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 { 247f9c78b2bSJens Axboe bio_disassociate_task(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 574f9c78b2bSJens Axboe inline 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 EXPORT_SYMBOL(bio_phys_segments); 582f9c78b2bSJens Axboe 583f9c78b2bSJens Axboe /** 584f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 585f9c78b2bSJens Axboe * @bio: destination bio 586f9c78b2bSJens Axboe * @bio_src: bio to clone 587f9c78b2bSJens Axboe * 588f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 589f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 590f9c78b2bSJens Axboe * bio will be one. 591f9c78b2bSJens Axboe * 592f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 593f9c78b2bSJens Axboe */ 594f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 595f9c78b2bSJens Axboe { 596ed996a52SChristoph Hellwig BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 597f9c78b2bSJens Axboe 598f9c78b2bSJens Axboe /* 59974d46992SChristoph Hellwig * most users will be overriding ->bi_disk with a new target, 600f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 601f9c78b2bSJens Axboe */ 60274d46992SChristoph Hellwig bio->bi_disk = bio_src->bi_disk; 60362530ed8SMichael Lyle bio->bi_partno = bio_src->bi_partno; 604b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 605111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 606111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 6071eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 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 61220bd723eSPaolo Valente bio_clone_blkcg_association(bio, bio_src); 613f9c78b2bSJens Axboe } 614f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 615f9c78b2bSJens Axboe 616f9c78b2bSJens Axboe /** 617f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 618f9c78b2bSJens Axboe * @bio: bio to clone 619f9c78b2bSJens Axboe * @gfp_mask: allocation priority 620f9c78b2bSJens Axboe * @bs: bio_set to allocate from 621f9c78b2bSJens Axboe * 622f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 623f9c78b2bSJens Axboe */ 624f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 625f9c78b2bSJens Axboe { 626f9c78b2bSJens Axboe struct bio *b; 627f9c78b2bSJens Axboe 628f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 629f9c78b2bSJens Axboe if (!b) 630f9c78b2bSJens Axboe return NULL; 631f9c78b2bSJens Axboe 632f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 633f9c78b2bSJens Axboe 634f9c78b2bSJens Axboe if (bio_integrity(bio)) { 635f9c78b2bSJens Axboe int ret; 636f9c78b2bSJens Axboe 637f9c78b2bSJens Axboe ret = bio_integrity_clone(b, bio, gfp_mask); 638f9c78b2bSJens Axboe 639f9c78b2bSJens Axboe if (ret < 0) { 640f9c78b2bSJens Axboe bio_put(b); 641f9c78b2bSJens Axboe return NULL; 642f9c78b2bSJens Axboe } 643f9c78b2bSJens Axboe } 644f9c78b2bSJens Axboe 645f9c78b2bSJens Axboe return b; 646f9c78b2bSJens Axboe } 647f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 648f9c78b2bSJens Axboe 649f4595875SShaohua Li /** 650c66a14d0SKent Overstreet * bio_add_pc_page - attempt to add page to bio 651c66a14d0SKent Overstreet * @q: the target queue 652c66a14d0SKent Overstreet * @bio: destination bio 653c66a14d0SKent Overstreet * @page: page to add 654c66a14d0SKent Overstreet * @len: vec entry length 655c66a14d0SKent Overstreet * @offset: vec entry offset 656f9c78b2bSJens Axboe * 657c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This can fail for a 658c66a14d0SKent Overstreet * number of reasons, such as the bio being full or target block device 659c66a14d0SKent Overstreet * limitations. The target block device must allow bio's up to PAGE_SIZE, 660c66a14d0SKent Overstreet * so it is always possible to add a single page to an empty bio. 661c66a14d0SKent Overstreet * 662c66a14d0SKent Overstreet * This should only be used by REQ_PC bios. 663f9c78b2bSJens Axboe */ 664c66a14d0SKent Overstreet int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page 665c66a14d0SKent Overstreet *page, unsigned int len, unsigned int offset) 666f9c78b2bSJens Axboe { 667f9c78b2bSJens Axboe int retried_segments = 0; 668f9c78b2bSJens Axboe struct bio_vec *bvec; 669f9c78b2bSJens Axboe 670f9c78b2bSJens Axboe /* 671f9c78b2bSJens Axboe * cloned bio must not modify vec list 672f9c78b2bSJens Axboe */ 673f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 674f9c78b2bSJens Axboe return 0; 675f9c78b2bSJens Axboe 676c66a14d0SKent Overstreet if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 677f9c78b2bSJens Axboe return 0; 678f9c78b2bSJens Axboe 679f9c78b2bSJens Axboe /* 680f9c78b2bSJens Axboe * For filesystems with a blocksize smaller than the pagesize 681f9c78b2bSJens Axboe * we will often be called with the same page as last time and 682f9c78b2bSJens Axboe * a consecutive offset. Optimize this special case. 683f9c78b2bSJens Axboe */ 684f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 685f9c78b2bSJens Axboe struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; 686f9c78b2bSJens Axboe 687f9c78b2bSJens Axboe if (page == prev->bv_page && 688f9c78b2bSJens Axboe offset == prev->bv_offset + prev->bv_len) { 689f9c78b2bSJens Axboe prev->bv_len += len; 690fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 691f9c78b2bSJens Axboe goto done; 692f9c78b2bSJens Axboe } 69366cb45aaSJens Axboe 69466cb45aaSJens Axboe /* 69566cb45aaSJens Axboe * If the queue doesn't support SG gaps and adding this 69666cb45aaSJens Axboe * offset would create a gap, disallow it. 69766cb45aaSJens Axboe */ 69803100aadSKeith Busch if (bvec_gap_to_prev(q, prev, offset)) 69966cb45aaSJens Axboe return 0; 700f9c78b2bSJens Axboe } 701f9c78b2bSJens Axboe 7020aa69fd3SChristoph Hellwig if (bio_full(bio)) 703f9c78b2bSJens Axboe return 0; 704f9c78b2bSJens Axboe 705f9c78b2bSJens Axboe /* 706f9c78b2bSJens Axboe * setup the new entry, we might clear it again later if we 707f9c78b2bSJens Axboe * cannot add the page 708f9c78b2bSJens Axboe */ 709f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 710f9c78b2bSJens Axboe bvec->bv_page = page; 711f9c78b2bSJens Axboe bvec->bv_len = len; 712f9c78b2bSJens Axboe bvec->bv_offset = offset; 713fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 714fcbf6a08SMaurizio Lombardi bio->bi_phys_segments++; 715fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 716fcbf6a08SMaurizio Lombardi 717fcbf6a08SMaurizio Lombardi /* 718fcbf6a08SMaurizio Lombardi * Perform a recount if the number of segments is greater 719fcbf6a08SMaurizio Lombardi * than queue_max_segments(q). 720fcbf6a08SMaurizio Lombardi */ 721fcbf6a08SMaurizio Lombardi 722fcbf6a08SMaurizio Lombardi while (bio->bi_phys_segments > queue_max_segments(q)) { 723fcbf6a08SMaurizio Lombardi 724fcbf6a08SMaurizio Lombardi if (retried_segments) 725fcbf6a08SMaurizio Lombardi goto failed; 726fcbf6a08SMaurizio Lombardi 727fcbf6a08SMaurizio Lombardi retried_segments = 1; 728fcbf6a08SMaurizio Lombardi blk_recount_segments(q, bio); 729fcbf6a08SMaurizio Lombardi } 730f9c78b2bSJens Axboe 731f9c78b2bSJens Axboe /* If we may be able to merge these biovecs, force a recount */ 732fcbf6a08SMaurizio Lombardi if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) 733b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_SEG_VALID); 734f9c78b2bSJens Axboe 735f9c78b2bSJens Axboe done: 736f9c78b2bSJens Axboe return len; 737fcbf6a08SMaurizio Lombardi 738fcbf6a08SMaurizio Lombardi failed: 739fcbf6a08SMaurizio Lombardi bvec->bv_page = NULL; 740fcbf6a08SMaurizio Lombardi bvec->bv_len = 0; 741fcbf6a08SMaurizio Lombardi bvec->bv_offset = 0; 742fcbf6a08SMaurizio Lombardi bio->bi_vcnt--; 743fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size -= len; 744fcbf6a08SMaurizio Lombardi blk_recount_segments(q, bio); 745fcbf6a08SMaurizio Lombardi return 0; 746f9c78b2bSJens Axboe } 747f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 748f9c78b2bSJens Axboe 749f9c78b2bSJens Axboe /** 7500aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 7510aa69fd3SChristoph Hellwig * @bio: destination bio 7520aa69fd3SChristoph Hellwig * @page: page to add 7530aa69fd3SChristoph Hellwig * @len: length of the data to add 7540aa69fd3SChristoph Hellwig * @off: offset of the data in @page 7550aa69fd3SChristoph Hellwig * 7560aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 7570aa69fd3SChristoph Hellwig * a useful optimisation for file systems with a block size smaller than the 7580aa69fd3SChristoph Hellwig * page size. 7590aa69fd3SChristoph Hellwig * 7600aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 7610aa69fd3SChristoph Hellwig */ 7620aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 7630aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 7640aa69fd3SChristoph Hellwig { 7650aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 7660aa69fd3SChristoph Hellwig return false; 7670aa69fd3SChristoph Hellwig 7680aa69fd3SChristoph Hellwig if (bio->bi_vcnt > 0) { 7690aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 7700aa69fd3SChristoph Hellwig 7710aa69fd3SChristoph Hellwig if (page == bv->bv_page && off == bv->bv_offset + bv->bv_len) { 7720aa69fd3SChristoph Hellwig bv->bv_len += len; 7730aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 7740aa69fd3SChristoph Hellwig return true; 7750aa69fd3SChristoph Hellwig } 7760aa69fd3SChristoph Hellwig } 7770aa69fd3SChristoph Hellwig return false; 7780aa69fd3SChristoph Hellwig } 7790aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 7800aa69fd3SChristoph Hellwig 7810aa69fd3SChristoph Hellwig /** 7820aa69fd3SChristoph Hellwig * __bio_add_page - add page to a bio in a new segment 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 7870aa69fd3SChristoph Hellwig * 7880aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 7890aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 7900aa69fd3SChristoph Hellwig */ 7910aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 7920aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 7930aa69fd3SChristoph Hellwig { 7940aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 7950aa69fd3SChristoph Hellwig 7960aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 7970aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_full(bio)); 7980aa69fd3SChristoph Hellwig 7990aa69fd3SChristoph Hellwig bv->bv_page = page; 8000aa69fd3SChristoph Hellwig bv->bv_offset = off; 8010aa69fd3SChristoph Hellwig bv->bv_len = len; 8020aa69fd3SChristoph Hellwig 8030aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8040aa69fd3SChristoph Hellwig bio->bi_vcnt++; 8050aa69fd3SChristoph Hellwig } 8060aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 8070aa69fd3SChristoph Hellwig 8080aa69fd3SChristoph Hellwig /** 809f9c78b2bSJens Axboe * bio_add_page - attempt to add page to bio 810f9c78b2bSJens Axboe * @bio: destination bio 811f9c78b2bSJens Axboe * @page: page to add 812f9c78b2bSJens Axboe * @len: vec entry length 813f9c78b2bSJens Axboe * @offset: vec entry offset 814f9c78b2bSJens Axboe * 815c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This will only fail 816c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 817f9c78b2bSJens Axboe */ 818c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 819c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 820f9c78b2bSJens Axboe { 8210aa69fd3SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset)) { 8220aa69fd3SChristoph Hellwig if (bio_full(bio)) 823c66a14d0SKent Overstreet return 0; 8240aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 825c66a14d0SKent Overstreet } 826c66a14d0SKent Overstreet return len; 827f9c78b2bSJens Axboe } 828f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 829f9c78b2bSJens Axboe 8302cefe4dbSKent Overstreet /** 8312cefe4dbSKent Overstreet * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 8322cefe4dbSKent Overstreet * @bio: bio to add pages to 8332cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 8342cefe4dbSKent Overstreet * 8352cefe4dbSKent Overstreet * Pins as many pages from *iter and appends them to @bio's bvec array. The 8362cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 8372cefe4dbSKent Overstreet */ 8382cefe4dbSKent Overstreet int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 8392cefe4dbSKent Overstreet { 8402cefe4dbSKent Overstreet unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 8412cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 8422cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 8432cefe4dbSKent Overstreet size_t offset, diff; 8442cefe4dbSKent Overstreet ssize_t size; 8452cefe4dbSKent Overstreet 8462cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 8472cefe4dbSKent Overstreet if (unlikely(size <= 0)) 8482cefe4dbSKent Overstreet return size ? size : -EFAULT; 8492cefe4dbSKent Overstreet nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE; 8502cefe4dbSKent Overstreet 8512cefe4dbSKent Overstreet /* 8522cefe4dbSKent Overstreet * Deep magic below: We need to walk the pinned pages backwards 8532cefe4dbSKent Overstreet * because we are abusing the space allocated for the bio_vecs 8542cefe4dbSKent Overstreet * for the page array. Because the bio_vecs are larger than the 8552cefe4dbSKent Overstreet * page pointers by definition this will always work. But it also 8562cefe4dbSKent Overstreet * means we can't use bio_add_page, so any changes to it's semantics 8572cefe4dbSKent Overstreet * need to be reflected here as well. 8582cefe4dbSKent Overstreet */ 8592cefe4dbSKent Overstreet bio->bi_iter.bi_size += size; 8602cefe4dbSKent Overstreet bio->bi_vcnt += nr_pages; 8612cefe4dbSKent Overstreet 8622cefe4dbSKent Overstreet diff = (nr_pages * PAGE_SIZE - offset) - size; 8632cefe4dbSKent Overstreet while (nr_pages--) { 8642cefe4dbSKent Overstreet bv[nr_pages].bv_page = pages[nr_pages]; 8652cefe4dbSKent Overstreet bv[nr_pages].bv_len = PAGE_SIZE; 8662cefe4dbSKent Overstreet bv[nr_pages].bv_offset = 0; 8672cefe4dbSKent Overstreet } 8682cefe4dbSKent Overstreet 8692cefe4dbSKent Overstreet bv[0].bv_offset += offset; 8702cefe4dbSKent Overstreet bv[0].bv_len -= offset; 8712cefe4dbSKent Overstreet if (diff) 8722cefe4dbSKent Overstreet bv[bio->bi_vcnt - 1].bv_len -= diff; 8732cefe4dbSKent Overstreet 8742cefe4dbSKent Overstreet iov_iter_advance(iter, size); 8752cefe4dbSKent Overstreet return 0; 8762cefe4dbSKent Overstreet } 8772cefe4dbSKent Overstreet EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages); 8782cefe4dbSKent Overstreet 8794246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 880f9c78b2bSJens Axboe { 88165e53aabSChristoph Hellwig complete(bio->bi_private); 882f9c78b2bSJens Axboe } 883f9c78b2bSJens Axboe 884f9c78b2bSJens Axboe /** 885f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 886f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 887f9c78b2bSJens Axboe * 888f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 889f9c78b2bSJens Axboe * bio_endio() on failure. 8903d289d68SJan Kara * 8913d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 8923d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 8933d289d68SJan Kara * on his own. 894f9c78b2bSJens Axboe */ 8954e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 896f9c78b2bSJens Axboe { 897e319e1fbSByungchul Park DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 898f9c78b2bSJens Axboe 89965e53aabSChristoph Hellwig bio->bi_private = &done; 900f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 9011eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 9024e49ea4aSMike Christie submit_bio(bio); 90365e53aabSChristoph Hellwig wait_for_completion_io(&done); 904f9c78b2bSJens Axboe 90565e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 906f9c78b2bSJens Axboe } 907f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 908f9c78b2bSJens Axboe 909f9c78b2bSJens Axboe /** 910f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 911f9c78b2bSJens Axboe * @bio: bio to advance 912f9c78b2bSJens Axboe * @bytes: number of bytes to complete 913f9c78b2bSJens Axboe * 914f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 915f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 916f9c78b2bSJens Axboe * be updated on the last bvec as well. 917f9c78b2bSJens Axboe * 918f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 919f9c78b2bSJens Axboe */ 920f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 921f9c78b2bSJens Axboe { 922f9c78b2bSJens Axboe if (bio_integrity(bio)) 923f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 924f9c78b2bSJens Axboe 925f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 926f9c78b2bSJens Axboe } 927f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 928f9c78b2bSJens Axboe 92945db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 93045db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 931f9c78b2bSJens Axboe { 932f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 933f9c78b2bSJens Axboe void *src_p, *dst_p; 934f9c78b2bSJens Axboe unsigned bytes; 935f9c78b2bSJens Axboe 93645db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 93745db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 93845db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 93945db54d5SKent Overstreet 94045db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 94145db54d5SKent Overstreet 94245db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 94345db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 94445db54d5SKent Overstreet 94545db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 94645db54d5SKent Overstreet src_p + src_bv.bv_offset, 94745db54d5SKent Overstreet bytes); 94845db54d5SKent Overstreet 94945db54d5SKent Overstreet kunmap_atomic(dst_p); 95045db54d5SKent Overstreet kunmap_atomic(src_p); 95145db54d5SKent Overstreet 9526e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 9536e6e811dSKent Overstreet 95445db54d5SKent Overstreet bio_advance_iter(src, src_iter, bytes); 95545db54d5SKent Overstreet bio_advance_iter(dst, dst_iter, bytes); 95645db54d5SKent Overstreet } 95745db54d5SKent Overstreet } 95845db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 95945db54d5SKent Overstreet 96045db54d5SKent Overstreet /** 96145db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 96245db54d5SKent Overstreet * @src: source bio 96345db54d5SKent Overstreet * @dst: destination bio 96445db54d5SKent Overstreet * 96545db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 96645db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 96745db54d5SKent Overstreet */ 96845db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 96945db54d5SKent Overstreet { 97045db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 97145db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 97245db54d5SKent Overstreet 97345db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 97445db54d5SKent Overstreet } 97545db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 97645db54d5SKent Overstreet 97745db54d5SKent Overstreet /** 97845db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 97945db54d5SKent Overstreet * another 98045db54d5SKent Overstreet * @src: source bio list 98145db54d5SKent Overstreet * @dst: destination bio list 98245db54d5SKent Overstreet * 98345db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 98445db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 98545db54d5SKent Overstreet * bios). 98645db54d5SKent Overstreet */ 98745db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 98845db54d5SKent Overstreet { 98945db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 99045db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 99145db54d5SKent Overstreet 992f9c78b2bSJens Axboe while (1) { 993f9c78b2bSJens Axboe if (!src_iter.bi_size) { 994f9c78b2bSJens Axboe src = src->bi_next; 995f9c78b2bSJens Axboe if (!src) 996f9c78b2bSJens Axboe break; 997f9c78b2bSJens Axboe 998f9c78b2bSJens Axboe src_iter = src->bi_iter; 999f9c78b2bSJens Axboe } 1000f9c78b2bSJens Axboe 1001f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1002f9c78b2bSJens Axboe dst = dst->bi_next; 1003f9c78b2bSJens Axboe if (!dst) 1004f9c78b2bSJens Axboe break; 1005f9c78b2bSJens Axboe 1006f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1007f9c78b2bSJens Axboe } 1008f9c78b2bSJens Axboe 100945db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1010f9c78b2bSJens Axboe } 1011f9c78b2bSJens Axboe } 101245db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1013f9c78b2bSJens Axboe 1014f9c78b2bSJens Axboe struct bio_map_data { 1015f9c78b2bSJens Axboe int is_our_pages; 101626e49cfcSKent Overstreet struct iov_iter iter; 101726e49cfcSKent Overstreet struct iovec iov[]; 1018f9c78b2bSJens Axboe }; 1019f9c78b2bSJens Axboe 10200e5b935dSAl Viro static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1021f9c78b2bSJens Axboe gfp_t gfp_mask) 1022f9c78b2bSJens Axboe { 10230e5b935dSAl Viro struct bio_map_data *bmd; 10240e5b935dSAl Viro if (data->nr_segs > UIO_MAXIOV) 1025f9c78b2bSJens Axboe return NULL; 1026f9c78b2bSJens Axboe 10270e5b935dSAl Viro bmd = kmalloc(sizeof(struct bio_map_data) + 10280e5b935dSAl Viro sizeof(struct iovec) * data->nr_segs, gfp_mask); 10290e5b935dSAl Viro if (!bmd) 10300e5b935dSAl Viro return NULL; 10310e5b935dSAl Viro memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 10320e5b935dSAl Viro bmd->iter = *data; 10330e5b935dSAl Viro bmd->iter.iov = bmd->iov; 10340e5b935dSAl Viro return bmd; 1035f9c78b2bSJens Axboe } 1036f9c78b2bSJens Axboe 10379124d3feSDongsu Park /** 10389124d3feSDongsu Park * bio_copy_from_iter - copy all pages from iov_iter to bio 10399124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as destination 10409124d3feSDongsu Park * @iter: iov_iter as source 10419124d3feSDongsu Park * 10429124d3feSDongsu Park * Copy all pages from iov_iter to bio. 10439124d3feSDongsu Park * Returns 0 on success, or error on failure. 10449124d3feSDongsu Park */ 104598a09d61SAl Viro static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1046f9c78b2bSJens Axboe { 10479124d3feSDongsu Park int i; 1048f9c78b2bSJens Axboe struct bio_vec *bvec; 1049f9c78b2bSJens Axboe 1050f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 10519124d3feSDongsu Park ssize_t ret; 1052f9c78b2bSJens Axboe 10539124d3feSDongsu Park ret = copy_page_from_iter(bvec->bv_page, 10549124d3feSDongsu Park bvec->bv_offset, 10559124d3feSDongsu Park bvec->bv_len, 105698a09d61SAl Viro iter); 1057f9c78b2bSJens Axboe 105898a09d61SAl Viro if (!iov_iter_count(iter)) 10599124d3feSDongsu Park break; 1060f9c78b2bSJens Axboe 10619124d3feSDongsu Park if (ret < bvec->bv_len) 10629124d3feSDongsu Park return -EFAULT; 1063f9c78b2bSJens Axboe } 1064f9c78b2bSJens Axboe 10659124d3feSDongsu Park return 0; 1066f9c78b2bSJens Axboe } 1067f9c78b2bSJens Axboe 10689124d3feSDongsu Park /** 10699124d3feSDongsu Park * bio_copy_to_iter - copy all pages from bio to iov_iter 10709124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as source 10719124d3feSDongsu Park * @iter: iov_iter as destination 10729124d3feSDongsu Park * 10739124d3feSDongsu Park * Copy all pages from bio to iov_iter. 10749124d3feSDongsu Park * Returns 0 on success, or error on failure. 10759124d3feSDongsu Park */ 10769124d3feSDongsu Park static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 10779124d3feSDongsu Park { 10789124d3feSDongsu Park int i; 10799124d3feSDongsu Park struct bio_vec *bvec; 10809124d3feSDongsu Park 10819124d3feSDongsu Park bio_for_each_segment_all(bvec, bio, i) { 10829124d3feSDongsu Park ssize_t ret; 10839124d3feSDongsu Park 10849124d3feSDongsu Park ret = copy_page_to_iter(bvec->bv_page, 10859124d3feSDongsu Park bvec->bv_offset, 10869124d3feSDongsu Park bvec->bv_len, 10879124d3feSDongsu Park &iter); 10889124d3feSDongsu Park 10899124d3feSDongsu Park if (!iov_iter_count(&iter)) 10909124d3feSDongsu Park break; 10919124d3feSDongsu Park 10929124d3feSDongsu Park if (ret < bvec->bv_len) 10939124d3feSDongsu Park return -EFAULT; 10949124d3feSDongsu Park } 10959124d3feSDongsu Park 10969124d3feSDongsu Park return 0; 1097f9c78b2bSJens Axboe } 1098f9c78b2bSJens Axboe 1099491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 11001dfa0f68SChristoph Hellwig { 11011dfa0f68SChristoph Hellwig struct bio_vec *bvec; 11021dfa0f68SChristoph Hellwig int i; 11031dfa0f68SChristoph Hellwig 11041dfa0f68SChristoph Hellwig bio_for_each_segment_all(bvec, bio, i) 11051dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 11061dfa0f68SChristoph Hellwig } 1107491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 11081dfa0f68SChristoph Hellwig 1109f9c78b2bSJens Axboe /** 1110f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1111f9c78b2bSJens Axboe * @bio: bio being terminated 1112f9c78b2bSJens Axboe * 1113ddad8dd0SChristoph Hellwig * Free pages allocated from bio_copy_user_iov() and write back data 1114f9c78b2bSJens Axboe * to user space in case of a read. 1115f9c78b2bSJens Axboe */ 1116f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1117f9c78b2bSJens Axboe { 1118f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 11191dfa0f68SChristoph Hellwig int ret = 0; 1120f9c78b2bSJens Axboe 1121f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1122f9c78b2bSJens Axboe /* 1123f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 11242d99b55dSHannes Reinecke * don't copy into a random user address space, just free 11252d99b55dSHannes Reinecke * and return -EINTR so user space doesn't expect any data. 1126f9c78b2bSJens Axboe */ 11272d99b55dSHannes Reinecke if (!current->mm) 11282d99b55dSHannes Reinecke ret = -EINTR; 11292d99b55dSHannes Reinecke else if (bio_data_dir(bio) == READ) 11309124d3feSDongsu Park ret = bio_copy_to_iter(bio, bmd->iter); 11311dfa0f68SChristoph Hellwig if (bmd->is_our_pages) 11321dfa0f68SChristoph Hellwig bio_free_pages(bio); 1133f9c78b2bSJens Axboe } 1134f9c78b2bSJens Axboe kfree(bmd); 1135f9c78b2bSJens Axboe bio_put(bio); 1136f9c78b2bSJens Axboe return ret; 1137f9c78b2bSJens Axboe } 1138f9c78b2bSJens Axboe 1139f9c78b2bSJens Axboe /** 1140f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1141f9c78b2bSJens Axboe * @q: destination block queue 1142f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 114326e49cfcSKent Overstreet * @iter: iovec iterator 1144f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1145f9c78b2bSJens Axboe * 1146f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1147f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1148f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1149f9c78b2bSJens Axboe */ 1150f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1151f9c78b2bSJens Axboe struct rq_map_data *map_data, 1152e81cef5dSAl Viro struct iov_iter *iter, 115326e49cfcSKent Overstreet gfp_t gfp_mask) 1154f9c78b2bSJens Axboe { 1155f9c78b2bSJens Axboe struct bio_map_data *bmd; 1156f9c78b2bSJens Axboe struct page *page; 1157f9c78b2bSJens Axboe struct bio *bio; 1158d16d44ebSAl Viro int i = 0, ret; 1159d16d44ebSAl Viro int nr_pages; 116026e49cfcSKent Overstreet unsigned int len = iter->count; 1161bd5ceceaSGeliang Tang unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1162f9c78b2bSJens Axboe 11630e5b935dSAl Viro bmd = bio_alloc_map_data(iter, gfp_mask); 1164f9c78b2bSJens Axboe if (!bmd) 1165f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1166f9c78b2bSJens Axboe 116726e49cfcSKent Overstreet /* 116826e49cfcSKent Overstreet * We need to do a deep copy of the iov_iter including the iovecs. 116926e49cfcSKent Overstreet * The caller provided iov might point to an on-stack or otherwise 117026e49cfcSKent Overstreet * shortlived one. 117126e49cfcSKent Overstreet */ 117226e49cfcSKent Overstreet bmd->is_our_pages = map_data ? 0 : 1; 117326e49cfcSKent Overstreet 1174d16d44ebSAl Viro nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1175d16d44ebSAl Viro if (nr_pages > BIO_MAX_PAGES) 1176d16d44ebSAl Viro nr_pages = BIO_MAX_PAGES; 1177f9c78b2bSJens Axboe 1178f9c78b2bSJens Axboe ret = -ENOMEM; 1179f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1180f9c78b2bSJens Axboe if (!bio) 1181f9c78b2bSJens Axboe goto out_bmd; 1182f9c78b2bSJens Axboe 1183f9c78b2bSJens Axboe ret = 0; 1184f9c78b2bSJens Axboe 1185f9c78b2bSJens Axboe if (map_data) { 1186f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1187f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1188f9c78b2bSJens Axboe } 1189f9c78b2bSJens Axboe while (len) { 1190f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1191f9c78b2bSJens Axboe 1192f9c78b2bSJens Axboe bytes -= offset; 1193f9c78b2bSJens Axboe 1194f9c78b2bSJens Axboe if (bytes > len) 1195f9c78b2bSJens Axboe bytes = len; 1196f9c78b2bSJens Axboe 1197f9c78b2bSJens Axboe if (map_data) { 1198f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1199f9c78b2bSJens Axboe ret = -ENOMEM; 1200f9c78b2bSJens Axboe break; 1201f9c78b2bSJens Axboe } 1202f9c78b2bSJens Axboe 1203f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1204f9c78b2bSJens Axboe page += (i % nr_pages); 1205f9c78b2bSJens Axboe 1206f9c78b2bSJens Axboe i++; 1207f9c78b2bSJens Axboe } else { 1208f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1209f9c78b2bSJens Axboe if (!page) { 1210f9c78b2bSJens Axboe ret = -ENOMEM; 1211f9c78b2bSJens Axboe break; 1212f9c78b2bSJens Axboe } 1213f9c78b2bSJens Axboe } 1214f9c78b2bSJens Axboe 1215f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) 1216f9c78b2bSJens Axboe break; 1217f9c78b2bSJens Axboe 1218f9c78b2bSJens Axboe len -= bytes; 1219f9c78b2bSJens Axboe offset = 0; 1220f9c78b2bSJens Axboe } 1221f9c78b2bSJens Axboe 1222f9c78b2bSJens Axboe if (ret) 1223f9c78b2bSJens Axboe goto cleanup; 1224f9c78b2bSJens Axboe 12252884d0beSAl Viro if (map_data) 12262884d0beSAl Viro map_data->offset += bio->bi_iter.bi_size; 12272884d0beSAl Viro 1228f9c78b2bSJens Axboe /* 1229f9c78b2bSJens Axboe * success 1230f9c78b2bSJens Axboe */ 123126e49cfcSKent Overstreet if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) || 1232f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 123398a09d61SAl Viro ret = bio_copy_from_iter(bio, iter); 1234f9c78b2bSJens Axboe if (ret) 1235f9c78b2bSJens Axboe goto cleanup; 123698a09d61SAl Viro } else { 1237e81cef5dSAl Viro iov_iter_advance(iter, bio->bi_iter.bi_size); 1238f9c78b2bSJens Axboe } 1239f9c78b2bSJens Axboe 124026e49cfcSKent Overstreet bio->bi_private = bmd; 12412884d0beSAl Viro if (map_data && map_data->null_mapped) 12422884d0beSAl Viro bio_set_flag(bio, BIO_NULL_MAPPED); 1243f9c78b2bSJens Axboe return bio; 1244f9c78b2bSJens Axboe cleanup: 1245f9c78b2bSJens Axboe if (!map_data) 12461dfa0f68SChristoph Hellwig bio_free_pages(bio); 1247f9c78b2bSJens Axboe bio_put(bio); 1248f9c78b2bSJens Axboe out_bmd: 1249f9c78b2bSJens Axboe kfree(bmd); 1250f9c78b2bSJens Axboe return ERR_PTR(ret); 1251f9c78b2bSJens Axboe } 1252f9c78b2bSJens Axboe 125337f19e57SChristoph Hellwig /** 125437f19e57SChristoph Hellwig * bio_map_user_iov - map user iovec into bio 125537f19e57SChristoph Hellwig * @q: the struct request_queue for the bio 125637f19e57SChristoph Hellwig * @iter: iovec iterator 125737f19e57SChristoph Hellwig * @gfp_mask: memory allocation flags 125837f19e57SChristoph Hellwig * 125937f19e57SChristoph Hellwig * Map the user space address into a bio suitable for io to a block 126037f19e57SChristoph Hellwig * device. Returns an error pointer in case of error. 126137f19e57SChristoph Hellwig */ 126237f19e57SChristoph Hellwig struct bio *bio_map_user_iov(struct request_queue *q, 1263e81cef5dSAl Viro struct iov_iter *iter, 126426e49cfcSKent Overstreet gfp_t gfp_mask) 1265f9c78b2bSJens Axboe { 126626e49cfcSKent Overstreet int j; 1267f9c78b2bSJens Axboe struct bio *bio; 1268076098e5SAl Viro int ret; 12692b04e8f6SAl Viro struct bio_vec *bvec; 1270f9c78b2bSJens Axboe 1271b282cc76SAl Viro if (!iov_iter_count(iter)) 1272f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1273f9c78b2bSJens Axboe 1274b282cc76SAl Viro bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1275f9c78b2bSJens Axboe if (!bio) 1276f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1277f9c78b2bSJens Axboe 12780a0f1513SAl Viro while (iov_iter_count(iter)) { 1279629e42bcSAl Viro struct page **pages; 1280076098e5SAl Viro ssize_t bytes; 1281076098e5SAl Viro size_t offs, added = 0; 1282076098e5SAl Viro int npages; 1283f9c78b2bSJens Axboe 12840a0f1513SAl Viro bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1285076098e5SAl Viro if (unlikely(bytes <= 0)) { 1286076098e5SAl Viro ret = bytes ? bytes : -EFAULT; 1287f9c78b2bSJens Axboe goto out_unmap; 1288f9c78b2bSJens Axboe } 1289f9c78b2bSJens Axboe 1290076098e5SAl Viro npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1291076098e5SAl Viro 129298f0bc99SAl Viro if (unlikely(offs & queue_dma_alignment(q))) { 129398f0bc99SAl Viro ret = -EINVAL; 129498f0bc99SAl Viro j = 0; 129598f0bc99SAl Viro } else { 1296629e42bcSAl Viro for (j = 0; j < npages; j++) { 129798f0bc99SAl Viro struct page *page = pages[j]; 1298076098e5SAl Viro unsigned int n = PAGE_SIZE - offs; 129995d78c28SVitaly Mayatskikh unsigned short prev_bi_vcnt = bio->bi_vcnt; 1300f9c78b2bSJens Axboe 1301076098e5SAl Viro if (n > bytes) 1302076098e5SAl Viro n = bytes; 1303f9c78b2bSJens Axboe 130498f0bc99SAl Viro if (!bio_add_pc_page(q, bio, page, n, offs)) 1305f9c78b2bSJens Axboe break; 1306f9c78b2bSJens Axboe 130795d78c28SVitaly Mayatskikh /* 130895d78c28SVitaly Mayatskikh * check if vector was merged with previous 130995d78c28SVitaly Mayatskikh * drop page reference if needed 131095d78c28SVitaly Mayatskikh */ 131195d78c28SVitaly Mayatskikh if (bio->bi_vcnt == prev_bi_vcnt) 131298f0bc99SAl Viro put_page(page); 131395d78c28SVitaly Mayatskikh 1314076098e5SAl Viro added += n; 1315076098e5SAl Viro bytes -= n; 1316076098e5SAl Viro offs = 0; 1317f9c78b2bSJens Axboe } 13180a0f1513SAl Viro iov_iter_advance(iter, added); 131998f0bc99SAl Viro } 1320f9c78b2bSJens Axboe /* 1321f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1322f9c78b2bSJens Axboe */ 1323629e42bcSAl Viro while (j < npages) 132409cbfeafSKirill A. Shutemov put_page(pages[j++]); 1325629e42bcSAl Viro kvfree(pages); 1326e2e115d1SAl Viro /* couldn't stuff something into bio? */ 1327e2e115d1SAl Viro if (bytes) 1328e2e115d1SAl Viro break; 1329f9c78b2bSJens Axboe } 1330f9c78b2bSJens Axboe 1331b7c44ed9SJens Axboe bio_set_flag(bio, BIO_USER_MAPPED); 133237f19e57SChristoph Hellwig 133337f19e57SChristoph Hellwig /* 13345fad1b64SBart Van Assche * subtle -- if bio_map_user_iov() ended up bouncing a bio, 133537f19e57SChristoph Hellwig * it would normally disappear when its bi_end_io is run. 133637f19e57SChristoph Hellwig * however, we need it for the unmap, so grab an extra 133737f19e57SChristoph Hellwig * reference to it 133837f19e57SChristoph Hellwig */ 133937f19e57SChristoph Hellwig bio_get(bio); 1340f9c78b2bSJens Axboe return bio; 1341f9c78b2bSJens Axboe 1342f9c78b2bSJens Axboe out_unmap: 13432b04e8f6SAl Viro bio_for_each_segment_all(bvec, bio, j) { 13442b04e8f6SAl Viro put_page(bvec->bv_page); 1345f9c78b2bSJens Axboe } 1346f9c78b2bSJens Axboe bio_put(bio); 1347f9c78b2bSJens Axboe return ERR_PTR(ret); 1348f9c78b2bSJens Axboe } 1349f9c78b2bSJens Axboe 1350f9c78b2bSJens Axboe static void __bio_unmap_user(struct bio *bio) 1351f9c78b2bSJens Axboe { 1352f9c78b2bSJens Axboe struct bio_vec *bvec; 1353f9c78b2bSJens Axboe int i; 1354f9c78b2bSJens Axboe 1355f9c78b2bSJens Axboe /* 1356f9c78b2bSJens Axboe * make sure we dirty pages we wrote to 1357f9c78b2bSJens Axboe */ 1358f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1359f9c78b2bSJens Axboe if (bio_data_dir(bio) == READ) 1360f9c78b2bSJens Axboe set_page_dirty_lock(bvec->bv_page); 1361f9c78b2bSJens Axboe 136209cbfeafSKirill A. Shutemov put_page(bvec->bv_page); 1363f9c78b2bSJens Axboe } 1364f9c78b2bSJens Axboe 1365f9c78b2bSJens Axboe bio_put(bio); 1366f9c78b2bSJens Axboe } 1367f9c78b2bSJens Axboe 1368f9c78b2bSJens Axboe /** 1369f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1370f9c78b2bSJens Axboe * @bio: the bio being unmapped 1371f9c78b2bSJens Axboe * 13725fad1b64SBart Van Assche * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 13735fad1b64SBart Van Assche * process context. 1374f9c78b2bSJens Axboe * 1375f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1376f9c78b2bSJens Axboe */ 1377f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1378f9c78b2bSJens Axboe { 1379f9c78b2bSJens Axboe __bio_unmap_user(bio); 1380f9c78b2bSJens Axboe bio_put(bio); 1381f9c78b2bSJens Axboe } 1382f9c78b2bSJens Axboe 13834246a0b6SChristoph Hellwig static void bio_map_kern_endio(struct bio *bio) 1384f9c78b2bSJens Axboe { 1385f9c78b2bSJens Axboe bio_put(bio); 1386f9c78b2bSJens Axboe } 1387f9c78b2bSJens Axboe 138875c72b83SChristoph Hellwig /** 138975c72b83SChristoph Hellwig * bio_map_kern - map kernel address into bio 139075c72b83SChristoph Hellwig * @q: the struct request_queue for the bio 139175c72b83SChristoph Hellwig * @data: pointer to buffer to map 139275c72b83SChristoph Hellwig * @len: length in bytes 139375c72b83SChristoph Hellwig * @gfp_mask: allocation flags for bio allocation 139475c72b83SChristoph Hellwig * 139575c72b83SChristoph Hellwig * Map the kernel address into a bio suitable for io to a block 139675c72b83SChristoph Hellwig * device. Returns an error pointer in case of error. 139775c72b83SChristoph Hellwig */ 139875c72b83SChristoph Hellwig struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 139975c72b83SChristoph Hellwig gfp_t gfp_mask) 1400f9c78b2bSJens Axboe { 1401f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1402f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1403f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1404f9c78b2bSJens Axboe const int nr_pages = end - start; 1405f9c78b2bSJens Axboe int offset, i; 1406f9c78b2bSJens Axboe struct bio *bio; 1407f9c78b2bSJens Axboe 1408f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1409f9c78b2bSJens Axboe if (!bio) 1410f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1411f9c78b2bSJens Axboe 1412f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1413f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1414f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1415f9c78b2bSJens Axboe 1416f9c78b2bSJens Axboe if (len <= 0) 1417f9c78b2bSJens Axboe break; 1418f9c78b2bSJens Axboe 1419f9c78b2bSJens Axboe if (bytes > len) 1420f9c78b2bSJens Axboe bytes = len; 1421f9c78b2bSJens Axboe 1422f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 142375c72b83SChristoph Hellwig offset) < bytes) { 142475c72b83SChristoph Hellwig /* we don't support partial mappings */ 142575c72b83SChristoph Hellwig bio_put(bio); 142675c72b83SChristoph Hellwig return ERR_PTR(-EINVAL); 142775c72b83SChristoph Hellwig } 1428f9c78b2bSJens Axboe 1429f9c78b2bSJens Axboe data += bytes; 1430f9c78b2bSJens Axboe len -= bytes; 1431f9c78b2bSJens Axboe offset = 0; 1432f9c78b2bSJens Axboe } 1433f9c78b2bSJens Axboe 1434f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1435f9c78b2bSJens Axboe return bio; 1436f9c78b2bSJens Axboe } 1437f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_map_kern); 1438f9c78b2bSJens Axboe 14394246a0b6SChristoph Hellwig static void bio_copy_kern_endio(struct bio *bio) 1440f9c78b2bSJens Axboe { 14411dfa0f68SChristoph Hellwig bio_free_pages(bio); 14421dfa0f68SChristoph Hellwig bio_put(bio); 14431dfa0f68SChristoph Hellwig } 14441dfa0f68SChristoph Hellwig 14454246a0b6SChristoph Hellwig static void bio_copy_kern_endio_read(struct bio *bio) 14461dfa0f68SChristoph Hellwig { 144742d2683aSChristoph Hellwig char *p = bio->bi_private; 14481dfa0f68SChristoph Hellwig struct bio_vec *bvec; 1449f9c78b2bSJens Axboe int i; 1450f9c78b2bSJens Axboe 1451f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 14521dfa0f68SChristoph Hellwig memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1453f9c78b2bSJens Axboe p += bvec->bv_len; 1454f9c78b2bSJens Axboe } 1455f9c78b2bSJens Axboe 14564246a0b6SChristoph Hellwig bio_copy_kern_endio(bio); 1457f9c78b2bSJens Axboe } 1458f9c78b2bSJens Axboe 1459f9c78b2bSJens Axboe /** 1460f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1461f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1462f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1463f9c78b2bSJens Axboe * @len: length in bytes 1464f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1465f9c78b2bSJens Axboe * @reading: data direction is READ 1466f9c78b2bSJens Axboe * 1467f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1468f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1469f9c78b2bSJens Axboe */ 1470f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1471f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1472f9c78b2bSJens Axboe { 147342d2683aSChristoph Hellwig unsigned long kaddr = (unsigned long)data; 147442d2683aSChristoph Hellwig unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 147542d2683aSChristoph Hellwig unsigned long start = kaddr >> PAGE_SHIFT; 147642d2683aSChristoph Hellwig struct bio *bio; 1477f9c78b2bSJens Axboe void *p = data; 14781dfa0f68SChristoph Hellwig int nr_pages = 0; 1479f9c78b2bSJens Axboe 148042d2683aSChristoph Hellwig /* 148142d2683aSChristoph Hellwig * Overflow, abort 148242d2683aSChristoph Hellwig */ 148342d2683aSChristoph Hellwig if (end < start) 148442d2683aSChristoph Hellwig return ERR_PTR(-EINVAL); 1485f9c78b2bSJens Axboe 148642d2683aSChristoph Hellwig nr_pages = end - start; 148742d2683aSChristoph Hellwig bio = bio_kmalloc(gfp_mask, nr_pages); 148842d2683aSChristoph Hellwig if (!bio) 148942d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 149042d2683aSChristoph Hellwig 149142d2683aSChristoph Hellwig while (len) { 149242d2683aSChristoph Hellwig struct page *page; 149342d2683aSChristoph Hellwig unsigned int bytes = PAGE_SIZE; 149442d2683aSChristoph Hellwig 149542d2683aSChristoph Hellwig if (bytes > len) 149642d2683aSChristoph Hellwig bytes = len; 149742d2683aSChristoph Hellwig 149842d2683aSChristoph Hellwig page = alloc_page(q->bounce_gfp | gfp_mask); 149942d2683aSChristoph Hellwig if (!page) 150042d2683aSChristoph Hellwig goto cleanup; 150142d2683aSChristoph Hellwig 150242d2683aSChristoph Hellwig if (!reading) 150342d2683aSChristoph Hellwig memcpy(page_address(page), p, bytes); 150442d2683aSChristoph Hellwig 150542d2683aSChristoph Hellwig if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 150642d2683aSChristoph Hellwig break; 150742d2683aSChristoph Hellwig 150842d2683aSChristoph Hellwig len -= bytes; 150942d2683aSChristoph Hellwig p += bytes; 1510f9c78b2bSJens Axboe } 1511f9c78b2bSJens Axboe 15121dfa0f68SChristoph Hellwig if (reading) { 15131dfa0f68SChristoph Hellwig bio->bi_end_io = bio_copy_kern_endio_read; 151442d2683aSChristoph Hellwig bio->bi_private = data; 15151dfa0f68SChristoph Hellwig } else { 1516f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 15171dfa0f68SChristoph Hellwig } 15181dfa0f68SChristoph Hellwig 1519f9c78b2bSJens Axboe return bio; 152042d2683aSChristoph Hellwig 152142d2683aSChristoph Hellwig cleanup: 15221dfa0f68SChristoph Hellwig bio_free_pages(bio); 152342d2683aSChristoph Hellwig bio_put(bio); 152442d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 1525f9c78b2bSJens Axboe } 1526f9c78b2bSJens Axboe 1527f9c78b2bSJens Axboe /* 1528f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1529f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1530f9c78b2bSJens Axboe * 1531f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1532f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1533f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1534f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1535f9c78b2bSJens Axboe * in process context. 1536f9c78b2bSJens Axboe * 1537f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1538f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1539f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1540f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1541f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1542f9c78b2bSJens Axboe * 1543f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1544f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1545f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1546f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1547f9c78b2bSJens Axboe * pagecache. 1548f9c78b2bSJens Axboe * 1549f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1550f9c78b2bSJens Axboe * deferred bio dirtying paths. 1551f9c78b2bSJens Axboe */ 1552f9c78b2bSJens Axboe 1553f9c78b2bSJens Axboe /* 1554f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1555f9c78b2bSJens Axboe */ 1556f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1557f9c78b2bSJens Axboe { 1558f9c78b2bSJens Axboe struct bio_vec *bvec; 1559f9c78b2bSJens Axboe int i; 1560f9c78b2bSJens Axboe 1561f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 15623bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 15633bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1564f9c78b2bSJens Axboe } 1565f9c78b2bSJens Axboe } 15661900fcc4SKent Overstreet EXPORT_SYMBOL_GPL(bio_set_pages_dirty); 1567f9c78b2bSJens Axboe 1568f9c78b2bSJens Axboe static void bio_release_pages(struct bio *bio) 1569f9c78b2bSJens Axboe { 1570f9c78b2bSJens Axboe struct bio_vec *bvec; 1571f9c78b2bSJens Axboe int i; 1572f9c78b2bSJens Axboe 157324d5493fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, i) 157424d5493fSChristoph Hellwig put_page(bvec->bv_page); 1575f9c78b2bSJens Axboe } 1576f9c78b2bSJens Axboe 1577f9c78b2bSJens Axboe /* 1578f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1579f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1580f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 158124d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1582f9c78b2bSJens Axboe * 1583f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1584ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1585ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1586f9c78b2bSJens Axboe */ 1587f9c78b2bSJens Axboe 1588f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1589f9c78b2bSJens Axboe 1590f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1591f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1592f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1593f9c78b2bSJens Axboe 1594f9c78b2bSJens Axboe /* 1595f9c78b2bSJens Axboe * This runs in process context 1596f9c78b2bSJens Axboe */ 1597f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1598f9c78b2bSJens Axboe { 159924d5493fSChristoph Hellwig struct bio *bio, *next; 1600f9c78b2bSJens Axboe 160124d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 160224d5493fSChristoph Hellwig next = bio_dirty_list; 1603f9c78b2bSJens Axboe bio_dirty_list = NULL; 160424d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1605f9c78b2bSJens Axboe 160624d5493fSChristoph Hellwig while ((bio = next) != NULL) { 160724d5493fSChristoph Hellwig next = bio->bi_private; 1608f9c78b2bSJens Axboe 1609f9c78b2bSJens Axboe bio_set_pages_dirty(bio); 1610f9c78b2bSJens Axboe bio_release_pages(bio); 1611f9c78b2bSJens Axboe bio_put(bio); 1612f9c78b2bSJens Axboe } 1613f9c78b2bSJens Axboe } 1614f9c78b2bSJens Axboe 1615f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1616f9c78b2bSJens Axboe { 1617f9c78b2bSJens Axboe struct bio_vec *bvec; 161824d5493fSChristoph Hellwig unsigned long flags; 1619f9c78b2bSJens Axboe int i; 1620f9c78b2bSJens Axboe 1621f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 162224d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 162324d5493fSChristoph Hellwig goto defer; 1624f9c78b2bSJens Axboe } 1625f9c78b2bSJens Axboe 162624d5493fSChristoph Hellwig bio_release_pages(bio); 162724d5493fSChristoph Hellwig bio_put(bio); 162824d5493fSChristoph Hellwig return; 162924d5493fSChristoph Hellwig defer: 1630f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1631f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1632f9c78b2bSJens Axboe bio_dirty_list = bio; 1633f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1634f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1635f9c78b2bSJens Axboe } 16361900fcc4SKent Overstreet EXPORT_SYMBOL_GPL(bio_check_pages_dirty); 1637f9c78b2bSJens Axboe 1638ddcf35d3SMichael Callahan void generic_start_io_acct(struct request_queue *q, int op, 1639d62e26b3SJens Axboe unsigned long sectors, struct hd_struct *part) 1640394ffa50SGu Zheng { 1641ddcf35d3SMichael Callahan const int sgrp = op_stat_group(op); 1642394ffa50SGu Zheng int cpu = part_stat_lock(); 1643394ffa50SGu Zheng 1644d62e26b3SJens Axboe part_round_stats(q, cpu, part); 1645ddcf35d3SMichael Callahan part_stat_inc(cpu, part, ios[sgrp]); 1646ddcf35d3SMichael Callahan part_stat_add(cpu, part, sectors[sgrp], sectors); 1647ddcf35d3SMichael Callahan part_inc_in_flight(q, part, op_is_write(op)); 1648394ffa50SGu Zheng 1649394ffa50SGu Zheng part_stat_unlock(); 1650394ffa50SGu Zheng } 1651394ffa50SGu Zheng EXPORT_SYMBOL(generic_start_io_acct); 1652394ffa50SGu Zheng 1653ddcf35d3SMichael Callahan void generic_end_io_acct(struct request_queue *q, int req_op, 1654d62e26b3SJens Axboe struct hd_struct *part, unsigned long start_time) 1655394ffa50SGu Zheng { 1656394ffa50SGu Zheng unsigned long duration = jiffies - start_time; 1657ddcf35d3SMichael Callahan const int sgrp = op_stat_group(req_op); 1658394ffa50SGu Zheng int cpu = part_stat_lock(); 1659394ffa50SGu Zheng 1660ddcf35d3SMichael Callahan part_stat_add(cpu, part, ticks[sgrp], duration); 1661d62e26b3SJens Axboe part_round_stats(q, cpu, part); 1662ddcf35d3SMichael Callahan part_dec_in_flight(q, part, op_is_write(req_op)); 1663394ffa50SGu Zheng 1664394ffa50SGu Zheng part_stat_unlock(); 1665394ffa50SGu Zheng } 1666394ffa50SGu Zheng EXPORT_SYMBOL(generic_end_io_acct); 1667394ffa50SGu Zheng 1668f9c78b2bSJens Axboe #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1669f9c78b2bSJens Axboe void bio_flush_dcache_pages(struct bio *bi) 1670f9c78b2bSJens Axboe { 1671f9c78b2bSJens Axboe struct bio_vec bvec; 1672f9c78b2bSJens Axboe struct bvec_iter iter; 1673f9c78b2bSJens Axboe 1674f9c78b2bSJens Axboe bio_for_each_segment(bvec, bi, iter) 1675f9c78b2bSJens Axboe flush_dcache_page(bvec.bv_page); 1676f9c78b2bSJens Axboe } 1677f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_flush_dcache_pages); 1678f9c78b2bSJens Axboe #endif 1679f9c78b2bSJens Axboe 1680c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1681c4cf5261SJens Axboe { 1682c4cf5261SJens Axboe /* 1683c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1684c4cf5261SJens Axboe * we always end io on the first invocation. 1685c4cf5261SJens Axboe */ 1686c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1687c4cf5261SJens Axboe return true; 1688c4cf5261SJens Axboe 1689c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1690c4cf5261SJens Axboe 1691326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1692b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1693c4cf5261SJens Axboe return true; 1694326e1dbbSMike Snitzer } 1695c4cf5261SJens Axboe 1696c4cf5261SJens Axboe return false; 1697c4cf5261SJens Axboe } 1698c4cf5261SJens Axboe 1699f9c78b2bSJens Axboe /** 1700f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1701f9c78b2bSJens Axboe * @bio: bio 1702f9c78b2bSJens Axboe * 1703f9c78b2bSJens Axboe * Description: 17044246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 17054246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 17064246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1707fbbaf700SNeilBrown * 1708fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1709fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1710fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1711fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1712f9c78b2bSJens Axboe **/ 17134246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1714f9c78b2bSJens Axboe { 1715ba8c6967SChristoph Hellwig again: 17162b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1717ba8c6967SChristoph Hellwig return; 17187c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 17197c20f116SChristoph Hellwig return; 1720f9c78b2bSJens Axboe 172167b42d0bSJosef Bacik if (bio->bi_disk) 172267b42d0bSJosef Bacik rq_qos_done_bio(bio->bi_disk->queue, bio); 172367b42d0bSJosef Bacik 1724f9c78b2bSJens Axboe /* 1725ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1726ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1727ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1728ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1729ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1730ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1731f9c78b2bSJens Axboe */ 1732f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 173338f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1734ba8c6967SChristoph Hellwig goto again; 1735ba8c6967SChristoph Hellwig } 1736ba8c6967SChristoph Hellwig 173774d46992SChristoph Hellwig if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 173874d46992SChristoph Hellwig trace_block_bio_complete(bio->bi_disk->queue, bio, 1739a462b950SBart Van Assche blk_status_to_errno(bio->bi_status)); 1740fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1741fbbaf700SNeilBrown } 1742fbbaf700SNeilBrown 17439e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1744b222dd2fSShaohua Li /* release cgroup info */ 1745b222dd2fSShaohua Li bio_uninit(bio); 1746f9c78b2bSJens Axboe if (bio->bi_end_io) 17474246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1748f9c78b2bSJens Axboe } 1749f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1750f9c78b2bSJens Axboe 1751f9c78b2bSJens Axboe /** 1752f9c78b2bSJens Axboe * bio_split - split a bio 1753f9c78b2bSJens Axboe * @bio: bio to split 1754f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1755f9c78b2bSJens Axboe * @gfp: gfp mask 1756f9c78b2bSJens Axboe * @bs: bio set to allocate from 1757f9c78b2bSJens Axboe * 1758f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1759f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1760f9c78b2bSJens Axboe * 1761f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1762f3f5da62SMartin K. Petersen * to @bio's bi_io_vec; it is the caller's responsibility to ensure that 1763f3f5da62SMartin K. Petersen * @bio is not freed before the split. 1764f9c78b2bSJens Axboe */ 1765f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1766f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1767f9c78b2bSJens Axboe { 1768f341a4d3SMikulas Patocka struct bio *split; 1769f9c78b2bSJens Axboe 1770f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1771f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1772f9c78b2bSJens Axboe 1773f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1774f9c78b2bSJens Axboe if (!split) 1775f9c78b2bSJens Axboe return NULL; 1776f9c78b2bSJens Axboe 1777f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1778f9c78b2bSJens Axboe 1779f9c78b2bSJens Axboe if (bio_integrity(split)) 1780fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1781f9c78b2bSJens Axboe 1782f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1783f9c78b2bSJens Axboe 1784fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 178520d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1786fbbaf700SNeilBrown 1787f9c78b2bSJens Axboe return split; 1788f9c78b2bSJens Axboe } 1789f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1790f9c78b2bSJens Axboe 1791f9c78b2bSJens Axboe /** 1792f9c78b2bSJens Axboe * bio_trim - trim a bio 1793f9c78b2bSJens Axboe * @bio: bio to trim 1794f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1795f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1796f9c78b2bSJens Axboe */ 1797f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1798f9c78b2bSJens Axboe { 1799f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1800f9c78b2bSJens Axboe * the given offset and size. 1801f9c78b2bSJens Axboe */ 1802f9c78b2bSJens Axboe 1803f9c78b2bSJens Axboe size <<= 9; 1804f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1805f9c78b2bSJens Axboe return; 1806f9c78b2bSJens Axboe 1807b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_SEG_VALID); 1808f9c78b2bSJens Axboe 1809f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1810f9c78b2bSJens Axboe 1811f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1812376a78abSDmitry Monakhov 1813376a78abSDmitry Monakhov if (bio_integrity(bio)) 1814fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1815376a78abSDmitry Monakhov 1816f9c78b2bSJens Axboe } 1817f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1818f9c78b2bSJens Axboe 1819f9c78b2bSJens Axboe /* 1820f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1821f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1822f9c78b2bSJens Axboe */ 18238aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1824f9c78b2bSJens Axboe { 1825ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1826f9c78b2bSJens Axboe 18278aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1828f9c78b2bSJens Axboe } 1829f9c78b2bSJens Axboe 1830917a38c7SKent Overstreet /* 1831917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1832917a38c7SKent Overstreet * 1833917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1834917a38c7SKent Overstreet * kzalloc()). 1835917a38c7SKent Overstreet */ 1836917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1837f9c78b2bSJens Axboe { 1838f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1839f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1840917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1841f9c78b2bSJens Axboe 18428aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 18438aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1844f9c78b2bSJens Axboe 1845f9c78b2bSJens Axboe bioset_integrity_free(bs); 1846917a38c7SKent Overstreet if (bs->bio_slab) 1847f9c78b2bSJens Axboe bio_put_slab(bs); 1848917a38c7SKent Overstreet bs->bio_slab = NULL; 1849917a38c7SKent Overstreet } 1850917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1851f9c78b2bSJens Axboe 1852011067b0SNeilBrown /** 1853917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1854dad08527SKent Overstreet * @bs: pool to initialize 1855917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1856917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1857917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1858917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1859917a38c7SKent Overstreet * 1860dad08527SKent Overstreet * Description: 1861dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1862dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1863dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1864dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1865dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1866dad08527SKent Overstreet * or things will break badly. 1867dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1868dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1869dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1870dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1871dad08527SKent Overstreet * 1872917a38c7SKent Overstreet */ 1873917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1874917a38c7SKent Overstreet unsigned int pool_size, 1875917a38c7SKent Overstreet unsigned int front_pad, 1876917a38c7SKent Overstreet int flags) 1877917a38c7SKent Overstreet { 1878917a38c7SKent Overstreet unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1879917a38c7SKent Overstreet 1880917a38c7SKent Overstreet bs->front_pad = front_pad; 1881917a38c7SKent Overstreet 1882917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1883917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1884917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1885917a38c7SKent Overstreet 1886917a38c7SKent Overstreet bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1887917a38c7SKent Overstreet if (!bs->bio_slab) 1888917a38c7SKent Overstreet return -ENOMEM; 1889917a38c7SKent Overstreet 1890917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1891917a38c7SKent Overstreet goto bad; 1892917a38c7SKent Overstreet 1893917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1894917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1895917a38c7SKent Overstreet goto bad; 1896917a38c7SKent Overstreet 1897917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 1898917a38c7SKent Overstreet return 0; 1899917a38c7SKent Overstreet 1900917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1901917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1902917a38c7SKent Overstreet goto bad; 1903917a38c7SKent Overstreet 1904917a38c7SKent Overstreet return 0; 1905917a38c7SKent Overstreet bad: 1906917a38c7SKent Overstreet bioset_exit(bs); 1907917a38c7SKent Overstreet return -ENOMEM; 1908917a38c7SKent Overstreet } 1909917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1910917a38c7SKent Overstreet 191128e89fd9SJens Axboe /* 191228e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 191328e89fd9SJens Axboe * another bio_set. 191428e89fd9SJens Axboe */ 191528e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 191628e89fd9SJens Axboe { 191728e89fd9SJens Axboe int flags; 191828e89fd9SJens Axboe 191928e89fd9SJens Axboe flags = 0; 192028e89fd9SJens Axboe if (src->bvec_pool.min_nr) 192128e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 192228e89fd9SJens Axboe if (src->rescue_workqueue) 192328e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 192428e89fd9SJens Axboe 192528e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 192628e89fd9SJens Axboe } 192728e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 192828e89fd9SJens Axboe 1929f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 19301d933cf0STejun Heo 19310d3bd88dSTejun Heo #ifdef CONFIG_MEMCG 19320d3bd88dSTejun Heo /** 19330d3bd88dSTejun Heo * bio_associate_blkcg_from_page - associate a bio with the page's blkcg 19340d3bd88dSTejun Heo * @bio: target bio 19350d3bd88dSTejun Heo * @page: the page to lookup the blkcg from 19360d3bd88dSTejun Heo * 19370d3bd88dSTejun Heo * Associate @bio with the blkcg from @page's owning memcg. This works like 19380d3bd88dSTejun Heo * every other associate function wrt references. 19390d3bd88dSTejun Heo */ 19400d3bd88dSTejun Heo int bio_associate_blkcg_from_page(struct bio *bio, struct page *page) 19410d3bd88dSTejun Heo { 19420d3bd88dSTejun Heo struct cgroup_subsys_state *blkcg_css; 19430d3bd88dSTejun Heo 19440d3bd88dSTejun Heo if (unlikely(bio->bi_css)) 19450d3bd88dSTejun Heo return -EBUSY; 19460d3bd88dSTejun Heo if (!page->mem_cgroup) 19470d3bd88dSTejun Heo return 0; 19480d3bd88dSTejun Heo blkcg_css = cgroup_get_e_css(page->mem_cgroup->css.cgroup, 19490d3bd88dSTejun Heo &io_cgrp_subsys); 19500d3bd88dSTejun Heo bio->bi_css = blkcg_css; 19510d3bd88dSTejun Heo return 0; 19520d3bd88dSTejun Heo } 19530d3bd88dSTejun Heo #endif /* CONFIG_MEMCG */ 19540d3bd88dSTejun Heo 19551d933cf0STejun Heo /** 19561d933cf0STejun Heo * bio_associate_blkcg - associate a bio with the specified blkcg 19571d933cf0STejun Heo * @bio: target bio 19581d933cf0STejun Heo * @blkcg_css: css of the blkcg to associate 19591d933cf0STejun Heo * 19601d933cf0STejun Heo * Associate @bio with the blkcg specified by @blkcg_css. Block layer will 19611d933cf0STejun Heo * treat @bio as if it were issued by a task which belongs to the blkcg. 19621d933cf0STejun Heo * 19631d933cf0STejun Heo * This function takes an extra reference of @blkcg_css which will be put 19641d933cf0STejun Heo * when @bio is released. The caller must own @bio and is responsible for 19651d933cf0STejun Heo * synchronizing calls to this function. 19661d933cf0STejun Heo */ 19671d933cf0STejun Heo int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css) 19681d933cf0STejun Heo { 19691d933cf0STejun Heo if (unlikely(bio->bi_css)) 19701d933cf0STejun Heo return -EBUSY; 19711d933cf0STejun Heo css_get(blkcg_css); 19721d933cf0STejun Heo bio->bi_css = blkcg_css; 19731d933cf0STejun Heo return 0; 19741d933cf0STejun Heo } 19755aa2a96bSTejun Heo EXPORT_SYMBOL_GPL(bio_associate_blkcg); 19761d933cf0STejun Heo 1977f9c78b2bSJens Axboe /** 197808e18eabSJosef Bacik * bio_associate_blkg - associate a bio with the specified blkg 197908e18eabSJosef Bacik * @bio: target bio 198008e18eabSJosef Bacik * @blkg: the blkg to associate 198108e18eabSJosef Bacik * 198208e18eabSJosef Bacik * Associate @bio with the blkg specified by @blkg. This is the queue specific 198308e18eabSJosef Bacik * blkcg information associated with the @bio, a reference will be taken on the 198408e18eabSJosef Bacik * @blkg and will be freed when the bio is freed. 198508e18eabSJosef Bacik */ 198608e18eabSJosef Bacik int bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 198708e18eabSJosef Bacik { 198808e18eabSJosef Bacik if (unlikely(bio->bi_blkg)) 198908e18eabSJosef Bacik return -EBUSY; 199008e18eabSJosef Bacik blkg_get(blkg); 199108e18eabSJosef Bacik bio->bi_blkg = blkg; 199208e18eabSJosef Bacik return 0; 199308e18eabSJosef Bacik } 199408e18eabSJosef Bacik 199508e18eabSJosef Bacik /** 1996f9c78b2bSJens Axboe * bio_disassociate_task - undo bio_associate_current() 1997f9c78b2bSJens Axboe * @bio: target bio 1998f9c78b2bSJens Axboe */ 1999f9c78b2bSJens Axboe void bio_disassociate_task(struct bio *bio) 2000f9c78b2bSJens Axboe { 2001f9c78b2bSJens Axboe if (bio->bi_ioc) { 2002f9c78b2bSJens Axboe put_io_context(bio->bi_ioc); 2003f9c78b2bSJens Axboe bio->bi_ioc = NULL; 2004f9c78b2bSJens Axboe } 2005f9c78b2bSJens Axboe if (bio->bi_css) { 2006f9c78b2bSJens Axboe css_put(bio->bi_css); 2007f9c78b2bSJens Axboe bio->bi_css = NULL; 2008f9c78b2bSJens Axboe } 200908e18eabSJosef Bacik if (bio->bi_blkg) { 201008e18eabSJosef Bacik blkg_put(bio->bi_blkg); 201108e18eabSJosef Bacik bio->bi_blkg = NULL; 201208e18eabSJosef Bacik } 2013f9c78b2bSJens Axboe } 2014f9c78b2bSJens Axboe 201520bd723eSPaolo Valente /** 201620bd723eSPaolo Valente * bio_clone_blkcg_association - clone blkcg association from src to dst bio 201720bd723eSPaolo Valente * @dst: destination bio 201820bd723eSPaolo Valente * @src: source bio 201920bd723eSPaolo Valente */ 202020bd723eSPaolo Valente void bio_clone_blkcg_association(struct bio *dst, struct bio *src) 202120bd723eSPaolo Valente { 202220bd723eSPaolo Valente if (src->bi_css) 202320bd723eSPaolo Valente WARN_ON(bio_associate_blkcg(dst, src->bi_css)); 202420bd723eSPaolo Valente } 20258a8e6f84SShaohua Li EXPORT_SYMBOL_GPL(bio_clone_blkcg_association); 2026f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 2027f9c78b2bSJens Axboe 2028f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2029f9c78b2bSJens Axboe { 2030f9c78b2bSJens Axboe int i; 2031f9c78b2bSJens Axboe 2032ed996a52SChristoph Hellwig for (i = 0; i < BVEC_POOL_NR; i++) { 2033f9c78b2bSJens Axboe int size; 2034f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2035f9c78b2bSJens Axboe 2036f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2037f9c78b2bSJens Axboe bvs->slab = NULL; 2038f9c78b2bSJens Axboe continue; 2039f9c78b2bSJens Axboe } 2040f9c78b2bSJens Axboe 2041f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2042f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2043f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2044f9c78b2bSJens Axboe } 2045f9c78b2bSJens Axboe } 2046f9c78b2bSJens Axboe 2047f9c78b2bSJens Axboe static int __init init_bio(void) 2048f9c78b2bSJens Axboe { 2049f9c78b2bSJens Axboe bio_slab_max = 2; 2050f9c78b2bSJens Axboe bio_slab_nr = 0; 20516396bb22SKees Cook bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 20526396bb22SKees Cook GFP_KERNEL); 2053f9c78b2bSJens Axboe if (!bio_slabs) 2054f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2055f9c78b2bSJens Axboe 2056f9c78b2bSJens Axboe bio_integrity_init(); 2057f9c78b2bSJens Axboe biovec_init_slabs(); 2058f9c78b2bSJens Axboe 2059f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2060f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2061f9c78b2bSJens Axboe 2062f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2063f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2064f9c78b2bSJens Axboe 2065f9c78b2bSJens Axboe return 0; 2066f9c78b2bSJens Axboe } 2067f9c78b2bSJens Axboe subsys_initcall(init_bio); 2068