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 286ac0b715SChristoph Hellwig struct biovec_slab { 296ac0b715SChristoph Hellwig int nr_vecs; 306ac0b715SChristoph Hellwig char *name; 316ac0b715SChristoph Hellwig struct kmem_cache *slab; 326ac0b715SChristoph Hellwig }; 336ac0b715SChristoph Hellwig 34f9c78b2bSJens Axboe /* 35f9c78b2bSJens Axboe * if you change this list, also change bvec_alloc or things will 36f9c78b2bSJens Axboe * break badly! cannot be bigger than what you can fit into an 37f9c78b2bSJens Axboe * unsigned short 38f9c78b2bSJens Axboe */ 39bd5c4facSMikulas Patocka #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 40ed996a52SChristoph Hellwig static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 41bd5c4facSMikulas Patocka BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 42f9c78b2bSJens Axboe }; 43f9c78b2bSJens Axboe #undef BV 44f9c78b2bSJens Axboe 45f9c78b2bSJens Axboe /* 46f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 47f9c78b2bSJens Axboe * IO code that does not need private memory pools. 48f9c78b2bSJens Axboe */ 49f4f8154aSKent Overstreet struct bio_set fs_bio_set; 50f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 51f9c78b2bSJens Axboe 52f9c78b2bSJens Axboe /* 53f9c78b2bSJens Axboe * Our slab pool management 54f9c78b2bSJens Axboe */ 55f9c78b2bSJens Axboe struct bio_slab { 56f9c78b2bSJens Axboe struct kmem_cache *slab; 57f9c78b2bSJens Axboe unsigned int slab_ref; 58f9c78b2bSJens Axboe unsigned int slab_size; 59f9c78b2bSJens Axboe char name[8]; 60f9c78b2bSJens Axboe }; 61f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 6249d1ec85SMing Lei static DEFINE_XARRAY(bio_slabs); 63f9c78b2bSJens Axboe 6449d1ec85SMing Lei static struct bio_slab *create_bio_slab(unsigned int size) 65f9c78b2bSJens Axboe { 6649d1ec85SMing Lei struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL); 6749d1ec85SMing Lei 6849d1ec85SMing Lei if (!bslab) 6949d1ec85SMing Lei return NULL; 7049d1ec85SMing Lei 7149d1ec85SMing Lei snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size); 7249d1ec85SMing Lei bslab->slab = kmem_cache_create(bslab->name, size, 7349d1ec85SMing Lei ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL); 7449d1ec85SMing Lei if (!bslab->slab) 7549d1ec85SMing Lei goto fail_alloc_slab; 7649d1ec85SMing Lei 7749d1ec85SMing Lei bslab->slab_ref = 1; 7849d1ec85SMing Lei bslab->slab_size = size; 7949d1ec85SMing Lei 8049d1ec85SMing Lei if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL))) 8149d1ec85SMing Lei return bslab; 8249d1ec85SMing Lei 8349d1ec85SMing Lei kmem_cache_destroy(bslab->slab); 8449d1ec85SMing Lei 8549d1ec85SMing Lei fail_alloc_slab: 8649d1ec85SMing Lei kfree(bslab); 8749d1ec85SMing Lei return NULL; 8849d1ec85SMing Lei } 8949d1ec85SMing Lei 9049d1ec85SMing Lei static inline unsigned int bs_bio_slab_size(struct bio_set *bs) 9149d1ec85SMing Lei { 929f180e31SMing Lei return bs->front_pad + sizeof(struct bio) + bs->back_pad; 9349d1ec85SMing Lei } 9449d1ec85SMing Lei 9549d1ec85SMing Lei static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs) 9649d1ec85SMing Lei { 9749d1ec85SMing Lei unsigned int size = bs_bio_slab_size(bs); 9849d1ec85SMing Lei struct bio_slab *bslab; 99f9c78b2bSJens Axboe 100f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 10149d1ec85SMing Lei bslab = xa_load(&bio_slabs, size); 10249d1ec85SMing Lei if (bslab) 103f9c78b2bSJens Axboe bslab->slab_ref++; 10449d1ec85SMing Lei else 10549d1ec85SMing Lei bslab = create_bio_slab(size); 106f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 10749d1ec85SMing Lei 10849d1ec85SMing Lei if (bslab) 10949d1ec85SMing Lei return bslab->slab; 11049d1ec85SMing Lei return NULL; 111f9c78b2bSJens Axboe } 112f9c78b2bSJens Axboe 113f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 114f9c78b2bSJens Axboe { 115f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 11649d1ec85SMing Lei unsigned int slab_size = bs_bio_slab_size(bs); 117f9c78b2bSJens Axboe 118f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 119f9c78b2bSJens Axboe 12049d1ec85SMing Lei bslab = xa_load(&bio_slabs, slab_size); 121f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 122f9c78b2bSJens Axboe goto out; 123f9c78b2bSJens Axboe 12449d1ec85SMing Lei WARN_ON_ONCE(bslab->slab != bs->bio_slab); 12549d1ec85SMing Lei 126f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 127f9c78b2bSJens Axboe 128f9c78b2bSJens Axboe if (--bslab->slab_ref) 129f9c78b2bSJens Axboe goto out; 130f9c78b2bSJens Axboe 13149d1ec85SMing Lei xa_erase(&bio_slabs, slab_size); 13249d1ec85SMing Lei 133f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 13449d1ec85SMing Lei kfree(bslab); 135f9c78b2bSJens Axboe 136f9c78b2bSJens Axboe out: 137f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 138f9c78b2bSJens Axboe } 139f9c78b2bSJens Axboe 140f9c78b2bSJens Axboe unsigned int bvec_nr_vecs(unsigned short idx) 141f9c78b2bSJens Axboe { 142d6c02a9bSGreg Edwards return bvec_slabs[--idx].nr_vecs; 143f9c78b2bSJens Axboe } 144f9c78b2bSJens Axboe 145f9c78b2bSJens Axboe void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 146f9c78b2bSJens Axboe { 147ed996a52SChristoph Hellwig if (!idx) 148ed996a52SChristoph Hellwig return; 149ed996a52SChristoph Hellwig idx--; 150f9c78b2bSJens Axboe 151ed996a52SChristoph Hellwig BIO_BUG_ON(idx >= BVEC_POOL_NR); 152ed996a52SChristoph Hellwig 153ed996a52SChristoph Hellwig if (idx == BVEC_POOL_MAX) { 154f9c78b2bSJens Axboe mempool_free(bv, pool); 155ed996a52SChristoph Hellwig } else { 156f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + idx; 157f9c78b2bSJens Axboe 158f9c78b2bSJens Axboe kmem_cache_free(bvs->slab, bv); 159f9c78b2bSJens Axboe } 160f9c78b2bSJens Axboe } 161f9c78b2bSJens Axboe 162*f2c3eb9bSChristoph Hellwig /* 163*f2c3eb9bSChristoph Hellwig * Make the first allocation restricted and don't dump info on allocation 164*f2c3eb9bSChristoph Hellwig * failures, since we'll fall back to the mempool in case of failure. 165*f2c3eb9bSChristoph Hellwig */ 166*f2c3eb9bSChristoph Hellwig static inline gfp_t bvec_alloc_gfp(gfp_t gfp) 167*f2c3eb9bSChristoph Hellwig { 168*f2c3eb9bSChristoph Hellwig return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) | 169*f2c3eb9bSChristoph Hellwig __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 170*f2c3eb9bSChristoph Hellwig } 171*f2c3eb9bSChristoph Hellwig 172f9c78b2bSJens Axboe struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 173f9c78b2bSJens Axboe mempool_t *pool) 174f9c78b2bSJens Axboe { 175f9c78b2bSJens Axboe struct bio_vec *bvl; 176f9c78b2bSJens Axboe 177f9c78b2bSJens Axboe /* 178f9c78b2bSJens Axboe * see comment near bvec_array define! 179f9c78b2bSJens Axboe */ 180f9c78b2bSJens Axboe switch (nr) { 181f9c78b2bSJens Axboe case 1: 182f9c78b2bSJens Axboe *idx = 0; 183f9c78b2bSJens Axboe break; 184f9c78b2bSJens Axboe case 2 ... 4: 185f9c78b2bSJens Axboe *idx = 1; 186f9c78b2bSJens Axboe break; 187f9c78b2bSJens Axboe case 5 ... 16: 188f9c78b2bSJens Axboe *idx = 2; 189f9c78b2bSJens Axboe break; 190f9c78b2bSJens Axboe case 17 ... 64: 191f9c78b2bSJens Axboe *idx = 3; 192f9c78b2bSJens Axboe break; 193f9c78b2bSJens Axboe case 65 ... 128: 194f9c78b2bSJens Axboe *idx = 4; 195f9c78b2bSJens Axboe break; 196f9c78b2bSJens Axboe case 129 ... BIO_MAX_PAGES: 197f9c78b2bSJens Axboe *idx = 5; 198f9c78b2bSJens Axboe break; 199f9c78b2bSJens Axboe default: 200f9c78b2bSJens Axboe return NULL; 201f9c78b2bSJens Axboe } 202f9c78b2bSJens Axboe 203f9c78b2bSJens Axboe /* 204f9c78b2bSJens Axboe * idx now points to the pool we want to allocate from. only the 205f9c78b2bSJens Axboe * 1-vec entry pool is mempool backed. 206f9c78b2bSJens Axboe */ 207ed996a52SChristoph Hellwig if (*idx == BVEC_POOL_MAX) { 208f9c78b2bSJens Axboe fallback: 209f9c78b2bSJens Axboe bvl = mempool_alloc(pool, gfp_mask); 210f9c78b2bSJens Axboe } else { 211f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + *idx; 212f9c78b2bSJens Axboe 213f9c78b2bSJens Axboe /* 214d0164adcSMel Gorman * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 215f9c78b2bSJens Axboe * is set, retry with the 1-entry mempool 216f9c78b2bSJens Axboe */ 217*f2c3eb9bSChristoph Hellwig bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask)); 218d0164adcSMel Gorman if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 219ed996a52SChristoph Hellwig *idx = BVEC_POOL_MAX; 220f9c78b2bSJens Axboe goto fallback; 221f9c78b2bSJens Axboe } 222f9c78b2bSJens Axboe } 223f9c78b2bSJens Axboe 224ed996a52SChristoph Hellwig (*idx)++; 225f9c78b2bSJens Axboe return bvl; 226f9c78b2bSJens Axboe } 227f9c78b2bSJens Axboe 2289ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 229f9c78b2bSJens Axboe { 230db9819c7SChristoph Hellwig #ifdef CONFIG_BLK_CGROUP 231db9819c7SChristoph Hellwig if (bio->bi_blkg) { 232db9819c7SChristoph Hellwig blkg_put(bio->bi_blkg); 233db9819c7SChristoph Hellwig bio->bi_blkg = NULL; 234db9819c7SChristoph Hellwig } 235db9819c7SChristoph Hellwig #endif 236ece841abSJustin Tee if (bio_integrity(bio)) 237ece841abSJustin Tee bio_integrity_free(bio); 238a892c8d5SSatya Tangirala 239a892c8d5SSatya Tangirala bio_crypt_free_ctx(bio); 240f9c78b2bSJens Axboe } 2419ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 242f9c78b2bSJens Axboe 243f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 244f9c78b2bSJens Axboe { 245f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 246f9c78b2bSJens Axboe void *p; 247f9c78b2bSJens Axboe 2489ae3b3f5SJens Axboe bio_uninit(bio); 249f9c78b2bSJens Axboe 250f9c78b2bSJens Axboe if (bs) { 2518aa6ba2fSKent Overstreet bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 252f9c78b2bSJens Axboe 253f9c78b2bSJens Axboe /* 254f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 255f9c78b2bSJens Axboe */ 256f9c78b2bSJens Axboe p = bio; 257f9c78b2bSJens Axboe p -= bs->front_pad; 258f9c78b2bSJens Axboe 2598aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 260f9c78b2bSJens Axboe } else { 261f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 262f9c78b2bSJens Axboe kfree(bio); 263f9c78b2bSJens Axboe } 264f9c78b2bSJens Axboe } 265f9c78b2bSJens Axboe 2669ae3b3f5SJens Axboe /* 2679ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2689ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2699ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2709ae3b3f5SJens Axboe */ 2713a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2723a83f467SMing Lei unsigned short max_vecs) 273f9c78b2bSJens Axboe { 274f9c78b2bSJens Axboe memset(bio, 0, sizeof(*bio)); 275c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 276dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2773a83f467SMing Lei 2783a83f467SMing Lei bio->bi_io_vec = table; 2793a83f467SMing Lei bio->bi_max_vecs = max_vecs; 280f9c78b2bSJens Axboe } 281f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 282f9c78b2bSJens Axboe 283f9c78b2bSJens Axboe /** 284f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 285f9c78b2bSJens Axboe * @bio: bio to reset 286f9c78b2bSJens Axboe * 287f9c78b2bSJens Axboe * Description: 288f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 289f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 290f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 291f9c78b2bSJens Axboe * comment in struct bio. 292f9c78b2bSJens Axboe */ 293f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 294f9c78b2bSJens Axboe { 295f9c78b2bSJens Axboe unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 296f9c78b2bSJens Axboe 2979ae3b3f5SJens Axboe bio_uninit(bio); 298f9c78b2bSJens Axboe 299f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 3004246a0b6SChristoph Hellwig bio->bi_flags = flags; 301c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 302f9c78b2bSJens Axboe } 303f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 304f9c78b2bSJens Axboe 30538f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 306f9c78b2bSJens Axboe { 3074246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3084246a0b6SChristoph Hellwig 3094e4cbee9SChristoph Hellwig if (!parent->bi_status) 3104e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 311f9c78b2bSJens Axboe bio_put(bio); 31238f8baaeSChristoph Hellwig return parent; 31338f8baaeSChristoph Hellwig } 31438f8baaeSChristoph Hellwig 31538f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 31638f8baaeSChristoph Hellwig { 31738f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 318f9c78b2bSJens Axboe } 319f9c78b2bSJens Axboe 320f9c78b2bSJens Axboe /** 321f9c78b2bSJens Axboe * bio_chain - chain bio completions 322f9c78b2bSJens Axboe * @bio: the target bio 3235b874af6SMauro Carvalho Chehab * @parent: the parent bio of @bio 324f9c78b2bSJens Axboe * 325f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 326f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 327f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 328f9c78b2bSJens Axboe * 329f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 330f9c78b2bSJens Axboe */ 331f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 332f9c78b2bSJens Axboe { 333f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 334f9c78b2bSJens Axboe 335f9c78b2bSJens Axboe bio->bi_private = parent; 336f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 337c4cf5261SJens Axboe bio_inc_remaining(parent); 338f9c78b2bSJens Axboe } 339f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 340f9c78b2bSJens Axboe 341f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 342f9c78b2bSJens Axboe { 343f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 344f9c78b2bSJens Axboe struct bio *bio; 345f9c78b2bSJens Axboe 346f9c78b2bSJens Axboe while (1) { 347f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 348f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 349f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 350f9c78b2bSJens Axboe 351f9c78b2bSJens Axboe if (!bio) 352f9c78b2bSJens Axboe break; 353f9c78b2bSJens Axboe 354ed00aabdSChristoph Hellwig submit_bio_noacct(bio); 355f9c78b2bSJens Axboe } 356f9c78b2bSJens Axboe } 357f9c78b2bSJens Axboe 358f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 359f9c78b2bSJens Axboe { 360f9c78b2bSJens Axboe struct bio_list punt, nopunt; 361f9c78b2bSJens Axboe struct bio *bio; 362f9c78b2bSJens Axboe 36347e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 36447e0fb46SNeilBrown return; 365f9c78b2bSJens Axboe /* 366f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 367f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 368f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 369f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 370f9c78b2bSJens Axboe * our own rescuer would be bad. 371f9c78b2bSJens Axboe * 372f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 373f9c78b2bSJens Axboe * remove from the middle of the list: 374f9c78b2bSJens Axboe */ 375f9c78b2bSJens Axboe 376f9c78b2bSJens Axboe bio_list_init(&punt); 377f9c78b2bSJens Axboe bio_list_init(&nopunt); 378f9c78b2bSJens Axboe 379f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 380f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 381f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 382f9c78b2bSJens Axboe 383f5fe1b51SNeilBrown bio_list_init(&nopunt); 384f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 385f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 386f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 387f9c78b2bSJens Axboe 388f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 389f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 390f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 391f9c78b2bSJens Axboe 392f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 393f9c78b2bSJens Axboe } 394f9c78b2bSJens Axboe 395f9c78b2bSJens Axboe /** 396f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 397519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 398f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 399f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 400f9c78b2bSJens Axboe * 4013175199aSChristoph Hellwig * Allocate a bio from the mempools in @bs. 402f9c78b2bSJens Axboe * 4033175199aSChristoph Hellwig * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to 4043175199aSChristoph Hellwig * allocate a bio. This is due to the mempool guarantees. To make this work, 4053175199aSChristoph Hellwig * callers must never allocate more than 1 bio at a time from the general pool. 4063175199aSChristoph Hellwig * Callers that need to allocate more than 1 bio must always submit the 4073175199aSChristoph Hellwig * previously allocated bio for IO before attempting to allocate a new one. 4083175199aSChristoph Hellwig * Failure to do so can cause deadlocks under memory pressure. 409f9c78b2bSJens Axboe * 4103175199aSChristoph Hellwig * Note that when running under submit_bio_noacct() (i.e. any block driver), 4113175199aSChristoph Hellwig * bios are not submitted until after you return - see the code in 412ed00aabdSChristoph Hellwig * submit_bio_noacct() that converts recursion into iteration, to prevent 413f9c78b2bSJens Axboe * stack overflows. 414f9c78b2bSJens Axboe * 4153175199aSChristoph Hellwig * This would normally mean allocating multiple bios under submit_bio_noacct() 4163175199aSChristoph Hellwig * would be susceptible to deadlocks, but we have 417f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 418f9c78b2bSJens Axboe * thread. 419f9c78b2bSJens Axboe * 420f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 421f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 422ed00aabdSChristoph Hellwig * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad 423f9c78b2bSJens Axboe * for per bio allocations. 424f9c78b2bSJens Axboe * 4253175199aSChristoph Hellwig * Returns: Pointer to new bio on success, NULL on failure. 426f9c78b2bSJens Axboe */ 4277a88fa19SDan Carpenter struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 4287a88fa19SDan Carpenter struct bio_set *bs) 429f9c78b2bSJens Axboe { 430f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 431f9c78b2bSJens Axboe struct bio *bio; 432f9c78b2bSJens Axboe void *p; 433f9c78b2bSJens Axboe 4343175199aSChristoph Hellwig /* should not use nobvec bioset for nr_iovecs > 0 */ 4353175199aSChristoph Hellwig if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0)) 436f9c78b2bSJens Axboe return NULL; 437f9c78b2bSJens Axboe 438f9c78b2bSJens Axboe /* 4393175199aSChristoph Hellwig * submit_bio_noacct() converts recursion to iteration; this means if 4403175199aSChristoph Hellwig * we're running beneath it, any bios we allocate and submit will not be 4413175199aSChristoph Hellwig * submitted (and thus freed) until after we return. 442f9c78b2bSJens Axboe * 4433175199aSChristoph Hellwig * This exposes us to a potential deadlock if we allocate multiple bios 4443175199aSChristoph Hellwig * from the same bio_set() while running underneath submit_bio_noacct(). 4453175199aSChristoph Hellwig * If we were to allocate multiple bios (say a stacking block driver 4463175199aSChristoph Hellwig * that was splitting bios), we would deadlock if we exhausted the 4473175199aSChristoph Hellwig * mempool's reserve. 448f9c78b2bSJens Axboe * 449f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 4503175199aSChristoph Hellwig * workqueue per bio_set. If we go to allocate and there are bios on 4513175199aSChristoph Hellwig * current->bio_list, we first try the allocation without 4523175199aSChristoph Hellwig * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be 4533175199aSChristoph Hellwig * blocking to the rescuer workqueue before we retry with the original 4543175199aSChristoph Hellwig * gfp_flags. 455f9c78b2bSJens Axboe */ 456f5fe1b51SNeilBrown if (current->bio_list && 457f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 45847e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 45947e0fb46SNeilBrown bs->rescue_workqueue) 460d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 461f9c78b2bSJens Axboe 4628aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 463f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 464f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 465f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4668aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 467f9c78b2bSJens Axboe } 468f9c78b2bSJens Axboe if (unlikely(!p)) 469f9c78b2bSJens Axboe return NULL; 470f9c78b2bSJens Axboe 4713175199aSChristoph Hellwig bio = p + bs->front_pad; 4723175199aSChristoph Hellwig if (nr_iovecs > BIO_INLINE_VECS) { 473ed996a52SChristoph Hellwig unsigned long idx = 0; 4743175199aSChristoph Hellwig struct bio_vec *bvl = NULL; 475ed996a52SChristoph Hellwig 4768aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 477f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 478f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 479f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4803175199aSChristoph Hellwig bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, 4813175199aSChristoph Hellwig &bs->bvec_pool); 482f9c78b2bSJens Axboe } 483f9c78b2bSJens Axboe 484f9c78b2bSJens Axboe if (unlikely(!bvl)) 485f9c78b2bSJens Axboe goto err_free; 486f9c78b2bSJens Axboe 4873175199aSChristoph Hellwig bio_init(bio, bvl, bvec_nr_vecs(idx)); 4888358c28aSMing Lei bio->bi_flags |= idx << BVEC_POOL_OFFSET; 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 */ 5133175199aSChristoph Hellwig struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned int 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 52938a72dacSKent Overstreet void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 530f9c78b2bSJens Axboe { 531f9c78b2bSJens Axboe unsigned long flags; 532f9c78b2bSJens Axboe struct bio_vec bv; 533f9c78b2bSJens Axboe struct bvec_iter iter; 534f9c78b2bSJens Axboe 53538a72dacSKent Overstreet __bio_for_each_segment(bv, bio, iter, start) { 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 } 54238a72dacSKent Overstreet EXPORT_SYMBOL(zero_fill_bio_iter); 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 62729125ed6SChristoph Hellwig /** 628f9c78b2bSJens Axboe * bio_put - release a reference to a bio 629f9c78b2bSJens Axboe * @bio: bio to release reference to 630f9c78b2bSJens Axboe * 631f9c78b2bSJens Axboe * Description: 632f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 6339b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 634f9c78b2bSJens Axboe **/ 635f9c78b2bSJens Axboe void bio_put(struct bio *bio) 636f9c78b2bSJens Axboe { 637dac56212SJens Axboe if (!bio_flagged(bio, BIO_REFFED)) 638dac56212SJens Axboe bio_free(bio); 639dac56212SJens Axboe else { 640dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 641f9c78b2bSJens Axboe 642f9c78b2bSJens Axboe /* 643f9c78b2bSJens Axboe * last put frees it 644f9c78b2bSJens Axboe */ 645dac56212SJens Axboe if (atomic_dec_and_test(&bio->__bi_cnt)) 646f9c78b2bSJens Axboe bio_free(bio); 647f9c78b2bSJens Axboe } 648dac56212SJens Axboe } 649f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 650f9c78b2bSJens Axboe 651f9c78b2bSJens Axboe /** 652f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 653f9c78b2bSJens Axboe * @bio: destination bio 654f9c78b2bSJens Axboe * @bio_src: bio to clone 655f9c78b2bSJens Axboe * 656f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 657f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 658f9c78b2bSJens Axboe * bio will be one. 659f9c78b2bSJens Axboe * 660f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 661f9c78b2bSJens Axboe */ 662f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 663f9c78b2bSJens Axboe { 664ed996a52SChristoph Hellwig BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 665f9c78b2bSJens Axboe 666f9c78b2bSJens Axboe /* 667309dca30SChristoph Hellwig * most users will be overriding ->bi_bdev with a new target, 668f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 669f9c78b2bSJens Axboe */ 670309dca30SChristoph Hellwig bio->bi_bdev = bio_src->bi_bdev; 671b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 672111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 673111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 67446bbf653SChristoph Hellwig if (bio_flagged(bio_src, BIO_REMAPPED)) 67546bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 6761eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 677ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 678cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 679f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 680f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 68120bd723eSPaolo Valente 682db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 683e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 684f9c78b2bSJens Axboe } 685f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 686f9c78b2bSJens Axboe 687f9c78b2bSJens Axboe /** 688f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 689f9c78b2bSJens Axboe * @bio: bio to clone 690f9c78b2bSJens Axboe * @gfp_mask: allocation priority 691f9c78b2bSJens Axboe * @bs: bio_set to allocate from 692f9c78b2bSJens Axboe * 693f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 694f9c78b2bSJens Axboe */ 695f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 696f9c78b2bSJens Axboe { 697f9c78b2bSJens Axboe struct bio *b; 698f9c78b2bSJens Axboe 699f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 700f9c78b2bSJens Axboe if (!b) 701f9c78b2bSJens Axboe return NULL; 702f9c78b2bSJens Axboe 703f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 704f9c78b2bSJens Axboe 70507560151SEric Biggers if (bio_crypt_clone(b, bio, gfp_mask) < 0) 70607560151SEric Biggers goto err_put; 707a892c8d5SSatya Tangirala 70807560151SEric Biggers if (bio_integrity(bio) && 70907560151SEric Biggers bio_integrity_clone(b, bio, gfp_mask) < 0) 71007560151SEric Biggers goto err_put; 711f9c78b2bSJens Axboe 712f9c78b2bSJens Axboe return b; 71307560151SEric Biggers 71407560151SEric Biggers err_put: 71507560151SEric Biggers bio_put(b); 71607560151SEric Biggers return NULL; 717f9c78b2bSJens Axboe } 718f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 719f9c78b2bSJens Axboe 7205cbd28e3SChristoph Hellwig const char *bio_devname(struct bio *bio, char *buf) 7215cbd28e3SChristoph Hellwig { 722309dca30SChristoph Hellwig return bdevname(bio->bi_bdev, buf); 7235cbd28e3SChristoph Hellwig } 7245cbd28e3SChristoph Hellwig EXPORT_SYMBOL(bio_devname); 7255cbd28e3SChristoph Hellwig 7265919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 7275919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 728ff896738SChristoph Hellwig bool *same_page) 7295919482eSMing Lei { 730d8166519SMatthew Wilcox (Oracle) size_t bv_end = bv->bv_offset + bv->bv_len; 731d8166519SMatthew Wilcox (Oracle) phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1; 7325919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 7335919482eSMing Lei 7345919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 7355919482eSMing Lei return false; 7365919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 7375919482eSMing Lei return false; 73852d52d1cSChristoph Hellwig 739ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 740d8166519SMatthew Wilcox (Oracle) if (*same_page) 7415919482eSMing Lei return true; 742d8166519SMatthew Wilcox (Oracle) return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE); 7435919482eSMing Lei } 7445919482eSMing Lei 745e4581105SChristoph Hellwig /* 746e4581105SChristoph Hellwig * Try to merge a page into a segment, while obeying the hardware segment 747e4581105SChristoph Hellwig * size limit. This is not for normal read/write bios, but for passthrough 748e4581105SChristoph Hellwig * or Zone Append operations that we can't split. 749e4581105SChristoph Hellwig */ 750e4581105SChristoph Hellwig static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio, 751e4581105SChristoph Hellwig struct page *page, unsigned len, 752e4581105SChristoph Hellwig unsigned offset, bool *same_page) 753489fbbcbSMing Lei { 754384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 755489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 756489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 757489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 758489fbbcbSMing Lei 759489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 760489fbbcbSMing Lei return false; 761489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 762489fbbcbSMing Lei return false; 763384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 764489fbbcbSMing Lei } 765489fbbcbSMing Lei 766f4595875SShaohua Li /** 767e4581105SChristoph Hellwig * bio_add_hw_page - attempt to add a page to a bio with hw constraints 768c66a14d0SKent Overstreet * @q: the target queue 769c66a14d0SKent Overstreet * @bio: destination bio 770c66a14d0SKent Overstreet * @page: page to add 771c66a14d0SKent Overstreet * @len: vec entry length 772c66a14d0SKent Overstreet * @offset: vec entry offset 773e4581105SChristoph Hellwig * @max_sectors: maximum number of sectors that can be added 774e4581105SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 775f9c78b2bSJens Axboe * 776e4581105SChristoph Hellwig * Add a page to a bio while respecting the hardware max_sectors, max_segment 777e4581105SChristoph Hellwig * and gap limitations. 778f9c78b2bSJens Axboe */ 779e4581105SChristoph Hellwig int bio_add_hw_page(struct request_queue *q, struct bio *bio, 78019047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 781e4581105SChristoph Hellwig unsigned int max_sectors, bool *same_page) 782f9c78b2bSJens Axboe { 783f9c78b2bSJens Axboe struct bio_vec *bvec; 784f9c78b2bSJens Axboe 785e4581105SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 786f9c78b2bSJens Axboe return 0; 787f9c78b2bSJens Axboe 788e4581105SChristoph Hellwig if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 789f9c78b2bSJens Axboe return 0; 790f9c78b2bSJens Axboe 791f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 792e4581105SChristoph Hellwig if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page)) 793384209cdSChristoph Hellwig return len; 794320ea869SChristoph Hellwig 795320ea869SChristoph Hellwig /* 796320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 797320ea869SChristoph Hellwig * would create a gap, disallow it. 798320ea869SChristoph Hellwig */ 799384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 800320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 801320ea869SChristoph Hellwig return 0; 802f9c78b2bSJens Axboe } 803f9c78b2bSJens Axboe 80479d08f89SMing Lei if (bio_full(bio, len)) 805f9c78b2bSJens Axboe return 0; 806f9c78b2bSJens Axboe 80714ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 808489fbbcbSMing Lei return 0; 809489fbbcbSMing Lei 810f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 811f9c78b2bSJens Axboe bvec->bv_page = page; 812f9c78b2bSJens Axboe bvec->bv_len = len; 813f9c78b2bSJens Axboe bvec->bv_offset = offset; 814fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 815dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 816f9c78b2bSJens Axboe return len; 817f9c78b2bSJens Axboe } 81819047087SMing Lei 819e4581105SChristoph Hellwig /** 820e4581105SChristoph Hellwig * bio_add_pc_page - attempt to add page to passthrough bio 821e4581105SChristoph Hellwig * @q: the target queue 822e4581105SChristoph Hellwig * @bio: destination bio 823e4581105SChristoph Hellwig * @page: page to add 824e4581105SChristoph Hellwig * @len: vec entry length 825e4581105SChristoph Hellwig * @offset: vec entry offset 826e4581105SChristoph Hellwig * 827e4581105SChristoph Hellwig * Attempt to add a page to the bio_vec maplist. This can fail for a 828e4581105SChristoph Hellwig * number of reasons, such as the bio being full or target block device 829e4581105SChristoph Hellwig * limitations. The target block device must allow bio's up to PAGE_SIZE, 830e4581105SChristoph Hellwig * so it is always possible to add a single page to an empty bio. 831e4581105SChristoph Hellwig * 832e4581105SChristoph Hellwig * This should only be used by passthrough bios. 833e4581105SChristoph Hellwig */ 83419047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 83519047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 83619047087SMing Lei { 837d1916c86SChristoph Hellwig bool same_page = false; 838e4581105SChristoph Hellwig return bio_add_hw_page(q, bio, page, len, offset, 839e4581105SChristoph Hellwig queue_max_hw_sectors(q), &same_page); 84019047087SMing Lei } 841f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 842f9c78b2bSJens Axboe 843f9c78b2bSJens Axboe /** 8440aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 8450aa69fd3SChristoph Hellwig * @bio: destination bio 846551879a4SMing Lei * @page: start page to add 8470aa69fd3SChristoph Hellwig * @len: length of the data to add 848551879a4SMing Lei * @off: offset of the data relative to @page 849ff896738SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 8500aa69fd3SChristoph Hellwig * 8510aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 8523cf14889SRandy Dunlap * useful optimisation for file systems with a block size smaller than the 8530aa69fd3SChristoph Hellwig * page size. 8540aa69fd3SChristoph Hellwig * 855551879a4SMing Lei * Warn if (@len, @off) crosses pages in case that @same_page is true. 856551879a4SMing Lei * 8570aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 8580aa69fd3SChristoph Hellwig */ 8590aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 860ff896738SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 8610aa69fd3SChristoph Hellwig { 8620aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8630aa69fd3SChristoph Hellwig return false; 8640aa69fd3SChristoph Hellwig 865cc90bc68SAndreas Gruenbacher if (bio->bi_vcnt > 0) { 8660aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 8670aa69fd3SChristoph Hellwig 8685919482eSMing Lei if (page_is_mergeable(bv, page, len, off, same_page)) { 8692cd896a5SRitesh Harjani if (bio->bi_iter.bi_size > UINT_MAX - len) { 8702cd896a5SRitesh Harjani *same_page = false; 871cc90bc68SAndreas Gruenbacher return false; 8722cd896a5SRitesh Harjani } 8730aa69fd3SChristoph Hellwig bv->bv_len += len; 8740aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8750aa69fd3SChristoph Hellwig return true; 8760aa69fd3SChristoph Hellwig } 8775919482eSMing Lei } 8780aa69fd3SChristoph Hellwig return false; 8790aa69fd3SChristoph Hellwig } 8800aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 8810aa69fd3SChristoph Hellwig 8820aa69fd3SChristoph Hellwig /** 883551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 8840aa69fd3SChristoph Hellwig * @bio: destination bio 885551879a4SMing Lei * @page: start page to add 886551879a4SMing Lei * @len: length of the data to add, may cross pages 887551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 8880aa69fd3SChristoph Hellwig * 8890aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 8900aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 8910aa69fd3SChristoph Hellwig */ 8920aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 8930aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 8940aa69fd3SChristoph Hellwig { 8950aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 8960aa69fd3SChristoph Hellwig 8970aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 89879d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 8990aa69fd3SChristoph Hellwig 9000aa69fd3SChristoph Hellwig bv->bv_page = page; 9010aa69fd3SChristoph Hellwig bv->bv_offset = off; 9020aa69fd3SChristoph Hellwig bv->bv_len = len; 9030aa69fd3SChristoph Hellwig 9040aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 9050aa69fd3SChristoph Hellwig bio->bi_vcnt++; 906b8e24a93SJohannes Weiner 907b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 908b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 9090aa69fd3SChristoph Hellwig } 9100aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 9110aa69fd3SChristoph Hellwig 9120aa69fd3SChristoph Hellwig /** 913551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 914f9c78b2bSJens Axboe * @bio: destination bio 915551879a4SMing Lei * @page: start page to add 916551879a4SMing Lei * @len: vec entry length, may cross pages 917551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 918f9c78b2bSJens Axboe * 919551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 920c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 921f9c78b2bSJens Axboe */ 922c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 923c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 924f9c78b2bSJens Axboe { 925ff896738SChristoph Hellwig bool same_page = false; 926ff896738SChristoph Hellwig 927ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 92879d08f89SMing Lei if (bio_full(bio, len)) 929c66a14d0SKent Overstreet return 0; 9300aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 931c66a14d0SKent Overstreet } 932c66a14d0SKent Overstreet return len; 933f9c78b2bSJens Axboe } 934f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 935f9c78b2bSJens Axboe 936d241a95fSChristoph Hellwig void bio_release_pages(struct bio *bio, bool mark_dirty) 9377321ecbfSChristoph Hellwig { 9387321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 9397321ecbfSChristoph Hellwig struct bio_vec *bvec; 9407321ecbfSChristoph Hellwig 941b2d0d991SChristoph Hellwig if (bio_flagged(bio, BIO_NO_PAGE_REF)) 942b2d0d991SChristoph Hellwig return; 943b2d0d991SChristoph Hellwig 944d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 945d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 946d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 9477321ecbfSChristoph Hellwig put_page(bvec->bv_page); 9487321ecbfSChristoph Hellwig } 949d241a95fSChristoph Hellwig } 95029b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_release_pages); 9517321ecbfSChristoph Hellwig 952c42bca92SPavel Begunkov static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter) 9536d0c48aeSJens Axboe { 954c42bca92SPavel Begunkov WARN_ON_ONCE(BVEC_POOL_IDX(bio) != 0); 9556d0c48aeSJens Axboe 956c42bca92SPavel Begunkov bio->bi_vcnt = iter->nr_segs; 957c42bca92SPavel Begunkov bio->bi_max_vecs = iter->nr_segs; 958c42bca92SPavel Begunkov bio->bi_io_vec = (struct bio_vec *)iter->bvec; 959c42bca92SPavel Begunkov bio->bi_iter.bi_bvec_done = iter->iov_offset; 960c42bca92SPavel Begunkov bio->bi_iter.bi_size = iter->count; 9616d0c48aeSJens Axboe 962c42bca92SPavel Begunkov iov_iter_advance(iter, iter->count); 9636d0c48aeSJens Axboe return 0; 9646d0c48aeSJens Axboe } 9656d0c48aeSJens Axboe 966576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 967576ed913SChristoph Hellwig 9682cefe4dbSKent Overstreet /** 96917d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 9702cefe4dbSKent Overstreet * @bio: bio to add pages to 9712cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 9722cefe4dbSKent Overstreet * 97317d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 9742cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 97517d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 9763cf14889SRandy Dunlap * next non-empty segment of the iov iterator. 9772cefe4dbSKent Overstreet */ 97817d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 9792cefe4dbSKent Overstreet { 980576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 981576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 9822cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 9832cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 98445691804SChristoph Hellwig bool same_page = false; 985576ed913SChristoph Hellwig ssize_t size, left; 986576ed913SChristoph Hellwig unsigned len, i; 987b403ea24SMartin Wilck size_t offset; 988576ed913SChristoph Hellwig 989576ed913SChristoph Hellwig /* 990576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 991576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 992576ed913SChristoph Hellwig * without overwriting the temporary page array. 993576ed913SChristoph Hellwig */ 994576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 995576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 9962cefe4dbSKent Overstreet 9972cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 9982cefe4dbSKent Overstreet if (unlikely(size <= 0)) 9992cefe4dbSKent Overstreet return size ? size : -EFAULT; 10002cefe4dbSKent Overstreet 1001576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 1002576ed913SChristoph Hellwig struct page *page = pages[i]; 10032cefe4dbSKent Overstreet 1004576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 100545691804SChristoph Hellwig 100645691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 100745691804SChristoph Hellwig if (same_page) 100845691804SChristoph Hellwig put_page(page); 100945691804SChristoph Hellwig } else { 101079d08f89SMing Lei if (WARN_ON_ONCE(bio_full(bio, len))) 1011576ed913SChristoph Hellwig return -EINVAL; 101245691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 101345691804SChristoph Hellwig } 1014576ed913SChristoph Hellwig offset = 0; 10152cefe4dbSKent Overstreet } 10162cefe4dbSKent Overstreet 10172cefe4dbSKent Overstreet iov_iter_advance(iter, size); 10182cefe4dbSKent Overstreet return 0; 10192cefe4dbSKent Overstreet } 102017d51b10SMartin Wilck 10210512a75bSKeith Busch static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter) 10220512a75bSKeith Busch { 10230512a75bSKeith Busch unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 10240512a75bSKeith Busch unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 1025309dca30SChristoph Hellwig struct request_queue *q = bio->bi_bdev->bd_disk->queue; 10260512a75bSKeith Busch unsigned int max_append_sectors = queue_max_zone_append_sectors(q); 10270512a75bSKeith Busch struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 10280512a75bSKeith Busch struct page **pages = (struct page **)bv; 10290512a75bSKeith Busch ssize_t size, left; 10300512a75bSKeith Busch unsigned len, i; 10310512a75bSKeith Busch size_t offset; 10324977d121SNaohiro Aota int ret = 0; 10330512a75bSKeith Busch 10340512a75bSKeith Busch if (WARN_ON_ONCE(!max_append_sectors)) 10350512a75bSKeith Busch return 0; 10360512a75bSKeith Busch 10370512a75bSKeith Busch /* 10380512a75bSKeith Busch * Move page array up in the allocated memory for the bio vecs as far as 10390512a75bSKeith Busch * possible so that we can start filling biovecs from the beginning 10400512a75bSKeith Busch * without overwriting the temporary page array. 10410512a75bSKeith Busch */ 10420512a75bSKeith Busch BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 10430512a75bSKeith Busch pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 10440512a75bSKeith Busch 10450512a75bSKeith Busch size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 10460512a75bSKeith Busch if (unlikely(size <= 0)) 10470512a75bSKeith Busch return size ? size : -EFAULT; 10480512a75bSKeith Busch 10490512a75bSKeith Busch for (left = size, i = 0; left > 0; left -= len, i++) { 10500512a75bSKeith Busch struct page *page = pages[i]; 10510512a75bSKeith Busch bool same_page = false; 10520512a75bSKeith Busch 10530512a75bSKeith Busch len = min_t(size_t, PAGE_SIZE - offset, left); 10540512a75bSKeith Busch if (bio_add_hw_page(q, bio, page, len, offset, 10554977d121SNaohiro Aota max_append_sectors, &same_page) != len) { 10564977d121SNaohiro Aota ret = -EINVAL; 10574977d121SNaohiro Aota break; 10584977d121SNaohiro Aota } 10590512a75bSKeith Busch if (same_page) 10600512a75bSKeith Busch put_page(page); 10610512a75bSKeith Busch offset = 0; 10620512a75bSKeith Busch } 10630512a75bSKeith Busch 10644977d121SNaohiro Aota iov_iter_advance(iter, size - left); 10654977d121SNaohiro Aota return ret; 10660512a75bSKeith Busch } 10670512a75bSKeith Busch 106817d51b10SMartin Wilck /** 10696d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 107017d51b10SMartin Wilck * @bio: bio to add pages to 10716d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 107217d51b10SMartin Wilck * 10736d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 10746d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 10756d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 1076c42bca92SPavel Begunkov * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided 1077c42bca92SPavel Begunkov * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs 1078c42bca92SPavel Begunkov * to ensure the bvecs and pages stay referenced until the submitted I/O is 1079c42bca92SPavel Begunkov * completed by a call to ->ki_complete() or returns with an error other than 1080c42bca92SPavel Begunkov * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF 1081c42bca92SPavel Begunkov * on IO completion. If it isn't, then pages should be released. 10826d0c48aeSJens Axboe * 108317d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 10845cd3ddc1SMauro Carvalho Chehab * fit into the bio, or are requested in @iter, whatever is smaller. If 10856d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 10866d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 10870cf41e5eSPavel Begunkov * 10880cf41e5eSPavel Begunkov * It's intended for direct IO, so doesn't do PSI tracking, the caller is 10890cf41e5eSPavel Begunkov * responsible for setting BIO_WORKINGSET if necessary. 109017d51b10SMartin Wilck */ 109117d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 109217d51b10SMartin Wilck { 1093c42bca92SPavel Begunkov int ret = 0; 109414eacf12SChristoph Hellwig 1095c42bca92SPavel Begunkov if (iov_iter_is_bvec(iter)) { 1096c42bca92SPavel Begunkov if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 109714eacf12SChristoph Hellwig return -EINVAL; 1098c42bca92SPavel Begunkov bio_iov_bvec_set(bio, iter); 1099c42bca92SPavel Begunkov bio_set_flag(bio, BIO_NO_PAGE_REF); 1100c42bca92SPavel Begunkov return 0; 11010512a75bSKeith Busch } else { 1102c42bca92SPavel Begunkov do { 1103c42bca92SPavel Begunkov if (bio_op(bio) == REQ_OP_ZONE_APPEND) 1104c42bca92SPavel Begunkov ret = __bio_iov_append_get_pages(bio, iter); 11056d0c48aeSJens Axboe else 11066d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 110779d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 1108c42bca92SPavel Begunkov } 11090cf41e5eSPavel Begunkov 11100cf41e5eSPavel Begunkov /* don't account direct I/O as memory stall */ 11110cf41e5eSPavel Begunkov bio_clear_flag(bio, BIO_WORKINGSET); 111214eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 111317d51b10SMartin Wilck } 111429b2a3aaSJohannes Thumshirn EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 11152cefe4dbSKent Overstreet 11164246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1117f9c78b2bSJens Axboe { 111865e53aabSChristoph Hellwig complete(bio->bi_private); 1119f9c78b2bSJens Axboe } 1120f9c78b2bSJens Axboe 1121f9c78b2bSJens Axboe /** 1122f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1123f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1124f9c78b2bSJens Axboe * 1125f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1126f9c78b2bSJens Axboe * bio_endio() on failure. 11273d289d68SJan Kara * 11283d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 11293d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 11303d289d68SJan Kara * on his own. 1131f9c78b2bSJens Axboe */ 11324e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1133f9c78b2bSJens Axboe { 1134309dca30SChristoph Hellwig DECLARE_COMPLETION_ONSTACK_MAP(done, 1135309dca30SChristoph Hellwig bio->bi_bdev->bd_disk->lockdep_map); 1136de6a78b6SMing Lei unsigned long hang_check; 1137f9c78b2bSJens Axboe 113865e53aabSChristoph Hellwig bio->bi_private = &done; 1139f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 11401eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 11414e49ea4aSMike Christie submit_bio(bio); 1142de6a78b6SMing Lei 1143de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1144de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1145de6a78b6SMing Lei if (hang_check) 1146de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1147de6a78b6SMing Lei hang_check * (HZ/2))) 1148de6a78b6SMing Lei ; 1149de6a78b6SMing Lei else 115065e53aabSChristoph Hellwig wait_for_completion_io(&done); 1151f9c78b2bSJens Axboe 115265e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1153f9c78b2bSJens Axboe } 1154f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1155f9c78b2bSJens Axboe 1156f9c78b2bSJens Axboe /** 1157f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 1158f9c78b2bSJens Axboe * @bio: bio to advance 1159f9c78b2bSJens Axboe * @bytes: number of bytes to complete 1160f9c78b2bSJens Axboe * 1161f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1162f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1163f9c78b2bSJens Axboe * be updated on the last bvec as well. 1164f9c78b2bSJens Axboe * 1165f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 1166f9c78b2bSJens Axboe */ 1167f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 1168f9c78b2bSJens Axboe { 1169f9c78b2bSJens Axboe if (bio_integrity(bio)) 1170f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1171f9c78b2bSJens Axboe 1172a892c8d5SSatya Tangirala bio_crypt_advance(bio, bytes); 1173f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1174f9c78b2bSJens Axboe } 1175f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 1176f9c78b2bSJens Axboe 117745db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 117845db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1179f9c78b2bSJens Axboe { 1180f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1181f9c78b2bSJens Axboe void *src_p, *dst_p; 1182f9c78b2bSJens Axboe unsigned bytes; 1183f9c78b2bSJens Axboe 118445db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 118545db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 118645db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 118745db54d5SKent Overstreet 118845db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 118945db54d5SKent Overstreet 119045db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 119145db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 119245db54d5SKent Overstreet 119345db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 119445db54d5SKent Overstreet src_p + src_bv.bv_offset, 119545db54d5SKent Overstreet bytes); 119645db54d5SKent Overstreet 119745db54d5SKent Overstreet kunmap_atomic(dst_p); 119845db54d5SKent Overstreet kunmap_atomic(src_p); 119945db54d5SKent Overstreet 12006e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 12016e6e811dSKent Overstreet 120222b56c29SPavel Begunkov bio_advance_iter_single(src, src_iter, bytes); 120322b56c29SPavel Begunkov bio_advance_iter_single(dst, dst_iter, bytes); 120445db54d5SKent Overstreet } 120545db54d5SKent Overstreet } 120645db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 120745db54d5SKent Overstreet 120845db54d5SKent Overstreet /** 120945db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 121045db54d5SKent Overstreet * @src: source bio 121145db54d5SKent Overstreet * @dst: destination bio 121245db54d5SKent Overstreet * 121345db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 121445db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 121545db54d5SKent Overstreet */ 121645db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 121745db54d5SKent Overstreet { 121845db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 121945db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 122045db54d5SKent Overstreet 122145db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 122245db54d5SKent Overstreet } 122345db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 122445db54d5SKent Overstreet 122545db54d5SKent Overstreet /** 122645db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 122745db54d5SKent Overstreet * another 122845db54d5SKent Overstreet * @src: source bio list 122945db54d5SKent Overstreet * @dst: destination bio list 123045db54d5SKent Overstreet * 123145db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 123245db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 123345db54d5SKent Overstreet * bios). 123445db54d5SKent Overstreet */ 123545db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 123645db54d5SKent Overstreet { 123745db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 123845db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 123945db54d5SKent Overstreet 1240f9c78b2bSJens Axboe while (1) { 1241f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1242f9c78b2bSJens Axboe src = src->bi_next; 1243f9c78b2bSJens Axboe if (!src) 1244f9c78b2bSJens Axboe break; 1245f9c78b2bSJens Axboe 1246f9c78b2bSJens Axboe src_iter = src->bi_iter; 1247f9c78b2bSJens Axboe } 1248f9c78b2bSJens Axboe 1249f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1250f9c78b2bSJens Axboe dst = dst->bi_next; 1251f9c78b2bSJens Axboe if (!dst) 1252f9c78b2bSJens Axboe break; 1253f9c78b2bSJens Axboe 1254f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1255f9c78b2bSJens Axboe } 1256f9c78b2bSJens Axboe 125745db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1258f9c78b2bSJens Axboe } 1259f9c78b2bSJens Axboe } 126045db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1261f9c78b2bSJens Axboe 1262491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 12631dfa0f68SChristoph Hellwig { 12641dfa0f68SChristoph Hellwig struct bio_vec *bvec; 12656dc4f100SMing Lei struct bvec_iter_all iter_all; 12661dfa0f68SChristoph Hellwig 12672b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 12681dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 12691dfa0f68SChristoph Hellwig } 1270491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 12711dfa0f68SChristoph Hellwig 1272f9c78b2bSJens Axboe /* 1273f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1274f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1275f9c78b2bSJens Axboe * 1276f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1277f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1278f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1279f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1280f9c78b2bSJens Axboe * in process context. 1281f9c78b2bSJens Axboe * 1282f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1283f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1284f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1285f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1286f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1287f9c78b2bSJens Axboe * 1288f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1289f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1290f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1291f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1292f9c78b2bSJens Axboe * pagecache. 1293f9c78b2bSJens Axboe * 1294f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1295f9c78b2bSJens Axboe * deferred bio dirtying paths. 1296f9c78b2bSJens Axboe */ 1297f9c78b2bSJens Axboe 1298f9c78b2bSJens Axboe /* 1299f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1300f9c78b2bSJens Axboe */ 1301f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1302f9c78b2bSJens Axboe { 1303f9c78b2bSJens Axboe struct bio_vec *bvec; 13046dc4f100SMing Lei struct bvec_iter_all iter_all; 1305f9c78b2bSJens Axboe 13062b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 13073bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 13083bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1309f9c78b2bSJens Axboe } 1310f9c78b2bSJens Axboe } 1311f9c78b2bSJens Axboe 1312f9c78b2bSJens Axboe /* 1313f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1314f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1315f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 131624d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1317f9c78b2bSJens Axboe * 1318f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1319ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1320ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1321f9c78b2bSJens Axboe */ 1322f9c78b2bSJens Axboe 1323f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1324f9c78b2bSJens Axboe 1325f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1326f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1327f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1328f9c78b2bSJens Axboe 1329f9c78b2bSJens Axboe /* 1330f9c78b2bSJens Axboe * This runs in process context 1331f9c78b2bSJens Axboe */ 1332f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1333f9c78b2bSJens Axboe { 133424d5493fSChristoph Hellwig struct bio *bio, *next; 1335f9c78b2bSJens Axboe 133624d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 133724d5493fSChristoph Hellwig next = bio_dirty_list; 1338f9c78b2bSJens Axboe bio_dirty_list = NULL; 133924d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1340f9c78b2bSJens Axboe 134124d5493fSChristoph Hellwig while ((bio = next) != NULL) { 134224d5493fSChristoph Hellwig next = bio->bi_private; 1343f9c78b2bSJens Axboe 1344d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1345f9c78b2bSJens Axboe bio_put(bio); 1346f9c78b2bSJens Axboe } 1347f9c78b2bSJens Axboe } 1348f9c78b2bSJens Axboe 1349f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1350f9c78b2bSJens Axboe { 1351f9c78b2bSJens Axboe struct bio_vec *bvec; 135224d5493fSChristoph Hellwig unsigned long flags; 13536dc4f100SMing Lei struct bvec_iter_all iter_all; 1354f9c78b2bSJens Axboe 13552b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 135624d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 135724d5493fSChristoph Hellwig goto defer; 1358f9c78b2bSJens Axboe } 1359f9c78b2bSJens Axboe 1360d241a95fSChristoph Hellwig bio_release_pages(bio, false); 136124d5493fSChristoph Hellwig bio_put(bio); 136224d5493fSChristoph Hellwig return; 136324d5493fSChristoph Hellwig defer: 1364f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1365f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1366f9c78b2bSJens Axboe bio_dirty_list = bio; 1367f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1368f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1369f9c78b2bSJens Axboe } 1370f9c78b2bSJens Axboe 1371c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1372c4cf5261SJens Axboe { 1373c4cf5261SJens Axboe /* 1374c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1375c4cf5261SJens Axboe * we always end io on the first invocation. 1376c4cf5261SJens Axboe */ 1377c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1378c4cf5261SJens Axboe return true; 1379c4cf5261SJens Axboe 1380c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1381c4cf5261SJens Axboe 1382326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1383b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1384c4cf5261SJens Axboe return true; 1385326e1dbbSMike Snitzer } 1386c4cf5261SJens Axboe 1387c4cf5261SJens Axboe return false; 1388c4cf5261SJens Axboe } 1389c4cf5261SJens Axboe 1390f9c78b2bSJens Axboe /** 1391f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1392f9c78b2bSJens Axboe * @bio: bio 1393f9c78b2bSJens Axboe * 1394f9c78b2bSJens Axboe * Description: 13954246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 13964246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 13974246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1398fbbaf700SNeilBrown * 1399fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1400fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1401fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1402fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1403f9c78b2bSJens Axboe **/ 14044246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1405f9c78b2bSJens Axboe { 1406ba8c6967SChristoph Hellwig again: 14072b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1408ba8c6967SChristoph Hellwig return; 14097c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 14107c20f116SChristoph Hellwig return; 1411f9c78b2bSJens Axboe 1412309dca30SChristoph Hellwig if (bio->bi_bdev) 1413309dca30SChristoph Hellwig rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio); 141467b42d0bSJosef Bacik 1415f9c78b2bSJens Axboe /* 1416ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1417ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1418ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1419ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1420ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1421ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1422f9c78b2bSJens Axboe */ 1423f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 142438f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1425ba8c6967SChristoph Hellwig goto again; 1426ba8c6967SChristoph Hellwig } 1427ba8c6967SChristoph Hellwig 1428309dca30SChristoph Hellwig if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 1429309dca30SChristoph Hellwig trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio); 1430fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1431fbbaf700SNeilBrown } 1432fbbaf700SNeilBrown 14339e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1434b222dd2fSShaohua Li /* release cgroup info */ 1435b222dd2fSShaohua Li bio_uninit(bio); 1436f9c78b2bSJens Axboe if (bio->bi_end_io) 14374246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1438f9c78b2bSJens Axboe } 1439f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1440f9c78b2bSJens Axboe 1441f9c78b2bSJens Axboe /** 1442f9c78b2bSJens Axboe * bio_split - split a bio 1443f9c78b2bSJens Axboe * @bio: bio to split 1444f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1445f9c78b2bSJens Axboe * @gfp: gfp mask 1446f9c78b2bSJens Axboe * @bs: bio set to allocate from 1447f9c78b2bSJens Axboe * 1448f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1449f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1450f9c78b2bSJens Axboe * 1451f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1452dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1453dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1454f9c78b2bSJens Axboe */ 1455f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1456f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1457f9c78b2bSJens Axboe { 1458f341a4d3SMikulas Patocka struct bio *split; 1459f9c78b2bSJens Axboe 1460f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1461f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1462f9c78b2bSJens Axboe 14630512a75bSKeith Busch /* Zone append commands cannot be split */ 14640512a75bSKeith Busch if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND)) 14650512a75bSKeith Busch return NULL; 14660512a75bSKeith Busch 1467f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1468f9c78b2bSJens Axboe if (!split) 1469f9c78b2bSJens Axboe return NULL; 1470f9c78b2bSJens Axboe 1471f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1472f9c78b2bSJens Axboe 1473f9c78b2bSJens Axboe if (bio_integrity(split)) 1474fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1475f9c78b2bSJens Axboe 1476f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1477f9c78b2bSJens Axboe 1478fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 147920d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1480fbbaf700SNeilBrown 1481f9c78b2bSJens Axboe return split; 1482f9c78b2bSJens Axboe } 1483f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1484f9c78b2bSJens Axboe 1485f9c78b2bSJens Axboe /** 1486f9c78b2bSJens Axboe * bio_trim - trim a bio 1487f9c78b2bSJens Axboe * @bio: bio to trim 1488f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1489f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1490f9c78b2bSJens Axboe */ 1491f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1492f9c78b2bSJens Axboe { 1493f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1494f9c78b2bSJens Axboe * the given offset and size. 1495f9c78b2bSJens Axboe */ 1496f9c78b2bSJens Axboe 1497f9c78b2bSJens Axboe size <<= 9; 1498f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1499f9c78b2bSJens Axboe return; 1500f9c78b2bSJens Axboe 1501f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1502f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1503376a78abSDmitry Monakhov 1504376a78abSDmitry Monakhov if (bio_integrity(bio)) 1505fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1506376a78abSDmitry Monakhov 1507f9c78b2bSJens Axboe } 1508f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1509f9c78b2bSJens Axboe 1510f9c78b2bSJens Axboe /* 1511f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1512f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1513f9c78b2bSJens Axboe */ 15148aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1515f9c78b2bSJens Axboe { 1516ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1517f9c78b2bSJens Axboe 15188aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1519f9c78b2bSJens Axboe } 1520f9c78b2bSJens Axboe 1521917a38c7SKent Overstreet /* 1522917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1523917a38c7SKent Overstreet * 1524917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1525917a38c7SKent Overstreet * kzalloc()). 1526917a38c7SKent Overstreet */ 1527917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1528f9c78b2bSJens Axboe { 1529f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1530f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1531917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1532f9c78b2bSJens Axboe 15338aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 15348aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1535f9c78b2bSJens Axboe 1536f9c78b2bSJens Axboe bioset_integrity_free(bs); 1537917a38c7SKent Overstreet if (bs->bio_slab) 1538f9c78b2bSJens Axboe bio_put_slab(bs); 1539917a38c7SKent Overstreet bs->bio_slab = NULL; 1540917a38c7SKent Overstreet } 1541917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1542f9c78b2bSJens Axboe 1543011067b0SNeilBrown /** 1544917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1545dad08527SKent Overstreet * @bs: pool to initialize 1546917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1547917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1548917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1549917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1550917a38c7SKent Overstreet * 1551dad08527SKent Overstreet * Description: 1552dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1553dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1554dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1555dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1556dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1557dad08527SKent Overstreet * or things will break badly. 1558dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1559dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1560dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1561dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1562dad08527SKent Overstreet * 1563917a38c7SKent Overstreet */ 1564917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1565917a38c7SKent Overstreet unsigned int pool_size, 1566917a38c7SKent Overstreet unsigned int front_pad, 1567917a38c7SKent Overstreet int flags) 1568917a38c7SKent Overstreet { 1569917a38c7SKent Overstreet bs->front_pad = front_pad; 15709f180e31SMing Lei if (flags & BIOSET_NEED_BVECS) 15719f180e31SMing Lei bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 15729f180e31SMing Lei else 15739f180e31SMing Lei bs->back_pad = 0; 1574917a38c7SKent Overstreet 1575917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1576917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1577917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1578917a38c7SKent Overstreet 157949d1ec85SMing Lei bs->bio_slab = bio_find_or_create_slab(bs); 1580917a38c7SKent Overstreet if (!bs->bio_slab) 1581917a38c7SKent Overstreet return -ENOMEM; 1582917a38c7SKent Overstreet 1583917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1584917a38c7SKent Overstreet goto bad; 1585917a38c7SKent Overstreet 1586917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1587917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1588917a38c7SKent Overstreet goto bad; 1589917a38c7SKent Overstreet 1590917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 1591917a38c7SKent Overstreet return 0; 1592917a38c7SKent Overstreet 1593917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1594917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1595917a38c7SKent Overstreet goto bad; 1596917a38c7SKent Overstreet 1597917a38c7SKent Overstreet return 0; 1598917a38c7SKent Overstreet bad: 1599917a38c7SKent Overstreet bioset_exit(bs); 1600917a38c7SKent Overstreet return -ENOMEM; 1601917a38c7SKent Overstreet } 1602917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1603917a38c7SKent Overstreet 160428e89fd9SJens Axboe /* 160528e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 160628e89fd9SJens Axboe * another bio_set. 160728e89fd9SJens Axboe */ 160828e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 160928e89fd9SJens Axboe { 161028e89fd9SJens Axboe int flags; 161128e89fd9SJens Axboe 161228e89fd9SJens Axboe flags = 0; 161328e89fd9SJens Axboe if (src->bvec_pool.min_nr) 161428e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 161528e89fd9SJens Axboe if (src->rescue_workqueue) 161628e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 161728e89fd9SJens Axboe 161828e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 161928e89fd9SJens Axboe } 162028e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 162128e89fd9SJens Axboe 1622f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 1623f9c78b2bSJens Axboe { 1624f9c78b2bSJens Axboe int i; 1625f9c78b2bSJens Axboe 1626ed996a52SChristoph Hellwig for (i = 0; i < BVEC_POOL_NR; i++) { 1627f9c78b2bSJens Axboe int size; 1628f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 1629f9c78b2bSJens Axboe 1630f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 1631f9c78b2bSJens Axboe bvs->slab = NULL; 1632f9c78b2bSJens Axboe continue; 1633f9c78b2bSJens Axboe } 1634f9c78b2bSJens Axboe 1635f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 1636f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 1637f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 1638f9c78b2bSJens Axboe } 1639f9c78b2bSJens Axboe } 1640f9c78b2bSJens Axboe 1641f9c78b2bSJens Axboe static int __init init_bio(void) 1642f9c78b2bSJens Axboe { 16432b24e6f6SJohannes Thumshirn BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 16442b24e6f6SJohannes Thumshirn 1645f9c78b2bSJens Axboe bio_integrity_init(); 1646f9c78b2bSJens Axboe biovec_init_slabs(); 1647f9c78b2bSJens Axboe 1648f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 1649f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 1650f9c78b2bSJens Axboe 1651f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 1652f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 1653f9c78b2bSJens Axboe 1654f9c78b2bSJens Axboe return 0; 1655f9c78b2bSJens Axboe } 1656f9c78b2bSJens Axboe subsys_initcall(init_bio); 1657