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