open.c

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