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 6505919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 6515919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 6525919482eSMing Lei bool same_page) 6535919482eSMing Lei { 6545919482eSMing Lei phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 6555919482eSMing Lei bv->bv_offset + bv->bv_len - 1; 6565919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 6575919482eSMing Lei 6585919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 6595919482eSMing Lei return false; 6605919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 6615919482eSMing Lei return false; 6625919482eSMing Lei if (same_page && (vec_end_addr & PAGE_MASK) != page_addr) 6635919482eSMing Lei return false; 6645919482eSMing Lei 6655919482eSMing Lei return true; 6665919482eSMing Lei } 6675919482eSMing Lei 668489fbbcbSMing Lei /* 669489fbbcbSMing Lei * Check if the @page can be added to the current segment(@bv), and make 670489fbbcbSMing Lei * sure to call it only if page_is_mergeable(@bv, @page) is true 671489fbbcbSMing Lei */ 672489fbbcbSMing Lei static bool can_add_page_to_seg(struct request_queue *q, 673489fbbcbSMing Lei struct bio_vec *bv, struct page *page, unsigned len, 674489fbbcbSMing Lei unsigned offset) 675489fbbcbSMing Lei { 676489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 677489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 678489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 679489fbbcbSMing Lei 680489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 681489fbbcbSMing Lei return false; 682489fbbcbSMing Lei 683489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 684489fbbcbSMing Lei return false; 685489fbbcbSMing Lei 686489fbbcbSMing Lei return true; 687489fbbcbSMing Lei } 688489fbbcbSMing Lei 689f4595875SShaohua Li /** 69019047087SMing Lei * __bio_add_pc_page - attempt to add page to passthrough bio 691c66a14d0SKent Overstreet * @q: the target queue 692c66a14d0SKent Overstreet * @bio: destination bio 693c66a14d0SKent Overstreet * @page: page to add 694c66a14d0SKent Overstreet * @len: vec entry length 695c66a14d0SKent Overstreet * @offset: vec entry offset 69619047087SMing Lei * @put_same_page: put the page if it is same with last added page 697f9c78b2bSJens Axboe * 698c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This can fail for a 699c66a14d0SKent Overstreet * number of reasons, such as the bio being full or target block device 700c66a14d0SKent Overstreet * limitations. The target block device must allow bio's up to PAGE_SIZE, 701c66a14d0SKent Overstreet * so it is always possible to add a single page to an empty bio. 702c66a14d0SKent Overstreet * 7035a8ce240SMing Lei * This should only be used by passthrough bios. 704f9c78b2bSJens Axboe */ 70519047087SMing Lei int __bio_add_pc_page(struct request_queue *q, struct bio *bio, 70619047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 70719047087SMing Lei bool put_same_page) 708f9c78b2bSJens Axboe { 709f9c78b2bSJens Axboe struct bio_vec *bvec; 710f9c78b2bSJens Axboe 711f9c78b2bSJens Axboe /* 712f9c78b2bSJens Axboe * cloned bio must not modify vec list 713f9c78b2bSJens Axboe */ 714f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 715f9c78b2bSJens Axboe return 0; 716f9c78b2bSJens Axboe 717c66a14d0SKent Overstreet if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 718f9c78b2bSJens Axboe return 0; 719f9c78b2bSJens Axboe 720f9c78b2bSJens Axboe /* 721f9c78b2bSJens Axboe * For filesystems with a blocksize smaller than the pagesize 722f9c78b2bSJens Axboe * we will often be called with the same page as last time and 723f9c78b2bSJens Axboe * a consecutive offset. Optimize this special case. 724f9c78b2bSJens Axboe */ 725f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 7265a8ce240SMing Lei bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 727f9c78b2bSJens Axboe 7285a8ce240SMing Lei if (page == bvec->bv_page && 7295a8ce240SMing Lei offset == bvec->bv_offset + bvec->bv_len) { 73019047087SMing Lei if (put_same_page) 73119047087SMing Lei put_page(page); 732489fbbcbSMing Lei bvec_merge: 7335a8ce240SMing Lei bvec->bv_len += len; 734fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 735f9c78b2bSJens Axboe goto done; 736f9c78b2bSJens Axboe } 73766cb45aaSJens Axboe 73866cb45aaSJens Axboe /* 73966cb45aaSJens Axboe * If the queue doesn't support SG gaps and adding this 74066cb45aaSJens Axboe * offset would create a gap, disallow it. 74166cb45aaSJens Axboe */ 7425a8ce240SMing Lei if (bvec_gap_to_prev(q, bvec, offset)) 74366cb45aaSJens Axboe return 0; 744489fbbcbSMing Lei 745489fbbcbSMing Lei if (page_is_mergeable(bvec, page, len, offset, false) && 746489fbbcbSMing Lei can_add_page_to_seg(q, bvec, page, len, offset)) 747489fbbcbSMing Lei goto bvec_merge; 748f9c78b2bSJens Axboe } 749f9c78b2bSJens Axboe 7500aa69fd3SChristoph Hellwig if (bio_full(bio)) 751f9c78b2bSJens Axboe return 0; 752f9c78b2bSJens Axboe 753489fbbcbSMing Lei if (bio->bi_phys_segments >= queue_max_segments(q)) 754489fbbcbSMing Lei return 0; 755489fbbcbSMing Lei 756f9c78b2bSJens Axboe /* 757f9c78b2bSJens Axboe * setup the new entry, we might clear it again later if we 758f9c78b2bSJens Axboe * cannot add the page 759f9c78b2bSJens Axboe */ 760f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 761f9c78b2bSJens Axboe bvec->bv_page = page; 762f9c78b2bSJens Axboe bvec->bv_len = len; 763f9c78b2bSJens Axboe bvec->bv_offset = offset; 764fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 765fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 766fcbf6a08SMaurizio Lombardi 767f9c78b2bSJens Axboe done: 768489fbbcbSMing Lei bio->bi_phys_segments = bio->bi_vcnt; 769489fbbcbSMing Lei bio_set_flag(bio, BIO_SEG_VALID); 770f9c78b2bSJens Axboe return len; 771f9c78b2bSJens Axboe } 77219047087SMing Lei EXPORT_SYMBOL(__bio_add_pc_page); 77319047087SMing Lei 77419047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 77519047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 77619047087SMing Lei { 77719047087SMing Lei return __bio_add_pc_page(q, bio, page, len, offset, false); 77819047087SMing Lei } 779f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 780f9c78b2bSJens Axboe 781f9c78b2bSJens Axboe /** 7820aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 7830aa69fd3SChristoph Hellwig * @bio: destination bio 7840aa69fd3SChristoph Hellwig * @page: page to add 7850aa69fd3SChristoph Hellwig * @len: length of the data to add 7860aa69fd3SChristoph Hellwig * @off: offset of the data in @page 78707173c3eSMing Lei * @same_page: if %true only merge if the new data is in the same physical 78807173c3eSMing Lei * page as the last segment of the bio. 7890aa69fd3SChristoph Hellwig * 7900aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 7910aa69fd3SChristoph Hellwig * a useful optimisation for file systems with a block size smaller than the 7920aa69fd3SChristoph Hellwig * page size. 7930aa69fd3SChristoph Hellwig * 7940aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 7950aa69fd3SChristoph Hellwig */ 7960aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 79707173c3eSMing Lei unsigned int len, unsigned int off, bool same_page) 7980aa69fd3SChristoph Hellwig { 7990aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8000aa69fd3SChristoph Hellwig return false; 8010aa69fd3SChristoph Hellwig 8020aa69fd3SChristoph Hellwig if (bio->bi_vcnt > 0) { 8030aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 8040aa69fd3SChristoph Hellwig 8055919482eSMing Lei if (page_is_mergeable(bv, page, len, off, same_page)) { 8060aa69fd3SChristoph Hellwig bv->bv_len += len; 8070aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8080aa69fd3SChristoph Hellwig return true; 8090aa69fd3SChristoph Hellwig } 8105919482eSMing Lei } 8110aa69fd3SChristoph Hellwig return false; 8120aa69fd3SChristoph Hellwig } 8130aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 8140aa69fd3SChristoph Hellwig 8150aa69fd3SChristoph Hellwig /** 8160aa69fd3SChristoph Hellwig * __bio_add_page - add page to a bio in a new segment 8170aa69fd3SChristoph Hellwig * @bio: destination bio 8180aa69fd3SChristoph Hellwig * @page: page to add 8190aa69fd3SChristoph Hellwig * @len: length of the data to add 8200aa69fd3SChristoph Hellwig * @off: offset of the data in @page 8210aa69fd3SChristoph Hellwig * 8220aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 8230aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 8240aa69fd3SChristoph Hellwig */ 8250aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 8260aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 8270aa69fd3SChristoph Hellwig { 8280aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 8290aa69fd3SChristoph Hellwig 8300aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 8310aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_full(bio)); 8320aa69fd3SChristoph Hellwig 8330aa69fd3SChristoph Hellwig bv->bv_page = page; 8340aa69fd3SChristoph Hellwig bv->bv_offset = off; 8350aa69fd3SChristoph Hellwig bv->bv_len = len; 8360aa69fd3SChristoph Hellwig 8370aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8380aa69fd3SChristoph Hellwig bio->bi_vcnt++; 8390aa69fd3SChristoph Hellwig } 8400aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 8410aa69fd3SChristoph Hellwig 8420aa69fd3SChristoph Hellwig /** 843f9c78b2bSJens Axboe * bio_add_page - attempt to add page to bio 844f9c78b2bSJens Axboe * @bio: destination bio 845f9c78b2bSJens Axboe * @page: page to add 846f9c78b2bSJens Axboe * @len: vec entry length 847f9c78b2bSJens Axboe * @offset: vec entry offset 848f9c78b2bSJens Axboe * 849c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This will only fail 850c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 851f9c78b2bSJens Axboe */ 852c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 853c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 854f9c78b2bSJens Axboe { 85507173c3eSMing Lei if (!__bio_try_merge_page(bio, page, len, offset, false)) { 8560aa69fd3SChristoph Hellwig if (bio_full(bio)) 857c66a14d0SKent Overstreet return 0; 8580aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 859c66a14d0SKent Overstreet } 860c66a14d0SKent Overstreet return len; 861f9c78b2bSJens Axboe } 862f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 863f9c78b2bSJens Axboe 8646d0c48aeSJens Axboe static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 8656d0c48aeSJens Axboe { 8666d0c48aeSJens Axboe const struct bio_vec *bv = iter->bvec; 8676d0c48aeSJens Axboe unsigned int len; 8686d0c48aeSJens Axboe size_t size; 8696d0c48aeSJens Axboe 8706d0c48aeSJens Axboe if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 8716d0c48aeSJens Axboe return -EINVAL; 8726d0c48aeSJens Axboe 8736d0c48aeSJens Axboe len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 8746d0c48aeSJens Axboe size = bio_add_page(bio, bv->bv_page, len, 8756d0c48aeSJens Axboe bv->bv_offset + iter->iov_offset); 876a10584c3SChristoph Hellwig if (unlikely(size != len)) 877a10584c3SChristoph Hellwig return -EINVAL; 878a10584c3SChristoph Hellwig 879399254aaSJens Axboe if (!bio_flagged(bio, BIO_NO_PAGE_REF)) { 8806d0c48aeSJens Axboe struct page *page; 8816d0c48aeSJens Axboe int i; 8826d0c48aeSJens Axboe 8836d0c48aeSJens Axboe mp_bvec_for_each_page(page, bv, i) 8846d0c48aeSJens Axboe get_page(page); 885399254aaSJens Axboe } 886399254aaSJens Axboe 8876d0c48aeSJens Axboe iov_iter_advance(iter, size); 8886d0c48aeSJens Axboe return 0; 8896d0c48aeSJens Axboe } 8906d0c48aeSJens Axboe 891576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 892576ed913SChristoph Hellwig 8932cefe4dbSKent Overstreet /** 89417d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 8952cefe4dbSKent Overstreet * @bio: bio to add pages to 8962cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 8972cefe4dbSKent Overstreet * 89817d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 8992cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 90017d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 90117d51b10SMartin Wilck * the next non-empty segment of the iov iterator. 9022cefe4dbSKent Overstreet */ 90317d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 9042cefe4dbSKent Overstreet { 905576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 906576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 9072cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 9082cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 909576ed913SChristoph Hellwig ssize_t size, left; 910576ed913SChristoph Hellwig unsigned len, i; 911b403ea24SMartin Wilck size_t offset; 912576ed913SChristoph Hellwig 913576ed913SChristoph Hellwig /* 914576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 915576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 916576ed913SChristoph Hellwig * without overwriting the temporary page array. 917576ed913SChristoph Hellwig */ 918576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 919576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 9202cefe4dbSKent Overstreet 9212cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 9222cefe4dbSKent Overstreet if (unlikely(size <= 0)) 9232cefe4dbSKent Overstreet return size ? size : -EFAULT; 9242cefe4dbSKent Overstreet 925576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 926576ed913SChristoph Hellwig struct page *page = pages[i]; 9272cefe4dbSKent Overstreet 928576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 929576ed913SChristoph Hellwig if (WARN_ON_ONCE(bio_add_page(bio, page, len, offset) != len)) 930576ed913SChristoph Hellwig return -EINVAL; 931576ed913SChristoph Hellwig offset = 0; 9322cefe4dbSKent Overstreet } 9332cefe4dbSKent Overstreet 9342cefe4dbSKent Overstreet iov_iter_advance(iter, size); 9352cefe4dbSKent Overstreet return 0; 9362cefe4dbSKent Overstreet } 93717d51b10SMartin Wilck 93817d51b10SMartin Wilck /** 9396d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 94017d51b10SMartin Wilck * @bio: bio to add pages to 9416d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 94217d51b10SMartin Wilck * 9436d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 9446d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 9456d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 946399254aaSJens Axboe * pages. If we're adding kernel pages, and the caller told us it's safe to 947399254aaSJens Axboe * do so, we just have to add the pages to the bio directly. We don't grab an 948399254aaSJens Axboe * extra reference to those pages (the user should already have that), and we 949399254aaSJens Axboe * don't put the page on IO completion. The caller needs to check if the bio is 950399254aaSJens Axboe * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 951399254aaSJens Axboe * released. 9526d0c48aeSJens Axboe * 95317d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 9546d0c48aeSJens Axboe * fit into the bio, or are requested in *iter, whatever is smaller. If 9556d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 9566d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 95717d51b10SMartin Wilck */ 95817d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 95917d51b10SMartin Wilck { 9606d0c48aeSJens Axboe const bool is_bvec = iov_iter_is_bvec(iter); 96114eacf12SChristoph Hellwig int ret; 96214eacf12SChristoph Hellwig 96314eacf12SChristoph Hellwig if (WARN_ON_ONCE(bio->bi_vcnt)) 96414eacf12SChristoph Hellwig return -EINVAL; 96517d51b10SMartin Wilck 966399254aaSJens Axboe /* 967399254aaSJens Axboe * If this is a BVEC iter, then the pages are kernel pages. Don't 968399254aaSJens Axboe * release them on IO completion, if the caller asked us to. 969399254aaSJens Axboe */ 970399254aaSJens Axboe if (is_bvec && iov_iter_bvec_no_ref(iter)) 971399254aaSJens Axboe bio_set_flag(bio, BIO_NO_PAGE_REF); 972399254aaSJens Axboe 97317d51b10SMartin Wilck do { 9746d0c48aeSJens Axboe if (is_bvec) 9756d0c48aeSJens Axboe ret = __bio_iov_bvec_add_pages(bio, iter); 9766d0c48aeSJens Axboe else 9776d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 97814eacf12SChristoph Hellwig } while (!ret && iov_iter_count(iter) && !bio_full(bio)); 97917d51b10SMartin Wilck 98014eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 98117d51b10SMartin Wilck } 9822cefe4dbSKent Overstreet 9834246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 984f9c78b2bSJens Axboe { 98565e53aabSChristoph Hellwig complete(bio->bi_private); 986f9c78b2bSJens Axboe } 987f9c78b2bSJens Axboe 988f9c78b2bSJens Axboe /** 989f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 990f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 991f9c78b2bSJens Axboe * 992f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 993f9c78b2bSJens Axboe * bio_endio() on failure. 9943d289d68SJan Kara * 9953d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 9963d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 9973d289d68SJan Kara * on his own. 998f9c78b2bSJens Axboe */ 9994e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1000f9c78b2bSJens Axboe { 1001e319e1fbSByungchul Park DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 1002f9c78b2bSJens Axboe 100365e53aabSChristoph Hellwig bio->bi_private = &done; 1004f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 10051eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 10064e49ea4aSMike Christie submit_bio(bio); 100765e53aabSChristoph Hellwig wait_for_completion_io(&done); 1008f9c78b2bSJens Axboe 100965e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1010f9c78b2bSJens Axboe } 1011f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1012f9c78b2bSJens Axboe 1013f9c78b2bSJens Axboe /** 1014f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 1015f9c78b2bSJens Axboe * @bio: bio to advance 1016f9c78b2bSJens Axboe * @bytes: number of bytes to complete 1017f9c78b2bSJens Axboe * 1018f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1019f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1020f9c78b2bSJens Axboe * be updated on the last bvec as well. 1021f9c78b2bSJens Axboe * 1022f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 1023f9c78b2bSJens Axboe */ 1024f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 1025f9c78b2bSJens Axboe { 1026f9c78b2bSJens Axboe if (bio_integrity(bio)) 1027f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1028f9c78b2bSJens Axboe 1029f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1030f9c78b2bSJens Axboe } 1031f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 1032f9c78b2bSJens Axboe 103345db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 103445db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1035f9c78b2bSJens Axboe { 1036f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1037f9c78b2bSJens Axboe void *src_p, *dst_p; 1038f9c78b2bSJens Axboe unsigned bytes; 1039f9c78b2bSJens Axboe 104045db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 104145db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 104245db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 104345db54d5SKent Overstreet 104445db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 104545db54d5SKent Overstreet 104645db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 104745db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 104845db54d5SKent Overstreet 104945db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 105045db54d5SKent Overstreet src_p + src_bv.bv_offset, 105145db54d5SKent Overstreet bytes); 105245db54d5SKent Overstreet 105345db54d5SKent Overstreet kunmap_atomic(dst_p); 105445db54d5SKent Overstreet kunmap_atomic(src_p); 105545db54d5SKent Overstreet 10566e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 10576e6e811dSKent Overstreet 105845db54d5SKent Overstreet bio_advance_iter(src, src_iter, bytes); 105945db54d5SKent Overstreet bio_advance_iter(dst, dst_iter, bytes); 106045db54d5SKent Overstreet } 106145db54d5SKent Overstreet } 106245db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 106345db54d5SKent Overstreet 106445db54d5SKent Overstreet /** 106545db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 106645db54d5SKent Overstreet * @src: source bio 106745db54d5SKent Overstreet * @dst: destination bio 106845db54d5SKent Overstreet * 106945db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 107045db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 107145db54d5SKent Overstreet */ 107245db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 107345db54d5SKent Overstreet { 107445db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 107545db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 107645db54d5SKent Overstreet 107745db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 107845db54d5SKent Overstreet } 107945db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 108045db54d5SKent Overstreet 108145db54d5SKent Overstreet /** 108245db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 108345db54d5SKent Overstreet * another 108445db54d5SKent Overstreet * @src: source bio list 108545db54d5SKent Overstreet * @dst: destination bio list 108645db54d5SKent Overstreet * 108745db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 108845db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 108945db54d5SKent Overstreet * bios). 109045db54d5SKent Overstreet */ 109145db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 109245db54d5SKent Overstreet { 109345db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 109445db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 109545db54d5SKent Overstreet 1096f9c78b2bSJens Axboe while (1) { 1097f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1098f9c78b2bSJens Axboe src = src->bi_next; 1099f9c78b2bSJens Axboe if (!src) 1100f9c78b2bSJens Axboe break; 1101f9c78b2bSJens Axboe 1102f9c78b2bSJens Axboe src_iter = src->bi_iter; 1103f9c78b2bSJens Axboe } 1104f9c78b2bSJens Axboe 1105f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1106f9c78b2bSJens Axboe dst = dst->bi_next; 1107f9c78b2bSJens Axboe if (!dst) 1108f9c78b2bSJens Axboe break; 1109f9c78b2bSJens Axboe 1110f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1111f9c78b2bSJens Axboe } 1112f9c78b2bSJens Axboe 111345db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1114f9c78b2bSJens Axboe } 1115f9c78b2bSJens Axboe } 111645db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1117f9c78b2bSJens Axboe 1118f9c78b2bSJens Axboe struct bio_map_data { 1119f9c78b2bSJens Axboe int is_our_pages; 112026e49cfcSKent Overstreet struct iov_iter iter; 112126e49cfcSKent Overstreet struct iovec iov[]; 1122f9c78b2bSJens Axboe }; 1123f9c78b2bSJens Axboe 11240e5b935dSAl Viro static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1125f9c78b2bSJens Axboe gfp_t gfp_mask) 1126f9c78b2bSJens Axboe { 11270e5b935dSAl Viro struct bio_map_data *bmd; 11280e5b935dSAl Viro if (data->nr_segs > UIO_MAXIOV) 1129f9c78b2bSJens Axboe return NULL; 1130f9c78b2bSJens Axboe 11310e5b935dSAl Viro bmd = kmalloc(sizeof(struct bio_map_data) + 11320e5b935dSAl Viro sizeof(struct iovec) * data->nr_segs, gfp_mask); 11330e5b935dSAl Viro if (!bmd) 11340e5b935dSAl Viro return NULL; 11350e5b935dSAl Viro memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 11360e5b935dSAl Viro bmd->iter = *data; 11370e5b935dSAl Viro bmd->iter.iov = bmd->iov; 11380e5b935dSAl Viro return bmd; 1139f9c78b2bSJens Axboe } 1140f9c78b2bSJens Axboe 11419124d3feSDongsu Park /** 11429124d3feSDongsu Park * bio_copy_from_iter - copy all pages from iov_iter to bio 11439124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as destination 11449124d3feSDongsu Park * @iter: iov_iter as source 11459124d3feSDongsu Park * 11469124d3feSDongsu Park * Copy all pages from iov_iter to bio. 11479124d3feSDongsu Park * Returns 0 on success, or error on failure. 11489124d3feSDongsu Park */ 114998a09d61SAl Viro static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1150f9c78b2bSJens Axboe { 11519124d3feSDongsu Park int i; 1152f9c78b2bSJens Axboe struct bio_vec *bvec; 11536dc4f100SMing Lei struct bvec_iter_all iter_all; 1154f9c78b2bSJens Axboe 11556dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 11569124d3feSDongsu Park ssize_t ret; 1157f9c78b2bSJens Axboe 11589124d3feSDongsu Park ret = copy_page_from_iter(bvec->bv_page, 11599124d3feSDongsu Park bvec->bv_offset, 11609124d3feSDongsu Park bvec->bv_len, 116198a09d61SAl Viro iter); 1162f9c78b2bSJens Axboe 116398a09d61SAl Viro if (!iov_iter_count(iter)) 11649124d3feSDongsu Park break; 1165f9c78b2bSJens Axboe 11669124d3feSDongsu Park if (ret < bvec->bv_len) 11679124d3feSDongsu Park return -EFAULT; 1168f9c78b2bSJens Axboe } 1169f9c78b2bSJens Axboe 11709124d3feSDongsu Park return 0; 1171f9c78b2bSJens Axboe } 1172f9c78b2bSJens Axboe 11739124d3feSDongsu Park /** 11749124d3feSDongsu Park * bio_copy_to_iter - copy all pages from bio to iov_iter 11759124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as source 11769124d3feSDongsu Park * @iter: iov_iter as destination 11779124d3feSDongsu Park * 11789124d3feSDongsu Park * Copy all pages from bio to iov_iter. 11799124d3feSDongsu Park * Returns 0 on success, or error on failure. 11809124d3feSDongsu Park */ 11819124d3feSDongsu Park static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 11829124d3feSDongsu Park { 11839124d3feSDongsu Park int i; 11849124d3feSDongsu Park struct bio_vec *bvec; 11856dc4f100SMing Lei struct bvec_iter_all iter_all; 11869124d3feSDongsu Park 11876dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 11889124d3feSDongsu Park ssize_t ret; 11899124d3feSDongsu Park 11909124d3feSDongsu Park ret = copy_page_to_iter(bvec->bv_page, 11919124d3feSDongsu Park bvec->bv_offset, 11929124d3feSDongsu Park bvec->bv_len, 11939124d3feSDongsu Park &iter); 11949124d3feSDongsu Park 11959124d3feSDongsu Park if (!iov_iter_count(&iter)) 11969124d3feSDongsu Park break; 11979124d3feSDongsu Park 11989124d3feSDongsu Park if (ret < bvec->bv_len) 11999124d3feSDongsu Park return -EFAULT; 12009124d3feSDongsu Park } 12019124d3feSDongsu Park 12029124d3feSDongsu Park return 0; 1203f9c78b2bSJens Axboe } 1204f9c78b2bSJens Axboe 1205491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 12061dfa0f68SChristoph Hellwig { 12071dfa0f68SChristoph Hellwig struct bio_vec *bvec; 12081dfa0f68SChristoph Hellwig int i; 12096dc4f100SMing Lei struct bvec_iter_all iter_all; 12101dfa0f68SChristoph Hellwig 12116dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) 12121dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 12131dfa0f68SChristoph Hellwig } 1214491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 12151dfa0f68SChristoph Hellwig 1216f9c78b2bSJens Axboe /** 1217f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1218f9c78b2bSJens Axboe * @bio: bio being terminated 1219f9c78b2bSJens Axboe * 1220ddad8dd0SChristoph Hellwig * Free pages allocated from bio_copy_user_iov() and write back data 1221f9c78b2bSJens Axboe * to user space in case of a read. 1222f9c78b2bSJens Axboe */ 1223f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1224f9c78b2bSJens Axboe { 1225f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 12261dfa0f68SChristoph Hellwig int ret = 0; 1227f9c78b2bSJens Axboe 1228f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1229f9c78b2bSJens Axboe /* 1230f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 12312d99b55dSHannes Reinecke * don't copy into a random user address space, just free 12322d99b55dSHannes Reinecke * and return -EINTR so user space doesn't expect any data. 1233f9c78b2bSJens Axboe */ 12342d99b55dSHannes Reinecke if (!current->mm) 12352d99b55dSHannes Reinecke ret = -EINTR; 12362d99b55dSHannes Reinecke else if (bio_data_dir(bio) == READ) 12379124d3feSDongsu Park ret = bio_copy_to_iter(bio, bmd->iter); 12381dfa0f68SChristoph Hellwig if (bmd->is_our_pages) 12391dfa0f68SChristoph Hellwig bio_free_pages(bio); 1240f9c78b2bSJens Axboe } 1241f9c78b2bSJens Axboe kfree(bmd); 1242f9c78b2bSJens Axboe bio_put(bio); 1243f9c78b2bSJens Axboe return ret; 1244f9c78b2bSJens Axboe } 1245f9c78b2bSJens Axboe 1246f9c78b2bSJens Axboe /** 1247f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1248f9c78b2bSJens Axboe * @q: destination block queue 1249f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 125026e49cfcSKent Overstreet * @iter: iovec iterator 1251f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1252f9c78b2bSJens Axboe * 1253f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1254f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1255f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1256f9c78b2bSJens Axboe */ 1257f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1258f9c78b2bSJens Axboe struct rq_map_data *map_data, 1259e81cef5dSAl Viro struct iov_iter *iter, 126026e49cfcSKent Overstreet gfp_t gfp_mask) 1261f9c78b2bSJens Axboe { 1262f9c78b2bSJens Axboe struct bio_map_data *bmd; 1263f9c78b2bSJens Axboe struct page *page; 1264f9c78b2bSJens Axboe struct bio *bio; 1265d16d44ebSAl Viro int i = 0, ret; 1266d16d44ebSAl Viro int nr_pages; 126726e49cfcSKent Overstreet unsigned int len = iter->count; 1268bd5ceceaSGeliang Tang unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1269f9c78b2bSJens Axboe 12700e5b935dSAl Viro bmd = bio_alloc_map_data(iter, gfp_mask); 1271f9c78b2bSJens Axboe if (!bmd) 1272f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1273f9c78b2bSJens Axboe 127426e49cfcSKent Overstreet /* 127526e49cfcSKent Overstreet * We need to do a deep copy of the iov_iter including the iovecs. 127626e49cfcSKent Overstreet * The caller provided iov might point to an on-stack or otherwise 127726e49cfcSKent Overstreet * shortlived one. 127826e49cfcSKent Overstreet */ 127926e49cfcSKent Overstreet bmd->is_our_pages = map_data ? 0 : 1; 128026e49cfcSKent Overstreet 1281d16d44ebSAl Viro nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1282d16d44ebSAl Viro if (nr_pages > BIO_MAX_PAGES) 1283d16d44ebSAl Viro nr_pages = BIO_MAX_PAGES; 1284f9c78b2bSJens Axboe 1285f9c78b2bSJens Axboe ret = -ENOMEM; 1286f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1287f9c78b2bSJens Axboe if (!bio) 1288f9c78b2bSJens Axboe goto out_bmd; 1289f9c78b2bSJens Axboe 1290f9c78b2bSJens Axboe ret = 0; 1291f9c78b2bSJens Axboe 1292f9c78b2bSJens Axboe if (map_data) { 1293f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1294f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1295f9c78b2bSJens Axboe } 1296f9c78b2bSJens Axboe while (len) { 1297f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1298f9c78b2bSJens Axboe 1299f9c78b2bSJens Axboe bytes -= offset; 1300f9c78b2bSJens Axboe 1301f9c78b2bSJens Axboe if (bytes > len) 1302f9c78b2bSJens Axboe bytes = len; 1303f9c78b2bSJens Axboe 1304f9c78b2bSJens Axboe if (map_data) { 1305f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1306f9c78b2bSJens Axboe ret = -ENOMEM; 1307f9c78b2bSJens Axboe break; 1308f9c78b2bSJens Axboe } 1309f9c78b2bSJens Axboe 1310f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1311f9c78b2bSJens Axboe page += (i % nr_pages); 1312f9c78b2bSJens Axboe 1313f9c78b2bSJens Axboe i++; 1314f9c78b2bSJens Axboe } else { 1315f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1316f9c78b2bSJens Axboe if (!page) { 1317f9c78b2bSJens Axboe ret = -ENOMEM; 1318f9c78b2bSJens Axboe break; 1319f9c78b2bSJens Axboe } 1320f9c78b2bSJens Axboe } 1321f9c78b2bSJens Axboe 1322f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) 1323f9c78b2bSJens Axboe break; 1324f9c78b2bSJens Axboe 1325f9c78b2bSJens Axboe len -= bytes; 1326f9c78b2bSJens Axboe offset = 0; 1327f9c78b2bSJens Axboe } 1328f9c78b2bSJens Axboe 1329f9c78b2bSJens Axboe if (ret) 1330f9c78b2bSJens Axboe goto cleanup; 1331f9c78b2bSJens Axboe 13322884d0beSAl Viro if (map_data) 13332884d0beSAl Viro map_data->offset += bio->bi_iter.bi_size; 13342884d0beSAl Viro 1335f9c78b2bSJens Axboe /* 1336f9c78b2bSJens Axboe * success 1337f9c78b2bSJens Axboe */ 133800e23707SDavid Howells if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1339f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 134098a09d61SAl Viro ret = bio_copy_from_iter(bio, iter); 1341f9c78b2bSJens Axboe if (ret) 1342f9c78b2bSJens Axboe goto cleanup; 134398a09d61SAl Viro } else { 1344f55adad6SKeith Busch if (bmd->is_our_pages) 1345f3587d76SKeith Busch zero_fill_bio(bio); 1346e81cef5dSAl Viro iov_iter_advance(iter, bio->bi_iter.bi_size); 1347f9c78b2bSJens Axboe } 1348f9c78b2bSJens Axboe 134926e49cfcSKent Overstreet bio->bi_private = bmd; 13502884d0beSAl Viro if (map_data && map_data->null_mapped) 13512884d0beSAl Viro bio_set_flag(bio, BIO_NULL_MAPPED); 1352f9c78b2bSJens Axboe return bio; 1353f9c78b2bSJens Axboe cleanup: 1354f9c78b2bSJens Axboe if (!map_data) 13551dfa0f68SChristoph Hellwig bio_free_pages(bio); 1356f9c78b2bSJens Axboe bio_put(bio); 1357f9c78b2bSJens Axboe out_bmd: 1358f9c78b2bSJens Axboe kfree(bmd); 1359f9c78b2bSJens Axboe return ERR_PTR(ret); 1360f9c78b2bSJens Axboe } 1361f9c78b2bSJens Axboe 136237f19e57SChristoph Hellwig /** 136337f19e57SChristoph Hellwig * bio_map_user_iov - map user iovec into bio 136437f19e57SChristoph Hellwig * @q: the struct request_queue for the bio 136537f19e57SChristoph Hellwig * @iter: iovec iterator 136637f19e57SChristoph Hellwig * @gfp_mask: memory allocation flags 136737f19e57SChristoph Hellwig * 136837f19e57SChristoph Hellwig * Map the user space address into a bio suitable for io to a block 136937f19e57SChristoph Hellwig * device. Returns an error pointer in case of error. 137037f19e57SChristoph Hellwig */ 137137f19e57SChristoph Hellwig struct bio *bio_map_user_iov(struct request_queue *q, 1372e81cef5dSAl Viro struct iov_iter *iter, 137326e49cfcSKent Overstreet gfp_t gfp_mask) 1374f9c78b2bSJens Axboe { 137526e49cfcSKent Overstreet int j; 1376f9c78b2bSJens Axboe struct bio *bio; 1377076098e5SAl Viro int ret; 13782b04e8f6SAl Viro struct bio_vec *bvec; 13796dc4f100SMing Lei struct bvec_iter_all iter_all; 1380f9c78b2bSJens Axboe 1381b282cc76SAl Viro if (!iov_iter_count(iter)) 1382f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1383f9c78b2bSJens Axboe 1384b282cc76SAl Viro bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1385f9c78b2bSJens Axboe if (!bio) 1386f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1387f9c78b2bSJens Axboe 13880a0f1513SAl Viro while (iov_iter_count(iter)) { 1389629e42bcSAl Viro struct page **pages; 1390076098e5SAl Viro ssize_t bytes; 1391076098e5SAl Viro size_t offs, added = 0; 1392076098e5SAl Viro int npages; 1393f9c78b2bSJens Axboe 13940a0f1513SAl Viro bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1395076098e5SAl Viro if (unlikely(bytes <= 0)) { 1396076098e5SAl Viro ret = bytes ? bytes : -EFAULT; 1397f9c78b2bSJens Axboe goto out_unmap; 1398f9c78b2bSJens Axboe } 1399f9c78b2bSJens Axboe 1400076098e5SAl Viro npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1401076098e5SAl Viro 140298f0bc99SAl Viro if (unlikely(offs & queue_dma_alignment(q))) { 140398f0bc99SAl Viro ret = -EINVAL; 140498f0bc99SAl Viro j = 0; 140598f0bc99SAl Viro } else { 1406629e42bcSAl Viro for (j = 0; j < npages; j++) { 140798f0bc99SAl Viro struct page *page = pages[j]; 1408076098e5SAl Viro unsigned int n = PAGE_SIZE - offs; 1409f9c78b2bSJens Axboe 1410076098e5SAl Viro if (n > bytes) 1411076098e5SAl Viro n = bytes; 1412f9c78b2bSJens Axboe 141319047087SMing Lei if (!__bio_add_pc_page(q, bio, page, n, offs, 141419047087SMing Lei true)) 1415f9c78b2bSJens Axboe break; 1416f9c78b2bSJens Axboe 1417076098e5SAl Viro added += n; 1418076098e5SAl Viro bytes -= n; 1419076098e5SAl Viro offs = 0; 1420f9c78b2bSJens Axboe } 14210a0f1513SAl Viro iov_iter_advance(iter, added); 142298f0bc99SAl Viro } 1423f9c78b2bSJens Axboe /* 1424f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1425f9c78b2bSJens Axboe */ 1426629e42bcSAl Viro while (j < npages) 142709cbfeafSKirill A. Shutemov put_page(pages[j++]); 1428629e42bcSAl Viro kvfree(pages); 1429e2e115d1SAl Viro /* couldn't stuff something into bio? */ 1430e2e115d1SAl Viro if (bytes) 1431e2e115d1SAl Viro break; 1432f9c78b2bSJens Axboe } 1433f9c78b2bSJens Axboe 1434b7c44ed9SJens Axboe bio_set_flag(bio, BIO_USER_MAPPED); 143537f19e57SChristoph Hellwig 143637f19e57SChristoph Hellwig /* 14375fad1b64SBart Van Assche * subtle -- if bio_map_user_iov() ended up bouncing a bio, 143837f19e57SChristoph Hellwig * it would normally disappear when its bi_end_io is run. 143937f19e57SChristoph Hellwig * however, we need it for the unmap, so grab an extra 144037f19e57SChristoph Hellwig * reference to it 144137f19e57SChristoph Hellwig */ 144237f19e57SChristoph Hellwig bio_get(bio); 1443f9c78b2bSJens Axboe return bio; 1444f9c78b2bSJens Axboe 1445f9c78b2bSJens Axboe out_unmap: 14466dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, j, iter_all) { 14472b04e8f6SAl Viro put_page(bvec->bv_page); 1448f9c78b2bSJens Axboe } 1449f9c78b2bSJens Axboe bio_put(bio); 1450f9c78b2bSJens Axboe return ERR_PTR(ret); 1451f9c78b2bSJens Axboe } 1452f9c78b2bSJens Axboe 1453f9c78b2bSJens Axboe static void __bio_unmap_user(struct bio *bio) 1454f9c78b2bSJens Axboe { 1455f9c78b2bSJens Axboe struct bio_vec *bvec; 1456f9c78b2bSJens Axboe int i; 14576dc4f100SMing Lei struct bvec_iter_all iter_all; 1458f9c78b2bSJens Axboe 1459f9c78b2bSJens Axboe /* 1460f9c78b2bSJens Axboe * make sure we dirty pages we wrote to 1461f9c78b2bSJens Axboe */ 14626dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 1463f9c78b2bSJens Axboe if (bio_data_dir(bio) == READ) 1464f9c78b2bSJens Axboe set_page_dirty_lock(bvec->bv_page); 1465f9c78b2bSJens Axboe 146609cbfeafSKirill A. Shutemov put_page(bvec->bv_page); 1467f9c78b2bSJens Axboe } 1468f9c78b2bSJens Axboe 1469f9c78b2bSJens Axboe bio_put(bio); 1470f9c78b2bSJens Axboe } 1471f9c78b2bSJens Axboe 1472f9c78b2bSJens Axboe /** 1473f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1474f9c78b2bSJens Axboe * @bio: the bio being unmapped 1475f9c78b2bSJens Axboe * 14765fad1b64SBart Van Assche * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 14775fad1b64SBart Van Assche * process context. 1478f9c78b2bSJens Axboe * 1479f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1480f9c78b2bSJens Axboe */ 1481f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1482f9c78b2bSJens Axboe { 1483f9c78b2bSJens Axboe __bio_unmap_user(bio); 1484f9c78b2bSJens Axboe bio_put(bio); 1485f9c78b2bSJens Axboe } 1486f9c78b2bSJens Axboe 14874246a0b6SChristoph Hellwig static void bio_map_kern_endio(struct bio *bio) 1488f9c78b2bSJens Axboe { 1489f9c78b2bSJens Axboe bio_put(bio); 1490f9c78b2bSJens Axboe } 1491f9c78b2bSJens Axboe 149275c72b83SChristoph Hellwig /** 149375c72b83SChristoph Hellwig * bio_map_kern - map kernel address into bio 149475c72b83SChristoph Hellwig * @q: the struct request_queue for the bio 149575c72b83SChristoph Hellwig * @data: pointer to buffer to map 149675c72b83SChristoph Hellwig * @len: length in bytes 149775c72b83SChristoph Hellwig * @gfp_mask: allocation flags for bio allocation 149875c72b83SChristoph Hellwig * 149975c72b83SChristoph Hellwig * Map the kernel address into a bio suitable for io to a block 150075c72b83SChristoph Hellwig * device. Returns an error pointer in case of error. 150175c72b83SChristoph Hellwig */ 150275c72b83SChristoph Hellwig struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 150375c72b83SChristoph Hellwig gfp_t gfp_mask) 1504f9c78b2bSJens Axboe { 1505f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1506f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1507f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1508f9c78b2bSJens Axboe const int nr_pages = end - start; 1509f9c78b2bSJens Axboe int offset, i; 1510f9c78b2bSJens Axboe struct bio *bio; 1511f9c78b2bSJens Axboe 1512f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1513f9c78b2bSJens Axboe if (!bio) 1514f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1515f9c78b2bSJens Axboe 1516f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1517f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1518f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1519f9c78b2bSJens Axboe 1520f9c78b2bSJens Axboe if (len <= 0) 1521f9c78b2bSJens Axboe break; 1522f9c78b2bSJens Axboe 1523f9c78b2bSJens Axboe if (bytes > len) 1524f9c78b2bSJens Axboe bytes = len; 1525f9c78b2bSJens Axboe 1526f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 152775c72b83SChristoph Hellwig offset) < bytes) { 152875c72b83SChristoph Hellwig /* we don't support partial mappings */ 152975c72b83SChristoph Hellwig bio_put(bio); 153075c72b83SChristoph Hellwig return ERR_PTR(-EINVAL); 153175c72b83SChristoph Hellwig } 1532f9c78b2bSJens Axboe 1533f9c78b2bSJens Axboe data += bytes; 1534f9c78b2bSJens Axboe len -= bytes; 1535f9c78b2bSJens Axboe offset = 0; 1536f9c78b2bSJens Axboe } 1537f9c78b2bSJens Axboe 1538f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1539f9c78b2bSJens Axboe return bio; 1540f9c78b2bSJens Axboe } 1541f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_map_kern); 1542f9c78b2bSJens Axboe 15434246a0b6SChristoph Hellwig static void bio_copy_kern_endio(struct bio *bio) 1544f9c78b2bSJens Axboe { 15451dfa0f68SChristoph Hellwig bio_free_pages(bio); 15461dfa0f68SChristoph Hellwig bio_put(bio); 15471dfa0f68SChristoph Hellwig } 15481dfa0f68SChristoph Hellwig 15494246a0b6SChristoph Hellwig static void bio_copy_kern_endio_read(struct bio *bio) 15501dfa0f68SChristoph Hellwig { 155142d2683aSChristoph Hellwig char *p = bio->bi_private; 15521dfa0f68SChristoph Hellwig struct bio_vec *bvec; 1553f9c78b2bSJens Axboe int i; 15546dc4f100SMing Lei struct bvec_iter_all iter_all; 1555f9c78b2bSJens Axboe 15566dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 15571dfa0f68SChristoph Hellwig memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1558f9c78b2bSJens Axboe p += bvec->bv_len; 1559f9c78b2bSJens Axboe } 1560f9c78b2bSJens Axboe 15614246a0b6SChristoph Hellwig bio_copy_kern_endio(bio); 1562f9c78b2bSJens Axboe } 1563f9c78b2bSJens Axboe 1564f9c78b2bSJens Axboe /** 1565f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1566f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1567f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1568f9c78b2bSJens Axboe * @len: length in bytes 1569f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1570f9c78b2bSJens Axboe * @reading: data direction is READ 1571f9c78b2bSJens Axboe * 1572f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1573f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1574f9c78b2bSJens Axboe */ 1575f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1576f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1577f9c78b2bSJens Axboe { 157842d2683aSChristoph Hellwig unsigned long kaddr = (unsigned long)data; 157942d2683aSChristoph Hellwig unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 158042d2683aSChristoph Hellwig unsigned long start = kaddr >> PAGE_SHIFT; 158142d2683aSChristoph Hellwig struct bio *bio; 1582f9c78b2bSJens Axboe void *p = data; 15831dfa0f68SChristoph Hellwig int nr_pages = 0; 1584f9c78b2bSJens Axboe 158542d2683aSChristoph Hellwig /* 158642d2683aSChristoph Hellwig * Overflow, abort 158742d2683aSChristoph Hellwig */ 158842d2683aSChristoph Hellwig if (end < start) 158942d2683aSChristoph Hellwig return ERR_PTR(-EINVAL); 1590f9c78b2bSJens Axboe 159142d2683aSChristoph Hellwig nr_pages = end - start; 159242d2683aSChristoph Hellwig bio = bio_kmalloc(gfp_mask, nr_pages); 159342d2683aSChristoph Hellwig if (!bio) 159442d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 159542d2683aSChristoph Hellwig 159642d2683aSChristoph Hellwig while (len) { 159742d2683aSChristoph Hellwig struct page *page; 159842d2683aSChristoph Hellwig unsigned int bytes = PAGE_SIZE; 159942d2683aSChristoph Hellwig 160042d2683aSChristoph Hellwig if (bytes > len) 160142d2683aSChristoph Hellwig bytes = len; 160242d2683aSChristoph Hellwig 160342d2683aSChristoph Hellwig page = alloc_page(q->bounce_gfp | gfp_mask); 160442d2683aSChristoph Hellwig if (!page) 160542d2683aSChristoph Hellwig goto cleanup; 160642d2683aSChristoph Hellwig 160742d2683aSChristoph Hellwig if (!reading) 160842d2683aSChristoph Hellwig memcpy(page_address(page), p, bytes); 160942d2683aSChristoph Hellwig 161042d2683aSChristoph Hellwig if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 161142d2683aSChristoph Hellwig break; 161242d2683aSChristoph Hellwig 161342d2683aSChristoph Hellwig len -= bytes; 161442d2683aSChristoph Hellwig p += bytes; 1615f9c78b2bSJens Axboe } 1616f9c78b2bSJens Axboe 16171dfa0f68SChristoph Hellwig if (reading) { 16181dfa0f68SChristoph Hellwig bio->bi_end_io = bio_copy_kern_endio_read; 161942d2683aSChristoph Hellwig bio->bi_private = data; 16201dfa0f68SChristoph Hellwig } else { 1621f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 16221dfa0f68SChristoph Hellwig } 16231dfa0f68SChristoph Hellwig 1624f9c78b2bSJens Axboe return bio; 162542d2683aSChristoph Hellwig 162642d2683aSChristoph Hellwig cleanup: 16271dfa0f68SChristoph Hellwig bio_free_pages(bio); 162842d2683aSChristoph Hellwig bio_put(bio); 162942d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 1630f9c78b2bSJens Axboe } 1631f9c78b2bSJens Axboe 1632f9c78b2bSJens Axboe /* 1633f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1634f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1635f9c78b2bSJens Axboe * 1636f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1637f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1638f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1639f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1640f9c78b2bSJens Axboe * in process context. 1641f9c78b2bSJens Axboe * 1642f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1643f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1644f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1645f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1646f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1647f9c78b2bSJens Axboe * 1648f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1649f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1650f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1651f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1652f9c78b2bSJens Axboe * pagecache. 1653f9c78b2bSJens Axboe * 1654f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1655f9c78b2bSJens Axboe * deferred bio dirtying paths. 1656f9c78b2bSJens Axboe */ 1657f9c78b2bSJens Axboe 1658f9c78b2bSJens Axboe /* 1659f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1660f9c78b2bSJens Axboe */ 1661f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1662f9c78b2bSJens Axboe { 1663f9c78b2bSJens Axboe struct bio_vec *bvec; 1664f9c78b2bSJens Axboe int i; 16656dc4f100SMing Lei struct bvec_iter_all iter_all; 1666f9c78b2bSJens Axboe 16676dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 16683bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 16693bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1670f9c78b2bSJens Axboe } 1671f9c78b2bSJens Axboe } 1672f9c78b2bSJens Axboe 1673f9c78b2bSJens Axboe static void bio_release_pages(struct bio *bio) 1674f9c78b2bSJens Axboe { 1675f9c78b2bSJens Axboe struct bio_vec *bvec; 1676f9c78b2bSJens Axboe int i; 16776dc4f100SMing Lei struct bvec_iter_all iter_all; 1678f9c78b2bSJens Axboe 16796dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) 168024d5493fSChristoph Hellwig put_page(bvec->bv_page); 1681f9c78b2bSJens Axboe } 1682f9c78b2bSJens Axboe 1683f9c78b2bSJens Axboe /* 1684f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1685f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1686f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 168724d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1688f9c78b2bSJens Axboe * 1689f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1690ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1691ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1692f9c78b2bSJens Axboe */ 1693f9c78b2bSJens Axboe 1694f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1695f9c78b2bSJens Axboe 1696f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1697f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1698f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1699f9c78b2bSJens Axboe 1700f9c78b2bSJens Axboe /* 1701f9c78b2bSJens Axboe * This runs in process context 1702f9c78b2bSJens Axboe */ 1703f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1704f9c78b2bSJens Axboe { 170524d5493fSChristoph Hellwig struct bio *bio, *next; 1706f9c78b2bSJens Axboe 170724d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 170824d5493fSChristoph Hellwig next = bio_dirty_list; 1709f9c78b2bSJens Axboe bio_dirty_list = NULL; 171024d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1711f9c78b2bSJens Axboe 171224d5493fSChristoph Hellwig while ((bio = next) != NULL) { 171324d5493fSChristoph Hellwig next = bio->bi_private; 1714f9c78b2bSJens Axboe 1715f9c78b2bSJens Axboe bio_set_pages_dirty(bio); 1716399254aaSJens Axboe if (!bio_flagged(bio, BIO_NO_PAGE_REF)) 1717f9c78b2bSJens Axboe bio_release_pages(bio); 1718f9c78b2bSJens Axboe bio_put(bio); 1719f9c78b2bSJens Axboe } 1720f9c78b2bSJens Axboe } 1721f9c78b2bSJens Axboe 1722f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1723f9c78b2bSJens Axboe { 1724f9c78b2bSJens Axboe struct bio_vec *bvec; 172524d5493fSChristoph Hellwig unsigned long flags; 1726f9c78b2bSJens Axboe int i; 17276dc4f100SMing Lei struct bvec_iter_all iter_all; 1728f9c78b2bSJens Axboe 17296dc4f100SMing Lei bio_for_each_segment_all(bvec, bio, i, iter_all) { 173024d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 173124d5493fSChristoph Hellwig goto defer; 1732f9c78b2bSJens Axboe } 1733f9c78b2bSJens Axboe 1734399254aaSJens Axboe if (!bio_flagged(bio, BIO_NO_PAGE_REF)) 173524d5493fSChristoph Hellwig bio_release_pages(bio); 173624d5493fSChristoph Hellwig bio_put(bio); 173724d5493fSChristoph Hellwig return; 173824d5493fSChristoph Hellwig defer: 1739f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1740f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1741f9c78b2bSJens Axboe bio_dirty_list = bio; 1742f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1743f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1744f9c78b2bSJens Axboe } 1745f9c78b2bSJens Axboe 17465b18b5a7SMikulas Patocka void update_io_ticks(struct hd_struct *part, unsigned long now) 17475b18b5a7SMikulas Patocka { 17485b18b5a7SMikulas Patocka unsigned long stamp; 17495b18b5a7SMikulas Patocka again: 17505b18b5a7SMikulas Patocka stamp = READ_ONCE(part->stamp); 17515b18b5a7SMikulas Patocka if (unlikely(stamp != now)) { 17525b18b5a7SMikulas Patocka if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 17535b18b5a7SMikulas Patocka __part_stat_add(part, io_ticks, 1); 17545b18b5a7SMikulas Patocka } 17555b18b5a7SMikulas Patocka } 17565b18b5a7SMikulas Patocka if (part->partno) { 17575b18b5a7SMikulas Patocka part = &part_to_disk(part)->part0; 17585b18b5a7SMikulas Patocka goto again; 17595b18b5a7SMikulas Patocka } 17605b18b5a7SMikulas Patocka } 1761f9c78b2bSJens Axboe 1762ddcf35d3SMichael Callahan void generic_start_io_acct(struct request_queue *q, int op, 1763d62e26b3SJens Axboe unsigned long sectors, struct hd_struct *part) 1764394ffa50SGu Zheng { 1765ddcf35d3SMichael Callahan const int sgrp = op_stat_group(op); 1766394ffa50SGu Zheng 1767112f158fSMike Snitzer part_stat_lock(); 1768112f158fSMike Snitzer 17695b18b5a7SMikulas Patocka update_io_ticks(part, jiffies); 1770112f158fSMike Snitzer part_stat_inc(part, ios[sgrp]); 1771112f158fSMike Snitzer part_stat_add(part, sectors[sgrp], sectors); 1772ddcf35d3SMichael Callahan part_inc_in_flight(q, part, op_is_write(op)); 1773394ffa50SGu Zheng 1774394ffa50SGu Zheng part_stat_unlock(); 1775394ffa50SGu Zheng } 1776394ffa50SGu Zheng EXPORT_SYMBOL(generic_start_io_acct); 1777394ffa50SGu Zheng 1778ddcf35d3SMichael Callahan void generic_end_io_acct(struct request_queue *q, int req_op, 1779d62e26b3SJens Axboe struct hd_struct *part, unsigned long start_time) 1780394ffa50SGu Zheng { 17815b18b5a7SMikulas Patocka unsigned long now = jiffies; 17825b18b5a7SMikulas Patocka unsigned long duration = now - start_time; 1783ddcf35d3SMichael Callahan const int sgrp = op_stat_group(req_op); 1784394ffa50SGu Zheng 1785112f158fSMike Snitzer part_stat_lock(); 1786112f158fSMike Snitzer 17875b18b5a7SMikulas Patocka update_io_ticks(part, now); 1788112f158fSMike Snitzer part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 17895b18b5a7SMikulas Patocka part_stat_add(part, time_in_queue, duration); 1790ddcf35d3SMichael Callahan part_dec_in_flight(q, part, op_is_write(req_op)); 1791394ffa50SGu Zheng 1792394ffa50SGu Zheng part_stat_unlock(); 1793394ffa50SGu Zheng } 1794394ffa50SGu Zheng EXPORT_SYMBOL(generic_end_io_acct); 1795394ffa50SGu Zheng 1796f9c78b2bSJens Axboe #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1797f9c78b2bSJens Axboe void bio_flush_dcache_pages(struct bio *bi) 1798f9c78b2bSJens Axboe { 1799f9c78b2bSJens Axboe struct bio_vec bvec; 1800f9c78b2bSJens Axboe struct bvec_iter iter; 1801f9c78b2bSJens Axboe 1802f9c78b2bSJens Axboe bio_for_each_segment(bvec, bi, iter) 1803f9c78b2bSJens Axboe flush_dcache_page(bvec.bv_page); 1804f9c78b2bSJens Axboe } 1805f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_flush_dcache_pages); 1806f9c78b2bSJens Axboe #endif 1807f9c78b2bSJens Axboe 1808c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1809c4cf5261SJens Axboe { 1810c4cf5261SJens Axboe /* 1811c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1812c4cf5261SJens Axboe * we always end io on the first invocation. 1813c4cf5261SJens Axboe */ 1814c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1815c4cf5261SJens Axboe return true; 1816c4cf5261SJens Axboe 1817c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1818c4cf5261SJens Axboe 1819326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1820b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1821c4cf5261SJens Axboe return true; 1822326e1dbbSMike Snitzer } 1823c4cf5261SJens Axboe 1824c4cf5261SJens Axboe return false; 1825c4cf5261SJens Axboe } 1826c4cf5261SJens Axboe 1827f9c78b2bSJens Axboe /** 1828f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1829f9c78b2bSJens Axboe * @bio: bio 1830f9c78b2bSJens Axboe * 1831f9c78b2bSJens Axboe * Description: 18324246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 18334246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 18344246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1835fbbaf700SNeilBrown * 1836fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1837fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1838fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1839fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1840f9c78b2bSJens Axboe **/ 18414246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1842f9c78b2bSJens Axboe { 1843ba8c6967SChristoph Hellwig again: 18442b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1845ba8c6967SChristoph Hellwig return; 18467c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 18477c20f116SChristoph Hellwig return; 1848f9c78b2bSJens Axboe 184967b42d0bSJosef Bacik if (bio->bi_disk) 185067b42d0bSJosef Bacik rq_qos_done_bio(bio->bi_disk->queue, bio); 185167b42d0bSJosef Bacik 1852f9c78b2bSJens Axboe /* 1853ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1854ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1855ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1856ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1857ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1858ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1859f9c78b2bSJens Axboe */ 1860f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 186138f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1862ba8c6967SChristoph Hellwig goto again; 1863ba8c6967SChristoph Hellwig } 1864ba8c6967SChristoph Hellwig 186574d46992SChristoph Hellwig if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 186674d46992SChristoph Hellwig trace_block_bio_complete(bio->bi_disk->queue, bio, 1867a462b950SBart Van Assche blk_status_to_errno(bio->bi_status)); 1868fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1869fbbaf700SNeilBrown } 1870fbbaf700SNeilBrown 18719e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1872b222dd2fSShaohua Li /* release cgroup info */ 1873b222dd2fSShaohua Li bio_uninit(bio); 1874f9c78b2bSJens Axboe if (bio->bi_end_io) 18754246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1876f9c78b2bSJens Axboe } 1877f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1878f9c78b2bSJens Axboe 1879f9c78b2bSJens Axboe /** 1880f9c78b2bSJens Axboe * bio_split - split a bio 1881f9c78b2bSJens Axboe * @bio: bio to split 1882f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1883f9c78b2bSJens Axboe * @gfp: gfp mask 1884f9c78b2bSJens Axboe * @bs: bio set to allocate from 1885f9c78b2bSJens Axboe * 1886f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1887f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1888f9c78b2bSJens Axboe * 1889f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1890f3f5da62SMartin K. Petersen * to @bio's bi_io_vec; it is the caller's responsibility to ensure that 1891f3f5da62SMartin K. Petersen * @bio is not freed before the split. 1892f9c78b2bSJens Axboe */ 1893f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1894f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1895f9c78b2bSJens Axboe { 1896f341a4d3SMikulas Patocka struct bio *split; 1897f9c78b2bSJens Axboe 1898f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1899f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1900f9c78b2bSJens Axboe 1901f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1902f9c78b2bSJens Axboe if (!split) 1903f9c78b2bSJens Axboe return NULL; 1904f9c78b2bSJens Axboe 1905f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1906f9c78b2bSJens Axboe 1907f9c78b2bSJens Axboe if (bio_integrity(split)) 1908fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1909f9c78b2bSJens Axboe 1910f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1911f9c78b2bSJens Axboe 1912fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 191320d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1914fbbaf700SNeilBrown 1915f9c78b2bSJens Axboe return split; 1916f9c78b2bSJens Axboe } 1917f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1918f9c78b2bSJens Axboe 1919f9c78b2bSJens Axboe /** 1920f9c78b2bSJens Axboe * bio_trim - trim a bio 1921f9c78b2bSJens Axboe * @bio: bio to trim 1922f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1923f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1924f9c78b2bSJens Axboe */ 1925f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1926f9c78b2bSJens Axboe { 1927f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1928f9c78b2bSJens Axboe * the given offset and size. 1929f9c78b2bSJens Axboe */ 1930f9c78b2bSJens Axboe 1931f9c78b2bSJens Axboe size <<= 9; 1932f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1933f9c78b2bSJens Axboe return; 1934f9c78b2bSJens Axboe 1935b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_SEG_VALID); 1936f9c78b2bSJens Axboe 1937f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1938f9c78b2bSJens Axboe 1939f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1940376a78abSDmitry Monakhov 1941376a78abSDmitry Monakhov if (bio_integrity(bio)) 1942fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1943376a78abSDmitry Monakhov 1944f9c78b2bSJens Axboe } 1945f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1946f9c78b2bSJens Axboe 1947f9c78b2bSJens Axboe /* 1948f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1949f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1950f9c78b2bSJens Axboe */ 19518aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1952f9c78b2bSJens Axboe { 1953ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1954f9c78b2bSJens Axboe 19558aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1956f9c78b2bSJens Axboe } 1957f9c78b2bSJens Axboe 1958917a38c7SKent Overstreet /* 1959917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1960917a38c7SKent Overstreet * 1961917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1962917a38c7SKent Overstreet * kzalloc()). 1963917a38c7SKent Overstreet */ 1964917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1965f9c78b2bSJens Axboe { 1966f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1967f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1968917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1969f9c78b2bSJens Axboe 19708aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 19718aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1972f9c78b2bSJens Axboe 1973f9c78b2bSJens Axboe bioset_integrity_free(bs); 1974917a38c7SKent Overstreet if (bs->bio_slab) 1975f9c78b2bSJens Axboe bio_put_slab(bs); 1976917a38c7SKent Overstreet bs->bio_slab = NULL; 1977917a38c7SKent Overstreet } 1978917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1979f9c78b2bSJens Axboe 1980011067b0SNeilBrown /** 1981917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1982dad08527SKent Overstreet * @bs: pool to initialize 1983917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1984917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1985917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1986917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1987917a38c7SKent Overstreet * 1988dad08527SKent Overstreet * Description: 1989dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1990dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1991dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1992dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1993dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1994dad08527SKent Overstreet * or things will break badly. 1995dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1996dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1997dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1998dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1999dad08527SKent Overstreet * 2000917a38c7SKent Overstreet */ 2001917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 2002917a38c7SKent Overstreet unsigned int pool_size, 2003917a38c7SKent Overstreet unsigned int front_pad, 2004917a38c7SKent Overstreet int flags) 2005917a38c7SKent Overstreet { 2006917a38c7SKent Overstreet unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 2007917a38c7SKent Overstreet 2008917a38c7SKent Overstreet bs->front_pad = front_pad; 2009917a38c7SKent Overstreet 2010917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 2011917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 2012917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 2013917a38c7SKent Overstreet 2014917a38c7SKent Overstreet bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 2015917a38c7SKent Overstreet if (!bs->bio_slab) 2016917a38c7SKent Overstreet return -ENOMEM; 2017917a38c7SKent Overstreet 2018917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2019917a38c7SKent Overstreet goto bad; 2020917a38c7SKent Overstreet 2021917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 2022917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 2023917a38c7SKent Overstreet goto bad; 2024917a38c7SKent Overstreet 2025917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 2026917a38c7SKent Overstreet return 0; 2027917a38c7SKent Overstreet 2028917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2029917a38c7SKent Overstreet if (!bs->rescue_workqueue) 2030917a38c7SKent Overstreet goto bad; 2031917a38c7SKent Overstreet 2032917a38c7SKent Overstreet return 0; 2033917a38c7SKent Overstreet bad: 2034917a38c7SKent Overstreet bioset_exit(bs); 2035917a38c7SKent Overstreet return -ENOMEM; 2036917a38c7SKent Overstreet } 2037917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 2038917a38c7SKent Overstreet 203928e89fd9SJens Axboe /* 204028e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 204128e89fd9SJens Axboe * another bio_set. 204228e89fd9SJens Axboe */ 204328e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 204428e89fd9SJens Axboe { 204528e89fd9SJens Axboe int flags; 204628e89fd9SJens Axboe 204728e89fd9SJens Axboe flags = 0; 204828e89fd9SJens Axboe if (src->bvec_pool.min_nr) 204928e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 205028e89fd9SJens Axboe if (src->rescue_workqueue) 205128e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 205228e89fd9SJens Axboe 205328e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 205428e89fd9SJens Axboe } 205528e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 205628e89fd9SJens Axboe 2057f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 20581d933cf0STejun Heo 20591d933cf0STejun Heo /** 20602268c0feSDennis Zhou * bio_disassociate_blkg - puts back the blkg reference if associated 2061b5f2954dSDennis Zhou * @bio: target bio 2062b5f2954dSDennis Zhou * 20632268c0feSDennis Zhou * Helper to disassociate the blkg from @bio if a blkg is associated. 2064b5f2954dSDennis Zhou */ 20652268c0feSDennis Zhou void bio_disassociate_blkg(struct bio *bio) 2066b5f2954dSDennis Zhou { 206708e18eabSJosef Bacik if (bio->bi_blkg) { 206808e18eabSJosef Bacik blkg_put(bio->bi_blkg); 206908e18eabSJosef Bacik bio->bi_blkg = NULL; 207008e18eabSJosef Bacik } 2071f9c78b2bSJens Axboe } 2072892ad71fSDennis Zhou EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2073f9c78b2bSJens Axboe 207420bd723eSPaolo Valente /** 20752268c0feSDennis Zhou * __bio_associate_blkg - associate a bio with the a blkg 2076f9c78b2bSJens Axboe * @bio: target bio 2077f9c78b2bSJens Axboe * @blkg: the blkg to associate 2078f9c78b2bSJens Axboe * 2079beea9da0SDennis Zhou * This tries to associate @bio with the specified @blkg. Association failure 2080beea9da0SDennis Zhou * is handled by walking up the blkg tree. Therefore, the blkg associated can 2081beea9da0SDennis Zhou * be anything between @blkg and the root_blkg. This situation only happens 2082beea9da0SDennis Zhou * when a cgroup is dying and then the remaining bios will spill to the closest 2083beea9da0SDennis Zhou * alive blkg. 2084beea9da0SDennis Zhou * 2085beea9da0SDennis Zhou * A reference will be taken on the @blkg and will be released when @bio is 2086beea9da0SDennis Zhou * freed. 2087f9c78b2bSJens Axboe */ 20882268c0feSDennis Zhou static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2089f9c78b2bSJens Axboe { 20902268c0feSDennis Zhou bio_disassociate_blkg(bio); 20912268c0feSDennis Zhou 20927754f669SDennis Zhou bio->bi_blkg = blkg_tryget_closest(blkg); 20932268c0feSDennis Zhou } 20942268c0feSDennis Zhou 2095fd42df30SDennis Zhou /** 2096fd42df30SDennis Zhou * bio_associate_blkg_from_css - associate a bio with a specified css 2097fd42df30SDennis Zhou * @bio: target bio 2098fd42df30SDennis Zhou * @css: target css 2099fd42df30SDennis Zhou * 2100fd42df30SDennis Zhou * Associate @bio with the blkg found by combining the css's blkg and the 2101fc5a828bSDennis Zhou * request_queue of the @bio. This falls back to the queue's root_blkg if 2102fc5a828bSDennis Zhou * the association fails with the css. 2103fd42df30SDennis Zhou */ 2104fd42df30SDennis Zhou void bio_associate_blkg_from_css(struct bio *bio, 2105fd42df30SDennis Zhou struct cgroup_subsys_state *css) 2106fd42df30SDennis Zhou { 2107fc5a828bSDennis Zhou struct request_queue *q = bio->bi_disk->queue; 2108fc5a828bSDennis Zhou struct blkcg_gq *blkg; 2109fc5a828bSDennis Zhou 2110fc5a828bSDennis Zhou rcu_read_lock(); 2111fc5a828bSDennis Zhou 2112fc5a828bSDennis Zhou if (!css || !css->parent) 2113fc5a828bSDennis Zhou blkg = q->root_blkg; 2114fc5a828bSDennis Zhou else 2115fc5a828bSDennis Zhou blkg = blkg_lookup_create(css_to_blkcg(css), q); 2116fc5a828bSDennis Zhou 2117fc5a828bSDennis Zhou __bio_associate_blkg(bio, blkg); 2118fc5a828bSDennis Zhou 2119fc5a828bSDennis Zhou rcu_read_unlock(); 2120fd42df30SDennis Zhou } 2121fd42df30SDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2122fd42df30SDennis Zhou 21236a7f6d86SDennis Zhou #ifdef CONFIG_MEMCG 21246a7f6d86SDennis Zhou /** 21256a7f6d86SDennis Zhou * bio_associate_blkg_from_page - associate a bio with the page's blkg 21266a7f6d86SDennis Zhou * @bio: target bio 21276a7f6d86SDennis Zhou * @page: the page to lookup the blkcg from 21286a7f6d86SDennis Zhou * 21296a7f6d86SDennis Zhou * Associate @bio with the blkg from @page's owning memcg and the respective 2130fc5a828bSDennis Zhou * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2131fc5a828bSDennis Zhou * root_blkg. 21326a7f6d86SDennis Zhou */ 21336a7f6d86SDennis Zhou void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 21346a7f6d86SDennis Zhou { 21356a7f6d86SDennis Zhou struct cgroup_subsys_state *css; 21366a7f6d86SDennis Zhou 21376a7f6d86SDennis Zhou if (!page->mem_cgroup) 21386a7f6d86SDennis Zhou return; 21396a7f6d86SDennis Zhou 2140fc5a828bSDennis Zhou rcu_read_lock(); 2141fc5a828bSDennis Zhou 2142fc5a828bSDennis Zhou css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2143fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 2144fc5a828bSDennis Zhou 2145fc5a828bSDennis Zhou rcu_read_unlock(); 21466a7f6d86SDennis Zhou } 21476a7f6d86SDennis Zhou #endif /* CONFIG_MEMCG */ 21486a7f6d86SDennis Zhou 21492268c0feSDennis Zhou /** 21502268c0feSDennis Zhou * bio_associate_blkg - associate a bio with a blkg 21512268c0feSDennis Zhou * @bio: target bio 21522268c0feSDennis Zhou * 21532268c0feSDennis Zhou * Associate @bio with the blkg found from the bio's css and request_queue. 21542268c0feSDennis Zhou * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 21552268c0feSDennis Zhou * already associated, the css is reused and association redone as the 21562268c0feSDennis Zhou * request_queue may have changed. 21572268c0feSDennis Zhou */ 21582268c0feSDennis Zhou void bio_associate_blkg(struct bio *bio) 21592268c0feSDennis Zhou { 2160fc5a828bSDennis Zhou struct cgroup_subsys_state *css; 21612268c0feSDennis Zhou 21622268c0feSDennis Zhou rcu_read_lock(); 21632268c0feSDennis Zhou 2164db6638d7SDennis Zhou if (bio->bi_blkg) 2165fc5a828bSDennis Zhou css = &bio_blkcg(bio)->css; 2166db6638d7SDennis Zhou else 2167fc5a828bSDennis Zhou css = blkcg_css(); 21682268c0feSDennis Zhou 2169fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 21702268c0feSDennis Zhou 21712268c0feSDennis Zhou rcu_read_unlock(); 2172f9c78b2bSJens Axboe } 21735cdf2e3fSDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg); 2174f9c78b2bSJens Axboe 217520bd723eSPaolo Valente /** 2176db6638d7SDennis Zhou * bio_clone_blkg_association - clone blkg association from src to dst bio 217720bd723eSPaolo Valente * @dst: destination bio 217820bd723eSPaolo Valente * @src: source bio 217920bd723eSPaolo Valente */ 2180db6638d7SDennis Zhou void bio_clone_blkg_association(struct bio *dst, struct bio *src) 218120bd723eSPaolo Valente { 21826ab21879SDennis Zhou rcu_read_lock(); 21836ab21879SDennis Zhou 2184fc5a828bSDennis Zhou if (src->bi_blkg) 21852268c0feSDennis Zhou __bio_associate_blkg(dst, src->bi_blkg); 21866ab21879SDennis Zhou 21876ab21879SDennis Zhou rcu_read_unlock(); 218820bd723eSPaolo Valente } 2189db6638d7SDennis Zhou EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2190f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 2191f9c78b2bSJens Axboe 2192f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2193f9c78b2bSJens Axboe { 2194f9c78b2bSJens Axboe int i; 2195f9c78b2bSJens Axboe 2196ed996a52SChristoph Hellwig for (i = 0; i < BVEC_POOL_NR; i++) { 2197f9c78b2bSJens Axboe int size; 2198f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2199f9c78b2bSJens Axboe 2200f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2201f9c78b2bSJens Axboe bvs->slab = NULL; 2202f9c78b2bSJens Axboe continue; 2203f9c78b2bSJens Axboe } 2204f9c78b2bSJens Axboe 2205f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2206f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2207f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2208f9c78b2bSJens Axboe } 2209f9c78b2bSJens Axboe } 2210f9c78b2bSJens Axboe 2211f9c78b2bSJens Axboe static int __init init_bio(void) 2212f9c78b2bSJens Axboe { 2213f9c78b2bSJens Axboe bio_slab_max = 2; 2214f9c78b2bSJens Axboe bio_slab_nr = 0; 22156396bb22SKees Cook bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 22166396bb22SKees Cook GFP_KERNEL); 22172b24e6f6SJohannes Thumshirn 22182b24e6f6SJohannes Thumshirn BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 22192b24e6f6SJohannes Thumshirn 2220f9c78b2bSJens Axboe if (!bio_slabs) 2221f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2222f9c78b2bSJens Axboe 2223f9c78b2bSJens Axboe bio_integrity_init(); 2224f9c78b2bSJens Axboe biovec_init_slabs(); 2225f9c78b2bSJens Axboe 2226f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2227f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2228f9c78b2bSJens Axboe 2229f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2230f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2231f9c78b2bSJens Axboe 2232f9c78b2bSJens Axboe return 0; 2233f9c78b2bSJens Axboe } 2234f9c78b2bSJens Axboe subsys_initcall(init_bio); 2235