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