1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 1999 Eric Youngdale 4 * Copyright (C) 2014 Christoph Hellwig 5 * 6 * SCSI queueing library. 7 * Initial versions: Eric Youngdale (eric@andante.org). 8 * Based upon conversations with large numbers 9 * of people at Linux Expo. 10 */ 11 12 #include <linux/bio.h> 13 #include <linux/bitops.h> 14 #include <linux/blkdev.h> 15 #include <linux/completion.h> 16 #include <linux/kernel.h> 17 #include <linux/export.h> 18 #include <linux/init.h> 19 #include <linux/pci.h> 20 #include <linux/delay.h> 21 #include <linux/hardirq.h> 22 #include <linux/scatterlist.h> 23 #include <linux/blk-mq.h> 24 #include <linux/blk-integrity.h> 25 #include <linux/ratelimit.h> 26 #include <asm/unaligned.h> 27 28 #include <scsi/scsi.h> 29 #include <scsi/scsi_cmnd.h> 30 #include <scsi/scsi_dbg.h> 31 #include <scsi/scsi_device.h> 32 #include <scsi/scsi_driver.h> 33 #include <scsi/scsi_eh.h> 34 #include <scsi/scsi_host.h> 35 #include <scsi/scsi_transport.h> /* __scsi_init_queue() */ 36 #include <scsi/scsi_dh.h> 37 38 #include <trace/events/scsi.h> 39 40 #include "scsi_debugfs.h" 41 #include "scsi_priv.h" 42 #include "scsi_logging.h" 43 44 /* 45 * Size of integrity metadata is usually small, 1 inline sg should 46 * cover normal cases. 47 */ 48 #ifdef CONFIG_ARCH_NO_SG_CHAIN 49 #define SCSI_INLINE_PROT_SG_CNT 0 50 #define SCSI_INLINE_SG_CNT 0 51 #else 52 #define SCSI_INLINE_PROT_SG_CNT 1 53 #define SCSI_INLINE_SG_CNT 2 54 #endif 55 56 static struct kmem_cache *scsi_sense_cache; 57 static DEFINE_MUTEX(scsi_sense_cache_mutex); 58 59 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd); 60 61 int scsi_init_sense_cache(struct Scsi_Host *shost) 62 { 63 int ret = 0; 64 65 mutex_lock(&scsi_sense_cache_mutex); 66 if (!scsi_sense_cache) { 67 scsi_sense_cache = 68 kmem_cache_create_usercopy("scsi_sense_cache", 69 SCSI_SENSE_BUFFERSIZE, 0, SLAB_HWCACHE_ALIGN, 70 0, SCSI_SENSE_BUFFERSIZE, NULL); 71 if (!scsi_sense_cache) 72 ret = -ENOMEM; 73 } 74 mutex_unlock(&scsi_sense_cache_mutex); 75 return ret; 76 } 77 78 static void 79 scsi_set_blocked(struct scsi_cmnd *cmd, int reason) 80 { 81 struct Scsi_Host *host = cmd->device->host; 82 struct scsi_device *device = cmd->device; 83 struct scsi_target *starget = scsi_target(device); 84 85 /* 86 * Set the appropriate busy bit for the device/host. 87 * 88 * If the host/device isn't busy, assume that something actually 89 * completed, and that we should be able to queue a command now. 90 * 91 * Note that the prior mid-layer assumption that any host could 92 * always queue at least one command is now broken. The mid-layer 93 * will implement a user specifiable stall (see 94 * scsi_host.max_host_blocked and scsi_device.max_device_blocked) 95 * if a command is requeued with no other commands outstanding 96 * either for the device or for the host. 97 */ 98 switch (reason) { 99 case SCSI_MLQUEUE_HOST_BUSY: 100 atomic_set(&host->host_blocked, host->max_host_blocked); 101 break; 102 case SCSI_MLQUEUE_DEVICE_BUSY: 103 case SCSI_MLQUEUE_EH_RETRY: 104 atomic_set(&device->device_blocked, 105 device->max_device_blocked); 106 break; 107 case SCSI_MLQUEUE_TARGET_BUSY: 108 atomic_set(&starget->target_blocked, 109 starget->max_target_blocked); 110 break; 111 } 112 } 113 114 static void scsi_mq_requeue_cmd(struct scsi_cmnd *cmd) 115 { 116 struct request *rq = scsi_cmd_to_rq(cmd); 117 118 if (rq->rq_flags & RQF_DONTPREP) { 119 rq->rq_flags &= ~RQF_DONTPREP; 120 scsi_mq_uninit_cmd(cmd); 121 } else { 122 WARN_ON_ONCE(true); 123 } 124 blk_mq_requeue_request(rq, true); 125 } 126 127 /** 128 * __scsi_queue_insert - private queue insertion 129 * @cmd: The SCSI command being requeued 130 * @reason: The reason for the requeue 131 * @unbusy: Whether the queue should be unbusied 132 * 133 * This is a private queue insertion. The public interface 134 * scsi_queue_insert() always assumes the queue should be unbusied 135 * because it's always called before the completion. This function is 136 * for a requeue after completion, which should only occur in this 137 * file. 138 */ 139 static void __scsi_queue_insert(struct scsi_cmnd *cmd, int reason, bool unbusy) 140 { 141 struct scsi_device *device = cmd->device; 142 143 SCSI_LOG_MLQUEUE(1, scmd_printk(KERN_INFO, cmd, 144 "Inserting command %p into mlqueue\n", cmd)); 145 146 scsi_set_blocked(cmd, reason); 147 148 /* 149 * Decrement the counters, since these commands are no longer 150 * active on the host/device. 151 */ 152 if (unbusy) 153 scsi_device_unbusy(device, cmd); 154 155 /* 156 * Requeue this command. It will go before all other commands 157 * that are already in the queue. Schedule requeue work under 158 * lock such that the kblockd_schedule_work() call happens 159 * before blk_mq_destroy_queue() finishes. 160 */ 161 cmd->result = 0; 162 163 blk_mq_requeue_request(scsi_cmd_to_rq(cmd), true); 164 } 165 166 /** 167 * scsi_queue_insert - Reinsert a command in the queue. 168 * @cmd: command that we are adding to queue. 169 * @reason: why we are inserting command to queue. 170 * 171 * We do this for one of two cases. Either the host is busy and it cannot accept 172 * any more commands for the time being, or the device returned QUEUE_FULL and 173 * can accept no more commands. 174 * 175 * Context: This could be called either from an interrupt context or a normal 176 * process context. 177 */ 178 void scsi_queue_insert(struct scsi_cmnd *cmd, int reason) 179 { 180 __scsi_queue_insert(cmd, reason, true); 181 } 182 183 184 /** 185 * __scsi_execute - insert request and wait for the result 186 * @sdev: scsi device 187 * @cmd: scsi command 188 * @data_direction: data direction 189 * @buffer: data buffer 190 * @bufflen: len of buffer 191 * @sense: optional sense buffer 192 * @sshdr: optional decoded sense header 193 * @timeout: request timeout in HZ 194 * @retries: number of times to retry request 195 * @flags: flags for ->cmd_flags 196 * @rq_flags: flags for ->rq_flags 197 * @resid: optional residual length 198 * 199 * Returns the scsi_cmnd result field if a command was executed, or a negative 200 * Linux error code if we didn't get that far. 201 */ 202 int __scsi_execute(struct scsi_device *sdev, const unsigned char *cmd, 203 int data_direction, void *buffer, unsigned bufflen, 204 unsigned char *sense, struct scsi_sense_hdr *sshdr, 205 int timeout, int retries, blk_opf_t flags, 206 req_flags_t rq_flags, int *resid) 207 { 208 struct request *req; 209 struct scsi_cmnd *scmd; 210 int ret; 211 212 req = scsi_alloc_request(sdev->request_queue, 213 data_direction == DMA_TO_DEVICE ? 214 REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 215 rq_flags & RQF_PM ? BLK_MQ_REQ_PM : 0); 216 if (IS_ERR(req)) 217 return PTR_ERR(req); 218 219 if (bufflen) { 220 ret = blk_rq_map_kern(sdev->request_queue, req, 221 buffer, bufflen, GFP_NOIO); 222 if (ret) 223 goto out; 224 } 225 scmd = blk_mq_rq_to_pdu(req); 226 scmd->cmd_len = COMMAND_SIZE(cmd[0]); 227 memcpy(scmd->cmnd, cmd, scmd->cmd_len); 228 scmd->allowed = retries; 229 req->timeout = timeout; 230 req->cmd_flags |= flags; 231 req->rq_flags |= rq_flags | RQF_QUIET; 232 233 /* 234 * head injection *required* here otherwise quiesce won't work 235 */ 236 blk_execute_rq(req, true); 237 238 /* 239 * Some devices (USB mass-storage in particular) may transfer 240 * garbage data together with a residue indicating that the data 241 * is invalid. Prevent the garbage from being misinterpreted 242 * and prevent security leaks by zeroing out the excess data. 243 */ 244 if (unlikely(scmd->resid_len > 0 && scmd->resid_len <= bufflen)) 245 memset(buffer + bufflen - scmd->resid_len, 0, scmd->resid_len); 246 247 if (resid) 248 *resid = scmd->resid_len; 249 if (sense && scmd->sense_len) 250 memcpy(sense, scmd->sense_buffer, SCSI_SENSE_BUFFERSIZE); 251 if (sshdr) 252 scsi_normalize_sense(scmd->sense_buffer, scmd->sense_len, 253 sshdr); 254 ret = scmd->result; 255 out: 256 blk_mq_free_request(req); 257 258 return ret; 259 } 260 EXPORT_SYMBOL(__scsi_execute); 261 262 /* 263 * Wake up the error handler if necessary. Avoid as follows that the error 264 * handler is not woken up if host in-flight requests number == 265 * shost->host_failed: use call_rcu() in scsi_eh_scmd_add() in combination 266 * with an RCU read lock in this function to ensure that this function in 267 * its entirety either finishes before scsi_eh_scmd_add() increases the 268 * host_failed counter or that it notices the shost state change made by 269 * scsi_eh_scmd_add(). 270 */ 271 static void scsi_dec_host_busy(struct Scsi_Host *shost, struct scsi_cmnd *cmd) 272 { 273 unsigned long flags; 274 275 rcu_read_lock(); 276 __clear_bit(SCMD_STATE_INFLIGHT, &cmd->state); 277 if (unlikely(scsi_host_in_recovery(shost))) { 278 spin_lock_irqsave(shost->host_lock, flags); 279 if (shost->host_failed || shost->host_eh_scheduled) 280 scsi_eh_wakeup(shost); 281 spin_unlock_irqrestore(shost->host_lock, flags); 282 } 283 rcu_read_unlock(); 284 } 285 286 void scsi_device_unbusy(struct scsi_device *sdev, struct scsi_cmnd *cmd) 287 { 288 struct Scsi_Host *shost = sdev->host; 289 struct scsi_target *starget = scsi_target(sdev); 290 291 scsi_dec_host_busy(shost, cmd); 292 293 if (starget->can_queue > 0) 294 atomic_dec(&starget->target_busy); 295 296 sbitmap_put(&sdev->budget_map, cmd->budget_token); 297 cmd->budget_token = -1; 298 } 299 300 static void scsi_kick_queue(struct request_queue *q) 301 { 302 blk_mq_run_hw_queues(q, false); 303 } 304 305 /* 306 * Called for single_lun devices on IO completion. Clear starget_sdev_user, 307 * and call blk_run_queue for all the scsi_devices on the target - 308 * including current_sdev first. 309 * 310 * Called with *no* scsi locks held. 311 */ 312 static void scsi_single_lun_run(struct scsi_device *current_sdev) 313 { 314 struct Scsi_Host *shost = current_sdev->host; 315 struct scsi_device *sdev, *tmp; 316 struct scsi_target *starget = scsi_target(current_sdev); 317 unsigned long flags; 318 319 spin_lock_irqsave(shost->host_lock, flags); 320 starget->starget_sdev_user = NULL; 321 spin_unlock_irqrestore(shost->host_lock, flags); 322 323 /* 324 * Call blk_run_queue for all LUNs on the target, starting with 325 * current_sdev. We race with others (to set starget_sdev_user), 326 * but in most cases, we will be first. Ideally, each LU on the 327 * target would get some limited time or requests on the target. 328 */ 329 scsi_kick_queue(current_sdev->request_queue); 330 331 spin_lock_irqsave(shost->host_lock, flags); 332 if (starget->starget_sdev_user) 333 goto out; 334 list_for_each_entry_safe(sdev, tmp, &starget->devices, 335 same_target_siblings) { 336 if (sdev == current_sdev) 337 continue; 338 if (scsi_device_get(sdev)) 339 continue; 340 341 spin_unlock_irqrestore(shost->host_lock, flags); 342 scsi_kick_queue(sdev->request_queue); 343 spin_lock_irqsave(shost->host_lock, flags); 344 345 scsi_device_put(sdev); 346 } 347 out: 348 spin_unlock_irqrestore(shost->host_lock, flags); 349 } 350 351 static inline bool scsi_device_is_busy(struct scsi_device *sdev) 352 { 353 if (scsi_device_busy(sdev) >= sdev->queue_depth) 354 return true; 355 if (atomic_read(&sdev->device_blocked) > 0) 356 return true; 357 return false; 358 } 359 360 static inline bool scsi_target_is_busy(struct scsi_target *starget) 361 { 362 if (starget->can_queue > 0) { 363 if (atomic_read(&starget->target_busy) >= starget->can_queue) 364 return true; 365 if (atomic_read(&starget->target_blocked) > 0) 366 return true; 367 } 368 return false; 369 } 370 371 static inline bool scsi_host_is_busy(struct Scsi_Host *shost) 372 { 373 if (atomic_read(&shost->host_blocked) > 0) 374 return true; 375 if (shost->host_self_blocked) 376 return true; 377 return false; 378 } 379 380 static void scsi_starved_list_run(struct Scsi_Host *shost) 381 { 382 LIST_HEAD(starved_list); 383 struct scsi_device *sdev; 384 unsigned long flags; 385 386 spin_lock_irqsave(shost->host_lock, flags); 387 list_splice_init(&shost->starved_list, &starved_list); 388 389 while (!list_empty(&starved_list)) { 390 struct request_queue *slq; 391 392 /* 393 * As long as shost is accepting commands and we have 394 * starved queues, call blk_run_queue. scsi_request_fn 395 * drops the queue_lock and can add us back to the 396 * starved_list. 397 * 398 * host_lock protects the starved_list and starved_entry. 399 * scsi_request_fn must get the host_lock before checking 400 * or modifying starved_list or starved_entry. 401 */ 402 if (scsi_host_is_busy(shost)) 403 break; 404 405 sdev = list_entry(starved_list.next, 406 struct scsi_device, starved_entry); 407 list_del_init(&sdev->starved_entry); 408 if (scsi_target_is_busy(scsi_target(sdev))) { 409 list_move_tail(&sdev->starved_entry, 410 &shost->starved_list); 411 continue; 412 } 413 414 /* 415 * Once we drop the host lock, a racing scsi_remove_device() 416 * call may remove the sdev from the starved list and destroy 417 * it and the queue. Mitigate by taking a reference to the 418 * queue and never touching the sdev again after we drop the 419 * host lock. Note: if __scsi_remove_device() invokes 420 * blk_mq_destroy_queue() before the queue is run from this 421 * function then blk_run_queue() will return immediately since 422 * blk_mq_destroy_queue() marks the queue with QUEUE_FLAG_DYING. 423 */ 424 slq = sdev->request_queue; 425 if (!blk_get_queue(slq)) 426 continue; 427 spin_unlock_irqrestore(shost->host_lock, flags); 428 429 scsi_kick_queue(slq); 430 blk_put_queue(slq); 431 432 spin_lock_irqsave(shost->host_lock, flags); 433 } 434 /* put any unprocessed entries back */ 435 list_splice(&starved_list, &shost->starved_list); 436 spin_unlock_irqrestore(shost->host_lock, flags); 437 } 438 439 /** 440 * scsi_run_queue - Select a proper request queue to serve next. 441 * @q: last request's queue 442 * 443 * The previous command was completely finished, start a new one if possible. 444 */ 445 static void scsi_run_queue(struct request_queue *q) 446 { 447 struct scsi_device *sdev = q->queuedata; 448 449 if (scsi_target(sdev)->single_lun) 450 scsi_single_lun_run(sdev); 451 if (!list_empty(&sdev->host->starved_list)) 452 scsi_starved_list_run(sdev->host); 453 454 blk_mq_run_hw_queues(q, false); 455 } 456 457 void scsi_requeue_run_queue(struct work_struct *work) 458 { 459 struct scsi_device *sdev; 460 struct request_queue *q; 461 462 sdev = container_of(work, struct scsi_device, requeue_work); 463 q = sdev->request_queue; 464 scsi_run_queue(q); 465 } 466 467 void scsi_run_host_queues(struct Scsi_Host *shost) 468 { 469 struct scsi_device *sdev; 470 471 shost_for_each_device(sdev, shost) 472 scsi_run_queue(sdev->request_queue); 473 } 474 475 static void scsi_uninit_cmd(struct scsi_cmnd *cmd) 476 { 477 if (!blk_rq_is_passthrough(scsi_cmd_to_rq(cmd))) { 478 struct scsi_driver *drv = scsi_cmd_to_driver(cmd); 479 480 if (drv->uninit_command) 481 drv->uninit_command(cmd); 482 } 483 } 484 485 void scsi_free_sgtables(struct scsi_cmnd *cmd) 486 { 487 if (cmd->sdb.table.nents) 488 sg_free_table_chained(&cmd->sdb.table, 489 SCSI_INLINE_SG_CNT); 490 if (scsi_prot_sg_count(cmd)) 491 sg_free_table_chained(&cmd->prot_sdb->table, 492 SCSI_INLINE_PROT_SG_CNT); 493 } 494 EXPORT_SYMBOL_GPL(scsi_free_sgtables); 495 496 static void scsi_mq_uninit_cmd(struct scsi_cmnd *cmd) 497 { 498 scsi_free_sgtables(cmd); 499 scsi_uninit_cmd(cmd); 500 } 501 502 static void scsi_run_queue_async(struct scsi_device *sdev) 503 { 504 if (scsi_target(sdev)->single_lun || 505 !list_empty(&sdev->host->starved_list)) { 506 kblockd_schedule_work(&sdev->requeue_work); 507 } else { 508 /* 509 * smp_mb() present in sbitmap_queue_clear() or implied in 510 * .end_io is for ordering writing .device_busy in 511 * scsi_device_unbusy() and reading sdev->restarts. 512 */ 513 int old = atomic_read(&sdev->restarts); 514 515 /* 516 * ->restarts has to be kept as non-zero if new budget 517 * contention occurs. 518 * 519 * No need to run queue when either another re-run 520 * queue wins in updating ->restarts or a new budget 521 * contention occurs. 522 */ 523 if (old && atomic_cmpxchg(&sdev->restarts, old, 0) == old) 524 blk_mq_run_hw_queues(sdev->request_queue, true); 525 } 526 } 527 528 /* Returns false when no more bytes to process, true if there are more */ 529 static bool scsi_end_request(struct request *req, blk_status_t error, 530 unsigned int bytes) 531 { 532 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 533 struct scsi_device *sdev = cmd->device; 534 struct request_queue *q = sdev->request_queue; 535 536 if (blk_update_request(req, error, bytes)) 537 return true; 538 539 // XXX: 540 if (blk_queue_add_random(q)) 541 add_disk_randomness(req->q->disk); 542 543 if (!blk_rq_is_passthrough(req)) { 544 WARN_ON_ONCE(!(cmd->flags & SCMD_INITIALIZED)); 545 cmd->flags &= ~SCMD_INITIALIZED; 546 } 547 548 /* 549 * Calling rcu_barrier() is not necessary here because the 550 * SCSI error handler guarantees that the function called by 551 * call_rcu() has been called before scsi_end_request() is 552 * called. 553 */ 554 destroy_rcu_head(&cmd->rcu); 555 556 /* 557 * In the MQ case the command gets freed by __blk_mq_end_request, 558 * so we have to do all cleanup that depends on it earlier. 559 * 560 * We also can't kick the queues from irq context, so we 561 * will have to defer it to a workqueue. 562 */ 563 scsi_mq_uninit_cmd(cmd); 564 565 /* 566 * queue is still alive, so grab the ref for preventing it 567 * from being cleaned up during running queue. 568 */ 569 percpu_ref_get(&q->q_usage_counter); 570 571 __blk_mq_end_request(req, error); 572 573 scsi_run_queue_async(sdev); 574 575 percpu_ref_put(&q->q_usage_counter); 576 return false; 577 } 578 579 static inline u8 get_scsi_ml_byte(int result) 580 { 581 return (result >> 8) & 0xff; 582 } 583 584 /** 585 * scsi_result_to_blk_status - translate a SCSI result code into blk_status_t 586 * @result: scsi error code 587 * 588 * Translate a SCSI result code into a blk_status_t value. 589 */ 590 static blk_status_t scsi_result_to_blk_status(int result) 591 { 592 /* 593 * Check the scsi-ml byte first in case we converted a host or status 594 * byte. 595 */ 596 switch (get_scsi_ml_byte(result)) { 597 case SCSIML_STAT_OK: 598 break; 599 case SCSIML_STAT_RESV_CONFLICT: 600 return BLK_STS_NEXUS; 601 case SCSIML_STAT_NOSPC: 602 return BLK_STS_NOSPC; 603 case SCSIML_STAT_MED_ERROR: 604 return BLK_STS_MEDIUM; 605 case SCSIML_STAT_TGT_FAILURE: 606 return BLK_STS_TARGET; 607 } 608 609 switch (host_byte(result)) { 610 case DID_OK: 611 if (scsi_status_is_good(result)) 612 return BLK_STS_OK; 613 return BLK_STS_IOERR; 614 case DID_TRANSPORT_FAILFAST: 615 case DID_TRANSPORT_MARGINAL: 616 return BLK_STS_TRANSPORT; 617 default: 618 return BLK_STS_IOERR; 619 } 620 } 621 622 /** 623 * scsi_rq_err_bytes - determine number of bytes till the next failure boundary 624 * @rq: request to examine 625 * 626 * Description: 627 * A request could be merge of IOs which require different failure 628 * handling. This function determines the number of bytes which 629 * can be failed from the beginning of the request without 630 * crossing into area which need to be retried further. 631 * 632 * Return: 633 * The number of bytes to fail. 634 */ 635 static unsigned int scsi_rq_err_bytes(const struct request *rq) 636 { 637 blk_opf_t ff = rq->cmd_flags & REQ_FAILFAST_MASK; 638 unsigned int bytes = 0; 639 struct bio *bio; 640 641 if (!(rq->rq_flags & RQF_MIXED_MERGE)) 642 return blk_rq_bytes(rq); 643 644 /* 645 * Currently the only 'mixing' which can happen is between 646 * different fastfail types. We can safely fail portions 647 * which have all the failfast bits that the first one has - 648 * the ones which are at least as eager to fail as the first 649 * one. 650 */ 651 for (bio = rq->bio; bio; bio = bio->bi_next) { 652 if ((bio->bi_opf & ff) != ff) 653 break; 654 bytes += bio->bi_iter.bi_size; 655 } 656 657 /* this could lead to infinite loop */ 658 BUG_ON(blk_rq_bytes(rq) && !bytes); 659 return bytes; 660 } 661 662 /* Helper for scsi_io_completion() when "reprep" action required. */ 663 static void scsi_io_completion_reprep(struct scsi_cmnd *cmd, 664 struct request_queue *q) 665 { 666 /* A new command will be prepared and issued. */ 667 scsi_mq_requeue_cmd(cmd); 668 } 669 670 static bool scsi_cmd_runtime_exceeced(struct scsi_cmnd *cmd) 671 { 672 struct request *req = scsi_cmd_to_rq(cmd); 673 unsigned long wait_for; 674 675 if (cmd->allowed == SCSI_CMD_RETRIES_NO_LIMIT) 676 return false; 677 678 wait_for = (cmd->allowed + 1) * req->timeout; 679 if (time_before(cmd->jiffies_at_alloc + wait_for, jiffies)) { 680 scmd_printk(KERN_ERR, cmd, "timing out command, waited %lus\n", 681 wait_for/HZ); 682 return true; 683 } 684 return false; 685 } 686 687 /* Helper for scsi_io_completion() when special action required. */ 688 static void scsi_io_completion_action(struct scsi_cmnd *cmd, int result) 689 { 690 struct request_queue *q = cmd->device->request_queue; 691 struct request *req = scsi_cmd_to_rq(cmd); 692 int level = 0; 693 enum {ACTION_FAIL, ACTION_REPREP, ACTION_RETRY, 694 ACTION_DELAYED_RETRY} action; 695 struct scsi_sense_hdr sshdr; 696 bool sense_valid; 697 bool sense_current = true; /* false implies "deferred sense" */ 698 blk_status_t blk_stat; 699 700 sense_valid = scsi_command_normalize_sense(cmd, &sshdr); 701 if (sense_valid) 702 sense_current = !scsi_sense_is_deferred(&sshdr); 703 704 blk_stat = scsi_result_to_blk_status(result); 705 706 if (host_byte(result) == DID_RESET) { 707 /* Third party bus reset or reset for error recovery 708 * reasons. Just retry the command and see what 709 * happens. 710 */ 711 action = ACTION_RETRY; 712 } else if (sense_valid && sense_current) { 713 switch (sshdr.sense_key) { 714 case UNIT_ATTENTION: 715 if (cmd->device->removable) { 716 /* Detected disc change. Set a bit 717 * and quietly refuse further access. 718 */ 719 cmd->device->changed = 1; 720 action = ACTION_FAIL; 721 } else { 722 /* Must have been a power glitch, or a 723 * bus reset. Could not have been a 724 * media change, so we just retry the 725 * command and see what happens. 726 */ 727 action = ACTION_RETRY; 728 } 729 break; 730 case ILLEGAL_REQUEST: 731 /* If we had an ILLEGAL REQUEST returned, then 732 * we may have performed an unsupported 733 * command. The only thing this should be 734 * would be a ten byte read where only a six 735 * byte read was supported. Also, on a system 736 * where READ CAPACITY failed, we may have 737 * read past the end of the disk. 738 */ 739 if ((cmd->device->use_10_for_rw && 740 sshdr.asc == 0x20 && sshdr.ascq == 0x00) && 741 (cmd->cmnd[0] == READ_10 || 742 cmd->cmnd[0] == WRITE_10)) { 743 /* This will issue a new 6-byte command. */ 744 cmd->device->use_10_for_rw = 0; 745 action = ACTION_REPREP; 746 } else if (sshdr.asc == 0x10) /* DIX */ { 747 action = ACTION_FAIL; 748 blk_stat = BLK_STS_PROTECTION; 749 /* INVALID COMMAND OPCODE or INVALID FIELD IN CDB */ 750 } else if (sshdr.asc == 0x20 || sshdr.asc == 0x24) { 751 action = ACTION_FAIL; 752 blk_stat = BLK_STS_TARGET; 753 } else 754 action = ACTION_FAIL; 755 break; 756 case ABORTED_COMMAND: 757 action = ACTION_FAIL; 758 if (sshdr.asc == 0x10) /* DIF */ 759 blk_stat = BLK_STS_PROTECTION; 760 break; 761 case NOT_READY: 762 /* If the device is in the process of becoming 763 * ready, or has a temporary blockage, retry. 764 */ 765 if (sshdr.asc == 0x04) { 766 switch (sshdr.ascq) { 767 case 0x01: /* becoming ready */ 768 case 0x04: /* format in progress */ 769 case 0x05: /* rebuild in progress */ 770 case 0x06: /* recalculation in progress */ 771 case 0x07: /* operation in progress */ 772 case 0x08: /* Long write in progress */ 773 case 0x09: /* self test in progress */ 774 case 0x11: /* notify (enable spinup) required */ 775 case 0x14: /* space allocation in progress */ 776 case 0x1a: /* start stop unit in progress */ 777 case 0x1b: /* sanitize in progress */ 778 case 0x1d: /* configuration in progress */ 779 case 0x24: /* depopulation in progress */ 780 action = ACTION_DELAYED_RETRY; 781 break; 782 case 0x0a: /* ALUA state transition */ 783 blk_stat = BLK_STS_TRANSPORT; 784 fallthrough; 785 default: 786 action = ACTION_FAIL; 787 break; 788 } 789 } else 790 action = ACTION_FAIL; 791 break; 792 case VOLUME_OVERFLOW: 793 /* See SSC3rXX or current. */ 794 action = ACTION_FAIL; 795 break; 796 case DATA_PROTECT: 797 action = ACTION_FAIL; 798 if ((sshdr.asc == 0x0C && sshdr.ascq == 0x12) || 799 (sshdr.asc == 0x55 && 800 (sshdr.ascq == 0x0E || sshdr.ascq == 0x0F))) { 801 /* Insufficient zone resources */ 802 blk_stat = BLK_STS_ZONE_OPEN_RESOURCE; 803 } 804 break; 805 default: 806 action = ACTION_FAIL; 807 break; 808 } 809 } else 810 action = ACTION_FAIL; 811 812 if (action != ACTION_FAIL && scsi_cmd_runtime_exceeced(cmd)) 813 action = ACTION_FAIL; 814 815 switch (action) { 816 case ACTION_FAIL: 817 /* Give up and fail the remainder of the request */ 818 if (!(req->rq_flags & RQF_QUIET)) { 819 static DEFINE_RATELIMIT_STATE(_rs, 820 DEFAULT_RATELIMIT_INTERVAL, 821 DEFAULT_RATELIMIT_BURST); 822 823 if (unlikely(scsi_logging_level)) 824 level = 825 SCSI_LOG_LEVEL(SCSI_LOG_MLCOMPLETE_SHIFT, 826 SCSI_LOG_MLCOMPLETE_BITS); 827 828 /* 829 * if logging is enabled the failure will be printed 830 * in scsi_log_completion(), so avoid duplicate messages 831 */ 832 if (!level && __ratelimit(&_rs)) { 833 scsi_print_result(cmd, NULL, FAILED); 834 if (sense_valid) 835 scsi_print_sense(cmd); 836 scsi_print_command(cmd); 837 } 838 } 839 if (!scsi_end_request(req, blk_stat, scsi_rq_err_bytes(req))) 840 return; 841 fallthrough; 842 case ACTION_REPREP: 843 scsi_io_completion_reprep(cmd, q); 844 break; 845 case ACTION_RETRY: 846 /* Retry the same command immediately */ 847 __scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY, false); 848 break; 849 case ACTION_DELAYED_RETRY: 850 /* Retry the same command after a delay */ 851 __scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY, false); 852 break; 853 } 854 } 855 856 /* 857 * Helper for scsi_io_completion() when cmd->result is non-zero. Returns a 858 * new result that may suppress further error checking. Also modifies 859 * *blk_statp in some cases. 860 */ 861 static int scsi_io_completion_nz_result(struct scsi_cmnd *cmd, int result, 862 blk_status_t *blk_statp) 863 { 864 bool sense_valid; 865 bool sense_current = true; /* false implies "deferred sense" */ 866 struct request *req = scsi_cmd_to_rq(cmd); 867 struct scsi_sense_hdr sshdr; 868 869 sense_valid = scsi_command_normalize_sense(cmd, &sshdr); 870 if (sense_valid) 871 sense_current = !scsi_sense_is_deferred(&sshdr); 872 873 if (blk_rq_is_passthrough(req)) { 874 if (sense_valid) { 875 /* 876 * SG_IO wants current and deferred errors 877 */ 878 cmd->sense_len = min(8 + cmd->sense_buffer[7], 879 SCSI_SENSE_BUFFERSIZE); 880 } 881 if (sense_current) 882 *blk_statp = scsi_result_to_blk_status(result); 883 } else if (blk_rq_bytes(req) == 0 && sense_current) { 884 /* 885 * Flush commands do not transfers any data, and thus cannot use 886 * good_bytes != blk_rq_bytes(req) as the signal for an error. 887 * This sets *blk_statp explicitly for the problem case. 888 */ 889 *blk_statp = scsi_result_to_blk_status(result); 890 } 891 /* 892 * Recovered errors need reporting, but they're always treated as 893 * success, so fiddle the result code here. For passthrough requests 894 * we already took a copy of the original into sreq->result which 895 * is what gets returned to the user 896 */ 897 if (sense_valid && (sshdr.sense_key == RECOVERED_ERROR)) { 898 bool do_print = true; 899 /* 900 * if ATA PASS-THROUGH INFORMATION AVAILABLE [0x0, 0x1d] 901 * skip print since caller wants ATA registers. Only occurs 902 * on SCSI ATA PASS_THROUGH commands when CK_COND=1 903 */ 904 if ((sshdr.asc == 0x0) && (sshdr.ascq == 0x1d)) 905 do_print = false; 906 else if (req->rq_flags & RQF_QUIET) 907 do_print = false; 908 if (do_print) 909 scsi_print_sense(cmd); 910 result = 0; 911 /* for passthrough, *blk_statp may be set */ 912 *blk_statp = BLK_STS_OK; 913 } 914 /* 915 * Another corner case: the SCSI status byte is non-zero but 'good'. 916 * Example: PRE-FETCH command returns SAM_STAT_CONDITION_MET when 917 * it is able to fit nominated LBs in its cache (and SAM_STAT_GOOD 918 * if it can't fit). Treat SAM_STAT_CONDITION_MET and the related 919 * intermediate statuses (both obsolete in SAM-4) as good. 920 */ 921 if ((result & 0xff) && scsi_status_is_good(result)) { 922 result = 0; 923 *blk_statp = BLK_STS_OK; 924 } 925 return result; 926 } 927 928 /** 929 * scsi_io_completion - Completion processing for SCSI commands. 930 * @cmd: command that is finished. 931 * @good_bytes: number of processed bytes. 932 * 933 * We will finish off the specified number of sectors. If we are done, the 934 * command block will be released and the queue function will be goosed. If we 935 * are not done then we have to figure out what to do next: 936 * 937 * a) We can call scsi_io_completion_reprep(). The request will be 938 * unprepared and put back on the queue. Then a new command will 939 * be created for it. This should be used if we made forward 940 * progress, or if we want to switch from READ(10) to READ(6) for 941 * example. 942 * 943 * b) We can call scsi_io_completion_action(). The request will be 944 * put back on the queue and retried using the same command as 945 * before, possibly after a delay. 946 * 947 * c) We can call scsi_end_request() with blk_stat other than 948 * BLK_STS_OK, to fail the remainder of the request. 949 */ 950 void scsi_io_completion(struct scsi_cmnd *cmd, unsigned int good_bytes) 951 { 952 int result = cmd->result; 953 struct request_queue *q = cmd->device->request_queue; 954 struct request *req = scsi_cmd_to_rq(cmd); 955 blk_status_t blk_stat = BLK_STS_OK; 956 957 if (unlikely(result)) /* a nz result may or may not be an error */ 958 result = scsi_io_completion_nz_result(cmd, result, &blk_stat); 959 960 /* 961 * Next deal with any sectors which we were able to correctly 962 * handle. 963 */ 964 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, cmd, 965 "%u sectors total, %d bytes done.\n", 966 blk_rq_sectors(req), good_bytes)); 967 968 /* 969 * Failed, zero length commands always need to drop down 970 * to retry code. Fast path should return in this block. 971 */ 972 if (likely(blk_rq_bytes(req) > 0 || blk_stat == BLK_STS_OK)) { 973 if (likely(!scsi_end_request(req, blk_stat, good_bytes))) 974 return; /* no bytes remaining */ 975 } 976 977 /* Kill remainder if no retries. */ 978 if (unlikely(blk_stat && scsi_noretry_cmd(cmd))) { 979 if (scsi_end_request(req, blk_stat, blk_rq_bytes(req))) 980 WARN_ONCE(true, 981 "Bytes remaining after failed, no-retry command"); 982 return; 983 } 984 985 /* 986 * If there had been no error, but we have leftover bytes in the 987 * request just queue the command up again. 988 */ 989 if (likely(result == 0)) 990 scsi_io_completion_reprep(cmd, q); 991 else 992 scsi_io_completion_action(cmd, result); 993 } 994 995 static inline bool scsi_cmd_needs_dma_drain(struct scsi_device *sdev, 996 struct request *rq) 997 { 998 return sdev->dma_drain_len && blk_rq_is_passthrough(rq) && 999 !op_is_write(req_op(rq)) && 1000 sdev->host->hostt->dma_need_drain(rq); 1001 } 1002 1003 /** 1004 * scsi_alloc_sgtables - Allocate and initialize data and integrity scatterlists 1005 * @cmd: SCSI command data structure to initialize. 1006 * 1007 * Initializes @cmd->sdb and also @cmd->prot_sdb if data integrity is enabled 1008 * for @cmd. 1009 * 1010 * Returns: 1011 * * BLK_STS_OK - on success 1012 * * BLK_STS_RESOURCE - if the failure is retryable 1013 * * BLK_STS_IOERR - if the failure is fatal 1014 */ 1015 blk_status_t scsi_alloc_sgtables(struct scsi_cmnd *cmd) 1016 { 1017 struct scsi_device *sdev = cmd->device; 1018 struct request *rq = scsi_cmd_to_rq(cmd); 1019 unsigned short nr_segs = blk_rq_nr_phys_segments(rq); 1020 struct scatterlist *last_sg = NULL; 1021 blk_status_t ret; 1022 bool need_drain = scsi_cmd_needs_dma_drain(sdev, rq); 1023 int count; 1024 1025 if (WARN_ON_ONCE(!nr_segs)) 1026 return BLK_STS_IOERR; 1027 1028 /* 1029 * Make sure there is space for the drain. The driver must adjust 1030 * max_hw_segments to be prepared for this. 1031 */ 1032 if (need_drain) 1033 nr_segs++; 1034 1035 /* 1036 * If sg table allocation fails, requeue request later. 1037 */ 1038 if (unlikely(sg_alloc_table_chained(&cmd->sdb.table, nr_segs, 1039 cmd->sdb.table.sgl, SCSI_INLINE_SG_CNT))) 1040 return BLK_STS_RESOURCE; 1041 1042 /* 1043 * Next, walk the list, and fill in the addresses and sizes of 1044 * each segment. 1045 */ 1046 count = __blk_rq_map_sg(rq->q, rq, cmd->sdb.table.sgl, &last_sg); 1047 1048 if (blk_rq_bytes(rq) & rq->q->dma_pad_mask) { 1049 unsigned int pad_len = 1050 (rq->q->dma_pad_mask & ~blk_rq_bytes(rq)) + 1; 1051 1052 last_sg->length += pad_len; 1053 cmd->extra_len += pad_len; 1054 } 1055 1056 if (need_drain) { 1057 sg_unmark_end(last_sg); 1058 last_sg = sg_next(last_sg); 1059 sg_set_buf(last_sg, sdev->dma_drain_buf, sdev->dma_drain_len); 1060 sg_mark_end(last_sg); 1061 1062 cmd->extra_len += sdev->dma_drain_len; 1063 count++; 1064 } 1065 1066 BUG_ON(count > cmd->sdb.table.nents); 1067 cmd->sdb.table.nents = count; 1068 cmd->sdb.length = blk_rq_payload_bytes(rq); 1069 1070 if (blk_integrity_rq(rq)) { 1071 struct scsi_data_buffer *prot_sdb = cmd->prot_sdb; 1072 int ivecs; 1073 1074 if (WARN_ON_ONCE(!prot_sdb)) { 1075 /* 1076 * This can happen if someone (e.g. multipath) 1077 * queues a command to a device on an adapter 1078 * that does not support DIX. 1079 */ 1080 ret = BLK_STS_IOERR; 1081 goto out_free_sgtables; 1082 } 1083 1084 ivecs = blk_rq_count_integrity_sg(rq->q, rq->bio); 1085 1086 if (sg_alloc_table_chained(&prot_sdb->table, ivecs, 1087 prot_sdb->table.sgl, 1088 SCSI_INLINE_PROT_SG_CNT)) { 1089 ret = BLK_STS_RESOURCE; 1090 goto out_free_sgtables; 1091 } 1092 1093 count = blk_rq_map_integrity_sg(rq->q, rq->bio, 1094 prot_sdb->table.sgl); 1095 BUG_ON(count > ivecs); 1096 BUG_ON(count > queue_max_integrity_segments(rq->q)); 1097 1098 cmd->prot_sdb = prot_sdb; 1099 cmd->prot_sdb->table.nents = count; 1100 } 1101 1102 return BLK_STS_OK; 1103 out_free_sgtables: 1104 scsi_free_sgtables(cmd); 1105 return ret; 1106 } 1107 EXPORT_SYMBOL(scsi_alloc_sgtables); 1108 1109 /** 1110 * scsi_initialize_rq - initialize struct scsi_cmnd partially 1111 * @rq: Request associated with the SCSI command to be initialized. 1112 * 1113 * This function initializes the members of struct scsi_cmnd that must be 1114 * initialized before request processing starts and that won't be 1115 * reinitialized if a SCSI command is requeued. 1116 */ 1117 static void scsi_initialize_rq(struct request *rq) 1118 { 1119 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1120 1121 memset(cmd->cmnd, 0, sizeof(cmd->cmnd)); 1122 cmd->cmd_len = MAX_COMMAND_SIZE; 1123 cmd->sense_len = 0; 1124 init_rcu_head(&cmd->rcu); 1125 cmd->jiffies_at_alloc = jiffies; 1126 cmd->retries = 0; 1127 } 1128 1129 struct request *scsi_alloc_request(struct request_queue *q, blk_opf_t opf, 1130 blk_mq_req_flags_t flags) 1131 { 1132 struct request *rq; 1133 1134 rq = blk_mq_alloc_request(q, opf, flags); 1135 if (!IS_ERR(rq)) 1136 scsi_initialize_rq(rq); 1137 return rq; 1138 } 1139 EXPORT_SYMBOL_GPL(scsi_alloc_request); 1140 1141 /* 1142 * Only called when the request isn't completed by SCSI, and not freed by 1143 * SCSI 1144 */ 1145 static void scsi_cleanup_rq(struct request *rq) 1146 { 1147 if (rq->rq_flags & RQF_DONTPREP) { 1148 scsi_mq_uninit_cmd(blk_mq_rq_to_pdu(rq)); 1149 rq->rq_flags &= ~RQF_DONTPREP; 1150 } 1151 } 1152 1153 /* Called before a request is prepared. See also scsi_mq_prep_fn(). */ 1154 void scsi_init_command(struct scsi_device *dev, struct scsi_cmnd *cmd) 1155 { 1156 struct request *rq = scsi_cmd_to_rq(cmd); 1157 1158 if (!blk_rq_is_passthrough(rq) && !(cmd->flags & SCMD_INITIALIZED)) { 1159 cmd->flags |= SCMD_INITIALIZED; 1160 scsi_initialize_rq(rq); 1161 } 1162 1163 cmd->device = dev; 1164 INIT_LIST_HEAD(&cmd->eh_entry); 1165 INIT_DELAYED_WORK(&cmd->abort_work, scmd_eh_abort_handler); 1166 } 1167 1168 static blk_status_t scsi_setup_scsi_cmnd(struct scsi_device *sdev, 1169 struct request *req) 1170 { 1171 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1172 1173 /* 1174 * Passthrough requests may transfer data, in which case they must 1175 * a bio attached to them. Or they might contain a SCSI command 1176 * that does not transfer data, in which case they may optionally 1177 * submit a request without an attached bio. 1178 */ 1179 if (req->bio) { 1180 blk_status_t ret = scsi_alloc_sgtables(cmd); 1181 if (unlikely(ret != BLK_STS_OK)) 1182 return ret; 1183 } else { 1184 BUG_ON(blk_rq_bytes(req)); 1185 1186 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1187 } 1188 1189 cmd->transfersize = blk_rq_bytes(req); 1190 return BLK_STS_OK; 1191 } 1192 1193 static blk_status_t 1194 scsi_device_state_check(struct scsi_device *sdev, struct request *req) 1195 { 1196 switch (sdev->sdev_state) { 1197 case SDEV_CREATED: 1198 return BLK_STS_OK; 1199 case SDEV_OFFLINE: 1200 case SDEV_TRANSPORT_OFFLINE: 1201 /* 1202 * If the device is offline we refuse to process any 1203 * commands. The device must be brought online 1204 * before trying any recovery commands. 1205 */ 1206 if (!sdev->offline_already) { 1207 sdev->offline_already = true; 1208 sdev_printk(KERN_ERR, sdev, 1209 "rejecting I/O to offline device\n"); 1210 } 1211 return BLK_STS_IOERR; 1212 case SDEV_DEL: 1213 /* 1214 * If the device is fully deleted, we refuse to 1215 * process any commands as well. 1216 */ 1217 sdev_printk(KERN_ERR, sdev, 1218 "rejecting I/O to dead device\n"); 1219 return BLK_STS_IOERR; 1220 case SDEV_BLOCK: 1221 case SDEV_CREATED_BLOCK: 1222 return BLK_STS_RESOURCE; 1223 case SDEV_QUIESCE: 1224 /* 1225 * If the device is blocked we only accept power management 1226 * commands. 1227 */ 1228 if (req && WARN_ON_ONCE(!(req->rq_flags & RQF_PM))) 1229 return BLK_STS_RESOURCE; 1230 return BLK_STS_OK; 1231 default: 1232 /* 1233 * For any other not fully online state we only allow 1234 * power management commands. 1235 */ 1236 if (req && !(req->rq_flags & RQF_PM)) 1237 return BLK_STS_OFFLINE; 1238 return BLK_STS_OK; 1239 } 1240 } 1241 1242 /* 1243 * scsi_dev_queue_ready: if we can send requests to sdev, assign one token 1244 * and return the token else return -1. 1245 */ 1246 static inline int scsi_dev_queue_ready(struct request_queue *q, 1247 struct scsi_device *sdev) 1248 { 1249 int token; 1250 1251 token = sbitmap_get(&sdev->budget_map); 1252 if (atomic_read(&sdev->device_blocked)) { 1253 if (token < 0) 1254 goto out; 1255 1256 if (scsi_device_busy(sdev) > 1) 1257 goto out_dec; 1258 1259 /* 1260 * unblock after device_blocked iterates to zero 1261 */ 1262 if (atomic_dec_return(&sdev->device_blocked) > 0) 1263 goto out_dec; 1264 SCSI_LOG_MLQUEUE(3, sdev_printk(KERN_INFO, sdev, 1265 "unblocking device at zero depth\n")); 1266 } 1267 1268 return token; 1269 out_dec: 1270 if (token >= 0) 1271 sbitmap_put(&sdev->budget_map, token); 1272 out: 1273 return -1; 1274 } 1275 1276 /* 1277 * scsi_target_queue_ready: checks if there we can send commands to target 1278 * @sdev: scsi device on starget to check. 1279 */ 1280 static inline int scsi_target_queue_ready(struct Scsi_Host *shost, 1281 struct scsi_device *sdev) 1282 { 1283 struct scsi_target *starget = scsi_target(sdev); 1284 unsigned int busy; 1285 1286 if (starget->single_lun) { 1287 spin_lock_irq(shost->host_lock); 1288 if (starget->starget_sdev_user && 1289 starget->starget_sdev_user != sdev) { 1290 spin_unlock_irq(shost->host_lock); 1291 return 0; 1292 } 1293 starget->starget_sdev_user = sdev; 1294 spin_unlock_irq(shost->host_lock); 1295 } 1296 1297 if (starget->can_queue <= 0) 1298 return 1; 1299 1300 busy = atomic_inc_return(&starget->target_busy) - 1; 1301 if (atomic_read(&starget->target_blocked) > 0) { 1302 if (busy) 1303 goto starved; 1304 1305 /* 1306 * unblock after target_blocked iterates to zero 1307 */ 1308 if (atomic_dec_return(&starget->target_blocked) > 0) 1309 goto out_dec; 1310 1311 SCSI_LOG_MLQUEUE(3, starget_printk(KERN_INFO, starget, 1312 "unblocking target at zero depth\n")); 1313 } 1314 1315 if (busy >= starget->can_queue) 1316 goto starved; 1317 1318 return 1; 1319 1320 starved: 1321 spin_lock_irq(shost->host_lock); 1322 list_move_tail(&sdev->starved_entry, &shost->starved_list); 1323 spin_unlock_irq(shost->host_lock); 1324 out_dec: 1325 if (starget->can_queue > 0) 1326 atomic_dec(&starget->target_busy); 1327 return 0; 1328 } 1329 1330 /* 1331 * scsi_host_queue_ready: if we can send requests to shost, return 1 else 1332 * return 0. We must end up running the queue again whenever 0 is 1333 * returned, else IO can hang. 1334 */ 1335 static inline int scsi_host_queue_ready(struct request_queue *q, 1336 struct Scsi_Host *shost, 1337 struct scsi_device *sdev, 1338 struct scsi_cmnd *cmd) 1339 { 1340 if (scsi_host_in_recovery(shost)) 1341 return 0; 1342 1343 if (atomic_read(&shost->host_blocked) > 0) { 1344 if (scsi_host_busy(shost) > 0) 1345 goto starved; 1346 1347 /* 1348 * unblock after host_blocked iterates to zero 1349 */ 1350 if (atomic_dec_return(&shost->host_blocked) > 0) 1351 goto out_dec; 1352 1353 SCSI_LOG_MLQUEUE(3, 1354 shost_printk(KERN_INFO, shost, 1355 "unblocking host at zero depth\n")); 1356 } 1357 1358 if (shost->host_self_blocked) 1359 goto starved; 1360 1361 /* We're OK to process the command, so we can't be starved */ 1362 if (!list_empty(&sdev->starved_entry)) { 1363 spin_lock_irq(shost->host_lock); 1364 if (!list_empty(&sdev->starved_entry)) 1365 list_del_init(&sdev->starved_entry); 1366 spin_unlock_irq(shost->host_lock); 1367 } 1368 1369 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state); 1370 1371 return 1; 1372 1373 starved: 1374 spin_lock_irq(shost->host_lock); 1375 if (list_empty(&sdev->starved_entry)) 1376 list_add_tail(&sdev->starved_entry, &shost->starved_list); 1377 spin_unlock_irq(shost->host_lock); 1378 out_dec: 1379 scsi_dec_host_busy(shost, cmd); 1380 return 0; 1381 } 1382 1383 /* 1384 * Busy state exporting function for request stacking drivers. 1385 * 1386 * For efficiency, no lock is taken to check the busy state of 1387 * shost/starget/sdev, since the returned value is not guaranteed and 1388 * may be changed after request stacking drivers call the function, 1389 * regardless of taking lock or not. 1390 * 1391 * When scsi can't dispatch I/Os anymore and needs to kill I/Os scsi 1392 * needs to return 'not busy'. Otherwise, request stacking drivers 1393 * may hold requests forever. 1394 */ 1395 static bool scsi_mq_lld_busy(struct request_queue *q) 1396 { 1397 struct scsi_device *sdev = q->queuedata; 1398 struct Scsi_Host *shost; 1399 1400 if (blk_queue_dying(q)) 1401 return false; 1402 1403 shost = sdev->host; 1404 1405 /* 1406 * Ignore host/starget busy state. 1407 * Since block layer does not have a concept of fairness across 1408 * multiple queues, congestion of host/starget needs to be handled 1409 * in SCSI layer. 1410 */ 1411 if (scsi_host_in_recovery(shost) || scsi_device_is_busy(sdev)) 1412 return true; 1413 1414 return false; 1415 } 1416 1417 /* 1418 * Block layer request completion callback. May be called from interrupt 1419 * context. 1420 */ 1421 static void scsi_complete(struct request *rq) 1422 { 1423 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1424 enum scsi_disposition disposition; 1425 1426 INIT_LIST_HEAD(&cmd->eh_entry); 1427 1428 atomic_inc(&cmd->device->iodone_cnt); 1429 if (cmd->result) 1430 atomic_inc(&cmd->device->ioerr_cnt); 1431 1432 disposition = scsi_decide_disposition(cmd); 1433 if (disposition != SUCCESS && scsi_cmd_runtime_exceeced(cmd)) 1434 disposition = SUCCESS; 1435 1436 scsi_log_completion(cmd, disposition); 1437 1438 switch (disposition) { 1439 case SUCCESS: 1440 scsi_finish_command(cmd); 1441 break; 1442 case NEEDS_RETRY: 1443 scsi_queue_insert(cmd, SCSI_MLQUEUE_EH_RETRY); 1444 break; 1445 case ADD_TO_MLQUEUE: 1446 scsi_queue_insert(cmd, SCSI_MLQUEUE_DEVICE_BUSY); 1447 break; 1448 default: 1449 scsi_eh_scmd_add(cmd); 1450 break; 1451 } 1452 } 1453 1454 /** 1455 * scsi_dispatch_cmd - Dispatch a command to the low-level driver. 1456 * @cmd: command block we are dispatching. 1457 * 1458 * Return: nonzero return request was rejected and device's queue needs to be 1459 * plugged. 1460 */ 1461 static int scsi_dispatch_cmd(struct scsi_cmnd *cmd) 1462 { 1463 struct Scsi_Host *host = cmd->device->host; 1464 int rtn = 0; 1465 1466 atomic_inc(&cmd->device->iorequest_cnt); 1467 1468 /* check if the device is still usable */ 1469 if (unlikely(cmd->device->sdev_state == SDEV_DEL)) { 1470 /* in SDEV_DEL we error all commands. DID_NO_CONNECT 1471 * returns an immediate error upwards, and signals 1472 * that the device is no longer present */ 1473 cmd->result = DID_NO_CONNECT << 16; 1474 goto done; 1475 } 1476 1477 /* Check to see if the scsi lld made this device blocked. */ 1478 if (unlikely(scsi_device_blocked(cmd->device))) { 1479 /* 1480 * in blocked state, the command is just put back on 1481 * the device queue. The suspend state has already 1482 * blocked the queue so future requests should not 1483 * occur until the device transitions out of the 1484 * suspend state. 1485 */ 1486 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd, 1487 "queuecommand : device blocked\n")); 1488 return SCSI_MLQUEUE_DEVICE_BUSY; 1489 } 1490 1491 /* Store the LUN value in cmnd, if needed. */ 1492 if (cmd->device->lun_in_cdb) 1493 cmd->cmnd[1] = (cmd->cmnd[1] & 0x1f) | 1494 (cmd->device->lun << 5 & 0xe0); 1495 1496 scsi_log_send(cmd); 1497 1498 /* 1499 * Before we queue this command, check if the command 1500 * length exceeds what the host adapter can handle. 1501 */ 1502 if (cmd->cmd_len > cmd->device->host->max_cmd_len) { 1503 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd, 1504 "queuecommand : command too long. " 1505 "cdb_size=%d host->max_cmd_len=%d\n", 1506 cmd->cmd_len, cmd->device->host->max_cmd_len)); 1507 cmd->result = (DID_ABORT << 16); 1508 goto done; 1509 } 1510 1511 if (unlikely(host->shost_state == SHOST_DEL)) { 1512 cmd->result = (DID_NO_CONNECT << 16); 1513 goto done; 1514 1515 } 1516 1517 trace_scsi_dispatch_cmd_start(cmd); 1518 rtn = host->hostt->queuecommand(host, cmd); 1519 if (rtn) { 1520 trace_scsi_dispatch_cmd_error(cmd, rtn); 1521 if (rtn != SCSI_MLQUEUE_DEVICE_BUSY && 1522 rtn != SCSI_MLQUEUE_TARGET_BUSY) 1523 rtn = SCSI_MLQUEUE_HOST_BUSY; 1524 1525 SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd, 1526 "queuecommand : request rejected\n")); 1527 } 1528 1529 return rtn; 1530 done: 1531 scsi_done(cmd); 1532 return 0; 1533 } 1534 1535 /* Size in bytes of the sg-list stored in the scsi-mq command-private data. */ 1536 static unsigned int scsi_mq_inline_sgl_size(struct Scsi_Host *shost) 1537 { 1538 return min_t(unsigned int, shost->sg_tablesize, SCSI_INLINE_SG_CNT) * 1539 sizeof(struct scatterlist); 1540 } 1541 1542 static blk_status_t scsi_prepare_cmd(struct request *req) 1543 { 1544 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1545 struct scsi_device *sdev = req->q->queuedata; 1546 struct Scsi_Host *shost = sdev->host; 1547 bool in_flight = test_bit(SCMD_STATE_INFLIGHT, &cmd->state); 1548 struct scatterlist *sg; 1549 1550 scsi_init_command(sdev, cmd); 1551 1552 cmd->eh_eflags = 0; 1553 cmd->allowed = 0; 1554 cmd->prot_type = 0; 1555 cmd->prot_flags = 0; 1556 cmd->submitter = 0; 1557 memset(&cmd->sdb, 0, sizeof(cmd->sdb)); 1558 cmd->underflow = 0; 1559 cmd->transfersize = 0; 1560 cmd->host_scribble = NULL; 1561 cmd->result = 0; 1562 cmd->extra_len = 0; 1563 cmd->state = 0; 1564 if (in_flight) 1565 __set_bit(SCMD_STATE_INFLIGHT, &cmd->state); 1566 1567 /* 1568 * Only clear the driver-private command data if the LLD does not supply 1569 * a function to initialize that data. 1570 */ 1571 if (!shost->hostt->init_cmd_priv) 1572 memset(cmd + 1, 0, shost->hostt->cmd_size); 1573 1574 cmd->prot_op = SCSI_PROT_NORMAL; 1575 if (blk_rq_bytes(req)) 1576 cmd->sc_data_direction = rq_dma_dir(req); 1577 else 1578 cmd->sc_data_direction = DMA_NONE; 1579 1580 sg = (void *)cmd + sizeof(struct scsi_cmnd) + shost->hostt->cmd_size; 1581 cmd->sdb.table.sgl = sg; 1582 1583 if (scsi_host_get_prot(shost)) { 1584 memset(cmd->prot_sdb, 0, sizeof(struct scsi_data_buffer)); 1585 1586 cmd->prot_sdb->table.sgl = 1587 (struct scatterlist *)(cmd->prot_sdb + 1); 1588 } 1589 1590 /* 1591 * Special handling for passthrough commands, which don't go to the ULP 1592 * at all: 1593 */ 1594 if (blk_rq_is_passthrough(req)) 1595 return scsi_setup_scsi_cmnd(sdev, req); 1596 1597 if (sdev->handler && sdev->handler->prep_fn) { 1598 blk_status_t ret = sdev->handler->prep_fn(sdev, req); 1599 1600 if (ret != BLK_STS_OK) 1601 return ret; 1602 } 1603 1604 memset(cmd->cmnd, 0, sizeof(cmd->cmnd)); 1605 return scsi_cmd_to_driver(cmd)->init_command(cmd); 1606 } 1607 1608 static void scsi_done_internal(struct scsi_cmnd *cmd, bool complete_directly) 1609 { 1610 struct request *req = scsi_cmd_to_rq(cmd); 1611 1612 switch (cmd->submitter) { 1613 case SUBMITTED_BY_BLOCK_LAYER: 1614 break; 1615 case SUBMITTED_BY_SCSI_ERROR_HANDLER: 1616 return scsi_eh_done(cmd); 1617 case SUBMITTED_BY_SCSI_RESET_IOCTL: 1618 return; 1619 } 1620 1621 if (unlikely(blk_should_fake_timeout(scsi_cmd_to_rq(cmd)->q))) 1622 return; 1623 if (unlikely(test_and_set_bit(SCMD_STATE_COMPLETE, &cmd->state))) 1624 return; 1625 trace_scsi_dispatch_cmd_done(cmd); 1626 1627 if (complete_directly) 1628 blk_mq_complete_request_direct(req, scsi_complete); 1629 else 1630 blk_mq_complete_request(req); 1631 } 1632 1633 void scsi_done(struct scsi_cmnd *cmd) 1634 { 1635 scsi_done_internal(cmd, false); 1636 } 1637 EXPORT_SYMBOL(scsi_done); 1638 1639 void scsi_done_direct(struct scsi_cmnd *cmd) 1640 { 1641 scsi_done_internal(cmd, true); 1642 } 1643 EXPORT_SYMBOL(scsi_done_direct); 1644 1645 static void scsi_mq_put_budget(struct request_queue *q, int budget_token) 1646 { 1647 struct scsi_device *sdev = q->queuedata; 1648 1649 sbitmap_put(&sdev->budget_map, budget_token); 1650 } 1651 1652 /* 1653 * When to reinvoke queueing after a resource shortage. It's 3 msecs to 1654 * not change behaviour from the previous unplug mechanism, experimentation 1655 * may prove this needs changing. 1656 */ 1657 #define SCSI_QUEUE_DELAY 3 1658 1659 static int scsi_mq_get_budget(struct request_queue *q) 1660 { 1661 struct scsi_device *sdev = q->queuedata; 1662 int token = scsi_dev_queue_ready(q, sdev); 1663 1664 if (token >= 0) 1665 return token; 1666 1667 atomic_inc(&sdev->restarts); 1668 1669 /* 1670 * Orders atomic_inc(&sdev->restarts) and atomic_read(&sdev->device_busy). 1671 * .restarts must be incremented before .device_busy is read because the 1672 * code in scsi_run_queue_async() depends on the order of these operations. 1673 */ 1674 smp_mb__after_atomic(); 1675 1676 /* 1677 * If all in-flight requests originated from this LUN are completed 1678 * before reading .device_busy, sdev->device_busy will be observed as 1679 * zero, then blk_mq_delay_run_hw_queues() will dispatch this request 1680 * soon. Otherwise, completion of one of these requests will observe 1681 * the .restarts flag, and the request queue will be run for handling 1682 * this request, see scsi_end_request(). 1683 */ 1684 if (unlikely(scsi_device_busy(sdev) == 0 && 1685 !scsi_device_blocked(sdev))) 1686 blk_mq_delay_run_hw_queues(sdev->request_queue, SCSI_QUEUE_DELAY); 1687 return -1; 1688 } 1689 1690 static void scsi_mq_set_rq_budget_token(struct request *req, int token) 1691 { 1692 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1693 1694 cmd->budget_token = token; 1695 } 1696 1697 static int scsi_mq_get_rq_budget_token(struct request *req) 1698 { 1699 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1700 1701 return cmd->budget_token; 1702 } 1703 1704 static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, 1705 const struct blk_mq_queue_data *bd) 1706 { 1707 struct request *req = bd->rq; 1708 struct request_queue *q = req->q; 1709 struct scsi_device *sdev = q->queuedata; 1710 struct Scsi_Host *shost = sdev->host; 1711 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(req); 1712 blk_status_t ret; 1713 int reason; 1714 1715 WARN_ON_ONCE(cmd->budget_token < 0); 1716 1717 /* 1718 * If the device is not in running state we will reject some or all 1719 * commands. 1720 */ 1721 if (unlikely(sdev->sdev_state != SDEV_RUNNING)) { 1722 ret = scsi_device_state_check(sdev, req); 1723 if (ret != BLK_STS_OK) 1724 goto out_put_budget; 1725 } 1726 1727 ret = BLK_STS_RESOURCE; 1728 if (!scsi_target_queue_ready(shost, sdev)) 1729 goto out_put_budget; 1730 if (!scsi_host_queue_ready(q, shost, sdev, cmd)) 1731 goto out_dec_target_busy; 1732 1733 if (!(req->rq_flags & RQF_DONTPREP)) { 1734 ret = scsi_prepare_cmd(req); 1735 if (ret != BLK_STS_OK) 1736 goto out_dec_host_busy; 1737 req->rq_flags |= RQF_DONTPREP; 1738 } else { 1739 clear_bit(SCMD_STATE_COMPLETE, &cmd->state); 1740 } 1741 1742 cmd->flags &= SCMD_PRESERVED_FLAGS; 1743 if (sdev->simple_tags) 1744 cmd->flags |= SCMD_TAGGED; 1745 if (bd->last) 1746 cmd->flags |= SCMD_LAST; 1747 1748 scsi_set_resid(cmd, 0); 1749 memset(cmd->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE); 1750 cmd->submitter = SUBMITTED_BY_BLOCK_LAYER; 1751 1752 blk_mq_start_request(req); 1753 reason = scsi_dispatch_cmd(cmd); 1754 if (reason) { 1755 scsi_set_blocked(cmd, reason); 1756 ret = BLK_STS_RESOURCE; 1757 goto out_dec_host_busy; 1758 } 1759 1760 return BLK_STS_OK; 1761 1762 out_dec_host_busy: 1763 scsi_dec_host_busy(shost, cmd); 1764 out_dec_target_busy: 1765 if (scsi_target(sdev)->can_queue > 0) 1766 atomic_dec(&scsi_target(sdev)->target_busy); 1767 out_put_budget: 1768 scsi_mq_put_budget(q, cmd->budget_token); 1769 cmd->budget_token = -1; 1770 switch (ret) { 1771 case BLK_STS_OK: 1772 break; 1773 case BLK_STS_RESOURCE: 1774 case BLK_STS_ZONE_RESOURCE: 1775 if (scsi_device_blocked(sdev)) 1776 ret = BLK_STS_DEV_RESOURCE; 1777 break; 1778 case BLK_STS_AGAIN: 1779 cmd->result = DID_BUS_BUSY << 16; 1780 if (req->rq_flags & RQF_DONTPREP) 1781 scsi_mq_uninit_cmd(cmd); 1782 break; 1783 default: 1784 if (unlikely(!scsi_device_online(sdev))) 1785 cmd->result = DID_NO_CONNECT << 16; 1786 else 1787 cmd->result = DID_ERROR << 16; 1788 /* 1789 * Make sure to release all allocated resources when 1790 * we hit an error, as we will never see this command 1791 * again. 1792 */ 1793 if (req->rq_flags & RQF_DONTPREP) 1794 scsi_mq_uninit_cmd(cmd); 1795 scsi_run_queue_async(sdev); 1796 break; 1797 } 1798 return ret; 1799 } 1800 1801 static int scsi_mq_init_request(struct blk_mq_tag_set *set, struct request *rq, 1802 unsigned int hctx_idx, unsigned int numa_node) 1803 { 1804 struct Scsi_Host *shost = set->driver_data; 1805 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1806 struct scatterlist *sg; 1807 int ret = 0; 1808 1809 cmd->sense_buffer = 1810 kmem_cache_alloc_node(scsi_sense_cache, GFP_KERNEL, numa_node); 1811 if (!cmd->sense_buffer) 1812 return -ENOMEM; 1813 1814 if (scsi_host_get_prot(shost)) { 1815 sg = (void *)cmd + sizeof(struct scsi_cmnd) + 1816 shost->hostt->cmd_size; 1817 cmd->prot_sdb = (void *)sg + scsi_mq_inline_sgl_size(shost); 1818 } 1819 1820 if (shost->hostt->init_cmd_priv) { 1821 ret = shost->hostt->init_cmd_priv(shost, cmd); 1822 if (ret < 0) 1823 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer); 1824 } 1825 1826 return ret; 1827 } 1828 1829 static void scsi_mq_exit_request(struct blk_mq_tag_set *set, struct request *rq, 1830 unsigned int hctx_idx) 1831 { 1832 struct Scsi_Host *shost = set->driver_data; 1833 struct scsi_cmnd *cmd = blk_mq_rq_to_pdu(rq); 1834 1835 if (shost->hostt->exit_cmd_priv) 1836 shost->hostt->exit_cmd_priv(shost, cmd); 1837 kmem_cache_free(scsi_sense_cache, cmd->sense_buffer); 1838 } 1839 1840 1841 static int scsi_mq_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob) 1842 { 1843 struct Scsi_Host *shost = hctx->driver_data; 1844 1845 if (shost->hostt->mq_poll) 1846 return shost->hostt->mq_poll(shost, hctx->queue_num); 1847 1848 return 0; 1849 } 1850 1851 static int scsi_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, 1852 unsigned int hctx_idx) 1853 { 1854 struct Scsi_Host *shost = data; 1855 1856 hctx->driver_data = shost; 1857 return 0; 1858 } 1859 1860 static int scsi_map_queues(struct blk_mq_tag_set *set) 1861 { 1862 struct Scsi_Host *shost = container_of(set, struct Scsi_Host, tag_set); 1863 1864 if (shost->hostt->map_queues) 1865 return shost->hostt->map_queues(shost); 1866 return blk_mq_map_queues(&set->map[HCTX_TYPE_DEFAULT]); 1867 } 1868 1869 void __scsi_init_queue(struct Scsi_Host *shost, struct request_queue *q) 1870 { 1871 struct device *dev = shost->dma_dev; 1872 1873 /* 1874 * this limit is imposed by hardware restrictions 1875 */ 1876 blk_queue_max_segments(q, min_t(unsigned short, shost->sg_tablesize, 1877 SG_MAX_SEGMENTS)); 1878 1879 if (scsi_host_prot_dma(shost)) { 1880 shost->sg_prot_tablesize = 1881 min_not_zero(shost->sg_prot_tablesize, 1882 (unsigned short)SCSI_MAX_PROT_SG_SEGMENTS); 1883 BUG_ON(shost->sg_prot_tablesize < shost->sg_tablesize); 1884 blk_queue_max_integrity_segments(q, shost->sg_prot_tablesize); 1885 } 1886 1887 blk_queue_max_hw_sectors(q, shost->max_sectors); 1888 blk_queue_segment_boundary(q, shost->dma_boundary); 1889 dma_set_seg_boundary(dev, shost->dma_boundary); 1890 1891 blk_queue_max_segment_size(q, shost->max_segment_size); 1892 blk_queue_virt_boundary(q, shost->virt_boundary_mask); 1893 dma_set_max_seg_size(dev, queue_max_segment_size(q)); 1894 1895 /* 1896 * Set a reasonable default alignment: The larger of 32-byte (dword), 1897 * which is a common minimum for HBAs, and the minimum DMA alignment, 1898 * which is set by the platform. 1899 * 1900 * Devices that require a bigger alignment can increase it later. 1901 */ 1902 blk_queue_dma_alignment(q, max(4, dma_get_cache_alignment()) - 1); 1903 } 1904 EXPORT_SYMBOL_GPL(__scsi_init_queue); 1905 1906 static const struct blk_mq_ops scsi_mq_ops_no_commit = { 1907 .get_budget = scsi_mq_get_budget, 1908 .put_budget = scsi_mq_put_budget, 1909 .queue_rq = scsi_queue_rq, 1910 .complete = scsi_complete, 1911 .timeout = scsi_timeout, 1912 #ifdef CONFIG_BLK_DEBUG_FS 1913 .show_rq = scsi_show_rq, 1914 #endif 1915 .init_request = scsi_mq_init_request, 1916 .exit_request = scsi_mq_exit_request, 1917 .cleanup_rq = scsi_cleanup_rq, 1918 .busy = scsi_mq_lld_busy, 1919 .map_queues = scsi_map_queues, 1920 .init_hctx = scsi_init_hctx, 1921 .poll = scsi_mq_poll, 1922 .set_rq_budget_token = scsi_mq_set_rq_budget_token, 1923 .get_rq_budget_token = scsi_mq_get_rq_budget_token, 1924 }; 1925 1926 1927 static void scsi_commit_rqs(struct blk_mq_hw_ctx *hctx) 1928 { 1929 struct Scsi_Host *shost = hctx->driver_data; 1930 1931 shost->hostt->commit_rqs(shost, hctx->queue_num); 1932 } 1933 1934 static const struct blk_mq_ops scsi_mq_ops = { 1935 .get_budget = scsi_mq_get_budget, 1936 .put_budget = scsi_mq_put_budget, 1937 .queue_rq = scsi_queue_rq, 1938 .commit_rqs = scsi_commit_rqs, 1939 .complete = scsi_complete, 1940 .timeout = scsi_timeout, 1941 #ifdef CONFIG_BLK_DEBUG_FS 1942 .show_rq = scsi_show_rq, 1943 #endif 1944 .init_request = scsi_mq_init_request, 1945 .exit_request = scsi_mq_exit_request, 1946 .cleanup_rq = scsi_cleanup_rq, 1947 .busy = scsi_mq_lld_busy, 1948 .map_queues = scsi_map_queues, 1949 .init_hctx = scsi_init_hctx, 1950 .poll = scsi_mq_poll, 1951 .set_rq_budget_token = scsi_mq_set_rq_budget_token, 1952 .get_rq_budget_token = scsi_mq_get_rq_budget_token, 1953 }; 1954 1955 int scsi_mq_setup_tags(struct Scsi_Host *shost) 1956 { 1957 unsigned int cmd_size, sgl_size; 1958 struct blk_mq_tag_set *tag_set = &shost->tag_set; 1959 1960 sgl_size = max_t(unsigned int, sizeof(struct scatterlist), 1961 scsi_mq_inline_sgl_size(shost)); 1962 cmd_size = sizeof(struct scsi_cmnd) + shost->hostt->cmd_size + sgl_size; 1963 if (scsi_host_get_prot(shost)) 1964 cmd_size += sizeof(struct scsi_data_buffer) + 1965 sizeof(struct scatterlist) * SCSI_INLINE_PROT_SG_CNT; 1966 1967 memset(tag_set, 0, sizeof(*tag_set)); 1968 if (shost->hostt->commit_rqs) 1969 tag_set->ops = &scsi_mq_ops; 1970 else 1971 tag_set->ops = &scsi_mq_ops_no_commit; 1972 tag_set->nr_hw_queues = shost->nr_hw_queues ? : 1; 1973 tag_set->nr_maps = shost->nr_maps ? : 1; 1974 tag_set->queue_depth = shost->can_queue; 1975 tag_set->cmd_size = cmd_size; 1976 tag_set->numa_node = dev_to_node(shost->dma_dev); 1977 tag_set->flags = BLK_MQ_F_SHOULD_MERGE; 1978 tag_set->flags |= 1979 BLK_ALLOC_POLICY_TO_MQ_FLAG(shost->hostt->tag_alloc_policy); 1980 tag_set->driver_data = shost; 1981 if (shost->host_tagset) 1982 tag_set->flags |= BLK_MQ_F_TAG_HCTX_SHARED; 1983 1984 return blk_mq_alloc_tag_set(tag_set); 1985 } 1986 1987 void scsi_mq_destroy_tags(struct Scsi_Host *shost) 1988 { 1989 blk_mq_free_tag_set(&shost->tag_set); 1990 } 1991 1992 /** 1993 * scsi_device_from_queue - return sdev associated with a request_queue 1994 * @q: The request queue to return the sdev from 1995 * 1996 * Return the sdev associated with a request queue or NULL if the 1997 * request_queue does not reference a SCSI device. 1998 */ 1999 struct scsi_device *scsi_device_from_queue(struct request_queue *q) 2000 { 2001 struct scsi_device *sdev = NULL; 2002 2003 if (q->mq_ops == &scsi_mq_ops_no_commit || 2004 q->mq_ops == &scsi_mq_ops) 2005 sdev = q->queuedata; 2006 if (!sdev || !get_device(&sdev->sdev_gendev)) 2007 sdev = NULL; 2008 2009 return sdev; 2010 } 2011 /* 2012 * pktcdvd should have been integrated into the SCSI layers, but for historical 2013 * reasons like the old IDE driver it isn't. This export allows it to safely 2014 * probe if a given device is a SCSI one and only attach to that. 2015 */ 2016 #ifdef CONFIG_CDROM_PKTCDVD_MODULE 2017 EXPORT_SYMBOL_GPL(scsi_device_from_queue); 2018 #endif 2019 2020 /** 2021 * scsi_block_requests - Utility function used by low-level drivers to prevent 2022 * further commands from being queued to the device. 2023 * @shost: host in question 2024 * 2025 * There is no timer nor any other means by which the requests get unblocked 2026 * other than the low-level driver calling scsi_unblock_requests(). 2027 */ 2028 void scsi_block_requests(struct Scsi_Host *shost) 2029 { 2030 shost->host_self_blocked = 1; 2031 } 2032 EXPORT_SYMBOL(scsi_block_requests); 2033 2034 /** 2035 * scsi_unblock_requests - Utility function used by low-level drivers to allow 2036 * further commands to be queued to the device. 2037 * @shost: host in question 2038 * 2039 * There is no timer nor any other means by which the requests get unblocked 2040 * other than the low-level driver calling scsi_unblock_requests(). This is done 2041 * as an API function so that changes to the internals of the scsi mid-layer 2042 * won't require wholesale changes to drivers that use this feature. 2043 */ 2044 void scsi_unblock_requests(struct Scsi_Host *shost) 2045 { 2046 shost->host_self_blocked = 0; 2047 scsi_run_host_queues(shost); 2048 } 2049 EXPORT_SYMBOL(scsi_unblock_requests); 2050 2051 void scsi_exit_queue(void) 2052 { 2053 kmem_cache_destroy(scsi_sense_cache); 2054 } 2055 2056 /** 2057 * scsi_mode_select - issue a mode select 2058 * @sdev: SCSI device to be queried 2059 * @pf: Page format bit (1 == standard, 0 == vendor specific) 2060 * @sp: Save page bit (0 == don't save, 1 == save) 2061 * @buffer: request buffer (may not be smaller than eight bytes) 2062 * @len: length of request buffer. 2063 * @timeout: command timeout 2064 * @retries: number of retries before failing 2065 * @data: returns a structure abstracting the mode header data 2066 * @sshdr: place to put sense data (or NULL if no sense to be collected). 2067 * must be SCSI_SENSE_BUFFERSIZE big. 2068 * 2069 * Returns zero if successful; negative error number or scsi 2070 * status on error 2071 * 2072 */ 2073 int scsi_mode_select(struct scsi_device *sdev, int pf, int sp, 2074 unsigned char *buffer, int len, int timeout, int retries, 2075 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 2076 { 2077 unsigned char cmd[10]; 2078 unsigned char *real_buffer; 2079 int ret; 2080 2081 memset(cmd, 0, sizeof(cmd)); 2082 cmd[1] = (pf ? 0x10 : 0) | (sp ? 0x01 : 0); 2083 2084 /* 2085 * Use MODE SELECT(10) if the device asked for it or if the mode page 2086 * and the mode select header cannot fit within the maximumm 255 bytes 2087 * of the MODE SELECT(6) command. 2088 */ 2089 if (sdev->use_10_for_ms || 2090 len + 4 > 255 || 2091 data->block_descriptor_length > 255) { 2092 if (len > 65535 - 8) 2093 return -EINVAL; 2094 real_buffer = kmalloc(8 + len, GFP_KERNEL); 2095 if (!real_buffer) 2096 return -ENOMEM; 2097 memcpy(real_buffer + 8, buffer, len); 2098 len += 8; 2099 real_buffer[0] = 0; 2100 real_buffer[1] = 0; 2101 real_buffer[2] = data->medium_type; 2102 real_buffer[3] = data->device_specific; 2103 real_buffer[4] = data->longlba ? 0x01 : 0; 2104 real_buffer[5] = 0; 2105 put_unaligned_be16(data->block_descriptor_length, 2106 &real_buffer[6]); 2107 2108 cmd[0] = MODE_SELECT_10; 2109 put_unaligned_be16(len, &cmd[7]); 2110 } else { 2111 if (data->longlba) 2112 return -EINVAL; 2113 2114 real_buffer = kmalloc(4 + len, GFP_KERNEL); 2115 if (!real_buffer) 2116 return -ENOMEM; 2117 memcpy(real_buffer + 4, buffer, len); 2118 len += 4; 2119 real_buffer[0] = 0; 2120 real_buffer[1] = data->medium_type; 2121 real_buffer[2] = data->device_specific; 2122 real_buffer[3] = data->block_descriptor_length; 2123 2124 cmd[0] = MODE_SELECT; 2125 cmd[4] = len; 2126 } 2127 2128 ret = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, real_buffer, len, 2129 sshdr, timeout, retries, NULL); 2130 kfree(real_buffer); 2131 return ret; 2132 } 2133 EXPORT_SYMBOL_GPL(scsi_mode_select); 2134 2135 /** 2136 * scsi_mode_sense - issue a mode sense, falling back from 10 to six bytes if necessary. 2137 * @sdev: SCSI device to be queried 2138 * @dbd: set to prevent mode sense from returning block descriptors 2139 * @modepage: mode page being requested 2140 * @buffer: request buffer (may not be smaller than eight bytes) 2141 * @len: length of request buffer. 2142 * @timeout: command timeout 2143 * @retries: number of retries before failing 2144 * @data: returns a structure abstracting the mode header data 2145 * @sshdr: place to put sense data (or NULL if no sense to be collected). 2146 * must be SCSI_SENSE_BUFFERSIZE big. 2147 * 2148 * Returns zero if successful, or a negative error number on failure 2149 */ 2150 int 2151 scsi_mode_sense(struct scsi_device *sdev, int dbd, int modepage, 2152 unsigned char *buffer, int len, int timeout, int retries, 2153 struct scsi_mode_data *data, struct scsi_sense_hdr *sshdr) 2154 { 2155 unsigned char cmd[12]; 2156 int use_10_for_ms; 2157 int header_length; 2158 int result, retry_count = retries; 2159 struct scsi_sense_hdr my_sshdr; 2160 2161 memset(data, 0, sizeof(*data)); 2162 memset(&cmd[0], 0, 12); 2163 2164 dbd = sdev->set_dbd_for_ms ? 8 : dbd; 2165 cmd[1] = dbd & 0x18; /* allows DBD and LLBA bits */ 2166 cmd[2] = modepage; 2167 2168 /* caller might not be interested in sense, but we need it */ 2169 if (!sshdr) 2170 sshdr = &my_sshdr; 2171 2172 retry: 2173 use_10_for_ms = sdev->use_10_for_ms || len > 255; 2174 2175 if (use_10_for_ms) { 2176 if (len < 8 || len > 65535) 2177 return -EINVAL; 2178 2179 cmd[0] = MODE_SENSE_10; 2180 put_unaligned_be16(len, &cmd[7]); 2181 header_length = 8; 2182 } else { 2183 if (len < 4) 2184 return -EINVAL; 2185 2186 cmd[0] = MODE_SENSE; 2187 cmd[4] = len; 2188 header_length = 4; 2189 } 2190 2191 memset(buffer, 0, len); 2192 2193 result = scsi_execute_req(sdev, cmd, DMA_FROM_DEVICE, buffer, len, 2194 sshdr, timeout, retries, NULL); 2195 if (result < 0) 2196 return result; 2197 2198 /* This code looks awful: what it's doing is making sure an 2199 * ILLEGAL REQUEST sense return identifies the actual command 2200 * byte as the problem. MODE_SENSE commands can return 2201 * ILLEGAL REQUEST if the code page isn't supported */ 2202 2203 if (!scsi_status_is_good(result)) { 2204 if (scsi_sense_valid(sshdr)) { 2205 if ((sshdr->sense_key == ILLEGAL_REQUEST) && 2206 (sshdr->asc == 0x20) && (sshdr->ascq == 0)) { 2207 /* 2208 * Invalid command operation code: retry using 2209 * MODE SENSE(6) if this was a MODE SENSE(10) 2210 * request, except if the request mode page is 2211 * too large for MODE SENSE single byte 2212 * allocation length field. 2213 */ 2214 if (use_10_for_ms) { 2215 if (len > 255) 2216 return -EIO; 2217 sdev->use_10_for_ms = 0; 2218 goto retry; 2219 } 2220 } 2221 if (scsi_status_is_check_condition(result) && 2222 sshdr->sense_key == UNIT_ATTENTION && 2223 retry_count) { 2224 retry_count--; 2225 goto retry; 2226 } 2227 } 2228 return -EIO; 2229 } 2230 if (unlikely(buffer[0] == 0x86 && buffer[1] == 0x0b && 2231 (modepage == 6 || modepage == 8))) { 2232 /* Initio breakage? */ 2233 header_length = 0; 2234 data->length = 13; 2235 data->medium_type = 0; 2236 data->device_specific = 0; 2237 data->longlba = 0; 2238 data->block_descriptor_length = 0; 2239 } else if (use_10_for_ms) { 2240 data->length = get_unaligned_be16(&buffer[0]) + 2; 2241 data->medium_type = buffer[2]; 2242 data->device_specific = buffer[3]; 2243 data->longlba = buffer[4] & 0x01; 2244 data->block_descriptor_length = get_unaligned_be16(&buffer[6]); 2245 } else { 2246 data->length = buffer[0] + 1; 2247 data->medium_type = buffer[1]; 2248 data->device_specific = buffer[2]; 2249 data->block_descriptor_length = buffer[3]; 2250 } 2251 data->header_length = header_length; 2252 2253 return 0; 2254 } 2255 EXPORT_SYMBOL(scsi_mode_sense); 2256 2257 /** 2258 * scsi_test_unit_ready - test if unit is ready 2259 * @sdev: scsi device to change the state of. 2260 * @timeout: command timeout 2261 * @retries: number of retries before failing 2262 * @sshdr: outpout pointer for decoded sense information. 2263 * 2264 * Returns zero if unsuccessful or an error if TUR failed. For 2265 * removable media, UNIT_ATTENTION sets ->changed flag. 2266 **/ 2267 int 2268 scsi_test_unit_ready(struct scsi_device *sdev, int timeout, int retries, 2269 struct scsi_sense_hdr *sshdr) 2270 { 2271 char cmd[] = { 2272 TEST_UNIT_READY, 0, 0, 0, 0, 0, 2273 }; 2274 int result; 2275 2276 /* try to eat the UNIT_ATTENTION if there are enough retries */ 2277 do { 2278 result = scsi_execute_req(sdev, cmd, DMA_NONE, NULL, 0, sshdr, 2279 timeout, 1, NULL); 2280 if (sdev->removable && scsi_sense_valid(sshdr) && 2281 sshdr->sense_key == UNIT_ATTENTION) 2282 sdev->changed = 1; 2283 } while (scsi_sense_valid(sshdr) && 2284 sshdr->sense_key == UNIT_ATTENTION && --retries); 2285 2286 return result; 2287 } 2288 EXPORT_SYMBOL(scsi_test_unit_ready); 2289 2290 /** 2291 * scsi_device_set_state - Take the given device through the device state model. 2292 * @sdev: scsi device to change the state of. 2293 * @state: state to change to. 2294 * 2295 * Returns zero if successful or an error if the requested 2296 * transition is illegal. 2297 */ 2298 int 2299 scsi_device_set_state(struct scsi_device *sdev, enum scsi_device_state state) 2300 { 2301 enum scsi_device_state oldstate = sdev->sdev_state; 2302 2303 if (state == oldstate) 2304 return 0; 2305 2306 switch (state) { 2307 case SDEV_CREATED: 2308 switch (oldstate) { 2309 case SDEV_CREATED_BLOCK: 2310 break; 2311 default: 2312 goto illegal; 2313 } 2314 break; 2315 2316 case SDEV_RUNNING: 2317 switch (oldstate) { 2318 case SDEV_CREATED: 2319 case SDEV_OFFLINE: 2320 case SDEV_TRANSPORT_OFFLINE: 2321 case SDEV_QUIESCE: 2322 case SDEV_BLOCK: 2323 break; 2324 default: 2325 goto illegal; 2326 } 2327 break; 2328 2329 case SDEV_QUIESCE: 2330 switch (oldstate) { 2331 case SDEV_RUNNING: 2332 case SDEV_OFFLINE: 2333 case SDEV_TRANSPORT_OFFLINE: 2334 break; 2335 default: 2336 goto illegal; 2337 } 2338 break; 2339 2340 case SDEV_OFFLINE: 2341 case SDEV_TRANSPORT_OFFLINE: 2342 switch (oldstate) { 2343 case SDEV_CREATED: 2344 case SDEV_RUNNING: 2345 case SDEV_QUIESCE: 2346 case SDEV_BLOCK: 2347 break; 2348 default: 2349 goto illegal; 2350 } 2351 break; 2352 2353 case SDEV_BLOCK: 2354 switch (oldstate) { 2355 case SDEV_RUNNING: 2356 case SDEV_CREATED_BLOCK: 2357 case SDEV_QUIESCE: 2358 case SDEV_OFFLINE: 2359 break; 2360 default: 2361 goto illegal; 2362 } 2363 break; 2364 2365 case SDEV_CREATED_BLOCK: 2366 switch (oldstate) { 2367 case SDEV_CREATED: 2368 break; 2369 default: 2370 goto illegal; 2371 } 2372 break; 2373 2374 case SDEV_CANCEL: 2375 switch (oldstate) { 2376 case SDEV_CREATED: 2377 case SDEV_RUNNING: 2378 case SDEV_QUIESCE: 2379 case SDEV_OFFLINE: 2380 case SDEV_TRANSPORT_OFFLINE: 2381 break; 2382 default: 2383 goto illegal; 2384 } 2385 break; 2386 2387 case SDEV_DEL: 2388 switch (oldstate) { 2389 case SDEV_CREATED: 2390 case SDEV_RUNNING: 2391 case SDEV_OFFLINE: 2392 case SDEV_TRANSPORT_OFFLINE: 2393 case SDEV_CANCEL: 2394 case SDEV_BLOCK: 2395 case SDEV_CREATED_BLOCK: 2396 break; 2397 default: 2398 goto illegal; 2399 } 2400 break; 2401 2402 } 2403 sdev->offline_already = false; 2404 sdev->sdev_state = state; 2405 return 0; 2406 2407 illegal: 2408 SCSI_LOG_ERROR_RECOVERY(1, 2409 sdev_printk(KERN_ERR, sdev, 2410 "Illegal state transition %s->%s", 2411 scsi_device_state_name(oldstate), 2412 scsi_device_state_name(state)) 2413 ); 2414 return -EINVAL; 2415 } 2416 EXPORT_SYMBOL(scsi_device_set_state); 2417 2418 /** 2419 * scsi_evt_emit - emit a single SCSI device uevent 2420 * @sdev: associated SCSI device 2421 * @evt: event to emit 2422 * 2423 * Send a single uevent (scsi_event) to the associated scsi_device. 2424 */ 2425 static void scsi_evt_emit(struct scsi_device *sdev, struct scsi_event *evt) 2426 { 2427 int idx = 0; 2428 char *envp[3]; 2429 2430 switch (evt->evt_type) { 2431 case SDEV_EVT_MEDIA_CHANGE: 2432 envp[idx++] = "SDEV_MEDIA_CHANGE=1"; 2433 break; 2434 case SDEV_EVT_INQUIRY_CHANGE_REPORTED: 2435 scsi_rescan_device(&sdev->sdev_gendev); 2436 envp[idx++] = "SDEV_UA=INQUIRY_DATA_HAS_CHANGED"; 2437 break; 2438 case SDEV_EVT_CAPACITY_CHANGE_REPORTED: 2439 envp[idx++] = "SDEV_UA=CAPACITY_DATA_HAS_CHANGED"; 2440 break; 2441 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED: 2442 envp[idx++] = "SDEV_UA=THIN_PROVISIONING_SOFT_THRESHOLD_REACHED"; 2443 break; 2444 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED: 2445 envp[idx++] = "SDEV_UA=MODE_PARAMETERS_CHANGED"; 2446 break; 2447 case SDEV_EVT_LUN_CHANGE_REPORTED: 2448 envp[idx++] = "SDEV_UA=REPORTED_LUNS_DATA_HAS_CHANGED"; 2449 break; 2450 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED: 2451 envp[idx++] = "SDEV_UA=ASYMMETRIC_ACCESS_STATE_CHANGED"; 2452 break; 2453 case SDEV_EVT_POWER_ON_RESET_OCCURRED: 2454 envp[idx++] = "SDEV_UA=POWER_ON_RESET_OCCURRED"; 2455 break; 2456 default: 2457 /* do nothing */ 2458 break; 2459 } 2460 2461 envp[idx++] = NULL; 2462 2463 kobject_uevent_env(&sdev->sdev_gendev.kobj, KOBJ_CHANGE, envp); 2464 } 2465 2466 /** 2467 * scsi_evt_thread - send a uevent for each scsi event 2468 * @work: work struct for scsi_device 2469 * 2470 * Dispatch queued events to their associated scsi_device kobjects 2471 * as uevents. 2472 */ 2473 void scsi_evt_thread(struct work_struct *work) 2474 { 2475 struct scsi_device *sdev; 2476 enum scsi_device_event evt_type; 2477 LIST_HEAD(event_list); 2478 2479 sdev = container_of(work, struct scsi_device, event_work); 2480 2481 for (evt_type = SDEV_EVT_FIRST; evt_type <= SDEV_EVT_LAST; evt_type++) 2482 if (test_and_clear_bit(evt_type, sdev->pending_events)) 2483 sdev_evt_send_simple(sdev, evt_type, GFP_KERNEL); 2484 2485 while (1) { 2486 struct scsi_event *evt; 2487 struct list_head *this, *tmp; 2488 unsigned long flags; 2489 2490 spin_lock_irqsave(&sdev->list_lock, flags); 2491 list_splice_init(&sdev->event_list, &event_list); 2492 spin_unlock_irqrestore(&sdev->list_lock, flags); 2493 2494 if (list_empty(&event_list)) 2495 break; 2496 2497 list_for_each_safe(this, tmp, &event_list) { 2498 evt = list_entry(this, struct scsi_event, node); 2499 list_del(&evt->node); 2500 scsi_evt_emit(sdev, evt); 2501 kfree(evt); 2502 } 2503 } 2504 } 2505 2506 /** 2507 * sdev_evt_send - send asserted event to uevent thread 2508 * @sdev: scsi_device event occurred on 2509 * @evt: event to send 2510 * 2511 * Assert scsi device event asynchronously. 2512 */ 2513 void sdev_evt_send(struct scsi_device *sdev, struct scsi_event *evt) 2514 { 2515 unsigned long flags; 2516 2517 #if 0 2518 /* FIXME: currently this check eliminates all media change events 2519 * for polled devices. Need to update to discriminate between AN 2520 * and polled events */ 2521 if (!test_bit(evt->evt_type, sdev->supported_events)) { 2522 kfree(evt); 2523 return; 2524 } 2525 #endif 2526 2527 spin_lock_irqsave(&sdev->list_lock, flags); 2528 list_add_tail(&evt->node, &sdev->event_list); 2529 schedule_work(&sdev->event_work); 2530 spin_unlock_irqrestore(&sdev->list_lock, flags); 2531 } 2532 EXPORT_SYMBOL_GPL(sdev_evt_send); 2533 2534 /** 2535 * sdev_evt_alloc - allocate a new scsi event 2536 * @evt_type: type of event to allocate 2537 * @gfpflags: GFP flags for allocation 2538 * 2539 * Allocates and returns a new scsi_event. 2540 */ 2541 struct scsi_event *sdev_evt_alloc(enum scsi_device_event evt_type, 2542 gfp_t gfpflags) 2543 { 2544 struct scsi_event *evt = kzalloc(sizeof(struct scsi_event), gfpflags); 2545 if (!evt) 2546 return NULL; 2547 2548 evt->evt_type = evt_type; 2549 INIT_LIST_HEAD(&evt->node); 2550 2551 /* evt_type-specific initialization, if any */ 2552 switch (evt_type) { 2553 case SDEV_EVT_MEDIA_CHANGE: 2554 case SDEV_EVT_INQUIRY_CHANGE_REPORTED: 2555 case SDEV_EVT_CAPACITY_CHANGE_REPORTED: 2556 case SDEV_EVT_SOFT_THRESHOLD_REACHED_REPORTED: 2557 case SDEV_EVT_MODE_PARAMETER_CHANGE_REPORTED: 2558 case SDEV_EVT_LUN_CHANGE_REPORTED: 2559 case SDEV_EVT_ALUA_STATE_CHANGE_REPORTED: 2560 case SDEV_EVT_POWER_ON_RESET_OCCURRED: 2561 default: 2562 /* do nothing */ 2563 break; 2564 } 2565 2566 return evt; 2567 } 2568 EXPORT_SYMBOL_GPL(sdev_evt_alloc); 2569 2570 /** 2571 * sdev_evt_send_simple - send asserted event to uevent thread 2572 * @sdev: scsi_device event occurred on 2573 * @evt_type: type of event to send 2574 * @gfpflags: GFP flags for allocation 2575 * 2576 * Assert scsi device event asynchronously, given an event type. 2577 */ 2578 void sdev_evt_send_simple(struct scsi_device *sdev, 2579 enum scsi_device_event evt_type, gfp_t gfpflags) 2580 { 2581 struct scsi_event *evt = sdev_evt_alloc(evt_type, gfpflags); 2582 if (!evt) { 2583 sdev_printk(KERN_ERR, sdev, "event %d eaten due to OOM\n", 2584 evt_type); 2585 return; 2586 } 2587 2588 sdev_evt_send(sdev, evt); 2589 } 2590 EXPORT_SYMBOL_GPL(sdev_evt_send_simple); 2591 2592 /** 2593 * scsi_device_quiesce - Block all commands except power management. 2594 * @sdev: scsi device to quiesce. 2595 * 2596 * This works by trying to transition to the SDEV_QUIESCE state 2597 * (which must be a legal transition). When the device is in this 2598 * state, only power management requests will be accepted, all others will 2599 * be deferred. 2600 * 2601 * Must be called with user context, may sleep. 2602 * 2603 * Returns zero if unsuccessful or an error if not. 2604 */ 2605 int 2606 scsi_device_quiesce(struct scsi_device *sdev) 2607 { 2608 struct request_queue *q = sdev->request_queue; 2609 int err; 2610 2611 /* 2612 * It is allowed to call scsi_device_quiesce() multiple times from 2613 * the same context but concurrent scsi_device_quiesce() calls are 2614 * not allowed. 2615 */ 2616 WARN_ON_ONCE(sdev->quiesced_by && sdev->quiesced_by != current); 2617 2618 if (sdev->quiesced_by == current) 2619 return 0; 2620 2621 blk_set_pm_only(q); 2622 2623 blk_mq_freeze_queue(q); 2624 /* 2625 * Ensure that the effect of blk_set_pm_only() will be visible 2626 * for percpu_ref_tryget() callers that occur after the queue 2627 * unfreeze even if the queue was already frozen before this function 2628 * was called. See also https://lwn.net/Articles/573497/. 2629 */ 2630 synchronize_rcu(); 2631 blk_mq_unfreeze_queue(q); 2632 2633 mutex_lock(&sdev->state_mutex); 2634 err = scsi_device_set_state(sdev, SDEV_QUIESCE); 2635 if (err == 0) 2636 sdev->quiesced_by = current; 2637 else 2638 blk_clear_pm_only(q); 2639 mutex_unlock(&sdev->state_mutex); 2640 2641 return err; 2642 } 2643 EXPORT_SYMBOL(scsi_device_quiesce); 2644 2645 /** 2646 * scsi_device_resume - Restart user issued commands to a quiesced device. 2647 * @sdev: scsi device to resume. 2648 * 2649 * Moves the device from quiesced back to running and restarts the 2650 * queues. 2651 * 2652 * Must be called with user context, may sleep. 2653 */ 2654 void scsi_device_resume(struct scsi_device *sdev) 2655 { 2656 /* check if the device state was mutated prior to resume, and if 2657 * so assume the state is being managed elsewhere (for example 2658 * device deleted during suspend) 2659 */ 2660 mutex_lock(&sdev->state_mutex); 2661 if (sdev->sdev_state == SDEV_QUIESCE) 2662 scsi_device_set_state(sdev, SDEV_RUNNING); 2663 if (sdev->quiesced_by) { 2664 sdev->quiesced_by = NULL; 2665 blk_clear_pm_only(sdev->request_queue); 2666 } 2667 mutex_unlock(&sdev->state_mutex); 2668 } 2669 EXPORT_SYMBOL(scsi_device_resume); 2670 2671 static void 2672 device_quiesce_fn(struct scsi_device *sdev, void *data) 2673 { 2674 scsi_device_quiesce(sdev); 2675 } 2676 2677 void 2678 scsi_target_quiesce(struct scsi_target *starget) 2679 { 2680 starget_for_each_device(starget, NULL, device_quiesce_fn); 2681 } 2682 EXPORT_SYMBOL(scsi_target_quiesce); 2683 2684 static void 2685 device_resume_fn(struct scsi_device *sdev, void *data) 2686 { 2687 scsi_device_resume(sdev); 2688 } 2689 2690 void 2691 scsi_target_resume(struct scsi_target *starget) 2692 { 2693 starget_for_each_device(starget, NULL, device_resume_fn); 2694 } 2695 EXPORT_SYMBOL(scsi_target_resume); 2696 2697 static int __scsi_internal_device_block_nowait(struct scsi_device *sdev) 2698 { 2699 if (scsi_device_set_state(sdev, SDEV_BLOCK)) 2700 return scsi_device_set_state(sdev, SDEV_CREATED_BLOCK); 2701 2702 return 0; 2703 } 2704 2705 void scsi_start_queue(struct scsi_device *sdev) 2706 { 2707 if (cmpxchg(&sdev->queue_stopped, 1, 0)) 2708 blk_mq_unquiesce_queue(sdev->request_queue); 2709 } 2710 2711 static void scsi_stop_queue(struct scsi_device *sdev, bool nowait) 2712 { 2713 /* 2714 * The atomic variable of ->queue_stopped covers that 2715 * blk_mq_quiesce_queue* is balanced with blk_mq_unquiesce_queue. 2716 * 2717 * However, we still need to wait until quiesce is done 2718 * in case that queue has been stopped. 2719 */ 2720 if (!cmpxchg(&sdev->queue_stopped, 0, 1)) { 2721 if (nowait) 2722 blk_mq_quiesce_queue_nowait(sdev->request_queue); 2723 else 2724 blk_mq_quiesce_queue(sdev->request_queue); 2725 } else { 2726 if (!nowait) 2727 blk_mq_wait_quiesce_done(sdev->request_queue); 2728 } 2729 } 2730 2731 /** 2732 * scsi_internal_device_block_nowait - try to transition to the SDEV_BLOCK state 2733 * @sdev: device to block 2734 * 2735 * Pause SCSI command processing on the specified device. Does not sleep. 2736 * 2737 * Returns zero if successful or a negative error code upon failure. 2738 * 2739 * Notes: 2740 * This routine transitions the device to the SDEV_BLOCK state (which must be 2741 * a legal transition). When the device is in this state, command processing 2742 * is paused until the device leaves the SDEV_BLOCK state. See also 2743 * scsi_internal_device_unblock_nowait(). 2744 */ 2745 int scsi_internal_device_block_nowait(struct scsi_device *sdev) 2746 { 2747 int ret = __scsi_internal_device_block_nowait(sdev); 2748 2749 /* 2750 * The device has transitioned to SDEV_BLOCK. Stop the 2751 * block layer from calling the midlayer with this device's 2752 * request queue. 2753 */ 2754 if (!ret) 2755 scsi_stop_queue(sdev, true); 2756 return ret; 2757 } 2758 EXPORT_SYMBOL_GPL(scsi_internal_device_block_nowait); 2759 2760 /** 2761 * scsi_internal_device_block - try to transition to the SDEV_BLOCK state 2762 * @sdev: device to block 2763 * 2764 * Pause SCSI command processing on the specified device and wait until all 2765 * ongoing scsi_request_fn() / scsi_queue_rq() calls have finished. May sleep. 2766 * 2767 * Returns zero if successful or a negative error code upon failure. 2768 * 2769 * Note: 2770 * This routine transitions the device to the SDEV_BLOCK state (which must be 2771 * a legal transition). When the device is in this state, command processing 2772 * is paused until the device leaves the SDEV_BLOCK state. See also 2773 * scsi_internal_device_unblock(). 2774 */ 2775 static int scsi_internal_device_block(struct scsi_device *sdev) 2776 { 2777 int err; 2778 2779 mutex_lock(&sdev->state_mutex); 2780 err = __scsi_internal_device_block_nowait(sdev); 2781 if (err == 0) 2782 scsi_stop_queue(sdev, false); 2783 mutex_unlock(&sdev->state_mutex); 2784 2785 return err; 2786 } 2787 2788 /** 2789 * scsi_internal_device_unblock_nowait - resume a device after a block request 2790 * @sdev: device to resume 2791 * @new_state: state to set the device to after unblocking 2792 * 2793 * Restart the device queue for a previously suspended SCSI device. Does not 2794 * sleep. 2795 * 2796 * Returns zero if successful or a negative error code upon failure. 2797 * 2798 * Notes: 2799 * This routine transitions the device to the SDEV_RUNNING state or to one of 2800 * the offline states (which must be a legal transition) allowing the midlayer 2801 * to goose the queue for this device. 2802 */ 2803 int scsi_internal_device_unblock_nowait(struct scsi_device *sdev, 2804 enum scsi_device_state new_state) 2805 { 2806 switch (new_state) { 2807 case SDEV_RUNNING: 2808 case SDEV_TRANSPORT_OFFLINE: 2809 break; 2810 default: 2811 return -EINVAL; 2812 } 2813 2814 /* 2815 * Try to transition the scsi device to SDEV_RUNNING or one of the 2816 * offlined states and goose the device queue if successful. 2817 */ 2818 switch (sdev->sdev_state) { 2819 case SDEV_BLOCK: 2820 case SDEV_TRANSPORT_OFFLINE: 2821 sdev->sdev_state = new_state; 2822 break; 2823 case SDEV_CREATED_BLOCK: 2824 if (new_state == SDEV_TRANSPORT_OFFLINE || 2825 new_state == SDEV_OFFLINE) 2826 sdev->sdev_state = new_state; 2827 else 2828 sdev->sdev_state = SDEV_CREATED; 2829 break; 2830 case SDEV_CANCEL: 2831 case SDEV_OFFLINE: 2832 break; 2833 default: 2834 return -EINVAL; 2835 } 2836 scsi_start_queue(sdev); 2837 2838 return 0; 2839 } 2840 EXPORT_SYMBOL_GPL(scsi_internal_device_unblock_nowait); 2841 2842 /** 2843 * scsi_internal_device_unblock - resume a device after a block request 2844 * @sdev: device to resume 2845 * @new_state: state to set the device to after unblocking 2846 * 2847 * Restart the device queue for a previously suspended SCSI device. May sleep. 2848 * 2849 * Returns zero if successful or a negative error code upon failure. 2850 * 2851 * Notes: 2852 * This routine transitions the device to the SDEV_RUNNING state or to one of 2853 * the offline states (which must be a legal transition) allowing the midlayer 2854 * to goose the queue for this device. 2855 */ 2856 static int scsi_internal_device_unblock(struct scsi_device *sdev, 2857 enum scsi_device_state new_state) 2858 { 2859 int ret; 2860 2861 mutex_lock(&sdev->state_mutex); 2862 ret = scsi_internal_device_unblock_nowait(sdev, new_state); 2863 mutex_unlock(&sdev->state_mutex); 2864 2865 return ret; 2866 } 2867 2868 static void 2869 device_block(struct scsi_device *sdev, void *data) 2870 { 2871 int ret; 2872 2873 ret = scsi_internal_device_block(sdev); 2874 2875 WARN_ONCE(ret, "scsi_internal_device_block(%s) failed: ret = %d\n", 2876 dev_name(&sdev->sdev_gendev), ret); 2877 } 2878 2879 static int 2880 target_block(struct device *dev, void *data) 2881 { 2882 if (scsi_is_target_device(dev)) 2883 starget_for_each_device(to_scsi_target(dev), NULL, 2884 device_block); 2885 return 0; 2886 } 2887 2888 void 2889 scsi_target_block(struct device *dev) 2890 { 2891 if (scsi_is_target_device(dev)) 2892 starget_for_each_device(to_scsi_target(dev), NULL, 2893 device_block); 2894 else 2895 device_for_each_child(dev, NULL, target_block); 2896 } 2897 EXPORT_SYMBOL_GPL(scsi_target_block); 2898 2899 static void 2900 device_unblock(struct scsi_device *sdev, void *data) 2901 { 2902 scsi_internal_device_unblock(sdev, *(enum scsi_device_state *)data); 2903 } 2904 2905 static int 2906 target_unblock(struct device *dev, void *data) 2907 { 2908 if (scsi_is_target_device(dev)) 2909 starget_for_each_device(to_scsi_target(dev), data, 2910 device_unblock); 2911 return 0; 2912 } 2913 2914 void 2915 scsi_target_unblock(struct device *dev, enum scsi_device_state new_state) 2916 { 2917 if (scsi_is_target_device(dev)) 2918 starget_for_each_device(to_scsi_target(dev), &new_state, 2919 device_unblock); 2920 else 2921 device_for_each_child(dev, &new_state, target_unblock); 2922 } 2923 EXPORT_SYMBOL_GPL(scsi_target_unblock); 2924 2925 int 2926 scsi_host_block(struct Scsi_Host *shost) 2927 { 2928 struct scsi_device *sdev; 2929 int ret = 0; 2930 2931 /* 2932 * Call scsi_internal_device_block_nowait so we can avoid 2933 * calling synchronize_rcu() for each LUN. 2934 */ 2935 shost_for_each_device(sdev, shost) { 2936 mutex_lock(&sdev->state_mutex); 2937 ret = scsi_internal_device_block_nowait(sdev); 2938 mutex_unlock(&sdev->state_mutex); 2939 if (ret) { 2940 scsi_device_put(sdev); 2941 break; 2942 } 2943 } 2944 2945 /* 2946 * SCSI never enables blk-mq's BLK_MQ_F_BLOCKING flag so 2947 * calling synchronize_rcu() once is enough. 2948 */ 2949 WARN_ON_ONCE(shost->tag_set.flags & BLK_MQ_F_BLOCKING); 2950 2951 if (!ret) 2952 synchronize_rcu(); 2953 2954 return ret; 2955 } 2956 EXPORT_SYMBOL_GPL(scsi_host_block); 2957 2958 int 2959 scsi_host_unblock(struct Scsi_Host *shost, int new_state) 2960 { 2961 struct scsi_device *sdev; 2962 int ret = 0; 2963 2964 shost_for_each_device(sdev, shost) { 2965 ret = scsi_internal_device_unblock(sdev, new_state); 2966 if (ret) { 2967 scsi_device_put(sdev); 2968 break; 2969 } 2970 } 2971 return ret; 2972 } 2973 EXPORT_SYMBOL_GPL(scsi_host_unblock); 2974 2975 /** 2976 * scsi_kmap_atomic_sg - find and atomically map an sg-elemnt 2977 * @sgl: scatter-gather list 2978 * @sg_count: number of segments in sg 2979 * @offset: offset in bytes into sg, on return offset into the mapped area 2980 * @len: bytes to map, on return number of bytes mapped 2981 * 2982 * Returns virtual address of the start of the mapped page 2983 */ 2984 void *scsi_kmap_atomic_sg(struct scatterlist *sgl, int sg_count, 2985 size_t *offset, size_t *len) 2986 { 2987 int i; 2988 size_t sg_len = 0, len_complete = 0; 2989 struct scatterlist *sg; 2990 struct page *page; 2991 2992 WARN_ON(!irqs_disabled()); 2993 2994 for_each_sg(sgl, sg, sg_count, i) { 2995 len_complete = sg_len; /* Complete sg-entries */ 2996 sg_len += sg->length; 2997 if (sg_len > *offset) 2998 break; 2999 } 3000 3001 if (unlikely(i == sg_count)) { 3002 printk(KERN_ERR "%s: Bytes in sg: %zu, requested offset %zu, " 3003 "elements %d\n", 3004 __func__, sg_len, *offset, sg_count); 3005 WARN_ON(1); 3006 return NULL; 3007 } 3008 3009 /* Offset starting from the beginning of first page in this sg-entry */ 3010 *offset = *offset - len_complete + sg->offset; 3011 3012 /* Assumption: contiguous pages can be accessed as "page + i" */ 3013 page = nth_page(sg_page(sg), (*offset >> PAGE_SHIFT)); 3014 *offset &= ~PAGE_MASK; 3015 3016 /* Bytes in this sg-entry from *offset to the end of the page */ 3017 sg_len = PAGE_SIZE - *offset; 3018 if (*len > sg_len) 3019 *len = sg_len; 3020 3021 return kmap_atomic(page); 3022 } 3023 EXPORT_SYMBOL(scsi_kmap_atomic_sg); 3024 3025 /** 3026 * scsi_kunmap_atomic_sg - atomically unmap a virtual address, previously mapped with scsi_kmap_atomic_sg 3027 * @virt: virtual address to be unmapped 3028 */ 3029 void scsi_kunmap_atomic_sg(void *virt) 3030 { 3031 kunmap_atomic(virt); 3032 } 3033 EXPORT_SYMBOL(scsi_kunmap_atomic_sg); 3034 3035 void sdev_disable_disk_events(struct scsi_device *sdev) 3036 { 3037 atomic_inc(&sdev->disk_events_disable_depth); 3038 } 3039 EXPORT_SYMBOL(sdev_disable_disk_events); 3040 3041 void sdev_enable_disk_events(struct scsi_device *sdev) 3042 { 3043 if (WARN_ON_ONCE(atomic_read(&sdev->disk_events_disable_depth) <= 0)) 3044 return; 3045 atomic_dec(&sdev->disk_events_disable_depth); 3046 } 3047 EXPORT_SYMBOL(sdev_enable_disk_events); 3048 3049 static unsigned char designator_prio(const unsigned char *d) 3050 { 3051 if (d[1] & 0x30) 3052 /* not associated with LUN */ 3053 return 0; 3054 3055 if (d[3] == 0) 3056 /* invalid length */ 3057 return 0; 3058 3059 /* 3060 * Order of preference for lun descriptor: 3061 * - SCSI name string 3062 * - NAA IEEE Registered Extended 3063 * - EUI-64 based 16-byte 3064 * - EUI-64 based 12-byte 3065 * - NAA IEEE Registered 3066 * - NAA IEEE Extended 3067 * - EUI-64 based 8-byte 3068 * - SCSI name string (truncated) 3069 * - T10 Vendor ID 3070 * as longer descriptors reduce the likelyhood 3071 * of identification clashes. 3072 */ 3073 3074 switch (d[1] & 0xf) { 3075 case 8: 3076 /* SCSI name string, variable-length UTF-8 */ 3077 return 9; 3078 case 3: 3079 switch (d[4] >> 4) { 3080 case 6: 3081 /* NAA registered extended */ 3082 return 8; 3083 case 5: 3084 /* NAA registered */ 3085 return 5; 3086 case 4: 3087 /* NAA extended */ 3088 return 4; 3089 case 3: 3090 /* NAA locally assigned */ 3091 return 1; 3092 default: 3093 break; 3094 } 3095 break; 3096 case 2: 3097 switch (d[3]) { 3098 case 16: 3099 /* EUI64-based, 16 byte */ 3100 return 7; 3101 case 12: 3102 /* EUI64-based, 12 byte */ 3103 return 6; 3104 case 8: 3105 /* EUI64-based, 8 byte */ 3106 return 3; 3107 default: 3108 break; 3109 } 3110 break; 3111 case 1: 3112 /* T10 vendor ID */ 3113 return 1; 3114 default: 3115 break; 3116 } 3117 3118 return 0; 3119 } 3120 3121 /** 3122 * scsi_vpd_lun_id - return a unique device identification 3123 * @sdev: SCSI device 3124 * @id: buffer for the identification 3125 * @id_len: length of the buffer 3126 * 3127 * Copies a unique device identification into @id based 3128 * on the information in the VPD page 0x83 of the device. 3129 * The string will be formatted as a SCSI name string. 3130 * 3131 * Returns the length of the identification or error on failure. 3132 * If the identifier is longer than the supplied buffer the actual 3133 * identifier length is returned and the buffer is not zero-padded. 3134 */ 3135 int scsi_vpd_lun_id(struct scsi_device *sdev, char *id, size_t id_len) 3136 { 3137 u8 cur_id_prio = 0; 3138 u8 cur_id_size = 0; 3139 const unsigned char *d, *cur_id_str; 3140 const struct scsi_vpd *vpd_pg83; 3141 int id_size = -EINVAL; 3142 3143 rcu_read_lock(); 3144 vpd_pg83 = rcu_dereference(sdev->vpd_pg83); 3145 if (!vpd_pg83) { 3146 rcu_read_unlock(); 3147 return -ENXIO; 3148 } 3149 3150 /* The id string must be at least 20 bytes + terminating NULL byte */ 3151 if (id_len < 21) { 3152 rcu_read_unlock(); 3153 return -EINVAL; 3154 } 3155 3156 memset(id, 0, id_len); 3157 for (d = vpd_pg83->data + 4; 3158 d < vpd_pg83->data + vpd_pg83->len; 3159 d += d[3] + 4) { 3160 u8 prio = designator_prio(d); 3161 3162 if (prio == 0 || cur_id_prio > prio) 3163 continue; 3164 3165 switch (d[1] & 0xf) { 3166 case 0x1: 3167 /* T10 Vendor ID */ 3168 if (cur_id_size > d[3]) 3169 break; 3170 cur_id_prio = prio; 3171 cur_id_size = d[3]; 3172 if (cur_id_size + 4 > id_len) 3173 cur_id_size = id_len - 4; 3174 cur_id_str = d + 4; 3175 id_size = snprintf(id, id_len, "t10.%*pE", 3176 cur_id_size, cur_id_str); 3177 break; 3178 case 0x2: 3179 /* EUI-64 */ 3180 cur_id_prio = prio; 3181 cur_id_size = d[3]; 3182 cur_id_str = d + 4; 3183 switch (cur_id_size) { 3184 case 8: 3185 id_size = snprintf(id, id_len, 3186 "eui.%8phN", 3187 cur_id_str); 3188 break; 3189 case 12: 3190 id_size = snprintf(id, id_len, 3191 "eui.%12phN", 3192 cur_id_str); 3193 break; 3194 case 16: 3195 id_size = snprintf(id, id_len, 3196 "eui.%16phN", 3197 cur_id_str); 3198 break; 3199 default: 3200 break; 3201 } 3202 break; 3203 case 0x3: 3204 /* NAA */ 3205 cur_id_prio = prio; 3206 cur_id_size = d[3]; 3207 cur_id_str = d + 4; 3208 switch (cur_id_size) { 3209 case 8: 3210 id_size = snprintf(id, id_len, 3211 "naa.%8phN", 3212 cur_id_str); 3213 break; 3214 case 16: 3215 id_size = snprintf(id, id_len, 3216 "naa.%16phN", 3217 cur_id_str); 3218 break; 3219 default: 3220 break; 3221 } 3222 break; 3223 case 0x8: 3224 /* SCSI name string */ 3225 if (cur_id_size > d[3]) 3226 break; 3227 /* Prefer others for truncated descriptor */ 3228 if (d[3] > id_len) { 3229 prio = 2; 3230 if (cur_id_prio > prio) 3231 break; 3232 } 3233 cur_id_prio = prio; 3234 cur_id_size = id_size = d[3]; 3235 cur_id_str = d + 4; 3236 if (cur_id_size >= id_len) 3237 cur_id_size = id_len - 1; 3238 memcpy(id, cur_id_str, cur_id_size); 3239 break; 3240 default: 3241 break; 3242 } 3243 } 3244 rcu_read_unlock(); 3245 3246 return id_size; 3247 } 3248 EXPORT_SYMBOL(scsi_vpd_lun_id); 3249 3250 /* 3251 * scsi_vpd_tpg_id - return a target port group identifier 3252 * @sdev: SCSI device 3253 * 3254 * Returns the Target Port Group identifier from the information 3255 * froom VPD page 0x83 of the device. 3256 * 3257 * Returns the identifier or error on failure. 3258 */ 3259 int scsi_vpd_tpg_id(struct scsi_device *sdev, int *rel_id) 3260 { 3261 const unsigned char *d; 3262 const struct scsi_vpd *vpd_pg83; 3263 int group_id = -EAGAIN, rel_port = -1; 3264 3265 rcu_read_lock(); 3266 vpd_pg83 = rcu_dereference(sdev->vpd_pg83); 3267 if (!vpd_pg83) { 3268 rcu_read_unlock(); 3269 return -ENXIO; 3270 } 3271 3272 d = vpd_pg83->data + 4; 3273 while (d < vpd_pg83->data + vpd_pg83->len) { 3274 switch (d[1] & 0xf) { 3275 case 0x4: 3276 /* Relative target port */ 3277 rel_port = get_unaligned_be16(&d[6]); 3278 break; 3279 case 0x5: 3280 /* Target port group */ 3281 group_id = get_unaligned_be16(&d[6]); 3282 break; 3283 default: 3284 break; 3285 } 3286 d += d[3] + 4; 3287 } 3288 rcu_read_unlock(); 3289 3290 if (group_id >= 0 && rel_id && rel_port != -1) 3291 *rel_id = rel_port; 3292 3293 return group_id; 3294 } 3295 EXPORT_SYMBOL(scsi_vpd_tpg_id); 3296 3297 /** 3298 * scsi_build_sense - build sense data for a command 3299 * @scmd: scsi command for which the sense should be formatted 3300 * @desc: Sense format (non-zero == descriptor format, 3301 * 0 == fixed format) 3302 * @key: Sense key 3303 * @asc: Additional sense code 3304 * @ascq: Additional sense code qualifier 3305 * 3306 **/ 3307 void scsi_build_sense(struct scsi_cmnd *scmd, int desc, u8 key, u8 asc, u8 ascq) 3308 { 3309 scsi_build_sense_buffer(desc, scmd->sense_buffer, key, asc, ascq); 3310 scmd->result = SAM_STAT_CHECK_CONDITION; 3311 } 3312 EXPORT_SYMBOL_GPL(scsi_build_sense); 3313