1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Functions related to generic helpers functions 4 */ 5 #include <linux/kernel.h> 6 #include <linux/module.h> 7 #include <linux/bio.h> 8 #include <linux/blkdev.h> 9 #include <linux/scatterlist.h> 10 11 #include "blk.h" 12 13 static sector_t bio_discard_limit(struct block_device *bdev, sector_t sector) 14 { 15 unsigned int discard_granularity = bdev_discard_granularity(bdev); 16 sector_t granularity_aligned_sector; 17 18 if (bdev_is_partition(bdev)) 19 sector += bdev->bd_start_sect; 20 21 granularity_aligned_sector = 22 round_up(sector, discard_granularity >> SECTOR_SHIFT); 23 24 /* 25 * Make sure subsequent bios start aligned to the discard granularity if 26 * it needs to be split. 27 */ 28 if (granularity_aligned_sector != sector) 29 return granularity_aligned_sector - sector; 30 31 /* 32 * Align the bio size to the discard granularity to make splitting the bio 33 * at discard granularity boundaries easier in the driver if needed. 34 */ 35 return round_down(UINT_MAX, discard_granularity) >> SECTOR_SHIFT; 36 } 37 38 int __blkdev_issue_discard(struct block_device *bdev, sector_t sector, 39 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop) 40 { 41 struct bio *bio = *biop; 42 sector_t bs_mask; 43 44 if (bdev_read_only(bdev)) 45 return -EPERM; 46 if (!bdev_max_discard_sectors(bdev)) 47 return -EOPNOTSUPP; 48 49 /* In case the discard granularity isn't set by buggy device driver */ 50 if (WARN_ON_ONCE(!bdev_discard_granularity(bdev))) { 51 pr_err_ratelimited("%pg: Error: discard_granularity is 0.\n", 52 bdev); 53 return -EOPNOTSUPP; 54 } 55 56 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1; 57 if ((sector | nr_sects) & bs_mask) 58 return -EINVAL; 59 60 if (!nr_sects) 61 return -EINVAL; 62 63 while (nr_sects) { 64 sector_t req_sects = 65 min(nr_sects, bio_discard_limit(bdev, sector)); 66 67 bio = blk_next_bio(bio, bdev, 0, REQ_OP_DISCARD, gfp_mask); 68 bio->bi_iter.bi_sector = sector; 69 bio->bi_iter.bi_size = req_sects << 9; 70 sector += req_sects; 71 nr_sects -= req_sects; 72 73 /* 74 * We can loop for a long time in here, if someone does 75 * full device discards (like mkfs). Be nice and allow 76 * us to schedule out to avoid softlocking if preempt 77 * is disabled. 78 */ 79 cond_resched(); 80 } 81 82 *biop = bio; 83 return 0; 84 } 85 EXPORT_SYMBOL(__blkdev_issue_discard); 86 87 /** 88 * blkdev_issue_discard - queue a discard 89 * @bdev: blockdev to issue discard for 90 * @sector: start sector 91 * @nr_sects: number of sectors to discard 92 * @gfp_mask: memory allocation flags (for bio_alloc) 93 * 94 * Description: 95 * Issue a discard request for the sectors in question. 96 */ 97 int blkdev_issue_discard(struct block_device *bdev, sector_t sector, 98 sector_t nr_sects, gfp_t gfp_mask) 99 { 100 struct bio *bio = NULL; 101 struct blk_plug plug; 102 int ret; 103 104 blk_start_plug(&plug); 105 ret = __blkdev_issue_discard(bdev, sector, nr_sects, gfp_mask, &bio); 106 if (!ret && bio) { 107 ret = submit_bio_wait(bio); 108 if (ret == -EOPNOTSUPP) 109 ret = 0; 110 bio_put(bio); 111 } 112 blk_finish_plug(&plug); 113 114 return ret; 115 } 116 EXPORT_SYMBOL(blkdev_issue_discard); 117 118 static int __blkdev_issue_write_zeroes(struct block_device *bdev, 119 sector_t sector, sector_t nr_sects, gfp_t gfp_mask, 120 struct bio **biop, unsigned flags) 121 { 122 struct bio *bio = *biop; 123 unsigned int max_sectors; 124 125 if (bdev_read_only(bdev)) 126 return -EPERM; 127 128 /* Ensure that max_sectors doesn't overflow bi_size */ 129 max_sectors = bdev_write_zeroes_sectors(bdev); 130 131 if (max_sectors == 0) 132 return -EOPNOTSUPP; 133 134 while (nr_sects) { 135 unsigned int len = min_t(sector_t, nr_sects, max_sectors); 136 137 bio = blk_next_bio(bio, bdev, 0, REQ_OP_WRITE_ZEROES, gfp_mask); 138 bio->bi_iter.bi_sector = sector; 139 if (flags & BLKDEV_ZERO_NOUNMAP) 140 bio->bi_opf |= REQ_NOUNMAP; 141 142 bio->bi_iter.bi_size = len << SECTOR_SHIFT; 143 nr_sects -= len; 144 sector += len; 145 cond_resched(); 146 } 147 148 *biop = bio; 149 return 0; 150 } 151 152 /* 153 * Convert a number of 512B sectors to a number of pages. 154 * The result is limited to a number of pages that can fit into a BIO. 155 * Also make sure that the result is always at least 1 (page) for the cases 156 * where nr_sects is lower than the number of sectors in a page. 157 */ 158 static unsigned int __blkdev_sectors_to_bio_pages(sector_t nr_sects) 159 { 160 sector_t pages = DIV_ROUND_UP_SECTOR_T(nr_sects, PAGE_SIZE / 512); 161 162 return min(pages, (sector_t)BIO_MAX_VECS); 163 } 164 165 static int __blkdev_issue_zero_pages(struct block_device *bdev, 166 sector_t sector, sector_t nr_sects, gfp_t gfp_mask, 167 struct bio **biop) 168 { 169 struct bio *bio = *biop; 170 int bi_size = 0; 171 unsigned int sz; 172 173 if (bdev_read_only(bdev)) 174 return -EPERM; 175 176 while (nr_sects != 0) { 177 bio = blk_next_bio(bio, bdev, __blkdev_sectors_to_bio_pages(nr_sects), 178 REQ_OP_WRITE, gfp_mask); 179 bio->bi_iter.bi_sector = sector; 180 181 while (nr_sects != 0) { 182 sz = min((sector_t) PAGE_SIZE, nr_sects << 9); 183 bi_size = bio_add_page(bio, ZERO_PAGE(0), sz, 0); 184 nr_sects -= bi_size >> 9; 185 sector += bi_size >> 9; 186 if (bi_size < sz) 187 break; 188 } 189 cond_resched(); 190 } 191 192 *biop = bio; 193 return 0; 194 } 195 196 /** 197 * __blkdev_issue_zeroout - generate number of zero filed write bios 198 * @bdev: blockdev to issue 199 * @sector: start sector 200 * @nr_sects: number of sectors to write 201 * @gfp_mask: memory allocation flags (for bio_alloc) 202 * @biop: pointer to anchor bio 203 * @flags: controls detailed behavior 204 * 205 * Description: 206 * Zero-fill a block range, either using hardware offload or by explicitly 207 * writing zeroes to the device. 208 * 209 * If a device is using logical block provisioning, the underlying space will 210 * not be released if %flags contains BLKDEV_ZERO_NOUNMAP. 211 * 212 * If %flags contains BLKDEV_ZERO_NOFALLBACK, the function will return 213 * -EOPNOTSUPP if no explicit hardware offload for zeroing is provided. 214 */ 215 int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector, 216 sector_t nr_sects, gfp_t gfp_mask, struct bio **biop, 217 unsigned flags) 218 { 219 int ret; 220 sector_t bs_mask; 221 222 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1; 223 if ((sector | nr_sects) & bs_mask) 224 return -EINVAL; 225 226 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, gfp_mask, 227 biop, flags); 228 if (ret != -EOPNOTSUPP || (flags & BLKDEV_ZERO_NOFALLBACK)) 229 return ret; 230 231 return __blkdev_issue_zero_pages(bdev, sector, nr_sects, gfp_mask, 232 biop); 233 } 234 EXPORT_SYMBOL(__blkdev_issue_zeroout); 235 236 /** 237 * blkdev_issue_zeroout - zero-fill a block range 238 * @bdev: blockdev to write 239 * @sector: start sector 240 * @nr_sects: number of sectors to write 241 * @gfp_mask: memory allocation flags (for bio_alloc) 242 * @flags: controls detailed behavior 243 * 244 * Description: 245 * Zero-fill a block range, either using hardware offload or by explicitly 246 * writing zeroes to the device. See __blkdev_issue_zeroout() for the 247 * valid values for %flags. 248 */ 249 int blkdev_issue_zeroout(struct block_device *bdev, sector_t sector, 250 sector_t nr_sects, gfp_t gfp_mask, unsigned flags) 251 { 252 int ret = 0; 253 sector_t bs_mask; 254 struct bio *bio; 255 struct blk_plug plug; 256 bool try_write_zeroes = !!bdev_write_zeroes_sectors(bdev); 257 258 bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1; 259 if ((sector | nr_sects) & bs_mask) 260 return -EINVAL; 261 262 retry: 263 bio = NULL; 264 blk_start_plug(&plug); 265 if (try_write_zeroes) { 266 ret = __blkdev_issue_write_zeroes(bdev, sector, nr_sects, 267 gfp_mask, &bio, flags); 268 } else if (!(flags & BLKDEV_ZERO_NOFALLBACK)) { 269 ret = __blkdev_issue_zero_pages(bdev, sector, nr_sects, 270 gfp_mask, &bio); 271 } else { 272 /* No zeroing offload support */ 273 ret = -EOPNOTSUPP; 274 } 275 if (ret == 0 && bio) { 276 ret = submit_bio_wait(bio); 277 bio_put(bio); 278 } 279 blk_finish_plug(&plug); 280 if (ret && try_write_zeroes) { 281 if (!(flags & BLKDEV_ZERO_NOFALLBACK)) { 282 try_write_zeroes = false; 283 goto retry; 284 } 285 if (!bdev_write_zeroes_sectors(bdev)) { 286 /* 287 * Zeroing offload support was indicated, but the 288 * device reported ILLEGAL REQUEST (for some devices 289 * there is no non-destructive way to verify whether 290 * WRITE ZEROES is actually supported). 291 */ 292 ret = -EOPNOTSUPP; 293 } 294 } 295 296 return ret; 297 } 298 EXPORT_SYMBOL(blkdev_issue_zeroout); 299 300 int blkdev_issue_secure_erase(struct block_device *bdev, sector_t sector, 301 sector_t nr_sects, gfp_t gfp) 302 { 303 sector_t bs_mask = (bdev_logical_block_size(bdev) >> 9) - 1; 304 unsigned int max_sectors = bdev_max_secure_erase_sectors(bdev); 305 struct bio *bio = NULL; 306 struct blk_plug plug; 307 int ret = 0; 308 309 /* make sure that "len << SECTOR_SHIFT" doesn't overflow */ 310 if (max_sectors > UINT_MAX >> SECTOR_SHIFT) 311 max_sectors = UINT_MAX >> SECTOR_SHIFT; 312 max_sectors &= ~bs_mask; 313 314 if (max_sectors == 0) 315 return -EOPNOTSUPP; 316 if ((sector | nr_sects) & bs_mask) 317 return -EINVAL; 318 if (bdev_read_only(bdev)) 319 return -EPERM; 320 321 blk_start_plug(&plug); 322 while (nr_sects) { 323 unsigned int len = min_t(sector_t, nr_sects, max_sectors); 324 325 bio = blk_next_bio(bio, bdev, 0, REQ_OP_SECURE_ERASE, gfp); 326 bio->bi_iter.bi_sector = sector; 327 bio->bi_iter.bi_size = len << SECTOR_SHIFT; 328 329 sector += len; 330 nr_sects -= len; 331 cond_resched(); 332 } 333 if (bio) { 334 ret = submit_bio_wait(bio); 335 bio_put(bio); 336 } 337 blk_finish_plug(&plug); 338 339 return ret; 340 } 341 EXPORT_SYMBOL(blkdev_issue_secure_erase); 342