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