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 = scsi_alloc_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_mq_free_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 static void scsi_initialize_rq(struct request *rq) 1084 { 1085 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1086 struct scsi_request *req = &cmd->req; 1087 1088 memset(req->__cmd, 0, sizeof(req->__cmd)); 1089 req->cmd = req->__cmd; 1090 req->cmd_len = BLK_MAX_CDB; 1091 req->sense_len = 0; 1092 1093 init_rcu_head(&cmd->rcu); 1094 cmd->jiffies_at_alloc = jiffies; 1095 cmd->retries = 0; 1096 } 1097 1098 struct request *scsi_alloc_request(struct request_queue *q, 1099 unsigned int op, blk_mq_req_flags_t flags) 1100 { 1101 struct request *rq; 1102 1103 rq = blk_mq_alloc_request(q, op, flags); 1104 if (!IS_ERR(rq)) 1105 scsi_initialize_rq(rq); 1106 return rq; 1107 } 1108 EXPORT_SYMBOL_GPL(scsi_alloc_request); 1109 1110 /* 1111 * Only called when the request isn't completed by SCSI, and not freed by 1112 * SCSI 1113 */ 1114 static void scsi_cleanup_rq(struct request *rq) 1115 { 1116 if (rq->rq_flags & RQF_DONTPREP) { 1117 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq)); 1118 rq->rq_flags &= ~RQF_DONTPREP; 1119 } 1120 } 1121 1122 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */ 1123 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd) 1124 { 1125 void *buf = cmd->sense_buffer; 1126 void *prot = cmd->prot_sdb; 1127 struct request *rq = scsi_cmd_to_rq(cmd); 1128 unsigned int flags = cmd->flags & SCMD_PRESERVED_FLAGS; 1129 unsigned long jiffies_at_alloc; 1130 int retries, to_clear; 1131 bool in_flight; 1132 int budget_token = cmd->budget_token; 1133 1134 if (!blk_rq_is_passthrough(rq) && !(flags & SCMD_INITIALIZED)) { 1135 flags |= SCMD_INITIALIZED; 1136 scsi_initialize_rq(rq); 1137 } 1138 1139 jiffies_at_alloc = cmd->jiffies_at_alloc; 1140 retries = cmd->retries; 1141 in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state); 1142 /* 1143 * Zero out the cmd, except for the embedded scsi_request. Only clear 1144 * the driver-private command data if the LLD does not supply a 1145 * function to initialize that data. 1146 */ 1147 to_clear = sizeof(*cmd) - sizeof(cmd->req); 1148 if (!dev->host->hostt->init_cmd_priv) 1149 to_clear += dev->host->hostt->cmd_size; 1150 memset((char *)cmd + sizeof(cmd->req), 0, to_clear); 1151 1152 cmd->device = dev; 1153 cmd->sense_buffer = buf; 1154 cmd->prot_sdb = prot; 1155 cmd->flags = flags; 1156 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler); 1157 cmd->jiffies_at_alloc = jiffies_at_alloc; 1158 cmd->retries = retries; 1159 if (in_flight) 1160 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state); 1161 cmd->budget_token = budget_token; 1162 1163 } 1164 1165 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev, 1166 struct request *req) 1167 { 1168 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1169 1170 /* 1171 * Passthrough requests may transfer data, in which case they must 1172 * a bio attached to them. Or they might contain a SCSI command 1173 * that does not transfer data, in which case they may optionally 1174 * submit a request without an attached bio. 1175 */ 1176 if (req->bio) { 1177 blk_status_t ret = scsi_alloc_sgtables(cmd); 1178 if (unlikely(ret != BLK_STS_OK)) 1179 return ret; 1180 } else { 1181 BUG_ON(blk_rq_bytes(req)); 1182 1183 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1184 } 1185 1186 cmd->cmd_len = scsi_req(req)->cmd_len; 1187 if (cmd->cmd_len == 0) 1188 cmd->cmd_len = scsi_command_size(cmd->cmnd); 1189 cmd->cmnd = scsi_req(req)->cmd; 1190 cmd->transfersize = blk_rq_bytes(req); 1191 cmd->allowed = scsi_req(req)->retries; 1192 return BLK_STS_OK; 1193 } 1194 1195 static blk_status_t 1196 scsi_device_state_check(struct scsi_device *sdev, struct request *req) 1197 { 1198 switch (sdev->sdev_state) { 1199 case SDEV_CREATED: 1200 return BLK_STS_OK; 1201 case SDEV_OFFLINE: 1202 case SDEV_TRANSPORT_OFFLINE: 1203 /* 1204 * If the device is offline we refuse to process any 1205 * commands. The device must be brought online 1206 * before trying any recovery commands. 1207 */ 1208 if (!sdev->offline_already) { 1209 sdev->offline_already = true; 1210 sdev_printk(KERN_ERR, sdev, 1211 "rejecting I/O to offline device\n"); 1212 } 1213 return BLK_STS_IOERR; 1214 case SDEV_DEL: 1215 /* 1216 * If the device is fully deleted, we refuse to 1217 * process any commands as well. 1218 */ 1219 sdev_printk(KERN_ERR, sdev, 1220 "rejecting I/O to dead device\n"); 1221 return BLK_STS_IOERR; 1222 case SDEV_BLOCK: 1223 case SDEV_CREATED_BLOCK: 1224 return BLK_STS_RESOURCE; 1225 case SDEV_QUIESCE: 1226 /* 1227 * If the device is blocked we only accept power management 1228 * commands. 1229 */ 1230 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM))) 1231 return BLK_STS_RESOURCE; 1232 return BLK_STS_OK; 1233 default: 1234 /* 1235 * For any other not fully online state we only allow 1236 * power management commands. 1237 */ 1238 if (req && !(req->rq_flags & RQF_PM)) 1239 return BLK_STS_IOERR; 1240 return BLK_STS_OK; 1241 } 1242 } 1243 1244 /* 1245 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token 1246 * and return the token else return -1. 1247 */ 1248 static inline int scsi_dev_queue_ready(struct request_queue *q, 1249 struct scsi_device *sdev) 1250 { 1251 int token; 1252 1253 token = sbitmap_get(&sdev->budget_map); 1254 if (atomic_read(&sdev->device_blocked)) { 1255 if (token < 0) 1256 goto out; 1257 1258 if (scsi_device_busy(sdev) > 1) 1259 goto out_dec; 1260 1261 /* 1262 * unblock after device_blocked iterates to zero 1263 */ 1264 if (atomic_dec_return(&sdev->device_blocked) > 0) 1265 goto out_dec; 1266 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev, 1267 "unblocking device at zero depth\n")); 1268 } 1269 1270 return token; 1271 out_dec: 1272 if (token >= 0) 1273 sbitmap_put(&sdev->budget_map, token); 1274 out: 1275 return -1; 1276 } 1277 1278 /* 1279 * scsi_target_queue_ready: checks if there we can send commands to target 1280 * @sdev: scsi device on starget to check. 1281 */ 1282 static inline int scsi_target_queue_ready(struct Scsi_Host *shost, 1283 struct scsi_device *sdev) 1284 { 1285 struct scsi_target *starget = scsi_target(sdev); 1286 unsigned int busy; 1287 1288 if (starget->single_lun) { 1289 spin_lock_irq(shost->host_lock); 1290 if (starget->starget_sdev_user && 1291 starget->starget_sdev_user != sdev) { 1292 spin_unlock_irq(shost->host_lock); 1293 return 0; 1294 } 1295 starget->starget_sdev_user = sdev; 1296 spin_unlock_irq(shost->host_lock); 1297 } 1298 1299 if (starget->can_queue <= 0) 1300 return 1; 1301 1302 busy = atomic_inc_return(&starget->target_busy) - 1; 1303 if (atomic_read(&starget->target_blocked) > 0) { 1304 if (busy) 1305 goto starved; 1306 1307 /* 1308 * unblock after target_blocked iterates to zero 1309 */ 1310 if (atomic_dec_return(&starget->target_blocked) > 0) 1311 goto out_dec; 1312 1313 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget, 1314 "unblocking target at zero depth\n")); 1315 } 1316 1317 if (busy >= starget->can_queue) 1318 goto starved; 1319 1320 return 1; 1321 1322 starved: 1323 spin_lock_irq(shost->host_lock); 1324 list_move_tail(&sdev->starved_entry, &shost->starved_list); 1325 spin_unlock_irq(shost->host_lock); 1326 out_dec: 1327 if (starget->can_queue > 0) 1328 atomic_dec(&starget->target_busy); 1329 return 0; 1330 } 1331 1332 /* 1333 * scsi_host_queue_ready: if we can send requests to shost, return 1 else 1334 * return 0. We must end up running the queue again whenever 0 is 1335 * returned, else IO can hang. 1336 */ 1337 static inline int scsi_host_queue_ready(struct request_queue *q, 1338 struct Scsi_Host *shost, 1339 struct scsi_device *sdev, 1340 struct scsi_cmnd *cmd) 1341 { 1342 if (scsi_host_in_recovery(shost)) 1343 return 0; 1344 1345 if (atomic_read(&shost->host_blocked) > 0) { 1346 if (scsi_host_busy(shost) > 0) 1347 goto starved; 1348 1349 /* 1350 * unblock after host_blocked iterates to zero 1351 */ 1352 if (atomic_dec_return(&shost->host_blocked) > 0) 1353 goto out_dec; 1354 1355 SCSI_LOG_MLQUEUE(3, 1356 shost_printk(KERN_INFO, shost, 1357 "unblocking host at zero depth\n")); 1358 } 1359 1360 if (shost->host_self_blocked) 1361 goto starved; 1362 1363 /* We're OK to process the command, so we can't be starved */ 1364 if (!list_empty(&sdev->starved_entry)) { 1365 spin_lock_irq(shost->host_lock); 1366 if (!list_empty(&sdev->starved_entry)) 1367 list_del_init(&sdev->starved_entry); 1368 spin_unlock_irq(shost->host_lock); 1369 } 1370 1371 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state); 1372 1373 return 1; 1374 1375 starved: 1376 spin_lock_irq(shost->host_lock); 1377 if (list_empty(&sdev->starved_entry)) 1378 list_add_tail(&sdev->starved_entry, &shost->starved_list); 1379 spin_unlock_irq(shost->host_lock); 1380 out_dec: 1381 scsi_dec_host_busy(shost, cmd); 1382 return 0; 1383 } 1384 1385 /* 1386 * Busy state exporting function for request stacking drivers. 1387 * 1388 * For efficiency, no lock is taken to check the busy state of 1389 * shost/starget/sdev, since the returned value is not guaranteed and 1390 * may be changed after request stacking drivers call the function, 1391 * regardless of taking lock or not. 1392 * 1393 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi 1394 * needs to return 'not busy'. Otherwise, request stacking drivers 1395 * may hold requests forever. 1396 */ 1397 static bool scsi_mq_lld_busy(struct request_queue *q) 1398 { 1399 struct scsi_device *sdev = q->queuedata; 1400 struct Scsi_Host *shost; 1401 1402 if (blk_queue_dying(q)) 1403 return false; 1404 1405 shost = sdev->host; 1406 1407 /* 1408 * Ignore host/starget busy state. 1409 * Since block layer does not have a concept of fairness across 1410 * multiple queues, congestion of host/starget needs to be handled 1411 * in SCSI layer. 1412 */ 1413 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev)) 1414 return true; 1415 1416 return false; 1417 } 1418 1419 /* 1420 * Block layer request completion callback. May be called from interrupt 1421 * context. 1422 */ 1423 static void scsi_complete(struct request *rq) 1424 { 1425 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1426 enum scsi_disposition disposition; 1427 1428 INIT_LIST_HEAD(&cmd->eh_entry); 1429 1430 atomic_inc(&cmd->device->iodone_cnt); 1431 if (cmd->result) 1432 atomic_inc(&cmd->device->ioerr_cnt); 1433 1434 disposition = scsi_decide_disposition(cmd); 1435 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd)) 1436 disposition = SUCCESS; 1437 1438 scsi_log_completion(cmd, disposition); 1439 1440 switch (disposition) { 1441 case SUCCESS: 1442 scsi_finish_command(cmd); 1443 break; 1444 case NEEDS_RETRY: 1445 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY); 1446 break; 1447 case ADD_TO_MLQUEUE: 1448 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY); 1449 break; 1450 default: 1451 scsi_eh_scmd_add(cmd); 1452 break; 1453 } 1454 } 1455 1456 /** 1457 * scsi_dispatch_cmd - Dispatch a command to the low-level driver. 1458 * @cmd: command block we are dispatching. 1459 * 1460 * Return: nonzero return request was rejected and device's queue needs to be 1461 * plugged. 1462 */ 1463 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd) 1464 { 1465 struct Scsi_Host *host = cmd->device->host; 1466 int rtn = 0; 1467 1468 atomic_inc(&cmd->device->iorequest_cnt); 1469 1470 /* check if the device is still usable */ 1471 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) { 1472 /* in SDEV_DEL we error all commands. DID_NO_CONNECT 1473 * returns an immediate error upwards, and signals 1474 * that the device is no longer present */ 1475 cmd->result = DID_NO_CONNECT << 16; 1476 goto done; 1477 } 1478 1479 /* Check to see if the scsi lld made this device blocked. */ 1480 if (unlikely(scsi_device_blocked(cmd->device))) { 1481 /* 1482 * in blocked state, the command is just put back on 1483 * the device queue. The suspend state has already 1484 * blocked the queue so future requests should not 1485 * occur until the device transitions out of the 1486 * suspend state. 1487 */ 1488 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd, 1489 "queuecommand : device blocked\n")); 1490 return SCSI_MLQUEUE_DEVICE_BUSY; 1491 } 1492 1493 /* Store the LUN value in cmnd, if needed. */ 1494 if (cmd->device->lun_in_cdb) 1495 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) | 1496 (cmd->device->lun << 5 & 0xe0); 1497 1498 scsi_log_send(cmd); 1499 1500 /* 1501 * Before we queue this command, check if the command 1502 * length exceeds what the host adapter can handle. 1503 */ 1504 if (cmd->cmd_len > cmd->device->host->max_cmd_len) { 1505 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd, 1506 "queuecommand : command too long. " 1507 "cdb_size=%d host->max_cmd_len=%d\n", 1508 cmd->cmd_len, cmd->device->host->max_cmd_len)); 1509 cmd->result = (DID_ABORT << 16); 1510 goto done; 1511 } 1512 1513 if (unlikely(host->shost_state == SHOST_DEL)) { 1514 cmd->result = (DID_NO_CONNECT << 16); 1515 goto done; 1516 1517 } 1518 1519 trace_scsi_dispatch_cmd_start(cmd); 1520 rtn = host->hostt->queuecommand(host, cmd); 1521 if (rtn) { 1522 trace_scsi_dispatch_cmd_error(cmd, rtn); 1523 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY && 1524 rtn != SCSI_MLQUEUE_TARGET_BUSY) 1525 rtn = SCSI_MLQUEUE_HOST_BUSY; 1526 1527 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd, 1528 "queuecommand : request rejected\n")); 1529 } 1530 1531 return rtn; 1532 done: 1533 cmd->scsi_done(cmd); 1534 return 0; 1535 } 1536 1537 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */ 1538 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost) 1539 { 1540 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) * 1541 sizeof(struct scatterlist); 1542 } 1543 1544 static blk_status_t scsi_prepare_cmd(struct request *req) 1545 { 1546 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1547 struct scsi_device *sdev = req->q->queuedata; 1548 struct Scsi_Host *shost = sdev->host; 1549 struct scatterlist *sg; 1550 1551 scsi_init_command(sdev, cmd); 1552 1553 cmd->prot_op = SCSI_PROT_NORMAL; 1554 if (blk_rq_bytes(req)) 1555 cmd->sc_data_direction = rq_dma_dir(req); 1556 else 1557 cmd->sc_data_direction = DMA_NONE; 1558 1559 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size; 1560 cmd->sdb.table.sgl = sg; 1561 1562 if (scsi_host_get_prot(shost)) { 1563 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer)); 1564 1565 cmd->prot_sdb->table.sgl = 1566 (struct scatterlist *)(cmd->prot_sdb + 1); 1567 } 1568 1569 /* 1570 * Special handling for passthrough commands, which don't go to the ULP 1571 * at all: 1572 */ 1573 if (blk_rq_is_passthrough(req)) 1574 return scsi_setup_scsi_cmnd(sdev, req); 1575 1576 if (sdev->handler && sdev->handler->prep_fn) { 1577 blk_status_t ret = sdev->handler->prep_fn(sdev, req); 1578 1579 if (ret != BLK_STS_OK) 1580 return ret; 1581 } 1582 1583 cmd->cmnd = scsi_req(req)->cmd = scsi_req(req)->__cmd; 1584 memset(cmd->cmnd, 0, BLK_MAX_CDB); 1585 return scsi_cmd_to_driver(cmd)->init_command(cmd); 1586 } 1587 1588 static void scsi_mq_done(struct scsi_cmnd *cmd) 1589 { 1590 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q))) 1591 return; 1592 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state))) 1593 return; 1594 trace_scsi_dispatch_cmd_done(cmd); 1595 blk_mq_complete_request(scsi_cmd_to_rq(cmd)); 1596 } 1597 1598 static void scsi_mq_put_budget(struct request_queue *q, int budget_token) 1599 { 1600 struct scsi_device *sdev = q->queuedata; 1601 1602 sbitmap_put(&sdev->budget_map, budget_token); 1603 } 1604 1605 static int scsi_mq_get_budget(struct request_queue *q) 1606 { 1607 struct scsi_device *sdev = q->queuedata; 1608 int token = scsi_dev_queue_ready(q, sdev); 1609 1610 if (token >= 0) 1611 return token; 1612 1613 atomic_inc(&sdev->restarts); 1614 1615 /* 1616 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy). 1617 * .restarts must be incremented before .device_busy is read because the 1618 * code in scsi_run_queue_async() depends on the order of these operations. 1619 */ 1620 smp_mb__after_atomic(); 1621 1622 /* 1623 * If all in-flight requests originated from this LUN are completed 1624 * before reading .device_busy, sdev->device_busy will be observed as 1625 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request 1626 * soon. Otherwise, completion of one of these requests will observe 1627 * the .restarts flag, and the request queue will be run for handling 1628 * this request, see scsi_end_request(). 1629 */ 1630 if (unlikely(scsi_device_busy(sdev) == 0 && 1631 !scsi_device_blocked(sdev))) 1632 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY); 1633 return -1; 1634 } 1635 1636 static void scsi_mq_set_rq_budget_token(struct request *req, int token) 1637 { 1638 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1639 1640 cmd->budget_token = token; 1641 } 1642 1643 static int scsi_mq_get_rq_budget_token(struct request *req) 1644 { 1645 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1646 1647 return cmd->budget_token; 1648 } 1649 1650 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, 1651 const struct blk_mq_queue_data *bd) 1652 { 1653 struct request *req = bd->rq; 1654 struct request_queue *q = req->q; 1655 struct scsi_device *sdev = q->queuedata; 1656 struct Scsi_Host *shost = sdev->host; 1657 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1658 blk_status_t ret; 1659 int reason; 1660 1661 WARN_ON_ONCE(cmd->budget_token < 0); 1662 1663 /* 1664 * If the device is not in running state we will reject some or all 1665 * commands. 1666 */ 1667 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) { 1668 ret = scsi_device_state_check(sdev, req); 1669 if (ret != BLK_STS_OK) 1670 goto out_put_budget; 1671 } 1672 1673 ret = BLK_STS_RESOURCE; 1674 if (!scsi_target_queue_ready(shost, sdev)) 1675 goto out_put_budget; 1676 if (!scsi_host_queue_ready(q, shost, sdev, cmd)) 1677 goto out_dec_target_busy; 1678 1679 if (!(req->rq_flags & RQF_DONTPREP)) { 1680 ret = scsi_prepare_cmd(req); 1681 if (ret != BLK_STS_OK) 1682 goto out_dec_host_busy; 1683 req->rq_flags |= RQF_DONTPREP; 1684 } else { 1685 clear_bit(SCMD_STATE_COMPLETE, &cmd->state); 1686 } 1687 1688 cmd->flags &= SCMD_PRESERVED_FLAGS; 1689 if (sdev->simple_tags) 1690 cmd->flags |= SCMD_TAGGED; 1691 if (bd->last) 1692 cmd->flags |= SCMD_LAST; 1693 1694 scsi_set_resid(cmd, 0); 1695 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1696 cmd->scsi_done = scsi_mq_done; 1697 1698 blk_mq_start_request(req); 1699 reason = scsi_dispatch_cmd(cmd); 1700 if (reason) { 1701 scsi_set_blocked(cmd, reason); 1702 ret = BLK_STS_RESOURCE; 1703 goto out_dec_host_busy; 1704 } 1705 1706 return BLK_STS_OK; 1707 1708 out_dec_host_busy: 1709 scsi_dec_host_busy(shost, cmd); 1710 out_dec_target_busy: 1711 if (scsi_target(sdev)->can_queue > 0) 1712 atomic_dec(&scsi_target(sdev)->target_busy); 1713 out_put_budget: 1714 scsi_mq_put_budget(q, cmd->budget_token); 1715 cmd->budget_token = -1; 1716 switch (ret) { 1717 case BLK_STS_OK: 1718 break; 1719 case BLK_STS_RESOURCE: 1720 case BLK_STS_ZONE_RESOURCE: 1721 if (scsi_device_blocked(sdev)) 1722 ret = BLK_STS_DEV_RESOURCE; 1723 break; 1724 case BLK_STS_AGAIN: 1725 scsi_req(req)->result = DID_BUS_BUSY << 16; 1726 if (req->rq_flags & RQF_DONTPREP) 1727 scsi_mq_uninit_cmd(cmd); 1728 break; 1729 default: 1730 if (unlikely(!scsi_device_online(sdev))) 1731 scsi_req(req)->result = DID_NO_CONNECT << 16; 1732 else 1733 scsi_req(req)->result = DID_ERROR << 16; 1734 /* 1735 * Make sure to release all allocated resources when 1736 * we hit an error, as we will never see this command 1737 * again. 1738 */ 1739 if (req->rq_flags & RQF_DONTPREP) 1740 scsi_mq_uninit_cmd(cmd); 1741 scsi_run_queue_async(sdev); 1742 break; 1743 } 1744 return ret; 1745 } 1746 1747 static enum blk_eh_timer_return scsi_timeout(struct request *req, 1748 bool reserved) 1749 { 1750 if (reserved) 1751 return BLK_EH_RESET_TIMER; 1752 return scsi_times_out(req); 1753 } 1754 1755 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq, 1756 unsigned int hctx_idx, unsigned int numa_node) 1757 { 1758 struct Scsi_Host *shost = set->driver_data; 1759 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1760 struct scatterlist *sg; 1761 int ret = 0; 1762 1763 cmd->sense_buffer = 1764 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node); 1765 if (!cmd->sense_buffer) 1766 return -ENOMEM; 1767 cmd->req.sense = cmd->sense_buffer; 1768 1769 if (scsi_host_get_prot(shost)) { 1770 sg = (void *)cmd + sizeof(struct scsi_cmnd) + 1771 shost->hostt->cmd_size; 1772 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost); 1773 } 1774 1775 if (shost->hostt->init_cmd_priv) { 1776 ret = shost->hostt->init_cmd_priv(shost, cmd); 1777 if (ret < 0) 1778 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer); 1779 } 1780 1781 return ret; 1782 } 1783 1784 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq, 1785 unsigned int hctx_idx) 1786 { 1787 struct Scsi_Host *shost = set->driver_data; 1788 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1789 1790 if (shost->hostt->exit_cmd_priv) 1791 shost->hostt->exit_cmd_priv(shost, cmd); 1792 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer); 1793 } 1794 1795 1796 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob) 1797 { 1798 struct Scsi_Host *shost = hctx->driver_data; 1799 1800 if (shost->hostt->mq_poll) 1801 return shost->hostt->mq_poll(shost, hctx->queue_num); 1802 1803 return 0; 1804 } 1805 1806 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, 1807 unsigned int hctx_idx) 1808 { 1809 struct Scsi_Host *shost = data; 1810 1811 hctx->driver_data = shost; 1812 return 0; 1813 } 1814 1815 static int scsi_map_queues(struct blk_mq_tag_set *set) 1816 { 1817 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set); 1818 1819 if (shost->hostt->map_queues) 1820 return shost->hostt->map_queues(shost); 1821 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]); 1822 } 1823 1824 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q) 1825 { 1826 struct device *dev = shost->dma_dev; 1827 1828 /* 1829 * this limit is imposed by hardware restrictions 1830 */ 1831 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize, 1832 SG_MAX_SEGMENTS)); 1833 1834 if (scsi_host_prot_dma(shost)) { 1835 shost->sg_prot_tablesize = 1836 min_not_zero(shost->sg_prot_tablesize, 1837 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS); 1838 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize); 1839 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize); 1840 } 1841 1842 if (dev->dma_mask) { 1843 shost->max_sectors = min_t(unsigned int, shost->max_sectors, 1844 dma_max_mapping_size(dev) >> SECTOR_SHIFT); 1845 } 1846 blk_queue_max_hw_sectors(q, shost->max_sectors); 1847 blk_queue_segment_boundary(q, shost->dma_boundary); 1848 dma_set_seg_boundary(dev, shost->dma_boundary); 1849 1850 blk_queue_max_segment_size(q, shost->max_segment_size); 1851 blk_queue_virt_boundary(q, shost->virt_boundary_mask); 1852 dma_set_max_seg_size(dev, queue_max_segment_size(q)); 1853 1854 /* 1855 * Set a reasonable default alignment: The larger of 32-byte (dword), 1856 * which is a common minimum for HBAs, and the minimum DMA alignment, 1857 * which is set by the platform. 1858 * 1859 * Devices that require a bigger alignment can increase it later. 1860 */ 1861 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1); 1862 } 1863 EXPORT_SYMBOL_GPL(__scsi_init_queue); 1864 1865 static const struct blk_mq_ops scsi_mq_ops_no_commit = { 1866 .get_budget = scsi_mq_get_budget, 1867 .put_budget = scsi_mq_put_budget, 1868 .queue_rq = scsi_queue_rq, 1869 .complete = scsi_complete, 1870 .timeout = scsi_timeout, 1871 #ifdef CONFIG_BLK_DEBUG_FS 1872 .show_rq = scsi_show_rq, 1873 #endif 1874 .init_request = scsi_mq_init_request, 1875 .exit_request = scsi_mq_exit_request, 1876 .cleanup_rq = scsi_cleanup_rq, 1877 .busy = scsi_mq_lld_busy, 1878 .map_queues = scsi_map_queues, 1879 .init_hctx = scsi_init_hctx, 1880 .poll = scsi_mq_poll, 1881 .set_rq_budget_token = scsi_mq_set_rq_budget_token, 1882 .get_rq_budget_token = scsi_mq_get_rq_budget_token, 1883 }; 1884 1885 1886 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx) 1887 { 1888 struct Scsi_Host *shost = hctx->driver_data; 1889 1890 shost->hostt->commit_rqs(shost, hctx->queue_num); 1891 } 1892 1893 static const struct blk_mq_ops scsi_mq_ops = { 1894 .get_budget = scsi_mq_get_budget, 1895 .put_budget = scsi_mq_put_budget, 1896 .queue_rq = scsi_queue_rq, 1897 .commit_rqs = scsi_commit_rqs, 1898 .complete = scsi_complete, 1899 .timeout = scsi_timeout, 1900 #ifdef CONFIG_BLK_DEBUG_FS 1901 .show_rq = scsi_show_rq, 1902 #endif 1903 .init_request = scsi_mq_init_request, 1904 .exit_request = scsi_mq_exit_request, 1905 .cleanup_rq = scsi_cleanup_rq, 1906 .busy = scsi_mq_lld_busy, 1907 .map_queues = scsi_map_queues, 1908 .init_hctx = scsi_init_hctx, 1909 .poll = scsi_mq_poll, 1910 .set_rq_budget_token = scsi_mq_set_rq_budget_token, 1911 .get_rq_budget_token = scsi_mq_get_rq_budget_token, 1912 }; 1913 1914 int scsi_mq_setup_tags(struct Scsi_Host *shost) 1915 { 1916 unsigned int cmd_size, sgl_size; 1917 struct blk_mq_tag_set *tag_set = &shost->tag_set; 1918 1919 sgl_size = max_t(unsigned int, sizeof(struct scatterlist), 1920 scsi_mq_inline_sgl_size(shost)); 1921 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size; 1922 if (scsi_host_get_prot(shost)) 1923 cmd_size += sizeof(struct scsi_data_buffer) + 1924 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT; 1925 1926 memset(tag_set, 0, sizeof(*tag_set)); 1927 if (shost->hostt->commit_rqs) 1928 tag_set->ops = &scsi_mq_ops; 1929 else 1930 tag_set->ops = &scsi_mq_ops_no_commit; 1931 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1; 1932 tag_set->nr_maps = shost->nr_maps ? : 1; 1933 tag_set->queue_depth = shost->can_queue; 1934 tag_set->cmd_size = cmd_size; 1935 tag_set->numa_node = NUMA_NO_NODE; 1936 tag_set->flags = BLK_MQ_F_SHOULD_MERGE; 1937 tag_set->flags |= 1938 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy); 1939 tag_set->driver_data = shost; 1940 if (shost->host_tagset) 1941 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED; 1942 1943 return blk_mq_alloc_tag_set(tag_set); 1944 } 1945 1946 void scsi_mq_destroy_tags(struct Scsi_Host *shost) 1947 { 1948 blk_mq_free_tag_set(&shost->tag_set); 1949 } 1950 1951 /** 1952 * scsi_device_from_queue - return sdev associated with a request_queue 1953 * @q: The request queue to return the sdev from 1954 * 1955 * Return the sdev associated with a request queue or NULL if the 1956 * request_queue does not reference a SCSI device. 1957 */ 1958 struct scsi_device *scsi_device_from_queue(struct request_queue *q) 1959 { 1960 struct scsi_device *sdev = NULL; 1961 1962 if (q->mq_ops == &scsi_mq_ops_no_commit || 1963 q->mq_ops == &scsi_mq_ops) 1964 sdev = q->queuedata; 1965 if (!sdev || !get_device(&sdev->sdev_gendev)) 1966 sdev = NULL; 1967 1968 return sdev; 1969 } 1970 /* 1971 * pktcdvd should have been integrated into the SCSI layers, but for historical 1972 * reasons like the old IDE driver it isn't. This export allows it to safely 1973 * probe if a given device is a SCSI one and only attach to that. 1974 */ 1975 #ifdef CONFIG_CDROM_PKTCDVD_MODULE 1976 EXPORT_SYMBOL_GPL(scsi_device_from_queue); 1977 #endif 1978 1979 /** 1980 * scsi_block_requests - Utility function used by low-level drivers to prevent 1981 * further commands from being queued to the device. 1982 * @shost: host in question 1983 * 1984 * There is no timer nor any other means by which the requests get unblocked 1985 * other than the low-level driver calling scsi_unblock_requests(). 1986 */ 1987 void scsi_block_requests(struct Scsi_Host *shost) 1988 { 1989 shost->host_self_blocked = 1; 1990 } 1991 EXPORT_SYMBOL(scsi_block_requests); 1992 1993 /** 1994 * scsi_unblock_requests - Utility function used by low-level drivers to allow 1995 * further commands to be queued to the device. 1996 * @shost: host in question 1997 * 1998 * There is no timer nor any other means by which the requests get unblocked 1999 * other than the low-level driver calling scsi_unblock_requests(). This is done 2000 * as an API function so that changes to the internals of the scsi mid-layer 2001 * won't require wholesale changes to drivers that use this feature. 2002 */ 2003 void scsi_unblock_requests(struct Scsi_Host *shost) 2004 { 2005 shost->host_self_blocked = 0; 2006 scsi_run_host_queues(shost); 2007 } 2008 EXPORT_SYMBOL(scsi_unblock_requests); 2009 2010 void scsi_exit_queue(void) 2011 { 2012 kmem_cache_destroy(scsi_sense_cache); 2013 } 2014 2015 /** 2016 * scsi_mode_select - issue a mode select 2017 * @sdev: SCSI device to be queried 2018 * @pf: Page format bit (1 == standard, 0 == vendor specific) 2019 * @sp: Save page bit (0 == don't save, 1 == save) 2020 * @modepage: mode page being requested 2021 * @buffer: request buffer (may not be smaller than eight bytes) 2022 * @len: length of request buffer. 2023 * @timeout: command timeout 2024 * @retries: number of retries before failing 2025 * @data: returns a structure abstracting the mode header data 2026 * @sshdr: place to put sense data (or NULL if no sense to be collected). 2027 * must be SCSI_SENSE_BUFFERSIZE big. 2028 * 2029 * Returns zero if successful; negative error number or scsi 2030 * status on error 2031 * 2032 */ 2033 int 2034 scsi_mode_select(struct scsi_device *sdev, int pf, int sp, int modepage, 2035 unsigned char *buffer, int len, int timeout, int retries, 2036 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 2037 { 2038 unsigned char cmd[10]; 2039 unsigned char *real_buffer; 2040 int ret; 2041 2042 memset(cmd, 0, sizeof(cmd)); 2043 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0); 2044 2045 if (sdev->use_10_for_ms) { 2046 if (len > 65535) 2047 return -EINVAL; 2048 real_buffer = kmalloc(8 + len, GFP_KERNEL); 2049 if (!real_buffer) 2050 return -ENOMEM; 2051 memcpy(real_buffer + 8, buffer, len); 2052 len += 8; 2053 real_buffer[0] = 0; 2054 real_buffer[1] = 0; 2055 real_buffer[2] = data->medium_type; 2056 real_buffer[3] = data->device_specific; 2057 real_buffer[4] = data->longlba ? 0x01 : 0; 2058 real_buffer[5] = 0; 2059 real_buffer[6] = data->block_descriptor_length >> 8; 2060 real_buffer[7] = data->block_descriptor_length; 2061 2062 cmd[0] = MODE_SELECT_10; 2063 cmd[7] = len >> 8; 2064 cmd[8] = len; 2065 } else { 2066 if (len > 255 || data->block_descriptor_length > 255 || 2067 data->longlba) 2068 return -EINVAL; 2069 2070 real_buffer = kmalloc(4 + len, GFP_KERNEL); 2071 if (!real_buffer) 2072 return -ENOMEM; 2073 memcpy(real_buffer + 4, buffer, len); 2074 len += 4; 2075 real_buffer[0] = 0; 2076 real_buffer[1] = data->medium_type; 2077 real_buffer[2] = data->device_specific; 2078 real_buffer[3] = data->block_descriptor_length; 2079 2080 cmd[0] = MODE_SELECT; 2081 cmd[4] = len; 2082 } 2083 2084 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len, 2085 sshdr, timeout, retries, NULL); 2086 kfree(real_buffer); 2087 return ret; 2088 } 2089 EXPORT_SYMBOL_GPL(scsi_mode_select); 2090 2091 /** 2092 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary. 2093 * @sdev: SCSI device to be queried 2094 * @dbd: set if mode sense will allow block descriptors to be returned 2095 * @modepage: mode page being requested 2096 * @buffer: request buffer (may not be smaller than eight bytes) 2097 * @len: length of request buffer. 2098 * @timeout: command timeout 2099 * @retries: number of retries before failing 2100 * @data: returns a structure abstracting the mode header data 2101 * @sshdr: place to put sense data (or NULL if no sense to be collected). 2102 * must be SCSI_SENSE_BUFFERSIZE big. 2103 * 2104 * Returns zero if successful, or a negative error number on failure 2105 */ 2106 int 2107 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, 2108 unsigned char *buffer, int len, int timeout, int retries, 2109 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 2110 { 2111 unsigned char cmd[12]; 2112 int use_10_for_ms; 2113 int header_length; 2114 int result, retry_count = retries; 2115 struct scsi_sense_hdr my_sshdr; 2116 2117 memset(data, 0, sizeof(*data)); 2118 memset(&cmd[0], 0, 12); 2119 2120 dbd = sdev->set_dbd_for_ms ? 8 : dbd; 2121 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */ 2122 cmd[2] = modepage; 2123 2124 /* caller might not be interested in sense, but we need it */ 2125 if (!sshdr) 2126 sshdr = &my_sshdr; 2127 2128 retry: 2129 use_10_for_ms = sdev->use_10_for_ms; 2130 2131 if (use_10_for_ms) { 2132 if (len < 8) 2133 len = 8; 2134 2135 cmd[0] = MODE_SENSE_10; 2136 cmd[8] = len; 2137 header_length = 8; 2138 } else { 2139 if (len < 4) 2140 len = 4; 2141 2142 cmd[0] = MODE_SENSE; 2143 cmd[4] = len; 2144 header_length = 4; 2145 } 2146 2147 memset(buffer, 0, len); 2148 2149 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len, 2150 sshdr, timeout, retries, NULL); 2151 if (result < 0) 2152 return result; 2153 2154 /* This code looks awful: what it's doing is making sure an 2155 * ILLEGAL REQUEST sense return identifies the actual command 2156 * byte as the problem. MODE_SENSE commands can return 2157 * ILLEGAL REQUEST if the code page isn't supported */ 2158 2159 if (!scsi_status_is_good(result)) { 2160 if (scsi_sense_valid(sshdr)) { 2161 if ((sshdr->sense_key == ILLEGAL_REQUEST) && 2162 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) { 2163 /* 2164 * Invalid command operation code 2165 */ 2166 if (use_10_for_ms) { 2167 sdev->use_10_for_ms = 0; 2168 goto retry; 2169 } 2170 } 2171 if (scsi_status_is_check_condition(result) && 2172 sshdr->sense_key == UNIT_ATTENTION && 2173 retry_count) { 2174 retry_count--; 2175 goto retry; 2176 } 2177 } 2178 return -EIO; 2179 } 2180 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b && 2181 (modepage == 6 || modepage == 8))) { 2182 /* Initio breakage? */ 2183 header_length = 0; 2184 data->length = 13; 2185 data->medium_type = 0; 2186 data->device_specific = 0; 2187 data->longlba = 0; 2188 data->block_descriptor_length = 0; 2189 } else if (use_10_for_ms) { 2190 data->length = buffer[0]*256 + buffer[1] + 2; 2191 data->medium_type = buffer[2]; 2192 data->device_specific = buffer[3]; 2193 data->longlba = buffer[4] & 0x01; 2194 data->block_descriptor_length = buffer[6]*256 2195 + buffer[7]; 2196 } else { 2197 data->length = buffer[0] + 1; 2198 data->medium_type = buffer[1]; 2199 data->device_specific = buffer[2]; 2200 data->block_descriptor_length = buffer[3]; 2201 } 2202 data->header_length = header_length; 2203 2204 return 0; 2205 } 2206 EXPORT_SYMBOL(scsi_mode_sense); 2207 2208 /** 2209 * scsi_test_unit_ready - test if unit is ready 2210 * @sdev: scsi device to change the state of. 2211 * @timeout: command timeout 2212 * @retries: number of retries before failing 2213 * @sshdr: outpout pointer for decoded sense information. 2214 * 2215 * Returns zero if unsuccessful or an error if TUR failed. For 2216 * removable media, UNIT_ATTENTION sets ->changed flag. 2217 **/ 2218 int 2219 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries, 2220 struct scsi_sense_hdr *sshdr) 2221 { 2222 char cmd[] = { 2223 TEST_UNIT_READY, 0, 0, 0, 0, 0, 2224 }; 2225 int result; 2226 2227 /* try to eat the UNIT_ATTENTION if there are enough retries */ 2228 do { 2229 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr, 2230 timeout, 1, NULL); 2231 if (sdev->removable && scsi_sense_valid(sshdr) && 2232 sshdr->sense_key == UNIT_ATTENTION) 2233 sdev->changed = 1; 2234 } while (scsi_sense_valid(sshdr) && 2235 sshdr->sense_key == UNIT_ATTENTION && --retries); 2236 2237 return result; 2238 } 2239 EXPORT_SYMBOL(scsi_test_unit_ready); 2240 2241 /** 2242 * scsi_device_set_state - Take the given device through the device state model. 2243 * @sdev: scsi device to change the state of. 2244 * @state: state to change to. 2245 * 2246 * Returns zero if successful or an error if the requested 2247 * transition is illegal. 2248 */ 2249 int 2250 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state) 2251 { 2252 enum scsi_device_state oldstate = sdev->sdev_state; 2253 2254 if (state == oldstate) 2255 return 0; 2256 2257 switch (state) { 2258 case SDEV_CREATED: 2259 switch (oldstate) { 2260 case SDEV_CREATED_BLOCK: 2261 break; 2262 default: 2263 goto illegal; 2264 } 2265 break; 2266 2267 case SDEV_RUNNING: 2268 switch (oldstate) { 2269 case SDEV_CREATED: 2270 case SDEV_OFFLINE: 2271 case SDEV_TRANSPORT_OFFLINE: 2272 case SDEV_QUIESCE: 2273 case SDEV_BLOCK: 2274 break; 2275 default: 2276 goto illegal; 2277 } 2278 break; 2279 2280 case SDEV_QUIESCE: 2281 switch (oldstate) { 2282 case SDEV_RUNNING: 2283 case SDEV_OFFLINE: 2284 case SDEV_TRANSPORT_OFFLINE: 2285 break; 2286 default: 2287 goto illegal; 2288 } 2289 break; 2290 2291 case SDEV_OFFLINE: 2292 case SDEV_TRANSPORT_OFFLINE: 2293 switch (oldstate) { 2294 case SDEV_CREATED: 2295 case SDEV_RUNNING: 2296 case SDEV_QUIESCE: 2297 case SDEV_BLOCK: 2298 break; 2299 default: 2300 goto illegal; 2301 } 2302 break; 2303 2304 case SDEV_BLOCK: 2305 switch (oldstate) { 2306 case SDEV_RUNNING: 2307 case SDEV_CREATED_BLOCK: 2308 case SDEV_QUIESCE: 2309 case SDEV_OFFLINE: 2310 break; 2311 default: 2312 goto illegal; 2313 } 2314 break; 2315 2316 case SDEV_CREATED_BLOCK: 2317 switch (oldstate) { 2318 case SDEV_CREATED: 2319 break; 2320 default: 2321 goto illegal; 2322 } 2323 break; 2324 2325 case SDEV_CANCEL: 2326 switch (oldstate) { 2327 case SDEV_CREATED: 2328 case SDEV_RUNNING: 2329 case SDEV_QUIESCE: 2330 case SDEV_OFFLINE: 2331 case SDEV_TRANSPORT_OFFLINE: 2332 break; 2333 default: 2334 goto illegal; 2335 } 2336 break; 2337 2338 case SDEV_DEL: 2339 switch (oldstate) { 2340 case SDEV_CREATED: 2341 case SDEV_RUNNING: 2342 case SDEV_OFFLINE: 2343 case SDEV_TRANSPORT_OFFLINE: 2344 case SDEV_CANCEL: 2345 case SDEV_BLOCK: 2346 case SDEV_CREATED_BLOCK: 2347 break; 2348 default: 2349 goto illegal; 2350 } 2351 break; 2352 2353 } 2354 sdev->offline_already = false; 2355 sdev->sdev_state = state; 2356 return 0; 2357 2358 illegal: 2359 SCSI_LOG_ERROR_RECOVERY(1, 2360 sdev_printk(KERN_ERR, sdev, 2361 "Illegal state transition %s->%s", 2362 scsi_device_state_name(oldstate), 2363 scsi_device_state_name(state)) 2364 ); 2365 return -EINVAL; 2366 } 2367 EXPORT_SYMBOL(scsi_device_set_state); 2368 2369 /** 2370 * scsi_evt_emit - emit a single SCSI device uevent 2371 * @sdev: associated SCSI device 2372 * @evt: event to emit 2373 * 2374 * Send a single uevent (scsi_event) to the associated scsi_device. 2375 */ 2376 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt) 2377 { 2378 int idx = 0; 2379 char *envp[3]; 2380 2381 switch (evt->evt_type) { 2382 case SDEV_EVT_MEDIA_CHANGE: 2383 envp[idx++] = "SDEV_MEDIA_CHANGE=1"; 2384 break; 2385 case SDEV_EVT_INQUIRY_CHANGE_REPORTED: 2386 scsi_rescan_device(&sdev->sdev_gendev); 2387 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED"; 2388 break; 2389 case SDEV_EVT_CAPACITY_CHANGE_REPORTED: 2390 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED"; 2391 break; 2392 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED: 2393 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED"; 2394 break; 2395 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED: 2396 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED"; 2397 break; 2398 case SDEV_EVT_LUN_CHANGE_REPORTED: 2399 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED"; 2400 break; 2401 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED: 2402 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED"; 2403 break; 2404 case SDEV_EVT_POWER_ON_RESET_OCCURRED: 2405 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED"; 2406 break; 2407 default: 2408 /* do nothing */ 2409 break; 2410 } 2411 2412 envp[idx++] = NULL; 2413 2414 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp); 2415 } 2416 2417 /** 2418 * scsi_evt_thread - send a uevent for each scsi event 2419 * @work: work struct for scsi_device 2420 * 2421 * Dispatch queued events to their associated scsi_device kobjects 2422 * as uevents. 2423 */ 2424 void scsi_evt_thread(struct work_struct *work) 2425 { 2426 struct scsi_device *sdev; 2427 enum scsi_device_event evt_type; 2428 LIST_HEAD(event_list); 2429 2430 sdev = container_of(work, struct scsi_device, event_work); 2431 2432 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++) 2433 if (test_and_clear_bit(evt_type, sdev->pending_events)) 2434 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL); 2435 2436 while (1) { 2437 struct scsi_event *evt; 2438 struct list_head *this, *tmp; 2439 unsigned long flags; 2440 2441 spin_lock_irqsave(&sdev->list_lock, flags); 2442 list_splice_init(&sdev->event_list, &event_list); 2443 spin_unlock_irqrestore(&sdev->list_lock, flags); 2444 2445 if (list_empty(&event_list)) 2446 break; 2447 2448 list_for_each_safe(this, tmp, &event_list) { 2449 evt = list_entry(this, struct scsi_event, node); 2450 list_del(&evt->node); 2451 scsi_evt_emit(sdev, evt); 2452 kfree(evt); 2453 } 2454 } 2455 } 2456 2457 /** 2458 * sdev_evt_send - send asserted event to uevent thread 2459 * @sdev: scsi_device event occurred on 2460 * @evt: event to send 2461 * 2462 * Assert scsi device event asynchronously. 2463 */ 2464 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt) 2465 { 2466 unsigned long flags; 2467 2468 #if 0 2469 /* FIXME: currently this check eliminates all media change events 2470 * for polled devices. Need to update to discriminate between AN 2471 * and polled events */ 2472 if (!test_bit(evt->evt_type, sdev->supported_events)) { 2473 kfree(evt); 2474 return; 2475 } 2476 #endif 2477 2478 spin_lock_irqsave(&sdev->list_lock, flags); 2479 list_add_tail(&evt->node, &sdev->event_list); 2480 schedule_work(&sdev->event_work); 2481 spin_unlock_irqrestore(&sdev->list_lock, flags); 2482 } 2483 EXPORT_SYMBOL_GPL(sdev_evt_send); 2484 2485 /** 2486 * sdev_evt_alloc - allocate a new scsi event 2487 * @evt_type: type of event to allocate 2488 * @gfpflags: GFP flags for allocation 2489 * 2490 * Allocates and returns a new scsi_event. 2491 */ 2492 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type, 2493 gfp_t gfpflags) 2494 { 2495 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags); 2496 if (!evt) 2497 return NULL; 2498 2499 evt->evt_type = evt_type; 2500 INIT_LIST_HEAD(&evt->node); 2501 2502 /* evt_type-specific initialization, if any */ 2503 switch (evt_type) { 2504 case SDEV_EVT_MEDIA_CHANGE: 2505 case SDEV_EVT_INQUIRY_CHANGE_REPORTED: 2506 case SDEV_EVT_CAPACITY_CHANGE_REPORTED: 2507 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED: 2508 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED: 2509 case SDEV_EVT_LUN_CHANGE_REPORTED: 2510 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED: 2511 case SDEV_EVT_POWER_ON_RESET_OCCURRED: 2512 default: 2513 /* do nothing */ 2514 break; 2515 } 2516 2517 return evt; 2518 } 2519 EXPORT_SYMBOL_GPL(sdev_evt_alloc); 2520 2521 /** 2522 * sdev_evt_send_simple - send asserted event to uevent thread 2523 * @sdev: scsi_device event occurred on 2524 * @evt_type: type of event to send 2525 * @gfpflags: GFP flags for allocation 2526 * 2527 * Assert scsi device event asynchronously, given an event type. 2528 */ 2529 void sdev_evt_send_simple(struct scsi_device *sdev, 2530 enum scsi_device_event evt_type, gfp_t gfpflags) 2531 { 2532 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags); 2533 if (!evt) { 2534 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n", 2535 evt_type); 2536 return; 2537 } 2538 2539 sdev_evt_send(sdev, evt); 2540 } 2541 EXPORT_SYMBOL_GPL(sdev_evt_send_simple); 2542 2543 /** 2544 * scsi_device_quiesce - Block all commands except power management. 2545 * @sdev: scsi device to quiesce. 2546 * 2547 * This works by trying to transition to the SDEV_QUIESCE state 2548 * (which must be a legal transition). When the device is in this 2549 * state, only power management requests will be accepted, all others will 2550 * be deferred. 2551 * 2552 * Must be called with user context, may sleep. 2553 * 2554 * Returns zero if unsuccessful or an error if not. 2555 */ 2556 int 2557 scsi_device_quiesce(struct scsi_device *sdev) 2558 { 2559 struct request_queue *q = sdev->request_queue; 2560 int err; 2561 2562 /* 2563 * It is allowed to call scsi_device_quiesce() multiple times from 2564 * the same context but concurrent scsi_device_quiesce() calls are 2565 * not allowed. 2566 */ 2567 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current); 2568 2569 if (sdev->quiesced_by == current) 2570 return 0; 2571 2572 blk_set_pm_only(q); 2573 2574 blk_mq_freeze_queue(q); 2575 /* 2576 * Ensure that the effect of blk_set_pm_only() will be visible 2577 * for percpu_ref_tryget() callers that occur after the queue 2578 * unfreeze even if the queue was already frozen before this function 2579 * was called. See also https://lwn.net/Articles/573497/. 2580 */ 2581 synchronize_rcu(); 2582 blk_mq_unfreeze_queue(q); 2583 2584 mutex_lock(&sdev->state_mutex); 2585 err = scsi_device_set_state(sdev, SDEV_QUIESCE); 2586 if (err == 0) 2587 sdev->quiesced_by = current; 2588 else 2589 blk_clear_pm_only(q); 2590 mutex_unlock(&sdev->state_mutex); 2591 2592 return err; 2593 } 2594 EXPORT_SYMBOL(scsi_device_quiesce); 2595 2596 /** 2597 * scsi_device_resume - Restart user issued commands to a quiesced device. 2598 * @sdev: scsi device to resume. 2599 * 2600 * Moves the device from quiesced back to running and restarts the 2601 * queues. 2602 * 2603 * Must be called with user context, may sleep. 2604 */ 2605 void scsi_device_resume(struct scsi_device *sdev) 2606 { 2607 /* check if the device state was mutated prior to resume, and if 2608 * so assume the state is being managed elsewhere (for example 2609 * device deleted during suspend) 2610 */ 2611 mutex_lock(&sdev->state_mutex); 2612 if (sdev->sdev_state == SDEV_QUIESCE) 2613 scsi_device_set_state(sdev, SDEV_RUNNING); 2614 if (sdev->quiesced_by) { 2615 sdev->quiesced_by = NULL; 2616 blk_clear_pm_only(sdev->request_queue); 2617 } 2618 mutex_unlock(&sdev->state_mutex); 2619 } 2620 EXPORT_SYMBOL(scsi_device_resume); 2621 2622 static void 2623 device_quiesce_fn(struct scsi_device *sdev, void *data) 2624 { 2625 scsi_device_quiesce(sdev); 2626 } 2627 2628 void 2629 scsi_target_quiesce(struct scsi_target *starget) 2630 { 2631 starget_for_each_device(starget, NULL, device_quiesce_fn); 2632 } 2633 EXPORT_SYMBOL(scsi_target_quiesce); 2634 2635 static void 2636 device_resume_fn(struct scsi_device *sdev, void *data) 2637 { 2638 scsi_device_resume(sdev); 2639 } 2640 2641 void 2642 scsi_target_resume(struct scsi_target *starget) 2643 { 2644 starget_for_each_device(starget, NULL, device_resume_fn); 2645 } 2646 EXPORT_SYMBOL(scsi_target_resume); 2647 2648 /** 2649 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state 2650 * @sdev: device to block 2651 * 2652 * Pause SCSI command processing on the specified device. Does not sleep. 2653 * 2654 * Returns zero if successful or a negative error code upon failure. 2655 * 2656 * Notes: 2657 * This routine transitions the device to the SDEV_BLOCK state (which must be 2658 * a legal transition). When the device is in this state, command processing 2659 * is paused until the device leaves the SDEV_BLOCK state. See also 2660 * scsi_internal_device_unblock_nowait(). 2661 */ 2662 int scsi_internal_device_block_nowait(struct scsi_device *sdev) 2663 { 2664 struct request_queue *q = sdev->request_queue; 2665 int err = 0; 2666 2667 err = scsi_device_set_state(sdev, SDEV_BLOCK); 2668 if (err) { 2669 err = scsi_device_set_state(sdev, SDEV_CREATED_BLOCK); 2670 2671 if (err) 2672 return err; 2673 } 2674 2675 /* 2676 * The device has transitioned to SDEV_BLOCK. Stop the 2677 * block layer from calling the midlayer with this device's 2678 * request queue. 2679 */ 2680 blk_mq_quiesce_queue_nowait(q); 2681 return 0; 2682 } 2683 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); 2684 2685 /** 2686 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state 2687 * @sdev: device to block 2688 * 2689 * Pause SCSI command processing on the specified device and wait until all 2690 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep. 2691 * 2692 * Returns zero if successful or a negative error code upon failure. 2693 * 2694 * Note: 2695 * This routine transitions the device to the SDEV_BLOCK state (which must be 2696 * a legal transition). When the device is in this state, command processing 2697 * is paused until the device leaves the SDEV_BLOCK state. See also 2698 * scsi_internal_device_unblock(). 2699 */ 2700 static int scsi_internal_device_block(struct scsi_device *sdev) 2701 { 2702 struct request_queue *q = sdev->request_queue; 2703 int err; 2704 2705 mutex_lock(&sdev->state_mutex); 2706 err = scsi_internal_device_block_nowait(sdev); 2707 if (err == 0) 2708 blk_mq_quiesce_queue(q); 2709 mutex_unlock(&sdev->state_mutex); 2710 2711 return err; 2712 } 2713 2714 void scsi_start_queue(struct scsi_device *sdev) 2715 { 2716 struct request_queue *q = sdev->request_queue; 2717 2718 blk_mq_unquiesce_queue(q); 2719 } 2720 2721 /** 2722 * scsi_internal_device_unblock_nowait - resume a device after a block request 2723 * @sdev: device to resume 2724 * @new_state: state to set the device to after unblocking 2725 * 2726 * Restart the device queue for a previously suspended SCSI device. Does not 2727 * sleep. 2728 * 2729 * Returns zero if successful or a negative error code upon failure. 2730 * 2731 * Notes: 2732 * This routine transitions the device to the SDEV_RUNNING state or to one of 2733 * the offline states (which must be a legal transition) allowing the midlayer 2734 * to goose the queue for this device. 2735 */ 2736 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev, 2737 enum scsi_device_state new_state) 2738 { 2739 switch (new_state) { 2740 case SDEV_RUNNING: 2741 case SDEV_TRANSPORT_OFFLINE: 2742 break; 2743 default: 2744 return -EINVAL; 2745 } 2746 2747 /* 2748 * Try to transition the scsi device to SDEV_RUNNING or one of the 2749 * offlined states and goose the device queue if successful. 2750 */ 2751 switch (sdev->sdev_state) { 2752 case SDEV_BLOCK: 2753 case SDEV_TRANSPORT_OFFLINE: 2754 sdev->sdev_state = new_state; 2755 break; 2756 case SDEV_CREATED_BLOCK: 2757 if (new_state == SDEV_TRANSPORT_OFFLINE || 2758 new_state == SDEV_OFFLINE) 2759 sdev->sdev_state = new_state; 2760 else 2761 sdev->sdev_state = SDEV_CREATED; 2762 break; 2763 case SDEV_CANCEL: 2764 case SDEV_OFFLINE: 2765 break; 2766 default: 2767 return -EINVAL; 2768 } 2769 scsi_start_queue(sdev); 2770 2771 return 0; 2772 } 2773 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait); 2774 2775 /** 2776 * scsi_internal_device_unblock - resume a device after a block request 2777 * @sdev: device to resume 2778 * @new_state: state to set the device to after unblocking 2779 * 2780 * Restart the device queue for a previously suspended SCSI device. May sleep. 2781 * 2782 * Returns zero if successful or a negative error code upon failure. 2783 * 2784 * Notes: 2785 * This routine transitions the device to the SDEV_RUNNING state or to one of 2786 * the offline states (which must be a legal transition) allowing the midlayer 2787 * to goose the queue for this device. 2788 */ 2789 static int scsi_internal_device_unblock(struct scsi_device *sdev, 2790 enum scsi_device_state new_state) 2791 { 2792 int ret; 2793 2794 mutex_lock(&sdev->state_mutex); 2795 ret = scsi_internal_device_unblock_nowait(sdev, new_state); 2796 mutex_unlock(&sdev->state_mutex); 2797 2798 return ret; 2799 } 2800 2801 static void 2802 device_block(struct scsi_device *sdev, void *data) 2803 { 2804 int ret; 2805 2806 ret = scsi_internal_device_block(sdev); 2807 2808 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n", 2809 dev_name(&sdev->sdev_gendev), ret); 2810 } 2811 2812 static int 2813 target_block(struct device *dev, void *data) 2814 { 2815 if (scsi_is_target_device(dev)) 2816 starget_for_each_device(to_scsi_target(dev), NULL, 2817 device_block); 2818 return 0; 2819 } 2820 2821 void 2822 scsi_target_block(struct device *dev) 2823 { 2824 if (scsi_is_target_device(dev)) 2825 starget_for_each_device(to_scsi_target(dev), NULL, 2826 device_block); 2827 else 2828 device_for_each_child(dev, NULL, target_block); 2829 } 2830 EXPORT_SYMBOL_GPL(scsi_target_block); 2831 2832 static void 2833 device_unblock(struct scsi_device *sdev, void *data) 2834 { 2835 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data); 2836 } 2837 2838 static int 2839 target_unblock(struct device *dev, void *data) 2840 { 2841 if (scsi_is_target_device(dev)) 2842 starget_for_each_device(to_scsi_target(dev), data, 2843 device_unblock); 2844 return 0; 2845 } 2846 2847 void 2848 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state) 2849 { 2850 if (scsi_is_target_device(dev)) 2851 starget_for_each_device(to_scsi_target(dev), &new_state, 2852 device_unblock); 2853 else 2854 device_for_each_child(dev, &new_state, target_unblock); 2855 } 2856 EXPORT_SYMBOL_GPL(scsi_target_unblock); 2857 2858 int 2859 scsi_host_block(struct Scsi_Host *shost) 2860 { 2861 struct scsi_device *sdev; 2862 int ret = 0; 2863 2864 /* 2865 * Call scsi_internal_device_block_nowait so we can avoid 2866 * calling synchronize_rcu() for each LUN. 2867 */ 2868 shost_for_each_device(sdev, shost) { 2869 mutex_lock(&sdev->state_mutex); 2870 ret = scsi_internal_device_block_nowait(sdev); 2871 mutex_unlock(&sdev->state_mutex); 2872 if (ret) { 2873 scsi_device_put(sdev); 2874 break; 2875 } 2876 } 2877 2878 /* 2879 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so 2880 * calling synchronize_rcu() once is enough. 2881 */ 2882 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING); 2883 2884 if (!ret) 2885 synchronize_rcu(); 2886 2887 return ret; 2888 } 2889 EXPORT_SYMBOL_GPL(scsi_host_block); 2890 2891 int 2892 scsi_host_unblock(struct Scsi_Host *shost, int new_state) 2893 { 2894 struct scsi_device *sdev; 2895 int ret = 0; 2896 2897 shost_for_each_device(sdev, shost) { 2898 ret = scsi_internal_device_unblock(sdev, new_state); 2899 if (ret) { 2900 scsi_device_put(sdev); 2901 break; 2902 } 2903 } 2904 return ret; 2905 } 2906 EXPORT_SYMBOL_GPL(scsi_host_unblock); 2907 2908 /** 2909 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt 2910 * @sgl: scatter-gather list 2911 * @sg_count: number of segments in sg 2912 * @offset: offset in bytes into sg, on return offset into the mapped area 2913 * @len: bytes to map, on return number of bytes mapped 2914 * 2915 * Returns virtual address of the start of the mapped page 2916 */ 2917 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count, 2918 size_t *offset, size_t *len) 2919 { 2920 int i; 2921 size_t sg_len = 0, len_complete = 0; 2922 struct scatterlist *sg; 2923 struct page *page; 2924 2925 WARN_ON(!irqs_disabled()); 2926 2927 for_each_sg(sgl, sg, sg_count, i) { 2928 len_complete = sg_len; /* Complete sg-entries */ 2929 sg_len += sg->length; 2930 if (sg_len > *offset) 2931 break; 2932 } 2933 2934 if (unlikely(i == sg_count)) { 2935 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, " 2936 "elements %d\n", 2937 __func__, sg_len, *offset, sg_count); 2938 WARN_ON(1); 2939 return NULL; 2940 } 2941 2942 /* Offset starting from the beginning of first page in this sg-entry */ 2943 *offset = *offset - len_complete + sg->offset; 2944 2945 /* Assumption: contiguous pages can be accessed as "page + i" */ 2946 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT)); 2947 *offset &= ~PAGE_MASK; 2948 2949 /* Bytes in this sg-entry from *offset to the end of the page */ 2950 sg_len = PAGE_SIZE - *offset; 2951 if (*len > sg_len) 2952 *len = sg_len; 2953 2954 return kmap_atomic(page); 2955 } 2956 EXPORT_SYMBOL(scsi_kmap_atomic_sg); 2957 2958 /** 2959 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg 2960 * @virt: virtual address to be unmapped 2961 */ 2962 void scsi_kunmap_atomic_sg(void *virt) 2963 { 2964 kunmap_atomic(virt); 2965 } 2966 EXPORT_SYMBOL(scsi_kunmap_atomic_sg); 2967 2968 void sdev_disable_disk_events(struct scsi_device *sdev) 2969 { 2970 atomic_inc(&sdev->disk_events_disable_depth); 2971 } 2972 EXPORT_SYMBOL(sdev_disable_disk_events); 2973 2974 void sdev_enable_disk_events(struct scsi_device *sdev) 2975 { 2976 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0)) 2977 return; 2978 atomic_dec(&sdev->disk_events_disable_depth); 2979 } 2980 EXPORT_SYMBOL(sdev_enable_disk_events); 2981 2982 static unsigned char designator_prio(const unsigned char *d) 2983 { 2984 if (d[1] & 0x30) 2985 /* not associated with LUN */ 2986 return 0; 2987 2988 if (d[3] == 0) 2989 /* invalid length */ 2990 return 0; 2991 2992 /* 2993 * Order of preference for lun descriptor: 2994 * - SCSI name string 2995 * - NAA IEEE Registered Extended 2996 * - EUI-64 based 16-byte 2997 * - EUI-64 based 12-byte 2998 * - NAA IEEE Registered 2999 * - NAA IEEE Extended 3000 * - EUI-64 based 8-byte 3001 * - SCSI name string (truncated) 3002 * - T10 Vendor ID 3003 * as longer descriptors reduce the likelyhood 3004 * of identification clashes. 3005 */ 3006 3007 switch (d[1] & 0xf) { 3008 case 8: 3009 /* SCSI name string, variable-length UTF-8 */ 3010 return 9; 3011 case 3: 3012 switch (d[4] >> 4) { 3013 case 6: 3014 /* NAA registered extended */ 3015 return 8; 3016 case 5: 3017 /* NAA registered */ 3018 return 5; 3019 case 4: 3020 /* NAA extended */ 3021 return 4; 3022 case 3: 3023 /* NAA locally assigned */ 3024 return 1; 3025 default: 3026 break; 3027 } 3028 break; 3029 case 2: 3030 switch (d[3]) { 3031 case 16: 3032 /* EUI64-based, 16 byte */ 3033 return 7; 3034 case 12: 3035 /* EUI64-based, 12 byte */ 3036 return 6; 3037 case 8: 3038 /* EUI64-based, 8 byte */ 3039 return 3; 3040 default: 3041 break; 3042 } 3043 break; 3044 case 1: 3045 /* T10 vendor ID */ 3046 return 1; 3047 default: 3048 break; 3049 } 3050 3051 return 0; 3052 } 3053 3054 /** 3055 * scsi_vpd_lun_id - return a unique device identification 3056 * @sdev: SCSI device 3057 * @id: buffer for the identification 3058 * @id_len: length of the buffer 3059 * 3060 * Copies a unique device identification into @id based 3061 * on the information in the VPD page 0x83 of the device. 3062 * The string will be formatted as a SCSI name string. 3063 * 3064 * Returns the length of the identification or error on failure. 3065 * If the identifier is longer than the supplied buffer the actual 3066 * identifier length is returned and the buffer is not zero-padded. 3067 */ 3068 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len) 3069 { 3070 u8 cur_id_prio = 0; 3071 u8 cur_id_size = 0; 3072 const unsigned char *d, *cur_id_str; 3073 const struct scsi_vpd *vpd_pg83; 3074 int id_size = -EINVAL; 3075 3076 rcu_read_lock(); 3077 vpd_pg83 = rcu_dereference(sdev->vpd_pg83); 3078 if (!vpd_pg83) { 3079 rcu_read_unlock(); 3080 return -ENXIO; 3081 } 3082 3083 /* The id string must be at least 20 bytes + terminating NULL byte */ 3084 if (id_len < 21) { 3085 rcu_read_unlock(); 3086 return -EINVAL; 3087 } 3088 3089 memset(id, 0, id_len); 3090 for (d = vpd_pg83->data + 4; 3091 d < vpd_pg83->data + vpd_pg83->len; 3092 d += d[3] + 4) { 3093 u8 prio = designator_prio(d); 3094 3095 if (prio == 0 || cur_id_prio > prio) 3096 continue; 3097 3098 switch (d[1] & 0xf) { 3099 case 0x1: 3100 /* T10 Vendor ID */ 3101 if (cur_id_size > d[3]) 3102 break; 3103 cur_id_prio = prio; 3104 cur_id_size = d[3]; 3105 if (cur_id_size + 4 > id_len) 3106 cur_id_size = id_len - 4; 3107 cur_id_str = d + 4; 3108 id_size = snprintf(id, id_len, "t10.%*pE", 3109 cur_id_size, cur_id_str); 3110 break; 3111 case 0x2: 3112 /* EUI-64 */ 3113 cur_id_prio = prio; 3114 cur_id_size = d[3]; 3115 cur_id_str = d + 4; 3116 switch (cur_id_size) { 3117 case 8: 3118 id_size = snprintf(id, id_len, 3119 "eui.%8phN", 3120 cur_id_str); 3121 break; 3122 case 12: 3123 id_size = snprintf(id, id_len, 3124 "eui.%12phN", 3125 cur_id_str); 3126 break; 3127 case 16: 3128 id_size = snprintf(id, id_len, 3129 "eui.%16phN", 3130 cur_id_str); 3131 break; 3132 default: 3133 break; 3134 } 3135 break; 3136 case 0x3: 3137 /* NAA */ 3138 cur_id_prio = prio; 3139 cur_id_size = d[3]; 3140 cur_id_str = d + 4; 3141 switch (cur_id_size) { 3142 case 8: 3143 id_size = snprintf(id, id_len, 3144 "naa.%8phN", 3145 cur_id_str); 3146 break; 3147 case 16: 3148 id_size = snprintf(id, id_len, 3149 "naa.%16phN", 3150 cur_id_str); 3151 break; 3152 default: 3153 break; 3154 } 3155 break; 3156 case 0x8: 3157 /* SCSI name string */ 3158 if (cur_id_size > d[3]) 3159 break; 3160 /* Prefer others for truncated descriptor */ 3161 if (d[3] > id_len) { 3162 prio = 2; 3163 if (cur_id_prio > prio) 3164 break; 3165 } 3166 cur_id_prio = prio; 3167 cur_id_size = id_size = d[3]; 3168 cur_id_str = d + 4; 3169 if (cur_id_size >= id_len) 3170 cur_id_size = id_len - 1; 3171 memcpy(id, cur_id_str, cur_id_size); 3172 break; 3173 default: 3174 break; 3175 } 3176 } 3177 rcu_read_unlock(); 3178 3179 return id_size; 3180 } 3181 EXPORT_SYMBOL(scsi_vpd_lun_id); 3182 3183 /* 3184 * scsi_vpd_tpg_id - return a target port group identifier 3185 * @sdev: SCSI device 3186 * 3187 * Returns the Target Port Group identifier from the information 3188 * froom VPD page 0x83 of the device. 3189 * 3190 * Returns the identifier or error on failure. 3191 */ 3192 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id) 3193 { 3194 const unsigned char *d; 3195 const struct scsi_vpd *vpd_pg83; 3196 int group_id = -EAGAIN, rel_port = -1; 3197 3198 rcu_read_lock(); 3199 vpd_pg83 = rcu_dereference(sdev->vpd_pg83); 3200 if (!vpd_pg83) { 3201 rcu_read_unlock(); 3202 return -ENXIO; 3203 } 3204 3205 d = vpd_pg83->data + 4; 3206 while (d < vpd_pg83->data + vpd_pg83->len) { 3207 switch (d[1] & 0xf) { 3208 case 0x4: 3209 /* Relative target port */ 3210 rel_port = get_unaligned_be16(&d[6]); 3211 break; 3212 case 0x5: 3213 /* Target port group */ 3214 group_id = get_unaligned_be16(&d[6]); 3215 break; 3216 default: 3217 break; 3218 } 3219 d += d[3] + 4; 3220 } 3221 rcu_read_unlock(); 3222 3223 if (group_id >= 0 && rel_id && rel_port != -1) 3224 *rel_id = rel_port; 3225 3226 return group_id; 3227 } 3228 EXPORT_SYMBOL(scsi_vpd_tpg_id); 3229 3230 /** 3231 * scsi_build_sense - build sense data for a command 3232 * @scmd: scsi command for which the sense should be formatted 3233 * @desc: Sense format (non-zero == descriptor format, 3234 * 0 == fixed format) 3235 * @key: Sense key 3236 * @asc: Additional sense code 3237 * @ascq: Additional sense code qualifier 3238 * 3239 **/ 3240 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq) 3241 { 3242 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq); 3243 scmd->result = SAM_STAT_CHECK_CONDITION; 3244 } 3245 EXPORT_SYMBOL_GPL(scsi_build_sense); 3246