xref: /openbmc/linux/drivers/scsi/scsi_lib.c (revision 0f9b4c3ca5fdf3e177266ef994071b1a03f07318)
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 	blk_mq_requeue_request(rq, false);
126 	if (!scsi_host_in_recovery(cmd->device->host))
127 		blk_mq_delay_kick_requeue_list(rq->q, msecs);
128 }
129 
130 /**
131  * __scsi_queue_insert - private queue insertion
132  * @cmd: The SCSI command being requeued
133  * @reason:  The reason for the requeue
134  * @unbusy: Whether the queue should be unbusied
135  *
136  * This is a private queue insertion.  The public interface
137  * scsi_queue_insert() always assumes the queue should be unbusied
138  * because it's always called before the completion.  This function is
139  * for a requeue after completion, which should only occur in this
140  * file.
141  */
__scsi_queue_insert(struct scsi_cmnd * cmd,int reason,bool unbusy)142 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
143 {
144 	struct scsi_device *device = cmd->device;
145 
146 	SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
147 		"Inserting command %p into mlqueue\n", cmd));
148 
149 	scsi_set_blocked(cmd, reason);
150 
151 	/*
152 	 * Decrement the counters, since these commands are no longer
153 	 * active on the host/device.
154 	 */
155 	if (unbusy)
156 		scsi_device_unbusy(device, cmd);
157 
158 	/*
159 	 * Requeue this command.  It will go before all other commands
160 	 * that are already in the queue. Schedule requeue work under
161 	 * lock such that the kblockd_schedule_work() call happens
162 	 * before blk_mq_destroy_queue() finishes.
163 	 */
164 	cmd->result = 0;
165 
166 	blk_mq_requeue_request(scsi_cmd_to_rq(cmd),
167 			       !scsi_host_in_recovery(cmd->device->host));
168 }
169 
170 /**
171  * scsi_queue_insert - Reinsert a command in the queue.
172  * @cmd:    command that we are adding to queue.
173  * @reason: why we are inserting command to queue.
174  *
175  * We do this for one of two cases. Either the host is busy and it cannot accept
176  * any more commands for the time being, or the device returned QUEUE_FULL and
177  * can accept no more commands.
178  *
179  * Context: This could be called either from an interrupt context or a normal
180  * process context.
181  */
scsi_queue_insert(struct scsi_cmnd * cmd,int reason)182 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
183 {
184 	__scsi_queue_insert(cmd, reason, true);
185 }
186 
187 /**
188  * scsi_execute_cmd - insert request and wait for the result
189  * @sdev:	scsi_device
190  * @cmd:	scsi command
191  * @opf:	block layer request cmd_flags
192  * @buffer:	data buffer
193  * @bufflen:	len of buffer
194  * @timeout:	request timeout in HZ
195  * @retries:	number of times to retry request
196  * @args:	Optional args. See struct definition for field descriptions
197  *
198  * Returns the scsi_cmnd result field if a command was executed, or a negative
199  * Linux error code if we didn't get that far.
200  */
scsi_execute_cmd(struct scsi_device * sdev,const unsigned char * cmd,blk_opf_t opf,void * buffer,unsigned int bufflen,int timeout,int retries,const struct scsi_exec_args * args)201 int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,
202 		     blk_opf_t opf, void *buffer, unsigned int bufflen,
203 		     int timeout, int retries,
204 		     const struct scsi_exec_args *args)
205 {
206 	static const struct scsi_exec_args default_args;
207 	struct request *req;
208 	struct scsi_cmnd *scmd;
209 	int ret;
210 
211 	if (!args)
212 		args = &default_args;
213 	else if (WARN_ON_ONCE(args->sense &&
214 			      args->sense_len != SCSI_SENSE_BUFFERSIZE))
215 		return -EINVAL;
216 
217 	req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);
218 	if (IS_ERR(req))
219 		return PTR_ERR(req);
220 
221 	if (bufflen) {
222 		ret = blk_rq_map_kern(sdev->request_queue, req,
223 				      buffer, bufflen, GFP_NOIO);
224 		if (ret)
225 			goto out;
226 	}
227 	scmd = blk_mq_rq_to_pdu(req);
228 	scmd->cmd_len = COMMAND_SIZE(cmd[0]);
229 	memcpy(scmd->cmnd, cmd, scmd->cmd_len);
230 	scmd->allowed = retries;
231 	scmd->flags |= args->scmd_flags;
232 	req->timeout = timeout;
233 	req->rq_flags |= RQF_QUIET;
234 
235 	/*
236 	 * head injection *required* here otherwise quiesce won't work
237 	 */
238 	blk_execute_rq(req, true);
239 
240 	/*
241 	 * Some devices (USB mass-storage in particular) may transfer
242 	 * garbage data together with a residue indicating that the data
243 	 * is invalid.  Prevent the garbage from being misinterpreted
244 	 * and prevent security leaks by zeroing out the excess data.
245 	 */
246 	if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))
247 		memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);
248 
249 	if (args->resid)
250 		*args->resid = scmd->resid_len;
251 	if (args->sense)
252 		memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);
253 	if (args->sshdr)
254 		scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,
255 				     args->sshdr);
256 
257 	ret = scmd->result;
258  out:
259 	blk_mq_free_request(req);
260 
261 	return ret;
262 }
263 EXPORT_SYMBOL(scsi_execute_cmd);
264 
265 /*
266  * Wake up the error handler if necessary. Avoid as follows that the error
267  * handler is not woken up if host in-flight requests number ==
268  * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
269  * with an RCU read lock in this function to ensure that this function in
270  * its entirety either finishes before scsi_eh_scmd_add() increases the
271  * host_failed counter or that it notices the shost state change made by
272  * scsi_eh_scmd_add().
273  */
scsi_dec_host_busy(struct Scsi_Host * shost,struct scsi_cmnd * cmd)274 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
275 {
276 	unsigned long flags;
277 
278 	rcu_read_lock();
279 	__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
280 	if (unlikely(scsi_host_in_recovery(shost))) {
281 		unsigned int busy = scsi_host_busy(shost);
282 
283 		spin_lock_irqsave(shost->host_lock, flags);
284 		if (shost->host_failed || shost->host_eh_scheduled)
285 			scsi_eh_wakeup(shost, busy);
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 
305 /*
306  * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with
307  * interrupts disabled.
308  */
scsi_kick_sdev_queue(struct scsi_device * sdev,void * data)309 static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)
310 {
311 	struct scsi_device *current_sdev = data;
312 
313 	if (sdev != current_sdev)
314 		blk_mq_run_hw_queues(sdev->request_queue, true);
315 }
316 
317 /*
318  * Called for single_lun devices on IO completion. Clear starget_sdev_user,
319  * and call blk_run_queue for all the scsi_devices on the target -
320  * including current_sdev first.
321  *
322  * Called with *no* scsi locks held.
323  */
scsi_single_lun_run(struct scsi_device * current_sdev)324 static void scsi_single_lun_run(struct scsi_device *current_sdev)
325 {
326 	struct Scsi_Host *shost = current_sdev->host;
327 	struct scsi_target *starget = scsi_target(current_sdev);
328 	unsigned long flags;
329 
330 	spin_lock_irqsave(shost->host_lock, flags);
331 	starget->starget_sdev_user = NULL;
332 	spin_unlock_irqrestore(shost->host_lock, flags);
333 
334 	/*
335 	 * Call blk_run_queue for all LUNs on the target, starting with
336 	 * current_sdev. We race with others (to set starget_sdev_user),
337 	 * but in most cases, we will be first. Ideally, each LU on the
338 	 * target would get some limited time or requests on the target.
339 	 */
340 	blk_mq_run_hw_queues(current_sdev->request_queue,
341 			     shost->queuecommand_may_block);
342 
343 	spin_lock_irqsave(shost->host_lock, flags);
344 	if (!starget->starget_sdev_user)
345 		__starget_for_each_device(starget, current_sdev,
346 					  scsi_kick_sdev_queue);
347 	spin_unlock_irqrestore(shost->host_lock, flags);
348 }
349 
scsi_device_is_busy(struct scsi_device * sdev)350 static inline bool scsi_device_is_busy(struct scsi_device *sdev)
351 {
352 	if (scsi_device_busy(sdev) >= sdev->queue_depth)
353 		return true;
354 	if (atomic_read(&sdev->device_blocked) > 0)
355 		return true;
356 	return false;
357 }
358 
scsi_target_is_busy(struct scsi_target * starget)359 static inline bool scsi_target_is_busy(struct scsi_target *starget)
360 {
361 	if (starget->can_queue > 0) {
362 		if (atomic_read(&starget->target_busy) >= starget->can_queue)
363 			return true;
364 		if (atomic_read(&starget->target_blocked) > 0)
365 			return true;
366 	}
367 	return false;
368 }
369 
scsi_host_is_busy(struct Scsi_Host * shost)370 static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
371 {
372 	if (atomic_read(&shost->host_blocked) > 0)
373 		return true;
374 	if (shost->host_self_blocked)
375 		return true;
376 	return false;
377 }
378 
scsi_starved_list_run(struct Scsi_Host * shost)379 static void scsi_starved_list_run(struct Scsi_Host *shost)
380 {
381 	LIST_HEAD(starved_list);
382 	struct scsi_device *sdev;
383 	unsigned long flags;
384 
385 	spin_lock_irqsave(shost->host_lock, flags);
386 	list_splice_init(&shost->starved_list, &starved_list);
387 
388 	while (!list_empty(&starved_list)) {
389 		struct request_queue *slq;
390 
391 		/*
392 		 * As long as shost is accepting commands and we have
393 		 * starved queues, call blk_run_queue. scsi_request_fn
394 		 * drops the queue_lock and can add us back to the
395 		 * starved_list.
396 		 *
397 		 * host_lock protects the starved_list and starved_entry.
398 		 * scsi_request_fn must get the host_lock before checking
399 		 * or modifying starved_list or starved_entry.
400 		 */
401 		if (scsi_host_is_busy(shost))
402 			break;
403 
404 		sdev = list_entry(starved_list.next,
405 				  struct scsi_device, starved_entry);
406 		list_del_init(&sdev->starved_entry);
407 		if (scsi_target_is_busy(scsi_target(sdev))) {
408 			list_move_tail(&sdev->starved_entry,
409 				       &shost->starved_list);
410 			continue;
411 		}
412 
413 		/*
414 		 * Once we drop the host lock, a racing scsi_remove_device()
415 		 * call may remove the sdev from the starved list and destroy
416 		 * it and the queue.  Mitigate by taking a reference to the
417 		 * queue and never touching the sdev again after we drop the
418 		 * host lock.  Note: if __scsi_remove_device() invokes
419 		 * blk_mq_destroy_queue() before the queue is run from this
420 		 * function then blk_run_queue() will return immediately since
421 		 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.
422 		 */
423 		slq = sdev->request_queue;
424 		if (!blk_get_queue(slq))
425 			continue;
426 		spin_unlock_irqrestore(shost->host_lock, flags);
427 
428 		blk_mq_run_hw_queues(slq, false);
429 		blk_put_queue(slq);
430 
431 		spin_lock_irqsave(shost->host_lock, flags);
432 	}
433 	/* put any unprocessed entries back */
434 	list_splice(&starved_list, &shost->starved_list);
435 	spin_unlock_irqrestore(shost->host_lock, flags);
436 }
437 
438 /**
439  * scsi_run_queue - Select a proper request queue to serve next.
440  * @q:  last request's queue
441  *
442  * The previous command was completely finished, start a new one if possible.
443  */
scsi_run_queue(struct request_queue * q)444 static void scsi_run_queue(struct request_queue *q)
445 {
446 	struct scsi_device *sdev = q->queuedata;
447 
448 	if (scsi_target(sdev)->single_lun)
449 		scsi_single_lun_run(sdev);
450 	if (!list_empty(&sdev->host->starved_list))
451 		scsi_starved_list_run(sdev->host);
452 
453 	/* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */
454 	blk_mq_kick_requeue_list(q);
455 }
456 
scsi_requeue_run_queue(struct work_struct * work)457 void scsi_requeue_run_queue(struct work_struct *work)
458 {
459 	struct scsi_device *sdev;
460 	struct request_queue *q;
461 
462 	sdev = container_of(work, struct scsi_device, requeue_work);
463 	q = sdev->request_queue;
464 	scsi_run_queue(q);
465 }
466 
scsi_run_host_queues(struct Scsi_Host * shost)467 void scsi_run_host_queues(struct Scsi_Host *shost)
468 {
469 	struct scsi_device *sdev;
470 
471 	shost_for_each_device(sdev, shost)
472 		scsi_run_queue(sdev->request_queue);
473 }
474 
scsi_uninit_cmd(struct scsi_cmnd * cmd)475 static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
476 {
477 	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {
478 		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
479 
480 		if (drv->uninit_command)
481 			drv->uninit_command(cmd);
482 	}
483 }
484 
scsi_free_sgtables(struct scsi_cmnd * cmd)485 void scsi_free_sgtables(struct scsi_cmnd *cmd)
486 {
487 	if (cmd->sdb.table.nents)
488 		sg_free_table_chained(&cmd->sdb.table,
489 				SCSI_INLINE_SG_CNT);
490 	if (scsi_prot_sg_count(cmd))
491 		sg_free_table_chained(&cmd->prot_sdb->table,
492 				SCSI_INLINE_PROT_SG_CNT);
493 }
494 EXPORT_SYMBOL_GPL(scsi_free_sgtables);
495 
scsi_mq_uninit_cmd(struct scsi_cmnd * cmd)496 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
497 {
498 	scsi_free_sgtables(cmd);
499 	scsi_uninit_cmd(cmd);
500 }
501 
scsi_run_queue_async(struct scsi_device * sdev)502 static void scsi_run_queue_async(struct scsi_device *sdev)
503 {
504 	if (scsi_host_in_recovery(sdev->host))
505 		return;
506 
507 	if (scsi_target(sdev)->single_lun ||
508 	    !list_empty(&sdev->host->starved_list)) {
509 		kblockd_schedule_work(&sdev->requeue_work);
510 	} else {
511 		/*
512 		 * smp_mb() present in sbitmap_queue_clear() or implied in
513 		 * .end_io is for ordering writing .device_busy in
514 		 * scsi_device_unbusy() and reading sdev->restarts.
515 		 */
516 		int old = atomic_read(&sdev->restarts);
517 
518 		/*
519 		 * ->restarts has to be kept as non-zero if new budget
520 		 *  contention occurs.
521 		 *
522 		 *  No need to run queue when either another re-run
523 		 *  queue wins in updating ->restarts or a new budget
524 		 *  contention occurs.
525 		 */
526 		if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)
527 			blk_mq_run_hw_queues(sdev->request_queue, true);
528 	}
529 }
530 
531 /* 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)532 static bool scsi_end_request(struct request *req, blk_status_t error,
533 		unsigned int bytes)
534 {
535 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
536 	struct scsi_device *sdev = cmd->device;
537 	struct request_queue *q = sdev->request_queue;
538 
539 	if (blk_update_request(req, error, bytes))
540 		return true;
541 
542 	// XXX:
543 	if (blk_queue_add_random(q))
544 		add_disk_randomness(req->q->disk);
545 
546 	WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&
547 		     !(cmd->flags & SCMD_INITIALIZED));
548 	cmd->flags = 0;
549 
550 	/*
551 	 * Calling rcu_barrier() is not necessary here because the
552 	 * SCSI error handler guarantees that the function called by
553 	 * call_rcu() has been called before scsi_end_request() is
554 	 * called.
555 	 */
556 	destroy_rcu_head(&cmd->rcu);
557 
558 	/*
559 	 * In the MQ case the command gets freed by __blk_mq_end_request,
560 	 * so we have to do all cleanup that depends on it earlier.
561 	 *
562 	 * We also can't kick the queues from irq context, so we
563 	 * will have to defer it to a workqueue.
564 	 */
565 	scsi_mq_uninit_cmd(cmd);
566 
567 	/*
568 	 * queue is still alive, so grab the ref for preventing it
569 	 * from being cleaned up during running queue.
570 	 */
571 	percpu_ref_get(&q->q_usage_counter);
572 
573 	__blk_mq_end_request(req, error);
574 
575 	scsi_run_queue_async(sdev);
576 
577 	percpu_ref_put(&q->q_usage_counter);
578 	return false;
579 }
580 
581 /**
582  * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
583  * @result:	scsi error code
584  *
585  * Translate a SCSI result code into a blk_status_t value.
586  */
scsi_result_to_blk_status(int result)587 static blk_status_t scsi_result_to_blk_status(int result)
588 {
589 	/*
590 	 * Check the scsi-ml byte first in case we converted a host or status
591 	 * byte.
592 	 */
593 	switch (scsi_ml_byte(result)) {
594 	case SCSIML_STAT_OK:
595 		break;
596 	case SCSIML_STAT_RESV_CONFLICT:
597 		return BLK_STS_RESV_CONFLICT;
598 	case SCSIML_STAT_NOSPC:
599 		return BLK_STS_NOSPC;
600 	case SCSIML_STAT_MED_ERROR:
601 		return BLK_STS_MEDIUM;
602 	case SCSIML_STAT_TGT_FAILURE:
603 		return BLK_STS_TARGET;
604 	case SCSIML_STAT_DL_TIMEOUT:
605 		return BLK_STS_DURATION_LIMIT;
606 	}
607 
608 	switch (host_byte(result)) {
609 	case DID_OK:
610 		if (scsi_status_is_good(result))
611 			return BLK_STS_OK;
612 		return BLK_STS_IOERR;
613 	case DID_TRANSPORT_FAILFAST:
614 	case DID_TRANSPORT_MARGINAL:
615 		return BLK_STS_TRANSPORT;
616 	default:
617 		return BLK_STS_IOERR;
618 	}
619 }
620 
621 /**
622  * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
623  * @rq: request to examine
624  *
625  * Description:
626  *     A request could be merge of IOs which require different failure
627  *     handling.  This function determines the number of bytes which
628  *     can be failed from the beginning of the request without
629  *     crossing into area which need to be retried further.
630  *
631  * Return:
632  *     The number of bytes to fail.
633  */
scsi_rq_err_bytes(const struct request * rq)634 static unsigned int scsi_rq_err_bytes(const struct request *rq)
635 {
636 	blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
637 	unsigned int bytes = 0;
638 	struct bio *bio;
639 
640 	if (!(rq->rq_flags & RQF_MIXED_MERGE))
641 		return blk_rq_bytes(rq);
642 
643 	/*
644 	 * Currently the only 'mixing' which can happen is between
645 	 * different fastfail types.  We can safely fail portions
646 	 * which have all the failfast bits that the first one has -
647 	 * the ones which are at least as eager to fail as the first
648 	 * one.
649 	 */
650 	for (bio = rq->bio; bio; bio = bio->bi_next) {
651 		if ((bio->bi_opf & ff) != ff)
652 			break;
653 		bytes += bio->bi_iter.bi_size;
654 	}
655 
656 	/* this could lead to infinite loop */
657 	BUG_ON(blk_rq_bytes(rq) && !bytes);
658 	return bytes;
659 }
660 
scsi_cmd_runtime_exceeced(struct scsi_cmnd * cmd)661 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
662 {
663 	struct request *req = scsi_cmd_to_rq(cmd);
664 	unsigned long wait_for;
665 
666 	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
667 		return false;
668 
669 	wait_for = (cmd->allowed + 1) * req->timeout;
670 	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
671 		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
672 			    wait_for/HZ);
673 		return true;
674 	}
675 	return false;
676 }
677 
678 /*
679  * When ALUA transition state is returned, reprep the cmd to
680  * use the ALUA handler's transition timeout. Delay the reprep
681  * 1 sec to avoid aggressive retries of the target in that
682  * state.
683  */
684 #define ALUA_TRANSITION_REPREP_DELAY	1000
685 
686 /* Helper for scsi_io_completion() when special action required. */
scsi_io_completion_action(struct scsi_cmnd * cmd,int result)687 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
688 {
689 	struct request *req = scsi_cmd_to_rq(cmd);
690 	int level = 0;
691 	enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
692 	      ACTION_RETRY, ACTION_DELAYED_RETRY} action;
693 	struct scsi_sense_hdr sshdr;
694 	bool sense_valid;
695 	bool sense_current = true;      /* false implies "deferred sense" */
696 	blk_status_t blk_stat;
697 
698 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
699 	if (sense_valid)
700 		sense_current = !scsi_sense_is_deferred(&sshdr);
701 
702 	blk_stat = scsi_result_to_blk_status(result);
703 
704 	if (host_byte(result) == DID_RESET) {
705 		/* Third party bus reset or reset for error recovery
706 		 * reasons.  Just retry the command and see what
707 		 * happens.
708 		 */
709 		action = ACTION_RETRY;
710 	} else if (sense_valid && sense_current) {
711 		switch (sshdr.sense_key) {
712 		case UNIT_ATTENTION:
713 			if (cmd->device->removable) {
714 				/* Detected disc change.  Set a bit
715 				 * and quietly refuse further access.
716 				 */
717 				cmd->device->changed = 1;
718 				action = ACTION_FAIL;
719 			} else {
720 				/* Must have been a power glitch, or a
721 				 * bus reset.  Could not have been a
722 				 * media change, so we just retry the
723 				 * command and see what happens.
724 				 */
725 				action = ACTION_RETRY;
726 			}
727 			break;
728 		case ILLEGAL_REQUEST:
729 			/* If we had an ILLEGAL REQUEST returned, then
730 			 * we may have performed an unsupported
731 			 * command.  The only thing this should be
732 			 * would be a ten byte read where only a six
733 			 * byte read was supported.  Also, on a system
734 			 * where READ CAPACITY failed, we may have
735 			 * read past the end of the disk.
736 			 */
737 			if ((cmd->device->use_10_for_rw &&
738 			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
739 			    (cmd->cmnd[0] == READ_10 ||
740 			     cmd->cmnd[0] == WRITE_10)) {
741 				/* This will issue a new 6-byte command. */
742 				cmd->device->use_10_for_rw = 0;
743 				action = ACTION_REPREP;
744 			} else if (sshdr.asc == 0x10) /* DIX */ {
745 				action = ACTION_FAIL;
746 				blk_stat = BLK_STS_PROTECTION;
747 			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
748 			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
749 				action = ACTION_FAIL;
750 				blk_stat = BLK_STS_TARGET;
751 			} else
752 				action = ACTION_FAIL;
753 			break;
754 		case ABORTED_COMMAND:
755 			action = ACTION_FAIL;
756 			if (sshdr.asc == 0x10) /* DIF */
757 				blk_stat = BLK_STS_PROTECTION;
758 			break;
759 		case NOT_READY:
760 			/* If the device is in the process of becoming
761 			 * ready, or has a temporary blockage, retry.
762 			 */
763 			if (sshdr.asc == 0x04) {
764 				switch (sshdr.ascq) {
765 				case 0x01: /* becoming ready */
766 				case 0x04: /* format in progress */
767 				case 0x05: /* rebuild in progress */
768 				case 0x06: /* recalculation in progress */
769 				case 0x07: /* operation in progress */
770 				case 0x08: /* Long write in progress */
771 				case 0x09: /* self test in progress */
772 				case 0x11: /* notify (enable spinup) required */
773 				case 0x14: /* space allocation in progress */
774 				case 0x1a: /* start stop unit in progress */
775 				case 0x1b: /* sanitize in progress */
776 				case 0x1d: /* configuration in progress */
777 					action = ACTION_DELAYED_RETRY;
778 					break;
779 				case 0x0a: /* ALUA state transition */
780 					action = ACTION_DELAYED_REPREP;
781 					break;
782 				/*
783 				 * Depopulation might take many hours,
784 				 * thus it is not worthwhile to retry.
785 				 */
786 				case 0x24: /* depopulation in progress */
787 				case 0x25: /* depopulation restore in progress */
788 					fallthrough;
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 		case COMPLETED:
810 			fallthrough;
811 		default:
812 			action = ACTION_FAIL;
813 			break;
814 		}
815 	} else
816 		action = ACTION_FAIL;
817 
818 	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
819 		action = ACTION_FAIL;
820 
821 	switch (action) {
822 	case ACTION_FAIL:
823 		/* Give up and fail the remainder of the request */
824 		if (!(req->rq_flags & RQF_QUIET)) {
825 			static DEFINE_RATELIMIT_STATE(_rs,
826 					DEFAULT_RATELIMIT_INTERVAL,
827 					DEFAULT_RATELIMIT_BURST);
828 
829 			if (unlikely(scsi_logging_level))
830 				level =
831 				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
832 						    SCSI_LOG_MLCOMPLETE_BITS);
833 
834 			/*
835 			 * if logging is enabled the failure will be printed
836 			 * in scsi_log_completion(), so avoid duplicate messages
837 			 */
838 			if (!level && __ratelimit(&_rs)) {
839 				scsi_print_result(cmd, NULL, FAILED);
840 				if (sense_valid)
841 					scsi_print_sense(cmd);
842 				scsi_print_command(cmd);
843 			}
844 		}
845 		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
846 			return;
847 		fallthrough;
848 	case ACTION_REPREP:
849 		scsi_mq_requeue_cmd(cmd, 0);
850 		break;
851 	case ACTION_DELAYED_REPREP:
852 		scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
853 		break;
854 	case ACTION_RETRY:
855 		/* Retry the same command immediately */
856 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
857 		break;
858 	case ACTION_DELAYED_RETRY:
859 		/* Retry the same command after a delay */
860 		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
861 		break;
862 	}
863 }
864 
865 /*
866  * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
867  * new result that may suppress further error checking. Also modifies
868  * *blk_statp in some cases.
869  */
scsi_io_completion_nz_result(struct scsi_cmnd * cmd,int result,blk_status_t * blk_statp)870 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
871 					blk_status_t *blk_statp)
872 {
873 	bool sense_valid;
874 	bool sense_current = true;	/* false implies "deferred sense" */
875 	struct request *req = scsi_cmd_to_rq(cmd);
876 	struct scsi_sense_hdr sshdr;
877 
878 	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
879 	if (sense_valid)
880 		sense_current = !scsi_sense_is_deferred(&sshdr);
881 
882 	if (blk_rq_is_passthrough(req)) {
883 		if (sense_valid) {
884 			/*
885 			 * SG_IO wants current and deferred errors
886 			 */
887 			cmd->sense_len = min(8 + cmd->sense_buffer[7],
888 					     SCSI_SENSE_BUFFERSIZE);
889 		}
890 		if (sense_current)
891 			*blk_statp = scsi_result_to_blk_status(result);
892 	} else if (blk_rq_bytes(req) == 0 && sense_current) {
893 		/*
894 		 * Flush commands do not transfers any data, and thus cannot use
895 		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
896 		 * This sets *blk_statp explicitly for the problem case.
897 		 */
898 		*blk_statp = scsi_result_to_blk_status(result);
899 	}
900 	/*
901 	 * Recovered errors need reporting, but they're always treated as
902 	 * success, so fiddle the result code here.  For passthrough requests
903 	 * we already took a copy of the original into sreq->result which
904 	 * is what gets returned to the user
905 	 */
906 	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
907 		bool do_print = true;
908 		/*
909 		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
910 		 * skip print since caller wants ATA registers. Only occurs
911 		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
912 		 */
913 		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
914 			do_print = false;
915 		else if (req->rq_flags & RQF_QUIET)
916 			do_print = false;
917 		if (do_print)
918 			scsi_print_sense(cmd);
919 		result = 0;
920 		/* for passthrough, *blk_statp may be set */
921 		*blk_statp = BLK_STS_OK;
922 	}
923 	/*
924 	 * Another corner case: the SCSI status byte is non-zero but 'good'.
925 	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
926 	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
927 	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
928 	 * intermediate statuses (both obsolete in SAM-4) as good.
929 	 */
930 	if ((result & 0xff) && scsi_status_is_good(result)) {
931 		result = 0;
932 		*blk_statp = BLK_STS_OK;
933 	}
934 	return result;
935 }
936 
937 /**
938  * scsi_io_completion - Completion processing for SCSI commands.
939  * @cmd:	command that is finished.
940  * @good_bytes:	number of processed bytes.
941  *
942  * We will finish off the specified number of sectors. If we are done, the
943  * command block will be released and the queue function will be goosed. If we
944  * are not done then we have to figure out what to do next:
945  *
946  *   a) We can call scsi_mq_requeue_cmd().  The request will be
947  *	unprepared and put back on the queue.  Then a new command will
948  *	be created for it.  This should be used if we made forward
949  *	progress, or if we want to switch from READ(10) to READ(6) for
950  *	example.
951  *
952  *   b) We can call scsi_io_completion_action().  The request will be
953  *	put back on the queue and retried using the same command as
954  *	before, possibly after a delay.
955  *
956  *   c) We can call scsi_end_request() with blk_stat other than
957  *	BLK_STS_OK, to fail the remainder of the request.
958  */
scsi_io_completion(struct scsi_cmnd * cmd,unsigned int good_bytes)959 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
960 {
961 	int result = cmd->result;
962 	struct request *req = scsi_cmd_to_rq(cmd);
963 	blk_status_t blk_stat = BLK_STS_OK;
964 
965 	if (unlikely(result))	/* a nz result may or may not be an error */
966 		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
967 
968 	/*
969 	 * Next deal with any sectors which we were able to correctly
970 	 * handle.
971 	 */
972 	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
973 		"%u sectors total, %d bytes done.\n",
974 		blk_rq_sectors(req), good_bytes));
975 
976 	/*
977 	 * Failed, zero length commands always need to drop down
978 	 * to retry code. Fast path should return in this block.
979 	 */
980 	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
981 		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
982 			return; /* no bytes remaining */
983 	}
984 
985 	/* Kill remainder if no retries. */
986 	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
987 		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
988 			WARN_ONCE(true,
989 			    "Bytes remaining after failed, no-retry command");
990 		return;
991 	}
992 
993 	/*
994 	 * If there had been no error, but we have leftover bytes in the
995 	 * request just queue the command up again.
996 	 */
997 	if (likely(result == 0))
998 		scsi_mq_requeue_cmd(cmd, 0);
999 	else
1000 		scsi_io_completion_action(cmd, result);
1001 }
1002 
scsi_cmd_needs_dma_drain(struct scsi_device * sdev,struct request * rq)1003 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1004 		struct request *rq)
1005 {
1006 	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1007 	       !op_is_write(req_op(rq)) &&
1008 	       sdev->host->hostt->dma_need_drain(rq);
1009 }
1010 
1011 /**
1012  * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1013  * @cmd: SCSI command data structure to initialize.
1014  *
1015  * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1016  * for @cmd.
1017  *
1018  * Returns:
1019  * * BLK_STS_OK       - on success
1020  * * BLK_STS_RESOURCE - if the failure is retryable
1021  * * BLK_STS_IOERR    - if the failure is fatal
1022  */
scsi_alloc_sgtables(struct scsi_cmnd * cmd)1023 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1024 {
1025 	struct scsi_device *sdev = cmd->device;
1026 	struct request *rq = scsi_cmd_to_rq(cmd);
1027 	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1028 	struct scatterlist *last_sg = NULL;
1029 	blk_status_t ret;
1030 	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1031 	int count;
1032 
1033 	if (WARN_ON_ONCE(!nr_segs))
1034 		return BLK_STS_IOERR;
1035 
1036 	/*
1037 	 * Make sure there is space for the drain.  The driver must adjust
1038 	 * max_hw_segments to be prepared for this.
1039 	 */
1040 	if (need_drain)
1041 		nr_segs++;
1042 
1043 	/*
1044 	 * If sg table allocation fails, requeue request later.
1045 	 */
1046 	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1047 			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1048 		return BLK_STS_RESOURCE;
1049 
1050 	/*
1051 	 * Next, walk the list, and fill in the addresses and sizes of
1052 	 * each segment.
1053 	 */
1054 	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1055 
1056 	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1057 		unsigned int pad_len =
1058 			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1059 
1060 		last_sg->length += pad_len;
1061 		cmd->extra_len += pad_len;
1062 	}
1063 
1064 	if (need_drain) {
1065 		sg_unmark_end(last_sg);
1066 		last_sg = sg_next(last_sg);
1067 		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1068 		sg_mark_end(last_sg);
1069 
1070 		cmd->extra_len += sdev->dma_drain_len;
1071 		count++;
1072 	}
1073 
1074 	BUG_ON(count > cmd->sdb.table.nents);
1075 	cmd->sdb.table.nents = count;
1076 	cmd->sdb.length = blk_rq_payload_bytes(rq);
1077 
1078 	if (blk_integrity_rq(rq)) {
1079 		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1080 		int ivecs;
1081 
1082 		if (WARN_ON_ONCE(!prot_sdb)) {
1083 			/*
1084 			 * This can happen if someone (e.g. multipath)
1085 			 * queues a command to a device on an adapter
1086 			 * that does not support DIX.
1087 			 */
1088 			ret = BLK_STS_IOERR;
1089 			goto out_free_sgtables;
1090 		}
1091 
1092 		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1093 
1094 		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1095 				prot_sdb->table.sgl,
1096 				SCSI_INLINE_PROT_SG_CNT)) {
1097 			ret = BLK_STS_RESOURCE;
1098 			goto out_free_sgtables;
1099 		}
1100 
1101 		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1102 						prot_sdb->table.sgl);
1103 		BUG_ON(count > ivecs);
1104 		BUG_ON(count > queue_max_integrity_segments(rq->q));
1105 
1106 		cmd->prot_sdb = prot_sdb;
1107 		cmd->prot_sdb->table.nents = count;
1108 	}
1109 
1110 	return BLK_STS_OK;
1111 out_free_sgtables:
1112 	scsi_free_sgtables(cmd);
1113 	return ret;
1114 }
1115 EXPORT_SYMBOL(scsi_alloc_sgtables);
1116 
1117 /**
1118  * scsi_initialize_rq - initialize struct scsi_cmnd partially
1119  * @rq: Request associated with the SCSI command to be initialized.
1120  *
1121  * This function initializes the members of struct scsi_cmnd that must be
1122  * initialized before request processing starts and that won't be
1123  * reinitialized if a SCSI command is requeued.
1124  */
scsi_initialize_rq(struct request * rq)1125 static void scsi_initialize_rq(struct request *rq)
1126 {
1127 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1128 
1129 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1130 	cmd->cmd_len = MAX_COMMAND_SIZE;
1131 	cmd->sense_len = 0;
1132 	init_rcu_head(&cmd->rcu);
1133 	cmd->jiffies_at_alloc = jiffies;
1134 	cmd->retries = 0;
1135 }
1136 
scsi_alloc_request(struct request_queue * q,blk_opf_t opf,blk_mq_req_flags_t flags)1137 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1138 				   blk_mq_req_flags_t flags)
1139 {
1140 	struct request *rq;
1141 
1142 	rq = blk_mq_alloc_request(q, opf, flags);
1143 	if (!IS_ERR(rq))
1144 		scsi_initialize_rq(rq);
1145 	return rq;
1146 }
1147 EXPORT_SYMBOL_GPL(scsi_alloc_request);
1148 
1149 /*
1150  * Only called when the request isn't completed by SCSI, and not freed by
1151  * SCSI
1152  */
scsi_cleanup_rq(struct request * rq)1153 static void scsi_cleanup_rq(struct request *rq)
1154 {
1155 	if (rq->rq_flags & RQF_DONTPREP) {
1156 		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1157 		rq->rq_flags &= ~RQF_DONTPREP;
1158 	}
1159 }
1160 
1161 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */
scsi_init_command(struct scsi_device * dev,struct scsi_cmnd * cmd)1162 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1163 {
1164 	struct request *rq = scsi_cmd_to_rq(cmd);
1165 
1166 	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1167 		cmd->flags |= SCMD_INITIALIZED;
1168 		scsi_initialize_rq(rq);
1169 	}
1170 
1171 	cmd->device = dev;
1172 	INIT_LIST_HEAD(&cmd->eh_entry);
1173 	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1174 }
1175 
scsi_setup_scsi_cmnd(struct scsi_device * sdev,struct request * req)1176 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1177 		struct request *req)
1178 {
1179 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1180 
1181 	/*
1182 	 * Passthrough requests may transfer data, in which case they must
1183 	 * a bio attached to them.  Or they might contain a SCSI command
1184 	 * that does not transfer data, in which case they may optionally
1185 	 * submit a request without an attached bio.
1186 	 */
1187 	if (req->bio) {
1188 		blk_status_t ret = scsi_alloc_sgtables(cmd);
1189 		if (unlikely(ret != BLK_STS_OK))
1190 			return ret;
1191 	} else {
1192 		BUG_ON(blk_rq_bytes(req));
1193 
1194 		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1195 	}
1196 
1197 	cmd->transfersize = blk_rq_bytes(req);
1198 	return BLK_STS_OK;
1199 }
1200 
1201 static blk_status_t
scsi_device_state_check(struct scsi_device * sdev,struct request * req)1202 scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1203 {
1204 	switch (sdev->sdev_state) {
1205 	case SDEV_CREATED:
1206 		return BLK_STS_OK;
1207 	case SDEV_OFFLINE:
1208 	case SDEV_TRANSPORT_OFFLINE:
1209 		/*
1210 		 * If the device is offline we refuse to process any
1211 		 * commands.  The device must be brought online
1212 		 * before trying any recovery commands.
1213 		 */
1214 		if (!sdev->offline_already) {
1215 			sdev->offline_already = true;
1216 			sdev_printk(KERN_ERR, sdev,
1217 				    "rejecting I/O to offline device\n");
1218 		}
1219 		return BLK_STS_IOERR;
1220 	case SDEV_DEL:
1221 		/*
1222 		 * If the device is fully deleted, we refuse to
1223 		 * process any commands as well.
1224 		 */
1225 		sdev_printk(KERN_ERR, sdev,
1226 			    "rejecting I/O to dead device\n");
1227 		return BLK_STS_IOERR;
1228 	case SDEV_BLOCK:
1229 	case SDEV_CREATED_BLOCK:
1230 		return BLK_STS_RESOURCE;
1231 	case SDEV_QUIESCE:
1232 		/*
1233 		 * If the device is blocked we only accept power management
1234 		 * commands.
1235 		 */
1236 		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1237 			return BLK_STS_RESOURCE;
1238 		return BLK_STS_OK;
1239 	default:
1240 		/*
1241 		 * For any other not fully online state we only allow
1242 		 * power management commands.
1243 		 */
1244 		if (req && !(req->rq_flags & RQF_PM))
1245 			return BLK_STS_OFFLINE;
1246 		return BLK_STS_OK;
1247 	}
1248 }
1249 
1250 /*
1251  * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1252  * and return the token else return -1.
1253  */
scsi_dev_queue_ready(struct request_queue * q,struct scsi_device * sdev)1254 static inline int scsi_dev_queue_ready(struct request_queue *q,
1255 				  struct scsi_device *sdev)
1256 {
1257 	int token;
1258 
1259 	token = sbitmap_get(&sdev->budget_map);
1260 	if (atomic_read(&sdev->device_blocked)) {
1261 		if (token < 0)
1262 			goto out;
1263 
1264 		if (scsi_device_busy(sdev) > 1)
1265 			goto out_dec;
1266 
1267 		/*
1268 		 * unblock after device_blocked iterates to zero
1269 		 */
1270 		if (atomic_dec_return(&sdev->device_blocked) > 0)
1271 			goto out_dec;
1272 		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1273 				   "unblocking device at zero depth\n"));
1274 	}
1275 
1276 	return token;
1277 out_dec:
1278 	if (token >= 0)
1279 		sbitmap_put(&sdev->budget_map, token);
1280 out:
1281 	return -1;
1282 }
1283 
1284 /*
1285  * scsi_target_queue_ready: checks if there we can send commands to target
1286  * @sdev: scsi device on starget to check.
1287  */
scsi_target_queue_ready(struct Scsi_Host * shost,struct scsi_device * sdev)1288 static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1289 					   struct scsi_device *sdev)
1290 {
1291 	struct scsi_target *starget = scsi_target(sdev);
1292 	unsigned int busy;
1293 
1294 	if (starget->single_lun) {
1295 		spin_lock_irq(shost->host_lock);
1296 		if (starget->starget_sdev_user &&
1297 		    starget->starget_sdev_user != sdev) {
1298 			spin_unlock_irq(shost->host_lock);
1299 			return 0;
1300 		}
1301 		starget->starget_sdev_user = sdev;
1302 		spin_unlock_irq(shost->host_lock);
1303 	}
1304 
1305 	if (starget->can_queue <= 0)
1306 		return 1;
1307 
1308 	busy = atomic_inc_return(&starget->target_busy) - 1;
1309 	if (atomic_read(&starget->target_blocked) > 0) {
1310 		if (busy)
1311 			goto starved;
1312 
1313 		/*
1314 		 * unblock after target_blocked iterates to zero
1315 		 */
1316 		if (atomic_dec_return(&starget->target_blocked) > 0)
1317 			goto out_dec;
1318 
1319 		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1320 				 "unblocking target at zero depth\n"));
1321 	}
1322 
1323 	if (busy >= starget->can_queue)
1324 		goto starved;
1325 
1326 	return 1;
1327 
1328 starved:
1329 	spin_lock_irq(shost->host_lock);
1330 	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1331 	spin_unlock_irq(shost->host_lock);
1332 out_dec:
1333 	if (starget->can_queue > 0)
1334 		atomic_dec(&starget->target_busy);
1335 	return 0;
1336 }
1337 
1338 /*
1339  * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1340  * return 0. We must end up running the queue again whenever 0 is
1341  * returned, else IO can hang.
1342  */
scsi_host_queue_ready(struct request_queue * q,struct Scsi_Host * shost,struct scsi_device * sdev,struct scsi_cmnd * cmd)1343 static inline int scsi_host_queue_ready(struct request_queue *q,
1344 				   struct Scsi_Host *shost,
1345 				   struct scsi_device *sdev,
1346 				   struct scsi_cmnd *cmd)
1347 {
1348 	if (atomic_read(&shost->host_blocked) > 0) {
1349 		if (scsi_host_busy(shost) > 0)
1350 			goto starved;
1351 
1352 		/*
1353 		 * unblock after host_blocked iterates to zero
1354 		 */
1355 		if (atomic_dec_return(&shost->host_blocked) > 0)
1356 			goto out_dec;
1357 
1358 		SCSI_LOG_MLQUEUE(3,
1359 			shost_printk(KERN_INFO, shost,
1360 				     "unblocking host at zero depth\n"));
1361 	}
1362 
1363 	if (shost->host_self_blocked)
1364 		goto starved;
1365 
1366 	/* We're OK to process the command, so we can't be starved */
1367 	if (!list_empty(&sdev->starved_entry)) {
1368 		spin_lock_irq(shost->host_lock);
1369 		if (!list_empty(&sdev->starved_entry))
1370 			list_del_init(&sdev->starved_entry);
1371 		spin_unlock_irq(shost->host_lock);
1372 	}
1373 
1374 	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1375 
1376 	return 1;
1377 
1378 starved:
1379 	spin_lock_irq(shost->host_lock);
1380 	if (list_empty(&sdev->starved_entry))
1381 		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1382 	spin_unlock_irq(shost->host_lock);
1383 out_dec:
1384 	scsi_dec_host_busy(shost, cmd);
1385 	return 0;
1386 }
1387 
1388 /*
1389  * Busy state exporting function for request stacking drivers.
1390  *
1391  * For efficiency, no lock is taken to check the busy state of
1392  * shost/starget/sdev, since the returned value is not guaranteed and
1393  * may be changed after request stacking drivers call the function,
1394  * regardless of taking lock or not.
1395  *
1396  * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1397  * needs to return 'not busy'. Otherwise, request stacking drivers
1398  * may hold requests forever.
1399  */
scsi_mq_lld_busy(struct request_queue * q)1400 static bool scsi_mq_lld_busy(struct request_queue *q)
1401 {
1402 	struct scsi_device *sdev = q->queuedata;
1403 	struct Scsi_Host *shost;
1404 
1405 	if (blk_queue_dying(q))
1406 		return false;
1407 
1408 	shost = sdev->host;
1409 
1410 	/*
1411 	 * Ignore host/starget busy state.
1412 	 * Since block layer does not have a concept of fairness across
1413 	 * multiple queues, congestion of host/starget needs to be handled
1414 	 * in SCSI layer.
1415 	 */
1416 	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1417 		return true;
1418 
1419 	return false;
1420 }
1421 
1422 /*
1423  * Block layer request completion callback. May be called from interrupt
1424  * context.
1425  */
scsi_complete(struct request * rq)1426 static void scsi_complete(struct request *rq)
1427 {
1428 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1429 	enum scsi_disposition disposition;
1430 
1431 	INIT_LIST_HEAD(&cmd->eh_entry);
1432 
1433 	atomic_inc(&cmd->device->iodone_cnt);
1434 	if (cmd->result)
1435 		atomic_inc(&cmd->device->ioerr_cnt);
1436 
1437 	disposition = scsi_decide_disposition(cmd);
1438 	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
1439 		disposition = SUCCESS;
1440 
1441 	scsi_log_completion(cmd, disposition);
1442 
1443 	switch (disposition) {
1444 	case SUCCESS:
1445 		scsi_finish_command(cmd);
1446 		break;
1447 	case NEEDS_RETRY:
1448 		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1449 		break;
1450 	case ADD_TO_MLQUEUE:
1451 		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1452 		break;
1453 	default:
1454 		scsi_eh_scmd_add(cmd);
1455 		break;
1456 	}
1457 }
1458 
1459 /**
1460  * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1461  * @cmd: command block we are dispatching.
1462  *
1463  * Return: nonzero return request was rejected and device's queue needs to be
1464  * plugged.
1465  */
scsi_dispatch_cmd(struct scsi_cmnd * cmd)1466 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1467 {
1468 	struct Scsi_Host *host = cmd->device->host;
1469 	int rtn = 0;
1470 
1471 	atomic_inc(&cmd->device->iorequest_cnt);
1472 
1473 	/* check if the device is still usable */
1474 	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1475 		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1476 		 * returns an immediate error upwards, and signals
1477 		 * that the device is no longer present */
1478 		cmd->result = DID_NO_CONNECT << 16;
1479 		goto done;
1480 	}
1481 
1482 	/* Check to see if the scsi lld made this device blocked. */
1483 	if (unlikely(scsi_device_blocked(cmd->device))) {
1484 		/*
1485 		 * in blocked state, the command is just put back on
1486 		 * the device queue.  The suspend state has already
1487 		 * blocked the queue so future requests should not
1488 		 * occur until the device transitions out of the
1489 		 * suspend state.
1490 		 */
1491 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1492 			"queuecommand : device blocked\n"));
1493 		atomic_dec(&cmd->device->iorequest_cnt);
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 		atomic_dec(&cmd->device->iorequest_cnt);
1527 		trace_scsi_dispatch_cmd_error(cmd, rtn);
1528 		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1529 		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1530 			rtn = SCSI_MLQUEUE_HOST_BUSY;
1531 
1532 		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1533 			"queuecommand : request rejected\n"));
1534 	}
1535 
1536 	return rtn;
1537  done:
1538 	scsi_done(cmd);
1539 	return 0;
1540 }
1541 
1542 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
scsi_mq_inline_sgl_size(struct Scsi_Host * shost)1543 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1544 {
1545 	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1546 		sizeof(struct scatterlist);
1547 }
1548 
scsi_prepare_cmd(struct request * req)1549 static blk_status_t scsi_prepare_cmd(struct request *req)
1550 {
1551 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1552 	struct scsi_device *sdev = req->q->queuedata;
1553 	struct Scsi_Host *shost = sdev->host;
1554 	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1555 	struct scatterlist *sg;
1556 
1557 	scsi_init_command(sdev, cmd);
1558 
1559 	cmd->eh_eflags = 0;
1560 	cmd->prot_type = 0;
1561 	cmd->prot_flags = 0;
1562 	cmd->submitter = 0;
1563 	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1564 	cmd->underflow = 0;
1565 	cmd->transfersize = 0;
1566 	cmd->host_scribble = NULL;
1567 	cmd->result = 0;
1568 	cmd->extra_len = 0;
1569 	cmd->state = 0;
1570 	if (in_flight)
1571 		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1572 
1573 	cmd->prot_op = SCSI_PROT_NORMAL;
1574 	if (blk_rq_bytes(req))
1575 		cmd->sc_data_direction = rq_dma_dir(req);
1576 	else
1577 		cmd->sc_data_direction = DMA_NONE;
1578 
1579 	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580 	cmd->sdb.table.sgl = sg;
1581 
1582 	if (scsi_host_get_prot(shost)) {
1583 		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1584 
1585 		cmd->prot_sdb->table.sgl =
1586 			(struct scatterlist *)(cmd->prot_sdb + 1);
1587 	}
1588 
1589 	/*
1590 	 * Special handling for passthrough commands, which don't go to the ULP
1591 	 * at all:
1592 	 */
1593 	if (blk_rq_is_passthrough(req))
1594 		return scsi_setup_scsi_cmnd(sdev, req);
1595 
1596 	if (sdev->handler && sdev->handler->prep_fn) {
1597 		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1598 
1599 		if (ret != BLK_STS_OK)
1600 			return ret;
1601 	}
1602 
1603 	/* Usually overridden by the ULP */
1604 	cmd->allowed = 0;
1605 	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1606 	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1607 }
1608 
scsi_done_internal(struct scsi_cmnd * cmd,bool complete_directly)1609 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1610 {
1611 	struct request *req = scsi_cmd_to_rq(cmd);
1612 
1613 	switch (cmd->submitter) {
1614 	case SUBMITTED_BY_BLOCK_LAYER:
1615 		break;
1616 	case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1617 		return scsi_eh_done(cmd);
1618 	case SUBMITTED_BY_SCSI_RESET_IOCTL:
1619 		return;
1620 	}
1621 
1622 	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1623 		return;
1624 	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1625 		return;
1626 	trace_scsi_dispatch_cmd_done(cmd);
1627 
1628 	if (complete_directly)
1629 		blk_mq_complete_request_direct(req, scsi_complete);
1630 	else
1631 		blk_mq_complete_request(req);
1632 }
1633 
scsi_done(struct scsi_cmnd * cmd)1634 void scsi_done(struct scsi_cmnd *cmd)
1635 {
1636 	scsi_done_internal(cmd, false);
1637 }
1638 EXPORT_SYMBOL(scsi_done);
1639 
scsi_done_direct(struct scsi_cmnd * cmd)1640 void scsi_done_direct(struct scsi_cmnd *cmd)
1641 {
1642 	scsi_done_internal(cmd, true);
1643 }
1644 EXPORT_SYMBOL(scsi_done_direct);
1645 
scsi_mq_put_budget(struct request_queue * q,int budget_token)1646 static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1647 {
1648 	struct scsi_device *sdev = q->queuedata;
1649 
1650 	sbitmap_put(&sdev->budget_map, budget_token);
1651 }
1652 
1653 /*
1654  * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1655  * not change behaviour from the previous unplug mechanism, experimentation
1656  * may prove this needs changing.
1657  */
1658 #define SCSI_QUEUE_DELAY 3
1659 
scsi_mq_get_budget(struct request_queue * q)1660 static int scsi_mq_get_budget(struct request_queue *q)
1661 {
1662 	struct scsi_device *sdev = q->queuedata;
1663 	int token = scsi_dev_queue_ready(q, sdev);
1664 
1665 	if (token >= 0)
1666 		return token;
1667 
1668 	atomic_inc(&sdev->restarts);
1669 
1670 	/*
1671 	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1672 	 * .restarts must be incremented before .device_busy is read because the
1673 	 * code in scsi_run_queue_async() depends on the order of these operations.
1674 	 */
1675 	smp_mb__after_atomic();
1676 
1677 	/*
1678 	 * If all in-flight requests originated from this LUN are completed
1679 	 * before reading .device_busy, sdev->device_busy will be observed as
1680 	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1681 	 * soon. Otherwise, completion of one of these requests will observe
1682 	 * the .restarts flag, and the request queue will be run for handling
1683 	 * this request, see scsi_end_request().
1684 	 */
1685 	if (unlikely(scsi_device_busy(sdev) == 0 &&
1686 				!scsi_device_blocked(sdev)))
1687 		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1688 	return -1;
1689 }
1690 
scsi_mq_set_rq_budget_token(struct request * req,int token)1691 static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1692 {
1693 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1694 
1695 	cmd->budget_token = token;
1696 }
1697 
scsi_mq_get_rq_budget_token(struct request * req)1698 static int scsi_mq_get_rq_budget_token(struct request *req)
1699 {
1700 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1701 
1702 	return cmd->budget_token;
1703 }
1704 
scsi_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)1705 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1706 			 const struct blk_mq_queue_data *bd)
1707 {
1708 	struct request *req = bd->rq;
1709 	struct request_queue *q = req->q;
1710 	struct scsi_device *sdev = q->queuedata;
1711 	struct Scsi_Host *shost = sdev->host;
1712 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1713 	blk_status_t ret;
1714 	int reason;
1715 
1716 	WARN_ON_ONCE(cmd->budget_token < 0);
1717 
1718 	/*
1719 	 * If the device is not in running state we will reject some or all
1720 	 * commands.
1721 	 */
1722 	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1723 		ret = scsi_device_state_check(sdev, req);
1724 		if (ret != BLK_STS_OK)
1725 			goto out_put_budget;
1726 	}
1727 
1728 	ret = BLK_STS_RESOURCE;
1729 	if (!scsi_target_queue_ready(shost, sdev))
1730 		goto out_put_budget;
1731 	if (unlikely(scsi_host_in_recovery(shost))) {
1732 		if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1733 			ret = BLK_STS_OFFLINE;
1734 		goto out_dec_target_busy;
1735 	}
1736 	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1737 		goto out_dec_target_busy;
1738 
1739 	/*
1740 	 * Only clear the driver-private command data if the LLD does not supply
1741 	 * a function to initialize that data.
1742 	 */
1743 	if (shost->hostt->cmd_size && !shost->hostt->init_cmd_priv)
1744 		memset(cmd + 1, 0, shost->hostt->cmd_size);
1745 
1746 	if (!(req->rq_flags & RQF_DONTPREP)) {
1747 		ret = scsi_prepare_cmd(req);
1748 		if (ret != BLK_STS_OK)
1749 			goto out_dec_host_busy;
1750 		req->rq_flags |= RQF_DONTPREP;
1751 	} else {
1752 		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1753 	}
1754 
1755 	cmd->flags &= SCMD_PRESERVED_FLAGS;
1756 	if (sdev->simple_tags)
1757 		cmd->flags |= SCMD_TAGGED;
1758 	if (bd->last)
1759 		cmd->flags |= SCMD_LAST;
1760 
1761 	scsi_set_resid(cmd, 0);
1762 	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1763 	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1764 
1765 	blk_mq_start_request(req);
1766 	reason = scsi_dispatch_cmd(cmd);
1767 	if (reason) {
1768 		scsi_set_blocked(cmd, reason);
1769 		ret = BLK_STS_RESOURCE;
1770 		goto out_dec_host_busy;
1771 	}
1772 
1773 	return BLK_STS_OK;
1774 
1775 out_dec_host_busy:
1776 	scsi_dec_host_busy(shost, cmd);
1777 out_dec_target_busy:
1778 	if (scsi_target(sdev)->can_queue > 0)
1779 		atomic_dec(&scsi_target(sdev)->target_busy);
1780 out_put_budget:
1781 	scsi_mq_put_budget(q, cmd->budget_token);
1782 	cmd->budget_token = -1;
1783 	switch (ret) {
1784 	case BLK_STS_OK:
1785 		break;
1786 	case BLK_STS_RESOURCE:
1787 	case BLK_STS_ZONE_RESOURCE:
1788 		if (scsi_device_blocked(sdev))
1789 			ret = BLK_STS_DEV_RESOURCE;
1790 		break;
1791 	case BLK_STS_AGAIN:
1792 		cmd->result = DID_BUS_BUSY << 16;
1793 		if (req->rq_flags & RQF_DONTPREP)
1794 			scsi_mq_uninit_cmd(cmd);
1795 		break;
1796 	default:
1797 		if (unlikely(!scsi_device_online(sdev)))
1798 			cmd->result = DID_NO_CONNECT << 16;
1799 		else
1800 			cmd->result = DID_ERROR << 16;
1801 		/*
1802 		 * Make sure to release all allocated resources when
1803 		 * we hit an error, as we will never see this command
1804 		 * again.
1805 		 */
1806 		if (req->rq_flags & RQF_DONTPREP)
1807 			scsi_mq_uninit_cmd(cmd);
1808 		scsi_run_queue_async(sdev);
1809 		break;
1810 	}
1811 	return ret;
1812 }
1813 
scsi_mq_init_request(struct blk_mq_tag_set * set,struct request * rq,unsigned int hctx_idx,unsigned int numa_node)1814 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1815 				unsigned int hctx_idx, unsigned int numa_node)
1816 {
1817 	struct Scsi_Host *shost = set->driver_data;
1818 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1819 	struct scatterlist *sg;
1820 	int ret = 0;
1821 
1822 	cmd->sense_buffer =
1823 		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
1824 	if (!cmd->sense_buffer)
1825 		return -ENOMEM;
1826 
1827 	if (scsi_host_get_prot(shost)) {
1828 		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1829 			shost->hostt->cmd_size;
1830 		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1831 	}
1832 
1833 	if (shost->hostt->init_cmd_priv) {
1834 		ret = shost->hostt->init_cmd_priv(shost, cmd);
1835 		if (ret < 0)
1836 			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1837 	}
1838 
1839 	return ret;
1840 }
1841 
scsi_mq_exit_request(struct blk_mq_tag_set * set,struct request * rq,unsigned int hctx_idx)1842 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1843 				 unsigned int hctx_idx)
1844 {
1845 	struct Scsi_Host *shost = set->driver_data;
1846 	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1847 
1848 	if (shost->hostt->exit_cmd_priv)
1849 		shost->hostt->exit_cmd_priv(shost, cmd);
1850 	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1851 }
1852 
1853 
scsi_mq_poll(struct blk_mq_hw_ctx * hctx,struct io_comp_batch * iob)1854 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1855 {
1856 	struct Scsi_Host *shost = hctx->driver_data;
1857 
1858 	if (shost->hostt->mq_poll)
1859 		return shost->hostt->mq_poll(shost, hctx->queue_num);
1860 
1861 	return 0;
1862 }
1863 
scsi_init_hctx(struct blk_mq_hw_ctx * hctx,void * data,unsigned int hctx_idx)1864 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1865 			  unsigned int hctx_idx)
1866 {
1867 	struct Scsi_Host *shost = data;
1868 
1869 	hctx->driver_data = shost;
1870 	return 0;
1871 }
1872 
scsi_map_queues(struct blk_mq_tag_set * set)1873 static void scsi_map_queues(struct blk_mq_tag_set *set)
1874 {
1875 	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1876 
1877 	if (shost->hostt->map_queues)
1878 		return shost->hostt->map_queues(shost);
1879 	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1880 }
1881 
__scsi_init_queue(struct Scsi_Host * shost,struct request_queue * q)1882 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1883 {
1884 	struct device *dev = shost->dma_dev;
1885 
1886 	/*
1887 	 * this limit is imposed by hardware restrictions
1888 	 */
1889 	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1890 					SG_MAX_SEGMENTS));
1891 
1892 	if (scsi_host_prot_dma(shost)) {
1893 		shost->sg_prot_tablesize =
1894 			min_not_zero(shost->sg_prot_tablesize,
1895 				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1896 		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1897 		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1898 	}
1899 
1900 	blk_queue_max_hw_sectors(q, shost->max_sectors);
1901 	blk_queue_segment_boundary(q, shost->dma_boundary);
1902 	dma_set_seg_boundary(dev, shost->dma_boundary);
1903 
1904 	blk_queue_max_segment_size(q, shost->max_segment_size);
1905 	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1906 	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1907 
1908 	/*
1909 	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1910 	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1911 	 * which is set by the platform.
1912 	 *
1913 	 * Devices that require a bigger alignment can increase it later.
1914 	 */
1915 	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1916 }
1917 EXPORT_SYMBOL_GPL(__scsi_init_queue);
1918 
1919 static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1920 	.get_budget	= scsi_mq_get_budget,
1921 	.put_budget	= scsi_mq_put_budget,
1922 	.queue_rq	= scsi_queue_rq,
1923 	.complete	= scsi_complete,
1924 	.timeout	= scsi_timeout,
1925 #ifdef CONFIG_BLK_DEBUG_FS
1926 	.show_rq	= scsi_show_rq,
1927 #endif
1928 	.init_request	= scsi_mq_init_request,
1929 	.exit_request	= scsi_mq_exit_request,
1930 	.cleanup_rq	= scsi_cleanup_rq,
1931 	.busy		= scsi_mq_lld_busy,
1932 	.map_queues	= scsi_map_queues,
1933 	.init_hctx	= scsi_init_hctx,
1934 	.poll		= scsi_mq_poll,
1935 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1936 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1937 };
1938 
1939 
scsi_commit_rqs(struct blk_mq_hw_ctx * hctx)1940 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1941 {
1942 	struct Scsi_Host *shost = hctx->driver_data;
1943 
1944 	shost->hostt->commit_rqs(shost, hctx->queue_num);
1945 }
1946 
1947 static const struct blk_mq_ops scsi_mq_ops = {
1948 	.get_budget	= scsi_mq_get_budget,
1949 	.put_budget	= scsi_mq_put_budget,
1950 	.queue_rq	= scsi_queue_rq,
1951 	.commit_rqs	= scsi_commit_rqs,
1952 	.complete	= scsi_complete,
1953 	.timeout	= scsi_timeout,
1954 #ifdef CONFIG_BLK_DEBUG_FS
1955 	.show_rq	= scsi_show_rq,
1956 #endif
1957 	.init_request	= scsi_mq_init_request,
1958 	.exit_request	= scsi_mq_exit_request,
1959 	.cleanup_rq	= scsi_cleanup_rq,
1960 	.busy		= scsi_mq_lld_busy,
1961 	.map_queues	= scsi_map_queues,
1962 	.init_hctx	= scsi_init_hctx,
1963 	.poll		= scsi_mq_poll,
1964 	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
1965 	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
1966 };
1967 
scsi_mq_setup_tags(struct Scsi_Host * shost)1968 int scsi_mq_setup_tags(struct Scsi_Host *shost)
1969 {
1970 	unsigned int cmd_size, sgl_size;
1971 	struct blk_mq_tag_set *tag_set = &shost->tag_set;
1972 
1973 	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1974 				scsi_mq_inline_sgl_size(shost));
1975 	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1976 	if (scsi_host_get_prot(shost))
1977 		cmd_size += sizeof(struct scsi_data_buffer) +
1978 			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1979 
1980 	memset(tag_set, 0, sizeof(*tag_set));
1981 	if (shost->hostt->commit_rqs)
1982 		tag_set->ops = &scsi_mq_ops;
1983 	else
1984 		tag_set->ops = &scsi_mq_ops_no_commit;
1985 	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
1986 	tag_set->nr_maps = shost->nr_maps ? : 1;
1987 	tag_set->queue_depth = shost->can_queue;
1988 	tag_set->cmd_size = cmd_size;
1989 	tag_set->numa_node = dev_to_node(shost->dma_dev);
1990 	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1991 	tag_set->flags |=
1992 		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
1993 	if (shost->queuecommand_may_block)
1994 		tag_set->flags |= BLK_MQ_F_BLOCKING;
1995 	tag_set->driver_data = shost;
1996 	if (shost->host_tagset)
1997 		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
1998 
1999 	return blk_mq_alloc_tag_set(tag_set);
2000 }
2001 
scsi_mq_free_tags(struct kref * kref)2002 void scsi_mq_free_tags(struct kref *kref)
2003 {
2004 	struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2005 					       tagset_refcnt);
2006 
2007 	blk_mq_free_tag_set(&shost->tag_set);
2008 	complete(&shost->tagset_freed);
2009 }
2010 
2011 /**
2012  * scsi_device_from_queue - return sdev associated with a request_queue
2013  * @q: The request queue to return the sdev from
2014  *
2015  * Return the sdev associated with a request queue or NULL if the
2016  * request_queue does not reference a SCSI device.
2017  */
scsi_device_from_queue(struct request_queue * q)2018 struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2019 {
2020 	struct scsi_device *sdev = NULL;
2021 
2022 	if (q->mq_ops == &scsi_mq_ops_no_commit ||
2023 	    q->mq_ops == &scsi_mq_ops)
2024 		sdev = q->queuedata;
2025 	if (!sdev || !get_device(&sdev->sdev_gendev))
2026 		sdev = NULL;
2027 
2028 	return sdev;
2029 }
2030 /*
2031  * pktcdvd should have been integrated into the SCSI layers, but for historical
2032  * reasons like the old IDE driver it isn't.  This export allows it to safely
2033  * probe if a given device is a SCSI one and only attach to that.
2034  */
2035 #ifdef CONFIG_CDROM_PKTCDVD_MODULE
2036 EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2037 #endif
2038 
2039 /**
2040  * scsi_block_requests - Utility function used by low-level drivers to prevent
2041  * further commands from being queued to the device.
2042  * @shost:  host in question
2043  *
2044  * There is no timer nor any other means by which the requests get unblocked
2045  * other than the low-level driver calling scsi_unblock_requests().
2046  */
scsi_block_requests(struct Scsi_Host * shost)2047 void scsi_block_requests(struct Scsi_Host *shost)
2048 {
2049 	shost->host_self_blocked = 1;
2050 }
2051 EXPORT_SYMBOL(scsi_block_requests);
2052 
2053 /**
2054  * scsi_unblock_requests - Utility function used by low-level drivers to allow
2055  * further commands to be queued to the device.
2056  * @shost:  host in question
2057  *
2058  * There is no timer nor any other means by which the requests get unblocked
2059  * other than the low-level driver calling scsi_unblock_requests(). This is done
2060  * as an API function so that changes to the internals of the scsi mid-layer
2061  * won't require wholesale changes to drivers that use this feature.
2062  */
scsi_unblock_requests(struct Scsi_Host * shost)2063 void scsi_unblock_requests(struct Scsi_Host *shost)
2064 {
2065 	shost->host_self_blocked = 0;
2066 	scsi_run_host_queues(shost);
2067 }
2068 EXPORT_SYMBOL(scsi_unblock_requests);
2069 
scsi_exit_queue(void)2070 void scsi_exit_queue(void)
2071 {
2072 	kmem_cache_destroy(scsi_sense_cache);
2073 }
2074 
2075 /**
2076  *	scsi_mode_select - issue a mode select
2077  *	@sdev:	SCSI device to be queried
2078  *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2079  *	@sp:	Save page bit (0 == don't save, 1 == save)
2080  *	@buffer: request buffer (may not be smaller than eight bytes)
2081  *	@len:	length of request buffer.
2082  *	@timeout: command timeout
2083  *	@retries: number of retries before failing
2084  *	@data: returns a structure abstracting the mode header data
2085  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2086  *		must be SCSI_SENSE_BUFFERSIZE big.
2087  *
2088  *	Returns zero if successful; negative error number or scsi
2089  *	status on error
2090  *
2091  */
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)2092 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2093 		     unsigned char *buffer, int len, int timeout, int retries,
2094 		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2095 {
2096 	unsigned char cmd[10];
2097 	unsigned char *real_buffer;
2098 	const struct scsi_exec_args exec_args = {
2099 		.sshdr = sshdr,
2100 	};
2101 	int ret;
2102 
2103 	memset(cmd, 0, sizeof(cmd));
2104 	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2105 
2106 	/*
2107 	 * Use MODE SELECT(10) if the device asked for it or if the mode page
2108 	 * and the mode select header cannot fit within the maximumm 255 bytes
2109 	 * of the MODE SELECT(6) command.
2110 	 */
2111 	if (sdev->use_10_for_ms ||
2112 	    len + 4 > 255 ||
2113 	    data->block_descriptor_length > 255) {
2114 		if (len > 65535 - 8)
2115 			return -EINVAL;
2116 		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2117 		if (!real_buffer)
2118 			return -ENOMEM;
2119 		memcpy(real_buffer + 8, buffer, len);
2120 		len += 8;
2121 		real_buffer[0] = 0;
2122 		real_buffer[1] = 0;
2123 		real_buffer[2] = data->medium_type;
2124 		real_buffer[3] = data->device_specific;
2125 		real_buffer[4] = data->longlba ? 0x01 : 0;
2126 		real_buffer[5] = 0;
2127 		put_unaligned_be16(data->block_descriptor_length,
2128 				   &real_buffer[6]);
2129 
2130 		cmd[0] = MODE_SELECT_10;
2131 		put_unaligned_be16(len, &cmd[7]);
2132 	} else {
2133 		if (data->longlba)
2134 			return -EINVAL;
2135 
2136 		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2137 		if (!real_buffer)
2138 			return -ENOMEM;
2139 		memcpy(real_buffer + 4, buffer, len);
2140 		len += 4;
2141 		real_buffer[0] = 0;
2142 		real_buffer[1] = data->medium_type;
2143 		real_buffer[2] = data->device_specific;
2144 		real_buffer[3] = data->block_descriptor_length;
2145 
2146 		cmd[0] = MODE_SELECT;
2147 		cmd[4] = len;
2148 	}
2149 
2150 	ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2151 			       timeout, retries, &exec_args);
2152 	kfree(real_buffer);
2153 	return ret;
2154 }
2155 EXPORT_SYMBOL_GPL(scsi_mode_select);
2156 
2157 /**
2158  *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2159  *	@sdev:	SCSI device to be queried
2160  *	@dbd:	set to prevent mode sense from returning block descriptors
2161  *	@modepage: mode page being requested
2162  *	@subpage: sub-page of the mode page being requested
2163  *	@buffer: request buffer (may not be smaller than eight bytes)
2164  *	@len:	length of request buffer.
2165  *	@timeout: command timeout
2166  *	@retries: number of retries before failing
2167  *	@data: returns a structure abstracting the mode header data
2168  *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2169  *		must be SCSI_SENSE_BUFFERSIZE big.
2170  *
2171  *	Returns zero if successful, or a negative error number on failure
2172  */
2173 int
scsi_mode_sense(struct scsi_device * sdev,int dbd,int modepage,int subpage,unsigned char * buffer,int len,int timeout,int retries,struct scsi_mode_data * data,struct scsi_sense_hdr * sshdr)2174 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2175 		  unsigned char *buffer, int len, int timeout, int retries,
2176 		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2177 {
2178 	unsigned char cmd[12];
2179 	int use_10_for_ms;
2180 	int header_length;
2181 	int result, retry_count = retries;
2182 	struct scsi_sense_hdr my_sshdr;
2183 	const struct scsi_exec_args exec_args = {
2184 		/* caller might not be interested in sense, but we need it */
2185 		.sshdr = sshdr ? : &my_sshdr,
2186 	};
2187 
2188 	memset(data, 0, sizeof(*data));
2189 	memset(&cmd[0], 0, 12);
2190 
2191 	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2192 	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2193 	cmd[2] = modepage;
2194 	cmd[3] = subpage;
2195 
2196 	sshdr = exec_args.sshdr;
2197 
2198  retry:
2199 	use_10_for_ms = sdev->use_10_for_ms || len > 255;
2200 
2201 	if (use_10_for_ms) {
2202 		if (len < 8 || len > 65535)
2203 			return -EINVAL;
2204 
2205 		cmd[0] = MODE_SENSE_10;
2206 		put_unaligned_be16(len, &cmd[7]);
2207 		header_length = 8;
2208 	} else {
2209 		if (len < 4)
2210 			return -EINVAL;
2211 
2212 		cmd[0] = MODE_SENSE;
2213 		cmd[4] = len;
2214 		header_length = 4;
2215 	}
2216 
2217 	memset(buffer, 0, len);
2218 
2219 	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2220 				  timeout, retries, &exec_args);
2221 	if (result < 0)
2222 		return result;
2223 
2224 	/* This code looks awful: what it's doing is making sure an
2225 	 * ILLEGAL REQUEST sense return identifies the actual command
2226 	 * byte as the problem.  MODE_SENSE commands can return
2227 	 * ILLEGAL REQUEST if the code page isn't supported */
2228 
2229 	if (!scsi_status_is_good(result)) {
2230 		if (scsi_sense_valid(sshdr)) {
2231 			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2232 			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2233 				/*
2234 				 * Invalid command operation code: retry using
2235 				 * MODE SENSE(6) if this was a MODE SENSE(10)
2236 				 * request, except if the request mode page is
2237 				 * too large for MODE SENSE single byte
2238 				 * allocation length field.
2239 				 */
2240 				if (use_10_for_ms) {
2241 					if (len > 255)
2242 						return -EIO;
2243 					sdev->use_10_for_ms = 0;
2244 					goto retry;
2245 				}
2246 			}
2247 			if (scsi_status_is_check_condition(result) &&
2248 			    sshdr->sense_key == UNIT_ATTENTION &&
2249 			    retry_count) {
2250 				retry_count--;
2251 				goto retry;
2252 			}
2253 		}
2254 		return -EIO;
2255 	}
2256 	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2257 		     (modepage == 6 || modepage == 8))) {
2258 		/* Initio breakage? */
2259 		header_length = 0;
2260 		data->length = 13;
2261 		data->medium_type = 0;
2262 		data->device_specific = 0;
2263 		data->longlba = 0;
2264 		data->block_descriptor_length = 0;
2265 	} else if (use_10_for_ms) {
2266 		data->length = get_unaligned_be16(&buffer[0]) + 2;
2267 		data->medium_type = buffer[2];
2268 		data->device_specific = buffer[3];
2269 		data->longlba = buffer[4] & 0x01;
2270 		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2271 	} else {
2272 		data->length = buffer[0] + 1;
2273 		data->medium_type = buffer[1];
2274 		data->device_specific = buffer[2];
2275 		data->block_descriptor_length = buffer[3];
2276 	}
2277 	data->header_length = header_length;
2278 
2279 	return 0;
2280 }
2281 EXPORT_SYMBOL(scsi_mode_sense);
2282 
2283 /**
2284  *	scsi_test_unit_ready - test if unit is ready
2285  *	@sdev:	scsi device to change the state of.
2286  *	@timeout: command timeout
2287  *	@retries: number of retries before failing
2288  *	@sshdr: outpout pointer for decoded sense information.
2289  *
2290  *	Returns zero if unsuccessful or an error if TUR failed.  For
2291  *	removable media, UNIT_ATTENTION sets ->changed flag.
2292  **/
2293 int
scsi_test_unit_ready(struct scsi_device * sdev,int timeout,int retries,struct scsi_sense_hdr * sshdr)2294 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2295 		     struct scsi_sense_hdr *sshdr)
2296 {
2297 	char cmd[] = {
2298 		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2299 	};
2300 	const struct scsi_exec_args exec_args = {
2301 		.sshdr = sshdr,
2302 	};
2303 	int result;
2304 
2305 	/* try to eat the UNIT_ATTENTION if there are enough retries */
2306 	do {
2307 		result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2308 					  timeout, 1, &exec_args);
2309 		if (sdev->removable && scsi_sense_valid(sshdr) &&
2310 		    sshdr->sense_key == UNIT_ATTENTION)
2311 			sdev->changed = 1;
2312 	} while (scsi_sense_valid(sshdr) &&
2313 		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2314 
2315 	return result;
2316 }
2317 EXPORT_SYMBOL(scsi_test_unit_ready);
2318 
2319 /**
2320  *	scsi_device_set_state - Take the given device through the device state model.
2321  *	@sdev:	scsi device to change the state of.
2322  *	@state:	state to change to.
2323  *
2324  *	Returns zero if successful or an error if the requested
2325  *	transition is illegal.
2326  */
2327 int
scsi_device_set_state(struct scsi_device * sdev,enum scsi_device_state state)2328 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2329 {
2330 	enum scsi_device_state oldstate = sdev->sdev_state;
2331 
2332 	if (state == oldstate)
2333 		return 0;
2334 
2335 	switch (state) {
2336 	case SDEV_CREATED:
2337 		switch (oldstate) {
2338 		case SDEV_CREATED_BLOCK:
2339 			break;
2340 		default:
2341 			goto illegal;
2342 		}
2343 		break;
2344 
2345 	case SDEV_RUNNING:
2346 		switch (oldstate) {
2347 		case SDEV_CREATED:
2348 		case SDEV_OFFLINE:
2349 		case SDEV_TRANSPORT_OFFLINE:
2350 		case SDEV_QUIESCE:
2351 		case SDEV_BLOCK:
2352 			break;
2353 		default:
2354 			goto illegal;
2355 		}
2356 		break;
2357 
2358 	case SDEV_QUIESCE:
2359 		switch (oldstate) {
2360 		case SDEV_RUNNING:
2361 		case SDEV_OFFLINE:
2362 		case SDEV_TRANSPORT_OFFLINE:
2363 			break;
2364 		default:
2365 			goto illegal;
2366 		}
2367 		break;
2368 
2369 	case SDEV_OFFLINE:
2370 	case SDEV_TRANSPORT_OFFLINE:
2371 		switch (oldstate) {
2372 		case SDEV_CREATED:
2373 		case SDEV_RUNNING:
2374 		case SDEV_QUIESCE:
2375 		case SDEV_BLOCK:
2376 			break;
2377 		default:
2378 			goto illegal;
2379 		}
2380 		break;
2381 
2382 	case SDEV_BLOCK:
2383 		switch (oldstate) {
2384 		case SDEV_RUNNING:
2385 		case SDEV_CREATED_BLOCK:
2386 		case SDEV_QUIESCE:
2387 		case SDEV_OFFLINE:
2388 			break;
2389 		default:
2390 			goto illegal;
2391 		}
2392 		break;
2393 
2394 	case SDEV_CREATED_BLOCK:
2395 		switch (oldstate) {
2396 		case SDEV_CREATED:
2397 			break;
2398 		default:
2399 			goto illegal;
2400 		}
2401 		break;
2402 
2403 	case SDEV_CANCEL:
2404 		switch (oldstate) {
2405 		case SDEV_CREATED:
2406 		case SDEV_RUNNING:
2407 		case SDEV_QUIESCE:
2408 		case SDEV_OFFLINE:
2409 		case SDEV_TRANSPORT_OFFLINE:
2410 			break;
2411 		default:
2412 			goto illegal;
2413 		}
2414 		break;
2415 
2416 	case SDEV_DEL:
2417 		switch (oldstate) {
2418 		case SDEV_CREATED:
2419 		case SDEV_RUNNING:
2420 		case SDEV_OFFLINE:
2421 		case SDEV_TRANSPORT_OFFLINE:
2422 		case SDEV_CANCEL:
2423 		case SDEV_BLOCK:
2424 		case SDEV_CREATED_BLOCK:
2425 			break;
2426 		default:
2427 			goto illegal;
2428 		}
2429 		break;
2430 
2431 	}
2432 	sdev->offline_already = false;
2433 	sdev->sdev_state = state;
2434 	return 0;
2435 
2436  illegal:
2437 	SCSI_LOG_ERROR_RECOVERY(1,
2438 				sdev_printk(KERN_ERR, sdev,
2439 					    "Illegal state transition %s->%s",
2440 					    scsi_device_state_name(oldstate),
2441 					    scsi_device_state_name(state))
2442 				);
2443 	return -EINVAL;
2444 }
2445 EXPORT_SYMBOL(scsi_device_set_state);
2446 
2447 /**
2448  *	scsi_evt_emit - emit a single SCSI device uevent
2449  *	@sdev: associated SCSI device
2450  *	@evt: event to emit
2451  *
2452  *	Send a single uevent (scsi_event) to the associated scsi_device.
2453  */
scsi_evt_emit(struct scsi_device * sdev,struct scsi_event * evt)2454 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2455 {
2456 	int idx = 0;
2457 	char *envp[3];
2458 
2459 	switch (evt->evt_type) {
2460 	case SDEV_EVT_MEDIA_CHANGE:
2461 		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2462 		break;
2463 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2464 		scsi_rescan_device(sdev);
2465 		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2466 		break;
2467 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2468 		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2469 		break;
2470 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2471 	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2472 		break;
2473 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2474 		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2475 		break;
2476 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2477 		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2478 		break;
2479 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2480 		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2481 		break;
2482 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2483 		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2484 		break;
2485 	default:
2486 		/* do nothing */
2487 		break;
2488 	}
2489 
2490 	envp[idx++] = NULL;
2491 
2492 	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2493 }
2494 
2495 /**
2496  *	scsi_evt_thread - send a uevent for each scsi event
2497  *	@work: work struct for scsi_device
2498  *
2499  *	Dispatch queued events to their associated scsi_device kobjects
2500  *	as uevents.
2501  */
scsi_evt_thread(struct work_struct * work)2502 void scsi_evt_thread(struct work_struct *work)
2503 {
2504 	struct scsi_device *sdev;
2505 	enum scsi_device_event evt_type;
2506 	LIST_HEAD(event_list);
2507 
2508 	sdev = container_of(work, struct scsi_device, event_work);
2509 
2510 	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2511 		if (test_and_clear_bit(evt_type, sdev->pending_events))
2512 			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2513 
2514 	while (1) {
2515 		struct scsi_event *evt;
2516 		struct list_head *this, *tmp;
2517 		unsigned long flags;
2518 
2519 		spin_lock_irqsave(&sdev->list_lock, flags);
2520 		list_splice_init(&sdev->event_list, &event_list);
2521 		spin_unlock_irqrestore(&sdev->list_lock, flags);
2522 
2523 		if (list_empty(&event_list))
2524 			break;
2525 
2526 		list_for_each_safe(this, tmp, &event_list) {
2527 			evt = list_entry(this, struct scsi_event, node);
2528 			list_del(&evt->node);
2529 			scsi_evt_emit(sdev, evt);
2530 			kfree(evt);
2531 		}
2532 	}
2533 }
2534 
2535 /**
2536  * 	sdev_evt_send - send asserted event to uevent thread
2537  *	@sdev: scsi_device event occurred on
2538  *	@evt: event to send
2539  *
2540  *	Assert scsi device event asynchronously.
2541  */
sdev_evt_send(struct scsi_device * sdev,struct scsi_event * evt)2542 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2543 {
2544 	unsigned long flags;
2545 
2546 #if 0
2547 	/* FIXME: currently this check eliminates all media change events
2548 	 * for polled devices.  Need to update to discriminate between AN
2549 	 * and polled events */
2550 	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2551 		kfree(evt);
2552 		return;
2553 	}
2554 #endif
2555 
2556 	spin_lock_irqsave(&sdev->list_lock, flags);
2557 	list_add_tail(&evt->node, &sdev->event_list);
2558 	schedule_work(&sdev->event_work);
2559 	spin_unlock_irqrestore(&sdev->list_lock, flags);
2560 }
2561 EXPORT_SYMBOL_GPL(sdev_evt_send);
2562 
2563 /**
2564  * 	sdev_evt_alloc - allocate a new scsi event
2565  *	@evt_type: type of event to allocate
2566  *	@gfpflags: GFP flags for allocation
2567  *
2568  *	Allocates and returns a new scsi_event.
2569  */
sdev_evt_alloc(enum scsi_device_event evt_type,gfp_t gfpflags)2570 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2571 				  gfp_t gfpflags)
2572 {
2573 	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2574 	if (!evt)
2575 		return NULL;
2576 
2577 	evt->evt_type = evt_type;
2578 	INIT_LIST_HEAD(&evt->node);
2579 
2580 	/* evt_type-specific initialization, if any */
2581 	switch (evt_type) {
2582 	case SDEV_EVT_MEDIA_CHANGE:
2583 	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2584 	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2585 	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2586 	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2587 	case SDEV_EVT_LUN_CHANGE_REPORTED:
2588 	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2589 	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2590 	default:
2591 		/* do nothing */
2592 		break;
2593 	}
2594 
2595 	return evt;
2596 }
2597 EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2598 
2599 /**
2600  * 	sdev_evt_send_simple - send asserted event to uevent thread
2601  *	@sdev: scsi_device event occurred on
2602  *	@evt_type: type of event to send
2603  *	@gfpflags: GFP flags for allocation
2604  *
2605  *	Assert scsi device event asynchronously, given an event type.
2606  */
sdev_evt_send_simple(struct scsi_device * sdev,enum scsi_device_event evt_type,gfp_t gfpflags)2607 void sdev_evt_send_simple(struct scsi_device *sdev,
2608 			  enum scsi_device_event evt_type, gfp_t gfpflags)
2609 {
2610 	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2611 	if (!evt) {
2612 		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2613 			    evt_type);
2614 		return;
2615 	}
2616 
2617 	sdev_evt_send(sdev, evt);
2618 }
2619 EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2620 
2621 /**
2622  *	scsi_device_quiesce - Block all commands except power management.
2623  *	@sdev:	scsi device to quiesce.
2624  *
2625  *	This works by trying to transition to the SDEV_QUIESCE state
2626  *	(which must be a legal transition).  When the device is in this
2627  *	state, only power management requests will be accepted, all others will
2628  *	be deferred.
2629  *
2630  *	Must be called with user context, may sleep.
2631  *
2632  *	Returns zero if unsuccessful or an error if not.
2633  */
2634 int
scsi_device_quiesce(struct scsi_device * sdev)2635 scsi_device_quiesce(struct scsi_device *sdev)
2636 {
2637 	struct request_queue *q = sdev->request_queue;
2638 	int err;
2639 
2640 	/*
2641 	 * It is allowed to call scsi_device_quiesce() multiple times from
2642 	 * the same context but concurrent scsi_device_quiesce() calls are
2643 	 * not allowed.
2644 	 */
2645 	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2646 
2647 	if (sdev->quiesced_by == current)
2648 		return 0;
2649 
2650 	blk_set_pm_only(q);
2651 
2652 	blk_mq_freeze_queue(q);
2653 	/*
2654 	 * Ensure that the effect of blk_set_pm_only() will be visible
2655 	 * for percpu_ref_tryget() callers that occur after the queue
2656 	 * unfreeze even if the queue was already frozen before this function
2657 	 * was called. See also https://lwn.net/Articles/573497/.
2658 	 */
2659 	synchronize_rcu();
2660 	blk_mq_unfreeze_queue(q);
2661 
2662 	mutex_lock(&sdev->state_mutex);
2663 	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2664 	if (err == 0)
2665 		sdev->quiesced_by = current;
2666 	else
2667 		blk_clear_pm_only(q);
2668 	mutex_unlock(&sdev->state_mutex);
2669 
2670 	return err;
2671 }
2672 EXPORT_SYMBOL(scsi_device_quiesce);
2673 
2674 /**
2675  *	scsi_device_resume - Restart user issued commands to a quiesced device.
2676  *	@sdev:	scsi device to resume.
2677  *
2678  *	Moves the device from quiesced back to running and restarts the
2679  *	queues.
2680  *
2681  *	Must be called with user context, may sleep.
2682  */
scsi_device_resume(struct scsi_device * sdev)2683 void scsi_device_resume(struct scsi_device *sdev)
2684 {
2685 	/* check if the device state was mutated prior to resume, and if
2686 	 * so assume the state is being managed elsewhere (for example
2687 	 * device deleted during suspend)
2688 	 */
2689 	mutex_lock(&sdev->state_mutex);
2690 	if (sdev->sdev_state == SDEV_QUIESCE)
2691 		scsi_device_set_state(sdev, SDEV_RUNNING);
2692 	if (sdev->quiesced_by) {
2693 		sdev->quiesced_by = NULL;
2694 		blk_clear_pm_only(sdev->request_queue);
2695 	}
2696 	mutex_unlock(&sdev->state_mutex);
2697 }
2698 EXPORT_SYMBOL(scsi_device_resume);
2699 
2700 static void
device_quiesce_fn(struct scsi_device * sdev,void * data)2701 device_quiesce_fn(struct scsi_device *sdev, void *data)
2702 {
2703 	scsi_device_quiesce(sdev);
2704 }
2705 
2706 void
scsi_target_quiesce(struct scsi_target * starget)2707 scsi_target_quiesce(struct scsi_target *starget)
2708 {
2709 	starget_for_each_device(starget, NULL, device_quiesce_fn);
2710 }
2711 EXPORT_SYMBOL(scsi_target_quiesce);
2712 
2713 static void
device_resume_fn(struct scsi_device * sdev,void * data)2714 device_resume_fn(struct scsi_device *sdev, void *data)
2715 {
2716 	scsi_device_resume(sdev);
2717 }
2718 
2719 void
scsi_target_resume(struct scsi_target * starget)2720 scsi_target_resume(struct scsi_target *starget)
2721 {
2722 	starget_for_each_device(starget, NULL, device_resume_fn);
2723 }
2724 EXPORT_SYMBOL(scsi_target_resume);
2725 
__scsi_internal_device_block_nowait(struct scsi_device * sdev)2726 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2727 {
2728 	if (scsi_device_set_state(sdev, SDEV_BLOCK))
2729 		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2730 
2731 	return 0;
2732 }
2733 
scsi_start_queue(struct scsi_device * sdev)2734 void scsi_start_queue(struct scsi_device *sdev)
2735 {
2736 	if (cmpxchg(&sdev->queue_stopped, 1, 0))
2737 		blk_mq_unquiesce_queue(sdev->request_queue);
2738 }
2739 
scsi_stop_queue(struct scsi_device * sdev)2740 static void scsi_stop_queue(struct scsi_device *sdev)
2741 {
2742 	/*
2743 	 * The atomic variable of ->queue_stopped covers that
2744 	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2745 	 *
2746 	 * The caller needs to wait until quiesce is done.
2747 	 */
2748 	if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2749 		blk_mq_quiesce_queue_nowait(sdev->request_queue);
2750 }
2751 
2752 /**
2753  * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2754  * @sdev: device to block
2755  *
2756  * Pause SCSI command processing on the specified device. Does not sleep.
2757  *
2758  * Returns zero if successful or a negative error code upon failure.
2759  *
2760  * Notes:
2761  * This routine transitions the device to the SDEV_BLOCK state (which must be
2762  * a legal transition). When the device is in this state, command processing
2763  * is paused until the device leaves the SDEV_BLOCK state. See also
2764  * scsi_internal_device_unblock_nowait().
2765  */
scsi_internal_device_block_nowait(struct scsi_device * sdev)2766 int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2767 {
2768 	int ret = __scsi_internal_device_block_nowait(sdev);
2769 
2770 	/*
2771 	 * The device has transitioned to SDEV_BLOCK.  Stop the
2772 	 * block layer from calling the midlayer with this device's
2773 	 * request queue.
2774 	 */
2775 	if (!ret)
2776 		scsi_stop_queue(sdev);
2777 	return ret;
2778 }
2779 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2780 
2781 /**
2782  * scsi_device_block - try to transition to the SDEV_BLOCK state
2783  * @sdev: device to block
2784  * @data: dummy argument, ignored
2785  *
2786  * Pause SCSI command processing on the specified device. Callers must wait
2787  * until all ongoing scsi_queue_rq() calls have finished after this function
2788  * returns.
2789  *
2790  * Note:
2791  * This routine transitions the device to the SDEV_BLOCK state (which must be
2792  * a legal transition). When the device is in this state, command processing
2793  * is paused until the device leaves the SDEV_BLOCK state. See also
2794  * scsi_internal_device_unblock().
2795  */
scsi_device_block(struct scsi_device * sdev,void * data)2796 static void scsi_device_block(struct scsi_device *sdev, void *data)
2797 {
2798 	int err;
2799 	enum scsi_device_state state;
2800 
2801 	mutex_lock(&sdev->state_mutex);
2802 	err = __scsi_internal_device_block_nowait(sdev);
2803 	state = sdev->sdev_state;
2804 	if (err == 0)
2805 		/*
2806 		 * scsi_stop_queue() must be called with the state_mutex
2807 		 * held. Otherwise a simultaneous scsi_start_queue() call
2808 		 * might unquiesce the queue before we quiesce it.
2809 		 */
2810 		scsi_stop_queue(sdev);
2811 
2812 	mutex_unlock(&sdev->state_mutex);
2813 
2814 	WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2815 		  __func__, dev_name(&sdev->sdev_gendev), state);
2816 }
2817 
2818 /**
2819  * scsi_internal_device_unblock_nowait - resume a device after a block request
2820  * @sdev:	device to resume
2821  * @new_state:	state to set the device to after unblocking
2822  *
2823  * Restart the device queue for a previously suspended SCSI device. Does not
2824  * sleep.
2825  *
2826  * Returns zero if successful or a negative error code upon failure.
2827  *
2828  * Notes:
2829  * This routine transitions the device to the SDEV_RUNNING state or to one of
2830  * the offline states (which must be a legal transition) allowing the midlayer
2831  * to goose the queue for this device.
2832  */
scsi_internal_device_unblock_nowait(struct scsi_device * sdev,enum scsi_device_state new_state)2833 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2834 					enum scsi_device_state new_state)
2835 {
2836 	switch (new_state) {
2837 	case SDEV_RUNNING:
2838 	case SDEV_TRANSPORT_OFFLINE:
2839 		break;
2840 	default:
2841 		return -EINVAL;
2842 	}
2843 
2844 	/*
2845 	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2846 	 * offlined states and goose the device queue if successful.
2847 	 */
2848 	switch (sdev->sdev_state) {
2849 	case SDEV_BLOCK:
2850 	case SDEV_TRANSPORT_OFFLINE:
2851 		sdev->sdev_state = new_state;
2852 		break;
2853 	case SDEV_CREATED_BLOCK:
2854 		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2855 		    new_state == SDEV_OFFLINE)
2856 			sdev->sdev_state = new_state;
2857 		else
2858 			sdev->sdev_state = SDEV_CREATED;
2859 		break;
2860 	case SDEV_CANCEL:
2861 	case SDEV_OFFLINE:
2862 		break;
2863 	default:
2864 		return -EINVAL;
2865 	}
2866 	scsi_start_queue(sdev);
2867 
2868 	return 0;
2869 }
2870 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2871 
2872 /**
2873  * scsi_internal_device_unblock - resume a device after a block request
2874  * @sdev:	device to resume
2875  * @new_state:	state to set the device to after unblocking
2876  *
2877  * Restart the device queue for a previously suspended SCSI device. May sleep.
2878  *
2879  * Returns zero if successful or a negative error code upon failure.
2880  *
2881  * Notes:
2882  * This routine transitions the device to the SDEV_RUNNING state or to one of
2883  * the offline states (which must be a legal transition) allowing the midlayer
2884  * to goose the queue for this device.
2885  */
scsi_internal_device_unblock(struct scsi_device * sdev,enum scsi_device_state new_state)2886 static int scsi_internal_device_unblock(struct scsi_device *sdev,
2887 					enum scsi_device_state new_state)
2888 {
2889 	int ret;
2890 
2891 	mutex_lock(&sdev->state_mutex);
2892 	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2893 	mutex_unlock(&sdev->state_mutex);
2894 
2895 	return ret;
2896 }
2897 
2898 static int
target_block(struct device * dev,void * data)2899 target_block(struct device *dev, void *data)
2900 {
2901 	if (scsi_is_target_device(dev))
2902 		starget_for_each_device(to_scsi_target(dev), NULL,
2903 					scsi_device_block);
2904 	return 0;
2905 }
2906 
2907 /**
2908  * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
2909  * @dev: a parent device of one or more scsi_target devices
2910  * @shost: the Scsi_Host to which this device belongs
2911  *
2912  * Iterate over all children of @dev, which should be scsi_target devices,
2913  * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
2914  * ongoing scsi_queue_rq() calls to finish. May sleep.
2915  *
2916  * Note:
2917  * @dev must not itself be a scsi_target device.
2918  */
2919 void
scsi_block_targets(struct Scsi_Host * shost,struct device * dev)2920 scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
2921 {
2922 	WARN_ON_ONCE(scsi_is_target_device(dev));
2923 	device_for_each_child(dev, NULL, target_block);
2924 	blk_mq_wait_quiesce_done(&shost->tag_set);
2925 }
2926 EXPORT_SYMBOL_GPL(scsi_block_targets);
2927 
2928 static void
device_unblock(struct scsi_device * sdev,void * data)2929 device_unblock(struct scsi_device *sdev, void *data)
2930 {
2931 	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2932 }
2933 
2934 static int
target_unblock(struct device * dev,void * data)2935 target_unblock(struct device *dev, void *data)
2936 {
2937 	if (scsi_is_target_device(dev))
2938 		starget_for_each_device(to_scsi_target(dev), data,
2939 					device_unblock);
2940 	return 0;
2941 }
2942 
2943 void
scsi_target_unblock(struct device * dev,enum scsi_device_state new_state)2944 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2945 {
2946 	if (scsi_is_target_device(dev))
2947 		starget_for_each_device(to_scsi_target(dev), &new_state,
2948 					device_unblock);
2949 	else
2950 		device_for_each_child(dev, &new_state, target_unblock);
2951 }
2952 EXPORT_SYMBOL_GPL(scsi_target_unblock);
2953 
2954 /**
2955  * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
2956  * @shost: device to block
2957  *
2958  * Pause SCSI command processing for all logical units associated with the SCSI
2959  * host and wait until pending scsi_queue_rq() calls have finished.
2960  *
2961  * Returns zero if successful or a negative error code upon failure.
2962  */
2963 int
scsi_host_block(struct Scsi_Host * shost)2964 scsi_host_block(struct Scsi_Host *shost)
2965 {
2966 	struct scsi_device *sdev;
2967 	int ret;
2968 
2969 	/*
2970 	 * Call scsi_internal_device_block_nowait so we can avoid
2971 	 * calling synchronize_rcu() for each LUN.
2972 	 */
2973 	shost_for_each_device(sdev, shost) {
2974 		mutex_lock(&sdev->state_mutex);
2975 		ret = scsi_internal_device_block_nowait(sdev);
2976 		mutex_unlock(&sdev->state_mutex);
2977 		if (ret) {
2978 			scsi_device_put(sdev);
2979 			return ret;
2980 		}
2981 	}
2982 
2983 	/* Wait for ongoing scsi_queue_rq() calls to finish. */
2984 	blk_mq_wait_quiesce_done(&shost->tag_set);
2985 
2986 	return 0;
2987 }
2988 EXPORT_SYMBOL_GPL(scsi_host_block);
2989 
2990 int
scsi_host_unblock(struct Scsi_Host * shost,int new_state)2991 scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2992 {
2993 	struct scsi_device *sdev;
2994 	int ret = 0;
2995 
2996 	shost_for_each_device(sdev, shost) {
2997 		ret = scsi_internal_device_unblock(sdev, new_state);
2998 		if (ret) {
2999 			scsi_device_put(sdev);
3000 			break;
3001 		}
3002 	}
3003 	return ret;
3004 }
3005 EXPORT_SYMBOL_GPL(scsi_host_unblock);
3006 
3007 /**
3008  * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3009  * @sgl:	scatter-gather list
3010  * @sg_count:	number of segments in sg
3011  * @offset:	offset in bytes into sg, on return offset into the mapped area
3012  * @len:	bytes to map, on return number of bytes mapped
3013  *
3014  * Returns virtual address of the start of the mapped page
3015  */
scsi_kmap_atomic_sg(struct scatterlist * sgl,int sg_count,size_t * offset,size_t * len)3016 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3017 			  size_t *offset, size_t *len)
3018 {
3019 	int i;
3020 	size_t sg_len = 0, len_complete = 0;
3021 	struct scatterlist *sg;
3022 	struct page *page;
3023 
3024 	WARN_ON(!irqs_disabled());
3025 
3026 	for_each_sg(sgl, sg, sg_count, i) {
3027 		len_complete = sg_len; /* Complete sg-entries */
3028 		sg_len += sg->length;
3029 		if (sg_len > *offset)
3030 			break;
3031 	}
3032 
3033 	if (unlikely(i == sg_count)) {
3034 		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3035 			"elements %d\n",
3036 		       __func__, sg_len, *offset, sg_count);
3037 		WARN_ON(1);
3038 		return NULL;
3039 	}
3040 
3041 	/* Offset starting from the beginning of first page in this sg-entry */
3042 	*offset = *offset - len_complete + sg->offset;
3043 
3044 	/* Assumption: contiguous pages can be accessed as "page + i" */
3045 	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3046 	*offset &= ~PAGE_MASK;
3047 
3048 	/* Bytes in this sg-entry from *offset to the end of the page */
3049 	sg_len = PAGE_SIZE - *offset;
3050 	if (*len > sg_len)
3051 		*len = sg_len;
3052 
3053 	return kmap_atomic(page);
3054 }
3055 EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3056 
3057 /**
3058  * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3059  * @virt:	virtual address to be unmapped
3060  */
scsi_kunmap_atomic_sg(void * virt)3061 void scsi_kunmap_atomic_sg(void *virt)
3062 {
3063 	kunmap_atomic(virt);
3064 }
3065 EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3066 
sdev_disable_disk_events(struct scsi_device * sdev)3067 void sdev_disable_disk_events(struct scsi_device *sdev)
3068 {
3069 	atomic_inc(&sdev->disk_events_disable_depth);
3070 }
3071 EXPORT_SYMBOL(sdev_disable_disk_events);
3072 
sdev_enable_disk_events(struct scsi_device * sdev)3073 void sdev_enable_disk_events(struct scsi_device *sdev)
3074 {
3075 	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3076 		return;
3077 	atomic_dec(&sdev->disk_events_disable_depth);
3078 }
3079 EXPORT_SYMBOL(sdev_enable_disk_events);
3080 
designator_prio(const unsigned char * d)3081 static unsigned char designator_prio(const unsigned char *d)
3082 {
3083 	if (d[1] & 0x30)
3084 		/* not associated with LUN */
3085 		return 0;
3086 
3087 	if (d[3] == 0)
3088 		/* invalid length */
3089 		return 0;
3090 
3091 	/*
3092 	 * Order of preference for lun descriptor:
3093 	 * - SCSI name string
3094 	 * - NAA IEEE Registered Extended
3095 	 * - EUI-64 based 16-byte
3096 	 * - EUI-64 based 12-byte
3097 	 * - NAA IEEE Registered
3098 	 * - NAA IEEE Extended
3099 	 * - EUI-64 based 8-byte
3100 	 * - SCSI name string (truncated)
3101 	 * - T10 Vendor ID
3102 	 * as longer descriptors reduce the likelyhood
3103 	 * of identification clashes.
3104 	 */
3105 
3106 	switch (d[1] & 0xf) {
3107 	case 8:
3108 		/* SCSI name string, variable-length UTF-8 */
3109 		return 9;
3110 	case 3:
3111 		switch (d[4] >> 4) {
3112 		case 6:
3113 			/* NAA registered extended */
3114 			return 8;
3115 		case 5:
3116 			/* NAA registered */
3117 			return 5;
3118 		case 4:
3119 			/* NAA extended */
3120 			return 4;
3121 		case 3:
3122 			/* NAA locally assigned */
3123 			return 1;
3124 		default:
3125 			break;
3126 		}
3127 		break;
3128 	case 2:
3129 		switch (d[3]) {
3130 		case 16:
3131 			/* EUI64-based, 16 byte */
3132 			return 7;
3133 		case 12:
3134 			/* EUI64-based, 12 byte */
3135 			return 6;
3136 		case 8:
3137 			/* EUI64-based, 8 byte */
3138 			return 3;
3139 		default:
3140 			break;
3141 		}
3142 		break;
3143 	case 1:
3144 		/* T10 vendor ID */
3145 		return 1;
3146 	default:
3147 		break;
3148 	}
3149 
3150 	return 0;
3151 }
3152 
3153 /**
3154  * scsi_vpd_lun_id - return a unique device identification
3155  * @sdev: SCSI device
3156  * @id:   buffer for the identification
3157  * @id_len:  length of the buffer
3158  *
3159  * Copies a unique device identification into @id based
3160  * on the information in the VPD page 0x83 of the device.
3161  * The string will be formatted as a SCSI name string.
3162  *
3163  * Returns the length of the identification or error on failure.
3164  * If the identifier is longer than the supplied buffer the actual
3165  * identifier length is returned and the buffer is not zero-padded.
3166  */
scsi_vpd_lun_id(struct scsi_device * sdev,char * id,size_t id_len)3167 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3168 {
3169 	u8 cur_id_prio = 0;
3170 	u8 cur_id_size = 0;
3171 	const unsigned char *d, *cur_id_str;
3172 	const struct scsi_vpd *vpd_pg83;
3173 	int id_size = -EINVAL;
3174 
3175 	rcu_read_lock();
3176 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3177 	if (!vpd_pg83) {
3178 		rcu_read_unlock();
3179 		return -ENXIO;
3180 	}
3181 
3182 	/* The id string must be at least 20 bytes + terminating NULL byte */
3183 	if (id_len < 21) {
3184 		rcu_read_unlock();
3185 		return -EINVAL;
3186 	}
3187 
3188 	memset(id, 0, id_len);
3189 	for (d = vpd_pg83->data + 4;
3190 	     d < vpd_pg83->data + vpd_pg83->len;
3191 	     d += d[3] + 4) {
3192 		u8 prio = designator_prio(d);
3193 
3194 		if (prio == 0 || cur_id_prio > prio)
3195 			continue;
3196 
3197 		switch (d[1] & 0xf) {
3198 		case 0x1:
3199 			/* T10 Vendor ID */
3200 			if (cur_id_size > d[3])
3201 				break;
3202 			cur_id_prio = prio;
3203 			cur_id_size = d[3];
3204 			if (cur_id_size + 4 > id_len)
3205 				cur_id_size = id_len - 4;
3206 			cur_id_str = d + 4;
3207 			id_size = snprintf(id, id_len, "t10.%*pE",
3208 					   cur_id_size, cur_id_str);
3209 			break;
3210 		case 0x2:
3211 			/* EUI-64 */
3212 			cur_id_prio = prio;
3213 			cur_id_size = d[3];
3214 			cur_id_str = d + 4;
3215 			switch (cur_id_size) {
3216 			case 8:
3217 				id_size = snprintf(id, id_len,
3218 						   "eui.%8phN",
3219 						   cur_id_str);
3220 				break;
3221 			case 12:
3222 				id_size = snprintf(id, id_len,
3223 						   "eui.%12phN",
3224 						   cur_id_str);
3225 				break;
3226 			case 16:
3227 				id_size = snprintf(id, id_len,
3228 						   "eui.%16phN",
3229 						   cur_id_str);
3230 				break;
3231 			default:
3232 				break;
3233 			}
3234 			break;
3235 		case 0x3:
3236 			/* NAA */
3237 			cur_id_prio = prio;
3238 			cur_id_size = d[3];
3239 			cur_id_str = d + 4;
3240 			switch (cur_id_size) {
3241 			case 8:
3242 				id_size = snprintf(id, id_len,
3243 						   "naa.%8phN",
3244 						   cur_id_str);
3245 				break;
3246 			case 16:
3247 				id_size = snprintf(id, id_len,
3248 						   "naa.%16phN",
3249 						   cur_id_str);
3250 				break;
3251 			default:
3252 				break;
3253 			}
3254 			break;
3255 		case 0x8:
3256 			/* SCSI name string */
3257 			if (cur_id_size > d[3])
3258 				break;
3259 			/* Prefer others for truncated descriptor */
3260 			if (d[3] > id_len) {
3261 				prio = 2;
3262 				if (cur_id_prio > prio)
3263 					break;
3264 			}
3265 			cur_id_prio = prio;
3266 			cur_id_size = id_size = d[3];
3267 			cur_id_str = d + 4;
3268 			if (cur_id_size >= id_len)
3269 				cur_id_size = id_len - 1;
3270 			memcpy(id, cur_id_str, cur_id_size);
3271 			break;
3272 		default:
3273 			break;
3274 		}
3275 	}
3276 	rcu_read_unlock();
3277 
3278 	return id_size;
3279 }
3280 EXPORT_SYMBOL(scsi_vpd_lun_id);
3281 
3282 /*
3283  * scsi_vpd_tpg_id - return a target port group identifier
3284  * @sdev: SCSI device
3285  *
3286  * Returns the Target Port Group identifier from the information
3287  * froom VPD page 0x83 of the device.
3288  *
3289  * Returns the identifier or error on failure.
3290  */
scsi_vpd_tpg_id(struct scsi_device * sdev,int * rel_id)3291 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3292 {
3293 	const unsigned char *d;
3294 	const struct scsi_vpd *vpd_pg83;
3295 	int group_id = -EAGAIN, rel_port = -1;
3296 
3297 	rcu_read_lock();
3298 	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3299 	if (!vpd_pg83) {
3300 		rcu_read_unlock();
3301 		return -ENXIO;
3302 	}
3303 
3304 	d = vpd_pg83->data + 4;
3305 	while (d < vpd_pg83->data + vpd_pg83->len) {
3306 		switch (d[1] & 0xf) {
3307 		case 0x4:
3308 			/* Relative target port */
3309 			rel_port = get_unaligned_be16(&d[6]);
3310 			break;
3311 		case 0x5:
3312 			/* Target port group */
3313 			group_id = get_unaligned_be16(&d[6]);
3314 			break;
3315 		default:
3316 			break;
3317 		}
3318 		d += d[3] + 4;
3319 	}
3320 	rcu_read_unlock();
3321 
3322 	if (group_id >= 0 && rel_id && rel_port != -1)
3323 		*rel_id = rel_port;
3324 
3325 	return group_id;
3326 }
3327 EXPORT_SYMBOL(scsi_vpd_tpg_id);
3328 
3329 /**
3330  * scsi_build_sense - build sense data for a command
3331  * @scmd:	scsi command for which the sense should be formatted
3332  * @desc:	Sense format (non-zero == descriptor format,
3333  *              0 == fixed format)
3334  * @key:	Sense key
3335  * @asc:	Additional sense code
3336  * @ascq:	Additional sense code qualifier
3337  *
3338  **/
scsi_build_sense(struct scsi_cmnd * scmd,int desc,u8 key,u8 asc,u8 ascq)3339 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3340 {
3341 	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3342 	scmd->result = SAM_STAT_CHECK_CONDITION;
3343 }
3344 EXPORT_SYMBOL_GPL(scsi_build_sense);
3345