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