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> 20de6a78b6SMing Lei #include <linux/sched/sysctl.h> 21f9c78b2bSJens Axboe 22f9c78b2bSJens Axboe #include <trace/events/block.h> 239e234eeaSShaohua Li #include "blk.h" 2467b42d0bSJosef Bacik #include "blk-rq-qos.h" 25f9c78b2bSJens Axboe 26f9c78b2bSJens Axboe /* 27f9c78b2bSJens Axboe * Test patch to inline a certain number of bi_io_vec's inside the bio 28f9c78b2bSJens Axboe * itself, to shrink a bio data allocation from two mempool calls to one 29f9c78b2bSJens Axboe */ 30f9c78b2bSJens Axboe #define BIO_INLINE_VECS 4 31f9c78b2bSJens Axboe 32f9c78b2bSJens Axboe /* 33f9c78b2bSJens Axboe * if you change this list, also change bvec_alloc or things will 34f9c78b2bSJens Axboe * break badly! cannot be bigger than what you can fit into an 35f9c78b2bSJens Axboe * unsigned short 36f9c78b2bSJens Axboe */ 37bd5c4facSMikulas Patocka #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n } 38ed996a52SChristoph Hellwig static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = { 39bd5c4facSMikulas Patocka BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max), 40f9c78b2bSJens Axboe }; 41f9c78b2bSJens Axboe #undef BV 42f9c78b2bSJens Axboe 43f9c78b2bSJens Axboe /* 44f9c78b2bSJens Axboe * fs_bio_set is the bio_set containing bio and iovec memory pools used by 45f9c78b2bSJens Axboe * IO code that does not need private memory pools. 46f9c78b2bSJens Axboe */ 47f4f8154aSKent Overstreet struct bio_set fs_bio_set; 48f9c78b2bSJens Axboe EXPORT_SYMBOL(fs_bio_set); 49f9c78b2bSJens Axboe 50f9c78b2bSJens Axboe /* 51f9c78b2bSJens Axboe * Our slab pool management 52f9c78b2bSJens Axboe */ 53f9c78b2bSJens Axboe struct bio_slab { 54f9c78b2bSJens Axboe struct kmem_cache *slab; 55f9c78b2bSJens Axboe unsigned int slab_ref; 56f9c78b2bSJens Axboe unsigned int slab_size; 57f9c78b2bSJens Axboe char name[8]; 58f9c78b2bSJens Axboe }; 59f9c78b2bSJens Axboe static DEFINE_MUTEX(bio_slab_lock); 60f9c78b2bSJens Axboe static struct bio_slab *bio_slabs; 61f9c78b2bSJens Axboe static unsigned int bio_slab_nr, bio_slab_max; 62f9c78b2bSJens Axboe 63f9c78b2bSJens Axboe static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size) 64f9c78b2bSJens Axboe { 65f9c78b2bSJens Axboe unsigned int sz = sizeof(struct bio) + extra_size; 66f9c78b2bSJens Axboe struct kmem_cache *slab = NULL; 67f9c78b2bSJens Axboe struct bio_slab *bslab, *new_bio_slabs; 68f9c78b2bSJens Axboe unsigned int new_bio_slab_max; 69f9c78b2bSJens Axboe unsigned int i, entry = -1; 70f9c78b2bSJens Axboe 71f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 72f9c78b2bSJens Axboe 73f9c78b2bSJens Axboe i = 0; 74f9c78b2bSJens Axboe while (i < bio_slab_nr) { 75f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 76f9c78b2bSJens Axboe 77f9c78b2bSJens Axboe if (!bslab->slab && entry == -1) 78f9c78b2bSJens Axboe entry = i; 79f9c78b2bSJens Axboe else if (bslab->slab_size == sz) { 80f9c78b2bSJens Axboe slab = bslab->slab; 81f9c78b2bSJens Axboe bslab->slab_ref++; 82f9c78b2bSJens Axboe break; 83f9c78b2bSJens Axboe } 84f9c78b2bSJens Axboe i++; 85f9c78b2bSJens Axboe } 86f9c78b2bSJens Axboe 87f9c78b2bSJens Axboe if (slab) 88f9c78b2bSJens Axboe goto out_unlock; 89f9c78b2bSJens Axboe 90f9c78b2bSJens Axboe if (bio_slab_nr == bio_slab_max && entry == -1) { 91f9c78b2bSJens Axboe new_bio_slab_max = bio_slab_max << 1; 92f9c78b2bSJens Axboe new_bio_slabs = krealloc(bio_slabs, 93f9c78b2bSJens Axboe new_bio_slab_max * sizeof(struct bio_slab), 94f9c78b2bSJens Axboe GFP_KERNEL); 95f9c78b2bSJens Axboe if (!new_bio_slabs) 96f9c78b2bSJens Axboe goto out_unlock; 97f9c78b2bSJens Axboe bio_slab_max = new_bio_slab_max; 98f9c78b2bSJens Axboe bio_slabs = new_bio_slabs; 99f9c78b2bSJens Axboe } 100f9c78b2bSJens Axboe if (entry == -1) 101f9c78b2bSJens Axboe entry = bio_slab_nr++; 102f9c78b2bSJens Axboe 103f9c78b2bSJens Axboe bslab = &bio_slabs[entry]; 104f9c78b2bSJens Axboe 105f9c78b2bSJens Axboe snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry); 1066a241483SMikulas Patocka slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN, 1076a241483SMikulas Patocka SLAB_HWCACHE_ALIGN, NULL); 108f9c78b2bSJens Axboe if (!slab) 109f9c78b2bSJens Axboe goto out_unlock; 110f9c78b2bSJens Axboe 111f9c78b2bSJens Axboe bslab->slab = slab; 112f9c78b2bSJens Axboe bslab->slab_ref = 1; 113f9c78b2bSJens Axboe bslab->slab_size = sz; 114f9c78b2bSJens Axboe out_unlock: 115f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 116f9c78b2bSJens Axboe return slab; 117f9c78b2bSJens Axboe } 118f9c78b2bSJens Axboe 119f9c78b2bSJens Axboe static void bio_put_slab(struct bio_set *bs) 120f9c78b2bSJens Axboe { 121f9c78b2bSJens Axboe struct bio_slab *bslab = NULL; 122f9c78b2bSJens Axboe unsigned int i; 123f9c78b2bSJens Axboe 124f9c78b2bSJens Axboe mutex_lock(&bio_slab_lock); 125f9c78b2bSJens Axboe 126f9c78b2bSJens Axboe for (i = 0; i < bio_slab_nr; i++) { 127f9c78b2bSJens Axboe if (bs->bio_slab == bio_slabs[i].slab) { 128f9c78b2bSJens Axboe bslab = &bio_slabs[i]; 129f9c78b2bSJens Axboe break; 130f9c78b2bSJens Axboe } 131f9c78b2bSJens Axboe } 132f9c78b2bSJens Axboe 133f9c78b2bSJens Axboe if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n")) 134f9c78b2bSJens Axboe goto out; 135f9c78b2bSJens Axboe 136f9c78b2bSJens Axboe WARN_ON(!bslab->slab_ref); 137f9c78b2bSJens Axboe 138f9c78b2bSJens Axboe if (--bslab->slab_ref) 139f9c78b2bSJens Axboe goto out; 140f9c78b2bSJens Axboe 141f9c78b2bSJens Axboe kmem_cache_destroy(bslab->slab); 142f9c78b2bSJens Axboe bslab->slab = NULL; 143f9c78b2bSJens Axboe 144f9c78b2bSJens Axboe out: 145f9c78b2bSJens Axboe mutex_unlock(&bio_slab_lock); 146f9c78b2bSJens Axboe } 147f9c78b2bSJens Axboe 148f9c78b2bSJens Axboe unsigned int bvec_nr_vecs(unsigned short idx) 149f9c78b2bSJens Axboe { 150d6c02a9bSGreg Edwards return bvec_slabs[--idx].nr_vecs; 151f9c78b2bSJens Axboe } 152f9c78b2bSJens Axboe 153f9c78b2bSJens Axboe void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx) 154f9c78b2bSJens Axboe { 155ed996a52SChristoph Hellwig if (!idx) 156ed996a52SChristoph Hellwig return; 157ed996a52SChristoph Hellwig idx--; 158f9c78b2bSJens Axboe 159ed996a52SChristoph Hellwig BIO_BUG_ON(idx >= BVEC_POOL_NR); 160ed996a52SChristoph Hellwig 161ed996a52SChristoph Hellwig if (idx == BVEC_POOL_MAX) { 162f9c78b2bSJens Axboe mempool_free(bv, pool); 163ed996a52SChristoph Hellwig } else { 164f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + idx; 165f9c78b2bSJens Axboe 166f9c78b2bSJens Axboe kmem_cache_free(bvs->slab, bv); 167f9c78b2bSJens Axboe } 168f9c78b2bSJens Axboe } 169f9c78b2bSJens Axboe 170f9c78b2bSJens Axboe struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx, 171f9c78b2bSJens Axboe mempool_t *pool) 172f9c78b2bSJens Axboe { 173f9c78b2bSJens Axboe struct bio_vec *bvl; 174f9c78b2bSJens Axboe 175f9c78b2bSJens Axboe /* 176f9c78b2bSJens Axboe * see comment near bvec_array define! 177f9c78b2bSJens Axboe */ 178f9c78b2bSJens Axboe switch (nr) { 179f9c78b2bSJens Axboe case 1: 180f9c78b2bSJens Axboe *idx = 0; 181f9c78b2bSJens Axboe break; 182f9c78b2bSJens Axboe case 2 ... 4: 183f9c78b2bSJens Axboe *idx = 1; 184f9c78b2bSJens Axboe break; 185f9c78b2bSJens Axboe case 5 ... 16: 186f9c78b2bSJens Axboe *idx = 2; 187f9c78b2bSJens Axboe break; 188f9c78b2bSJens Axboe case 17 ... 64: 189f9c78b2bSJens Axboe *idx = 3; 190f9c78b2bSJens Axboe break; 191f9c78b2bSJens Axboe case 65 ... 128: 192f9c78b2bSJens Axboe *idx = 4; 193f9c78b2bSJens Axboe break; 194f9c78b2bSJens Axboe case 129 ... BIO_MAX_PAGES: 195f9c78b2bSJens Axboe *idx = 5; 196f9c78b2bSJens Axboe break; 197f9c78b2bSJens Axboe default: 198f9c78b2bSJens Axboe return NULL; 199f9c78b2bSJens Axboe } 200f9c78b2bSJens Axboe 201f9c78b2bSJens Axboe /* 202f9c78b2bSJens Axboe * idx now points to the pool we want to allocate from. only the 203f9c78b2bSJens Axboe * 1-vec entry pool is mempool backed. 204f9c78b2bSJens Axboe */ 205ed996a52SChristoph Hellwig if (*idx == BVEC_POOL_MAX) { 206f9c78b2bSJens Axboe fallback: 207f9c78b2bSJens Axboe bvl = mempool_alloc(pool, gfp_mask); 208f9c78b2bSJens Axboe } else { 209f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + *idx; 210d0164adcSMel Gorman gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO); 211f9c78b2bSJens Axboe 212f9c78b2bSJens Axboe /* 213f9c78b2bSJens Axboe * Make this allocation restricted and don't dump info on 214f9c78b2bSJens Axboe * allocation failures, since we'll fallback to the mempool 215f9c78b2bSJens Axboe * in case of failure. 216f9c78b2bSJens Axboe */ 217f9c78b2bSJens Axboe __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN; 218f9c78b2bSJens Axboe 219f9c78b2bSJens Axboe /* 220d0164adcSMel Gorman * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM 221f9c78b2bSJens Axboe * is set, retry with the 1-entry mempool 222f9c78b2bSJens Axboe */ 223f9c78b2bSJens Axboe bvl = kmem_cache_alloc(bvs->slab, __gfp_mask); 224d0164adcSMel Gorman if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) { 225ed996a52SChristoph Hellwig *idx = BVEC_POOL_MAX; 226f9c78b2bSJens Axboe goto fallback; 227f9c78b2bSJens Axboe } 228f9c78b2bSJens Axboe } 229f9c78b2bSJens Axboe 230ed996a52SChristoph Hellwig (*idx)++; 231f9c78b2bSJens Axboe return bvl; 232f9c78b2bSJens Axboe } 233f9c78b2bSJens Axboe 2349ae3b3f5SJens Axboe void bio_uninit(struct bio *bio) 235f9c78b2bSJens Axboe { 2366f70fb66SDennis Zhou bio_disassociate_blkg(bio); 237ece841abSJustin Tee 238ece841abSJustin Tee if (bio_integrity(bio)) 239ece841abSJustin Tee bio_integrity_free(bio); 240f9c78b2bSJens Axboe } 2419ae3b3f5SJens Axboe EXPORT_SYMBOL(bio_uninit); 242f9c78b2bSJens Axboe 243f9c78b2bSJens Axboe static void bio_free(struct bio *bio) 244f9c78b2bSJens Axboe { 245f9c78b2bSJens Axboe struct bio_set *bs = bio->bi_pool; 246f9c78b2bSJens Axboe void *p; 247f9c78b2bSJens Axboe 2489ae3b3f5SJens Axboe bio_uninit(bio); 249f9c78b2bSJens Axboe 250f9c78b2bSJens Axboe if (bs) { 2518aa6ba2fSKent Overstreet bvec_free(&bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio)); 252f9c78b2bSJens Axboe 253f9c78b2bSJens Axboe /* 254f9c78b2bSJens Axboe * If we have front padding, adjust the bio pointer before freeing 255f9c78b2bSJens Axboe */ 256f9c78b2bSJens Axboe p = bio; 257f9c78b2bSJens Axboe p -= bs->front_pad; 258f9c78b2bSJens Axboe 2598aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 260f9c78b2bSJens Axboe } else { 261f9c78b2bSJens Axboe /* Bio was allocated by bio_kmalloc() */ 262f9c78b2bSJens Axboe kfree(bio); 263f9c78b2bSJens Axboe } 264f9c78b2bSJens Axboe } 265f9c78b2bSJens Axboe 2669ae3b3f5SJens Axboe /* 2679ae3b3f5SJens Axboe * Users of this function have their own bio allocation. Subsequently, 2689ae3b3f5SJens Axboe * they must remember to pair any call to bio_init() with bio_uninit() 2699ae3b3f5SJens Axboe * when IO has completed, or when the bio is released. 2709ae3b3f5SJens Axboe */ 2713a83f467SMing Lei void bio_init(struct bio *bio, struct bio_vec *table, 2723a83f467SMing Lei unsigned short max_vecs) 273f9c78b2bSJens Axboe { 274f9c78b2bSJens Axboe memset(bio, 0, sizeof(*bio)); 275c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 276dac56212SJens Axboe atomic_set(&bio->__bi_cnt, 1); 2773a83f467SMing Lei 2783a83f467SMing Lei bio->bi_io_vec = table; 2793a83f467SMing Lei bio->bi_max_vecs = max_vecs; 280f9c78b2bSJens Axboe } 281f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_init); 282f9c78b2bSJens Axboe 283f9c78b2bSJens Axboe /** 284f9c78b2bSJens Axboe * bio_reset - reinitialize a bio 285f9c78b2bSJens Axboe * @bio: bio to reset 286f9c78b2bSJens Axboe * 287f9c78b2bSJens Axboe * Description: 288f9c78b2bSJens Axboe * After calling bio_reset(), @bio will be in the same state as a freshly 289f9c78b2bSJens Axboe * allocated bio returned bio bio_alloc_bioset() - the only fields that are 290f9c78b2bSJens Axboe * preserved are the ones that are initialized by bio_alloc_bioset(). See 291f9c78b2bSJens Axboe * comment in struct bio. 292f9c78b2bSJens Axboe */ 293f9c78b2bSJens Axboe void bio_reset(struct bio *bio) 294f9c78b2bSJens Axboe { 295f9c78b2bSJens Axboe unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS); 296f9c78b2bSJens Axboe 2979ae3b3f5SJens Axboe bio_uninit(bio); 298f9c78b2bSJens Axboe 299f9c78b2bSJens Axboe memset(bio, 0, BIO_RESET_BYTES); 3004246a0b6SChristoph Hellwig bio->bi_flags = flags; 301c4cf5261SJens Axboe atomic_set(&bio->__bi_remaining, 1); 302f9c78b2bSJens Axboe } 303f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_reset); 304f9c78b2bSJens Axboe 30538f8baaeSChristoph Hellwig static struct bio *__bio_chain_endio(struct bio *bio) 306f9c78b2bSJens Axboe { 3074246a0b6SChristoph Hellwig struct bio *parent = bio->bi_private; 3084246a0b6SChristoph Hellwig 3094e4cbee9SChristoph Hellwig if (!parent->bi_status) 3104e4cbee9SChristoph Hellwig parent->bi_status = bio->bi_status; 311f9c78b2bSJens Axboe bio_put(bio); 31238f8baaeSChristoph Hellwig return parent; 31338f8baaeSChristoph Hellwig } 31438f8baaeSChristoph Hellwig 31538f8baaeSChristoph Hellwig static void bio_chain_endio(struct bio *bio) 31638f8baaeSChristoph Hellwig { 31738f8baaeSChristoph Hellwig bio_endio(__bio_chain_endio(bio)); 318f9c78b2bSJens Axboe } 319f9c78b2bSJens Axboe 320f9c78b2bSJens Axboe /** 321f9c78b2bSJens Axboe * bio_chain - chain bio completions 322f9c78b2bSJens Axboe * @bio: the target bio 323f9c78b2bSJens Axboe * @parent: the @bio's parent bio 324f9c78b2bSJens Axboe * 325f9c78b2bSJens Axboe * The caller won't have a bi_end_io called when @bio completes - instead, 326f9c78b2bSJens Axboe * @parent's bi_end_io won't be called until both @parent and @bio have 327f9c78b2bSJens Axboe * completed; the chained bio will also be freed when it completes. 328f9c78b2bSJens Axboe * 329f9c78b2bSJens Axboe * The caller must not set bi_private or bi_end_io in @bio. 330f9c78b2bSJens Axboe */ 331f9c78b2bSJens Axboe void bio_chain(struct bio *bio, struct bio *parent) 332f9c78b2bSJens Axboe { 333f9c78b2bSJens Axboe BUG_ON(bio->bi_private || bio->bi_end_io); 334f9c78b2bSJens Axboe 335f9c78b2bSJens Axboe bio->bi_private = parent; 336f9c78b2bSJens Axboe bio->bi_end_io = bio_chain_endio; 337c4cf5261SJens Axboe bio_inc_remaining(parent); 338f9c78b2bSJens Axboe } 339f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_chain); 340f9c78b2bSJens Axboe 341f9c78b2bSJens Axboe static void bio_alloc_rescue(struct work_struct *work) 342f9c78b2bSJens Axboe { 343f9c78b2bSJens Axboe struct bio_set *bs = container_of(work, struct bio_set, rescue_work); 344f9c78b2bSJens Axboe struct bio *bio; 345f9c78b2bSJens Axboe 346f9c78b2bSJens Axboe while (1) { 347f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 348f9c78b2bSJens Axboe bio = bio_list_pop(&bs->rescue_list); 349f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 350f9c78b2bSJens Axboe 351f9c78b2bSJens Axboe if (!bio) 352f9c78b2bSJens Axboe break; 353f9c78b2bSJens Axboe 354f9c78b2bSJens Axboe generic_make_request(bio); 355f9c78b2bSJens Axboe } 356f9c78b2bSJens Axboe } 357f9c78b2bSJens Axboe 358f9c78b2bSJens Axboe static void punt_bios_to_rescuer(struct bio_set *bs) 359f9c78b2bSJens Axboe { 360f9c78b2bSJens Axboe struct bio_list punt, nopunt; 361f9c78b2bSJens Axboe struct bio *bio; 362f9c78b2bSJens Axboe 36347e0fb46SNeilBrown if (WARN_ON_ONCE(!bs->rescue_workqueue)) 36447e0fb46SNeilBrown return; 365f9c78b2bSJens Axboe /* 366f9c78b2bSJens Axboe * In order to guarantee forward progress we must punt only bios that 367f9c78b2bSJens Axboe * were allocated from this bio_set; otherwise, if there was a bio on 368f9c78b2bSJens Axboe * there for a stacking driver higher up in the stack, processing it 369f9c78b2bSJens Axboe * could require allocating bios from this bio_set, and doing that from 370f9c78b2bSJens Axboe * our own rescuer would be bad. 371f9c78b2bSJens Axboe * 372f9c78b2bSJens Axboe * Since bio lists are singly linked, pop them all instead of trying to 373f9c78b2bSJens Axboe * remove from the middle of the list: 374f9c78b2bSJens Axboe */ 375f9c78b2bSJens Axboe 376f9c78b2bSJens Axboe bio_list_init(&punt); 377f9c78b2bSJens Axboe bio_list_init(&nopunt); 378f9c78b2bSJens Axboe 379f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[0]))) 380f9c78b2bSJens Axboe bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 381f5fe1b51SNeilBrown current->bio_list[0] = nopunt; 382f9c78b2bSJens Axboe 383f5fe1b51SNeilBrown bio_list_init(&nopunt); 384f5fe1b51SNeilBrown while ((bio = bio_list_pop(¤t->bio_list[1]))) 385f5fe1b51SNeilBrown bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio); 386f5fe1b51SNeilBrown current->bio_list[1] = nopunt; 387f9c78b2bSJens Axboe 388f9c78b2bSJens Axboe spin_lock(&bs->rescue_lock); 389f9c78b2bSJens Axboe bio_list_merge(&bs->rescue_list, &punt); 390f9c78b2bSJens Axboe spin_unlock(&bs->rescue_lock); 391f9c78b2bSJens Axboe 392f9c78b2bSJens Axboe queue_work(bs->rescue_workqueue, &bs->rescue_work); 393f9c78b2bSJens Axboe } 394f9c78b2bSJens Axboe 395f9c78b2bSJens Axboe /** 396f9c78b2bSJens Axboe * bio_alloc_bioset - allocate a bio for I/O 397519c8e9fSRandy Dunlap * @gfp_mask: the GFP_* mask given to the slab allocator 398f9c78b2bSJens Axboe * @nr_iovecs: number of iovecs to pre-allocate 399f9c78b2bSJens Axboe * @bs: the bio_set to allocate from. 400f9c78b2bSJens Axboe * 401f9c78b2bSJens Axboe * Description: 402f9c78b2bSJens Axboe * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is 403f9c78b2bSJens Axboe * backed by the @bs's mempool. 404f9c78b2bSJens Axboe * 405d0164adcSMel Gorman * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will 406d0164adcSMel Gorman * always be able to allocate a bio. This is due to the mempool guarantees. 407d0164adcSMel Gorman * To make this work, callers must never allocate more than 1 bio at a time 408d0164adcSMel Gorman * from this pool. Callers that need to allocate more than 1 bio must always 409d0164adcSMel Gorman * submit the previously allocated bio for IO before attempting to allocate 410d0164adcSMel Gorman * a new one. Failure to do so can cause deadlocks under memory pressure. 411f9c78b2bSJens Axboe * 412f9c78b2bSJens Axboe * Note that when running under generic_make_request() (i.e. any block 413f9c78b2bSJens Axboe * driver), bios are not submitted until after you return - see the code in 414f9c78b2bSJens Axboe * generic_make_request() that converts recursion into iteration, to prevent 415f9c78b2bSJens Axboe * stack overflows. 416f9c78b2bSJens Axboe * 417f9c78b2bSJens Axboe * This would normally mean allocating multiple bios under 418f9c78b2bSJens Axboe * generic_make_request() would be susceptible to deadlocks, but we have 419f9c78b2bSJens Axboe * deadlock avoidance code that resubmits any blocked bios from a rescuer 420f9c78b2bSJens Axboe * thread. 421f9c78b2bSJens Axboe * 422f9c78b2bSJens Axboe * However, we do not guarantee forward progress for allocations from other 423f9c78b2bSJens Axboe * mempools. Doing multiple allocations from the same mempool under 424f9c78b2bSJens Axboe * generic_make_request() should be avoided - instead, use bio_set's front_pad 425f9c78b2bSJens Axboe * for per bio allocations. 426f9c78b2bSJens Axboe * 427f9c78b2bSJens Axboe * RETURNS: 428f9c78b2bSJens Axboe * Pointer to new bio on success, NULL on failure. 429f9c78b2bSJens Axboe */ 4307a88fa19SDan Carpenter struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs, 4317a88fa19SDan Carpenter struct bio_set *bs) 432f9c78b2bSJens Axboe { 433f9c78b2bSJens Axboe gfp_t saved_gfp = gfp_mask; 434f9c78b2bSJens Axboe unsigned front_pad; 435f9c78b2bSJens Axboe unsigned inline_vecs; 436f9c78b2bSJens Axboe struct bio_vec *bvl = NULL; 437f9c78b2bSJens Axboe struct bio *bio; 438f9c78b2bSJens Axboe void *p; 439f9c78b2bSJens Axboe 440f9c78b2bSJens Axboe if (!bs) { 441f9c78b2bSJens Axboe if (nr_iovecs > UIO_MAXIOV) 442f9c78b2bSJens Axboe return NULL; 443f9c78b2bSJens Axboe 444f9c78b2bSJens Axboe p = kmalloc(sizeof(struct bio) + 445f9c78b2bSJens Axboe nr_iovecs * sizeof(struct bio_vec), 446f9c78b2bSJens Axboe gfp_mask); 447f9c78b2bSJens Axboe front_pad = 0; 448f9c78b2bSJens Axboe inline_vecs = nr_iovecs; 449f9c78b2bSJens Axboe } else { 450d8f429e1SJunichi Nomura /* should not use nobvec bioset for nr_iovecs > 0 */ 4518aa6ba2fSKent Overstreet if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && 4528aa6ba2fSKent Overstreet nr_iovecs > 0)) 453d8f429e1SJunichi Nomura return NULL; 454f9c78b2bSJens Axboe /* 455f9c78b2bSJens Axboe * generic_make_request() converts recursion to iteration; this 456f9c78b2bSJens Axboe * means if we're running beneath it, any bios we allocate and 457f9c78b2bSJens Axboe * submit will not be submitted (and thus freed) until after we 458f9c78b2bSJens Axboe * return. 459f9c78b2bSJens Axboe * 460f9c78b2bSJens Axboe * This exposes us to a potential deadlock if we allocate 461f9c78b2bSJens Axboe * multiple bios from the same bio_set() while running 462f9c78b2bSJens Axboe * underneath generic_make_request(). If we were to allocate 463f9c78b2bSJens Axboe * multiple bios (say a stacking block driver that was splitting 464f9c78b2bSJens Axboe * bios), we would deadlock if we exhausted the mempool's 465f9c78b2bSJens Axboe * reserve. 466f9c78b2bSJens Axboe * 467f9c78b2bSJens Axboe * We solve this, and guarantee forward progress, with a rescuer 468f9c78b2bSJens Axboe * workqueue per bio_set. If we go to allocate and there are 469f9c78b2bSJens Axboe * bios on current->bio_list, we first try the allocation 470d0164adcSMel Gorman * without __GFP_DIRECT_RECLAIM; if that fails, we punt those 471d0164adcSMel Gorman * bios we would be blocking to the rescuer workqueue before 472d0164adcSMel Gorman * we retry with the original gfp_flags. 473f9c78b2bSJens Axboe */ 474f9c78b2bSJens Axboe 475f5fe1b51SNeilBrown if (current->bio_list && 476f5fe1b51SNeilBrown (!bio_list_empty(¤t->bio_list[0]) || 47747e0fb46SNeilBrown !bio_list_empty(¤t->bio_list[1])) && 47847e0fb46SNeilBrown bs->rescue_workqueue) 479d0164adcSMel Gorman gfp_mask &= ~__GFP_DIRECT_RECLAIM; 480f9c78b2bSJens Axboe 4818aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 482f9c78b2bSJens Axboe if (!p && gfp_mask != saved_gfp) { 483f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 484f9c78b2bSJens Axboe gfp_mask = saved_gfp; 4858aa6ba2fSKent Overstreet p = mempool_alloc(&bs->bio_pool, gfp_mask); 486f9c78b2bSJens Axboe } 487f9c78b2bSJens Axboe 488f9c78b2bSJens Axboe front_pad = bs->front_pad; 489f9c78b2bSJens Axboe inline_vecs = BIO_INLINE_VECS; 490f9c78b2bSJens Axboe } 491f9c78b2bSJens Axboe 492f9c78b2bSJens Axboe if (unlikely(!p)) 493f9c78b2bSJens Axboe return NULL; 494f9c78b2bSJens Axboe 495f9c78b2bSJens Axboe bio = p + front_pad; 4963a83f467SMing Lei bio_init(bio, NULL, 0); 497f9c78b2bSJens Axboe 498f9c78b2bSJens Axboe if (nr_iovecs > inline_vecs) { 499ed996a52SChristoph Hellwig unsigned long idx = 0; 500ed996a52SChristoph Hellwig 5018aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 502f9c78b2bSJens Axboe if (!bvl && gfp_mask != saved_gfp) { 503f9c78b2bSJens Axboe punt_bios_to_rescuer(bs); 504f9c78b2bSJens Axboe gfp_mask = saved_gfp; 5058aa6ba2fSKent Overstreet bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, &bs->bvec_pool); 506f9c78b2bSJens Axboe } 507f9c78b2bSJens Axboe 508f9c78b2bSJens Axboe if (unlikely(!bvl)) 509f9c78b2bSJens Axboe goto err_free; 510f9c78b2bSJens Axboe 511ed996a52SChristoph Hellwig bio->bi_flags |= idx << BVEC_POOL_OFFSET; 512f9c78b2bSJens Axboe } else if (nr_iovecs) { 513f9c78b2bSJens Axboe bvl = bio->bi_inline_vecs; 514f9c78b2bSJens Axboe } 515f9c78b2bSJens Axboe 516f9c78b2bSJens Axboe bio->bi_pool = bs; 517f9c78b2bSJens Axboe bio->bi_max_vecs = nr_iovecs; 518f9c78b2bSJens Axboe bio->bi_io_vec = bvl; 519f9c78b2bSJens Axboe return bio; 520f9c78b2bSJens Axboe 521f9c78b2bSJens Axboe err_free: 5228aa6ba2fSKent Overstreet mempool_free(p, &bs->bio_pool); 523f9c78b2bSJens Axboe return NULL; 524f9c78b2bSJens Axboe } 525f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_alloc_bioset); 526f9c78b2bSJens Axboe 52738a72dacSKent Overstreet void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start) 528f9c78b2bSJens Axboe { 529f9c78b2bSJens Axboe unsigned long flags; 530f9c78b2bSJens Axboe struct bio_vec bv; 531f9c78b2bSJens Axboe struct bvec_iter iter; 532f9c78b2bSJens Axboe 53338a72dacSKent Overstreet __bio_for_each_segment(bv, bio, iter, start) { 534f9c78b2bSJens Axboe char *data = bvec_kmap_irq(&bv, &flags); 535f9c78b2bSJens Axboe memset(data, 0, bv.bv_len); 536f9c78b2bSJens Axboe flush_dcache_page(bv.bv_page); 537f9c78b2bSJens Axboe bvec_kunmap_irq(data, &flags); 538f9c78b2bSJens Axboe } 539f9c78b2bSJens Axboe } 54038a72dacSKent Overstreet EXPORT_SYMBOL(zero_fill_bio_iter); 541f9c78b2bSJens Axboe 54283c9c547SMing Lei /** 54383c9c547SMing Lei * bio_truncate - truncate the bio to small size of @new_size 54483c9c547SMing Lei * @bio: the bio to be truncated 54583c9c547SMing Lei * @new_size: new size for truncating the bio 54683c9c547SMing Lei * 54783c9c547SMing Lei * Description: 54883c9c547SMing Lei * Truncate the bio to new size of @new_size. If bio_op(bio) is 54983c9c547SMing Lei * REQ_OP_READ, zero the truncated part. This function should only 55083c9c547SMing Lei * be used for handling corner cases, such as bio eod. 55183c9c547SMing Lei */ 55285a8ce62SMing Lei void bio_truncate(struct bio *bio, unsigned new_size) 55385a8ce62SMing Lei { 55485a8ce62SMing Lei struct bio_vec bv; 55585a8ce62SMing Lei struct bvec_iter iter; 55685a8ce62SMing Lei unsigned int done = 0; 55785a8ce62SMing Lei bool truncated = false; 55885a8ce62SMing Lei 55985a8ce62SMing Lei if (new_size >= bio->bi_iter.bi_size) 56085a8ce62SMing Lei return; 56185a8ce62SMing Lei 56283c9c547SMing Lei if (bio_op(bio) != REQ_OP_READ) 56385a8ce62SMing Lei goto exit; 56485a8ce62SMing Lei 56585a8ce62SMing Lei bio_for_each_segment(bv, bio, iter) { 56685a8ce62SMing Lei if (done + bv.bv_len > new_size) { 56785a8ce62SMing Lei unsigned offset; 56885a8ce62SMing Lei 56985a8ce62SMing Lei if (!truncated) 57085a8ce62SMing Lei offset = new_size - done; 57185a8ce62SMing Lei else 57285a8ce62SMing Lei offset = 0; 57385a8ce62SMing Lei zero_user(bv.bv_page, offset, bv.bv_len - offset); 57485a8ce62SMing Lei truncated = true; 57585a8ce62SMing Lei } 57685a8ce62SMing Lei done += bv.bv_len; 57785a8ce62SMing Lei } 57885a8ce62SMing Lei 57985a8ce62SMing Lei exit: 58085a8ce62SMing Lei /* 58185a8ce62SMing Lei * Don't touch bvec table here and make it really immutable, since 58285a8ce62SMing Lei * fs bio user has to retrieve all pages via bio_for_each_segment_all 58385a8ce62SMing Lei * in its .end_bio() callback. 58485a8ce62SMing Lei * 58585a8ce62SMing Lei * It is enough to truncate bio by updating .bi_size since we can make 58685a8ce62SMing Lei * correct bvec with the updated .bi_size for drivers. 58785a8ce62SMing Lei */ 58885a8ce62SMing Lei bio->bi_iter.bi_size = new_size; 58985a8ce62SMing Lei } 59085a8ce62SMing Lei 591f9c78b2bSJens Axboe /** 592f9c78b2bSJens Axboe * bio_put - release a reference to a bio 593f9c78b2bSJens Axboe * @bio: bio to release reference to 594f9c78b2bSJens Axboe * 595f9c78b2bSJens Axboe * Description: 596f9c78b2bSJens Axboe * Put a reference to a &struct bio, either one you have gotten with 5979b10f6a9SNeilBrown * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it. 598f9c78b2bSJens Axboe **/ 599f9c78b2bSJens Axboe void bio_put(struct bio *bio) 600f9c78b2bSJens Axboe { 601dac56212SJens Axboe if (!bio_flagged(bio, BIO_REFFED)) 602dac56212SJens Axboe bio_free(bio); 603dac56212SJens Axboe else { 604dac56212SJens Axboe BIO_BUG_ON(!atomic_read(&bio->__bi_cnt)); 605f9c78b2bSJens Axboe 606f9c78b2bSJens Axboe /* 607f9c78b2bSJens Axboe * last put frees it 608f9c78b2bSJens Axboe */ 609dac56212SJens Axboe if (atomic_dec_and_test(&bio->__bi_cnt)) 610f9c78b2bSJens Axboe bio_free(bio); 611f9c78b2bSJens Axboe } 612dac56212SJens Axboe } 613f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_put); 614f9c78b2bSJens Axboe 615f9c78b2bSJens Axboe /** 616f9c78b2bSJens Axboe * __bio_clone_fast - clone a bio that shares the original bio's biovec 617f9c78b2bSJens Axboe * @bio: destination bio 618f9c78b2bSJens Axboe * @bio_src: bio to clone 619f9c78b2bSJens Axboe * 620f9c78b2bSJens Axboe * Clone a &bio. Caller will own the returned bio, but not 621f9c78b2bSJens Axboe * the actual data it points to. Reference count of returned 622f9c78b2bSJens Axboe * bio will be one. 623f9c78b2bSJens Axboe * 624f9c78b2bSJens Axboe * Caller must ensure that @bio_src is not freed before @bio. 625f9c78b2bSJens Axboe */ 626f9c78b2bSJens Axboe void __bio_clone_fast(struct bio *bio, struct bio *bio_src) 627f9c78b2bSJens Axboe { 628ed996a52SChristoph Hellwig BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio)); 629f9c78b2bSJens Axboe 630f9c78b2bSJens Axboe /* 63174d46992SChristoph Hellwig * most users will be overriding ->bi_disk with a new target, 632f9c78b2bSJens Axboe * so we don't set nor calculate new physical/hw segment counts here 633f9c78b2bSJens Axboe */ 63474d46992SChristoph Hellwig bio->bi_disk = bio_src->bi_disk; 63562530ed8SMichael Lyle bio->bi_partno = bio_src->bi_partno; 636b7c44ed9SJens Axboe bio_set_flag(bio, BIO_CLONED); 637111be883SShaohua Li if (bio_flagged(bio_src, BIO_THROTTLED)) 638111be883SShaohua Li bio_set_flag(bio, BIO_THROTTLED); 6391eff9d32SJens Axboe bio->bi_opf = bio_src->bi_opf; 640ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 641cb6934f8SJens Axboe bio->bi_write_hint = bio_src->bi_write_hint; 642f9c78b2bSJens Axboe bio->bi_iter = bio_src->bi_iter; 643f9c78b2bSJens Axboe bio->bi_io_vec = bio_src->bi_io_vec; 64420bd723eSPaolo Valente 645db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 646e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 647f9c78b2bSJens Axboe } 648f9c78b2bSJens Axboe EXPORT_SYMBOL(__bio_clone_fast); 649f9c78b2bSJens Axboe 650f9c78b2bSJens Axboe /** 651f9c78b2bSJens Axboe * bio_clone_fast - clone a bio that shares the original bio's biovec 652f9c78b2bSJens Axboe * @bio: bio to clone 653f9c78b2bSJens Axboe * @gfp_mask: allocation priority 654f9c78b2bSJens Axboe * @bs: bio_set to allocate from 655f9c78b2bSJens Axboe * 656f9c78b2bSJens Axboe * Like __bio_clone_fast, only also allocates the returned bio 657f9c78b2bSJens Axboe */ 658f9c78b2bSJens Axboe struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs) 659f9c78b2bSJens Axboe { 660f9c78b2bSJens Axboe struct bio *b; 661f9c78b2bSJens Axboe 662f9c78b2bSJens Axboe b = bio_alloc_bioset(gfp_mask, 0, bs); 663f9c78b2bSJens Axboe if (!b) 664f9c78b2bSJens Axboe return NULL; 665f9c78b2bSJens Axboe 666f9c78b2bSJens Axboe __bio_clone_fast(b, bio); 667f9c78b2bSJens Axboe 668f9c78b2bSJens Axboe if (bio_integrity(bio)) { 669f9c78b2bSJens Axboe int ret; 670f9c78b2bSJens Axboe 671f9c78b2bSJens Axboe ret = bio_integrity_clone(b, bio, gfp_mask); 672f9c78b2bSJens Axboe 673f9c78b2bSJens Axboe if (ret < 0) { 674f9c78b2bSJens Axboe bio_put(b); 675f9c78b2bSJens Axboe return NULL; 676f9c78b2bSJens Axboe } 677f9c78b2bSJens Axboe } 678f9c78b2bSJens Axboe 679f9c78b2bSJens Axboe return b; 680f9c78b2bSJens Axboe } 681f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_clone_fast); 682f9c78b2bSJens Axboe 6835919482eSMing Lei static inline bool page_is_mergeable(const struct bio_vec *bv, 6845919482eSMing Lei struct page *page, unsigned int len, unsigned int off, 685ff896738SChristoph Hellwig bool *same_page) 6865919482eSMing Lei { 6875919482eSMing Lei phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + 6885919482eSMing Lei bv->bv_offset + bv->bv_len - 1; 6895919482eSMing Lei phys_addr_t page_addr = page_to_phys(page); 6905919482eSMing Lei 6915919482eSMing Lei if (vec_end_addr + 1 != page_addr + off) 6925919482eSMing Lei return false; 6935919482eSMing Lei if (xen_domain() && !xen_biovec_phys_mergeable(bv, page)) 6945919482eSMing Lei return false; 69552d52d1cSChristoph Hellwig 696ff896738SChristoph Hellwig *same_page = ((vec_end_addr & PAGE_MASK) == page_addr); 697ff896738SChristoph Hellwig if (!*same_page && pfn_to_page(PFN_DOWN(vec_end_addr)) + 1 != page) 6985919482eSMing Lei return false; 6995919482eSMing Lei return true; 7005919482eSMing Lei } 7015919482eSMing Lei 702384209cdSChristoph Hellwig static bool bio_try_merge_pc_page(struct request_queue *q, struct bio *bio, 703384209cdSChristoph Hellwig struct page *page, unsigned len, unsigned offset, 704384209cdSChristoph Hellwig bool *same_page) 705489fbbcbSMing Lei { 706384209cdSChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 707489fbbcbSMing Lei unsigned long mask = queue_segment_boundary(q); 708489fbbcbSMing Lei phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset; 709489fbbcbSMing Lei phys_addr_t addr2 = page_to_phys(page) + offset + len - 1; 710489fbbcbSMing Lei 711489fbbcbSMing Lei if ((addr1 | mask) != (addr2 | mask)) 712489fbbcbSMing Lei return false; 713489fbbcbSMing Lei if (bv->bv_len + len > queue_max_segment_size(q)) 714489fbbcbSMing Lei return false; 715384209cdSChristoph Hellwig return __bio_try_merge_page(bio, page, len, offset, same_page); 716489fbbcbSMing Lei } 717489fbbcbSMing Lei 718f4595875SShaohua Li /** 71919047087SMing Lei * __bio_add_pc_page - attempt to add page to passthrough bio 720c66a14d0SKent Overstreet * @q: the target queue 721c66a14d0SKent Overstreet * @bio: destination bio 722c66a14d0SKent Overstreet * @page: page to add 723c66a14d0SKent Overstreet * @len: vec entry length 724c66a14d0SKent Overstreet * @offset: vec entry offset 725d1916c86SChristoph Hellwig * @same_page: return if the merge happen inside the same page 726f9c78b2bSJens Axboe * 727c66a14d0SKent Overstreet * Attempt to add a page to the bio_vec maplist. This can fail for a 728c66a14d0SKent Overstreet * number of reasons, such as the bio being full or target block device 729c66a14d0SKent Overstreet * limitations. The target block device must allow bio's up to PAGE_SIZE, 730c66a14d0SKent Overstreet * so it is always possible to add a single page to an empty bio. 731c66a14d0SKent Overstreet * 7325a8ce240SMing Lei * This should only be used by passthrough bios. 733f9c78b2bSJens Axboe */ 7344713839dSChristoph Hellwig static int __bio_add_pc_page(struct request_queue *q, struct bio *bio, 73519047087SMing Lei struct page *page, unsigned int len, unsigned int offset, 736d1916c86SChristoph Hellwig bool *same_page) 737f9c78b2bSJens Axboe { 738f9c78b2bSJens Axboe struct bio_vec *bvec; 739f9c78b2bSJens Axboe 740f9c78b2bSJens Axboe /* 741f9c78b2bSJens Axboe * cloned bio must not modify vec list 742f9c78b2bSJens Axboe */ 743f9c78b2bSJens Axboe if (unlikely(bio_flagged(bio, BIO_CLONED))) 744f9c78b2bSJens Axboe return 0; 745f9c78b2bSJens Axboe 746c66a14d0SKent Overstreet if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q)) 747f9c78b2bSJens Axboe return 0; 748f9c78b2bSJens Axboe 749f9c78b2bSJens Axboe if (bio->bi_vcnt > 0) { 750d1916c86SChristoph Hellwig if (bio_try_merge_pc_page(q, bio, page, len, offset, same_page)) 751384209cdSChristoph Hellwig return len; 752320ea869SChristoph Hellwig 753320ea869SChristoph Hellwig /* 754320ea869SChristoph Hellwig * If the queue doesn't support SG gaps and adding this segment 755320ea869SChristoph Hellwig * would create a gap, disallow it. 756320ea869SChristoph Hellwig */ 757384209cdSChristoph Hellwig bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; 758320ea869SChristoph Hellwig if (bvec_gap_to_prev(q, bvec, offset)) 759320ea869SChristoph Hellwig return 0; 760f9c78b2bSJens Axboe } 761f9c78b2bSJens Axboe 76279d08f89SMing Lei if (bio_full(bio, len)) 763f9c78b2bSJens Axboe return 0; 764f9c78b2bSJens Axboe 76514ccb66bSChristoph Hellwig if (bio->bi_vcnt >= queue_max_segments(q)) 766489fbbcbSMing Lei return 0; 767489fbbcbSMing Lei 768f9c78b2bSJens Axboe bvec = &bio->bi_io_vec[bio->bi_vcnt]; 769f9c78b2bSJens Axboe bvec->bv_page = page; 770f9c78b2bSJens Axboe bvec->bv_len = len; 771f9c78b2bSJens Axboe bvec->bv_offset = offset; 772fcbf6a08SMaurizio Lombardi bio->bi_vcnt++; 773dcdca753SChristoph Hellwig bio->bi_iter.bi_size += len; 774f9c78b2bSJens Axboe return len; 775f9c78b2bSJens Axboe } 77619047087SMing Lei 77719047087SMing Lei int bio_add_pc_page(struct request_queue *q, struct bio *bio, 77819047087SMing Lei struct page *page, unsigned int len, unsigned int offset) 77919047087SMing Lei { 780d1916c86SChristoph Hellwig bool same_page = false; 781d1916c86SChristoph Hellwig return __bio_add_pc_page(q, bio, page, len, offset, &same_page); 78219047087SMing Lei } 783f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_pc_page); 784f9c78b2bSJens Axboe 785f9c78b2bSJens Axboe /** 7860aa69fd3SChristoph Hellwig * __bio_try_merge_page - try appending data to an existing bvec. 7870aa69fd3SChristoph Hellwig * @bio: destination bio 788551879a4SMing Lei * @page: start page to add 7890aa69fd3SChristoph Hellwig * @len: length of the data to add 790551879a4SMing Lei * @off: offset of the data relative to @page 791ff896738SChristoph Hellwig * @same_page: return if the segment has been merged inside the same page 7920aa69fd3SChristoph Hellwig * 7930aa69fd3SChristoph Hellwig * Try to add the data at @page + @off to the last bvec of @bio. This is a 7940aa69fd3SChristoph Hellwig * a useful optimisation for file systems with a block size smaller than the 7950aa69fd3SChristoph Hellwig * page size. 7960aa69fd3SChristoph Hellwig * 797551879a4SMing Lei * Warn if (@len, @off) crosses pages in case that @same_page is true. 798551879a4SMing Lei * 7990aa69fd3SChristoph Hellwig * Return %true on success or %false on failure. 8000aa69fd3SChristoph Hellwig */ 8010aa69fd3SChristoph Hellwig bool __bio_try_merge_page(struct bio *bio, struct page *page, 802ff896738SChristoph Hellwig unsigned int len, unsigned int off, bool *same_page) 8030aa69fd3SChristoph Hellwig { 8040aa69fd3SChristoph Hellwig if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED))) 8050aa69fd3SChristoph Hellwig return false; 8060aa69fd3SChristoph Hellwig 807cc90bc68SAndreas Gruenbacher if (bio->bi_vcnt > 0) { 8080aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1]; 8090aa69fd3SChristoph Hellwig 8105919482eSMing Lei if (page_is_mergeable(bv, page, len, off, same_page)) { 811cc90bc68SAndreas Gruenbacher if (bio->bi_iter.bi_size > UINT_MAX - len) 812cc90bc68SAndreas Gruenbacher return false; 8130aa69fd3SChristoph Hellwig bv->bv_len += len; 8140aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8150aa69fd3SChristoph Hellwig return true; 8160aa69fd3SChristoph Hellwig } 8175919482eSMing Lei } 8180aa69fd3SChristoph Hellwig return false; 8190aa69fd3SChristoph Hellwig } 8200aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_try_merge_page); 8210aa69fd3SChristoph Hellwig 8220aa69fd3SChristoph Hellwig /** 823551879a4SMing Lei * __bio_add_page - add page(s) to a bio in a new segment 8240aa69fd3SChristoph Hellwig * @bio: destination bio 825551879a4SMing Lei * @page: start page to add 826551879a4SMing Lei * @len: length of the data to add, may cross pages 827551879a4SMing Lei * @off: offset of the data relative to @page, may cross pages 8280aa69fd3SChristoph Hellwig * 8290aa69fd3SChristoph Hellwig * Add the data at @page + @off to @bio as a new bvec. The caller must ensure 8300aa69fd3SChristoph Hellwig * that @bio has space for another bvec. 8310aa69fd3SChristoph Hellwig */ 8320aa69fd3SChristoph Hellwig void __bio_add_page(struct bio *bio, struct page *page, 8330aa69fd3SChristoph Hellwig unsigned int len, unsigned int off) 8340aa69fd3SChristoph Hellwig { 8350aa69fd3SChristoph Hellwig struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt]; 8360aa69fd3SChristoph Hellwig 8370aa69fd3SChristoph Hellwig WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)); 83879d08f89SMing Lei WARN_ON_ONCE(bio_full(bio, len)); 8390aa69fd3SChristoph Hellwig 8400aa69fd3SChristoph Hellwig bv->bv_page = page; 8410aa69fd3SChristoph Hellwig bv->bv_offset = off; 8420aa69fd3SChristoph Hellwig bv->bv_len = len; 8430aa69fd3SChristoph Hellwig 8440aa69fd3SChristoph Hellwig bio->bi_iter.bi_size += len; 8450aa69fd3SChristoph Hellwig bio->bi_vcnt++; 846b8e24a93SJohannes Weiner 847b8e24a93SJohannes Weiner if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page))) 848b8e24a93SJohannes Weiner bio_set_flag(bio, BIO_WORKINGSET); 8490aa69fd3SChristoph Hellwig } 8500aa69fd3SChristoph Hellwig EXPORT_SYMBOL_GPL(__bio_add_page); 8510aa69fd3SChristoph Hellwig 8520aa69fd3SChristoph Hellwig /** 853551879a4SMing Lei * bio_add_page - attempt to add page(s) to bio 854f9c78b2bSJens Axboe * @bio: destination bio 855551879a4SMing Lei * @page: start page to add 856551879a4SMing Lei * @len: vec entry length, may cross pages 857551879a4SMing Lei * @offset: vec entry offset relative to @page, may cross pages 858f9c78b2bSJens Axboe * 859551879a4SMing Lei * Attempt to add page(s) to the bio_vec maplist. This will only fail 860c66a14d0SKent Overstreet * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio. 861f9c78b2bSJens Axboe */ 862c66a14d0SKent Overstreet int bio_add_page(struct bio *bio, struct page *page, 863c66a14d0SKent Overstreet unsigned int len, unsigned int offset) 864f9c78b2bSJens Axboe { 865ff896738SChristoph Hellwig bool same_page = false; 866ff896738SChristoph Hellwig 867ff896738SChristoph Hellwig if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) { 86879d08f89SMing Lei if (bio_full(bio, len)) 869c66a14d0SKent Overstreet return 0; 8700aa69fd3SChristoph Hellwig __bio_add_page(bio, page, len, offset); 871c66a14d0SKent Overstreet } 872c66a14d0SKent Overstreet return len; 873f9c78b2bSJens Axboe } 874f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_add_page); 875f9c78b2bSJens Axboe 876d241a95fSChristoph Hellwig void bio_release_pages(struct bio *bio, bool mark_dirty) 8777321ecbfSChristoph Hellwig { 8787321ecbfSChristoph Hellwig struct bvec_iter_all iter_all; 8797321ecbfSChristoph Hellwig struct bio_vec *bvec; 8807321ecbfSChristoph Hellwig 881b2d0d991SChristoph Hellwig if (bio_flagged(bio, BIO_NO_PAGE_REF)) 882b2d0d991SChristoph Hellwig return; 883b2d0d991SChristoph Hellwig 884d241a95fSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 885d241a95fSChristoph Hellwig if (mark_dirty && !PageCompound(bvec->bv_page)) 886d241a95fSChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 8877321ecbfSChristoph Hellwig put_page(bvec->bv_page); 8887321ecbfSChristoph Hellwig } 889d241a95fSChristoph Hellwig } 8907321ecbfSChristoph Hellwig 8916d0c48aeSJens Axboe static int __bio_iov_bvec_add_pages(struct bio *bio, struct iov_iter *iter) 8926d0c48aeSJens Axboe { 8936d0c48aeSJens Axboe const struct bio_vec *bv = iter->bvec; 8946d0c48aeSJens Axboe unsigned int len; 8956d0c48aeSJens Axboe size_t size; 8966d0c48aeSJens Axboe 8976d0c48aeSJens Axboe if (WARN_ON_ONCE(iter->iov_offset > bv->bv_len)) 8986d0c48aeSJens Axboe return -EINVAL; 8996d0c48aeSJens Axboe 9006d0c48aeSJens Axboe len = min_t(size_t, bv->bv_len - iter->iov_offset, iter->count); 9016d0c48aeSJens Axboe size = bio_add_page(bio, bv->bv_page, len, 9026d0c48aeSJens Axboe bv->bv_offset + iter->iov_offset); 903a10584c3SChristoph Hellwig if (unlikely(size != len)) 904a10584c3SChristoph Hellwig return -EINVAL; 9056d0c48aeSJens Axboe iov_iter_advance(iter, size); 9066d0c48aeSJens Axboe return 0; 9076d0c48aeSJens Axboe } 9086d0c48aeSJens Axboe 909576ed913SChristoph Hellwig #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *)) 910576ed913SChristoph Hellwig 9112cefe4dbSKent Overstreet /** 91217d51b10SMartin Wilck * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio 9132cefe4dbSKent Overstreet * @bio: bio to add pages to 9142cefe4dbSKent Overstreet * @iter: iov iterator describing the region to be mapped 9152cefe4dbSKent Overstreet * 91617d51b10SMartin Wilck * Pins pages from *iter and appends them to @bio's bvec array. The 9172cefe4dbSKent Overstreet * pages will have to be released using put_page() when done. 91817d51b10SMartin Wilck * For multi-segment *iter, this function only adds pages from the 91917d51b10SMartin Wilck * the next non-empty segment of the iov iterator. 9202cefe4dbSKent Overstreet */ 92117d51b10SMartin Wilck static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 9222cefe4dbSKent Overstreet { 923576ed913SChristoph Hellwig unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt; 924576ed913SChristoph Hellwig unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt; 9252cefe4dbSKent Overstreet struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt; 9262cefe4dbSKent Overstreet struct page **pages = (struct page **)bv; 92745691804SChristoph Hellwig bool same_page = false; 928576ed913SChristoph Hellwig ssize_t size, left; 929576ed913SChristoph Hellwig unsigned len, i; 930b403ea24SMartin Wilck size_t offset; 931576ed913SChristoph Hellwig 932576ed913SChristoph Hellwig /* 933576ed913SChristoph Hellwig * Move page array up in the allocated memory for the bio vecs as far as 934576ed913SChristoph Hellwig * possible so that we can start filling biovecs from the beginning 935576ed913SChristoph Hellwig * without overwriting the temporary page array. 936576ed913SChristoph Hellwig */ 937576ed913SChristoph Hellwig BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2); 938576ed913SChristoph Hellwig pages += entries_left * (PAGE_PTRS_PER_BVEC - 1); 9392cefe4dbSKent Overstreet 9402cefe4dbSKent Overstreet size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset); 9412cefe4dbSKent Overstreet if (unlikely(size <= 0)) 9422cefe4dbSKent Overstreet return size ? size : -EFAULT; 9432cefe4dbSKent Overstreet 944576ed913SChristoph Hellwig for (left = size, i = 0; left > 0; left -= len, i++) { 945576ed913SChristoph Hellwig struct page *page = pages[i]; 9462cefe4dbSKent Overstreet 947576ed913SChristoph Hellwig len = min_t(size_t, PAGE_SIZE - offset, left); 94845691804SChristoph Hellwig 94945691804SChristoph Hellwig if (__bio_try_merge_page(bio, page, len, offset, &same_page)) { 95045691804SChristoph Hellwig if (same_page) 95145691804SChristoph Hellwig put_page(page); 95245691804SChristoph Hellwig } else { 95379d08f89SMing Lei if (WARN_ON_ONCE(bio_full(bio, len))) 954576ed913SChristoph Hellwig return -EINVAL; 95545691804SChristoph Hellwig __bio_add_page(bio, page, len, offset); 95645691804SChristoph Hellwig } 957576ed913SChristoph Hellwig offset = 0; 9582cefe4dbSKent Overstreet } 9592cefe4dbSKent Overstreet 9602cefe4dbSKent Overstreet iov_iter_advance(iter, size); 9612cefe4dbSKent Overstreet return 0; 9622cefe4dbSKent Overstreet } 96317d51b10SMartin Wilck 96417d51b10SMartin Wilck /** 9656d0c48aeSJens Axboe * bio_iov_iter_get_pages - add user or kernel pages to a bio 96617d51b10SMartin Wilck * @bio: bio to add pages to 9676d0c48aeSJens Axboe * @iter: iov iterator describing the region to be added 96817d51b10SMartin Wilck * 9696d0c48aeSJens Axboe * This takes either an iterator pointing to user memory, or one pointing to 9706d0c48aeSJens Axboe * kernel pages (BVEC iterator). If we're adding user pages, we pin them and 9716d0c48aeSJens Axboe * map them into the kernel. On IO completion, the caller should put those 972399254aaSJens Axboe * pages. If we're adding kernel pages, and the caller told us it's safe to 973399254aaSJens Axboe * do so, we just have to add the pages to the bio directly. We don't grab an 974399254aaSJens Axboe * extra reference to those pages (the user should already have that), and we 975399254aaSJens Axboe * don't put the page on IO completion. The caller needs to check if the bio is 976399254aaSJens Axboe * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be 977399254aaSJens Axboe * released. 9786d0c48aeSJens Axboe * 97917d51b10SMartin Wilck * The function tries, but does not guarantee, to pin as many pages as 9806d0c48aeSJens Axboe * fit into the bio, or are requested in *iter, whatever is smaller. If 9816d0c48aeSJens Axboe * MM encounters an error pinning the requested pages, it stops. Error 9826d0c48aeSJens Axboe * is returned only if 0 pages could be pinned. 98317d51b10SMartin Wilck */ 98417d51b10SMartin Wilck int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter) 98517d51b10SMartin Wilck { 9866d0c48aeSJens Axboe const bool is_bvec = iov_iter_is_bvec(iter); 98714eacf12SChristoph Hellwig int ret; 98814eacf12SChristoph Hellwig 98914eacf12SChristoph Hellwig if (WARN_ON_ONCE(bio->bi_vcnt)) 99014eacf12SChristoph Hellwig return -EINVAL; 99117d51b10SMartin Wilck 99217d51b10SMartin Wilck do { 9936d0c48aeSJens Axboe if (is_bvec) 9946d0c48aeSJens Axboe ret = __bio_iov_bvec_add_pages(bio, iter); 9956d0c48aeSJens Axboe else 9966d0c48aeSJens Axboe ret = __bio_iov_iter_get_pages(bio, iter); 99779d08f89SMing Lei } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0)); 99817d51b10SMartin Wilck 999b6207430SChristoph Hellwig if (is_bvec) 10007321ecbfSChristoph Hellwig bio_set_flag(bio, BIO_NO_PAGE_REF); 100114eacf12SChristoph Hellwig return bio->bi_vcnt ? 0 : ret; 100217d51b10SMartin Wilck } 10032cefe4dbSKent Overstreet 10044246a0b6SChristoph Hellwig static void submit_bio_wait_endio(struct bio *bio) 1005f9c78b2bSJens Axboe { 100665e53aabSChristoph Hellwig complete(bio->bi_private); 1007f9c78b2bSJens Axboe } 1008f9c78b2bSJens Axboe 1009f9c78b2bSJens Axboe /** 1010f9c78b2bSJens Axboe * submit_bio_wait - submit a bio, and wait until it completes 1011f9c78b2bSJens Axboe * @bio: The &struct bio which describes the I/O 1012f9c78b2bSJens Axboe * 1013f9c78b2bSJens Axboe * Simple wrapper around submit_bio(). Returns 0 on success, or the error from 1014f9c78b2bSJens Axboe * bio_endio() on failure. 10153d289d68SJan Kara * 10163d289d68SJan Kara * WARNING: Unlike to how submit_bio() is usually used, this function does not 10173d289d68SJan Kara * result in bio reference to be consumed. The caller must drop the reference 10183d289d68SJan Kara * on his own. 1019f9c78b2bSJens Axboe */ 10204e49ea4aSMike Christie int submit_bio_wait(struct bio *bio) 1021f9c78b2bSJens Axboe { 1022e319e1fbSByungchul Park DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map); 1023de6a78b6SMing Lei unsigned long hang_check; 1024f9c78b2bSJens Axboe 102565e53aabSChristoph Hellwig bio->bi_private = &done; 1026f9c78b2bSJens Axboe bio->bi_end_io = submit_bio_wait_endio; 10271eff9d32SJens Axboe bio->bi_opf |= REQ_SYNC; 10284e49ea4aSMike Christie submit_bio(bio); 1029de6a78b6SMing Lei 1030de6a78b6SMing Lei /* Prevent hang_check timer from firing at us during very long I/O */ 1031de6a78b6SMing Lei hang_check = sysctl_hung_task_timeout_secs; 1032de6a78b6SMing Lei if (hang_check) 1033de6a78b6SMing Lei while (!wait_for_completion_io_timeout(&done, 1034de6a78b6SMing Lei hang_check * (HZ/2))) 1035de6a78b6SMing Lei ; 1036de6a78b6SMing Lei else 103765e53aabSChristoph Hellwig wait_for_completion_io(&done); 1038f9c78b2bSJens Axboe 103965e53aabSChristoph Hellwig return blk_status_to_errno(bio->bi_status); 1040f9c78b2bSJens Axboe } 1041f9c78b2bSJens Axboe EXPORT_SYMBOL(submit_bio_wait); 1042f9c78b2bSJens Axboe 1043f9c78b2bSJens Axboe /** 1044f9c78b2bSJens Axboe * bio_advance - increment/complete a bio by some number of bytes 1045f9c78b2bSJens Axboe * @bio: bio to advance 1046f9c78b2bSJens Axboe * @bytes: number of bytes to complete 1047f9c78b2bSJens Axboe * 1048f9c78b2bSJens Axboe * This updates bi_sector, bi_size and bi_idx; if the number of bytes to 1049f9c78b2bSJens Axboe * complete doesn't align with a bvec boundary, then bv_len and bv_offset will 1050f9c78b2bSJens Axboe * be updated on the last bvec as well. 1051f9c78b2bSJens Axboe * 1052f9c78b2bSJens Axboe * @bio will then represent the remaining, uncompleted portion of the io. 1053f9c78b2bSJens Axboe */ 1054f9c78b2bSJens Axboe void bio_advance(struct bio *bio, unsigned bytes) 1055f9c78b2bSJens Axboe { 1056f9c78b2bSJens Axboe if (bio_integrity(bio)) 1057f9c78b2bSJens Axboe bio_integrity_advance(bio, bytes); 1058f9c78b2bSJens Axboe 1059f9c78b2bSJens Axboe bio_advance_iter(bio, &bio->bi_iter, bytes); 1060f9c78b2bSJens Axboe } 1061f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_advance); 1062f9c78b2bSJens Axboe 106345db54d5SKent Overstreet void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter, 106445db54d5SKent Overstreet struct bio *src, struct bvec_iter *src_iter) 1065f9c78b2bSJens Axboe { 1066f9c78b2bSJens Axboe struct bio_vec src_bv, dst_bv; 1067f9c78b2bSJens Axboe void *src_p, *dst_p; 1068f9c78b2bSJens Axboe unsigned bytes; 1069f9c78b2bSJens Axboe 107045db54d5SKent Overstreet while (src_iter->bi_size && dst_iter->bi_size) { 107145db54d5SKent Overstreet src_bv = bio_iter_iovec(src, *src_iter); 107245db54d5SKent Overstreet dst_bv = bio_iter_iovec(dst, *dst_iter); 107345db54d5SKent Overstreet 107445db54d5SKent Overstreet bytes = min(src_bv.bv_len, dst_bv.bv_len); 107545db54d5SKent Overstreet 107645db54d5SKent Overstreet src_p = kmap_atomic(src_bv.bv_page); 107745db54d5SKent Overstreet dst_p = kmap_atomic(dst_bv.bv_page); 107845db54d5SKent Overstreet 107945db54d5SKent Overstreet memcpy(dst_p + dst_bv.bv_offset, 108045db54d5SKent Overstreet src_p + src_bv.bv_offset, 108145db54d5SKent Overstreet bytes); 108245db54d5SKent Overstreet 108345db54d5SKent Overstreet kunmap_atomic(dst_p); 108445db54d5SKent Overstreet kunmap_atomic(src_p); 108545db54d5SKent Overstreet 10866e6e811dSKent Overstreet flush_dcache_page(dst_bv.bv_page); 10876e6e811dSKent Overstreet 108845db54d5SKent Overstreet bio_advance_iter(src, src_iter, bytes); 108945db54d5SKent Overstreet bio_advance_iter(dst, dst_iter, bytes); 109045db54d5SKent Overstreet } 109145db54d5SKent Overstreet } 109245db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data_iter); 109345db54d5SKent Overstreet 109445db54d5SKent Overstreet /** 109545db54d5SKent Overstreet * bio_copy_data - copy contents of data buffers from one bio to another 109645db54d5SKent Overstreet * @src: source bio 109745db54d5SKent Overstreet * @dst: destination bio 109845db54d5SKent Overstreet * 109945db54d5SKent Overstreet * Stops when it reaches the end of either @src or @dst - that is, copies 110045db54d5SKent Overstreet * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios). 110145db54d5SKent Overstreet */ 110245db54d5SKent Overstreet void bio_copy_data(struct bio *dst, struct bio *src) 110345db54d5SKent Overstreet { 110445db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 110545db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 110645db54d5SKent Overstreet 110745db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 110845db54d5SKent Overstreet } 110945db54d5SKent Overstreet EXPORT_SYMBOL(bio_copy_data); 111045db54d5SKent Overstreet 111145db54d5SKent Overstreet /** 111245db54d5SKent Overstreet * bio_list_copy_data - copy contents of data buffers from one chain of bios to 111345db54d5SKent Overstreet * another 111445db54d5SKent Overstreet * @src: source bio list 111545db54d5SKent Overstreet * @dst: destination bio list 111645db54d5SKent Overstreet * 111745db54d5SKent Overstreet * Stops when it reaches the end of either the @src list or @dst list - that is, 111845db54d5SKent Overstreet * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of 111945db54d5SKent Overstreet * bios). 112045db54d5SKent Overstreet */ 112145db54d5SKent Overstreet void bio_list_copy_data(struct bio *dst, struct bio *src) 112245db54d5SKent Overstreet { 112345db54d5SKent Overstreet struct bvec_iter src_iter = src->bi_iter; 112445db54d5SKent Overstreet struct bvec_iter dst_iter = dst->bi_iter; 112545db54d5SKent Overstreet 1126f9c78b2bSJens Axboe while (1) { 1127f9c78b2bSJens Axboe if (!src_iter.bi_size) { 1128f9c78b2bSJens Axboe src = src->bi_next; 1129f9c78b2bSJens Axboe if (!src) 1130f9c78b2bSJens Axboe break; 1131f9c78b2bSJens Axboe 1132f9c78b2bSJens Axboe src_iter = src->bi_iter; 1133f9c78b2bSJens Axboe } 1134f9c78b2bSJens Axboe 1135f9c78b2bSJens Axboe if (!dst_iter.bi_size) { 1136f9c78b2bSJens Axboe dst = dst->bi_next; 1137f9c78b2bSJens Axboe if (!dst) 1138f9c78b2bSJens Axboe break; 1139f9c78b2bSJens Axboe 1140f9c78b2bSJens Axboe dst_iter = dst->bi_iter; 1141f9c78b2bSJens Axboe } 1142f9c78b2bSJens Axboe 114345db54d5SKent Overstreet bio_copy_data_iter(dst, &dst_iter, src, &src_iter); 1144f9c78b2bSJens Axboe } 1145f9c78b2bSJens Axboe } 114645db54d5SKent Overstreet EXPORT_SYMBOL(bio_list_copy_data); 1147f9c78b2bSJens Axboe 1148f9c78b2bSJens Axboe struct bio_map_data { 1149f9c78b2bSJens Axboe int is_our_pages; 115026e49cfcSKent Overstreet struct iov_iter iter; 115126e49cfcSKent Overstreet struct iovec iov[]; 1152f9c78b2bSJens Axboe }; 1153f9c78b2bSJens Axboe 11540e5b935dSAl Viro static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, 1155f9c78b2bSJens Axboe gfp_t gfp_mask) 1156f9c78b2bSJens Axboe { 11570e5b935dSAl Viro struct bio_map_data *bmd; 11580e5b935dSAl Viro if (data->nr_segs > UIO_MAXIOV) 1159f9c78b2bSJens Axboe return NULL; 1160f9c78b2bSJens Axboe 1161f1f8f292SGustavo A. R. Silva bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); 11620e5b935dSAl Viro if (!bmd) 11630e5b935dSAl Viro return NULL; 11640e5b935dSAl Viro memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs); 11650e5b935dSAl Viro bmd->iter = *data; 11660e5b935dSAl Viro bmd->iter.iov = bmd->iov; 11670e5b935dSAl Viro return bmd; 1168f9c78b2bSJens Axboe } 1169f9c78b2bSJens Axboe 11709124d3feSDongsu Park /** 11719124d3feSDongsu Park * bio_copy_from_iter - copy all pages from iov_iter to bio 11729124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as destination 11739124d3feSDongsu Park * @iter: iov_iter as source 11749124d3feSDongsu Park * 11759124d3feSDongsu Park * Copy all pages from iov_iter to bio. 11769124d3feSDongsu Park * Returns 0 on success, or error on failure. 11779124d3feSDongsu Park */ 117898a09d61SAl Viro static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) 1179f9c78b2bSJens Axboe { 1180f9c78b2bSJens Axboe struct bio_vec *bvec; 11816dc4f100SMing Lei struct bvec_iter_all iter_all; 1182f9c78b2bSJens Axboe 11832b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 11849124d3feSDongsu Park ssize_t ret; 1185f9c78b2bSJens Axboe 11869124d3feSDongsu Park ret = copy_page_from_iter(bvec->bv_page, 11879124d3feSDongsu Park bvec->bv_offset, 11889124d3feSDongsu Park bvec->bv_len, 118998a09d61SAl Viro iter); 1190f9c78b2bSJens Axboe 119198a09d61SAl Viro if (!iov_iter_count(iter)) 11929124d3feSDongsu Park break; 1193f9c78b2bSJens Axboe 11949124d3feSDongsu Park if (ret < bvec->bv_len) 11959124d3feSDongsu Park return -EFAULT; 1196f9c78b2bSJens Axboe } 1197f9c78b2bSJens Axboe 11989124d3feSDongsu Park return 0; 1199f9c78b2bSJens Axboe } 1200f9c78b2bSJens Axboe 12019124d3feSDongsu Park /** 12029124d3feSDongsu Park * bio_copy_to_iter - copy all pages from bio to iov_iter 12039124d3feSDongsu Park * @bio: The &struct bio which describes the I/O as source 12049124d3feSDongsu Park * @iter: iov_iter as destination 12059124d3feSDongsu Park * 12069124d3feSDongsu Park * Copy all pages from bio to iov_iter. 12079124d3feSDongsu Park * Returns 0 on success, or error on failure. 12089124d3feSDongsu Park */ 12099124d3feSDongsu Park static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) 12109124d3feSDongsu Park { 12119124d3feSDongsu Park struct bio_vec *bvec; 12126dc4f100SMing Lei struct bvec_iter_all iter_all; 12139124d3feSDongsu Park 12142b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 12159124d3feSDongsu Park ssize_t ret; 12169124d3feSDongsu Park 12179124d3feSDongsu Park ret = copy_page_to_iter(bvec->bv_page, 12189124d3feSDongsu Park bvec->bv_offset, 12199124d3feSDongsu Park bvec->bv_len, 12209124d3feSDongsu Park &iter); 12219124d3feSDongsu Park 12229124d3feSDongsu Park if (!iov_iter_count(&iter)) 12239124d3feSDongsu Park break; 12249124d3feSDongsu Park 12259124d3feSDongsu Park if (ret < bvec->bv_len) 12269124d3feSDongsu Park return -EFAULT; 12279124d3feSDongsu Park } 12289124d3feSDongsu Park 12299124d3feSDongsu Park return 0; 1230f9c78b2bSJens Axboe } 1231f9c78b2bSJens Axboe 1232491221f8SGuoqing Jiang void bio_free_pages(struct bio *bio) 12331dfa0f68SChristoph Hellwig { 12341dfa0f68SChristoph Hellwig struct bio_vec *bvec; 12356dc4f100SMing Lei struct bvec_iter_all iter_all; 12361dfa0f68SChristoph Hellwig 12372b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) 12381dfa0f68SChristoph Hellwig __free_page(bvec->bv_page); 12391dfa0f68SChristoph Hellwig } 1240491221f8SGuoqing Jiang EXPORT_SYMBOL(bio_free_pages); 12411dfa0f68SChristoph Hellwig 1242f9c78b2bSJens Axboe /** 1243f9c78b2bSJens Axboe * bio_uncopy_user - finish previously mapped bio 1244f9c78b2bSJens Axboe * @bio: bio being terminated 1245f9c78b2bSJens Axboe * 1246ddad8dd0SChristoph Hellwig * Free pages allocated from bio_copy_user_iov() and write back data 1247f9c78b2bSJens Axboe * to user space in case of a read. 1248f9c78b2bSJens Axboe */ 1249f9c78b2bSJens Axboe int bio_uncopy_user(struct bio *bio) 1250f9c78b2bSJens Axboe { 1251f9c78b2bSJens Axboe struct bio_map_data *bmd = bio->bi_private; 12521dfa0f68SChristoph Hellwig int ret = 0; 1253f9c78b2bSJens Axboe 1254f9c78b2bSJens Axboe if (!bio_flagged(bio, BIO_NULL_MAPPED)) { 1255f9c78b2bSJens Axboe /* 1256f9c78b2bSJens Axboe * if we're in a workqueue, the request is orphaned, so 12572d99b55dSHannes Reinecke * don't copy into a random user address space, just free 12582d99b55dSHannes Reinecke * and return -EINTR so user space doesn't expect any data. 1259f9c78b2bSJens Axboe */ 12602d99b55dSHannes Reinecke if (!current->mm) 12612d99b55dSHannes Reinecke ret = -EINTR; 12622d99b55dSHannes Reinecke else if (bio_data_dir(bio) == READ) 12639124d3feSDongsu Park ret = bio_copy_to_iter(bio, bmd->iter); 12641dfa0f68SChristoph Hellwig if (bmd->is_our_pages) 12651dfa0f68SChristoph Hellwig bio_free_pages(bio); 1266f9c78b2bSJens Axboe } 1267f9c78b2bSJens Axboe kfree(bmd); 1268f9c78b2bSJens Axboe bio_put(bio); 1269f9c78b2bSJens Axboe return ret; 1270f9c78b2bSJens Axboe } 1271f9c78b2bSJens Axboe 1272f9c78b2bSJens Axboe /** 1273f9c78b2bSJens Axboe * bio_copy_user_iov - copy user data to bio 1274f9c78b2bSJens Axboe * @q: destination block queue 1275f9c78b2bSJens Axboe * @map_data: pointer to the rq_map_data holding pages (if necessary) 127626e49cfcSKent Overstreet * @iter: iovec iterator 1277f9c78b2bSJens Axboe * @gfp_mask: memory allocation flags 1278f9c78b2bSJens Axboe * 1279f9c78b2bSJens Axboe * Prepares and returns a bio for indirect user io, bouncing data 1280f9c78b2bSJens Axboe * to/from kernel pages as necessary. Must be paired with 1281f9c78b2bSJens Axboe * call bio_uncopy_user() on io completion. 1282f9c78b2bSJens Axboe */ 1283f9c78b2bSJens Axboe struct bio *bio_copy_user_iov(struct request_queue *q, 1284f9c78b2bSJens Axboe struct rq_map_data *map_data, 1285e81cef5dSAl Viro struct iov_iter *iter, 128626e49cfcSKent Overstreet gfp_t gfp_mask) 1287f9c78b2bSJens Axboe { 1288f9c78b2bSJens Axboe struct bio_map_data *bmd; 1289f9c78b2bSJens Axboe struct page *page; 1290f9c78b2bSJens Axboe struct bio *bio; 1291d16d44ebSAl Viro int i = 0, ret; 1292d16d44ebSAl Viro int nr_pages; 129326e49cfcSKent Overstreet unsigned int len = iter->count; 1294bd5ceceaSGeliang Tang unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; 1295f9c78b2bSJens Axboe 12960e5b935dSAl Viro bmd = bio_alloc_map_data(iter, gfp_mask); 1297f9c78b2bSJens Axboe if (!bmd) 1298f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1299f9c78b2bSJens Axboe 130026e49cfcSKent Overstreet /* 130126e49cfcSKent Overstreet * We need to do a deep copy of the iov_iter including the iovecs. 130226e49cfcSKent Overstreet * The caller provided iov might point to an on-stack or otherwise 130326e49cfcSKent Overstreet * shortlived one. 130426e49cfcSKent Overstreet */ 130526e49cfcSKent Overstreet bmd->is_our_pages = map_data ? 0 : 1; 130626e49cfcSKent Overstreet 1307d16d44ebSAl Viro nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE); 1308d16d44ebSAl Viro if (nr_pages > BIO_MAX_PAGES) 1309d16d44ebSAl Viro nr_pages = BIO_MAX_PAGES; 1310f9c78b2bSJens Axboe 1311f9c78b2bSJens Axboe ret = -ENOMEM; 1312f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1313f9c78b2bSJens Axboe if (!bio) 1314f9c78b2bSJens Axboe goto out_bmd; 1315f9c78b2bSJens Axboe 1316f9c78b2bSJens Axboe ret = 0; 1317f9c78b2bSJens Axboe 1318f9c78b2bSJens Axboe if (map_data) { 1319f9c78b2bSJens Axboe nr_pages = 1 << map_data->page_order; 1320f9c78b2bSJens Axboe i = map_data->offset / PAGE_SIZE; 1321f9c78b2bSJens Axboe } 1322f9c78b2bSJens Axboe while (len) { 1323f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE; 1324f9c78b2bSJens Axboe 1325f9c78b2bSJens Axboe bytes -= offset; 1326f9c78b2bSJens Axboe 1327f9c78b2bSJens Axboe if (bytes > len) 1328f9c78b2bSJens Axboe bytes = len; 1329f9c78b2bSJens Axboe 1330f9c78b2bSJens Axboe if (map_data) { 1331f9c78b2bSJens Axboe if (i == map_data->nr_entries * nr_pages) { 1332f9c78b2bSJens Axboe ret = -ENOMEM; 1333f9c78b2bSJens Axboe break; 1334f9c78b2bSJens Axboe } 1335f9c78b2bSJens Axboe 1336f9c78b2bSJens Axboe page = map_data->pages[i / nr_pages]; 1337f9c78b2bSJens Axboe page += (i % nr_pages); 1338f9c78b2bSJens Axboe 1339f9c78b2bSJens Axboe i++; 1340f9c78b2bSJens Axboe } else { 1341f9c78b2bSJens Axboe page = alloc_page(q->bounce_gfp | gfp_mask); 1342f9c78b2bSJens Axboe if (!page) { 1343f9c78b2bSJens Axboe ret = -ENOMEM; 1344f9c78b2bSJens Axboe break; 1345f9c78b2bSJens Axboe } 1346f9c78b2bSJens Axboe } 1347f9c78b2bSJens Axboe 1348a3761c3cSJérôme Glisse if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes) { 1349a3761c3cSJérôme Glisse if (!map_data) 1350a3761c3cSJérôme Glisse __free_page(page); 1351f9c78b2bSJens Axboe break; 1352a3761c3cSJérôme Glisse } 1353f9c78b2bSJens Axboe 1354f9c78b2bSJens Axboe len -= bytes; 1355f9c78b2bSJens Axboe offset = 0; 1356f9c78b2bSJens Axboe } 1357f9c78b2bSJens Axboe 1358f9c78b2bSJens Axboe if (ret) 1359f9c78b2bSJens Axboe goto cleanup; 1360f9c78b2bSJens Axboe 13612884d0beSAl Viro if (map_data) 13622884d0beSAl Viro map_data->offset += bio->bi_iter.bi_size; 13632884d0beSAl Viro 1364f9c78b2bSJens Axboe /* 1365f9c78b2bSJens Axboe * success 1366f9c78b2bSJens Axboe */ 136700e23707SDavid Howells if ((iov_iter_rw(iter) == WRITE && (!map_data || !map_data->null_mapped)) || 1368f9c78b2bSJens Axboe (map_data && map_data->from_user)) { 136998a09d61SAl Viro ret = bio_copy_from_iter(bio, iter); 1370f9c78b2bSJens Axboe if (ret) 1371f9c78b2bSJens Axboe goto cleanup; 137298a09d61SAl Viro } else { 1373f55adad6SKeith Busch if (bmd->is_our_pages) 1374f3587d76SKeith Busch zero_fill_bio(bio); 1375e81cef5dSAl Viro iov_iter_advance(iter, bio->bi_iter.bi_size); 1376f9c78b2bSJens Axboe } 1377f9c78b2bSJens Axboe 137826e49cfcSKent Overstreet bio->bi_private = bmd; 13792884d0beSAl Viro if (map_data && map_data->null_mapped) 13802884d0beSAl Viro bio_set_flag(bio, BIO_NULL_MAPPED); 1381f9c78b2bSJens Axboe return bio; 1382f9c78b2bSJens Axboe cleanup: 1383f9c78b2bSJens Axboe if (!map_data) 13841dfa0f68SChristoph Hellwig bio_free_pages(bio); 1385f9c78b2bSJens Axboe bio_put(bio); 1386f9c78b2bSJens Axboe out_bmd: 1387f9c78b2bSJens Axboe kfree(bmd); 1388f9c78b2bSJens Axboe return ERR_PTR(ret); 1389f9c78b2bSJens Axboe } 1390f9c78b2bSJens Axboe 139137f19e57SChristoph Hellwig /** 139237f19e57SChristoph Hellwig * bio_map_user_iov - map user iovec into bio 139337f19e57SChristoph Hellwig * @q: the struct request_queue for the bio 139437f19e57SChristoph Hellwig * @iter: iovec iterator 139537f19e57SChristoph Hellwig * @gfp_mask: memory allocation flags 139637f19e57SChristoph Hellwig * 139737f19e57SChristoph Hellwig * Map the user space address into a bio suitable for io to a block 139837f19e57SChristoph Hellwig * device. Returns an error pointer in case of error. 139937f19e57SChristoph Hellwig */ 140037f19e57SChristoph Hellwig struct bio *bio_map_user_iov(struct request_queue *q, 1401e81cef5dSAl Viro struct iov_iter *iter, 140226e49cfcSKent Overstreet gfp_t gfp_mask) 1403f9c78b2bSJens Axboe { 140426e49cfcSKent Overstreet int j; 1405f9c78b2bSJens Axboe struct bio *bio; 1406076098e5SAl Viro int ret; 1407f9c78b2bSJens Axboe 1408b282cc76SAl Viro if (!iov_iter_count(iter)) 1409f9c78b2bSJens Axboe return ERR_PTR(-EINVAL); 1410f9c78b2bSJens Axboe 1411b282cc76SAl Viro bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES)); 1412f9c78b2bSJens Axboe if (!bio) 1413f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1414f9c78b2bSJens Axboe 14150a0f1513SAl Viro while (iov_iter_count(iter)) { 1416629e42bcSAl Viro struct page **pages; 1417076098e5SAl Viro ssize_t bytes; 1418076098e5SAl Viro size_t offs, added = 0; 1419076098e5SAl Viro int npages; 1420f9c78b2bSJens Axboe 14210a0f1513SAl Viro bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs); 1422076098e5SAl Viro if (unlikely(bytes <= 0)) { 1423076098e5SAl Viro ret = bytes ? bytes : -EFAULT; 1424f9c78b2bSJens Axboe goto out_unmap; 1425f9c78b2bSJens Axboe } 1426f9c78b2bSJens Axboe 1427076098e5SAl Viro npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); 1428076098e5SAl Viro 142998f0bc99SAl Viro if (unlikely(offs & queue_dma_alignment(q))) { 143098f0bc99SAl Viro ret = -EINVAL; 143198f0bc99SAl Viro j = 0; 143298f0bc99SAl Viro } else { 1433629e42bcSAl Viro for (j = 0; j < npages; j++) { 143498f0bc99SAl Viro struct page *page = pages[j]; 1435076098e5SAl Viro unsigned int n = PAGE_SIZE - offs; 1436d1916c86SChristoph Hellwig bool same_page = false; 1437f9c78b2bSJens Axboe 1438076098e5SAl Viro if (n > bytes) 1439076098e5SAl Viro n = bytes; 1440f9c78b2bSJens Axboe 144119047087SMing Lei if (!__bio_add_pc_page(q, bio, page, n, offs, 1442d1916c86SChristoph Hellwig &same_page)) { 1443d1916c86SChristoph Hellwig if (same_page) 1444d1916c86SChristoph Hellwig put_page(page); 1445f9c78b2bSJens Axboe break; 1446d1916c86SChristoph Hellwig } 1447f9c78b2bSJens Axboe 1448076098e5SAl Viro added += n; 1449076098e5SAl Viro bytes -= n; 1450076098e5SAl Viro offs = 0; 1451f9c78b2bSJens Axboe } 14520a0f1513SAl Viro iov_iter_advance(iter, added); 145398f0bc99SAl Viro } 1454f9c78b2bSJens Axboe /* 1455f9c78b2bSJens Axboe * release the pages we didn't map into the bio, if any 1456f9c78b2bSJens Axboe */ 1457629e42bcSAl Viro while (j < npages) 145809cbfeafSKirill A. Shutemov put_page(pages[j++]); 1459629e42bcSAl Viro kvfree(pages); 1460e2e115d1SAl Viro /* couldn't stuff something into bio? */ 1461e2e115d1SAl Viro if (bytes) 1462e2e115d1SAl Viro break; 1463f9c78b2bSJens Axboe } 1464f9c78b2bSJens Axboe 1465b7c44ed9SJens Axboe bio_set_flag(bio, BIO_USER_MAPPED); 146637f19e57SChristoph Hellwig 146737f19e57SChristoph Hellwig /* 14685fad1b64SBart Van Assche * subtle -- if bio_map_user_iov() ended up bouncing a bio, 146937f19e57SChristoph Hellwig * it would normally disappear when its bi_end_io is run. 147037f19e57SChristoph Hellwig * however, we need it for the unmap, so grab an extra 147137f19e57SChristoph Hellwig * reference to it 147237f19e57SChristoph Hellwig */ 147337f19e57SChristoph Hellwig bio_get(bio); 1474f9c78b2bSJens Axboe return bio; 1475f9c78b2bSJens Axboe 1476f9c78b2bSJens Axboe out_unmap: 1477506e0798SChristoph Hellwig bio_release_pages(bio, false); 1478f9c78b2bSJens Axboe bio_put(bio); 1479f9c78b2bSJens Axboe return ERR_PTR(ret); 1480f9c78b2bSJens Axboe } 1481f9c78b2bSJens Axboe 1482f9c78b2bSJens Axboe /** 1483f9c78b2bSJens Axboe * bio_unmap_user - unmap a bio 1484f9c78b2bSJens Axboe * @bio: the bio being unmapped 1485f9c78b2bSJens Axboe * 14865fad1b64SBart Van Assche * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from 14875fad1b64SBart Van Assche * process context. 1488f9c78b2bSJens Axboe * 1489f9c78b2bSJens Axboe * bio_unmap_user() may sleep. 1490f9c78b2bSJens Axboe */ 1491f9c78b2bSJens Axboe void bio_unmap_user(struct bio *bio) 1492f9c78b2bSJens Axboe { 1493163cc2d3SChristoph Hellwig bio_release_pages(bio, bio_data_dir(bio) == READ); 1494163cc2d3SChristoph Hellwig bio_put(bio); 1495f9c78b2bSJens Axboe bio_put(bio); 1496f9c78b2bSJens Axboe } 1497f9c78b2bSJens Axboe 1498b4c5875dSDamien Le Moal static void bio_invalidate_vmalloc_pages(struct bio *bio) 1499b4c5875dSDamien Le Moal { 1500b4c5875dSDamien Le Moal #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE 1501b4c5875dSDamien Le Moal if (bio->bi_private && !op_is_write(bio_op(bio))) { 1502b4c5875dSDamien Le Moal unsigned long i, len = 0; 1503b4c5875dSDamien Le Moal 1504b4c5875dSDamien Le Moal for (i = 0; i < bio->bi_vcnt; i++) 1505b4c5875dSDamien Le Moal len += bio->bi_io_vec[i].bv_len; 1506b4c5875dSDamien Le Moal invalidate_kernel_vmap_range(bio->bi_private, len); 1507b4c5875dSDamien Le Moal } 1508b4c5875dSDamien Le Moal #endif 1509b4c5875dSDamien Le Moal } 1510b4c5875dSDamien Le Moal 15114246a0b6SChristoph Hellwig static void bio_map_kern_endio(struct bio *bio) 1512f9c78b2bSJens Axboe { 1513b4c5875dSDamien Le Moal bio_invalidate_vmalloc_pages(bio); 1514f9c78b2bSJens Axboe bio_put(bio); 1515f9c78b2bSJens Axboe } 1516f9c78b2bSJens Axboe 151775c72b83SChristoph Hellwig /** 151875c72b83SChristoph Hellwig * bio_map_kern - map kernel address into bio 151975c72b83SChristoph Hellwig * @q: the struct request_queue for the bio 152075c72b83SChristoph Hellwig * @data: pointer to buffer to map 152175c72b83SChristoph Hellwig * @len: length in bytes 152275c72b83SChristoph Hellwig * @gfp_mask: allocation flags for bio allocation 152375c72b83SChristoph Hellwig * 152475c72b83SChristoph Hellwig * Map the kernel address into a bio suitable for io to a block 152575c72b83SChristoph Hellwig * device. Returns an error pointer in case of error. 152675c72b83SChristoph Hellwig */ 152775c72b83SChristoph Hellwig struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len, 152875c72b83SChristoph Hellwig gfp_t gfp_mask) 1529f9c78b2bSJens Axboe { 1530f9c78b2bSJens Axboe unsigned long kaddr = (unsigned long)data; 1531f9c78b2bSJens Axboe unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 1532f9c78b2bSJens Axboe unsigned long start = kaddr >> PAGE_SHIFT; 1533f9c78b2bSJens Axboe const int nr_pages = end - start; 1534b4c5875dSDamien Le Moal bool is_vmalloc = is_vmalloc_addr(data); 1535b4c5875dSDamien Le Moal struct page *page; 1536f9c78b2bSJens Axboe int offset, i; 1537f9c78b2bSJens Axboe struct bio *bio; 1538f9c78b2bSJens Axboe 1539f9c78b2bSJens Axboe bio = bio_kmalloc(gfp_mask, nr_pages); 1540f9c78b2bSJens Axboe if (!bio) 1541f9c78b2bSJens Axboe return ERR_PTR(-ENOMEM); 1542f9c78b2bSJens Axboe 1543b4c5875dSDamien Le Moal if (is_vmalloc) { 1544b4c5875dSDamien Le Moal flush_kernel_vmap_range(data, len); 1545b4c5875dSDamien Le Moal bio->bi_private = data; 1546b4c5875dSDamien Le Moal } 1547b4c5875dSDamien Le Moal 1548f9c78b2bSJens Axboe offset = offset_in_page(kaddr); 1549f9c78b2bSJens Axboe for (i = 0; i < nr_pages; i++) { 1550f9c78b2bSJens Axboe unsigned int bytes = PAGE_SIZE - offset; 1551f9c78b2bSJens Axboe 1552f9c78b2bSJens Axboe if (len <= 0) 1553f9c78b2bSJens Axboe break; 1554f9c78b2bSJens Axboe 1555f9c78b2bSJens Axboe if (bytes > len) 1556f9c78b2bSJens Axboe bytes = len; 1557f9c78b2bSJens Axboe 1558b4c5875dSDamien Le Moal if (!is_vmalloc) 1559b4c5875dSDamien Le Moal page = virt_to_page(data); 1560b4c5875dSDamien Le Moal else 1561b4c5875dSDamien Le Moal page = vmalloc_to_page(data); 1562b4c5875dSDamien Le Moal if (bio_add_pc_page(q, bio, page, bytes, 156375c72b83SChristoph Hellwig offset) < bytes) { 156475c72b83SChristoph Hellwig /* we don't support partial mappings */ 156575c72b83SChristoph Hellwig bio_put(bio); 156675c72b83SChristoph Hellwig return ERR_PTR(-EINVAL); 156775c72b83SChristoph Hellwig } 1568f9c78b2bSJens Axboe 1569f9c78b2bSJens Axboe data += bytes; 1570f9c78b2bSJens Axboe len -= bytes; 1571f9c78b2bSJens Axboe offset = 0; 1572f9c78b2bSJens Axboe } 1573f9c78b2bSJens Axboe 1574f9c78b2bSJens Axboe bio->bi_end_io = bio_map_kern_endio; 1575f9c78b2bSJens Axboe return bio; 1576f9c78b2bSJens Axboe } 1577f9c78b2bSJens Axboe 15784246a0b6SChristoph Hellwig static void bio_copy_kern_endio(struct bio *bio) 1579f9c78b2bSJens Axboe { 15801dfa0f68SChristoph Hellwig bio_free_pages(bio); 15811dfa0f68SChristoph Hellwig bio_put(bio); 15821dfa0f68SChristoph Hellwig } 15831dfa0f68SChristoph Hellwig 15844246a0b6SChristoph Hellwig static void bio_copy_kern_endio_read(struct bio *bio) 15851dfa0f68SChristoph Hellwig { 158642d2683aSChristoph Hellwig char *p = bio->bi_private; 15871dfa0f68SChristoph Hellwig struct bio_vec *bvec; 15886dc4f100SMing Lei struct bvec_iter_all iter_all; 1589f9c78b2bSJens Axboe 15902b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 15911dfa0f68SChristoph Hellwig memcpy(p, page_address(bvec->bv_page), bvec->bv_len); 1592f9c78b2bSJens Axboe p += bvec->bv_len; 1593f9c78b2bSJens Axboe } 1594f9c78b2bSJens Axboe 15954246a0b6SChristoph Hellwig bio_copy_kern_endio(bio); 1596f9c78b2bSJens Axboe } 1597f9c78b2bSJens Axboe 1598f9c78b2bSJens Axboe /** 1599f9c78b2bSJens Axboe * bio_copy_kern - copy kernel address into bio 1600f9c78b2bSJens Axboe * @q: the struct request_queue for the bio 1601f9c78b2bSJens Axboe * @data: pointer to buffer to copy 1602f9c78b2bSJens Axboe * @len: length in bytes 1603f9c78b2bSJens Axboe * @gfp_mask: allocation flags for bio and page allocation 1604f9c78b2bSJens Axboe * @reading: data direction is READ 1605f9c78b2bSJens Axboe * 1606f9c78b2bSJens Axboe * copy the kernel address into a bio suitable for io to a block 1607f9c78b2bSJens Axboe * device. Returns an error pointer in case of error. 1608f9c78b2bSJens Axboe */ 1609f9c78b2bSJens Axboe struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len, 1610f9c78b2bSJens Axboe gfp_t gfp_mask, int reading) 1611f9c78b2bSJens Axboe { 161242d2683aSChristoph Hellwig unsigned long kaddr = (unsigned long)data; 161342d2683aSChristoph Hellwig unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; 161442d2683aSChristoph Hellwig unsigned long start = kaddr >> PAGE_SHIFT; 161542d2683aSChristoph Hellwig struct bio *bio; 1616f9c78b2bSJens Axboe void *p = data; 16171dfa0f68SChristoph Hellwig int nr_pages = 0; 1618f9c78b2bSJens Axboe 161942d2683aSChristoph Hellwig /* 162042d2683aSChristoph Hellwig * Overflow, abort 162142d2683aSChristoph Hellwig */ 162242d2683aSChristoph Hellwig if (end < start) 162342d2683aSChristoph Hellwig return ERR_PTR(-EINVAL); 1624f9c78b2bSJens Axboe 162542d2683aSChristoph Hellwig nr_pages = end - start; 162642d2683aSChristoph Hellwig bio = bio_kmalloc(gfp_mask, nr_pages); 162742d2683aSChristoph Hellwig if (!bio) 162842d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 162942d2683aSChristoph Hellwig 163042d2683aSChristoph Hellwig while (len) { 163142d2683aSChristoph Hellwig struct page *page; 163242d2683aSChristoph Hellwig unsigned int bytes = PAGE_SIZE; 163342d2683aSChristoph Hellwig 163442d2683aSChristoph Hellwig if (bytes > len) 163542d2683aSChristoph Hellwig bytes = len; 163642d2683aSChristoph Hellwig 163742d2683aSChristoph Hellwig page = alloc_page(q->bounce_gfp | gfp_mask); 163842d2683aSChristoph Hellwig if (!page) 163942d2683aSChristoph Hellwig goto cleanup; 164042d2683aSChristoph Hellwig 164142d2683aSChristoph Hellwig if (!reading) 164242d2683aSChristoph Hellwig memcpy(page_address(page), p, bytes); 164342d2683aSChristoph Hellwig 164442d2683aSChristoph Hellwig if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) 164542d2683aSChristoph Hellwig break; 164642d2683aSChristoph Hellwig 164742d2683aSChristoph Hellwig len -= bytes; 164842d2683aSChristoph Hellwig p += bytes; 1649f9c78b2bSJens Axboe } 1650f9c78b2bSJens Axboe 16511dfa0f68SChristoph Hellwig if (reading) { 16521dfa0f68SChristoph Hellwig bio->bi_end_io = bio_copy_kern_endio_read; 165342d2683aSChristoph Hellwig bio->bi_private = data; 16541dfa0f68SChristoph Hellwig } else { 1655f9c78b2bSJens Axboe bio->bi_end_io = bio_copy_kern_endio; 16561dfa0f68SChristoph Hellwig } 16571dfa0f68SChristoph Hellwig 1658f9c78b2bSJens Axboe return bio; 165942d2683aSChristoph Hellwig 166042d2683aSChristoph Hellwig cleanup: 16611dfa0f68SChristoph Hellwig bio_free_pages(bio); 166242d2683aSChristoph Hellwig bio_put(bio); 166342d2683aSChristoph Hellwig return ERR_PTR(-ENOMEM); 1664f9c78b2bSJens Axboe } 1665f9c78b2bSJens Axboe 1666f9c78b2bSJens Axboe /* 1667f9c78b2bSJens Axboe * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions 1668f9c78b2bSJens Axboe * for performing direct-IO in BIOs. 1669f9c78b2bSJens Axboe * 1670f9c78b2bSJens Axboe * The problem is that we cannot run set_page_dirty() from interrupt context 1671f9c78b2bSJens Axboe * because the required locks are not interrupt-safe. So what we can do is to 1672f9c78b2bSJens Axboe * mark the pages dirty _before_ performing IO. And in interrupt context, 1673f9c78b2bSJens Axboe * check that the pages are still dirty. If so, fine. If not, redirty them 1674f9c78b2bSJens Axboe * in process context. 1675f9c78b2bSJens Axboe * 1676f9c78b2bSJens Axboe * We special-case compound pages here: normally this means reads into hugetlb 1677f9c78b2bSJens Axboe * pages. The logic in here doesn't really work right for compound pages 1678f9c78b2bSJens Axboe * because the VM does not uniformly chase down the head page in all cases. 1679f9c78b2bSJens Axboe * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't 1680f9c78b2bSJens Axboe * handle them at all. So we skip compound pages here at an early stage. 1681f9c78b2bSJens Axboe * 1682f9c78b2bSJens Axboe * Note that this code is very hard to test under normal circumstances because 1683f9c78b2bSJens Axboe * direct-io pins the pages with get_user_pages(). This makes 1684f9c78b2bSJens Axboe * is_page_cache_freeable return false, and the VM will not clean the pages. 1685f9c78b2bSJens Axboe * But other code (eg, flusher threads) could clean the pages if they are mapped 1686f9c78b2bSJens Axboe * pagecache. 1687f9c78b2bSJens Axboe * 1688f9c78b2bSJens Axboe * Simply disabling the call to bio_set_pages_dirty() is a good way to test the 1689f9c78b2bSJens Axboe * deferred bio dirtying paths. 1690f9c78b2bSJens Axboe */ 1691f9c78b2bSJens Axboe 1692f9c78b2bSJens Axboe /* 1693f9c78b2bSJens Axboe * bio_set_pages_dirty() will mark all the bio's pages as dirty. 1694f9c78b2bSJens Axboe */ 1695f9c78b2bSJens Axboe void bio_set_pages_dirty(struct bio *bio) 1696f9c78b2bSJens Axboe { 1697f9c78b2bSJens Axboe struct bio_vec *bvec; 16986dc4f100SMing Lei struct bvec_iter_all iter_all; 1699f9c78b2bSJens Axboe 17002b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 17013bb50983SChristoph Hellwig if (!PageCompound(bvec->bv_page)) 17023bb50983SChristoph Hellwig set_page_dirty_lock(bvec->bv_page); 1703f9c78b2bSJens Axboe } 1704f9c78b2bSJens Axboe } 1705f9c78b2bSJens Axboe 1706f9c78b2bSJens Axboe /* 1707f9c78b2bSJens Axboe * bio_check_pages_dirty() will check that all the BIO's pages are still dirty. 1708f9c78b2bSJens Axboe * If they are, then fine. If, however, some pages are clean then they must 1709f9c78b2bSJens Axboe * have been written out during the direct-IO read. So we take another ref on 171024d5493fSChristoph Hellwig * the BIO and re-dirty the pages in process context. 1711f9c78b2bSJens Axboe * 1712f9c78b2bSJens Axboe * It is expected that bio_check_pages_dirty() will wholly own the BIO from 1713ea1754a0SKirill A. Shutemov * here on. It will run one put_page() against each page and will run one 1714ea1754a0SKirill A. Shutemov * bio_put() against the BIO. 1715f9c78b2bSJens Axboe */ 1716f9c78b2bSJens Axboe 1717f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work); 1718f9c78b2bSJens Axboe 1719f9c78b2bSJens Axboe static DECLARE_WORK(bio_dirty_work, bio_dirty_fn); 1720f9c78b2bSJens Axboe static DEFINE_SPINLOCK(bio_dirty_lock); 1721f9c78b2bSJens Axboe static struct bio *bio_dirty_list; 1722f9c78b2bSJens Axboe 1723f9c78b2bSJens Axboe /* 1724f9c78b2bSJens Axboe * This runs in process context 1725f9c78b2bSJens Axboe */ 1726f9c78b2bSJens Axboe static void bio_dirty_fn(struct work_struct *work) 1727f9c78b2bSJens Axboe { 172824d5493fSChristoph Hellwig struct bio *bio, *next; 1729f9c78b2bSJens Axboe 173024d5493fSChristoph Hellwig spin_lock_irq(&bio_dirty_lock); 173124d5493fSChristoph Hellwig next = bio_dirty_list; 1732f9c78b2bSJens Axboe bio_dirty_list = NULL; 173324d5493fSChristoph Hellwig spin_unlock_irq(&bio_dirty_lock); 1734f9c78b2bSJens Axboe 173524d5493fSChristoph Hellwig while ((bio = next) != NULL) { 173624d5493fSChristoph Hellwig next = bio->bi_private; 1737f9c78b2bSJens Axboe 1738d241a95fSChristoph Hellwig bio_release_pages(bio, true); 1739f9c78b2bSJens Axboe bio_put(bio); 1740f9c78b2bSJens Axboe } 1741f9c78b2bSJens Axboe } 1742f9c78b2bSJens Axboe 1743f9c78b2bSJens Axboe void bio_check_pages_dirty(struct bio *bio) 1744f9c78b2bSJens Axboe { 1745f9c78b2bSJens Axboe struct bio_vec *bvec; 174624d5493fSChristoph Hellwig unsigned long flags; 17476dc4f100SMing Lei struct bvec_iter_all iter_all; 1748f9c78b2bSJens Axboe 17492b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 175024d5493fSChristoph Hellwig if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page)) 175124d5493fSChristoph Hellwig goto defer; 1752f9c78b2bSJens Axboe } 1753f9c78b2bSJens Axboe 1754d241a95fSChristoph Hellwig bio_release_pages(bio, false); 175524d5493fSChristoph Hellwig bio_put(bio); 175624d5493fSChristoph Hellwig return; 175724d5493fSChristoph Hellwig defer: 1758f9c78b2bSJens Axboe spin_lock_irqsave(&bio_dirty_lock, flags); 1759f9c78b2bSJens Axboe bio->bi_private = bio_dirty_list; 1760f9c78b2bSJens Axboe bio_dirty_list = bio; 1761f9c78b2bSJens Axboe spin_unlock_irqrestore(&bio_dirty_lock, flags); 1762f9c78b2bSJens Axboe schedule_work(&bio_dirty_work); 1763f9c78b2bSJens Axboe } 1764f9c78b2bSJens Axboe 17655b18b5a7SMikulas Patocka void update_io_ticks(struct hd_struct *part, unsigned long now) 17665b18b5a7SMikulas Patocka { 17675b18b5a7SMikulas Patocka unsigned long stamp; 17685b18b5a7SMikulas Patocka again: 17695b18b5a7SMikulas Patocka stamp = READ_ONCE(part->stamp); 17705b18b5a7SMikulas Patocka if (unlikely(stamp != now)) { 17715b18b5a7SMikulas Patocka if (likely(cmpxchg(&part->stamp, stamp, now) == stamp)) { 17725b18b5a7SMikulas Patocka __part_stat_add(part, io_ticks, 1); 17735b18b5a7SMikulas Patocka } 17745b18b5a7SMikulas Patocka } 17755b18b5a7SMikulas Patocka if (part->partno) { 17765b18b5a7SMikulas Patocka part = &part_to_disk(part)->part0; 17775b18b5a7SMikulas Patocka goto again; 17785b18b5a7SMikulas Patocka } 17795b18b5a7SMikulas Patocka } 1780f9c78b2bSJens Axboe 1781ddcf35d3SMichael Callahan void generic_start_io_acct(struct request_queue *q, int op, 1782d62e26b3SJens Axboe unsigned long sectors, struct hd_struct *part) 1783394ffa50SGu Zheng { 1784ddcf35d3SMichael Callahan const int sgrp = op_stat_group(op); 1785394ffa50SGu Zheng 1786112f158fSMike Snitzer part_stat_lock(); 1787112f158fSMike Snitzer 17885b18b5a7SMikulas Patocka update_io_ticks(part, jiffies); 1789112f158fSMike Snitzer part_stat_inc(part, ios[sgrp]); 1790112f158fSMike Snitzer part_stat_add(part, sectors[sgrp], sectors); 1791ddcf35d3SMichael Callahan part_inc_in_flight(q, part, op_is_write(op)); 1792394ffa50SGu Zheng 1793394ffa50SGu Zheng part_stat_unlock(); 1794394ffa50SGu Zheng } 1795394ffa50SGu Zheng EXPORT_SYMBOL(generic_start_io_acct); 1796394ffa50SGu Zheng 1797ddcf35d3SMichael Callahan void generic_end_io_acct(struct request_queue *q, int req_op, 1798d62e26b3SJens Axboe struct hd_struct *part, unsigned long start_time) 1799394ffa50SGu Zheng { 18005b18b5a7SMikulas Patocka unsigned long now = jiffies; 18015b18b5a7SMikulas Patocka unsigned long duration = now - start_time; 1802ddcf35d3SMichael Callahan const int sgrp = op_stat_group(req_op); 1803394ffa50SGu Zheng 1804112f158fSMike Snitzer part_stat_lock(); 1805112f158fSMike Snitzer 18065b18b5a7SMikulas Patocka update_io_ticks(part, now); 1807112f158fSMike Snitzer part_stat_add(part, nsecs[sgrp], jiffies_to_nsecs(duration)); 18085b18b5a7SMikulas Patocka part_stat_add(part, time_in_queue, duration); 1809ddcf35d3SMichael Callahan part_dec_in_flight(q, part, op_is_write(req_op)); 1810394ffa50SGu Zheng 1811394ffa50SGu Zheng part_stat_unlock(); 1812394ffa50SGu Zheng } 1813394ffa50SGu Zheng EXPORT_SYMBOL(generic_end_io_acct); 1814394ffa50SGu Zheng 1815c4cf5261SJens Axboe static inline bool bio_remaining_done(struct bio *bio) 1816c4cf5261SJens Axboe { 1817c4cf5261SJens Axboe /* 1818c4cf5261SJens Axboe * If we're not chaining, then ->__bi_remaining is always 1 and 1819c4cf5261SJens Axboe * we always end io on the first invocation. 1820c4cf5261SJens Axboe */ 1821c4cf5261SJens Axboe if (!bio_flagged(bio, BIO_CHAIN)) 1822c4cf5261SJens Axboe return true; 1823c4cf5261SJens Axboe 1824c4cf5261SJens Axboe BUG_ON(atomic_read(&bio->__bi_remaining) <= 0); 1825c4cf5261SJens Axboe 1826326e1dbbSMike Snitzer if (atomic_dec_and_test(&bio->__bi_remaining)) { 1827b7c44ed9SJens Axboe bio_clear_flag(bio, BIO_CHAIN); 1828c4cf5261SJens Axboe return true; 1829326e1dbbSMike Snitzer } 1830c4cf5261SJens Axboe 1831c4cf5261SJens Axboe return false; 1832c4cf5261SJens Axboe } 1833c4cf5261SJens Axboe 1834f9c78b2bSJens Axboe /** 1835f9c78b2bSJens Axboe * bio_endio - end I/O on a bio 1836f9c78b2bSJens Axboe * @bio: bio 1837f9c78b2bSJens Axboe * 1838f9c78b2bSJens Axboe * Description: 18394246a0b6SChristoph Hellwig * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred 18404246a0b6SChristoph Hellwig * way to end I/O on a bio. No one should call bi_end_io() directly on a 18414246a0b6SChristoph Hellwig * bio unless they own it and thus know that it has an end_io function. 1842fbbaf700SNeilBrown * 1843fbbaf700SNeilBrown * bio_endio() can be called several times on a bio that has been chained 1844fbbaf700SNeilBrown * using bio_chain(). The ->bi_end_io() function will only be called the 1845fbbaf700SNeilBrown * last time. At this point the BLK_TA_COMPLETE tracing event will be 1846fbbaf700SNeilBrown * generated if BIO_TRACE_COMPLETION is set. 1847f9c78b2bSJens Axboe **/ 18484246a0b6SChristoph Hellwig void bio_endio(struct bio *bio) 1849f9c78b2bSJens Axboe { 1850ba8c6967SChristoph Hellwig again: 18512b885517SChristoph Hellwig if (!bio_remaining_done(bio)) 1852ba8c6967SChristoph Hellwig return; 18537c20f116SChristoph Hellwig if (!bio_integrity_endio(bio)) 18547c20f116SChristoph Hellwig return; 1855f9c78b2bSJens Axboe 185667b42d0bSJosef Bacik if (bio->bi_disk) 185767b42d0bSJosef Bacik rq_qos_done_bio(bio->bi_disk->queue, bio); 185867b42d0bSJosef Bacik 1859f9c78b2bSJens Axboe /* 1860ba8c6967SChristoph Hellwig * Need to have a real endio function for chained bios, otherwise 1861ba8c6967SChristoph Hellwig * various corner cases will break (like stacking block devices that 1862ba8c6967SChristoph Hellwig * save/restore bi_end_io) - however, we want to avoid unbounded 1863ba8c6967SChristoph Hellwig * recursion and blowing the stack. Tail call optimization would 1864ba8c6967SChristoph Hellwig * handle this, but compiling with frame pointers also disables 1865ba8c6967SChristoph Hellwig * gcc's sibling call optimization. 1866f9c78b2bSJens Axboe */ 1867f9c78b2bSJens Axboe if (bio->bi_end_io == bio_chain_endio) { 186838f8baaeSChristoph Hellwig bio = __bio_chain_endio(bio); 1869ba8c6967SChristoph Hellwig goto again; 1870ba8c6967SChristoph Hellwig } 1871ba8c6967SChristoph Hellwig 187274d46992SChristoph Hellwig if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) { 187374d46992SChristoph Hellwig trace_block_bio_complete(bio->bi_disk->queue, bio, 1874a462b950SBart Van Assche blk_status_to_errno(bio->bi_status)); 1875fbbaf700SNeilBrown bio_clear_flag(bio, BIO_TRACE_COMPLETION); 1876fbbaf700SNeilBrown } 1877fbbaf700SNeilBrown 18789e234eeaSShaohua Li blk_throtl_bio_endio(bio); 1879b222dd2fSShaohua Li /* release cgroup info */ 1880b222dd2fSShaohua Li bio_uninit(bio); 1881f9c78b2bSJens Axboe if (bio->bi_end_io) 18824246a0b6SChristoph Hellwig bio->bi_end_io(bio); 1883f9c78b2bSJens Axboe } 1884f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_endio); 1885f9c78b2bSJens Axboe 1886f9c78b2bSJens Axboe /** 1887f9c78b2bSJens Axboe * bio_split - split a bio 1888f9c78b2bSJens Axboe * @bio: bio to split 1889f9c78b2bSJens Axboe * @sectors: number of sectors to split from the front of @bio 1890f9c78b2bSJens Axboe * @gfp: gfp mask 1891f9c78b2bSJens Axboe * @bs: bio set to allocate from 1892f9c78b2bSJens Axboe * 1893f9c78b2bSJens Axboe * Allocates and returns a new bio which represents @sectors from the start of 1894f9c78b2bSJens Axboe * @bio, and updates @bio to represent the remaining sectors. 1895f9c78b2bSJens Axboe * 1896f3f5da62SMartin K. Petersen * Unless this is a discard request the newly allocated bio will point 1897dad77584SBart Van Assche * to @bio's bi_io_vec. It is the caller's responsibility to ensure that 1898dad77584SBart Van Assche * neither @bio nor @bs are freed before the split bio. 1899f9c78b2bSJens Axboe */ 1900f9c78b2bSJens Axboe struct bio *bio_split(struct bio *bio, int sectors, 1901f9c78b2bSJens Axboe gfp_t gfp, struct bio_set *bs) 1902f9c78b2bSJens Axboe { 1903f341a4d3SMikulas Patocka struct bio *split; 1904f9c78b2bSJens Axboe 1905f9c78b2bSJens Axboe BUG_ON(sectors <= 0); 1906f9c78b2bSJens Axboe BUG_ON(sectors >= bio_sectors(bio)); 1907f9c78b2bSJens Axboe 1908f9c78b2bSJens Axboe split = bio_clone_fast(bio, gfp, bs); 1909f9c78b2bSJens Axboe if (!split) 1910f9c78b2bSJens Axboe return NULL; 1911f9c78b2bSJens Axboe 1912f9c78b2bSJens Axboe split->bi_iter.bi_size = sectors << 9; 1913f9c78b2bSJens Axboe 1914f9c78b2bSJens Axboe if (bio_integrity(split)) 1915fbd08e76SDmitry Monakhov bio_integrity_trim(split); 1916f9c78b2bSJens Axboe 1917f9c78b2bSJens Axboe bio_advance(bio, split->bi_iter.bi_size); 1918f9c78b2bSJens Axboe 1919fbbaf700SNeilBrown if (bio_flagged(bio, BIO_TRACE_COMPLETION)) 192020d59023SGoldwyn Rodrigues bio_set_flag(split, BIO_TRACE_COMPLETION); 1921fbbaf700SNeilBrown 1922f9c78b2bSJens Axboe return split; 1923f9c78b2bSJens Axboe } 1924f9c78b2bSJens Axboe EXPORT_SYMBOL(bio_split); 1925f9c78b2bSJens Axboe 1926f9c78b2bSJens Axboe /** 1927f9c78b2bSJens Axboe * bio_trim - trim a bio 1928f9c78b2bSJens Axboe * @bio: bio to trim 1929f9c78b2bSJens Axboe * @offset: number of sectors to trim from the front of @bio 1930f9c78b2bSJens Axboe * @size: size we want to trim @bio to, in sectors 1931f9c78b2bSJens Axboe */ 1932f9c78b2bSJens Axboe void bio_trim(struct bio *bio, int offset, int size) 1933f9c78b2bSJens Axboe { 1934f9c78b2bSJens Axboe /* 'bio' is a cloned bio which we need to trim to match 1935f9c78b2bSJens Axboe * the given offset and size. 1936f9c78b2bSJens Axboe */ 1937f9c78b2bSJens Axboe 1938f9c78b2bSJens Axboe size <<= 9; 1939f9c78b2bSJens Axboe if (offset == 0 && size == bio->bi_iter.bi_size) 1940f9c78b2bSJens Axboe return; 1941f9c78b2bSJens Axboe 1942f9c78b2bSJens Axboe bio_advance(bio, offset << 9); 1943f9c78b2bSJens Axboe bio->bi_iter.bi_size = size; 1944376a78abSDmitry Monakhov 1945376a78abSDmitry Monakhov if (bio_integrity(bio)) 1946fbd08e76SDmitry Monakhov bio_integrity_trim(bio); 1947376a78abSDmitry Monakhov 1948f9c78b2bSJens Axboe } 1949f9c78b2bSJens Axboe EXPORT_SYMBOL_GPL(bio_trim); 1950f9c78b2bSJens Axboe 1951f9c78b2bSJens Axboe /* 1952f9c78b2bSJens Axboe * create memory pools for biovec's in a bio_set. 1953f9c78b2bSJens Axboe * use the global biovec slabs created for general use. 1954f9c78b2bSJens Axboe */ 19558aa6ba2fSKent Overstreet int biovec_init_pool(mempool_t *pool, int pool_entries) 1956f9c78b2bSJens Axboe { 1957ed996a52SChristoph Hellwig struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX; 1958f9c78b2bSJens Axboe 19598aa6ba2fSKent Overstreet return mempool_init_slab_pool(pool, pool_entries, bp->slab); 1960f9c78b2bSJens Axboe } 1961f9c78b2bSJens Axboe 1962917a38c7SKent Overstreet /* 1963917a38c7SKent Overstreet * bioset_exit - exit a bioset initialized with bioset_init() 1964917a38c7SKent Overstreet * 1965917a38c7SKent Overstreet * May be called on a zeroed but uninitialized bioset (i.e. allocated with 1966917a38c7SKent Overstreet * kzalloc()). 1967917a38c7SKent Overstreet */ 1968917a38c7SKent Overstreet void bioset_exit(struct bio_set *bs) 1969f9c78b2bSJens Axboe { 1970f9c78b2bSJens Axboe if (bs->rescue_workqueue) 1971f9c78b2bSJens Axboe destroy_workqueue(bs->rescue_workqueue); 1972917a38c7SKent Overstreet bs->rescue_workqueue = NULL; 1973f9c78b2bSJens Axboe 19748aa6ba2fSKent Overstreet mempool_exit(&bs->bio_pool); 19758aa6ba2fSKent Overstreet mempool_exit(&bs->bvec_pool); 1976f9c78b2bSJens Axboe 1977f9c78b2bSJens Axboe bioset_integrity_free(bs); 1978917a38c7SKent Overstreet if (bs->bio_slab) 1979f9c78b2bSJens Axboe bio_put_slab(bs); 1980917a38c7SKent Overstreet bs->bio_slab = NULL; 1981917a38c7SKent Overstreet } 1982917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_exit); 1983f9c78b2bSJens Axboe 1984011067b0SNeilBrown /** 1985917a38c7SKent Overstreet * bioset_init - Initialize a bio_set 1986dad08527SKent Overstreet * @bs: pool to initialize 1987917a38c7SKent Overstreet * @pool_size: Number of bio and bio_vecs to cache in the mempool 1988917a38c7SKent Overstreet * @front_pad: Number of bytes to allocate in front of the returned bio 1989917a38c7SKent Overstreet * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS 1990917a38c7SKent Overstreet * and %BIOSET_NEED_RESCUER 1991917a38c7SKent Overstreet * 1992dad08527SKent Overstreet * Description: 1993dad08527SKent Overstreet * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller 1994dad08527SKent Overstreet * to ask for a number of bytes to be allocated in front of the bio. 1995dad08527SKent Overstreet * Front pad allocation is useful for embedding the bio inside 1996dad08527SKent Overstreet * another structure, to avoid allocating extra data to go with the bio. 1997dad08527SKent Overstreet * Note that the bio must be embedded at the END of that structure always, 1998dad08527SKent Overstreet * or things will break badly. 1999dad08527SKent Overstreet * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated 2000dad08527SKent Overstreet * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast(). 2001dad08527SKent Overstreet * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to 2002dad08527SKent Overstreet * dispatch queued requests when the mempool runs out of space. 2003dad08527SKent Overstreet * 2004917a38c7SKent Overstreet */ 2005917a38c7SKent Overstreet int bioset_init(struct bio_set *bs, 2006917a38c7SKent Overstreet unsigned int pool_size, 2007917a38c7SKent Overstreet unsigned int front_pad, 2008917a38c7SKent Overstreet int flags) 2009917a38c7SKent Overstreet { 2010917a38c7SKent Overstreet unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec); 2011917a38c7SKent Overstreet 2012917a38c7SKent Overstreet bs->front_pad = front_pad; 2013917a38c7SKent Overstreet 2014917a38c7SKent Overstreet spin_lock_init(&bs->rescue_lock); 2015917a38c7SKent Overstreet bio_list_init(&bs->rescue_list); 2016917a38c7SKent Overstreet INIT_WORK(&bs->rescue_work, bio_alloc_rescue); 2017917a38c7SKent Overstreet 2018917a38c7SKent Overstreet bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad); 2019917a38c7SKent Overstreet if (!bs->bio_slab) 2020917a38c7SKent Overstreet return -ENOMEM; 2021917a38c7SKent Overstreet 2022917a38c7SKent Overstreet if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab)) 2023917a38c7SKent Overstreet goto bad; 2024917a38c7SKent Overstreet 2025917a38c7SKent Overstreet if ((flags & BIOSET_NEED_BVECS) && 2026917a38c7SKent Overstreet biovec_init_pool(&bs->bvec_pool, pool_size)) 2027917a38c7SKent Overstreet goto bad; 2028917a38c7SKent Overstreet 2029917a38c7SKent Overstreet if (!(flags & BIOSET_NEED_RESCUER)) 2030917a38c7SKent Overstreet return 0; 2031917a38c7SKent Overstreet 2032917a38c7SKent Overstreet bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0); 2033917a38c7SKent Overstreet if (!bs->rescue_workqueue) 2034917a38c7SKent Overstreet goto bad; 2035917a38c7SKent Overstreet 2036917a38c7SKent Overstreet return 0; 2037917a38c7SKent Overstreet bad: 2038917a38c7SKent Overstreet bioset_exit(bs); 2039917a38c7SKent Overstreet return -ENOMEM; 2040917a38c7SKent Overstreet } 2041917a38c7SKent Overstreet EXPORT_SYMBOL(bioset_init); 2042917a38c7SKent Overstreet 204328e89fd9SJens Axboe /* 204428e89fd9SJens Axboe * Initialize and setup a new bio_set, based on the settings from 204528e89fd9SJens Axboe * another bio_set. 204628e89fd9SJens Axboe */ 204728e89fd9SJens Axboe int bioset_init_from_src(struct bio_set *bs, struct bio_set *src) 204828e89fd9SJens Axboe { 204928e89fd9SJens Axboe int flags; 205028e89fd9SJens Axboe 205128e89fd9SJens Axboe flags = 0; 205228e89fd9SJens Axboe if (src->bvec_pool.min_nr) 205328e89fd9SJens Axboe flags |= BIOSET_NEED_BVECS; 205428e89fd9SJens Axboe if (src->rescue_workqueue) 205528e89fd9SJens Axboe flags |= BIOSET_NEED_RESCUER; 205628e89fd9SJens Axboe 205728e89fd9SJens Axboe return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags); 205828e89fd9SJens Axboe } 205928e89fd9SJens Axboe EXPORT_SYMBOL(bioset_init_from_src); 206028e89fd9SJens Axboe 2061f9c78b2bSJens Axboe #ifdef CONFIG_BLK_CGROUP 20621d933cf0STejun Heo 20631d933cf0STejun Heo /** 20642268c0feSDennis Zhou * bio_disassociate_blkg - puts back the blkg reference if associated 2065b5f2954dSDennis Zhou * @bio: target bio 2066b5f2954dSDennis Zhou * 20672268c0feSDennis Zhou * Helper to disassociate the blkg from @bio if a blkg is associated. 2068b5f2954dSDennis Zhou */ 20692268c0feSDennis Zhou void bio_disassociate_blkg(struct bio *bio) 2070b5f2954dSDennis Zhou { 207108e18eabSJosef Bacik if (bio->bi_blkg) { 207208e18eabSJosef Bacik blkg_put(bio->bi_blkg); 207308e18eabSJosef Bacik bio->bi_blkg = NULL; 207408e18eabSJosef Bacik } 2075f9c78b2bSJens Axboe } 2076892ad71fSDennis Zhou EXPORT_SYMBOL_GPL(bio_disassociate_blkg); 2077f9c78b2bSJens Axboe 207820bd723eSPaolo Valente /** 20792268c0feSDennis Zhou * __bio_associate_blkg - associate a bio with the a blkg 2080f9c78b2bSJens Axboe * @bio: target bio 2081f9c78b2bSJens Axboe * @blkg: the blkg to associate 2082f9c78b2bSJens Axboe * 2083beea9da0SDennis Zhou * This tries to associate @bio with the specified @blkg. Association failure 2084beea9da0SDennis Zhou * is handled by walking up the blkg tree. Therefore, the blkg associated can 2085beea9da0SDennis Zhou * be anything between @blkg and the root_blkg. This situation only happens 2086beea9da0SDennis Zhou * when a cgroup is dying and then the remaining bios will spill to the closest 2087beea9da0SDennis Zhou * alive blkg. 2088beea9da0SDennis Zhou * 2089beea9da0SDennis Zhou * A reference will be taken on the @blkg and will be released when @bio is 2090beea9da0SDennis Zhou * freed. 2091f9c78b2bSJens Axboe */ 20922268c0feSDennis Zhou static void __bio_associate_blkg(struct bio *bio, struct blkcg_gq *blkg) 2093f9c78b2bSJens Axboe { 20942268c0feSDennis Zhou bio_disassociate_blkg(bio); 20952268c0feSDennis Zhou 20967754f669SDennis Zhou bio->bi_blkg = blkg_tryget_closest(blkg); 20972268c0feSDennis Zhou } 20982268c0feSDennis Zhou 2099fd42df30SDennis Zhou /** 2100fd42df30SDennis Zhou * bio_associate_blkg_from_css - associate a bio with a specified css 2101fd42df30SDennis Zhou * @bio: target bio 2102fd42df30SDennis Zhou * @css: target css 2103fd42df30SDennis Zhou * 2104fd42df30SDennis Zhou * Associate @bio with the blkg found by combining the css's blkg and the 2105fc5a828bSDennis Zhou * request_queue of the @bio. This falls back to the queue's root_blkg if 2106fc5a828bSDennis Zhou * the association fails with the css. 2107fd42df30SDennis Zhou */ 2108fd42df30SDennis Zhou void bio_associate_blkg_from_css(struct bio *bio, 2109fd42df30SDennis Zhou struct cgroup_subsys_state *css) 2110fd42df30SDennis Zhou { 2111fc5a828bSDennis Zhou struct request_queue *q = bio->bi_disk->queue; 2112fc5a828bSDennis Zhou struct blkcg_gq *blkg; 2113fc5a828bSDennis Zhou 2114fc5a828bSDennis Zhou rcu_read_lock(); 2115fc5a828bSDennis Zhou 2116fc5a828bSDennis Zhou if (!css || !css->parent) 2117fc5a828bSDennis Zhou blkg = q->root_blkg; 2118fc5a828bSDennis Zhou else 2119fc5a828bSDennis Zhou blkg = blkg_lookup_create(css_to_blkcg(css), q); 2120fc5a828bSDennis Zhou 2121fc5a828bSDennis Zhou __bio_associate_blkg(bio, blkg); 2122fc5a828bSDennis Zhou 2123fc5a828bSDennis Zhou rcu_read_unlock(); 2124fd42df30SDennis Zhou } 2125fd42df30SDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css); 2126fd42df30SDennis Zhou 21276a7f6d86SDennis Zhou #ifdef CONFIG_MEMCG 21286a7f6d86SDennis Zhou /** 21296a7f6d86SDennis Zhou * bio_associate_blkg_from_page - associate a bio with the page's blkg 21306a7f6d86SDennis Zhou * @bio: target bio 21316a7f6d86SDennis Zhou * @page: the page to lookup the blkcg from 21326a7f6d86SDennis Zhou * 21336a7f6d86SDennis Zhou * Associate @bio with the blkg from @page's owning memcg and the respective 2134fc5a828bSDennis Zhou * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's 2135fc5a828bSDennis Zhou * root_blkg. 21366a7f6d86SDennis Zhou */ 21376a7f6d86SDennis Zhou void bio_associate_blkg_from_page(struct bio *bio, struct page *page) 21386a7f6d86SDennis Zhou { 21396a7f6d86SDennis Zhou struct cgroup_subsys_state *css; 21406a7f6d86SDennis Zhou 21416a7f6d86SDennis Zhou if (!page->mem_cgroup) 21426a7f6d86SDennis Zhou return; 21436a7f6d86SDennis Zhou 2144fc5a828bSDennis Zhou rcu_read_lock(); 2145fc5a828bSDennis Zhou 2146fc5a828bSDennis Zhou css = cgroup_e_css(page->mem_cgroup->css.cgroup, &io_cgrp_subsys); 2147fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 2148fc5a828bSDennis Zhou 2149fc5a828bSDennis Zhou rcu_read_unlock(); 21506a7f6d86SDennis Zhou } 21516a7f6d86SDennis Zhou #endif /* CONFIG_MEMCG */ 21526a7f6d86SDennis Zhou 21532268c0feSDennis Zhou /** 21542268c0feSDennis Zhou * bio_associate_blkg - associate a bio with a blkg 21552268c0feSDennis Zhou * @bio: target bio 21562268c0feSDennis Zhou * 21572268c0feSDennis Zhou * Associate @bio with the blkg found from the bio's css and request_queue. 21582268c0feSDennis Zhou * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is 21592268c0feSDennis Zhou * already associated, the css is reused and association redone as the 21602268c0feSDennis Zhou * request_queue may have changed. 21612268c0feSDennis Zhou */ 21622268c0feSDennis Zhou void bio_associate_blkg(struct bio *bio) 21632268c0feSDennis Zhou { 2164fc5a828bSDennis Zhou struct cgroup_subsys_state *css; 21652268c0feSDennis Zhou 21662268c0feSDennis Zhou rcu_read_lock(); 21672268c0feSDennis Zhou 2168db6638d7SDennis Zhou if (bio->bi_blkg) 2169fc5a828bSDennis Zhou css = &bio_blkcg(bio)->css; 2170db6638d7SDennis Zhou else 2171fc5a828bSDennis Zhou css = blkcg_css(); 21722268c0feSDennis Zhou 2173fc5a828bSDennis Zhou bio_associate_blkg_from_css(bio, css); 21742268c0feSDennis Zhou 21752268c0feSDennis Zhou rcu_read_unlock(); 2176f9c78b2bSJens Axboe } 21775cdf2e3fSDennis Zhou EXPORT_SYMBOL_GPL(bio_associate_blkg); 2178f9c78b2bSJens Axboe 217920bd723eSPaolo Valente /** 2180db6638d7SDennis Zhou * bio_clone_blkg_association - clone blkg association from src to dst bio 218120bd723eSPaolo Valente * @dst: destination bio 218220bd723eSPaolo Valente * @src: source bio 218320bd723eSPaolo Valente */ 2184db6638d7SDennis Zhou void bio_clone_blkg_association(struct bio *dst, struct bio *src) 218520bd723eSPaolo Valente { 21866ab21879SDennis Zhou rcu_read_lock(); 21876ab21879SDennis Zhou 2188fc5a828bSDennis Zhou if (src->bi_blkg) 21892268c0feSDennis Zhou __bio_associate_blkg(dst, src->bi_blkg); 21906ab21879SDennis Zhou 21916ab21879SDennis Zhou rcu_read_unlock(); 219220bd723eSPaolo Valente } 2193db6638d7SDennis Zhou EXPORT_SYMBOL_GPL(bio_clone_blkg_association); 2194f9c78b2bSJens Axboe #endif /* CONFIG_BLK_CGROUP */ 2195f9c78b2bSJens Axboe 2196f9c78b2bSJens Axboe static void __init biovec_init_slabs(void) 2197f9c78b2bSJens Axboe { 2198f9c78b2bSJens Axboe int i; 2199f9c78b2bSJens Axboe 2200ed996a52SChristoph Hellwig for (i = 0; i < BVEC_POOL_NR; i++) { 2201f9c78b2bSJens Axboe int size; 2202f9c78b2bSJens Axboe struct biovec_slab *bvs = bvec_slabs + i; 2203f9c78b2bSJens Axboe 2204f9c78b2bSJens Axboe if (bvs->nr_vecs <= BIO_INLINE_VECS) { 2205f9c78b2bSJens Axboe bvs->slab = NULL; 2206f9c78b2bSJens Axboe continue; 2207f9c78b2bSJens Axboe } 2208f9c78b2bSJens Axboe 2209f9c78b2bSJens Axboe size = bvs->nr_vecs * sizeof(struct bio_vec); 2210f9c78b2bSJens Axboe bvs->slab = kmem_cache_create(bvs->name, size, 0, 2211f9c78b2bSJens Axboe SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 2212f9c78b2bSJens Axboe } 2213f9c78b2bSJens Axboe } 2214f9c78b2bSJens Axboe 2215f9c78b2bSJens Axboe static int __init init_bio(void) 2216f9c78b2bSJens Axboe { 2217f9c78b2bSJens Axboe bio_slab_max = 2; 2218f9c78b2bSJens Axboe bio_slab_nr = 0; 22196396bb22SKees Cook bio_slabs = kcalloc(bio_slab_max, sizeof(struct bio_slab), 22206396bb22SKees Cook GFP_KERNEL); 22212b24e6f6SJohannes Thumshirn 22222b24e6f6SJohannes Thumshirn BUILD_BUG_ON(BIO_FLAG_LAST > BVEC_POOL_OFFSET); 22232b24e6f6SJohannes Thumshirn 2224f9c78b2bSJens Axboe if (!bio_slabs) 2225f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2226f9c78b2bSJens Axboe 2227f9c78b2bSJens Axboe bio_integrity_init(); 2228f9c78b2bSJens Axboe biovec_init_slabs(); 2229f9c78b2bSJens Axboe 2230f4f8154aSKent Overstreet if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS)) 2231f9c78b2bSJens Axboe panic("bio: can't allocate bios\n"); 2232f9c78b2bSJens Axboe 2233f4f8154aSKent Overstreet if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE)) 2234f9c78b2bSJens Axboe panic("bio: can't create integrity pool\n"); 2235f9c78b2bSJens Axboe 2236f9c78b2bSJens Axboe return 0; 2237f9c78b2bSJens Axboe } 2238f9c78b2bSJens Axboe subsys_initcall(init_bio); 2239