1 // SPDX-License-Identifier: GPL-2.0 2 /* bounce buffer handling for block devices 3 * 4 * - Split from highmem.c 5 */ 6 7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 8 9 #include <linux/mm.h> 10 #include <linux/export.h> 11 #include <linux/swap.h> 12 #include <linux/gfp.h> 13 #include <linux/bio.h> 14 #include <linux/pagemap.h> 15 #include <linux/mempool.h> 16 #include <linux/blkdev.h> 17 #include <linux/backing-dev.h> 18 #include <linux/init.h> 19 #include <linux/hash.h> 20 #include <linux/highmem.h> 21 #include <linux/printk.h> 22 #include <asm/tlbflush.h> 23 24 #include <trace/events/block.h> 25 #include "blk.h" 26 27 #define POOL_SIZE 64 28 #define ISA_POOL_SIZE 16 29 30 static struct bio_set bounce_bio_set, bounce_bio_split; 31 static mempool_t page_pool; 32 33 static void init_bounce_bioset(void) 34 { 35 static bool bounce_bs_setup; 36 int ret; 37 38 if (bounce_bs_setup) 39 return; 40 41 ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS); 42 BUG_ON(ret); 43 if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE)) 44 BUG_ON(1); 45 46 ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0); 47 BUG_ON(ret); 48 bounce_bs_setup = true; 49 } 50 51 static __init int init_emergency_pool(void) 52 { 53 int ret; 54 55 #ifndef CONFIG_MEMORY_HOTPLUG 56 if (max_pfn <= max_low_pfn) 57 return 0; 58 #endif 59 60 ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0); 61 BUG_ON(ret); 62 pr_info("pool size: %d pages\n", POOL_SIZE); 63 64 init_bounce_bioset(); 65 return 0; 66 } 67 68 __initcall(init_emergency_pool); 69 70 /* 71 * highmem version, map in to vec 72 */ 73 static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom) 74 { 75 unsigned char *vto; 76 77 vto = kmap_atomic(to->bv_page); 78 memcpy(vto + to->bv_offset, vfrom, to->bv_len); 79 kunmap_atomic(vto); 80 } 81 82 /* 83 * Simple bounce buffer support for highmem pages. Depending on the 84 * queue gfp mask set, *to may or may not be a highmem page. kmap it 85 * always, it will do the Right Thing 86 */ 87 static void copy_to_high_bio_irq(struct bio *to, struct bio *from) 88 { 89 unsigned char *vfrom; 90 struct bio_vec tovec, fromvec; 91 struct bvec_iter iter; 92 /* 93 * The bio of @from is created by bounce, so we can iterate 94 * its bvec from start to end, but the @from->bi_iter can't be 95 * trusted because it might be changed by splitting. 96 */ 97 struct bvec_iter from_iter = BVEC_ITER_ALL_INIT; 98 99 bio_for_each_segment(tovec, to, iter) { 100 fromvec = bio_iter_iovec(from, from_iter); 101 if (tovec.bv_page != fromvec.bv_page) { 102 /* 103 * fromvec->bv_offset and fromvec->bv_len might have 104 * been modified by the block layer, so use the original 105 * copy, bounce_copy_vec already uses tovec->bv_len 106 */ 107 vfrom = page_address(fromvec.bv_page) + 108 tovec.bv_offset; 109 110 bounce_copy_vec(&tovec, vfrom); 111 flush_dcache_page(tovec.bv_page); 112 } 113 bio_advance_iter(from, &from_iter, tovec.bv_len); 114 } 115 } 116 117 static void bounce_end_io(struct bio *bio) 118 { 119 struct bio *bio_orig = bio->bi_private; 120 struct bio_vec *bvec, orig_vec; 121 struct bvec_iter orig_iter = bio_orig->bi_iter; 122 struct bvec_iter_all iter_all; 123 124 /* 125 * free up bounce indirect pages used 126 */ 127 bio_for_each_segment_all(bvec, bio, iter_all) { 128 orig_vec = bio_iter_iovec(bio_orig, orig_iter); 129 if (bvec->bv_page != orig_vec.bv_page) { 130 dec_zone_page_state(bvec->bv_page, NR_BOUNCE); 131 mempool_free(bvec->bv_page, &page_pool); 132 } 133 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len); 134 } 135 136 bio_orig->bi_status = bio->bi_status; 137 bio_endio(bio_orig); 138 bio_put(bio); 139 } 140 141 static void bounce_end_io_write(struct bio *bio) 142 { 143 bounce_end_io(bio); 144 } 145 146 static void bounce_end_io_read(struct bio *bio) 147 { 148 struct bio *bio_orig = bio->bi_private; 149 150 if (!bio->bi_status) 151 copy_to_high_bio_irq(bio_orig, bio); 152 153 bounce_end_io(bio); 154 } 155 156 static struct bio *bounce_clone_bio(struct bio *bio_src) 157 { 158 struct bvec_iter iter; 159 struct bio_vec bv; 160 struct bio *bio; 161 162 /* 163 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from 164 * bio_src->bi_io_vec to bio->bi_io_vec. 165 * 166 * We can't do that anymore, because: 167 * 168 * - The point of cloning the biovec is to produce a bio with a biovec 169 * the caller can modify: bi_idx and bi_bvec_done should be 0. 170 * 171 * - The original bio could've had more than BIO_MAX_VECS biovecs; if 172 * we tried to clone the whole thing bio_alloc_bioset() would fail. 173 * But the clone should succeed as long as the number of biovecs we 174 * actually need to allocate is fewer than BIO_MAX_VECS. 175 * 176 * - Lastly, bi_vcnt should not be looked at or relied upon by code 177 * that does not own the bio - reason being drivers don't use it for 178 * iterating over the biovec anymore, so expecting it to be kept up 179 * to date (i.e. for clones that share the parent biovec) is just 180 * asking for trouble and would force extra work on 181 * __bio_clone_fast() anyways. 182 */ 183 if (bio_is_passthrough(bio_src)) 184 bio = bio_kmalloc(GFP_NOIO | __GFP_NOFAIL, 185 bio_segments(bio_src)); 186 else 187 bio = bio_alloc_bioset(GFP_NOIO, bio_segments(bio_src), 188 &bounce_bio_set); 189 bio->bi_bdev = bio_src->bi_bdev; 190 if (bio_flagged(bio_src, BIO_REMAPPED)) 191 bio_set_flag(bio, BIO_REMAPPED); 192 bio->bi_opf = bio_src->bi_opf; 193 bio->bi_ioprio = bio_src->bi_ioprio; 194 bio->bi_write_hint = bio_src->bi_write_hint; 195 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; 196 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size; 197 198 switch (bio_op(bio)) { 199 case REQ_OP_DISCARD: 200 case REQ_OP_SECURE_ERASE: 201 case REQ_OP_WRITE_ZEROES: 202 break; 203 case REQ_OP_WRITE_SAME: 204 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0]; 205 break; 206 default: 207 bio_for_each_segment(bv, bio_src, iter) 208 bio->bi_io_vec[bio->bi_vcnt++] = bv; 209 break; 210 } 211 212 if (bio_crypt_clone(bio, bio_src, GFP_NOIO) < 0) 213 goto err_put; 214 215 if (bio_integrity(bio_src) && 216 bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0) 217 goto err_put; 218 219 bio_clone_blkg_association(bio, bio_src); 220 blkcg_bio_issue_init(bio); 221 222 return bio; 223 224 err_put: 225 bio_put(bio); 226 return NULL; 227 } 228 229 void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig) 230 { 231 struct bio *bio; 232 int rw = bio_data_dir(*bio_orig); 233 struct bio_vec *to, from; 234 struct bvec_iter iter; 235 unsigned i = 0; 236 bool bounce = false; 237 int sectors = 0; 238 239 bio_for_each_segment(from, *bio_orig, iter) { 240 if (i++ < BIO_MAX_VECS) 241 sectors += from.bv_len >> 9; 242 if (PageHighMem(from.bv_page)) 243 bounce = true; 244 } 245 if (!bounce) 246 return; 247 248 if (!bio_is_passthrough(*bio_orig) && 249 sectors < bio_sectors(*bio_orig)) { 250 bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split); 251 bio_chain(bio, *bio_orig); 252 submit_bio_noacct(*bio_orig); 253 *bio_orig = bio; 254 } 255 bio = bounce_clone_bio(*bio_orig); 256 257 /* 258 * Bvec table can't be updated by bio_for_each_segment_all(), 259 * so retrieve bvec from the table directly. This way is safe 260 * because the 'bio' is single-page bvec. 261 */ 262 for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) { 263 struct page *page = to->bv_page; 264 265 if (!PageHighMem(page)) 266 continue; 267 268 to->bv_page = mempool_alloc(&page_pool, GFP_NOIO); 269 inc_zone_page_state(to->bv_page, NR_BOUNCE); 270 271 if (rw == WRITE) { 272 char *vto, *vfrom; 273 274 flush_dcache_page(page); 275 276 vto = page_address(to->bv_page) + to->bv_offset; 277 vfrom = kmap_atomic(page) + to->bv_offset; 278 memcpy(vto, vfrom, to->bv_len); 279 kunmap_atomic(vfrom); 280 } 281 } 282 283 trace_block_bio_bounce(*bio_orig); 284 285 bio->bi_flags |= (1 << BIO_BOUNCED); 286 287 if (rw == READ) 288 bio->bi_end_io = bounce_end_io_read; 289 else 290 bio->bi_end_io = bounce_end_io_write; 291 292 bio->bi_private = *bio_orig; 293 *bio_orig = bio; 294 } 295