1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_BLKDEV_H 3 #define _LINUX_BLKDEV_H 4 5 #include <linux/sched.h> 6 #include <linux/sched/clock.h> 7 8 #ifdef CONFIG_BLOCK 9 10 #include <linux/major.h> 11 #include <linux/genhd.h> 12 #include <linux/list.h> 13 #include <linux/llist.h> 14 #include <linux/timer.h> 15 #include <linux/workqueue.h> 16 #include <linux/pagemap.h> 17 #include <linux/backing-dev-defs.h> 18 #include <linux/wait.h> 19 #include <linux/mempool.h> 20 #include <linux/pfn.h> 21 #include <linux/bio.h> 22 #include <linux/stringify.h> 23 #include <linux/gfp.h> 24 #include <linux/bsg.h> 25 #include <linux/smp.h> 26 #include <linux/rcupdate.h> 27 #include <linux/percpu-refcount.h> 28 #include <linux/scatterlist.h> 29 #include <linux/blkzoned.h> 30 31 struct module; 32 struct scsi_ioctl_command; 33 34 struct request_queue; 35 struct elevator_queue; 36 struct blk_trace; 37 struct request; 38 struct sg_io_hdr; 39 struct bsg_job; 40 struct blkcg_gq; 41 struct blk_flush_queue; 42 struct pr_ops; 43 struct rq_qos; 44 struct blk_queue_stats; 45 struct blk_stat_callback; 46 47 #define BLKDEV_MIN_RQ 4 48 #define BLKDEV_MAX_RQ 128 /* Default maximum */ 49 50 /* Must be consistent with blk_mq_poll_stats_bkt() */ 51 #define BLK_MQ_POLL_STATS_BKTS 16 52 53 /* Doing classic polling */ 54 #define BLK_MQ_POLL_CLASSIC -1 55 56 /* 57 * Maximum number of blkcg policies allowed to be registered concurrently. 58 * Defined here to simplify include dependency. 59 */ 60 #define BLKCG_MAX_POLS 5 61 62 typedef void (rq_end_io_fn)(struct request *, blk_status_t); 63 64 /* 65 * request flags */ 66 typedef __u32 __bitwise req_flags_t; 67 68 /* elevator knows about this request */ 69 #define RQF_SORTED ((__force req_flags_t)(1 << 0)) 70 /* drive already may have started this one */ 71 #define RQF_STARTED ((__force req_flags_t)(1 << 1)) 72 /* may not be passed by ioscheduler */ 73 #define RQF_SOFTBARRIER ((__force req_flags_t)(1 << 3)) 74 /* request for flush sequence */ 75 #define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << 4)) 76 /* merge of different types, fail separately */ 77 #define RQF_MIXED_MERGE ((__force req_flags_t)(1 << 5)) 78 /* track inflight for MQ */ 79 #define RQF_MQ_INFLIGHT ((__force req_flags_t)(1 << 6)) 80 /* don't call prep for this one */ 81 #define RQF_DONTPREP ((__force req_flags_t)(1 << 7)) 82 /* set for "ide_preempt" requests and also for requests for which the SCSI 83 "quiesce" state must be ignored. */ 84 #define RQF_PREEMPT ((__force req_flags_t)(1 << 8)) 85 /* contains copies of user pages */ 86 #define RQF_COPY_USER ((__force req_flags_t)(1 << 9)) 87 /* vaguely specified driver internal error. Ignored by the block layer */ 88 #define RQF_FAILED ((__force req_flags_t)(1 << 10)) 89 /* don't warn about errors */ 90 #define RQF_QUIET ((__force req_flags_t)(1 << 11)) 91 /* elevator private data attached */ 92 #define RQF_ELVPRIV ((__force req_flags_t)(1 << 12)) 93 /* account into disk and partition IO statistics */ 94 #define RQF_IO_STAT ((__force req_flags_t)(1 << 13)) 95 /* request came from our alloc pool */ 96 #define RQF_ALLOCED ((__force req_flags_t)(1 << 14)) 97 /* runtime pm request */ 98 #define RQF_PM ((__force req_flags_t)(1 << 15)) 99 /* on IO scheduler merge hash */ 100 #define RQF_HASHED ((__force req_flags_t)(1 << 16)) 101 /* track IO completion time */ 102 #define RQF_STATS ((__force req_flags_t)(1 << 17)) 103 /* Look at ->special_vec for the actual data payload instead of the 104 bio chain. */ 105 #define RQF_SPECIAL_PAYLOAD ((__force req_flags_t)(1 << 18)) 106 /* The per-zone write lock is held for this request */ 107 #define RQF_ZONE_WRITE_LOCKED ((__force req_flags_t)(1 << 19)) 108 /* already slept for hybrid poll */ 109 #define RQF_MQ_POLL_SLEPT ((__force req_flags_t)(1 << 20)) 110 /* ->timeout has been called, don't expire again */ 111 #define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21)) 112 113 /* flags that prevent us from merging requests: */ 114 #define RQF_NOMERGE_FLAGS \ 115 (RQF_STARTED | RQF_SOFTBARRIER | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD) 116 117 /* 118 * Request state for blk-mq. 119 */ 120 enum mq_rq_state { 121 MQ_RQ_IDLE = 0, 122 MQ_RQ_IN_FLIGHT = 1, 123 MQ_RQ_COMPLETE = 2, 124 }; 125 126 /* 127 * Try to put the fields that are referenced together in the same cacheline. 128 * 129 * If you modify this structure, make sure to update blk_rq_init() and 130 * especially blk_mq_rq_ctx_init() to take care of the added fields. 131 */ 132 struct request { 133 struct request_queue *q; 134 struct blk_mq_ctx *mq_ctx; 135 struct blk_mq_hw_ctx *mq_hctx; 136 137 unsigned int cmd_flags; /* op and common flags */ 138 req_flags_t rq_flags; 139 140 int tag; 141 int internal_tag; 142 143 /* the following two fields are internal, NEVER access directly */ 144 unsigned int __data_len; /* total data len */ 145 sector_t __sector; /* sector cursor */ 146 147 struct bio *bio; 148 struct bio *biotail; 149 150 struct list_head queuelist; 151 152 /* 153 * The hash is used inside the scheduler, and killed once the 154 * request reaches the dispatch list. The ipi_list is only used 155 * to queue the request for softirq completion, which is long 156 * after the request has been unhashed (and even removed from 157 * the dispatch list). 158 */ 159 union { 160 struct hlist_node hash; /* merge hash */ 161 struct list_head ipi_list; 162 }; 163 164 /* 165 * The rb_node is only used inside the io scheduler, requests 166 * are pruned when moved to the dispatch queue. So let the 167 * completion_data share space with the rb_node. 168 */ 169 union { 170 struct rb_node rb_node; /* sort/lookup */ 171 struct bio_vec special_vec; 172 void *completion_data; 173 int error_count; /* for legacy drivers, don't use */ 174 }; 175 176 /* 177 * Three pointers are available for the IO schedulers, if they need 178 * more they have to dynamically allocate it. Flush requests are 179 * never put on the IO scheduler. So let the flush fields share 180 * space with the elevator data. 181 */ 182 union { 183 struct { 184 struct io_cq *icq; 185 void *priv[2]; 186 } elv; 187 188 struct { 189 unsigned int seq; 190 struct list_head list; 191 rq_end_io_fn *saved_end_io; 192 } flush; 193 }; 194 195 struct gendisk *rq_disk; 196 struct hd_struct *part; 197 #ifdef CONFIG_BLK_RQ_ALLOC_TIME 198 /* Time that the first bio started allocating this request. */ 199 u64 alloc_time_ns; 200 #endif 201 /* Time that this request was allocated for this IO. */ 202 u64 start_time_ns; 203 /* Time that I/O was submitted to the device. */ 204 u64 io_start_time_ns; 205 206 #ifdef CONFIG_BLK_WBT 207 unsigned short wbt_flags; 208 #endif 209 /* 210 * rq sectors used for blk stats. It has the same value 211 * with blk_rq_sectors(rq), except that it never be zeroed 212 * by completion. 213 */ 214 unsigned short stats_sectors; 215 216 /* 217 * Number of scatter-gather DMA addr+len pairs after 218 * physical address coalescing is performed. 219 */ 220 unsigned short nr_phys_segments; 221 222 #if defined(CONFIG_BLK_DEV_INTEGRITY) 223 unsigned short nr_integrity_segments; 224 #endif 225 226 unsigned short write_hint; 227 unsigned short ioprio; 228 229 unsigned int extra_len; /* length of alignment and padding */ 230 231 enum mq_rq_state state; 232 refcount_t ref; 233 234 unsigned int timeout; 235 unsigned long deadline; 236 237 union { 238 struct __call_single_data csd; 239 u64 fifo_time; 240 }; 241 242 /* 243 * completion callback. 244 */ 245 rq_end_io_fn *end_io; 246 void *end_io_data; 247 }; 248 249 static inline bool blk_op_is_scsi(unsigned int op) 250 { 251 return op == REQ_OP_SCSI_IN || op == REQ_OP_SCSI_OUT; 252 } 253 254 static inline bool blk_op_is_private(unsigned int op) 255 { 256 return op == REQ_OP_DRV_IN || op == REQ_OP_DRV_OUT; 257 } 258 259 static inline bool blk_rq_is_scsi(struct request *rq) 260 { 261 return blk_op_is_scsi(req_op(rq)); 262 } 263 264 static inline bool blk_rq_is_private(struct request *rq) 265 { 266 return blk_op_is_private(req_op(rq)); 267 } 268 269 static inline bool blk_rq_is_passthrough(struct request *rq) 270 { 271 return blk_rq_is_scsi(rq) || blk_rq_is_private(rq); 272 } 273 274 static inline bool bio_is_passthrough(struct bio *bio) 275 { 276 unsigned op = bio_op(bio); 277 278 return blk_op_is_scsi(op) || blk_op_is_private(op); 279 } 280 281 static inline unsigned short req_get_ioprio(struct request *req) 282 { 283 return req->ioprio; 284 } 285 286 #include <linux/elevator.h> 287 288 struct blk_queue_ctx; 289 290 typedef blk_qc_t (make_request_fn) (struct request_queue *q, struct bio *bio); 291 292 struct bio_vec; 293 typedef int (dma_drain_needed_fn)(struct request *); 294 295 enum blk_eh_timer_return { 296 BLK_EH_DONE, /* drivers has completed the command */ 297 BLK_EH_RESET_TIMER, /* reset timer and try again */ 298 }; 299 300 enum blk_queue_state { 301 Queue_down, 302 Queue_up, 303 }; 304 305 #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */ 306 #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */ 307 308 #define BLK_SCSI_MAX_CMDS (256) 309 #define BLK_SCSI_CMD_PER_LONG (BLK_SCSI_MAX_CMDS / (sizeof(long) * 8)) 310 311 /* 312 * Zoned block device models (zoned limit). 313 */ 314 enum blk_zoned_model { 315 BLK_ZONED_NONE, /* Regular block device */ 316 BLK_ZONED_HA, /* Host-aware zoned block device */ 317 BLK_ZONED_HM, /* Host-managed zoned block device */ 318 }; 319 320 struct queue_limits { 321 unsigned long bounce_pfn; 322 unsigned long seg_boundary_mask; 323 unsigned long virt_boundary_mask; 324 325 unsigned int max_hw_sectors; 326 unsigned int max_dev_sectors; 327 unsigned int chunk_sectors; 328 unsigned int max_sectors; 329 unsigned int max_segment_size; 330 unsigned int physical_block_size; 331 unsigned int logical_block_size; 332 unsigned int alignment_offset; 333 unsigned int io_min; 334 unsigned int io_opt; 335 unsigned int max_discard_sectors; 336 unsigned int max_hw_discard_sectors; 337 unsigned int max_write_same_sectors; 338 unsigned int max_write_zeroes_sectors; 339 unsigned int discard_granularity; 340 unsigned int discard_alignment; 341 342 unsigned short max_segments; 343 unsigned short max_integrity_segments; 344 unsigned short max_discard_segments; 345 346 unsigned char misaligned; 347 unsigned char discard_misaligned; 348 unsigned char raid_partial_stripes_expensive; 349 enum blk_zoned_model zoned; 350 }; 351 352 typedef int (*report_zones_cb)(struct blk_zone *zone, unsigned int idx, 353 void *data); 354 355 #ifdef CONFIG_BLK_DEV_ZONED 356 357 #define BLK_ALL_ZONES ((unsigned int)-1) 358 int blkdev_report_zones(struct block_device *bdev, sector_t sector, 359 unsigned int nr_zones, report_zones_cb cb, void *data); 360 unsigned int blkdev_nr_zones(struct gendisk *disk); 361 extern int blkdev_zone_mgmt(struct block_device *bdev, enum req_opf op, 362 sector_t sectors, sector_t nr_sectors, 363 gfp_t gfp_mask); 364 extern int blk_revalidate_disk_zones(struct gendisk *disk); 365 366 extern int blkdev_report_zones_ioctl(struct block_device *bdev, fmode_t mode, 367 unsigned int cmd, unsigned long arg); 368 extern int blkdev_zone_mgmt_ioctl(struct block_device *bdev, fmode_t mode, 369 unsigned int cmd, unsigned long arg); 370 371 #else /* CONFIG_BLK_DEV_ZONED */ 372 373 static inline unsigned int blkdev_nr_zones(struct gendisk *disk) 374 { 375 return 0; 376 } 377 378 static inline int blkdev_report_zones_ioctl(struct block_device *bdev, 379 fmode_t mode, unsigned int cmd, 380 unsigned long arg) 381 { 382 return -ENOTTY; 383 } 384 385 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev, 386 fmode_t mode, unsigned int cmd, 387 unsigned long arg) 388 { 389 return -ENOTTY; 390 } 391 392 #endif /* CONFIG_BLK_DEV_ZONED */ 393 394 struct request_queue { 395 struct request *last_merge; 396 struct elevator_queue *elevator; 397 398 struct blk_queue_stats *stats; 399 struct rq_qos *rq_qos; 400 401 make_request_fn *make_request_fn; 402 dma_drain_needed_fn *dma_drain_needed; 403 404 const struct blk_mq_ops *mq_ops; 405 406 /* sw queues */ 407 struct blk_mq_ctx __percpu *queue_ctx; 408 409 unsigned int queue_depth; 410 411 /* hw dispatch queues */ 412 struct blk_mq_hw_ctx **queue_hw_ctx; 413 unsigned int nr_hw_queues; 414 415 struct backing_dev_info *backing_dev_info; 416 417 /* 418 * The queue owner gets to use this for whatever they like. 419 * ll_rw_blk doesn't touch it. 420 */ 421 void *queuedata; 422 423 /* 424 * various queue flags, see QUEUE_* below 425 */ 426 unsigned long queue_flags; 427 /* 428 * Number of contexts that have called blk_set_pm_only(). If this 429 * counter is above zero then only RQF_PM and RQF_PREEMPT requests are 430 * processed. 431 */ 432 atomic_t pm_only; 433 434 /* 435 * ida allocated id for this queue. Used to index queues from 436 * ioctx. 437 */ 438 int id; 439 440 /* 441 * queue needs bounce pages for pages above this limit 442 */ 443 gfp_t bounce_gfp; 444 445 spinlock_t queue_lock; 446 447 /* 448 * queue kobject 449 */ 450 struct kobject kobj; 451 452 /* 453 * mq queue kobject 454 */ 455 struct kobject *mq_kobj; 456 457 #ifdef CONFIG_BLK_DEV_INTEGRITY 458 struct blk_integrity integrity; 459 #endif /* CONFIG_BLK_DEV_INTEGRITY */ 460 461 #ifdef CONFIG_PM 462 struct device *dev; 463 int rpm_status; 464 unsigned int nr_pending; 465 #endif 466 467 /* 468 * queue settings 469 */ 470 unsigned long nr_requests; /* Max # of requests */ 471 472 unsigned int dma_drain_size; 473 void *dma_drain_buffer; 474 unsigned int dma_pad_mask; 475 unsigned int dma_alignment; 476 477 unsigned int rq_timeout; 478 int poll_nsec; 479 480 struct blk_stat_callback *poll_cb; 481 struct blk_rq_stat poll_stat[BLK_MQ_POLL_STATS_BKTS]; 482 483 struct timer_list timeout; 484 struct work_struct timeout_work; 485 486 struct list_head icq_list; 487 #ifdef CONFIG_BLK_CGROUP 488 DECLARE_BITMAP (blkcg_pols, BLKCG_MAX_POLS); 489 struct blkcg_gq *root_blkg; 490 struct list_head blkg_list; 491 #endif 492 493 struct queue_limits limits; 494 495 unsigned int required_elevator_features; 496 497 #ifdef CONFIG_BLK_DEV_ZONED 498 /* 499 * Zoned block device information for request dispatch control. 500 * nr_zones is the total number of zones of the device. This is always 501 * 0 for regular block devices. conv_zones_bitmap is a bitmap of nr_zones 502 * bits which indicates if a zone is conventional (bit set) or 503 * sequential (bit clear). seq_zones_wlock is a bitmap of nr_zones 504 * bits which indicates if a zone is write locked, that is, if a write 505 * request targeting the zone was dispatched. All three fields are 506 * initialized by the low level device driver (e.g. scsi/sd.c). 507 * Stacking drivers (device mappers) may or may not initialize 508 * these fields. 509 * 510 * Reads of this information must be protected with blk_queue_enter() / 511 * blk_queue_exit(). Modifying this information is only allowed while 512 * no requests are being processed. See also blk_mq_freeze_queue() and 513 * blk_mq_unfreeze_queue(). 514 */ 515 unsigned int nr_zones; 516 unsigned long *conv_zones_bitmap; 517 unsigned long *seq_zones_wlock; 518 #endif /* CONFIG_BLK_DEV_ZONED */ 519 520 /* 521 * sg stuff 522 */ 523 unsigned int sg_timeout; 524 unsigned int sg_reserved_size; 525 int node; 526 #ifdef CONFIG_BLK_DEV_IO_TRACE 527 struct blk_trace __rcu *blk_trace; 528 struct mutex blk_trace_mutex; 529 #endif 530 /* 531 * for flush operations 532 */ 533 struct blk_flush_queue *fq; 534 535 struct list_head requeue_list; 536 spinlock_t requeue_lock; 537 struct delayed_work requeue_work; 538 539 struct mutex sysfs_lock; 540 struct mutex sysfs_dir_lock; 541 542 /* 543 * for reusing dead hctx instance in case of updating 544 * nr_hw_queues 545 */ 546 struct list_head unused_hctx_list; 547 spinlock_t unused_hctx_lock; 548 549 int mq_freeze_depth; 550 551 #if defined(CONFIG_BLK_DEV_BSG) 552 struct bsg_class_device bsg_dev; 553 #endif 554 555 #ifdef CONFIG_BLK_DEV_THROTTLING 556 /* Throttle data */ 557 struct throtl_data *td; 558 #endif 559 struct rcu_head rcu_head; 560 wait_queue_head_t mq_freeze_wq; 561 /* 562 * Protect concurrent access to q_usage_counter by 563 * percpu_ref_kill() and percpu_ref_reinit(). 564 */ 565 struct mutex mq_freeze_lock; 566 struct percpu_ref q_usage_counter; 567 568 struct blk_mq_tag_set *tag_set; 569 struct list_head tag_set_list; 570 struct bio_set bio_split; 571 572 #ifdef CONFIG_BLK_DEBUG_FS 573 struct dentry *debugfs_dir; 574 struct dentry *sched_debugfs_dir; 575 struct dentry *rqos_debugfs_dir; 576 #endif 577 578 bool mq_sysfs_init_done; 579 580 size_t cmd_size; 581 582 struct work_struct release_work; 583 584 #define BLK_MAX_WRITE_HINTS 5 585 u64 write_hints[BLK_MAX_WRITE_HINTS]; 586 }; 587 588 #define QUEUE_FLAG_STOPPED 0 /* queue is stopped */ 589 #define QUEUE_FLAG_DYING 1 /* queue being torn down */ 590 #define QUEUE_FLAG_NOMERGES 3 /* disable merge attempts */ 591 #define QUEUE_FLAG_SAME_COMP 4 /* complete on same CPU-group */ 592 #define QUEUE_FLAG_FAIL_IO 5 /* fake timeout */ 593 #define QUEUE_FLAG_NONROT 6 /* non-rotational device (SSD) */ 594 #define QUEUE_FLAG_VIRT QUEUE_FLAG_NONROT /* paravirt device */ 595 #define QUEUE_FLAG_IO_STAT 7 /* do disk/partitions IO accounting */ 596 #define QUEUE_FLAG_DISCARD 8 /* supports DISCARD */ 597 #define QUEUE_FLAG_NOXMERGES 9 /* No extended merges */ 598 #define QUEUE_FLAG_ADD_RANDOM 10 /* Contributes to random pool */ 599 #define QUEUE_FLAG_SECERASE 11 /* supports secure erase */ 600 #define QUEUE_FLAG_SAME_FORCE 12 /* force complete on same CPU */ 601 #define QUEUE_FLAG_DEAD 13 /* queue tear-down finished */ 602 #define QUEUE_FLAG_INIT_DONE 14 /* queue is initialized */ 603 #define QUEUE_FLAG_POLL 16 /* IO polling enabled if set */ 604 #define QUEUE_FLAG_WC 17 /* Write back caching */ 605 #define QUEUE_FLAG_FUA 18 /* device supports FUA writes */ 606 #define QUEUE_FLAG_DAX 19 /* device supports DAX */ 607 #define QUEUE_FLAG_STATS 20 /* track IO start and completion times */ 608 #define QUEUE_FLAG_POLL_STATS 21 /* collecting stats for hybrid polling */ 609 #define QUEUE_FLAG_REGISTERED 22 /* queue has been registered to a disk */ 610 #define QUEUE_FLAG_SCSI_PASSTHROUGH 23 /* queue supports SCSI commands */ 611 #define QUEUE_FLAG_QUIESCED 24 /* queue has been quiesced */ 612 #define QUEUE_FLAG_PCI_P2PDMA 25 /* device supports PCI p2p requests */ 613 #define QUEUE_FLAG_ZONE_RESETALL 26 /* supports Zone Reset All */ 614 #define QUEUE_FLAG_RQ_ALLOC_TIME 27 /* record rq->alloc_time_ns */ 615 616 #define QUEUE_FLAG_MQ_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \ 617 (1 << QUEUE_FLAG_SAME_COMP)) 618 619 void blk_queue_flag_set(unsigned int flag, struct request_queue *q); 620 void blk_queue_flag_clear(unsigned int flag, struct request_queue *q); 621 bool blk_queue_flag_test_and_set(unsigned int flag, struct request_queue *q); 622 623 #define blk_queue_stopped(q) test_bit(QUEUE_FLAG_STOPPED, &(q)->queue_flags) 624 #define blk_queue_dying(q) test_bit(QUEUE_FLAG_DYING, &(q)->queue_flags) 625 #define blk_queue_dead(q) test_bit(QUEUE_FLAG_DEAD, &(q)->queue_flags) 626 #define blk_queue_init_done(q) test_bit(QUEUE_FLAG_INIT_DONE, &(q)->queue_flags) 627 #define blk_queue_nomerges(q) test_bit(QUEUE_FLAG_NOMERGES, &(q)->queue_flags) 628 #define blk_queue_noxmerges(q) \ 629 test_bit(QUEUE_FLAG_NOXMERGES, &(q)->queue_flags) 630 #define blk_queue_nonrot(q) test_bit(QUEUE_FLAG_NONROT, &(q)->queue_flags) 631 #define blk_queue_io_stat(q) test_bit(QUEUE_FLAG_IO_STAT, &(q)->queue_flags) 632 #define blk_queue_add_random(q) test_bit(QUEUE_FLAG_ADD_RANDOM, &(q)->queue_flags) 633 #define blk_queue_discard(q) test_bit(QUEUE_FLAG_DISCARD, &(q)->queue_flags) 634 #define blk_queue_zone_resetall(q) \ 635 test_bit(QUEUE_FLAG_ZONE_RESETALL, &(q)->queue_flags) 636 #define blk_queue_secure_erase(q) \ 637 (test_bit(QUEUE_FLAG_SECERASE, &(q)->queue_flags)) 638 #define blk_queue_dax(q) test_bit(QUEUE_FLAG_DAX, &(q)->queue_flags) 639 #define blk_queue_scsi_passthrough(q) \ 640 test_bit(QUEUE_FLAG_SCSI_PASSTHROUGH, &(q)->queue_flags) 641 #define blk_queue_pci_p2pdma(q) \ 642 test_bit(QUEUE_FLAG_PCI_P2PDMA, &(q)->queue_flags) 643 #ifdef CONFIG_BLK_RQ_ALLOC_TIME 644 #define blk_queue_rq_alloc_time(q) \ 645 test_bit(QUEUE_FLAG_RQ_ALLOC_TIME, &(q)->queue_flags) 646 #else 647 #define blk_queue_rq_alloc_time(q) false 648 #endif 649 650 #define blk_noretry_request(rq) \ 651 ((rq)->cmd_flags & (REQ_FAILFAST_DEV|REQ_FAILFAST_TRANSPORT| \ 652 REQ_FAILFAST_DRIVER)) 653 #define blk_queue_quiesced(q) test_bit(QUEUE_FLAG_QUIESCED, &(q)->queue_flags) 654 #define blk_queue_pm_only(q) atomic_read(&(q)->pm_only) 655 #define blk_queue_fua(q) test_bit(QUEUE_FLAG_FUA, &(q)->queue_flags) 656 #define blk_queue_registered(q) test_bit(QUEUE_FLAG_REGISTERED, &(q)->queue_flags) 657 658 extern void blk_set_pm_only(struct request_queue *q); 659 extern void blk_clear_pm_only(struct request_queue *q); 660 661 static inline bool blk_account_rq(struct request *rq) 662 { 663 return (rq->rq_flags & RQF_STARTED) && !blk_rq_is_passthrough(rq); 664 } 665 666 #define list_entry_rq(ptr) list_entry((ptr), struct request, queuelist) 667 668 #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ) 669 670 #define rq_dma_dir(rq) \ 671 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE) 672 673 #define dma_map_bvec(dev, bv, dir, attrs) \ 674 dma_map_page_attrs(dev, (bv)->bv_page, (bv)->bv_offset, (bv)->bv_len, \ 675 (dir), (attrs)) 676 677 static inline bool queue_is_mq(struct request_queue *q) 678 { 679 return q->mq_ops; 680 } 681 682 static inline enum blk_zoned_model 683 blk_queue_zoned_model(struct request_queue *q) 684 { 685 return q->limits.zoned; 686 } 687 688 static inline bool blk_queue_is_zoned(struct request_queue *q) 689 { 690 switch (blk_queue_zoned_model(q)) { 691 case BLK_ZONED_HA: 692 case BLK_ZONED_HM: 693 return true; 694 default: 695 return false; 696 } 697 } 698 699 static inline sector_t blk_queue_zone_sectors(struct request_queue *q) 700 { 701 return blk_queue_is_zoned(q) ? q->limits.chunk_sectors : 0; 702 } 703 704 #ifdef CONFIG_BLK_DEV_ZONED 705 static inline unsigned int blk_queue_nr_zones(struct request_queue *q) 706 { 707 return blk_queue_is_zoned(q) ? q->nr_zones : 0; 708 } 709 710 static inline unsigned int blk_queue_zone_no(struct request_queue *q, 711 sector_t sector) 712 { 713 if (!blk_queue_is_zoned(q)) 714 return 0; 715 return sector >> ilog2(q->limits.chunk_sectors); 716 } 717 718 static inline bool blk_queue_zone_is_seq(struct request_queue *q, 719 sector_t sector) 720 { 721 if (!blk_queue_is_zoned(q)) 722 return false; 723 if (!q->conv_zones_bitmap) 724 return true; 725 return !test_bit(blk_queue_zone_no(q, sector), q->conv_zones_bitmap); 726 } 727 #else /* CONFIG_BLK_DEV_ZONED */ 728 static inline unsigned int blk_queue_nr_zones(struct request_queue *q) 729 { 730 return 0; 731 } 732 #endif /* CONFIG_BLK_DEV_ZONED */ 733 734 static inline bool rq_is_sync(struct request *rq) 735 { 736 return op_is_sync(rq->cmd_flags); 737 } 738 739 static inline bool rq_mergeable(struct request *rq) 740 { 741 if (blk_rq_is_passthrough(rq)) 742 return false; 743 744 if (req_op(rq) == REQ_OP_FLUSH) 745 return false; 746 747 if (req_op(rq) == REQ_OP_WRITE_ZEROES) 748 return false; 749 750 if (rq->cmd_flags & REQ_NOMERGE_FLAGS) 751 return false; 752 if (rq->rq_flags & RQF_NOMERGE_FLAGS) 753 return false; 754 755 return true; 756 } 757 758 static inline bool blk_write_same_mergeable(struct bio *a, struct bio *b) 759 { 760 if (bio_page(a) == bio_page(b) && 761 bio_offset(a) == bio_offset(b)) 762 return true; 763 764 return false; 765 } 766 767 static inline unsigned int blk_queue_depth(struct request_queue *q) 768 { 769 if (q->queue_depth) 770 return q->queue_depth; 771 772 return q->nr_requests; 773 } 774 775 extern unsigned long blk_max_low_pfn, blk_max_pfn; 776 777 /* 778 * standard bounce addresses: 779 * 780 * BLK_BOUNCE_HIGH : bounce all highmem pages 781 * BLK_BOUNCE_ANY : don't bounce anything 782 * BLK_BOUNCE_ISA : bounce pages above ISA DMA boundary 783 */ 784 785 #if BITS_PER_LONG == 32 786 #define BLK_BOUNCE_HIGH ((u64)blk_max_low_pfn << PAGE_SHIFT) 787 #else 788 #define BLK_BOUNCE_HIGH -1ULL 789 #endif 790 #define BLK_BOUNCE_ANY (-1ULL) 791 #define BLK_BOUNCE_ISA (DMA_BIT_MASK(24)) 792 793 /* 794 * default timeout for SG_IO if none specified 795 */ 796 #define BLK_DEFAULT_SG_TIMEOUT (60 * HZ) 797 #define BLK_MIN_SG_TIMEOUT (7 * HZ) 798 799 struct rq_map_data { 800 struct page **pages; 801 int page_order; 802 int nr_entries; 803 unsigned long offset; 804 int null_mapped; 805 int from_user; 806 }; 807 808 struct req_iterator { 809 struct bvec_iter iter; 810 struct bio *bio; 811 }; 812 813 /* This should not be used directly - use rq_for_each_segment */ 814 #define for_each_bio(_bio) \ 815 for (; _bio; _bio = _bio->bi_next) 816 #define __rq_for_each_bio(_bio, rq) \ 817 if ((rq->bio)) \ 818 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next) 819 820 #define rq_for_each_segment(bvl, _rq, _iter) \ 821 __rq_for_each_bio(_iter.bio, _rq) \ 822 bio_for_each_segment(bvl, _iter.bio, _iter.iter) 823 824 #define rq_for_each_bvec(bvl, _rq, _iter) \ 825 __rq_for_each_bio(_iter.bio, _rq) \ 826 bio_for_each_bvec(bvl, _iter.bio, _iter.iter) 827 828 #define rq_iter_last(bvec, _iter) \ 829 (_iter.bio->bi_next == NULL && \ 830 bio_iter_last(bvec, _iter.iter)) 831 832 #ifndef ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 833 # error "You should define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE for your platform" 834 #endif 835 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 836 extern void rq_flush_dcache_pages(struct request *rq); 837 #else 838 static inline void rq_flush_dcache_pages(struct request *rq) 839 { 840 } 841 #endif 842 843 extern int blk_register_queue(struct gendisk *disk); 844 extern void blk_unregister_queue(struct gendisk *disk); 845 extern blk_qc_t generic_make_request(struct bio *bio); 846 extern blk_qc_t direct_make_request(struct bio *bio); 847 extern void blk_rq_init(struct request_queue *q, struct request *rq); 848 extern void blk_put_request(struct request *); 849 extern struct request *blk_get_request(struct request_queue *, unsigned int op, 850 blk_mq_req_flags_t flags); 851 extern int blk_lld_busy(struct request_queue *q); 852 extern int blk_rq_prep_clone(struct request *rq, struct request *rq_src, 853 struct bio_set *bs, gfp_t gfp_mask, 854 int (*bio_ctr)(struct bio *, struct bio *, void *), 855 void *data); 856 extern void blk_rq_unprep_clone(struct request *rq); 857 extern blk_status_t blk_insert_cloned_request(struct request_queue *q, 858 struct request *rq); 859 extern int blk_rq_append_bio(struct request *rq, struct bio **bio); 860 extern void blk_queue_split(struct request_queue *, struct bio **); 861 extern int scsi_verify_blk_ioctl(struct block_device *, unsigned int); 862 extern int scsi_cmd_blk_ioctl(struct block_device *, fmode_t, 863 unsigned int, void __user *); 864 extern int scsi_cmd_ioctl(struct request_queue *, struct gendisk *, fmode_t, 865 unsigned int, void __user *); 866 extern int sg_scsi_ioctl(struct request_queue *, struct gendisk *, fmode_t, 867 struct scsi_ioctl_command __user *); 868 extern int get_sg_io_hdr(struct sg_io_hdr *hdr, const void __user *argp); 869 extern int put_sg_io_hdr(const struct sg_io_hdr *hdr, void __user *argp); 870 871 extern int blk_queue_enter(struct request_queue *q, blk_mq_req_flags_t flags); 872 extern void blk_queue_exit(struct request_queue *q); 873 extern void blk_sync_queue(struct request_queue *q); 874 extern int blk_rq_map_user(struct request_queue *, struct request *, 875 struct rq_map_data *, void __user *, unsigned long, 876 gfp_t); 877 extern int blk_rq_unmap_user(struct bio *); 878 extern int blk_rq_map_kern(struct request_queue *, struct request *, void *, unsigned int, gfp_t); 879 extern int blk_rq_map_user_iov(struct request_queue *, struct request *, 880 struct rq_map_data *, const struct iov_iter *, 881 gfp_t); 882 extern void blk_execute_rq(struct request_queue *, struct gendisk *, 883 struct request *, int); 884 extern void blk_execute_rq_nowait(struct request_queue *, struct gendisk *, 885 struct request *, int, rq_end_io_fn *); 886 887 /* Helper to convert REQ_OP_XXX to its string format XXX */ 888 extern const char *blk_op_str(unsigned int op); 889 890 int blk_status_to_errno(blk_status_t status); 891 blk_status_t errno_to_blk_status(int errno); 892 893 int blk_poll(struct request_queue *q, blk_qc_t cookie, bool spin); 894 895 static inline struct request_queue *bdev_get_queue(struct block_device *bdev) 896 { 897 return bdev->bd_disk->queue; /* this is never NULL */ 898 } 899 900 /* 901 * The basic unit of block I/O is a sector. It is used in a number of contexts 902 * in Linux (blk, bio, genhd). The size of one sector is 512 = 2**9 903 * bytes. Variables of type sector_t represent an offset or size that is a 904 * multiple of 512 bytes. Hence these two constants. 905 */ 906 #ifndef SECTOR_SHIFT 907 #define SECTOR_SHIFT 9 908 #endif 909 #ifndef SECTOR_SIZE 910 #define SECTOR_SIZE (1 << SECTOR_SHIFT) 911 #endif 912 913 /* 914 * blk_rq_pos() : the current sector 915 * blk_rq_bytes() : bytes left in the entire request 916 * blk_rq_cur_bytes() : bytes left in the current segment 917 * blk_rq_err_bytes() : bytes left till the next error boundary 918 * blk_rq_sectors() : sectors left in the entire request 919 * blk_rq_cur_sectors() : sectors left in the current segment 920 * blk_rq_stats_sectors() : sectors of the entire request used for stats 921 */ 922 static inline sector_t blk_rq_pos(const struct request *rq) 923 { 924 return rq->__sector; 925 } 926 927 static inline unsigned int blk_rq_bytes(const struct request *rq) 928 { 929 return rq->__data_len; 930 } 931 932 static inline int blk_rq_cur_bytes(const struct request *rq) 933 { 934 return rq->bio ? bio_cur_bytes(rq->bio) : 0; 935 } 936 937 extern unsigned int blk_rq_err_bytes(const struct request *rq); 938 939 static inline unsigned int blk_rq_sectors(const struct request *rq) 940 { 941 return blk_rq_bytes(rq) >> SECTOR_SHIFT; 942 } 943 944 static inline unsigned int blk_rq_cur_sectors(const struct request *rq) 945 { 946 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT; 947 } 948 949 static inline unsigned int blk_rq_stats_sectors(const struct request *rq) 950 { 951 return rq->stats_sectors; 952 } 953 954 #ifdef CONFIG_BLK_DEV_ZONED 955 956 /* Helper to convert BLK_ZONE_ZONE_XXX to its string format XXX */ 957 const char *blk_zone_cond_str(enum blk_zone_cond zone_cond); 958 959 static inline unsigned int blk_rq_zone_no(struct request *rq) 960 { 961 return blk_queue_zone_no(rq->q, blk_rq_pos(rq)); 962 } 963 964 static inline unsigned int blk_rq_zone_is_seq(struct request *rq) 965 { 966 return blk_queue_zone_is_seq(rq->q, blk_rq_pos(rq)); 967 } 968 #endif /* CONFIG_BLK_DEV_ZONED */ 969 970 /* 971 * Some commands like WRITE SAME have a payload or data transfer size which 972 * is different from the size of the request. Any driver that supports such 973 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to 974 * calculate the data transfer size. 975 */ 976 static inline unsigned int blk_rq_payload_bytes(struct request *rq) 977 { 978 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 979 return rq->special_vec.bv_len; 980 return blk_rq_bytes(rq); 981 } 982 983 /* 984 * Return the first full biovec in the request. The caller needs to check that 985 * there are any bvecs before calling this helper. 986 */ 987 static inline struct bio_vec req_bvec(struct request *rq) 988 { 989 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 990 return rq->special_vec; 991 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter); 992 } 993 994 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q, 995 int op) 996 { 997 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)) 998 return min(q->limits.max_discard_sectors, 999 UINT_MAX >> SECTOR_SHIFT); 1000 1001 if (unlikely(op == REQ_OP_WRITE_SAME)) 1002 return q->limits.max_write_same_sectors; 1003 1004 if (unlikely(op == REQ_OP_WRITE_ZEROES)) 1005 return q->limits.max_write_zeroes_sectors; 1006 1007 return q->limits.max_sectors; 1008 } 1009 1010 /* 1011 * Return maximum size of a request at given offset. Only valid for 1012 * file system requests. 1013 */ 1014 static inline unsigned int blk_max_size_offset(struct request_queue *q, 1015 sector_t offset) 1016 { 1017 if (!q->limits.chunk_sectors) 1018 return q->limits.max_sectors; 1019 1020 return min(q->limits.max_sectors, (unsigned int)(q->limits.chunk_sectors - 1021 (offset & (q->limits.chunk_sectors - 1)))); 1022 } 1023 1024 static inline unsigned int blk_rq_get_max_sectors(struct request *rq, 1025 sector_t offset) 1026 { 1027 struct request_queue *q = rq->q; 1028 1029 if (blk_rq_is_passthrough(rq)) 1030 return q->limits.max_hw_sectors; 1031 1032 if (!q->limits.chunk_sectors || 1033 req_op(rq) == REQ_OP_DISCARD || 1034 req_op(rq) == REQ_OP_SECURE_ERASE) 1035 return blk_queue_get_max_sectors(q, req_op(rq)); 1036 1037 return min(blk_max_size_offset(q, offset), 1038 blk_queue_get_max_sectors(q, req_op(rq))); 1039 } 1040 1041 static inline unsigned int blk_rq_count_bios(struct request *rq) 1042 { 1043 unsigned int nr_bios = 0; 1044 struct bio *bio; 1045 1046 __rq_for_each_bio(bio, rq) 1047 nr_bios++; 1048 1049 return nr_bios; 1050 } 1051 1052 void blk_steal_bios(struct bio_list *list, struct request *rq); 1053 1054 /* 1055 * Request completion related functions. 1056 * 1057 * blk_update_request() completes given number of bytes and updates 1058 * the request without completing it. 1059 */ 1060 extern bool blk_update_request(struct request *rq, blk_status_t error, 1061 unsigned int nr_bytes); 1062 1063 extern void __blk_complete_request(struct request *); 1064 extern void blk_abort_request(struct request *); 1065 1066 /* 1067 * Access functions for manipulating queue properties 1068 */ 1069 extern void blk_cleanup_queue(struct request_queue *); 1070 extern void blk_queue_bounce_limit(struct request_queue *, u64); 1071 extern void blk_queue_max_hw_sectors(struct request_queue *, unsigned int); 1072 extern void blk_queue_chunk_sectors(struct request_queue *, unsigned int); 1073 extern void blk_queue_max_segments(struct request_queue *, unsigned short); 1074 extern void blk_queue_max_discard_segments(struct request_queue *, 1075 unsigned short); 1076 extern void blk_queue_max_segment_size(struct request_queue *, unsigned int); 1077 extern void blk_queue_max_discard_sectors(struct request_queue *q, 1078 unsigned int max_discard_sectors); 1079 extern void blk_queue_max_write_same_sectors(struct request_queue *q, 1080 unsigned int max_write_same_sectors); 1081 extern void blk_queue_max_write_zeroes_sectors(struct request_queue *q, 1082 unsigned int max_write_same_sectors); 1083 extern void blk_queue_logical_block_size(struct request_queue *, unsigned int); 1084 extern void blk_queue_physical_block_size(struct request_queue *, unsigned int); 1085 extern void blk_queue_alignment_offset(struct request_queue *q, 1086 unsigned int alignment); 1087 extern void blk_limits_io_min(struct queue_limits *limits, unsigned int min); 1088 extern void blk_queue_io_min(struct request_queue *q, unsigned int min); 1089 extern void blk_limits_io_opt(struct queue_limits *limits, unsigned int opt); 1090 extern void blk_queue_io_opt(struct request_queue *q, unsigned int opt); 1091 extern void blk_set_queue_depth(struct request_queue *q, unsigned int depth); 1092 extern void blk_set_default_limits(struct queue_limits *lim); 1093 extern void blk_set_stacking_limits(struct queue_limits *lim); 1094 extern int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, 1095 sector_t offset); 1096 extern int bdev_stack_limits(struct queue_limits *t, struct block_device *bdev, 1097 sector_t offset); 1098 extern void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, 1099 sector_t offset); 1100 extern void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b); 1101 extern void blk_queue_update_dma_pad(struct request_queue *, unsigned int); 1102 extern int blk_queue_dma_drain(struct request_queue *q, 1103 dma_drain_needed_fn *dma_drain_needed, 1104 void *buf, unsigned int size); 1105 extern void blk_queue_segment_boundary(struct request_queue *, unsigned long); 1106 extern void blk_queue_virt_boundary(struct request_queue *, unsigned long); 1107 extern void blk_queue_dma_alignment(struct request_queue *, int); 1108 extern void blk_queue_update_dma_alignment(struct request_queue *, int); 1109 extern void blk_queue_rq_timeout(struct request_queue *, unsigned int); 1110 extern void blk_queue_write_cache(struct request_queue *q, bool enabled, bool fua); 1111 extern void blk_queue_required_elevator_features(struct request_queue *q, 1112 unsigned int features); 1113 extern bool blk_queue_can_use_dma_map_merging(struct request_queue *q, 1114 struct device *dev); 1115 1116 /* 1117 * Number of physical segments as sent to the device. 1118 * 1119 * Normally this is the number of discontiguous data segments sent by the 1120 * submitter. But for data-less command like discard we might have no 1121 * actual data segments submitted, but the driver might have to add it's 1122 * own special payload. In that case we still return 1 here so that this 1123 * special payload will be mapped. 1124 */ 1125 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq) 1126 { 1127 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 1128 return 1; 1129 return rq->nr_phys_segments; 1130 } 1131 1132 /* 1133 * Number of discard segments (or ranges) the driver needs to fill in. 1134 * Each discard bio merged into a request is counted as one segment. 1135 */ 1136 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq) 1137 { 1138 return max_t(unsigned short, rq->nr_phys_segments, 1); 1139 } 1140 1141 extern int blk_rq_map_sg(struct request_queue *, struct request *, struct scatterlist *); 1142 extern void blk_dump_rq_flags(struct request *, char *); 1143 extern long nr_blockdev_pages(void); 1144 1145 bool __must_check blk_get_queue(struct request_queue *); 1146 struct request_queue *blk_alloc_queue(make_request_fn make_request, int node_id); 1147 extern void blk_put_queue(struct request_queue *); 1148 extern void blk_set_queue_dying(struct request_queue *); 1149 1150 /* 1151 * blk_plug permits building a queue of related requests by holding the I/O 1152 * fragments for a short period. This allows merging of sequential requests 1153 * into single larger request. As the requests are moved from a per-task list to 1154 * the device's request_queue in a batch, this results in improved scalability 1155 * as the lock contention for request_queue lock is reduced. 1156 * 1157 * It is ok not to disable preemption when adding the request to the plug list 1158 * or when attempting a merge, because blk_schedule_flush_list() will only flush 1159 * the plug list when the task sleeps by itself. For details, please see 1160 * schedule() where blk_schedule_flush_plug() is called. 1161 */ 1162 struct blk_plug { 1163 struct list_head mq_list; /* blk-mq requests */ 1164 struct list_head cb_list; /* md requires an unplug callback */ 1165 unsigned short rq_count; 1166 bool multiple_queues; 1167 }; 1168 #define BLK_MAX_REQUEST_COUNT 16 1169 #define BLK_PLUG_FLUSH_SIZE (128 * 1024) 1170 1171 struct blk_plug_cb; 1172 typedef void (*blk_plug_cb_fn)(struct blk_plug_cb *, bool); 1173 struct blk_plug_cb { 1174 struct list_head list; 1175 blk_plug_cb_fn callback; 1176 void *data; 1177 }; 1178 extern struct blk_plug_cb *blk_check_plugged(blk_plug_cb_fn unplug, 1179 void *data, int size); 1180 extern void blk_start_plug(struct blk_plug *); 1181 extern void blk_finish_plug(struct blk_plug *); 1182 extern void blk_flush_plug_list(struct blk_plug *, bool); 1183 1184 static inline void blk_flush_plug(struct task_struct *tsk) 1185 { 1186 struct blk_plug *plug = tsk->plug; 1187 1188 if (plug) 1189 blk_flush_plug_list(plug, false); 1190 } 1191 1192 static inline void blk_schedule_flush_plug(struct task_struct *tsk) 1193 { 1194 struct blk_plug *plug = tsk->plug; 1195 1196 if (plug) 1197 blk_flush_plug_list(plug, true); 1198 } 1199 1200 static inline bool blk_needs_flush_plug(struct task_struct *tsk) 1201 { 1202 struct blk_plug *plug = tsk->plug; 1203 1204 return plug && 1205 (!list_empty(&plug->mq_list) || 1206 !list_empty(&plug->cb_list)); 1207 } 1208 1209 extern int blkdev_issue_flush(struct block_device *, gfp_t, sector_t *); 1210 extern int blkdev_issue_write_same(struct block_device *bdev, sector_t sector, 1211 sector_t nr_sects, gfp_t gfp_mask, struct page *page); 1212 1213 #define BLKDEV_DISCARD_SECURE (1 << 0) /* issue a secure erase */ 1214 1215 extern int blkdev_issue_discard(struct block_device *bdev, sector_t sector, 1216 sector_t nr_sects, gfp_t gfp_mask, unsigned long flags); 1217 extern int __blkdev_issue_discard(struct block_device *bdev, sector_t sector, 1218 sector_t nr_sects, gfp_t gfp_mask, int flags, 1219 struct bio **biop); 1220 1221 #define BLKDEV_ZERO_NOUNMAP (1 << 0) /* do not free blocks */ 1222 #define BLKDEV_ZERO_NOFALLBACK (1 << 1) /* don't write explicit zeroes */ 1223 1224 extern int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector, 1225 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop, 1226 unsigned flags); 1227 extern int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector, 1228 sector_t nr_sects, gfp_t gfp_mask, unsigned flags); 1229 1230 static inline int sb_issue_discard(struct super_block *sb, sector_t block, 1231 sector_t nr_blocks, gfp_t gfp_mask, unsigned long flags) 1232 { 1233 return blkdev_issue_discard(sb->s_bdev, 1234 block << (sb->s_blocksize_bits - 1235 SECTOR_SHIFT), 1236 nr_blocks << (sb->s_blocksize_bits - 1237 SECTOR_SHIFT), 1238 gfp_mask, flags); 1239 } 1240 static inline int sb_issue_zeroout(struct super_block *sb, sector_t block, 1241 sector_t nr_blocks, gfp_t gfp_mask) 1242 { 1243 return blkdev_issue_zeroout(sb->s_bdev, 1244 block << (sb->s_blocksize_bits - 1245 SECTOR_SHIFT), 1246 nr_blocks << (sb->s_blocksize_bits - 1247 SECTOR_SHIFT), 1248 gfp_mask, 0); 1249 } 1250 1251 extern int blk_verify_command(unsigned char *cmd, fmode_t mode); 1252 1253 enum blk_default_limits { 1254 BLK_MAX_SEGMENTS = 128, 1255 BLK_SAFE_MAX_SECTORS = 255, 1256 BLK_DEF_MAX_SECTORS = 2560, 1257 BLK_MAX_SEGMENT_SIZE = 65536, 1258 BLK_SEG_BOUNDARY_MASK = 0xFFFFFFFFUL, 1259 }; 1260 1261 static inline unsigned long queue_segment_boundary(const struct request_queue *q) 1262 { 1263 return q->limits.seg_boundary_mask; 1264 } 1265 1266 static inline unsigned long queue_virt_boundary(const struct request_queue *q) 1267 { 1268 return q->limits.virt_boundary_mask; 1269 } 1270 1271 static inline unsigned int queue_max_sectors(const struct request_queue *q) 1272 { 1273 return q->limits.max_sectors; 1274 } 1275 1276 static inline unsigned int queue_max_hw_sectors(const struct request_queue *q) 1277 { 1278 return q->limits.max_hw_sectors; 1279 } 1280 1281 static inline unsigned short queue_max_segments(const struct request_queue *q) 1282 { 1283 return q->limits.max_segments; 1284 } 1285 1286 static inline unsigned short queue_max_discard_segments(const struct request_queue *q) 1287 { 1288 return q->limits.max_discard_segments; 1289 } 1290 1291 static inline unsigned int queue_max_segment_size(const struct request_queue *q) 1292 { 1293 return q->limits.max_segment_size; 1294 } 1295 1296 static inline unsigned queue_logical_block_size(const struct request_queue *q) 1297 { 1298 int retval = 512; 1299 1300 if (q && q->limits.logical_block_size) 1301 retval = q->limits.logical_block_size; 1302 1303 return retval; 1304 } 1305 1306 static inline unsigned int bdev_logical_block_size(struct block_device *bdev) 1307 { 1308 return queue_logical_block_size(bdev_get_queue(bdev)); 1309 } 1310 1311 static inline unsigned int queue_physical_block_size(const struct request_queue *q) 1312 { 1313 return q->limits.physical_block_size; 1314 } 1315 1316 static inline unsigned int bdev_physical_block_size(struct block_device *bdev) 1317 { 1318 return queue_physical_block_size(bdev_get_queue(bdev)); 1319 } 1320 1321 static inline unsigned int queue_io_min(const struct request_queue *q) 1322 { 1323 return q->limits.io_min; 1324 } 1325 1326 static inline int bdev_io_min(struct block_device *bdev) 1327 { 1328 return queue_io_min(bdev_get_queue(bdev)); 1329 } 1330 1331 static inline unsigned int queue_io_opt(const struct request_queue *q) 1332 { 1333 return q->limits.io_opt; 1334 } 1335 1336 static inline int bdev_io_opt(struct block_device *bdev) 1337 { 1338 return queue_io_opt(bdev_get_queue(bdev)); 1339 } 1340 1341 static inline int queue_alignment_offset(const struct request_queue *q) 1342 { 1343 if (q->limits.misaligned) 1344 return -1; 1345 1346 return q->limits.alignment_offset; 1347 } 1348 1349 static inline int queue_limit_alignment_offset(struct queue_limits *lim, sector_t sector) 1350 { 1351 unsigned int granularity = max(lim->physical_block_size, lim->io_min); 1352 unsigned int alignment = sector_div(sector, granularity >> SECTOR_SHIFT) 1353 << SECTOR_SHIFT; 1354 1355 return (granularity + lim->alignment_offset - alignment) % granularity; 1356 } 1357 1358 static inline int bdev_alignment_offset(struct block_device *bdev) 1359 { 1360 struct request_queue *q = bdev_get_queue(bdev); 1361 1362 if (q->limits.misaligned) 1363 return -1; 1364 1365 if (bdev != bdev->bd_contains) 1366 return bdev->bd_part->alignment_offset; 1367 1368 return q->limits.alignment_offset; 1369 } 1370 1371 static inline int queue_discard_alignment(const struct request_queue *q) 1372 { 1373 if (q->limits.discard_misaligned) 1374 return -1; 1375 1376 return q->limits.discard_alignment; 1377 } 1378 1379 static inline int queue_limit_discard_alignment(struct queue_limits *lim, sector_t sector) 1380 { 1381 unsigned int alignment, granularity, offset; 1382 1383 if (!lim->max_discard_sectors) 1384 return 0; 1385 1386 /* Why are these in bytes, not sectors? */ 1387 alignment = lim->discard_alignment >> SECTOR_SHIFT; 1388 granularity = lim->discard_granularity >> SECTOR_SHIFT; 1389 if (!granularity) 1390 return 0; 1391 1392 /* Offset of the partition start in 'granularity' sectors */ 1393 offset = sector_div(sector, granularity); 1394 1395 /* And why do we do this modulus *again* in blkdev_issue_discard()? */ 1396 offset = (granularity + alignment - offset) % granularity; 1397 1398 /* Turn it back into bytes, gaah */ 1399 return offset << SECTOR_SHIFT; 1400 } 1401 1402 static inline int bdev_discard_alignment(struct block_device *bdev) 1403 { 1404 struct request_queue *q = bdev_get_queue(bdev); 1405 1406 if (bdev != bdev->bd_contains) 1407 return bdev->bd_part->discard_alignment; 1408 1409 return q->limits.discard_alignment; 1410 } 1411 1412 static inline unsigned int bdev_write_same(struct block_device *bdev) 1413 { 1414 struct request_queue *q = bdev_get_queue(bdev); 1415 1416 if (q) 1417 return q->limits.max_write_same_sectors; 1418 1419 return 0; 1420 } 1421 1422 static inline unsigned int bdev_write_zeroes_sectors(struct block_device *bdev) 1423 { 1424 struct request_queue *q = bdev_get_queue(bdev); 1425 1426 if (q) 1427 return q->limits.max_write_zeroes_sectors; 1428 1429 return 0; 1430 } 1431 1432 static inline enum blk_zoned_model bdev_zoned_model(struct block_device *bdev) 1433 { 1434 struct request_queue *q = bdev_get_queue(bdev); 1435 1436 if (q) 1437 return blk_queue_zoned_model(q); 1438 1439 return BLK_ZONED_NONE; 1440 } 1441 1442 static inline bool bdev_is_zoned(struct block_device *bdev) 1443 { 1444 struct request_queue *q = bdev_get_queue(bdev); 1445 1446 if (q) 1447 return blk_queue_is_zoned(q); 1448 1449 return false; 1450 } 1451 1452 static inline sector_t bdev_zone_sectors(struct block_device *bdev) 1453 { 1454 struct request_queue *q = bdev_get_queue(bdev); 1455 1456 if (q) 1457 return blk_queue_zone_sectors(q); 1458 return 0; 1459 } 1460 1461 static inline int queue_dma_alignment(const struct request_queue *q) 1462 { 1463 return q ? q->dma_alignment : 511; 1464 } 1465 1466 static inline int blk_rq_aligned(struct request_queue *q, unsigned long addr, 1467 unsigned int len) 1468 { 1469 unsigned int alignment = queue_dma_alignment(q) | q->dma_pad_mask; 1470 return !(addr & alignment) && !(len & alignment); 1471 } 1472 1473 /* assumes size > 256 */ 1474 static inline unsigned int blksize_bits(unsigned int size) 1475 { 1476 unsigned int bits = 8; 1477 do { 1478 bits++; 1479 size >>= 1; 1480 } while (size > 256); 1481 return bits; 1482 } 1483 1484 static inline unsigned int block_size(struct block_device *bdev) 1485 { 1486 return bdev->bd_block_size; 1487 } 1488 1489 int kblockd_schedule_work(struct work_struct *work); 1490 int kblockd_mod_delayed_work_on(int cpu, struct delayed_work *dwork, unsigned long delay); 1491 1492 #define MODULE_ALIAS_BLOCKDEV(major,minor) \ 1493 MODULE_ALIAS("block-major-" __stringify(major) "-" __stringify(minor)) 1494 #define MODULE_ALIAS_BLOCKDEV_MAJOR(major) \ 1495 MODULE_ALIAS("block-major-" __stringify(major) "-*") 1496 1497 #if defined(CONFIG_BLK_DEV_INTEGRITY) 1498 1499 enum blk_integrity_flags { 1500 BLK_INTEGRITY_VERIFY = 1 << 0, 1501 BLK_INTEGRITY_GENERATE = 1 << 1, 1502 BLK_INTEGRITY_DEVICE_CAPABLE = 1 << 2, 1503 BLK_INTEGRITY_IP_CHECKSUM = 1 << 3, 1504 }; 1505 1506 struct blk_integrity_iter { 1507 void *prot_buf; 1508 void *data_buf; 1509 sector_t seed; 1510 unsigned int data_size; 1511 unsigned short interval; 1512 const char *disk_name; 1513 }; 1514 1515 typedef blk_status_t (integrity_processing_fn) (struct blk_integrity_iter *); 1516 typedef void (integrity_prepare_fn) (struct request *); 1517 typedef void (integrity_complete_fn) (struct request *, unsigned int); 1518 1519 struct blk_integrity_profile { 1520 integrity_processing_fn *generate_fn; 1521 integrity_processing_fn *verify_fn; 1522 integrity_prepare_fn *prepare_fn; 1523 integrity_complete_fn *complete_fn; 1524 const char *name; 1525 }; 1526 1527 extern void blk_integrity_register(struct gendisk *, struct blk_integrity *); 1528 extern void blk_integrity_unregister(struct gendisk *); 1529 extern int blk_integrity_compare(struct gendisk *, struct gendisk *); 1530 extern int blk_rq_map_integrity_sg(struct request_queue *, struct bio *, 1531 struct scatterlist *); 1532 extern int blk_rq_count_integrity_sg(struct request_queue *, struct bio *); 1533 extern bool blk_integrity_merge_rq(struct request_queue *, struct request *, 1534 struct request *); 1535 extern bool blk_integrity_merge_bio(struct request_queue *, struct request *, 1536 struct bio *); 1537 1538 static inline struct blk_integrity *blk_get_integrity(struct gendisk *disk) 1539 { 1540 struct blk_integrity *bi = &disk->queue->integrity; 1541 1542 if (!bi->profile) 1543 return NULL; 1544 1545 return bi; 1546 } 1547 1548 static inline 1549 struct blk_integrity *bdev_get_integrity(struct block_device *bdev) 1550 { 1551 return blk_get_integrity(bdev->bd_disk); 1552 } 1553 1554 static inline bool blk_integrity_rq(struct request *rq) 1555 { 1556 return rq->cmd_flags & REQ_INTEGRITY; 1557 } 1558 1559 static inline void blk_queue_max_integrity_segments(struct request_queue *q, 1560 unsigned int segs) 1561 { 1562 q->limits.max_integrity_segments = segs; 1563 } 1564 1565 static inline unsigned short 1566 queue_max_integrity_segments(const struct request_queue *q) 1567 { 1568 return q->limits.max_integrity_segments; 1569 } 1570 1571 /** 1572 * bio_integrity_intervals - Return number of integrity intervals for a bio 1573 * @bi: blk_integrity profile for device 1574 * @sectors: Size of the bio in 512-byte sectors 1575 * 1576 * Description: The block layer calculates everything in 512 byte 1577 * sectors but integrity metadata is done in terms of the data integrity 1578 * interval size of the storage device. Convert the block layer sectors 1579 * to the appropriate number of integrity intervals. 1580 */ 1581 static inline unsigned int bio_integrity_intervals(struct blk_integrity *bi, 1582 unsigned int sectors) 1583 { 1584 return sectors >> (bi->interval_exp - 9); 1585 } 1586 1587 static inline unsigned int bio_integrity_bytes(struct blk_integrity *bi, 1588 unsigned int sectors) 1589 { 1590 return bio_integrity_intervals(bi, sectors) * bi->tuple_size; 1591 } 1592 1593 /* 1594 * Return the first bvec that contains integrity data. Only drivers that are 1595 * limited to a single integrity segment should use this helper. 1596 */ 1597 static inline struct bio_vec *rq_integrity_vec(struct request *rq) 1598 { 1599 if (WARN_ON_ONCE(queue_max_integrity_segments(rq->q) > 1)) 1600 return NULL; 1601 return rq->bio->bi_integrity->bip_vec; 1602 } 1603 1604 #else /* CONFIG_BLK_DEV_INTEGRITY */ 1605 1606 struct bio; 1607 struct block_device; 1608 struct gendisk; 1609 struct blk_integrity; 1610 1611 static inline int blk_integrity_rq(struct request *rq) 1612 { 1613 return 0; 1614 } 1615 static inline int blk_rq_count_integrity_sg(struct request_queue *q, 1616 struct bio *b) 1617 { 1618 return 0; 1619 } 1620 static inline int blk_rq_map_integrity_sg(struct request_queue *q, 1621 struct bio *b, 1622 struct scatterlist *s) 1623 { 1624 return 0; 1625 } 1626 static inline struct blk_integrity *bdev_get_integrity(struct block_device *b) 1627 { 1628 return NULL; 1629 } 1630 static inline struct blk_integrity *blk_get_integrity(struct gendisk *disk) 1631 { 1632 return NULL; 1633 } 1634 static inline int blk_integrity_compare(struct gendisk *a, struct gendisk *b) 1635 { 1636 return 0; 1637 } 1638 static inline void blk_integrity_register(struct gendisk *d, 1639 struct blk_integrity *b) 1640 { 1641 } 1642 static inline void blk_integrity_unregister(struct gendisk *d) 1643 { 1644 } 1645 static inline void blk_queue_max_integrity_segments(struct request_queue *q, 1646 unsigned int segs) 1647 { 1648 } 1649 static inline unsigned short queue_max_integrity_segments(const struct request_queue *q) 1650 { 1651 return 0; 1652 } 1653 static inline bool blk_integrity_merge_rq(struct request_queue *rq, 1654 struct request *r1, 1655 struct request *r2) 1656 { 1657 return true; 1658 } 1659 static inline bool blk_integrity_merge_bio(struct request_queue *rq, 1660 struct request *r, 1661 struct bio *b) 1662 { 1663 return true; 1664 } 1665 1666 static inline unsigned int bio_integrity_intervals(struct blk_integrity *bi, 1667 unsigned int sectors) 1668 { 1669 return 0; 1670 } 1671 1672 static inline unsigned int bio_integrity_bytes(struct blk_integrity *bi, 1673 unsigned int sectors) 1674 { 1675 return 0; 1676 } 1677 1678 static inline struct bio_vec *rq_integrity_vec(struct request *rq) 1679 { 1680 return NULL; 1681 } 1682 1683 #endif /* CONFIG_BLK_DEV_INTEGRITY */ 1684 1685 struct block_device_operations { 1686 int (*open) (struct block_device *, fmode_t); 1687 void (*release) (struct gendisk *, fmode_t); 1688 int (*rw_page)(struct block_device *, sector_t, struct page *, unsigned int); 1689 int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long); 1690 int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long); 1691 unsigned int (*check_events) (struct gendisk *disk, 1692 unsigned int clearing); 1693 /* ->media_changed() is DEPRECATED, use ->check_events() instead */ 1694 int (*media_changed) (struct gendisk *); 1695 void (*unlock_native_capacity) (struct gendisk *); 1696 int (*revalidate_disk) (struct gendisk *); 1697 int (*getgeo)(struct block_device *, struct hd_geometry *); 1698 /* this callback is with swap_lock and sometimes page table lock held */ 1699 void (*swap_slot_free_notify) (struct block_device *, unsigned long); 1700 int (*report_zones)(struct gendisk *, sector_t sector, 1701 unsigned int nr_zones, report_zones_cb cb, void *data); 1702 char *(*devnode)(struct gendisk *disk, umode_t *mode); 1703 struct module *owner; 1704 const struct pr_ops *pr_ops; 1705 }; 1706 1707 #ifdef CONFIG_COMPAT 1708 extern int blkdev_compat_ptr_ioctl(struct block_device *, fmode_t, 1709 unsigned int, unsigned long); 1710 #else 1711 #define blkdev_compat_ptr_ioctl NULL 1712 #endif 1713 1714 extern int __blkdev_driver_ioctl(struct block_device *, fmode_t, unsigned int, 1715 unsigned long); 1716 extern int bdev_read_page(struct block_device *, sector_t, struct page *); 1717 extern int bdev_write_page(struct block_device *, sector_t, struct page *, 1718 struct writeback_control *); 1719 1720 #ifdef CONFIG_BLK_DEV_ZONED 1721 bool blk_req_needs_zone_write_lock(struct request *rq); 1722 void __blk_req_zone_write_lock(struct request *rq); 1723 void __blk_req_zone_write_unlock(struct request *rq); 1724 1725 static inline void blk_req_zone_write_lock(struct request *rq) 1726 { 1727 if (blk_req_needs_zone_write_lock(rq)) 1728 __blk_req_zone_write_lock(rq); 1729 } 1730 1731 static inline void blk_req_zone_write_unlock(struct request *rq) 1732 { 1733 if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED) 1734 __blk_req_zone_write_unlock(rq); 1735 } 1736 1737 static inline bool blk_req_zone_is_write_locked(struct request *rq) 1738 { 1739 return rq->q->seq_zones_wlock && 1740 test_bit(blk_rq_zone_no(rq), rq->q->seq_zones_wlock); 1741 } 1742 1743 static inline bool blk_req_can_dispatch_to_zone(struct request *rq) 1744 { 1745 if (!blk_req_needs_zone_write_lock(rq)) 1746 return true; 1747 return !blk_req_zone_is_write_locked(rq); 1748 } 1749 #else 1750 static inline bool blk_req_needs_zone_write_lock(struct request *rq) 1751 { 1752 return false; 1753 } 1754 1755 static inline void blk_req_zone_write_lock(struct request *rq) 1756 { 1757 } 1758 1759 static inline void blk_req_zone_write_unlock(struct request *rq) 1760 { 1761 } 1762 static inline bool blk_req_zone_is_write_locked(struct request *rq) 1763 { 1764 return false; 1765 } 1766 1767 static inline bool blk_req_can_dispatch_to_zone(struct request *rq) 1768 { 1769 return true; 1770 } 1771 #endif /* CONFIG_BLK_DEV_ZONED */ 1772 1773 #else /* CONFIG_BLOCK */ 1774 1775 struct block_device; 1776 1777 /* 1778 * stubs for when the block layer is configured out 1779 */ 1780 #define buffer_heads_over_limit 0 1781 1782 static inline long nr_blockdev_pages(void) 1783 { 1784 return 0; 1785 } 1786 1787 struct blk_plug { 1788 }; 1789 1790 static inline void blk_start_plug(struct blk_plug *plug) 1791 { 1792 } 1793 1794 static inline void blk_finish_plug(struct blk_plug *plug) 1795 { 1796 } 1797 1798 static inline void blk_flush_plug(struct task_struct *task) 1799 { 1800 } 1801 1802 static inline void blk_schedule_flush_plug(struct task_struct *task) 1803 { 1804 } 1805 1806 1807 static inline bool blk_needs_flush_plug(struct task_struct *tsk) 1808 { 1809 return false; 1810 } 1811 1812 static inline int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask, 1813 sector_t *error_sector) 1814 { 1815 return 0; 1816 } 1817 1818 #endif /* CONFIG_BLOCK */ 1819 1820 static inline void blk_wake_io_task(struct task_struct *waiter) 1821 { 1822 /* 1823 * If we're polling, the task itself is doing the completions. For 1824 * that case, we don't need to signal a wakeup, it's enough to just 1825 * mark us as RUNNING. 1826 */ 1827 if (waiter == current) 1828 __set_current_state(TASK_RUNNING); 1829 else 1830 wake_up_process(waiter); 1831 } 1832 1833 #endif 1834