually gets
1593 * a run when a tag is freed.
1594 */
1595 if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1596 if (list_empty(&sdev->starved_entry))
1597 list_add_tail(&sdev->starved_entry,
1598 &shost->starved_list);
1599 goto not_ready;
1600 }
1601
1602 if (!scsi_target_queue_ready(shost, sdev))
1603 goto not_ready;
1604
1605 if (!scsi_host_queue_ready(q, shost, sdev))
1606 goto not_ready;
1607
1608 scsi_target(sdev)->target_busy++;
1609 shost->host_busy++;
1610
1611 /*
1612 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1613 * take the lock again.
1614 */
1615 spin_unlock_irq(shost->host_lock);
1616
1617 /*
1618 * Finally, initialize any error handling parameters, and set up
1619 * the timers for timeouts.
1620 */
1621 scsi_init_cmd_errh(cmd);
1622
1623 /*
1624 * Dispatch the command to the low-level driver.
1625 */
1626 rtn = scsi_dispatch_cmd(cmd);
1627 spin_lock_irq(q->queue_lock);
1628 if (rtn)
1629 goto out_delay;
1630 }
1631
1632 return;
1633
1634 not_ready:
1635 spin_unlock_irq(shost->host_lock);
1636
1637 /*
1638 * lock q, handle tag, requeue req, and decrement device_busy. We
1639 * must return with queue_lock held.
1640 *
1641 * Decrementing device_busy without checking it is OK, as all such
1642 * cases (host limits or settings) should run the queue at some
1643 * later time.
1644 */
1645 spin_lock_irq(q->queue_lock);
1646 blk_requeue_request(q, req);
1647 sdev->device_busy--;
1648out_delay:
1649 if (sdev->device_busy == 0)
1650 blk_delay_queue(q, SCSI_QUEUE_DELAY);
1651}
1652
1653u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1654{
1655 struct device *host_dev;
1656 u64 bounce_limit = 0xffffffff;
1657
1658 if (shost->unchecked_isa_dma)
1659 return BLK_BOUNCE_ISA;
1660 /*
1661 * Platforms with virtual-DMA translation
1662 * hardware have no practical limit.
1663 */
1664 if (!PCI_DMA_BUS_IS_PHYS)
1665 return BLK_BOUNCE_ANY;
1666
1667 host_dev = scsi_get_device(shost);
1668 if (host_dev && host_dev->dma_mask)
1669 bounce_limit = (u64)dma_max_pfn(host_dev) << PAGE_SHIFT;
1670
1671 return bounce_limit;
1672}
1673EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1674
1675struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1676 request_fn_proc *request_fn)
1677{
1678 struct request_queue *q;
1679 struct device *dev = shost->dma_dev;
1680
1681 q = blk_init_queue(request_fn, NULL);
1682 if (!q)
1683 return NULL;
1684
1685 /*
1686 * this limit is imposed by hardware restrictions
1687 */
1688 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1689 SCSI_MAX_SG_CHAIN_SEGMENTS));
1690
1691 if (scsi_host_prot_dma(shost)) {
1692 shost->sg_prot_tablesize =
1693 min_not_zero(shost->sg_prot_tablesize,
1694 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1695 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1696 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1697 }
1698
1699 blk_queue_max_hw_sectors(q, shost->max_sectors);
1700 blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1701 blk_queue_segment_boundary(q, shost->dma_boundary);
1702 dma_set_seg_boundary(dev, shost->dma_boundary);
1703
1704 blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1705
1706 if (!shost->use_clustering)
1707 q->limits.cluster = 0;
1708
1709 /*
1710 * set a reasonable default alignment on word boundaries: the
1711 * host and device may alter it using
1712 * blk_queue_update_dma_alignment() later.
1713 */
1714 blk_queue_dma_alignment(q, 0x03);
1715
1716 return q;
1717}
1718EXPORT_SYMBOL(__scsi_alloc_queue);
1719
1720struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1721{
1722 struct request_queue *q;
1723
1724 q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1725 if (!q)
1726 return NULL;
1727
1728 blk_queue_prep_rq(q, scsi_prep_fn);
1729 blk_queue_softirq_done(q, scsi_softirq_done);
1730 blk_queue_rq_timed_out(q, scsi_times_out);
1731 blk_queue_lld_busy(q, scsi_lld_busy);
1732 return q;
1733}
1734
1735/*
1736 * Function: scsi_block_requests()
1737 *
1738 * Purpose: Utility function used by low-level drivers to prevent further
1739 * commands from being queued to the device.
1740 *
1741 * Arguments: shost - Host in question
1742 *
1743 * Returns: Nothing
1744 *
1745 * Lock status: No locks are assumed held.
1746 *
1747 * Notes: There is no timer nor any other means by which the requests
1748 * get unblocked other than the low-level driver calling
1749 * scsi_unblock_requests().
1750 */
1751void scsi_block_requests(struct Scsi_Host *shost)
1752{
1753 shost->host_self_blocked = 1;
1754}
1755EXPORT_SYMBOL(scsi_block_requests);
1756
1757/*
1758 * Function: scsi_unblock_requests()
1759 *
1760 * Purpose: Utility function used by low-level drivers to allow further
1761 * commands from being queued to the device.
1762 *
1763 * Arguments: shost - Host in question
1764 *
1765 * Returns: Nothing
1766 *
1767 * Lock status: No locks are assumed held.
1768 *
1769 * Notes: There is no timer nor any other means by which the requests
1770 * get unblocked other than the low-level driver calling
1771 * scsi_unblock_requests().
1772 *
1773 * This is done as an API function so that changes to the
1774 * internals of the scsi mid-layer won't require wholesale
1775 * changes to drivers that use this feature.
1776 */
1777void scsi_unblock_requests(struct Scsi_Host *shost)
1778{
1779 shost->host_self_blocked = 0;
1780 scsi_run_host_queues(shost);
1781}
1782EXPORT_SYMBOL(scsi_unblock_requests);
1783
1784int __init scsi_init_queue(void)
1785{
1786 int i;
1787
1788 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1789 sizeof(struct scsi_data_buffer),
1790 0, 0, NULL);
1791 if (!scsi_sdb_cache) {
1792 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1793 return -ENOMEM;
1794 }
1795
1796 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1797 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1798 int size = sgp->size * sizeof(struct scatterlist);
1799
1800 sgp->slab = kmem_cache_create(sgp->name, size, 0,
1801 SLAB_HWCACHE_ALIGN, NULL);
1802 if (!sgp->slab) {
1803 printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1804 sgp->name);
1805 goto cleanup_sdb;
1806 }
1807
1808 sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1809 sgp->slab);
1810 if (!sgp->pool) {
1811 printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1812 sgp->name);
1813 goto cleanup_sdb;
1814 }
1815 }
1816
1817 return 0;
1818
1819cleanup_sdb:
1820 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1821 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1822 if (sgp->pool)
1823 mempool_destroy(sgp->pool);
1824 if (sgp->slab)
1825 kmem_cache_destroy(sgp->slab);
1826 }
1827 kmem_cache_destroy(scsi_sdb_cache);
1828
1829 return -ENOMEM;
1830}
1831
1832void scsi_exit_queue(void)
1833{
1834 int i;
1835
1836 kmem_cache_destroy(scsi_sdb_cache);
1837
1838 for (i = 0; i < SG_MEMPOOL_NR; i++) {
1839 struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1840 mempool_destroy(sgp->pool);
1841 kmem_cache_destroy(sgp->slab);
1842 }
1843}
1844
1845/**
1846 * scsi_mode_select - issue a mode select
1847 * @sdev: SCSI device to be queried
1848 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1849 * @sp: Save page bit (0 == don't save, 1 == save)
1850 * @modepage: mode page being requested
1851 * @buffer: request buffer (may not be smaller than eight bytes)
1852 * @len: length of request buffer.
1853 * @timeout: command timeout
1854 * @retries: number of retries before failing
1855 * @data: returns a structure abstracting the mode header data
1856 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1857 * must be SCSI_SENSE_BUFFERSIZE big.
1858 *
1859 * Returns zero if successful; negative error number or scsi
1860 * status on error
1861 *
1862 */
1863int
1864scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1865 unsigned char *buffer, int len, int timeout, int retries,
1866 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1867{
1868 unsigned char cmd[10];
1869 unsigned char *real_buffer;
1870 int ret;
1871
1872 memset(cmd, 0, sizeof(cmd));
1873 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1874
1875 if (sdev->use_10_for_ms) {
1876 if (len > 65535)
1877 return -EINVAL;
1878 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1879 if (!real_buffer)
1880 return -ENOMEM;
1881 memcpy(real_buffer + 8, buffer, len);
1882 len += 8;
1883 real_buffer[0] = 0;
1884 real_buffer[1] = 0;
1885 real_buffer[2] = data->medium_type;
1886 real_buffer[3] = data->device_specific;
1887 real_buffer[4] = data->longlba ? 0x01 : 0;
1888 real_buffer[5] = 0;
1889 real_buffer[6] = data->block_descriptor_length >> 8;
1890 real_buffer[7] = data->block_descriptor_length;
1891
1892 cmd[0] = MODE_SELECT_10;
1893 cmd[7] = len >> 8;
1894 cmd[8] = len;
1895 } else {
1896 if (len > 255 || data->block_descriptor_length > 255 ||
1897 data->longlba)
1898 return -EINVAL;
1899
1900 real_buffer = kmalloc(4 + len, GFP_KERNEL);
1901 if (!real_buffer)
1902 return -ENOMEM;
1903 memcpy(real_buffer + 4, buffer, len);
1904 len += 4;
1905 real_buffer[0] = 0;
1906 real_buffer[1] = data->medium_type;
1907 real_buffer[2] = data->device_specific;
1908 real_buffer[3] = data->block_descriptor_length;
1909
1910
1911 cmd[0] = MODE_SELECT;
1912 cmd[4] = len;
1913 }
1914
1915 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1916 sshdr, timeout, retries, NULL);
1917 kfree(real_buffer);
1918 return ret;
1919}
1920EXPORT_SYMBOL_GPL(scsi_mode_select);
1921
1922/**
1923 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1924 * @sdev: SCSI device to be queried
1925 * @dbd: set if mode sense will allow block descriptors to be returned
1926 * @modepage: mode page being requested
1927 * @buffer: request buffer (may not be smaller than eight bytes)
1928 * @len: length of request buffer.
1929 * @timeout: command timeout
1930 * @retries: number of retries before failing
1931 * @data: returns a structure abstracting the mode header data
1932 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1933 * must be SCSI_SENSE_BUFFERSIZE big.
1934 *
1935 * Returns zero if unsuccessful, or the header offset (either 4
1936 * or 8 depending on whether a six or ten byte command was
1937 * issued) if successful.
1938 */
1939int
1940scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1941 unsigned char *buffer, int len, int timeout, int retries,
1942 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1943{
1944 unsigned char cmd[12];
1945 int use_10_for_ms;
1946 int header_length;
1947 int result;
1948 struct scsi_sense_hdr my_sshdr;
1949
1950 memset(data, 0, sizeof(*data));
1951 memset(&cmd[0], 0, 12);
1952 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
1953 cmd[2] = modepage;
1954
1955 /* caller might not be interested in sense, but we need it */
1956 if (!sshdr)
1957 sshdr = &my_sshdr;
1958
1959 retry:
1960 use_10_for_ms = sdev->use_10_for_ms;
1961
1962 if (use_10_for_ms) {
1963 if (len < 8)
1964 len = 8;
1965
1966 cmd[0] = MODE_SENSE_10;
1967 cmd[8] = len;
1968 header_length = 8;
1969 } else {
1970 if (len < 4)
1971 len = 4;
1972
1973 cmd[0] = MODE_SENSE;
1974 cmd[4] = len;
1975 header_length = 4;
1976 }
1977
1978 memset(buffer, 0, len);
1979
1980 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1981 sshdr, timeout, retries, NULL);
1982
1983 /* This code looks awful: what it's doing is making sure an
1984 * ILLEGAL REQUEST sense return identifies the actual command
1985 * byte as the problem. MODE_SENSE commands can return
1986 * ILLEGAL REQUEST if the code page isn't supported */
1987
1988 if (use_10_for_ms && !scsi_status_is_good(result) &&
1989 (driver_byte(result) & DRIVER_SENSE)) {
1990 if (scsi_sense_valid(sshdr)) {
1991 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1992 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1993 /*
1994 * Invalid command operation code
1995 */
1996 sdev->use_10_for_ms = 0;
1997 goto retry;
1998 }
1999 }
2000 }
2001
2002 if(scsi_status_is_good(result)) {
2003 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2004 (modepage == 6 || modepage == 8))) {
2005 /* Initio breakage? */
2006 header_length = 0;
2007 data->length = 13;
2008 data->medium_type = 0;
2009 data->device_specific = 0;
2010 data->longlba = 0;
2011 data->block_descriptor_length = 0;
2012 } else if(use_10_for_ms) {
2013 data->length = buffer[0]*256 + buffer[1] + 2;
2014 data->medium_type = buffer[2];
2015 data->device_specific = buffer[3];
2016 data->longlba = buffer[4] & 0x01;
2017 data->block_descriptor_length = buffer[6]*256
2018 + buffer[7];
2019 } else {
2020 data->length = buffer[0] + 1;
2021 data->medium_type = buffer[1];
2022 data->device_specific = buffer[2];
2023 data->block_descriptor_length = buffer[3];
2024 }
2025 data->header_length = header_length;
2026 }
2027
2028 return result;
2029}
2030EXPORT_SYMBOL(scsi_mode_sense);
2031
2032/**
2033 * scsi_test_unit_ready - test if unit is ready
2034 * @sdev: scsi device to change the state of.
2035 * @timeout: command timeout
2036 * @retries: number of retries before failing
2037 * @sshdr_external: Optional pointer to struct scsi_sense_hdr for
2038 * returning sense. Make sure that this is cleared before passing
2039 * in.
2040 *
2041 * Returns zero if unsuccessful or an error if TUR failed. For
2042 * removable media, UNIT_ATTENTION sets ->changed flag.
2043 **/
2044int
2045scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2046 struct scsi_sense_hdr *sshdr_external)
2047{
2048 char cmd[] = {
2049 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2050 };
2051 struct scsi_sense_hdr *sshdr;
2052 int result;
2053
2054 if (!sshdr_external)
2055 sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2056 else
2057 sshdr = sshdr_external;
2058
2059 /* try to eat the UNIT_ATTENTION if there are enough retries */
2060 do {
2061 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2062 timeout, retries, NULL);
2063 if (sdev->removable && scsi_sense_valid(sshdr) &&
2064 sshdr->sense_key == UNIT_ATTENTION)
2065 sdev->changed = 1;
2066 } while (scsi_sense_valid(sshdr) &&
2067 sshdr->sense_key == UNIT_ATTENTION && --retries);
2068
2069 if (!sshdr_external)
2070 kfree(sshdr);
2071 return result;
2072}
2073EXPORT_SYMBOL(scsi_test_unit_ready);
2074
2075/**
2076 * scsi_device_set_state - Take the given device through the device state model.
2077 * @sdev: scsi device to change the state of.
2078 * @state: state to change to.
2079 *
2080 * Returns zero if unsuccessful or an error if the requested
2081 * transition is illegal.
2082 */
2083int
2084scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2085{
2086 enum scsi_device_state oldstate = sdev->sdev_state;
2087
2088 if (state == oldstate)
2089 return 0;
2090
2091 switch (state) {
2092 case SDEV_CREATED:
2093 switch (oldstate) {
2094 case SDEV_CREATED_BLOCK:
2095 break;
2096 default:
2097 goto illegal;
2098 }
2099 break;
2100
2101 case SDEV_RUNNING:
2102 switch (oldstate) {
2103 case SDEV_CREATED:
2104 case SDEV_OFFLINE:
2105 case SDEV_TRANSPORT_OFFLINE:
2106 case SDEV_QUIESCE:
2107 case SDEV_BLOCK:
2108 break;
2109 default:
2110 goto illegal;
2111 }
2112 break;
2113
2114 case SDEV_QUIESCE:
2115 switch (oldstate) {
2116 case SDEV_RUNNING:
2117 case SDEV_OFFLINE:
2118 case SDEV_TRANSPORT_OFFLINE:
2119 break;
2120 default:
2121 goto illegal;
2122 }
2123 break;
2124
2125 case SDEV_OFFLINE:
2126 case SDEV_TRANSPORT_OFFLINE:
2127 switch (oldstate) {
2128 case SDEV_CREATED:
2129 case SDEV_RUNNING:
2130 case SDEV_QUIESCE:
2131 case SDEV_BLOCK:
2132 break;
2133 default:
2134 goto illegal;
2135 }
2136 break;
2137
2138 case SDEV_BLOCK:
2139 switch (oldstate) {
2140 case SDEV_RUNNING:
2141 case SDEV_CREATED_BLOCK:
2142 break;
2143 default:
2144 goto illegal;
2145 }
2146 break;
2147
2148 case SDEV_CREATED_BLOCK:
2149 switch (oldstate) {
2150 case SDEV_CREATED:
2151 break;
2152 default:
2153 goto illegal;
2154 }
2155 break;
2156
2157 case SDEV_CANCEL:
2158 switch (oldstate) {
2159 case SDEV_CREATED:
2160 case SDEV_RUNNING:
2161 case SDEV_QUIESCE:
2162 case SDEV_OFFLINE:
2163 case SDEV_TRANSPORT_OFFLINE:
2164 case SDEV_BLOCK:
2165 break;
2166 default:
2167 goto illegal;
2168 }
2169 break;
2170
2171 case SDEV_DEL:
2172 switch (oldstate) {
2173 case SDEV_CREATED:
2174 case SDEV_RUNNING:
2175 case SDEV_OFFLINE:
2176 case SDEV_TRANSPORT_OFFLINE:
2177 case SDEV_CANCEL:
2178 case SDEV_CREATED_BLOCK:
2179 break;
2180 default:
2181 goto illegal;
2182 }
2183 break;
2184
2185 }
2186 sdev->sdev_state = state;
2187 return 0;
2188
2189 illegal:
2190 SCSI_LOG_ERROR_RECOVERY(1,
2191 sdev_printk(KERN_ERR, sdev,
2192 "Illegal state transition %s->%s\n",
2193 scsi_device_state_name(oldstate),
2194 scsi_device_state_name(state))
2195 );
2196 return -EINVAL;
2197}
2198EXPORT_SYMBOL(scsi_device_set_state);
2199
2200/**
2201 * sdev_evt_emit - emit a single SCSI device uevent
2202 * @sdev: associated SCSI device
2203 * @evt: event to emit
2204 *
2205 * Send a single uevent (scsi_event) to the associated scsi_device.
2206 */
2207static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2208{
2209 int idx = 0;
2210 char *envp[3];
2211
2212 switch (evt->evt_type) {
2213 case SDEV_EVT_MEDIA_CHANGE:
2214 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2215 break;
2216 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2217 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2218 break;
2219 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2220 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2221 break;
2222 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2223 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2224 break;
2225 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2226 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2227 break;
2228 case SDEV_EVT_LUN_CHANGE_REPORTED:
2229 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2230 break;
2231 default:
2232 /* do nothing */
2233 break;
2234 }
2235
2236 envp[idx++] = NULL;
2237
2238 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2239}
2240
2241/**
2242 * sdev_evt_thread - send a uevent for each scsi event
2243 * @work: work struct for scsi_device
2244 *
2245 * Dispatch queued events to their associated scsi_device kobjects
2246 * as uevents.
2247 */
2248void scsi_evt_thread(struct work_struct *work)
2249{
2250 struct scsi_device *sdev;
2251 enum scsi_device_event evt_type;
2252 LIST_HEAD(event_list);
2253
2254 sdev = container_of(work, struct scsi_device, event_work);
2255
2256 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2257 if (test_and_clear_bit(evt_type, sdev->pending_events))
2258 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2259
2260 while (1) {
2261 struct scsi_event *evt;
2262 struct list_head *this, *tmp;
2263 unsigned long flags;
2264
2265 spin_lock_irqsave(&sdev->list_lock, flags);
2266 list_splice_init(&sdev->event_list, &event_list);
2267 spin_unlock_irqrestore(&sdev->list_lock, flags);
2268
2269 if (list_empty(&event_list))
2270 break;
2271
2272 list_for_each_safe(this, tmp, &event_list) {
2273 evt = list_entry(this, struct scsi_event, node);
2274 list_del(&evt->node);
2275 scsi_evt_emit(sdev, evt);
2276 kfree(evt);
2277 }
2278 }
2279}
2280
2281/**
2282 * sdev_evt_send - send asserted event to uevent thread
2283 * @sdev: scsi_device event occurred on
2284 * @evt: event to send
2285 *
2286 * Assert scsi device event asynchronously.
2287 */
2288void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2289{
2290 unsigned long flags;
2291
2292#if 0
2293 /* FIXME: currently this check eliminates all media change events
2294 * for polled devices. Need to update to discriminate between AN
2295 * and polled events */
2296 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2297 kfree(evt);
2298 return;
2299 }
2300#endif
2301
2302 spin_lock_irqsave(&sdev->list_lock, flags);
2303 list_add_tail(&evt->node, &sdev->event_list);
2304 schedule_work(&sdev->event_work);
2305 spin_unlock_irqrestore(&sdev->list_lock, flags);
2306}
2307EXPORT_SYMBOL_GPL(sdev_evt_send);
2308
2309/**
2310 * sdev_evt_alloc - allocate a new scsi event
2311 * @evt_type: type of event to allocate
2312 * @gfpflags: GFP flags for allocation
2313 *
2314 * Allocates and returns a new scsi_event.
2315 */
2316struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2317 gfp_t gfpflags)
2318{
2319 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2320 if (!evt)
2321 return NULL;
2322
2323 evt->evt_type = evt_type;
2324 INIT_LIST_HEAD(&evt->node);
2325
2326 /* evt_type-specific initialization, if any */
2327 switch (evt_type) {
2328 case SDEV_EVT_MEDIA_CHANGE:
2329 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2330 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2331 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2332 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2333 case SDEV_EVT_LUN_CHANGE_REPORTED:
2334 default:
2335 /* do nothing */
2336 break;
2337 }
2338
2339 return evt;
2340}
2341EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2342
2343/**
2344 * sdev_evt_send_simple - send asserted event to uevent thread
2345 * @sdev: scsi_device event occurred on
2346 * @evt_type: type of event to send
2347 * @gfpflags: GFP flags for allocation
2348 *
2349 * Assert scsi device event asynchronously, given an event type.
2350 */
2351void sdev_evt_send_simple(struct scsi_device *sdev,
2352 enum scsi_device_event evt_type, gfp_t gfpflags)
2353{
2354 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2355 if (!evt) {
2356 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2357 evt_type);
2358 return;
2359 }
2360
2361 sdev_evt_send(sdev, evt);
2362}
2363EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2364
2365/**
2366 * scsi_device_quiesce - Block user issued commands.
2367 * @sdev: scsi device to quiesce.
2368 *
2369 * This works by trying to transition to the SDEV_QUIESCE state
2370 * (which must be a legal transition). When the device is in this
2371 * state, only special requests will be accepted, all others will
2372 * be deferred. Since special requests may also be requeued requests,
2373 * a successful return doesn't guarantee the device will be
2374 * totally quiescent.
2375 *
2376 * Must be called with user context, may sleep.
2377 *
2378 * Returns zero if unsuccessful or an error if not.
2379 */
2380int
2381scsi_device_quiesce(struct scsi_device *sdev)
2382{
2383 int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2384 if (err)
2385 return err;
2386
2387 scsi_run_queue(sdev->request_queue);
2388 while (sdev->device_busy) {
2389 msleep_interruptible(200);
2390 scsi_run_queue(sdev->request_queue);
2391 }
2392 return 0;
2393}
2394EXPORT_SYMBOL(scsi_device_quiesce);
2395
2396/**
2397 * scsi_device_resume - Restart user issued commands to a quiesced device.
2398 * @sdev: scsi device to resume.
2399 *
2400 * Moves the device from quiesced back to running and restarts the
2401 * queues.
2402 *
2403 * Must be called with user context, may sleep.
2404 */
2405void scsi_device_resume(struct scsi_device *sdev)
2406{
2407 /* check if the device state was mutated prior to resume, and if
2408 * so assume the state is being managed elsewhere (for example
2409 * device deleted during suspend)
2410 */
2411 if (sdev->sdev_state != SDEV_QUIESCE ||
2412 scsi_device_set_state(sdev, SDEV_RUNNING))
2413 return;
2414 scsi_run_queue(sdev->request_queue);
2415}
2416EXPORT_SYMBOL(scsi_device_resume);
2417
2418static void
2419device_quiesce_fn(struct scsi_device *sdev, void *data)
2420{
2421 scsi_device_quiesce(sdev);
2422}
2423
2424void
2425scsi_target_quiesce(struct scsi_target *starget)
2426{
2427 starget_for_each_device(starget, NULL, device_quiesce_fn);
2428}
2429EXPORT_SYMBOL(scsi_target_quiesce);
2430
2431static void
2432device_resume_fn(struct scsi_device *sdev, void *data)
2433{
2434 scsi_device_resume(sdev);
2435}
2436
2437void
2438scsi_target_resume(struct scsi_target *starget)
2439{
2440 starget_for_each_device(starget, NULL, device_resume_fn);
2441}
2442EXPORT_SYMBOL(scsi_target_resume);
2443
2444/**
2445 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2446 * @sdev: device to block
2447 *
2448 * Block request made by scsi lld's to temporarily stop all
2449 * scsi commands on the specified device. Called from interrupt
2450 * or normal process context.
2451 *
2452 * Returns zero if successful or error if not
2453 *
2454 * Notes:
2455 * This routine transitions the device to the SDEV_BLOCK state
2456 * (which must be a legal transition). When the device is in this
2457 * state, all commands are deferred until the scsi lld reenables
2458 * the device with scsi_device_unblock or device_block_tmo fires.
2459 */
2460int
2461scsi_internal_device_block(struct scsi_device *sdev)
2462{
2463 struct request_queue *q = sdev->request_queue;
2464 unsigned long flags;
2465 int err = 0;
2466
2467 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2468 if (err) {
2469 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2470
2471 if (err)
2472 return err;
2473 }
2474
2475 /*
2476 * The device has transitioned to SDEV_BLOCK. Stop the
2477 * block layer from calling the midlayer with this device's
2478 * request queue.
2479 */
2480 spin_lock_irqsave(q->queue_lock, flags);
2481 blk_stop_queue(q);
2482 spin_unlock_irqrestore(q->queue_lock, flags);
2483
2484 return 0;
2485}
2486EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2487
2488/**
2489 * scsi_internal_device_unblock - resume a device after a block request
2490 * @sdev: device to resume
2491 * @new_state: state to set devices to after unblocking
2492 *
2493 * Called by scsi lld's or the midlayer to restart the device queue
2494 * for the previously suspended scsi device. Called from interrupt or
2495 * normal process context.
2496 *
2497 * Returns zero if successful or error if not.
2498 *
2499 * Notes:
2500 * This routine transitions the device to the SDEV_RUNNING state
2501 * or to one of the offline states (which must be a legal transition)
2502 * allowing the midlayer to goose the queue for this device.
2503 */
2504int
2505scsi_internal_device_unblock(struct scsi_device *sdev,
2506 enum scsi_device_state new_state)
2507{
2508 struct request_queue *q = sdev->request_queue;
2509 unsigned long flags;
2510
2511 /*
2512 * Try to transition the scsi device to SDEV_RUNNING or one of the
2513 * offlined states and goose the device queue if successful.
2514 */
2515 if ((sdev->sdev_state == SDEV_BLOCK) ||
2516 (sdev->sdev_state == SDEV_TRANSPORT_OFFLINE))
2517 sdev->sdev_state = new_state;
2518 else if (sdev->sdev_state == SDEV_CREATED_BLOCK) {
2519 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2520 new_state == SDEV_OFFLINE)
2521 sdev->sdev_state = new_state;
2522 else
2523 sdev->sdev_state = SDEV_CREATED;
2524 } else if (sdev->sdev_state != SDEV_CANCEL &&
2525 sdev->sdev_state != SDEV_OFFLINE)
2526 return -EINVAL;
2527
2528 spin_lock_irqsave(q->queue_lock, flags);
2529 blk_start_queue(q);
2530 spin_unlock_irqrestore(q->queue_lock, flags);
2531
2532 return 0;
2533}
2534EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2535
2536static void
2537device_block(struct scsi_device *sdev, void *data)
2538{
2539 scsi_internal_device_block(sdev);
2540}
2541
2542static int
2543target_block(struct device *dev, void *data)
2544{
2545 if (scsi_is_target_device(dev))
2546 starget_for_each_device(to_scsi_target(dev), NULL,
2547 device_block);
2548 return 0;
2549}
2550
2551void
2552scsi_target_block(struct device *dev)
2553{
2554 if (scsi_is_target_device(dev))
2555 starget_for_each_device(to_scsi_target(dev), NULL,
2556 device_block);
2557 else
2558 device_for_each_child(dev, NULL, target_block);
2559}
2560EXPORT_SYMBOL_GPL(scsi_target_block);
2561
2562static void
2563device_unblock(struct scsi_device *sdev, void *data)
2564{
2565 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2566}
2567
2568static int
2569target_unblock(struct device *dev, void *data)
2570{
2571 if (scsi_is_target_device(dev))
2572 starget_for_each_device(to_scsi_target(dev), data,
2573 device_unblock);
2574 return 0;
2575}
2576
2577void
2578scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2579{
2580 if (scsi_is_target_device(dev))
2581 starget_for_each_device(to_scsi_target(dev), &new_state,
2582 device_unblock);
2583 else
2584 device_for_each_child(dev, &new_state, target_unblock);
2585}
2586EXPORT_SYMBOL_GPL(scsi_target_unblock);
2587
2588/**
2589 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2590 * @sgl: scatter-gather list
2591 * @sg_count: number of segments in sg
2592 * @offset: offset in bytes into sg, on return offset into the mapped area
2593 * @len: bytes to map, on return number of bytes mapped
2594 *
2595 * Returns virtual address of the start of the mapped page
2596 */
2597void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2598 size_t *offset, size_t *len)
2599{
2600 int i;
2601 size_t sg_len = 0, len_complete = 0;
2602 struct scatterlist *sg;
2603 struct page *page;
2604
2605 WARN_ON(!irqs_disabled());
2606
2607 for_each_sg(sgl, sg, sg_count, i) {
2608 len_complete = sg_len; /* Complete sg-entries */
2609 sg_len += sg->length;
2610 if (sg_len > *offset)
2611 break;
2612 }
2613
2614 if (unlikely(i == sg_count)) {
2615 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2616 "elements %d\n",
2617 __func__, sg_len, *offset, sg_count);
2618 WARN_ON(1);
2619 return NULL;
2620 }
2621
2622 /* Offset starting from the beginning of first page in this sg-entry */
2623 *offset = *offset - len_complete + sg->offset;
2624
2625 /* Assumption: contiguous pages can be accessed as "page + i" */
2626 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2627 *offset &= ~PAGE_MASK;
2628
2629 /* Bytes in this sg-entry from *offset to the end of the page */
2630 sg_len = PAGE_SIZE - *offset;
2631 if (*len > sg_len)
2632 *len = sg_len;
2633
2634 return kmap_atomic(page);
2635}
2636EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2637
2638/**
2639 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2640 * @virt: virtual address to be unmapped
2641 */
2642void scsi_kunmap_atomic_sg(void *virt)
2643{
2644 kunmap_atomic(virt);
2645}
2646EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2647
2648void sdev_disable_disk_events(struct scsi_device *sdev)
2649{
2650 atomic_inc(&sdev->disk_events_disable_depth);
2651}
2652EXPORT_SYMBOL(sdev_disable_disk_events);
2653
2654void sdev_enable_disk_events(struct scsi_device *sdev)
2655{
2656 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2657 return;
2658 atomic_dec(&sdev->disk_events_disable_depth);
2659}
2660EXPORT_SYMBOL(sdev_enable_disk_events);