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