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