]>
Commit | Line | Data |
---|---|---|
1 | // SPDX-License-Identifier: GPL-2.0 | |
2 | /* | |
3 | * fs/ext4/verity.c: fs-verity support for ext4 | |
4 | * | |
5 | * Copyright 2019 Google LLC | |
6 | */ | |
7 | ||
8 | /* | |
9 | * Implementation of fsverity_operations for ext4. | |
10 | * | |
11 | * ext4 stores the verity metadata (Merkle tree and fsverity_descriptor) past | |
12 | * the end of the file, starting at the first 64K boundary beyond i_size. This | |
13 | * approach works because (a) verity files are readonly, and (b) pages fully | |
14 | * beyond i_size aren't visible to userspace but can be read/written internally | |
15 | * by ext4 with only some relatively small changes to ext4. This approach | |
16 | * avoids having to depend on the EA_INODE feature and on rearchitecturing | |
17 | * ext4's xattr support to support paging multi-gigabyte xattrs into memory, and | |
18 | * to support encrypting xattrs. Note that the verity metadata *must* be | |
19 | * encrypted when the file is, since it contains hashes of the plaintext data. | |
20 | * | |
21 | * Using a 64K boundary rather than a 4K one keeps things ready for | |
22 | * architectures with 64K pages, and it doesn't necessarily waste space on-disk | |
23 | * since there can be a hole between i_size and the start of the Merkle tree. | |
24 | */ | |
25 | ||
26 | #include <linux/quotaops.h> | |
27 | ||
28 | #include "ext4.h" | |
29 | #include "ext4_extents.h" | |
30 | #include "ext4_jbd2.h" | |
31 | ||
32 | static inline loff_t ext4_verity_metadata_pos(const struct inode *inode) | |
33 | { | |
34 | return round_up(inode->i_size, 65536); | |
35 | } | |
36 | ||
37 | /* | |
38 | * Read some verity metadata from the inode. __vfs_read() can't be used because | |
39 | * we need to read beyond i_size. | |
40 | */ | |
41 | static int pagecache_read(struct inode *inode, void *buf, size_t count, | |
42 | loff_t pos) | |
43 | { | |
44 | while (count) { | |
45 | size_t n = min_t(size_t, count, | |
46 | PAGE_SIZE - offset_in_page(pos)); | |
47 | struct page *page; | |
48 | void *addr; | |
49 | ||
50 | page = read_mapping_page(inode->i_mapping, pos >> PAGE_SHIFT, | |
51 | NULL); | |
52 | if (IS_ERR(page)) | |
53 | return PTR_ERR(page); | |
54 | ||
55 | addr = kmap_atomic(page); | |
56 | memcpy(buf, addr + offset_in_page(pos), n); | |
57 | kunmap_atomic(addr); | |
58 | ||
59 | put_page(page); | |
60 | ||
61 | buf += n; | |
62 | pos += n; | |
63 | count -= n; | |
64 | } | |
65 | return 0; | |
66 | } | |
67 | ||
68 | /* | |
69 | * Write some verity metadata to the inode for FS_IOC_ENABLE_VERITY. | |
70 | * kernel_write() can't be used because the file descriptor is readonly. | |
71 | */ | |
72 | static int pagecache_write(struct inode *inode, const void *buf, size_t count, | |
73 | loff_t pos) | |
74 | { | |
75 | if (pos + count > inode->i_sb->s_maxbytes) | |
76 | return -EFBIG; | |
77 | ||
78 | while (count) { | |
79 | size_t n = min_t(size_t, count, | |
80 | PAGE_SIZE - offset_in_page(pos)); | |
81 | struct page *page; | |
82 | void *fsdata; | |
83 | void *addr; | |
84 | int res; | |
85 | ||
86 | res = pagecache_write_begin(NULL, inode->i_mapping, pos, n, 0, | |
87 | &page, &fsdata); | |
88 | if (res) | |
89 | return res; | |
90 | ||
91 | addr = kmap_atomic(page); | |
92 | memcpy(addr + offset_in_page(pos), buf, n); | |
93 | kunmap_atomic(addr); | |
94 | ||
95 | res = pagecache_write_end(NULL, inode->i_mapping, pos, n, n, | |
96 | page, fsdata); | |
97 | if (res < 0) | |
98 | return res; | |
99 | if (res != n) | |
100 | return -EIO; | |
101 | ||
102 | buf += n; | |
103 | pos += n; | |
104 | count -= n; | |
105 | } | |
106 | return 0; | |
107 | } | |
108 | ||
109 | static int ext4_begin_enable_verity(struct file *filp) | |
110 | { | |
111 | struct inode *inode = file_inode(filp); | |
112 | const int credits = 2; /* superblock and inode for ext4_orphan_add() */ | |
113 | handle_t *handle; | |
114 | int err; | |
115 | ||
116 | if (ext4_verity_in_progress(inode)) | |
117 | return -EBUSY; | |
118 | ||
119 | /* | |
120 | * Since the file was opened readonly, we have to initialize the jbd | |
121 | * inode and quotas here and not rely on ->open() doing it. This must | |
122 | * be done before evicting the inline data. | |
123 | */ | |
124 | ||
125 | err = ext4_inode_attach_jinode(inode); | |
126 | if (err) | |
127 | return err; | |
128 | ||
129 | err = dquot_initialize(inode); | |
130 | if (err) | |
131 | return err; | |
132 | ||
133 | err = ext4_convert_inline_data(inode); | |
134 | if (err) | |
135 | return err; | |
136 | ||
137 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
138 | ext4_warning_inode(inode, | |
139 | "verity is only allowed on extent-based files"); | |
140 | return -EOPNOTSUPP; | |
141 | } | |
142 | ||
143 | /* | |
144 | * ext4 uses the last allocated block to find the verity descriptor, so | |
145 | * we must remove any other blocks past EOF which might confuse things. | |
146 | */ | |
147 | err = ext4_truncate(inode); | |
148 | if (err) | |
149 | return err; | |
150 | ||
151 | handle = ext4_journal_start(inode, EXT4_HT_INODE, credits); | |
152 | if (IS_ERR(handle)) | |
153 | return PTR_ERR(handle); | |
154 | ||
155 | err = ext4_orphan_add(handle, inode); | |
156 | if (err == 0) | |
157 | ext4_set_inode_state(inode, EXT4_STATE_VERITY_IN_PROGRESS); | |
158 | ||
159 | ext4_journal_stop(handle); | |
160 | return err; | |
161 | } | |
162 | ||
163 | /* | |
164 | * ext4 stores the verity descriptor beginning on the next filesystem block | |
165 | * boundary after the Merkle tree. Then, the descriptor size is stored in the | |
166 | * last 4 bytes of the last allocated filesystem block --- which is either the | |
167 | * block in which the descriptor ends, or the next block after that if there | |
168 | * weren't at least 4 bytes remaining. | |
169 | * | |
170 | * We can't simply store the descriptor in an xattr because it *must* be | |
171 | * encrypted when ext4 encryption is used, but ext4 encryption doesn't encrypt | |
172 | * xattrs. Also, if the descriptor includes a large signature blob it may be | |
173 | * too large to store in an xattr without the EA_INODE feature. | |
174 | */ | |
175 | static int ext4_write_verity_descriptor(struct inode *inode, const void *desc, | |
176 | size_t desc_size, u64 merkle_tree_size) | |
177 | { | |
178 | const u64 desc_pos = round_up(ext4_verity_metadata_pos(inode) + | |
179 | merkle_tree_size, i_blocksize(inode)); | |
180 | const u64 desc_end = desc_pos + desc_size; | |
181 | const __le32 desc_size_disk = cpu_to_le32(desc_size); | |
182 | const u64 desc_size_pos = round_up(desc_end + sizeof(desc_size_disk), | |
183 | i_blocksize(inode)) - | |
184 | sizeof(desc_size_disk); | |
185 | int err; | |
186 | ||
187 | err = pagecache_write(inode, desc, desc_size, desc_pos); | |
188 | if (err) | |
189 | return err; | |
190 | ||
191 | return pagecache_write(inode, &desc_size_disk, sizeof(desc_size_disk), | |
192 | desc_size_pos); | |
193 | } | |
194 | ||
195 | static int ext4_end_enable_verity(struct file *filp, const void *desc, | |
196 | size_t desc_size, u64 merkle_tree_size) | |
197 | { | |
198 | struct inode *inode = file_inode(filp); | |
199 | const int credits = 2; /* superblock and inode for ext4_orphan_del() */ | |
200 | handle_t *handle; | |
201 | int err = 0; | |
202 | int err2; | |
203 | ||
204 | if (desc != NULL) { | |
205 | /* Succeeded; write the verity descriptor. */ | |
206 | err = ext4_write_verity_descriptor(inode, desc, desc_size, | |
207 | merkle_tree_size); | |
208 | ||
209 | /* Write all pages before clearing VERITY_IN_PROGRESS. */ | |
210 | if (!err) | |
211 | err = filemap_write_and_wait(inode->i_mapping); | |
212 | } | |
213 | ||
214 | /* If we failed, truncate anything we wrote past i_size. */ | |
215 | if (desc == NULL || err) | |
216 | ext4_truncate(inode); | |
217 | ||
218 | /* | |
219 | * We must always clean up by clearing EXT4_STATE_VERITY_IN_PROGRESS and | |
220 | * deleting the inode from the orphan list, even if something failed. | |
221 | * If everything succeeded, we'll also set the verity bit in the same | |
222 | * transaction. | |
223 | */ | |
224 | ||
225 | ext4_clear_inode_state(inode, EXT4_STATE_VERITY_IN_PROGRESS); | |
226 | ||
227 | handle = ext4_journal_start(inode, EXT4_HT_INODE, credits); | |
228 | if (IS_ERR(handle)) { | |
229 | ext4_orphan_del(NULL, inode); | |
230 | return PTR_ERR(handle); | |
231 | } | |
232 | ||
233 | err2 = ext4_orphan_del(handle, inode); | |
234 | if (err2) | |
235 | goto out_stop; | |
236 | ||
237 | if (desc != NULL && !err) { | |
238 | struct ext4_iloc iloc; | |
239 | ||
240 | err = ext4_reserve_inode_write(handle, inode, &iloc); | |
241 | if (err) | |
242 | goto out_stop; | |
243 | ext4_set_inode_flag(inode, EXT4_INODE_VERITY); | |
244 | ext4_set_inode_flags(inode); | |
245 | err = ext4_mark_iloc_dirty(handle, inode, &iloc); | |
246 | } | |
247 | out_stop: | |
248 | ext4_journal_stop(handle); | |
249 | return err ?: err2; | |
250 | } | |
251 | ||
252 | static int ext4_get_verity_descriptor_location(struct inode *inode, | |
253 | size_t *desc_size_ret, | |
254 | u64 *desc_pos_ret) | |
255 | { | |
256 | struct ext4_ext_path *path; | |
257 | struct ext4_extent *last_extent; | |
258 | u32 end_lblk; | |
259 | u64 desc_size_pos; | |
260 | __le32 desc_size_disk; | |
261 | u32 desc_size; | |
262 | u64 desc_pos; | |
263 | int err; | |
264 | ||
265 | /* | |
266 | * Descriptor size is in last 4 bytes of last allocated block. | |
267 | * See ext4_write_verity_descriptor(). | |
268 | */ | |
269 | ||
270 | if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) { | |
271 | EXT4_ERROR_INODE(inode, "verity file doesn't use extents"); | |
272 | return -EFSCORRUPTED; | |
273 | } | |
274 | ||
275 | path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0); | |
276 | if (IS_ERR(path)) | |
277 | return PTR_ERR(path); | |
278 | ||
279 | last_extent = path[path->p_depth].p_ext; | |
280 | if (!last_extent) { | |
281 | EXT4_ERROR_INODE(inode, "verity file has no extents"); | |
282 | ext4_ext_drop_refs(path); | |
283 | kfree(path); | |
284 | return -EFSCORRUPTED; | |
285 | } | |
286 | ||
287 | end_lblk = le32_to_cpu(last_extent->ee_block) + | |
288 | ext4_ext_get_actual_len(last_extent); | |
289 | desc_size_pos = (u64)end_lblk << inode->i_blkbits; | |
290 | ext4_ext_drop_refs(path); | |
291 | kfree(path); | |
292 | ||
293 | if (desc_size_pos < sizeof(desc_size_disk)) | |
294 | goto bad; | |
295 | desc_size_pos -= sizeof(desc_size_disk); | |
296 | ||
297 | err = pagecache_read(inode, &desc_size_disk, sizeof(desc_size_disk), | |
298 | desc_size_pos); | |
299 | if (err) | |
300 | return err; | |
301 | desc_size = le32_to_cpu(desc_size_disk); | |
302 | ||
303 | /* | |
304 | * The descriptor is stored just before the desc_size_disk, but starting | |
305 | * on a filesystem block boundary. | |
306 | */ | |
307 | ||
308 | if (desc_size > INT_MAX || desc_size > desc_size_pos) | |
309 | goto bad; | |
310 | ||
311 | desc_pos = round_down(desc_size_pos - desc_size, i_blocksize(inode)); | |
312 | if (desc_pos < ext4_verity_metadata_pos(inode)) | |
313 | goto bad; | |
314 | ||
315 | *desc_size_ret = desc_size; | |
316 | *desc_pos_ret = desc_pos; | |
317 | return 0; | |
318 | ||
319 | bad: | |
320 | EXT4_ERROR_INODE(inode, "verity file corrupted; can't find descriptor"); | |
321 | return -EFSCORRUPTED; | |
322 | } | |
323 | ||
324 | static int ext4_get_verity_descriptor(struct inode *inode, void *buf, | |
325 | size_t buf_size) | |
326 | { | |
327 | size_t desc_size = 0; | |
328 | u64 desc_pos = 0; | |
329 | int err; | |
330 | ||
331 | err = ext4_get_verity_descriptor_location(inode, &desc_size, &desc_pos); | |
332 | if (err) | |
333 | return err; | |
334 | ||
335 | if (buf_size) { | |
336 | if (desc_size > buf_size) | |
337 | return -ERANGE; | |
338 | err = pagecache_read(inode, buf, desc_size, desc_pos); | |
339 | if (err) | |
340 | return err; | |
341 | } | |
342 | return desc_size; | |
343 | } | |
344 | ||
345 | /* | |
346 | * Prefetch some pages from the file's Merkle tree. | |
347 | * | |
348 | * This is basically a stripped-down version of __do_page_cache_readahead() | |
349 | * which works on pages past i_size. | |
350 | */ | |
351 | static void ext4_merkle_tree_readahead(struct address_space *mapping, | |
352 | pgoff_t start_index, unsigned long count) | |
353 | { | |
354 | LIST_HEAD(pages); | |
355 | unsigned int nr_pages = 0; | |
356 | struct page *page; | |
357 | pgoff_t index; | |
358 | struct blk_plug plug; | |
359 | ||
360 | for (index = start_index; index < start_index + count; index++) { | |
361 | page = xa_load(&mapping->i_pages, index); | |
362 | if (!page || xa_is_value(page)) { | |
363 | page = __page_cache_alloc(readahead_gfp_mask(mapping)); | |
364 | if (!page) | |
365 | break; | |
366 | page->index = index; | |
367 | list_add(&page->lru, &pages); | |
368 | nr_pages++; | |
369 | } | |
370 | } | |
371 | blk_start_plug(&plug); | |
372 | ext4_mpage_readpages(mapping, &pages, NULL, nr_pages, true); | |
373 | blk_finish_plug(&plug); | |
374 | } | |
375 | ||
376 | static struct page *ext4_read_merkle_tree_page(struct inode *inode, | |
377 | pgoff_t index, | |
378 | unsigned long num_ra_pages) | |
379 | { | |
380 | struct page *page; | |
381 | ||
382 | index += ext4_verity_metadata_pos(inode) >> PAGE_SHIFT; | |
383 | ||
384 | page = find_get_page_flags(inode->i_mapping, index, FGP_ACCESSED); | |
385 | if (!page || !PageUptodate(page)) { | |
386 | if (page) | |
387 | put_page(page); | |
388 | else if (num_ra_pages > 1) | |
389 | ext4_merkle_tree_readahead(inode->i_mapping, index, | |
390 | num_ra_pages); | |
391 | page = read_mapping_page(inode->i_mapping, index, NULL); | |
392 | } | |
393 | return page; | |
394 | } | |
395 | ||
396 | static int ext4_write_merkle_tree_block(struct inode *inode, const void *buf, | |
397 | u64 index, int log_blocksize) | |
398 | { | |
399 | loff_t pos = ext4_verity_metadata_pos(inode) + (index << log_blocksize); | |
400 | ||
401 | return pagecache_write(inode, buf, 1 << log_blocksize, pos); | |
402 | } | |
403 | ||
404 | const struct fsverity_operations ext4_verityops = { | |
405 | .begin_enable_verity = ext4_begin_enable_verity, | |
406 | .end_enable_verity = ext4_end_enable_verity, | |
407 | .get_verity_descriptor = ext4_get_verity_descriptor, | |
408 | .read_merkle_tree_page = ext4_read_merkle_tree_page, | |
409 | .write_merkle_tree_block = ext4_write_merkle_tree_block, | |
410 | }; |