xref: /openbmc/linux/drivers/scsi/scsi_lib.c (revision 9c1f8594)
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 
38 struct 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
49 static 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 
68 struct kmem_cache *scsi_sdb_cache;
69 
70 /*
71  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
72  * not change behaviour from the previous unplug mechanism, experimentation
73  * may prove this needs changing.
74  */
75 #define SCSI_QUEUE_DELAY	3
76 
77 /*
78  * Function:	scsi_unprep_request()
79  *
80  * Purpose:	Remove all preparation done for a request, including its
81  *		associated scsi_cmnd, so that it can be requeued.
82  *
83  * Arguments:	req	- request to unprepare
84  *
85  * Lock status:	Assumed that no locks are held upon entry.
86  *
87  * Returns:	Nothing.
88  */
89 static void scsi_unprep_request(struct request *req)
90 {
91 	struct scsi_cmnd *cmd = req->special;
92 
93 	blk_unprep_request(req);
94 	req->special = NULL;
95 
96 	scsi_put_command(cmd);
97 }
98 
99 /**
100  * __scsi_queue_insert - private queue insertion
101  * @cmd: The SCSI command being requeued
102  * @reason:  The reason for the requeue
103  * @unbusy: Whether the queue should be unbusied
104  *
105  * This is a private queue insertion.  The public interface
106  * scsi_queue_insert() always assumes the queue should be unbusied
107  * because it's always called before the completion.  This function is
108  * for a requeue after completion, which should only occur in this
109  * file.
110  */
111 static int __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, int unbusy)
112 {
113 	struct Scsi_Host *host = cmd->device->host;
114 	struct scsi_device *device = cmd->device;
115 	struct scsi_target *starget = scsi_target(device);
116 	struct request_queue *q = device->request_queue;
117 	unsigned long flags;
118 
119 	SCSI_LOG_MLQUEUE(1,
120 		 printk("Inserting command %p into mlqueue\n", cmd));
121 
122 	/*
123 	 * Set the appropriate busy bit for the device/host.
124 	 *
125 	 * If the host/device isn't busy, assume that something actually
126 	 * completed, and that we should be able to queue a command now.
127 	 *
128 	 * Note that the prior mid-layer assumption that any host could
129 	 * always queue at least one command is now broken.  The mid-layer
130 	 * will implement a user specifiable stall (see
131 	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
132 	 * if a command is requeued with no other commands outstanding
133 	 * either for the device or for the host.
134 	 */
135 	switch (reason) {
136 	case SCSI_MLQUEUE_HOST_BUSY:
137 		host->host_blocked = host->max_host_blocked;
138 		break;
139 	case SCSI_MLQUEUE_DEVICE_BUSY:
140 	case SCSI_MLQUEUE_EH_RETRY:
141 		device->device_blocked = device->max_device_blocked;
142 		break;
143 	case SCSI_MLQUEUE_TARGET_BUSY:
144 		starget->target_blocked = starget->max_target_blocked;
145 		break;
146 	}
147 
148 	/*
149 	 * Decrement the counters, since these commands are no longer
150 	 * active on the host/device.
151 	 */
152 	if (unbusy)
153 		scsi_device_unbusy(device);
154 
155 	/*
156 	 * Requeue this command.  It will go before all other commands
157 	 * that are already in the queue.
158 	 */
159 	spin_lock_irqsave(q->queue_lock, flags);
160 	blk_requeue_request(q, cmd->request);
161 	spin_unlock_irqrestore(q->queue_lock, flags);
162 
163 	kblockd_schedule_work(q, &device->requeue_work);
164 
165 	return 0;
166 }
167 
168 /*
169  * Function:    scsi_queue_insert()
170  *
171  * Purpose:     Insert a command in the midlevel queue.
172  *
173  * Arguments:   cmd    - command that we are adding to queue.
174  *              reason - why we are inserting command to queue.
175  *
176  * Lock status: Assumed that lock is not held upon entry.
177  *
178  * Returns:     Nothing.
179  *
180  * Notes:       We do this for one of two cases.  Either the host is busy
181  *              and it cannot accept any more commands for the time being,
182  *              or the device returned QUEUE_FULL and can accept no more
183  *              commands.
184  * Notes:       This could be called either from an interrupt context or a
185  *              normal process context.
186  */
187 int scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
188 {
189 	return __scsi_queue_insert(cmd, reason, 1);
190 }
191 /**
192  * scsi_execute - insert request and wait for the result
193  * @sdev:	scsi device
194  * @cmd:	scsi command
195  * @data_direction: data direction
196  * @buffer:	data buffer
197  * @bufflen:	len of buffer
198  * @sense:	optional sense buffer
199  * @timeout:	request timeout in seconds
200  * @retries:	number of times to retry request
201  * @flags:	or into request flags;
202  * @resid:	optional residual length
203  *
204  * returns the req->errors value which is the scsi_cmnd result
205  * field.
206  */
207 int scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
208 		 int data_direction, void *buffer, unsigned bufflen,
209 		 unsigned char *sense, int timeout, int retries, int flags,
210 		 int *resid)
211 {
212 	struct request *req;
213 	int write = (data_direction == DMA_TO_DEVICE);
214 	int ret = DRIVER_ERROR << 24;
215 
216 	req = blk_get_request(sdev->request_queue, write, __GFP_WAIT);
217 	if (!req)
218 		return ret;
219 
220 	if (bufflen &&	blk_rq_map_kern(sdev->request_queue, req,
221 					buffer, bufflen, __GFP_WAIT))
222 		goto out;
223 
224 	req->cmd_len = COMMAND_SIZE(cmd[0]);
225 	memcpy(req->cmd, cmd, req->cmd_len);
226 	req->sense = sense;
227 	req->sense_len = 0;
228 	req->retries = retries;
229 	req->timeout = timeout;
230 	req->cmd_type = REQ_TYPE_BLOCK_PC;
231 	req->cmd_flags |= flags | REQ_QUIET | REQ_PREEMPT;
232 
233 	/*
234 	 * head injection *required* here otherwise quiesce won't work
235 	 */
236 	blk_execute_rq(req->q, NULL, req, 1);
237 
238 	/*
239 	 * Some devices (USB mass-storage in particular) may transfer
240 	 * garbage data together with a residue indicating that the data
241 	 * is invalid.  Prevent the garbage from being misinterpreted
242 	 * and prevent security leaks by zeroing out the excess data.
243 	 */
244 	if (unlikely(req->resid_len > 0 && req->resid_len <= bufflen))
245 		memset(buffer + (bufflen - req->resid_len), 0, req->resid_len);
246 
247 	if (resid)
248 		*resid = req->resid_len;
249 	ret = req->errors;
250  out:
251 	blk_put_request(req);
252 
253 	return ret;
254 }
255 EXPORT_SYMBOL(scsi_execute);
256 
257 
258 int scsi_execute_req(struct scsi_device *sdev, const unsigned char *cmd,
259 		     int data_direction, void *buffer, unsigned bufflen,
260 		     struct scsi_sense_hdr *sshdr, int timeout, int retries,
261 		     int *resid)
262 {
263 	char *sense = NULL;
264 	int result;
265 
266 	if (sshdr) {
267 		sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_NOIO);
268 		if (!sense)
269 			return DRIVER_ERROR << 24;
270 	}
271 	result = scsi_execute(sdev, cmd, data_direction, buffer, bufflen,
272 			      sense, timeout, retries, 0, resid);
273 	if (sshdr)
274 		scsi_normalize_sense(sense, SCSI_SENSE_BUFFERSIZE, sshdr);
275 
276 	kfree(sense);
277 	return result;
278 }
279 EXPORT_SYMBOL(scsi_execute_req);
280 
281 /*
282  * Function:    scsi_init_cmd_errh()
283  *
284  * Purpose:     Initialize cmd fields related to error handling.
285  *
286  * Arguments:   cmd	- command that is ready to be queued.
287  *
288  * Notes:       This function has the job of initializing a number of
289  *              fields related to error handling.   Typically this will
290  *              be called once for each command, as required.
291  */
292 static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
293 {
294 	cmd->serial_number = 0;
295 	scsi_set_resid(cmd, 0);
296 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
297 	if (cmd->cmd_len == 0)
298 		cmd->cmd_len = scsi_command_size(cmd->cmnd);
299 }
300 
301 void scsi_device_unbusy(struct scsi_device *sdev)
302 {
303 	struct Scsi_Host *shost = sdev->host;
304 	struct scsi_target *starget = scsi_target(sdev);
305 	unsigned long flags;
306 
307 	spin_lock_irqsave(shost->host_lock, flags);
308 	shost->host_busy--;
309 	starget->target_busy--;
310 	if (unlikely(scsi_host_in_recovery(shost) &&
311 		     (shost->host_failed || shost->host_eh_scheduled)))
312 		scsi_eh_wakeup(shost);
313 	spin_unlock(shost->host_lock);
314 	spin_lock(sdev->request_queue->queue_lock);
315 	sdev->device_busy--;
316 	spin_unlock_irqrestore(sdev->request_queue->queue_lock, flags);
317 }
318 
319 /*
320  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
321  * and call blk_run_queue for all the scsi_devices on the target -
322  * including current_sdev first.
323  *
324  * Called with *no* scsi locks held.
325  */
326 static void scsi_single_lun_run(struct scsi_device *current_sdev)
327 {
328 	struct Scsi_Host *shost = current_sdev->host;
329 	struct scsi_device *sdev, *tmp;
330 	struct scsi_target *starget = scsi_target(current_sdev);
331 	unsigned long flags;
332 
333 	spin_lock_irqsave(shost->host_lock, flags);
334 	starget->starget_sdev_user = NULL;
335 	spin_unlock_irqrestore(shost->host_lock, flags);
336 
337 	/*
338 	 * Call blk_run_queue for all LUNs on the target, starting with
339 	 * current_sdev. We race with others (to set starget_sdev_user),
340 	 * but in most cases, we will be first. Ideally, each LU on the
341 	 * target would get some limited time or requests on the target.
342 	 */
343 	blk_run_queue(current_sdev->request_queue);
344 
345 	spin_lock_irqsave(shost->host_lock, flags);
346 	if (starget->starget_sdev_user)
347 		goto out;
348 	list_for_each_entry_safe(sdev, tmp, &starget->devices,
349 			same_target_siblings) {
350 		if (sdev == current_sdev)
351 			continue;
352 		if (scsi_device_get(sdev))
353 			continue;
354 
355 		spin_unlock_irqrestore(shost->host_lock, flags);
356 		blk_run_queue(sdev->request_queue);
357 		spin_lock_irqsave(shost->host_lock, flags);
358 
359 		scsi_device_put(sdev);
360 	}
361  out:
362 	spin_unlock_irqrestore(shost->host_lock, flags);
363 }
364 
365 static inline int scsi_device_is_busy(struct scsi_device *sdev)
366 {
367 	if (sdev->device_busy >= sdev->queue_depth || sdev->device_blocked)
368 		return 1;
369 
370 	return 0;
371 }
372 
373 static inline int scsi_target_is_busy(struct scsi_target *starget)
374 {
375 	return ((starget->can_queue > 0 &&
376 		 starget->target_busy >= starget->can_queue) ||
377 		 starget->target_blocked);
378 }
379 
380 static inline int scsi_host_is_busy(struct Scsi_Host *shost)
381 {
382 	if ((shost->can_queue > 0 && shost->host_busy >= shost->can_queue) ||
383 	    shost->host_blocked || shost->host_self_blocked)
384 		return 1;
385 
386 	return 0;
387 }
388 
389 /*
390  * Function:	scsi_run_queue()
391  *
392  * Purpose:	Select a proper request queue to serve next
393  *
394  * Arguments:	q	- last request's queue
395  *
396  * Returns:     Nothing
397  *
398  * Notes:	The previous command was completely finished, start
399  *		a new one if possible.
400  */
401 static void scsi_run_queue(struct request_queue *q)
402 {
403 	struct scsi_device *sdev = q->queuedata;
404 	struct Scsi_Host *shost;
405 	LIST_HEAD(starved_list);
406 	unsigned long flags;
407 
408 	/* if the device is dead, sdev will be NULL, so no queue to run */
409 	if (!sdev)
410 		return;
411 
412 	shost = sdev->host;
413 	if (scsi_target(sdev)->single_lun)
414 		scsi_single_lun_run(sdev);
415 
416 	spin_lock_irqsave(shost->host_lock, flags);
417 	list_splice_init(&shost->starved_list, &starved_list);
418 
419 	while (!list_empty(&starved_list)) {
420 		/*
421 		 * As long as shost is accepting commands and we have
422 		 * starved queues, call blk_run_queue. scsi_request_fn
423 		 * drops the queue_lock and can add us back to the
424 		 * starved_list.
425 		 *
426 		 * host_lock protects the starved_list and starved_entry.
427 		 * scsi_request_fn must get the host_lock before checking
428 		 * or modifying starved_list or starved_entry.
429 		 */
430 		if (scsi_host_is_busy(shost))
431 			break;
432 
433 		sdev = list_entry(starved_list.next,
434 				  struct scsi_device, starved_entry);
435 		list_del_init(&sdev->starved_entry);
436 		if (scsi_target_is_busy(scsi_target(sdev))) {
437 			list_move_tail(&sdev->starved_entry,
438 				       &shost->starved_list);
439 			continue;
440 		}
441 
442 		spin_unlock(shost->host_lock);
443 		spin_lock(sdev->request_queue->queue_lock);
444 		__blk_run_queue(sdev->request_queue);
445 		spin_unlock(sdev->request_queue->queue_lock);
446 		spin_lock(shost->host_lock);
447 	}
448 	/* put any unprocessed entries back */
449 	list_splice(&starved_list, &shost->starved_list);
450 	spin_unlock_irqrestore(shost->host_lock, flags);
451 
452 	blk_run_queue(q);
453 }
454 
455 void scsi_requeue_run_queue(struct work_struct *work)
456 {
457 	struct scsi_device *sdev;
458 	struct request_queue *q;
459 
460 	sdev = container_of(work, struct scsi_device, requeue_work);
461 	q = sdev->request_queue;
462 	scsi_run_queue(q);
463 }
464 
465 /*
466  * Function:	scsi_requeue_command()
467  *
468  * Purpose:	Handle post-processing of completed commands.
469  *
470  * Arguments:	q	- queue to operate on
471  *		cmd	- command that may need to be requeued.
472  *
473  * Returns:	Nothing
474  *
475  * Notes:	After command completion, there may be blocks left
476  *		over which weren't finished by the previous command
477  *		this can be for a number of reasons - the main one is
478  *		I/O errors in the middle of the request, in which case
479  *		we need to request the blocks that come after the bad
480  *		sector.
481  * Notes:	Upon return, cmd is a stale pointer.
482  */
483 static void scsi_requeue_command(struct request_queue *q, struct scsi_cmnd *cmd)
484 {
485 	struct request *req = cmd->request;
486 	unsigned long flags;
487 
488 	spin_lock_irqsave(q->queue_lock, flags);
489 	scsi_unprep_request(req);
490 	blk_requeue_request(q, req);
491 	spin_unlock_irqrestore(q->queue_lock, flags);
492 
493 	scsi_run_queue(q);
494 }
495 
496 void scsi_next_command(struct scsi_cmnd *cmd)
497 {
498 	struct scsi_device *sdev = cmd->device;
499 	struct request_queue *q = sdev->request_queue;
500 
501 	/* need to hold a reference on the device before we let go of the cmd */
502 	get_device(&sdev->sdev_gendev);
503 
504 	scsi_put_command(cmd);
505 	scsi_run_queue(q);
506 
507 	/* ok to remove device now */
508 	put_device(&sdev->sdev_gendev);
509 }
510 
511 void scsi_run_host_queues(struct Scsi_Host *shost)
512 {
513 	struct scsi_device *sdev;
514 
515 	shost_for_each_device(sdev, shost)
516 		scsi_run_queue(sdev->request_queue);
517 }
518 
519 static void __scsi_release_buffers(struct scsi_cmnd *, int);
520 
521 /*
522  * Function:    scsi_end_request()
523  *
524  * Purpose:     Post-processing of completed commands (usually invoked at end
525  *		of upper level post-processing and scsi_io_completion).
526  *
527  * Arguments:   cmd	 - command that is complete.
528  *              error    - 0 if I/O indicates success, < 0 for I/O error.
529  *              bytes    - number of bytes of completed I/O
530  *		requeue  - indicates whether we should requeue leftovers.
531  *
532  * Lock status: Assumed that lock is not held upon entry.
533  *
534  * Returns:     cmd if requeue required, NULL otherwise.
535  *
536  * Notes:       This is called for block device requests in order to
537  *              mark some number of sectors as complete.
538  *
539  *		We are guaranteeing that the request queue will be goosed
540  *		at some point during this call.
541  * Notes:	If cmd was requeued, upon return it will be a stale pointer.
542  */
543 static struct scsi_cmnd *scsi_end_request(struct scsi_cmnd *cmd, int error,
544 					  int bytes, int requeue)
545 {
546 	struct request_queue *q = cmd->device->request_queue;
547 	struct request *req = cmd->request;
548 
549 	/*
550 	 * If there are blocks left over at the end, set up the command
551 	 * to queue the remainder of them.
552 	 */
553 	if (blk_end_request(req, error, bytes)) {
554 		/* kill remainder if no retrys */
555 		if (error && scsi_noretry_cmd(cmd))
556 			blk_end_request_all(req, error);
557 		else {
558 			if (requeue) {
559 				/*
560 				 * Bleah.  Leftovers again.  Stick the
561 				 * leftovers in the front of the
562 				 * queue, and goose the queue again.
563 				 */
564 				scsi_release_buffers(cmd);
565 				scsi_requeue_command(q, cmd);
566 				cmd = NULL;
567 			}
568 			return cmd;
569 		}
570 	}
571 
572 	/*
573 	 * This will goose the queue request function at the end, so we don't
574 	 * need to worry about launching another command.
575 	 */
576 	__scsi_release_buffers(cmd, 0);
577 	scsi_next_command(cmd);
578 	return NULL;
579 }
580 
581 static inline unsigned int scsi_sgtable_index(unsigned short nents)
582 {
583 	unsigned int index;
584 
585 	BUG_ON(nents > SCSI_MAX_SG_SEGMENTS);
586 
587 	if (nents <= 8)
588 		index = 0;
589 	else
590 		index = get_count_order(nents) - 3;
591 
592 	return index;
593 }
594 
595 static void scsi_sg_free(struct scatterlist *sgl, unsigned int nents)
596 {
597 	struct scsi_host_sg_pool *sgp;
598 
599 	sgp = scsi_sg_pools + scsi_sgtable_index(nents);
600 	mempool_free(sgl, sgp->pool);
601 }
602 
603 static struct scatterlist *scsi_sg_alloc(unsigned int nents, gfp_t gfp_mask)
604 {
605 	struct scsi_host_sg_pool *sgp;
606 
607 	sgp = scsi_sg_pools + scsi_sgtable_index(nents);
608 	return mempool_alloc(sgp->pool, gfp_mask);
609 }
610 
611 static int scsi_alloc_sgtable(struct scsi_data_buffer *sdb, int nents,
612 			      gfp_t gfp_mask)
613 {
614 	int ret;
615 
616 	BUG_ON(!nents);
617 
618 	ret = __sg_alloc_table(&sdb->table, nents, SCSI_MAX_SG_SEGMENTS,
619 			       gfp_mask, scsi_sg_alloc);
620 	if (unlikely(ret))
621 		__sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS,
622 				scsi_sg_free);
623 
624 	return ret;
625 }
626 
627 static void scsi_free_sgtable(struct scsi_data_buffer *sdb)
628 {
629 	__sg_free_table(&sdb->table, SCSI_MAX_SG_SEGMENTS, scsi_sg_free);
630 }
631 
632 static void __scsi_release_buffers(struct scsi_cmnd *cmd, int do_bidi_check)
633 {
634 
635 	if (cmd->sdb.table.nents)
636 		scsi_free_sgtable(&cmd->sdb);
637 
638 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
639 
640 	if (do_bidi_check && scsi_bidi_cmnd(cmd)) {
641 		struct scsi_data_buffer *bidi_sdb =
642 			cmd->request->next_rq->special;
643 		scsi_free_sgtable(bidi_sdb);
644 		kmem_cache_free(scsi_sdb_cache, bidi_sdb);
645 		cmd->request->next_rq->special = NULL;
646 	}
647 
648 	if (scsi_prot_sg_count(cmd))
649 		scsi_free_sgtable(cmd->prot_sdb);
650 }
651 
652 /*
653  * Function:    scsi_release_buffers()
654  *
655  * Purpose:     Completion processing for block device I/O requests.
656  *
657  * Arguments:   cmd	- command that we are bailing.
658  *
659  * Lock status: Assumed that no lock is held upon entry.
660  *
661  * Returns:     Nothing
662  *
663  * Notes:       In the event that an upper level driver rejects a
664  *		command, we must release resources allocated during
665  *		the __init_io() function.  Primarily this would involve
666  *		the scatter-gather table, and potentially any bounce
667  *		buffers.
668  */
669 void scsi_release_buffers(struct scsi_cmnd *cmd)
670 {
671 	__scsi_release_buffers(cmd, 1);
672 }
673 EXPORT_SYMBOL(scsi_release_buffers);
674 
675 static int __scsi_error_from_host_byte(struct scsi_cmnd *cmd, int result)
676 {
677 	int error = 0;
678 
679 	switch(host_byte(result)) {
680 	case DID_TRANSPORT_FAILFAST:
681 		error = -ENOLINK;
682 		break;
683 	case DID_TARGET_FAILURE:
684 		cmd->result |= (DID_OK << 16);
685 		error = -EREMOTEIO;
686 		break;
687 	case DID_NEXUS_FAILURE:
688 		cmd->result |= (DID_OK << 16);
689 		error = -EBADE;
690 		break;
691 	default:
692 		error = -EIO;
693 		break;
694 	}
695 
696 	return error;
697 }
698 
699 /*
700  * Function:    scsi_io_completion()
701  *
702  * Purpose:     Completion processing for block device I/O requests.
703  *
704  * Arguments:   cmd   - command that is finished.
705  *
706  * Lock status: Assumed that no lock is held upon entry.
707  *
708  * Returns:     Nothing
709  *
710  * Notes:       This function is matched in terms of capabilities to
711  *              the function that created the scatter-gather list.
712  *              In other words, if there are no bounce buffers
713  *              (the normal case for most drivers), we don't need
714  *              the logic to deal with cleaning up afterwards.
715  *
716  *		We must call scsi_end_request().  This will finish off
717  *		the specified number of sectors.  If we are done, the
718  *		command block will be released and the queue function
719  *		will be goosed.  If we are not done then we have to
720  *		figure out what to do next:
721  *
722  *		a) We can call scsi_requeue_command().  The request
723  *		   will be unprepared and put back on the queue.  Then
724  *		   a new command will be created for it.  This should
725  *		   be used if we made forward progress, or if we want
726  *		   to switch from READ(10) to READ(6) for example.
727  *
728  *		b) We can call scsi_queue_insert().  The request will
729  *		   be put back on the queue and retried using the same
730  *		   command as before, possibly after a delay.
731  *
732  *		c) We can call blk_end_request() with -EIO to fail
733  *		   the remainder of the request.
734  */
735 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
736 {
737 	int result = cmd->result;
738 	struct request_queue *q = cmd->device->request_queue;
739 	struct request *req = cmd->request;
740 	int error = 0;
741 	struct scsi_sense_hdr sshdr;
742 	int sense_valid = 0;
743 	int sense_deferred = 0;
744 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
745 	      ACTION_DELAYED_RETRY} action;
746 	char *description = NULL;
747 
748 	if (result) {
749 		sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
750 		if (sense_valid)
751 			sense_deferred = scsi_sense_is_deferred(&sshdr);
752 	}
753 
754 	if (req->cmd_type == REQ_TYPE_BLOCK_PC) { /* SG_IO ioctl from block level */
755 		req->errors = result;
756 		if (result) {
757 			if (sense_valid && req->sense) {
758 				/*
759 				 * SG_IO wants current and deferred errors
760 				 */
761 				int len = 8 + cmd->sense_buffer[7];
762 
763 				if (len > SCSI_SENSE_BUFFERSIZE)
764 					len = SCSI_SENSE_BUFFERSIZE;
765 				memcpy(req->sense, cmd->sense_buffer,  len);
766 				req->sense_len = len;
767 			}
768 			if (!sense_deferred)
769 				error = __scsi_error_from_host_byte(cmd, result);
770 		}
771 
772 		req->resid_len = scsi_get_resid(cmd);
773 
774 		if (scsi_bidi_cmnd(cmd)) {
775 			/*
776 			 * Bidi commands Must be complete as a whole,
777 			 * both sides at once.
778 			 */
779 			req->next_rq->resid_len = scsi_in(cmd)->resid;
780 
781 			scsi_release_buffers(cmd);
782 			blk_end_request_all(req, 0);
783 
784 			scsi_next_command(cmd);
785 			return;
786 		}
787 	}
788 
789 	/* no bidi support for !REQ_TYPE_BLOCK_PC yet */
790 	BUG_ON(blk_bidi_rq(req));
791 
792 	/*
793 	 * Next deal with any sectors which we were able to correctly
794 	 * handle.
795 	 */
796 	SCSI_LOG_HLCOMPLETE(1, printk("%u sectors total, "
797 				      "%d bytes done.\n",
798 				      blk_rq_sectors(req), good_bytes));
799 
800 	/*
801 	 * Recovered errors need reporting, but they're always treated
802 	 * as success, so fiddle the result code here.  For BLOCK_PC
803 	 * we already took a copy of the original into rq->errors which
804 	 * is what gets returned to the user
805 	 */
806 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
807 		/* if ATA PASS-THROUGH INFORMATION AVAILABLE skip
808 		 * print since caller wants ATA registers. Only occurs on
809 		 * SCSI ATA PASS_THROUGH commands when CK_COND=1
810 		 */
811 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
812 			;
813 		else if (!(req->cmd_flags & REQ_QUIET))
814 			scsi_print_sense("", cmd);
815 		result = 0;
816 		/* BLOCK_PC may have set error */
817 		error = 0;
818 	}
819 
820 	/*
821 	 * A number of bytes were successfully read.  If there
822 	 * are leftovers and there is some kind of error
823 	 * (result != 0), retry the rest.
824 	 */
825 	if (scsi_end_request(cmd, error, good_bytes, result == 0) == NULL)
826 		return;
827 
828 	error = __scsi_error_from_host_byte(cmd, result);
829 
830 	if (host_byte(result) == DID_RESET) {
831 		/* Third party bus reset or reset for error recovery
832 		 * reasons.  Just retry the command and see what
833 		 * happens.
834 		 */
835 		action = ACTION_RETRY;
836 	} else if (sense_valid && !sense_deferred) {
837 		switch (sshdr.sense_key) {
838 		case UNIT_ATTENTION:
839 			if (cmd->device->removable) {
840 				/* Detected disc change.  Set a bit
841 				 * and quietly refuse further access.
842 				 */
843 				cmd->device->changed = 1;
844 				description = "Media Changed";
845 				action = ACTION_FAIL;
846 			} else {
847 				/* Must have been a power glitch, or a
848 				 * bus reset.  Could not have been a
849 				 * media change, so we just retry the
850 				 * command and see what happens.
851 				 */
852 				action = ACTION_RETRY;
853 			}
854 			break;
855 		case ILLEGAL_REQUEST:
856 			/* If we had an ILLEGAL REQUEST returned, then
857 			 * we may have performed an unsupported
858 			 * command.  The only thing this should be
859 			 * would be a ten byte read where only a six
860 			 * byte read was supported.  Also, on a system
861 			 * where READ CAPACITY failed, we may have
862 			 * read past the end of the disk.
863 			 */
864 			if ((cmd->device->use_10_for_rw &&
865 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
866 			    (cmd->cmnd[0] == READ_10 ||
867 			     cmd->cmnd[0] == WRITE_10)) {
868 				/* This will issue a new 6-byte command. */
869 				cmd->device->use_10_for_rw = 0;
870 				action = ACTION_REPREP;
871 			} else if (sshdr.asc == 0x10) /* DIX */ {
872 				description = "Host Data Integrity Failure";
873 				action = ACTION_FAIL;
874 				error = -EILSEQ;
875 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
876 			} else if ((sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
877 				   (cmd->cmnd[0] == UNMAP ||
878 				    cmd->cmnd[0] == WRITE_SAME_16 ||
879 				    cmd->cmnd[0] == WRITE_SAME)) {
880 				description = "Discard failure";
881 				action = ACTION_FAIL;
882 			} else
883 				action = ACTION_FAIL;
884 			break;
885 		case ABORTED_COMMAND:
886 			action = ACTION_FAIL;
887 			if (sshdr.asc == 0x10) { /* DIF */
888 				description = "Target Data Integrity Failure";
889 				error = -EILSEQ;
890 			}
891 			break;
892 		case NOT_READY:
893 			/* If the device is in the process of becoming
894 			 * ready, or has a temporary blockage, retry.
895 			 */
896 			if (sshdr.asc == 0x04) {
897 				switch (sshdr.ascq) {
898 				case 0x01: /* becoming ready */
899 				case 0x04: /* format in progress */
900 				case 0x05: /* rebuild in progress */
901 				case 0x06: /* recalculation in progress */
902 				case 0x07: /* operation in progress */
903 				case 0x08: /* Long write in progress */
904 				case 0x09: /* self test in progress */
905 				case 0x14: /* space allocation in progress */
906 					action = ACTION_DELAYED_RETRY;
907 					break;
908 				default:
909 					description = "Device not ready";
910 					action = ACTION_FAIL;
911 					break;
912 				}
913 			} else {
914 				description = "Device not ready";
915 				action = ACTION_FAIL;
916 			}
917 			break;
918 		case VOLUME_OVERFLOW:
919 			/* See SSC3rXX or current. */
920 			action = ACTION_FAIL;
921 			break;
922 		default:
923 			description = "Unhandled sense code";
924 			action = ACTION_FAIL;
925 			break;
926 		}
927 	} else {
928 		description = "Unhandled error code";
929 		action = ACTION_FAIL;
930 	}
931 
932 	switch (action) {
933 	case ACTION_FAIL:
934 		/* Give up and fail the remainder of the request */
935 		scsi_release_buffers(cmd);
936 		if (!(req->cmd_flags & REQ_QUIET)) {
937 			if (description)
938 				scmd_printk(KERN_INFO, cmd, "%s\n",
939 					    description);
940 			scsi_print_result(cmd);
941 			if (driver_byte(result) & DRIVER_SENSE)
942 				scsi_print_sense("", cmd);
943 			scsi_print_command(cmd);
944 		}
945 		if (blk_end_request_err(req, error))
946 			scsi_requeue_command(q, cmd);
947 		else
948 			scsi_next_command(cmd);
949 		break;
950 	case ACTION_REPREP:
951 		/* Unprep the request and put it back at the head of the queue.
952 		 * A new command will be prepared and issued.
953 		 */
954 		scsi_release_buffers(cmd);
955 		scsi_requeue_command(q, cmd);
956 		break;
957 	case ACTION_RETRY:
958 		/* Retry the same command immediately */
959 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, 0);
960 		break;
961 	case ACTION_DELAYED_RETRY:
962 		/* Retry the same command after a delay */
963 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, 0);
964 		break;
965 	}
966 }
967 
968 static int scsi_init_sgtable(struct request *req, struct scsi_data_buffer *sdb,
969 			     gfp_t gfp_mask)
970 {
971 	int count;
972 
973 	/*
974 	 * If sg table allocation fails, requeue request later.
975 	 */
976 	if (unlikely(scsi_alloc_sgtable(sdb, req->nr_phys_segments,
977 					gfp_mask))) {
978 		return BLKPREP_DEFER;
979 	}
980 
981 	req->buffer = NULL;
982 
983 	/*
984 	 * Next, walk the list, and fill in the addresses and sizes of
985 	 * each segment.
986 	 */
987 	count = blk_rq_map_sg(req->q, req, sdb->table.sgl);
988 	BUG_ON(count > sdb->table.nents);
989 	sdb->table.nents = count;
990 	sdb->length = blk_rq_bytes(req);
991 	return BLKPREP_OK;
992 }
993 
994 /*
995  * Function:    scsi_init_io()
996  *
997  * Purpose:     SCSI I/O initialize function.
998  *
999  * Arguments:   cmd   - Command descriptor we wish to initialize
1000  *
1001  * Returns:     0 on success
1002  *		BLKPREP_DEFER if the failure is retryable
1003  *		BLKPREP_KILL if the failure is fatal
1004  */
1005 int scsi_init_io(struct scsi_cmnd *cmd, gfp_t gfp_mask)
1006 {
1007 	struct request *rq = cmd->request;
1008 
1009 	int error = scsi_init_sgtable(rq, &cmd->sdb, gfp_mask);
1010 	if (error)
1011 		goto err_exit;
1012 
1013 	if (blk_bidi_rq(rq)) {
1014 		struct scsi_data_buffer *bidi_sdb = kmem_cache_zalloc(
1015 			scsi_sdb_cache, GFP_ATOMIC);
1016 		if (!bidi_sdb) {
1017 			error = BLKPREP_DEFER;
1018 			goto err_exit;
1019 		}
1020 
1021 		rq->next_rq->special = bidi_sdb;
1022 		error = scsi_init_sgtable(rq->next_rq, bidi_sdb, GFP_ATOMIC);
1023 		if (error)
1024 			goto err_exit;
1025 	}
1026 
1027 	if (blk_integrity_rq(rq)) {
1028 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1029 		int ivecs, count;
1030 
1031 		BUG_ON(prot_sdb == NULL);
1032 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1033 
1034 		if (scsi_alloc_sgtable(prot_sdb, ivecs, gfp_mask)) {
1035 			error = BLKPREP_DEFER;
1036 			goto err_exit;
1037 		}
1038 
1039 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1040 						prot_sdb->table.sgl);
1041 		BUG_ON(unlikely(count > ivecs));
1042 		BUG_ON(unlikely(count > queue_max_integrity_segments(rq->q)));
1043 
1044 		cmd->prot_sdb = prot_sdb;
1045 		cmd->prot_sdb->table.nents = count;
1046 	}
1047 
1048 	return BLKPREP_OK ;
1049 
1050 err_exit:
1051 	scsi_release_buffers(cmd);
1052 	cmd->request->special = NULL;
1053 	scsi_put_command(cmd);
1054 	return error;
1055 }
1056 EXPORT_SYMBOL(scsi_init_io);
1057 
1058 static struct scsi_cmnd *scsi_get_cmd_from_req(struct scsi_device *sdev,
1059 		struct request *req)
1060 {
1061 	struct scsi_cmnd *cmd;
1062 
1063 	if (!req->special) {
1064 		cmd = scsi_get_command(sdev, GFP_ATOMIC);
1065 		if (unlikely(!cmd))
1066 			return NULL;
1067 		req->special = cmd;
1068 	} else {
1069 		cmd = req->special;
1070 	}
1071 
1072 	/* pull a tag out of the request if we have one */
1073 	cmd->tag = req->tag;
1074 	cmd->request = req;
1075 
1076 	cmd->cmnd = req->cmd;
1077 	cmd->prot_op = SCSI_PROT_NORMAL;
1078 
1079 	return cmd;
1080 }
1081 
1082 int scsi_setup_blk_pc_cmnd(struct scsi_device *sdev, struct request *req)
1083 {
1084 	struct scsi_cmnd *cmd;
1085 	int ret = scsi_prep_state_check(sdev, req);
1086 
1087 	if (ret != BLKPREP_OK)
1088 		return ret;
1089 
1090 	cmd = scsi_get_cmd_from_req(sdev, req);
1091 	if (unlikely(!cmd))
1092 		return BLKPREP_DEFER;
1093 
1094 	/*
1095 	 * BLOCK_PC requests may transfer data, in which case they must
1096 	 * a bio attached to them.  Or they might contain a SCSI command
1097 	 * that does not transfer data, in which case they may optionally
1098 	 * submit a request without an attached bio.
1099 	 */
1100 	if (req->bio) {
1101 		int ret;
1102 
1103 		BUG_ON(!req->nr_phys_segments);
1104 
1105 		ret = scsi_init_io(cmd, GFP_ATOMIC);
1106 		if (unlikely(ret))
1107 			return ret;
1108 	} else {
1109 		BUG_ON(blk_rq_bytes(req));
1110 
1111 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1112 		req->buffer = NULL;
1113 	}
1114 
1115 	cmd->cmd_len = req->cmd_len;
1116 	if (!blk_rq_bytes(req))
1117 		cmd->sc_data_direction = DMA_NONE;
1118 	else if (rq_data_dir(req) == WRITE)
1119 		cmd->sc_data_direction = DMA_TO_DEVICE;
1120 	else
1121 		cmd->sc_data_direction = DMA_FROM_DEVICE;
1122 
1123 	cmd->transfersize = blk_rq_bytes(req);
1124 	cmd->allowed = req->retries;
1125 	return BLKPREP_OK;
1126 }
1127 EXPORT_SYMBOL(scsi_setup_blk_pc_cmnd);
1128 
1129 /*
1130  * Setup a REQ_TYPE_FS command.  These are simple read/write request
1131  * from filesystems that still need to be translated to SCSI CDBs from
1132  * the ULD.
1133  */
1134 int scsi_setup_fs_cmnd(struct scsi_device *sdev, struct request *req)
1135 {
1136 	struct scsi_cmnd *cmd;
1137 	int ret = scsi_prep_state_check(sdev, req);
1138 
1139 	if (ret != BLKPREP_OK)
1140 		return ret;
1141 
1142 	if (unlikely(sdev->scsi_dh_data && sdev->scsi_dh_data->scsi_dh
1143 			 && sdev->scsi_dh_data->scsi_dh->prep_fn)) {
1144 		ret = sdev->scsi_dh_data->scsi_dh->prep_fn(sdev, req);
1145 		if (ret != BLKPREP_OK)
1146 			return ret;
1147 	}
1148 
1149 	/*
1150 	 * Filesystem requests must transfer data.
1151 	 */
1152 	BUG_ON(!req->nr_phys_segments);
1153 
1154 	cmd = scsi_get_cmd_from_req(sdev, req);
1155 	if (unlikely(!cmd))
1156 		return BLKPREP_DEFER;
1157 
1158 	memset(cmd->cmnd, 0, BLK_MAX_CDB);
1159 	return scsi_init_io(cmd, GFP_ATOMIC);
1160 }
1161 EXPORT_SYMBOL(scsi_setup_fs_cmnd);
1162 
1163 int scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1164 {
1165 	int ret = BLKPREP_OK;
1166 
1167 	/*
1168 	 * If the device is not in running state we will reject some
1169 	 * or all commands.
1170 	 */
1171 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1172 		switch (sdev->sdev_state) {
1173 		case SDEV_OFFLINE:
1174 			/*
1175 			 * If the device is offline we refuse to process any
1176 			 * commands.  The device must be brought online
1177 			 * before trying any recovery commands.
1178 			 */
1179 			sdev_printk(KERN_ERR, sdev,
1180 				    "rejecting I/O to offline device\n");
1181 			ret = BLKPREP_KILL;
1182 			break;
1183 		case SDEV_DEL:
1184 			/*
1185 			 * If the device is fully deleted, we refuse to
1186 			 * process any commands as well.
1187 			 */
1188 			sdev_printk(KERN_ERR, sdev,
1189 				    "rejecting I/O to dead device\n");
1190 			ret = BLKPREP_KILL;
1191 			break;
1192 		case SDEV_QUIESCE:
1193 		case SDEV_BLOCK:
1194 		case SDEV_CREATED_BLOCK:
1195 			/*
1196 			 * If the devices is blocked we defer normal commands.
1197 			 */
1198 			if (!(req->cmd_flags & REQ_PREEMPT))
1199 				ret = BLKPREP_DEFER;
1200 			break;
1201 		default:
1202 			/*
1203 			 * For any other not fully online state we only allow
1204 			 * special commands.  In particular any user initiated
1205 			 * command is not allowed.
1206 			 */
1207 			if (!(req->cmd_flags & REQ_PREEMPT))
1208 				ret = BLKPREP_KILL;
1209 			break;
1210 		}
1211 	}
1212 	return ret;
1213 }
1214 EXPORT_SYMBOL(scsi_prep_state_check);
1215 
1216 int scsi_prep_return(struct request_queue *q, struct request *req, int ret)
1217 {
1218 	struct scsi_device *sdev = q->queuedata;
1219 
1220 	switch (ret) {
1221 	case BLKPREP_KILL:
1222 		req->errors = DID_NO_CONNECT << 16;
1223 		/* release the command and kill it */
1224 		if (req->special) {
1225 			struct scsi_cmnd *cmd = req->special;
1226 			scsi_release_buffers(cmd);
1227 			scsi_put_command(cmd);
1228 			req->special = NULL;
1229 		}
1230 		break;
1231 	case BLKPREP_DEFER:
1232 		/*
1233 		 * If we defer, the blk_peek_request() returns NULL, but the
1234 		 * queue must be restarted, so we schedule a callback to happen
1235 		 * shortly.
1236 		 */
1237 		if (sdev->device_busy == 0)
1238 			blk_delay_queue(q, SCSI_QUEUE_DELAY);
1239 		break;
1240 	default:
1241 		req->cmd_flags |= REQ_DONTPREP;
1242 	}
1243 
1244 	return ret;
1245 }
1246 EXPORT_SYMBOL(scsi_prep_return);
1247 
1248 int scsi_prep_fn(struct request_queue *q, struct request *req)
1249 {
1250 	struct scsi_device *sdev = q->queuedata;
1251 	int ret = BLKPREP_KILL;
1252 
1253 	if (req->cmd_type == REQ_TYPE_BLOCK_PC)
1254 		ret = scsi_setup_blk_pc_cmnd(sdev, req);
1255 	return scsi_prep_return(q, req, ret);
1256 }
1257 EXPORT_SYMBOL(scsi_prep_fn);
1258 
1259 /*
1260  * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1261  * return 0.
1262  *
1263  * Called with the queue_lock held.
1264  */
1265 static inline int scsi_dev_queue_ready(struct request_queue *q,
1266 				  struct scsi_device *sdev)
1267 {
1268 	if (sdev->device_busy == 0 && sdev->device_blocked) {
1269 		/*
1270 		 * unblock after device_blocked iterates to zero
1271 		 */
1272 		if (--sdev->device_blocked == 0) {
1273 			SCSI_LOG_MLQUEUE(3,
1274 				   sdev_printk(KERN_INFO, sdev,
1275 				   "unblocking device at zero depth\n"));
1276 		} else {
1277 			blk_delay_queue(q, SCSI_QUEUE_DELAY);
1278 			return 0;
1279 		}
1280 	}
1281 	if (scsi_device_is_busy(sdev))
1282 		return 0;
1283 
1284 	return 1;
1285 }
1286 
1287 
1288 /*
1289  * scsi_target_queue_ready: checks if there we can send commands to target
1290  * @sdev: scsi device on starget to check.
1291  *
1292  * Called with the host lock held.
1293  */
1294 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1295 					   struct scsi_device *sdev)
1296 {
1297 	struct scsi_target *starget = scsi_target(sdev);
1298 
1299 	if (starget->single_lun) {
1300 		if (starget->starget_sdev_user &&
1301 		    starget->starget_sdev_user != sdev)
1302 			return 0;
1303 		starget->starget_sdev_user = sdev;
1304 	}
1305 
1306 	if (starget->target_busy == 0 && starget->target_blocked) {
1307 		/*
1308 		 * unblock after target_blocked iterates to zero
1309 		 */
1310 		if (--starget->target_blocked == 0) {
1311 			SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1312 					 "unblocking target at zero depth\n"));
1313 		} else
1314 			return 0;
1315 	}
1316 
1317 	if (scsi_target_is_busy(starget)) {
1318 		if (list_empty(&sdev->starved_entry))
1319 			list_add_tail(&sdev->starved_entry,
1320 				      &shost->starved_list);
1321 		return 0;
1322 	}
1323 
1324 	/* We're OK to process the command, so we can't be starved */
1325 	if (!list_empty(&sdev->starved_entry))
1326 		list_del_init(&sdev->starved_entry);
1327 	return 1;
1328 }
1329 
1330 /*
1331  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1332  * return 0. We must end up running the queue again whenever 0 is
1333  * returned, else IO can hang.
1334  *
1335  * Called with host_lock held.
1336  */
1337 static inline int scsi_host_queue_ready(struct request_queue *q,
1338 				   struct Scsi_Host *shost,
1339 				   struct scsi_device *sdev)
1340 {
1341 	if (scsi_host_in_recovery(shost))
1342 		return 0;
1343 	if (shost->host_busy == 0 && shost->host_blocked) {
1344 		/*
1345 		 * unblock after host_blocked iterates to zero
1346 		 */
1347 		if (--shost->host_blocked == 0) {
1348 			SCSI_LOG_MLQUEUE(3,
1349 				printk("scsi%d unblocking host at zero depth\n",
1350 					shost->host_no));
1351 		} else {
1352 			return 0;
1353 		}
1354 	}
1355 	if (scsi_host_is_busy(shost)) {
1356 		if (list_empty(&sdev->starved_entry))
1357 			list_add_tail(&sdev->starved_entry, &shost->starved_list);
1358 		return 0;
1359 	}
1360 
1361 	/* We're OK to process the command, so we can't be starved */
1362 	if (!list_empty(&sdev->starved_entry))
1363 		list_del_init(&sdev->starved_entry);
1364 
1365 	return 1;
1366 }
1367 
1368 /*
1369  * Busy state exporting function for request stacking drivers.
1370  *
1371  * For efficiency, no lock is taken to check the busy state of
1372  * shost/starget/sdev, since the returned value is not guaranteed and
1373  * may be changed after request stacking drivers call the function,
1374  * regardless of taking lock or not.
1375  *
1376  * When scsi can't dispatch I/Os anymore and needs to kill I/Os
1377  * (e.g. !sdev), scsi needs to return 'not busy'.
1378  * Otherwise, request stacking drivers may hold requests forever.
1379  */
1380 static int scsi_lld_busy(struct request_queue *q)
1381 {
1382 	struct scsi_device *sdev = q->queuedata;
1383 	struct Scsi_Host *shost;
1384 	struct scsi_target *starget;
1385 
1386 	if (!sdev)
1387 		return 0;
1388 
1389 	shost = sdev->host;
1390 	starget = scsi_target(sdev);
1391 
1392 	if (scsi_host_in_recovery(shost) || scsi_host_is_busy(shost) ||
1393 	    scsi_target_is_busy(starget) || scsi_device_is_busy(sdev))
1394 		return 1;
1395 
1396 	return 0;
1397 }
1398 
1399 /*
1400  * Kill a request for a dead device
1401  */
1402 static void scsi_kill_request(struct request *req, struct request_queue *q)
1403 {
1404 	struct scsi_cmnd *cmd = req->special;
1405 	struct scsi_device *sdev;
1406 	struct scsi_target *starget;
1407 	struct Scsi_Host *shost;
1408 
1409 	blk_start_request(req);
1410 
1411 	sdev = cmd->device;
1412 	starget = scsi_target(sdev);
1413 	shost = sdev->host;
1414 	scsi_init_cmd_errh(cmd);
1415 	cmd->result = DID_NO_CONNECT << 16;
1416 	atomic_inc(&cmd->device->iorequest_cnt);
1417 
1418 	/*
1419 	 * SCSI request completion path will do scsi_device_unbusy(),
1420 	 * bump busy counts.  To bump the counters, we need to dance
1421 	 * with the locks as normal issue path does.
1422 	 */
1423 	sdev->device_busy++;
1424 	spin_unlock(sdev->request_queue->queue_lock);
1425 	spin_lock(shost->host_lock);
1426 	shost->host_busy++;
1427 	starget->target_busy++;
1428 	spin_unlock(shost->host_lock);
1429 	spin_lock(sdev->request_queue->queue_lock);
1430 
1431 	blk_complete_request(req);
1432 }
1433 
1434 static void scsi_softirq_done(struct request *rq)
1435 {
1436 	struct scsi_cmnd *cmd = rq->special;
1437 	unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1438 	int disposition;
1439 
1440 	INIT_LIST_HEAD(&cmd->eh_entry);
1441 
1442 	atomic_inc(&cmd->device->iodone_cnt);
1443 	if (cmd->result)
1444 		atomic_inc(&cmd->device->ioerr_cnt);
1445 
1446 	disposition = scsi_decide_disposition(cmd);
1447 	if (disposition != SUCCESS &&
1448 	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1449 		sdev_printk(KERN_ERR, cmd->device,
1450 			    "timing out command, waited %lus\n",
1451 			    wait_for/HZ);
1452 		disposition = SUCCESS;
1453 	}
1454 
1455 	scsi_log_completion(cmd, disposition);
1456 
1457 	switch (disposition) {
1458 		case SUCCESS:
1459 			scsi_finish_command(cmd);
1460 			break;
1461 		case NEEDS_RETRY:
1462 			scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1463 			break;
1464 		case ADD_TO_MLQUEUE:
1465 			scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1466 			break;
1467 		default:
1468 			if (!scsi_eh_scmd_add(cmd, 0))
1469 				scsi_finish_command(cmd);
1470 	}
1471 }
1472 
1473 /*
1474  * Function:    scsi_request_fn()
1475  *
1476  * Purpose:     Main strategy routine for SCSI.
1477  *
1478  * Arguments:   q       - Pointer to actual queue.
1479  *
1480  * Returns:     Nothing
1481  *
1482  * Lock status: IO request lock assumed to be held when called.
1483  */
1484 static void scsi_request_fn(struct request_queue *q)
1485 {
1486 	struct scsi_device *sdev = q->queuedata;
1487 	struct Scsi_Host *shost;
1488 	struct scsi_cmnd *cmd;
1489 	struct request *req;
1490 
1491 	if (!sdev) {
1492 		printk("scsi: killing requests for dead queue\n");
1493 		while ((req = blk_peek_request(q)) != NULL)
1494 			scsi_kill_request(req, q);
1495 		return;
1496 	}
1497 
1498 	if(!get_device(&sdev->sdev_gendev))
1499 		/* We must be tearing the block queue down already */
1500 		return;
1501 
1502 	/*
1503 	 * To start with, we keep looping until the queue is empty, or until
1504 	 * the host is no longer able to accept any more requests.
1505 	 */
1506 	shost = sdev->host;
1507 	for (;;) {
1508 		int rtn;
1509 		/*
1510 		 * get next queueable request.  We do this early to make sure
1511 		 * that the request is fully prepared even if we cannot
1512 		 * accept it.
1513 		 */
1514 		req = blk_peek_request(q);
1515 		if (!req || !scsi_dev_queue_ready(q, sdev))
1516 			break;
1517 
1518 		if (unlikely(!scsi_device_online(sdev))) {
1519 			sdev_printk(KERN_ERR, sdev,
1520 				    "rejecting I/O to offline device\n");
1521 			scsi_kill_request(req, q);
1522 			continue;
1523 		}
1524 
1525 
1526 		/*
1527 		 * Remove the request from the request list.
1528 		 */
1529 		if (!(blk_queue_tagged(q) && !blk_queue_start_tag(q, req)))
1530 			blk_start_request(req);
1531 		sdev->device_busy++;
1532 
1533 		spin_unlock(q->queue_lock);
1534 		cmd = req->special;
1535 		if (unlikely(cmd == NULL)) {
1536 			printk(KERN_CRIT "impossible request in %s.\n"
1537 					 "please mail a stack trace to "
1538 					 "linux-scsi@vger.kernel.org\n",
1539 					 __func__);
1540 			blk_dump_rq_flags(req, "foo");
1541 			BUG();
1542 		}
1543 		spin_lock(shost->host_lock);
1544 
1545 		/*
1546 		 * We hit this when the driver is using a host wide
1547 		 * tag map. For device level tag maps the queue_depth check
1548 		 * in the device ready fn would prevent us from trying
1549 		 * to allocate a tag. Since the map is a shared host resource
1550 		 * we add the dev to the starved list so it eventually gets
1551 		 * a run when a tag is freed.
1552 		 */
1553 		if (blk_queue_tagged(q) && !blk_rq_tagged(req)) {
1554 			if (list_empty(&sdev->starved_entry))
1555 				list_add_tail(&sdev->starved_entry,
1556 					      &shost->starved_list);
1557 			goto not_ready;
1558 		}
1559 
1560 		if (!scsi_target_queue_ready(shost, sdev))
1561 			goto not_ready;
1562 
1563 		if (!scsi_host_queue_ready(q, shost, sdev))
1564 			goto not_ready;
1565 
1566 		scsi_target(sdev)->target_busy++;
1567 		shost->host_busy++;
1568 
1569 		/*
1570 		 * XXX(hch): This is rather suboptimal, scsi_dispatch_cmd will
1571 		 *		take the lock again.
1572 		 */
1573 		spin_unlock_irq(shost->host_lock);
1574 
1575 		/*
1576 		 * Finally, initialize any error handling parameters, and set up
1577 		 * the timers for timeouts.
1578 		 */
1579 		scsi_init_cmd_errh(cmd);
1580 
1581 		/*
1582 		 * Dispatch the command to the low-level driver.
1583 		 */
1584 		rtn = scsi_dispatch_cmd(cmd);
1585 		spin_lock_irq(q->queue_lock);
1586 		if (rtn)
1587 			goto out_delay;
1588 	}
1589 
1590 	goto out;
1591 
1592  not_ready:
1593 	spin_unlock_irq(shost->host_lock);
1594 
1595 	/*
1596 	 * lock q, handle tag, requeue req, and decrement device_busy. We
1597 	 * must return with queue_lock held.
1598 	 *
1599 	 * Decrementing device_busy without checking it is OK, as all such
1600 	 * cases (host limits or settings) should run the queue at some
1601 	 * later time.
1602 	 */
1603 	spin_lock_irq(q->queue_lock);
1604 	blk_requeue_request(q, req);
1605 	sdev->device_busy--;
1606 out_delay:
1607 	if (sdev->device_busy == 0)
1608 		blk_delay_queue(q, SCSI_QUEUE_DELAY);
1609 out:
1610 	/* must be careful here...if we trigger the ->remove() function
1611 	 * we cannot be holding the q lock */
1612 	spin_unlock_irq(q->queue_lock);
1613 	put_device(&sdev->sdev_gendev);
1614 	spin_lock_irq(q->queue_lock);
1615 }
1616 
1617 u64 scsi_calculate_bounce_limit(struct Scsi_Host *shost)
1618 {
1619 	struct device *host_dev;
1620 	u64 bounce_limit = 0xffffffff;
1621 
1622 	if (shost->unchecked_isa_dma)
1623 		return BLK_BOUNCE_ISA;
1624 	/*
1625 	 * Platforms with virtual-DMA translation
1626 	 * hardware have no practical limit.
1627 	 */
1628 	if (!PCI_DMA_BUS_IS_PHYS)
1629 		return BLK_BOUNCE_ANY;
1630 
1631 	host_dev = scsi_get_device(shost);
1632 	if (host_dev && host_dev->dma_mask)
1633 		bounce_limit = *host_dev->dma_mask;
1634 
1635 	return bounce_limit;
1636 }
1637 EXPORT_SYMBOL(scsi_calculate_bounce_limit);
1638 
1639 struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost,
1640 					 request_fn_proc *request_fn)
1641 {
1642 	struct request_queue *q;
1643 	struct device *dev = shost->shost_gendev.parent;
1644 
1645 	q = blk_init_queue(request_fn, NULL);
1646 	if (!q)
1647 		return NULL;
1648 
1649 	/*
1650 	 * this limit is imposed by hardware restrictions
1651 	 */
1652 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1653 					SCSI_MAX_SG_CHAIN_SEGMENTS));
1654 
1655 	if (scsi_host_prot_dma(shost)) {
1656 		shost->sg_prot_tablesize =
1657 			min_not_zero(shost->sg_prot_tablesize,
1658 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1659 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1660 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1661 	}
1662 
1663 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1664 	blk_queue_bounce_limit(q, scsi_calculate_bounce_limit(shost));
1665 	blk_queue_segment_boundary(q, shost->dma_boundary);
1666 	dma_set_seg_boundary(dev, shost->dma_boundary);
1667 
1668 	blk_queue_max_segment_size(q, dma_get_max_seg_size(dev));
1669 
1670 	if (!shost->use_clustering)
1671 		q->limits.cluster = 0;
1672 
1673 	/*
1674 	 * set a reasonable default alignment on word boundaries: the
1675 	 * host and device may alter it using
1676 	 * blk_queue_update_dma_alignment() later.
1677 	 */
1678 	blk_queue_dma_alignment(q, 0x03);
1679 
1680 	return q;
1681 }
1682 EXPORT_SYMBOL(__scsi_alloc_queue);
1683 
1684 struct request_queue *scsi_alloc_queue(struct scsi_device *sdev)
1685 {
1686 	struct request_queue *q;
1687 
1688 	q = __scsi_alloc_queue(sdev->host, scsi_request_fn);
1689 	if (!q)
1690 		return NULL;
1691 
1692 	blk_queue_prep_rq(q, scsi_prep_fn);
1693 	blk_queue_softirq_done(q, scsi_softirq_done);
1694 	blk_queue_rq_timed_out(q, scsi_times_out);
1695 	blk_queue_lld_busy(q, scsi_lld_busy);
1696 	return q;
1697 }
1698 
1699 void scsi_free_queue(struct request_queue *q)
1700 {
1701 	blk_cleanup_queue(q);
1702 }
1703 
1704 /*
1705  * Function:    scsi_block_requests()
1706  *
1707  * Purpose:     Utility function used by low-level drivers to prevent further
1708  *		commands from being queued to the device.
1709  *
1710  * Arguments:   shost       - Host in question
1711  *
1712  * Returns:     Nothing
1713  *
1714  * Lock status: No locks are assumed held.
1715  *
1716  * Notes:       There is no timer nor any other means by which the requests
1717  *		get unblocked other than the low-level driver calling
1718  *		scsi_unblock_requests().
1719  */
1720 void scsi_block_requests(struct Scsi_Host *shost)
1721 {
1722 	shost->host_self_blocked = 1;
1723 }
1724 EXPORT_SYMBOL(scsi_block_requests);
1725 
1726 /*
1727  * Function:    scsi_unblock_requests()
1728  *
1729  * Purpose:     Utility function used by low-level drivers to allow further
1730  *		commands from being queued to the device.
1731  *
1732  * Arguments:   shost       - Host in question
1733  *
1734  * Returns:     Nothing
1735  *
1736  * Lock status: No locks are assumed held.
1737  *
1738  * Notes:       There is no timer nor any other means by which the requests
1739  *		get unblocked other than the low-level driver calling
1740  *		scsi_unblock_requests().
1741  *
1742  *		This is done as an API function so that changes to the
1743  *		internals of the scsi mid-layer won't require wholesale
1744  *		changes to drivers that use this feature.
1745  */
1746 void scsi_unblock_requests(struct Scsi_Host *shost)
1747 {
1748 	shost->host_self_blocked = 0;
1749 	scsi_run_host_queues(shost);
1750 }
1751 EXPORT_SYMBOL(scsi_unblock_requests);
1752 
1753 int __init scsi_init_queue(void)
1754 {
1755 	int i;
1756 
1757 	scsi_sdb_cache = kmem_cache_create("scsi_data_buffer",
1758 					   sizeof(struct scsi_data_buffer),
1759 					   0, 0, NULL);
1760 	if (!scsi_sdb_cache) {
1761 		printk(KERN_ERR "SCSI: can't init scsi sdb cache\n");
1762 		return -ENOMEM;
1763 	}
1764 
1765 	for (i = 0; i < SG_MEMPOOL_NR; i++) {
1766 		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1767 		int size = sgp->size * sizeof(struct scatterlist);
1768 
1769 		sgp->slab = kmem_cache_create(sgp->name, size, 0,
1770 				SLAB_HWCACHE_ALIGN, NULL);
1771 		if (!sgp->slab) {
1772 			printk(KERN_ERR "SCSI: can't init sg slab %s\n",
1773 					sgp->name);
1774 			goto cleanup_sdb;
1775 		}
1776 
1777 		sgp->pool = mempool_create_slab_pool(SG_MEMPOOL_SIZE,
1778 						     sgp->slab);
1779 		if (!sgp->pool) {
1780 			printk(KERN_ERR "SCSI: can't init sg mempool %s\n",
1781 					sgp->name);
1782 			goto cleanup_sdb;
1783 		}
1784 	}
1785 
1786 	return 0;
1787 
1788 cleanup_sdb:
1789 	for (i = 0; i < SG_MEMPOOL_NR; i++) {
1790 		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1791 		if (sgp->pool)
1792 			mempool_destroy(sgp->pool);
1793 		if (sgp->slab)
1794 			kmem_cache_destroy(sgp->slab);
1795 	}
1796 	kmem_cache_destroy(scsi_sdb_cache);
1797 
1798 	return -ENOMEM;
1799 }
1800 
1801 void scsi_exit_queue(void)
1802 {
1803 	int i;
1804 
1805 	kmem_cache_destroy(scsi_sdb_cache);
1806 
1807 	for (i = 0; i < SG_MEMPOOL_NR; i++) {
1808 		struct scsi_host_sg_pool *sgp = scsi_sg_pools + i;
1809 		mempool_destroy(sgp->pool);
1810 		kmem_cache_destroy(sgp->slab);
1811 	}
1812 }
1813 
1814 /**
1815  *	scsi_mode_select - issue a mode select
1816  *	@sdev:	SCSI device to be queried
1817  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
1818  *	@sp:	Save page bit (0 == don't save, 1 == save)
1819  *	@modepage: mode page being requested
1820  *	@buffer: request buffer (may not be smaller than eight bytes)
1821  *	@len:	length of request buffer.
1822  *	@timeout: command timeout
1823  *	@retries: number of retries before failing
1824  *	@data: returns a structure abstracting the mode header data
1825  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
1826  *		must be SCSI_SENSE_BUFFERSIZE big.
1827  *
1828  *	Returns zero if successful; negative error number or scsi
1829  *	status on error
1830  *
1831  */
1832 int
1833 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1834 		 unsigned char *buffer, int len, int timeout, int retries,
1835 		 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1836 {
1837 	unsigned char cmd[10];
1838 	unsigned char *real_buffer;
1839 	int ret;
1840 
1841 	memset(cmd, 0, sizeof(cmd));
1842 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1843 
1844 	if (sdev->use_10_for_ms) {
1845 		if (len > 65535)
1846 			return -EINVAL;
1847 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
1848 		if (!real_buffer)
1849 			return -ENOMEM;
1850 		memcpy(real_buffer + 8, buffer, len);
1851 		len += 8;
1852 		real_buffer[0] = 0;
1853 		real_buffer[1] = 0;
1854 		real_buffer[2] = data->medium_type;
1855 		real_buffer[3] = data->device_specific;
1856 		real_buffer[4] = data->longlba ? 0x01 : 0;
1857 		real_buffer[5] = 0;
1858 		real_buffer[6] = data->block_descriptor_length >> 8;
1859 		real_buffer[7] = data->block_descriptor_length;
1860 
1861 		cmd[0] = MODE_SELECT_10;
1862 		cmd[7] = len >> 8;
1863 		cmd[8] = len;
1864 	} else {
1865 		if (len > 255 || data->block_descriptor_length > 255 ||
1866 		    data->longlba)
1867 			return -EINVAL;
1868 
1869 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
1870 		if (!real_buffer)
1871 			return -ENOMEM;
1872 		memcpy(real_buffer + 4, buffer, len);
1873 		len += 4;
1874 		real_buffer[0] = 0;
1875 		real_buffer[1] = data->medium_type;
1876 		real_buffer[2] = data->device_specific;
1877 		real_buffer[3] = data->block_descriptor_length;
1878 
1879 
1880 		cmd[0] = MODE_SELECT;
1881 		cmd[4] = len;
1882 	}
1883 
1884 	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
1885 			       sshdr, timeout, retries, NULL);
1886 	kfree(real_buffer);
1887 	return ret;
1888 }
1889 EXPORT_SYMBOL_GPL(scsi_mode_select);
1890 
1891 /**
1892  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
1893  *	@sdev:	SCSI device to be queried
1894  *	@dbd:	set if mode sense will allow block descriptors to be returned
1895  *	@modepage: mode page being requested
1896  *	@buffer: request buffer (may not be smaller than eight bytes)
1897  *	@len:	length of request buffer.
1898  *	@timeout: command timeout
1899  *	@retries: number of retries before failing
1900  *	@data: returns a structure abstracting the mode header data
1901  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
1902  *		must be SCSI_SENSE_BUFFERSIZE big.
1903  *
1904  *	Returns zero if unsuccessful, or the header offset (either 4
1905  *	or 8 depending on whether a six or ten byte command was
1906  *	issued) if successful.
1907  */
1908 int
1909 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
1910 		  unsigned char *buffer, int len, int timeout, int retries,
1911 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1912 {
1913 	unsigned char cmd[12];
1914 	int use_10_for_ms;
1915 	int header_length;
1916 	int result;
1917 	struct scsi_sense_hdr my_sshdr;
1918 
1919 	memset(data, 0, sizeof(*data));
1920 	memset(&cmd[0], 0, 12);
1921 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
1922 	cmd[2] = modepage;
1923 
1924 	/* caller might not be interested in sense, but we need it */
1925 	if (!sshdr)
1926 		sshdr = &my_sshdr;
1927 
1928  retry:
1929 	use_10_for_ms = sdev->use_10_for_ms;
1930 
1931 	if (use_10_for_ms) {
1932 		if (len < 8)
1933 			len = 8;
1934 
1935 		cmd[0] = MODE_SENSE_10;
1936 		cmd[8] = len;
1937 		header_length = 8;
1938 	} else {
1939 		if (len < 4)
1940 			len = 4;
1941 
1942 		cmd[0] = MODE_SENSE;
1943 		cmd[4] = len;
1944 		header_length = 4;
1945 	}
1946 
1947 	memset(buffer, 0, len);
1948 
1949 	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
1950 				  sshdr, timeout, retries, NULL);
1951 
1952 	/* This code looks awful: what it's doing is making sure an
1953 	 * ILLEGAL REQUEST sense return identifies the actual command
1954 	 * byte as the problem.  MODE_SENSE commands can return
1955 	 * ILLEGAL REQUEST if the code page isn't supported */
1956 
1957 	if (use_10_for_ms && !scsi_status_is_good(result) &&
1958 	    (driver_byte(result) & DRIVER_SENSE)) {
1959 		if (scsi_sense_valid(sshdr)) {
1960 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
1961 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
1962 				/*
1963 				 * Invalid command operation code
1964 				 */
1965 				sdev->use_10_for_ms = 0;
1966 				goto retry;
1967 			}
1968 		}
1969 	}
1970 
1971 	if(scsi_status_is_good(result)) {
1972 		if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
1973 			     (modepage == 6 || modepage == 8))) {
1974 			/* Initio breakage? */
1975 			header_length = 0;
1976 			data->length = 13;
1977 			data->medium_type = 0;
1978 			data->device_specific = 0;
1979 			data->longlba = 0;
1980 			data->block_descriptor_length = 0;
1981 		} else if(use_10_for_ms) {
1982 			data->length = buffer[0]*256 + buffer[1] + 2;
1983 			data->medium_type = buffer[2];
1984 			data->device_specific = buffer[3];
1985 			data->longlba = buffer[4] & 0x01;
1986 			data->block_descriptor_length = buffer[6]*256
1987 				+ buffer[7];
1988 		} else {
1989 			data->length = buffer[0] + 1;
1990 			data->medium_type = buffer[1];
1991 			data->device_specific = buffer[2];
1992 			data->block_descriptor_length = buffer[3];
1993 		}
1994 		data->header_length = header_length;
1995 	}
1996 
1997 	return result;
1998 }
1999 EXPORT_SYMBOL(scsi_mode_sense);
2000 
2001 /**
2002  *	scsi_test_unit_ready - test if unit is ready
2003  *	@sdev:	scsi device to change the state of.
2004  *	@timeout: command timeout
2005  *	@retries: number of retries before failing
2006  *	@sshdr_external: Optional pointer to struct scsi_sense_hdr for
2007  *		returning sense. Make sure that this is cleared before passing
2008  *		in.
2009  *
2010  *	Returns zero if unsuccessful or an error if TUR failed.  For
2011  *	removable media, UNIT_ATTENTION sets ->changed flag.
2012  **/
2013 int
2014 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2015 		     struct scsi_sense_hdr *sshdr_external)
2016 {
2017 	char cmd[] = {
2018 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2019 	};
2020 	struct scsi_sense_hdr *sshdr;
2021 	int result;
2022 
2023 	if (!sshdr_external)
2024 		sshdr = kzalloc(sizeof(*sshdr), GFP_KERNEL);
2025 	else
2026 		sshdr = sshdr_external;
2027 
2028 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2029 	do {
2030 		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2031 					  timeout, retries, NULL);
2032 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2033 		    sshdr->sense_key == UNIT_ATTENTION)
2034 			sdev->changed = 1;
2035 	} while (scsi_sense_valid(sshdr) &&
2036 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2037 
2038 	if (!sshdr_external)
2039 		kfree(sshdr);
2040 	return result;
2041 }
2042 EXPORT_SYMBOL(scsi_test_unit_ready);
2043 
2044 /**
2045  *	scsi_device_set_state - Take the given device through the device state model.
2046  *	@sdev:	scsi device to change the state of.
2047  *	@state:	state to change to.
2048  *
2049  *	Returns zero if unsuccessful or an error if the requested
2050  *	transition is illegal.
2051  */
2052 int
2053 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2054 {
2055 	enum scsi_device_state oldstate = sdev->sdev_state;
2056 
2057 	if (state == oldstate)
2058 		return 0;
2059 
2060 	switch (state) {
2061 	case SDEV_CREATED:
2062 		switch (oldstate) {
2063 		case SDEV_CREATED_BLOCK:
2064 			break;
2065 		default:
2066 			goto illegal;
2067 		}
2068 		break;
2069 
2070 	case SDEV_RUNNING:
2071 		switch (oldstate) {
2072 		case SDEV_CREATED:
2073 		case SDEV_OFFLINE:
2074 		case SDEV_QUIESCE:
2075 		case SDEV_BLOCK:
2076 			break;
2077 		default:
2078 			goto illegal;
2079 		}
2080 		break;
2081 
2082 	case SDEV_QUIESCE:
2083 		switch (oldstate) {
2084 		case SDEV_RUNNING:
2085 		case SDEV_OFFLINE:
2086 			break;
2087 		default:
2088 			goto illegal;
2089 		}
2090 		break;
2091 
2092 	case SDEV_OFFLINE:
2093 		switch (oldstate) {
2094 		case SDEV_CREATED:
2095 		case SDEV_RUNNING:
2096 		case SDEV_QUIESCE:
2097 		case SDEV_BLOCK:
2098 			break;
2099 		default:
2100 			goto illegal;
2101 		}
2102 		break;
2103 
2104 	case SDEV_BLOCK:
2105 		switch (oldstate) {
2106 		case SDEV_RUNNING:
2107 		case SDEV_CREATED_BLOCK:
2108 			break;
2109 		default:
2110 			goto illegal;
2111 		}
2112 		break;
2113 
2114 	case SDEV_CREATED_BLOCK:
2115 		switch (oldstate) {
2116 		case SDEV_CREATED:
2117 			break;
2118 		default:
2119 			goto illegal;
2120 		}
2121 		break;
2122 
2123 	case SDEV_CANCEL:
2124 		switch (oldstate) {
2125 		case SDEV_CREATED:
2126 		case SDEV_RUNNING:
2127 		case SDEV_QUIESCE:
2128 		case SDEV_OFFLINE:
2129 		case SDEV_BLOCK:
2130 			break;
2131 		default:
2132 			goto illegal;
2133 		}
2134 		break;
2135 
2136 	case SDEV_DEL:
2137 		switch (oldstate) {
2138 		case SDEV_CREATED:
2139 		case SDEV_RUNNING:
2140 		case SDEV_OFFLINE:
2141 		case SDEV_CANCEL:
2142 			break;
2143 		default:
2144 			goto illegal;
2145 		}
2146 		break;
2147 
2148 	}
2149 	sdev->sdev_state = state;
2150 	return 0;
2151 
2152  illegal:
2153 	SCSI_LOG_ERROR_RECOVERY(1,
2154 				sdev_printk(KERN_ERR, sdev,
2155 					    "Illegal state transition %s->%s\n",
2156 					    scsi_device_state_name(oldstate),
2157 					    scsi_device_state_name(state))
2158 				);
2159 	return -EINVAL;
2160 }
2161 EXPORT_SYMBOL(scsi_device_set_state);
2162 
2163 /**
2164  * 	sdev_evt_emit - emit a single SCSI device uevent
2165  *	@sdev: associated SCSI device
2166  *	@evt: event to emit
2167  *
2168  *	Send a single uevent (scsi_event) to the associated scsi_device.
2169  */
2170 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2171 {
2172 	int idx = 0;
2173 	char *envp[3];
2174 
2175 	switch (evt->evt_type) {
2176 	case SDEV_EVT_MEDIA_CHANGE:
2177 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2178 		break;
2179 
2180 	default:
2181 		/* do nothing */
2182 		break;
2183 	}
2184 
2185 	envp[idx++] = NULL;
2186 
2187 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2188 }
2189 
2190 /**
2191  * 	sdev_evt_thread - send a uevent for each scsi event
2192  *	@work: work struct for scsi_device
2193  *
2194  *	Dispatch queued events to their associated scsi_device kobjects
2195  *	as uevents.
2196  */
2197 void scsi_evt_thread(struct work_struct *work)
2198 {
2199 	struct scsi_device *sdev;
2200 	LIST_HEAD(event_list);
2201 
2202 	sdev = container_of(work, struct scsi_device, event_work);
2203 
2204 	while (1) {
2205 		struct scsi_event *evt;
2206 		struct list_head *this, *tmp;
2207 		unsigned long flags;
2208 
2209 		spin_lock_irqsave(&sdev->list_lock, flags);
2210 		list_splice_init(&sdev->event_list, &event_list);
2211 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2212 
2213 		if (list_empty(&event_list))
2214 			break;
2215 
2216 		list_for_each_safe(this, tmp, &event_list) {
2217 			evt = list_entry(this, struct scsi_event, node);
2218 			list_del(&evt->node);
2219 			scsi_evt_emit(sdev, evt);
2220 			kfree(evt);
2221 		}
2222 	}
2223 }
2224 
2225 /**
2226  * 	sdev_evt_send - send asserted event to uevent thread
2227  *	@sdev: scsi_device event occurred on
2228  *	@evt: event to send
2229  *
2230  *	Assert scsi device event asynchronously.
2231  */
2232 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2233 {
2234 	unsigned long flags;
2235 
2236 #if 0
2237 	/* FIXME: currently this check eliminates all media change events
2238 	 * for polled devices.  Need to update to discriminate between AN
2239 	 * and polled events */
2240 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2241 		kfree(evt);
2242 		return;
2243 	}
2244 #endif
2245 
2246 	spin_lock_irqsave(&sdev->list_lock, flags);
2247 	list_add_tail(&evt->node, &sdev->event_list);
2248 	schedule_work(&sdev->event_work);
2249 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2250 }
2251 EXPORT_SYMBOL_GPL(sdev_evt_send);
2252 
2253 /**
2254  * 	sdev_evt_alloc - allocate a new scsi event
2255  *	@evt_type: type of event to allocate
2256  *	@gfpflags: GFP flags for allocation
2257  *
2258  *	Allocates and returns a new scsi_event.
2259  */
2260 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2261 				  gfp_t gfpflags)
2262 {
2263 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2264 	if (!evt)
2265 		return NULL;
2266 
2267 	evt->evt_type = evt_type;
2268 	INIT_LIST_HEAD(&evt->node);
2269 
2270 	/* evt_type-specific initialization, if any */
2271 	switch (evt_type) {
2272 	case SDEV_EVT_MEDIA_CHANGE:
2273 	default:
2274 		/* do nothing */
2275 		break;
2276 	}
2277 
2278 	return evt;
2279 }
2280 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2281 
2282 /**
2283  * 	sdev_evt_send_simple - send asserted event to uevent thread
2284  *	@sdev: scsi_device event occurred on
2285  *	@evt_type: type of event to send
2286  *	@gfpflags: GFP flags for allocation
2287  *
2288  *	Assert scsi device event asynchronously, given an event type.
2289  */
2290 void sdev_evt_send_simple(struct scsi_device *sdev,
2291 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2292 {
2293 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2294 	if (!evt) {
2295 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2296 			    evt_type);
2297 		return;
2298 	}
2299 
2300 	sdev_evt_send(sdev, evt);
2301 }
2302 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2303 
2304 /**
2305  *	scsi_device_quiesce - Block user issued commands.
2306  *	@sdev:	scsi device to quiesce.
2307  *
2308  *	This works by trying to transition to the SDEV_QUIESCE state
2309  *	(which must be a legal transition).  When the device is in this
2310  *	state, only special requests will be accepted, all others will
2311  *	be deferred.  Since special requests may also be requeued requests,
2312  *	a successful return doesn't guarantee the device will be
2313  *	totally quiescent.
2314  *
2315  *	Must be called with user context, may sleep.
2316  *
2317  *	Returns zero if unsuccessful or an error if not.
2318  */
2319 int
2320 scsi_device_quiesce(struct scsi_device *sdev)
2321 {
2322 	int err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2323 	if (err)
2324 		return err;
2325 
2326 	scsi_run_queue(sdev->request_queue);
2327 	while (sdev->device_busy) {
2328 		msleep_interruptible(200);
2329 		scsi_run_queue(sdev->request_queue);
2330 	}
2331 	return 0;
2332 }
2333 EXPORT_SYMBOL(scsi_device_quiesce);
2334 
2335 /**
2336  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2337  *	@sdev:	scsi device to resume.
2338  *
2339  *	Moves the device from quiesced back to running and restarts the
2340  *	queues.
2341  *
2342  *	Must be called with user context, may sleep.
2343  */
2344 void
2345 scsi_device_resume(struct scsi_device *sdev)
2346 {
2347 	if(scsi_device_set_state(sdev, SDEV_RUNNING))
2348 		return;
2349 	scsi_run_queue(sdev->request_queue);
2350 }
2351 EXPORT_SYMBOL(scsi_device_resume);
2352 
2353 static void
2354 device_quiesce_fn(struct scsi_device *sdev, void *data)
2355 {
2356 	scsi_device_quiesce(sdev);
2357 }
2358 
2359 void
2360 scsi_target_quiesce(struct scsi_target *starget)
2361 {
2362 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2363 }
2364 EXPORT_SYMBOL(scsi_target_quiesce);
2365 
2366 static void
2367 device_resume_fn(struct scsi_device *sdev, void *data)
2368 {
2369 	scsi_device_resume(sdev);
2370 }
2371 
2372 void
2373 scsi_target_resume(struct scsi_target *starget)
2374 {
2375 	starget_for_each_device(starget, NULL, device_resume_fn);
2376 }
2377 EXPORT_SYMBOL(scsi_target_resume);
2378 
2379 /**
2380  * scsi_internal_device_block - internal function to put a device temporarily into the SDEV_BLOCK state
2381  * @sdev:	device to block
2382  *
2383  * Block request made by scsi lld's to temporarily stop all
2384  * scsi commands on the specified device.  Called from interrupt
2385  * or normal process context.
2386  *
2387  * Returns zero if successful or error if not
2388  *
2389  * Notes:
2390  *	This routine transitions the device to the SDEV_BLOCK state
2391  *	(which must be a legal transition).  When the device is in this
2392  *	state, all commands are deferred until the scsi lld reenables
2393  *	the device with scsi_device_unblock or device_block_tmo fires.
2394  *	This routine assumes the host_lock is held on entry.
2395  */
2396 int
2397 scsi_internal_device_block(struct scsi_device *sdev)
2398 {
2399 	struct request_queue *q = sdev->request_queue;
2400 	unsigned long flags;
2401 	int err = 0;
2402 
2403 	err = scsi_device_set_state(sdev, SDEV_BLOCK);
2404 	if (err) {
2405 		err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2406 
2407 		if (err)
2408 			return err;
2409 	}
2410 
2411 	/*
2412 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2413 	 * block layer from calling the midlayer with this device's
2414 	 * request queue.
2415 	 */
2416 	spin_lock_irqsave(q->queue_lock, flags);
2417 	blk_stop_queue(q);
2418 	spin_unlock_irqrestore(q->queue_lock, flags);
2419 
2420 	return 0;
2421 }
2422 EXPORT_SYMBOL_GPL(scsi_internal_device_block);
2423 
2424 /**
2425  * scsi_internal_device_unblock - resume a device after a block request
2426  * @sdev:	device to resume
2427  *
2428  * Called by scsi lld's or the midlayer to restart the device queue
2429  * for the previously suspended scsi device.  Called from interrupt or
2430  * normal process context.
2431  *
2432  * Returns zero if successful or error if not.
2433  *
2434  * Notes:
2435  *	This routine transitions the device to the SDEV_RUNNING state
2436  *	(which must be a legal transition) allowing the midlayer to
2437  *	goose the queue for this device.  This routine assumes the
2438  *	host_lock is held upon entry.
2439  */
2440 int
2441 scsi_internal_device_unblock(struct scsi_device *sdev)
2442 {
2443 	struct request_queue *q = sdev->request_queue;
2444 	unsigned long flags;
2445 
2446 	/*
2447 	 * Try to transition the scsi device to SDEV_RUNNING
2448 	 * and goose the device queue if successful.
2449 	 */
2450 	if (sdev->sdev_state == SDEV_BLOCK)
2451 		sdev->sdev_state = SDEV_RUNNING;
2452 	else if (sdev->sdev_state == SDEV_CREATED_BLOCK)
2453 		sdev->sdev_state = SDEV_CREATED;
2454 	else if (sdev->sdev_state != SDEV_CANCEL &&
2455 		 sdev->sdev_state != SDEV_OFFLINE)
2456 		return -EINVAL;
2457 
2458 	spin_lock_irqsave(q->queue_lock, flags);
2459 	blk_start_queue(q);
2460 	spin_unlock_irqrestore(q->queue_lock, flags);
2461 
2462 	return 0;
2463 }
2464 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock);
2465 
2466 static void
2467 device_block(struct scsi_device *sdev, void *data)
2468 {
2469 	scsi_internal_device_block(sdev);
2470 }
2471 
2472 static int
2473 target_block(struct device *dev, void *data)
2474 {
2475 	if (scsi_is_target_device(dev))
2476 		starget_for_each_device(to_scsi_target(dev), NULL,
2477 					device_block);
2478 	return 0;
2479 }
2480 
2481 void
2482 scsi_target_block(struct device *dev)
2483 {
2484 	if (scsi_is_target_device(dev))
2485 		starget_for_each_device(to_scsi_target(dev), NULL,
2486 					device_block);
2487 	else
2488 		device_for_each_child(dev, NULL, target_block);
2489 }
2490 EXPORT_SYMBOL_GPL(scsi_target_block);
2491 
2492 static void
2493 device_unblock(struct scsi_device *sdev, void *data)
2494 {
2495 	scsi_internal_device_unblock(sdev);
2496 }
2497 
2498 static int
2499 target_unblock(struct device *dev, void *data)
2500 {
2501 	if (scsi_is_target_device(dev))
2502 		starget_for_each_device(to_scsi_target(dev), NULL,
2503 					device_unblock);
2504 	return 0;
2505 }
2506 
2507 void
2508 scsi_target_unblock(struct device *dev)
2509 {
2510 	if (scsi_is_target_device(dev))
2511 		starget_for_each_device(to_scsi_target(dev), NULL,
2512 					device_unblock);
2513 	else
2514 		device_for_each_child(dev, NULL, target_unblock);
2515 }
2516 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2517 
2518 /**
2519  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2520  * @sgl:	scatter-gather list
2521  * @sg_count:	number of segments in sg
2522  * @offset:	offset in bytes into sg, on return offset into the mapped area
2523  * @len:	bytes to map, on return number of bytes mapped
2524  *
2525  * Returns virtual address of the start of the mapped page
2526  */
2527 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2528 			  size_t *offset, size_t *len)
2529 {
2530 	int i;
2531 	size_t sg_len = 0, len_complete = 0;
2532 	struct scatterlist *sg;
2533 	struct page *page;
2534 
2535 	WARN_ON(!irqs_disabled());
2536 
2537 	for_each_sg(sgl, sg, sg_count, i) {
2538 		len_complete = sg_len; /* Complete sg-entries */
2539 		sg_len += sg->length;
2540 		if (sg_len > *offset)
2541 			break;
2542 	}
2543 
2544 	if (unlikely(i == sg_count)) {
2545 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2546 			"elements %d\n",
2547 		       __func__, sg_len, *offset, sg_count);
2548 		WARN_ON(1);
2549 		return NULL;
2550 	}
2551 
2552 	/* Offset starting from the beginning of first page in this sg-entry */
2553 	*offset = *offset - len_complete + sg->offset;
2554 
2555 	/* Assumption: contiguous pages can be accessed as "page + i" */
2556 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2557 	*offset &= ~PAGE_MASK;
2558 
2559 	/* Bytes in this sg-entry from *offset to the end of the page */
2560 	sg_len = PAGE_SIZE - *offset;
2561 	if (*len > sg_len)
2562 		*len = sg_len;
2563 
2564 	return kmap_atomic(page, KM_BIO_SRC_IRQ);
2565 }
2566 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2567 
2568 /**
2569  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2570  * @virt:	virtual address to be unmapped
2571  */
2572 void scsi_kunmap_atomic_sg(void *virt)
2573 {
2574 	kunmap_atomic(virt, KM_BIO_SRC_IRQ);
2575 }
2576 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2577