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> 31f9c78b2bSJens Axboe 32f9c78b2bSJens Axboe #include <trace/events/block.h> 33f9c78b2bSJens Axboe 34f9c78b2bSJens Axboe /* 35f9c78b2bSJens Axboe * Test patch to inline a certain number of bi_io_vec's inside the bio 36f9c78b2bSJens Axboe * itself, to shrink a bio data allocation from two mempool calls to one 37f9c78b2bSJens Axboe */ 38f9c78b2bSJens Axboe #define BIO_INLINE_VECS 4 39f9c78b2bSJens Axboe 40f9c78b2bSJens Axboe /* 41f9c78b2bSJens Axboe * if you change this list, also change bvec_alloc or things will 42f9c78b2bSJens Axboe * break badly! cannot be bigger than what you can fit into an 43f9c78b2bSJens Axboe * unsigned short 44f9c78b2bSJens Axboe */ 45f9c78b2bSJens Axboe #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } 46f9c78b2bSJens Axboe static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { 47f9c78b2bSJens Axboe BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), 48f9c78b2bSJens Axboe }; 49f9c78b2bSJens Axboe #undef BV 50f9c78b2bSJens Axboe 51f9c78b2bSJens Axboe /* 52f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 53f9c78b2bSJens Axboe * IO code that does not need private memory pools. 54f9c78b2bSJens Axboe */ 55f9c78b2bSJens Axboe struct bio_set *fs_bio_set; 56f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 57f9c78b2bSJens Axboe 58f9c78b2bSJens Axboe /* 59f9c78b2bSJens Axboe * Our slab pool management 60f9c78b2bSJens Axboe */ 61f9c78b2bSJens Axboe struct bio_slab { 62f9c78b2bSJens Axboe struct kmem_cache *slab; 63f9c78b2bSJens Axboe unsigned int slab_ref; 64f9c78b2bSJens Axboe unsigned int slab_size; 65f9c78b2bSJens Axboe char name[8]; 66f9c78b2bSJens Axboe }; 67f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 68f9c78b2bSJens Axboe static struct bio_slab *bio_slabs; 69f9c78b2bSJens Axboe static unsigned int bio_slab_nr, bio_slab_max; 70f9c78b2bSJens Axboe 71f9c78b2bSJens Axboe static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 72f9c78b2bSJens Axboe { 73f9c78b2bSJens Axboe unsigned int sz = sizeof(struct bio) + extra_size; 74f9c78b2bSJens Axboe struct kmem_cache *slab = NULL; 75f9c78b2bSJens Axboe struct bio_slab *bslab, *new_bio_slabs; 76f9c78b2bSJens Axboe unsigned int new_bio_slab_max; 77f9c78b2bSJens Axboe unsigned int i, entry = -1; 78f9c78b2bSJens Axboe 79f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 80f9c78b2bSJens Axboe 81f9c78b2bSJens Axboe i = 0; 82f9c78b2bSJens Axboe while (i < bio_slab_nr) { 83f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 84f9c78b2bSJens Axboe 85f9c78b2bSJens Axboe if (!bslab->slab && entry == -1) 86f9c78b2bSJens Axboe entry = i; 87f9c78b2bSJens Axboe else if (bslab->slab_size == sz) { 88f9c78b2bSJens Axboe slab = bslab->slab; 89f9c78b2bSJens Axboe bslab->slab_ref++; 90f9c78b2bSJens Axboe break; 91f9c78b2bSJens Axboe } 92f9c78b2bSJens Axboe i++; 93f9c78b2bSJens Axboe } 94f9c78b2bSJens Axboe 95f9c78b2bSJens Axboe if (slab) 96f9c78b2bSJens Axboe goto out_unlock; 97f9c78b2bSJens Axboe 98f9c78b2bSJens Axboe if (bio_slab_nr == bio_slab_max && entry == -1) { 99f9c78b2bSJens Axboe new_bio_slab_max = bio_slab_max << 1; 100f9c78b2bSJens Axboe new_bio_slabs = krealloc(bio_slabs, 101f9c78b2bSJens Axboe new_bio_slab_max * sizeof(struct bio_slab), 102f9c78b2bSJens Axboe GFP_KERNEL); 103f9c78b2bSJens Axboe if (!new_bio_slabs) 104f9c78b2bSJens Axboe goto out_unlock; 105f9c78b2bSJens Axboe bio_slab_max = new_bio_slab_max; 106f9c78b2bSJens Axboe bio_slabs = new_bio_slabs; 107f9c78b2bSJens Axboe } 108f9c78b2bSJens Axboe if (entry == -1) 109f9c78b2bSJens Axboe entry = bio_slab_nr++; 110f9c78b2bSJens Axboe 111f9c78b2bSJens Axboe bslab = &bio_slabs[entry]; 112f9c78b2bSJens Axboe 113f9c78b2bSJens Axboe snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 1146a241483SMikulas Patocka slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 1156a241483SMikulas Patocka SLAB_HWCACHE_ALIGN, NULL); 116f9c78b2bSJens Axboe if (!slab) 117f9c78b2bSJens Axboe goto out_unlock; 118f9c78b2bSJens Axboe 119f9c78b2bSJens Axboe bslab->slab = slab; 120f9c78b2bSJens Axboe bslab->slab_ref = 1; 121f9c78b2bSJens Axboe bslab->slab_size = sz; 122f9c78b2bSJens Axboe out_unlock: 123f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 124f9c78b2bSJens Axboe return slab; 125f9c78b2bSJens Axboe } 126f9c78b2bSJens Axboe 127f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 128f9c78b2bSJens Axboe { 129f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 130f9c78b2bSJens Axboe unsigned int i; 131f9c78b2bSJens Axboe 132f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 133f9c78b2bSJens Axboe 134f9c78b2bSJens Axboe for (i = 0; i < bio_slab_nr; i++) { 135f9c78b2bSJens Axboe if (bs->bio_slab == bio_slabs[i].slab) { 136f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 137f9c78b2bSJens Axboe break; 138f9c78b2bSJens Axboe } 139f9c78b2bSJens Axboe } 140f9c78b2bSJens Axboe 141f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 142f9c78b2bSJens Axboe goto out; 143f9c78b2bSJens Axboe 144f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 145f9c78b2bSJens Axboe 146f9c78b2bSJens Axboe if (--bslab->slab_ref) 147f9c78b2bSJens Axboe goto out; 148f9c78b2bSJens Axboe 149f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 150f9c78b2bSJens Axboe bslab->slab = NULL; 151f9c78b2bSJens Axboe 152f9c78b2bSJens Axboe out: 153f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 154f9c78b2bSJens Axboe } 155f9c78b2bSJens Axboe 156f9c78b2bSJens Axboe unsigned int bvec_nr_vecs(unsigned short idx) 157f9c78b2bSJens Axboe { 158f9c78b2bSJens Axboe return bvec_slabs[idx].nr_vecs; 159f9c78b2bSJens Axboe } 160f9c78b2bSJens Axboe 161f9c78b2bSJens Axboe void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 162f9c78b2bSJens Axboe { 163f9c78b2bSJens Axboe BIO_BUG_ON(idx >= BIOVEC_NR_POOLS); 164f9c78b2bSJens Axboe 165f9c78b2bSJens Axboe if (idx == BIOVEC_MAX_IDX) 166f9c78b2bSJens Axboe mempool_free(bv, pool); 167f9c78b2bSJens Axboe else { 168f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + idx; 169f9c78b2bSJens Axboe 170f9c78b2bSJens Axboe kmem_cache_free(bvs->slab, bv); 171f9c78b2bSJens Axboe } 172f9c78b2bSJens Axboe } 173f9c78b2bSJens Axboe 174f9c78b2bSJens Axboe struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 175f9c78b2bSJens Axboe mempool_t *pool) 176f9c78b2bSJens Axboe { 177f9c78b2bSJens Axboe struct bio_vec *bvl; 178f9c78b2bSJens Axboe 179f9c78b2bSJens Axboe /* 180f9c78b2bSJens Axboe * see comment near bvec_array define! 181f9c78b2bSJens Axboe */ 182f9c78b2bSJens Axboe switch (nr) { 183f9c78b2bSJens Axboe case 1: 184f9c78b2bSJens Axboe *idx = 0; 185f9c78b2bSJens Axboe break; 186f9c78b2bSJens Axboe case 2 ... 4: 187f9c78b2bSJens Axboe *idx = 1; 188f9c78b2bSJens Axboe break; 189f9c78b2bSJens Axboe case 5 ... 16: 190f9c78b2bSJens Axboe *idx = 2; 191f9c78b2bSJens Axboe break; 192f9c78b2bSJens Axboe case 17 ... 64: 193f9c78b2bSJens Axboe *idx = 3; 194f9c78b2bSJens Axboe break; 195f9c78b2bSJens Axboe case 65 ... 128: 196f9c78b2bSJens Axboe *idx = 4; 197f9c78b2bSJens Axboe break; 198f9c78b2bSJens Axboe case 129 ... BIO_MAX_PAGES: 199f9c78b2bSJens Axboe *idx = 5; 200f9c78b2bSJens Axboe break; 201f9c78b2bSJens Axboe default: 202f9c78b2bSJens Axboe return NULL; 203f9c78b2bSJens Axboe } 204f9c78b2bSJens Axboe 205f9c78b2bSJens Axboe /* 206f9c78b2bSJens Axboe * idx now points to the pool we want to allocate from. only the 207f9c78b2bSJens Axboe * 1-vec entry pool is mempool backed. 208f9c78b2bSJens Axboe */ 209f9c78b2bSJens Axboe if (*idx == BIOVEC_MAX_IDX) { 210f9c78b2bSJens Axboe fallback: 211f9c78b2bSJens Axboe bvl = mempool_alloc(pool, gfp_mask); 212f9c78b2bSJens Axboe } else { 213f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + *idx; 214f9c78b2bSJens Axboe gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO); 215f9c78b2bSJens Axboe 216f9c78b2bSJens Axboe /* 217f9c78b2bSJens Axboe * Make this allocation restricted and don't dump info on 218f9c78b2bSJens Axboe * allocation failures, since we'll fallback to the mempool 219f9c78b2bSJens Axboe * in case of failure. 220f9c78b2bSJens Axboe */ 221f9c78b2bSJens Axboe __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 222f9c78b2bSJens Axboe 223f9c78b2bSJens Axboe /* 224f9c78b2bSJens Axboe * Try a slab allocation. If this fails and __GFP_WAIT 225f9c78b2bSJens Axboe * is set, retry with the 1-entry mempool 226f9c78b2bSJens Axboe */ 227f9c78b2bSJens Axboe bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 228f9c78b2bSJens Axboe if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) { 229f9c78b2bSJens Axboe *idx = BIOVEC_MAX_IDX; 230f9c78b2bSJens Axboe goto fallback; 231f9c78b2bSJens Axboe } 232f9c78b2bSJens Axboe } 233f9c78b2bSJens Axboe 234f9c78b2bSJens Axboe return bvl; 235f9c78b2bSJens Axboe } 236f9c78b2bSJens Axboe 237f9c78b2bSJens Axboe static void __bio_free(struct bio *bio) 238f9c78b2bSJens Axboe { 239f9c78b2bSJens Axboe bio_disassociate_task(bio); 240f9c78b2bSJens Axboe 241f9c78b2bSJens Axboe if (bio_integrity(bio)) 242f9c78b2bSJens Axboe bio_integrity_free(bio); 243f9c78b2bSJens Axboe } 244f9c78b2bSJens Axboe 245f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 246f9c78b2bSJens Axboe { 247f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 248f9c78b2bSJens Axboe void *p; 249f9c78b2bSJens Axboe 250f9c78b2bSJens Axboe __bio_free(bio); 251f9c78b2bSJens Axboe 252f9c78b2bSJens Axboe if (bs) { 253f9c78b2bSJens Axboe if (bio_flagged(bio, BIO_OWNS_VEC)) 254f9c78b2bSJens Axboe bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio)); 255f9c78b2bSJens Axboe 256f9c78b2bSJens Axboe /* 257f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 258f9c78b2bSJens Axboe */ 259f9c78b2bSJens Axboe p = bio; 260f9c78b2bSJens Axboe p -= bs->front_pad; 261f9c78b2bSJens Axboe 262f9c78b2bSJens Axboe mempool_free(p, bs->bio_pool); 263f9c78b2bSJens Axboe } else { 264f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 265f9c78b2bSJens Axboe kfree(bio); 266f9c78b2bSJens Axboe } 267f9c78b2bSJens Axboe } 268f9c78b2bSJens Axboe 269f9c78b2bSJens Axboe void bio_init(struct bio *bio) 270f9c78b2bSJens Axboe { 271f9c78b2bSJens Axboe memset(bio, 0, sizeof(*bio)); 272f9c78b2bSJens Axboe bio->bi_flags = 1 << BIO_UPTODATE; 273c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 274dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 275f9c78b2bSJens Axboe } 276f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 277f9c78b2bSJens Axboe 278f9c78b2bSJens Axboe /** 279f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 280f9c78b2bSJens Axboe * @bio: bio to reset 281f9c78b2bSJens Axboe * 282f9c78b2bSJens Axboe * Description: 283f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 284f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 285f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 286f9c78b2bSJens Axboe * comment in struct bio. 287f9c78b2bSJens Axboe */ 288f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 289f9c78b2bSJens Axboe { 290f9c78b2bSJens Axboe unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 291f9c78b2bSJens Axboe 292f9c78b2bSJens Axboe __bio_free(bio); 293f9c78b2bSJens Axboe 294f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 295f9c78b2bSJens Axboe bio->bi_flags = flags | (1 << BIO_UPTODATE); 296c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 297f9c78b2bSJens Axboe } 298f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 299f9c78b2bSJens Axboe 300f9c78b2bSJens Axboe static void bio_chain_endio(struct bio *bio, int error) 301f9c78b2bSJens Axboe { 302f9c78b2bSJens Axboe bio_endio(bio->bi_private, error); 303f9c78b2bSJens Axboe bio_put(bio); 304f9c78b2bSJens Axboe } 305f9c78b2bSJens Axboe 306326e1dbbSMike Snitzer /* 307326e1dbbSMike Snitzer * Increment chain count for the bio. Make sure the CHAIN flag update 308326e1dbbSMike Snitzer * is visible before the raised count. 309326e1dbbSMike Snitzer */ 310326e1dbbSMike Snitzer static inline void bio_inc_remaining(struct bio *bio) 311326e1dbbSMike Snitzer { 312326e1dbbSMike Snitzer bio->bi_flags |= (1 << BIO_CHAIN); 313326e1dbbSMike Snitzer smp_mb__before_atomic(); 314326e1dbbSMike Snitzer atomic_inc(&bio->__bi_remaining); 315326e1dbbSMike Snitzer } 316326e1dbbSMike Snitzer 317f9c78b2bSJens Axboe /** 318f9c78b2bSJens Axboe * bio_chain - chain bio completions 319f9c78b2bSJens Axboe * @bio: the target bio 320f9c78b2bSJens Axboe * @parent: the @bio's parent bio 321f9c78b2bSJens Axboe * 322f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 323f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 324f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 325f9c78b2bSJens Axboe * 326f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 327f9c78b2bSJens Axboe */ 328f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 329f9c78b2bSJens Axboe { 330f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 331f9c78b2bSJens Axboe 332f9c78b2bSJens Axboe bio->bi_private = parent; 333f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 334c4cf5261SJens Axboe bio_inc_remaining(parent); 335f9c78b2bSJens Axboe } 336f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 337f9c78b2bSJens Axboe 338f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 339f9c78b2bSJens Axboe { 340f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 341f9c78b2bSJens Axboe struct bio *bio; 342f9c78b2bSJens Axboe 343f9c78b2bSJens Axboe while (1) { 344f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 345f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 346f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 347f9c78b2bSJens Axboe 348f9c78b2bSJens Axboe if (!bio) 349f9c78b2bSJens Axboe break; 350f9c78b2bSJens Axboe 351f9c78b2bSJens Axboe generic_make_request(bio); 352f9c78b2bSJens Axboe } 353f9c78b2bSJens Axboe } 354f9c78b2bSJens Axboe 355f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 356f9c78b2bSJens Axboe { 357f9c78b2bSJens Axboe struct bio_list punt, nopunt; 358f9c78b2bSJens Axboe struct bio *bio; 359f9c78b2bSJens Axboe 360f9c78b2bSJens Axboe /* 361f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 362f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 363f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 364f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 365f9c78b2bSJens Axboe * our own rescuer would be bad. 366f9c78b2bSJens Axboe * 367f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 368f9c78b2bSJens Axboe * remove from the middle of the list: 369f9c78b2bSJens Axboe */ 370f9c78b2bSJens Axboe 371f9c78b2bSJens Axboe bio_list_init(&punt); 372f9c78b2bSJens Axboe bio_list_init(&nopunt); 373f9c78b2bSJens Axboe 374f9c78b2bSJens Axboe while ((bio = bio_list_pop(current->bio_list))) 375f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 376f9c78b2bSJens Axboe 377f9c78b2bSJens Axboe *current->bio_list = nopunt; 378f9c78b2bSJens Axboe 379f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 380f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 381f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 382f9c78b2bSJens Axboe 383f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 384f9c78b2bSJens Axboe } 385f9c78b2bSJens Axboe 386f9c78b2bSJens Axboe /** 387f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 388f9c78b2bSJens Axboe * @gfp_mask: the GFP_ mask given to the slab allocator 389f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 390f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 391f9c78b2bSJens Axboe * 392f9c78b2bSJens Axboe * Description: 393f9c78b2bSJens Axboe * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 394f9c78b2bSJens Axboe * backed by the @bs's mempool. 395f9c78b2bSJens Axboe * 396f9c78b2bSJens Axboe * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be 397f9c78b2bSJens Axboe * able to allocate a bio. This is due to the mempool guarantees. To make this 398f9c78b2bSJens Axboe * work, callers must never allocate more than 1 bio at a time from this pool. 399f9c78b2bSJens Axboe * Callers that need to allocate more than 1 bio must always submit the 400f9c78b2bSJens Axboe * previously allocated bio for IO before attempting to allocate a new one. 401f9c78b2bSJens Axboe * Failure to do so can cause deadlocks under memory pressure. 402f9c78b2bSJens Axboe * 403f9c78b2bSJens Axboe * Note that when running under generic_make_request() (i.e. any block 404f9c78b2bSJens Axboe * driver), bios are not submitted until after you return - see the code in 405f9c78b2bSJens Axboe * generic_make_request() that converts recursion into iteration, to prevent 406f9c78b2bSJens Axboe * stack overflows. 407f9c78b2bSJens Axboe * 408f9c78b2bSJens Axboe * This would normally mean allocating multiple bios under 409f9c78b2bSJens Axboe * generic_make_request() would be susceptible to deadlocks, but we have 410f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 411f9c78b2bSJens Axboe * thread. 412f9c78b2bSJens Axboe * 413f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 414f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 415f9c78b2bSJens Axboe * generic_make_request() should be avoided - instead, use bio_set's front_pad 416f9c78b2bSJens Axboe * for per bio allocations. 417f9c78b2bSJens Axboe * 418f9c78b2bSJens Axboe * RETURNS: 419f9c78b2bSJens Axboe * Pointer to new bio on success, NULL on failure. 420f9c78b2bSJens Axboe */ 421f9c78b2bSJens Axboe struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) 422f9c78b2bSJens Axboe { 423f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 424f9c78b2bSJens Axboe unsigned front_pad; 425f9c78b2bSJens Axboe unsigned inline_vecs; 426f9c78b2bSJens Axboe unsigned long idx = BIO_POOL_NONE; 427f9c78b2bSJens Axboe struct bio_vec *bvl = NULL; 428f9c78b2bSJens Axboe struct bio *bio; 429f9c78b2bSJens Axboe void *p; 430f9c78b2bSJens Axboe 431f9c78b2bSJens Axboe if (!bs) { 432f9c78b2bSJens Axboe if (nr_iovecs > UIO_MAXIOV) 433f9c78b2bSJens Axboe return NULL; 434f9c78b2bSJens Axboe 435f9c78b2bSJens Axboe p = kmalloc(sizeof(struct bio) + 436f9c78b2bSJens Axboe nr_iovecs * sizeof(struct bio_vec), 437f9c78b2bSJens Axboe gfp_mask); 438f9c78b2bSJens Axboe front_pad = 0; 439f9c78b2bSJens Axboe inline_vecs = nr_iovecs; 440f9c78b2bSJens Axboe } else { 441d8f429e1SJunichi Nomura /* should not use nobvec bioset for nr_iovecs > 0 */ 442d8f429e1SJunichi Nomura if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0)) 443d8f429e1SJunichi Nomura return NULL; 444f9c78b2bSJens Axboe /* 445f9c78b2bSJens Axboe * generic_make_request() converts recursion to iteration; this 446f9c78b2bSJens Axboe * means if we're running beneath it, any bios we allocate and 447f9c78b2bSJens Axboe * submit will not be submitted (and thus freed) until after we 448f9c78b2bSJens Axboe * return. 449f9c78b2bSJens Axboe * 450f9c78b2bSJens Axboe * This exposes us to a potential deadlock if we allocate 451f9c78b2bSJens Axboe * multiple bios from the same bio_set() while running 452f9c78b2bSJens Axboe * underneath generic_make_request(). If we were to allocate 453f9c78b2bSJens Axboe * multiple bios (say a stacking block driver that was splitting 454f9c78b2bSJens Axboe * bios), we would deadlock if we exhausted the mempool's 455f9c78b2bSJens Axboe * reserve. 456f9c78b2bSJens Axboe * 457f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 458f9c78b2bSJens Axboe * workqueue per bio_set. If we go to allocate and there are 459f9c78b2bSJens Axboe * bios on current->bio_list, we first try the allocation 460f9c78b2bSJens Axboe * without __GFP_WAIT; if that fails, we punt those bios we 461f9c78b2bSJens Axboe * would be blocking to the rescuer workqueue before we retry 462f9c78b2bSJens Axboe * with the original gfp_flags. 463f9c78b2bSJens Axboe */ 464f9c78b2bSJens Axboe 465f9c78b2bSJens Axboe if (current->bio_list && !bio_list_empty(current->bio_list)) 466f9c78b2bSJens Axboe gfp_mask &= ~__GFP_WAIT; 467f9c78b2bSJens Axboe 468f9c78b2bSJens Axboe p = mempool_alloc(bs->bio_pool, gfp_mask); 469f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 470f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 471f9c78b2bSJens Axboe gfp_mask = saved_gfp; 472f9c78b2bSJens Axboe p = mempool_alloc(bs->bio_pool, gfp_mask); 473f9c78b2bSJens Axboe } 474f9c78b2bSJens Axboe 475f9c78b2bSJens Axboe front_pad = bs->front_pad; 476f9c78b2bSJens Axboe inline_vecs = BIO_INLINE_VECS; 477f9c78b2bSJens Axboe } 478f9c78b2bSJens Axboe 479f9c78b2bSJens Axboe if (unlikely(!p)) 480f9c78b2bSJens Axboe return NULL; 481f9c78b2bSJens Axboe 482f9c78b2bSJens Axboe bio = p + front_pad; 483f9c78b2bSJens Axboe bio_init(bio); 484f9c78b2bSJens Axboe 485f9c78b2bSJens Axboe if (nr_iovecs > inline_vecs) { 486f9c78b2bSJens Axboe bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); 487f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 488f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 489f9c78b2bSJens Axboe gfp_mask = saved_gfp; 490f9c78b2bSJens Axboe bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); 491f9c78b2bSJens Axboe } 492f9c78b2bSJens Axboe 493f9c78b2bSJens Axboe if (unlikely(!bvl)) 494f9c78b2bSJens Axboe goto err_free; 495f9c78b2bSJens Axboe 496f9c78b2bSJens Axboe bio->bi_flags |= 1 << BIO_OWNS_VEC; 497f9c78b2bSJens Axboe } else if (nr_iovecs) { 498f9c78b2bSJens Axboe bvl = bio->bi_inline_vecs; 499f9c78b2bSJens Axboe } 500f9c78b2bSJens Axboe 501f9c78b2bSJens Axboe bio->bi_pool = bs; 502f9c78b2bSJens Axboe bio->bi_flags |= idx << BIO_POOL_OFFSET; 503f9c78b2bSJens Axboe bio->bi_max_vecs = nr_iovecs; 504f9c78b2bSJens Axboe bio->bi_io_vec = bvl; 505f9c78b2bSJens Axboe return bio; 506f9c78b2bSJens Axboe 507f9c78b2bSJens Axboe err_free: 508f9c78b2bSJens Axboe mempool_free(p, bs->bio_pool); 509f9c78b2bSJens Axboe return NULL; 510f9c78b2bSJens Axboe } 511f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 512f9c78b2bSJens Axboe 513f9c78b2bSJens Axboe void zero_fill_bio(struct bio *bio) 514f9c78b2bSJens Axboe { 515f9c78b2bSJens Axboe unsigned long flags; 516f9c78b2bSJens Axboe struct bio_vec bv; 517f9c78b2bSJens Axboe struct bvec_iter iter; 518f9c78b2bSJens Axboe 519f9c78b2bSJens Axboe bio_for_each_segment(bv, bio, iter) { 520f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 521f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 522f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 523f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 524f9c78b2bSJens Axboe } 525f9c78b2bSJens Axboe } 526f9c78b2bSJens Axboe EXPORT_SYMBOL(zero_fill_bio); 527f9c78b2bSJens Axboe 528f9c78b2bSJens Axboe /** 529f9c78b2bSJens Axboe * bio_put - release a reference to a bio 530f9c78b2bSJens Axboe * @bio: bio to release reference to 531f9c78b2bSJens Axboe * 532f9c78b2bSJens Axboe * Description: 533f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 534f9c78b2bSJens Axboe * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. 535f9c78b2bSJens Axboe **/ 536f9c78b2bSJens Axboe void bio_put(struct bio *bio) 537f9c78b2bSJens Axboe { 538dac56212SJens Axboe if (!bio_flagged(bio, BIO_REFFED)) 539dac56212SJens Axboe bio_free(bio); 540dac56212SJens Axboe else { 541dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 542f9c78b2bSJens Axboe 543f9c78b2bSJens Axboe /* 544f9c78b2bSJens Axboe * last put frees it 545f9c78b2bSJens Axboe */ 546dac56212SJens Axboe if (atomic_dec_and_test(&bio->__bi_cnt)) 547f9c78b2bSJens Axboe bio_free(bio); 548f9c78b2bSJens Axboe } 549dac56212SJens Axboe } 550f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 551f9c78b2bSJens Axboe 552f9c78b2bSJens Axboe inline int bio_phys_segments(struct request_queue *q, struct bio *bio) 553f9c78b2bSJens Axboe { 554f9c78b2bSJens Axboe if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 555f9c78b2bSJens Axboe blk_recount_segments(q, bio); 556f9c78b2bSJens Axboe 557f9c78b2bSJens Axboe return bio->bi_phys_segments; 558f9c78b2bSJens Axboe } 559f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_phys_segments); 560f9c78b2bSJens Axboe 561f9c78b2bSJens Axboe /** 562f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 563f9c78b2bSJens Axboe * @bio: destination bio 564f9c78b2bSJens Axboe * @bio_src: bio to clone 565f9c78b2bSJens Axboe * 566f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 567f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 568f9c78b2bSJens Axboe * bio will be one. 569f9c78b2bSJens Axboe * 570f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 571f9c78b2bSJens Axboe */ 572f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 573f9c78b2bSJens Axboe { 574f9c78b2bSJens Axboe BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE); 575f9c78b2bSJens Axboe 576f9c78b2bSJens Axboe /* 577f9c78b2bSJens Axboe * most users will be overriding ->bi_bdev with a new target, 578f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 579f9c78b2bSJens Axboe */ 580f9c78b2bSJens Axboe bio->bi_bdev = bio_src->bi_bdev; 581f9c78b2bSJens Axboe bio->bi_flags |= 1 << BIO_CLONED; 582f9c78b2bSJens Axboe bio->bi_rw = bio_src->bi_rw; 583f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 584f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 585f9c78b2bSJens Axboe } 586f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 587f9c78b2bSJens Axboe 588f9c78b2bSJens Axboe /** 589f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 590f9c78b2bSJens Axboe * @bio: bio to clone 591f9c78b2bSJens Axboe * @gfp_mask: allocation priority 592f9c78b2bSJens Axboe * @bs: bio_set to allocate from 593f9c78b2bSJens Axboe * 594f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 595f9c78b2bSJens Axboe */ 596f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 597f9c78b2bSJens Axboe { 598f9c78b2bSJens Axboe struct bio *b; 599f9c78b2bSJens Axboe 600f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 601f9c78b2bSJens Axboe if (!b) 602f9c78b2bSJens Axboe return NULL; 603f9c78b2bSJens Axboe 604f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 605f9c78b2bSJens Axboe 606f9c78b2bSJens Axboe if (bio_integrity(bio)) { 607f9c78b2bSJens Axboe int ret; 608f9c78b2bSJens Axboe 609f9c78b2bSJens Axboe ret = bio_integrity_clone(b, bio, gfp_mask); 610f9c78b2bSJens Axboe 611f9c78b2bSJens Axboe if (ret < 0) { 612f9c78b2bSJens Axboe bio_put(b); 613f9c78b2bSJens Axboe return NULL; 614f9c78b2bSJens Axboe } 615f9c78b2bSJens Axboe } 616f9c78b2bSJens Axboe 617f9c78b2bSJens Axboe return b; 618f9c78b2bSJens Axboe } 619f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 620f9c78b2bSJens Axboe 621f9c78b2bSJens Axboe /** 622f9c78b2bSJens Axboe * bio_clone_bioset - clone a bio 623f9c78b2bSJens Axboe * @bio_src: bio to clone 624f9c78b2bSJens Axboe * @gfp_mask: allocation priority 625f9c78b2bSJens Axboe * @bs: bio_set to allocate from 626f9c78b2bSJens Axboe * 627f9c78b2bSJens Axboe * Clone bio. Caller will own the returned bio, but not the actual data it 628f9c78b2bSJens Axboe * points to. Reference count of returned bio will be one. 629f9c78b2bSJens Axboe */ 630f9c78b2bSJens Axboe struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, 631f9c78b2bSJens Axboe struct bio_set *bs) 632f9c78b2bSJens Axboe { 633f9c78b2bSJens Axboe struct bvec_iter iter; 634f9c78b2bSJens Axboe struct bio_vec bv; 635f9c78b2bSJens Axboe struct bio *bio; 636f9c78b2bSJens Axboe 637f9c78b2bSJens Axboe /* 638f9c78b2bSJens Axboe * Pre immutable biovecs, __bio_clone() used to just do a memcpy from 639f9c78b2bSJens Axboe * bio_src->bi_io_vec to bio->bi_io_vec. 640f9c78b2bSJens Axboe * 641f9c78b2bSJens Axboe * We can't do that anymore, because: 642f9c78b2bSJens Axboe * 643f9c78b2bSJens Axboe * - The point of cloning the biovec is to produce a bio with a biovec 644f9c78b2bSJens Axboe * the caller can modify: bi_idx and bi_bvec_done should be 0. 645f9c78b2bSJens Axboe * 646f9c78b2bSJens Axboe * - The original bio could've had more than BIO_MAX_PAGES biovecs; if 647f9c78b2bSJens Axboe * we tried to clone the whole thing bio_alloc_bioset() would fail. 648f9c78b2bSJens Axboe * But the clone should succeed as long as the number of biovecs we 649f9c78b2bSJens Axboe * actually need to allocate is fewer than BIO_MAX_PAGES. 650f9c78b2bSJens Axboe * 651f9c78b2bSJens Axboe * - Lastly, bi_vcnt should not be looked at or relied upon by code 652f9c78b2bSJens Axboe * that does not own the bio - reason being drivers don't use it for 653f9c78b2bSJens Axboe * iterating over the biovec anymore, so expecting it to be kept up 654f9c78b2bSJens Axboe * to date (i.e. for clones that share the parent biovec) is just 655f9c78b2bSJens Axboe * asking for trouble and would force extra work on 656f9c78b2bSJens Axboe * __bio_clone_fast() anyways. 657f9c78b2bSJens Axboe */ 658f9c78b2bSJens Axboe 659f9c78b2bSJens Axboe bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs); 660f9c78b2bSJens Axboe if (!bio) 661f9c78b2bSJens Axboe return NULL; 662f9c78b2bSJens Axboe 663f9c78b2bSJens Axboe bio->bi_bdev = bio_src->bi_bdev; 664f9c78b2bSJens Axboe bio->bi_rw = bio_src->bi_rw; 665f9c78b2bSJens Axboe bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; 666f9c78b2bSJens Axboe bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; 667f9c78b2bSJens Axboe 668f9c78b2bSJens Axboe if (bio->bi_rw & REQ_DISCARD) 669f9c78b2bSJens Axboe goto integrity_clone; 670f9c78b2bSJens Axboe 671f9c78b2bSJens Axboe if (bio->bi_rw & REQ_WRITE_SAME) { 672f9c78b2bSJens Axboe bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; 673f9c78b2bSJens Axboe goto integrity_clone; 674f9c78b2bSJens Axboe } 675f9c78b2bSJens Axboe 676f9c78b2bSJens Axboe bio_for_each_segment(bv, bio_src, iter) 677f9c78b2bSJens Axboe bio->bi_io_vec[bio->bi_vcnt++] = bv; 678f9c78b2bSJens Axboe 679f9c78b2bSJens Axboe integrity_clone: 680f9c78b2bSJens Axboe if (bio_integrity(bio_src)) { 681f9c78b2bSJens Axboe int ret; 682f9c78b2bSJens Axboe 683f9c78b2bSJens Axboe ret = bio_integrity_clone(bio, bio_src, gfp_mask); 684f9c78b2bSJens Axboe if (ret < 0) { 685f9c78b2bSJens Axboe bio_put(bio); 686f9c78b2bSJens Axboe return NULL; 687f9c78b2bSJens Axboe } 688f9c78b2bSJens Axboe } 689f9c78b2bSJens Axboe 690f9c78b2bSJens Axboe return bio; 691f9c78b2bSJens Axboe } 692f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_bioset); 693f9c78b2bSJens Axboe 694f9c78b2bSJens Axboe /** 695f9c78b2bSJens Axboe * bio_get_nr_vecs - return approx number of vecs 696f9c78b2bSJens Axboe * @bdev: I/O target 697f9c78b2bSJens Axboe * 698f9c78b2bSJens Axboe * Return the approximate number of pages we can send to this target. 699f9c78b2bSJens Axboe * There's no guarantee that you will be able to fit this number of pages 700f9c78b2bSJens Axboe * into a bio, it does not account for dynamic restrictions that vary 701f9c78b2bSJens Axboe * on offset. 702f9c78b2bSJens Axboe */ 703f9c78b2bSJens Axboe int bio_get_nr_vecs(struct block_device *bdev) 704f9c78b2bSJens Axboe { 705f9c78b2bSJens Axboe struct request_queue *q = bdev_get_queue(bdev); 706f9c78b2bSJens Axboe int nr_pages; 707f9c78b2bSJens Axboe 708f9c78b2bSJens Axboe nr_pages = min_t(unsigned, 709f9c78b2bSJens Axboe queue_max_segments(q), 710f9c78b2bSJens Axboe queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1); 711f9c78b2bSJens Axboe 712f9c78b2bSJens Axboe return min_t(unsigned, nr_pages, BIO_MAX_PAGES); 713f9c78b2bSJens Axboe 714f9c78b2bSJens Axboe } 715f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_get_nr_vecs); 716f9c78b2bSJens Axboe 717f9c78b2bSJens Axboe static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page 718f9c78b2bSJens Axboe *page, unsigned int len, unsigned int offset, 719f9c78b2bSJens Axboe unsigned int max_sectors) 720f9c78b2bSJens Axboe { 721f9c78b2bSJens Axboe int retried_segments = 0; 722f9c78b2bSJens Axboe struct bio_vec *bvec; 723f9c78b2bSJens Axboe 724f9c78b2bSJens Axboe /* 725f9c78b2bSJens Axboe * cloned bio must not modify vec list 726f9c78b2bSJens Axboe */ 727f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 728f9c78b2bSJens Axboe return 0; 729f9c78b2bSJens Axboe 730f9c78b2bSJens Axboe if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 731f9c78b2bSJens Axboe return 0; 732f9c78b2bSJens Axboe 733f9c78b2bSJens Axboe /* 734f9c78b2bSJens Axboe * For filesystems with a blocksize smaller than the pagesize 735f9c78b2bSJens Axboe * we will often be called with the same page as last time and 736f9c78b2bSJens Axboe * a consecutive offset. Optimize this special case. 737f9c78b2bSJens Axboe */ 738f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 739f9c78b2bSJens Axboe struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; 740f9c78b2bSJens Axboe 741f9c78b2bSJens Axboe if (page == prev->bv_page && 742f9c78b2bSJens Axboe offset == prev->bv_offset + prev->bv_len) { 743f9c78b2bSJens Axboe unsigned int prev_bv_len = prev->bv_len; 744f9c78b2bSJens Axboe prev->bv_len += len; 745f9c78b2bSJens Axboe 746f9c78b2bSJens Axboe if (q->merge_bvec_fn) { 747f9c78b2bSJens Axboe struct bvec_merge_data bvm = { 748f9c78b2bSJens Axboe /* prev_bvec is already charged in 749f9c78b2bSJens Axboe bi_size, discharge it in order to 750f9c78b2bSJens Axboe simulate merging updated prev_bvec 751f9c78b2bSJens Axboe as new bvec. */ 752f9c78b2bSJens Axboe .bi_bdev = bio->bi_bdev, 753f9c78b2bSJens Axboe .bi_sector = bio->bi_iter.bi_sector, 754f9c78b2bSJens Axboe .bi_size = bio->bi_iter.bi_size - 755f9c78b2bSJens Axboe prev_bv_len, 756f9c78b2bSJens Axboe .bi_rw = bio->bi_rw, 757f9c78b2bSJens Axboe }; 758f9c78b2bSJens Axboe 759f9c78b2bSJens Axboe if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { 760f9c78b2bSJens Axboe prev->bv_len -= len; 761f9c78b2bSJens Axboe return 0; 762f9c78b2bSJens Axboe } 763f9c78b2bSJens Axboe } 764f9c78b2bSJens Axboe 765fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 766f9c78b2bSJens Axboe goto done; 767f9c78b2bSJens Axboe } 76866cb45aaSJens Axboe 76966cb45aaSJens Axboe /* 77066cb45aaSJens Axboe * If the queue doesn't support SG gaps and adding this 77166cb45aaSJens Axboe * offset would create a gap, disallow it. 77266cb45aaSJens Axboe */ 77366cb45aaSJens Axboe if (q->queue_flags & (1 << QUEUE_FLAG_SG_GAPS) && 77466cb45aaSJens Axboe bvec_gap_to_prev(prev, offset)) 77566cb45aaSJens Axboe return 0; 776f9c78b2bSJens Axboe } 777f9c78b2bSJens Axboe 778f9c78b2bSJens Axboe if (bio->bi_vcnt >= bio->bi_max_vecs) 779f9c78b2bSJens Axboe return 0; 780f9c78b2bSJens Axboe 781f9c78b2bSJens Axboe /* 782f9c78b2bSJens Axboe * setup the new entry, we might clear it again later if we 783f9c78b2bSJens Axboe * cannot add the page 784f9c78b2bSJens Axboe */ 785f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 786f9c78b2bSJens Axboe bvec->bv_page = page; 787f9c78b2bSJens Axboe bvec->bv_len = len; 788f9c78b2bSJens Axboe bvec->bv_offset = offset; 789fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 790fcbf6a08SMaurizio Lombardi bio->bi_phys_segments++; 791fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size += len; 792fcbf6a08SMaurizio Lombardi 793fcbf6a08SMaurizio Lombardi /* 794fcbf6a08SMaurizio Lombardi * Perform a recount if the number of segments is greater 795fcbf6a08SMaurizio Lombardi * than queue_max_segments(q). 796fcbf6a08SMaurizio Lombardi */ 797fcbf6a08SMaurizio Lombardi 798fcbf6a08SMaurizio Lombardi while (bio->bi_phys_segments > queue_max_segments(q)) { 799fcbf6a08SMaurizio Lombardi 800fcbf6a08SMaurizio Lombardi if (retried_segments) 801fcbf6a08SMaurizio Lombardi goto failed; 802fcbf6a08SMaurizio Lombardi 803fcbf6a08SMaurizio Lombardi retried_segments = 1; 804fcbf6a08SMaurizio Lombardi blk_recount_segments(q, bio); 805fcbf6a08SMaurizio Lombardi } 806f9c78b2bSJens Axboe 807f9c78b2bSJens Axboe /* 808f9c78b2bSJens Axboe * if queue has other restrictions (eg varying max sector size 809f9c78b2bSJens Axboe * depending on offset), it can specify a merge_bvec_fn in the 810f9c78b2bSJens Axboe * queue to get further control 811f9c78b2bSJens Axboe */ 812f9c78b2bSJens Axboe if (q->merge_bvec_fn) { 813f9c78b2bSJens Axboe struct bvec_merge_data bvm = { 814f9c78b2bSJens Axboe .bi_bdev = bio->bi_bdev, 815f9c78b2bSJens Axboe .bi_sector = bio->bi_iter.bi_sector, 816fcbf6a08SMaurizio Lombardi .bi_size = bio->bi_iter.bi_size - len, 817f9c78b2bSJens Axboe .bi_rw = bio->bi_rw, 818f9c78b2bSJens Axboe }; 819f9c78b2bSJens Axboe 820f9c78b2bSJens Axboe /* 821f9c78b2bSJens Axboe * merge_bvec_fn() returns number of bytes it can accept 822f9c78b2bSJens Axboe * at this offset 823f9c78b2bSJens Axboe */ 824fcbf6a08SMaurizio Lombardi if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) 825fcbf6a08SMaurizio Lombardi goto failed; 826f9c78b2bSJens Axboe } 827f9c78b2bSJens Axboe 828f9c78b2bSJens Axboe /* If we may be able to merge these biovecs, force a recount */ 829fcbf6a08SMaurizio Lombardi if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) 830f9c78b2bSJens Axboe bio->bi_flags &= ~(1 << BIO_SEG_VALID); 831f9c78b2bSJens Axboe 832f9c78b2bSJens Axboe done: 833f9c78b2bSJens Axboe return len; 834fcbf6a08SMaurizio Lombardi 835fcbf6a08SMaurizio Lombardi failed: 836fcbf6a08SMaurizio Lombardi bvec->bv_page = NULL; 837fcbf6a08SMaurizio Lombardi bvec->bv_len = 0; 838fcbf6a08SMaurizio Lombardi bvec->bv_offset = 0; 839fcbf6a08SMaurizio Lombardi bio->bi_vcnt--; 840fcbf6a08SMaurizio Lombardi bio->bi_iter.bi_size -= len; 841fcbf6a08SMaurizio Lombardi blk_recount_segments(q, bio); 842fcbf6a08SMaurizio Lombardi return 0; 843f9c78b2bSJens Axboe } 844f9c78b2bSJens Axboe 845f9c78b2bSJens Axboe /** 846f9c78b2bSJens Axboe * bio_add_pc_page - attempt to add page to bio 847f9c78b2bSJens Axboe * @q: the target queue 848f9c78b2bSJens Axboe * @bio: destination bio 849f9c78b2bSJens Axboe * @page: page to add 850f9c78b2bSJens Axboe * @len: vec entry length 851f9c78b2bSJens Axboe * @offset: vec entry offset 852f9c78b2bSJens Axboe * 853f9c78b2bSJens Axboe * Attempt to add a page to the bio_vec maplist. This can fail for a 854f9c78b2bSJens Axboe * number of reasons, such as the bio being full or target block device 855f9c78b2bSJens Axboe * limitations. The target block device must allow bio's up to PAGE_SIZE, 856f9c78b2bSJens Axboe * so it is always possible to add a single page to an empty bio. 857f9c78b2bSJens Axboe * 858f9c78b2bSJens Axboe * This should only be used by REQ_PC bios. 859f9c78b2bSJens Axboe */ 860f9c78b2bSJens Axboe int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, 861f9c78b2bSJens Axboe unsigned int len, unsigned int offset) 862f9c78b2bSJens Axboe { 863f9c78b2bSJens Axboe return __bio_add_page(q, bio, page, len, offset, 864f9c78b2bSJens Axboe queue_max_hw_sectors(q)); 865f9c78b2bSJens Axboe } 866f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 867f9c78b2bSJens Axboe 868f9c78b2bSJens Axboe /** 869f9c78b2bSJens Axboe * bio_add_page - attempt to add page to bio 870f9c78b2bSJens Axboe * @bio: destination bio 871f9c78b2bSJens Axboe * @page: page to add 872f9c78b2bSJens Axboe * @len: vec entry length 873f9c78b2bSJens Axboe * @offset: vec entry offset 874f9c78b2bSJens Axboe * 875f9c78b2bSJens Axboe * Attempt to add a page to the bio_vec maplist. This can fail for a 876f9c78b2bSJens Axboe * number of reasons, such as the bio being full or target block device 877f9c78b2bSJens Axboe * limitations. The target block device must allow bio's up to PAGE_SIZE, 878f9c78b2bSJens Axboe * so it is always possible to add a single page to an empty bio. 879f9c78b2bSJens Axboe */ 880f9c78b2bSJens Axboe int bio_add_page(struct bio *bio, struct page *page, unsigned int len, 881f9c78b2bSJens Axboe unsigned int offset) 882f9c78b2bSJens Axboe { 883f9c78b2bSJens Axboe struct request_queue *q = bdev_get_queue(bio->bi_bdev); 88458a4915aSJens Axboe unsigned int max_sectors; 885762380adSJens Axboe 88658a4915aSJens Axboe max_sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector); 88758a4915aSJens Axboe if ((max_sectors < (len >> 9)) && !bio->bi_iter.bi_size) 88858a4915aSJens Axboe max_sectors = len >> 9; 88958a4915aSJens Axboe 89058a4915aSJens Axboe return __bio_add_page(q, bio, page, len, offset, max_sectors); 891f9c78b2bSJens Axboe } 892f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 893f9c78b2bSJens Axboe 894f9c78b2bSJens Axboe struct submit_bio_ret { 895f9c78b2bSJens Axboe struct completion event; 896f9c78b2bSJens Axboe int error; 897f9c78b2bSJens Axboe }; 898f9c78b2bSJens Axboe 899f9c78b2bSJens Axboe static void submit_bio_wait_endio(struct bio *bio, int error) 900f9c78b2bSJens Axboe { 901f9c78b2bSJens Axboe struct submit_bio_ret *ret = bio->bi_private; 902f9c78b2bSJens Axboe 903f9c78b2bSJens Axboe ret->error = error; 904f9c78b2bSJens Axboe complete(&ret->event); 905f9c78b2bSJens Axboe } 906f9c78b2bSJens Axboe 907f9c78b2bSJens Axboe /** 908f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 909f9c78b2bSJens Axboe * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 910f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 911f9c78b2bSJens Axboe * 912f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 913f9c78b2bSJens Axboe * bio_endio() on failure. 914f9c78b2bSJens Axboe */ 915f9c78b2bSJens Axboe int submit_bio_wait(int rw, struct bio *bio) 916f9c78b2bSJens Axboe { 917f9c78b2bSJens Axboe struct submit_bio_ret ret; 918f9c78b2bSJens Axboe 919f9c78b2bSJens Axboe rw |= REQ_SYNC; 920f9c78b2bSJens Axboe init_completion(&ret.event); 921f9c78b2bSJens Axboe bio->bi_private = &ret; 922f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 923f9c78b2bSJens Axboe submit_bio(rw, bio); 924f9c78b2bSJens Axboe wait_for_completion(&ret.event); 925f9c78b2bSJens Axboe 926f9c78b2bSJens Axboe return ret.error; 927f9c78b2bSJens Axboe } 928f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 929f9c78b2bSJens Axboe 930f9c78b2bSJens Axboe /** 931f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 932f9c78b2bSJens Axboe * @bio: bio to advance 933f9c78b2bSJens Axboe * @bytes: number of bytes to complete 934f9c78b2bSJens Axboe * 935f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 936f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 937f9c78b2bSJens Axboe * be updated on the last bvec as well. 938f9c78b2bSJens Axboe * 939f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 940f9c78b2bSJens Axboe */ 941f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 942f9c78b2bSJens Axboe { 943f9c78b2bSJens Axboe if (bio_integrity(bio)) 944f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 945f9c78b2bSJens Axboe 946f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 947f9c78b2bSJens Axboe } 948f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 949f9c78b2bSJens Axboe 950f9c78b2bSJens Axboe /** 951f9c78b2bSJens Axboe * bio_alloc_pages - allocates a single page for each bvec in a bio 952f9c78b2bSJens Axboe * @bio: bio to allocate pages for 953f9c78b2bSJens Axboe * @gfp_mask: flags for allocation 954f9c78b2bSJens Axboe * 955f9c78b2bSJens Axboe * Allocates pages up to @bio->bi_vcnt. 956f9c78b2bSJens Axboe * 957f9c78b2bSJens Axboe * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are 958f9c78b2bSJens Axboe * freed. 959f9c78b2bSJens Axboe */ 960f9c78b2bSJens Axboe int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) 961f9c78b2bSJens Axboe { 962f9c78b2bSJens Axboe int i; 963f9c78b2bSJens Axboe struct bio_vec *bv; 964f9c78b2bSJens Axboe 965f9c78b2bSJens Axboe bio_for_each_segment_all(bv, bio, i) { 966f9c78b2bSJens Axboe bv->bv_page = alloc_page(gfp_mask); 967f9c78b2bSJens Axboe if (!bv->bv_page) { 968f9c78b2bSJens Axboe while (--bv >= bio->bi_io_vec) 969f9c78b2bSJens Axboe __free_page(bv->bv_page); 970f9c78b2bSJens Axboe return -ENOMEM; 971f9c78b2bSJens Axboe } 972f9c78b2bSJens Axboe } 973f9c78b2bSJens Axboe 974f9c78b2bSJens Axboe return 0; 975f9c78b2bSJens Axboe } 976f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_pages); 977f9c78b2bSJens Axboe 978f9c78b2bSJens Axboe /** 979f9c78b2bSJens Axboe * bio_copy_data - copy contents of data buffers from one chain of bios to 980f9c78b2bSJens Axboe * another 981f9c78b2bSJens Axboe * @src: source bio list 982f9c78b2bSJens Axboe * @dst: destination bio list 983f9c78b2bSJens Axboe * 984f9c78b2bSJens Axboe * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats 985f9c78b2bSJens Axboe * @src and @dst as linked lists of bios. 986f9c78b2bSJens Axboe * 987f9c78b2bSJens Axboe * Stops when it reaches the end of either @src or @dst - that is, copies 988f9c78b2bSJens Axboe * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 989f9c78b2bSJens Axboe */ 990f9c78b2bSJens Axboe void bio_copy_data(struct bio *dst, struct bio *src) 991f9c78b2bSJens Axboe { 992f9c78b2bSJens Axboe struct bvec_iter src_iter, dst_iter; 993f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 994f9c78b2bSJens Axboe void *src_p, *dst_p; 995f9c78b2bSJens Axboe unsigned bytes; 996f9c78b2bSJens Axboe 997f9c78b2bSJens Axboe src_iter = src->bi_iter; 998f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 999f9c78b2bSJens Axboe 1000f9c78b2bSJens Axboe while (1) { 1001f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1002f9c78b2bSJens Axboe src = src->bi_next; 1003f9c78b2bSJens Axboe if (!src) 1004f9c78b2bSJens Axboe break; 1005f9c78b2bSJens Axboe 1006f9c78b2bSJens Axboe src_iter = src->bi_iter; 1007f9c78b2bSJens Axboe } 1008f9c78b2bSJens Axboe 1009f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1010f9c78b2bSJens Axboe dst = dst->bi_next; 1011f9c78b2bSJens Axboe if (!dst) 1012f9c78b2bSJens Axboe break; 1013f9c78b2bSJens Axboe 1014f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1015f9c78b2bSJens Axboe } 1016f9c78b2bSJens Axboe 1017f9c78b2bSJens Axboe src_bv = bio_iter_iovec(src, src_iter); 1018f9c78b2bSJens Axboe dst_bv = bio_iter_iovec(dst, dst_iter); 1019f9c78b2bSJens Axboe 1020f9c78b2bSJens Axboe bytes = min(src_bv.bv_len, dst_bv.bv_len); 1021f9c78b2bSJens Axboe 1022f9c78b2bSJens Axboe src_p = kmap_atomic(src_bv.bv_page); 1023f9c78b2bSJens Axboe dst_p = kmap_atomic(dst_bv.bv_page); 1024f9c78b2bSJens Axboe 1025f9c78b2bSJens Axboe memcpy(dst_p + dst_bv.bv_offset, 1026f9c78b2bSJens Axboe src_p + src_bv.bv_offset, 1027f9c78b2bSJens Axboe bytes); 1028f9c78b2bSJens Axboe 1029f9c78b2bSJens Axboe kunmap_atomic(dst_p); 1030f9c78b2bSJens Axboe kunmap_atomic(src_p); 1031f9c78b2bSJens Axboe 1032f9c78b2bSJens Axboe bio_advance_iter(src, &src_iter, bytes); 1033f9c78b2bSJens Axboe bio_advance_iter(dst, &dst_iter, bytes); 1034f9c78b2bSJens Axboe } 1035f9c78b2bSJens Axboe } 1036f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_copy_data); 1037f9c78b2bSJens Axboe 1038f9c78b2bSJens Axboe struct bio_map_data { 1039f9c78b2bSJens Axboe int is_our_pages; 104026e49cfcSKent Overstreet struct iov_iter iter; 104126e49cfcSKent Overstreet struct iovec iov[]; 1042f9c78b2bSJens Axboe }; 1043f9c78b2bSJens Axboe 1044f9c78b2bSJens Axboe static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count, 1045f9c78b2bSJens Axboe gfp_t gfp_mask) 1046f9c78b2bSJens Axboe { 1047f9c78b2bSJens Axboe if (iov_count > UIO_MAXIOV) 1048f9c78b2bSJens Axboe return NULL; 1049f9c78b2bSJens Axboe 1050f9c78b2bSJens Axboe return kmalloc(sizeof(struct bio_map_data) + 105126e49cfcSKent Overstreet sizeof(struct iovec) * iov_count, gfp_mask); 1052f9c78b2bSJens Axboe } 1053f9c78b2bSJens Axboe 10549124d3feSDongsu Park /** 10559124d3feSDongsu Park * bio_copy_from_iter - copy all pages from iov_iter to bio 10569124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as destination 10579124d3feSDongsu Park * @iter: iov_iter as source 10589124d3feSDongsu Park * 10599124d3feSDongsu Park * Copy all pages from iov_iter to bio. 10609124d3feSDongsu Park * Returns 0 on success, or error on failure. 10619124d3feSDongsu Park */ 10629124d3feSDongsu Park static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter) 1063f9c78b2bSJens Axboe { 10649124d3feSDongsu Park int i; 1065f9c78b2bSJens Axboe struct bio_vec *bvec; 1066f9c78b2bSJens Axboe 1067f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 10689124d3feSDongsu Park ssize_t ret; 1069f9c78b2bSJens Axboe 10709124d3feSDongsu Park ret = copy_page_from_iter(bvec->bv_page, 10719124d3feSDongsu Park bvec->bv_offset, 10729124d3feSDongsu Park bvec->bv_len, 10739124d3feSDongsu Park &iter); 1074f9c78b2bSJens Axboe 10759124d3feSDongsu Park if (!iov_iter_count(&iter)) 10769124d3feSDongsu Park break; 1077f9c78b2bSJens Axboe 10789124d3feSDongsu Park if (ret < bvec->bv_len) 10799124d3feSDongsu Park return -EFAULT; 1080f9c78b2bSJens Axboe } 1081f9c78b2bSJens Axboe 10829124d3feSDongsu Park return 0; 1083f9c78b2bSJens Axboe } 1084f9c78b2bSJens Axboe 10859124d3feSDongsu Park /** 10869124d3feSDongsu Park * bio_copy_to_iter - copy all pages from bio to iov_iter 10879124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as source 10889124d3feSDongsu Park * @iter: iov_iter as destination 10899124d3feSDongsu Park * 10909124d3feSDongsu Park * Copy all pages from bio to iov_iter. 10919124d3feSDongsu Park * Returns 0 on success, or error on failure. 10929124d3feSDongsu Park */ 10939124d3feSDongsu Park static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 10949124d3feSDongsu Park { 10959124d3feSDongsu Park int i; 10969124d3feSDongsu Park struct bio_vec *bvec; 10979124d3feSDongsu Park 10989124d3feSDongsu Park bio_for_each_segment_all(bvec, bio, i) { 10999124d3feSDongsu Park ssize_t ret; 11009124d3feSDongsu Park 11019124d3feSDongsu Park ret = copy_page_to_iter(bvec->bv_page, 11029124d3feSDongsu Park bvec->bv_offset, 11039124d3feSDongsu Park bvec->bv_len, 11049124d3feSDongsu Park &iter); 11059124d3feSDongsu Park 11069124d3feSDongsu Park if (!iov_iter_count(&iter)) 11079124d3feSDongsu Park break; 11089124d3feSDongsu Park 11099124d3feSDongsu Park if (ret < bvec->bv_len) 11109124d3feSDongsu Park return -EFAULT; 11119124d3feSDongsu Park } 11129124d3feSDongsu Park 11139124d3feSDongsu Park return 0; 1114f9c78b2bSJens Axboe } 1115f9c78b2bSJens Axboe 11161dfa0f68SChristoph Hellwig static void bio_free_pages(struct bio *bio) 11171dfa0f68SChristoph Hellwig { 11181dfa0f68SChristoph Hellwig struct bio_vec *bvec; 11191dfa0f68SChristoph Hellwig int i; 11201dfa0f68SChristoph Hellwig 11211dfa0f68SChristoph Hellwig bio_for_each_segment_all(bvec, bio, i) 11221dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 11231dfa0f68SChristoph Hellwig } 11241dfa0f68SChristoph Hellwig 1125f9c78b2bSJens Axboe /** 1126f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1127f9c78b2bSJens Axboe * @bio: bio being terminated 1128f9c78b2bSJens Axboe * 1129ddad8dd0SChristoph Hellwig * Free pages allocated from bio_copy_user_iov() and write back data 1130f9c78b2bSJens Axboe * to user space in case of a read. 1131f9c78b2bSJens Axboe */ 1132f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1133f9c78b2bSJens Axboe { 1134f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 11351dfa0f68SChristoph Hellwig int ret = 0; 1136f9c78b2bSJens Axboe 1137f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1138f9c78b2bSJens Axboe /* 1139f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 1140f9c78b2bSJens Axboe * don't copy into a random user address space, just free. 1141f9c78b2bSJens Axboe */ 11429124d3feSDongsu Park if (current->mm && bio_data_dir(bio) == READ) 11439124d3feSDongsu Park ret = bio_copy_to_iter(bio, bmd->iter); 11441dfa0f68SChristoph Hellwig if (bmd->is_our_pages) 11451dfa0f68SChristoph Hellwig bio_free_pages(bio); 1146f9c78b2bSJens Axboe } 1147f9c78b2bSJens Axboe kfree(bmd); 1148f9c78b2bSJens Axboe bio_put(bio); 1149f9c78b2bSJens Axboe return ret; 1150f9c78b2bSJens Axboe } 1151f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_uncopy_user); 1152f9c78b2bSJens Axboe 1153f9c78b2bSJens Axboe /** 1154f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1155f9c78b2bSJens Axboe * @q: destination block queue 1156f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 115726e49cfcSKent Overstreet * @iter: iovec iterator 1158f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1159f9c78b2bSJens Axboe * 1160f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1161f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1162f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1163f9c78b2bSJens Axboe */ 1164f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1165f9c78b2bSJens Axboe struct rq_map_data *map_data, 116626e49cfcSKent Overstreet const struct iov_iter *iter, 116726e49cfcSKent Overstreet gfp_t gfp_mask) 1168f9c78b2bSJens Axboe { 1169f9c78b2bSJens Axboe struct bio_map_data *bmd; 1170f9c78b2bSJens Axboe struct page *page; 1171f9c78b2bSJens Axboe struct bio *bio; 1172f9c78b2bSJens Axboe int i, ret; 1173f9c78b2bSJens Axboe int nr_pages = 0; 117426e49cfcSKent Overstreet unsigned int len = iter->count; 1175f9c78b2bSJens Axboe unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; 1176f9c78b2bSJens Axboe 117726e49cfcSKent Overstreet for (i = 0; i < iter->nr_segs; i++) { 1178f9c78b2bSJens Axboe unsigned long uaddr; 1179f9c78b2bSJens Axboe unsigned long end; 1180f9c78b2bSJens Axboe unsigned long start; 1181f9c78b2bSJens Axboe 118226e49cfcSKent Overstreet uaddr = (unsigned long) iter->iov[i].iov_base; 118326e49cfcSKent Overstreet end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1) 118426e49cfcSKent Overstreet >> PAGE_SHIFT; 1185f9c78b2bSJens Axboe start = uaddr >> PAGE_SHIFT; 1186f9c78b2bSJens Axboe 1187f9c78b2bSJens Axboe /* 1188f9c78b2bSJens Axboe * Overflow, abort 1189f9c78b2bSJens Axboe */ 1190f9c78b2bSJens Axboe if (end < start) 1191f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1192f9c78b2bSJens Axboe 1193f9c78b2bSJens Axboe nr_pages += end - start; 1194f9c78b2bSJens Axboe } 1195f9c78b2bSJens Axboe 1196f9c78b2bSJens Axboe if (offset) 1197f9c78b2bSJens Axboe nr_pages++; 1198f9c78b2bSJens Axboe 119926e49cfcSKent Overstreet bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask); 1200f9c78b2bSJens Axboe if (!bmd) 1201f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1202f9c78b2bSJens Axboe 120326e49cfcSKent Overstreet /* 120426e49cfcSKent Overstreet * We need to do a deep copy of the iov_iter including the iovecs. 120526e49cfcSKent Overstreet * The caller provided iov might point to an on-stack or otherwise 120626e49cfcSKent Overstreet * shortlived one. 120726e49cfcSKent Overstreet */ 120826e49cfcSKent Overstreet bmd->is_our_pages = map_data ? 0 : 1; 120926e49cfcSKent Overstreet memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs); 121026e49cfcSKent Overstreet iov_iter_init(&bmd->iter, iter->type, bmd->iov, 121126e49cfcSKent Overstreet iter->nr_segs, iter->count); 121226e49cfcSKent Overstreet 1213f9c78b2bSJens Axboe ret = -ENOMEM; 1214f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1215f9c78b2bSJens Axboe if (!bio) 1216f9c78b2bSJens Axboe goto out_bmd; 1217f9c78b2bSJens Axboe 121826e49cfcSKent Overstreet if (iter->type & WRITE) 1219f9c78b2bSJens Axboe bio->bi_rw |= REQ_WRITE; 1220f9c78b2bSJens Axboe 1221f9c78b2bSJens Axboe ret = 0; 1222f9c78b2bSJens Axboe 1223f9c78b2bSJens Axboe if (map_data) { 1224f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1225f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1226f9c78b2bSJens Axboe } 1227f9c78b2bSJens Axboe while (len) { 1228f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1229f9c78b2bSJens Axboe 1230f9c78b2bSJens Axboe bytes -= offset; 1231f9c78b2bSJens Axboe 1232f9c78b2bSJens Axboe if (bytes > len) 1233f9c78b2bSJens Axboe bytes = len; 1234f9c78b2bSJens Axboe 1235f9c78b2bSJens Axboe if (map_data) { 1236f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1237f9c78b2bSJens Axboe ret = -ENOMEM; 1238f9c78b2bSJens Axboe break; 1239f9c78b2bSJens Axboe } 1240f9c78b2bSJens Axboe 1241f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1242f9c78b2bSJens Axboe page += (i % nr_pages); 1243f9c78b2bSJens Axboe 1244f9c78b2bSJens Axboe i++; 1245f9c78b2bSJens Axboe } else { 1246f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1247f9c78b2bSJens Axboe if (!page) { 1248f9c78b2bSJens Axboe ret = -ENOMEM; 1249f9c78b2bSJens Axboe break; 1250f9c78b2bSJens Axboe } 1251f9c78b2bSJens Axboe } 1252f9c78b2bSJens Axboe 1253f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) 1254f9c78b2bSJens Axboe break; 1255f9c78b2bSJens Axboe 1256f9c78b2bSJens Axboe len -= bytes; 1257f9c78b2bSJens Axboe offset = 0; 1258f9c78b2bSJens Axboe } 1259f9c78b2bSJens Axboe 1260f9c78b2bSJens Axboe if (ret) 1261f9c78b2bSJens Axboe goto cleanup; 1262f9c78b2bSJens Axboe 1263f9c78b2bSJens Axboe /* 1264f9c78b2bSJens Axboe * success 1265f9c78b2bSJens Axboe */ 126626e49cfcSKent Overstreet if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) || 1267f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 12689124d3feSDongsu Park ret = bio_copy_from_iter(bio, *iter); 1269f9c78b2bSJens Axboe if (ret) 1270f9c78b2bSJens Axboe goto cleanup; 1271f9c78b2bSJens Axboe } 1272f9c78b2bSJens Axboe 127326e49cfcSKent Overstreet bio->bi_private = bmd; 1274f9c78b2bSJens Axboe return bio; 1275f9c78b2bSJens Axboe cleanup: 1276f9c78b2bSJens Axboe if (!map_data) 12771dfa0f68SChristoph Hellwig bio_free_pages(bio); 1278f9c78b2bSJens Axboe bio_put(bio); 1279f9c78b2bSJens Axboe out_bmd: 1280f9c78b2bSJens Axboe kfree(bmd); 1281f9c78b2bSJens Axboe return ERR_PTR(ret); 1282f9c78b2bSJens Axboe } 1283f9c78b2bSJens Axboe 128437f19e57SChristoph Hellwig /** 128537f19e57SChristoph Hellwig * bio_map_user_iov - map user iovec into bio 128637f19e57SChristoph Hellwig * @q: the struct request_queue for the bio 128737f19e57SChristoph Hellwig * @iter: iovec iterator 128837f19e57SChristoph Hellwig * @gfp_mask: memory allocation flags 128937f19e57SChristoph Hellwig * 129037f19e57SChristoph Hellwig * Map the user space address into a bio suitable for io to a block 129137f19e57SChristoph Hellwig * device. Returns an error pointer in case of error. 129237f19e57SChristoph Hellwig */ 129337f19e57SChristoph Hellwig struct bio *bio_map_user_iov(struct request_queue *q, 129426e49cfcSKent Overstreet const struct iov_iter *iter, 129526e49cfcSKent Overstreet gfp_t gfp_mask) 1296f9c78b2bSJens Axboe { 129726e49cfcSKent Overstreet int j; 1298f9c78b2bSJens Axboe int nr_pages = 0; 1299f9c78b2bSJens Axboe struct page **pages; 1300f9c78b2bSJens Axboe struct bio *bio; 1301f9c78b2bSJens Axboe int cur_page = 0; 1302f9c78b2bSJens Axboe int ret, offset; 130326e49cfcSKent Overstreet struct iov_iter i; 130426e49cfcSKent Overstreet struct iovec iov; 1305f9c78b2bSJens Axboe 130626e49cfcSKent Overstreet iov_for_each(iov, i, *iter) { 130726e49cfcSKent Overstreet unsigned long uaddr = (unsigned long) iov.iov_base; 130826e49cfcSKent Overstreet unsigned long len = iov.iov_len; 1309f9c78b2bSJens Axboe unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1310f9c78b2bSJens Axboe unsigned long start = uaddr >> PAGE_SHIFT; 1311f9c78b2bSJens Axboe 1312f9c78b2bSJens Axboe /* 1313f9c78b2bSJens Axboe * Overflow, abort 1314f9c78b2bSJens Axboe */ 1315f9c78b2bSJens Axboe if (end < start) 1316f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1317f9c78b2bSJens Axboe 1318f9c78b2bSJens Axboe nr_pages += end - start; 1319f9c78b2bSJens Axboe /* 1320f9c78b2bSJens Axboe * buffer must be aligned to at least hardsector size for now 1321f9c78b2bSJens Axboe */ 1322f9c78b2bSJens Axboe if (uaddr & queue_dma_alignment(q)) 1323f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1324f9c78b2bSJens Axboe } 1325f9c78b2bSJens Axboe 1326f9c78b2bSJens Axboe if (!nr_pages) 1327f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1328f9c78b2bSJens Axboe 1329f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1330f9c78b2bSJens Axboe if (!bio) 1331f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1332f9c78b2bSJens Axboe 1333f9c78b2bSJens Axboe ret = -ENOMEM; 1334f9c78b2bSJens Axboe pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); 1335f9c78b2bSJens Axboe if (!pages) 1336f9c78b2bSJens Axboe goto out; 1337f9c78b2bSJens Axboe 133826e49cfcSKent Overstreet iov_for_each(iov, i, *iter) { 133926e49cfcSKent Overstreet unsigned long uaddr = (unsigned long) iov.iov_base; 134026e49cfcSKent Overstreet unsigned long len = iov.iov_len; 1341f9c78b2bSJens Axboe unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1342f9c78b2bSJens Axboe unsigned long start = uaddr >> PAGE_SHIFT; 1343f9c78b2bSJens Axboe const int local_nr_pages = end - start; 1344f9c78b2bSJens Axboe const int page_limit = cur_page + local_nr_pages; 1345f9c78b2bSJens Axboe 1346f9c78b2bSJens Axboe ret = get_user_pages_fast(uaddr, local_nr_pages, 134726e49cfcSKent Overstreet (iter->type & WRITE) != WRITE, 134826e49cfcSKent Overstreet &pages[cur_page]); 1349f9c78b2bSJens Axboe if (ret < local_nr_pages) { 1350f9c78b2bSJens Axboe ret = -EFAULT; 1351f9c78b2bSJens Axboe goto out_unmap; 1352f9c78b2bSJens Axboe } 1353f9c78b2bSJens Axboe 1354f9c78b2bSJens Axboe offset = uaddr & ~PAGE_MASK; 1355f9c78b2bSJens Axboe for (j = cur_page; j < page_limit; j++) { 1356f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1357f9c78b2bSJens Axboe 1358f9c78b2bSJens Axboe if (len <= 0) 1359f9c78b2bSJens Axboe break; 1360f9c78b2bSJens Axboe 1361f9c78b2bSJens Axboe if (bytes > len) 1362f9c78b2bSJens Axboe bytes = len; 1363f9c78b2bSJens Axboe 1364f9c78b2bSJens Axboe /* 1365f9c78b2bSJens Axboe * sorry... 1366f9c78b2bSJens Axboe */ 1367f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < 1368f9c78b2bSJens Axboe bytes) 1369f9c78b2bSJens Axboe break; 1370f9c78b2bSJens Axboe 1371f9c78b2bSJens Axboe len -= bytes; 1372f9c78b2bSJens Axboe offset = 0; 1373f9c78b2bSJens Axboe } 1374f9c78b2bSJens Axboe 1375f9c78b2bSJens Axboe cur_page = j; 1376f9c78b2bSJens Axboe /* 1377f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1378f9c78b2bSJens Axboe */ 1379f9c78b2bSJens Axboe while (j < page_limit) 1380f9c78b2bSJens Axboe page_cache_release(pages[j++]); 1381f9c78b2bSJens Axboe } 1382f9c78b2bSJens Axboe 1383f9c78b2bSJens Axboe kfree(pages); 1384f9c78b2bSJens Axboe 1385f9c78b2bSJens Axboe /* 1386f9c78b2bSJens Axboe * set data direction, and check if mapped pages need bouncing 1387f9c78b2bSJens Axboe */ 138826e49cfcSKent Overstreet if (iter->type & WRITE) 1389f9c78b2bSJens Axboe bio->bi_rw |= REQ_WRITE; 1390f9c78b2bSJens Axboe 1391f9c78b2bSJens Axboe bio->bi_flags |= (1 << BIO_USER_MAPPED); 139237f19e57SChristoph Hellwig 139337f19e57SChristoph Hellwig /* 139437f19e57SChristoph Hellwig * subtle -- if __bio_map_user() ended up bouncing a bio, 139537f19e57SChristoph Hellwig * it would normally disappear when its bi_end_io is run. 139637f19e57SChristoph Hellwig * however, we need it for the unmap, so grab an extra 139737f19e57SChristoph Hellwig * reference to it 139837f19e57SChristoph Hellwig */ 139937f19e57SChristoph Hellwig bio_get(bio); 1400f9c78b2bSJens Axboe return bio; 1401f9c78b2bSJens Axboe 1402f9c78b2bSJens Axboe out_unmap: 140326e49cfcSKent Overstreet for (j = 0; j < nr_pages; j++) { 140426e49cfcSKent Overstreet if (!pages[j]) 1405f9c78b2bSJens Axboe break; 140626e49cfcSKent Overstreet page_cache_release(pages[j]); 1407f9c78b2bSJens Axboe } 1408f9c78b2bSJens Axboe out: 1409f9c78b2bSJens Axboe kfree(pages); 1410f9c78b2bSJens Axboe bio_put(bio); 1411f9c78b2bSJens Axboe return ERR_PTR(ret); 1412f9c78b2bSJens Axboe } 1413f9c78b2bSJens Axboe 1414f9c78b2bSJens Axboe static void __bio_unmap_user(struct bio *bio) 1415f9c78b2bSJens Axboe { 1416f9c78b2bSJens Axboe struct bio_vec *bvec; 1417f9c78b2bSJens Axboe int i; 1418f9c78b2bSJens Axboe 1419f9c78b2bSJens Axboe /* 1420f9c78b2bSJens Axboe * make sure we dirty pages we wrote to 1421f9c78b2bSJens Axboe */ 1422f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1423f9c78b2bSJens Axboe if (bio_data_dir(bio) == READ) 1424f9c78b2bSJens Axboe set_page_dirty_lock(bvec->bv_page); 1425f9c78b2bSJens Axboe 1426f9c78b2bSJens Axboe page_cache_release(bvec->bv_page); 1427f9c78b2bSJens Axboe } 1428f9c78b2bSJens Axboe 1429f9c78b2bSJens Axboe bio_put(bio); 1430f9c78b2bSJens Axboe } 1431f9c78b2bSJens Axboe 1432f9c78b2bSJens Axboe /** 1433f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1434f9c78b2bSJens Axboe * @bio: the bio being unmapped 1435f9c78b2bSJens Axboe * 1436f9c78b2bSJens Axboe * Unmap a bio previously mapped by bio_map_user(). Must be called with 1437f9c78b2bSJens Axboe * a process context. 1438f9c78b2bSJens Axboe * 1439f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1440f9c78b2bSJens Axboe */ 1441f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1442f9c78b2bSJens Axboe { 1443f9c78b2bSJens Axboe __bio_unmap_user(bio); 1444f9c78b2bSJens Axboe bio_put(bio); 1445f9c78b2bSJens Axboe } 1446f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_unmap_user); 1447f9c78b2bSJens Axboe 1448f9c78b2bSJens Axboe static void bio_map_kern_endio(struct bio *bio, int err) 1449f9c78b2bSJens Axboe { 1450f9c78b2bSJens Axboe bio_put(bio); 1451f9c78b2bSJens Axboe } 1452f9c78b2bSJens Axboe 145375c72b83SChristoph Hellwig /** 145475c72b83SChristoph Hellwig * bio_map_kern - map kernel address into bio 145575c72b83SChristoph Hellwig * @q: the struct request_queue for the bio 145675c72b83SChristoph Hellwig * @data: pointer to buffer to map 145775c72b83SChristoph Hellwig * @len: length in bytes 145875c72b83SChristoph Hellwig * @gfp_mask: allocation flags for bio allocation 145975c72b83SChristoph Hellwig * 146075c72b83SChristoph Hellwig * Map the kernel address into a bio suitable for io to a block 146175c72b83SChristoph Hellwig * device. Returns an error pointer in case of error. 146275c72b83SChristoph Hellwig */ 146375c72b83SChristoph Hellwig struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 146475c72b83SChristoph Hellwig gfp_t gfp_mask) 1465f9c78b2bSJens Axboe { 1466f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1467f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1468f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1469f9c78b2bSJens Axboe const int nr_pages = end - start; 1470f9c78b2bSJens Axboe int offset, i; 1471f9c78b2bSJens Axboe struct bio *bio; 1472f9c78b2bSJens Axboe 1473f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1474f9c78b2bSJens Axboe if (!bio) 1475f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1476f9c78b2bSJens Axboe 1477f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1478f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1479f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1480f9c78b2bSJens Axboe 1481f9c78b2bSJens Axboe if (len <= 0) 1482f9c78b2bSJens Axboe break; 1483f9c78b2bSJens Axboe 1484f9c78b2bSJens Axboe if (bytes > len) 1485f9c78b2bSJens Axboe bytes = len; 1486f9c78b2bSJens Axboe 1487f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 148875c72b83SChristoph Hellwig offset) < bytes) { 148975c72b83SChristoph Hellwig /* we don't support partial mappings */ 149075c72b83SChristoph Hellwig bio_put(bio); 149175c72b83SChristoph Hellwig return ERR_PTR(-EINVAL); 149275c72b83SChristoph Hellwig } 1493f9c78b2bSJens Axboe 1494f9c78b2bSJens Axboe data += bytes; 1495f9c78b2bSJens Axboe len -= bytes; 1496f9c78b2bSJens Axboe offset = 0; 1497f9c78b2bSJens Axboe } 1498f9c78b2bSJens Axboe 1499f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1500f9c78b2bSJens Axboe return bio; 1501f9c78b2bSJens Axboe } 1502f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_map_kern); 1503f9c78b2bSJens Axboe 1504f9c78b2bSJens Axboe static void bio_copy_kern_endio(struct bio *bio, int err) 1505f9c78b2bSJens Axboe { 15061dfa0f68SChristoph Hellwig bio_free_pages(bio); 15071dfa0f68SChristoph Hellwig bio_put(bio); 15081dfa0f68SChristoph Hellwig } 15091dfa0f68SChristoph Hellwig 15101dfa0f68SChristoph Hellwig static void bio_copy_kern_endio_read(struct bio *bio, int err) 15111dfa0f68SChristoph Hellwig { 151242d2683aSChristoph Hellwig char *p = bio->bi_private; 15131dfa0f68SChristoph Hellwig struct bio_vec *bvec; 1514f9c78b2bSJens Axboe int i; 1515f9c78b2bSJens Axboe 1516f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 15171dfa0f68SChristoph Hellwig memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1518f9c78b2bSJens Axboe p += bvec->bv_len; 1519f9c78b2bSJens Axboe } 1520f9c78b2bSJens Axboe 15211dfa0f68SChristoph Hellwig bio_copy_kern_endio(bio, err); 1522f9c78b2bSJens Axboe } 1523f9c78b2bSJens Axboe 1524f9c78b2bSJens Axboe /** 1525f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1526f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1527f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1528f9c78b2bSJens Axboe * @len: length in bytes 1529f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1530f9c78b2bSJens Axboe * @reading: data direction is READ 1531f9c78b2bSJens Axboe * 1532f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1533f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1534f9c78b2bSJens Axboe */ 1535f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1536f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1537f9c78b2bSJens Axboe { 153842d2683aSChristoph Hellwig unsigned long kaddr = (unsigned long)data; 153942d2683aSChristoph Hellwig unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 154042d2683aSChristoph Hellwig unsigned long start = kaddr >> PAGE_SHIFT; 154142d2683aSChristoph Hellwig struct bio *bio; 1542f9c78b2bSJens Axboe void *p = data; 15431dfa0f68SChristoph Hellwig int nr_pages = 0; 1544f9c78b2bSJens Axboe 154542d2683aSChristoph Hellwig /* 154642d2683aSChristoph Hellwig * Overflow, abort 154742d2683aSChristoph Hellwig */ 154842d2683aSChristoph Hellwig if (end < start) 154942d2683aSChristoph Hellwig return ERR_PTR(-EINVAL); 1550f9c78b2bSJens Axboe 155142d2683aSChristoph Hellwig nr_pages = end - start; 155242d2683aSChristoph Hellwig bio = bio_kmalloc(gfp_mask, nr_pages); 155342d2683aSChristoph Hellwig if (!bio) 155442d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 155542d2683aSChristoph Hellwig 155642d2683aSChristoph Hellwig while (len) { 155742d2683aSChristoph Hellwig struct page *page; 155842d2683aSChristoph Hellwig unsigned int bytes = PAGE_SIZE; 155942d2683aSChristoph Hellwig 156042d2683aSChristoph Hellwig if (bytes > len) 156142d2683aSChristoph Hellwig bytes = len; 156242d2683aSChristoph Hellwig 156342d2683aSChristoph Hellwig page = alloc_page(q->bounce_gfp | gfp_mask); 156442d2683aSChristoph Hellwig if (!page) 156542d2683aSChristoph Hellwig goto cleanup; 156642d2683aSChristoph Hellwig 156742d2683aSChristoph Hellwig if (!reading) 156842d2683aSChristoph Hellwig memcpy(page_address(page), p, bytes); 156942d2683aSChristoph Hellwig 157042d2683aSChristoph Hellwig if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 157142d2683aSChristoph Hellwig break; 157242d2683aSChristoph Hellwig 157342d2683aSChristoph Hellwig len -= bytes; 157442d2683aSChristoph Hellwig p += bytes; 1575f9c78b2bSJens Axboe } 1576f9c78b2bSJens Axboe 15771dfa0f68SChristoph Hellwig if (reading) { 15781dfa0f68SChristoph Hellwig bio->bi_end_io = bio_copy_kern_endio_read; 157942d2683aSChristoph Hellwig bio->bi_private = data; 15801dfa0f68SChristoph Hellwig } else { 1581f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 15821dfa0f68SChristoph Hellwig bio->bi_rw |= REQ_WRITE; 15831dfa0f68SChristoph Hellwig } 15841dfa0f68SChristoph Hellwig 1585f9c78b2bSJens Axboe return bio; 158642d2683aSChristoph Hellwig 158742d2683aSChristoph Hellwig cleanup: 15881dfa0f68SChristoph Hellwig bio_free_pages(bio); 158942d2683aSChristoph Hellwig bio_put(bio); 159042d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 1591f9c78b2bSJens Axboe } 1592f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_copy_kern); 1593f9c78b2bSJens Axboe 1594f9c78b2bSJens Axboe /* 1595f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1596f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1597f9c78b2bSJens Axboe * 1598f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1599f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1600f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1601f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1602f9c78b2bSJens Axboe * in process context. 1603f9c78b2bSJens Axboe * 1604f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1605f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1606f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1607f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1608f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1609f9c78b2bSJens Axboe * 1610f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1611f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1612f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1613f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1614f9c78b2bSJens Axboe * pagecache. 1615f9c78b2bSJens Axboe * 1616f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1617f9c78b2bSJens Axboe * deferred bio dirtying paths. 1618f9c78b2bSJens Axboe */ 1619f9c78b2bSJens Axboe 1620f9c78b2bSJens Axboe /* 1621f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1622f9c78b2bSJens Axboe */ 1623f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1624f9c78b2bSJens Axboe { 1625f9c78b2bSJens Axboe struct bio_vec *bvec; 1626f9c78b2bSJens Axboe int i; 1627f9c78b2bSJens Axboe 1628f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1629f9c78b2bSJens Axboe struct page *page = bvec->bv_page; 1630f9c78b2bSJens Axboe 1631f9c78b2bSJens Axboe if (page && !PageCompound(page)) 1632f9c78b2bSJens Axboe set_page_dirty_lock(page); 1633f9c78b2bSJens Axboe } 1634f9c78b2bSJens Axboe } 1635f9c78b2bSJens Axboe 1636f9c78b2bSJens Axboe static void bio_release_pages(struct bio *bio) 1637f9c78b2bSJens Axboe { 1638f9c78b2bSJens Axboe struct bio_vec *bvec; 1639f9c78b2bSJens Axboe int i; 1640f9c78b2bSJens Axboe 1641f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1642f9c78b2bSJens Axboe struct page *page = bvec->bv_page; 1643f9c78b2bSJens Axboe 1644f9c78b2bSJens Axboe if (page) 1645f9c78b2bSJens Axboe put_page(page); 1646f9c78b2bSJens Axboe } 1647f9c78b2bSJens Axboe } 1648f9c78b2bSJens Axboe 1649f9c78b2bSJens Axboe /* 1650f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1651f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1652f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 1653f9c78b2bSJens Axboe * the BIO and the offending pages and re-dirty the pages in process context. 1654f9c78b2bSJens Axboe * 1655f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1656f9c78b2bSJens Axboe * here on. It will run one page_cache_release() against each page and will 1657f9c78b2bSJens Axboe * run one bio_put() against the BIO. 1658f9c78b2bSJens Axboe */ 1659f9c78b2bSJens Axboe 1660f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1661f9c78b2bSJens Axboe 1662f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1663f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1664f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1665f9c78b2bSJens Axboe 1666f9c78b2bSJens Axboe /* 1667f9c78b2bSJens Axboe * This runs in process context 1668f9c78b2bSJens Axboe */ 1669f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1670f9c78b2bSJens Axboe { 1671f9c78b2bSJens Axboe unsigned long flags; 1672f9c78b2bSJens Axboe struct bio *bio; 1673f9c78b2bSJens Axboe 1674f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1675f9c78b2bSJens Axboe bio = bio_dirty_list; 1676f9c78b2bSJens Axboe bio_dirty_list = NULL; 1677f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1678f9c78b2bSJens Axboe 1679f9c78b2bSJens Axboe while (bio) { 1680f9c78b2bSJens Axboe struct bio *next = bio->bi_private; 1681f9c78b2bSJens Axboe 1682f9c78b2bSJens Axboe bio_set_pages_dirty(bio); 1683f9c78b2bSJens Axboe bio_release_pages(bio); 1684f9c78b2bSJens Axboe bio_put(bio); 1685f9c78b2bSJens Axboe bio = next; 1686f9c78b2bSJens Axboe } 1687f9c78b2bSJens Axboe } 1688f9c78b2bSJens Axboe 1689f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1690f9c78b2bSJens Axboe { 1691f9c78b2bSJens Axboe struct bio_vec *bvec; 1692f9c78b2bSJens Axboe int nr_clean_pages = 0; 1693f9c78b2bSJens Axboe int i; 1694f9c78b2bSJens Axboe 1695f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1696f9c78b2bSJens Axboe struct page *page = bvec->bv_page; 1697f9c78b2bSJens Axboe 1698f9c78b2bSJens Axboe if (PageDirty(page) || PageCompound(page)) { 1699f9c78b2bSJens Axboe page_cache_release(page); 1700f9c78b2bSJens Axboe bvec->bv_page = NULL; 1701f9c78b2bSJens Axboe } else { 1702f9c78b2bSJens Axboe nr_clean_pages++; 1703f9c78b2bSJens Axboe } 1704f9c78b2bSJens Axboe } 1705f9c78b2bSJens Axboe 1706f9c78b2bSJens Axboe if (nr_clean_pages) { 1707f9c78b2bSJens Axboe unsigned long flags; 1708f9c78b2bSJens Axboe 1709f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1710f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1711f9c78b2bSJens Axboe bio_dirty_list = bio; 1712f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1713f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1714f9c78b2bSJens Axboe } else { 1715f9c78b2bSJens Axboe bio_put(bio); 1716f9c78b2bSJens Axboe } 1717f9c78b2bSJens Axboe } 1718f9c78b2bSJens Axboe 1719394ffa50SGu Zheng void generic_start_io_acct(int rw, unsigned long sectors, 1720394ffa50SGu Zheng struct hd_struct *part) 1721394ffa50SGu Zheng { 1722394ffa50SGu Zheng int cpu = part_stat_lock(); 1723394ffa50SGu Zheng 1724394ffa50SGu Zheng part_round_stats(cpu, part); 1725394ffa50SGu Zheng part_stat_inc(cpu, part, ios[rw]); 1726394ffa50SGu Zheng part_stat_add(cpu, part, sectors[rw], sectors); 1727394ffa50SGu Zheng part_inc_in_flight(part, rw); 1728394ffa50SGu Zheng 1729394ffa50SGu Zheng part_stat_unlock(); 1730394ffa50SGu Zheng } 1731394ffa50SGu Zheng EXPORT_SYMBOL(generic_start_io_acct); 1732394ffa50SGu Zheng 1733394ffa50SGu Zheng void generic_end_io_acct(int rw, struct hd_struct *part, 1734394ffa50SGu Zheng unsigned long start_time) 1735394ffa50SGu Zheng { 1736394ffa50SGu Zheng unsigned long duration = jiffies - start_time; 1737394ffa50SGu Zheng int cpu = part_stat_lock(); 1738394ffa50SGu Zheng 1739394ffa50SGu Zheng part_stat_add(cpu, part, ticks[rw], duration); 1740394ffa50SGu Zheng part_round_stats(cpu, part); 1741394ffa50SGu Zheng part_dec_in_flight(part, rw); 1742394ffa50SGu Zheng 1743394ffa50SGu Zheng part_stat_unlock(); 1744394ffa50SGu Zheng } 1745394ffa50SGu Zheng EXPORT_SYMBOL(generic_end_io_acct); 1746394ffa50SGu Zheng 1747f9c78b2bSJens Axboe #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1748f9c78b2bSJens Axboe void bio_flush_dcache_pages(struct bio *bi) 1749f9c78b2bSJens Axboe { 1750f9c78b2bSJens Axboe struct bio_vec bvec; 1751f9c78b2bSJens Axboe struct bvec_iter iter; 1752f9c78b2bSJens Axboe 1753f9c78b2bSJens Axboe bio_for_each_segment(bvec, bi, iter) 1754f9c78b2bSJens Axboe flush_dcache_page(bvec.bv_page); 1755f9c78b2bSJens Axboe } 1756f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_flush_dcache_pages); 1757f9c78b2bSJens Axboe #endif 1758f9c78b2bSJens Axboe 1759c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1760c4cf5261SJens Axboe { 1761c4cf5261SJens Axboe /* 1762c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1763c4cf5261SJens Axboe * we always end io on the first invocation. 1764c4cf5261SJens Axboe */ 1765c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1766c4cf5261SJens Axboe return true; 1767c4cf5261SJens Axboe 1768c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1769c4cf5261SJens Axboe 1770326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1771326e1dbbSMike Snitzer clear_bit(BIO_CHAIN, &bio->bi_flags); 1772c4cf5261SJens Axboe return true; 1773326e1dbbSMike Snitzer } 1774c4cf5261SJens Axboe 1775c4cf5261SJens Axboe return false; 1776c4cf5261SJens Axboe } 1777c4cf5261SJens Axboe 1778f9c78b2bSJens Axboe /** 1779f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1780f9c78b2bSJens Axboe * @bio: bio 1781f9c78b2bSJens Axboe * @error: error, if any 1782f9c78b2bSJens Axboe * 1783f9c78b2bSJens Axboe * Description: 1784f9c78b2bSJens Axboe * bio_endio() will end I/O on the whole bio. bio_endio() is the 1785f9c78b2bSJens Axboe * preferred way to end I/O on a bio, it takes care of clearing 1786f9c78b2bSJens Axboe * BIO_UPTODATE on error. @error is 0 on success, and and one of the 1787f9c78b2bSJens Axboe * established -Exxxx (-EIO, for instance) error values in case 1788f9c78b2bSJens Axboe * something went wrong. No one should call bi_end_io() directly on a 1789f9c78b2bSJens Axboe * bio unless they own it and thus know that it has an end_io 1790f9c78b2bSJens Axboe * function. 1791f9c78b2bSJens Axboe **/ 1792f9c78b2bSJens Axboe void bio_endio(struct bio *bio, int error) 1793f9c78b2bSJens Axboe { 1794f9c78b2bSJens Axboe while (bio) { 1795f9c78b2bSJens Axboe if (error) 1796f9c78b2bSJens Axboe clear_bit(BIO_UPTODATE, &bio->bi_flags); 1797f9c78b2bSJens Axboe else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) 1798f9c78b2bSJens Axboe error = -EIO; 1799f9c78b2bSJens Axboe 1800c4cf5261SJens Axboe if (unlikely(!bio_remaining_done(bio))) 1801c4cf5261SJens Axboe break; 1802f9c78b2bSJens Axboe 1803f9c78b2bSJens Axboe /* 1804f9c78b2bSJens Axboe * Need to have a real endio function for chained bios, 1805f9c78b2bSJens Axboe * otherwise various corner cases will break (like stacking 1806f9c78b2bSJens Axboe * block devices that save/restore bi_end_io) - however, we want 1807f9c78b2bSJens Axboe * to avoid unbounded recursion and blowing the stack. Tail call 1808f9c78b2bSJens Axboe * optimization would handle this, but compiling with frame 1809f9c78b2bSJens Axboe * pointers also disables gcc's sibling call optimization. 1810f9c78b2bSJens Axboe */ 1811f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 1812f9c78b2bSJens Axboe struct bio *parent = bio->bi_private; 1813f9c78b2bSJens Axboe bio_put(bio); 1814f9c78b2bSJens Axboe bio = parent; 1815f9c78b2bSJens Axboe } else { 1816f9c78b2bSJens Axboe if (bio->bi_end_io) 1817f9c78b2bSJens Axboe bio->bi_end_io(bio, error); 1818f9c78b2bSJens Axboe bio = NULL; 1819f9c78b2bSJens Axboe } 1820f9c78b2bSJens Axboe } 1821f9c78b2bSJens Axboe } 1822f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1823f9c78b2bSJens Axboe 1824f9c78b2bSJens Axboe /** 1825f9c78b2bSJens Axboe * bio_split - split a bio 1826f9c78b2bSJens Axboe * @bio: bio to split 1827f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1828f9c78b2bSJens Axboe * @gfp: gfp mask 1829f9c78b2bSJens Axboe * @bs: bio set to allocate from 1830f9c78b2bSJens Axboe * 1831f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1832f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1833f9c78b2bSJens Axboe * 1834f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1835f3f5da62SMartin K. Petersen * to @bio's bi_io_vec; it is the caller's responsibility to ensure that 1836f3f5da62SMartin K. Petersen * @bio is not freed before the split. 1837f9c78b2bSJens Axboe */ 1838f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1839f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1840f9c78b2bSJens Axboe { 1841f9c78b2bSJens Axboe struct bio *split = NULL; 1842f9c78b2bSJens Axboe 1843f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1844f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1845f9c78b2bSJens Axboe 1846f3f5da62SMartin K. Petersen /* 1847f3f5da62SMartin K. Petersen * Discards need a mutable bio_vec to accommodate the payload 1848f3f5da62SMartin K. Petersen * required by the DSM TRIM and UNMAP commands. 1849f3f5da62SMartin K. Petersen */ 1850f3f5da62SMartin K. Petersen if (bio->bi_rw & REQ_DISCARD) 1851f3f5da62SMartin K. Petersen split = bio_clone_bioset(bio, gfp, bs); 1852f3f5da62SMartin K. Petersen else 1853f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1854f3f5da62SMartin K. Petersen 1855f9c78b2bSJens Axboe if (!split) 1856f9c78b2bSJens Axboe return NULL; 1857f9c78b2bSJens Axboe 1858f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1859f9c78b2bSJens Axboe 1860f9c78b2bSJens Axboe if (bio_integrity(split)) 1861f9c78b2bSJens Axboe bio_integrity_trim(split, 0, sectors); 1862f9c78b2bSJens Axboe 1863f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1864f9c78b2bSJens Axboe 1865f9c78b2bSJens Axboe return split; 1866f9c78b2bSJens Axboe } 1867f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1868f9c78b2bSJens Axboe 1869f9c78b2bSJens Axboe /** 1870f9c78b2bSJens Axboe * bio_trim - trim a bio 1871f9c78b2bSJens Axboe * @bio: bio to trim 1872f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1873f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1874f9c78b2bSJens Axboe */ 1875f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1876f9c78b2bSJens Axboe { 1877f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1878f9c78b2bSJens Axboe * the given offset and size. 1879f9c78b2bSJens Axboe */ 1880f9c78b2bSJens Axboe 1881f9c78b2bSJens Axboe size <<= 9; 1882f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1883f9c78b2bSJens Axboe return; 1884f9c78b2bSJens Axboe 1885f9c78b2bSJens Axboe clear_bit(BIO_SEG_VALID, &bio->bi_flags); 1886f9c78b2bSJens Axboe 1887f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1888f9c78b2bSJens Axboe 1889f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1890f9c78b2bSJens Axboe } 1891f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1892f9c78b2bSJens Axboe 1893f9c78b2bSJens Axboe /* 1894f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1895f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1896f9c78b2bSJens Axboe */ 1897f9c78b2bSJens Axboe mempool_t *biovec_create_pool(int pool_entries) 1898f9c78b2bSJens Axboe { 1899f9c78b2bSJens Axboe struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; 1900f9c78b2bSJens Axboe 1901f9c78b2bSJens Axboe return mempool_create_slab_pool(pool_entries, bp->slab); 1902f9c78b2bSJens Axboe } 1903f9c78b2bSJens Axboe 1904f9c78b2bSJens Axboe void bioset_free(struct bio_set *bs) 1905f9c78b2bSJens Axboe { 1906f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1907f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1908f9c78b2bSJens Axboe 1909f9c78b2bSJens Axboe if (bs->bio_pool) 1910f9c78b2bSJens Axboe mempool_destroy(bs->bio_pool); 1911f9c78b2bSJens Axboe 1912f9c78b2bSJens Axboe if (bs->bvec_pool) 1913f9c78b2bSJens Axboe mempool_destroy(bs->bvec_pool); 1914f9c78b2bSJens Axboe 1915f9c78b2bSJens Axboe bioset_integrity_free(bs); 1916f9c78b2bSJens Axboe bio_put_slab(bs); 1917f9c78b2bSJens Axboe 1918f9c78b2bSJens Axboe kfree(bs); 1919f9c78b2bSJens Axboe } 1920f9c78b2bSJens Axboe EXPORT_SYMBOL(bioset_free); 1921f9c78b2bSJens Axboe 1922d8f429e1SJunichi Nomura static struct bio_set *__bioset_create(unsigned int pool_size, 1923d8f429e1SJunichi Nomura unsigned int front_pad, 1924d8f429e1SJunichi Nomura bool create_bvec_pool) 1925f9c78b2bSJens Axboe { 1926f9c78b2bSJens Axboe unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1927f9c78b2bSJens Axboe struct bio_set *bs; 1928f9c78b2bSJens Axboe 1929f9c78b2bSJens Axboe bs = kzalloc(sizeof(*bs), GFP_KERNEL); 1930f9c78b2bSJens Axboe if (!bs) 1931f9c78b2bSJens Axboe return NULL; 1932f9c78b2bSJens Axboe 1933f9c78b2bSJens Axboe bs->front_pad = front_pad; 1934f9c78b2bSJens Axboe 1935f9c78b2bSJens Axboe spin_lock_init(&bs->rescue_lock); 1936f9c78b2bSJens Axboe bio_list_init(&bs->rescue_list); 1937f9c78b2bSJens Axboe INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1938f9c78b2bSJens Axboe 1939f9c78b2bSJens Axboe bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1940f9c78b2bSJens Axboe if (!bs->bio_slab) { 1941f9c78b2bSJens Axboe kfree(bs); 1942f9c78b2bSJens Axboe return NULL; 1943f9c78b2bSJens Axboe } 1944f9c78b2bSJens Axboe 1945f9c78b2bSJens Axboe bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); 1946f9c78b2bSJens Axboe if (!bs->bio_pool) 1947f9c78b2bSJens Axboe goto bad; 1948f9c78b2bSJens Axboe 1949d8f429e1SJunichi Nomura if (create_bvec_pool) { 1950f9c78b2bSJens Axboe bs->bvec_pool = biovec_create_pool(pool_size); 1951f9c78b2bSJens Axboe if (!bs->bvec_pool) 1952f9c78b2bSJens Axboe goto bad; 1953d8f429e1SJunichi Nomura } 1954f9c78b2bSJens Axboe 1955f9c78b2bSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1956f9c78b2bSJens Axboe if (!bs->rescue_workqueue) 1957f9c78b2bSJens Axboe goto bad; 1958f9c78b2bSJens Axboe 1959f9c78b2bSJens Axboe return bs; 1960f9c78b2bSJens Axboe bad: 1961f9c78b2bSJens Axboe bioset_free(bs); 1962f9c78b2bSJens Axboe return NULL; 1963f9c78b2bSJens Axboe } 1964d8f429e1SJunichi Nomura 1965d8f429e1SJunichi Nomura /** 1966d8f429e1SJunichi Nomura * bioset_create - Create a bio_set 1967d8f429e1SJunichi Nomura * @pool_size: Number of bio and bio_vecs to cache in the mempool 1968d8f429e1SJunichi Nomura * @front_pad: Number of bytes to allocate in front of the returned bio 1969d8f429e1SJunichi Nomura * 1970d8f429e1SJunichi Nomura * Description: 1971d8f429e1SJunichi Nomura * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1972d8f429e1SJunichi Nomura * to ask for a number of bytes to be allocated in front of the bio. 1973d8f429e1SJunichi Nomura * Front pad allocation is useful for embedding the bio inside 1974d8f429e1SJunichi Nomura * another structure, to avoid allocating extra data to go with the bio. 1975d8f429e1SJunichi Nomura * Note that the bio must be embedded at the END of that structure always, 1976d8f429e1SJunichi Nomura * or things will break badly. 1977d8f429e1SJunichi Nomura */ 1978d8f429e1SJunichi Nomura struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) 1979d8f429e1SJunichi Nomura { 1980d8f429e1SJunichi Nomura return __bioset_create(pool_size, front_pad, true); 1981d8f429e1SJunichi Nomura } 1982f9c78b2bSJens Axboe EXPORT_SYMBOL(bioset_create); 1983f9c78b2bSJens Axboe 1984d8f429e1SJunichi Nomura /** 1985d8f429e1SJunichi Nomura * bioset_create_nobvec - Create a bio_set without bio_vec mempool 1986d8f429e1SJunichi Nomura * @pool_size: Number of bio to cache in the mempool 1987d8f429e1SJunichi Nomura * @front_pad: Number of bytes to allocate in front of the returned bio 1988d8f429e1SJunichi Nomura * 1989d8f429e1SJunichi Nomura * Description: 1990d8f429e1SJunichi Nomura * Same functionality as bioset_create() except that mempool is not 1991d8f429e1SJunichi Nomura * created for bio_vecs. Saving some memory for bio_clone_fast() users. 1992d8f429e1SJunichi Nomura */ 1993d8f429e1SJunichi Nomura struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad) 1994d8f429e1SJunichi Nomura { 1995d8f429e1SJunichi Nomura return __bioset_create(pool_size, front_pad, false); 1996d8f429e1SJunichi Nomura } 1997d8f429e1SJunichi Nomura EXPORT_SYMBOL(bioset_create_nobvec); 1998d8f429e1SJunichi Nomura 1999f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 20001d933cf0STejun Heo 20011d933cf0STejun Heo /** 20021d933cf0STejun Heo * bio_associate_blkcg - associate a bio with the specified blkcg 20031d933cf0STejun Heo * @bio: target bio 20041d933cf0STejun Heo * @blkcg_css: css of the blkcg to associate 20051d933cf0STejun Heo * 20061d933cf0STejun Heo * Associate @bio with the blkcg specified by @blkcg_css. Block layer will 20071d933cf0STejun Heo * treat @bio as if it were issued by a task which belongs to the blkcg. 20081d933cf0STejun Heo * 20091d933cf0STejun Heo * This function takes an extra reference of @blkcg_css which will be put 20101d933cf0STejun Heo * when @bio is released. The caller must own @bio and is responsible for 20111d933cf0STejun Heo * synchronizing calls to this function. 20121d933cf0STejun Heo */ 20131d933cf0STejun Heo int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css) 20141d933cf0STejun Heo { 20151d933cf0STejun Heo if (unlikely(bio->bi_css)) 20161d933cf0STejun Heo return -EBUSY; 20171d933cf0STejun Heo css_get(blkcg_css); 20181d933cf0STejun Heo bio->bi_css = blkcg_css; 20191d933cf0STejun Heo return 0; 20201d933cf0STejun Heo } 20215aa2a96bSTejun Heo EXPORT_SYMBOL_GPL(bio_associate_blkcg); 20221d933cf0STejun Heo 2023f9c78b2bSJens Axboe /** 2024f9c78b2bSJens Axboe * bio_associate_current - associate a bio with %current 2025f9c78b2bSJens Axboe * @bio: target bio 2026f9c78b2bSJens Axboe * 2027f9c78b2bSJens Axboe * Associate @bio with %current if it hasn't been associated yet. Block 2028f9c78b2bSJens Axboe * layer will treat @bio as if it were issued by %current no matter which 2029f9c78b2bSJens Axboe * task actually issues it. 2030f9c78b2bSJens Axboe * 2031f9c78b2bSJens Axboe * This function takes an extra reference of @task's io_context and blkcg 2032f9c78b2bSJens Axboe * which will be put when @bio is released. The caller must own @bio, 2033f9c78b2bSJens Axboe * ensure %current->io_context exists, and is responsible for synchronizing 2034f9c78b2bSJens Axboe * calls to this function. 2035f9c78b2bSJens Axboe */ 2036f9c78b2bSJens Axboe int bio_associate_current(struct bio *bio) 2037f9c78b2bSJens Axboe { 2038f9c78b2bSJens Axboe struct io_context *ioc; 2039f9c78b2bSJens Axboe 20401d933cf0STejun Heo if (bio->bi_css) 2041f9c78b2bSJens Axboe return -EBUSY; 2042f9c78b2bSJens Axboe 2043f9c78b2bSJens Axboe ioc = current->io_context; 2044f9c78b2bSJens Axboe if (!ioc) 2045f9c78b2bSJens Axboe return -ENOENT; 2046f9c78b2bSJens Axboe 2047f9c78b2bSJens Axboe get_io_context_active(ioc); 2048f9c78b2bSJens Axboe bio->bi_ioc = ioc; 2049c165b3e3STejun Heo bio->bi_css = task_get_css(current, io_cgrp_id); 2050f9c78b2bSJens Axboe return 0; 2051f9c78b2bSJens Axboe } 20525aa2a96bSTejun Heo EXPORT_SYMBOL_GPL(bio_associate_current); 2053f9c78b2bSJens Axboe 2054f9c78b2bSJens Axboe /** 2055f9c78b2bSJens Axboe * bio_disassociate_task - undo bio_associate_current() 2056f9c78b2bSJens Axboe * @bio: target bio 2057f9c78b2bSJens Axboe */ 2058f9c78b2bSJens Axboe void bio_disassociate_task(struct bio *bio) 2059f9c78b2bSJens Axboe { 2060f9c78b2bSJens Axboe if (bio->bi_ioc) { 2061f9c78b2bSJens Axboe put_io_context(bio->bi_ioc); 2062f9c78b2bSJens Axboe bio->bi_ioc = NULL; 2063f9c78b2bSJens Axboe } 2064f9c78b2bSJens Axboe if (bio->bi_css) { 2065f9c78b2bSJens Axboe css_put(bio->bi_css); 2066f9c78b2bSJens Axboe bio->bi_css = NULL; 2067f9c78b2bSJens Axboe } 2068f9c78b2bSJens Axboe } 2069f9c78b2bSJens Axboe 2070f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 2071f9c78b2bSJens Axboe 2072f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2073f9c78b2bSJens Axboe { 2074f9c78b2bSJens Axboe int i; 2075f9c78b2bSJens Axboe 2076f9c78b2bSJens Axboe for (i = 0; i < BIOVEC_NR_POOLS; i++) { 2077f9c78b2bSJens Axboe int size; 2078f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2079f9c78b2bSJens Axboe 2080f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2081f9c78b2bSJens Axboe bvs->slab = NULL; 2082f9c78b2bSJens Axboe continue; 2083f9c78b2bSJens Axboe } 2084f9c78b2bSJens Axboe 2085f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2086f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2087f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2088f9c78b2bSJens Axboe } 2089f9c78b2bSJens Axboe } 2090f9c78b2bSJens Axboe 2091f9c78b2bSJens Axboe static int __init init_bio(void) 2092f9c78b2bSJens Axboe { 2093f9c78b2bSJens Axboe bio_slab_max = 2; 2094f9c78b2bSJens Axboe bio_slab_nr = 0; 2095f9c78b2bSJens Axboe bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); 2096f9c78b2bSJens Axboe if (!bio_slabs) 2097f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2098f9c78b2bSJens Axboe 2099f9c78b2bSJens Axboe bio_integrity_init(); 2100f9c78b2bSJens Axboe biovec_init_slabs(); 2101f9c78b2bSJens Axboe 2102f9c78b2bSJens Axboe fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); 2103f9c78b2bSJens Axboe if (!fs_bio_set) 2104f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2105f9c78b2bSJens Axboe 2106f9c78b2bSJens Axboe if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) 2107f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2108f9c78b2bSJens Axboe 2109f9c78b2bSJens Axboe return 0; 2110f9c78b2bSJens Axboe } 2111f9c78b2bSJens Axboe subsys_initcall(init_bio); 2112