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 #include <scsi/sg.h> /* for struct sg_iovec */ 32f9c78b2bSJens Axboe 33f9c78b2bSJens Axboe #include <trace/events/block.h> 34f9c78b2bSJens Axboe 35f9c78b2bSJens Axboe /* 36f9c78b2bSJens Axboe * Test patch to inline a certain number of bi_io_vec's inside the bio 37f9c78b2bSJens Axboe * itself, to shrink a bio data allocation from two mempool calls to one 38f9c78b2bSJens Axboe */ 39f9c78b2bSJens Axboe #define BIO_INLINE_VECS 4 40f9c78b2bSJens Axboe 41f9c78b2bSJens Axboe /* 42f9c78b2bSJens Axboe * if you change this list, also change bvec_alloc or things will 43f9c78b2bSJens Axboe * break badly! cannot be bigger than what you can fit into an 44f9c78b2bSJens Axboe * unsigned short 45f9c78b2bSJens Axboe */ 46f9c78b2bSJens Axboe #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) } 47f9c78b2bSJens Axboe static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = { 48f9c78b2bSJens Axboe BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES), 49f9c78b2bSJens Axboe }; 50f9c78b2bSJens Axboe #undef BV 51f9c78b2bSJens Axboe 52f9c78b2bSJens Axboe /* 53f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 54f9c78b2bSJens Axboe * IO code that does not need private memory pools. 55f9c78b2bSJens Axboe */ 56f9c78b2bSJens Axboe struct bio_set *fs_bio_set; 57f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 58f9c78b2bSJens Axboe 59f9c78b2bSJens Axboe /* 60f9c78b2bSJens Axboe * Our slab pool management 61f9c78b2bSJens Axboe */ 62f9c78b2bSJens Axboe struct bio_slab { 63f9c78b2bSJens Axboe struct kmem_cache *slab; 64f9c78b2bSJens Axboe unsigned int slab_ref; 65f9c78b2bSJens Axboe unsigned int slab_size; 66f9c78b2bSJens Axboe char name[8]; 67f9c78b2bSJens Axboe }; 68f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 69f9c78b2bSJens Axboe static struct bio_slab *bio_slabs; 70f9c78b2bSJens Axboe static unsigned int bio_slab_nr, bio_slab_max; 71f9c78b2bSJens Axboe 72f9c78b2bSJens Axboe static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 73f9c78b2bSJens Axboe { 74f9c78b2bSJens Axboe unsigned int sz = sizeof(struct bio) + extra_size; 75f9c78b2bSJens Axboe struct kmem_cache *slab = NULL; 76f9c78b2bSJens Axboe struct bio_slab *bslab, *new_bio_slabs; 77f9c78b2bSJens Axboe unsigned int new_bio_slab_max; 78f9c78b2bSJens Axboe unsigned int i, entry = -1; 79f9c78b2bSJens Axboe 80f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 81f9c78b2bSJens Axboe 82f9c78b2bSJens Axboe i = 0; 83f9c78b2bSJens Axboe while (i < bio_slab_nr) { 84f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 85f9c78b2bSJens Axboe 86f9c78b2bSJens Axboe if (!bslab->slab && entry == -1) 87f9c78b2bSJens Axboe entry = i; 88f9c78b2bSJens Axboe else if (bslab->slab_size == sz) { 89f9c78b2bSJens Axboe slab = bslab->slab; 90f9c78b2bSJens Axboe bslab->slab_ref++; 91f9c78b2bSJens Axboe break; 92f9c78b2bSJens Axboe } 93f9c78b2bSJens Axboe i++; 94f9c78b2bSJens Axboe } 95f9c78b2bSJens Axboe 96f9c78b2bSJens Axboe if (slab) 97f9c78b2bSJens Axboe goto out_unlock; 98f9c78b2bSJens Axboe 99f9c78b2bSJens Axboe if (bio_slab_nr == bio_slab_max && entry == -1) { 100f9c78b2bSJens Axboe new_bio_slab_max = bio_slab_max << 1; 101f9c78b2bSJens Axboe new_bio_slabs = krealloc(bio_slabs, 102f9c78b2bSJens Axboe new_bio_slab_max * sizeof(struct bio_slab), 103f9c78b2bSJens Axboe GFP_KERNEL); 104f9c78b2bSJens Axboe if (!new_bio_slabs) 105f9c78b2bSJens Axboe goto out_unlock; 106f9c78b2bSJens Axboe bio_slab_max = new_bio_slab_max; 107f9c78b2bSJens Axboe bio_slabs = new_bio_slabs; 108f9c78b2bSJens Axboe } 109f9c78b2bSJens Axboe if (entry == -1) 110f9c78b2bSJens Axboe entry = bio_slab_nr++; 111f9c78b2bSJens Axboe 112f9c78b2bSJens Axboe bslab = &bio_slabs[entry]; 113f9c78b2bSJens Axboe 114f9c78b2bSJens Axboe snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 115f9c78b2bSJens Axboe slab = kmem_cache_create(bslab->name, sz, 0, 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; 273f9c78b2bSJens Axboe atomic_set(&bio->bi_remaining, 1); 274f9c78b2bSJens 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); 296f9c78b2bSJens 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 306f9c78b2bSJens Axboe /** 307f9c78b2bSJens Axboe * bio_chain - chain bio completions 308f9c78b2bSJens Axboe * @bio: the target bio 309f9c78b2bSJens Axboe * @parent: the @bio's parent bio 310f9c78b2bSJens Axboe * 311f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 312f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 313f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 314f9c78b2bSJens Axboe * 315f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 316f9c78b2bSJens Axboe */ 317f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 318f9c78b2bSJens Axboe { 319f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 320f9c78b2bSJens Axboe 321f9c78b2bSJens Axboe bio->bi_private = parent; 322f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 323f9c78b2bSJens Axboe atomic_inc(&parent->bi_remaining); 324f9c78b2bSJens Axboe } 325f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 326f9c78b2bSJens Axboe 327f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 328f9c78b2bSJens Axboe { 329f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 330f9c78b2bSJens Axboe struct bio *bio; 331f9c78b2bSJens Axboe 332f9c78b2bSJens Axboe while (1) { 333f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 334f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 335f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 336f9c78b2bSJens Axboe 337f9c78b2bSJens Axboe if (!bio) 338f9c78b2bSJens Axboe break; 339f9c78b2bSJens Axboe 340f9c78b2bSJens Axboe generic_make_request(bio); 341f9c78b2bSJens Axboe } 342f9c78b2bSJens Axboe } 343f9c78b2bSJens Axboe 344f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 345f9c78b2bSJens Axboe { 346f9c78b2bSJens Axboe struct bio_list punt, nopunt; 347f9c78b2bSJens Axboe struct bio *bio; 348f9c78b2bSJens Axboe 349f9c78b2bSJens Axboe /* 350f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 351f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 352f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 353f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 354f9c78b2bSJens Axboe * our own rescuer would be bad. 355f9c78b2bSJens Axboe * 356f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 357f9c78b2bSJens Axboe * remove from the middle of the list: 358f9c78b2bSJens Axboe */ 359f9c78b2bSJens Axboe 360f9c78b2bSJens Axboe bio_list_init(&punt); 361f9c78b2bSJens Axboe bio_list_init(&nopunt); 362f9c78b2bSJens Axboe 363f9c78b2bSJens Axboe while ((bio = bio_list_pop(current->bio_list))) 364f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 365f9c78b2bSJens Axboe 366f9c78b2bSJens Axboe *current->bio_list = nopunt; 367f9c78b2bSJens Axboe 368f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 369f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 370f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 371f9c78b2bSJens Axboe 372f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 373f9c78b2bSJens Axboe } 374f9c78b2bSJens Axboe 375f9c78b2bSJens Axboe /** 376f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 377f9c78b2bSJens Axboe * @gfp_mask: the GFP_ mask given to the slab allocator 378f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 379f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 380f9c78b2bSJens Axboe * 381f9c78b2bSJens Axboe * Description: 382f9c78b2bSJens Axboe * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 383f9c78b2bSJens Axboe * backed by the @bs's mempool. 384f9c78b2bSJens Axboe * 385f9c78b2bSJens Axboe * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be 386f9c78b2bSJens Axboe * able to allocate a bio. This is due to the mempool guarantees. To make this 387f9c78b2bSJens Axboe * work, callers must never allocate more than 1 bio at a time from this pool. 388f9c78b2bSJens Axboe * Callers that need to allocate more than 1 bio must always submit the 389f9c78b2bSJens Axboe * previously allocated bio for IO before attempting to allocate a new one. 390f9c78b2bSJens Axboe * Failure to do so can cause deadlocks under memory pressure. 391f9c78b2bSJens Axboe * 392f9c78b2bSJens Axboe * Note that when running under generic_make_request() (i.e. any block 393f9c78b2bSJens Axboe * driver), bios are not submitted until after you return - see the code in 394f9c78b2bSJens Axboe * generic_make_request() that converts recursion into iteration, to prevent 395f9c78b2bSJens Axboe * stack overflows. 396f9c78b2bSJens Axboe * 397f9c78b2bSJens Axboe * This would normally mean allocating multiple bios under 398f9c78b2bSJens Axboe * generic_make_request() would be susceptible to deadlocks, but we have 399f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 400f9c78b2bSJens Axboe * thread. 401f9c78b2bSJens Axboe * 402f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 403f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 404f9c78b2bSJens Axboe * generic_make_request() should be avoided - instead, use bio_set's front_pad 405f9c78b2bSJens Axboe * for per bio allocations. 406f9c78b2bSJens Axboe * 407f9c78b2bSJens Axboe * RETURNS: 408f9c78b2bSJens Axboe * Pointer to new bio on success, NULL on failure. 409f9c78b2bSJens Axboe */ 410f9c78b2bSJens Axboe struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs) 411f9c78b2bSJens Axboe { 412f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 413f9c78b2bSJens Axboe unsigned front_pad; 414f9c78b2bSJens Axboe unsigned inline_vecs; 415f9c78b2bSJens Axboe unsigned long idx = BIO_POOL_NONE; 416f9c78b2bSJens Axboe struct bio_vec *bvl = NULL; 417f9c78b2bSJens Axboe struct bio *bio; 418f9c78b2bSJens Axboe void *p; 419f9c78b2bSJens Axboe 420f9c78b2bSJens Axboe if (!bs) { 421f9c78b2bSJens Axboe if (nr_iovecs > UIO_MAXIOV) 422f9c78b2bSJens Axboe return NULL; 423f9c78b2bSJens Axboe 424f9c78b2bSJens Axboe p = kmalloc(sizeof(struct bio) + 425f9c78b2bSJens Axboe nr_iovecs * sizeof(struct bio_vec), 426f9c78b2bSJens Axboe gfp_mask); 427f9c78b2bSJens Axboe front_pad = 0; 428f9c78b2bSJens Axboe inline_vecs = nr_iovecs; 429f9c78b2bSJens Axboe } else { 430f9c78b2bSJens Axboe /* 431f9c78b2bSJens Axboe * generic_make_request() converts recursion to iteration; this 432f9c78b2bSJens Axboe * means if we're running beneath it, any bios we allocate and 433f9c78b2bSJens Axboe * submit will not be submitted (and thus freed) until after we 434f9c78b2bSJens Axboe * return. 435f9c78b2bSJens Axboe * 436f9c78b2bSJens Axboe * This exposes us to a potential deadlock if we allocate 437f9c78b2bSJens Axboe * multiple bios from the same bio_set() while running 438f9c78b2bSJens Axboe * underneath generic_make_request(). If we were to allocate 439f9c78b2bSJens Axboe * multiple bios (say a stacking block driver that was splitting 440f9c78b2bSJens Axboe * bios), we would deadlock if we exhausted the mempool's 441f9c78b2bSJens Axboe * reserve. 442f9c78b2bSJens Axboe * 443f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 444f9c78b2bSJens Axboe * workqueue per bio_set. If we go to allocate and there are 445f9c78b2bSJens Axboe * bios on current->bio_list, we first try the allocation 446f9c78b2bSJens Axboe * without __GFP_WAIT; if that fails, we punt those bios we 447f9c78b2bSJens Axboe * would be blocking to the rescuer workqueue before we retry 448f9c78b2bSJens Axboe * with the original gfp_flags. 449f9c78b2bSJens Axboe */ 450f9c78b2bSJens Axboe 451f9c78b2bSJens Axboe if (current->bio_list && !bio_list_empty(current->bio_list)) 452f9c78b2bSJens Axboe gfp_mask &= ~__GFP_WAIT; 453f9c78b2bSJens Axboe 454f9c78b2bSJens Axboe p = mempool_alloc(bs->bio_pool, gfp_mask); 455f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 456f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 457f9c78b2bSJens Axboe gfp_mask = saved_gfp; 458f9c78b2bSJens Axboe p = mempool_alloc(bs->bio_pool, gfp_mask); 459f9c78b2bSJens Axboe } 460f9c78b2bSJens Axboe 461f9c78b2bSJens Axboe front_pad = bs->front_pad; 462f9c78b2bSJens Axboe inline_vecs = BIO_INLINE_VECS; 463f9c78b2bSJens Axboe } 464f9c78b2bSJens Axboe 465f9c78b2bSJens Axboe if (unlikely(!p)) 466f9c78b2bSJens Axboe return NULL; 467f9c78b2bSJens Axboe 468f9c78b2bSJens Axboe bio = p + front_pad; 469f9c78b2bSJens Axboe bio_init(bio); 470f9c78b2bSJens Axboe 471f9c78b2bSJens Axboe if (nr_iovecs > inline_vecs) { 472f9c78b2bSJens Axboe bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); 473f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 474f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 475f9c78b2bSJens Axboe gfp_mask = saved_gfp; 476f9c78b2bSJens Axboe bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool); 477f9c78b2bSJens Axboe } 478f9c78b2bSJens Axboe 479f9c78b2bSJens Axboe if (unlikely(!bvl)) 480f9c78b2bSJens Axboe goto err_free; 481f9c78b2bSJens Axboe 482f9c78b2bSJens Axboe bio->bi_flags |= 1 << BIO_OWNS_VEC; 483f9c78b2bSJens Axboe } else if (nr_iovecs) { 484f9c78b2bSJens Axboe bvl = bio->bi_inline_vecs; 485f9c78b2bSJens Axboe } 486f9c78b2bSJens Axboe 487f9c78b2bSJens Axboe bio->bi_pool = bs; 488f9c78b2bSJens Axboe bio->bi_flags |= idx << BIO_POOL_OFFSET; 489f9c78b2bSJens Axboe bio->bi_max_vecs = nr_iovecs; 490f9c78b2bSJens Axboe bio->bi_io_vec = bvl; 491f9c78b2bSJens Axboe return bio; 492f9c78b2bSJens Axboe 493f9c78b2bSJens Axboe err_free: 494f9c78b2bSJens Axboe mempool_free(p, bs->bio_pool); 495f9c78b2bSJens Axboe return NULL; 496f9c78b2bSJens Axboe } 497f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 498f9c78b2bSJens Axboe 499f9c78b2bSJens Axboe void zero_fill_bio(struct bio *bio) 500f9c78b2bSJens Axboe { 501f9c78b2bSJens Axboe unsigned long flags; 502f9c78b2bSJens Axboe struct bio_vec bv; 503f9c78b2bSJens Axboe struct bvec_iter iter; 504f9c78b2bSJens Axboe 505f9c78b2bSJens Axboe bio_for_each_segment(bv, bio, iter) { 506f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 507f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 508f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 509f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 510f9c78b2bSJens Axboe } 511f9c78b2bSJens Axboe } 512f9c78b2bSJens Axboe EXPORT_SYMBOL(zero_fill_bio); 513f9c78b2bSJens Axboe 514f9c78b2bSJens Axboe /** 515f9c78b2bSJens Axboe * bio_put - release a reference to a bio 516f9c78b2bSJens Axboe * @bio: bio to release reference to 517f9c78b2bSJens Axboe * 518f9c78b2bSJens Axboe * Description: 519f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 520f9c78b2bSJens Axboe * bio_alloc, bio_get or bio_clone. The last put of a bio will free it. 521f9c78b2bSJens Axboe **/ 522f9c78b2bSJens Axboe void bio_put(struct bio *bio) 523f9c78b2bSJens Axboe { 524f9c78b2bSJens Axboe BIO_BUG_ON(!atomic_read(&bio->bi_cnt)); 525f9c78b2bSJens Axboe 526f9c78b2bSJens Axboe /* 527f9c78b2bSJens Axboe * last put frees it 528f9c78b2bSJens Axboe */ 529f9c78b2bSJens Axboe if (atomic_dec_and_test(&bio->bi_cnt)) 530f9c78b2bSJens Axboe bio_free(bio); 531f9c78b2bSJens Axboe } 532f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 533f9c78b2bSJens Axboe 534f9c78b2bSJens Axboe inline int bio_phys_segments(struct request_queue *q, struct bio *bio) 535f9c78b2bSJens Axboe { 536f9c78b2bSJens Axboe if (unlikely(!bio_flagged(bio, BIO_SEG_VALID))) 537f9c78b2bSJens Axboe blk_recount_segments(q, bio); 538f9c78b2bSJens Axboe 539f9c78b2bSJens Axboe return bio->bi_phys_segments; 540f9c78b2bSJens Axboe } 541f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_phys_segments); 542f9c78b2bSJens Axboe 543f9c78b2bSJens Axboe /** 544f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 545f9c78b2bSJens Axboe * @bio: destination bio 546f9c78b2bSJens Axboe * @bio_src: bio to clone 547f9c78b2bSJens Axboe * 548f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 549f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 550f9c78b2bSJens Axboe * bio will be one. 551f9c78b2bSJens Axboe * 552f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 553f9c78b2bSJens Axboe */ 554f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 555f9c78b2bSJens Axboe { 556f9c78b2bSJens Axboe BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE); 557f9c78b2bSJens Axboe 558f9c78b2bSJens Axboe /* 559f9c78b2bSJens Axboe * most users will be overriding ->bi_bdev with a new target, 560f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 561f9c78b2bSJens Axboe */ 562f9c78b2bSJens Axboe bio->bi_bdev = bio_src->bi_bdev; 563f9c78b2bSJens Axboe bio->bi_flags |= 1 << BIO_CLONED; 564f9c78b2bSJens Axboe bio->bi_rw = bio_src->bi_rw; 565f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 566f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 567f9c78b2bSJens Axboe } 568f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 569f9c78b2bSJens Axboe 570f9c78b2bSJens Axboe /** 571f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 572f9c78b2bSJens Axboe * @bio: bio to clone 573f9c78b2bSJens Axboe * @gfp_mask: allocation priority 574f9c78b2bSJens Axboe * @bs: bio_set to allocate from 575f9c78b2bSJens Axboe * 576f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 577f9c78b2bSJens Axboe */ 578f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 579f9c78b2bSJens Axboe { 580f9c78b2bSJens Axboe struct bio *b; 581f9c78b2bSJens Axboe 582f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 583f9c78b2bSJens Axboe if (!b) 584f9c78b2bSJens Axboe return NULL; 585f9c78b2bSJens Axboe 586f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 587f9c78b2bSJens Axboe 588f9c78b2bSJens Axboe if (bio_integrity(bio)) { 589f9c78b2bSJens Axboe int ret; 590f9c78b2bSJens Axboe 591f9c78b2bSJens Axboe ret = bio_integrity_clone(b, bio, gfp_mask); 592f9c78b2bSJens Axboe 593f9c78b2bSJens Axboe if (ret < 0) { 594f9c78b2bSJens Axboe bio_put(b); 595f9c78b2bSJens Axboe return NULL; 596f9c78b2bSJens Axboe } 597f9c78b2bSJens Axboe } 598f9c78b2bSJens Axboe 599f9c78b2bSJens Axboe return b; 600f9c78b2bSJens Axboe } 601f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 602f9c78b2bSJens Axboe 603f9c78b2bSJens Axboe /** 604f9c78b2bSJens Axboe * bio_clone_bioset - clone a bio 605f9c78b2bSJens Axboe * @bio_src: bio to clone 606f9c78b2bSJens Axboe * @gfp_mask: allocation priority 607f9c78b2bSJens Axboe * @bs: bio_set to allocate from 608f9c78b2bSJens Axboe * 609f9c78b2bSJens Axboe * Clone bio. Caller will own the returned bio, but not the actual data it 610f9c78b2bSJens Axboe * points to. Reference count of returned bio will be one. 611f9c78b2bSJens Axboe */ 612f9c78b2bSJens Axboe struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask, 613f9c78b2bSJens Axboe struct bio_set *bs) 614f9c78b2bSJens Axboe { 615f9c78b2bSJens Axboe struct bvec_iter iter; 616f9c78b2bSJens Axboe struct bio_vec bv; 617f9c78b2bSJens Axboe struct bio *bio; 618f9c78b2bSJens Axboe 619f9c78b2bSJens Axboe /* 620f9c78b2bSJens Axboe * Pre immutable biovecs, __bio_clone() used to just do a memcpy from 621f9c78b2bSJens Axboe * bio_src->bi_io_vec to bio->bi_io_vec. 622f9c78b2bSJens Axboe * 623f9c78b2bSJens Axboe * We can't do that anymore, because: 624f9c78b2bSJens Axboe * 625f9c78b2bSJens Axboe * - The point of cloning the biovec is to produce a bio with a biovec 626f9c78b2bSJens Axboe * the caller can modify: bi_idx and bi_bvec_done should be 0. 627f9c78b2bSJens Axboe * 628f9c78b2bSJens Axboe * - The original bio could've had more than BIO_MAX_PAGES biovecs; if 629f9c78b2bSJens Axboe * we tried to clone the whole thing bio_alloc_bioset() would fail. 630f9c78b2bSJens Axboe * But the clone should succeed as long as the number of biovecs we 631f9c78b2bSJens Axboe * actually need to allocate is fewer than BIO_MAX_PAGES. 632f9c78b2bSJens Axboe * 633f9c78b2bSJens Axboe * - Lastly, bi_vcnt should not be looked at or relied upon by code 634f9c78b2bSJens Axboe * that does not own the bio - reason being drivers don't use it for 635f9c78b2bSJens Axboe * iterating over the biovec anymore, so expecting it to be kept up 636f9c78b2bSJens Axboe * to date (i.e. for clones that share the parent biovec) is just 637f9c78b2bSJens Axboe * asking for trouble and would force extra work on 638f9c78b2bSJens Axboe * __bio_clone_fast() anyways. 639f9c78b2bSJens Axboe */ 640f9c78b2bSJens Axboe 641f9c78b2bSJens Axboe bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs); 642f9c78b2bSJens Axboe if (!bio) 643f9c78b2bSJens Axboe return NULL; 644f9c78b2bSJens Axboe 645f9c78b2bSJens Axboe bio->bi_bdev = bio_src->bi_bdev; 646f9c78b2bSJens Axboe bio->bi_rw = bio_src->bi_rw; 647f9c78b2bSJens Axboe bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; 648f9c78b2bSJens Axboe bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; 649f9c78b2bSJens Axboe 650f9c78b2bSJens Axboe if (bio->bi_rw & REQ_DISCARD) 651f9c78b2bSJens Axboe goto integrity_clone; 652f9c78b2bSJens Axboe 653f9c78b2bSJens Axboe if (bio->bi_rw & REQ_WRITE_SAME) { 654f9c78b2bSJens Axboe bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; 655f9c78b2bSJens Axboe goto integrity_clone; 656f9c78b2bSJens Axboe } 657f9c78b2bSJens Axboe 658f9c78b2bSJens Axboe bio_for_each_segment(bv, bio_src, iter) 659f9c78b2bSJens Axboe bio->bi_io_vec[bio->bi_vcnt++] = bv; 660f9c78b2bSJens Axboe 661f9c78b2bSJens Axboe integrity_clone: 662f9c78b2bSJens Axboe if (bio_integrity(bio_src)) { 663f9c78b2bSJens Axboe int ret; 664f9c78b2bSJens Axboe 665f9c78b2bSJens Axboe ret = bio_integrity_clone(bio, bio_src, gfp_mask); 666f9c78b2bSJens Axboe if (ret < 0) { 667f9c78b2bSJens Axboe bio_put(bio); 668f9c78b2bSJens Axboe return NULL; 669f9c78b2bSJens Axboe } 670f9c78b2bSJens Axboe } 671f9c78b2bSJens Axboe 672f9c78b2bSJens Axboe return bio; 673f9c78b2bSJens Axboe } 674f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_bioset); 675f9c78b2bSJens Axboe 676f9c78b2bSJens Axboe /** 677f9c78b2bSJens Axboe * bio_get_nr_vecs - return approx number of vecs 678f9c78b2bSJens Axboe * @bdev: I/O target 679f9c78b2bSJens Axboe * 680f9c78b2bSJens Axboe * Return the approximate number of pages we can send to this target. 681f9c78b2bSJens Axboe * There's no guarantee that you will be able to fit this number of pages 682f9c78b2bSJens Axboe * into a bio, it does not account for dynamic restrictions that vary 683f9c78b2bSJens Axboe * on offset. 684f9c78b2bSJens Axboe */ 685f9c78b2bSJens Axboe int bio_get_nr_vecs(struct block_device *bdev) 686f9c78b2bSJens Axboe { 687f9c78b2bSJens Axboe struct request_queue *q = bdev_get_queue(bdev); 688f9c78b2bSJens Axboe int nr_pages; 689f9c78b2bSJens Axboe 690f9c78b2bSJens Axboe nr_pages = min_t(unsigned, 691f9c78b2bSJens Axboe queue_max_segments(q), 692f9c78b2bSJens Axboe queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1); 693f9c78b2bSJens Axboe 694f9c78b2bSJens Axboe return min_t(unsigned, nr_pages, BIO_MAX_PAGES); 695f9c78b2bSJens Axboe 696f9c78b2bSJens Axboe } 697f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_get_nr_vecs); 698f9c78b2bSJens Axboe 699f9c78b2bSJens Axboe static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page 700f9c78b2bSJens Axboe *page, unsigned int len, unsigned int offset, 701f9c78b2bSJens Axboe unsigned int max_sectors) 702f9c78b2bSJens Axboe { 703f9c78b2bSJens Axboe int retried_segments = 0; 704f9c78b2bSJens Axboe struct bio_vec *bvec; 705f9c78b2bSJens Axboe 706f9c78b2bSJens Axboe /* 707f9c78b2bSJens Axboe * cloned bio must not modify vec list 708f9c78b2bSJens Axboe */ 709f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 710f9c78b2bSJens Axboe return 0; 711f9c78b2bSJens Axboe 712f9c78b2bSJens Axboe if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors) 713f9c78b2bSJens Axboe return 0; 714f9c78b2bSJens Axboe 715f9c78b2bSJens Axboe /* 716f9c78b2bSJens Axboe * For filesystems with a blocksize smaller than the pagesize 717f9c78b2bSJens Axboe * we will often be called with the same page as last time and 718f9c78b2bSJens Axboe * a consecutive offset. Optimize this special case. 719f9c78b2bSJens Axboe */ 720f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 721f9c78b2bSJens Axboe struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; 722f9c78b2bSJens Axboe 723f9c78b2bSJens Axboe if (page == prev->bv_page && 724f9c78b2bSJens Axboe offset == prev->bv_offset + prev->bv_len) { 725f9c78b2bSJens Axboe unsigned int prev_bv_len = prev->bv_len; 726f9c78b2bSJens Axboe prev->bv_len += len; 727f9c78b2bSJens Axboe 728f9c78b2bSJens Axboe if (q->merge_bvec_fn) { 729f9c78b2bSJens Axboe struct bvec_merge_data bvm = { 730f9c78b2bSJens Axboe /* prev_bvec is already charged in 731f9c78b2bSJens Axboe bi_size, discharge it in order to 732f9c78b2bSJens Axboe simulate merging updated prev_bvec 733f9c78b2bSJens Axboe as new bvec. */ 734f9c78b2bSJens Axboe .bi_bdev = bio->bi_bdev, 735f9c78b2bSJens Axboe .bi_sector = bio->bi_iter.bi_sector, 736f9c78b2bSJens Axboe .bi_size = bio->bi_iter.bi_size - 737f9c78b2bSJens Axboe prev_bv_len, 738f9c78b2bSJens Axboe .bi_rw = bio->bi_rw, 739f9c78b2bSJens Axboe }; 740f9c78b2bSJens Axboe 741f9c78b2bSJens Axboe if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) { 742f9c78b2bSJens Axboe prev->bv_len -= len; 743f9c78b2bSJens Axboe return 0; 744f9c78b2bSJens Axboe } 745f9c78b2bSJens Axboe } 746f9c78b2bSJens Axboe 747f9c78b2bSJens Axboe goto done; 748f9c78b2bSJens Axboe } 749f9c78b2bSJens Axboe } 750f9c78b2bSJens Axboe 751f9c78b2bSJens Axboe if (bio->bi_vcnt >= bio->bi_max_vecs) 752f9c78b2bSJens Axboe return 0; 753f9c78b2bSJens Axboe 754f9c78b2bSJens Axboe /* 755f9c78b2bSJens Axboe * we might lose a segment or two here, but rather that than 756f9c78b2bSJens Axboe * make this too complex. 757f9c78b2bSJens Axboe */ 758f9c78b2bSJens Axboe 759f9c78b2bSJens Axboe while (bio->bi_phys_segments >= queue_max_segments(q)) { 760f9c78b2bSJens Axboe 761f9c78b2bSJens Axboe if (retried_segments) 762f9c78b2bSJens Axboe return 0; 763f9c78b2bSJens Axboe 764f9c78b2bSJens Axboe retried_segments = 1; 765f9c78b2bSJens Axboe blk_recount_segments(q, bio); 766f9c78b2bSJens Axboe } 767f9c78b2bSJens Axboe 768f9c78b2bSJens Axboe /* 769f9c78b2bSJens Axboe * setup the new entry, we might clear it again later if we 770f9c78b2bSJens Axboe * cannot add the page 771f9c78b2bSJens Axboe */ 772f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 773f9c78b2bSJens Axboe bvec->bv_page = page; 774f9c78b2bSJens Axboe bvec->bv_len = len; 775f9c78b2bSJens Axboe bvec->bv_offset = offset; 776f9c78b2bSJens Axboe 777f9c78b2bSJens Axboe /* 778f9c78b2bSJens Axboe * if queue has other restrictions (eg varying max sector size 779f9c78b2bSJens Axboe * depending on offset), it can specify a merge_bvec_fn in the 780f9c78b2bSJens Axboe * queue to get further control 781f9c78b2bSJens Axboe */ 782f9c78b2bSJens Axboe if (q->merge_bvec_fn) { 783f9c78b2bSJens Axboe struct bvec_merge_data bvm = { 784f9c78b2bSJens Axboe .bi_bdev = bio->bi_bdev, 785f9c78b2bSJens Axboe .bi_sector = bio->bi_iter.bi_sector, 786f9c78b2bSJens Axboe .bi_size = bio->bi_iter.bi_size, 787f9c78b2bSJens Axboe .bi_rw = bio->bi_rw, 788f9c78b2bSJens Axboe }; 789f9c78b2bSJens Axboe 790f9c78b2bSJens Axboe /* 791f9c78b2bSJens Axboe * merge_bvec_fn() returns number of bytes it can accept 792f9c78b2bSJens Axboe * at this offset 793f9c78b2bSJens Axboe */ 794f9c78b2bSJens Axboe if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) { 795f9c78b2bSJens Axboe bvec->bv_page = NULL; 796f9c78b2bSJens Axboe bvec->bv_len = 0; 797f9c78b2bSJens Axboe bvec->bv_offset = 0; 798f9c78b2bSJens Axboe return 0; 799f9c78b2bSJens Axboe } 800f9c78b2bSJens Axboe } 801f9c78b2bSJens Axboe 802f9c78b2bSJens Axboe /* If we may be able to merge these biovecs, force a recount */ 803f9c78b2bSJens Axboe if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec))) 804f9c78b2bSJens Axboe bio->bi_flags &= ~(1 << BIO_SEG_VALID); 805f9c78b2bSJens Axboe 806f9c78b2bSJens Axboe bio->bi_vcnt++; 807f9c78b2bSJens Axboe bio->bi_phys_segments++; 808f9c78b2bSJens Axboe done: 809f9c78b2bSJens Axboe bio->bi_iter.bi_size += len; 810f9c78b2bSJens Axboe return len; 811f9c78b2bSJens Axboe } 812f9c78b2bSJens Axboe 813f9c78b2bSJens Axboe /** 814f9c78b2bSJens Axboe * bio_add_pc_page - attempt to add page to bio 815f9c78b2bSJens Axboe * @q: the target queue 816f9c78b2bSJens Axboe * @bio: destination bio 817f9c78b2bSJens Axboe * @page: page to add 818f9c78b2bSJens Axboe * @len: vec entry length 819f9c78b2bSJens Axboe * @offset: vec entry offset 820f9c78b2bSJens Axboe * 821f9c78b2bSJens Axboe * Attempt to add a page to the bio_vec maplist. This can fail for a 822f9c78b2bSJens Axboe * number of reasons, such as the bio being full or target block device 823f9c78b2bSJens Axboe * limitations. The target block device must allow bio's up to PAGE_SIZE, 824f9c78b2bSJens Axboe * so it is always possible to add a single page to an empty bio. 825f9c78b2bSJens Axboe * 826f9c78b2bSJens Axboe * This should only be used by REQ_PC bios. 827f9c78b2bSJens Axboe */ 828f9c78b2bSJens Axboe int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page, 829f9c78b2bSJens Axboe unsigned int len, unsigned int offset) 830f9c78b2bSJens Axboe { 831f9c78b2bSJens Axboe return __bio_add_page(q, bio, page, len, offset, 832f9c78b2bSJens Axboe queue_max_hw_sectors(q)); 833f9c78b2bSJens Axboe } 834f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 835f9c78b2bSJens Axboe 836f9c78b2bSJens Axboe /** 837f9c78b2bSJens Axboe * bio_add_page - attempt to add page to bio 838f9c78b2bSJens Axboe * @bio: destination bio 839f9c78b2bSJens Axboe * @page: page to add 840f9c78b2bSJens Axboe * @len: vec entry length 841f9c78b2bSJens Axboe * @offset: vec entry offset 842f9c78b2bSJens Axboe * 843f9c78b2bSJens Axboe * Attempt to add a page to the bio_vec maplist. This can fail for a 844f9c78b2bSJens Axboe * number of reasons, such as the bio being full or target block device 845f9c78b2bSJens Axboe * limitations. The target block device must allow bio's up to PAGE_SIZE, 846f9c78b2bSJens Axboe * so it is always possible to add a single page to an empty bio. 847f9c78b2bSJens Axboe */ 848f9c78b2bSJens Axboe int bio_add_page(struct bio *bio, struct page *page, unsigned int len, 849f9c78b2bSJens Axboe unsigned int offset) 850f9c78b2bSJens Axboe { 851f9c78b2bSJens Axboe struct request_queue *q = bdev_get_queue(bio->bi_bdev); 852f9c78b2bSJens Axboe return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q)); 853f9c78b2bSJens Axboe } 854f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 855f9c78b2bSJens Axboe 856f9c78b2bSJens Axboe struct submit_bio_ret { 857f9c78b2bSJens Axboe struct completion event; 858f9c78b2bSJens Axboe int error; 859f9c78b2bSJens Axboe }; 860f9c78b2bSJens Axboe 861f9c78b2bSJens Axboe static void submit_bio_wait_endio(struct bio *bio, int error) 862f9c78b2bSJens Axboe { 863f9c78b2bSJens Axboe struct submit_bio_ret *ret = bio->bi_private; 864f9c78b2bSJens Axboe 865f9c78b2bSJens Axboe ret->error = error; 866f9c78b2bSJens Axboe complete(&ret->event); 867f9c78b2bSJens Axboe } 868f9c78b2bSJens Axboe 869f9c78b2bSJens Axboe /** 870f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 871f9c78b2bSJens Axboe * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead) 872f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 873f9c78b2bSJens Axboe * 874f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 875f9c78b2bSJens Axboe * bio_endio() on failure. 876f9c78b2bSJens Axboe */ 877f9c78b2bSJens Axboe int submit_bio_wait(int rw, struct bio *bio) 878f9c78b2bSJens Axboe { 879f9c78b2bSJens Axboe struct submit_bio_ret ret; 880f9c78b2bSJens Axboe 881f9c78b2bSJens Axboe rw |= REQ_SYNC; 882f9c78b2bSJens Axboe init_completion(&ret.event); 883f9c78b2bSJens Axboe bio->bi_private = &ret; 884f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 885f9c78b2bSJens Axboe submit_bio(rw, bio); 886f9c78b2bSJens Axboe wait_for_completion(&ret.event); 887f9c78b2bSJens Axboe 888f9c78b2bSJens Axboe return ret.error; 889f9c78b2bSJens Axboe } 890f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 891f9c78b2bSJens Axboe 892f9c78b2bSJens Axboe /** 893f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 894f9c78b2bSJens Axboe * @bio: bio to advance 895f9c78b2bSJens Axboe * @bytes: number of bytes to complete 896f9c78b2bSJens Axboe * 897f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 898f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 899f9c78b2bSJens Axboe * be updated on the last bvec as well. 900f9c78b2bSJens Axboe * 901f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 902f9c78b2bSJens Axboe */ 903f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 904f9c78b2bSJens Axboe { 905f9c78b2bSJens Axboe if (bio_integrity(bio)) 906f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 907f9c78b2bSJens Axboe 908f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 909f9c78b2bSJens Axboe } 910f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 911f9c78b2bSJens Axboe 912f9c78b2bSJens Axboe /** 913f9c78b2bSJens Axboe * bio_alloc_pages - allocates a single page for each bvec in a bio 914f9c78b2bSJens Axboe * @bio: bio to allocate pages for 915f9c78b2bSJens Axboe * @gfp_mask: flags for allocation 916f9c78b2bSJens Axboe * 917f9c78b2bSJens Axboe * Allocates pages up to @bio->bi_vcnt. 918f9c78b2bSJens Axboe * 919f9c78b2bSJens Axboe * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are 920f9c78b2bSJens Axboe * freed. 921f9c78b2bSJens Axboe */ 922f9c78b2bSJens Axboe int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask) 923f9c78b2bSJens Axboe { 924f9c78b2bSJens Axboe int i; 925f9c78b2bSJens Axboe struct bio_vec *bv; 926f9c78b2bSJens Axboe 927f9c78b2bSJens Axboe bio_for_each_segment_all(bv, bio, i) { 928f9c78b2bSJens Axboe bv->bv_page = alloc_page(gfp_mask); 929f9c78b2bSJens Axboe if (!bv->bv_page) { 930f9c78b2bSJens Axboe while (--bv >= bio->bi_io_vec) 931f9c78b2bSJens Axboe __free_page(bv->bv_page); 932f9c78b2bSJens Axboe return -ENOMEM; 933f9c78b2bSJens Axboe } 934f9c78b2bSJens Axboe } 935f9c78b2bSJens Axboe 936f9c78b2bSJens Axboe return 0; 937f9c78b2bSJens Axboe } 938f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_pages); 939f9c78b2bSJens Axboe 940f9c78b2bSJens Axboe /** 941f9c78b2bSJens Axboe * bio_copy_data - copy contents of data buffers from one chain of bios to 942f9c78b2bSJens Axboe * another 943f9c78b2bSJens Axboe * @src: source bio list 944f9c78b2bSJens Axboe * @dst: destination bio list 945f9c78b2bSJens Axboe * 946f9c78b2bSJens Axboe * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats 947f9c78b2bSJens Axboe * @src and @dst as linked lists of bios. 948f9c78b2bSJens Axboe * 949f9c78b2bSJens Axboe * Stops when it reaches the end of either @src or @dst - that is, copies 950f9c78b2bSJens Axboe * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 951f9c78b2bSJens Axboe */ 952f9c78b2bSJens Axboe void bio_copy_data(struct bio *dst, struct bio *src) 953f9c78b2bSJens Axboe { 954f9c78b2bSJens Axboe struct bvec_iter src_iter, dst_iter; 955f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 956f9c78b2bSJens Axboe void *src_p, *dst_p; 957f9c78b2bSJens Axboe unsigned bytes; 958f9c78b2bSJens Axboe 959f9c78b2bSJens Axboe src_iter = src->bi_iter; 960f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 961f9c78b2bSJens Axboe 962f9c78b2bSJens Axboe while (1) { 963f9c78b2bSJens Axboe if (!src_iter.bi_size) { 964f9c78b2bSJens Axboe src = src->bi_next; 965f9c78b2bSJens Axboe if (!src) 966f9c78b2bSJens Axboe break; 967f9c78b2bSJens Axboe 968f9c78b2bSJens Axboe src_iter = src->bi_iter; 969f9c78b2bSJens Axboe } 970f9c78b2bSJens Axboe 971f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 972f9c78b2bSJens Axboe dst = dst->bi_next; 973f9c78b2bSJens Axboe if (!dst) 974f9c78b2bSJens Axboe break; 975f9c78b2bSJens Axboe 976f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 977f9c78b2bSJens Axboe } 978f9c78b2bSJens Axboe 979f9c78b2bSJens Axboe src_bv = bio_iter_iovec(src, src_iter); 980f9c78b2bSJens Axboe dst_bv = bio_iter_iovec(dst, dst_iter); 981f9c78b2bSJens Axboe 982f9c78b2bSJens Axboe bytes = min(src_bv.bv_len, dst_bv.bv_len); 983f9c78b2bSJens Axboe 984f9c78b2bSJens Axboe src_p = kmap_atomic(src_bv.bv_page); 985f9c78b2bSJens Axboe dst_p = kmap_atomic(dst_bv.bv_page); 986f9c78b2bSJens Axboe 987f9c78b2bSJens Axboe memcpy(dst_p + dst_bv.bv_offset, 988f9c78b2bSJens Axboe src_p + src_bv.bv_offset, 989f9c78b2bSJens Axboe bytes); 990f9c78b2bSJens Axboe 991f9c78b2bSJens Axboe kunmap_atomic(dst_p); 992f9c78b2bSJens Axboe kunmap_atomic(src_p); 993f9c78b2bSJens Axboe 994f9c78b2bSJens Axboe bio_advance_iter(src, &src_iter, bytes); 995f9c78b2bSJens Axboe bio_advance_iter(dst, &dst_iter, bytes); 996f9c78b2bSJens Axboe } 997f9c78b2bSJens Axboe } 998f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_copy_data); 999f9c78b2bSJens Axboe 1000f9c78b2bSJens Axboe struct bio_map_data { 1001f9c78b2bSJens Axboe int nr_sgvecs; 1002f9c78b2bSJens Axboe int is_our_pages; 1003f9c78b2bSJens Axboe struct sg_iovec sgvecs[]; 1004f9c78b2bSJens Axboe }; 1005f9c78b2bSJens Axboe 1006f9c78b2bSJens Axboe static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio, 1007f9c78b2bSJens Axboe const struct sg_iovec *iov, int iov_count, 1008f9c78b2bSJens Axboe int is_our_pages) 1009f9c78b2bSJens Axboe { 1010f9c78b2bSJens Axboe memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count); 1011f9c78b2bSJens Axboe bmd->nr_sgvecs = iov_count; 1012f9c78b2bSJens Axboe bmd->is_our_pages = is_our_pages; 1013f9c78b2bSJens Axboe bio->bi_private = bmd; 1014f9c78b2bSJens Axboe } 1015f9c78b2bSJens Axboe 1016f9c78b2bSJens Axboe static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count, 1017f9c78b2bSJens Axboe gfp_t gfp_mask) 1018f9c78b2bSJens Axboe { 1019f9c78b2bSJens Axboe if (iov_count > UIO_MAXIOV) 1020f9c78b2bSJens Axboe return NULL; 1021f9c78b2bSJens Axboe 1022f9c78b2bSJens Axboe return kmalloc(sizeof(struct bio_map_data) + 1023f9c78b2bSJens Axboe sizeof(struct sg_iovec) * iov_count, gfp_mask); 1024f9c78b2bSJens Axboe } 1025f9c78b2bSJens Axboe 1026f9c78b2bSJens Axboe static int __bio_copy_iov(struct bio *bio, const struct sg_iovec *iov, int iov_count, 1027f9c78b2bSJens Axboe int to_user, int from_user, int do_free_page) 1028f9c78b2bSJens Axboe { 1029f9c78b2bSJens Axboe int ret = 0, i; 1030f9c78b2bSJens Axboe struct bio_vec *bvec; 1031f9c78b2bSJens Axboe int iov_idx = 0; 1032f9c78b2bSJens Axboe unsigned int iov_off = 0; 1033f9c78b2bSJens Axboe 1034f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1035f9c78b2bSJens Axboe char *bv_addr = page_address(bvec->bv_page); 1036f9c78b2bSJens Axboe unsigned int bv_len = bvec->bv_len; 1037f9c78b2bSJens Axboe 1038f9c78b2bSJens Axboe while (bv_len && iov_idx < iov_count) { 1039f9c78b2bSJens Axboe unsigned int bytes; 1040f9c78b2bSJens Axboe char __user *iov_addr; 1041f9c78b2bSJens Axboe 1042f9c78b2bSJens Axboe bytes = min_t(unsigned int, 1043f9c78b2bSJens Axboe iov[iov_idx].iov_len - iov_off, bv_len); 1044f9c78b2bSJens Axboe iov_addr = iov[iov_idx].iov_base + iov_off; 1045f9c78b2bSJens Axboe 1046f9c78b2bSJens Axboe if (!ret) { 1047f9c78b2bSJens Axboe if (to_user) 1048f9c78b2bSJens Axboe ret = copy_to_user(iov_addr, bv_addr, 1049f9c78b2bSJens Axboe bytes); 1050f9c78b2bSJens Axboe 1051f9c78b2bSJens Axboe if (from_user) 1052f9c78b2bSJens Axboe ret = copy_from_user(bv_addr, iov_addr, 1053f9c78b2bSJens Axboe bytes); 1054f9c78b2bSJens Axboe 1055f9c78b2bSJens Axboe if (ret) 1056f9c78b2bSJens Axboe ret = -EFAULT; 1057f9c78b2bSJens Axboe } 1058f9c78b2bSJens Axboe 1059f9c78b2bSJens Axboe bv_len -= bytes; 1060f9c78b2bSJens Axboe bv_addr += bytes; 1061f9c78b2bSJens Axboe iov_addr += bytes; 1062f9c78b2bSJens Axboe iov_off += bytes; 1063f9c78b2bSJens Axboe 1064f9c78b2bSJens Axboe if (iov[iov_idx].iov_len == iov_off) { 1065f9c78b2bSJens Axboe iov_idx++; 1066f9c78b2bSJens Axboe iov_off = 0; 1067f9c78b2bSJens Axboe } 1068f9c78b2bSJens Axboe } 1069f9c78b2bSJens Axboe 1070f9c78b2bSJens Axboe if (do_free_page) 1071f9c78b2bSJens Axboe __free_page(bvec->bv_page); 1072f9c78b2bSJens Axboe } 1073f9c78b2bSJens Axboe 1074f9c78b2bSJens Axboe return ret; 1075f9c78b2bSJens Axboe } 1076f9c78b2bSJens Axboe 1077f9c78b2bSJens Axboe /** 1078f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1079f9c78b2bSJens Axboe * @bio: bio being terminated 1080f9c78b2bSJens Axboe * 1081f9c78b2bSJens Axboe * Free pages allocated from bio_copy_user() and write back data 1082f9c78b2bSJens Axboe * to user space in case of a read. 1083f9c78b2bSJens Axboe */ 1084f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1085f9c78b2bSJens Axboe { 1086f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 1087f9c78b2bSJens Axboe struct bio_vec *bvec; 1088f9c78b2bSJens Axboe int ret = 0, i; 1089f9c78b2bSJens Axboe 1090f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1091f9c78b2bSJens Axboe /* 1092f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 1093f9c78b2bSJens Axboe * don't copy into a random user address space, just free. 1094f9c78b2bSJens Axboe */ 1095f9c78b2bSJens Axboe if (current->mm) 1096f9c78b2bSJens Axboe ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1097f9c78b2bSJens Axboe bio_data_dir(bio) == READ, 1098f9c78b2bSJens Axboe 0, bmd->is_our_pages); 1099f9c78b2bSJens Axboe else if (bmd->is_our_pages) 1100f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) 1101f9c78b2bSJens Axboe __free_page(bvec->bv_page); 1102f9c78b2bSJens Axboe } 1103f9c78b2bSJens Axboe kfree(bmd); 1104f9c78b2bSJens Axboe bio_put(bio); 1105f9c78b2bSJens Axboe return ret; 1106f9c78b2bSJens Axboe } 1107f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_uncopy_user); 1108f9c78b2bSJens Axboe 1109f9c78b2bSJens Axboe /** 1110f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1111f9c78b2bSJens Axboe * @q: destination block queue 1112f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 1113f9c78b2bSJens Axboe * @iov: the iovec. 1114f9c78b2bSJens Axboe * @iov_count: number of elements in the iovec 1115f9c78b2bSJens Axboe * @write_to_vm: bool indicating writing to pages or not 1116f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1117f9c78b2bSJens Axboe * 1118f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1119f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1120f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1121f9c78b2bSJens Axboe */ 1122f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1123f9c78b2bSJens Axboe struct rq_map_data *map_data, 1124f9c78b2bSJens Axboe const struct sg_iovec *iov, int iov_count, 1125f9c78b2bSJens Axboe int write_to_vm, gfp_t gfp_mask) 1126f9c78b2bSJens Axboe { 1127f9c78b2bSJens Axboe struct bio_map_data *bmd; 1128f9c78b2bSJens Axboe struct bio_vec *bvec; 1129f9c78b2bSJens Axboe struct page *page; 1130f9c78b2bSJens Axboe struct bio *bio; 1131f9c78b2bSJens Axboe int i, ret; 1132f9c78b2bSJens Axboe int nr_pages = 0; 1133f9c78b2bSJens Axboe unsigned int len = 0; 1134f9c78b2bSJens Axboe unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0; 1135f9c78b2bSJens Axboe 1136f9c78b2bSJens Axboe for (i = 0; i < iov_count; i++) { 1137f9c78b2bSJens Axboe unsigned long uaddr; 1138f9c78b2bSJens Axboe unsigned long end; 1139f9c78b2bSJens Axboe unsigned long start; 1140f9c78b2bSJens Axboe 1141f9c78b2bSJens Axboe uaddr = (unsigned long)iov[i].iov_base; 1142f9c78b2bSJens Axboe end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1143f9c78b2bSJens Axboe start = uaddr >> PAGE_SHIFT; 1144f9c78b2bSJens Axboe 1145f9c78b2bSJens Axboe /* 1146f9c78b2bSJens Axboe * Overflow, abort 1147f9c78b2bSJens Axboe */ 1148f9c78b2bSJens Axboe if (end < start) 1149f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1150f9c78b2bSJens Axboe 1151f9c78b2bSJens Axboe nr_pages += end - start; 1152f9c78b2bSJens Axboe len += iov[i].iov_len; 1153f9c78b2bSJens Axboe } 1154f9c78b2bSJens Axboe 1155f9c78b2bSJens Axboe if (offset) 1156f9c78b2bSJens Axboe nr_pages++; 1157f9c78b2bSJens Axboe 1158f9c78b2bSJens Axboe bmd = bio_alloc_map_data(iov_count, gfp_mask); 1159f9c78b2bSJens Axboe if (!bmd) 1160f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1161f9c78b2bSJens Axboe 1162f9c78b2bSJens Axboe ret = -ENOMEM; 1163f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1164f9c78b2bSJens Axboe if (!bio) 1165f9c78b2bSJens Axboe goto out_bmd; 1166f9c78b2bSJens Axboe 1167f9c78b2bSJens Axboe if (!write_to_vm) 1168f9c78b2bSJens Axboe bio->bi_rw |= REQ_WRITE; 1169f9c78b2bSJens Axboe 1170f9c78b2bSJens Axboe ret = 0; 1171f9c78b2bSJens Axboe 1172f9c78b2bSJens Axboe if (map_data) { 1173f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1174f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1175f9c78b2bSJens Axboe } 1176f9c78b2bSJens Axboe while (len) { 1177f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1178f9c78b2bSJens Axboe 1179f9c78b2bSJens Axboe bytes -= offset; 1180f9c78b2bSJens Axboe 1181f9c78b2bSJens Axboe if (bytes > len) 1182f9c78b2bSJens Axboe bytes = len; 1183f9c78b2bSJens Axboe 1184f9c78b2bSJens Axboe if (map_data) { 1185f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1186f9c78b2bSJens Axboe ret = -ENOMEM; 1187f9c78b2bSJens Axboe break; 1188f9c78b2bSJens Axboe } 1189f9c78b2bSJens Axboe 1190f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1191f9c78b2bSJens Axboe page += (i % nr_pages); 1192f9c78b2bSJens Axboe 1193f9c78b2bSJens Axboe i++; 1194f9c78b2bSJens Axboe } else { 1195f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1196f9c78b2bSJens Axboe if (!page) { 1197f9c78b2bSJens Axboe ret = -ENOMEM; 1198f9c78b2bSJens Axboe break; 1199f9c78b2bSJens Axboe } 1200f9c78b2bSJens Axboe } 1201f9c78b2bSJens Axboe 1202f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) 1203f9c78b2bSJens Axboe break; 1204f9c78b2bSJens Axboe 1205f9c78b2bSJens Axboe len -= bytes; 1206f9c78b2bSJens Axboe offset = 0; 1207f9c78b2bSJens Axboe } 1208f9c78b2bSJens Axboe 1209f9c78b2bSJens Axboe if (ret) 1210f9c78b2bSJens Axboe goto cleanup; 1211f9c78b2bSJens Axboe 1212f9c78b2bSJens Axboe /* 1213f9c78b2bSJens Axboe * success 1214f9c78b2bSJens Axboe */ 1215f9c78b2bSJens Axboe if ((!write_to_vm && (!map_data || !map_data->null_mapped)) || 1216f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 1217f9c78b2bSJens Axboe ret = __bio_copy_iov(bio, iov, iov_count, 0, 1, 0); 1218f9c78b2bSJens Axboe if (ret) 1219f9c78b2bSJens Axboe goto cleanup; 1220f9c78b2bSJens Axboe } 1221f9c78b2bSJens Axboe 1222f9c78b2bSJens Axboe bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1); 1223f9c78b2bSJens Axboe return bio; 1224f9c78b2bSJens Axboe cleanup: 1225f9c78b2bSJens Axboe if (!map_data) 1226f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) 1227f9c78b2bSJens Axboe __free_page(bvec->bv_page); 1228f9c78b2bSJens Axboe 1229f9c78b2bSJens Axboe bio_put(bio); 1230f9c78b2bSJens Axboe out_bmd: 1231f9c78b2bSJens Axboe kfree(bmd); 1232f9c78b2bSJens Axboe return ERR_PTR(ret); 1233f9c78b2bSJens Axboe } 1234f9c78b2bSJens Axboe 1235f9c78b2bSJens Axboe /** 1236f9c78b2bSJens Axboe * bio_copy_user - copy user data to bio 1237f9c78b2bSJens Axboe * @q: destination block queue 1238f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 1239f9c78b2bSJens Axboe * @uaddr: start of user address 1240f9c78b2bSJens Axboe * @len: length in bytes 1241f9c78b2bSJens Axboe * @write_to_vm: bool indicating writing to pages or not 1242f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1243f9c78b2bSJens Axboe * 1244f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1245f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1246f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1247f9c78b2bSJens Axboe */ 1248f9c78b2bSJens Axboe struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data, 1249f9c78b2bSJens Axboe unsigned long uaddr, unsigned int len, 1250f9c78b2bSJens Axboe int write_to_vm, gfp_t gfp_mask) 1251f9c78b2bSJens Axboe { 1252f9c78b2bSJens Axboe struct sg_iovec iov; 1253f9c78b2bSJens Axboe 1254f9c78b2bSJens Axboe iov.iov_base = (void __user *)uaddr; 1255f9c78b2bSJens Axboe iov.iov_len = len; 1256f9c78b2bSJens Axboe 1257f9c78b2bSJens Axboe return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask); 1258f9c78b2bSJens Axboe } 1259f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_copy_user); 1260f9c78b2bSJens Axboe 1261f9c78b2bSJens Axboe static struct bio *__bio_map_user_iov(struct request_queue *q, 1262f9c78b2bSJens Axboe struct block_device *bdev, 1263f9c78b2bSJens Axboe const struct sg_iovec *iov, int iov_count, 1264f9c78b2bSJens Axboe int write_to_vm, gfp_t gfp_mask) 1265f9c78b2bSJens Axboe { 1266f9c78b2bSJens Axboe int i, j; 1267f9c78b2bSJens Axboe int nr_pages = 0; 1268f9c78b2bSJens Axboe struct page **pages; 1269f9c78b2bSJens Axboe struct bio *bio; 1270f9c78b2bSJens Axboe int cur_page = 0; 1271f9c78b2bSJens Axboe int ret, offset; 1272f9c78b2bSJens Axboe 1273f9c78b2bSJens Axboe for (i = 0; i < iov_count; i++) { 1274f9c78b2bSJens Axboe unsigned long uaddr = (unsigned long)iov[i].iov_base; 1275f9c78b2bSJens Axboe unsigned long len = iov[i].iov_len; 1276f9c78b2bSJens Axboe unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1277f9c78b2bSJens Axboe unsigned long start = uaddr >> PAGE_SHIFT; 1278f9c78b2bSJens Axboe 1279f9c78b2bSJens Axboe /* 1280f9c78b2bSJens Axboe * Overflow, abort 1281f9c78b2bSJens Axboe */ 1282f9c78b2bSJens Axboe if (end < start) 1283f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1284f9c78b2bSJens Axboe 1285f9c78b2bSJens Axboe nr_pages += end - start; 1286f9c78b2bSJens Axboe /* 1287f9c78b2bSJens Axboe * buffer must be aligned to at least hardsector size for now 1288f9c78b2bSJens Axboe */ 1289f9c78b2bSJens Axboe if (uaddr & queue_dma_alignment(q)) 1290f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1291f9c78b2bSJens Axboe } 1292f9c78b2bSJens Axboe 1293f9c78b2bSJens Axboe if (!nr_pages) 1294f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1295f9c78b2bSJens Axboe 1296f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1297f9c78b2bSJens Axboe if (!bio) 1298f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1299f9c78b2bSJens Axboe 1300f9c78b2bSJens Axboe ret = -ENOMEM; 1301f9c78b2bSJens Axboe pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask); 1302f9c78b2bSJens Axboe if (!pages) 1303f9c78b2bSJens Axboe goto out; 1304f9c78b2bSJens Axboe 1305f9c78b2bSJens Axboe for (i = 0; i < iov_count; i++) { 1306f9c78b2bSJens Axboe unsigned long uaddr = (unsigned long)iov[i].iov_base; 1307f9c78b2bSJens Axboe unsigned long len = iov[i].iov_len; 1308f9c78b2bSJens Axboe unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1309f9c78b2bSJens Axboe unsigned long start = uaddr >> PAGE_SHIFT; 1310f9c78b2bSJens Axboe const int local_nr_pages = end - start; 1311f9c78b2bSJens Axboe const int page_limit = cur_page + local_nr_pages; 1312f9c78b2bSJens Axboe 1313f9c78b2bSJens Axboe ret = get_user_pages_fast(uaddr, local_nr_pages, 1314f9c78b2bSJens Axboe write_to_vm, &pages[cur_page]); 1315f9c78b2bSJens Axboe if (ret < local_nr_pages) { 1316f9c78b2bSJens Axboe ret = -EFAULT; 1317f9c78b2bSJens Axboe goto out_unmap; 1318f9c78b2bSJens Axboe } 1319f9c78b2bSJens Axboe 1320f9c78b2bSJens Axboe offset = uaddr & ~PAGE_MASK; 1321f9c78b2bSJens Axboe for (j = cur_page; j < page_limit; j++) { 1322f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1323f9c78b2bSJens Axboe 1324f9c78b2bSJens Axboe if (len <= 0) 1325f9c78b2bSJens Axboe break; 1326f9c78b2bSJens Axboe 1327f9c78b2bSJens Axboe if (bytes > len) 1328f9c78b2bSJens Axboe bytes = len; 1329f9c78b2bSJens Axboe 1330f9c78b2bSJens Axboe /* 1331f9c78b2bSJens Axboe * sorry... 1332f9c78b2bSJens Axboe */ 1333f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, pages[j], bytes, offset) < 1334f9c78b2bSJens Axboe bytes) 1335f9c78b2bSJens Axboe break; 1336f9c78b2bSJens Axboe 1337f9c78b2bSJens Axboe len -= bytes; 1338f9c78b2bSJens Axboe offset = 0; 1339f9c78b2bSJens Axboe } 1340f9c78b2bSJens Axboe 1341f9c78b2bSJens Axboe cur_page = j; 1342f9c78b2bSJens Axboe /* 1343f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1344f9c78b2bSJens Axboe */ 1345f9c78b2bSJens Axboe while (j < page_limit) 1346f9c78b2bSJens Axboe page_cache_release(pages[j++]); 1347f9c78b2bSJens Axboe } 1348f9c78b2bSJens Axboe 1349f9c78b2bSJens Axboe kfree(pages); 1350f9c78b2bSJens Axboe 1351f9c78b2bSJens Axboe /* 1352f9c78b2bSJens Axboe * set data direction, and check if mapped pages need bouncing 1353f9c78b2bSJens Axboe */ 1354f9c78b2bSJens Axboe if (!write_to_vm) 1355f9c78b2bSJens Axboe bio->bi_rw |= REQ_WRITE; 1356f9c78b2bSJens Axboe 1357f9c78b2bSJens Axboe bio->bi_bdev = bdev; 1358f9c78b2bSJens Axboe bio->bi_flags |= (1 << BIO_USER_MAPPED); 1359f9c78b2bSJens Axboe return bio; 1360f9c78b2bSJens Axboe 1361f9c78b2bSJens Axboe out_unmap: 1362f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1363f9c78b2bSJens Axboe if(!pages[i]) 1364f9c78b2bSJens Axboe break; 1365f9c78b2bSJens Axboe page_cache_release(pages[i]); 1366f9c78b2bSJens Axboe } 1367f9c78b2bSJens Axboe out: 1368f9c78b2bSJens Axboe kfree(pages); 1369f9c78b2bSJens Axboe bio_put(bio); 1370f9c78b2bSJens Axboe return ERR_PTR(ret); 1371f9c78b2bSJens Axboe } 1372f9c78b2bSJens Axboe 1373f9c78b2bSJens Axboe /** 1374f9c78b2bSJens Axboe * bio_map_user - map user address into bio 1375f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1376f9c78b2bSJens Axboe * @bdev: destination block device 1377f9c78b2bSJens Axboe * @uaddr: start of user address 1378f9c78b2bSJens Axboe * @len: length in bytes 1379f9c78b2bSJens Axboe * @write_to_vm: bool indicating writing to pages or not 1380f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1381f9c78b2bSJens Axboe * 1382f9c78b2bSJens Axboe * Map the user space address into a bio suitable for io to a block 1383f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1384f9c78b2bSJens Axboe */ 1385f9c78b2bSJens Axboe struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev, 1386f9c78b2bSJens Axboe unsigned long uaddr, unsigned int len, int write_to_vm, 1387f9c78b2bSJens Axboe gfp_t gfp_mask) 1388f9c78b2bSJens Axboe { 1389f9c78b2bSJens Axboe struct sg_iovec iov; 1390f9c78b2bSJens Axboe 1391f9c78b2bSJens Axboe iov.iov_base = (void __user *)uaddr; 1392f9c78b2bSJens Axboe iov.iov_len = len; 1393f9c78b2bSJens Axboe 1394f9c78b2bSJens Axboe return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask); 1395f9c78b2bSJens Axboe } 1396f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_map_user); 1397f9c78b2bSJens Axboe 1398f9c78b2bSJens Axboe /** 1399f9c78b2bSJens Axboe * bio_map_user_iov - map user sg_iovec table into bio 1400f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1401f9c78b2bSJens Axboe * @bdev: destination block device 1402f9c78b2bSJens Axboe * @iov: the iovec. 1403f9c78b2bSJens Axboe * @iov_count: number of elements in the iovec 1404f9c78b2bSJens Axboe * @write_to_vm: bool indicating writing to pages or not 1405f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1406f9c78b2bSJens Axboe * 1407f9c78b2bSJens Axboe * Map the user space address into a bio suitable for io to a block 1408f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1409f9c78b2bSJens Axboe */ 1410f9c78b2bSJens Axboe struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev, 1411f9c78b2bSJens Axboe const struct sg_iovec *iov, int iov_count, 1412f9c78b2bSJens Axboe int write_to_vm, gfp_t gfp_mask) 1413f9c78b2bSJens Axboe { 1414f9c78b2bSJens Axboe struct bio *bio; 1415f9c78b2bSJens Axboe 1416f9c78b2bSJens Axboe bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm, 1417f9c78b2bSJens Axboe gfp_mask); 1418f9c78b2bSJens Axboe if (IS_ERR(bio)) 1419f9c78b2bSJens Axboe return bio; 1420f9c78b2bSJens Axboe 1421f9c78b2bSJens Axboe /* 1422f9c78b2bSJens Axboe * subtle -- if __bio_map_user() ended up bouncing a bio, 1423f9c78b2bSJens Axboe * it would normally disappear when its bi_end_io is run. 1424f9c78b2bSJens Axboe * however, we need it for the unmap, so grab an extra 1425f9c78b2bSJens Axboe * reference to it 1426f9c78b2bSJens Axboe */ 1427f9c78b2bSJens Axboe bio_get(bio); 1428f9c78b2bSJens Axboe 1429f9c78b2bSJens Axboe return bio; 1430f9c78b2bSJens Axboe } 1431f9c78b2bSJens Axboe 1432f9c78b2bSJens Axboe static void __bio_unmap_user(struct bio *bio) 1433f9c78b2bSJens Axboe { 1434f9c78b2bSJens Axboe struct bio_vec *bvec; 1435f9c78b2bSJens Axboe int i; 1436f9c78b2bSJens Axboe 1437f9c78b2bSJens Axboe /* 1438f9c78b2bSJens Axboe * make sure we dirty pages we wrote to 1439f9c78b2bSJens Axboe */ 1440f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1441f9c78b2bSJens Axboe if (bio_data_dir(bio) == READ) 1442f9c78b2bSJens Axboe set_page_dirty_lock(bvec->bv_page); 1443f9c78b2bSJens Axboe 1444f9c78b2bSJens Axboe page_cache_release(bvec->bv_page); 1445f9c78b2bSJens Axboe } 1446f9c78b2bSJens Axboe 1447f9c78b2bSJens Axboe bio_put(bio); 1448f9c78b2bSJens Axboe } 1449f9c78b2bSJens Axboe 1450f9c78b2bSJens Axboe /** 1451f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1452f9c78b2bSJens Axboe * @bio: the bio being unmapped 1453f9c78b2bSJens Axboe * 1454f9c78b2bSJens Axboe * Unmap a bio previously mapped by bio_map_user(). Must be called with 1455f9c78b2bSJens Axboe * a process context. 1456f9c78b2bSJens Axboe * 1457f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1458f9c78b2bSJens Axboe */ 1459f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1460f9c78b2bSJens Axboe { 1461f9c78b2bSJens Axboe __bio_unmap_user(bio); 1462f9c78b2bSJens Axboe bio_put(bio); 1463f9c78b2bSJens Axboe } 1464f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_unmap_user); 1465f9c78b2bSJens Axboe 1466f9c78b2bSJens Axboe static void bio_map_kern_endio(struct bio *bio, int err) 1467f9c78b2bSJens Axboe { 1468f9c78b2bSJens Axboe bio_put(bio); 1469f9c78b2bSJens Axboe } 1470f9c78b2bSJens Axboe 1471f9c78b2bSJens Axboe static struct bio *__bio_map_kern(struct request_queue *q, void *data, 1472f9c78b2bSJens Axboe unsigned int len, gfp_t gfp_mask) 1473f9c78b2bSJens Axboe { 1474f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1475f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1476f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1477f9c78b2bSJens Axboe const int nr_pages = end - start; 1478f9c78b2bSJens Axboe int offset, i; 1479f9c78b2bSJens Axboe struct bio *bio; 1480f9c78b2bSJens Axboe 1481f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1482f9c78b2bSJens Axboe if (!bio) 1483f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1484f9c78b2bSJens Axboe 1485f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1486f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1487f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1488f9c78b2bSJens Axboe 1489f9c78b2bSJens Axboe if (len <= 0) 1490f9c78b2bSJens Axboe break; 1491f9c78b2bSJens Axboe 1492f9c78b2bSJens Axboe if (bytes > len) 1493f9c78b2bSJens Axboe bytes = len; 1494f9c78b2bSJens Axboe 1495f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 1496f9c78b2bSJens Axboe offset) < bytes) 1497f9c78b2bSJens Axboe break; 1498f9c78b2bSJens Axboe 1499f9c78b2bSJens Axboe data += bytes; 1500f9c78b2bSJens Axboe len -= bytes; 1501f9c78b2bSJens Axboe offset = 0; 1502f9c78b2bSJens Axboe } 1503f9c78b2bSJens Axboe 1504f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1505f9c78b2bSJens Axboe return bio; 1506f9c78b2bSJens Axboe } 1507f9c78b2bSJens Axboe 1508f9c78b2bSJens Axboe /** 1509f9c78b2bSJens Axboe * bio_map_kern - map kernel address into bio 1510f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1511f9c78b2bSJens Axboe * @data: pointer to buffer to map 1512f9c78b2bSJens Axboe * @len: length in bytes 1513f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio allocation 1514f9c78b2bSJens Axboe * 1515f9c78b2bSJens Axboe * Map the kernel address into a bio suitable for io to a block 1516f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1517f9c78b2bSJens Axboe */ 1518f9c78b2bSJens Axboe struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 1519f9c78b2bSJens Axboe gfp_t gfp_mask) 1520f9c78b2bSJens Axboe { 1521f9c78b2bSJens Axboe struct bio *bio; 1522f9c78b2bSJens Axboe 1523f9c78b2bSJens Axboe bio = __bio_map_kern(q, data, len, gfp_mask); 1524f9c78b2bSJens Axboe if (IS_ERR(bio)) 1525f9c78b2bSJens Axboe return bio; 1526f9c78b2bSJens Axboe 1527f9c78b2bSJens Axboe if (bio->bi_iter.bi_size == len) 1528f9c78b2bSJens Axboe return bio; 1529f9c78b2bSJens Axboe 1530f9c78b2bSJens Axboe /* 1531f9c78b2bSJens Axboe * Don't support partial mappings. 1532f9c78b2bSJens Axboe */ 1533f9c78b2bSJens Axboe bio_put(bio); 1534f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1535f9c78b2bSJens Axboe } 1536f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_map_kern); 1537f9c78b2bSJens Axboe 1538f9c78b2bSJens Axboe static void bio_copy_kern_endio(struct bio *bio, int err) 1539f9c78b2bSJens Axboe { 1540f9c78b2bSJens Axboe struct bio_vec *bvec; 1541f9c78b2bSJens Axboe const int read = bio_data_dir(bio) == READ; 1542f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 1543f9c78b2bSJens Axboe int i; 1544f9c78b2bSJens Axboe char *p = bmd->sgvecs[0].iov_base; 1545f9c78b2bSJens Axboe 1546f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1547f9c78b2bSJens Axboe char *addr = page_address(bvec->bv_page); 1548f9c78b2bSJens Axboe 1549f9c78b2bSJens Axboe if (read) 1550f9c78b2bSJens Axboe memcpy(p, addr, bvec->bv_len); 1551f9c78b2bSJens Axboe 1552f9c78b2bSJens Axboe __free_page(bvec->bv_page); 1553f9c78b2bSJens Axboe p += bvec->bv_len; 1554f9c78b2bSJens Axboe } 1555f9c78b2bSJens Axboe 1556f9c78b2bSJens Axboe kfree(bmd); 1557f9c78b2bSJens Axboe bio_put(bio); 1558f9c78b2bSJens Axboe } 1559f9c78b2bSJens Axboe 1560f9c78b2bSJens Axboe /** 1561f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1562f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1563f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1564f9c78b2bSJens Axboe * @len: length in bytes 1565f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1566f9c78b2bSJens Axboe * @reading: data direction is READ 1567f9c78b2bSJens Axboe * 1568f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1569f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1570f9c78b2bSJens Axboe */ 1571f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1572f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1573f9c78b2bSJens Axboe { 1574f9c78b2bSJens Axboe struct bio *bio; 1575f9c78b2bSJens Axboe struct bio_vec *bvec; 1576f9c78b2bSJens Axboe int i; 1577f9c78b2bSJens Axboe 1578f9c78b2bSJens Axboe bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask); 1579f9c78b2bSJens Axboe if (IS_ERR(bio)) 1580f9c78b2bSJens Axboe return bio; 1581f9c78b2bSJens Axboe 1582f9c78b2bSJens Axboe if (!reading) { 1583f9c78b2bSJens Axboe void *p = data; 1584f9c78b2bSJens Axboe 1585f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1586f9c78b2bSJens Axboe char *addr = page_address(bvec->bv_page); 1587f9c78b2bSJens Axboe 1588f9c78b2bSJens Axboe memcpy(addr, p, bvec->bv_len); 1589f9c78b2bSJens Axboe p += bvec->bv_len; 1590f9c78b2bSJens Axboe } 1591f9c78b2bSJens Axboe } 1592f9c78b2bSJens Axboe 1593f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 1594f9c78b2bSJens Axboe 1595f9c78b2bSJens Axboe return bio; 1596f9c78b2bSJens Axboe } 1597f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_copy_kern); 1598f9c78b2bSJens Axboe 1599f9c78b2bSJens Axboe /* 1600f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1601f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1602f9c78b2bSJens Axboe * 1603f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1604f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1605f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1606f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1607f9c78b2bSJens Axboe * in process context. 1608f9c78b2bSJens Axboe * 1609f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1610f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1611f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1612f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1613f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1614f9c78b2bSJens Axboe * 1615f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1616f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1617f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1618f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1619f9c78b2bSJens Axboe * pagecache. 1620f9c78b2bSJens Axboe * 1621f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1622f9c78b2bSJens Axboe * deferred bio dirtying paths. 1623f9c78b2bSJens Axboe */ 1624f9c78b2bSJens Axboe 1625f9c78b2bSJens Axboe /* 1626f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1627f9c78b2bSJens Axboe */ 1628f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1629f9c78b2bSJens Axboe { 1630f9c78b2bSJens Axboe struct bio_vec *bvec; 1631f9c78b2bSJens Axboe int i; 1632f9c78b2bSJens Axboe 1633f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1634f9c78b2bSJens Axboe struct page *page = bvec->bv_page; 1635f9c78b2bSJens Axboe 1636f9c78b2bSJens Axboe if (page && !PageCompound(page)) 1637f9c78b2bSJens Axboe set_page_dirty_lock(page); 1638f9c78b2bSJens Axboe } 1639f9c78b2bSJens Axboe } 1640f9c78b2bSJens Axboe 1641f9c78b2bSJens Axboe static void bio_release_pages(struct bio *bio) 1642f9c78b2bSJens Axboe { 1643f9c78b2bSJens Axboe struct bio_vec *bvec; 1644f9c78b2bSJens Axboe int i; 1645f9c78b2bSJens Axboe 1646f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1647f9c78b2bSJens Axboe struct page *page = bvec->bv_page; 1648f9c78b2bSJens Axboe 1649f9c78b2bSJens Axboe if (page) 1650f9c78b2bSJens Axboe put_page(page); 1651f9c78b2bSJens Axboe } 1652f9c78b2bSJens Axboe } 1653f9c78b2bSJens Axboe 1654f9c78b2bSJens Axboe /* 1655f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1656f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1657f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 1658f9c78b2bSJens Axboe * the BIO and the offending pages and re-dirty the pages in process context. 1659f9c78b2bSJens Axboe * 1660f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1661f9c78b2bSJens Axboe * here on. It will run one page_cache_release() against each page and will 1662f9c78b2bSJens Axboe * run one bio_put() against the BIO. 1663f9c78b2bSJens Axboe */ 1664f9c78b2bSJens Axboe 1665f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1666f9c78b2bSJens Axboe 1667f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1668f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1669f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1670f9c78b2bSJens Axboe 1671f9c78b2bSJens Axboe /* 1672f9c78b2bSJens Axboe * This runs in process context 1673f9c78b2bSJens Axboe */ 1674f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1675f9c78b2bSJens Axboe { 1676f9c78b2bSJens Axboe unsigned long flags; 1677f9c78b2bSJens Axboe struct bio *bio; 1678f9c78b2bSJens Axboe 1679f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1680f9c78b2bSJens Axboe bio = bio_dirty_list; 1681f9c78b2bSJens Axboe bio_dirty_list = NULL; 1682f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1683f9c78b2bSJens Axboe 1684f9c78b2bSJens Axboe while (bio) { 1685f9c78b2bSJens Axboe struct bio *next = bio->bi_private; 1686f9c78b2bSJens Axboe 1687f9c78b2bSJens Axboe bio_set_pages_dirty(bio); 1688f9c78b2bSJens Axboe bio_release_pages(bio); 1689f9c78b2bSJens Axboe bio_put(bio); 1690f9c78b2bSJens Axboe bio = next; 1691f9c78b2bSJens Axboe } 1692f9c78b2bSJens Axboe } 1693f9c78b2bSJens Axboe 1694f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1695f9c78b2bSJens Axboe { 1696f9c78b2bSJens Axboe struct bio_vec *bvec; 1697f9c78b2bSJens Axboe int nr_clean_pages = 0; 1698f9c78b2bSJens Axboe int i; 1699f9c78b2bSJens Axboe 1700f9c78b2bSJens Axboe bio_for_each_segment_all(bvec, bio, i) { 1701f9c78b2bSJens Axboe struct page *page = bvec->bv_page; 1702f9c78b2bSJens Axboe 1703f9c78b2bSJens Axboe if (PageDirty(page) || PageCompound(page)) { 1704f9c78b2bSJens Axboe page_cache_release(page); 1705f9c78b2bSJens Axboe bvec->bv_page = NULL; 1706f9c78b2bSJens Axboe } else { 1707f9c78b2bSJens Axboe nr_clean_pages++; 1708f9c78b2bSJens Axboe } 1709f9c78b2bSJens Axboe } 1710f9c78b2bSJens Axboe 1711f9c78b2bSJens Axboe if (nr_clean_pages) { 1712f9c78b2bSJens Axboe unsigned long flags; 1713f9c78b2bSJens Axboe 1714f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1715f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1716f9c78b2bSJens Axboe bio_dirty_list = bio; 1717f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1718f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1719f9c78b2bSJens Axboe } else { 1720f9c78b2bSJens Axboe bio_put(bio); 1721f9c78b2bSJens Axboe } 1722f9c78b2bSJens Axboe } 1723f9c78b2bSJens Axboe 1724f9c78b2bSJens Axboe #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1725f9c78b2bSJens Axboe void bio_flush_dcache_pages(struct bio *bi) 1726f9c78b2bSJens Axboe { 1727f9c78b2bSJens Axboe struct bio_vec bvec; 1728f9c78b2bSJens Axboe struct bvec_iter iter; 1729f9c78b2bSJens Axboe 1730f9c78b2bSJens Axboe bio_for_each_segment(bvec, bi, iter) 1731f9c78b2bSJens Axboe flush_dcache_page(bvec.bv_page); 1732f9c78b2bSJens Axboe } 1733f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_flush_dcache_pages); 1734f9c78b2bSJens Axboe #endif 1735f9c78b2bSJens Axboe 1736f9c78b2bSJens Axboe /** 1737f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1738f9c78b2bSJens Axboe * @bio: bio 1739f9c78b2bSJens Axboe * @error: error, if any 1740f9c78b2bSJens Axboe * 1741f9c78b2bSJens Axboe * Description: 1742f9c78b2bSJens Axboe * bio_endio() will end I/O on the whole bio. bio_endio() is the 1743f9c78b2bSJens Axboe * preferred way to end I/O on a bio, it takes care of clearing 1744f9c78b2bSJens Axboe * BIO_UPTODATE on error. @error is 0 on success, and and one of the 1745f9c78b2bSJens Axboe * established -Exxxx (-EIO, for instance) error values in case 1746f9c78b2bSJens Axboe * something went wrong. No one should call bi_end_io() directly on a 1747f9c78b2bSJens Axboe * bio unless they own it and thus know that it has an end_io 1748f9c78b2bSJens Axboe * function. 1749f9c78b2bSJens Axboe **/ 1750f9c78b2bSJens Axboe void bio_endio(struct bio *bio, int error) 1751f9c78b2bSJens Axboe { 1752f9c78b2bSJens Axboe while (bio) { 1753f9c78b2bSJens Axboe BUG_ON(atomic_read(&bio->bi_remaining) <= 0); 1754f9c78b2bSJens Axboe 1755f9c78b2bSJens Axboe if (error) 1756f9c78b2bSJens Axboe clear_bit(BIO_UPTODATE, &bio->bi_flags); 1757f9c78b2bSJens Axboe else if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) 1758f9c78b2bSJens Axboe error = -EIO; 1759f9c78b2bSJens Axboe 1760f9c78b2bSJens Axboe if (!atomic_dec_and_test(&bio->bi_remaining)) 1761f9c78b2bSJens Axboe return; 1762f9c78b2bSJens Axboe 1763f9c78b2bSJens Axboe /* 1764f9c78b2bSJens Axboe * Need to have a real endio function for chained bios, 1765f9c78b2bSJens Axboe * otherwise various corner cases will break (like stacking 1766f9c78b2bSJens Axboe * block devices that save/restore bi_end_io) - however, we want 1767f9c78b2bSJens Axboe * to avoid unbounded recursion and blowing the stack. Tail call 1768f9c78b2bSJens Axboe * optimization would handle this, but compiling with frame 1769f9c78b2bSJens Axboe * pointers also disables gcc's sibling call optimization. 1770f9c78b2bSJens Axboe */ 1771f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 1772f9c78b2bSJens Axboe struct bio *parent = bio->bi_private; 1773f9c78b2bSJens Axboe bio_put(bio); 1774f9c78b2bSJens Axboe bio = parent; 1775f9c78b2bSJens Axboe } else { 1776f9c78b2bSJens Axboe if (bio->bi_end_io) 1777f9c78b2bSJens Axboe bio->bi_end_io(bio, error); 1778f9c78b2bSJens Axboe bio = NULL; 1779f9c78b2bSJens Axboe } 1780f9c78b2bSJens Axboe } 1781f9c78b2bSJens Axboe } 1782f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1783f9c78b2bSJens Axboe 1784f9c78b2bSJens Axboe /** 1785f9c78b2bSJens Axboe * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining 1786f9c78b2bSJens Axboe * @bio: bio 1787f9c78b2bSJens Axboe * @error: error, if any 1788f9c78b2bSJens Axboe * 1789f9c78b2bSJens Axboe * For code that has saved and restored bi_end_io; thing hard before using this 1790f9c78b2bSJens Axboe * function, probably you should've cloned the entire bio. 1791f9c78b2bSJens Axboe **/ 1792f9c78b2bSJens Axboe void bio_endio_nodec(struct bio *bio, int error) 1793f9c78b2bSJens Axboe { 1794f9c78b2bSJens Axboe atomic_inc(&bio->bi_remaining); 1795f9c78b2bSJens Axboe bio_endio(bio, error); 1796f9c78b2bSJens Axboe } 1797f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio_nodec); 1798f9c78b2bSJens Axboe 1799f9c78b2bSJens Axboe /** 1800f9c78b2bSJens Axboe * bio_split - split a bio 1801f9c78b2bSJens Axboe * @bio: bio to split 1802f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1803f9c78b2bSJens Axboe * @gfp: gfp mask 1804f9c78b2bSJens Axboe * @bs: bio set to allocate from 1805f9c78b2bSJens Axboe * 1806f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1807f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1808f9c78b2bSJens Axboe * 1809f9c78b2bSJens Axboe * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's 1810f9c78b2bSJens Axboe * responsibility to ensure that @bio is not freed before the split. 1811f9c78b2bSJens Axboe */ 1812f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1813f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1814f9c78b2bSJens Axboe { 1815f9c78b2bSJens Axboe struct bio *split = NULL; 1816f9c78b2bSJens Axboe 1817f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1818f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1819f9c78b2bSJens Axboe 1820f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1821f9c78b2bSJens Axboe if (!split) 1822f9c78b2bSJens Axboe return NULL; 1823f9c78b2bSJens Axboe 1824f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1825f9c78b2bSJens Axboe 1826f9c78b2bSJens Axboe if (bio_integrity(split)) 1827f9c78b2bSJens Axboe bio_integrity_trim(split, 0, sectors); 1828f9c78b2bSJens Axboe 1829f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1830f9c78b2bSJens Axboe 1831f9c78b2bSJens Axboe return split; 1832f9c78b2bSJens Axboe } 1833f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1834f9c78b2bSJens Axboe 1835f9c78b2bSJens Axboe /** 1836f9c78b2bSJens Axboe * bio_trim - trim a bio 1837f9c78b2bSJens Axboe * @bio: bio to trim 1838f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1839f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1840f9c78b2bSJens Axboe */ 1841f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1842f9c78b2bSJens Axboe { 1843f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1844f9c78b2bSJens Axboe * the given offset and size. 1845f9c78b2bSJens Axboe */ 1846f9c78b2bSJens Axboe 1847f9c78b2bSJens Axboe size <<= 9; 1848f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1849f9c78b2bSJens Axboe return; 1850f9c78b2bSJens Axboe 1851f9c78b2bSJens Axboe clear_bit(BIO_SEG_VALID, &bio->bi_flags); 1852f9c78b2bSJens Axboe 1853f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1854f9c78b2bSJens Axboe 1855f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1856f9c78b2bSJens Axboe } 1857f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1858f9c78b2bSJens Axboe 1859f9c78b2bSJens Axboe /* 1860f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1861f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1862f9c78b2bSJens Axboe */ 1863f9c78b2bSJens Axboe mempool_t *biovec_create_pool(int pool_entries) 1864f9c78b2bSJens Axboe { 1865f9c78b2bSJens Axboe struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX; 1866f9c78b2bSJens Axboe 1867f9c78b2bSJens Axboe return mempool_create_slab_pool(pool_entries, bp->slab); 1868f9c78b2bSJens Axboe } 1869f9c78b2bSJens Axboe 1870f9c78b2bSJens Axboe void bioset_free(struct bio_set *bs) 1871f9c78b2bSJens Axboe { 1872f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1873f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1874f9c78b2bSJens Axboe 1875f9c78b2bSJens Axboe if (bs->bio_pool) 1876f9c78b2bSJens Axboe mempool_destroy(bs->bio_pool); 1877f9c78b2bSJens Axboe 1878f9c78b2bSJens Axboe if (bs->bvec_pool) 1879f9c78b2bSJens Axboe mempool_destroy(bs->bvec_pool); 1880f9c78b2bSJens Axboe 1881f9c78b2bSJens Axboe bioset_integrity_free(bs); 1882f9c78b2bSJens Axboe bio_put_slab(bs); 1883f9c78b2bSJens Axboe 1884f9c78b2bSJens Axboe kfree(bs); 1885f9c78b2bSJens Axboe } 1886f9c78b2bSJens Axboe EXPORT_SYMBOL(bioset_free); 1887f9c78b2bSJens Axboe 1888f9c78b2bSJens Axboe /** 1889f9c78b2bSJens Axboe * bioset_create - Create a bio_set 1890f9c78b2bSJens Axboe * @pool_size: Number of bio and bio_vecs to cache in the mempool 1891f9c78b2bSJens Axboe * @front_pad: Number of bytes to allocate in front of the returned bio 1892f9c78b2bSJens Axboe * 1893f9c78b2bSJens Axboe * Description: 1894f9c78b2bSJens Axboe * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1895f9c78b2bSJens Axboe * to ask for a number of bytes to be allocated in front of the bio. 1896f9c78b2bSJens Axboe * Front pad allocation is useful for embedding the bio inside 1897f9c78b2bSJens Axboe * another structure, to avoid allocating extra data to go with the bio. 1898f9c78b2bSJens Axboe * Note that the bio must be embedded at the END of that structure always, 1899f9c78b2bSJens Axboe * or things will break badly. 1900f9c78b2bSJens Axboe */ 1901f9c78b2bSJens Axboe struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad) 1902f9c78b2bSJens Axboe { 1903f9c78b2bSJens Axboe unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1904f9c78b2bSJens Axboe struct bio_set *bs; 1905f9c78b2bSJens Axboe 1906f9c78b2bSJens Axboe bs = kzalloc(sizeof(*bs), GFP_KERNEL); 1907f9c78b2bSJens Axboe if (!bs) 1908f9c78b2bSJens Axboe return NULL; 1909f9c78b2bSJens Axboe 1910f9c78b2bSJens Axboe bs->front_pad = front_pad; 1911f9c78b2bSJens Axboe 1912f9c78b2bSJens Axboe spin_lock_init(&bs->rescue_lock); 1913f9c78b2bSJens Axboe bio_list_init(&bs->rescue_list); 1914f9c78b2bSJens Axboe INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1915f9c78b2bSJens Axboe 1916f9c78b2bSJens Axboe bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1917f9c78b2bSJens Axboe if (!bs->bio_slab) { 1918f9c78b2bSJens Axboe kfree(bs); 1919f9c78b2bSJens Axboe return NULL; 1920f9c78b2bSJens Axboe } 1921f9c78b2bSJens Axboe 1922f9c78b2bSJens Axboe bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab); 1923f9c78b2bSJens Axboe if (!bs->bio_pool) 1924f9c78b2bSJens Axboe goto bad; 1925f9c78b2bSJens Axboe 1926f9c78b2bSJens Axboe bs->bvec_pool = biovec_create_pool(pool_size); 1927f9c78b2bSJens Axboe if (!bs->bvec_pool) 1928f9c78b2bSJens Axboe goto bad; 1929f9c78b2bSJens Axboe 1930f9c78b2bSJens Axboe bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1931f9c78b2bSJens Axboe if (!bs->rescue_workqueue) 1932f9c78b2bSJens Axboe goto bad; 1933f9c78b2bSJens Axboe 1934f9c78b2bSJens Axboe return bs; 1935f9c78b2bSJens Axboe bad: 1936f9c78b2bSJens Axboe bioset_free(bs); 1937f9c78b2bSJens Axboe return NULL; 1938f9c78b2bSJens Axboe } 1939f9c78b2bSJens Axboe EXPORT_SYMBOL(bioset_create); 1940f9c78b2bSJens Axboe 1941f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 1942f9c78b2bSJens Axboe /** 1943f9c78b2bSJens Axboe * bio_associate_current - associate a bio with %current 1944f9c78b2bSJens Axboe * @bio: target bio 1945f9c78b2bSJens Axboe * 1946f9c78b2bSJens Axboe * Associate @bio with %current if it hasn't been associated yet. Block 1947f9c78b2bSJens Axboe * layer will treat @bio as if it were issued by %current no matter which 1948f9c78b2bSJens Axboe * task actually issues it. 1949f9c78b2bSJens Axboe * 1950f9c78b2bSJens Axboe * This function takes an extra reference of @task's io_context and blkcg 1951f9c78b2bSJens Axboe * which will be put when @bio is released. The caller must own @bio, 1952f9c78b2bSJens Axboe * ensure %current->io_context exists, and is responsible for synchronizing 1953f9c78b2bSJens Axboe * calls to this function. 1954f9c78b2bSJens Axboe */ 1955f9c78b2bSJens Axboe int bio_associate_current(struct bio *bio) 1956f9c78b2bSJens Axboe { 1957f9c78b2bSJens Axboe struct io_context *ioc; 1958f9c78b2bSJens Axboe struct cgroup_subsys_state *css; 1959f9c78b2bSJens Axboe 1960f9c78b2bSJens Axboe if (bio->bi_ioc) 1961f9c78b2bSJens Axboe return -EBUSY; 1962f9c78b2bSJens Axboe 1963f9c78b2bSJens Axboe ioc = current->io_context; 1964f9c78b2bSJens Axboe if (!ioc) 1965f9c78b2bSJens Axboe return -ENOENT; 1966f9c78b2bSJens Axboe 1967f9c78b2bSJens Axboe /* acquire active ref on @ioc and associate */ 1968f9c78b2bSJens Axboe get_io_context_active(ioc); 1969f9c78b2bSJens Axboe bio->bi_ioc = ioc; 1970f9c78b2bSJens Axboe 1971f9c78b2bSJens Axboe /* associate blkcg if exists */ 1972f9c78b2bSJens Axboe rcu_read_lock(); 1973f9c78b2bSJens Axboe css = task_css(current, blkio_cgrp_id); 1974f9c78b2bSJens Axboe if (css && css_tryget(css)) 1975f9c78b2bSJens Axboe bio->bi_css = css; 1976f9c78b2bSJens Axboe rcu_read_unlock(); 1977f9c78b2bSJens Axboe 1978f9c78b2bSJens Axboe return 0; 1979f9c78b2bSJens Axboe } 1980f9c78b2bSJens Axboe 1981f9c78b2bSJens Axboe /** 1982f9c78b2bSJens Axboe * bio_disassociate_task - undo bio_associate_current() 1983f9c78b2bSJens Axboe * @bio: target bio 1984f9c78b2bSJens Axboe */ 1985f9c78b2bSJens Axboe void bio_disassociate_task(struct bio *bio) 1986f9c78b2bSJens Axboe { 1987f9c78b2bSJens Axboe if (bio->bi_ioc) { 1988f9c78b2bSJens Axboe put_io_context(bio->bi_ioc); 1989f9c78b2bSJens Axboe bio->bi_ioc = NULL; 1990f9c78b2bSJens Axboe } 1991f9c78b2bSJens Axboe if (bio->bi_css) { 1992f9c78b2bSJens Axboe css_put(bio->bi_css); 1993f9c78b2bSJens Axboe bio->bi_css = NULL; 1994f9c78b2bSJens Axboe } 1995f9c78b2bSJens Axboe } 1996f9c78b2bSJens Axboe 1997f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 1998f9c78b2bSJens Axboe 1999f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2000f9c78b2bSJens Axboe { 2001f9c78b2bSJens Axboe int i; 2002f9c78b2bSJens Axboe 2003f9c78b2bSJens Axboe for (i = 0; i < BIOVEC_NR_POOLS; i++) { 2004f9c78b2bSJens Axboe int size; 2005f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2006f9c78b2bSJens Axboe 2007f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2008f9c78b2bSJens Axboe bvs->slab = NULL; 2009f9c78b2bSJens Axboe continue; 2010f9c78b2bSJens Axboe } 2011f9c78b2bSJens Axboe 2012f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2013f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2014f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2015f9c78b2bSJens Axboe } 2016f9c78b2bSJens Axboe } 2017f9c78b2bSJens Axboe 2018f9c78b2bSJens Axboe static int __init init_bio(void) 2019f9c78b2bSJens Axboe { 2020f9c78b2bSJens Axboe bio_slab_max = 2; 2021f9c78b2bSJens Axboe bio_slab_nr = 0; 2022f9c78b2bSJens Axboe bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL); 2023f9c78b2bSJens Axboe if (!bio_slabs) 2024f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2025f9c78b2bSJens Axboe 2026f9c78b2bSJens Axboe bio_integrity_init(); 2027f9c78b2bSJens Axboe biovec_init_slabs(); 2028f9c78b2bSJens Axboe 2029f9c78b2bSJens Axboe fs_bio_set = bioset_create(BIO_POOL_SIZE, 0); 2030f9c78b2bSJens Axboe if (!fs_bio_set) 2031f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2032f9c78b2bSJens Axboe 2033f9c78b2bSJens Axboe if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE)) 2034f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2035f9c78b2bSJens Axboe 2036f9c78b2bSJens Axboe return 0; 2037f9c78b2bSJens Axboe } 2038f9c78b2bSJens Axboe subsys_initcall(init_bio); 2039