18c16567dSChristoph Hellwig // SPDX-License-Identifier: GPL-2.0 2f9c78b2bSJens Axboe /* 3f9c78b2bSJens Axboe * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 4f9c78b2bSJens Axboe */ 5f9c78b2bSJens Axboe #include <linux/mm.h> 6f9c78b2bSJens Axboe #include <linux/swap.h> 7f9c78b2bSJens Axboe #include <linux/bio.h> 8f9c78b2bSJens Axboe #include <linux/blkdev.h> 9f9c78b2bSJens Axboe #include <linux/uio.h> 10f9c78b2bSJens Axboe #include <linux/iocontext.h> 11f9c78b2bSJens Axboe #include <linux/slab.h> 12f9c78b2bSJens Axboe #include <linux/init.h> 13f9c78b2bSJens Axboe #include <linux/kernel.h> 14f9c78b2bSJens Axboe #include <linux/export.h> 15f9c78b2bSJens Axboe #include <linux/mempool.h> 16f9c78b2bSJens Axboe #include <linux/workqueue.h> 17f9c78b2bSJens Axboe #include <linux/cgroup.h> 18b4c5875dSDamien Le Moal #include <linux/highmem.h> 19de6a78b6SMing Lei #include <linux/sched/sysctl.h> 20a892c8d5SSatya Tangirala #include <linux/blk-crypto.h> 2149d1ec85SMing Lei #include <linux/xarray.h> 22f9c78b2bSJens Axboe 23f9c78b2bSJens Axboe #include <trace/events/block.h> 249e234eeaSShaohua Li #include "blk.h" 2567b42d0bSJosef Bacik #include "blk-rq-qos.h" 26672fdcf0SMing Lei #include "blk-cgroup.h" 27f9c78b2bSJens Axboe 28b99182c5SPavel Begunkov #define ALLOC_CACHE_THRESHOLD 16 2942b2b2fbSPavel Begunkov #define ALLOC_CACHE_MAX 256 30b99182c5SPavel Begunkov 31be4d234dSJens Axboe struct bio_alloc_cache { 32fcade2ceSJens Axboe struct bio *free_list; 33b99182c5SPavel Begunkov struct bio *free_list_irq; 34be4d234dSJens Axboe unsigned int nr; 35b99182c5SPavel Begunkov unsigned int nr_irq; 36be4d234dSJens Axboe }; 37be4d234dSJens Axboe 38de76fd89SChristoph Hellwig static struct biovec_slab { 396ac0b715SChristoph Hellwig int nr_vecs; 406ac0b715SChristoph Hellwig char *name; 416ac0b715SChristoph Hellwig struct kmem_cache *slab; 42de76fd89SChristoph Hellwig } bvec_slabs[] __read_mostly = { 43de76fd89SChristoph Hellwig { .nr_vecs = 16, .name = "biovec-16" }, 44de76fd89SChristoph Hellwig { .nr_vecs = 64, .name = "biovec-64" }, 45de76fd89SChristoph Hellwig { .nr_vecs = 128, .name = "biovec-128" }, 46a8affc03SChristoph Hellwig { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" }, 47f9c78b2bSJens Axboe }; 486ac0b715SChristoph Hellwig 497a800a20SChristoph Hellwig static struct biovec_slab *biovec_slab(unsigned short nr_vecs) 507a800a20SChristoph Hellwig { 517a800a20SChristoph Hellwig switch (nr_vecs) { 527a800a20SChristoph Hellwig /* smaller bios use inline vecs */ 537a800a20SChristoph Hellwig case 5 ... 16: 547a800a20SChristoph Hellwig return &bvec_slabs[0]; 557a800a20SChristoph Hellwig case 17 ... 64: 567a800a20SChristoph Hellwig return &bvec_slabs[1]; 577a800a20SChristoph Hellwig case 65 ... 128: 587a800a20SChristoph Hellwig return &bvec_slabs[2]; 59a8affc03SChristoph Hellwig case 129 ... BIO_MAX_VECS: 607a800a20SChristoph Hellwig return &bvec_slabs[3]; 617a800a20SChristoph Hellwig default: 627a800a20SChristoph Hellwig BUG(); 637a800a20SChristoph Hellwig return NULL; 647a800a20SChristoph Hellwig } 657a800a20SChristoph Hellwig } 66f9c78b2bSJens Axboe 67f9c78b2bSJens Axboe /* 68f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 69f9c78b2bSJens Axboe * IO code that does not need private memory pools. 70f9c78b2bSJens Axboe */ 71f4f8154aSKent Overstreet struct bio_set fs_bio_set; 72f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 73f9c78b2bSJens Axboe 74f9c78b2bSJens Axboe /* 75f9c78b2bSJens Axboe * Our slab pool management 76f9c78b2bSJens Axboe */ 77f9c78b2bSJens Axboe struct bio_slab { 78f9c78b2bSJens Axboe struct kmem_cache *slab; 79f9c78b2bSJens Axboe unsigned int slab_ref; 80f9c78b2bSJens Axboe unsigned int slab_size; 81f9c78b2bSJens Axboe char name[8]; 82f9c78b2bSJens Axboe }; 83f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 8449d1ec85SMing Lei static DEFINE_XARRAY(bio_slabs); 85f9c78b2bSJens Axboe 8649d1ec85SMing Lei static struct bio_slab *create_bio_slab(unsigned int size) 87f9c78b2bSJens Axboe { 8849d1ec85SMing Lei struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); 8949d1ec85SMing Lei 9049d1ec85SMing Lei if (!bslab) 9149d1ec85SMing Lei return NULL; 9249d1ec85SMing Lei 9349d1ec85SMing Lei snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); 9449d1ec85SMing Lei bslab->slab = kmem_cache_create(bslab->name, size, 951a7e76e4SChristoph Hellwig ARCH_KMALLOC_MINALIGN, 961a7e76e4SChristoph Hellwig SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL); 9749d1ec85SMing Lei if (!bslab->slab) 9849d1ec85SMing Lei goto fail_alloc_slab; 9949d1ec85SMing Lei 10049d1ec85SMing Lei bslab->slab_ref = 1; 10149d1ec85SMing Lei bslab->slab_size = size; 10249d1ec85SMing Lei 10349d1ec85SMing Lei if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) 10449d1ec85SMing Lei return bslab; 10549d1ec85SMing Lei 10649d1ec85SMing Lei kmem_cache_destroy(bslab->slab); 10749d1ec85SMing Lei 10849d1ec85SMing Lei fail_alloc_slab: 10949d1ec85SMing Lei kfree(bslab); 11049d1ec85SMing Lei return NULL; 11149d1ec85SMing Lei } 11249d1ec85SMing Lei 11349d1ec85SMing Lei static inline unsigned int bs_bio_slab_size(struct bio_set *bs) 11449d1ec85SMing Lei { 1159f180e31SMing Lei return bs->front_pad + sizeof(struct bio) + bs->back_pad; 11649d1ec85SMing Lei } 11749d1ec85SMing Lei 11849d1ec85SMing Lei static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) 11949d1ec85SMing Lei { 12049d1ec85SMing Lei unsigned int size = bs_bio_slab_size(bs); 12149d1ec85SMing Lei struct bio_slab *bslab; 122f9c78b2bSJens Axboe 123f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 12449d1ec85SMing Lei bslab = xa_load(&bio_slabs, size); 12549d1ec85SMing Lei if (bslab) 126f9c78b2bSJens Axboe bslab->slab_ref++; 12749d1ec85SMing Lei else 12849d1ec85SMing Lei bslab = create_bio_slab(size); 129f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 13049d1ec85SMing Lei 13149d1ec85SMing Lei if (bslab) 13249d1ec85SMing Lei return bslab->slab; 13349d1ec85SMing Lei return NULL; 134f9c78b2bSJens Axboe } 135f9c78b2bSJens Axboe 136f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 137f9c78b2bSJens Axboe { 138f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 13949d1ec85SMing Lei unsigned int slab_size = bs_bio_slab_size(bs); 140f9c78b2bSJens Axboe 141f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 142f9c78b2bSJens Axboe 14349d1ec85SMing Lei bslab = xa_load(&bio_slabs, slab_size); 144f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 145f9c78b2bSJens Axboe goto out; 146f9c78b2bSJens Axboe 14749d1ec85SMing Lei WARN_ON_ONCE(bslab->slab != bs->bio_slab); 14849d1ec85SMing Lei 149f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 150f9c78b2bSJens Axboe 151f9c78b2bSJens Axboe if (--bslab->slab_ref) 152f9c78b2bSJens Axboe goto out; 153f9c78b2bSJens Axboe 15449d1ec85SMing Lei xa_erase(&bio_slabs, slab_size); 15549d1ec85SMing Lei 156f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 15749d1ec85SMing Lei kfree(bslab); 158f9c78b2bSJens Axboe 159f9c78b2bSJens Axboe out: 160f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 161f9c78b2bSJens Axboe } 162f9c78b2bSJens Axboe 1637a800a20SChristoph Hellwig void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs) 164f9c78b2bSJens Axboe { 1659e8c0d0dSChristoph Hellwig BUG_ON(nr_vecs > BIO_MAX_VECS); 166f9c78b2bSJens Axboe 167a8affc03SChristoph Hellwig if (nr_vecs == BIO_MAX_VECS) 168f9c78b2bSJens Axboe mempool_free(bv, pool); 1697a800a20SChristoph Hellwig else if (nr_vecs > BIO_INLINE_VECS) 1707a800a20SChristoph Hellwig kmem_cache_free(biovec_slab(nr_vecs)->slab, bv); 171f9c78b2bSJens Axboe } 172f9c78b2bSJens Axboe 173f2c3eb9bSChristoph Hellwig /* 174f2c3eb9bSChristoph Hellwig * Make the first allocation restricted and don't dump info on allocation 175f2c3eb9bSChristoph Hellwig * failures, since we'll fall back to the mempool in case of failure. 176f2c3eb9bSChristoph Hellwig */ 177f2c3eb9bSChristoph Hellwig static inline gfp_t bvec_alloc_gfp(gfp_t gfp) 178f9c78b2bSJens Axboe { 179f2c3eb9bSChristoph Hellwig return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | 180f2c3eb9bSChristoph Hellwig __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 181f2c3eb9bSChristoph Hellwig } 182f2c3eb9bSChristoph Hellwig 1837a800a20SChristoph Hellwig struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, 1847a800a20SChristoph Hellwig gfp_t gfp_mask) 185f9c78b2bSJens Axboe { 1867a800a20SChristoph Hellwig struct biovec_slab *bvs = biovec_slab(*nr_vecs); 1877a800a20SChristoph Hellwig 1887a800a20SChristoph Hellwig if (WARN_ON_ONCE(!bvs)) 189f9c78b2bSJens Axboe return NULL; 1907a800a20SChristoph Hellwig 1917a800a20SChristoph Hellwig /* 1927a800a20SChristoph Hellwig * Upgrade the nr_vecs request to take full advantage of the allocation. 1937a800a20SChristoph Hellwig * We also rely on this in the bvec_free path. 1947a800a20SChristoph Hellwig */ 1957a800a20SChristoph Hellwig *nr_vecs = bvs->nr_vecs; 196f9c78b2bSJens Axboe 197f9c78b2bSJens Axboe /* 198f007a3d6SChristoph Hellwig * Try a slab allocation first for all smaller allocations. If that 199f007a3d6SChristoph Hellwig * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. 200a8affc03SChristoph Hellwig * The mempool is sized to handle up to BIO_MAX_VECS entries. 201f9c78b2bSJens Axboe */ 202a8affc03SChristoph Hellwig if (*nr_vecs < BIO_MAX_VECS) { 203f9c78b2bSJens Axboe struct bio_vec *bvl; 204f9c78b2bSJens Axboe 205f2c3eb9bSChristoph Hellwig bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); 2067a800a20SChristoph Hellwig if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) 207f9c78b2bSJens Axboe return bvl; 208a8affc03SChristoph Hellwig *nr_vecs = BIO_MAX_VECS; 209f9c78b2bSJens Axboe } 210f9c78b2bSJens Axboe 211f007a3d6SChristoph Hellwig return mempool_alloc(pool, gfp_mask); 212f9c78b2bSJens Axboe } 213f9c78b2bSJens Axboe 2149ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 215f9c78b2bSJens Axboe { 216db9819c7SChristoph Hellwig #ifdef CONFIG_BLK_CGROUP 217db9819c7SChristoph Hellwig if (bio->bi_blkg) { 218db9819c7SChristoph Hellwig blkg_put(bio->bi_blkg); 219db9819c7SChristoph Hellwig bio->bi_blkg = NULL; 220db9819c7SChristoph Hellwig } 221db9819c7SChristoph Hellwig #endif 222ece841abSJustin Tee if (bio_integrity(bio)) 223ece841abSJustin Tee bio_integrity_free(bio); 224a892c8d5SSatya Tangirala 225a892c8d5SSatya Tangirala bio_crypt_free_ctx(bio); 226f9c78b2bSJens Axboe } 2279ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 228f9c78b2bSJens Axboe 229f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 230f9c78b2bSJens Axboe { 231f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 232066ff571SChristoph Hellwig void *p = bio; 233066ff571SChristoph Hellwig 234066ff571SChristoph Hellwig WARN_ON_ONCE(!bs); 235f9c78b2bSJens Axboe 2369ae3b3f5SJens Axboe bio_uninit(bio); 2377a800a20SChristoph Hellwig bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs); 238066ff571SChristoph Hellwig mempool_free(p - bs->front_pad, &bs->bio_pool); 239f9c78b2bSJens Axboe } 240f9c78b2bSJens Axboe 2419ae3b3f5SJens Axboe /* 2429ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2439ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2449ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2459ae3b3f5SJens Axboe */ 24649add496SChristoph Hellwig void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table, 24716458cf3SBart Van Assche unsigned short max_vecs, blk_opf_t opf) 248f9c78b2bSJens Axboe { 249da521626SJens Axboe bio->bi_next = NULL; 25049add496SChristoph Hellwig bio->bi_bdev = bdev; 25149add496SChristoph Hellwig bio->bi_opf = opf; 252da521626SJens Axboe bio->bi_flags = 0; 253da521626SJens Axboe bio->bi_ioprio = 0; 254da521626SJens Axboe bio->bi_status = 0; 255da521626SJens Axboe bio->bi_iter.bi_sector = 0; 256da521626SJens Axboe bio->bi_iter.bi_size = 0; 257da521626SJens Axboe bio->bi_iter.bi_idx = 0; 258da521626SJens Axboe bio->bi_iter.bi_bvec_done = 0; 259da521626SJens Axboe bio->bi_end_io = NULL; 260da521626SJens Axboe bio->bi_private = NULL; 261da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP 262da521626SJens Axboe bio->bi_blkg = NULL; 263da521626SJens Axboe bio->bi_issue.value = 0; 26449add496SChristoph Hellwig if (bdev) 26549add496SChristoph Hellwig bio_associate_blkg(bio); 266da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP_IOCOST 267da521626SJens Axboe bio->bi_iocost_cost = 0; 268da521626SJens Axboe #endif 269da521626SJens Axboe #endif 270da521626SJens Axboe #ifdef CONFIG_BLK_INLINE_ENCRYPTION 271da521626SJens Axboe bio->bi_crypt_context = NULL; 272da521626SJens Axboe #endif 273da521626SJens Axboe #ifdef CONFIG_BLK_DEV_INTEGRITY 274da521626SJens Axboe bio->bi_integrity = NULL; 275da521626SJens Axboe #endif 276da521626SJens Axboe bio->bi_vcnt = 0; 277da521626SJens Axboe 278c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 279dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2803e08773cSChristoph Hellwig bio->bi_cookie = BLK_QC_T_NONE; 2813a83f467SMing Lei 2823a83f467SMing Lei bio->bi_max_vecs = max_vecs; 283da521626SJens Axboe bio->bi_io_vec = table; 284da521626SJens Axboe bio->bi_pool = NULL; 285f9c78b2bSJens Axboe } 286f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 287f9c78b2bSJens Axboe 288f9c78b2bSJens Axboe /** 289f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 290f9c78b2bSJens Axboe * @bio: bio to reset 291a7c50c94SChristoph Hellwig * @bdev: block device to use the bio for 292a7c50c94SChristoph Hellwig * @opf: operation and flags for bio 293f9c78b2bSJens Axboe * 294f9c78b2bSJens Axboe * Description: 295f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 296f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 297f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 298f9c78b2bSJens Axboe * comment in struct bio. 299f9c78b2bSJens Axboe */ 30016458cf3SBart Van Assche void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf) 301f9c78b2bSJens Axboe { 3029ae3b3f5SJens Axboe bio_uninit(bio); 303f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 304c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 305a7c50c94SChristoph Hellwig bio->bi_bdev = bdev; 30678e34374SChristoph Hellwig if (bio->bi_bdev) 30778e34374SChristoph Hellwig bio_associate_blkg(bio); 308a7c50c94SChristoph Hellwig bio->bi_opf = opf; 309f9c78b2bSJens Axboe } 310f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 311f9c78b2bSJens Axboe 31238f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 313f9c78b2bSJens Axboe { 3144246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3154246a0b6SChristoph Hellwig 3163edf5346SYufen Yu if (bio->bi_status && !parent->bi_status) 3174e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 318f9c78b2bSJens Axboe bio_put(bio); 31938f8baaeSChristoph Hellwig return parent; 32038f8baaeSChristoph Hellwig } 32138f8baaeSChristoph Hellwig 32238f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 32338f8baaeSChristoph Hellwig { 32438f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 325f9c78b2bSJens Axboe } 326f9c78b2bSJens Axboe 327f9c78b2bSJens Axboe /** 328f9c78b2bSJens Axboe * bio_chain - chain bio completions 329f9c78b2bSJens Axboe * @bio: the target bio 3305b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 331f9c78b2bSJens Axboe * 332f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 333f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 334f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 335f9c78b2bSJens Axboe * 336f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 337f9c78b2bSJens Axboe */ 338f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 339f9c78b2bSJens Axboe { 340f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 341f9c78b2bSJens Axboe 342f9c78b2bSJens Axboe bio->bi_private = parent; 343f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 344c4cf5261SJens Axboe bio_inc_remaining(parent); 345f9c78b2bSJens Axboe } 346f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 347f9c78b2bSJens Axboe 3480a3140eaSChaitanya Kulkarni struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev, 34916458cf3SBart Van Assche unsigned int nr_pages, blk_opf_t opf, gfp_t gfp) 3503b005bf6SChristoph Hellwig { 35107888c66SChristoph Hellwig struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp); 3520a3140eaSChaitanya Kulkarni 3533b005bf6SChristoph Hellwig if (bio) { 3543b005bf6SChristoph Hellwig bio_chain(bio, new); 3553b005bf6SChristoph Hellwig submit_bio(bio); 3563b005bf6SChristoph Hellwig } 3573b005bf6SChristoph Hellwig 3583b005bf6SChristoph Hellwig return new; 3593b005bf6SChristoph Hellwig } 3603b005bf6SChristoph Hellwig EXPORT_SYMBOL_GPL(blk_next_bio); 3613b005bf6SChristoph Hellwig 362f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 363f9c78b2bSJens Axboe { 364f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 365f9c78b2bSJens Axboe struct bio *bio; 366f9c78b2bSJens Axboe 367f9c78b2bSJens Axboe while (1) { 368f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 369f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 370f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 371f9c78b2bSJens Axboe 372f9c78b2bSJens Axboe if (!bio) 373f9c78b2bSJens Axboe break; 374f9c78b2bSJens Axboe 375ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 376f9c78b2bSJens Axboe } 377f9c78b2bSJens Axboe } 378f9c78b2bSJens Axboe 379f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 380f9c78b2bSJens Axboe { 381f9c78b2bSJens Axboe struct bio_list punt, nopunt; 382f9c78b2bSJens Axboe struct bio *bio; 383f9c78b2bSJens Axboe 38447e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 38547e0fb46SNeilBrown return; 386f9c78b2bSJens Axboe /* 387f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 388f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 389f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 390f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 391f9c78b2bSJens Axboe * our own rescuer would be bad. 392f9c78b2bSJens Axboe * 393f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 394f9c78b2bSJens Axboe * remove from the middle of the list: 395f9c78b2bSJens Axboe */ 396f9c78b2bSJens Axboe 397f9c78b2bSJens Axboe bio_list_init(&punt); 398f9c78b2bSJens Axboe bio_list_init(&nopunt); 399f9c78b2bSJens Axboe 400f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 401f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 402f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 403f9c78b2bSJens Axboe 404f5fe1b51SNeilBrown bio_list_init(&nopunt); 405f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 406f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 407f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 408f9c78b2bSJens Axboe 409f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 410f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 411f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 412f9c78b2bSJens Axboe 413f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 414f9c78b2bSJens Axboe } 415f9c78b2bSJens Axboe 416b99182c5SPavel Begunkov static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache) 417b99182c5SPavel Begunkov { 418b99182c5SPavel Begunkov unsigned long flags; 419b99182c5SPavel Begunkov 420b99182c5SPavel Begunkov /* cache->free_list must be empty */ 421b99182c5SPavel Begunkov if (WARN_ON_ONCE(cache->free_list)) 422b99182c5SPavel Begunkov return; 423b99182c5SPavel Begunkov 424b99182c5SPavel Begunkov local_irq_save(flags); 425b99182c5SPavel Begunkov cache->free_list = cache->free_list_irq; 426b99182c5SPavel Begunkov cache->free_list_irq = NULL; 427b99182c5SPavel Begunkov cache->nr += cache->nr_irq; 428b99182c5SPavel Begunkov cache->nr_irq = 0; 429b99182c5SPavel Begunkov local_irq_restore(flags); 430b99182c5SPavel Begunkov } 431b99182c5SPavel Begunkov 4320df71650SMike Snitzer static struct bio *bio_alloc_percpu_cache(struct block_device *bdev, 43316458cf3SBart Van Assche unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp, 4340df71650SMike Snitzer struct bio_set *bs) 4350df71650SMike Snitzer { 4360df71650SMike Snitzer struct bio_alloc_cache *cache; 4370df71650SMike Snitzer struct bio *bio; 4380df71650SMike Snitzer 4390df71650SMike Snitzer cache = per_cpu_ptr(bs->cache, get_cpu()); 4400df71650SMike Snitzer if (!cache->free_list) { 441b99182c5SPavel Begunkov if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD) 442b99182c5SPavel Begunkov bio_alloc_irq_cache_splice(cache); 443b99182c5SPavel Begunkov if (!cache->free_list) { 4440df71650SMike Snitzer put_cpu(); 4450df71650SMike Snitzer return NULL; 4460df71650SMike Snitzer } 447b99182c5SPavel Begunkov } 4480df71650SMike Snitzer bio = cache->free_list; 4490df71650SMike Snitzer cache->free_list = bio->bi_next; 4500df71650SMike Snitzer cache->nr--; 4510df71650SMike Snitzer put_cpu(); 4520df71650SMike Snitzer 4530df71650SMike Snitzer bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf); 4540df71650SMike Snitzer bio->bi_pool = bs; 4550df71650SMike Snitzer return bio; 4560df71650SMike Snitzer } 4570df71650SMike Snitzer 458f9c78b2bSJens Axboe /** 459f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 460609be106SChristoph Hellwig * @bdev: block device to allocate the bio for (can be %NULL) 461609be106SChristoph Hellwig * @nr_vecs: number of bvecs to pre-allocate 462609be106SChristoph Hellwig * @opf: operation and flags for bio 463519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 464f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 465f9c78b2bSJens Axboe * 4663175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 467f9c78b2bSJens Axboe * 4683175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4693175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4703175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4713175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4723175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4733175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 474f9c78b2bSJens Axboe * 4753175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4763175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 477ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 478f9c78b2bSJens Axboe * stack overflows. 479f9c78b2bSJens Axboe * 4803175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4813175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 482f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 483f9c78b2bSJens Axboe * thread. 484f9c78b2bSJens Axboe * 485f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 486f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 487ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 488f9c78b2bSJens Axboe * for per bio allocations. 489f9c78b2bSJens Axboe * 4903175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 491f9c78b2bSJens Axboe */ 492609be106SChristoph Hellwig struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs, 49316458cf3SBart Van Assche blk_opf_t opf, gfp_t gfp_mask, 4947a88fa19SDan Carpenter struct bio_set *bs) 495f9c78b2bSJens Axboe { 496f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 497f9c78b2bSJens Axboe struct bio *bio; 498f9c78b2bSJens Axboe void *p; 499f9c78b2bSJens Axboe 500609be106SChristoph Hellwig /* should not use nobvec bioset for nr_vecs > 0 */ 501609be106SChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0)) 502f9c78b2bSJens Axboe return NULL; 503f9c78b2bSJens Axboe 5040df71650SMike Snitzer if (opf & REQ_ALLOC_CACHE) { 5050df71650SMike Snitzer if (bs->cache && nr_vecs <= BIO_INLINE_VECS) { 5060df71650SMike Snitzer bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf, 5070df71650SMike Snitzer gfp_mask, bs); 5080df71650SMike Snitzer if (bio) 5090df71650SMike Snitzer return bio; 5100df71650SMike Snitzer /* 5110df71650SMike Snitzer * No cached bio available, bio returned below marked with 5120df71650SMike Snitzer * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache. 5130df71650SMike Snitzer */ 5140df71650SMike Snitzer } else { 5150df71650SMike Snitzer opf &= ~REQ_ALLOC_CACHE; 5160df71650SMike Snitzer } 5170df71650SMike Snitzer } 5180df71650SMike Snitzer 519f9c78b2bSJens Axboe /* 5203175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 5213175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 5223175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 523f9c78b2bSJens Axboe * 5243175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 5253175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 5263175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 5273175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 5283175199aSChristoph Hellwig * mempool's reserve. 529f9c78b2bSJens Axboe * 530f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 5313175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 5323175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 5333175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 5343175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 5353175199aSChristoph Hellwig * gfp_flags. 536f9c78b2bSJens Axboe */ 537f5fe1b51SNeilBrown if (current->bio_list && 538f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 53947e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 54047e0fb46SNeilBrown bs->rescue_workqueue) 541d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 542f9c78b2bSJens Axboe 5438aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 544f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 545f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 546f9c78b2bSJens Axboe gfp_mask = saved_gfp; 5478aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 548f9c78b2bSJens Axboe } 549f9c78b2bSJens Axboe if (unlikely(!p)) 550f9c78b2bSJens Axboe return NULL; 551759aa12fSPavel Begunkov if (!mempool_is_saturated(&bs->bio_pool)) 552759aa12fSPavel Begunkov opf &= ~REQ_ALLOC_CACHE; 553f9c78b2bSJens Axboe 5543175199aSChristoph Hellwig bio = p + bs->front_pad; 555609be106SChristoph Hellwig if (nr_vecs > BIO_INLINE_VECS) { 5563175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 557f9c78b2bSJens Axboe 558609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 559f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 560f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 561f9c78b2bSJens Axboe gfp_mask = saved_gfp; 562609be106SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask); 563f9c78b2bSJens Axboe } 564f9c78b2bSJens Axboe if (unlikely(!bvl)) 565f9c78b2bSJens Axboe goto err_free; 566f9c78b2bSJens Axboe 56749add496SChristoph Hellwig bio_init(bio, bdev, bvl, nr_vecs, opf); 568609be106SChristoph Hellwig } else if (nr_vecs) { 56949add496SChristoph Hellwig bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf); 5703175199aSChristoph Hellwig } else { 57149add496SChristoph Hellwig bio_init(bio, bdev, NULL, 0, opf); 572f9c78b2bSJens Axboe } 573f9c78b2bSJens Axboe 574f9c78b2bSJens Axboe bio->bi_pool = bs; 575f9c78b2bSJens Axboe return bio; 576f9c78b2bSJens Axboe 577f9c78b2bSJens Axboe err_free: 5788aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 579f9c78b2bSJens Axboe return NULL; 580f9c78b2bSJens Axboe } 581f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 582f9c78b2bSJens Axboe 5833175199aSChristoph Hellwig /** 584066ff571SChristoph Hellwig * bio_kmalloc - kmalloc a bio 585066ff571SChristoph Hellwig * @nr_vecs: number of bio_vecs to allocate 5863175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 5873175199aSChristoph Hellwig * 588066ff571SChristoph Hellwig * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized 589066ff571SChristoph Hellwig * using bio_init() before use. To free a bio returned from this function use 590066ff571SChristoph Hellwig * kfree() after calling bio_uninit(). A bio returned from this function can 591066ff571SChristoph Hellwig * be reused by calling bio_uninit() before calling bio_init() again. 592066ff571SChristoph Hellwig * 593066ff571SChristoph Hellwig * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this 594340e1347SDeming Wang * function are not backed by a mempool can fail. Do not use this function 595066ff571SChristoph Hellwig * for allocations in the file system I/O path. 5963175199aSChristoph Hellwig * 5973175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 5983175199aSChristoph Hellwig */ 599066ff571SChristoph Hellwig struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask) 6003175199aSChristoph Hellwig { 6013175199aSChristoph Hellwig struct bio *bio; 6023175199aSChristoph Hellwig 603066ff571SChristoph Hellwig if (nr_vecs > UIO_MAXIOV) 6043175199aSChristoph Hellwig return NULL; 605066ff571SChristoph Hellwig return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask); 6063175199aSChristoph Hellwig } 6073175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 6083175199aSChristoph Hellwig 6096f822e1bSChristoph Hellwig void zero_fill_bio(struct bio *bio) 610f9c78b2bSJens Axboe { 611f9c78b2bSJens Axboe struct bio_vec bv; 612f9c78b2bSJens Axboe struct bvec_iter iter; 613f9c78b2bSJens Axboe 614ab6c340eSChristoph Hellwig bio_for_each_segment(bv, bio, iter) 615ab6c340eSChristoph Hellwig memzero_bvec(&bv); 616f9c78b2bSJens Axboe } 6176f822e1bSChristoph Hellwig EXPORT_SYMBOL(zero_fill_bio); 618f9c78b2bSJens Axboe 61983c9c547SMing Lei /** 62083c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 62183c9c547SMing Lei * @bio: the bio to be truncated 62283c9c547SMing Lei * @new_size: new size for truncating the bio 62383c9c547SMing Lei * 62483c9c547SMing Lei * Description: 62583c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 62683c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 62783c9c547SMing Lei * be used for handling corner cases, such as bio eod. 62883c9c547SMing Lei */ 6294f7ab09aSChristoph Hellwig static void bio_truncate(struct bio *bio, unsigned new_size) 63085a8ce62SMing Lei { 63185a8ce62SMing Lei struct bio_vec bv; 63285a8ce62SMing Lei struct bvec_iter iter; 63385a8ce62SMing Lei unsigned int done = 0; 63485a8ce62SMing Lei bool truncated = false; 63585a8ce62SMing Lei 63685a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 63785a8ce62SMing Lei return; 63885a8ce62SMing Lei 63983c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 64085a8ce62SMing Lei goto exit; 64185a8ce62SMing Lei 64285a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 64385a8ce62SMing Lei if (done + bv.bv_len > new_size) { 64485a8ce62SMing Lei unsigned offset; 64585a8ce62SMing Lei 64685a8ce62SMing Lei if (!truncated) 64785a8ce62SMing Lei offset = new_size - done; 64885a8ce62SMing Lei else 64985a8ce62SMing Lei offset = 0; 6503ee859e3SOGAWA Hirofumi zero_user(bv.bv_page, bv.bv_offset + offset, 6513ee859e3SOGAWA Hirofumi bv.bv_len - offset); 65285a8ce62SMing Lei truncated = true; 65385a8ce62SMing Lei } 65485a8ce62SMing Lei done += bv.bv_len; 65585a8ce62SMing Lei } 65685a8ce62SMing Lei 65785a8ce62SMing Lei exit: 65885a8ce62SMing Lei /* 65985a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 66085a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 66185a8ce62SMing Lei * in its .end_bio() callback. 66285a8ce62SMing Lei * 66385a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 66485a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 66585a8ce62SMing Lei */ 66685a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 66785a8ce62SMing Lei } 66885a8ce62SMing Lei 669f9c78b2bSJens Axboe /** 67029125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 67129125ed6SChristoph Hellwig * @bio: bio to truncate 67229125ed6SChristoph Hellwig * 67329125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 67429125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 67529125ed6SChristoph Hellwig * 67629125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 67729125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 67829125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 67929125ed6SChristoph Hellwig * sector" case. 68029125ed6SChristoph Hellwig */ 68129125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 68229125ed6SChristoph Hellwig { 683309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 68429125ed6SChristoph Hellwig 68529125ed6SChristoph Hellwig if (!maxsector) 68629125ed6SChristoph Hellwig return; 68729125ed6SChristoph Hellwig 68829125ed6SChristoph Hellwig /* 68929125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 69029125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 69129125ed6SChristoph Hellwig * an EIO. 69229125ed6SChristoph Hellwig */ 69329125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 69429125ed6SChristoph Hellwig return; 69529125ed6SChristoph Hellwig 69629125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 69729125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 69829125ed6SChristoph Hellwig return; 69929125ed6SChristoph Hellwig 70029125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 70129125ed6SChristoph Hellwig } 70229125ed6SChristoph Hellwig 703b99182c5SPavel Begunkov static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache, 704be4d234dSJens Axboe unsigned int nr) 705be4d234dSJens Axboe { 706be4d234dSJens Axboe unsigned int i = 0; 707be4d234dSJens Axboe struct bio *bio; 708be4d234dSJens Axboe 709fcade2ceSJens Axboe while ((bio = cache->free_list) != NULL) { 710fcade2ceSJens Axboe cache->free_list = bio->bi_next; 711be4d234dSJens Axboe cache->nr--; 712be4d234dSJens Axboe bio_free(bio); 713be4d234dSJens Axboe if (++i == nr) 714be4d234dSJens Axboe break; 715be4d234dSJens Axboe } 716b99182c5SPavel Begunkov return i; 717b99182c5SPavel Begunkov } 718b99182c5SPavel Begunkov 719b99182c5SPavel Begunkov static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 720b99182c5SPavel Begunkov unsigned int nr) 721b99182c5SPavel Begunkov { 722b99182c5SPavel Begunkov nr -= __bio_alloc_cache_prune(cache, nr); 723b99182c5SPavel Begunkov if (!READ_ONCE(cache->free_list)) { 724b99182c5SPavel Begunkov bio_alloc_irq_cache_splice(cache); 725b99182c5SPavel Begunkov __bio_alloc_cache_prune(cache, nr); 726b99182c5SPavel Begunkov } 727be4d234dSJens Axboe } 728be4d234dSJens Axboe 729be4d234dSJens Axboe static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 730be4d234dSJens Axboe { 731be4d234dSJens Axboe struct bio_set *bs; 732be4d234dSJens Axboe 733be4d234dSJens Axboe bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 734be4d234dSJens Axboe if (bs->cache) { 735be4d234dSJens Axboe struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 736be4d234dSJens Axboe 737be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 738be4d234dSJens Axboe } 739be4d234dSJens Axboe return 0; 740be4d234dSJens Axboe } 741be4d234dSJens Axboe 742be4d234dSJens Axboe static void bio_alloc_cache_destroy(struct bio_set *bs) 743be4d234dSJens Axboe { 744be4d234dSJens Axboe int cpu; 745be4d234dSJens Axboe 746be4d234dSJens Axboe if (!bs->cache) 747be4d234dSJens Axboe return; 748be4d234dSJens Axboe 749be4d234dSJens Axboe cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 750be4d234dSJens Axboe for_each_possible_cpu(cpu) { 751be4d234dSJens Axboe struct bio_alloc_cache *cache; 752be4d234dSJens Axboe 753be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, cpu); 754be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 755be4d234dSJens Axboe } 756be4d234dSJens Axboe free_percpu(bs->cache); 757605f7415SJens Axboe bs->cache = NULL; 758be4d234dSJens Axboe } 759be4d234dSJens Axboe 760f25cf75aSPavel Begunkov static inline void bio_put_percpu_cache(struct bio *bio) 761f25cf75aSPavel Begunkov { 762f25cf75aSPavel Begunkov struct bio_alloc_cache *cache; 763f25cf75aSPavel Begunkov 764f25cf75aSPavel Begunkov cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 765b99182c5SPavel Begunkov if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) { 766b99182c5SPavel Begunkov put_cpu(); 767b99182c5SPavel Begunkov bio_free(bio); 768b99182c5SPavel Begunkov return; 769b99182c5SPavel Begunkov } 770b99182c5SPavel Begunkov 771f25cf75aSPavel Begunkov bio_uninit(bio); 772f25cf75aSPavel Begunkov 773f25cf75aSPavel Begunkov if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) { 774f25cf75aSPavel Begunkov bio->bi_next = cache->free_list; 77511eb695fSJens Axboe bio->bi_bdev = NULL; 776f25cf75aSPavel Begunkov cache->free_list = bio; 777f25cf75aSPavel Begunkov cache->nr++; 778f25cf75aSPavel Begunkov } else { 779b99182c5SPavel Begunkov unsigned long flags; 780f25cf75aSPavel Begunkov 781b99182c5SPavel Begunkov local_irq_save(flags); 782b99182c5SPavel Begunkov bio->bi_next = cache->free_list_irq; 783b99182c5SPavel Begunkov cache->free_list_irq = bio; 784b99182c5SPavel Begunkov cache->nr_irq++; 785b99182c5SPavel Begunkov local_irq_restore(flags); 786b99182c5SPavel Begunkov } 787f25cf75aSPavel Begunkov put_cpu(); 788f25cf75aSPavel Begunkov } 789f25cf75aSPavel Begunkov 79029125ed6SChristoph Hellwig /** 791f9c78b2bSJens Axboe * bio_put - release a reference to a bio 792f9c78b2bSJens Axboe * @bio: bio to release reference to 793f9c78b2bSJens Axboe * 794f9c78b2bSJens Axboe * Description: 795f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 7969b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 797f9c78b2bSJens Axboe **/ 798f9c78b2bSJens Axboe void bio_put(struct bio *bio) 799f9c78b2bSJens Axboe { 800be4d234dSJens Axboe if (unlikely(bio_flagged(bio, BIO_REFFED))) { 8019e8c0d0dSChristoph Hellwig BUG_ON(!atomic_read(&bio->__bi_cnt)); 802be4d234dSJens Axboe if (!atomic_dec_and_test(&bio->__bi_cnt)) 803be4d234dSJens Axboe return; 804be4d234dSJens Axboe } 805f25cf75aSPavel Begunkov if (bio->bi_opf & REQ_ALLOC_CACHE) 806f25cf75aSPavel Begunkov bio_put_percpu_cache(bio); 807f25cf75aSPavel Begunkov else 808f9c78b2bSJens Axboe bio_free(bio); 809f9c78b2bSJens Axboe } 810f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 811f9c78b2bSJens Axboe 812a0e8de79SChristoph Hellwig static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp) 813f9c78b2bSJens Axboe { 814b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 815ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 816f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 81720bd723eSPaolo Valente 8187ecc56c6SChristoph Hellwig if (bio->bi_bdev) { 8197ecc56c6SChristoph Hellwig if (bio->bi_bdev == bio_src->bi_bdev && 8207ecc56c6SChristoph Hellwig bio_flagged(bio_src, BIO_REMAPPED)) 8217ecc56c6SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 822db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 8237ecc56c6SChristoph Hellwig } 82456b4b5abSChristoph Hellwig 82556b4b5abSChristoph Hellwig if (bio_crypt_clone(bio, bio_src, gfp) < 0) 82656b4b5abSChristoph Hellwig return -ENOMEM; 82756b4b5abSChristoph Hellwig if (bio_integrity(bio_src) && 82856b4b5abSChristoph Hellwig bio_integrity_clone(bio, bio_src, gfp) < 0) 82956b4b5abSChristoph Hellwig return -ENOMEM; 83056b4b5abSChristoph Hellwig return 0; 831f9c78b2bSJens Axboe } 832f9c78b2bSJens Axboe 833f9c78b2bSJens Axboe /** 834abfc426dSChristoph Hellwig * bio_alloc_clone - clone a bio that shares the original bio's biovec 835abfc426dSChristoph Hellwig * @bdev: block_device to clone onto 836a0e8de79SChristoph Hellwig * @bio_src: bio to clone from 837a0e8de79SChristoph Hellwig * @gfp: allocation priority 838f9c78b2bSJens Axboe * @bs: bio_set to allocate from 839f9c78b2bSJens Axboe * 840a0e8de79SChristoph Hellwig * Allocate a new bio that is a clone of @bio_src. The caller owns the returned 841a0e8de79SChristoph Hellwig * bio, but not the actual data it points to. 842a0e8de79SChristoph Hellwig * 843a0e8de79SChristoph Hellwig * The caller must ensure that the return bio is not freed before @bio_src. 844f9c78b2bSJens Axboe */ 845abfc426dSChristoph Hellwig struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src, 846abfc426dSChristoph Hellwig gfp_t gfp, struct bio_set *bs) 847f9c78b2bSJens Axboe { 848a0e8de79SChristoph Hellwig struct bio *bio; 849f9c78b2bSJens Axboe 850abfc426dSChristoph Hellwig bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs); 851a0e8de79SChristoph Hellwig if (!bio) 852f9c78b2bSJens Axboe return NULL; 853f9c78b2bSJens Axboe 854a0e8de79SChristoph Hellwig if (__bio_clone(bio, bio_src, gfp) < 0) { 855a0e8de79SChristoph Hellwig bio_put(bio); 85607560151SEric Biggers return NULL; 857f9c78b2bSJens Axboe } 858a0e8de79SChristoph Hellwig bio->bi_io_vec = bio_src->bi_io_vec; 85956b4b5abSChristoph Hellwig 860a0e8de79SChristoph Hellwig return bio; 86156b4b5abSChristoph Hellwig } 862abfc426dSChristoph Hellwig EXPORT_SYMBOL(bio_alloc_clone); 863f9c78b2bSJens Axboe 864a0e8de79SChristoph Hellwig /** 865abfc426dSChristoph Hellwig * bio_init_clone - clone a bio that shares the original bio's biovec 866abfc426dSChristoph Hellwig * @bdev: block_device to clone onto 867a0e8de79SChristoph Hellwig * @bio: bio to clone into 868a0e8de79SChristoph Hellwig * @bio_src: bio to clone from 869a0e8de79SChristoph Hellwig * @gfp: allocation priority 870a0e8de79SChristoph Hellwig * 871a0e8de79SChristoph Hellwig * Initialize a new bio in caller provided memory that is a clone of @bio_src. 872a0e8de79SChristoph Hellwig * The caller owns the returned bio, but not the actual data it points to. 873a0e8de79SChristoph Hellwig * 874a0e8de79SChristoph Hellwig * The caller must ensure that @bio_src is not freed before @bio. 875a0e8de79SChristoph Hellwig */ 876abfc426dSChristoph Hellwig int bio_init_clone(struct block_device *bdev, struct bio *bio, 877abfc426dSChristoph Hellwig struct bio *bio_src, gfp_t gfp) 878a0e8de79SChristoph Hellwig { 879a0e8de79SChristoph Hellwig int ret; 880a0e8de79SChristoph Hellwig 881abfc426dSChristoph Hellwig bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf); 882a0e8de79SChristoph Hellwig ret = __bio_clone(bio, bio_src, gfp); 883a0e8de79SChristoph Hellwig if (ret) 884a0e8de79SChristoph Hellwig bio_uninit(bio); 885a0e8de79SChristoph Hellwig return ret; 886a0e8de79SChristoph Hellwig } 887abfc426dSChristoph Hellwig EXPORT_SYMBOL(bio_init_clone); 888a0e8de79SChristoph Hellwig 8899a6083beSChristoph Hellwig /** 8909a6083beSChristoph Hellwig * bio_full - check if the bio is full 8919a6083beSChristoph Hellwig * @bio: bio to check 8929a6083beSChristoph Hellwig * @len: length of one segment to be added 8939a6083beSChristoph Hellwig * 8949a6083beSChristoph Hellwig * Return true if @bio is full and one segment with @len bytes can't be 8959a6083beSChristoph Hellwig * added to the bio, otherwise return false 8969a6083beSChristoph Hellwig */ 8979a6083beSChristoph Hellwig static inline bool bio_full(struct bio *bio, unsigned len) 8989a6083beSChristoph Hellwig { 8999a6083beSChristoph Hellwig if (bio->bi_vcnt >= bio->bi_max_vecs) 9009a6083beSChristoph Hellwig return true; 9019a6083beSChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) 9029a6083beSChristoph Hellwig return true; 9039a6083beSChristoph Hellwig return false; 9049a6083beSChristoph Hellwig } 9059a6083beSChristoph Hellwig 9065919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 9075919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 908ff896738SChristoph Hellwig bool *same_page) 9095919482eSMing Lei { 910d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 911d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 9125919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 9135919482eSMing Lei 9145919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 9155919482eSMing Lei return false; 9165919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 9175919482eSMing Lei return false; 91849580e69SLogan Gunthorpe if (!zone_device_pages_have_same_pgmap(bv->bv_page, page)) 91949580e69SLogan Gunthorpe return false; 92052d52d1cSChristoph Hellwig 921ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 922d8166519SMatthew Wilcox (Oracle) if (*same_page) 9235919482eSMing Lei return true; 92411b331f8SAlexander Potapenko else if (IS_ENABLED(CONFIG_KMSAN)) 92511b331f8SAlexander Potapenko return false; 926d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 9275919482eSMing Lei } 9285919482eSMing Lei 9299774b391SChristoph Hellwig /** 9309774b391SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 9319774b391SChristoph Hellwig * @bio: destination bio 9329774b391SChristoph Hellwig * @page: start page to add 9339774b391SChristoph Hellwig * @len: length of the data to add 9349774b391SChristoph Hellwig * @off: offset of the data relative to @page 9359774b391SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 9369774b391SChristoph Hellwig * 9379774b391SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 9389774b391SChristoph Hellwig * useful optimisation for file systems with a block size smaller than the 9399774b391SChristoph Hellwig * page size. 9409774b391SChristoph Hellwig * 9419774b391SChristoph Hellwig * Warn if (@len, @off) crosses pages in case that @same_page is true. 9429774b391SChristoph Hellwig * 9439774b391SChristoph Hellwig * Return %true on success or %false on failure. 9449774b391SChristoph Hellwig */ 9459774b391SChristoph Hellwig static bool __bio_try_merge_page(struct bio *bio, struct page *page, 9469774b391SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 9479774b391SChristoph Hellwig { 9489774b391SChristoph Hellwig if (bio->bi_vcnt > 0) { 9499774b391SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 9509774b391SChristoph Hellwig 9519774b391SChristoph Hellwig if (page_is_mergeable(bv, page, len, off, same_page)) { 9529774b391SChristoph Hellwig if (bio->bi_iter.bi_size > UINT_MAX - len) { 9539774b391SChristoph Hellwig *same_page = false; 9549774b391SChristoph Hellwig return false; 9559774b391SChristoph Hellwig } 9569774b391SChristoph Hellwig bv->bv_len += len; 9579774b391SChristoph Hellwig bio->bi_iter.bi_size += len; 9589774b391SChristoph Hellwig return true; 9599774b391SChristoph Hellwig } 9609774b391SChristoph Hellwig } 9619774b391SChristoph Hellwig return false; 9629774b391SChristoph Hellwig } 9639774b391SChristoph Hellwig 964e4581105SChristoph Hellwig /* 965e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 966e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 967e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 968e4581105SChristoph Hellwig */ 969e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 970e4581105SChristoph Hellwig struct page *page, unsigned len, 971e4581105SChristoph Hellwig unsigned offset, bool *same_page) 972489fbbcbSMing Lei { 973384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 974489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 975489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 976489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 977489fbbcbSMing Lei 978489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 979489fbbcbSMing Lei return false; 980489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 981489fbbcbSMing Lei return false; 982384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 983489fbbcbSMing Lei } 984489fbbcbSMing Lei 985f4595875SShaohua Li /** 986e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 987c66a14d0SKent Overstreet * @q: the target queue 988c66a14d0SKent Overstreet * @bio: destination bio 989c66a14d0SKent Overstreet * @page: page to add 990c66a14d0SKent Overstreet * @len: vec entry length 991c66a14d0SKent Overstreet * @offset: vec entry offset 992e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 993e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 994f9c78b2bSJens Axboe * 995e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 996e4581105SChristoph Hellwig * and gap limitations. 997f9c78b2bSJens Axboe */ 998e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 99919047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 1000e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 1001f9c78b2bSJens Axboe { 1002f9c78b2bSJens Axboe struct bio_vec *bvec; 1003f9c78b2bSJens Axboe 1004e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 1005f9c78b2bSJens Axboe return 0; 1006f9c78b2bSJens Axboe 10076850b2ddSChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> SECTOR_SHIFT) > max_sectors) 1008f9c78b2bSJens Axboe return 0; 1009f9c78b2bSJens Axboe 1010f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 1011e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 1012384209cdSChristoph Hellwig return len; 1013320ea869SChristoph Hellwig 1014cd1d83e2SChristoph Hellwig if (bio->bi_vcnt >= 1015cd1d83e2SChristoph Hellwig min(bio->bi_max_vecs, queue_max_segments(q))) 1016cd1d83e2SChristoph Hellwig return 0; 1017cd1d83e2SChristoph Hellwig 1018320ea869SChristoph Hellwig /* 1019320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 1020320ea869SChristoph Hellwig * would create a gap, disallow it. 1021320ea869SChristoph Hellwig */ 1022384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 1023c55ddd90SChristoph Hellwig if (bvec_gap_to_prev(&q->limits, bvec, offset)) 1024320ea869SChristoph Hellwig return 0; 1025f9c78b2bSJens Axboe } 1026f9c78b2bSJens Axboe 1027d58cdfaeSChristoph Hellwig bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset); 1028fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 1029dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 1030f9c78b2bSJens Axboe return len; 1031f9c78b2bSJens Axboe } 103219047087SMing Lei 1033e4581105SChristoph Hellwig /** 1034e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 1035e4581105SChristoph Hellwig * @q: the target queue 1036e4581105SChristoph Hellwig * @bio: destination bio 1037e4581105SChristoph Hellwig * @page: page to add 1038e4581105SChristoph Hellwig * @len: vec entry length 1039e4581105SChristoph Hellwig * @offset: vec entry offset 1040e4581105SChristoph Hellwig * 1041e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 1042e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 1043e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 1044e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 1045e4581105SChristoph Hellwig * 1046e4581105SChristoph Hellwig * This should only be used by passthrough bios. 1047e4581105SChristoph Hellwig */ 104819047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 104919047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 105019047087SMing Lei { 1051d1916c86SChristoph Hellwig bool same_page = false; 1052e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 1053e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 105419047087SMing Lei } 1055f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 1056f9c78b2bSJens Axboe 1057f9c78b2bSJens Axboe /** 1058ae29333fSJohannes Thumshirn * bio_add_zone_append_page - attempt to add page to zone-append bio 1059ae29333fSJohannes Thumshirn * @bio: destination bio 1060ae29333fSJohannes Thumshirn * @page: page to add 1061ae29333fSJohannes Thumshirn * @len: vec entry length 1062ae29333fSJohannes Thumshirn * @offset: vec entry offset 1063ae29333fSJohannes Thumshirn * 1064ae29333fSJohannes Thumshirn * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 1065ae29333fSJohannes Thumshirn * for a zone-append request. This can fail for a number of reasons, such as the 1066ae29333fSJohannes Thumshirn * bio being full or the target block device is not a zoned block device or 1067ae29333fSJohannes Thumshirn * other limitations of the target block device. The target block device must 1068ae29333fSJohannes Thumshirn * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 1069ae29333fSJohannes Thumshirn * to an empty bio. 1070ae29333fSJohannes Thumshirn * 1071ae29333fSJohannes Thumshirn * Returns: number of bytes added to the bio, or 0 in case of a failure. 1072ae29333fSJohannes Thumshirn */ 1073ae29333fSJohannes Thumshirn int bio_add_zone_append_page(struct bio *bio, struct page *page, 1074ae29333fSJohannes Thumshirn unsigned int len, unsigned int offset) 1075ae29333fSJohannes Thumshirn { 10763caee463SPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1077ae29333fSJohannes Thumshirn bool same_page = false; 1078ae29333fSJohannes Thumshirn 1079ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 1080ae29333fSJohannes Thumshirn return 0; 1081ae29333fSJohannes Thumshirn 1082edd1dbc8SChristoph Hellwig if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev))) 1083ae29333fSJohannes Thumshirn return 0; 1084ae29333fSJohannes Thumshirn 1085ae29333fSJohannes Thumshirn return bio_add_hw_page(q, bio, page, len, offset, 1086ae29333fSJohannes Thumshirn queue_max_zone_append_sectors(q), &same_page); 1087ae29333fSJohannes Thumshirn } 1088ae29333fSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 1089ae29333fSJohannes Thumshirn 1090ae29333fSJohannes Thumshirn /** 1091551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 10920aa69fd3SChristoph Hellwig * @bio: destination bio 1093551879a4SMing Lei * @page: start page to add 1094551879a4SMing Lei * @len: length of the data to add, may cross pages 1095551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 10960aa69fd3SChristoph Hellwig * 10970aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 10980aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 10990aa69fd3SChristoph Hellwig */ 11000aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 11010aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 11020aa69fd3SChristoph Hellwig { 11030aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 110479d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 11050aa69fd3SChristoph Hellwig 1106d58cdfaeSChristoph Hellwig bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off); 11070aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 11080aa69fd3SChristoph Hellwig bio->bi_vcnt++; 11090aa69fd3SChristoph Hellwig } 11100aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 11110aa69fd3SChristoph Hellwig 11120aa69fd3SChristoph Hellwig /** 1113551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 1114f9c78b2bSJens Axboe * @bio: destination bio 1115551879a4SMing Lei * @page: start page to add 1116551879a4SMing Lei * @len: vec entry length, may cross pages 1117551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 1118f9c78b2bSJens Axboe * 1119551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 1120c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1121f9c78b2bSJens Axboe */ 1122c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 1123c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 1124f9c78b2bSJens Axboe { 1125ff896738SChristoph Hellwig bool same_page = false; 1126ff896738SChristoph Hellwig 1127*939e1a37SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 1128*939e1a37SChristoph Hellwig return 0; 1129*939e1a37SChristoph Hellwig 1130ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 113179d08f89SMing Lei if (bio_full(bio, len)) 1132c66a14d0SKent Overstreet return 0; 11330aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 1134c66a14d0SKent Overstreet } 1135c66a14d0SKent Overstreet return len; 1136f9c78b2bSJens Axboe } 1137f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 1138f9c78b2bSJens Axboe 11397a150f1eSJohannes Thumshirn void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len, 11407a150f1eSJohannes Thumshirn size_t off) 11417a150f1eSJohannes Thumshirn { 11427a150f1eSJohannes Thumshirn WARN_ON_ONCE(len > UINT_MAX); 11437a150f1eSJohannes Thumshirn WARN_ON_ONCE(off > UINT_MAX); 11447a150f1eSJohannes Thumshirn __bio_add_page(bio, &folio->page, len, off); 11457a150f1eSJohannes Thumshirn } 11467a150f1eSJohannes Thumshirn 114785f5a74cSMatthew Wilcox (Oracle) /** 114885f5a74cSMatthew Wilcox (Oracle) * bio_add_folio - Attempt to add part of a folio to a bio. 114985f5a74cSMatthew Wilcox (Oracle) * @bio: BIO to add to. 115085f5a74cSMatthew Wilcox (Oracle) * @folio: Folio to add. 115185f5a74cSMatthew Wilcox (Oracle) * @len: How many bytes from the folio to add. 115285f5a74cSMatthew Wilcox (Oracle) * @off: First byte in this folio to add. 115385f5a74cSMatthew Wilcox (Oracle) * 115485f5a74cSMatthew Wilcox (Oracle) * Filesystems that use folios can call this function instead of calling 115585f5a74cSMatthew Wilcox (Oracle) * bio_add_page() for each page in the folio. If @off is bigger than 115685f5a74cSMatthew Wilcox (Oracle) * PAGE_SIZE, this function can create a bio_vec that starts in a page 115785f5a74cSMatthew Wilcox (Oracle) * after the bv_page. BIOs do not support folios that are 4GiB or larger. 115885f5a74cSMatthew Wilcox (Oracle) * 115985f5a74cSMatthew Wilcox (Oracle) * Return: Whether the addition was successful. 116085f5a74cSMatthew Wilcox (Oracle) */ 116185f5a74cSMatthew Wilcox (Oracle) bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len, 116285f5a74cSMatthew Wilcox (Oracle) size_t off) 116385f5a74cSMatthew Wilcox (Oracle) { 116485f5a74cSMatthew Wilcox (Oracle) if (len > UINT_MAX || off > UINT_MAX) 1165455a844dSJiapeng Chong return false; 116685f5a74cSMatthew Wilcox (Oracle) return bio_add_page(bio, &folio->page, len, off) > 0; 116785f5a74cSMatthew Wilcox (Oracle) } 1168cd57b771SMatthew Wilcox EXPORT_SYMBOL(bio_add_folio); 116985f5a74cSMatthew Wilcox (Oracle) 1170c809084aSPavel Begunkov void __bio_release_pages(struct bio *bio, bool mark_dirty) 11717321ecbfSChristoph Hellwig { 11727321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 11737321ecbfSChristoph Hellwig struct bio_vec *bvec; 11747321ecbfSChristoph Hellwig 1175d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1176d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 1177d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1178fd363244SDavid Howells bio_release_page(bio, bvec->bv_page); 11797321ecbfSChristoph Hellwig } 1180d241a95fSChristoph Hellwig } 1181c809084aSPavel Begunkov EXPORT_SYMBOL_GPL(__bio_release_pages); 11827321ecbfSChristoph Hellwig 11831bb6b810SPavel Begunkov void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 11846d0c48aeSJens Axboe { 1185fa5fa8ecSPavel Begunkov size_t size = iov_iter_count(iter); 1186fa5fa8ecSPavel Begunkov 11877a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_max_vecs); 11886d0c48aeSJens Axboe 1189fa5fa8ecSPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1190fa5fa8ecSPavel Begunkov struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1191fa5fa8ecSPavel Begunkov size_t max_sectors = queue_max_zone_append_sectors(q); 1192fa5fa8ecSPavel Begunkov 1193fa5fa8ecSPavel Begunkov size = min(size, max_sectors << SECTOR_SHIFT); 1194fa5fa8ecSPavel Begunkov } 1195fa5fa8ecSPavel Begunkov 1196c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 1197c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1198c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 1199fa5fa8ecSPavel Begunkov bio->bi_iter.bi_size = size; 1200977be012SChristoph Hellwig bio_set_flag(bio, BIO_CLONED); 12017de55b7dSJohannes Thumshirn } 12026d0c48aeSJens Axboe 1203c58c0074SKeith Busch static int bio_iov_add_page(struct bio *bio, struct page *page, 1204c58c0074SKeith Busch unsigned int len, unsigned int offset) 1205c58c0074SKeith Busch { 1206c58c0074SKeith Busch bool same_page = false; 1207c58c0074SKeith Busch 1208c58c0074SKeith Busch if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 1209c58c0074SKeith Busch __bio_add_page(bio, page, len, offset); 1210c58c0074SKeith Busch return 0; 1211c58c0074SKeith Busch } 1212c58c0074SKeith Busch 1213c58c0074SKeith Busch if (same_page) 1214a7e689ddSDavid Howells bio_release_page(bio, page); 1215c58c0074SKeith Busch return 0; 1216c58c0074SKeith Busch } 1217c58c0074SKeith Busch 1218c58c0074SKeith Busch static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page, 1219c58c0074SKeith Busch unsigned int len, unsigned int offset) 1220c58c0074SKeith Busch { 1221c58c0074SKeith Busch struct request_queue *q = bdev_get_queue(bio->bi_bdev); 1222c58c0074SKeith Busch bool same_page = false; 1223c58c0074SKeith Busch 1224c58c0074SKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 1225c58c0074SKeith Busch queue_max_zone_append_sectors(q), &same_page) != len) 1226c58c0074SKeith Busch return -EINVAL; 1227c58c0074SKeith Busch if (same_page) 1228a7e689ddSDavid Howells bio_release_page(bio, page); 1229c58c0074SKeith Busch return 0; 1230c58c0074SKeith Busch } 1231c58c0074SKeith Busch 1232576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1233576ed913SChristoph Hellwig 12342cefe4dbSKent Overstreet /** 123517d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 12362cefe4dbSKent Overstreet * @bio: bio to add pages to 12372cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 12382cefe4dbSKent Overstreet * 1239a7e689ddSDavid Howells * Extracts pages from *iter and appends them to @bio's bvec array. The pages 1240a7e689ddSDavid Howells * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag. 1241a7e689ddSDavid Howells * For a multi-segment *iter, this function only adds pages from the next 1242a7e689ddSDavid Howells * non-empty segment of the iov iterator. 12432cefe4dbSKent Overstreet */ 124417d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 12452cefe4dbSKent Overstreet { 1246f62e52d1SDavid Howells iov_iter_extraction_t extraction_flags = 0; 1247576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1248576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 12492cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 12502cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 1251576ed913SChristoph Hellwig ssize_t size, left; 1252e97424fdSKeith Busch unsigned len, i = 0; 1253480cb846SAl Viro size_t offset, trim; 1254325347d9SKeith Busch int ret = 0; 1255576ed913SChristoph Hellwig 1256576ed913SChristoph Hellwig /* 1257576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 1258576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 1259576ed913SChristoph Hellwig * without overwriting the temporary page array. 1260576ed913SChristoph Hellwig */ 1261576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1262576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 12632cefe4dbSKent Overstreet 12645e3e3f2eSLogan Gunthorpe if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue)) 1265f62e52d1SDavid Howells extraction_flags |= ITER_ALLOW_P2PDMA; 12665e3e3f2eSLogan Gunthorpe 1267b1a000d3SKeith Busch /* 1268b1a000d3SKeith Busch * Each segment in the iov is required to be a block size multiple. 1269b1a000d3SKeith Busch * However, we may not be able to get the entire segment if it spans 1270b1a000d3SKeith Busch * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the 1271b1a000d3SKeith Busch * result to ensure the bio's total size is correct. The remainder of 1272b1a000d3SKeith Busch * the iov data will be picked up in the next bio iteration. 1273b1a000d3SKeith Busch */ 1274a7e689ddSDavid Howells size = iov_iter_extract_pages(iter, &pages, 12755e3e3f2eSLogan Gunthorpe UINT_MAX - bio->bi_iter.bi_size, 1276a7e689ddSDavid Howells nr_pages, extraction_flags, &offset); 1277480cb846SAl Viro if (unlikely(size <= 0)) 1278480cb846SAl Viro return size ? size : -EFAULT; 1279e97424fdSKeith Busch 1280480cb846SAl Viro nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE); 1281480cb846SAl Viro 1282480cb846SAl Viro trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1); 1283480cb846SAl Viro iov_iter_revert(iter, trim); 1284480cb846SAl Viro 1285480cb846SAl Viro size -= trim; 1286480cb846SAl Viro if (unlikely(!size)) { 1287480cb846SAl Viro ret = -EFAULT; 1288e97424fdSKeith Busch goto out; 1289e97424fdSKeith Busch } 12902cefe4dbSKent Overstreet 1291576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1292576ed913SChristoph Hellwig struct page *page = pages[i]; 12932cefe4dbSKent Overstreet 1294576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 129534cdb8c8SKeith Busch if (bio_op(bio) == REQ_OP_ZONE_APPEND) { 1296c58c0074SKeith Busch ret = bio_iov_add_zone_append_page(bio, page, len, 1297c58c0074SKeith Busch offset); 1298e97424fdSKeith Busch if (ret) 1299325347d9SKeith Busch break; 130034cdb8c8SKeith Busch } else 130134cdb8c8SKeith Busch bio_iov_add_page(bio, page, len, offset); 130234cdb8c8SKeith Busch 1303576ed913SChristoph Hellwig offset = 0; 13042cefe4dbSKent Overstreet } 13052cefe4dbSKent Overstreet 1306480cb846SAl Viro iov_iter_revert(iter, left); 1307e97424fdSKeith Busch out: 1308e97424fdSKeith Busch while (i < nr_pages) 1309a7e689ddSDavid Howells bio_release_page(bio, pages[i++]); 1310e97424fdSKeith Busch 1311325347d9SKeith Busch return ret; 13122cefe4dbSKent Overstreet } 131317d51b10SMartin Wilck 131417d51b10SMartin Wilck /** 13156d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 131617d51b10SMartin Wilck * @bio: bio to add pages to 13176d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 131817d51b10SMartin Wilck * 13196d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 13206d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 13216d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1322c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1323c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1324c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1325c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1326c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1327c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 13286d0c48aeSJens Axboe * 132917d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 13305cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 13316d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 13326d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 133317d51b10SMartin Wilck */ 133417d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 133517d51b10SMartin Wilck { 1336c42bca92SPavel Begunkov int ret = 0; 133714eacf12SChristoph Hellwig 1338*939e1a37SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 1339*939e1a37SChristoph Hellwig return -EIO; 1340*939e1a37SChristoph Hellwig 1341c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1342fa5fa8ecSPavel Begunkov bio_iov_bvec_set(bio, iter); 1343fa5fa8ecSPavel Begunkov iov_iter_advance(iter, bio->bi_iter.bi_size); 1344fa5fa8ecSPavel Begunkov return 0; 134586004515SChristoph Hellwig } 134617d51b10SMartin Wilck 1347a7e689ddSDavid Howells if (iov_iter_extract_will_pin(iter)) 1348a7e689ddSDavid Howells bio_set_flag(bio, BIO_PAGE_PINNED); 134917d51b10SMartin Wilck do { 13506d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 135179d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 135217d51b10SMartin Wilck 135314eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 135417d51b10SMartin Wilck } 135529b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 13562cefe4dbSKent Overstreet 13574246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1358f9c78b2bSJens Axboe { 135965e53aabSChristoph Hellwig complete(bio->bi_private); 1360f9c78b2bSJens Axboe } 1361f9c78b2bSJens Axboe 1362f9c78b2bSJens Axboe /** 1363f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1364f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1365f9c78b2bSJens Axboe * 1366f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1367f9c78b2bSJens Axboe * bio_endio() on failure. 13683d289d68SJan Kara * 13693d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 13703d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 13713d289d68SJan Kara * on his own. 1372f9c78b2bSJens Axboe */ 13734e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1374f9c78b2bSJens Axboe { 1375309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1376309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1377de6a78b6SMing Lei unsigned long hang_check; 1378f9c78b2bSJens Axboe 137965e53aabSChristoph Hellwig bio->bi_private = &done; 1380f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 13811eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 13824e49ea4aSMike Christie submit_bio(bio); 1383de6a78b6SMing Lei 1384de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1385de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1386de6a78b6SMing Lei if (hang_check) 1387de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1388de6a78b6SMing Lei hang_check * (HZ/2))) 1389de6a78b6SMing Lei ; 1390de6a78b6SMing Lei else 139165e53aabSChristoph Hellwig wait_for_completion_io(&done); 1392f9c78b2bSJens Axboe 139365e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1394f9c78b2bSJens Axboe } 1395f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1396f9c78b2bSJens Axboe 1397d4aa57a1SJens Axboe void __bio_advance(struct bio *bio, unsigned bytes) 1398f9c78b2bSJens Axboe { 1399f9c78b2bSJens Axboe if (bio_integrity(bio)) 1400f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1401f9c78b2bSJens Axboe 1402a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1403f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1404f9c78b2bSJens Axboe } 1405d4aa57a1SJens Axboe EXPORT_SYMBOL(__bio_advance); 1406f9c78b2bSJens Axboe 1407ee4b4e22SJens Axboe void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 1408ee4b4e22SJens Axboe struct bio *src, struct bvec_iter *src_iter) 1409ee4b4e22SJens Axboe { 1410ee4b4e22SJens Axboe while (src_iter->bi_size && dst_iter->bi_size) { 1411ee4b4e22SJens Axboe struct bio_vec src_bv = bio_iter_iovec(src, *src_iter); 1412ee4b4e22SJens Axboe struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter); 1413ee4b4e22SJens Axboe unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len); 1414ee4b4e22SJens Axboe void *src_buf = bvec_kmap_local(&src_bv); 1415ee4b4e22SJens Axboe void *dst_buf = bvec_kmap_local(&dst_bv); 1416ee4b4e22SJens Axboe 1417ee4b4e22SJens Axboe memcpy(dst_buf, src_buf, bytes); 1418ee4b4e22SJens Axboe 1419ee4b4e22SJens Axboe kunmap_local(dst_buf); 1420ee4b4e22SJens Axboe kunmap_local(src_buf); 1421ee4b4e22SJens Axboe 1422ee4b4e22SJens Axboe bio_advance_iter_single(src, src_iter, bytes); 1423ee4b4e22SJens Axboe bio_advance_iter_single(dst, dst_iter, bytes); 1424ee4b4e22SJens Axboe } 1425ee4b4e22SJens Axboe } 1426ee4b4e22SJens Axboe EXPORT_SYMBOL(bio_copy_data_iter); 1427ee4b4e22SJens Axboe 142845db54d5SKent Overstreet /** 142945db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 143045db54d5SKent Overstreet * @src: source bio 143145db54d5SKent Overstreet * @dst: destination bio 143245db54d5SKent Overstreet * 143345db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 143445db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 143545db54d5SKent Overstreet */ 143645db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 143745db54d5SKent Overstreet { 143845db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 143945db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 144045db54d5SKent Overstreet 1441ee4b4e22SJens Axboe bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 144245db54d5SKent Overstreet } 144345db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 144445db54d5SKent Overstreet 1445491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 14461dfa0f68SChristoph Hellwig { 14471dfa0f68SChristoph Hellwig struct bio_vec *bvec; 14486dc4f100SMing Lei struct bvec_iter_all iter_all; 14491dfa0f68SChristoph Hellwig 14502b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 14511dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 14521dfa0f68SChristoph Hellwig } 1453491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 14541dfa0f68SChristoph Hellwig 1455f9c78b2bSJens Axboe /* 1456f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1457f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1458f9c78b2bSJens Axboe * 1459f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1460f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1461f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1462f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1463f9c78b2bSJens Axboe * in process context. 1464f9c78b2bSJens Axboe * 1465f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1466f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1467f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1468f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1469f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1470f9c78b2bSJens Axboe * 1471f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1472f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1473f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1474f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1475f9c78b2bSJens Axboe * pagecache. 1476f9c78b2bSJens Axboe * 1477f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1478f9c78b2bSJens Axboe * deferred bio dirtying paths. 1479f9c78b2bSJens Axboe */ 1480f9c78b2bSJens Axboe 1481f9c78b2bSJens Axboe /* 1482f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1483f9c78b2bSJens Axboe */ 1484f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1485f9c78b2bSJens Axboe { 1486f9c78b2bSJens Axboe struct bio_vec *bvec; 14876dc4f100SMing Lei struct bvec_iter_all iter_all; 1488f9c78b2bSJens Axboe 14892b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 14903bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 14913bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1492f9c78b2bSJens Axboe } 1493f9c78b2bSJens Axboe } 1494f9c78b2bSJens Axboe 1495f9c78b2bSJens Axboe /* 1496f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1497f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1498f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 149924d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1500f9c78b2bSJens Axboe * 1501f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1502fd363244SDavid Howells * here on. It will unpin each page and will run one bio_put() against the 1503fd363244SDavid Howells * BIO. 1504f9c78b2bSJens Axboe */ 1505f9c78b2bSJens Axboe 1506f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1507f9c78b2bSJens Axboe 1508f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1509f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1510f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1511f9c78b2bSJens Axboe 1512f9c78b2bSJens Axboe /* 1513f9c78b2bSJens Axboe * This runs in process context 1514f9c78b2bSJens Axboe */ 1515f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1516f9c78b2bSJens Axboe { 151724d5493fSChristoph Hellwig struct bio *bio, *next; 1518f9c78b2bSJens Axboe 151924d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 152024d5493fSChristoph Hellwig next = bio_dirty_list; 1521f9c78b2bSJens Axboe bio_dirty_list = NULL; 152224d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1523f9c78b2bSJens Axboe 152424d5493fSChristoph Hellwig while ((bio = next) != NULL) { 152524d5493fSChristoph Hellwig next = bio->bi_private; 1526f9c78b2bSJens Axboe 1527d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1528f9c78b2bSJens Axboe bio_put(bio); 1529f9c78b2bSJens Axboe } 1530f9c78b2bSJens Axboe } 1531f9c78b2bSJens Axboe 1532f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1533f9c78b2bSJens Axboe { 1534f9c78b2bSJens Axboe struct bio_vec *bvec; 153524d5493fSChristoph Hellwig unsigned long flags; 15366dc4f100SMing Lei struct bvec_iter_all iter_all; 1537f9c78b2bSJens Axboe 15382b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 153924d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 154024d5493fSChristoph Hellwig goto defer; 1541f9c78b2bSJens Axboe } 1542f9c78b2bSJens Axboe 1543d241a95fSChristoph Hellwig bio_release_pages(bio, false); 154424d5493fSChristoph Hellwig bio_put(bio); 154524d5493fSChristoph Hellwig return; 154624d5493fSChristoph Hellwig defer: 1547f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1548f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1549f9c78b2bSJens Axboe bio_dirty_list = bio; 1550f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1551f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1552f9c78b2bSJens Axboe } 1553f9c78b2bSJens Axboe 1554c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1555c4cf5261SJens Axboe { 1556c4cf5261SJens Axboe /* 1557c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1558c4cf5261SJens Axboe * we always end io on the first invocation. 1559c4cf5261SJens Axboe */ 1560c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1561c4cf5261SJens Axboe return true; 1562c4cf5261SJens Axboe 1563c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1564c4cf5261SJens Axboe 1565326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1566b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1567c4cf5261SJens Axboe return true; 1568326e1dbbSMike Snitzer } 1569c4cf5261SJens Axboe 1570c4cf5261SJens Axboe return false; 1571c4cf5261SJens Axboe } 1572c4cf5261SJens Axboe 1573f9c78b2bSJens Axboe /** 1574f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1575f9c78b2bSJens Axboe * @bio: bio 1576f9c78b2bSJens Axboe * 1577f9c78b2bSJens Axboe * Description: 15784246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 15794246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 15804246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1581fbbaf700SNeilBrown * 1582fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1583fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 158460b6a7e6SEdward Hsieh * last time. 1585f9c78b2bSJens Axboe **/ 15864246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1587f9c78b2bSJens Axboe { 1588ba8c6967SChristoph Hellwig again: 15892b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1590ba8c6967SChristoph Hellwig return; 15917c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 15927c20f116SChristoph Hellwig return; 1593f9c78b2bSJens Axboe 1594aa1b46dcSTejun Heo rq_qos_done_bio(bio); 159567b42d0bSJosef Bacik 159660b6a7e6SEdward Hsieh if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 15973caee463SPavel Begunkov trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio); 159860b6a7e6SEdward Hsieh bio_clear_flag(bio, BIO_TRACE_COMPLETION); 159960b6a7e6SEdward Hsieh } 160060b6a7e6SEdward Hsieh 1601f9c78b2bSJens Axboe /* 1602ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1603ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1604ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1605ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1606ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1607ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1608f9c78b2bSJens Axboe */ 1609f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 161038f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1611ba8c6967SChristoph Hellwig goto again; 1612ba8c6967SChristoph Hellwig } 1613ba8c6967SChristoph Hellwig 16149e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1615b222dd2fSShaohua Li /* release cgroup info */ 1616b222dd2fSShaohua Li bio_uninit(bio); 1617f9c78b2bSJens Axboe if (bio->bi_end_io) 16184246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1619f9c78b2bSJens Axboe } 1620f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1621f9c78b2bSJens Axboe 1622f9c78b2bSJens Axboe /** 1623f9c78b2bSJens Axboe * bio_split - split a bio 1624f9c78b2bSJens Axboe * @bio: bio to split 1625f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1626f9c78b2bSJens Axboe * @gfp: gfp mask 1627f9c78b2bSJens Axboe * @bs: bio set to allocate from 1628f9c78b2bSJens Axboe * 1629f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1630f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1631f9c78b2bSJens Axboe * 1632f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1633dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1634dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1635f9c78b2bSJens Axboe */ 1636f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1637f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1638f9c78b2bSJens Axboe { 1639f341a4d3SMikulas Patocka struct bio *split; 1640f9c78b2bSJens Axboe 1641f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1642f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1643f9c78b2bSJens Axboe 16440512a75bSKeith Busch /* Zone append commands cannot be split */ 16450512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 16460512a75bSKeith Busch return NULL; 16470512a75bSKeith Busch 1648abfc426dSChristoph Hellwig split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs); 1649f9c78b2bSJens Axboe if (!split) 1650f9c78b2bSJens Axboe return NULL; 1651f9c78b2bSJens Axboe 1652f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1653f9c78b2bSJens Axboe 1654f9c78b2bSJens Axboe if (bio_integrity(split)) 1655fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1656f9c78b2bSJens Axboe 1657f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1658f9c78b2bSJens Axboe 1659fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 166020d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1661fbbaf700SNeilBrown 1662f9c78b2bSJens Axboe return split; 1663f9c78b2bSJens Axboe } 1664f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1665f9c78b2bSJens Axboe 1666f9c78b2bSJens Axboe /** 1667f9c78b2bSJens Axboe * bio_trim - trim a bio 1668f9c78b2bSJens Axboe * @bio: bio to trim 1669f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1670f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1671e83502caSChaitanya Kulkarni * 1672e83502caSChaitanya Kulkarni * This function is typically used for bios that are cloned and submitted 1673e83502caSChaitanya Kulkarni * to the underlying device in parts. 1674f9c78b2bSJens Axboe */ 1675e83502caSChaitanya Kulkarni void bio_trim(struct bio *bio, sector_t offset, sector_t size) 1676f9c78b2bSJens Axboe { 1677e83502caSChaitanya Kulkarni if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS || 16788535c018SMing Lei offset + size > bio_sectors(bio))) 1679e83502caSChaitanya Kulkarni return; 1680f9c78b2bSJens Axboe 1681f9c78b2bSJens Axboe size <<= 9; 1682f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1683f9c78b2bSJens Axboe return; 1684f9c78b2bSJens Axboe 1685f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1686f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1687376a78abSDmitry Monakhov 1688376a78abSDmitry Monakhov if (bio_integrity(bio)) 1689fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1690f9c78b2bSJens Axboe } 1691f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1692f9c78b2bSJens Axboe 1693f9c78b2bSJens Axboe /* 1694f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1695f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1696f9c78b2bSJens Axboe */ 16978aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1698f9c78b2bSJens Axboe { 16997a800a20SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1700f9c78b2bSJens Axboe 17018aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1702f9c78b2bSJens Axboe } 1703f9c78b2bSJens Axboe 1704917a38c7SKent Overstreet /* 1705917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1706917a38c7SKent Overstreet * 1707917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1708917a38c7SKent Overstreet * kzalloc()). 1709917a38c7SKent Overstreet */ 1710917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1711f9c78b2bSJens Axboe { 1712be4d234dSJens Axboe bio_alloc_cache_destroy(bs); 1713f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1714f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1715917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1716f9c78b2bSJens Axboe 17178aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 17188aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1719f9c78b2bSJens Axboe 1720f9c78b2bSJens Axboe bioset_integrity_free(bs); 1721917a38c7SKent Overstreet if (bs->bio_slab) 1722f9c78b2bSJens Axboe bio_put_slab(bs); 1723917a38c7SKent Overstreet bs->bio_slab = NULL; 1724917a38c7SKent Overstreet } 1725917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1726f9c78b2bSJens Axboe 1727011067b0SNeilBrown /** 1728917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1729dad08527SKent Overstreet * @bs: pool to initialize 1730917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1731917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1732917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1733917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1734917a38c7SKent Overstreet * 1735dad08527SKent Overstreet * Description: 1736dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1737dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1738dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1739dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1740dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1741dad08527SKent Overstreet * or things will break badly. 1742dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1743abfc426dSChristoph Hellwig * for allocating iovecs. This pool is not needed e.g. for bio_init_clone(). 1744abfc426dSChristoph Hellwig * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used 1745abfc426dSChristoph Hellwig * to dispatch queued requests when the mempool runs out of space. 1746dad08527SKent Overstreet * 1747917a38c7SKent Overstreet */ 1748917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1749917a38c7SKent Overstreet unsigned int pool_size, 1750917a38c7SKent Overstreet unsigned int front_pad, 1751917a38c7SKent Overstreet int flags) 1752917a38c7SKent Overstreet { 1753917a38c7SKent Overstreet bs->front_pad = front_pad; 17549f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 17559f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 17569f180e31SMing Lei else 17579f180e31SMing Lei bs->back_pad = 0; 1758917a38c7SKent Overstreet 1759917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1760917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1761917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1762917a38c7SKent Overstreet 176349d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1764917a38c7SKent Overstreet if (!bs->bio_slab) 1765917a38c7SKent Overstreet return -ENOMEM; 1766917a38c7SKent Overstreet 1767917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1768917a38c7SKent Overstreet goto bad; 1769917a38c7SKent Overstreet 1770917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1771917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1772917a38c7SKent Overstreet goto bad; 1773917a38c7SKent Overstreet 1774be4d234dSJens Axboe if (flags & BIOSET_NEED_RESCUER) { 1775be4d234dSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", 1776be4d234dSJens Axboe WQ_MEM_RECLAIM, 0); 1777917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1778917a38c7SKent Overstreet goto bad; 1779be4d234dSJens Axboe } 1780be4d234dSJens Axboe if (flags & BIOSET_PERCPU_CACHE) { 1781be4d234dSJens Axboe bs->cache = alloc_percpu(struct bio_alloc_cache); 1782be4d234dSJens Axboe if (!bs->cache) 1783be4d234dSJens Axboe goto bad; 1784be4d234dSJens Axboe cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1785be4d234dSJens Axboe } 1786917a38c7SKent Overstreet 1787917a38c7SKent Overstreet return 0; 1788917a38c7SKent Overstreet bad: 1789917a38c7SKent Overstreet bioset_exit(bs); 1790917a38c7SKent Overstreet return -ENOMEM; 1791917a38c7SKent Overstreet } 1792917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1793917a38c7SKent Overstreet 1794de76fd89SChristoph Hellwig static int __init init_bio(void) 1795f9c78b2bSJens Axboe { 1796f9c78b2bSJens Axboe int i; 1797f9c78b2bSJens Axboe 1798a3df2e45SJens Axboe BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags)); 1799a3df2e45SJens Axboe 1800f9c78b2bSJens Axboe bio_integrity_init(); 1801de76fd89SChristoph Hellwig 1802de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1803f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1804f9c78b2bSJens Axboe 1805de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1806de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1807f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1808f9c78b2bSJens Axboe } 1809f9c78b2bSJens Axboe 1810be4d234dSJens Axboe cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1811be4d234dSJens Axboe bio_cpu_dead); 1812be4d234dSJens Axboe 181312c5b70cSJens Axboe if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, 181412c5b70cSJens Axboe BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE)) 1815f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1816f9c78b2bSJens Axboe 1817f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1818f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1819f9c78b2bSJens Axboe 1820f9c78b2bSJens Axboe return 0; 1821f9c78b2bSJens Axboe } 1822f9c78b2bSJens Axboe subsys_initcall(init_bio); 1823