1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef BLK_INTERNAL_H 3 #define BLK_INTERNAL_H 4 5 #include <linux/blk-crypto.h> 6 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */ 7 #include <xen/xen.h> 8 #include "blk-crypto-internal.h" 9 10 struct elevator_type; 11 12 /* Max future timer expiry for timeouts */ 13 #define BLK_MAX_TIMEOUT (5 * HZ) 14 15 extern struct dentry *blk_debugfs_root; 16 17 struct blk_flush_queue { 18 unsigned int flush_pending_idx:1; 19 unsigned int flush_running_idx:1; 20 blk_status_t rq_status; 21 unsigned long flush_pending_since; 22 struct list_head flush_queue[2]; 23 struct list_head flush_data_in_flight; 24 struct request *flush_rq; 25 26 spinlock_t mq_flush_lock; 27 }; 28 29 bool is_flush_rq(struct request *req); 30 31 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size, 32 gfp_t flags); 33 void blk_free_flush_queue(struct blk_flush_queue *q); 34 35 void blk_freeze_queue(struct request_queue *q); 36 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic); 37 void blk_queue_start_drain(struct request_queue *q); 38 int __bio_queue_enter(struct request_queue *q, struct bio *bio); 39 void submit_bio_noacct_nocheck(struct bio *bio); 40 41 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm) 42 { 43 rcu_read_lock(); 44 if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter)) 45 goto fail; 46 47 /* 48 * The code that increments the pm_only counter must ensure that the 49 * counter is globally visible before the queue is unfrozen. 50 */ 51 if (blk_queue_pm_only(q) && 52 (!pm || queue_rpm_status(q) == RPM_SUSPENDED)) 53 goto fail_put; 54 55 rcu_read_unlock(); 56 return true; 57 58 fail_put: 59 blk_queue_exit(q); 60 fail: 61 rcu_read_unlock(); 62 return false; 63 } 64 65 static inline int bio_queue_enter(struct bio *bio) 66 { 67 struct request_queue *q = bdev_get_queue(bio->bi_bdev); 68 69 if (blk_try_enter_queue(q, false)) 70 return 0; 71 return __bio_queue_enter(q, bio); 72 } 73 74 #define BIO_INLINE_VECS 4 75 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, 76 gfp_t gfp_mask); 77 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs); 78 79 static inline bool biovec_phys_mergeable(struct request_queue *q, 80 struct bio_vec *vec1, struct bio_vec *vec2) 81 { 82 unsigned long mask = queue_segment_boundary(q); 83 phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset; 84 phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset; 85 86 /* 87 * Merging adjacent physical pages may not work correctly under KMSAN 88 * if their metadata pages aren't adjacent. Just disable merging. 89 */ 90 if (IS_ENABLED(CONFIG_KMSAN)) 91 return false; 92 93 if (addr1 + vec1->bv_len != addr2) 94 return false; 95 if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page)) 96 return false; 97 if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask)) 98 return false; 99 return true; 100 } 101 102 static inline bool __bvec_gap_to_prev(const struct queue_limits *lim, 103 struct bio_vec *bprv, unsigned int offset) 104 { 105 return (offset & lim->virt_boundary_mask) || 106 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask); 107 } 108 109 /* 110 * Check if adding a bio_vec after bprv with offset would create a gap in 111 * the SG list. Most drivers don't care about this, but some do. 112 */ 113 static inline bool bvec_gap_to_prev(const struct queue_limits *lim, 114 struct bio_vec *bprv, unsigned int offset) 115 { 116 if (!lim->virt_boundary_mask) 117 return false; 118 return __bvec_gap_to_prev(lim, bprv, offset); 119 } 120 121 static inline bool rq_mergeable(struct request *rq) 122 { 123 if (blk_rq_is_passthrough(rq)) 124 return false; 125 126 if (req_op(rq) == REQ_OP_FLUSH) 127 return false; 128 129 if (req_op(rq) == REQ_OP_WRITE_ZEROES) 130 return false; 131 132 if (req_op(rq) == REQ_OP_ZONE_APPEND) 133 return false; 134 135 if (rq->cmd_flags & REQ_NOMERGE_FLAGS) 136 return false; 137 if (rq->rq_flags & RQF_NOMERGE_FLAGS) 138 return false; 139 140 return true; 141 } 142 143 /* 144 * There are two different ways to handle DISCARD merges: 145 * 1) If max_discard_segments > 1, the driver treats every bio as a range and 146 * send the bios to controller together. The ranges don't need to be 147 * contiguous. 148 * 2) Otherwise, the request will be normal read/write requests. The ranges 149 * need to be contiguous. 150 */ 151 static inline bool blk_discard_mergable(struct request *req) 152 { 153 if (req_op(req) == REQ_OP_DISCARD && 154 queue_max_discard_segments(req->q) > 1) 155 return true; 156 return false; 157 } 158 159 static inline unsigned int blk_rq_get_max_segments(struct request *rq) 160 { 161 if (req_op(rq) == REQ_OP_DISCARD) 162 return queue_max_discard_segments(rq->q); 163 return queue_max_segments(rq->q); 164 } 165 166 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q, 167 enum req_op op) 168 { 169 if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE)) 170 return min(q->limits.max_discard_sectors, 171 UINT_MAX >> SECTOR_SHIFT); 172 173 if (unlikely(op == REQ_OP_WRITE_ZEROES)) 174 return q->limits.max_write_zeroes_sectors; 175 176 return q->limits.max_sectors; 177 } 178 179 #ifdef CONFIG_BLK_DEV_INTEGRITY 180 void blk_flush_integrity(void); 181 bool __bio_integrity_endio(struct bio *); 182 void bio_integrity_free(struct bio *bio); 183 static inline bool bio_integrity_endio(struct bio *bio) 184 { 185 if (bio_integrity(bio)) 186 return __bio_integrity_endio(bio); 187 return true; 188 } 189 190 bool blk_integrity_merge_rq(struct request_queue *, struct request *, 191 struct request *); 192 bool blk_integrity_merge_bio(struct request_queue *, struct request *, 193 struct bio *); 194 195 static inline bool integrity_req_gap_back_merge(struct request *req, 196 struct bio *next) 197 { 198 struct bio_integrity_payload *bip = bio_integrity(req->bio); 199 struct bio_integrity_payload *bip_next = bio_integrity(next); 200 201 return bvec_gap_to_prev(&req->q->limits, 202 &bip->bip_vec[bip->bip_vcnt - 1], 203 bip_next->bip_vec[0].bv_offset); 204 } 205 206 static inline bool integrity_req_gap_front_merge(struct request *req, 207 struct bio *bio) 208 { 209 struct bio_integrity_payload *bip = bio_integrity(bio); 210 struct bio_integrity_payload *bip_next = bio_integrity(req->bio); 211 212 return bvec_gap_to_prev(&req->q->limits, 213 &bip->bip_vec[bip->bip_vcnt - 1], 214 bip_next->bip_vec[0].bv_offset); 215 } 216 217 extern const struct attribute_group blk_integrity_attr_group; 218 #else /* CONFIG_BLK_DEV_INTEGRITY */ 219 static inline bool blk_integrity_merge_rq(struct request_queue *rq, 220 struct request *r1, struct request *r2) 221 { 222 return true; 223 } 224 static inline bool blk_integrity_merge_bio(struct request_queue *rq, 225 struct request *r, struct bio *b) 226 { 227 return true; 228 } 229 static inline bool integrity_req_gap_back_merge(struct request *req, 230 struct bio *next) 231 { 232 return false; 233 } 234 static inline bool integrity_req_gap_front_merge(struct request *req, 235 struct bio *bio) 236 { 237 return false; 238 } 239 240 static inline void blk_flush_integrity(void) 241 { 242 } 243 static inline bool bio_integrity_endio(struct bio *bio) 244 { 245 return true; 246 } 247 static inline void bio_integrity_free(struct bio *bio) 248 { 249 } 250 #endif /* CONFIG_BLK_DEV_INTEGRITY */ 251 252 unsigned long blk_rq_timeout(unsigned long timeout); 253 void blk_add_timer(struct request *req); 254 const char *blk_status_to_str(blk_status_t status); 255 256 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio, 257 unsigned int nr_segs); 258 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list, 259 struct bio *bio, unsigned int nr_segs); 260 261 /* 262 * Plug flush limits 263 */ 264 #define BLK_MAX_REQUEST_COUNT 32 265 #define BLK_PLUG_FLUSH_SIZE (128 * 1024) 266 267 /* 268 * Internal elevator interface 269 */ 270 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED) 271 272 void blk_insert_flush(struct request *rq); 273 274 int elevator_switch(struct request_queue *q, struct elevator_type *new_e); 275 void elevator_disable(struct request_queue *q); 276 void elevator_exit(struct request_queue *q); 277 int elv_register_queue(struct request_queue *q, bool uevent); 278 void elv_unregister_queue(struct request_queue *q); 279 280 ssize_t part_size_show(struct device *dev, struct device_attribute *attr, 281 char *buf); 282 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr, 283 char *buf); 284 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr, 285 char *buf); 286 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr, 287 char *buf); 288 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr, 289 const char *buf, size_t count); 290 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *); 291 ssize_t part_timeout_store(struct device *, struct device_attribute *, 292 const char *, size_t); 293 294 static inline bool bio_may_exceed_limits(struct bio *bio, 295 const struct queue_limits *lim) 296 { 297 switch (bio_op(bio)) { 298 case REQ_OP_DISCARD: 299 case REQ_OP_SECURE_ERASE: 300 case REQ_OP_WRITE_ZEROES: 301 return true; /* non-trivial splitting decisions */ 302 default: 303 break; 304 } 305 306 /* 307 * All drivers must accept single-segments bios that are <= PAGE_SIZE. 308 * This is a quick and dirty check that relies on the fact that 309 * bi_io_vec[0] is always valid if a bio has data. The check might 310 * lead to occasional false negatives when bios are cloned, but compared 311 * to the performance impact of cloned bios themselves the loop below 312 * doesn't matter anyway. 313 */ 314 return lim->chunk_sectors || bio->bi_vcnt != 1 || 315 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE; 316 } 317 318 struct bio *__bio_split_to_limits(struct bio *bio, 319 const struct queue_limits *lim, 320 unsigned int *nr_segs); 321 int ll_back_merge_fn(struct request *req, struct bio *bio, 322 unsigned int nr_segs); 323 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq, 324 struct request *next); 325 unsigned int blk_recalc_rq_segments(struct request *rq); 326 void blk_rq_set_mixed_merge(struct request *rq); 327 bool blk_rq_merge_ok(struct request *rq, struct bio *bio); 328 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio); 329 330 void blk_set_default_limits(struct queue_limits *lim); 331 int blk_dev_init(void); 332 333 /* 334 * Contribute to IO statistics IFF: 335 * 336 * a) it's attached to a gendisk, and 337 * b) the queue had IO stats enabled when this request was started 338 */ 339 static inline bool blk_do_io_stat(struct request *rq) 340 { 341 return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq); 342 } 343 344 void update_io_ticks(struct block_device *part, unsigned long now, bool end); 345 346 static inline void req_set_nomerge(struct request_queue *q, struct request *req) 347 { 348 req->cmd_flags |= REQ_NOMERGE; 349 if (req == q->last_merge) 350 q->last_merge = NULL; 351 } 352 353 /* 354 * Internal io_context interface 355 */ 356 struct io_cq *ioc_find_get_icq(struct request_queue *q); 357 struct io_cq *ioc_lookup_icq(struct request_queue *q); 358 #ifdef CONFIG_BLK_ICQ 359 void ioc_clear_queue(struct request_queue *q); 360 #else 361 static inline void ioc_clear_queue(struct request_queue *q) 362 { 363 } 364 #endif /* CONFIG_BLK_ICQ */ 365 366 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW 367 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page); 368 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q, 369 const char *page, size_t count); 370 extern void blk_throtl_bio_endio(struct bio *bio); 371 extern void blk_throtl_stat_add(struct request *rq, u64 time); 372 #else 373 static inline void blk_throtl_bio_endio(struct bio *bio) { } 374 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { } 375 #endif 376 377 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q); 378 379 static inline bool blk_queue_may_bounce(struct request_queue *q) 380 { 381 return IS_ENABLED(CONFIG_BOUNCE) && 382 q->limits.bounce == BLK_BOUNCE_HIGH && 383 max_low_pfn >= max_pfn; 384 } 385 386 static inline struct bio *blk_queue_bounce(struct bio *bio, 387 struct request_queue *q) 388 { 389 if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio))) 390 return __blk_queue_bounce(bio, q); 391 return bio; 392 } 393 394 #ifdef CONFIG_BLK_DEV_ZONED 395 void disk_free_zone_bitmaps(struct gendisk *disk); 396 void disk_clear_zone_settings(struct gendisk *disk); 397 #else 398 static inline void disk_free_zone_bitmaps(struct gendisk *disk) {} 399 static inline void disk_clear_zone_settings(struct gendisk *disk) {} 400 #endif 401 402 int blk_alloc_ext_minor(void); 403 void blk_free_ext_minor(unsigned int minor); 404 #define ADDPART_FLAG_NONE 0 405 #define ADDPART_FLAG_RAID 1 406 #define ADDPART_FLAG_WHOLEDISK 2 407 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start, 408 sector_t length); 409 int bdev_del_partition(struct gendisk *disk, int partno); 410 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start, 411 sector_t length); 412 void blk_drop_partitions(struct gendisk *disk); 413 414 void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors); 415 416 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id, 417 struct lock_class_key *lkclass); 418 419 int bio_add_hw_page(struct request_queue *q, struct bio *bio, 420 struct page *page, unsigned int len, unsigned int offset, 421 unsigned int max_sectors, bool *same_page); 422 423 struct request_queue *blk_alloc_queue(int node_id); 424 425 int disk_scan_partitions(struct gendisk *disk, fmode_t mode); 426 427 int disk_alloc_events(struct gendisk *disk); 428 void disk_add_events(struct gendisk *disk); 429 void disk_del_events(struct gendisk *disk); 430 void disk_release_events(struct gendisk *disk); 431 void disk_block_events(struct gendisk *disk); 432 void disk_unblock_events(struct gendisk *disk); 433 void disk_flush_events(struct gendisk *disk, unsigned int mask); 434 extern struct device_attribute dev_attr_events; 435 extern struct device_attribute dev_attr_events_async; 436 extern struct device_attribute dev_attr_events_poll_msecs; 437 438 extern struct attribute_group blk_trace_attr_group; 439 440 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg); 441 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg); 442 443 extern const struct address_space_operations def_blk_aops; 444 445 int disk_register_independent_access_ranges(struct gendisk *disk); 446 void disk_unregister_independent_access_ranges(struct gendisk *disk); 447 448 #ifdef CONFIG_FAIL_MAKE_REQUEST 449 bool should_fail_request(struct block_device *part, unsigned int bytes); 450 #else /* CONFIG_FAIL_MAKE_REQUEST */ 451 static inline bool should_fail_request(struct block_device *part, 452 unsigned int bytes) 453 { 454 return false; 455 } 456 #endif /* CONFIG_FAIL_MAKE_REQUEST */ 457 458 /* 459 * Optimized request reference counting. Ideally we'd make timeouts be more 460 * clever, as that's the only reason we need references at all... But until 461 * this happens, this is faster than using refcount_t. Also see: 462 * 463 * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count") 464 */ 465 #define req_ref_zero_or_close_to_overflow(req) \ 466 ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u) 467 468 static inline bool req_ref_inc_not_zero(struct request *req) 469 { 470 return atomic_inc_not_zero(&req->ref); 471 } 472 473 static inline bool req_ref_put_and_test(struct request *req) 474 { 475 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req)); 476 return atomic_dec_and_test(&req->ref); 477 } 478 479 static inline void req_ref_set(struct request *req, int value) 480 { 481 atomic_set(&req->ref, value); 482 } 483 484 static inline int req_ref_read(struct request *req) 485 { 486 return atomic_read(&req->ref); 487 } 488 489 #endif /* BLK_INTERNAL_H */ 490