1 // SPDX-License-Identifier: GPL-2.0
3 * Inline encryption support for fscrypt
5 * Copyright 2019 Google LLC
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.
15 #include <linux/blk-crypto-profile.h>
16 #include <linux/blkdev.h>
17 #include <linux/buffer_head.h>
18 #include <linux/sched/mm.h>
19 #include <linux/slab.h>
20 #include <linux/uio.h>
22 #include "fscrypt_private.h"
24 static int fscrypt_get_num_devices(struct super_block *sb)
26 if (sb->s_cop->get_num_devices)
27 return sb->s_cop->get_num_devices(sb);
31 static void fscrypt_get_devices(struct super_block *sb, int num_devs,
32 struct request_queue **devs)
35 devs[0] = bdev_get_queue(sb->s_bdev);
37 sb->s_cop->get_devices(sb, devs);
40 static unsigned int fscrypt_get_dun_bytes(const struct fscrypt_info *ci)
42 struct super_block *sb = ci->ci_inode->i_sb;
43 unsigned int flags = fscrypt_policy_flags(&ci->ci_policy);
44 int ino_bits = 64, lblk_bits = 64;
46 if (flags & FSCRYPT_POLICY_FLAG_DIRECT_KEY)
47 return offsetofend(union fscrypt_iv, nonce);
49 if (flags & FSCRYPT_POLICY_FLAG_IV_INO_LBLK_64)
50 return sizeof(__le64);
52 if (flags & FSCRYPT_POLICY_FLAG_IV_INO_LBLK_32)
53 return sizeof(__le32);
55 /* Default case: IVs are just the file logical block number */
56 if (sb->s_cop->get_ino_and_lblk_bits)
57 sb->s_cop->get_ino_and_lblk_bits(sb, &ino_bits, &lblk_bits);
58 return DIV_ROUND_UP(lblk_bits, 8);
62 * Log a message when starting to use blk-crypto (native) or blk-crypto-fallback
63 * for an encryption mode for the first time. This is the blk-crypto
64 * counterpart to the message logged when starting to use the crypto API for the
65 * first time. A limitation is that these messages don't convey which specific
66 * filesystems or files are using each implementation. However, *usually*
67 * systems use just one implementation per mode, which makes these messages
68 * helpful for debugging problems where the "wrong" implementation is used.
70 static void fscrypt_log_blk_crypto_impl(struct fscrypt_mode *mode,
71 struct request_queue **devs,
73 const struct blk_crypto_config *cfg)
77 for (i = 0; i < num_devs; i++) {
78 if (!IS_ENABLED(CONFIG_BLK_INLINE_ENCRYPTION_FALLBACK) ||
79 __blk_crypto_cfg_supported(devs[i]->crypto_profile, cfg)) {
80 if (!xchg(&mode->logged_blk_crypto_native, 1))
81 pr_info("fscrypt: %s using blk-crypto (native)\n",
83 } else if (!xchg(&mode->logged_blk_crypto_fallback, 1)) {
84 pr_info("fscrypt: %s using blk-crypto-fallback\n",
90 /* Enable inline encryption for this file if supported. */
91 int fscrypt_select_encryption_impl(struct fscrypt_info *ci)
93 const struct inode *inode = ci->ci_inode;
94 struct super_block *sb = inode->i_sb;
95 struct blk_crypto_config crypto_cfg;
97 struct request_queue **devs;
100 /* The file must need contents encryption, not filenames encryption */
101 if (!S_ISREG(inode->i_mode))
104 /* The crypto mode must have a blk-crypto counterpart */
105 if (ci->ci_mode->blk_crypto_mode == BLK_ENCRYPTION_MODE_INVALID)
108 /* The filesystem must be mounted with -o inlinecrypt */
109 if (!(sb->s_flags & SB_INLINECRYPT))
113 * When a page contains multiple logically contiguous filesystem blocks,
114 * some filesystem code only calls fscrypt_mergeable_bio() for the first
115 * block in the page. This is fine for most of fscrypt's IV generation
116 * strategies, where contiguous blocks imply contiguous IVs. But it
117 * doesn't work with IV_INO_LBLK_32. For now, simply exclude
118 * IV_INO_LBLK_32 with blocksize != PAGE_SIZE from inline encryption.
120 if ((fscrypt_policy_flags(&ci->ci_policy) &
121 FSCRYPT_POLICY_FLAG_IV_INO_LBLK_32) &&
122 sb->s_blocksize != PAGE_SIZE)
126 * On all the filesystem's devices, blk-crypto must support the crypto
127 * configuration that the file would use.
129 crypto_cfg.crypto_mode = ci->ci_mode->blk_crypto_mode;
130 crypto_cfg.data_unit_size = sb->s_blocksize;
131 crypto_cfg.dun_bytes = fscrypt_get_dun_bytes(ci);
132 num_devs = fscrypt_get_num_devices(sb);
133 devs = kmalloc_array(num_devs, sizeof(*devs), GFP_KERNEL);
136 fscrypt_get_devices(sb, num_devs, devs);
138 for (i = 0; i < num_devs; i++) {
139 if (!blk_crypto_config_supported(devs[i], &crypto_cfg))
143 fscrypt_log_blk_crypto_impl(ci->ci_mode, devs, num_devs, &crypto_cfg);
145 ci->ci_inlinecrypt = true;
152 int fscrypt_prepare_inline_crypt_key(struct fscrypt_prepared_key *prep_key,
154 const struct fscrypt_info *ci)
156 const struct inode *inode = ci->ci_inode;
157 struct super_block *sb = inode->i_sb;
158 enum blk_crypto_mode_num crypto_mode = ci->ci_mode->blk_crypto_mode;
159 struct blk_crypto_key *blk_key;
160 struct request_queue **devs;
161 unsigned int num_devs;
165 blk_key = kmalloc(sizeof(*blk_key), GFP_KERNEL);
169 err = blk_crypto_init_key(blk_key, raw_key, crypto_mode,
170 fscrypt_get_dun_bytes(ci), sb->s_blocksize);
172 fscrypt_err(inode, "error %d initializing blk-crypto key", err);
176 /* Start using blk-crypto on all the filesystem's block devices. */
177 num_devs = fscrypt_get_num_devices(sb);
178 devs = kmalloc_array(num_devs, sizeof(*devs), GFP_KERNEL);
183 fscrypt_get_devices(sb, num_devs, devs);
184 for (i = 0; i < num_devs; i++) {
185 err = blk_crypto_start_using_key(blk_key, devs[i]);
191 fscrypt_err(inode, "error %d starting to use blk-crypto", err);
196 * Pairs with the smp_load_acquire() in fscrypt_is_key_prepared().
197 * I.e., here we publish ->blk_key with a RELEASE barrier so that
198 * concurrent tasks can ACQUIRE it. Note that this concurrency is only
199 * possible for per-mode keys, not for per-file keys.
201 smp_store_release(&prep_key->blk_key, blk_key);
205 kfree_sensitive(blk_key);
209 void fscrypt_destroy_inline_crypt_key(struct super_block *sb,
210 struct fscrypt_prepared_key *prep_key)
212 struct blk_crypto_key *blk_key = prep_key->blk_key;
213 struct request_queue **devs;
214 unsigned int num_devs;
220 /* Evict the key from all the filesystem's block devices. */
221 num_devs = fscrypt_get_num_devices(sb);
222 devs = kmalloc_array(num_devs, sizeof(*devs), GFP_KERNEL);
224 fscrypt_get_devices(sb, num_devs, devs);
225 for (i = 0; i < num_devs; i++)
226 blk_crypto_evict_key(devs[i], blk_key);
229 kfree_sensitive(blk_key);
232 bool __fscrypt_inode_uses_inline_crypto(const struct inode *inode)
234 return inode->i_crypt_info->ci_inlinecrypt;
236 EXPORT_SYMBOL_GPL(__fscrypt_inode_uses_inline_crypto);
238 static void fscrypt_generate_dun(const struct fscrypt_info *ci, u64 lblk_num,
239 u64 dun[BLK_CRYPTO_DUN_ARRAY_SIZE])
244 fscrypt_generate_iv(&iv, lblk_num, ci);
246 BUILD_BUG_ON(FSCRYPT_MAX_IV_SIZE > BLK_CRYPTO_MAX_IV_SIZE);
247 memset(dun, 0, BLK_CRYPTO_MAX_IV_SIZE);
248 for (i = 0; i < ci->ci_mode->ivsize/sizeof(dun[0]); i++)
249 dun[i] = le64_to_cpu(iv.dun[i]);
253 * fscrypt_set_bio_crypt_ctx() - prepare a file contents bio for inline crypto
254 * @bio: a bio which will eventually be submitted to the file
255 * @inode: the file's inode
256 * @first_lblk: the first file logical block number in the I/O
257 * @gfp_mask: memory allocation flags - these must be a waiting mask so that
258 * bio_crypt_set_ctx can't fail.
260 * If the contents of the file should be encrypted (or decrypted) with inline
261 * encryption, then assign the appropriate encryption context to the bio.
263 * Normally the bio should be newly allocated (i.e. no pages added yet), as
264 * otherwise fscrypt_mergeable_bio() won't work as intended.
266 * The encryption context will be freed automatically when the bio is freed.
268 void fscrypt_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
269 u64 first_lblk, gfp_t gfp_mask)
271 const struct fscrypt_info *ci;
272 u64 dun[BLK_CRYPTO_DUN_ARRAY_SIZE];
274 if (!fscrypt_inode_uses_inline_crypto(inode))
276 ci = inode->i_crypt_info;
278 fscrypt_generate_dun(ci, first_lblk, dun);
279 bio_crypt_set_ctx(bio, ci->ci_enc_key.blk_key, dun, gfp_mask);
281 EXPORT_SYMBOL_GPL(fscrypt_set_bio_crypt_ctx);
283 /* Extract the inode and logical block number from a buffer_head. */
284 static bool bh_get_inode_and_lblk_num(const struct buffer_head *bh,
285 const struct inode **inode_ret,
288 struct page *page = bh->b_page;
289 const struct address_space *mapping;
290 const struct inode *inode;
293 * The ext4 journal (jbd2) can submit a buffer_head it directly created
294 * for a non-pagecache page. fscrypt doesn't care about these.
296 mapping = page_mapping(page);
299 inode = mapping->host;
302 *lblk_num_ret = ((u64)page->index << (PAGE_SHIFT - inode->i_blkbits)) +
303 (bh_offset(bh) >> inode->i_blkbits);
308 * fscrypt_set_bio_crypt_ctx_bh() - prepare a file contents bio for inline
310 * @bio: a bio which will eventually be submitted to the file
311 * @first_bh: the first buffer_head for which I/O will be submitted
312 * @gfp_mask: memory allocation flags
314 * Same as fscrypt_set_bio_crypt_ctx(), except this takes a buffer_head instead
315 * of an inode and block number directly.
317 void fscrypt_set_bio_crypt_ctx_bh(struct bio *bio,
318 const struct buffer_head *first_bh,
321 const struct inode *inode;
324 if (bh_get_inode_and_lblk_num(first_bh, &inode, &first_lblk))
325 fscrypt_set_bio_crypt_ctx(bio, inode, first_lblk, gfp_mask);
327 EXPORT_SYMBOL_GPL(fscrypt_set_bio_crypt_ctx_bh);
330 * fscrypt_mergeable_bio() - test whether data can be added to a bio
331 * @bio: the bio being built up
332 * @inode: the inode for the next part of the I/O
333 * @next_lblk: the next file logical block number in the I/O
335 * When building a bio which may contain data which should undergo inline
336 * encryption (or decryption) via fscrypt, filesystems should call this function
337 * to ensure that the resulting bio contains only contiguous data unit numbers.
338 * This will return false if the next part of the I/O cannot be merged with the
339 * bio because either the encryption key would be different or the encryption
340 * data unit numbers would be discontiguous.
342 * fscrypt_set_bio_crypt_ctx() must have already been called on the bio.
344 * This function isn't required in cases where crypto-mergeability is ensured in
345 * another way, such as I/O targeting only a single file (and thus a single key)
346 * combined with fscrypt_limit_io_blocks() to ensure DUN contiguity.
348 * Return: true iff the I/O is mergeable
350 bool fscrypt_mergeable_bio(struct bio *bio, const struct inode *inode,
353 const struct bio_crypt_ctx *bc = bio->bi_crypt_context;
354 u64 next_dun[BLK_CRYPTO_DUN_ARRAY_SIZE];
356 if (!!bc != fscrypt_inode_uses_inline_crypto(inode))
362 * Comparing the key pointers is good enough, as all I/O for each key
363 * uses the same pointer. I.e., there's currently no need to support
364 * merging requests where the keys are the same but the pointers differ.
366 if (bc->bc_key != inode->i_crypt_info->ci_enc_key.blk_key)
369 fscrypt_generate_dun(inode->i_crypt_info, next_lblk, next_dun);
370 return bio_crypt_dun_is_contiguous(bc, bio->bi_iter.bi_size, next_dun);
372 EXPORT_SYMBOL_GPL(fscrypt_mergeable_bio);
375 * fscrypt_mergeable_bio_bh() - test whether data can be added to a bio
376 * @bio: the bio being built up
377 * @next_bh: the next buffer_head for which I/O will be submitted
379 * Same as fscrypt_mergeable_bio(), except this takes a buffer_head instead of
380 * an inode and block number directly.
382 * Return: true iff the I/O is mergeable
384 bool fscrypt_mergeable_bio_bh(struct bio *bio,
385 const struct buffer_head *next_bh)
387 const struct inode *inode;
390 if (!bh_get_inode_and_lblk_num(next_bh, &inode, &next_lblk))
391 return !bio->bi_crypt_context;
393 return fscrypt_mergeable_bio(bio, inode, next_lblk);
395 EXPORT_SYMBOL_GPL(fscrypt_mergeable_bio_bh);
398 * fscrypt_dio_supported() - check whether a DIO (direct I/O) request is
399 * supported as far as encryption is concerned
400 * @iocb: the file and position the I/O is targeting
401 * @iter: the I/O data segment(s)
403 * Return: %true if there are no encryption constraints that prevent DIO from
404 * being supported; %false if DIO is unsupported. (Note that in the
405 * %true case, the filesystem might have other, non-encryption-related
406 * constraints that prevent DIO from actually being supported.)
408 bool fscrypt_dio_supported(struct kiocb *iocb, struct iov_iter *iter)
410 const struct inode *inode = file_inode(iocb->ki_filp);
411 const unsigned int blocksize = i_blocksize(inode);
413 /* If the file is unencrypted, no veto from us. */
414 if (!fscrypt_needs_contents_encryption(inode))
417 /* We only support DIO with inline crypto, not fs-layer crypto. */
418 if (!fscrypt_inode_uses_inline_crypto(inode))
422 * Since the granularity of encryption is filesystem blocks, the file
423 * position and total I/O length must be aligned to the filesystem block
424 * size -- not just to the block device's logical block size as is
425 * traditionally the case for DIO on many filesystems.
427 * We require that the user-provided memory buffers be filesystem block
428 * aligned too. It is simpler to have a single alignment value required
429 * for all properties of the I/O, as is normally the case for DIO.
430 * Also, allowing less aligned buffers would imply that data units could
431 * cross bvecs, which would greatly complicate the I/O stack, which
432 * assumes that bios can be split at any bvec boundary.
434 if (!IS_ALIGNED(iocb->ki_pos | iov_iter_alignment(iter), blocksize))
439 EXPORT_SYMBOL_GPL(fscrypt_dio_supported);
442 * fscrypt_limit_io_blocks() - limit I/O blocks to avoid discontiguous DUNs
443 * @inode: the file on which I/O is being done
444 * @lblk: the block at which the I/O is being started from
445 * @nr_blocks: the number of blocks we want to submit starting at @lblk
447 * Determine the limit to the number of blocks that can be submitted in a bio
448 * targeting @lblk without causing a data unit number (DUN) discontiguity.
450 * This is normally just @nr_blocks, as normally the DUNs just increment along
451 * with the logical blocks. (Or the file is not encrypted.)
453 * In rare cases, fscrypt can be using an IV generation method that allows the
454 * DUN to wrap around within logically contiguous blocks, and that wraparound
455 * will occur. If this happens, a value less than @nr_blocks will be returned
456 * so that the wraparound doesn't occur in the middle of a bio, which would
457 * cause encryption/decryption to produce wrong results.
459 * Return: the actual number of blocks that can be submitted
461 u64 fscrypt_limit_io_blocks(const struct inode *inode, u64 lblk, u64 nr_blocks)
463 const struct fscrypt_info *ci;
466 if (!fscrypt_inode_uses_inline_crypto(inode))
472 ci = inode->i_crypt_info;
473 if (!(fscrypt_policy_flags(&ci->ci_policy) &
474 FSCRYPT_POLICY_FLAG_IV_INO_LBLK_32))
477 /* With IV_INO_LBLK_32, the DUN can wrap around from U32_MAX to 0. */
479 dun = ci->ci_hashed_ino + lblk;
481 return min_t(u64, nr_blocks, (u64)U32_MAX + 1 - dun);
483 EXPORT_SYMBOL_GPL(fscrypt_limit_io_blocks);