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