1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 1999 Eric Youngdale
4  * Copyright (C) 2014 Christoph Hellwig
5  *
6  *  SCSI queueing library.
7  *      Initial versions: Eric Youngdale (eric@andante.org).
8  *                        Based upon conversations with large numbers
9  *                        of people at Linux Expo.
10  */
11 
12 #include <linux/bio.h>
13 #include <linux/bitops.h>
14 #include <linux/blkdev.h>
15 #include <linux/completion.h>
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/init.h>
19 #include <linux/pci.h>
20 #include <linux/delay.h>
21 #include <linux/hardirq.h>
22 #include <linux/scatterlist.h>
23 #include <linux/blk-mq.h>
24 #include <linux/blk-integrity.h>
25 #include <linux/ratelimit.h>
26 #include <asm/unaligned.h>
27 
28 #include <scsi/scsi.h>
29 #include <scsi/scsi_cmnd.h>
30 #include <scsi/scsi_dbg.h>
31 #include <scsi/scsi_device.h>
32 #include <scsi/scsi_driver.h>
33 #include <scsi/scsi_eh.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */
36 #include <scsi/scsi_dh.h>
37 
38 #include <trace/events/scsi.h>
39 
40 #include "scsi_debugfs.h"
41 #include "scsi_priv.h"
42 #include "scsi_logging.h"
43 
44 /*
45  * Size of integrity metadata is usually small, 1 inline sg should
46  * cover normal cases.
47  */
48 #ifdef CONFIG_ARCH_NO_SG_CHAIN
49 #define  SCSI_INLINE_PROT_SG_CNT  0
50 #define  SCSI_INLINE_SG_CNT  0
51 #else
52 #define  SCSI_INLINE_PROT_SG_CNT  1
53 #define  SCSI_INLINE_SG_CNT  2
54 #endif
55 
56 static struct kmem_cache *scsi_sense_cache;
57 static DEFINE_MUTEX(scsi_sense_cache_mutex);
58 
59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
60 
scsi_init_sense_cache(struct Scsi_Host * shost)61 int scsi_init_sense_cache(struct Scsi_Host *shost)
62 {
63 	int ret = 0;
64 
65 	mutex_lock(&scsi_sense_cache_mutex);
66 	if (!scsi_sense_cache) {
67 		scsi_sense_cache =
68 			kmem_cache_create_usercopy("scsi_sense_cache",
69 				SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
70 				0, SCSI_SENSE_BUFFERSIZE, NULL);
71 		if (!scsi_sense_cache)
72 			ret = -ENOMEM;
73 	}
74 	mutex_unlock(&scsi_sense_cache_mutex);
75 	return ret;
76 }
77 
78 static void
scsi_set_blocked(struct scsi_cmnd * cmd,int reason)79 scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
80 {
81 	struct Scsi_Host *host = cmd->device->host;
82 	struct scsi_device *device = cmd->device;
83 	struct scsi_target *starget = scsi_target(device);
84 
85 	/*
86 	 * Set the appropriate busy bit for the device/host.
87 	 *
88 	 * If the host/device isn't busy, assume that something actually
89 	 * completed, and that we should be able to queue a command now.
90 	 *
91 	 * Note that the prior mid-layer assumption that any host could
92 	 * always queue at least one command is now broken.  The mid-layer
93 	 * will implement a user specifiable stall (see
94 	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
95 	 * if a command is requeued with no other commands outstanding
96 	 * either for the device or for the host.
97 	 */
98 	switch (reason) {
99 	case SCSI_MLQUEUE_HOST_BUSY:
100 		atomic_set(&host->host_blocked, host->max_host_blocked);
101 		break;
102 	case SCSI_MLQUEUE_DEVICE_BUSY:
103 	case SCSI_MLQUEUE_EH_RETRY:
104 		atomic_set(&device->device_blocked,
105 			   device->max_device_blocked);
106 		break;
107 	case SCSI_MLQUEUE_TARGET_BUSY:
108 		atomic_set(&starget->target_blocked,
109 			   starget->max_target_blocked);
110 		break;
111 	}
112 }
113 
scsi_mq_requeue_cmd(struct scsi_cmnd * cmd,unsigned long msecs)114 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd, unsigned long msecs)
115 {
116 	struct request *rq = scsi_cmd_to_rq(cmd);
117 
118 	if (rq->rq_flags & RQF_DONTPREP) {
119 		rq->rq_flags &= ~RQF_DONTPREP;
120 		scsi_mq_uninit_cmd(cmd);
121 	} else {
122 		WARN_ON_ONCE(true);
123 	}
124 
125 	if (msecs) {
126 		blk_mq_requeue_request(rq, false);
127 		blk_mq_delay_kick_requeue_list(rq->q, msecs);
128 	} else
129 		blk_mq_requeue_request(rq, true);
130 }
131 
132 /**
133  * __scsi_queue_insert - private queue insertion
134  * @cmd: The SCSI command being requeued
135  * @reason:  The reason for the requeue
136  * @unbusy: Whether the queue should be unbusied
137  *
138  * This is a private queue insertion.  The public interface
139  * scsi_queue_insert() always assumes the queue should be unbusied
140  * because it's always called before the completion.  This function is
141  * for a requeue after completion, which should only occur in this
142  * file.
143  */
__scsi_queue_insert(struct scsi_cmnd * cmd,int reason,bool unbusy)144 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
145 {
146 	struct scsi_device *device = cmd->device;
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, cmd);
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_mq_destroy_queue() finishes.
165 	 */
166 	cmd->result = 0;
167 
168 	blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true);
169 }
170 
171 /**
172  * scsi_queue_insert - Reinsert a command in the queue.
173  * @cmd:    command that we are adding to queue.
174  * @reason: why we are inserting command to queue.
175  *
176  * We do this for one of two cases. Either the host is busy and it cannot accept
177  * any more commands for the time being, or the device returned QUEUE_FULL and
178  * can accept no more commands.
179  *
180  * Context: This could be called either from an interrupt context or a normal
181  * process context.
182  */
scsi_queue_insert(struct scsi_cmnd * cmd,int reason)183 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
184 {
185 	__scsi_queue_insert(cmd, reason, true);
186 }
187 
188 
189 /**
190  * __scsi_execute - insert request and wait for the result
191  * @sdev:	scsi device
192  * @cmd:	scsi command
193  * @data_direction: data direction
194  * @buffer:	data buffer
195  * @bufflen:	len of buffer
196  * @sense:	optional sense buffer
197  * @sshdr:	optional decoded sense header
198  * @timeout:	request timeout in HZ
199  * @retries:	number of times to retry request
200  * @flags:	flags for ->cmd_flags
201  * @rq_flags:	flags for ->rq_flags
202  * @resid:	optional residual length
203  *
204  * Returns the scsi_cmnd result field if a command was executed, or a negative
205  * Linux error code if we didn't get that far.
206  */
__scsi_execute(struct scsi_device * sdev,const unsigned char * cmd,int data_direction,void * buffer,unsigned bufflen,unsigned char * sense,struct scsi_sense_hdr * sshdr,int timeout,int retries,blk_opf_t flags,req_flags_t rq_flags,int * resid)207 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
208 		 int data_direction, void *buffer, unsigned bufflen,
209 		 unsigned char *sense, struct scsi_sense_hdr *sshdr,
210 		 int timeout, int retries, blk_opf_t flags,
211 		 req_flags_t rq_flags, int *resid)
212 {
213 	struct request *req;
214 	struct scsi_cmnd *scmd;
215 	int ret;
216 
217 	req = scsi_alloc_request(sdev->request_queue,
218 			data_direction == DMA_TO_DEVICE ?
219 			REQ_OP_DRV_OUT : REQ_OP_DRV_IN,
220 			rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0);
221 	if (IS_ERR(req))
222 		return PTR_ERR(req);
223 
224 	if (bufflen) {
225 		ret = blk_rq_map_kern(sdev->request_queue, req,
226 				      buffer, bufflen, GFP_NOIO);
227 		if (ret)
228 			goto out;
229 	}
230 	scmd = blk_mq_rq_to_pdu(req);
231 	scmd->cmd_len = COMMAND_SIZE(cmd[0]);
232 	memcpy(scmd->cmnd, cmd, scmd->cmd_len);
233 	scmd->allowed = retries;
234 	req->timeout = timeout;
235 	req->cmd_flags |= flags;
236 	req->rq_flags |= rq_flags | RQF_QUIET;
237 
238 	/*
239 	 * head injection *required* here otherwise quiesce won't work
240 	 */
241 	blk_execute_rq(req, true);
242 
243 	/*
244 	 * Some devices (USB mass-storage in particular) may transfer
245 	 * garbage data together with a residue indicating that the data
246 	 * is invalid.  Prevent the garbage from being misinterpreted
247 	 * and prevent security leaks by zeroing out the excess data.
248 	 */
249 	if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
250 		memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
251 
252 	if (resid)
253 		*resid = scmd->resid_len;
254 	if (sense && scmd->sense_len)
255 		memcpy(sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
256 	if (sshdr)
257 		scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
258 				     sshdr);
259 	ret = scmd->result;
260  out:
261 	blk_mq_free_request(req);
262 
263 	return ret;
264 }
265 EXPORT_SYMBOL(__scsi_execute);
266 
267 /*
268  * Wake up the error handler if necessary. Avoid as follows that the error
269  * handler is not woken up if host in-flight requests number ==
270  * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
271  * with an RCU read lock in this function to ensure that this function in
272  * its entirety either finishes before scsi_eh_scmd_add() increases the
273  * host_failed counter or that it notices the shost state change made by
274  * scsi_eh_scmd_add().
275  */
scsi_dec_host_busy(struct Scsi_Host * shost,struct scsi_cmnd * cmd)276 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
277 {
278 	unsigned long flags;
279 
280 	rcu_read_lock();
281 	__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
282 	if (unlikely(scsi_host_in_recovery(shost))) {
283 		spin_lock_irqsave(shost->host_lock, flags);
284 		if (shost->host_failed || shost->host_eh_scheduled)
285 			scsi_eh_wakeup(shost);
286 		spin_unlock_irqrestore(shost->host_lock, flags);
287 	}
288 	rcu_read_unlock();
289 }
290 
scsi_device_unbusy(struct scsi_device * sdev,struct scsi_cmnd * cmd)291 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
292 {
293 	struct Scsi_Host *shost = sdev->host;
294 	struct scsi_target *starget = scsi_target(sdev);
295 
296 	scsi_dec_host_busy(shost, cmd);
297 
298 	if (starget->can_queue > 0)
299 		atomic_dec(&starget->target_busy);
300 
301 	sbitmap_put(&sdev->budget_map, cmd->budget_token);
302 	cmd->budget_token = -1;
303 }
304 
scsi_kick_queue(struct request_queue * q)305 static void scsi_kick_queue(struct request_queue *q)
306 {
307 	blk_mq_run_hw_queues(q, false);
308 }
309 
310 /*
311  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
312  * and call blk_run_queue for all the scsi_devices on the target -
313  * including current_sdev first.
314  *
315  * Called with *no* scsi locks held.
316  */
scsi_single_lun_run(struct scsi_device * current_sdev)317 static void scsi_single_lun_run(struct scsi_device *current_sdev)
318 {
319 	struct Scsi_Host *shost = current_sdev->host;
320 	struct scsi_device *sdev, *tmp;
321 	struct scsi_target *starget = scsi_target(current_sdev);
322 	unsigned long flags;
323 
324 	spin_lock_irqsave(shost->host_lock, flags);
325 	starget->starget_sdev_user = NULL;
326 	spin_unlock_irqrestore(shost->host_lock, flags);
327 
328 	/*
329 	 * Call blk_run_queue for all LUNs on the target, starting with
330 	 * current_sdev. We race with others (to set starget_sdev_user),
331 	 * but in most cases, we will be first. Ideally, each LU on the
332 	 * target would get some limited time or requests on the target.
333 	 */
334 	scsi_kick_queue(current_sdev->request_queue);
335 
336 	spin_lock_irqsave(shost->host_lock, flags);
337 	if (starget->starget_sdev_user)
338 		goto out;
339 	list_for_each_entry_safe(sdev, tmp, &starget->devices,
340 			same_target_siblings) {
341 		if (sdev == current_sdev)
342 			continue;
343 		if (scsi_device_get(sdev))
344 			continue;
345 
346 		spin_unlock_irqrestore(shost->host_lock, flags);
347 		scsi_kick_queue(sdev->request_queue);
348 		spin_lock_irqsave(shost->host_lock, flags);
349 
350 		scsi_device_put(sdev);
351 	}
352  out:
353 	spin_unlock_irqrestore(shost->host_lock, flags);
354 }
355 
scsi_device_is_busy(struct scsi_device * sdev)356 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
357 {
358 	if (scsi_device_busy(sdev) >= sdev->queue_depth)
359 		return true;
360 	if (atomic_read(&sdev->device_blocked) > 0)
361 		return true;
362 	return false;
363 }
364 
scsi_target_is_busy(struct scsi_target * starget)365 static inline bool scsi_target_is_busy(struct scsi_target *starget)
366 {
367 	if (starget->can_queue > 0) {
368 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
369 			return true;
370 		if (atomic_read(&starget->target_blocked) > 0)
371 			return true;
372 	}
373 	return false;
374 }
375 
scsi_host_is_busy(struct Scsi_Host * shost)376 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
377 {
378 	if (atomic_read(&shost->host_blocked) > 0)
379 		return true;
380 	if (shost->host_self_blocked)
381 		return true;
382 	return false;
383 }
384 
scsi_starved_list_run(struct Scsi_Host * shost)385 static void scsi_starved_list_run(struct Scsi_Host *shost)
386 {
387 	LIST_HEAD(starved_list);
388 	struct scsi_device *sdev;
389 	unsigned long flags;
390 
391 	spin_lock_irqsave(shost->host_lock, flags);
392 	list_splice_init(&shost->starved_list, &starved_list);
393 
394 	while (!list_empty(&starved_list)) {
395 		struct request_queue *slq;
396 
397 		/*
398 		 * As long as shost is accepting commands and we have
399 		 * starved queues, call blk_run_queue. scsi_request_fn
400 		 * drops the queue_lock and can add us back to the
401 		 * starved_list.
402 		 *
403 		 * host_lock protects the starved_list and starved_entry.
404 		 * scsi_request_fn must get the host_lock before checking
405 		 * or modifying starved_list or starved_entry.
406 		 */
407 		if (scsi_host_is_busy(shost))
408 			break;
409 
410 		sdev = list_entry(starved_list.next,
411 				  struct scsi_device, starved_entry);
412 		list_del_init(&sdev->starved_entry);
413 		if (scsi_target_is_busy(scsi_target(sdev))) {
414 			list_move_tail(&sdev->starved_entry,
415 				       &shost->starved_list);
416 			continue;
417 		}
418 
419 		/*
420 		 * Once we drop the host lock, a racing scsi_remove_device()
421 		 * call may remove the sdev from the starved list and destroy
422 		 * it and the queue.  Mitigate by taking a reference to the
423 		 * queue and never touching the sdev again after we drop the
424 		 * host lock.  Note: if __scsi_remove_device() invokes
425 		 * blk_mq_destroy_queue() before the queue is run from this
426 		 * function then blk_run_queue() will return immediately since
427 		 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
428 		 */
429 		slq = sdev->request_queue;
430 		if (!blk_get_queue(slq))
431 			continue;
432 		spin_unlock_irqrestore(shost->host_lock, flags);
433 
434 		scsi_kick_queue(slq);
435 		blk_put_queue(slq);
436 
437 		spin_lock_irqsave(shost->host_lock, flags);
438 	}
439 	/* put any unprocessed entries back */
440 	list_splice(&starved_list, &shost->starved_list);
441 	spin_unlock_irqrestore(shost->host_lock, flags);
442 }
443 
444 /**
445  * scsi_run_queue - Select a proper request queue to serve next.
446  * @q:  last request's queue
447  *
448  * The previous command was completely finished, start a new one if possible.
449  */
scsi_run_queue(struct request_queue * q)450 static void scsi_run_queue(struct request_queue *q)
451 {
452 	struct scsi_device *sdev = q->queuedata;
453 
454 	if (scsi_target(sdev)->single_lun)
455 		scsi_single_lun_run(sdev);
456 	if (!list_empty(&sdev->host->starved_list))
457 		scsi_starved_list_run(sdev->host);
458 
459 	blk_mq_run_hw_queues(q, false);
460 }
461 
scsi_requeue_run_queue(struct work_struct * work)462 void scsi_requeue_run_queue(struct work_struct *work)
463 {
464 	struct scsi_device *sdev;
465 	struct request_queue *q;
466 
467 	sdev = container_of(work, struct scsi_device, requeue_work);
468 	q = sdev->request_queue;
469 	scsi_run_queue(q);
470 }
471 
scsi_run_host_queues(struct Scsi_Host * shost)472 void scsi_run_host_queues(struct Scsi_Host *shost)
473 {
474 	struct scsi_device *sdev;
475 
476 	shost_for_each_device(sdev, shost)
477 		scsi_run_queue(sdev->request_queue);
478 }
479 
scsi_uninit_cmd(struct scsi_cmnd * cmd)480 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
481 {
482 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
483 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
484 
485 		if (drv->uninit_command)
486 			drv->uninit_command(cmd);
487 	}
488 }
489 
scsi_free_sgtables(struct scsi_cmnd * cmd)490 void scsi_free_sgtables(struct scsi_cmnd *cmd)
491 {
492 	if (cmd->sdb.table.nents)
493 		sg_free_table_chained(&cmd->sdb.table,
494 				SCSI_INLINE_SG_CNT);
495 	if (scsi_prot_sg_count(cmd))
496 		sg_free_table_chained(&cmd->prot_sdb->table,
497 				SCSI_INLINE_PROT_SG_CNT);
498 }
499 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
500 
scsi_mq_uninit_cmd(struct scsi_cmnd * cmd)501 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
502 {
503 	scsi_free_sgtables(cmd);
504 	scsi_uninit_cmd(cmd);
505 }
506 
scsi_run_queue_async(struct scsi_device * sdev)507 static void scsi_run_queue_async(struct scsi_device *sdev)
508 {
509 	if (scsi_target(sdev)->single_lun ||
510 	    !list_empty(&sdev->host->starved_list)) {
511 		kblockd_schedule_work(&sdev->requeue_work);
512 	} else {
513 		/*
514 		 * smp_mb() present in sbitmap_queue_clear() or implied in
515 		 * .end_io is for ordering writing .device_busy in
516 		 * scsi_device_unbusy() and reading sdev->restarts.
517 		 */
518 		int old = atomic_read(&sdev->restarts);
519 
520 		/*
521 		 * ->restarts has to be kept as non-zero if new budget
522 		 *  contention occurs.
523 		 *
524 		 *  No need to run queue when either another re-run
525 		 *  queue wins in updating ->restarts or a new budget
526 		 *  contention occurs.
527 		 */
528 		if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
529 			blk_mq_run_hw_queues(sdev->request_queue, true);
530 	}
531 }
532 
533 /* Returns false when no more bytes to process, true if there are more */
scsi_end_request(struct request * req,blk_status_t error,unsigned int bytes)534 static bool scsi_end_request(struct request *req, blk_status_t error,
535 		unsigned int bytes)
536 {
537 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
538 	struct scsi_device *sdev = cmd->device;
539 	struct request_queue *q = sdev->request_queue;
540 
541 	if (blk_update_request(req, error, bytes))
542 		return true;
543 
544 	// XXX:
545 	if (blk_queue_add_random(q))
546 		add_disk_randomness(req->q->disk);
547 
548 	if (!blk_rq_is_passthrough(req)) {
549 		WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
550 		cmd->flags &= ~SCMD_INITIALIZED;
551 	}
552 
553 	/*
554 	 * Calling rcu_barrier() is not necessary here because the
555 	 * SCSI error handler guarantees that the function called by
556 	 * call_rcu() has been called before scsi_end_request() is
557 	 * called.
558 	 */
559 	destroy_rcu_head(&cmd->rcu);
560 
561 	/*
562 	 * In the MQ case the command gets freed by __blk_mq_end_request,
563 	 * so we have to do all cleanup that depends on it earlier.
564 	 *
565 	 * We also can't kick the queues from irq context, so we
566 	 * will have to defer it to a workqueue.
567 	 */
568 	scsi_mq_uninit_cmd(cmd);
569 
570 	/*
571 	 * queue is still alive, so grab the ref for preventing it
572 	 * from being cleaned up during running queue.
573 	 */
574 	percpu_ref_get(&q->q_usage_counter);
575 
576 	__blk_mq_end_request(req, error);
577 
578 	scsi_run_queue_async(sdev);
579 
580 	percpu_ref_put(&q->q_usage_counter);
581 	return false;
582 }
583 
get_scsi_ml_byte(int result)584 static inline u8 get_scsi_ml_byte(int result)
585 {
586 	return (result >> 8) & 0xff;
587 }
588 
589 /**
590  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
591  * @result:	scsi error code
592  *
593  * Translate a SCSI result code into a blk_status_t value.
594  */
scsi_result_to_blk_status(int result)595 static blk_status_t scsi_result_to_blk_status(int result)
596 {
597 	/*
598 	 * Check the scsi-ml byte first in case we converted a host or status
599 	 * byte.
600 	 */
601 	switch (get_scsi_ml_byte(result)) {
602 	case SCSIML_STAT_OK:
603 		break;
604 	case SCSIML_STAT_RESV_CONFLICT:
605 		return BLK_STS_NEXUS;
606 	case SCSIML_STAT_NOSPC:
607 		return BLK_STS_NOSPC;
608 	case SCSIML_STAT_MED_ERROR:
609 		return BLK_STS_MEDIUM;
610 	case SCSIML_STAT_TGT_FAILURE:
611 		return BLK_STS_TARGET;
612 	}
613 
614 	switch (host_byte(result)) {
615 	case DID_OK:
616 		if (scsi_status_is_good(result))
617 			return BLK_STS_OK;
618 		return BLK_STS_IOERR;
619 	case DID_TRANSPORT_FAILFAST:
620 	case DID_TRANSPORT_MARGINAL:
621 		return BLK_STS_TRANSPORT;
622 	default:
623 		return BLK_STS_IOERR;
624 	}
625 }
626 
627 /**
628  * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
629  * @rq: request to examine
630  *
631  * Description:
632  *     A request could be merge of IOs which require different failure
633  *     handling.  This function determines the number of bytes which
634  *     can be failed from the beginning of the request without
635  *     crossing into area which need to be retried further.
636  *
637  * Return:
638  *     The number of bytes to fail.
639  */
scsi_rq_err_bytes(const struct request * rq)640 static unsigned int scsi_rq_err_bytes(const struct request *rq)
641 {
642 	blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
643 	unsigned int bytes = 0;
644 	struct bio *bio;
645 
646 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
647 		return blk_rq_bytes(rq);
648 
649 	/*
650 	 * Currently the only 'mixing' which can happen is between
651 	 * different fastfail types.  We can safely fail portions
652 	 * which have all the failfast bits that the first one has -
653 	 * the ones which are at least as eager to fail as the first
654 	 * one.
655 	 */
656 	for (bio = rq->bio; bio; bio = bio->bi_next) {
657 		if ((bio->bi_opf & ff) != ff)
658 			break;
659 		bytes += bio->bi_iter.bi_size;
660 	}
661 
662 	/* this could lead to infinite loop */
663 	BUG_ON(blk_rq_bytes(rq) && !bytes);
664 	return bytes;
665 }
666 
scsi_cmd_runtime_exceeced(struct scsi_cmnd * cmd)667 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
668 {
669 	struct request *req = scsi_cmd_to_rq(cmd);
670 	unsigned long wait_for;
671 
672 	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
673 		return false;
674 
675 	wait_for = (cmd->allowed + 1) * req->timeout;
676 	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
677 		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
678 			    wait_for/HZ);
679 		return true;
680 	}
681 	return false;
682 }
683 
684 /*
685  * When ALUA transition state is returned, reprep the cmd to
686  * use the ALUA handler's transition timeout. Delay the reprep
687  * 1 sec to avoid aggressive retries of the target in that
688  * state.
689  */
690 #define ALUA_TRANSITION_REPREP_DELAY	1000
691 
692 /* Helper for scsi_io_completion() when special action required. */
scsi_io_completion_action(struct scsi_cmnd * cmd,int result)693 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
694 {
695 	struct request *req = scsi_cmd_to_rq(cmd);
696 	int level = 0;
697 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
698 	      ACTION_RETRY, ACTION_DELAYED_RETRY} action;
699 	struct scsi_sense_hdr sshdr;
700 	bool sense_valid;
701 	bool sense_current = true;      /* false implies "deferred sense" */
702 	blk_status_t blk_stat;
703 
704 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
705 	if (sense_valid)
706 		sense_current = !scsi_sense_is_deferred(&sshdr);
707 
708 	blk_stat = scsi_result_to_blk_status(result);
709 
710 	if (host_byte(result) == DID_RESET) {
711 		/* Third party bus reset or reset for error recovery
712 		 * reasons.  Just retry the command and see what
713 		 * happens.
714 		 */
715 		action = ACTION_RETRY;
716 	} else if (sense_valid && sense_current) {
717 		switch (sshdr.sense_key) {
718 		case UNIT_ATTENTION:
719 			if (cmd->device->removable) {
720 				/* Detected disc change.  Set a bit
721 				 * and quietly refuse further access.
722 				 */
723 				cmd->device->changed = 1;
724 				action = ACTION_FAIL;
725 			} else {
726 				/* Must have been a power glitch, or a
727 				 * bus reset.  Could not have been a
728 				 * media change, so we just retry the
729 				 * command and see what happens.
730 				 */
731 				action = ACTION_RETRY;
732 			}
733 			break;
734 		case ILLEGAL_REQUEST:
735 			/* If we had an ILLEGAL REQUEST returned, then
736 			 * we may have performed an unsupported
737 			 * command.  The only thing this should be
738 			 * would be a ten byte read where only a six
739 			 * byte read was supported.  Also, on a system
740 			 * where READ CAPACITY failed, we may have
741 			 * read past the end of the disk.
742 			 */
743 			if ((cmd->device->use_10_for_rw &&
744 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
745 			    (cmd->cmnd[0] == READ_10 ||
746 			     cmd->cmnd[0] == WRITE_10)) {
747 				/* This will issue a new 6-byte command. */
748 				cmd->device->use_10_for_rw = 0;
749 				action = ACTION_REPREP;
750 			} else if (sshdr.asc == 0x10) /* DIX */ {
751 				action = ACTION_FAIL;
752 				blk_stat = BLK_STS_PROTECTION;
753 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
754 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
755 				action = ACTION_FAIL;
756 				blk_stat = BLK_STS_TARGET;
757 			} else
758 				action = ACTION_FAIL;
759 			break;
760 		case ABORTED_COMMAND:
761 			action = ACTION_FAIL;
762 			if (sshdr.asc == 0x10) /* DIF */
763 				blk_stat = BLK_STS_PROTECTION;
764 			break;
765 		case NOT_READY:
766 			/* If the device is in the process of becoming
767 			 * ready, or has a temporary blockage, retry.
768 			 */
769 			if (sshdr.asc == 0x04) {
770 				switch (sshdr.ascq) {
771 				case 0x01: /* becoming ready */
772 				case 0x04: /* format in progress */
773 				case 0x05: /* rebuild in progress */
774 				case 0x06: /* recalculation in progress */
775 				case 0x07: /* operation in progress */
776 				case 0x08: /* Long write in progress */
777 				case 0x09: /* self test in progress */
778 				case 0x11: /* notify (enable spinup) required */
779 				case 0x14: /* space allocation in progress */
780 				case 0x1a: /* start stop unit in progress */
781 				case 0x1b: /* sanitize in progress */
782 				case 0x1d: /* configuration in progress */
783 				case 0x24: /* depopulation in progress */
784 					action = ACTION_DELAYED_RETRY;
785 					break;
786 				case 0x0a: /* ALUA state transition */
787 					action = ACTION_DELAYED_REPREP;
788 					break;
789 				default:
790 					action = ACTION_FAIL;
791 					break;
792 				}
793 			} else
794 				action = ACTION_FAIL;
795 			break;
796 		case VOLUME_OVERFLOW:
797 			/* See SSC3rXX or current. */
798 			action = ACTION_FAIL;
799 			break;
800 		case DATA_PROTECT:
801 			action = ACTION_FAIL;
802 			if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
803 			    (sshdr.asc == 0x55 &&
804 			     (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
805 				/* Insufficient zone resources */
806 				blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
807 			}
808 			break;
809 		default:
810 			action = ACTION_FAIL;
811 			break;
812 		}
813 	} else
814 		action = ACTION_FAIL;
815 
816 	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
817 		action = ACTION_FAIL;
818 
819 	switch (action) {
820 	case ACTION_FAIL:
821 		/* Give up and fail the remainder of the request */
822 		if (!(req->rq_flags & RQF_QUIET)) {
823 			static DEFINE_RATELIMIT_STATE(_rs,
824 					DEFAULT_RATELIMIT_INTERVAL,
825 					DEFAULT_RATELIMIT_BURST);
826 
827 			if (unlikely(scsi_logging_level))
828 				level =
829 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
830 						    SCSI_LOG_MLCOMPLETE_BITS);
831 
832 			/*
833 			 * if logging is enabled the failure will be printed
834 			 * in scsi_log_completion(), so avoid duplicate messages
835 			 */
836 			if (!level && __ratelimit(&_rs)) {
837 				scsi_print_result(cmd, NULL, FAILED);
838 				if (sense_valid)
839 					scsi_print_sense(cmd);
840 				scsi_print_command(cmd);
841 			}
842 		}
843 		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
844 			return;
845 		fallthrough;
846 	case ACTION_REPREP:
847 		scsi_mq_requeue_cmd(cmd, 0);
848 		break;
849 	case ACTION_DELAYED_REPREP:
850 		scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
851 		break;
852 	case ACTION_RETRY:
853 		/* Retry the same command immediately */
854 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
855 		break;
856 	case ACTION_DELAYED_RETRY:
857 		/* Retry the same command after a delay */
858 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
859 		break;
860 	}
861 }
862 
863 /*
864  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
865  * new result that may suppress further error checking. Also modifies
866  * *blk_statp in some cases.
867  */
scsi_io_completion_nz_result(struct scsi_cmnd * cmd,int result,blk_status_t * blk_statp)868 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
869 					blk_status_t *blk_statp)
870 {
871 	bool sense_valid;
872 	bool sense_current = true;	/* false implies "deferred sense" */
873 	struct request *req = scsi_cmd_to_rq(cmd);
874 	struct scsi_sense_hdr sshdr;
875 
876 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
877 	if (sense_valid)
878 		sense_current = !scsi_sense_is_deferred(&sshdr);
879 
880 	if (blk_rq_is_passthrough(req)) {
881 		if (sense_valid) {
882 			/*
883 			 * SG_IO wants current and deferred errors
884 			 */
885 			cmd->sense_len = min(8 + cmd->sense_buffer[7],
886 					     SCSI_SENSE_BUFFERSIZE);
887 		}
888 		if (sense_current)
889 			*blk_statp = scsi_result_to_blk_status(result);
890 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
891 		/*
892 		 * Flush commands do not transfers any data, and thus cannot use
893 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
894 		 * This sets *blk_statp explicitly for the problem case.
895 		 */
896 		*blk_statp = scsi_result_to_blk_status(result);
897 	}
898 	/*
899 	 * Recovered errors need reporting, but they're always treated as
900 	 * success, so fiddle the result code here.  For passthrough requests
901 	 * we already took a copy of the original into sreq->result which
902 	 * is what gets returned to the user
903 	 */
904 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
905 		bool do_print = true;
906 		/*
907 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
908 		 * skip print since caller wants ATA registers. Only occurs
909 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
910 		 */
911 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
912 			do_print = false;
913 		else if (req->rq_flags & RQF_QUIET)
914 			do_print = false;
915 		if (do_print)
916 			scsi_print_sense(cmd);
917 		result = 0;
918 		/* for passthrough, *blk_statp may be set */
919 		*blk_statp = BLK_STS_OK;
920 	}
921 	/*
922 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
923 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
924 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
925 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
926 	 * intermediate statuses (both obsolete in SAM-4) as good.
927 	 */
928 	if ((result & 0xff) && scsi_status_is_good(result)) {
929 		result = 0;
930 		*blk_statp = BLK_STS_OK;
931 	}
932 	return result;
933 }
934 
935 /**
936  * scsi_io_completion - Completion processing for SCSI commands.
937  * @cmd:	command that is finished.
938  * @good_bytes:	number of processed bytes.
939  *
940  * We will finish off the specified number of sectors. If we are done, the
941  * command block will be released and the queue function will be goosed. If we
942  * are not done then we have to figure out what to do next:
943  *
944  *   a) We can call scsi_mq_requeue_cmd().  The request will be
945  *	unprepared and put back on the queue.  Then a new command will
946  *	be created for it.  This should be used if we made forward
947  *	progress, or if we want to switch from READ(10) to READ(6) for
948  *	example.
949  *
950  *   b) We can call scsi_io_completion_action().  The request will be
951  *	put back on the queue and retried using the same command as
952  *	before, possibly after a delay.
953  *
954  *   c) We can call scsi_end_request() with blk_stat other than
955  *	BLK_STS_OK, to fail the remainder of the request.
956  */
scsi_io_completion(struct scsi_cmnd * cmd,unsigned int good_bytes)957 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
958 {
959 	int result = cmd->result;
960 	struct request *req = scsi_cmd_to_rq(cmd);
961 	blk_status_t blk_stat = BLK_STS_OK;
962 
963 	if (unlikely(result))	/* a nz result may or may not be an error */
964 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
965 
966 	/*
967 	 * Next deal with any sectors which we were able to correctly
968 	 * handle.
969 	 */
970 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
971 		"%u sectors total, %d bytes done.\n",
972 		blk_rq_sectors(req), good_bytes));
973 
974 	/*
975 	 * Failed, zero length commands always need to drop down
976 	 * to retry code. Fast path should return in this block.
977 	 */
978 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
979 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
980 			return; /* no bytes remaining */
981 	}
982 
983 	/* Kill remainder if no retries. */
984 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
985 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
986 			WARN_ONCE(true,
987 			    "Bytes remaining after failed, no-retry command");
988 		return;
989 	}
990 
991 	/*
992 	 * If there had been no error, but we have leftover bytes in the
993 	 * request just queue the command up again.
994 	 */
995 	if (likely(result == 0))
996 		scsi_mq_requeue_cmd(cmd, 0);
997 	else
998 		scsi_io_completion_action(cmd, result);
999 }
1000 
scsi_cmd_needs_dma_drain(struct scsi_device * sdev,struct request * rq)1001 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1002 		struct request *rq)
1003 {
1004 	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1005 	       !op_is_write(req_op(rq)) &&
1006 	       sdev->host->hostt->dma_need_drain(rq);
1007 }
1008 
1009 /**
1010  * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1011  * @cmd: SCSI command data structure to initialize.
1012  *
1013  * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1014  * for @cmd.
1015  *
1016  * Returns:
1017  * * BLK_STS_OK       - on success
1018  * * BLK_STS_RESOURCE - if the failure is retryable
1019  * * BLK_STS_IOERR    - if the failure is fatal
1020  */
scsi_alloc_sgtables(struct scsi_cmnd * cmd)1021 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1022 {
1023 	struct scsi_device *sdev = cmd->device;
1024 	struct request *rq = scsi_cmd_to_rq(cmd);
1025 	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1026 	struct scatterlist *last_sg = NULL;
1027 	blk_status_t ret;
1028 	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1029 	int count;
1030 
1031 	if (WARN_ON_ONCE(!nr_segs))
1032 		return BLK_STS_IOERR;
1033 
1034 	/*
1035 	 * Make sure there is space for the drain.  The driver must adjust
1036 	 * max_hw_segments to be prepared for this.
1037 	 */
1038 	if (need_drain)
1039 		nr_segs++;
1040 
1041 	/*
1042 	 * If sg table allocation fails, requeue request later.
1043 	 */
1044 	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1045 			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1046 		return BLK_STS_RESOURCE;
1047 
1048 	/*
1049 	 * Next, walk the list, and fill in the addresses and sizes of
1050 	 * each segment.
1051 	 */
1052 	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1053 
1054 	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1055 		unsigned int pad_len =
1056 			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1057 
1058 		last_sg->length += pad_len;
1059 		cmd->extra_len += pad_len;
1060 	}
1061 
1062 	if (need_drain) {
1063 		sg_unmark_end(last_sg);
1064 		last_sg = sg_next(last_sg);
1065 		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1066 		sg_mark_end(last_sg);
1067 
1068 		cmd->extra_len += sdev->dma_drain_len;
1069 		count++;
1070 	}
1071 
1072 	BUG_ON(count > cmd->sdb.table.nents);
1073 	cmd->sdb.table.nents = count;
1074 	cmd->sdb.length = blk_rq_payload_bytes(rq);
1075 
1076 	if (blk_integrity_rq(rq)) {
1077 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1078 		int ivecs;
1079 
1080 		if (WARN_ON_ONCE(!prot_sdb)) {
1081 			/*
1082 			 * This can happen if someone (e.g. multipath)
1083 			 * queues a command to a device on an adapter
1084 			 * that does not support DIX.
1085 			 */
1086 			ret = BLK_STS_IOERR;
1087 			goto out_free_sgtables;
1088 		}
1089 
1090 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1091 
1092 		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1093 				prot_sdb->table.sgl,
1094 				SCSI_INLINE_PROT_SG_CNT)) {
1095 			ret = BLK_STS_RESOURCE;
1096 			goto out_free_sgtables;
1097 		}
1098 
1099 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1100 						prot_sdb->table.sgl);
1101 		BUG_ON(count > ivecs);
1102 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1103 
1104 		cmd->prot_sdb = prot_sdb;
1105 		cmd->prot_sdb->table.nents = count;
1106 	}
1107 
1108 	return BLK_STS_OK;
1109 out_free_sgtables:
1110 	scsi_free_sgtables(cmd);
1111 	return ret;
1112 }
1113 EXPORT_SYMBOL(scsi_alloc_sgtables);
1114 
1115 /**
1116  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1117  * @rq: Request associated with the SCSI command to be initialized.
1118  *
1119  * This function initializes the members of struct scsi_cmnd that must be
1120  * initialized before request processing starts and that won't be
1121  * reinitialized if a SCSI command is requeued.
1122  */
scsi_initialize_rq(struct request * rq)1123 static void scsi_initialize_rq(struct request *rq)
1124 {
1125 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1126 
1127 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1128 	cmd->cmd_len = MAX_COMMAND_SIZE;
1129 	cmd->sense_len = 0;
1130 	init_rcu_head(&cmd->rcu);
1131 	cmd->jiffies_at_alloc = jiffies;
1132 	cmd->retries = 0;
1133 }
1134 
scsi_alloc_request(struct request_queue * q,blk_opf_t opf,blk_mq_req_flags_t flags)1135 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1136 				   blk_mq_req_flags_t flags)
1137 {
1138 	struct request *rq;
1139 
1140 	rq = blk_mq_alloc_request(q, opf, flags);
1141 	if (!IS_ERR(rq))
1142 		scsi_initialize_rq(rq);
1143 	return rq;
1144 }
1145 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1146 
1147 /*
1148  * Only called when the request isn't completed by SCSI, and not freed by
1149  * SCSI
1150  */
scsi_cleanup_rq(struct request * rq)1151 static void scsi_cleanup_rq(struct request *rq)
1152 {
1153 	if (rq->rq_flags & RQF_DONTPREP) {
1154 		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1155 		rq->rq_flags &= ~RQF_DONTPREP;
1156 	}
1157 }
1158 
1159 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
scsi_init_command(struct scsi_device * dev,struct scsi_cmnd * cmd)1160 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1161 {
1162 	struct request *rq = scsi_cmd_to_rq(cmd);
1163 
1164 	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1165 		cmd->flags |= SCMD_INITIALIZED;
1166 		scsi_initialize_rq(rq);
1167 	}
1168 
1169 	cmd->device = dev;
1170 	INIT_LIST_HEAD(&cmd->eh_entry);
1171 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1172 }
1173 
scsi_setup_scsi_cmnd(struct scsi_device * sdev,struct request * req)1174 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1175 		struct request *req)
1176 {
1177 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1178 
1179 	/*
1180 	 * Passthrough requests may transfer data, in which case they must
1181 	 * a bio attached to them.  Or they might contain a SCSI command
1182 	 * that does not transfer data, in which case they may optionally
1183 	 * submit a request without an attached bio.
1184 	 */
1185 	if (req->bio) {
1186 		blk_status_t ret = scsi_alloc_sgtables(cmd);
1187 		if (unlikely(ret != BLK_STS_OK))
1188 			return ret;
1189 	} else {
1190 		BUG_ON(blk_rq_bytes(req));
1191 
1192 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1193 	}
1194 
1195 	cmd->transfersize = blk_rq_bytes(req);
1196 	return BLK_STS_OK;
1197 }
1198 
1199 static blk_status_t
scsi_device_state_check(struct scsi_device * sdev,struct request * req)1200 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1201 {
1202 	switch (sdev->sdev_state) {
1203 	case SDEV_CREATED:
1204 		return BLK_STS_OK;
1205 	case SDEV_OFFLINE:
1206 	case SDEV_TRANSPORT_OFFLINE:
1207 		/*
1208 		 * If the device is offline we refuse to process any
1209 		 * commands.  The device must be brought online
1210 		 * before trying any recovery commands.
1211 		 */
1212 		if (!sdev->offline_already) {
1213 			sdev->offline_already = true;
1214 			sdev_printk(KERN_ERR, sdev,
1215 				    "rejecting I/O to offline device\n");
1216 		}
1217 		return BLK_STS_IOERR;
1218 	case SDEV_DEL:
1219 		/*
1220 		 * If the device is fully deleted, we refuse to
1221 		 * process any commands as well.
1222 		 */
1223 		sdev_printk(KERN_ERR, sdev,
1224 			    "rejecting I/O to dead device\n");
1225 		return BLK_STS_IOERR;
1226 	case SDEV_BLOCK:
1227 	case SDEV_CREATED_BLOCK:
1228 		return BLK_STS_RESOURCE;
1229 	case SDEV_QUIESCE:
1230 		/*
1231 		 * If the device is blocked we only accept power management
1232 		 * commands.
1233 		 */
1234 		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1235 			return BLK_STS_RESOURCE;
1236 		return BLK_STS_OK;
1237 	default:
1238 		/*
1239 		 * For any other not fully online state we only allow
1240 		 * power management commands.
1241 		 */
1242 		if (req && !(req->rq_flags & RQF_PM))
1243 			return BLK_STS_OFFLINE;
1244 		return BLK_STS_OK;
1245 	}
1246 }
1247 
1248 /*
1249  * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1250  * and return the token else return -1.
1251  */
scsi_dev_queue_ready(struct request_queue * q,struct scsi_device * sdev)1252 static inline int scsi_dev_queue_ready(struct request_queue *q,
1253 				  struct scsi_device *sdev)
1254 {
1255 	int token;
1256 
1257 	token = sbitmap_get(&sdev->budget_map);
1258 	if (atomic_read(&sdev->device_blocked)) {
1259 		if (token < 0)
1260 			goto out;
1261 
1262 		if (scsi_device_busy(sdev) > 1)
1263 			goto out_dec;
1264 
1265 		/*
1266 		 * unblock after device_blocked iterates to zero
1267 		 */
1268 		if (atomic_dec_return(&sdev->device_blocked) > 0)
1269 			goto out_dec;
1270 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1271 				   "unblocking device at zero depth\n"));
1272 	}
1273 
1274 	return token;
1275 out_dec:
1276 	if (token >= 0)
1277 		sbitmap_put(&sdev->budget_map, token);
1278 out:
1279 	return -1;
1280 }
1281 
1282 /*
1283  * scsi_target_queue_ready: checks if there we can send commands to target
1284  * @sdev: scsi device on starget to check.
1285  */
scsi_target_queue_ready(struct Scsi_Host * shost,struct scsi_device * sdev)1286 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1287 					   struct scsi_device *sdev)
1288 {
1289 	struct scsi_target *starget = scsi_target(sdev);
1290 	unsigned int busy;
1291 
1292 	if (starget->single_lun) {
1293 		spin_lock_irq(shost->host_lock);
1294 		if (starget->starget_sdev_user &&
1295 		    starget->starget_sdev_user != sdev) {
1296 			spin_unlock_irq(shost->host_lock);
1297 			return 0;
1298 		}
1299 		starget->starget_sdev_user = sdev;
1300 		spin_unlock_irq(shost->host_lock);
1301 	}
1302 
1303 	if (starget->can_queue <= 0)
1304 		return 1;
1305 
1306 	busy = atomic_inc_return(&starget->target_busy) - 1;
1307 	if (atomic_read(&starget->target_blocked) > 0) {
1308 		if (busy)
1309 			goto starved;
1310 
1311 		/*
1312 		 * unblock after target_blocked iterates to zero
1313 		 */
1314 		if (atomic_dec_return(&starget->target_blocked) > 0)
1315 			goto out_dec;
1316 
1317 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1318 				 "unblocking target at zero depth\n"));
1319 	}
1320 
1321 	if (busy >= starget->can_queue)
1322 		goto starved;
1323 
1324 	return 1;
1325 
1326 starved:
1327 	spin_lock_irq(shost->host_lock);
1328 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1329 	spin_unlock_irq(shost->host_lock);
1330 out_dec:
1331 	if (starget->can_queue > 0)
1332 		atomic_dec(&starget->target_busy);
1333 	return 0;
1334 }
1335 
1336 /*
1337  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1338  * return 0. We must end up running the queue again whenever 0 is
1339  * returned, else IO can hang.
1340  */
scsi_host_queue_ready(struct request_queue * q,struct Scsi_Host * shost,struct scsi_device * sdev,struct scsi_cmnd * cmd)1341 static inline int scsi_host_queue_ready(struct request_queue *q,
1342 				   struct Scsi_Host *shost,
1343 				   struct scsi_device *sdev,
1344 				   struct scsi_cmnd *cmd)
1345 {
1346 	if (scsi_host_in_recovery(shost))
1347 		return 0;
1348 
1349 	if (atomic_read(&shost->host_blocked) > 0) {
1350 		if (scsi_host_busy(shost) > 0)
1351 			goto starved;
1352 
1353 		/*
1354 		 * unblock after host_blocked iterates to zero
1355 		 */
1356 		if (atomic_dec_return(&shost->host_blocked) > 0)
1357 			goto out_dec;
1358 
1359 		SCSI_LOG_MLQUEUE(3,
1360 			shost_printk(KERN_INFO, shost,
1361 				     "unblocking host at zero depth\n"));
1362 	}
1363 
1364 	if (shost->host_self_blocked)
1365 		goto starved;
1366 
1367 	/* We're OK to process the command, so we can't be starved */
1368 	if (!list_empty(&sdev->starved_entry)) {
1369 		spin_lock_irq(shost->host_lock);
1370 		if (!list_empty(&sdev->starved_entry))
1371 			list_del_init(&sdev->starved_entry);
1372 		spin_unlock_irq(shost->host_lock);
1373 	}
1374 
1375 	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1376 
1377 	return 1;
1378 
1379 starved:
1380 	spin_lock_irq(shost->host_lock);
1381 	if (list_empty(&sdev->starved_entry))
1382 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1383 	spin_unlock_irq(shost->host_lock);
1384 out_dec:
1385 	scsi_dec_host_busy(shost, cmd);
1386 	return 0;
1387 }
1388 
1389 /*
1390  * Busy state exporting function for request stacking drivers.
1391  *
1392  * For efficiency, no lock is taken to check the busy state of
1393  * shost/starget/sdev, since the returned value is not guaranteed and
1394  * may be changed after request stacking drivers call the function,
1395  * regardless of taking lock or not.
1396  *
1397  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1398  * needs to return 'not busy'. Otherwise, request stacking drivers
1399  * may hold requests forever.
1400  */
scsi_mq_lld_busy(struct request_queue * q)1401 static bool scsi_mq_lld_busy(struct request_queue *q)
1402 {
1403 	struct scsi_device *sdev = q->queuedata;
1404 	struct Scsi_Host *shost;
1405 
1406 	if (blk_queue_dying(q))
1407 		return false;
1408 
1409 	shost = sdev->host;
1410 
1411 	/*
1412 	 * Ignore host/starget busy state.
1413 	 * Since block layer does not have a concept of fairness across
1414 	 * multiple queues, congestion of host/starget needs to be handled
1415 	 * in SCSI layer.
1416 	 */
1417 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1418 		return true;
1419 
1420 	return false;
1421 }
1422 
1423 /*
1424  * Block layer request completion callback. May be called from interrupt
1425  * context.
1426  */
scsi_complete(struct request * rq)1427 static void scsi_complete(struct request *rq)
1428 {
1429 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1430 	enum scsi_disposition disposition;
1431 
1432 	INIT_LIST_HEAD(&cmd->eh_entry);
1433 
1434 	atomic_inc(&cmd->device->iodone_cnt);
1435 	if (cmd->result)
1436 		atomic_inc(&cmd->device->ioerr_cnt);
1437 
1438 	disposition = scsi_decide_disposition(cmd);
1439 	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1440 		disposition = SUCCESS;
1441 
1442 	scsi_log_completion(cmd, disposition);
1443 
1444 	switch (disposition) {
1445 	case SUCCESS:
1446 		scsi_finish_command(cmd);
1447 		break;
1448 	case NEEDS_RETRY:
1449 		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1450 		break;
1451 	case ADD_TO_MLQUEUE:
1452 		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1453 		break;
1454 	default:
1455 		scsi_eh_scmd_add(cmd);
1456 		break;
1457 	}
1458 }
1459 
1460 /**
1461  * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1462  * @cmd: command block we are dispatching.
1463  *
1464  * Return: nonzero return request was rejected and device's queue needs to be
1465  * plugged.
1466  */
scsi_dispatch_cmd(struct scsi_cmnd * cmd)1467 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1468 {
1469 	struct Scsi_Host *host = cmd->device->host;
1470 	int rtn = 0;
1471 
1472 	atomic_inc(&cmd->device->iorequest_cnt);
1473 
1474 	/* check if the device is still usable */
1475 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1476 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1477 		 * returns an immediate error upwards, and signals
1478 		 * that the device is no longer present */
1479 		cmd->result = DID_NO_CONNECT << 16;
1480 		goto done;
1481 	}
1482 
1483 	/* Check to see if the scsi lld made this device blocked. */
1484 	if (unlikely(scsi_device_blocked(cmd->device))) {
1485 		/*
1486 		 * in blocked state, the command is just put back on
1487 		 * the device queue.  The suspend state has already
1488 		 * blocked the queue so future requests should not
1489 		 * occur until the device transitions out of the
1490 		 * suspend state.
1491 		 */
1492 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1493 			"queuecommand : device blocked\n"));
1494 		return SCSI_MLQUEUE_DEVICE_BUSY;
1495 	}
1496 
1497 	/* Store the LUN value in cmnd, if needed. */
1498 	if (cmd->device->lun_in_cdb)
1499 		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1500 			       (cmd->device->lun << 5 & 0xe0);
1501 
1502 	scsi_log_send(cmd);
1503 
1504 	/*
1505 	 * Before we queue this command, check if the command
1506 	 * length exceeds what the host adapter can handle.
1507 	 */
1508 	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1509 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1510 			       "queuecommand : command too long. "
1511 			       "cdb_size=%d host->max_cmd_len=%d\n",
1512 			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1513 		cmd->result = (DID_ABORT << 16);
1514 		goto done;
1515 	}
1516 
1517 	if (unlikely(host->shost_state == SHOST_DEL)) {
1518 		cmd->result = (DID_NO_CONNECT << 16);
1519 		goto done;
1520 
1521 	}
1522 
1523 	trace_scsi_dispatch_cmd_start(cmd);
1524 	rtn = host->hostt->queuecommand(host, cmd);
1525 	if (rtn) {
1526 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1527 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1528 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1529 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1530 
1531 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1532 			"queuecommand : request rejected\n"));
1533 	}
1534 
1535 	return rtn;
1536  done:
1537 	scsi_done(cmd);
1538 	return 0;
1539 }
1540 
1541 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
scsi_mq_inline_sgl_size(struct Scsi_Host * shost)1542 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1543 {
1544 	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1545 		sizeof(struct scatterlist);
1546 }
1547 
scsi_prepare_cmd(struct request * req)1548 static blk_status_t scsi_prepare_cmd(struct request *req)
1549 {
1550 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1551 	struct scsi_device *sdev = req->q->queuedata;
1552 	struct Scsi_Host *shost = sdev->host;
1553 	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1554 	struct scatterlist *sg;
1555 
1556 	scsi_init_command(sdev, cmd);
1557 
1558 	cmd->eh_eflags = 0;
1559 	cmd->prot_type = 0;
1560 	cmd->prot_flags = 0;
1561 	cmd->submitter = 0;
1562 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1563 	cmd->underflow = 0;
1564 	cmd->transfersize = 0;
1565 	cmd->host_scribble = NULL;
1566 	cmd->result = 0;
1567 	cmd->extra_len = 0;
1568 	cmd->state = 0;
1569 	if (in_flight)
1570 		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1571 
1572 	/*
1573 	 * Only clear the driver-private command data if the LLD does not supply
1574 	 * a function to initialize that data.
1575 	 */
1576 	if (!shost->hostt->init_cmd_priv)
1577 		memset(cmd + 1, 0, shost->hostt->cmd_size);
1578 
1579 	cmd->prot_op = SCSI_PROT_NORMAL;
1580 	if (blk_rq_bytes(req))
1581 		cmd->sc_data_direction = rq_dma_dir(req);
1582 	else
1583 		cmd->sc_data_direction = DMA_NONE;
1584 
1585 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1586 	cmd->sdb.table.sgl = sg;
1587 
1588 	if (scsi_host_get_prot(shost)) {
1589 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1590 
1591 		cmd->prot_sdb->table.sgl =
1592 			(struct scatterlist *)(cmd->prot_sdb + 1);
1593 	}
1594 
1595 	/*
1596 	 * Special handling for passthrough commands, which don't go to the ULP
1597 	 * at all:
1598 	 */
1599 	if (blk_rq_is_passthrough(req))
1600 		return scsi_setup_scsi_cmnd(sdev, req);
1601 
1602 	if (sdev->handler && sdev->handler->prep_fn) {
1603 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1604 
1605 		if (ret != BLK_STS_OK)
1606 			return ret;
1607 	}
1608 
1609 	/* Usually overridden by the ULP */
1610 	cmd->allowed = 0;
1611 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1612 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1613 }
1614 
scsi_done_internal(struct scsi_cmnd * cmd,bool complete_directly)1615 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1616 {
1617 	struct request *req = scsi_cmd_to_rq(cmd);
1618 
1619 	switch (cmd->submitter) {
1620 	case SUBMITTED_BY_BLOCK_LAYER:
1621 		break;
1622 	case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1623 		return scsi_eh_done(cmd);
1624 	case SUBMITTED_BY_SCSI_RESET_IOCTL:
1625 		return;
1626 	}
1627 
1628 	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1629 		return;
1630 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1631 		return;
1632 	trace_scsi_dispatch_cmd_done(cmd);
1633 
1634 	if (complete_directly)
1635 		blk_mq_complete_request_direct(req, scsi_complete);
1636 	else
1637 		blk_mq_complete_request(req);
1638 }
1639 
scsi_done(struct scsi_cmnd * cmd)1640 void scsi_done(struct scsi_cmnd *cmd)
1641 {
1642 	scsi_done_internal(cmd, false);
1643 }
1644 EXPORT_SYMBOL(scsi_done);
1645 
scsi_done_direct(struct scsi_cmnd * cmd)1646 void scsi_done_direct(struct scsi_cmnd *cmd)
1647 {
1648 	scsi_done_internal(cmd, true);
1649 }
1650 EXPORT_SYMBOL(scsi_done_direct);
1651 
scsi_mq_put_budget(struct request_queue * q,int budget_token)1652 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1653 {
1654 	struct scsi_device *sdev = q->queuedata;
1655 
1656 	sbitmap_put(&sdev->budget_map, budget_token);
1657 }
1658 
1659 /*
1660  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1661  * not change behaviour from the previous unplug mechanism, experimentation
1662  * may prove this needs changing.
1663  */
1664 #define SCSI_QUEUE_DELAY 3
1665 
scsi_mq_get_budget(struct request_queue * q)1666 static int scsi_mq_get_budget(struct request_queue *q)
1667 {
1668 	struct scsi_device *sdev = q->queuedata;
1669 	int token = scsi_dev_queue_ready(q, sdev);
1670 
1671 	if (token >= 0)
1672 		return token;
1673 
1674 	atomic_inc(&sdev->restarts);
1675 
1676 	/*
1677 	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1678 	 * .restarts must be incremented before .device_busy is read because the
1679 	 * code in scsi_run_queue_async() depends on the order of these operations.
1680 	 */
1681 	smp_mb__after_atomic();
1682 
1683 	/*
1684 	 * If all in-flight requests originated from this LUN are completed
1685 	 * before reading .device_busy, sdev->device_busy will be observed as
1686 	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1687 	 * soon. Otherwise, completion of one of these requests will observe
1688 	 * the .restarts flag, and the request queue will be run for handling
1689 	 * this request, see scsi_end_request().
1690 	 */
1691 	if (unlikely(scsi_device_busy(sdev) == 0 &&
1692 				!scsi_device_blocked(sdev)))
1693 		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1694 	return -1;
1695 }
1696 
scsi_mq_set_rq_budget_token(struct request * req,int token)1697 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1698 {
1699 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1700 
1701 	cmd->budget_token = token;
1702 }
1703 
scsi_mq_get_rq_budget_token(struct request * req)1704 static int scsi_mq_get_rq_budget_token(struct request *req)
1705 {
1706 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1707 
1708 	return cmd->budget_token;
1709 }
1710 
scsi_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)1711 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1712 			 const struct blk_mq_queue_data *bd)
1713 {
1714 	struct request *req = bd->rq;
1715 	struct request_queue *q = req->q;
1716 	struct scsi_device *sdev = q->queuedata;
1717 	struct Scsi_Host *shost = sdev->host;
1718 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1719 	blk_status_t ret;
1720 	int reason;
1721 
1722 	WARN_ON_ONCE(cmd->budget_token < 0);
1723 
1724 	/*
1725 	 * If the device is not in running state we will reject some or all
1726 	 * commands.
1727 	 */
1728 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1729 		ret = scsi_device_state_check(sdev, req);
1730 		if (ret != BLK_STS_OK)
1731 			goto out_put_budget;
1732 	}
1733 
1734 	ret = BLK_STS_RESOURCE;
1735 	if (!scsi_target_queue_ready(shost, sdev))
1736 		goto out_put_budget;
1737 	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1738 		goto out_dec_target_busy;
1739 
1740 	if (!(req->rq_flags & RQF_DONTPREP)) {
1741 		ret = scsi_prepare_cmd(req);
1742 		if (ret != BLK_STS_OK)
1743 			goto out_dec_host_busy;
1744 		req->rq_flags |= RQF_DONTPREP;
1745 	} else {
1746 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1747 	}
1748 
1749 	cmd->flags &= SCMD_PRESERVED_FLAGS;
1750 	if (sdev->simple_tags)
1751 		cmd->flags |= SCMD_TAGGED;
1752 	if (bd->last)
1753 		cmd->flags |= SCMD_LAST;
1754 
1755 	scsi_set_resid(cmd, 0);
1756 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1757 	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1758 
1759 	blk_mq_start_request(req);
1760 	reason = scsi_dispatch_cmd(cmd);
1761 	if (reason) {
1762 		scsi_set_blocked(cmd, reason);
1763 		ret = BLK_STS_RESOURCE;
1764 		goto out_dec_host_busy;
1765 	}
1766 
1767 	return BLK_STS_OK;
1768 
1769 out_dec_host_busy:
1770 	scsi_dec_host_busy(shost, cmd);
1771 out_dec_target_busy:
1772 	if (scsi_target(sdev)->can_queue > 0)
1773 		atomic_dec(&scsi_target(sdev)->target_busy);
1774 out_put_budget:
1775 	scsi_mq_put_budget(q, cmd->budget_token);
1776 	cmd->budget_token = -1;
1777 	switch (ret) {
1778 	case BLK_STS_OK:
1779 		break;
1780 	case BLK_STS_RESOURCE:
1781 	case BLK_STS_ZONE_RESOURCE:
1782 		if (scsi_device_blocked(sdev))
1783 			ret = BLK_STS_DEV_RESOURCE;
1784 		break;
1785 	case BLK_STS_AGAIN:
1786 		cmd->result = DID_BUS_BUSY << 16;
1787 		if (req->rq_flags & RQF_DONTPREP)
1788 			scsi_mq_uninit_cmd(cmd);
1789 		break;
1790 	default:
1791 		if (unlikely(!scsi_device_online(sdev)))
1792 			cmd->result = DID_NO_CONNECT << 16;
1793 		else
1794 			cmd->result = DID_ERROR << 16;
1795 		/*
1796 		 * Make sure to release all allocated resources when
1797 		 * we hit an error, as we will never see this command
1798 		 * again.
1799 		 */
1800 		if (req->rq_flags & RQF_DONTPREP)
1801 			scsi_mq_uninit_cmd(cmd);
1802 		scsi_run_queue_async(sdev);
1803 		break;
1804 	}
1805 	return ret;
1806 }
1807 
scsi_mq_init_request(struct blk_mq_tag_set * set,struct request * rq,unsigned int hctx_idx,unsigned int numa_node)1808 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1809 				unsigned int hctx_idx, unsigned int numa_node)
1810 {
1811 	struct Scsi_Host *shost = set->driver_data;
1812 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1813 	struct scatterlist *sg;
1814 	int ret = 0;
1815 
1816 	cmd->sense_buffer =
1817 		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1818 	if (!cmd->sense_buffer)
1819 		return -ENOMEM;
1820 
1821 	if (scsi_host_get_prot(shost)) {
1822 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1823 			shost->hostt->cmd_size;
1824 		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1825 	}
1826 
1827 	if (shost->hostt->init_cmd_priv) {
1828 		ret = shost->hostt->init_cmd_priv(shost, cmd);
1829 		if (ret < 0)
1830 			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1831 	}
1832 
1833 	return ret;
1834 }
1835 
scsi_mq_exit_request(struct blk_mq_tag_set * set,struct request * rq,unsigned int hctx_idx)1836 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1837 				 unsigned int hctx_idx)
1838 {
1839 	struct Scsi_Host *shost = set->driver_data;
1840 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1841 
1842 	if (shost->hostt->exit_cmd_priv)
1843 		shost->hostt->exit_cmd_priv(shost, cmd);
1844 	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1845 }
1846 
1847 
scsi_mq_poll(struct blk_mq_hw_ctx * hctx,struct io_comp_batch * iob)1848 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1849 {
1850 	struct Scsi_Host *shost = hctx->driver_data;
1851 
1852 	if (shost->hostt->mq_poll)
1853 		return shost->hostt->mq_poll(shost, hctx->queue_num);
1854 
1855 	return 0;
1856 }
1857 
scsi_init_hctx(struct blk_mq_hw_ctx * hctx,void * data,unsigned int hctx_idx)1858 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1859 			  unsigned int hctx_idx)
1860 {
1861 	struct Scsi_Host *shost = data;
1862 
1863 	hctx->driver_data = shost;
1864 	return 0;
1865 }
1866 
scsi_map_queues(struct blk_mq_tag_set * set)1867 static void scsi_map_queues(struct blk_mq_tag_set *set)
1868 {
1869 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1870 
1871 	if (shost->hostt->map_queues)
1872 		return shost->hostt->map_queues(shost);
1873 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1874 }
1875 
__scsi_init_queue(struct Scsi_Host * shost,struct request_queue * q)1876 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1877 {
1878 	struct device *dev = shost->dma_dev;
1879 
1880 	/*
1881 	 * this limit is imposed by hardware restrictions
1882 	 */
1883 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1884 					SG_MAX_SEGMENTS));
1885 
1886 	if (scsi_host_prot_dma(shost)) {
1887 		shost->sg_prot_tablesize =
1888 			min_not_zero(shost->sg_prot_tablesize,
1889 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1890 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1891 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1892 	}
1893 
1894 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1895 	blk_queue_segment_boundary(q, shost->dma_boundary);
1896 	dma_set_seg_boundary(dev, shost->dma_boundary);
1897 
1898 	blk_queue_max_segment_size(q, shost->max_segment_size);
1899 	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1900 	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1901 
1902 	/*
1903 	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1904 	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1905 	 * which is set by the platform.
1906 	 *
1907 	 * Devices that require a bigger alignment can increase it later.
1908 	 */
1909 	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1910 }
1911 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1912 
1913 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1914 	.get_budget	= scsi_mq_get_budget,
1915 	.put_budget	= scsi_mq_put_budget,
1916 	.queue_rq	= scsi_queue_rq,
1917 	.complete	= scsi_complete,
1918 	.timeout	= scsi_timeout,
1919 #ifdef CONFIG_BLK_DEBUG_FS
1920 	.show_rq	= scsi_show_rq,
1921 #endif
1922 	.init_request	= scsi_mq_init_request,
1923 	.exit_request	= scsi_mq_exit_request,
1924 	.cleanup_rq	= scsi_cleanup_rq,
1925 	.busy		= scsi_mq_lld_busy,
1926 	.map_queues	= scsi_map_queues,
1927 	.init_hctx	= scsi_init_hctx,
1928 	.poll		= scsi_mq_poll,
1929 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1930 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1931 };
1932 
1933 
scsi_commit_rqs(struct blk_mq_hw_ctx * hctx)1934 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1935 {
1936 	struct Scsi_Host *shost = hctx->driver_data;
1937 
1938 	shost->hostt->commit_rqs(shost, hctx->queue_num);
1939 }
1940 
1941 static const struct blk_mq_ops scsi_mq_ops = {
1942 	.get_budget	= scsi_mq_get_budget,
1943 	.put_budget	= scsi_mq_put_budget,
1944 	.queue_rq	= scsi_queue_rq,
1945 	.commit_rqs	= scsi_commit_rqs,
1946 	.complete	= scsi_complete,
1947 	.timeout	= scsi_timeout,
1948 #ifdef CONFIG_BLK_DEBUG_FS
1949 	.show_rq	= scsi_show_rq,
1950 #endif
1951 	.init_request	= scsi_mq_init_request,
1952 	.exit_request	= scsi_mq_exit_request,
1953 	.cleanup_rq	= scsi_cleanup_rq,
1954 	.busy		= scsi_mq_lld_busy,
1955 	.map_queues	= scsi_map_queues,
1956 	.init_hctx	= scsi_init_hctx,
1957 	.poll		= scsi_mq_poll,
1958 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1959 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1960 };
1961 
scsi_mq_setup_tags(struct Scsi_Host * shost)1962 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1963 {
1964 	unsigned int cmd_size, sgl_size;
1965 	struct blk_mq_tag_set *tag_set = &shost->tag_set;
1966 
1967 	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1968 				scsi_mq_inline_sgl_size(shost));
1969 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1970 	if (scsi_host_get_prot(shost))
1971 		cmd_size += sizeof(struct scsi_data_buffer) +
1972 			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1973 
1974 	memset(tag_set, 0, sizeof(*tag_set));
1975 	if (shost->hostt->commit_rqs)
1976 		tag_set->ops = &scsi_mq_ops;
1977 	else
1978 		tag_set->ops = &scsi_mq_ops_no_commit;
1979 	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1980 	tag_set->nr_maps = shost->nr_maps ? : 1;
1981 	tag_set->queue_depth = shost->can_queue;
1982 	tag_set->cmd_size = cmd_size;
1983 	tag_set->numa_node = dev_to_node(shost->dma_dev);
1984 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1985 	tag_set->flags |=
1986 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1987 	tag_set->driver_data = shost;
1988 	if (shost->host_tagset)
1989 		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1990 
1991 	return blk_mq_alloc_tag_set(tag_set);
1992 }
1993 
scsi_mq_free_tags(struct kref * kref)1994 void scsi_mq_free_tags(struct kref *kref)
1995 {
1996 	struct Scsi_Host *shost = container_of(kref, typeof(*shost),
1997 					       tagset_refcnt);
1998 
1999 	blk_mq_free_tag_set(&shost->tag_set);
2000 	complete(&shost->tagset_freed);
2001 }
2002 
2003 /**
2004  * scsi_device_from_queue - return sdev associated with a request_queue
2005  * @q: The request queue to return the sdev from
2006  *
2007  * Return the sdev associated with a request queue or NULL if the
2008  * request_queue does not reference a SCSI device.
2009  */
scsi_device_from_queue(struct request_queue * q)2010 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2011 {
2012 	struct scsi_device *sdev = NULL;
2013 
2014 	if (q->mq_ops == &scsi_mq_ops_no_commit ||
2015 	    q->mq_ops == &scsi_mq_ops)
2016 		sdev = q->queuedata;
2017 	if (!sdev || !get_device(&sdev->sdev_gendev))
2018 		sdev = NULL;
2019 
2020 	return sdev;
2021 }
2022 /*
2023  * pktcdvd should have been integrated into the SCSI layers, but for historical
2024  * reasons like the old IDE driver it isn't.  This export allows it to safely
2025  * probe if a given device is a SCSI one and only attach to that.
2026  */
2027 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2028 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2029 #endif
2030 
2031 /**
2032  * scsi_block_requests - Utility function used by low-level drivers to prevent
2033  * further commands from being queued to the device.
2034  * @shost:  host in question
2035  *
2036  * There is no timer nor any other means by which the requests get unblocked
2037  * other than the low-level driver calling scsi_unblock_requests().
2038  */
scsi_block_requests(struct Scsi_Host * shost)2039 void scsi_block_requests(struct Scsi_Host *shost)
2040 {
2041 	shost->host_self_blocked = 1;
2042 }
2043 EXPORT_SYMBOL(scsi_block_requests);
2044 
2045 /**
2046  * scsi_unblock_requests - Utility function used by low-level drivers to allow
2047  * further commands to be queued to the device.
2048  * @shost:  host in question
2049  *
2050  * There is no timer nor any other means by which the requests get unblocked
2051  * other than the low-level driver calling scsi_unblock_requests(). This is done
2052  * as an API function so that changes to the internals of the scsi mid-layer
2053  * won't require wholesale changes to drivers that use this feature.
2054  */
scsi_unblock_requests(struct Scsi_Host * shost)2055 void scsi_unblock_requests(struct Scsi_Host *shost)
2056 {
2057 	shost->host_self_blocked = 0;
2058 	scsi_run_host_queues(shost);
2059 }
2060 EXPORT_SYMBOL(scsi_unblock_requests);
2061 
scsi_exit_queue(void)2062 void scsi_exit_queue(void)
2063 {
2064 	kmem_cache_destroy(scsi_sense_cache);
2065 }
2066 
2067 /**
2068  *	scsi_mode_select - issue a mode select
2069  *	@sdev:	SCSI device to be queried
2070  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2071  *	@sp:	Save page bit (0 == don't save, 1 == save)
2072  *	@buffer: request buffer (may not be smaller than eight bytes)
2073  *	@len:	length of request buffer.
2074  *	@timeout: command timeout
2075  *	@retries: number of retries before failing
2076  *	@data: returns a structure abstracting the mode header data
2077  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2078  *		must be SCSI_SENSE_BUFFERSIZE big.
2079  *
2080  *	Returns zero if successful; negative error number or scsi
2081  *	status on error
2082  *
2083  */
scsi_mode_select(struct scsi_device * sdev,int pf,int sp,unsigned char * buffer,int len,int timeout,int retries,struct scsi_mode_data * data,struct scsi_sense_hdr * sshdr)2084 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2085 		     unsigned char *buffer, int len, int timeout, int retries,
2086 		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2087 {
2088 	unsigned char cmd[10];
2089 	unsigned char *real_buffer;
2090 	int ret;
2091 
2092 	memset(cmd, 0, sizeof(cmd));
2093 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2094 
2095 	/*
2096 	 * Use MODE SELECT(10) if the device asked for it or if the mode page
2097 	 * and the mode select header cannot fit within the maximumm 255 bytes
2098 	 * of the MODE SELECT(6) command.
2099 	 */
2100 	if (sdev->use_10_for_ms ||
2101 	    len + 4 > 255 ||
2102 	    data->block_descriptor_length > 255) {
2103 		if (len > 65535 - 8)
2104 			return -EINVAL;
2105 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2106 		if (!real_buffer)
2107 			return -ENOMEM;
2108 		memcpy(real_buffer + 8, buffer, len);
2109 		len += 8;
2110 		real_buffer[0] = 0;
2111 		real_buffer[1] = 0;
2112 		real_buffer[2] = data->medium_type;
2113 		real_buffer[3] = data->device_specific;
2114 		real_buffer[4] = data->longlba ? 0x01 : 0;
2115 		real_buffer[5] = 0;
2116 		put_unaligned_be16(data->block_descriptor_length,
2117 				   &real_buffer[6]);
2118 
2119 		cmd[0] = MODE_SELECT_10;
2120 		put_unaligned_be16(len, &cmd[7]);
2121 	} else {
2122 		if (data->longlba)
2123 			return -EINVAL;
2124 
2125 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2126 		if (!real_buffer)
2127 			return -ENOMEM;
2128 		memcpy(real_buffer + 4, buffer, len);
2129 		len += 4;
2130 		real_buffer[0] = 0;
2131 		real_buffer[1] = data->medium_type;
2132 		real_buffer[2] = data->device_specific;
2133 		real_buffer[3] = data->block_descriptor_length;
2134 
2135 		cmd[0] = MODE_SELECT;
2136 		cmd[4] = len;
2137 	}
2138 
2139 	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2140 			       sshdr, timeout, retries, NULL);
2141 	kfree(real_buffer);
2142 	return ret;
2143 }
2144 EXPORT_SYMBOL_GPL(scsi_mode_select);
2145 
2146 /**
2147  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2148  *	@sdev:	SCSI device to be queried
2149  *	@dbd:	set to prevent mode sense from returning block descriptors
2150  *	@modepage: mode page being requested
2151  *	@buffer: request buffer (may not be smaller than eight bytes)
2152  *	@len:	length of request buffer.
2153  *	@timeout: command timeout
2154  *	@retries: number of retries before failing
2155  *	@data: returns a structure abstracting the mode header data
2156  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2157  *		must be SCSI_SENSE_BUFFERSIZE big.
2158  *
2159  *	Returns zero if successful, or a negative error number on failure
2160  */
2161 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)2162 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2163 		  unsigned char *buffer, int len, int timeout, int retries,
2164 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2165 {
2166 	unsigned char cmd[12];
2167 	int use_10_for_ms;
2168 	int header_length;
2169 	int result, retry_count = retries;
2170 	struct scsi_sense_hdr my_sshdr;
2171 
2172 	memset(data, 0, sizeof(*data));
2173 	memset(&cmd[0], 0, 12);
2174 
2175 	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2176 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2177 	cmd[2] = modepage;
2178 
2179 	/* caller might not be interested in sense, but we need it */
2180 	if (!sshdr)
2181 		sshdr = &my_sshdr;
2182 
2183  retry:
2184 	use_10_for_ms = sdev->use_10_for_ms || len > 255;
2185 
2186 	if (use_10_for_ms) {
2187 		if (len < 8 || len > 65535)
2188 			return -EINVAL;
2189 
2190 		cmd[0] = MODE_SENSE_10;
2191 		put_unaligned_be16(len, &cmd[7]);
2192 		header_length = 8;
2193 	} else {
2194 		if (len < 4)
2195 			return -EINVAL;
2196 
2197 		cmd[0] = MODE_SENSE;
2198 		cmd[4] = len;
2199 		header_length = 4;
2200 	}
2201 
2202 	memset(buffer, 0, len);
2203 
2204 	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2205 				  sshdr, timeout, retries, NULL);
2206 	if (result < 0)
2207 		return result;
2208 
2209 	/* This code looks awful: what it's doing is making sure an
2210 	 * ILLEGAL REQUEST sense return identifies the actual command
2211 	 * byte as the problem.  MODE_SENSE commands can return
2212 	 * ILLEGAL REQUEST if the code page isn't supported */
2213 
2214 	if (!scsi_status_is_good(result)) {
2215 		if (scsi_sense_valid(sshdr)) {
2216 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2217 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2218 				/*
2219 				 * Invalid command operation code: retry using
2220 				 * MODE SENSE(6) if this was a MODE SENSE(10)
2221 				 * request, except if the request mode page is
2222 				 * too large for MODE SENSE single byte
2223 				 * allocation length field.
2224 				 */
2225 				if (use_10_for_ms) {
2226 					if (len > 255)
2227 						return -EIO;
2228 					sdev->use_10_for_ms = 0;
2229 					goto retry;
2230 				}
2231 			}
2232 			if (scsi_status_is_check_condition(result) &&
2233 			    sshdr->sense_key == UNIT_ATTENTION &&
2234 			    retry_count) {
2235 				retry_count--;
2236 				goto retry;
2237 			}
2238 		}
2239 		return -EIO;
2240 	}
2241 	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2242 		     (modepage == 6 || modepage == 8))) {
2243 		/* Initio breakage? */
2244 		header_length = 0;
2245 		data->length = 13;
2246 		data->medium_type = 0;
2247 		data->device_specific = 0;
2248 		data->longlba = 0;
2249 		data->block_descriptor_length = 0;
2250 	} else if (use_10_for_ms) {
2251 		data->length = get_unaligned_be16(&buffer[0]) + 2;
2252 		data->medium_type = buffer[2];
2253 		data->device_specific = buffer[3];
2254 		data->longlba = buffer[4] & 0x01;
2255 		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2256 	} else {
2257 		data->length = buffer[0] + 1;
2258 		data->medium_type = buffer[1];
2259 		data->device_specific = buffer[2];
2260 		data->block_descriptor_length = buffer[3];
2261 	}
2262 	data->header_length = header_length;
2263 
2264 	return 0;
2265 }
2266 EXPORT_SYMBOL(scsi_mode_sense);
2267 
2268 /**
2269  *	scsi_test_unit_ready - test if unit is ready
2270  *	@sdev:	scsi device to change the state of.
2271  *	@timeout: command timeout
2272  *	@retries: number of retries before failing
2273  *	@sshdr: outpout pointer for decoded sense information.
2274  *
2275  *	Returns zero if unsuccessful or an error if TUR failed.  For
2276  *	removable media, UNIT_ATTENTION sets ->changed flag.
2277  **/
2278 int
scsi_test_unit_ready(struct scsi_device * sdev,int timeout,int retries,struct scsi_sense_hdr * sshdr)2279 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2280 		     struct scsi_sense_hdr *sshdr)
2281 {
2282 	char cmd[] = {
2283 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2284 	};
2285 	int result;
2286 
2287 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2288 	do {
2289 		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2290 					  timeout, 1, NULL);
2291 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2292 		    sshdr->sense_key == UNIT_ATTENTION)
2293 			sdev->changed = 1;
2294 	} while (scsi_sense_valid(sshdr) &&
2295 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2296 
2297 	return result;
2298 }
2299 EXPORT_SYMBOL(scsi_test_unit_ready);
2300 
2301 /**
2302  *	scsi_device_set_state - Take the given device through the device state model.
2303  *	@sdev:	scsi device to change the state of.
2304  *	@state:	state to change to.
2305  *
2306  *	Returns zero if successful or an error if the requested
2307  *	transition is illegal.
2308  */
2309 int
scsi_device_set_state(struct scsi_device * sdev,enum scsi_device_state state)2310 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2311 {
2312 	enum scsi_device_state oldstate = sdev->sdev_state;
2313 
2314 	if (state == oldstate)
2315 		return 0;
2316 
2317 	switch (state) {
2318 	case SDEV_CREATED:
2319 		switch (oldstate) {
2320 		case SDEV_CREATED_BLOCK:
2321 			break;
2322 		default:
2323 			goto illegal;
2324 		}
2325 		break;
2326 
2327 	case SDEV_RUNNING:
2328 		switch (oldstate) {
2329 		case SDEV_CREATED:
2330 		case SDEV_OFFLINE:
2331 		case SDEV_TRANSPORT_OFFLINE:
2332 		case SDEV_QUIESCE:
2333 		case SDEV_BLOCK:
2334 			break;
2335 		default:
2336 			goto illegal;
2337 		}
2338 		break;
2339 
2340 	case SDEV_QUIESCE:
2341 		switch (oldstate) {
2342 		case SDEV_RUNNING:
2343 		case SDEV_OFFLINE:
2344 		case SDEV_TRANSPORT_OFFLINE:
2345 			break;
2346 		default:
2347 			goto illegal;
2348 		}
2349 		break;
2350 
2351 	case SDEV_OFFLINE:
2352 	case SDEV_TRANSPORT_OFFLINE:
2353 		switch (oldstate) {
2354 		case SDEV_CREATED:
2355 		case SDEV_RUNNING:
2356 		case SDEV_QUIESCE:
2357 		case SDEV_BLOCK:
2358 			break;
2359 		default:
2360 			goto illegal;
2361 		}
2362 		break;
2363 
2364 	case SDEV_BLOCK:
2365 		switch (oldstate) {
2366 		case SDEV_RUNNING:
2367 		case SDEV_CREATED_BLOCK:
2368 		case SDEV_QUIESCE:
2369 		case SDEV_OFFLINE:
2370 			break;
2371 		default:
2372 			goto illegal;
2373 		}
2374 		break;
2375 
2376 	case SDEV_CREATED_BLOCK:
2377 		switch (oldstate) {
2378 		case SDEV_CREATED:
2379 			break;
2380 		default:
2381 			goto illegal;
2382 		}
2383 		break;
2384 
2385 	case SDEV_CANCEL:
2386 		switch (oldstate) {
2387 		case SDEV_CREATED:
2388 		case SDEV_RUNNING:
2389 		case SDEV_QUIESCE:
2390 		case SDEV_OFFLINE:
2391 		case SDEV_TRANSPORT_OFFLINE:
2392 			break;
2393 		default:
2394 			goto illegal;
2395 		}
2396 		break;
2397 
2398 	case SDEV_DEL:
2399 		switch (oldstate) {
2400 		case SDEV_CREATED:
2401 		case SDEV_RUNNING:
2402 		case SDEV_OFFLINE:
2403 		case SDEV_TRANSPORT_OFFLINE:
2404 		case SDEV_CANCEL:
2405 		case SDEV_BLOCK:
2406 		case SDEV_CREATED_BLOCK:
2407 			break;
2408 		default:
2409 			goto illegal;
2410 		}
2411 		break;
2412 
2413 	}
2414 	sdev->offline_already = false;
2415 	sdev->sdev_state = state;
2416 	return 0;
2417 
2418  illegal:
2419 	SCSI_LOG_ERROR_RECOVERY(1,
2420 				sdev_printk(KERN_ERR, sdev,
2421 					    "Illegal state transition %s->%s",
2422 					    scsi_device_state_name(oldstate),
2423 					    scsi_device_state_name(state))
2424 				);
2425 	return -EINVAL;
2426 }
2427 EXPORT_SYMBOL(scsi_device_set_state);
2428 
2429 /**
2430  *	scsi_evt_emit - emit a single SCSI device uevent
2431  *	@sdev: associated SCSI device
2432  *	@evt: event to emit
2433  *
2434  *	Send a single uevent (scsi_event) to the associated scsi_device.
2435  */
scsi_evt_emit(struct scsi_device * sdev,struct scsi_event * evt)2436 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2437 {
2438 	int idx = 0;
2439 	char *envp[3];
2440 
2441 	switch (evt->evt_type) {
2442 	case SDEV_EVT_MEDIA_CHANGE:
2443 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2444 		break;
2445 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2446 		scsi_rescan_device(&sdev->sdev_gendev);
2447 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2448 		break;
2449 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2450 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2451 		break;
2452 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2453 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2454 		break;
2455 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2456 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2457 		break;
2458 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2459 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2460 		break;
2461 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2462 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2463 		break;
2464 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2465 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2466 		break;
2467 	default:
2468 		/* do nothing */
2469 		break;
2470 	}
2471 
2472 	envp[idx++] = NULL;
2473 
2474 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2475 }
2476 
2477 /**
2478  *	scsi_evt_thread - send a uevent for each scsi event
2479  *	@work: work struct for scsi_device
2480  *
2481  *	Dispatch queued events to their associated scsi_device kobjects
2482  *	as uevents.
2483  */
scsi_evt_thread(struct work_struct * work)2484 void scsi_evt_thread(struct work_struct *work)
2485 {
2486 	struct scsi_device *sdev;
2487 	enum scsi_device_event evt_type;
2488 	LIST_HEAD(event_list);
2489 
2490 	sdev = container_of(work, struct scsi_device, event_work);
2491 
2492 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2493 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2494 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2495 
2496 	while (1) {
2497 		struct scsi_event *evt;
2498 		struct list_head *this, *tmp;
2499 		unsigned long flags;
2500 
2501 		spin_lock_irqsave(&sdev->list_lock, flags);
2502 		list_splice_init(&sdev->event_list, &event_list);
2503 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2504 
2505 		if (list_empty(&event_list))
2506 			break;
2507 
2508 		list_for_each_safe(this, tmp, &event_list) {
2509 			evt = list_entry(this, struct scsi_event, node);
2510 			list_del(&evt->node);
2511 			scsi_evt_emit(sdev, evt);
2512 			kfree(evt);
2513 		}
2514 	}
2515 }
2516 
2517 /**
2518  * 	sdev_evt_send - send asserted event to uevent thread
2519  *	@sdev: scsi_device event occurred on
2520  *	@evt: event to send
2521  *
2522  *	Assert scsi device event asynchronously.
2523  */
sdev_evt_send(struct scsi_device * sdev,struct scsi_event * evt)2524 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2525 {
2526 	unsigned long flags;
2527 
2528 #if 0
2529 	/* FIXME: currently this check eliminates all media change events
2530 	 * for polled devices.  Need to update to discriminate between AN
2531 	 * and polled events */
2532 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2533 		kfree(evt);
2534 		return;
2535 	}
2536 #endif
2537 
2538 	spin_lock_irqsave(&sdev->list_lock, flags);
2539 	list_add_tail(&evt->node, &sdev->event_list);
2540 	schedule_work(&sdev->event_work);
2541 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2542 }
2543 EXPORT_SYMBOL_GPL(sdev_evt_send);
2544 
2545 /**
2546  * 	sdev_evt_alloc - allocate a new scsi event
2547  *	@evt_type: type of event to allocate
2548  *	@gfpflags: GFP flags for allocation
2549  *
2550  *	Allocates and returns a new scsi_event.
2551  */
sdev_evt_alloc(enum scsi_device_event evt_type,gfp_t gfpflags)2552 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2553 				  gfp_t gfpflags)
2554 {
2555 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2556 	if (!evt)
2557 		return NULL;
2558 
2559 	evt->evt_type = evt_type;
2560 	INIT_LIST_HEAD(&evt->node);
2561 
2562 	/* evt_type-specific initialization, if any */
2563 	switch (evt_type) {
2564 	case SDEV_EVT_MEDIA_CHANGE:
2565 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2566 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2567 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2568 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2569 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2570 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2571 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2572 	default:
2573 		/* do nothing */
2574 		break;
2575 	}
2576 
2577 	return evt;
2578 }
2579 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2580 
2581 /**
2582  * 	sdev_evt_send_simple - send asserted event to uevent thread
2583  *	@sdev: scsi_device event occurred on
2584  *	@evt_type: type of event to send
2585  *	@gfpflags: GFP flags for allocation
2586  *
2587  *	Assert scsi device event asynchronously, given an event type.
2588  */
sdev_evt_send_simple(struct scsi_device * sdev,enum scsi_device_event evt_type,gfp_t gfpflags)2589 void sdev_evt_send_simple(struct scsi_device *sdev,
2590 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2591 {
2592 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2593 	if (!evt) {
2594 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2595 			    evt_type);
2596 		return;
2597 	}
2598 
2599 	sdev_evt_send(sdev, evt);
2600 }
2601 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2602 
2603 /**
2604  *	scsi_device_quiesce - Block all commands except power management.
2605  *	@sdev:	scsi device to quiesce.
2606  *
2607  *	This works by trying to transition to the SDEV_QUIESCE state
2608  *	(which must be a legal transition).  When the device is in this
2609  *	state, only power management requests will be accepted, all others will
2610  *	be deferred.
2611  *
2612  *	Must be called with user context, may sleep.
2613  *
2614  *	Returns zero if unsuccessful or an error if not.
2615  */
2616 int
scsi_device_quiesce(struct scsi_device * sdev)2617 scsi_device_quiesce(struct scsi_device *sdev)
2618 {
2619 	struct request_queue *q = sdev->request_queue;
2620 	int err;
2621 
2622 	/*
2623 	 * It is allowed to call scsi_device_quiesce() multiple times from
2624 	 * the same context but concurrent scsi_device_quiesce() calls are
2625 	 * not allowed.
2626 	 */
2627 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2628 
2629 	if (sdev->quiesced_by == current)
2630 		return 0;
2631 
2632 	blk_set_pm_only(q);
2633 
2634 	blk_mq_freeze_queue(q);
2635 	/*
2636 	 * Ensure that the effect of blk_set_pm_only() will be visible
2637 	 * for percpu_ref_tryget() callers that occur after the queue
2638 	 * unfreeze even if the queue was already frozen before this function
2639 	 * was called. See also https://lwn.net/Articles/573497/.
2640 	 */
2641 	synchronize_rcu();
2642 	blk_mq_unfreeze_queue(q);
2643 
2644 	mutex_lock(&sdev->state_mutex);
2645 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2646 	if (err == 0)
2647 		sdev->quiesced_by = current;
2648 	else
2649 		blk_clear_pm_only(q);
2650 	mutex_unlock(&sdev->state_mutex);
2651 
2652 	return err;
2653 }
2654 EXPORT_SYMBOL(scsi_device_quiesce);
2655 
2656 /**
2657  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2658  *	@sdev:	scsi device to resume.
2659  *
2660  *	Moves the device from quiesced back to running and restarts the
2661  *	queues.
2662  *
2663  *	Must be called with user context, may sleep.
2664  */
scsi_device_resume(struct scsi_device * sdev)2665 void scsi_device_resume(struct scsi_device *sdev)
2666 {
2667 	/* check if the device state was mutated prior to resume, and if
2668 	 * so assume the state is being managed elsewhere (for example
2669 	 * device deleted during suspend)
2670 	 */
2671 	mutex_lock(&sdev->state_mutex);
2672 	if (sdev->sdev_state == SDEV_QUIESCE)
2673 		scsi_device_set_state(sdev, SDEV_RUNNING);
2674 	if (sdev->quiesced_by) {
2675 		sdev->quiesced_by = NULL;
2676 		blk_clear_pm_only(sdev->request_queue);
2677 	}
2678 	mutex_unlock(&sdev->state_mutex);
2679 }
2680 EXPORT_SYMBOL(scsi_device_resume);
2681 
2682 static void
device_quiesce_fn(struct scsi_device * sdev,void * data)2683 device_quiesce_fn(struct scsi_device *sdev, void *data)
2684 {
2685 	scsi_device_quiesce(sdev);
2686 }
2687 
2688 void
scsi_target_quiesce(struct scsi_target * starget)2689 scsi_target_quiesce(struct scsi_target *starget)
2690 {
2691 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2692 }
2693 EXPORT_SYMBOL(scsi_target_quiesce);
2694 
2695 static void
device_resume_fn(struct scsi_device * sdev,void * data)2696 device_resume_fn(struct scsi_device *sdev, void *data)
2697 {
2698 	scsi_device_resume(sdev);
2699 }
2700 
2701 void
scsi_target_resume(struct scsi_target * starget)2702 scsi_target_resume(struct scsi_target *starget)
2703 {
2704 	starget_for_each_device(starget, NULL, device_resume_fn);
2705 }
2706 EXPORT_SYMBOL(scsi_target_resume);
2707 
__scsi_internal_device_block_nowait(struct scsi_device * sdev)2708 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2709 {
2710 	if (scsi_device_set_state(sdev, SDEV_BLOCK))
2711 		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2712 
2713 	return 0;
2714 }
2715 
scsi_start_queue(struct scsi_device * sdev)2716 void scsi_start_queue(struct scsi_device *sdev)
2717 {
2718 	if (cmpxchg(&sdev->queue_stopped, 1, 0))
2719 		blk_mq_unquiesce_queue(sdev->request_queue);
2720 }
2721 
scsi_stop_queue(struct scsi_device * sdev,bool nowait)2722 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait)
2723 {
2724 	/*
2725 	 * The atomic variable of ->queue_stopped covers that
2726 	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2727 	 *
2728 	 * However, we still need to wait until quiesce is done
2729 	 * in case that queue has been stopped.
2730 	 */
2731 	if (!cmpxchg(&sdev->queue_stopped, 0, 1)) {
2732 		if (nowait)
2733 			blk_mq_quiesce_queue_nowait(sdev->request_queue);
2734 		else
2735 			blk_mq_quiesce_queue(sdev->request_queue);
2736 	} else {
2737 		if (!nowait)
2738 			blk_mq_wait_quiesce_done(sdev->request_queue);
2739 	}
2740 }
2741 
2742 /**
2743  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2744  * @sdev: device to block
2745  *
2746  * Pause SCSI command processing on the specified device. Does not sleep.
2747  *
2748  * Returns zero if successful or a negative error code upon failure.
2749  *
2750  * Notes:
2751  * This routine transitions the device to the SDEV_BLOCK state (which must be
2752  * a legal transition). When the device is in this state, command processing
2753  * is paused until the device leaves the SDEV_BLOCK state. See also
2754  * scsi_internal_device_unblock_nowait().
2755  */
scsi_internal_device_block_nowait(struct scsi_device * sdev)2756 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2757 {
2758 	int ret = __scsi_internal_device_block_nowait(sdev);
2759 
2760 	/*
2761 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2762 	 * block layer from calling the midlayer with this device's
2763 	 * request queue.
2764 	 */
2765 	if (!ret)
2766 		scsi_stop_queue(sdev, true);
2767 	return ret;
2768 }
2769 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2770 
2771 /**
2772  * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2773  * @sdev: device to block
2774  *
2775  * Pause SCSI command processing on the specified device and wait until all
2776  * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2777  *
2778  * Returns zero if successful or a negative error code upon failure.
2779  *
2780  * Note:
2781  * This routine transitions the device to the SDEV_BLOCK state (which must be
2782  * a legal transition). When the device is in this state, command processing
2783  * is paused until the device leaves the SDEV_BLOCK state. See also
2784  * scsi_internal_device_unblock().
2785  */
scsi_internal_device_block(struct scsi_device * sdev)2786 static int scsi_internal_device_block(struct scsi_device *sdev)
2787 {
2788 	int err;
2789 
2790 	mutex_lock(&sdev->state_mutex);
2791 	err = __scsi_internal_device_block_nowait(sdev);
2792 	if (err == 0)
2793 		scsi_stop_queue(sdev, false);
2794 	mutex_unlock(&sdev->state_mutex);
2795 
2796 	return err;
2797 }
2798 
2799 /**
2800  * scsi_internal_device_unblock_nowait - resume a device after a block request
2801  * @sdev:	device to resume
2802  * @new_state:	state to set the device to after unblocking
2803  *
2804  * Restart the device queue for a previously suspended SCSI device. Does not
2805  * sleep.
2806  *
2807  * Returns zero if successful or a negative error code upon failure.
2808  *
2809  * Notes:
2810  * This routine transitions the device to the SDEV_RUNNING state or to one of
2811  * the offline states (which must be a legal transition) allowing the midlayer
2812  * to goose the queue for this device.
2813  */
scsi_internal_device_unblock_nowait(struct scsi_device * sdev,enum scsi_device_state new_state)2814 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2815 					enum scsi_device_state new_state)
2816 {
2817 	switch (new_state) {
2818 	case SDEV_RUNNING:
2819 	case SDEV_TRANSPORT_OFFLINE:
2820 		break;
2821 	default:
2822 		return -EINVAL;
2823 	}
2824 
2825 	/*
2826 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2827 	 * offlined states and goose the device queue if successful.
2828 	 */
2829 	switch (sdev->sdev_state) {
2830 	case SDEV_BLOCK:
2831 	case SDEV_TRANSPORT_OFFLINE:
2832 		sdev->sdev_state = new_state;
2833 		break;
2834 	case SDEV_CREATED_BLOCK:
2835 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2836 		    new_state == SDEV_OFFLINE)
2837 			sdev->sdev_state = new_state;
2838 		else
2839 			sdev->sdev_state = SDEV_CREATED;
2840 		break;
2841 	case SDEV_CANCEL:
2842 	case SDEV_OFFLINE:
2843 		break;
2844 	default:
2845 		return -EINVAL;
2846 	}
2847 	scsi_start_queue(sdev);
2848 
2849 	return 0;
2850 }
2851 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2852 
2853 /**
2854  * scsi_internal_device_unblock - resume a device after a block request
2855  * @sdev:	device to resume
2856  * @new_state:	state to set the device to after unblocking
2857  *
2858  * Restart the device queue for a previously suspended SCSI device. May sleep.
2859  *
2860  * Returns zero if successful or a negative error code upon failure.
2861  *
2862  * Notes:
2863  * This routine transitions the device to the SDEV_RUNNING state or to one of
2864  * the offline states (which must be a legal transition) allowing the midlayer
2865  * to goose the queue for this device.
2866  */
scsi_internal_device_unblock(struct scsi_device * sdev,enum scsi_device_state new_state)2867 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2868 					enum scsi_device_state new_state)
2869 {
2870 	int ret;
2871 
2872 	mutex_lock(&sdev->state_mutex);
2873 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2874 	mutex_unlock(&sdev->state_mutex);
2875 
2876 	return ret;
2877 }
2878 
2879 static void
device_block(struct scsi_device * sdev,void * data)2880 device_block(struct scsi_device *sdev, void *data)
2881 {
2882 	int ret;
2883 
2884 	ret = scsi_internal_device_block(sdev);
2885 
2886 	WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2887 		  dev_name(&sdev->sdev_gendev), ret);
2888 }
2889 
2890 static int
target_block(struct device * dev,void * data)2891 target_block(struct device *dev, void *data)
2892 {
2893 	if (scsi_is_target_device(dev))
2894 		starget_for_each_device(to_scsi_target(dev), NULL,
2895 					device_block);
2896 	return 0;
2897 }
2898 
2899 void
scsi_target_block(struct device * dev)2900 scsi_target_block(struct device *dev)
2901 {
2902 	if (scsi_is_target_device(dev))
2903 		starget_for_each_device(to_scsi_target(dev), NULL,
2904 					device_block);
2905 	else
2906 		device_for_each_child(dev, NULL, target_block);
2907 }
2908 EXPORT_SYMBOL_GPL(scsi_target_block);
2909 
2910 static void
device_unblock(struct scsi_device * sdev,void * data)2911 device_unblock(struct scsi_device *sdev, void *data)
2912 {
2913 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2914 }
2915 
2916 static int
target_unblock(struct device * dev,void * data)2917 target_unblock(struct device *dev, void *data)
2918 {
2919 	if (scsi_is_target_device(dev))
2920 		starget_for_each_device(to_scsi_target(dev), data,
2921 					device_unblock);
2922 	return 0;
2923 }
2924 
2925 void
scsi_target_unblock(struct device * dev,enum scsi_device_state new_state)2926 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2927 {
2928 	if (scsi_is_target_device(dev))
2929 		starget_for_each_device(to_scsi_target(dev), &new_state,
2930 					device_unblock);
2931 	else
2932 		device_for_each_child(dev, &new_state, target_unblock);
2933 }
2934 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2935 
2936 int
scsi_host_block(struct Scsi_Host * shost)2937 scsi_host_block(struct Scsi_Host *shost)
2938 {
2939 	struct scsi_device *sdev;
2940 	int ret = 0;
2941 
2942 	/*
2943 	 * Call scsi_internal_device_block_nowait so we can avoid
2944 	 * calling synchronize_rcu() for each LUN.
2945 	 */
2946 	shost_for_each_device(sdev, shost) {
2947 		mutex_lock(&sdev->state_mutex);
2948 		ret = scsi_internal_device_block_nowait(sdev);
2949 		mutex_unlock(&sdev->state_mutex);
2950 		if (ret) {
2951 			scsi_device_put(sdev);
2952 			break;
2953 		}
2954 	}
2955 
2956 	/*
2957 	 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2958 	 * calling synchronize_rcu() once is enough.
2959 	 */
2960 	WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2961 
2962 	if (!ret)
2963 		synchronize_rcu();
2964 
2965 	return ret;
2966 }
2967 EXPORT_SYMBOL_GPL(scsi_host_block);
2968 
2969 int
scsi_host_unblock(struct Scsi_Host * shost,int new_state)2970 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2971 {
2972 	struct scsi_device *sdev;
2973 	int ret = 0;
2974 
2975 	shost_for_each_device(sdev, shost) {
2976 		ret = scsi_internal_device_unblock(sdev, new_state);
2977 		if (ret) {
2978 			scsi_device_put(sdev);
2979 			break;
2980 		}
2981 	}
2982 	return ret;
2983 }
2984 EXPORT_SYMBOL_GPL(scsi_host_unblock);
2985 
2986 /**
2987  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2988  * @sgl:	scatter-gather list
2989  * @sg_count:	number of segments in sg
2990  * @offset:	offset in bytes into sg, on return offset into the mapped area
2991  * @len:	bytes to map, on return number of bytes mapped
2992  *
2993  * Returns virtual address of the start of the mapped page
2994  */
scsi_kmap_atomic_sg(struct scatterlist * sgl,int sg_count,size_t * offset,size_t * len)2995 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2996 			  size_t *offset, size_t *len)
2997 {
2998 	int i;
2999 	size_t sg_len = 0, len_complete = 0;
3000 	struct scatterlist *sg;
3001 	struct page *page;
3002 
3003 	WARN_ON(!irqs_disabled());
3004 
3005 	for_each_sg(sgl, sg, sg_count, i) {
3006 		len_complete = sg_len; /* Complete sg-entries */
3007 		sg_len += sg->length;
3008 		if (sg_len > *offset)
3009 			break;
3010 	}
3011 
3012 	if (unlikely(i == sg_count)) {
3013 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3014 			"elements %d\n",
3015 		       __func__, sg_len, *offset, sg_count);
3016 		WARN_ON(1);
3017 		return NULL;
3018 	}
3019 
3020 	/* Offset starting from the beginning of first page in this sg-entry */
3021 	*offset = *offset - len_complete + sg->offset;
3022 
3023 	/* Assumption: contiguous pages can be accessed as "page + i" */
3024 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3025 	*offset &= ~PAGE_MASK;
3026 
3027 	/* Bytes in this sg-entry from *offset to the end of the page */
3028 	sg_len = PAGE_SIZE - *offset;
3029 	if (*len > sg_len)
3030 		*len = sg_len;
3031 
3032 	return kmap_atomic(page);
3033 }
3034 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3035 
3036 /**
3037  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3038  * @virt:	virtual address to be unmapped
3039  */
scsi_kunmap_atomic_sg(void * virt)3040 void scsi_kunmap_atomic_sg(void *virt)
3041 {
3042 	kunmap_atomic(virt);
3043 }
3044 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3045 
sdev_disable_disk_events(struct scsi_device * sdev)3046 void sdev_disable_disk_events(struct scsi_device *sdev)
3047 {
3048 	atomic_inc(&sdev->disk_events_disable_depth);
3049 }
3050 EXPORT_SYMBOL(sdev_disable_disk_events);
3051 
sdev_enable_disk_events(struct scsi_device * sdev)3052 void sdev_enable_disk_events(struct scsi_device *sdev)
3053 {
3054 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3055 		return;
3056 	atomic_dec(&sdev->disk_events_disable_depth);
3057 }
3058 EXPORT_SYMBOL(sdev_enable_disk_events);
3059 
designator_prio(const unsigned char * d)3060 static unsigned char designator_prio(const unsigned char *d)
3061 {
3062 	if (d[1] & 0x30)
3063 		/* not associated with LUN */
3064 		return 0;
3065 
3066 	if (d[3] == 0)
3067 		/* invalid length */
3068 		return 0;
3069 
3070 	/*
3071 	 * Order of preference for lun descriptor:
3072 	 * - SCSI name string
3073 	 * - NAA IEEE Registered Extended
3074 	 * - EUI-64 based 16-byte
3075 	 * - EUI-64 based 12-byte
3076 	 * - NAA IEEE Registered
3077 	 * - NAA IEEE Extended
3078 	 * - EUI-64 based 8-byte
3079 	 * - SCSI name string (truncated)
3080 	 * - T10 Vendor ID
3081 	 * as longer descriptors reduce the likelyhood
3082 	 * of identification clashes.
3083 	 */
3084 
3085 	switch (d[1] & 0xf) {
3086 	case 8:
3087 		/* SCSI name string, variable-length UTF-8 */
3088 		return 9;
3089 	case 3:
3090 		switch (d[4] >> 4) {
3091 		case 6:
3092 			/* NAA registered extended */
3093 			return 8;
3094 		case 5:
3095 			/* NAA registered */
3096 			return 5;
3097 		case 4:
3098 			/* NAA extended */
3099 			return 4;
3100 		case 3:
3101 			/* NAA locally assigned */
3102 			return 1;
3103 		default:
3104 			break;
3105 		}
3106 		break;
3107 	case 2:
3108 		switch (d[3]) {
3109 		case 16:
3110 			/* EUI64-based, 16 byte */
3111 			return 7;
3112 		case 12:
3113 			/* EUI64-based, 12 byte */
3114 			return 6;
3115 		case 8:
3116 			/* EUI64-based, 8 byte */
3117 			return 3;
3118 		default:
3119 			break;
3120 		}
3121 		break;
3122 	case 1:
3123 		/* T10 vendor ID */
3124 		return 1;
3125 	default:
3126 		break;
3127 	}
3128 
3129 	return 0;
3130 }
3131 
3132 /**
3133  * scsi_vpd_lun_id - return a unique device identification
3134  * @sdev: SCSI device
3135  * @id:   buffer for the identification
3136  * @id_len:  length of the buffer
3137  *
3138  * Copies a unique device identification into @id based
3139  * on the information in the VPD page 0x83 of the device.
3140  * The string will be formatted as a SCSI name string.
3141  *
3142  * Returns the length of the identification or error on failure.
3143  * If the identifier is longer than the supplied buffer the actual
3144  * identifier length is returned and the buffer is not zero-padded.
3145  */
scsi_vpd_lun_id(struct scsi_device * sdev,char * id,size_t id_len)3146 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3147 {
3148 	u8 cur_id_prio = 0;
3149 	u8 cur_id_size = 0;
3150 	const unsigned char *d, *cur_id_str;
3151 	const struct scsi_vpd *vpd_pg83;
3152 	int id_size = -EINVAL;
3153 
3154 	rcu_read_lock();
3155 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3156 	if (!vpd_pg83) {
3157 		rcu_read_unlock();
3158 		return -ENXIO;
3159 	}
3160 
3161 	/* The id string must be at least 20 bytes + terminating NULL byte */
3162 	if (id_len < 21) {
3163 		rcu_read_unlock();
3164 		return -EINVAL;
3165 	}
3166 
3167 	memset(id, 0, id_len);
3168 	for (d = vpd_pg83->data + 4;
3169 	     d < vpd_pg83->data + vpd_pg83->len;
3170 	     d += d[3] + 4) {
3171 		u8 prio = designator_prio(d);
3172 
3173 		if (prio == 0 || cur_id_prio > prio)
3174 			continue;
3175 
3176 		switch (d[1] & 0xf) {
3177 		case 0x1:
3178 			/* T10 Vendor ID */
3179 			if (cur_id_size > d[3])
3180 				break;
3181 			cur_id_prio = prio;
3182 			cur_id_size = d[3];
3183 			if (cur_id_size + 4 > id_len)
3184 				cur_id_size = id_len - 4;
3185 			cur_id_str = d + 4;
3186 			id_size = snprintf(id, id_len, "t10.%*pE",
3187 					   cur_id_size, cur_id_str);
3188 			break;
3189 		case 0x2:
3190 			/* EUI-64 */
3191 			cur_id_prio = prio;
3192 			cur_id_size = d[3];
3193 			cur_id_str = d + 4;
3194 			switch (cur_id_size) {
3195 			case 8:
3196 				id_size = snprintf(id, id_len,
3197 						   "eui.%8phN",
3198 						   cur_id_str);
3199 				break;
3200 			case 12:
3201 				id_size = snprintf(id, id_len,
3202 						   "eui.%12phN",
3203 						   cur_id_str);
3204 				break;
3205 			case 16:
3206 				id_size = snprintf(id, id_len,
3207 						   "eui.%16phN",
3208 						   cur_id_str);
3209 				break;
3210 			default:
3211 				break;
3212 			}
3213 			break;
3214 		case 0x3:
3215 			/* NAA */
3216 			cur_id_prio = prio;
3217 			cur_id_size = d[3];
3218 			cur_id_str = d + 4;
3219 			switch (cur_id_size) {
3220 			case 8:
3221 				id_size = snprintf(id, id_len,
3222 						   "naa.%8phN",
3223 						   cur_id_str);
3224 				break;
3225 			case 16:
3226 				id_size = snprintf(id, id_len,
3227 						   "naa.%16phN",
3228 						   cur_id_str);
3229 				break;
3230 			default:
3231 				break;
3232 			}
3233 			break;
3234 		case 0x8:
3235 			/* SCSI name string */
3236 			if (cur_id_size > d[3])
3237 				break;
3238 			/* Prefer others for truncated descriptor */
3239 			if (d[3] > id_len) {
3240 				prio = 2;
3241 				if (cur_id_prio > prio)
3242 					break;
3243 			}
3244 			cur_id_prio = prio;
3245 			cur_id_size = id_size = d[3];
3246 			cur_id_str = d + 4;
3247 			if (cur_id_size >= id_len)
3248 				cur_id_size = id_len - 1;
3249 			memcpy(id, cur_id_str, cur_id_size);
3250 			break;
3251 		default:
3252 			break;
3253 		}
3254 	}
3255 	rcu_read_unlock();
3256 
3257 	return id_size;
3258 }
3259 EXPORT_SYMBOL(scsi_vpd_lun_id);
3260 
3261 /*
3262  * scsi_vpd_tpg_id - return a target port group identifier
3263  * @sdev: SCSI device
3264  *
3265  * Returns the Target Port Group identifier from the information
3266  * froom VPD page 0x83 of the device.
3267  *
3268  * Returns the identifier or error on failure.
3269  */
scsi_vpd_tpg_id(struct scsi_device * sdev,int * rel_id)3270 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3271 {
3272 	const unsigned char *d;
3273 	const struct scsi_vpd *vpd_pg83;
3274 	int group_id = -EAGAIN, rel_port = -1;
3275 
3276 	rcu_read_lock();
3277 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3278 	if (!vpd_pg83) {
3279 		rcu_read_unlock();
3280 		return -ENXIO;
3281 	}
3282 
3283 	d = vpd_pg83->data + 4;
3284 	while (d < vpd_pg83->data + vpd_pg83->len) {
3285 		switch (d[1] & 0xf) {
3286 		case 0x4:
3287 			/* Relative target port */
3288 			rel_port = get_unaligned_be16(&d[6]);
3289 			break;
3290 		case 0x5:
3291 			/* Target port group */
3292 			group_id = get_unaligned_be16(&d[6]);
3293 			break;
3294 		default:
3295 			break;
3296 		}
3297 		d += d[3] + 4;
3298 	}
3299 	rcu_read_unlock();
3300 
3301 	if (group_id >= 0 && rel_id && rel_port != -1)
3302 		*rel_id = rel_port;
3303 
3304 	return group_id;
3305 }
3306 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3307 
3308 /**
3309  * scsi_build_sense - build sense data for a command
3310  * @scmd:	scsi command for which the sense should be formatted
3311  * @desc:	Sense format (non-zero == descriptor format,
3312  *              0 == fixed format)
3313  * @key:	Sense key
3314  * @asc:	Additional sense code
3315  * @ascq:	Additional sense code qualifier
3316  *
3317  **/
scsi_build_sense(struct scsi_cmnd * scmd,int desc,u8 key,u8 asc,u8 ascq)3318 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3319 {
3320 	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3321 	scmd->result = SAM_STAT_CHECK_CONDITION;
3322 }
3323 EXPORT_SYMBOL_GPL(scsi_build_sense);
3324