18c16567dSChristoph Hellwig // SPDX-License-Identifier: GPL-2.0 2f9c78b2bSJens Axboe /* 3f9c78b2bSJens Axboe * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 4f9c78b2bSJens Axboe */ 5f9c78b2bSJens Axboe #include <linux/mm.h> 6f9c78b2bSJens Axboe #include <linux/swap.h> 7f9c78b2bSJens Axboe #include <linux/bio.h> 8f9c78b2bSJens Axboe #include <linux/blkdev.h> 9f9c78b2bSJens Axboe #include <linux/uio.h> 10f9c78b2bSJens Axboe #include <linux/iocontext.h> 11f9c78b2bSJens Axboe #include <linux/slab.h> 12f9c78b2bSJens Axboe #include <linux/init.h> 13f9c78b2bSJens Axboe #include <linux/kernel.h> 14f9c78b2bSJens Axboe #include <linux/export.h> 15f9c78b2bSJens Axboe #include <linux/mempool.h> 16f9c78b2bSJens Axboe #include <linux/workqueue.h> 17f9c78b2bSJens Axboe #include <linux/cgroup.h> 18b4c5875dSDamien Le Moal #include <linux/highmem.h> 19de6a78b6SMing Lei #include <linux/sched/sysctl.h> 20a892c8d5SSatya Tangirala #include <linux/blk-crypto.h> 2149d1ec85SMing Lei #include <linux/xarray.h> 22f9c78b2bSJens Axboe 23f9c78b2bSJens Axboe #include <trace/events/block.h> 249e234eeaSShaohua Li #include "blk.h" 2567b42d0bSJosef Bacik #include "blk-rq-qos.h" 26672fdcf0SMing Lei #include "blk-cgroup.h" 27f9c78b2bSJens Axboe 28be4d234dSJens Axboe struct bio_alloc_cache { 29fcade2ceSJens Axboe struct bio *free_list; 30be4d234dSJens Axboe unsigned int nr; 31be4d234dSJens Axboe }; 32be4d234dSJens Axboe 33de76fd89SChristoph Hellwig static struct biovec_slab { 346ac0b715SChristoph Hellwig int nr_vecs; 356ac0b715SChristoph Hellwig char *name; 366ac0b715SChristoph Hellwig struct kmem_cache *slab; 37de76fd89SChristoph Hellwig } bvec_slabs[] __read_mostly = { 38de76fd89SChristoph Hellwig { .nr_vecs = 16, .name = "biovec-16" }, 39de76fd89SChristoph Hellwig { .nr_vecs = 64, .name = "biovec-64" }, 40de76fd89SChristoph Hellwig { .nr_vecs = 128, .name = "biovec-128" }, 41a8affc03SChristoph Hellwig { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" }, 42f9c78b2bSJens Axboe }; 436ac0b715SChristoph Hellwig 447a800a20SChristoph Hellwig static struct biovec_slab *biovec_slab(unsigned short nr_vecs) 457a800a20SChristoph Hellwig { 467a800a20SChristoph Hellwig switch (nr_vecs) { 477a800a20SChristoph Hellwig /* smaller bios use inline vecs */ 487a800a20SChristoph Hellwig case 5 ... 16: 497a800a20SChristoph Hellwig return &bvec_slabs[0]; 507a800a20SChristoph Hellwig case 17 ... 64: 517a800a20SChristoph Hellwig return &bvec_slabs[1]; 527a800a20SChristoph Hellwig case 65 ... 128: 537a800a20SChristoph Hellwig return &bvec_slabs[2]; 54a8affc03SChristoph Hellwig case 129 ... BIO_MAX_VECS: 557a800a20SChristoph Hellwig return &bvec_slabs[3]; 567a800a20SChristoph Hellwig default: 577a800a20SChristoph Hellwig BUG(); 587a800a20SChristoph Hellwig return NULL; 597a800a20SChristoph Hellwig } 607a800a20SChristoph Hellwig } 61f9c78b2bSJens Axboe 62f9c78b2bSJens Axboe /* 63f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 64f9c78b2bSJens Axboe * IO code that does not need private memory pools. 65f9c78b2bSJens Axboe */ 66f4f8154aSKent Overstreet struct bio_set fs_bio_set; 67f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 68f9c78b2bSJens Axboe 69f9c78b2bSJens Axboe /* 70f9c78b2bSJens Axboe * Our slab pool management 71f9c78b2bSJens Axboe */ 72f9c78b2bSJens Axboe struct bio_slab { 73f9c78b2bSJens Axboe struct kmem_cache *slab; 74f9c78b2bSJens Axboe unsigned int slab_ref; 75f9c78b2bSJens Axboe unsigned int slab_size; 76f9c78b2bSJens Axboe char name[8]; 77f9c78b2bSJens Axboe }; 78f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 7949d1ec85SMing Lei static DEFINE_XARRAY(bio_slabs); 80f9c78b2bSJens Axboe 8149d1ec85SMing Lei static struct bio_slab *create_bio_slab(unsigned int size) 82f9c78b2bSJens Axboe { 8349d1ec85SMing Lei struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); 8449d1ec85SMing Lei 8549d1ec85SMing Lei if (!bslab) 8649d1ec85SMing Lei return NULL; 8749d1ec85SMing Lei 8849d1ec85SMing Lei snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); 8949d1ec85SMing Lei bslab->slab = kmem_cache_create(bslab->name, size, 901a7e76e4SChristoph Hellwig ARCH_KMALLOC_MINALIGN, 911a7e76e4SChristoph Hellwig SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL); 9249d1ec85SMing Lei if (!bslab->slab) 9349d1ec85SMing Lei goto fail_alloc_slab; 9449d1ec85SMing Lei 9549d1ec85SMing Lei bslab->slab_ref = 1; 9649d1ec85SMing Lei bslab->slab_size = size; 9749d1ec85SMing Lei 9849d1ec85SMing Lei if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) 9949d1ec85SMing Lei return bslab; 10049d1ec85SMing Lei 10149d1ec85SMing Lei kmem_cache_destroy(bslab->slab); 10249d1ec85SMing Lei 10349d1ec85SMing Lei fail_alloc_slab: 10449d1ec85SMing Lei kfree(bslab); 10549d1ec85SMing Lei return NULL; 10649d1ec85SMing Lei } 10749d1ec85SMing Lei 10849d1ec85SMing Lei static inline unsigned int bs_bio_slab_size(struct bio_set *bs) 10949d1ec85SMing Lei { 1109f180e31SMing Lei return bs->front_pad + sizeof(struct bio) + bs->back_pad; 11149d1ec85SMing Lei } 11249d1ec85SMing Lei 11349d1ec85SMing Lei static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) 11449d1ec85SMing Lei { 11549d1ec85SMing Lei unsigned int size = bs_bio_slab_size(bs); 11649d1ec85SMing Lei struct bio_slab *bslab; 117f9c78b2bSJens Axboe 118f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 11949d1ec85SMing Lei bslab = xa_load(&bio_slabs, size); 12049d1ec85SMing Lei if (bslab) 121f9c78b2bSJens Axboe bslab->slab_ref++; 12249d1ec85SMing Lei else 12349d1ec85SMing Lei bslab = create_bio_slab(size); 124f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 12549d1ec85SMing Lei 12649d1ec85SMing Lei if (bslab) 12749d1ec85SMing Lei return bslab->slab; 12849d1ec85SMing Lei return NULL; 129f9c78b2bSJens Axboe } 130f9c78b2bSJens Axboe 131f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 132f9c78b2bSJens Axboe { 133f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 13449d1ec85SMing Lei unsigned int slab_size = bs_bio_slab_size(bs); 135f9c78b2bSJens Axboe 136f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 137f9c78b2bSJens Axboe 13849d1ec85SMing Lei bslab = xa_load(&bio_slabs, slab_size); 139f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 140f9c78b2bSJens Axboe goto out; 141f9c78b2bSJens Axboe 14249d1ec85SMing Lei WARN_ON_ONCE(bslab->slab != bs->bio_slab); 14349d1ec85SMing Lei 144f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 145f9c78b2bSJens Axboe 146f9c78b2bSJens Axboe if (--bslab->slab_ref) 147f9c78b2bSJens Axboe goto out; 148f9c78b2bSJens Axboe 14949d1ec85SMing Lei xa_erase(&bio_slabs, slab_size); 15049d1ec85SMing Lei 151f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 15249d1ec85SMing Lei kfree(bslab); 153f9c78b2bSJens Axboe 154f9c78b2bSJens Axboe out: 155f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 156f9c78b2bSJens Axboe } 157f9c78b2bSJens Axboe 1587a800a20SChristoph Hellwig void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs) 159f9c78b2bSJens Axboe { 1609e8c0d0dSChristoph Hellwig BUG_ON(nr_vecs > BIO_MAX_VECS); 161f9c78b2bSJens Axboe 162a8affc03SChristoph Hellwig if (nr_vecs == BIO_MAX_VECS) 163f9c78b2bSJens Axboe mempool_free(bv, pool); 1647a800a20SChristoph Hellwig else if (nr_vecs > BIO_INLINE_VECS) 1657a800a20SChristoph Hellwig kmem_cache_free(biovec_slab(nr_vecs)->slab, bv); 166f9c78b2bSJens Axboe } 167f9c78b2bSJens Axboe 168f2c3eb9bSChristoph Hellwig /* 169f2c3eb9bSChristoph Hellwig * Make the first allocation restricted and don't dump info on allocation 170f2c3eb9bSChristoph Hellwig * failures, since we'll fall back to the mempool in case of failure. 171f2c3eb9bSChristoph Hellwig */ 172f2c3eb9bSChristoph Hellwig static inline gfp_t bvec_alloc_gfp(gfp_t gfp) 173f9c78b2bSJens Axboe { 174f2c3eb9bSChristoph Hellwig return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | 175f2c3eb9bSChristoph Hellwig __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 176f2c3eb9bSChristoph Hellwig } 177f2c3eb9bSChristoph Hellwig 1787a800a20SChristoph Hellwig struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, 1797a800a20SChristoph Hellwig gfp_t gfp_mask) 180f9c78b2bSJens Axboe { 1817a800a20SChristoph Hellwig struct biovec_slab *bvs = biovec_slab(*nr_vecs); 1827a800a20SChristoph Hellwig 1837a800a20SChristoph Hellwig if (WARN_ON_ONCE(!bvs)) 184f9c78b2bSJens Axboe return NULL; 1857a800a20SChristoph Hellwig 1867a800a20SChristoph Hellwig /* 1877a800a20SChristoph Hellwig * Upgrade the nr_vecs request to take full advantage of the allocation. 1887a800a20SChristoph Hellwig * We also rely on this in the bvec_free path. 1897a800a20SChristoph Hellwig */ 1907a800a20SChristoph Hellwig *nr_vecs = bvs->nr_vecs; 191f9c78b2bSJens Axboe 192f9c78b2bSJens Axboe /* 193f007a3d6SChristoph Hellwig * Try a slab allocation first for all smaller allocations. If that 194f007a3d6SChristoph Hellwig * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. 195a8affc03SChristoph Hellwig * The mempool is sized to handle up to BIO_MAX_VECS entries. 196f9c78b2bSJens Axboe */ 197a8affc03SChristoph Hellwig if (*nr_vecs < BIO_MAX_VECS) { 198f9c78b2bSJens Axboe struct bio_vec *bvl; 199f9c78b2bSJens Axboe 200f2c3eb9bSChristoph Hellwig bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); 2017a800a20SChristoph Hellwig if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) 202f9c78b2bSJens Axboe return bvl; 203a8affc03SChristoph Hellwig *nr_vecs = BIO_MAX_VECS; 204f9c78b2bSJens Axboe } 205f9c78b2bSJens Axboe 206f007a3d6SChristoph Hellwig return mempool_alloc(pool, gfp_mask); 207f9c78b2bSJens Axboe } 208f9c78b2bSJens Axboe 2099ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 210f9c78b2bSJens Axboe { 211db9819c7SChristoph Hellwig #ifdef CONFIG_BLK_CGROUP 212db9819c7SChristoph Hellwig if (bio->bi_blkg) { 213db9819c7SChristoph Hellwig blkg_put(bio->bi_blkg); 214db9819c7SChristoph Hellwig bio->bi_blkg = NULL; 215db9819c7SChristoph Hellwig } 216db9819c7SChristoph Hellwig #endif 217ece841abSJustin Tee if (bio_integrity(bio)) 218ece841abSJustin Tee bio_integrity_free(bio); 219a892c8d5SSatya Tangirala 220a892c8d5SSatya Tangirala bio_crypt_free_ctx(bio); 221f9c78b2bSJens Axboe } 2229ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 223f9c78b2bSJens Axboe 224f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 225f9c78b2bSJens Axboe { 226f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 227066ff571SChristoph Hellwig void *p = bio; 228066ff571SChristoph Hellwig 229066ff571SChristoph Hellwig WARN_ON_ONCE(!bs); 230f9c78b2bSJens Axboe 2319ae3b3f5SJens Axboe bio_uninit(bio); 2327a800a20SChristoph Hellwig bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs); 233066ff571SChristoph Hellwig mempool_free(p - bs->front_pad, &bs->bio_pool); 234f9c78b2bSJens Axboe } 235f9c78b2bSJens Axboe 2369ae3b3f5SJens Axboe /* 2379ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2389ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2399ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2409ae3b3f5SJens Axboe */ 24149add496SChristoph Hellwig void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table, 24216458cf3SBart Van Assche unsigned short max_vecs, blk_opf_t opf) 243f9c78b2bSJens Axboe { 244da521626SJens Axboe bio->bi_next = NULL; 24549add496SChristoph Hellwig bio->bi_bdev = bdev; 24649add496SChristoph Hellwig bio->bi_opf = opf; 247da521626SJens Axboe bio->bi_flags = 0; 248da521626SJens Axboe bio->bi_ioprio = 0; 249da521626SJens Axboe bio->bi_status = 0; 250da521626SJens Axboe bio->bi_iter.bi_sector = 0; 251da521626SJens Axboe bio->bi_iter.bi_size = 0; 252da521626SJens Axboe bio->bi_iter.bi_idx = 0; 253da521626SJens Axboe bio->bi_iter.bi_bvec_done = 0; 254da521626SJens Axboe bio->bi_end_io = NULL; 255da521626SJens Axboe bio->bi_private = NULL; 256da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP 257da521626SJens Axboe bio->bi_blkg = NULL; 258da521626SJens Axboe bio->bi_issue.value = 0; 25949add496SChristoph Hellwig if (bdev) 26049add496SChristoph Hellwig bio_associate_blkg(bio); 261da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP_IOCOST 262da521626SJens Axboe bio->bi_iocost_cost = 0; 263da521626SJens Axboe #endif 264da521626SJens Axboe #endif 265da521626SJens Axboe #ifdef CONFIG_BLK_INLINE_ENCRYPTION 266da521626SJens Axboe bio->bi_crypt_context = NULL; 267da521626SJens Axboe #endif 268da521626SJens Axboe #ifdef CONFIG_BLK_DEV_INTEGRITY 269da521626SJens Axboe bio->bi_integrity = NULL; 270da521626SJens Axboe #endif 271da521626SJens Axboe bio->bi_vcnt = 0; 272da521626SJens Axboe 273c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 274dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2753e08773cSChristoph Hellwig bio->bi_cookie = BLK_QC_T_NONE; 2763a83f467SMing Lei 2773a83f467SMing Lei bio->bi_max_vecs = max_vecs; 278da521626SJens Axboe bio->bi_io_vec = table; 279da521626SJens Axboe bio->bi_pool = NULL; 280f9c78b2bSJens Axboe } 281f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 282f9c78b2bSJens Axboe 283f9c78b2bSJens Axboe /** 284f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 285f9c78b2bSJens Axboe * @bio: bio to reset 286a7c50c94SChristoph Hellwig * @bdev: block device to use the bio for 287a7c50c94SChristoph Hellwig * @opf: operation and flags for bio 288f9c78b2bSJens Axboe * 289f9c78b2bSJens Axboe * Description: 290f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 291f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 292f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 293f9c78b2bSJens Axboe * comment in struct bio. 294f9c78b2bSJens Axboe */ 29516458cf3SBart Van Assche void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf) 296f9c78b2bSJens Axboe { 2979ae3b3f5SJens Axboe bio_uninit(bio); 298f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 299c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 300a7c50c94SChristoph Hellwig bio->bi_bdev = bdev; 30178e34374SChristoph Hellwig if (bio->bi_bdev) 30278e34374SChristoph Hellwig bio_associate_blkg(bio); 303a7c50c94SChristoph Hellwig bio->bi_opf = opf; 304f9c78b2bSJens Axboe } 305f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 306f9c78b2bSJens Axboe 30738f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 308f9c78b2bSJens Axboe { 3094246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3104246a0b6SChristoph Hellwig 3113edf5346SYufen Yu if (bio->bi_status && !parent->bi_status) 3124e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 313f9c78b2bSJens Axboe bio_put(bio); 31438f8baaeSChristoph Hellwig return parent; 31538f8baaeSChristoph Hellwig } 31638f8baaeSChristoph Hellwig 31738f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 31838f8baaeSChristoph Hellwig { 31938f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 320f9c78b2bSJens Axboe } 321f9c78b2bSJens Axboe 322f9c78b2bSJens Axboe /** 323f9c78b2bSJens Axboe * bio_chain - chain bio completions 324f9c78b2bSJens Axboe * @bio: the target bio 3255b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 326f9c78b2bSJens Axboe * 327f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 328f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 329f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 330f9c78b2bSJens Axboe * 331f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 332f9c78b2bSJens Axboe */ 333f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 334f9c78b2bSJens Axboe { 335f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 336f9c78b2bSJens Axboe 337f9c78b2bSJens Axboe bio->bi_private = parent; 338f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 339c4cf5261SJens Axboe bio_inc_remaining(parent); 340f9c78b2bSJens Axboe } 341f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 342f9c78b2bSJens Axboe 3430a3140eaSChaitanya Kulkarni struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev, 34416458cf3SBart Van Assche unsigned int nr_pages, blk_opf_t opf, gfp_t gfp) 3453b005bf6SChristoph Hellwig { 34607888c66SChristoph Hellwig struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp); 3470a3140eaSChaitanya Kulkarni 3483b005bf6SChristoph Hellwig if (bio) { 3493b005bf6SChristoph Hellwig bio_chain(bio, new); 3503b005bf6SChristoph Hellwig submit_bio(bio); 3513b005bf6SChristoph Hellwig } 3523b005bf6SChristoph Hellwig 3533b005bf6SChristoph Hellwig return new; 3543b005bf6SChristoph Hellwig } 3553b005bf6SChristoph Hellwig EXPORT_SYMBOL_GPL(blk_next_bio); 3563b005bf6SChristoph Hellwig 357f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 358f9c78b2bSJens Axboe { 359f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 360f9c78b2bSJens Axboe struct bio *bio; 361f9c78b2bSJens Axboe 362f9c78b2bSJens Axboe while (1) { 363f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 364f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 365f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 366f9c78b2bSJens Axboe 367f9c78b2bSJens Axboe if (!bio) 368f9c78b2bSJens Axboe break; 369f9c78b2bSJens Axboe 370ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 371f9c78b2bSJens Axboe } 372f9c78b2bSJens Axboe } 373f9c78b2bSJens Axboe 374f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 375f9c78b2bSJens Axboe { 376f9c78b2bSJens Axboe struct bio_list punt, nopunt; 377f9c78b2bSJens Axboe struct bio *bio; 378f9c78b2bSJens Axboe 37947e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 38047e0fb46SNeilBrown return; 381f9c78b2bSJens Axboe /* 382f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 383f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 384f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 385f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 386f9c78b2bSJens Axboe * our own rescuer would be bad. 387f9c78b2bSJens Axboe * 388f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 389f9c78b2bSJens Axboe * remove from the middle of the list: 390f9c78b2bSJens Axboe */ 391f9c78b2bSJens Axboe 392f9c78b2bSJens Axboe bio_list_init(&punt); 393f9c78b2bSJens Axboe bio_list_init(&nopunt); 394f9c78b2bSJens Axboe 395f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 396f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 397f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 398f9c78b2bSJens Axboe 399f5fe1b51SNeilBrown bio_list_init(&nopunt); 400f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 401f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 402f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 403f9c78b2bSJens Axboe 404f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 405f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 406f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 407f9c78b2bSJens Axboe 408f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 409f9c78b2bSJens Axboe } 410f9c78b2bSJens Axboe 4110df71650SMike Snitzer static struct bio *bio_alloc_percpu_cache(struct block_device *bdev, 41216458cf3SBart Van Assche unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp, 4130df71650SMike Snitzer struct bio_set *bs) 4140df71650SMike Snitzer { 4150df71650SMike Snitzer struct bio_alloc_cache *cache; 4160df71650SMike Snitzer struct bio *bio; 4170df71650SMike Snitzer 4180df71650SMike Snitzer cache = per_cpu_ptr(bs->cache, get_cpu()); 4190df71650SMike Snitzer if (!cache->free_list) { 4200df71650SMike Snitzer put_cpu(); 4210df71650SMike Snitzer return NULL; 4220df71650SMike Snitzer } 4230df71650SMike Snitzer bio = cache->free_list; 4240df71650SMike Snitzer cache->free_list = bio->bi_next; 4250df71650SMike Snitzer cache->nr--; 4260df71650SMike Snitzer put_cpu(); 4270df71650SMike Snitzer 4280df71650SMike Snitzer bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf); 4290df71650SMike Snitzer bio->bi_pool = bs; 4300df71650SMike Snitzer return bio; 4310df71650SMike Snitzer } 4320df71650SMike Snitzer 433f9c78b2bSJens Axboe /** 434f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 435609be106SChristoph Hellwig * @bdev: block device to allocate the bio for (can be %NULL) 436609be106SChristoph Hellwig * @nr_vecs: number of bvecs to pre-allocate 437609be106SChristoph Hellwig * @opf: operation and flags for bio 438519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 439f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 440f9c78b2bSJens Axboe * 4413175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 442f9c78b2bSJens Axboe * 4433175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4443175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4453175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4463175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4473175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4483175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 449f9c78b2bSJens Axboe * 4503175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4513175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 452ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 453f9c78b2bSJens Axboe * stack overflows. 454f9c78b2bSJens Axboe * 4553175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4563175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 457f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 458f9c78b2bSJens Axboe * thread. 459f9c78b2bSJens Axboe * 460f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 461f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 462ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 463f9c78b2bSJens Axboe * for per bio allocations. 464f9c78b2bSJens Axboe * 4650df71650SMike Snitzer * If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done from process 4660df71650SMike Snitzer * context, not hard/soft IRQ. 4670df71650SMike Snitzer * 4683175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 469f9c78b2bSJens Axboe */ 470609be106SChristoph Hellwig struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs, 47116458cf3SBart Van Assche blk_opf_t opf, gfp_t gfp_mask, 4727a88fa19SDan Carpenter struct bio_set *bs) 473f9c78b2bSJens Axboe { 474f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 475f9c78b2bSJens Axboe struct bio *bio; 476f9c78b2bSJens Axboe void *p; 477f9c78b2bSJens Axboe 478609be106SChristoph Hellwig /* should not use nobvec bioset for nr_vecs > 0 */ 479609be106SChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0)) 480f9c78b2bSJens Axboe return NULL; 481f9c78b2bSJens Axboe 4820df71650SMike Snitzer if (opf & REQ_ALLOC_CACHE) { 4830df71650SMike Snitzer if (bs->cache && nr_vecs <= BIO_INLINE_VECS) { 4840df71650SMike Snitzer bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf, 4850df71650SMike Snitzer gfp_mask, bs); 4860df71650SMike Snitzer if (bio) 4870df71650SMike Snitzer return bio; 4880df71650SMike Snitzer /* 4890df71650SMike Snitzer * No cached bio available, bio returned below marked with 4900df71650SMike Snitzer * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache. 4910df71650SMike Snitzer */ 4920df71650SMike Snitzer } else { 4930df71650SMike Snitzer opf &= ~REQ_ALLOC_CACHE; 4940df71650SMike Snitzer } 4950df71650SMike Snitzer } 4960df71650SMike Snitzer 497f9c78b2bSJens Axboe /* 4983175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 4993175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 5003175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 501f9c78b2bSJens Axboe * 5023175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 5033175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 5043175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 5053175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 5063175199aSChristoph Hellwig * mempool's reserve. 507f9c78b2bSJens Axboe * 508f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 5093175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 5103175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 5113175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 5123175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 5133175199aSChristoph Hellwig * gfp_flags. 514f9c78b2bSJens Axboe */ 515f5fe1b51SNeilBrown if (current->bio_list && 516f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 51747e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 51847e0fb46SNeilBrown bs->rescue_workqueue) 519d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 520f9c78b2bSJens Axboe 5218aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 522f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 523f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 524f9c78b2bSJens Axboe gfp_mask = saved_gfp; 5258aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 526f9c78b2bSJens Axboe } 527f9c78b2bSJens Axboe if (unlikely(!p)) 528f9c78b2bSJens Axboe return NULL; 529*759aa12fSPavel Begunkov if (!mempool_is_saturated(&bs->bio_pool)) 530*759aa12fSPavel Begunkov opf &= ~REQ_ALLOC_CACHE; 531f9c78b2bSJens Axboe 5323175199aSChristoph Hellwig bio = p + bs->front_pad; 533609be106SChristoph Hellwig if (nr_vecs > BIO_INLINE_VECS) { 5343175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 535f9c78b2bSJens Axboe 536609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 537f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 538f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 539f9c78b2bSJens Axboe gfp_mask = saved_gfp; 540609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 541f9c78b2bSJens Axboe } 542f9c78b2bSJens Axboe if (unlikely(!bvl)) 543f9c78b2bSJens Axboe goto err_free; 544f9c78b2bSJens Axboe 54549add496SChristoph Hellwig bio_init(bio, bdev, bvl, nr_vecs, opf); 546609be106SChristoph Hellwig } else if (nr_vecs) { 54749add496SChristoph Hellwig bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf); 5483175199aSChristoph Hellwig } else { 54949add496SChristoph Hellwig bio_init(bio, bdev, NULL, 0, opf); 550f9c78b2bSJens Axboe } 551f9c78b2bSJens Axboe 552f9c78b2bSJens Axboe bio->bi_pool = bs; 553f9c78b2bSJens Axboe return bio; 554f9c78b2bSJens Axboe 555f9c78b2bSJens Axboe err_free: 5568aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 557f9c78b2bSJens Axboe return NULL; 558f9c78b2bSJens Axboe } 559f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 560f9c78b2bSJens Axboe 5613175199aSChristoph Hellwig /** 562066ff571SChristoph Hellwig * bio_kmalloc - kmalloc a bio 563066ff571SChristoph Hellwig * @nr_vecs: number of bio_vecs to allocate 5643175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 5653175199aSChristoph Hellwig * 566066ff571SChristoph Hellwig * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized 567066ff571SChristoph Hellwig * using bio_init() before use. To free a bio returned from this function use 568066ff571SChristoph Hellwig * kfree() after calling bio_uninit(). A bio returned from this function can 569066ff571SChristoph Hellwig * be reused by calling bio_uninit() before calling bio_init() again. 570066ff571SChristoph Hellwig * 571066ff571SChristoph Hellwig * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this 572340e1347SDeming Wang * function are not backed by a mempool can fail. Do not use this function 573066ff571SChristoph Hellwig * for allocations in the file system I/O path. 5743175199aSChristoph Hellwig * 5753175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 5763175199aSChristoph Hellwig */ 577066ff571SChristoph Hellwig struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask) 5783175199aSChristoph Hellwig { 5793175199aSChristoph Hellwig struct bio *bio; 5803175199aSChristoph Hellwig 581066ff571SChristoph Hellwig if (nr_vecs > UIO_MAXIOV) 5823175199aSChristoph Hellwig return NULL; 583066ff571SChristoph Hellwig return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask); 5843175199aSChristoph Hellwig } 5853175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 5863175199aSChristoph Hellwig 5876f822e1bSChristoph Hellwig void zero_fill_bio(struct bio *bio) 588f9c78b2bSJens Axboe { 589f9c78b2bSJens Axboe struct bio_vec bv; 590f9c78b2bSJens Axboe struct bvec_iter iter; 591f9c78b2bSJens Axboe 592ab6c340eSChristoph Hellwig bio_for_each_segment(bv, bio, iter) 593ab6c340eSChristoph Hellwig memzero_bvec(&bv); 594f9c78b2bSJens Axboe } 5956f822e1bSChristoph Hellwig EXPORT_SYMBOL(zero_fill_bio); 596f9c78b2bSJens Axboe 59783c9c547SMing Lei /** 59883c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 59983c9c547SMing Lei * @bio: the bio to be truncated 60083c9c547SMing Lei * @new_size: new size for truncating the bio 60183c9c547SMing Lei * 60283c9c547SMing Lei * Description: 60383c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 60483c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 60583c9c547SMing Lei * be used for handling corner cases, such as bio eod. 60683c9c547SMing Lei */ 6074f7ab09aSChristoph Hellwig static void bio_truncate(struct bio *bio, unsigned new_size) 60885a8ce62SMing Lei { 60985a8ce62SMing Lei struct bio_vec bv; 61085a8ce62SMing Lei struct bvec_iter iter; 61185a8ce62SMing Lei unsigned int done = 0; 61285a8ce62SMing Lei bool truncated = false; 61385a8ce62SMing Lei 61485a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 61585a8ce62SMing Lei return; 61685a8ce62SMing Lei 61783c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 61885a8ce62SMing Lei goto exit; 61985a8ce62SMing Lei 62085a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 62185a8ce62SMing Lei if (done + bv.bv_len > new_size) { 62285a8ce62SMing Lei unsigned offset; 62385a8ce62SMing Lei 62485a8ce62SMing Lei if (!truncated) 62585a8ce62SMing Lei offset = new_size - done; 62685a8ce62SMing Lei else 62785a8ce62SMing Lei offset = 0; 6283ee859e3SOGAWA Hirofumi zero_user(bv.bv_page, bv.bv_offset + offset, 6293ee859e3SOGAWA Hirofumi bv.bv_len - offset); 63085a8ce62SMing Lei truncated = true; 63185a8ce62SMing Lei } 63285a8ce62SMing Lei done += bv.bv_len; 63385a8ce62SMing Lei } 63485a8ce62SMing Lei 63585a8ce62SMing Lei exit: 63685a8ce62SMing Lei /* 63785a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 63885a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 63985a8ce62SMing Lei * in its .end_bio() callback. 64085a8ce62SMing Lei * 64185a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 64285a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 64385a8ce62SMing Lei */ 64485a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 64585a8ce62SMing Lei } 64685a8ce62SMing Lei 647f9c78b2bSJens Axboe /** 64829125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 64929125ed6SChristoph Hellwig * @bio: bio to truncate 65029125ed6SChristoph Hellwig * 65129125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 65229125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 65329125ed6SChristoph Hellwig * 65429125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 65529125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 65629125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 65729125ed6SChristoph Hellwig * sector" case. 65829125ed6SChristoph Hellwig */ 65929125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 66029125ed6SChristoph Hellwig { 661309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 66229125ed6SChristoph Hellwig 66329125ed6SChristoph Hellwig if (!maxsector) 66429125ed6SChristoph Hellwig return; 66529125ed6SChristoph Hellwig 66629125ed6SChristoph Hellwig /* 66729125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 66829125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 66929125ed6SChristoph Hellwig * an EIO. 67029125ed6SChristoph Hellwig */ 67129125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 67229125ed6SChristoph Hellwig return; 67329125ed6SChristoph Hellwig 67429125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 67529125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 67629125ed6SChristoph Hellwig return; 67729125ed6SChristoph Hellwig 67829125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 67929125ed6SChristoph Hellwig } 68029125ed6SChristoph Hellwig 681be4d234dSJens Axboe #define ALLOC_CACHE_MAX 512 682be4d234dSJens Axboe #define ALLOC_CACHE_SLACK 64 683be4d234dSJens Axboe 684be4d234dSJens Axboe static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 685be4d234dSJens Axboe unsigned int nr) 686be4d234dSJens Axboe { 687be4d234dSJens Axboe unsigned int i = 0; 688be4d234dSJens Axboe struct bio *bio; 689be4d234dSJens Axboe 690fcade2ceSJens Axboe while ((bio = cache->free_list) != NULL) { 691fcade2ceSJens Axboe cache->free_list = bio->bi_next; 692be4d234dSJens Axboe cache->nr--; 693be4d234dSJens Axboe bio_free(bio); 694be4d234dSJens Axboe if (++i == nr) 695be4d234dSJens Axboe break; 696be4d234dSJens Axboe } 697be4d234dSJens Axboe } 698be4d234dSJens Axboe 699be4d234dSJens Axboe static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 700be4d234dSJens Axboe { 701be4d234dSJens Axboe struct bio_set *bs; 702be4d234dSJens Axboe 703be4d234dSJens Axboe bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 704be4d234dSJens Axboe if (bs->cache) { 705be4d234dSJens Axboe struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 706be4d234dSJens Axboe 707be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 708be4d234dSJens Axboe } 709be4d234dSJens Axboe return 0; 710be4d234dSJens Axboe } 711be4d234dSJens Axboe 712be4d234dSJens Axboe static void bio_alloc_cache_destroy(struct bio_set *bs) 713be4d234dSJens Axboe { 714be4d234dSJens Axboe int cpu; 715be4d234dSJens Axboe 716be4d234dSJens Axboe if (!bs->cache) 717be4d234dSJens Axboe return; 718be4d234dSJens Axboe 719be4d234dSJens Axboe cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 720be4d234dSJens Axboe for_each_possible_cpu(cpu) { 721be4d234dSJens Axboe struct bio_alloc_cache *cache; 722be4d234dSJens Axboe 723be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, cpu); 724be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 725be4d234dSJens Axboe } 726be4d234dSJens Axboe free_percpu(bs->cache); 727605f7415SJens Axboe bs->cache = NULL; 728be4d234dSJens Axboe } 729be4d234dSJens Axboe 73029125ed6SChristoph Hellwig /** 731f9c78b2bSJens Axboe * bio_put - release a reference to a bio 732f9c78b2bSJens Axboe * @bio: bio to release reference to 733f9c78b2bSJens Axboe * 734f9c78b2bSJens Axboe * Description: 735f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 7369b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 737f9c78b2bSJens Axboe **/ 738f9c78b2bSJens Axboe void bio_put(struct bio *bio) 739f9c78b2bSJens Axboe { 740be4d234dSJens Axboe if (unlikely(bio_flagged(bio, BIO_REFFED))) { 7419e8c0d0dSChristoph Hellwig BUG_ON(!atomic_read(&bio->__bi_cnt)); 742be4d234dSJens Axboe if (!atomic_dec_and_test(&bio->__bi_cnt)) 743be4d234dSJens Axboe return; 744be4d234dSJens Axboe } 745f9c78b2bSJens Axboe 746d4347d50SPavel Begunkov if ((bio->bi_opf & REQ_ALLOC_CACHE) && !WARN_ON_ONCE(in_interrupt())) { 747be4d234dSJens Axboe struct bio_alloc_cache *cache; 748be4d234dSJens Axboe 749be4d234dSJens Axboe bio_uninit(bio); 750be4d234dSJens Axboe cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 751fcade2ceSJens Axboe bio->bi_next = cache->free_list; 752fcade2ceSJens Axboe cache->free_list = bio; 753be4d234dSJens Axboe if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 754be4d234dSJens Axboe bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 755be4d234dSJens Axboe put_cpu(); 756be4d234dSJens Axboe } else { 757f9c78b2bSJens Axboe bio_free(bio); 758f9c78b2bSJens Axboe } 759dac56212SJens Axboe } 760f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 761f9c78b2bSJens Axboe 762a0e8de79SChristoph Hellwig static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp) 763f9c78b2bSJens Axboe { 764b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 765ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 766f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 76720bd723eSPaolo Valente 7687ecc56c6SChristoph Hellwig if (bio->bi_bdev) { 7697ecc56c6SChristoph Hellwig if (bio->bi_bdev == bio_src->bi_bdev && 7707ecc56c6SChristoph Hellwig bio_flagged(bio_src, BIO_REMAPPED)) 7717ecc56c6SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 772db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 7737ecc56c6SChristoph Hellwig } 77456b4b5abSChristoph Hellwig 77556b4b5abSChristoph Hellwig if (bio_crypt_clone(bio, bio_src, gfp) < 0) 77656b4b5abSChristoph Hellwig return -ENOMEM; 77756b4b5abSChristoph Hellwig if (bio_integrity(bio_src) && 77856b4b5abSChristoph Hellwig bio_integrity_clone(bio, bio_src, gfp) < 0) 77956b4b5abSChristoph Hellwig return -ENOMEM; 78056b4b5abSChristoph Hellwig return 0; 781f9c78b2bSJens Axboe } 782f9c78b2bSJens Axboe 783f9c78b2bSJens Axboe /** 784abfc426dSChristoph Hellwig * bio_alloc_clone - clone a bio that shares the original bio's biovec 785abfc426dSChristoph Hellwig * @bdev: block_device to clone onto 786a0e8de79SChristoph Hellwig * @bio_src: bio to clone from 787a0e8de79SChristoph Hellwig * @gfp: allocation priority 788f9c78b2bSJens Axboe * @bs: bio_set to allocate from 789f9c78b2bSJens Axboe * 790a0e8de79SChristoph Hellwig * Allocate a new bio that is a clone of @bio_src. The caller owns the returned 791a0e8de79SChristoph Hellwig * bio, but not the actual data it points to. 792a0e8de79SChristoph Hellwig * 793a0e8de79SChristoph Hellwig * The caller must ensure that the return bio is not freed before @bio_src. 794f9c78b2bSJens Axboe */ 795abfc426dSChristoph Hellwig struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src, 796abfc426dSChristoph Hellwig gfp_t gfp, struct bio_set *bs) 797f9c78b2bSJens Axboe { 798a0e8de79SChristoph Hellwig struct bio *bio; 799f9c78b2bSJens Axboe 800abfc426dSChristoph Hellwig bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs); 801a0e8de79SChristoph Hellwig if (!bio) 802f9c78b2bSJens Axboe return NULL; 803f9c78b2bSJens Axboe 804a0e8de79SChristoph Hellwig if (__bio_clone(bio, bio_src, gfp) < 0) { 805a0e8de79SChristoph Hellwig bio_put(bio); 80607560151SEric Biggers return NULL; 807f9c78b2bSJens Axboe } 808a0e8de79SChristoph Hellwig bio->bi_io_vec = bio_src->bi_io_vec; 80956b4b5abSChristoph Hellwig 810a0e8de79SChristoph Hellwig return bio; 81156b4b5abSChristoph Hellwig } 812abfc426dSChristoph Hellwig EXPORT_SYMBOL(bio_alloc_clone); 813f9c78b2bSJens Axboe 814a0e8de79SChristoph Hellwig /** 815abfc426dSChristoph Hellwig * bio_init_clone - clone a bio that shares the original bio's biovec 816abfc426dSChristoph Hellwig * @bdev: block_device to clone onto 817a0e8de79SChristoph Hellwig * @bio: bio to clone into 818a0e8de79SChristoph Hellwig * @bio_src: bio to clone from 819a0e8de79SChristoph Hellwig * @gfp: allocation priority 820a0e8de79SChristoph Hellwig * 821a0e8de79SChristoph Hellwig * Initialize a new bio in caller provided memory that is a clone of @bio_src. 822a0e8de79SChristoph Hellwig * The caller owns the returned bio, but not the actual data it points to. 823a0e8de79SChristoph Hellwig * 824a0e8de79SChristoph Hellwig * The caller must ensure that @bio_src is not freed before @bio. 825a0e8de79SChristoph Hellwig */ 826abfc426dSChristoph Hellwig int bio_init_clone(struct block_device *bdev, struct bio *bio, 827abfc426dSChristoph Hellwig struct bio *bio_src, gfp_t gfp) 828a0e8de79SChristoph Hellwig { 829a0e8de79SChristoph Hellwig int ret; 830a0e8de79SChristoph Hellwig 831abfc426dSChristoph Hellwig bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf); 832a0e8de79SChristoph Hellwig ret = __bio_clone(bio, bio_src, gfp); 833a0e8de79SChristoph Hellwig if (ret) 834a0e8de79SChristoph Hellwig bio_uninit(bio); 835a0e8de79SChristoph Hellwig return ret; 836a0e8de79SChristoph Hellwig } 837abfc426dSChristoph Hellwig EXPORT_SYMBOL(bio_init_clone); 838a0e8de79SChristoph Hellwig 8399a6083beSChristoph Hellwig /** 8409a6083beSChristoph Hellwig * bio_full - check if the bio is full 8419a6083beSChristoph Hellwig * @bio: bio to check 8429a6083beSChristoph Hellwig * @len: length of one segment to be added 8439a6083beSChristoph Hellwig * 8449a6083beSChristoph Hellwig * Return true if @bio is full and one segment with @len bytes can't be 8459a6083beSChristoph Hellwig * added to the bio, otherwise return false 8469a6083beSChristoph Hellwig */ 8479a6083beSChristoph Hellwig static inline bool bio_full(struct bio *bio, unsigned len) 8489a6083beSChristoph Hellwig { 8499a6083beSChristoph Hellwig if (bio->bi_vcnt >= bio->bi_max_vecs) 8509a6083beSChristoph Hellwig return true; 8519a6083beSChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) 8529a6083beSChristoph Hellwig return true; 8539a6083beSChristoph Hellwig return false; 8549a6083beSChristoph Hellwig } 8559a6083beSChristoph Hellwig 8565919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 8575919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 858ff896738SChristoph Hellwig bool *same_page) 8595919482eSMing Lei { 860d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 861d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 8625919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 8635919482eSMing Lei 8645919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 8655919482eSMing Lei return false; 8665919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 8675919482eSMing Lei return false; 86849580e69SLogan Gunthorpe if (!zone_device_pages_have_same_pgmap(bv->bv_page, page)) 86949580e69SLogan Gunthorpe return false; 87052d52d1cSChristoph Hellwig 871ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 872d8166519SMatthew Wilcox (Oracle) if (*same_page) 8735919482eSMing Lei return true; 87411b331f8SAlexander Potapenko else if (IS_ENABLED(CONFIG_KMSAN)) 87511b331f8SAlexander Potapenko return false; 876d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 8775919482eSMing Lei } 8785919482eSMing Lei 8799774b391SChristoph Hellwig /** 8809774b391SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 8819774b391SChristoph Hellwig * @bio: destination bio 8829774b391SChristoph Hellwig * @page: start page to add 8839774b391SChristoph Hellwig * @len: length of the data to add 8849774b391SChristoph Hellwig * @off: offset of the data relative to @page 8859774b391SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 8869774b391SChristoph Hellwig * 8879774b391SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 8889774b391SChristoph Hellwig * useful optimisation for file systems with a block size smaller than the 8899774b391SChristoph Hellwig * page size. 8909774b391SChristoph Hellwig * 8919774b391SChristoph Hellwig * Warn if (@len, @off) crosses pages in case that @same_page is true. 8929774b391SChristoph Hellwig * 8939774b391SChristoph Hellwig * Return %true on success or %false on failure. 8949774b391SChristoph Hellwig */ 8959774b391SChristoph Hellwig static bool __bio_try_merge_page(struct bio *bio, struct page *page, 8969774b391SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 8979774b391SChristoph Hellwig { 8989774b391SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8999774b391SChristoph Hellwig return false; 9009774b391SChristoph Hellwig 9019774b391SChristoph Hellwig if (bio->bi_vcnt > 0) { 9029774b391SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 9039774b391SChristoph Hellwig 9049774b391SChristoph Hellwig if (page_is_mergeable(bv, page, len, off, same_page)) { 9059774b391SChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) { 9069774b391SChristoph Hellwig *same_page = false; 9079774b391SChristoph Hellwig return false; 9089774b391SChristoph Hellwig } 9099774b391SChristoph Hellwig bv->bv_len += len; 9109774b391SChristoph Hellwig bio->bi_iter.bi_size += len; 9119774b391SChristoph Hellwig return true; 9129774b391SChristoph Hellwig } 9139774b391SChristoph Hellwig } 9149774b391SChristoph Hellwig return false; 9159774b391SChristoph Hellwig } 9169774b391SChristoph Hellwig 917e4581105SChristoph Hellwig /* 918e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 919e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 920e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 921e4581105SChristoph Hellwig */ 922e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 923e4581105SChristoph Hellwig struct page *page, unsigned len, 924e4581105SChristoph Hellwig unsigned offset, bool *same_page) 925489fbbcbSMing Lei { 926384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 927489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 928489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 929489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 930489fbbcbSMing Lei 931489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 932489fbbcbSMing Lei return false; 933489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 934489fbbcbSMing Lei return false; 935384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 936489fbbcbSMing Lei } 937489fbbcbSMing Lei 938f4595875SShaohua Li /** 939e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 940c66a14d0SKent Overstreet * @q: the target queue 941c66a14d0SKent Overstreet * @bio: destination bio 942c66a14d0SKent Overstreet * @page: page to add 943c66a14d0SKent Overstreet * @len: vec entry length 944c66a14d0SKent Overstreet * @offset: vec entry offset 945e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 946e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 947f9c78b2bSJens Axboe * 948e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 949e4581105SChristoph Hellwig * and gap limitations. 950f9c78b2bSJens Axboe */ 951e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 95219047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 953e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 954f9c78b2bSJens Axboe { 955f9c78b2bSJens Axboe struct bio_vec *bvec; 956f9c78b2bSJens Axboe 957e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 958f9c78b2bSJens Axboe return 0; 959f9c78b2bSJens Axboe 960e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 961f9c78b2bSJens Axboe return 0; 962f9c78b2bSJens Axboe 963f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 964e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 965384209cdSChristoph Hellwig return len; 966320ea869SChristoph Hellwig 967320ea869SChristoph Hellwig /* 968320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 969320ea869SChristoph Hellwig * would create a gap, disallow it. 970320ea869SChristoph Hellwig */ 971384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 972c55ddd90SChristoph Hellwig if (bvec_gap_to_prev(&q->limits, bvec, offset)) 973320ea869SChristoph Hellwig return 0; 974f9c78b2bSJens Axboe } 975f9c78b2bSJens Axboe 97679d08f89SMing Lei if (bio_full(bio, len)) 977f9c78b2bSJens Axboe return 0; 978f9c78b2bSJens Axboe 97914ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 980489fbbcbSMing Lei return 0; 981489fbbcbSMing Lei 982f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 983f9c78b2bSJens Axboe bvec->bv_page = page; 984f9c78b2bSJens Axboe bvec->bv_len = len; 985f9c78b2bSJens Axboe bvec->bv_offset = offset; 986fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 987dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 988f9c78b2bSJens Axboe return len; 989f9c78b2bSJens Axboe } 99019047087SMing Lei 991e4581105SChristoph Hellwig /** 992e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 993e4581105SChristoph Hellwig * @q: the target queue 994e4581105SChristoph Hellwig * @bio: destination bio 995e4581105SChristoph Hellwig * @page: page to add 996e4581105SChristoph Hellwig * @len: vec entry length 997e4581105SChristoph Hellwig * @offset: vec entry offset 998e4581105SChristoph Hellwig * 999e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 1000e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 1001e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 1002e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 1003e4581105SChristoph Hellwig * 1004e4581105SChristoph Hellwig * This should only be used by passthrough bios. 1005e4581105SChristoph Hellwig */ 100619047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 100719047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 100819047087SMing Lei { 1009d1916c86SChristoph Hellwig bool same_page = false; 1010e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 1011e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 101219047087SMing Lei } 1013f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 1014f9c78b2bSJens Axboe 1015f9c78b2bSJens Axboe /** 1016ae29333fSJohannes Thumshirn * bio_add_zone_append_page - attempt to add page to zone-append bio 1017ae29333fSJohannes Thumshirn * @bio: destination bio 1018ae29333fSJohannes Thumshirn * @page: page to add 1019ae29333fSJohannes Thumshirn * @len: vec entry length 1020ae29333fSJohannes Thumshirn * @offset: vec entry offset 1021ae29333fSJohannes Thumshirn * 1022ae29333fSJohannes Thumshirn * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 1023ae29333fSJohannes Thumshirn * for a zone-append request. This can fail for a number of reasons, such as the 1024ae29333fSJohannes Thumshirn * bio being full or the target block device is not a zoned block device or 1025ae29333fSJohannes Thumshirn * other limitations of the target block device. The target block device must 1026ae29333fSJohannes Thumshirn * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 1027ae29333fSJohannes Thumshirn * to an empty bio. 1028ae29333fSJohannes Thumshirn * 1029ae29333fSJohannes Thumshirn * Returns: number of bytes added to the bio, or 0 in case of a failure. 1030ae29333fSJohannes Thumshirn */ 1031ae29333fSJohannes Thumshirn int bio_add_zone_append_page(struct bio *bio, struct page *page, 1032ae29333fSJohannes Thumshirn unsigned int len, unsigned int offset) 1033ae29333fSJohannes Thumshirn { 10343caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1035ae29333fSJohannes Thumshirn bool same_page = false; 1036ae29333fSJohannes Thumshirn 1037ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 1038ae29333fSJohannes Thumshirn return 0; 1039ae29333fSJohannes Thumshirn 1040edd1dbc8SChristoph Hellwig if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev))) 1041ae29333fSJohannes Thumshirn return 0; 1042ae29333fSJohannes Thumshirn 1043ae29333fSJohannes Thumshirn return bio_add_hw_page(q, bio, page, len, offset, 1044ae29333fSJohannes Thumshirn queue_max_zone_append_sectors(q), &same_page); 1045ae29333fSJohannes Thumshirn } 1046ae29333fSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 1047ae29333fSJohannes Thumshirn 1048ae29333fSJohannes Thumshirn /** 1049551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 10500aa69fd3SChristoph Hellwig * @bio: destination bio 1051551879a4SMing Lei * @page: start page to add 1052551879a4SMing Lei * @len: length of the data to add, may cross pages 1053551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 10540aa69fd3SChristoph Hellwig * 10550aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 10560aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 10570aa69fd3SChristoph Hellwig */ 10580aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 10590aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 10600aa69fd3SChristoph Hellwig { 10610aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 10620aa69fd3SChristoph Hellwig 10630aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 106479d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 10650aa69fd3SChristoph Hellwig 10660aa69fd3SChristoph Hellwig bv->bv_page = page; 10670aa69fd3SChristoph Hellwig bv->bv_offset = off; 10680aa69fd3SChristoph Hellwig bv->bv_len = len; 10690aa69fd3SChristoph Hellwig 10700aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 10710aa69fd3SChristoph Hellwig bio->bi_vcnt++; 10720aa69fd3SChristoph Hellwig } 10730aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 10740aa69fd3SChristoph Hellwig 10750aa69fd3SChristoph Hellwig /** 1076551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 1077f9c78b2bSJens Axboe * @bio: destination bio 1078551879a4SMing Lei * @page: start page to add 1079551879a4SMing Lei * @len: vec entry length, may cross pages 1080551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 1081f9c78b2bSJens Axboe * 1082551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 1083c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1084f9c78b2bSJens Axboe */ 1085c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 1086c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 1087f9c78b2bSJens Axboe { 1088ff896738SChristoph Hellwig bool same_page = false; 1089ff896738SChristoph Hellwig 1090ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 109179d08f89SMing Lei if (bio_full(bio, len)) 1092c66a14d0SKent Overstreet return 0; 10930aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 1094c66a14d0SKent Overstreet } 1095c66a14d0SKent Overstreet return len; 1096f9c78b2bSJens Axboe } 1097f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 1098f9c78b2bSJens Axboe 109985f5a74cSMatthew Wilcox (Oracle) /** 110085f5a74cSMatthew Wilcox (Oracle) * bio_add_folio - Attempt to add part of a folio to a bio. 110185f5a74cSMatthew Wilcox (Oracle) * @bio: BIO to add to. 110285f5a74cSMatthew Wilcox (Oracle) * @folio: Folio to add. 110385f5a74cSMatthew Wilcox (Oracle) * @len: How many bytes from the folio to add. 110485f5a74cSMatthew Wilcox (Oracle) * @off: First byte in this folio to add. 110585f5a74cSMatthew Wilcox (Oracle) * 110685f5a74cSMatthew Wilcox (Oracle) * Filesystems that use folios can call this function instead of calling 110785f5a74cSMatthew Wilcox (Oracle) * bio_add_page() for each page in the folio. If @off is bigger than 110885f5a74cSMatthew Wilcox (Oracle) * PAGE_SIZE, this function can create a bio_vec that starts in a page 110985f5a74cSMatthew Wilcox (Oracle) * after the bv_page. BIOs do not support folios that are 4GiB or larger. 111085f5a74cSMatthew Wilcox (Oracle) * 111185f5a74cSMatthew Wilcox (Oracle) * Return: Whether the addition was successful. 111285f5a74cSMatthew Wilcox (Oracle) */ 111385f5a74cSMatthew Wilcox (Oracle) bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, 111485f5a74cSMatthew Wilcox (Oracle) size_t off) 111585f5a74cSMatthew Wilcox (Oracle) { 111685f5a74cSMatthew Wilcox (Oracle) if (len > UINT_MAX || off > UINT_MAX) 1117455a844dSJiapeng Chong return false; 111885f5a74cSMatthew Wilcox (Oracle) return bio_add_page(bio, &folio->page, len, off) > 0; 111985f5a74cSMatthew Wilcox (Oracle) } 112085f5a74cSMatthew Wilcox (Oracle) 1121c809084aSPavel Begunkov void __bio_release_pages(struct bio *bio, bool mark_dirty) 11227321ecbfSChristoph Hellwig { 11237321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 11247321ecbfSChristoph Hellwig struct bio_vec *bvec; 11257321ecbfSChristoph Hellwig 1126d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1127d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 1128d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 11297321ecbfSChristoph Hellwig put_page(bvec->bv_page); 11307321ecbfSChristoph Hellwig } 1131d241a95fSChristoph Hellwig } 1132c809084aSPavel Begunkov EXPORT_SYMBOL_GPL(__bio_release_pages); 11337321ecbfSChristoph Hellwig 11341bb6b810SPavel Begunkov void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 11356d0c48aeSJens Axboe { 1136fa5fa8ecSPavel Begunkov size_t size = iov_iter_count(iter); 1137fa5fa8ecSPavel Begunkov 11387a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_max_vecs); 11396d0c48aeSJens Axboe 1140fa5fa8ecSPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1141fa5fa8ecSPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1142fa5fa8ecSPavel Begunkov size_t max_sectors = queue_max_zone_append_sectors(q); 1143fa5fa8ecSPavel Begunkov 1144fa5fa8ecSPavel Begunkov size = min(size, max_sectors << SECTOR_SHIFT); 1145fa5fa8ecSPavel Begunkov } 1146fa5fa8ecSPavel Begunkov 1147c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 1148c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1149c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 1150fa5fa8ecSPavel Begunkov bio->bi_iter.bi_size = size; 1151ed97ce5eSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 1152977be012SChristoph Hellwig bio_set_flag(bio, BIO_CLONED); 11537de55b7dSJohannes Thumshirn } 11546d0c48aeSJens Axboe 1155c58c0074SKeith Busch static int bio_iov_add_page(struct bio *bio, struct page *page, 1156c58c0074SKeith Busch unsigned int len, unsigned int offset) 1157c58c0074SKeith Busch { 1158c58c0074SKeith Busch bool same_page = false; 1159c58c0074SKeith Busch 1160c58c0074SKeith Busch if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1161c58c0074SKeith Busch __bio_add_page(bio, page, len, offset); 1162c58c0074SKeith Busch return 0; 1163c58c0074SKeith Busch } 1164c58c0074SKeith Busch 1165c58c0074SKeith Busch if (same_page) 1166c58c0074SKeith Busch put_page(page); 1167c58c0074SKeith Busch return 0; 1168c58c0074SKeith Busch } 1169c58c0074SKeith Busch 1170c58c0074SKeith Busch static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page, 1171c58c0074SKeith Busch unsigned int len, unsigned int offset) 1172c58c0074SKeith Busch { 1173c58c0074SKeith Busch struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1174c58c0074SKeith Busch bool same_page = false; 1175c58c0074SKeith Busch 1176c58c0074SKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 1177c58c0074SKeith Busch queue_max_zone_append_sectors(q), &same_page) != len) 1178c58c0074SKeith Busch return -EINVAL; 1179c58c0074SKeith Busch if (same_page) 1180c58c0074SKeith Busch put_page(page); 1181c58c0074SKeith Busch return 0; 1182c58c0074SKeith Busch } 1183c58c0074SKeith Busch 1184576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1185576ed913SChristoph Hellwig 11862cefe4dbSKent Overstreet /** 118717d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 11882cefe4dbSKent Overstreet * @bio: bio to add pages to 11892cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 11902cefe4dbSKent Overstreet * 119117d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 11922cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 119317d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 11943cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 11952cefe4dbSKent Overstreet */ 119617d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 11972cefe4dbSKent Overstreet { 1198576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1199576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 12002cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 12012cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 12025e3e3f2eSLogan Gunthorpe unsigned int gup_flags = 0; 1203576ed913SChristoph Hellwig ssize_t size, left; 1204e97424fdSKeith Busch unsigned len, i = 0; 1205480cb846SAl Viro size_t offset, trim; 1206325347d9SKeith Busch int ret = 0; 1207576ed913SChristoph Hellwig 1208576ed913SChristoph Hellwig /* 1209576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 1210576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 1211576ed913SChristoph Hellwig * without overwriting the temporary page array. 1212576ed913SChristoph Hellwig */ 1213576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1214576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 12152cefe4dbSKent Overstreet 12165e3e3f2eSLogan Gunthorpe if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue)) 12175e3e3f2eSLogan Gunthorpe gup_flags |= FOLL_PCI_P2PDMA; 12185e3e3f2eSLogan Gunthorpe 1219b1a000d3SKeith Busch /* 1220b1a000d3SKeith Busch * Each segment in the iov is required to be a block size multiple. 1221b1a000d3SKeith Busch * However, we may not be able to get the entire segment if it spans 1222b1a000d3SKeith Busch * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the 1223b1a000d3SKeith Busch * result to ensure the bio's total size is correct. The remainder of 1224b1a000d3SKeith Busch * the iov data will be picked up in the next bio iteration. 1225b1a000d3SKeith Busch */ 12265e3e3f2eSLogan Gunthorpe size = iov_iter_get_pages(iter, pages, 12275e3e3f2eSLogan Gunthorpe UINT_MAX - bio->bi_iter.bi_size, 12285e3e3f2eSLogan Gunthorpe nr_pages, &offset, gup_flags); 1229480cb846SAl Viro if (unlikely(size <= 0)) 1230480cb846SAl Viro return size ? size : -EFAULT; 1231e97424fdSKeith Busch 1232480cb846SAl Viro nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE); 1233480cb846SAl Viro 1234480cb846SAl Viro trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1); 1235480cb846SAl Viro iov_iter_revert(iter, trim); 1236480cb846SAl Viro 1237480cb846SAl Viro size -= trim; 1238480cb846SAl Viro if (unlikely(!size)) { 1239480cb846SAl Viro ret = -EFAULT; 1240e97424fdSKeith Busch goto out; 1241e97424fdSKeith Busch } 12422cefe4dbSKent Overstreet 1243576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1244576ed913SChristoph Hellwig struct page *page = pages[i]; 12452cefe4dbSKent Overstreet 1246576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 124734cdb8c8SKeith Busch if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1248c58c0074SKeith Busch ret = bio_iov_add_zone_append_page(bio, page, len, 1249c58c0074SKeith Busch offset); 1250e97424fdSKeith Busch if (ret) 1251325347d9SKeith Busch break; 125234cdb8c8SKeith Busch } else 125334cdb8c8SKeith Busch bio_iov_add_page(bio, page, len, offset); 125434cdb8c8SKeith Busch 1255576ed913SChristoph Hellwig offset = 0; 12562cefe4dbSKent Overstreet } 12572cefe4dbSKent Overstreet 1258480cb846SAl Viro iov_iter_revert(iter, left); 1259e97424fdSKeith Busch out: 1260e97424fdSKeith Busch while (i < nr_pages) 1261e97424fdSKeith Busch put_page(pages[i++]); 1262e97424fdSKeith Busch 1263325347d9SKeith Busch return ret; 12642cefe4dbSKent Overstreet } 126517d51b10SMartin Wilck 126617d51b10SMartin Wilck /** 12676d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 126817d51b10SMartin Wilck * @bio: bio to add pages to 12696d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 127017d51b10SMartin Wilck * 12716d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 12726d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 12736d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1274c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1275c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1276c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1277c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1278c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1279c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 12806d0c48aeSJens Axboe * 128117d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 12825cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 12836d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 12846d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 128517d51b10SMartin Wilck */ 128617d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 128717d51b10SMartin Wilck { 1288c42bca92SPavel Begunkov int ret = 0; 128914eacf12SChristoph Hellwig 1290c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1291fa5fa8ecSPavel Begunkov bio_iov_bvec_set(bio, iter); 1292fa5fa8ecSPavel Begunkov iov_iter_advance(iter, bio->bi_iter.bi_size); 1293fa5fa8ecSPavel Begunkov return 0; 129486004515SChristoph Hellwig } 129517d51b10SMartin Wilck 129617d51b10SMartin Wilck do { 12976d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 129879d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 129917d51b10SMartin Wilck 130014eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 130117d51b10SMartin Wilck } 130229b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 13032cefe4dbSKent Overstreet 13044246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1305f9c78b2bSJens Axboe { 130665e53aabSChristoph Hellwig complete(bio->bi_private); 1307f9c78b2bSJens Axboe } 1308f9c78b2bSJens Axboe 1309f9c78b2bSJens Axboe /** 1310f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1311f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1312f9c78b2bSJens Axboe * 1313f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1314f9c78b2bSJens Axboe * bio_endio() on failure. 13153d289d68SJan Kara * 13163d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 13173d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 13183d289d68SJan Kara * on his own. 1319f9c78b2bSJens Axboe */ 13204e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1321f9c78b2bSJens Axboe { 1322309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1323309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1324de6a78b6SMing Lei unsigned long hang_check; 1325f9c78b2bSJens Axboe 132665e53aabSChristoph Hellwig bio->bi_private = &done; 1327f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 13281eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 13294e49ea4aSMike Christie submit_bio(bio); 1330de6a78b6SMing Lei 1331de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1332de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1333de6a78b6SMing Lei if (hang_check) 1334de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1335de6a78b6SMing Lei hang_check * (HZ/2))) 1336de6a78b6SMing Lei ; 1337de6a78b6SMing Lei else 133865e53aabSChristoph Hellwig wait_for_completion_io(&done); 1339f9c78b2bSJens Axboe 134065e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1341f9c78b2bSJens Axboe } 1342f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1343f9c78b2bSJens Axboe 1344d4aa57a1SJens Axboe void __bio_advance(struct bio *bio, unsigned bytes) 1345f9c78b2bSJens Axboe { 1346f9c78b2bSJens Axboe if (bio_integrity(bio)) 1347f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1348f9c78b2bSJens Axboe 1349a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1350f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1351f9c78b2bSJens Axboe } 1352d4aa57a1SJens Axboe EXPORT_SYMBOL(__bio_advance); 1353f9c78b2bSJens Axboe 135445db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 135545db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1356f9c78b2bSJens Axboe { 135745db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 1358f8b679a0SChristoph Hellwig struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1359f8b679a0SChristoph Hellwig struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1360f8b679a0SChristoph Hellwig unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1361403d5034SChristoph Hellwig void *src_buf = bvec_kmap_local(&src_bv); 1362403d5034SChristoph Hellwig void *dst_buf = bvec_kmap_local(&dst_bv); 136345db54d5SKent Overstreet 1364403d5034SChristoph Hellwig memcpy(dst_buf, src_buf, bytes); 1365403d5034SChristoph Hellwig 1366403d5034SChristoph Hellwig kunmap_local(dst_buf); 1367f8b679a0SChristoph Hellwig kunmap_local(src_buf); 13686e6e811dSKent Overstreet 136922b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 137022b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 137145db54d5SKent Overstreet } 137245db54d5SKent Overstreet } 137345db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 137445db54d5SKent Overstreet 137545db54d5SKent Overstreet /** 137645db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 137745db54d5SKent Overstreet * @src: source bio 137845db54d5SKent Overstreet * @dst: destination bio 137945db54d5SKent Overstreet * 138045db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 138145db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 138245db54d5SKent Overstreet */ 138345db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 138445db54d5SKent Overstreet { 138545db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 138645db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 138745db54d5SKent Overstreet 138845db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 138945db54d5SKent Overstreet } 139045db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 139145db54d5SKent Overstreet 1392491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 13931dfa0f68SChristoph Hellwig { 13941dfa0f68SChristoph Hellwig struct bio_vec *bvec; 13956dc4f100SMing Lei struct bvec_iter_all iter_all; 13961dfa0f68SChristoph Hellwig 13972b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 13981dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 13991dfa0f68SChristoph Hellwig } 1400491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 14011dfa0f68SChristoph Hellwig 1402f9c78b2bSJens Axboe /* 1403f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1404f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1405f9c78b2bSJens Axboe * 1406f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1407f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1408f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1409f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1410f9c78b2bSJens Axboe * in process context. 1411f9c78b2bSJens Axboe * 1412f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1413f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1414f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1415f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1416f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1417f9c78b2bSJens Axboe * 1418f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1419f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1420f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1421f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1422f9c78b2bSJens Axboe * pagecache. 1423f9c78b2bSJens Axboe * 1424f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1425f9c78b2bSJens Axboe * deferred bio dirtying paths. 1426f9c78b2bSJens Axboe */ 1427f9c78b2bSJens Axboe 1428f9c78b2bSJens Axboe /* 1429f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1430f9c78b2bSJens Axboe */ 1431f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1432f9c78b2bSJens Axboe { 1433f9c78b2bSJens Axboe struct bio_vec *bvec; 14346dc4f100SMing Lei struct bvec_iter_all iter_all; 1435f9c78b2bSJens Axboe 14362b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 14373bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 14383bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1439f9c78b2bSJens Axboe } 1440f9c78b2bSJens Axboe } 1441f9c78b2bSJens Axboe 1442f9c78b2bSJens Axboe /* 1443f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1444f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1445f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 144624d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1447f9c78b2bSJens Axboe * 1448f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1449ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1450ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1451f9c78b2bSJens Axboe */ 1452f9c78b2bSJens Axboe 1453f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1454f9c78b2bSJens Axboe 1455f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1456f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1457f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1458f9c78b2bSJens Axboe 1459f9c78b2bSJens Axboe /* 1460f9c78b2bSJens Axboe * This runs in process context 1461f9c78b2bSJens Axboe */ 1462f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1463f9c78b2bSJens Axboe { 146424d5493fSChristoph Hellwig struct bio *bio, *next; 1465f9c78b2bSJens Axboe 146624d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 146724d5493fSChristoph Hellwig next = bio_dirty_list; 1468f9c78b2bSJens Axboe bio_dirty_list = NULL; 146924d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1470f9c78b2bSJens Axboe 147124d5493fSChristoph Hellwig while ((bio = next) != NULL) { 147224d5493fSChristoph Hellwig next = bio->bi_private; 1473f9c78b2bSJens Axboe 1474d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1475f9c78b2bSJens Axboe bio_put(bio); 1476f9c78b2bSJens Axboe } 1477f9c78b2bSJens Axboe } 1478f9c78b2bSJens Axboe 1479f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1480f9c78b2bSJens Axboe { 1481f9c78b2bSJens Axboe struct bio_vec *bvec; 148224d5493fSChristoph Hellwig unsigned long flags; 14836dc4f100SMing Lei struct bvec_iter_all iter_all; 1484f9c78b2bSJens Axboe 14852b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 148624d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 148724d5493fSChristoph Hellwig goto defer; 1488f9c78b2bSJens Axboe } 1489f9c78b2bSJens Axboe 1490d241a95fSChristoph Hellwig bio_release_pages(bio, false); 149124d5493fSChristoph Hellwig bio_put(bio); 149224d5493fSChristoph Hellwig return; 149324d5493fSChristoph Hellwig defer: 1494f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1495f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1496f9c78b2bSJens Axboe bio_dirty_list = bio; 1497f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1498f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1499f9c78b2bSJens Axboe } 1500f9c78b2bSJens Axboe 1501c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1502c4cf5261SJens Axboe { 1503c4cf5261SJens Axboe /* 1504c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1505c4cf5261SJens Axboe * we always end io on the first invocation. 1506c4cf5261SJens Axboe */ 1507c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1508c4cf5261SJens Axboe return true; 1509c4cf5261SJens Axboe 1510c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1511c4cf5261SJens Axboe 1512326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1513b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1514c4cf5261SJens Axboe return true; 1515326e1dbbSMike Snitzer } 1516c4cf5261SJens Axboe 1517c4cf5261SJens Axboe return false; 1518c4cf5261SJens Axboe } 1519c4cf5261SJens Axboe 1520f9c78b2bSJens Axboe /** 1521f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1522f9c78b2bSJens Axboe * @bio: bio 1523f9c78b2bSJens Axboe * 1524f9c78b2bSJens Axboe * Description: 15254246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 15264246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 15274246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1528fbbaf700SNeilBrown * 1529fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1530fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 153160b6a7e6SEdward Hsieh * last time. 1532f9c78b2bSJens Axboe **/ 15334246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1534f9c78b2bSJens Axboe { 1535ba8c6967SChristoph Hellwig again: 15362b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1537ba8c6967SChristoph Hellwig return; 15387c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 15397c20f116SChristoph Hellwig return; 1540f9c78b2bSJens Axboe 1541aa1b46dcSTejun Heo rq_qos_done_bio(bio); 154267b42d0bSJosef Bacik 154360b6a7e6SEdward Hsieh if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 15443caee463SPavel Begunkov trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 154560b6a7e6SEdward Hsieh bio_clear_flag(bio, BIO_TRACE_COMPLETION); 154660b6a7e6SEdward Hsieh } 154760b6a7e6SEdward Hsieh 1548f9c78b2bSJens Axboe /* 1549ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1550ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1551ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1552ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1553ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1554ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1555f9c78b2bSJens Axboe */ 1556f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 155738f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1558ba8c6967SChristoph Hellwig goto again; 1559ba8c6967SChristoph Hellwig } 1560ba8c6967SChristoph Hellwig 15619e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1562b222dd2fSShaohua Li /* release cgroup info */ 1563b222dd2fSShaohua Li bio_uninit(bio); 1564f9c78b2bSJens Axboe if (bio->bi_end_io) 15654246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1566f9c78b2bSJens Axboe } 1567f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1568f9c78b2bSJens Axboe 1569f9c78b2bSJens Axboe /** 1570f9c78b2bSJens Axboe * bio_split - split a bio 1571f9c78b2bSJens Axboe * @bio: bio to split 1572f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1573f9c78b2bSJens Axboe * @gfp: gfp mask 1574f9c78b2bSJens Axboe * @bs: bio set to allocate from 1575f9c78b2bSJens Axboe * 1576f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1577f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1578f9c78b2bSJens Axboe * 1579f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1580dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1581dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1582f9c78b2bSJens Axboe */ 1583f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1584f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1585f9c78b2bSJens Axboe { 1586f341a4d3SMikulas Patocka struct bio *split; 1587f9c78b2bSJens Axboe 1588f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1589f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1590f9c78b2bSJens Axboe 15910512a75bSKeith Busch /* Zone append commands cannot be split */ 15920512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 15930512a75bSKeith Busch return NULL; 15940512a75bSKeith Busch 1595abfc426dSChristoph Hellwig split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs); 1596f9c78b2bSJens Axboe if (!split) 1597f9c78b2bSJens Axboe return NULL; 1598f9c78b2bSJens Axboe 1599f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1600f9c78b2bSJens Axboe 1601f9c78b2bSJens Axboe if (bio_integrity(split)) 1602fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1603f9c78b2bSJens Axboe 1604f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1605f9c78b2bSJens Axboe 1606fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 160720d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1608fbbaf700SNeilBrown 1609f9c78b2bSJens Axboe return split; 1610f9c78b2bSJens Axboe } 1611f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1612f9c78b2bSJens Axboe 1613f9c78b2bSJens Axboe /** 1614f9c78b2bSJens Axboe * bio_trim - trim a bio 1615f9c78b2bSJens Axboe * @bio: bio to trim 1616f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1617f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1618e83502caSChaitanya Kulkarni * 1619e83502caSChaitanya Kulkarni * This function is typically used for bios that are cloned and submitted 1620e83502caSChaitanya Kulkarni * to the underlying device in parts. 1621f9c78b2bSJens Axboe */ 1622e83502caSChaitanya Kulkarni void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1623f9c78b2bSJens Axboe { 1624e83502caSChaitanya Kulkarni if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 16258535c018SMing Lei offset + size > bio_sectors(bio))) 1626e83502caSChaitanya Kulkarni return; 1627f9c78b2bSJens Axboe 1628f9c78b2bSJens Axboe size <<= 9; 1629f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1630f9c78b2bSJens Axboe return; 1631f9c78b2bSJens Axboe 1632f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1633f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1634376a78abSDmitry Monakhov 1635376a78abSDmitry Monakhov if (bio_integrity(bio)) 1636fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1637f9c78b2bSJens Axboe } 1638f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1639f9c78b2bSJens Axboe 1640f9c78b2bSJens Axboe /* 1641f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1642f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1643f9c78b2bSJens Axboe */ 16448aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1645f9c78b2bSJens Axboe { 16467a800a20SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1647f9c78b2bSJens Axboe 16488aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1649f9c78b2bSJens Axboe } 1650f9c78b2bSJens Axboe 1651917a38c7SKent Overstreet /* 1652917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1653917a38c7SKent Overstreet * 1654917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1655917a38c7SKent Overstreet * kzalloc()). 1656917a38c7SKent Overstreet */ 1657917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1658f9c78b2bSJens Axboe { 1659be4d234dSJens Axboe bio_alloc_cache_destroy(bs); 1660f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1661f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1662917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1663f9c78b2bSJens Axboe 16648aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 16658aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1666f9c78b2bSJens Axboe 1667f9c78b2bSJens Axboe bioset_integrity_free(bs); 1668917a38c7SKent Overstreet if (bs->bio_slab) 1669f9c78b2bSJens Axboe bio_put_slab(bs); 1670917a38c7SKent Overstreet bs->bio_slab = NULL; 1671917a38c7SKent Overstreet } 1672917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1673f9c78b2bSJens Axboe 1674011067b0SNeilBrown /** 1675917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1676dad08527SKent Overstreet * @bs: pool to initialize 1677917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1678917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1679917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1680917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1681917a38c7SKent Overstreet * 1682dad08527SKent Overstreet * Description: 1683dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1684dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1685dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1686dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1687dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1688dad08527SKent Overstreet * or things will break badly. 1689dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1690abfc426dSChristoph Hellwig * for allocating iovecs. This pool is not needed e.g. for bio_init_clone(). 1691abfc426dSChristoph Hellwig * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used 1692abfc426dSChristoph Hellwig * to dispatch queued requests when the mempool runs out of space. 1693dad08527SKent Overstreet * 1694917a38c7SKent Overstreet */ 1695917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1696917a38c7SKent Overstreet unsigned int pool_size, 1697917a38c7SKent Overstreet unsigned int front_pad, 1698917a38c7SKent Overstreet int flags) 1699917a38c7SKent Overstreet { 1700917a38c7SKent Overstreet bs->front_pad = front_pad; 17019f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 17029f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 17039f180e31SMing Lei else 17049f180e31SMing Lei bs->back_pad = 0; 1705917a38c7SKent Overstreet 1706917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1707917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1708917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1709917a38c7SKent Overstreet 171049d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1711917a38c7SKent Overstreet if (!bs->bio_slab) 1712917a38c7SKent Overstreet return -ENOMEM; 1713917a38c7SKent Overstreet 1714917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1715917a38c7SKent Overstreet goto bad; 1716917a38c7SKent Overstreet 1717917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1718917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1719917a38c7SKent Overstreet goto bad; 1720917a38c7SKent Overstreet 1721be4d234dSJens Axboe if (flags & BIOSET_NEED_RESCUER) { 1722be4d234dSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", 1723be4d234dSJens Axboe WQ_MEM_RECLAIM, 0); 1724917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1725917a38c7SKent Overstreet goto bad; 1726be4d234dSJens Axboe } 1727be4d234dSJens Axboe if (flags & BIOSET_PERCPU_CACHE) { 1728be4d234dSJens Axboe bs->cache = alloc_percpu(struct bio_alloc_cache); 1729be4d234dSJens Axboe if (!bs->cache) 1730be4d234dSJens Axboe goto bad; 1731be4d234dSJens Axboe cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1732be4d234dSJens Axboe } 1733917a38c7SKent Overstreet 1734917a38c7SKent Overstreet return 0; 1735917a38c7SKent Overstreet bad: 1736917a38c7SKent Overstreet bioset_exit(bs); 1737917a38c7SKent Overstreet return -ENOMEM; 1738917a38c7SKent Overstreet } 1739917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1740917a38c7SKent Overstreet 1741de76fd89SChristoph Hellwig static int __init init_bio(void) 1742f9c78b2bSJens Axboe { 1743f9c78b2bSJens Axboe int i; 1744f9c78b2bSJens Axboe 1745f9c78b2bSJens Axboe bio_integrity_init(); 1746de76fd89SChristoph Hellwig 1747de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1748f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1749f9c78b2bSJens Axboe 1750de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1751de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1752f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1753f9c78b2bSJens Axboe } 1754f9c78b2bSJens Axboe 1755be4d234dSJens Axboe cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1756be4d234dSJens Axboe bio_cpu_dead); 1757be4d234dSJens Axboe 175812c5b70cSJens Axboe if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, 175912c5b70cSJens Axboe BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE)) 1760f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1761f9c78b2bSJens Axboe 1762f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1763f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1764f9c78b2bSJens Axboe 1765f9c78b2bSJens Axboe return 0; 1766f9c78b2bSJens Axboe } 1767f9c78b2bSJens Axboe subsys_initcall(init_bio); 1768