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 28de76fd89SChristoph Hellwig static struct biovec_slab { 296ac0b715SChristoph Hellwig int nr_vecs; 306ac0b715SChristoph Hellwig char *name; 316ac0b715SChristoph Hellwig struct kmem_cache *slab; 32de76fd89SChristoph Hellwig } bvec_slabs[] __read_mostly = { 33de76fd89SChristoph Hellwig { .nr_vecs = 16, .name = "biovec-16" }, 34de76fd89SChristoph Hellwig { .nr_vecs = 64, .name = "biovec-64" }, 35de76fd89SChristoph Hellwig { .nr_vecs = 128, .name = "biovec-128" }, 36de76fd89SChristoph Hellwig { .nr_vecs = BIO_MAX_PAGES, .name = "biovec-max" }, 376ac0b715SChristoph Hellwig }; 386ac0b715SChristoph Hellwig 39f9c78b2bSJens Axboe /* 40f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 41f9c78b2bSJens Axboe * IO code that does not need private memory pools. 42f9c78b2bSJens Axboe */ 43f4f8154aSKent Overstreet struct bio_set fs_bio_set; 44f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 45f9c78b2bSJens Axboe 46f9c78b2bSJens Axboe /* 47f9c78b2bSJens Axboe * Our slab pool management 48f9c78b2bSJens Axboe */ 49f9c78b2bSJens Axboe struct bio_slab { 50f9c78b2bSJens Axboe struct kmem_cache *slab; 51f9c78b2bSJens Axboe unsigned int slab_ref; 52f9c78b2bSJens Axboe unsigned int slab_size; 53f9c78b2bSJens Axboe char name[8]; 54f9c78b2bSJens Axboe }; 55f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 5649d1ec85SMing Lei static DEFINE_XARRAY(bio_slabs); 57f9c78b2bSJens Axboe 5849d1ec85SMing Lei static struct bio_slab *create_bio_slab(unsigned int size) 59f9c78b2bSJens Axboe { 6049d1ec85SMing Lei struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); 6149d1ec85SMing Lei 6249d1ec85SMing Lei if (!bslab) 6349d1ec85SMing Lei return NULL; 6449d1ec85SMing Lei 6549d1ec85SMing Lei snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); 6649d1ec85SMing Lei bslab->slab = kmem_cache_create(bslab->name, size, 6749d1ec85SMing Lei ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL); 6849d1ec85SMing Lei if (!bslab->slab) 6949d1ec85SMing Lei goto fail_alloc_slab; 7049d1ec85SMing Lei 7149d1ec85SMing Lei bslab->slab_ref = 1; 7249d1ec85SMing Lei bslab->slab_size = size; 7349d1ec85SMing Lei 7449d1ec85SMing Lei if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) 7549d1ec85SMing Lei return bslab; 7649d1ec85SMing Lei 7749d1ec85SMing Lei kmem_cache_destroy(bslab->slab); 7849d1ec85SMing Lei 7949d1ec85SMing Lei fail_alloc_slab: 8049d1ec85SMing Lei kfree(bslab); 8149d1ec85SMing Lei return NULL; 8249d1ec85SMing Lei } 8349d1ec85SMing Lei 8449d1ec85SMing Lei static inline unsigned int bs_bio_slab_size(struct bio_set *bs) 8549d1ec85SMing Lei { 869f180e31SMing Lei return bs->front_pad + sizeof(struct bio) + bs->back_pad; 8749d1ec85SMing Lei } 8849d1ec85SMing Lei 8949d1ec85SMing Lei static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) 9049d1ec85SMing Lei { 9149d1ec85SMing Lei unsigned int size = bs_bio_slab_size(bs); 9249d1ec85SMing Lei struct bio_slab *bslab; 93f9c78b2bSJens Axboe 94f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 9549d1ec85SMing Lei bslab = xa_load(&bio_slabs, size); 9649d1ec85SMing Lei if (bslab) 97f9c78b2bSJens Axboe bslab->slab_ref++; 9849d1ec85SMing Lei else 9949d1ec85SMing Lei bslab = create_bio_slab(size); 100f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 10149d1ec85SMing Lei 10249d1ec85SMing Lei if (bslab) 10349d1ec85SMing Lei return bslab->slab; 10449d1ec85SMing Lei return NULL; 105f9c78b2bSJens Axboe } 106f9c78b2bSJens Axboe 107f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 108f9c78b2bSJens Axboe { 109f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 11049d1ec85SMing Lei unsigned int slab_size = bs_bio_slab_size(bs); 111f9c78b2bSJens Axboe 112f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 113f9c78b2bSJens Axboe 11449d1ec85SMing Lei bslab = xa_load(&bio_slabs, slab_size); 115f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 116f9c78b2bSJens Axboe goto out; 117f9c78b2bSJens Axboe 11849d1ec85SMing Lei WARN_ON_ONCE(bslab->slab != bs->bio_slab); 11949d1ec85SMing Lei 120f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 121f9c78b2bSJens Axboe 122f9c78b2bSJens Axboe if (--bslab->slab_ref) 123f9c78b2bSJens Axboe goto out; 124f9c78b2bSJens Axboe 12549d1ec85SMing Lei xa_erase(&bio_slabs, slab_size); 12649d1ec85SMing Lei 127f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 12849d1ec85SMing Lei kfree(bslab); 129f9c78b2bSJens Axboe 130f9c78b2bSJens Axboe out: 131f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 132f9c78b2bSJens Axboe } 133f9c78b2bSJens Axboe 134f9c78b2bSJens Axboe unsigned int bvec_nr_vecs(unsigned short idx) 135f9c78b2bSJens Axboe { 136d6c02a9bSGreg Edwards return bvec_slabs[--idx].nr_vecs; 137f9c78b2bSJens Axboe } 138f9c78b2bSJens Axboe 139f9c78b2bSJens Axboe void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 140f9c78b2bSJens Axboe { 141ed996a52SChristoph Hellwig if (!idx) 142ed996a52SChristoph Hellwig return; 143ed996a52SChristoph Hellwig idx--; 144f9c78b2bSJens Axboe 145ed996a52SChristoph Hellwig BIO_BUG_ON(idx >= BVEC_POOL_NR); 146ed996a52SChristoph Hellwig 147ed996a52SChristoph Hellwig if (idx == BVEC_POOL_MAX) { 148f9c78b2bSJens Axboe mempool_free(bv, pool); 149ed996a52SChristoph Hellwig } else { 150f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + idx; 151f9c78b2bSJens Axboe 152f9c78b2bSJens Axboe kmem_cache_free(bvs->slab, bv); 153f9c78b2bSJens Axboe } 154f9c78b2bSJens Axboe } 155f9c78b2bSJens Axboe 156f2c3eb9bSChristoph Hellwig /* 157f2c3eb9bSChristoph Hellwig * Make the first allocation restricted and don't dump info on allocation 158f2c3eb9bSChristoph Hellwig * failures, since we'll fall back to the mempool in case of failure. 159f2c3eb9bSChristoph Hellwig */ 160f2c3eb9bSChristoph Hellwig static inline gfp_t bvec_alloc_gfp(gfp_t gfp) 161f2c3eb9bSChristoph Hellwig { 162f2c3eb9bSChristoph Hellwig return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | 163f2c3eb9bSChristoph Hellwig __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 164f2c3eb9bSChristoph Hellwig } 165f2c3eb9bSChristoph Hellwig 166f9c78b2bSJens Axboe struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 167f9c78b2bSJens Axboe mempool_t *pool) 168f9c78b2bSJens Axboe { 169f9c78b2bSJens Axboe /* 170f9c78b2bSJens Axboe * see comment near bvec_array define! 171f9c78b2bSJens Axboe */ 172f9c78b2bSJens Axboe switch (nr) { 173de76fd89SChristoph Hellwig /* smaller bios use inline vecs */ 174f9c78b2bSJens Axboe case 5 ... 16: 175f9c78b2bSJens Axboe *idx = 2; 176f9c78b2bSJens Axboe break; 177f9c78b2bSJens Axboe case 17 ... 64: 178f9c78b2bSJens Axboe *idx = 3; 179f9c78b2bSJens Axboe break; 180f9c78b2bSJens Axboe case 65 ... 128: 181f9c78b2bSJens Axboe *idx = 4; 182f9c78b2bSJens Axboe break; 183f9c78b2bSJens Axboe case 129 ... BIO_MAX_PAGES: 184f9c78b2bSJens Axboe *idx = 5; 185f9c78b2bSJens Axboe break; 186f9c78b2bSJens Axboe default: 187f9c78b2bSJens Axboe return NULL; 188f9c78b2bSJens Axboe } 189f9c78b2bSJens Axboe 190f9c78b2bSJens Axboe /* 191f007a3d6SChristoph Hellwig * Try a slab allocation first for all smaller allocations. If that 192f007a3d6SChristoph Hellwig * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool. 193f007a3d6SChristoph Hellwig * The mempool is sized to handle up to BIO_MAX_PAGES entries. 194f9c78b2bSJens Axboe */ 195f007a3d6SChristoph Hellwig if (*idx < BVEC_POOL_MAX) { 196f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + *idx; 197f007a3d6SChristoph Hellwig struct bio_vec *bvl; 198f9c78b2bSJens Axboe 199f2c3eb9bSChristoph Hellwig bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); 200f007a3d6SChristoph Hellwig if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM)) { 201f007a3d6SChristoph Hellwig (*idx)++; 202f007a3d6SChristoph Hellwig return bvl; 203f9c78b2bSJens Axboe } 204f007a3d6SChristoph Hellwig *idx = BVEC_POOL_MAX; 205f9c78b2bSJens Axboe } 206f9c78b2bSJens Axboe 207ed996a52SChristoph Hellwig (*idx)++; 208f007a3d6SChristoph Hellwig return mempool_alloc(pool, gfp_mask); 209f9c78b2bSJens Axboe } 210f9c78b2bSJens Axboe 2119ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 212f9c78b2bSJens Axboe { 213db9819c7SChristoph Hellwig #ifdef CONFIG_BLK_CGROUP 214db9819c7SChristoph Hellwig if (bio->bi_blkg) { 215db9819c7SChristoph Hellwig blkg_put(bio->bi_blkg); 216db9819c7SChristoph Hellwig bio->bi_blkg = NULL; 217db9819c7SChristoph Hellwig } 218db9819c7SChristoph Hellwig #endif 219ece841abSJustin Tee if (bio_integrity(bio)) 220ece841abSJustin Tee bio_integrity_free(bio); 221a892c8d5SSatya Tangirala 222a892c8d5SSatya Tangirala bio_crypt_free_ctx(bio); 223f9c78b2bSJens Axboe } 2249ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 225f9c78b2bSJens Axboe 226f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 227f9c78b2bSJens Axboe { 228f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 229f9c78b2bSJens Axboe void *p; 230f9c78b2bSJens Axboe 2319ae3b3f5SJens Axboe bio_uninit(bio); 232f9c78b2bSJens Axboe 233f9c78b2bSJens Axboe if (bs) { 2348aa6ba2fSKent Overstreet bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 235f9c78b2bSJens Axboe 236f9c78b2bSJens Axboe /* 237f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 238f9c78b2bSJens Axboe */ 239f9c78b2bSJens Axboe p = bio; 240f9c78b2bSJens Axboe p -= bs->front_pad; 241f9c78b2bSJens Axboe 2428aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 243f9c78b2bSJens Axboe } else { 244f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 245f9c78b2bSJens Axboe kfree(bio); 246f9c78b2bSJens Axboe } 247f9c78b2bSJens Axboe } 248f9c78b2bSJens Axboe 2499ae3b3f5SJens Axboe /* 2509ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2519ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2529ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2539ae3b3f5SJens Axboe */ 2543a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2553a83f467SMing Lei unsigned short max_vecs) 256f9c78b2bSJens Axboe { 257f9c78b2bSJens Axboe memset(bio, 0, sizeof(*bio)); 258c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 259dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2603a83f467SMing Lei 2613a83f467SMing Lei bio->bi_io_vec = table; 2623a83f467SMing Lei bio->bi_max_vecs = max_vecs; 263f9c78b2bSJens Axboe } 264f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 265f9c78b2bSJens Axboe 266f9c78b2bSJens Axboe /** 267f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 268f9c78b2bSJens Axboe * @bio: bio to reset 269f9c78b2bSJens Axboe * 270f9c78b2bSJens Axboe * Description: 271f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 272f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 273f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 274f9c78b2bSJens Axboe * comment in struct bio. 275f9c78b2bSJens Axboe */ 276f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 277f9c78b2bSJens Axboe { 278f9c78b2bSJens Axboe unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 279f9c78b2bSJens Axboe 2809ae3b3f5SJens Axboe bio_uninit(bio); 281f9c78b2bSJens Axboe 282f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 2834246a0b6SChristoph Hellwig bio->bi_flags = flags; 284c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 285f9c78b2bSJens Axboe } 286f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 287f9c78b2bSJens Axboe 28838f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 289f9c78b2bSJens Axboe { 2904246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 2914246a0b6SChristoph Hellwig 2924e4cbee9SChristoph Hellwig if (!parent->bi_status) 2934e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 294f9c78b2bSJens Axboe bio_put(bio); 29538f8baaeSChristoph Hellwig return parent; 29638f8baaeSChristoph Hellwig } 29738f8baaeSChristoph Hellwig 29838f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 29938f8baaeSChristoph Hellwig { 30038f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 301f9c78b2bSJens Axboe } 302f9c78b2bSJens Axboe 303f9c78b2bSJens Axboe /** 304f9c78b2bSJens Axboe * bio_chain - chain bio completions 305f9c78b2bSJens Axboe * @bio: the target bio 3065b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 307f9c78b2bSJens Axboe * 308f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 309f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 310f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 311f9c78b2bSJens Axboe * 312f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 313f9c78b2bSJens Axboe */ 314f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 315f9c78b2bSJens Axboe { 316f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 317f9c78b2bSJens Axboe 318f9c78b2bSJens Axboe bio->bi_private = parent; 319f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 320c4cf5261SJens Axboe bio_inc_remaining(parent); 321f9c78b2bSJens Axboe } 322f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 323f9c78b2bSJens Axboe 324f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 325f9c78b2bSJens Axboe { 326f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 327f9c78b2bSJens Axboe struct bio *bio; 328f9c78b2bSJens Axboe 329f9c78b2bSJens Axboe while (1) { 330f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 331f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 332f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 333f9c78b2bSJens Axboe 334f9c78b2bSJens Axboe if (!bio) 335f9c78b2bSJens Axboe break; 336f9c78b2bSJens Axboe 337ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 338f9c78b2bSJens Axboe } 339f9c78b2bSJens Axboe } 340f9c78b2bSJens Axboe 341f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 342f9c78b2bSJens Axboe { 343f9c78b2bSJens Axboe struct bio_list punt, nopunt; 344f9c78b2bSJens Axboe struct bio *bio; 345f9c78b2bSJens Axboe 34647e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 34747e0fb46SNeilBrown return; 348f9c78b2bSJens Axboe /* 349f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 350f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 351f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 352f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 353f9c78b2bSJens Axboe * our own rescuer would be bad. 354f9c78b2bSJens Axboe * 355f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 356f9c78b2bSJens Axboe * remove from the middle of the list: 357f9c78b2bSJens Axboe */ 358f9c78b2bSJens Axboe 359f9c78b2bSJens Axboe bio_list_init(&punt); 360f9c78b2bSJens Axboe bio_list_init(&nopunt); 361f9c78b2bSJens Axboe 362f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 363f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 364f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 365f9c78b2bSJens Axboe 366f5fe1b51SNeilBrown bio_list_init(&nopunt); 367f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 368f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 369f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 370f9c78b2bSJens Axboe 371f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 372f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 373f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 374f9c78b2bSJens Axboe 375f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 376f9c78b2bSJens Axboe } 377f9c78b2bSJens Axboe 378f9c78b2bSJens Axboe /** 379f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 380519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 381f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 382f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 383f9c78b2bSJens Axboe * 3843175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 385f9c78b2bSJens Axboe * 3863175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 3873175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 3883175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 3893175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 3903175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 3913175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 392f9c78b2bSJens Axboe * 3933175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 3943175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 395ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 396f9c78b2bSJens Axboe * stack overflows. 397f9c78b2bSJens Axboe * 3983175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 3993175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 400f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 401f9c78b2bSJens Axboe * thread. 402f9c78b2bSJens Axboe * 403f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 404f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 405ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 406f9c78b2bSJens Axboe * for per bio allocations. 407f9c78b2bSJens Axboe * 4083175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 409f9c78b2bSJens Axboe */ 410*0f2e6ab8SChristoph Hellwig struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs, 4117a88fa19SDan Carpenter struct bio_set *bs) 412f9c78b2bSJens Axboe { 413f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 414f9c78b2bSJens Axboe struct bio *bio; 415f9c78b2bSJens Axboe void *p; 416f9c78b2bSJens Axboe 4173175199aSChristoph Hellwig /* should not use nobvec bioset for nr_iovecs > 0 */ 4183175199aSChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) 419f9c78b2bSJens Axboe return NULL; 420f9c78b2bSJens Axboe 421f9c78b2bSJens Axboe /* 4223175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 4233175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 4243175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 425f9c78b2bSJens Axboe * 4263175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 4273175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 4283175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 4293175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 4303175199aSChristoph Hellwig * mempool's reserve. 431f9c78b2bSJens Axboe * 432f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 4333175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 4343175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 4353175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 4363175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 4373175199aSChristoph Hellwig * gfp_flags. 438f9c78b2bSJens Axboe */ 439f5fe1b51SNeilBrown if (current->bio_list && 440f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 44147e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 44247e0fb46SNeilBrown bs->rescue_workqueue) 443d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 444f9c78b2bSJens Axboe 4458aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 446f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 447f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 448f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4498aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 450f9c78b2bSJens Axboe } 451f9c78b2bSJens Axboe if (unlikely(!p)) 452f9c78b2bSJens Axboe return NULL; 453f9c78b2bSJens Axboe 4543175199aSChristoph Hellwig bio = p + bs->front_pad; 4553175199aSChristoph Hellwig if (nr_iovecs > BIO_INLINE_VECS) { 456ed996a52SChristoph Hellwig unsigned long idx = 0; 4573175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 458ed996a52SChristoph Hellwig 4598aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 460f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 461f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 462f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4633175199aSChristoph Hellwig bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, 4643175199aSChristoph Hellwig &bs->bvec_pool); 465f9c78b2bSJens Axboe } 466f9c78b2bSJens Axboe 467f9c78b2bSJens Axboe if (unlikely(!bvl)) 468f9c78b2bSJens Axboe goto err_free; 469f9c78b2bSJens Axboe 4703175199aSChristoph Hellwig bio_init(bio, bvl, bvec_nr_vecs(idx)); 4718358c28aSMing Lei bio->bi_flags |= idx << BVEC_POOL_OFFSET; 472f9c78b2bSJens Axboe } else if (nr_iovecs) { 4733175199aSChristoph Hellwig bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS); 4743175199aSChristoph Hellwig } else { 4753175199aSChristoph Hellwig bio_init(bio, NULL, 0); 476f9c78b2bSJens Axboe } 477f9c78b2bSJens Axboe 478f9c78b2bSJens Axboe bio->bi_pool = bs; 479f9c78b2bSJens Axboe return bio; 480f9c78b2bSJens Axboe 481f9c78b2bSJens Axboe err_free: 4828aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 483f9c78b2bSJens Axboe return NULL; 484f9c78b2bSJens Axboe } 485f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 486f9c78b2bSJens Axboe 4873175199aSChristoph Hellwig /** 4883175199aSChristoph Hellwig * bio_kmalloc - kmalloc a bio for I/O 4893175199aSChristoph Hellwig * @gfp_mask: the GFP_* mask given to the slab allocator 4903175199aSChristoph Hellwig * @nr_iovecs: number of iovecs to pre-allocate 4913175199aSChristoph Hellwig * 4923175199aSChristoph Hellwig * Use kmalloc to allocate and initialize a bio. 4933175199aSChristoph Hellwig * 4943175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 4953175199aSChristoph Hellwig */ 496*0f2e6ab8SChristoph Hellwig struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs) 4973175199aSChristoph Hellwig { 4983175199aSChristoph Hellwig struct bio *bio; 4993175199aSChristoph Hellwig 5003175199aSChristoph Hellwig if (nr_iovecs > UIO_MAXIOV) 5013175199aSChristoph Hellwig return NULL; 5023175199aSChristoph Hellwig 5033175199aSChristoph Hellwig bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask); 5043175199aSChristoph Hellwig if (unlikely(!bio)) 5053175199aSChristoph Hellwig return NULL; 5063175199aSChristoph Hellwig bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs); 5073175199aSChristoph Hellwig bio->bi_pool = NULL; 5083175199aSChristoph Hellwig return bio; 5093175199aSChristoph Hellwig } 5103175199aSChristoph Hellwig EXPORT_SYMBOL(bio_kmalloc); 5113175199aSChristoph Hellwig 51238a72dacSKent Overstreet void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 513f9c78b2bSJens Axboe { 514f9c78b2bSJens Axboe unsigned long flags; 515f9c78b2bSJens Axboe struct bio_vec bv; 516f9c78b2bSJens Axboe struct bvec_iter iter; 517f9c78b2bSJens Axboe 51838a72dacSKent Overstreet __bio_for_each_segment(bv, bio, iter, start) { 519f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 520f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 521f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 522f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 523f9c78b2bSJens Axboe } 524f9c78b2bSJens Axboe } 52538a72dacSKent Overstreet EXPORT_SYMBOL(zero_fill_bio_iter); 526f9c78b2bSJens Axboe 52783c9c547SMing Lei /** 52883c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 52983c9c547SMing Lei * @bio: the bio to be truncated 53083c9c547SMing Lei * @new_size: new size for truncating the bio 53183c9c547SMing Lei * 53283c9c547SMing Lei * Description: 53383c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 53483c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 53583c9c547SMing Lei * be used for handling corner cases, such as bio eod. 53683c9c547SMing Lei */ 53785a8ce62SMing Lei void bio_truncate(struct bio *bio, unsigned new_size) 53885a8ce62SMing Lei { 53985a8ce62SMing Lei struct bio_vec bv; 54085a8ce62SMing Lei struct bvec_iter iter; 54185a8ce62SMing Lei unsigned int done = 0; 54285a8ce62SMing Lei bool truncated = false; 54385a8ce62SMing Lei 54485a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 54585a8ce62SMing Lei return; 54685a8ce62SMing Lei 54783c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 54885a8ce62SMing Lei goto exit; 54985a8ce62SMing Lei 55085a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 55185a8ce62SMing Lei if (done + bv.bv_len > new_size) { 55285a8ce62SMing Lei unsigned offset; 55385a8ce62SMing Lei 55485a8ce62SMing Lei if (!truncated) 55585a8ce62SMing Lei offset = new_size - done; 55685a8ce62SMing Lei else 55785a8ce62SMing Lei offset = 0; 55885a8ce62SMing Lei zero_user(bv.bv_page, offset, bv.bv_len - offset); 55985a8ce62SMing Lei truncated = true; 56085a8ce62SMing Lei } 56185a8ce62SMing Lei done += bv.bv_len; 56285a8ce62SMing Lei } 56385a8ce62SMing Lei 56485a8ce62SMing Lei exit: 56585a8ce62SMing Lei /* 56685a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 56785a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 56885a8ce62SMing Lei * in its .end_bio() callback. 56985a8ce62SMing Lei * 57085a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 57185a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 57285a8ce62SMing Lei */ 57385a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 57485a8ce62SMing Lei } 57585a8ce62SMing Lei 576f9c78b2bSJens Axboe /** 57729125ed6SChristoph Hellwig * guard_bio_eod - truncate a BIO to fit the block device 57829125ed6SChristoph Hellwig * @bio: bio to truncate 57929125ed6SChristoph Hellwig * 58029125ed6SChristoph Hellwig * This allows us to do IO even on the odd last sectors of a device, even if the 58129125ed6SChristoph Hellwig * block size is some multiple of the physical sector size. 58229125ed6SChristoph Hellwig * 58329125ed6SChristoph Hellwig * We'll just truncate the bio to the size of the device, and clear the end of 58429125ed6SChristoph Hellwig * the buffer head manually. Truly out-of-range accesses will turn into actual 58529125ed6SChristoph Hellwig * I/O errors, this only handles the "we need to be able to do I/O at the final 58629125ed6SChristoph Hellwig * sector" case. 58729125ed6SChristoph Hellwig */ 58829125ed6SChristoph Hellwig void guard_bio_eod(struct bio *bio) 58929125ed6SChristoph Hellwig { 590309dca30SChristoph Hellwig sector_t maxsector = bdev_nr_sectors(bio->bi_bdev); 59129125ed6SChristoph Hellwig 59229125ed6SChristoph Hellwig if (!maxsector) 59329125ed6SChristoph Hellwig return; 59429125ed6SChristoph Hellwig 59529125ed6SChristoph Hellwig /* 59629125ed6SChristoph Hellwig * If the *whole* IO is past the end of the device, 59729125ed6SChristoph Hellwig * let it through, and the IO layer will turn it into 59829125ed6SChristoph Hellwig * an EIO. 59929125ed6SChristoph Hellwig */ 60029125ed6SChristoph Hellwig if (unlikely(bio->bi_iter.bi_sector >= maxsector)) 60129125ed6SChristoph Hellwig return; 60229125ed6SChristoph Hellwig 60329125ed6SChristoph Hellwig maxsector -= bio->bi_iter.bi_sector; 60429125ed6SChristoph Hellwig if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) 60529125ed6SChristoph Hellwig return; 60629125ed6SChristoph Hellwig 60729125ed6SChristoph Hellwig bio_truncate(bio, maxsector << 9); 60829125ed6SChristoph Hellwig } 60929125ed6SChristoph Hellwig 61029125ed6SChristoph Hellwig /** 611f9c78b2bSJens Axboe * bio_put - release a reference to a bio 612f9c78b2bSJens Axboe * @bio: bio to release reference to 613f9c78b2bSJens Axboe * 614f9c78b2bSJens Axboe * Description: 615f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 6169b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 617f9c78b2bSJens Axboe **/ 618f9c78b2bSJens Axboe void bio_put(struct bio *bio) 619f9c78b2bSJens Axboe { 620dac56212SJens Axboe if (!bio_flagged(bio, BIO_REFFED)) 621dac56212SJens Axboe bio_free(bio); 622dac56212SJens Axboe else { 623dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 624f9c78b2bSJens Axboe 625f9c78b2bSJens Axboe /* 626f9c78b2bSJens Axboe * last put frees it 627f9c78b2bSJens Axboe */ 628dac56212SJens Axboe if (atomic_dec_and_test(&bio->__bi_cnt)) 629f9c78b2bSJens Axboe bio_free(bio); 630f9c78b2bSJens Axboe } 631dac56212SJens Axboe } 632f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 633f9c78b2bSJens Axboe 634f9c78b2bSJens Axboe /** 635f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 636f9c78b2bSJens Axboe * @bio: destination bio 637f9c78b2bSJens Axboe * @bio_src: bio to clone 638f9c78b2bSJens Axboe * 639f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 640f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 641f9c78b2bSJens Axboe * bio will be one. 642f9c78b2bSJens Axboe * 643f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 644f9c78b2bSJens Axboe */ 645f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 646f9c78b2bSJens Axboe { 647ed996a52SChristoph Hellwig BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 648f9c78b2bSJens Axboe 649f9c78b2bSJens Axboe /* 650309dca30SChristoph Hellwig * most users will be overriding ->bi_bdev with a new target, 651f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 652f9c78b2bSJens Axboe */ 653309dca30SChristoph Hellwig bio->bi_bdev = bio_src->bi_bdev; 654b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 655111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 656111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 65746bbf653SChristoph Hellwig if (bio_flagged(bio_src, BIO_REMAPPED)) 65846bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 6591eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 660ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 661cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 662f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 663f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 66420bd723eSPaolo Valente 665db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 666e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 667f9c78b2bSJens Axboe } 668f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 669f9c78b2bSJens Axboe 670f9c78b2bSJens Axboe /** 671f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 672f9c78b2bSJens Axboe * @bio: bio to clone 673f9c78b2bSJens Axboe * @gfp_mask: allocation priority 674f9c78b2bSJens Axboe * @bs: bio_set to allocate from 675f9c78b2bSJens Axboe * 676f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 677f9c78b2bSJens Axboe */ 678f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 679f9c78b2bSJens Axboe { 680f9c78b2bSJens Axboe struct bio *b; 681f9c78b2bSJens Axboe 682f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 683f9c78b2bSJens Axboe if (!b) 684f9c78b2bSJens Axboe return NULL; 685f9c78b2bSJens Axboe 686f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 687f9c78b2bSJens Axboe 68807560151SEric Biggers if (bio_crypt_clone(b, bio, gfp_mask) < 0) 68907560151SEric Biggers goto err_put; 690a892c8d5SSatya Tangirala 69107560151SEric Biggers if (bio_integrity(bio) && 69207560151SEric Biggers bio_integrity_clone(b, bio, gfp_mask) < 0) 69307560151SEric Biggers goto err_put; 694f9c78b2bSJens Axboe 695f9c78b2bSJens Axboe return b; 69607560151SEric Biggers 69707560151SEric Biggers err_put: 69807560151SEric Biggers bio_put(b); 69907560151SEric Biggers return NULL; 700f9c78b2bSJens Axboe } 701f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 702f9c78b2bSJens Axboe 7035cbd28e3SChristoph Hellwig const char *bio_devname(struct bio *bio, char *buf) 7045cbd28e3SChristoph Hellwig { 705309dca30SChristoph Hellwig return bdevname(bio->bi_bdev, buf); 7065cbd28e3SChristoph Hellwig } 7075cbd28e3SChristoph Hellwig EXPORT_SYMBOL(bio_devname); 7085cbd28e3SChristoph Hellwig 7095919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 7105919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 711ff896738SChristoph Hellwig bool *same_page) 7125919482eSMing Lei { 713d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 714d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 7155919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 7165919482eSMing Lei 7175919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 7185919482eSMing Lei return false; 7195919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 7205919482eSMing Lei return false; 72152d52d1cSChristoph Hellwig 722ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 723d8166519SMatthew Wilcox (Oracle) if (*same_page) 7245919482eSMing Lei return true; 725d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 7265919482eSMing Lei } 7275919482eSMing Lei 728e4581105SChristoph Hellwig /* 729e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 730e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 731e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 732e4581105SChristoph Hellwig */ 733e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 734e4581105SChristoph Hellwig struct page *page, unsigned len, 735e4581105SChristoph Hellwig unsigned offset, bool *same_page) 736489fbbcbSMing Lei { 737384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 738489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 739489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 740489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 741489fbbcbSMing Lei 742489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 743489fbbcbSMing Lei return false; 744489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 745489fbbcbSMing Lei return false; 746384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 747489fbbcbSMing Lei } 748489fbbcbSMing Lei 749f4595875SShaohua Li /** 750e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 751c66a14d0SKent Overstreet * @q: the target queue 752c66a14d0SKent Overstreet * @bio: destination bio 753c66a14d0SKent Overstreet * @page: page to add 754c66a14d0SKent Overstreet * @len: vec entry length 755c66a14d0SKent Overstreet * @offset: vec entry offset 756e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 757e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 758f9c78b2bSJens Axboe * 759e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 760e4581105SChristoph Hellwig * and gap limitations. 761f9c78b2bSJens Axboe */ 762e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 76319047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 764e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 765f9c78b2bSJens Axboe { 766f9c78b2bSJens Axboe struct bio_vec *bvec; 767f9c78b2bSJens Axboe 768e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 769f9c78b2bSJens Axboe return 0; 770f9c78b2bSJens Axboe 771e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 772f9c78b2bSJens Axboe return 0; 773f9c78b2bSJens Axboe 774f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 775e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 776384209cdSChristoph Hellwig return len; 777320ea869SChristoph Hellwig 778320ea869SChristoph Hellwig /* 779320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 780320ea869SChristoph Hellwig * would create a gap, disallow it. 781320ea869SChristoph Hellwig */ 782384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 783320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 784320ea869SChristoph Hellwig return 0; 785f9c78b2bSJens Axboe } 786f9c78b2bSJens Axboe 78779d08f89SMing Lei if (bio_full(bio, len)) 788f9c78b2bSJens Axboe return 0; 789f9c78b2bSJens Axboe 79014ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 791489fbbcbSMing Lei return 0; 792489fbbcbSMing Lei 793f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 794f9c78b2bSJens Axboe bvec->bv_page = page; 795f9c78b2bSJens Axboe bvec->bv_len = len; 796f9c78b2bSJens Axboe bvec->bv_offset = offset; 797fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 798dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 799f9c78b2bSJens Axboe return len; 800f9c78b2bSJens Axboe } 80119047087SMing Lei 802e4581105SChristoph Hellwig /** 803e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 804e4581105SChristoph Hellwig * @q: the target queue 805e4581105SChristoph Hellwig * @bio: destination bio 806e4581105SChristoph Hellwig * @page: page to add 807e4581105SChristoph Hellwig * @len: vec entry length 808e4581105SChristoph Hellwig * @offset: vec entry offset 809e4581105SChristoph Hellwig * 810e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 811e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 812e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 813e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 814e4581105SChristoph Hellwig * 815e4581105SChristoph Hellwig * This should only be used by passthrough bios. 816e4581105SChristoph Hellwig */ 81719047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 81819047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 81919047087SMing Lei { 820d1916c86SChristoph Hellwig bool same_page = false; 821e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 822e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 82319047087SMing Lei } 824f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 825f9c78b2bSJens Axboe 826f9c78b2bSJens Axboe /** 8270aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 8280aa69fd3SChristoph Hellwig * @bio: destination bio 829551879a4SMing Lei * @page: start page to add 8300aa69fd3SChristoph Hellwig * @len: length of the data to add 831551879a4SMing Lei * @off: offset of the data relative to @page 832ff896738SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 8330aa69fd3SChristoph Hellwig * 8340aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 8353cf14889SRandy Dunlap * useful optimisation for file systems with a block size smaller than the 8360aa69fd3SChristoph Hellwig * page size. 8370aa69fd3SChristoph Hellwig * 838551879a4SMing Lei * Warn if (@len, @off) crosses pages in case that @same_page is true. 839551879a4SMing Lei * 8400aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 8410aa69fd3SChristoph Hellwig */ 8420aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 843ff896738SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 8440aa69fd3SChristoph Hellwig { 8450aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8460aa69fd3SChristoph Hellwig return false; 8470aa69fd3SChristoph Hellwig 848cc90bc68SAndreas Gruenbacher if (bio->bi_vcnt > 0) { 8490aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 8500aa69fd3SChristoph Hellwig 8515919482eSMing Lei if (page_is_mergeable(bv, page, len, off, same_page)) { 8522cd896a5SRitesh Harjani if (bio->bi_iter.bi_size > UINT_MAX - len) { 8532cd896a5SRitesh Harjani *same_page = false; 854cc90bc68SAndreas Gruenbacher return false; 8552cd896a5SRitesh Harjani } 8560aa69fd3SChristoph Hellwig bv->bv_len += len; 8570aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8580aa69fd3SChristoph Hellwig return true; 8590aa69fd3SChristoph Hellwig } 8605919482eSMing Lei } 8610aa69fd3SChristoph Hellwig return false; 8620aa69fd3SChristoph Hellwig } 8630aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 8640aa69fd3SChristoph Hellwig 8650aa69fd3SChristoph Hellwig /** 866551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 8670aa69fd3SChristoph Hellwig * @bio: destination bio 868551879a4SMing Lei * @page: start page to add 869551879a4SMing Lei * @len: length of the data to add, may cross pages 870551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 8710aa69fd3SChristoph Hellwig * 8720aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 8730aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 8740aa69fd3SChristoph Hellwig */ 8750aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 8760aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 8770aa69fd3SChristoph Hellwig { 8780aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 8790aa69fd3SChristoph Hellwig 8800aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 88179d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 8820aa69fd3SChristoph Hellwig 8830aa69fd3SChristoph Hellwig bv->bv_page = page; 8840aa69fd3SChristoph Hellwig bv->bv_offset = off; 8850aa69fd3SChristoph Hellwig bv->bv_len = len; 8860aa69fd3SChristoph Hellwig 8870aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8880aa69fd3SChristoph Hellwig bio->bi_vcnt++; 889b8e24a93SJohannes Weiner 890b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 891b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 8920aa69fd3SChristoph Hellwig } 8930aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 8940aa69fd3SChristoph Hellwig 8950aa69fd3SChristoph Hellwig /** 896551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 897f9c78b2bSJens Axboe * @bio: destination bio 898551879a4SMing Lei * @page: start page to add 899551879a4SMing Lei * @len: vec entry length, may cross pages 900551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 901f9c78b2bSJens Axboe * 902551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 903c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 904f9c78b2bSJens Axboe */ 905c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 906c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 907f9c78b2bSJens Axboe { 908ff896738SChristoph Hellwig bool same_page = false; 909ff896738SChristoph Hellwig 910ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 91179d08f89SMing Lei if (bio_full(bio, len)) 912c66a14d0SKent Overstreet return 0; 9130aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 914c66a14d0SKent Overstreet } 915c66a14d0SKent Overstreet return len; 916f9c78b2bSJens Axboe } 917f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 918f9c78b2bSJens Axboe 919d241a95fSChristoph Hellwig void bio_release_pages(struct bio *bio, bool mark_dirty) 9207321ecbfSChristoph Hellwig { 9217321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 9227321ecbfSChristoph Hellwig struct bio_vec *bvec; 9237321ecbfSChristoph Hellwig 924b2d0d991SChristoph Hellwig if (bio_flagged(bio, BIO_NO_PAGE_REF)) 925b2d0d991SChristoph Hellwig return; 926b2d0d991SChristoph Hellwig 927d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 928d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 929d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 9307321ecbfSChristoph Hellwig put_page(bvec->bv_page); 9317321ecbfSChristoph Hellwig } 932d241a95fSChristoph Hellwig } 93329b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_release_pages); 9347321ecbfSChristoph Hellwig 935c42bca92SPavel Begunkov static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 9366d0c48aeSJens Axboe { 937c42bca92SPavel Begunkov WARN_ON_ONCE(BVEC_POOL_IDX(bio) != 0); 9386d0c48aeSJens Axboe 939c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 940c42bca92SPavel Begunkov bio->bi_max_vecs = iter->nr_segs; 941c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 942c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 943c42bca92SPavel Begunkov bio->bi_iter.bi_size = iter->count; 9446d0c48aeSJens Axboe 945c42bca92SPavel Begunkov iov_iter_advance(iter, iter->count); 9466d0c48aeSJens Axboe return 0; 9476d0c48aeSJens Axboe } 9486d0c48aeSJens Axboe 949576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 950576ed913SChristoph Hellwig 9512cefe4dbSKent Overstreet /** 95217d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 9532cefe4dbSKent Overstreet * @bio: bio to add pages to 9542cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 9552cefe4dbSKent Overstreet * 95617d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 9572cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 95817d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 9593cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 9602cefe4dbSKent Overstreet */ 96117d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 9622cefe4dbSKent Overstreet { 963576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 964576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 9652cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 9662cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 96745691804SChristoph Hellwig bool same_page = false; 968576ed913SChristoph Hellwig ssize_t size, left; 969576ed913SChristoph Hellwig unsigned len, i; 970b403ea24SMartin Wilck size_t offset; 971576ed913SChristoph Hellwig 972576ed913SChristoph Hellwig /* 973576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 974576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 975576ed913SChristoph Hellwig * without overwriting the temporary page array. 976576ed913SChristoph Hellwig */ 977576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 978576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 9792cefe4dbSKent Overstreet 9802cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 9812cefe4dbSKent Overstreet if (unlikely(size <= 0)) 9822cefe4dbSKent Overstreet return size ? size : -EFAULT; 9832cefe4dbSKent Overstreet 984576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 985576ed913SChristoph Hellwig struct page *page = pages[i]; 9862cefe4dbSKent Overstreet 987576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 98845691804SChristoph Hellwig 98945691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 99045691804SChristoph Hellwig if (same_page) 99145691804SChristoph Hellwig put_page(page); 99245691804SChristoph Hellwig } else { 99379d08f89SMing Lei if (WARN_ON_ONCE(bio_full(bio, len))) 994576ed913SChristoph Hellwig return -EINVAL; 99545691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 99645691804SChristoph Hellwig } 997576ed913SChristoph Hellwig offset = 0; 9982cefe4dbSKent Overstreet } 9992cefe4dbSKent Overstreet 10002cefe4dbSKent Overstreet iov_iter_advance(iter, size); 10012cefe4dbSKent Overstreet return 0; 10022cefe4dbSKent Overstreet } 100317d51b10SMartin Wilck 10040512a75bSKeith Busch static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 10050512a75bSKeith Busch { 10060512a75bSKeith Busch unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 10070512a75bSKeith Busch unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1008309dca30SChristoph Hellwig struct request_queue *q = bio->bi_bdev->bd_disk->queue; 10090512a75bSKeith Busch unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 10100512a75bSKeith Busch struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 10110512a75bSKeith Busch struct page **pages = (struct page **)bv; 10120512a75bSKeith Busch ssize_t size, left; 10130512a75bSKeith Busch unsigned len, i; 10140512a75bSKeith Busch size_t offset; 10154977d121SNaohiro Aota int ret = 0; 10160512a75bSKeith Busch 10170512a75bSKeith Busch if (WARN_ON_ONCE(!max_append_sectors)) 10180512a75bSKeith Busch return 0; 10190512a75bSKeith Busch 10200512a75bSKeith Busch /* 10210512a75bSKeith Busch * Move page array up in the allocated memory for the bio vecs as far as 10220512a75bSKeith Busch * possible so that we can start filling biovecs from the beginning 10230512a75bSKeith Busch * without overwriting the temporary page array. 10240512a75bSKeith Busch */ 10250512a75bSKeith Busch BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 10260512a75bSKeith Busch pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 10270512a75bSKeith Busch 10280512a75bSKeith Busch size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 10290512a75bSKeith Busch if (unlikely(size <= 0)) 10300512a75bSKeith Busch return size ? size : -EFAULT; 10310512a75bSKeith Busch 10320512a75bSKeith Busch for (left = size, i = 0; left > 0; left -= len, i++) { 10330512a75bSKeith Busch struct page *page = pages[i]; 10340512a75bSKeith Busch bool same_page = false; 10350512a75bSKeith Busch 10360512a75bSKeith Busch len = min_t(size_t, PAGE_SIZE - offset, left); 10370512a75bSKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 10384977d121SNaohiro Aota max_append_sectors, &same_page) != len) { 10394977d121SNaohiro Aota ret = -EINVAL; 10404977d121SNaohiro Aota break; 10414977d121SNaohiro Aota } 10420512a75bSKeith Busch if (same_page) 10430512a75bSKeith Busch put_page(page); 10440512a75bSKeith Busch offset = 0; 10450512a75bSKeith Busch } 10460512a75bSKeith Busch 10474977d121SNaohiro Aota iov_iter_advance(iter, size - left); 10484977d121SNaohiro Aota return ret; 10490512a75bSKeith Busch } 10500512a75bSKeith Busch 105117d51b10SMartin Wilck /** 10526d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 105317d51b10SMartin Wilck * @bio: bio to add pages to 10546d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 105517d51b10SMartin Wilck * 10566d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 10576d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 10586d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1059c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1060c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1061c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1062c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1063c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1064c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 10656d0c48aeSJens Axboe * 106617d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 10675cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 10686d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 10696d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 10700cf41e5eSPavel Begunkov * 10710cf41e5eSPavel Begunkov * It's intended for direct IO, so doesn't do PSI tracking, the caller is 10720cf41e5eSPavel Begunkov * responsible for setting BIO_WORKINGSET if necessary. 107317d51b10SMartin Wilck */ 107417d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 107517d51b10SMartin Wilck { 1076c42bca92SPavel Begunkov int ret = 0; 107714eacf12SChristoph Hellwig 1078c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1079c42bca92SPavel Begunkov if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 108014eacf12SChristoph Hellwig return -EINVAL; 1081c42bca92SPavel Begunkov bio_iov_bvec_set(bio, iter); 1082c42bca92SPavel Begunkov bio_set_flag(bio, BIO_NO_PAGE_REF); 1083c42bca92SPavel Begunkov return 0; 10840512a75bSKeith Busch } else { 1085c42bca92SPavel Begunkov do { 1086c42bca92SPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) 1087c42bca92SPavel Begunkov ret = __bio_iov_append_get_pages(bio, iter); 10886d0c48aeSJens Axboe else 10896d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 109079d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1091c42bca92SPavel Begunkov } 10920cf41e5eSPavel Begunkov 10930cf41e5eSPavel Begunkov /* don't account direct I/O as memory stall */ 10940cf41e5eSPavel Begunkov bio_clear_flag(bio, BIO_WORKINGSET); 109514eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 109617d51b10SMartin Wilck } 109729b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 10982cefe4dbSKent Overstreet 10994246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1100f9c78b2bSJens Axboe { 110165e53aabSChristoph Hellwig complete(bio->bi_private); 1102f9c78b2bSJens Axboe } 1103f9c78b2bSJens Axboe 1104f9c78b2bSJens Axboe /** 1105f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1106f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1107f9c78b2bSJens Axboe * 1108f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1109f9c78b2bSJens Axboe * bio_endio() on failure. 11103d289d68SJan Kara * 11113d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 11123d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 11133d289d68SJan Kara * on his own. 1114f9c78b2bSJens Axboe */ 11154e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1116f9c78b2bSJens Axboe { 1117309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1118309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1119de6a78b6SMing Lei unsigned long hang_check; 1120f9c78b2bSJens Axboe 112165e53aabSChristoph Hellwig bio->bi_private = &done; 1122f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 11231eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 11244e49ea4aSMike Christie submit_bio(bio); 1125de6a78b6SMing Lei 1126de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1127de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1128de6a78b6SMing Lei if (hang_check) 1129de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1130de6a78b6SMing Lei hang_check * (HZ/2))) 1131de6a78b6SMing Lei ; 1132de6a78b6SMing Lei else 113365e53aabSChristoph Hellwig wait_for_completion_io(&done); 1134f9c78b2bSJens Axboe 113565e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1136f9c78b2bSJens Axboe } 1137f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1138f9c78b2bSJens Axboe 1139f9c78b2bSJens Axboe /** 1140f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 1141f9c78b2bSJens Axboe * @bio: bio to advance 1142f9c78b2bSJens Axboe * @bytes: number of bytes to complete 1143f9c78b2bSJens Axboe * 1144f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1145f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1146f9c78b2bSJens Axboe * be updated on the last bvec as well. 1147f9c78b2bSJens Axboe * 1148f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 1149f9c78b2bSJens Axboe */ 1150f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 1151f9c78b2bSJens Axboe { 1152f9c78b2bSJens Axboe if (bio_integrity(bio)) 1153f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1154f9c78b2bSJens Axboe 1155a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1156f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1157f9c78b2bSJens Axboe } 1158f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 1159f9c78b2bSJens Axboe 116045db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 116145db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1162f9c78b2bSJens Axboe { 1163f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1164f9c78b2bSJens Axboe void *src_p, *dst_p; 1165f9c78b2bSJens Axboe unsigned bytes; 1166f9c78b2bSJens Axboe 116745db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 116845db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 116945db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 117045db54d5SKent Overstreet 117145db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 117245db54d5SKent Overstreet 117345db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 117445db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 117545db54d5SKent Overstreet 117645db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 117745db54d5SKent Overstreet src_p + src_bv.bv_offset, 117845db54d5SKent Overstreet bytes); 117945db54d5SKent Overstreet 118045db54d5SKent Overstreet kunmap_atomic(dst_p); 118145db54d5SKent Overstreet kunmap_atomic(src_p); 118245db54d5SKent Overstreet 11836e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 11846e6e811dSKent Overstreet 118522b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 118622b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 118745db54d5SKent Overstreet } 118845db54d5SKent Overstreet } 118945db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 119045db54d5SKent Overstreet 119145db54d5SKent Overstreet /** 119245db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 119345db54d5SKent Overstreet * @src: source bio 119445db54d5SKent Overstreet * @dst: destination bio 119545db54d5SKent Overstreet * 119645db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 119745db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 119845db54d5SKent Overstreet */ 119945db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 120045db54d5SKent Overstreet { 120145db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 120245db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 120345db54d5SKent Overstreet 120445db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 120545db54d5SKent Overstreet } 120645db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 120745db54d5SKent Overstreet 120845db54d5SKent Overstreet /** 120945db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 121045db54d5SKent Overstreet * another 121145db54d5SKent Overstreet * @src: source bio list 121245db54d5SKent Overstreet * @dst: destination bio list 121345db54d5SKent Overstreet * 121445db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 121545db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 121645db54d5SKent Overstreet * bios). 121745db54d5SKent Overstreet */ 121845db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 121945db54d5SKent Overstreet { 122045db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 122145db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 122245db54d5SKent Overstreet 1223f9c78b2bSJens Axboe while (1) { 1224f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1225f9c78b2bSJens Axboe src = src->bi_next; 1226f9c78b2bSJens Axboe if (!src) 1227f9c78b2bSJens Axboe break; 1228f9c78b2bSJens Axboe 1229f9c78b2bSJens Axboe src_iter = src->bi_iter; 1230f9c78b2bSJens Axboe } 1231f9c78b2bSJens Axboe 1232f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1233f9c78b2bSJens Axboe dst = dst->bi_next; 1234f9c78b2bSJens Axboe if (!dst) 1235f9c78b2bSJens Axboe break; 1236f9c78b2bSJens Axboe 1237f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1238f9c78b2bSJens Axboe } 1239f9c78b2bSJens Axboe 124045db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1241f9c78b2bSJens Axboe } 1242f9c78b2bSJens Axboe } 124345db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1244f9c78b2bSJens Axboe 1245491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 12461dfa0f68SChristoph Hellwig { 12471dfa0f68SChristoph Hellwig struct bio_vec *bvec; 12486dc4f100SMing Lei struct bvec_iter_all iter_all; 12491dfa0f68SChristoph Hellwig 12502b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 12511dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 12521dfa0f68SChristoph Hellwig } 1253491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 12541dfa0f68SChristoph Hellwig 1255f9c78b2bSJens Axboe /* 1256f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1257f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1258f9c78b2bSJens Axboe * 1259f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1260f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1261f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1262f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1263f9c78b2bSJens Axboe * in process context. 1264f9c78b2bSJens Axboe * 1265f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1266f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1267f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1268f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1269f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1270f9c78b2bSJens Axboe * 1271f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1272f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1273f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1274f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1275f9c78b2bSJens Axboe * pagecache. 1276f9c78b2bSJens Axboe * 1277f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1278f9c78b2bSJens Axboe * deferred bio dirtying paths. 1279f9c78b2bSJens Axboe */ 1280f9c78b2bSJens Axboe 1281f9c78b2bSJens Axboe /* 1282f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1283f9c78b2bSJens Axboe */ 1284f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1285f9c78b2bSJens Axboe { 1286f9c78b2bSJens Axboe struct bio_vec *bvec; 12876dc4f100SMing Lei struct bvec_iter_all iter_all; 1288f9c78b2bSJens Axboe 12892b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 12903bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 12913bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1292f9c78b2bSJens Axboe } 1293f9c78b2bSJens Axboe } 1294f9c78b2bSJens Axboe 1295f9c78b2bSJens Axboe /* 1296f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1297f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1298f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 129924d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1300f9c78b2bSJens Axboe * 1301f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1302ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1303ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1304f9c78b2bSJens Axboe */ 1305f9c78b2bSJens Axboe 1306f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1307f9c78b2bSJens Axboe 1308f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1309f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1310f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1311f9c78b2bSJens Axboe 1312f9c78b2bSJens Axboe /* 1313f9c78b2bSJens Axboe * This runs in process context 1314f9c78b2bSJens Axboe */ 1315f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1316f9c78b2bSJens Axboe { 131724d5493fSChristoph Hellwig struct bio *bio, *next; 1318f9c78b2bSJens Axboe 131924d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 132024d5493fSChristoph Hellwig next = bio_dirty_list; 1321f9c78b2bSJens Axboe bio_dirty_list = NULL; 132224d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1323f9c78b2bSJens Axboe 132424d5493fSChristoph Hellwig while ((bio = next) != NULL) { 132524d5493fSChristoph Hellwig next = bio->bi_private; 1326f9c78b2bSJens Axboe 1327d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1328f9c78b2bSJens Axboe bio_put(bio); 1329f9c78b2bSJens Axboe } 1330f9c78b2bSJens Axboe } 1331f9c78b2bSJens Axboe 1332f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1333f9c78b2bSJens Axboe { 1334f9c78b2bSJens Axboe struct bio_vec *bvec; 133524d5493fSChristoph Hellwig unsigned long flags; 13366dc4f100SMing Lei struct bvec_iter_all iter_all; 1337f9c78b2bSJens Axboe 13382b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 133924d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 134024d5493fSChristoph Hellwig goto defer; 1341f9c78b2bSJens Axboe } 1342f9c78b2bSJens Axboe 1343d241a95fSChristoph Hellwig bio_release_pages(bio, false); 134424d5493fSChristoph Hellwig bio_put(bio); 134524d5493fSChristoph Hellwig return; 134624d5493fSChristoph Hellwig defer: 1347f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1348f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1349f9c78b2bSJens Axboe bio_dirty_list = bio; 1350f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1351f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1352f9c78b2bSJens Axboe } 1353f9c78b2bSJens Axboe 1354c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1355c4cf5261SJens Axboe { 1356c4cf5261SJens Axboe /* 1357c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1358c4cf5261SJens Axboe * we always end io on the first invocation. 1359c4cf5261SJens Axboe */ 1360c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1361c4cf5261SJens Axboe return true; 1362c4cf5261SJens Axboe 1363c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1364c4cf5261SJens Axboe 1365326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1366b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1367c4cf5261SJens Axboe return true; 1368326e1dbbSMike Snitzer } 1369c4cf5261SJens Axboe 1370c4cf5261SJens Axboe return false; 1371c4cf5261SJens Axboe } 1372c4cf5261SJens Axboe 1373f9c78b2bSJens Axboe /** 1374f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1375f9c78b2bSJens Axboe * @bio: bio 1376f9c78b2bSJens Axboe * 1377f9c78b2bSJens Axboe * Description: 13784246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 13794246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 13804246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1381fbbaf700SNeilBrown * 1382fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1383fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1384fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1385fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1386f9c78b2bSJens Axboe **/ 13874246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1388f9c78b2bSJens Axboe { 1389ba8c6967SChristoph Hellwig again: 13902b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1391ba8c6967SChristoph Hellwig return; 13927c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 13937c20f116SChristoph Hellwig return; 1394f9c78b2bSJens Axboe 1395309dca30SChristoph Hellwig if (bio->bi_bdev) 1396309dca30SChristoph Hellwig rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio); 139767b42d0bSJosef Bacik 1398f9c78b2bSJens Axboe /* 1399ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1400ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1401ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1402ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1403ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1404ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1405f9c78b2bSJens Axboe */ 1406f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 140738f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1408ba8c6967SChristoph Hellwig goto again; 1409ba8c6967SChristoph Hellwig } 1410ba8c6967SChristoph Hellwig 1411309dca30SChristoph Hellwig if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1412309dca30SChristoph Hellwig trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio); 1413fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1414fbbaf700SNeilBrown } 1415fbbaf700SNeilBrown 14169e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1417b222dd2fSShaohua Li /* release cgroup info */ 1418b222dd2fSShaohua Li bio_uninit(bio); 1419f9c78b2bSJens Axboe if (bio->bi_end_io) 14204246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1421f9c78b2bSJens Axboe } 1422f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1423f9c78b2bSJens Axboe 1424f9c78b2bSJens Axboe /** 1425f9c78b2bSJens Axboe * bio_split - split a bio 1426f9c78b2bSJens Axboe * @bio: bio to split 1427f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1428f9c78b2bSJens Axboe * @gfp: gfp mask 1429f9c78b2bSJens Axboe * @bs: bio set to allocate from 1430f9c78b2bSJens Axboe * 1431f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1432f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1433f9c78b2bSJens Axboe * 1434f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1435dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1436dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1437f9c78b2bSJens Axboe */ 1438f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1439f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1440f9c78b2bSJens Axboe { 1441f341a4d3SMikulas Patocka struct bio *split; 1442f9c78b2bSJens Axboe 1443f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1444f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1445f9c78b2bSJens Axboe 14460512a75bSKeith Busch /* Zone append commands cannot be split */ 14470512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 14480512a75bSKeith Busch return NULL; 14490512a75bSKeith Busch 1450f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1451f9c78b2bSJens Axboe if (!split) 1452f9c78b2bSJens Axboe return NULL; 1453f9c78b2bSJens Axboe 1454f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1455f9c78b2bSJens Axboe 1456f9c78b2bSJens Axboe if (bio_integrity(split)) 1457fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1458f9c78b2bSJens Axboe 1459f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1460f9c78b2bSJens Axboe 1461fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 146220d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1463fbbaf700SNeilBrown 1464f9c78b2bSJens Axboe return split; 1465f9c78b2bSJens Axboe } 1466f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1467f9c78b2bSJens Axboe 1468f9c78b2bSJens Axboe /** 1469f9c78b2bSJens Axboe * bio_trim - trim a bio 1470f9c78b2bSJens Axboe * @bio: bio to trim 1471f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1472f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1473f9c78b2bSJens Axboe */ 1474f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1475f9c78b2bSJens Axboe { 1476f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1477f9c78b2bSJens Axboe * the given offset and size. 1478f9c78b2bSJens Axboe */ 1479f9c78b2bSJens Axboe 1480f9c78b2bSJens Axboe size <<= 9; 1481f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1482f9c78b2bSJens Axboe return; 1483f9c78b2bSJens Axboe 1484f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1485f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1486376a78abSDmitry Monakhov 1487376a78abSDmitry Monakhov if (bio_integrity(bio)) 1488fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1489376a78abSDmitry Monakhov 1490f9c78b2bSJens Axboe } 1491f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1492f9c78b2bSJens Axboe 1493f9c78b2bSJens Axboe /* 1494f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1495f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1496f9c78b2bSJens Axboe */ 14978aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1498f9c78b2bSJens Axboe { 1499ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1500f9c78b2bSJens Axboe 15018aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1502f9c78b2bSJens Axboe } 1503f9c78b2bSJens Axboe 1504917a38c7SKent Overstreet /* 1505917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1506917a38c7SKent Overstreet * 1507917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1508917a38c7SKent Overstreet * kzalloc()). 1509917a38c7SKent Overstreet */ 1510917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1511f9c78b2bSJens Axboe { 1512f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1513f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1514917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1515f9c78b2bSJens Axboe 15168aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 15178aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1518f9c78b2bSJens Axboe 1519f9c78b2bSJens Axboe bioset_integrity_free(bs); 1520917a38c7SKent Overstreet if (bs->bio_slab) 1521f9c78b2bSJens Axboe bio_put_slab(bs); 1522917a38c7SKent Overstreet bs->bio_slab = NULL; 1523917a38c7SKent Overstreet } 1524917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1525f9c78b2bSJens Axboe 1526011067b0SNeilBrown /** 1527917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1528dad08527SKent Overstreet * @bs: pool to initialize 1529917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1530917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1531917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1532917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1533917a38c7SKent Overstreet * 1534dad08527SKent Overstreet * Description: 1535dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1536dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1537dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1538dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1539dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1540dad08527SKent Overstreet * or things will break badly. 1541dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1542dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1543dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1544dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1545dad08527SKent Overstreet * 1546917a38c7SKent Overstreet */ 1547917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1548917a38c7SKent Overstreet unsigned int pool_size, 1549917a38c7SKent Overstreet unsigned int front_pad, 1550917a38c7SKent Overstreet int flags) 1551917a38c7SKent Overstreet { 1552917a38c7SKent Overstreet bs->front_pad = front_pad; 15539f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 15549f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 15559f180e31SMing Lei else 15569f180e31SMing Lei bs->back_pad = 0; 1557917a38c7SKent Overstreet 1558917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1559917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1560917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1561917a38c7SKent Overstreet 156249d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1563917a38c7SKent Overstreet if (!bs->bio_slab) 1564917a38c7SKent Overstreet return -ENOMEM; 1565917a38c7SKent Overstreet 1566917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1567917a38c7SKent Overstreet goto bad; 1568917a38c7SKent Overstreet 1569917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1570917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1571917a38c7SKent Overstreet goto bad; 1572917a38c7SKent Overstreet 1573917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 1574917a38c7SKent Overstreet return 0; 1575917a38c7SKent Overstreet 1576917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1577917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1578917a38c7SKent Overstreet goto bad; 1579917a38c7SKent Overstreet 1580917a38c7SKent Overstreet return 0; 1581917a38c7SKent Overstreet bad: 1582917a38c7SKent Overstreet bioset_exit(bs); 1583917a38c7SKent Overstreet return -ENOMEM; 1584917a38c7SKent Overstreet } 1585917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1586917a38c7SKent Overstreet 158728e89fd9SJens Axboe /* 158828e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 158928e89fd9SJens Axboe * another bio_set. 159028e89fd9SJens Axboe */ 159128e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 159228e89fd9SJens Axboe { 159328e89fd9SJens Axboe int flags; 159428e89fd9SJens Axboe 159528e89fd9SJens Axboe flags = 0; 159628e89fd9SJens Axboe if (src->bvec_pool.min_nr) 159728e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 159828e89fd9SJens Axboe if (src->rescue_workqueue) 159928e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 160028e89fd9SJens Axboe 160128e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 160228e89fd9SJens Axboe } 160328e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 160428e89fd9SJens Axboe 1605de76fd89SChristoph Hellwig static int __init init_bio(void) 1606f9c78b2bSJens Axboe { 1607f9c78b2bSJens Axboe int i; 1608f9c78b2bSJens Axboe 16092b24e6f6SJohannes Thumshirn BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 16102b24e6f6SJohannes Thumshirn 1611f9c78b2bSJens Axboe bio_integrity_init(); 1612de76fd89SChristoph Hellwig 1613de76fd89SChristoph Hellwig for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) { 1614de76fd89SChristoph Hellwig struct biovec_slab *bvs = bvec_slabs + i; 1615de76fd89SChristoph Hellwig 1616de76fd89SChristoph Hellwig bvs->slab = kmem_cache_create(bvs->name, 1617de76fd89SChristoph Hellwig bvs->nr_vecs * sizeof(struct bio_vec), 0, 1618de76fd89SChristoph Hellwig SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 1619de76fd89SChristoph Hellwig } 1620f9c78b2bSJens Axboe 1621f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1622f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1623f9c78b2bSJens Axboe 1624f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1625f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1626f9c78b2bSJens Axboe 1627f9c78b2bSJens Axboe return 0; 1628f9c78b2bSJens Axboe } 1629f9c78b2bSJens Axboe subsys_initcall(init_bio); 1630