the-tree/block/blk.h

0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 #ifndef BLK_INTERNAL_H
0003 #define BLK_INTERNAL_H
0004
0005 #include <linux/blk-crypto.h>
0006 #include <linux/memblock.h> /* for max_pfn/max_low_pfn */
0007 #include <xen/xen.h>
0008 #include "blk-crypto-internal.h"
0009
0010 struct elevator_type;
0011
0012 /* Max future timer expiry for timeouts */
0013 #define BLK_MAX_TIMEOUT     (5 * HZ)
0014
0015 extern struct dentry *blk_debugfs_root;
0016
0017 struct blk_flush_queue {
0018     unsigned int        flush_pending_idx:1;
0019     unsigned int        flush_running_idx:1;
0020     blk_status_t        rq_status;
0021     unsigned long       flush_pending_since;
0022     struct list_head    flush_queue[2];
0023     struct list_head    flush_data_in_flight;
0024     struct request      *flush_rq;
0025
0026     spinlock_t      mq_flush_lock;
0027 };
0028
0029 extern struct kmem_cache *blk_requestq_cachep;
0030 extern struct kmem_cache *blk_requestq_srcu_cachep;
0031 extern struct kobj_type blk_queue_ktype;
0032 extern struct ida blk_queue_ida;
0033
0034 bool is_flush_rq(struct request *req);
0035
0036 struct blk_flush_queue *blk_alloc_flush_queue(int node, int cmd_size,
0037                           gfp_t flags);
0038 void blk_free_flush_queue(struct blk_flush_queue *q);
0039
0040 void blk_freeze_queue(struct request_queue *q);
0041 void __blk_mq_unfreeze_queue(struct request_queue *q, bool force_atomic);
0042 void blk_queue_start_drain(struct request_queue *q);
0043 int __bio_queue_enter(struct request_queue *q, struct bio *bio);
0044 void submit_bio_noacct_nocheck(struct bio *bio);
0045
0046 static inline bool blk_try_enter_queue(struct request_queue *q, bool pm)
0047 {
0048     rcu_read_lock();
0049     if (!percpu_ref_tryget_live_rcu(&q->q_usage_counter))
0050         goto fail;
0051
0052     /*
0053      * The code that increments the pm_only counter must ensure that the
0054      * counter is globally visible before the queue is unfrozen.
0055      */
0056     if (blk_queue_pm_only(q) &&
0057         (!pm || queue_rpm_status(q) == RPM_SUSPENDED))
0058         goto fail_put;
0059
0060     rcu_read_unlock();
0061     return true;
0062
0063 fail_put:
0064     blk_queue_exit(q);
0065 fail:
0066     rcu_read_unlock();
0067     return false;
0068 }
0069
0070 static inline int bio_queue_enter(struct bio *bio)
0071 {
0072     struct request_queue *q = bdev_get_queue(bio->bi_bdev);
0073
0074     if (blk_try_enter_queue(q, false))
0075         return 0;
0076     return __bio_queue_enter(q, bio);
0077 }
0078
0079 #define BIO_INLINE_VECS 4
0080 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
0081         gfp_t gfp_mask);
0082 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs);
0083
0084 static inline bool biovec_phys_mergeable(struct request_queue *q,
0085         struct bio_vec *vec1, struct bio_vec *vec2)
0086 {
0087     unsigned long mask = queue_segment_boundary(q);
0088     phys_addr_t addr1 = page_to_phys(vec1->bv_page) + vec1->bv_offset;
0089     phys_addr_t addr2 = page_to_phys(vec2->bv_page) + vec2->bv_offset;
0090
0091     if (addr1 + vec1->bv_len != addr2)
0092         return false;
0093     if (xen_domain() && !xen_biovec_phys_mergeable(vec1, vec2->bv_page))
0094         return false;
0095     if ((addr1 | mask) != ((addr2 + vec2->bv_len - 1) | mask))
0096         return false;
0097     return true;
0098 }
0099
0100 static inline bool __bvec_gap_to_prev(struct queue_limits *lim,
0101         struct bio_vec *bprv, unsigned int offset)
0102 {
0103     return (offset & lim->virt_boundary_mask) ||
0104         ((bprv->bv_offset + bprv->bv_len) & lim->virt_boundary_mask);
0105 }
0106
0107 /*
0108  * Check if adding a bio_vec after bprv with offset would create a gap in
0109  * the SG list. Most drivers don't care about this, but some do.
0110  */
0111 static inline bool bvec_gap_to_prev(struct queue_limits *lim,
0112         struct bio_vec *bprv, unsigned int offset)
0113 {
0114     if (!lim->virt_boundary_mask)
0115         return false;
0116     return __bvec_gap_to_prev(lim, bprv, offset);
0117 }
0118
0119 static inline bool rq_mergeable(struct request *rq)
0120 {
0121     if (blk_rq_is_passthrough(rq))
0122         return false;
0123
0124     if (req_op(rq) == REQ_OP_FLUSH)
0125         return false;
0126
0127     if (req_op(rq) == REQ_OP_WRITE_ZEROES)
0128         return false;
0129
0130     if (req_op(rq) == REQ_OP_ZONE_APPEND)
0131         return false;
0132
0133     if (rq->cmd_flags & REQ_NOMERGE_FLAGS)
0134         return false;
0135     if (rq->rq_flags & RQF_NOMERGE_FLAGS)
0136         return false;
0137
0138     return true;
0139 }
0140
0141 /*
0142  * There are two different ways to handle DISCARD merges:
0143  *  1) If max_discard_segments > 1, the driver treats every bio as a range and
0144  *     send the bios to controller together. The ranges don't need to be
0145  *     contiguous.
0146  *  2) Otherwise, the request will be normal read/write requests.  The ranges
0147  *     need to be contiguous.
0148  */
0149 static inline bool blk_discard_mergable(struct request *req)
0150 {
0151     if (req_op(req) == REQ_OP_DISCARD &&
0152         queue_max_discard_segments(req->q) > 1)
0153         return true;
0154     return false;
0155 }
0156
0157 static inline unsigned int blk_queue_get_max_sectors(struct request_queue *q,
0158                              enum req_op op)
0159 {
0160     if (unlikely(op == REQ_OP_DISCARD || op == REQ_OP_SECURE_ERASE))
0161         return min(q->limits.max_discard_sectors,
0162                UINT_MAX >> SECTOR_SHIFT);
0163
0164     if (unlikely(op == REQ_OP_WRITE_ZEROES))
0165         return q->limits.max_write_zeroes_sectors;
0166
0167     return q->limits.max_sectors;
0168 }
0169
0170 #ifdef CONFIG_BLK_DEV_INTEGRITY
0171 void blk_flush_integrity(void);
0172 bool __bio_integrity_endio(struct bio *);
0173 void bio_integrity_free(struct bio *bio);
0174 static inline bool bio_integrity_endio(struct bio *bio)
0175 {
0176     if (bio_integrity(bio))
0177         return __bio_integrity_endio(bio);
0178     return true;
0179 }
0180
0181 bool blk_integrity_merge_rq(struct request_queue *, struct request *,
0182         struct request *);
0183 bool blk_integrity_merge_bio(struct request_queue *, struct request *,
0184         struct bio *);
0185
0186 static inline bool integrity_req_gap_back_merge(struct request *req,
0187         struct bio *next)
0188 {
0189     struct bio_integrity_payload *bip = bio_integrity(req->bio);
0190     struct bio_integrity_payload *bip_next = bio_integrity(next);
0191
0192     return bvec_gap_to_prev(&req->q->limits,
0193                 &bip->bip_vec[bip->bip_vcnt - 1],
0194                 bip_next->bip_vec[0].bv_offset);
0195 }
0196
0197 static inline bool integrity_req_gap_front_merge(struct request *req,
0198         struct bio *bio)
0199 {
0200     struct bio_integrity_payload *bip = bio_integrity(bio);
0201     struct bio_integrity_payload *bip_next = bio_integrity(req->bio);
0202
0203     return bvec_gap_to_prev(&req->q->limits,
0204                 &bip->bip_vec[bip->bip_vcnt - 1],
0205                 bip_next->bip_vec[0].bv_offset);
0206 }
0207
0208 int blk_integrity_add(struct gendisk *disk);
0209 void blk_integrity_del(struct gendisk *);
0210 #else /* CONFIG_BLK_DEV_INTEGRITY */
0211 static inline bool blk_integrity_merge_rq(struct request_queue *rq,
0212         struct request *r1, struct request *r2)
0213 {
0214     return true;
0215 }
0216 static inline bool blk_integrity_merge_bio(struct request_queue *rq,
0217         struct request *r, struct bio *b)
0218 {
0219     return true;
0220 }
0221 static inline bool integrity_req_gap_back_merge(struct request *req,
0222         struct bio *next)
0223 {
0224     return false;
0225 }
0226 static inline bool integrity_req_gap_front_merge(struct request *req,
0227         struct bio *bio)
0228 {
0229     return false;
0230 }
0231
0232 static inline void blk_flush_integrity(void)
0233 {
0234 }
0235 static inline bool bio_integrity_endio(struct bio *bio)
0236 {
0237     return true;
0238 }
0239 static inline void bio_integrity_free(struct bio *bio)
0240 {
0241 }
0242 static inline int blk_integrity_add(struct gendisk *disk)
0243 {
0244     return 0;
0245 }
0246 static inline void blk_integrity_del(struct gendisk *disk)
0247 {
0248 }
0249 #endif /* CONFIG_BLK_DEV_INTEGRITY */
0250
0251 unsigned long blk_rq_timeout(unsigned long timeout);
0252 void blk_add_timer(struct request *req);
0253 const char *blk_status_to_str(blk_status_t status);
0254
0255 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
0256         unsigned int nr_segs);
0257 bool blk_bio_list_merge(struct request_queue *q, struct list_head *list,
0258             struct bio *bio, unsigned int nr_segs);
0259
0260 /*
0261  * Plug flush limits
0262  */
0263 #define BLK_MAX_REQUEST_COUNT   32
0264 #define BLK_PLUG_FLUSH_SIZE (128 * 1024)
0265
0266 /*
0267  * Internal elevator interface
0268  */
0269 #define ELV_ON_HASH(rq) ((rq)->rq_flags & RQF_HASHED)
0270
0271 void blk_insert_flush(struct request *rq);
0272
0273 int elevator_switch_mq(struct request_queue *q,
0274                   struct elevator_type *new_e);
0275 void elevator_exit(struct request_queue *q);
0276 int elv_register_queue(struct request_queue *q, bool uevent);
0277 void elv_unregister_queue(struct request_queue *q);
0278
0279 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
0280         char *buf);
0281 ssize_t part_stat_show(struct device *dev, struct device_attribute *attr,
0282         char *buf);
0283 ssize_t part_inflight_show(struct device *dev, struct device_attribute *attr,
0284         char *buf);
0285 ssize_t part_fail_show(struct device *dev, struct device_attribute *attr,
0286         char *buf);
0287 ssize_t part_fail_store(struct device *dev, struct device_attribute *attr,
0288         const char *buf, size_t count);
0289 ssize_t part_timeout_show(struct device *, struct device_attribute *, char *);
0290 ssize_t part_timeout_store(struct device *, struct device_attribute *,
0291                 const char *, size_t);
0292
0293 static inline bool bio_may_exceed_limits(struct bio *bio,
0294         struct queue_limits *lim)
0295 {
0296     switch (bio_op(bio)) {
0297     case REQ_OP_DISCARD:
0298     case REQ_OP_SECURE_ERASE:
0299     case REQ_OP_WRITE_ZEROES:
0300         return true; /* non-trivial splitting decisions */
0301     default:
0302         break;
0303     }
0304
0305     /*
0306      * All drivers must accept single-segments bios that are <= PAGE_SIZE.
0307      * This is a quick and dirty check that relies on the fact that
0308      * bi_io_vec[0] is always valid if a bio has data.  The check might
0309      * lead to occasional false negatives when bios are cloned, but compared
0310      * to the performance impact of cloned bios themselves the loop below
0311      * doesn't matter anyway.
0312      */
0313     return lim->chunk_sectors || bio->bi_vcnt != 1 ||
0314         bio->bi_io_vec->bv_len + bio->bi_io_vec->bv_offset > PAGE_SIZE;
0315 }
0316
0317 struct bio *__bio_split_to_limits(struct bio *bio, struct queue_limits *lim,
0318                unsigned int *nr_segs);
0319 int ll_back_merge_fn(struct request *req, struct bio *bio,
0320         unsigned int nr_segs);
0321 bool blk_attempt_req_merge(struct request_queue *q, struct request *rq,
0322                 struct request *next);
0323 unsigned int blk_recalc_rq_segments(struct request *rq);
0324 void blk_rq_set_mixed_merge(struct request *rq);
0325 bool blk_rq_merge_ok(struct request *rq, struct bio *bio);
0326 enum elv_merge blk_try_merge(struct request *rq, struct bio *bio);
0327
0328 int blk_dev_init(void);
0329
0330 /*
0331  * Contribute to IO statistics IFF:
0332  *
0333  *  a) it's attached to a gendisk, and
0334  *  b) the queue had IO stats enabled when this request was started
0335  */
0336 static inline bool blk_do_io_stat(struct request *rq)
0337 {
0338     return (rq->rq_flags & RQF_IO_STAT) && !blk_rq_is_passthrough(rq);
0339 }
0340
0341 void update_io_ticks(struct block_device *part, unsigned long now, bool end);
0342
0343 static inline void req_set_nomerge(struct request_queue *q, struct request *req)
0344 {
0345     req->cmd_flags |= REQ_NOMERGE;
0346     if (req == q->last_merge)
0347         q->last_merge = NULL;
0348 }
0349
0350 /*
0351  * Internal io_context interface
0352  */
0353 struct io_cq *ioc_find_get_icq(struct request_queue *q);
0354 struct io_cq *ioc_lookup_icq(struct request_queue *q);
0355 #ifdef CONFIG_BLK_ICQ
0356 void ioc_clear_queue(struct request_queue *q);
0357 #else
0358 static inline void ioc_clear_queue(struct request_queue *q)
0359 {
0360 }
0361 #endif /* CONFIG_BLK_ICQ */
0362
0363 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
0364 extern ssize_t blk_throtl_sample_time_show(struct request_queue *q, char *page);
0365 extern ssize_t blk_throtl_sample_time_store(struct request_queue *q,
0366     const char *page, size_t count);
0367 extern void blk_throtl_bio_endio(struct bio *bio);
0368 extern void blk_throtl_stat_add(struct request *rq, u64 time);
0369 #else
0370 static inline void blk_throtl_bio_endio(struct bio *bio) { }
0371 static inline void blk_throtl_stat_add(struct request *rq, u64 time) { }
0372 #endif
0373
0374 struct bio *__blk_queue_bounce(struct bio *bio, struct request_queue *q);
0375
0376 static inline bool blk_queue_may_bounce(struct request_queue *q)
0377 {
0378     return IS_ENABLED(CONFIG_BOUNCE) &&
0379         q->limits.bounce == BLK_BOUNCE_HIGH &&
0380         max_low_pfn >= max_pfn;
0381 }
0382
0383 static inline struct bio *blk_queue_bounce(struct bio *bio,
0384         struct request_queue *q)
0385 {
0386     if (unlikely(blk_queue_may_bounce(q) && bio_has_data(bio)))
0387         return __blk_queue_bounce(bio, q);
0388     return bio;
0389 }
0390
0391 #ifdef CONFIG_BLK_CGROUP_IOLATENCY
0392 extern int blk_iolatency_init(struct request_queue *q);
0393 #else
0394 static inline int blk_iolatency_init(struct request_queue *q) { return 0; }
0395 #endif
0396
0397 #ifdef CONFIG_BLK_DEV_ZONED
0398 void disk_free_zone_bitmaps(struct gendisk *disk);
0399 void disk_clear_zone_settings(struct gendisk *disk);
0400 #else
0401 static inline void disk_free_zone_bitmaps(struct gendisk *disk) {}
0402 static inline void disk_clear_zone_settings(struct gendisk *disk) {}
0403 #endif
0404
0405 int blk_alloc_ext_minor(void);
0406 void blk_free_ext_minor(unsigned int minor);
0407 #define ADDPART_FLAG_NONE   0
0408 #define ADDPART_FLAG_RAID   1
0409 #define ADDPART_FLAG_WHOLEDISK  2
0410 int bdev_add_partition(struct gendisk *disk, int partno, sector_t start,
0411         sector_t length);
0412 int bdev_del_partition(struct gendisk *disk, int partno);
0413 int bdev_resize_partition(struct gendisk *disk, int partno, sector_t start,
0414         sector_t length);
0415 void blk_drop_partitions(struct gendisk *disk);
0416
0417 struct gendisk *__alloc_disk_node(struct request_queue *q, int node_id,
0418         struct lock_class_key *lkclass);
0419
0420 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
0421         struct page *page, unsigned int len, unsigned int offset,
0422         unsigned int max_sectors, bool *same_page);
0423
0424 static inline struct kmem_cache *blk_get_queue_kmem_cache(bool srcu)
0425 {
0426     if (srcu)
0427         return blk_requestq_srcu_cachep;
0428     return blk_requestq_cachep;
0429 }
0430 struct request_queue *blk_alloc_queue(int node_id, bool alloc_srcu);
0431
0432 int disk_scan_partitions(struct gendisk *disk, fmode_t mode);
0433
0434 int disk_alloc_events(struct gendisk *disk);
0435 void disk_add_events(struct gendisk *disk);
0436 void disk_del_events(struct gendisk *disk);
0437 void disk_release_events(struct gendisk *disk);
0438 void disk_block_events(struct gendisk *disk);
0439 void disk_unblock_events(struct gendisk *disk);
0440 void disk_flush_events(struct gendisk *disk, unsigned int mask);
0441 extern struct device_attribute dev_attr_events;
0442 extern struct device_attribute dev_attr_events_async;
0443 extern struct device_attribute dev_attr_events_poll_msecs;
0444
0445 extern struct attribute_group blk_trace_attr_group;
0446
0447 long blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
0448 long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg);
0449
0450 extern const struct address_space_operations def_blk_aops;
0451
0452 int disk_register_independent_access_ranges(struct gendisk *disk);
0453 void disk_unregister_independent_access_ranges(struct gendisk *disk);
0454
0455 #ifdef CONFIG_FAIL_MAKE_REQUEST
0456 bool should_fail_request(struct block_device *part, unsigned int bytes);
0457 #else /* CONFIG_FAIL_MAKE_REQUEST */
0458 static inline bool should_fail_request(struct block_device *part,
0459                     unsigned int bytes)
0460 {
0461     return false;
0462 }
0463 #endif /* CONFIG_FAIL_MAKE_REQUEST */
0464
0465 /*
0466  * Optimized request reference counting. Ideally we'd make timeouts be more
0467  * clever, as that's the only reason we need references at all... But until
0468  * this happens, this is faster than using refcount_t. Also see:
0469  *
0470  * abc54d634334 ("io_uring: switch to atomic_t for io_kiocb reference count")
0471  */
0472 #define req_ref_zero_or_close_to_overflow(req)  \
0473     ((unsigned int) atomic_read(&(req->ref)) + 127u <= 127u)
0474
0475 static inline bool req_ref_inc_not_zero(struct request *req)
0476 {
0477     return atomic_inc_not_zero(&req->ref);
0478 }
0479
0480 static inline bool req_ref_put_and_test(struct request *req)
0481 {
0482     WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
0483     return atomic_dec_and_test(&req->ref);
0484 }
0485
0486 static inline void req_ref_set(struct request *req, int value)
0487 {
0488     atomic_set(&req->ref, value);
0489 }
0490
0491 static inline int req_ref_read(struct request *req)
0492 {
0493     return atomic_read(&req->ref);
0494 }
0495
0496 #endif /* BLK_INTERNAL_H */