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> 19b4c5875dSDamien Le Moal #include <linux/highmem.h> 20f9c78b2bSJens Axboe 21f9c78b2bSJens Axboe #include <trace/events/block.h> 229e234eeaSShaohua Li #include "blk.h" 2367b42d0bSJosef Bacik #include "blk-rq-qos.h" 24f9c78b2bSJens Axboe 25f9c78b2bSJens Axboe /* 26f9c78b2bSJens Axboe * Test patch to inline a certain number of bi_io_vec's inside the bio 27f9c78b2bSJens Axboe * itself, to shrink a bio data allocation from two mempool calls to one 28f9c78b2bSJens Axboe */ 29f9c78b2bSJens Axboe #define BIO_INLINE_VECS 4 30f9c78b2bSJens Axboe 31f9c78b2bSJens Axboe /* 32f9c78b2bSJens Axboe * if you change this list, also change bvec_alloc or things will 33f9c78b2bSJens Axboe * break badly! cannot be bigger than what you can fit into an 34f9c78b2bSJens Axboe * unsigned short 35f9c78b2bSJens Axboe */ 36bd5c4facSMikulas Patocka #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 37ed996a52SChristoph Hellwig static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 38bd5c4facSMikulas Patocka BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 39f9c78b2bSJens Axboe }; 40f9c78b2bSJens Axboe #undef BV 41f9c78b2bSJens Axboe 42f9c78b2bSJens Axboe /* 43f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 44f9c78b2bSJens Axboe * IO code that does not need private memory pools. 45f9c78b2bSJens Axboe */ 46f4f8154aSKent Overstreet struct bio_set fs_bio_set; 47f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 48f9c78b2bSJens Axboe 49f9c78b2bSJens Axboe /* 50f9c78b2bSJens Axboe * Our slab pool management 51f9c78b2bSJens Axboe */ 52f9c78b2bSJens Axboe struct bio_slab { 53f9c78b2bSJens Axboe struct kmem_cache *slab; 54f9c78b2bSJens Axboe unsigned int slab_ref; 55f9c78b2bSJens Axboe unsigned int slab_size; 56f9c78b2bSJens Axboe char name[8]; 57f9c78b2bSJens Axboe }; 58f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 59f9c78b2bSJens Axboe static struct bio_slab *bio_slabs; 60f9c78b2bSJens Axboe static unsigned int bio_slab_nr, bio_slab_max; 61f9c78b2bSJens Axboe 62f9c78b2bSJens Axboe static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 63f9c78b2bSJens Axboe { 64f9c78b2bSJens Axboe unsigned int sz = sizeof(struct bio) + extra_size; 65f9c78b2bSJens Axboe struct kmem_cache *slab = NULL; 66f9c78b2bSJens Axboe struct bio_slab *bslab, *new_bio_slabs; 67f9c78b2bSJens Axboe unsigned int new_bio_slab_max; 68f9c78b2bSJens Axboe unsigned int i, entry = -1; 69f9c78b2bSJens Axboe 70f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 71f9c78b2bSJens Axboe 72f9c78b2bSJens Axboe i = 0; 73f9c78b2bSJens Axboe while (i < bio_slab_nr) { 74f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 75f9c78b2bSJens Axboe 76f9c78b2bSJens Axboe if (!bslab->slab && entry == -1) 77f9c78b2bSJens Axboe entry = i; 78f9c78b2bSJens Axboe else if (bslab->slab_size == sz) { 79f9c78b2bSJens Axboe slab = bslab->slab; 80f9c78b2bSJens Axboe bslab->slab_ref++; 81f9c78b2bSJens Axboe break; 82f9c78b2bSJens Axboe } 83f9c78b2bSJens Axboe i++; 84f9c78b2bSJens Axboe } 85f9c78b2bSJens Axboe 86f9c78b2bSJens Axboe if (slab) 87f9c78b2bSJens Axboe goto out_unlock; 88f9c78b2bSJens Axboe 89f9c78b2bSJens Axboe if (bio_slab_nr == bio_slab_max && entry == -1) { 90f9c78b2bSJens Axboe new_bio_slab_max = bio_slab_max << 1; 91f9c78b2bSJens Axboe new_bio_slabs = krealloc(bio_slabs, 92f9c78b2bSJens Axboe new_bio_slab_max * sizeof(struct bio_slab), 93f9c78b2bSJens Axboe GFP_KERNEL); 94f9c78b2bSJens Axboe if (!new_bio_slabs) 95f9c78b2bSJens Axboe goto out_unlock; 96f9c78b2bSJens Axboe bio_slab_max = new_bio_slab_max; 97f9c78b2bSJens Axboe bio_slabs = new_bio_slabs; 98f9c78b2bSJens Axboe } 99f9c78b2bSJens Axboe if (entry == -1) 100f9c78b2bSJens Axboe entry = bio_slab_nr++; 101f9c78b2bSJens Axboe 102f9c78b2bSJens Axboe bslab = &bio_slabs[entry]; 103f9c78b2bSJens Axboe 104f9c78b2bSJens Axboe snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 1056a241483SMikulas Patocka slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 1066a241483SMikulas Patocka SLAB_HWCACHE_ALIGN, NULL); 107f9c78b2bSJens Axboe if (!slab) 108f9c78b2bSJens Axboe goto out_unlock; 109f9c78b2bSJens Axboe 110f9c78b2bSJens Axboe bslab->slab = slab; 111f9c78b2bSJens Axboe bslab->slab_ref = 1; 112f9c78b2bSJens Axboe bslab->slab_size = sz; 113f9c78b2bSJens Axboe out_unlock: 114f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 115f9c78b2bSJens Axboe return slab; 116f9c78b2bSJens Axboe } 117f9c78b2bSJens Axboe 118f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 119f9c78b2bSJens Axboe { 120f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 121f9c78b2bSJens Axboe unsigned int i; 122f9c78b2bSJens Axboe 123f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 124f9c78b2bSJens Axboe 125f9c78b2bSJens Axboe for (i = 0; i < bio_slab_nr; i++) { 126f9c78b2bSJens Axboe if (bs->bio_slab == bio_slabs[i].slab) { 127f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 128f9c78b2bSJens Axboe break; 129f9c78b2bSJens Axboe } 130f9c78b2bSJens Axboe } 131f9c78b2bSJens Axboe 132f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 133f9c78b2bSJens Axboe goto out; 134f9c78b2bSJens Axboe 135f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 136f9c78b2bSJens Axboe 137f9c78b2bSJens Axboe if (--bslab->slab_ref) 138f9c78b2bSJens Axboe goto out; 139f9c78b2bSJens Axboe 140f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 141f9c78b2bSJens Axboe bslab->slab = NULL; 142f9c78b2bSJens Axboe 143f9c78b2bSJens Axboe out: 144f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 145f9c78b2bSJens Axboe } 146f9c78b2bSJens Axboe 147f9c78b2bSJens Axboe unsigned int bvec_nr_vecs(unsigned short idx) 148f9c78b2bSJens Axboe { 149d6c02a9bSGreg Edwards return bvec_slabs[--idx].nr_vecs; 150f9c78b2bSJens Axboe } 151f9c78b2bSJens Axboe 152f9c78b2bSJens Axboe void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 153f9c78b2bSJens Axboe { 154ed996a52SChristoph Hellwig if (!idx) 155ed996a52SChristoph Hellwig return; 156ed996a52SChristoph Hellwig idx--; 157f9c78b2bSJens Axboe 158ed996a52SChristoph Hellwig BIO_BUG_ON(idx >= BVEC_POOL_NR); 159ed996a52SChristoph Hellwig 160ed996a52SChristoph Hellwig if (idx == BVEC_POOL_MAX) { 161f9c78b2bSJens Axboe mempool_free(bv, pool); 162ed996a52SChristoph Hellwig } else { 163f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + idx; 164f9c78b2bSJens Axboe 165f9c78b2bSJens Axboe kmem_cache_free(bvs->slab, bv); 166f9c78b2bSJens Axboe } 167f9c78b2bSJens Axboe } 168f9c78b2bSJens Axboe 169f9c78b2bSJens Axboe struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 170f9c78b2bSJens Axboe mempool_t *pool) 171f9c78b2bSJens Axboe { 172f9c78b2bSJens Axboe struct bio_vec *bvl; 173f9c78b2bSJens Axboe 174f9c78b2bSJens Axboe /* 175f9c78b2bSJens Axboe * see comment near bvec_array define! 176f9c78b2bSJens Axboe */ 177f9c78b2bSJens Axboe switch (nr) { 178f9c78b2bSJens Axboe case 1: 179f9c78b2bSJens Axboe *idx = 0; 180f9c78b2bSJens Axboe break; 181f9c78b2bSJens Axboe case 2 ... 4: 182f9c78b2bSJens Axboe *idx = 1; 183f9c78b2bSJens Axboe break; 184f9c78b2bSJens Axboe case 5 ... 16: 185f9c78b2bSJens Axboe *idx = 2; 186f9c78b2bSJens Axboe break; 187f9c78b2bSJens Axboe case 17 ... 64: 188f9c78b2bSJens Axboe *idx = 3; 189f9c78b2bSJens Axboe break; 190f9c78b2bSJens Axboe case 65 ... 128: 191f9c78b2bSJens Axboe *idx = 4; 192f9c78b2bSJens Axboe break; 193f9c78b2bSJens Axboe case 129 ... BIO_MAX_PAGES: 194f9c78b2bSJens Axboe *idx = 5; 195f9c78b2bSJens Axboe break; 196f9c78b2bSJens Axboe default: 197f9c78b2bSJens Axboe return NULL; 198f9c78b2bSJens Axboe } 199f9c78b2bSJens Axboe 200f9c78b2bSJens Axboe /* 201f9c78b2bSJens Axboe * idx now points to the pool we want to allocate from. only the 202f9c78b2bSJens Axboe * 1-vec entry pool is mempool backed. 203f9c78b2bSJens Axboe */ 204ed996a52SChristoph Hellwig if (*idx == BVEC_POOL_MAX) { 205f9c78b2bSJens Axboe fallback: 206f9c78b2bSJens Axboe bvl = mempool_alloc(pool, gfp_mask); 207f9c78b2bSJens Axboe } else { 208f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + *idx; 209d0164adcSMel Gorman gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 210f9c78b2bSJens Axboe 211f9c78b2bSJens Axboe /* 212f9c78b2bSJens Axboe * Make this allocation restricted and don't dump info on 213f9c78b2bSJens Axboe * allocation failures, since we'll fallback to the mempool 214f9c78b2bSJens Axboe * in case of failure. 215f9c78b2bSJens Axboe */ 216f9c78b2bSJens Axboe __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 217f9c78b2bSJens Axboe 218f9c78b2bSJens Axboe /* 219d0164adcSMel Gorman * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 220f9c78b2bSJens Axboe * is set, retry with the 1-entry mempool 221f9c78b2bSJens Axboe */ 222f9c78b2bSJens Axboe bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 223d0164adcSMel Gorman if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 224ed996a52SChristoph Hellwig *idx = BVEC_POOL_MAX; 225f9c78b2bSJens Axboe goto fallback; 226f9c78b2bSJens Axboe } 227f9c78b2bSJens Axboe } 228f9c78b2bSJens Axboe 229ed996a52SChristoph Hellwig (*idx)++; 230f9c78b2bSJens Axboe return bvl; 231f9c78b2bSJens Axboe } 232f9c78b2bSJens Axboe 2339ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 234f9c78b2bSJens Axboe { 2356f70fb66SDennis Zhou bio_disassociate_blkg(bio); 236f9c78b2bSJens Axboe } 2379ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 238f9c78b2bSJens Axboe 239f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 240f9c78b2bSJens Axboe { 241f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 242f9c78b2bSJens Axboe void *p; 243f9c78b2bSJens Axboe 2449ae3b3f5SJens Axboe bio_uninit(bio); 245f9c78b2bSJens Axboe 246f9c78b2bSJens Axboe if (bs) { 2478aa6ba2fSKent Overstreet bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 248f9c78b2bSJens Axboe 249f9c78b2bSJens Axboe /* 250f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 251f9c78b2bSJens Axboe */ 252f9c78b2bSJens Axboe p = bio; 253f9c78b2bSJens Axboe p -= bs->front_pad; 254f9c78b2bSJens Axboe 2558aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 256f9c78b2bSJens Axboe } else { 257f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 258f9c78b2bSJens Axboe kfree(bio); 259f9c78b2bSJens Axboe } 260f9c78b2bSJens Axboe } 261f9c78b2bSJens Axboe 2629ae3b3f5SJens Axboe /* 2639ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2649ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2659ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2669ae3b3f5SJens Axboe */ 2673a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2683a83f467SMing Lei unsigned short max_vecs) 269f9c78b2bSJens Axboe { 270f9c78b2bSJens Axboe memset(bio, 0, sizeof(*bio)); 271c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 272dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2733a83f467SMing Lei 2743a83f467SMing Lei bio->bi_io_vec = table; 2753a83f467SMing Lei bio->bi_max_vecs = max_vecs; 276f9c78b2bSJens Axboe } 277f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 278f9c78b2bSJens Axboe 279f9c78b2bSJens Axboe /** 280f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 281f9c78b2bSJens Axboe * @bio: bio to reset 282f9c78b2bSJens Axboe * 283f9c78b2bSJens Axboe * Description: 284f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 285f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 286f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 287f9c78b2bSJens Axboe * comment in struct bio. 288f9c78b2bSJens Axboe */ 289f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 290f9c78b2bSJens Axboe { 291f9c78b2bSJens Axboe unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 292f9c78b2bSJens Axboe 2939ae3b3f5SJens Axboe bio_uninit(bio); 294f9c78b2bSJens Axboe 295f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 2964246a0b6SChristoph Hellwig bio->bi_flags = flags; 297c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 298f9c78b2bSJens Axboe } 299f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 300f9c78b2bSJens Axboe 30138f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 302f9c78b2bSJens Axboe { 3034246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3044246a0b6SChristoph Hellwig 3054e4cbee9SChristoph Hellwig if (!parent->bi_status) 3064e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 307f9c78b2bSJens Axboe bio_put(bio); 30838f8baaeSChristoph Hellwig return parent; 30938f8baaeSChristoph Hellwig } 31038f8baaeSChristoph Hellwig 31138f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 31238f8baaeSChristoph Hellwig { 31338f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 314f9c78b2bSJens Axboe } 315f9c78b2bSJens Axboe 316f9c78b2bSJens Axboe /** 317f9c78b2bSJens Axboe * bio_chain - chain bio completions 318f9c78b2bSJens Axboe * @bio: the target bio 319f9c78b2bSJens Axboe * @parent: the @bio's parent bio 320f9c78b2bSJens Axboe * 321f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 322f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 323f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 324f9c78b2bSJens Axboe * 325f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 326f9c78b2bSJens Axboe */ 327f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 328f9c78b2bSJens Axboe { 329f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 330f9c78b2bSJens Axboe 331f9c78b2bSJens Axboe bio->bi_private = parent; 332f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 333c4cf5261SJens Axboe bio_inc_remaining(parent); 334f9c78b2bSJens Axboe } 335f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 336f9c78b2bSJens Axboe 337f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 338f9c78b2bSJens Axboe { 339f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 340f9c78b2bSJens Axboe struct bio *bio; 341f9c78b2bSJens Axboe 342f9c78b2bSJens Axboe while (1) { 343f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 344f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 345f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 346f9c78b2bSJens Axboe 347f9c78b2bSJens Axboe if (!bio) 348f9c78b2bSJens Axboe break; 349f9c78b2bSJens Axboe 350f9c78b2bSJens Axboe generic_make_request(bio); 351f9c78b2bSJens Axboe } 352f9c78b2bSJens Axboe } 353f9c78b2bSJens Axboe 354f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 355f9c78b2bSJens Axboe { 356f9c78b2bSJens Axboe struct bio_list punt, nopunt; 357f9c78b2bSJens Axboe struct bio *bio; 358f9c78b2bSJens Axboe 35947e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 36047e0fb46SNeilBrown return; 361f9c78b2bSJens Axboe /* 362f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 363f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 364f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 365f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 366f9c78b2bSJens Axboe * our own rescuer would be bad. 367f9c78b2bSJens Axboe * 368f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 369f9c78b2bSJens Axboe * remove from the middle of the list: 370f9c78b2bSJens Axboe */ 371f9c78b2bSJens Axboe 372f9c78b2bSJens Axboe bio_list_init(&punt); 373f9c78b2bSJens Axboe bio_list_init(&nopunt); 374f9c78b2bSJens Axboe 375f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 376f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 377f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 378f9c78b2bSJens Axboe 379f5fe1b51SNeilBrown bio_list_init(&nopunt); 380f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 381f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 382f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 383f9c78b2bSJens Axboe 384f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 385f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 386f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 387f9c78b2bSJens Axboe 388f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 389f9c78b2bSJens Axboe } 390f9c78b2bSJens Axboe 391f9c78b2bSJens Axboe /** 392f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 393519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 394f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 395f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 396f9c78b2bSJens Axboe * 397f9c78b2bSJens Axboe * Description: 398f9c78b2bSJens Axboe * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 399f9c78b2bSJens Axboe * backed by the @bs's mempool. 400f9c78b2bSJens Axboe * 401d0164adcSMel Gorman * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 402d0164adcSMel Gorman * always be able to allocate a bio. This is due to the mempool guarantees. 403d0164adcSMel Gorman * To make this work, callers must never allocate more than 1 bio at a time 404d0164adcSMel Gorman * from this pool. Callers that need to allocate more than 1 bio must always 405d0164adcSMel Gorman * submit the previously allocated bio for IO before attempting to allocate 406d0164adcSMel Gorman * a new one. Failure to do so can cause deadlocks under memory pressure. 407f9c78b2bSJens Axboe * 408f9c78b2bSJens Axboe * Note that when running under generic_make_request() (i.e. any block 409f9c78b2bSJens Axboe * driver), bios are not submitted until after you return - see the code in 410f9c78b2bSJens Axboe * generic_make_request() that converts recursion into iteration, to prevent 411f9c78b2bSJens Axboe * stack overflows. 412f9c78b2bSJens Axboe * 413f9c78b2bSJens Axboe * This would normally mean allocating multiple bios under 414f9c78b2bSJens Axboe * generic_make_request() would be susceptible to deadlocks, but we have 415f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 416f9c78b2bSJens Axboe * thread. 417f9c78b2bSJens Axboe * 418f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 419f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 420f9c78b2bSJens Axboe * generic_make_request() should be avoided - instead, use bio_set's front_pad 421f9c78b2bSJens Axboe * for per bio allocations. 422f9c78b2bSJens Axboe * 423f9c78b2bSJens Axboe * RETURNS: 424f9c78b2bSJens Axboe * Pointer to new bio on success, NULL on failure. 425f9c78b2bSJens Axboe */ 4267a88fa19SDan Carpenter struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 4277a88fa19SDan Carpenter struct bio_set *bs) 428f9c78b2bSJens Axboe { 429f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 430f9c78b2bSJens Axboe unsigned front_pad; 431f9c78b2bSJens Axboe unsigned inline_vecs; 432f9c78b2bSJens Axboe struct bio_vec *bvl = NULL; 433f9c78b2bSJens Axboe struct bio *bio; 434f9c78b2bSJens Axboe void *p; 435f9c78b2bSJens Axboe 436f9c78b2bSJens Axboe if (!bs) { 437f9c78b2bSJens Axboe if (nr_iovecs > UIO_MAXIOV) 438f9c78b2bSJens Axboe return NULL; 439f9c78b2bSJens Axboe 440f9c78b2bSJens Axboe p = kmalloc(sizeof(struct bio) + 441f9c78b2bSJens Axboe nr_iovecs * sizeof(struct bio_vec), 442f9c78b2bSJens Axboe gfp_mask); 443f9c78b2bSJens Axboe front_pad = 0; 444f9c78b2bSJens Axboe inline_vecs = nr_iovecs; 445f9c78b2bSJens Axboe } else { 446d8f429e1SJunichi Nomura /* should not use nobvec bioset for nr_iovecs > 0 */ 4478aa6ba2fSKent Overstreet if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 4488aa6ba2fSKent Overstreet nr_iovecs > 0)) 449d8f429e1SJunichi Nomura return NULL; 450f9c78b2bSJens Axboe /* 451f9c78b2bSJens Axboe * generic_make_request() converts recursion to iteration; this 452f9c78b2bSJens Axboe * means if we're running beneath it, any bios we allocate and 453f9c78b2bSJens Axboe * submit will not be submitted (and thus freed) until after we 454f9c78b2bSJens Axboe * return. 455f9c78b2bSJens Axboe * 456f9c78b2bSJens Axboe * This exposes us to a potential deadlock if we allocate 457f9c78b2bSJens Axboe * multiple bios from the same bio_set() while running 458f9c78b2bSJens Axboe * underneath generic_make_request(). If we were to allocate 459f9c78b2bSJens Axboe * multiple bios (say a stacking block driver that was splitting 460f9c78b2bSJens Axboe * bios), we would deadlock if we exhausted the mempool's 461f9c78b2bSJens Axboe * reserve. 462f9c78b2bSJens Axboe * 463f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 464f9c78b2bSJens Axboe * workqueue per bio_set. If we go to allocate and there are 465f9c78b2bSJens Axboe * bios on current->bio_list, we first try the allocation 466d0164adcSMel Gorman * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 467d0164adcSMel Gorman * bios we would be blocking to the rescuer workqueue before 468d0164adcSMel Gorman * we retry with the original gfp_flags. 469f9c78b2bSJens Axboe */ 470f9c78b2bSJens Axboe 471f5fe1b51SNeilBrown if (current->bio_list && 472f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 47347e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 47447e0fb46SNeilBrown bs->rescue_workqueue) 475d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 476f9c78b2bSJens Axboe 4778aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 478f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 479f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 480f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4818aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 482f9c78b2bSJens Axboe } 483f9c78b2bSJens Axboe 484f9c78b2bSJens Axboe front_pad = bs->front_pad; 485f9c78b2bSJens Axboe inline_vecs = BIO_INLINE_VECS; 486f9c78b2bSJens Axboe } 487f9c78b2bSJens Axboe 488f9c78b2bSJens Axboe if (unlikely(!p)) 489f9c78b2bSJens Axboe return NULL; 490f9c78b2bSJens Axboe 491f9c78b2bSJens Axboe bio = p + front_pad; 4923a83f467SMing Lei bio_init(bio, NULL, 0); 493f9c78b2bSJens Axboe 494f9c78b2bSJens Axboe if (nr_iovecs > inline_vecs) { 495ed996a52SChristoph Hellwig unsigned long idx = 0; 496ed996a52SChristoph Hellwig 4978aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 498f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 499f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 500f9c78b2bSJens Axboe gfp_mask = saved_gfp; 5018aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 502f9c78b2bSJens Axboe } 503f9c78b2bSJens Axboe 504f9c78b2bSJens Axboe if (unlikely(!bvl)) 505f9c78b2bSJens Axboe goto err_free; 506f9c78b2bSJens Axboe 507ed996a52SChristoph Hellwig bio->bi_flags |= idx << BVEC_POOL_OFFSET; 508f9c78b2bSJens Axboe } else if (nr_iovecs) { 509f9c78b2bSJens Axboe bvl = bio->bi_inline_vecs; 510f9c78b2bSJens Axboe } 511f9c78b2bSJens Axboe 512f9c78b2bSJens Axboe bio->bi_pool = bs; 513f9c78b2bSJens Axboe bio->bi_max_vecs = nr_iovecs; 514f9c78b2bSJens Axboe bio->bi_io_vec = bvl; 515f9c78b2bSJens Axboe return bio; 516f9c78b2bSJens Axboe 517f9c78b2bSJens Axboe err_free: 5188aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 519f9c78b2bSJens Axboe return NULL; 520f9c78b2bSJens Axboe } 521f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 522f9c78b2bSJens Axboe 52338a72dacSKent Overstreet void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 524f9c78b2bSJens Axboe { 525f9c78b2bSJens Axboe unsigned long flags; 526f9c78b2bSJens Axboe struct bio_vec bv; 527f9c78b2bSJens Axboe struct bvec_iter iter; 528f9c78b2bSJens Axboe 52938a72dacSKent Overstreet __bio_for_each_segment(bv, bio, iter, start) { 530f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 531f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 532f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 533f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 534f9c78b2bSJens Axboe } 535f9c78b2bSJens Axboe } 53638a72dacSKent Overstreet EXPORT_SYMBOL(zero_fill_bio_iter); 537f9c78b2bSJens Axboe 538f9c78b2bSJens Axboe /** 539f9c78b2bSJens Axboe * bio_put - release a reference to a bio 540f9c78b2bSJens Axboe * @bio: bio to release reference to 541f9c78b2bSJens Axboe * 542f9c78b2bSJens Axboe * Description: 543f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 5449b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 545f9c78b2bSJens Axboe **/ 546f9c78b2bSJens Axboe void bio_put(struct bio *bio) 547f9c78b2bSJens Axboe { 548dac56212SJens Axboe if (!bio_flagged(bio, BIO_REFFED)) 549dac56212SJens Axboe bio_free(bio); 550dac56212SJens Axboe else { 551dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 552f9c78b2bSJens Axboe 553f9c78b2bSJens Axboe /* 554f9c78b2bSJens Axboe * last put frees it 555f9c78b2bSJens Axboe */ 556dac56212SJens Axboe if (atomic_dec_and_test(&bio->__bi_cnt)) 557f9c78b2bSJens Axboe bio_free(bio); 558f9c78b2bSJens Axboe } 559dac56212SJens Axboe } 560f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 561f9c78b2bSJens Axboe 562f9c78b2bSJens Axboe /** 563f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 564f9c78b2bSJens Axboe * @bio: destination bio 565f9c78b2bSJens Axboe * @bio_src: bio to clone 566f9c78b2bSJens Axboe * 567f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 568f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 569f9c78b2bSJens Axboe * bio will be one. 570f9c78b2bSJens Axboe * 571f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 572f9c78b2bSJens Axboe */ 573f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 574f9c78b2bSJens Axboe { 575ed996a52SChristoph Hellwig BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 576f9c78b2bSJens Axboe 577f9c78b2bSJens Axboe /* 57874d46992SChristoph Hellwig * most users will be overriding ->bi_disk with a new target, 579f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 580f9c78b2bSJens Axboe */ 58174d46992SChristoph Hellwig bio->bi_disk = bio_src->bi_disk; 58262530ed8SMichael Lyle bio->bi_partno = bio_src->bi_partno; 583b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 584111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 585111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 5861eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 587ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 588cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 589f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 590f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 59120bd723eSPaolo Valente 592db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 593e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 594f9c78b2bSJens Axboe } 595f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 596f9c78b2bSJens Axboe 597f9c78b2bSJens Axboe /** 598f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 599f9c78b2bSJens Axboe * @bio: bio to clone 600f9c78b2bSJens Axboe * @gfp_mask: allocation priority 601f9c78b2bSJens Axboe * @bs: bio_set to allocate from 602f9c78b2bSJens Axboe * 603f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 604f9c78b2bSJens Axboe */ 605f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 606f9c78b2bSJens Axboe { 607f9c78b2bSJens Axboe struct bio *b; 608f9c78b2bSJens Axboe 609f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 610f9c78b2bSJens Axboe if (!b) 611f9c78b2bSJens Axboe return NULL; 612f9c78b2bSJens Axboe 613f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 614f9c78b2bSJens Axboe 615f9c78b2bSJens Axboe if (bio_integrity(bio)) { 616f9c78b2bSJens Axboe int ret; 617f9c78b2bSJens Axboe 618f9c78b2bSJens Axboe ret = bio_integrity_clone(b, bio, gfp_mask); 619f9c78b2bSJens Axboe 620f9c78b2bSJens Axboe if (ret < 0) { 621f9c78b2bSJens Axboe bio_put(b); 622f9c78b2bSJens Axboe return NULL; 623f9c78b2bSJens Axboe } 624f9c78b2bSJens Axboe } 625f9c78b2bSJens Axboe 626f9c78b2bSJens Axboe return b; 627f9c78b2bSJens Axboe } 628f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 629f9c78b2bSJens Axboe 6305919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 6315919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 632ff896738SChristoph Hellwig bool *same_page) 6335919482eSMing Lei { 6345919482eSMing Lei phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 6355919482eSMing Lei bv->bv_offset + bv->bv_len - 1; 6365919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 6375919482eSMing Lei 6385919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 6395919482eSMing Lei return false; 6405919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 6415919482eSMing Lei return false; 64252d52d1cSChristoph Hellwig 643ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 644ff896738SChristoph Hellwig if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) 6455919482eSMing Lei return false; 6465919482eSMing Lei return true; 6475919482eSMing Lei } 6485919482eSMing Lei 649384209cdSChristoph Hellwig static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio, 650384209cdSChristoph Hellwig struct page *page, unsigned len, unsigned offset, 651384209cdSChristoph Hellwig bool *same_page) 652489fbbcbSMing Lei { 653384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 654489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 655489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 656489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 657489fbbcbSMing Lei 658489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 659489fbbcbSMing Lei return false; 660489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 661489fbbcbSMing Lei return false; 662384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 663489fbbcbSMing Lei } 664489fbbcbSMing Lei 665f4595875SShaohua Li /** 66619047087SMing Lei * __bio_add_pc_page - attempt to add page to passthrough bio 667c66a14d0SKent Overstreet * @q: the target queue 668c66a14d0SKent Overstreet * @bio: destination bio 669c66a14d0SKent Overstreet * @page: page to add 670c66a14d0SKent Overstreet * @len: vec entry length 671c66a14d0SKent Overstreet * @offset: vec entry offset 672d1916c86SChristoph Hellwig * @same_page: return if the merge happen inside the same page 673f9c78b2bSJens Axboe * 674c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This can fail for a 675c66a14d0SKent Overstreet * number of reasons, such as the bio being full or target block device 676c66a14d0SKent Overstreet * limitations. The target block device must allow bio's up to PAGE_SIZE, 677c66a14d0SKent Overstreet * so it is always possible to add a single page to an empty bio. 678c66a14d0SKent Overstreet * 6795a8ce240SMing Lei * This should only be used by passthrough bios. 680f9c78b2bSJens Axboe */ 6814713839dSChristoph Hellwig static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, 68219047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 683d1916c86SChristoph Hellwig bool *same_page) 684f9c78b2bSJens Axboe { 685f9c78b2bSJens Axboe struct bio_vec *bvec; 686f9c78b2bSJens Axboe 687f9c78b2bSJens Axboe /* 688f9c78b2bSJens Axboe * cloned bio must not modify vec list 689f9c78b2bSJens Axboe */ 690f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 691f9c78b2bSJens Axboe return 0; 692f9c78b2bSJens Axboe 693c66a14d0SKent Overstreet if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 694f9c78b2bSJens Axboe return 0; 695f9c78b2bSJens Axboe 696f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 697d1916c86SChristoph Hellwig if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page)) 698384209cdSChristoph Hellwig return len; 699320ea869SChristoph Hellwig 700320ea869SChristoph Hellwig /* 701320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 702320ea869SChristoph Hellwig * would create a gap, disallow it. 703320ea869SChristoph Hellwig */ 704384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 705320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 706320ea869SChristoph Hellwig return 0; 707f9c78b2bSJens Axboe } 708f9c78b2bSJens Axboe 70979d08f89SMing Lei if (bio_full(bio, len)) 710f9c78b2bSJens Axboe return 0; 711f9c78b2bSJens Axboe 71214ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 713489fbbcbSMing Lei return 0; 714489fbbcbSMing Lei 715f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 716f9c78b2bSJens Axboe bvec->bv_page = page; 717f9c78b2bSJens Axboe bvec->bv_len = len; 718f9c78b2bSJens Axboe bvec->bv_offset = offset; 719fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 720dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 721f9c78b2bSJens Axboe return len; 722f9c78b2bSJens Axboe } 72319047087SMing Lei 72419047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 72519047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 72619047087SMing Lei { 727d1916c86SChristoph Hellwig bool same_page = false; 728d1916c86SChristoph Hellwig return __bio_add_pc_page(q, bio, page, len, offset, &same_page); 72919047087SMing Lei } 730f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 731f9c78b2bSJens Axboe 732f9c78b2bSJens Axboe /** 7330aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 7340aa69fd3SChristoph Hellwig * @bio: destination bio 735551879a4SMing Lei * @page: start page to add 7360aa69fd3SChristoph Hellwig * @len: length of the data to add 737551879a4SMing Lei * @off: offset of the data relative to @page 738ff896738SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 7390aa69fd3SChristoph Hellwig * 7400aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 7410aa69fd3SChristoph Hellwig * a useful optimisation for file systems with a block size smaller than the 7420aa69fd3SChristoph Hellwig * page size. 7430aa69fd3SChristoph Hellwig * 744551879a4SMing Lei * Warn if (@len, @off) crosses pages in case that @same_page is true. 745551879a4SMing Lei * 7460aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 7470aa69fd3SChristoph Hellwig */ 7480aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 749ff896738SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 7500aa69fd3SChristoph Hellwig { 7510aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 7520aa69fd3SChristoph Hellwig return false; 7530aa69fd3SChristoph Hellwig 7540aa69fd3SChristoph Hellwig if (bio->bi_vcnt > 0) { 7550aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 7560aa69fd3SChristoph Hellwig 7575919482eSMing Lei if (page_is_mergeable(bv, page, len, off, same_page)) { 7580aa69fd3SChristoph Hellwig bv->bv_len += len; 7590aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 7600aa69fd3SChristoph Hellwig return true; 7610aa69fd3SChristoph Hellwig } 7625919482eSMing Lei } 7630aa69fd3SChristoph Hellwig return false; 7640aa69fd3SChristoph Hellwig } 7650aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 7660aa69fd3SChristoph Hellwig 7670aa69fd3SChristoph Hellwig /** 768551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 7690aa69fd3SChristoph Hellwig * @bio: destination bio 770551879a4SMing Lei * @page: start page to add 771551879a4SMing Lei * @len: length of the data to add, may cross pages 772551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 7730aa69fd3SChristoph Hellwig * 7740aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 7750aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 7760aa69fd3SChristoph Hellwig */ 7770aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 7780aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 7790aa69fd3SChristoph Hellwig { 7800aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 7810aa69fd3SChristoph Hellwig 7820aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 78379d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 7840aa69fd3SChristoph Hellwig 7850aa69fd3SChristoph Hellwig bv->bv_page = page; 7860aa69fd3SChristoph Hellwig bv->bv_offset = off; 7870aa69fd3SChristoph Hellwig bv->bv_len = len; 7880aa69fd3SChristoph Hellwig 7890aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 7900aa69fd3SChristoph Hellwig bio->bi_vcnt++; 791b8e24a93SJohannes Weiner 792b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 793b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 7940aa69fd3SChristoph Hellwig } 7950aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 7960aa69fd3SChristoph Hellwig 7970aa69fd3SChristoph Hellwig /** 798551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 799f9c78b2bSJens Axboe * @bio: destination bio 800551879a4SMing Lei * @page: start page to add 801551879a4SMing Lei * @len: vec entry length, may cross pages 802551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 803f9c78b2bSJens Axboe * 804551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 805c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 806f9c78b2bSJens Axboe */ 807c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 808c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 809f9c78b2bSJens Axboe { 810ff896738SChristoph Hellwig bool same_page = false; 811ff896738SChristoph Hellwig 812ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 81379d08f89SMing Lei if (bio_full(bio, len)) 814c66a14d0SKent Overstreet return 0; 8150aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 816c66a14d0SKent Overstreet } 817c66a14d0SKent Overstreet return len; 818f9c78b2bSJens Axboe } 819f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 820f9c78b2bSJens Axboe 821d241a95fSChristoph Hellwig void bio_release_pages(struct bio *bio, bool mark_dirty) 8227321ecbfSChristoph Hellwig { 8237321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 8247321ecbfSChristoph Hellwig struct bio_vec *bvec; 8257321ecbfSChristoph Hellwig 826b2d0d991SChristoph Hellwig if (bio_flagged(bio, BIO_NO_PAGE_REF)) 827b2d0d991SChristoph Hellwig return; 828b2d0d991SChristoph Hellwig 829d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 830d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 831d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 8327321ecbfSChristoph Hellwig put_page(bvec->bv_page); 8337321ecbfSChristoph Hellwig } 834d241a95fSChristoph Hellwig } 8357321ecbfSChristoph Hellwig 8366d0c48aeSJens Axboe static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 8376d0c48aeSJens Axboe { 8386d0c48aeSJens Axboe const struct bio_vec *bv = iter->bvec; 8396d0c48aeSJens Axboe unsigned int len; 8406d0c48aeSJens Axboe size_t size; 8416d0c48aeSJens Axboe 8426d0c48aeSJens Axboe if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 8436d0c48aeSJens Axboe return -EINVAL; 8446d0c48aeSJens Axboe 8456d0c48aeSJens Axboe len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 8466d0c48aeSJens Axboe size = bio_add_page(bio, bv->bv_page, len, 8476d0c48aeSJens Axboe bv->bv_offset + iter->iov_offset); 848a10584c3SChristoph Hellwig if (unlikely(size != len)) 849a10584c3SChristoph Hellwig return -EINVAL; 8506d0c48aeSJens Axboe iov_iter_advance(iter, size); 8516d0c48aeSJens Axboe return 0; 8526d0c48aeSJens Axboe } 8536d0c48aeSJens Axboe 854576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 855576ed913SChristoph Hellwig 8562cefe4dbSKent Overstreet /** 85717d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 8582cefe4dbSKent Overstreet * @bio: bio to add pages to 8592cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 8602cefe4dbSKent Overstreet * 86117d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 8622cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 86317d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 86417d51b10SMartin Wilck * the next non-empty segment of the iov iterator. 8652cefe4dbSKent Overstreet */ 86617d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 8672cefe4dbSKent Overstreet { 868576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 869576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 8702cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 8712cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 87245691804SChristoph Hellwig bool same_page = false; 873576ed913SChristoph Hellwig ssize_t size, left; 874576ed913SChristoph Hellwig unsigned len, i; 875b403ea24SMartin Wilck size_t offset; 876576ed913SChristoph Hellwig 877576ed913SChristoph Hellwig /* 878576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 879576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 880576ed913SChristoph Hellwig * without overwriting the temporary page array. 881576ed913SChristoph Hellwig */ 882576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 883576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 8842cefe4dbSKent Overstreet 8852cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 8862cefe4dbSKent Overstreet if (unlikely(size <= 0)) 8872cefe4dbSKent Overstreet return size ? size : -EFAULT; 8882cefe4dbSKent Overstreet 889576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 890576ed913SChristoph Hellwig struct page *page = pages[i]; 8912cefe4dbSKent Overstreet 892576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 89345691804SChristoph Hellwig 89445691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 89545691804SChristoph Hellwig if (same_page) 89645691804SChristoph Hellwig put_page(page); 89745691804SChristoph Hellwig } else { 89879d08f89SMing Lei if (WARN_ON_ONCE(bio_full(bio, len))) 899576ed913SChristoph Hellwig return -EINVAL; 90045691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 90145691804SChristoph Hellwig } 902576ed913SChristoph Hellwig offset = 0; 9032cefe4dbSKent Overstreet } 9042cefe4dbSKent Overstreet 9052cefe4dbSKent Overstreet iov_iter_advance(iter, size); 9062cefe4dbSKent Overstreet return 0; 9072cefe4dbSKent Overstreet } 90817d51b10SMartin Wilck 90917d51b10SMartin Wilck /** 9106d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 91117d51b10SMartin Wilck * @bio: bio to add pages to 9126d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 91317d51b10SMartin Wilck * 9146d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 9156d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 9166d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 917399254aaSJens Axboe * pages. If we're adding kernel pages, and the caller told us it's safe to 918399254aaSJens Axboe * do so, we just have to add the pages to the bio directly. We don't grab an 919399254aaSJens Axboe * extra reference to those pages (the user should already have that), and we 920399254aaSJens Axboe * don't put the page on IO completion. The caller needs to check if the bio is 921399254aaSJens Axboe * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 922399254aaSJens Axboe * released. 9236d0c48aeSJens Axboe * 92417d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 9256d0c48aeSJens Axboe * fit into the bio, or are requested in *iter, whatever is smaller. If 9266d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 9276d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 92817d51b10SMartin Wilck */ 92917d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 93017d51b10SMartin Wilck { 9316d0c48aeSJens Axboe const bool is_bvec = iov_iter_is_bvec(iter); 93214eacf12SChristoph Hellwig int ret; 93314eacf12SChristoph Hellwig 93414eacf12SChristoph Hellwig if (WARN_ON_ONCE(bio->bi_vcnt)) 93514eacf12SChristoph Hellwig return -EINVAL; 93617d51b10SMartin Wilck 93717d51b10SMartin Wilck do { 9386d0c48aeSJens Axboe if (is_bvec) 9396d0c48aeSJens Axboe ret = __bio_iov_bvec_add_pages(bio, iter); 9406d0c48aeSJens Axboe else 9416d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 94279d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 94317d51b10SMartin Wilck 944b6207430SChristoph Hellwig if (is_bvec) 9457321ecbfSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 94614eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 94717d51b10SMartin Wilck } 9482cefe4dbSKent Overstreet 9494246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 950f9c78b2bSJens Axboe { 95165e53aabSChristoph Hellwig complete(bio->bi_private); 952f9c78b2bSJens Axboe } 953f9c78b2bSJens Axboe 954f9c78b2bSJens Axboe /** 955f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 956f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 957f9c78b2bSJens Axboe * 958f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 959f9c78b2bSJens Axboe * bio_endio() on failure. 9603d289d68SJan Kara * 9613d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 9623d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 9633d289d68SJan Kara * on his own. 964f9c78b2bSJens Axboe */ 9654e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 966f9c78b2bSJens Axboe { 967e319e1fbSByungchul Park DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 968f9c78b2bSJens Axboe 96965e53aabSChristoph Hellwig bio->bi_private = &done; 970f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 9711eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 9724e49ea4aSMike Christie submit_bio(bio); 97365e53aabSChristoph Hellwig wait_for_completion_io(&done); 974f9c78b2bSJens Axboe 97565e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 976f9c78b2bSJens Axboe } 977f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 978f9c78b2bSJens Axboe 979f9c78b2bSJens Axboe /** 980f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 981f9c78b2bSJens Axboe * @bio: bio to advance 982f9c78b2bSJens Axboe * @bytes: number of bytes to complete 983f9c78b2bSJens Axboe * 984f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 985f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 986f9c78b2bSJens Axboe * be updated on the last bvec as well. 987f9c78b2bSJens Axboe * 988f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 989f9c78b2bSJens Axboe */ 990f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 991f9c78b2bSJens Axboe { 992f9c78b2bSJens Axboe if (bio_integrity(bio)) 993f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 994f9c78b2bSJens Axboe 995f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 996f9c78b2bSJens Axboe } 997f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 998f9c78b2bSJens Axboe 99945db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 100045db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1001f9c78b2bSJens Axboe { 1002f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1003f9c78b2bSJens Axboe void *src_p, *dst_p; 1004f9c78b2bSJens Axboe unsigned bytes; 1005f9c78b2bSJens Axboe 100645db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 100745db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 100845db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 100945db54d5SKent Overstreet 101045db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 101145db54d5SKent Overstreet 101245db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 101345db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 101445db54d5SKent Overstreet 101545db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 101645db54d5SKent Overstreet src_p + src_bv.bv_offset, 101745db54d5SKent Overstreet bytes); 101845db54d5SKent Overstreet 101945db54d5SKent Overstreet kunmap_atomic(dst_p); 102045db54d5SKent Overstreet kunmap_atomic(src_p); 102145db54d5SKent Overstreet 10226e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 10236e6e811dSKent Overstreet 102445db54d5SKent Overstreet bio_advance_iter(src, src_iter, bytes); 102545db54d5SKent Overstreet bio_advance_iter(dst, dst_iter, bytes); 102645db54d5SKent Overstreet } 102745db54d5SKent Overstreet } 102845db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 102945db54d5SKent Overstreet 103045db54d5SKent Overstreet /** 103145db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 103245db54d5SKent Overstreet * @src: source bio 103345db54d5SKent Overstreet * @dst: destination bio 103445db54d5SKent Overstreet * 103545db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 103645db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 103745db54d5SKent Overstreet */ 103845db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 103945db54d5SKent Overstreet { 104045db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 104145db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 104245db54d5SKent Overstreet 104345db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 104445db54d5SKent Overstreet } 104545db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 104645db54d5SKent Overstreet 104745db54d5SKent Overstreet /** 104845db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 104945db54d5SKent Overstreet * another 105045db54d5SKent Overstreet * @src: source bio list 105145db54d5SKent Overstreet * @dst: destination bio list 105245db54d5SKent Overstreet * 105345db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 105445db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 105545db54d5SKent Overstreet * bios). 105645db54d5SKent Overstreet */ 105745db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 105845db54d5SKent Overstreet { 105945db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 106045db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 106145db54d5SKent Overstreet 1062f9c78b2bSJens Axboe while (1) { 1063f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1064f9c78b2bSJens Axboe src = src->bi_next; 1065f9c78b2bSJens Axboe if (!src) 1066f9c78b2bSJens Axboe break; 1067f9c78b2bSJens Axboe 1068f9c78b2bSJens Axboe src_iter = src->bi_iter; 1069f9c78b2bSJens Axboe } 1070f9c78b2bSJens Axboe 1071f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1072f9c78b2bSJens Axboe dst = dst->bi_next; 1073f9c78b2bSJens Axboe if (!dst) 1074f9c78b2bSJens Axboe break; 1075f9c78b2bSJens Axboe 1076f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1077f9c78b2bSJens Axboe } 1078f9c78b2bSJens Axboe 107945db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1080f9c78b2bSJens Axboe } 1081f9c78b2bSJens Axboe } 108245db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1083f9c78b2bSJens Axboe 1084f9c78b2bSJens Axboe struct bio_map_data { 1085f9c78b2bSJens Axboe int is_our_pages; 108626e49cfcSKent Overstreet struct iov_iter iter; 108726e49cfcSKent Overstreet struct iovec iov[]; 1088f9c78b2bSJens Axboe }; 1089f9c78b2bSJens Axboe 10900e5b935dSAl Viro static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1091f9c78b2bSJens Axboe gfp_t gfp_mask) 1092f9c78b2bSJens Axboe { 10930e5b935dSAl Viro struct bio_map_data *bmd; 10940e5b935dSAl Viro if (data->nr_segs > UIO_MAXIOV) 1095f9c78b2bSJens Axboe return NULL; 1096f9c78b2bSJens Axboe 1097f1f8f292SGustavo A. R. Silva bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); 10980e5b935dSAl Viro if (!bmd) 10990e5b935dSAl Viro return NULL; 11000e5b935dSAl Viro memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 11010e5b935dSAl Viro bmd->iter = *data; 11020e5b935dSAl Viro bmd->iter.iov = bmd->iov; 11030e5b935dSAl Viro return bmd; 1104f9c78b2bSJens Axboe } 1105f9c78b2bSJens Axboe 11069124d3feSDongsu Park /** 11079124d3feSDongsu Park * bio_copy_from_iter - copy all pages from iov_iter to bio 11089124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as destination 11099124d3feSDongsu Park * @iter: iov_iter as source 11109124d3feSDongsu Park * 11119124d3feSDongsu Park * Copy all pages from iov_iter to bio. 11129124d3feSDongsu Park * Returns 0 on success, or error on failure. 11139124d3feSDongsu Park */ 111498a09d61SAl Viro static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1115f9c78b2bSJens Axboe { 1116f9c78b2bSJens Axboe struct bio_vec *bvec; 11176dc4f100SMing Lei struct bvec_iter_all iter_all; 1118f9c78b2bSJens Axboe 11192b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 11209124d3feSDongsu Park ssize_t ret; 1121f9c78b2bSJens Axboe 11229124d3feSDongsu Park ret = copy_page_from_iter(bvec->bv_page, 11239124d3feSDongsu Park bvec->bv_offset, 11249124d3feSDongsu Park bvec->bv_len, 112598a09d61SAl Viro iter); 1126f9c78b2bSJens Axboe 112798a09d61SAl Viro if (!iov_iter_count(iter)) 11289124d3feSDongsu Park break; 1129f9c78b2bSJens Axboe 11309124d3feSDongsu Park if (ret < bvec->bv_len) 11319124d3feSDongsu Park return -EFAULT; 1132f9c78b2bSJens Axboe } 1133f9c78b2bSJens Axboe 11349124d3feSDongsu Park return 0; 1135f9c78b2bSJens Axboe } 1136f9c78b2bSJens Axboe 11379124d3feSDongsu Park /** 11389124d3feSDongsu Park * bio_copy_to_iter - copy all pages from bio to iov_iter 11399124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as source 11409124d3feSDongsu Park * @iter: iov_iter as destination 11419124d3feSDongsu Park * 11429124d3feSDongsu Park * Copy all pages from bio to iov_iter. 11439124d3feSDongsu Park * Returns 0 on success, or error on failure. 11449124d3feSDongsu Park */ 11459124d3feSDongsu Park static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 11469124d3feSDongsu Park { 11479124d3feSDongsu Park struct bio_vec *bvec; 11486dc4f100SMing Lei struct bvec_iter_all iter_all; 11499124d3feSDongsu Park 11502b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 11519124d3feSDongsu Park ssize_t ret; 11529124d3feSDongsu Park 11539124d3feSDongsu Park ret = copy_page_to_iter(bvec->bv_page, 11549124d3feSDongsu Park bvec->bv_offset, 11559124d3feSDongsu Park bvec->bv_len, 11569124d3feSDongsu Park &iter); 11579124d3feSDongsu Park 11589124d3feSDongsu Park if (!iov_iter_count(&iter)) 11599124d3feSDongsu Park break; 11609124d3feSDongsu Park 11619124d3feSDongsu Park if (ret < bvec->bv_len) 11629124d3feSDongsu Park return -EFAULT; 11639124d3feSDongsu Park } 11649124d3feSDongsu Park 11659124d3feSDongsu Park return 0; 1166f9c78b2bSJens Axboe } 1167f9c78b2bSJens Axboe 1168491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 11691dfa0f68SChristoph Hellwig { 11701dfa0f68SChristoph Hellwig struct bio_vec *bvec; 11716dc4f100SMing Lei struct bvec_iter_all iter_all; 11721dfa0f68SChristoph Hellwig 11732b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 11741dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 11751dfa0f68SChristoph Hellwig } 1176491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 11771dfa0f68SChristoph Hellwig 1178f9c78b2bSJens Axboe /** 1179f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1180f9c78b2bSJens Axboe * @bio: bio being terminated 1181f9c78b2bSJens Axboe * 1182ddad8dd0SChristoph Hellwig * Free pages allocated from bio_copy_user_iov() and write back data 1183f9c78b2bSJens Axboe * to user space in case of a read. 1184f9c78b2bSJens Axboe */ 1185f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1186f9c78b2bSJens Axboe { 1187f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 11881dfa0f68SChristoph Hellwig int ret = 0; 1189f9c78b2bSJens Axboe 1190f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1191f9c78b2bSJens Axboe /* 1192f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 11932d99b55dSHannes Reinecke * don't copy into a random user address space, just free 11942d99b55dSHannes Reinecke * and return -EINTR so user space doesn't expect any data. 1195f9c78b2bSJens Axboe */ 11962d99b55dSHannes Reinecke if (!current->mm) 11972d99b55dSHannes Reinecke ret = -EINTR; 11982d99b55dSHannes Reinecke else if (bio_data_dir(bio) == READ) 11999124d3feSDongsu Park ret = bio_copy_to_iter(bio, bmd->iter); 12001dfa0f68SChristoph Hellwig if (bmd->is_our_pages) 12011dfa0f68SChristoph Hellwig bio_free_pages(bio); 1202f9c78b2bSJens Axboe } 1203f9c78b2bSJens Axboe kfree(bmd); 1204f9c78b2bSJens Axboe bio_put(bio); 1205f9c78b2bSJens Axboe return ret; 1206f9c78b2bSJens Axboe } 1207f9c78b2bSJens Axboe 1208f9c78b2bSJens Axboe /** 1209f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1210f9c78b2bSJens Axboe * @q: destination block queue 1211f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 121226e49cfcSKent Overstreet * @iter: iovec iterator 1213f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1214f9c78b2bSJens Axboe * 1215f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1216f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1217f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1218f9c78b2bSJens Axboe */ 1219f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1220f9c78b2bSJens Axboe struct rq_map_data *map_data, 1221e81cef5dSAl Viro struct iov_iter *iter, 122226e49cfcSKent Overstreet gfp_t gfp_mask) 1223f9c78b2bSJens Axboe { 1224f9c78b2bSJens Axboe struct bio_map_data *bmd; 1225f9c78b2bSJens Axboe struct page *page; 1226f9c78b2bSJens Axboe struct bio *bio; 1227d16d44ebSAl Viro int i = 0, ret; 1228d16d44ebSAl Viro int nr_pages; 122926e49cfcSKent Overstreet unsigned int len = iter->count; 1230bd5ceceaSGeliang Tang unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1231f9c78b2bSJens Axboe 12320e5b935dSAl Viro bmd = bio_alloc_map_data(iter, gfp_mask); 1233f9c78b2bSJens Axboe if (!bmd) 1234f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1235f9c78b2bSJens Axboe 123626e49cfcSKent Overstreet /* 123726e49cfcSKent Overstreet * We need to do a deep copy of the iov_iter including the iovecs. 123826e49cfcSKent Overstreet * The caller provided iov might point to an on-stack or otherwise 123926e49cfcSKent Overstreet * shortlived one. 124026e49cfcSKent Overstreet */ 124126e49cfcSKent Overstreet bmd->is_our_pages = map_data ? 0 : 1; 124226e49cfcSKent Overstreet 1243d16d44ebSAl Viro nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1244d16d44ebSAl Viro if (nr_pages > BIO_MAX_PAGES) 1245d16d44ebSAl Viro nr_pages = BIO_MAX_PAGES; 1246f9c78b2bSJens Axboe 1247f9c78b2bSJens Axboe ret = -ENOMEM; 1248f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1249f9c78b2bSJens Axboe if (!bio) 1250f9c78b2bSJens Axboe goto out_bmd; 1251f9c78b2bSJens Axboe 1252f9c78b2bSJens Axboe ret = 0; 1253f9c78b2bSJens Axboe 1254f9c78b2bSJens Axboe if (map_data) { 1255f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1256f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1257f9c78b2bSJens Axboe } 1258f9c78b2bSJens Axboe while (len) { 1259f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1260f9c78b2bSJens Axboe 1261f9c78b2bSJens Axboe bytes -= offset; 1262f9c78b2bSJens Axboe 1263f9c78b2bSJens Axboe if (bytes > len) 1264f9c78b2bSJens Axboe bytes = len; 1265f9c78b2bSJens Axboe 1266f9c78b2bSJens Axboe if (map_data) { 1267f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1268f9c78b2bSJens Axboe ret = -ENOMEM; 1269f9c78b2bSJens Axboe break; 1270f9c78b2bSJens Axboe } 1271f9c78b2bSJens Axboe 1272f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1273f9c78b2bSJens Axboe page += (i % nr_pages); 1274f9c78b2bSJens Axboe 1275f9c78b2bSJens Axboe i++; 1276f9c78b2bSJens Axboe } else { 1277f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1278f9c78b2bSJens Axboe if (!page) { 1279f9c78b2bSJens Axboe ret = -ENOMEM; 1280f9c78b2bSJens Axboe break; 1281f9c78b2bSJens Axboe } 1282f9c78b2bSJens Axboe } 1283f9c78b2bSJens Axboe 1284a3761c3cSJérôme Glisse if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { 1285a3761c3cSJérôme Glisse if (!map_data) 1286a3761c3cSJérôme Glisse __free_page(page); 1287f9c78b2bSJens Axboe break; 1288a3761c3cSJérôme Glisse } 1289f9c78b2bSJens Axboe 1290f9c78b2bSJens Axboe len -= bytes; 1291f9c78b2bSJens Axboe offset = 0; 1292f9c78b2bSJens Axboe } 1293f9c78b2bSJens Axboe 1294f9c78b2bSJens Axboe if (ret) 1295f9c78b2bSJens Axboe goto cleanup; 1296f9c78b2bSJens Axboe 12972884d0beSAl Viro if (map_data) 12982884d0beSAl Viro map_data->offset += bio->bi_iter.bi_size; 12992884d0beSAl Viro 1300f9c78b2bSJens Axboe /* 1301f9c78b2bSJens Axboe * success 1302f9c78b2bSJens Axboe */ 130300e23707SDavid Howells if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1304f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 130598a09d61SAl Viro ret = bio_copy_from_iter(bio, iter); 1306f9c78b2bSJens Axboe if (ret) 1307f9c78b2bSJens Axboe goto cleanup; 130898a09d61SAl Viro } else { 1309f55adad6SKeith Busch if (bmd->is_our_pages) 1310f3587d76SKeith Busch zero_fill_bio(bio); 1311e81cef5dSAl Viro iov_iter_advance(iter, bio->bi_iter.bi_size); 1312f9c78b2bSJens Axboe } 1313f9c78b2bSJens Axboe 131426e49cfcSKent Overstreet bio->bi_private = bmd; 13152884d0beSAl Viro if (map_data && map_data->null_mapped) 13162884d0beSAl Viro bio_set_flag(bio, BIO_NULL_MAPPED); 1317f9c78b2bSJens Axboe return bio; 1318f9c78b2bSJens Axboe cleanup: 1319f9c78b2bSJens Axboe if (!map_data) 13201dfa0f68SChristoph Hellwig bio_free_pages(bio); 1321f9c78b2bSJens Axboe bio_put(bio); 1322f9c78b2bSJens Axboe out_bmd: 1323f9c78b2bSJens Axboe kfree(bmd); 1324f9c78b2bSJens Axboe return ERR_PTR(ret); 1325f9c78b2bSJens Axboe } 1326f9c78b2bSJens Axboe 132737f19e57SChristoph Hellwig /** 132837f19e57SChristoph Hellwig * bio_map_user_iov - map user iovec into bio 132937f19e57SChristoph Hellwig * @q: the struct request_queue for the bio 133037f19e57SChristoph Hellwig * @iter: iovec iterator 133137f19e57SChristoph Hellwig * @gfp_mask: memory allocation flags 133237f19e57SChristoph Hellwig * 133337f19e57SChristoph Hellwig * Map the user space address into a bio suitable for io to a block 133437f19e57SChristoph Hellwig * device. Returns an error pointer in case of error. 133537f19e57SChristoph Hellwig */ 133637f19e57SChristoph Hellwig struct bio *bio_map_user_iov(struct request_queue *q, 1337e81cef5dSAl Viro struct iov_iter *iter, 133826e49cfcSKent Overstreet gfp_t gfp_mask) 1339f9c78b2bSJens Axboe { 134026e49cfcSKent Overstreet int j; 1341f9c78b2bSJens Axboe struct bio *bio; 1342076098e5SAl Viro int ret; 1343f9c78b2bSJens Axboe 1344b282cc76SAl Viro if (!iov_iter_count(iter)) 1345f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1346f9c78b2bSJens Axboe 1347b282cc76SAl Viro bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1348f9c78b2bSJens Axboe if (!bio) 1349f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1350f9c78b2bSJens Axboe 13510a0f1513SAl Viro while (iov_iter_count(iter)) { 1352629e42bcSAl Viro struct page **pages; 1353076098e5SAl Viro ssize_t bytes; 1354076098e5SAl Viro size_t offs, added = 0; 1355076098e5SAl Viro int npages; 1356f9c78b2bSJens Axboe 13570a0f1513SAl Viro bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1358076098e5SAl Viro if (unlikely(bytes <= 0)) { 1359076098e5SAl Viro ret = bytes ? bytes : -EFAULT; 1360f9c78b2bSJens Axboe goto out_unmap; 1361f9c78b2bSJens Axboe } 1362f9c78b2bSJens Axboe 1363076098e5SAl Viro npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1364076098e5SAl Viro 136598f0bc99SAl Viro if (unlikely(offs & queue_dma_alignment(q))) { 136698f0bc99SAl Viro ret = -EINVAL; 136798f0bc99SAl Viro j = 0; 136898f0bc99SAl Viro } else { 1369629e42bcSAl Viro for (j = 0; j < npages; j++) { 137098f0bc99SAl Viro struct page *page = pages[j]; 1371076098e5SAl Viro unsigned int n = PAGE_SIZE - offs; 1372d1916c86SChristoph Hellwig bool same_page = false; 1373f9c78b2bSJens Axboe 1374076098e5SAl Viro if (n > bytes) 1375076098e5SAl Viro n = bytes; 1376f9c78b2bSJens Axboe 137719047087SMing Lei if (!__bio_add_pc_page(q, bio, page, n, offs, 1378d1916c86SChristoph Hellwig &same_page)) { 1379d1916c86SChristoph Hellwig if (same_page) 1380d1916c86SChristoph Hellwig put_page(page); 1381f9c78b2bSJens Axboe break; 1382d1916c86SChristoph Hellwig } 1383f9c78b2bSJens Axboe 1384076098e5SAl Viro added += n; 1385076098e5SAl Viro bytes -= n; 1386076098e5SAl Viro offs = 0; 1387f9c78b2bSJens Axboe } 13880a0f1513SAl Viro iov_iter_advance(iter, added); 138998f0bc99SAl Viro } 1390f9c78b2bSJens Axboe /* 1391f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1392f9c78b2bSJens Axboe */ 1393629e42bcSAl Viro while (j < npages) 139409cbfeafSKirill A. Shutemov put_page(pages[j++]); 1395629e42bcSAl Viro kvfree(pages); 1396e2e115d1SAl Viro /* couldn't stuff something into bio? */ 1397e2e115d1SAl Viro if (bytes) 1398e2e115d1SAl Viro break; 1399f9c78b2bSJens Axboe } 1400f9c78b2bSJens Axboe 1401b7c44ed9SJens Axboe bio_set_flag(bio, BIO_USER_MAPPED); 140237f19e57SChristoph Hellwig 140337f19e57SChristoph Hellwig /* 14045fad1b64SBart Van Assche * subtle -- if bio_map_user_iov() ended up bouncing a bio, 140537f19e57SChristoph Hellwig * it would normally disappear when its bi_end_io is run. 140637f19e57SChristoph Hellwig * however, we need it for the unmap, so grab an extra 140737f19e57SChristoph Hellwig * reference to it 140837f19e57SChristoph Hellwig */ 140937f19e57SChristoph Hellwig bio_get(bio); 1410f9c78b2bSJens Axboe return bio; 1411f9c78b2bSJens Axboe 1412f9c78b2bSJens Axboe out_unmap: 1413506e0798SChristoph Hellwig bio_release_pages(bio, false); 1414f9c78b2bSJens Axboe bio_put(bio); 1415f9c78b2bSJens Axboe return ERR_PTR(ret); 1416f9c78b2bSJens Axboe } 1417f9c78b2bSJens Axboe 1418f9c78b2bSJens Axboe /** 1419f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1420f9c78b2bSJens Axboe * @bio: the bio being unmapped 1421f9c78b2bSJens Axboe * 14225fad1b64SBart Van Assche * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 14235fad1b64SBart Van Assche * process context. 1424f9c78b2bSJens Axboe * 1425f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1426f9c78b2bSJens Axboe */ 1427f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1428f9c78b2bSJens Axboe { 1429163cc2d3SChristoph Hellwig bio_release_pages(bio, bio_data_dir(bio) == READ); 1430163cc2d3SChristoph Hellwig bio_put(bio); 1431f9c78b2bSJens Axboe bio_put(bio); 1432f9c78b2bSJens Axboe } 1433f9c78b2bSJens Axboe 1434b4c5875dSDamien Le Moal static void bio_invalidate_vmalloc_pages(struct bio *bio) 1435b4c5875dSDamien Le Moal { 1436b4c5875dSDamien Le Moal #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE 1437b4c5875dSDamien Le Moal if (bio->bi_private && !op_is_write(bio_op(bio))) { 1438b4c5875dSDamien Le Moal unsigned long i, len = 0; 1439b4c5875dSDamien Le Moal 1440b4c5875dSDamien Le Moal for (i = 0; i < bio->bi_vcnt; i++) 1441b4c5875dSDamien Le Moal len += bio->bi_io_vec[i].bv_len; 1442b4c5875dSDamien Le Moal invalidate_kernel_vmap_range(bio->bi_private, len); 1443b4c5875dSDamien Le Moal } 1444b4c5875dSDamien Le Moal #endif 1445b4c5875dSDamien Le Moal } 1446b4c5875dSDamien Le Moal 14474246a0b6SChristoph Hellwig static void bio_map_kern_endio(struct bio *bio) 1448f9c78b2bSJens Axboe { 1449b4c5875dSDamien Le Moal bio_invalidate_vmalloc_pages(bio); 1450f9c78b2bSJens Axboe bio_put(bio); 1451f9c78b2bSJens Axboe } 1452f9c78b2bSJens Axboe 145375c72b83SChristoph Hellwig /** 145475c72b83SChristoph Hellwig * bio_map_kern - map kernel address into bio 145575c72b83SChristoph Hellwig * @q: the struct request_queue for the bio 145675c72b83SChristoph Hellwig * @data: pointer to buffer to map 145775c72b83SChristoph Hellwig * @len: length in bytes 145875c72b83SChristoph Hellwig * @gfp_mask: allocation flags for bio allocation 145975c72b83SChristoph Hellwig * 146075c72b83SChristoph Hellwig * Map the kernel address into a bio suitable for io to a block 146175c72b83SChristoph Hellwig * device. Returns an error pointer in case of error. 146275c72b83SChristoph Hellwig */ 146375c72b83SChristoph Hellwig struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 146475c72b83SChristoph Hellwig gfp_t gfp_mask) 1465f9c78b2bSJens Axboe { 1466f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1467f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1468f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1469f9c78b2bSJens Axboe const int nr_pages = end - start; 1470b4c5875dSDamien Le Moal bool is_vmalloc = is_vmalloc_addr(data); 1471b4c5875dSDamien Le Moal struct page *page; 1472f9c78b2bSJens Axboe int offset, i; 1473f9c78b2bSJens Axboe struct bio *bio; 1474f9c78b2bSJens Axboe 1475f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1476f9c78b2bSJens Axboe if (!bio) 1477f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1478f9c78b2bSJens Axboe 1479b4c5875dSDamien Le Moal if (is_vmalloc) { 1480b4c5875dSDamien Le Moal flush_kernel_vmap_range(data, len); 1481b4c5875dSDamien Le Moal bio->bi_private = data; 1482b4c5875dSDamien Le Moal } 1483b4c5875dSDamien Le Moal 1484f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1485f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1486f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1487f9c78b2bSJens Axboe 1488f9c78b2bSJens Axboe if (len <= 0) 1489f9c78b2bSJens Axboe break; 1490f9c78b2bSJens Axboe 1491f9c78b2bSJens Axboe if (bytes > len) 1492f9c78b2bSJens Axboe bytes = len; 1493f9c78b2bSJens Axboe 1494b4c5875dSDamien Le Moal if (!is_vmalloc) 1495b4c5875dSDamien Le Moal page = virt_to_page(data); 1496b4c5875dSDamien Le Moal else 1497b4c5875dSDamien Le Moal page = vmalloc_to_page(data); 1498b4c5875dSDamien Le Moal if (bio_add_pc_page(q, bio, page, bytes, 149975c72b83SChristoph Hellwig offset) < bytes) { 150075c72b83SChristoph Hellwig /* we don't support partial mappings */ 150175c72b83SChristoph Hellwig bio_put(bio); 150275c72b83SChristoph Hellwig return ERR_PTR(-EINVAL); 150375c72b83SChristoph Hellwig } 1504f9c78b2bSJens Axboe 1505f9c78b2bSJens Axboe data += bytes; 1506f9c78b2bSJens Axboe len -= bytes; 1507f9c78b2bSJens Axboe offset = 0; 1508f9c78b2bSJens Axboe } 1509f9c78b2bSJens Axboe 1510f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1511f9c78b2bSJens Axboe return bio; 1512f9c78b2bSJens Axboe } 1513f9c78b2bSJens Axboe 15144246a0b6SChristoph Hellwig static void bio_copy_kern_endio(struct bio *bio) 1515f9c78b2bSJens Axboe { 15161dfa0f68SChristoph Hellwig bio_free_pages(bio); 15171dfa0f68SChristoph Hellwig bio_put(bio); 15181dfa0f68SChristoph Hellwig } 15191dfa0f68SChristoph Hellwig 15204246a0b6SChristoph Hellwig static void bio_copy_kern_endio_read(struct bio *bio) 15211dfa0f68SChristoph Hellwig { 152242d2683aSChristoph Hellwig char *p = bio->bi_private; 15231dfa0f68SChristoph Hellwig struct bio_vec *bvec; 15246dc4f100SMing Lei struct bvec_iter_all iter_all; 1525f9c78b2bSJens Axboe 15262b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 15271dfa0f68SChristoph Hellwig memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1528f9c78b2bSJens Axboe p += bvec->bv_len; 1529f9c78b2bSJens Axboe } 1530f9c78b2bSJens Axboe 15314246a0b6SChristoph Hellwig bio_copy_kern_endio(bio); 1532f9c78b2bSJens Axboe } 1533f9c78b2bSJens Axboe 1534f9c78b2bSJens Axboe /** 1535f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1536f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1537f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1538f9c78b2bSJens Axboe * @len: length in bytes 1539f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1540f9c78b2bSJens Axboe * @reading: data direction is READ 1541f9c78b2bSJens Axboe * 1542f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1543f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1544f9c78b2bSJens Axboe */ 1545f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1546f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1547f9c78b2bSJens Axboe { 154842d2683aSChristoph Hellwig unsigned long kaddr = (unsigned long)data; 154942d2683aSChristoph Hellwig unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 155042d2683aSChristoph Hellwig unsigned long start = kaddr >> PAGE_SHIFT; 155142d2683aSChristoph Hellwig struct bio *bio; 1552f9c78b2bSJens Axboe void *p = data; 15531dfa0f68SChristoph Hellwig int nr_pages = 0; 1554f9c78b2bSJens Axboe 155542d2683aSChristoph Hellwig /* 155642d2683aSChristoph Hellwig * Overflow, abort 155742d2683aSChristoph Hellwig */ 155842d2683aSChristoph Hellwig if (end < start) 155942d2683aSChristoph Hellwig return ERR_PTR(-EINVAL); 1560f9c78b2bSJens Axboe 156142d2683aSChristoph Hellwig nr_pages = end - start; 156242d2683aSChristoph Hellwig bio = bio_kmalloc(gfp_mask, nr_pages); 156342d2683aSChristoph Hellwig if (!bio) 156442d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 156542d2683aSChristoph Hellwig 156642d2683aSChristoph Hellwig while (len) { 156742d2683aSChristoph Hellwig struct page *page; 156842d2683aSChristoph Hellwig unsigned int bytes = PAGE_SIZE; 156942d2683aSChristoph Hellwig 157042d2683aSChristoph Hellwig if (bytes > len) 157142d2683aSChristoph Hellwig bytes = len; 157242d2683aSChristoph Hellwig 157342d2683aSChristoph Hellwig page = alloc_page(q->bounce_gfp | gfp_mask); 157442d2683aSChristoph Hellwig if (!page) 157542d2683aSChristoph Hellwig goto cleanup; 157642d2683aSChristoph Hellwig 157742d2683aSChristoph Hellwig if (!reading) 157842d2683aSChristoph Hellwig memcpy(page_address(page), p, bytes); 157942d2683aSChristoph Hellwig 158042d2683aSChristoph Hellwig if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 158142d2683aSChristoph Hellwig break; 158242d2683aSChristoph Hellwig 158342d2683aSChristoph Hellwig len -= bytes; 158442d2683aSChristoph Hellwig p += bytes; 1585f9c78b2bSJens Axboe } 1586f9c78b2bSJens Axboe 15871dfa0f68SChristoph Hellwig if (reading) { 15881dfa0f68SChristoph Hellwig bio->bi_end_io = bio_copy_kern_endio_read; 158942d2683aSChristoph Hellwig bio->bi_private = data; 15901dfa0f68SChristoph Hellwig } else { 1591f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 15921dfa0f68SChristoph Hellwig } 15931dfa0f68SChristoph Hellwig 1594f9c78b2bSJens Axboe return bio; 159542d2683aSChristoph Hellwig 159642d2683aSChristoph Hellwig cleanup: 15971dfa0f68SChristoph Hellwig bio_free_pages(bio); 159842d2683aSChristoph Hellwig bio_put(bio); 159942d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 1600f9c78b2bSJens Axboe } 1601f9c78b2bSJens Axboe 1602f9c78b2bSJens Axboe /* 1603f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1604f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1605f9c78b2bSJens Axboe * 1606f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1607f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1608f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1609f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1610f9c78b2bSJens Axboe * in process context. 1611f9c78b2bSJens Axboe * 1612f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1613f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1614f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1615f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1616f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1617f9c78b2bSJens Axboe * 1618f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1619f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1620f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1621f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1622f9c78b2bSJens Axboe * pagecache. 1623f9c78b2bSJens Axboe * 1624f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1625f9c78b2bSJens Axboe * deferred bio dirtying paths. 1626f9c78b2bSJens Axboe */ 1627f9c78b2bSJens Axboe 1628f9c78b2bSJens Axboe /* 1629f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1630f9c78b2bSJens Axboe */ 1631f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1632f9c78b2bSJens Axboe { 1633f9c78b2bSJens Axboe struct bio_vec *bvec; 16346dc4f100SMing Lei struct bvec_iter_all iter_all; 1635f9c78b2bSJens Axboe 16362b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 16373bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 16383bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1639f9c78b2bSJens Axboe } 1640f9c78b2bSJens Axboe } 1641f9c78b2bSJens Axboe 1642f9c78b2bSJens Axboe /* 1643f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1644f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1645f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 164624d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1647f9c78b2bSJens Axboe * 1648f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1649ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1650ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1651f9c78b2bSJens Axboe */ 1652f9c78b2bSJens Axboe 1653f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1654f9c78b2bSJens Axboe 1655f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1656f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1657f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1658f9c78b2bSJens Axboe 1659f9c78b2bSJens Axboe /* 1660f9c78b2bSJens Axboe * This runs in process context 1661f9c78b2bSJens Axboe */ 1662f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1663f9c78b2bSJens Axboe { 166424d5493fSChristoph Hellwig struct bio *bio, *next; 1665f9c78b2bSJens Axboe 166624d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 166724d5493fSChristoph Hellwig next = bio_dirty_list; 1668f9c78b2bSJens Axboe bio_dirty_list = NULL; 166924d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1670f9c78b2bSJens Axboe 167124d5493fSChristoph Hellwig while ((bio = next) != NULL) { 167224d5493fSChristoph Hellwig next = bio->bi_private; 1673f9c78b2bSJens Axboe 1674d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1675f9c78b2bSJens Axboe bio_put(bio); 1676f9c78b2bSJens Axboe } 1677f9c78b2bSJens Axboe } 1678f9c78b2bSJens Axboe 1679f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1680f9c78b2bSJens Axboe { 1681f9c78b2bSJens Axboe struct bio_vec *bvec; 168224d5493fSChristoph Hellwig unsigned long flags; 16836dc4f100SMing Lei struct bvec_iter_all iter_all; 1684f9c78b2bSJens Axboe 16852b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 168624d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 168724d5493fSChristoph Hellwig goto defer; 1688f9c78b2bSJens Axboe } 1689f9c78b2bSJens Axboe 1690d241a95fSChristoph Hellwig bio_release_pages(bio, false); 169124d5493fSChristoph Hellwig bio_put(bio); 169224d5493fSChristoph Hellwig return; 169324d5493fSChristoph Hellwig defer: 1694f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1695f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1696f9c78b2bSJens Axboe bio_dirty_list = bio; 1697f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1698f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1699f9c78b2bSJens Axboe } 1700f9c78b2bSJens Axboe 17015b18b5a7SMikulas Patocka void update_io_ticks(struct hd_struct *part, unsigned long now) 17025b18b5a7SMikulas Patocka { 17035b18b5a7SMikulas Patocka unsigned long stamp; 17045b18b5a7SMikulas Patocka again: 17055b18b5a7SMikulas Patocka stamp = READ_ONCE(part->stamp); 17065b18b5a7SMikulas Patocka if (unlikely(stamp != now)) { 17075b18b5a7SMikulas Patocka if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 17085b18b5a7SMikulas Patocka __part_stat_add(part, io_ticks, 1); 17095b18b5a7SMikulas Patocka } 17105b18b5a7SMikulas Patocka } 17115b18b5a7SMikulas Patocka if (part->partno) { 17125b18b5a7SMikulas Patocka part = &part_to_disk(part)->part0; 17135b18b5a7SMikulas Patocka goto again; 17145b18b5a7SMikulas Patocka } 17155b18b5a7SMikulas Patocka } 1716f9c78b2bSJens Axboe 1717ddcf35d3SMichael Callahan void generic_start_io_acct(struct request_queue *q, int op, 1718d62e26b3SJens Axboe unsigned long sectors, struct hd_struct *part) 1719394ffa50SGu Zheng { 1720ddcf35d3SMichael Callahan const int sgrp = op_stat_group(op); 1721394ffa50SGu Zheng 1722112f158fSMike Snitzer part_stat_lock(); 1723112f158fSMike Snitzer 17245b18b5a7SMikulas Patocka update_io_ticks(part, jiffies); 1725112f158fSMike Snitzer part_stat_inc(part, ios[sgrp]); 1726112f158fSMike Snitzer part_stat_add(part, sectors[sgrp], sectors); 1727ddcf35d3SMichael Callahan part_inc_in_flight(q, part, op_is_write(op)); 1728394ffa50SGu Zheng 1729394ffa50SGu Zheng part_stat_unlock(); 1730394ffa50SGu Zheng } 1731394ffa50SGu Zheng EXPORT_SYMBOL(generic_start_io_acct); 1732394ffa50SGu Zheng 1733ddcf35d3SMichael Callahan void generic_end_io_acct(struct request_queue *q, int req_op, 1734d62e26b3SJens Axboe struct hd_struct *part, unsigned long start_time) 1735394ffa50SGu Zheng { 17365b18b5a7SMikulas Patocka unsigned long now = jiffies; 17375b18b5a7SMikulas Patocka unsigned long duration = now - start_time; 1738ddcf35d3SMichael Callahan const int sgrp = op_stat_group(req_op); 1739394ffa50SGu Zheng 1740112f158fSMike Snitzer part_stat_lock(); 1741112f158fSMike Snitzer 17425b18b5a7SMikulas Patocka update_io_ticks(part, now); 1743112f158fSMike Snitzer part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 17445b18b5a7SMikulas Patocka part_stat_add(part, time_in_queue, duration); 1745ddcf35d3SMichael Callahan part_dec_in_flight(q, part, op_is_write(req_op)); 1746394ffa50SGu Zheng 1747394ffa50SGu Zheng part_stat_unlock(); 1748394ffa50SGu Zheng } 1749394ffa50SGu Zheng EXPORT_SYMBOL(generic_end_io_acct); 1750394ffa50SGu Zheng 1751c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1752c4cf5261SJens Axboe { 1753c4cf5261SJens Axboe /* 1754c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1755c4cf5261SJens Axboe * we always end io on the first invocation. 1756c4cf5261SJens Axboe */ 1757c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1758c4cf5261SJens Axboe return true; 1759c4cf5261SJens Axboe 1760c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1761c4cf5261SJens Axboe 1762326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1763b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1764c4cf5261SJens Axboe return true; 1765326e1dbbSMike Snitzer } 1766c4cf5261SJens Axboe 1767c4cf5261SJens Axboe return false; 1768c4cf5261SJens Axboe } 1769c4cf5261SJens Axboe 1770f9c78b2bSJens Axboe /** 1771f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1772f9c78b2bSJens Axboe * @bio: bio 1773f9c78b2bSJens Axboe * 1774f9c78b2bSJens Axboe * Description: 17754246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 17764246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 17774246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1778fbbaf700SNeilBrown * 1779fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1780fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1781fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1782fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1783f9c78b2bSJens Axboe **/ 17844246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1785f9c78b2bSJens Axboe { 1786ba8c6967SChristoph Hellwig again: 17872b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1788ba8c6967SChristoph Hellwig return; 17897c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 17907c20f116SChristoph Hellwig return; 1791f9c78b2bSJens Axboe 179267b42d0bSJosef Bacik if (bio->bi_disk) 179367b42d0bSJosef Bacik rq_qos_done_bio(bio->bi_disk->queue, bio); 179467b42d0bSJosef Bacik 1795f9c78b2bSJens Axboe /* 1796ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1797ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1798ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1799ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1800ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1801ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1802f9c78b2bSJens Axboe */ 1803f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 180438f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1805ba8c6967SChristoph Hellwig goto again; 1806ba8c6967SChristoph Hellwig } 1807ba8c6967SChristoph Hellwig 180874d46992SChristoph Hellwig if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 180974d46992SChristoph Hellwig trace_block_bio_complete(bio->bi_disk->queue, bio, 1810a462b950SBart Van Assche blk_status_to_errno(bio->bi_status)); 1811fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1812fbbaf700SNeilBrown } 1813fbbaf700SNeilBrown 18149e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1815b222dd2fSShaohua Li /* release cgroup info */ 1816b222dd2fSShaohua Li bio_uninit(bio); 1817f9c78b2bSJens Axboe if (bio->bi_end_io) 18184246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1819f9c78b2bSJens Axboe } 1820f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1821f9c78b2bSJens Axboe 1822f9c78b2bSJens Axboe /** 1823f9c78b2bSJens Axboe * bio_split - split a bio 1824f9c78b2bSJens Axboe * @bio: bio to split 1825f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1826f9c78b2bSJens Axboe * @gfp: gfp mask 1827f9c78b2bSJens Axboe * @bs: bio set to allocate from 1828f9c78b2bSJens Axboe * 1829f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1830f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1831f9c78b2bSJens Axboe * 1832f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1833dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1834dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1835f9c78b2bSJens Axboe */ 1836f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1837f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1838f9c78b2bSJens Axboe { 1839f341a4d3SMikulas Patocka struct bio *split; 1840f9c78b2bSJens Axboe 1841f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1842f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1843f9c78b2bSJens Axboe 1844f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1845f9c78b2bSJens Axboe if (!split) 1846f9c78b2bSJens Axboe return NULL; 1847f9c78b2bSJens Axboe 1848f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1849f9c78b2bSJens Axboe 1850f9c78b2bSJens Axboe if (bio_integrity(split)) 1851fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1852f9c78b2bSJens Axboe 1853f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1854f9c78b2bSJens Axboe 1855fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 185620d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1857fbbaf700SNeilBrown 1858f9c78b2bSJens Axboe return split; 1859f9c78b2bSJens Axboe } 1860f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1861f9c78b2bSJens Axboe 1862f9c78b2bSJens Axboe /** 1863f9c78b2bSJens Axboe * bio_trim - trim a bio 1864f9c78b2bSJens Axboe * @bio: bio to trim 1865f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1866f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1867f9c78b2bSJens Axboe */ 1868f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1869f9c78b2bSJens Axboe { 1870f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1871f9c78b2bSJens Axboe * the given offset and size. 1872f9c78b2bSJens Axboe */ 1873f9c78b2bSJens Axboe 1874f9c78b2bSJens Axboe size <<= 9; 1875f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1876f9c78b2bSJens Axboe return; 1877f9c78b2bSJens Axboe 1878f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1879f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1880376a78abSDmitry Monakhov 1881376a78abSDmitry Monakhov if (bio_integrity(bio)) 1882fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1883376a78abSDmitry Monakhov 1884f9c78b2bSJens Axboe } 1885f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1886f9c78b2bSJens Axboe 1887f9c78b2bSJens Axboe /* 1888f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1889f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1890f9c78b2bSJens Axboe */ 18918aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1892f9c78b2bSJens Axboe { 1893ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1894f9c78b2bSJens Axboe 18958aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1896f9c78b2bSJens Axboe } 1897f9c78b2bSJens Axboe 1898917a38c7SKent Overstreet /* 1899917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1900917a38c7SKent Overstreet * 1901917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1902917a38c7SKent Overstreet * kzalloc()). 1903917a38c7SKent Overstreet */ 1904917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1905f9c78b2bSJens Axboe { 1906f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1907f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1908917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1909f9c78b2bSJens Axboe 19108aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 19118aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1912f9c78b2bSJens Axboe 1913f9c78b2bSJens Axboe bioset_integrity_free(bs); 1914917a38c7SKent Overstreet if (bs->bio_slab) 1915f9c78b2bSJens Axboe bio_put_slab(bs); 1916917a38c7SKent Overstreet bs->bio_slab = NULL; 1917917a38c7SKent Overstreet } 1918917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1919f9c78b2bSJens Axboe 1920011067b0SNeilBrown /** 1921917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1922dad08527SKent Overstreet * @bs: pool to initialize 1923917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1924917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1925917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1926917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1927917a38c7SKent Overstreet * 1928dad08527SKent Overstreet * Description: 1929dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1930dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1931dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1932dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1933dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1934dad08527SKent Overstreet * or things will break badly. 1935dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 1936dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 1937dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 1938dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 1939dad08527SKent Overstreet * 1940917a38c7SKent Overstreet */ 1941917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 1942917a38c7SKent Overstreet unsigned int pool_size, 1943917a38c7SKent Overstreet unsigned int front_pad, 1944917a38c7SKent Overstreet int flags) 1945917a38c7SKent Overstreet { 1946917a38c7SKent Overstreet unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 1947917a38c7SKent Overstreet 1948917a38c7SKent Overstreet bs->front_pad = front_pad; 1949917a38c7SKent Overstreet 1950917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 1951917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 1952917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 1953917a38c7SKent Overstreet 1954917a38c7SKent Overstreet bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 1955917a38c7SKent Overstreet if (!bs->bio_slab) 1956917a38c7SKent Overstreet return -ENOMEM; 1957917a38c7SKent Overstreet 1958917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 1959917a38c7SKent Overstreet goto bad; 1960917a38c7SKent Overstreet 1961917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 1962917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 1963917a38c7SKent Overstreet goto bad; 1964917a38c7SKent Overstreet 1965917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 1966917a38c7SKent Overstreet return 0; 1967917a38c7SKent Overstreet 1968917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 1969917a38c7SKent Overstreet if (!bs->rescue_workqueue) 1970917a38c7SKent Overstreet goto bad; 1971917a38c7SKent Overstreet 1972917a38c7SKent Overstreet return 0; 1973917a38c7SKent Overstreet bad: 1974917a38c7SKent Overstreet bioset_exit(bs); 1975917a38c7SKent Overstreet return -ENOMEM; 1976917a38c7SKent Overstreet } 1977917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 1978917a38c7SKent Overstreet 197928e89fd9SJens Axboe /* 198028e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 198128e89fd9SJens Axboe * another bio_set. 198228e89fd9SJens Axboe */ 198328e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 198428e89fd9SJens Axboe { 198528e89fd9SJens Axboe int flags; 198628e89fd9SJens Axboe 198728e89fd9SJens Axboe flags = 0; 198828e89fd9SJens Axboe if (src->bvec_pool.min_nr) 198928e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 199028e89fd9SJens Axboe if (src->rescue_workqueue) 199128e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 199228e89fd9SJens Axboe 199328e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 199428e89fd9SJens Axboe } 199528e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 199628e89fd9SJens Axboe 1997f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 19981d933cf0STejun Heo 19991d933cf0STejun Heo /** 20002268c0feSDennis Zhou * bio_disassociate_blkg - puts back the blkg reference if associated 2001b5f2954dSDennis Zhou * @bio: target bio 2002b5f2954dSDennis Zhou * 20032268c0feSDennis Zhou * Helper to disassociate the blkg from @bio if a blkg is associated. 2004b5f2954dSDennis Zhou */ 20052268c0feSDennis Zhou void bio_disassociate_blkg(struct bio *bio) 2006b5f2954dSDennis Zhou { 200708e18eabSJosef Bacik if (bio->bi_blkg) { 200808e18eabSJosef Bacik blkg_put(bio->bi_blkg); 200908e18eabSJosef Bacik bio->bi_blkg = NULL; 201008e18eabSJosef Bacik } 2011f9c78b2bSJens Axboe } 2012892ad71fSDennis Zhou EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2013f9c78b2bSJens Axboe 201420bd723eSPaolo Valente /** 20152268c0feSDennis Zhou * __bio_associate_blkg - associate a bio with the a blkg 2016f9c78b2bSJens Axboe * @bio: target bio 2017f9c78b2bSJens Axboe * @blkg: the blkg to associate 2018f9c78b2bSJens Axboe * 2019beea9da0SDennis Zhou * This tries to associate @bio with the specified @blkg. Association failure 2020beea9da0SDennis Zhou * is handled by walking up the blkg tree. Therefore, the blkg associated can 2021beea9da0SDennis Zhou * be anything between @blkg and the root_blkg. This situation only happens 2022beea9da0SDennis Zhou * when a cgroup is dying and then the remaining bios will spill to the closest 2023beea9da0SDennis Zhou * alive blkg. 2024beea9da0SDennis Zhou * 2025beea9da0SDennis Zhou * A reference will be taken on the @blkg and will be released when @bio is 2026beea9da0SDennis Zhou * freed. 2027f9c78b2bSJens Axboe */ 20282268c0feSDennis Zhou static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2029f9c78b2bSJens Axboe { 20302268c0feSDennis Zhou bio_disassociate_blkg(bio); 20312268c0feSDennis Zhou 20327754f669SDennis Zhou bio->bi_blkg = blkg_tryget_closest(blkg); 20332268c0feSDennis Zhou } 20342268c0feSDennis Zhou 2035fd42df30SDennis Zhou /** 2036fd42df30SDennis Zhou * bio_associate_blkg_from_css - associate a bio with a specified css 2037fd42df30SDennis Zhou * @bio: target bio 2038fd42df30SDennis Zhou * @css: target css 2039fd42df30SDennis Zhou * 2040fd42df30SDennis Zhou * Associate @bio with the blkg found by combining the css's blkg and the 2041fc5a828bSDennis Zhou * request_queue of the @bio. This falls back to the queue's root_blkg if 2042fc5a828bSDennis Zhou * the association fails with the css. 2043fd42df30SDennis Zhou */ 2044fd42df30SDennis Zhou void bio_associate_blkg_from_css(struct bio *bio, 2045fd42df30SDennis Zhou struct cgroup_subsys_state *css) 2046fd42df30SDennis Zhou { 2047fc5a828bSDennis Zhou struct request_queue *q = bio->bi_disk->queue; 2048fc5a828bSDennis Zhou struct blkcg_gq *blkg; 2049fc5a828bSDennis Zhou 2050fc5a828bSDennis Zhou rcu_read_lock(); 2051fc5a828bSDennis Zhou 2052fc5a828bSDennis Zhou if (!css || !css->parent) 2053fc5a828bSDennis Zhou blkg = q->root_blkg; 2054fc5a828bSDennis Zhou else 2055fc5a828bSDennis Zhou blkg = blkg_lookup_create(css_to_blkcg(css), q); 2056fc5a828bSDennis Zhou 2057fc5a828bSDennis Zhou __bio_associate_blkg(bio, blkg); 2058fc5a828bSDennis Zhou 2059fc5a828bSDennis Zhou rcu_read_unlock(); 2060fd42df30SDennis Zhou } 2061fd42df30SDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2062fd42df30SDennis Zhou 20636a7f6d86SDennis Zhou #ifdef CONFIG_MEMCG 20646a7f6d86SDennis Zhou /** 20656a7f6d86SDennis Zhou * bio_associate_blkg_from_page - associate a bio with the page's blkg 20666a7f6d86SDennis Zhou * @bio: target bio 20676a7f6d86SDennis Zhou * @page: the page to lookup the blkcg from 20686a7f6d86SDennis Zhou * 20696a7f6d86SDennis Zhou * Associate @bio with the blkg from @page's owning memcg and the respective 2070fc5a828bSDennis Zhou * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2071fc5a828bSDennis Zhou * root_blkg. 20726a7f6d86SDennis Zhou */ 20736a7f6d86SDennis Zhou void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 20746a7f6d86SDennis Zhou { 20756a7f6d86SDennis Zhou struct cgroup_subsys_state *css; 20766a7f6d86SDennis Zhou 20776a7f6d86SDennis Zhou if (!page->mem_cgroup) 20786a7f6d86SDennis Zhou return; 20796a7f6d86SDennis Zhou 2080fc5a828bSDennis Zhou rcu_read_lock(); 2081fc5a828bSDennis Zhou 2082fc5a828bSDennis Zhou css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2083fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 2084fc5a828bSDennis Zhou 2085fc5a828bSDennis Zhou rcu_read_unlock(); 20866a7f6d86SDennis Zhou } 20876a7f6d86SDennis Zhou #endif /* CONFIG_MEMCG */ 20886a7f6d86SDennis Zhou 20892268c0feSDennis Zhou /** 20902268c0feSDennis Zhou * bio_associate_blkg - associate a bio with a blkg 20912268c0feSDennis Zhou * @bio: target bio 20922268c0feSDennis Zhou * 20932268c0feSDennis Zhou * Associate @bio with the blkg found from the bio's css and request_queue. 20942268c0feSDennis Zhou * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 20952268c0feSDennis Zhou * already associated, the css is reused and association redone as the 20962268c0feSDennis Zhou * request_queue may have changed. 20972268c0feSDennis Zhou */ 20982268c0feSDennis Zhou void bio_associate_blkg(struct bio *bio) 20992268c0feSDennis Zhou { 2100fc5a828bSDennis Zhou struct cgroup_subsys_state *css; 21012268c0feSDennis Zhou 21022268c0feSDennis Zhou rcu_read_lock(); 21032268c0feSDennis Zhou 2104db6638d7SDennis Zhou if (bio->bi_blkg) 2105fc5a828bSDennis Zhou css = &bio_blkcg(bio)->css; 2106db6638d7SDennis Zhou else 2107fc5a828bSDennis Zhou css = blkcg_css(); 21082268c0feSDennis Zhou 2109fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 21102268c0feSDennis Zhou 21112268c0feSDennis Zhou rcu_read_unlock(); 2112f9c78b2bSJens Axboe } 21135cdf2e3fSDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg); 2114f9c78b2bSJens Axboe 211520bd723eSPaolo Valente /** 2116db6638d7SDennis Zhou * bio_clone_blkg_association - clone blkg association from src to dst bio 211720bd723eSPaolo Valente * @dst: destination bio 211820bd723eSPaolo Valente * @src: source bio 211920bd723eSPaolo Valente */ 2120db6638d7SDennis Zhou void bio_clone_blkg_association(struct bio *dst, struct bio *src) 212120bd723eSPaolo Valente { 21226ab21879SDennis Zhou rcu_read_lock(); 21236ab21879SDennis Zhou 2124fc5a828bSDennis Zhou if (src->bi_blkg) 21252268c0feSDennis Zhou __bio_associate_blkg(dst, src->bi_blkg); 21266ab21879SDennis Zhou 21276ab21879SDennis Zhou rcu_read_unlock(); 212820bd723eSPaolo Valente } 2129db6638d7SDennis Zhou EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2130f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 2131f9c78b2bSJens Axboe 2132f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2133f9c78b2bSJens Axboe { 2134f9c78b2bSJens Axboe int i; 2135f9c78b2bSJens Axboe 2136ed996a52SChristoph Hellwig for (i = 0; i < BVEC_POOL_NR; i++) { 2137f9c78b2bSJens Axboe int size; 2138f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2139f9c78b2bSJens Axboe 2140f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2141f9c78b2bSJens Axboe bvs->slab = NULL; 2142f9c78b2bSJens Axboe continue; 2143f9c78b2bSJens Axboe } 2144f9c78b2bSJens Axboe 2145f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2146f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2147f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2148f9c78b2bSJens Axboe } 2149f9c78b2bSJens Axboe } 2150f9c78b2bSJens Axboe 2151f9c78b2bSJens Axboe static int __init init_bio(void) 2152f9c78b2bSJens Axboe { 2153f9c78b2bSJens Axboe bio_slab_max = 2; 2154f9c78b2bSJens Axboe bio_slab_nr = 0; 21556396bb22SKees Cook bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 21566396bb22SKees Cook GFP_KERNEL); 21572b24e6f6SJohannes Thumshirn 21582b24e6f6SJohannes Thumshirn BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 21592b24e6f6SJohannes Thumshirn 2160f9c78b2bSJens Axboe if (!bio_slabs) 2161f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2162f9c78b2bSJens Axboe 2163f9c78b2bSJens Axboe bio_integrity_init(); 2164f9c78b2bSJens Axboe biovec_init_slabs(); 2165f9c78b2bSJens Axboe 2166f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2167f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2168f9c78b2bSJens Axboe 2169f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2170f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2171f9c78b2bSJens Axboe 2172f9c78b2bSJens Axboe return 0; 2173f9c78b2bSJens Axboe } 2174f9c78b2bSJens Axboe subsys_initcall(init_bio); 2175