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> 1808e18eabSJosef Bacik #include <linux/blk-cgroup.h> 19b4c5875dSDamien Le Moal #include <linux/highmem.h> 20de6a78b6SMing Lei #include <linux/sched/sysctl.h> 21a892c8d5SSatya Tangirala #include <linux/blk-crypto.h> 2249d1ec85SMing Lei #include <linux/xarray.h> 23f9c78b2bSJens Axboe 24f9c78b2bSJens Axboe #include <trace/events/block.h> 259e234eeaSShaohua Li #include "blk.h" 2667b42d0bSJosef Bacik #include "blk-rq-qos.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_write_hint = 0; 261da521626SJens Axboe bio->bi_status = 0; 262da521626SJens Axboe bio->bi_iter.bi_sector = 0; 263da521626SJens Axboe bio->bi_iter.bi_size = 0; 264da521626SJens Axboe bio->bi_iter.bi_idx = 0; 265da521626SJens Axboe bio->bi_iter.bi_bvec_done = 0; 266da521626SJens Axboe bio->bi_end_io = NULL; 267da521626SJens Axboe bio->bi_private = NULL; 268da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP 269da521626SJens Axboe bio->bi_blkg = NULL; 270da521626SJens Axboe bio->bi_issue.value = 0; 27149add496SChristoph Hellwig if (bdev) 27249add496SChristoph Hellwig bio_associate_blkg(bio); 273da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP_IOCOST 274da521626SJens Axboe bio->bi_iocost_cost = 0; 275da521626SJens Axboe #endif 276da521626SJens Axboe #endif 277da521626SJens Axboe #ifdef CONFIG_BLK_INLINE_ENCRYPTION 278da521626SJens Axboe bio->bi_crypt_context = NULL; 279da521626SJens Axboe #endif 280da521626SJens Axboe #ifdef CONFIG_BLK_DEV_INTEGRITY 281da521626SJens Axboe bio->bi_integrity = NULL; 282da521626SJens Axboe #endif 283da521626SJens Axboe bio->bi_vcnt = 0; 284da521626SJens Axboe 285c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 286dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2873e08773cSChristoph Hellwig bio->bi_cookie = BLK_QC_T_NONE; 2883a83f467SMing Lei 2893a83f467SMing Lei bio->bi_max_vecs = max_vecs; 290da521626SJens Axboe bio->bi_io_vec = table; 291da521626SJens Axboe bio->bi_pool = NULL; 292f9c78b2bSJens Axboe } 293f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 294f9c78b2bSJens Axboe 295f9c78b2bSJens Axboe /** 296f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 297f9c78b2bSJens Axboe * @bio: bio to reset 298a7c50c94SChristoph Hellwig * @bdev: block device to use the bio for 299a7c50c94SChristoph Hellwig * @opf: operation and flags for bio 300f9c78b2bSJens Axboe * 301f9c78b2bSJens Axboe * Description: 302f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 303f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 304f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 305f9c78b2bSJens Axboe * comment in struct bio. 306f9c78b2bSJens Axboe */ 307a7c50c94SChristoph Hellwig void bio_reset(struct bio *bio, struct block_device *bdev, unsigned int opf) 308f9c78b2bSJens Axboe { 3099ae3b3f5SJens Axboe bio_uninit(bio); 310f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 311c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 312a7c50c94SChristoph Hellwig bio->bi_bdev = bdev; 313a7c50c94SChristoph Hellwig bio->bi_opf = opf; 314f9c78b2bSJens Axboe } 315f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 316f9c78b2bSJens Axboe 31738f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 318f9c78b2bSJens Axboe { 3194246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3204246a0b6SChristoph Hellwig 3213edf5346SYufen Yu if (bio->bi_status && !parent->bi_status) 3224e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 323f9c78b2bSJens Axboe bio_put(bio); 32438f8baaeSChristoph Hellwig return parent; 32538f8baaeSChristoph Hellwig } 32638f8baaeSChristoph Hellwig 32738f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 32838f8baaeSChristoph Hellwig { 32938f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 330f9c78b2bSJens Axboe } 331f9c78b2bSJens Axboe 332f9c78b2bSJens Axboe /** 333f9c78b2bSJens Axboe * bio_chain - chain bio completions 334f9c78b2bSJens Axboe * @bio: the target bio 3355b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 336f9c78b2bSJens Axboe * 337f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 338f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 339f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 340f9c78b2bSJens Axboe * 341f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 342f9c78b2bSJens Axboe */ 343f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 344f9c78b2bSJens Axboe { 345f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 346f9c78b2bSJens Axboe 347f9c78b2bSJens Axboe bio->bi_private = parent; 348f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 349c4cf5261SJens Axboe bio_inc_remaining(parent); 350f9c78b2bSJens Axboe } 351f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 352f9c78b2bSJens Axboe 3530a3140eaSChaitanya Kulkarni struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev, 3540a3140eaSChaitanya Kulkarni unsigned int nr_pages, unsigned int opf, gfp_t gfp) 3553b005bf6SChristoph Hellwig { 35607888c66SChristoph Hellwig struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp); 3570a3140eaSChaitanya Kulkarni 3583b005bf6SChristoph Hellwig if (bio) { 3593b005bf6SChristoph Hellwig bio_chain(bio, new); 3603b005bf6SChristoph Hellwig submit_bio(bio); 3613b005bf6SChristoph Hellwig } 3623b005bf6SChristoph Hellwig 3633b005bf6SChristoph Hellwig return new; 3643b005bf6SChristoph Hellwig } 3653b005bf6SChristoph Hellwig EXPORT_SYMBOL_GPL(blk_next_bio); 3663b005bf6SChristoph Hellwig 367f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 368f9c78b2bSJens Axboe { 369f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 370f9c78b2bSJens Axboe struct bio *bio; 371f9c78b2bSJens Axboe 372f9c78b2bSJens Axboe while (1) { 373f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 374f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 375f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 376f9c78b2bSJens Axboe 377f9c78b2bSJens Axboe if (!bio) 378f9c78b2bSJens Axboe break; 379f9c78b2bSJens Axboe 380ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 381f9c78b2bSJens Axboe } 382f9c78b2bSJens Axboe } 383f9c78b2bSJens Axboe 384f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 385f9c78b2bSJens Axboe { 386f9c78b2bSJens Axboe struct bio_list punt, nopunt; 387f9c78b2bSJens Axboe struct bio *bio; 388f9c78b2bSJens Axboe 38947e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 39047e0fb46SNeilBrown return; 391f9c78b2bSJens Axboe /* 392f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 393f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 394f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 395f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 396f9c78b2bSJens Axboe * our own rescuer would be bad. 397f9c78b2bSJens Axboe * 398f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 399f9c78b2bSJens Axboe * remove from the middle of the list: 400f9c78b2bSJens Axboe */ 401f9c78b2bSJens Axboe 402f9c78b2bSJens Axboe bio_list_init(&punt); 403f9c78b2bSJens Axboe bio_list_init(&nopunt); 404f9c78b2bSJens Axboe 405f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 406f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 407f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 408f9c78b2bSJens Axboe 409f5fe1b51SNeilBrown bio_list_init(&nopunt); 410f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 411f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 412f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 413f9c78b2bSJens Axboe 414f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 415f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 416f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 417f9c78b2bSJens Axboe 418f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 419f9c78b2bSJens Axboe } 420f9c78b2bSJens Axboe 421f9c78b2bSJens Axboe /** 422f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 423609be106SChristoph Hellwig * @bdev: block device to allocate the bio for (can be %NULL) 424609be106SChristoph Hellwig * @nr_vecs: number of bvecs to pre-allocate 425609be106SChristoph Hellwig * @opf: operation and flags for bio 426519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 427f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 428f9c78b2bSJens Axboe * 4293175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 430f9c78b2bSJens Axboe * 4313175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4323175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4333175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4343175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4353175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4363175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 437f9c78b2bSJens Axboe * 4383175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4393175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 440ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 441f9c78b2bSJens Axboe * stack overflows. 442f9c78b2bSJens Axboe * 4433175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4443175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 445f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 446f9c78b2bSJens Axboe * thread. 447f9c78b2bSJens Axboe * 448f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 449f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 450ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 451f9c78b2bSJens Axboe * for per bio allocations. 452f9c78b2bSJens Axboe * 4533175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 454f9c78b2bSJens Axboe */ 455609be106SChristoph Hellwig struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs, 456609be106SChristoph Hellwig unsigned int opf, gfp_t gfp_mask, 4577a88fa19SDan Carpenter struct bio_set *bs) 458f9c78b2bSJens Axboe { 459f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 460f9c78b2bSJens Axboe struct bio *bio; 461f9c78b2bSJens Axboe void *p; 462f9c78b2bSJens Axboe 463609be106SChristoph Hellwig /* should not use nobvec bioset for nr_vecs > 0 */ 464609be106SChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0)) 465f9c78b2bSJens Axboe return NULL; 466f9c78b2bSJens Axboe 467f9c78b2bSJens Axboe /* 4683175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 4693175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 4703175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 471f9c78b2bSJens Axboe * 4723175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 4733175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 4743175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 4753175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 4763175199aSChristoph Hellwig * mempool's reserve. 477f9c78b2bSJens Axboe * 478f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 4793175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 4803175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 4813175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 4823175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 4833175199aSChristoph Hellwig * gfp_flags. 484f9c78b2bSJens Axboe */ 485f5fe1b51SNeilBrown if (current->bio_list && 486f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 48747e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 48847e0fb46SNeilBrown bs->rescue_workqueue) 489d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 490f9c78b2bSJens Axboe 4918aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 492f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 493f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 494f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4958aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 496f9c78b2bSJens Axboe } 497f9c78b2bSJens Axboe if (unlikely(!p)) 498f9c78b2bSJens Axboe return NULL; 499f9c78b2bSJens Axboe 5003175199aSChristoph Hellwig bio = p + bs->front_pad; 501609be106SChristoph Hellwig if (nr_vecs > BIO_INLINE_VECS) { 5023175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 503f9c78b2bSJens Axboe 504609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 505f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 506f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 507f9c78b2bSJens Axboe gfp_mask = saved_gfp; 508609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 509f9c78b2bSJens Axboe } 510f9c78b2bSJens Axboe if (unlikely(!bvl)) 511f9c78b2bSJens Axboe goto err_free; 512f9c78b2bSJens Axboe 51349add496SChristoph Hellwig bio_init(bio, bdev, bvl, nr_vecs, opf); 514609be106SChristoph Hellwig } else if (nr_vecs) { 51549add496SChristoph Hellwig bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf); 5163175199aSChristoph Hellwig } else { 51749add496SChristoph Hellwig bio_init(bio, bdev, NULL, 0, opf); 518f9c78b2bSJens Axboe } 519f9c78b2bSJens Axboe 520f9c78b2bSJens Axboe bio->bi_pool = bs; 521f9c78b2bSJens Axboe return bio; 522f9c78b2bSJens Axboe 523f9c78b2bSJens Axboe err_free: 5248aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 525f9c78b2bSJens Axboe return NULL; 526f9c78b2bSJens Axboe } 527f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 528f9c78b2bSJens Axboe 5293175199aSChristoph Hellwig /** 5303175199aSChristoph Hellwig * bio_kmalloc - kmalloc a bio for I/O 5313175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 5323175199aSChristoph Hellwig * @nr_iovecs: number of iovecs to pre-allocate 5333175199aSChristoph Hellwig * 5343175199aSChristoph Hellwig * Use kmalloc to allocate and initialize a bio. 5353175199aSChristoph Hellwig * 5363175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 5373175199aSChristoph Hellwig */ 5380f2e6ab8SChristoph Hellwig struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) 5393175199aSChristoph Hellwig { 5403175199aSChristoph Hellwig struct bio *bio; 5413175199aSChristoph Hellwig 5423175199aSChristoph Hellwig if (nr_iovecs > UIO_MAXIOV) 5433175199aSChristoph Hellwig return NULL; 5443175199aSChristoph Hellwig 5453175199aSChristoph Hellwig bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 5463175199aSChristoph Hellwig if (unlikely(!bio)) 5473175199aSChristoph Hellwig return NULL; 54849add496SChristoph Hellwig bio_init(bio, NULL, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs, 54949add496SChristoph Hellwig 0); 5503175199aSChristoph Hellwig bio->bi_pool = NULL; 5513175199aSChristoph Hellwig return bio; 5523175199aSChristoph Hellwig } 5533175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 5543175199aSChristoph Hellwig 5556f822e1bSChristoph Hellwig void zero_fill_bio(struct bio *bio) 556f9c78b2bSJens Axboe { 557f9c78b2bSJens Axboe struct bio_vec bv; 558f9c78b2bSJens Axboe struct bvec_iter iter; 559f9c78b2bSJens Axboe 560ab6c340eSChristoph Hellwig bio_for_each_segment(bv, bio, iter) 561ab6c340eSChristoph Hellwig memzero_bvec(&bv); 562f9c78b2bSJens Axboe } 5636f822e1bSChristoph Hellwig EXPORT_SYMBOL(zero_fill_bio); 564f9c78b2bSJens Axboe 56583c9c547SMing Lei /** 56683c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 56783c9c547SMing Lei * @bio: the bio to be truncated 56883c9c547SMing Lei * @new_size: new size for truncating the bio 56983c9c547SMing Lei * 57083c9c547SMing Lei * Description: 57183c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 57283c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 57383c9c547SMing Lei * be used for handling corner cases, such as bio eod. 57483c9c547SMing Lei */ 5754f7ab09aSChristoph Hellwig static void bio_truncate(struct bio *bio, unsigned new_size) 57685a8ce62SMing Lei { 57785a8ce62SMing Lei struct bio_vec bv; 57885a8ce62SMing Lei struct bvec_iter iter; 57985a8ce62SMing Lei unsigned int done = 0; 58085a8ce62SMing Lei bool truncated = false; 58185a8ce62SMing Lei 58285a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 58385a8ce62SMing Lei return; 58485a8ce62SMing Lei 58583c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 58685a8ce62SMing Lei goto exit; 58785a8ce62SMing Lei 58885a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 58985a8ce62SMing Lei if (done + bv.bv_len > new_size) { 59085a8ce62SMing Lei unsigned offset; 59185a8ce62SMing Lei 59285a8ce62SMing Lei if (!truncated) 59385a8ce62SMing Lei offset = new_size - done; 59485a8ce62SMing Lei else 59585a8ce62SMing Lei offset = 0; 5963ee859e3SOGAWA Hirofumi zero_user(bv.bv_page, bv.bv_offset + offset, 5973ee859e3SOGAWA Hirofumi bv.bv_len - offset); 59885a8ce62SMing Lei truncated = true; 59985a8ce62SMing Lei } 60085a8ce62SMing Lei done += bv.bv_len; 60185a8ce62SMing Lei } 60285a8ce62SMing Lei 60385a8ce62SMing Lei exit: 60485a8ce62SMing Lei /* 60585a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 60685a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 60785a8ce62SMing Lei * in its .end_bio() callback. 60885a8ce62SMing Lei * 60985a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 61085a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 61185a8ce62SMing Lei */ 61285a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 61385a8ce62SMing Lei } 61485a8ce62SMing Lei 615f9c78b2bSJens Axboe /** 61629125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 61729125ed6SChristoph Hellwig * @bio: bio to truncate 61829125ed6SChristoph Hellwig * 61929125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 62029125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 62129125ed6SChristoph Hellwig * 62229125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 62329125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 62429125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 62529125ed6SChristoph Hellwig * sector" case. 62629125ed6SChristoph Hellwig */ 62729125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 62829125ed6SChristoph Hellwig { 629309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 63029125ed6SChristoph Hellwig 63129125ed6SChristoph Hellwig if (!maxsector) 63229125ed6SChristoph Hellwig return; 63329125ed6SChristoph Hellwig 63429125ed6SChristoph Hellwig /* 63529125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 63629125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 63729125ed6SChristoph Hellwig * an EIO. 63829125ed6SChristoph Hellwig */ 63929125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 64029125ed6SChristoph Hellwig return; 64129125ed6SChristoph Hellwig 64229125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 64329125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 64429125ed6SChristoph Hellwig return; 64529125ed6SChristoph Hellwig 64629125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 64729125ed6SChristoph Hellwig } 64829125ed6SChristoph Hellwig 649be4d234dSJens Axboe #define ALLOC_CACHE_MAX 512 650be4d234dSJens Axboe #define ALLOC_CACHE_SLACK 64 651be4d234dSJens Axboe 652be4d234dSJens Axboe static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 653be4d234dSJens Axboe unsigned int nr) 654be4d234dSJens Axboe { 655be4d234dSJens Axboe unsigned int i = 0; 656be4d234dSJens Axboe struct bio *bio; 657be4d234dSJens Axboe 658fcade2ceSJens Axboe while ((bio = cache->free_list) != NULL) { 659fcade2ceSJens Axboe cache->free_list = bio->bi_next; 660be4d234dSJens Axboe cache->nr--; 661be4d234dSJens Axboe bio_free(bio); 662be4d234dSJens Axboe if (++i == nr) 663be4d234dSJens Axboe break; 664be4d234dSJens Axboe } 665be4d234dSJens Axboe } 666be4d234dSJens Axboe 667be4d234dSJens Axboe static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 668be4d234dSJens Axboe { 669be4d234dSJens Axboe struct bio_set *bs; 670be4d234dSJens Axboe 671be4d234dSJens Axboe bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 672be4d234dSJens Axboe if (bs->cache) { 673be4d234dSJens Axboe struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 674be4d234dSJens Axboe 675be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 676be4d234dSJens Axboe } 677be4d234dSJens Axboe return 0; 678be4d234dSJens Axboe } 679be4d234dSJens Axboe 680be4d234dSJens Axboe static void bio_alloc_cache_destroy(struct bio_set *bs) 681be4d234dSJens Axboe { 682be4d234dSJens Axboe int cpu; 683be4d234dSJens Axboe 684be4d234dSJens Axboe if (!bs->cache) 685be4d234dSJens Axboe return; 686be4d234dSJens Axboe 687be4d234dSJens Axboe cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 688be4d234dSJens Axboe for_each_possible_cpu(cpu) { 689be4d234dSJens Axboe struct bio_alloc_cache *cache; 690be4d234dSJens Axboe 691be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, cpu); 692be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 693be4d234dSJens Axboe } 694be4d234dSJens Axboe free_percpu(bs->cache); 695be4d234dSJens Axboe } 696be4d234dSJens Axboe 69729125ed6SChristoph Hellwig /** 698f9c78b2bSJens Axboe * bio_put - release a reference to a bio 699f9c78b2bSJens Axboe * @bio: bio to release reference to 700f9c78b2bSJens Axboe * 701f9c78b2bSJens Axboe * Description: 702f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 7039b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 704f9c78b2bSJens Axboe **/ 705f9c78b2bSJens Axboe void bio_put(struct bio *bio) 706f9c78b2bSJens Axboe { 707be4d234dSJens Axboe if (unlikely(bio_flagged(bio, BIO_REFFED))) { 7089e8c0d0dSChristoph Hellwig BUG_ON(!atomic_read(&bio->__bi_cnt)); 709be4d234dSJens Axboe if (!atomic_dec_and_test(&bio->__bi_cnt)) 710be4d234dSJens Axboe return; 711be4d234dSJens Axboe } 712f9c78b2bSJens Axboe 713be4d234dSJens Axboe if (bio_flagged(bio, BIO_PERCPU_CACHE)) { 714be4d234dSJens Axboe struct bio_alloc_cache *cache; 715be4d234dSJens Axboe 716be4d234dSJens Axboe bio_uninit(bio); 717be4d234dSJens Axboe cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 718fcade2ceSJens Axboe bio->bi_next = cache->free_list; 719fcade2ceSJens Axboe cache->free_list = bio; 720be4d234dSJens Axboe if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 721be4d234dSJens Axboe bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 722be4d234dSJens Axboe put_cpu(); 723be4d234dSJens Axboe } else { 724f9c78b2bSJens Axboe bio_free(bio); 725f9c78b2bSJens Axboe } 726dac56212SJens Axboe } 727f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 728f9c78b2bSJens Axboe 729f9c78b2bSJens Axboe /** 730f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 731f9c78b2bSJens Axboe * @bio: destination bio 732f9c78b2bSJens Axboe * @bio_src: bio to clone 733f9c78b2bSJens Axboe * 734f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 735f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 736f9c78b2bSJens Axboe * bio will be one. 737f9c78b2bSJens Axboe * 738f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 739f9c78b2bSJens Axboe */ 740f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 741f9c78b2bSJens Axboe { 7427a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs); 743f9c78b2bSJens Axboe 744f9c78b2bSJens Axboe /* 745309dca30SChristoph Hellwig * most users will be overriding ->bi_bdev with a new target, 746f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 747f9c78b2bSJens Axboe */ 748309dca30SChristoph Hellwig bio->bi_bdev = bio_src->bi_bdev; 749b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 750111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 751111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 75246bbf653SChristoph Hellwig if (bio_flagged(bio_src, BIO_REMAPPED)) 75346bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 7541eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 755ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 756cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 757f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 758f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 75920bd723eSPaolo Valente 760db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 761e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 762f9c78b2bSJens Axboe } 763f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 764f9c78b2bSJens Axboe 765f9c78b2bSJens Axboe /** 766f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 767f9c78b2bSJens Axboe * @bio: bio to clone 768f9c78b2bSJens Axboe * @gfp_mask: allocation priority 769f9c78b2bSJens Axboe * @bs: bio_set to allocate from 770f9c78b2bSJens Axboe * 771f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 772f9c78b2bSJens Axboe */ 773f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 774f9c78b2bSJens Axboe { 775f9c78b2bSJens Axboe struct bio *b; 776f9c78b2bSJens Axboe 777609be106SChristoph Hellwig b = bio_alloc_bioset(NULL, 0, 0, gfp_mask, bs); 778f9c78b2bSJens Axboe if (!b) 779f9c78b2bSJens Axboe return NULL; 780f9c78b2bSJens Axboe 781f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 782f9c78b2bSJens Axboe 78307560151SEric Biggers if (bio_crypt_clone(b, bio, gfp_mask) < 0) 78407560151SEric Biggers goto err_put; 785a892c8d5SSatya Tangirala 78607560151SEric Biggers if (bio_integrity(bio) && 78707560151SEric Biggers bio_integrity_clone(b, bio, gfp_mask) < 0) 78807560151SEric Biggers goto err_put; 789f9c78b2bSJens Axboe 790f9c78b2bSJens Axboe return b; 79107560151SEric Biggers 79207560151SEric Biggers err_put: 79307560151SEric Biggers bio_put(b); 79407560151SEric Biggers return NULL; 795f9c78b2bSJens Axboe } 796f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 797f9c78b2bSJens Axboe 7985cbd28e3SChristoph Hellwig const char *bio_devname(struct bio *bio, char *buf) 7995cbd28e3SChristoph Hellwig { 800309dca30SChristoph Hellwig return bdevname(bio->bi_bdev, buf); 8015cbd28e3SChristoph Hellwig } 8025cbd28e3SChristoph Hellwig EXPORT_SYMBOL(bio_devname); 8035cbd28e3SChristoph Hellwig 8049a6083beSChristoph Hellwig /** 8059a6083beSChristoph Hellwig * bio_full - check if the bio is full 8069a6083beSChristoph Hellwig * @bio: bio to check 8079a6083beSChristoph Hellwig * @len: length of one segment to be added 8089a6083beSChristoph Hellwig * 8099a6083beSChristoph Hellwig * Return true if @bio is full and one segment with @len bytes can't be 8109a6083beSChristoph Hellwig * added to the bio, otherwise return false 8119a6083beSChristoph Hellwig */ 8129a6083beSChristoph Hellwig static inline bool bio_full(struct bio *bio, unsigned len) 8139a6083beSChristoph Hellwig { 8149a6083beSChristoph Hellwig if (bio->bi_vcnt >= bio->bi_max_vecs) 8159a6083beSChristoph Hellwig return true; 8169a6083beSChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) 8179a6083beSChristoph Hellwig return true; 8189a6083beSChristoph Hellwig return false; 8199a6083beSChristoph Hellwig } 8209a6083beSChristoph Hellwig 8215919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 8225919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 823ff896738SChristoph Hellwig bool *same_page) 8245919482eSMing Lei { 825d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 826d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 8275919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 8285919482eSMing Lei 8295919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 8305919482eSMing Lei return false; 8315919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 8325919482eSMing Lei return false; 83352d52d1cSChristoph Hellwig 834ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 835d8166519SMatthew Wilcox (Oracle) if (*same_page) 8365919482eSMing Lei return true; 837d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 8385919482eSMing Lei } 8395919482eSMing Lei 8409774b391SChristoph Hellwig /** 8419774b391SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 8429774b391SChristoph Hellwig * @bio: destination bio 8439774b391SChristoph Hellwig * @page: start page to add 8449774b391SChristoph Hellwig * @len: length of the data to add 8459774b391SChristoph Hellwig * @off: offset of the data relative to @page 8469774b391SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 8479774b391SChristoph Hellwig * 8489774b391SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 8499774b391SChristoph Hellwig * useful optimisation for file systems with a block size smaller than the 8509774b391SChristoph Hellwig * page size. 8519774b391SChristoph Hellwig * 8529774b391SChristoph Hellwig * Warn if (@len, @off) crosses pages in case that @same_page is true. 8539774b391SChristoph Hellwig * 8549774b391SChristoph Hellwig * Return %true on success or %false on failure. 8559774b391SChristoph Hellwig */ 8569774b391SChristoph Hellwig static bool __bio_try_merge_page(struct bio *bio, struct page *page, 8579774b391SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 8589774b391SChristoph Hellwig { 8599774b391SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8609774b391SChristoph Hellwig return false; 8619774b391SChristoph Hellwig 8629774b391SChristoph Hellwig if (bio->bi_vcnt > 0) { 8639774b391SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 8649774b391SChristoph Hellwig 8659774b391SChristoph Hellwig if (page_is_mergeable(bv, page, len, off, same_page)) { 8669774b391SChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) { 8679774b391SChristoph Hellwig *same_page = false; 8689774b391SChristoph Hellwig return false; 8699774b391SChristoph Hellwig } 8709774b391SChristoph Hellwig bv->bv_len += len; 8719774b391SChristoph Hellwig bio->bi_iter.bi_size += len; 8729774b391SChristoph Hellwig return true; 8739774b391SChristoph Hellwig } 8749774b391SChristoph Hellwig } 8759774b391SChristoph Hellwig return false; 8769774b391SChristoph Hellwig } 8779774b391SChristoph Hellwig 878e4581105SChristoph Hellwig /* 879e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 880e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 881e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 882e4581105SChristoph Hellwig */ 883e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 884e4581105SChristoph Hellwig struct page *page, unsigned len, 885e4581105SChristoph Hellwig unsigned offset, bool *same_page) 886489fbbcbSMing Lei { 887384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 888489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 889489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 890489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 891489fbbcbSMing Lei 892489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 893489fbbcbSMing Lei return false; 894489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 895489fbbcbSMing Lei return false; 896384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 897489fbbcbSMing Lei } 898489fbbcbSMing Lei 899f4595875SShaohua Li /** 900e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 901c66a14d0SKent Overstreet * @q: the target queue 902c66a14d0SKent Overstreet * @bio: destination bio 903c66a14d0SKent Overstreet * @page: page to add 904c66a14d0SKent Overstreet * @len: vec entry length 905c66a14d0SKent Overstreet * @offset: vec entry offset 906e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 907e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 908f9c78b2bSJens Axboe * 909e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 910e4581105SChristoph Hellwig * and gap limitations. 911f9c78b2bSJens Axboe */ 912e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 91319047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 914e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 915f9c78b2bSJens Axboe { 916f9c78b2bSJens Axboe struct bio_vec *bvec; 917f9c78b2bSJens Axboe 918e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 919f9c78b2bSJens Axboe return 0; 920f9c78b2bSJens Axboe 921e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 922f9c78b2bSJens Axboe return 0; 923f9c78b2bSJens Axboe 924f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 925e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 926384209cdSChristoph Hellwig return len; 927320ea869SChristoph Hellwig 928320ea869SChristoph Hellwig /* 929320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 930320ea869SChristoph Hellwig * would create a gap, disallow it. 931320ea869SChristoph Hellwig */ 932384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 933320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 934320ea869SChristoph Hellwig return 0; 935f9c78b2bSJens Axboe } 936f9c78b2bSJens Axboe 93779d08f89SMing Lei if (bio_full(bio, len)) 938f9c78b2bSJens Axboe return 0; 939f9c78b2bSJens Axboe 94014ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 941489fbbcbSMing Lei return 0; 942489fbbcbSMing Lei 943f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 944f9c78b2bSJens Axboe bvec->bv_page = page; 945f9c78b2bSJens Axboe bvec->bv_len = len; 946f9c78b2bSJens Axboe bvec->bv_offset = offset; 947fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 948dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 949f9c78b2bSJens Axboe return len; 950f9c78b2bSJens Axboe } 95119047087SMing Lei 952e4581105SChristoph Hellwig /** 953e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 954e4581105SChristoph Hellwig * @q: the target queue 955e4581105SChristoph Hellwig * @bio: destination bio 956e4581105SChristoph Hellwig * @page: page to add 957e4581105SChristoph Hellwig * @len: vec entry length 958e4581105SChristoph Hellwig * @offset: vec entry offset 959e4581105SChristoph Hellwig * 960e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 961e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 962e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 963e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 964e4581105SChristoph Hellwig * 965e4581105SChristoph Hellwig * This should only be used by passthrough bios. 966e4581105SChristoph Hellwig */ 96719047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 96819047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 96919047087SMing Lei { 970d1916c86SChristoph Hellwig bool same_page = false; 971e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 972e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 97319047087SMing Lei } 974f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 975f9c78b2bSJens Axboe 976f9c78b2bSJens Axboe /** 977ae29333fSJohannes Thumshirn * bio_add_zone_append_page - attempt to add page to zone-append bio 978ae29333fSJohannes Thumshirn * @bio: destination bio 979ae29333fSJohannes Thumshirn * @page: page to add 980ae29333fSJohannes Thumshirn * @len: vec entry length 981ae29333fSJohannes Thumshirn * @offset: vec entry offset 982ae29333fSJohannes Thumshirn * 983ae29333fSJohannes Thumshirn * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 984ae29333fSJohannes Thumshirn * for a zone-append request. This can fail for a number of reasons, such as the 985ae29333fSJohannes Thumshirn * bio being full or the target block device is not a zoned block device or 986ae29333fSJohannes Thumshirn * other limitations of the target block device. The target block device must 987ae29333fSJohannes Thumshirn * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 988ae29333fSJohannes Thumshirn * to an empty bio. 989ae29333fSJohannes Thumshirn * 990ae29333fSJohannes Thumshirn * Returns: number of bytes added to the bio, or 0 in case of a failure. 991ae29333fSJohannes Thumshirn */ 992ae29333fSJohannes Thumshirn int bio_add_zone_append_page(struct bio *bio, struct page *page, 993ae29333fSJohannes Thumshirn unsigned int len, unsigned int offset) 994ae29333fSJohannes Thumshirn { 9953caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 996ae29333fSJohannes Thumshirn bool same_page = false; 997ae29333fSJohannes Thumshirn 998ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 999ae29333fSJohannes Thumshirn return 0; 1000ae29333fSJohannes Thumshirn 1001ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 1002ae29333fSJohannes Thumshirn return 0; 1003ae29333fSJohannes Thumshirn 1004ae29333fSJohannes Thumshirn return bio_add_hw_page(q, bio, page, len, offset, 1005ae29333fSJohannes Thumshirn queue_max_zone_append_sectors(q), &same_page); 1006ae29333fSJohannes Thumshirn } 1007ae29333fSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 1008ae29333fSJohannes Thumshirn 1009ae29333fSJohannes Thumshirn /** 1010551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 10110aa69fd3SChristoph Hellwig * @bio: destination bio 1012551879a4SMing Lei * @page: start page to add 1013551879a4SMing Lei * @len: length of the data to add, may cross pages 1014551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 10150aa69fd3SChristoph Hellwig * 10160aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 10170aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 10180aa69fd3SChristoph Hellwig */ 10190aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 10200aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 10210aa69fd3SChristoph Hellwig { 10220aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 10230aa69fd3SChristoph Hellwig 10240aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 102579d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 10260aa69fd3SChristoph Hellwig 10270aa69fd3SChristoph Hellwig bv->bv_page = page; 10280aa69fd3SChristoph Hellwig bv->bv_offset = off; 10290aa69fd3SChristoph Hellwig bv->bv_len = len; 10300aa69fd3SChristoph Hellwig 10310aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 10320aa69fd3SChristoph Hellwig bio->bi_vcnt++; 1033b8e24a93SJohannes Weiner 1034b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 1035b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 10360aa69fd3SChristoph Hellwig } 10370aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 10380aa69fd3SChristoph Hellwig 10390aa69fd3SChristoph Hellwig /** 1040551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 1041f9c78b2bSJens Axboe * @bio: destination bio 1042551879a4SMing Lei * @page: start page to add 1043551879a4SMing Lei * @len: vec entry length, may cross pages 1044551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 1045f9c78b2bSJens Axboe * 1046551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 1047c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1048f9c78b2bSJens Axboe */ 1049c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 1050c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 1051f9c78b2bSJens Axboe { 1052ff896738SChristoph Hellwig bool same_page = false; 1053ff896738SChristoph Hellwig 1054ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 105579d08f89SMing Lei if (bio_full(bio, len)) 1056c66a14d0SKent Overstreet return 0; 10570aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 1058c66a14d0SKent Overstreet } 1059c66a14d0SKent Overstreet return len; 1060f9c78b2bSJens Axboe } 1061f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 1062f9c78b2bSJens Axboe 106385f5a74cSMatthew Wilcox (Oracle) /** 106485f5a74cSMatthew Wilcox (Oracle) * bio_add_folio - Attempt to add part of a folio to a bio. 106585f5a74cSMatthew Wilcox (Oracle) * @bio: BIO to add to. 106685f5a74cSMatthew Wilcox (Oracle) * @folio: Folio to add. 106785f5a74cSMatthew Wilcox (Oracle) * @len: How many bytes from the folio to add. 106885f5a74cSMatthew Wilcox (Oracle) * @off: First byte in this folio to add. 106985f5a74cSMatthew Wilcox (Oracle) * 107085f5a74cSMatthew Wilcox (Oracle) * Filesystems that use folios can call this function instead of calling 107185f5a74cSMatthew Wilcox (Oracle) * bio_add_page() for each page in the folio. If @off is bigger than 107285f5a74cSMatthew Wilcox (Oracle) * PAGE_SIZE, this function can create a bio_vec that starts in a page 107385f5a74cSMatthew Wilcox (Oracle) * after the bv_page. BIOs do not support folios that are 4GiB or larger. 107485f5a74cSMatthew Wilcox (Oracle) * 107585f5a74cSMatthew Wilcox (Oracle) * Return: Whether the addition was successful. 107685f5a74cSMatthew Wilcox (Oracle) */ 107785f5a74cSMatthew Wilcox (Oracle) bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, 107885f5a74cSMatthew Wilcox (Oracle) size_t off) 107985f5a74cSMatthew Wilcox (Oracle) { 108085f5a74cSMatthew Wilcox (Oracle) if (len > UINT_MAX || off > UINT_MAX) 1081*455a844dSJiapeng Chong return false; 108285f5a74cSMatthew Wilcox (Oracle) return bio_add_page(bio, &folio->page, len, off) > 0; 108385f5a74cSMatthew Wilcox (Oracle) } 108485f5a74cSMatthew Wilcox (Oracle) 1085c809084aSPavel Begunkov void __bio_release_pages(struct bio *bio, bool mark_dirty) 10867321ecbfSChristoph Hellwig { 10877321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 10887321ecbfSChristoph Hellwig struct bio_vec *bvec; 10897321ecbfSChristoph Hellwig 1090d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1091d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 1092d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 10937321ecbfSChristoph Hellwig put_page(bvec->bv_page); 10947321ecbfSChristoph Hellwig } 1095d241a95fSChristoph Hellwig } 1096c809084aSPavel Begunkov EXPORT_SYMBOL_GPL(__bio_release_pages); 10977321ecbfSChristoph Hellwig 10981bb6b810SPavel Begunkov void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 10996d0c48aeSJens Axboe { 1100fa5fa8ecSPavel Begunkov size_t size = iov_iter_count(iter); 1101fa5fa8ecSPavel Begunkov 11027a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_max_vecs); 11036d0c48aeSJens Axboe 1104fa5fa8ecSPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1105fa5fa8ecSPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1106fa5fa8ecSPavel Begunkov size_t max_sectors = queue_max_zone_append_sectors(q); 1107fa5fa8ecSPavel Begunkov 1108fa5fa8ecSPavel Begunkov size = min(size, max_sectors << SECTOR_SHIFT); 1109fa5fa8ecSPavel Begunkov } 1110fa5fa8ecSPavel Begunkov 1111c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 1112c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1113c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 1114fa5fa8ecSPavel Begunkov bio->bi_iter.bi_size = size; 1115ed97ce5eSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 1116977be012SChristoph Hellwig bio_set_flag(bio, BIO_CLONED); 11177de55b7dSJohannes Thumshirn } 11186d0c48aeSJens Axboe 1119d9cf3bd5SPavel Begunkov static void bio_put_pages(struct page **pages, size_t size, size_t off) 1120d9cf3bd5SPavel Begunkov { 1121d9cf3bd5SPavel Begunkov size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE); 1122d9cf3bd5SPavel Begunkov 1123d9cf3bd5SPavel Begunkov for (i = 0; i < nr; i++) 1124d9cf3bd5SPavel Begunkov put_page(pages[i]); 1125d9cf3bd5SPavel Begunkov } 1126d9cf3bd5SPavel Begunkov 1127576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1128576ed913SChristoph Hellwig 11292cefe4dbSKent Overstreet /** 113017d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 11312cefe4dbSKent Overstreet * @bio: bio to add pages to 11322cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 11332cefe4dbSKent Overstreet * 113417d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 11352cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 113617d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 11373cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 11382cefe4dbSKent Overstreet */ 113917d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 11402cefe4dbSKent Overstreet { 1141576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1142576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 11432cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 11442cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 114545691804SChristoph Hellwig bool same_page = false; 1146576ed913SChristoph Hellwig ssize_t size, left; 1147576ed913SChristoph Hellwig unsigned len, i; 1148b403ea24SMartin Wilck size_t offset; 1149576ed913SChristoph Hellwig 1150576ed913SChristoph Hellwig /* 1151576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 1152576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 1153576ed913SChristoph Hellwig * without overwriting the temporary page array. 1154576ed913SChristoph Hellwig */ 1155576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1156576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 11572cefe4dbSKent Overstreet 115835c820e7SJens Axboe size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 11592cefe4dbSKent Overstreet if (unlikely(size <= 0)) 11602cefe4dbSKent Overstreet return size ? size : -EFAULT; 11612cefe4dbSKent Overstreet 1162576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1163576ed913SChristoph Hellwig struct page *page = pages[i]; 11642cefe4dbSKent Overstreet 1165576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 116645691804SChristoph Hellwig 116745691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 116845691804SChristoph Hellwig if (same_page) 116945691804SChristoph Hellwig put_page(page); 117045691804SChristoph Hellwig } else { 1171d9cf3bd5SPavel Begunkov if (WARN_ON_ONCE(bio_full(bio, len))) { 1172d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 1173576ed913SChristoph Hellwig return -EINVAL; 1174d9cf3bd5SPavel Begunkov } 117545691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 117645691804SChristoph Hellwig } 1177576ed913SChristoph Hellwig offset = 0; 11782cefe4dbSKent Overstreet } 11792cefe4dbSKent Overstreet 11802cefe4dbSKent Overstreet iov_iter_advance(iter, size); 11812cefe4dbSKent Overstreet return 0; 11822cefe4dbSKent Overstreet } 118317d51b10SMartin Wilck 11840512a75bSKeith Busch static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 11850512a75bSKeith Busch { 11860512a75bSKeith Busch unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 11870512a75bSKeith Busch unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 11883caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 11890512a75bSKeith Busch unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 11900512a75bSKeith Busch struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 11910512a75bSKeith Busch struct page **pages = (struct page **)bv; 11920512a75bSKeith Busch ssize_t size, left; 11930512a75bSKeith Busch unsigned len, i; 11940512a75bSKeith Busch size_t offset; 11954977d121SNaohiro Aota int ret = 0; 11960512a75bSKeith Busch 11970512a75bSKeith Busch if (WARN_ON_ONCE(!max_append_sectors)) 11980512a75bSKeith Busch return 0; 11990512a75bSKeith Busch 12000512a75bSKeith Busch /* 12010512a75bSKeith Busch * Move page array up in the allocated memory for the bio vecs as far as 12020512a75bSKeith Busch * possible so that we can start filling biovecs from the beginning 12030512a75bSKeith Busch * without overwriting the temporary page array. 12040512a75bSKeith Busch */ 12050512a75bSKeith Busch BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 12060512a75bSKeith Busch pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 12070512a75bSKeith Busch 12080512a75bSKeith Busch size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 12090512a75bSKeith Busch if (unlikely(size <= 0)) 12100512a75bSKeith Busch return size ? size : -EFAULT; 12110512a75bSKeith Busch 12120512a75bSKeith Busch for (left = size, i = 0; left > 0; left -= len, i++) { 12130512a75bSKeith Busch struct page *page = pages[i]; 12140512a75bSKeith Busch bool same_page = false; 12150512a75bSKeith Busch 12160512a75bSKeith Busch len = min_t(size_t, PAGE_SIZE - offset, left); 12170512a75bSKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 12184977d121SNaohiro Aota max_append_sectors, &same_page) != len) { 1219d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 12204977d121SNaohiro Aota ret = -EINVAL; 12214977d121SNaohiro Aota break; 12224977d121SNaohiro Aota } 12230512a75bSKeith Busch if (same_page) 12240512a75bSKeith Busch put_page(page); 12250512a75bSKeith Busch offset = 0; 12260512a75bSKeith Busch } 12270512a75bSKeith Busch 12284977d121SNaohiro Aota iov_iter_advance(iter, size - left); 12294977d121SNaohiro Aota return ret; 12300512a75bSKeith Busch } 12310512a75bSKeith Busch 123217d51b10SMartin Wilck /** 12336d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 123417d51b10SMartin Wilck * @bio: bio to add pages to 12356d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 123617d51b10SMartin Wilck * 12376d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 12386d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 12396d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1240c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1241c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1242c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1243c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1244c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1245c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 12466d0c48aeSJens Axboe * 124717d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 12485cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 12496d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 12506d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 12510cf41e5eSPavel Begunkov * 12520cf41e5eSPavel Begunkov * It's intended for direct IO, so doesn't do PSI tracking, the caller is 12530cf41e5eSPavel Begunkov * responsible for setting BIO_WORKINGSET if necessary. 125417d51b10SMartin Wilck */ 125517d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 125617d51b10SMartin Wilck { 1257c42bca92SPavel Begunkov int ret = 0; 125814eacf12SChristoph Hellwig 1259c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1260fa5fa8ecSPavel Begunkov bio_iov_bvec_set(bio, iter); 1261fa5fa8ecSPavel Begunkov iov_iter_advance(iter, bio->bi_iter.bi_size); 1262fa5fa8ecSPavel Begunkov return 0; 126386004515SChristoph Hellwig } 126417d51b10SMartin Wilck 126517d51b10SMartin Wilck do { 1266c42bca92SPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) 12670512a75bSKeith Busch ret = __bio_iov_append_get_pages(bio, iter); 12686d0c48aeSJens Axboe else 12696d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 127079d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 127117d51b10SMartin Wilck 12720cf41e5eSPavel Begunkov /* don't account direct I/O as memory stall */ 12730cf41e5eSPavel Begunkov bio_clear_flag(bio, BIO_WORKINGSET); 127414eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 127517d51b10SMartin Wilck } 127629b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 12772cefe4dbSKent Overstreet 12784246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1279f9c78b2bSJens Axboe { 128065e53aabSChristoph Hellwig complete(bio->bi_private); 1281f9c78b2bSJens Axboe } 1282f9c78b2bSJens Axboe 1283f9c78b2bSJens Axboe /** 1284f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1285f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1286f9c78b2bSJens Axboe * 1287f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1288f9c78b2bSJens Axboe * bio_endio() on failure. 12893d289d68SJan Kara * 12903d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 12913d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 12923d289d68SJan Kara * on his own. 1293f9c78b2bSJens Axboe */ 12944e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1295f9c78b2bSJens Axboe { 1296309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1297309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1298de6a78b6SMing Lei unsigned long hang_check; 1299f9c78b2bSJens Axboe 130065e53aabSChristoph Hellwig bio->bi_private = &done; 1301f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 13021eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 13034e49ea4aSMike Christie submit_bio(bio); 1304de6a78b6SMing Lei 1305de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1306de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1307de6a78b6SMing Lei if (hang_check) 1308de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1309de6a78b6SMing Lei hang_check * (HZ/2))) 1310de6a78b6SMing Lei ; 1311de6a78b6SMing Lei else 131265e53aabSChristoph Hellwig wait_for_completion_io(&done); 1313f9c78b2bSJens Axboe 131465e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1315f9c78b2bSJens Axboe } 1316f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1317f9c78b2bSJens Axboe 1318d4aa57a1SJens Axboe void __bio_advance(struct bio *bio, unsigned bytes) 1319f9c78b2bSJens Axboe { 1320f9c78b2bSJens Axboe if (bio_integrity(bio)) 1321f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1322f9c78b2bSJens Axboe 1323a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1324f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1325f9c78b2bSJens Axboe } 1326d4aa57a1SJens Axboe EXPORT_SYMBOL(__bio_advance); 1327f9c78b2bSJens Axboe 132845db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 132945db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1330f9c78b2bSJens Axboe { 133145db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 1332f8b679a0SChristoph Hellwig struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1333f8b679a0SChristoph Hellwig struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1334f8b679a0SChristoph Hellwig unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1335f8b679a0SChristoph Hellwig void *src_buf; 133645db54d5SKent Overstreet 1337f8b679a0SChristoph Hellwig src_buf = bvec_kmap_local(&src_bv); 1338f8b679a0SChristoph Hellwig memcpy_to_bvec(&dst_bv, src_buf); 1339f8b679a0SChristoph Hellwig kunmap_local(src_buf); 13406e6e811dSKent Overstreet 134122b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 134222b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 134345db54d5SKent Overstreet } 134445db54d5SKent Overstreet } 134545db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 134645db54d5SKent Overstreet 134745db54d5SKent Overstreet /** 134845db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 134945db54d5SKent Overstreet * @src: source bio 135045db54d5SKent Overstreet * @dst: destination bio 135145db54d5SKent Overstreet * 135245db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 135345db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 135445db54d5SKent Overstreet */ 135545db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 135645db54d5SKent Overstreet { 135745db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 135845db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 135945db54d5SKent Overstreet 136045db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 136145db54d5SKent Overstreet } 136245db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 136345db54d5SKent Overstreet 1364491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 13651dfa0f68SChristoph Hellwig { 13661dfa0f68SChristoph Hellwig struct bio_vec *bvec; 13676dc4f100SMing Lei struct bvec_iter_all iter_all; 13681dfa0f68SChristoph Hellwig 13692b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 13701dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 13711dfa0f68SChristoph Hellwig } 1372491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 13731dfa0f68SChristoph Hellwig 1374f9c78b2bSJens Axboe /* 1375f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1376f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1377f9c78b2bSJens Axboe * 1378f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1379f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1380f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1381f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1382f9c78b2bSJens Axboe * in process context. 1383f9c78b2bSJens Axboe * 1384f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1385f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1386f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1387f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1388f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1389f9c78b2bSJens Axboe * 1390f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1391f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1392f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1393f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1394f9c78b2bSJens Axboe * pagecache. 1395f9c78b2bSJens Axboe * 1396f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1397f9c78b2bSJens Axboe * deferred bio dirtying paths. 1398f9c78b2bSJens Axboe */ 1399f9c78b2bSJens Axboe 1400f9c78b2bSJens Axboe /* 1401f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1402f9c78b2bSJens Axboe */ 1403f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1404f9c78b2bSJens Axboe { 1405f9c78b2bSJens Axboe struct bio_vec *bvec; 14066dc4f100SMing Lei struct bvec_iter_all iter_all; 1407f9c78b2bSJens Axboe 14082b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 14093bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 14103bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1411f9c78b2bSJens Axboe } 1412f9c78b2bSJens Axboe } 1413f9c78b2bSJens Axboe 1414f9c78b2bSJens Axboe /* 1415f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1416f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1417f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 141824d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1419f9c78b2bSJens Axboe * 1420f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1421ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1422ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1423f9c78b2bSJens Axboe */ 1424f9c78b2bSJens Axboe 1425f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1426f9c78b2bSJens Axboe 1427f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1428f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1429f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1430f9c78b2bSJens Axboe 1431f9c78b2bSJens Axboe /* 1432f9c78b2bSJens Axboe * This runs in process context 1433f9c78b2bSJens Axboe */ 1434f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1435f9c78b2bSJens Axboe { 143624d5493fSChristoph Hellwig struct bio *bio, *next; 1437f9c78b2bSJens Axboe 143824d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 143924d5493fSChristoph Hellwig next = bio_dirty_list; 1440f9c78b2bSJens Axboe bio_dirty_list = NULL; 144124d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1442f9c78b2bSJens Axboe 144324d5493fSChristoph Hellwig while ((bio = next) != NULL) { 144424d5493fSChristoph Hellwig next = bio->bi_private; 1445f9c78b2bSJens Axboe 1446d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1447f9c78b2bSJens Axboe bio_put(bio); 1448f9c78b2bSJens Axboe } 1449f9c78b2bSJens Axboe } 1450f9c78b2bSJens Axboe 1451f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1452f9c78b2bSJens Axboe { 1453f9c78b2bSJens Axboe struct bio_vec *bvec; 145424d5493fSChristoph Hellwig unsigned long flags; 14556dc4f100SMing Lei struct bvec_iter_all iter_all; 1456f9c78b2bSJens Axboe 14572b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 145824d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 145924d5493fSChristoph Hellwig goto defer; 1460f9c78b2bSJens Axboe } 1461f9c78b2bSJens Axboe 1462d241a95fSChristoph Hellwig bio_release_pages(bio, false); 146324d5493fSChristoph Hellwig bio_put(bio); 146424d5493fSChristoph Hellwig return; 146524d5493fSChristoph Hellwig defer: 1466f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1467f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1468f9c78b2bSJens Axboe bio_dirty_list = bio; 1469f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1470f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1471f9c78b2bSJens Axboe } 1472f9c78b2bSJens Axboe 1473c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1474c4cf5261SJens Axboe { 1475c4cf5261SJens Axboe /* 1476c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1477c4cf5261SJens Axboe * we always end io on the first invocation. 1478c4cf5261SJens Axboe */ 1479c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1480c4cf5261SJens Axboe return true; 1481c4cf5261SJens Axboe 1482c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1483c4cf5261SJens Axboe 1484326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1485b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1486c4cf5261SJens Axboe return true; 1487326e1dbbSMike Snitzer } 1488c4cf5261SJens Axboe 1489c4cf5261SJens Axboe return false; 1490c4cf5261SJens Axboe } 1491c4cf5261SJens Axboe 1492f9c78b2bSJens Axboe /** 1493f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1494f9c78b2bSJens Axboe * @bio: bio 1495f9c78b2bSJens Axboe * 1496f9c78b2bSJens Axboe * Description: 14974246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 14984246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 14994246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1500fbbaf700SNeilBrown * 1501fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1502fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 150360b6a7e6SEdward Hsieh * last time. 1504f9c78b2bSJens Axboe **/ 15054246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1506f9c78b2bSJens Axboe { 1507ba8c6967SChristoph Hellwig again: 15082b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1509ba8c6967SChristoph Hellwig return; 15107c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 15117c20f116SChristoph Hellwig return; 1512f9c78b2bSJens Axboe 1513a647a524SMing Lei if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED)) 15143caee463SPavel Begunkov rq_qos_done_bio(bdev_get_queue(bio->bi_bdev), bio); 151567b42d0bSJosef Bacik 151660b6a7e6SEdward Hsieh if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 15173caee463SPavel Begunkov trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 151860b6a7e6SEdward Hsieh bio_clear_flag(bio, BIO_TRACE_COMPLETION); 151960b6a7e6SEdward Hsieh } 152060b6a7e6SEdward Hsieh 1521f9c78b2bSJens Axboe /* 1522ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1523ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1524ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1525ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1526ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1527ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1528f9c78b2bSJens Axboe */ 1529f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 153038f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1531ba8c6967SChristoph Hellwig goto again; 1532ba8c6967SChristoph Hellwig } 1533ba8c6967SChristoph Hellwig 15349e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1535b222dd2fSShaohua Li /* release cgroup info */ 1536b222dd2fSShaohua Li bio_uninit(bio); 1537f9c78b2bSJens Axboe if (bio->bi_end_io) 15384246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1539f9c78b2bSJens Axboe } 1540f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1541f9c78b2bSJens Axboe 1542f9c78b2bSJens Axboe /** 1543f9c78b2bSJens Axboe * bio_split - split a bio 1544f9c78b2bSJens Axboe * @bio: bio to split 1545f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1546f9c78b2bSJens Axboe * @gfp: gfp mask 1547f9c78b2bSJens Axboe * @bs: bio set to allocate from 1548f9c78b2bSJens Axboe * 1549f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1550f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1551f9c78b2bSJens Axboe * 1552f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1553dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1554dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1555f9c78b2bSJens Axboe */ 1556f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1557f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1558f9c78b2bSJens Axboe { 1559f341a4d3SMikulas Patocka struct bio *split; 1560f9c78b2bSJens Axboe 1561f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1562f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1563f9c78b2bSJens Axboe 15640512a75bSKeith Busch /* Zone append commands cannot be split */ 15650512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 15660512a75bSKeith Busch return NULL; 15670512a75bSKeith Busch 1568f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1569f9c78b2bSJens Axboe if (!split) 1570f9c78b2bSJens Axboe return NULL; 1571f9c78b2bSJens Axboe 1572f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1573f9c78b2bSJens Axboe 1574f9c78b2bSJens Axboe if (bio_integrity(split)) 1575fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1576f9c78b2bSJens Axboe 1577f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1578f9c78b2bSJens Axboe 1579fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 158020d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1581fbbaf700SNeilBrown 1582f9c78b2bSJens Axboe return split; 1583f9c78b2bSJens Axboe } 1584f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1585f9c78b2bSJens Axboe 1586f9c78b2bSJens Axboe /** 1587f9c78b2bSJens Axboe * bio_trim - trim a bio 1588f9c78b2bSJens Axboe * @bio: bio to trim 1589f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1590f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1591e83502caSChaitanya Kulkarni * 1592e83502caSChaitanya Kulkarni * This function is typically used for bios that are cloned and submitted 1593e83502caSChaitanya Kulkarni * to the underlying device in parts. 1594f9c78b2bSJens Axboe */ 1595e83502caSChaitanya Kulkarni void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1596f9c78b2bSJens Axboe { 1597e83502caSChaitanya Kulkarni if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 1598e83502caSChaitanya Kulkarni offset + size > bio->bi_iter.bi_size)) 1599e83502caSChaitanya Kulkarni return; 1600f9c78b2bSJens Axboe 1601f9c78b2bSJens Axboe size <<= 9; 1602f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1603f9c78b2bSJens Axboe return; 1604f9c78b2bSJens Axboe 1605f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1606f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1607376a78abSDmitry Monakhov 1608376a78abSDmitry Monakhov if (bio_integrity(bio)) 1609fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1610f9c78b2bSJens Axboe } 1611f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1612f9c78b2bSJens Axboe 1613f9c78b2bSJens Axboe /* 1614f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1615f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1616f9c78b2bSJens Axboe */ 16178aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1618f9c78b2bSJens Axboe { 16197a800a20SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1620f9c78b2bSJens Axboe 16218aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1622f9c78b2bSJens Axboe } 1623f9c78b2bSJens Axboe 1624917a38c7SKent Overstreet /* 1625917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1626917a38c7SKent Overstreet * 1627917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1628917a38c7SKent Overstreet * kzalloc()). 1629917a38c7SKent Overstreet */ 1630917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1631f9c78b2bSJens Axboe { 1632be4d234dSJens Axboe bio_alloc_cache_destroy(bs); 1633f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1634f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1635917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1636f9c78b2bSJens Axboe 16378aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 16388aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1639f9c78b2bSJens Axboe 1640f9c78b2bSJens Axboe bioset_integrity_free(bs); 1641917a38c7SKent Overstreet if (bs->bio_slab) 1642f9c78b2bSJens Axboe bio_put_slab(bs); 1643917a38c7SKent Overstreet bs->bio_slab = NULL; 1644917a38c7SKent Overstreet } 1645917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1646f9c78b2bSJens Axboe 1647011067b0SNeilBrown /** 1648917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1649dad08527SKent Overstreet * @bs: pool to initialize 1650917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1651917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1652917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1653917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1654917a38c7SKent Overstreet * 1655dad08527SKent Overstreet * Description: 1656dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1657dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1658dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1659dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1660dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1661dad08527SKent Overstreet * or things will break badly. 1662dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1663dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1664dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1665dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1666dad08527SKent Overstreet * 1667917a38c7SKent Overstreet */ 1668917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1669917a38c7SKent Overstreet unsigned int pool_size, 1670917a38c7SKent Overstreet unsigned int front_pad, 1671917a38c7SKent Overstreet int flags) 1672917a38c7SKent Overstreet { 1673917a38c7SKent Overstreet bs->front_pad = front_pad; 16749f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 16759f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 16769f180e31SMing Lei else 16779f180e31SMing Lei bs->back_pad = 0; 1678917a38c7SKent Overstreet 1679917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1680917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1681917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1682917a38c7SKent Overstreet 168349d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1684917a38c7SKent Overstreet if (!bs->bio_slab) 1685917a38c7SKent Overstreet return -ENOMEM; 1686917a38c7SKent Overstreet 1687917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1688917a38c7SKent Overstreet goto bad; 1689917a38c7SKent Overstreet 1690917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1691917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1692917a38c7SKent Overstreet goto bad; 1693917a38c7SKent Overstreet 1694be4d234dSJens Axboe if (flags & BIOSET_NEED_RESCUER) { 1695be4d234dSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", 1696be4d234dSJens Axboe WQ_MEM_RECLAIM, 0); 1697917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1698917a38c7SKent Overstreet goto bad; 1699be4d234dSJens Axboe } 1700be4d234dSJens Axboe if (flags & BIOSET_PERCPU_CACHE) { 1701be4d234dSJens Axboe bs->cache = alloc_percpu(struct bio_alloc_cache); 1702be4d234dSJens Axboe if (!bs->cache) 1703be4d234dSJens Axboe goto bad; 1704be4d234dSJens Axboe cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1705be4d234dSJens Axboe } 1706917a38c7SKent Overstreet 1707917a38c7SKent Overstreet return 0; 1708917a38c7SKent Overstreet bad: 1709917a38c7SKent Overstreet bioset_exit(bs); 1710917a38c7SKent Overstreet return -ENOMEM; 1711917a38c7SKent Overstreet } 1712917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1713917a38c7SKent Overstreet 171428e89fd9SJens Axboe /* 171528e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 171628e89fd9SJens Axboe * another bio_set. 171728e89fd9SJens Axboe */ 171828e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 171928e89fd9SJens Axboe { 172028e89fd9SJens Axboe int flags; 172128e89fd9SJens Axboe 172228e89fd9SJens Axboe flags = 0; 172328e89fd9SJens Axboe if (src->bvec_pool.min_nr) 172428e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 172528e89fd9SJens Axboe if (src->rescue_workqueue) 172628e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 172728e89fd9SJens Axboe 172828e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 172928e89fd9SJens Axboe } 173028e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 173128e89fd9SJens Axboe 1732be4d234dSJens Axboe /** 1733be4d234dSJens Axboe * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb 1734be4d234dSJens Axboe * @kiocb: kiocb describing the IO 1735b77c88c2SChristoph Hellwig * @bdev: block device to allocate the bio for (can be %NULL) 17360ef47db1SJens Axboe * @nr_vecs: number of iovecs to pre-allocate 1737b77c88c2SChristoph Hellwig * @opf: operation and flags for bio 1738be4d234dSJens Axboe * @bs: bio_set to allocate from 1739be4d234dSJens Axboe * 1740be4d234dSJens Axboe * Description: 1741be4d234dSJens Axboe * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only 1742be4d234dSJens Axboe * used to check if we should dip into the per-cpu bio_set allocation 17433d5b3fbeSJens Axboe * cache. The allocation uses GFP_KERNEL internally. On return, the 17443d5b3fbeSJens Axboe * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio 17453d5b3fbeSJens Axboe * MUST be done from process context, not hard/soft IRQ. 1746be4d234dSJens Axboe * 1747be4d234dSJens Axboe */ 1748b77c88c2SChristoph Hellwig struct bio *bio_alloc_kiocb(struct kiocb *kiocb, struct block_device *bdev, 1749b77c88c2SChristoph Hellwig unsigned short nr_vecs, unsigned int opf, struct bio_set *bs) 1750be4d234dSJens Axboe { 1751be4d234dSJens Axboe struct bio_alloc_cache *cache; 1752be4d234dSJens Axboe struct bio *bio; 1753be4d234dSJens Axboe 1754be4d234dSJens Axboe if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS) 1755b77c88c2SChristoph Hellwig return bio_alloc_bioset(bdev, nr_vecs, opf, GFP_KERNEL, bs); 1756be4d234dSJens Axboe 1757be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, get_cpu()); 1758fcade2ceSJens Axboe if (cache->free_list) { 1759fcade2ceSJens Axboe bio = cache->free_list; 1760fcade2ceSJens Axboe cache->free_list = bio->bi_next; 1761be4d234dSJens Axboe cache->nr--; 1762be4d234dSJens Axboe put_cpu(); 176349add496SChristoph Hellwig bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, 176449add496SChristoph Hellwig nr_vecs, opf); 1765be4d234dSJens Axboe bio->bi_pool = bs; 1766be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1767be4d234dSJens Axboe return bio; 1768be4d234dSJens Axboe } 1769be4d234dSJens Axboe put_cpu(); 1770b77c88c2SChristoph Hellwig bio = bio_alloc_bioset(bdev, nr_vecs, opf, GFP_KERNEL, bs); 1771be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1772be4d234dSJens Axboe return bio; 1773be4d234dSJens Axboe } 1774be4d234dSJens Axboe EXPORT_SYMBOL_GPL(bio_alloc_kiocb); 1775be4d234dSJens Axboe 1776de76fd89SChristoph Hellwig static int __init init_bio(void) 1777f9c78b2bSJens Axboe { 1778f9c78b2bSJens Axboe int i; 1779f9c78b2bSJens Axboe 1780f9c78b2bSJens Axboe bio_integrity_init(); 1781de76fd89SChristoph Hellwig 1782de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1783f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1784f9c78b2bSJens Axboe 1785de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1786de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1787f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1788f9c78b2bSJens Axboe } 1789f9c78b2bSJens Axboe 1790be4d234dSJens Axboe cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1791be4d234dSJens Axboe bio_cpu_dead); 1792be4d234dSJens Axboe 1793f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1794f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1795f9c78b2bSJens Axboe 1796f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1797f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1798f9c78b2bSJens Axboe 1799f9c78b2bSJens Axboe return 0; 1800f9c78b2bSJens Axboe } 1801f9c78b2bSJens Axboe subsys_initcall(init_bio); 1802