1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 2719c555fSJens Axboe /* bounce buffer handling for block devices 3719c555fSJens Axboe * 4719c555fSJens Axboe * - Split from highmem.c 5719c555fSJens Axboe */ 6719c555fSJens Axboe 7b1de0d13SMitchel Humpherys #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 8b1de0d13SMitchel Humpherys 9719c555fSJens Axboe #include <linux/mm.h> 10719c555fSJens Axboe #include <linux/export.h> 11719c555fSJens Axboe #include <linux/swap.h> 12719c555fSJens Axboe #include <linux/gfp.h> 13719c555fSJens Axboe #include <linux/bio.h> 14719c555fSJens Axboe #include <linux/pagemap.h> 15719c555fSJens Axboe #include <linux/mempool.h> 16719c555fSJens Axboe #include <linux/blkdev.h> 1766114cadSTejun Heo #include <linux/backing-dev.h> 18719c555fSJens Axboe #include <linux/init.h> 19719c555fSJens Axboe #include <linux/hash.h> 20719c555fSJens Axboe #include <linux/highmem.h> 21b1de0d13SMitchel Humpherys #include <linux/printk.h> 22719c555fSJens Axboe #include <asm/tlbflush.h> 23719c555fSJens Axboe 24719c555fSJens Axboe #include <trace/events/block.h> 253bce016aSChristoph Hellwig #include "blk.h" 26719c555fSJens Axboe 27719c555fSJens Axboe #define POOL_SIZE 64 28719c555fSJens Axboe #define ISA_POOL_SIZE 16 29719c555fSJens Axboe 30338aa96dSKent Overstreet static struct bio_set bounce_bio_set, bounce_bio_split; 31ce288e05SChristoph Hellwig static mempool_t page_pool; 32719c555fSJens Axboe 3352990a5fSJens Axboe static void init_bounce_bioset(void) 3452990a5fSJens Axboe { 3552990a5fSJens Axboe static bool bounce_bs_setup; 3652990a5fSJens Axboe int ret; 3752990a5fSJens Axboe 3852990a5fSJens Axboe if (bounce_bs_setup) 3952990a5fSJens Axboe return; 4052990a5fSJens Axboe 4152990a5fSJens Axboe ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); 4252990a5fSJens Axboe BUG_ON(ret); 4352990a5fSJens Axboe if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE)) 4452990a5fSJens Axboe BUG_ON(1); 4552990a5fSJens Axboe 4652990a5fSJens Axboe ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0); 4752990a5fSJens Axboe BUG_ON(ret); 4852990a5fSJens Axboe bounce_bs_setup = true; 4952990a5fSJens Axboe } 5052990a5fSJens Axboe 51719c555fSJens Axboe static __init int init_emergency_pool(void) 52719c555fSJens Axboe { 53338aa96dSKent Overstreet int ret; 54*9bb33f24SChristoph Hellwig 55*9bb33f24SChristoph Hellwig #ifndef CONFIG_MEMORY_HOTPLUG 56719c555fSJens Axboe if (max_pfn <= max_low_pfn) 57719c555fSJens Axboe return 0; 58719c555fSJens Axboe #endif 59719c555fSJens Axboe 60338aa96dSKent Overstreet ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0); 61338aa96dSKent Overstreet BUG_ON(ret); 62b1de0d13SMitchel Humpherys pr_info("pool size: %d pages\n", POOL_SIZE); 63719c555fSJens Axboe 6452990a5fSJens Axboe init_bounce_bioset(); 65719c555fSJens Axboe return 0; 66719c555fSJens Axboe } 67719c555fSJens Axboe 68719c555fSJens Axboe __initcall(init_emergency_pool); 69719c555fSJens Axboe 70719c555fSJens Axboe /* 71719c555fSJens Axboe * highmem version, map in to vec 72719c555fSJens Axboe */ 73719c555fSJens Axboe static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom) 74719c555fSJens Axboe { 75719c555fSJens Axboe unsigned char *vto; 76719c555fSJens Axboe 77719c555fSJens Axboe vto = kmap_atomic(to->bv_page); 78719c555fSJens Axboe memcpy(vto + to->bv_offset, vfrom, to->bv_len); 79719c555fSJens Axboe kunmap_atomic(vto); 80719c555fSJens Axboe } 81719c555fSJens Axboe 82719c555fSJens Axboe /* 83719c555fSJens Axboe * Simple bounce buffer support for highmem pages. Depending on the 84719c555fSJens Axboe * queue gfp mask set, *to may or may not be a highmem page. kmap it 85719c555fSJens Axboe * always, it will do the Right Thing 86719c555fSJens Axboe */ 87719c555fSJens Axboe static void copy_to_high_bio_irq(struct bio *to, struct bio *from) 88719c555fSJens Axboe { 89719c555fSJens Axboe unsigned char *vfrom; 903c892a09SMing Lei struct bio_vec tovec, fromvec; 91719c555fSJens Axboe struct bvec_iter iter; 923c892a09SMing Lei /* 933c892a09SMing Lei * The bio of @from is created by bounce, so we can iterate 943c892a09SMing Lei * its bvec from start to end, but the @from->bi_iter can't be 953c892a09SMing Lei * trusted because it might be changed by splitting. 963c892a09SMing Lei */ 973c892a09SMing Lei struct bvec_iter from_iter = BVEC_ITER_ALL_INIT; 98719c555fSJens Axboe 99719c555fSJens Axboe bio_for_each_segment(tovec, to, iter) { 1003c892a09SMing Lei fromvec = bio_iter_iovec(from, from_iter); 1013c892a09SMing Lei if (tovec.bv_page != fromvec.bv_page) { 102719c555fSJens Axboe /* 103719c555fSJens Axboe * fromvec->bv_offset and fromvec->bv_len might have 104719c555fSJens Axboe * been modified by the block layer, so use the original 105719c555fSJens Axboe * copy, bounce_copy_vec already uses tovec->bv_len 106719c555fSJens Axboe */ 1073c892a09SMing Lei vfrom = page_address(fromvec.bv_page) + 108719c555fSJens Axboe tovec.bv_offset; 109719c555fSJens Axboe 110719c555fSJens Axboe bounce_copy_vec(&tovec, vfrom); 111719c555fSJens Axboe flush_dcache_page(tovec.bv_page); 112719c555fSJens Axboe } 1133c892a09SMing Lei bio_advance_iter(from, &from_iter, tovec.bv_len); 114719c555fSJens Axboe } 115719c555fSJens Axboe } 116719c555fSJens Axboe 117ce288e05SChristoph Hellwig static void bounce_end_io(struct bio *bio) 118719c555fSJens Axboe { 119719c555fSJens Axboe struct bio *bio_orig = bio->bi_private; 1207891f05cSMing Lei struct bio_vec *bvec, orig_vec; 1217891f05cSMing Lei struct bvec_iter orig_iter = bio_orig->bi_iter; 1226dc4f100SMing Lei struct bvec_iter_all iter_all; 123719c555fSJens Axboe 124719c555fSJens Axboe /* 125719c555fSJens Axboe * free up bounce indirect pages used 126719c555fSJens Axboe */ 1272b070cfeSChristoph Hellwig bio_for_each_segment_all(bvec, bio, iter_all) { 1287891f05cSMing Lei orig_vec = bio_iter_iovec(bio_orig, orig_iter); 1297891f05cSMing Lei if (bvec->bv_page != orig_vec.bv_page) { 130719c555fSJens Axboe dec_zone_page_state(bvec->bv_page, NR_BOUNCE); 131ce288e05SChristoph Hellwig mempool_free(bvec->bv_page, &page_pool); 132719c555fSJens Axboe } 1337891f05cSMing Lei bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len); 1347891f05cSMing Lei } 135719c555fSJens Axboe 1364e4cbee9SChristoph Hellwig bio_orig->bi_status = bio->bi_status; 1374246a0b6SChristoph Hellwig bio_endio(bio_orig); 138719c555fSJens Axboe bio_put(bio); 139719c555fSJens Axboe } 140719c555fSJens Axboe 1414246a0b6SChristoph Hellwig static void bounce_end_io_write(struct bio *bio) 142719c555fSJens Axboe { 143ce288e05SChristoph Hellwig bounce_end_io(bio); 144719c555fSJens Axboe } 145719c555fSJens Axboe 146ce288e05SChristoph Hellwig static void bounce_end_io_read(struct bio *bio) 147719c555fSJens Axboe { 148719c555fSJens Axboe struct bio *bio_orig = bio->bi_private; 149719c555fSJens Axboe 1504e4cbee9SChristoph Hellwig if (!bio->bi_status) 151719c555fSJens Axboe copy_to_high_bio_irq(bio_orig, bio); 152719c555fSJens Axboe 153ce288e05SChristoph Hellwig bounce_end_io(bio); 154719c555fSJens Axboe } 155719c555fSJens Axboe 156ebfe4183SChristoph Hellwig static struct bio *bounce_clone_bio(struct bio *bio_src) 157c55183c9SChristoph Hellwig { 158c55183c9SChristoph Hellwig struct bvec_iter iter; 159c55183c9SChristoph Hellwig struct bio_vec bv; 160c55183c9SChristoph Hellwig struct bio *bio; 161c55183c9SChristoph Hellwig 162c55183c9SChristoph Hellwig /* 163c55183c9SChristoph Hellwig * Pre immutable biovecs, __bio_clone() used to just do a memcpy from 164c55183c9SChristoph Hellwig * bio_src->bi_io_vec to bio->bi_io_vec. 165c55183c9SChristoph Hellwig * 166c55183c9SChristoph Hellwig * We can't do that anymore, because: 167c55183c9SChristoph Hellwig * 168c55183c9SChristoph Hellwig * - The point of cloning the biovec is to produce a bio with a biovec 169c55183c9SChristoph Hellwig * the caller can modify: bi_idx and bi_bvec_done should be 0. 170c55183c9SChristoph Hellwig * 171a8affc03SChristoph Hellwig * - The original bio could've had more than BIO_MAX_VECS biovecs; if 172c55183c9SChristoph Hellwig * we tried to clone the whole thing bio_alloc_bioset() would fail. 173c55183c9SChristoph Hellwig * But the clone should succeed as long as the number of biovecs we 174a8affc03SChristoph Hellwig * actually need to allocate is fewer than BIO_MAX_VECS. 175c55183c9SChristoph Hellwig * 176c55183c9SChristoph Hellwig * - Lastly, bi_vcnt should not be looked at or relied upon by code 177c55183c9SChristoph Hellwig * that does not own the bio - reason being drivers don't use it for 178c55183c9SChristoph Hellwig * iterating over the biovec anymore, so expecting it to be kept up 179c55183c9SChristoph Hellwig * to date (i.e. for clones that share the parent biovec) is just 180c55183c9SChristoph Hellwig * asking for trouble and would force extra work on 181c55183c9SChristoph Hellwig * __bio_clone_fast() anyways. 182c55183c9SChristoph Hellwig */ 183b90994c6SChristoph Hellwig if (bio_is_passthrough(bio_src)) 18447dc096aSChristoph Hellwig bio = bio_kmalloc(GFP_NOIO | __GFP_NOFAIL, 18547dc096aSChristoph Hellwig bio_segments(bio_src)); 186b90994c6SChristoph Hellwig else 187ebfe4183SChristoph Hellwig bio = bio_alloc_bioset(GFP_NOIO, bio_segments(bio_src), 188b90994c6SChristoph Hellwig &bounce_bio_set); 189309dca30SChristoph Hellwig bio->bi_bdev = bio_src->bi_bdev; 19046bbf653SChristoph Hellwig if (bio_flagged(bio_src, BIO_REMAPPED)) 19146bbf653SChristoph Hellwig bio_set_flag(bio, BIO_REMAPPED); 192c55183c9SChristoph Hellwig bio->bi_opf = bio_src->bi_opf; 193ca474b73SHannes Reinecke bio->bi_ioprio = bio_src->bi_ioprio; 194c55183c9SChristoph Hellwig bio->bi_write_hint = bio_src->bi_write_hint; 195c55183c9SChristoph Hellwig bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; 196c55183c9SChristoph Hellwig bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; 197c55183c9SChristoph Hellwig 198c55183c9SChristoph Hellwig switch (bio_op(bio)) { 199c55183c9SChristoph Hellwig case REQ_OP_DISCARD: 200c55183c9SChristoph Hellwig case REQ_OP_SECURE_ERASE: 201c55183c9SChristoph Hellwig case REQ_OP_WRITE_ZEROES: 202c55183c9SChristoph Hellwig break; 203c55183c9SChristoph Hellwig case REQ_OP_WRITE_SAME: 204c55183c9SChristoph Hellwig bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; 205c55183c9SChristoph Hellwig break; 206c55183c9SChristoph Hellwig default: 207c55183c9SChristoph Hellwig bio_for_each_segment(bv, bio_src, iter) 208c55183c9SChristoph Hellwig bio->bi_io_vec[bio->bi_vcnt++] = bv; 209c55183c9SChristoph Hellwig break; 210c55183c9SChristoph Hellwig } 211c55183c9SChristoph Hellwig 212ebfe4183SChristoph Hellwig if (bio_crypt_clone(bio, bio_src, GFP_NOIO) < 0) 21307560151SEric Biggers goto err_put; 214a892c8d5SSatya Tangirala 21507560151SEric Biggers if (bio_integrity(bio_src) && 216ebfe4183SChristoph Hellwig bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0) 21707560151SEric Biggers goto err_put; 218c55183c9SChristoph Hellwig 219db6638d7SDennis Zhou bio_clone_blkg_association(bio, bio_src); 220e439bedfSDennis Zhou blkcg_bio_issue_init(bio); 2215bf9a1f3SDennis Zhou (Facebook) 222c55183c9SChristoph Hellwig return bio; 22307560151SEric Biggers 22407560151SEric Biggers err_put: 22507560151SEric Biggers bio_put(bio); 22607560151SEric Biggers return NULL; 227c55183c9SChristoph Hellwig } 228c55183c9SChristoph Hellwig 229*9bb33f24SChristoph Hellwig void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig) 230719c555fSJens Axboe { 231719c555fSJens Axboe struct bio *bio; 232719c555fSJens Axboe int rw = bio_data_dir(*bio_orig); 233719c555fSJens Axboe struct bio_vec *to, from; 234719c555fSJens Axboe struct bvec_iter iter; 235a8821f3fSNeilBrown unsigned i = 0; 236a8821f3fSNeilBrown bool bounce = false; 237a8821f3fSNeilBrown int sectors = 0; 238719c555fSJens Axboe 239a8821f3fSNeilBrown bio_for_each_segment(from, *bio_orig, iter) { 240a8affc03SChristoph Hellwig if (i++ < BIO_MAX_VECS) 241a8821f3fSNeilBrown sectors += from.bv_len >> 9; 242*9bb33f24SChristoph Hellwig if (PageHighMem(from.bv_page)) 243a8821f3fSNeilBrown bounce = true; 244a8821f3fSNeilBrown } 245a8821f3fSNeilBrown if (!bounce) 246719c555fSJens Axboe return; 247a8821f3fSNeilBrown 248b90994c6SChristoph Hellwig if (!bio_is_passthrough(*bio_orig) && 249b90994c6SChristoph Hellwig sectors < bio_sectors(*bio_orig)) { 250338aa96dSKent Overstreet bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split); 251a8821f3fSNeilBrown bio_chain(bio, *bio_orig); 252ed00aabdSChristoph Hellwig submit_bio_noacct(*bio_orig); 253a8821f3fSNeilBrown *bio_orig = bio; 254a8821f3fSNeilBrown } 255ebfe4183SChristoph Hellwig bio = bounce_clone_bio(*bio_orig); 256719c555fSJens Axboe 2578f4e80daSMing Lei /* 2588f4e80daSMing Lei * Bvec table can't be updated by bio_for_each_segment_all(), 2598f4e80daSMing Lei * so retrieve bvec from the table directly. This way is safe 2608f4e80daSMing Lei * because the 'bio' is single-page bvec. 2618f4e80daSMing Lei */ 2628f4e80daSMing Lei for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) { 263719c555fSJens Axboe struct page *page = to->bv_page; 264719c555fSJens Axboe 265*9bb33f24SChristoph Hellwig if (!PageHighMem(page)) 266719c555fSJens Axboe continue; 267719c555fSJens Axboe 268ce288e05SChristoph Hellwig to->bv_page = mempool_alloc(&page_pool, GFP_NOIO); 269393a3397SWang YanQing inc_zone_page_state(to->bv_page, NR_BOUNCE); 270719c555fSJens Axboe 271719c555fSJens Axboe if (rw == WRITE) { 272719c555fSJens Axboe char *vto, *vfrom; 273719c555fSJens Axboe 274719c555fSJens Axboe flush_dcache_page(page); 275719c555fSJens Axboe 276719c555fSJens Axboe vto = page_address(to->bv_page) + to->bv_offset; 277719c555fSJens Axboe vfrom = kmap_atomic(page) + to->bv_offset; 278719c555fSJens Axboe memcpy(vto, vfrom, to->bv_len); 279719c555fSJens Axboe kunmap_atomic(vfrom); 280719c555fSJens Axboe } 281719c555fSJens Axboe } 282719c555fSJens Axboe 283e8a676d6SChristoph Hellwig trace_block_bio_bounce(*bio_orig); 284719c555fSJens Axboe 285719c555fSJens Axboe bio->bi_flags |= (1 << BIO_BOUNCED); 286719c555fSJens Axboe 287719c555fSJens Axboe if (rw == READ) 288719c555fSJens Axboe bio->bi_end_io = bounce_end_io_read; 289ce288e05SChristoph Hellwig else 290ce288e05SChristoph Hellwig bio->bi_end_io = bounce_end_io_write; 291719c555fSJens Axboe 292719c555fSJens Axboe bio->bi_private = *bio_orig; 293719c555fSJens Axboe *bio_orig = bio; 294719c555fSJens Axboe } 295