• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Ioctl to enable verity on a file
4  *
5  * Copyright 2019 Google LLC
6  */
7 
8 #include "fsverity_private.h"
9 
10 #include <linux/mount.h>
11 #include <linux/pagemap.h>
12 #include <linux/slab.h>
13 #include <linux/sched/signal.h>
14 #include <linux/uaccess.h>
15 
16 struct block_buffer {
17 	u32 filled;
18 	bool is_root_hash;
19 	u8 *data;
20 };
21 
22 /* Hash a block, writing the result to the next level's pending block buffer. */
hash_one_block(struct inode * inode,const struct merkle_tree_params * params,struct ahash_request * req,struct block_buffer * cur)23 static int hash_one_block(struct inode *inode,
24 			  const struct merkle_tree_params *params,
25 			  struct ahash_request *req, struct block_buffer *cur)
26 {
27 	struct block_buffer *next = cur + 1;
28 	int err;
29 
30 	/*
31 	 * Safety check to prevent a buffer overflow in case of a filesystem bug
32 	 * that allows the file size to change despite deny_write_access(), or a
33 	 * bug in the Merkle tree logic itself
34 	 */
35 	if (WARN_ON_ONCE(next->is_root_hash && next->filled != 0))
36 		return -EINVAL;
37 
38 	/* Zero-pad the block if it's shorter than the block size. */
39 	memset(&cur->data[cur->filled], 0, params->block_size - cur->filled);
40 
41 	err = fsverity_hash_block(params, inode, req, virt_to_page(cur->data),
42 				  offset_in_page(cur->data),
43 				  &next->data[next->filled]);
44 	if (err)
45 		return err;
46 	next->filled += params->digest_size;
47 	cur->filled = 0;
48 	return 0;
49 }
50 
write_merkle_tree_block(struct inode * inode,const u8 * buf,unsigned long index,const struct merkle_tree_params * params)51 static int write_merkle_tree_block(struct inode *inode, const u8 *buf,
52 				   unsigned long index,
53 				   const struct merkle_tree_params *params)
54 {
55 	u64 pos = (u64)index << params->log_blocksize;
56 	int err;
57 
58 	err = inode->i_sb->s_vop->write_merkle_tree_block(inode, buf, pos,
59 							  params->block_size);
60 	if (err)
61 		fsverity_err(inode, "Error %d writing Merkle tree block %lu",
62 			     err, index);
63 	return err;
64 }
65 
66 /*
67  * Build the Merkle tree for the given file using the given parameters, and
68  * return the root hash in @root_hash.
69  *
70  * The tree is written to a filesystem-specific location as determined by the
71  * ->write_merkle_tree_block() method.  However, the blocks that comprise the
72  * tree are the same for all filesystems.
73  */
build_merkle_tree(struct file * filp,const struct merkle_tree_params * params,u8 * root_hash)74 static int build_merkle_tree(struct file *filp,
75 			     const struct merkle_tree_params *params,
76 			     u8 *root_hash)
77 {
78 	struct inode *inode = file_inode(filp);
79 	const u64 data_size = inode->i_size;
80 	const int num_levels = params->num_levels;
81 	struct ahash_request *req;
82 	struct block_buffer _buffers[1 + FS_VERITY_MAX_LEVELS + 1] = {};
83 	struct block_buffer *buffers = &_buffers[1];
84 	unsigned long level_offset[FS_VERITY_MAX_LEVELS];
85 	int level;
86 	u64 offset;
87 	int err;
88 
89 	if (data_size == 0) {
90 		/* Empty file is a special case; root hash is all 0's */
91 		memset(root_hash, 0, params->digest_size);
92 		return 0;
93 	}
94 
95 	/* This allocation never fails, since it's mempool-backed. */
96 	req = fsverity_alloc_hash_request(params->hash_alg, GFP_KERNEL);
97 
98 	/*
99 	 * Allocate the block buffers.  Buffer "-1" is for data blocks.
100 	 * Buffers 0 <= level < num_levels are for the actual tree levels.
101 	 * Buffer 'num_levels' is for the root hash.
102 	 */
103 	for (level = -1; level < num_levels; level++) {
104 		buffers[level].data = kzalloc(params->block_size, GFP_KERNEL);
105 		if (!buffers[level].data) {
106 			err = -ENOMEM;
107 			goto out;
108 		}
109 	}
110 	buffers[num_levels].data = root_hash;
111 	buffers[num_levels].is_root_hash = true;
112 
113 	BUILD_BUG_ON(sizeof(level_offset) != sizeof(params->level_start));
114 	memcpy(level_offset, params->level_start, sizeof(level_offset));
115 
116 	/* Hash each data block, also hashing the tree blocks as they fill up */
117 	for (offset = 0; offset < data_size; offset += params->block_size) {
118 		ssize_t bytes_read;
119 		loff_t pos = offset;
120 
121 		buffers[-1].filled = min_t(u64, params->block_size,
122 					   data_size - offset);
123 		bytes_read = __kernel_read(filp, buffers[-1].data,
124 					   buffers[-1].filled, &pos);
125 		if (bytes_read < 0) {
126 			err = bytes_read;
127 			fsverity_err(inode, "Error %d reading file data", err);
128 			goto out;
129 		}
130 		if (bytes_read != buffers[-1].filled) {
131 			err = -EINVAL;
132 			fsverity_err(inode, "Short read of file data");
133 			goto out;
134 		}
135 		err = hash_one_block(inode, params, req, &buffers[-1]);
136 		if (err)
137 			goto out;
138 		for (level = 0; level < num_levels; level++) {
139 			if (buffers[level].filled + params->digest_size <=
140 			    params->block_size) {
141 				/* Next block at @level isn't full yet */
142 				break;
143 			}
144 			/* Next block at @level is full */
145 
146 			err = hash_one_block(inode, params, req,
147 					     &buffers[level]);
148 			if (err)
149 				goto out;
150 			err = write_merkle_tree_block(inode,
151 						      buffers[level].data,
152 						      level_offset[level],
153 						      params);
154 			if (err)
155 				goto out;
156 			level_offset[level]++;
157 		}
158 		if (fatal_signal_pending(current)) {
159 			err = -EINTR;
160 			goto out;
161 		}
162 		cond_resched();
163 	}
164 	/* Finish all nonempty pending tree blocks. */
165 	for (level = 0; level < num_levels; level++) {
166 		if (buffers[level].filled != 0) {
167 			err = hash_one_block(inode, params, req,
168 					     &buffers[level]);
169 			if (err)
170 				goto out;
171 			err = write_merkle_tree_block(inode,
172 						      buffers[level].data,
173 						      level_offset[level],
174 						      params);
175 			if (err)
176 				goto out;
177 		}
178 	}
179 	/* The root hash was filled by the last call to hash_one_block(). */
180 	if (WARN_ON(buffers[num_levels].filled != params->digest_size)) {
181 		err = -EINVAL;
182 		goto out;
183 	}
184 	err = 0;
185 out:
186 	for (level = -1; level < num_levels; level++)
187 		kfree(buffers[level].data);
188 	fsverity_free_hash_request(params->hash_alg, req);
189 	return err;
190 }
191 
enable_verity(struct file * filp,const struct fsverity_enable_arg * arg)192 static int enable_verity(struct file *filp,
193 			 const struct fsverity_enable_arg *arg)
194 {
195 	struct inode *inode = file_inode(filp);
196 	const struct fsverity_operations *vops = inode->i_sb->s_vop;
197 	struct merkle_tree_params params = { };
198 	struct fsverity_descriptor *desc;
199 	size_t desc_size = struct_size(desc, signature, arg->sig_size);
200 	struct fsverity_info *vi;
201 	int err;
202 
203 	/* Start initializing the fsverity_descriptor */
204 	desc = kzalloc(desc_size, GFP_KERNEL);
205 	if (!desc)
206 		return -ENOMEM;
207 	desc->version = 1;
208 	desc->hash_algorithm = arg->hash_algorithm;
209 	desc->log_blocksize = ilog2(arg->block_size);
210 
211 	/* Get the salt if the user provided one */
212 	if (arg->salt_size &&
213 	    copy_from_user(desc->salt, u64_to_user_ptr(arg->salt_ptr),
214 			   arg->salt_size)) {
215 		err = -EFAULT;
216 		goto out;
217 	}
218 	desc->salt_size = arg->salt_size;
219 
220 	/* Get the signature if the user provided one */
221 	if (arg->sig_size &&
222 	    copy_from_user(desc->signature, u64_to_user_ptr(arg->sig_ptr),
223 			   arg->sig_size)) {
224 		err = -EFAULT;
225 		goto out;
226 	}
227 	desc->sig_size = cpu_to_le32(arg->sig_size);
228 
229 	desc->data_size = cpu_to_le64(inode->i_size);
230 
231 	/* Prepare the Merkle tree parameters */
232 	err = fsverity_init_merkle_tree_params(&params, inode,
233 					       arg->hash_algorithm,
234 					       desc->log_blocksize,
235 					       desc->salt, desc->salt_size);
236 	if (err)
237 		goto out;
238 
239 	/*
240 	 * Start enabling verity on this file, serialized by the inode lock.
241 	 * Fail if verity is already enabled or is already being enabled.
242 	 */
243 	inode_lock(inode);
244 	if (IS_VERITY(inode))
245 		err = -EEXIST;
246 	else
247 		err = vops->begin_enable_verity(filp);
248 	inode_unlock(inode);
249 	if (err)
250 		goto out;
251 
252 	/*
253 	 * Build the Merkle tree.  Don't hold the inode lock during this, since
254 	 * on huge files this may take a very long time and we don't want to
255 	 * force unrelated syscalls like chown() to block forever.  We don't
256 	 * need the inode lock here because deny_write_access() already prevents
257 	 * the file from being written to or truncated, and we still serialize
258 	 * ->begin_enable_verity() and ->end_enable_verity() using the inode
259 	 * lock and only allow one process to be here at a time on a given file.
260 	 */
261 	BUILD_BUG_ON(sizeof(desc->root_hash) < FS_VERITY_MAX_DIGEST_SIZE);
262 	err = build_merkle_tree(filp, &params, desc->root_hash);
263 	if (err) {
264 		fsverity_err(inode, "Error %d building Merkle tree", err);
265 		goto rollback;
266 	}
267 
268 	/*
269 	 * Create the fsverity_info.  Don't bother trying to save work by
270 	 * reusing the merkle_tree_params from above.  Instead, just create the
271 	 * fsverity_info from the fsverity_descriptor as if it were just loaded
272 	 * from disk.  This is simpler, and it serves as an extra check that the
273 	 * metadata we're writing is valid before actually enabling verity.
274 	 */
275 	vi = fsverity_create_info(inode, desc);
276 	if (IS_ERR(vi)) {
277 		err = PTR_ERR(vi);
278 		goto rollback;
279 	}
280 
281 	/*
282 	 * Tell the filesystem to finish enabling verity on the file.
283 	 * Serialized with ->begin_enable_verity() by the inode lock.
284 	 */
285 	inode_lock(inode);
286 	err = vops->end_enable_verity(filp, desc, desc_size, params.tree_size);
287 	inode_unlock(inode);
288 	if (err) {
289 		fsverity_err(inode, "%ps() failed with err %d",
290 			     vops->end_enable_verity, err);
291 		fsverity_free_info(vi);
292 	} else if (WARN_ON(!IS_VERITY(inode))) {
293 		err = -EINVAL;
294 		fsverity_free_info(vi);
295 	} else {
296 		/* Successfully enabled verity */
297 
298 		/*
299 		 * Readers can start using ->i_verity_info immediately, so it
300 		 * can't be rolled back once set.  So don't set it until just
301 		 * after the filesystem has successfully enabled verity.
302 		 */
303 		fsverity_set_info(inode, vi);
304 	}
305 out:
306 	kfree(params.hashstate);
307 	kfree(desc);
308 	return err;
309 
310 rollback:
311 	inode_lock(inode);
312 	(void)vops->end_enable_verity(filp, NULL, 0, params.tree_size);
313 	inode_unlock(inode);
314 	goto out;
315 }
316 
317 /**
318  * fsverity_ioctl_enable() - enable verity on a file
319  * @filp: file to enable verity on
320  * @uarg: user pointer to fsverity_enable_arg
321  *
322  * Enable fs-verity on a file.  See the "FS_IOC_ENABLE_VERITY" section of
323  * Documentation/filesystems/fsverity.rst for the documentation.
324  *
325  * Return: 0 on success, -errno on failure
326  */
fsverity_ioctl_enable(struct file * filp,const void __user * uarg)327 int fsverity_ioctl_enable(struct file *filp, const void __user *uarg)
328 {
329 	struct inode *inode = file_inode(filp);
330 	struct fsverity_enable_arg arg;
331 	int err;
332 
333 	if (copy_from_user(&arg, uarg, sizeof(arg)))
334 		return -EFAULT;
335 
336 	if (arg.version != 1)
337 		return -EINVAL;
338 
339 	if (arg.__reserved1 ||
340 	    memchr_inv(arg.__reserved2, 0, sizeof(arg.__reserved2)))
341 		return -EINVAL;
342 
343 	if (!is_power_of_2(arg.block_size))
344 		return -EINVAL;
345 
346 	if (arg.salt_size > sizeof_field(struct fsverity_descriptor, salt))
347 		return -EMSGSIZE;
348 
349 	if (arg.sig_size > FS_VERITY_MAX_SIGNATURE_SIZE)
350 		return -EMSGSIZE;
351 
352 	/*
353 	 * Require a regular file with write access.  But the actual fd must
354 	 * still be readonly so that we can lock out all writers.  This is
355 	 * needed to guarantee that no writable fds exist to the file once it
356 	 * has verity enabled, and to stabilize the data being hashed.
357 	 */
358 
359 	err = file_permission(filp, MAY_WRITE);
360 	if (err)
361 		return err;
362 	/*
363 	 * __kernel_read() is used while building the Merkle tree.  So, we can't
364 	 * allow file descriptors that were opened for ioctl access only, using
365 	 * the special nonstandard access mode 3.  O_RDONLY only, please!
366 	 */
367 	if (!(filp->f_mode & FMODE_READ))
368 		return -EBADF;
369 
370 	if (IS_APPEND(inode))
371 		return -EPERM;
372 
373 	if (S_ISDIR(inode->i_mode))
374 		return -EISDIR;
375 
376 	if (!S_ISREG(inode->i_mode))
377 		return -EINVAL;
378 
379 	err = mnt_want_write_file(filp);
380 	if (err) /* -EROFS */
381 		return err;
382 
383 	err = deny_write_access(filp);
384 	if (err) /* -ETXTBSY */
385 		goto out_drop_write;
386 
387 	err = enable_verity(filp, &arg);
388 
389 	/*
390 	 * We no longer drop the inode's pagecache after enabling verity.  This
391 	 * used to be done to try to avoid a race condition where pages could be
392 	 * evicted after being used in the Merkle tree construction, then
393 	 * re-instantiated by a concurrent read.  Such pages are unverified, and
394 	 * the backing storage could have filled them with different content, so
395 	 * they shouldn't be used to fulfill reads once verity is enabled.
396 	 *
397 	 * But, dropping the pagecache has a big performance impact, and it
398 	 * doesn't fully solve the race condition anyway.  So for those reasons,
399 	 * and also because this race condition isn't very important relatively
400 	 * speaking (especially for small-ish files, where the chance of a page
401 	 * being used, evicted, *and* re-instantiated all while enabling verity
402 	 * is quite small), we no longer drop the inode's pagecache.
403 	 */
404 
405 	/*
406 	 * allow_write_access() is needed to pair with deny_write_access().
407 	 * Regardless, the filesystem won't allow writing to verity files.
408 	 */
409 	allow_write_access(filp);
410 out_drop_write:
411 	mnt_drop_write_file(filp);
412 	return err;
413 }
414 EXPORT_SYMBOL_GPL(fsverity_ioctl_enable);
415