media: stv06xx: add missing descriptor sanity checks
[linux/fpc-iii.git] / drivers / scsi / scsi_lib.c
blob3e7a45d0dacad97f8828eb01fdc997759705470a
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 1999 Eric Youngdale
4 * Copyright (C) 2014 Christoph Hellwig
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.
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>
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>
37 #include <trace/events/scsi.h>
39 #include "scsi_debugfs.h"
40 #include "scsi_priv.h"
41 #include "scsi_logging.h"
44 * Size of integrity metadata is usually small, 1 inline sg should
45 * cover normal cases.
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
55 static struct kmem_cache *scsi_sdb_cache;
56 static struct kmem_cache *scsi_sense_cache;
57 static struct kmem_cache *scsi_sense_isadma_cache;
58 static DEFINE_MUTEX(scsi_sense_cache_mutex);
60 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
62 static inline struct kmem_cache *
63 scsi_select_sense_cache(bool unchecked_isa_dma)
65 return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
68 static void scsi_free_sense_buffer(bool unchecked_isa_dma,
69 unsigned char *sense_buffer)
71 kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
72 sense_buffer);
75 static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
76 gfp_t gfp_mask, int numa_node)
78 return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
79 gfp_mask, numa_node);
82 int scsi_init_sense_cache(struct Scsi_Host *shost)
84 struct kmem_cache *cache;
85 int ret = 0;
87 mutex_lock(&scsi_sense_cache_mutex);
88 cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
89 if (cache)
90 goto exit;
92 if (shost->unchecked_isa_dma) {
93 scsi_sense_isadma_cache =
94 kmem_cache_create("scsi_sense_cache(DMA)",
95 SCSI_SENSE_BUFFERSIZE, 0,
96 SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
97 if (!scsi_sense_isadma_cache)
98 ret = -ENOMEM;
99 } else {
100 scsi_sense_cache =
101 kmem_cache_create_usercopy("scsi_sense_cache",
102 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
103 0, SCSI_SENSE_BUFFERSIZE, NULL);
104 if (!scsi_sense_cache)
105 ret = -ENOMEM;
107 exit:
108 mutex_unlock(&scsi_sense_cache_mutex);
109 return ret;
113 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
114 * not change behaviour from the previous unplug mechanism, experimentation
115 * may prove this needs changing.
117 #define SCSI_QUEUE_DELAY 3
119 static void
120 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
122 struct Scsi_Host *host = cmd->device->host;
123 struct scsi_device *device = cmd->device;
124 struct scsi_target *starget = scsi_target(device);
127 * Set the appropriate busy bit for the device/host.
129 * If the host/device isn't busy, assume that something actually
130 * completed, and that we should be able to queue a command now.
132 * Note that the prior mid-layer assumption that any host could
133 * always queue at least one command is now broken. The mid-layer
134 * will implement a user specifiable stall (see
135 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
136 * if a command is requeued with no other commands outstanding
137 * either for the device or for the host.
139 switch (reason) {
140 case SCSI_MLQUEUE_HOST_BUSY:
141 atomic_set(&host->host_blocked, host->max_host_blocked);
142 break;
143 case SCSI_MLQUEUE_DEVICE_BUSY:
144 case SCSI_MLQUEUE_EH_RETRY:
145 atomic_set(&device->device_blocked,
146 device->max_device_blocked);
147 break;
148 case SCSI_MLQUEUE_TARGET_BUSY:
149 atomic_set(&starget->target_blocked,
150 starget->max_target_blocked);
151 break;
155 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
157 if (cmd->request->rq_flags & RQF_DONTPREP) {
158 cmd->request->rq_flags &= ~RQF_DONTPREP;
159 scsi_mq_uninit_cmd(cmd);
160 } else {
161 WARN_ON_ONCE(true);
163 blk_mq_requeue_request(cmd->request, true);
167 * __scsi_queue_insert - private queue insertion
168 * @cmd: The SCSI command being requeued
169 * @reason: The reason for the requeue
170 * @unbusy: Whether the queue should be unbusied
172 * This is a private queue insertion. The public interface
173 * scsi_queue_insert() always assumes the queue should be unbusied
174 * because it's always called before the completion. This function is
175 * for a requeue after completion, which should only occur in this
176 * file.
178 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
180 struct scsi_device *device = cmd->device;
182 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
183 "Inserting command %p into mlqueue\n", cmd));
185 scsi_set_blocked(cmd, reason);
188 * Decrement the counters, since these commands are no longer
189 * active on the host/device.
191 if (unbusy)
192 scsi_device_unbusy(device, cmd);
195 * Requeue this command. It will go before all other commands
196 * that are already in the queue. Schedule requeue work under
197 * lock such that the kblockd_schedule_work() call happens
198 * before blk_cleanup_queue() finishes.
200 cmd->result = 0;
202 blk_mq_requeue_request(cmd->request, true);
206 * Function: scsi_queue_insert()
208 * Purpose: Insert a command in the midlevel queue.
210 * Arguments: cmd - command that we are adding to queue.
211 * reason - why we are inserting command to queue.
213 * Lock status: Assumed that lock is not held upon entry.
215 * Returns: Nothing.
217 * Notes: We do this for one of two cases. Either the host is busy
218 * and it cannot accept any more commands for the time being,
219 * or the device returned QUEUE_FULL and can accept no more
220 * commands.
221 * Notes: This could be called either from an interrupt context or a
222 * normal process context.
224 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
226 __scsi_queue_insert(cmd, reason, true);
231 * __scsi_execute - insert request and wait for the result
232 * @sdev: scsi device
233 * @cmd: scsi command
234 * @data_direction: data direction
235 * @buffer: data buffer
236 * @bufflen: len of buffer
237 * @sense: optional sense buffer
238 * @sshdr: optional decoded sense header
239 * @timeout: request timeout in seconds
240 * @retries: number of times to retry request
241 * @flags: flags for ->cmd_flags
242 * @rq_flags: flags for ->rq_flags
243 * @resid: optional residual length
245 * Returns the scsi_cmnd result field if a command was executed, or a negative
246 * Linux error code if we didn't get that far.
248 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
249 int data_direction, void *buffer, unsigned bufflen,
250 unsigned char *sense, struct scsi_sense_hdr *sshdr,
251 int timeout, int retries, u64 flags, req_flags_t rq_flags,
252 int *resid)
254 struct request *req;
255 struct scsi_request *rq;
256 int ret = DRIVER_ERROR << 24;
258 req = blk_get_request(sdev->request_queue,
259 data_direction == DMA_TO_DEVICE ?
260 REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
261 if (IS_ERR(req))
262 return ret;
263 rq = scsi_req(req);
265 if (bufflen && blk_rq_map_kern(sdev->request_queue, req,
266 buffer, bufflen, GFP_NOIO))
267 goto out;
269 rq->cmd_len = COMMAND_SIZE(cmd[0]);
270 memcpy(rq->cmd, cmd, rq->cmd_len);
271 rq->retries = retries;
272 req->timeout = timeout;
273 req->cmd_flags |= flags;
274 req->rq_flags |= rq_flags | RQF_QUIET;
277 * head injection *required* here otherwise quiesce won't work
279 blk_execute_rq(req->q, NULL, req, 1);
282 * Some devices (USB mass-storage in particular) may transfer
283 * garbage data together with a residue indicating that the data
284 * is invalid. Prevent the garbage from being misinterpreted
285 * and prevent security leaks by zeroing out the excess data.
287 if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
288 memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
290 if (resid)
291 *resid = rq->resid_len;
292 if (sense && rq->sense_len)
293 memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
294 if (sshdr)
295 scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
296 ret = rq->result;
297 out:
298 blk_put_request(req);
300 return ret;
302 EXPORT_SYMBOL(__scsi_execute);
305 * Function: scsi_init_cmd_errh()
307 * Purpose: Initialize cmd fields related to error handling.
309 * Arguments: cmd - command that is ready to be queued.
311 * Notes: This function has the job of initializing a number of
312 * fields related to error handling. Typically this will
313 * be called once for each command, as required.
315 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
317 scsi_set_resid(cmd, 0);
318 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
319 if (cmd->cmd_len == 0)
320 cmd->cmd_len = scsi_command_size(cmd->cmnd);
324 * Wake up the error handler if necessary. Avoid as follows that the error
325 * handler is not woken up if host in-flight requests number ==
326 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
327 * with an RCU read lock in this function to ensure that this function in
328 * its entirety either finishes before scsi_eh_scmd_add() increases the
329 * host_failed counter or that it notices the shost state change made by
330 * scsi_eh_scmd_add().
332 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
334 unsigned long flags;
336 rcu_read_lock();
337 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
338 if (unlikely(scsi_host_in_recovery(shost))) {
339 spin_lock_irqsave(shost->host_lock, flags);
340 if (shost->host_failed || shost->host_eh_scheduled)
341 scsi_eh_wakeup(shost);
342 spin_unlock_irqrestore(shost->host_lock, flags);
344 rcu_read_unlock();
347 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
349 struct Scsi_Host *shost = sdev->host;
350 struct scsi_target *starget = scsi_target(sdev);
352 scsi_dec_host_busy(shost, cmd);
354 if (starget->can_queue > 0)
355 atomic_dec(&starget->target_busy);
357 atomic_dec(&sdev->device_busy);
360 static void scsi_kick_queue(struct request_queue *q)
362 blk_mq_run_hw_queues(q, false);
366 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
367 * and call blk_run_queue for all the scsi_devices on the target -
368 * including current_sdev first.
370 * Called with *no* scsi locks held.
372 static void scsi_single_lun_run(struct scsi_device *current_sdev)
374 struct Scsi_Host *shost = current_sdev->host;
375 struct scsi_device *sdev, *tmp;
376 struct scsi_target *starget = scsi_target(current_sdev);
377 unsigned long flags;
379 spin_lock_irqsave(shost->host_lock, flags);
380 starget->starget_sdev_user = NULL;
381 spin_unlock_irqrestore(shost->host_lock, flags);
384 * Call blk_run_queue for all LUNs on the target, starting with
385 * current_sdev. We race with others (to set starget_sdev_user),
386 * but in most cases, we will be first. Ideally, each LU on the
387 * target would get some limited time or requests on the target.
389 scsi_kick_queue(current_sdev->request_queue);
391 spin_lock_irqsave(shost->host_lock, flags);
392 if (starget->starget_sdev_user)
393 goto out;
394 list_for_each_entry_safe(sdev, tmp, &starget->devices,
395 same_target_siblings) {
396 if (sdev == current_sdev)
397 continue;
398 if (scsi_device_get(sdev))
399 continue;
401 spin_unlock_irqrestore(shost->host_lock, flags);
402 scsi_kick_queue(sdev->request_queue);
403 spin_lock_irqsave(shost->host_lock, flags);
405 scsi_device_put(sdev);
407 out:
408 spin_unlock_irqrestore(shost->host_lock, flags);
411 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
413 if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
414 return true;
415 if (atomic_read(&sdev->device_blocked) > 0)
416 return true;
417 return false;
420 static inline bool scsi_target_is_busy(struct scsi_target *starget)
422 if (starget->can_queue > 0) {
423 if (atomic_read(&starget->target_busy) >= starget->can_queue)
424 return true;
425 if (atomic_read(&starget->target_blocked) > 0)
426 return true;
428 return false;
431 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
433 if (atomic_read(&shost->host_blocked) > 0)
434 return true;
435 if (shost->host_self_blocked)
436 return true;
437 return false;
440 static void scsi_starved_list_run(struct Scsi_Host *shost)
442 LIST_HEAD(starved_list);
443 struct scsi_device *sdev;
444 unsigned long flags;
446 spin_lock_irqsave(shost->host_lock, flags);
447 list_splice_init(&shost->starved_list, &starved_list);
449 while (!list_empty(&starved_list)) {
450 struct request_queue *slq;
453 * As long as shost is accepting commands and we have
454 * starved queues, call blk_run_queue. scsi_request_fn
455 * drops the queue_lock and can add us back to the
456 * starved_list.
458 * host_lock protects the starved_list and starved_entry.
459 * scsi_request_fn must get the host_lock before checking
460 * or modifying starved_list or starved_entry.
462 if (scsi_host_is_busy(shost))
463 break;
465 sdev = list_entry(starved_list.next,
466 struct scsi_device, starved_entry);
467 list_del_init(&sdev->starved_entry);
468 if (scsi_target_is_busy(scsi_target(sdev))) {
469 list_move_tail(&sdev->starved_entry,
470 &shost->starved_list);
471 continue;
475 * Once we drop the host lock, a racing scsi_remove_device()
476 * call may remove the sdev from the starved list and destroy
477 * it and the queue. Mitigate by taking a reference to the
478 * queue and never touching the sdev again after we drop the
479 * host lock. Note: if __scsi_remove_device() invokes
480 * blk_cleanup_queue() before the queue is run from this
481 * function then blk_run_queue() will return immediately since
482 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
484 slq = sdev->request_queue;
485 if (!blk_get_queue(slq))
486 continue;
487 spin_unlock_irqrestore(shost->host_lock, flags);
489 scsi_kick_queue(slq);
490 blk_put_queue(slq);
492 spin_lock_irqsave(shost->host_lock, flags);
494 /* put any unprocessed entries back */
495 list_splice(&starved_list, &shost->starved_list);
496 spin_unlock_irqrestore(shost->host_lock, flags);
500 * Function: scsi_run_queue()
502 * Purpose: Select a proper request queue to serve next
504 * Arguments: q - last request's queue
506 * Returns: Nothing
508 * Notes: The previous command was completely finished, start
509 * a new one if possible.
511 static void scsi_run_queue(struct request_queue *q)
513 struct scsi_device *sdev = q->queuedata;
515 if (scsi_target(sdev)->single_lun)
516 scsi_single_lun_run(sdev);
517 if (!list_empty(&sdev->host->starved_list))
518 scsi_starved_list_run(sdev->host);
520 blk_mq_run_hw_queues(q, false);
523 void scsi_requeue_run_queue(struct work_struct *work)
525 struct scsi_device *sdev;
526 struct request_queue *q;
528 sdev = container_of(work, struct scsi_device, requeue_work);
529 q = sdev->request_queue;
530 scsi_run_queue(q);
533 void scsi_run_host_queues(struct Scsi_Host *shost)
535 struct scsi_device *sdev;
537 shost_for_each_device(sdev, shost)
538 scsi_run_queue(sdev->request_queue);
541 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
543 if (!blk_rq_is_passthrough(cmd->request)) {
544 struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
546 if (drv->uninit_command)
547 drv->uninit_command(cmd);
551 static void scsi_mq_free_sgtables(struct scsi_cmnd *cmd)
553 if (cmd->sdb.table.nents)
554 sg_free_table_chained(&cmd->sdb.table,
555 SCSI_INLINE_SG_CNT);
556 if (scsi_prot_sg_count(cmd))
557 sg_free_table_chained(&cmd->prot_sdb->table,
558 SCSI_INLINE_PROT_SG_CNT);
561 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
563 scsi_mq_free_sgtables(cmd);
564 scsi_uninit_cmd(cmd);
565 scsi_del_cmd_from_list(cmd);
568 /* Returns false when no more bytes to process, true if there are more */
569 static bool scsi_end_request(struct request *req, blk_status_t error,
570 unsigned int bytes)
572 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
573 struct scsi_device *sdev = cmd->device;
574 struct request_queue *q = sdev->request_queue;
576 if (blk_update_request(req, error, bytes))
577 return true;
579 if (blk_queue_add_random(q))
580 add_disk_randomness(req->rq_disk);
582 if (!blk_rq_is_scsi(req)) {
583 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
584 cmd->flags &= ~SCMD_INITIALIZED;
588 * Calling rcu_barrier() is not necessary here because the
589 * SCSI error handler guarantees that the function called by
590 * call_rcu() has been called before scsi_end_request() is
591 * called.
593 destroy_rcu_head(&cmd->rcu);
596 * In the MQ case the command gets freed by __blk_mq_end_request,
597 * so we have to do all cleanup that depends on it earlier.
599 * We also can't kick the queues from irq context, so we
600 * will have to defer it to a workqueue.
602 scsi_mq_uninit_cmd(cmd);
605 * queue is still alive, so grab the ref for preventing it
606 * from being cleaned up during running queue.
608 percpu_ref_get(&q->q_usage_counter);
610 __blk_mq_end_request(req, error);
612 if (scsi_target(sdev)->single_lun ||
613 !list_empty(&sdev->host->starved_list))
614 kblockd_schedule_work(&sdev->requeue_work);
615 else
616 blk_mq_run_hw_queues(q, true);
618 percpu_ref_put(&q->q_usage_counter);
619 return false;
623 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
624 * @cmd: SCSI command
625 * @result: scsi error code
627 * Translate a SCSI result code into a blk_status_t value. May reset the host
628 * byte of @cmd->result.
630 static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
632 switch (host_byte(result)) {
633 case DID_OK:
635 * Also check the other bytes than the status byte in result
636 * to handle the case when a SCSI LLD sets result to
637 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
639 if (scsi_status_is_good(result) && (result & ~0xff) == 0)
640 return BLK_STS_OK;
641 return BLK_STS_IOERR;
642 case DID_TRANSPORT_FAILFAST:
643 return BLK_STS_TRANSPORT;
644 case DID_TARGET_FAILURE:
645 set_host_byte(cmd, DID_OK);
646 return BLK_STS_TARGET;
647 case DID_NEXUS_FAILURE:
648 set_host_byte(cmd, DID_OK);
649 return BLK_STS_NEXUS;
650 case DID_ALLOC_FAILURE:
651 set_host_byte(cmd, DID_OK);
652 return BLK_STS_NOSPC;
653 case DID_MEDIUM_ERROR:
654 set_host_byte(cmd, DID_OK);
655 return BLK_STS_MEDIUM;
656 default:
657 return BLK_STS_IOERR;
661 /* Helper for scsi_io_completion() when "reprep" action required. */
662 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
663 struct request_queue *q)
665 /* A new command will be prepared and issued. */
666 scsi_mq_requeue_cmd(cmd);
669 /* Helper for scsi_io_completion() when special action required. */
670 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
672 struct request_queue *q = cmd->device->request_queue;
673 struct request *req = cmd->request;
674 int level = 0;
675 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
676 ACTION_DELAYED_RETRY} action;
677 unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
678 struct scsi_sense_hdr sshdr;
679 bool sense_valid;
680 bool sense_current = true; /* false implies "deferred sense" */
681 blk_status_t blk_stat;
683 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
684 if (sense_valid)
685 sense_current = !scsi_sense_is_deferred(&sshdr);
687 blk_stat = scsi_result_to_blk_status(cmd, result);
689 if (host_byte(result) == DID_RESET) {
690 /* Third party bus reset or reset for error recovery
691 * reasons. Just retry the command and see what
692 * happens.
694 action = ACTION_RETRY;
695 } else if (sense_valid && sense_current) {
696 switch (sshdr.sense_key) {
697 case UNIT_ATTENTION:
698 if (cmd->device->removable) {
699 /* Detected disc change. Set a bit
700 * and quietly refuse further access.
702 cmd->device->changed = 1;
703 action = ACTION_FAIL;
704 } else {
705 /* Must have been a power glitch, or a
706 * bus reset. Could not have been a
707 * media change, so we just retry the
708 * command and see what happens.
710 action = ACTION_RETRY;
712 break;
713 case ILLEGAL_REQUEST:
714 /* If we had an ILLEGAL REQUEST returned, then
715 * we may have performed an unsupported
716 * command. The only thing this should be
717 * would be a ten byte read where only a six
718 * byte read was supported. Also, on a system
719 * where READ CAPACITY failed, we may have
720 * read past the end of the disk.
722 if ((cmd->device->use_10_for_rw &&
723 sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
724 (cmd->cmnd[0] == READ_10 ||
725 cmd->cmnd[0] == WRITE_10)) {
726 /* This will issue a new 6-byte command. */
727 cmd->device->use_10_for_rw = 0;
728 action = ACTION_REPREP;
729 } else if (sshdr.asc == 0x10) /* DIX */ {
730 action = ACTION_FAIL;
731 blk_stat = BLK_STS_PROTECTION;
732 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
733 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
734 action = ACTION_FAIL;
735 blk_stat = BLK_STS_TARGET;
736 } else
737 action = ACTION_FAIL;
738 break;
739 case ABORTED_COMMAND:
740 action = ACTION_FAIL;
741 if (sshdr.asc == 0x10) /* DIF */
742 blk_stat = BLK_STS_PROTECTION;
743 break;
744 case NOT_READY:
745 /* If the device is in the process of becoming
746 * ready, or has a temporary blockage, retry.
748 if (sshdr.asc == 0x04) {
749 switch (sshdr.ascq) {
750 case 0x01: /* becoming ready */
751 case 0x04: /* format in progress */
752 case 0x05: /* rebuild in progress */
753 case 0x06: /* recalculation in progress */
754 case 0x07: /* operation in progress */
755 case 0x08: /* Long write in progress */
756 case 0x09: /* self test in progress */
757 case 0x14: /* space allocation in progress */
758 case 0x1a: /* start stop unit in progress */
759 case 0x1b: /* sanitize in progress */
760 case 0x1d: /* configuration in progress */
761 case 0x24: /* depopulation in progress */
762 action = ACTION_DELAYED_RETRY;
763 break;
764 default:
765 action = ACTION_FAIL;
766 break;
768 } else
769 action = ACTION_FAIL;
770 break;
771 case VOLUME_OVERFLOW:
772 /* See SSC3rXX or current. */
773 action = ACTION_FAIL;
774 break;
775 default:
776 action = ACTION_FAIL;
777 break;
779 } else
780 action = ACTION_FAIL;
782 if (action != ACTION_FAIL &&
783 time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
784 action = ACTION_FAIL;
786 switch (action) {
787 case ACTION_FAIL:
788 /* Give up and fail the remainder of the request */
789 if (!(req->rq_flags & RQF_QUIET)) {
790 static DEFINE_RATELIMIT_STATE(_rs,
791 DEFAULT_RATELIMIT_INTERVAL,
792 DEFAULT_RATELIMIT_BURST);
794 if (unlikely(scsi_logging_level))
795 level =
796 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
797 SCSI_LOG_MLCOMPLETE_BITS);
800 * if logging is enabled the failure will be printed
801 * in scsi_log_completion(), so avoid duplicate messages
803 if (!level && __ratelimit(&_rs)) {
804 scsi_print_result(cmd, NULL, FAILED);
805 if (driver_byte(result) == DRIVER_SENSE)
806 scsi_print_sense(cmd);
807 scsi_print_command(cmd);
810 if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
811 return;
812 /*FALLTHRU*/
813 case ACTION_REPREP:
814 scsi_io_completion_reprep(cmd, q);
815 break;
816 case ACTION_RETRY:
817 /* Retry the same command immediately */
818 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
819 break;
820 case ACTION_DELAYED_RETRY:
821 /* Retry the same command after a delay */
822 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
823 break;
828 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
829 * new result that may suppress further error checking. Also modifies
830 * *blk_statp in some cases.
832 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
833 blk_status_t *blk_statp)
835 bool sense_valid;
836 bool sense_current = true; /* false implies "deferred sense" */
837 struct request *req = cmd->request;
838 struct scsi_sense_hdr sshdr;
840 sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
841 if (sense_valid)
842 sense_current = !scsi_sense_is_deferred(&sshdr);
844 if (blk_rq_is_passthrough(req)) {
845 if (sense_valid) {
847 * SG_IO wants current and deferred errors
849 scsi_req(req)->sense_len =
850 min(8 + cmd->sense_buffer[7],
851 SCSI_SENSE_BUFFERSIZE);
853 if (sense_current)
854 *blk_statp = scsi_result_to_blk_status(cmd, result);
855 } else if (blk_rq_bytes(req) == 0 && sense_current) {
857 * Flush commands do not transfers any data, and thus cannot use
858 * good_bytes != blk_rq_bytes(req) as the signal for an error.
859 * This sets *blk_statp explicitly for the problem case.
861 *blk_statp = scsi_result_to_blk_status(cmd, result);
864 * Recovered errors need reporting, but they're always treated as
865 * success, so fiddle the result code here. For passthrough requests
866 * we already took a copy of the original into sreq->result which
867 * is what gets returned to the user
869 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
870 bool do_print = true;
872 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
873 * skip print since caller wants ATA registers. Only occurs
874 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
876 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
877 do_print = false;
878 else if (req->rq_flags & RQF_QUIET)
879 do_print = false;
880 if (do_print)
881 scsi_print_sense(cmd);
882 result = 0;
883 /* for passthrough, *blk_statp may be set */
884 *blk_statp = BLK_STS_OK;
887 * Another corner case: the SCSI status byte is non-zero but 'good'.
888 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
889 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
890 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
891 * intermediate statuses (both obsolete in SAM-4) as good.
893 if (status_byte(result) && scsi_status_is_good(result)) {
894 result = 0;
895 *blk_statp = BLK_STS_OK;
897 return result;
901 * Function: scsi_io_completion()
903 * Purpose: Completion processing for block device I/O requests.
905 * Arguments: cmd - command that is finished.
907 * Lock status: Assumed that no lock is held upon entry.
909 * Returns: Nothing
911 * Notes: We will finish off the specified number of sectors. If we
912 * are done, the command block will be released and the queue
913 * function will be goosed. If we are not done then we have to
914 * figure out what to do next:
916 * a) We can call scsi_requeue_command(). The request
917 * will be unprepared and put back on the queue. Then
918 * a new command will be created for it. This should
919 * be used if we made forward progress, or if we want
920 * to switch from READ(10) to READ(6) for example.
922 * b) We can call __scsi_queue_insert(). The request will
923 * be put back on the queue and retried using the same
924 * command as before, possibly after a delay.
926 * c) We can call scsi_end_request() with blk_stat other than
927 * BLK_STS_OK, to fail the remainder of the request.
929 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
931 int result = cmd->result;
932 struct request_queue *q = cmd->device->request_queue;
933 struct request *req = cmd->request;
934 blk_status_t blk_stat = BLK_STS_OK;
936 if (unlikely(result)) /* a nz result may or may not be an error */
937 result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
939 if (unlikely(blk_rq_is_passthrough(req))) {
941 * scsi_result_to_blk_status may have reset the host_byte
943 scsi_req(req)->result = cmd->result;
947 * Next deal with any sectors which we were able to correctly
948 * handle.
950 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
951 "%u sectors total, %d bytes done.\n",
952 blk_rq_sectors(req), good_bytes));
955 * Next deal with any sectors which we were able to correctly
956 * handle. Failed, zero length commands always need to drop down
957 * to retry code. Fast path should return in this block.
959 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
960 if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
961 return; /* no bytes remaining */
964 /* Kill remainder if no retries. */
965 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
966 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
967 WARN_ONCE(true,
968 "Bytes remaining after failed, no-retry command");
969 return;
973 * If there had been no error, but we have leftover bytes in the
974 * requeues just queue the command up again.
976 if (likely(result == 0))
977 scsi_io_completion_reprep(cmd, q);
978 else
979 scsi_io_completion_action(cmd, result);
982 static blk_status_t scsi_init_sgtable(struct request *req,
983 struct scsi_data_buffer *sdb)
985 int count;
988 * If sg table allocation fails, requeue request later.
990 if (unlikely(sg_alloc_table_chained(&sdb->table,
991 blk_rq_nr_phys_segments(req), sdb->table.sgl,
992 SCSI_INLINE_SG_CNT)))
993 return BLK_STS_RESOURCE;
996 * Next, walk the list, and fill in the addresses and sizes of
997 * each segment.
999 count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
1000 BUG_ON(count > sdb->table.nents);
1001 sdb->table.nents = count;
1002 sdb->length = blk_rq_payload_bytes(req);
1003 return BLK_STS_OK;
1007 * Function: scsi_init_io()
1009 * Purpose: SCSI I/O initialize function.
1011 * Arguments: cmd - Command descriptor we wish to initialize
1013 * Returns: BLK_STS_OK on success
1014 * BLK_STS_RESOURCE if the failure is retryable
1015 * BLK_STS_IOERR if the failure is fatal
1017 blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
1019 struct request *rq = cmd->request;
1020 blk_status_t ret;
1022 if (WARN_ON_ONCE(!blk_rq_nr_phys_segments(rq)))
1023 return BLK_STS_IOERR;
1025 ret = scsi_init_sgtable(rq, &cmd->sdb);
1026 if (ret)
1027 return ret;
1029 if (blk_integrity_rq(rq)) {
1030 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1031 int ivecs, count;
1033 if (WARN_ON_ONCE(!prot_sdb)) {
1035 * This can happen if someone (e.g. multipath)
1036 * queues a command to a device on an adapter
1037 * that does not support DIX.
1039 ret = BLK_STS_IOERR;
1040 goto out_free_sgtables;
1043 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1045 if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1046 prot_sdb->table.sgl,
1047 SCSI_INLINE_PROT_SG_CNT)) {
1048 ret = BLK_STS_RESOURCE;
1049 goto out_free_sgtables;
1052 count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1053 prot_sdb->table.sgl);
1054 BUG_ON(count > ivecs);
1055 BUG_ON(count > queue_max_integrity_segments(rq->q));
1057 cmd->prot_sdb = prot_sdb;
1058 cmd->prot_sdb->table.nents = count;
1061 return BLK_STS_OK;
1062 out_free_sgtables:
1063 scsi_mq_free_sgtables(cmd);
1064 return ret;
1066 EXPORT_SYMBOL(scsi_init_io);
1069 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1070 * @rq: Request associated with the SCSI command to be initialized.
1072 * This function initializes the members of struct scsi_cmnd that must be
1073 * initialized before request processing starts and that won't be
1074 * reinitialized if a SCSI command is requeued.
1076 * Called from inside blk_get_request() for pass-through requests and from
1077 * inside scsi_init_command() for filesystem requests.
1079 static void scsi_initialize_rq(struct request *rq)
1081 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1083 scsi_req_init(&cmd->req);
1084 init_rcu_head(&cmd->rcu);
1085 cmd->jiffies_at_alloc = jiffies;
1086 cmd->retries = 0;
1090 * Only called when the request isn't completed by SCSI, and not freed by
1091 * SCSI
1093 static void scsi_cleanup_rq(struct request *rq)
1095 if (rq->rq_flags & RQF_DONTPREP) {
1096 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1097 rq->rq_flags &= ~RQF_DONTPREP;
1101 /* Add a command to the list used by the aacraid and dpt_i2o drivers */
1102 void scsi_add_cmd_to_list(struct scsi_cmnd *cmd)
1104 struct scsi_device *sdev = cmd->device;
1105 struct Scsi_Host *shost = sdev->host;
1106 unsigned long flags;
1108 if (shost->use_cmd_list) {
1109 spin_lock_irqsave(&sdev->list_lock, flags);
1110 list_add_tail(&cmd->list, &sdev->cmd_list);
1111 spin_unlock_irqrestore(&sdev->list_lock, flags);
1115 /* Remove a command from the list used by the aacraid and dpt_i2o drivers */
1116 void scsi_del_cmd_from_list(struct scsi_cmnd *cmd)
1118 struct scsi_device *sdev = cmd->device;
1119 struct Scsi_Host *shost = sdev->host;
1120 unsigned long flags;
1122 if (shost->use_cmd_list) {
1123 spin_lock_irqsave(&sdev->list_lock, flags);
1124 BUG_ON(list_empty(&cmd->list));
1125 list_del_init(&cmd->list);
1126 spin_unlock_irqrestore(&sdev->list_lock, flags);
1130 /* Called after a request has been started. */
1131 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1133 void *buf = cmd->sense_buffer;
1134 void *prot = cmd->prot_sdb;
1135 struct request *rq = blk_mq_rq_from_pdu(cmd);
1136 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1137 unsigned long jiffies_at_alloc;
1138 int retries;
1139 bool in_flight;
1141 if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1142 flags |= SCMD_INITIALIZED;
1143 scsi_initialize_rq(rq);
1146 jiffies_at_alloc = cmd->jiffies_at_alloc;
1147 retries = cmd->retries;
1148 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1149 /* zero out the cmd, except for the embedded scsi_request */
1150 memset((char *)cmd + sizeof(cmd->req), 0,
1151 sizeof(*cmd) - sizeof(cmd->req) + dev->host->hostt->cmd_size);
1153 cmd->device = dev;
1154 cmd->sense_buffer = buf;
1155 cmd->prot_sdb = prot;
1156 cmd->flags = flags;
1157 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1158 cmd->jiffies_at_alloc = jiffies_at_alloc;
1159 cmd->retries = retries;
1160 if (in_flight)
1161 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1163 scsi_add_cmd_to_list(cmd);
1166 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1167 struct request *req)
1169 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1172 * Passthrough requests may transfer data, in which case they must
1173 * a bio attached to them. Or they might contain a SCSI command
1174 * that does not transfer data, in which case they may optionally
1175 * submit a request without an attached bio.
1177 if (req->bio) {
1178 blk_status_t ret = scsi_init_io(cmd);
1179 if (unlikely(ret != BLK_STS_OK))
1180 return ret;
1181 } else {
1182 BUG_ON(blk_rq_bytes(req));
1184 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1187 cmd->cmd_len = scsi_req(req)->cmd_len;
1188 cmd->cmnd = scsi_req(req)->cmd;
1189 cmd->transfersize = blk_rq_bytes(req);
1190 cmd->allowed = scsi_req(req)->retries;
1191 return BLK_STS_OK;
1195 * Setup a normal block command. These are simple request from filesystems
1196 * that still need to be translated to SCSI CDBs from the ULD.
1198 static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1199 struct request *req)
1201 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1203 if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1204 blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1205 if (ret != BLK_STS_OK)
1206 return ret;
1209 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1210 memset(cmd->cmnd, 0, BLK_MAX_CDB);
1211 return scsi_cmd_to_driver(cmd)->init_command(cmd);
1214 static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1215 struct request *req)
1217 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1219 if (!blk_rq_bytes(req))
1220 cmd->sc_data_direction = DMA_NONE;
1221 else if (rq_data_dir(req) == WRITE)
1222 cmd->sc_data_direction = DMA_TO_DEVICE;
1223 else
1224 cmd->sc_data_direction = DMA_FROM_DEVICE;
1226 if (blk_rq_is_scsi(req))
1227 return scsi_setup_scsi_cmnd(sdev, req);
1228 else
1229 return scsi_setup_fs_cmnd(sdev, req);
1232 static blk_status_t
1233 scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1235 switch (sdev->sdev_state) {
1236 case SDEV_OFFLINE:
1237 case SDEV_TRANSPORT_OFFLINE:
1239 * If the device is offline we refuse to process any
1240 * commands. The device must be brought online
1241 * before trying any recovery commands.
1243 sdev_printk(KERN_ERR, sdev,
1244 "rejecting I/O to offline device\n");
1245 return BLK_STS_IOERR;
1246 case SDEV_DEL:
1248 * If the device is fully deleted, we refuse to
1249 * process any commands as well.
1251 sdev_printk(KERN_ERR, sdev,
1252 "rejecting I/O to dead device\n");
1253 return BLK_STS_IOERR;
1254 case SDEV_BLOCK:
1255 case SDEV_CREATED_BLOCK:
1256 return BLK_STS_RESOURCE;
1257 case SDEV_QUIESCE:
1259 * If the devices is blocked we defer normal commands.
1261 if (req && !(req->rq_flags & RQF_PREEMPT))
1262 return BLK_STS_RESOURCE;
1263 return BLK_STS_OK;
1264 default:
1266 * For any other not fully online state we only allow
1267 * special commands. In particular any user initiated
1268 * command is not allowed.
1270 if (req && !(req->rq_flags & RQF_PREEMPT))
1271 return BLK_STS_IOERR;
1272 return BLK_STS_OK;
1277 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1278 * return 0.
1280 * Called with the queue_lock held.
1282 static inline int scsi_dev_queue_ready(struct request_queue *q,
1283 struct scsi_device *sdev)
1285 unsigned int busy;
1287 busy = atomic_inc_return(&sdev->device_busy) - 1;
1288 if (atomic_read(&sdev->device_blocked)) {
1289 if (busy)
1290 goto out_dec;
1293 * unblock after device_blocked iterates to zero
1295 if (atomic_dec_return(&sdev->device_blocked) > 0)
1296 goto out_dec;
1297 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1298 "unblocking device at zero depth\n"));
1301 if (busy >= sdev->queue_depth)
1302 goto out_dec;
1304 return 1;
1305 out_dec:
1306 atomic_dec(&sdev->device_busy);
1307 return 0;
1311 * scsi_target_queue_ready: checks if there we can send commands to target
1312 * @sdev: scsi device on starget to check.
1314 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1315 struct scsi_device *sdev)
1317 struct scsi_target *starget = scsi_target(sdev);
1318 unsigned int busy;
1320 if (starget->single_lun) {
1321 spin_lock_irq(shost->host_lock);
1322 if (starget->starget_sdev_user &&
1323 starget->starget_sdev_user != sdev) {
1324 spin_unlock_irq(shost->host_lock);
1325 return 0;
1327 starget->starget_sdev_user = sdev;
1328 spin_unlock_irq(shost->host_lock);
1331 if (starget->can_queue <= 0)
1332 return 1;
1334 busy = atomic_inc_return(&starget->target_busy) - 1;
1335 if (atomic_read(&starget->target_blocked) > 0) {
1336 if (busy)
1337 goto starved;
1340 * unblock after target_blocked iterates to zero
1342 if (atomic_dec_return(&starget->target_blocked) > 0)
1343 goto out_dec;
1345 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1346 "unblocking target at zero depth\n"));
1349 if (busy >= starget->can_queue)
1350 goto starved;
1352 return 1;
1354 starved:
1355 spin_lock_irq(shost->host_lock);
1356 list_move_tail(&sdev->starved_entry, &shost->starved_list);
1357 spin_unlock_irq(shost->host_lock);
1358 out_dec:
1359 if (starget->can_queue > 0)
1360 atomic_dec(&starget->target_busy);
1361 return 0;
1365 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1366 * return 0. We must end up running the queue again whenever 0 is
1367 * returned, else IO can hang.
1369 static inline int scsi_host_queue_ready(struct request_queue *q,
1370 struct Scsi_Host *shost,
1371 struct scsi_device *sdev,
1372 struct scsi_cmnd *cmd)
1374 if (scsi_host_in_recovery(shost))
1375 return 0;
1377 if (atomic_read(&shost->host_blocked) > 0) {
1378 if (scsi_host_busy(shost) > 0)
1379 goto starved;
1382 * unblock after host_blocked iterates to zero
1384 if (atomic_dec_return(&shost->host_blocked) > 0)
1385 goto out_dec;
1387 SCSI_LOG_MLQUEUE(3,
1388 shost_printk(KERN_INFO, shost,
1389 "unblocking host at zero depth\n"));
1392 if (shost->host_self_blocked)
1393 goto starved;
1395 /* We're OK to process the command, so we can't be starved */
1396 if (!list_empty(&sdev->starved_entry)) {
1397 spin_lock_irq(shost->host_lock);
1398 if (!list_empty(&sdev->starved_entry))
1399 list_del_init(&sdev->starved_entry);
1400 spin_unlock_irq(shost->host_lock);
1403 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1405 return 1;
1407 starved:
1408 spin_lock_irq(shost->host_lock);
1409 if (list_empty(&sdev->starved_entry))
1410 list_add_tail(&sdev->starved_entry, &shost->starved_list);
1411 spin_unlock_irq(shost->host_lock);
1412 out_dec:
1413 scsi_dec_host_busy(shost, cmd);
1414 return 0;
1418 * Busy state exporting function for request stacking drivers.
1420 * For efficiency, no lock is taken to check the busy state of
1421 * shost/starget/sdev, since the returned value is not guaranteed and
1422 * may be changed after request stacking drivers call the function,
1423 * regardless of taking lock or not.
1425 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1426 * needs to return 'not busy'. Otherwise, request stacking drivers
1427 * may hold requests forever.
1429 static bool scsi_mq_lld_busy(struct request_queue *q)
1431 struct scsi_device *sdev = q->queuedata;
1432 struct Scsi_Host *shost;
1434 if (blk_queue_dying(q))
1435 return false;
1437 shost = sdev->host;
1440 * Ignore host/starget busy state.
1441 * Since block layer does not have a concept of fairness across
1442 * multiple queues, congestion of host/starget needs to be handled
1443 * in SCSI layer.
1445 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1446 return true;
1448 return false;
1451 static void scsi_softirq_done(struct request *rq)
1453 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1454 unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1455 int disposition;
1457 INIT_LIST_HEAD(&cmd->eh_entry);
1459 atomic_inc(&cmd->device->iodone_cnt);
1460 if (cmd->result)
1461 atomic_inc(&cmd->device->ioerr_cnt);
1463 disposition = scsi_decide_disposition(cmd);
1464 if (disposition != SUCCESS &&
1465 time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1466 scmd_printk(KERN_ERR, cmd,
1467 "timing out command, waited %lus\n",
1468 wait_for/HZ);
1469 disposition = SUCCESS;
1472 scsi_log_completion(cmd, disposition);
1474 switch (disposition) {
1475 case SUCCESS:
1476 scsi_finish_command(cmd);
1477 break;
1478 case NEEDS_RETRY:
1479 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1480 break;
1481 case ADD_TO_MLQUEUE:
1482 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1483 break;
1484 default:
1485 scsi_eh_scmd_add(cmd);
1486 break;
1491 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1492 * @cmd: command block we are dispatching.
1494 * Return: nonzero return request was rejected and device's queue needs to be
1495 * plugged.
1497 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1499 struct Scsi_Host *host = cmd->device->host;
1500 int rtn = 0;
1502 atomic_inc(&cmd->device->iorequest_cnt);
1504 /* check if the device is still usable */
1505 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1506 /* in SDEV_DEL we error all commands. DID_NO_CONNECT
1507 * returns an immediate error upwards, and signals
1508 * that the device is no longer present */
1509 cmd->result = DID_NO_CONNECT << 16;
1510 goto done;
1513 /* Check to see if the scsi lld made this device blocked. */
1514 if (unlikely(scsi_device_blocked(cmd->device))) {
1516 * in blocked state, the command is just put back on
1517 * the device queue. The suspend state has already
1518 * blocked the queue so future requests should not
1519 * occur until the device transitions out of the
1520 * suspend state.
1522 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1523 "queuecommand : device blocked\n"));
1524 return SCSI_MLQUEUE_DEVICE_BUSY;
1527 /* Store the LUN value in cmnd, if needed. */
1528 if (cmd->device->lun_in_cdb)
1529 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1530 (cmd->device->lun << 5 & 0xe0);
1532 scsi_log_send(cmd);
1535 * Before we queue this command, check if the command
1536 * length exceeds what the host adapter can handle.
1538 if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1539 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1540 "queuecommand : command too long. "
1541 "cdb_size=%d host->max_cmd_len=%d\n",
1542 cmd->cmd_len, cmd->device->host->max_cmd_len));
1543 cmd->result = (DID_ABORT << 16);
1544 goto done;
1547 if (unlikely(host->shost_state == SHOST_DEL)) {
1548 cmd->result = (DID_NO_CONNECT << 16);
1549 goto done;
1553 trace_scsi_dispatch_cmd_start(cmd);
1554 rtn = host->hostt->queuecommand(host, cmd);
1555 if (rtn) {
1556 trace_scsi_dispatch_cmd_error(cmd, rtn);
1557 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1558 rtn != SCSI_MLQUEUE_TARGET_BUSY)
1559 rtn = SCSI_MLQUEUE_HOST_BUSY;
1561 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1562 "queuecommand : request rejected\n"));
1565 return rtn;
1566 done:
1567 cmd->scsi_done(cmd);
1568 return 0;
1571 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1572 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1574 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1575 sizeof(struct scatterlist);
1578 static blk_status_t scsi_mq_prep_fn(struct request *req)
1580 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1581 struct scsi_device *sdev = req->q->queuedata;
1582 struct Scsi_Host *shost = sdev->host;
1583 struct scatterlist *sg;
1585 scsi_init_command(sdev, cmd);
1587 cmd->request = req;
1588 cmd->tag = req->tag;
1589 cmd->prot_op = SCSI_PROT_NORMAL;
1591 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1592 cmd->sdb.table.sgl = sg;
1594 if (scsi_host_get_prot(shost)) {
1595 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1597 cmd->prot_sdb->table.sgl =
1598 (struct scatterlist *)(cmd->prot_sdb + 1);
1601 blk_mq_start_request(req);
1603 return scsi_setup_cmnd(sdev, req);
1606 static void scsi_mq_done(struct scsi_cmnd *cmd)
1608 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1609 return;
1610 trace_scsi_dispatch_cmd_done(cmd);
1613 * If the block layer didn't complete the request due to a timeout
1614 * injection, scsi must clear its internal completed state so that the
1615 * timeout handler will see it needs to escalate its own error
1616 * recovery.
1618 if (unlikely(!blk_mq_complete_request(cmd->request)))
1619 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1622 static void scsi_mq_put_budget(struct blk_mq_hw_ctx *hctx)
1624 struct request_queue *q = hctx->queue;
1625 struct scsi_device *sdev = q->queuedata;
1627 atomic_dec(&sdev->device_busy);
1630 static bool scsi_mq_get_budget(struct blk_mq_hw_ctx *hctx)
1632 struct request_queue *q = hctx->queue;
1633 struct scsi_device *sdev = q->queuedata;
1635 if (scsi_dev_queue_ready(q, sdev))
1636 return true;
1638 if (atomic_read(&sdev->device_busy) == 0 && !scsi_device_blocked(sdev))
1639 blk_mq_delay_run_hw_queue(hctx, SCSI_QUEUE_DELAY);
1640 return false;
1643 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1644 const struct blk_mq_queue_data *bd)
1646 struct request *req = bd->rq;
1647 struct request_queue *q = req->q;
1648 struct scsi_device *sdev = q->queuedata;
1649 struct Scsi_Host *shost = sdev->host;
1650 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1651 blk_status_t ret;
1652 int reason;
1655 * If the device is not in running state we will reject some or all
1656 * commands.
1658 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1659 ret = scsi_prep_state_check(sdev, req);
1660 if (ret != BLK_STS_OK)
1661 goto out_put_budget;
1664 ret = BLK_STS_RESOURCE;
1665 if (!scsi_target_queue_ready(shost, sdev))
1666 goto out_put_budget;
1667 if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1668 goto out_dec_target_busy;
1670 if (!(req->rq_flags & RQF_DONTPREP)) {
1671 ret = scsi_mq_prep_fn(req);
1672 if (ret != BLK_STS_OK)
1673 goto out_dec_host_busy;
1674 req->rq_flags |= RQF_DONTPREP;
1675 } else {
1676 clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1677 blk_mq_start_request(req);
1680 cmd->flags &= SCMD_PRESERVED_FLAGS;
1681 if (sdev->simple_tags)
1682 cmd->flags |= SCMD_TAGGED;
1683 if (bd->last)
1684 cmd->flags |= SCMD_LAST;
1686 scsi_init_cmd_errh(cmd);
1687 cmd->scsi_done = scsi_mq_done;
1689 reason = scsi_dispatch_cmd(cmd);
1690 if (reason) {
1691 scsi_set_blocked(cmd, reason);
1692 ret = BLK_STS_RESOURCE;
1693 goto out_dec_host_busy;
1696 return BLK_STS_OK;
1698 out_dec_host_busy:
1699 scsi_dec_host_busy(shost, cmd);
1700 out_dec_target_busy:
1701 if (scsi_target(sdev)->can_queue > 0)
1702 atomic_dec(&scsi_target(sdev)->target_busy);
1703 out_put_budget:
1704 scsi_mq_put_budget(hctx);
1705 switch (ret) {
1706 case BLK_STS_OK:
1707 break;
1708 case BLK_STS_RESOURCE:
1709 if (atomic_read(&sdev->device_busy) ||
1710 scsi_device_blocked(sdev))
1711 ret = BLK_STS_DEV_RESOURCE;
1712 break;
1713 default:
1714 if (unlikely(!scsi_device_online(sdev)))
1715 scsi_req(req)->result = DID_NO_CONNECT << 16;
1716 else
1717 scsi_req(req)->result = DID_ERROR << 16;
1719 * Make sure to release all allocated resources when
1720 * we hit an error, as we will never see this command
1721 * again.
1723 if (req->rq_flags & RQF_DONTPREP)
1724 scsi_mq_uninit_cmd(cmd);
1725 break;
1727 return ret;
1730 static enum blk_eh_timer_return scsi_timeout(struct request *req,
1731 bool reserved)
1733 if (reserved)
1734 return BLK_EH_RESET_TIMER;
1735 return scsi_times_out(req);
1738 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1739 unsigned int hctx_idx, unsigned int numa_node)
1741 struct Scsi_Host *shost = set->driver_data;
1742 const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1743 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1744 struct scatterlist *sg;
1746 if (unchecked_isa_dma)
1747 cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1748 cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1749 GFP_KERNEL, numa_node);
1750 if (!cmd->sense_buffer)
1751 return -ENOMEM;
1752 cmd->req.sense = cmd->sense_buffer;
1754 if (scsi_host_get_prot(shost)) {
1755 sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1756 shost->hostt->cmd_size;
1757 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1760 return 0;
1763 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1764 unsigned int hctx_idx)
1766 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1768 scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1769 cmd->sense_buffer);
1772 static int scsi_map_queues(struct blk_mq_tag_set *set)
1774 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1776 if (shost->hostt->map_queues)
1777 return shost->hostt->map_queues(shost);
1778 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1781 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1783 struct device *dev = shost->dma_dev;
1786 * this limit is imposed by hardware restrictions
1788 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1789 SG_MAX_SEGMENTS));
1791 if (scsi_host_prot_dma(shost)) {
1792 shost->sg_prot_tablesize =
1793 min_not_zero(shost->sg_prot_tablesize,
1794 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1795 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1796 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1799 if (dev->dma_mask) {
1800 shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1801 dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1803 blk_queue_max_hw_sectors(q, shost->max_sectors);
1804 if (shost->unchecked_isa_dma)
1805 blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1806 blk_queue_segment_boundary(q, shost->dma_boundary);
1807 dma_set_seg_boundary(dev, shost->dma_boundary);
1809 blk_queue_max_segment_size(q, shost->max_segment_size);
1810 blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1811 dma_set_max_seg_size(dev, queue_max_segment_size(q));
1814 * Set a reasonable default alignment: The larger of 32-byte (dword),
1815 * which is a common minimum for HBAs, and the minimum DMA alignment,
1816 * which is set by the platform.
1818 * Devices that require a bigger alignment can increase it later.
1820 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1822 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1824 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1825 .get_budget = scsi_mq_get_budget,
1826 .put_budget = scsi_mq_put_budget,
1827 .queue_rq = scsi_queue_rq,
1828 .complete = scsi_softirq_done,
1829 .timeout = scsi_timeout,
1830 #ifdef CONFIG_BLK_DEBUG_FS
1831 .show_rq = scsi_show_rq,
1832 #endif
1833 .init_request = scsi_mq_init_request,
1834 .exit_request = scsi_mq_exit_request,
1835 .initialize_rq_fn = scsi_initialize_rq,
1836 .cleanup_rq = scsi_cleanup_rq,
1837 .busy = scsi_mq_lld_busy,
1838 .map_queues = scsi_map_queues,
1842 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1844 struct request_queue *q = hctx->queue;
1845 struct scsi_device *sdev = q->queuedata;
1846 struct Scsi_Host *shost = sdev->host;
1848 shost->hostt->commit_rqs(shost, hctx->queue_num);
1851 static const struct blk_mq_ops scsi_mq_ops = {
1852 .get_budget = scsi_mq_get_budget,
1853 .put_budget = scsi_mq_put_budget,
1854 .queue_rq = scsi_queue_rq,
1855 .commit_rqs = scsi_commit_rqs,
1856 .complete = scsi_softirq_done,
1857 .timeout = scsi_timeout,
1858 #ifdef CONFIG_BLK_DEBUG_FS
1859 .show_rq = scsi_show_rq,
1860 #endif
1861 .init_request = scsi_mq_init_request,
1862 .exit_request = scsi_mq_exit_request,
1863 .initialize_rq_fn = scsi_initialize_rq,
1864 .cleanup_rq = scsi_cleanup_rq,
1865 .busy = scsi_mq_lld_busy,
1866 .map_queues = scsi_map_queues,
1869 struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1871 sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1872 if (IS_ERR(sdev->request_queue))
1873 return NULL;
1875 sdev->request_queue->queuedata = sdev;
1876 __scsi_init_queue(sdev->host, sdev->request_queue);
1877 blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1878 return sdev->request_queue;
1881 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1883 unsigned int cmd_size, sgl_size;
1885 sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1886 scsi_mq_inline_sgl_size(shost));
1887 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1888 if (scsi_host_get_prot(shost))
1889 cmd_size += sizeof(struct scsi_data_buffer) +
1890 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1892 memset(&shost->tag_set, 0, sizeof(shost->tag_set));
1893 if (shost->hostt->commit_rqs)
1894 shost->tag_set.ops = &scsi_mq_ops;
1895 else
1896 shost->tag_set.ops = &scsi_mq_ops_no_commit;
1897 shost->tag_set.nr_hw_queues = shost->nr_hw_queues ? : 1;
1898 shost->tag_set.queue_depth = shost->can_queue;
1899 shost->tag_set.cmd_size = cmd_size;
1900 shost->tag_set.numa_node = NUMA_NO_NODE;
1901 shost->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
1902 shost->tag_set.flags |=
1903 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1904 shost->tag_set.driver_data = shost;
1906 return blk_mq_alloc_tag_set(&shost->tag_set);
1909 void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1911 blk_mq_free_tag_set(&shost->tag_set);
1915 * scsi_device_from_queue - return sdev associated with a request_queue
1916 * @q: The request queue to return the sdev from
1918 * Return the sdev associated with a request queue or NULL if the
1919 * request_queue does not reference a SCSI device.
1921 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1923 struct scsi_device *sdev = NULL;
1925 if (q->mq_ops == &scsi_mq_ops_no_commit ||
1926 q->mq_ops == &scsi_mq_ops)
1927 sdev = q->queuedata;
1928 if (!sdev || !get_device(&sdev->sdev_gendev))
1929 sdev = NULL;
1931 return sdev;
1933 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
1936 * Function: scsi_block_requests()
1938 * Purpose: Utility function used by low-level drivers to prevent further
1939 * commands from being queued to the device.
1941 * Arguments: shost - Host in question
1943 * Returns: Nothing
1945 * Lock status: No locks are assumed held.
1947 * Notes: There is no timer nor any other means by which the requests
1948 * get unblocked other than the low-level driver calling
1949 * scsi_unblock_requests().
1951 void scsi_block_requests(struct Scsi_Host *shost)
1953 shost->host_self_blocked = 1;
1955 EXPORT_SYMBOL(scsi_block_requests);
1958 * Function: scsi_unblock_requests()
1960 * Purpose: Utility function used by low-level drivers to allow further
1961 * commands from being queued to the device.
1963 * Arguments: shost - Host in question
1965 * Returns: Nothing
1967 * Lock status: No locks are assumed held.
1969 * Notes: There is no timer nor any other means by which the requests
1970 * get unblocked other than the low-level driver calling
1971 * scsi_unblock_requests().
1973 * This is done as an API function so that changes to the
1974 * internals of the scsi mid-layer won't require wholesale
1975 * changes to drivers that use this feature.
1977 void scsi_unblock_requests(struct Scsi_Host *shost)
1979 shost->host_self_blocked = 0;
1980 scsi_run_host_queues(shost);
1982 EXPORT_SYMBOL(scsi_unblock_requests);
1984 int __init scsi_init_queue(void)
1986 scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1987 sizeof(struct scsi_data_buffer),
1988 0, 0, NULL);
1989 if (!scsi_sdb_cache) {
1990 printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1991 return -ENOMEM;
1994 return 0;
1997 void scsi_exit_queue(void)
1999 kmem_cache_destroy(scsi_sense_cache);
2000 kmem_cache_destroy(scsi_sense_isadma_cache);
2001 kmem_cache_destroy(scsi_sdb_cache);
2005 * scsi_mode_select - issue a mode select
2006 * @sdev: SCSI device to be queried
2007 * @pf: Page format bit (1 == standard, 0 == vendor specific)
2008 * @sp: Save page bit (0 == don't save, 1 == save)
2009 * @modepage: mode page being requested
2010 * @buffer: request buffer (may not be smaller than eight bytes)
2011 * @len: length of request buffer.
2012 * @timeout: command timeout
2013 * @retries: number of retries before failing
2014 * @data: returns a structure abstracting the mode header data
2015 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2016 * must be SCSI_SENSE_BUFFERSIZE big.
2018 * Returns zero if successful; negative error number or scsi
2019 * status on error
2023 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
2024 unsigned char *buffer, int len, int timeout, int retries,
2025 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2027 unsigned char cmd[10];
2028 unsigned char *real_buffer;
2029 int ret;
2031 memset(cmd, 0, sizeof(cmd));
2032 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2034 if (sdev->use_10_for_ms) {
2035 if (len > 65535)
2036 return -EINVAL;
2037 real_buffer = kmalloc(8 + len, GFP_KERNEL);
2038 if (!real_buffer)
2039 return -ENOMEM;
2040 memcpy(real_buffer + 8, buffer, len);
2041 len += 8;
2042 real_buffer[0] = 0;
2043 real_buffer[1] = 0;
2044 real_buffer[2] = data->medium_type;
2045 real_buffer[3] = data->device_specific;
2046 real_buffer[4] = data->longlba ? 0x01 : 0;
2047 real_buffer[5] = 0;
2048 real_buffer[6] = data->block_descriptor_length >> 8;
2049 real_buffer[7] = data->block_descriptor_length;
2051 cmd[0] = MODE_SELECT_10;
2052 cmd[7] = len >> 8;
2053 cmd[8] = len;
2054 } else {
2055 if (len > 255 || data->block_descriptor_length > 255 ||
2056 data->longlba)
2057 return -EINVAL;
2059 real_buffer = kmalloc(4 + len, GFP_KERNEL);
2060 if (!real_buffer)
2061 return -ENOMEM;
2062 memcpy(real_buffer + 4, buffer, len);
2063 len += 4;
2064 real_buffer[0] = 0;
2065 real_buffer[1] = data->medium_type;
2066 real_buffer[2] = data->device_specific;
2067 real_buffer[3] = data->block_descriptor_length;
2070 cmd[0] = MODE_SELECT;
2071 cmd[4] = len;
2074 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2075 sshdr, timeout, retries, NULL);
2076 kfree(real_buffer);
2077 return ret;
2079 EXPORT_SYMBOL_GPL(scsi_mode_select);
2082 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2083 * @sdev: SCSI device to be queried
2084 * @dbd: set if mode sense will allow block descriptors to be returned
2085 * @modepage: mode page being requested
2086 * @buffer: request buffer (may not be smaller than eight bytes)
2087 * @len: length of request buffer.
2088 * @timeout: command timeout
2089 * @retries: number of retries before failing
2090 * @data: returns a structure abstracting the mode header data
2091 * @sshdr: place to put sense data (or NULL if no sense to be collected).
2092 * must be SCSI_SENSE_BUFFERSIZE big.
2094 * Returns zero if unsuccessful, or the header offset (either 4
2095 * or 8 depending on whether a six or ten byte command was
2096 * issued) if successful.
2099 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2100 unsigned char *buffer, int len, int timeout, int retries,
2101 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2103 unsigned char cmd[12];
2104 int use_10_for_ms;
2105 int header_length;
2106 int result, retry_count = retries;
2107 struct scsi_sense_hdr my_sshdr;
2109 memset(data, 0, sizeof(*data));
2110 memset(&cmd[0], 0, 12);
2111 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */
2112 cmd[2] = modepage;
2114 /* caller might not be interested in sense, but we need it */
2115 if (!sshdr)
2116 sshdr = &my_sshdr;
2118 retry:
2119 use_10_for_ms = sdev->use_10_for_ms;
2121 if (use_10_for_ms) {
2122 if (len < 8)
2123 len = 8;
2125 cmd[0] = MODE_SENSE_10;
2126 cmd[8] = len;
2127 header_length = 8;
2128 } else {
2129 if (len < 4)
2130 len = 4;
2132 cmd[0] = MODE_SENSE;
2133 cmd[4] = len;
2134 header_length = 4;
2137 memset(buffer, 0, len);
2139 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2140 sshdr, timeout, retries, NULL);
2142 /* This code looks awful: what it's doing is making sure an
2143 * ILLEGAL REQUEST sense return identifies the actual command
2144 * byte as the problem. MODE_SENSE commands can return
2145 * ILLEGAL REQUEST if the code page isn't supported */
2147 if (use_10_for_ms && !scsi_status_is_good(result) &&
2148 driver_byte(result) == DRIVER_SENSE) {
2149 if (scsi_sense_valid(sshdr)) {
2150 if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2151 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2153 * Invalid command operation code
2155 sdev->use_10_for_ms = 0;
2156 goto retry;
2161 if(scsi_status_is_good(result)) {
2162 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2163 (modepage == 6 || modepage == 8))) {
2164 /* Initio breakage? */
2165 header_length = 0;
2166 data->length = 13;
2167 data->medium_type = 0;
2168 data->device_specific = 0;
2169 data->longlba = 0;
2170 data->block_descriptor_length = 0;
2171 } else if(use_10_for_ms) {
2172 data->length = buffer[0]*256 + buffer[1] + 2;
2173 data->medium_type = buffer[2];
2174 data->device_specific = buffer[3];
2175 data->longlba = buffer[4] & 0x01;
2176 data->block_descriptor_length = buffer[6]*256
2177 + buffer[7];
2178 } else {
2179 data->length = buffer[0] + 1;
2180 data->medium_type = buffer[1];
2181 data->device_specific = buffer[2];
2182 data->block_descriptor_length = buffer[3];
2184 data->header_length = header_length;
2185 } else if ((status_byte(result) == CHECK_CONDITION) &&
2186 scsi_sense_valid(sshdr) &&
2187 sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2188 retry_count--;
2189 goto retry;
2192 return result;
2194 EXPORT_SYMBOL(scsi_mode_sense);
2197 * scsi_test_unit_ready - test if unit is ready
2198 * @sdev: scsi device to change the state of.
2199 * @timeout: command timeout
2200 * @retries: number of retries before failing
2201 * @sshdr: outpout pointer for decoded sense information.
2203 * Returns zero if unsuccessful or an error if TUR failed. For
2204 * removable media, UNIT_ATTENTION sets ->changed flag.
2207 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2208 struct scsi_sense_hdr *sshdr)
2210 char cmd[] = {
2211 TEST_UNIT_READY, 0, 0, 0, 0, 0,
2213 int result;
2215 /* try to eat the UNIT_ATTENTION if there are enough retries */
2216 do {
2217 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2218 timeout, 1, NULL);
2219 if (sdev->removable && scsi_sense_valid(sshdr) &&
2220 sshdr->sense_key == UNIT_ATTENTION)
2221 sdev->changed = 1;
2222 } while (scsi_sense_valid(sshdr) &&
2223 sshdr->sense_key == UNIT_ATTENTION && --retries);
2225 return result;
2227 EXPORT_SYMBOL(scsi_test_unit_ready);
2230 * scsi_device_set_state - Take the given device through the device state model.
2231 * @sdev: scsi device to change the state of.
2232 * @state: state to change to.
2234 * Returns zero if successful or an error if the requested
2235 * transition is illegal.
2238 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2240 enum scsi_device_state oldstate = sdev->sdev_state;
2242 if (state == oldstate)
2243 return 0;
2245 switch (state) {
2246 case SDEV_CREATED:
2247 switch (oldstate) {
2248 case SDEV_CREATED_BLOCK:
2249 break;
2250 default:
2251 goto illegal;
2253 break;
2255 case SDEV_RUNNING:
2256 switch (oldstate) {
2257 case SDEV_CREATED:
2258 case SDEV_OFFLINE:
2259 case SDEV_TRANSPORT_OFFLINE:
2260 case SDEV_QUIESCE:
2261 case SDEV_BLOCK:
2262 break;
2263 default:
2264 goto illegal;
2266 break;
2268 case SDEV_QUIESCE:
2269 switch (oldstate) {
2270 case SDEV_RUNNING:
2271 case SDEV_OFFLINE:
2272 case SDEV_TRANSPORT_OFFLINE:
2273 break;
2274 default:
2275 goto illegal;
2277 break;
2279 case SDEV_OFFLINE:
2280 case SDEV_TRANSPORT_OFFLINE:
2281 switch (oldstate) {
2282 case SDEV_CREATED:
2283 case SDEV_RUNNING:
2284 case SDEV_QUIESCE:
2285 case SDEV_BLOCK:
2286 break;
2287 default:
2288 goto illegal;
2290 break;
2292 case SDEV_BLOCK:
2293 switch (oldstate) {
2294 case SDEV_RUNNING:
2295 case SDEV_CREATED_BLOCK:
2296 case SDEV_OFFLINE:
2297 break;
2298 default:
2299 goto illegal;
2301 break;
2303 case SDEV_CREATED_BLOCK:
2304 switch (oldstate) {
2305 case SDEV_CREATED:
2306 break;
2307 default:
2308 goto illegal;
2310 break;
2312 case SDEV_CANCEL:
2313 switch (oldstate) {
2314 case SDEV_CREATED:
2315 case SDEV_RUNNING:
2316 case SDEV_QUIESCE:
2317 case SDEV_OFFLINE:
2318 case SDEV_TRANSPORT_OFFLINE:
2319 break;
2320 default:
2321 goto illegal;
2323 break;
2325 case SDEV_DEL:
2326 switch (oldstate) {
2327 case SDEV_CREATED:
2328 case SDEV_RUNNING:
2329 case SDEV_OFFLINE:
2330 case SDEV_TRANSPORT_OFFLINE:
2331 case SDEV_CANCEL:
2332 case SDEV_BLOCK:
2333 case SDEV_CREATED_BLOCK:
2334 break;
2335 default:
2336 goto illegal;
2338 break;
2341 sdev->sdev_state = state;
2342 return 0;
2344 illegal:
2345 SCSI_LOG_ERROR_RECOVERY(1,
2346 sdev_printk(KERN_ERR, sdev,
2347 "Illegal state transition %s->%s",
2348 scsi_device_state_name(oldstate),
2349 scsi_device_state_name(state))
2351 return -EINVAL;
2353 EXPORT_SYMBOL(scsi_device_set_state);
2356 * sdev_evt_emit - emit a single SCSI device uevent
2357 * @sdev: associated SCSI device
2358 * @evt: event to emit
2360 * Send a single uevent (scsi_event) to the associated scsi_device.
2362 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2364 int idx = 0;
2365 char *envp[3];
2367 switch (evt->evt_type) {
2368 case SDEV_EVT_MEDIA_CHANGE:
2369 envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2370 break;
2371 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2372 scsi_rescan_device(&sdev->sdev_gendev);
2373 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2374 break;
2375 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2376 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2377 break;
2378 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2379 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2380 break;
2381 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2382 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2383 break;
2384 case SDEV_EVT_LUN_CHANGE_REPORTED:
2385 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2386 break;
2387 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2388 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2389 break;
2390 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2391 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2392 break;
2393 default:
2394 /* do nothing */
2395 break;
2398 envp[idx++] = NULL;
2400 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2404 * sdev_evt_thread - send a uevent for each scsi event
2405 * @work: work struct for scsi_device
2407 * Dispatch queued events to their associated scsi_device kobjects
2408 * as uevents.
2410 void scsi_evt_thread(struct work_struct *work)
2412 struct scsi_device *sdev;
2413 enum scsi_device_event evt_type;
2414 LIST_HEAD(event_list);
2416 sdev = container_of(work, struct scsi_device, event_work);
2418 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2419 if (test_and_clear_bit(evt_type, sdev->pending_events))
2420 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2422 while (1) {
2423 struct scsi_event *evt;
2424 struct list_head *this, *tmp;
2425 unsigned long flags;
2427 spin_lock_irqsave(&sdev->list_lock, flags);
2428 list_splice_init(&sdev->event_list, &event_list);
2429 spin_unlock_irqrestore(&sdev->list_lock, flags);
2431 if (list_empty(&event_list))
2432 break;
2434 list_for_each_safe(this, tmp, &event_list) {
2435 evt = list_entry(this, struct scsi_event, node);
2436 list_del(&evt->node);
2437 scsi_evt_emit(sdev, evt);
2438 kfree(evt);
2444 * sdev_evt_send - send asserted event to uevent thread
2445 * @sdev: scsi_device event occurred on
2446 * @evt: event to send
2448 * Assert scsi device event asynchronously.
2450 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2452 unsigned long flags;
2454 #if 0
2455 /* FIXME: currently this check eliminates all media change events
2456 * for polled devices. Need to update to discriminate between AN
2457 * and polled events */
2458 if (!test_bit(evt->evt_type, sdev->supported_events)) {
2459 kfree(evt);
2460 return;
2462 #endif
2464 spin_lock_irqsave(&sdev->list_lock, flags);
2465 list_add_tail(&evt->node, &sdev->event_list);
2466 schedule_work(&sdev->event_work);
2467 spin_unlock_irqrestore(&sdev->list_lock, flags);
2469 EXPORT_SYMBOL_GPL(sdev_evt_send);
2472 * sdev_evt_alloc - allocate a new scsi event
2473 * @evt_type: type of event to allocate
2474 * @gfpflags: GFP flags for allocation
2476 * Allocates and returns a new scsi_event.
2478 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2479 gfp_t gfpflags)
2481 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2482 if (!evt)
2483 return NULL;
2485 evt->evt_type = evt_type;
2486 INIT_LIST_HEAD(&evt->node);
2488 /* evt_type-specific initialization, if any */
2489 switch (evt_type) {
2490 case SDEV_EVT_MEDIA_CHANGE:
2491 case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2492 case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2493 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2494 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2495 case SDEV_EVT_LUN_CHANGE_REPORTED:
2496 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2497 case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2498 default:
2499 /* do nothing */
2500 break;
2503 return evt;
2505 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2508 * sdev_evt_send_simple - send asserted event to uevent thread
2509 * @sdev: scsi_device event occurred on
2510 * @evt_type: type of event to send
2511 * @gfpflags: GFP flags for allocation
2513 * Assert scsi device event asynchronously, given an event type.
2515 void sdev_evt_send_simple(struct scsi_device *sdev,
2516 enum scsi_device_event evt_type, gfp_t gfpflags)
2518 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2519 if (!evt) {
2520 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2521 evt_type);
2522 return;
2525 sdev_evt_send(sdev, evt);
2527 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2530 * scsi_device_quiesce - Block user issued commands.
2531 * @sdev: scsi device to quiesce.
2533 * This works by trying to transition to the SDEV_QUIESCE state
2534 * (which must be a legal transition). When the device is in this
2535 * state, only special requests will be accepted, all others will
2536 * be deferred. Since special requests may also be requeued requests,
2537 * a successful return doesn't guarantee the device will be
2538 * totally quiescent.
2540 * Must be called with user context, may sleep.
2542 * Returns zero if unsuccessful or an error if not.
2545 scsi_device_quiesce(struct scsi_device *sdev)
2547 struct request_queue *q = sdev->request_queue;
2548 int err;
2551 * It is allowed to call scsi_device_quiesce() multiple times from
2552 * the same context but concurrent scsi_device_quiesce() calls are
2553 * not allowed.
2555 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2557 if (sdev->quiesced_by == current)
2558 return 0;
2560 blk_set_pm_only(q);
2562 blk_mq_freeze_queue(q);
2564 * Ensure that the effect of blk_set_pm_only() will be visible
2565 * for percpu_ref_tryget() callers that occur after the queue
2566 * unfreeze even if the queue was already frozen before this function
2567 * was called. See also https://lwn.net/Articles/573497/.
2569 synchronize_rcu();
2570 blk_mq_unfreeze_queue(q);
2572 mutex_lock(&sdev->state_mutex);
2573 err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2574 if (err == 0)
2575 sdev->quiesced_by = current;
2576 else
2577 blk_clear_pm_only(q);
2578 mutex_unlock(&sdev->state_mutex);
2580 return err;
2582 EXPORT_SYMBOL(scsi_device_quiesce);
2585 * scsi_device_resume - Restart user issued commands to a quiesced device.
2586 * @sdev: scsi device to resume.
2588 * Moves the device from quiesced back to running and restarts the
2589 * queues.
2591 * Must be called with user context, may sleep.
2593 void scsi_device_resume(struct scsi_device *sdev)
2595 /* check if the device state was mutated prior to resume, and if
2596 * so assume the state is being managed elsewhere (for example
2597 * device deleted during suspend)
2599 mutex_lock(&sdev->state_mutex);
2600 if (sdev->quiesced_by) {
2601 sdev->quiesced_by = NULL;
2602 blk_clear_pm_only(sdev->request_queue);
2604 if (sdev->sdev_state == SDEV_QUIESCE)
2605 scsi_device_set_state(sdev, SDEV_RUNNING);
2606 mutex_unlock(&sdev->state_mutex);
2608 EXPORT_SYMBOL(scsi_device_resume);
2610 static void
2611 device_quiesce_fn(struct scsi_device *sdev, void *data)
2613 scsi_device_quiesce(sdev);
2616 void
2617 scsi_target_quiesce(struct scsi_target *starget)
2619 starget_for_each_device(starget, NULL, device_quiesce_fn);
2621 EXPORT_SYMBOL(scsi_target_quiesce);
2623 static void
2624 device_resume_fn(struct scsi_device *sdev, void *data)
2626 scsi_device_resume(sdev);
2629 void
2630 scsi_target_resume(struct scsi_target *starget)
2632 starget_for_each_device(starget, NULL, device_resume_fn);
2634 EXPORT_SYMBOL(scsi_target_resume);
2637 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2638 * @sdev: device to block
2640 * Pause SCSI command processing on the specified device. Does not sleep.
2642 * Returns zero if successful or a negative error code upon failure.
2644 * Notes:
2645 * This routine transitions the device to the SDEV_BLOCK state (which must be
2646 * a legal transition). When the device is in this state, command processing
2647 * is paused until the device leaves the SDEV_BLOCK state. See also
2648 * scsi_internal_device_unblock_nowait().
2650 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2652 struct request_queue *q = sdev->request_queue;
2653 int err = 0;
2655 err = scsi_device_set_state(sdev, SDEV_BLOCK);
2656 if (err) {
2657 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2659 if (err)
2660 return err;
2664 * The device has transitioned to SDEV_BLOCK. Stop the
2665 * block layer from calling the midlayer with this device's
2666 * request queue.
2668 blk_mq_quiesce_queue_nowait(q);
2669 return 0;
2671 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2674 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2675 * @sdev: device to block
2677 * Pause SCSI command processing on the specified device and wait until all
2678 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2680 * Returns zero if successful or a negative error code upon failure.
2682 * Note:
2683 * This routine transitions the device to the SDEV_BLOCK state (which must be
2684 * a legal transition). When the device is in this state, command processing
2685 * is paused until the device leaves the SDEV_BLOCK state. See also
2686 * scsi_internal_device_unblock().
2688 static int scsi_internal_device_block(struct scsi_device *sdev)
2690 struct request_queue *q = sdev->request_queue;
2691 int err;
2693 mutex_lock(&sdev->state_mutex);
2694 err = scsi_internal_device_block_nowait(sdev);
2695 if (err == 0)
2696 blk_mq_quiesce_queue(q);
2697 mutex_unlock(&sdev->state_mutex);
2699 return err;
2702 void scsi_start_queue(struct scsi_device *sdev)
2704 struct request_queue *q = sdev->request_queue;
2706 blk_mq_unquiesce_queue(q);
2710 * scsi_internal_device_unblock_nowait - resume a device after a block request
2711 * @sdev: device to resume
2712 * @new_state: state to set the device to after unblocking
2714 * Restart the device queue for a previously suspended SCSI device. Does not
2715 * sleep.
2717 * Returns zero if successful or a negative error code upon failure.
2719 * Notes:
2720 * This routine transitions the device to the SDEV_RUNNING state or to one of
2721 * the offline states (which must be a legal transition) allowing the midlayer
2722 * to goose the queue for this device.
2724 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2725 enum scsi_device_state new_state)
2727 switch (new_state) {
2728 case SDEV_RUNNING:
2729 case SDEV_TRANSPORT_OFFLINE:
2730 break;
2731 default:
2732 return -EINVAL;
2736 * Try to transition the scsi device to SDEV_RUNNING or one of the
2737 * offlined states and goose the device queue if successful.
2739 switch (sdev->sdev_state) {
2740 case SDEV_BLOCK:
2741 case SDEV_TRANSPORT_OFFLINE:
2742 sdev->sdev_state = new_state;
2743 break;
2744 case SDEV_CREATED_BLOCK:
2745 if (new_state == SDEV_TRANSPORT_OFFLINE ||
2746 new_state == SDEV_OFFLINE)
2747 sdev->sdev_state = new_state;
2748 else
2749 sdev->sdev_state = SDEV_CREATED;
2750 break;
2751 case SDEV_CANCEL:
2752 case SDEV_OFFLINE:
2753 break;
2754 default:
2755 return -EINVAL;
2757 scsi_start_queue(sdev);
2759 return 0;
2761 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2764 * scsi_internal_device_unblock - resume a device after a block request
2765 * @sdev: device to resume
2766 * @new_state: state to set the device to after unblocking
2768 * Restart the device queue for a previously suspended SCSI device. May sleep.
2770 * Returns zero if successful or a negative error code upon failure.
2772 * Notes:
2773 * This routine transitions the device to the SDEV_RUNNING state or to one of
2774 * the offline states (which must be a legal transition) allowing the midlayer
2775 * to goose the queue for this device.
2777 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2778 enum scsi_device_state new_state)
2780 int ret;
2782 mutex_lock(&sdev->state_mutex);
2783 ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2784 mutex_unlock(&sdev->state_mutex);
2786 return ret;
2789 static void
2790 device_block(struct scsi_device *sdev, void *data)
2792 int ret;
2794 ret = scsi_internal_device_block(sdev);
2796 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2797 dev_name(&sdev->sdev_gendev), ret);
2800 static int
2801 target_block(struct device *dev, void *data)
2803 if (scsi_is_target_device(dev))
2804 starget_for_each_device(to_scsi_target(dev), NULL,
2805 device_block);
2806 return 0;
2809 void
2810 scsi_target_block(struct device *dev)
2812 if (scsi_is_target_device(dev))
2813 starget_for_each_device(to_scsi_target(dev), NULL,
2814 device_block);
2815 else
2816 device_for_each_child(dev, NULL, target_block);
2818 EXPORT_SYMBOL_GPL(scsi_target_block);
2820 static void
2821 device_unblock(struct scsi_device *sdev, void *data)
2823 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2826 static int
2827 target_unblock(struct device *dev, void *data)
2829 if (scsi_is_target_device(dev))
2830 starget_for_each_device(to_scsi_target(dev), data,
2831 device_unblock);
2832 return 0;
2835 void
2836 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2838 if (scsi_is_target_device(dev))
2839 starget_for_each_device(to_scsi_target(dev), &new_state,
2840 device_unblock);
2841 else
2842 device_for_each_child(dev, &new_state, target_unblock);
2844 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2847 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2848 * @sgl: scatter-gather list
2849 * @sg_count: number of segments in sg
2850 * @offset: offset in bytes into sg, on return offset into the mapped area
2851 * @len: bytes to map, on return number of bytes mapped
2853 * Returns virtual address of the start of the mapped page
2855 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2856 size_t *offset, size_t *len)
2858 int i;
2859 size_t sg_len = 0, len_complete = 0;
2860 struct scatterlist *sg;
2861 struct page *page;
2863 WARN_ON(!irqs_disabled());
2865 for_each_sg(sgl, sg, sg_count, i) {
2866 len_complete = sg_len; /* Complete sg-entries */
2867 sg_len += sg->length;
2868 if (sg_len > *offset)
2869 break;
2872 if (unlikely(i == sg_count)) {
2873 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2874 "elements %d\n",
2875 __func__, sg_len, *offset, sg_count);
2876 WARN_ON(1);
2877 return NULL;
2880 /* Offset starting from the beginning of first page in this sg-entry */
2881 *offset = *offset - len_complete + sg->offset;
2883 /* Assumption: contiguous pages can be accessed as "page + i" */
2884 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2885 *offset &= ~PAGE_MASK;
2887 /* Bytes in this sg-entry from *offset to the end of the page */
2888 sg_len = PAGE_SIZE - *offset;
2889 if (*len > sg_len)
2890 *len = sg_len;
2892 return kmap_atomic(page);
2894 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2897 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2898 * @virt: virtual address to be unmapped
2900 void scsi_kunmap_atomic_sg(void *virt)
2902 kunmap_atomic(virt);
2904 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2906 void sdev_disable_disk_events(struct scsi_device *sdev)
2908 atomic_inc(&sdev->disk_events_disable_depth);
2910 EXPORT_SYMBOL(sdev_disable_disk_events);
2912 void sdev_enable_disk_events(struct scsi_device *sdev)
2914 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2915 return;
2916 atomic_dec(&sdev->disk_events_disable_depth);
2918 EXPORT_SYMBOL(sdev_enable_disk_events);
2921 * scsi_vpd_lun_id - return a unique device identification
2922 * @sdev: SCSI device
2923 * @id: buffer for the identification
2924 * @id_len: length of the buffer
2926 * Copies a unique device identification into @id based
2927 * on the information in the VPD page 0x83 of the device.
2928 * The string will be formatted as a SCSI name string.
2930 * Returns the length of the identification or error on failure.
2931 * If the identifier is longer than the supplied buffer the actual
2932 * identifier length is returned and the buffer is not zero-padded.
2934 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2936 u8 cur_id_type = 0xff;
2937 u8 cur_id_size = 0;
2938 const unsigned char *d, *cur_id_str;
2939 const struct scsi_vpd *vpd_pg83;
2940 int id_size = -EINVAL;
2942 rcu_read_lock();
2943 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2944 if (!vpd_pg83) {
2945 rcu_read_unlock();
2946 return -ENXIO;
2950 * Look for the correct descriptor.
2951 * Order of preference for lun descriptor:
2952 * - SCSI name string
2953 * - NAA IEEE Registered Extended
2954 * - EUI-64 based 16-byte
2955 * - EUI-64 based 12-byte
2956 * - NAA IEEE Registered
2957 * - NAA IEEE Extended
2958 * - T10 Vendor ID
2959 * as longer descriptors reduce the likelyhood
2960 * of identification clashes.
2963 /* The id string must be at least 20 bytes + terminating NULL byte */
2964 if (id_len < 21) {
2965 rcu_read_unlock();
2966 return -EINVAL;
2969 memset(id, 0, id_len);
2970 d = vpd_pg83->data + 4;
2971 while (d < vpd_pg83->data + vpd_pg83->len) {
2972 /* Skip designators not referring to the LUN */
2973 if ((d[1] & 0x30) != 0x00)
2974 goto next_desig;
2976 switch (d[1] & 0xf) {
2977 case 0x1:
2978 /* T10 Vendor ID */
2979 if (cur_id_size > d[3])
2980 break;
2981 /* Prefer anything */
2982 if (cur_id_type > 0x01 && cur_id_type != 0xff)
2983 break;
2984 cur_id_size = d[3];
2985 if (cur_id_size + 4 > id_len)
2986 cur_id_size = id_len - 4;
2987 cur_id_str = d + 4;
2988 cur_id_type = d[1] & 0xf;
2989 id_size = snprintf(id, id_len, "t10.%*pE",
2990 cur_id_size, cur_id_str);
2991 break;
2992 case 0x2:
2993 /* EUI-64 */
2994 if (cur_id_size > d[3])
2995 break;
2996 /* Prefer NAA IEEE Registered Extended */
2997 if (cur_id_type == 0x3 &&
2998 cur_id_size == d[3])
2999 break;
3000 cur_id_size = d[3];
3001 cur_id_str = d + 4;
3002 cur_id_type = d[1] & 0xf;
3003 switch (cur_id_size) {
3004 case 8:
3005 id_size = snprintf(id, id_len,
3006 "eui.%8phN",
3007 cur_id_str);
3008 break;
3009 case 12:
3010 id_size = snprintf(id, id_len,
3011 "eui.%12phN",
3012 cur_id_str);
3013 break;
3014 case 16:
3015 id_size = snprintf(id, id_len,
3016 "eui.%16phN",
3017 cur_id_str);
3018 break;
3019 default:
3020 cur_id_size = 0;
3021 break;
3023 break;
3024 case 0x3:
3025 /* NAA */
3026 if (cur_id_size > d[3])
3027 break;
3028 cur_id_size = d[3];
3029 cur_id_str = d + 4;
3030 cur_id_type = d[1] & 0xf;
3031 switch (cur_id_size) {
3032 case 8:
3033 id_size = snprintf(id, id_len,
3034 "naa.%8phN",
3035 cur_id_str);
3036 break;
3037 case 16:
3038 id_size = snprintf(id, id_len,
3039 "naa.%16phN",
3040 cur_id_str);
3041 break;
3042 default:
3043 cur_id_size = 0;
3044 break;
3046 break;
3047 case 0x8:
3048 /* SCSI name string */
3049 if (cur_id_size + 4 > d[3])
3050 break;
3051 /* Prefer others for truncated descriptor */
3052 if (cur_id_size && d[3] > id_len)
3053 break;
3054 cur_id_size = id_size = d[3];
3055 cur_id_str = d + 4;
3056 cur_id_type = d[1] & 0xf;
3057 if (cur_id_size >= id_len)
3058 cur_id_size = id_len - 1;
3059 memcpy(id, cur_id_str, cur_id_size);
3060 /* Decrease priority for truncated descriptor */
3061 if (cur_id_size != id_size)
3062 cur_id_size = 6;
3063 break;
3064 default:
3065 break;
3067 next_desig:
3068 d += d[3] + 4;
3070 rcu_read_unlock();
3072 return id_size;
3074 EXPORT_SYMBOL(scsi_vpd_lun_id);
3077 * scsi_vpd_tpg_id - return a target port group identifier
3078 * @sdev: SCSI device
3080 * Returns the Target Port Group identifier from the information
3081 * froom VPD page 0x83 of the device.
3083 * Returns the identifier or error on failure.
3085 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3087 const unsigned char *d;
3088 const struct scsi_vpd *vpd_pg83;
3089 int group_id = -EAGAIN, rel_port = -1;
3091 rcu_read_lock();
3092 vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3093 if (!vpd_pg83) {
3094 rcu_read_unlock();
3095 return -ENXIO;
3098 d = vpd_pg83->data + 4;
3099 while (d < vpd_pg83->data + vpd_pg83->len) {
3100 switch (d[1] & 0xf) {
3101 case 0x4:
3102 /* Relative target port */
3103 rel_port = get_unaligned_be16(&d[6]);
3104 break;
3105 case 0x5:
3106 /* Target port group */
3107 group_id = get_unaligned_be16(&d[6]);
3108 break;
3109 default:
3110 break;
3112 d += d[3] + 4;
3114 rcu_read_unlock();
3116 if (group_id >= 0 && rel_id && rel_port != -1)
3117 *rel_id = rel_port;
3119 return group_id;
3121 EXPORT_SYMBOL(scsi_vpd_tpg_id);