| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * t10_pi.c - Functions for generating and verifying T10 Protection |
| 4 | * Information. |
| 5 | */ |
| 6 | |
| 7 | #include <linux/t10-pi.h> |
| 8 | #include <linux/blk-integrity.h> |
| 9 | #include <linux/crc-t10dif.h> |
| 10 | #include <linux/crc64.h> |
| 11 | #include <net/checksum.h> |
| 12 | #include <linux/unaligned.h> |
| 13 | #include "blk.h" |
| 14 | |
| 15 | struct blk_integrity_iter { |
| 16 | void *prot_buf; |
| 17 | void *data_buf; |
| 18 | sector_t seed; |
| 19 | unsigned int data_size; |
| 20 | unsigned short interval; |
| 21 | const char *disk_name; |
| 22 | }; |
| 23 | |
| 24 | static __be16 t10_pi_csum(__be16 csum, void *data, unsigned int len, |
| 25 | unsigned char csum_type) |
| 26 | { |
| 27 | if (csum_type == BLK_INTEGRITY_CSUM_IP) |
| 28 | return (__force __be16)ip_compute_csum(data, len); |
| 29 | return cpu_to_be16(crc_t10dif_update(be16_to_cpu(csum), data, len)); |
| 30 | } |
| 31 | |
| 32 | /* |
| 33 | * Type 1 and Type 2 protection use the same format: 16 bit guard tag, |
| 34 | * 16 bit app tag, 32 bit reference tag. Type 3 does not define the ref |
| 35 | * tag. |
| 36 | */ |
| 37 | static void t10_pi_generate(struct blk_integrity_iter *iter, |
| 38 | struct blk_integrity *bi) |
| 39 | { |
| 40 | u8 offset = bi->pi_offset; |
| 41 | unsigned int i; |
| 42 | |
| 43 | for (i = 0 ; i < iter->data_size ; i += iter->interval) { |
| 44 | struct t10_pi_tuple *pi = iter->prot_buf + offset; |
| 45 | |
| 46 | pi->guard_tag = t10_pi_csum(0, iter->data_buf, iter->interval, |
| 47 | bi->csum_type); |
| 48 | if (offset) |
| 49 | pi->guard_tag = t10_pi_csum(pi->guard_tag, |
| 50 | iter->prot_buf, offset, bi->csum_type); |
| 51 | pi->app_tag = 0; |
| 52 | |
| 53 | if (bi->flags & BLK_INTEGRITY_REF_TAG) |
| 54 | pi->ref_tag = cpu_to_be32(lower_32_bits(iter->seed)); |
| 55 | else |
| 56 | pi->ref_tag = 0; |
| 57 | |
| 58 | iter->data_buf += iter->interval; |
| 59 | iter->prot_buf += bi->tuple_size; |
| 60 | iter->seed++; |
| 61 | } |
| 62 | } |
| 63 | |
| 64 | static blk_status_t t10_pi_verify(struct blk_integrity_iter *iter, |
| 65 | struct blk_integrity *bi) |
| 66 | { |
| 67 | u8 offset = bi->pi_offset; |
| 68 | unsigned int i; |
| 69 | |
| 70 | for (i = 0 ; i < iter->data_size ; i += iter->interval) { |
| 71 | struct t10_pi_tuple *pi = iter->prot_buf + offset; |
| 72 | __be16 csum; |
| 73 | |
| 74 | if (bi->flags & BLK_INTEGRITY_REF_TAG) { |
| 75 | if (pi->app_tag == T10_PI_APP_ESCAPE) |
| 76 | goto next; |
| 77 | |
| 78 | if (be32_to_cpu(pi->ref_tag) != |
| 79 | lower_32_bits(iter->seed)) { |
| 80 | pr_err("%s: ref tag error at location %llu " \ |
| 81 | "(rcvd %u)\n", iter->disk_name, |
| 82 | (unsigned long long) |
| 83 | iter->seed, be32_to_cpu(pi->ref_tag)); |
| 84 | return BLK_STS_PROTECTION; |
| 85 | } |
| 86 | } else { |
| 87 | if (pi->app_tag == T10_PI_APP_ESCAPE && |
| 88 | pi->ref_tag == T10_PI_REF_ESCAPE) |
| 89 | goto next; |
| 90 | } |
| 91 | |
| 92 | csum = t10_pi_csum(0, iter->data_buf, iter->interval, |
| 93 | bi->csum_type); |
| 94 | if (offset) |
| 95 | csum = t10_pi_csum(csum, iter->prot_buf, offset, |
| 96 | bi->csum_type); |
| 97 | |
| 98 | if (pi->guard_tag != csum) { |
| 99 | pr_err("%s: guard tag error at sector %llu " \ |
| 100 | "(rcvd %04x, want %04x)\n", iter->disk_name, |
| 101 | (unsigned long long)iter->seed, |
| 102 | be16_to_cpu(pi->guard_tag), be16_to_cpu(csum)); |
| 103 | return BLK_STS_PROTECTION; |
| 104 | } |
| 105 | |
| 106 | next: |
| 107 | iter->data_buf += iter->interval; |
| 108 | iter->prot_buf += bi->tuple_size; |
| 109 | iter->seed++; |
| 110 | } |
| 111 | |
| 112 | return BLK_STS_OK; |
| 113 | } |
| 114 | |
| 115 | /** |
| 116 | * t10_pi_type1_prepare - prepare PI prior submitting request to device |
| 117 | * @rq: request with PI that should be prepared |
| 118 | * |
| 119 | * For Type 1/Type 2, the virtual start sector is the one that was |
| 120 | * originally submitted by the block layer for the ref_tag usage. Due to |
| 121 | * partitioning, MD/DM cloning, etc. the actual physical start sector is |
| 122 | * likely to be different. Remap protection information to match the |
| 123 | * physical LBA. |
| 124 | */ |
| 125 | static void t10_pi_type1_prepare(struct request *rq) |
| 126 | { |
| 127 | struct blk_integrity *bi = &rq->q->limits.integrity; |
| 128 | const int tuple_sz = bi->tuple_size; |
| 129 | u32 ref_tag = t10_pi_ref_tag(rq); |
| 130 | u8 offset = bi->pi_offset; |
| 131 | struct bio *bio; |
| 132 | |
| 133 | __rq_for_each_bio(bio, rq) { |
| 134 | struct bio_integrity_payload *bip = bio_integrity(bio); |
| 135 | u32 virt = bip_get_seed(bip) & 0xffffffff; |
| 136 | struct bio_vec iv; |
| 137 | struct bvec_iter iter; |
| 138 | |
| 139 | /* Already remapped? */ |
| 140 | if (bip->bip_flags & BIP_MAPPED_INTEGRITY) |
| 141 | break; |
| 142 | |
| 143 | bip_for_each_vec(iv, bip, iter) { |
| 144 | unsigned int j; |
| 145 | void *p; |
| 146 | |
| 147 | p = bvec_kmap_local(&iv); |
| 148 | for (j = 0; j < iv.bv_len; j += tuple_sz) { |
| 149 | struct t10_pi_tuple *pi = p + offset; |
| 150 | |
| 151 | if (be32_to_cpu(pi->ref_tag) == virt) |
| 152 | pi->ref_tag = cpu_to_be32(ref_tag); |
| 153 | virt++; |
| 154 | ref_tag++; |
| 155 | p += tuple_sz; |
| 156 | } |
| 157 | kunmap_local(p); |
| 158 | } |
| 159 | |
| 160 | bip->bip_flags |= BIP_MAPPED_INTEGRITY; |
| 161 | } |
| 162 | } |
| 163 | |
| 164 | /** |
| 165 | * t10_pi_type1_complete - prepare PI prior returning request to the blk layer |
| 166 | * @rq: request with PI that should be prepared |
| 167 | * @nr_bytes: total bytes to prepare |
| 168 | * |
| 169 | * For Type 1/Type 2, the virtual start sector is the one that was |
| 170 | * originally submitted by the block layer for the ref_tag usage. Due to |
| 171 | * partitioning, MD/DM cloning, etc. the actual physical start sector is |
| 172 | * likely to be different. Since the physical start sector was submitted |
| 173 | * to the device, we should remap it back to virtual values expected by the |
| 174 | * block layer. |
| 175 | */ |
| 176 | static void t10_pi_type1_complete(struct request *rq, unsigned int nr_bytes) |
| 177 | { |
| 178 | struct blk_integrity *bi = &rq->q->limits.integrity; |
| 179 | unsigned intervals = nr_bytes >> bi->interval_exp; |
| 180 | const int tuple_sz = bi->tuple_size; |
| 181 | u32 ref_tag = t10_pi_ref_tag(rq); |
| 182 | u8 offset = bi->pi_offset; |
| 183 | struct bio *bio; |
| 184 | |
| 185 | __rq_for_each_bio(bio, rq) { |
| 186 | struct bio_integrity_payload *bip = bio_integrity(bio); |
| 187 | u32 virt = bip_get_seed(bip) & 0xffffffff; |
| 188 | struct bio_vec iv; |
| 189 | struct bvec_iter iter; |
| 190 | |
| 191 | bip_for_each_vec(iv, bip, iter) { |
| 192 | unsigned int j; |
| 193 | void *p; |
| 194 | |
| 195 | p = bvec_kmap_local(&iv); |
| 196 | for (j = 0; j < iv.bv_len && intervals; j += tuple_sz) { |
| 197 | struct t10_pi_tuple *pi = p + offset; |
| 198 | |
| 199 | if (be32_to_cpu(pi->ref_tag) == ref_tag) |
| 200 | pi->ref_tag = cpu_to_be32(virt); |
| 201 | virt++; |
| 202 | ref_tag++; |
| 203 | intervals--; |
| 204 | p += tuple_sz; |
| 205 | } |
| 206 | kunmap_local(p); |
| 207 | } |
| 208 | } |
| 209 | } |
| 210 | |
| 211 | static __be64 ext_pi_crc64(u64 crc, void *data, unsigned int len) |
| 212 | { |
| 213 | return cpu_to_be64(crc64_nvme(crc, data, len)); |
| 214 | } |
| 215 | |
| 216 | static void ext_pi_crc64_generate(struct blk_integrity_iter *iter, |
| 217 | struct blk_integrity *bi) |
| 218 | { |
| 219 | u8 offset = bi->pi_offset; |
| 220 | unsigned int i; |
| 221 | |
| 222 | for (i = 0 ; i < iter->data_size ; i += iter->interval) { |
| 223 | struct crc64_pi_tuple *pi = iter->prot_buf + offset; |
| 224 | |
| 225 | pi->guard_tag = ext_pi_crc64(0, iter->data_buf, iter->interval); |
| 226 | if (offset) |
| 227 | pi->guard_tag = ext_pi_crc64(be64_to_cpu(pi->guard_tag), |
| 228 | iter->prot_buf, offset); |
| 229 | pi->app_tag = 0; |
| 230 | |
| 231 | if (bi->flags & BLK_INTEGRITY_REF_TAG) |
| 232 | put_unaligned_be48(iter->seed, pi->ref_tag); |
| 233 | else |
| 234 | put_unaligned_be48(0ULL, pi->ref_tag); |
| 235 | |
| 236 | iter->data_buf += iter->interval; |
| 237 | iter->prot_buf += bi->tuple_size; |
| 238 | iter->seed++; |
| 239 | } |
| 240 | } |
| 241 | |
| 242 | static bool ext_pi_ref_escape(const u8 ref_tag[6]) |
| 243 | { |
| 244 | static const u8 ref_escape[6] = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff }; |
| 245 | |
| 246 | return memcmp(ref_tag, ref_escape, sizeof(ref_escape)) == 0; |
| 247 | } |
| 248 | |
| 249 | static blk_status_t ext_pi_crc64_verify(struct blk_integrity_iter *iter, |
| 250 | struct blk_integrity *bi) |
| 251 | { |
| 252 | u8 offset = bi->pi_offset; |
| 253 | unsigned int i; |
| 254 | |
| 255 | for (i = 0; i < iter->data_size; i += iter->interval) { |
| 256 | struct crc64_pi_tuple *pi = iter->prot_buf + offset; |
| 257 | u64 ref, seed; |
| 258 | __be64 csum; |
| 259 | |
| 260 | if (bi->flags & BLK_INTEGRITY_REF_TAG) { |
| 261 | if (pi->app_tag == T10_PI_APP_ESCAPE) |
| 262 | goto next; |
| 263 | |
| 264 | ref = get_unaligned_be48(pi->ref_tag); |
| 265 | seed = lower_48_bits(iter->seed); |
| 266 | if (ref != seed) { |
| 267 | pr_err("%s: ref tag error at location %llu (rcvd %llu)\n", |
| 268 | iter->disk_name, seed, ref); |
| 269 | return BLK_STS_PROTECTION; |
| 270 | } |
| 271 | } else { |
| 272 | if (pi->app_tag == T10_PI_APP_ESCAPE && |
| 273 | ext_pi_ref_escape(pi->ref_tag)) |
| 274 | goto next; |
| 275 | } |
| 276 | |
| 277 | csum = ext_pi_crc64(0, iter->data_buf, iter->interval); |
| 278 | if (offset) |
| 279 | csum = ext_pi_crc64(be64_to_cpu(csum), iter->prot_buf, |
| 280 | offset); |
| 281 | |
| 282 | if (pi->guard_tag != csum) { |
| 283 | pr_err("%s: guard tag error at sector %llu " \ |
| 284 | "(rcvd %016llx, want %016llx)\n", |
| 285 | iter->disk_name, (unsigned long long)iter->seed, |
| 286 | be64_to_cpu(pi->guard_tag), be64_to_cpu(csum)); |
| 287 | return BLK_STS_PROTECTION; |
| 288 | } |
| 289 | |
| 290 | next: |
| 291 | iter->data_buf += iter->interval; |
| 292 | iter->prot_buf += bi->tuple_size; |
| 293 | iter->seed++; |
| 294 | } |
| 295 | |
| 296 | return BLK_STS_OK; |
| 297 | } |
| 298 | |
| 299 | static void ext_pi_type1_prepare(struct request *rq) |
| 300 | { |
| 301 | struct blk_integrity *bi = &rq->q->limits.integrity; |
| 302 | const int tuple_sz = bi->tuple_size; |
| 303 | u64 ref_tag = ext_pi_ref_tag(rq); |
| 304 | u8 offset = bi->pi_offset; |
| 305 | struct bio *bio; |
| 306 | |
| 307 | __rq_for_each_bio(bio, rq) { |
| 308 | struct bio_integrity_payload *bip = bio_integrity(bio); |
| 309 | u64 virt = lower_48_bits(bip_get_seed(bip)); |
| 310 | struct bio_vec iv; |
| 311 | struct bvec_iter iter; |
| 312 | |
| 313 | /* Already remapped? */ |
| 314 | if (bip->bip_flags & BIP_MAPPED_INTEGRITY) |
| 315 | break; |
| 316 | |
| 317 | bip_for_each_vec(iv, bip, iter) { |
| 318 | unsigned int j; |
| 319 | void *p; |
| 320 | |
| 321 | p = bvec_kmap_local(&iv); |
| 322 | for (j = 0; j < iv.bv_len; j += tuple_sz) { |
| 323 | struct crc64_pi_tuple *pi = p + offset; |
| 324 | u64 ref = get_unaligned_be48(pi->ref_tag); |
| 325 | |
| 326 | if (ref == virt) |
| 327 | put_unaligned_be48(ref_tag, pi->ref_tag); |
| 328 | virt++; |
| 329 | ref_tag++; |
| 330 | p += tuple_sz; |
| 331 | } |
| 332 | kunmap_local(p); |
| 333 | } |
| 334 | |
| 335 | bip->bip_flags |= BIP_MAPPED_INTEGRITY; |
| 336 | } |
| 337 | } |
| 338 | |
| 339 | static void ext_pi_type1_complete(struct request *rq, unsigned int nr_bytes) |
| 340 | { |
| 341 | struct blk_integrity *bi = &rq->q->limits.integrity; |
| 342 | unsigned intervals = nr_bytes >> bi->interval_exp; |
| 343 | const int tuple_sz = bi->tuple_size; |
| 344 | u64 ref_tag = ext_pi_ref_tag(rq); |
| 345 | u8 offset = bi->pi_offset; |
| 346 | struct bio *bio; |
| 347 | |
| 348 | __rq_for_each_bio(bio, rq) { |
| 349 | struct bio_integrity_payload *bip = bio_integrity(bio); |
| 350 | u64 virt = lower_48_bits(bip_get_seed(bip)); |
| 351 | struct bio_vec iv; |
| 352 | struct bvec_iter iter; |
| 353 | |
| 354 | bip_for_each_vec(iv, bip, iter) { |
| 355 | unsigned int j; |
| 356 | void *p; |
| 357 | |
| 358 | p = bvec_kmap_local(&iv); |
| 359 | for (j = 0; j < iv.bv_len && intervals; j += tuple_sz) { |
| 360 | struct crc64_pi_tuple *pi = p + offset; |
| 361 | u64 ref = get_unaligned_be48(pi->ref_tag); |
| 362 | |
| 363 | if (ref == ref_tag) |
| 364 | put_unaligned_be48(virt, pi->ref_tag); |
| 365 | virt++; |
| 366 | ref_tag++; |
| 367 | intervals--; |
| 368 | p += tuple_sz; |
| 369 | } |
| 370 | kunmap_local(p); |
| 371 | } |
| 372 | } |
| 373 | } |
| 374 | |
| 375 | void blk_integrity_generate(struct bio *bio) |
| 376 | { |
| 377 | struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk); |
| 378 | struct bio_integrity_payload *bip = bio_integrity(bio); |
| 379 | struct blk_integrity_iter iter; |
| 380 | struct bvec_iter bviter; |
| 381 | struct bio_vec bv; |
| 382 | |
| 383 | iter.disk_name = bio->bi_bdev->bd_disk->disk_name; |
| 384 | iter.interval = 1 << bi->interval_exp; |
| 385 | iter.seed = bio->bi_iter.bi_sector; |
| 386 | iter.prot_buf = bvec_virt(bip->bip_vec); |
| 387 | bio_for_each_segment(bv, bio, bviter) { |
| 388 | void *kaddr = bvec_kmap_local(&bv); |
| 389 | |
| 390 | iter.data_buf = kaddr; |
| 391 | iter.data_size = bv.bv_len; |
| 392 | switch (bi->csum_type) { |
| 393 | case BLK_INTEGRITY_CSUM_CRC64: |
| 394 | ext_pi_crc64_generate(&iter, bi); |
| 395 | break; |
| 396 | case BLK_INTEGRITY_CSUM_CRC: |
| 397 | case BLK_INTEGRITY_CSUM_IP: |
| 398 | t10_pi_generate(&iter, bi); |
| 399 | break; |
| 400 | default: |
| 401 | break; |
| 402 | } |
| 403 | kunmap_local(kaddr); |
| 404 | } |
| 405 | } |
| 406 | |
| 407 | void blk_integrity_verify_iter(struct bio *bio, struct bvec_iter *saved_iter) |
| 408 | { |
| 409 | struct blk_integrity *bi = blk_get_integrity(bio->bi_bdev->bd_disk); |
| 410 | struct bio_integrity_payload *bip = bio_integrity(bio); |
| 411 | struct blk_integrity_iter iter; |
| 412 | struct bvec_iter bviter; |
| 413 | struct bio_vec bv; |
| 414 | |
| 415 | /* |
| 416 | * At the moment verify is called bi_iter has been advanced during split |
| 417 | * and completion, so use the copy created during submission here. |
| 418 | */ |
| 419 | iter.disk_name = bio->bi_bdev->bd_disk->disk_name; |
| 420 | iter.interval = 1 << bi->interval_exp; |
| 421 | iter.seed = saved_iter->bi_sector; |
| 422 | iter.prot_buf = bvec_virt(bip->bip_vec); |
| 423 | __bio_for_each_segment(bv, bio, bviter, *saved_iter) { |
| 424 | void *kaddr = bvec_kmap_local(&bv); |
| 425 | blk_status_t ret = BLK_STS_OK; |
| 426 | |
| 427 | iter.data_buf = kaddr; |
| 428 | iter.data_size = bv.bv_len; |
| 429 | switch (bi->csum_type) { |
| 430 | case BLK_INTEGRITY_CSUM_CRC64: |
| 431 | ret = ext_pi_crc64_verify(&iter, bi); |
| 432 | break; |
| 433 | case BLK_INTEGRITY_CSUM_CRC: |
| 434 | case BLK_INTEGRITY_CSUM_IP: |
| 435 | ret = t10_pi_verify(&iter, bi); |
| 436 | break; |
| 437 | default: |
| 438 | break; |
| 439 | } |
| 440 | kunmap_local(kaddr); |
| 441 | |
| 442 | if (ret) { |
| 443 | bio->bi_status = ret; |
| 444 | return; |
| 445 | } |
| 446 | } |
| 447 | } |
| 448 | |
| 449 | void blk_integrity_prepare(struct request *rq) |
| 450 | { |
| 451 | struct blk_integrity *bi = &rq->q->limits.integrity; |
| 452 | |
| 453 | if (!(bi->flags & BLK_INTEGRITY_REF_TAG)) |
| 454 | return; |
| 455 | |
| 456 | if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC64) |
| 457 | ext_pi_type1_prepare(rq); |
| 458 | else |
| 459 | t10_pi_type1_prepare(rq); |
| 460 | } |
| 461 | |
| 462 | void blk_integrity_complete(struct request *rq, unsigned int nr_bytes) |
| 463 | { |
| 464 | struct blk_integrity *bi = &rq->q->limits.integrity; |
| 465 | |
| 466 | if (!(bi->flags & BLK_INTEGRITY_REF_TAG)) |
| 467 | return; |
| 468 | |
| 469 | if (bi->csum_type == BLK_INTEGRITY_CSUM_CRC64) |
| 470 | ext_pi_type1_complete(rq, nr_bytes); |
| 471 | else |
| 472 | t10_pi_type1_complete(rq, nr_bytes); |
| 473 | } |