1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef BLK_INTERNAL_H 3 #define BLK_INTERNAL_H 4 5 #include <linux/bio-integrity.h> 6 #include <linux/blk-crypto.h> 7 #include <linux/lockdep.h> 8 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */ 9 #include <linux/sched/sysctl.h> 10 #include <linux/timekeeping.h> 11 #include <xen/xen.h> 12 #include "blk-crypto-internal.h" 13 14 struct elv_change_ctx; 15 16 /* 17 * Default upper limit for the software max_sectors limit used for regular I/Os. 18 * This can be increased through sysfs. 19 * 20 * This should not be confused with the max_hw_sector limit that is entirely 21 * controlled by the block device driver, usually based on hardware limits. 22 */ 23 #define BLK_DEF_MAX_SECTORS_CAP (SZ_4M >> SECTOR_SHIFT) 24 25 #define BLK_DEV_MAX_SECTORS (LLONG_MAX >> 9) 26 #define BLK_MIN_SEGMENT_SIZE 4096 27 28 /* Max future timer expiry for timeouts */ 29 #define BLK_MAX_TIMEOUT (5 * HZ) 30 31 extern const struct kobj_type blk_queue_ktype; 32 extern struct dentry *blk_debugfs_root; 33 34 struct blk_flush_queue { 35 spinlock_t mq_flush_lock; 36 unsigned int flush_pending_idx:1; 37 unsigned int flush_running_idx:1; 38 blk_status_t rq_status; 39 unsigned long flush_pending_since; 40 struct list_head flush_queue[2]; 41 unsigned long flush_data_in_flight; 42 struct request *flush_rq; 43 struct rcu_head rcu_head; 44 }; 45 46 bool is_flush_rq(struct request *req); 47 48 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size, 49 gfp_t flags); 50 void blk_free_flush_queue(struct blk_flush_queue *q); 51 52 bool __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic); 53 bool blk_queue_start_drain(struct request_queue *q); 54 bool __blk_freeze_queue_start(struct request_queue *q, 55 struct task_struct *owner); 56 int __bio_queue_enter(struct request_queue *q, struct bio *bio); 57 void submit_bio_noacct_nocheck(struct bio *bio, bool split); 58 int bio_submit_or_kill(struct bio *bio, unsigned int flags); 59 60 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm) 61 { 62 rcu_read_lock(); 63 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter)) 64 goto fail; 65 66 /* 67 * The code that increments the pm_only counter must ensure that the 68 * counter is globally visible before the queue is unfrozen. 69 */ 70 if (blk_queue_pm_only(q) && 71 (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) 72 goto fail_put; 73 74 rcu_read_unlock(); 75 return true; 76 77 fail_put: 78 blk_queue_exit(q); 79 fail: 80 rcu_read_unlock(); 81 return false; 82 } 83 84 static inline int bio_queue_enter(struct bio *bio) 85 { 86 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 87 88 if (blk_try_enter_queue(q, false)) { 89 rwsem_acquire_read(&q->io_lockdep_map, 0, 0, _RET_IP_); 90 rwsem_release(&q->io_lockdep_map, _RET_IP_); 91 return 0; 92 } 93 return __bio_queue_enter(q, bio); 94 } 95 96 static inline void blk_wait_io(struct completion *done) 97 { 98 /* Prevent hang_check timer from firing at us during very long I/O */ 99 unsigned long timeout = sysctl_hung_task_timeout_secs * HZ / 2; 100 101 if (timeout) 102 while (!wait_for_completion_io_timeout(done, timeout)) 103 ; 104 else 105 wait_for_completion_io(done); 106 } 107 108 struct block_device *blkdev_get_no_open(dev_t dev, bool autoload); 109 void blkdev_put_no_open(struct block_device *bdev); 110 111 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv, 112 struct page *page, unsigned len, unsigned offset); 113 114 static inline bool biovec_phys_mergeable(struct request_queue *q, 115 struct bio_vec *vec1, struct bio_vec *vec2) 116 { 117 unsigned long mask = queue_segment_boundary(q); 118 phys_addr_t addr1 = bvec_phys(vec1); 119 phys_addr_t addr2 = bvec_phys(vec2); 120 121 /* 122 * Merging adjacent physical pages may not work correctly under KMSAN 123 * if their metadata pages aren't adjacent. Just disable merging. 124 */ 125 if (IS_ENABLED(CONFIG_KMSAN)) 126 return false; 127 128 if (addr1 + vec1->bv_len != addr2) 129 return false; 130 if (!zone_device_pages_have_same_pgmap(vec1->bv_page, vec2->bv_page)) 131 return false; 132 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page)) 133 return false; 134 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask)) 135 return false; 136 return true; 137 } 138 139 /* 140 * Check if two pages from potentially different zone device pgmaps can 141 * coexist as separate bvec entries in the same bio. 142 * 143 * The block DMA iterator (blk_dma_map_iter_start) caches the P2PDMA mapping 144 * state from the first segment and applies it to all subsequent segments, so 145 * P2PDMA pages from different pgmaps must not be mixed in the same bio. 146 * 147 * Other zone device types (FS_DAX, GENERIC) use the same dma_map_phys() path 148 * as normal RAM. PRIVATE and COHERENT pages never appear in bios. 149 */ 150 static inline bool zone_device_pages_compatible(const struct page *a, 151 const struct page *b) 152 { 153 if (is_pci_p2pdma_page(a) || is_pci_p2pdma_page(b)) 154 return zone_device_pages_have_same_pgmap(a, b); 155 return true; 156 } 157 158 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim, 159 struct bio_vec *bprv, unsigned int offset) 160 { 161 return (offset & lim->virt_boundary_mask) || 162 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask); 163 } 164 165 /* 166 * Check if adding a bio_vec after bprv with offset would create a gap in 167 * the SG list. Most drivers don't care about this, but some do. 168 */ 169 static inline bool bvec_gap_to_prev(const struct queue_limits *lim, 170 struct bio_vec *bprv, unsigned int offset) 171 { 172 if (!lim->virt_boundary_mask) 173 return false; 174 return __bvec_gap_to_prev(lim, bprv, offset); 175 } 176 177 static inline bool rq_mergeable(struct request *rq) 178 { 179 if (blk_rq_is_passthrough(rq)) 180 return false; 181 182 if (req_op(rq) == REQ_OP_FLUSH) 183 return false; 184 185 if (req_op(rq) == REQ_OP_WRITE_ZEROES) 186 return false; 187 188 if (req_op(rq) == REQ_OP_ZONE_APPEND) 189 return false; 190 191 if (rq->cmd_flags & REQ_NOMERGE_FLAGS) 192 return false; 193 if (rq->rq_flags & RQF_NOMERGE_FLAGS) 194 return false; 195 196 return true; 197 } 198 199 /* 200 * There are two different ways to handle DISCARD merges: 201 * 1) If max_discard_segments > 1, the driver treats every bio as a range and 202 * send the bios to controller together. The ranges don't need to be 203 * contiguous. 204 * 2) Otherwise, the request will be normal read/write requests. The ranges 205 * need to be contiguous. 206 */ 207 static inline bool blk_discard_mergable(struct request *req) 208 { 209 if (req_op(req) == REQ_OP_DISCARD && 210 queue_max_discard_segments(req->q) > 1) 211 return true; 212 return false; 213 } 214 215 static inline unsigned int blk_rq_get_max_segments(struct request *rq) 216 { 217 if (req_op(rq) == REQ_OP_DISCARD) 218 return queue_max_discard_segments(rq->q); 219 return queue_max_segments(rq->q); 220 } 221 222 static inline unsigned int blk_queue_get_max_sectors(struct request *rq) 223 { 224 struct request_queue *q = rq->q; 225 enum req_op op = req_op(rq); 226 227 if (unlikely(op == REQ_OP_DISCARD)) 228 return min(q->limits.max_discard_sectors, 229 UINT_MAX >> SECTOR_SHIFT); 230 231 if (unlikely(op == REQ_OP_SECURE_ERASE)) 232 return min(q->limits.max_secure_erase_sectors, 233 UINT_MAX >> SECTOR_SHIFT); 234 235 if (unlikely(op == REQ_OP_WRITE_ZEROES)) 236 return q->limits.max_write_zeroes_sectors; 237 238 if (rq->cmd_flags & REQ_ATOMIC) 239 return q->limits.atomic_write_max_sectors; 240 241 return q->limits.max_sectors; 242 } 243 244 #ifdef CONFIG_BLK_DEV_INTEGRITY 245 void blk_flush_integrity(void); 246 void bio_integrity_free(struct bio *bio); 247 248 /* 249 * Integrity payloads can either be owned by the submitter, in which case 250 * bio_uninit will free them, or owned and generated by the block layer, 251 * in which case we'll verify them here (for reads) and free them before 252 * the bio is handed back to the submitted. 253 */ 254 bool __bio_integrity_endio(struct bio *bio); 255 static inline bool bio_integrity_endio(struct bio *bio) 256 { 257 struct bio_integrity_payload *bip = bio_integrity(bio); 258 259 if (bip && (bip->bip_flags & BIP_BLOCK_INTEGRITY)) 260 return __bio_integrity_endio(bio); 261 return true; 262 } 263 264 bool blk_integrity_merge_rq(struct request_queue *, struct request *, 265 struct request *); 266 bool blk_integrity_merge_bio(struct request_queue *, struct request *, 267 struct bio *); 268 269 static inline bool integrity_req_gap_back_merge(struct request *req, 270 struct bio *next) 271 { 272 struct bio_integrity_payload *bip = bio_integrity(req->bio); 273 struct bio_integrity_payload *bip_next = bio_integrity(next); 274 275 return bvec_gap_to_prev(&req->q->limits, 276 &bip->bip_vec[bip->bip_vcnt - 1], 277 bip_next->bip_vec[0].bv_offset); 278 } 279 280 static inline bool integrity_req_gap_front_merge(struct request *req, 281 struct bio *bio) 282 { 283 struct bio_integrity_payload *bip = bio_integrity(bio); 284 struct bio_integrity_payload *bip_next = bio_integrity(req->bio); 285 286 return bvec_gap_to_prev(&req->q->limits, 287 &bip->bip_vec[bip->bip_vcnt - 1], 288 bip_next->bip_vec[0].bv_offset); 289 } 290 291 extern const struct attribute_group blk_integrity_attr_group; 292 #else /* CONFIG_BLK_DEV_INTEGRITY */ 293 static inline bool blk_integrity_merge_rq(struct request_queue *rq, 294 struct request *r1, struct request *r2) 295 { 296 return true; 297 } 298 static inline bool blk_integrity_merge_bio(struct request_queue *rq, 299 struct request *r, struct bio *b) 300 { 301 return true; 302 } 303 static inline bool integrity_req_gap_back_merge(struct request *req, 304 struct bio *next) 305 { 306 return false; 307 } 308 static inline bool integrity_req_gap_front_merge(struct request *req, 309 struct bio *bio) 310 { 311 return false; 312 } 313 314 static inline void blk_flush_integrity(void) 315 { 316 } 317 static inline bool bio_integrity_endio(struct bio *bio) 318 { 319 return true; 320 } 321 static inline void bio_integrity_free(struct bio *bio) 322 { 323 } 324 #endif /* CONFIG_BLK_DEV_INTEGRITY */ 325 326 unsigned long blk_rq_timeout(unsigned long timeout); 327 void blk_add_timer(struct request *req); 328 329 enum bio_merge_status { 330 BIO_MERGE_OK, 331 BIO_MERGE_NONE, 332 BIO_MERGE_FAILED, 333 }; 334 335 enum bio_merge_status bio_attempt_back_merge(struct request *req, 336 struct bio *bio, unsigned int nr_segs); 337 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, 338 unsigned int nr_segs); 339 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, 340 struct bio *bio, unsigned int nr_segs); 341 342 /* 343 * Plug flush limits 344 */ 345 #define BLK_MAX_REQUEST_COUNT 32 346 #define BLK_PLUG_FLUSH_SIZE (128 * 1024) 347 348 /* 349 * Internal elevator interface 350 */ 351 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED) 352 353 bool blk_insert_flush(struct request *rq); 354 355 void elv_update_nr_hw_queues(struct request_queue *q, 356 struct elv_change_ctx *ctx); 357 void elevator_set_default(struct request_queue *q); 358 void elevator_set_none(struct request_queue *q); 359 360 ssize_t part_size_show(struct device *dev, struct device_attribute *attr, 361 char *buf); 362 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr, 363 char *buf); 364 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr, 365 char *buf); 366 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr, 367 char *buf); 368 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr, 369 const char *buf, size_t count); 370 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *); 371 ssize_t part_timeout_store(struct device *, struct device_attribute *, 372 const char *, size_t); 373 374 struct bio *bio_split_discard(struct bio *bio, const struct queue_limits *lim, 375 unsigned *nsegs); 376 struct bio *bio_split_write_zeroes(struct bio *bio, 377 const struct queue_limits *lim, unsigned *nsegs); 378 struct bio *bio_split_rw(struct bio *bio, const struct queue_limits *lim, 379 unsigned *nr_segs); 380 struct bio *bio_split_zone_append(struct bio *bio, 381 const struct queue_limits *lim, unsigned *nr_segs); 382 383 /* 384 * All drivers must accept single-segments bios that are smaller than PAGE_SIZE. 385 * 386 * This is a quick and dirty check that relies on the fact that bi_io_vec[0] is 387 * always valid if a bio has data. The check might lead to occasional false 388 * positives when bios are cloned, but compared to the performance impact of 389 * cloned bios themselves the loop below doesn't matter anyway. 390 */ 391 static inline bool bio_may_need_split(struct bio *bio, 392 const struct queue_limits *lim) 393 { 394 const struct bio_vec *bv; 395 396 if (lim->chunk_sectors) 397 return true; 398 399 if (!bio->bi_io_vec) 400 return true; 401 402 bv = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 403 if (bio->bi_iter.bi_size > bv->bv_len - bio->bi_iter.bi_bvec_done) 404 return true; 405 return bv->bv_len + bv->bv_offset > lim->max_fast_segment_size; 406 } 407 408 /** 409 * __bio_split_to_limits - split a bio to fit the queue limits 410 * @bio: bio to be split 411 * @lim: queue limits to split based on 412 * @nr_segs: returns the number of segments in the returned bio 413 * 414 * Check if @bio needs splitting based on the queue limits, and if so split off 415 * a bio fitting the limits from the beginning of @bio and return it. @bio is 416 * shortened to the remainder and re-submitted. 417 * 418 * The split bio is allocated from @q->bio_split, which is provided by the 419 * block layer. 420 */ 421 static inline struct bio *__bio_split_to_limits(struct bio *bio, 422 const struct queue_limits *lim, unsigned int *nr_segs) 423 { 424 switch (bio_op(bio)) { 425 case REQ_OP_READ: 426 case REQ_OP_WRITE: 427 if (bio_may_need_split(bio, lim)) 428 return bio_split_rw(bio, lim, nr_segs); 429 *nr_segs = 1; 430 return bio; 431 case REQ_OP_ZONE_APPEND: 432 return bio_split_zone_append(bio, lim, nr_segs); 433 case REQ_OP_DISCARD: 434 case REQ_OP_SECURE_ERASE: 435 return bio_split_discard(bio, lim, nr_segs); 436 case REQ_OP_WRITE_ZEROES: 437 return bio_split_write_zeroes(bio, lim, nr_segs); 438 default: 439 /* other operations can't be split */ 440 *nr_segs = 0; 441 return bio; 442 } 443 } 444 445 /** 446 * get_max_segment_size() - maximum number of bytes to add as a single segment 447 * @lim: Request queue limits. 448 * @paddr: address of the range to add 449 * @len: maximum length available to add at @paddr 450 * 451 * Returns the maximum number of bytes of the range starting at @paddr that can 452 * be added to a single segment. 453 */ 454 static inline unsigned get_max_segment_size(const struct queue_limits *lim, 455 phys_addr_t paddr, unsigned int len) 456 { 457 /* 458 * Prevent an overflow if mask = ULONG_MAX and offset = 0 by adding 1 459 * after having calculated the minimum. 460 */ 461 return min_t(unsigned long, len, 462 min(lim->seg_boundary_mask - (lim->seg_boundary_mask & paddr), 463 (unsigned long)lim->max_segment_size - 1) + 1); 464 } 465 466 int ll_back_merge_fn(struct request *req, struct bio *bio, 467 unsigned int nr_segs); 468 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, 469 struct request *next); 470 unsigned int blk_recalc_rq_segments(struct request *rq); 471 bool blk_rq_merge_ok(struct request *rq, struct bio *bio); 472 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio); 473 474 int blk_set_default_limits(struct queue_limits *lim); 475 void blk_apply_bdi_limits(struct backing_dev_info *bdi, 476 struct queue_limits *lim); 477 int blk_dev_init(void); 478 479 void update_io_ticks(struct block_device *part, unsigned long now, bool end); 480 481 static inline void req_set_nomerge(struct request_queue *q, struct request *req) 482 { 483 req->cmd_flags |= REQ_NOMERGE; 484 if (req == q->last_merge) 485 q->last_merge = NULL; 486 } 487 488 /* 489 * Internal io_context interface 490 */ 491 struct io_cq *ioc_find_get_icq(struct request_queue *q); 492 struct io_cq *ioc_lookup_icq(struct request_queue *q); 493 #ifdef CONFIG_BLK_ICQ 494 void ioc_clear_queue(struct request_queue *q); 495 #else 496 static inline void ioc_clear_queue(struct request_queue *q) 497 { 498 } 499 #endif /* CONFIG_BLK_ICQ */ 500 501 #ifdef CONFIG_BLK_DEV_ZONED 502 void disk_init_zone_resources(struct gendisk *disk); 503 void disk_free_zone_resources(struct gendisk *disk); 504 static inline bool bio_zone_write_plugging(struct bio *bio) 505 { 506 return bio_flagged(bio, BIO_ZONE_WRITE_PLUGGING); 507 } 508 static inline bool blk_req_bio_is_zone_append(struct request *rq, 509 struct bio *bio) 510 { 511 return req_op(rq) == REQ_OP_ZONE_APPEND || 512 bio_flagged(bio, BIO_EMULATES_ZONE_APPEND); 513 } 514 void blk_zone_write_plug_bio_merged(struct bio *bio); 515 void blk_zone_write_plug_init_request(struct request *rq); 516 void blk_zone_append_update_request_bio(struct request *rq, struct bio *bio); 517 void blk_zone_mgmt_bio_endio(struct bio *bio); 518 void blk_zone_write_plug_bio_endio(struct bio *bio); 519 static inline void blk_zone_bio_endio(struct bio *bio) 520 { 521 /* 522 * Zone management BIOs may impact zone write plugs (e.g. a zone reset 523 * changes a zone write plug zone write pointer offset), but these 524 * operation do not go through zone write plugging as they may operate 525 * on zones that do not have a zone write 526 * plug. blk_zone_mgmt_bio_endio() handles the potential changes to zone 527 * write plugs that are present. 528 */ 529 if (op_is_zone_mgmt(bio_op(bio))) { 530 blk_zone_mgmt_bio_endio(bio); 531 return; 532 } 533 534 /* 535 * For write BIOs to zoned devices, signal the completion of the BIO so 536 * that the next write BIO can be submitted by zone write plugging. 537 */ 538 if (bio_zone_write_plugging(bio)) 539 blk_zone_write_plug_bio_endio(bio); 540 } 541 542 void blk_zone_write_plug_finish_request(struct request *rq); 543 static inline void blk_zone_finish_request(struct request *rq) 544 { 545 if (rq->rq_flags & RQF_ZONE_WRITE_PLUGGING) 546 blk_zone_write_plug_finish_request(rq); 547 } 548 int blkdev_report_zones_ioctl(struct block_device *bdev, unsigned int cmd, 549 unsigned long arg); 550 int blkdev_zone_mgmt_ioctl(struct block_device *bdev, blk_mode_t mode, 551 unsigned int cmd, unsigned long arg); 552 #else /* CONFIG_BLK_DEV_ZONED */ 553 static inline void disk_init_zone_resources(struct gendisk *disk) 554 { 555 } 556 static inline void disk_free_zone_resources(struct gendisk *disk) 557 { 558 } 559 static inline bool bio_zone_write_plugging(struct bio *bio) 560 { 561 return false; 562 } 563 static inline bool blk_req_bio_is_zone_append(struct request *req, 564 struct bio *bio) 565 { 566 return false; 567 } 568 static inline void blk_zone_write_plug_bio_merged(struct bio *bio) 569 { 570 } 571 static inline void blk_zone_write_plug_init_request(struct request *rq) 572 { 573 } 574 static inline void blk_zone_append_update_request_bio(struct request *rq, 575 struct bio *bio) 576 { 577 } 578 static inline void blk_zone_bio_endio(struct bio *bio) 579 { 580 } 581 static inline void blk_zone_finish_request(struct request *rq) 582 { 583 } 584 static inline int blkdev_report_zones_ioctl(struct block_device *bdev, 585 unsigned int cmd, unsigned long arg) 586 { 587 return -ENOTTY; 588 } 589 static inline int blkdev_zone_mgmt_ioctl(struct block_device *bdev, 590 blk_mode_t mode, unsigned int cmd, unsigned long arg) 591 { 592 return -ENOTTY; 593 } 594 #endif /* CONFIG_BLK_DEV_ZONED */ 595 596 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno); 597 void bdev_add(struct block_device *bdev, dev_t dev); 598 void bdev_unhash(struct block_device *bdev); 599 void bdev_drop(struct block_device *bdev); 600 601 int blk_alloc_ext_minor(void); 602 void blk_free_ext_minor(unsigned int minor); 603 #define ADDPART_FLAG_NONE 0 604 #define ADDPART_FLAG_RAID 1 605 #define ADDPART_FLAG_WHOLEDISK 2 606 #define ADDPART_FLAG_READONLY 4 607 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 608 sector_t length); 609 int bdev_del_partition(struct gendisk *disk, int partno); 610 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, 611 sector_t length); 612 void drop_partition(struct block_device *part); 613 614 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors); 615 616 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id, 617 struct lock_class_key *lkclass); 618 struct request_queue *blk_alloc_queue(struct queue_limits *lim, int node_id); 619 620 int disk_scan_partitions(struct gendisk *disk, blk_mode_t mode); 621 622 int disk_alloc_events(struct gendisk *disk); 623 void disk_add_events(struct gendisk *disk); 624 void disk_del_events(struct gendisk *disk); 625 void disk_release_events(struct gendisk *disk); 626 void disk_block_events(struct gendisk *disk); 627 void disk_unblock_events(struct gendisk *disk); 628 void disk_flush_events(struct gendisk *disk, unsigned int mask); 629 extern struct device_attribute dev_attr_events; 630 extern struct device_attribute dev_attr_events_async; 631 extern struct device_attribute dev_attr_events_poll_msecs; 632 633 extern struct attribute_group blk_trace_attr_group; 634 635 blk_mode_t file_to_blk_mode(struct file *file); 636 int truncate_bdev_range(struct block_device *bdev, blk_mode_t mode, 637 loff_t lstart, loff_t lend); 638 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg); 639 int blkdev_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags); 640 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg); 641 642 extern const struct address_space_operations def_blk_aops; 643 644 int disk_register_independent_access_ranges(struct gendisk *disk); 645 void disk_unregister_independent_access_ranges(struct gendisk *disk); 646 647 int should_fail_bio(struct bio *bio); 648 #ifdef CONFIG_FAIL_MAKE_REQUEST 649 bool should_fail_request(struct block_device *part, unsigned int bytes); 650 #else /* CONFIG_FAIL_MAKE_REQUEST */ 651 static inline bool should_fail_request(struct block_device *part, 652 unsigned int bytes) 653 { 654 return false; 655 } 656 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 657 658 /* 659 * Optimized request reference counting. Ideally we'd make timeouts be more 660 * clever, as that's the only reason we need references at all... But until 661 * this happens, this is faster than using refcount_t. Also see: 662 * 663 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count") 664 */ 665 #define req_ref_zero_or_close_to_overflow(req) \ 666 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u) 667 668 static inline bool req_ref_inc_not_zero(struct request *req) 669 { 670 return atomic_inc_not_zero(&req->ref); 671 } 672 673 static inline bool req_ref_put_and_test(struct request *req) 674 { 675 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); 676 return atomic_dec_and_test(&req->ref); 677 } 678 679 static inline void req_ref_set(struct request *req, int value) 680 { 681 atomic_set(&req->ref, value); 682 } 683 684 static inline int req_ref_read(struct request *req) 685 { 686 return atomic_read(&req->ref); 687 } 688 689 static inline u64 blk_time_get_ns(void) 690 { 691 struct blk_plug *plug = current->plug; 692 693 if (!plug || !in_task()) 694 return ktime_get_ns(); 695 696 /* 697 * 0 could very well be a valid time, but rather than flag "this is 698 * a valid timestamp" separately, just accept that we'll do an extra 699 * ktime_get_ns() if we just happen to get 0 as the current time. 700 */ 701 if (!plug->cur_ktime) { 702 plug->cur_ktime = ktime_get_ns(); 703 current->flags |= PF_BLOCK_TS; 704 } 705 return plug->cur_ktime; 706 } 707 708 static inline ktime_t blk_time_get(void) 709 { 710 return ns_to_ktime(blk_time_get_ns()); 711 } 712 713 void bdev_release(struct file *bdev_file); 714 int bdev_open(struct block_device *bdev, blk_mode_t mode, void *holder, 715 const struct blk_holder_ops *hops, struct file *bdev_file); 716 int bdev_permission(dev_t dev, blk_mode_t mode, void *holder); 717 718 void bio_integrity_generate(struct bio *bio); 719 blk_status_t bio_integrity_verify(struct bio *bio, 720 struct bvec_iter *saved_iter); 721 722 void blk_integrity_prepare(struct request *rq); 723 void blk_integrity_complete(struct request *rq, unsigned int nr_bytes); 724 725 #ifdef CONFIG_LOCKDEP 726 static inline void blk_freeze_acquire_lock(struct request_queue *q) 727 { 728 if (!q->mq_freeze_disk_dead) 729 rwsem_acquire(&q->io_lockdep_map, 0, 1, _RET_IP_); 730 if (!q->mq_freeze_queue_dying) 731 rwsem_acquire(&q->q_lockdep_map, 0, 1, _RET_IP_); 732 } 733 734 static inline void blk_unfreeze_release_lock(struct request_queue *q) 735 { 736 if (!q->mq_freeze_queue_dying) 737 rwsem_release(&q->q_lockdep_map, _RET_IP_); 738 if (!q->mq_freeze_disk_dead) 739 rwsem_release(&q->io_lockdep_map, _RET_IP_); 740 } 741 #else 742 static inline void blk_freeze_acquire_lock(struct request_queue *q) 743 { 744 } 745 static inline void blk_unfreeze_release_lock(struct request_queue *q) 746 { 747 } 748 #endif 749 750 /* 751 * debugfs directory and file creation can trigger fs reclaim, which can enter 752 * back into the block layer request_queue. This can cause deadlock if the 753 * queue is frozen. Use NOIO context together with debugfs_mutex to prevent fs 754 * reclaim from triggering block I/O. 755 */ 756 static inline void blk_debugfs_lock_nomemsave(struct request_queue *q) 757 { 758 mutex_lock(&q->debugfs_mutex); 759 } 760 761 static inline void blk_debugfs_unlock_nomemrestore(struct request_queue *q) 762 { 763 mutex_unlock(&q->debugfs_mutex); 764 } 765 766 static inline unsigned int __must_check blk_debugfs_lock(struct request_queue *q) 767 { 768 unsigned int memflags = memalloc_noio_save(); 769 770 blk_debugfs_lock_nomemsave(q); 771 return memflags; 772 } 773 774 static inline void blk_debugfs_unlock(struct request_queue *q, 775 unsigned int memflags) 776 { 777 blk_debugfs_unlock_nomemrestore(q); 778 memalloc_noio_restore(memflags); 779 } 780 781 #endif /* BLK_INTERNAL_H */ 782