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 */ 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 347*3b005bf6SChristoph Hellwig struct bio *blk_next_bio(struct bio *bio, unsigned int nr_pages, gfp_t gfp) 348*3b005bf6SChristoph Hellwig { 349*3b005bf6SChristoph Hellwig struct bio *new = bio_alloc(gfp, nr_pages); 350*3b005bf6SChristoph Hellwig 351*3b005bf6SChristoph Hellwig if (bio) { 352*3b005bf6SChristoph Hellwig bio_chain(bio, new); 353*3b005bf6SChristoph Hellwig submit_bio(bio); 354*3b005bf6SChristoph Hellwig } 355*3b005bf6SChristoph Hellwig 356*3b005bf6SChristoph Hellwig return new; 357*3b005bf6SChristoph Hellwig } 358*3b005bf6SChristoph Hellwig EXPORT_SYMBOL_GPL(blk_next_bio); 359*3b005bf6SChristoph Hellwig 360f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 361f9c78b2bSJens Axboe { 362f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 363f9c78b2bSJens Axboe struct bio *bio; 364f9c78b2bSJens Axboe 365f9c78b2bSJens Axboe while (1) { 366f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 367f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 368f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 369f9c78b2bSJens Axboe 370f9c78b2bSJens Axboe if (!bio) 371f9c78b2bSJens Axboe break; 372f9c78b2bSJens Axboe 373ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 374f9c78b2bSJens Axboe } 375f9c78b2bSJens Axboe } 376f9c78b2bSJens Axboe 377f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 378f9c78b2bSJens Axboe { 379f9c78b2bSJens Axboe struct bio_list punt, nopunt; 380f9c78b2bSJens Axboe struct bio *bio; 381f9c78b2bSJens Axboe 38247e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 38347e0fb46SNeilBrown return; 384f9c78b2bSJens Axboe /* 385f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 386f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 387f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 388f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 389f9c78b2bSJens Axboe * our own rescuer would be bad. 390f9c78b2bSJens Axboe * 391f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 392f9c78b2bSJens Axboe * remove from the middle of the list: 393f9c78b2bSJens Axboe */ 394f9c78b2bSJens Axboe 395f9c78b2bSJens Axboe bio_list_init(&punt); 396f9c78b2bSJens Axboe bio_list_init(&nopunt); 397f9c78b2bSJens Axboe 398f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 399f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 400f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 401f9c78b2bSJens Axboe 402f5fe1b51SNeilBrown bio_list_init(&nopunt); 403f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 404f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 405f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 406f9c78b2bSJens Axboe 407f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 408f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 409f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 410f9c78b2bSJens Axboe 411f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 412f9c78b2bSJens Axboe } 413f9c78b2bSJens Axboe 414f9c78b2bSJens Axboe /** 415f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 416519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 417f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 418f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 419f9c78b2bSJens Axboe * 4203175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 421f9c78b2bSJens Axboe * 4223175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4233175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4243175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4253175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4263175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4273175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 428f9c78b2bSJens Axboe * 4293175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4303175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 431ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 432f9c78b2bSJens Axboe * stack overflows. 433f9c78b2bSJens Axboe * 4343175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4353175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 436f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 437f9c78b2bSJens Axboe * thread. 438f9c78b2bSJens Axboe * 439f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 440f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 441ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 442f9c78b2bSJens Axboe * for per bio allocations. 443f9c78b2bSJens Axboe * 4443175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 445f9c78b2bSJens Axboe */ 4460f2e6ab8SChristoph Hellwig struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs, 4477a88fa19SDan Carpenter struct bio_set *bs) 448f9c78b2bSJens Axboe { 449f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 450f9c78b2bSJens Axboe struct bio *bio; 451f9c78b2bSJens Axboe void *p; 452f9c78b2bSJens Axboe 4533175199aSChristoph Hellwig /* should not use nobvec bioset for nr_iovecs > 0 */ 4543175199aSChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) 455f9c78b2bSJens Axboe return NULL; 456f9c78b2bSJens Axboe 457f9c78b2bSJens Axboe /* 4583175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 4593175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 4603175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 461f9c78b2bSJens Axboe * 4623175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 4633175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 4643175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 4653175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 4663175199aSChristoph Hellwig * mempool's reserve. 467f9c78b2bSJens Axboe * 468f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 4693175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 4703175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 4713175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 4723175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 4733175199aSChristoph Hellwig * gfp_flags. 474f9c78b2bSJens Axboe */ 475f5fe1b51SNeilBrown if (current->bio_list && 476f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 47747e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 47847e0fb46SNeilBrown bs->rescue_workqueue) 479d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 480f9c78b2bSJens Axboe 4818aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 482f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 483f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 484f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4858aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 486f9c78b2bSJens Axboe } 487f9c78b2bSJens Axboe if (unlikely(!p)) 488f9c78b2bSJens Axboe return NULL; 489f9c78b2bSJens Axboe 4903175199aSChristoph Hellwig bio = p + bs->front_pad; 4913175199aSChristoph Hellwig if (nr_iovecs > BIO_INLINE_VECS) { 4923175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 493f9c78b2bSJens Axboe 4947a800a20SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 495f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 496f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 497f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4987a800a20SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 499f9c78b2bSJens Axboe } 500f9c78b2bSJens Axboe if (unlikely(!bvl)) 501f9c78b2bSJens Axboe goto err_free; 502f9c78b2bSJens Axboe 5037a800a20SChristoph Hellwig bio_init(bio, bvl, nr_iovecs); 504f9c78b2bSJens Axboe } else if (nr_iovecs) { 5053175199aSChristoph Hellwig bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS); 5063175199aSChristoph Hellwig } else { 5073175199aSChristoph Hellwig bio_init(bio, NULL, 0); 508f9c78b2bSJens Axboe } 509f9c78b2bSJens Axboe 510f9c78b2bSJens Axboe bio->bi_pool = bs; 511f9c78b2bSJens Axboe return bio; 512f9c78b2bSJens Axboe 513f9c78b2bSJens Axboe err_free: 5148aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 515f9c78b2bSJens Axboe return NULL; 516f9c78b2bSJens Axboe } 517f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 518f9c78b2bSJens Axboe 5193175199aSChristoph Hellwig /** 5203175199aSChristoph Hellwig * bio_kmalloc - kmalloc a bio for I/O 5213175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 5223175199aSChristoph Hellwig * @nr_iovecs: number of iovecs to pre-allocate 5233175199aSChristoph Hellwig * 5243175199aSChristoph Hellwig * Use kmalloc to allocate and initialize a bio. 5253175199aSChristoph Hellwig * 5263175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 5273175199aSChristoph Hellwig */ 5280f2e6ab8SChristoph Hellwig struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) 5293175199aSChristoph Hellwig { 5303175199aSChristoph Hellwig struct bio *bio; 5313175199aSChristoph Hellwig 5323175199aSChristoph Hellwig if (nr_iovecs > UIO_MAXIOV) 5333175199aSChristoph Hellwig return NULL; 5343175199aSChristoph Hellwig 5353175199aSChristoph Hellwig bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 5363175199aSChristoph Hellwig if (unlikely(!bio)) 5373175199aSChristoph Hellwig return NULL; 5383175199aSChristoph Hellwig bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs); 5393175199aSChristoph Hellwig bio->bi_pool = NULL; 5403175199aSChristoph Hellwig return bio; 5413175199aSChristoph Hellwig } 5423175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 5433175199aSChristoph Hellwig 5446f822e1bSChristoph Hellwig void zero_fill_bio(struct bio *bio) 545f9c78b2bSJens Axboe { 546f9c78b2bSJens Axboe struct bio_vec bv; 547f9c78b2bSJens Axboe struct bvec_iter iter; 548f9c78b2bSJens Axboe 549ab6c340eSChristoph Hellwig bio_for_each_segment(bv, bio, iter) 550ab6c340eSChristoph Hellwig memzero_bvec(&bv); 551f9c78b2bSJens Axboe } 5526f822e1bSChristoph Hellwig EXPORT_SYMBOL(zero_fill_bio); 553f9c78b2bSJens Axboe 55483c9c547SMing Lei /** 55583c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 55683c9c547SMing Lei * @bio: the bio to be truncated 55783c9c547SMing Lei * @new_size: new size for truncating the bio 55883c9c547SMing Lei * 55983c9c547SMing Lei * Description: 56083c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 56183c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 56283c9c547SMing Lei * be used for handling corner cases, such as bio eod. 56383c9c547SMing Lei */ 5644f7ab09aSChristoph Hellwig static void bio_truncate(struct bio *bio, unsigned new_size) 56585a8ce62SMing Lei { 56685a8ce62SMing Lei struct bio_vec bv; 56785a8ce62SMing Lei struct bvec_iter iter; 56885a8ce62SMing Lei unsigned int done = 0; 56985a8ce62SMing Lei bool truncated = false; 57085a8ce62SMing Lei 57185a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 57285a8ce62SMing Lei return; 57385a8ce62SMing Lei 57483c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 57585a8ce62SMing Lei goto exit; 57685a8ce62SMing Lei 57785a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 57885a8ce62SMing Lei if (done + bv.bv_len > new_size) { 57985a8ce62SMing Lei unsigned offset; 58085a8ce62SMing Lei 58185a8ce62SMing Lei if (!truncated) 58285a8ce62SMing Lei offset = new_size - done; 58385a8ce62SMing Lei else 58485a8ce62SMing Lei offset = 0; 5853ee859e3SOGAWA Hirofumi zero_user(bv.bv_page, bv.bv_offset + offset, 5863ee859e3SOGAWA Hirofumi bv.bv_len - offset); 58785a8ce62SMing Lei truncated = true; 58885a8ce62SMing Lei } 58985a8ce62SMing Lei done += bv.bv_len; 59085a8ce62SMing Lei } 59185a8ce62SMing Lei 59285a8ce62SMing Lei exit: 59385a8ce62SMing Lei /* 59485a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 59585a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 59685a8ce62SMing Lei * in its .end_bio() callback. 59785a8ce62SMing Lei * 59885a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 59985a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 60085a8ce62SMing Lei */ 60185a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 60285a8ce62SMing Lei } 60385a8ce62SMing Lei 604f9c78b2bSJens Axboe /** 60529125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 60629125ed6SChristoph Hellwig * @bio: bio to truncate 60729125ed6SChristoph Hellwig * 60829125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 60929125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 61029125ed6SChristoph Hellwig * 61129125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 61229125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 61329125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 61429125ed6SChristoph Hellwig * sector" case. 61529125ed6SChristoph Hellwig */ 61629125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 61729125ed6SChristoph Hellwig { 618309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 61929125ed6SChristoph Hellwig 62029125ed6SChristoph Hellwig if (!maxsector) 62129125ed6SChristoph Hellwig return; 62229125ed6SChristoph Hellwig 62329125ed6SChristoph Hellwig /* 62429125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 62529125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 62629125ed6SChristoph Hellwig * an EIO. 62729125ed6SChristoph Hellwig */ 62829125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 62929125ed6SChristoph Hellwig return; 63029125ed6SChristoph Hellwig 63129125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 63229125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 63329125ed6SChristoph Hellwig return; 63429125ed6SChristoph Hellwig 63529125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 63629125ed6SChristoph Hellwig } 63729125ed6SChristoph Hellwig 638be4d234dSJens Axboe #define ALLOC_CACHE_MAX 512 639be4d234dSJens Axboe #define ALLOC_CACHE_SLACK 64 640be4d234dSJens Axboe 641be4d234dSJens Axboe static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 642be4d234dSJens Axboe unsigned int nr) 643be4d234dSJens Axboe { 644be4d234dSJens Axboe unsigned int i = 0; 645be4d234dSJens Axboe struct bio *bio; 646be4d234dSJens Axboe 647fcade2ceSJens Axboe while ((bio = cache->free_list) != NULL) { 648fcade2ceSJens Axboe cache->free_list = bio->bi_next; 649be4d234dSJens Axboe cache->nr--; 650be4d234dSJens Axboe bio_free(bio); 651be4d234dSJens Axboe if (++i == nr) 652be4d234dSJens Axboe break; 653be4d234dSJens Axboe } 654be4d234dSJens Axboe } 655be4d234dSJens Axboe 656be4d234dSJens Axboe static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 657be4d234dSJens Axboe { 658be4d234dSJens Axboe struct bio_set *bs; 659be4d234dSJens Axboe 660be4d234dSJens Axboe bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 661be4d234dSJens Axboe if (bs->cache) { 662be4d234dSJens Axboe struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 663be4d234dSJens Axboe 664be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 665be4d234dSJens Axboe } 666be4d234dSJens Axboe return 0; 667be4d234dSJens Axboe } 668be4d234dSJens Axboe 669be4d234dSJens Axboe static void bio_alloc_cache_destroy(struct bio_set *bs) 670be4d234dSJens Axboe { 671be4d234dSJens Axboe int cpu; 672be4d234dSJens Axboe 673be4d234dSJens Axboe if (!bs->cache) 674be4d234dSJens Axboe return; 675be4d234dSJens Axboe 676be4d234dSJens Axboe cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 677be4d234dSJens Axboe for_each_possible_cpu(cpu) { 678be4d234dSJens Axboe struct bio_alloc_cache *cache; 679be4d234dSJens Axboe 680be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, cpu); 681be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 682be4d234dSJens Axboe } 683be4d234dSJens Axboe free_percpu(bs->cache); 684be4d234dSJens Axboe } 685be4d234dSJens Axboe 68629125ed6SChristoph Hellwig /** 687f9c78b2bSJens Axboe * bio_put - release a reference to a bio 688f9c78b2bSJens Axboe * @bio: bio to release reference to 689f9c78b2bSJens Axboe * 690f9c78b2bSJens Axboe * Description: 691f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 6929b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 693f9c78b2bSJens Axboe **/ 694f9c78b2bSJens Axboe void bio_put(struct bio *bio) 695f9c78b2bSJens Axboe { 696be4d234dSJens Axboe if (unlikely(bio_flagged(bio, BIO_REFFED))) { 6979e8c0d0dSChristoph Hellwig BUG_ON(!atomic_read(&bio->__bi_cnt)); 698be4d234dSJens Axboe if (!atomic_dec_and_test(&bio->__bi_cnt)) 699be4d234dSJens Axboe return; 700be4d234dSJens Axboe } 701f9c78b2bSJens Axboe 702be4d234dSJens Axboe if (bio_flagged(bio, BIO_PERCPU_CACHE)) { 703be4d234dSJens Axboe struct bio_alloc_cache *cache; 704be4d234dSJens Axboe 705be4d234dSJens Axboe bio_uninit(bio); 706be4d234dSJens Axboe cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 707fcade2ceSJens Axboe bio->bi_next = cache->free_list; 708fcade2ceSJens Axboe cache->free_list = bio; 709be4d234dSJens Axboe if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 710be4d234dSJens Axboe bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 711be4d234dSJens Axboe put_cpu(); 712be4d234dSJens Axboe } else { 713f9c78b2bSJens Axboe bio_free(bio); 714f9c78b2bSJens Axboe } 715dac56212SJens Axboe } 716f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 717f9c78b2bSJens Axboe 718f9c78b2bSJens Axboe /** 719f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 720f9c78b2bSJens Axboe * @bio: destination bio 721f9c78b2bSJens Axboe * @bio_src: bio to clone 722f9c78b2bSJens Axboe * 723f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 724f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 725f9c78b2bSJens Axboe * bio will be one. 726f9c78b2bSJens Axboe * 727f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 728f9c78b2bSJens Axboe */ 729f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 730f9c78b2bSJens Axboe { 7317a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs); 732f9c78b2bSJens Axboe 733f9c78b2bSJens Axboe /* 734309dca30SChristoph Hellwig * most users will be overriding ->bi_bdev with a new target, 735f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 736f9c78b2bSJens Axboe */ 737309dca30SChristoph Hellwig bio->bi_bdev = bio_src->bi_bdev; 738b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 739111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 740111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 74146bbf653SChristoph Hellwig if (bio_flagged(bio_src, BIO_REMAPPED)) 74246bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 7431eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 744ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 745cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 746f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 747f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 74820bd723eSPaolo Valente 749db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 750e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 751f9c78b2bSJens Axboe } 752f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 753f9c78b2bSJens Axboe 754f9c78b2bSJens Axboe /** 755f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 756f9c78b2bSJens Axboe * @bio: bio to clone 757f9c78b2bSJens Axboe * @gfp_mask: allocation priority 758f9c78b2bSJens Axboe * @bs: bio_set to allocate from 759f9c78b2bSJens Axboe * 760f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 761f9c78b2bSJens Axboe */ 762f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 763f9c78b2bSJens Axboe { 764f9c78b2bSJens Axboe struct bio *b; 765f9c78b2bSJens Axboe 766f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 767f9c78b2bSJens Axboe if (!b) 768f9c78b2bSJens Axboe return NULL; 769f9c78b2bSJens Axboe 770f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 771f9c78b2bSJens Axboe 77207560151SEric Biggers if (bio_crypt_clone(b, bio, gfp_mask) < 0) 77307560151SEric Biggers goto err_put; 774a892c8d5SSatya Tangirala 77507560151SEric Biggers if (bio_integrity(bio) && 77607560151SEric Biggers bio_integrity_clone(b, bio, gfp_mask) < 0) 77707560151SEric Biggers goto err_put; 778f9c78b2bSJens Axboe 779f9c78b2bSJens Axboe return b; 78007560151SEric Biggers 78107560151SEric Biggers err_put: 78207560151SEric Biggers bio_put(b); 78307560151SEric Biggers return NULL; 784f9c78b2bSJens Axboe } 785f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 786f9c78b2bSJens Axboe 7875cbd28e3SChristoph Hellwig const char *bio_devname(struct bio *bio, char *buf) 7885cbd28e3SChristoph Hellwig { 789309dca30SChristoph Hellwig return bdevname(bio->bi_bdev, buf); 7905cbd28e3SChristoph Hellwig } 7915cbd28e3SChristoph Hellwig EXPORT_SYMBOL(bio_devname); 7925cbd28e3SChristoph Hellwig 7939a6083beSChristoph Hellwig /** 7949a6083beSChristoph Hellwig * bio_full - check if the bio is full 7959a6083beSChristoph Hellwig * @bio: bio to check 7969a6083beSChristoph Hellwig * @len: length of one segment to be added 7979a6083beSChristoph Hellwig * 7989a6083beSChristoph Hellwig * Return true if @bio is full and one segment with @len bytes can't be 7999a6083beSChristoph Hellwig * added to the bio, otherwise return false 8009a6083beSChristoph Hellwig */ 8019a6083beSChristoph Hellwig static inline bool bio_full(struct bio *bio, unsigned len) 8029a6083beSChristoph Hellwig { 8039a6083beSChristoph Hellwig if (bio->bi_vcnt >= bio->bi_max_vecs) 8049a6083beSChristoph Hellwig return true; 8059a6083beSChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) 8069a6083beSChristoph Hellwig return true; 8079a6083beSChristoph Hellwig return false; 8089a6083beSChristoph Hellwig } 8099a6083beSChristoph Hellwig 8105919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 8115919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 812ff896738SChristoph Hellwig bool *same_page) 8135919482eSMing Lei { 814d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 815d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 8165919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 8175919482eSMing Lei 8185919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 8195919482eSMing Lei return false; 8205919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 8215919482eSMing Lei return false; 82252d52d1cSChristoph Hellwig 823ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 824d8166519SMatthew Wilcox (Oracle) if (*same_page) 8255919482eSMing Lei return true; 826d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 8275919482eSMing Lei } 8285919482eSMing Lei 8299774b391SChristoph Hellwig /** 8309774b391SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 8319774b391SChristoph Hellwig * @bio: destination bio 8329774b391SChristoph Hellwig * @page: start page to add 8339774b391SChristoph Hellwig * @len: length of the data to add 8349774b391SChristoph Hellwig * @off: offset of the data relative to @page 8359774b391SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 8369774b391SChristoph Hellwig * 8379774b391SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 8389774b391SChristoph Hellwig * useful optimisation for file systems with a block size smaller than the 8399774b391SChristoph Hellwig * page size. 8409774b391SChristoph Hellwig * 8419774b391SChristoph Hellwig * Warn if (@len, @off) crosses pages in case that @same_page is true. 8429774b391SChristoph Hellwig * 8439774b391SChristoph Hellwig * Return %true on success or %false on failure. 8449774b391SChristoph Hellwig */ 8459774b391SChristoph Hellwig static bool __bio_try_merge_page(struct bio *bio, struct page *page, 8469774b391SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 8479774b391SChristoph Hellwig { 8489774b391SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8499774b391SChristoph Hellwig return false; 8509774b391SChristoph Hellwig 8519774b391SChristoph Hellwig if (bio->bi_vcnt > 0) { 8529774b391SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 8539774b391SChristoph Hellwig 8549774b391SChristoph Hellwig if (page_is_mergeable(bv, page, len, off, same_page)) { 8559774b391SChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) { 8569774b391SChristoph Hellwig *same_page = false; 8579774b391SChristoph Hellwig return false; 8589774b391SChristoph Hellwig } 8599774b391SChristoph Hellwig bv->bv_len += len; 8609774b391SChristoph Hellwig bio->bi_iter.bi_size += len; 8619774b391SChristoph Hellwig return true; 8629774b391SChristoph Hellwig } 8639774b391SChristoph Hellwig } 8649774b391SChristoph Hellwig return false; 8659774b391SChristoph Hellwig } 8669774b391SChristoph Hellwig 867e4581105SChristoph Hellwig /* 868e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 869e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 870e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 871e4581105SChristoph Hellwig */ 872e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 873e4581105SChristoph Hellwig struct page *page, unsigned len, 874e4581105SChristoph Hellwig unsigned offset, bool *same_page) 875489fbbcbSMing Lei { 876384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 877489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 878489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 879489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 880489fbbcbSMing Lei 881489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 882489fbbcbSMing Lei return false; 883489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 884489fbbcbSMing Lei return false; 885384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 886489fbbcbSMing Lei } 887489fbbcbSMing Lei 888f4595875SShaohua Li /** 889e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 890c66a14d0SKent Overstreet * @q: the target queue 891c66a14d0SKent Overstreet * @bio: destination bio 892c66a14d0SKent Overstreet * @page: page to add 893c66a14d0SKent Overstreet * @len: vec entry length 894c66a14d0SKent Overstreet * @offset: vec entry offset 895e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 896e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 897f9c78b2bSJens Axboe * 898e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 899e4581105SChristoph Hellwig * and gap limitations. 900f9c78b2bSJens Axboe */ 901e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 90219047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 903e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 904f9c78b2bSJens Axboe { 905f9c78b2bSJens Axboe struct bio_vec *bvec; 906f9c78b2bSJens Axboe 907e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 908f9c78b2bSJens Axboe return 0; 909f9c78b2bSJens Axboe 910e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 911f9c78b2bSJens Axboe return 0; 912f9c78b2bSJens Axboe 913f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 914e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 915384209cdSChristoph Hellwig return len; 916320ea869SChristoph Hellwig 917320ea869SChristoph Hellwig /* 918320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 919320ea869SChristoph Hellwig * would create a gap, disallow it. 920320ea869SChristoph Hellwig */ 921384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 922320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 923320ea869SChristoph Hellwig return 0; 924f9c78b2bSJens Axboe } 925f9c78b2bSJens Axboe 92679d08f89SMing Lei if (bio_full(bio, len)) 927f9c78b2bSJens Axboe return 0; 928f9c78b2bSJens Axboe 92914ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 930489fbbcbSMing Lei return 0; 931489fbbcbSMing Lei 932f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 933f9c78b2bSJens Axboe bvec->bv_page = page; 934f9c78b2bSJens Axboe bvec->bv_len = len; 935f9c78b2bSJens Axboe bvec->bv_offset = offset; 936fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 937dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 938f9c78b2bSJens Axboe return len; 939f9c78b2bSJens Axboe } 94019047087SMing Lei 941e4581105SChristoph Hellwig /** 942e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 943e4581105SChristoph Hellwig * @q: the target queue 944e4581105SChristoph Hellwig * @bio: destination bio 945e4581105SChristoph Hellwig * @page: page to add 946e4581105SChristoph Hellwig * @len: vec entry length 947e4581105SChristoph Hellwig * @offset: vec entry offset 948e4581105SChristoph Hellwig * 949e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 950e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 951e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 952e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 953e4581105SChristoph Hellwig * 954e4581105SChristoph Hellwig * This should only be used by passthrough bios. 955e4581105SChristoph Hellwig */ 95619047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 95719047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 95819047087SMing Lei { 959d1916c86SChristoph Hellwig bool same_page = false; 960e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 961e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 96219047087SMing Lei } 963f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 964f9c78b2bSJens Axboe 965f9c78b2bSJens Axboe /** 966ae29333fSJohannes Thumshirn * bio_add_zone_append_page - attempt to add page to zone-append bio 967ae29333fSJohannes Thumshirn * @bio: destination bio 968ae29333fSJohannes Thumshirn * @page: page to add 969ae29333fSJohannes Thumshirn * @len: vec entry length 970ae29333fSJohannes Thumshirn * @offset: vec entry offset 971ae29333fSJohannes Thumshirn * 972ae29333fSJohannes Thumshirn * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 973ae29333fSJohannes Thumshirn * for a zone-append request. This can fail for a number of reasons, such as the 974ae29333fSJohannes Thumshirn * bio being full or the target block device is not a zoned block device or 975ae29333fSJohannes Thumshirn * other limitations of the target block device. The target block device must 976ae29333fSJohannes Thumshirn * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 977ae29333fSJohannes Thumshirn * to an empty bio. 978ae29333fSJohannes Thumshirn * 979ae29333fSJohannes Thumshirn * Returns: number of bytes added to the bio, or 0 in case of a failure. 980ae29333fSJohannes Thumshirn */ 981ae29333fSJohannes Thumshirn int bio_add_zone_append_page(struct bio *bio, struct page *page, 982ae29333fSJohannes Thumshirn unsigned int len, unsigned int offset) 983ae29333fSJohannes Thumshirn { 9843caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 985ae29333fSJohannes Thumshirn bool same_page = false; 986ae29333fSJohannes Thumshirn 987ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 988ae29333fSJohannes Thumshirn return 0; 989ae29333fSJohannes Thumshirn 990ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 991ae29333fSJohannes Thumshirn return 0; 992ae29333fSJohannes Thumshirn 993ae29333fSJohannes Thumshirn return bio_add_hw_page(q, bio, page, len, offset, 994ae29333fSJohannes Thumshirn queue_max_zone_append_sectors(q), &same_page); 995ae29333fSJohannes Thumshirn } 996ae29333fSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 997ae29333fSJohannes Thumshirn 998ae29333fSJohannes Thumshirn /** 999551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 10000aa69fd3SChristoph Hellwig * @bio: destination bio 1001551879a4SMing Lei * @page: start page to add 1002551879a4SMing Lei * @len: length of the data to add, may cross pages 1003551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 10040aa69fd3SChristoph Hellwig * 10050aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 10060aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 10070aa69fd3SChristoph Hellwig */ 10080aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 10090aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 10100aa69fd3SChristoph Hellwig { 10110aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 10120aa69fd3SChristoph Hellwig 10130aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 101479d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 10150aa69fd3SChristoph Hellwig 10160aa69fd3SChristoph Hellwig bv->bv_page = page; 10170aa69fd3SChristoph Hellwig bv->bv_offset = off; 10180aa69fd3SChristoph Hellwig bv->bv_len = len; 10190aa69fd3SChristoph Hellwig 10200aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 10210aa69fd3SChristoph Hellwig bio->bi_vcnt++; 1022b8e24a93SJohannes Weiner 1023b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 1024b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 10250aa69fd3SChristoph Hellwig } 10260aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 10270aa69fd3SChristoph Hellwig 10280aa69fd3SChristoph Hellwig /** 1029551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 1030f9c78b2bSJens Axboe * @bio: destination bio 1031551879a4SMing Lei * @page: start page to add 1032551879a4SMing Lei * @len: vec entry length, may cross pages 1033551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 1034f9c78b2bSJens Axboe * 1035551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 1036c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1037f9c78b2bSJens Axboe */ 1038c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 1039c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 1040f9c78b2bSJens Axboe { 1041ff896738SChristoph Hellwig bool same_page = false; 1042ff896738SChristoph Hellwig 1043ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 104479d08f89SMing Lei if (bio_full(bio, len)) 1045c66a14d0SKent Overstreet return 0; 10460aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 1047c66a14d0SKent Overstreet } 1048c66a14d0SKent Overstreet return len; 1049f9c78b2bSJens Axboe } 1050f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 1051f9c78b2bSJens Axboe 105285f5a74cSMatthew Wilcox (Oracle) /** 105385f5a74cSMatthew Wilcox (Oracle) * bio_add_folio - Attempt to add part of a folio to a bio. 105485f5a74cSMatthew Wilcox (Oracle) * @bio: BIO to add to. 105585f5a74cSMatthew Wilcox (Oracle) * @folio: Folio to add. 105685f5a74cSMatthew Wilcox (Oracle) * @len: How many bytes from the folio to add. 105785f5a74cSMatthew Wilcox (Oracle) * @off: First byte in this folio to add. 105885f5a74cSMatthew Wilcox (Oracle) * 105985f5a74cSMatthew Wilcox (Oracle) * Filesystems that use folios can call this function instead of calling 106085f5a74cSMatthew Wilcox (Oracle) * bio_add_page() for each page in the folio. If @off is bigger than 106185f5a74cSMatthew Wilcox (Oracle) * PAGE_SIZE, this function can create a bio_vec that starts in a page 106285f5a74cSMatthew Wilcox (Oracle) * after the bv_page. BIOs do not support folios that are 4GiB or larger. 106385f5a74cSMatthew Wilcox (Oracle) * 106485f5a74cSMatthew Wilcox (Oracle) * Return: Whether the addition was successful. 106585f5a74cSMatthew Wilcox (Oracle) */ 106685f5a74cSMatthew Wilcox (Oracle) bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, 106785f5a74cSMatthew Wilcox (Oracle) size_t off) 106885f5a74cSMatthew Wilcox (Oracle) { 106985f5a74cSMatthew Wilcox (Oracle) if (len > UINT_MAX || off > UINT_MAX) 107085f5a74cSMatthew Wilcox (Oracle) return 0; 107185f5a74cSMatthew Wilcox (Oracle) return bio_add_page(bio, &folio->page, len, off) > 0; 107285f5a74cSMatthew Wilcox (Oracle) } 107385f5a74cSMatthew Wilcox (Oracle) 1074c809084aSPavel Begunkov void __bio_release_pages(struct bio *bio, bool mark_dirty) 10757321ecbfSChristoph Hellwig { 10767321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 10777321ecbfSChristoph Hellwig struct bio_vec *bvec; 10787321ecbfSChristoph Hellwig 1079d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1080d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 1081d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 10827321ecbfSChristoph Hellwig put_page(bvec->bv_page); 10837321ecbfSChristoph Hellwig } 1084d241a95fSChristoph Hellwig } 1085c809084aSPavel Begunkov EXPORT_SYMBOL_GPL(__bio_release_pages); 10867321ecbfSChristoph Hellwig 10871bb6b810SPavel Begunkov void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 10886d0c48aeSJens Axboe { 1089fa5fa8ecSPavel Begunkov size_t size = iov_iter_count(iter); 1090fa5fa8ecSPavel Begunkov 10917a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_max_vecs); 10926d0c48aeSJens Axboe 1093fa5fa8ecSPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1094fa5fa8ecSPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1095fa5fa8ecSPavel Begunkov size_t max_sectors = queue_max_zone_append_sectors(q); 1096fa5fa8ecSPavel Begunkov 1097fa5fa8ecSPavel Begunkov size = min(size, max_sectors << SECTOR_SHIFT); 1098fa5fa8ecSPavel Begunkov } 1099fa5fa8ecSPavel Begunkov 1100c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 1101c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1102c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 1103fa5fa8ecSPavel Begunkov bio->bi_iter.bi_size = size; 1104ed97ce5eSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 1105977be012SChristoph Hellwig bio_set_flag(bio, BIO_CLONED); 11067de55b7dSJohannes Thumshirn } 11076d0c48aeSJens Axboe 1108d9cf3bd5SPavel Begunkov static void bio_put_pages(struct page **pages, size_t size, size_t off) 1109d9cf3bd5SPavel Begunkov { 1110d9cf3bd5SPavel Begunkov size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE); 1111d9cf3bd5SPavel Begunkov 1112d9cf3bd5SPavel Begunkov for (i = 0; i < nr; i++) 1113d9cf3bd5SPavel Begunkov put_page(pages[i]); 1114d9cf3bd5SPavel Begunkov } 1115d9cf3bd5SPavel Begunkov 1116576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1117576ed913SChristoph Hellwig 11182cefe4dbSKent Overstreet /** 111917d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 11202cefe4dbSKent Overstreet * @bio: bio to add pages to 11212cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 11222cefe4dbSKent Overstreet * 112317d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 11242cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 112517d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 11263cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 11272cefe4dbSKent Overstreet */ 112817d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 11292cefe4dbSKent Overstreet { 1130576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1131576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 11322cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 11332cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 113445691804SChristoph Hellwig bool same_page = false; 1135576ed913SChristoph Hellwig ssize_t size, left; 1136576ed913SChristoph Hellwig unsigned len, i; 1137b403ea24SMartin Wilck size_t offset; 1138576ed913SChristoph Hellwig 1139576ed913SChristoph Hellwig /* 1140576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 1141576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 1142576ed913SChristoph Hellwig * without overwriting the temporary page array. 1143576ed913SChristoph Hellwig */ 1144576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1145576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 11462cefe4dbSKent Overstreet 114735c820e7SJens Axboe size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 11482cefe4dbSKent Overstreet if (unlikely(size <= 0)) 11492cefe4dbSKent Overstreet return size ? size : -EFAULT; 11502cefe4dbSKent Overstreet 1151576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1152576ed913SChristoph Hellwig struct page *page = pages[i]; 11532cefe4dbSKent Overstreet 1154576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 115545691804SChristoph Hellwig 115645691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 115745691804SChristoph Hellwig if (same_page) 115845691804SChristoph Hellwig put_page(page); 115945691804SChristoph Hellwig } else { 1160d9cf3bd5SPavel Begunkov if (WARN_ON_ONCE(bio_full(bio, len))) { 1161d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 1162576ed913SChristoph Hellwig return -EINVAL; 1163d9cf3bd5SPavel Begunkov } 116445691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 116545691804SChristoph Hellwig } 1166576ed913SChristoph Hellwig offset = 0; 11672cefe4dbSKent Overstreet } 11682cefe4dbSKent Overstreet 11692cefe4dbSKent Overstreet iov_iter_advance(iter, size); 11702cefe4dbSKent Overstreet return 0; 11712cefe4dbSKent Overstreet } 117217d51b10SMartin Wilck 11730512a75bSKeith Busch static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 11740512a75bSKeith Busch { 11750512a75bSKeith Busch unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 11760512a75bSKeith Busch unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 11773caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 11780512a75bSKeith Busch unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 11790512a75bSKeith Busch struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 11800512a75bSKeith Busch struct page **pages = (struct page **)bv; 11810512a75bSKeith Busch ssize_t size, left; 11820512a75bSKeith Busch unsigned len, i; 11830512a75bSKeith Busch size_t offset; 11844977d121SNaohiro Aota int ret = 0; 11850512a75bSKeith Busch 11860512a75bSKeith Busch if (WARN_ON_ONCE(!max_append_sectors)) 11870512a75bSKeith Busch return 0; 11880512a75bSKeith Busch 11890512a75bSKeith Busch /* 11900512a75bSKeith Busch * Move page array up in the allocated memory for the bio vecs as far as 11910512a75bSKeith Busch * possible so that we can start filling biovecs from the beginning 11920512a75bSKeith Busch * without overwriting the temporary page array. 11930512a75bSKeith Busch */ 11940512a75bSKeith Busch BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 11950512a75bSKeith Busch pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 11960512a75bSKeith Busch 11970512a75bSKeith Busch size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 11980512a75bSKeith Busch if (unlikely(size <= 0)) 11990512a75bSKeith Busch return size ? size : -EFAULT; 12000512a75bSKeith Busch 12010512a75bSKeith Busch for (left = size, i = 0; left > 0; left -= len, i++) { 12020512a75bSKeith Busch struct page *page = pages[i]; 12030512a75bSKeith Busch bool same_page = false; 12040512a75bSKeith Busch 12050512a75bSKeith Busch len = min_t(size_t, PAGE_SIZE - offset, left); 12060512a75bSKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 12074977d121SNaohiro Aota max_append_sectors, &same_page) != len) { 1208d9cf3bd5SPavel Begunkov bio_put_pages(pages + i, left, offset); 12094977d121SNaohiro Aota ret = -EINVAL; 12104977d121SNaohiro Aota break; 12114977d121SNaohiro Aota } 12120512a75bSKeith Busch if (same_page) 12130512a75bSKeith Busch put_page(page); 12140512a75bSKeith Busch offset = 0; 12150512a75bSKeith Busch } 12160512a75bSKeith Busch 12174977d121SNaohiro Aota iov_iter_advance(iter, size - left); 12184977d121SNaohiro Aota return ret; 12190512a75bSKeith Busch } 12200512a75bSKeith Busch 122117d51b10SMartin Wilck /** 12226d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 122317d51b10SMartin Wilck * @bio: bio to add pages to 12246d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 122517d51b10SMartin Wilck * 12266d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 12276d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 12286d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1229c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1230c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1231c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1232c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1233c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1234c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 12356d0c48aeSJens Axboe * 123617d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 12375cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 12386d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 12396d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 12400cf41e5eSPavel Begunkov * 12410cf41e5eSPavel Begunkov * It's intended for direct IO, so doesn't do PSI tracking, the caller is 12420cf41e5eSPavel Begunkov * responsible for setting BIO_WORKINGSET if necessary. 124317d51b10SMartin Wilck */ 124417d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 124517d51b10SMartin Wilck { 1246c42bca92SPavel Begunkov int ret = 0; 124714eacf12SChristoph Hellwig 1248c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1249fa5fa8ecSPavel Begunkov bio_iov_bvec_set(bio, iter); 1250fa5fa8ecSPavel Begunkov iov_iter_advance(iter, bio->bi_iter.bi_size); 1251fa5fa8ecSPavel Begunkov return 0; 125286004515SChristoph Hellwig } 125317d51b10SMartin Wilck 125417d51b10SMartin Wilck do { 1255c42bca92SPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) 12560512a75bSKeith Busch ret = __bio_iov_append_get_pages(bio, iter); 12576d0c48aeSJens Axboe else 12586d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 125979d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 126017d51b10SMartin Wilck 12610cf41e5eSPavel Begunkov /* don't account direct I/O as memory stall */ 12620cf41e5eSPavel Begunkov bio_clear_flag(bio, BIO_WORKINGSET); 126314eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 126417d51b10SMartin Wilck } 126529b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 12662cefe4dbSKent Overstreet 12674246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1268f9c78b2bSJens Axboe { 126965e53aabSChristoph Hellwig complete(bio->bi_private); 1270f9c78b2bSJens Axboe } 1271f9c78b2bSJens Axboe 1272f9c78b2bSJens Axboe /** 1273f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1274f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1275f9c78b2bSJens Axboe * 1276f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1277f9c78b2bSJens Axboe * bio_endio() on failure. 12783d289d68SJan Kara * 12793d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 12803d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 12813d289d68SJan Kara * on his own. 1282f9c78b2bSJens Axboe */ 12834e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1284f9c78b2bSJens Axboe { 1285309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1286309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1287de6a78b6SMing Lei unsigned long hang_check; 1288f9c78b2bSJens Axboe 128965e53aabSChristoph Hellwig bio->bi_private = &done; 1290f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 12911eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 12924e49ea4aSMike Christie submit_bio(bio); 1293de6a78b6SMing Lei 1294de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1295de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1296de6a78b6SMing Lei if (hang_check) 1297de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1298de6a78b6SMing Lei hang_check * (HZ/2))) 1299de6a78b6SMing Lei ; 1300de6a78b6SMing Lei else 130165e53aabSChristoph Hellwig wait_for_completion_io(&done); 1302f9c78b2bSJens Axboe 130365e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1304f9c78b2bSJens Axboe } 1305f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1306f9c78b2bSJens Axboe 1307d4aa57a1SJens Axboe void __bio_advance(struct bio *bio, unsigned bytes) 1308f9c78b2bSJens Axboe { 1309f9c78b2bSJens Axboe if (bio_integrity(bio)) 1310f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1311f9c78b2bSJens Axboe 1312a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1313f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1314f9c78b2bSJens Axboe } 1315d4aa57a1SJens Axboe EXPORT_SYMBOL(__bio_advance); 1316f9c78b2bSJens Axboe 131745db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 131845db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1319f9c78b2bSJens Axboe { 132045db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 1321f8b679a0SChristoph Hellwig struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1322f8b679a0SChristoph Hellwig struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1323f8b679a0SChristoph Hellwig unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1324f8b679a0SChristoph Hellwig void *src_buf; 132545db54d5SKent Overstreet 1326f8b679a0SChristoph Hellwig src_buf = bvec_kmap_local(&src_bv); 1327f8b679a0SChristoph Hellwig memcpy_to_bvec(&dst_bv, src_buf); 1328f8b679a0SChristoph Hellwig kunmap_local(src_buf); 13296e6e811dSKent Overstreet 133022b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 133122b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 133245db54d5SKent Overstreet } 133345db54d5SKent Overstreet } 133445db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 133545db54d5SKent Overstreet 133645db54d5SKent Overstreet /** 133745db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 133845db54d5SKent Overstreet * @src: source bio 133945db54d5SKent Overstreet * @dst: destination bio 134045db54d5SKent Overstreet * 134145db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 134245db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 134345db54d5SKent Overstreet */ 134445db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 134545db54d5SKent Overstreet { 134645db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 134745db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 134845db54d5SKent Overstreet 134945db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 135045db54d5SKent Overstreet } 135145db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 135245db54d5SKent Overstreet 1353491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 13541dfa0f68SChristoph Hellwig { 13551dfa0f68SChristoph Hellwig struct bio_vec *bvec; 13566dc4f100SMing Lei struct bvec_iter_all iter_all; 13571dfa0f68SChristoph Hellwig 13582b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 13591dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 13601dfa0f68SChristoph Hellwig } 1361491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 13621dfa0f68SChristoph Hellwig 1363f9c78b2bSJens Axboe /* 1364f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1365f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1366f9c78b2bSJens Axboe * 1367f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1368f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1369f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1370f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1371f9c78b2bSJens Axboe * in process context. 1372f9c78b2bSJens Axboe * 1373f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1374f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1375f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1376f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1377f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1378f9c78b2bSJens Axboe * 1379f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1380f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1381f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1382f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1383f9c78b2bSJens Axboe * pagecache. 1384f9c78b2bSJens Axboe * 1385f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1386f9c78b2bSJens Axboe * deferred bio dirtying paths. 1387f9c78b2bSJens Axboe */ 1388f9c78b2bSJens Axboe 1389f9c78b2bSJens Axboe /* 1390f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1391f9c78b2bSJens Axboe */ 1392f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1393f9c78b2bSJens Axboe { 1394f9c78b2bSJens Axboe struct bio_vec *bvec; 13956dc4f100SMing Lei struct bvec_iter_all iter_all; 1396f9c78b2bSJens Axboe 13972b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 13983bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 13993bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1400f9c78b2bSJens Axboe } 1401f9c78b2bSJens Axboe } 1402f9c78b2bSJens Axboe 1403f9c78b2bSJens Axboe /* 1404f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1405f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1406f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 140724d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1408f9c78b2bSJens Axboe * 1409f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1410ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1411ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1412f9c78b2bSJens Axboe */ 1413f9c78b2bSJens Axboe 1414f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1415f9c78b2bSJens Axboe 1416f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1417f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1418f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1419f9c78b2bSJens Axboe 1420f9c78b2bSJens Axboe /* 1421f9c78b2bSJens Axboe * This runs in process context 1422f9c78b2bSJens Axboe */ 1423f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1424f9c78b2bSJens Axboe { 142524d5493fSChristoph Hellwig struct bio *bio, *next; 1426f9c78b2bSJens Axboe 142724d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 142824d5493fSChristoph Hellwig next = bio_dirty_list; 1429f9c78b2bSJens Axboe bio_dirty_list = NULL; 143024d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1431f9c78b2bSJens Axboe 143224d5493fSChristoph Hellwig while ((bio = next) != NULL) { 143324d5493fSChristoph Hellwig next = bio->bi_private; 1434f9c78b2bSJens Axboe 1435d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1436f9c78b2bSJens Axboe bio_put(bio); 1437f9c78b2bSJens Axboe } 1438f9c78b2bSJens Axboe } 1439f9c78b2bSJens Axboe 1440f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1441f9c78b2bSJens Axboe { 1442f9c78b2bSJens Axboe struct bio_vec *bvec; 144324d5493fSChristoph Hellwig unsigned long flags; 14446dc4f100SMing Lei struct bvec_iter_all iter_all; 1445f9c78b2bSJens Axboe 14462b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 144724d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 144824d5493fSChristoph Hellwig goto defer; 1449f9c78b2bSJens Axboe } 1450f9c78b2bSJens Axboe 1451d241a95fSChristoph Hellwig bio_release_pages(bio, false); 145224d5493fSChristoph Hellwig bio_put(bio); 145324d5493fSChristoph Hellwig return; 145424d5493fSChristoph Hellwig defer: 1455f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1456f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1457f9c78b2bSJens Axboe bio_dirty_list = bio; 1458f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1459f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1460f9c78b2bSJens Axboe } 1461f9c78b2bSJens Axboe 1462c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1463c4cf5261SJens Axboe { 1464c4cf5261SJens Axboe /* 1465c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1466c4cf5261SJens Axboe * we always end io on the first invocation. 1467c4cf5261SJens Axboe */ 1468c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1469c4cf5261SJens Axboe return true; 1470c4cf5261SJens Axboe 1471c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1472c4cf5261SJens Axboe 1473326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1474b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1475c4cf5261SJens Axboe return true; 1476326e1dbbSMike Snitzer } 1477c4cf5261SJens Axboe 1478c4cf5261SJens Axboe return false; 1479c4cf5261SJens Axboe } 1480c4cf5261SJens Axboe 1481f9c78b2bSJens Axboe /** 1482f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1483f9c78b2bSJens Axboe * @bio: bio 1484f9c78b2bSJens Axboe * 1485f9c78b2bSJens Axboe * Description: 14864246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 14874246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 14884246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1489fbbaf700SNeilBrown * 1490fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1491fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 149260b6a7e6SEdward Hsieh * last time. 1493f9c78b2bSJens Axboe **/ 14944246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1495f9c78b2bSJens Axboe { 1496ba8c6967SChristoph Hellwig again: 14972b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1498ba8c6967SChristoph Hellwig return; 14997c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 15007c20f116SChristoph Hellwig return; 1501f9c78b2bSJens Axboe 1502a647a524SMing Lei if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED)) 15033caee463SPavel Begunkov rq_qos_done_bio(bdev_get_queue(bio->bi_bdev), bio); 150467b42d0bSJosef Bacik 150560b6a7e6SEdward Hsieh if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 15063caee463SPavel Begunkov trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 150760b6a7e6SEdward Hsieh bio_clear_flag(bio, BIO_TRACE_COMPLETION); 150860b6a7e6SEdward Hsieh } 150960b6a7e6SEdward Hsieh 1510f9c78b2bSJens Axboe /* 1511ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1512ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1513ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1514ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1515ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1516ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1517f9c78b2bSJens Axboe */ 1518f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 151938f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1520ba8c6967SChristoph Hellwig goto again; 1521ba8c6967SChristoph Hellwig } 1522ba8c6967SChristoph Hellwig 15239e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1524b222dd2fSShaohua Li /* release cgroup info */ 1525b222dd2fSShaohua Li bio_uninit(bio); 1526f9c78b2bSJens Axboe if (bio->bi_end_io) 15274246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1528f9c78b2bSJens Axboe } 1529f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1530f9c78b2bSJens Axboe 1531f9c78b2bSJens Axboe /** 1532f9c78b2bSJens Axboe * bio_split - split a bio 1533f9c78b2bSJens Axboe * @bio: bio to split 1534f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1535f9c78b2bSJens Axboe * @gfp: gfp mask 1536f9c78b2bSJens Axboe * @bs: bio set to allocate from 1537f9c78b2bSJens Axboe * 1538f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1539f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1540f9c78b2bSJens Axboe * 1541f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1542dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1543dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1544f9c78b2bSJens Axboe */ 1545f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1546f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1547f9c78b2bSJens Axboe { 1548f341a4d3SMikulas Patocka struct bio *split; 1549f9c78b2bSJens Axboe 1550f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1551f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1552f9c78b2bSJens Axboe 15530512a75bSKeith Busch /* Zone append commands cannot be split */ 15540512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 15550512a75bSKeith Busch return NULL; 15560512a75bSKeith Busch 1557f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1558f9c78b2bSJens Axboe if (!split) 1559f9c78b2bSJens Axboe return NULL; 1560f9c78b2bSJens Axboe 1561f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1562f9c78b2bSJens Axboe 1563f9c78b2bSJens Axboe if (bio_integrity(split)) 1564fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1565f9c78b2bSJens Axboe 1566f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1567f9c78b2bSJens Axboe 1568fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 156920d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1570fbbaf700SNeilBrown 1571f9c78b2bSJens Axboe return split; 1572f9c78b2bSJens Axboe } 1573f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1574f9c78b2bSJens Axboe 1575f9c78b2bSJens Axboe /** 1576f9c78b2bSJens Axboe * bio_trim - trim a bio 1577f9c78b2bSJens Axboe * @bio: bio to trim 1578f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1579f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1580e83502caSChaitanya Kulkarni * 1581e83502caSChaitanya Kulkarni * This function is typically used for bios that are cloned and submitted 1582e83502caSChaitanya Kulkarni * to the underlying device in parts. 1583f9c78b2bSJens Axboe */ 1584e83502caSChaitanya Kulkarni void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1585f9c78b2bSJens Axboe { 1586e83502caSChaitanya Kulkarni if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 1587e83502caSChaitanya Kulkarni offset + size > bio->bi_iter.bi_size)) 1588e83502caSChaitanya Kulkarni return; 1589f9c78b2bSJens Axboe 1590f9c78b2bSJens Axboe size <<= 9; 1591f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1592f9c78b2bSJens Axboe return; 1593f9c78b2bSJens Axboe 1594f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1595f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1596376a78abSDmitry Monakhov 1597376a78abSDmitry Monakhov if (bio_integrity(bio)) 1598fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1599f9c78b2bSJens Axboe } 1600f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1601f9c78b2bSJens Axboe 1602f9c78b2bSJens Axboe /* 1603f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1604f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1605f9c78b2bSJens Axboe */ 16068aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1607f9c78b2bSJens Axboe { 16087a800a20SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1609f9c78b2bSJens Axboe 16108aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1611f9c78b2bSJens Axboe } 1612f9c78b2bSJens Axboe 1613917a38c7SKent Overstreet /* 1614917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1615917a38c7SKent Overstreet * 1616917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1617917a38c7SKent Overstreet * kzalloc()). 1618917a38c7SKent Overstreet */ 1619917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1620f9c78b2bSJens Axboe { 1621be4d234dSJens Axboe bio_alloc_cache_destroy(bs); 1622f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1623f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1624917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1625f9c78b2bSJens Axboe 16268aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 16278aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1628f9c78b2bSJens Axboe 1629f9c78b2bSJens Axboe bioset_integrity_free(bs); 1630917a38c7SKent Overstreet if (bs->bio_slab) 1631f9c78b2bSJens Axboe bio_put_slab(bs); 1632917a38c7SKent Overstreet bs->bio_slab = NULL; 1633917a38c7SKent Overstreet } 1634917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1635f9c78b2bSJens Axboe 1636011067b0SNeilBrown /** 1637917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1638dad08527SKent Overstreet * @bs: pool to initialize 1639917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1640917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1641917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1642917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1643917a38c7SKent Overstreet * 1644dad08527SKent Overstreet * Description: 1645dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1646dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1647dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1648dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1649dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1650dad08527SKent Overstreet * or things will break badly. 1651dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1652dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1653dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1654dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1655dad08527SKent Overstreet * 1656917a38c7SKent Overstreet */ 1657917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1658917a38c7SKent Overstreet unsigned int pool_size, 1659917a38c7SKent Overstreet unsigned int front_pad, 1660917a38c7SKent Overstreet int flags) 1661917a38c7SKent Overstreet { 1662917a38c7SKent Overstreet bs->front_pad = front_pad; 16639f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 16649f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 16659f180e31SMing Lei else 16669f180e31SMing Lei bs->back_pad = 0; 1667917a38c7SKent Overstreet 1668917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1669917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1670917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1671917a38c7SKent Overstreet 167249d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1673917a38c7SKent Overstreet if (!bs->bio_slab) 1674917a38c7SKent Overstreet return -ENOMEM; 1675917a38c7SKent Overstreet 1676917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1677917a38c7SKent Overstreet goto bad; 1678917a38c7SKent Overstreet 1679917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1680917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1681917a38c7SKent Overstreet goto bad; 1682917a38c7SKent Overstreet 1683be4d234dSJens Axboe if (flags & BIOSET_NEED_RESCUER) { 1684be4d234dSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", 1685be4d234dSJens Axboe WQ_MEM_RECLAIM, 0); 1686917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1687917a38c7SKent Overstreet goto bad; 1688be4d234dSJens Axboe } 1689be4d234dSJens Axboe if (flags & BIOSET_PERCPU_CACHE) { 1690be4d234dSJens Axboe bs->cache = alloc_percpu(struct bio_alloc_cache); 1691be4d234dSJens Axboe if (!bs->cache) 1692be4d234dSJens Axboe goto bad; 1693be4d234dSJens Axboe cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1694be4d234dSJens Axboe } 1695917a38c7SKent Overstreet 1696917a38c7SKent Overstreet return 0; 1697917a38c7SKent Overstreet bad: 1698917a38c7SKent Overstreet bioset_exit(bs); 1699917a38c7SKent Overstreet return -ENOMEM; 1700917a38c7SKent Overstreet } 1701917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1702917a38c7SKent Overstreet 170328e89fd9SJens Axboe /* 170428e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 170528e89fd9SJens Axboe * another bio_set. 170628e89fd9SJens Axboe */ 170728e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 170828e89fd9SJens Axboe { 170928e89fd9SJens Axboe int flags; 171028e89fd9SJens Axboe 171128e89fd9SJens Axboe flags = 0; 171228e89fd9SJens Axboe if (src->bvec_pool.min_nr) 171328e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 171428e89fd9SJens Axboe if (src->rescue_workqueue) 171528e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 171628e89fd9SJens Axboe 171728e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 171828e89fd9SJens Axboe } 171928e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 172028e89fd9SJens Axboe 1721be4d234dSJens Axboe /** 1722be4d234dSJens Axboe * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb 1723be4d234dSJens Axboe * @kiocb: kiocb describing the IO 17240ef47db1SJens Axboe * @nr_vecs: number of iovecs to pre-allocate 1725be4d234dSJens Axboe * @bs: bio_set to allocate from 1726be4d234dSJens Axboe * 1727be4d234dSJens Axboe * Description: 1728be4d234dSJens Axboe * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only 1729be4d234dSJens Axboe * used to check if we should dip into the per-cpu bio_set allocation 17303d5b3fbeSJens Axboe * cache. The allocation uses GFP_KERNEL internally. On return, the 17313d5b3fbeSJens Axboe * bio is marked BIO_PERCPU_CACHEABLE, and the final put of the bio 17323d5b3fbeSJens Axboe * MUST be done from process context, not hard/soft IRQ. 1733be4d234dSJens Axboe * 1734be4d234dSJens Axboe */ 1735be4d234dSJens Axboe struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs, 1736be4d234dSJens Axboe struct bio_set *bs) 1737be4d234dSJens Axboe { 1738be4d234dSJens Axboe struct bio_alloc_cache *cache; 1739be4d234dSJens Axboe struct bio *bio; 1740be4d234dSJens Axboe 1741be4d234dSJens Axboe if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS) 1742be4d234dSJens Axboe return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1743be4d234dSJens Axboe 1744be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, get_cpu()); 1745fcade2ceSJens Axboe if (cache->free_list) { 1746fcade2ceSJens Axboe bio = cache->free_list; 1747fcade2ceSJens Axboe cache->free_list = bio->bi_next; 1748be4d234dSJens Axboe cache->nr--; 1749be4d234dSJens Axboe put_cpu(); 1750be4d234dSJens Axboe bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs); 1751be4d234dSJens Axboe bio->bi_pool = bs; 1752be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1753be4d234dSJens Axboe return bio; 1754be4d234dSJens Axboe } 1755be4d234dSJens Axboe put_cpu(); 1756be4d234dSJens Axboe bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1757be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1758be4d234dSJens Axboe return bio; 1759be4d234dSJens Axboe } 1760be4d234dSJens Axboe EXPORT_SYMBOL_GPL(bio_alloc_kiocb); 1761be4d234dSJens Axboe 1762de76fd89SChristoph Hellwig static int __init init_bio(void) 1763f9c78b2bSJens Axboe { 1764f9c78b2bSJens Axboe int i; 1765f9c78b2bSJens Axboe 1766f9c78b2bSJens Axboe bio_integrity_init(); 1767de76fd89SChristoph Hellwig 1768de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1769f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1770f9c78b2bSJens Axboe 1771de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1772de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1773f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1774f9c78b2bSJens Axboe } 1775f9c78b2bSJens Axboe 1776be4d234dSJens Axboe cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1777be4d234dSJens Axboe bio_cpu_dead); 1778be4d234dSJens Axboe 1779f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1780f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1781f9c78b2bSJens Axboe 1782f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1783f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1784f9c78b2bSJens Axboe 1785f9c78b2bSJens Axboe return 0; 1786f9c78b2bSJens Axboe } 1787f9c78b2bSJens Axboe subsys_initcall(init_bio); 1788