1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Opening fs-verity files
4 *
5 * Copyright 2019 Google LLC
6 */
7
8#include "fsverity_private.h"
9
10#include <linux/mm.h>
11#include <linux/slab.h>
12
13static struct kmem_cache *fsverity_info_cachep;
14
15/**
16 * fsverity_init_merkle_tree_params() - initialize Merkle tree parameters
17 * @params: the parameters struct to initialize
18 * @inode: the inode for which the Merkle tree is being built
19 * @hash_algorithm: number of hash algorithm to use
20 * @log_blocksize: log base 2 of block size to use
21 * @salt: pointer to salt (optional)
22 * @salt_size: size of salt, possibly 0
23 *
24 * Validate the hash algorithm and block size, then compute the tree topology
25 * (num levels, num blocks in each level, etc.) and initialize @params.
26 *
27 * Return: 0 on success, -errno on failure
28 */
29int fsverity_init_merkle_tree_params(struct merkle_tree_params *params,
30 const struct inode *inode,
31 unsigned int hash_algorithm,
32 unsigned int log_blocksize,
33 const u8 *salt, size_t salt_size)
34{
35 const struct fsverity_hash_alg *hash_alg;
36 int err;
37 u64 blocks;
38 u64 blocks_in_level[FS_VERITY_MAX_LEVELS];
39 u64 offset;
40 int level;
41
42 memset(params, 0, sizeof(*params));
43
44 hash_alg = fsverity_get_hash_alg(inode, num: hash_algorithm);
45 if (IS_ERR(ptr: hash_alg))
46 return PTR_ERR(ptr: hash_alg);
47 params->hash_alg = hash_alg;
48 params->digest_size = hash_alg->digest_size;
49
50 params->hashstate = fsverity_prepare_hash_state(alg: hash_alg, salt,
51 salt_size);
52 if (IS_ERR(ptr: params->hashstate)) {
53 err = PTR_ERR(ptr: params->hashstate);
54 params->hashstate = NULL;
55 fsverity_err(inode, "Error %d preparing hash state", err);
56 goto out_err;
57 }
58
59 /*
60 * fs/verity/ directly assumes that the Merkle tree block size is a
61 * power of 2 less than or equal to PAGE_SIZE. Another restriction
62 * arises from the interaction between fs/verity/ and the filesystems
63 * themselves: filesystems expect to be able to verify a single
64 * filesystem block of data at a time. Therefore, the Merkle tree block
65 * size must also be less than or equal to the filesystem block size.
66 *
67 * The above are the only hard limitations, so in theory the Merkle tree
68 * block size could be as small as twice the digest size. However,
69 * that's not useful, and it would result in some unusually deep and
70 * large Merkle trees. So we currently require that the Merkle tree
71 * block size be at least 1024 bytes. That's small enough to test the
72 * sub-page block case on systems with 4K pages, but not too small.
73 */
74 if (log_blocksize < 10 || log_blocksize > PAGE_SHIFT ||
75 log_blocksize > inode->i_blkbits) {
76 fsverity_warn(inode, "Unsupported log_blocksize: %u",
77 log_blocksize);
78 err = -EINVAL;
79 goto out_err;
80 }
81 params->log_blocksize = log_blocksize;
82 params->block_size = 1 << log_blocksize;
83 params->log_blocks_per_page = PAGE_SHIFT - log_blocksize;
84 params->blocks_per_page = 1 << params->log_blocks_per_page;
85
86 if (WARN_ON_ONCE(!is_power_of_2(params->digest_size))) {
87 err = -EINVAL;
88 goto out_err;
89 }
90 if (params->block_size < 2 * params->digest_size) {
91 fsverity_warn(inode,
92 "Merkle tree block size (%u) too small for hash algorithm \"%s\"",
93 params->block_size, hash_alg->name);
94 err = -EINVAL;
95 goto out_err;
96 }
97 params->log_digestsize = ilog2(params->digest_size);
98 params->log_arity = log_blocksize - params->log_digestsize;
99 params->hashes_per_block = 1 << params->log_arity;
100
101 /*
102 * Compute the number of levels in the Merkle tree and create a map from
103 * level to the starting block of that level. Level 'num_levels - 1' is
104 * the root and is stored first. Level 0 is the level directly "above"
105 * the data blocks and is stored last.
106 */
107
108 /* Compute number of levels and the number of blocks in each level */
109 blocks = ((u64)inode->i_size + params->block_size - 1) >> log_blocksize;
110 while (blocks > 1) {
111 if (params->num_levels >= FS_VERITY_MAX_LEVELS) {
112 fsverity_err(inode, "Too many levels in Merkle tree");
113 err = -EFBIG;
114 goto out_err;
115 }
116 blocks = (blocks + params->hashes_per_block - 1) >>
117 params->log_arity;
118 blocks_in_level[params->num_levels++] = blocks;
119 }
120
121 /* Compute the starting block of each level */
122 offset = 0;
123 for (level = (int)params->num_levels - 1; level >= 0; level--) {
124 params->level_start[level] = offset;
125 offset += blocks_in_level[level];
126 }
127
128 /*
129 * With block_size != PAGE_SIZE, an in-memory bitmap will need to be
130 * allocated to track the "verified" status of hash blocks. Don't allow
131 * this bitmap to get too large. For now, limit it to 1 MiB, which
132 * limits the file size to about 4.4 TB with SHA-256 and 4K blocks.
133 *
134 * Together with the fact that the data, and thus also the Merkle tree,
135 * cannot have more than ULONG_MAX pages, this implies that hash block
136 * indices can always fit in an 'unsigned long'. But to be safe, we
137 * explicitly check for that too. Note, this is only for hash block
138 * indices; data block indices might not fit in an 'unsigned long'.
139 */
140 if ((params->block_size != PAGE_SIZE && offset > 1 << 23) ||
141 offset > ULONG_MAX) {
142 fsverity_err(inode, "Too many blocks in Merkle tree");
143 err = -EFBIG;
144 goto out_err;
145 }
146
147 params->tree_size = offset << log_blocksize;
148 params->tree_pages = PAGE_ALIGN(params->tree_size) >> PAGE_SHIFT;
149 return 0;
150
151out_err:
152 kfree(objp: params->hashstate);
153 memset(params, 0, sizeof(*params));
154 return err;
155}
156
157/*
158 * Compute the file digest by hashing the fsverity_descriptor excluding the
159 * builtin signature and with the sig_size field set to 0.
160 */
161static int compute_file_digest(const struct fsverity_hash_alg *hash_alg,
162 struct fsverity_descriptor *desc,
163 u8 *file_digest)
164{
165 __le32 sig_size = desc->sig_size;
166 int err;
167
168 desc->sig_size = 0;
169 err = fsverity_hash_buffer(alg: hash_alg, data: desc, size: sizeof(*desc), out: file_digest);
170 desc->sig_size = sig_size;
171
172 return err;
173}
174
175/*
176 * Create a new fsverity_info from the given fsverity_descriptor (with optional
177 * appended builtin signature), and check the signature if present. The
178 * fsverity_descriptor must have already undergone basic validation.
179 */
180struct fsverity_info *fsverity_create_info(const struct inode *inode,
181 struct fsverity_descriptor *desc)
182{
183 struct fsverity_info *vi;
184 int err;
185
186 vi = kmem_cache_zalloc(k: fsverity_info_cachep, GFP_KERNEL);
187 if (!vi)
188 return ERR_PTR(error: -ENOMEM);
189 vi->inode = inode;
190
191 err = fsverity_init_merkle_tree_params(params: &vi->tree_params, inode,
192 hash_algorithm: desc->hash_algorithm,
193 log_blocksize: desc->log_blocksize,
194 salt: desc->salt, salt_size: desc->salt_size);
195 if (err) {
196 fsverity_err(inode,
197 "Error %d initializing Merkle tree parameters",
198 err);
199 goto fail;
200 }
201
202 memcpy(vi->root_hash, desc->root_hash, vi->tree_params.digest_size);
203
204 err = compute_file_digest(hash_alg: vi->tree_params.hash_alg, desc,
205 file_digest: vi->file_digest);
206 if (err) {
207 fsverity_err(inode, "Error %d computing file digest", err);
208 goto fail;
209 }
210
211 err = fsverity_verify_signature(vi, signature: desc->signature,
212 le32_to_cpu(desc->sig_size));
213 if (err)
214 goto fail;
215
216 if (vi->tree_params.block_size != PAGE_SIZE) {
217 /*
218 * When the Merkle tree block size and page size differ, we use
219 * a bitmap to keep track of which hash blocks have been
220 * verified. This bitmap must contain one bit per hash block,
221 * including alignment to a page boundary at the end.
222 *
223 * Eventually, to support extremely large files in an efficient
224 * way, it might be necessary to make pages of this bitmap
225 * reclaimable. But for now, simply allocating the whole bitmap
226 * is a simple solution that works well on the files on which
227 * fsverity is realistically used. E.g., with SHA-256 and 4K
228 * blocks, a 100MB file only needs a 24-byte bitmap, and the
229 * bitmap for any file under 17GB fits in a 4K page.
230 */
231 unsigned long num_bits =
232 vi->tree_params.tree_pages <<
233 vi->tree_params.log_blocks_per_page;
234
235 vi->hash_block_verified = kvcalloc(BITS_TO_LONGS(num_bits),
236 size: sizeof(unsigned long),
237 GFP_KERNEL);
238 if (!vi->hash_block_verified) {
239 err = -ENOMEM;
240 goto fail;
241 }
242 spin_lock_init(&vi->hash_page_init_lock);
243 }
244
245 return vi;
246
247fail:
248 fsverity_free_info(vi);
249 return ERR_PTR(error: err);
250}
251
252void fsverity_set_info(struct inode *inode, struct fsverity_info *vi)
253{
254 /*
255 * Multiple tasks may race to set ->i_verity_info, so use
256 * cmpxchg_release(). This pairs with the smp_load_acquire() in
257 * fsverity_get_info(). I.e., here we publish ->i_verity_info with a
258 * RELEASE barrier so that other tasks can ACQUIRE it.
259 */
260 if (cmpxchg_release(&inode->i_verity_info, NULL, vi) != NULL) {
261 /* Lost the race, so free the fsverity_info we allocated. */
262 fsverity_free_info(vi);
263 /*
264 * Afterwards, the caller may access ->i_verity_info directly,
265 * so make sure to ACQUIRE the winning fsverity_info.
266 */
267 (void)fsverity_get_info(inode);
268 }
269}
270
271void fsverity_free_info(struct fsverity_info *vi)
272{
273 if (!vi)
274 return;
275 kfree(objp: vi->tree_params.hashstate);
276 kvfree(addr: vi->hash_block_verified);
277 kmem_cache_free(s: fsverity_info_cachep, objp: vi);
278}
279
280static bool validate_fsverity_descriptor(struct inode *inode,
281 const struct fsverity_descriptor *desc,
282 size_t desc_size)
283{
284 if (desc_size < sizeof(*desc)) {
285 fsverity_err(inode, "Unrecognized descriptor size: %zu bytes",
286 desc_size);
287 return false;
288 }
289
290 if (desc->version != 1) {
291 fsverity_err(inode, "Unrecognized descriptor version: %u",
292 desc->version);
293 return false;
294 }
295
296 if (memchr_inv(p: desc->__reserved, c: 0, size: sizeof(desc->__reserved))) {
297 fsverity_err(inode, "Reserved bits set in descriptor");
298 return false;
299 }
300
301 if (desc->salt_size > sizeof(desc->salt)) {
302 fsverity_err(inode, "Invalid salt_size: %u", desc->salt_size);
303 return false;
304 }
305
306 if (le64_to_cpu(desc->data_size) != inode->i_size) {
307 fsverity_err(inode,
308 "Wrong data_size: %llu (desc) != %lld (inode)",
309 le64_to_cpu(desc->data_size), inode->i_size);
310 return false;
311 }
312
313 if (le32_to_cpu(desc->sig_size) > desc_size - sizeof(*desc)) {
314 fsverity_err(inode, "Signature overflows verity descriptor");
315 return false;
316 }
317
318 return true;
319}
320
321/*
322 * Read the inode's fsverity_descriptor (with optional appended builtin
323 * signature) from the filesystem, and do basic validation of it.
324 */
325int fsverity_get_descriptor(struct inode *inode,
326 struct fsverity_descriptor **desc_ret)
327{
328 int res;
329 struct fsverity_descriptor *desc;
330
331 res = inode->i_sb->s_vop->get_verity_descriptor(inode, NULL, 0);
332 if (res < 0) {
333 fsverity_err(inode,
334 "Error %d getting verity descriptor size", res);
335 return res;
336 }
337 if (res > FS_VERITY_MAX_DESCRIPTOR_SIZE) {
338 fsverity_err(inode, "Verity descriptor is too large (%d bytes)",
339 res);
340 return -EMSGSIZE;
341 }
342 desc = kmalloc(size: res, GFP_KERNEL);
343 if (!desc)
344 return -ENOMEM;
345 res = inode->i_sb->s_vop->get_verity_descriptor(inode, desc, res);
346 if (res < 0) {
347 fsverity_err(inode, "Error %d reading verity descriptor", res);
348 kfree(objp: desc);
349 return res;
350 }
351
352 if (!validate_fsverity_descriptor(inode, desc, desc_size: res)) {
353 kfree(objp: desc);
354 return -EINVAL;
355 }
356
357 *desc_ret = desc;
358 return 0;
359}
360
361/* Ensure the inode has an ->i_verity_info */
362static int ensure_verity_info(struct inode *inode)
363{
364 struct fsverity_info *vi = fsverity_get_info(inode);
365 struct fsverity_descriptor *desc;
366 int err;
367
368 if (vi)
369 return 0;
370
371 err = fsverity_get_descriptor(inode, desc_ret: &desc);
372 if (err)
373 return err;
374
375 vi = fsverity_create_info(inode, desc);
376 if (IS_ERR(ptr: vi)) {
377 err = PTR_ERR(ptr: vi);
378 goto out_free_desc;
379 }
380
381 fsverity_set_info(inode, vi);
382 err = 0;
383out_free_desc:
384 kfree(objp: desc);
385 return err;
386}
387
388int __fsverity_file_open(struct inode *inode, struct file *filp)
389{
390 if (filp->f_mode & FMODE_WRITE)
391 return -EPERM;
392 return ensure_verity_info(inode);
393}
394EXPORT_SYMBOL_GPL(__fsverity_file_open);
395
396int __fsverity_prepare_setattr(struct dentry *dentry, struct iattr *attr)
397{
398 if (attr->ia_valid & ATTR_SIZE)
399 return -EPERM;
400 return 0;
401}
402EXPORT_SYMBOL_GPL(__fsverity_prepare_setattr);
403
404void __fsverity_cleanup_inode(struct inode *inode)
405{
406 fsverity_free_info(vi: inode->i_verity_info);
407 inode->i_verity_info = NULL;
408}
409EXPORT_SYMBOL_GPL(__fsverity_cleanup_inode);
410
411void __init fsverity_init_info_cache(void)
412{
413 fsverity_info_cachep = KMEM_CACHE_USERCOPY(
414 fsverity_info,
415 SLAB_RECLAIM_ACCOUNT | SLAB_PANIC,
416 file_digest);
417}
418

source code of linux/fs/verity/open.c