Linux Audio

Check our new training course

Loading...
v5.9
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 1999 Eric Youngdale
   4 * Copyright (C) 2014 Christoph Hellwig
   5 *
   6 *  SCSI queueing library.
   7 *      Initial versions: Eric Youngdale (eric@andante.org).
   8 *                        Based upon conversations with large numbers
   9 *                        of people at Linux Expo.
  10 */
  11
  12#include <linux/bio.h>
  13#include <linux/bitops.h>
  14#include <linux/blkdev.h>
  15#include <linux/completion.h>
  16#include <linux/kernel.h>
  17#include <linux/export.h>
  18#include <linux/init.h>
  19#include <linux/pci.h>
  20#include <linux/delay.h>
  21#include <linux/hardirq.h>
  22#include <linux/scatterlist.h>
  23#include <linux/blk-mq.h>
 
  24#include <linux/ratelimit.h>
  25#include <asm/unaligned.h>
  26
  27#include <scsi/scsi.h>
  28#include <scsi/scsi_cmnd.h>
  29#include <scsi/scsi_dbg.h>
  30#include <scsi/scsi_device.h>
  31#include <scsi/scsi_driver.h>
  32#include <scsi/scsi_eh.h>
  33#include <scsi/scsi_host.h>
  34#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
  35#include <scsi/scsi_dh.h>
  36
  37#include <trace/events/scsi.h>
  38
  39#include "scsi_debugfs.h"
  40#include "scsi_priv.h"
  41#include "scsi_logging.h"
  42
  43/*
  44 * Size of integrity metadata is usually small, 1 inline sg should
  45 * cover normal cases.
  46 */
  47#ifdef CONFIG_ARCH_NO_SG_CHAIN
  48#define  SCSI_INLINE_PROT_SG_CNT  0
  49#define  SCSI_INLINE_SG_CNT  0
  50#else
  51#define  SCSI_INLINE_PROT_SG_CNT  1
  52#define  SCSI_INLINE_SG_CNT  2
  53#endif
  54
  55static struct kmem_cache *scsi_sense_cache;
  56static struct kmem_cache *scsi_sense_isadma_cache;
  57static DEFINE_MUTEX(scsi_sense_cache_mutex);
  58
  59static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd);
  60
  61static inline struct kmem_cache *
  62scsi_select_sense_cache(bool unchecked_isa_dma)
  63{
  64	return unchecked_isa_dma ? scsi_sense_isadma_cache : scsi_sense_cache;
  65}
  66
  67static void scsi_free_sense_buffer(bool unchecked_isa_dma,
  68				   unsigned char *sense_buffer)
  69{
  70	kmem_cache_free(scsi_select_sense_cache(unchecked_isa_dma),
  71			sense_buffer);
  72}
  73
  74static unsigned char *scsi_alloc_sense_buffer(bool unchecked_isa_dma,
  75	gfp_t gfp_mask, int numa_node)
  76{
  77	return kmem_cache_alloc_node(scsi_select_sense_cache(unchecked_isa_dma),
  78				     gfp_mask, numa_node);
  79}
  80
  81int scsi_init_sense_cache(struct Scsi_Host *shost)
  82{
  83	struct kmem_cache *cache;
  84	int ret = 0;
  85
  86	mutex_lock(&scsi_sense_cache_mutex);
  87	cache = scsi_select_sense_cache(shost->unchecked_isa_dma);
  88	if (cache)
  89		goto exit;
  90
  91	if (shost->unchecked_isa_dma) {
  92		scsi_sense_isadma_cache =
  93			kmem_cache_create("scsi_sense_cache(DMA)",
  94				SCSI_SENSE_BUFFERSIZE, 0,
  95				SLAB_HWCACHE_ALIGN | SLAB_CACHE_DMA, NULL);
  96		if (!scsi_sense_isadma_cache)
  97			ret = -ENOMEM;
  98	} else {
  99		scsi_sense_cache =
 100			kmem_cache_create_usercopy("scsi_sense_cache",
 101				SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN,
 102				0, SCSI_SENSE_BUFFERSIZE, NULL);
 103		if (!scsi_sense_cache)
 104			ret = -ENOMEM;
 105	}
 106 exit:
 107	mutex_unlock(&scsi_sense_cache_mutex);
 108	return ret;
 109}
 110
 111/*
 112 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
 113 * not change behaviour from the previous unplug mechanism, experimentation
 114 * may prove this needs changing.
 115 */
 116#define SCSI_QUEUE_DELAY	3
 117
 118static void
 119scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
 120{
 121	struct Scsi_Host *host = cmd->device->host;
 122	struct scsi_device *device = cmd->device;
 123	struct scsi_target *starget = scsi_target(device);
 124
 125	/*
 126	 * Set the appropriate busy bit for the device/host.
 127	 *
 128	 * If the host/device isn't busy, assume that something actually
 129	 * completed, and that we should be able to queue a command now.
 130	 *
 131	 * Note that the prior mid-layer assumption that any host could
 132	 * always queue at least one command is now broken.  The mid-layer
 133	 * will implement a user specifiable stall (see
 134	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
 135	 * if a command is requeued with no other commands outstanding
 136	 * either for the device or for the host.
 137	 */
 138	switch (reason) {
 139	case SCSI_MLQUEUE_HOST_BUSY:
 140		atomic_set(&host->host_blocked, host->max_host_blocked);
 141		break;
 142	case SCSI_MLQUEUE_DEVICE_BUSY:
 143	case SCSI_MLQUEUE_EH_RETRY:
 144		atomic_set(&device->device_blocked,
 145			   device->max_device_blocked);
 146		break;
 147	case SCSI_MLQUEUE_TARGET_BUSY:
 148		atomic_set(&starget->target_blocked,
 149			   starget->max_target_blocked);
 150		break;
 151	}
 152}
 153
 154static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd)
 155{
 156	if (cmd->request->rq_flags & RQF_DONTPREP) {
 157		cmd->request->rq_flags &= ~RQF_DONTPREP;
 
 
 158		scsi_mq_uninit_cmd(cmd);
 159	} else {
 160		WARN_ON_ONCE(true);
 161	}
 162	blk_mq_requeue_request(cmd->request, true);
 
 
 
 163}
 164
 165/**
 166 * __scsi_queue_insert - private queue insertion
 167 * @cmd: The SCSI command being requeued
 168 * @reason:  The reason for the requeue
 169 * @unbusy: Whether the queue should be unbusied
 170 *
 171 * This is a private queue insertion.  The public interface
 172 * scsi_queue_insert() always assumes the queue should be unbusied
 173 * because it's always called before the completion.  This function is
 174 * for a requeue after completion, which should only occur in this
 175 * file.
 176 */
 177static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy)
 178{
 179	struct scsi_device *device = cmd->device;
 180
 181	SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd,
 182		"Inserting command %p into mlqueue\n", cmd));
 183
 184	scsi_set_blocked(cmd, reason);
 185
 186	/*
 187	 * Decrement the counters, since these commands are no longer
 188	 * active on the host/device.
 189	 */
 190	if (unbusy)
 191		scsi_device_unbusy(device, cmd);
 192
 193	/*
 194	 * Requeue this command.  It will go before all other commands
 195	 * that are already in the queue. Schedule requeue work under
 196	 * lock such that the kblockd_schedule_work() call happens
 197	 * before blk_cleanup_queue() finishes.
 198	 */
 199	cmd->result = 0;
 200
 201	blk_mq_requeue_request(cmd->request, true);
 
 202}
 203
 204/**
 205 * scsi_queue_insert - Reinsert a command in the queue.
 206 * @cmd:    command that we are adding to queue.
 207 * @reason: why we are inserting command to queue.
 208 *
 209 * We do this for one of two cases. Either the host is busy and it cannot accept
 210 * any more commands for the time being, or the device returned QUEUE_FULL and
 211 * can accept no more commands.
 212 *
 213 * Context: This could be called either from an interrupt context or a normal
 214 * process context.
 215 */
 216void scsi_queue_insert(struct scsi_cmnd *cmd, int reason)
 217{
 218	__scsi_queue_insert(cmd, reason, true);
 219}
 220
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 221
 222/**
 223 * __scsi_execute - insert request and wait for the result
 224 * @sdev:	scsi device
 225 * @cmd:	scsi command
 226 * @data_direction: data direction
 227 * @buffer:	data buffer
 228 * @bufflen:	len of buffer
 229 * @sense:	optional sense buffer
 230 * @sshdr:	optional decoded sense header
 231 * @timeout:	request timeout in seconds
 232 * @retries:	number of times to retry request
 233 * @flags:	flags for ->cmd_flags
 234 * @rq_flags:	flags for ->rq_flags
 235 * @resid:	optional residual length
 236 *
 237 * Returns the scsi_cmnd result field if a command was executed, or a negative
 238 * Linux error code if we didn't get that far.
 239 */
 240int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd,
 241		 int data_direction, void *buffer, unsigned bufflen,
 242		 unsigned char *sense, struct scsi_sense_hdr *sshdr,
 243		 int timeout, int retries, u64 flags, req_flags_t rq_flags,
 244		 int *resid)
 245{
 
 246	struct request *req;
 247	struct scsi_request *rq;
 248	int ret = DRIVER_ERROR << 24;
 
 
 
 
 
 
 249
 250	req = blk_get_request(sdev->request_queue,
 251			data_direction == DMA_TO_DEVICE ?
 252			REQ_OP_SCSI_OUT : REQ_OP_SCSI_IN, BLK_MQ_REQ_PREEMPT);
 253	if (IS_ERR(req))
 254		return ret;
 255	rq = scsi_req(req);
 256
 257	if (bufflen &&	blk_rq_map_kern(sdev->request_queue, req,
 258					buffer, bufflen, GFP_NOIO))
 259		goto out;
 260
 261	rq->cmd_len = COMMAND_SIZE(cmd[0]);
 262	memcpy(rq->cmd, cmd, rq->cmd_len);
 263	rq->retries = retries;
 
 
 
 
 264	req->timeout = timeout;
 265	req->cmd_flags |= flags;
 266	req->rq_flags |= rq_flags | RQF_QUIET;
 267
 268	/*
 269	 * head injection *required* here otherwise quiesce won't work
 270	 */
 271	blk_execute_rq(req->q, NULL, req, 1);
 
 
 
 
 
 272
 273	/*
 274	 * Some devices (USB mass-storage in particular) may transfer
 275	 * garbage data together with a residue indicating that the data
 276	 * is invalid.  Prevent the garbage from being misinterpreted
 277	 * and prevent security leaks by zeroing out the excess data.
 278	 */
 279	if (unlikely(rq->resid_len > 0 && rq->resid_len <= bufflen))
 280		memset(buffer + (bufflen - rq->resid_len), 0, rq->resid_len);
 
 
 
 
 
 
 
 
 281
 282	if (resid)
 283		*resid = rq->resid_len;
 284	if (sense && rq->sense_len)
 285		memcpy(sense, rq->sense, SCSI_SENSE_BUFFERSIZE);
 286	if (sshdr)
 287		scsi_normalize_sense(rq->sense, rq->sense_len, sshdr);
 288	ret = rq->result;
 289 out:
 290	blk_put_request(req);
 291
 292	return ret;
 293}
 294EXPORT_SYMBOL(__scsi_execute);
 295
 296/**
 297 * scsi_init_cmd_errh - Initialize cmd fields related to error handling.
 298 * @cmd:  command that is ready to be queued.
 299 *
 300 * This function has the job of initializing a number of fields related to error
 301 * handling. Typically this will be called once for each command, as required.
 302 */
 303static void scsi_init_cmd_errh(struct scsi_cmnd *cmd)
 304{
 305	scsi_set_resid(cmd, 0);
 306	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
 307	if (cmd->cmd_len == 0)
 308		cmd->cmd_len = scsi_command_size(cmd->cmnd);
 309}
 310
 311/*
 312 * Wake up the error handler if necessary. Avoid as follows that the error
 313 * handler is not woken up if host in-flight requests number ==
 314 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
 315 * with an RCU read lock in this function to ensure that this function in
 316 * its entirety either finishes before scsi_eh_scmd_add() increases the
 317 * host_failed counter or that it notices the shost state change made by
 318 * scsi_eh_scmd_add().
 319 */
 320static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd)
 321{
 322	unsigned long flags;
 323
 324	rcu_read_lock();
 325	__clear_bit(SCMD_STATE_INFLIGHT, &cmd->state);
 326	if (unlikely(scsi_host_in_recovery(shost))) {
 
 
 327		spin_lock_irqsave(shost->host_lock, flags);
 328		if (shost->host_failed || shost->host_eh_scheduled)
 329			scsi_eh_wakeup(shost);
 330		spin_unlock_irqrestore(shost->host_lock, flags);
 331	}
 332	rcu_read_unlock();
 333}
 334
 335void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd)
 336{
 337	struct Scsi_Host *shost = sdev->host;
 338	struct scsi_target *starget = scsi_target(sdev);
 339
 340	scsi_dec_host_busy(shost, cmd);
 341
 342	if (starget->can_queue > 0)
 343		atomic_dec(&starget->target_busy);
 344
 345	atomic_dec(&sdev->device_busy);
 
 346}
 347
 348static void scsi_kick_queue(struct request_queue *q)
 
 
 
 
 349{
 350	blk_mq_run_hw_queues(q, false);
 
 
 
 351}
 352
 353/*
 354 * Called for single_lun devices on IO completion. Clear starget_sdev_user,
 355 * and call blk_run_queue for all the scsi_devices on the target -
 356 * including current_sdev first.
 357 *
 358 * Called with *no* scsi locks held.
 359 */
 360static void scsi_single_lun_run(struct scsi_device *current_sdev)
 361{
 362	struct Scsi_Host *shost = current_sdev->host;
 363	struct scsi_device *sdev, *tmp;
 364	struct scsi_target *starget = scsi_target(current_sdev);
 365	unsigned long flags;
 366
 367	spin_lock_irqsave(shost->host_lock, flags);
 368	starget->starget_sdev_user = NULL;
 369	spin_unlock_irqrestore(shost->host_lock, flags);
 370
 371	/*
 372	 * Call blk_run_queue for all LUNs on the target, starting with
 373	 * current_sdev. We race with others (to set starget_sdev_user),
 374	 * but in most cases, we will be first. Ideally, each LU on the
 375	 * target would get some limited time or requests on the target.
 376	 */
 377	scsi_kick_queue(current_sdev->request_queue);
 
 378
 379	spin_lock_irqsave(shost->host_lock, flags);
 380	if (starget->starget_sdev_user)
 381		goto out;
 382	list_for_each_entry_safe(sdev, tmp, &starget->devices,
 383			same_target_siblings) {
 384		if (sdev == current_sdev)
 385			continue;
 386		if (scsi_device_get(sdev))
 387			continue;
 388
 389		spin_unlock_irqrestore(shost->host_lock, flags);
 390		scsi_kick_queue(sdev->request_queue);
 391		spin_lock_irqsave(shost->host_lock, flags);
 392
 393		scsi_device_put(sdev);
 394	}
 395 out:
 396	spin_unlock_irqrestore(shost->host_lock, flags);
 397}
 398
 399static inline bool scsi_device_is_busy(struct scsi_device *sdev)
 400{
 401	if (atomic_read(&sdev->device_busy) >= sdev->queue_depth)
 402		return true;
 403	if (atomic_read(&sdev->device_blocked) > 0)
 404		return true;
 405	return false;
 406}
 407
 408static inline bool scsi_target_is_busy(struct scsi_target *starget)
 409{
 410	if (starget->can_queue > 0) {
 411		if (atomic_read(&starget->target_busy) >= starget->can_queue)
 412			return true;
 413		if (atomic_read(&starget->target_blocked) > 0)
 414			return true;
 415	}
 416	return false;
 417}
 418
 419static inline bool scsi_host_is_busy(struct Scsi_Host *shost)
 420{
 421	if (atomic_read(&shost->host_blocked) > 0)
 422		return true;
 423	if (shost->host_self_blocked)
 424		return true;
 425	return false;
 426}
 427
 428static void scsi_starved_list_run(struct Scsi_Host *shost)
 429{
 430	LIST_HEAD(starved_list);
 431	struct scsi_device *sdev;
 432	unsigned long flags;
 433
 434	spin_lock_irqsave(shost->host_lock, flags);
 435	list_splice_init(&shost->starved_list, &starved_list);
 436
 437	while (!list_empty(&starved_list)) {
 438		struct request_queue *slq;
 439
 440		/*
 441		 * As long as shost is accepting commands and we have
 442		 * starved queues, call blk_run_queue. scsi_request_fn
 443		 * drops the queue_lock and can add us back to the
 444		 * starved_list.
 445		 *
 446		 * host_lock protects the starved_list and starved_entry.
 447		 * scsi_request_fn must get the host_lock before checking
 448		 * or modifying starved_list or starved_entry.
 449		 */
 450		if (scsi_host_is_busy(shost))
 451			break;
 452
 453		sdev = list_entry(starved_list.next,
 454				  struct scsi_device, starved_entry);
 455		list_del_init(&sdev->starved_entry);
 456		if (scsi_target_is_busy(scsi_target(sdev))) {
 457			list_move_tail(&sdev->starved_entry,
 458				       &shost->starved_list);
 459			continue;
 460		}
 461
 462		/*
 463		 * Once we drop the host lock, a racing scsi_remove_device()
 464		 * call may remove the sdev from the starved list and destroy
 465		 * it and the queue.  Mitigate by taking a reference to the
 466		 * queue and never touching the sdev again after we drop the
 467		 * host lock.  Note: if __scsi_remove_device() invokes
 468		 * blk_cleanup_queue() before the queue is run from this
 469		 * function then blk_run_queue() will return immediately since
 470		 * blk_cleanup_queue() marks the queue with QUEUE_FLAG_DYING.
 471		 */
 472		slq = sdev->request_queue;
 473		if (!blk_get_queue(slq))
 474			continue;
 475		spin_unlock_irqrestore(shost->host_lock, flags);
 476
 477		scsi_kick_queue(slq);
 478		blk_put_queue(slq);
 479
 480		spin_lock_irqsave(shost->host_lock, flags);
 481	}
 482	/* put any unprocessed entries back */
 483	list_splice(&starved_list, &shost->starved_list);
 484	spin_unlock_irqrestore(shost->host_lock, flags);
 485}
 486
 487/**
 488 * scsi_run_queue - Select a proper request queue to serve next.
 489 * @q:  last request's queue
 490 *
 491 * The previous command was completely finished, start a new one if possible.
 492 */
 493static void scsi_run_queue(struct request_queue *q)
 494{
 495	struct scsi_device *sdev = q->queuedata;
 496
 497	if (scsi_target(sdev)->single_lun)
 498		scsi_single_lun_run(sdev);
 499	if (!list_empty(&sdev->host->starved_list))
 500		scsi_starved_list_run(sdev->host);
 501
 502	blk_mq_run_hw_queues(q, false);
 
 503}
 504
 505void scsi_requeue_run_queue(struct work_struct *work)
 506{
 507	struct scsi_device *sdev;
 508	struct request_queue *q;
 509
 510	sdev = container_of(work, struct scsi_device, requeue_work);
 511	q = sdev->request_queue;
 512	scsi_run_queue(q);
 513}
 514
 515void scsi_run_host_queues(struct Scsi_Host *shost)
 516{
 517	struct scsi_device *sdev;
 518
 519	shost_for_each_device(sdev, shost)
 520		scsi_run_queue(sdev->request_queue);
 521}
 522
 523static void scsi_uninit_cmd(struct scsi_cmnd *cmd)
 524{
 525	if (!blk_rq_is_passthrough(cmd->request)) {
 526		struct scsi_driver *drv = scsi_cmd_to_driver(cmd);
 527
 528		if (drv->uninit_command)
 529			drv->uninit_command(cmd);
 530	}
 531}
 532
 533static void scsi_free_sgtables(struct scsi_cmnd *cmd)
 534{
 535	if (cmd->sdb.table.nents)
 536		sg_free_table_chained(&cmd->sdb.table,
 537				SCSI_INLINE_SG_CNT);
 538	if (scsi_prot_sg_count(cmd))
 539		sg_free_table_chained(&cmd->prot_sdb->table,
 540				SCSI_INLINE_PROT_SG_CNT);
 541}
 
 542
 543static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd)
 544{
 545	scsi_free_sgtables(cmd);
 546	scsi_uninit_cmd(cmd);
 547}
 548
 549static void scsi_run_queue_async(struct scsi_device *sdev)
 550{
 
 
 
 551	if (scsi_target(sdev)->single_lun ||
 552	    !list_empty(&sdev->host->starved_list))
 553		kblockd_schedule_work(&sdev->requeue_work);
 554	else
 555		blk_mq_run_hw_queues(sdev->request_queue, true);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 556}
 557
 558/* Returns false when no more bytes to process, true if there are more */
 559static bool scsi_end_request(struct request *req, blk_status_t error,
 560		unsigned int bytes)
 561{
 562	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
 563	struct scsi_device *sdev = cmd->device;
 564	struct request_queue *q = sdev->request_queue;
 565
 566	if (blk_update_request(req, error, bytes))
 567		return true;
 568
 
 569	if (blk_queue_add_random(q))
 570		add_disk_randomness(req->rq_disk);
 571
 572	if (!blk_rq_is_scsi(req)) {
 573		WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED));
 574		cmd->flags &= ~SCMD_INITIALIZED;
 575	}
 576
 577	/*
 578	 * Calling rcu_barrier() is not necessary here because the
 579	 * SCSI error handler guarantees that the function called by
 580	 * call_rcu() has been called before scsi_end_request() is
 581	 * called.
 582	 */
 583	destroy_rcu_head(&cmd->rcu);
 584
 585	/*
 586	 * In the MQ case the command gets freed by __blk_mq_end_request,
 587	 * so we have to do all cleanup that depends on it earlier.
 588	 *
 589	 * We also can't kick the queues from irq context, so we
 590	 * will have to defer it to a workqueue.
 591	 */
 592	scsi_mq_uninit_cmd(cmd);
 593
 594	/*
 595	 * queue is still alive, so grab the ref for preventing it
 596	 * from being cleaned up during running queue.
 597	 */
 598	percpu_ref_get(&q->q_usage_counter);
 599
 600	__blk_mq_end_request(req, error);
 601
 602	scsi_run_queue_async(sdev);
 603
 604	percpu_ref_put(&q->q_usage_counter);
 605	return false;
 606}
 607
 608/**
 609 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
 610 * @cmd:	SCSI command
 611 * @result:	scsi error code
 612 *
 613 * Translate a SCSI result code into a blk_status_t value. May reset the host
 614 * byte of @cmd->result.
 615 */
 616static blk_status_t scsi_result_to_blk_status(struct scsi_cmnd *cmd, int result)
 617{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 618	switch (host_byte(result)) {
 619	case DID_OK:
 620		/*
 621		 * Also check the other bytes than the status byte in result
 622		 * to handle the case when a SCSI LLD sets result to
 623		 * DRIVER_SENSE << 24 without setting SAM_STAT_CHECK_CONDITION.
 624		 */
 625		if (scsi_status_is_good(result) && (result & ~0xff) == 0)
 626			return BLK_STS_OK;
 627		return BLK_STS_IOERR;
 628	case DID_TRANSPORT_FAILFAST:
 
 629		return BLK_STS_TRANSPORT;
 630	case DID_TARGET_FAILURE:
 631		set_host_byte(cmd, DID_OK);
 632		return BLK_STS_TARGET;
 633	case DID_NEXUS_FAILURE:
 634		set_host_byte(cmd, DID_OK);
 635		return BLK_STS_NEXUS;
 636	case DID_ALLOC_FAILURE:
 637		set_host_byte(cmd, DID_OK);
 638		return BLK_STS_NOSPC;
 639	case DID_MEDIUM_ERROR:
 640		set_host_byte(cmd, DID_OK);
 641		return BLK_STS_MEDIUM;
 642	default:
 643		return BLK_STS_IOERR;
 644	}
 645}
 646
 647/* Helper for scsi_io_completion() when "reprep" action required. */
 648static void scsi_io_completion_reprep(struct scsi_cmnd *cmd,
 649				      struct request_queue *q)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 650{
 651	/* A new command will be prepared and issued. */
 652	scsi_mq_requeue_cmd(cmd);
 
 
 
 
 
 
 
 
 
 
 
 653}
 654
 
 
 
 
 
 
 
 
 655/* Helper for scsi_io_completion() when special action required. */
 656static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
 657{
 658	struct request_queue *q = cmd->device->request_queue;
 659	struct request *req = cmd->request;
 660	int level = 0;
 661	enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY,
 662	      ACTION_DELAYED_RETRY} action;
 663	unsigned long wait_for = (cmd->allowed + 1) * req->timeout;
 664	struct scsi_sense_hdr sshdr;
 665	bool sense_valid;
 666	bool sense_current = true;      /* false implies "deferred sense" */
 667	blk_status_t blk_stat;
 668
 669	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
 670	if (sense_valid)
 671		sense_current = !scsi_sense_is_deferred(&sshdr);
 672
 673	blk_stat = scsi_result_to_blk_status(cmd, result);
 674
 675	if (host_byte(result) == DID_RESET) {
 676		/* Third party bus reset or reset for error recovery
 677		 * reasons.  Just retry the command and see what
 678		 * happens.
 679		 */
 680		action = ACTION_RETRY;
 681	} else if (sense_valid && sense_current) {
 682		switch (sshdr.sense_key) {
 683		case UNIT_ATTENTION:
 684			if (cmd->device->removable) {
 685				/* Detected disc change.  Set a bit
 686				 * and quietly refuse further access.
 687				 */
 688				cmd->device->changed = 1;
 689				action = ACTION_FAIL;
 690			} else {
 691				/* Must have been a power glitch, or a
 692				 * bus reset.  Could not have been a
 693				 * media change, so we just retry the
 694				 * command and see what happens.
 695				 */
 696				action = ACTION_RETRY;
 697			}
 698			break;
 699		case ILLEGAL_REQUEST:
 700			/* If we had an ILLEGAL REQUEST returned, then
 701			 * we may have performed an unsupported
 702			 * command.  The only thing this should be
 703			 * would be a ten byte read where only a six
 704			 * byte read was supported.  Also, on a system
 705			 * where READ CAPACITY failed, we may have
 706			 * read past the end of the disk.
 707			 */
 708			if ((cmd->device->use_10_for_rw &&
 709			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
 710			    (cmd->cmnd[0] == READ_10 ||
 711			     cmd->cmnd[0] == WRITE_10)) {
 712				/* This will issue a new 6-byte command. */
 713				cmd->device->use_10_for_rw = 0;
 714				action = ACTION_REPREP;
 715			} else if (sshdr.asc == 0x10) /* DIX */ {
 716				action = ACTION_FAIL;
 717				blk_stat = BLK_STS_PROTECTION;
 718			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
 719			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
 720				action = ACTION_FAIL;
 721				blk_stat = BLK_STS_TARGET;
 722			} else
 723				action = ACTION_FAIL;
 724			break;
 725		case ABORTED_COMMAND:
 726			action = ACTION_FAIL;
 727			if (sshdr.asc == 0x10) /* DIF */
 728				blk_stat = BLK_STS_PROTECTION;
 729			break;
 730		case NOT_READY:
 731			/* If the device is in the process of becoming
 732			 * ready, or has a temporary blockage, retry.
 733			 */
 734			if (sshdr.asc == 0x04) {
 735				switch (sshdr.ascq) {
 736				case 0x01: /* becoming ready */
 737				case 0x04: /* format in progress */
 738				case 0x05: /* rebuild in progress */
 739				case 0x06: /* recalculation in progress */
 740				case 0x07: /* operation in progress */
 741				case 0x08: /* Long write in progress */
 742				case 0x09: /* self test in progress */
 
 743				case 0x14: /* space allocation in progress */
 744				case 0x1a: /* start stop unit in progress */
 745				case 0x1b: /* sanitize in progress */
 746				case 0x1d: /* configuration in progress */
 747				case 0x24: /* depopulation in progress */
 
 748					action = ACTION_DELAYED_RETRY;
 749					break;
 
 
 
 750				default:
 751					action = ACTION_FAIL;
 752					break;
 753				}
 754			} else
 755				action = ACTION_FAIL;
 756			break;
 757		case VOLUME_OVERFLOW:
 758			/* See SSC3rXX or current. */
 759			action = ACTION_FAIL;
 760			break;
 
 
 
 
 
 
 
 
 
 
 
 761		default:
 762			action = ACTION_FAIL;
 763			break;
 764		}
 765	} else
 766		action = ACTION_FAIL;
 767
 768	if (action != ACTION_FAIL &&
 769	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies))
 770		action = ACTION_FAIL;
 771
 772	switch (action) {
 773	case ACTION_FAIL:
 774		/* Give up and fail the remainder of the request */
 775		if (!(req->rq_flags & RQF_QUIET)) {
 776			static DEFINE_RATELIMIT_STATE(_rs,
 777					DEFAULT_RATELIMIT_INTERVAL,
 778					DEFAULT_RATELIMIT_BURST);
 779
 780			if (unlikely(scsi_logging_level))
 781				level =
 782				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
 783						    SCSI_LOG_MLCOMPLETE_BITS);
 784
 785			/*
 786			 * if logging is enabled the failure will be printed
 787			 * in scsi_log_completion(), so avoid duplicate messages
 788			 */
 789			if (!level && __ratelimit(&_rs)) {
 790				scsi_print_result(cmd, NULL, FAILED);
 791				if (driver_byte(result) == DRIVER_SENSE)
 792					scsi_print_sense(cmd);
 793				scsi_print_command(cmd);
 794			}
 795		}
 796		if (!scsi_end_request(req, blk_stat, blk_rq_err_bytes(req)))
 797			return;
 798		fallthrough;
 799	case ACTION_REPREP:
 800		scsi_io_completion_reprep(cmd, q);
 
 
 
 801		break;
 802	case ACTION_RETRY:
 803		/* Retry the same command immediately */
 804		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
 805		break;
 806	case ACTION_DELAYED_RETRY:
 807		/* Retry the same command after a delay */
 808		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
 809		break;
 810	}
 811}
 812
 813/*
 814 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
 815 * new result that may suppress further error checking. Also modifies
 816 * *blk_statp in some cases.
 817 */
 818static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
 819					blk_status_t *blk_statp)
 820{
 821	bool sense_valid;
 822	bool sense_current = true;	/* false implies "deferred sense" */
 823	struct request *req = cmd->request;
 824	struct scsi_sense_hdr sshdr;
 825
 826	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
 827	if (sense_valid)
 828		sense_current = !scsi_sense_is_deferred(&sshdr);
 829
 830	if (blk_rq_is_passthrough(req)) {
 831		if (sense_valid) {
 832			/*
 833			 * SG_IO wants current and deferred errors
 834			 */
 835			scsi_req(req)->sense_len =
 836				min(8 + cmd->sense_buffer[7],
 837				    SCSI_SENSE_BUFFERSIZE);
 838		}
 839		if (sense_current)
 840			*blk_statp = scsi_result_to_blk_status(cmd, result);
 841	} else if (blk_rq_bytes(req) == 0 && sense_current) {
 842		/*
 843		 * Flush commands do not transfers any data, and thus cannot use
 844		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
 845		 * This sets *blk_statp explicitly for the problem case.
 846		 */
 847		*blk_statp = scsi_result_to_blk_status(cmd, result);
 848	}
 849	/*
 850	 * Recovered errors need reporting, but they're always treated as
 851	 * success, so fiddle the result code here.  For passthrough requests
 852	 * we already took a copy of the original into sreq->result which
 853	 * is what gets returned to the user
 854	 */
 855	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
 856		bool do_print = true;
 857		/*
 858		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
 859		 * skip print since caller wants ATA registers. Only occurs
 860		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
 861		 */
 862		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
 863			do_print = false;
 864		else if (req->rq_flags & RQF_QUIET)
 865			do_print = false;
 866		if (do_print)
 867			scsi_print_sense(cmd);
 868		result = 0;
 869		/* for passthrough, *blk_statp may be set */
 870		*blk_statp = BLK_STS_OK;
 871	}
 872	/*
 873	 * Another corner case: the SCSI status byte is non-zero but 'good'.
 874	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
 875	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
 876	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
 877	 * intermediate statuses (both obsolete in SAM-4) as good.
 878	 */
 879	if (status_byte(result) && scsi_status_is_good(result)) {
 880		result = 0;
 881		*blk_statp = BLK_STS_OK;
 882	}
 883	return result;
 884}
 885
 886/**
 887 * scsi_io_completion - Completion processing for SCSI commands.
 888 * @cmd:	command that is finished.
 889 * @good_bytes:	number of processed bytes.
 890 *
 891 * We will finish off the specified number of sectors. If we are done, the
 892 * command block will be released and the queue function will be goosed. If we
 893 * are not done then we have to figure out what to do next:
 894 *
 895 *   a) We can call scsi_io_completion_reprep().  The request will be
 896 *	unprepared and put back on the queue.  Then a new command will
 897 *	be created for it.  This should be used if we made forward
 898 *	progress, or if we want to switch from READ(10) to READ(6) for
 899 *	example.
 900 *
 901 *   b) We can call scsi_io_completion_action().  The request will be
 902 *	put back on the queue and retried using the same command as
 903 *	before, possibly after a delay.
 904 *
 905 *   c) We can call scsi_end_request() with blk_stat other than
 906 *	BLK_STS_OK, to fail the remainder of the request.
 907 */
 908void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
 909{
 910	int result = cmd->result;
 911	struct request_queue *q = cmd->device->request_queue;
 912	struct request *req = cmd->request;
 913	blk_status_t blk_stat = BLK_STS_OK;
 914
 915	if (unlikely(result))	/* a nz result may or may not be an error */
 916		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
 917
 918	if (unlikely(blk_rq_is_passthrough(req))) {
 919		/*
 920		 * scsi_result_to_blk_status may have reset the host_byte
 921		 */
 922		scsi_req(req)->result = cmd->result;
 923	}
 924
 925	/*
 926	 * Next deal with any sectors which we were able to correctly
 927	 * handle.
 928	 */
 929	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
 930		"%u sectors total, %d bytes done.\n",
 931		blk_rq_sectors(req), good_bytes));
 932
 933	/*
 934	 * Failed, zero length commands always need to drop down
 935	 * to retry code. Fast path should return in this block.
 936	 */
 937	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
 938		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
 939			return; /* no bytes remaining */
 940	}
 941
 942	/* Kill remainder if no retries. */
 943	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
 944		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
 945			WARN_ONCE(true,
 946			    "Bytes remaining after failed, no-retry command");
 947		return;
 948	}
 949
 950	/*
 951	 * If there had been no error, but we have leftover bytes in the
 952	 * requeues just queue the command up again.
 953	 */
 954	if (likely(result == 0))
 955		scsi_io_completion_reprep(cmd, q);
 956	else
 957		scsi_io_completion_action(cmd, result);
 958}
 959
 960static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
 961		struct request *rq)
 962{
 963	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
 964	       !op_is_write(req_op(rq)) &&
 965	       sdev->host->hostt->dma_need_drain(rq);
 966}
 967
 968/**
 969 * scsi_init_io - SCSI I/O initialization function.
 970 * @cmd:  command descriptor we wish to initialize
 
 
 
 971 *
 972 * Returns:
 973 * * BLK_STS_OK       - on success
 974 * * BLK_STS_RESOURCE - if the failure is retryable
 975 * * BLK_STS_IOERR    - if the failure is fatal
 976 */
 977blk_status_t scsi_init_io(struct scsi_cmnd *cmd)
 978{
 979	struct scsi_device *sdev = cmd->device;
 980	struct request *rq = cmd->request;
 981	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
 982	struct scatterlist *last_sg = NULL;
 983	blk_status_t ret;
 984	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
 985	int count;
 986
 987	if (WARN_ON_ONCE(!nr_segs))
 988		return BLK_STS_IOERR;
 989
 990	/*
 991	 * Make sure there is space for the drain.  The driver must adjust
 992	 * max_hw_segments to be prepared for this.
 993	 */
 994	if (need_drain)
 995		nr_segs++;
 996
 997	/*
 998	 * If sg table allocation fails, requeue request later.
 999	 */
1000	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1001			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1002		return BLK_STS_RESOURCE;
1003
1004	/*
1005	 * Next, walk the list, and fill in the addresses and sizes of
1006	 * each segment.
1007	 */
1008	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1009
1010	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1011		unsigned int pad_len =
1012			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1013
1014		last_sg->length += pad_len;
1015		cmd->extra_len += pad_len;
1016	}
1017
1018	if (need_drain) {
1019		sg_unmark_end(last_sg);
1020		last_sg = sg_next(last_sg);
1021		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1022		sg_mark_end(last_sg);
1023
1024		cmd->extra_len += sdev->dma_drain_len;
1025		count++;
1026	}
1027
1028	BUG_ON(count > cmd->sdb.table.nents);
1029	cmd->sdb.table.nents = count;
1030	cmd->sdb.length = blk_rq_payload_bytes(rq);
1031
1032	if (blk_integrity_rq(rq)) {
1033		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1034		int ivecs;
1035
1036		if (WARN_ON_ONCE(!prot_sdb)) {
1037			/*
1038			 * This can happen if someone (e.g. multipath)
1039			 * queues a command to a device on an adapter
1040			 * that does not support DIX.
1041			 */
1042			ret = BLK_STS_IOERR;
1043			goto out_free_sgtables;
1044		}
1045
1046		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1047
1048		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1049				prot_sdb->table.sgl,
1050				SCSI_INLINE_PROT_SG_CNT)) {
1051			ret = BLK_STS_RESOURCE;
1052			goto out_free_sgtables;
1053		}
1054
1055		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1056						prot_sdb->table.sgl);
1057		BUG_ON(count > ivecs);
1058		BUG_ON(count > queue_max_integrity_segments(rq->q));
1059
1060		cmd->prot_sdb = prot_sdb;
1061		cmd->prot_sdb->table.nents = count;
1062	}
1063
1064	return BLK_STS_OK;
1065out_free_sgtables:
1066	scsi_free_sgtables(cmd);
1067	return ret;
1068}
1069EXPORT_SYMBOL(scsi_init_io);
1070
1071/**
1072 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1073 * @rq: Request associated with the SCSI command to be initialized.
1074 *
1075 * This function initializes the members of struct scsi_cmnd that must be
1076 * initialized before request processing starts and that won't be
1077 * reinitialized if a SCSI command is requeued.
1078 *
1079 * Called from inside blk_get_request() for pass-through requests and from
1080 * inside scsi_init_command() for filesystem requests.
1081 */
1082static void scsi_initialize_rq(struct request *rq)
1083{
1084	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1085
1086	scsi_req_init(&cmd->req);
 
 
1087	init_rcu_head(&cmd->rcu);
1088	cmd->jiffies_at_alloc = jiffies;
1089	cmd->retries = 0;
1090}
1091
 
 
 
 
 
 
 
 
 
 
 
 
1092/*
1093 * Only called when the request isn't completed by SCSI, and not freed by
1094 * SCSI
1095 */
1096static void scsi_cleanup_rq(struct request *rq)
1097{
1098	if (rq->rq_flags & RQF_DONTPREP) {
1099		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1100		rq->rq_flags &= ~RQF_DONTPREP;
1101	}
1102}
1103
1104/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1105void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1106{
1107	void *buf = cmd->sense_buffer;
1108	void *prot = cmd->prot_sdb;
1109	struct request *rq = blk_mq_rq_from_pdu(cmd);
1110	unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS;
1111	unsigned long jiffies_at_alloc;
1112	int retries, to_clear;
1113	bool in_flight;
1114
1115	if (!blk_rq_is_scsi(rq) && !(flags & SCMD_INITIALIZED)) {
1116		flags |= SCMD_INITIALIZED;
1117		scsi_initialize_rq(rq);
1118	}
1119
1120	jiffies_at_alloc = cmd->jiffies_at_alloc;
1121	retries = cmd->retries;
1122	in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1123	/*
1124	 * Zero out the cmd, except for the embedded scsi_request. Only clear
1125	 * the driver-private command data if the LLD does not supply a
1126	 * function to initialize that data.
1127	 */
1128	to_clear = sizeof(*cmd) - sizeof(cmd->req);
1129	if (!dev->host->hostt->init_cmd_priv)
1130		to_clear += dev->host->hostt->cmd_size;
1131	memset((char *)cmd + sizeof(cmd->req), 0, to_clear);
1132
1133	cmd->device = dev;
1134	cmd->sense_buffer = buf;
1135	cmd->prot_sdb = prot;
1136	cmd->flags = flags;
1137	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
1138	cmd->jiffies_at_alloc = jiffies_at_alloc;
1139	cmd->retries = retries;
1140	if (in_flight)
1141		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1142
1143}
1144
1145static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1146		struct request *req)
1147{
1148	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1149
1150	/*
1151	 * Passthrough requests may transfer data, in which case they must
1152	 * a bio attached to them.  Or they might contain a SCSI command
1153	 * that does not transfer data, in which case they may optionally
1154	 * submit a request without an attached bio.
1155	 */
1156	if (req->bio) {
1157		blk_status_t ret = scsi_init_io(cmd);
1158		if (unlikely(ret != BLK_STS_OK))
1159			return ret;
1160	} else {
1161		BUG_ON(blk_rq_bytes(req));
1162
1163		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1164	}
1165
1166	cmd->cmd_len = scsi_req(req)->cmd_len;
1167	cmd->cmnd = scsi_req(req)->cmd;
1168	cmd->transfersize = blk_rq_bytes(req);
1169	cmd->allowed = scsi_req(req)->retries;
1170	return BLK_STS_OK;
1171}
1172
1173/*
1174 * Setup a normal block command.  These are simple request from filesystems
1175 * that still need to be translated to SCSI CDBs from the ULD.
1176 */
1177static blk_status_t scsi_setup_fs_cmnd(struct scsi_device *sdev,
1178		struct request *req)
1179{
1180	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1181
1182	if (unlikely(sdev->handler && sdev->handler->prep_fn)) {
1183		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1184		if (ret != BLK_STS_OK)
1185			return ret;
1186	}
1187
1188	cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd;
1189	memset(cmd->cmnd, 0, BLK_MAX_CDB);
1190	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1191}
1192
1193static blk_status_t scsi_setup_cmnd(struct scsi_device *sdev,
1194		struct request *req)
1195{
1196	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1197	blk_status_t ret;
1198
1199	if (!blk_rq_bytes(req))
1200		cmd->sc_data_direction = DMA_NONE;
1201	else if (rq_data_dir(req) == WRITE)
1202		cmd->sc_data_direction = DMA_TO_DEVICE;
1203	else
1204		cmd->sc_data_direction = DMA_FROM_DEVICE;
1205
1206	if (blk_rq_is_scsi(req))
1207		ret = scsi_setup_scsi_cmnd(sdev, req);
1208	else
1209		ret = scsi_setup_fs_cmnd(sdev, req);
1210
1211	if (ret != BLK_STS_OK)
1212		scsi_free_sgtables(cmd);
1213
1214	return ret;
1215}
1216
1217static blk_status_t
1218scsi_prep_state_check(struct scsi_device *sdev, struct request *req)
1219{
1220	switch (sdev->sdev_state) {
 
 
1221	case SDEV_OFFLINE:
1222	case SDEV_TRANSPORT_OFFLINE:
1223		/*
1224		 * If the device is offline we refuse to process any
1225		 * commands.  The device must be brought online
1226		 * before trying any recovery commands.
1227		 */
1228		if (!sdev->offline_already) {
1229			sdev->offline_already = true;
1230			sdev_printk(KERN_ERR, sdev,
1231				    "rejecting I/O to offline device\n");
1232		}
1233		return BLK_STS_IOERR;
1234	case SDEV_DEL:
1235		/*
1236		 * If the device is fully deleted, we refuse to
1237		 * process any commands as well.
1238		 */
1239		sdev_printk(KERN_ERR, sdev,
1240			    "rejecting I/O to dead device\n");
1241		return BLK_STS_IOERR;
1242	case SDEV_BLOCK:
1243	case SDEV_CREATED_BLOCK:
1244		return BLK_STS_RESOURCE;
1245	case SDEV_QUIESCE:
1246		/*
1247		 * If the devices is blocked we defer normal commands.
 
1248		 */
1249		if (req && !(req->rq_flags & RQF_PREEMPT))
1250			return BLK_STS_RESOURCE;
1251		return BLK_STS_OK;
1252	default:
1253		/*
1254		 * For any other not fully online state we only allow
1255		 * special commands.  In particular any user initiated
1256		 * command is not allowed.
1257		 */
1258		if (req && !(req->rq_flags & RQF_PREEMPT))
1259			return BLK_STS_IOERR;
1260		return BLK_STS_OK;
1261	}
1262}
1263
1264/*
1265 * scsi_dev_queue_ready: if we can send requests to sdev, return 1 else
1266 * return 0.
1267 *
1268 * Called with the queue_lock held.
1269 */
1270static inline int scsi_dev_queue_ready(struct request_queue *q,
1271				  struct scsi_device *sdev)
1272{
1273	unsigned int busy;
1274
1275	busy = atomic_inc_return(&sdev->device_busy) - 1;
1276	if (atomic_read(&sdev->device_blocked)) {
1277		if (busy)
1278			goto out_dec;
1279
1280		/*
1281		 * unblock after device_blocked iterates to zero
1282		 */
1283		if (atomic_dec_return(&sdev->device_blocked) > 0)
1284			goto out_dec;
1285		SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1286				   "unblocking device at zero depth\n"));
 
 
 
 
1287	}
1288
1289	if (busy >= sdev->queue_depth)
1290		goto out_dec;
1291
1292	return 1;
1293out_dec:
1294	atomic_dec(&sdev->device_busy);
1295	return 0;
1296}
1297
1298/*
1299 * scsi_target_queue_ready: checks if there we can send commands to target
1300 * @sdev: scsi device on starget to check.
1301 */
1302static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1303					   struct scsi_device *sdev)
1304{
1305	struct scsi_target *starget = scsi_target(sdev);
1306	unsigned int busy;
1307
1308	if (starget->single_lun) {
1309		spin_lock_irq(shost->host_lock);
1310		if (starget->starget_sdev_user &&
1311		    starget->starget_sdev_user != sdev) {
1312			spin_unlock_irq(shost->host_lock);
1313			return 0;
1314		}
1315		starget->starget_sdev_user = sdev;
1316		spin_unlock_irq(shost->host_lock);
1317	}
1318
1319	if (starget->can_queue <= 0)
1320		return 1;
1321
1322	busy = atomic_inc_return(&starget->target_busy) - 1;
1323	if (atomic_read(&starget->target_blocked) > 0) {
1324		if (busy)
1325			goto starved;
1326
1327		/*
1328		 * unblock after target_blocked iterates to zero
1329		 */
1330		if (atomic_dec_return(&starget->target_blocked) > 0)
1331			goto out_dec;
1332
1333		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1334				 "unblocking target at zero depth\n"));
1335	}
1336
1337	if (busy >= starget->can_queue)
1338		goto starved;
1339
1340	return 1;
1341
1342starved:
1343	spin_lock_irq(shost->host_lock);
1344	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1345	spin_unlock_irq(shost->host_lock);
1346out_dec:
1347	if (starget->can_queue > 0)
1348		atomic_dec(&starget->target_busy);
1349	return 0;
1350}
1351
1352/*
1353 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1354 * return 0. We must end up running the queue again whenever 0 is
1355 * returned, else IO can hang.
1356 */
1357static inline int scsi_host_queue_ready(struct request_queue *q,
1358				   struct Scsi_Host *shost,
1359				   struct scsi_device *sdev,
1360				   struct scsi_cmnd *cmd)
1361{
1362	if (scsi_host_in_recovery(shost))
1363		return 0;
1364
1365	if (atomic_read(&shost->host_blocked) > 0) {
1366		if (scsi_host_busy(shost) > 0)
1367			goto starved;
1368
1369		/*
1370		 * unblock after host_blocked iterates to zero
1371		 */
1372		if (atomic_dec_return(&shost->host_blocked) > 0)
1373			goto out_dec;
1374
1375		SCSI_LOG_MLQUEUE(3,
1376			shost_printk(KERN_INFO, shost,
1377				     "unblocking host at zero depth\n"));
1378	}
1379
1380	if (shost->host_self_blocked)
1381		goto starved;
1382
1383	/* We're OK to process the command, so we can't be starved */
1384	if (!list_empty(&sdev->starved_entry)) {
1385		spin_lock_irq(shost->host_lock);
1386		if (!list_empty(&sdev->starved_entry))
1387			list_del_init(&sdev->starved_entry);
1388		spin_unlock_irq(shost->host_lock);
1389	}
1390
1391	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1392
1393	return 1;
1394
1395starved:
1396	spin_lock_irq(shost->host_lock);
1397	if (list_empty(&sdev->starved_entry))
1398		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1399	spin_unlock_irq(shost->host_lock);
1400out_dec:
1401	scsi_dec_host_busy(shost, cmd);
1402	return 0;
1403}
1404
1405/*
1406 * Busy state exporting function for request stacking drivers.
1407 *
1408 * For efficiency, no lock is taken to check the busy state of
1409 * shost/starget/sdev, since the returned value is not guaranteed and
1410 * may be changed after request stacking drivers call the function,
1411 * regardless of taking lock or not.
1412 *
1413 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1414 * needs to return 'not busy'. Otherwise, request stacking drivers
1415 * may hold requests forever.
1416 */
1417static bool scsi_mq_lld_busy(struct request_queue *q)
1418{
1419	struct scsi_device *sdev = q->queuedata;
1420	struct Scsi_Host *shost;
1421
1422	if (blk_queue_dying(q))
1423		return false;
1424
1425	shost = sdev->host;
1426
1427	/*
1428	 * Ignore host/starget busy state.
1429	 * Since block layer does not have a concept of fairness across
1430	 * multiple queues, congestion of host/starget needs to be handled
1431	 * in SCSI layer.
1432	 */
1433	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1434		return true;
1435
1436	return false;
1437}
1438
1439static void scsi_softirq_done(struct request *rq)
 
 
 
 
1440{
1441	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1442	unsigned long wait_for = (cmd->allowed + 1) * rq->timeout;
1443	int disposition;
1444
1445	INIT_LIST_HEAD(&cmd->eh_entry);
1446
1447	atomic_inc(&cmd->device->iodone_cnt);
1448	if (cmd->result)
1449		atomic_inc(&cmd->device->ioerr_cnt);
1450
1451	disposition = scsi_decide_disposition(cmd);
1452	if (disposition != SUCCESS &&
1453	    time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
1454		scmd_printk(KERN_ERR, cmd,
1455			    "timing out command, waited %lus\n",
1456			    wait_for/HZ);
1457		disposition = SUCCESS;
1458	}
1459
1460	scsi_log_completion(cmd, disposition);
1461
1462	switch (disposition) {
1463	case SUCCESS:
1464		scsi_finish_command(cmd);
1465		break;
1466	case NEEDS_RETRY:
1467		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1468		break;
1469	case ADD_TO_MLQUEUE:
1470		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1471		break;
1472	default:
1473		scsi_eh_scmd_add(cmd);
1474		break;
1475	}
1476}
1477
1478/**
1479 * scsi_dispatch_command - Dispatch a command to the low-level driver.
1480 * @cmd: command block we are dispatching.
1481 *
1482 * Return: nonzero return request was rejected and device's queue needs to be
1483 * plugged.
1484 */
1485static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1486{
1487	struct Scsi_Host *host = cmd->device->host;
1488	int rtn = 0;
1489
1490	atomic_inc(&cmd->device->iorequest_cnt);
1491
1492	/* check if the device is still usable */
1493	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1494		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1495		 * returns an immediate error upwards, and signals
1496		 * that the device is no longer present */
1497		cmd->result = DID_NO_CONNECT << 16;
1498		goto done;
1499	}
1500
1501	/* Check to see if the scsi lld made this device blocked. */
1502	if (unlikely(scsi_device_blocked(cmd->device))) {
1503		/*
1504		 * in blocked state, the command is just put back on
1505		 * the device queue.  The suspend state has already
1506		 * blocked the queue so future requests should not
1507		 * occur until the device transitions out of the
1508		 * suspend state.
1509		 */
1510		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1511			"queuecommand : device blocked\n"));
 
1512		return SCSI_MLQUEUE_DEVICE_BUSY;
1513	}
1514
1515	/* Store the LUN value in cmnd, if needed. */
1516	if (cmd->device->lun_in_cdb)
1517		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1518			       (cmd->device->lun << 5 & 0xe0);
1519
1520	scsi_log_send(cmd);
1521
1522	/*
1523	 * Before we queue this command, check if the command
1524	 * length exceeds what the host adapter can handle.
1525	 */
1526	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1527		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1528			       "queuecommand : command too long. "
1529			       "cdb_size=%d host->max_cmd_len=%d\n",
1530			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1531		cmd->result = (DID_ABORT << 16);
1532		goto done;
1533	}
1534
1535	if (unlikely(host->shost_state == SHOST_DEL)) {
1536		cmd->result = (DID_NO_CONNECT << 16);
1537		goto done;
1538
1539	}
1540
1541	trace_scsi_dispatch_cmd_start(cmd);
1542	rtn = host->hostt->queuecommand(host, cmd);
1543	if (rtn) {
 
1544		trace_scsi_dispatch_cmd_error(cmd, rtn);
1545		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1546		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1547			rtn = SCSI_MLQUEUE_HOST_BUSY;
1548
1549		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1550			"queuecommand : request rejected\n"));
1551	}
1552
1553	return rtn;
1554 done:
1555	cmd->scsi_done(cmd);
1556	return 0;
1557}
1558
1559/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1560static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1561{
1562	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1563		sizeof(struct scatterlist);
1564}
1565
1566static blk_status_t scsi_mq_prep_fn(struct request *req)
1567{
1568	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1569	struct scsi_device *sdev = req->q->queuedata;
1570	struct Scsi_Host *shost = sdev->host;
 
1571	struct scatterlist *sg;
1572
1573	scsi_init_command(sdev, cmd);
1574
1575	cmd->request = req;
1576	cmd->tag = req->tag;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1577	cmd->prot_op = SCSI_PROT_NORMAL;
 
 
 
 
1578
1579	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1580	cmd->sdb.table.sgl = sg;
1581
1582	if (scsi_host_get_prot(shost)) {
1583		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1584
1585		cmd->prot_sdb->table.sgl =
1586			(struct scatterlist *)(cmd->prot_sdb + 1);
1587	}
1588
1589	blk_mq_start_request(req);
 
 
 
 
 
 
 
 
1590
1591	return scsi_setup_cmnd(sdev, req);
 
 
 
 
 
 
 
1592}
1593
1594static void scsi_mq_done(struct scsi_cmnd *cmd)
1595{
1596	if (unlikely(blk_should_fake_timeout(cmd->request->q)))
 
 
 
 
 
 
 
 
 
 
 
1597		return;
1598	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1599		return;
1600	trace_scsi_dispatch_cmd_done(cmd);
1601	blk_mq_complete_request(cmd->request);
 
 
 
 
 
 
 
 
 
1602}
 
1603
1604static void scsi_mq_put_budget(struct request_queue *q)
 
 
 
 
 
 
1605{
1606	struct scsi_device *sdev = q->queuedata;
1607
1608	atomic_dec(&sdev->device_busy);
1609}
1610
1611static bool scsi_mq_get_budget(struct request_queue *q)
 
 
 
 
 
 
 
1612{
1613	struct scsi_device *sdev = q->queuedata;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1614
1615	return scsi_dev_queue_ready(q, sdev);
 
 
 
 
 
 
 
1616}
1617
1618static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1619			 const struct blk_mq_queue_data *bd)
1620{
1621	struct request *req = bd->rq;
1622	struct request_queue *q = req->q;
1623	struct scsi_device *sdev = q->queuedata;
1624	struct Scsi_Host *shost = sdev->host;
1625	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1626	blk_status_t ret;
1627	int reason;
1628
 
 
1629	/*
1630	 * If the device is not in running state we will reject some or all
1631	 * commands.
1632	 */
1633	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1634		ret = scsi_prep_state_check(sdev, req);
1635		if (ret != BLK_STS_OK)
1636			goto out_put_budget;
1637	}
1638
1639	ret = BLK_STS_RESOURCE;
1640	if (!scsi_target_queue_ready(shost, sdev))
1641		goto out_put_budget;
 
 
 
 
 
1642	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1643		goto out_dec_target_busy;
1644
1645	if (!(req->rq_flags & RQF_DONTPREP)) {
1646		ret = scsi_mq_prep_fn(req);
1647		if (ret != BLK_STS_OK)
1648			goto out_dec_host_busy;
1649		req->rq_flags |= RQF_DONTPREP;
1650	} else {
1651		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
1652		blk_mq_start_request(req);
1653	}
1654
1655	cmd->flags &= SCMD_PRESERVED_FLAGS;
1656	if (sdev->simple_tags)
1657		cmd->flags |= SCMD_TAGGED;
1658	if (bd->last)
1659		cmd->flags |= SCMD_LAST;
1660
1661	scsi_init_cmd_errh(cmd);
1662	cmd->scsi_done = scsi_mq_done;
 
1663
 
1664	reason = scsi_dispatch_cmd(cmd);
1665	if (reason) {
1666		scsi_set_blocked(cmd, reason);
1667		ret = BLK_STS_RESOURCE;
1668		goto out_dec_host_busy;
1669	}
1670
1671	return BLK_STS_OK;
1672
1673out_dec_host_busy:
1674	scsi_dec_host_busy(shost, cmd);
1675out_dec_target_busy:
1676	if (scsi_target(sdev)->can_queue > 0)
1677		atomic_dec(&scsi_target(sdev)->target_busy);
1678out_put_budget:
1679	scsi_mq_put_budget(q);
 
1680	switch (ret) {
1681	case BLK_STS_OK:
1682		break;
1683	case BLK_STS_RESOURCE:
1684	case BLK_STS_ZONE_RESOURCE:
1685		if (atomic_read(&sdev->device_busy) ||
1686		    scsi_device_blocked(sdev))
1687			ret = BLK_STS_DEV_RESOURCE;
1688		break;
 
 
 
 
 
1689	default:
1690		if (unlikely(!scsi_device_online(sdev)))
1691			scsi_req(req)->result = DID_NO_CONNECT << 16;
1692		else
1693			scsi_req(req)->result = DID_ERROR << 16;
1694		/*
1695		 * Make sure to release all allocated resources when
1696		 * we hit an error, as we will never see this command
1697		 * again.
1698		 */
1699		if (req->rq_flags & RQF_DONTPREP)
1700			scsi_mq_uninit_cmd(cmd);
1701		scsi_run_queue_async(sdev);
1702		break;
1703	}
1704	return ret;
1705}
1706
1707static enum blk_eh_timer_return scsi_timeout(struct request *req,
1708		bool reserved)
1709{
1710	if (reserved)
1711		return BLK_EH_RESET_TIMER;
1712	return scsi_times_out(req);
1713}
1714
1715static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1716				unsigned int hctx_idx, unsigned int numa_node)
1717{
1718	struct Scsi_Host *shost = set->driver_data;
1719	const bool unchecked_isa_dma = shost->unchecked_isa_dma;
1720	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1721	struct scatterlist *sg;
1722	int ret = 0;
1723
1724	if (unchecked_isa_dma)
1725		cmd->flags |= SCMD_UNCHECKED_ISA_DMA;
1726	cmd->sense_buffer = scsi_alloc_sense_buffer(unchecked_isa_dma,
1727						    GFP_KERNEL, numa_node);
1728	if (!cmd->sense_buffer)
1729		return -ENOMEM;
1730	cmd->req.sense = cmd->sense_buffer;
1731
1732	if (scsi_host_get_prot(shost)) {
1733		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1734			shost->hostt->cmd_size;
1735		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1736	}
1737
1738	if (shost->hostt->init_cmd_priv) {
1739		ret = shost->hostt->init_cmd_priv(shost, cmd);
1740		if (ret < 0)
1741			scsi_free_sense_buffer(unchecked_isa_dma,
1742					       cmd->sense_buffer);
1743	}
1744
1745	return ret;
1746}
1747
1748static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1749				 unsigned int hctx_idx)
1750{
1751	struct Scsi_Host *shost = set->driver_data;
1752	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1753
1754	if (shost->hostt->exit_cmd_priv)
1755		shost->hostt->exit_cmd_priv(shost, cmd);
1756	scsi_free_sense_buffer(cmd->flags & SCMD_UNCHECKED_ISA_DMA,
1757			       cmd->sense_buffer);
 
 
 
 
 
 
 
 
 
 
1758}
1759
1760static int scsi_map_queues(struct blk_mq_tag_set *set)
 
 
 
 
 
 
 
 
 
1761{
1762	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1763
1764	if (shost->hostt->map_queues)
1765		return shost->hostt->map_queues(shost);
1766	return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1767}
1768
1769void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1770{
1771	struct device *dev = shost->dma_dev;
1772
1773	/*
1774	 * this limit is imposed by hardware restrictions
1775	 */
1776	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1777					SG_MAX_SEGMENTS));
1778
1779	if (scsi_host_prot_dma(shost)) {
1780		shost->sg_prot_tablesize =
1781			min_not_zero(shost->sg_prot_tablesize,
1782				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1783		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1784		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1785	}
1786
1787	if (dev->dma_mask) {
1788		shost->max_sectors = min_t(unsigned int, shost->max_sectors,
1789				dma_max_mapping_size(dev) >> SECTOR_SHIFT);
1790	}
1791	blk_queue_max_hw_sectors(q, shost->max_sectors);
1792	if (shost->unchecked_isa_dma)
1793		blk_queue_bounce_limit(q, BLK_BOUNCE_ISA);
1794	blk_queue_segment_boundary(q, shost->dma_boundary);
1795	dma_set_seg_boundary(dev, shost->dma_boundary);
1796
1797	blk_queue_max_segment_size(q, shost->max_segment_size);
1798	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1799	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1800
1801	/*
1802	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1803	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1804	 * which is set by the platform.
1805	 *
1806	 * Devices that require a bigger alignment can increase it later.
1807	 */
1808	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
1809}
1810EXPORT_SYMBOL_GPL(__scsi_init_queue);
1811
1812static const struct blk_mq_ops scsi_mq_ops_no_commit = {
1813	.get_budget	= scsi_mq_get_budget,
1814	.put_budget	= scsi_mq_put_budget,
1815	.queue_rq	= scsi_queue_rq,
1816	.complete	= scsi_softirq_done,
1817	.timeout	= scsi_timeout,
1818#ifdef CONFIG_BLK_DEBUG_FS
1819	.show_rq	= scsi_show_rq,
1820#endif
1821	.init_request	= scsi_mq_init_request,
1822	.exit_request	= scsi_mq_exit_request,
1823	.initialize_rq_fn = scsi_initialize_rq,
1824	.cleanup_rq	= scsi_cleanup_rq,
1825	.busy		= scsi_mq_lld_busy,
1826	.map_queues	= scsi_map_queues,
 
 
 
 
1827};
1828
1829
1830static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
1831{
1832	struct request_queue *q = hctx->queue;
1833	struct scsi_device *sdev = q->queuedata;
1834	struct Scsi_Host *shost = sdev->host;
1835
1836	shost->hostt->commit_rqs(shost, hctx->queue_num);
1837}
1838
1839static const struct blk_mq_ops scsi_mq_ops = {
1840	.get_budget	= scsi_mq_get_budget,
1841	.put_budget	= scsi_mq_put_budget,
1842	.queue_rq	= scsi_queue_rq,
1843	.commit_rqs	= scsi_commit_rqs,
1844	.complete	= scsi_softirq_done,
1845	.timeout	= scsi_timeout,
1846#ifdef CONFIG_BLK_DEBUG_FS
1847	.show_rq	= scsi_show_rq,
1848#endif
1849	.init_request	= scsi_mq_init_request,
1850	.exit_request	= scsi_mq_exit_request,
1851	.initialize_rq_fn = scsi_initialize_rq,
1852	.cleanup_rq	= scsi_cleanup_rq,
1853	.busy		= scsi_mq_lld_busy,
1854	.map_queues	= scsi_map_queues,
 
 
 
 
1855};
1856
1857struct request_queue *scsi_mq_alloc_queue(struct scsi_device *sdev)
1858{
1859	sdev->request_queue = blk_mq_init_queue(&sdev->host->tag_set);
1860	if (IS_ERR(sdev->request_queue))
1861		return NULL;
1862
1863	sdev->request_queue->queuedata = sdev;
1864	__scsi_init_queue(sdev->host, sdev->request_queue);
1865	blk_queue_flag_set(QUEUE_FLAG_SCSI_PASSTHROUGH, sdev->request_queue);
1866	return sdev->request_queue;
1867}
1868
1869int scsi_mq_setup_tags(struct Scsi_Host *shost)
1870{
1871	unsigned int cmd_size, sgl_size;
1872	struct blk_mq_tag_set *tag_set = &shost->tag_set;
1873
1874	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
1875				scsi_mq_inline_sgl_size(shost));
1876	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
1877	if (scsi_host_get_prot(shost))
1878		cmd_size += sizeof(struct scsi_data_buffer) +
1879			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
1880
1881	memset(tag_set, 0, sizeof(*tag_set));
1882	if (shost->hostt->commit_rqs)
1883		tag_set->ops = &scsi_mq_ops;
1884	else
1885		tag_set->ops = &scsi_mq_ops_no_commit;
1886	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
 
1887	tag_set->queue_depth = shost->can_queue;
1888	tag_set->cmd_size = cmd_size;
1889	tag_set->numa_node = NUMA_NO_NODE;
1890	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
1891	tag_set->flags |=
1892		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
 
 
1893	tag_set->driver_data = shost;
 
 
1894
1895	return blk_mq_alloc_tag_set(tag_set);
1896}
1897
1898void scsi_mq_destroy_tags(struct Scsi_Host *shost)
1899{
 
 
 
1900	blk_mq_free_tag_set(&shost->tag_set);
 
1901}
1902
1903/**
1904 * scsi_device_from_queue - return sdev associated with a request_queue
1905 * @q: The request queue to return the sdev from
1906 *
1907 * Return the sdev associated with a request queue or NULL if the
1908 * request_queue does not reference a SCSI device.
1909 */
1910struct scsi_device *scsi_device_from_queue(struct request_queue *q)
1911{
1912	struct scsi_device *sdev = NULL;
1913
1914	if (q->mq_ops == &scsi_mq_ops_no_commit ||
1915	    q->mq_ops == &scsi_mq_ops)
1916		sdev = q->queuedata;
1917	if (!sdev || !get_device(&sdev->sdev_gendev))
1918		sdev = NULL;
1919
1920	return sdev;
1921}
 
 
 
 
 
 
1922EXPORT_SYMBOL_GPL(scsi_device_from_queue);
 
1923
1924/**
1925 * scsi_block_requests - Utility function used by low-level drivers to prevent
1926 * further commands from being queued to the device.
1927 * @shost:  host in question
1928 *
1929 * There is no timer nor any other means by which the requests get unblocked
1930 * other than the low-level driver calling scsi_unblock_requests().
1931 */
1932void scsi_block_requests(struct Scsi_Host *shost)
1933{
1934	shost->host_self_blocked = 1;
1935}
1936EXPORT_SYMBOL(scsi_block_requests);
1937
1938/**
1939 * scsi_unblock_requests - Utility function used by low-level drivers to allow
1940 * further commands to be queued to the device.
1941 * @shost:  host in question
1942 *
1943 * There is no timer nor any other means by which the requests get unblocked
1944 * other than the low-level driver calling scsi_unblock_requests(). This is done
1945 * as an API function so that changes to the internals of the scsi mid-layer
1946 * won't require wholesale changes to drivers that use this feature.
1947 */
1948void scsi_unblock_requests(struct Scsi_Host *shost)
1949{
1950	shost->host_self_blocked = 0;
1951	scsi_run_host_queues(shost);
1952}
1953EXPORT_SYMBOL(scsi_unblock_requests);
1954
1955void scsi_exit_queue(void)
1956{
1957	kmem_cache_destroy(scsi_sense_cache);
1958	kmem_cache_destroy(scsi_sense_isadma_cache);
1959}
1960
1961/**
1962 *	scsi_mode_select - issue a mode select
1963 *	@sdev:	SCSI device to be queried
1964 *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
1965 *	@sp:	Save page bit (0 == don't save, 1 == save)
1966 *	@modepage: mode page being requested
1967 *	@buffer: request buffer (may not be smaller than eight bytes)
1968 *	@len:	length of request buffer.
1969 *	@timeout: command timeout
1970 *	@retries: number of retries before failing
1971 *	@data: returns a structure abstracting the mode header data
1972 *	@sshdr: place to put sense data (or NULL if no sense to be collected).
1973 *		must be SCSI_SENSE_BUFFERSIZE big.
1974 *
1975 *	Returns zero if successful; negative error number or scsi
1976 *	status on error
1977 *
1978 */
1979int
1980scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage,
1981		 unsigned char *buffer, int len, int timeout, int retries,
1982		 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
1983{
1984	unsigned char cmd[10];
1985	unsigned char *real_buffer;
 
 
 
1986	int ret;
1987
1988	memset(cmd, 0, sizeof(cmd));
1989	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
1990
1991	if (sdev->use_10_for_ms) {
1992		if (len > 65535)
 
 
 
 
 
 
 
1993			return -EINVAL;
1994		real_buffer = kmalloc(8 + len, GFP_KERNEL);
1995		if (!real_buffer)
1996			return -ENOMEM;
1997		memcpy(real_buffer + 8, buffer, len);
1998		len += 8;
1999		real_buffer[0] = 0;
2000		real_buffer[1] = 0;
2001		real_buffer[2] = data->medium_type;
2002		real_buffer[3] = data->device_specific;
2003		real_buffer[4] = data->longlba ? 0x01 : 0;
2004		real_buffer[5] = 0;
2005		real_buffer[6] = data->block_descriptor_length >> 8;
2006		real_buffer[7] = data->block_descriptor_length;
2007
2008		cmd[0] = MODE_SELECT_10;
2009		cmd[7] = len >> 8;
2010		cmd[8] = len;
2011	} else {
2012		if (len > 255 || data->block_descriptor_length > 255 ||
2013		    data->longlba)
2014			return -EINVAL;
2015
2016		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2017		if (!real_buffer)
2018			return -ENOMEM;
2019		memcpy(real_buffer + 4, buffer, len);
2020		len += 4;
2021		real_buffer[0] = 0;
2022		real_buffer[1] = data->medium_type;
2023		real_buffer[2] = data->device_specific;
2024		real_buffer[3] = data->block_descriptor_length;
2025
2026		cmd[0] = MODE_SELECT;
2027		cmd[4] = len;
2028	}
2029
2030	ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len,
2031			       sshdr, timeout, retries, NULL);
2032	kfree(real_buffer);
2033	return ret;
2034}
2035EXPORT_SYMBOL_GPL(scsi_mode_select);
2036
2037/**
2038 *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2039 *	@sdev:	SCSI device to be queried
2040 *	@dbd:	set if mode sense will allow block descriptors to be returned
2041 *	@modepage: mode page being requested
 
2042 *	@buffer: request buffer (may not be smaller than eight bytes)
2043 *	@len:	length of request buffer.
2044 *	@timeout: command timeout
2045 *	@retries: number of retries before failing
2046 *	@data: returns a structure abstracting the mode header data
2047 *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2048 *		must be SCSI_SENSE_BUFFERSIZE big.
2049 *
2050 *	Returns zero if unsuccessful, or the header offset (either 4
2051 *	or 8 depending on whether a six or ten byte command was
2052 *	issued) if successful.
2053 */
2054int
2055scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage,
2056		  unsigned char *buffer, int len, int timeout, int retries,
2057		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2058{
2059	unsigned char cmd[12];
2060	int use_10_for_ms;
2061	int header_length;
2062	int result, retry_count = retries;
2063	struct scsi_sense_hdr my_sshdr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2064
2065	memset(data, 0, sizeof(*data));
2066	memset(&cmd[0], 0, 12);
2067
2068	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2069	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2070	cmd[2] = modepage;
 
2071
2072	/* caller might not be interested in sense, but we need it */
2073	if (!sshdr)
2074		sshdr = &my_sshdr;
2075
2076 retry:
2077	use_10_for_ms = sdev->use_10_for_ms;
2078
2079	if (use_10_for_ms) {
2080		if (len < 8)
2081			len = 8;
2082
2083		cmd[0] = MODE_SENSE_10;
2084		cmd[8] = len;
2085		header_length = 8;
2086	} else {
2087		if (len < 4)
2088			len = 4;
2089
2090		cmd[0] = MODE_SENSE;
2091		cmd[4] = len;
2092		header_length = 4;
2093	}
2094
2095	memset(buffer, 0, len);
2096
2097	result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len,
2098				  sshdr, timeout, retries, NULL);
 
 
2099
2100	/* This code looks awful: what it's doing is making sure an
2101	 * ILLEGAL REQUEST sense return identifies the actual command
2102	 * byte as the problem.  MODE_SENSE commands can return
2103	 * ILLEGAL REQUEST if the code page isn't supported */
2104
2105	if (use_10_for_ms && !scsi_status_is_good(result) &&
2106	    driver_byte(result) == DRIVER_SENSE) {
2107		if (scsi_sense_valid(sshdr)) {
2108			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2109			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2110				/*
2111				 * Invalid command operation code
 
 
 
 
2112				 */
2113				sdev->use_10_for_ms = 0;
2114				goto retry;
 
 
 
 
2115			}
2116		}
 
2117	}
2118
2119	if (scsi_status_is_good(result)) {
2120		if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2121			     (modepage == 6 || modepage == 8))) {
2122			/* Initio breakage? */
2123			header_length = 0;
2124			data->length = 13;
2125			data->medium_type = 0;
2126			data->device_specific = 0;
2127			data->longlba = 0;
2128			data->block_descriptor_length = 0;
2129		} else if (use_10_for_ms) {
2130			data->length = buffer[0]*256 + buffer[1] + 2;
2131			data->medium_type = buffer[2];
2132			data->device_specific = buffer[3];
2133			data->longlba = buffer[4] & 0x01;
2134			data->block_descriptor_length = buffer[6]*256
2135				+ buffer[7];
2136		} else {
2137			data->length = buffer[0] + 1;
2138			data->medium_type = buffer[1];
2139			data->device_specific = buffer[2];
2140			data->block_descriptor_length = buffer[3];
2141		}
2142		data->header_length = header_length;
2143	} else if ((status_byte(result) == CHECK_CONDITION) &&
2144		   scsi_sense_valid(sshdr) &&
2145		   sshdr->sense_key == UNIT_ATTENTION && retry_count) {
2146		retry_count--;
2147		goto retry;
2148	}
 
2149
2150	return result;
2151}
2152EXPORT_SYMBOL(scsi_mode_sense);
2153
2154/**
2155 *	scsi_test_unit_ready - test if unit is ready
2156 *	@sdev:	scsi device to change the state of.
2157 *	@timeout: command timeout
2158 *	@retries: number of retries before failing
2159 *	@sshdr: outpout pointer for decoded sense information.
2160 *
2161 *	Returns zero if unsuccessful or an error if TUR failed.  For
2162 *	removable media, UNIT_ATTENTION sets ->changed flag.
2163 **/
2164int
2165scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2166		     struct scsi_sense_hdr *sshdr)
2167{
2168	char cmd[] = {
2169		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2170	};
 
 
 
2171	int result;
2172
2173	/* try to eat the UNIT_ATTENTION if there are enough retries */
2174	do {
2175		result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr,
2176					  timeout, 1, NULL);
2177		if (sdev->removable && scsi_sense_valid(sshdr) &&
2178		    sshdr->sense_key == UNIT_ATTENTION)
2179			sdev->changed = 1;
2180	} while (scsi_sense_valid(sshdr) &&
2181		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2182
2183	return result;
2184}
2185EXPORT_SYMBOL(scsi_test_unit_ready);
2186
2187/**
2188 *	scsi_device_set_state - Take the given device through the device state model.
2189 *	@sdev:	scsi device to change the state of.
2190 *	@state:	state to change to.
2191 *
2192 *	Returns zero if successful or an error if the requested
2193 *	transition is illegal.
2194 */
2195int
2196scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2197{
2198	enum scsi_device_state oldstate = sdev->sdev_state;
2199
2200	if (state == oldstate)
2201		return 0;
2202
2203	switch (state) {
2204	case SDEV_CREATED:
2205		switch (oldstate) {
2206		case SDEV_CREATED_BLOCK:
2207			break;
2208		default:
2209			goto illegal;
2210		}
2211		break;
2212
2213	case SDEV_RUNNING:
2214		switch (oldstate) {
2215		case SDEV_CREATED:
2216		case SDEV_OFFLINE:
2217		case SDEV_TRANSPORT_OFFLINE:
2218		case SDEV_QUIESCE:
2219		case SDEV_BLOCK:
2220			break;
2221		default:
2222			goto illegal;
2223		}
2224		break;
2225
2226	case SDEV_QUIESCE:
2227		switch (oldstate) {
2228		case SDEV_RUNNING:
2229		case SDEV_OFFLINE:
2230		case SDEV_TRANSPORT_OFFLINE:
2231			break;
2232		default:
2233			goto illegal;
2234		}
2235		break;
2236
2237	case SDEV_OFFLINE:
2238	case SDEV_TRANSPORT_OFFLINE:
2239		switch (oldstate) {
2240		case SDEV_CREATED:
2241		case SDEV_RUNNING:
2242		case SDEV_QUIESCE:
2243		case SDEV_BLOCK:
2244			break;
2245		default:
2246			goto illegal;
2247		}
2248		break;
2249
2250	case SDEV_BLOCK:
2251		switch (oldstate) {
2252		case SDEV_RUNNING:
2253		case SDEV_CREATED_BLOCK:
2254		case SDEV_QUIESCE:
2255		case SDEV_OFFLINE:
2256			break;
2257		default:
2258			goto illegal;
2259		}
2260		break;
2261
2262	case SDEV_CREATED_BLOCK:
2263		switch (oldstate) {
2264		case SDEV_CREATED:
2265			break;
2266		default:
2267			goto illegal;
2268		}
2269		break;
2270
2271	case SDEV_CANCEL:
2272		switch (oldstate) {
2273		case SDEV_CREATED:
2274		case SDEV_RUNNING:
2275		case SDEV_QUIESCE:
2276		case SDEV_OFFLINE:
2277		case SDEV_TRANSPORT_OFFLINE:
2278			break;
2279		default:
2280			goto illegal;
2281		}
2282		break;
2283
2284	case SDEV_DEL:
2285		switch (oldstate) {
2286		case SDEV_CREATED:
2287		case SDEV_RUNNING:
2288		case SDEV_OFFLINE:
2289		case SDEV_TRANSPORT_OFFLINE:
2290		case SDEV_CANCEL:
2291		case SDEV_BLOCK:
2292		case SDEV_CREATED_BLOCK:
2293			break;
2294		default:
2295			goto illegal;
2296		}
2297		break;
2298
2299	}
2300	sdev->offline_already = false;
2301	sdev->sdev_state = state;
2302	return 0;
2303
2304 illegal:
2305	SCSI_LOG_ERROR_RECOVERY(1,
2306				sdev_printk(KERN_ERR, sdev,
2307					    "Illegal state transition %s->%s",
2308					    scsi_device_state_name(oldstate),
2309					    scsi_device_state_name(state))
2310				);
2311	return -EINVAL;
2312}
2313EXPORT_SYMBOL(scsi_device_set_state);
2314
2315/**
2316 * 	sdev_evt_emit - emit a single SCSI device uevent
2317 *	@sdev: associated SCSI device
2318 *	@evt: event to emit
2319 *
2320 *	Send a single uevent (scsi_event) to the associated scsi_device.
2321 */
2322static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2323{
2324	int idx = 0;
2325	char *envp[3];
2326
2327	switch (evt->evt_type) {
2328	case SDEV_EVT_MEDIA_CHANGE:
2329		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2330		break;
2331	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2332		scsi_rescan_device(&sdev->sdev_gendev);
2333		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2334		break;
2335	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2336		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2337		break;
2338	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2339	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2340		break;
2341	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2342		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2343		break;
2344	case SDEV_EVT_LUN_CHANGE_REPORTED:
2345		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2346		break;
2347	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2348		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2349		break;
2350	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2351		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2352		break;
2353	default:
2354		/* do nothing */
2355		break;
2356	}
2357
2358	envp[idx++] = NULL;
2359
2360	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2361}
2362
2363/**
2364 * 	sdev_evt_thread - send a uevent for each scsi event
2365 *	@work: work struct for scsi_device
2366 *
2367 *	Dispatch queued events to their associated scsi_device kobjects
2368 *	as uevents.
2369 */
2370void scsi_evt_thread(struct work_struct *work)
2371{
2372	struct scsi_device *sdev;
2373	enum scsi_device_event evt_type;
2374	LIST_HEAD(event_list);
2375
2376	sdev = container_of(work, struct scsi_device, event_work);
2377
2378	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2379		if (test_and_clear_bit(evt_type, sdev->pending_events))
2380			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2381
2382	while (1) {
2383		struct scsi_event *evt;
2384		struct list_head *this, *tmp;
2385		unsigned long flags;
2386
2387		spin_lock_irqsave(&sdev->list_lock, flags);
2388		list_splice_init(&sdev->event_list, &event_list);
2389		spin_unlock_irqrestore(&sdev->list_lock, flags);
2390
2391		if (list_empty(&event_list))
2392			break;
2393
2394		list_for_each_safe(this, tmp, &event_list) {
2395			evt = list_entry(this, struct scsi_event, node);
2396			list_del(&evt->node);
2397			scsi_evt_emit(sdev, evt);
2398			kfree(evt);
2399		}
2400	}
2401}
2402
2403/**
2404 * 	sdev_evt_send - send asserted event to uevent thread
2405 *	@sdev: scsi_device event occurred on
2406 *	@evt: event to send
2407 *
2408 *	Assert scsi device event asynchronously.
2409 */
2410void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2411{
2412	unsigned long flags;
2413
2414#if 0
2415	/* FIXME: currently this check eliminates all media change events
2416	 * for polled devices.  Need to update to discriminate between AN
2417	 * and polled events */
2418	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2419		kfree(evt);
2420		return;
2421	}
2422#endif
2423
2424	spin_lock_irqsave(&sdev->list_lock, flags);
2425	list_add_tail(&evt->node, &sdev->event_list);
2426	schedule_work(&sdev->event_work);
2427	spin_unlock_irqrestore(&sdev->list_lock, flags);
2428}
2429EXPORT_SYMBOL_GPL(sdev_evt_send);
2430
2431/**
2432 * 	sdev_evt_alloc - allocate a new scsi event
2433 *	@evt_type: type of event to allocate
2434 *	@gfpflags: GFP flags for allocation
2435 *
2436 *	Allocates and returns a new scsi_event.
2437 */
2438struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2439				  gfp_t gfpflags)
2440{
2441	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2442	if (!evt)
2443		return NULL;
2444
2445	evt->evt_type = evt_type;
2446	INIT_LIST_HEAD(&evt->node);
2447
2448	/* evt_type-specific initialization, if any */
2449	switch (evt_type) {
2450	case SDEV_EVT_MEDIA_CHANGE:
2451	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2452	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2453	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2454	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2455	case SDEV_EVT_LUN_CHANGE_REPORTED:
2456	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2457	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2458	default:
2459		/* do nothing */
2460		break;
2461	}
2462
2463	return evt;
2464}
2465EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2466
2467/**
2468 * 	sdev_evt_send_simple - send asserted event to uevent thread
2469 *	@sdev: scsi_device event occurred on
2470 *	@evt_type: type of event to send
2471 *	@gfpflags: GFP flags for allocation
2472 *
2473 *	Assert scsi device event asynchronously, given an event type.
2474 */
2475void sdev_evt_send_simple(struct scsi_device *sdev,
2476			  enum scsi_device_event evt_type, gfp_t gfpflags)
2477{
2478	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2479	if (!evt) {
2480		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2481			    evt_type);
2482		return;
2483	}
2484
2485	sdev_evt_send(sdev, evt);
2486}
2487EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2488
2489/**
2490 *	scsi_device_quiesce - Block user issued commands.
2491 *	@sdev:	scsi device to quiesce.
2492 *
2493 *	This works by trying to transition to the SDEV_QUIESCE state
2494 *	(which must be a legal transition).  When the device is in this
2495 *	state, only special requests will be accepted, all others will
2496 *	be deferred.  Since special requests may also be requeued requests,
2497 *	a successful return doesn't guarantee the device will be
2498 *	totally quiescent.
2499 *
2500 *	Must be called with user context, may sleep.
2501 *
2502 *	Returns zero if unsuccessful or an error if not.
2503 */
2504int
2505scsi_device_quiesce(struct scsi_device *sdev)
2506{
2507	struct request_queue *q = sdev->request_queue;
2508	int err;
2509
2510	/*
2511	 * It is allowed to call scsi_device_quiesce() multiple times from
2512	 * the same context but concurrent scsi_device_quiesce() calls are
2513	 * not allowed.
2514	 */
2515	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2516
2517	if (sdev->quiesced_by == current)
2518		return 0;
2519
2520	blk_set_pm_only(q);
2521
2522	blk_mq_freeze_queue(q);
2523	/*
2524	 * Ensure that the effect of blk_set_pm_only() will be visible
2525	 * for percpu_ref_tryget() callers that occur after the queue
2526	 * unfreeze even if the queue was already frozen before this function
2527	 * was called. See also https://lwn.net/Articles/573497/.
2528	 */
2529	synchronize_rcu();
2530	blk_mq_unfreeze_queue(q);
2531
2532	mutex_lock(&sdev->state_mutex);
2533	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2534	if (err == 0)
2535		sdev->quiesced_by = current;
2536	else
2537		blk_clear_pm_only(q);
2538	mutex_unlock(&sdev->state_mutex);
2539
2540	return err;
2541}
2542EXPORT_SYMBOL(scsi_device_quiesce);
2543
2544/**
2545 *	scsi_device_resume - Restart user issued commands to a quiesced device.
2546 *	@sdev:	scsi device to resume.
2547 *
2548 *	Moves the device from quiesced back to running and restarts the
2549 *	queues.
2550 *
2551 *	Must be called with user context, may sleep.
2552 */
2553void scsi_device_resume(struct scsi_device *sdev)
2554{
2555	/* check if the device state was mutated prior to resume, and if
2556	 * so assume the state is being managed elsewhere (for example
2557	 * device deleted during suspend)
2558	 */
2559	mutex_lock(&sdev->state_mutex);
 
 
2560	if (sdev->quiesced_by) {
2561		sdev->quiesced_by = NULL;
2562		blk_clear_pm_only(sdev->request_queue);
2563	}
2564	if (sdev->sdev_state == SDEV_QUIESCE)
2565		scsi_device_set_state(sdev, SDEV_RUNNING);
2566	mutex_unlock(&sdev->state_mutex);
2567}
2568EXPORT_SYMBOL(scsi_device_resume);
2569
2570static void
2571device_quiesce_fn(struct scsi_device *sdev, void *data)
2572{
2573	scsi_device_quiesce(sdev);
2574}
2575
2576void
2577scsi_target_quiesce(struct scsi_target *starget)
2578{
2579	starget_for_each_device(starget, NULL, device_quiesce_fn);
2580}
2581EXPORT_SYMBOL(scsi_target_quiesce);
2582
2583static void
2584device_resume_fn(struct scsi_device *sdev, void *data)
2585{
2586	scsi_device_resume(sdev);
2587}
2588
2589void
2590scsi_target_resume(struct scsi_target *starget)
2591{
2592	starget_for_each_device(starget, NULL, device_resume_fn);
2593}
2594EXPORT_SYMBOL(scsi_target_resume);
2595
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2596/**
2597 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2598 * @sdev: device to block
2599 *
2600 * Pause SCSI command processing on the specified device. Does not sleep.
2601 *
2602 * Returns zero if successful or a negative error code upon failure.
2603 *
2604 * Notes:
2605 * This routine transitions the device to the SDEV_BLOCK state (which must be
2606 * a legal transition). When the device is in this state, command processing
2607 * is paused until the device leaves the SDEV_BLOCK state. See also
2608 * scsi_internal_device_unblock_nowait().
2609 */
2610int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2611{
2612	struct request_queue *q = sdev->request_queue;
2613	int err = 0;
2614
2615	err = scsi_device_set_state(sdev, SDEV_BLOCK);
2616	if (err) {
2617		err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2618
2619		if (err)
2620			return err;
2621	}
2622
2623	/*
2624	 * The device has transitioned to SDEV_BLOCK.  Stop the
2625	 * block layer from calling the midlayer with this device's
2626	 * request queue.
2627	 */
2628	blk_mq_quiesce_queue_nowait(q);
2629	return 0;
 
2630}
2631EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2632
2633/**
2634 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state
2635 * @sdev: device to block
 
2636 *
2637 * Pause SCSI command processing on the specified device and wait until all
2638 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep.
2639 *
2640 * Returns zero if successful or a negative error code upon failure.
2641 *
2642 * Note:
2643 * This routine transitions the device to the SDEV_BLOCK state (which must be
2644 * a legal transition). When the device is in this state, command processing
2645 * is paused until the device leaves the SDEV_BLOCK state. See also
2646 * scsi_internal_device_unblock().
2647 */
2648static int scsi_internal_device_block(struct scsi_device *sdev)
2649{
2650	struct request_queue *q = sdev->request_queue;
2651	int err;
 
2652
2653	mutex_lock(&sdev->state_mutex);
2654	err = scsi_internal_device_block_nowait(sdev);
 
2655	if (err == 0)
2656		blk_mq_quiesce_queue(q);
2657	mutex_unlock(&sdev->state_mutex);
2658
2659	return err;
2660}
 
2661
2662void scsi_start_queue(struct scsi_device *sdev)
2663{
2664	struct request_queue *q = sdev->request_queue;
2665
2666	blk_mq_unquiesce_queue(q);
 
2667}
2668
2669/**
2670 * scsi_internal_device_unblock_nowait - resume a device after a block request
2671 * @sdev:	device to resume
2672 * @new_state:	state to set the device to after unblocking
2673 *
2674 * Restart the device queue for a previously suspended SCSI device. Does not
2675 * sleep.
2676 *
2677 * Returns zero if successful or a negative error code upon failure.
2678 *
2679 * Notes:
2680 * This routine transitions the device to the SDEV_RUNNING state or to one of
2681 * the offline states (which must be a legal transition) allowing the midlayer
2682 * to goose the queue for this device.
2683 */
2684int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2685					enum scsi_device_state new_state)
2686{
2687	switch (new_state) {
2688	case SDEV_RUNNING:
2689	case SDEV_TRANSPORT_OFFLINE:
2690		break;
2691	default:
2692		return -EINVAL;
2693	}
2694
2695	/*
2696	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2697	 * offlined states and goose the device queue if successful.
2698	 */
2699	switch (sdev->sdev_state) {
2700	case SDEV_BLOCK:
2701	case SDEV_TRANSPORT_OFFLINE:
2702		sdev->sdev_state = new_state;
2703		break;
2704	case SDEV_CREATED_BLOCK:
2705		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2706		    new_state == SDEV_OFFLINE)
2707			sdev->sdev_state = new_state;
2708		else
2709			sdev->sdev_state = SDEV_CREATED;
2710		break;
2711	case SDEV_CANCEL:
2712	case SDEV_OFFLINE:
2713		break;
2714	default:
2715		return -EINVAL;
2716	}
2717	scsi_start_queue(sdev);
2718
2719	return 0;
2720}
2721EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2722
2723/**
2724 * scsi_internal_device_unblock - resume a device after a block request
2725 * @sdev:	device to resume
2726 * @new_state:	state to set the device to after unblocking
2727 *
2728 * Restart the device queue for a previously suspended SCSI device. May sleep.
2729 *
2730 * Returns zero if successful or a negative error code upon failure.
2731 *
2732 * Notes:
2733 * This routine transitions the device to the SDEV_RUNNING state or to one of
2734 * the offline states (which must be a legal transition) allowing the midlayer
2735 * to goose the queue for this device.
2736 */
2737static int scsi_internal_device_unblock(struct scsi_device *sdev,
2738					enum scsi_device_state new_state)
2739{
2740	int ret;
2741
2742	mutex_lock(&sdev->state_mutex);
2743	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2744	mutex_unlock(&sdev->state_mutex);
2745
2746	return ret;
2747}
2748
2749static void
2750device_block(struct scsi_device *sdev, void *data)
2751{
2752	int ret;
2753
2754	ret = scsi_internal_device_block(sdev);
2755
2756	WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n",
2757		  dev_name(&sdev->sdev_gendev), ret);
2758}
2759
2760static int
2761target_block(struct device *dev, void *data)
2762{
2763	if (scsi_is_target_device(dev))
2764		starget_for_each_device(to_scsi_target(dev), NULL,
2765					device_block);
2766	return 0;
2767}
2768
 
 
 
 
 
 
 
 
 
 
 
 
2769void
2770scsi_target_block(struct device *dev)
2771{
2772	if (scsi_is_target_device(dev))
2773		starget_for_each_device(to_scsi_target(dev), NULL,
2774					device_block);
2775	else
2776		device_for_each_child(dev, NULL, target_block);
2777}
2778EXPORT_SYMBOL_GPL(scsi_target_block);
2779
2780static void
2781device_unblock(struct scsi_device *sdev, void *data)
2782{
2783	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
2784}
2785
2786static int
2787target_unblock(struct device *dev, void *data)
2788{
2789	if (scsi_is_target_device(dev))
2790		starget_for_each_device(to_scsi_target(dev), data,
2791					device_unblock);
2792	return 0;
2793}
2794
2795void
2796scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
2797{
2798	if (scsi_is_target_device(dev))
2799		starget_for_each_device(to_scsi_target(dev), &new_state,
2800					device_unblock);
2801	else
2802		device_for_each_child(dev, &new_state, target_unblock);
2803}
2804EXPORT_SYMBOL_GPL(scsi_target_unblock);
2805
 
 
 
 
 
 
 
 
 
2806int
2807scsi_host_block(struct Scsi_Host *shost)
2808{
2809	struct scsi_device *sdev;
2810	int ret = 0;
2811
2812	/*
2813	 * Call scsi_internal_device_block_nowait so we can avoid
2814	 * calling synchronize_rcu() for each LUN.
2815	 */
2816	shost_for_each_device(sdev, shost) {
2817		mutex_lock(&sdev->state_mutex);
2818		ret = scsi_internal_device_block_nowait(sdev);
2819		mutex_unlock(&sdev->state_mutex);
2820		if (ret) {
2821			scsi_device_put(sdev);
2822			break;
2823		}
2824	}
2825
2826	/*
2827	 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so
2828	 * calling synchronize_rcu() once is enough.
2829	 */
2830	WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING);
2831
2832	if (!ret)
2833		synchronize_rcu();
2834
2835	return ret;
2836}
2837EXPORT_SYMBOL_GPL(scsi_host_block);
2838
2839int
2840scsi_host_unblock(struct Scsi_Host *shost, int new_state)
2841{
2842	struct scsi_device *sdev;
2843	int ret = 0;
2844
2845	shost_for_each_device(sdev, shost) {
2846		ret = scsi_internal_device_unblock(sdev, new_state);
2847		if (ret) {
2848			scsi_device_put(sdev);
2849			break;
2850		}
2851	}
2852	return ret;
2853}
2854EXPORT_SYMBOL_GPL(scsi_host_unblock);
2855
2856/**
2857 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
2858 * @sgl:	scatter-gather list
2859 * @sg_count:	number of segments in sg
2860 * @offset:	offset in bytes into sg, on return offset into the mapped area
2861 * @len:	bytes to map, on return number of bytes mapped
2862 *
2863 * Returns virtual address of the start of the mapped page
2864 */
2865void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
2866			  size_t *offset, size_t *len)
2867{
2868	int i;
2869	size_t sg_len = 0, len_complete = 0;
2870	struct scatterlist *sg;
2871	struct page *page;
2872
2873	WARN_ON(!irqs_disabled());
2874
2875	for_each_sg(sgl, sg, sg_count, i) {
2876		len_complete = sg_len; /* Complete sg-entries */
2877		sg_len += sg->length;
2878		if (sg_len > *offset)
2879			break;
2880	}
2881
2882	if (unlikely(i == sg_count)) {
2883		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
2884			"elements %d\n",
2885		       __func__, sg_len, *offset, sg_count);
2886		WARN_ON(1);
2887		return NULL;
2888	}
2889
2890	/* Offset starting from the beginning of first page in this sg-entry */
2891	*offset = *offset - len_complete + sg->offset;
2892
2893	/* Assumption: contiguous pages can be accessed as "page + i" */
2894	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
2895	*offset &= ~PAGE_MASK;
2896
2897	/* Bytes in this sg-entry from *offset to the end of the page */
2898	sg_len = PAGE_SIZE - *offset;
2899	if (*len > sg_len)
2900		*len = sg_len;
2901
2902	return kmap_atomic(page);
2903}
2904EXPORT_SYMBOL(scsi_kmap_atomic_sg);
2905
2906/**
2907 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
2908 * @virt:	virtual address to be unmapped
2909 */
2910void scsi_kunmap_atomic_sg(void *virt)
2911{
2912	kunmap_atomic(virt);
2913}
2914EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
2915
2916void sdev_disable_disk_events(struct scsi_device *sdev)
2917{
2918	atomic_inc(&sdev->disk_events_disable_depth);
2919}
2920EXPORT_SYMBOL(sdev_disable_disk_events);
2921
2922void sdev_enable_disk_events(struct scsi_device *sdev)
2923{
2924	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
2925		return;
2926	atomic_dec(&sdev->disk_events_disable_depth);
2927}
2928EXPORT_SYMBOL(sdev_enable_disk_events);
2929
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2930/**
2931 * scsi_vpd_lun_id - return a unique device identification
2932 * @sdev: SCSI device
2933 * @id:   buffer for the identification
2934 * @id_len:  length of the buffer
2935 *
2936 * Copies a unique device identification into @id based
2937 * on the information in the VPD page 0x83 of the device.
2938 * The string will be formatted as a SCSI name string.
2939 *
2940 * Returns the length of the identification or error on failure.
2941 * If the identifier is longer than the supplied buffer the actual
2942 * identifier length is returned and the buffer is not zero-padded.
2943 */
2944int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
2945{
2946	u8 cur_id_type = 0xff;
2947	u8 cur_id_size = 0;
2948	const unsigned char *d, *cur_id_str;
2949	const struct scsi_vpd *vpd_pg83;
2950	int id_size = -EINVAL;
2951
2952	rcu_read_lock();
2953	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
2954	if (!vpd_pg83) {
2955		rcu_read_unlock();
2956		return -ENXIO;
2957	}
2958
2959	/*
2960	 * Look for the correct descriptor.
2961	 * Order of preference for lun descriptor:
2962	 * - SCSI name string
2963	 * - NAA IEEE Registered Extended
2964	 * - EUI-64 based 16-byte
2965	 * - EUI-64 based 12-byte
2966	 * - NAA IEEE Registered
2967	 * - NAA IEEE Extended
2968	 * - T10 Vendor ID
2969	 * as longer descriptors reduce the likelyhood
2970	 * of identification clashes.
2971	 */
2972
2973	/* The id string must be at least 20 bytes + terminating NULL byte */
2974	if (id_len < 21) {
2975		rcu_read_unlock();
2976		return -EINVAL;
2977	}
2978
2979	memset(id, 0, id_len);
2980	d = vpd_pg83->data + 4;
2981	while (d < vpd_pg83->data + vpd_pg83->len) {
2982		/* Skip designators not referring to the LUN */
2983		if ((d[1] & 0x30) != 0x00)
2984			goto next_desig;
 
 
2985
2986		switch (d[1] & 0xf) {
2987		case 0x1:
2988			/* T10 Vendor ID */
2989			if (cur_id_size > d[3])
2990				break;
2991			/* Prefer anything */
2992			if (cur_id_type > 0x01 && cur_id_type != 0xff)
2993				break;
2994			cur_id_size = d[3];
2995			if (cur_id_size + 4 > id_len)
2996				cur_id_size = id_len - 4;
2997			cur_id_str = d + 4;
2998			cur_id_type = d[1] & 0xf;
2999			id_size = snprintf(id, id_len, "t10.%*pE",
3000					   cur_id_size, cur_id_str);
3001			break;
3002		case 0x2:
3003			/* EUI-64 */
3004			if (cur_id_size > d[3])
3005				break;
3006			/* Prefer NAA IEEE Registered Extended */
3007			if (cur_id_type == 0x3 &&
3008			    cur_id_size == d[3])
3009				break;
3010			cur_id_size = d[3];
3011			cur_id_str = d + 4;
3012			cur_id_type = d[1] & 0xf;
3013			switch (cur_id_size) {
3014			case 8:
3015				id_size = snprintf(id, id_len,
3016						   "eui.%8phN",
3017						   cur_id_str);
3018				break;
3019			case 12:
3020				id_size = snprintf(id, id_len,
3021						   "eui.%12phN",
3022						   cur_id_str);
3023				break;
3024			case 16:
3025				id_size = snprintf(id, id_len,
3026						   "eui.%16phN",
3027						   cur_id_str);
3028				break;
3029			default:
3030				cur_id_size = 0;
3031				break;
3032			}
3033			break;
3034		case 0x3:
3035			/* NAA */
3036			if (cur_id_size > d[3])
3037				break;
3038			cur_id_size = d[3];
3039			cur_id_str = d + 4;
3040			cur_id_type = d[1] & 0xf;
3041			switch (cur_id_size) {
3042			case 8:
3043				id_size = snprintf(id, id_len,
3044						   "naa.%8phN",
3045						   cur_id_str);
3046				break;
3047			case 16:
3048				id_size = snprintf(id, id_len,
3049						   "naa.%16phN",
3050						   cur_id_str);
3051				break;
3052			default:
3053				cur_id_size = 0;
3054				break;
3055			}
3056			break;
3057		case 0x8:
3058			/* SCSI name string */
3059			if (cur_id_size + 4 > d[3])
3060				break;
3061			/* Prefer others for truncated descriptor */
3062			if (cur_id_size && d[3] > id_len)
3063				break;
 
 
 
 
3064			cur_id_size = id_size = d[3];
3065			cur_id_str = d + 4;
3066			cur_id_type = d[1] & 0xf;
3067			if (cur_id_size >= id_len)
3068				cur_id_size = id_len - 1;
3069			memcpy(id, cur_id_str, cur_id_size);
3070			/* Decrease priority for truncated descriptor */
3071			if (cur_id_size != id_size)
3072				cur_id_size = 6;
3073			break;
3074		default:
3075			break;
3076		}
3077next_desig:
3078		d += d[3] + 4;
3079	}
3080	rcu_read_unlock();
3081
3082	return id_size;
3083}
3084EXPORT_SYMBOL(scsi_vpd_lun_id);
3085
3086/*
3087 * scsi_vpd_tpg_id - return a target port group identifier
3088 * @sdev: SCSI device
3089 *
3090 * Returns the Target Port Group identifier from the information
3091 * froom VPD page 0x83 of the device.
3092 *
3093 * Returns the identifier or error on failure.
3094 */
3095int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3096{
3097	const unsigned char *d;
3098	const struct scsi_vpd *vpd_pg83;
3099	int group_id = -EAGAIN, rel_port = -1;
3100
3101	rcu_read_lock();
3102	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3103	if (!vpd_pg83) {
3104		rcu_read_unlock();
3105		return -ENXIO;
3106	}
3107
3108	d = vpd_pg83->data + 4;
3109	while (d < vpd_pg83->data + vpd_pg83->len) {
3110		switch (d[1] & 0xf) {
3111		case 0x4:
3112			/* Relative target port */
3113			rel_port = get_unaligned_be16(&d[6]);
3114			break;
3115		case 0x5:
3116			/* Target port group */
3117			group_id = get_unaligned_be16(&d[6]);
3118			break;
3119		default:
3120			break;
3121		}
3122		d += d[3] + 4;
3123	}
3124	rcu_read_unlock();
3125
3126	if (group_id >= 0 && rel_id && rel_port != -1)
3127		*rel_id = rel_port;
3128
3129	return group_id;
3130}
3131EXPORT_SYMBOL(scsi_vpd_tpg_id);
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 1999 Eric Youngdale
   4 * Copyright (C) 2014 Christoph Hellwig
   5 *
   6 *  SCSI queueing library.
   7 *      Initial versions: Eric Youngdale (eric@andante.org).
   8 *                        Based upon conversations with large numbers
   9 *                        of people at Linux Expo.
  10 */
  11
  12#include <linux/bio.h>
  13#include <linux/bitops.h>
  14#include <linux/blkdev.h>
  15#include <linux/completion.h>
  16#include <linux/kernel.h>
  17#include <linux/export.h>
  18#include <linux/init.h>
  19#include <linux/pci.h>
  20#include <linux/delay.h>
  21#include <linux/hardirq.h>
  22#include <linux/scatterlist.h>
  23#include <linux/blk-mq.h>
  24#include <linux/blk-integrity.h>
  25#include <linux/ratelimit.h>
  26#include <asm/unaligned.h>
  27
  28#include <scsi/scsi.h>
  29#include <scsi/scsi_cmnd.h>
  30#include <scsi/scsi_dbg.h>
  31#include <scsi/scsi_device.h>
  32#include <scsi/scsi_driver.h>
  33#include <scsi/scsi_eh.h>
  34#include <scsi/scsi_host.h>
  35#include <scsi/scsi_transport.h> /* __scsi_init_queue() */
  36#include <scsi/scsi_dh.h>
  37
  38#include <trace/events/scsi.h>
  39
  40#include "scsi_debugfs.h"
  41#include "scsi_priv.h"
  42#include "scsi_logging.h"
  43
  44/*
  45 * Size of integrity metadata is usually small, 1 inline sg should
  46 * cover normal cases.
  47 */
  48#ifdef CONFIG_ARCH_NO_SG_CHAIN
  49#define  SCSI_INLINE_PROT_SG_CNT  0
  50#define  SCSI_INLINE_SG_CNT  0
  51#else
  52#define  SCSI_INLINE_PROT_SG_CNT  1
  53#define  SCSI_INLINE_SG_CNT  2
  54#endif
  55
  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 void
  79scsi_set_blocked(struct scsi_cmnd *cmd, int reason)
  80{
  81	struct Scsi_Host *host = cmd->device->host;
  82	struct scsi_device *device = cmd->device;
  83	struct scsi_target *starget = scsi_target(device);
  84
  85	/*
  86	 * Set the appropriate busy bit for the device/host.
  87	 *
  88	 * If the host/device isn't busy, assume that something actually
  89	 * completed, and that we should be able to queue a command now.
  90	 *
  91	 * Note that the prior mid-layer assumption that any host could
  92	 * always queue at least one command is now broken.  The mid-layer
  93	 * will implement a user specifiable stall (see
  94	 * scsi_host.max_host_blocked and scsi_device.max_device_blocked)
  95	 * if a command is requeued with no other commands outstanding
  96	 * either for the device or for the host.
  97	 */
  98	switch (reason) {
  99	case SCSI_MLQUEUE_HOST_BUSY:
 100		atomic_set(&host->host_blocked, host->max_host_blocked);
 101		break;
 102	case SCSI_MLQUEUE_DEVICE_BUSY:
 103	case SCSI_MLQUEUE_EH_RETRY:
 104		atomic_set(&device->device_blocked,
 105			   device->max_device_blocked);
 106		break;
 107	case SCSI_MLQUEUE_TARGET_BUSY:
 108		atomic_set(&starget->target_blocked,
 109			   starget->max_target_blocked);
 110		break;
 111	}
 112}
 113
 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 insertion
 132 * @cmd: The SCSI command being requeued
 133 * @reason:  The reason for the requeue
 134 * @unbusy: Whether the queue should be unbusied
 135 *
 136 * This is a private queue insertion.  The public interface
 137 * scsi_queue_insert() always assumes the queue should be unbusied
 138 * because it's always called before the completion.  This function is
 139 * for a requeue after completion, which should only occur in this
 140 * file.
 141 */
 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 longer
 153	 * active on the host/device.
 154	 */
 155	if (unbusy)
 156		scsi_device_unbusy(device, cmd);
 157
 158	/*
 159	 * Requeue this command.  It will go before all other commands
 160	 * that are already in the queue. Schedule requeue work under
 161	 * lock such that the kblockd_schedule_work() call happens
 162	 * before blk_mq_destroy_queue() finishes.
 163	 */
 164	cmd->result = 0;
 165
 166	blk_mq_requeue_request(scsi_cmd_to_rq(cmd),
 167			       !scsi_host_in_recovery(cmd->device->host));
 168}
 169
 170/**
 171 * scsi_queue_insert - Reinsert a command in the queue.
 172 * @cmd:    command that we are adding to queue.
 173 * @reason: why we are inserting command to queue.
 174 *
 175 * We do this for one of two cases. Either the host is busy and it cannot accept
 176 * any more commands for the time being, or the device returned QUEUE_FULL and
 177 * can accept no more commands.
 178 *
 179 * Context: This could be called either from an interrupt context or a normal
 180 * process context.
 181 */
 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 result
 275 * @sdev:	scsi_device
 276 * @cmd:	scsi command
 277 * @opf:	block layer request cmd_flags
 278 * @buffer:	data buffer
 279 * @bufflen:	len of buffer
 280 * @timeout:	request timeout in HZ
 281 * @ml_retries:	number of times SCSI midlayer will retry request
 282 * @args:	Optional args. See struct definition for field descriptions
 
 
 
 
 283 *
 284 * Returns the scsi_cmnd result field if a command was executed, or a negative
 285 * 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 work
 324	 */
 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 transfer
 334	 * garbage data together with a residue indicating that the data
 335	 * is invalid.  Prevent the garbage from being misinterpreted
 336	 * 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 error
 359 * handler is not woken up if host in-flight requests number ==
 360 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination
 361 * with an RCU read lock in this function to ensure that this function in
 362 * its entirety either finishes before scsi_eh_scmd_add() increases the
 363 * host_failed counter or that it notices the shost state change made by
 364 * 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 with
 399 * 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 with
 428	 * current_sdev. We race with others (to set starget_sdev_user),
 429	 * but in most cases, we will be first. Ideally, each LU on the
 430	 * 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 have
 485		 * starved queues, call blk_run_queue. scsi_request_fn
 486		 * drops the queue_lock and can add us back to the
 487		 * starved_list.
 488		 *
 489		 * host_lock protects the starved_list and starved_entry.
 490		 * scsi_request_fn must get the host_lock before checking
 491		 * 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 destroy
 508		 * it and the queue.  Mitigate by taking a reference to the
 509		 * queue and never touching the sdev again after we drop the
 510		 * host lock.  Note: if __scsi_remove_device() invokes
 511		 * blk_mq_destroy_queue() before the queue is run from this
 512		 * function then blk_run_queue() will return immediately since
 513		 * 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 queue
 533 *
 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 in
 605		 * .end_io is for ordering writing .device_busy in
 606		 * 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 budget
 612		 *  contention occurs.
 613		 *
 614		 *  No need to run queue when either another re-run
 615		 *  queue wins in updating ->restarts or a new budget
 616		 *  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	// XXX:
 635	if (blk_queue_add_random(q))
 636		add_disk_randomness(req->q->disk);
 637
 638	WARN_ON_ONCE(!blk_rq_is_passthrough(req) &&
 639		     !(cmd->flags & SCMD_INITIALIZED));
 640	cmd->flags = 0;
 
 641
 642	/*
 643	 * Calling rcu_barrier() is not necessary here because the
 644	 * SCSI error handler guarantees that the function called by
 645	 * call_rcu() has been called before scsi_end_request() is
 646	 * called.
 647	 */
 648	destroy_rcu_head(&cmd->rcu);
 649
 650	/*
 651	 * In the MQ case the command gets freed by __blk_mq_end_request,
 652	 * so we have to do all cleanup that depends on it earlier.
 653	 *
 654	 * We also can't kick the queues from irq context, so we
 655	 * will have to defer it to a workqueue.
 656	 */
 657	scsi_mq_uninit_cmd(cmd);
 658
 659	/*
 660	 * queue is still alive, so grab the ref for preventing it
 661	 * from being cleaned up during running queue.
 662	 */
 663	percpu_ref_get(&q->q_usage_counter);
 664
 665	__blk_mq_end_request(req, error);
 666
 667	scsi_run_queue_async(sdev);
 668
 669	percpu_ref_put(&q->q_usage_counter);
 670	return false;
 671}
 672
 673/**
 674 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t
 
 675 * @result:	scsi error code
 676 *
 677 * Translate a SCSI result code into a blk_status_t value.
 
 678 */
 679static blk_status_t scsi_result_to_blk_status(int result)
 680{
 681	/*
 682	 * Check the scsi-ml byte first in case we converted a host or status
 683	 * byte.
 684	 */
 685	switch (scsi_ml_byte(result)) {
 686	case SCSIML_STAT_OK:
 687		break;
 688	case SCSIML_STAT_RESV_CONFLICT:
 689		return BLK_STS_RESV_CONFLICT;
 690	case SCSIML_STAT_NOSPC:
 691		return BLK_STS_NOSPC;
 692	case SCSIML_STAT_MED_ERROR:
 693		return BLK_STS_MEDIUM;
 694	case SCSIML_STAT_TGT_FAILURE:
 695		return BLK_STS_TARGET;
 696	case SCSIML_STAT_DL_TIMEOUT:
 697		return BLK_STS_DURATION_LIMIT;
 698	}
 699
 700	switch (host_byte(result)) {
 701	case DID_OK:
 702		if (scsi_status_is_good(result))
 
 
 
 
 
 703			return BLK_STS_OK;
 704		return BLK_STS_IOERR;
 705	case DID_TRANSPORT_FAILFAST:
 706	case DID_TRANSPORT_MARGINAL:
 707		return BLK_STS_TRANSPORT;
 
 
 
 
 
 
 
 
 
 
 
 
 708	default:
 709		return BLK_STS_IOERR;
 710	}
 711}
 712
 713/**
 714 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary
 715 * @rq: request to examine
 716 *
 717 * Description:
 718 *     A request could be merge of IOs which require different failure
 719 *     handling.  This function determines the number of bytes which
 720 *     can be failed from the beginning of the request without
 721 *     crossing into area which need to be retried further.
 722 *
 723 * Return:
 724 *     The number of bytes to fail.
 725 */
 726static unsigned int scsi_rq_err_bytes(const struct request *rq)
 727{
 728	blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK;
 729	unsigned int bytes = 0;
 730	struct bio *bio;
 731
 732	if (!(rq->rq_flags & RQF_MIXED_MERGE))
 733		return blk_rq_bytes(rq);
 734
 735	/*
 736	 * Currently the only 'mixing' which can happen is between
 737	 * different fastfail types.  We can safely fail portions
 738	 * which have all the failfast bits that the first one has -
 739	 * the ones which are at least as eager to fail as the first
 740	 * one.
 741	 */
 742	for (bio = rq->bio; bio; bio = bio->bi_next) {
 743		if ((bio->bi_opf & ff) != ff)
 744			break;
 745		bytes += bio->bi_iter.bi_size;
 746	}
 747
 748	/* this could lead to infinite loop */
 749	BUG_ON(blk_rq_bytes(rq) && !bytes);
 750	return bytes;
 751}
 752
 753static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd)
 754{
 755	struct request *req = scsi_cmd_to_rq(cmd);
 756	unsigned long wait_for;
 757
 758	if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT)
 759		return false;
 760
 761	wait_for = (cmd->allowed + 1) * req->timeout;
 762	if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) {
 763		scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n",
 764			    wait_for/HZ);
 765		return true;
 766	}
 767	return false;
 768}
 769
 770/*
 771 * When ALUA transition state is returned, reprep the cmd to
 772 * use the ALUA handler's transition timeout. Delay the reprep
 773 * 1 sec to avoid aggressive retries of the target in that
 774 * state.
 775 */
 776#define ALUA_TRANSITION_REPREP_DELAY	1000
 777
 778/* Helper for scsi_io_completion() when special action required. */
 779static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result)
 780{
 781	struct request *req = scsi_cmd_to_rq(cmd);
 
 782	int level = 0;
 783	enum {ACTION_FAIL, ACTION_REPREP, ACTION_DELAYED_REPREP,
 784	      ACTION_RETRY, ACTION_DELAYED_RETRY} action;
 
 785	struct scsi_sense_hdr sshdr;
 786	bool sense_valid;
 787	bool sense_current = true;      /* false implies "deferred sense" */
 788	blk_status_t blk_stat;
 789
 790	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
 791	if (sense_valid)
 792		sense_current = !scsi_sense_is_deferred(&sshdr);
 793
 794	blk_stat = scsi_result_to_blk_status(result);
 795
 796	if (host_byte(result) == DID_RESET) {
 797		/* Third party bus reset or reset for error recovery
 798		 * reasons.  Just retry the command and see what
 799		 * happens.
 800		 */
 801		action = ACTION_RETRY;
 802	} else if (sense_valid && sense_current) {
 803		switch (sshdr.sense_key) {
 804		case UNIT_ATTENTION:
 805			if (cmd->device->removable) {
 806				/* Detected disc change.  Set a bit
 807				 * and quietly refuse further access.
 808				 */
 809				cmd->device->changed = 1;
 810				action = ACTION_FAIL;
 811			} else {
 812				/* Must have been a power glitch, or a
 813				 * bus reset.  Could not have been a
 814				 * media change, so we just retry the
 815				 * command and see what happens.
 816				 */
 817				action = ACTION_RETRY;
 818			}
 819			break;
 820		case ILLEGAL_REQUEST:
 821			/* If we had an ILLEGAL REQUEST returned, then
 822			 * we may have performed an unsupported
 823			 * command.  The only thing this should be
 824			 * would be a ten byte read where only a six
 825			 * byte read was supported.  Also, on a system
 826			 * where READ CAPACITY failed, we may have
 827			 * read past the end of the disk.
 828			 */
 829			if ((cmd->device->use_10_for_rw &&
 830			    sshdr.asc == 0x20 && sshdr.ascq == 0x00) &&
 831			    (cmd->cmnd[0] == READ_10 ||
 832			     cmd->cmnd[0] == WRITE_10)) {
 833				/* This will issue a new 6-byte command. */
 834				cmd->device->use_10_for_rw = 0;
 835				action = ACTION_REPREP;
 836			} else if (sshdr.asc == 0x10) /* DIX */ {
 837				action = ACTION_FAIL;
 838				blk_stat = BLK_STS_PROTECTION;
 839			/* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */
 840			} else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) {
 841				action = ACTION_FAIL;
 842				blk_stat = BLK_STS_TARGET;
 843			} else
 844				action = ACTION_FAIL;
 845			break;
 846		case ABORTED_COMMAND:
 847			action = ACTION_FAIL;
 848			if (sshdr.asc == 0x10) /* DIF */
 849				blk_stat = BLK_STS_PROTECTION;
 850			break;
 851		case NOT_READY:
 852			/* If the device is in the process of becoming
 853			 * ready, or has a temporary blockage, retry.
 854			 */
 855			if (sshdr.asc == 0x04) {
 856				switch (sshdr.ascq) {
 857				case 0x01: /* becoming ready */
 858				case 0x04: /* format in progress */
 859				case 0x05: /* rebuild in progress */
 860				case 0x06: /* recalculation in progress */
 861				case 0x07: /* operation in progress */
 862				case 0x08: /* Long write in progress */
 863				case 0x09: /* self test in progress */
 864				case 0x11: /* notify (enable spinup) required */
 865				case 0x14: /* space allocation in progress */
 866				case 0x1a: /* start stop unit in progress */
 867				case 0x1b: /* sanitize in progress */
 868				case 0x1d: /* configuration in progress */
 869				case 0x24: /* depopulation in progress */
 870				case 0x25: /* depopulation restore in progress */
 871					action = ACTION_DELAYED_RETRY;
 872					break;
 873				case 0x0a: /* ALUA state transition */
 874					action = ACTION_DELAYED_REPREP;
 875					break;
 876				default:
 877					action = ACTION_FAIL;
 878					break;
 879				}
 880			} else
 881				action = ACTION_FAIL;
 882			break;
 883		case VOLUME_OVERFLOW:
 884			/* See SSC3rXX or current. */
 885			action = ACTION_FAIL;
 886			break;
 887		case DATA_PROTECT:
 888			action = ACTION_FAIL;
 889			if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) ||
 890			    (sshdr.asc == 0x55 &&
 891			     (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) {
 892				/* Insufficient zone resources */
 893				blk_stat = BLK_STS_ZONE_OPEN_RESOURCE;
 894			}
 895			break;
 896		case COMPLETED:
 897			fallthrough;
 898		default:
 899			action = ACTION_FAIL;
 900			break;
 901		}
 902	} else
 903		action = ACTION_FAIL;
 904
 905	if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd))
 
 906		action = ACTION_FAIL;
 907
 908	switch (action) {
 909	case ACTION_FAIL:
 910		/* Give up and fail the remainder of the request */
 911		if (!(req->rq_flags & RQF_QUIET)) {
 912			static DEFINE_RATELIMIT_STATE(_rs,
 913					DEFAULT_RATELIMIT_INTERVAL,
 914					DEFAULT_RATELIMIT_BURST);
 915
 916			if (unlikely(scsi_logging_level))
 917				level =
 918				     SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT,
 919						    SCSI_LOG_MLCOMPLETE_BITS);
 920
 921			/*
 922			 * if logging is enabled the failure will be printed
 923			 * in scsi_log_completion(), so avoid duplicate messages
 924			 */
 925			if (!level && __ratelimit(&_rs)) {
 926				scsi_print_result(cmd, NULL, FAILED);
 927				if (sense_valid)
 928					scsi_print_sense(cmd);
 929				scsi_print_command(cmd);
 930			}
 931		}
 932		if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req)))
 933			return;
 934		fallthrough;
 935	case ACTION_REPREP:
 936		scsi_mq_requeue_cmd(cmd, 0);
 937		break;
 938	case ACTION_DELAYED_REPREP:
 939		scsi_mq_requeue_cmd(cmd, ALUA_TRANSITION_REPREP_DELAY);
 940		break;
 941	case ACTION_RETRY:
 942		/* Retry the same command immediately */
 943		__scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false);
 944		break;
 945	case ACTION_DELAYED_RETRY:
 946		/* Retry the same command after a delay */
 947		__scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false);
 948		break;
 949	}
 950}
 951
 952/*
 953 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a
 954 * new result that may suppress further error checking. Also modifies
 955 * *blk_statp in some cases.
 956 */
 957static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result,
 958					blk_status_t *blk_statp)
 959{
 960	bool sense_valid;
 961	bool sense_current = true;	/* false implies "deferred sense" */
 962	struct request *req = scsi_cmd_to_rq(cmd);
 963	struct scsi_sense_hdr sshdr;
 964
 965	sense_valid = scsi_command_normalize_sense(cmd, &sshdr);
 966	if (sense_valid)
 967		sense_current = !scsi_sense_is_deferred(&sshdr);
 968
 969	if (blk_rq_is_passthrough(req)) {
 970		if (sense_valid) {
 971			/*
 972			 * SG_IO wants current and deferred errors
 973			 */
 974			cmd->sense_len = min(8 + cmd->sense_buffer[7],
 975					     SCSI_SENSE_BUFFERSIZE);
 
 976		}
 977		if (sense_current)
 978			*blk_statp = scsi_result_to_blk_status(result);
 979	} else if (blk_rq_bytes(req) == 0 && sense_current) {
 980		/*
 981		 * Flush commands do not transfers any data, and thus cannot use
 982		 * good_bytes != blk_rq_bytes(req) as the signal for an error.
 983		 * This sets *blk_statp explicitly for the problem case.
 984		 */
 985		*blk_statp = scsi_result_to_blk_status(result);
 986	}
 987	/*
 988	 * Recovered errors need reporting, but they're always treated as
 989	 * success, so fiddle the result code here.  For passthrough requests
 990	 * we already took a copy of the original into sreq->result which
 991	 * is what gets returned to the user
 992	 */
 993	if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) {
 994		bool do_print = true;
 995		/*
 996		 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d]
 997		 * skip print since caller wants ATA registers. Only occurs
 998		 * on SCSI ATA PASS_THROUGH commands when CK_COND=1
 999		 */
1000		if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d))
1001			do_print = false;
1002		else if (req->rq_flags & RQF_QUIET)
1003			do_print = false;
1004		if (do_print)
1005			scsi_print_sense(cmd);
1006		result = 0;
1007		/* for passthrough, *blk_statp may be set */
1008		*blk_statp = BLK_STS_OK;
1009	}
1010	/*
1011	 * Another corner case: the SCSI status byte is non-zero but 'good'.
1012	 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when
1013	 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD
1014	 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related
1015	 * intermediate statuses (both obsolete in SAM-4) as good.
1016	 */
1017	if ((result & 0xff) && scsi_status_is_good(result)) {
1018		result = 0;
1019		*blk_statp = BLK_STS_OK;
1020	}
1021	return result;
1022}
1023
1024/**
1025 * scsi_io_completion - Completion processing for SCSI commands.
1026 * @cmd:	command that is finished.
1027 * @good_bytes:	number of processed bytes.
1028 *
1029 * We will finish off the specified number of sectors. If we are done, the
1030 * command block will be released and the queue function will be goosed. If we
1031 * are not done then we have to figure out what to do next:
1032 *
1033 *   a) We can call scsi_mq_requeue_cmd().  The request will be
1034 *	unprepared and put back on the queue.  Then a new command will
1035 *	be created for it.  This should be used if we made forward
1036 *	progress, or if we want to switch from READ(10) to READ(6) for
1037 *	example.
1038 *
1039 *   b) We can call scsi_io_completion_action().  The request will be
1040 *	put back on the queue and retried using the same command as
1041 *	before, possibly after a delay.
1042 *
1043 *   c) We can call scsi_end_request() with blk_stat other than
1044 *	BLK_STS_OK, to fail the remainder of the request.
1045 */
1046void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes)
1047{
1048	int result = cmd->result;
1049	struct request *req = scsi_cmd_to_rq(cmd);
 
1050	blk_status_t blk_stat = BLK_STS_OK;
1051
1052	if (unlikely(result))	/* a nz result may or may not be an error */
1053		result = scsi_io_completion_nz_result(cmd, result, &blk_stat);
1054
 
 
 
 
 
 
 
1055	/*
1056	 * Next deal with any sectors which we were able to correctly
1057	 * handle.
1058	 */
1059	SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd,
1060		"%u sectors total, %d bytes done.\n",
1061		blk_rq_sectors(req), good_bytes));
1062
1063	/*
1064	 * Failed, zero length commands always need to drop down
1065	 * to retry code. Fast path should return in this block.
1066	 */
1067	if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) {
1068		if (likely(!scsi_end_request(req, blk_stat, good_bytes)))
1069			return; /* no bytes remaining */
1070	}
1071
1072	/* Kill remainder if no retries. */
1073	if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) {
1074		if (scsi_end_request(req, blk_stat, blk_rq_bytes(req)))
1075			WARN_ONCE(true,
1076			    "Bytes remaining after failed, no-retry command");
1077		return;
1078	}
1079
1080	/*
1081	 * If there had been no error, but we have leftover bytes in the
1082	 * request just queue the command up again.
1083	 */
1084	if (likely(result == 0))
1085		scsi_mq_requeue_cmd(cmd, 0);
1086	else
1087		scsi_io_completion_action(cmd, result);
1088}
1089
1090static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev,
1091		struct request *rq)
1092{
1093	return sdev->dma_drain_len && blk_rq_is_passthrough(rq) &&
1094	       !op_is_write(req_op(rq)) &&
1095	       sdev->host->hostt->dma_need_drain(rq);
1096}
1097
1098/**
1099 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists
1100 * @cmd: SCSI command data structure to initialize.
1101 *
1102 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled
1103 * for @cmd.
1104 *
1105 * Returns:
1106 * * BLK_STS_OK       - on success
1107 * * BLK_STS_RESOURCE - if the failure is retryable
1108 * * BLK_STS_IOERR    - if the failure is fatal
1109 */
1110blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd)
1111{
1112	struct scsi_device *sdev = cmd->device;
1113	struct request *rq = scsi_cmd_to_rq(cmd);
1114	unsigned short nr_segs = blk_rq_nr_phys_segments(rq);
1115	struct scatterlist *last_sg = NULL;
1116	blk_status_t ret;
1117	bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq);
1118	int count;
1119
1120	if (WARN_ON_ONCE(!nr_segs))
1121		return BLK_STS_IOERR;
1122
1123	/*
1124	 * Make sure there is space for the drain.  The driver must adjust
1125	 * max_hw_segments to be prepared for this.
1126	 */
1127	if (need_drain)
1128		nr_segs++;
1129
1130	/*
1131	 * If sg table allocation fails, requeue request later.
1132	 */
1133	if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs,
1134			cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT)))
1135		return BLK_STS_RESOURCE;
1136
1137	/*
1138	 * Next, walk the list, and fill in the addresses and sizes of
1139	 * each segment.
1140	 */
1141	count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg);
1142
1143	if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) {
1144		unsigned int pad_len =
1145			(rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1;
1146
1147		last_sg->length += pad_len;
1148		cmd->extra_len += pad_len;
1149	}
1150
1151	if (need_drain) {
1152		sg_unmark_end(last_sg);
1153		last_sg = sg_next(last_sg);
1154		sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len);
1155		sg_mark_end(last_sg);
1156
1157		cmd->extra_len += sdev->dma_drain_len;
1158		count++;
1159	}
1160
1161	BUG_ON(count > cmd->sdb.table.nents);
1162	cmd->sdb.table.nents = count;
1163	cmd->sdb.length = blk_rq_payload_bytes(rq);
1164
1165	if (blk_integrity_rq(rq)) {
1166		struct scsi_data_buffer *prot_sdb = cmd->prot_sdb;
1167		int ivecs;
1168
1169		if (WARN_ON_ONCE(!prot_sdb)) {
1170			/*
1171			 * This can happen if someone (e.g. multipath)
1172			 * queues a command to a device on an adapter
1173			 * that does not support DIX.
1174			 */
1175			ret = BLK_STS_IOERR;
1176			goto out_free_sgtables;
1177		}
1178
1179		ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio);
1180
1181		if (sg_alloc_table_chained(&prot_sdb->table, ivecs,
1182				prot_sdb->table.sgl,
1183				SCSI_INLINE_PROT_SG_CNT)) {
1184			ret = BLK_STS_RESOURCE;
1185			goto out_free_sgtables;
1186		}
1187
1188		count = blk_rq_map_integrity_sg(rq->q, rq->bio,
1189						prot_sdb->table.sgl);
1190		BUG_ON(count > ivecs);
1191		BUG_ON(count > queue_max_integrity_segments(rq->q));
1192
1193		cmd->prot_sdb = prot_sdb;
1194		cmd->prot_sdb->table.nents = count;
1195	}
1196
1197	return BLK_STS_OK;
1198out_free_sgtables:
1199	scsi_free_sgtables(cmd);
1200	return ret;
1201}
1202EXPORT_SYMBOL(scsi_alloc_sgtables);
1203
1204/**
1205 * scsi_initialize_rq - initialize struct scsi_cmnd partially
1206 * @rq: Request associated with the SCSI command to be initialized.
1207 *
1208 * This function initializes the members of struct scsi_cmnd that must be
1209 * initialized before request processing starts and that won't be
1210 * reinitialized if a SCSI command is requeued.
 
 
 
1211 */
1212static void scsi_initialize_rq(struct request *rq)
1213{
1214	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1215
1216	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1217	cmd->cmd_len = MAX_COMMAND_SIZE;
1218	cmd->sense_len = 0;
1219	init_rcu_head(&cmd->rcu);
1220	cmd->jiffies_at_alloc = jiffies;
1221	cmd->retries = 0;
1222}
1223
1224struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf,
1225				   blk_mq_req_flags_t flags)
1226{
1227	struct request *rq;
1228
1229	rq = blk_mq_alloc_request(q, opf, flags);
1230	if (!IS_ERR(rq))
1231		scsi_initialize_rq(rq);
1232	return rq;
1233}
1234EXPORT_SYMBOL_GPL(scsi_alloc_request);
1235
1236/*
1237 * Only called when the request isn't completed by SCSI, and not freed by
1238 * SCSI
1239 */
1240static void scsi_cleanup_rq(struct request *rq)
1241{
1242	if (rq->rq_flags & RQF_DONTPREP) {
1243		scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq));
1244		rq->rq_flags &= ~RQF_DONTPREP;
1245	}
1246}
1247
1248/* Called before a request is prepared. See also scsi_mq_prep_fn(). */
1249void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd)
1250{
1251	struct request *rq = scsi_cmd_to_rq(cmd);
 
 
 
 
 
 
1252
1253	if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) {
1254		cmd->flags |= SCMD_INITIALIZED;
1255		scsi_initialize_rq(rq);
1256	}
1257
 
 
 
 
 
 
 
 
 
 
 
 
 
1258	cmd->device = dev;
1259	INIT_LIST_HEAD(&cmd->eh_entry);
 
 
1260	INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler);
 
 
 
 
 
1261}
1262
1263static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev,
1264		struct request *req)
1265{
1266	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1267
1268	/*
1269	 * Passthrough requests may transfer data, in which case they must
1270	 * a bio attached to them.  Or they might contain a SCSI command
1271	 * that does not transfer data, in which case they may optionally
1272	 * submit a request without an attached bio.
1273	 */
1274	if (req->bio) {
1275		blk_status_t ret = scsi_alloc_sgtables(cmd);
1276		if (unlikely(ret != BLK_STS_OK))
1277			return ret;
1278	} else {
1279		BUG_ON(blk_rq_bytes(req));
1280
1281		memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1282	}
1283
 
 
1284	cmd->transfersize = blk_rq_bytes(req);
 
1285	return BLK_STS_OK;
1286}
1287
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1288static blk_status_t
1289scsi_device_state_check(struct scsi_device *sdev, struct request *req)
1290{
1291	switch (sdev->sdev_state) {
1292	case SDEV_CREATED:
1293		return BLK_STS_OK;
1294	case SDEV_OFFLINE:
1295	case SDEV_TRANSPORT_OFFLINE:
1296		/*
1297		 * If the device is offline we refuse to process any
1298		 * commands.  The device must be brought online
1299		 * before trying any recovery commands.
1300		 */
1301		if (!sdev->offline_already) {
1302			sdev->offline_already = true;
1303			sdev_printk(KERN_ERR, sdev,
1304				    "rejecting I/O to offline device\n");
1305		}
1306		return BLK_STS_IOERR;
1307	case SDEV_DEL:
1308		/*
1309		 * If the device is fully deleted, we refuse to
1310		 * process any commands as well.
1311		 */
1312		sdev_printk(KERN_ERR, sdev,
1313			    "rejecting I/O to dead device\n");
1314		return BLK_STS_IOERR;
1315	case SDEV_BLOCK:
1316	case SDEV_CREATED_BLOCK:
1317		return BLK_STS_RESOURCE;
1318	case SDEV_QUIESCE:
1319		/*
1320		 * If the device is blocked we only accept power management
1321		 * commands.
1322		 */
1323		if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM)))
1324			return BLK_STS_RESOURCE;
1325		return BLK_STS_OK;
1326	default:
1327		/*
1328		 * For any other not fully online state we only allow
1329		 * power management commands.
 
1330		 */
1331		if (req && !(req->rq_flags & RQF_PM))
1332			return BLK_STS_OFFLINE;
1333		return BLK_STS_OK;
1334	}
1335}
1336
1337/*
1338 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token
1339 * and return the token else return -1.
 
 
1340 */
1341static inline int scsi_dev_queue_ready(struct request_queue *q,
1342				  struct scsi_device *sdev)
1343{
1344	int token;
1345
1346	token = sbitmap_get(&sdev->budget_map);
1347	if (token < 0)
1348		return -1;
 
1349
1350	if (!atomic_read(&sdev->device_blocked))
1351		return token;
1352
1353	/*
1354	 * Only unblock if no other commands are pending and
1355	 * if device_blocked has decreased to zero
1356	 */
1357	if (scsi_device_busy(sdev) > 1 ||
1358	    atomic_dec_return(&sdev->device_blocked) > 0) {
1359		sbitmap_put(&sdev->budget_map, token);
1360		return -1;
1361	}
1362
1363	SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev,
1364			 "unblocking device at zero depth\n"));
1365
1366	return token;
 
 
 
1367}
1368
1369/*
1370 * scsi_target_queue_ready: checks if there we can send commands to target
1371 * @sdev: scsi device on starget to check.
1372 */
1373static inline int scsi_target_queue_ready(struct Scsi_Host *shost,
1374					   struct scsi_device *sdev)
1375{
1376	struct scsi_target *starget = scsi_target(sdev);
1377	unsigned int busy;
1378
1379	if (starget->single_lun) {
1380		spin_lock_irq(shost->host_lock);
1381		if (starget->starget_sdev_user &&
1382		    starget->starget_sdev_user != sdev) {
1383			spin_unlock_irq(shost->host_lock);
1384			return 0;
1385		}
1386		starget->starget_sdev_user = sdev;
1387		spin_unlock_irq(shost->host_lock);
1388	}
1389
1390	if (starget->can_queue <= 0)
1391		return 1;
1392
1393	busy = atomic_inc_return(&starget->target_busy) - 1;
1394	if (atomic_read(&starget->target_blocked) > 0) {
1395		if (busy)
1396			goto starved;
1397
1398		/*
1399		 * unblock after target_blocked iterates to zero
1400		 */
1401		if (atomic_dec_return(&starget->target_blocked) > 0)
1402			goto out_dec;
1403
1404		SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget,
1405				 "unblocking target at zero depth\n"));
1406	}
1407
1408	if (busy >= starget->can_queue)
1409		goto starved;
1410
1411	return 1;
1412
1413starved:
1414	spin_lock_irq(shost->host_lock);
1415	list_move_tail(&sdev->starved_entry, &shost->starved_list);
1416	spin_unlock_irq(shost->host_lock);
1417out_dec:
1418	if (starget->can_queue > 0)
1419		atomic_dec(&starget->target_busy);
1420	return 0;
1421}
1422
1423/*
1424 * scsi_host_queue_ready: if we can send requests to shost, return 1 else
1425 * return 0. We must end up running the queue again whenever 0 is
1426 * returned, else IO can hang.
1427 */
1428static inline int scsi_host_queue_ready(struct request_queue *q,
1429				   struct Scsi_Host *shost,
1430				   struct scsi_device *sdev,
1431				   struct scsi_cmnd *cmd)
1432{
 
 
 
1433	if (atomic_read(&shost->host_blocked) > 0) {
1434		if (scsi_host_busy(shost) > 0)
1435			goto starved;
1436
1437		/*
1438		 * unblock after host_blocked iterates to zero
1439		 */
1440		if (atomic_dec_return(&shost->host_blocked) > 0)
1441			goto out_dec;
1442
1443		SCSI_LOG_MLQUEUE(3,
1444			shost_printk(KERN_INFO, shost,
1445				     "unblocking host at zero depth\n"));
1446	}
1447
1448	if (shost->host_self_blocked)
1449		goto starved;
1450
1451	/* We're OK to process the command, so we can't be starved */
1452	if (!list_empty(&sdev->starved_entry)) {
1453		spin_lock_irq(shost->host_lock);
1454		if (!list_empty(&sdev->starved_entry))
1455			list_del_init(&sdev->starved_entry);
1456		spin_unlock_irq(shost->host_lock);
1457	}
1458
1459	__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1460
1461	return 1;
1462
1463starved:
1464	spin_lock_irq(shost->host_lock);
1465	if (list_empty(&sdev->starved_entry))
1466		list_add_tail(&sdev->starved_entry, &shost->starved_list);
1467	spin_unlock_irq(shost->host_lock);
1468out_dec:
1469	scsi_dec_host_busy(shost, cmd);
1470	return 0;
1471}
1472
1473/*
1474 * Busy state exporting function for request stacking drivers.
1475 *
1476 * For efficiency, no lock is taken to check the busy state of
1477 * shost/starget/sdev, since the returned value is not guaranteed and
1478 * may be changed after request stacking drivers call the function,
1479 * regardless of taking lock or not.
1480 *
1481 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi
1482 * needs to return 'not busy'. Otherwise, request stacking drivers
1483 * may hold requests forever.
1484 */
1485static bool scsi_mq_lld_busy(struct request_queue *q)
1486{
1487	struct scsi_device *sdev = q->queuedata;
1488	struct Scsi_Host *shost;
1489
1490	if (blk_queue_dying(q))
1491		return false;
1492
1493	shost = sdev->host;
1494
1495	/*
1496	 * Ignore host/starget busy state.
1497	 * Since block layer does not have a concept of fairness across
1498	 * multiple queues, congestion of host/starget needs to be handled
1499	 * in SCSI layer.
1500	 */
1501	if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev))
1502		return true;
1503
1504	return false;
1505}
1506
1507/*
1508 * Block layer request completion callback. May be called from interrupt
1509 * context.
1510 */
1511static void scsi_complete(struct request *rq)
1512{
1513	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1514	enum scsi_disposition disposition;
 
1515
1516	INIT_LIST_HEAD(&cmd->eh_entry);
1517
1518	atomic_inc(&cmd->device->iodone_cnt);
1519	if (cmd->result)
1520		atomic_inc(&cmd->device->ioerr_cnt);
1521
1522	disposition = scsi_decide_disposition(cmd);
1523	if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd))
 
 
 
 
1524		disposition = SUCCESS;
 
1525
1526	scsi_log_completion(cmd, disposition);
1527
1528	switch (disposition) {
1529	case SUCCESS:
1530		scsi_finish_command(cmd);
1531		break;
1532	case NEEDS_RETRY:
1533		scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY);
1534		break;
1535	case ADD_TO_MLQUEUE:
1536		scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY);
1537		break;
1538	default:
1539		scsi_eh_scmd_add(cmd);
1540		break;
1541	}
1542}
1543
1544/**
1545 * scsi_dispatch_cmd - Dispatch a command to the low-level driver.
1546 * @cmd: command block we are dispatching.
1547 *
1548 * Return: nonzero return request was rejected and device's queue needs to be
1549 * plugged.
1550 */
1551static int scsi_dispatch_cmd(struct scsi_cmnd *cmd)
1552{
1553	struct Scsi_Host *host = cmd->device->host;
1554	int rtn = 0;
1555
1556	atomic_inc(&cmd->device->iorequest_cnt);
1557
1558	/* check if the device is still usable */
1559	if (unlikely(cmd->device->sdev_state == SDEV_DEL)) {
1560		/* in SDEV_DEL we error all commands. DID_NO_CONNECT
1561		 * returns an immediate error upwards, and signals
1562		 * that the device is no longer present */
1563		cmd->result = DID_NO_CONNECT << 16;
1564		goto done;
1565	}
1566
1567	/* Check to see if the scsi lld made this device blocked. */
1568	if (unlikely(scsi_device_blocked(cmd->device))) {
1569		/*
1570		 * in blocked state, the command is just put back on
1571		 * the device queue.  The suspend state has already
1572		 * blocked the queue so future requests should not
1573		 * occur until the device transitions out of the
1574		 * suspend state.
1575		 */
1576		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1577			"queuecommand : device blocked\n"));
1578		atomic_dec(&cmd->device->iorequest_cnt);
1579		return SCSI_MLQUEUE_DEVICE_BUSY;
1580	}
1581
1582	/* Store the LUN value in cmnd, if needed. */
1583	if (cmd->device->lun_in_cdb)
1584		cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) |
1585			       (cmd->device->lun << 5 & 0xe0);
1586
1587	scsi_log_send(cmd);
1588
1589	/*
1590	 * Before we queue this command, check if the command
1591	 * length exceeds what the host adapter can handle.
1592	 */
1593	if (cmd->cmd_len > cmd->device->host->max_cmd_len) {
1594		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1595			       "queuecommand : command too long. "
1596			       "cdb_size=%d host->max_cmd_len=%d\n",
1597			       cmd->cmd_len, cmd->device->host->max_cmd_len));
1598		cmd->result = (DID_ABORT << 16);
1599		goto done;
1600	}
1601
1602	if (unlikely(host->shost_state == SHOST_DEL)) {
1603		cmd->result = (DID_NO_CONNECT << 16);
1604		goto done;
1605
1606	}
1607
1608	trace_scsi_dispatch_cmd_start(cmd);
1609	rtn = host->hostt->queuecommand(host, cmd);
1610	if (rtn) {
1611		atomic_dec(&cmd->device->iorequest_cnt);
1612		trace_scsi_dispatch_cmd_error(cmd, rtn);
1613		if (rtn != SCSI_MLQUEUE_DEVICE_BUSY &&
1614		    rtn != SCSI_MLQUEUE_TARGET_BUSY)
1615			rtn = SCSI_MLQUEUE_HOST_BUSY;
1616
1617		SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd,
1618			"queuecommand : request rejected\n"));
1619	}
1620
1621	return rtn;
1622 done:
1623	scsi_done(cmd);
1624	return 0;
1625}
1626
1627/* Size in bytes of the sg-list stored in the scsi-mq command-private data. */
1628static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost)
1629{
1630	return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) *
1631		sizeof(struct scatterlist);
1632}
1633
1634static blk_status_t scsi_prepare_cmd(struct request *req)
1635{
1636	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1637	struct scsi_device *sdev = req->q->queuedata;
1638	struct Scsi_Host *shost = sdev->host;
1639	bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1640	struct scatterlist *sg;
1641
1642	scsi_init_command(sdev, cmd);
1643
1644	cmd->eh_eflags = 0;
1645	cmd->prot_type = 0;
1646	cmd->prot_flags = 0;
1647	cmd->submitter = 0;
1648	memset(&cmd->sdb, 0, sizeof(cmd->sdb));
1649	cmd->underflow = 0;
1650	cmd->transfersize = 0;
1651	cmd->host_scribble = NULL;
1652	cmd->result = 0;
1653	cmd->extra_len = 0;
1654	cmd->state = 0;
1655	if (in_flight)
1656		__set_bit(SCMD_STATE_INFLIGHT, &cmd->state);
1657
1658	/*
1659	 * Only clear the driver-private command data if the LLD does not supply
1660	 * a function to initialize that data.
1661	 */
1662	if (!shost->hostt->init_cmd_priv)
1663		memset(cmd + 1, 0, shost->hostt->cmd_size);
1664
1665	cmd->prot_op = SCSI_PROT_NORMAL;
1666	if (blk_rq_bytes(req))
1667		cmd->sc_data_direction = rq_dma_dir(req);
1668	else
1669		cmd->sc_data_direction = DMA_NONE;
1670
1671	sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size;
1672	cmd->sdb.table.sgl = sg;
1673
1674	if (scsi_host_get_prot(shost)) {
1675		memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer));
1676
1677		cmd->prot_sdb->table.sgl =
1678			(struct scatterlist *)(cmd->prot_sdb + 1);
1679	}
1680
1681	/*
1682	 * Special handling for passthrough commands, which don't go to the ULP
1683	 * at all:
1684	 */
1685	if (blk_rq_is_passthrough(req))
1686		return scsi_setup_scsi_cmnd(sdev, req);
1687
1688	if (sdev->handler && sdev->handler->prep_fn) {
1689		blk_status_t ret = sdev->handler->prep_fn(sdev, req);
1690
1691		if (ret != BLK_STS_OK)
1692			return ret;
1693	}
1694
1695	/* Usually overridden by the ULP */
1696	cmd->allowed = 0;
1697	memset(cmd->cmnd, 0, sizeof(cmd->cmnd));
1698	return scsi_cmd_to_driver(cmd)->init_command(cmd);
1699}
1700
1701static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly)
1702{
1703	struct request *req = scsi_cmd_to_rq(cmd);
1704
1705	switch (cmd->submitter) {
1706	case SUBMITTED_BY_BLOCK_LAYER:
1707		break;
1708	case SUBMITTED_BY_SCSI_ERROR_HANDLER:
1709		return scsi_eh_done(cmd);
1710	case SUBMITTED_BY_SCSI_RESET_IOCTL:
1711		return;
1712	}
1713
1714	if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q)))
1715		return;
1716	if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state)))
1717		return;
1718	trace_scsi_dispatch_cmd_done(cmd);
1719
1720	if (complete_directly)
1721		blk_mq_complete_request_direct(req, scsi_complete);
1722	else
1723		blk_mq_complete_request(req);
1724}
1725
1726void scsi_done(struct scsi_cmnd *cmd)
1727{
1728	scsi_done_internal(cmd, false);
1729}
1730EXPORT_SYMBOL(scsi_done);
1731
1732void scsi_done_direct(struct scsi_cmnd *cmd)
1733{
1734	scsi_done_internal(cmd, true);
1735}
1736EXPORT_SYMBOL(scsi_done_direct);
1737
1738static void scsi_mq_put_budget(struct request_queue *q, int budget_token)
1739{
1740	struct scsi_device *sdev = q->queuedata;
1741
1742	sbitmap_put(&sdev->budget_map, budget_token);
1743}
1744
1745/*
1746 * When to reinvoke queueing after a resource shortage. It's 3 msecs to
1747 * not change behaviour from the previous unplug mechanism, experimentation
1748 * may prove this needs changing.
1749 */
1750#define SCSI_QUEUE_DELAY 3
1751
1752static int scsi_mq_get_budget(struct request_queue *q)
1753{
1754	struct scsi_device *sdev = q->queuedata;
1755	int token = scsi_dev_queue_ready(q, sdev);
1756
1757	if (token >= 0)
1758		return token;
1759
1760	atomic_inc(&sdev->restarts);
1761
1762	/*
1763	 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy).
1764	 * .restarts must be incremented before .device_busy is read because the
1765	 * code in scsi_run_queue_async() depends on the order of these operations.
1766	 */
1767	smp_mb__after_atomic();
1768
1769	/*
1770	 * If all in-flight requests originated from this LUN are completed
1771	 * before reading .device_busy, sdev->device_busy will be observed as
1772	 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request
1773	 * soon. Otherwise, completion of one of these requests will observe
1774	 * the .restarts flag, and the request queue will be run for handling
1775	 * this request, see scsi_end_request().
1776	 */
1777	if (unlikely(scsi_device_busy(sdev) == 0 &&
1778				!scsi_device_blocked(sdev)))
1779		blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY);
1780	return -1;
1781}
1782
1783static void scsi_mq_set_rq_budget_token(struct request *req, int token)
1784{
1785	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1786
1787	cmd->budget_token = token;
1788}
1789
1790static int scsi_mq_get_rq_budget_token(struct request *req)
1791{
1792	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1793
1794	return cmd->budget_token;
1795}
1796
1797static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx,
1798			 const struct blk_mq_queue_data *bd)
1799{
1800	struct request *req = bd->rq;
1801	struct request_queue *q = req->q;
1802	struct scsi_device *sdev = q->queuedata;
1803	struct Scsi_Host *shost = sdev->host;
1804	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req);
1805	blk_status_t ret;
1806	int reason;
1807
1808	WARN_ON_ONCE(cmd->budget_token < 0);
1809
1810	/*
1811	 * If the device is not in running state we will reject some or all
1812	 * commands.
1813	 */
1814	if (unlikely(sdev->sdev_state != SDEV_RUNNING)) {
1815		ret = scsi_device_state_check(sdev, req);
1816		if (ret != BLK_STS_OK)
1817			goto out_put_budget;
1818	}
1819
1820	ret = BLK_STS_RESOURCE;
1821	if (!scsi_target_queue_ready(shost, sdev))
1822		goto out_put_budget;
1823	if (unlikely(scsi_host_in_recovery(shost))) {
1824		if (cmd->flags & SCMD_FAIL_IF_RECOVERING)
1825			ret = BLK_STS_OFFLINE;
1826		goto out_dec_target_busy;
1827	}
1828	if (!scsi_host_queue_ready(q, shost, sdev, cmd))
1829		goto out_dec_target_busy;
1830
1831	if (!(req->rq_flags & RQF_DONTPREP)) {
1832		ret = scsi_prepare_cmd(req);
1833		if (ret != BLK_STS_OK)
1834			goto out_dec_host_busy;
1835		req->rq_flags |= RQF_DONTPREP;
1836	} else {
1837		clear_bit(SCMD_STATE_COMPLETE, &cmd->state);
 
1838	}
1839
1840	cmd->flags &= SCMD_PRESERVED_FLAGS;
1841	if (sdev->simple_tags)
1842		cmd->flags |= SCMD_TAGGED;
1843	if (bd->last)
1844		cmd->flags |= SCMD_LAST;
1845
1846	scsi_set_resid(cmd, 0);
1847	memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
1848	cmd->submitter = SUBMITTED_BY_BLOCK_LAYER;
1849
1850	blk_mq_start_request(req);
1851	reason = scsi_dispatch_cmd(cmd);
1852	if (reason) {
1853		scsi_set_blocked(cmd, reason);
1854		ret = BLK_STS_RESOURCE;
1855		goto out_dec_host_busy;
1856	}
1857
1858	return BLK_STS_OK;
1859
1860out_dec_host_busy:
1861	scsi_dec_host_busy(shost, cmd);
1862out_dec_target_busy:
1863	if (scsi_target(sdev)->can_queue > 0)
1864		atomic_dec(&scsi_target(sdev)->target_busy);
1865out_put_budget:
1866	scsi_mq_put_budget(q, cmd->budget_token);
1867	cmd->budget_token = -1;
1868	switch (ret) {
1869	case BLK_STS_OK:
1870		break;
1871	case BLK_STS_RESOURCE:
1872	case BLK_STS_ZONE_RESOURCE:
1873		if (scsi_device_blocked(sdev))
 
1874			ret = BLK_STS_DEV_RESOURCE;
1875		break;
1876	case BLK_STS_AGAIN:
1877		cmd->result = DID_BUS_BUSY << 16;
1878		if (req->rq_flags & RQF_DONTPREP)
1879			scsi_mq_uninit_cmd(cmd);
1880		break;
1881	default:
1882		if (unlikely(!scsi_device_online(sdev)))
1883			cmd->result = DID_NO_CONNECT << 16;
1884		else
1885			cmd->result = DID_ERROR << 16;
1886		/*
1887		 * Make sure to release all allocated resources when
1888		 * we hit an error, as we will never see this command
1889		 * again.
1890		 */
1891		if (req->rq_flags & RQF_DONTPREP)
1892			scsi_mq_uninit_cmd(cmd);
1893		scsi_run_queue_async(sdev);
1894		break;
1895	}
1896	return ret;
1897}
1898
 
 
 
 
 
 
 
 
1899static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq,
1900				unsigned int hctx_idx, unsigned int numa_node)
1901{
1902	struct Scsi_Host *shost = set->driver_data;
 
1903	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1904	struct scatterlist *sg;
1905	int ret = 0;
1906
1907	cmd->sense_buffer =
1908		kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node);
 
 
1909	if (!cmd->sense_buffer)
1910		return -ENOMEM;
 
1911
1912	if (scsi_host_get_prot(shost)) {
1913		sg = (void *)cmd + sizeof(struct scsi_cmnd) +
1914			shost->hostt->cmd_size;
1915		cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost);
1916	}
1917
1918	if (shost->hostt->init_cmd_priv) {
1919		ret = shost->hostt->init_cmd_priv(shost, cmd);
1920		if (ret < 0)
1921			kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
 
1922	}
1923
1924	return ret;
1925}
1926
1927static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq,
1928				 unsigned int hctx_idx)
1929{
1930	struct Scsi_Host *shost = set->driver_data;
1931	struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq);
1932
1933	if (shost->hostt->exit_cmd_priv)
1934		shost->hostt->exit_cmd_priv(shost, cmd);
1935	kmem_cache_free(scsi_sense_cache, cmd->sense_buffer);
1936}
1937
1938
1939static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
1940{
1941	struct Scsi_Host *shost = hctx->driver_data;
1942
1943	if (shost->hostt->mq_poll)
1944		return shost->hostt->mq_poll(shost, hctx->queue_num);
1945
1946	return 0;
1947}
1948
1949static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data,
1950			  unsigned int hctx_idx)
1951{
1952	struct Scsi_Host *shost = data;
1953
1954	hctx->driver_data = shost;
1955	return 0;
1956}
1957
1958static void scsi_map_queues(struct blk_mq_tag_set *set)
1959{
1960	struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set);
1961
1962	if (shost->hostt->map_queues)
1963		return shost->hostt->map_queues(shost);
1964	blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]);
1965}
1966
1967void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q)
1968{
1969	struct device *dev = shost->dma_dev;
1970
1971	/*
1972	 * this limit is imposed by hardware restrictions
1973	 */
1974	blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize,
1975					SG_MAX_SEGMENTS));
1976
1977	if (scsi_host_prot_dma(shost)) {
1978		shost->sg_prot_tablesize =
1979			min_not_zero(shost->sg_prot_tablesize,
1980				     (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS);
1981		BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize);
1982		blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize);
1983	}
1984
 
 
 
 
1985	blk_queue_max_hw_sectors(q, shost->max_sectors);
 
 
1986	blk_queue_segment_boundary(q, shost->dma_boundary);
1987	dma_set_seg_boundary(dev, shost->dma_boundary);
1988
1989	blk_queue_max_segment_size(q, shost->max_segment_size);
1990	blk_queue_virt_boundary(q, shost->virt_boundary_mask);
1991	dma_set_max_seg_size(dev, queue_max_segment_size(q));
1992
1993	/*
1994	 * Set a reasonable default alignment:  The larger of 32-byte (dword),
1995	 * which is a common minimum for HBAs, and the minimum DMA alignment,
1996	 * which is set by the platform.
1997	 *
1998	 * Devices that require a bigger alignment can increase it later.
1999	 */
2000	blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1);
2001}
2002EXPORT_SYMBOL_GPL(__scsi_init_queue);
2003
2004static const struct blk_mq_ops scsi_mq_ops_no_commit = {
2005	.get_budget	= scsi_mq_get_budget,
2006	.put_budget	= scsi_mq_put_budget,
2007	.queue_rq	= scsi_queue_rq,
2008	.complete	= scsi_complete,
2009	.timeout	= scsi_timeout,
2010#ifdef CONFIG_BLK_DEBUG_FS
2011	.show_rq	= scsi_show_rq,
2012#endif
2013	.init_request	= scsi_mq_init_request,
2014	.exit_request	= scsi_mq_exit_request,
 
2015	.cleanup_rq	= scsi_cleanup_rq,
2016	.busy		= scsi_mq_lld_busy,
2017	.map_queues	= scsi_map_queues,
2018	.init_hctx	= scsi_init_hctx,
2019	.poll		= scsi_mq_poll,
2020	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
2021	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
2022};
2023
2024
2025static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx)
2026{
2027	struct Scsi_Host *shost = hctx->driver_data;
 
 
2028
2029	shost->hostt->commit_rqs(shost, hctx->queue_num);
2030}
2031
2032static const struct blk_mq_ops scsi_mq_ops = {
2033	.get_budget	= scsi_mq_get_budget,
2034	.put_budget	= scsi_mq_put_budget,
2035	.queue_rq	= scsi_queue_rq,
2036	.commit_rqs	= scsi_commit_rqs,
2037	.complete	= scsi_complete,
2038	.timeout	= scsi_timeout,
2039#ifdef CONFIG_BLK_DEBUG_FS
2040	.show_rq	= scsi_show_rq,
2041#endif
2042	.init_request	= scsi_mq_init_request,
2043	.exit_request	= scsi_mq_exit_request,
 
2044	.cleanup_rq	= scsi_cleanup_rq,
2045	.busy		= scsi_mq_lld_busy,
2046	.map_queues	= scsi_map_queues,
2047	.init_hctx	= scsi_init_hctx,
2048	.poll		= scsi_mq_poll,
2049	.set_rq_budget_token = scsi_mq_set_rq_budget_token,
2050	.get_rq_budget_token = scsi_mq_get_rq_budget_token,
2051};
2052
 
 
 
 
 
 
 
 
 
 
 
 
2053int scsi_mq_setup_tags(struct Scsi_Host *shost)
2054{
2055	unsigned int cmd_size, sgl_size;
2056	struct blk_mq_tag_set *tag_set = &shost->tag_set;
2057
2058	sgl_size = max_t(unsigned int, sizeof(struct scatterlist),
2059				scsi_mq_inline_sgl_size(shost));
2060	cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size;
2061	if (scsi_host_get_prot(shost))
2062		cmd_size += sizeof(struct scsi_data_buffer) +
2063			sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT;
2064
2065	memset(tag_set, 0, sizeof(*tag_set));
2066	if (shost->hostt->commit_rqs)
2067		tag_set->ops = &scsi_mq_ops;
2068	else
2069		tag_set->ops = &scsi_mq_ops_no_commit;
2070	tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1;
2071	tag_set->nr_maps = shost->nr_maps ? : 1;
2072	tag_set->queue_depth = shost->can_queue;
2073	tag_set->cmd_size = cmd_size;
2074	tag_set->numa_node = dev_to_node(shost->dma_dev);
2075	tag_set->flags = BLK_MQ_F_SHOULD_MERGE;
2076	tag_set->flags |=
2077		BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy);
2078	if (shost->queuecommand_may_block)
2079		tag_set->flags |= BLK_MQ_F_BLOCKING;
2080	tag_set->driver_data = shost;
2081	if (shost->host_tagset)
2082		tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED;
2083
2084	return blk_mq_alloc_tag_set(tag_set);
2085}
2086
2087void scsi_mq_free_tags(struct kref *kref)
2088{
2089	struct Scsi_Host *shost = container_of(kref, typeof(*shost),
2090					       tagset_refcnt);
2091
2092	blk_mq_free_tag_set(&shost->tag_set);
2093	complete(&shost->tagset_freed);
2094}
2095
2096/**
2097 * scsi_device_from_queue - return sdev associated with a request_queue
2098 * @q: The request queue to return the sdev from
2099 *
2100 * Return the sdev associated with a request queue or NULL if the
2101 * request_queue does not reference a SCSI device.
2102 */
2103struct scsi_device *scsi_device_from_queue(struct request_queue *q)
2104{
2105	struct scsi_device *sdev = NULL;
2106
2107	if (q->mq_ops == &scsi_mq_ops_no_commit ||
2108	    q->mq_ops == &scsi_mq_ops)
2109		sdev = q->queuedata;
2110	if (!sdev || !get_device(&sdev->sdev_gendev))
2111		sdev = NULL;
2112
2113	return sdev;
2114}
2115/*
2116 * pktcdvd should have been integrated into the SCSI layers, but for historical
2117 * reasons like the old IDE driver it isn't.  This export allows it to safely
2118 * probe if a given device is a SCSI one and only attach to that.
2119 */
2120#ifdef CONFIG_CDROM_PKTCDVD_MODULE
2121EXPORT_SYMBOL_GPL(scsi_device_from_queue);
2122#endif
2123
2124/**
2125 * scsi_block_requests - Utility function used by low-level drivers to prevent
2126 * further commands from being queued to the device.
2127 * @shost:  host in question
2128 *
2129 * There is no timer nor any other means by which the requests get unblocked
2130 * other than the low-level driver calling scsi_unblock_requests().
2131 */
2132void scsi_block_requests(struct Scsi_Host *shost)
2133{
2134	shost->host_self_blocked = 1;
2135}
2136EXPORT_SYMBOL(scsi_block_requests);
2137
2138/**
2139 * scsi_unblock_requests - Utility function used by low-level drivers to allow
2140 * further commands to be queued to the device.
2141 * @shost:  host in question
2142 *
2143 * There is no timer nor any other means by which the requests get unblocked
2144 * other than the low-level driver calling scsi_unblock_requests(). This is done
2145 * as an API function so that changes to the internals of the scsi mid-layer
2146 * won't require wholesale changes to drivers that use this feature.
2147 */
2148void scsi_unblock_requests(struct Scsi_Host *shost)
2149{
2150	shost->host_self_blocked = 0;
2151	scsi_run_host_queues(shost);
2152}
2153EXPORT_SYMBOL(scsi_unblock_requests);
2154
2155void scsi_exit_queue(void)
2156{
2157	kmem_cache_destroy(scsi_sense_cache);
 
2158}
2159
2160/**
2161 *	scsi_mode_select - issue a mode select
2162 *	@sdev:	SCSI device to be queried
2163 *	@pf:	Page format bit (1 == standard, 0 == vendor specific)
2164 *	@sp:	Save page bit (0 == don't save, 1 == save)
 
2165 *	@buffer: request buffer (may not be smaller than eight bytes)
2166 *	@len:	length of request buffer.
2167 *	@timeout: command timeout
2168 *	@retries: number of retries before failing
2169 *	@data: returns a structure abstracting the mode header data
2170 *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2171 *		must be SCSI_SENSE_BUFFERSIZE big.
2172 *
2173 *	Returns zero if successful; negative error number or scsi
2174 *	status on error
2175 *
2176 */
2177int scsi_mode_select(struct scsi_device *sdev, int pf, int sp,
2178		     unsigned char *buffer, int len, int timeout, int retries,
2179		     struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
 
2180{
2181	unsigned char cmd[10];
2182	unsigned char *real_buffer;
2183	const struct scsi_exec_args exec_args = {
2184		.sshdr = sshdr,
2185	};
2186	int ret;
2187
2188	memset(cmd, 0, sizeof(cmd));
2189	cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0);
2190
2191	/*
2192	 * Use MODE SELECT(10) if the device asked for it or if the mode page
2193	 * and the mode select header cannot fit within the maximumm 255 bytes
2194	 * of the MODE SELECT(6) command.
2195	 */
2196	if (sdev->use_10_for_ms ||
2197	    len + 4 > 255 ||
2198	    data->block_descriptor_length > 255) {
2199		if (len > 65535 - 8)
2200			return -EINVAL;
2201		real_buffer = kmalloc(8 + len, GFP_KERNEL);
2202		if (!real_buffer)
2203			return -ENOMEM;
2204		memcpy(real_buffer + 8, buffer, len);
2205		len += 8;
2206		real_buffer[0] = 0;
2207		real_buffer[1] = 0;
2208		real_buffer[2] = data->medium_type;
2209		real_buffer[3] = data->device_specific;
2210		real_buffer[4] = data->longlba ? 0x01 : 0;
2211		real_buffer[5] = 0;
2212		put_unaligned_be16(data->block_descriptor_length,
2213				   &real_buffer[6]);
2214
2215		cmd[0] = MODE_SELECT_10;
2216		put_unaligned_be16(len, &cmd[7]);
 
2217	} else {
2218		if (data->longlba)
 
2219			return -EINVAL;
2220
2221		real_buffer = kmalloc(4 + len, GFP_KERNEL);
2222		if (!real_buffer)
2223			return -ENOMEM;
2224		memcpy(real_buffer + 4, buffer, len);
2225		len += 4;
2226		real_buffer[0] = 0;
2227		real_buffer[1] = data->medium_type;
2228		real_buffer[2] = data->device_specific;
2229		real_buffer[3] = data->block_descriptor_length;
2230
2231		cmd[0] = MODE_SELECT;
2232		cmd[4] = len;
2233	}
2234
2235	ret = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_OUT, real_buffer, len,
2236			       timeout, retries, &exec_args);
2237	kfree(real_buffer);
2238	return ret;
2239}
2240EXPORT_SYMBOL_GPL(scsi_mode_select);
2241
2242/**
2243 *	scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary.
2244 *	@sdev:	SCSI device to be queried
2245 *	@dbd:	set to prevent mode sense from returning block descriptors
2246 *	@modepage: mode page being requested
2247 *	@subpage: sub-page of the mode page being requested
2248 *	@buffer: request buffer (may not be smaller than eight bytes)
2249 *	@len:	length of request buffer.
2250 *	@timeout: command timeout
2251 *	@retries: number of retries before failing
2252 *	@data: returns a structure abstracting the mode header data
2253 *	@sshdr: place to put sense data (or NULL if no sense to be collected).
2254 *		must be SCSI_SENSE_BUFFERSIZE big.
2255 *
2256 *	Returns zero if successful, or a negative error number on failure
 
 
2257 */
2258int
2259scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, int subpage,
2260		  unsigned char *buffer, int len, int timeout, int retries,
2261		  struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr)
2262{
2263	unsigned char cmd[12];
2264	int use_10_for_ms;
2265	int header_length;
2266	int result;
2267	struct scsi_sense_hdr my_sshdr;
2268	struct scsi_failure failure_defs[] = {
2269		{
2270			.sense = UNIT_ATTENTION,
2271			.asc = SCMD_FAILURE_ASC_ANY,
2272			.ascq = SCMD_FAILURE_ASCQ_ANY,
2273			.allowed = retries,
2274			.result = SAM_STAT_CHECK_CONDITION,
2275		},
2276		{}
2277	};
2278	struct scsi_failures failures = {
2279		.failure_definitions = failure_defs,
2280	};
2281	const struct scsi_exec_args exec_args = {
2282		/* caller might not be interested in sense, but we need it */
2283		.sshdr = sshdr ? : &my_sshdr,
2284		.failures = &failures,
2285	};
2286
2287	memset(data, 0, sizeof(*data));
2288	memset(&cmd[0], 0, 12);
2289
2290	dbd = sdev->set_dbd_for_ms ? 8 : dbd;
2291	cmd[1] = dbd & 0x18;	/* allows DBD and LLBA bits */
2292	cmd[2] = modepage;
2293	cmd[3] = subpage;
2294
2295	sshdr = exec_args.sshdr;
 
 
2296
2297 retry:
2298	use_10_for_ms = sdev->use_10_for_ms || len > 255;
2299
2300	if (use_10_for_ms) {
2301		if (len < 8 || len > 65535)
2302			return -EINVAL;
2303
2304		cmd[0] = MODE_SENSE_10;
2305		put_unaligned_be16(len, &cmd[7]);
2306		header_length = 8;
2307	} else {
2308		if (len < 4)
2309			return -EINVAL;
2310
2311		cmd[0] = MODE_SENSE;
2312		cmd[4] = len;
2313		header_length = 4;
2314	}
2315
2316	memset(buffer, 0, len);
2317
2318	result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, buffer, len,
2319				  timeout, retries, &exec_args);
2320	if (result < 0)
2321		return result;
2322
2323	/* This code looks awful: what it's doing is making sure an
2324	 * ILLEGAL REQUEST sense return identifies the actual command
2325	 * byte as the problem.  MODE_SENSE commands can return
2326	 * ILLEGAL REQUEST if the code page isn't supported */
2327
2328	if (!scsi_status_is_good(result)) {
 
2329		if (scsi_sense_valid(sshdr)) {
2330			if ((sshdr->sense_key == ILLEGAL_REQUEST) &&
2331			    (sshdr->asc == 0x20) && (sshdr->ascq == 0)) {
2332				/*
2333				 * Invalid command operation code: retry using
2334				 * MODE SENSE(6) if this was a MODE SENSE(10)
2335				 * request, except if the request mode page is
2336				 * too large for MODE SENSE single byte
2337				 * allocation length field.
2338				 */
2339				if (use_10_for_ms) {
2340					if (len > 255)
2341						return -EIO;
2342					sdev->use_10_for_ms = 0;
2343					goto retry;
2344				}
2345			}
2346		}
2347		return -EIO;
2348	}
2349	if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b &&
2350		     (modepage == 6 || modepage == 8))) {
2351		/* Initio breakage? */
2352		header_length = 0;
2353		data->length = 13;
2354		data->medium_type = 0;
2355		data->device_specific = 0;
2356		data->longlba = 0;
2357		data->block_descriptor_length = 0;
2358	} else if (use_10_for_ms) {
2359		data->length = get_unaligned_be16(&buffer[0]) + 2;
2360		data->medium_type = buffer[2];
2361		data->device_specific = buffer[3];
2362		data->longlba = buffer[4] & 0x01;
2363		data->block_descriptor_length = get_unaligned_be16(&buffer[6]);
2364	} else {
2365		data->length = buffer[0] + 1;
2366		data->medium_type = buffer[1];
2367		data->device_specific = buffer[2];
2368		data->block_descriptor_length = buffer[3];
 
 
 
 
 
 
 
 
 
 
2369	}
2370	data->header_length = header_length;
2371
2372	return 0;
2373}
2374EXPORT_SYMBOL(scsi_mode_sense);
2375
2376/**
2377 *	scsi_test_unit_ready - test if unit is ready
2378 *	@sdev:	scsi device to change the state of.
2379 *	@timeout: command timeout
2380 *	@retries: number of retries before failing
2381 *	@sshdr: outpout pointer for decoded sense information.
2382 *
2383 *	Returns zero if unsuccessful or an error if TUR failed.  For
2384 *	removable media, UNIT_ATTENTION sets ->changed flag.
2385 **/
2386int
2387scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries,
2388		     struct scsi_sense_hdr *sshdr)
2389{
2390	char cmd[] = {
2391		TEST_UNIT_READY, 0, 0, 0, 0, 0,
2392	};
2393	const struct scsi_exec_args exec_args = {
2394		.sshdr = sshdr,
2395	};
2396	int result;
2397
2398	/* try to eat the UNIT_ATTENTION if there are enough retries */
2399	do {
2400		result = scsi_execute_cmd(sdev, cmd, REQ_OP_DRV_IN, NULL, 0,
2401					  timeout, 1, &exec_args);
2402		if (sdev->removable && result > 0 && scsi_sense_valid(sshdr) &&
2403		    sshdr->sense_key == UNIT_ATTENTION)
2404			sdev->changed = 1;
2405	} while (result > 0 && scsi_sense_valid(sshdr) &&
2406		 sshdr->sense_key == UNIT_ATTENTION && --retries);
2407
2408	return result;
2409}
2410EXPORT_SYMBOL(scsi_test_unit_ready);
2411
2412/**
2413 *	scsi_device_set_state - Take the given device through the device state model.
2414 *	@sdev:	scsi device to change the state of.
2415 *	@state:	state to change to.
2416 *
2417 *	Returns zero if successful or an error if the requested
2418 *	transition is illegal.
2419 */
2420int
2421scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state)
2422{
2423	enum scsi_device_state oldstate = sdev->sdev_state;
2424
2425	if (state == oldstate)
2426		return 0;
2427
2428	switch (state) {
2429	case SDEV_CREATED:
2430		switch (oldstate) {
2431		case SDEV_CREATED_BLOCK:
2432			break;
2433		default:
2434			goto illegal;
2435		}
2436		break;
2437
2438	case SDEV_RUNNING:
2439		switch (oldstate) {
2440		case SDEV_CREATED:
2441		case SDEV_OFFLINE:
2442		case SDEV_TRANSPORT_OFFLINE:
2443		case SDEV_QUIESCE:
2444		case SDEV_BLOCK:
2445			break;
2446		default:
2447			goto illegal;
2448		}
2449		break;
2450
2451	case SDEV_QUIESCE:
2452		switch (oldstate) {
2453		case SDEV_RUNNING:
2454		case SDEV_OFFLINE:
2455		case SDEV_TRANSPORT_OFFLINE:
2456			break;
2457		default:
2458			goto illegal;
2459		}
2460		break;
2461
2462	case SDEV_OFFLINE:
2463	case SDEV_TRANSPORT_OFFLINE:
2464		switch (oldstate) {
2465		case SDEV_CREATED:
2466		case SDEV_RUNNING:
2467		case SDEV_QUIESCE:
2468		case SDEV_BLOCK:
2469			break;
2470		default:
2471			goto illegal;
2472		}
2473		break;
2474
2475	case SDEV_BLOCK:
2476		switch (oldstate) {
2477		case SDEV_RUNNING:
2478		case SDEV_CREATED_BLOCK:
2479		case SDEV_QUIESCE:
2480		case SDEV_OFFLINE:
2481			break;
2482		default:
2483			goto illegal;
2484		}
2485		break;
2486
2487	case SDEV_CREATED_BLOCK:
2488		switch (oldstate) {
2489		case SDEV_CREATED:
2490			break;
2491		default:
2492			goto illegal;
2493		}
2494		break;
2495
2496	case SDEV_CANCEL:
2497		switch (oldstate) {
2498		case SDEV_CREATED:
2499		case SDEV_RUNNING:
2500		case SDEV_QUIESCE:
2501		case SDEV_OFFLINE:
2502		case SDEV_TRANSPORT_OFFLINE:
2503			break;
2504		default:
2505			goto illegal;
2506		}
2507		break;
2508
2509	case SDEV_DEL:
2510		switch (oldstate) {
2511		case SDEV_CREATED:
2512		case SDEV_RUNNING:
2513		case SDEV_OFFLINE:
2514		case SDEV_TRANSPORT_OFFLINE:
2515		case SDEV_CANCEL:
2516		case SDEV_BLOCK:
2517		case SDEV_CREATED_BLOCK:
2518			break;
2519		default:
2520			goto illegal;
2521		}
2522		break;
2523
2524	}
2525	sdev->offline_already = false;
2526	sdev->sdev_state = state;
2527	return 0;
2528
2529 illegal:
2530	SCSI_LOG_ERROR_RECOVERY(1,
2531				sdev_printk(KERN_ERR, sdev,
2532					    "Illegal state transition %s->%s",
2533					    scsi_device_state_name(oldstate),
2534					    scsi_device_state_name(state))
2535				);
2536	return -EINVAL;
2537}
2538EXPORT_SYMBOL(scsi_device_set_state);
2539
2540/**
2541 *	scsi_evt_emit - emit a single SCSI device uevent
2542 *	@sdev: associated SCSI device
2543 *	@evt: event to emit
2544 *
2545 *	Send a single uevent (scsi_event) to the associated scsi_device.
2546 */
2547static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt)
2548{
2549	int idx = 0;
2550	char *envp[3];
2551
2552	switch (evt->evt_type) {
2553	case SDEV_EVT_MEDIA_CHANGE:
2554		envp[idx++] = "SDEV_MEDIA_CHANGE=1";
2555		break;
2556	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2557		scsi_rescan_device(sdev);
2558		envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED";
2559		break;
2560	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2561		envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED";
2562		break;
2563	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2564	       envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED";
2565		break;
2566	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2567		envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED";
2568		break;
2569	case SDEV_EVT_LUN_CHANGE_REPORTED:
2570		envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED";
2571		break;
2572	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2573		envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED";
2574		break;
2575	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2576		envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED";
2577		break;
2578	default:
2579		/* do nothing */
2580		break;
2581	}
2582
2583	envp[idx++] = NULL;
2584
2585	kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp);
2586}
2587
2588/**
2589 *	scsi_evt_thread - send a uevent for each scsi event
2590 *	@work: work struct for scsi_device
2591 *
2592 *	Dispatch queued events to their associated scsi_device kobjects
2593 *	as uevents.
2594 */
2595void scsi_evt_thread(struct work_struct *work)
2596{
2597	struct scsi_device *sdev;
2598	enum scsi_device_event evt_type;
2599	LIST_HEAD(event_list);
2600
2601	sdev = container_of(work, struct scsi_device, event_work);
2602
2603	for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++)
2604		if (test_and_clear_bit(evt_type, sdev->pending_events))
2605			sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL);
2606
2607	while (1) {
2608		struct scsi_event *evt;
2609		struct list_head *this, *tmp;
2610		unsigned long flags;
2611
2612		spin_lock_irqsave(&sdev->list_lock, flags);
2613		list_splice_init(&sdev->event_list, &event_list);
2614		spin_unlock_irqrestore(&sdev->list_lock, flags);
2615
2616		if (list_empty(&event_list))
2617			break;
2618
2619		list_for_each_safe(this, tmp, &event_list) {
2620			evt = list_entry(this, struct scsi_event, node);
2621			list_del(&evt->node);
2622			scsi_evt_emit(sdev, evt);
2623			kfree(evt);
2624		}
2625	}
2626}
2627
2628/**
2629 * 	sdev_evt_send - send asserted event to uevent thread
2630 *	@sdev: scsi_device event occurred on
2631 *	@evt: event to send
2632 *
2633 *	Assert scsi device event asynchronously.
2634 */
2635void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt)
2636{
2637	unsigned long flags;
2638
2639#if 0
2640	/* FIXME: currently this check eliminates all media change events
2641	 * for polled devices.  Need to update to discriminate between AN
2642	 * and polled events */
2643	if (!test_bit(evt->evt_type, sdev->supported_events)) {
2644		kfree(evt);
2645		return;
2646	}
2647#endif
2648
2649	spin_lock_irqsave(&sdev->list_lock, flags);
2650	list_add_tail(&evt->node, &sdev->event_list);
2651	schedule_work(&sdev->event_work);
2652	spin_unlock_irqrestore(&sdev->list_lock, flags);
2653}
2654EXPORT_SYMBOL_GPL(sdev_evt_send);
2655
2656/**
2657 * 	sdev_evt_alloc - allocate a new scsi event
2658 *	@evt_type: type of event to allocate
2659 *	@gfpflags: GFP flags for allocation
2660 *
2661 *	Allocates and returns a new scsi_event.
2662 */
2663struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type,
2664				  gfp_t gfpflags)
2665{
2666	struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags);
2667	if (!evt)
2668		return NULL;
2669
2670	evt->evt_type = evt_type;
2671	INIT_LIST_HEAD(&evt->node);
2672
2673	/* evt_type-specific initialization, if any */
2674	switch (evt_type) {
2675	case SDEV_EVT_MEDIA_CHANGE:
2676	case SDEV_EVT_INQUIRY_CHANGE_REPORTED:
2677	case SDEV_EVT_CAPACITY_CHANGE_REPORTED:
2678	case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED:
2679	case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED:
2680	case SDEV_EVT_LUN_CHANGE_REPORTED:
2681	case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED:
2682	case SDEV_EVT_POWER_ON_RESET_OCCURRED:
2683	default:
2684		/* do nothing */
2685		break;
2686	}
2687
2688	return evt;
2689}
2690EXPORT_SYMBOL_GPL(sdev_evt_alloc);
2691
2692/**
2693 * 	sdev_evt_send_simple - send asserted event to uevent thread
2694 *	@sdev: scsi_device event occurred on
2695 *	@evt_type: type of event to send
2696 *	@gfpflags: GFP flags for allocation
2697 *
2698 *	Assert scsi device event asynchronously, given an event type.
2699 */
2700void sdev_evt_send_simple(struct scsi_device *sdev,
2701			  enum scsi_device_event evt_type, gfp_t gfpflags)
2702{
2703	struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags);
2704	if (!evt) {
2705		sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n",
2706			    evt_type);
2707		return;
2708	}
2709
2710	sdev_evt_send(sdev, evt);
2711}
2712EXPORT_SYMBOL_GPL(sdev_evt_send_simple);
2713
2714/**
2715 *	scsi_device_quiesce - Block all commands except power management.
2716 *	@sdev:	scsi device to quiesce.
2717 *
2718 *	This works by trying to transition to the SDEV_QUIESCE state
2719 *	(which must be a legal transition).  When the device is in this
2720 *	state, only power management requests will be accepted, all others will
2721 *	be deferred.
 
 
2722 *
2723 *	Must be called with user context, may sleep.
2724 *
2725 *	Returns zero if unsuccessful or an error if not.
2726 */
2727int
2728scsi_device_quiesce(struct scsi_device *sdev)
2729{
2730	struct request_queue *q = sdev->request_queue;
2731	int err;
2732
2733	/*
2734	 * It is allowed to call scsi_device_quiesce() multiple times from
2735	 * the same context but concurrent scsi_device_quiesce() calls are
2736	 * not allowed.
2737	 */
2738	WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current);
2739
2740	if (sdev->quiesced_by == current)
2741		return 0;
2742
2743	blk_set_pm_only(q);
2744
2745	blk_mq_freeze_queue(q);
2746	/*
2747	 * Ensure that the effect of blk_set_pm_only() will be visible
2748	 * for percpu_ref_tryget() callers that occur after the queue
2749	 * unfreeze even if the queue was already frozen before this function
2750	 * was called. See also https://lwn.net/Articles/573497/.
2751	 */
2752	synchronize_rcu();
2753	blk_mq_unfreeze_queue(q);
2754
2755	mutex_lock(&sdev->state_mutex);
2756	err = scsi_device_set_state(sdev, SDEV_QUIESCE);
2757	if (err == 0)
2758		sdev->quiesced_by = current;
2759	else
2760		blk_clear_pm_only(q);
2761	mutex_unlock(&sdev->state_mutex);
2762
2763	return err;
2764}
2765EXPORT_SYMBOL(scsi_device_quiesce);
2766
2767/**
2768 *	scsi_device_resume - Restart user issued commands to a quiesced device.
2769 *	@sdev:	scsi device to resume.
2770 *
2771 *	Moves the device from quiesced back to running and restarts the
2772 *	queues.
2773 *
2774 *	Must be called with user context, may sleep.
2775 */
2776void scsi_device_resume(struct scsi_device *sdev)
2777{
2778	/* check if the device state was mutated prior to resume, and if
2779	 * so assume the state is being managed elsewhere (for example
2780	 * device deleted during suspend)
2781	 */
2782	mutex_lock(&sdev->state_mutex);
2783	if (sdev->sdev_state == SDEV_QUIESCE)
2784		scsi_device_set_state(sdev, SDEV_RUNNING);
2785	if (sdev->quiesced_by) {
2786		sdev->quiesced_by = NULL;
2787		blk_clear_pm_only(sdev->request_queue);
2788	}
 
 
2789	mutex_unlock(&sdev->state_mutex);
2790}
2791EXPORT_SYMBOL(scsi_device_resume);
2792
2793static void
2794device_quiesce_fn(struct scsi_device *sdev, void *data)
2795{
2796	scsi_device_quiesce(sdev);
2797}
2798
2799void
2800scsi_target_quiesce(struct scsi_target *starget)
2801{
2802	starget_for_each_device(starget, NULL, device_quiesce_fn);
2803}
2804EXPORT_SYMBOL(scsi_target_quiesce);
2805
2806static void
2807device_resume_fn(struct scsi_device *sdev, void *data)
2808{
2809	scsi_device_resume(sdev);
2810}
2811
2812void
2813scsi_target_resume(struct scsi_target *starget)
2814{
2815	starget_for_each_device(starget, NULL, device_resume_fn);
2816}
2817EXPORT_SYMBOL(scsi_target_resume);
2818
2819static int __scsi_internal_device_block_nowait(struct scsi_device *sdev)
2820{
2821	if (scsi_device_set_state(sdev, SDEV_BLOCK))
2822		return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK);
2823
2824	return 0;
2825}
2826
2827void scsi_start_queue(struct scsi_device *sdev)
2828{
2829	if (cmpxchg(&sdev->queue_stopped, 1, 0))
2830		blk_mq_unquiesce_queue(sdev->request_queue);
2831}
2832
2833static void scsi_stop_queue(struct scsi_device *sdev)
2834{
2835	/*
2836	 * The atomic variable of ->queue_stopped covers that
2837	 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue.
2838	 *
2839	 * The caller needs to wait until quiesce is done.
2840	 */
2841	if (!cmpxchg(&sdev->queue_stopped, 0, 1))
2842		blk_mq_quiesce_queue_nowait(sdev->request_queue);
2843}
2844
2845/**
2846 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state
2847 * @sdev: device to block
2848 *
2849 * Pause SCSI command processing on the specified device. Does not sleep.
2850 *
2851 * Returns zero if successful or a negative error code upon failure.
2852 *
2853 * Notes:
2854 * This routine transitions the device to the SDEV_BLOCK state (which must be
2855 * a legal transition). When the device is in this state, command processing
2856 * is paused until the device leaves the SDEV_BLOCK state. See also
2857 * scsi_internal_device_unblock_nowait().
2858 */
2859int scsi_internal_device_block_nowait(struct scsi_device *sdev)
2860{
2861	int ret = __scsi_internal_device_block_nowait(sdev);
 
 
 
 
 
 
 
 
 
2862
2863	/*
2864	 * The device has transitioned to SDEV_BLOCK.  Stop the
2865	 * block layer from calling the midlayer with this device's
2866	 * request queue.
2867	 */
2868	if (!ret)
2869		scsi_stop_queue(sdev);
2870	return ret;
2871}
2872EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait);
2873
2874/**
2875 * scsi_device_block - try to transition to the SDEV_BLOCK state
2876 * @sdev: device to block
2877 * @data: dummy argument, ignored
2878 *
2879 * Pause SCSI command processing on the specified device. Callers must wait
2880 * until all ongoing scsi_queue_rq() calls have finished after this function
2881 * returns.
 
2882 *
2883 * Note:
2884 * This routine transitions the device to the SDEV_BLOCK state (which must be
2885 * a legal transition). When the device is in this state, command processing
2886 * is paused until the device leaves the SDEV_BLOCK state. See also
2887 * scsi_internal_device_unblock().
2888 */
2889static void scsi_device_block(struct scsi_device *sdev, void *data)
2890{
 
2891	int err;
2892	enum scsi_device_state state;
2893
2894	mutex_lock(&sdev->state_mutex);
2895	err = __scsi_internal_device_block_nowait(sdev);
2896	state = sdev->sdev_state;
2897	if (err == 0)
2898		/*
2899		 * scsi_stop_queue() must be called with the state_mutex
2900		 * held. Otherwise a simultaneous scsi_start_queue() call
2901		 * might unquiesce the queue before we quiesce it.
2902		 */
2903		scsi_stop_queue(sdev);
2904
2905	mutex_unlock(&sdev->state_mutex);
 
 
2906
2907	WARN_ONCE(err, "%s: failed to block %s in state %d\n",
2908		  __func__, dev_name(&sdev->sdev_gendev), state);
2909}
2910
2911/**
2912 * scsi_internal_device_unblock_nowait - resume a device after a block request
2913 * @sdev:	device to resume
2914 * @new_state:	state to set the device to after unblocking
2915 *
2916 * Restart the device queue for a previously suspended SCSI device. Does not
2917 * sleep.
2918 *
2919 * Returns zero if successful or a negative error code upon failure.
2920 *
2921 * Notes:
2922 * This routine transitions the device to the SDEV_RUNNING state or to one of
2923 * the offline states (which must be a legal transition) allowing the midlayer
2924 * to goose the queue for this device.
2925 */
2926int scsi_internal_device_unblock_nowait(struct scsi_device *sdev,
2927					enum scsi_device_state new_state)
2928{
2929	switch (new_state) {
2930	case SDEV_RUNNING:
2931	case SDEV_TRANSPORT_OFFLINE:
2932		break;
2933	default:
2934		return -EINVAL;
2935	}
2936
2937	/*
2938	 * Try to transition the scsi device to SDEV_RUNNING or one of the
2939	 * offlined states and goose the device queue if successful.
2940	 */
2941	switch (sdev->sdev_state) {
2942	case SDEV_BLOCK:
2943	case SDEV_TRANSPORT_OFFLINE:
2944		sdev->sdev_state = new_state;
2945		break;
2946	case SDEV_CREATED_BLOCK:
2947		if (new_state == SDEV_TRANSPORT_OFFLINE ||
2948		    new_state == SDEV_OFFLINE)
2949			sdev->sdev_state = new_state;
2950		else
2951			sdev->sdev_state = SDEV_CREATED;
2952		break;
2953	case SDEV_CANCEL:
2954	case SDEV_OFFLINE:
2955		break;
2956	default:
2957		return -EINVAL;
2958	}
2959	scsi_start_queue(sdev);
2960
2961	return 0;
2962}
2963EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait);
2964
2965/**
2966 * scsi_internal_device_unblock - resume a device after a block request
2967 * @sdev:	device to resume
2968 * @new_state:	state to set the device to after unblocking
2969 *
2970 * Restart the device queue for a previously suspended SCSI device. May sleep.
2971 *
2972 * Returns zero if successful or a negative error code upon failure.
2973 *
2974 * Notes:
2975 * This routine transitions the device to the SDEV_RUNNING state or to one of
2976 * the offline states (which must be a legal transition) allowing the midlayer
2977 * to goose the queue for this device.
2978 */
2979static int scsi_internal_device_unblock(struct scsi_device *sdev,
2980					enum scsi_device_state new_state)
2981{
2982	int ret;
2983
2984	mutex_lock(&sdev->state_mutex);
2985	ret = scsi_internal_device_unblock_nowait(sdev, new_state);
2986	mutex_unlock(&sdev->state_mutex);
2987
2988	return ret;
2989}
2990
 
 
 
 
 
 
 
 
 
 
 
2991static int
2992target_block(struct device *dev, void *data)
2993{
2994	if (scsi_is_target_device(dev))
2995		starget_for_each_device(to_scsi_target(dev), NULL,
2996					scsi_device_block);
2997	return 0;
2998}
2999
3000/**
3001 * scsi_block_targets - transition all SCSI child devices to SDEV_BLOCK state
3002 * @dev: a parent device of one or more scsi_target devices
3003 * @shost: the Scsi_Host to which this device belongs
3004 *
3005 * Iterate over all children of @dev, which should be scsi_target devices,
3006 * and switch all subordinate scsi devices to SDEV_BLOCK state. Wait for
3007 * ongoing scsi_queue_rq() calls to finish. May sleep.
3008 *
3009 * Note:
3010 * @dev must not itself be a scsi_target device.
3011 */
3012void
3013scsi_block_targets(struct Scsi_Host *shost, struct device *dev)
3014{
3015	WARN_ON_ONCE(scsi_is_target_device(dev));
3016	device_for_each_child(dev, NULL, target_block);
3017	blk_mq_wait_quiesce_done(&shost->tag_set);
 
 
3018}
3019EXPORT_SYMBOL_GPL(scsi_block_targets);
3020
3021static void
3022device_unblock(struct scsi_device *sdev, void *data)
3023{
3024	scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data);
3025}
3026
3027static int
3028target_unblock(struct device *dev, void *data)
3029{
3030	if (scsi_is_target_device(dev))
3031		starget_for_each_device(to_scsi_target(dev), data,
3032					device_unblock);
3033	return 0;
3034}
3035
3036void
3037scsi_target_unblock(struct device *dev, enum scsi_device_state new_state)
3038{
3039	if (scsi_is_target_device(dev))
3040		starget_for_each_device(to_scsi_target(dev), &new_state,
3041					device_unblock);
3042	else
3043		device_for_each_child(dev, &new_state, target_unblock);
3044}
3045EXPORT_SYMBOL_GPL(scsi_target_unblock);
3046
3047/**
3048 * scsi_host_block - Try to transition all logical units to the SDEV_BLOCK state
3049 * @shost: device to block
3050 *
3051 * Pause SCSI command processing for all logical units associated with the SCSI
3052 * host and wait until pending scsi_queue_rq() calls have finished.
3053 *
3054 * Returns zero if successful or a negative error code upon failure.
3055 */
3056int
3057scsi_host_block(struct Scsi_Host *shost)
3058{
3059	struct scsi_device *sdev;
3060	int ret;
3061
3062	/*
3063	 * Call scsi_internal_device_block_nowait so we can avoid
3064	 * calling synchronize_rcu() for each LUN.
3065	 */
3066	shost_for_each_device(sdev, shost) {
3067		mutex_lock(&sdev->state_mutex);
3068		ret = scsi_internal_device_block_nowait(sdev);
3069		mutex_unlock(&sdev->state_mutex);
3070		if (ret) {
3071			scsi_device_put(sdev);
3072			return ret;
3073		}
3074	}
3075
3076	/* Wait for ongoing scsi_queue_rq() calls to finish. */
3077	blk_mq_wait_quiesce_done(&shost->tag_set);
 
 
 
3078
3079	return 0;
 
 
 
3080}
3081EXPORT_SYMBOL_GPL(scsi_host_block);
3082
3083int
3084scsi_host_unblock(struct Scsi_Host *shost, int new_state)
3085{
3086	struct scsi_device *sdev;
3087	int ret = 0;
3088
3089	shost_for_each_device(sdev, shost) {
3090		ret = scsi_internal_device_unblock(sdev, new_state);
3091		if (ret) {
3092			scsi_device_put(sdev);
3093			break;
3094		}
3095	}
3096	return ret;
3097}
3098EXPORT_SYMBOL_GPL(scsi_host_unblock);
3099
3100/**
3101 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt
3102 * @sgl:	scatter-gather list
3103 * @sg_count:	number of segments in sg
3104 * @offset:	offset in bytes into sg, on return offset into the mapped area
3105 * @len:	bytes to map, on return number of bytes mapped
3106 *
3107 * Returns virtual address of the start of the mapped page
3108 */
3109void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count,
3110			  size_t *offset, size_t *len)
3111{
3112	int i;
3113	size_t sg_len = 0, len_complete = 0;
3114	struct scatterlist *sg;
3115	struct page *page;
3116
3117	WARN_ON(!irqs_disabled());
3118
3119	for_each_sg(sgl, sg, sg_count, i) {
3120		len_complete = sg_len; /* Complete sg-entries */
3121		sg_len += sg->length;
3122		if (sg_len > *offset)
3123			break;
3124	}
3125
3126	if (unlikely(i == sg_count)) {
3127		printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, "
3128			"elements %d\n",
3129		       __func__, sg_len, *offset, sg_count);
3130		WARN_ON(1);
3131		return NULL;
3132	}
3133
3134	/* Offset starting from the beginning of first page in this sg-entry */
3135	*offset = *offset - len_complete + sg->offset;
3136
3137	/* Assumption: contiguous pages can be accessed as "page + i" */
3138	page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT));
3139	*offset &= ~PAGE_MASK;
3140
3141	/* Bytes in this sg-entry from *offset to the end of the page */
3142	sg_len = PAGE_SIZE - *offset;
3143	if (*len > sg_len)
3144		*len = sg_len;
3145
3146	return kmap_atomic(page);
3147}
3148EXPORT_SYMBOL(scsi_kmap_atomic_sg);
3149
3150/**
3151 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg
3152 * @virt:	virtual address to be unmapped
3153 */
3154void scsi_kunmap_atomic_sg(void *virt)
3155{
3156	kunmap_atomic(virt);
3157}
3158EXPORT_SYMBOL(scsi_kunmap_atomic_sg);
3159
3160void sdev_disable_disk_events(struct scsi_device *sdev)
3161{
3162	atomic_inc(&sdev->disk_events_disable_depth);
3163}
3164EXPORT_SYMBOL(sdev_disable_disk_events);
3165
3166void sdev_enable_disk_events(struct scsi_device *sdev)
3167{
3168	if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0))
3169		return;
3170	atomic_dec(&sdev->disk_events_disable_depth);
3171}
3172EXPORT_SYMBOL(sdev_enable_disk_events);
3173
3174static unsigned char designator_prio(const unsigned char *d)
3175{
3176	if (d[1] & 0x30)
3177		/* not associated with LUN */
3178		return 0;
3179
3180	if (d[3] == 0)
3181		/* invalid length */
3182		return 0;
3183
3184	/*
3185	 * Order of preference for lun descriptor:
3186	 * - SCSI name string
3187	 * - NAA IEEE Registered Extended
3188	 * - EUI-64 based 16-byte
3189	 * - EUI-64 based 12-byte
3190	 * - NAA IEEE Registered
3191	 * - NAA IEEE Extended
3192	 * - EUI-64 based 8-byte
3193	 * - SCSI name string (truncated)
3194	 * - T10 Vendor ID
3195	 * as longer descriptors reduce the likelyhood
3196	 * of identification clashes.
3197	 */
3198
3199	switch (d[1] & 0xf) {
3200	case 8:
3201		/* SCSI name string, variable-length UTF-8 */
3202		return 9;
3203	case 3:
3204		switch (d[4] >> 4) {
3205		case 6:
3206			/* NAA registered extended */
3207			return 8;
3208		case 5:
3209			/* NAA registered */
3210			return 5;
3211		case 4:
3212			/* NAA extended */
3213			return 4;
3214		case 3:
3215			/* NAA locally assigned */
3216			return 1;
3217		default:
3218			break;
3219		}
3220		break;
3221	case 2:
3222		switch (d[3]) {
3223		case 16:
3224			/* EUI64-based, 16 byte */
3225			return 7;
3226		case 12:
3227			/* EUI64-based, 12 byte */
3228			return 6;
3229		case 8:
3230			/* EUI64-based, 8 byte */
3231			return 3;
3232		default:
3233			break;
3234		}
3235		break;
3236	case 1:
3237		/* T10 vendor ID */
3238		return 1;
3239	default:
3240		break;
3241	}
3242
3243	return 0;
3244}
3245
3246/**
3247 * scsi_vpd_lun_id - return a unique device identification
3248 * @sdev: SCSI device
3249 * @id:   buffer for the identification
3250 * @id_len:  length of the buffer
3251 *
3252 * Copies a unique device identification into @id based
3253 * on the information in the VPD page 0x83 of the device.
3254 * The string will be formatted as a SCSI name string.
3255 *
3256 * Returns the length of the identification or error on failure.
3257 * If the identifier is longer than the supplied buffer the actual
3258 * identifier length is returned and the buffer is not zero-padded.
3259 */
3260int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len)
3261{
3262	u8 cur_id_prio = 0;
3263	u8 cur_id_size = 0;
3264	const unsigned char *d, *cur_id_str;
3265	const struct scsi_vpd *vpd_pg83;
3266	int id_size = -EINVAL;
3267
3268	rcu_read_lock();
3269	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3270	if (!vpd_pg83) {
3271		rcu_read_unlock();
3272		return -ENXIO;
3273	}
3274
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3275	/* The id string must be at least 20 bytes + terminating NULL byte */
3276	if (id_len < 21) {
3277		rcu_read_unlock();
3278		return -EINVAL;
3279	}
3280
3281	memset(id, 0, id_len);
3282	for (d = vpd_pg83->data + 4;
3283	     d < vpd_pg83->data + vpd_pg83->len;
3284	     d += d[3] + 4) {
3285		u8 prio = designator_prio(d);
3286
3287		if (prio == 0 || cur_id_prio > prio)
3288			continue;
3289
3290		switch (d[1] & 0xf) {
3291		case 0x1:
3292			/* T10 Vendor ID */
3293			if (cur_id_size > d[3])
3294				break;
3295			cur_id_prio = prio;
 
 
3296			cur_id_size = d[3];
3297			if (cur_id_size + 4 > id_len)
3298				cur_id_size = id_len - 4;
3299			cur_id_str = d + 4;
 
3300			id_size = snprintf(id, id_len, "t10.%*pE",
3301					   cur_id_size, cur_id_str);
3302			break;
3303		case 0x2:
3304			/* EUI-64 */
3305			cur_id_prio = prio;
 
 
 
 
 
3306			cur_id_size = d[3];
3307			cur_id_str = d + 4;
 
3308			switch (cur_id_size) {
3309			case 8:
3310				id_size = snprintf(id, id_len,
3311						   "eui.%8phN",
3312						   cur_id_str);
3313				break;
3314			case 12:
3315				id_size = snprintf(id, id_len,
3316						   "eui.%12phN",
3317						   cur_id_str);
3318				break;
3319			case 16:
3320				id_size = snprintf(id, id_len,
3321						   "eui.%16phN",
3322						   cur_id_str);
3323				break;
3324			default:
 
3325				break;
3326			}
3327			break;
3328		case 0x3:
3329			/* NAA */
3330			cur_id_prio = prio;
 
3331			cur_id_size = d[3];
3332			cur_id_str = d + 4;
 
3333			switch (cur_id_size) {
3334			case 8:
3335				id_size = snprintf(id, id_len,
3336						   "naa.%8phN",
3337						   cur_id_str);
3338				break;
3339			case 16:
3340				id_size = snprintf(id, id_len,
3341						   "naa.%16phN",
3342						   cur_id_str);
3343				break;
3344			default:
 
3345				break;
3346			}
3347			break;
3348		case 0x8:
3349			/* SCSI name string */
3350			if (cur_id_size > d[3])
3351				break;
3352			/* Prefer others for truncated descriptor */
3353			if (d[3] > id_len) {
3354				prio = 2;
3355				if (cur_id_prio > prio)
3356					break;
3357			}
3358			cur_id_prio = prio;
3359			cur_id_size = id_size = d[3];
3360			cur_id_str = d + 4;
 
3361			if (cur_id_size >= id_len)
3362				cur_id_size = id_len - 1;
3363			memcpy(id, cur_id_str, cur_id_size);
 
 
 
3364			break;
3365		default:
3366			break;
3367		}
 
 
3368	}
3369	rcu_read_unlock();
3370
3371	return id_size;
3372}
3373EXPORT_SYMBOL(scsi_vpd_lun_id);
3374
3375/*
3376 * scsi_vpd_tpg_id - return a target port group identifier
3377 * @sdev: SCSI device
3378 *
3379 * Returns the Target Port Group identifier from the information
3380 * froom VPD page 0x83 of the device.
3381 *
3382 * Returns the identifier or error on failure.
3383 */
3384int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id)
3385{
3386	const unsigned char *d;
3387	const struct scsi_vpd *vpd_pg83;
3388	int group_id = -EAGAIN, rel_port = -1;
3389
3390	rcu_read_lock();
3391	vpd_pg83 = rcu_dereference(sdev->vpd_pg83);
3392	if (!vpd_pg83) {
3393		rcu_read_unlock();
3394		return -ENXIO;
3395	}
3396
3397	d = vpd_pg83->data + 4;
3398	while (d < vpd_pg83->data + vpd_pg83->len) {
3399		switch (d[1] & 0xf) {
3400		case 0x4:
3401			/* Relative target port */
3402			rel_port = get_unaligned_be16(&d[6]);
3403			break;
3404		case 0x5:
3405			/* Target port group */
3406			group_id = get_unaligned_be16(&d[6]);
3407			break;
3408		default:
3409			break;
3410		}
3411		d += d[3] + 4;
3412	}
3413	rcu_read_unlock();
3414
3415	if (group_id >= 0 && rel_id && rel_port != -1)
3416		*rel_id = rel_port;
3417
3418	return group_id;
3419}
3420EXPORT_SYMBOL(scsi_vpd_tpg_id);
3421
3422/**
3423 * scsi_build_sense - build sense data for a command
3424 * @scmd:	scsi command for which the sense should be formatted
3425 * @desc:	Sense format (non-zero == descriptor format,
3426 *              0 == fixed format)
3427 * @key:	Sense key
3428 * @asc:	Additional sense code
3429 * @ascq:	Additional sense code qualifier
3430 *
3431 **/
3432void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq)
3433{
3434	scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq);
3435	scmd->result = SAM_STAT_CHECK_CONDITION;
3436}
3437EXPORT_SYMBOL_GPL(scsi_build_sense);
3438
3439#ifdef CONFIG_SCSI_LIB_KUNIT_TEST
3440#include "scsi_lib_test.c"
3441#endif