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 { 29*fcade2ceSJens 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 */ 2523a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2533a83f467SMing Lei unsigned short max_vecs) 254f9c78b2bSJens Axboe { 255da521626SJens Axboe bio->bi_next = NULL; 256da521626SJens Axboe bio->bi_bdev = NULL; 257da521626SJens Axboe bio->bi_opf = 0; 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; 271da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP_IOCOST 272da521626SJens Axboe bio->bi_iocost_cost = 0; 273da521626SJens Axboe #endif 274da521626SJens Axboe #endif 275da521626SJens Axboe #ifdef CONFIG_BLK_INLINE_ENCRYPTION 276da521626SJens Axboe bio->bi_crypt_context = NULL; 277da521626SJens Axboe #endif 278da521626SJens Axboe #ifdef CONFIG_BLK_DEV_INTEGRITY 279da521626SJens Axboe bio->bi_integrity = NULL; 280da521626SJens Axboe #endif 281da521626SJens Axboe bio->bi_vcnt = 0; 282da521626SJens Axboe 283c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 284dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2853e08773cSChristoph Hellwig bio->bi_cookie = BLK_QC_T_NONE; 2863a83f467SMing Lei 2873a83f467SMing Lei bio->bi_max_vecs = max_vecs; 288da521626SJens Axboe bio->bi_io_vec = table; 289da521626SJens Axboe bio->bi_pool = NULL; 290f9c78b2bSJens Axboe } 291f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 292f9c78b2bSJens Axboe 293f9c78b2bSJens Axboe /** 294f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 295f9c78b2bSJens Axboe * @bio: bio to reset 296f9c78b2bSJens Axboe * 297f9c78b2bSJens Axboe * Description: 298f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 299f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 300f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 301f9c78b2bSJens Axboe * comment in struct bio. 302f9c78b2bSJens Axboe */ 303f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 304f9c78b2bSJens Axboe { 3059ae3b3f5SJens Axboe bio_uninit(bio); 306f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 307c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 308f9c78b2bSJens Axboe } 309f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 310f9c78b2bSJens Axboe 31138f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 312f9c78b2bSJens Axboe { 3134246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3144246a0b6SChristoph Hellwig 3153edf5346SYufen Yu if (bio->bi_status && !parent->bi_status) 3164e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 317f9c78b2bSJens Axboe bio_put(bio); 31838f8baaeSChristoph Hellwig return parent; 31938f8baaeSChristoph Hellwig } 32038f8baaeSChristoph Hellwig 32138f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 32238f8baaeSChristoph Hellwig { 32338f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 324f9c78b2bSJens Axboe } 325f9c78b2bSJens Axboe 326f9c78b2bSJens Axboe /** 327f9c78b2bSJens Axboe * bio_chain - chain bio completions 328f9c78b2bSJens Axboe * @bio: the target bio 3295b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 330f9c78b2bSJens Axboe * 331f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 332f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 333f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 334f9c78b2bSJens Axboe * 335f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 336f9c78b2bSJens Axboe */ 337f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 338f9c78b2bSJens Axboe { 339f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 340f9c78b2bSJens Axboe 341f9c78b2bSJens Axboe bio->bi_private = parent; 342f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 343c4cf5261SJens Axboe bio_inc_remaining(parent); 344f9c78b2bSJens Axboe } 345f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 346f9c78b2bSJens Axboe 347f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 348f9c78b2bSJens Axboe { 349f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 350f9c78b2bSJens Axboe struct bio *bio; 351f9c78b2bSJens Axboe 352f9c78b2bSJens Axboe while (1) { 353f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 354f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 355f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 356f9c78b2bSJens Axboe 357f9c78b2bSJens Axboe if (!bio) 358f9c78b2bSJens Axboe break; 359f9c78b2bSJens Axboe 360ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 361f9c78b2bSJens Axboe } 362f9c78b2bSJens Axboe } 363f9c78b2bSJens Axboe 364f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 365f9c78b2bSJens Axboe { 366f9c78b2bSJens Axboe struct bio_list punt, nopunt; 367f9c78b2bSJens Axboe struct bio *bio; 368f9c78b2bSJens Axboe 36947e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 37047e0fb46SNeilBrown return; 371f9c78b2bSJens Axboe /* 372f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 373f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 374f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 375f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 376f9c78b2bSJens Axboe * our own rescuer would be bad. 377f9c78b2bSJens Axboe * 378f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 379f9c78b2bSJens Axboe * remove from the middle of the list: 380f9c78b2bSJens Axboe */ 381f9c78b2bSJens Axboe 382f9c78b2bSJens Axboe bio_list_init(&punt); 383f9c78b2bSJens Axboe bio_list_init(&nopunt); 384f9c78b2bSJens Axboe 385f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 386f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 387f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 388f9c78b2bSJens Axboe 389f5fe1b51SNeilBrown bio_list_init(&nopunt); 390f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 391f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 392f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 393f9c78b2bSJens Axboe 394f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 395f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 396f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 397f9c78b2bSJens Axboe 398f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 399f9c78b2bSJens Axboe } 400f9c78b2bSJens Axboe 401f9c78b2bSJens Axboe /** 402f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 403519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 404f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 405f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 406f9c78b2bSJens Axboe * 4073175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 408f9c78b2bSJens Axboe * 4093175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4103175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4113175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4123175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4133175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4143175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 415f9c78b2bSJens Axboe * 4163175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4173175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 418ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 419f9c78b2bSJens Axboe * stack overflows. 420f9c78b2bSJens Axboe * 4213175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4223175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 423f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 424f9c78b2bSJens Axboe * thread. 425f9c78b2bSJens Axboe * 426f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 427f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 428ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 429f9c78b2bSJens Axboe * for per bio allocations. 430f9c78b2bSJens Axboe * 4313175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 432f9c78b2bSJens Axboe */ 4330f2e6ab8SChristoph Hellwig struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs, 4347a88fa19SDan Carpenter struct bio_set *bs) 435f9c78b2bSJens Axboe { 436f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 437f9c78b2bSJens Axboe struct bio *bio; 438f9c78b2bSJens Axboe void *p; 439f9c78b2bSJens Axboe 4403175199aSChristoph Hellwig /* should not use nobvec bioset for nr_iovecs > 0 */ 4413175199aSChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) 442f9c78b2bSJens Axboe return NULL; 443f9c78b2bSJens Axboe 444f9c78b2bSJens Axboe /* 4453175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 4463175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 4473175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 448f9c78b2bSJens Axboe * 4493175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 4503175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 4513175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 4523175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 4533175199aSChristoph Hellwig * mempool's reserve. 454f9c78b2bSJens Axboe * 455f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 4563175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 4573175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 4583175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 4593175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 4603175199aSChristoph Hellwig * gfp_flags. 461f9c78b2bSJens Axboe */ 462f5fe1b51SNeilBrown if (current->bio_list && 463f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 46447e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 46547e0fb46SNeilBrown bs->rescue_workqueue) 466d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 467f9c78b2bSJens Axboe 4688aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 469f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 470f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 471f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4728aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 473f9c78b2bSJens Axboe } 474f9c78b2bSJens Axboe if (unlikely(!p)) 475f9c78b2bSJens Axboe return NULL; 476f9c78b2bSJens Axboe 4773175199aSChristoph Hellwig bio = p + bs->front_pad; 4783175199aSChristoph Hellwig if (nr_iovecs > BIO_INLINE_VECS) { 4793175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 480f9c78b2bSJens Axboe 4817a800a20SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 482f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 483f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 484f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4857a800a20SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 486f9c78b2bSJens Axboe } 487f9c78b2bSJens Axboe if (unlikely(!bvl)) 488f9c78b2bSJens Axboe goto err_free; 489f9c78b2bSJens Axboe 4907a800a20SChristoph Hellwig bio_init(bio, bvl, nr_iovecs); 491f9c78b2bSJens Axboe } else if (nr_iovecs) { 4923175199aSChristoph Hellwig bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS); 4933175199aSChristoph Hellwig } else { 4943175199aSChristoph Hellwig bio_init(bio, NULL, 0); 495f9c78b2bSJens Axboe } 496f9c78b2bSJens Axboe 497f9c78b2bSJens Axboe bio->bi_pool = bs; 498f9c78b2bSJens Axboe return bio; 499f9c78b2bSJens Axboe 500f9c78b2bSJens Axboe err_free: 5018aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 502f9c78b2bSJens Axboe return NULL; 503f9c78b2bSJens Axboe } 504f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 505f9c78b2bSJens Axboe 5063175199aSChristoph Hellwig /** 5073175199aSChristoph Hellwig * bio_kmalloc - kmalloc a bio for I/O 5083175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 5093175199aSChristoph Hellwig * @nr_iovecs: number of iovecs to pre-allocate 5103175199aSChristoph Hellwig * 5113175199aSChristoph Hellwig * Use kmalloc to allocate and initialize a bio. 5123175199aSChristoph Hellwig * 5133175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 5143175199aSChristoph Hellwig */ 5150f2e6ab8SChristoph Hellwig struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) 5163175199aSChristoph Hellwig { 5173175199aSChristoph Hellwig struct bio *bio; 5183175199aSChristoph Hellwig 5193175199aSChristoph Hellwig if (nr_iovecs > UIO_MAXIOV) 5203175199aSChristoph Hellwig return NULL; 5213175199aSChristoph Hellwig 5223175199aSChristoph Hellwig bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 5233175199aSChristoph Hellwig if (unlikely(!bio)) 5243175199aSChristoph Hellwig return NULL; 5253175199aSChristoph Hellwig bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs); 5263175199aSChristoph Hellwig bio->bi_pool = NULL; 5273175199aSChristoph Hellwig return bio; 5283175199aSChristoph Hellwig } 5293175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 5303175199aSChristoph Hellwig 5316f822e1bSChristoph Hellwig void zero_fill_bio(struct bio *bio) 532f9c78b2bSJens Axboe { 533f9c78b2bSJens Axboe struct bio_vec bv; 534f9c78b2bSJens Axboe struct bvec_iter iter; 535f9c78b2bSJens Axboe 536ab6c340eSChristoph Hellwig bio_for_each_segment(bv, bio, iter) 537ab6c340eSChristoph Hellwig memzero_bvec(&bv); 538f9c78b2bSJens Axboe } 5396f822e1bSChristoph Hellwig EXPORT_SYMBOL(zero_fill_bio); 540f9c78b2bSJens Axboe 54183c9c547SMing Lei /** 54283c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 54383c9c547SMing Lei * @bio: the bio to be truncated 54483c9c547SMing Lei * @new_size: new size for truncating the bio 54583c9c547SMing Lei * 54683c9c547SMing Lei * Description: 54783c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 54883c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 54983c9c547SMing Lei * be used for handling corner cases, such as bio eod. 55083c9c547SMing Lei */ 5514f7ab09aSChristoph Hellwig static void bio_truncate(struct bio *bio, unsigned new_size) 55285a8ce62SMing Lei { 55385a8ce62SMing Lei struct bio_vec bv; 55485a8ce62SMing Lei struct bvec_iter iter; 55585a8ce62SMing Lei unsigned int done = 0; 55685a8ce62SMing Lei bool truncated = false; 55785a8ce62SMing Lei 55885a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 55985a8ce62SMing Lei return; 56085a8ce62SMing Lei 56183c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 56285a8ce62SMing Lei goto exit; 56385a8ce62SMing Lei 56485a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 56585a8ce62SMing Lei if (done + bv.bv_len > new_size) { 56685a8ce62SMing Lei unsigned offset; 56785a8ce62SMing Lei 56885a8ce62SMing Lei if (!truncated) 56985a8ce62SMing Lei offset = new_size - done; 57085a8ce62SMing Lei else 57185a8ce62SMing Lei offset = 0; 57285a8ce62SMing Lei zero_user(bv.bv_page, offset, bv.bv_len - offset); 57385a8ce62SMing Lei truncated = true; 57485a8ce62SMing Lei } 57585a8ce62SMing Lei done += bv.bv_len; 57685a8ce62SMing Lei } 57785a8ce62SMing Lei 57885a8ce62SMing Lei exit: 57985a8ce62SMing Lei /* 58085a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 58185a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 58285a8ce62SMing Lei * in its .end_bio() callback. 58385a8ce62SMing Lei * 58485a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 58585a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 58685a8ce62SMing Lei */ 58785a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 58885a8ce62SMing Lei } 58985a8ce62SMing Lei 590f9c78b2bSJens Axboe /** 59129125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 59229125ed6SChristoph Hellwig * @bio: bio to truncate 59329125ed6SChristoph Hellwig * 59429125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 59529125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 59629125ed6SChristoph Hellwig * 59729125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 59829125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 59929125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 60029125ed6SChristoph Hellwig * sector" case. 60129125ed6SChristoph Hellwig */ 60229125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 60329125ed6SChristoph Hellwig { 604309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 60529125ed6SChristoph Hellwig 60629125ed6SChristoph Hellwig if (!maxsector) 60729125ed6SChristoph Hellwig return; 60829125ed6SChristoph Hellwig 60929125ed6SChristoph Hellwig /* 61029125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 61129125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 61229125ed6SChristoph Hellwig * an EIO. 61329125ed6SChristoph Hellwig */ 61429125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 61529125ed6SChristoph Hellwig return; 61629125ed6SChristoph Hellwig 61729125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 61829125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 61929125ed6SChristoph Hellwig return; 62029125ed6SChristoph Hellwig 62129125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 62229125ed6SChristoph Hellwig } 62329125ed6SChristoph Hellwig 624be4d234dSJens Axboe #define ALLOC_CACHE_MAX 512 625be4d234dSJens Axboe #define ALLOC_CACHE_SLACK 64 626be4d234dSJens Axboe 627be4d234dSJens Axboe static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 628be4d234dSJens Axboe unsigned int nr) 629be4d234dSJens Axboe { 630be4d234dSJens Axboe unsigned int i = 0; 631be4d234dSJens Axboe struct bio *bio; 632be4d234dSJens Axboe 633*fcade2ceSJens Axboe while ((bio = cache->free_list) != NULL) { 634*fcade2ceSJens Axboe cache->free_list = bio->bi_next; 635be4d234dSJens Axboe cache->nr--; 636be4d234dSJens Axboe bio_free(bio); 637be4d234dSJens Axboe if (++i == nr) 638be4d234dSJens Axboe break; 639be4d234dSJens Axboe } 640be4d234dSJens Axboe } 641be4d234dSJens Axboe 642be4d234dSJens Axboe static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 643be4d234dSJens Axboe { 644be4d234dSJens Axboe struct bio_set *bs; 645be4d234dSJens Axboe 646be4d234dSJens Axboe bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 647be4d234dSJens Axboe if (bs->cache) { 648be4d234dSJens Axboe struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 649be4d234dSJens Axboe 650be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 651be4d234dSJens Axboe } 652be4d234dSJens Axboe return 0; 653be4d234dSJens Axboe } 654be4d234dSJens Axboe 655be4d234dSJens Axboe static void bio_alloc_cache_destroy(struct bio_set *bs) 656be4d234dSJens Axboe { 657be4d234dSJens Axboe int cpu; 658be4d234dSJens Axboe 659be4d234dSJens Axboe if (!bs->cache) 660be4d234dSJens Axboe return; 661be4d234dSJens Axboe 662be4d234dSJens Axboe cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 663be4d234dSJens Axboe for_each_possible_cpu(cpu) { 664be4d234dSJens Axboe struct bio_alloc_cache *cache; 665be4d234dSJens Axboe 666be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, cpu); 667be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 668be4d234dSJens Axboe } 669be4d234dSJens Axboe free_percpu(bs->cache); 670be4d234dSJens Axboe } 671be4d234dSJens Axboe 67229125ed6SChristoph Hellwig /** 673f9c78b2bSJens Axboe * bio_put - release a reference to a bio 674f9c78b2bSJens Axboe * @bio: bio to release reference to 675f9c78b2bSJens Axboe * 676f9c78b2bSJens Axboe * Description: 677f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 6789b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 679f9c78b2bSJens Axboe **/ 680f9c78b2bSJens Axboe void bio_put(struct bio *bio) 681f9c78b2bSJens Axboe { 682be4d234dSJens Axboe if (unlikely(bio_flagged(bio, BIO_REFFED))) { 6839e8c0d0dSChristoph Hellwig BUG_ON(!atomic_read(&bio->__bi_cnt)); 684be4d234dSJens Axboe if (!atomic_dec_and_test(&bio->__bi_cnt)) 685be4d234dSJens Axboe return; 686be4d234dSJens Axboe } 687f9c78b2bSJens Axboe 688be4d234dSJens Axboe if (bio_flagged(bio, BIO_PERCPU_CACHE)) { 689be4d234dSJens Axboe struct bio_alloc_cache *cache; 690be4d234dSJens Axboe 691be4d234dSJens Axboe bio_uninit(bio); 692be4d234dSJens Axboe cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 693*fcade2ceSJens Axboe bio->bi_next = cache->free_list; 694*fcade2ceSJens Axboe cache->free_list = bio; 695be4d234dSJens Axboe if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 696be4d234dSJens Axboe bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 697be4d234dSJens Axboe put_cpu(); 698be4d234dSJens Axboe } else { 699f9c78b2bSJens Axboe bio_free(bio); 700f9c78b2bSJens Axboe } 701dac56212SJens Axboe } 702f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 703f9c78b2bSJens Axboe 704f9c78b2bSJens Axboe /** 705f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 706f9c78b2bSJens Axboe * @bio: destination bio 707f9c78b2bSJens Axboe * @bio_src: bio to clone 708f9c78b2bSJens Axboe * 709f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 710f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 711f9c78b2bSJens Axboe * bio will be one. 712f9c78b2bSJens Axboe * 713f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 714f9c78b2bSJens Axboe */ 715f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 716f9c78b2bSJens Axboe { 7177a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs); 718f9c78b2bSJens Axboe 719f9c78b2bSJens Axboe /* 720309dca30SChristoph Hellwig * most users will be overriding ->bi_bdev with a new target, 721f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 722f9c78b2bSJens Axboe */ 723309dca30SChristoph Hellwig bio->bi_bdev = bio_src->bi_bdev; 724b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 725111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 726111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 72746bbf653SChristoph Hellwig if (bio_flagged(bio_src, BIO_REMAPPED)) 72846bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 7291eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 730ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 731cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 732f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 733f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 73420bd723eSPaolo Valente 735db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 736e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 737f9c78b2bSJens Axboe } 738f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 739f9c78b2bSJens Axboe 740f9c78b2bSJens Axboe /** 741f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 742f9c78b2bSJens Axboe * @bio: bio to clone 743f9c78b2bSJens Axboe * @gfp_mask: allocation priority 744f9c78b2bSJens Axboe * @bs: bio_set to allocate from 745f9c78b2bSJens Axboe * 746f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 747f9c78b2bSJens Axboe */ 748f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 749f9c78b2bSJens Axboe { 750f9c78b2bSJens Axboe struct bio *b; 751f9c78b2bSJens Axboe 752f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 753f9c78b2bSJens Axboe if (!b) 754f9c78b2bSJens Axboe return NULL; 755f9c78b2bSJens Axboe 756f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 757f9c78b2bSJens Axboe 75807560151SEric Biggers if (bio_crypt_clone(b, bio, gfp_mask) < 0) 75907560151SEric Biggers goto err_put; 760a892c8d5SSatya Tangirala 76107560151SEric Biggers if (bio_integrity(bio) && 76207560151SEric Biggers bio_integrity_clone(b, bio, gfp_mask) < 0) 76307560151SEric Biggers goto err_put; 764f9c78b2bSJens Axboe 765f9c78b2bSJens Axboe return b; 76607560151SEric Biggers 76707560151SEric Biggers err_put: 76807560151SEric Biggers bio_put(b); 76907560151SEric Biggers return NULL; 770f9c78b2bSJens Axboe } 771f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 772f9c78b2bSJens Axboe 7735cbd28e3SChristoph Hellwig const char *bio_devname(struct bio *bio, char *buf) 7745cbd28e3SChristoph Hellwig { 775309dca30SChristoph Hellwig return bdevname(bio->bi_bdev, buf); 7765cbd28e3SChristoph Hellwig } 7775cbd28e3SChristoph Hellwig EXPORT_SYMBOL(bio_devname); 7785cbd28e3SChristoph Hellwig 7799a6083beSChristoph Hellwig /** 7809a6083beSChristoph Hellwig * bio_full - check if the bio is full 7819a6083beSChristoph Hellwig * @bio: bio to check 7829a6083beSChristoph Hellwig * @len: length of one segment to be added 7839a6083beSChristoph Hellwig * 7849a6083beSChristoph Hellwig * Return true if @bio is full and one segment with @len bytes can't be 7859a6083beSChristoph Hellwig * added to the bio, otherwise return false 7869a6083beSChristoph Hellwig */ 7879a6083beSChristoph Hellwig static inline bool bio_full(struct bio *bio, unsigned len) 7889a6083beSChristoph Hellwig { 7899a6083beSChristoph Hellwig if (bio->bi_vcnt >= bio->bi_max_vecs) 7909a6083beSChristoph Hellwig return true; 7919a6083beSChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) 7929a6083beSChristoph Hellwig return true; 7939a6083beSChristoph Hellwig return false; 7949a6083beSChristoph Hellwig } 7959a6083beSChristoph Hellwig 7965919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 7975919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 798ff896738SChristoph Hellwig bool *same_page) 7995919482eSMing Lei { 800d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 801d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 8025919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 8035919482eSMing Lei 8045919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 8055919482eSMing Lei return false; 8065919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 8075919482eSMing Lei return false; 80852d52d1cSChristoph Hellwig 809ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 810d8166519SMatthew Wilcox (Oracle) if (*same_page) 8115919482eSMing Lei return true; 812d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 8135919482eSMing Lei } 8145919482eSMing Lei 8159774b391SChristoph Hellwig /** 8169774b391SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 8179774b391SChristoph Hellwig * @bio: destination bio 8189774b391SChristoph Hellwig * @page: start page to add 8199774b391SChristoph Hellwig * @len: length of the data to add 8209774b391SChristoph Hellwig * @off: offset of the data relative to @page 8219774b391SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 8229774b391SChristoph Hellwig * 8239774b391SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 8249774b391SChristoph Hellwig * useful optimisation for file systems with a block size smaller than the 8259774b391SChristoph Hellwig * page size. 8269774b391SChristoph Hellwig * 8279774b391SChristoph Hellwig * Warn if (@len, @off) crosses pages in case that @same_page is true. 8289774b391SChristoph Hellwig * 8299774b391SChristoph Hellwig * Return %true on success or %false on failure. 8309774b391SChristoph Hellwig */ 8319774b391SChristoph Hellwig static bool __bio_try_merge_page(struct bio *bio, struct page *page, 8329774b391SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 8339774b391SChristoph Hellwig { 8349774b391SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8359774b391SChristoph Hellwig return false; 8369774b391SChristoph Hellwig 8379774b391SChristoph Hellwig if (bio->bi_vcnt > 0) { 8389774b391SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 8399774b391SChristoph Hellwig 8409774b391SChristoph Hellwig if (page_is_mergeable(bv, page, len, off, same_page)) { 8419774b391SChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) { 8429774b391SChristoph Hellwig *same_page = false; 8439774b391SChristoph Hellwig return false; 8449774b391SChristoph Hellwig } 8459774b391SChristoph Hellwig bv->bv_len += len; 8469774b391SChristoph Hellwig bio->bi_iter.bi_size += len; 8479774b391SChristoph Hellwig return true; 8489774b391SChristoph Hellwig } 8499774b391SChristoph Hellwig } 8509774b391SChristoph Hellwig return false; 8519774b391SChristoph Hellwig } 8529774b391SChristoph Hellwig 853e4581105SChristoph Hellwig /* 854e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 855e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 856e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 857e4581105SChristoph Hellwig */ 858e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 859e4581105SChristoph Hellwig struct page *page, unsigned len, 860e4581105SChristoph Hellwig unsigned offset, bool *same_page) 861489fbbcbSMing Lei { 862384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 863489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 864489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 865489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 866489fbbcbSMing Lei 867489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 868489fbbcbSMing Lei return false; 869489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 870489fbbcbSMing Lei return false; 871384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 872489fbbcbSMing Lei } 873489fbbcbSMing Lei 874f4595875SShaohua Li /** 875e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 876c66a14d0SKent Overstreet * @q: the target queue 877c66a14d0SKent Overstreet * @bio: destination bio 878c66a14d0SKent Overstreet * @page: page to add 879c66a14d0SKent Overstreet * @len: vec entry length 880c66a14d0SKent Overstreet * @offset: vec entry offset 881e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 882e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 883f9c78b2bSJens Axboe * 884e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 885e4581105SChristoph Hellwig * and gap limitations. 886f9c78b2bSJens Axboe */ 887e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 88819047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 889e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 890f9c78b2bSJens Axboe { 891f9c78b2bSJens Axboe struct bio_vec *bvec; 892f9c78b2bSJens Axboe 893e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 894f9c78b2bSJens Axboe return 0; 895f9c78b2bSJens Axboe 896e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 897f9c78b2bSJens Axboe return 0; 898f9c78b2bSJens Axboe 899f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 900e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 901384209cdSChristoph Hellwig return len; 902320ea869SChristoph Hellwig 903320ea869SChristoph Hellwig /* 904320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 905320ea869SChristoph Hellwig * would create a gap, disallow it. 906320ea869SChristoph Hellwig */ 907384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 908320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 909320ea869SChristoph Hellwig return 0; 910f9c78b2bSJens Axboe } 911f9c78b2bSJens Axboe 91279d08f89SMing Lei if (bio_full(bio, len)) 913f9c78b2bSJens Axboe return 0; 914f9c78b2bSJens Axboe 91514ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 916489fbbcbSMing Lei return 0; 917489fbbcbSMing Lei 918f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 919f9c78b2bSJens Axboe bvec->bv_page = page; 920f9c78b2bSJens Axboe bvec->bv_len = len; 921f9c78b2bSJens Axboe bvec->bv_offset = offset; 922fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 923dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 924f9c78b2bSJens Axboe return len; 925f9c78b2bSJens Axboe } 92619047087SMing Lei 927e4581105SChristoph Hellwig /** 928e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 929e4581105SChristoph Hellwig * @q: the target queue 930e4581105SChristoph Hellwig * @bio: destination bio 931e4581105SChristoph Hellwig * @page: page to add 932e4581105SChristoph Hellwig * @len: vec entry length 933e4581105SChristoph Hellwig * @offset: vec entry offset 934e4581105SChristoph Hellwig * 935e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 936e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 937e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 938e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 939e4581105SChristoph Hellwig * 940e4581105SChristoph Hellwig * This should only be used by passthrough bios. 941e4581105SChristoph Hellwig */ 94219047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 94319047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 94419047087SMing Lei { 945d1916c86SChristoph Hellwig bool same_page = false; 946e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 947e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 94819047087SMing Lei } 949f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 950f9c78b2bSJens Axboe 951f9c78b2bSJens Axboe /** 952ae29333fSJohannes Thumshirn * bio_add_zone_append_page - attempt to add page to zone-append bio 953ae29333fSJohannes Thumshirn * @bio: destination bio 954ae29333fSJohannes Thumshirn * @page: page to add 955ae29333fSJohannes Thumshirn * @len: vec entry length 956ae29333fSJohannes Thumshirn * @offset: vec entry offset 957ae29333fSJohannes Thumshirn * 958ae29333fSJohannes Thumshirn * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 959ae29333fSJohannes Thumshirn * for a zone-append request. This can fail for a number of reasons, such as the 960ae29333fSJohannes Thumshirn * bio being full or the target block device is not a zoned block device or 961ae29333fSJohannes Thumshirn * other limitations of the target block device. The target block device must 962ae29333fSJohannes Thumshirn * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 963ae29333fSJohannes Thumshirn * to an empty bio. 964ae29333fSJohannes Thumshirn * 965ae29333fSJohannes Thumshirn * Returns: number of bytes added to the bio, or 0 in case of a failure. 966ae29333fSJohannes Thumshirn */ 967ae29333fSJohannes Thumshirn int bio_add_zone_append_page(struct bio *bio, struct page *page, 968ae29333fSJohannes Thumshirn unsigned int len, unsigned int offset) 969ae29333fSJohannes Thumshirn { 9703caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 971ae29333fSJohannes Thumshirn bool same_page = false; 972ae29333fSJohannes Thumshirn 973ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 974ae29333fSJohannes Thumshirn return 0; 975ae29333fSJohannes Thumshirn 976ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 977ae29333fSJohannes Thumshirn return 0; 978ae29333fSJohannes Thumshirn 979ae29333fSJohannes Thumshirn return bio_add_hw_page(q, bio, page, len, offset, 980ae29333fSJohannes Thumshirn queue_max_zone_append_sectors(q), &same_page); 981ae29333fSJohannes Thumshirn } 982ae29333fSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 983ae29333fSJohannes Thumshirn 984ae29333fSJohannes Thumshirn /** 985551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 9860aa69fd3SChristoph Hellwig * @bio: destination bio 987551879a4SMing Lei * @page: start page to add 988551879a4SMing Lei * @len: length of the data to add, may cross pages 989551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 9900aa69fd3SChristoph Hellwig * 9910aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 9920aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 9930aa69fd3SChristoph Hellwig */ 9940aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 9950aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 9960aa69fd3SChristoph Hellwig { 9970aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 9980aa69fd3SChristoph Hellwig 9990aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 100079d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 10010aa69fd3SChristoph Hellwig 10020aa69fd3SChristoph Hellwig bv->bv_page = page; 10030aa69fd3SChristoph Hellwig bv->bv_offset = off; 10040aa69fd3SChristoph Hellwig bv->bv_len = len; 10050aa69fd3SChristoph Hellwig 10060aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 10070aa69fd3SChristoph Hellwig bio->bi_vcnt++; 1008b8e24a93SJohannes Weiner 1009b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 1010b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 10110aa69fd3SChristoph Hellwig } 10120aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 10130aa69fd3SChristoph Hellwig 10140aa69fd3SChristoph Hellwig /** 1015551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 1016f9c78b2bSJens Axboe * @bio: destination bio 1017551879a4SMing Lei * @page: start page to add 1018551879a4SMing Lei * @len: vec entry length, may cross pages 1019551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 1020f9c78b2bSJens Axboe * 1021551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 1022c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1023f9c78b2bSJens Axboe */ 1024c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 1025c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 1026f9c78b2bSJens Axboe { 1027ff896738SChristoph Hellwig bool same_page = false; 1028ff896738SChristoph Hellwig 1029ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 103079d08f89SMing Lei if (bio_full(bio, len)) 1031c66a14d0SKent Overstreet return 0; 10320aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 1033c66a14d0SKent Overstreet } 1034c66a14d0SKent Overstreet return len; 1035f9c78b2bSJens Axboe } 1036f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 1037f9c78b2bSJens Axboe 1038c809084aSPavel Begunkov void __bio_release_pages(struct bio *bio, bool mark_dirty) 10397321ecbfSChristoph Hellwig { 10407321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 10417321ecbfSChristoph Hellwig struct bio_vec *bvec; 10427321ecbfSChristoph Hellwig 1043d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1044d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 1045d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 10467321ecbfSChristoph Hellwig put_page(bvec->bv_page); 10477321ecbfSChristoph Hellwig } 1048d241a95fSChristoph Hellwig } 1049c809084aSPavel Begunkov EXPORT_SYMBOL_GPL(__bio_release_pages); 10507321ecbfSChristoph Hellwig 10511bb6b810SPavel Begunkov void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 10526d0c48aeSJens Axboe { 1053fa5fa8ecSPavel Begunkov size_t size = iov_iter_count(iter); 1054fa5fa8ecSPavel Begunkov 10557a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_max_vecs); 10566d0c48aeSJens Axboe 1057fa5fa8ecSPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1058fa5fa8ecSPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1059fa5fa8ecSPavel Begunkov size_t max_sectors = queue_max_zone_append_sectors(q); 1060fa5fa8ecSPavel Begunkov 1061fa5fa8ecSPavel Begunkov size = min(size, max_sectors << SECTOR_SHIFT); 1062fa5fa8ecSPavel Begunkov } 1063fa5fa8ecSPavel Begunkov 1064c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 1065c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1066c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 1067fa5fa8ecSPavel Begunkov bio->bi_iter.bi_size = size; 1068ed97ce5eSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 1069977be012SChristoph Hellwig bio_set_flag(bio, BIO_CLONED); 10707de55b7dSJohannes Thumshirn } 10716d0c48aeSJens Axboe 1072d9cf3bd5SPavel Begunkov static void bio_put_pages(struct page **pages, size_t size, size_t off) 1073d9cf3bd5SPavel Begunkov { 1074d9cf3bd5SPavel Begunkov size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE); 1075d9cf3bd5SPavel Begunkov 1076d9cf3bd5SPavel Begunkov for (i = 0; i < nr; i++) 1077d9cf3bd5SPavel Begunkov put_page(pages[i]); 1078d9cf3bd5SPavel Begunkov } 1079d9cf3bd5SPavel Begunkov 1080576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1081576ed913SChristoph Hellwig 10822cefe4dbSKent Overstreet /** 108317d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 10842cefe4dbSKent Overstreet * @bio: bio to add pages to 10852cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 10862cefe4dbSKent Overstreet * 108717d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 10882cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 108917d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 10903cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 10912cefe4dbSKent Overstreet */ 109217d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 10932cefe4dbSKent Overstreet { 1094576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1095576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 10962cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 10972cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 109845691804SChristoph Hellwig bool same_page = false; 1099576ed913SChristoph Hellwig ssize_t size, left; 1100576ed913SChristoph Hellwig unsigned len, i; 1101b403ea24SMartin Wilck size_t offset; 1102576ed913SChristoph Hellwig 1103576ed913SChristoph Hellwig /* 1104576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 1105576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 1106576ed913SChristoph Hellwig * without overwriting the temporary page array. 1107576ed913SChristoph Hellwig */ 1108576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1109576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 11102cefe4dbSKent Overstreet 111135c820e7SJens Axboe size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 11122cefe4dbSKent Overstreet if (unlikely(size <= 0)) 11132cefe4dbSKent Overstreet return size ? size : -EFAULT; 11142cefe4dbSKent Overstreet 1115576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1116576ed913SChristoph Hellwig struct page *page = pages[i]; 11172cefe4dbSKent Overstreet 1118576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 111945691804SChristoph Hellwig 112045691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 112145691804SChristoph Hellwig if (same_page) 112245691804SChristoph Hellwig put_page(page); 112345691804SChristoph Hellwig } else { 1124d9cf3bd5SPavel Begunkov if (WARN_ON_ONCE(bio_full(bio, len))) { 1125d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 1126576ed913SChristoph Hellwig return -EINVAL; 1127d9cf3bd5SPavel Begunkov } 112845691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 112945691804SChristoph Hellwig } 1130576ed913SChristoph Hellwig offset = 0; 11312cefe4dbSKent Overstreet } 11322cefe4dbSKent Overstreet 11332cefe4dbSKent Overstreet iov_iter_advance(iter, size); 11342cefe4dbSKent Overstreet return 0; 11352cefe4dbSKent Overstreet } 113617d51b10SMartin Wilck 11370512a75bSKeith Busch static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 11380512a75bSKeith Busch { 11390512a75bSKeith Busch unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 11400512a75bSKeith Busch unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 11413caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 11420512a75bSKeith Busch unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 11430512a75bSKeith Busch struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 11440512a75bSKeith Busch struct page **pages = (struct page **)bv; 11450512a75bSKeith Busch ssize_t size, left; 11460512a75bSKeith Busch unsigned len, i; 11470512a75bSKeith Busch size_t offset; 11484977d121SNaohiro Aota int ret = 0; 11490512a75bSKeith Busch 11500512a75bSKeith Busch if (WARN_ON_ONCE(!max_append_sectors)) 11510512a75bSKeith Busch return 0; 11520512a75bSKeith Busch 11530512a75bSKeith Busch /* 11540512a75bSKeith Busch * Move page array up in the allocated memory for the bio vecs as far as 11550512a75bSKeith Busch * possible so that we can start filling biovecs from the beginning 11560512a75bSKeith Busch * without overwriting the temporary page array. 11570512a75bSKeith Busch */ 11580512a75bSKeith Busch BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 11590512a75bSKeith Busch pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 11600512a75bSKeith Busch 11610512a75bSKeith Busch size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 11620512a75bSKeith Busch if (unlikely(size <= 0)) 11630512a75bSKeith Busch return size ? size : -EFAULT; 11640512a75bSKeith Busch 11650512a75bSKeith Busch for (left = size, i = 0; left > 0; left -= len, i++) { 11660512a75bSKeith Busch struct page *page = pages[i]; 11670512a75bSKeith Busch bool same_page = false; 11680512a75bSKeith Busch 11690512a75bSKeith Busch len = min_t(size_t, PAGE_SIZE - offset, left); 11700512a75bSKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 11714977d121SNaohiro Aota max_append_sectors, &same_page) != len) { 1172d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 11734977d121SNaohiro Aota ret = -EINVAL; 11744977d121SNaohiro Aota break; 11754977d121SNaohiro Aota } 11760512a75bSKeith Busch if (same_page) 11770512a75bSKeith Busch put_page(page); 11780512a75bSKeith Busch offset = 0; 11790512a75bSKeith Busch } 11800512a75bSKeith Busch 11814977d121SNaohiro Aota iov_iter_advance(iter, size - left); 11824977d121SNaohiro Aota return ret; 11830512a75bSKeith Busch } 11840512a75bSKeith Busch 118517d51b10SMartin Wilck /** 11866d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 118717d51b10SMartin Wilck * @bio: bio to add pages to 11886d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 118917d51b10SMartin Wilck * 11906d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 11916d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 11926d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1193c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1194c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1195c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1196c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1197c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1198c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 11996d0c48aeSJens Axboe * 120017d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 12015cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 12026d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 12036d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 12040cf41e5eSPavel Begunkov * 12050cf41e5eSPavel Begunkov * It's intended for direct IO, so doesn't do PSI tracking, the caller is 12060cf41e5eSPavel Begunkov * responsible for setting BIO_WORKINGSET if necessary. 120717d51b10SMartin Wilck */ 120817d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 120917d51b10SMartin Wilck { 1210c42bca92SPavel Begunkov int ret = 0; 121114eacf12SChristoph Hellwig 1212c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1213fa5fa8ecSPavel Begunkov bio_iov_bvec_set(bio, iter); 1214fa5fa8ecSPavel Begunkov iov_iter_advance(iter, bio->bi_iter.bi_size); 1215fa5fa8ecSPavel Begunkov return 0; 121686004515SChristoph Hellwig } 121717d51b10SMartin Wilck 121817d51b10SMartin Wilck do { 1219c42bca92SPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) 12200512a75bSKeith Busch ret = __bio_iov_append_get_pages(bio, iter); 12216d0c48aeSJens Axboe else 12226d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 122379d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 122417d51b10SMartin Wilck 12250cf41e5eSPavel Begunkov /* don't account direct I/O as memory stall */ 12260cf41e5eSPavel Begunkov bio_clear_flag(bio, BIO_WORKINGSET); 122714eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 122817d51b10SMartin Wilck } 122929b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 12302cefe4dbSKent Overstreet 12314246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1232f9c78b2bSJens Axboe { 123365e53aabSChristoph Hellwig complete(bio->bi_private); 1234f9c78b2bSJens Axboe } 1235f9c78b2bSJens Axboe 1236f9c78b2bSJens Axboe /** 1237f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1238f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1239f9c78b2bSJens Axboe * 1240f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1241f9c78b2bSJens Axboe * bio_endio() on failure. 12423d289d68SJan Kara * 12433d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 12443d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 12453d289d68SJan Kara * on his own. 1246f9c78b2bSJens Axboe */ 12474e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1248f9c78b2bSJens Axboe { 1249309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1250309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1251de6a78b6SMing Lei unsigned long hang_check; 1252f9c78b2bSJens Axboe 125365e53aabSChristoph Hellwig bio->bi_private = &done; 1254f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 12551eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 12564e49ea4aSMike Christie submit_bio(bio); 1257de6a78b6SMing Lei 1258de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1259de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1260de6a78b6SMing Lei if (hang_check) 1261de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1262de6a78b6SMing Lei hang_check * (HZ/2))) 1263de6a78b6SMing Lei ; 1264de6a78b6SMing Lei else 126565e53aabSChristoph Hellwig wait_for_completion_io(&done); 1266f9c78b2bSJens Axboe 126765e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1268f9c78b2bSJens Axboe } 1269f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1270f9c78b2bSJens Axboe 1271d4aa57a1SJens Axboe void __bio_advance(struct bio *bio, unsigned bytes) 1272f9c78b2bSJens Axboe { 1273f9c78b2bSJens Axboe if (bio_integrity(bio)) 1274f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1275f9c78b2bSJens Axboe 1276a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1277f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1278f9c78b2bSJens Axboe } 1279d4aa57a1SJens Axboe EXPORT_SYMBOL(__bio_advance); 1280f9c78b2bSJens Axboe 128145db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 128245db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1283f9c78b2bSJens Axboe { 128445db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 1285f8b679a0SChristoph Hellwig struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1286f8b679a0SChristoph Hellwig struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1287f8b679a0SChristoph Hellwig unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1288f8b679a0SChristoph Hellwig void *src_buf; 128945db54d5SKent Overstreet 1290f8b679a0SChristoph Hellwig src_buf = bvec_kmap_local(&src_bv); 1291f8b679a0SChristoph Hellwig memcpy_to_bvec(&dst_bv, src_buf); 1292f8b679a0SChristoph Hellwig kunmap_local(src_buf); 12936e6e811dSKent Overstreet 129422b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 129522b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 129645db54d5SKent Overstreet } 129745db54d5SKent Overstreet } 129845db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 129945db54d5SKent Overstreet 130045db54d5SKent Overstreet /** 130145db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 130245db54d5SKent Overstreet * @src: source bio 130345db54d5SKent Overstreet * @dst: destination bio 130445db54d5SKent Overstreet * 130545db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 130645db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 130745db54d5SKent Overstreet */ 130845db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 130945db54d5SKent Overstreet { 131045db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 131145db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 131245db54d5SKent Overstreet 131345db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 131445db54d5SKent Overstreet } 131545db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 131645db54d5SKent Overstreet 1317491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 13181dfa0f68SChristoph Hellwig { 13191dfa0f68SChristoph Hellwig struct bio_vec *bvec; 13206dc4f100SMing Lei struct bvec_iter_all iter_all; 13211dfa0f68SChristoph Hellwig 13222b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 13231dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 13241dfa0f68SChristoph Hellwig } 1325491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 13261dfa0f68SChristoph Hellwig 1327f9c78b2bSJens Axboe /* 1328f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1329f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1330f9c78b2bSJens Axboe * 1331f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1332f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1333f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1334f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1335f9c78b2bSJens Axboe * in process context. 1336f9c78b2bSJens Axboe * 1337f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1338f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1339f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1340f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1341f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1342f9c78b2bSJens Axboe * 1343f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1344f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1345f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1346f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1347f9c78b2bSJens Axboe * pagecache. 1348f9c78b2bSJens Axboe * 1349f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1350f9c78b2bSJens Axboe * deferred bio dirtying paths. 1351f9c78b2bSJens Axboe */ 1352f9c78b2bSJens Axboe 1353f9c78b2bSJens Axboe /* 1354f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1355f9c78b2bSJens Axboe */ 1356f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1357f9c78b2bSJens Axboe { 1358f9c78b2bSJens Axboe struct bio_vec *bvec; 13596dc4f100SMing Lei struct bvec_iter_all iter_all; 1360f9c78b2bSJens Axboe 13612b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 13623bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 13633bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1364f9c78b2bSJens Axboe } 1365f9c78b2bSJens Axboe } 1366f9c78b2bSJens Axboe 1367f9c78b2bSJens Axboe /* 1368f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1369f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1370f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 137124d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1372f9c78b2bSJens Axboe * 1373f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1374ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1375ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1376f9c78b2bSJens Axboe */ 1377f9c78b2bSJens Axboe 1378f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1379f9c78b2bSJens Axboe 1380f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1381f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1382f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1383f9c78b2bSJens Axboe 1384f9c78b2bSJens Axboe /* 1385f9c78b2bSJens Axboe * This runs in process context 1386f9c78b2bSJens Axboe */ 1387f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1388f9c78b2bSJens Axboe { 138924d5493fSChristoph Hellwig struct bio *bio, *next; 1390f9c78b2bSJens Axboe 139124d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 139224d5493fSChristoph Hellwig next = bio_dirty_list; 1393f9c78b2bSJens Axboe bio_dirty_list = NULL; 139424d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1395f9c78b2bSJens Axboe 139624d5493fSChristoph Hellwig while ((bio = next) != NULL) { 139724d5493fSChristoph Hellwig next = bio->bi_private; 1398f9c78b2bSJens Axboe 1399d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1400f9c78b2bSJens Axboe bio_put(bio); 1401f9c78b2bSJens Axboe } 1402f9c78b2bSJens Axboe } 1403f9c78b2bSJens Axboe 1404f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1405f9c78b2bSJens Axboe { 1406f9c78b2bSJens Axboe struct bio_vec *bvec; 140724d5493fSChristoph Hellwig unsigned long flags; 14086dc4f100SMing Lei struct bvec_iter_all iter_all; 1409f9c78b2bSJens Axboe 14102b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 141124d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 141224d5493fSChristoph Hellwig goto defer; 1413f9c78b2bSJens Axboe } 1414f9c78b2bSJens Axboe 1415d241a95fSChristoph Hellwig bio_release_pages(bio, false); 141624d5493fSChristoph Hellwig bio_put(bio); 141724d5493fSChristoph Hellwig return; 141824d5493fSChristoph Hellwig defer: 1419f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1420f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1421f9c78b2bSJens Axboe bio_dirty_list = bio; 1422f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1423f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1424f9c78b2bSJens Axboe } 1425f9c78b2bSJens Axboe 1426c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1427c4cf5261SJens Axboe { 1428c4cf5261SJens Axboe /* 1429c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1430c4cf5261SJens Axboe * we always end io on the first invocation. 1431c4cf5261SJens Axboe */ 1432c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1433c4cf5261SJens Axboe return true; 1434c4cf5261SJens Axboe 1435c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1436c4cf5261SJens Axboe 1437326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1438b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1439c4cf5261SJens Axboe return true; 1440326e1dbbSMike Snitzer } 1441c4cf5261SJens Axboe 1442c4cf5261SJens Axboe return false; 1443c4cf5261SJens Axboe } 1444c4cf5261SJens Axboe 1445f9c78b2bSJens Axboe /** 1446f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1447f9c78b2bSJens Axboe * @bio: bio 1448f9c78b2bSJens Axboe * 1449f9c78b2bSJens Axboe * Description: 14504246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 14514246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 14524246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1453fbbaf700SNeilBrown * 1454fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1455fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 145660b6a7e6SEdward Hsieh * last time. 1457f9c78b2bSJens Axboe **/ 14584246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1459f9c78b2bSJens Axboe { 1460ba8c6967SChristoph Hellwig again: 14612b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1462ba8c6967SChristoph Hellwig return; 14637c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 14647c20f116SChristoph Hellwig return; 1465f9c78b2bSJens Axboe 1466a647a524SMing Lei if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED)) 14673caee463SPavel Begunkov rq_qos_done_bio(bdev_get_queue(bio->bi_bdev), bio); 146867b42d0bSJosef Bacik 146960b6a7e6SEdward Hsieh if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 14703caee463SPavel Begunkov trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 147160b6a7e6SEdward Hsieh bio_clear_flag(bio, BIO_TRACE_COMPLETION); 147260b6a7e6SEdward Hsieh } 147360b6a7e6SEdward Hsieh 1474f9c78b2bSJens Axboe /* 1475ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1476ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1477ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1478ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1479ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1480ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1481f9c78b2bSJens Axboe */ 1482f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 148338f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1484ba8c6967SChristoph Hellwig goto again; 1485ba8c6967SChristoph Hellwig } 1486ba8c6967SChristoph Hellwig 14879e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1488b222dd2fSShaohua Li /* release cgroup info */ 1489b222dd2fSShaohua Li bio_uninit(bio); 1490f9c78b2bSJens Axboe if (bio->bi_end_io) 14914246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1492f9c78b2bSJens Axboe } 1493f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1494f9c78b2bSJens Axboe 1495f9c78b2bSJens Axboe /** 1496f9c78b2bSJens Axboe * bio_split - split a bio 1497f9c78b2bSJens Axboe * @bio: bio to split 1498f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1499f9c78b2bSJens Axboe * @gfp: gfp mask 1500f9c78b2bSJens Axboe * @bs: bio set to allocate from 1501f9c78b2bSJens Axboe * 1502f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1503f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1504f9c78b2bSJens Axboe * 1505f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1506dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1507dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1508f9c78b2bSJens Axboe */ 1509f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1510f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1511f9c78b2bSJens Axboe { 1512f341a4d3SMikulas Patocka struct bio *split; 1513f9c78b2bSJens Axboe 1514f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1515f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1516f9c78b2bSJens Axboe 15170512a75bSKeith Busch /* Zone append commands cannot be split */ 15180512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 15190512a75bSKeith Busch return NULL; 15200512a75bSKeith Busch 1521f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1522f9c78b2bSJens Axboe if (!split) 1523f9c78b2bSJens Axboe return NULL; 1524f9c78b2bSJens Axboe 1525f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1526f9c78b2bSJens Axboe 1527f9c78b2bSJens Axboe if (bio_integrity(split)) 1528fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1529f9c78b2bSJens Axboe 1530f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1531f9c78b2bSJens Axboe 1532fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 153320d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1534fbbaf700SNeilBrown 1535f9c78b2bSJens Axboe return split; 1536f9c78b2bSJens Axboe } 1537f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1538f9c78b2bSJens Axboe 1539f9c78b2bSJens Axboe /** 1540f9c78b2bSJens Axboe * bio_trim - trim a bio 1541f9c78b2bSJens Axboe * @bio: bio to trim 1542f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1543f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1544e83502caSChaitanya Kulkarni * 1545e83502caSChaitanya Kulkarni * This function is typically used for bios that are cloned and submitted 1546e83502caSChaitanya Kulkarni * to the underlying device in parts. 1547f9c78b2bSJens Axboe */ 1548e83502caSChaitanya Kulkarni void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1549f9c78b2bSJens Axboe { 1550e83502caSChaitanya Kulkarni if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 1551e83502caSChaitanya Kulkarni offset + size > bio->bi_iter.bi_size)) 1552e83502caSChaitanya Kulkarni return; 1553f9c78b2bSJens Axboe 1554f9c78b2bSJens Axboe size <<= 9; 1555f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1556f9c78b2bSJens Axboe return; 1557f9c78b2bSJens Axboe 1558f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1559f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1560376a78abSDmitry Monakhov 1561376a78abSDmitry Monakhov if (bio_integrity(bio)) 1562fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1563f9c78b2bSJens Axboe } 1564f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1565f9c78b2bSJens Axboe 1566f9c78b2bSJens Axboe /* 1567f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1568f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1569f9c78b2bSJens Axboe */ 15708aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1571f9c78b2bSJens Axboe { 15727a800a20SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1573f9c78b2bSJens Axboe 15748aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1575f9c78b2bSJens Axboe } 1576f9c78b2bSJens Axboe 1577917a38c7SKent Overstreet /* 1578917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1579917a38c7SKent Overstreet * 1580917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1581917a38c7SKent Overstreet * kzalloc()). 1582917a38c7SKent Overstreet */ 1583917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1584f9c78b2bSJens Axboe { 1585be4d234dSJens Axboe bio_alloc_cache_destroy(bs); 1586f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1587f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1588917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1589f9c78b2bSJens Axboe 15908aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 15918aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1592f9c78b2bSJens Axboe 1593f9c78b2bSJens Axboe bioset_integrity_free(bs); 1594917a38c7SKent Overstreet if (bs->bio_slab) 1595f9c78b2bSJens Axboe bio_put_slab(bs); 1596917a38c7SKent Overstreet bs->bio_slab = NULL; 1597917a38c7SKent Overstreet } 1598917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1599f9c78b2bSJens Axboe 1600011067b0SNeilBrown /** 1601917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1602dad08527SKent Overstreet * @bs: pool to initialize 1603917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1604917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1605917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1606917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1607917a38c7SKent Overstreet * 1608dad08527SKent Overstreet * Description: 1609dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1610dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1611dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1612dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1613dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1614dad08527SKent Overstreet * or things will break badly. 1615dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1616dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1617dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1618dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1619dad08527SKent Overstreet * 1620917a38c7SKent Overstreet */ 1621917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1622917a38c7SKent Overstreet unsigned int pool_size, 1623917a38c7SKent Overstreet unsigned int front_pad, 1624917a38c7SKent Overstreet int flags) 1625917a38c7SKent Overstreet { 1626917a38c7SKent Overstreet bs->front_pad = front_pad; 16279f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 16289f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 16299f180e31SMing Lei else 16309f180e31SMing Lei bs->back_pad = 0; 1631917a38c7SKent Overstreet 1632917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1633917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1634917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1635917a38c7SKent Overstreet 163649d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1637917a38c7SKent Overstreet if (!bs->bio_slab) 1638917a38c7SKent Overstreet return -ENOMEM; 1639917a38c7SKent Overstreet 1640917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1641917a38c7SKent Overstreet goto bad; 1642917a38c7SKent Overstreet 1643917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1644917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1645917a38c7SKent Overstreet goto bad; 1646917a38c7SKent Overstreet 1647be4d234dSJens Axboe if (flags & BIOSET_NEED_RESCUER) { 1648be4d234dSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", 1649be4d234dSJens Axboe WQ_MEM_RECLAIM, 0); 1650917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1651917a38c7SKent Overstreet goto bad; 1652be4d234dSJens Axboe } 1653be4d234dSJens Axboe if (flags & BIOSET_PERCPU_CACHE) { 1654be4d234dSJens Axboe bs->cache = alloc_percpu(struct bio_alloc_cache); 1655be4d234dSJens Axboe if (!bs->cache) 1656be4d234dSJens Axboe goto bad; 1657be4d234dSJens Axboe cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1658be4d234dSJens Axboe } 1659917a38c7SKent Overstreet 1660917a38c7SKent Overstreet return 0; 1661917a38c7SKent Overstreet bad: 1662917a38c7SKent Overstreet bioset_exit(bs); 1663917a38c7SKent Overstreet return -ENOMEM; 1664917a38c7SKent Overstreet } 1665917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1666917a38c7SKent Overstreet 166728e89fd9SJens Axboe /* 166828e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 166928e89fd9SJens Axboe * another bio_set. 167028e89fd9SJens Axboe */ 167128e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 167228e89fd9SJens Axboe { 167328e89fd9SJens Axboe int flags; 167428e89fd9SJens Axboe 167528e89fd9SJens Axboe flags = 0; 167628e89fd9SJens Axboe if (src->bvec_pool.min_nr) 167728e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 167828e89fd9SJens Axboe if (src->rescue_workqueue) 167928e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 168028e89fd9SJens Axboe 168128e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 168228e89fd9SJens Axboe } 168328e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 168428e89fd9SJens Axboe 1685be4d234dSJens Axboe /** 1686be4d234dSJens Axboe * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb 1687be4d234dSJens Axboe * @kiocb: kiocb describing the IO 16880ef47db1SJens Axboe * @nr_vecs: number of iovecs to pre-allocate 1689be4d234dSJens Axboe * @bs: bio_set to allocate from 1690be4d234dSJens Axboe * 1691be4d234dSJens Axboe * Description: 1692be4d234dSJens Axboe * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only 1693be4d234dSJens Axboe * used to check if we should dip into the per-cpu bio_set allocation 16943d5b3fbeSJens Axboe * cache. The allocation uses GFP_KERNEL internally. On return, the 16953d5b3fbeSJens Axboe * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio 16963d5b3fbeSJens Axboe * MUST be done from process context, not hard/soft IRQ. 1697be4d234dSJens Axboe * 1698be4d234dSJens Axboe */ 1699be4d234dSJens Axboe struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs, 1700be4d234dSJens Axboe struct bio_set *bs) 1701be4d234dSJens Axboe { 1702be4d234dSJens Axboe struct bio_alloc_cache *cache; 1703be4d234dSJens Axboe struct bio *bio; 1704be4d234dSJens Axboe 1705be4d234dSJens Axboe if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS) 1706be4d234dSJens Axboe return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1707be4d234dSJens Axboe 1708be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, get_cpu()); 1709*fcade2ceSJens Axboe if (cache->free_list) { 1710*fcade2ceSJens Axboe bio = cache->free_list; 1711*fcade2ceSJens Axboe cache->free_list = bio->bi_next; 1712be4d234dSJens Axboe cache->nr--; 1713be4d234dSJens Axboe put_cpu(); 1714be4d234dSJens Axboe bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs); 1715be4d234dSJens Axboe bio->bi_pool = bs; 1716be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1717be4d234dSJens Axboe return bio; 1718be4d234dSJens Axboe } 1719be4d234dSJens Axboe put_cpu(); 1720be4d234dSJens Axboe bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1721be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1722be4d234dSJens Axboe return bio; 1723be4d234dSJens Axboe } 1724be4d234dSJens Axboe EXPORT_SYMBOL_GPL(bio_alloc_kiocb); 1725be4d234dSJens Axboe 1726de76fd89SChristoph Hellwig static int __init init_bio(void) 1727f9c78b2bSJens Axboe { 1728f9c78b2bSJens Axboe int i; 1729f9c78b2bSJens Axboe 1730f9c78b2bSJens Axboe bio_integrity_init(); 1731de76fd89SChristoph Hellwig 1732de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1733f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1734f9c78b2bSJens Axboe 1735de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1736de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1737f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1738f9c78b2bSJens Axboe } 1739f9c78b2bSJens Axboe 1740be4d234dSJens Axboe cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1741be4d234dSJens Axboe bio_cpu_dead); 1742be4d234dSJens Axboe 1743f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1744f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1745f9c78b2bSJens Axboe 1746f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1747f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1748f9c78b2bSJens Axboe 1749f9c78b2bSJens Axboe return 0; 1750f9c78b2bSJens Axboe } 1751f9c78b2bSJens Axboe subsys_initcall(init_bio); 1752