18c16567dSChristoph Hellwig // SPDX-License-Identifier: GPL-2.0 2f9c78b2bSJens Axboe /* 3f9c78b2bSJens Axboe * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 4f9c78b2bSJens Axboe */ 5f9c78b2bSJens Axboe #include <linux/mm.h> 6f9c78b2bSJens Axboe #include <linux/swap.h> 7f9c78b2bSJens Axboe #include <linux/bio.h> 8f9c78b2bSJens Axboe #include <linux/blkdev.h> 9f9c78b2bSJens Axboe #include <linux/uio.h> 10f9c78b2bSJens Axboe #include <linux/iocontext.h> 11f9c78b2bSJens Axboe #include <linux/slab.h> 12f9c78b2bSJens Axboe #include <linux/init.h> 13f9c78b2bSJens Axboe #include <linux/kernel.h> 14f9c78b2bSJens Axboe #include <linux/export.h> 15f9c78b2bSJens Axboe #include <linux/mempool.h> 16f9c78b2bSJens Axboe #include <linux/workqueue.h> 17f9c78b2bSJens Axboe #include <linux/cgroup.h> 1808e18eabSJosef Bacik #include <linux/blk-cgroup.h> 19b4c5875dSDamien Le Moal #include <linux/highmem.h> 20de6a78b6SMing Lei #include <linux/sched/sysctl.h> 21a892c8d5SSatya Tangirala #include <linux/blk-crypto.h> 2249d1ec85SMing Lei #include <linux/xarray.h> 23f9c78b2bSJens Axboe 24f9c78b2bSJens Axboe #include <trace/events/block.h> 259e234eeaSShaohua Li #include "blk.h" 2667b42d0bSJosef Bacik #include "blk-rq-qos.h" 27f9c78b2bSJens Axboe 28*be4d234dSJens Axboe struct bio_alloc_cache { 29*be4d234dSJens Axboe struct bio_list free_list; 30*be4d234dSJens Axboe unsigned int nr; 31*be4d234dSJens Axboe }; 32*be4d234dSJens Axboe 33de76fd89SChristoph Hellwig static struct biovec_slab { 346ac0b715SChristoph Hellwig int nr_vecs; 356ac0b715SChristoph Hellwig char *name; 366ac0b715SChristoph Hellwig struct kmem_cache *slab; 37de76fd89SChristoph Hellwig } bvec_slabs[] __read_mostly = { 38de76fd89SChristoph Hellwig { .nr_vecs = 16, .name = "biovec-16" }, 39de76fd89SChristoph Hellwig { .nr_vecs = 64, .name = "biovec-64" }, 40de76fd89SChristoph Hellwig { .nr_vecs = 128, .name = "biovec-128" }, 41a8affc03SChristoph Hellwig { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" }, 42f9c78b2bSJens Axboe }; 436ac0b715SChristoph Hellwig 447a800a20SChristoph Hellwig static struct biovec_slab *biovec_slab(unsigned short nr_vecs) 457a800a20SChristoph Hellwig { 467a800a20SChristoph Hellwig switch (nr_vecs) { 477a800a20SChristoph Hellwig /* smaller bios use inline vecs */ 487a800a20SChristoph Hellwig case 5 ... 16: 497a800a20SChristoph Hellwig return &bvec_slabs[0]; 507a800a20SChristoph Hellwig case 17 ... 64: 517a800a20SChristoph Hellwig return &bvec_slabs[1]; 527a800a20SChristoph Hellwig case 65 ... 128: 537a800a20SChristoph Hellwig return &bvec_slabs[2]; 54a8affc03SChristoph Hellwig case 129 ... BIO_MAX_VECS: 557a800a20SChristoph Hellwig return &bvec_slabs[3]; 567a800a20SChristoph Hellwig default: 577a800a20SChristoph Hellwig BUG(); 587a800a20SChristoph Hellwig return NULL; 597a800a20SChristoph Hellwig } 607a800a20SChristoph Hellwig } 61f9c78b2bSJens Axboe 62f9c78b2bSJens Axboe /* 63f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 64f9c78b2bSJens Axboe * IO code that does not need private memory pools. 65f9c78b2bSJens Axboe */ 66f4f8154aSKent Overstreet struct bio_set fs_bio_set; 67f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 68f9c78b2bSJens Axboe 69f9c78b2bSJens Axboe /* 70f9c78b2bSJens Axboe * Our slab pool management 71f9c78b2bSJens Axboe */ 72f9c78b2bSJens Axboe struct bio_slab { 73f9c78b2bSJens Axboe struct kmem_cache *slab; 74f9c78b2bSJens Axboe unsigned int slab_ref; 75f9c78b2bSJens Axboe unsigned int slab_size; 76f9c78b2bSJens Axboe char name[8]; 77f9c78b2bSJens Axboe }; 78f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 7949d1ec85SMing Lei static DEFINE_XARRAY(bio_slabs); 80f9c78b2bSJens Axboe 8149d1ec85SMing Lei static struct bio_slab *create_bio_slab(unsigned int size) 82f9c78b2bSJens Axboe { 8349d1ec85SMing Lei struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); 8449d1ec85SMing Lei 8549d1ec85SMing Lei if (!bslab) 8649d1ec85SMing Lei return NULL; 8749d1ec85SMing Lei 8849d1ec85SMing Lei snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); 8949d1ec85SMing Lei bslab->slab = kmem_cache_create(bslab->name, size, 9049d1ec85SMing Lei ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL); 9149d1ec85SMing Lei if (!bslab->slab) 9249d1ec85SMing Lei goto fail_alloc_slab; 9349d1ec85SMing Lei 9449d1ec85SMing Lei bslab->slab_ref = 1; 9549d1ec85SMing Lei bslab->slab_size = size; 9649d1ec85SMing Lei 9749d1ec85SMing Lei if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) 9849d1ec85SMing Lei return bslab; 9949d1ec85SMing Lei 10049d1ec85SMing Lei kmem_cache_destroy(bslab->slab); 10149d1ec85SMing Lei 10249d1ec85SMing Lei fail_alloc_slab: 10349d1ec85SMing Lei kfree(bslab); 10449d1ec85SMing Lei return NULL; 10549d1ec85SMing Lei } 10649d1ec85SMing Lei 10749d1ec85SMing Lei static inline unsigned int bs_bio_slab_size(struct bio_set *bs) 10849d1ec85SMing Lei { 1099f180e31SMing Lei return bs->front_pad + sizeof(struct bio) + bs->back_pad; 11049d1ec85SMing Lei } 11149d1ec85SMing Lei 11249d1ec85SMing Lei static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) 11349d1ec85SMing Lei { 11449d1ec85SMing Lei unsigned int size = bs_bio_slab_size(bs); 11549d1ec85SMing Lei struct bio_slab *bslab; 116f9c78b2bSJens Axboe 117f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 11849d1ec85SMing Lei bslab = xa_load(&bio_slabs, size); 11949d1ec85SMing Lei if (bslab) 120f9c78b2bSJens Axboe bslab->slab_ref++; 12149d1ec85SMing Lei else 12249d1ec85SMing Lei bslab = create_bio_slab(size); 123f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 12449d1ec85SMing Lei 12549d1ec85SMing Lei if (bslab) 12649d1ec85SMing Lei return bslab->slab; 12749d1ec85SMing Lei return NULL; 128f9c78b2bSJens Axboe } 129f9c78b2bSJens Axboe 130f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 131f9c78b2bSJens Axboe { 132f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 13349d1ec85SMing Lei unsigned int slab_size = bs_bio_slab_size(bs); 134f9c78b2bSJens Axboe 135f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 136f9c78b2bSJens Axboe 13749d1ec85SMing Lei bslab = xa_load(&bio_slabs, slab_size); 138f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 139f9c78b2bSJens Axboe goto out; 140f9c78b2bSJens Axboe 14149d1ec85SMing Lei WARN_ON_ONCE(bslab->slab != bs->bio_slab); 14249d1ec85SMing Lei 143f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 144f9c78b2bSJens Axboe 145f9c78b2bSJens Axboe if (--bslab->slab_ref) 146f9c78b2bSJens Axboe goto out; 147f9c78b2bSJens Axboe 14849d1ec85SMing Lei xa_erase(&bio_slabs, slab_size); 14949d1ec85SMing Lei 150f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 15149d1ec85SMing Lei kfree(bslab); 152f9c78b2bSJens Axboe 153f9c78b2bSJens Axboe out: 154f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 155f9c78b2bSJens Axboe } 156f9c78b2bSJens Axboe 1577a800a20SChristoph Hellwig void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs) 158f9c78b2bSJens Axboe { 159a8affc03SChristoph Hellwig BIO_BUG_ON(nr_vecs > BIO_MAX_VECS); 160f9c78b2bSJens Axboe 161a8affc03SChristoph Hellwig if (nr_vecs == BIO_MAX_VECS) 162f9c78b2bSJens Axboe mempool_free(bv, pool); 1637a800a20SChristoph Hellwig else if (nr_vecs > BIO_INLINE_VECS) 1647a800a20SChristoph Hellwig kmem_cache_free(biovec_slab(nr_vecs)->slab, bv); 165f9c78b2bSJens Axboe } 166f9c78b2bSJens Axboe 167f2c3eb9bSChristoph Hellwig /* 168f2c3eb9bSChristoph Hellwig * Make the first allocation restricted and don't dump info on allocation 169f2c3eb9bSChristoph Hellwig * failures, since we'll fall back to the mempool in case of failure. 170f2c3eb9bSChristoph Hellwig */ 171f2c3eb9bSChristoph Hellwig static inline gfp_t bvec_alloc_gfp(gfp_t gfp) 172f9c78b2bSJens Axboe { 173f2c3eb9bSChristoph Hellwig return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | 174f2c3eb9bSChristoph Hellwig __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 175f2c3eb9bSChristoph Hellwig } 176f2c3eb9bSChristoph Hellwig 1777a800a20SChristoph Hellwig struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs, 1787a800a20SChristoph Hellwig gfp_t gfp_mask) 179f9c78b2bSJens Axboe { 1807a800a20SChristoph Hellwig struct biovec_slab *bvs = biovec_slab(*nr_vecs); 1817a800a20SChristoph Hellwig 1827a800a20SChristoph Hellwig if (WARN_ON_ONCE(!bvs)) 183f9c78b2bSJens Axboe return NULL; 1847a800a20SChristoph Hellwig 1857a800a20SChristoph Hellwig /* 1867a800a20SChristoph Hellwig * Upgrade the nr_vecs request to take full advantage of the allocation. 1877a800a20SChristoph Hellwig * We also rely on this in the bvec_free path. 1887a800a20SChristoph Hellwig */ 1897a800a20SChristoph Hellwig *nr_vecs = bvs->nr_vecs; 190f9c78b2bSJens Axboe 191f9c78b2bSJens Axboe /* 192f007a3d6SChristoph Hellwig * Try a slab allocation first for all smaller allocations. If that 193f007a3d6SChristoph Hellwig * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. 194a8affc03SChristoph Hellwig * The mempool is sized to handle up to BIO_MAX_VECS entries. 195f9c78b2bSJens Axboe */ 196a8affc03SChristoph Hellwig if (*nr_vecs < BIO_MAX_VECS) { 197f9c78b2bSJens Axboe struct bio_vec *bvl; 198f9c78b2bSJens Axboe 199f2c3eb9bSChristoph Hellwig bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); 2007a800a20SChristoph Hellwig if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) 201f9c78b2bSJens Axboe return bvl; 202a8affc03SChristoph Hellwig *nr_vecs = BIO_MAX_VECS; 203f9c78b2bSJens Axboe } 204f9c78b2bSJens Axboe 205f007a3d6SChristoph Hellwig return mempool_alloc(pool, gfp_mask); 206f9c78b2bSJens Axboe } 207f9c78b2bSJens Axboe 2089ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 209f9c78b2bSJens Axboe { 210db9819c7SChristoph Hellwig #ifdef CONFIG_BLK_CGROUP 211db9819c7SChristoph Hellwig if (bio->bi_blkg) { 212db9819c7SChristoph Hellwig blkg_put(bio->bi_blkg); 213db9819c7SChristoph Hellwig bio->bi_blkg = NULL; 214db9819c7SChristoph Hellwig } 215db9819c7SChristoph Hellwig #endif 216ece841abSJustin Tee if (bio_integrity(bio)) 217ece841abSJustin Tee bio_integrity_free(bio); 218a892c8d5SSatya Tangirala 219a892c8d5SSatya Tangirala bio_crypt_free_ctx(bio); 220f9c78b2bSJens Axboe } 2219ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 222f9c78b2bSJens Axboe 223f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 224f9c78b2bSJens Axboe { 225f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 226f9c78b2bSJens Axboe void *p; 227f9c78b2bSJens Axboe 2289ae3b3f5SJens Axboe bio_uninit(bio); 229f9c78b2bSJens Axboe 230f9c78b2bSJens Axboe if (bs) { 2317a800a20SChristoph Hellwig bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs); 232f9c78b2bSJens Axboe 233f9c78b2bSJens Axboe /* 234f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 235f9c78b2bSJens Axboe */ 236f9c78b2bSJens Axboe p = bio; 237f9c78b2bSJens Axboe p -= bs->front_pad; 238f9c78b2bSJens Axboe 2398aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 240f9c78b2bSJens Axboe } else { 241f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 242f9c78b2bSJens Axboe kfree(bio); 243f9c78b2bSJens Axboe } 244f9c78b2bSJens Axboe } 245f9c78b2bSJens Axboe 2469ae3b3f5SJens Axboe /* 2479ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2489ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2499ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2509ae3b3f5SJens Axboe */ 2513a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2523a83f467SMing Lei unsigned short max_vecs) 253f9c78b2bSJens Axboe { 254da521626SJens Axboe bio->bi_next = NULL; 255da521626SJens Axboe bio->bi_bdev = NULL; 256da521626SJens Axboe bio->bi_opf = 0; 257da521626SJens Axboe bio->bi_flags = 0; 258da521626SJens Axboe bio->bi_ioprio = 0; 259da521626SJens Axboe bio->bi_write_hint = 0; 260da521626SJens Axboe bio->bi_status = 0; 261da521626SJens Axboe bio->bi_iter.bi_sector = 0; 262da521626SJens Axboe bio->bi_iter.bi_size = 0; 263da521626SJens Axboe bio->bi_iter.bi_idx = 0; 264da521626SJens Axboe bio->bi_iter.bi_bvec_done = 0; 265da521626SJens Axboe bio->bi_end_io = NULL; 266da521626SJens Axboe bio->bi_private = NULL; 267da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP 268da521626SJens Axboe bio->bi_blkg = NULL; 269da521626SJens Axboe bio->bi_issue.value = 0; 270da521626SJens Axboe #ifdef CONFIG_BLK_CGROUP_IOCOST 271da521626SJens Axboe bio->bi_iocost_cost = 0; 272da521626SJens Axboe #endif 273da521626SJens Axboe #endif 274da521626SJens Axboe #ifdef CONFIG_BLK_INLINE_ENCRYPTION 275da521626SJens Axboe bio->bi_crypt_context = NULL; 276da521626SJens Axboe #endif 277da521626SJens Axboe #ifdef CONFIG_BLK_DEV_INTEGRITY 278da521626SJens Axboe bio->bi_integrity = NULL; 279da521626SJens Axboe #endif 280da521626SJens Axboe bio->bi_vcnt = 0; 281da521626SJens Axboe 282c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 283dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2843a83f467SMing Lei 2853a83f467SMing Lei bio->bi_max_vecs = max_vecs; 286da521626SJens Axboe bio->bi_io_vec = table; 287da521626SJens Axboe bio->bi_pool = NULL; 288f9c78b2bSJens Axboe } 289f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 290f9c78b2bSJens Axboe 291f9c78b2bSJens Axboe /** 292f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 293f9c78b2bSJens Axboe * @bio: bio to reset 294f9c78b2bSJens Axboe * 295f9c78b2bSJens Axboe * Description: 296f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 297f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 298f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 299f9c78b2bSJens Axboe * comment in struct bio. 300f9c78b2bSJens Axboe */ 301f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 302f9c78b2bSJens Axboe { 3039ae3b3f5SJens Axboe bio_uninit(bio); 304f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 305c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 306f9c78b2bSJens Axboe } 307f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 308f9c78b2bSJens Axboe 30938f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 310f9c78b2bSJens Axboe { 3114246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3124246a0b6SChristoph Hellwig 3133edf5346SYufen Yu if (bio->bi_status && !parent->bi_status) 3144e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 315f9c78b2bSJens Axboe bio_put(bio); 31638f8baaeSChristoph Hellwig return parent; 31738f8baaeSChristoph Hellwig } 31838f8baaeSChristoph Hellwig 31938f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 32038f8baaeSChristoph Hellwig { 32138f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 322f9c78b2bSJens Axboe } 323f9c78b2bSJens Axboe 324f9c78b2bSJens Axboe /** 325f9c78b2bSJens Axboe * bio_chain - chain bio completions 326f9c78b2bSJens Axboe * @bio: the target bio 3275b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 328f9c78b2bSJens Axboe * 329f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 330f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 331f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 332f9c78b2bSJens Axboe * 333f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 334f9c78b2bSJens Axboe */ 335f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 336f9c78b2bSJens Axboe { 337f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 338f9c78b2bSJens Axboe 339f9c78b2bSJens Axboe bio->bi_private = parent; 340f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 341c4cf5261SJens Axboe bio_inc_remaining(parent); 342f9c78b2bSJens Axboe } 343f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 344f9c78b2bSJens Axboe 345f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 346f9c78b2bSJens Axboe { 347f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 348f9c78b2bSJens Axboe struct bio *bio; 349f9c78b2bSJens Axboe 350f9c78b2bSJens Axboe while (1) { 351f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 352f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 353f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 354f9c78b2bSJens Axboe 355f9c78b2bSJens Axboe if (!bio) 356f9c78b2bSJens Axboe break; 357f9c78b2bSJens Axboe 358ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 359f9c78b2bSJens Axboe } 360f9c78b2bSJens Axboe } 361f9c78b2bSJens Axboe 362f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 363f9c78b2bSJens Axboe { 364f9c78b2bSJens Axboe struct bio_list punt, nopunt; 365f9c78b2bSJens Axboe struct bio *bio; 366f9c78b2bSJens Axboe 36747e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 36847e0fb46SNeilBrown return; 369f9c78b2bSJens Axboe /* 370f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 371f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 372f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 373f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 374f9c78b2bSJens Axboe * our own rescuer would be bad. 375f9c78b2bSJens Axboe * 376f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 377f9c78b2bSJens Axboe * remove from the middle of the list: 378f9c78b2bSJens Axboe */ 379f9c78b2bSJens Axboe 380f9c78b2bSJens Axboe bio_list_init(&punt); 381f9c78b2bSJens Axboe bio_list_init(&nopunt); 382f9c78b2bSJens Axboe 383f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 384f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 385f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 386f9c78b2bSJens Axboe 387f5fe1b51SNeilBrown bio_list_init(&nopunt); 388f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 389f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 390f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 391f9c78b2bSJens Axboe 392f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 393f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 394f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 395f9c78b2bSJens Axboe 396f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 397f9c78b2bSJens Axboe } 398f9c78b2bSJens Axboe 399f9c78b2bSJens Axboe /** 400f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 401519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 402f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 403f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 404f9c78b2bSJens Axboe * 4053175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 406f9c78b2bSJens Axboe * 4073175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4083175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4093175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4103175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4113175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4123175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 413f9c78b2bSJens Axboe * 4143175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4153175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 416ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 417f9c78b2bSJens Axboe * stack overflows. 418f9c78b2bSJens Axboe * 4193175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4203175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 421f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 422f9c78b2bSJens Axboe * thread. 423f9c78b2bSJens Axboe * 424f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 425f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 426ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 427f9c78b2bSJens Axboe * for per bio allocations. 428f9c78b2bSJens Axboe * 4293175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 430f9c78b2bSJens Axboe */ 4310f2e6ab8SChristoph Hellwig struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs, 4327a88fa19SDan Carpenter struct bio_set *bs) 433f9c78b2bSJens Axboe { 434f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 435f9c78b2bSJens Axboe struct bio *bio; 436f9c78b2bSJens Axboe void *p; 437f9c78b2bSJens Axboe 4383175199aSChristoph Hellwig /* should not use nobvec bioset for nr_iovecs > 0 */ 4393175199aSChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) 440f9c78b2bSJens Axboe return NULL; 441f9c78b2bSJens Axboe 442f9c78b2bSJens Axboe /* 4433175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 4443175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 4453175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 446f9c78b2bSJens Axboe * 4473175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 4483175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 4493175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 4503175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 4513175199aSChristoph Hellwig * mempool's reserve. 452f9c78b2bSJens Axboe * 453f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 4543175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 4553175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 4563175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 4573175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 4583175199aSChristoph Hellwig * gfp_flags. 459f9c78b2bSJens Axboe */ 460f5fe1b51SNeilBrown if (current->bio_list && 461f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 46247e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 46347e0fb46SNeilBrown bs->rescue_workqueue) 464d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 465f9c78b2bSJens Axboe 4668aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 467f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 468f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 469f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4708aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 471f9c78b2bSJens Axboe } 472f9c78b2bSJens Axboe if (unlikely(!p)) 473f9c78b2bSJens Axboe return NULL; 474f9c78b2bSJens Axboe 4753175199aSChristoph Hellwig bio = p + bs->front_pad; 4763175199aSChristoph Hellwig if (nr_iovecs > BIO_INLINE_VECS) { 4773175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 478f9c78b2bSJens Axboe 4797a800a20SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 480f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 481f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 482f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4837a800a20SChristoph Hellwig bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask); 484f9c78b2bSJens Axboe } 485f9c78b2bSJens Axboe if (unlikely(!bvl)) 486f9c78b2bSJens Axboe goto err_free; 487f9c78b2bSJens Axboe 4887a800a20SChristoph Hellwig bio_init(bio, bvl, nr_iovecs); 489f9c78b2bSJens Axboe } else if (nr_iovecs) { 4903175199aSChristoph Hellwig bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS); 4913175199aSChristoph Hellwig } else { 4923175199aSChristoph Hellwig bio_init(bio, NULL, 0); 493f9c78b2bSJens Axboe } 494f9c78b2bSJens Axboe 495f9c78b2bSJens Axboe bio->bi_pool = bs; 496f9c78b2bSJens Axboe return bio; 497f9c78b2bSJens Axboe 498f9c78b2bSJens Axboe err_free: 4998aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 500f9c78b2bSJens Axboe return NULL; 501f9c78b2bSJens Axboe } 502f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 503f9c78b2bSJens Axboe 5043175199aSChristoph Hellwig /** 5053175199aSChristoph Hellwig * bio_kmalloc - kmalloc a bio for I/O 5063175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 5073175199aSChristoph Hellwig * @nr_iovecs: number of iovecs to pre-allocate 5083175199aSChristoph Hellwig * 5093175199aSChristoph Hellwig * Use kmalloc to allocate and initialize a bio. 5103175199aSChristoph Hellwig * 5113175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 5123175199aSChristoph Hellwig */ 5130f2e6ab8SChristoph Hellwig struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) 5143175199aSChristoph Hellwig { 5153175199aSChristoph Hellwig struct bio *bio; 5163175199aSChristoph Hellwig 5173175199aSChristoph Hellwig if (nr_iovecs > UIO_MAXIOV) 5183175199aSChristoph Hellwig return NULL; 5193175199aSChristoph Hellwig 5203175199aSChristoph Hellwig bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 5213175199aSChristoph Hellwig if (unlikely(!bio)) 5223175199aSChristoph Hellwig return NULL; 5233175199aSChristoph Hellwig bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs); 5243175199aSChristoph Hellwig bio->bi_pool = NULL; 5253175199aSChristoph Hellwig return bio; 5263175199aSChristoph Hellwig } 5273175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 5283175199aSChristoph Hellwig 5296f822e1bSChristoph Hellwig void zero_fill_bio(struct bio *bio) 530f9c78b2bSJens Axboe { 531f9c78b2bSJens Axboe unsigned long flags; 532f9c78b2bSJens Axboe struct bio_vec bv; 533f9c78b2bSJens Axboe struct bvec_iter iter; 534f9c78b2bSJens Axboe 5356f822e1bSChristoph Hellwig bio_for_each_segment(bv, bio, iter) { 536f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 537f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 538f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 539f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 540f9c78b2bSJens Axboe } 541f9c78b2bSJens Axboe } 5426f822e1bSChristoph Hellwig EXPORT_SYMBOL(zero_fill_bio); 543f9c78b2bSJens Axboe 54483c9c547SMing Lei /** 54583c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 54683c9c547SMing Lei * @bio: the bio to be truncated 54783c9c547SMing Lei * @new_size: new size for truncating the bio 54883c9c547SMing Lei * 54983c9c547SMing Lei * Description: 55083c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 55183c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 55283c9c547SMing Lei * be used for handling corner cases, such as bio eod. 55383c9c547SMing Lei */ 55485a8ce62SMing Lei void bio_truncate(struct bio *bio, unsigned new_size) 55585a8ce62SMing Lei { 55685a8ce62SMing Lei struct bio_vec bv; 55785a8ce62SMing Lei struct bvec_iter iter; 55885a8ce62SMing Lei unsigned int done = 0; 55985a8ce62SMing Lei bool truncated = false; 56085a8ce62SMing Lei 56185a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 56285a8ce62SMing Lei return; 56385a8ce62SMing Lei 56483c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 56585a8ce62SMing Lei goto exit; 56685a8ce62SMing Lei 56785a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 56885a8ce62SMing Lei if (done + bv.bv_len > new_size) { 56985a8ce62SMing Lei unsigned offset; 57085a8ce62SMing Lei 57185a8ce62SMing Lei if (!truncated) 57285a8ce62SMing Lei offset = new_size - done; 57385a8ce62SMing Lei else 57485a8ce62SMing Lei offset = 0; 57585a8ce62SMing Lei zero_user(bv.bv_page, offset, bv.bv_len - offset); 57685a8ce62SMing Lei truncated = true; 57785a8ce62SMing Lei } 57885a8ce62SMing Lei done += bv.bv_len; 57985a8ce62SMing Lei } 58085a8ce62SMing Lei 58185a8ce62SMing Lei exit: 58285a8ce62SMing Lei /* 58385a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 58485a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 58585a8ce62SMing Lei * in its .end_bio() callback. 58685a8ce62SMing Lei * 58785a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 58885a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 58985a8ce62SMing Lei */ 59085a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 59185a8ce62SMing Lei } 59285a8ce62SMing Lei 593f9c78b2bSJens Axboe /** 59429125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 59529125ed6SChristoph Hellwig * @bio: bio to truncate 59629125ed6SChristoph Hellwig * 59729125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 59829125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 59929125ed6SChristoph Hellwig * 60029125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 60129125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 60229125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 60329125ed6SChristoph Hellwig * sector" case. 60429125ed6SChristoph Hellwig */ 60529125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 60629125ed6SChristoph Hellwig { 607309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 60829125ed6SChristoph Hellwig 60929125ed6SChristoph Hellwig if (!maxsector) 61029125ed6SChristoph Hellwig return; 61129125ed6SChristoph Hellwig 61229125ed6SChristoph Hellwig /* 61329125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 61429125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 61529125ed6SChristoph Hellwig * an EIO. 61629125ed6SChristoph Hellwig */ 61729125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 61829125ed6SChristoph Hellwig return; 61929125ed6SChristoph Hellwig 62029125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 62129125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 62229125ed6SChristoph Hellwig return; 62329125ed6SChristoph Hellwig 62429125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 62529125ed6SChristoph Hellwig } 62629125ed6SChristoph Hellwig 627*be4d234dSJens Axboe #define ALLOC_CACHE_MAX 512 628*be4d234dSJens Axboe #define ALLOC_CACHE_SLACK 64 629*be4d234dSJens Axboe 630*be4d234dSJens Axboe static void bio_alloc_cache_prune(struct bio_alloc_cache *cache, 631*be4d234dSJens Axboe unsigned int nr) 632*be4d234dSJens Axboe { 633*be4d234dSJens Axboe unsigned int i = 0; 634*be4d234dSJens Axboe struct bio *bio; 635*be4d234dSJens Axboe 636*be4d234dSJens Axboe while ((bio = bio_list_pop(&cache->free_list)) != NULL) { 637*be4d234dSJens Axboe cache->nr--; 638*be4d234dSJens Axboe bio_free(bio); 639*be4d234dSJens Axboe if (++i == nr) 640*be4d234dSJens Axboe break; 641*be4d234dSJens Axboe } 642*be4d234dSJens Axboe } 643*be4d234dSJens Axboe 644*be4d234dSJens Axboe static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node) 645*be4d234dSJens Axboe { 646*be4d234dSJens Axboe struct bio_set *bs; 647*be4d234dSJens Axboe 648*be4d234dSJens Axboe bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead); 649*be4d234dSJens Axboe if (bs->cache) { 650*be4d234dSJens Axboe struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu); 651*be4d234dSJens Axboe 652*be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 653*be4d234dSJens Axboe } 654*be4d234dSJens Axboe return 0; 655*be4d234dSJens Axboe } 656*be4d234dSJens Axboe 657*be4d234dSJens Axboe static void bio_alloc_cache_destroy(struct bio_set *bs) 658*be4d234dSJens Axboe { 659*be4d234dSJens Axboe int cpu; 660*be4d234dSJens Axboe 661*be4d234dSJens Axboe if (!bs->cache) 662*be4d234dSJens Axboe return; 663*be4d234dSJens Axboe 664*be4d234dSJens Axboe cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 665*be4d234dSJens Axboe for_each_possible_cpu(cpu) { 666*be4d234dSJens Axboe struct bio_alloc_cache *cache; 667*be4d234dSJens Axboe 668*be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, cpu); 669*be4d234dSJens Axboe bio_alloc_cache_prune(cache, -1U); 670*be4d234dSJens Axboe } 671*be4d234dSJens Axboe free_percpu(bs->cache); 672*be4d234dSJens Axboe } 673*be4d234dSJens Axboe 67429125ed6SChristoph Hellwig /** 675f9c78b2bSJens Axboe * bio_put - release a reference to a bio 676f9c78b2bSJens Axboe * @bio: bio to release reference to 677f9c78b2bSJens Axboe * 678f9c78b2bSJens Axboe * Description: 679f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 6809b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 681f9c78b2bSJens Axboe **/ 682f9c78b2bSJens Axboe void bio_put(struct bio *bio) 683f9c78b2bSJens Axboe { 684*be4d234dSJens Axboe if (unlikely(bio_flagged(bio, BIO_REFFED))) { 685dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 686*be4d234dSJens Axboe if (!atomic_dec_and_test(&bio->__bi_cnt)) 687*be4d234dSJens Axboe return; 688*be4d234dSJens Axboe } 689f9c78b2bSJens Axboe 690*be4d234dSJens Axboe if (bio_flagged(bio, BIO_PERCPU_CACHE)) { 691*be4d234dSJens Axboe struct bio_alloc_cache *cache; 692*be4d234dSJens Axboe 693*be4d234dSJens Axboe bio_uninit(bio); 694*be4d234dSJens Axboe cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu()); 695*be4d234dSJens Axboe bio_list_add_head(&cache->free_list, bio); 696*be4d234dSJens Axboe if (++cache->nr > ALLOC_CACHE_MAX + ALLOC_CACHE_SLACK) 697*be4d234dSJens Axboe bio_alloc_cache_prune(cache, ALLOC_CACHE_SLACK); 698*be4d234dSJens Axboe put_cpu(); 699*be4d234dSJens Axboe } else { 700f9c78b2bSJens Axboe bio_free(bio); 701f9c78b2bSJens Axboe } 702dac56212SJens Axboe } 703f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 704f9c78b2bSJens Axboe 705f9c78b2bSJens Axboe /** 706f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 707f9c78b2bSJens Axboe * @bio: destination bio 708f9c78b2bSJens Axboe * @bio_src: bio to clone 709f9c78b2bSJens Axboe * 710f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 711f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 712f9c78b2bSJens Axboe * bio will be one. 713f9c78b2bSJens Axboe * 714f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 715f9c78b2bSJens Axboe */ 716f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 717f9c78b2bSJens Axboe { 7187a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs); 719f9c78b2bSJens Axboe 720f9c78b2bSJens Axboe /* 721309dca30SChristoph Hellwig * most users will be overriding ->bi_bdev with a new target, 722f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 723f9c78b2bSJens Axboe */ 724309dca30SChristoph Hellwig bio->bi_bdev = bio_src->bi_bdev; 725b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 726111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 727111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 72846bbf653SChristoph Hellwig if (bio_flagged(bio_src, BIO_REMAPPED)) 72946bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 7301eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 731ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 732cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 733f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 734f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 73520bd723eSPaolo Valente 736db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 737e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 738f9c78b2bSJens Axboe } 739f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 740f9c78b2bSJens Axboe 741f9c78b2bSJens Axboe /** 742f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 743f9c78b2bSJens Axboe * @bio: bio to clone 744f9c78b2bSJens Axboe * @gfp_mask: allocation priority 745f9c78b2bSJens Axboe * @bs: bio_set to allocate from 746f9c78b2bSJens Axboe * 747f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 748f9c78b2bSJens Axboe */ 749f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 750f9c78b2bSJens Axboe { 751f9c78b2bSJens Axboe struct bio *b; 752f9c78b2bSJens Axboe 753f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 754f9c78b2bSJens Axboe if (!b) 755f9c78b2bSJens Axboe return NULL; 756f9c78b2bSJens Axboe 757f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 758f9c78b2bSJens Axboe 75907560151SEric Biggers if (bio_crypt_clone(b, bio, gfp_mask) < 0) 76007560151SEric Biggers goto err_put; 761a892c8d5SSatya Tangirala 76207560151SEric Biggers if (bio_integrity(bio) && 76307560151SEric Biggers bio_integrity_clone(b, bio, gfp_mask) < 0) 76407560151SEric Biggers goto err_put; 765f9c78b2bSJens Axboe 766f9c78b2bSJens Axboe return b; 76707560151SEric Biggers 76807560151SEric Biggers err_put: 76907560151SEric Biggers bio_put(b); 77007560151SEric Biggers return NULL; 771f9c78b2bSJens Axboe } 772f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 773f9c78b2bSJens Axboe 7745cbd28e3SChristoph Hellwig const char *bio_devname(struct bio *bio, char *buf) 7755cbd28e3SChristoph Hellwig { 776309dca30SChristoph Hellwig return bdevname(bio->bi_bdev, buf); 7775cbd28e3SChristoph Hellwig } 7785cbd28e3SChristoph Hellwig EXPORT_SYMBOL(bio_devname); 7795cbd28e3SChristoph Hellwig 7805919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 7815919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 782ff896738SChristoph Hellwig bool *same_page) 7835919482eSMing Lei { 784d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 785d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 7865919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 7875919482eSMing Lei 7885919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 7895919482eSMing Lei return false; 7905919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 7915919482eSMing Lei return false; 79252d52d1cSChristoph Hellwig 793ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 794d8166519SMatthew Wilcox (Oracle) if (*same_page) 7955919482eSMing Lei return true; 796d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 7975919482eSMing Lei } 7985919482eSMing Lei 799e4581105SChristoph Hellwig /* 800e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 801e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 802e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 803e4581105SChristoph Hellwig */ 804e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 805e4581105SChristoph Hellwig struct page *page, unsigned len, 806e4581105SChristoph Hellwig unsigned offset, bool *same_page) 807489fbbcbSMing Lei { 808384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 809489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 810489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 811489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 812489fbbcbSMing Lei 813489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 814489fbbcbSMing Lei return false; 815489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 816489fbbcbSMing Lei return false; 817384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 818489fbbcbSMing Lei } 819489fbbcbSMing Lei 820f4595875SShaohua Li /** 821e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 822c66a14d0SKent Overstreet * @q: the target queue 823c66a14d0SKent Overstreet * @bio: destination bio 824c66a14d0SKent Overstreet * @page: page to add 825c66a14d0SKent Overstreet * @len: vec entry length 826c66a14d0SKent Overstreet * @offset: vec entry offset 827e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 828e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 829f9c78b2bSJens Axboe * 830e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 831e4581105SChristoph Hellwig * and gap limitations. 832f9c78b2bSJens Axboe */ 833e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 83419047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 835e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 836f9c78b2bSJens Axboe { 837f9c78b2bSJens Axboe struct bio_vec *bvec; 838f9c78b2bSJens Axboe 839e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 840f9c78b2bSJens Axboe return 0; 841f9c78b2bSJens Axboe 842e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 843f9c78b2bSJens Axboe return 0; 844f9c78b2bSJens Axboe 845f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 846e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 847384209cdSChristoph Hellwig return len; 848320ea869SChristoph Hellwig 849320ea869SChristoph Hellwig /* 850320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 851320ea869SChristoph Hellwig * would create a gap, disallow it. 852320ea869SChristoph Hellwig */ 853384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 854320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 855320ea869SChristoph Hellwig return 0; 856f9c78b2bSJens Axboe } 857f9c78b2bSJens Axboe 85879d08f89SMing Lei if (bio_full(bio, len)) 859f9c78b2bSJens Axboe return 0; 860f9c78b2bSJens Axboe 86114ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 862489fbbcbSMing Lei return 0; 863489fbbcbSMing Lei 864f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 865f9c78b2bSJens Axboe bvec->bv_page = page; 866f9c78b2bSJens Axboe bvec->bv_len = len; 867f9c78b2bSJens Axboe bvec->bv_offset = offset; 868fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 869dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 870f9c78b2bSJens Axboe return len; 871f9c78b2bSJens Axboe } 87219047087SMing Lei 873e4581105SChristoph Hellwig /** 874e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 875e4581105SChristoph Hellwig * @q: the target queue 876e4581105SChristoph Hellwig * @bio: destination bio 877e4581105SChristoph Hellwig * @page: page to add 878e4581105SChristoph Hellwig * @len: vec entry length 879e4581105SChristoph Hellwig * @offset: vec entry offset 880e4581105SChristoph Hellwig * 881e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 882e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 883e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 884e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 885e4581105SChristoph Hellwig * 886e4581105SChristoph Hellwig * This should only be used by passthrough bios. 887e4581105SChristoph Hellwig */ 88819047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 88919047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 89019047087SMing Lei { 891d1916c86SChristoph Hellwig bool same_page = false; 892e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 893e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 89419047087SMing Lei } 895f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 896f9c78b2bSJens Axboe 897f9c78b2bSJens Axboe /** 898ae29333fSJohannes Thumshirn * bio_add_zone_append_page - attempt to add page to zone-append bio 899ae29333fSJohannes Thumshirn * @bio: destination bio 900ae29333fSJohannes Thumshirn * @page: page to add 901ae29333fSJohannes Thumshirn * @len: vec entry length 902ae29333fSJohannes Thumshirn * @offset: vec entry offset 903ae29333fSJohannes Thumshirn * 904ae29333fSJohannes Thumshirn * Attempt to add a page to the bio_vec maplist of a bio that will be submitted 905ae29333fSJohannes Thumshirn * for a zone-append request. This can fail for a number of reasons, such as the 906ae29333fSJohannes Thumshirn * bio being full or the target block device is not a zoned block device or 907ae29333fSJohannes Thumshirn * other limitations of the target block device. The target block device must 908ae29333fSJohannes Thumshirn * allow bio's up to PAGE_SIZE, so it is always possible to add a single page 909ae29333fSJohannes Thumshirn * to an empty bio. 910ae29333fSJohannes Thumshirn * 911ae29333fSJohannes Thumshirn * Returns: number of bytes added to the bio, or 0 in case of a failure. 912ae29333fSJohannes Thumshirn */ 913ae29333fSJohannes Thumshirn int bio_add_zone_append_page(struct bio *bio, struct page *page, 914ae29333fSJohannes Thumshirn unsigned int len, unsigned int offset) 915ae29333fSJohannes Thumshirn { 916582cd91fSLinus Torvalds struct request_queue *q = bio->bi_bdev->bd_disk->queue; 917ae29333fSJohannes Thumshirn bool same_page = false; 918ae29333fSJohannes Thumshirn 919ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND)) 920ae29333fSJohannes Thumshirn return 0; 921ae29333fSJohannes Thumshirn 922ae29333fSJohannes Thumshirn if (WARN_ON_ONCE(!blk_queue_is_zoned(q))) 923ae29333fSJohannes Thumshirn return 0; 924ae29333fSJohannes Thumshirn 925ae29333fSJohannes Thumshirn return bio_add_hw_page(q, bio, page, len, offset, 926ae29333fSJohannes Thumshirn queue_max_zone_append_sectors(q), &same_page); 927ae29333fSJohannes Thumshirn } 928ae29333fSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_add_zone_append_page); 929ae29333fSJohannes Thumshirn 930ae29333fSJohannes Thumshirn /** 9310aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 9320aa69fd3SChristoph Hellwig * @bio: destination bio 933551879a4SMing Lei * @page: start page to add 9340aa69fd3SChristoph Hellwig * @len: length of the data to add 935551879a4SMing Lei * @off: offset of the data relative to @page 936ff896738SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 9370aa69fd3SChristoph Hellwig * 9380aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 9393cf14889SRandy Dunlap * useful optimisation for file systems with a block size smaller than the 9400aa69fd3SChristoph Hellwig * page size. 9410aa69fd3SChristoph Hellwig * 942551879a4SMing Lei * Warn if (@len, @off) crosses pages in case that @same_page is true. 943551879a4SMing Lei * 9440aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 9450aa69fd3SChristoph Hellwig */ 9460aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 947ff896738SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 9480aa69fd3SChristoph Hellwig { 9490aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 9500aa69fd3SChristoph Hellwig return false; 9510aa69fd3SChristoph Hellwig 952cc90bc68SAndreas Gruenbacher if (bio->bi_vcnt > 0) { 9530aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 9540aa69fd3SChristoph Hellwig 9555919482eSMing Lei if (page_is_mergeable(bv, page, len, off, same_page)) { 95635c820e7SJens Axboe if (bio->bi_iter.bi_size > UINT_MAX - len) { 9572cd896a5SRitesh Harjani *same_page = false; 958cc90bc68SAndreas Gruenbacher return false; 9592cd896a5SRitesh Harjani } 9600aa69fd3SChristoph Hellwig bv->bv_len += len; 9610aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 9620aa69fd3SChristoph Hellwig return true; 9630aa69fd3SChristoph Hellwig } 9645919482eSMing Lei } 9650aa69fd3SChristoph Hellwig return false; 9660aa69fd3SChristoph Hellwig } 9670aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 9680aa69fd3SChristoph Hellwig 9690aa69fd3SChristoph Hellwig /** 970551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 9710aa69fd3SChristoph Hellwig * @bio: destination bio 972551879a4SMing Lei * @page: start page to add 973551879a4SMing Lei * @len: length of the data to add, may cross pages 974551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 9750aa69fd3SChristoph Hellwig * 9760aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 9770aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 9780aa69fd3SChristoph Hellwig */ 9790aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 9800aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 9810aa69fd3SChristoph Hellwig { 9820aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 9830aa69fd3SChristoph Hellwig 9840aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 98579d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 9860aa69fd3SChristoph Hellwig 9870aa69fd3SChristoph Hellwig bv->bv_page = page; 9880aa69fd3SChristoph Hellwig bv->bv_offset = off; 9890aa69fd3SChristoph Hellwig bv->bv_len = len; 9900aa69fd3SChristoph Hellwig 9910aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 9920aa69fd3SChristoph Hellwig bio->bi_vcnt++; 993b8e24a93SJohannes Weiner 994b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 995b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 9960aa69fd3SChristoph Hellwig } 9970aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 9980aa69fd3SChristoph Hellwig 9990aa69fd3SChristoph Hellwig /** 1000551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 1001f9c78b2bSJens Axboe * @bio: destination bio 1002551879a4SMing Lei * @page: start page to add 1003551879a4SMing Lei * @len: vec entry length, may cross pages 1004551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 1005f9c78b2bSJens Axboe * 1006551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 1007c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 1008f9c78b2bSJens Axboe */ 1009c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 1010c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 1011f9c78b2bSJens Axboe { 1012ff896738SChristoph Hellwig bool same_page = false; 1013ff896738SChristoph Hellwig 1014ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 101579d08f89SMing Lei if (bio_full(bio, len)) 1016c66a14d0SKent Overstreet return 0; 10170aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 1018c66a14d0SKent Overstreet } 1019c66a14d0SKent Overstreet return len; 1020f9c78b2bSJens Axboe } 1021f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 1022f9c78b2bSJens Axboe 1023d241a95fSChristoph Hellwig void bio_release_pages(struct bio *bio, bool mark_dirty) 10247321ecbfSChristoph Hellwig { 10257321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 10267321ecbfSChristoph Hellwig struct bio_vec *bvec; 10277321ecbfSChristoph Hellwig 1028b2d0d991SChristoph Hellwig if (bio_flagged(bio, BIO_NO_PAGE_REF)) 1029b2d0d991SChristoph Hellwig return; 1030b2d0d991SChristoph Hellwig 1031d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1032d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 1033d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 10347321ecbfSChristoph Hellwig put_page(bvec->bv_page); 10357321ecbfSChristoph Hellwig } 1036d241a95fSChristoph Hellwig } 103729b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_release_pages); 10387321ecbfSChristoph Hellwig 10397de55b7dSJohannes Thumshirn static void __bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 10406d0c48aeSJens Axboe { 10417a800a20SChristoph Hellwig WARN_ON_ONCE(bio->bi_max_vecs); 10426d0c48aeSJens Axboe 1043c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 1044c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 1045c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 1046c42bca92SPavel Begunkov bio->bi_iter.bi_size = iter->count; 1047ed97ce5eSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 1048977be012SChristoph Hellwig bio_set_flag(bio, BIO_CLONED); 10497de55b7dSJohannes Thumshirn } 10506d0c48aeSJens Axboe 10517de55b7dSJohannes Thumshirn static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 10527de55b7dSJohannes Thumshirn { 10537de55b7dSJohannes Thumshirn __bio_iov_bvec_set(bio, iter); 1054c42bca92SPavel Begunkov iov_iter_advance(iter, iter->count); 10556d0c48aeSJens Axboe return 0; 10566d0c48aeSJens Axboe } 10576d0c48aeSJens Axboe 10587de55b7dSJohannes Thumshirn static int bio_iov_bvec_set_append(struct bio *bio, struct iov_iter *iter) 10597de55b7dSJohannes Thumshirn { 10607de55b7dSJohannes Thumshirn struct request_queue *q = bio->bi_bdev->bd_disk->queue; 10617de55b7dSJohannes Thumshirn struct iov_iter i = *iter; 10627de55b7dSJohannes Thumshirn 10637de55b7dSJohannes Thumshirn iov_iter_truncate(&i, queue_max_zone_append_sectors(q) << 9); 10647de55b7dSJohannes Thumshirn __bio_iov_bvec_set(bio, &i); 10657de55b7dSJohannes Thumshirn iov_iter_advance(iter, i.count); 10667de55b7dSJohannes Thumshirn return 0; 10677de55b7dSJohannes Thumshirn } 10687de55b7dSJohannes Thumshirn 1069576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 1070576ed913SChristoph Hellwig 10712cefe4dbSKent Overstreet /** 107217d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 10732cefe4dbSKent Overstreet * @bio: bio to add pages to 10742cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 10752cefe4dbSKent Overstreet * 107617d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 10772cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 107817d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 10793cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 10802cefe4dbSKent Overstreet */ 108117d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 10822cefe4dbSKent Overstreet { 1083576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 1084576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 10852cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 10862cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 108745691804SChristoph Hellwig bool same_page = false; 1088576ed913SChristoph Hellwig ssize_t size, left; 1089576ed913SChristoph Hellwig unsigned len, i; 1090b403ea24SMartin Wilck size_t offset; 1091576ed913SChristoph Hellwig 1092576ed913SChristoph Hellwig /* 1093576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 1094576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 1095576ed913SChristoph Hellwig * without overwriting the temporary page array. 1096576ed913SChristoph Hellwig */ 1097576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 1098576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 10992cefe4dbSKent Overstreet 110035c820e7SJens Axboe size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 11012cefe4dbSKent Overstreet if (unlikely(size <= 0)) 11022cefe4dbSKent Overstreet return size ? size : -EFAULT; 11032cefe4dbSKent Overstreet 1104576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1105576ed913SChristoph Hellwig struct page *page = pages[i]; 11062cefe4dbSKent Overstreet 1107576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 110845691804SChristoph Hellwig 110945691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 111045691804SChristoph Hellwig if (same_page) 111145691804SChristoph Hellwig put_page(page); 111245691804SChristoph Hellwig } else { 111379d08f89SMing Lei if (WARN_ON_ONCE(bio_full(bio, len))) 1114576ed913SChristoph Hellwig return -EINVAL; 111545691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 111645691804SChristoph Hellwig } 1117576ed913SChristoph Hellwig offset = 0; 11182cefe4dbSKent Overstreet } 11192cefe4dbSKent Overstreet 11202cefe4dbSKent Overstreet iov_iter_advance(iter, size); 11212cefe4dbSKent Overstreet return 0; 11222cefe4dbSKent Overstreet } 112317d51b10SMartin Wilck 11240512a75bSKeith Busch static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 11250512a75bSKeith Busch { 11260512a75bSKeith Busch unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 11270512a75bSKeith Busch unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1128309dca30SChristoph Hellwig struct request_queue *q = bio->bi_bdev->bd_disk->queue; 11290512a75bSKeith Busch unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 11300512a75bSKeith Busch struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 11310512a75bSKeith Busch struct page **pages = (struct page **)bv; 11320512a75bSKeith Busch ssize_t size, left; 11330512a75bSKeith Busch unsigned len, i; 11340512a75bSKeith Busch size_t offset; 11354977d121SNaohiro Aota int ret = 0; 11360512a75bSKeith Busch 11370512a75bSKeith Busch if (WARN_ON_ONCE(!max_append_sectors)) 11380512a75bSKeith Busch return 0; 11390512a75bSKeith Busch 11400512a75bSKeith Busch /* 11410512a75bSKeith Busch * Move page array up in the allocated memory for the bio vecs as far as 11420512a75bSKeith Busch * possible so that we can start filling biovecs from the beginning 11430512a75bSKeith Busch * without overwriting the temporary page array. 11440512a75bSKeith Busch */ 11450512a75bSKeith Busch BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 11460512a75bSKeith Busch pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 11470512a75bSKeith Busch 11480512a75bSKeith Busch size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 11490512a75bSKeith Busch if (unlikely(size <= 0)) 11500512a75bSKeith Busch return size ? size : -EFAULT; 11510512a75bSKeith Busch 11520512a75bSKeith Busch for (left = size, i = 0; left > 0; left -= len, i++) { 11530512a75bSKeith Busch struct page *page = pages[i]; 11540512a75bSKeith Busch bool same_page = false; 11550512a75bSKeith Busch 11560512a75bSKeith Busch len = min_t(size_t, PAGE_SIZE - offset, left); 11570512a75bSKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 11584977d121SNaohiro Aota max_append_sectors, &same_page) != len) { 11594977d121SNaohiro Aota ret = -EINVAL; 11604977d121SNaohiro Aota break; 11614977d121SNaohiro Aota } 11620512a75bSKeith Busch if (same_page) 11630512a75bSKeith Busch put_page(page); 11640512a75bSKeith Busch offset = 0; 11650512a75bSKeith Busch } 11660512a75bSKeith Busch 11674977d121SNaohiro Aota iov_iter_advance(iter, size - left); 11684977d121SNaohiro Aota return ret; 11690512a75bSKeith Busch } 11700512a75bSKeith Busch 117117d51b10SMartin Wilck /** 11726d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 117317d51b10SMartin Wilck * @bio: bio to add pages to 11746d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 117517d51b10SMartin Wilck * 11766d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 11776d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 11786d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1179c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1180c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1181c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1182c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1183c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1184c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 11856d0c48aeSJens Axboe * 118617d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 11875cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 11886d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 11896d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 11900cf41e5eSPavel Begunkov * 11910cf41e5eSPavel Begunkov * It's intended for direct IO, so doesn't do PSI tracking, the caller is 11920cf41e5eSPavel Begunkov * responsible for setting BIO_WORKINGSET if necessary. 119317d51b10SMartin Wilck */ 119417d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 119517d51b10SMartin Wilck { 1196c42bca92SPavel Begunkov int ret = 0; 119714eacf12SChristoph Hellwig 1198c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 11997de55b7dSJohannes Thumshirn if (bio_op(bio) == REQ_OP_ZONE_APPEND) 12007de55b7dSJohannes Thumshirn return bio_iov_bvec_set_append(bio, iter); 1201ed97ce5eSChristoph Hellwig return bio_iov_bvec_set(bio, iter); 120286004515SChristoph Hellwig } 120317d51b10SMartin Wilck 120417d51b10SMartin Wilck do { 1205c42bca92SPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) 12060512a75bSKeith Busch ret = __bio_iov_append_get_pages(bio, iter); 12076d0c48aeSJens Axboe else 12086d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 120979d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 121017d51b10SMartin Wilck 12110cf41e5eSPavel Begunkov /* don't account direct I/O as memory stall */ 12120cf41e5eSPavel Begunkov bio_clear_flag(bio, BIO_WORKINGSET); 121314eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 121417d51b10SMartin Wilck } 121529b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 12162cefe4dbSKent Overstreet 12174246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1218f9c78b2bSJens Axboe { 121965e53aabSChristoph Hellwig complete(bio->bi_private); 1220f9c78b2bSJens Axboe } 1221f9c78b2bSJens Axboe 1222f9c78b2bSJens Axboe /** 1223f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1224f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1225f9c78b2bSJens Axboe * 1226f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1227f9c78b2bSJens Axboe * bio_endio() on failure. 12283d289d68SJan Kara * 12293d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 12303d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 12313d289d68SJan Kara * on his own. 1232f9c78b2bSJens Axboe */ 12334e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1234f9c78b2bSJens Axboe { 1235309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1236309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1237de6a78b6SMing Lei unsigned long hang_check; 1238f9c78b2bSJens Axboe 123965e53aabSChristoph Hellwig bio->bi_private = &done; 1240f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 12411eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 12424e49ea4aSMike Christie submit_bio(bio); 1243de6a78b6SMing Lei 1244de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1245de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1246de6a78b6SMing Lei if (hang_check) 1247de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1248de6a78b6SMing Lei hang_check * (HZ/2))) 1249de6a78b6SMing Lei ; 1250de6a78b6SMing Lei else 125165e53aabSChristoph Hellwig wait_for_completion_io(&done); 1252f9c78b2bSJens Axboe 125365e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1254f9c78b2bSJens Axboe } 1255f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1256f9c78b2bSJens Axboe 1257f9c78b2bSJens Axboe /** 1258f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 1259f9c78b2bSJens Axboe * @bio: bio to advance 1260f9c78b2bSJens Axboe * @bytes: number of bytes to complete 1261f9c78b2bSJens Axboe * 1262f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1263f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1264f9c78b2bSJens Axboe * be updated on the last bvec as well. 1265f9c78b2bSJens Axboe * 1266f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 1267f9c78b2bSJens Axboe */ 1268f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 1269f9c78b2bSJens Axboe { 1270f9c78b2bSJens Axboe if (bio_integrity(bio)) 1271f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1272f9c78b2bSJens Axboe 1273a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1274f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1275f9c78b2bSJens Axboe } 1276f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 1277f9c78b2bSJens Axboe 127845db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 127945db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1280f9c78b2bSJens Axboe { 1281f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1282f9c78b2bSJens Axboe void *src_p, *dst_p; 1283f9c78b2bSJens Axboe unsigned bytes; 1284f9c78b2bSJens Axboe 128545db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 128645db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 128745db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 128845db54d5SKent Overstreet 128945db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 129045db54d5SKent Overstreet 129145db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 129245db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 129345db54d5SKent Overstreet 129445db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 129545db54d5SKent Overstreet src_p + src_bv.bv_offset, 129645db54d5SKent Overstreet bytes); 129745db54d5SKent Overstreet 129845db54d5SKent Overstreet kunmap_atomic(dst_p); 129945db54d5SKent Overstreet kunmap_atomic(src_p); 130045db54d5SKent Overstreet 13016e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 13026e6e811dSKent Overstreet 130322b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 130422b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 130545db54d5SKent Overstreet } 130645db54d5SKent Overstreet } 130745db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 130845db54d5SKent Overstreet 130945db54d5SKent Overstreet /** 131045db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 131145db54d5SKent Overstreet * @src: source bio 131245db54d5SKent Overstreet * @dst: destination bio 131345db54d5SKent Overstreet * 131445db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 131545db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 131645db54d5SKent Overstreet */ 131745db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 131845db54d5SKent Overstreet { 131945db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 132045db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 132145db54d5SKent Overstreet 132245db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 132345db54d5SKent Overstreet } 132445db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 132545db54d5SKent Overstreet 1326491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 13271dfa0f68SChristoph Hellwig { 13281dfa0f68SChristoph Hellwig struct bio_vec *bvec; 13296dc4f100SMing Lei struct bvec_iter_all iter_all; 13301dfa0f68SChristoph Hellwig 13312b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 13321dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 13331dfa0f68SChristoph Hellwig } 1334491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 13351dfa0f68SChristoph Hellwig 1336f9c78b2bSJens Axboe /* 1337f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1338f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1339f9c78b2bSJens Axboe * 1340f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1341f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1342f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1343f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1344f9c78b2bSJens Axboe * in process context. 1345f9c78b2bSJens Axboe * 1346f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1347f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1348f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1349f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1350f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1351f9c78b2bSJens Axboe * 1352f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1353f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1354f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1355f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1356f9c78b2bSJens Axboe * pagecache. 1357f9c78b2bSJens Axboe * 1358f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1359f9c78b2bSJens Axboe * deferred bio dirtying paths. 1360f9c78b2bSJens Axboe */ 1361f9c78b2bSJens Axboe 1362f9c78b2bSJens Axboe /* 1363f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1364f9c78b2bSJens Axboe */ 1365f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1366f9c78b2bSJens Axboe { 1367f9c78b2bSJens Axboe struct bio_vec *bvec; 13686dc4f100SMing Lei struct bvec_iter_all iter_all; 1369f9c78b2bSJens Axboe 13702b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 13713bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 13723bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1373f9c78b2bSJens Axboe } 1374f9c78b2bSJens Axboe } 1375f9c78b2bSJens Axboe 1376f9c78b2bSJens Axboe /* 1377f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1378f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1379f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 138024d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1381f9c78b2bSJens Axboe * 1382f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1383ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1384ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1385f9c78b2bSJens Axboe */ 1386f9c78b2bSJens Axboe 1387f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1388f9c78b2bSJens Axboe 1389f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1390f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1391f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1392f9c78b2bSJens Axboe 1393f9c78b2bSJens Axboe /* 1394f9c78b2bSJens Axboe * This runs in process context 1395f9c78b2bSJens Axboe */ 1396f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1397f9c78b2bSJens Axboe { 139824d5493fSChristoph Hellwig struct bio *bio, *next; 1399f9c78b2bSJens Axboe 140024d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 140124d5493fSChristoph Hellwig next = bio_dirty_list; 1402f9c78b2bSJens Axboe bio_dirty_list = NULL; 140324d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1404f9c78b2bSJens Axboe 140524d5493fSChristoph Hellwig while ((bio = next) != NULL) { 140624d5493fSChristoph Hellwig next = bio->bi_private; 1407f9c78b2bSJens Axboe 1408d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1409f9c78b2bSJens Axboe bio_put(bio); 1410f9c78b2bSJens Axboe } 1411f9c78b2bSJens Axboe } 1412f9c78b2bSJens Axboe 1413f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1414f9c78b2bSJens Axboe { 1415f9c78b2bSJens Axboe struct bio_vec *bvec; 141624d5493fSChristoph Hellwig unsigned long flags; 14176dc4f100SMing Lei struct bvec_iter_all iter_all; 1418f9c78b2bSJens Axboe 14192b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 142024d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 142124d5493fSChristoph Hellwig goto defer; 1422f9c78b2bSJens Axboe } 1423f9c78b2bSJens Axboe 1424d241a95fSChristoph Hellwig bio_release_pages(bio, false); 142524d5493fSChristoph Hellwig bio_put(bio); 142624d5493fSChristoph Hellwig return; 142724d5493fSChristoph Hellwig defer: 1428f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1429f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1430f9c78b2bSJens Axboe bio_dirty_list = bio; 1431f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1432f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1433f9c78b2bSJens Axboe } 1434f9c78b2bSJens Axboe 1435c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1436c4cf5261SJens Axboe { 1437c4cf5261SJens Axboe /* 1438c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1439c4cf5261SJens Axboe * we always end io on the first invocation. 1440c4cf5261SJens Axboe */ 1441c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1442c4cf5261SJens Axboe return true; 1443c4cf5261SJens Axboe 1444c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1445c4cf5261SJens Axboe 1446326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1447b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1448c4cf5261SJens Axboe return true; 1449326e1dbbSMike Snitzer } 1450c4cf5261SJens Axboe 1451c4cf5261SJens Axboe return false; 1452c4cf5261SJens Axboe } 1453c4cf5261SJens Axboe 1454f9c78b2bSJens Axboe /** 1455f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1456f9c78b2bSJens Axboe * @bio: bio 1457f9c78b2bSJens Axboe * 1458f9c78b2bSJens Axboe * Description: 14594246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 14604246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 14614246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1462fbbaf700SNeilBrown * 1463fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1464fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 146560b6a7e6SEdward Hsieh * last time. 1466f9c78b2bSJens Axboe **/ 14674246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1468f9c78b2bSJens Axboe { 1469ba8c6967SChristoph Hellwig again: 14702b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1471ba8c6967SChristoph Hellwig return; 14727c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 14737c20f116SChristoph Hellwig return; 1474f9c78b2bSJens Axboe 1475309dca30SChristoph Hellwig if (bio->bi_bdev) 1476309dca30SChristoph Hellwig rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio); 147767b42d0bSJosef Bacik 147860b6a7e6SEdward Hsieh if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 147960b6a7e6SEdward Hsieh trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio); 148060b6a7e6SEdward Hsieh bio_clear_flag(bio, BIO_TRACE_COMPLETION); 148160b6a7e6SEdward Hsieh } 148260b6a7e6SEdward Hsieh 1483f9c78b2bSJens Axboe /* 1484ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1485ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1486ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1487ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1488ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1489ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1490f9c78b2bSJens Axboe */ 1491f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 149238f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1493ba8c6967SChristoph Hellwig goto again; 1494ba8c6967SChristoph Hellwig } 1495ba8c6967SChristoph Hellwig 14969e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1497b222dd2fSShaohua Li /* release cgroup info */ 1498b222dd2fSShaohua Li bio_uninit(bio); 1499f9c78b2bSJens Axboe if (bio->bi_end_io) 15004246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1501f9c78b2bSJens Axboe } 1502f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1503f9c78b2bSJens Axboe 1504f9c78b2bSJens Axboe /** 1505f9c78b2bSJens Axboe * bio_split - split a bio 1506f9c78b2bSJens Axboe * @bio: bio to split 1507f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1508f9c78b2bSJens Axboe * @gfp: gfp mask 1509f9c78b2bSJens Axboe * @bs: bio set to allocate from 1510f9c78b2bSJens Axboe * 1511f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1512f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1513f9c78b2bSJens Axboe * 1514f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1515dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1516dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1517f9c78b2bSJens Axboe */ 1518f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1519f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1520f9c78b2bSJens Axboe { 1521f341a4d3SMikulas Patocka struct bio *split; 1522f9c78b2bSJens Axboe 1523f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1524f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1525f9c78b2bSJens Axboe 15260512a75bSKeith Busch /* Zone append commands cannot be split */ 15270512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 15280512a75bSKeith Busch return NULL; 15290512a75bSKeith Busch 1530f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1531f9c78b2bSJens Axboe if (!split) 1532f9c78b2bSJens Axboe return NULL; 1533f9c78b2bSJens Axboe 1534f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1535f9c78b2bSJens Axboe 1536f9c78b2bSJens Axboe if (bio_integrity(split)) 1537fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1538f9c78b2bSJens Axboe 1539f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1540f9c78b2bSJens Axboe 1541fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 154220d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1543fbbaf700SNeilBrown 1544f9c78b2bSJens Axboe return split; 1545f9c78b2bSJens Axboe } 1546f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1547f9c78b2bSJens Axboe 1548f9c78b2bSJens Axboe /** 1549f9c78b2bSJens Axboe * bio_trim - trim a bio 1550f9c78b2bSJens Axboe * @bio: bio to trim 1551f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1552f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1553f9c78b2bSJens Axboe */ 1554f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1555f9c78b2bSJens Axboe { 1556f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1557f9c78b2bSJens Axboe * the given offset and size. 1558f9c78b2bSJens Axboe */ 1559f9c78b2bSJens Axboe 1560f9c78b2bSJens Axboe size <<= 9; 1561f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1562f9c78b2bSJens Axboe return; 1563f9c78b2bSJens Axboe 1564f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1565f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1566376a78abSDmitry Monakhov 1567376a78abSDmitry Monakhov if (bio_integrity(bio)) 1568fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1569376a78abSDmitry Monakhov 1570f9c78b2bSJens Axboe } 1571f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1572f9c78b2bSJens Axboe 1573f9c78b2bSJens Axboe /* 1574f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1575f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1576f9c78b2bSJens Axboe */ 15778aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1578f9c78b2bSJens Axboe { 15797a800a20SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1; 1580f9c78b2bSJens Axboe 15818aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1582f9c78b2bSJens Axboe } 1583f9c78b2bSJens Axboe 1584917a38c7SKent Overstreet /* 1585917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1586917a38c7SKent Overstreet * 1587917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1588917a38c7SKent Overstreet * kzalloc()). 1589917a38c7SKent Overstreet */ 1590917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1591f9c78b2bSJens Axboe { 1592*be4d234dSJens Axboe bio_alloc_cache_destroy(bs); 1593f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1594f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1595917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1596f9c78b2bSJens Axboe 15978aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 15988aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1599f9c78b2bSJens Axboe 1600f9c78b2bSJens Axboe bioset_integrity_free(bs); 1601917a38c7SKent Overstreet if (bs->bio_slab) 1602f9c78b2bSJens Axboe bio_put_slab(bs); 1603917a38c7SKent Overstreet bs->bio_slab = NULL; 1604917a38c7SKent Overstreet } 1605917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1606f9c78b2bSJens Axboe 1607011067b0SNeilBrown /** 1608917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1609dad08527SKent Overstreet * @bs: pool to initialize 1610917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1611917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1612917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1613917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1614917a38c7SKent Overstreet * 1615dad08527SKent Overstreet * Description: 1616dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1617dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1618dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1619dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1620dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1621dad08527SKent Overstreet * or things will break badly. 1622dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1623dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1624dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1625dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1626dad08527SKent Overstreet * 1627917a38c7SKent Overstreet */ 1628917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1629917a38c7SKent Overstreet unsigned int pool_size, 1630917a38c7SKent Overstreet unsigned int front_pad, 1631917a38c7SKent Overstreet int flags) 1632917a38c7SKent Overstreet { 1633917a38c7SKent Overstreet bs->front_pad = front_pad; 16349f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 16359f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 16369f180e31SMing Lei else 16379f180e31SMing Lei bs->back_pad = 0; 1638917a38c7SKent Overstreet 1639917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1640917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1641917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1642917a38c7SKent Overstreet 164349d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1644917a38c7SKent Overstreet if (!bs->bio_slab) 1645917a38c7SKent Overstreet return -ENOMEM; 1646917a38c7SKent Overstreet 1647917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1648917a38c7SKent Overstreet goto bad; 1649917a38c7SKent Overstreet 1650917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1651917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1652917a38c7SKent Overstreet goto bad; 1653917a38c7SKent Overstreet 1654*be4d234dSJens Axboe if (flags & BIOSET_NEED_RESCUER) { 1655*be4d234dSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", 1656*be4d234dSJens Axboe WQ_MEM_RECLAIM, 0); 1657917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1658917a38c7SKent Overstreet goto bad; 1659*be4d234dSJens Axboe } 1660*be4d234dSJens Axboe if (flags & BIOSET_PERCPU_CACHE) { 1661*be4d234dSJens Axboe bs->cache = alloc_percpu(struct bio_alloc_cache); 1662*be4d234dSJens Axboe if (!bs->cache) 1663*be4d234dSJens Axboe goto bad; 1664*be4d234dSJens Axboe cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead); 1665*be4d234dSJens Axboe } 1666917a38c7SKent Overstreet 1667917a38c7SKent Overstreet return 0; 1668917a38c7SKent Overstreet bad: 1669917a38c7SKent Overstreet bioset_exit(bs); 1670917a38c7SKent Overstreet return -ENOMEM; 1671917a38c7SKent Overstreet } 1672917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1673917a38c7SKent Overstreet 167428e89fd9SJens Axboe /* 167528e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 167628e89fd9SJens Axboe * another bio_set. 167728e89fd9SJens Axboe */ 167828e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 167928e89fd9SJens Axboe { 168028e89fd9SJens Axboe int flags; 168128e89fd9SJens Axboe 168228e89fd9SJens Axboe flags = 0; 168328e89fd9SJens Axboe if (src->bvec_pool.min_nr) 168428e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 168528e89fd9SJens Axboe if (src->rescue_workqueue) 168628e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 168728e89fd9SJens Axboe 168828e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 168928e89fd9SJens Axboe } 169028e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 169128e89fd9SJens Axboe 1692*be4d234dSJens Axboe /** 1693*be4d234dSJens Axboe * bio_alloc_kiocb - Allocate a bio from bio_set based on kiocb 1694*be4d234dSJens Axboe * @kiocb: kiocb describing the IO 1695*be4d234dSJens Axboe * @bs: bio_set to allocate from 1696*be4d234dSJens Axboe * 1697*be4d234dSJens Axboe * Description: 1698*be4d234dSJens Axboe * Like @bio_alloc_bioset, but pass in the kiocb. The kiocb is only 1699*be4d234dSJens Axboe * used to check if we should dip into the per-cpu bio_set allocation 1700*be4d234dSJens Axboe * cache. The allocation uses GFP_KERNEL internally. 1701*be4d234dSJens Axboe * 1702*be4d234dSJens Axboe */ 1703*be4d234dSJens Axboe struct bio *bio_alloc_kiocb(struct kiocb *kiocb, unsigned short nr_vecs, 1704*be4d234dSJens Axboe struct bio_set *bs) 1705*be4d234dSJens Axboe { 1706*be4d234dSJens Axboe struct bio_alloc_cache *cache; 1707*be4d234dSJens Axboe struct bio *bio; 1708*be4d234dSJens Axboe 1709*be4d234dSJens Axboe if (!(kiocb->ki_flags & IOCB_ALLOC_CACHE) || nr_vecs > BIO_INLINE_VECS) 1710*be4d234dSJens Axboe return bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1711*be4d234dSJens Axboe 1712*be4d234dSJens Axboe cache = per_cpu_ptr(bs->cache, get_cpu()); 1713*be4d234dSJens Axboe bio = bio_list_pop(&cache->free_list); 1714*be4d234dSJens Axboe if (bio) { 1715*be4d234dSJens Axboe cache->nr--; 1716*be4d234dSJens Axboe put_cpu(); 1717*be4d234dSJens Axboe bio_init(bio, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs); 1718*be4d234dSJens Axboe bio->bi_pool = bs; 1719*be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1720*be4d234dSJens Axboe return bio; 1721*be4d234dSJens Axboe } 1722*be4d234dSJens Axboe put_cpu(); 1723*be4d234dSJens Axboe bio = bio_alloc_bioset(GFP_KERNEL, nr_vecs, bs); 1724*be4d234dSJens Axboe bio_set_flag(bio, BIO_PERCPU_CACHE); 1725*be4d234dSJens Axboe return bio; 1726*be4d234dSJens Axboe } 1727*be4d234dSJens Axboe EXPORT_SYMBOL_GPL(bio_alloc_kiocb); 1728*be4d234dSJens Axboe 1729de76fd89SChristoph Hellwig static int __init init_bio(void) 1730f9c78b2bSJens Axboe { 1731f9c78b2bSJens Axboe int i; 1732f9c78b2bSJens Axboe 1733f9c78b2bSJens Axboe bio_integrity_init(); 1734de76fd89SChristoph Hellwig 1735de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1736f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1737f9c78b2bSJens Axboe 1738de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1739de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1740f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1741f9c78b2bSJens Axboe } 1742f9c78b2bSJens Axboe 1743*be4d234dSJens Axboe cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL, 1744*be4d234dSJens Axboe bio_cpu_dead); 1745*be4d234dSJens Axboe 1746f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1747f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1748f9c78b2bSJens Axboe 1749f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1750f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1751f9c78b2bSJens Axboe 1752f9c78b2bSJens Axboe return 0; 1753f9c78b2bSJens Axboe } 1754f9c78b2bSJens Axboe subsys_initcall(init_bio); 1755