1f9c78b2bSJens Axboe /* 2f9c78b2bSJens Axboe * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 3f9c78b2bSJens Axboe * 4f9c78b2bSJens Axboe * This program is free software; you can redistribute it and/or modify 5f9c78b2bSJens Axboe * it under the terms of the GNU General Public License version 2 as 6f9c78b2bSJens Axboe * published by the Free Software Foundation. 7f9c78b2bSJens Axboe * 8f9c78b2bSJens Axboe * This program is distributed in the hope that it will be useful, 9f9c78b2bSJens Axboe * but WITHOUT ANY WARRANTY; without even the implied warranty of 10f9c78b2bSJens Axboe * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 11f9c78b2bSJens Axboe * GNU General Public License for more details. 12f9c78b2bSJens Axboe * 13f9c78b2bSJens Axboe * You should have received a copy of the GNU General Public Licens 14f9c78b2bSJens Axboe * along with this program; if not, write to the Free Software 15f9c78b2bSJens Axboe * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111- 16f9c78b2bSJens Axboe * 17f9c78b2bSJens Axboe */ 18f9c78b2bSJens Axboe #include <linux/mm.h> 19f9c78b2bSJens Axboe #include <linux/swap.h> 20f9c78b2bSJens Axboe #include <linux/bio.h> 21f9c78b2bSJens Axboe #include <linux/blkdev.h> 22f9c78b2bSJens Axboe #include <linux/uio.h> 23f9c78b2bSJens Axboe #include <linux/iocontext.h> 24f9c78b2bSJens Axboe #include <linux/slab.h> 25f9c78b2bSJens Axboe #include <linux/init.h> 26f9c78b2bSJens Axboe #include <linux/kernel.h> 27f9c78b2bSJens Axboe #include <linux/export.h> 28f9c78b2bSJens Axboe #include <linux/mempool.h> 29f9c78b2bSJens Axboe #include <linux/workqueue.h> 30f9c78b2bSJens Axboe #include <linux/cgroup.h> 3108e18eabSJosef Bacik #include <linux/blk-cgroup.h> 32f9c78b2bSJens Axboe 33f9c78b2bSJens Axboe #include <trace/events/block.h> 349e234eeaSShaohua Li #include "blk.h" 3567b42d0bSJosef Bacik #include "blk-rq-qos.h" 36f9c78b2bSJens Axboe 37f9c78b2bSJens Axboe /* 38f9c78b2bSJens Axboe * Test patch to inline a certain number of bi_io_vec's inside the bio 39f9c78b2bSJens Axboe * itself, to shrink a bio data allocation from two mempool calls to one 40f9c78b2bSJens Axboe */ 41f9c78b2bSJens Axboe #define BIO_INLINE_VECS 4 42f9c78b2bSJens Axboe 43f9c78b2bSJens Axboe /* 44f9c78b2bSJens Axboe * if you change this list, also change bvec_alloc or things will 45f9c78b2bSJens Axboe * break badly! cannot be bigger than what you can fit into an 46f9c78b2bSJens Axboe * unsigned short 47f9c78b2bSJens Axboe */ 48bd5c4facSMikulas Patocka #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 49ed996a52SChristoph Hellwig static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 50bd5c4facSMikulas Patocka BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 51f9c78b2bSJens Axboe }; 52f9c78b2bSJens Axboe #undef BV 53f9c78b2bSJens Axboe 54f9c78b2bSJens Axboe /* 55f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 56f9c78b2bSJens Axboe * IO code that does not need private memory pools. 57f9c78b2bSJens Axboe */ 58f4f8154aSKent Overstreet struct bio_set fs_bio_set; 59f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 60f9c78b2bSJens Axboe 61f9c78b2bSJens Axboe /* 62f9c78b2bSJens Axboe * Our slab pool management 63f9c78b2bSJens Axboe */ 64f9c78b2bSJens Axboe struct bio_slab { 65f9c78b2bSJens Axboe struct kmem_cache *slab; 66f9c78b2bSJens Axboe unsigned int slab_ref; 67f9c78b2bSJens Axboe unsigned int slab_size; 68f9c78b2bSJens Axboe char name[8]; 69f9c78b2bSJens Axboe }; 70f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 71f9c78b2bSJens Axboe static struct bio_slab *bio_slabs; 72f9c78b2bSJens Axboe static unsigned int bio_slab_nr, bio_slab_max; 73f9c78b2bSJens Axboe 74f9c78b2bSJens Axboe static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 75f9c78b2bSJens Axboe { 76f9c78b2bSJens Axboe unsigned int sz = sizeof(struct bio) + extra_size; 77f9c78b2bSJens Axboe struct kmem_cache *slab = NULL; 78f9c78b2bSJens Axboe struct bio_slab *bslab, *new_bio_slabs; 79f9c78b2bSJens Axboe unsigned int new_bio_slab_max; 80f9c78b2bSJens Axboe unsigned int i, entry = -1; 81f9c78b2bSJens Axboe 82f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 83f9c78b2bSJens Axboe 84f9c78b2bSJens Axboe i = 0; 85f9c78b2bSJens Axboe while (i < bio_slab_nr) { 86f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 87f9c78b2bSJens Axboe 88f9c78b2bSJens Axboe if (!bslab->slab && entry == -1) 89f9c78b2bSJens Axboe entry = i; 90f9c78b2bSJens Axboe else if (bslab->slab_size == sz) { 91f9c78b2bSJens Axboe slab = bslab->slab; 92f9c78b2bSJens Axboe bslab->slab_ref++; 93f9c78b2bSJens Axboe break; 94f9c78b2bSJens Axboe } 95f9c78b2bSJens Axboe i++; 96f9c78b2bSJens Axboe } 97f9c78b2bSJens Axboe 98f9c78b2bSJens Axboe if (slab) 99f9c78b2bSJens Axboe goto out_unlock; 100f9c78b2bSJens Axboe 101f9c78b2bSJens Axboe if (bio_slab_nr == bio_slab_max && entry == -1) { 102f9c78b2bSJens Axboe new_bio_slab_max = bio_slab_max << 1; 103f9c78b2bSJens Axboe new_bio_slabs = krealloc(bio_slabs, 104f9c78b2bSJens Axboe new_bio_slab_max * sizeof(struct bio_slab), 105f9c78b2bSJens Axboe GFP_KERNEL); 106f9c78b2bSJens Axboe if (!new_bio_slabs) 107f9c78b2bSJens Axboe goto out_unlock; 108f9c78b2bSJens Axboe bio_slab_max = new_bio_slab_max; 109f9c78b2bSJens Axboe bio_slabs = new_bio_slabs; 110f9c78b2bSJens Axboe } 111f9c78b2bSJens Axboe if (entry == -1) 112f9c78b2bSJens Axboe entry = bio_slab_nr++; 113f9c78b2bSJens Axboe 114f9c78b2bSJens Axboe bslab = &bio_slabs[entry]; 115f9c78b2bSJens Axboe 116f9c78b2bSJens Axboe snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 1176a241483SMikulas Patocka slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 1186a241483SMikulas Patocka SLAB_HWCACHE_ALIGN, NULL); 119f9c78b2bSJens Axboe if (!slab) 120f9c78b2bSJens Axboe goto out_unlock; 121f9c78b2bSJens Axboe 122f9c78b2bSJens Axboe bslab->slab = slab; 123f9c78b2bSJens Axboe bslab->slab_ref = 1; 124f9c78b2bSJens Axboe bslab->slab_size = sz; 125f9c78b2bSJens Axboe out_unlock: 126f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 127f9c78b2bSJens Axboe return slab; 128f9c78b2bSJens Axboe } 129f9c78b2bSJens Axboe 130f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 131f9c78b2bSJens Axboe { 132f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 133f9c78b2bSJens Axboe unsigned int i; 134f9c78b2bSJens Axboe 135f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 136f9c78b2bSJens Axboe 137f9c78b2bSJens Axboe for (i = 0; i < bio_slab_nr; i++) { 138f9c78b2bSJens Axboe if (bs->bio_slab == bio_slabs[i].slab) { 139f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 140f9c78b2bSJens Axboe break; 141f9c78b2bSJens Axboe } 142f9c78b2bSJens Axboe } 143f9c78b2bSJens Axboe 144f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 145f9c78b2bSJens Axboe goto out; 146f9c78b2bSJens Axboe 147f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 148f9c78b2bSJens Axboe 149f9c78b2bSJens Axboe if (--bslab->slab_ref) 150f9c78b2bSJens Axboe goto out; 151f9c78b2bSJens Axboe 152f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 153f9c78b2bSJens Axboe bslab->slab = NULL; 154f9c78b2bSJens Axboe 155f9c78b2bSJens Axboe out: 156f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 157f9c78b2bSJens Axboe } 158f9c78b2bSJens Axboe 159f9c78b2bSJens Axboe unsigned int bvec_nr_vecs(unsigned short idx) 160f9c78b2bSJens Axboe { 161d6c02a9bSGreg Edwards return bvec_slabs[--idx].nr_vecs; 162f9c78b2bSJens Axboe } 163f9c78b2bSJens Axboe 164f9c78b2bSJens Axboe void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 165f9c78b2bSJens Axboe { 166ed996a52SChristoph Hellwig if (!idx) 167ed996a52SChristoph Hellwig return; 168ed996a52SChristoph Hellwig idx--; 169f9c78b2bSJens Axboe 170ed996a52SChristoph Hellwig BIO_BUG_ON(idx >= BVEC_POOL_NR); 171ed996a52SChristoph Hellwig 172ed996a52SChristoph Hellwig if (idx == BVEC_POOL_MAX) { 173f9c78b2bSJens Axboe mempool_free(bv, pool); 174ed996a52SChristoph Hellwig } else { 175f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + idx; 176f9c78b2bSJens Axboe 177f9c78b2bSJens Axboe kmem_cache_free(bvs->slab, bv); 178f9c78b2bSJens Axboe } 179f9c78b2bSJens Axboe } 180f9c78b2bSJens Axboe 181f9c78b2bSJens Axboe struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 182f9c78b2bSJens Axboe mempool_t *pool) 183f9c78b2bSJens Axboe { 184f9c78b2bSJens Axboe struct bio_vec *bvl; 185f9c78b2bSJens Axboe 186f9c78b2bSJens Axboe /* 187f9c78b2bSJens Axboe * see comment near bvec_array define! 188f9c78b2bSJens Axboe */ 189f9c78b2bSJens Axboe switch (nr) { 190f9c78b2bSJens Axboe case 1: 191f9c78b2bSJens Axboe *idx = 0; 192f9c78b2bSJens Axboe break; 193f9c78b2bSJens Axboe case 2 ... 4: 194f9c78b2bSJens Axboe *idx = 1; 195f9c78b2bSJens Axboe break; 196f9c78b2bSJens Axboe case 5 ... 16: 197f9c78b2bSJens Axboe *idx = 2; 198f9c78b2bSJens Axboe break; 199f9c78b2bSJens Axboe case 17 ... 64: 200f9c78b2bSJens Axboe *idx = 3; 201f9c78b2bSJens Axboe break; 202f9c78b2bSJens Axboe case 65 ... 128: 203f9c78b2bSJens Axboe *idx = 4; 204f9c78b2bSJens Axboe break; 205f9c78b2bSJens Axboe case 129 ... BIO_MAX_PAGES: 206f9c78b2bSJens Axboe *idx = 5; 207f9c78b2bSJens Axboe break; 208f9c78b2bSJens Axboe default: 209f9c78b2bSJens Axboe return NULL; 210f9c78b2bSJens Axboe } 211f9c78b2bSJens Axboe 212f9c78b2bSJens Axboe /* 213f9c78b2bSJens Axboe * idx now points to the pool we want to allocate from. only the 214f9c78b2bSJens Axboe * 1-vec entry pool is mempool backed. 215f9c78b2bSJens Axboe */ 216ed996a52SChristoph Hellwig if (*idx == BVEC_POOL_MAX) { 217f9c78b2bSJens Axboe fallback: 218f9c78b2bSJens Axboe bvl = mempool_alloc(pool, gfp_mask); 219f9c78b2bSJens Axboe } else { 220f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + *idx; 221d0164adcSMel Gorman gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 222f9c78b2bSJens Axboe 223f9c78b2bSJens Axboe /* 224f9c78b2bSJens Axboe * Make this allocation restricted and don't dump info on 225f9c78b2bSJens Axboe * allocation failures, since we'll fallback to the mempool 226f9c78b2bSJens Axboe * in case of failure. 227f9c78b2bSJens Axboe */ 228f9c78b2bSJens Axboe __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 229f9c78b2bSJens Axboe 230f9c78b2bSJens Axboe /* 231d0164adcSMel Gorman * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 232f9c78b2bSJens Axboe * is set, retry with the 1-entry mempool 233f9c78b2bSJens Axboe */ 234f9c78b2bSJens Axboe bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 235d0164adcSMel Gorman if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 236ed996a52SChristoph Hellwig *idx = BVEC_POOL_MAX; 237f9c78b2bSJens Axboe goto fallback; 238f9c78b2bSJens Axboe } 239f9c78b2bSJens Axboe } 240f9c78b2bSJens Axboe 241ed996a52SChristoph Hellwig (*idx)++; 242f9c78b2bSJens Axboe return bvl; 243f9c78b2bSJens Axboe } 244f9c78b2bSJens Axboe 2459ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 246f9c78b2bSJens Axboe { 2476f70fb66SDennis Zhou bio_disassociate_blkg(bio); 248f9c78b2bSJens Axboe } 2499ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 250f9c78b2bSJens Axboe 251f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 252f9c78b2bSJens Axboe { 253f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 254f9c78b2bSJens Axboe void *p; 255f9c78b2bSJens Axboe 2569ae3b3f5SJens Axboe bio_uninit(bio); 257f9c78b2bSJens Axboe 258f9c78b2bSJens Axboe if (bs) { 2598aa6ba2fSKent Overstreet bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 260f9c78b2bSJens Axboe 261f9c78b2bSJens Axboe /* 262f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 263f9c78b2bSJens Axboe */ 264f9c78b2bSJens Axboe p = bio; 265f9c78b2bSJens Axboe p -= bs->front_pad; 266f9c78b2bSJens Axboe 2678aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 268f9c78b2bSJens Axboe } else { 269f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 270f9c78b2bSJens Axboe kfree(bio); 271f9c78b2bSJens Axboe } 272f9c78b2bSJens Axboe } 273f9c78b2bSJens Axboe 2749ae3b3f5SJens Axboe /* 2759ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2769ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2779ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2789ae3b3f5SJens Axboe */ 2793a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2803a83f467SMing Lei unsigned short max_vecs) 281f9c78b2bSJens Axboe { 282f9c78b2bSJens Axboe memset(bio, 0, sizeof(*bio)); 283c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 284dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2853a83f467SMing Lei 2863a83f467SMing Lei bio->bi_io_vec = table; 2873a83f467SMing Lei bio->bi_max_vecs = max_vecs; 288f9c78b2bSJens Axboe } 289f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 290f9c78b2bSJens Axboe 291f9c78b2bSJens Axboe /** 292f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 293f9c78b2bSJens Axboe * @bio: bio to reset 294f9c78b2bSJens Axboe * 295f9c78b2bSJens Axboe * Description: 296f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 297f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 298f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 299f9c78b2bSJens Axboe * comment in struct bio. 300f9c78b2bSJens Axboe */ 301f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 302f9c78b2bSJens Axboe { 303f9c78b2bSJens Axboe unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 304f9c78b2bSJens Axboe 3059ae3b3f5SJens Axboe bio_uninit(bio); 306f9c78b2bSJens Axboe 307f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 3084246a0b6SChristoph Hellwig bio->bi_flags = flags; 309c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 310f9c78b2bSJens Axboe } 311f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 312f9c78b2bSJens Axboe 31338f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 314f9c78b2bSJens Axboe { 3154246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3164246a0b6SChristoph Hellwig 3174e4cbee9SChristoph Hellwig if (!parent->bi_status) 3184e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 319f9c78b2bSJens Axboe bio_put(bio); 32038f8baaeSChristoph Hellwig return parent; 32138f8baaeSChristoph Hellwig } 32238f8baaeSChristoph Hellwig 32338f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 32438f8baaeSChristoph Hellwig { 32538f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 326f9c78b2bSJens Axboe } 327f9c78b2bSJens Axboe 328f9c78b2bSJens Axboe /** 329f9c78b2bSJens Axboe * bio_chain - chain bio completions 330f9c78b2bSJens Axboe * @bio: the target bio 331f9c78b2bSJens Axboe * @parent: the @bio's parent bio 332f9c78b2bSJens Axboe * 333f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 334f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 335f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 336f9c78b2bSJens Axboe * 337f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 338f9c78b2bSJens Axboe */ 339f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 340f9c78b2bSJens Axboe { 341f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 342f9c78b2bSJens Axboe 343f9c78b2bSJens Axboe bio->bi_private = parent; 344f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 345c4cf5261SJens Axboe bio_inc_remaining(parent); 346f9c78b2bSJens Axboe } 347f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 348f9c78b2bSJens Axboe 349f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 350f9c78b2bSJens Axboe { 351f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 352f9c78b2bSJens Axboe struct bio *bio; 353f9c78b2bSJens Axboe 354f9c78b2bSJens Axboe while (1) { 355f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 356f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 357f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 358f9c78b2bSJens Axboe 359f9c78b2bSJens Axboe if (!bio) 360f9c78b2bSJens Axboe break; 361f9c78b2bSJens Axboe 362f9c78b2bSJens Axboe generic_make_request(bio); 363f9c78b2bSJens Axboe } 364f9c78b2bSJens Axboe } 365f9c78b2bSJens Axboe 366f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 367f9c78b2bSJens Axboe { 368f9c78b2bSJens Axboe struct bio_list punt, nopunt; 369f9c78b2bSJens Axboe struct bio *bio; 370f9c78b2bSJens Axboe 37147e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 37247e0fb46SNeilBrown return; 373f9c78b2bSJens Axboe /* 374f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 375f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 376f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 377f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 378f9c78b2bSJens Axboe * our own rescuer would be bad. 379f9c78b2bSJens Axboe * 380f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 381f9c78b2bSJens Axboe * remove from the middle of the list: 382f9c78b2bSJens Axboe */ 383f9c78b2bSJens Axboe 384f9c78b2bSJens Axboe bio_list_init(&punt); 385f9c78b2bSJens Axboe bio_list_init(&nopunt); 386f9c78b2bSJens Axboe 387f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 388f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 389f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 390f9c78b2bSJens Axboe 391f5fe1b51SNeilBrown bio_list_init(&nopunt); 392f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 393f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 394f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 395f9c78b2bSJens Axboe 396f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 397f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 398f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 399f9c78b2bSJens Axboe 400f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 401f9c78b2bSJens Axboe } 402f9c78b2bSJens Axboe 403f9c78b2bSJens Axboe /** 404f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 405519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 406f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 407f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 408f9c78b2bSJens Axboe * 409f9c78b2bSJens Axboe * Description: 410f9c78b2bSJens Axboe * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 411f9c78b2bSJens Axboe * backed by the @bs's mempool. 412f9c78b2bSJens Axboe * 413d0164adcSMel Gorman * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 414d0164adcSMel Gorman * always be able to allocate a bio. This is due to the mempool guarantees. 415d0164adcSMel Gorman * To make this work, callers must never allocate more than 1 bio at a time 416d0164adcSMel Gorman * from this pool. Callers that need to allocate more than 1 bio must always 417d0164adcSMel Gorman * submit the previously allocated bio for IO before attempting to allocate 418d0164adcSMel Gorman * a new one. Failure to do so can cause deadlocks under memory pressure. 419f9c78b2bSJens Axboe * 420f9c78b2bSJens Axboe * Note that when running under generic_make_request() (i.e. any block 421f9c78b2bSJens Axboe * driver), bios are not submitted until after you return - see the code in 422f9c78b2bSJens Axboe * generic_make_request() that converts recursion into iteration, to prevent 423f9c78b2bSJens Axboe * stack overflows. 424f9c78b2bSJens Axboe * 425f9c78b2bSJens Axboe * This would normally mean allocating multiple bios under 426f9c78b2bSJens Axboe * generic_make_request() would be susceptible to deadlocks, but we have 427f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 428f9c78b2bSJens Axboe * thread. 429f9c78b2bSJens Axboe * 430f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 431f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 432f9c78b2bSJens Axboe * generic_make_request() should be avoided - instead, use bio_set's front_pad 433f9c78b2bSJens Axboe * for per bio allocations. 434f9c78b2bSJens Axboe * 435f9c78b2bSJens Axboe * RETURNS: 436f9c78b2bSJens Axboe * Pointer to new bio on success, NULL on failure. 437f9c78b2bSJens Axboe */ 4387a88fa19SDan Carpenter struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 4397a88fa19SDan Carpenter struct bio_set *bs) 440f9c78b2bSJens Axboe { 441f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 442f9c78b2bSJens Axboe unsigned front_pad; 443f9c78b2bSJens Axboe unsigned inline_vecs; 444f9c78b2bSJens Axboe struct bio_vec *bvl = NULL; 445f9c78b2bSJens Axboe struct bio *bio; 446f9c78b2bSJens Axboe void *p; 447f9c78b2bSJens Axboe 448f9c78b2bSJens Axboe if (!bs) { 449f9c78b2bSJens Axboe if (nr_iovecs > UIO_MAXIOV) 450f9c78b2bSJens Axboe return NULL; 451f9c78b2bSJens Axboe 452f9c78b2bSJens Axboe p = kmalloc(sizeof(struct bio) + 453f9c78b2bSJens Axboe nr_iovecs * sizeof(struct bio_vec), 454f9c78b2bSJens Axboe gfp_mask); 455f9c78b2bSJens Axboe front_pad = 0; 456f9c78b2bSJens Axboe inline_vecs = nr_iovecs; 457f9c78b2bSJens Axboe } else { 458d8f429e1SJunichi Nomura /* should not use nobvec bioset for nr_iovecs > 0 */ 4598aa6ba2fSKent Overstreet if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 4608aa6ba2fSKent Overstreet nr_iovecs > 0)) 461d8f429e1SJunichi Nomura return NULL; 462f9c78b2bSJens Axboe /* 463f9c78b2bSJens Axboe * generic_make_request() converts recursion to iteration; this 464f9c78b2bSJens Axboe * means if we're running beneath it, any bios we allocate and 465f9c78b2bSJens Axboe * submit will not be submitted (and thus freed) until after we 466f9c78b2bSJens Axboe * return. 467f9c78b2bSJens Axboe * 468f9c78b2bSJens Axboe * This exposes us to a potential deadlock if we allocate 469f9c78b2bSJens Axboe * multiple bios from the same bio_set() while running 470f9c78b2bSJens Axboe * underneath generic_make_request(). If we were to allocate 471f9c78b2bSJens Axboe * multiple bios (say a stacking block driver that was splitting 472f9c78b2bSJens Axboe * bios), we would deadlock if we exhausted the mempool's 473f9c78b2bSJens Axboe * reserve. 474f9c78b2bSJens Axboe * 475f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 476f9c78b2bSJens Axboe * workqueue per bio_set. If we go to allocate and there are 477f9c78b2bSJens Axboe * bios on current->bio_list, we first try the allocation 478d0164adcSMel Gorman * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 479d0164adcSMel Gorman * bios we would be blocking to the rescuer workqueue before 480d0164adcSMel Gorman * we retry with the original gfp_flags. 481f9c78b2bSJens Axboe */ 482f9c78b2bSJens Axboe 483f5fe1b51SNeilBrown if (current->bio_list && 484f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 48547e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 48647e0fb46SNeilBrown bs->rescue_workqueue) 487d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 488f9c78b2bSJens Axboe 4898aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 490f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 491f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 492f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4938aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 494f9c78b2bSJens Axboe } 495f9c78b2bSJens Axboe 496f9c78b2bSJens Axboe front_pad = bs->front_pad; 497f9c78b2bSJens Axboe inline_vecs = BIO_INLINE_VECS; 498f9c78b2bSJens Axboe } 499f9c78b2bSJens Axboe 500f9c78b2bSJens Axboe if (unlikely(!p)) 501f9c78b2bSJens Axboe return NULL; 502f9c78b2bSJens Axboe 503f9c78b2bSJens Axboe bio = p + front_pad; 5043a83f467SMing Lei bio_init(bio, NULL, 0); 505f9c78b2bSJens Axboe 506f9c78b2bSJens Axboe if (nr_iovecs > inline_vecs) { 507ed996a52SChristoph Hellwig unsigned long idx = 0; 508ed996a52SChristoph Hellwig 5098aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 510f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 511f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 512f9c78b2bSJens Axboe gfp_mask = saved_gfp; 5138aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 514f9c78b2bSJens Axboe } 515f9c78b2bSJens Axboe 516f9c78b2bSJens Axboe if (unlikely(!bvl)) 517f9c78b2bSJens Axboe goto err_free; 518f9c78b2bSJens Axboe 519ed996a52SChristoph Hellwig bio->bi_flags |= idx << BVEC_POOL_OFFSET; 520f9c78b2bSJens Axboe } else if (nr_iovecs) { 521f9c78b2bSJens Axboe bvl = bio->bi_inline_vecs; 522f9c78b2bSJens Axboe } 523f9c78b2bSJens Axboe 524f9c78b2bSJens Axboe bio->bi_pool = bs; 525f9c78b2bSJens Axboe bio->bi_max_vecs = nr_iovecs; 526f9c78b2bSJens Axboe bio->bi_io_vec = bvl; 527f9c78b2bSJens Axboe return bio; 528f9c78b2bSJens Axboe 529f9c78b2bSJens Axboe err_free: 5308aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 531f9c78b2bSJens Axboe return NULL; 532f9c78b2bSJens Axboe } 533f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 534f9c78b2bSJens Axboe 53538a72dacSKent Overstreet void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 536f9c78b2bSJens Axboe { 537f9c78b2bSJens Axboe unsigned long flags; 538f9c78b2bSJens Axboe struct bio_vec bv; 539f9c78b2bSJens Axboe struct bvec_iter iter; 540f9c78b2bSJens Axboe 54138a72dacSKent Overstreet __bio_for_each_segment(bv, bio, iter, start) { 542f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 543f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 544f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 545f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 546f9c78b2bSJens Axboe } 547f9c78b2bSJens Axboe } 54838a72dacSKent Overstreet EXPORT_SYMBOL(zero_fill_bio_iter); 549f9c78b2bSJens Axboe 550f9c78b2bSJens Axboe /** 551f9c78b2bSJens Axboe * bio_put - release a reference to a bio 552f9c78b2bSJens Axboe * @bio: bio to release reference to 553f9c78b2bSJens Axboe * 554f9c78b2bSJens Axboe * Description: 555f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 5569b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 557f9c78b2bSJens Axboe **/ 558f9c78b2bSJens Axboe void bio_put(struct bio *bio) 559f9c78b2bSJens Axboe { 560dac56212SJens Axboe if (!bio_flagged(bio, BIO_REFFED)) 561dac56212SJens Axboe bio_free(bio); 562dac56212SJens Axboe else { 563dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 564f9c78b2bSJens Axboe 565f9c78b2bSJens Axboe /* 566f9c78b2bSJens Axboe * last put frees it 567f9c78b2bSJens Axboe */ 568dac56212SJens Axboe if (atomic_dec_and_test(&bio->__bi_cnt)) 569f9c78b2bSJens Axboe bio_free(bio); 570f9c78b2bSJens Axboe } 571dac56212SJens Axboe } 572f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 573f9c78b2bSJens Axboe 5746c210aa5SChristoph Hellwig int bio_phys_segments(struct request_queue *q, struct bio *bio) 575f9c78b2bSJens Axboe { 576f9c78b2bSJens Axboe if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 577f9c78b2bSJens Axboe blk_recount_segments(q, bio); 578f9c78b2bSJens Axboe 579f9c78b2bSJens Axboe return bio->bi_phys_segments; 580f9c78b2bSJens Axboe } 581f9c78b2bSJens Axboe 582f9c78b2bSJens Axboe /** 583f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 584f9c78b2bSJens Axboe * @bio: destination bio 585f9c78b2bSJens Axboe * @bio_src: bio to clone 586f9c78b2bSJens Axboe * 587f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 588f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 589f9c78b2bSJens Axboe * bio will be one. 590f9c78b2bSJens Axboe * 591f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 592f9c78b2bSJens Axboe */ 593f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 594f9c78b2bSJens Axboe { 595ed996a52SChristoph Hellwig BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 596f9c78b2bSJens Axboe 597f9c78b2bSJens Axboe /* 59874d46992SChristoph Hellwig * most users will be overriding ->bi_disk with a new target, 599f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 600f9c78b2bSJens Axboe */ 60174d46992SChristoph Hellwig bio->bi_disk = bio_src->bi_disk; 60262530ed8SMichael Lyle bio->bi_partno = bio_src->bi_partno; 603b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 604111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 605111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 6061eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 607ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 608cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 609f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 610f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 61120bd723eSPaolo Valente 612db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 613e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 614f9c78b2bSJens Axboe } 615f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 616f9c78b2bSJens Axboe 617f9c78b2bSJens Axboe /** 618f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 619f9c78b2bSJens Axboe * @bio: bio to clone 620f9c78b2bSJens Axboe * @gfp_mask: allocation priority 621f9c78b2bSJens Axboe * @bs: bio_set to allocate from 622f9c78b2bSJens Axboe * 623f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 624f9c78b2bSJens Axboe */ 625f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 626f9c78b2bSJens Axboe { 627f9c78b2bSJens Axboe struct bio *b; 628f9c78b2bSJens Axboe 629f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 630f9c78b2bSJens Axboe if (!b) 631f9c78b2bSJens Axboe return NULL; 632f9c78b2bSJens Axboe 633f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 634f9c78b2bSJens Axboe 635f9c78b2bSJens Axboe if (bio_integrity(bio)) { 636f9c78b2bSJens Axboe int ret; 637f9c78b2bSJens Axboe 638f9c78b2bSJens Axboe ret = bio_integrity_clone(b, bio, gfp_mask); 639f9c78b2bSJens Axboe 640f9c78b2bSJens Axboe if (ret < 0) { 641f9c78b2bSJens Axboe bio_put(b); 642f9c78b2bSJens Axboe return NULL; 643f9c78b2bSJens Axboe } 644f9c78b2bSJens Axboe } 645f9c78b2bSJens Axboe 646f9c78b2bSJens Axboe return b; 647f9c78b2bSJens Axboe } 648f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 649f9c78b2bSJens Axboe 650f4595875SShaohua Li /** 651c66a14d0SKent Overstreet * bio_add_pc_page - attempt to add page to bio 652c66a14d0SKent Overstreet * @q: the target queue 653c66a14d0SKent Overstreet * @bio: destination bio 654c66a14d0SKent Overstreet * @page: page to add 655c66a14d0SKent Overstreet * @len: vec entry length 656c66a14d0SKent Overstreet * @offset: vec entry offset 657f9c78b2bSJens Axboe * 658c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This can fail for a 659c66a14d0SKent Overstreet * number of reasons, such as the bio being full or target block device 660c66a14d0SKent Overstreet * limitations. The target block device must allow bio's up to PAGE_SIZE, 661c66a14d0SKent Overstreet * so it is always possible to add a single page to an empty bio. 662c66a14d0SKent Overstreet * 663c66a14d0SKent Overstreet * This should only be used by REQ_PC bios. 664f9c78b2bSJens Axboe */ 665c66a14d0SKent Overstreet int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page 666c66a14d0SKent Overstreet *page, unsigned int len, unsigned int offset) 667f9c78b2bSJens Axboe { 668f9c78b2bSJens Axboe int retried_segments = 0; 669f9c78b2bSJens Axboe struct bio_vec *bvec; 670f9c78b2bSJens Axboe 671f9c78b2bSJens Axboe /* 672f9c78b2bSJens Axboe * cloned bio must not modify vec list 673f9c78b2bSJens Axboe */ 674f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 675f9c78b2bSJens Axboe return 0; 676f9c78b2bSJens Axboe 677c66a14d0SKent Overstreet if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 678f9c78b2bSJens Axboe return 0; 679f9c78b2bSJens Axboe 680f9c78b2bSJens Axboe /* 681f9c78b2bSJens Axboe * For filesystems with a blocksize smaller than the pagesize 682f9c78b2bSJens Axboe * we will often be called with the same page as last time and 683f9c78b2bSJens Axboe * a consecutive offset. Optimize this special case. 684f9c78b2bSJens Axboe */ 685f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 686f9c78b2bSJens Axboe struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; 687f9c78b2bSJens Axboe 688f9c78b2bSJens Axboe if (page == prev->bv_page && 689f9c78b2bSJens Axboe offset == prev->bv_offset + prev->bv_len) { 690f9c78b2bSJens Axboe prev->bv_len += len; 691fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 692f9c78b2bSJens Axboe goto done; 693f9c78b2bSJens Axboe } 69466cb45aaSJens Axboe 69566cb45aaSJens Axboe /* 69666cb45aaSJens Axboe * If the queue doesn't support SG gaps and adding this 69766cb45aaSJens Axboe * offset would create a gap, disallow it. 69866cb45aaSJens Axboe */ 69903100aadSKeith Busch if (bvec_gap_to_prev(q, prev, offset)) 70066cb45aaSJens Axboe return 0; 701f9c78b2bSJens Axboe } 702f9c78b2bSJens Axboe 7030aa69fd3SChristoph Hellwig if (bio_full(bio)) 704f9c78b2bSJens Axboe return 0; 705f9c78b2bSJens Axboe 706f9c78b2bSJens Axboe /* 707f9c78b2bSJens Axboe * setup the new entry, we might clear it again later if we 708f9c78b2bSJens Axboe * cannot add the page 709f9c78b2bSJens Axboe */ 710f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 711f9c78b2bSJens Axboe bvec->bv_page = page; 712f9c78b2bSJens Axboe bvec->bv_len = len; 713f9c78b2bSJens Axboe bvec->bv_offset = offset; 714fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 715fcbf6a08SMaurizio Lombardi bio->bi_phys_segments++; 716fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 717fcbf6a08SMaurizio Lombardi 718fcbf6a08SMaurizio Lombardi /* 719fcbf6a08SMaurizio Lombardi * Perform a recount if the number of segments is greater 720fcbf6a08SMaurizio Lombardi * than queue_max_segments(q). 721fcbf6a08SMaurizio Lombardi */ 722fcbf6a08SMaurizio Lombardi 723fcbf6a08SMaurizio Lombardi while (bio->bi_phys_segments > queue_max_segments(q)) { 724fcbf6a08SMaurizio Lombardi 725fcbf6a08SMaurizio Lombardi if (retried_segments) 726fcbf6a08SMaurizio Lombardi goto failed; 727fcbf6a08SMaurizio Lombardi 728fcbf6a08SMaurizio Lombardi retried_segments = 1; 729fcbf6a08SMaurizio Lombardi blk_recount_segments(q, bio); 730fcbf6a08SMaurizio Lombardi } 731f9c78b2bSJens Axboe 732f9c78b2bSJens Axboe /* If we may be able to merge these biovecs, force a recount */ 7333dccdae5SChristoph Hellwig if (bio->bi_vcnt > 1 && biovec_phys_mergeable(q, bvec - 1, bvec)) 734b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_SEG_VALID); 735f9c78b2bSJens Axboe 736f9c78b2bSJens Axboe done: 737f9c78b2bSJens Axboe return len; 738fcbf6a08SMaurizio Lombardi 739fcbf6a08SMaurizio Lombardi failed: 740fcbf6a08SMaurizio Lombardi bvec->bv_page = NULL; 741fcbf6a08SMaurizio Lombardi bvec->bv_len = 0; 742fcbf6a08SMaurizio Lombardi bvec->bv_offset = 0; 743fcbf6a08SMaurizio Lombardi bio->bi_vcnt--; 744fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size -= len; 745fcbf6a08SMaurizio Lombardi blk_recount_segments(q, bio); 746fcbf6a08SMaurizio Lombardi return 0; 747f9c78b2bSJens Axboe } 748f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 749f9c78b2bSJens Axboe 750f9c78b2bSJens Axboe /** 7510aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 7520aa69fd3SChristoph Hellwig * @bio: destination bio 7530aa69fd3SChristoph Hellwig * @page: page to add 7540aa69fd3SChristoph Hellwig * @len: length of the data to add 7550aa69fd3SChristoph Hellwig * @off: offset of the data in @page 75607173c3eSMing Lei * @same_page: if %true only merge if the new data is in the same physical 75707173c3eSMing Lei * page as the last segment of the bio. 7580aa69fd3SChristoph Hellwig * 7590aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 7600aa69fd3SChristoph Hellwig * a useful optimisation for file systems with a block size smaller than the 7610aa69fd3SChristoph Hellwig * page size. 7620aa69fd3SChristoph Hellwig * 7630aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 7640aa69fd3SChristoph Hellwig */ 7650aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 76607173c3eSMing Lei unsigned int len, unsigned int off, bool same_page) 7670aa69fd3SChristoph Hellwig { 7680aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 7690aa69fd3SChristoph Hellwig return false; 7700aa69fd3SChristoph Hellwig 7710aa69fd3SChristoph Hellwig if (bio->bi_vcnt > 0) { 7720aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 77307173c3eSMing Lei phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 77407173c3eSMing Lei bv->bv_offset + bv->bv_len - 1; 77507173c3eSMing Lei phys_addr_t page_addr = page_to_phys(page); 7760aa69fd3SChristoph Hellwig 77707173c3eSMing Lei if (vec_end_addr + 1 != page_addr + off) 77807173c3eSMing Lei return false; 77907173c3eSMing Lei if (same_page && (vec_end_addr & PAGE_MASK) != page_addr) 78007173c3eSMing Lei return false; 78107173c3eSMing Lei 7820aa69fd3SChristoph Hellwig bv->bv_len += len; 7830aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 7840aa69fd3SChristoph Hellwig return true; 7850aa69fd3SChristoph Hellwig } 7860aa69fd3SChristoph Hellwig return false; 7870aa69fd3SChristoph Hellwig } 7880aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 7890aa69fd3SChristoph Hellwig 7900aa69fd3SChristoph Hellwig /** 7910aa69fd3SChristoph Hellwig * __bio_add_page - add page to a bio in a new segment 7920aa69fd3SChristoph Hellwig * @bio: destination bio 7930aa69fd3SChristoph Hellwig * @page: page to add 7940aa69fd3SChristoph Hellwig * @len: length of the data to add 7950aa69fd3SChristoph Hellwig * @off: offset of the data in @page 7960aa69fd3SChristoph Hellwig * 7970aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 7980aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 7990aa69fd3SChristoph Hellwig */ 8000aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 8010aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 8020aa69fd3SChristoph Hellwig { 8030aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 8040aa69fd3SChristoph Hellwig 8050aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 8060aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_full(bio)); 8070aa69fd3SChristoph Hellwig 8080aa69fd3SChristoph Hellwig bv->bv_page = page; 8090aa69fd3SChristoph Hellwig bv->bv_offset = off; 8100aa69fd3SChristoph Hellwig bv->bv_len = len; 8110aa69fd3SChristoph Hellwig 8120aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8130aa69fd3SChristoph Hellwig bio->bi_vcnt++; 8140aa69fd3SChristoph Hellwig } 8150aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 8160aa69fd3SChristoph Hellwig 8170aa69fd3SChristoph Hellwig /** 818f9c78b2bSJens Axboe * bio_add_page - attempt to add page to bio 819f9c78b2bSJens Axboe * @bio: destination bio 820f9c78b2bSJens Axboe * @page: page to add 821f9c78b2bSJens Axboe * @len: vec entry length 822f9c78b2bSJens Axboe * @offset: vec entry offset 823f9c78b2bSJens Axboe * 824c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This will only fail 825c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 826f9c78b2bSJens Axboe */ 827c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 828c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 829f9c78b2bSJens Axboe { 83007173c3eSMing Lei if (!__bio_try_merge_page(bio, page, len, offset, false)) { 8310aa69fd3SChristoph Hellwig if (bio_full(bio)) 832c66a14d0SKent Overstreet return 0; 8330aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 834c66a14d0SKent Overstreet } 835c66a14d0SKent Overstreet return len; 836f9c78b2bSJens Axboe } 837f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 838f9c78b2bSJens Axboe 8396d0c48aeSJens Axboe static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 8406d0c48aeSJens Axboe { 8416d0c48aeSJens Axboe const struct bio_vec *bv = iter->bvec; 8426d0c48aeSJens Axboe unsigned int len; 8436d0c48aeSJens Axboe size_t size; 8446d0c48aeSJens Axboe 8456d0c48aeSJens Axboe if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 8466d0c48aeSJens Axboe return -EINVAL; 8476d0c48aeSJens Axboe 8486d0c48aeSJens Axboe len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 8496d0c48aeSJens Axboe size = bio_add_page(bio, bv->bv_page, len, 8506d0c48aeSJens Axboe bv->bv_offset + iter->iov_offset); 8516d0c48aeSJens Axboe if (size == len) { 8526d0c48aeSJens Axboe struct page *page; 8536d0c48aeSJens Axboe int i; 8546d0c48aeSJens Axboe 8556d0c48aeSJens Axboe /* 8566d0c48aeSJens Axboe * For the normal O_DIRECT case, we could skip grabbing this 8576d0c48aeSJens Axboe * reference and then not have to put them again when IO 8586d0c48aeSJens Axboe * completes. But this breaks some in-kernel users, like 8596d0c48aeSJens Axboe * splicing to/from a loop device, where we release the pipe 8606d0c48aeSJens Axboe * pages unconditionally. If we can fix that case, we can 8616d0c48aeSJens Axboe * get rid of the get here and the need to call 8626d0c48aeSJens Axboe * bio_release_pages() at IO completion time. 8636d0c48aeSJens Axboe */ 8646d0c48aeSJens Axboe mp_bvec_for_each_page(page, bv, i) 8656d0c48aeSJens Axboe get_page(page); 8666d0c48aeSJens Axboe iov_iter_advance(iter, size); 8676d0c48aeSJens Axboe return 0; 8686d0c48aeSJens Axboe } 8696d0c48aeSJens Axboe 8706d0c48aeSJens Axboe return -EINVAL; 8716d0c48aeSJens Axboe } 8726d0c48aeSJens Axboe 873576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 874576ed913SChristoph Hellwig 8752cefe4dbSKent Overstreet /** 87617d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 8772cefe4dbSKent Overstreet * @bio: bio to add pages to 8782cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 8792cefe4dbSKent Overstreet * 88017d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 8812cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 88217d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 88317d51b10SMartin Wilck * the next non-empty segment of the iov iterator. 8842cefe4dbSKent Overstreet */ 88517d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 8862cefe4dbSKent Overstreet { 887576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 888576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 8892cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 8902cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 891576ed913SChristoph Hellwig ssize_t size, left; 892576ed913SChristoph Hellwig unsigned len, i; 893b403ea24SMartin Wilck size_t offset; 894576ed913SChristoph Hellwig 895576ed913SChristoph Hellwig /* 896576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 897576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 898576ed913SChristoph Hellwig * without overwriting the temporary page array. 899576ed913SChristoph Hellwig */ 900576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 901576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 9022cefe4dbSKent Overstreet 9032cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 9042cefe4dbSKent Overstreet if (unlikely(size <= 0)) 9052cefe4dbSKent Overstreet return size ? size : -EFAULT; 9062cefe4dbSKent Overstreet 907576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 908576ed913SChristoph Hellwig struct page *page = pages[i]; 9092cefe4dbSKent Overstreet 910576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 911576ed913SChristoph Hellwig if (WARN_ON_ONCE(bio_add_page(bio, page, len, offset) != len)) 912576ed913SChristoph Hellwig return -EINVAL; 913576ed913SChristoph Hellwig offset = 0; 9142cefe4dbSKent Overstreet } 9152cefe4dbSKent Overstreet 9162cefe4dbSKent Overstreet iov_iter_advance(iter, size); 9172cefe4dbSKent Overstreet return 0; 9182cefe4dbSKent Overstreet } 91917d51b10SMartin Wilck 92017d51b10SMartin Wilck /** 9216d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 92217d51b10SMartin Wilck * @bio: bio to add pages to 9236d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 92417d51b10SMartin Wilck * 9256d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 9266d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 9276d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 9286d0c48aeSJens Axboe * pages. For now, when adding kernel pages, we still grab a reference to the 9296d0c48aeSJens Axboe * page. This isn't strictly needed for the common case, but some call paths 9306d0c48aeSJens Axboe * end up releasing pages from eg a pipe and we can't easily control these. 9316d0c48aeSJens Axboe * See comment in __bio_iov_bvec_add_pages(). 9326d0c48aeSJens Axboe * 93317d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 9346d0c48aeSJens Axboe * fit into the bio, or are requested in *iter, whatever is smaller. If 9356d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 9366d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 93717d51b10SMartin Wilck */ 93817d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 93917d51b10SMartin Wilck { 9406d0c48aeSJens Axboe const bool is_bvec = iov_iter_is_bvec(iter); 94117d51b10SMartin Wilck unsigned short orig_vcnt = bio->bi_vcnt; 94217d51b10SMartin Wilck 94317d51b10SMartin Wilck do { 9446d0c48aeSJens Axboe int ret; 9456d0c48aeSJens Axboe 9466d0c48aeSJens Axboe if (is_bvec) 9476d0c48aeSJens Axboe ret = __bio_iov_bvec_add_pages(bio, iter); 9486d0c48aeSJens Axboe else 9496d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 95017d51b10SMartin Wilck 95117d51b10SMartin Wilck if (unlikely(ret)) 95217d51b10SMartin Wilck return bio->bi_vcnt > orig_vcnt ? 0 : ret; 95317d51b10SMartin Wilck 95417d51b10SMartin Wilck } while (iov_iter_count(iter) && !bio_full(bio)); 95517d51b10SMartin Wilck 95617d51b10SMartin Wilck return 0; 95717d51b10SMartin Wilck } 9582cefe4dbSKent Overstreet 9594246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 960f9c78b2bSJens Axboe { 96165e53aabSChristoph Hellwig complete(bio->bi_private); 962f9c78b2bSJens Axboe } 963f9c78b2bSJens Axboe 964f9c78b2bSJens Axboe /** 965f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 966f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 967f9c78b2bSJens Axboe * 968f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 969f9c78b2bSJens Axboe * bio_endio() on failure. 9703d289d68SJan Kara * 9713d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 9723d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 9733d289d68SJan Kara * on his own. 974f9c78b2bSJens Axboe */ 9754e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 976f9c78b2bSJens Axboe { 977e319e1fbSByungchul Park DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 978f9c78b2bSJens Axboe 97965e53aabSChristoph Hellwig bio->bi_private = &done; 980f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 9811eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 9824e49ea4aSMike Christie submit_bio(bio); 98365e53aabSChristoph Hellwig wait_for_completion_io(&done); 984f9c78b2bSJens Axboe 98565e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 986f9c78b2bSJens Axboe } 987f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 988f9c78b2bSJens Axboe 989f9c78b2bSJens Axboe /** 990f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 991f9c78b2bSJens Axboe * @bio: bio to advance 992f9c78b2bSJens Axboe * @bytes: number of bytes to complete 993f9c78b2bSJens Axboe * 994f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 995f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 996f9c78b2bSJens Axboe * be updated on the last bvec as well. 997f9c78b2bSJens Axboe * 998f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 999f9c78b2bSJens Axboe */ 1000f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 1001f9c78b2bSJens Axboe { 1002f9c78b2bSJens Axboe if (bio_integrity(bio)) 1003f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1004f9c78b2bSJens Axboe 1005f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1006f9c78b2bSJens Axboe } 1007f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 1008f9c78b2bSJens Axboe 100945db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 101045db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1011f9c78b2bSJens Axboe { 1012f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1013f9c78b2bSJens Axboe void *src_p, *dst_p; 1014f9c78b2bSJens Axboe unsigned bytes; 1015f9c78b2bSJens Axboe 101645db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 101745db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 101845db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 101945db54d5SKent Overstreet 102045db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 102145db54d5SKent Overstreet 102245db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 102345db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 102445db54d5SKent Overstreet 102545db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 102645db54d5SKent Overstreet src_p + src_bv.bv_offset, 102745db54d5SKent Overstreet bytes); 102845db54d5SKent Overstreet 102945db54d5SKent Overstreet kunmap_atomic(dst_p); 103045db54d5SKent Overstreet kunmap_atomic(src_p); 103145db54d5SKent Overstreet 10326e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 10336e6e811dSKent Overstreet 103445db54d5SKent Overstreet bio_advance_iter(src, src_iter, bytes); 103545db54d5SKent Overstreet bio_advance_iter(dst, dst_iter, bytes); 103645db54d5SKent Overstreet } 103745db54d5SKent Overstreet } 103845db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 103945db54d5SKent Overstreet 104045db54d5SKent Overstreet /** 104145db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 104245db54d5SKent Overstreet * @src: source bio 104345db54d5SKent Overstreet * @dst: destination bio 104445db54d5SKent Overstreet * 104545db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 104645db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 104745db54d5SKent Overstreet */ 104845db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 104945db54d5SKent Overstreet { 105045db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 105145db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 105245db54d5SKent Overstreet 105345db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 105445db54d5SKent Overstreet } 105545db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 105645db54d5SKent Overstreet 105745db54d5SKent Overstreet /** 105845db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 105945db54d5SKent Overstreet * another 106045db54d5SKent Overstreet * @src: source bio list 106145db54d5SKent Overstreet * @dst: destination bio list 106245db54d5SKent Overstreet * 106345db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 106445db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 106545db54d5SKent Overstreet * bios). 106645db54d5SKent Overstreet */ 106745db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 106845db54d5SKent Overstreet { 106945db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 107045db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 107145db54d5SKent Overstreet 1072f9c78b2bSJens Axboe while (1) { 1073f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1074f9c78b2bSJens Axboe src = src->bi_next; 1075f9c78b2bSJens Axboe if (!src) 1076f9c78b2bSJens Axboe break; 1077f9c78b2bSJens Axboe 1078f9c78b2bSJens Axboe src_iter = src->bi_iter; 1079f9c78b2bSJens Axboe } 1080f9c78b2bSJens Axboe 1081f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1082f9c78b2bSJens Axboe dst = dst->bi_next; 1083f9c78b2bSJens Axboe if (!dst) 1084f9c78b2bSJens Axboe break; 1085f9c78b2bSJens Axboe 1086f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1087f9c78b2bSJens Axboe } 1088f9c78b2bSJens Axboe 108945db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1090f9c78b2bSJens Axboe } 1091f9c78b2bSJens Axboe } 109245db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1093f9c78b2bSJens Axboe 1094f9c78b2bSJens Axboe struct bio_map_data { 1095f9c78b2bSJens Axboe int is_our_pages; 109626e49cfcSKent Overstreet struct iov_iter iter; 109726e49cfcSKent Overstreet struct iovec iov[]; 1098f9c78b2bSJens Axboe }; 1099f9c78b2bSJens Axboe 11000e5b935dSAl Viro static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1101f9c78b2bSJens Axboe gfp_t gfp_mask) 1102f9c78b2bSJens Axboe { 11030e5b935dSAl Viro struct bio_map_data *bmd; 11040e5b935dSAl Viro if (data->nr_segs > UIO_MAXIOV) 1105f9c78b2bSJens Axboe return NULL; 1106f9c78b2bSJens Axboe 11070e5b935dSAl Viro bmd = kmalloc(sizeof(struct bio_map_data) + 11080e5b935dSAl Viro sizeof(struct iovec) * data->nr_segs, gfp_mask); 11090e5b935dSAl Viro if (!bmd) 11100e5b935dSAl Viro return NULL; 11110e5b935dSAl Viro memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 11120e5b935dSAl Viro bmd->iter = *data; 11130e5b935dSAl Viro bmd->iter.iov = bmd->iov; 11140e5b935dSAl Viro return bmd; 1115f9c78b2bSJens Axboe } 1116f9c78b2bSJens Axboe 11179124d3feSDongsu Park /** 11189124d3feSDongsu Park * bio_copy_from_iter - copy all pages from iov_iter to bio 11199124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as destination 11209124d3feSDongsu Park * @iter: iov_iter as source 11219124d3feSDongsu Park * 11229124d3feSDongsu Park * Copy all pages from iov_iter to bio. 11239124d3feSDongsu Park * Returns 0 on success, or error on failure. 11249124d3feSDongsu Park */ 112598a09d61SAl Viro static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1126f9c78b2bSJens Axboe { 11279124d3feSDongsu Park int i; 1128f9c78b2bSJens Axboe struct bio_vec *bvec; 11296dc4f100SMing Lei struct bvec_iter_all iter_all; 1130f9c78b2bSJens Axboe 11316dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 11329124d3feSDongsu Park ssize_t ret; 1133f9c78b2bSJens Axboe 11349124d3feSDongsu Park ret = copy_page_from_iter(bvec->bv_page, 11359124d3feSDongsu Park bvec->bv_offset, 11369124d3feSDongsu Park bvec->bv_len, 113798a09d61SAl Viro iter); 1138f9c78b2bSJens Axboe 113998a09d61SAl Viro if (!iov_iter_count(iter)) 11409124d3feSDongsu Park break; 1141f9c78b2bSJens Axboe 11429124d3feSDongsu Park if (ret < bvec->bv_len) 11439124d3feSDongsu Park return -EFAULT; 1144f9c78b2bSJens Axboe } 1145f9c78b2bSJens Axboe 11469124d3feSDongsu Park return 0; 1147f9c78b2bSJens Axboe } 1148f9c78b2bSJens Axboe 11499124d3feSDongsu Park /** 11509124d3feSDongsu Park * bio_copy_to_iter - copy all pages from bio to iov_iter 11519124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as source 11529124d3feSDongsu Park * @iter: iov_iter as destination 11539124d3feSDongsu Park * 11549124d3feSDongsu Park * Copy all pages from bio to iov_iter. 11559124d3feSDongsu Park * Returns 0 on success, or error on failure. 11569124d3feSDongsu Park */ 11579124d3feSDongsu Park static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 11589124d3feSDongsu Park { 11599124d3feSDongsu Park int i; 11609124d3feSDongsu Park struct bio_vec *bvec; 11616dc4f100SMing Lei struct bvec_iter_all iter_all; 11629124d3feSDongsu Park 11636dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 11649124d3feSDongsu Park ssize_t ret; 11659124d3feSDongsu Park 11669124d3feSDongsu Park ret = copy_page_to_iter(bvec->bv_page, 11679124d3feSDongsu Park bvec->bv_offset, 11689124d3feSDongsu Park bvec->bv_len, 11699124d3feSDongsu Park &iter); 11709124d3feSDongsu Park 11719124d3feSDongsu Park if (!iov_iter_count(&iter)) 11729124d3feSDongsu Park break; 11739124d3feSDongsu Park 11749124d3feSDongsu Park if (ret < bvec->bv_len) 11759124d3feSDongsu Park return -EFAULT; 11769124d3feSDongsu Park } 11779124d3feSDongsu Park 11789124d3feSDongsu Park return 0; 1179f9c78b2bSJens Axboe } 1180f9c78b2bSJens Axboe 1181491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 11821dfa0f68SChristoph Hellwig { 11831dfa0f68SChristoph Hellwig struct bio_vec *bvec; 11841dfa0f68SChristoph Hellwig int i; 11856dc4f100SMing Lei struct bvec_iter_all iter_all; 11861dfa0f68SChristoph Hellwig 11876dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) 11881dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 11891dfa0f68SChristoph Hellwig } 1190491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 11911dfa0f68SChristoph Hellwig 1192f9c78b2bSJens Axboe /** 1193f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1194f9c78b2bSJens Axboe * @bio: bio being terminated 1195f9c78b2bSJens Axboe * 1196ddad8dd0SChristoph Hellwig * Free pages allocated from bio_copy_user_iov() and write back data 1197f9c78b2bSJens Axboe * to user space in case of a read. 1198f9c78b2bSJens Axboe */ 1199f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1200f9c78b2bSJens Axboe { 1201f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 12021dfa0f68SChristoph Hellwig int ret = 0; 1203f9c78b2bSJens Axboe 1204f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1205f9c78b2bSJens Axboe /* 1206f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 12072d99b55dSHannes Reinecke * don't copy into a random user address space, just free 12082d99b55dSHannes Reinecke * and return -EINTR so user space doesn't expect any data. 1209f9c78b2bSJens Axboe */ 12102d99b55dSHannes Reinecke if (!current->mm) 12112d99b55dSHannes Reinecke ret = -EINTR; 12122d99b55dSHannes Reinecke else if (bio_data_dir(bio) == READ) 12139124d3feSDongsu Park ret = bio_copy_to_iter(bio, bmd->iter); 12141dfa0f68SChristoph Hellwig if (bmd->is_our_pages) 12151dfa0f68SChristoph Hellwig bio_free_pages(bio); 1216f9c78b2bSJens Axboe } 1217f9c78b2bSJens Axboe kfree(bmd); 1218f9c78b2bSJens Axboe bio_put(bio); 1219f9c78b2bSJens Axboe return ret; 1220f9c78b2bSJens Axboe } 1221f9c78b2bSJens Axboe 1222f9c78b2bSJens Axboe /** 1223f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1224f9c78b2bSJens Axboe * @q: destination block queue 1225f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 122626e49cfcSKent Overstreet * @iter: iovec iterator 1227f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1228f9c78b2bSJens Axboe * 1229f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1230f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1231f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1232f9c78b2bSJens Axboe */ 1233f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1234f9c78b2bSJens Axboe struct rq_map_data *map_data, 1235e81cef5dSAl Viro struct iov_iter *iter, 123626e49cfcSKent Overstreet gfp_t gfp_mask) 1237f9c78b2bSJens Axboe { 1238f9c78b2bSJens Axboe struct bio_map_data *bmd; 1239f9c78b2bSJens Axboe struct page *page; 1240f9c78b2bSJens Axboe struct bio *bio; 1241d16d44ebSAl Viro int i = 0, ret; 1242d16d44ebSAl Viro int nr_pages; 124326e49cfcSKent Overstreet unsigned int len = iter->count; 1244bd5ceceaSGeliang Tang unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1245f9c78b2bSJens Axboe 12460e5b935dSAl Viro bmd = bio_alloc_map_data(iter, gfp_mask); 1247f9c78b2bSJens Axboe if (!bmd) 1248f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1249f9c78b2bSJens Axboe 125026e49cfcSKent Overstreet /* 125126e49cfcSKent Overstreet * We need to do a deep copy of the iov_iter including the iovecs. 125226e49cfcSKent Overstreet * The caller provided iov might point to an on-stack or otherwise 125326e49cfcSKent Overstreet * shortlived one. 125426e49cfcSKent Overstreet */ 125526e49cfcSKent Overstreet bmd->is_our_pages = map_data ? 0 : 1; 125626e49cfcSKent Overstreet 1257d16d44ebSAl Viro nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1258d16d44ebSAl Viro if (nr_pages > BIO_MAX_PAGES) 1259d16d44ebSAl Viro nr_pages = BIO_MAX_PAGES; 1260f9c78b2bSJens Axboe 1261f9c78b2bSJens Axboe ret = -ENOMEM; 1262f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1263f9c78b2bSJens Axboe if (!bio) 1264f9c78b2bSJens Axboe goto out_bmd; 1265f9c78b2bSJens Axboe 1266f9c78b2bSJens Axboe ret = 0; 1267f9c78b2bSJens Axboe 1268f9c78b2bSJens Axboe if (map_data) { 1269f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1270f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1271f9c78b2bSJens Axboe } 1272f9c78b2bSJens Axboe while (len) { 1273f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1274f9c78b2bSJens Axboe 1275f9c78b2bSJens Axboe bytes -= offset; 1276f9c78b2bSJens Axboe 1277f9c78b2bSJens Axboe if (bytes > len) 1278f9c78b2bSJens Axboe bytes = len; 1279f9c78b2bSJens Axboe 1280f9c78b2bSJens Axboe if (map_data) { 1281f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1282f9c78b2bSJens Axboe ret = -ENOMEM; 1283f9c78b2bSJens Axboe break; 1284f9c78b2bSJens Axboe } 1285f9c78b2bSJens Axboe 1286f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1287f9c78b2bSJens Axboe page += (i % nr_pages); 1288f9c78b2bSJens Axboe 1289f9c78b2bSJens Axboe i++; 1290f9c78b2bSJens Axboe } else { 1291f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1292f9c78b2bSJens Axboe if (!page) { 1293f9c78b2bSJens Axboe ret = -ENOMEM; 1294f9c78b2bSJens Axboe break; 1295f9c78b2bSJens Axboe } 1296f9c78b2bSJens Axboe } 1297f9c78b2bSJens Axboe 1298f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) 1299f9c78b2bSJens Axboe break; 1300f9c78b2bSJens Axboe 1301f9c78b2bSJens Axboe len -= bytes; 1302f9c78b2bSJens Axboe offset = 0; 1303f9c78b2bSJens Axboe } 1304f9c78b2bSJens Axboe 1305f9c78b2bSJens Axboe if (ret) 1306f9c78b2bSJens Axboe goto cleanup; 1307f9c78b2bSJens Axboe 13082884d0beSAl Viro if (map_data) 13092884d0beSAl Viro map_data->offset += bio->bi_iter.bi_size; 13102884d0beSAl Viro 1311f9c78b2bSJens Axboe /* 1312f9c78b2bSJens Axboe * success 1313f9c78b2bSJens Axboe */ 131400e23707SDavid Howells if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1315f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 131698a09d61SAl Viro ret = bio_copy_from_iter(bio, iter); 1317f9c78b2bSJens Axboe if (ret) 1318f9c78b2bSJens Axboe goto cleanup; 131998a09d61SAl Viro } else { 1320f55adad6SKeith Busch if (bmd->is_our_pages) 1321f3587d76SKeith Busch zero_fill_bio(bio); 1322e81cef5dSAl Viro iov_iter_advance(iter, bio->bi_iter.bi_size); 1323f9c78b2bSJens Axboe } 1324f9c78b2bSJens Axboe 132526e49cfcSKent Overstreet bio->bi_private = bmd; 13262884d0beSAl Viro if (map_data && map_data->null_mapped) 13272884d0beSAl Viro bio_set_flag(bio, BIO_NULL_MAPPED); 1328f9c78b2bSJens Axboe return bio; 1329f9c78b2bSJens Axboe cleanup: 1330f9c78b2bSJens Axboe if (!map_data) 13311dfa0f68SChristoph Hellwig bio_free_pages(bio); 1332f9c78b2bSJens Axboe bio_put(bio); 1333f9c78b2bSJens Axboe out_bmd: 1334f9c78b2bSJens Axboe kfree(bmd); 1335f9c78b2bSJens Axboe return ERR_PTR(ret); 1336f9c78b2bSJens Axboe } 1337f9c78b2bSJens Axboe 133837f19e57SChristoph Hellwig /** 133937f19e57SChristoph Hellwig * bio_map_user_iov - map user iovec into bio 134037f19e57SChristoph Hellwig * @q: the struct request_queue for the bio 134137f19e57SChristoph Hellwig * @iter: iovec iterator 134237f19e57SChristoph Hellwig * @gfp_mask: memory allocation flags 134337f19e57SChristoph Hellwig * 134437f19e57SChristoph Hellwig * Map the user space address into a bio suitable for io to a block 134537f19e57SChristoph Hellwig * device. Returns an error pointer in case of error. 134637f19e57SChristoph Hellwig */ 134737f19e57SChristoph Hellwig struct bio *bio_map_user_iov(struct request_queue *q, 1348e81cef5dSAl Viro struct iov_iter *iter, 134926e49cfcSKent Overstreet gfp_t gfp_mask) 1350f9c78b2bSJens Axboe { 135126e49cfcSKent Overstreet int j; 1352f9c78b2bSJens Axboe struct bio *bio; 1353076098e5SAl Viro int ret; 13542b04e8f6SAl Viro struct bio_vec *bvec; 13556dc4f100SMing Lei struct bvec_iter_all iter_all; 1356f9c78b2bSJens Axboe 1357b282cc76SAl Viro if (!iov_iter_count(iter)) 1358f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1359f9c78b2bSJens Axboe 1360b282cc76SAl Viro bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1361f9c78b2bSJens Axboe if (!bio) 1362f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1363f9c78b2bSJens Axboe 13640a0f1513SAl Viro while (iov_iter_count(iter)) { 1365629e42bcSAl Viro struct page **pages; 1366076098e5SAl Viro ssize_t bytes; 1367076098e5SAl Viro size_t offs, added = 0; 1368076098e5SAl Viro int npages; 1369f9c78b2bSJens Axboe 13700a0f1513SAl Viro bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1371076098e5SAl Viro if (unlikely(bytes <= 0)) { 1372076098e5SAl Viro ret = bytes ? bytes : -EFAULT; 1373f9c78b2bSJens Axboe goto out_unmap; 1374f9c78b2bSJens Axboe } 1375f9c78b2bSJens Axboe 1376076098e5SAl Viro npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1377076098e5SAl Viro 137898f0bc99SAl Viro if (unlikely(offs & queue_dma_alignment(q))) { 137998f0bc99SAl Viro ret = -EINVAL; 138098f0bc99SAl Viro j = 0; 138198f0bc99SAl Viro } else { 1382629e42bcSAl Viro for (j = 0; j < npages; j++) { 138398f0bc99SAl Viro struct page *page = pages[j]; 1384076098e5SAl Viro unsigned int n = PAGE_SIZE - offs; 138595d78c28SVitaly Mayatskikh unsigned short prev_bi_vcnt = bio->bi_vcnt; 1386f9c78b2bSJens Axboe 1387076098e5SAl Viro if (n > bytes) 1388076098e5SAl Viro n = bytes; 1389f9c78b2bSJens Axboe 139098f0bc99SAl Viro if (!bio_add_pc_page(q, bio, page, n, offs)) 1391f9c78b2bSJens Axboe break; 1392f9c78b2bSJens Axboe 139395d78c28SVitaly Mayatskikh /* 139495d78c28SVitaly Mayatskikh * check if vector was merged with previous 139595d78c28SVitaly Mayatskikh * drop page reference if needed 139695d78c28SVitaly Mayatskikh */ 139795d78c28SVitaly Mayatskikh if (bio->bi_vcnt == prev_bi_vcnt) 139898f0bc99SAl Viro put_page(page); 139995d78c28SVitaly Mayatskikh 1400076098e5SAl Viro added += n; 1401076098e5SAl Viro bytes -= n; 1402076098e5SAl Viro offs = 0; 1403f9c78b2bSJens Axboe } 14040a0f1513SAl Viro iov_iter_advance(iter, added); 140598f0bc99SAl Viro } 1406f9c78b2bSJens Axboe /* 1407f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1408f9c78b2bSJens Axboe */ 1409629e42bcSAl Viro while (j < npages) 141009cbfeafSKirill A. Shutemov put_page(pages[j++]); 1411629e42bcSAl Viro kvfree(pages); 1412e2e115d1SAl Viro /* couldn't stuff something into bio? */ 1413e2e115d1SAl Viro if (bytes) 1414e2e115d1SAl Viro break; 1415f9c78b2bSJens Axboe } 1416f9c78b2bSJens Axboe 1417b7c44ed9SJens Axboe bio_set_flag(bio, BIO_USER_MAPPED); 141837f19e57SChristoph Hellwig 141937f19e57SChristoph Hellwig /* 14205fad1b64SBart Van Assche * subtle -- if bio_map_user_iov() ended up bouncing a bio, 142137f19e57SChristoph Hellwig * it would normally disappear when its bi_end_io is run. 142237f19e57SChristoph Hellwig * however, we need it for the unmap, so grab an extra 142337f19e57SChristoph Hellwig * reference to it 142437f19e57SChristoph Hellwig */ 142537f19e57SChristoph Hellwig bio_get(bio); 1426f9c78b2bSJens Axboe return bio; 1427f9c78b2bSJens Axboe 1428f9c78b2bSJens Axboe out_unmap: 14296dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, j, iter_all) { 14302b04e8f6SAl Viro put_page(bvec->bv_page); 1431f9c78b2bSJens Axboe } 1432f9c78b2bSJens Axboe bio_put(bio); 1433f9c78b2bSJens Axboe return ERR_PTR(ret); 1434f9c78b2bSJens Axboe } 1435f9c78b2bSJens Axboe 1436f9c78b2bSJens Axboe static void __bio_unmap_user(struct bio *bio) 1437f9c78b2bSJens Axboe { 1438f9c78b2bSJens Axboe struct bio_vec *bvec; 1439f9c78b2bSJens Axboe int i; 14406dc4f100SMing Lei struct bvec_iter_all iter_all; 1441f9c78b2bSJens Axboe 1442f9c78b2bSJens Axboe /* 1443f9c78b2bSJens Axboe * make sure we dirty pages we wrote to 1444f9c78b2bSJens Axboe */ 14456dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 1446f9c78b2bSJens Axboe if (bio_data_dir(bio) == READ) 1447f9c78b2bSJens Axboe set_page_dirty_lock(bvec->bv_page); 1448f9c78b2bSJens Axboe 144909cbfeafSKirill A. Shutemov put_page(bvec->bv_page); 1450f9c78b2bSJens Axboe } 1451f9c78b2bSJens Axboe 1452f9c78b2bSJens Axboe bio_put(bio); 1453f9c78b2bSJens Axboe } 1454f9c78b2bSJens Axboe 1455f9c78b2bSJens Axboe /** 1456f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1457f9c78b2bSJens Axboe * @bio: the bio being unmapped 1458f9c78b2bSJens Axboe * 14595fad1b64SBart Van Assche * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 14605fad1b64SBart Van Assche * process context. 1461f9c78b2bSJens Axboe * 1462f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1463f9c78b2bSJens Axboe */ 1464f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1465f9c78b2bSJens Axboe { 1466f9c78b2bSJens Axboe __bio_unmap_user(bio); 1467f9c78b2bSJens Axboe bio_put(bio); 1468f9c78b2bSJens Axboe } 1469f9c78b2bSJens Axboe 14704246a0b6SChristoph Hellwig static void bio_map_kern_endio(struct bio *bio) 1471f9c78b2bSJens Axboe { 1472f9c78b2bSJens Axboe bio_put(bio); 1473f9c78b2bSJens Axboe } 1474f9c78b2bSJens Axboe 147575c72b83SChristoph Hellwig /** 147675c72b83SChristoph Hellwig * bio_map_kern - map kernel address into bio 147775c72b83SChristoph Hellwig * @q: the struct request_queue for the bio 147875c72b83SChristoph Hellwig * @data: pointer to buffer to map 147975c72b83SChristoph Hellwig * @len: length in bytes 148075c72b83SChristoph Hellwig * @gfp_mask: allocation flags for bio allocation 148175c72b83SChristoph Hellwig * 148275c72b83SChristoph Hellwig * Map the kernel address into a bio suitable for io to a block 148375c72b83SChristoph Hellwig * device. Returns an error pointer in case of error. 148475c72b83SChristoph Hellwig */ 148575c72b83SChristoph Hellwig struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 148675c72b83SChristoph Hellwig gfp_t gfp_mask) 1487f9c78b2bSJens Axboe { 1488f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1489f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1490f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1491f9c78b2bSJens Axboe const int nr_pages = end - start; 1492f9c78b2bSJens Axboe int offset, i; 1493f9c78b2bSJens Axboe struct bio *bio; 1494f9c78b2bSJens Axboe 1495f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1496f9c78b2bSJens Axboe if (!bio) 1497f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1498f9c78b2bSJens Axboe 1499f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1500f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1501f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1502f9c78b2bSJens Axboe 1503f9c78b2bSJens Axboe if (len <= 0) 1504f9c78b2bSJens Axboe break; 1505f9c78b2bSJens Axboe 1506f9c78b2bSJens Axboe if (bytes > len) 1507f9c78b2bSJens Axboe bytes = len; 1508f9c78b2bSJens Axboe 1509f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 151075c72b83SChristoph Hellwig offset) < bytes) { 151175c72b83SChristoph Hellwig /* we don't support partial mappings */ 151275c72b83SChristoph Hellwig bio_put(bio); 151375c72b83SChristoph Hellwig return ERR_PTR(-EINVAL); 151475c72b83SChristoph Hellwig } 1515f9c78b2bSJens Axboe 1516f9c78b2bSJens Axboe data += bytes; 1517f9c78b2bSJens Axboe len -= bytes; 1518f9c78b2bSJens Axboe offset = 0; 1519f9c78b2bSJens Axboe } 1520f9c78b2bSJens Axboe 1521f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1522f9c78b2bSJens Axboe return bio; 1523f9c78b2bSJens Axboe } 1524f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_map_kern); 1525f9c78b2bSJens Axboe 15264246a0b6SChristoph Hellwig static void bio_copy_kern_endio(struct bio *bio) 1527f9c78b2bSJens Axboe { 15281dfa0f68SChristoph Hellwig bio_free_pages(bio); 15291dfa0f68SChristoph Hellwig bio_put(bio); 15301dfa0f68SChristoph Hellwig } 15311dfa0f68SChristoph Hellwig 15324246a0b6SChristoph Hellwig static void bio_copy_kern_endio_read(struct bio *bio) 15331dfa0f68SChristoph Hellwig { 153442d2683aSChristoph Hellwig char *p = bio->bi_private; 15351dfa0f68SChristoph Hellwig struct bio_vec *bvec; 1536f9c78b2bSJens Axboe int i; 15376dc4f100SMing Lei struct bvec_iter_all iter_all; 1538f9c78b2bSJens Axboe 15396dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 15401dfa0f68SChristoph Hellwig memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1541f9c78b2bSJens Axboe p += bvec->bv_len; 1542f9c78b2bSJens Axboe } 1543f9c78b2bSJens Axboe 15444246a0b6SChristoph Hellwig bio_copy_kern_endio(bio); 1545f9c78b2bSJens Axboe } 1546f9c78b2bSJens Axboe 1547f9c78b2bSJens Axboe /** 1548f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1549f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1550f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1551f9c78b2bSJens Axboe * @len: length in bytes 1552f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1553f9c78b2bSJens Axboe * @reading: data direction is READ 1554f9c78b2bSJens Axboe * 1555f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1556f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1557f9c78b2bSJens Axboe */ 1558f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1559f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1560f9c78b2bSJens Axboe { 156142d2683aSChristoph Hellwig unsigned long kaddr = (unsigned long)data; 156242d2683aSChristoph Hellwig unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 156342d2683aSChristoph Hellwig unsigned long start = kaddr >> PAGE_SHIFT; 156442d2683aSChristoph Hellwig struct bio *bio; 1565f9c78b2bSJens Axboe void *p = data; 15661dfa0f68SChristoph Hellwig int nr_pages = 0; 1567f9c78b2bSJens Axboe 156842d2683aSChristoph Hellwig /* 156942d2683aSChristoph Hellwig * Overflow, abort 157042d2683aSChristoph Hellwig */ 157142d2683aSChristoph Hellwig if (end < start) 157242d2683aSChristoph Hellwig return ERR_PTR(-EINVAL); 1573f9c78b2bSJens Axboe 157442d2683aSChristoph Hellwig nr_pages = end - start; 157542d2683aSChristoph Hellwig bio = bio_kmalloc(gfp_mask, nr_pages); 157642d2683aSChristoph Hellwig if (!bio) 157742d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 157842d2683aSChristoph Hellwig 157942d2683aSChristoph Hellwig while (len) { 158042d2683aSChristoph Hellwig struct page *page; 158142d2683aSChristoph Hellwig unsigned int bytes = PAGE_SIZE; 158242d2683aSChristoph Hellwig 158342d2683aSChristoph Hellwig if (bytes > len) 158442d2683aSChristoph Hellwig bytes = len; 158542d2683aSChristoph Hellwig 158642d2683aSChristoph Hellwig page = alloc_page(q->bounce_gfp | gfp_mask); 158742d2683aSChristoph Hellwig if (!page) 158842d2683aSChristoph Hellwig goto cleanup; 158942d2683aSChristoph Hellwig 159042d2683aSChristoph Hellwig if (!reading) 159142d2683aSChristoph Hellwig memcpy(page_address(page), p, bytes); 159242d2683aSChristoph Hellwig 159342d2683aSChristoph Hellwig if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 159442d2683aSChristoph Hellwig break; 159542d2683aSChristoph Hellwig 159642d2683aSChristoph Hellwig len -= bytes; 159742d2683aSChristoph Hellwig p += bytes; 1598f9c78b2bSJens Axboe } 1599f9c78b2bSJens Axboe 16001dfa0f68SChristoph Hellwig if (reading) { 16011dfa0f68SChristoph Hellwig bio->bi_end_io = bio_copy_kern_endio_read; 160242d2683aSChristoph Hellwig bio->bi_private = data; 16031dfa0f68SChristoph Hellwig } else { 1604f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 16051dfa0f68SChristoph Hellwig } 16061dfa0f68SChristoph Hellwig 1607f9c78b2bSJens Axboe return bio; 160842d2683aSChristoph Hellwig 160942d2683aSChristoph Hellwig cleanup: 16101dfa0f68SChristoph Hellwig bio_free_pages(bio); 161142d2683aSChristoph Hellwig bio_put(bio); 161242d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 1613f9c78b2bSJens Axboe } 1614f9c78b2bSJens Axboe 1615f9c78b2bSJens Axboe /* 1616f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1617f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1618f9c78b2bSJens Axboe * 1619f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1620f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1621f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1622f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1623f9c78b2bSJens Axboe * in process context. 1624f9c78b2bSJens Axboe * 1625f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1626f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1627f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1628f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1629f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1630f9c78b2bSJens Axboe * 1631f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1632f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1633f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1634f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1635f9c78b2bSJens Axboe * pagecache. 1636f9c78b2bSJens Axboe * 1637f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1638f9c78b2bSJens Axboe * deferred bio dirtying paths. 1639f9c78b2bSJens Axboe */ 1640f9c78b2bSJens Axboe 1641f9c78b2bSJens Axboe /* 1642f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1643f9c78b2bSJens Axboe */ 1644f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1645f9c78b2bSJens Axboe { 1646f9c78b2bSJens Axboe struct bio_vec *bvec; 1647f9c78b2bSJens Axboe int i; 16486dc4f100SMing Lei struct bvec_iter_all iter_all; 1649f9c78b2bSJens Axboe 16506dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 16513bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 16523bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1653f9c78b2bSJens Axboe } 1654f9c78b2bSJens Axboe } 1655f9c78b2bSJens Axboe 1656f9c78b2bSJens Axboe static void bio_release_pages(struct bio *bio) 1657f9c78b2bSJens Axboe { 1658f9c78b2bSJens Axboe struct bio_vec *bvec; 1659f9c78b2bSJens Axboe int i; 16606dc4f100SMing Lei struct bvec_iter_all iter_all; 1661f9c78b2bSJens Axboe 16626dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) 166324d5493fSChristoph Hellwig put_page(bvec->bv_page); 1664f9c78b2bSJens Axboe } 1665f9c78b2bSJens Axboe 1666f9c78b2bSJens Axboe /* 1667f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1668f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1669f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 167024d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1671f9c78b2bSJens Axboe * 1672f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1673ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1674ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1675f9c78b2bSJens Axboe */ 1676f9c78b2bSJens Axboe 1677f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1678f9c78b2bSJens Axboe 1679f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1680f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1681f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1682f9c78b2bSJens Axboe 1683f9c78b2bSJens Axboe /* 1684f9c78b2bSJens Axboe * This runs in process context 1685f9c78b2bSJens Axboe */ 1686f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1687f9c78b2bSJens Axboe { 168824d5493fSChristoph Hellwig struct bio *bio, *next; 1689f9c78b2bSJens Axboe 169024d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 169124d5493fSChristoph Hellwig next = bio_dirty_list; 1692f9c78b2bSJens Axboe bio_dirty_list = NULL; 169324d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1694f9c78b2bSJens Axboe 169524d5493fSChristoph Hellwig while ((bio = next) != NULL) { 169624d5493fSChristoph Hellwig next = bio->bi_private; 1697f9c78b2bSJens Axboe 1698f9c78b2bSJens Axboe bio_set_pages_dirty(bio); 1699f9c78b2bSJens Axboe bio_release_pages(bio); 1700f9c78b2bSJens Axboe bio_put(bio); 1701f9c78b2bSJens Axboe } 1702f9c78b2bSJens Axboe } 1703f9c78b2bSJens Axboe 1704f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1705f9c78b2bSJens Axboe { 1706f9c78b2bSJens Axboe struct bio_vec *bvec; 170724d5493fSChristoph Hellwig unsigned long flags; 1708f9c78b2bSJens Axboe int i; 17096dc4f100SMing Lei struct bvec_iter_all iter_all; 1710f9c78b2bSJens Axboe 17116dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 171224d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 171324d5493fSChristoph Hellwig goto defer; 1714f9c78b2bSJens Axboe } 1715f9c78b2bSJens Axboe 171624d5493fSChristoph Hellwig bio_release_pages(bio); 171724d5493fSChristoph Hellwig bio_put(bio); 171824d5493fSChristoph Hellwig return; 171924d5493fSChristoph Hellwig defer: 1720f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1721f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1722f9c78b2bSJens Axboe bio_dirty_list = bio; 1723f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1724f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1725f9c78b2bSJens Axboe } 1726f9c78b2bSJens Axboe 17275b18b5a7SMikulas Patocka void update_io_ticks(struct hd_struct *part, unsigned long now) 17285b18b5a7SMikulas Patocka { 17295b18b5a7SMikulas Patocka unsigned long stamp; 17305b18b5a7SMikulas Patocka again: 17315b18b5a7SMikulas Patocka stamp = READ_ONCE(part->stamp); 17325b18b5a7SMikulas Patocka if (unlikely(stamp != now)) { 17335b18b5a7SMikulas Patocka if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 17345b18b5a7SMikulas Patocka __part_stat_add(part, io_ticks, 1); 17355b18b5a7SMikulas Patocka } 17365b18b5a7SMikulas Patocka } 17375b18b5a7SMikulas Patocka if (part->partno) { 17385b18b5a7SMikulas Patocka part = &part_to_disk(part)->part0; 17395b18b5a7SMikulas Patocka goto again; 17405b18b5a7SMikulas Patocka } 17415b18b5a7SMikulas Patocka } 1742f9c78b2bSJens Axboe 1743ddcf35d3SMichael Callahan void generic_start_io_acct(struct request_queue *q, int op, 1744d62e26b3SJens Axboe unsigned long sectors, struct hd_struct *part) 1745394ffa50SGu Zheng { 1746ddcf35d3SMichael Callahan const int sgrp = op_stat_group(op); 1747394ffa50SGu Zheng 1748112f158fSMike Snitzer part_stat_lock(); 1749112f158fSMike Snitzer 17505b18b5a7SMikulas Patocka update_io_ticks(part, jiffies); 1751112f158fSMike Snitzer part_stat_inc(part, ios[sgrp]); 1752112f158fSMike Snitzer part_stat_add(part, sectors[sgrp], sectors); 1753ddcf35d3SMichael Callahan part_inc_in_flight(q, part, op_is_write(op)); 1754394ffa50SGu Zheng 1755394ffa50SGu Zheng part_stat_unlock(); 1756394ffa50SGu Zheng } 1757394ffa50SGu Zheng EXPORT_SYMBOL(generic_start_io_acct); 1758394ffa50SGu Zheng 1759ddcf35d3SMichael Callahan void generic_end_io_acct(struct request_queue *q, int req_op, 1760d62e26b3SJens Axboe struct hd_struct *part, unsigned long start_time) 1761394ffa50SGu Zheng { 17625b18b5a7SMikulas Patocka unsigned long now = jiffies; 17635b18b5a7SMikulas Patocka unsigned long duration = now - start_time; 1764ddcf35d3SMichael Callahan const int sgrp = op_stat_group(req_op); 1765394ffa50SGu Zheng 1766112f158fSMike Snitzer part_stat_lock(); 1767112f158fSMike Snitzer 17685b18b5a7SMikulas Patocka update_io_ticks(part, now); 1769112f158fSMike Snitzer part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 17705b18b5a7SMikulas Patocka part_stat_add(part, time_in_queue, duration); 1771ddcf35d3SMichael Callahan part_dec_in_flight(q, part, op_is_write(req_op)); 1772394ffa50SGu Zheng 1773394ffa50SGu Zheng part_stat_unlock(); 1774394ffa50SGu Zheng } 1775394ffa50SGu Zheng EXPORT_SYMBOL(generic_end_io_acct); 1776394ffa50SGu Zheng 1777f9c78b2bSJens Axboe #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1778f9c78b2bSJens Axboe void bio_flush_dcache_pages(struct bio *bi) 1779f9c78b2bSJens Axboe { 1780f9c78b2bSJens Axboe struct bio_vec bvec; 1781f9c78b2bSJens Axboe struct bvec_iter iter; 1782f9c78b2bSJens Axboe 1783f9c78b2bSJens Axboe bio_for_each_segment(bvec, bi, iter) 1784f9c78b2bSJens Axboe flush_dcache_page(bvec.bv_page); 1785f9c78b2bSJens Axboe } 1786f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_flush_dcache_pages); 1787f9c78b2bSJens Axboe #endif 1788f9c78b2bSJens Axboe 1789c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1790c4cf5261SJens Axboe { 1791c4cf5261SJens Axboe /* 1792c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1793c4cf5261SJens Axboe * we always end io on the first invocation. 1794c4cf5261SJens Axboe */ 1795c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1796c4cf5261SJens Axboe return true; 1797c4cf5261SJens Axboe 1798c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1799c4cf5261SJens Axboe 1800326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1801b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1802c4cf5261SJens Axboe return true; 1803326e1dbbSMike Snitzer } 1804c4cf5261SJens Axboe 1805c4cf5261SJens Axboe return false; 1806c4cf5261SJens Axboe } 1807c4cf5261SJens Axboe 1808f9c78b2bSJens Axboe /** 1809f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1810f9c78b2bSJens Axboe * @bio: bio 1811f9c78b2bSJens Axboe * 1812f9c78b2bSJens Axboe * Description: 18134246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 18144246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 18154246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1816fbbaf700SNeilBrown * 1817fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1818fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1819fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1820fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1821f9c78b2bSJens Axboe **/ 18224246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1823f9c78b2bSJens Axboe { 1824ba8c6967SChristoph Hellwig again: 18252b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1826ba8c6967SChristoph Hellwig return; 18277c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 18287c20f116SChristoph Hellwig return; 1829f9c78b2bSJens Axboe 183067b42d0bSJosef Bacik if (bio->bi_disk) 183167b42d0bSJosef Bacik rq_qos_done_bio(bio->bi_disk->queue, bio); 183267b42d0bSJosef Bacik 1833f9c78b2bSJens Axboe /* 1834ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1835ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1836ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1837ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1838ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1839ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1840f9c78b2bSJens Axboe */ 1841f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 184238f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1843ba8c6967SChristoph Hellwig goto again; 1844ba8c6967SChristoph Hellwig } 1845ba8c6967SChristoph Hellwig 184674d46992SChristoph Hellwig if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 184774d46992SChristoph Hellwig trace_block_bio_complete(bio->bi_disk->queue, bio, 1848a462b950SBart Van Assche blk_status_to_errno(bio->bi_status)); 1849fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1850fbbaf700SNeilBrown } 1851fbbaf700SNeilBrown 18529e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1853b222dd2fSShaohua Li /* release cgroup info */ 1854b222dd2fSShaohua Li bio_uninit(bio); 1855f9c78b2bSJens Axboe if (bio->bi_end_io) 18564246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1857f9c78b2bSJens Axboe } 1858f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1859f9c78b2bSJens Axboe 1860f9c78b2bSJens Axboe /** 1861f9c78b2bSJens Axboe * bio_split - split a bio 1862f9c78b2bSJens Axboe * @bio: bio to split 1863f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1864f9c78b2bSJens Axboe * @gfp: gfp mask 1865f9c78b2bSJens Axboe * @bs: bio set to allocate from 1866f9c78b2bSJens Axboe * 1867f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1868f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1869f9c78b2bSJens Axboe * 1870f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1871f3f5da62SMartin K. Petersen * to @bio's bi_io_vec; it is the caller's responsibility to ensure that 1872f3f5da62SMartin K. Petersen * @bio is not freed before the split. 1873f9c78b2bSJens Axboe */ 1874f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1875f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1876f9c78b2bSJens Axboe { 1877f341a4d3SMikulas Patocka struct bio *split; 1878f9c78b2bSJens Axboe 1879f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1880f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1881f9c78b2bSJens Axboe 1882f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1883f9c78b2bSJens Axboe if (!split) 1884f9c78b2bSJens Axboe return NULL; 1885f9c78b2bSJens Axboe 1886f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1887f9c78b2bSJens Axboe 1888f9c78b2bSJens Axboe if (bio_integrity(split)) 1889fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1890f9c78b2bSJens Axboe 1891f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1892f9c78b2bSJens Axboe 1893fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 189420d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1895fbbaf700SNeilBrown 1896f9c78b2bSJens Axboe return split; 1897f9c78b2bSJens Axboe } 1898f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1899f9c78b2bSJens Axboe 1900f9c78b2bSJens Axboe /** 1901f9c78b2bSJens Axboe * bio_trim - trim a bio 1902f9c78b2bSJens Axboe * @bio: bio to trim 1903f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1904f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1905f9c78b2bSJens Axboe */ 1906f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1907f9c78b2bSJens Axboe { 1908f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1909f9c78b2bSJens Axboe * the given offset and size. 1910f9c78b2bSJens Axboe */ 1911f9c78b2bSJens Axboe 1912f9c78b2bSJens Axboe size <<= 9; 1913f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1914f9c78b2bSJens Axboe return; 1915f9c78b2bSJens Axboe 1916b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_SEG_VALID); 1917f9c78b2bSJens Axboe 1918f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1919f9c78b2bSJens Axboe 1920f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1921376a78abSDmitry Monakhov 1922376a78abSDmitry Monakhov if (bio_integrity(bio)) 1923fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1924376a78abSDmitry Monakhov 1925f9c78b2bSJens Axboe } 1926f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1927f9c78b2bSJens Axboe 1928f9c78b2bSJens Axboe /* 1929f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1930f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1931f9c78b2bSJens Axboe */ 19328aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1933f9c78b2bSJens Axboe { 1934ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1935f9c78b2bSJens Axboe 19368aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1937f9c78b2bSJens Axboe } 1938f9c78b2bSJens Axboe 1939917a38c7SKent Overstreet /* 1940917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1941917a38c7SKent Overstreet * 1942917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1943917a38c7SKent Overstreet * kzalloc()). 1944917a38c7SKent Overstreet */ 1945917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1946f9c78b2bSJens Axboe { 1947f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1948f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1949917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1950f9c78b2bSJens Axboe 19518aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 19528aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1953f9c78b2bSJens Axboe 1954f9c78b2bSJens Axboe bioset_integrity_free(bs); 1955917a38c7SKent Overstreet if (bs->bio_slab) 1956f9c78b2bSJens Axboe bio_put_slab(bs); 1957917a38c7SKent Overstreet bs->bio_slab = NULL; 1958917a38c7SKent Overstreet } 1959917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1960f9c78b2bSJens Axboe 1961011067b0SNeilBrown /** 1962917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1963dad08527SKent Overstreet * @bs: pool to initialize 1964917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1965917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1966917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1967917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1968917a38c7SKent Overstreet * 1969dad08527SKent Overstreet * Description: 1970dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1971dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1972dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1973dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1974dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1975dad08527SKent Overstreet * or things will break badly. 1976dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1977dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1978dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1979dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1980dad08527SKent Overstreet * 1981917a38c7SKent Overstreet */ 1982917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1983917a38c7SKent Overstreet unsigned int pool_size, 1984917a38c7SKent Overstreet unsigned int front_pad, 1985917a38c7SKent Overstreet int flags) 1986917a38c7SKent Overstreet { 1987917a38c7SKent Overstreet unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1988917a38c7SKent Overstreet 1989917a38c7SKent Overstreet bs->front_pad = front_pad; 1990917a38c7SKent Overstreet 1991917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1992917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1993917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1994917a38c7SKent Overstreet 1995917a38c7SKent Overstreet bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1996917a38c7SKent Overstreet if (!bs->bio_slab) 1997917a38c7SKent Overstreet return -ENOMEM; 1998917a38c7SKent Overstreet 1999917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2000917a38c7SKent Overstreet goto bad; 2001917a38c7SKent Overstreet 2002917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 2003917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 2004917a38c7SKent Overstreet goto bad; 2005917a38c7SKent Overstreet 2006917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 2007917a38c7SKent Overstreet return 0; 2008917a38c7SKent Overstreet 2009917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2010917a38c7SKent Overstreet if (!bs->rescue_workqueue) 2011917a38c7SKent Overstreet goto bad; 2012917a38c7SKent Overstreet 2013917a38c7SKent Overstreet return 0; 2014917a38c7SKent Overstreet bad: 2015917a38c7SKent Overstreet bioset_exit(bs); 2016917a38c7SKent Overstreet return -ENOMEM; 2017917a38c7SKent Overstreet } 2018917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 2019917a38c7SKent Overstreet 202028e89fd9SJens Axboe /* 202128e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 202228e89fd9SJens Axboe * another bio_set. 202328e89fd9SJens Axboe */ 202428e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 202528e89fd9SJens Axboe { 202628e89fd9SJens Axboe int flags; 202728e89fd9SJens Axboe 202828e89fd9SJens Axboe flags = 0; 202928e89fd9SJens Axboe if (src->bvec_pool.min_nr) 203028e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 203128e89fd9SJens Axboe if (src->rescue_workqueue) 203228e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 203328e89fd9SJens Axboe 203428e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 203528e89fd9SJens Axboe } 203628e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 203728e89fd9SJens Axboe 2038f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 20391d933cf0STejun Heo 20401d933cf0STejun Heo /** 20412268c0feSDennis Zhou * bio_disassociate_blkg - puts back the blkg reference if associated 2042b5f2954dSDennis Zhou * @bio: target bio 2043b5f2954dSDennis Zhou * 20442268c0feSDennis Zhou * Helper to disassociate the blkg from @bio if a blkg is associated. 2045b5f2954dSDennis Zhou */ 20462268c0feSDennis Zhou void bio_disassociate_blkg(struct bio *bio) 2047b5f2954dSDennis Zhou { 204808e18eabSJosef Bacik if (bio->bi_blkg) { 204908e18eabSJosef Bacik blkg_put(bio->bi_blkg); 205008e18eabSJosef Bacik bio->bi_blkg = NULL; 205108e18eabSJosef Bacik } 2052f9c78b2bSJens Axboe } 2053892ad71fSDennis Zhou EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2054f9c78b2bSJens Axboe 205520bd723eSPaolo Valente /** 20562268c0feSDennis Zhou * __bio_associate_blkg - associate a bio with the a blkg 2057f9c78b2bSJens Axboe * @bio: target bio 2058f9c78b2bSJens Axboe * @blkg: the blkg to associate 2059f9c78b2bSJens Axboe * 2060beea9da0SDennis Zhou * This tries to associate @bio with the specified @blkg. Association failure 2061beea9da0SDennis Zhou * is handled by walking up the blkg tree. Therefore, the blkg associated can 2062beea9da0SDennis Zhou * be anything between @blkg and the root_blkg. This situation only happens 2063beea9da0SDennis Zhou * when a cgroup is dying and then the remaining bios will spill to the closest 2064beea9da0SDennis Zhou * alive blkg. 2065beea9da0SDennis Zhou * 2066beea9da0SDennis Zhou * A reference will be taken on the @blkg and will be released when @bio is 2067beea9da0SDennis Zhou * freed. 2068f9c78b2bSJens Axboe */ 20692268c0feSDennis Zhou static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2070f9c78b2bSJens Axboe { 20712268c0feSDennis Zhou bio_disassociate_blkg(bio); 20722268c0feSDennis Zhou 20737754f669SDennis Zhou bio->bi_blkg = blkg_tryget_closest(blkg); 20742268c0feSDennis Zhou } 20752268c0feSDennis Zhou 2076fd42df30SDennis Zhou /** 2077fd42df30SDennis Zhou * bio_associate_blkg_from_css - associate a bio with a specified css 2078fd42df30SDennis Zhou * @bio: target bio 2079fd42df30SDennis Zhou * @css: target css 2080fd42df30SDennis Zhou * 2081fd42df30SDennis Zhou * Associate @bio with the blkg found by combining the css's blkg and the 2082fc5a828bSDennis Zhou * request_queue of the @bio. This falls back to the queue's root_blkg if 2083fc5a828bSDennis Zhou * the association fails with the css. 2084fd42df30SDennis Zhou */ 2085fd42df30SDennis Zhou void bio_associate_blkg_from_css(struct bio *bio, 2086fd42df30SDennis Zhou struct cgroup_subsys_state *css) 2087fd42df30SDennis Zhou { 2088fc5a828bSDennis Zhou struct request_queue *q = bio->bi_disk->queue; 2089fc5a828bSDennis Zhou struct blkcg_gq *blkg; 2090fc5a828bSDennis Zhou 2091fc5a828bSDennis Zhou rcu_read_lock(); 2092fc5a828bSDennis Zhou 2093fc5a828bSDennis Zhou if (!css || !css->parent) 2094fc5a828bSDennis Zhou blkg = q->root_blkg; 2095fc5a828bSDennis Zhou else 2096fc5a828bSDennis Zhou blkg = blkg_lookup_create(css_to_blkcg(css), q); 2097fc5a828bSDennis Zhou 2098fc5a828bSDennis Zhou __bio_associate_blkg(bio, blkg); 2099fc5a828bSDennis Zhou 2100fc5a828bSDennis Zhou rcu_read_unlock(); 2101fd42df30SDennis Zhou } 2102fd42df30SDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2103fd42df30SDennis Zhou 21046a7f6d86SDennis Zhou #ifdef CONFIG_MEMCG 21056a7f6d86SDennis Zhou /** 21066a7f6d86SDennis Zhou * bio_associate_blkg_from_page - associate a bio with the page's blkg 21076a7f6d86SDennis Zhou * @bio: target bio 21086a7f6d86SDennis Zhou * @page: the page to lookup the blkcg from 21096a7f6d86SDennis Zhou * 21106a7f6d86SDennis Zhou * Associate @bio with the blkg from @page's owning memcg and the respective 2111fc5a828bSDennis Zhou * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2112fc5a828bSDennis Zhou * root_blkg. 21136a7f6d86SDennis Zhou */ 21146a7f6d86SDennis Zhou void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 21156a7f6d86SDennis Zhou { 21166a7f6d86SDennis Zhou struct cgroup_subsys_state *css; 21176a7f6d86SDennis Zhou 21186a7f6d86SDennis Zhou if (!page->mem_cgroup) 21196a7f6d86SDennis Zhou return; 21206a7f6d86SDennis Zhou 2121fc5a828bSDennis Zhou rcu_read_lock(); 2122fc5a828bSDennis Zhou 2123fc5a828bSDennis Zhou css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2124fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 2125fc5a828bSDennis Zhou 2126fc5a828bSDennis Zhou rcu_read_unlock(); 21276a7f6d86SDennis Zhou } 21286a7f6d86SDennis Zhou #endif /* CONFIG_MEMCG */ 21296a7f6d86SDennis Zhou 21302268c0feSDennis Zhou /** 21312268c0feSDennis Zhou * bio_associate_blkg - associate a bio with a blkg 21322268c0feSDennis Zhou * @bio: target bio 21332268c0feSDennis Zhou * 21342268c0feSDennis Zhou * Associate @bio with the blkg found from the bio's css and request_queue. 21352268c0feSDennis Zhou * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 21362268c0feSDennis Zhou * already associated, the css is reused and association redone as the 21372268c0feSDennis Zhou * request_queue may have changed. 21382268c0feSDennis Zhou */ 21392268c0feSDennis Zhou void bio_associate_blkg(struct bio *bio) 21402268c0feSDennis Zhou { 2141fc5a828bSDennis Zhou struct cgroup_subsys_state *css; 21422268c0feSDennis Zhou 21432268c0feSDennis Zhou rcu_read_lock(); 21442268c0feSDennis Zhou 2145db6638d7SDennis Zhou if (bio->bi_blkg) 2146fc5a828bSDennis Zhou css = &bio_blkcg(bio)->css; 2147db6638d7SDennis Zhou else 2148fc5a828bSDennis Zhou css = blkcg_css(); 21492268c0feSDennis Zhou 2150fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 21512268c0feSDennis Zhou 21522268c0feSDennis Zhou rcu_read_unlock(); 2153f9c78b2bSJens Axboe } 21545cdf2e3fSDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg); 2155f9c78b2bSJens Axboe 215620bd723eSPaolo Valente /** 2157db6638d7SDennis Zhou * bio_clone_blkg_association - clone blkg association from src to dst bio 215820bd723eSPaolo Valente * @dst: destination bio 215920bd723eSPaolo Valente * @src: source bio 216020bd723eSPaolo Valente */ 2161db6638d7SDennis Zhou void bio_clone_blkg_association(struct bio *dst, struct bio *src) 216220bd723eSPaolo Valente { 21636ab21879SDennis Zhou rcu_read_lock(); 21646ab21879SDennis Zhou 2165fc5a828bSDennis Zhou if (src->bi_blkg) 21662268c0feSDennis Zhou __bio_associate_blkg(dst, src->bi_blkg); 21676ab21879SDennis Zhou 21686ab21879SDennis Zhou rcu_read_unlock(); 216920bd723eSPaolo Valente } 2170db6638d7SDennis Zhou EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2171f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 2172f9c78b2bSJens Axboe 2173f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2174f9c78b2bSJens Axboe { 2175f9c78b2bSJens Axboe int i; 2176f9c78b2bSJens Axboe 2177ed996a52SChristoph Hellwig for (i = 0; i < BVEC_POOL_NR; i++) { 2178f9c78b2bSJens Axboe int size; 2179f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2180f9c78b2bSJens Axboe 2181f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2182f9c78b2bSJens Axboe bvs->slab = NULL; 2183f9c78b2bSJens Axboe continue; 2184f9c78b2bSJens Axboe } 2185f9c78b2bSJens Axboe 2186f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2187f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2188f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2189f9c78b2bSJens Axboe } 2190f9c78b2bSJens Axboe } 2191f9c78b2bSJens Axboe 2192f9c78b2bSJens Axboe static int __init init_bio(void) 2193f9c78b2bSJens Axboe { 2194f9c78b2bSJens Axboe bio_slab_max = 2; 2195f9c78b2bSJens Axboe bio_slab_nr = 0; 21966396bb22SKees Cook bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 21976396bb22SKees Cook GFP_KERNEL); 2198f9c78b2bSJens Axboe if (!bio_slabs) 2199f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2200f9c78b2bSJens Axboe 2201f9c78b2bSJens Axboe bio_integrity_init(); 2202f9c78b2bSJens Axboe biovec_init_slabs(); 2203f9c78b2bSJens Axboe 2204f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2205f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2206f9c78b2bSJens Axboe 2207f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2208f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2209f9c78b2bSJens Axboe 2210f9c78b2bSJens Axboe return 0; 2211f9c78b2bSJens Axboe } 2212f9c78b2bSJens Axboe subsys_initcall(init_bio); 2213