block/blk.h

   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 extern struct kmem_cache *blk_requestq_cachep;
  30 extern struct kmem_cache *blk_requestq_srcu_cachep;
  31 extern struct kobj_type blk_queue_ktype;
  32 extern struct ida blk_queue_ida;
  33
  34 bool is_flush_rq(struct request *req);
  35
  36 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
  37                                               gfp_t flags);
  38 void blk_free_flush_queue(struct blk_flush_queue *q);
  39
  40 void blk_freeze_queue(struct request_queue *q);
  41 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
  42 void blk_queue_start_drain(struct request_queue *q);
  43 int __bio_queue_enter(struct request_queue *q, struct bio *bio);
  44 void submit_bio_noacct_nocheck(struct bio *bio);
  45
  46 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
  47 {
  48         rcu_read_lock();
  49         if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
  50                 goto fail;
  51
  52         /*
  53          * The code that increments the pm_only counter must ensure that the
  54          * counter is globally visible before the queue is unfrozen.
  55          */
  56         if (blk_queue_pm_only(q) &&
  57             (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
  58                 goto fail_put;
  59
  60         rcu_read_unlock();
  61         return true;
  62
  63 fail_put:
  64         blk_queue_exit(q);
  65 fail:
  66         rcu_read_unlock();
  67         return false;
  68 }
  69
  70 static inline int bio_queue_enter(struct bio *bio)
  71 {
  72         struct request_queue *q = bdev_get_queue(bio->bi_bdev);
  73
  74         if (blk_try_enter_queue(q, false))
  75                 return 0;
  76         return __bio_queue_enter(q, bio);
  77 }
  78
  79 #define BIO_INLINE_VECS 4
  80 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
  81                 gfp_t gfp_mask);
  82 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
  83
  84 static inline bool biovec_phys_mergeable(struct request_queue *q,
  85                 struct bio_vec *vec1, struct bio_vec *vec2)
  86 {
  87         unsigned long mask = queue_segment_boundary(q);
  88         phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
  89         phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
  90
  91         if (addr1 + vec1->bv_len != addr2)
  92                 return false;
  93         if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
  94                 return false;
  95         if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
  96                 return false;
  97         return true;
  98 }
  99
 100 static inline bool __bvec_gap_to_prev(struct queue_limits *lim,
 101                 struct bio_vec *bprv, unsigned int offset)
 102 {
 103         return (offset & lim->virt_boundary_mask) ||
 104                 ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
 105 }
 106
 107 /*
 108  * Check if adding a bio_vec after bprv with offset would create a gap in
 109  * the SG list. Most drivers don't care about this, but some do.
 110  */
 111 static inline bool bvec_gap_to_prev(struct queue_limits *lim,
 112                 struct bio_vec *bprv, unsigned int offset)
 113 {
 114         if (!lim->virt_boundary_mask)
 115                 return false;
 116         return __bvec_gap_to_prev(lim, bprv, offset);
 117 }
 118
 119 static inline bool rq_mergeable(struct request *rq)
 120 {
 121         if (blk_rq_is_passthrough(rq))
 122                 return false;
 123
 124         if (req_op(rq) == REQ_OP_FLUSH)
 125                 return false;
 126
 127         if (req_op(rq) == REQ_OP_WRITE_ZEROES)
 128                 return false;
 129
 130         if (req_op(rq) == REQ_OP_ZONE_APPEND)
 131                 return false;
 132
 133         if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
 134                 return false;
 135         if (rq->rq_flags & RQF_NOMERGE_FLAGS)
 136                 return false;
 137
 138         return true;
 139 }
 140
 141 /*
 142  * There are two different ways to handle DISCARD merges:
 143  *  1) If max_discard_segments > 1, the driver treats every bio as a range and
 144  *     send the bios to controller together. The ranges don't need to be
 145  *     contiguous.
 146  *  2) Otherwise, the request will be normal read/write requests.  The ranges
 147  *     need to be contiguous.
 148  */
 149 static inline bool blk_discard_mergable(struct request *req)
 150 {
 151         if (req_op(req) == REQ_OP_DISCARD &&
 152             queue_max_discard_segments(req->q) > 1)
 153                 return true;
 154         return false;
 155 }
 156
 157 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
 158                                                      enum req_op op)
 159 {
 160         if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
 161                 return min(q->limits.max_discard_sectors,
 162                            UINT_MAX >> SECTOR_SHIFT);
 163
 164         if (unlikely(op == REQ_OP_WRITE_ZEROES))
 165                 return q->limits.max_write_zeroes_sectors;
 166
 167         return q->limits.max_sectors;
 168 }
 169
 170 #ifdef CONFIG_BLK_DEV_INTEGRITY
 171 void blk_flush_integrity(void);
 172 bool __bio_integrity_endio(struct bio *);
 173 void bio_integrity_free(struct bio *bio);
 174 static inline bool bio_integrity_endio(struct bio *bio)
 175 {
 176         if (bio_integrity(bio))
 177                 return __bio_integrity_endio(bio);
 178         return true;
 179 }
 180
 181 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
 182                 struct request *);
 183 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
 184                 struct bio *);
 185
 186 static inline bool integrity_req_gap_back_merge(struct request *req,
 187                 struct bio *next)
 188 {
 189         struct bio_integrity_payload *bip = bio_integrity(req->bio);
 190         struct bio_integrity_payload *bip_next = bio_integrity(next);
 191
 192         return bvec_gap_to_prev(&req->q->limits,
 193                                 &bip->bip_vec[bip->bip_vcnt - 1],
 194                                 bip_next->bip_vec[0].bv_offset);
 195 }
 196
 197 static inline bool integrity_req_gap_front_merge(struct request *req,
 198                 struct bio *bio)
 199 {
 200         struct bio_integrity_payload *bip = bio_integrity(bio);
 201         struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
 202
 203         return bvec_gap_to_prev(&req->q->limits,
 204                                 &bip->bip_vec[bip->bip_vcnt - 1],
 205                                 bip_next->bip_vec[0].bv_offset);
 206 }
 207
 208 int blk_integrity_add(struct gendisk *disk);
 209 void blk_integrity_del(struct gendisk *);
 210 #else /* CONFIG_BLK_DEV_INTEGRITY */
 211 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
 212                 struct request *r1, struct request *r2)
 213 {
 214         return true;
 215 }
 216 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
 217                 struct request *r, struct bio *b)
 218 {
 219         return true;
 220 }
 221 static inline bool integrity_req_gap_back_merge(struct request *req,
 222                 struct bio *next)
 223 {
 224         return false;
 225 }
 226 static inline bool integrity_req_gap_front_merge(struct request *req,
 227                 struct bio *bio)
 228 {
 229         return false;
 230 }
 231
 232 static inline void blk_flush_integrity(void)
 233 {
 234 }
 235 static inline bool bio_integrity_endio(struct bio *bio)
 236 {
 237         return true;
 238 }
 239 static inline void bio_integrity_free(struct bio *bio)
 240 {
 241 }
 242 static inline int blk_integrity_add(struct gendisk *disk)
 243 {
 244         return 0;
 245 }
 246 static inline void blk_integrity_del(struct gendisk *disk)
 247 {
 248 }
 249 #endif /* CONFIG_BLK_DEV_INTEGRITY */
 250
 251 unsigned long blk_rq_timeout(unsigned long timeout);
 252 void blk_add_timer(struct request *req);
 253 const char *blk_status_to_str(blk_status_t status);
 254
 255 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
 256                 unsigned int nr_segs);
 257 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
 258                         struct bio *bio, unsigned int nr_segs);
 259
 260 /*
 261  * Plug flush limits
 262  */
 263 #define BLK_MAX_REQUEST_COUNT   32
 264 #define BLK_PLUG_FLUSH_SIZE     (128 * 1024)
 265
 266 /*
 267  * Internal elevator interface
 268  */
 269 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
 270
 271 void blk_insert_flush(struct request *rq);
 272
 273 int elevator_switch_mq(struct request_queue *q,
 274                               struct elevator_type *new_e);
 275 void elevator_exit(struct request_queue *q);
 276 int elv_register_queue(struct request_queue *q, bool uevent);
 277 void elv_unregister_queue(struct request_queue *q);
 278
 279 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
 280                 char *buf);
 281 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
 282                 char *buf);
 283 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
 284                 char *buf);
 285 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
 286                 char *buf);
 287 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
 288                 const char *buf, size_t count);
 289 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
 290 ssize_t part_timeout_store(struct device *, struct device_attribute *,
 291                                 const char *, size_t);
 292
 293 static inline bool bio_may_exceed_limits(struct bio *bio,
 294                 struct queue_limits *lim)
 295 {
 296         switch (bio_op(bio)) {
 297         case REQ_OP_DISCARD:
 298         case REQ_OP_SECURE_ERASE:
 299         case REQ_OP_WRITE_ZEROES:
 300                 return true; /* non-trivial splitting decisions */
 301         default:
 302                 break;
 303         }
 304
 305         /*
 306          * All drivers must accept single-segments bios that are <= PAGE_SIZE.
 307          * This is a quick and dirty check that relies on the fact that
 308          * bi_io_vec[0] is always valid if a bio has data.  The check might
 309          * lead to occasional false negatives when bios are cloned, but compared
 310          * to the performance impact of cloned bios themselves the loop below
 311          * doesn't matter anyway.
 312          */
 313         return lim->chunk_sectors || bio->bi_vcnt != 1 ||
 314                 bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
 315 }
 316
 317 struct bio *__bio_split_to_limits(struct bio *bio, struct queue_limits *lim,
 318                        unsigned int *nr_segs);
 319 int ll_back_merge_fn(struct request *req, struct bio *bio,
 320                 unsigned int nr_segs);
 321 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
 322                                 struct request *next);
 323 unsigned int blk_recalc_rq_segments(struct request *rq);
 324 void blk_rq_set_mixed_merge(struct request *rq);
 325 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
 326 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
 327
 328 int blk_dev_init(void);
 329
 330 /*
 331  * Contribute to IO statistics IFF:
 332  *
 333  *      a) it's attached to a gendisk, and
 334  *      b) the queue had IO stats enabled when this request was started
 335  */
 336 static inline bool blk_do_io_stat(struct request *rq)
 337 {
 338         return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
 339 }
 340
 341 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
 342
 343 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
 344 {
 345         req->cmd_flags |= REQ_NOMERGE;
 346         if (req == q->last_merge)
 347                 q->last_merge = NULL;
 348 }
 349
 350 /*
 351  * Internal io_context interface
 352  */
 353 struct io_cq *ioc_find_get_icq(struct request_queue *q);
 354 struct io_cq *ioc_lookup_icq(struct request_queue *q);
 355 #ifdef CONFIG_BLK_ICQ
 356 void ioc_clear_queue(struct request_queue *q);
 357 #else
 358 static inline void ioc_clear_queue(struct request_queue *q)
 359 {
 360 }
 361 #endif /* CONFIG_BLK_ICQ */
 362
 363 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 364 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
 365 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
 366         const char *page, size_t count);
 367 extern void blk_throtl_bio_endio(struct bio *bio);
 368 extern void blk_throtl_stat_add(struct request *rq, u64 time);
 369 #else
 370 static inline void blk_throtl_bio_endio(struct bio *bio) { }
 371 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
 372 #endif
 373
 374 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
 375
 376 static inline bool blk_queue_may_bounce(struct request_queue *q)
 377 {
 378         return IS_ENABLED(CONFIG_BOUNCE) &&
 379                 q->limits.bounce == BLK_BOUNCE_HIGH &&
 380                 max_low_pfn >= max_pfn;
 381 }
 382
 383 static inline struct bio *blk_queue_bounce(struct bio *bio,
 384                 struct request_queue *q)
 385 {
 386         if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
 387                 return __blk_queue_bounce(bio, q);
 388         return bio;
 389 }
 390
 391 #ifdef CONFIG_BLK_CGROUP_IOLATENCY
 392 extern int blk_iolatency_init(struct request_queue *q);
 393 #else
 394 static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
 395 #endif
 396
 397 #ifdef CONFIG_BLK_DEV_ZONED
 398 void disk_free_zone_bitmaps(struct gendisk *disk);
 399 void disk_clear_zone_settings(struct gendisk *disk);
 400 #else
 401 static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
 402 static inline void disk_clear_zone_settings(struct gendisk *disk) {}
 403 #endif
 404
 405 int blk_alloc_ext_minor(void);
 406 void blk_free_ext_minor(unsigned int minor);
 407 #define ADDPART_FLAG_NONE       0
 408 #define ADDPART_FLAG_RAID       1
 409 #define ADDPART_FLAG_WHOLEDISK  2
 410 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
 411                 sector_t length);
 412 int bdev_del_partition(struct gendisk *disk, int partno);
 413 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
 414                 sector_t length);
 415 void blk_drop_partitions(struct gendisk *disk);
 416
 417 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
 418                 struct lock_class_key *lkclass);
 419
 420 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
 421                 struct page *page, unsigned int len, unsigned int offset,
 422                 unsigned int max_sectors, bool *same_page);
 423
 424 static inline struct kmem_cache *blk_get_queue_kmem_cache(bool srcu)
 425 {
 426         if (srcu)
 427                 return blk_requestq_srcu_cachep;
 428         return blk_requestq_cachep;
 429 }
 430 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu);
 431
 432 int disk_scan_partitions(struct gendisk *disk, fmode_t mode);
 433
 434 int disk_alloc_events(struct gendisk *disk);
 435 void disk_add_events(struct gendisk *disk);
 436 void disk_del_events(struct gendisk *disk);
 437 void disk_release_events(struct gendisk *disk);
 438 void disk_block_events(struct gendisk *disk);
 439 void disk_unblock_events(struct gendisk *disk);
 440 void disk_flush_events(struct gendisk *disk, unsigned int mask);
 441 extern struct device_attribute dev_attr_events;
 442 extern struct device_attribute dev_attr_events_async;
 443 extern struct device_attribute dev_attr_events_poll_msecs;
 444
 445 extern struct attribute_group blk_trace_attr_group;
 446
 447 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
 448 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
 449
 450 extern const struct address_space_operations def_blk_aops;
 451
 452 int disk_register_independent_access_ranges(struct gendisk *disk);
 453 void disk_unregister_independent_access_ranges(struct gendisk *disk);
 454
 455 #ifdef CONFIG_FAIL_MAKE_REQUEST
 456 bool should_fail_request(struct block_device *part, unsigned int bytes);
 457 #else /* CONFIG_FAIL_MAKE_REQUEST */
 458 static inline bool should_fail_request(struct block_device *part,
 459                                         unsigned int bytes)
 460 {
 461         return false;
 462 }
 463 #endif /* CONFIG_FAIL_MAKE_REQUEST */
 464
 465 /*
 466  * Optimized request reference counting. Ideally we'd make timeouts be more
 467  * clever, as that's the only reason we need references at all... But until
 468  * this happens, this is faster than using refcount_t. Also see:
 469  *
 470  * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
 471  */
 472 #define req_ref_zero_or_close_to_overflow(req)  \
 473         ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
 474
 475 static inline bool req_ref_inc_not_zero(struct request *req)
 476 {
 477         return atomic_inc_not_zero(&req->ref);
 478 }
 479
 480 static inline bool req_ref_put_and_test(struct request *req)
 481 {
 482         WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
 483         return atomic_dec_and_test(&req->ref);
 484 }
 485
 486 static inline void req_ref_set(struct request *req, int value)
 487 {
 488         atomic_set(&req->ref, value);
 489 }
 490
 491 static inline int req_ref_read(struct request *req)
 492 {
 493         return atomic_read(&req->ref);
 494 }
 495
 496 #endif /* BLK_INTERNAL_H */