brintos

brintos / linux-shallow public Read only

0
0
Text · 89.7 KiB · adee6f6 Raw
3435 lines · c
1// SPDX-License-Identifier: GPL-2.0-only2/*3 * Copyright (C) 1999 Eric Youngdale4 * Copyright (C) 2014 Christoph Hellwig5 *6 *  SCSI queueing library.7 *      Initial versions: Eric Youngdale (eric@andante.org).8 *                        Based upon conversations with large numbers9 *                        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 <linux/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_limits() */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 should46 * cover normal cases.47 */48#ifdef CONFIG_ARCH_NO_SG_CHAIN49#define  SCSI_INLINE_PROT_SG_CNT  050#define  SCSI_INLINE_SG_CNT  051#else52#define  SCSI_INLINE_PROT_SG_CNT  153#define  SCSI_INLINE_SG_CNT  254#endif55 56static struct kmem_cache *scsi_sense_cache;57static DEFINE_MUTEX(scsi_sense_cache_mutex);58 59static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);60 61int 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 78static void79scsi_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 actually89	 * completed, and that we should be able to queue a command now.90	 *91	 * Note that the prior mid-layer assumption that any host could92	 * always queue at least one command is now broken.  The mid-layer93	 * will implement a user specifiable stall (see94	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)95	 * if a command is requeued with no other commands outstanding96	 * 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 114static 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 insertion132 * @cmd: The SCSI command being requeued133 * @reason:  The reason for the requeue134 * @unbusy: Whether the queue should be unbusied135 *136 * This is a private queue insertion.  The public interface137 * scsi_queue_insert() always assumes the queue should be unbusied138 * because it's always called before the completion.  This function is139 * for a requeue after completion, which should only occur in this140 * file.141 */142static 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 longer153	 * 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 commands160	 * that are already in the queue. Schedule requeue work under161	 * lock such that the kblockd_schedule_work() call happens162	 * 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 accept176 * any more commands for the time being, or the device returned QUEUE_FULL and177 * can accept no more commands.178 *179 * Context: This could be called either from an interrupt context or a normal180 * process context.181 */182void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)183{184	__scsi_queue_insert(cmd, reason, true);185}186 187void scsi_failures_reset_retries(struct scsi_failures *failures)188{189	struct scsi_failure *failure;190 191	failures->total_retries = 0;192 193	for (failure = failures->failure_definitions; failure->result;194	     failure++)195		failure->retries = 0;196}197EXPORT_SYMBOL_GPL(scsi_failures_reset_retries);198 199/**200 * scsi_check_passthrough - Determine if passthrough scsi_cmnd needs a retry.201 * @scmd: scsi_cmnd to check.202 * @failures: scsi_failures struct that lists failures to check for.203 *204 * Returns -EAGAIN if the caller should retry else 0.205 */206static int scsi_check_passthrough(struct scsi_cmnd *scmd,207				  struct scsi_failures *failures)208{209	struct scsi_failure *failure;210	struct scsi_sense_hdr sshdr;211	enum sam_status status;212 213	if (!failures)214		return 0;215 216	for (failure = failures->failure_definitions; failure->result;217	     failure++) {218		if (failure->result == SCMD_FAILURE_RESULT_ANY)219			goto maybe_retry;220 221		if (host_byte(scmd->result) &&222		    host_byte(scmd->result) == host_byte(failure->result))223			goto maybe_retry;224 225		status = status_byte(scmd->result);226		if (!status)227			continue;228 229		if (failure->result == SCMD_FAILURE_STAT_ANY &&230		    !scsi_status_is_good(scmd->result))231			goto maybe_retry;232 233		if (status != status_byte(failure->result))234			continue;235 236		if (status_byte(failure->result) != SAM_STAT_CHECK_CONDITION ||237		    failure->sense == SCMD_FAILURE_SENSE_ANY)238			goto maybe_retry;239 240		if (!scsi_command_normalize_sense(scmd, &sshdr))241			return 0;242 243		if (failure->sense != sshdr.sense_key)244			continue;245 246		if (failure->asc == SCMD_FAILURE_ASC_ANY)247			goto maybe_retry;248 249		if (failure->asc != sshdr.asc)250			continue;251 252		if (failure->ascq == SCMD_FAILURE_ASCQ_ANY ||253		    failure->ascq == sshdr.ascq)254			goto maybe_retry;255	}256 257	return 0;258 259maybe_retry:260	if (failure->allowed) {261		if (failure->allowed == SCMD_FAILURE_NO_LIMIT ||262		    ++failure->retries <= failure->allowed)263			return -EAGAIN;264	} else {265		if (failures->total_allowed == SCMD_FAILURE_NO_LIMIT ||266		    ++failures->total_retries <= failures->total_allowed)267			return -EAGAIN;268	}269 270	return 0;271}272 273/**274 * scsi_execute_cmd - insert request and wait for the result275 * @sdev:	scsi_device276 * @cmd:	scsi command277 * @opf:	block layer request cmd_flags278 * @buffer:	data buffer279 * @bufflen:	len of buffer280 * @timeout:	request timeout in HZ281 * @ml_retries:	number of times SCSI midlayer will retry request282 * @args:	Optional args. See struct definition for field descriptions283 *284 * Returns the scsi_cmnd result field if a command was executed, or a negative285 * Linux error code if we didn't get that far.286 */287int scsi_execute_cmd(struct scsi_device *sdev, const unsigned char *cmd,288		     blk_opf_t opf, void *buffer, unsigned int bufflen,289		     int timeout, int ml_retries,290		     const struct scsi_exec_args *args)291{292	static const struct scsi_exec_args default_args;293	struct request *req;294	struct scsi_cmnd *scmd;295	int ret;296 297	if (!args)298		args = &default_args;299	else if (WARN_ON_ONCE(args->sense &&300			      args->sense_len != SCSI_SENSE_BUFFERSIZE))301		return -EINVAL;302 303retry:304	req = scsi_alloc_request(sdev->request_queue, opf, args->req_flags);305	if (IS_ERR(req))306		return PTR_ERR(req);307 308	if (bufflen) {309		ret = blk_rq_map_kern(sdev->request_queue, req,310				      buffer, bufflen, GFP_NOIO);311		if (ret)312			goto out;313	}314	scmd = blk_mq_rq_to_pdu(req);315	scmd->cmd_len = COMMAND_SIZE(cmd[0]);316	memcpy(scmd->cmnd, cmd, scmd->cmd_len);317	scmd->allowed = ml_retries;318	scmd->flags |= args->scmd_flags;319	req->timeout = timeout;320	req->rq_flags |= RQF_QUIET;321 322	/*323	 * head injection *required* here otherwise quiesce won't work324	 */325	blk_execute_rq(req, true);326 327	if (scsi_check_passthrough(scmd, args->failures) == -EAGAIN) {328		blk_mq_free_request(req);329		goto retry;330	}331 332	/*333	 * Some devices (USB mass-storage in particular) may transfer334	 * garbage data together with a residue indicating that the data335	 * is invalid.  Prevent the garbage from being misinterpreted336	 * and prevent security leaks by zeroing out the excess data.337	 */338	if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen))339		memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len);340 341	if (args->resid)342		*args->resid = scmd->resid_len;343	if (args->sense)344		memcpy(args->sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE);345	if (args->sshdr)346		scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len,347				     args->sshdr);348 349	ret = scmd->result;350 out:351	blk_mq_free_request(req);352 353	return ret;354}355EXPORT_SYMBOL(scsi_execute_cmd);356 357/*358 * Wake up the error handler if necessary. Avoid as follows that the error359 * handler is not woken up if host in-flight requests number ==360 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination361 * with an RCU read lock in this function to ensure that this function in362 * its entirety either finishes before scsi_eh_scmd_add() increases the363 * host_failed counter or that it notices the shost state change made by364 * scsi_eh_scmd_add().365 */366static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)367{368	unsigned long flags;369 370	rcu_read_lock();371	__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);372	if (unlikely(scsi_host_in_recovery(shost))) {373		unsigned int busy = scsi_host_busy(shost);374 375		spin_lock_irqsave(shost->host_lock, flags);376		if (shost->host_failed || shost->host_eh_scheduled)377			scsi_eh_wakeup(shost, busy);378		spin_unlock_irqrestore(shost->host_lock, flags);379	}380	rcu_read_unlock();381}382 383void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)384{385	struct Scsi_Host *shost = sdev->host;386	struct scsi_target *starget = scsi_target(sdev);387 388	scsi_dec_host_busy(shost, cmd);389 390	if (starget->can_queue > 0)391		atomic_dec(&starget->target_busy);392 393	sbitmap_put(&sdev->budget_map, cmd->budget_token);394	cmd->budget_token = -1;395}396 397/*398 * Kick the queue of SCSI device @sdev if @sdev != current_sdev. Called with399 * interrupts disabled.400 */401static void scsi_kick_sdev_queue(struct scsi_device *sdev, void *data)402{403	struct scsi_device *current_sdev = data;404 405	if (sdev != current_sdev)406		blk_mq_run_hw_queues(sdev->request_queue, true);407}408 409/*410 * Called for single_lun devices on IO completion. Clear starget_sdev_user,411 * and call blk_run_queue for all the scsi_devices on the target -412 * including current_sdev first.413 *414 * Called with *no* scsi locks held.415 */416static void scsi_single_lun_run(struct scsi_device *current_sdev)417{418	struct Scsi_Host *shost = current_sdev->host;419	struct scsi_target *starget = scsi_target(current_sdev);420	unsigned long flags;421 422	spin_lock_irqsave(shost->host_lock, flags);423	starget->starget_sdev_user = NULL;424	spin_unlock_irqrestore(shost->host_lock, flags);425 426	/*427	 * Call blk_run_queue for all LUNs on the target, starting with428	 * current_sdev. We race with others (to set starget_sdev_user),429	 * but in most cases, we will be first. Ideally, each LU on the430	 * target would get some limited time or requests on the target.431	 */432	blk_mq_run_hw_queues(current_sdev->request_queue,433			     shost->queuecommand_may_block);434 435	spin_lock_irqsave(shost->host_lock, flags);436	if (!starget->starget_sdev_user)437		__starget_for_each_device(starget, current_sdev,438					  scsi_kick_sdev_queue);439	spin_unlock_irqrestore(shost->host_lock, flags);440}441 442static inline bool scsi_device_is_busy(struct scsi_device *sdev)443{444	if (scsi_device_busy(sdev) >= sdev->queue_depth)445		return true;446	if (atomic_read(&sdev->device_blocked) > 0)447		return true;448	return false;449}450 451static inline bool scsi_target_is_busy(struct scsi_target *starget)452{453	if (starget->can_queue > 0) {454		if (atomic_read(&starget->target_busy) >= starget->can_queue)455			return true;456		if (atomic_read(&starget->target_blocked) > 0)457			return true;458	}459	return false;460}461 462static inline bool scsi_host_is_busy(struct Scsi_Host *shost)463{464	if (atomic_read(&shost->host_blocked) > 0)465		return true;466	if (shost->host_self_blocked)467		return true;468	return false;469}470 471static void scsi_starved_list_run(struct Scsi_Host *shost)472{473	LIST_HEAD(starved_list);474	struct scsi_device *sdev;475	unsigned long flags;476 477	spin_lock_irqsave(shost->host_lock, flags);478	list_splice_init(&shost->starved_list, &starved_list);479 480	while (!list_empty(&starved_list)) {481		struct request_queue *slq;482 483		/*484		 * As long as shost is accepting commands and we have485		 * starved queues, call blk_run_queue. scsi_request_fn486		 * drops the queue_lock and can add us back to the487		 * starved_list.488		 *489		 * host_lock protects the starved_list and starved_entry.490		 * scsi_request_fn must get the host_lock before checking491		 * or modifying starved_list or starved_entry.492		 */493		if (scsi_host_is_busy(shost))494			break;495 496		sdev = list_entry(starved_list.next,497				  struct scsi_device, starved_entry);498		list_del_init(&sdev->starved_entry);499		if (scsi_target_is_busy(scsi_target(sdev))) {500			list_move_tail(&sdev->starved_entry,501				       &shost->starved_list);502			continue;503		}504 505		/*506		 * Once we drop the host lock, a racing scsi_remove_device()507		 * call may remove the sdev from the starved list and destroy508		 * it and the queue.  Mitigate by taking a reference to the509		 * queue and never touching the sdev again after we drop the510		 * host lock.  Note: if __scsi_remove_device() invokes511		 * blk_mq_destroy_queue() before the queue is run from this512		 * function then blk_run_queue() will return immediately since513		 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING.514		 */515		slq = sdev->request_queue;516		if (!blk_get_queue(slq))517			continue;518		spin_unlock_irqrestore(shost->host_lock, flags);519 520		blk_mq_run_hw_queues(slq, false);521		blk_put_queue(slq);522 523		spin_lock_irqsave(shost->host_lock, flags);524	}525	/* put any unprocessed entries back */526	list_splice(&starved_list, &shost->starved_list);527	spin_unlock_irqrestore(shost->host_lock, flags);528}529 530/**531 * scsi_run_queue - Select a proper request queue to serve next.532 * @q:  last request's queue533 *534 * The previous command was completely finished, start a new one if possible.535 */536static void scsi_run_queue(struct request_queue *q)537{538	struct scsi_device *sdev = q->queuedata;539 540	if (scsi_target(sdev)->single_lun)541		scsi_single_lun_run(sdev);542	if (!list_empty(&sdev->host->starved_list))543		scsi_starved_list_run(sdev->host);544 545	/* Note: blk_mq_kick_requeue_list() runs the queue asynchronously. */546	blk_mq_kick_requeue_list(q);547}548 549void scsi_requeue_run_queue(struct work_struct *work)550{551	struct scsi_device *sdev;552	struct request_queue *q;553 554	sdev = container_of(work, struct scsi_device, requeue_work);555	q = sdev->request_queue;556	scsi_run_queue(q);557}558 559void scsi_run_host_queues(struct Scsi_Host *shost)560{561	struct scsi_device *sdev;562 563	shost_for_each_device(sdev, shost)564		scsi_run_queue(sdev->request_queue);565}566 567static void scsi_uninit_cmd(struct scsi_cmnd *cmd)568{569	if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) {570		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);571 572		if (drv->uninit_command)573			drv->uninit_command(cmd);574	}575}576 577void scsi_free_sgtables(struct scsi_cmnd *cmd)578{579	if (cmd->sdb.table.nents)580		sg_free_table_chained(&cmd->sdb.table,581				SCSI_INLINE_SG_CNT);582	if (scsi_prot_sg_count(cmd))583		sg_free_table_chained(&cmd->prot_sdb->table,584				SCSI_INLINE_PROT_SG_CNT);585}586EXPORT_SYMBOL_GPL(scsi_free_sgtables);587 588static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)589{590	scsi_free_sgtables(cmd);591	scsi_uninit_cmd(cmd);592}593 594static void scsi_run_queue_async(struct scsi_device *sdev)595{596	if (scsi_host_in_recovery(sdev->host))597		return;598 599	if (scsi_target(sdev)->single_lun ||600	    !list_empty(&sdev->host->starved_list)) {601		kblockd_schedule_work(&sdev->requeue_work);602	} else {603		/*604		 * smp_mb() present in sbitmap_queue_clear() or implied in605		 * .end_io is for ordering writing .device_busy in606		 * scsi_device_unbusy() and reading sdev->restarts.607		 */608		int old = atomic_read(&sdev->restarts);609 610		/*611		 * ->restarts has to be kept as non-zero if new budget612		 *  contention occurs.613		 *614		 *  No need to run queue when either another re-run615		 *  queue wins in updating ->restarts or a new budget616		 *  contention occurs.617		 */618		if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old)619			blk_mq_run_hw_queues(sdev->request_queue, true);620	}621}622 623/* Returns false when no more bytes to process, true if there are more */624static bool scsi_end_request(struct request *req, blk_status_t error,625		unsigned int bytes)626{627	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);628	struct scsi_device *sdev = cmd->device;629	struct request_queue *q = sdev->request_queue;630 631	if (blk_update_request(req, error, bytes))632		return true;633 634	if (q->limits.features & BLK_FEAT_ADD_RANDOM)635		add_disk_randomness(req->q->disk);636 637	WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&638		     !(cmd->flags & SCMD_INITIALIZED));639	cmd->flags = 0;640 641	/*642	 * Calling rcu_barrier() is not necessary here because the643	 * SCSI error handler guarantees that the function called by644	 * call_rcu() has been called before scsi_end_request() is645	 * called.646	 */647	destroy_rcu_head(&cmd->rcu);648 649	/*650	 * In the MQ case the command gets freed by __blk_mq_end_request,651	 * so we have to do all cleanup that depends on it earlier.652	 *653	 * We also can't kick the queues from irq context, so we654	 * will have to defer it to a workqueue.655	 */656	scsi_mq_uninit_cmd(cmd);657 658	/*659	 * queue is still alive, so grab the ref for preventing it660	 * from being cleaned up during running queue.661	 */662	percpu_ref_get(&q->q_usage_counter);663 664	__blk_mq_end_request(req, error);665 666	scsi_run_queue_async(sdev);667 668	percpu_ref_put(&q->q_usage_counter);669	return false;670}671 672/**673 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t674 * @result:	scsi error code675 *676 * Translate a SCSI result code into a blk_status_t value.677 */678static blk_status_t scsi_result_to_blk_status(int result)679{680	/*681	 * Check the scsi-ml byte first in case we converted a host or status682	 * byte.683	 */684	switch (scsi_ml_byte(result)) {685	case SCSIML_STAT_OK:686		break;687	case SCSIML_STAT_RESV_CONFLICT:688		return BLK_STS_RESV_CONFLICT;689	case SCSIML_STAT_NOSPC:690		return BLK_STS_NOSPC;691	case SCSIML_STAT_MED_ERROR:692		return BLK_STS_MEDIUM;693	case SCSIML_STAT_TGT_FAILURE:694		return BLK_STS_TARGET;695	case SCSIML_STAT_DL_TIMEOUT:696		return BLK_STS_DURATION_LIMIT;697	}698 699	switch (host_byte(result)) {700	case DID_OK:701		if (scsi_status_is_good(result))702			return BLK_STS_OK;703		return BLK_STS_IOERR;704	case DID_TRANSPORT_FAILFAST:705	case DID_TRANSPORT_MARGINAL:706		return BLK_STS_TRANSPORT;707	default:708		return BLK_STS_IOERR;709	}710}711 712/**713 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary714 * @rq: request to examine715 *716 * Description:717 *     A request could be merge of IOs which require different failure718 *     handling.  This function determines the number of bytes which719 *     can be failed from the beginning of the request without720 *     crossing into area which need to be retried further.721 *722 * Return:723 *     The number of bytes to fail.724 */725static unsigned int scsi_rq_err_bytes(const struct request *rq)726{727	blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;728	unsigned int bytes = 0;729	struct bio *bio;730 731	if (!(rq->rq_flags & RQF_MIXED_MERGE))732		return blk_rq_bytes(rq);733 734	/*735	 * Currently the only 'mixing' which can happen is between736	 * different fastfail types.  We can safely fail portions737	 * which have all the failfast bits that the first one has -738	 * the ones which are at least as eager to fail as the first739	 * one.740	 */741	for (bio = rq->bio; bio; bio = bio->bi_next) {742		if ((bio->bi_opf & ff) != ff)743			break;744		bytes += bio->bi_iter.bi_size;745	}746 747	/* this could lead to infinite loop */748	BUG_ON(blk_rq_bytes(rq) && !bytes);749	return bytes;750}751 752static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)753{754	struct request *req = scsi_cmd_to_rq(cmd);755	unsigned long wait_for;756 757	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)758		return false;759 760	wait_for = (cmd->allowed + 1) * req->timeout;761	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {762		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",763			    wait_for/HZ);764		return true;765	}766	return false;767}768 769/*770 * When ALUA transition state is returned, reprep the cmd to771 * use the ALUA handler's transition timeout. Delay the reprep772 * 1 sec to avoid aggressive retries of the target in that773 * state.774 */775#define ALUA_TRANSITION_REPREP_DELAY	1000776 777/* Helper for scsi_io_completion() when special action required. */778static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)779{780	struct request *req = scsi_cmd_to_rq(cmd);781	int level = 0;782	enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,783	      ACTION_RETRY, ACTION_DELAYED_RETRY} action;784	struct scsi_sense_hdr sshdr;785	bool sense_valid;786	bool sense_current = true;      /* false implies "deferred sense" */787	blk_status_t blk_stat;788 789	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);790	if (sense_valid)791		sense_current = !scsi_sense_is_deferred(&sshdr);792 793	blk_stat = scsi_result_to_blk_status(result);794 795	if (host_byte(result) == DID_RESET) {796		/* Third party bus reset or reset for error recovery797		 * reasons.  Just retry the command and see what798		 * happens.799		 */800		action = ACTION_RETRY;801	} else if (sense_valid && sense_current) {802		switch (sshdr.sense_key) {803		case UNIT_ATTENTION:804			if (cmd->device->removable) {805				/* Detected disc change.  Set a bit806				 * and quietly refuse further access.807				 */808				cmd->device->changed = 1;809				action = ACTION_FAIL;810			} else {811				/* Must have been a power glitch, or a812				 * bus reset.  Could not have been a813				 * media change, so we just retry the814				 * command and see what happens.815				 */816				action = ACTION_RETRY;817			}818			break;819		case ILLEGAL_REQUEST:820			/* If we had an ILLEGAL REQUEST returned, then821			 * we may have performed an unsupported822			 * command.  The only thing this should be823			 * would be a ten byte read where only a six824			 * byte read was supported.  Also, on a system825			 * where READ CAPACITY failed, we may have826			 * read past the end of the disk.827			 */828			if ((cmd->device->use_10_for_rw &&829			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&830			    (cmd->cmnd[0] == READ_10 ||831			     cmd->cmnd[0] == WRITE_10)) {832				/* This will issue a new 6-byte command. */833				cmd->device->use_10_for_rw = 0;834				action = ACTION_REPREP;835			} else if (sshdr.asc == 0x10) /* DIX */ {836				action = ACTION_FAIL;837				blk_stat = BLK_STS_PROTECTION;838			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */839			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {840				action = ACTION_FAIL;841				blk_stat = BLK_STS_TARGET;842			} else843				action = ACTION_FAIL;844			break;845		case ABORTED_COMMAND:846			action = ACTION_FAIL;847			if (sshdr.asc == 0x10) /* DIF */848				blk_stat = BLK_STS_PROTECTION;849			break;850		case NOT_READY:851			/* If the device is in the process of becoming852			 * ready, or has a temporary blockage, retry.853			 */854			if (sshdr.asc == 0x04) {855				switch (sshdr.ascq) {856				case 0x01: /* becoming ready */857				case 0x04: /* format in progress */858				case 0x05: /* rebuild in progress */859				case 0x06: /* recalculation in progress */860				case 0x07: /* operation in progress */861				case 0x08: /* Long write in progress */862				case 0x09: /* self test in progress */863				case 0x11: /* notify (enable spinup) required */864				case 0x14: /* space allocation in progress */865				case 0x1a: /* start stop unit in progress */866				case 0x1b: /* sanitize in progress */867				case 0x1d: /* configuration in progress */868				case 0x24: /* depopulation in progress */869				case 0x25: /* depopulation restore in progress */870					action = ACTION_DELAYED_RETRY;871					break;872				case 0x0a: /* ALUA state transition */873					action = ACTION_DELAYED_REPREP;874					break;875				default:876					action = ACTION_FAIL;877					break;878				}879			} else880				action = ACTION_FAIL;881			break;882		case VOLUME_OVERFLOW:883			/* See SSC3rXX or current. */884			action = ACTION_FAIL;885			break;886		case DATA_PROTECT:887			action = ACTION_FAIL;888			if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||889			    (sshdr.asc == 0x55 &&890			     (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {891				/* Insufficient zone resources */892				blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;893			}894			break;895		case COMPLETED:896			fallthrough;897		default:898			action = ACTION_FAIL;899			break;900		}901	} else902		action = ACTION_FAIL;903 904	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))905		action = ACTION_FAIL;906 907	switch (action) {908	case ACTION_FAIL:909		/* Give up and fail the remainder of the request */910		if (!(req->rq_flags & RQF_QUIET)) {911			static DEFINE_RATELIMIT_STATE(_rs,912					DEFAULT_RATELIMIT_INTERVAL,913					DEFAULT_RATELIMIT_BURST);914 915			if (unlikely(scsi_logging_level))916				level =917				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,918						    SCSI_LOG_MLCOMPLETE_BITS);919 920			/*921			 * if logging is enabled the failure will be printed922			 * in scsi_log_completion(), so avoid duplicate messages923			 */924			if (!level && __ratelimit(&_rs)) {925				scsi_print_result(cmd, NULL, FAILED);926				if (sense_valid)927					scsi_print_sense(cmd);928				scsi_print_command(cmd);929			}930		}931		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))932			return;933		fallthrough;934	case ACTION_REPREP:935		scsi_mq_requeue_cmd(cmd, 0);936		break;937	case ACTION_DELAYED_REPREP:938		scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);939		break;940	case ACTION_RETRY:941		/* Retry the same command immediately */942		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);943		break;944	case ACTION_DELAYED_RETRY:945		/* Retry the same command after a delay */946		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);947		break;948	}949}950 951/*952 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a953 * new result that may suppress further error checking. Also modifies954 * *blk_statp in some cases.955 */956static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,957					blk_status_t *blk_statp)958{959	bool sense_valid;960	bool sense_current = true;	/* false implies "deferred sense" */961	struct request *req = scsi_cmd_to_rq(cmd);962	struct scsi_sense_hdr sshdr;963 964	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);965	if (sense_valid)966		sense_current = !scsi_sense_is_deferred(&sshdr);967 968	if (blk_rq_is_passthrough(req)) {969		if (sense_valid) {970			/*971			 * SG_IO wants current and deferred errors972			 */973			cmd->sense_len = min(8 + cmd->sense_buffer[7],974					     SCSI_SENSE_BUFFERSIZE);975		}976		if (sense_current)977			*blk_statp = scsi_result_to_blk_status(result);978	} else if (blk_rq_bytes(req) == 0 && sense_current) {979		/*980		 * Flush commands do not transfers any data, and thus cannot use981		 * good_bytes != blk_rq_bytes(req) as the signal for an error.982		 * This sets *blk_statp explicitly for the problem case.983		 */984		*blk_statp = scsi_result_to_blk_status(result);985	}986	/*987	 * Recovered errors need reporting, but they're always treated as988	 * success, so fiddle the result code here.  For passthrough requests989	 * we already took a copy of the original into sreq->result which990	 * is what gets returned to the user991	 */992	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {993		bool do_print = true;994		/*995		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]996		 * skip print since caller wants ATA registers. Only occurs997		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1998		 */999		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))1000			do_print = false;1001		else if (req->rq_flags & RQF_QUIET)1002			do_print = false;1003		if (do_print)1004			scsi_print_sense(cmd);1005		result = 0;1006		/* for passthrough, *blk_statp may be set */1007		*blk_statp = BLK_STS_OK;1008	}1009	/*1010	 * Another corner case: the SCSI status byte is non-zero but 'good'.1011	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when1012	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD1013	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related1014	 * intermediate statuses (both obsolete in SAM-4) as good.1015	 */1016	if ((result & 0xff) && scsi_status_is_good(result)) {1017		result = 0;1018		*blk_statp = BLK_STS_OK;1019	}1020	return result;1021}1022 1023/**1024 * scsi_io_completion - Completion processing for SCSI commands.1025 * @cmd:	command that is finished.1026 * @good_bytes:	number of processed bytes.1027 *1028 * We will finish off the specified number of sectors. If we are done, the1029 * command block will be released and the queue function will be goosed. If we1030 * are not done then we have to figure out what to do next:1031 *1032 *   a) We can call scsi_mq_requeue_cmd().  The request will be1033 *	unprepared and put back on the queue.  Then a new command will1034 *	be created for it.  This should be used if we made forward1035 *	progress, or if we want to switch from READ(10) to READ(6) for1036 *	example.1037 *1038 *   b) We can call scsi_io_completion_action().  The request will be1039 *	put back on the queue and retried using the same command as1040 *	before, possibly after a delay.1041 *1042 *   c) We can call scsi_end_request() with blk_stat other than1043 *	BLK_STS_OK, to fail the remainder of the request.1044 */1045void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)1046{1047	int result = cmd->result;1048	struct request *req = scsi_cmd_to_rq(cmd);1049	blk_status_t blk_stat = BLK_STS_OK;1050 1051	if (unlikely(result))	/* a nz result may or may not be an error */1052		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);1053 1054	/*1055	 * Next deal with any sectors which we were able to correctly1056	 * handle.1057	 */1058	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,1059		"%u sectors total, %d bytes done.\n",1060		blk_rq_sectors(req), good_bytes));1061 1062	/*1063	 * Failed, zero length commands always need to drop down1064	 * to retry code. Fast path should return in this block.1065	 */1066	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {1067		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))1068			return; /* no bytes remaining */1069	}1070 1071	/* Kill remainder if no retries. */1072	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {1073		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))1074			WARN_ONCE(true,1075			    "Bytes remaining after failed, no-retry command");1076		return;1077	}1078 1079	/*1080	 * If there had been no error, but we have leftover bytes in the1081	 * request just queue the command up again.1082	 */1083	if (likely(result == 0))1084		scsi_mq_requeue_cmd(cmd, 0);1085	else1086		scsi_io_completion_action(cmd, result);1087}1088 1089static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,1090		struct request *rq)1091{1092	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&1093	       !op_is_write(req_op(rq)) &&1094	       sdev->host->hostt->dma_need_drain(rq);1095}1096 1097/**1098 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists1099 * @cmd: SCSI command data structure to initialize.1100 *1101 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled1102 * for @cmd.1103 *1104 * Returns:1105 * * BLK_STS_OK       - on success1106 * * BLK_STS_RESOURCE - if the failure is retryable1107 * * BLK_STS_IOERR    - if the failure is fatal1108 */1109blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)1110{1111	struct scsi_device *sdev = cmd->device;1112	struct request *rq = scsi_cmd_to_rq(cmd);1113	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);1114	struct scatterlist *last_sg = NULL;1115	blk_status_t ret;1116	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);1117	int count;1118 1119	if (WARN_ON_ONCE(!nr_segs))1120		return BLK_STS_IOERR;1121 1122	/*1123	 * Make sure there is space for the drain.  The driver must adjust1124	 * max_hw_segments to be prepared for this.1125	 */1126	if (need_drain)1127		nr_segs++;1128 1129	/*1130	 * If sg table allocation fails, requeue request later.1131	 */1132	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,1133			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))1134		return BLK_STS_RESOURCE;1135 1136	/*1137	 * Next, walk the list, and fill in the addresses and sizes of1138	 * each segment.1139	 */1140	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);1141 1142	if (blk_rq_bytes(rq) & rq->q->limits.dma_pad_mask) {1143		unsigned int pad_len =1144			(rq->q->limits.dma_pad_mask & ~blk_rq_bytes(rq)) + 1;1145 1146		last_sg->length += pad_len;1147		cmd->extra_len += pad_len;1148	}1149 1150	if (need_drain) {1151		sg_unmark_end(last_sg);1152		last_sg = sg_next(last_sg);1153		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);1154		sg_mark_end(last_sg);1155 1156		cmd->extra_len += sdev->dma_drain_len;1157		count++;1158	}1159 1160	BUG_ON(count > cmd->sdb.table.nents);1161	cmd->sdb.table.nents = count;1162	cmd->sdb.length = blk_rq_payload_bytes(rq);1163 1164	if (blk_integrity_rq(rq)) {1165		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;1166 1167		if (WARN_ON_ONCE(!prot_sdb)) {1168			/*1169			 * This can happen if someone (e.g. multipath)1170			 * queues a command to a device on an adapter1171			 * that does not support DIX.1172			 */1173			ret = BLK_STS_IOERR;1174			goto out_free_sgtables;1175		}1176 1177		if (sg_alloc_table_chained(&prot_sdb->table,1178				rq->nr_integrity_segments,1179				prot_sdb->table.sgl,1180				SCSI_INLINE_PROT_SG_CNT)) {1181			ret = BLK_STS_RESOURCE;1182			goto out_free_sgtables;1183		}1184 1185		count = blk_rq_map_integrity_sg(rq, prot_sdb->table.sgl);1186		cmd->prot_sdb = prot_sdb;1187		cmd->prot_sdb->table.nents = count;1188	}1189 1190	return BLK_STS_OK;1191out_free_sgtables:1192	scsi_free_sgtables(cmd);1193	return ret;1194}1195EXPORT_SYMBOL(scsi_alloc_sgtables);1196 1197/**1198 * scsi_initialize_rq - initialize struct scsi_cmnd partially1199 * @rq: Request associated with the SCSI command to be initialized.1200 *1201 * This function initializes the members of struct scsi_cmnd that must be1202 * initialized before request processing starts and that won't be1203 * reinitialized if a SCSI command is requeued.1204 */1205static void scsi_initialize_rq(struct request *rq)1206{1207	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);1208 1209	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));1210	cmd->cmd_len = MAX_COMMAND_SIZE;1211	cmd->sense_len = 0;1212	init_rcu_head(&cmd->rcu);1213	cmd->jiffies_at_alloc = jiffies;1214	cmd->retries = 0;1215}1216 1217struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,1218				   blk_mq_req_flags_t flags)1219{1220	struct request *rq;1221 1222	rq = blk_mq_alloc_request(q, opf, flags);1223	if (!IS_ERR(rq))1224		scsi_initialize_rq(rq);1225	return rq;1226}1227EXPORT_SYMBOL_GPL(scsi_alloc_request);1228 1229/*1230 * Only called when the request isn't completed by SCSI, and not freed by1231 * SCSI1232 */1233static void scsi_cleanup_rq(struct request *rq)1234{1235	if (rq->rq_flags & RQF_DONTPREP) {1236		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));1237		rq->rq_flags &= ~RQF_DONTPREP;1238	}1239}1240 1241/* Called before a request is prepared. See also scsi_mq_prep_fn(). */1242void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)1243{1244	struct request *rq = scsi_cmd_to_rq(cmd);1245 1246	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {1247		cmd->flags |= SCMD_INITIALIZED;1248		scsi_initialize_rq(rq);1249	}1250 1251	cmd->device = dev;1252	INIT_LIST_HEAD(&cmd->eh_entry);1253	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);1254}1255 1256static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,1257		struct request *req)1258{1259	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);1260 1261	/*1262	 * Passthrough requests may transfer data, in which case they must1263	 * a bio attached to them.  Or they might contain a SCSI command1264	 * that does not transfer data, in which case they may optionally1265	 * submit a request without an attached bio.1266	 */1267	if (req->bio) {1268		blk_status_t ret = scsi_alloc_sgtables(cmd);1269		if (unlikely(ret != BLK_STS_OK))1270			return ret;1271	} else {1272		BUG_ON(blk_rq_bytes(req));1273 1274		memset(&cmd->sdb, 0, sizeof(cmd->sdb));1275	}1276 1277	cmd->transfersize = blk_rq_bytes(req);1278	return BLK_STS_OK;1279}1280 1281static blk_status_t1282scsi_device_state_check(struct scsi_device *sdev, struct request *req)1283{1284	switch (sdev->sdev_state) {1285	case SDEV_CREATED:1286		return BLK_STS_OK;1287	case SDEV_OFFLINE:1288	case SDEV_TRANSPORT_OFFLINE:1289		/*1290		 * If the device is offline we refuse to process any1291		 * commands.  The device must be brought online1292		 * before trying any recovery commands.1293		 */1294		if (!sdev->offline_already) {1295			sdev->offline_already = true;1296			sdev_printk(KERN_ERR, sdev,1297				    "rejecting I/O to offline device\n");1298		}1299		return BLK_STS_IOERR;1300	case SDEV_DEL:1301		/*1302		 * If the device is fully deleted, we refuse to1303		 * process any commands as well.1304		 */1305		sdev_printk(KERN_ERR, sdev,1306			    "rejecting I/O to dead device\n");1307		return BLK_STS_IOERR;1308	case SDEV_BLOCK:1309	case SDEV_CREATED_BLOCK:1310		return BLK_STS_RESOURCE;1311	case SDEV_QUIESCE:1312		/*1313		 * If the device is blocked we only accept power management1314		 * commands.1315		 */1316		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))1317			return BLK_STS_RESOURCE;1318		return BLK_STS_OK;1319	default:1320		/*1321		 * For any other not fully online state we only allow1322		 * power management commands.1323		 */1324		if (req && !(req->rq_flags & RQF_PM))1325			return BLK_STS_OFFLINE;1326		return BLK_STS_OK;1327	}1328}1329 1330/*1331 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token1332 * and return the token else return -1.1333 */1334static inline int scsi_dev_queue_ready(struct request_queue *q,1335				  struct scsi_device *sdev)1336{1337	int token;1338 1339	token = sbitmap_get(&sdev->budget_map);1340	if (token < 0)1341		return -1;1342 1343	if (!atomic_read(&sdev->device_blocked))1344		return token;1345 1346	/*1347	 * Only unblock if no other commands are pending and1348	 * if device_blocked has decreased to zero1349	 */1350	if (scsi_device_busy(sdev) > 1 ||1351	    atomic_dec_return(&sdev->device_blocked) > 0) {1352		sbitmap_put(&sdev->budget_map, token);1353		return -1;1354	}1355 1356	SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,1357			 "unblocking device at zero depth\n"));1358 1359	return token;1360}1361 1362/*1363 * scsi_target_queue_ready: checks if there we can send commands to target1364 * @sdev: scsi device on starget to check.1365 */1366static inline int scsi_target_queue_ready(struct Scsi_Host *shost,1367					   struct scsi_device *sdev)1368{1369	struct scsi_target *starget = scsi_target(sdev);1370	unsigned int busy;1371 1372	if (starget->single_lun) {1373		spin_lock_irq(shost->host_lock);1374		if (starget->starget_sdev_user &&1375		    starget->starget_sdev_user != sdev) {1376			spin_unlock_irq(shost->host_lock);1377			return 0;1378		}1379		starget->starget_sdev_user = sdev;1380		spin_unlock_irq(shost->host_lock);1381	}1382 1383	if (starget->can_queue <= 0)1384		return 1;1385 1386	busy = atomic_inc_return(&starget->target_busy) - 1;1387	if (atomic_read(&starget->target_blocked) > 0) {1388		if (busy)1389			goto starved;1390 1391		/*1392		 * unblock after target_blocked iterates to zero1393		 */1394		if (atomic_dec_return(&starget->target_blocked) > 0)1395			goto out_dec;1396 1397		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,1398				 "unblocking target at zero depth\n"));1399	}1400 1401	if (busy >= starget->can_queue)1402		goto starved;1403 1404	return 1;1405 1406starved:1407	spin_lock_irq(shost->host_lock);1408	list_move_tail(&sdev->starved_entry, &shost->starved_list);1409	spin_unlock_irq(shost->host_lock);1410out_dec:1411	if (starget->can_queue > 0)1412		atomic_dec(&starget->target_busy);1413	return 0;1414}1415 1416/*1417 * scsi_host_queue_ready: if we can send requests to shost, return 1 else1418 * return 0. We must end up running the queue again whenever 0 is1419 * returned, else IO can hang.1420 */1421static inline int scsi_host_queue_ready(struct request_queue *q,1422				   struct Scsi_Host *shost,1423				   struct scsi_device *sdev,1424				   struct scsi_cmnd *cmd)1425{1426	if (atomic_read(&shost->host_blocked) > 0) {1427		if (scsi_host_busy(shost) > 0)1428			goto starved;1429 1430		/*1431		 * unblock after host_blocked iterates to zero1432		 */1433		if (atomic_dec_return(&shost->host_blocked) > 0)1434			goto out_dec;1435 1436		SCSI_LOG_MLQUEUE(3,1437			shost_printk(KERN_INFO, shost,1438				     "unblocking host at zero depth\n"));1439	}1440 1441	if (shost->host_self_blocked)1442		goto starved;1443 1444	/* We're OK to process the command, so we can't be starved */1445	if (!list_empty(&sdev->starved_entry)) {1446		spin_lock_irq(shost->host_lock);1447		if (!list_empty(&sdev->starved_entry))1448			list_del_init(&sdev->starved_entry);1449		spin_unlock_irq(shost->host_lock);1450	}1451 1452	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);1453 1454	return 1;1455 1456starved:1457	spin_lock_irq(shost->host_lock);1458	if (list_empty(&sdev->starved_entry))1459		list_add_tail(&sdev->starved_entry, &shost->starved_list);1460	spin_unlock_irq(shost->host_lock);1461out_dec:1462	scsi_dec_host_busy(shost, cmd);1463	return 0;1464}1465 1466/*1467 * Busy state exporting function for request stacking drivers.1468 *1469 * For efficiency, no lock is taken to check the busy state of1470 * shost/starget/sdev, since the returned value is not guaranteed and1471 * may be changed after request stacking drivers call the function,1472 * regardless of taking lock or not.1473 *1474 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi1475 * needs to return 'not busy'. Otherwise, request stacking drivers1476 * may hold requests forever.1477 */1478static bool scsi_mq_lld_busy(struct request_queue *q)1479{1480	struct scsi_device *sdev = q->queuedata;1481	struct Scsi_Host *shost;1482 1483	if (blk_queue_dying(q))1484		return false;1485 1486	shost = sdev->host;1487 1488	/*1489	 * Ignore host/starget busy state.1490	 * Since block layer does not have a concept of fairness across1491	 * multiple queues, congestion of host/starget needs to be handled1492	 * in SCSI layer.1493	 */1494	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))1495		return true;1496 1497	return false;1498}1499 1500/*1501 * Block layer request completion callback. May be called from interrupt1502 * context.1503 */1504static void scsi_complete(struct request *rq)1505{1506	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);1507	enum scsi_disposition disposition;1508 1509	INIT_LIST_HEAD(&cmd->eh_entry);1510 1511	atomic_inc(&cmd->device->iodone_cnt);1512	if (cmd->result)1513		atomic_inc(&cmd->device->ioerr_cnt);1514 1515	disposition = scsi_decide_disposition(cmd);1516	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))1517		disposition = SUCCESS;1518 1519	scsi_log_completion(cmd, disposition);1520 1521	switch (disposition) {1522	case SUCCESS:1523		scsi_finish_command(cmd);1524		break;1525	case NEEDS_RETRY:1526		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);1527		break;1528	case ADD_TO_MLQUEUE:1529		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);1530		break;1531	default:1532		scsi_eh_scmd_add(cmd);1533		break;1534	}1535}1536 1537/**1538 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.1539 * @cmd: command block we are dispatching.1540 *1541 * Return: nonzero return request was rejected and device's queue needs to be1542 * plugged.1543 */1544static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)1545{1546	struct Scsi_Host *host = cmd->device->host;1547	int rtn = 0;1548 1549	atomic_inc(&cmd->device->iorequest_cnt);1550 1551	/* check if the device is still usable */1552	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {1553		/* in SDEV_DEL we error all commands. DID_NO_CONNECT1554		 * returns an immediate error upwards, and signals1555		 * that the device is no longer present */1556		cmd->result = DID_NO_CONNECT << 16;1557		goto done;1558	}1559 1560	/* Check to see if the scsi lld made this device blocked. */1561	if (unlikely(scsi_device_blocked(cmd->device))) {1562		/*1563		 * in blocked state, the command is just put back on1564		 * the device queue.  The suspend state has already1565		 * blocked the queue so future requests should not1566		 * occur until the device transitions out of the1567		 * suspend state.1568		 */1569		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,1570			"queuecommand : device blocked\n"));1571		atomic_dec(&cmd->device->iorequest_cnt);1572		return SCSI_MLQUEUE_DEVICE_BUSY;1573	}1574 1575	/* Store the LUN value in cmnd, if needed. */1576	if (cmd->device->lun_in_cdb)1577		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |1578			       (cmd->device->lun << 5 & 0xe0);1579 1580	scsi_log_send(cmd);1581 1582	/*1583	 * Before we queue this command, check if the command1584	 * length exceeds what the host adapter can handle.1585	 */1586	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {1587		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,1588			       "queuecommand : command too long. "1589			       "cdb_size=%d host->max_cmd_len=%d\n",1590			       cmd->cmd_len, cmd->device->host->max_cmd_len));1591		cmd->result = (DID_ABORT << 16);1592		goto done;1593	}1594 1595	if (unlikely(host->shost_state == SHOST_DEL)) {1596		cmd->result = (DID_NO_CONNECT << 16);1597		goto done;1598 1599	}1600 1601	trace_scsi_dispatch_cmd_start(cmd);1602	rtn = host->hostt->queuecommand(host, cmd);1603	if (rtn) {1604		atomic_dec(&cmd->device->iorequest_cnt);1605		trace_scsi_dispatch_cmd_error(cmd, rtn);1606		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&1607		    rtn != SCSI_MLQUEUE_TARGET_BUSY)1608			rtn = SCSI_MLQUEUE_HOST_BUSY;1609 1610		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,1611			"queuecommand : request rejected\n"));1612	}1613 1614	return rtn;1615 done:1616	scsi_done(cmd);1617	return 0;1618}1619 1620/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */1621static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)1622{1623	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *1624		sizeof(struct scatterlist);1625}1626 1627static blk_status_t scsi_prepare_cmd(struct request *req)1628{1629	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);1630	struct scsi_device *sdev = req->q->queuedata;1631	struct Scsi_Host *shost = sdev->host;1632	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);1633	struct scatterlist *sg;1634 1635	scsi_init_command(sdev, cmd);1636 1637	cmd->eh_eflags = 0;1638	cmd->prot_type = 0;1639	cmd->prot_flags = 0;1640	cmd->submitter = 0;1641	memset(&cmd->sdb, 0, sizeof(cmd->sdb));1642	cmd->underflow = 0;1643	cmd->transfersize = 0;1644	cmd->host_scribble = NULL;1645	cmd->result = 0;1646	cmd->extra_len = 0;1647	cmd->state = 0;1648	if (in_flight)1649		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);1650 1651	/*1652	 * Only clear the driver-private command data if the LLD does not supply1653	 * a function to initialize that data.1654	 */1655	if (!shost->hostt->init_cmd_priv)1656		memset(cmd + 1, 0, shost->hostt->cmd_size);1657 1658	cmd->prot_op = SCSI_PROT_NORMAL;1659	if (blk_rq_bytes(req))1660		cmd->sc_data_direction = rq_dma_dir(req);1661	else1662		cmd->sc_data_direction = DMA_NONE;1663 1664	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;1665	cmd->sdb.table.sgl = sg;1666 1667	if (scsi_host_get_prot(shost)) {1668		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));1669 1670		cmd->prot_sdb->table.sgl =1671			(struct scatterlist *)(cmd->prot_sdb + 1);1672	}1673 1674	/*1675	 * Special handling for passthrough commands, which don't go to the ULP1676	 * at all:1677	 */1678	if (blk_rq_is_passthrough(req))1679		return scsi_setup_scsi_cmnd(sdev, req);1680 1681	if (sdev->handler && sdev->handler->prep_fn) {1682		blk_status_t ret = sdev->handler->prep_fn(sdev, req);1683 1684		if (ret != BLK_STS_OK)1685			return ret;1686	}1687 1688	/* Usually overridden by the ULP */1689	cmd->allowed = 0;1690	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));1691	return scsi_cmd_to_driver(cmd)->init_command(cmd);1692}1693 1694static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)1695{1696	struct request *req = scsi_cmd_to_rq(cmd);1697 1698	switch (cmd->submitter) {1699	case SUBMITTED_BY_BLOCK_LAYER:1700		break;1701	case SUBMITTED_BY_SCSI_ERROR_HANDLER:1702		return scsi_eh_done(cmd);1703	case SUBMITTED_BY_SCSI_RESET_IOCTL:1704		return;1705	}1706 1707	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))1708		return;1709	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))1710		return;1711	trace_scsi_dispatch_cmd_done(cmd);1712 1713	if (complete_directly)1714		blk_mq_complete_request_direct(req, scsi_complete);1715	else1716		blk_mq_complete_request(req);1717}1718 1719void scsi_done(struct scsi_cmnd *cmd)1720{1721	scsi_done_internal(cmd, false);1722}1723EXPORT_SYMBOL(scsi_done);1724 1725void scsi_done_direct(struct scsi_cmnd *cmd)1726{1727	scsi_done_internal(cmd, true);1728}1729EXPORT_SYMBOL(scsi_done_direct);1730 1731static void scsi_mq_put_budget(struct request_queue *q, int budget_token)1732{1733	struct scsi_device *sdev = q->queuedata;1734 1735	sbitmap_put(&sdev->budget_map, budget_token);1736}1737 1738/*1739 * When to reinvoke queueing after a resource shortage. It's 3 msecs to1740 * not change behaviour from the previous unplug mechanism, experimentation1741 * may prove this needs changing.1742 */1743#define SCSI_QUEUE_DELAY 31744 1745static int scsi_mq_get_budget(struct request_queue *q)1746{1747	struct scsi_device *sdev = q->queuedata;1748	int token = scsi_dev_queue_ready(q, sdev);1749 1750	if (token >= 0)1751		return token;1752 1753	atomic_inc(&sdev->restarts);1754 1755	/*1756	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).1757	 * .restarts must be incremented before .device_busy is read because the1758	 * code in scsi_run_queue_async() depends on the order of these operations.1759	 */1760	smp_mb__after_atomic();1761 1762	/*1763	 * If all in-flight requests originated from this LUN are completed1764	 * before reading .device_busy, sdev->device_busy will be observed as1765	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request1766	 * soon. Otherwise, completion of one of these requests will observe1767	 * the .restarts flag, and the request queue will be run for handling1768	 * this request, see scsi_end_request().1769	 */1770	if (unlikely(scsi_device_busy(sdev) == 0 &&1771				!scsi_device_blocked(sdev)))1772		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);1773	return -1;1774}1775 1776static void scsi_mq_set_rq_budget_token(struct request *req, int token)1777{1778	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);1779 1780	cmd->budget_token = token;1781}1782 1783static int scsi_mq_get_rq_budget_token(struct request *req)1784{1785	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);1786 1787	return cmd->budget_token;1788}1789 1790static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,1791			 const struct blk_mq_queue_data *bd)1792{1793	struct request *req = bd->rq;1794	struct request_queue *q = req->q;1795	struct scsi_device *sdev = q->queuedata;1796	struct Scsi_Host *shost = sdev->host;1797	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);1798	blk_status_t ret;1799	int reason;1800 1801	WARN_ON_ONCE(cmd->budget_token < 0);1802 1803	/*1804	 * If the device is not in running state we will reject some or all1805	 * commands.1806	 */1807	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {1808		ret = scsi_device_state_check(sdev, req);1809		if (ret != BLK_STS_OK)1810			goto out_put_budget;1811	}1812 1813	ret = BLK_STS_RESOURCE;1814	if (!scsi_target_queue_ready(shost, sdev))1815		goto out_put_budget;1816	if (unlikely(scsi_host_in_recovery(shost))) {1817		if (cmd->flags & SCMD_FAIL_IF_RECOVERING)1818			ret = BLK_STS_OFFLINE;1819		goto out_dec_target_busy;1820	}1821	if (!scsi_host_queue_ready(q, shost, sdev, cmd))1822		goto out_dec_target_busy;1823 1824	if (!(req->rq_flags & RQF_DONTPREP)) {1825		ret = scsi_prepare_cmd(req);1826		if (ret != BLK_STS_OK)1827			goto out_dec_host_busy;1828		req->rq_flags |= RQF_DONTPREP;1829	} else {1830		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);1831	}1832 1833	cmd->flags &= SCMD_PRESERVED_FLAGS;1834	if (sdev->simple_tags)1835		cmd->flags |= SCMD_TAGGED;1836	if (bd->last)1837		cmd->flags |= SCMD_LAST;1838 1839	scsi_set_resid(cmd, 0);1840	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);1841	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;1842 1843	blk_mq_start_request(req);1844	reason = scsi_dispatch_cmd(cmd);1845	if (reason) {1846		scsi_set_blocked(cmd, reason);1847		ret = BLK_STS_RESOURCE;1848		goto out_dec_host_busy;1849	}1850 1851	return BLK_STS_OK;1852 1853out_dec_host_busy:1854	scsi_dec_host_busy(shost, cmd);1855out_dec_target_busy:1856	if (scsi_target(sdev)->can_queue > 0)1857		atomic_dec(&scsi_target(sdev)->target_busy);1858out_put_budget:1859	scsi_mq_put_budget(q, cmd->budget_token);1860	cmd->budget_token = -1;1861	switch (ret) {1862	case BLK_STS_OK:1863		break;1864	case BLK_STS_RESOURCE:1865		if (scsi_device_blocked(sdev))1866			ret = BLK_STS_DEV_RESOURCE;1867		break;1868	case BLK_STS_AGAIN:1869		cmd->result = DID_BUS_BUSY << 16;1870		if (req->rq_flags & RQF_DONTPREP)1871			scsi_mq_uninit_cmd(cmd);1872		break;1873	default:1874		if (unlikely(!scsi_device_online(sdev)))1875			cmd->result = DID_NO_CONNECT << 16;1876		else1877			cmd->result = DID_ERROR << 16;1878		/*1879		 * Make sure to release all allocated resources when1880		 * we hit an error, as we will never see this command1881		 * again.1882		 */1883		if (req->rq_flags & RQF_DONTPREP)1884			scsi_mq_uninit_cmd(cmd);1885		scsi_run_queue_async(sdev);1886		break;1887	}1888	return ret;1889}1890 1891static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,1892				unsigned int hctx_idx, unsigned int numa_node)1893{1894	struct Scsi_Host *shost = set->driver_data;1895	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);1896	struct scatterlist *sg;1897	int ret = 0;1898 1899	cmd->sense_buffer =1900		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);1901	if (!cmd->sense_buffer)1902		return -ENOMEM;1903 1904	if (scsi_host_get_prot(shost)) {1905		sg = (void *)cmd + sizeof(struct scsi_cmnd) +1906			shost->hostt->cmd_size;1907		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);1908	}1909 1910	if (shost->hostt->init_cmd_priv) {1911		ret = shost->hostt->init_cmd_priv(shost, cmd);1912		if (ret < 0)1913			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);1914	}1915 1916	return ret;1917}1918 1919static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,1920				 unsigned int hctx_idx)1921{1922	struct Scsi_Host *shost = set->driver_data;1923	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);1924 1925	if (shost->hostt->exit_cmd_priv)1926		shost->hostt->exit_cmd_priv(shost, cmd);1927	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);1928}1929 1930 1931static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)1932{1933	struct Scsi_Host *shost = hctx->driver_data;1934 1935	if (shost->hostt->mq_poll)1936		return shost->hostt->mq_poll(shost, hctx->queue_num);1937 1938	return 0;1939}1940 1941static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,1942			  unsigned int hctx_idx)1943{1944	struct Scsi_Host *shost = data;1945 1946	hctx->driver_data = shost;1947	return 0;1948}1949 1950static void scsi_map_queues(struct blk_mq_tag_set *set)1951{1952	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);1953 1954	if (shost->hostt->map_queues)1955		return shost->hostt->map_queues(shost);1956	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);1957}1958 1959void scsi_init_limits(struct Scsi_Host *shost, struct queue_limits *lim)1960{1961	struct device *dev = shost->dma_dev;1962 1963	memset(lim, 0, sizeof(*lim));1964	lim->max_segments =1965		min_t(unsigned short, shost->sg_tablesize, SG_MAX_SEGMENTS);1966 1967	if (scsi_host_prot_dma(shost)) {1968		shost->sg_prot_tablesize =1969			min_not_zero(shost->sg_prot_tablesize,1970				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);1971		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);1972		lim->max_integrity_segments = shost->sg_prot_tablesize;1973	}1974 1975	lim->max_hw_sectors = shost->max_sectors;1976	lim->seg_boundary_mask = shost->dma_boundary;1977	lim->max_segment_size = shost->max_segment_size;1978	lim->virt_boundary_mask = shost->virt_boundary_mask;1979	lim->dma_alignment = max_t(unsigned int,1980		shost->dma_alignment, dma_get_cache_alignment() - 1);1981 1982	if (shost->no_highmem)1983		lim->features |= BLK_FEAT_BOUNCE_HIGH;1984 1985	/*1986	 * Propagate the DMA formation properties to the dma-mapping layer as1987	 * a courtesy service to the LLDDs.  This needs to check that the buses1988	 * actually support the DMA API first, though.1989	 */1990	if (dev->dma_parms) {1991		dma_set_seg_boundary(dev, shost->dma_boundary);1992		dma_set_max_seg_size(dev, shost->max_segment_size);1993	}1994}1995EXPORT_SYMBOL_GPL(scsi_init_limits);1996 1997static const struct blk_mq_ops scsi_mq_ops_no_commit = {1998	.get_budget	= scsi_mq_get_budget,1999	.put_budget	= scsi_mq_put_budget,2000	.queue_rq	= scsi_queue_rq,2001	.complete	= scsi_complete,2002	.timeout	= scsi_timeout,2003#ifdef CONFIG_BLK_DEBUG_FS2004	.show_rq	= scsi_show_rq,2005#endif2006	.init_request	= scsi_mq_init_request,2007	.exit_request	= scsi_mq_exit_request,2008	.cleanup_rq	= scsi_cleanup_rq,2009	.busy		= scsi_mq_lld_busy,2010	.map_queues	= scsi_map_queues,2011	.init_hctx	= scsi_init_hctx,2012	.poll		= scsi_mq_poll,2013	.set_rq_budget_token = scsi_mq_set_rq_budget_token,2014	.get_rq_budget_token = scsi_mq_get_rq_budget_token,2015};2016 2017 2018static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)2019{2020	struct Scsi_Host *shost = hctx->driver_data;2021 2022	shost->hostt->commit_rqs(shost, hctx->queue_num);2023}2024 2025static const struct blk_mq_ops scsi_mq_ops = {2026	.get_budget	= scsi_mq_get_budget,2027	.put_budget	= scsi_mq_put_budget,2028	.queue_rq	= scsi_queue_rq,2029	.commit_rqs	= scsi_commit_rqs,2030	.complete	= scsi_complete,2031	.timeout	= scsi_timeout,2032#ifdef CONFIG_BLK_DEBUG_FS2033	.show_rq	= scsi_show_rq,2034#endif2035	.init_request	= scsi_mq_init_request,2036	.exit_request	= scsi_mq_exit_request,2037	.cleanup_rq	= scsi_cleanup_rq,2038	.busy		= scsi_mq_lld_busy,2039	.map_queues	= scsi_map_queues,2040	.init_hctx	= scsi_init_hctx,2041	.poll		= scsi_mq_poll,2042	.set_rq_budget_token = scsi_mq_set_rq_budget_token,2043	.get_rq_budget_token = scsi_mq_get_rq_budget_token,2044};2045 2046int scsi_mq_setup_tags(struct Scsi_Host *shost)2047{2048	unsigned int cmd_size, sgl_size;2049	struct blk_mq_tag_set *tag_set = &shost->tag_set;2050 2051	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),2052				scsi_mq_inline_sgl_size(shost));2053	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;2054	if (scsi_host_get_prot(shost))2055		cmd_size += sizeof(struct scsi_data_buffer) +2056			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;2057 2058	memset(tag_set, 0, sizeof(*tag_set));2059	if (shost->hostt->commit_rqs)2060		tag_set->ops = &scsi_mq_ops;2061	else2062		tag_set->ops = &scsi_mq_ops_no_commit;2063	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;2064	tag_set->nr_maps = shost->nr_maps ? : 1;2065	tag_set->queue_depth = shost->can_queue;2066	tag_set->cmd_size = cmd_size;2067	tag_set->numa_node = dev_to_node(shost->dma_dev);2068	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;2069	tag_set->flags |=2070		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);2071	if (shost->queuecommand_may_block)2072		tag_set->flags |= BLK_MQ_F_BLOCKING;2073	tag_set->driver_data = shost;2074	if (shost->host_tagset)2075		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;2076 2077	return blk_mq_alloc_tag_set(tag_set);2078}2079 2080void scsi_mq_free_tags(struct kref *kref)2081{2082	struct Scsi_Host *shost = container_of(kref, typeof(*shost),2083					       tagset_refcnt);2084 2085	blk_mq_free_tag_set(&shost->tag_set);2086	complete(&shost->tagset_freed);2087}2088 2089/**2090 * scsi_device_from_queue - return sdev associated with a request_queue2091 * @q: The request queue to return the sdev from2092 *2093 * Return the sdev associated with a request queue or NULL if the2094 * request_queue does not reference a SCSI device.2095 */2096struct scsi_device *scsi_device_from_queue(struct request_queue *q)2097{2098	struct scsi_device *sdev = NULL;2099 2100	if (q->mq_ops == &scsi_mq_ops_no_commit ||2101	    q->mq_ops == &scsi_mq_ops)2102		sdev = q->queuedata;2103	if (!sdev || !get_device(&sdev->sdev_gendev))2104		sdev = NULL;2105 2106	return sdev;2107}2108/*2109 * pktcdvd should have been integrated into the SCSI layers, but for historical2110 * reasons like the old IDE driver it isn't.  This export allows it to safely2111 * probe if a given device is a SCSI one and only attach to that.2112 */2113#ifdef CONFIG_CDROM_PKTCDVD_MODULE2114EXPORT_SYMBOL_GPL(scsi_device_from_queue);2115#endif2116 2117/**2118 * scsi_block_requests - Utility function used by low-level drivers to prevent2119 * further commands from being queued to the device.2120 * @shost:  host in question2121 *2122 * There is no timer nor any other means by which the requests get unblocked2123 * other than the low-level driver calling scsi_unblock_requests().2124 */2125void scsi_block_requests(struct Scsi_Host *shost)2126{2127	shost->host_self_blocked = 1;2128}2129EXPORT_SYMBOL(scsi_block_requests);2130 2131/**2132 * scsi_unblock_requests - Utility function used by low-level drivers to allow2133 * further commands to be queued to the device.2134 * @shost:  host in question2135 *2136 * There is no timer nor any other means by which the requests get unblocked2137 * other than the low-level driver calling scsi_unblock_requests(). This is done2138 * as an API function so that changes to the internals of the scsi mid-layer2139 * won't require wholesale changes to drivers that use this feature.2140 */2141void scsi_unblock_requests(struct Scsi_Host *shost)2142{2143	shost->host_self_blocked = 0;2144	scsi_run_host_queues(shost);2145}2146EXPORT_SYMBOL(scsi_unblock_requests);2147 2148void scsi_exit_queue(void)2149{2150	kmem_cache_destroy(scsi_sense_cache);2151}2152 2153/**2154 *	scsi_mode_select - issue a mode select2155 *	@sdev:	SCSI device to be queried2156 *	@pf:	Page format bit (1 == standard, 0 == vendor specific)2157 *	@sp:	Save page bit (0 == don't save, 1 == save)2158 *	@buffer: request buffer (may not be smaller than eight bytes)2159 *	@len:	length of request buffer.2160 *	@timeout: command timeout2161 *	@retries: number of retries before failing2162 *	@data: returns a structure abstracting the mode header data2163 *	@sshdr: place to put sense data (or NULL if no sense to be collected).2164 *		must be SCSI_SENSE_BUFFERSIZE big.2165 *2166 *	Returns zero if successful; negative error number or scsi2167 *	status on error2168 *2169 */2170int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,2171		     unsigned char *buffer, int len, int timeout, int retries,2172		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)2173{2174	unsigned char cmd[10];2175	unsigned char *real_buffer;2176	const struct scsi_exec_args exec_args = {2177		.sshdr = sshdr,2178	};2179	int ret;2180 2181	memset(cmd, 0, sizeof(cmd));2182	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);2183 2184	/*2185	 * Use MODE SELECT(10) if the device asked for it or if the mode page2186	 * and the mode select header cannot fit within the maximumm 255 bytes2187	 * of the MODE SELECT(6) command.2188	 */2189	if (sdev->use_10_for_ms ||2190	    len + 4 > 255 ||2191	    data->block_descriptor_length > 255) {2192		if (len > 65535 - 8)2193			return -EINVAL;2194		real_buffer = kmalloc(8 + len, GFP_KERNEL);2195		if (!real_buffer)2196			return -ENOMEM;2197		memcpy(real_buffer + 8, buffer, len);2198		len += 8;2199		real_buffer[0] = 0;2200		real_buffer[1] = 0;2201		real_buffer[2] = data->medium_type;2202		real_buffer[3] = data->device_specific;2203		real_buffer[4] = data->longlba ? 0x01 : 0;2204		real_buffer[5] = 0;2205		put_unaligned_be16(data->block_descriptor_length,2206				   &real_buffer[6]);2207 2208		cmd[0] = MODE_SELECT_10;2209		put_unaligned_be16(len, &cmd[7]);2210	} else {2211		if (data->longlba)2212			return -EINVAL;2213 2214		real_buffer = kmalloc(4 + len, GFP_KERNEL);2215		if (!real_buffer)2216			return -ENOMEM;2217		memcpy(real_buffer + 4, buffer, len);2218		len += 4;2219		real_buffer[0] = 0;2220		real_buffer[1] = data->medium_type;2221		real_buffer[2] = data->device_specific;2222		real_buffer[3] = data->block_descriptor_length;2223 2224		cmd[0] = MODE_SELECT;2225		cmd[4] = len;2226	}2227 2228	ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,2229			       timeout, retries, &exec_args);2230	kfree(real_buffer);2231	return ret;2232}2233EXPORT_SYMBOL_GPL(scsi_mode_select);2234 2235/**2236 *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.2237 *	@sdev:	SCSI device to be queried2238 *	@dbd:	set to prevent mode sense from returning block descriptors2239 *	@modepage: mode page being requested2240 *	@subpage: sub-page of the mode page being requested2241 *	@buffer: request buffer (may not be smaller than eight bytes)2242 *	@len:	length of request buffer.2243 *	@timeout: command timeout2244 *	@retries: number of retries before failing2245 *	@data: returns a structure abstracting the mode header data2246 *	@sshdr: place to put sense data (or NULL if no sense to be collected).2247 *		must be SCSI_SENSE_BUFFERSIZE big.2248 *2249 *	Returns zero if successful, or a negative error number on failure2250 */2251int2252scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,2253		  unsigned char *buffer, int len, int timeout, int retries,2254		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)2255{2256	unsigned char cmd[12];2257	int use_10_for_ms;2258	int header_length;2259	int result;2260	struct scsi_sense_hdr my_sshdr;2261	struct scsi_failure failure_defs[] = {2262		{2263			.sense = UNIT_ATTENTION,2264			.asc = SCMD_FAILURE_ASC_ANY,2265			.ascq = SCMD_FAILURE_ASCQ_ANY,2266			.allowed = retries,2267			.result = SAM_STAT_CHECK_CONDITION,2268		},2269		{}2270	};2271	struct scsi_failures failures = {2272		.failure_definitions = failure_defs,2273	};2274	const struct scsi_exec_args exec_args = {2275		/* caller might not be interested in sense, but we need it */2276		.sshdr = sshdr ? : &my_sshdr,2277		.failures = &failures,2278	};2279 2280	memset(data, 0, sizeof(*data));2281	memset(&cmd[0], 0, 12);2282 2283	dbd = sdev->set_dbd_for_ms ? 8 : dbd;2284	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */2285	cmd[2] = modepage;2286	cmd[3] = subpage;2287 2288	sshdr = exec_args.sshdr;2289 2290 retry:2291	use_10_for_ms = sdev->use_10_for_ms || len > 255;2292 2293	if (use_10_for_ms) {2294		if (len < 8 || len > 65535)2295			return -EINVAL;2296 2297		cmd[0] = MODE_SENSE_10;2298		put_unaligned_be16(len, &cmd[7]);2299		header_length = 8;2300	} else {2301		if (len < 4)2302			return -EINVAL;2303 2304		cmd[0] = MODE_SENSE;2305		cmd[4] = len;2306		header_length = 4;2307	}2308 2309	memset(buffer, 0, len);2310 2311	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,2312				  timeout, retries, &exec_args);2313	if (result < 0)2314		return result;2315 2316	/* This code looks awful: what it's doing is making sure an2317	 * ILLEGAL REQUEST sense return identifies the actual command2318	 * byte as the problem.  MODE_SENSE commands can return2319	 * ILLEGAL REQUEST if the code page isn't supported */2320 2321	if (!scsi_status_is_good(result)) {2322		if (scsi_sense_valid(sshdr)) {2323			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&2324			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {2325				/*2326				 * Invalid command operation code: retry using2327				 * MODE SENSE(6) if this was a MODE SENSE(10)2328				 * request, except if the request mode page is2329				 * too large for MODE SENSE single byte2330				 * allocation length field.2331				 */2332				if (use_10_for_ms) {2333					if (len > 255)2334						return -EIO;2335					sdev->use_10_for_ms = 0;2336					goto retry;2337				}2338			}2339		}2340		return -EIO;2341	}2342	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&2343		     (modepage == 6 || modepage == 8))) {2344		/* Initio breakage? */2345		header_length = 0;2346		data->length = 13;2347		data->medium_type = 0;2348		data->device_specific = 0;2349		data->longlba = 0;2350		data->block_descriptor_length = 0;2351	} else if (use_10_for_ms) {2352		data->length = get_unaligned_be16(&buffer[0]) + 2;2353		data->medium_type = buffer[2];2354		data->device_specific = buffer[3];2355		data->longlba = buffer[4] & 0x01;2356		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);2357	} else {2358		data->length = buffer[0] + 1;2359		data->medium_type = buffer[1];2360		data->device_specific = buffer[2];2361		data->block_descriptor_length = buffer[3];2362	}2363	data->header_length = header_length;2364 2365	return 0;2366}2367EXPORT_SYMBOL(scsi_mode_sense);2368 2369/**2370 *	scsi_test_unit_ready - test if unit is ready2371 *	@sdev:	scsi device to change the state of.2372 *	@timeout: command timeout2373 *	@retries: number of retries before failing2374 *	@sshdr: outpout pointer for decoded sense information.2375 *2376 *	Returns zero if unsuccessful or an error if TUR failed.  For2377 *	removable media, UNIT_ATTENTION sets ->changed flag.2378 **/2379int2380scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,2381		     struct scsi_sense_hdr *sshdr)2382{2383	char cmd[] = {2384		TEST_UNIT_READY, 0, 0, 0, 0, 0,2385	};2386	const struct scsi_exec_args exec_args = {2387		.sshdr = sshdr,2388	};2389	int result;2390 2391	/* try to eat the UNIT_ATTENTION if there are enough retries */2392	do {2393		result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,2394					  timeout, 1, &exec_args);2395		if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&2396		    sshdr->sense_key == UNIT_ATTENTION)2397			sdev->changed = 1;2398	} while (result > 0 && scsi_sense_valid(sshdr) &&2399		 sshdr->sense_key == UNIT_ATTENTION && --retries);2400 2401	return result;2402}2403EXPORT_SYMBOL(scsi_test_unit_ready);2404 2405/**2406 *	scsi_device_set_state - Take the given device through the device state model.2407 *	@sdev:	scsi device to change the state of.2408 *	@state:	state to change to.2409 *2410 *	Returns zero if successful or an error if the requested2411 *	transition is illegal.2412 */2413int2414scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)2415{2416	enum scsi_device_state oldstate = sdev->sdev_state;2417 2418	if (state == oldstate)2419		return 0;2420 2421	switch (state) {2422	case SDEV_CREATED:2423		switch (oldstate) {2424		case SDEV_CREATED_BLOCK:2425			break;2426		default:2427			goto illegal;2428		}2429		break;2430 2431	case SDEV_RUNNING:2432		switch (oldstate) {2433		case SDEV_CREATED:2434		case SDEV_OFFLINE:2435		case SDEV_TRANSPORT_OFFLINE:2436		case SDEV_QUIESCE:2437		case SDEV_BLOCK:2438			break;2439		default:2440			goto illegal;2441		}2442		break;2443 2444	case SDEV_QUIESCE:2445		switch (oldstate) {2446		case SDEV_RUNNING:2447		case SDEV_OFFLINE:2448		case SDEV_TRANSPORT_OFFLINE:2449			break;2450		default:2451			goto illegal;2452		}2453		break;2454 2455	case SDEV_OFFLINE:2456	case SDEV_TRANSPORT_OFFLINE:2457		switch (oldstate) {2458		case SDEV_CREATED:2459		case SDEV_RUNNING:2460		case SDEV_QUIESCE:2461		case SDEV_BLOCK:2462			break;2463		default:2464			goto illegal;2465		}2466		break;2467 2468	case SDEV_BLOCK:2469		switch (oldstate) {2470		case SDEV_RUNNING:2471		case SDEV_CREATED_BLOCK:2472		case SDEV_QUIESCE:2473		case SDEV_OFFLINE:2474			break;2475		default:2476			goto illegal;2477		}2478		break;2479 2480	case SDEV_CREATED_BLOCK:2481		switch (oldstate) {2482		case SDEV_CREATED:2483			break;2484		default:2485			goto illegal;2486		}2487		break;2488 2489	case SDEV_CANCEL:2490		switch (oldstate) {2491		case SDEV_CREATED:2492		case SDEV_RUNNING:2493		case SDEV_QUIESCE:2494		case SDEV_OFFLINE:2495		case SDEV_TRANSPORT_OFFLINE:2496			break;2497		default:2498			goto illegal;2499		}2500		break;2501 2502	case SDEV_DEL:2503		switch (oldstate) {2504		case SDEV_CREATED:2505		case SDEV_RUNNING:2506		case SDEV_OFFLINE:2507		case SDEV_TRANSPORT_OFFLINE:2508		case SDEV_CANCEL:2509		case SDEV_BLOCK:2510		case SDEV_CREATED_BLOCK:2511			break;2512		default:2513			goto illegal;2514		}2515		break;2516 2517	}2518	sdev->offline_already = false;2519	sdev->sdev_state = state;2520	return 0;2521 2522 illegal:2523	SCSI_LOG_ERROR_RECOVERY(1,2524				sdev_printk(KERN_ERR, sdev,2525					    "Illegal state transition %s->%s",2526					    scsi_device_state_name(oldstate),2527					    scsi_device_state_name(state))2528				);2529	return -EINVAL;2530}2531EXPORT_SYMBOL(scsi_device_set_state);2532 2533/**2534 *	scsi_evt_emit - emit a single SCSI device uevent2535 *	@sdev: associated SCSI device2536 *	@evt: event to emit2537 *2538 *	Send a single uevent (scsi_event) to the associated scsi_device.2539 */2540static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)2541{2542	int idx = 0;2543	char *envp[3];2544 2545	switch (evt->evt_type) {2546	case SDEV_EVT_MEDIA_CHANGE:2547		envp[idx++] = "SDEV_MEDIA_CHANGE=1";2548		break;2549	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:2550		scsi_rescan_device(sdev);2551		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";2552		break;2553	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:2554		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";2555		break;2556	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:2557	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";2558		break;2559	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:2560		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";2561		break;2562	case SDEV_EVT_LUN_CHANGE_REPORTED:2563		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";2564		break;2565	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:2566		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";2567		break;2568	case SDEV_EVT_POWER_ON_RESET_OCCURRED:2569		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";2570		break;2571	default:2572		/* do nothing */2573		break;2574	}2575 2576	envp[idx++] = NULL;2577 2578	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);2579}2580 2581/**2582 *	scsi_evt_thread - send a uevent for each scsi event2583 *	@work: work struct for scsi_device2584 *2585 *	Dispatch queued events to their associated scsi_device kobjects2586 *	as uevents.2587 */2588void scsi_evt_thread(struct work_struct *work)2589{2590	struct scsi_device *sdev;2591	enum scsi_device_event evt_type;2592	LIST_HEAD(event_list);2593 2594	sdev = container_of(work, struct scsi_device, event_work);2595 2596	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)2597		if (test_and_clear_bit(evt_type, sdev->pending_events))2598			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);2599 2600	while (1) {2601		struct scsi_event *evt;2602		struct list_head *this, *tmp;2603		unsigned long flags;2604 2605		spin_lock_irqsave(&sdev->list_lock, flags);2606		list_splice_init(&sdev->event_list, &event_list);2607		spin_unlock_irqrestore(&sdev->list_lock, flags);2608 2609		if (list_empty(&event_list))2610			break;2611 2612		list_for_each_safe(this, tmp, &event_list) {2613			evt = list_entry(this, struct scsi_event, node);2614			list_del(&evt->node);2615			scsi_evt_emit(sdev, evt);2616			kfree(evt);2617		}2618	}2619}2620 2621/**2622 * 	sdev_evt_send - send asserted event to uevent thread2623 *	@sdev: scsi_device event occurred on2624 *	@evt: event to send2625 *2626 *	Assert scsi device event asynchronously.2627 */2628void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)2629{2630	unsigned long flags;2631 2632#if 02633	/* FIXME: currently this check eliminates all media change events2634	 * for polled devices.  Need to update to discriminate between AN2635	 * and polled events */2636	if (!test_bit(evt->evt_type, sdev->supported_events)) {2637		kfree(evt);2638		return;2639	}2640#endif2641 2642	spin_lock_irqsave(&sdev->list_lock, flags);2643	list_add_tail(&evt->node, &sdev->event_list);2644	schedule_work(&sdev->event_work);2645	spin_unlock_irqrestore(&sdev->list_lock, flags);2646}2647EXPORT_SYMBOL_GPL(sdev_evt_send);2648 2649/**2650 * 	sdev_evt_alloc - allocate a new scsi event2651 *	@evt_type: type of event to allocate2652 *	@gfpflags: GFP flags for allocation2653 *2654 *	Allocates and returns a new scsi_event.2655 */2656struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,2657				  gfp_t gfpflags)2658{2659	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);2660	if (!evt)2661		return NULL;2662 2663	evt->evt_type = evt_type;2664	INIT_LIST_HEAD(&evt->node);2665 2666	/* evt_type-specific initialization, if any */2667	switch (evt_type) {2668	case SDEV_EVT_MEDIA_CHANGE:2669	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:2670	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:2671	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:2672	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:2673	case SDEV_EVT_LUN_CHANGE_REPORTED:2674	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:2675	case SDEV_EVT_POWER_ON_RESET_OCCURRED:2676	default:2677		/* do nothing */2678		break;2679	}2680 2681	return evt;2682}2683EXPORT_SYMBOL_GPL(sdev_evt_alloc);2684 2685/**2686 * 	sdev_evt_send_simple - send asserted event to uevent thread2687 *	@sdev: scsi_device event occurred on2688 *	@evt_type: type of event to send2689 *	@gfpflags: GFP flags for allocation2690 *2691 *	Assert scsi device event asynchronously, given an event type.2692 */2693void sdev_evt_send_simple(struct scsi_device *sdev,2694			  enum scsi_device_event evt_type, gfp_t gfpflags)2695{2696	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);2697	if (!evt) {2698		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",2699			    evt_type);2700		return;2701	}2702 2703	sdev_evt_send(sdev, evt);2704}2705EXPORT_SYMBOL_GPL(sdev_evt_send_simple);2706 2707/**2708 *	scsi_device_quiesce - Block all commands except power management.2709 *	@sdev:	scsi device to quiesce.2710 *2711 *	This works by trying to transition to the SDEV_QUIESCE state2712 *	(which must be a legal transition).  When the device is in this2713 *	state, only power management requests will be accepted, all others will2714 *	be deferred.2715 *2716 *	Must be called with user context, may sleep.2717 *2718 *	Returns zero if unsuccessful or an error if not.2719 */2720int2721scsi_device_quiesce(struct scsi_device *sdev)2722{2723	struct request_queue *q = sdev->request_queue;2724	int err;2725 2726	/*2727	 * It is allowed to call scsi_device_quiesce() multiple times from2728	 * the same context but concurrent scsi_device_quiesce() calls are2729	 * not allowed.2730	 */2731	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);2732 2733	if (sdev->quiesced_by == current)2734		return 0;2735 2736	blk_set_pm_only(q);2737 2738	blk_mq_freeze_queue(q);2739	/*2740	 * Ensure that the effect of blk_set_pm_only() will be visible2741	 * for percpu_ref_tryget() callers that occur after the queue2742	 * unfreeze even if the queue was already frozen before this function2743	 * was called. See also https://lwn.net/Articles/573497/.2744	 */2745	synchronize_rcu();2746	blk_mq_unfreeze_queue(q);2747 2748	mutex_lock(&sdev->state_mutex);2749	err = scsi_device_set_state(sdev, SDEV_QUIESCE);2750	if (err == 0)2751		sdev->quiesced_by = current;2752	else2753		blk_clear_pm_only(q);2754	mutex_unlock(&sdev->state_mutex);2755 2756	return err;2757}2758EXPORT_SYMBOL(scsi_device_quiesce);2759 2760/**2761 *	scsi_device_resume - Restart user issued commands to a quiesced device.2762 *	@sdev:	scsi device to resume.2763 *2764 *	Moves the device from quiesced back to running and restarts the2765 *	queues.2766 *2767 *	Must be called with user context, may sleep.2768 */2769void scsi_device_resume(struct scsi_device *sdev)2770{2771	/* check if the device state was mutated prior to resume, and if2772	 * so assume the state is being managed elsewhere (for example2773	 * device deleted during suspend)2774	 */2775	mutex_lock(&sdev->state_mutex);2776	if (sdev->sdev_state == SDEV_QUIESCE)2777		scsi_device_set_state(sdev, SDEV_RUNNING);2778	if (sdev->quiesced_by) {2779		sdev->quiesced_by = NULL;2780		blk_clear_pm_only(sdev->request_queue);2781	}2782	mutex_unlock(&sdev->state_mutex);2783}2784EXPORT_SYMBOL(scsi_device_resume);2785 2786static void2787device_quiesce_fn(struct scsi_device *sdev, void *data)2788{2789	scsi_device_quiesce(sdev);2790}2791 2792void2793scsi_target_quiesce(struct scsi_target *starget)2794{2795	starget_for_each_device(starget, NULL, device_quiesce_fn);2796}2797EXPORT_SYMBOL(scsi_target_quiesce);2798 2799static void2800device_resume_fn(struct scsi_device *sdev, void *data)2801{2802	scsi_device_resume(sdev);2803}2804 2805void2806scsi_target_resume(struct scsi_target *starget)2807{2808	starget_for_each_device(starget, NULL, device_resume_fn);2809}2810EXPORT_SYMBOL(scsi_target_resume);2811 2812static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)2813{2814	if (scsi_device_set_state(sdev, SDEV_BLOCK))2815		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);2816 2817	return 0;2818}2819 2820void scsi_start_queue(struct scsi_device *sdev)2821{2822	if (cmpxchg(&sdev->queue_stopped, 1, 0))2823		blk_mq_unquiesce_queue(sdev->request_queue);2824}2825 2826static void scsi_stop_queue(struct scsi_device *sdev)2827{2828	/*2829	 * The atomic variable of ->queue_stopped covers that2830	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.2831	 *2832	 * The caller needs to wait until quiesce is done.2833	 */2834	if (!cmpxchg(&sdev->queue_stopped, 0, 1))2835		blk_mq_quiesce_queue_nowait(sdev->request_queue);2836}2837 2838/**2839 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state2840 * @sdev: device to block2841 *2842 * Pause SCSI command processing on the specified device. Does not sleep.2843 *2844 * Returns zero if successful or a negative error code upon failure.2845 *2846 * Notes:2847 * This routine transitions the device to the SDEV_BLOCK state (which must be2848 * a legal transition). When the device is in this state, command processing2849 * is paused until the device leaves the SDEV_BLOCK state. See also2850 * scsi_internal_device_unblock_nowait().2851 */2852int scsi_internal_device_block_nowait(struct scsi_device *sdev)2853{2854	int ret = __scsi_internal_device_block_nowait(sdev);2855 2856	/*2857	 * The device has transitioned to SDEV_BLOCK.  Stop the2858	 * block layer from calling the midlayer with this device's2859	 * request queue.2860	 */2861	if (!ret)2862		scsi_stop_queue(sdev);2863	return ret;2864}2865EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);2866 2867/**2868 * scsi_device_block - try to transition to the SDEV_BLOCK state2869 * @sdev: device to block2870 * @data: dummy argument, ignored2871 *2872 * Pause SCSI command processing on the specified device. Callers must wait2873 * until all ongoing scsi_queue_rq() calls have finished after this function2874 * returns.2875 *2876 * Note:2877 * This routine transitions the device to the SDEV_BLOCK state (which must be2878 * a legal transition). When the device is in this state, command processing2879 * is paused until the device leaves the SDEV_BLOCK state. See also2880 * scsi_internal_device_unblock().2881 */2882static void scsi_device_block(struct scsi_device *sdev, void *data)2883{2884	int err;2885	enum scsi_device_state state;2886 2887	mutex_lock(&sdev->state_mutex);2888	err = __scsi_internal_device_block_nowait(sdev);2889	state = sdev->sdev_state;2890	if (err == 0)2891		/*2892		 * scsi_stop_queue() must be called with the state_mutex2893		 * held. Otherwise a simultaneous scsi_start_queue() call2894		 * might unquiesce the queue before we quiesce it.2895		 */2896		scsi_stop_queue(sdev);2897 2898	mutex_unlock(&sdev->state_mutex);2899 2900	WARN_ONCE(err, "%s: failed to block %s in state %d\n",2901		  __func__, dev_name(&sdev->sdev_gendev), state);2902}2903 2904/**2905 * scsi_internal_device_unblock_nowait - resume a device after a block request2906 * @sdev:	device to resume2907 * @new_state:	state to set the device to after unblocking2908 *2909 * Restart the device queue for a previously suspended SCSI device. Does not2910 * sleep.2911 *2912 * Returns zero if successful or a negative error code upon failure.2913 *2914 * Notes:2915 * This routine transitions the device to the SDEV_RUNNING state or to one of2916 * the offline states (which must be a legal transition) allowing the midlayer2917 * to goose the queue for this device.2918 */2919int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,2920					enum scsi_device_state new_state)2921{2922	switch (new_state) {2923	case SDEV_RUNNING:2924	case SDEV_TRANSPORT_OFFLINE:2925		break;2926	default:2927		return -EINVAL;2928	}2929 2930	/*2931	 * Try to transition the scsi device to SDEV_RUNNING or one of the2932	 * offlined states and goose the device queue if successful.2933	 */2934	switch (sdev->sdev_state) {2935	case SDEV_BLOCK:2936	case SDEV_TRANSPORT_OFFLINE:2937		sdev->sdev_state = new_state;2938		break;2939	case SDEV_CREATED_BLOCK:2940		if (new_state == SDEV_TRANSPORT_OFFLINE ||2941		    new_state == SDEV_OFFLINE)2942			sdev->sdev_state = new_state;2943		else2944			sdev->sdev_state = SDEV_CREATED;2945		break;2946	case SDEV_CANCEL:2947	case SDEV_OFFLINE:2948		break;2949	default:2950		return -EINVAL;2951	}2952	scsi_start_queue(sdev);2953 2954	return 0;2955}2956EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);2957 2958/**2959 * scsi_internal_device_unblock - resume a device after a block request2960 * @sdev:	device to resume2961 * @new_state:	state to set the device to after unblocking2962 *2963 * Restart the device queue for a previously suspended SCSI device. May sleep.2964 *2965 * Returns zero if successful or a negative error code upon failure.2966 *2967 * Notes:2968 * This routine transitions the device to the SDEV_RUNNING state or to one of2969 * the offline states (which must be a legal transition) allowing the midlayer2970 * to goose the queue for this device.2971 */2972static int scsi_internal_device_unblock(struct scsi_device *sdev,2973					enum scsi_device_state new_state)2974{2975	int ret;2976 2977	mutex_lock(&sdev->state_mutex);2978	ret = scsi_internal_device_unblock_nowait(sdev, new_state);2979	mutex_unlock(&sdev->state_mutex);2980 2981	return ret;2982}2983 2984static int2985target_block(struct device *dev, void *data)2986{2987	if (scsi_is_target_device(dev))2988		starget_for_each_device(to_scsi_target(dev), NULL,2989					scsi_device_block);2990	return 0;2991}2992 2993/**2994 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state2995 * @dev: a parent device of one or more scsi_target devices2996 * @shost: the Scsi_Host to which this device belongs2997 *2998 * Iterate over all children of @dev, which should be scsi_target devices,2999 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for3000 * ongoing scsi_queue_rq() calls to finish. May sleep.3001 *3002 * Note:3003 * @dev must not itself be a scsi_target device.3004 */3005void3006scsi_block_targets(struct Scsi_Host *shost, struct device *dev)3007{3008	WARN_ON_ONCE(scsi_is_target_device(dev));3009	device_for_each_child(dev, NULL, target_block);3010	blk_mq_wait_quiesce_done(&shost->tag_set);3011}3012EXPORT_SYMBOL_GPL(scsi_block_targets);3013 3014static void3015device_unblock(struct scsi_device *sdev, void *data)3016{3017	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);3018}3019 3020static int3021target_unblock(struct device *dev, void *data)3022{3023	if (scsi_is_target_device(dev))3024		starget_for_each_device(to_scsi_target(dev), data,3025					device_unblock);3026	return 0;3027}3028 3029void3030scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)3031{3032	if (scsi_is_target_device(dev))3033		starget_for_each_device(to_scsi_target(dev), &new_state,3034					device_unblock);3035	else3036		device_for_each_child(dev, &new_state, target_unblock);3037}3038EXPORT_SYMBOL_GPL(scsi_target_unblock);3039 3040/**3041 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state3042 * @shost: device to block3043 *3044 * Pause SCSI command processing for all logical units associated with the SCSI3045 * host and wait until pending scsi_queue_rq() calls have finished.3046 *3047 * Returns zero if successful or a negative error code upon failure.3048 */3049int3050scsi_host_block(struct Scsi_Host *shost)3051{3052	struct scsi_device *sdev;3053	int ret;3054 3055	/*3056	 * Call scsi_internal_device_block_nowait so we can avoid3057	 * calling synchronize_rcu() for each LUN.3058	 */3059	shost_for_each_device(sdev, shost) {3060		mutex_lock(&sdev->state_mutex);3061		ret = scsi_internal_device_block_nowait(sdev);3062		mutex_unlock(&sdev->state_mutex);3063		if (ret) {3064			scsi_device_put(sdev);3065			return ret;3066		}3067	}3068 3069	/* Wait for ongoing scsi_queue_rq() calls to finish. */3070	blk_mq_wait_quiesce_done(&shost->tag_set);3071 3072	return 0;3073}3074EXPORT_SYMBOL_GPL(scsi_host_block);3075 3076int3077scsi_host_unblock(struct Scsi_Host *shost, int new_state)3078{3079	struct scsi_device *sdev;3080	int ret = 0;3081 3082	shost_for_each_device(sdev, shost) {3083		ret = scsi_internal_device_unblock(sdev, new_state);3084		if (ret) {3085			scsi_device_put(sdev);3086			break;3087		}3088	}3089	return ret;3090}3091EXPORT_SYMBOL_GPL(scsi_host_unblock);3092 3093/**3094 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt3095 * @sgl:	scatter-gather list3096 * @sg_count:	number of segments in sg3097 * @offset:	offset in bytes into sg, on return offset into the mapped area3098 * @len:	bytes to map, on return number of bytes mapped3099 *3100 * Returns virtual address of the start of the mapped page3101 */3102void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,3103			  size_t *offset, size_t *len)3104{3105	int i;3106	size_t sg_len = 0, len_complete = 0;3107	struct scatterlist *sg;3108	struct page *page;3109 3110	WARN_ON(!irqs_disabled());3111 3112	for_each_sg(sgl, sg, sg_count, i) {3113		len_complete = sg_len; /* Complete sg-entries */3114		sg_len += sg->length;3115		if (sg_len > *offset)3116			break;3117	}3118 3119	if (unlikely(i == sg_count)) {3120		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "3121			"elements %d\n",3122		       __func__, sg_len, *offset, sg_count);3123		WARN_ON(1);3124		return NULL;3125	}3126 3127	/* Offset starting from the beginning of first page in this sg-entry */3128	*offset = *offset - len_complete + sg->offset;3129 3130	/* Assumption: contiguous pages can be accessed as "page + i" */3131	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));3132	*offset &= ~PAGE_MASK;3133 3134	/* Bytes in this sg-entry from *offset to the end of the page */3135	sg_len = PAGE_SIZE - *offset;3136	if (*len > sg_len)3137		*len = sg_len;3138 3139	return kmap_atomic(page);3140}3141EXPORT_SYMBOL(scsi_kmap_atomic_sg);3142 3143/**3144 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg3145 * @virt:	virtual address to be unmapped3146 */3147void scsi_kunmap_atomic_sg(void *virt)3148{3149	kunmap_atomic(virt);3150}3151EXPORT_SYMBOL(scsi_kunmap_atomic_sg);3152 3153void sdev_disable_disk_events(struct scsi_device *sdev)3154{3155	atomic_inc(&sdev->disk_events_disable_depth);3156}3157EXPORT_SYMBOL(sdev_disable_disk_events);3158 3159void sdev_enable_disk_events(struct scsi_device *sdev)3160{3161	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))3162		return;3163	atomic_dec(&sdev->disk_events_disable_depth);3164}3165EXPORT_SYMBOL(sdev_enable_disk_events);3166 3167static unsigned char designator_prio(const unsigned char *d)3168{3169	if (d[1] & 0x30)3170		/* not associated with LUN */3171		return 0;3172 3173	if (d[3] == 0)3174		/* invalid length */3175		return 0;3176 3177	/*3178	 * Order of preference for lun descriptor:3179	 * - SCSI name string3180	 * - NAA IEEE Registered Extended3181	 * - EUI-64 based 16-byte3182	 * - EUI-64 based 12-byte3183	 * - NAA IEEE Registered3184	 * - NAA IEEE Extended3185	 * - EUI-64 based 8-byte3186	 * - SCSI name string (truncated)3187	 * - T10 Vendor ID3188	 * as longer descriptors reduce the likelyhood3189	 * of identification clashes.3190	 */3191 3192	switch (d[1] & 0xf) {3193	case 8:3194		/* SCSI name string, variable-length UTF-8 */3195		return 9;3196	case 3:3197		switch (d[4] >> 4) {3198		case 6:3199			/* NAA registered extended */3200			return 8;3201		case 5:3202			/* NAA registered */3203			return 5;3204		case 4:3205			/* NAA extended */3206			return 4;3207		case 3:3208			/* NAA locally assigned */3209			return 1;3210		default:3211			break;3212		}3213		break;3214	case 2:3215		switch (d[3]) {3216		case 16:3217			/* EUI64-based, 16 byte */3218			return 7;3219		case 12:3220			/* EUI64-based, 12 byte */3221			return 6;3222		case 8:3223			/* EUI64-based, 8 byte */3224			return 3;3225		default:3226			break;3227		}3228		break;3229	case 1:3230		/* T10 vendor ID */3231		return 1;3232	default:3233		break;3234	}3235 3236	return 0;3237}3238 3239/**3240 * scsi_vpd_lun_id - return a unique device identification3241 * @sdev: SCSI device3242 * @id:   buffer for the identification3243 * @id_len:  length of the buffer3244 *3245 * Copies a unique device identification into @id based3246 * on the information in the VPD page 0x83 of the device.3247 * The string will be formatted as a SCSI name string.3248 *3249 * Returns the length of the identification or error on failure.3250 * If the identifier is longer than the supplied buffer the actual3251 * identifier length is returned and the buffer is not zero-padded.3252 */3253int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)3254{3255	u8 cur_id_prio = 0;3256	u8 cur_id_size = 0;3257	const unsigned char *d, *cur_id_str;3258	const struct scsi_vpd *vpd_pg83;3259	int id_size = -EINVAL;3260 3261	rcu_read_lock();3262	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);3263	if (!vpd_pg83) {3264		rcu_read_unlock();3265		return -ENXIO;3266	}3267 3268	/* The id string must be at least 20 bytes + terminating NULL byte */3269	if (id_len < 21) {3270		rcu_read_unlock();3271		return -EINVAL;3272	}3273 3274	memset(id, 0, id_len);3275	for (d = vpd_pg83->data + 4;3276	     d < vpd_pg83->data + vpd_pg83->len;3277	     d += d[3] + 4) {3278		u8 prio = designator_prio(d);3279 3280		if (prio == 0 || cur_id_prio > prio)3281			continue;3282 3283		switch (d[1] & 0xf) {3284		case 0x1:3285			/* T10 Vendor ID */3286			if (cur_id_size > d[3])3287				break;3288			cur_id_prio = prio;3289			cur_id_size = d[3];3290			if (cur_id_size + 4 > id_len)3291				cur_id_size = id_len - 4;3292			cur_id_str = d + 4;3293			id_size = snprintf(id, id_len, "t10.%*pE",3294					   cur_id_size, cur_id_str);3295			break;3296		case 0x2:3297			/* EUI-64 */3298			cur_id_prio = prio;3299			cur_id_size = d[3];3300			cur_id_str = d + 4;3301			switch (cur_id_size) {3302			case 8:3303				id_size = snprintf(id, id_len,3304						   "eui.%8phN",3305						   cur_id_str);3306				break;3307			case 12:3308				id_size = snprintf(id, id_len,3309						   "eui.%12phN",3310						   cur_id_str);3311				break;3312			case 16:3313				id_size = snprintf(id, id_len,3314						   "eui.%16phN",3315						   cur_id_str);3316				break;3317			default:3318				break;3319			}3320			break;3321		case 0x3:3322			/* NAA */3323			cur_id_prio = prio;3324			cur_id_size = d[3];3325			cur_id_str = d + 4;3326			switch (cur_id_size) {3327			case 8:3328				id_size = snprintf(id, id_len,3329						   "naa.%8phN",3330						   cur_id_str);3331				break;3332			case 16:3333				id_size = snprintf(id, id_len,3334						   "naa.%16phN",3335						   cur_id_str);3336				break;3337			default:3338				break;3339			}3340			break;3341		case 0x8:3342			/* SCSI name string */3343			if (cur_id_size > d[3])3344				break;3345			/* Prefer others for truncated descriptor */3346			if (d[3] > id_len) {3347				prio = 2;3348				if (cur_id_prio > prio)3349					break;3350			}3351			cur_id_prio = prio;3352			cur_id_size = id_size = d[3];3353			cur_id_str = d + 4;3354			if (cur_id_size >= id_len)3355				cur_id_size = id_len - 1;3356			memcpy(id, cur_id_str, cur_id_size);3357			break;3358		default:3359			break;3360		}3361	}3362	rcu_read_unlock();3363 3364	return id_size;3365}3366EXPORT_SYMBOL(scsi_vpd_lun_id);3367 3368/*3369 * scsi_vpd_tpg_id - return a target port group identifier3370 * @sdev: SCSI device3371 *3372 * Returns the Target Port Group identifier from the information3373 * froom VPD page 0x83 of the device.3374 *3375 * Returns the identifier or error on failure.3376 */3377int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)3378{3379	const unsigned char *d;3380	const struct scsi_vpd *vpd_pg83;3381	int group_id = -EAGAIN, rel_port = -1;3382 3383	rcu_read_lock();3384	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);3385	if (!vpd_pg83) {3386		rcu_read_unlock();3387		return -ENXIO;3388	}3389 3390	d = vpd_pg83->data + 4;3391	while (d < vpd_pg83->data + vpd_pg83->len) {3392		switch (d[1] & 0xf) {3393		case 0x4:3394			/* Relative target port */3395			rel_port = get_unaligned_be16(&d[6]);3396			break;3397		case 0x5:3398			/* Target port group */3399			group_id = get_unaligned_be16(&d[6]);3400			break;3401		default:3402			break;3403		}3404		d += d[3] + 4;3405	}3406	rcu_read_unlock();3407 3408	if (group_id >= 0 && rel_id && rel_port != -1)3409		*rel_id = rel_port;3410 3411	return group_id;3412}3413EXPORT_SYMBOL(scsi_vpd_tpg_id);3414 3415/**3416 * scsi_build_sense - build sense data for a command3417 * @scmd:	scsi command for which the sense should be formatted3418 * @desc:	Sense format (non-zero == descriptor format,3419 *              0 == fixed format)3420 * @key:	Sense key3421 * @asc:	Additional sense code3422 * @ascq:	Additional sense code qualifier3423 *3424 **/3425void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)3426{3427	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);3428	scmd->result = SAM_STAT_CHECK_CONDITION;3429}3430EXPORT_SYMBOL_GPL(scsi_build_sense);3431 3432#ifdef CONFIG_SCSI_LIB_KUNIT_TEST3433#include "scsi_lib_test.c"3434#endif3435