Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
5 *
6 * SCSI queueing library.
7 * Initial versions: Eric Youngdale (eric@andante.org).
8 * Based upon conversations with large numbers
9 * of people at Linux Expo.
10 */
11
12#include <linux/bio.h>
13#include <linux/bitops.h>
14#include <linux/blkdev.h>
15#include <linux/completion.h>
16#include <linux/kernel.h>
17#include <linux/export.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#include <linux/blk-mq.h>
24#include <linux/ratelimit.h>
25#include <asm/unaligned.h>
26
27#include <scsi/scsi.h>
28#include <scsi/scsi_cmnd.h>
29#include <scsi/scsi_dbg.h>
30#include <scsi/scsi_device.h>
31#include <scsi/scsi_driver.h>
32#include <scsi/scsi_eh.h>
33#include <scsi/scsi_host.h>
34#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
35#include <scsi/scsi_dh.h>
36
37#include <trace/events/scsi.h>
38
39#include "scsi_debugfs.h"
40#include "scsi_priv.h"
41#include "scsi_logging.h"
42
43/*
44 * Size of integrity metadata is usually small, 1 inline sg should
45 * cover normal cases.
46 */
47#ifdef CONFIG_ARCH_NO_SG_CHAIN
48#define SCSI_INLINE_PROT_SG_CNT 0
49#define SCSI_INLINE_SG_CNT 0
50#else
51#define SCSI_INLINE_PROT_SG_CNT 1
52#define SCSI_INLINE_SG_CNT 2
53#endif
54
55static struct kmem_cache *scsi_sense_cache;
56static struct kmem_cache *scsi_sense_isadma_cache;
57static DEFINE_MUTEX(scsi_sense_cache_mutex);
58
59static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60
61static inline struct kmem_cache *
62scsi_select_sense_cache(bool unchecked_isa_dma)
63{
64 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
65}
66
67static void scsi_free_sense_buffer(bool unchecked_isa_dma,
68 unsigned char *sense_buffer)
69{
70 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
71 sense_buffer);
72}
73
74static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
75 gfp_t gfp_mask, int numa_node)
76{
77 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
78 gfp_mask, numa_node);
79}
80
81int scsi_init_sense_cache(struct Scsi_Host *shost)
82{
83 struct kmem_cache *cache;
84 int ret = 0;
85
86 mutex_lock(&scsi_sense_cache_mutex);
87 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
88 if (cache)
89 goto exit;
90
91 if (shost->unchecked_isa_dma) {
92 scsi_sense_isadma_cache =
93 kmem_cache_create("scsi_sense_cache(DMA)",
94 SCSI_SENSE_BUFFERSIZE, 0,
95 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
96 if (!scsi_sense_isadma_cache)
97 ret = -ENOMEM;
98 } else {
99 scsi_sense_cache =
100 kmem_cache_create_usercopy("scsi_sense_cache",
101 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
102 0, SCSI_SENSE_BUFFERSIZE, NULL);
103 if (!scsi_sense_cache)
104 ret = -ENOMEM;
105 }
106 exit:
107 mutex_unlock(&scsi_sense_cache_mutex);
108 return ret;
109}
110
111/*
112 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
113 * not change behaviour from the previous unplug mechanism, experimentation
114 * may prove this needs changing.
115 */
116#define SCSI_QUEUE_DELAY 3
117
118static void
119scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
120{
121 struct Scsi_Host *host = cmd->device->host;
122 struct scsi_device *device = cmd->device;
123 struct scsi_target *starget = scsi_target(device);
124
125 /*
126 * Set the appropriate busy bit for the device/host.
127 *
128 * If the host/device isn't busy, assume that something actually
129 * completed, and that we should be able to queue a command now.
130 *
131 * Note that the prior mid-layer assumption that any host could
132 * always queue at least one command is now broken. The mid-layer
133 * will implement a user specifiable stall (see
134 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
135 * if a command is requeued with no other commands outstanding
136 * either for the device or for the host.
137 */
138 switch (reason) {
139 case SCSI_MLQUEUE_HOST_BUSY:
140 atomic_set(&host->host_blocked, host->max_host_blocked);
141 break;
142 case SCSI_MLQUEUE_DEVICE_BUSY:
143 case SCSI_MLQUEUE_EH_RETRY:
144 atomic_set(&device->device_blocked,
145 device->max_device_blocked);
146 break;
147 case SCSI_MLQUEUE_TARGET_BUSY:
148 atomic_set(&starget->target_blocked,
149 starget->max_target_blocked);
150 break;
151 }
152}
153
154static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
155{
156 if (cmd->request->rq_flags & RQF_DONTPREP) {
157 cmd->request->rq_flags &= ~RQF_DONTPREP;
158 scsi_mq_uninit_cmd(cmd);
159 } else {
160 WARN_ON_ONCE(true);
161 }
162 blk_mq_requeue_request(cmd->request, true);
163}
164
165/**
166 * __scsi_queue_insert - private queue insertion
167 * @cmd: The SCSI command being requeued
168 * @reason: The reason for the requeue
169 * @unbusy: Whether the queue should be unbusied
170 *
171 * This is a private queue insertion. The public interface
172 * scsi_queue_insert() always assumes the queue should be unbusied
173 * because it's always called before the completion. This function is
174 * for a requeue after completion, which should only occur in this
175 * file.
176 */
177static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
178{
179 struct scsi_device *device = cmd->device;
180
181 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
182 "Inserting command %p into mlqueue\n", cmd));
183
184 scsi_set_blocked(cmd, reason);
185
186 /*
187 * Decrement the counters, since these commands are no longer
188 * active on the host/device.
189 */
190 if (unbusy)
191 scsi_device_unbusy(device, cmd);
192
193 /*
194 * Requeue this command. It will go before all other commands
195 * that are already in the queue. Schedule requeue work under
196 * lock such that the kblockd_schedule_work() call happens
197 * before blk_cleanup_queue() finishes.
198 */
199 cmd->result = 0;
200
201 blk_mq_requeue_request(cmd->request, true);
202}
203
204/**
205 * scsi_queue_insert - Reinsert a command in the queue.
206 * @cmd: command that we are adding to queue.
207 * @reason: why we are inserting command to queue.
208 *
209 * We do this for one of two cases. Either the host is busy and it cannot accept
210 * any more commands for the time being, or the device returned QUEUE_FULL and
211 * can accept no more commands.
212 *
213 * Context: This could be called either from an interrupt context or a normal
214 * process context.
215 */
216void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
217{
218 __scsi_queue_insert(cmd, reason, true);
219}
220
221
222/**
223 * __scsi_execute - insert request and wait for the result
224 * @sdev: scsi device
225 * @cmd: scsi command
226 * @data_direction: data direction
227 * @buffer: data buffer
228 * @bufflen: len of buffer
229 * @sense: optional sense buffer
230 * @sshdr: optional decoded sense header
231 * @timeout: request timeout in seconds
232 * @retries: number of times to retry request
233 * @flags: flags for ->cmd_flags
234 * @rq_flags: flags for ->rq_flags
235 * @resid: optional residual length
236 *
237 * Returns the scsi_cmnd result field if a command was executed, or a negative
238 * Linux error code if we didn't get that far.
239 */
240int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
241 int data_direction, void *buffer, unsigned bufflen,
242 unsigned char *sense, struct scsi_sense_hdr *sshdr,
243 int timeout, int retries, u64 flags, req_flags_t rq_flags,
244 int *resid)
245{
246 struct request *req;
247 struct scsi_request *rq;
248 int ret = DRIVER_ERROR << 24;
249
250 req = blk_get_request(sdev->request_queue,
251 data_direction == DMA_TO_DEVICE ?
252 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
253 if (IS_ERR(req))
254 return ret;
255 rq = scsi_req(req);
256
257 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
258 buffer, bufflen, GFP_NOIO))
259 goto out;
260
261 rq->cmd_len = COMMAND_SIZE(cmd[0]);
262 memcpy(rq->cmd, cmd, rq->cmd_len);
263 rq->retries = retries;
264 req->timeout = timeout;
265 req->cmd_flags |= flags;
266 req->rq_flags |= rq_flags | RQF_QUIET;
267
268 /*
269 * head injection *required* here otherwise quiesce won't work
270 */
271 blk_execute_rq(req->q, NULL, req, 1);
272
273 /*
274 * Some devices (USB mass-storage in particular) may transfer
275 * garbage data together with a residue indicating that the data
276 * is invalid. Prevent the garbage from being misinterpreted
277 * and prevent security leaks by zeroing out the excess data.
278 */
279 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
280 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
281
282 if (resid)
283 *resid = rq->resid_len;
284 if (sense && rq->sense_len)
285 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
286 if (sshdr)
287 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
288 ret = rq->result;
289 out:
290 blk_put_request(req);
291
292 return ret;
293}
294EXPORT_SYMBOL(__scsi_execute);
295
296/**
297 * scsi_init_cmd_errh - Initialize cmd fields related to error handling.
298 * @cmd: command that is ready to be queued.
299 *
300 * This function has the job of initializing a number of fields related to error
301 * handling. Typically this will be called once for each command, as required.
302 */
303static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
304{
305 scsi_set_resid(cmd, 0);
306 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
307 if (cmd->cmd_len == 0)
308 cmd->cmd_len = scsi_command_size(cmd->cmnd);
309}
310
311/*
312 * Wake up the error handler if necessary. Avoid as follows that the error
313 * handler is not woken up if host in-flight requests number ==
314 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
315 * with an RCU read lock in this function to ensure that this function in
316 * its entirety either finishes before scsi_eh_scmd_add() increases the
317 * host_failed counter or that it notices the shost state change made by
318 * scsi_eh_scmd_add().
319 */
320static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
321{
322 unsigned long flags;
323
324 rcu_read_lock();
325 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
326 if (unlikely(scsi_host_in_recovery(shost))) {
327 spin_lock_irqsave(shost->host_lock, flags);
328 if (shost->host_failed || shost->host_eh_scheduled)
329 scsi_eh_wakeup(shost);
330 spin_unlock_irqrestore(shost->host_lock, flags);
331 }
332 rcu_read_unlock();
333}
334
335void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
336{
337 struct Scsi_Host *shost = sdev->host;
338 struct scsi_target *starget = scsi_target(sdev);
339
340 scsi_dec_host_busy(shost, cmd);
341
342 if (starget->can_queue > 0)
343 atomic_dec(&starget->target_busy);
344
345 atomic_dec(&sdev->device_busy);
346}
347
348static void scsi_kick_queue(struct request_queue *q)
349{
350 blk_mq_run_hw_queues(q, false);
351}
352
353/*
354 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
355 * and call blk_run_queue for all the scsi_devices on the target -
356 * including current_sdev first.
357 *
358 * Called with *no* scsi locks held.
359 */
360static void scsi_single_lun_run(struct scsi_device *current_sdev)
361{
362 struct Scsi_Host *shost = current_sdev->host;
363 struct scsi_device *sdev, *tmp;
364 struct scsi_target *starget = scsi_target(current_sdev);
365 unsigned long flags;
366
367 spin_lock_irqsave(shost->host_lock, flags);
368 starget->starget_sdev_user = NULL;
369 spin_unlock_irqrestore(shost->host_lock, flags);
370
371 /*
372 * Call blk_run_queue for all LUNs on the target, starting with
373 * current_sdev. We race with others (to set starget_sdev_user),
374 * but in most cases, we will be first. Ideally, each LU on the
375 * target would get some limited time or requests on the target.
376 */
377 scsi_kick_queue(current_sdev->request_queue);
378
379 spin_lock_irqsave(shost->host_lock, flags);
380 if (starget->starget_sdev_user)
381 goto out;
382 list_for_each_entry_safe(sdev, tmp, &starget->devices,
383 same_target_siblings) {
384 if (sdev == current_sdev)
385 continue;
386 if (scsi_device_get(sdev))
387 continue;
388
389 spin_unlock_irqrestore(shost->host_lock, flags);
390 scsi_kick_queue(sdev->request_queue);
391 spin_lock_irqsave(shost->host_lock, flags);
392
393 scsi_device_put(sdev);
394 }
395 out:
396 spin_unlock_irqrestore(shost->host_lock, flags);
397}
398
399static inline bool scsi_device_is_busy(struct scsi_device *sdev)
400{
401 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
402 return true;
403 if (atomic_read(&sdev->device_blocked) > 0)
404 return true;
405 return false;
406}
407
408static inline bool scsi_target_is_busy(struct scsi_target *starget)
409{
410 if (starget->can_queue > 0) {
411 if (atomic_read(&starget->target_busy) >= starget->can_queue)
412 return true;
413 if (atomic_read(&starget->target_blocked) > 0)
414 return true;
415 }
416 return false;
417}
418
419static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
420{
421 if (atomic_read(&shost->host_blocked) > 0)
422 return true;
423 if (shost->host_self_blocked)
424 return true;
425 return false;
426}
427
428static void scsi_starved_list_run(struct Scsi_Host *shost)
429{
430 LIST_HEAD(starved_list);
431 struct scsi_device *sdev;
432 unsigned long flags;
433
434 spin_lock_irqsave(shost->host_lock, flags);
435 list_splice_init(&shost->starved_list, &starved_list);
436
437 while (!list_empty(&starved_list)) {
438 struct request_queue *slq;
439
440 /*
441 * As long as shost is accepting commands and we have
442 * starved queues, call blk_run_queue. scsi_request_fn
443 * drops the queue_lock and can add us back to the
444 * starved_list.
445 *
446 * host_lock protects the starved_list and starved_entry.
447 * scsi_request_fn must get the host_lock before checking
448 * or modifying starved_list or starved_entry.
449 */
450 if (scsi_host_is_busy(shost))
451 break;
452
453 sdev = list_entry(starved_list.next,
454 struct scsi_device, starved_entry);
455 list_del_init(&sdev->starved_entry);
456 if (scsi_target_is_busy(scsi_target(sdev))) {
457 list_move_tail(&sdev->starved_entry,
458 &shost->starved_list);
459 continue;
460 }
461
462 /*
463 * Once we drop the host lock, a racing scsi_remove_device()
464 * call may remove the sdev from the starved list and destroy
465 * it and the queue. Mitigate by taking a reference to the
466 * queue and never touching the sdev again after we drop the
467 * host lock. Note: if __scsi_remove_device() invokes
468 * blk_cleanup_queue() before the queue is run from this
469 * function then blk_run_queue() will return immediately since
470 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
471 */
472 slq = sdev->request_queue;
473 if (!blk_get_queue(slq))
474 continue;
475 spin_unlock_irqrestore(shost->host_lock, flags);
476
477 scsi_kick_queue(slq);
478 blk_put_queue(slq);
479
480 spin_lock_irqsave(shost->host_lock, flags);
481 }
482 /* put any unprocessed entries back */
483 list_splice(&starved_list, &shost->starved_list);
484 spin_unlock_irqrestore(shost->host_lock, flags);
485}
486
487/**
488 * scsi_run_queue - Select a proper request queue to serve next.
489 * @q: last request's queue
490 *
491 * The previous command was completely finished, start a new one if possible.
492 */
493static void scsi_run_queue(struct request_queue *q)
494{
495 struct scsi_device *sdev = q->queuedata;
496
497 if (scsi_target(sdev)->single_lun)
498 scsi_single_lun_run(sdev);
499 if (!list_empty(&sdev->host->starved_list))
500 scsi_starved_list_run(sdev->host);
501
502 blk_mq_run_hw_queues(q, false);
503}
504
505void scsi_requeue_run_queue(struct work_struct *work)
506{
507 struct scsi_device *sdev;
508 struct request_queue *q;
509
510 sdev = container_of(work, struct scsi_device, requeue_work);
511 q = sdev->request_queue;
512 scsi_run_queue(q);
513}
514
515void scsi_run_host_queues(struct Scsi_Host *shost)
516{
517 struct scsi_device *sdev;
518
519 shost_for_each_device(sdev, shost)
520 scsi_run_queue(sdev->request_queue);
521}
522
523static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
524{
525 if (!blk_rq_is_passthrough(cmd->request)) {
526 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
527
528 if (drv->uninit_command)
529 drv->uninit_command(cmd);
530 }
531}
532
533static void scsi_free_sgtables(struct scsi_cmnd *cmd)
534{
535 if (cmd->sdb.table.nents)
536 sg_free_table_chained(&cmd->sdb.table,
537 SCSI_INLINE_SG_CNT);
538 if (scsi_prot_sg_count(cmd))
539 sg_free_table_chained(&cmd->prot_sdb->table,
540 SCSI_INLINE_PROT_SG_CNT);
541}
542
543static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
544{
545 scsi_free_sgtables(cmd);
546 scsi_uninit_cmd(cmd);
547}
548
549static void scsi_run_queue_async(struct scsi_device *sdev)
550{
551 if (scsi_target(sdev)->single_lun ||
552 !list_empty(&sdev->host->starved_list))
553 kblockd_schedule_work(&sdev->requeue_work);
554 else
555 blk_mq_run_hw_queues(sdev->request_queue, true);
556}
557
558/* Returns false when no more bytes to process, true if there are more */
559static bool scsi_end_request(struct request *req, blk_status_t error,
560 unsigned int bytes)
561{
562 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
563 struct scsi_device *sdev = cmd->device;
564 struct request_queue *q = sdev->request_queue;
565
566 if (blk_update_request(req, error, bytes))
567 return true;
568
569 if (blk_queue_add_random(q))
570 add_disk_randomness(req->rq_disk);
571
572 if (!blk_rq_is_scsi(req)) {
573 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
574 cmd->flags &= ~SCMD_INITIALIZED;
575 }
576
577 /*
578 * Calling rcu_barrier() is not necessary here because the
579 * SCSI error handler guarantees that the function called by
580 * call_rcu() has been called before scsi_end_request() is
581 * called.
582 */
583 destroy_rcu_head(&cmd->rcu);
584
585 /*
586 * In the MQ case the command gets freed by __blk_mq_end_request,
587 * so we have to do all cleanup that depends on it earlier.
588 *
589 * We also can't kick the queues from irq context, so we
590 * will have to defer it to a workqueue.
591 */
592 scsi_mq_uninit_cmd(cmd);
593
594 /*
595 * queue is still alive, so grab the ref for preventing it
596 * from being cleaned up during running queue.
597 */
598 percpu_ref_get(&q->q_usage_counter);
599
600 __blk_mq_end_request(req, error);
601
602 scsi_run_queue_async(sdev);
603
604 percpu_ref_put(&q->q_usage_counter);
605 return false;
606}
607
608/**
609 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
610 * @cmd: SCSI command
611 * @result: scsi error code
612 *
613 * Translate a SCSI result code into a blk_status_t value. May reset the host
614 * byte of @cmd->result.
615 */
616static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
617{
618 switch (host_byte(result)) {
619 case DID_OK:
620 /*
621 * Also check the other bytes than the status byte in result
622 * to handle the case when a SCSI LLD sets result to
623 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
624 */
625 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
626 return BLK_STS_OK;
627 return BLK_STS_IOERR;
628 case DID_TRANSPORT_FAILFAST:
629 return BLK_STS_TRANSPORT;
630 case DID_TARGET_FAILURE:
631 set_host_byte(cmd, DID_OK);
632 return BLK_STS_TARGET;
633 case DID_NEXUS_FAILURE:
634 set_host_byte(cmd, DID_OK);
635 return BLK_STS_NEXUS;
636 case DID_ALLOC_FAILURE:
637 set_host_byte(cmd, DID_OK);
638 return BLK_STS_NOSPC;
639 case DID_MEDIUM_ERROR:
640 set_host_byte(cmd, DID_OK);
641 return BLK_STS_MEDIUM;
642 default:
643 return BLK_STS_IOERR;
644 }
645}
646
647/* Helper for scsi_io_completion() when "reprep" action required. */
648static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
649 struct request_queue *q)
650{
651 /* A new command will be prepared and issued. */
652 scsi_mq_requeue_cmd(cmd);
653}
654
655/* Helper for scsi_io_completion() when special action required. */
656static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
657{
658 struct request_queue *q = cmd->device->request_queue;
659 struct request *req = cmd->request;
660 int level = 0;
661 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
662 ACTION_DELAYED_RETRY} action;
663 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
664 struct scsi_sense_hdr sshdr;
665 bool sense_valid;
666 bool sense_current = true; /* false implies "deferred sense" */
667 blk_status_t blk_stat;
668
669 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
670 if (sense_valid)
671 sense_current = !scsi_sense_is_deferred(&sshdr);
672
673 blk_stat = scsi_result_to_blk_status(cmd, result);
674
675 if (host_byte(result) == DID_RESET) {
676 /* Third party bus reset or reset for error recovery
677 * reasons. Just retry the command and see what
678 * happens.
679 */
680 action = ACTION_RETRY;
681 } else if (sense_valid && sense_current) {
682 switch (sshdr.sense_key) {
683 case UNIT_ATTENTION:
684 if (cmd->device->removable) {
685 /* Detected disc change. Set a bit
686 * and quietly refuse further access.
687 */
688 cmd->device->changed = 1;
689 action = ACTION_FAIL;
690 } else {
691 /* Must have been a power glitch, or a
692 * bus reset. Could not have been a
693 * media change, so we just retry the
694 * command and see what happens.
695 */
696 action = ACTION_RETRY;
697 }
698 break;
699 case ILLEGAL_REQUEST:
700 /* If we had an ILLEGAL REQUEST returned, then
701 * we may have performed an unsupported
702 * command. The only thing this should be
703 * would be a ten byte read where only a six
704 * byte read was supported. Also, on a system
705 * where READ CAPACITY failed, we may have
706 * read past the end of the disk.
707 */
708 if ((cmd->device->use_10_for_rw &&
709 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
710 (cmd->cmnd[0] == READ_10 ||
711 cmd->cmnd[0] == WRITE_10)) {
712 /* This will issue a new 6-byte command. */
713 cmd->device->use_10_for_rw = 0;
714 action = ACTION_REPREP;
715 } else if (sshdr.asc == 0x10) /* DIX */ {
716 action = ACTION_FAIL;
717 blk_stat = BLK_STS_PROTECTION;
718 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
719 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
720 action = ACTION_FAIL;
721 blk_stat = BLK_STS_TARGET;
722 } else
723 action = ACTION_FAIL;
724 break;
725 case ABORTED_COMMAND:
726 action = ACTION_FAIL;
727 if (sshdr.asc == 0x10) /* DIF */
728 blk_stat = BLK_STS_PROTECTION;
729 break;
730 case NOT_READY:
731 /* If the device is in the process of becoming
732 * ready, or has a temporary blockage, retry.
733 */
734 if (sshdr.asc == 0x04) {
735 switch (sshdr.ascq) {
736 case 0x01: /* becoming ready */
737 case 0x04: /* format in progress */
738 case 0x05: /* rebuild in progress */
739 case 0x06: /* recalculation in progress */
740 case 0x07: /* operation in progress */
741 case 0x08: /* Long write in progress */
742 case 0x09: /* self test in progress */
743 case 0x14: /* space allocation in progress */
744 case 0x1a: /* start stop unit in progress */
745 case 0x1b: /* sanitize in progress */
746 case 0x1d: /* configuration in progress */
747 case 0x24: /* depopulation in progress */
748 action = ACTION_DELAYED_RETRY;
749 break;
750 default:
751 action = ACTION_FAIL;
752 break;
753 }
754 } else
755 action = ACTION_FAIL;
756 break;
757 case VOLUME_OVERFLOW:
758 /* See SSC3rXX or current. */
759 action = ACTION_FAIL;
760 break;
761 default:
762 action = ACTION_FAIL;
763 break;
764 }
765 } else
766 action = ACTION_FAIL;
767
768 if (action != ACTION_FAIL &&
769 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
770 action = ACTION_FAIL;
771
772 switch (action) {
773 case ACTION_FAIL:
774 /* Give up and fail the remainder of the request */
775 if (!(req->rq_flags & RQF_QUIET)) {
776 static DEFINE_RATELIMIT_STATE(_rs,
777 DEFAULT_RATELIMIT_INTERVAL,
778 DEFAULT_RATELIMIT_BURST);
779
780 if (unlikely(scsi_logging_level))
781 level =
782 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
783 SCSI_LOG_MLCOMPLETE_BITS);
784
785 /*
786 * if logging is enabled the failure will be printed
787 * in scsi_log_completion(), so avoid duplicate messages
788 */
789 if (!level && __ratelimit(&_rs)) {
790 scsi_print_result(cmd, NULL, FAILED);
791 if (driver_byte(result) == DRIVER_SENSE)
792 scsi_print_sense(cmd);
793 scsi_print_command(cmd);
794 }
795 }
796 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
797 return;
798 fallthrough;
799 case ACTION_REPREP:
800 scsi_io_completion_reprep(cmd, q);
801 break;
802 case ACTION_RETRY:
803 /* Retry the same command immediately */
804 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
805 break;
806 case ACTION_DELAYED_RETRY:
807 /* Retry the same command after a delay */
808 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
809 break;
810 }
811}
812
813/*
814 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
815 * new result that may suppress further error checking. Also modifies
816 * *blk_statp in some cases.
817 */
818static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
819 blk_status_t *blk_statp)
820{
821 bool sense_valid;
822 bool sense_current = true; /* false implies "deferred sense" */
823 struct request *req = cmd->request;
824 struct scsi_sense_hdr sshdr;
825
826 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
827 if (sense_valid)
828 sense_current = !scsi_sense_is_deferred(&sshdr);
829
830 if (blk_rq_is_passthrough(req)) {
831 if (sense_valid) {
832 /*
833 * SG_IO wants current and deferred errors
834 */
835 scsi_req(req)->sense_len =
836 min(8 + cmd->sense_buffer[7],
837 SCSI_SENSE_BUFFERSIZE);
838 }
839 if (sense_current)
840 *blk_statp = scsi_result_to_blk_status(cmd, result);
841 } else if (blk_rq_bytes(req) == 0 && sense_current) {
842 /*
843 * Flush commands do not transfers any data, and thus cannot use
844 * good_bytes != blk_rq_bytes(req) as the signal for an error.
845 * This sets *blk_statp explicitly for the problem case.
846 */
847 *blk_statp = scsi_result_to_blk_status(cmd, result);
848 }
849 /*
850 * Recovered errors need reporting, but they're always treated as
851 * success, so fiddle the result code here. For passthrough requests
852 * we already took a copy of the original into sreq->result which
853 * is what gets returned to the user
854 */
855 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
856 bool do_print = true;
857 /*
858 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
859 * skip print since caller wants ATA registers. Only occurs
860 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
861 */
862 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
863 do_print = false;
864 else if (req->rq_flags & RQF_QUIET)
865 do_print = false;
866 if (do_print)
867 scsi_print_sense(cmd);
868 result = 0;
869 /* for passthrough, *blk_statp may be set */
870 *blk_statp = BLK_STS_OK;
871 }
872 /*
873 * Another corner case: the SCSI status byte is non-zero but 'good'.
874 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
875 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
876 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
877 * intermediate statuses (both obsolete in SAM-4) as good.
878 */
879 if (status_byte(result) && scsi_status_is_good(result)) {
880 result = 0;
881 *blk_statp = BLK_STS_OK;
882 }
883 return result;
884}
885
886/**
887 * scsi_io_completion - Completion processing for SCSI commands.
888 * @cmd: command that is finished.
889 * @good_bytes: number of processed bytes.
890 *
891 * We will finish off the specified number of sectors. If we are done, the
892 * command block will be released and the queue function will be goosed. If we
893 * are not done then we have to figure out what to do next:
894 *
895 * a) We can call scsi_io_completion_reprep(). The request will be
896 * unprepared and put back on the queue. Then a new command will
897 * be created for it. This should be used if we made forward
898 * progress, or if we want to switch from READ(10) to READ(6) for
899 * example.
900 *
901 * b) We can call scsi_io_completion_action(). The request will be
902 * put back on the queue and retried using the same command as
903 * before, possibly after a delay.
904 *
905 * c) We can call scsi_end_request() with blk_stat other than
906 * BLK_STS_OK, to fail the remainder of the request.
907 */
908void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
909{
910 int result = cmd->result;
911 struct request_queue *q = cmd->device->request_queue;
912 struct request *req = cmd->request;
913 blk_status_t blk_stat = BLK_STS_OK;
914
915 if (unlikely(result)) /* a nz result may or may not be an error */
916 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
917
918 if (unlikely(blk_rq_is_passthrough(req))) {
919 /*
920 * scsi_result_to_blk_status may have reset the host_byte
921 */
922 scsi_req(req)->result = cmd->result;
923 }
924
925 /*
926 * Next deal with any sectors which we were able to correctly
927 * handle.
928 */
929 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
930 "%u sectors total, %d bytes done.\n",
931 blk_rq_sectors(req), good_bytes));
932
933 /*
934 * Failed, zero length commands always need to drop down
935 * to retry code. Fast path should return in this block.
936 */
937 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
938 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
939 return; /* no bytes remaining */
940 }
941
942 /* Kill remainder if no retries. */
943 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
944 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
945 WARN_ONCE(true,
946 "Bytes remaining after failed, no-retry command");
947 return;
948 }
949
950 /*
951 * If there had been no error, but we have leftover bytes in the
952 * requeues just queue the command up again.
953 */
954 if (likely(result == 0))
955 scsi_io_completion_reprep(cmd, q);
956 else
957 scsi_io_completion_action(cmd, result);
958}
959
960static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
961 struct request *rq)
962{
963 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
964 !op_is_write(req_op(rq)) &&
965 sdev->host->hostt->dma_need_drain(rq);
966}
967
968/**
969 * scsi_init_io - SCSI I/O initialization function.
970 * @cmd: command descriptor we wish to initialize
971 *
972 * Returns:
973 * * BLK_STS_OK - on success
974 * * BLK_STS_RESOURCE - if the failure is retryable
975 * * BLK_STS_IOERR - if the failure is fatal
976 */
977blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
978{
979 struct scsi_device *sdev = cmd->device;
980 struct request *rq = cmd->request;
981 unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
982 struct scatterlist *last_sg = NULL;
983 blk_status_t ret;
984 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
985 int count;
986
987 if (WARN_ON_ONCE(!nr_segs))
988 return BLK_STS_IOERR;
989
990 /*
991 * Make sure there is space for the drain. The driver must adjust
992 * max_hw_segments to be prepared for this.
993 */
994 if (need_drain)
995 nr_segs++;
996
997 /*
998 * If sg table allocation fails, requeue request later.
999 */
1000 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1001 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1002 return BLK_STS_RESOURCE;
1003
1004 /*
1005 * Next, walk the list, and fill in the addresses and sizes of
1006 * each segment.
1007 */
1008 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1009
1010 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1011 unsigned int pad_len =
1012 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1013
1014 last_sg->length += pad_len;
1015 cmd->extra_len += pad_len;
1016 }
1017
1018 if (need_drain) {
1019 sg_unmark_end(last_sg);
1020 last_sg = sg_next(last_sg);
1021 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1022 sg_mark_end(last_sg);
1023
1024 cmd->extra_len += sdev->dma_drain_len;
1025 count++;
1026 }
1027
1028 BUG_ON(count > cmd->sdb.table.nents);
1029 cmd->sdb.table.nents = count;
1030 cmd->sdb.length = blk_rq_payload_bytes(rq);
1031
1032 if (blk_integrity_rq(rq)) {
1033 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1034 int ivecs;
1035
1036 if (WARN_ON_ONCE(!prot_sdb)) {
1037 /*
1038 * This can happen if someone (e.g. multipath)
1039 * queues a command to a device on an adapter
1040 * that does not support DIX.
1041 */
1042 ret = BLK_STS_IOERR;
1043 goto out_free_sgtables;
1044 }
1045
1046 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1047
1048 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1049 prot_sdb->table.sgl,
1050 SCSI_INLINE_PROT_SG_CNT)) {
1051 ret = BLK_STS_RESOURCE;
1052 goto out_free_sgtables;
1053 }
1054
1055 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1056 prot_sdb->table.sgl);
1057 BUG_ON(count > ivecs);
1058 BUG_ON(count > queue_max_integrity_segments(rq->q));
1059
1060 cmd->prot_sdb = prot_sdb;
1061 cmd->prot_sdb->table.nents = count;
1062 }
1063
1064 return BLK_STS_OK;
1065out_free_sgtables:
1066 scsi_free_sgtables(cmd);
1067 return ret;
1068}
1069EXPORT_SYMBOL(scsi_init_io);
1070
1071/**
1072 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1073 * @rq: Request associated with the SCSI command to be initialized.
1074 *
1075 * This function initializes the members of struct scsi_cmnd that must be
1076 * initialized before request processing starts and that won't be
1077 * reinitialized if a SCSI command is requeued.
1078 *
1079 * Called from inside blk_get_request() for pass-through requests and from
1080 * inside scsi_init_command() for filesystem requests.
1081 */
1082static void scsi_initialize_rq(struct request *rq)
1083{
1084 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1085
1086 scsi_req_init(&cmd->req);
1087 init_rcu_head(&cmd->rcu);
1088 cmd->jiffies_at_alloc = jiffies;
1089 cmd->retries = 0;
1090}
1091
1092/*
1093 * Only called when the request isn't completed by SCSI, and not freed by
1094 * SCSI
1095 */
1096static void scsi_cleanup_rq(struct request *rq)
1097{
1098 if (rq->rq_flags & RQF_DONTPREP) {
1099 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1100 rq->rq_flags &= ~RQF_DONTPREP;
1101 }
1102}
1103
1104/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1105void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1106{
1107 void *buf = cmd->sense_buffer;
1108 void *prot = cmd->prot_sdb;
1109 struct request *rq = blk_mq_rq_from_pdu(cmd);
1110 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1111 unsigned long jiffies_at_alloc;
1112 int retries, to_clear;
1113 bool in_flight;
1114
1115 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1116 flags |= SCMD_INITIALIZED;
1117 scsi_initialize_rq(rq);
1118 }
1119
1120 jiffies_at_alloc = cmd->jiffies_at_alloc;
1121 retries = cmd->retries;
1122 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1123 /*
1124 * Zero out the cmd, except for the embedded scsi_request. Only clear
1125 * the driver-private command data if the LLD does not supply a
1126 * function to initialize that data.
1127 */
1128 to_clear = sizeof(*cmd) - sizeof(cmd->req);
1129 if (!dev->host->hostt->init_cmd_priv)
1130 to_clear += dev->host->hostt->cmd_size;
1131 memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1132
1133 cmd->device = dev;
1134 cmd->sense_buffer = buf;
1135 cmd->prot_sdb = prot;
1136 cmd->flags = flags;
1137 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1138 cmd->jiffies_at_alloc = jiffies_at_alloc;
1139 cmd->retries = retries;
1140 if (in_flight)
1141 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1142
1143}
1144
1145static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1146 struct request *req)
1147{
1148 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1149
1150 /*
1151 * Passthrough requests may transfer data, in which case they must
1152 * a bio attached to them. Or they might contain a SCSI command
1153 * that does not transfer data, in which case they may optionally
1154 * submit a request without an attached bio.
1155 */
1156 if (req->bio) {
1157 blk_status_t ret = scsi_init_io(cmd);
1158 if (unlikely(ret != BLK_STS_OK))
1159 return ret;
1160 } else {
1161 BUG_ON(blk_rq_bytes(req));
1162
1163 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1164 }
1165
1166 cmd->cmd_len = scsi_req(req)->cmd_len;
1167 cmd->cmnd = scsi_req(req)->cmd;
1168 cmd->transfersize = blk_rq_bytes(req);
1169 cmd->allowed = scsi_req(req)->retries;
1170 return BLK_STS_OK;
1171}
1172
1173/*
1174 * Setup a normal block command. These are simple request from filesystems
1175 * that still need to be translated to SCSI CDBs from the ULD.
1176 */
1177static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1178 struct request *req)
1179{
1180 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1181
1182 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1183 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1184 if (ret != BLK_STS_OK)
1185 return ret;
1186 }
1187
1188 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1189 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1190 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1191}
1192
1193static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1194 struct request *req)
1195{
1196 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1197 blk_status_t ret;
1198
1199 if (!blk_rq_bytes(req))
1200 cmd->sc_data_direction = DMA_NONE;
1201 else if (rq_data_dir(req) == WRITE)
1202 cmd->sc_data_direction = DMA_TO_DEVICE;
1203 else
1204 cmd->sc_data_direction = DMA_FROM_DEVICE;
1205
1206 if (blk_rq_is_scsi(req))
1207 ret = scsi_setup_scsi_cmnd(sdev, req);
1208 else
1209 ret = scsi_setup_fs_cmnd(sdev, req);
1210
1211 if (ret != BLK_STS_OK)
1212 scsi_free_sgtables(cmd);
1213
1214 return ret;
1215}
1216
1217static blk_status_t
1218scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1219{
1220 switch (sdev->sdev_state) {
1221 case SDEV_OFFLINE:
1222 case SDEV_TRANSPORT_OFFLINE:
1223 /*
1224 * If the device is offline we refuse to process any
1225 * commands. The device must be brought online
1226 * before trying any recovery commands.
1227 */
1228 if (!sdev->offline_already) {
1229 sdev->offline_already = true;
1230 sdev_printk(KERN_ERR, sdev,
1231 "rejecting I/O to offline device\n");
1232 }
1233 return BLK_STS_IOERR;
1234 case SDEV_DEL:
1235 /*
1236 * If the device is fully deleted, we refuse to
1237 * process any commands as well.
1238 */
1239 sdev_printk(KERN_ERR, sdev,
1240 "rejecting I/O to dead device\n");
1241 return BLK_STS_IOERR;
1242 case SDEV_BLOCK:
1243 case SDEV_CREATED_BLOCK:
1244 return BLK_STS_RESOURCE;
1245 case SDEV_QUIESCE:
1246 /*
1247 * If the devices is blocked we defer normal commands.
1248 */
1249 if (req && !(req->rq_flags & RQF_PREEMPT))
1250 return BLK_STS_RESOURCE;
1251 return BLK_STS_OK;
1252 default:
1253 /*
1254 * For any other not fully online state we only allow
1255 * special commands. In particular any user initiated
1256 * command is not allowed.
1257 */
1258 if (req && !(req->rq_flags & RQF_PREEMPT))
1259 return BLK_STS_IOERR;
1260 return BLK_STS_OK;
1261 }
1262}
1263
1264/*
1265 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1266 * return 0.
1267 *
1268 * Called with the queue_lock held.
1269 */
1270static inline int scsi_dev_queue_ready(struct request_queue *q,
1271 struct scsi_device *sdev)
1272{
1273 unsigned int busy;
1274
1275 busy = atomic_inc_return(&sdev->device_busy) - 1;
1276 if (atomic_read(&sdev->device_blocked)) {
1277 if (busy)
1278 goto out_dec;
1279
1280 /*
1281 * unblock after device_blocked iterates to zero
1282 */
1283 if (atomic_dec_return(&sdev->device_blocked) > 0)
1284 goto out_dec;
1285 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1286 "unblocking device at zero depth\n"));
1287 }
1288
1289 if (busy >= sdev->queue_depth)
1290 goto out_dec;
1291
1292 return 1;
1293out_dec:
1294 atomic_dec(&sdev->device_busy);
1295 return 0;
1296}
1297
1298/*
1299 * scsi_target_queue_ready: checks if there we can send commands to target
1300 * @sdev: scsi device on starget to check.
1301 */
1302static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1303 struct scsi_device *sdev)
1304{
1305 struct scsi_target *starget = scsi_target(sdev);
1306 unsigned int busy;
1307
1308 if (starget->single_lun) {
1309 spin_lock_irq(shost->host_lock);
1310 if (starget->starget_sdev_user &&
1311 starget->starget_sdev_user != sdev) {
1312 spin_unlock_irq(shost->host_lock);
1313 return 0;
1314 }
1315 starget->starget_sdev_user = sdev;
1316 spin_unlock_irq(shost->host_lock);
1317 }
1318
1319 if (starget->can_queue <= 0)
1320 return 1;
1321
1322 busy = atomic_inc_return(&starget->target_busy) - 1;
1323 if (atomic_read(&starget->target_blocked) > 0) {
1324 if (busy)
1325 goto starved;
1326
1327 /*
1328 * unblock after target_blocked iterates to zero
1329 */
1330 if (atomic_dec_return(&starget->target_blocked) > 0)
1331 goto out_dec;
1332
1333 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1334 "unblocking target at zero depth\n"));
1335 }
1336
1337 if (busy >= starget->can_queue)
1338 goto starved;
1339
1340 return 1;
1341
1342starved:
1343 spin_lock_irq(shost->host_lock);
1344 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1345 spin_unlock_irq(shost->host_lock);
1346out_dec:
1347 if (starget->can_queue > 0)
1348 atomic_dec(&starget->target_busy);
1349 return 0;
1350}
1351
1352/*
1353 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1354 * return 0. We must end up running the queue again whenever 0 is
1355 * returned, else IO can hang.
1356 */
1357static inline int scsi_host_queue_ready(struct request_queue *q,
1358 struct Scsi_Host *shost,
1359 struct scsi_device *sdev,
1360 struct scsi_cmnd *cmd)
1361{
1362 if (scsi_host_in_recovery(shost))
1363 return 0;
1364
1365 if (atomic_read(&shost->host_blocked) > 0) {
1366 if (scsi_host_busy(shost) > 0)
1367 goto starved;
1368
1369 /*
1370 * unblock after host_blocked iterates to zero
1371 */
1372 if (atomic_dec_return(&shost->host_blocked) > 0)
1373 goto out_dec;
1374
1375 SCSI_LOG_MLQUEUE(3,
1376 shost_printk(KERN_INFO, shost,
1377 "unblocking host at zero depth\n"));
1378 }
1379
1380 if (shost->host_self_blocked)
1381 goto starved;
1382
1383 /* We're OK to process the command, so we can't be starved */
1384 if (!list_empty(&sdev->starved_entry)) {
1385 spin_lock_irq(shost->host_lock);
1386 if (!list_empty(&sdev->starved_entry))
1387 list_del_init(&sdev->starved_entry);
1388 spin_unlock_irq(shost->host_lock);
1389 }
1390
1391 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1392
1393 return 1;
1394
1395starved:
1396 spin_lock_irq(shost->host_lock);
1397 if (list_empty(&sdev->starved_entry))
1398 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1399 spin_unlock_irq(shost->host_lock);
1400out_dec:
1401 scsi_dec_host_busy(shost, cmd);
1402 return 0;
1403}
1404
1405/*
1406 * Busy state exporting function for request stacking drivers.
1407 *
1408 * For efficiency, no lock is taken to check the busy state of
1409 * shost/starget/sdev, since the returned value is not guaranteed and
1410 * may be changed after request stacking drivers call the function,
1411 * regardless of taking lock or not.
1412 *
1413 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1414 * needs to return 'not busy'. Otherwise, request stacking drivers
1415 * may hold requests forever.
1416 */
1417static bool scsi_mq_lld_busy(struct request_queue *q)
1418{
1419 struct scsi_device *sdev = q->queuedata;
1420 struct Scsi_Host *shost;
1421
1422 if (blk_queue_dying(q))
1423 return false;
1424
1425 shost = sdev->host;
1426
1427 /*
1428 * Ignore host/starget busy state.
1429 * Since block layer does not have a concept of fairness across
1430 * multiple queues, congestion of host/starget needs to be handled
1431 * in SCSI layer.
1432 */
1433 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1434 return true;
1435
1436 return false;
1437}
1438
1439static void scsi_softirq_done(struct request *rq)
1440{
1441 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1442 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1443 int disposition;
1444
1445 INIT_LIST_HEAD(&cmd->eh_entry);
1446
1447 atomic_inc(&cmd->device->iodone_cnt);
1448 if (cmd->result)
1449 atomic_inc(&cmd->device->ioerr_cnt);
1450
1451 disposition = scsi_decide_disposition(cmd);
1452 if (disposition != SUCCESS &&
1453 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1454 scmd_printk(KERN_ERR, cmd,
1455 "timing out command, waited %lus\n",
1456 wait_for/HZ);
1457 disposition = SUCCESS;
1458 }
1459
1460 scsi_log_completion(cmd, disposition);
1461
1462 switch (disposition) {
1463 case SUCCESS:
1464 scsi_finish_command(cmd);
1465 break;
1466 case NEEDS_RETRY:
1467 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1468 break;
1469 case ADD_TO_MLQUEUE:
1470 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1471 break;
1472 default:
1473 scsi_eh_scmd_add(cmd);
1474 break;
1475 }
1476}
1477
1478/**
1479 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1480 * @cmd: command block we are dispatching.
1481 *
1482 * Return: nonzero return request was rejected and device's queue needs to be
1483 * plugged.
1484 */
1485static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1486{
1487 struct Scsi_Host *host = cmd->device->host;
1488 int rtn = 0;
1489
1490 atomic_inc(&cmd->device->iorequest_cnt);
1491
1492 /* check if the device is still usable */
1493 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1494 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1495 * returns an immediate error upwards, and signals
1496 * that the device is no longer present */
1497 cmd->result = DID_NO_CONNECT << 16;
1498 goto done;
1499 }
1500
1501 /* Check to see if the scsi lld made this device blocked. */
1502 if (unlikely(scsi_device_blocked(cmd->device))) {
1503 /*
1504 * in blocked state, the command is just put back on
1505 * the device queue. The suspend state has already
1506 * blocked the queue so future requests should not
1507 * occur until the device transitions out of the
1508 * suspend state.
1509 */
1510 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1511 "queuecommand : device blocked\n"));
1512 return SCSI_MLQUEUE_DEVICE_BUSY;
1513 }
1514
1515 /* Store the LUN value in cmnd, if needed. */
1516 if (cmd->device->lun_in_cdb)
1517 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1518 (cmd->device->lun << 5 & 0xe0);
1519
1520 scsi_log_send(cmd);
1521
1522 /*
1523 * Before we queue this command, check if the command
1524 * length exceeds what the host adapter can handle.
1525 */
1526 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1527 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1528 "queuecommand : command too long. "
1529 "cdb_size=%d host->max_cmd_len=%d\n",
1530 cmd->cmd_len, cmd->device->host->max_cmd_len));
1531 cmd->result = (DID_ABORT << 16);
1532 goto done;
1533 }
1534
1535 if (unlikely(host->shost_state == SHOST_DEL)) {
1536 cmd->result = (DID_NO_CONNECT << 16);
1537 goto done;
1538
1539 }
1540
1541 trace_scsi_dispatch_cmd_start(cmd);
1542 rtn = host->hostt->queuecommand(host, cmd);
1543 if (rtn) {
1544 trace_scsi_dispatch_cmd_error(cmd, rtn);
1545 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1546 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1547 rtn = SCSI_MLQUEUE_HOST_BUSY;
1548
1549 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1550 "queuecommand : request rejected\n"));
1551 }
1552
1553 return rtn;
1554 done:
1555 cmd->scsi_done(cmd);
1556 return 0;
1557}
1558
1559/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1560static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1561{
1562 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1563 sizeof(struct scatterlist);
1564}
1565
1566static blk_status_t scsi_mq_prep_fn(struct request *req)
1567{
1568 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1569 struct scsi_device *sdev = req->q->queuedata;
1570 struct Scsi_Host *shost = sdev->host;
1571 struct scatterlist *sg;
1572
1573 scsi_init_command(sdev, cmd);
1574
1575 cmd->request = req;
1576 cmd->tag = req->tag;
1577 cmd->prot_op = SCSI_PROT_NORMAL;
1578
1579 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580 cmd->sdb.table.sgl = sg;
1581
1582 if (scsi_host_get_prot(shost)) {
1583 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1584
1585 cmd->prot_sdb->table.sgl =
1586 (struct scatterlist *)(cmd->prot_sdb + 1);
1587 }
1588
1589 blk_mq_start_request(req);
1590
1591 return scsi_setup_cmnd(sdev, req);
1592}
1593
1594static void scsi_mq_done(struct scsi_cmnd *cmd)
1595{
1596 if (unlikely(blk_should_fake_timeout(cmd->request->q)))
1597 return;
1598 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1599 return;
1600 trace_scsi_dispatch_cmd_done(cmd);
1601 blk_mq_complete_request(cmd->request);
1602}
1603
1604static void scsi_mq_put_budget(struct request_queue *q)
1605{
1606 struct scsi_device *sdev = q->queuedata;
1607
1608 atomic_dec(&sdev->device_busy);
1609}
1610
1611static bool scsi_mq_get_budget(struct request_queue *q)
1612{
1613 struct scsi_device *sdev = q->queuedata;
1614
1615 return scsi_dev_queue_ready(q, sdev);
1616}
1617
1618static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1619 const struct blk_mq_queue_data *bd)
1620{
1621 struct request *req = bd->rq;
1622 struct request_queue *q = req->q;
1623 struct scsi_device *sdev = q->queuedata;
1624 struct Scsi_Host *shost = sdev->host;
1625 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1626 blk_status_t ret;
1627 int reason;
1628
1629 /*
1630 * If the device is not in running state we will reject some or all
1631 * commands.
1632 */
1633 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1634 ret = scsi_prep_state_check(sdev, req);
1635 if (ret != BLK_STS_OK)
1636 goto out_put_budget;
1637 }
1638
1639 ret = BLK_STS_RESOURCE;
1640 if (!scsi_target_queue_ready(shost, sdev))
1641 goto out_put_budget;
1642 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1643 goto out_dec_target_busy;
1644
1645 if (!(req->rq_flags & RQF_DONTPREP)) {
1646 ret = scsi_mq_prep_fn(req);
1647 if (ret != BLK_STS_OK)
1648 goto out_dec_host_busy;
1649 req->rq_flags |= RQF_DONTPREP;
1650 } else {
1651 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1652 blk_mq_start_request(req);
1653 }
1654
1655 cmd->flags &= SCMD_PRESERVED_FLAGS;
1656 if (sdev->simple_tags)
1657 cmd->flags |= SCMD_TAGGED;
1658 if (bd->last)
1659 cmd->flags |= SCMD_LAST;
1660
1661 scsi_init_cmd_errh(cmd);
1662 cmd->scsi_done = scsi_mq_done;
1663
1664 reason = scsi_dispatch_cmd(cmd);
1665 if (reason) {
1666 scsi_set_blocked(cmd, reason);
1667 ret = BLK_STS_RESOURCE;
1668 goto out_dec_host_busy;
1669 }
1670
1671 return BLK_STS_OK;
1672
1673out_dec_host_busy:
1674 scsi_dec_host_busy(shost, cmd);
1675out_dec_target_busy:
1676 if (scsi_target(sdev)->can_queue > 0)
1677 atomic_dec(&scsi_target(sdev)->target_busy);
1678out_put_budget:
1679 scsi_mq_put_budget(q);
1680 switch (ret) {
1681 case BLK_STS_OK:
1682 break;
1683 case BLK_STS_RESOURCE:
1684 case BLK_STS_ZONE_RESOURCE:
1685 if (atomic_read(&sdev->device_busy) ||
1686 scsi_device_blocked(sdev))
1687 ret = BLK_STS_DEV_RESOURCE;
1688 break;
1689 default:
1690 if (unlikely(!scsi_device_online(sdev)))
1691 scsi_req(req)->result = DID_NO_CONNECT << 16;
1692 else
1693 scsi_req(req)->result = DID_ERROR << 16;
1694 /*
1695 * Make sure to release all allocated resources when
1696 * we hit an error, as we will never see this command
1697 * again.
1698 */
1699 if (req->rq_flags & RQF_DONTPREP)
1700 scsi_mq_uninit_cmd(cmd);
1701 scsi_run_queue_async(sdev);
1702 break;
1703 }
1704 return ret;
1705}
1706
1707static enum blk_eh_timer_return scsi_timeout(struct request *req,
1708 bool reserved)
1709{
1710 if (reserved)
1711 return BLK_EH_RESET_TIMER;
1712 return scsi_times_out(req);
1713}
1714
1715static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1716 unsigned int hctx_idx, unsigned int numa_node)
1717{
1718 struct Scsi_Host *shost = set->driver_data;
1719 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1720 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1721 struct scatterlist *sg;
1722 int ret = 0;
1723
1724 if (unchecked_isa_dma)
1725 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1726 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1727 GFP_KERNEL, numa_node);
1728 if (!cmd->sense_buffer)
1729 return -ENOMEM;
1730 cmd->req.sense = cmd->sense_buffer;
1731
1732 if (scsi_host_get_prot(shost)) {
1733 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1734 shost->hostt->cmd_size;
1735 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1736 }
1737
1738 if (shost->hostt->init_cmd_priv) {
1739 ret = shost->hostt->init_cmd_priv(shost, cmd);
1740 if (ret < 0)
1741 scsi_free_sense_buffer(unchecked_isa_dma,
1742 cmd->sense_buffer);
1743 }
1744
1745 return ret;
1746}
1747
1748static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1749 unsigned int hctx_idx)
1750{
1751 struct Scsi_Host *shost = set->driver_data;
1752 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1753
1754 if (shost->hostt->exit_cmd_priv)
1755 shost->hostt->exit_cmd_priv(shost, cmd);
1756 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1757 cmd->sense_buffer);
1758}
1759
1760static int scsi_map_queues(struct blk_mq_tag_set *set)
1761{
1762 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1763
1764 if (shost->hostt->map_queues)
1765 return shost->hostt->map_queues(shost);
1766 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1767}
1768
1769void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1770{
1771 struct device *dev = shost->dma_dev;
1772
1773 /*
1774 * this limit is imposed by hardware restrictions
1775 */
1776 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1777 SG_MAX_SEGMENTS));
1778
1779 if (scsi_host_prot_dma(shost)) {
1780 shost->sg_prot_tablesize =
1781 min_not_zero(shost->sg_prot_tablesize,
1782 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1783 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1784 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1785 }
1786
1787 if (dev->dma_mask) {
1788 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1789 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1790 }
1791 blk_queue_max_hw_sectors(q, shost->max_sectors);
1792 if (shost->unchecked_isa_dma)
1793 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1794 blk_queue_segment_boundary(q, shost->dma_boundary);
1795 dma_set_seg_boundary(dev, shost->dma_boundary);
1796
1797 blk_queue_max_segment_size(q, shost->max_segment_size);
1798 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1799 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1800
1801 /*
1802 * Set a reasonable default alignment: The larger of 32-byte (dword),
1803 * which is a common minimum for HBAs, and the minimum DMA alignment,
1804 * which is set by the platform.
1805 *
1806 * Devices that require a bigger alignment can increase it later.
1807 */
1808 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1809}
1810EXPORT_SYMBOL_GPL(__scsi_init_queue);
1811
1812static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1813 .get_budget = scsi_mq_get_budget,
1814 .put_budget = scsi_mq_put_budget,
1815 .queue_rq = scsi_queue_rq,
1816 .complete = scsi_softirq_done,
1817 .timeout = scsi_timeout,
1818#ifdef CONFIG_BLK_DEBUG_FS
1819 .show_rq = scsi_show_rq,
1820#endif
1821 .init_request = scsi_mq_init_request,
1822 .exit_request = scsi_mq_exit_request,
1823 .initialize_rq_fn = scsi_initialize_rq,
1824 .cleanup_rq = scsi_cleanup_rq,
1825 .busy = scsi_mq_lld_busy,
1826 .map_queues = scsi_map_queues,
1827};
1828
1829
1830static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1831{
1832 struct request_queue *q = hctx->queue;
1833 struct scsi_device *sdev = q->queuedata;
1834 struct Scsi_Host *shost = sdev->host;
1835
1836 shost->hostt->commit_rqs(shost, hctx->queue_num);
1837}
1838
1839static const struct blk_mq_ops scsi_mq_ops = {
1840 .get_budget = scsi_mq_get_budget,
1841 .put_budget = scsi_mq_put_budget,
1842 .queue_rq = scsi_queue_rq,
1843 .commit_rqs = scsi_commit_rqs,
1844 .complete = scsi_softirq_done,
1845 .timeout = scsi_timeout,
1846#ifdef CONFIG_BLK_DEBUG_FS
1847 .show_rq = scsi_show_rq,
1848#endif
1849 .init_request = scsi_mq_init_request,
1850 .exit_request = scsi_mq_exit_request,
1851 .initialize_rq_fn = scsi_initialize_rq,
1852 .cleanup_rq = scsi_cleanup_rq,
1853 .busy = scsi_mq_lld_busy,
1854 .map_queues = scsi_map_queues,
1855};
1856
1857struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1858{
1859 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1860 if (IS_ERR(sdev->request_queue))
1861 return NULL;
1862
1863 sdev->request_queue->queuedata = sdev;
1864 __scsi_init_queue(sdev->host, sdev->request_queue);
1865 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1866 return sdev->request_queue;
1867}
1868
1869int scsi_mq_setup_tags(struct Scsi_Host *shost)
1870{
1871 unsigned int cmd_size, sgl_size;
1872 struct blk_mq_tag_set *tag_set = &shost->tag_set;
1873
1874 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1875 scsi_mq_inline_sgl_size(shost));
1876 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1877 if (scsi_host_get_prot(shost))
1878 cmd_size += sizeof(struct scsi_data_buffer) +
1879 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1880
1881 memset(tag_set, 0, sizeof(*tag_set));
1882 if (shost->hostt->commit_rqs)
1883 tag_set->ops = &scsi_mq_ops;
1884 else
1885 tag_set->ops = &scsi_mq_ops_no_commit;
1886 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1887 tag_set->queue_depth = shost->can_queue;
1888 tag_set->cmd_size = cmd_size;
1889 tag_set->numa_node = NUMA_NO_NODE;
1890 tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1891 tag_set->flags |=
1892 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1893 tag_set->driver_data = shost;
1894
1895 return blk_mq_alloc_tag_set(tag_set);
1896}
1897
1898void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1899{
1900 blk_mq_free_tag_set(&shost->tag_set);
1901}
1902
1903/**
1904 * scsi_device_from_queue - return sdev associated with a request_queue
1905 * @q: The request queue to return the sdev from
1906 *
1907 * Return the sdev associated with a request queue or NULL if the
1908 * request_queue does not reference a SCSI device.
1909 */
1910struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1911{
1912 struct scsi_device *sdev = NULL;
1913
1914 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1915 q->mq_ops == &scsi_mq_ops)
1916 sdev = q->queuedata;
1917 if (!sdev || !get_device(&sdev->sdev_gendev))
1918 sdev = NULL;
1919
1920 return sdev;
1921}
1922EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1923
1924/**
1925 * scsi_block_requests - Utility function used by low-level drivers to prevent
1926 * further commands from being queued to the device.
1927 * @shost: host in question
1928 *
1929 * There is no timer nor any other means by which the requests get unblocked
1930 * other than the low-level driver calling scsi_unblock_requests().
1931 */
1932void scsi_block_requests(struct Scsi_Host *shost)
1933{
1934 shost->host_self_blocked = 1;
1935}
1936EXPORT_SYMBOL(scsi_block_requests);
1937
1938/**
1939 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1940 * further commands to be queued to the device.
1941 * @shost: host in question
1942 *
1943 * There is no timer nor any other means by which the requests get unblocked
1944 * other than the low-level driver calling scsi_unblock_requests(). This is done
1945 * as an API function so that changes to the internals of the scsi mid-layer
1946 * won't require wholesale changes to drivers that use this feature.
1947 */
1948void scsi_unblock_requests(struct Scsi_Host *shost)
1949{
1950 shost->host_self_blocked = 0;
1951 scsi_run_host_queues(shost);
1952}
1953EXPORT_SYMBOL(scsi_unblock_requests);
1954
1955void scsi_exit_queue(void)
1956{
1957 kmem_cache_destroy(scsi_sense_cache);
1958 kmem_cache_destroy(scsi_sense_isadma_cache);
1959}
1960
1961/**
1962 * scsi_mode_select - issue a mode select
1963 * @sdev: SCSI device to be queried
1964 * @pf: Page format bit (1 == standard, 0 == vendor specific)
1965 * @sp: Save page bit (0 == don't save, 1 == save)
1966 * @modepage: mode page being requested
1967 * @buffer: request buffer (may not be smaller than eight bytes)
1968 * @len: length of request buffer.
1969 * @timeout: command timeout
1970 * @retries: number of retries before failing
1971 * @data: returns a structure abstracting the mode header data
1972 * @sshdr: place to put sense data (or NULL if no sense to be collected).
1973 * must be SCSI_SENSE_BUFFERSIZE big.
1974 *
1975 * Returns zero if successful; negative error number or scsi
1976 * status on error
1977 *
1978 */
1979int
1980scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1981 unsigned char *buffer, int len, int timeout, int retries,
1982 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1983{
1984 unsigned char cmd[10];
1985 unsigned char *real_buffer;
1986 int ret;
1987
1988 memset(cmd, 0, sizeof(cmd));
1989 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1990
1991 if (sdev->use_10_for_ms) {
1992 if (len > 65535)
1993 return -EINVAL;
1994 real_buffer = kmalloc(8 + len, GFP_KERNEL);
1995 if (!real_buffer)
1996 return -ENOMEM;
1997 memcpy(real_buffer + 8, buffer, len);
1998 len += 8;
1999 real_buffer[0] = 0;
2000 real_buffer[1] = 0;
2001 real_buffer[2] = data->medium_type;
2002 real_buffer[3] = data->device_specific;
2003 real_buffer[4] = data->longlba ? 0x01 : 0;
2004 real_buffer[5] = 0;
2005 real_buffer[6] = data->block_descriptor_length >> 8;
2006 real_buffer[7] = data->block_descriptor_length;
2007
2008 cmd[0] = MODE_SELECT_10;
2009 cmd[7] = len >> 8;
2010 cmd[8] = len;
2011 } else {
2012 if (len > 255 || data->block_descriptor_length > 255 ||
2013 data->longlba)
2014 return -EINVAL;
2015
2016 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2017 if (!real_buffer)
2018 return -ENOMEM;
2019 memcpy(real_buffer + 4, buffer, len);
2020 len += 4;
2021 real_buffer[0] = 0;
2022 real_buffer[1] = data->medium_type;
2023 real_buffer[2] = data->device_specific;
2024 real_buffer[3] = data->block_descriptor_length;
2025
2026 cmd[0] = MODE_SELECT;
2027 cmd[4] = len;
2028 }
2029
2030 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2031 sshdr, timeout, retries, NULL);
2032 kfree(real_buffer);
2033 return ret;
2034}
2035EXPORT_SYMBOL_GPL(scsi_mode_select);
2036
2037/**
2038 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2039 * @sdev: SCSI device to be queried
2040 * @dbd: set if mode sense will allow block descriptors to be returned
2041 * @modepage: mode page being requested
2042 * @buffer: request buffer (may not be smaller than eight bytes)
2043 * @len: length of request buffer.
2044 * @timeout: command timeout
2045 * @retries: number of retries before failing
2046 * @data: returns a structure abstracting the mode header data
2047 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2048 * must be SCSI_SENSE_BUFFERSIZE big.
2049 *
2050 * Returns zero if unsuccessful, or the header offset (either 4
2051 * or 8 depending on whether a six or ten byte command was
2052 * issued) if successful.
2053 */
2054int
2055scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2056 unsigned char *buffer, int len, int timeout, int retries,
2057 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2058{
2059 unsigned char cmd[12];
2060 int use_10_for_ms;
2061 int header_length;
2062 int result, retry_count = retries;
2063 struct scsi_sense_hdr my_sshdr;
2064
2065 memset(data, 0, sizeof(*data));
2066 memset(&cmd[0], 0, 12);
2067
2068 dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2069 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2070 cmd[2] = modepage;
2071
2072 /* caller might not be interested in sense, but we need it */
2073 if (!sshdr)
2074 sshdr = &my_sshdr;
2075
2076 retry:
2077 use_10_for_ms = sdev->use_10_for_ms;
2078
2079 if (use_10_for_ms) {
2080 if (len < 8)
2081 len = 8;
2082
2083 cmd[0] = MODE_SENSE_10;
2084 cmd[8] = len;
2085 header_length = 8;
2086 } else {
2087 if (len < 4)
2088 len = 4;
2089
2090 cmd[0] = MODE_SENSE;
2091 cmd[4] = len;
2092 header_length = 4;
2093 }
2094
2095 memset(buffer, 0, len);
2096
2097 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2098 sshdr, timeout, retries, NULL);
2099
2100 /* This code looks awful: what it's doing is making sure an
2101 * ILLEGAL REQUEST sense return identifies the actual command
2102 * byte as the problem. MODE_SENSE commands can return
2103 * ILLEGAL REQUEST if the code page isn't supported */
2104
2105 if (use_10_for_ms && !scsi_status_is_good(result) &&
2106 driver_byte(result) == DRIVER_SENSE) {
2107 if (scsi_sense_valid(sshdr)) {
2108 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2109 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2110 /*
2111 * Invalid command operation code
2112 */
2113 sdev->use_10_for_ms = 0;
2114 goto retry;
2115 }
2116 }
2117 }
2118
2119 if (scsi_status_is_good(result)) {
2120 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2121 (modepage == 6 || modepage == 8))) {
2122 /* Initio breakage? */
2123 header_length = 0;
2124 data->length = 13;
2125 data->medium_type = 0;
2126 data->device_specific = 0;
2127 data->longlba = 0;
2128 data->block_descriptor_length = 0;
2129 } else if (use_10_for_ms) {
2130 data->length = buffer[0]*256 + buffer[1] + 2;
2131 data->medium_type = buffer[2];
2132 data->device_specific = buffer[3];
2133 data->longlba = buffer[4] & 0x01;
2134 data->block_descriptor_length = buffer[6]*256
2135 + buffer[7];
2136 } else {
2137 data->length = buffer[0] + 1;
2138 data->medium_type = buffer[1];
2139 data->device_specific = buffer[2];
2140 data->block_descriptor_length = buffer[3];
2141 }
2142 data->header_length = header_length;
2143 } else if ((status_byte(result) == CHECK_CONDITION) &&
2144 scsi_sense_valid(sshdr) &&
2145 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2146 retry_count--;
2147 goto retry;
2148 }
2149
2150 return result;
2151}
2152EXPORT_SYMBOL(scsi_mode_sense);
2153
2154/**
2155 * scsi_test_unit_ready - test if unit is ready
2156 * @sdev: scsi device to change the state of.
2157 * @timeout: command timeout
2158 * @retries: number of retries before failing
2159 * @sshdr: outpout pointer for decoded sense information.
2160 *
2161 * Returns zero if unsuccessful or an error if TUR failed. For
2162 * removable media, UNIT_ATTENTION sets ->changed flag.
2163 **/
2164int
2165scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2166 struct scsi_sense_hdr *sshdr)
2167{
2168 char cmd[] = {
2169 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2170 };
2171 int result;
2172
2173 /* try to eat the UNIT_ATTENTION if there are enough retries */
2174 do {
2175 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2176 timeout, 1, NULL);
2177 if (sdev->removable && scsi_sense_valid(sshdr) &&
2178 sshdr->sense_key == UNIT_ATTENTION)
2179 sdev->changed = 1;
2180 } while (scsi_sense_valid(sshdr) &&
2181 sshdr->sense_key == UNIT_ATTENTION && --retries);
2182
2183 return result;
2184}
2185EXPORT_SYMBOL(scsi_test_unit_ready);
2186
2187/**
2188 * scsi_device_set_state - Take the given device through the device state model.
2189 * @sdev: scsi device to change the state of.
2190 * @state: state to change to.
2191 *
2192 * Returns zero if successful or an error if the requested
2193 * transition is illegal.
2194 */
2195int
2196scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2197{
2198 enum scsi_device_state oldstate = sdev->sdev_state;
2199
2200 if (state == oldstate)
2201 return 0;
2202
2203 switch (state) {
2204 case SDEV_CREATED:
2205 switch (oldstate) {
2206 case SDEV_CREATED_BLOCK:
2207 break;
2208 default:
2209 goto illegal;
2210 }
2211 break;
2212
2213 case SDEV_RUNNING:
2214 switch (oldstate) {
2215 case SDEV_CREATED:
2216 case SDEV_OFFLINE:
2217 case SDEV_TRANSPORT_OFFLINE:
2218 case SDEV_QUIESCE:
2219 case SDEV_BLOCK:
2220 break;
2221 default:
2222 goto illegal;
2223 }
2224 break;
2225
2226 case SDEV_QUIESCE:
2227 switch (oldstate) {
2228 case SDEV_RUNNING:
2229 case SDEV_OFFLINE:
2230 case SDEV_TRANSPORT_OFFLINE:
2231 break;
2232 default:
2233 goto illegal;
2234 }
2235 break;
2236
2237 case SDEV_OFFLINE:
2238 case SDEV_TRANSPORT_OFFLINE:
2239 switch (oldstate) {
2240 case SDEV_CREATED:
2241 case SDEV_RUNNING:
2242 case SDEV_QUIESCE:
2243 case SDEV_BLOCK:
2244 break;
2245 default:
2246 goto illegal;
2247 }
2248 break;
2249
2250 case SDEV_BLOCK:
2251 switch (oldstate) {
2252 case SDEV_RUNNING:
2253 case SDEV_CREATED_BLOCK:
2254 case SDEV_QUIESCE:
2255 case SDEV_OFFLINE:
2256 break;
2257 default:
2258 goto illegal;
2259 }
2260 break;
2261
2262 case SDEV_CREATED_BLOCK:
2263 switch (oldstate) {
2264 case SDEV_CREATED:
2265 break;
2266 default:
2267 goto illegal;
2268 }
2269 break;
2270
2271 case SDEV_CANCEL:
2272 switch (oldstate) {
2273 case SDEV_CREATED:
2274 case SDEV_RUNNING:
2275 case SDEV_QUIESCE:
2276 case SDEV_OFFLINE:
2277 case SDEV_TRANSPORT_OFFLINE:
2278 break;
2279 default:
2280 goto illegal;
2281 }
2282 break;
2283
2284 case SDEV_DEL:
2285 switch (oldstate) {
2286 case SDEV_CREATED:
2287 case SDEV_RUNNING:
2288 case SDEV_OFFLINE:
2289 case SDEV_TRANSPORT_OFFLINE:
2290 case SDEV_CANCEL:
2291 case SDEV_BLOCK:
2292 case SDEV_CREATED_BLOCK:
2293 break;
2294 default:
2295 goto illegal;
2296 }
2297 break;
2298
2299 }
2300 sdev->offline_already = false;
2301 sdev->sdev_state = state;
2302 return 0;
2303
2304 illegal:
2305 SCSI_LOG_ERROR_RECOVERY(1,
2306 sdev_printk(KERN_ERR, sdev,
2307 "Illegal state transition %s->%s",
2308 scsi_device_state_name(oldstate),
2309 scsi_device_state_name(state))
2310 );
2311 return -EINVAL;
2312}
2313EXPORT_SYMBOL(scsi_device_set_state);
2314
2315/**
2316 * sdev_evt_emit - emit a single SCSI device uevent
2317 * @sdev: associated SCSI device
2318 * @evt: event to emit
2319 *
2320 * Send a single uevent (scsi_event) to the associated scsi_device.
2321 */
2322static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2323{
2324 int idx = 0;
2325 char *envp[3];
2326
2327 switch (evt->evt_type) {
2328 case SDEV_EVT_MEDIA_CHANGE:
2329 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2330 break;
2331 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2332 scsi_rescan_device(&sdev->sdev_gendev);
2333 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2334 break;
2335 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2336 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2337 break;
2338 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2339 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2340 break;
2341 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2342 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2343 break;
2344 case SDEV_EVT_LUN_CHANGE_REPORTED:
2345 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2346 break;
2347 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2348 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2349 break;
2350 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2351 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2352 break;
2353 default:
2354 /* do nothing */
2355 break;
2356 }
2357
2358 envp[idx++] = NULL;
2359
2360 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2361}
2362
2363/**
2364 * sdev_evt_thread - send a uevent for each scsi event
2365 * @work: work struct for scsi_device
2366 *
2367 * Dispatch queued events to their associated scsi_device kobjects
2368 * as uevents.
2369 */
2370void scsi_evt_thread(struct work_struct *work)
2371{
2372 struct scsi_device *sdev;
2373 enum scsi_device_event evt_type;
2374 LIST_HEAD(event_list);
2375
2376 sdev = container_of(work, struct scsi_device, event_work);
2377
2378 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2379 if (test_and_clear_bit(evt_type, sdev->pending_events))
2380 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2381
2382 while (1) {
2383 struct scsi_event *evt;
2384 struct list_head *this, *tmp;
2385 unsigned long flags;
2386
2387 spin_lock_irqsave(&sdev->list_lock, flags);
2388 list_splice_init(&sdev->event_list, &event_list);
2389 spin_unlock_irqrestore(&sdev->list_lock, flags);
2390
2391 if (list_empty(&event_list))
2392 break;
2393
2394 list_for_each_safe(this, tmp, &event_list) {
2395 evt = list_entry(this, struct scsi_event, node);
2396 list_del(&evt->node);
2397 scsi_evt_emit(sdev, evt);
2398 kfree(evt);
2399 }
2400 }
2401}
2402
2403/**
2404 * sdev_evt_send - send asserted event to uevent thread
2405 * @sdev: scsi_device event occurred on
2406 * @evt: event to send
2407 *
2408 * Assert scsi device event asynchronously.
2409 */
2410void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2411{
2412 unsigned long flags;
2413
2414#if 0
2415 /* FIXME: currently this check eliminates all media change events
2416 * for polled devices. Need to update to discriminate between AN
2417 * and polled events */
2418 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2419 kfree(evt);
2420 return;
2421 }
2422#endif
2423
2424 spin_lock_irqsave(&sdev->list_lock, flags);
2425 list_add_tail(&evt->node, &sdev->event_list);
2426 schedule_work(&sdev->event_work);
2427 spin_unlock_irqrestore(&sdev->list_lock, flags);
2428}
2429EXPORT_SYMBOL_GPL(sdev_evt_send);
2430
2431/**
2432 * sdev_evt_alloc - allocate a new scsi event
2433 * @evt_type: type of event to allocate
2434 * @gfpflags: GFP flags for allocation
2435 *
2436 * Allocates and returns a new scsi_event.
2437 */
2438struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2439 gfp_t gfpflags)
2440{
2441 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2442 if (!evt)
2443 return NULL;
2444
2445 evt->evt_type = evt_type;
2446 INIT_LIST_HEAD(&evt->node);
2447
2448 /* evt_type-specific initialization, if any */
2449 switch (evt_type) {
2450 case SDEV_EVT_MEDIA_CHANGE:
2451 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2452 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2453 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2454 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2455 case SDEV_EVT_LUN_CHANGE_REPORTED:
2456 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2457 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2458 default:
2459 /* do nothing */
2460 break;
2461 }
2462
2463 return evt;
2464}
2465EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2466
2467/**
2468 * sdev_evt_send_simple - send asserted event to uevent thread
2469 * @sdev: scsi_device event occurred on
2470 * @evt_type: type of event to send
2471 * @gfpflags: GFP flags for allocation
2472 *
2473 * Assert scsi device event asynchronously, given an event type.
2474 */
2475void sdev_evt_send_simple(struct scsi_device *sdev,
2476 enum scsi_device_event evt_type, gfp_t gfpflags)
2477{
2478 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2479 if (!evt) {
2480 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2481 evt_type);
2482 return;
2483 }
2484
2485 sdev_evt_send(sdev, evt);
2486}
2487EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2488
2489/**
2490 * scsi_device_quiesce - Block user issued commands.
2491 * @sdev: scsi device to quiesce.
2492 *
2493 * This works by trying to transition to the SDEV_QUIESCE state
2494 * (which must be a legal transition). When the device is in this
2495 * state, only special requests will be accepted, all others will
2496 * be deferred. Since special requests may also be requeued requests,
2497 * a successful return doesn't guarantee the device will be
2498 * totally quiescent.
2499 *
2500 * Must be called with user context, may sleep.
2501 *
2502 * Returns zero if unsuccessful or an error if not.
2503 */
2504int
2505scsi_device_quiesce(struct scsi_device *sdev)
2506{
2507 struct request_queue *q = sdev->request_queue;
2508 int err;
2509
2510 /*
2511 * It is allowed to call scsi_device_quiesce() multiple times from
2512 * the same context but concurrent scsi_device_quiesce() calls are
2513 * not allowed.
2514 */
2515 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2516
2517 if (sdev->quiesced_by == current)
2518 return 0;
2519
2520 blk_set_pm_only(q);
2521
2522 blk_mq_freeze_queue(q);
2523 /*
2524 * Ensure that the effect of blk_set_pm_only() will be visible
2525 * for percpu_ref_tryget() callers that occur after the queue
2526 * unfreeze even if the queue was already frozen before this function
2527 * was called. See also https://lwn.net/Articles/573497/.
2528 */
2529 synchronize_rcu();
2530 blk_mq_unfreeze_queue(q);
2531
2532 mutex_lock(&sdev->state_mutex);
2533 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2534 if (err == 0)
2535 sdev->quiesced_by = current;
2536 else
2537 blk_clear_pm_only(q);
2538 mutex_unlock(&sdev->state_mutex);
2539
2540 return err;
2541}
2542EXPORT_SYMBOL(scsi_device_quiesce);
2543
2544/**
2545 * scsi_device_resume - Restart user issued commands to a quiesced device.
2546 * @sdev: scsi device to resume.
2547 *
2548 * Moves the device from quiesced back to running and restarts the
2549 * queues.
2550 *
2551 * Must be called with user context, may sleep.
2552 */
2553void scsi_device_resume(struct scsi_device *sdev)
2554{
2555 /* check if the device state was mutated prior to resume, and if
2556 * so assume the state is being managed elsewhere (for example
2557 * device deleted during suspend)
2558 */
2559 mutex_lock(&sdev->state_mutex);
2560 if (sdev->quiesced_by) {
2561 sdev->quiesced_by = NULL;
2562 blk_clear_pm_only(sdev->request_queue);
2563 }
2564 if (sdev->sdev_state == SDEV_QUIESCE)
2565 scsi_device_set_state(sdev, SDEV_RUNNING);
2566 mutex_unlock(&sdev->state_mutex);
2567}
2568EXPORT_SYMBOL(scsi_device_resume);
2569
2570static void
2571device_quiesce_fn(struct scsi_device *sdev, void *data)
2572{
2573 scsi_device_quiesce(sdev);
2574}
2575
2576void
2577scsi_target_quiesce(struct scsi_target *starget)
2578{
2579 starget_for_each_device(starget, NULL, device_quiesce_fn);
2580}
2581EXPORT_SYMBOL(scsi_target_quiesce);
2582
2583static void
2584device_resume_fn(struct scsi_device *sdev, void *data)
2585{
2586 scsi_device_resume(sdev);
2587}
2588
2589void
2590scsi_target_resume(struct scsi_target *starget)
2591{
2592 starget_for_each_device(starget, NULL, device_resume_fn);
2593}
2594EXPORT_SYMBOL(scsi_target_resume);
2595
2596/**
2597 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2598 * @sdev: device to block
2599 *
2600 * Pause SCSI command processing on the specified device. Does not sleep.
2601 *
2602 * Returns zero if successful or a negative error code upon failure.
2603 *
2604 * Notes:
2605 * This routine transitions the device to the SDEV_BLOCK state (which must be
2606 * a legal transition). When the device is in this state, command processing
2607 * is paused until the device leaves the SDEV_BLOCK state. See also
2608 * scsi_internal_device_unblock_nowait().
2609 */
2610int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2611{
2612 struct request_queue *q = sdev->request_queue;
2613 int err = 0;
2614
2615 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2616 if (err) {
2617 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2618
2619 if (err)
2620 return err;
2621 }
2622
2623 /*
2624 * The device has transitioned to SDEV_BLOCK. Stop the
2625 * block layer from calling the midlayer with this device's
2626 * request queue.
2627 */
2628 blk_mq_quiesce_queue_nowait(q);
2629 return 0;
2630}
2631EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2632
2633/**
2634 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2635 * @sdev: device to block
2636 *
2637 * Pause SCSI command processing on the specified device and wait until all
2638 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2639 *
2640 * Returns zero if successful or a negative error code upon failure.
2641 *
2642 * Note:
2643 * This routine transitions the device to the SDEV_BLOCK state (which must be
2644 * a legal transition). When the device is in this state, command processing
2645 * is paused until the device leaves the SDEV_BLOCK state. See also
2646 * scsi_internal_device_unblock().
2647 */
2648static int scsi_internal_device_block(struct scsi_device *sdev)
2649{
2650 struct request_queue *q = sdev->request_queue;
2651 int err;
2652
2653 mutex_lock(&sdev->state_mutex);
2654 err = scsi_internal_device_block_nowait(sdev);
2655 if (err == 0)
2656 blk_mq_quiesce_queue(q);
2657 mutex_unlock(&sdev->state_mutex);
2658
2659 return err;
2660}
2661
2662void scsi_start_queue(struct scsi_device *sdev)
2663{
2664 struct request_queue *q = sdev->request_queue;
2665
2666 blk_mq_unquiesce_queue(q);
2667}
2668
2669/**
2670 * scsi_internal_device_unblock_nowait - resume a device after a block request
2671 * @sdev: device to resume
2672 * @new_state: state to set the device to after unblocking
2673 *
2674 * Restart the device queue for a previously suspended SCSI device. Does not
2675 * sleep.
2676 *
2677 * Returns zero if successful or a negative error code upon failure.
2678 *
2679 * Notes:
2680 * This routine transitions the device to the SDEV_RUNNING state or to one of
2681 * the offline states (which must be a legal transition) allowing the midlayer
2682 * to goose the queue for this device.
2683 */
2684int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2685 enum scsi_device_state new_state)
2686{
2687 switch (new_state) {
2688 case SDEV_RUNNING:
2689 case SDEV_TRANSPORT_OFFLINE:
2690 break;
2691 default:
2692 return -EINVAL;
2693 }
2694
2695 /*
2696 * Try to transition the scsi device to SDEV_RUNNING or one of the
2697 * offlined states and goose the device queue if successful.
2698 */
2699 switch (sdev->sdev_state) {
2700 case SDEV_BLOCK:
2701 case SDEV_TRANSPORT_OFFLINE:
2702 sdev->sdev_state = new_state;
2703 break;
2704 case SDEV_CREATED_BLOCK:
2705 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2706 new_state == SDEV_OFFLINE)
2707 sdev->sdev_state = new_state;
2708 else
2709 sdev->sdev_state = SDEV_CREATED;
2710 break;
2711 case SDEV_CANCEL:
2712 case SDEV_OFFLINE:
2713 break;
2714 default:
2715 return -EINVAL;
2716 }
2717 scsi_start_queue(sdev);
2718
2719 return 0;
2720}
2721EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2722
2723/**
2724 * scsi_internal_device_unblock - resume a device after a block request
2725 * @sdev: device to resume
2726 * @new_state: state to set the device to after unblocking
2727 *
2728 * Restart the device queue for a previously suspended SCSI device. May sleep.
2729 *
2730 * Returns zero if successful or a negative error code upon failure.
2731 *
2732 * Notes:
2733 * This routine transitions the device to the SDEV_RUNNING state or to one of
2734 * the offline states (which must be a legal transition) allowing the midlayer
2735 * to goose the queue for this device.
2736 */
2737static int scsi_internal_device_unblock(struct scsi_device *sdev,
2738 enum scsi_device_state new_state)
2739{
2740 int ret;
2741
2742 mutex_lock(&sdev->state_mutex);
2743 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2744 mutex_unlock(&sdev->state_mutex);
2745
2746 return ret;
2747}
2748
2749static void
2750device_block(struct scsi_device *sdev, void *data)
2751{
2752 int ret;
2753
2754 ret = scsi_internal_device_block(sdev);
2755
2756 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2757 dev_name(&sdev->sdev_gendev), ret);
2758}
2759
2760static int
2761target_block(struct device *dev, void *data)
2762{
2763 if (scsi_is_target_device(dev))
2764 starget_for_each_device(to_scsi_target(dev), NULL,
2765 device_block);
2766 return 0;
2767}
2768
2769void
2770scsi_target_block(struct device *dev)
2771{
2772 if (scsi_is_target_device(dev))
2773 starget_for_each_device(to_scsi_target(dev), NULL,
2774 device_block);
2775 else
2776 device_for_each_child(dev, NULL, target_block);
2777}
2778EXPORT_SYMBOL_GPL(scsi_target_block);
2779
2780static void
2781device_unblock(struct scsi_device *sdev, void *data)
2782{
2783 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2784}
2785
2786static int
2787target_unblock(struct device *dev, void *data)
2788{
2789 if (scsi_is_target_device(dev))
2790 starget_for_each_device(to_scsi_target(dev), data,
2791 device_unblock);
2792 return 0;
2793}
2794
2795void
2796scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2797{
2798 if (scsi_is_target_device(dev))
2799 starget_for_each_device(to_scsi_target(dev), &new_state,
2800 device_unblock);
2801 else
2802 device_for_each_child(dev, &new_state, target_unblock);
2803}
2804EXPORT_SYMBOL_GPL(scsi_target_unblock);
2805
2806int
2807scsi_host_block(struct Scsi_Host *shost)
2808{
2809 struct scsi_device *sdev;
2810 int ret = 0;
2811
2812 /*
2813 * Call scsi_internal_device_block_nowait so we can avoid
2814 * calling synchronize_rcu() for each LUN.
2815 */
2816 shost_for_each_device(sdev, shost) {
2817 mutex_lock(&sdev->state_mutex);
2818 ret = scsi_internal_device_block_nowait(sdev);
2819 mutex_unlock(&sdev->state_mutex);
2820 if (ret) {
2821 scsi_device_put(sdev);
2822 break;
2823 }
2824 }
2825
2826 /*
2827 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2828 * calling synchronize_rcu() once is enough.
2829 */
2830 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2831
2832 if (!ret)
2833 synchronize_rcu();
2834
2835 return ret;
2836}
2837EXPORT_SYMBOL_GPL(scsi_host_block);
2838
2839int
2840scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2841{
2842 struct scsi_device *sdev;
2843 int ret = 0;
2844
2845 shost_for_each_device(sdev, shost) {
2846 ret = scsi_internal_device_unblock(sdev, new_state);
2847 if (ret) {
2848 scsi_device_put(sdev);
2849 break;
2850 }
2851 }
2852 return ret;
2853}
2854EXPORT_SYMBOL_GPL(scsi_host_unblock);
2855
2856/**
2857 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2858 * @sgl: scatter-gather list
2859 * @sg_count: number of segments in sg
2860 * @offset: offset in bytes into sg, on return offset into the mapped area
2861 * @len: bytes to map, on return number of bytes mapped
2862 *
2863 * Returns virtual address of the start of the mapped page
2864 */
2865void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2866 size_t *offset, size_t *len)
2867{
2868 int i;
2869 size_t sg_len = 0, len_complete = 0;
2870 struct scatterlist *sg;
2871 struct page *page;
2872
2873 WARN_ON(!irqs_disabled());
2874
2875 for_each_sg(sgl, sg, sg_count, i) {
2876 len_complete = sg_len; /* Complete sg-entries */
2877 sg_len += sg->length;
2878 if (sg_len > *offset)
2879 break;
2880 }
2881
2882 if (unlikely(i == sg_count)) {
2883 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2884 "elements %d\n",
2885 __func__, sg_len, *offset, sg_count);
2886 WARN_ON(1);
2887 return NULL;
2888 }
2889
2890 /* Offset starting from the beginning of first page in this sg-entry */
2891 *offset = *offset - len_complete + sg->offset;
2892
2893 /* Assumption: contiguous pages can be accessed as "page + i" */
2894 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2895 *offset &= ~PAGE_MASK;
2896
2897 /* Bytes in this sg-entry from *offset to the end of the page */
2898 sg_len = PAGE_SIZE - *offset;
2899 if (*len > sg_len)
2900 *len = sg_len;
2901
2902 return kmap_atomic(page);
2903}
2904EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2905
2906/**
2907 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2908 * @virt: virtual address to be unmapped
2909 */
2910void scsi_kunmap_atomic_sg(void *virt)
2911{
2912 kunmap_atomic(virt);
2913}
2914EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2915
2916void sdev_disable_disk_events(struct scsi_device *sdev)
2917{
2918 atomic_inc(&sdev->disk_events_disable_depth);
2919}
2920EXPORT_SYMBOL(sdev_disable_disk_events);
2921
2922void sdev_enable_disk_events(struct scsi_device *sdev)
2923{
2924 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2925 return;
2926 atomic_dec(&sdev->disk_events_disable_depth);
2927}
2928EXPORT_SYMBOL(sdev_enable_disk_events);
2929
2930/**
2931 * scsi_vpd_lun_id - return a unique device identification
2932 * @sdev: SCSI device
2933 * @id: buffer for the identification
2934 * @id_len: length of the buffer
2935 *
2936 * Copies a unique device identification into @id based
2937 * on the information in the VPD page 0x83 of the device.
2938 * The string will be formatted as a SCSI name string.
2939 *
2940 * Returns the length of the identification or error on failure.
2941 * If the identifier is longer than the supplied buffer the actual
2942 * identifier length is returned and the buffer is not zero-padded.
2943 */
2944int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2945{
2946 u8 cur_id_type = 0xff;
2947 u8 cur_id_size = 0;
2948 const unsigned char *d, *cur_id_str;
2949 const struct scsi_vpd *vpd_pg83;
2950 int id_size = -EINVAL;
2951
2952 rcu_read_lock();
2953 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2954 if (!vpd_pg83) {
2955 rcu_read_unlock();
2956 return -ENXIO;
2957 }
2958
2959 /*
2960 * Look for the correct descriptor.
2961 * Order of preference for lun descriptor:
2962 * - SCSI name string
2963 * - NAA IEEE Registered Extended
2964 * - EUI-64 based 16-byte
2965 * - EUI-64 based 12-byte
2966 * - NAA IEEE Registered
2967 * - NAA IEEE Extended
2968 * - T10 Vendor ID
2969 * as longer descriptors reduce the likelyhood
2970 * of identification clashes.
2971 */
2972
2973 /* The id string must be at least 20 bytes + terminating NULL byte */
2974 if (id_len < 21) {
2975 rcu_read_unlock();
2976 return -EINVAL;
2977 }
2978
2979 memset(id, 0, id_len);
2980 d = vpd_pg83->data + 4;
2981 while (d < vpd_pg83->data + vpd_pg83->len) {
2982 /* Skip designators not referring to the LUN */
2983 if ((d[1] & 0x30) != 0x00)
2984 goto next_desig;
2985
2986 switch (d[1] & 0xf) {
2987 case 0x1:
2988 /* T10 Vendor ID */
2989 if (cur_id_size > d[3])
2990 break;
2991 /* Prefer anything */
2992 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2993 break;
2994 cur_id_size = d[3];
2995 if (cur_id_size + 4 > id_len)
2996 cur_id_size = id_len - 4;
2997 cur_id_str = d + 4;
2998 cur_id_type = d[1] & 0xf;
2999 id_size = snprintf(id, id_len, "t10.%*pE",
3000 cur_id_size, cur_id_str);
3001 break;
3002 case 0x2:
3003 /* EUI-64 */
3004 if (cur_id_size > d[3])
3005 break;
3006 /* Prefer NAA IEEE Registered Extended */
3007 if (cur_id_type == 0x3 &&
3008 cur_id_size == d[3])
3009 break;
3010 cur_id_size = d[3];
3011 cur_id_str = d + 4;
3012 cur_id_type = d[1] & 0xf;
3013 switch (cur_id_size) {
3014 case 8:
3015 id_size = snprintf(id, id_len,
3016 "eui.%8phN",
3017 cur_id_str);
3018 break;
3019 case 12:
3020 id_size = snprintf(id, id_len,
3021 "eui.%12phN",
3022 cur_id_str);
3023 break;
3024 case 16:
3025 id_size = snprintf(id, id_len,
3026 "eui.%16phN",
3027 cur_id_str);
3028 break;
3029 default:
3030 cur_id_size = 0;
3031 break;
3032 }
3033 break;
3034 case 0x3:
3035 /* NAA */
3036 if (cur_id_size > d[3])
3037 break;
3038 cur_id_size = d[3];
3039 cur_id_str = d + 4;
3040 cur_id_type = d[1] & 0xf;
3041 switch (cur_id_size) {
3042 case 8:
3043 id_size = snprintf(id, id_len,
3044 "naa.%8phN",
3045 cur_id_str);
3046 break;
3047 case 16:
3048 id_size = snprintf(id, id_len,
3049 "naa.%16phN",
3050 cur_id_str);
3051 break;
3052 default:
3053 cur_id_size = 0;
3054 break;
3055 }
3056 break;
3057 case 0x8:
3058 /* SCSI name string */
3059 if (cur_id_size + 4 > d[3])
3060 break;
3061 /* Prefer others for truncated descriptor */
3062 if (cur_id_size && d[3] > id_len)
3063 break;
3064 cur_id_size = id_size = d[3];
3065 cur_id_str = d + 4;
3066 cur_id_type = d[1] & 0xf;
3067 if (cur_id_size >= id_len)
3068 cur_id_size = id_len - 1;
3069 memcpy(id, cur_id_str, cur_id_size);
3070 /* Decrease priority for truncated descriptor */
3071 if (cur_id_size != id_size)
3072 cur_id_size = 6;
3073 break;
3074 default:
3075 break;
3076 }
3077next_desig:
3078 d += d[3] + 4;
3079 }
3080 rcu_read_unlock();
3081
3082 return id_size;
3083}
3084EXPORT_SYMBOL(scsi_vpd_lun_id);
3085
3086/*
3087 * scsi_vpd_tpg_id - return a target port group identifier
3088 * @sdev: SCSI device
3089 *
3090 * Returns the Target Port Group identifier from the information
3091 * froom VPD page 0x83 of the device.
3092 *
3093 * Returns the identifier or error on failure.
3094 */
3095int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3096{
3097 const unsigned char *d;
3098 const struct scsi_vpd *vpd_pg83;
3099 int group_id = -EAGAIN, rel_port = -1;
3100
3101 rcu_read_lock();
3102 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3103 if (!vpd_pg83) {
3104 rcu_read_unlock();
3105 return -ENXIO;
3106 }
3107
3108 d = vpd_pg83->data + 4;
3109 while (d < vpd_pg83->data + vpd_pg83->len) {
3110 switch (d[1] & 0xf) {
3111 case 0x4:
3112 /* Relative target port */
3113 rel_port = get_unaligned_be16(&d[6]);
3114 break;
3115 case 0x5:
3116 /* Target port group */
3117 group_id = get_unaligned_be16(&d[6]);
3118 break;
3119 default:
3120 break;
3121 }
3122 d += d[3] + 4;
3123 }
3124 rcu_read_unlock();
3125
3126 if (group_id >= 0 && rel_id && rel_port != -1)
3127 *rel_id = rel_port;
3128
3129 return group_id;
3130}
3131EXPORT_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);