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; 227f9c78b2bSJens Axboe void *p; 228f9c78b2bSJens Axboe 2299ae3b3f5SJens Axboe bio_uninit(bio); 230f9c78b2bSJens Axboe 231f9c78b2bSJens Axboe if (bs) { 2327a800a20SChristoph Hellwig bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs); 233f9c78b2bSJens Axboe 234f9c78b2bSJens Axboe /* 235f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 236f9c78b2bSJens Axboe */ 237f9c78b2bSJens Axboe p = bio; 238f9c78b2bSJens Axboe p -= bs->front_pad; 239f9c78b2bSJens Axboe 2408aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 241f9c78b2bSJens Axboe } else { 242f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 243f9c78b2bSJens Axboe kfree(bio); 244f9c78b2bSJens Axboe } 245f9c78b2bSJens Axboe } 246f9c78b2bSJens Axboe 2479ae3b3f5SJens Axboe /* 2489ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2499ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2509ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2519ae3b3f5SJens Axboe */ 25249add496SChristoph Hellwig void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table, 25349add496SChristoph Hellwig unsigned short max_vecs, unsigned int opf) 254f9c78b2bSJens Axboe { 255da521626SJens Axboe bio->bi_next = NULL; 25649add496SChristoph Hellwig bio->bi_bdev = bdev; 25749add496SChristoph Hellwig bio->bi_opf = opf; 258da521626SJens Axboe bio->bi_flags = 0; 259da521626SJens Axboe bio->bi_ioprio = 0; 260da521626SJens Axboe bio->bi_status = 0; 261da521626SJens Axboe bio->bi_iter.bi_sector = 0; 262da521626SJens Axboe bio->bi_iter.bi_size = 0; 263da521626SJens Axboe bio->bi_iter.bi_idx = 0; 264da521626SJens Axboe bio->bi_iter.bi_bvec_done = 0; 265da521626SJens Axboe bio->bi_end_io = NULL; 266da521626SJens Axboe bio->bi_private = NULL; 267da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP 268da521626SJens Axboe bio->bi_blkg = NULL; 269da521626SJens Axboe bio->bi_issue.value = 0; 27049add496SChristoph Hellwig if (bdev) 27149add496SChristoph Hellwig bio_associate_blkg(bio); 272da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP_IOCOST 273da521626SJens Axboe bio->bi_iocost_cost = 0; 274da521626SJens Axboe #endif 275da521626SJens Axboe #endif 276da521626SJens Axboe #ifdef CONFIG_BLK_INLINE_ENCRYPTION 277da521626SJens Axboe bio->bi_crypt_context = NULL; 278da521626SJens Axboe #endif 279da521626SJens Axboe #ifdef CONFIG_BLK_DEV_INTEGRITY 280da521626SJens Axboe bio->bi_integrity = NULL; 281da521626SJens Axboe #endif 282da521626SJens Axboe bio->bi_vcnt = 0; 283da521626SJens Axboe 284c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 285dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2863e08773cSChristoph Hellwig bio->bi_cookie = BLK_QC_T_NONE; 2873a83f467SMing Lei 2883a83f467SMing Lei bio->bi_max_vecs = max_vecs; 289da521626SJens Axboe bio->bi_io_vec = table; 290da521626SJens Axboe bio->bi_pool = NULL; 291f9c78b2bSJens Axboe } 292f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 293f9c78b2bSJens Axboe 294f9c78b2bSJens Axboe /** 295f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 296f9c78b2bSJens Axboe * @bio: bio to reset 297a7c50c94SChristoph Hellwig * @bdev: block device to use the bio for 298a7c50c94SChristoph Hellwig * @opf: operation and flags for bio 299f9c78b2bSJens Axboe * 300f9c78b2bSJens Axboe * Description: 301f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 302f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 303f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 304f9c78b2bSJens Axboe * comment in struct bio. 305f9c78b2bSJens Axboe */ 306a7c50c94SChristoph Hellwig void bio_reset(struct bio *bio, struct block_device *bdev, unsigned int opf) 307f9c78b2bSJens Axboe { 3089ae3b3f5SJens Axboe bio_uninit(bio); 309f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 310c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 311a7c50c94SChristoph Hellwig bio->bi_bdev = bdev; 31278e34374SChristoph Hellwig if (bio->bi_bdev) 31378e34374SChristoph Hellwig bio_associate_blkg(bio); 314a7c50c94SChristoph Hellwig bio->bi_opf = opf; 315f9c78b2bSJens Axboe } 316f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 317f9c78b2bSJens Axboe 31838f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 319f9c78b2bSJens Axboe { 3204246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3214246a0b6SChristoph Hellwig 3223edf5346SYufen Yu if (bio->bi_status && !parent->bi_status) 3234e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 324f9c78b2bSJens Axboe bio_put(bio); 32538f8baaeSChristoph Hellwig return parent; 32638f8baaeSChristoph Hellwig } 32738f8baaeSChristoph Hellwig 32838f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 32938f8baaeSChristoph Hellwig { 33038f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 331f9c78b2bSJens Axboe } 332f9c78b2bSJens Axboe 333f9c78b2bSJens Axboe /** 334f9c78b2bSJens Axboe * bio_chain - chain bio completions 335f9c78b2bSJens Axboe * @bio: the target bio 3365b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 337f9c78b2bSJens Axboe * 338f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 339f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 340f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 341f9c78b2bSJens Axboe * 342f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 343f9c78b2bSJens Axboe */ 344f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 345f9c78b2bSJens Axboe { 346f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 347f9c78b2bSJens Axboe 348f9c78b2bSJens Axboe bio->bi_private = parent; 349f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 350c4cf5261SJens Axboe bio_inc_remaining(parent); 351f9c78b2bSJens Axboe } 352f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 353f9c78b2bSJens Axboe 3540a3140eaSChaitanya Kulkarni struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev, 3550a3140eaSChaitanya Kulkarni unsigned int nr_pages, unsigned int opf, gfp_t gfp) 3563b005bf6SChristoph Hellwig { 35707888c66SChristoph Hellwig struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp); 3580a3140eaSChaitanya Kulkarni 3593b005bf6SChristoph Hellwig if (bio) { 3603b005bf6SChristoph Hellwig bio_chain(bio, new); 3613b005bf6SChristoph Hellwig submit_bio(bio); 3623b005bf6SChristoph Hellwig } 3633b005bf6SChristoph Hellwig 3643b005bf6SChristoph Hellwig return new; 3653b005bf6SChristoph Hellwig } 3663b005bf6SChristoph Hellwig EXPORT_SYMBOL_GPL(blk_next_bio); 3673b005bf6SChristoph Hellwig 368f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 369f9c78b2bSJens Axboe { 370f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 371f9c78b2bSJens Axboe struct bio *bio; 372f9c78b2bSJens Axboe 373f9c78b2bSJens Axboe while (1) { 374f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 375f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 376f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 377f9c78b2bSJens Axboe 378f9c78b2bSJens Axboe if (!bio) 379f9c78b2bSJens Axboe break; 380f9c78b2bSJens Axboe 381ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 382f9c78b2bSJens Axboe } 383f9c78b2bSJens Axboe } 384f9c78b2bSJens Axboe 385f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 386f9c78b2bSJens Axboe { 387f9c78b2bSJens Axboe struct bio_list punt, nopunt; 388f9c78b2bSJens Axboe struct bio *bio; 389f9c78b2bSJens Axboe 39047e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 39147e0fb46SNeilBrown return; 392f9c78b2bSJens Axboe /* 393f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 394f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 395f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 396f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 397f9c78b2bSJens Axboe * our own rescuer would be bad. 398f9c78b2bSJens Axboe * 399f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 400f9c78b2bSJens Axboe * remove from the middle of the list: 401f9c78b2bSJens Axboe */ 402f9c78b2bSJens Axboe 403f9c78b2bSJens Axboe bio_list_init(&punt); 404f9c78b2bSJens Axboe bio_list_init(&nopunt); 405f9c78b2bSJens Axboe 406f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 407f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 408f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 409f9c78b2bSJens Axboe 410f5fe1b51SNeilBrown bio_list_init(&nopunt); 411f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 412f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 413f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 414f9c78b2bSJens Axboe 415f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 416f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 417f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 418f9c78b2bSJens Axboe 419f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 420f9c78b2bSJens Axboe } 421f9c78b2bSJens Axboe 4220df71650SMike Snitzer static struct bio *bio_alloc_percpu_cache(struct block_device *bdev, 4230df71650SMike Snitzer unsigned short nr_vecs, unsigned int opf, gfp_t gfp, 4240df71650SMike Snitzer struct bio_set *bs) 4250df71650SMike Snitzer { 4260df71650SMike Snitzer struct bio_alloc_cache *cache; 4270df71650SMike Snitzer struct bio *bio; 4280df71650SMike Snitzer 4290df71650SMike Snitzer cache = per_cpu_ptr(bs->cache, get_cpu()); 4300df71650SMike Snitzer if (!cache->free_list) { 4310df71650SMike Snitzer put_cpu(); 4320df71650SMike Snitzer return NULL; 4330df71650SMike Snitzer } 4340df71650SMike Snitzer bio = cache->free_list; 4350df71650SMike Snitzer cache->free_list = bio->bi_next; 4360df71650SMike Snitzer cache->nr--; 4370df71650SMike Snitzer put_cpu(); 4380df71650SMike Snitzer 4390df71650SMike Snitzer bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf); 4400df71650SMike Snitzer bio->bi_pool = bs; 4410df71650SMike Snitzer return bio; 4420df71650SMike Snitzer } 4430df71650SMike Snitzer 444f9c78b2bSJens Axboe /** 445f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 446609be106SChristoph Hellwig * @bdev: block device to allocate the bio for (can be %NULL) 447609be106SChristoph Hellwig * @nr_vecs: number of bvecs to pre-allocate 448609be106SChristoph Hellwig * @opf: operation and flags for bio 449519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 450f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 451f9c78b2bSJens Axboe * 4523175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 453f9c78b2bSJens Axboe * 4543175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4553175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4563175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4573175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4583175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4593175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 460f9c78b2bSJens Axboe * 4613175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4623175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 463ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 464f9c78b2bSJens Axboe * stack overflows. 465f9c78b2bSJens Axboe * 4663175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4673175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 468f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 469f9c78b2bSJens Axboe * thread. 470f9c78b2bSJens Axboe * 471f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 472f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 473ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 474f9c78b2bSJens Axboe * for per bio allocations. 475f9c78b2bSJens Axboe * 4760df71650SMike Snitzer * If REQ_ALLOC_CACHE is set, the final put of the bio MUST be done from process 4770df71650SMike Snitzer * context, not hard/soft IRQ. 4780df71650SMike Snitzer * 4793175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 480f9c78b2bSJens Axboe */ 481609be106SChristoph Hellwig struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs, 482609be106SChristoph Hellwig unsigned int opf, gfp_t gfp_mask, 4837a88fa19SDan Carpenter struct bio_set *bs) 484f9c78b2bSJens Axboe { 485f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 486f9c78b2bSJens Axboe struct bio *bio; 487f9c78b2bSJens Axboe void *p; 488f9c78b2bSJens Axboe 489609be106SChristoph Hellwig /* should not use nobvec bioset for nr_vecs > 0 */ 490609be106SChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0)) 491f9c78b2bSJens Axboe return NULL; 492f9c78b2bSJens Axboe 4930df71650SMike Snitzer if (opf & REQ_ALLOC_CACHE) { 4940df71650SMike Snitzer if (bs->cache && nr_vecs <= BIO_INLINE_VECS) { 4950df71650SMike Snitzer bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf, 4960df71650SMike Snitzer gfp_mask, bs); 4970df71650SMike Snitzer if (bio) 4980df71650SMike Snitzer return bio; 4990df71650SMike Snitzer /* 5000df71650SMike Snitzer * No cached bio available, bio returned below marked with 5010df71650SMike Snitzer * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache. 5020df71650SMike Snitzer */ 5030df71650SMike Snitzer } else { 5040df71650SMike Snitzer opf &= ~REQ_ALLOC_CACHE; 5050df71650SMike Snitzer } 5060df71650SMike Snitzer } 5070df71650SMike Snitzer 508f9c78b2bSJens Axboe /* 5093175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 5103175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 5113175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 512f9c78b2bSJens Axboe * 5133175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 5143175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 5153175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 5163175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 5173175199aSChristoph Hellwig * mempool's reserve. 518f9c78b2bSJens Axboe * 519f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 5203175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 5213175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 5223175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 5233175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 5243175199aSChristoph Hellwig * gfp_flags. 525f9c78b2bSJens Axboe */ 526f5fe1b51SNeilBrown if (current->bio_list && 527f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 52847e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 52947e0fb46SNeilBrown bs->rescue_workqueue) 530d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 531f9c78b2bSJens Axboe 5328aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 533f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 534f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 535f9c78b2bSJens Axboe gfp_mask = saved_gfp; 5368aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 537f9c78b2bSJens Axboe } 538f9c78b2bSJens Axboe if (unlikely(!p)) 539f9c78b2bSJens Axboe return NULL; 540f9c78b2bSJens Axboe 5413175199aSChristoph Hellwig bio = p + bs->front_pad; 542609be106SChristoph Hellwig if (nr_vecs > BIO_INLINE_VECS) { 5433175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 544f9c78b2bSJens Axboe 545609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 546f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 547f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 548f9c78b2bSJens Axboe gfp_mask = saved_gfp; 549609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 550f9c78b2bSJens Axboe } 551f9c78b2bSJens Axboe if (unlikely(!bvl)) 552f9c78b2bSJens Axboe goto err_free; 553f9c78b2bSJens Axboe 55449add496SChristoph Hellwig bio_init(bio, bdev, bvl, nr_vecs, opf); 555609be106SChristoph Hellwig } else if (nr_vecs) { 55649add496SChristoph Hellwig bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf); 5573175199aSChristoph Hellwig } else { 55849add496SChristoph Hellwig bio_init(bio, bdev, NULL, 0, opf); 559f9c78b2bSJens Axboe } 560f9c78b2bSJens Axboe 561f9c78b2bSJens Axboe bio->bi_pool = bs; 562f9c78b2bSJens Axboe return bio; 563f9c78b2bSJens Axboe 564f9c78b2bSJens Axboe err_free: 5658aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 566f9c78b2bSJens Axboe return NULL; 567f9c78b2bSJens Axboe } 568f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 569f9c78b2bSJens Axboe 5703175199aSChristoph Hellwig /** 5713175199aSChristoph Hellwig * bio_kmalloc - kmalloc a bio for I/O 5723175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 5733175199aSChristoph Hellwig * @nr_iovecs: number of iovecs to pre-allocate 5743175199aSChristoph Hellwig * 5753175199aSChristoph Hellwig * Use kmalloc to allocate and initialize a bio. 5763175199aSChristoph Hellwig * 5773175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 5783175199aSChristoph Hellwig */ 5790f2e6ab8SChristoph Hellwig struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) 5803175199aSChristoph Hellwig { 5813175199aSChristoph Hellwig struct bio *bio; 5823175199aSChristoph Hellwig 5833175199aSChristoph Hellwig if (nr_iovecs > UIO_MAXIOV) 5843175199aSChristoph Hellwig return NULL; 5853175199aSChristoph Hellwig 5863175199aSChristoph Hellwig bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 5873175199aSChristoph Hellwig if (unlikely(!bio)) 5883175199aSChristoph Hellwig return NULL; 58949add496SChristoph Hellwig bio_init(bio, NULL, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs, 59049add496SChristoph Hellwig 0); 5913175199aSChristoph Hellwig bio->bi_pool = NULL; 5923175199aSChristoph Hellwig return bio; 5933175199aSChristoph Hellwig } 5943175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 5953175199aSChristoph Hellwig 5966f822e1bSChristoph Hellwig void zero_fill_bio(struct bio *bio) 597f9c78b2bSJens Axboe { 598f9c78b2bSJens Axboe struct bio_vec bv; 599f9c78b2bSJens Axboe struct bvec_iter iter; 600f9c78b2bSJens Axboe 601ab6c340eSChristoph Hellwig bio_for_each_segment(bv, bio, iter) 602ab6c340eSChristoph Hellwig memzero_bvec(&bv); 603f9c78b2bSJens Axboe } 6046f822e1bSChristoph Hellwig EXPORT_SYMBOL(zero_fill_bio); 605f9c78b2bSJens Axboe 60683c9c547SMing Lei /** 60783c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 60883c9c547SMing Lei * @bio: the bio to be truncated 60983c9c547SMing Lei * @new_size: new size for truncating the bio 61083c9c547SMing Lei * 61183c9c547SMing Lei * Description: 61283c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 61383c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 61483c9c547SMing Lei * be used for handling corner cases, such as bio eod. 61583c9c547SMing Lei */ 6164f7ab09aSChristoph Hellwig static void bio_truncate(struct bio *bio, unsigned new_size) 61785a8ce62SMing Lei { 61885a8ce62SMing Lei struct bio_vec bv; 61985a8ce62SMing Lei struct bvec_iter iter; 62085a8ce62SMing Lei unsigned int done = 0; 62185a8ce62SMing Lei bool truncated = false; 62285a8ce62SMing Lei 62385a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 62485a8ce62SMing Lei return; 62585a8ce62SMing Lei 62683c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 62785a8ce62SMing Lei goto exit; 62885a8ce62SMing Lei 62985a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 63085a8ce62SMing Lei if (done + bv.bv_len > new_size) { 63185a8ce62SMing Lei unsigned offset; 63285a8ce62SMing Lei 63385a8ce62SMing Lei if (!truncated) 63485a8ce62SMing Lei offset = new_size - done; 63585a8ce62SMing Lei else 63685a8ce62SMing Lei offset = 0; 6373ee859e3SOGAWA Hirofumi zero_user(bv.bv_page, bv.bv_offset + offset, 6383ee859e3SOGAWA Hirofumi bv.bv_len - offset); 63985a8ce62SMing Lei truncated = true; 64085a8ce62SMing Lei } 64185a8ce62SMing Lei done += bv.bv_len; 64285a8ce62SMing Lei } 64385a8ce62SMing Lei 64485a8ce62SMing Lei exit: 64585a8ce62SMing Lei /* 64685a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 64785a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 64885a8ce62SMing Lei * in its .end_bio() callback. 64985a8ce62SMing Lei * 65085a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 65185a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 65285a8ce62SMing Lei */ 65385a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 65485a8ce62SMing Lei } 65585a8ce62SMing Lei 656f9c78b2bSJens Axboe /** 65729125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 65829125ed6SChristoph Hellwig * @bio: bio to truncate 65929125ed6SChristoph Hellwig * 66029125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 66129125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 66229125ed6SChristoph Hellwig * 66329125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 66429125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 66529125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 66629125ed6SChristoph Hellwig * sector" case. 66729125ed6SChristoph Hellwig */ 66829125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 66929125ed6SChristoph Hellwig { 670309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 67129125ed6SChristoph Hellwig 67229125ed6SChristoph Hellwig if (!maxsector) 67329125ed6SChristoph Hellwig return; 67429125ed6SChristoph Hellwig 67529125ed6SChristoph Hellwig /* 67629125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 67729125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 67829125ed6SChristoph Hellwig * an EIO. 67929125ed6SChristoph Hellwig */ 68029125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 68129125ed6SChristoph Hellwig return; 68229125ed6SChristoph Hellwig 68329125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 68429125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 68529125ed6SChristoph Hellwig return; 68629125ed6SChristoph Hellwig 68729125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 68829125ed6SChristoph Hellwig } 68929125ed6SChristoph Hellwig 690be4d234dSJens Axboe #define ALLOC_CACHE_MAX 512 691be4d234dSJens Axboe #define ALLOC_CACHE_SLACK 64 692be4d234dSJens Axboe 693be4d234dSJens Axboe static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 694be4d234dSJens Axboe unsigned int nr) 695be4d234dSJens Axboe { 696be4d234dSJens Axboe unsigned int i = 0; 697be4d234dSJens Axboe struct bio *bio; 698be4d234dSJens Axboe 699fcade2ceSJens Axboe while ((bio = cache->free_list) != NULL) { 700fcade2ceSJens Axboe cache->free_list = bio->bi_next; 701be4d234dSJens Axboe cache->nr--; 702be4d234dSJens Axboe bio_free(bio); 703be4d234dSJens Axboe if (++i == nr) 704be4d234dSJens Axboe break; 705be4d234dSJens Axboe } 706be4d234dSJens Axboe } 707be4d234dSJens Axboe 708be4d234dSJens Axboe static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 709be4d234dSJens Axboe { 710be4d234dSJens Axboe struct bio_set *bs; 711be4d234dSJens Axboe 712be4d234dSJens Axboe bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 713be4d234dSJens Axboe if (bs->cache) { 714be4d234dSJens Axboe struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 715be4d234dSJens Axboe 716be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 717be4d234dSJens Axboe } 718be4d234dSJens Axboe return 0; 719be4d234dSJens Axboe } 720be4d234dSJens Axboe 721be4d234dSJens Axboe static void bio_alloc_cache_destroy(struct bio_set *bs) 722be4d234dSJens Axboe { 723be4d234dSJens Axboe int cpu; 724be4d234dSJens Axboe 725be4d234dSJens Axboe if (!bs->cache) 726be4d234dSJens Axboe return; 727be4d234dSJens Axboe 728be4d234dSJens Axboe cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 729be4d234dSJens Axboe for_each_possible_cpu(cpu) { 730be4d234dSJens Axboe struct bio_alloc_cache *cache; 731be4d234dSJens Axboe 732be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, cpu); 733be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 734be4d234dSJens Axboe } 735be4d234dSJens Axboe free_percpu(bs->cache); 736be4d234dSJens Axboe } 737be4d234dSJens Axboe 73829125ed6SChristoph Hellwig /** 739f9c78b2bSJens Axboe * bio_put - release a reference to a bio 740f9c78b2bSJens Axboe * @bio: bio to release reference to 741f9c78b2bSJens Axboe * 742f9c78b2bSJens Axboe * Description: 743f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 7449b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 745f9c78b2bSJens Axboe **/ 746f9c78b2bSJens Axboe void bio_put(struct bio *bio) 747f9c78b2bSJens Axboe { 748be4d234dSJens Axboe if (unlikely(bio_flagged(bio, BIO_REFFED))) { 7499e8c0d0dSChristoph Hellwig BUG_ON(!atomic_read(&bio->__bi_cnt)); 750be4d234dSJens Axboe if (!atomic_dec_and_test(&bio->__bi_cnt)) 751be4d234dSJens Axboe return; 752be4d234dSJens Axboe } 753f9c78b2bSJens Axboe 7540df71650SMike Snitzer if (bio->bi_opf & REQ_ALLOC_CACHE) { 755be4d234dSJens Axboe struct bio_alloc_cache *cache; 756be4d234dSJens Axboe 757be4d234dSJens Axboe bio_uninit(bio); 758be4d234dSJens Axboe cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 759fcade2ceSJens Axboe bio->bi_next = cache->free_list; 760fcade2ceSJens Axboe cache->free_list = bio; 761be4d234dSJens Axboe if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 762be4d234dSJens Axboe bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 763be4d234dSJens Axboe put_cpu(); 764be4d234dSJens Axboe } else { 765f9c78b2bSJens Axboe bio_free(bio); 766f9c78b2bSJens Axboe } 767dac56212SJens Axboe } 768f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 769f9c78b2bSJens Axboe 770a0e8de79SChristoph Hellwig static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp) 771f9c78b2bSJens Axboe { 772b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 773111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 774111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 775abfc426dSChristoph Hellwig if (bio->bi_bdev == bio_src->bi_bdev && 776abfc426dSChristoph Hellwig bio_flagged(bio_src, BIO_REMAPPED)) 77746bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 778ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 779f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 78020bd723eSPaolo Valente 781db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 782e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 78356b4b5abSChristoph Hellwig 78456b4b5abSChristoph Hellwig if (bio_crypt_clone(bio, bio_src, gfp) < 0) 78556b4b5abSChristoph Hellwig return -ENOMEM; 78656b4b5abSChristoph Hellwig if (bio_integrity(bio_src) && 78756b4b5abSChristoph Hellwig bio_integrity_clone(bio, bio_src, gfp) < 0) 78856b4b5abSChristoph Hellwig return -ENOMEM; 78956b4b5abSChristoph Hellwig return 0; 790f9c78b2bSJens Axboe } 791f9c78b2bSJens Axboe 792f9c78b2bSJens Axboe /** 793abfc426dSChristoph Hellwig * bio_alloc_clone - clone a bio that shares the original bio's biovec 794abfc426dSChristoph Hellwig * @bdev: block_device to clone onto 795a0e8de79SChristoph Hellwig * @bio_src: bio to clone from 796a0e8de79SChristoph Hellwig * @gfp: allocation priority 797f9c78b2bSJens Axboe * @bs: bio_set to allocate from 798f9c78b2bSJens Axboe * 799a0e8de79SChristoph Hellwig * Allocate a new bio that is a clone of @bio_src. The caller owns the returned 800a0e8de79SChristoph Hellwig * bio, but not the actual data it points to. 801a0e8de79SChristoph Hellwig * 802a0e8de79SChristoph Hellwig * The caller must ensure that the return bio is not freed before @bio_src. 803f9c78b2bSJens Axboe */ 804abfc426dSChristoph Hellwig struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src, 805abfc426dSChristoph Hellwig gfp_t gfp, struct bio_set *bs) 806f9c78b2bSJens Axboe { 807a0e8de79SChristoph Hellwig struct bio *bio; 808f9c78b2bSJens Axboe 809abfc426dSChristoph Hellwig bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs); 810a0e8de79SChristoph Hellwig if (!bio) 811f9c78b2bSJens Axboe return NULL; 812f9c78b2bSJens Axboe 813a0e8de79SChristoph Hellwig if (__bio_clone(bio, bio_src, gfp) < 0) { 814a0e8de79SChristoph Hellwig bio_put(bio); 81507560151SEric Biggers return NULL; 816f9c78b2bSJens Axboe } 817a0e8de79SChristoph Hellwig bio->bi_io_vec = bio_src->bi_io_vec; 81856b4b5abSChristoph Hellwig 819a0e8de79SChristoph Hellwig return bio; 82056b4b5abSChristoph Hellwig } 821abfc426dSChristoph Hellwig EXPORT_SYMBOL(bio_alloc_clone); 822f9c78b2bSJens Axboe 823a0e8de79SChristoph Hellwig /** 824abfc426dSChristoph Hellwig * bio_init_clone - clone a bio that shares the original bio's biovec 825abfc426dSChristoph Hellwig * @bdev: block_device to clone onto 826a0e8de79SChristoph Hellwig * @bio: bio to clone into 827a0e8de79SChristoph Hellwig * @bio_src: bio to clone from 828a0e8de79SChristoph Hellwig * @gfp: allocation priority 829a0e8de79SChristoph Hellwig * 830a0e8de79SChristoph Hellwig * Initialize a new bio in caller provided memory that is a clone of @bio_src. 831a0e8de79SChristoph Hellwig * The caller owns the returned bio, but not the actual data it points to. 832a0e8de79SChristoph Hellwig * 833a0e8de79SChristoph Hellwig * The caller must ensure that @bio_src is not freed before @bio. 834a0e8de79SChristoph Hellwig */ 835abfc426dSChristoph Hellwig int bio_init_clone(struct block_device *bdev, struct bio *bio, 836abfc426dSChristoph Hellwig struct bio *bio_src, gfp_t gfp) 837a0e8de79SChristoph Hellwig { 838a0e8de79SChristoph Hellwig int ret; 839a0e8de79SChristoph Hellwig 840abfc426dSChristoph Hellwig bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf); 841a0e8de79SChristoph Hellwig ret = __bio_clone(bio, bio_src, gfp); 842a0e8de79SChristoph Hellwig if (ret) 843a0e8de79SChristoph Hellwig bio_uninit(bio); 844a0e8de79SChristoph Hellwig return ret; 845a0e8de79SChristoph Hellwig } 846abfc426dSChristoph Hellwig EXPORT_SYMBOL(bio_init_clone); 847a0e8de79SChristoph Hellwig 8489a6083beSChristoph Hellwig /** 8499a6083beSChristoph Hellwig * bio_full - check if the bio is full 8509a6083beSChristoph Hellwig * @bio: bio to check 8519a6083beSChristoph Hellwig * @len: length of one segment to be added 8529a6083beSChristoph Hellwig * 8539a6083beSChristoph Hellwig * Return true if @bio is full and one segment with @len bytes can't be 8549a6083beSChristoph Hellwig * added to the bio, otherwise return false 8559a6083beSChristoph Hellwig */ 8569a6083beSChristoph Hellwig static inline bool bio_full(struct bio *bio, unsigned len) 8579a6083beSChristoph Hellwig { 8589a6083beSChristoph Hellwig if (bio->bi_vcnt >= bio->bi_max_vecs) 8599a6083beSChristoph Hellwig return true; 8609a6083beSChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) 8619a6083beSChristoph Hellwig return true; 8629a6083beSChristoph Hellwig return false; 8639a6083beSChristoph Hellwig } 8649a6083beSChristoph Hellwig 8655919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 8665919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 867ff896738SChristoph Hellwig bool *same_page) 8685919482eSMing Lei { 869d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 870d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 8715919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 8725919482eSMing Lei 8735919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 8745919482eSMing Lei return false; 8755919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 8765919482eSMing Lei return false; 87752d52d1cSChristoph Hellwig 878ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 879d8166519SMatthew Wilcox (Oracle) if (*same_page) 8805919482eSMing Lei return true; 881d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 8825919482eSMing Lei } 8835919482eSMing Lei 8849774b391SChristoph Hellwig /** 8859774b391SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 8869774b391SChristoph Hellwig * @bio: destination bio 8879774b391SChristoph Hellwig * @page: start page to add 8889774b391SChristoph Hellwig * @len: length of the data to add 8899774b391SChristoph Hellwig * @off: offset of the data relative to @page 8909774b391SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 8919774b391SChristoph Hellwig * 8929774b391SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 8939774b391SChristoph Hellwig * useful optimisation for file systems with a block size smaller than the 8949774b391SChristoph Hellwig * page size. 8959774b391SChristoph Hellwig * 8969774b391SChristoph Hellwig * Warn if (@len, @off) crosses pages in case that @same_page is true. 8979774b391SChristoph Hellwig * 8989774b391SChristoph Hellwig * Return %true on success or %false on failure. 8999774b391SChristoph Hellwig */ 9009774b391SChristoph Hellwig static bool __bio_try_merge_page(struct bio *bio, struct page *page, 9019774b391SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 9029774b391SChristoph Hellwig { 9039774b391SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 9049774b391SChristoph Hellwig return false; 9059774b391SChristoph Hellwig 9069774b391SChristoph Hellwig if (bio->bi_vcnt > 0) { 9079774b391SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 9089774b391SChristoph Hellwig 9099774b391SChristoph Hellwig if (page_is_mergeable(bv, page, len, off, same_page)) { 9109774b391SChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) { 9119774b391SChristoph Hellwig *same_page = false; 9129774b391SChristoph Hellwig return false; 9139774b391SChristoph Hellwig } 9149774b391SChristoph Hellwig bv->bv_len += len; 9159774b391SChristoph Hellwig bio->bi_iter.bi_size += len; 9169774b391SChristoph Hellwig return true; 9179774b391SChristoph Hellwig } 9189774b391SChristoph Hellwig } 9199774b391SChristoph Hellwig return false; 9209774b391SChristoph Hellwig } 9219774b391SChristoph Hellwig 922e4581105SChristoph Hellwig /* 923e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 924e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 925e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 926e4581105SChristoph Hellwig */ 927e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 928e4581105SChristoph Hellwig struct page *page, unsigned len, 929e4581105SChristoph Hellwig unsigned offset, bool *same_page) 930489fbbcbSMing Lei { 931384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 932489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 933489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 934489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 935489fbbcbSMing Lei 936489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 937489fbbcbSMing Lei return false; 938489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 939489fbbcbSMing Lei return false; 940384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 941489fbbcbSMing Lei } 942489fbbcbSMing Lei 943f4595875SShaohua Li /** 944e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 945c66a14d0SKent Overstreet * @q: the target queue 946c66a14d0SKent Overstreet * @bio: destination bio 947c66a14d0SKent Overstreet * @page: page to add 948c66a14d0SKent Overstreet * @len: vec entry length 949c66a14d0SKent Overstreet * @offset: vec entry offset 950e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 951e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 952f9c78b2bSJens Axboe * 953e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 954e4581105SChristoph Hellwig * and gap limitations. 955f9c78b2bSJens Axboe */ 956e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 95719047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 958e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 959f9c78b2bSJens Axboe { 960f9c78b2bSJens Axboe struct bio_vec *bvec; 961f9c78b2bSJens Axboe 962e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 963f9c78b2bSJens Axboe return 0; 964f9c78b2bSJens Axboe 965e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 966f9c78b2bSJens Axboe return 0; 967f9c78b2bSJens Axboe 968f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 969e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 970384209cdSChristoph Hellwig return len; 971320ea869SChristoph Hellwig 972320ea869SChristoph Hellwig /* 973320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 974320ea869SChristoph Hellwig * would create a gap, disallow it. 975320ea869SChristoph Hellwig */ 976384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 977320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 978320ea869SChristoph Hellwig return 0; 979f9c78b2bSJens Axboe } 980f9c78b2bSJens Axboe 98179d08f89SMing Lei if (bio_full(bio, len)) 982f9c78b2bSJens Axboe return 0; 983f9c78b2bSJens Axboe 98414ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 985489fbbcbSMing Lei return 0; 986489fbbcbSMing Lei 987f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 988f9c78b2bSJens Axboe bvec->bv_page = page; 989f9c78b2bSJens Axboe bvec->bv_len = len; 990f9c78b2bSJens Axboe bvec->bv_offset = offset; 991fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 992dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 993f9c78b2bSJens Axboe return len; 994f9c78b2bSJens Axboe } 99519047087SMing Lei 996e4581105SChristoph Hellwig /** 997e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 998e4581105SChristoph Hellwig * @q: the target queue 999e4581105SChristoph Hellwig * @bio: destination bio 1000e4581105SChristoph Hellwig * @page: page to add 1001e4581105SChristoph Hellwig * @len: vec entry length 1002e4581105SChristoph Hellwig * @offset: vec entry offset 1003e4581105SChristoph Hellwig * 1004e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 1005e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 1006e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 1007e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 1008e4581105SChristoph Hellwig * 1009e4581105SChristoph Hellwig * This should only be used by passthrough bios. 1010e4581105SChristoph Hellwig */ 101119047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 101219047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 101319047087SMing Lei { 1014d1916c86SChristoph Hellwig bool same_page = false; 1015e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 1016e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 101719047087SMing Lei } 1018f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 1019f9c78b2bSJens Axboe 1020f9c78b2bSJens Axboe /** 1021ae29333fSJohannes Thumshirn * bio_add_zone_append_page - attempt to add page to zone-append bio 1022ae29333fSJohannes Thumshirn * @bio: destination bio 1023ae29333fSJohannes Thumshirn * @page: page to add 1024ae29333fSJohannes Thumshirn * @len: vec entry length 1025ae29333fSJohannes Thumshirn * @offset: vec entry offset 1026ae29333fSJohannes Thumshirn * 1027ae29333fSJohannes Thumshirn * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 1028ae29333fSJohannes Thumshirn * for a zone-append request. This can fail for a number of reasons, such as the 1029ae29333fSJohannes Thumshirn * bio being full or the target block device is not a zoned block device or 1030ae29333fSJohannes Thumshirn * other limitations of the target block device. The target block device must 1031ae29333fSJohannes Thumshirn * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 1032ae29333fSJohannes Thumshirn * to an empty bio. 1033ae29333fSJohannes Thumshirn * 1034ae29333fSJohannes Thumshirn * Returns: number of bytes added to the bio, or 0 in case of a failure. 1035ae29333fSJohannes Thumshirn */ 1036ae29333fSJohannes Thumshirn int bio_add_zone_append_page(struct bio *bio, struct page *page, 1037ae29333fSJohannes Thumshirn unsigned int len, unsigned int offset) 1038ae29333fSJohannes Thumshirn { 10393caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1040ae29333fSJohannes Thumshirn bool same_page = false; 1041ae29333fSJohannes Thumshirn 1042ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 1043ae29333fSJohannes Thumshirn return 0; 1044ae29333fSJohannes Thumshirn 1045ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 1046ae29333fSJohannes Thumshirn return 0; 1047ae29333fSJohannes Thumshirn 1048ae29333fSJohannes Thumshirn return bio_add_hw_page(q, bio, page, len, offset, 1049ae29333fSJohannes Thumshirn queue_max_zone_append_sectors(q), &same_page); 1050ae29333fSJohannes Thumshirn } 1051ae29333fSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 1052ae29333fSJohannes Thumshirn 1053ae29333fSJohannes Thumshirn /** 1054551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 10550aa69fd3SChristoph Hellwig * @bio: destination bio 1056551879a4SMing Lei * @page: start page to add 1057551879a4SMing Lei * @len: length of the data to add, may cross pages 1058551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 10590aa69fd3SChristoph Hellwig * 10600aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 10610aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 10620aa69fd3SChristoph Hellwig */ 10630aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 10640aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 10650aa69fd3SChristoph Hellwig { 10660aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 10670aa69fd3SChristoph Hellwig 10680aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 106979d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 10700aa69fd3SChristoph Hellwig 10710aa69fd3SChristoph Hellwig bv->bv_page = page; 10720aa69fd3SChristoph Hellwig bv->bv_offset = off; 10730aa69fd3SChristoph Hellwig bv->bv_len = len; 10740aa69fd3SChristoph Hellwig 10750aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 10760aa69fd3SChristoph Hellwig bio->bi_vcnt++; 1077b8e24a93SJohannes Weiner 1078b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 1079b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 10800aa69fd3SChristoph Hellwig } 10810aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 10820aa69fd3SChristoph Hellwig 10830aa69fd3SChristoph Hellwig /** 1084551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 1085f9c78b2bSJens Axboe * @bio: destination bio 1086551879a4SMing Lei * @page: start page to add 1087551879a4SMing Lei * @len: vec entry length, may cross pages 1088551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 1089f9c78b2bSJens Axboe * 1090551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 1091c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1092f9c78b2bSJens Axboe */ 1093c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 1094c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 1095f9c78b2bSJens Axboe { 1096ff896738SChristoph Hellwig bool same_page = false; 1097ff896738SChristoph Hellwig 1098ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 109979d08f89SMing Lei if (bio_full(bio, len)) 1100c66a14d0SKent Overstreet return 0; 11010aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 1102c66a14d0SKent Overstreet } 1103c66a14d0SKent Overstreet return len; 1104f9c78b2bSJens Axboe } 1105f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 1106f9c78b2bSJens Axboe 110785f5a74cSMatthew Wilcox (Oracle) /** 110885f5a74cSMatthew Wilcox (Oracle) * bio_add_folio - Attempt to add part of a folio to a bio. 110985f5a74cSMatthew Wilcox (Oracle) * @bio: BIO to add to. 111085f5a74cSMatthew Wilcox (Oracle) * @folio: Folio to add. 111185f5a74cSMatthew Wilcox (Oracle) * @len: How many bytes from the folio to add. 111285f5a74cSMatthew Wilcox (Oracle) * @off: First byte in this folio to add. 111385f5a74cSMatthew Wilcox (Oracle) * 111485f5a74cSMatthew Wilcox (Oracle) * Filesystems that use folios can call this function instead of calling 111585f5a74cSMatthew Wilcox (Oracle) * bio_add_page() for each page in the folio. If @off is bigger than 111685f5a74cSMatthew Wilcox (Oracle) * PAGE_SIZE, this function can create a bio_vec that starts in a page 111785f5a74cSMatthew Wilcox (Oracle) * after the bv_page. BIOs do not support folios that are 4GiB or larger. 111885f5a74cSMatthew Wilcox (Oracle) * 111985f5a74cSMatthew Wilcox (Oracle) * Return: Whether the addition was successful. 112085f5a74cSMatthew Wilcox (Oracle) */ 112185f5a74cSMatthew Wilcox (Oracle) bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, 112285f5a74cSMatthew Wilcox (Oracle) size_t off) 112385f5a74cSMatthew Wilcox (Oracle) { 112485f5a74cSMatthew Wilcox (Oracle) if (len > UINT_MAX || off > UINT_MAX) 1125455a844dSJiapeng Chong return false; 112685f5a74cSMatthew Wilcox (Oracle) return bio_add_page(bio, &folio->page, len, off) > 0; 112785f5a74cSMatthew Wilcox (Oracle) } 112885f5a74cSMatthew Wilcox (Oracle) 1129c809084aSPavel Begunkov void __bio_release_pages(struct bio *bio, bool mark_dirty) 11307321ecbfSChristoph Hellwig { 11317321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 11327321ecbfSChristoph Hellwig struct bio_vec *bvec; 11337321ecbfSChristoph Hellwig 1134d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1135d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 1136d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 11377321ecbfSChristoph Hellwig put_page(bvec->bv_page); 11387321ecbfSChristoph Hellwig } 1139d241a95fSChristoph Hellwig } 1140c809084aSPavel Begunkov EXPORT_SYMBOL_GPL(__bio_release_pages); 11417321ecbfSChristoph Hellwig 11421bb6b810SPavel Begunkov void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 11436d0c48aeSJens Axboe { 1144fa5fa8ecSPavel Begunkov size_t size = iov_iter_count(iter); 1145fa5fa8ecSPavel Begunkov 11467a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_max_vecs); 11476d0c48aeSJens Axboe 1148fa5fa8ecSPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1149fa5fa8ecSPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1150fa5fa8ecSPavel Begunkov size_t max_sectors = queue_max_zone_append_sectors(q); 1151fa5fa8ecSPavel Begunkov 1152fa5fa8ecSPavel Begunkov size = min(size, max_sectors << SECTOR_SHIFT); 1153fa5fa8ecSPavel Begunkov } 1154fa5fa8ecSPavel Begunkov 1155c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 1156c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1157c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 1158fa5fa8ecSPavel Begunkov bio->bi_iter.bi_size = size; 1159ed97ce5eSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 1160977be012SChristoph Hellwig bio_set_flag(bio, BIO_CLONED); 11617de55b7dSJohannes Thumshirn } 11626d0c48aeSJens Axboe 1163d9cf3bd5SPavel Begunkov static void bio_put_pages(struct page **pages, size_t size, size_t off) 1164d9cf3bd5SPavel Begunkov { 1165d9cf3bd5SPavel Begunkov size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE); 1166d9cf3bd5SPavel Begunkov 1167d9cf3bd5SPavel Begunkov for (i = 0; i < nr; i++) 1168d9cf3bd5SPavel Begunkov put_page(pages[i]); 1169d9cf3bd5SPavel Begunkov } 1170d9cf3bd5SPavel Begunkov 1171576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1172576ed913SChristoph Hellwig 11732cefe4dbSKent Overstreet /** 117417d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 11752cefe4dbSKent Overstreet * @bio: bio to add pages to 11762cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 11772cefe4dbSKent Overstreet * 117817d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 11792cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 118017d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 11813cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 11822cefe4dbSKent Overstreet */ 118317d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 11842cefe4dbSKent Overstreet { 1185576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1186576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 11872cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 11882cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 118945691804SChristoph Hellwig bool same_page = false; 1190576ed913SChristoph Hellwig ssize_t size, left; 1191576ed913SChristoph Hellwig unsigned len, i; 1192b403ea24SMartin Wilck size_t offset; 1193576ed913SChristoph Hellwig 1194576ed913SChristoph Hellwig /* 1195576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 1196576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 1197576ed913SChristoph Hellwig * without overwriting the temporary page array. 1198576ed913SChristoph Hellwig */ 1199576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1200576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 12012cefe4dbSKent Overstreet 120235c820e7SJens Axboe size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 12032cefe4dbSKent Overstreet if (unlikely(size <= 0)) 12042cefe4dbSKent Overstreet return size ? size : -EFAULT; 12052cefe4dbSKent Overstreet 1206576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1207576ed913SChristoph Hellwig struct page *page = pages[i]; 12082cefe4dbSKent Overstreet 1209576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 121045691804SChristoph Hellwig 121145691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 121245691804SChristoph Hellwig if (same_page) 121345691804SChristoph Hellwig put_page(page); 121445691804SChristoph Hellwig } else { 1215d9cf3bd5SPavel Begunkov if (WARN_ON_ONCE(bio_full(bio, len))) { 1216d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 1217576ed913SChristoph Hellwig return -EINVAL; 1218d9cf3bd5SPavel Begunkov } 121945691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 122045691804SChristoph Hellwig } 1221576ed913SChristoph Hellwig offset = 0; 12222cefe4dbSKent Overstreet } 12232cefe4dbSKent Overstreet 12242cefe4dbSKent Overstreet iov_iter_advance(iter, size); 12252cefe4dbSKent Overstreet return 0; 12262cefe4dbSKent Overstreet } 122717d51b10SMartin Wilck 12280512a75bSKeith Busch static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 12290512a75bSKeith Busch { 12300512a75bSKeith Busch unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 12310512a75bSKeith Busch unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 12323caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 12330512a75bSKeith Busch unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 12340512a75bSKeith Busch struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 12350512a75bSKeith Busch struct page **pages = (struct page **)bv; 12360512a75bSKeith Busch ssize_t size, left; 12370512a75bSKeith Busch unsigned len, i; 12380512a75bSKeith Busch size_t offset; 12394977d121SNaohiro Aota int ret = 0; 12400512a75bSKeith Busch 12410512a75bSKeith Busch if (WARN_ON_ONCE(!max_append_sectors)) 12420512a75bSKeith Busch return 0; 12430512a75bSKeith Busch 12440512a75bSKeith Busch /* 12450512a75bSKeith Busch * Move page array up in the allocated memory for the bio vecs as far as 12460512a75bSKeith Busch * possible so that we can start filling biovecs from the beginning 12470512a75bSKeith Busch * without overwriting the temporary page array. 12480512a75bSKeith Busch */ 12490512a75bSKeith Busch BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 12500512a75bSKeith Busch pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 12510512a75bSKeith Busch 12520512a75bSKeith Busch size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 12530512a75bSKeith Busch if (unlikely(size <= 0)) 12540512a75bSKeith Busch return size ? size : -EFAULT; 12550512a75bSKeith Busch 12560512a75bSKeith Busch for (left = size, i = 0; left > 0; left -= len, i++) { 12570512a75bSKeith Busch struct page *page = pages[i]; 12580512a75bSKeith Busch bool same_page = false; 12590512a75bSKeith Busch 12600512a75bSKeith Busch len = min_t(size_t, PAGE_SIZE - offset, left); 12610512a75bSKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 12624977d121SNaohiro Aota max_append_sectors, &same_page) != len) { 1263d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 12644977d121SNaohiro Aota ret = -EINVAL; 12654977d121SNaohiro Aota break; 12664977d121SNaohiro Aota } 12670512a75bSKeith Busch if (same_page) 12680512a75bSKeith Busch put_page(page); 12690512a75bSKeith Busch offset = 0; 12700512a75bSKeith Busch } 12710512a75bSKeith Busch 12724977d121SNaohiro Aota iov_iter_advance(iter, size - left); 12734977d121SNaohiro Aota return ret; 12740512a75bSKeith Busch } 12750512a75bSKeith Busch 127617d51b10SMartin Wilck /** 12776d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 127817d51b10SMartin Wilck * @bio: bio to add pages to 12796d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 128017d51b10SMartin Wilck * 12816d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 12826d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 12836d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1284c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1285c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1286c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1287c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1288c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1289c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 12906d0c48aeSJens Axboe * 129117d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 12925cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 12936d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 12946d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 12950cf41e5eSPavel Begunkov * 12960cf41e5eSPavel Begunkov * It's intended for direct IO, so doesn't do PSI tracking, the caller is 12970cf41e5eSPavel Begunkov * responsible for setting BIO_WORKINGSET if necessary. 129817d51b10SMartin Wilck */ 129917d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 130017d51b10SMartin Wilck { 1301c42bca92SPavel Begunkov int ret = 0; 130214eacf12SChristoph Hellwig 1303c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1304fa5fa8ecSPavel Begunkov bio_iov_bvec_set(bio, iter); 1305fa5fa8ecSPavel Begunkov iov_iter_advance(iter, bio->bi_iter.bi_size); 1306fa5fa8ecSPavel Begunkov return 0; 130786004515SChristoph Hellwig } 130817d51b10SMartin Wilck 130917d51b10SMartin Wilck do { 1310c42bca92SPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) 13110512a75bSKeith Busch ret = __bio_iov_append_get_pages(bio, iter); 13126d0c48aeSJens Axboe else 13136d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 131479d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 131517d51b10SMartin Wilck 13160cf41e5eSPavel Begunkov /* don't account direct I/O as memory stall */ 13170cf41e5eSPavel Begunkov bio_clear_flag(bio, BIO_WORKINGSET); 131814eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 131917d51b10SMartin Wilck } 132029b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 13212cefe4dbSKent Overstreet 13224246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1323f9c78b2bSJens Axboe { 132465e53aabSChristoph Hellwig complete(bio->bi_private); 1325f9c78b2bSJens Axboe } 1326f9c78b2bSJens Axboe 1327f9c78b2bSJens Axboe /** 1328f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1329f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1330f9c78b2bSJens Axboe * 1331f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1332f9c78b2bSJens Axboe * bio_endio() on failure. 13333d289d68SJan Kara * 13343d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 13353d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 13363d289d68SJan Kara * on his own. 1337f9c78b2bSJens Axboe */ 13384e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1339f9c78b2bSJens Axboe { 1340309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1341309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1342de6a78b6SMing Lei unsigned long hang_check; 1343f9c78b2bSJens Axboe 134465e53aabSChristoph Hellwig bio->bi_private = &done; 1345f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 13461eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 13474e49ea4aSMike Christie submit_bio(bio); 1348de6a78b6SMing Lei 1349de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1350de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1351de6a78b6SMing Lei if (hang_check) 1352de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1353de6a78b6SMing Lei hang_check * (HZ/2))) 1354de6a78b6SMing Lei ; 1355de6a78b6SMing Lei else 135665e53aabSChristoph Hellwig wait_for_completion_io(&done); 1357f9c78b2bSJens Axboe 135865e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1359f9c78b2bSJens Axboe } 1360f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1361f9c78b2bSJens Axboe 1362d4aa57a1SJens Axboe void __bio_advance(struct bio *bio, unsigned bytes) 1363f9c78b2bSJens Axboe { 1364f9c78b2bSJens Axboe if (bio_integrity(bio)) 1365f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1366f9c78b2bSJens Axboe 1367a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1368f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1369f9c78b2bSJens Axboe } 1370d4aa57a1SJens Axboe EXPORT_SYMBOL(__bio_advance); 1371f9c78b2bSJens Axboe 137245db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 137345db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1374f9c78b2bSJens Axboe { 137545db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 1376f8b679a0SChristoph Hellwig struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1377f8b679a0SChristoph Hellwig struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1378f8b679a0SChristoph Hellwig unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1379f8b679a0SChristoph Hellwig void *src_buf; 138045db54d5SKent Overstreet 1381f8b679a0SChristoph Hellwig src_buf = bvec_kmap_local(&src_bv); 1382f8b679a0SChristoph Hellwig memcpy_to_bvec(&dst_bv, src_buf); 1383f8b679a0SChristoph Hellwig kunmap_local(src_buf); 13846e6e811dSKent Overstreet 138522b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 138622b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 138745db54d5SKent Overstreet } 138845db54d5SKent Overstreet } 138945db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 139045db54d5SKent Overstreet 139145db54d5SKent Overstreet /** 139245db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 139345db54d5SKent Overstreet * @src: source bio 139445db54d5SKent Overstreet * @dst: destination bio 139545db54d5SKent Overstreet * 139645db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 139745db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 139845db54d5SKent Overstreet */ 139945db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 140045db54d5SKent Overstreet { 140145db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 140245db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 140345db54d5SKent Overstreet 140445db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 140545db54d5SKent Overstreet } 140645db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 140745db54d5SKent Overstreet 1408491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 14091dfa0f68SChristoph Hellwig { 14101dfa0f68SChristoph Hellwig struct bio_vec *bvec; 14116dc4f100SMing Lei struct bvec_iter_all iter_all; 14121dfa0f68SChristoph Hellwig 14132b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 14141dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 14151dfa0f68SChristoph Hellwig } 1416491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 14171dfa0f68SChristoph Hellwig 1418f9c78b2bSJens Axboe /* 1419f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1420f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1421f9c78b2bSJens Axboe * 1422f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1423f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1424f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1425f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1426f9c78b2bSJens Axboe * in process context. 1427f9c78b2bSJens Axboe * 1428f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1429f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1430f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1431f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1432f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1433f9c78b2bSJens Axboe * 1434f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1435f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1436f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1437f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1438f9c78b2bSJens Axboe * pagecache. 1439f9c78b2bSJens Axboe * 1440f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1441f9c78b2bSJens Axboe * deferred bio dirtying paths. 1442f9c78b2bSJens Axboe */ 1443f9c78b2bSJens Axboe 1444f9c78b2bSJens Axboe /* 1445f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1446f9c78b2bSJens Axboe */ 1447f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1448f9c78b2bSJens Axboe { 1449f9c78b2bSJens Axboe struct bio_vec *bvec; 14506dc4f100SMing Lei struct bvec_iter_all iter_all; 1451f9c78b2bSJens Axboe 14522b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 14533bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 14543bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1455f9c78b2bSJens Axboe } 1456f9c78b2bSJens Axboe } 1457f9c78b2bSJens Axboe 1458f9c78b2bSJens Axboe /* 1459f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1460f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1461f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 146224d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1463f9c78b2bSJens Axboe * 1464f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1465ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1466ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1467f9c78b2bSJens Axboe */ 1468f9c78b2bSJens Axboe 1469f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1470f9c78b2bSJens Axboe 1471f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1472f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1473f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1474f9c78b2bSJens Axboe 1475f9c78b2bSJens Axboe /* 1476f9c78b2bSJens Axboe * This runs in process context 1477f9c78b2bSJens Axboe */ 1478f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1479f9c78b2bSJens Axboe { 148024d5493fSChristoph Hellwig struct bio *bio, *next; 1481f9c78b2bSJens Axboe 148224d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 148324d5493fSChristoph Hellwig next = bio_dirty_list; 1484f9c78b2bSJens Axboe bio_dirty_list = NULL; 148524d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1486f9c78b2bSJens Axboe 148724d5493fSChristoph Hellwig while ((bio = next) != NULL) { 148824d5493fSChristoph Hellwig next = bio->bi_private; 1489f9c78b2bSJens Axboe 1490d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1491f9c78b2bSJens Axboe bio_put(bio); 1492f9c78b2bSJens Axboe } 1493f9c78b2bSJens Axboe } 1494f9c78b2bSJens Axboe 1495f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1496f9c78b2bSJens Axboe { 1497f9c78b2bSJens Axboe struct bio_vec *bvec; 149824d5493fSChristoph Hellwig unsigned long flags; 14996dc4f100SMing Lei struct bvec_iter_all iter_all; 1500f9c78b2bSJens Axboe 15012b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 150224d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 150324d5493fSChristoph Hellwig goto defer; 1504f9c78b2bSJens Axboe } 1505f9c78b2bSJens Axboe 1506d241a95fSChristoph Hellwig bio_release_pages(bio, false); 150724d5493fSChristoph Hellwig bio_put(bio); 150824d5493fSChristoph Hellwig return; 150924d5493fSChristoph Hellwig defer: 1510f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1511f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1512f9c78b2bSJens Axboe bio_dirty_list = bio; 1513f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1514f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1515f9c78b2bSJens Axboe } 1516f9c78b2bSJens Axboe 1517c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1518c4cf5261SJens Axboe { 1519c4cf5261SJens Axboe /* 1520c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1521c4cf5261SJens Axboe * we always end io on the first invocation. 1522c4cf5261SJens Axboe */ 1523c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1524c4cf5261SJens Axboe return true; 1525c4cf5261SJens Axboe 1526c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1527c4cf5261SJens Axboe 1528326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1529b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1530c4cf5261SJens Axboe return true; 1531326e1dbbSMike Snitzer } 1532c4cf5261SJens Axboe 1533c4cf5261SJens Axboe return false; 1534c4cf5261SJens Axboe } 1535c4cf5261SJens Axboe 1536f9c78b2bSJens Axboe /** 1537f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1538f9c78b2bSJens Axboe * @bio: bio 1539f9c78b2bSJens Axboe * 1540f9c78b2bSJens Axboe * Description: 15414246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 15424246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 15434246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1544fbbaf700SNeilBrown * 1545fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1546fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 154760b6a7e6SEdward Hsieh * last time. 1548f9c78b2bSJens Axboe **/ 15494246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1550f9c78b2bSJens Axboe { 1551ba8c6967SChristoph Hellwig again: 15522b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1553ba8c6967SChristoph Hellwig return; 15547c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 15557c20f116SChristoph Hellwig return; 1556f9c78b2bSJens Axboe 1557aa1b46dcSTejun Heo rq_qos_done_bio(bio); 155867b42d0bSJosef Bacik 155960b6a7e6SEdward Hsieh if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 15603caee463SPavel Begunkov trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 156160b6a7e6SEdward Hsieh bio_clear_flag(bio, BIO_TRACE_COMPLETION); 156260b6a7e6SEdward Hsieh } 156360b6a7e6SEdward Hsieh 1564f9c78b2bSJens Axboe /* 1565ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1566ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1567ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1568ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1569ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1570ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1571f9c78b2bSJens Axboe */ 1572f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 157338f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1574ba8c6967SChristoph Hellwig goto again; 1575ba8c6967SChristoph Hellwig } 1576ba8c6967SChristoph Hellwig 15779e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1578b222dd2fSShaohua Li /* release cgroup info */ 1579b222dd2fSShaohua Li bio_uninit(bio); 1580f9c78b2bSJens Axboe if (bio->bi_end_io) 15814246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1582f9c78b2bSJens Axboe } 1583f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1584f9c78b2bSJens Axboe 1585f9c78b2bSJens Axboe /** 1586f9c78b2bSJens Axboe * bio_split - split a bio 1587f9c78b2bSJens Axboe * @bio: bio to split 1588f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1589f9c78b2bSJens Axboe * @gfp: gfp mask 1590f9c78b2bSJens Axboe * @bs: bio set to allocate from 1591f9c78b2bSJens Axboe * 1592f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1593f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1594f9c78b2bSJens Axboe * 1595f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1596dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1597dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1598f9c78b2bSJens Axboe */ 1599f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1600f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1601f9c78b2bSJens Axboe { 1602f341a4d3SMikulas Patocka struct bio *split; 1603f9c78b2bSJens Axboe 1604f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1605f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1606f9c78b2bSJens Axboe 16070512a75bSKeith Busch /* Zone append commands cannot be split */ 16080512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 16090512a75bSKeith Busch return NULL; 16100512a75bSKeith Busch 1611abfc426dSChristoph Hellwig split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs); 1612f9c78b2bSJens Axboe if (!split) 1613f9c78b2bSJens Axboe return NULL; 1614f9c78b2bSJens Axboe 1615f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1616f9c78b2bSJens Axboe 1617f9c78b2bSJens Axboe if (bio_integrity(split)) 1618fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1619f9c78b2bSJens Axboe 1620f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1621f9c78b2bSJens Axboe 1622fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 162320d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1624fbbaf700SNeilBrown 1625f9c78b2bSJens Axboe return split; 1626f9c78b2bSJens Axboe } 1627f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1628f9c78b2bSJens Axboe 1629f9c78b2bSJens Axboe /** 1630f9c78b2bSJens Axboe * bio_trim - trim a bio 1631f9c78b2bSJens Axboe * @bio: bio to trim 1632f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1633f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1634e83502caSChaitanya Kulkarni * 1635e83502caSChaitanya Kulkarni * This function is typically used for bios that are cloned and submitted 1636e83502caSChaitanya Kulkarni * to the underlying device in parts. 1637f9c78b2bSJens Axboe */ 1638e83502caSChaitanya Kulkarni void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1639f9c78b2bSJens Axboe { 1640e83502caSChaitanya Kulkarni if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 16418535c018SMing Lei offset + size > bio_sectors(bio))) 1642e83502caSChaitanya Kulkarni return; 1643f9c78b2bSJens Axboe 1644f9c78b2bSJens Axboe size <<= 9; 1645f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1646f9c78b2bSJens Axboe return; 1647f9c78b2bSJens Axboe 1648f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1649f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1650376a78abSDmitry Monakhov 1651376a78abSDmitry Monakhov if (bio_integrity(bio)) 1652fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1653f9c78b2bSJens Axboe } 1654f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1655f9c78b2bSJens Axboe 1656f9c78b2bSJens Axboe /* 1657f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1658f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1659f9c78b2bSJens Axboe */ 16608aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1661f9c78b2bSJens Axboe { 16627a800a20SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1663f9c78b2bSJens Axboe 16648aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1665f9c78b2bSJens Axboe } 1666f9c78b2bSJens Axboe 1667917a38c7SKent Overstreet /* 1668917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1669917a38c7SKent Overstreet * 1670917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1671917a38c7SKent Overstreet * kzalloc()). 1672917a38c7SKent Overstreet */ 1673917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1674f9c78b2bSJens Axboe { 1675be4d234dSJens Axboe bio_alloc_cache_destroy(bs); 1676f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1677f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1678917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1679f9c78b2bSJens Axboe 16808aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 16818aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1682f9c78b2bSJens Axboe 1683f9c78b2bSJens Axboe bioset_integrity_free(bs); 1684917a38c7SKent Overstreet if (bs->bio_slab) 1685f9c78b2bSJens Axboe bio_put_slab(bs); 1686917a38c7SKent Overstreet bs->bio_slab = NULL; 1687917a38c7SKent Overstreet } 1688917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1689f9c78b2bSJens Axboe 1690011067b0SNeilBrown /** 1691917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1692dad08527SKent Overstreet * @bs: pool to initialize 1693917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1694917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1695917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1696917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1697917a38c7SKent Overstreet * 1698dad08527SKent Overstreet * Description: 1699dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1700dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1701dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1702dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1703dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1704dad08527SKent Overstreet * or things will break badly. 1705dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1706abfc426dSChristoph Hellwig * for allocating iovecs. This pool is not needed e.g. for bio_init_clone(). 1707abfc426dSChristoph Hellwig * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used 1708abfc426dSChristoph Hellwig * to dispatch queued requests when the mempool runs out of space. 1709dad08527SKent Overstreet * 1710917a38c7SKent Overstreet */ 1711917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1712917a38c7SKent Overstreet unsigned int pool_size, 1713917a38c7SKent Overstreet unsigned int front_pad, 1714917a38c7SKent Overstreet int flags) 1715917a38c7SKent Overstreet { 1716917a38c7SKent Overstreet bs->front_pad = front_pad; 17179f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 17189f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 17199f180e31SMing Lei else 17209f180e31SMing Lei bs->back_pad = 0; 1721917a38c7SKent Overstreet 1722917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1723917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1724917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1725917a38c7SKent Overstreet 172649d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1727917a38c7SKent Overstreet if (!bs->bio_slab) 1728917a38c7SKent Overstreet return -ENOMEM; 1729917a38c7SKent Overstreet 1730917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1731917a38c7SKent Overstreet goto bad; 1732917a38c7SKent Overstreet 1733917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1734917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1735917a38c7SKent Overstreet goto bad; 1736917a38c7SKent Overstreet 1737be4d234dSJens Axboe if (flags & BIOSET_NEED_RESCUER) { 1738be4d234dSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", 1739be4d234dSJens Axboe WQ_MEM_RECLAIM, 0); 1740917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1741917a38c7SKent Overstreet goto bad; 1742be4d234dSJens Axboe } 1743be4d234dSJens Axboe if (flags & BIOSET_PERCPU_CACHE) { 1744be4d234dSJens Axboe bs->cache = alloc_percpu(struct bio_alloc_cache); 1745be4d234dSJens Axboe if (!bs->cache) 1746be4d234dSJens Axboe goto bad; 1747be4d234dSJens Axboe cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1748be4d234dSJens Axboe } 1749917a38c7SKent Overstreet 1750917a38c7SKent Overstreet return 0; 1751917a38c7SKent Overstreet bad: 1752917a38c7SKent Overstreet bioset_exit(bs); 1753917a38c7SKent Overstreet return -ENOMEM; 1754917a38c7SKent Overstreet } 1755917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1756917a38c7SKent Overstreet 175728e89fd9SJens Axboe /* 175828e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 175928e89fd9SJens Axboe * another bio_set. 176028e89fd9SJens Axboe */ 176128e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 176228e89fd9SJens Axboe { 176328e89fd9SJens Axboe int flags; 176428e89fd9SJens Axboe 176528e89fd9SJens Axboe flags = 0; 176628e89fd9SJens Axboe if (src->bvec_pool.min_nr) 176728e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 176828e89fd9SJens Axboe if (src->rescue_workqueue) 176928e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 1770*b53f3dcdSMike Snitzer if (src->cache) 1771*b53f3dcdSMike Snitzer flags |= BIOSET_PERCPU_CACHE; 177228e89fd9SJens Axboe 177328e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 177428e89fd9SJens Axboe } 177528e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 177628e89fd9SJens Axboe 1777de76fd89SChristoph Hellwig static int __init init_bio(void) 1778f9c78b2bSJens Axboe { 1779f9c78b2bSJens Axboe int i; 1780f9c78b2bSJens Axboe 1781f9c78b2bSJens Axboe bio_integrity_init(); 1782de76fd89SChristoph Hellwig 1783de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1784f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1785f9c78b2bSJens Axboe 1786de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1787de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1788f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1789f9c78b2bSJens Axboe } 1790f9c78b2bSJens Axboe 1791be4d234dSJens Axboe cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1792be4d234dSJens Axboe bio_cpu_dead); 1793be4d234dSJens Axboe 1794f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1795f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1796f9c78b2bSJens Axboe 1797f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1798f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1799f9c78b2bSJens Axboe 1800f9c78b2bSJens Axboe return 0; 1801f9c78b2bSJens Axboe } 1802f9c78b2bSJens Axboe subsys_initcall(init_bio); 1803