1 // SPDX-License-Identifier: GPL-2.0-only
3 * This contains encryption functions for per-file encryption.
5 * Copyright (C) 2015, Google, Inc.
6 * Copyright (C) 2015, Motorola Mobility
8 * Written by Michael Halcrow, 2014.
10 * Filename encryption additions
11 * Uday Savagaonkar, 2014
12 * Encryption policy handling additions
13 * Ildar Muslukhov, 2014
14 * Add fscrypt_pullback_bio_page()
17 * This has not yet undergone a rigorous security audit.
19 * The usage of AES-XTS should conform to recommendations in NIST
20 * Special Publication 800-38E and IEEE P1619/D16.
23 #include <linux/pagemap.h>
24 #include <linux/mempool.h>
25 #include <linux/module.h>
26 #include <linux/scatterlist.h>
27 #include <linux/ratelimit.h>
28 #include <linux/dcache.h>
29 #include <linux/namei.h>
30 #include <crypto/aes.h>
31 #include <crypto/skcipher.h>
32 #include "fscrypt_private.h"
34 static unsigned int num_prealloc_crypto_pages = 32;
35 static unsigned int num_prealloc_crypto_ctxs = 128;
37 module_param(num_prealloc_crypto_pages, uint, 0444);
38 MODULE_PARM_DESC(num_prealloc_crypto_pages,
39 "Number of crypto pages to preallocate");
40 module_param(num_prealloc_crypto_ctxs, uint, 0444);
41 MODULE_PARM_DESC(num_prealloc_crypto_ctxs,
42 "Number of crypto contexts to preallocate");
44 static mempool_t *fscrypt_bounce_page_pool = NULL;
46 static LIST_HEAD(fscrypt_free_ctxs);
47 static DEFINE_SPINLOCK(fscrypt_ctx_lock);
49 static struct workqueue_struct *fscrypt_read_workqueue;
50 static DEFINE_MUTEX(fscrypt_init_mutex);
52 static struct kmem_cache *fscrypt_ctx_cachep;
53 struct kmem_cache *fscrypt_info_cachep;
55 void fscrypt_enqueue_decrypt_work(struct work_struct *work)
57 queue_work(fscrypt_read_workqueue, work);
59 EXPORT_SYMBOL(fscrypt_enqueue_decrypt_work);
62 * fscrypt_release_ctx() - Release a decryption context
63 * @ctx: The decryption context to release.
65 * If the decryption context was allocated from the pre-allocated pool, return
66 * it to that pool. Else, free it.
68 void fscrypt_release_ctx(struct fscrypt_ctx *ctx)
72 if (ctx->flags & FS_CTX_REQUIRES_FREE_ENCRYPT_FL) {
73 kmem_cache_free(fscrypt_ctx_cachep, ctx);
75 spin_lock_irqsave(&fscrypt_ctx_lock, flags);
76 list_add(&ctx->free_list, &fscrypt_free_ctxs);
77 spin_unlock_irqrestore(&fscrypt_ctx_lock, flags);
80 EXPORT_SYMBOL(fscrypt_release_ctx);
83 * fscrypt_get_ctx() - Get a decryption context
84 * @gfp_flags: The gfp flag for memory allocation
86 * Allocate and initialize a decryption context.
88 * Return: A new decryption context on success; an ERR_PTR() otherwise.
90 struct fscrypt_ctx *fscrypt_get_ctx(gfp_t gfp_flags)
92 struct fscrypt_ctx *ctx;
96 * First try getting a ctx from the free list so that we don't have to
97 * call into the slab allocator.
99 spin_lock_irqsave(&fscrypt_ctx_lock, flags);
100 ctx = list_first_entry_or_null(&fscrypt_free_ctxs,
101 struct fscrypt_ctx, free_list);
103 list_del(&ctx->free_list);
104 spin_unlock_irqrestore(&fscrypt_ctx_lock, flags);
106 ctx = kmem_cache_zalloc(fscrypt_ctx_cachep, gfp_flags);
108 return ERR_PTR(-ENOMEM);
109 ctx->flags |= FS_CTX_REQUIRES_FREE_ENCRYPT_FL;
111 ctx->flags &= ~FS_CTX_REQUIRES_FREE_ENCRYPT_FL;
115 EXPORT_SYMBOL(fscrypt_get_ctx);
117 struct page *fscrypt_alloc_bounce_page(gfp_t gfp_flags)
119 return mempool_alloc(fscrypt_bounce_page_pool, gfp_flags);
123 * fscrypt_free_bounce_page() - free a ciphertext bounce page
125 * Free a bounce page that was allocated by fscrypt_encrypt_pagecache_blocks(),
126 * or by fscrypt_alloc_bounce_page() directly.
128 void fscrypt_free_bounce_page(struct page *bounce_page)
132 set_page_private(bounce_page, (unsigned long)NULL);
133 ClearPagePrivate(bounce_page);
134 mempool_free(bounce_page, fscrypt_bounce_page_pool);
136 EXPORT_SYMBOL(fscrypt_free_bounce_page);
138 void fscrypt_generate_iv(union fscrypt_iv *iv, u64 lblk_num,
139 const struct fscrypt_info *ci)
141 memset(iv, 0, ci->ci_mode->ivsize);
142 iv->lblk_num = cpu_to_le64(lblk_num);
144 if (fscrypt_is_direct_key_policy(&ci->ci_policy))
145 memcpy(iv->nonce, ci->ci_nonce, FS_KEY_DERIVATION_NONCE_SIZE);
147 if (ci->ci_essiv_tfm != NULL)
148 crypto_cipher_encrypt_one(ci->ci_essiv_tfm, iv->raw, iv->raw);
151 /* Encrypt or decrypt a single filesystem block of file contents */
152 int fscrypt_crypt_block(const struct inode *inode, fscrypt_direction_t rw,
153 u64 lblk_num, struct page *src_page,
154 struct page *dest_page, unsigned int len,
155 unsigned int offs, gfp_t gfp_flags)
158 struct skcipher_request *req = NULL;
159 DECLARE_CRYPTO_WAIT(wait);
160 struct scatterlist dst, src;
161 struct fscrypt_info *ci = inode->i_crypt_info;
162 struct crypto_skcipher *tfm = ci->ci_ctfm;
165 if (WARN_ON_ONCE(len <= 0))
167 if (WARN_ON_ONCE(len % FS_CRYPTO_BLOCK_SIZE != 0))
170 fscrypt_generate_iv(&iv, lblk_num, ci);
172 req = skcipher_request_alloc(tfm, gfp_flags);
176 skcipher_request_set_callback(
177 req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
178 crypto_req_done, &wait);
180 sg_init_table(&dst, 1);
181 sg_set_page(&dst, dest_page, len, offs);
182 sg_init_table(&src, 1);
183 sg_set_page(&src, src_page, len, offs);
184 skcipher_request_set_crypt(req, &src, &dst, len, &iv);
185 if (rw == FS_DECRYPT)
186 res = crypto_wait_req(crypto_skcipher_decrypt(req), &wait);
188 res = crypto_wait_req(crypto_skcipher_encrypt(req), &wait);
189 skcipher_request_free(req);
191 fscrypt_err(inode, "%scryption failed for block %llu: %d",
192 (rw == FS_DECRYPT ? "De" : "En"), lblk_num, res);
199 * fscrypt_encrypt_pagecache_blocks() - Encrypt filesystem blocks from a pagecache page
200 * @page: The locked pagecache page containing the block(s) to encrypt
201 * @len: Total size of the block(s) to encrypt. Must be a nonzero
202 * multiple of the filesystem's block size.
203 * @offs: Byte offset within @page of the first block to encrypt. Must be
204 * a multiple of the filesystem's block size.
205 * @gfp_flags: Memory allocation flags
207 * A new bounce page is allocated, and the specified block(s) are encrypted into
208 * it. In the bounce page, the ciphertext block(s) will be located at the same
209 * offsets at which the plaintext block(s) were located in the source page; any
210 * other parts of the bounce page will be left uninitialized. However, normally
211 * blocksize == PAGE_SIZE and the whole page is encrypted at once.
213 * This is for use by the filesystem's ->writepages() method.
215 * Return: the new encrypted bounce page on success; an ERR_PTR() on failure
217 struct page *fscrypt_encrypt_pagecache_blocks(struct page *page,
223 const struct inode *inode = page->mapping->host;
224 const unsigned int blockbits = inode->i_blkbits;
225 const unsigned int blocksize = 1 << blockbits;
226 struct page *ciphertext_page;
227 u64 lblk_num = ((u64)page->index << (PAGE_SHIFT - blockbits)) +
232 if (WARN_ON_ONCE(!PageLocked(page)))
233 return ERR_PTR(-EINVAL);
235 if (WARN_ON_ONCE(len <= 0 || !IS_ALIGNED(len | offs, blocksize)))
236 return ERR_PTR(-EINVAL);
238 ciphertext_page = fscrypt_alloc_bounce_page(gfp_flags);
239 if (!ciphertext_page)
240 return ERR_PTR(-ENOMEM);
242 for (i = offs; i < offs + len; i += blocksize, lblk_num++) {
243 err = fscrypt_crypt_block(inode, FS_ENCRYPT, lblk_num,
244 page, ciphertext_page,
245 blocksize, i, gfp_flags);
247 fscrypt_free_bounce_page(ciphertext_page);
251 SetPagePrivate(ciphertext_page);
252 set_page_private(ciphertext_page, (unsigned long)page);
253 return ciphertext_page;
255 EXPORT_SYMBOL(fscrypt_encrypt_pagecache_blocks);
258 * fscrypt_encrypt_block_inplace() - Encrypt a filesystem block in-place
259 * @inode: The inode to which this block belongs
260 * @page: The page containing the block to encrypt
261 * @len: Size of block to encrypt. Doesn't need to be a multiple of the
262 * fs block size, but must be a multiple of FS_CRYPTO_BLOCK_SIZE.
263 * @offs: Byte offset within @page at which the block to encrypt begins
264 * @lblk_num: Filesystem logical block number of the block, i.e. the 0-based
265 * number of the block within the file
266 * @gfp_flags: Memory allocation flags
268 * Encrypt a possibly-compressed filesystem block that is located in an
269 * arbitrary page, not necessarily in the original pagecache page. The @inode
270 * and @lblk_num must be specified, as they can't be determined from @page.
272 * Return: 0 on success; -errno on failure
274 int fscrypt_encrypt_block_inplace(const struct inode *inode, struct page *page,
275 unsigned int len, unsigned int offs,
276 u64 lblk_num, gfp_t gfp_flags)
278 return fscrypt_crypt_block(inode, FS_ENCRYPT, lblk_num, page, page,
279 len, offs, gfp_flags);
281 EXPORT_SYMBOL(fscrypt_encrypt_block_inplace);
284 * fscrypt_decrypt_pagecache_blocks() - Decrypt filesystem blocks in a pagecache page
285 * @page: The locked pagecache page containing the block(s) to decrypt
286 * @len: Total size of the block(s) to decrypt. Must be a nonzero
287 * multiple of the filesystem's block size.
288 * @offs: Byte offset within @page of the first block to decrypt. Must be
289 * a multiple of the filesystem's block size.
291 * The specified block(s) are decrypted in-place within the pagecache page,
292 * which must still be locked and not uptodate. Normally, blocksize ==
293 * PAGE_SIZE and the whole page is decrypted at once.
295 * This is for use by the filesystem's ->readpages() method.
297 * Return: 0 on success; -errno on failure
299 int fscrypt_decrypt_pagecache_blocks(struct page *page, unsigned int len,
302 const struct inode *inode = page->mapping->host;
303 const unsigned int blockbits = inode->i_blkbits;
304 const unsigned int blocksize = 1 << blockbits;
305 u64 lblk_num = ((u64)page->index << (PAGE_SHIFT - blockbits)) +
310 if (WARN_ON_ONCE(!PageLocked(page)))
313 if (WARN_ON_ONCE(len <= 0 || !IS_ALIGNED(len | offs, blocksize)))
316 for (i = offs; i < offs + len; i += blocksize, lblk_num++) {
317 err = fscrypt_crypt_block(inode, FS_DECRYPT, lblk_num, page,
318 page, blocksize, i, GFP_NOFS);
324 EXPORT_SYMBOL(fscrypt_decrypt_pagecache_blocks);
327 * fscrypt_decrypt_block_inplace() - Decrypt a filesystem block in-place
328 * @inode: The inode to which this block belongs
329 * @page: The page containing the block to decrypt
330 * @len: Size of block to decrypt. Doesn't need to be a multiple of the
331 * fs block size, but must be a multiple of FS_CRYPTO_BLOCK_SIZE.
332 * @offs: Byte offset within @page at which the block to decrypt begins
333 * @lblk_num: Filesystem logical block number of the block, i.e. the 0-based
334 * number of the block within the file
336 * Decrypt a possibly-compressed filesystem block that is located in an
337 * arbitrary page, not necessarily in the original pagecache page. The @inode
338 * and @lblk_num must be specified, as they can't be determined from @page.
340 * Return: 0 on success; -errno on failure
342 int fscrypt_decrypt_block_inplace(const struct inode *inode, struct page *page,
343 unsigned int len, unsigned int offs,
346 return fscrypt_crypt_block(inode, FS_DECRYPT, lblk_num, page, page,
347 len, offs, GFP_NOFS);
349 EXPORT_SYMBOL(fscrypt_decrypt_block_inplace);
352 * Validate dentries in encrypted directories to make sure we aren't potentially
353 * caching stale dentries after a key has been added.
355 static int fscrypt_d_revalidate(struct dentry *dentry, unsigned int flags)
362 * Plaintext names are always valid, since fscrypt doesn't support
363 * reverting to ciphertext names without evicting the directory's inode
364 * -- which implies eviction of the dentries in the directory.
366 if (!(dentry->d_flags & DCACHE_ENCRYPTED_NAME))
370 * Ciphertext name; valid if the directory's key is still unavailable.
372 * Although fscrypt forbids rename() on ciphertext names, we still must
373 * use dget_parent() here rather than use ->d_parent directly. That's
374 * because a corrupted fs image may contain directory hard links, which
375 * the VFS handles by moving the directory's dentry tree in the dcache
376 * each time ->lookup() finds the directory and it already has a dentry
377 * elsewhere. Thus ->d_parent can be changing, and we must safely grab
378 * a reference to some ->d_parent to prevent it from being freed.
381 if (flags & LOOKUP_RCU)
384 dir = dget_parent(dentry);
385 err = fscrypt_get_encryption_info(d_inode(dir));
386 valid = !fscrypt_has_encryption_key(d_inode(dir));
395 const struct dentry_operations fscrypt_d_ops = {
396 .d_revalidate = fscrypt_d_revalidate,
399 static void fscrypt_destroy(void)
401 struct fscrypt_ctx *pos, *n;
403 list_for_each_entry_safe(pos, n, &fscrypt_free_ctxs, free_list)
404 kmem_cache_free(fscrypt_ctx_cachep, pos);
405 INIT_LIST_HEAD(&fscrypt_free_ctxs);
406 mempool_destroy(fscrypt_bounce_page_pool);
407 fscrypt_bounce_page_pool = NULL;
411 * fscrypt_initialize() - allocate major buffers for fs encryption.
412 * @cop_flags: fscrypt operations flags
414 * We only call this when we start accessing encrypted files, since it
415 * results in memory getting allocated that wouldn't otherwise be used.
417 * Return: Zero on success, non-zero otherwise.
419 int fscrypt_initialize(unsigned int cop_flags)
421 int i, res = -ENOMEM;
423 /* No need to allocate a bounce page pool if this FS won't use it. */
424 if (cop_flags & FS_CFLG_OWN_PAGES)
427 mutex_lock(&fscrypt_init_mutex);
428 if (fscrypt_bounce_page_pool)
429 goto already_initialized;
431 for (i = 0; i < num_prealloc_crypto_ctxs; i++) {
432 struct fscrypt_ctx *ctx;
434 ctx = kmem_cache_zalloc(fscrypt_ctx_cachep, GFP_NOFS);
437 list_add(&ctx->free_list, &fscrypt_free_ctxs);
440 fscrypt_bounce_page_pool =
441 mempool_create_page_pool(num_prealloc_crypto_pages, 0);
442 if (!fscrypt_bounce_page_pool)
446 mutex_unlock(&fscrypt_init_mutex);
450 mutex_unlock(&fscrypt_init_mutex);
454 void fscrypt_msg(const struct inode *inode, const char *level,
455 const char *fmt, ...)
457 static DEFINE_RATELIMIT_STATE(rs, DEFAULT_RATELIMIT_INTERVAL,
458 DEFAULT_RATELIMIT_BURST);
459 struct va_format vaf;
462 if (!__ratelimit(&rs))
469 printk("%sfscrypt (%s, inode %lu): %pV\n",
470 level, inode->i_sb->s_id, inode->i_ino, &vaf);
472 printk("%sfscrypt: %pV\n", level, &vaf);
477 * fscrypt_init() - Set up for fs encryption.
479 static int __init fscrypt_init(void)
484 * Use an unbound workqueue to allow bios to be decrypted in parallel
485 * even when they happen to complete on the same CPU. This sacrifices
486 * locality, but it's worthwhile since decryption is CPU-intensive.
488 * Also use a high-priority workqueue to prioritize decryption work,
489 * which blocks reads from completing, over regular application tasks.
491 fscrypt_read_workqueue = alloc_workqueue("fscrypt_read_queue",
492 WQ_UNBOUND | WQ_HIGHPRI,
494 if (!fscrypt_read_workqueue)
497 fscrypt_ctx_cachep = KMEM_CACHE(fscrypt_ctx, SLAB_RECLAIM_ACCOUNT);
498 if (!fscrypt_ctx_cachep)
499 goto fail_free_queue;
501 fscrypt_info_cachep = KMEM_CACHE(fscrypt_info, SLAB_RECLAIM_ACCOUNT);
502 if (!fscrypt_info_cachep)
505 err = fscrypt_init_keyring();
512 kmem_cache_destroy(fscrypt_info_cachep);
514 kmem_cache_destroy(fscrypt_ctx_cachep);
516 destroy_workqueue(fscrypt_read_workqueue);
520 late_initcall(fscrypt_init)