b.liu | e958203 | 2025-04-17 19:18:16 +0800 | [diff] [blame^] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Inline encryption support for fscrypt |
| 4 | * |
| 5 | * Copyright 2019 Google LLC |
| 6 | */ |
| 7 | |
| 8 | /* |
| 9 | * With "inline encryption", the block layer handles the decryption/encryption |
| 10 | * as part of the bio, instead of the filesystem doing the crypto itself via |
| 11 | * crypto API. See Documentation/block/inline-encryption.rst. fscrypt still |
| 12 | * provides the key and IV to use. |
| 13 | */ |
| 14 | |
| 15 | #include <linux/blk-crypto.h> |
| 16 | #include <linux/blkdev.h> |
| 17 | #include <linux/buffer_head.h> |
| 18 | #include <linux/keyslot-manager.h> |
| 19 | #include <linux/sched/mm.h> |
| 20 | #include <linux/uio.h> |
| 21 | |
| 22 | #include "fscrypt_private.h" |
| 23 | |
| 24 | struct fscrypt_blk_crypto_key { |
| 25 | struct blk_crypto_key base; |
| 26 | int num_devs; |
| 27 | struct request_queue *devs[]; |
| 28 | }; |
| 29 | |
| 30 | static int fscrypt_get_num_devices(struct super_block *sb) |
| 31 | { |
| 32 | if (sb->s_cop->get_num_devices) |
| 33 | return sb->s_cop->get_num_devices(sb); |
| 34 | return 1; |
| 35 | } |
| 36 | |
| 37 | static void fscrypt_get_devices(struct super_block *sb, int num_devs, |
| 38 | struct request_queue **devs) |
| 39 | { |
| 40 | if (num_devs == 1) |
| 41 | devs[0] = bdev_get_queue(sb->s_bdev); |
| 42 | else |
| 43 | sb->s_cop->get_devices(sb, devs); |
| 44 | } |
| 45 | |
| 46 | static unsigned int fscrypt_get_dun_bytes(const struct fscrypt_info *ci) |
| 47 | { |
| 48 | struct super_block *sb = ci->ci_inode->i_sb; |
| 49 | unsigned int flags = fscrypt_policy_flags(&ci->ci_policy); |
| 50 | int ino_bits = 64, lblk_bits = 64; |
| 51 | |
| 52 | if (flags & FSCRYPT_POLICY_FLAG_DIRECT_KEY) |
| 53 | return offsetofend(union fscrypt_iv, nonce); |
| 54 | |
| 55 | if (flags & FSCRYPT_POLICY_FLAG_IV_INO_LBLK_64) |
| 56 | return sizeof(__le64); |
| 57 | |
| 58 | if (flags & FSCRYPT_POLICY_FLAG_IV_INO_LBLK_32) |
| 59 | return sizeof(__le32); |
| 60 | |
| 61 | /* Default case: IVs are just the file logical block number */ |
| 62 | if (sb->s_cop->get_ino_and_lblk_bits) |
| 63 | sb->s_cop->get_ino_and_lblk_bits(sb, &ino_bits, &lblk_bits); |
| 64 | return DIV_ROUND_UP(lblk_bits, 8); |
| 65 | } |
| 66 | |
| 67 | /* Enable inline encryption for this file if supported. */ |
| 68 | int fscrypt_select_encryption_impl(struct fscrypt_info *ci, |
| 69 | bool is_hw_wrapped_key) |
| 70 | { |
| 71 | const struct inode *inode = ci->ci_inode; |
| 72 | struct super_block *sb = inode->i_sb; |
| 73 | struct blk_crypto_config crypto_cfg; |
| 74 | int num_devs; |
| 75 | struct request_queue **devs; |
| 76 | int i; |
| 77 | |
| 78 | /* The file must need contents encryption, not filenames encryption */ |
| 79 | if (!S_ISREG(inode->i_mode)) |
| 80 | return 0; |
| 81 | |
| 82 | /* The crypto mode must have a blk-crypto counterpart */ |
| 83 | if (ci->ci_mode->blk_crypto_mode == BLK_ENCRYPTION_MODE_INVALID) |
| 84 | return 0; |
| 85 | |
| 86 | /* The filesystem must be mounted with -o inlinecrypt */ |
| 87 | if (!(sb->s_flags & SB_INLINECRYPT)) |
| 88 | return 0; |
| 89 | |
| 90 | /* |
| 91 | * When a page contains multiple logically contiguous filesystem blocks, |
| 92 | * some filesystem code only calls fscrypt_mergeable_bio() for the first |
| 93 | * block in the page. This is fine for most of fscrypt's IV generation |
| 94 | * strategies, where contiguous blocks imply contiguous IVs. But it |
| 95 | * doesn't work with IV_INO_LBLK_32. For now, simply exclude |
| 96 | * IV_INO_LBLK_32 with blocksize != PAGE_SIZE from inline encryption. |
| 97 | */ |
| 98 | if ((fscrypt_policy_flags(&ci->ci_policy) & |
| 99 | FSCRYPT_POLICY_FLAG_IV_INO_LBLK_32) && |
| 100 | sb->s_blocksize != PAGE_SIZE) |
| 101 | return 0; |
| 102 | |
| 103 | /* |
| 104 | * On all the filesystem's devices, blk-crypto must support the crypto |
| 105 | * configuration that the file would use. |
| 106 | */ |
| 107 | crypto_cfg.crypto_mode = ci->ci_mode->blk_crypto_mode; |
| 108 | crypto_cfg.data_unit_size = sb->s_blocksize; |
| 109 | crypto_cfg.dun_bytes = fscrypt_get_dun_bytes(ci); |
| 110 | crypto_cfg.is_hw_wrapped = is_hw_wrapped_key; |
| 111 | num_devs = fscrypt_get_num_devices(sb); |
| 112 | devs = kmalloc_array(num_devs, sizeof(*devs), GFP_KERNEL); |
| 113 | if (!devs) |
| 114 | return -ENOMEM; |
| 115 | fscrypt_get_devices(sb, num_devs, devs); |
| 116 | |
| 117 | for (i = 0; i < num_devs; i++) { |
| 118 | if (!blk_crypto_config_supported(devs[i], &crypto_cfg)) |
| 119 | goto out_free_devs; |
| 120 | } |
| 121 | |
| 122 | ci->ci_inlinecrypt = true; |
| 123 | out_free_devs: |
| 124 | kfree(devs); |
| 125 | |
| 126 | return 0; |
| 127 | } |
| 128 | |
| 129 | int fscrypt_prepare_inline_crypt_key(struct fscrypt_prepared_key *prep_key, |
| 130 | const u8 *raw_key, |
| 131 | unsigned int raw_key_size, |
| 132 | bool is_hw_wrapped, |
| 133 | const struct fscrypt_info *ci) |
| 134 | { |
| 135 | const struct inode *inode = ci->ci_inode; |
| 136 | struct super_block *sb = inode->i_sb; |
| 137 | enum blk_crypto_mode_num crypto_mode = ci->ci_mode->blk_crypto_mode; |
| 138 | int num_devs = fscrypt_get_num_devices(sb); |
| 139 | int queue_refs = 0; |
| 140 | struct fscrypt_blk_crypto_key *blk_key; |
| 141 | int err; |
| 142 | int i; |
| 143 | |
| 144 | blk_key = kzalloc(struct_size(blk_key, devs, num_devs), GFP_KERNEL); |
| 145 | if (!blk_key) |
| 146 | return -ENOMEM; |
| 147 | |
| 148 | blk_key->num_devs = num_devs; |
| 149 | fscrypt_get_devices(sb, num_devs, blk_key->devs); |
| 150 | |
| 151 | BUILD_BUG_ON(FSCRYPT_MAX_HW_WRAPPED_KEY_SIZE > |
| 152 | BLK_CRYPTO_MAX_WRAPPED_KEY_SIZE); |
| 153 | |
| 154 | err = blk_crypto_init_key(&blk_key->base, raw_key, raw_key_size, |
| 155 | is_hw_wrapped, crypto_mode, |
| 156 | fscrypt_get_dun_bytes(ci), sb->s_blocksize); |
| 157 | if (err) { |
| 158 | fscrypt_err(inode, "error %d initializing blk-crypto key", err); |
| 159 | goto fail; |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * We have to start using blk-crypto on all the filesystem's devices. |
| 164 | * We also have to save all the request_queue's for later so that the |
| 165 | * key can be evicted from them. This is needed because some keys |
| 166 | * aren't destroyed until after the filesystem was already unmounted |
| 167 | * (namely, the per-mode keys in struct fscrypt_master_key). |
| 168 | */ |
| 169 | for (i = 0; i < num_devs; i++) { |
| 170 | if (!blk_get_queue(blk_key->devs[i])) { |
| 171 | fscrypt_err(inode, "couldn't get request_queue"); |
| 172 | err = -EAGAIN; |
| 173 | goto fail; |
| 174 | } |
| 175 | queue_refs++; |
| 176 | |
| 177 | err = blk_crypto_start_using_key(&blk_key->base, |
| 178 | blk_key->devs[i]); |
| 179 | if (err) { |
| 180 | fscrypt_err(inode, |
| 181 | "error %d starting to use blk-crypto", err); |
| 182 | goto fail; |
| 183 | } |
| 184 | } |
| 185 | /* |
| 186 | * Pairs with the smp_load_acquire() in fscrypt_is_key_prepared(). |
| 187 | * I.e., here we publish ->blk_key with a RELEASE barrier so that |
| 188 | * concurrent tasks can ACQUIRE it. Note that this concurrency is only |
| 189 | * possible for per-mode keys, not for per-file keys. |
| 190 | */ |
| 191 | smp_store_release(&prep_key->blk_key, blk_key); |
| 192 | return 0; |
| 193 | |
| 194 | fail: |
| 195 | for (i = 0; i < queue_refs; i++) |
| 196 | blk_put_queue(blk_key->devs[i]); |
| 197 | kzfree(blk_key); |
| 198 | return err; |
| 199 | } |
| 200 | |
| 201 | void fscrypt_destroy_inline_crypt_key(struct fscrypt_prepared_key *prep_key) |
| 202 | { |
| 203 | struct fscrypt_blk_crypto_key *blk_key = prep_key->blk_key; |
| 204 | int i; |
| 205 | |
| 206 | if (blk_key) { |
| 207 | for (i = 0; i < blk_key->num_devs; i++) { |
| 208 | blk_crypto_evict_key(blk_key->devs[i], &blk_key->base); |
| 209 | blk_put_queue(blk_key->devs[i]); |
| 210 | } |
| 211 | kzfree(blk_key); |
| 212 | } |
| 213 | } |
| 214 | |
| 215 | int fscrypt_derive_raw_secret(struct super_block *sb, |
| 216 | const u8 *wrapped_key, |
| 217 | unsigned int wrapped_key_size, |
| 218 | u8 *raw_secret, unsigned int raw_secret_size) |
| 219 | { |
| 220 | struct request_queue *q; |
| 221 | |
| 222 | q = bdev_get_queue(sb->s_bdev); |
| 223 | if (!q->ksm) |
| 224 | return -EOPNOTSUPP; |
| 225 | |
| 226 | return blk_ksm_derive_raw_secret(q->ksm, wrapped_key, wrapped_key_size, |
| 227 | raw_secret, raw_secret_size); |
| 228 | } |
| 229 | |
| 230 | bool __fscrypt_inode_uses_inline_crypto(const struct inode *inode) |
| 231 | { |
| 232 | return inode->i_crypt_info->ci_inlinecrypt; |
| 233 | } |
| 234 | EXPORT_SYMBOL_GPL(__fscrypt_inode_uses_inline_crypto); |
| 235 | |
| 236 | static void fscrypt_generate_dun(const struct fscrypt_info *ci, u64 lblk_num, |
| 237 | u64 dun[BLK_CRYPTO_DUN_ARRAY_SIZE]) |
| 238 | { |
| 239 | union fscrypt_iv iv; |
| 240 | int i; |
| 241 | |
| 242 | fscrypt_generate_iv(&iv, lblk_num, ci); |
| 243 | |
| 244 | BUILD_BUG_ON(FSCRYPT_MAX_IV_SIZE > BLK_CRYPTO_MAX_IV_SIZE); |
| 245 | memset(dun, 0, BLK_CRYPTO_MAX_IV_SIZE); |
| 246 | for (i = 0; i < ci->ci_mode->ivsize/sizeof(dun[0]); i++) |
| 247 | dun[i] = le64_to_cpu(iv.dun[i]); |
| 248 | } |
| 249 | |
| 250 | /** |
| 251 | * fscrypt_set_bio_crypt_ctx() - prepare a file contents bio for inline crypto |
| 252 | * @bio: a bio which will eventually be submitted to the file |
| 253 | * @inode: the file's inode |
| 254 | * @first_lblk: the first file logical block number in the I/O |
| 255 | * @gfp_mask: memory allocation flags - these must be a waiting mask so that |
| 256 | * bio_crypt_set_ctx can't fail. |
| 257 | * |
| 258 | * If the contents of the file should be encrypted (or decrypted) with inline |
| 259 | * encryption, then assign the appropriate encryption context to the bio. |
| 260 | * |
| 261 | * Normally the bio should be newly allocated (i.e. no pages added yet), as |
| 262 | * otherwise fscrypt_mergeable_bio() won't work as intended. |
| 263 | * |
| 264 | * The encryption context will be freed automatically when the bio is freed. |
| 265 | * |
| 266 | * This function also handles setting bi_skip_dm_default_key when needed. |
| 267 | */ |
| 268 | void fscrypt_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode, |
| 269 | u64 first_lblk, gfp_t gfp_mask) |
| 270 | { |
| 271 | const struct fscrypt_info *ci; |
| 272 | u64 dun[BLK_CRYPTO_DUN_ARRAY_SIZE]; |
| 273 | |
| 274 | if (fscrypt_inode_should_skip_dm_default_key(inode)) |
| 275 | bio_set_skip_dm_default_key(bio); |
| 276 | |
| 277 | if (!fscrypt_inode_uses_inline_crypto(inode)) |
| 278 | return; |
| 279 | ci = inode->i_crypt_info; |
| 280 | |
| 281 | fscrypt_generate_dun(ci, first_lblk, dun); |
| 282 | bio_crypt_set_ctx(bio, &ci->ci_enc_key.blk_key->base, dun, gfp_mask); |
| 283 | } |
| 284 | EXPORT_SYMBOL_GPL(fscrypt_set_bio_crypt_ctx); |
| 285 | |
| 286 | /* Extract the inode and logical block number from a buffer_head. */ |
| 287 | static bool bh_get_inode_and_lblk_num(const struct buffer_head *bh, |
| 288 | const struct inode **inode_ret, |
| 289 | u64 *lblk_num_ret) |
| 290 | { |
| 291 | struct page *page = bh->b_page; |
| 292 | const struct address_space *mapping; |
| 293 | const struct inode *inode; |
| 294 | |
| 295 | /* |
| 296 | * The ext4 journal (jbd2) can submit a buffer_head it directly created |
| 297 | * for a non-pagecache page. fscrypt doesn't care about these. |
| 298 | */ |
| 299 | mapping = page_mapping(page); |
| 300 | if (!mapping) |
| 301 | return false; |
| 302 | inode = mapping->host; |
| 303 | |
| 304 | *inode_ret = inode; |
| 305 | *lblk_num_ret = ((u64)page->index << (PAGE_SHIFT - inode->i_blkbits)) + |
| 306 | (bh_offset(bh) >> inode->i_blkbits); |
| 307 | return true; |
| 308 | } |
| 309 | |
| 310 | /** |
| 311 | * fscrypt_set_bio_crypt_ctx_bh() - prepare a file contents bio for inline |
| 312 | * crypto |
| 313 | * @bio: a bio which will eventually be submitted to the file |
| 314 | * @first_bh: the first buffer_head for which I/O will be submitted |
| 315 | * @gfp_mask: memory allocation flags |
| 316 | * |
| 317 | * Same as fscrypt_set_bio_crypt_ctx(), except this takes a buffer_head instead |
| 318 | * of an inode and block number directly. |
| 319 | */ |
| 320 | void fscrypt_set_bio_crypt_ctx_bh(struct bio *bio, |
| 321 | const struct buffer_head *first_bh, |
| 322 | gfp_t gfp_mask) |
| 323 | { |
| 324 | const struct inode *inode; |
| 325 | u64 first_lblk; |
| 326 | |
| 327 | if (bh_get_inode_and_lblk_num(first_bh, &inode, &first_lblk)) |
| 328 | fscrypt_set_bio_crypt_ctx(bio, inode, first_lblk, gfp_mask); |
| 329 | } |
| 330 | EXPORT_SYMBOL_GPL(fscrypt_set_bio_crypt_ctx_bh); |
| 331 | |
| 332 | /** |
| 333 | * fscrypt_mergeable_bio() - test whether data can be added to a bio |
| 334 | * @bio: the bio being built up |
| 335 | * @inode: the inode for the next part of the I/O |
| 336 | * @next_lblk: the next file logical block number in the I/O |
| 337 | * |
| 338 | * When building a bio which may contain data which should undergo inline |
| 339 | * encryption (or decryption) via fscrypt, filesystems should call this function |
| 340 | * to ensure that the resulting bio contains only contiguous data unit numbers. |
| 341 | * This will return false if the next part of the I/O cannot be merged with the |
| 342 | * bio because either the encryption key would be different or the encryption |
| 343 | * data unit numbers would be discontiguous. |
| 344 | * |
| 345 | * fscrypt_set_bio_crypt_ctx() must have already been called on the bio. |
| 346 | * |
| 347 | * This function also returns false if the next part of the I/O would need to |
| 348 | * have a different value for the bi_skip_dm_default_key flag. |
| 349 | * |
| 350 | * Return: true iff the I/O is mergeable |
| 351 | */ |
| 352 | bool fscrypt_mergeable_bio(struct bio *bio, const struct inode *inode, |
| 353 | u64 next_lblk) |
| 354 | { |
| 355 | const struct bio_crypt_ctx *bc = bio->bi_crypt_context; |
| 356 | u64 next_dun[BLK_CRYPTO_DUN_ARRAY_SIZE]; |
| 357 | |
| 358 | if (!!bc != fscrypt_inode_uses_inline_crypto(inode)) |
| 359 | return false; |
| 360 | if (bio_should_skip_dm_default_key(bio) != |
| 361 | fscrypt_inode_should_skip_dm_default_key(inode)) |
| 362 | return false; |
| 363 | if (!bc) |
| 364 | return true; |
| 365 | |
| 366 | /* |
| 367 | * Comparing the key pointers is good enough, as all I/O for each key |
| 368 | * uses the same pointer. I.e., there's currently no need to support |
| 369 | * merging requests where the keys are the same but the pointers differ. |
| 370 | */ |
| 371 | if (bc->bc_key != &inode->i_crypt_info->ci_enc_key.blk_key->base) |
| 372 | return false; |
| 373 | |
| 374 | fscrypt_generate_dun(inode->i_crypt_info, next_lblk, next_dun); |
| 375 | return bio_crypt_dun_is_contiguous(bc, bio->bi_iter.bi_size, next_dun); |
| 376 | } |
| 377 | EXPORT_SYMBOL_GPL(fscrypt_mergeable_bio); |
| 378 | |
| 379 | /** |
| 380 | * fscrypt_mergeable_bio_bh() - test whether data can be added to a bio |
| 381 | * @bio: the bio being built up |
| 382 | * @next_bh: the next buffer_head for which I/O will be submitted |
| 383 | * |
| 384 | * Same as fscrypt_mergeable_bio(), except this takes a buffer_head instead of |
| 385 | * an inode and block number directly. |
| 386 | * |
| 387 | * Return: true iff the I/O is mergeable |
| 388 | */ |
| 389 | bool fscrypt_mergeable_bio_bh(struct bio *bio, |
| 390 | const struct buffer_head *next_bh) |
| 391 | { |
| 392 | const struct inode *inode; |
| 393 | u64 next_lblk; |
| 394 | |
| 395 | if (!bh_get_inode_and_lblk_num(next_bh, &inode, &next_lblk)) |
| 396 | return !bio->bi_crypt_context && |
| 397 | !bio_should_skip_dm_default_key(bio); |
| 398 | |
| 399 | return fscrypt_mergeable_bio(bio, inode, next_lblk); |
| 400 | } |
| 401 | EXPORT_SYMBOL_GPL(fscrypt_mergeable_bio_bh); |
| 402 | |
| 403 | /** |
| 404 | * fscrypt_dio_supported() - check whether a direct I/O request is unsupported |
| 405 | * due to encryption constraints |
| 406 | * @iocb: the file and position the I/O is targeting |
| 407 | * @iter: the I/O data segment(s) |
| 408 | * |
| 409 | * Return: true if direct I/O is supported |
| 410 | */ |
| 411 | bool fscrypt_dio_supported(struct kiocb *iocb, struct iov_iter *iter) |
| 412 | { |
| 413 | const struct inode *inode = file_inode(iocb->ki_filp); |
| 414 | const unsigned int blocksize = i_blocksize(inode); |
| 415 | |
| 416 | /* If the file is unencrypted, no veto from us. */ |
| 417 | if (!fscrypt_needs_contents_encryption(inode)) |
| 418 | return true; |
| 419 | |
| 420 | /* We only support direct I/O with inline crypto, not fs-layer crypto */ |
| 421 | if (!fscrypt_inode_uses_inline_crypto(inode)) |
| 422 | return false; |
| 423 | |
| 424 | /* |
| 425 | * Since the granularity of encryption is filesystem blocks, the I/O |
| 426 | * must be block aligned -- not just disk sector aligned. |
| 427 | */ |
| 428 | if (!IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(iter), blocksize)) |
| 429 | return false; |
| 430 | |
| 431 | return true; |
| 432 | } |
| 433 | EXPORT_SYMBOL_GPL(fscrypt_dio_supported); |
| 434 | |
| 435 | /** |
| 436 | * fscrypt_limit_io_blocks() - limit I/O blocks to avoid discontiguous DUNs |
| 437 | * @inode: the file on which I/O is being done |
| 438 | * @lblk: the block at which the I/O is being started from |
| 439 | * @nr_blocks: the number of blocks we want to submit starting at @pos |
| 440 | * |
| 441 | * Determine the limit to the number of blocks that can be submitted in the bio |
| 442 | * targeting @pos without causing a data unit number (DUN) discontinuity. |
| 443 | * |
| 444 | * This is normally just @nr_blocks, as normally the DUNs just increment along |
| 445 | * with the logical blocks. (Or the file is not encrypted.) |
| 446 | * |
| 447 | * In rare cases, fscrypt can be using an IV generation method that allows the |
| 448 | * DUN to wrap around within logically continuous blocks, and that wraparound |
| 449 | * will occur. If this happens, a value less than @nr_blocks will be returned |
| 450 | * so that the wraparound doesn't occur in the middle of the bio. |
| 451 | * |
| 452 | * Return: the actual number of blocks that can be submitted |
| 453 | */ |
| 454 | u64 fscrypt_limit_io_blocks(const struct inode *inode, u64 lblk, u64 nr_blocks) |
| 455 | { |
| 456 | const struct fscrypt_info *ci = inode->i_crypt_info; |
| 457 | u32 dun; |
| 458 | |
| 459 | if (!fscrypt_inode_uses_inline_crypto(inode)) |
| 460 | return nr_blocks; |
| 461 | |
| 462 | if (nr_blocks <= 1) |
| 463 | return nr_blocks; |
| 464 | |
| 465 | if (!(fscrypt_policy_flags(&ci->ci_policy) & |
| 466 | FSCRYPT_POLICY_FLAG_IV_INO_LBLK_32)) |
| 467 | return nr_blocks; |
| 468 | |
| 469 | /* With IV_INO_LBLK_32, the DUN can wrap around from U32_MAX to 0. */ |
| 470 | |
| 471 | dun = ci->ci_hashed_ino + lblk; |
| 472 | |
| 473 | return min_t(u64, nr_blocks, (u64)U32_MAX + 1 - dun); |
| 474 | } |