18c16567dSChristoph Hellwig // SPDX-License-Identifier: GPL-2.0 2f9c78b2bSJens Axboe /* 3f9c78b2bSJens Axboe * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk> 4f9c78b2bSJens Axboe */ 5f9c78b2bSJens Axboe #include <linux/mm.h> 6f9c78b2bSJens Axboe #include <linux/swap.h> 7f9c78b2bSJens Axboe #include <linux/bio.h> 8f9c78b2bSJens Axboe #include <linux/blkdev.h> 9f9c78b2bSJens Axboe #include <linux/uio.h> 10f9c78b2bSJens Axboe #include <linux/iocontext.h> 11f9c78b2bSJens Axboe #include <linux/slab.h> 12f9c78b2bSJens Axboe #include <linux/init.h> 13f9c78b2bSJens Axboe #include <linux/kernel.h> 14f9c78b2bSJens Axboe #include <linux/export.h> 15f9c78b2bSJens Axboe #include <linux/mempool.h> 16f9c78b2bSJens Axboe #include <linux/workqueue.h> 17f9c78b2bSJens Axboe #include <linux/cgroup.h> 1808e18eabSJosef Bacik #include <linux/blk-cgroup.h> 19f9c78b2bSJens Axboe 20f9c78b2bSJens Axboe #include <trace/events/block.h> 219e234eeaSShaohua Li #include "blk.h" 2267b42d0bSJosef Bacik #include "blk-rq-qos.h" 23f9c78b2bSJens Axboe 24f9c78b2bSJens Axboe /* 25f9c78b2bSJens Axboe * Test patch to inline a certain number of bi_io_vec's inside the bio 26f9c78b2bSJens Axboe * itself, to shrink a bio data allocation from two mempool calls to one 27f9c78b2bSJens Axboe */ 28f9c78b2bSJens Axboe #define BIO_INLINE_VECS 4 29f9c78b2bSJens Axboe 30f9c78b2bSJens Axboe /* 31f9c78b2bSJens Axboe * if you change this list, also change bvec_alloc or things will 32f9c78b2bSJens Axboe * break badly! cannot be bigger than what you can fit into an 33f9c78b2bSJens Axboe * unsigned short 34f9c78b2bSJens Axboe */ 35bd5c4facSMikulas Patocka #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 36ed996a52SChristoph Hellwig static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 37bd5c4facSMikulas Patocka BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 38f9c78b2bSJens Axboe }; 39f9c78b2bSJens Axboe #undef BV 40f9c78b2bSJens Axboe 41f9c78b2bSJens Axboe /* 42f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 43f9c78b2bSJens Axboe * IO code that does not need private memory pools. 44f9c78b2bSJens Axboe */ 45f4f8154aSKent Overstreet struct bio_set fs_bio_set; 46f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 47f9c78b2bSJens Axboe 48f9c78b2bSJens Axboe /* 49f9c78b2bSJens Axboe * Our slab pool management 50f9c78b2bSJens Axboe */ 51f9c78b2bSJens Axboe struct bio_slab { 52f9c78b2bSJens Axboe struct kmem_cache *slab; 53f9c78b2bSJens Axboe unsigned int slab_ref; 54f9c78b2bSJens Axboe unsigned int slab_size; 55f9c78b2bSJens Axboe char name[8]; 56f9c78b2bSJens Axboe }; 57f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 58f9c78b2bSJens Axboe static struct bio_slab *bio_slabs; 59f9c78b2bSJens Axboe static unsigned int bio_slab_nr, bio_slab_max; 60f9c78b2bSJens Axboe 61f9c78b2bSJens Axboe static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 62f9c78b2bSJens Axboe { 63f9c78b2bSJens Axboe unsigned int sz = sizeof(struct bio) + extra_size; 64f9c78b2bSJens Axboe struct kmem_cache *slab = NULL; 65f9c78b2bSJens Axboe struct bio_slab *bslab, *new_bio_slabs; 66f9c78b2bSJens Axboe unsigned int new_bio_slab_max; 67f9c78b2bSJens Axboe unsigned int i, entry = -1; 68f9c78b2bSJens Axboe 69f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 70f9c78b2bSJens Axboe 71f9c78b2bSJens Axboe i = 0; 72f9c78b2bSJens Axboe while (i < bio_slab_nr) { 73f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 74f9c78b2bSJens Axboe 75f9c78b2bSJens Axboe if (!bslab->slab && entry == -1) 76f9c78b2bSJens Axboe entry = i; 77f9c78b2bSJens Axboe else if (bslab->slab_size == sz) { 78f9c78b2bSJens Axboe slab = bslab->slab; 79f9c78b2bSJens Axboe bslab->slab_ref++; 80f9c78b2bSJens Axboe break; 81f9c78b2bSJens Axboe } 82f9c78b2bSJens Axboe i++; 83f9c78b2bSJens Axboe } 84f9c78b2bSJens Axboe 85f9c78b2bSJens Axboe if (slab) 86f9c78b2bSJens Axboe goto out_unlock; 87f9c78b2bSJens Axboe 88f9c78b2bSJens Axboe if (bio_slab_nr == bio_slab_max && entry == -1) { 89f9c78b2bSJens Axboe new_bio_slab_max = bio_slab_max << 1; 90f9c78b2bSJens Axboe new_bio_slabs = krealloc(bio_slabs, 91f9c78b2bSJens Axboe new_bio_slab_max * sizeof(struct bio_slab), 92f9c78b2bSJens Axboe GFP_KERNEL); 93f9c78b2bSJens Axboe if (!new_bio_slabs) 94f9c78b2bSJens Axboe goto out_unlock; 95f9c78b2bSJens Axboe bio_slab_max = new_bio_slab_max; 96f9c78b2bSJens Axboe bio_slabs = new_bio_slabs; 97f9c78b2bSJens Axboe } 98f9c78b2bSJens Axboe if (entry == -1) 99f9c78b2bSJens Axboe entry = bio_slab_nr++; 100f9c78b2bSJens Axboe 101f9c78b2bSJens Axboe bslab = &bio_slabs[entry]; 102f9c78b2bSJens Axboe 103f9c78b2bSJens Axboe snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 1046a241483SMikulas Patocka slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 1056a241483SMikulas Patocka SLAB_HWCACHE_ALIGN, NULL); 106f9c78b2bSJens Axboe if (!slab) 107f9c78b2bSJens Axboe goto out_unlock; 108f9c78b2bSJens Axboe 109f9c78b2bSJens Axboe bslab->slab = slab; 110f9c78b2bSJens Axboe bslab->slab_ref = 1; 111f9c78b2bSJens Axboe bslab->slab_size = sz; 112f9c78b2bSJens Axboe out_unlock: 113f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 114f9c78b2bSJens Axboe return slab; 115f9c78b2bSJens Axboe } 116f9c78b2bSJens Axboe 117f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 118f9c78b2bSJens Axboe { 119f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 120f9c78b2bSJens Axboe unsigned int i; 121f9c78b2bSJens Axboe 122f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 123f9c78b2bSJens Axboe 124f9c78b2bSJens Axboe for (i = 0; i < bio_slab_nr; i++) { 125f9c78b2bSJens Axboe if (bs->bio_slab == bio_slabs[i].slab) { 126f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 127f9c78b2bSJens Axboe break; 128f9c78b2bSJens Axboe } 129f9c78b2bSJens Axboe } 130f9c78b2bSJens Axboe 131f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 132f9c78b2bSJens Axboe goto out; 133f9c78b2bSJens Axboe 134f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 135f9c78b2bSJens Axboe 136f9c78b2bSJens Axboe if (--bslab->slab_ref) 137f9c78b2bSJens Axboe goto out; 138f9c78b2bSJens Axboe 139f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 140f9c78b2bSJens Axboe bslab->slab = NULL; 141f9c78b2bSJens Axboe 142f9c78b2bSJens Axboe out: 143f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 144f9c78b2bSJens Axboe } 145f9c78b2bSJens Axboe 146f9c78b2bSJens Axboe unsigned int bvec_nr_vecs(unsigned short idx) 147f9c78b2bSJens Axboe { 148d6c02a9bSGreg Edwards return bvec_slabs[--idx].nr_vecs; 149f9c78b2bSJens Axboe } 150f9c78b2bSJens Axboe 151f9c78b2bSJens Axboe void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 152f9c78b2bSJens Axboe { 153ed996a52SChristoph Hellwig if (!idx) 154ed996a52SChristoph Hellwig return; 155ed996a52SChristoph Hellwig idx--; 156f9c78b2bSJens Axboe 157ed996a52SChristoph Hellwig BIO_BUG_ON(idx >= BVEC_POOL_NR); 158ed996a52SChristoph Hellwig 159ed996a52SChristoph Hellwig if (idx == BVEC_POOL_MAX) { 160f9c78b2bSJens Axboe mempool_free(bv, pool); 161ed996a52SChristoph Hellwig } else { 162f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + idx; 163f9c78b2bSJens Axboe 164f9c78b2bSJens Axboe kmem_cache_free(bvs->slab, bv); 165f9c78b2bSJens Axboe } 166f9c78b2bSJens Axboe } 167f9c78b2bSJens Axboe 168f9c78b2bSJens Axboe struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 169f9c78b2bSJens Axboe mempool_t *pool) 170f9c78b2bSJens Axboe { 171f9c78b2bSJens Axboe struct bio_vec *bvl; 172f9c78b2bSJens Axboe 173f9c78b2bSJens Axboe /* 174f9c78b2bSJens Axboe * see comment near bvec_array define! 175f9c78b2bSJens Axboe */ 176f9c78b2bSJens Axboe switch (nr) { 177f9c78b2bSJens Axboe case 1: 178f9c78b2bSJens Axboe *idx = 0; 179f9c78b2bSJens Axboe break; 180f9c78b2bSJens Axboe case 2 ... 4: 181f9c78b2bSJens Axboe *idx = 1; 182f9c78b2bSJens Axboe break; 183f9c78b2bSJens Axboe case 5 ... 16: 184f9c78b2bSJens Axboe *idx = 2; 185f9c78b2bSJens Axboe break; 186f9c78b2bSJens Axboe case 17 ... 64: 187f9c78b2bSJens Axboe *idx = 3; 188f9c78b2bSJens Axboe break; 189f9c78b2bSJens Axboe case 65 ... 128: 190f9c78b2bSJens Axboe *idx = 4; 191f9c78b2bSJens Axboe break; 192f9c78b2bSJens Axboe case 129 ... BIO_MAX_PAGES: 193f9c78b2bSJens Axboe *idx = 5; 194f9c78b2bSJens Axboe break; 195f9c78b2bSJens Axboe default: 196f9c78b2bSJens Axboe return NULL; 197f9c78b2bSJens Axboe } 198f9c78b2bSJens Axboe 199f9c78b2bSJens Axboe /* 200f9c78b2bSJens Axboe * idx now points to the pool we want to allocate from. only the 201f9c78b2bSJens Axboe * 1-vec entry pool is mempool backed. 202f9c78b2bSJens Axboe */ 203ed996a52SChristoph Hellwig if (*idx == BVEC_POOL_MAX) { 204f9c78b2bSJens Axboe fallback: 205f9c78b2bSJens Axboe bvl = mempool_alloc(pool, gfp_mask); 206f9c78b2bSJens Axboe } else { 207f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + *idx; 208d0164adcSMel Gorman gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 209f9c78b2bSJens Axboe 210f9c78b2bSJens Axboe /* 211f9c78b2bSJens Axboe * Make this allocation restricted and don't dump info on 212f9c78b2bSJens Axboe * allocation failures, since we'll fallback to the mempool 213f9c78b2bSJens Axboe * in case of failure. 214f9c78b2bSJens Axboe */ 215f9c78b2bSJens Axboe __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 216f9c78b2bSJens Axboe 217f9c78b2bSJens Axboe /* 218d0164adcSMel Gorman * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 219f9c78b2bSJens Axboe * is set, retry with the 1-entry mempool 220f9c78b2bSJens Axboe */ 221f9c78b2bSJens Axboe bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 222d0164adcSMel Gorman if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 223ed996a52SChristoph Hellwig *idx = BVEC_POOL_MAX; 224f9c78b2bSJens Axboe goto fallback; 225f9c78b2bSJens Axboe } 226f9c78b2bSJens Axboe } 227f9c78b2bSJens Axboe 228ed996a52SChristoph Hellwig (*idx)++; 229f9c78b2bSJens Axboe return bvl; 230f9c78b2bSJens Axboe } 231f9c78b2bSJens Axboe 2329ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 233f9c78b2bSJens Axboe { 2346f70fb66SDennis Zhou bio_disassociate_blkg(bio); 235f9c78b2bSJens Axboe } 2369ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 237f9c78b2bSJens Axboe 238f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 239f9c78b2bSJens Axboe { 240f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 241f9c78b2bSJens Axboe void *p; 242f9c78b2bSJens Axboe 2439ae3b3f5SJens Axboe bio_uninit(bio); 244f9c78b2bSJens Axboe 245f9c78b2bSJens Axboe if (bs) { 2468aa6ba2fSKent Overstreet bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 247f9c78b2bSJens Axboe 248f9c78b2bSJens Axboe /* 249f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 250f9c78b2bSJens Axboe */ 251f9c78b2bSJens Axboe p = bio; 252f9c78b2bSJens Axboe p -= bs->front_pad; 253f9c78b2bSJens Axboe 2548aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 255f9c78b2bSJens Axboe } else { 256f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 257f9c78b2bSJens Axboe kfree(bio); 258f9c78b2bSJens Axboe } 259f9c78b2bSJens Axboe } 260f9c78b2bSJens Axboe 2619ae3b3f5SJens Axboe /* 2629ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2639ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2649ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2659ae3b3f5SJens Axboe */ 2663a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2673a83f467SMing Lei unsigned short max_vecs) 268f9c78b2bSJens Axboe { 269f9c78b2bSJens Axboe memset(bio, 0, sizeof(*bio)); 270c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 271dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2723a83f467SMing Lei 2733a83f467SMing Lei bio->bi_io_vec = table; 2743a83f467SMing Lei bio->bi_max_vecs = max_vecs; 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 2929ae3b3f5SJens Axboe bio_uninit(bio); 293f9c78b2bSJens Axboe 294f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 2954246a0b6SChristoph Hellwig bio->bi_flags = flags; 296c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 297f9c78b2bSJens Axboe } 298f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 299f9c78b2bSJens Axboe 30038f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 301f9c78b2bSJens Axboe { 3024246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3034246a0b6SChristoph Hellwig 3044e4cbee9SChristoph Hellwig if (!parent->bi_status) 3054e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 306f9c78b2bSJens Axboe bio_put(bio); 30738f8baaeSChristoph Hellwig return parent; 30838f8baaeSChristoph Hellwig } 30938f8baaeSChristoph Hellwig 31038f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 31138f8baaeSChristoph Hellwig { 31238f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 313f9c78b2bSJens Axboe } 314f9c78b2bSJens Axboe 315f9c78b2bSJens Axboe /** 316f9c78b2bSJens Axboe * bio_chain - chain bio completions 317f9c78b2bSJens Axboe * @bio: the target bio 318f9c78b2bSJens Axboe * @parent: the @bio's parent bio 319f9c78b2bSJens Axboe * 320f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 321f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 322f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 323f9c78b2bSJens Axboe * 324f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 325f9c78b2bSJens Axboe */ 326f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 327f9c78b2bSJens Axboe { 328f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 329f9c78b2bSJens Axboe 330f9c78b2bSJens Axboe bio->bi_private = parent; 331f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 332c4cf5261SJens Axboe bio_inc_remaining(parent); 333f9c78b2bSJens Axboe } 334f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 335f9c78b2bSJens Axboe 336f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 337f9c78b2bSJens Axboe { 338f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 339f9c78b2bSJens Axboe struct bio *bio; 340f9c78b2bSJens Axboe 341f9c78b2bSJens Axboe while (1) { 342f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 343f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 344f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 345f9c78b2bSJens Axboe 346f9c78b2bSJens Axboe if (!bio) 347f9c78b2bSJens Axboe break; 348f9c78b2bSJens Axboe 349f9c78b2bSJens Axboe generic_make_request(bio); 350f9c78b2bSJens Axboe } 351f9c78b2bSJens Axboe } 352f9c78b2bSJens Axboe 353f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 354f9c78b2bSJens Axboe { 355f9c78b2bSJens Axboe struct bio_list punt, nopunt; 356f9c78b2bSJens Axboe struct bio *bio; 357f9c78b2bSJens Axboe 35847e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 35947e0fb46SNeilBrown return; 360f9c78b2bSJens Axboe /* 361f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 362f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 363f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 364f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 365f9c78b2bSJens Axboe * our own rescuer would be bad. 366f9c78b2bSJens Axboe * 367f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 368f9c78b2bSJens Axboe * remove from the middle of the list: 369f9c78b2bSJens Axboe */ 370f9c78b2bSJens Axboe 371f9c78b2bSJens Axboe bio_list_init(&punt); 372f9c78b2bSJens Axboe bio_list_init(&nopunt); 373f9c78b2bSJens Axboe 374f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 375f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 376f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 377f9c78b2bSJens Axboe 378f5fe1b51SNeilBrown bio_list_init(&nopunt); 379f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 380f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 381f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 382f9c78b2bSJens Axboe 383f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 384f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 385f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 386f9c78b2bSJens Axboe 387f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 388f9c78b2bSJens Axboe } 389f9c78b2bSJens Axboe 390f9c78b2bSJens Axboe /** 391f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 392519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 393f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 394f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 395f9c78b2bSJens Axboe * 396f9c78b2bSJens Axboe * Description: 397f9c78b2bSJens Axboe * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 398f9c78b2bSJens Axboe * backed by the @bs's mempool. 399f9c78b2bSJens Axboe * 400d0164adcSMel Gorman * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 401d0164adcSMel Gorman * always be able to allocate a bio. This is due to the mempool guarantees. 402d0164adcSMel Gorman * To make this work, callers must never allocate more than 1 bio at a time 403d0164adcSMel Gorman * from this pool. Callers that need to allocate more than 1 bio must always 404d0164adcSMel Gorman * submit the previously allocated bio for IO before attempting to allocate 405d0164adcSMel Gorman * a new one. Failure to do so can cause deadlocks under memory pressure. 406f9c78b2bSJens Axboe * 407f9c78b2bSJens Axboe * Note that when running under generic_make_request() (i.e. any block 408f9c78b2bSJens Axboe * driver), bios are not submitted until after you return - see the code in 409f9c78b2bSJens Axboe * generic_make_request() that converts recursion into iteration, to prevent 410f9c78b2bSJens Axboe * stack overflows. 411f9c78b2bSJens Axboe * 412f9c78b2bSJens Axboe * This would normally mean allocating multiple bios under 413f9c78b2bSJens Axboe * generic_make_request() would be susceptible to deadlocks, but we have 414f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 415f9c78b2bSJens Axboe * thread. 416f9c78b2bSJens Axboe * 417f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 418f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 419f9c78b2bSJens Axboe * generic_make_request() should be avoided - instead, use bio_set's front_pad 420f9c78b2bSJens Axboe * for per bio allocations. 421f9c78b2bSJens Axboe * 422f9c78b2bSJens Axboe * RETURNS: 423f9c78b2bSJens Axboe * Pointer to new bio on success, NULL on failure. 424f9c78b2bSJens Axboe */ 4257a88fa19SDan Carpenter struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 4267a88fa19SDan Carpenter struct bio_set *bs) 427f9c78b2bSJens Axboe { 428f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 429f9c78b2bSJens Axboe unsigned front_pad; 430f9c78b2bSJens Axboe unsigned inline_vecs; 431f9c78b2bSJens Axboe struct bio_vec *bvl = NULL; 432f9c78b2bSJens Axboe struct bio *bio; 433f9c78b2bSJens Axboe void *p; 434f9c78b2bSJens Axboe 435f9c78b2bSJens Axboe if (!bs) { 436f9c78b2bSJens Axboe if (nr_iovecs > UIO_MAXIOV) 437f9c78b2bSJens Axboe return NULL; 438f9c78b2bSJens Axboe 439f9c78b2bSJens Axboe p = kmalloc(sizeof(struct bio) + 440f9c78b2bSJens Axboe nr_iovecs * sizeof(struct bio_vec), 441f9c78b2bSJens Axboe gfp_mask); 442f9c78b2bSJens Axboe front_pad = 0; 443f9c78b2bSJens Axboe inline_vecs = nr_iovecs; 444f9c78b2bSJens Axboe } else { 445d8f429e1SJunichi Nomura /* should not use nobvec bioset for nr_iovecs > 0 */ 4468aa6ba2fSKent Overstreet if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 4478aa6ba2fSKent Overstreet nr_iovecs > 0)) 448d8f429e1SJunichi Nomura return NULL; 449f9c78b2bSJens Axboe /* 450f9c78b2bSJens Axboe * generic_make_request() converts recursion to iteration; this 451f9c78b2bSJens Axboe * means if we're running beneath it, any bios we allocate and 452f9c78b2bSJens Axboe * submit will not be submitted (and thus freed) until after we 453f9c78b2bSJens Axboe * return. 454f9c78b2bSJens Axboe * 455f9c78b2bSJens Axboe * This exposes us to a potential deadlock if we allocate 456f9c78b2bSJens Axboe * multiple bios from the same bio_set() while running 457f9c78b2bSJens Axboe * underneath generic_make_request(). If we were to allocate 458f9c78b2bSJens Axboe * multiple bios (say a stacking block driver that was splitting 459f9c78b2bSJens Axboe * bios), we would deadlock if we exhausted the mempool's 460f9c78b2bSJens Axboe * reserve. 461f9c78b2bSJens Axboe * 462f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 463f9c78b2bSJens Axboe * workqueue per bio_set. If we go to allocate and there are 464f9c78b2bSJens Axboe * bios on current->bio_list, we first try the allocation 465d0164adcSMel Gorman * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 466d0164adcSMel Gorman * bios we would be blocking to the rescuer workqueue before 467d0164adcSMel Gorman * we retry with the original gfp_flags. 468f9c78b2bSJens Axboe */ 469f9c78b2bSJens Axboe 470f5fe1b51SNeilBrown if (current->bio_list && 471f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 47247e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 47347e0fb46SNeilBrown bs->rescue_workqueue) 474d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 475f9c78b2bSJens Axboe 4768aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 477f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 478f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 479f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4808aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 481f9c78b2bSJens Axboe } 482f9c78b2bSJens Axboe 483f9c78b2bSJens Axboe front_pad = bs->front_pad; 484f9c78b2bSJens Axboe inline_vecs = BIO_INLINE_VECS; 485f9c78b2bSJens Axboe } 486f9c78b2bSJens Axboe 487f9c78b2bSJens Axboe if (unlikely(!p)) 488f9c78b2bSJens Axboe return NULL; 489f9c78b2bSJens Axboe 490f9c78b2bSJens Axboe bio = p + front_pad; 4913a83f467SMing Lei bio_init(bio, NULL, 0); 492f9c78b2bSJens Axboe 493f9c78b2bSJens Axboe if (nr_iovecs > inline_vecs) { 494ed996a52SChristoph Hellwig unsigned long idx = 0; 495ed996a52SChristoph Hellwig 4968aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 497f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 498f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 499f9c78b2bSJens Axboe gfp_mask = saved_gfp; 5008aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 501f9c78b2bSJens Axboe } 502f9c78b2bSJens Axboe 503f9c78b2bSJens Axboe if (unlikely(!bvl)) 504f9c78b2bSJens Axboe goto err_free; 505f9c78b2bSJens Axboe 506ed996a52SChristoph Hellwig bio->bi_flags |= idx << BVEC_POOL_OFFSET; 507f9c78b2bSJens Axboe } else if (nr_iovecs) { 508f9c78b2bSJens Axboe bvl = bio->bi_inline_vecs; 509f9c78b2bSJens Axboe } 510f9c78b2bSJens Axboe 511f9c78b2bSJens Axboe bio->bi_pool = bs; 512f9c78b2bSJens Axboe bio->bi_max_vecs = nr_iovecs; 513f9c78b2bSJens Axboe bio->bi_io_vec = bvl; 514f9c78b2bSJens Axboe return bio; 515f9c78b2bSJens Axboe 516f9c78b2bSJens Axboe err_free: 5178aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 518f9c78b2bSJens Axboe return NULL; 519f9c78b2bSJens Axboe } 520f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 521f9c78b2bSJens Axboe 52238a72dacSKent Overstreet void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 523f9c78b2bSJens Axboe { 524f9c78b2bSJens Axboe unsigned long flags; 525f9c78b2bSJens Axboe struct bio_vec bv; 526f9c78b2bSJens Axboe struct bvec_iter iter; 527f9c78b2bSJens Axboe 52838a72dacSKent Overstreet __bio_for_each_segment(bv, bio, iter, start) { 529f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 530f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 531f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 532f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 533f9c78b2bSJens Axboe } 534f9c78b2bSJens Axboe } 53538a72dacSKent Overstreet EXPORT_SYMBOL(zero_fill_bio_iter); 536f9c78b2bSJens Axboe 537f9c78b2bSJens Axboe /** 538f9c78b2bSJens Axboe * bio_put - release a reference to a bio 539f9c78b2bSJens Axboe * @bio: bio to release reference to 540f9c78b2bSJens Axboe * 541f9c78b2bSJens Axboe * Description: 542f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 5439b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 544f9c78b2bSJens Axboe **/ 545f9c78b2bSJens Axboe void bio_put(struct bio *bio) 546f9c78b2bSJens Axboe { 547dac56212SJens Axboe if (!bio_flagged(bio, BIO_REFFED)) 548dac56212SJens Axboe bio_free(bio); 549dac56212SJens Axboe else { 550dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 551f9c78b2bSJens Axboe 552f9c78b2bSJens Axboe /* 553f9c78b2bSJens Axboe * last put frees it 554f9c78b2bSJens Axboe */ 555dac56212SJens Axboe if (atomic_dec_and_test(&bio->__bi_cnt)) 556f9c78b2bSJens Axboe bio_free(bio); 557f9c78b2bSJens Axboe } 558dac56212SJens Axboe } 559f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 560f9c78b2bSJens Axboe 561f9c78b2bSJens Axboe /** 562f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 563f9c78b2bSJens Axboe * @bio: destination bio 564f9c78b2bSJens Axboe * @bio_src: bio to clone 565f9c78b2bSJens Axboe * 566f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 567f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 568f9c78b2bSJens Axboe * bio will be one. 569f9c78b2bSJens Axboe * 570f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 571f9c78b2bSJens Axboe */ 572f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 573f9c78b2bSJens Axboe { 574ed996a52SChristoph Hellwig BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 575f9c78b2bSJens Axboe 576f9c78b2bSJens Axboe /* 57774d46992SChristoph Hellwig * most users will be overriding ->bi_disk with a new target, 578f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 579f9c78b2bSJens Axboe */ 58074d46992SChristoph Hellwig bio->bi_disk = bio_src->bi_disk; 58162530ed8SMichael Lyle bio->bi_partno = bio_src->bi_partno; 582b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 583111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 584111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 5851eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 586ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 587cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 588f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 589f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 59020bd723eSPaolo Valente 591db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 592e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 593f9c78b2bSJens Axboe } 594f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 595f9c78b2bSJens Axboe 596f9c78b2bSJens Axboe /** 597f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 598f9c78b2bSJens Axboe * @bio: bio to clone 599f9c78b2bSJens Axboe * @gfp_mask: allocation priority 600f9c78b2bSJens Axboe * @bs: bio_set to allocate from 601f9c78b2bSJens Axboe * 602f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 603f9c78b2bSJens Axboe */ 604f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 605f9c78b2bSJens Axboe { 606f9c78b2bSJens Axboe struct bio *b; 607f9c78b2bSJens Axboe 608f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 609f9c78b2bSJens Axboe if (!b) 610f9c78b2bSJens Axboe return NULL; 611f9c78b2bSJens Axboe 612f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 613f9c78b2bSJens Axboe 614f9c78b2bSJens Axboe if (bio_integrity(bio)) { 615f9c78b2bSJens Axboe int ret; 616f9c78b2bSJens Axboe 617f9c78b2bSJens Axboe ret = bio_integrity_clone(b, bio, gfp_mask); 618f9c78b2bSJens Axboe 619f9c78b2bSJens Axboe if (ret < 0) { 620f9c78b2bSJens Axboe bio_put(b); 621f9c78b2bSJens Axboe return NULL; 622f9c78b2bSJens Axboe } 623f9c78b2bSJens Axboe } 624f9c78b2bSJens Axboe 625f9c78b2bSJens Axboe return b; 626f9c78b2bSJens Axboe } 627f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 628f9c78b2bSJens Axboe 6295919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 6305919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 631ff896738SChristoph Hellwig bool *same_page) 6325919482eSMing Lei { 6335919482eSMing Lei phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 6345919482eSMing Lei bv->bv_offset + bv->bv_len - 1; 6355919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 6365919482eSMing Lei 6375919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 6385919482eSMing Lei return false; 6395919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 6405919482eSMing Lei return false; 64152d52d1cSChristoph Hellwig 642ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 643ff896738SChristoph Hellwig if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) 6445919482eSMing Lei return false; 6455919482eSMing Lei return true; 6465919482eSMing Lei } 6475919482eSMing Lei 648489fbbcbSMing Lei /* 649489fbbcbSMing Lei * Check if the @page can be added to the current segment(@bv), and make 650489fbbcbSMing Lei * sure to call it only if page_is_mergeable(@bv, @page) is true 651489fbbcbSMing Lei */ 652489fbbcbSMing Lei static bool can_add_page_to_seg(struct request_queue *q, 653489fbbcbSMing Lei struct bio_vec *bv, struct page *page, unsigned len, 654489fbbcbSMing Lei unsigned offset) 655489fbbcbSMing Lei { 656489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 657489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 658489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 659489fbbcbSMing Lei 660489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 661489fbbcbSMing Lei return false; 662489fbbcbSMing Lei 663489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 664489fbbcbSMing Lei return false; 665489fbbcbSMing Lei 666489fbbcbSMing Lei return true; 667489fbbcbSMing Lei } 668489fbbcbSMing Lei 669f4595875SShaohua Li /** 67019047087SMing Lei * __bio_add_pc_page - attempt to add page to passthrough bio 671c66a14d0SKent Overstreet * @q: the target queue 672c66a14d0SKent Overstreet * @bio: destination bio 673c66a14d0SKent Overstreet * @page: page to add 674c66a14d0SKent Overstreet * @len: vec entry length 675c66a14d0SKent Overstreet * @offset: vec entry offset 67619047087SMing Lei * @put_same_page: put the page if it is same with last added page 677f9c78b2bSJens Axboe * 678c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This can fail for a 679c66a14d0SKent Overstreet * number of reasons, such as the bio being full or target block device 680c66a14d0SKent Overstreet * limitations. The target block device must allow bio's up to PAGE_SIZE, 681c66a14d0SKent Overstreet * so it is always possible to add a single page to an empty bio. 682c66a14d0SKent Overstreet * 6835a8ce240SMing Lei * This should only be used by passthrough bios. 684f9c78b2bSJens Axboe */ 6854713839dSChristoph Hellwig static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, 68619047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 68719047087SMing Lei bool put_same_page) 688f9c78b2bSJens Axboe { 689f9c78b2bSJens Axboe struct bio_vec *bvec; 690ff896738SChristoph Hellwig bool same_page = false; 691f9c78b2bSJens Axboe 692f9c78b2bSJens Axboe /* 693f9c78b2bSJens Axboe * cloned bio must not modify vec list 694f9c78b2bSJens Axboe */ 695f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 696f9c78b2bSJens Axboe return 0; 697f9c78b2bSJens Axboe 698c66a14d0SKent Overstreet if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 699f9c78b2bSJens Axboe return 0; 700f9c78b2bSJens Axboe 701f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 7025a8ce240SMing Lei bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 703f9c78b2bSJens Axboe 7045a8ce240SMing Lei if (page == bvec->bv_page && 7055a8ce240SMing Lei offset == bvec->bv_offset + bvec->bv_len) { 70619047087SMing Lei if (put_same_page) 70719047087SMing Lei put_page(page); 7085a8ce240SMing Lei bvec->bv_len += len; 709f9c78b2bSJens Axboe goto done; 710f9c78b2bSJens Axboe } 71166cb45aaSJens Axboe 71266cb45aaSJens Axboe /* 71366cb45aaSJens Axboe * If the queue doesn't support SG gaps and adding this 71466cb45aaSJens Axboe * offset would create a gap, disallow it. 71566cb45aaSJens Axboe */ 7165a8ce240SMing Lei if (bvec_gap_to_prev(q, bvec, offset)) 71766cb45aaSJens Axboe return 0; 718489fbbcbSMing Lei 719ff896738SChristoph Hellwig if (page_is_mergeable(bvec, page, len, offset, &same_page) && 720dcdca753SChristoph Hellwig can_add_page_to_seg(q, bvec, page, len, offset)) { 721dcdca753SChristoph Hellwig bvec->bv_len += len; 722dcdca753SChristoph Hellwig goto done; 723dcdca753SChristoph Hellwig } 724f9c78b2bSJens Axboe } 725f9c78b2bSJens Axboe 72679d08f89SMing Lei if (bio_full(bio, len)) 727f9c78b2bSJens Axboe return 0; 728f9c78b2bSJens Axboe 72914ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 730489fbbcbSMing Lei return 0; 731489fbbcbSMing Lei 732f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 733f9c78b2bSJens Axboe bvec->bv_page = page; 734f9c78b2bSJens Axboe bvec->bv_len = len; 735f9c78b2bSJens Axboe bvec->bv_offset = offset; 736fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 737f9c78b2bSJens Axboe done: 738dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 739f9c78b2bSJens Axboe return len; 740f9c78b2bSJens Axboe } 74119047087SMing Lei 74219047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 74319047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 74419047087SMing Lei { 74519047087SMing Lei return __bio_add_pc_page(q, bio, page, len, offset, false); 74619047087SMing Lei } 747f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 748f9c78b2bSJens Axboe 749f9c78b2bSJens Axboe /** 7500aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 7510aa69fd3SChristoph Hellwig * @bio: destination bio 752551879a4SMing Lei * @page: start page to add 7530aa69fd3SChristoph Hellwig * @len: length of the data to add 754551879a4SMing Lei * @off: offset of the data relative to @page 755ff896738SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 7560aa69fd3SChristoph Hellwig * 7570aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 7580aa69fd3SChristoph Hellwig * a useful optimisation for file systems with a block size smaller than the 7590aa69fd3SChristoph Hellwig * page size. 7600aa69fd3SChristoph Hellwig * 761551879a4SMing Lei * Warn if (@len, @off) crosses pages in case that @same_page is true. 762551879a4SMing Lei * 7630aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 7640aa69fd3SChristoph Hellwig */ 7650aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 766ff896738SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 7670aa69fd3SChristoph Hellwig { 7680aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 7690aa69fd3SChristoph Hellwig return false; 7700aa69fd3SChristoph Hellwig 7710aa69fd3SChristoph Hellwig if (bio->bi_vcnt > 0) { 7720aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 7730aa69fd3SChristoph Hellwig 7745919482eSMing Lei if (page_is_mergeable(bv, page, len, off, same_page)) { 7750aa69fd3SChristoph Hellwig bv->bv_len += len; 7760aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 7770aa69fd3SChristoph Hellwig return true; 7780aa69fd3SChristoph Hellwig } 7795919482eSMing Lei } 7800aa69fd3SChristoph Hellwig return false; 7810aa69fd3SChristoph Hellwig } 7820aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 7830aa69fd3SChristoph Hellwig 7840aa69fd3SChristoph Hellwig /** 785551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 7860aa69fd3SChristoph Hellwig * @bio: destination bio 787551879a4SMing Lei * @page: start page to add 788551879a4SMing Lei * @len: length of the data to add, may cross pages 789551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 7900aa69fd3SChristoph Hellwig * 7910aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 7920aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 7930aa69fd3SChristoph Hellwig */ 7940aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 7950aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 7960aa69fd3SChristoph Hellwig { 7970aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 7980aa69fd3SChristoph Hellwig 7990aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 80079d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 8010aa69fd3SChristoph Hellwig 8020aa69fd3SChristoph Hellwig bv->bv_page = page; 8030aa69fd3SChristoph Hellwig bv->bv_offset = off; 8040aa69fd3SChristoph Hellwig bv->bv_len = len; 8050aa69fd3SChristoph Hellwig 8060aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8070aa69fd3SChristoph Hellwig bio->bi_vcnt++; 8080aa69fd3SChristoph Hellwig } 8090aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 8100aa69fd3SChristoph Hellwig 8110aa69fd3SChristoph Hellwig /** 812551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 813f9c78b2bSJens Axboe * @bio: destination bio 814551879a4SMing Lei * @page: start page to add 815551879a4SMing Lei * @len: vec entry length, may cross pages 816551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 817f9c78b2bSJens Axboe * 818551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 819c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 820f9c78b2bSJens Axboe */ 821c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 822c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 823f9c78b2bSJens Axboe { 824ff896738SChristoph Hellwig bool same_page = false; 825ff896738SChristoph Hellwig 826ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 82779d08f89SMing Lei if (bio_full(bio, len)) 828c66a14d0SKent Overstreet return 0; 8290aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 830c66a14d0SKent Overstreet } 831c66a14d0SKent Overstreet return len; 832f9c78b2bSJens Axboe } 833f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 834f9c78b2bSJens Axboe 835d241a95fSChristoph Hellwig void bio_release_pages(struct bio *bio, bool mark_dirty) 8367321ecbfSChristoph Hellwig { 8377321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 8387321ecbfSChristoph Hellwig struct bio_vec *bvec; 8397321ecbfSChristoph Hellwig 840b2d0d991SChristoph Hellwig if (bio_flagged(bio, BIO_NO_PAGE_REF)) 841b2d0d991SChristoph Hellwig return; 842b2d0d991SChristoph Hellwig 843d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 844d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 845d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 8467321ecbfSChristoph Hellwig put_page(bvec->bv_page); 8477321ecbfSChristoph Hellwig } 848d241a95fSChristoph Hellwig } 8497321ecbfSChristoph Hellwig 8506d0c48aeSJens Axboe static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 8516d0c48aeSJens Axboe { 8526d0c48aeSJens Axboe const struct bio_vec *bv = iter->bvec; 8536d0c48aeSJens Axboe unsigned int len; 8546d0c48aeSJens Axboe size_t size; 8556d0c48aeSJens Axboe 8566d0c48aeSJens Axboe if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 8576d0c48aeSJens Axboe return -EINVAL; 8586d0c48aeSJens Axboe 8596d0c48aeSJens Axboe len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 8606d0c48aeSJens Axboe size = bio_add_page(bio, bv->bv_page, len, 8616d0c48aeSJens Axboe bv->bv_offset + iter->iov_offset); 862a10584c3SChristoph Hellwig if (unlikely(size != len)) 863a10584c3SChristoph Hellwig return -EINVAL; 8646d0c48aeSJens Axboe iov_iter_advance(iter, size); 8656d0c48aeSJens Axboe return 0; 8666d0c48aeSJens Axboe } 8676d0c48aeSJens Axboe 868576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 869576ed913SChristoph Hellwig 8702cefe4dbSKent Overstreet /** 87117d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 8722cefe4dbSKent Overstreet * @bio: bio to add pages to 8732cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 8742cefe4dbSKent Overstreet * 87517d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 8762cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 87717d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 87817d51b10SMartin Wilck * the next non-empty segment of the iov iterator. 8792cefe4dbSKent Overstreet */ 88017d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 8812cefe4dbSKent Overstreet { 882576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 883576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 8842cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 8852cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 88645691804SChristoph Hellwig bool same_page = false; 887576ed913SChristoph Hellwig ssize_t size, left; 888576ed913SChristoph Hellwig unsigned len, i; 889b403ea24SMartin Wilck size_t offset; 890576ed913SChristoph Hellwig 891576ed913SChristoph Hellwig /* 892576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 893576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 894576ed913SChristoph Hellwig * without overwriting the temporary page array. 895576ed913SChristoph Hellwig */ 896576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 897576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 8982cefe4dbSKent Overstreet 8992cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 9002cefe4dbSKent Overstreet if (unlikely(size <= 0)) 9012cefe4dbSKent Overstreet return size ? size : -EFAULT; 9022cefe4dbSKent Overstreet 903576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 904576ed913SChristoph Hellwig struct page *page = pages[i]; 9052cefe4dbSKent Overstreet 906576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 90745691804SChristoph Hellwig 90845691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 90945691804SChristoph Hellwig if (same_page) 91045691804SChristoph Hellwig put_page(page); 91145691804SChristoph Hellwig } else { 91279d08f89SMing Lei if (WARN_ON_ONCE(bio_full(bio, len))) 913576ed913SChristoph Hellwig return -EINVAL; 91445691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 91545691804SChristoph Hellwig } 916576ed913SChristoph Hellwig offset = 0; 9172cefe4dbSKent Overstreet } 9182cefe4dbSKent Overstreet 9192cefe4dbSKent Overstreet iov_iter_advance(iter, size); 9202cefe4dbSKent Overstreet return 0; 9212cefe4dbSKent Overstreet } 92217d51b10SMartin Wilck 92317d51b10SMartin Wilck /** 9246d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 92517d51b10SMartin Wilck * @bio: bio to add pages to 9266d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 92717d51b10SMartin Wilck * 9286d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 9296d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 9306d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 931399254aaSJens Axboe * pages. If we're adding kernel pages, and the caller told us it's safe to 932399254aaSJens Axboe * do so, we just have to add the pages to the bio directly. We don't grab an 933399254aaSJens Axboe * extra reference to those pages (the user should already have that), and we 934399254aaSJens Axboe * don't put the page on IO completion. The caller needs to check if the bio is 935399254aaSJens Axboe * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 936399254aaSJens Axboe * released. 9376d0c48aeSJens Axboe * 93817d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 9396d0c48aeSJens Axboe * fit into the bio, or are requested in *iter, whatever is smaller. If 9406d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 9416d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 94217d51b10SMartin Wilck */ 94317d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 94417d51b10SMartin Wilck { 9456d0c48aeSJens Axboe const bool is_bvec = iov_iter_is_bvec(iter); 94614eacf12SChristoph Hellwig int ret; 94714eacf12SChristoph Hellwig 94814eacf12SChristoph Hellwig if (WARN_ON_ONCE(bio->bi_vcnt)) 94914eacf12SChristoph Hellwig return -EINVAL; 95017d51b10SMartin Wilck 95117d51b10SMartin Wilck do { 9526d0c48aeSJens Axboe if (is_bvec) 9536d0c48aeSJens Axboe ret = __bio_iov_bvec_add_pages(bio, iter); 9546d0c48aeSJens Axboe else 9556d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 95679d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 95717d51b10SMartin Wilck 958b6207430SChristoph Hellwig if (is_bvec) 9597321ecbfSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 96014eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 96117d51b10SMartin Wilck } 9622cefe4dbSKent Overstreet 9634246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 964f9c78b2bSJens Axboe { 96565e53aabSChristoph Hellwig complete(bio->bi_private); 966f9c78b2bSJens Axboe } 967f9c78b2bSJens Axboe 968f9c78b2bSJens Axboe /** 969f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 970f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 971f9c78b2bSJens Axboe * 972f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 973f9c78b2bSJens Axboe * bio_endio() on failure. 9743d289d68SJan Kara * 9753d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 9763d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 9773d289d68SJan Kara * on his own. 978f9c78b2bSJens Axboe */ 9794e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 980f9c78b2bSJens Axboe { 981e319e1fbSByungchul Park DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 982f9c78b2bSJens Axboe 98365e53aabSChristoph Hellwig bio->bi_private = &done; 984f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 9851eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 9864e49ea4aSMike Christie submit_bio(bio); 98765e53aabSChristoph Hellwig wait_for_completion_io(&done); 988f9c78b2bSJens Axboe 98965e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 990f9c78b2bSJens Axboe } 991f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 992f9c78b2bSJens Axboe 993f9c78b2bSJens Axboe /** 994f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 995f9c78b2bSJens Axboe * @bio: bio to advance 996f9c78b2bSJens Axboe * @bytes: number of bytes to complete 997f9c78b2bSJens Axboe * 998f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 999f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1000f9c78b2bSJens Axboe * be updated on the last bvec as well. 1001f9c78b2bSJens Axboe * 1002f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 1003f9c78b2bSJens Axboe */ 1004f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 1005f9c78b2bSJens Axboe { 1006f9c78b2bSJens Axboe if (bio_integrity(bio)) 1007f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1008f9c78b2bSJens Axboe 1009f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1010f9c78b2bSJens Axboe } 1011f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 1012f9c78b2bSJens Axboe 101345db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 101445db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1015f9c78b2bSJens Axboe { 1016f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1017f9c78b2bSJens Axboe void *src_p, *dst_p; 1018f9c78b2bSJens Axboe unsigned bytes; 1019f9c78b2bSJens Axboe 102045db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 102145db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 102245db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 102345db54d5SKent Overstreet 102445db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 102545db54d5SKent Overstreet 102645db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 102745db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 102845db54d5SKent Overstreet 102945db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 103045db54d5SKent Overstreet src_p + src_bv.bv_offset, 103145db54d5SKent Overstreet bytes); 103245db54d5SKent Overstreet 103345db54d5SKent Overstreet kunmap_atomic(dst_p); 103445db54d5SKent Overstreet kunmap_atomic(src_p); 103545db54d5SKent Overstreet 10366e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 10376e6e811dSKent Overstreet 103845db54d5SKent Overstreet bio_advance_iter(src, src_iter, bytes); 103945db54d5SKent Overstreet bio_advance_iter(dst, dst_iter, bytes); 104045db54d5SKent Overstreet } 104145db54d5SKent Overstreet } 104245db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 104345db54d5SKent Overstreet 104445db54d5SKent Overstreet /** 104545db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 104645db54d5SKent Overstreet * @src: source bio 104745db54d5SKent Overstreet * @dst: destination bio 104845db54d5SKent Overstreet * 104945db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 105045db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 105145db54d5SKent Overstreet */ 105245db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 105345db54d5SKent Overstreet { 105445db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 105545db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 105645db54d5SKent Overstreet 105745db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 105845db54d5SKent Overstreet } 105945db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 106045db54d5SKent Overstreet 106145db54d5SKent Overstreet /** 106245db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 106345db54d5SKent Overstreet * another 106445db54d5SKent Overstreet * @src: source bio list 106545db54d5SKent Overstreet * @dst: destination bio list 106645db54d5SKent Overstreet * 106745db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 106845db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 106945db54d5SKent Overstreet * bios). 107045db54d5SKent Overstreet */ 107145db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 107245db54d5SKent Overstreet { 107345db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 107445db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 107545db54d5SKent Overstreet 1076f9c78b2bSJens Axboe while (1) { 1077f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1078f9c78b2bSJens Axboe src = src->bi_next; 1079f9c78b2bSJens Axboe if (!src) 1080f9c78b2bSJens Axboe break; 1081f9c78b2bSJens Axboe 1082f9c78b2bSJens Axboe src_iter = src->bi_iter; 1083f9c78b2bSJens Axboe } 1084f9c78b2bSJens Axboe 1085f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1086f9c78b2bSJens Axboe dst = dst->bi_next; 1087f9c78b2bSJens Axboe if (!dst) 1088f9c78b2bSJens Axboe break; 1089f9c78b2bSJens Axboe 1090f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1091f9c78b2bSJens Axboe } 1092f9c78b2bSJens Axboe 109345db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1094f9c78b2bSJens Axboe } 1095f9c78b2bSJens Axboe } 109645db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1097f9c78b2bSJens Axboe 1098f9c78b2bSJens Axboe struct bio_map_data { 1099f9c78b2bSJens Axboe int is_our_pages; 110026e49cfcSKent Overstreet struct iov_iter iter; 110126e49cfcSKent Overstreet struct iovec iov[]; 1102f9c78b2bSJens Axboe }; 1103f9c78b2bSJens Axboe 11040e5b935dSAl Viro static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1105f9c78b2bSJens Axboe gfp_t gfp_mask) 1106f9c78b2bSJens Axboe { 11070e5b935dSAl Viro struct bio_map_data *bmd; 11080e5b935dSAl Viro if (data->nr_segs > UIO_MAXIOV) 1109f9c78b2bSJens Axboe return NULL; 1110f9c78b2bSJens Axboe 1111f1f8f292SGustavo A. R. Silva bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); 11120e5b935dSAl Viro if (!bmd) 11130e5b935dSAl Viro return NULL; 11140e5b935dSAl Viro memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 11150e5b935dSAl Viro bmd->iter = *data; 11160e5b935dSAl Viro bmd->iter.iov = bmd->iov; 11170e5b935dSAl Viro return bmd; 1118f9c78b2bSJens Axboe } 1119f9c78b2bSJens Axboe 11209124d3feSDongsu Park /** 11219124d3feSDongsu Park * bio_copy_from_iter - copy all pages from iov_iter to bio 11229124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as destination 11239124d3feSDongsu Park * @iter: iov_iter as source 11249124d3feSDongsu Park * 11259124d3feSDongsu Park * Copy all pages from iov_iter to bio. 11269124d3feSDongsu Park * Returns 0 on success, or error on failure. 11279124d3feSDongsu Park */ 112898a09d61SAl Viro static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1129f9c78b2bSJens Axboe { 1130f9c78b2bSJens Axboe struct bio_vec *bvec; 11316dc4f100SMing Lei struct bvec_iter_all iter_all; 1132f9c78b2bSJens Axboe 11332b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 11349124d3feSDongsu Park ssize_t ret; 1135f9c78b2bSJens Axboe 11369124d3feSDongsu Park ret = copy_page_from_iter(bvec->bv_page, 11379124d3feSDongsu Park bvec->bv_offset, 11389124d3feSDongsu Park bvec->bv_len, 113998a09d61SAl Viro iter); 1140f9c78b2bSJens Axboe 114198a09d61SAl Viro if (!iov_iter_count(iter)) 11429124d3feSDongsu Park break; 1143f9c78b2bSJens Axboe 11449124d3feSDongsu Park if (ret < bvec->bv_len) 11459124d3feSDongsu Park return -EFAULT; 1146f9c78b2bSJens Axboe } 1147f9c78b2bSJens Axboe 11489124d3feSDongsu Park return 0; 1149f9c78b2bSJens Axboe } 1150f9c78b2bSJens Axboe 11519124d3feSDongsu Park /** 11529124d3feSDongsu Park * bio_copy_to_iter - copy all pages from bio to iov_iter 11539124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as source 11549124d3feSDongsu Park * @iter: iov_iter as destination 11559124d3feSDongsu Park * 11569124d3feSDongsu Park * Copy all pages from bio to iov_iter. 11579124d3feSDongsu Park * Returns 0 on success, or error on failure. 11589124d3feSDongsu Park */ 11599124d3feSDongsu Park static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 11609124d3feSDongsu Park { 11619124d3feSDongsu Park struct bio_vec *bvec; 11626dc4f100SMing Lei struct bvec_iter_all iter_all; 11639124d3feSDongsu Park 11642b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 11659124d3feSDongsu Park ssize_t ret; 11669124d3feSDongsu Park 11679124d3feSDongsu Park ret = copy_page_to_iter(bvec->bv_page, 11689124d3feSDongsu Park bvec->bv_offset, 11699124d3feSDongsu Park bvec->bv_len, 11709124d3feSDongsu Park &iter); 11719124d3feSDongsu Park 11729124d3feSDongsu Park if (!iov_iter_count(&iter)) 11739124d3feSDongsu Park break; 11749124d3feSDongsu Park 11759124d3feSDongsu Park if (ret < bvec->bv_len) 11769124d3feSDongsu Park return -EFAULT; 11779124d3feSDongsu Park } 11789124d3feSDongsu Park 11799124d3feSDongsu Park return 0; 1180f9c78b2bSJens Axboe } 1181f9c78b2bSJens Axboe 1182491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 11831dfa0f68SChristoph Hellwig { 11841dfa0f68SChristoph Hellwig struct bio_vec *bvec; 11856dc4f100SMing Lei struct bvec_iter_all iter_all; 11861dfa0f68SChristoph Hellwig 11872b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 11881dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 11891dfa0f68SChristoph Hellwig } 1190491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 11911dfa0f68SChristoph Hellwig 1192f9c78b2bSJens Axboe /** 1193f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1194f9c78b2bSJens Axboe * @bio: bio being terminated 1195f9c78b2bSJens Axboe * 1196ddad8dd0SChristoph Hellwig * Free pages allocated from bio_copy_user_iov() and write back data 1197f9c78b2bSJens Axboe * to user space in case of a read. 1198f9c78b2bSJens Axboe */ 1199f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1200f9c78b2bSJens Axboe { 1201f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 12021dfa0f68SChristoph Hellwig int ret = 0; 1203f9c78b2bSJens Axboe 1204f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1205f9c78b2bSJens Axboe /* 1206f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 12072d99b55dSHannes Reinecke * don't copy into a random user address space, just free 12082d99b55dSHannes Reinecke * and return -EINTR so user space doesn't expect any data. 1209f9c78b2bSJens Axboe */ 12102d99b55dSHannes Reinecke if (!current->mm) 12112d99b55dSHannes Reinecke ret = -EINTR; 12122d99b55dSHannes Reinecke else if (bio_data_dir(bio) == READ) 12139124d3feSDongsu Park ret = bio_copy_to_iter(bio, bmd->iter); 12141dfa0f68SChristoph Hellwig if (bmd->is_our_pages) 12151dfa0f68SChristoph Hellwig bio_free_pages(bio); 1216f9c78b2bSJens Axboe } 1217f9c78b2bSJens Axboe kfree(bmd); 1218f9c78b2bSJens Axboe bio_put(bio); 1219f9c78b2bSJens Axboe return ret; 1220f9c78b2bSJens Axboe } 1221f9c78b2bSJens Axboe 1222f9c78b2bSJens Axboe /** 1223f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1224f9c78b2bSJens Axboe * @q: destination block queue 1225f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 122626e49cfcSKent Overstreet * @iter: iovec iterator 1227f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1228f9c78b2bSJens Axboe * 1229f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1230f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1231f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1232f9c78b2bSJens Axboe */ 1233f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1234f9c78b2bSJens Axboe struct rq_map_data *map_data, 1235e81cef5dSAl Viro struct iov_iter *iter, 123626e49cfcSKent Overstreet gfp_t gfp_mask) 1237f9c78b2bSJens Axboe { 1238f9c78b2bSJens Axboe struct bio_map_data *bmd; 1239f9c78b2bSJens Axboe struct page *page; 1240f9c78b2bSJens Axboe struct bio *bio; 1241d16d44ebSAl Viro int i = 0, ret; 1242d16d44ebSAl Viro int nr_pages; 124326e49cfcSKent Overstreet unsigned int len = iter->count; 1244bd5ceceaSGeliang Tang unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1245f9c78b2bSJens Axboe 12460e5b935dSAl Viro bmd = bio_alloc_map_data(iter, gfp_mask); 1247f9c78b2bSJens Axboe if (!bmd) 1248f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1249f9c78b2bSJens Axboe 125026e49cfcSKent Overstreet /* 125126e49cfcSKent Overstreet * We need to do a deep copy of the iov_iter including the iovecs. 125226e49cfcSKent Overstreet * The caller provided iov might point to an on-stack or otherwise 125326e49cfcSKent Overstreet * shortlived one. 125426e49cfcSKent Overstreet */ 125526e49cfcSKent Overstreet bmd->is_our_pages = map_data ? 0 : 1; 125626e49cfcSKent Overstreet 1257d16d44ebSAl Viro nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1258d16d44ebSAl Viro if (nr_pages > BIO_MAX_PAGES) 1259d16d44ebSAl Viro nr_pages = BIO_MAX_PAGES; 1260f9c78b2bSJens Axboe 1261f9c78b2bSJens Axboe ret = -ENOMEM; 1262f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1263f9c78b2bSJens Axboe if (!bio) 1264f9c78b2bSJens Axboe goto out_bmd; 1265f9c78b2bSJens Axboe 1266f9c78b2bSJens Axboe ret = 0; 1267f9c78b2bSJens Axboe 1268f9c78b2bSJens Axboe if (map_data) { 1269f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1270f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1271f9c78b2bSJens Axboe } 1272f9c78b2bSJens Axboe while (len) { 1273f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1274f9c78b2bSJens Axboe 1275f9c78b2bSJens Axboe bytes -= offset; 1276f9c78b2bSJens Axboe 1277f9c78b2bSJens Axboe if (bytes > len) 1278f9c78b2bSJens Axboe bytes = len; 1279f9c78b2bSJens Axboe 1280f9c78b2bSJens Axboe if (map_data) { 1281f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1282f9c78b2bSJens Axboe ret = -ENOMEM; 1283f9c78b2bSJens Axboe break; 1284f9c78b2bSJens Axboe } 1285f9c78b2bSJens Axboe 1286f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1287f9c78b2bSJens Axboe page += (i % nr_pages); 1288f9c78b2bSJens Axboe 1289f9c78b2bSJens Axboe i++; 1290f9c78b2bSJens Axboe } else { 1291f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1292f9c78b2bSJens Axboe if (!page) { 1293f9c78b2bSJens Axboe ret = -ENOMEM; 1294f9c78b2bSJens Axboe break; 1295f9c78b2bSJens Axboe } 1296f9c78b2bSJens Axboe } 1297f9c78b2bSJens Axboe 1298a3761c3cSJérôme Glisse if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { 1299a3761c3cSJérôme Glisse if (!map_data) 1300a3761c3cSJérôme Glisse __free_page(page); 1301f9c78b2bSJens Axboe break; 1302a3761c3cSJérôme Glisse } 1303f9c78b2bSJens Axboe 1304f9c78b2bSJens Axboe len -= bytes; 1305f9c78b2bSJens Axboe offset = 0; 1306f9c78b2bSJens Axboe } 1307f9c78b2bSJens Axboe 1308f9c78b2bSJens Axboe if (ret) 1309f9c78b2bSJens Axboe goto cleanup; 1310f9c78b2bSJens Axboe 13112884d0beSAl Viro if (map_data) 13122884d0beSAl Viro map_data->offset += bio->bi_iter.bi_size; 13132884d0beSAl Viro 1314f9c78b2bSJens Axboe /* 1315f9c78b2bSJens Axboe * success 1316f9c78b2bSJens Axboe */ 131700e23707SDavid Howells if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1318f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 131998a09d61SAl Viro ret = bio_copy_from_iter(bio, iter); 1320f9c78b2bSJens Axboe if (ret) 1321f9c78b2bSJens Axboe goto cleanup; 132298a09d61SAl Viro } else { 1323f55adad6SKeith Busch if (bmd->is_our_pages) 1324f3587d76SKeith Busch zero_fill_bio(bio); 1325e81cef5dSAl Viro iov_iter_advance(iter, bio->bi_iter.bi_size); 1326f9c78b2bSJens Axboe } 1327f9c78b2bSJens Axboe 132826e49cfcSKent Overstreet bio->bi_private = bmd; 13292884d0beSAl Viro if (map_data && map_data->null_mapped) 13302884d0beSAl Viro bio_set_flag(bio, BIO_NULL_MAPPED); 1331f9c78b2bSJens Axboe return bio; 1332f9c78b2bSJens Axboe cleanup: 1333f9c78b2bSJens Axboe if (!map_data) 13341dfa0f68SChristoph Hellwig bio_free_pages(bio); 1335f9c78b2bSJens Axboe bio_put(bio); 1336f9c78b2bSJens Axboe out_bmd: 1337f9c78b2bSJens Axboe kfree(bmd); 1338f9c78b2bSJens Axboe return ERR_PTR(ret); 1339f9c78b2bSJens Axboe } 1340f9c78b2bSJens Axboe 134137f19e57SChristoph Hellwig /** 134237f19e57SChristoph Hellwig * bio_map_user_iov - map user iovec into bio 134337f19e57SChristoph Hellwig * @q: the struct request_queue for the bio 134437f19e57SChristoph Hellwig * @iter: iovec iterator 134537f19e57SChristoph Hellwig * @gfp_mask: memory allocation flags 134637f19e57SChristoph Hellwig * 134737f19e57SChristoph Hellwig * Map the user space address into a bio suitable for io to a block 134837f19e57SChristoph Hellwig * device. Returns an error pointer in case of error. 134937f19e57SChristoph Hellwig */ 135037f19e57SChristoph Hellwig struct bio *bio_map_user_iov(struct request_queue *q, 1351e81cef5dSAl Viro struct iov_iter *iter, 135226e49cfcSKent Overstreet gfp_t gfp_mask) 1353f9c78b2bSJens Axboe { 135426e49cfcSKent Overstreet int j; 1355f9c78b2bSJens Axboe struct bio *bio; 1356076098e5SAl Viro int ret; 1357f9c78b2bSJens Axboe 1358b282cc76SAl Viro if (!iov_iter_count(iter)) 1359f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1360f9c78b2bSJens Axboe 1361b282cc76SAl Viro bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1362f9c78b2bSJens Axboe if (!bio) 1363f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1364f9c78b2bSJens Axboe 13650a0f1513SAl Viro while (iov_iter_count(iter)) { 1366629e42bcSAl Viro struct page **pages; 1367076098e5SAl Viro ssize_t bytes; 1368076098e5SAl Viro size_t offs, added = 0; 1369076098e5SAl Viro int npages; 1370f9c78b2bSJens Axboe 13710a0f1513SAl Viro bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1372076098e5SAl Viro if (unlikely(bytes <= 0)) { 1373076098e5SAl Viro ret = bytes ? bytes : -EFAULT; 1374f9c78b2bSJens Axboe goto out_unmap; 1375f9c78b2bSJens Axboe } 1376f9c78b2bSJens Axboe 1377076098e5SAl Viro npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1378076098e5SAl Viro 137998f0bc99SAl Viro if (unlikely(offs & queue_dma_alignment(q))) { 138098f0bc99SAl Viro ret = -EINVAL; 138198f0bc99SAl Viro j = 0; 138298f0bc99SAl Viro } else { 1383629e42bcSAl Viro for (j = 0; j < npages; j++) { 138498f0bc99SAl Viro struct page *page = pages[j]; 1385076098e5SAl Viro unsigned int n = PAGE_SIZE - offs; 1386f9c78b2bSJens Axboe 1387076098e5SAl Viro if (n > bytes) 1388076098e5SAl Viro n = bytes; 1389f9c78b2bSJens Axboe 139019047087SMing Lei if (!__bio_add_pc_page(q, bio, page, n, offs, 139119047087SMing Lei true)) 1392f9c78b2bSJens Axboe break; 1393f9c78b2bSJens Axboe 1394076098e5SAl Viro added += n; 1395076098e5SAl Viro bytes -= n; 1396076098e5SAl Viro offs = 0; 1397f9c78b2bSJens Axboe } 13980a0f1513SAl Viro iov_iter_advance(iter, added); 139998f0bc99SAl Viro } 1400f9c78b2bSJens Axboe /* 1401f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1402f9c78b2bSJens Axboe */ 1403629e42bcSAl Viro while (j < npages) 140409cbfeafSKirill A. Shutemov put_page(pages[j++]); 1405629e42bcSAl Viro kvfree(pages); 1406e2e115d1SAl Viro /* couldn't stuff something into bio? */ 1407e2e115d1SAl Viro if (bytes) 1408e2e115d1SAl Viro break; 1409f9c78b2bSJens Axboe } 1410f9c78b2bSJens Axboe 1411b7c44ed9SJens Axboe bio_set_flag(bio, BIO_USER_MAPPED); 141237f19e57SChristoph Hellwig 141337f19e57SChristoph Hellwig /* 14145fad1b64SBart Van Assche * subtle -- if bio_map_user_iov() ended up bouncing a bio, 141537f19e57SChristoph Hellwig * it would normally disappear when its bi_end_io is run. 141637f19e57SChristoph Hellwig * however, we need it for the unmap, so grab an extra 141737f19e57SChristoph Hellwig * reference to it 141837f19e57SChristoph Hellwig */ 141937f19e57SChristoph Hellwig bio_get(bio); 1420f9c78b2bSJens Axboe return bio; 1421f9c78b2bSJens Axboe 1422f9c78b2bSJens Axboe out_unmap: 1423506e0798SChristoph Hellwig bio_release_pages(bio, false); 1424f9c78b2bSJens Axboe bio_put(bio); 1425f9c78b2bSJens Axboe return ERR_PTR(ret); 1426f9c78b2bSJens Axboe } 1427f9c78b2bSJens Axboe 1428f9c78b2bSJens Axboe /** 1429f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1430f9c78b2bSJens Axboe * @bio: the bio being unmapped 1431f9c78b2bSJens Axboe * 14325fad1b64SBart Van Assche * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 14335fad1b64SBart Van Assche * process context. 1434f9c78b2bSJens Axboe * 1435f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1436f9c78b2bSJens Axboe */ 1437f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1438f9c78b2bSJens Axboe { 1439163cc2d3SChristoph Hellwig bio_release_pages(bio, bio_data_dir(bio) == READ); 1440163cc2d3SChristoph Hellwig bio_put(bio); 1441f9c78b2bSJens Axboe bio_put(bio); 1442f9c78b2bSJens Axboe } 1443f9c78b2bSJens Axboe 14444246a0b6SChristoph Hellwig static void bio_map_kern_endio(struct bio *bio) 1445f9c78b2bSJens Axboe { 1446f9c78b2bSJens Axboe bio_put(bio); 1447f9c78b2bSJens Axboe } 1448f9c78b2bSJens Axboe 144975c72b83SChristoph Hellwig /** 145075c72b83SChristoph Hellwig * bio_map_kern - map kernel address into bio 145175c72b83SChristoph Hellwig * @q: the struct request_queue for the bio 145275c72b83SChristoph Hellwig * @data: pointer to buffer to map 145375c72b83SChristoph Hellwig * @len: length in bytes 145475c72b83SChristoph Hellwig * @gfp_mask: allocation flags for bio allocation 145575c72b83SChristoph Hellwig * 145675c72b83SChristoph Hellwig * Map the kernel address into a bio suitable for io to a block 145775c72b83SChristoph Hellwig * device. Returns an error pointer in case of error. 145875c72b83SChristoph Hellwig */ 145975c72b83SChristoph Hellwig struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 146075c72b83SChristoph Hellwig gfp_t gfp_mask) 1461f9c78b2bSJens Axboe { 1462f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1463f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1464f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1465f9c78b2bSJens Axboe const int nr_pages = end - start; 1466f9c78b2bSJens Axboe int offset, i; 1467f9c78b2bSJens Axboe struct bio *bio; 1468f9c78b2bSJens Axboe 1469f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1470f9c78b2bSJens Axboe if (!bio) 1471f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1472f9c78b2bSJens Axboe 1473f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1474f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1475f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1476f9c78b2bSJens Axboe 1477f9c78b2bSJens Axboe if (len <= 0) 1478f9c78b2bSJens Axboe break; 1479f9c78b2bSJens Axboe 1480f9c78b2bSJens Axboe if (bytes > len) 1481f9c78b2bSJens Axboe bytes = len; 1482f9c78b2bSJens Axboe 1483f9c78b2bSJens Axboe if (bio_add_pc_page(q, bio, virt_to_page(data), bytes, 148475c72b83SChristoph Hellwig offset) < bytes) { 148575c72b83SChristoph Hellwig /* we don't support partial mappings */ 148675c72b83SChristoph Hellwig bio_put(bio); 148775c72b83SChristoph Hellwig return ERR_PTR(-EINVAL); 148875c72b83SChristoph Hellwig } 1489f9c78b2bSJens Axboe 1490f9c78b2bSJens Axboe data += bytes; 1491f9c78b2bSJens Axboe len -= bytes; 1492f9c78b2bSJens Axboe offset = 0; 1493f9c78b2bSJens Axboe } 1494f9c78b2bSJens Axboe 1495f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1496f9c78b2bSJens Axboe return bio; 1497f9c78b2bSJens Axboe } 1498f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_map_kern); 1499f9c78b2bSJens Axboe 15004246a0b6SChristoph Hellwig static void bio_copy_kern_endio(struct bio *bio) 1501f9c78b2bSJens Axboe { 15021dfa0f68SChristoph Hellwig bio_free_pages(bio); 15031dfa0f68SChristoph Hellwig bio_put(bio); 15041dfa0f68SChristoph Hellwig } 15051dfa0f68SChristoph Hellwig 15064246a0b6SChristoph Hellwig static void bio_copy_kern_endio_read(struct bio *bio) 15071dfa0f68SChristoph Hellwig { 150842d2683aSChristoph Hellwig char *p = bio->bi_private; 15091dfa0f68SChristoph Hellwig struct bio_vec *bvec; 15106dc4f100SMing Lei struct bvec_iter_all iter_all; 1511f9c78b2bSJens Axboe 15122b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 15131dfa0f68SChristoph Hellwig memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1514f9c78b2bSJens Axboe p += bvec->bv_len; 1515f9c78b2bSJens Axboe } 1516f9c78b2bSJens Axboe 15174246a0b6SChristoph Hellwig bio_copy_kern_endio(bio); 1518f9c78b2bSJens Axboe } 1519f9c78b2bSJens Axboe 1520f9c78b2bSJens Axboe /** 1521f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1522f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1523f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1524f9c78b2bSJens Axboe * @len: length in bytes 1525f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1526f9c78b2bSJens Axboe * @reading: data direction is READ 1527f9c78b2bSJens Axboe * 1528f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1529f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1530f9c78b2bSJens Axboe */ 1531f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1532f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1533f9c78b2bSJens Axboe { 153442d2683aSChristoph Hellwig unsigned long kaddr = (unsigned long)data; 153542d2683aSChristoph Hellwig unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 153642d2683aSChristoph Hellwig unsigned long start = kaddr >> PAGE_SHIFT; 153742d2683aSChristoph Hellwig struct bio *bio; 1538f9c78b2bSJens Axboe void *p = data; 15391dfa0f68SChristoph Hellwig int nr_pages = 0; 1540f9c78b2bSJens Axboe 154142d2683aSChristoph Hellwig /* 154242d2683aSChristoph Hellwig * Overflow, abort 154342d2683aSChristoph Hellwig */ 154442d2683aSChristoph Hellwig if (end < start) 154542d2683aSChristoph Hellwig return ERR_PTR(-EINVAL); 1546f9c78b2bSJens Axboe 154742d2683aSChristoph Hellwig nr_pages = end - start; 154842d2683aSChristoph Hellwig bio = bio_kmalloc(gfp_mask, nr_pages); 154942d2683aSChristoph Hellwig if (!bio) 155042d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 155142d2683aSChristoph Hellwig 155242d2683aSChristoph Hellwig while (len) { 155342d2683aSChristoph Hellwig struct page *page; 155442d2683aSChristoph Hellwig unsigned int bytes = PAGE_SIZE; 155542d2683aSChristoph Hellwig 155642d2683aSChristoph Hellwig if (bytes > len) 155742d2683aSChristoph Hellwig bytes = len; 155842d2683aSChristoph Hellwig 155942d2683aSChristoph Hellwig page = alloc_page(q->bounce_gfp | gfp_mask); 156042d2683aSChristoph Hellwig if (!page) 156142d2683aSChristoph Hellwig goto cleanup; 156242d2683aSChristoph Hellwig 156342d2683aSChristoph Hellwig if (!reading) 156442d2683aSChristoph Hellwig memcpy(page_address(page), p, bytes); 156542d2683aSChristoph Hellwig 156642d2683aSChristoph Hellwig if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 156742d2683aSChristoph Hellwig break; 156842d2683aSChristoph Hellwig 156942d2683aSChristoph Hellwig len -= bytes; 157042d2683aSChristoph Hellwig p += bytes; 1571f9c78b2bSJens Axboe } 1572f9c78b2bSJens Axboe 15731dfa0f68SChristoph Hellwig if (reading) { 15741dfa0f68SChristoph Hellwig bio->bi_end_io = bio_copy_kern_endio_read; 157542d2683aSChristoph Hellwig bio->bi_private = data; 15761dfa0f68SChristoph Hellwig } else { 1577f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 15781dfa0f68SChristoph Hellwig } 15791dfa0f68SChristoph Hellwig 1580f9c78b2bSJens Axboe return bio; 158142d2683aSChristoph Hellwig 158242d2683aSChristoph Hellwig cleanup: 15831dfa0f68SChristoph Hellwig bio_free_pages(bio); 158442d2683aSChristoph Hellwig bio_put(bio); 158542d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 1586f9c78b2bSJens Axboe } 1587f9c78b2bSJens Axboe 1588f9c78b2bSJens Axboe /* 1589f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1590f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1591f9c78b2bSJens Axboe * 1592f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1593f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1594f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1595f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1596f9c78b2bSJens Axboe * in process context. 1597f9c78b2bSJens Axboe * 1598f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1599f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1600f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1601f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1602f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1603f9c78b2bSJens Axboe * 1604f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1605f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1606f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1607f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1608f9c78b2bSJens Axboe * pagecache. 1609f9c78b2bSJens Axboe * 1610f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1611f9c78b2bSJens Axboe * deferred bio dirtying paths. 1612f9c78b2bSJens Axboe */ 1613f9c78b2bSJens Axboe 1614f9c78b2bSJens Axboe /* 1615f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1616f9c78b2bSJens Axboe */ 1617f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1618f9c78b2bSJens Axboe { 1619f9c78b2bSJens Axboe struct bio_vec *bvec; 16206dc4f100SMing Lei struct bvec_iter_all iter_all; 1621f9c78b2bSJens Axboe 16222b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 16233bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 16243bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1625f9c78b2bSJens Axboe } 1626f9c78b2bSJens Axboe } 1627f9c78b2bSJens Axboe 1628f9c78b2bSJens Axboe /* 1629f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1630f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1631f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 163224d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1633f9c78b2bSJens Axboe * 1634f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1635ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1636ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1637f9c78b2bSJens Axboe */ 1638f9c78b2bSJens Axboe 1639f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1640f9c78b2bSJens Axboe 1641f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1642f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1643f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1644f9c78b2bSJens Axboe 1645f9c78b2bSJens Axboe /* 1646f9c78b2bSJens Axboe * This runs in process context 1647f9c78b2bSJens Axboe */ 1648f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1649f9c78b2bSJens Axboe { 165024d5493fSChristoph Hellwig struct bio *bio, *next; 1651f9c78b2bSJens Axboe 165224d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 165324d5493fSChristoph Hellwig next = bio_dirty_list; 1654f9c78b2bSJens Axboe bio_dirty_list = NULL; 165524d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1656f9c78b2bSJens Axboe 165724d5493fSChristoph Hellwig while ((bio = next) != NULL) { 165824d5493fSChristoph Hellwig next = bio->bi_private; 1659f9c78b2bSJens Axboe 1660d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1661f9c78b2bSJens Axboe bio_put(bio); 1662f9c78b2bSJens Axboe } 1663f9c78b2bSJens Axboe } 1664f9c78b2bSJens Axboe 1665f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1666f9c78b2bSJens Axboe { 1667f9c78b2bSJens Axboe struct bio_vec *bvec; 166824d5493fSChristoph Hellwig unsigned long flags; 16696dc4f100SMing Lei struct bvec_iter_all iter_all; 1670f9c78b2bSJens Axboe 16712b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 167224d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 167324d5493fSChristoph Hellwig goto defer; 1674f9c78b2bSJens Axboe } 1675f9c78b2bSJens Axboe 1676d241a95fSChristoph Hellwig bio_release_pages(bio, false); 167724d5493fSChristoph Hellwig bio_put(bio); 167824d5493fSChristoph Hellwig return; 167924d5493fSChristoph Hellwig defer: 1680f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1681f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1682f9c78b2bSJens Axboe bio_dirty_list = bio; 1683f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1684f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1685f9c78b2bSJens Axboe } 1686f9c78b2bSJens Axboe 16875b18b5a7SMikulas Patocka void update_io_ticks(struct hd_struct *part, unsigned long now) 16885b18b5a7SMikulas Patocka { 16895b18b5a7SMikulas Patocka unsigned long stamp; 16905b18b5a7SMikulas Patocka again: 16915b18b5a7SMikulas Patocka stamp = READ_ONCE(part->stamp); 16925b18b5a7SMikulas Patocka if (unlikely(stamp != now)) { 16935b18b5a7SMikulas Patocka if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 16945b18b5a7SMikulas Patocka __part_stat_add(part, io_ticks, 1); 16955b18b5a7SMikulas Patocka } 16965b18b5a7SMikulas Patocka } 16975b18b5a7SMikulas Patocka if (part->partno) { 16985b18b5a7SMikulas Patocka part = &part_to_disk(part)->part0; 16995b18b5a7SMikulas Patocka goto again; 17005b18b5a7SMikulas Patocka } 17015b18b5a7SMikulas Patocka } 1702f9c78b2bSJens Axboe 1703ddcf35d3SMichael Callahan void generic_start_io_acct(struct request_queue *q, int op, 1704d62e26b3SJens Axboe unsigned long sectors, struct hd_struct *part) 1705394ffa50SGu Zheng { 1706ddcf35d3SMichael Callahan const int sgrp = op_stat_group(op); 1707394ffa50SGu Zheng 1708112f158fSMike Snitzer part_stat_lock(); 1709112f158fSMike Snitzer 17105b18b5a7SMikulas Patocka update_io_ticks(part, jiffies); 1711112f158fSMike Snitzer part_stat_inc(part, ios[sgrp]); 1712112f158fSMike Snitzer part_stat_add(part, sectors[sgrp], sectors); 1713ddcf35d3SMichael Callahan part_inc_in_flight(q, part, op_is_write(op)); 1714394ffa50SGu Zheng 1715394ffa50SGu Zheng part_stat_unlock(); 1716394ffa50SGu Zheng } 1717394ffa50SGu Zheng EXPORT_SYMBOL(generic_start_io_acct); 1718394ffa50SGu Zheng 1719ddcf35d3SMichael Callahan void generic_end_io_acct(struct request_queue *q, int req_op, 1720d62e26b3SJens Axboe struct hd_struct *part, unsigned long start_time) 1721394ffa50SGu Zheng { 17225b18b5a7SMikulas Patocka unsigned long now = jiffies; 17235b18b5a7SMikulas Patocka unsigned long duration = now - start_time; 1724ddcf35d3SMichael Callahan const int sgrp = op_stat_group(req_op); 1725394ffa50SGu Zheng 1726112f158fSMike Snitzer part_stat_lock(); 1727112f158fSMike Snitzer 17285b18b5a7SMikulas Patocka update_io_ticks(part, now); 1729112f158fSMike Snitzer part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 17305b18b5a7SMikulas Patocka part_stat_add(part, time_in_queue, duration); 1731ddcf35d3SMichael Callahan part_dec_in_flight(q, part, op_is_write(req_op)); 1732394ffa50SGu Zheng 1733394ffa50SGu Zheng part_stat_unlock(); 1734394ffa50SGu Zheng } 1735394ffa50SGu Zheng EXPORT_SYMBOL(generic_end_io_acct); 1736394ffa50SGu Zheng 1737c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1738c4cf5261SJens Axboe { 1739c4cf5261SJens Axboe /* 1740c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1741c4cf5261SJens Axboe * we always end io on the first invocation. 1742c4cf5261SJens Axboe */ 1743c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1744c4cf5261SJens Axboe return true; 1745c4cf5261SJens Axboe 1746c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1747c4cf5261SJens Axboe 1748326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1749b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1750c4cf5261SJens Axboe return true; 1751326e1dbbSMike Snitzer } 1752c4cf5261SJens Axboe 1753c4cf5261SJens Axboe return false; 1754c4cf5261SJens Axboe } 1755c4cf5261SJens Axboe 1756f9c78b2bSJens Axboe /** 1757f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1758f9c78b2bSJens Axboe * @bio: bio 1759f9c78b2bSJens Axboe * 1760f9c78b2bSJens Axboe * Description: 17614246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 17624246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 17634246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1764fbbaf700SNeilBrown * 1765fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1766fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1767fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1768fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1769f9c78b2bSJens Axboe **/ 17704246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1771f9c78b2bSJens Axboe { 1772ba8c6967SChristoph Hellwig again: 17732b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1774ba8c6967SChristoph Hellwig return; 17757c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 17767c20f116SChristoph Hellwig return; 1777f9c78b2bSJens Axboe 177867b42d0bSJosef Bacik if (bio->bi_disk) 177967b42d0bSJosef Bacik rq_qos_done_bio(bio->bi_disk->queue, bio); 178067b42d0bSJosef Bacik 1781f9c78b2bSJens Axboe /* 1782ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1783ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1784ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1785ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1786ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1787ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1788f9c78b2bSJens Axboe */ 1789f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 179038f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1791ba8c6967SChristoph Hellwig goto again; 1792ba8c6967SChristoph Hellwig } 1793ba8c6967SChristoph Hellwig 179474d46992SChristoph Hellwig if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 179574d46992SChristoph Hellwig trace_block_bio_complete(bio->bi_disk->queue, bio, 1796a462b950SBart Van Assche blk_status_to_errno(bio->bi_status)); 1797fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1798fbbaf700SNeilBrown } 1799fbbaf700SNeilBrown 18009e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1801b222dd2fSShaohua Li /* release cgroup info */ 1802b222dd2fSShaohua Li bio_uninit(bio); 1803f9c78b2bSJens Axboe if (bio->bi_end_io) 18044246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1805f9c78b2bSJens Axboe } 1806f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1807f9c78b2bSJens Axboe 1808f9c78b2bSJens Axboe /** 1809f9c78b2bSJens Axboe * bio_split - split a bio 1810f9c78b2bSJens Axboe * @bio: bio to split 1811f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1812f9c78b2bSJens Axboe * @gfp: gfp mask 1813f9c78b2bSJens Axboe * @bs: bio set to allocate from 1814f9c78b2bSJens Axboe * 1815f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1816f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1817f9c78b2bSJens Axboe * 1818f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1819f3f5da62SMartin K. Petersen * to @bio's bi_io_vec; it is the caller's responsibility to ensure that 1820f3f5da62SMartin K. Petersen * @bio is not freed before the split. 1821f9c78b2bSJens Axboe */ 1822f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1823f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1824f9c78b2bSJens Axboe { 1825f341a4d3SMikulas Patocka struct bio *split; 1826f9c78b2bSJens Axboe 1827f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1828f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1829f9c78b2bSJens Axboe 1830f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1831f9c78b2bSJens Axboe if (!split) 1832f9c78b2bSJens Axboe return NULL; 1833f9c78b2bSJens Axboe 1834f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1835f9c78b2bSJens Axboe 1836f9c78b2bSJens Axboe if (bio_integrity(split)) 1837fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1838f9c78b2bSJens Axboe 1839f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1840f9c78b2bSJens Axboe 1841fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 184220d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1843fbbaf700SNeilBrown 1844f9c78b2bSJens Axboe return split; 1845f9c78b2bSJens Axboe } 1846f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1847f9c78b2bSJens Axboe 1848f9c78b2bSJens Axboe /** 1849f9c78b2bSJens Axboe * bio_trim - trim a bio 1850f9c78b2bSJens Axboe * @bio: bio to trim 1851f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1852f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1853f9c78b2bSJens Axboe */ 1854f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1855f9c78b2bSJens Axboe { 1856f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1857f9c78b2bSJens Axboe * the given offset and size. 1858f9c78b2bSJens Axboe */ 1859f9c78b2bSJens Axboe 1860f9c78b2bSJens Axboe size <<= 9; 1861f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1862f9c78b2bSJens Axboe return; 1863f9c78b2bSJens Axboe 1864f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1865f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1866376a78abSDmitry Monakhov 1867376a78abSDmitry Monakhov if (bio_integrity(bio)) 1868fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1869376a78abSDmitry Monakhov 1870f9c78b2bSJens Axboe } 1871f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1872f9c78b2bSJens Axboe 1873f9c78b2bSJens Axboe /* 1874f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1875f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1876f9c78b2bSJens Axboe */ 18778aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1878f9c78b2bSJens Axboe { 1879ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1880f9c78b2bSJens Axboe 18818aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1882f9c78b2bSJens Axboe } 1883f9c78b2bSJens Axboe 1884917a38c7SKent Overstreet /* 1885917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1886917a38c7SKent Overstreet * 1887917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1888917a38c7SKent Overstreet * kzalloc()). 1889917a38c7SKent Overstreet */ 1890917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1891f9c78b2bSJens Axboe { 1892f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1893f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1894917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1895f9c78b2bSJens Axboe 18968aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 18978aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1898f9c78b2bSJens Axboe 1899f9c78b2bSJens Axboe bioset_integrity_free(bs); 1900917a38c7SKent Overstreet if (bs->bio_slab) 1901f9c78b2bSJens Axboe bio_put_slab(bs); 1902917a38c7SKent Overstreet bs->bio_slab = NULL; 1903917a38c7SKent Overstreet } 1904917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1905f9c78b2bSJens Axboe 1906011067b0SNeilBrown /** 1907917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1908dad08527SKent Overstreet * @bs: pool to initialize 1909917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1910917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1911917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1912917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1913917a38c7SKent Overstreet * 1914dad08527SKent Overstreet * Description: 1915dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1916dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1917dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1918dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1919dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1920dad08527SKent Overstreet * or things will break badly. 1921dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1922dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1923dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1924dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1925dad08527SKent Overstreet * 1926917a38c7SKent Overstreet */ 1927917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1928917a38c7SKent Overstreet unsigned int pool_size, 1929917a38c7SKent Overstreet unsigned int front_pad, 1930917a38c7SKent Overstreet int flags) 1931917a38c7SKent Overstreet { 1932917a38c7SKent Overstreet unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1933917a38c7SKent Overstreet 1934917a38c7SKent Overstreet bs->front_pad = front_pad; 1935917a38c7SKent Overstreet 1936917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1937917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1938917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1939917a38c7SKent Overstreet 1940917a38c7SKent Overstreet bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1941917a38c7SKent Overstreet if (!bs->bio_slab) 1942917a38c7SKent Overstreet return -ENOMEM; 1943917a38c7SKent Overstreet 1944917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1945917a38c7SKent Overstreet goto bad; 1946917a38c7SKent Overstreet 1947917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1948917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1949917a38c7SKent Overstreet goto bad; 1950917a38c7SKent Overstreet 1951917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 1952917a38c7SKent Overstreet return 0; 1953917a38c7SKent Overstreet 1954917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1955917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1956917a38c7SKent Overstreet goto bad; 1957917a38c7SKent Overstreet 1958917a38c7SKent Overstreet return 0; 1959917a38c7SKent Overstreet bad: 1960917a38c7SKent Overstreet bioset_exit(bs); 1961917a38c7SKent Overstreet return -ENOMEM; 1962917a38c7SKent Overstreet } 1963917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1964917a38c7SKent Overstreet 196528e89fd9SJens Axboe /* 196628e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 196728e89fd9SJens Axboe * another bio_set. 196828e89fd9SJens Axboe */ 196928e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 197028e89fd9SJens Axboe { 197128e89fd9SJens Axboe int flags; 197228e89fd9SJens Axboe 197328e89fd9SJens Axboe flags = 0; 197428e89fd9SJens Axboe if (src->bvec_pool.min_nr) 197528e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 197628e89fd9SJens Axboe if (src->rescue_workqueue) 197728e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 197828e89fd9SJens Axboe 197928e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 198028e89fd9SJens Axboe } 198128e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 198228e89fd9SJens Axboe 1983f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 19841d933cf0STejun Heo 19851d933cf0STejun Heo /** 19862268c0feSDennis Zhou * bio_disassociate_blkg - puts back the blkg reference if associated 1987b5f2954dSDennis Zhou * @bio: target bio 1988b5f2954dSDennis Zhou * 19892268c0feSDennis Zhou * Helper to disassociate the blkg from @bio if a blkg is associated. 1990b5f2954dSDennis Zhou */ 19912268c0feSDennis Zhou void bio_disassociate_blkg(struct bio *bio) 1992b5f2954dSDennis Zhou { 199308e18eabSJosef Bacik if (bio->bi_blkg) { 199408e18eabSJosef Bacik blkg_put(bio->bi_blkg); 199508e18eabSJosef Bacik bio->bi_blkg = NULL; 199608e18eabSJosef Bacik } 1997f9c78b2bSJens Axboe } 1998892ad71fSDennis Zhou EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 1999f9c78b2bSJens Axboe 200020bd723eSPaolo Valente /** 20012268c0feSDennis Zhou * __bio_associate_blkg - associate a bio with the a blkg 2002f9c78b2bSJens Axboe * @bio: target bio 2003f9c78b2bSJens Axboe * @blkg: the blkg to associate 2004f9c78b2bSJens Axboe * 2005beea9da0SDennis Zhou * This tries to associate @bio with the specified @blkg. Association failure 2006beea9da0SDennis Zhou * is handled by walking up the blkg tree. Therefore, the blkg associated can 2007beea9da0SDennis Zhou * be anything between @blkg and the root_blkg. This situation only happens 2008beea9da0SDennis Zhou * when a cgroup is dying and then the remaining bios will spill to the closest 2009beea9da0SDennis Zhou * alive blkg. 2010beea9da0SDennis Zhou * 2011beea9da0SDennis Zhou * A reference will be taken on the @blkg and will be released when @bio is 2012beea9da0SDennis Zhou * freed. 2013f9c78b2bSJens Axboe */ 20142268c0feSDennis Zhou static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2015f9c78b2bSJens Axboe { 20162268c0feSDennis Zhou bio_disassociate_blkg(bio); 20172268c0feSDennis Zhou 20187754f669SDennis Zhou bio->bi_blkg = blkg_tryget_closest(blkg); 20192268c0feSDennis Zhou } 20202268c0feSDennis Zhou 2021fd42df30SDennis Zhou /** 2022fd42df30SDennis Zhou * bio_associate_blkg_from_css - associate a bio with a specified css 2023fd42df30SDennis Zhou * @bio: target bio 2024fd42df30SDennis Zhou * @css: target css 2025fd42df30SDennis Zhou * 2026fd42df30SDennis Zhou * Associate @bio with the blkg found by combining the css's blkg and the 2027fc5a828bSDennis Zhou * request_queue of the @bio. This falls back to the queue's root_blkg if 2028fc5a828bSDennis Zhou * the association fails with the css. 2029fd42df30SDennis Zhou */ 2030fd42df30SDennis Zhou void bio_associate_blkg_from_css(struct bio *bio, 2031fd42df30SDennis Zhou struct cgroup_subsys_state *css) 2032fd42df30SDennis Zhou { 2033fc5a828bSDennis Zhou struct request_queue *q = bio->bi_disk->queue; 2034fc5a828bSDennis Zhou struct blkcg_gq *blkg; 2035fc5a828bSDennis Zhou 2036fc5a828bSDennis Zhou rcu_read_lock(); 2037fc5a828bSDennis Zhou 2038fc5a828bSDennis Zhou if (!css || !css->parent) 2039fc5a828bSDennis Zhou blkg = q->root_blkg; 2040fc5a828bSDennis Zhou else 2041fc5a828bSDennis Zhou blkg = blkg_lookup_create(css_to_blkcg(css), q); 2042fc5a828bSDennis Zhou 2043fc5a828bSDennis Zhou __bio_associate_blkg(bio, blkg); 2044fc5a828bSDennis Zhou 2045fc5a828bSDennis Zhou rcu_read_unlock(); 2046fd42df30SDennis Zhou } 2047fd42df30SDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2048fd42df30SDennis Zhou 20496a7f6d86SDennis Zhou #ifdef CONFIG_MEMCG 20506a7f6d86SDennis Zhou /** 20516a7f6d86SDennis Zhou * bio_associate_blkg_from_page - associate a bio with the page's blkg 20526a7f6d86SDennis Zhou * @bio: target bio 20536a7f6d86SDennis Zhou * @page: the page to lookup the blkcg from 20546a7f6d86SDennis Zhou * 20556a7f6d86SDennis Zhou * Associate @bio with the blkg from @page's owning memcg and the respective 2056fc5a828bSDennis Zhou * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2057fc5a828bSDennis Zhou * root_blkg. 20586a7f6d86SDennis Zhou */ 20596a7f6d86SDennis Zhou void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 20606a7f6d86SDennis Zhou { 20616a7f6d86SDennis Zhou struct cgroup_subsys_state *css; 20626a7f6d86SDennis Zhou 20636a7f6d86SDennis Zhou if (!page->mem_cgroup) 20646a7f6d86SDennis Zhou return; 20656a7f6d86SDennis Zhou 2066fc5a828bSDennis Zhou rcu_read_lock(); 2067fc5a828bSDennis Zhou 2068fc5a828bSDennis Zhou css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2069fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 2070fc5a828bSDennis Zhou 2071fc5a828bSDennis Zhou rcu_read_unlock(); 20726a7f6d86SDennis Zhou } 20736a7f6d86SDennis Zhou #endif /* CONFIG_MEMCG */ 20746a7f6d86SDennis Zhou 20752268c0feSDennis Zhou /** 20762268c0feSDennis Zhou * bio_associate_blkg - associate a bio with a blkg 20772268c0feSDennis Zhou * @bio: target bio 20782268c0feSDennis Zhou * 20792268c0feSDennis Zhou * Associate @bio with the blkg found from the bio's css and request_queue. 20802268c0feSDennis Zhou * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 20812268c0feSDennis Zhou * already associated, the css is reused and association redone as the 20822268c0feSDennis Zhou * request_queue may have changed. 20832268c0feSDennis Zhou */ 20842268c0feSDennis Zhou void bio_associate_blkg(struct bio *bio) 20852268c0feSDennis Zhou { 2086fc5a828bSDennis Zhou struct cgroup_subsys_state *css; 20872268c0feSDennis Zhou 20882268c0feSDennis Zhou rcu_read_lock(); 20892268c0feSDennis Zhou 2090db6638d7SDennis Zhou if (bio->bi_blkg) 2091fc5a828bSDennis Zhou css = &bio_blkcg(bio)->css; 2092db6638d7SDennis Zhou else 2093fc5a828bSDennis Zhou css = blkcg_css(); 20942268c0feSDennis Zhou 2095fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 20962268c0feSDennis Zhou 20972268c0feSDennis Zhou rcu_read_unlock(); 2098f9c78b2bSJens Axboe } 20995cdf2e3fSDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg); 2100f9c78b2bSJens Axboe 210120bd723eSPaolo Valente /** 2102db6638d7SDennis Zhou * bio_clone_blkg_association - clone blkg association from src to dst bio 210320bd723eSPaolo Valente * @dst: destination bio 210420bd723eSPaolo Valente * @src: source bio 210520bd723eSPaolo Valente */ 2106db6638d7SDennis Zhou void bio_clone_blkg_association(struct bio *dst, struct bio *src) 210720bd723eSPaolo Valente { 21086ab21879SDennis Zhou rcu_read_lock(); 21096ab21879SDennis Zhou 2110fc5a828bSDennis Zhou if (src->bi_blkg) 21112268c0feSDennis Zhou __bio_associate_blkg(dst, src->bi_blkg); 21126ab21879SDennis Zhou 21136ab21879SDennis Zhou rcu_read_unlock(); 211420bd723eSPaolo Valente } 2115db6638d7SDennis Zhou EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2116f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 2117f9c78b2bSJens Axboe 2118f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2119f9c78b2bSJens Axboe { 2120f9c78b2bSJens Axboe int i; 2121f9c78b2bSJens Axboe 2122ed996a52SChristoph Hellwig for (i = 0; i < BVEC_POOL_NR; i++) { 2123f9c78b2bSJens Axboe int size; 2124f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2125f9c78b2bSJens Axboe 2126f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2127f9c78b2bSJens Axboe bvs->slab = NULL; 2128f9c78b2bSJens Axboe continue; 2129f9c78b2bSJens Axboe } 2130f9c78b2bSJens Axboe 2131f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2132f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2133f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2134f9c78b2bSJens Axboe } 2135f9c78b2bSJens Axboe } 2136f9c78b2bSJens Axboe 2137f9c78b2bSJens Axboe static int __init init_bio(void) 2138f9c78b2bSJens Axboe { 2139f9c78b2bSJens Axboe bio_slab_max = 2; 2140f9c78b2bSJens Axboe bio_slab_nr = 0; 21416396bb22SKees Cook bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 21426396bb22SKees Cook GFP_KERNEL); 21432b24e6f6SJohannes Thumshirn 21442b24e6f6SJohannes Thumshirn BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 21452b24e6f6SJohannes Thumshirn 2146f9c78b2bSJens Axboe if (!bio_slabs) 2147f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2148f9c78b2bSJens Axboe 2149f9c78b2bSJens Axboe bio_integrity_init(); 2150f9c78b2bSJens Axboe biovec_init_slabs(); 2151f9c78b2bSJens Axboe 2152f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2153f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2154f9c78b2bSJens Axboe 2155f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2156f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2157f9c78b2bSJens Axboe 2158f9c78b2bSJens Axboe return 0; 2159f9c78b2bSJens Axboe } 2160f9c78b2bSJens Axboe subsys_initcall(init_bio); 2161