• Home
  • History
  • Annotate
  • Line#
  • Navigate
  • Raw
  • Download
  • only in /asuswrt-rt-n18u-9.0.0.4.380.2695/release/src-rt-6.x.4708/linux/linux-2.6.36/drivers/scsi/
1/*
2 *  scsi_lib.c Copyright (C) 1999 Eric Youngdale
3 *
4 *  SCSI queueing library.
5 *      Initial versions: Eric Youngdale (eric@andante.org).
6 *                        Based upon conversations with large numbers
7 *                        of people at Linux Expo.
8 */
9
10#include <linux/bio.h>
11#include <linux/bitops.h>
12#include <linux/blkdev.h>
13#include <linux/completion.h>
14#include <linux/kernel.h>
15#include <linux/mempool.h>
16#include <linux/slab.h>
17#include <linux/init.h>
18#include <linux/pci.h>
19#include <linux/delay.h>
20#include <linux/hardirq.h>
21#include <linux/scatterlist.h>
22
23#include <scsi/scsi.h>
24#include <scsi/scsi_cmnd.h>
25#include <scsi/scsi_dbg.h>
26#include <scsi/scsi_device.h>
27#include <scsi/scsi_driver.h>
28#include <scsi/scsi_eh.h>
29#include <scsi/scsi_host.h>
30
31#include "scsi_priv.h"
32#include "scsi_logging.h"
33
34
35#define SG_MEMPOOL_NR		ARRAY_SIZE(scsi_sg_pools)
36#define SG_MEMPOOL_SIZE		2
37
38struct scsi_host_sg_pool {
39	size_t		size;
40	char		*name;
41	struct kmem_cache	*slab;
42	mempool_t	*pool;
43};
44
45#define SP(x) { x, "sgpool-" __stringify(x) }
46#if (SCSI_MAX_SG_SEGMENTS < 32)
47#error SCSI_MAX_SG_SEGMENTS is too small (must be 32 or greater)
48#endif
49static struct scsi_host_sg_pool scsi_sg_pools[] = {
50	SP(8),
51	SP(16),
52#if (SCSI_MAX_SG_SEGMENTS > 32)
53	SP(32),
54#if (SCSI_MAX_SG_SEGMENTS > 64)
55	SP(64),
56#if (SCSI_MAX_SG_SEGMENTS > 128)
57	SP(128),
58#if (SCSI_MAX_SG_SEGMENTS > 256)
59#error SCSI_MAX_SG_SEGMENTS is too large (256 MAX)
60#endif
61#endif
62#endif
63#endif
64	SP(SCSI_MAX_SG_SEGMENTS)
65};
66#undef SP
67
68struct kmem_cache *scsi_sdb_cache;
69
70static void scsi_run_queue(struct request_queue *q);
71
72/*
73 * Function:	scsi_unprep_request()
74 *
75 * Purpose:	Remove all preparation done for a request, including its
76 *		associated scsi_cmnd, so that it can be requeued.
77 *
78 * Arguments:	req	- request to unprepare
79 *
80 * Lock status:	Assumed that no locks are held upon entry.
81 *
82 * Returns:	Nothing.
83 */
84static void scsi_unprep_request(struct request *req)
85{
86	struct scsi_cmnd *cmd = req->special;
87
88	blk_unprep_request(req);
89	req->special = NULL;
90
91	scsi_put_command(cmd);
92}
93
94/**
95 * __scsi_queue_insert - private queue insertion
96 * @cmd: The SCSI command being requeued
97 * @reason:  The reason for the requeue
98 * @unbusy: Whether the queue should be unbusied
99 *
100 * This is a private queue insertion.  The public interface
101 * scsi_queue_insert() always assumes the queue should be unbusied
102 * because it's always called before the completion.  This function is
103 * for a requeue after completion, which should only occur in this
104 * file.
105 */
106static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
107{
108	struct Scsi_Host *host = cmd->device->host;
109	struct scsi_device *device = cmd->device;
110	struct scsi_target *starget = scsi_target(device);
111	struct request_queue *q = device->request_queue;
112	unsigned long flags;
113
114	SCSI_LOG_MLQUEUE(1,
115		 printk("Inserting command %p into mlqueue\n", cmd));
116
117	/*
118	 * Set the appropriate busy bit for the device/host.
119	 *
120	 * If the host/device isn't busy, assume that something actually
121	 * completed, and that we should be able to queue a command now.
122	 *
123	 * Note that the prior mid-layer assumption that any host could
124	 * always queue at least one command is now broken.  The mid-layer
125	 * will implement a user specifiable stall (see
126	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
127	 * if a command is requeued with no other commands outstanding
128	 * either for the device or for the host.
129	 */
130	switch (reason) {
131	case SCSI_MLQUEUE_HOST_BUSY:
132		host->host_blocked = host->max_host_blocked;
133		break;
134	case SCSI_MLQUEUE_DEVICE_BUSY:
135		device->device_blocked = device->max_device_blocked;
136		break;
137	case SCSI_MLQUEUE_TARGET_BUSY:
138		starget->target_blocked = starget->max_target_blocked;
139		break;
140	}
141
142	/*
143	 * Decrement the counters, since these commands are no longer
144	 * active on the host/device.
145	 */
146	if (unbusy)
147		scsi_device_unbusy(device);
148
149	/*
150	 * Requeue this command.  It will go before all other commands
151	 * that are already in the queue.
152	 *
153	 * NOTE: there is magic here about the way the queue is plugged if
154	 * we have no outstanding commands.
155	 *
156	 * Although we *don't* plug the queue, we call the request
157	 * function.  The SCSI request function detects the blocked condition
158	 * and plugs the queue appropriately.
159         */
160	spin_lock_irqsave(q->queue_lock, flags);
161	blk_requeue_request(q, cmd->request);
162	spin_unlock_irqrestore(q->queue_lock, flags);
163
164	scsi_run_queue(q);
165
166	return 0;
167}
168
169/*
170 * Function:    scsi_queue_insert()
171 *
172 * Purpose:     Insert a command in the midlevel queue.
173 *
174 * Arguments:   cmd    - command that we are adding to queue.
175 *              reason - why we are inserting command to queue.
176 *
177 * Lock status: Assumed that lock is not held upon entry.
178 *
179 * Returns:     Nothing.
180 *
181 * Notes:       We do this for one of two cases.  Either the host is busy
182 *              and it cannot accept any more commands for the time being,
183 *              or the device returned QUEUE_FULL and can accept no more
184 *              commands.
185 * Notes:       This could be called either from an interrupt context or a
186 *              normal process context.
187 */
188int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
189{
190	return __scsi_queue_insert(cmd, reason, 1);
191}
192/**
193 * scsi_execute - insert request and wait for the result
194 * @sdev:	scsi device
195 * @cmd:	scsi command
196 * @data_direction: data direction
197 * @buffer:	data buffer
198 * @bufflen:	len of buffer
199 * @sense:	optional sense buffer
200 * @timeout:	request timeout in seconds
201 * @retries:	number of times to retry request
202 * @flags:	or into request flags;
203 * @resid:	optional residual length
204 *
205 * returns the req->errors value which is the scsi_cmnd result
206 * field.
207 */
208int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
209		 int data_direction, void *buffer, unsigned bufflen,
210		 unsigned char *sense, int timeout, int retries, int flags,
211		 int *resid)
212{
213	struct request *req;
214	int write = (data_direction == DMA_TO_DEVICE);
215	int ret = DRIVER_ERROR << 24;
216
217	req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
218
219	if (bufflen &&	blk_rq_map_kern(sdev->request_queue, req,
220					buffer, bufflen, __GFP_WAIT))
221		goto out;
222
223	req->cmd_len = COMMAND_SIZE(cmd[0]);
224	memcpy(req->cmd, cmd, req->cmd_len);
225	req->sense = sense;
226	req->sense_len = 0;
227	req->retries = retries;
228	req->timeout = timeout;
229	req->cmd_type = REQ_TYPE_BLOCK_PC;
230	req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
231
232	/*
233	 * head injection *required* here otherwise quiesce won't work
234	 */
235	blk_execute_rq(req->q, NULL, req, 1);
236
237	/*
238	 * Some devices (USB mass-storage in particular) may transfer
239	 * garbage data together with a residue indicating that the data
240	 * is invalid.  Prevent the garbage from being misinterpreted
241	 * and prevent security leaks by zeroing out the excess data.
242	 */
243	if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
244		memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
245
246	if (resid)
247		*resid = req->resid_len;
248	ret = req->errors;
249 out:
250	blk_put_request(req);
251
252	return ret;
253}
254EXPORT_SYMBOL(scsi_execute);
255
256
257int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
258		     int data_direction, void *buffer, unsigned bufflen,
259		     struct scsi_sense_hdr *sshdr, int timeout, int retries,
260		     int *resid)
261{
262	char *sense = NULL;
263	int result;
264
265	if (sshdr) {
266		sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
267		if (!sense)
268			return DRIVER_ERROR << 24;
269	}
270	result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
271			      sense, timeout, retries, 0, resid);
272	if (sshdr)
273		scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
274
275	kfree(sense);
276	return result;
277}
278EXPORT_SYMBOL(scsi_execute_req);
279
280/*
281 * Function:    scsi_init_cmd_errh()
282 *
283 * Purpose:     Initialize cmd fields related to error handling.
284 *
285 * Arguments:   cmd	- command that is ready to be queued.
286 *
287 * Notes:       This function has the job of initializing a number of
288 *              fields related to error handling.   Typically this will
289 *              be called once for each command, as required.
290 */
291static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
292{
293	cmd->serial_number = 0;
294	scsi_set_resid(cmd, 0);
295	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
296	if (cmd->cmd_len == 0)
297		cmd->cmd_len = scsi_command_size(cmd->cmnd);
298}
299
300void scsi_device_unbusy(struct scsi_device *sdev)
301{
302	struct Scsi_Host *shost = sdev->host;
303	struct scsi_target *starget = scsi_target(sdev);
304	unsigned long flags;
305
306	spin_lock_irqsave(shost->host_lock, flags);
307	shost->host_busy--;
308	starget->target_busy--;
309	if (unlikely(scsi_host_in_recovery(shost) &&
310		     (shost->host_failed || shost->host_eh_scheduled)))
311		scsi_eh_wakeup(shost);
312	spin_unlock(shost->host_lock);
313	spin_lock(sdev->request_queue->queue_lock);
314	sdev->device_busy--;
315	spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
316}
317
318/*
319 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
320 * and call blk_run_queue for all the scsi_devices on the target -
321 * including current_sdev first.
322 *
323 * Called with *no* scsi locks held.
324 */
325static void scsi_single_lun_run(struct scsi_device *current_sdev)
326{
327	struct Scsi_Host *shost = current_sdev->host;
328	struct scsi_device *sdev, *tmp;
329	struct scsi_target *starget = scsi_target(current_sdev);
330	unsigned long flags;
331
332	spin_lock_irqsave(shost->host_lock, flags);
333	starget->starget_sdev_user = NULL;
334	spin_unlock_irqrestore(shost->host_lock, flags);
335
336	/*
337	 * Call blk_run_queue for all LUNs on the target, starting with
338	 * current_sdev. We race with others (to set starget_sdev_user),
339	 * but in most cases, we will be first. Ideally, each LU on the
340	 * target would get some limited time or requests on the target.
341	 */
342	blk_run_queue(current_sdev->request_queue);
343
344	spin_lock_irqsave(shost->host_lock, flags);
345	if (starget->starget_sdev_user)
346		goto out;
347	list_for_each_entry_safe(sdev, tmp, &starget->devices,
348			same_target_siblings) {
349		if (sdev == current_sdev)
350			continue;
351		if (scsi_device_get(sdev))
352			continue;
353
354		spin_unlock_irqrestore(shost->host_lock, flags);
355		blk_run_queue(sdev->request_queue);
356		spin_lock_irqsave(shost->host_lock, flags);
357
358		scsi_device_put(sdev);
359	}
360 out:
361	spin_unlock_irqrestore(shost->host_lock, flags);
362}
363
364static inline int scsi_device_is_busy(struct scsi_device *sdev)
365{
366	if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
367		return 1;
368
369	return 0;
370}
371
372static inline int scsi_target_is_busy(struct scsi_target *starget)
373{
374	return ((starget->can_queue > 0 &&
375		 starget->target_busy >= starget->can_queue) ||
376		 starget->target_blocked);
377}
378
379static inline int scsi_host_is_busy(struct Scsi_Host *shost)
380{
381	if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
382	    shost->host_blocked || shost->host_self_blocked)
383		return 1;
384
385	return 0;
386}
387
388/*
389 * Function:	scsi_run_queue()
390 *
391 * Purpose:	Select a proper request queue to serve next
392 *
393 * Arguments:	q	- last request's queue
394 *
395 * Returns:     Nothing
396 *
397 * Notes:	The previous command was completely finished, start
398 *		a new one if possible.
399 */
400static void scsi_run_queue(struct request_queue *q)
401{
402	struct scsi_device *sdev = q->queuedata;
403	struct Scsi_Host *shost = sdev->host;
404	LIST_HEAD(starved_list);
405	unsigned long flags;
406
407	if (scsi_target(sdev)->single_lun)
408		scsi_single_lun_run(sdev);
409
410	spin_lock_irqsave(shost->host_lock, flags);
411	list_splice_init(&shost->starved_list, &starved_list);
412
413	while (!list_empty(&starved_list)) {
414		int flagset;
415
416		/*
417		 * As long as shost is accepting commands and we have
418		 * starved queues, call blk_run_queue. scsi_request_fn
419		 * drops the queue_lock and can add us back to the
420		 * starved_list.
421		 *
422		 * host_lock protects the starved_list and starved_entry.
423		 * scsi_request_fn must get the host_lock before checking
424		 * or modifying starved_list or starved_entry.
425		 */
426		if (scsi_host_is_busy(shost))
427			break;
428
429		sdev = list_entry(starved_list.next,
430				  struct scsi_device, starved_entry);
431		list_del_init(&sdev->starved_entry);
432		if (scsi_target_is_busy(scsi_target(sdev))) {
433			list_move_tail(&sdev->starved_entry,
434				       &shost->starved_list);
435			continue;
436		}
437
438		spin_unlock(shost->host_lock);
439
440		spin_lock(sdev->request_queue->queue_lock);
441		flagset = test_bit(QUEUE_FLAG_REENTER, &q->queue_flags) &&
442				!test_bit(QUEUE_FLAG_REENTER,
443					&sdev->request_queue->queue_flags);
444		if (flagset)
445			queue_flag_set(QUEUE_FLAG_REENTER, sdev->request_queue);
446		__blk_run_queue(sdev->request_queue);
447		if (flagset)
448			queue_flag_clear(QUEUE_FLAG_REENTER, sdev->request_queue);
449		spin_unlock(sdev->request_queue->queue_lock);
450
451		spin_lock(shost->host_lock);
452	}
453	/* put any unprocessed entries back */
454	list_splice(&starved_list, &shost->starved_list);
455	spin_unlock_irqrestore(shost->host_lock, flags);
456
457	blk_run_queue(q);
458}
459
460/*
461 * Function:	scsi_requeue_command()
462 *
463 * Purpose:	Handle post-processing of completed commands.
464 *
465 * Arguments:	q	- queue to operate on
466 *		cmd	- command that may need to be requeued.
467 *
468 * Returns:	Nothing
469 *
470 * Notes:	After command completion, there may be blocks left
471 *		over which weren't finished by the previous command
472 *		this can be for a number of reasons - the main one is
473 *		I/O errors in the middle of the request, in which case
474 *		we need to request the blocks that come after the bad
475 *		sector.
476 * Notes:	Upon return, cmd is a stale pointer.
477 */
478static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
479{
480	struct request *req = cmd->request;
481	unsigned long flags;
482
483	spin_lock_irqsave(q->queue_lock, flags);
484	scsi_unprep_request(req);
485	blk_requeue_request(q, req);
486	spin_unlock_irqrestore(q->queue_lock, flags);
487
488	scsi_run_queue(q);
489}
490
491void scsi_next_command(struct scsi_cmnd *cmd)
492{
493	struct scsi_device *sdev = cmd->device;
494	struct request_queue *q = sdev->request_queue;
495
496	/* need to hold a reference on the device before we let go of the cmd */
497	get_device(&sdev->sdev_gendev);
498
499	scsi_put_command(cmd);
500	scsi_run_queue(q);
501
502	/* ok to remove device now */
503	put_device(&sdev->sdev_gendev);
504}
505
506void scsi_run_host_queues(struct Scsi_Host *shost)
507{
508	struct scsi_device *sdev;
509
510	shost_for_each_device(sdev, shost)
511		scsi_run_queue(sdev->request_queue);
512}
513
514static void __scsi_release_buffers(struct scsi_cmnd *, int);
515
516/*
517 * Function:    scsi_end_request()
518 *
519 * Purpose:     Post-processing of completed commands (usually invoked at end
520 *		of upper level post-processing and scsi_io_completion).
521 *
522 * Arguments:   cmd	 - command that is complete.
523 *              error    - 0 if I/O indicates success, < 0 for I/O error.
524 *              bytes    - number of bytes of completed I/O
525 *		requeue  - indicates whether we should requeue leftovers.
526 *
527 * Lock status: Assumed that lock is not held upon entry.
528 *
529 * Returns:     cmd if requeue required, NULL otherwise.
530 *
531 * Notes:       This is called for block device requests in order to
532 *              mark some number of sectors as complete.
533 *
534 *		We are guaranteeing that the request queue will be goosed
535 *		at some point during this call.
536 * Notes:	If cmd was requeued, upon return it will be a stale pointer.
537 */
538static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
539					  int bytes, int requeue)
540{
541	struct request_queue *q = cmd->device->request_queue;
542	struct request *req = cmd->request;
543
544	/*
545	 * If there are blocks left over at the end, set up the command
546	 * to queue the remainder of them.
547	 */
548	if (blk_end_request(req, error, bytes)) {
549		/* kill remainder if no retrys */
550		if (error && scsi_noretry_cmd(cmd))
551			blk_end_request_all(req, error);
552		else {
553			if (requeue) {
554				/*
555				 * Bleah.  Leftovers again.  Stick the
556				 * leftovers in the front of the
557				 * queue, and goose the queue again.
558				 */
559				scsi_release_buffers(cmd);
560				scsi_requeue_command(q, cmd);
561				cmd = NULL;
562			}
563			return cmd;
564		}
565	}
566
567	/*
568	 * This will goose the queue request function at the end, so we don't
569	 * need to worry about launching another command.
570	 */
571	__scsi_release_buffers(cmd, 0);
572	scsi_next_command(cmd);
573	return NULL;
574}
575
576static inline unsigned int scsi_sgtable_index(unsigned short nents)
577{
578	unsigned int index;
579
580	BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
581
582	if (nents <= 8)
583		index = 0;
584	else
585		index = get_count_order(nents) - 3;
586
587	return index;
588}
589
590static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
591{
592	struct scsi_host_sg_pool *sgp;
593
594	sgp = scsi_sg_pools + scsi_sgtable_index(nents);
595	mempool_free(sgl, sgp->pool);
596}
597
598static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
599{
600	struct scsi_host_sg_pool *sgp;
601
602	sgp = scsi_sg_pools + scsi_sgtable_index(nents);
603	return mempool_alloc(sgp->pool, gfp_mask);
604}
605
606static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
607			      gfp_t gfp_mask)
608{
609	int ret;
610
611	BUG_ON(!nents);
612
613	ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
614			       gfp_mask, scsi_sg_alloc);
615	if (unlikely(ret))
616		__sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
617				scsi_sg_free);
618
619	return ret;
620}
621
622static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
623{
624	__sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
625}
626
627static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
628{
629
630	if (cmd->sdb.table.nents)
631		scsi_free_sgtable(&cmd->sdb);
632
633	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
634
635	if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
636		struct scsi_data_buffer *bidi_sdb =
637			cmd->request->next_rq->special;
638		scsi_free_sgtable(bidi_sdb);
639		kmem_cache_free(scsi_sdb_cache, bidi_sdb);
640		cmd->request->next_rq->special = NULL;
641	}
642
643	if (scsi_prot_sg_count(cmd))
644		scsi_free_sgtable(cmd->prot_sdb);
645}
646
647/*
648 * Function:    scsi_release_buffers()
649 *
650 * Purpose:     Completion processing for block device I/O requests.
651 *
652 * Arguments:   cmd	- command that we are bailing.
653 *
654 * Lock status: Assumed that no lock is held upon entry.
655 *
656 * Returns:     Nothing
657 *
658 * Notes:       In the event that an upper level driver rejects a
659 *		command, we must release resources allocated during
660 *		the __init_io() function.  Primarily this would involve
661 *		the scatter-gather table, and potentially any bounce
662 *		buffers.
663 */
664void scsi_release_buffers(struct scsi_cmnd *cmd)
665{
666	__scsi_release_buffers(cmd, 1);
667}
668EXPORT_SYMBOL(scsi_release_buffers);
669
670/*
671 * Function:    scsi_io_completion()
672 *
673 * Purpose:     Completion processing for block device I/O requests.
674 *
675 * Arguments:   cmd   - command that is finished.
676 *
677 * Lock status: Assumed that no lock is held upon entry.
678 *
679 * Returns:     Nothing
680 *
681 * Notes:       This function is matched in terms of capabilities to
682 *              the function that created the scatter-gather list.
683 *              In other words, if there are no bounce buffers
684 *              (the normal case for most drivers), we don't need
685 *              the logic to deal with cleaning up afterwards.
686 *
687 *		We must call scsi_end_request().  This will finish off
688 *		the specified number of sectors.  If we are done, the
689 *		command block will be released and the queue function
690 *		will be goosed.  If we are not done then we have to
691 *		figure out what to do next:
692 *
693 *		a) We can call scsi_requeue_command().  The request
694 *		   will be unprepared and put back on the queue.  Then
695 *		   a new command will be created for it.  This should
696 *		   be used if we made forward progress, or if we want
697 *		   to switch from READ(10) to READ(6) for example.
698 *
699 *		b) We can call scsi_queue_insert().  The request will
700 *		   be put back on the queue and retried using the same
701 *		   command as before, possibly after a delay.
702 *
703 *		c) We can call blk_end_request() with -EIO to fail
704 *		   the remainder of the request.
705 */
706void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
707{
708	int result = cmd->result;
709	struct request_queue *q = cmd->device->request_queue;
710	struct request *req = cmd->request;
711	int error = 0;
712	struct scsi_sense_hdr sshdr;
713	int sense_valid = 0;
714	int sense_deferred = 0;
715	enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
716	      ACTION_DELAYED_RETRY} action;
717	char *description = NULL;
718
719	if (result) {
720		sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
721		if (sense_valid)
722			sense_deferred = scsi_sense_is_deferred(&sshdr);
723	}
724
725	if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
726		req->errors = result;
727		if (result) {
728			if (sense_valid && req->sense) {
729				/*
730				 * SG_IO wants current and deferred errors
731				 */
732				int len = 8 + cmd->sense_buffer[7];
733
734				if (len > SCSI_SENSE_BUFFERSIZE)
735					len = SCSI_SENSE_BUFFERSIZE;
736				memcpy(req->sense, cmd->sense_buffer,  len);
737				req->sense_len = len;
738			}
739			if (!sense_deferred)
740				error = -EIO;
741		}
742
743		req->resid_len = scsi_get_resid(cmd);
744
745		if (scsi_bidi_cmnd(cmd)) {
746			/*
747			 * Bidi commands Must be complete as a whole,
748			 * both sides at once.
749			 */
750			req->next_rq->resid_len = scsi_in(cmd)->resid;
751
752			scsi_release_buffers(cmd);
753			blk_end_request_all(req, 0);
754
755			scsi_next_command(cmd);
756			return;
757		}
758	}
759
760	/* no bidi support for !REQ_TYPE_BLOCK_PC yet */
761	BUG_ON(blk_bidi_rq(req));
762
763	/*
764	 * Next deal with any sectors which we were able to correctly
765	 * handle.
766	 */
767	SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
768				      "%d bytes done.\n",
769				      blk_rq_sectors(req), good_bytes));
770
771	/*
772	 * Recovered errors need reporting, but they're always treated
773	 * as success, so fiddle the result code here.  For BLOCK_PC
774	 * we already took a copy of the original into rq->errors which
775	 * is what gets returned to the user
776	 */
777	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
778		/* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
779		 * print since caller wants ATA registers. Only occurs on
780		 * SCSI ATA PASS_THROUGH commands when CK_COND=1
781		 */
782		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
783			;
784		else if (!(req->cmd_flags & REQ_QUIET))
785			scsi_print_sense("", cmd);
786		result = 0;
787		/* BLOCK_PC may have set error */
788		error = 0;
789	}
790
791	/*
792	 * A number of bytes were successfully read.  If there
793	 * are leftovers and there is some kind of error
794	 * (result != 0), retry the rest.
795	 */
796	if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
797		return;
798
799	error = -EIO;
800
801	if (host_byte(result) == DID_RESET) {
802		/* Third party bus reset or reset for error recovery
803		 * reasons.  Just retry the command and see what
804		 * happens.
805		 */
806		action = ACTION_RETRY;
807	} else if (sense_valid && !sense_deferred) {
808		switch (sshdr.sense_key) {
809		case UNIT_ATTENTION:
810			if (cmd->device->removable) {
811				/* Detected disc change.  Set a bit
812				 * and quietly refuse further access.
813				 */
814				cmd->device->changed = 1;
815				description = "Media Changed";
816				action = ACTION_FAIL;
817			} else {
818				/* Must have been a power glitch, or a
819				 * bus reset.  Could not have been a
820				 * media change, so we just retry the
821				 * command and see what happens.
822				 */
823				action = ACTION_RETRY;
824			}
825			break;
826		case ILLEGAL_REQUEST:
827			/* If we had an ILLEGAL REQUEST returned, then
828			 * we may have performed an unsupported
829			 * command.  The only thing this should be
830			 * would be a ten byte read where only a six
831			 * byte read was supported.  Also, on a system
832			 * where READ CAPACITY failed, we may have
833			 * read past the end of the disk.
834			 */
835			if ((cmd->device->use_10_for_rw &&
836			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
837			    (cmd->cmnd[0] == READ_10 ||
838			     cmd->cmnd[0] == WRITE_10)) {
839				/* This will issue a new 6-byte command. */
840				cmd->device->use_10_for_rw = 0;
841				action = ACTION_REPREP;
842			} else if (sshdr.asc == 0x10) /* DIX */ {
843				description = "Host Data Integrity Failure";
844				action = ACTION_FAIL;
845				error = -EILSEQ;
846			} else
847				action = ACTION_FAIL;
848			break;
849		case ABORTED_COMMAND:
850			action = ACTION_FAIL;
851			if (sshdr.asc == 0x10) { /* DIF */
852				description = "Target Data Integrity Failure";
853				error = -EILSEQ;
854			}
855			break;
856		case NOT_READY:
857			/* If the device is in the process of becoming
858			 * ready, or has a temporary blockage, retry.
859			 */
860			if (sshdr.asc == 0x04) {
861				switch (sshdr.ascq) {
862				case 0x01: /* becoming ready */
863				case 0x04: /* format in progress */
864				case 0x05: /* rebuild in progress */
865				case 0x06: /* recalculation in progress */
866				case 0x07: /* operation in progress */
867				case 0x08: /* Long write in progress */
868				case 0x09: /* self test in progress */
869				case 0x14: /* space allocation in progress */
870					action = ACTION_DELAYED_RETRY;
871					break;
872				default:
873					description = "Device not ready";
874					action = ACTION_FAIL;
875					break;
876				}
877			} else {
878				description = "Device not ready";
879				action = ACTION_FAIL;
880			}
881			break;
882		case VOLUME_OVERFLOW:
883			/* See SSC3rXX or current. */
884			action = ACTION_FAIL;
885			break;
886		default:
887			description = "Unhandled sense code";
888			action = ACTION_FAIL;
889			break;
890		}
891	} else {
892		description = "Unhandled error code";
893		action = ACTION_FAIL;
894	}
895
896	switch (action) {
897	case ACTION_FAIL:
898		/* Give up and fail the remainder of the request */
899		scsi_release_buffers(cmd);
900		if (!(req->cmd_flags & REQ_QUIET)) {
901			if (description)
902				scmd_printk(KERN_INFO, cmd, "%s\n",
903					    description);
904			scsi_print_result(cmd);
905			if (driver_byte(result) & DRIVER_SENSE)
906				scsi_print_sense("", cmd);
907			scsi_print_command(cmd);
908		}
909		if (blk_end_request_err(req, error))
910			scsi_requeue_command(q, cmd);
911		else
912			scsi_next_command(cmd);
913		break;
914	case ACTION_REPREP:
915		/* Unprep the request and put it back at the head of the queue.
916		 * A new command will be prepared and issued.
917		 */
918		scsi_release_buffers(cmd);
919		scsi_requeue_command(q, cmd);
920		break;
921	case ACTION_RETRY:
922		/* Retry the same command immediately */
923		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
924		break;
925	case ACTION_DELAYED_RETRY:
926		/* Retry the same command after a delay */
927		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
928		break;
929	}
930}
931
932static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
933			     gfp_t gfp_mask)
934{
935	int count;
936
937	/*
938	 * If sg table allocation fails, requeue request later.
939	 */
940	if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
941					gfp_mask))) {
942		return BLKPREP_DEFER;
943	}
944
945	req->buffer = NULL;
946
947	/*
948	 * Next, walk the list, and fill in the addresses and sizes of
949	 * each segment.
950	 */
951	count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
952	BUG_ON(count > sdb->table.nents);
953	sdb->table.nents = count;
954	sdb->length = blk_rq_bytes(req);
955	return BLKPREP_OK;
956}
957
958/*
959 * Function:    scsi_init_io()
960 *
961 * Purpose:     SCSI I/O initialize function.
962 *
963 * Arguments:   cmd   - Command descriptor we wish to initialize
964 *
965 * Returns:     0 on success
966 *		BLKPREP_DEFER if the failure is retryable
967 *		BLKPREP_KILL if the failure is fatal
968 */
969int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
970{
971	int error = scsi_init_sgtable(cmd->request, &cmd->sdb, gfp_mask);
972	if (error)
973		goto err_exit;
974
975	if (blk_bidi_rq(cmd->request)) {
976		struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
977			scsi_sdb_cache, GFP_ATOMIC);
978		if (!bidi_sdb) {
979			error = BLKPREP_DEFER;
980			goto err_exit;
981		}
982
983		cmd->request->next_rq->special = bidi_sdb;
984		error = scsi_init_sgtable(cmd->request->next_rq, bidi_sdb,
985								    GFP_ATOMIC);
986		if (error)
987			goto err_exit;
988	}
989
990	if (blk_integrity_rq(cmd->request)) {
991		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
992		int ivecs, count;
993
994		BUG_ON(prot_sdb == NULL);
995		ivecs = blk_rq_count_integrity_sg(cmd->request);
996
997		if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
998			error = BLKPREP_DEFER;
999			goto err_exit;
1000		}
1001
1002		count = blk_rq_map_integrity_sg(cmd->request,
1003						prot_sdb->table.sgl);
1004		BUG_ON(unlikely(count > ivecs));
1005
1006		cmd->prot_sdb = prot_sdb;
1007		cmd->prot_sdb->table.nents = count;
1008	}
1009
1010	return BLKPREP_OK ;
1011
1012err_exit:
1013	scsi_release_buffers(cmd);
1014	cmd->request->special = NULL;
1015	scsi_put_command(cmd);
1016	return error;
1017}
1018EXPORT_SYMBOL(scsi_init_io);
1019
1020static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1021		struct request *req)
1022{
1023	struct scsi_cmnd *cmd;
1024
1025	if (!req->special) {
1026		cmd = scsi_get_command(sdev, GFP_ATOMIC);
1027		if (unlikely(!cmd))
1028			return NULL;
1029		req->special = cmd;
1030	} else {
1031		cmd = req->special;
1032	}
1033
1034	/* pull a tag out of the request if we have one */
1035	cmd->tag = req->tag;
1036	cmd->request = req;
1037
1038	cmd->cmnd = req->cmd;
1039
1040	return cmd;
1041}
1042
1043int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1044{
1045	struct scsi_cmnd *cmd;
1046	int ret = scsi_prep_state_check(sdev, req);
1047
1048	if (ret != BLKPREP_OK)
1049		return ret;
1050
1051	cmd = scsi_get_cmd_from_req(sdev, req);
1052	if (unlikely(!cmd))
1053		return BLKPREP_DEFER;
1054
1055	/*
1056	 * BLOCK_PC requests may transfer data, in which case they must
1057	 * a bio attached to them.  Or they might contain a SCSI command
1058	 * that does not transfer data, in which case they may optionally
1059	 * submit a request without an attached bio.
1060	 */
1061	if (req->bio) {
1062		int ret;
1063
1064		BUG_ON(!req->nr_phys_segments);
1065
1066		ret = scsi_init_io(cmd, GFP_ATOMIC);
1067		if (unlikely(ret))
1068			return ret;
1069	} else {
1070		BUG_ON(blk_rq_bytes(req));
1071
1072		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1073		req->buffer = NULL;
1074	}
1075
1076	cmd->cmd_len = req->cmd_len;
1077	if (!blk_rq_bytes(req))
1078		cmd->sc_data_direction = DMA_NONE;
1079	else if (rq_data_dir(req) == WRITE)
1080		cmd->sc_data_direction = DMA_TO_DEVICE;
1081	else
1082		cmd->sc_data_direction = DMA_FROM_DEVICE;
1083
1084	cmd->transfersize = blk_rq_bytes(req);
1085	cmd->allowed = req->retries;
1086	return BLKPREP_OK;
1087}
1088EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1089
1090/*
1091 * Setup a REQ_TYPE_FS command.  These are simple read/write request
1092 * from filesystems that still need to be translated to SCSI CDBs from
1093 * the ULD.
1094 */
1095int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1096{
1097	struct scsi_cmnd *cmd;
1098	int ret = scsi_prep_state_check(sdev, req);
1099
1100	if (ret != BLKPREP_OK)
1101		return ret;
1102
1103	if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1104			 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1105		ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1106		if (ret != BLKPREP_OK)
1107			return ret;
1108	}
1109
1110	/*
1111	 * Filesystem requests must transfer data.
1112	 */
1113	BUG_ON(!req->nr_phys_segments);
1114
1115	cmd = scsi_get_cmd_from_req(sdev, req);
1116	if (unlikely(!cmd))
1117		return BLKPREP_DEFER;
1118
1119	memset(cmd->cmnd, 0, BLK_MAX_CDB);
1120	return scsi_init_io(cmd, GFP_ATOMIC);
1121}
1122EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1123
1124int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1125{
1126	int ret = BLKPREP_OK;
1127
1128	/*
1129	 * If the device is not in running state we will reject some
1130	 * or all commands.
1131	 */
1132	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1133		switch (sdev->sdev_state) {
1134		case SDEV_OFFLINE:
1135			/*
1136			 * If the device is offline we refuse to process any
1137			 * commands.  The device must be brought online
1138			 * before trying any recovery commands.
1139			 */
1140			sdev_printk(KERN_ERR, sdev,
1141				    "rejecting I/O to offline device\n");
1142			ret = BLKPREP_KILL;
1143			break;
1144		case SDEV_DEL:
1145			/*
1146			 * If the device is fully deleted, we refuse to
1147			 * process any commands as well.
1148			 */
1149			sdev_printk(KERN_ERR, sdev,
1150				    "rejecting I/O to dead device\n");
1151			ret = BLKPREP_KILL;
1152			break;
1153		case SDEV_QUIESCE:
1154		case SDEV_BLOCK:
1155		case SDEV_CREATED_BLOCK:
1156			/*
1157			 * If the devices is blocked we defer normal commands.
1158			 */
1159			if (!(req->cmd_flags & REQ_PREEMPT))
1160				ret = BLKPREP_DEFER;
1161			break;
1162		default:
1163			/*
1164			 * For any other not fully online state we only allow
1165			 * special commands.  In particular any user initiated
1166			 * command is not allowed.
1167			 */
1168			if (!(req->cmd_flags & REQ_PREEMPT))
1169				ret = BLKPREP_KILL;
1170			break;
1171		}
1172	}
1173	return ret;
1174}
1175EXPORT_SYMBOL(scsi_prep_state_check);
1176
1177int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1178{
1179	struct scsi_device *sdev = q->queuedata;
1180
1181	switch (ret) {
1182	case BLKPREP_KILL:
1183		req->errors = DID_NO_CONNECT << 16;
1184		/* release the command and kill it */
1185		if (req->special) {
1186			struct scsi_cmnd *cmd = req->special;
1187			scsi_release_buffers(cmd);
1188			scsi_put_command(cmd);
1189			req->special = NULL;
1190		}
1191		break;
1192	case BLKPREP_DEFER:
1193		/*
1194		 * If we defer, the blk_peek_request() returns NULL, but the
1195		 * queue must be restarted, so we plug here if no returning
1196		 * command will automatically do that.
1197		 */
1198		if (sdev->device_busy == 0)
1199			blk_plug_device(q);
1200		break;
1201	default:
1202		req->cmd_flags |= REQ_DONTPREP;
1203	}
1204
1205	return ret;
1206}
1207EXPORT_SYMBOL(scsi_prep_return);
1208
1209int scsi_prep_fn(struct request_queue *q, struct request *req)
1210{
1211	struct scsi_device *sdev = q->queuedata;
1212	int ret = BLKPREP_KILL;
1213
1214	if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1215		ret = scsi_setup_blk_pc_cmnd(sdev, req);
1216	return scsi_prep_return(q, req, ret);
1217}
1218EXPORT_SYMBOL(scsi_prep_fn);
1219
1220/*
1221 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1222 * return 0.
1223 *
1224 * Called with the queue_lock held.
1225 */
1226static inline int scsi_dev_queue_ready(struct request_queue *q,
1227				  struct scsi_device *sdev)
1228{
1229	if (sdev->device_busy == 0 && sdev->device_blocked) {
1230		/*
1231		 * unblock after device_blocked iterates to zero
1232		 */
1233		if (--sdev->device_blocked == 0) {
1234			SCSI_LOG_MLQUEUE(3,
1235				   sdev_printk(KERN_INFO, sdev,
1236				   "unblocking device at zero depth\n"));
1237		} else {
1238			blk_plug_device(q);
1239			return 0;
1240		}
1241	}
1242	if (scsi_device_is_busy(sdev))
1243		return 0;
1244
1245	return 1;
1246}
1247
1248
1249/*
1250 * scsi_target_queue_ready: checks if there we can send commands to target
1251 * @sdev: scsi device on starget to check.
1252 *
1253 * Called with the host lock held.
1254 */
1255static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1256					   struct scsi_device *sdev)
1257{
1258	struct scsi_target *starget = scsi_target(sdev);
1259
1260	if (starget->single_lun) {
1261		if (starget->starget_sdev_user &&
1262		    starget->starget_sdev_user != sdev)
1263			return 0;
1264		starget->starget_sdev_user = sdev;
1265	}
1266
1267	if (starget->target_busy == 0 && starget->target_blocked) {
1268		/*
1269		 * unblock after target_blocked iterates to zero
1270		 */
1271		if (--starget->target_blocked == 0) {
1272			SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1273					 "unblocking target at zero depth\n"));
1274		} else
1275			return 0;
1276	}
1277
1278	if (scsi_target_is_busy(starget)) {
1279		if (list_empty(&sdev->starved_entry)) {
1280			list_add_tail(&sdev->starved_entry,
1281				      &shost->starved_list);
1282			return 0;
1283		}
1284	}
1285
1286	/* We're OK to process the command, so we can't be starved */
1287	if (!list_empty(&sdev->starved_entry))
1288		list_del_init(&sdev->starved_entry);
1289	return 1;
1290}
1291
1292/*
1293 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1294 * return 0. We must end up running the queue again whenever 0 is
1295 * returned, else IO can hang.
1296 *
1297 * Called with host_lock held.
1298 */
1299static inline int scsi_host_queue_ready(struct request_queue *q,
1300				   struct Scsi_Host *shost,
1301				   struct scsi_device *sdev)
1302{
1303	if (scsi_host_in_recovery(shost))
1304		return 0;
1305	if (shost->host_busy == 0 && shost->host_blocked) {
1306		/*
1307		 * unblock after host_blocked iterates to zero
1308		 */
1309		if (--shost->host_blocked == 0) {
1310			SCSI_LOG_MLQUEUE(3,
1311				printk("scsi%d unblocking host at zero depth\n",
1312					shost->host_no));
1313		} else {
1314			return 0;
1315		}
1316	}
1317	if (scsi_host_is_busy(shost)) {
1318		if (list_empty(&sdev->starved_entry))
1319			list_add_tail(&sdev->starved_entry, &shost->starved_list);
1320		return 0;
1321	}
1322
1323	/* We're OK to process the command, so we can't be starved */
1324	if (!list_empty(&sdev->starved_entry))
1325		list_del_init(&sdev->starved_entry);
1326
1327	return 1;
1328}
1329
1330/*
1331 * Busy state exporting function for request stacking drivers.
1332 *
1333 * For efficiency, no lock is taken to check the busy state of
1334 * shost/starget/sdev, since the returned value is not guaranteed and
1335 * may be changed after request stacking drivers call the function,
1336 * regardless of taking lock or not.
1337 *
1338 * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1339 * (e.g. !sdev), scsi needs to return 'not busy'.
1340 * Otherwise, request stacking drivers may hold requests forever.
1341 */
1342static int scsi_lld_busy(struct request_queue *q)
1343{
1344	struct scsi_device *sdev = q->queuedata;
1345	struct Scsi_Host *shost;
1346	struct scsi_target *starget;
1347
1348	if (!sdev)
1349		return 0;
1350
1351	shost = sdev->host;
1352	starget = scsi_target(sdev);
1353
1354	if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) ||
1355	    scsi_target_is_busy(starget) || scsi_device_is_busy(sdev))
1356		return 1;
1357
1358	return 0;
1359}
1360
1361/*
1362 * Kill a request for a dead device
1363 */
1364static void scsi_kill_request(struct request *req, struct request_queue *q)
1365{
1366	struct scsi_cmnd *cmd = req->special;
1367	struct scsi_device *sdev;
1368	struct scsi_target *starget;
1369	struct Scsi_Host *shost;
1370
1371	blk_start_request(req);
1372
1373	sdev = cmd->device;
1374	starget = scsi_target(sdev);
1375	shost = sdev->host;
1376	scsi_init_cmd_errh(cmd);
1377	cmd->result = DID_NO_CONNECT << 16;
1378	atomic_inc(&cmd->device->iorequest_cnt);
1379
1380	/*
1381	 * SCSI request completion path will do scsi_device_unbusy(),
1382	 * bump busy counts.  To bump the counters, we need to dance
1383	 * with the locks as normal issue path does.
1384	 */
1385	sdev->device_busy++;
1386	spin_unlock(sdev->request_queue->queue_lock);
1387	spin_lock(shost->host_lock);
1388	shost->host_busy++;
1389	starget->target_busy++;
1390	spin_unlock(shost->host_lock);
1391	spin_lock(sdev->request_queue->queue_lock);
1392
1393	blk_complete_request(req);
1394}
1395
1396static void scsi_softirq_done(struct request *rq)
1397{
1398	struct scsi_cmnd *cmd = rq->special;
1399	unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1400	int disposition;
1401
1402	INIT_LIST_HEAD(&cmd->eh_entry);
1403
1404	/*
1405	 * Set the serial numbers back to zero
1406	 */
1407	cmd->serial_number = 0;
1408
1409	atomic_inc(&cmd->device->iodone_cnt);
1410	if (cmd->result)
1411		atomic_inc(&cmd->device->ioerr_cnt);
1412
1413	disposition = scsi_decide_disposition(cmd);
1414	if (disposition != SUCCESS &&
1415	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1416		sdev_printk(KERN_ERR, cmd->device,
1417			    "timing out command, waited %lus\n",
1418			    wait_for/HZ);
1419		disposition = SUCCESS;
1420	}
1421
1422	scsi_log_completion(cmd, disposition);
1423
1424	switch (disposition) {
1425		case SUCCESS:
1426			scsi_finish_command(cmd);
1427			break;
1428		case NEEDS_RETRY:
1429			scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1430			break;
1431		case ADD_TO_MLQUEUE:
1432			scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1433			break;
1434		default:
1435			if (!scsi_eh_scmd_add(cmd, 0))
1436				scsi_finish_command(cmd);
1437	}
1438}
1439
1440/*
1441 * Function:    scsi_request_fn()
1442 *
1443 * Purpose:     Main strategy routine for SCSI.
1444 *
1445 * Arguments:   q       - Pointer to actual queue.
1446 *
1447 * Returns:     Nothing
1448 *
1449 * Lock status: IO request lock assumed to be held when called.
1450 */
1451static void scsi_request_fn(struct request_queue *q)
1452{
1453	struct scsi_device *sdev = q->queuedata;
1454	struct Scsi_Host *shost;
1455	struct scsi_cmnd *cmd;
1456	struct request *req;
1457
1458	if (!sdev) {
1459		printk("scsi: killing requests for dead queue\n");
1460		while ((req = blk_peek_request(q)) != NULL)
1461			scsi_kill_request(req, q);
1462		return;
1463	}
1464
1465	if(!get_device(&sdev->sdev_gendev))
1466		/* We must be tearing the block queue down already */
1467		return;
1468
1469	/*
1470	 * To start with, we keep looping until the queue is empty, or until
1471	 * the host is no longer able to accept any more requests.
1472	 */
1473	shost = sdev->host;
1474	while (!blk_queue_plugged(q)) {
1475		int rtn;
1476		/*
1477		 * get next queueable request.  We do this early to make sure
1478		 * that the request is fully prepared even if we cannot
1479		 * accept it.
1480		 */
1481		req = blk_peek_request(q);
1482		if (!req || !scsi_dev_queue_ready(q, sdev))
1483			break;
1484
1485		if (unlikely(!scsi_device_online(sdev))) {
1486			sdev_printk(KERN_ERR, sdev,
1487				    "rejecting I/O to offline device\n");
1488			scsi_kill_request(req, q);
1489			continue;
1490		}
1491
1492
1493		/*
1494		 * Remove the request from the request list.
1495		 */
1496		if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1497			blk_start_request(req);
1498		sdev->device_busy++;
1499
1500		spin_unlock(q->queue_lock);
1501		cmd = req->special;
1502		if (unlikely(cmd == NULL)) {
1503			printk(KERN_CRIT "impossible request in %s.\n"
1504					 "please mail a stack trace to "
1505					 "linux-scsi@vger.kernel.org\n",
1506					 __func__);
1507			blk_dump_rq_flags(req, "foo");
1508			BUG();
1509		}
1510		spin_lock(shost->host_lock);
1511
1512		/*
1513		 * We hit this when the driver is using a host wide
1514		 * tag map. For device level tag maps the queue_depth check
1515		 * in the device ready fn would prevent us from trying
1516		 * to allocate a tag. Since the map is a shared host resource
1517		 * we add the dev to the starved list so it eventually gets
1518		 * a run when a tag is freed.
1519		 */
1520		if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1521			if (list_empty(&sdev->starved_entry))
1522				list_add_tail(&sdev->starved_entry,
1523					      &shost->starved_list);
1524			goto not_ready;
1525		}
1526
1527		if (!scsi_target_queue_ready(shost, sdev))
1528			goto not_ready;
1529
1530		if (!scsi_host_queue_ready(q, shost, sdev))
1531			goto not_ready;
1532
1533		scsi_target(sdev)->target_busy++;
1534		shost->host_busy++;
1535
1536		spin_unlock_irq(shost->host_lock);
1537
1538		/*
1539		 * Finally, initialize any error handling parameters, and set up
1540		 * the timers for timeouts.
1541		 */
1542		scsi_init_cmd_errh(cmd);
1543
1544		/*
1545		 * Dispatch the command to the low-level driver.
1546		 */
1547		rtn = scsi_dispatch_cmd(cmd);
1548		spin_lock_irq(q->queue_lock);
1549		if(rtn) {
1550			/* we're refusing the command; because of
1551			 * the way locks get dropped, we need to
1552			 * check here if plugging is required */
1553			if(sdev->device_busy == 0)
1554				blk_plug_device(q);
1555
1556			break;
1557		}
1558	}
1559
1560	goto out;
1561
1562 not_ready:
1563	spin_unlock_irq(shost->host_lock);
1564
1565	/*
1566	 * lock q, handle tag, requeue req, and decrement device_busy. We
1567	 * must return with queue_lock held.
1568	 *
1569	 * Decrementing device_busy without checking it is OK, as all such
1570	 * cases (host limits or settings) should run the queue at some
1571	 * later time.
1572	 */
1573	spin_lock_irq(q->queue_lock);
1574	blk_requeue_request(q, req);
1575	sdev->device_busy--;
1576	if(sdev->device_busy == 0)
1577		blk_plug_device(q);
1578 out:
1579	/* must be careful here...if we trigger the ->remove() function
1580	 * we cannot be holding the q lock */
1581	spin_unlock_irq(q->queue_lock);
1582	put_device(&sdev->sdev_gendev);
1583	spin_lock_irq(q->queue_lock);
1584}
1585
1586u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1587{
1588	struct device *host_dev;
1589	u64 bounce_limit = 0xffffffff;
1590
1591	if (shost->unchecked_isa_dma)
1592		return BLK_BOUNCE_ISA;
1593	/*
1594	 * Platforms with virtual-DMA translation
1595	 * hardware have no practical limit.
1596	 */
1597	if (!PCI_DMA_BUS_IS_PHYS)
1598		return BLK_BOUNCE_ANY;
1599
1600	host_dev = scsi_get_device(shost);
1601	if (host_dev && host_dev->dma_mask)
1602		bounce_limit = *host_dev->dma_mask;
1603
1604	return bounce_limit;
1605}
1606EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1607
1608struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1609					 request_fn_proc *request_fn)
1610{
1611	struct request_queue *q;
1612	struct device *dev = shost->shost_gendev.parent;
1613
1614	q = blk_init_queue(request_fn, NULL);
1615	if (!q)
1616		return NULL;
1617
1618	/*
1619	 * this limit is imposed by hardware restrictions
1620	 */
1621	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1622					SCSI_MAX_SG_CHAIN_SEGMENTS));
1623
1624	blk_queue_max_hw_sectors(q, shost->max_sectors);
1625	blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1626	blk_queue_segment_boundary(q, shost->dma_boundary);
1627	dma_set_seg_boundary(dev, shost->dma_boundary);
1628
1629	blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1630
1631	if (!shost->use_clustering)
1632		q->limits.cluster = 0;
1633
1634	/*
1635	 * set a reasonable default alignment on word boundaries: the
1636	 * host and device may alter it using
1637	 * blk_queue_update_dma_alignment() later.
1638	 */
1639	blk_queue_dma_alignment(q, 0x03);
1640
1641	return q;
1642}
1643EXPORT_SYMBOL(__scsi_alloc_queue);
1644
1645struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1646{
1647	struct request_queue *q;
1648
1649	q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1650	if (!q)
1651		return NULL;
1652
1653	blk_queue_prep_rq(q, scsi_prep_fn);
1654	blk_queue_softirq_done(q, scsi_softirq_done);
1655	blk_queue_rq_timed_out(q, scsi_times_out);
1656	blk_queue_lld_busy(q, scsi_lld_busy);
1657	return q;
1658}
1659
1660void scsi_free_queue(struct request_queue *q)
1661{
1662	blk_cleanup_queue(q);
1663}
1664
1665/*
1666 * Function:    scsi_block_requests()
1667 *
1668 * Purpose:     Utility function used by low-level drivers to prevent further
1669 *		commands from being queued to the device.
1670 *
1671 * Arguments:   shost       - Host in question
1672 *
1673 * Returns:     Nothing
1674 *
1675 * Lock status: No locks are assumed held.
1676 *
1677 * Notes:       There is no timer nor any other means by which the requests
1678 *		get unblocked other than the low-level driver calling
1679 *		scsi_unblock_requests().
1680 */
1681void scsi_block_requests(struct Scsi_Host *shost)
1682{
1683	shost->host_self_blocked = 1;
1684}
1685EXPORT_SYMBOL(scsi_block_requests);
1686
1687/*
1688 * Function:    scsi_unblock_requests()
1689 *
1690 * Purpose:     Utility function used by low-level drivers to allow further
1691 *		commands from being queued to the device.
1692 *
1693 * Arguments:   shost       - Host in question
1694 *
1695 * Returns:     Nothing
1696 *
1697 * Lock status: No locks are assumed held.
1698 *
1699 * Notes:       There is no timer nor any other means by which the requests
1700 *		get unblocked other than the low-level driver calling
1701 *		scsi_unblock_requests().
1702 *
1703 *		This is done as an API function so that changes to the
1704 *		internals of the scsi mid-layer won't require wholesale
1705 *		changes to drivers that use this feature.
1706 */
1707void scsi_unblock_requests(struct Scsi_Host *shost)
1708{
1709	shost->host_self_blocked = 0;
1710	scsi_run_host_queues(shost);
1711}
1712EXPORT_SYMBOL(scsi_unblock_requests);
1713
1714int __init scsi_init_queue(void)
1715{
1716	int i;
1717
1718	scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1719					   sizeof(struct scsi_data_buffer),
1720					   0, 0, NULL);
1721	if (!scsi_sdb_cache) {
1722		printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1723		return -ENOMEM;
1724	}
1725
1726	for (i = 0; i < SG_MEMPOOL_NR; i++) {
1727		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1728		int size = sgp->size * sizeof(struct scatterlist);
1729
1730		sgp->slab = kmem_cache_create(sgp->name, size, 0,
1731				SLAB_HWCACHE_ALIGN, NULL);
1732		if (!sgp->slab) {
1733			printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1734					sgp->name);
1735			goto cleanup_sdb;
1736		}
1737
1738		sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1739						     sgp->slab);
1740		if (!sgp->pool) {
1741			printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1742					sgp->name);
1743			goto cleanup_sdb;
1744		}
1745	}
1746
1747	return 0;
1748
1749cleanup_sdb:
1750	for (i = 0; i < SG_MEMPOOL_NR; i++) {
1751		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1752		if (sgp->pool)
1753			mempool_destroy(sgp->pool);
1754		if (sgp->slab)
1755			kmem_cache_destroy(sgp->slab);
1756	}
1757	kmem_cache_destroy(scsi_sdb_cache);
1758
1759	return -ENOMEM;
1760}
1761
1762void scsi_exit_queue(void)
1763{
1764	int i;
1765
1766	kmem_cache_destroy(scsi_sdb_cache);
1767
1768	for (i = 0; i < SG_MEMPOOL_NR; i++) {
1769		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1770		mempool_destroy(sgp->pool);
1771		kmem_cache_destroy(sgp->slab);
1772	}
1773}
1774
1775/**
1776 *	scsi_mode_select - issue a mode select
1777 *	@sdev:	SCSI device to be queried
1778 *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
1779 *	@sp:	Save page bit (0 == don't save, 1 == save)
1780 *	@modepage: mode page being requested
1781 *	@buffer: request buffer (may not be smaller than eight bytes)
1782 *	@len:	length of request buffer.
1783 *	@timeout: command timeout
1784 *	@retries: number of retries before failing
1785 *	@data: returns a structure abstracting the mode header data
1786 *	@sshdr: place to put sense data (or NULL if no sense to be collected).
1787 *		must be SCSI_SENSE_BUFFERSIZE big.
1788 *
1789 *	Returns zero if successful; negative error number or scsi
1790 *	status on error
1791 *
1792 */
1793int
1794scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1795		 unsigned char *buffer, int len, int timeout, int retries,
1796		 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1797{
1798	unsigned char cmd[10];
1799	unsigned char *real_buffer;
1800	int ret;
1801
1802	memset(cmd, 0, sizeof(cmd));
1803	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1804
1805	if (sdev->use_10_for_ms) {
1806		if (len > 65535)
1807			return -EINVAL;
1808		real_buffer = kmalloc(8 + len, GFP_KERNEL);
1809		if (!real_buffer)
1810			return -ENOMEM;
1811		memcpy(real_buffer + 8, buffer, len);
1812		len += 8;
1813		real_buffer[0] = 0;
1814		real_buffer[1] = 0;
1815		real_buffer[2] = data->medium_type;
1816		real_buffer[3] = data->device_specific;
1817		real_buffer[4] = data->longlba ? 0x01 : 0;
1818		real_buffer[5] = 0;
1819		real_buffer[6] = data->block_descriptor_length >> 8;
1820		real_buffer[7] = data->block_descriptor_length;
1821
1822		cmd[0] = MODE_SELECT_10;
1823		cmd[7] = len >> 8;
1824		cmd[8] = len;
1825	} else {
1826		if (len > 255 || data->block_descriptor_length > 255 ||
1827		    data->longlba)
1828			return -EINVAL;
1829
1830		real_buffer = kmalloc(4 + len, GFP_KERNEL);
1831		if (!real_buffer)
1832			return -ENOMEM;
1833		memcpy(real_buffer + 4, buffer, len);
1834		len += 4;
1835		real_buffer[0] = 0;
1836		real_buffer[1] = data->medium_type;
1837		real_buffer[2] = data->device_specific;
1838		real_buffer[3] = data->block_descriptor_length;
1839
1840
1841		cmd[0] = MODE_SELECT;
1842		cmd[4] = len;
1843	}
1844
1845	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1846			       sshdr, timeout, retries, NULL);
1847	kfree(real_buffer);
1848	return ret;
1849}
1850EXPORT_SYMBOL_GPL(scsi_mode_select);
1851
1852/**
1853 *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1854 *	@sdev:	SCSI device to be queried
1855 *	@dbd:	set if mode sense will allow block descriptors to be returned
1856 *	@modepage: mode page being requested
1857 *	@buffer: request buffer (may not be smaller than eight bytes)
1858 *	@len:	length of request buffer.
1859 *	@timeout: command timeout
1860 *	@retries: number of retries before failing
1861 *	@data: returns a structure abstracting the mode header data
1862 *	@sshdr: place to put sense data (or NULL if no sense to be collected).
1863 *		must be SCSI_SENSE_BUFFERSIZE big.
1864 *
1865 *	Returns zero if unsuccessful, or the header offset (either 4
1866 *	or 8 depending on whether a six or ten byte command was
1867 *	issued) if successful.
1868 */
1869int
1870scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1871		  unsigned char *buffer, int len, int timeout, int retries,
1872		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1873{
1874	unsigned char cmd[12];
1875	int use_10_for_ms;
1876	int header_length;
1877	int result;
1878	struct scsi_sense_hdr my_sshdr;
1879
1880	memset(data, 0, sizeof(*data));
1881	memset(&cmd[0], 0, 12);
1882	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
1883	cmd[2] = modepage;
1884
1885	/* caller might not be interested in sense, but we need it */
1886	if (!sshdr)
1887		sshdr = &my_sshdr;
1888
1889 retry:
1890	use_10_for_ms = sdev->use_10_for_ms;
1891
1892	if (use_10_for_ms) {
1893		if (len < 8)
1894			len = 8;
1895
1896		cmd[0] = MODE_SENSE_10;
1897		cmd[8] = len;
1898		header_length = 8;
1899	} else {
1900		if (len < 4)
1901			len = 4;
1902
1903		cmd[0] = MODE_SENSE;
1904		cmd[4] = len;
1905		header_length = 4;
1906	}
1907
1908	memset(buffer, 0, len);
1909
1910	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1911				  sshdr, timeout, retries, NULL);
1912
1913	/* This code looks awful: what it's doing is making sure an
1914	 * ILLEGAL REQUEST sense return identifies the actual command
1915	 * byte as the problem.  MODE_SENSE commands can return
1916	 * ILLEGAL REQUEST if the code page isn't supported */
1917
1918	if (use_10_for_ms && !scsi_status_is_good(result) &&
1919	    (driver_byte(result) & DRIVER_SENSE)) {
1920		if (scsi_sense_valid(sshdr)) {
1921			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1922			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1923				/*
1924				 * Invalid command operation code
1925				 */
1926				sdev->use_10_for_ms = 0;
1927				goto retry;
1928			}
1929		}
1930	}
1931
1932	if(scsi_status_is_good(result)) {
1933		if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1934			     (modepage == 6 || modepage == 8))) {
1935			/* Initio breakage? */
1936			header_length = 0;
1937			data->length = 13;
1938			data->medium_type = 0;
1939			data->device_specific = 0;
1940			data->longlba = 0;
1941			data->block_descriptor_length = 0;
1942		} else if(use_10_for_ms) {
1943			data->length = buffer[0]*256 + buffer[1] + 2;
1944			data->medium_type = buffer[2];
1945			data->device_specific = buffer[3];
1946			data->longlba = buffer[4] & 0x01;
1947			data->block_descriptor_length = buffer[6]*256
1948				+ buffer[7];
1949		} else {
1950			data->length = buffer[0] + 1;
1951			data->medium_type = buffer[1];
1952			data->device_specific = buffer[2];
1953			data->block_descriptor_length = buffer[3];
1954		}
1955		data->header_length = header_length;
1956	}
1957
1958	return result;
1959}
1960EXPORT_SYMBOL(scsi_mode_sense);
1961
1962/**
1963 *	scsi_test_unit_ready - test if unit is ready
1964 *	@sdev:	scsi device to change the state of.
1965 *	@timeout: command timeout
1966 *	@retries: number of retries before failing
1967 *	@sshdr_external: Optional pointer to struct scsi_sense_hdr for
1968 *		returning sense. Make sure that this is cleared before passing
1969 *		in.
1970 *
1971 *	Returns zero if unsuccessful or an error if TUR failed.  For
1972 *	removable media, a return of NOT_READY or UNIT_ATTENTION is
1973 *	translated to success, with the ->changed flag updated.
1974 **/
1975int
1976scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
1977		     struct scsi_sense_hdr *sshdr_external)
1978{
1979	char cmd[] = {
1980		TEST_UNIT_READY, 0, 0, 0, 0, 0,
1981	};
1982	struct scsi_sense_hdr *sshdr;
1983	int result;
1984
1985	if (!sshdr_external)
1986		sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
1987	else
1988		sshdr = sshdr_external;
1989
1990	/* try to eat the UNIT_ATTENTION if there are enough retries */
1991	do {
1992		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
1993					  timeout, retries, NULL);
1994		if (sdev->removable && scsi_sense_valid(sshdr) &&
1995		    sshdr->sense_key == UNIT_ATTENTION)
1996			sdev->changed = 1;
1997	} while (scsi_sense_valid(sshdr) &&
1998		 sshdr->sense_key == UNIT_ATTENTION && --retries);
1999
2000	if (!sshdr)
2001		/* could not allocate sense buffer, so can't process it */
2002		return result;
2003
2004	if (sdev->removable && scsi_sense_valid(sshdr) &&
2005	    (sshdr->sense_key == UNIT_ATTENTION ||
2006	     sshdr->sense_key == NOT_READY)) {
2007		sdev->changed = 1;
2008		result = 0;
2009	}
2010	if (!sshdr_external)
2011		kfree(sshdr);
2012	return result;
2013}
2014EXPORT_SYMBOL(scsi_test_unit_ready);
2015
2016/**
2017 *	scsi_device_set_state - Take the given device through the device state model.
2018 *	@sdev:	scsi device to change the state of.
2019 *	@state:	state to change to.
2020 *
2021 *	Returns zero if unsuccessful or an error if the requested
2022 *	transition is illegal.
2023 */
2024int
2025scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2026{
2027	enum scsi_device_state oldstate = sdev->sdev_state;
2028
2029	if (state == oldstate)
2030		return 0;
2031
2032	switch (state) {
2033	case SDEV_CREATED:
2034		switch (oldstate) {
2035		case SDEV_CREATED_BLOCK:
2036			break;
2037		default:
2038			goto illegal;
2039		}
2040		break;
2041
2042	case SDEV_RUNNING:
2043		switch (oldstate) {
2044		case SDEV_CREATED:
2045		case SDEV_OFFLINE:
2046		case SDEV_QUIESCE:
2047		case SDEV_BLOCK:
2048			break;
2049		default:
2050			goto illegal;
2051		}
2052		break;
2053
2054	case SDEV_QUIESCE:
2055		switch (oldstate) {
2056		case SDEV_RUNNING:
2057		case SDEV_OFFLINE:
2058			break;
2059		default:
2060			goto illegal;
2061		}
2062		break;
2063
2064	case SDEV_OFFLINE:
2065		switch (oldstate) {
2066		case SDEV_CREATED:
2067		case SDEV_RUNNING:
2068		case SDEV_QUIESCE:
2069		case SDEV_BLOCK:
2070			break;
2071		default:
2072			goto illegal;
2073		}
2074		break;
2075
2076	case SDEV_BLOCK:
2077		switch (oldstate) {
2078		case SDEV_RUNNING:
2079		case SDEV_CREATED_BLOCK:
2080			break;
2081		default:
2082			goto illegal;
2083		}
2084		break;
2085
2086	case SDEV_CREATED_BLOCK:
2087		switch (oldstate) {
2088		case SDEV_CREATED:
2089			break;
2090		default:
2091			goto illegal;
2092		}
2093		break;
2094
2095	case SDEV_CANCEL:
2096		switch (oldstate) {
2097		case SDEV_CREATED:
2098		case SDEV_RUNNING:
2099		case SDEV_QUIESCE:
2100		case SDEV_OFFLINE:
2101		case SDEV_BLOCK:
2102			break;
2103		default:
2104			goto illegal;
2105		}
2106		break;
2107
2108	case SDEV_DEL:
2109		switch (oldstate) {
2110		case SDEV_CREATED:
2111		case SDEV_RUNNING:
2112		case SDEV_OFFLINE:
2113		case SDEV_CANCEL:
2114			break;
2115		default:
2116			goto illegal;
2117		}
2118		break;
2119
2120	}
2121	sdev->sdev_state = state;
2122	return 0;
2123
2124 illegal:
2125	SCSI_LOG_ERROR_RECOVERY(1,
2126				sdev_printk(KERN_ERR, sdev,
2127					    "Illegal state transition %s->%s\n",
2128					    scsi_device_state_name(oldstate),
2129					    scsi_device_state_name(state))
2130				);
2131	return -EINVAL;
2132}
2133EXPORT_SYMBOL(scsi_device_set_state);
2134
2135/**
2136 * 	sdev_evt_emit - emit a single SCSI device uevent
2137 *	@sdev: associated SCSI device
2138 *	@evt: event to emit
2139 *
2140 *	Send a single uevent (scsi_event) to the associated scsi_device.
2141 */
2142static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2143{
2144	int idx = 0;
2145	char *envp[3];
2146
2147	switch (evt->evt_type) {
2148	case SDEV_EVT_MEDIA_CHANGE:
2149		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2150		break;
2151
2152	default:
2153		/* do nothing */
2154		break;
2155	}
2156
2157	envp[idx++] = NULL;
2158
2159	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2160}
2161
2162/**
2163 * 	sdev_evt_thread - send a uevent for each scsi event
2164 *	@work: work struct for scsi_device
2165 *
2166 *	Dispatch queued events to their associated scsi_device kobjects
2167 *	as uevents.
2168 */
2169void scsi_evt_thread(struct work_struct *work)
2170{
2171	struct scsi_device *sdev;
2172	LIST_HEAD(event_list);
2173
2174	sdev = container_of(work, struct scsi_device, event_work);
2175
2176	while (1) {
2177		struct scsi_event *evt;
2178		struct list_head *this, *tmp;
2179		unsigned long flags;
2180
2181		spin_lock_irqsave(&sdev->list_lock, flags);
2182		list_splice_init(&sdev->event_list, &event_list);
2183		spin_unlock_irqrestore(&sdev->list_lock, flags);
2184
2185		if (list_empty(&event_list))
2186			break;
2187
2188		list_for_each_safe(this, tmp, &event_list) {
2189			evt = list_entry(this, struct scsi_event, node);
2190			list_del(&evt->node);
2191			scsi_evt_emit(sdev, evt);
2192			kfree(evt);
2193		}
2194	}
2195}
2196
2197/**
2198 * 	sdev_evt_send - send asserted event to uevent thread
2199 *	@sdev: scsi_device event occurred on
2200 *	@evt: event to send
2201 *
2202 *	Assert scsi device event asynchronously.
2203 */
2204void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2205{
2206	unsigned long flags;
2207
2208
2209	spin_lock_irqsave(&sdev->list_lock, flags);
2210	list_add_tail(&evt->node, &sdev->event_list);
2211	schedule_work(&sdev->event_work);
2212	spin_unlock_irqrestore(&sdev->list_lock, flags);
2213}
2214EXPORT_SYMBOL_GPL(sdev_evt_send);
2215
2216/**
2217 * 	sdev_evt_alloc - allocate a new scsi event
2218 *	@evt_type: type of event to allocate
2219 *	@gfpflags: GFP flags for allocation
2220 *
2221 *	Allocates and returns a new scsi_event.
2222 */
2223struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2224				  gfp_t gfpflags)
2225{
2226	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2227	if (!evt)
2228		return NULL;
2229
2230	evt->evt_type = evt_type;
2231	INIT_LIST_HEAD(&evt->node);
2232
2233	/* evt_type-specific initialization, if any */
2234	switch (evt_type) {
2235	case SDEV_EVT_MEDIA_CHANGE:
2236	default:
2237		/* do nothing */
2238		break;
2239	}
2240
2241	return evt;
2242}
2243EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2244
2245/**
2246 * 	sdev_evt_send_simple - send asserted event to uevent thread
2247 *	@sdev: scsi_device event occurred on
2248 *	@evt_type: type of event to send
2249 *	@gfpflags: GFP flags for allocation
2250 *
2251 *	Assert scsi device event asynchronously, given an event type.
2252 */
2253void sdev_evt_send_simple(struct scsi_device *sdev,
2254			  enum scsi_device_event evt_type, gfp_t gfpflags)
2255{
2256	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2257	if (!evt) {
2258		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2259			    evt_type);
2260		return;
2261	}
2262
2263	sdev_evt_send(sdev, evt);
2264}
2265EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2266
2267/**
2268 *	scsi_device_quiesce - Block user issued commands.
2269 *	@sdev:	scsi device to quiesce.
2270 *
2271 *	This works by trying to transition to the SDEV_QUIESCE state
2272 *	(which must be a legal transition).  When the device is in this
2273 *	state, only special requests will be accepted, all others will
2274 *	be deferred.  Since special requests may also be requeued requests,
2275 *	a successful return doesn't guarantee the device will be
2276 *	totally quiescent.
2277 *
2278 *	Must be called with user context, may sleep.
2279 *
2280 *	Returns zero if unsuccessful or an error if not.
2281 */
2282int
2283scsi_device_quiesce(struct scsi_device *sdev)
2284{
2285	int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2286	if (err)
2287		return err;
2288
2289	scsi_run_queue(sdev->request_queue);
2290	while (sdev->device_busy) {
2291		msleep_interruptible(200);
2292		scsi_run_queue(sdev->request_queue);
2293	}
2294	return 0;
2295}
2296EXPORT_SYMBOL(scsi_device_quiesce);
2297
2298/**
2299 *	scsi_device_resume - Restart user issued commands to a quiesced device.
2300 *	@sdev:	scsi device to resume.
2301 *
2302 *	Moves the device from quiesced back to running and restarts the
2303 *	queues.
2304 *
2305 *	Must be called with user context, may sleep.
2306 */
2307void
2308scsi_device_resume(struct scsi_device *sdev)
2309{
2310	if(scsi_device_set_state(sdev, SDEV_RUNNING))
2311		return;
2312	scsi_run_queue(sdev->request_queue);
2313}
2314EXPORT_SYMBOL(scsi_device_resume);
2315
2316static void
2317device_quiesce_fn(struct scsi_device *sdev, void *data)
2318{
2319	scsi_device_quiesce(sdev);
2320}
2321
2322void
2323scsi_target_quiesce(struct scsi_target *starget)
2324{
2325	starget_for_each_device(starget, NULL, device_quiesce_fn);
2326}
2327EXPORT_SYMBOL(scsi_target_quiesce);
2328
2329static void
2330device_resume_fn(struct scsi_device *sdev, void *data)
2331{
2332	scsi_device_resume(sdev);
2333}
2334
2335void
2336scsi_target_resume(struct scsi_target *starget)
2337{
2338	starget_for_each_device(starget, NULL, device_resume_fn);
2339}
2340EXPORT_SYMBOL(scsi_target_resume);
2341
2342/**
2343 * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2344 * @sdev:	device to block
2345 *
2346 * Block request made by scsi lld's to temporarily stop all
2347 * scsi commands on the specified device.  Called from interrupt
2348 * or normal process context.
2349 *
2350 * Returns zero if successful or error if not
2351 *
2352 * Notes:
2353 *	This routine transitions the device to the SDEV_BLOCK state
2354 *	(which must be a legal transition).  When the device is in this
2355 *	state, all commands are deferred until the scsi lld reenables
2356 *	the device with scsi_device_unblock or device_block_tmo fires.
2357 *	This routine assumes the host_lock is held on entry.
2358 */
2359int
2360scsi_internal_device_block(struct scsi_device *sdev)
2361{
2362	struct request_queue *q = sdev->request_queue;
2363	unsigned long flags;
2364	int err = 0;
2365
2366	err = scsi_device_set_state(sdev, SDEV_BLOCK);
2367	if (err) {
2368		err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2369
2370		if (err)
2371			return err;
2372	}
2373
2374	/*
2375	 * The device has transitioned to SDEV_BLOCK.  Stop the
2376	 * block layer from calling the midlayer with this device's
2377	 * request queue.
2378	 */
2379	spin_lock_irqsave(q->queue_lock, flags);
2380	blk_stop_queue(q);
2381	spin_unlock_irqrestore(q->queue_lock, flags);
2382
2383	return 0;
2384}
2385EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2386
2387/**
2388 * scsi_internal_device_unblock - resume a device after a block request
2389 * @sdev:	device to resume
2390 *
2391 * Called by scsi lld's or the midlayer to restart the device queue
2392 * for the previously suspended scsi device.  Called from interrupt or
2393 * normal process context.
2394 *
2395 * Returns zero if successful or error if not.
2396 *
2397 * Notes:
2398 *	This routine transitions the device to the SDEV_RUNNING state
2399 *	(which must be a legal transition) allowing the midlayer to
2400 *	goose the queue for this device.  This routine assumes the
2401 *	host_lock is held upon entry.
2402 */
2403int
2404scsi_internal_device_unblock(struct scsi_device *sdev)
2405{
2406	struct request_queue *q = sdev->request_queue;
2407	unsigned long flags;
2408
2409	/*
2410	 * Try to transition the scsi device to SDEV_RUNNING
2411	 * and goose the device queue if successful.
2412	 */
2413	if (sdev->sdev_state == SDEV_BLOCK)
2414		sdev->sdev_state = SDEV_RUNNING;
2415	else if (sdev->sdev_state == SDEV_CREATED_BLOCK)
2416		sdev->sdev_state = SDEV_CREATED;
2417	else if (sdev->sdev_state != SDEV_CANCEL &&
2418		 sdev->sdev_state != SDEV_OFFLINE)
2419		return -EINVAL;
2420
2421	spin_lock_irqsave(q->queue_lock, flags);
2422	blk_start_queue(q);
2423	spin_unlock_irqrestore(q->queue_lock, flags);
2424
2425	return 0;
2426}
2427EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2428
2429static void
2430device_block(struct scsi_device *sdev, void *data)
2431{
2432	scsi_internal_device_block(sdev);
2433}
2434
2435static int
2436target_block(struct device *dev, void *data)
2437{
2438	if (scsi_is_target_device(dev))
2439		starget_for_each_device(to_scsi_target(dev), NULL,
2440					device_block);
2441	return 0;
2442}
2443
2444void
2445scsi_target_block(struct device *dev)
2446{
2447	if (scsi_is_target_device(dev))
2448		starget_for_each_device(to_scsi_target(dev), NULL,
2449					device_block);
2450	else
2451		device_for_each_child(dev, NULL, target_block);
2452}
2453EXPORT_SYMBOL_GPL(scsi_target_block);
2454
2455static void
2456device_unblock(struct scsi_device *sdev, void *data)
2457{
2458	scsi_internal_device_unblock(sdev);
2459}
2460
2461static int
2462target_unblock(struct device *dev, void *data)
2463{
2464	if (scsi_is_target_device(dev))
2465		starget_for_each_device(to_scsi_target(dev), NULL,
2466					device_unblock);
2467	return 0;
2468}
2469
2470void
2471scsi_target_unblock(struct device *dev)
2472{
2473	if (scsi_is_target_device(dev))
2474		starget_for_each_device(to_scsi_target(dev), NULL,
2475					device_unblock);
2476	else
2477		device_for_each_child(dev, NULL, target_unblock);
2478}
2479EXPORT_SYMBOL_GPL(scsi_target_unblock);
2480
2481/**
2482 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2483 * @sgl:	scatter-gather list
2484 * @sg_count:	number of segments in sg
2485 * @offset:	offset in bytes into sg, on return offset into the mapped area
2486 * @len:	bytes to map, on return number of bytes mapped
2487 *
2488 * Returns virtual address of the start of the mapped page
2489 */
2490void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2491			  size_t *offset, size_t *len)
2492{
2493	int i;
2494	size_t sg_len = 0, len_complete = 0;
2495	struct scatterlist *sg;
2496	struct page *page;
2497
2498	WARN_ON(!irqs_disabled());
2499
2500	for_each_sg(sgl, sg, sg_count, i) {
2501		len_complete = sg_len; /* Complete sg-entries */
2502		sg_len += sg->length;
2503		if (sg_len > *offset)
2504			break;
2505	}
2506
2507	if (unlikely(i == sg_count)) {
2508		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2509			"elements %d\n",
2510		       __func__, sg_len, *offset, sg_count);
2511		WARN_ON(1);
2512		return NULL;
2513	}
2514
2515	/* Offset starting from the beginning of first page in this sg-entry */
2516	*offset = *offset - len_complete + sg->offset;
2517
2518	/* Assumption: contiguous pages can be accessed as "page + i" */
2519	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2520	*offset &= ~PAGE_MASK;
2521
2522	/* Bytes in this sg-entry from *offset to the end of the page */
2523	sg_len = PAGE_SIZE - *offset;
2524	if (*len > sg_len)
2525		*len = sg_len;
2526
2527	return kmap_atomic(page, KM_BIO_SRC_IRQ);
2528}
2529EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2530
2531/**
2532 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2533 * @virt:	virtual address to be unmapped
2534 */
2535void scsi_kunmap_atomic_sg(void *virt)
2536{
2537	kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2538}
2539EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2540