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
 32static 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 */
 41static 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
 49		page = read_mapping_page(inode->i_mapping, pos >> PAGE_SHIFT,
 50					 NULL);
 51		if (IS_ERR(page))
 52			return PTR_ERR(page);
 53
 54		memcpy_from_page(buf, page, offset_in_page(pos), n);
 55
 56		put_page(page);
 57
 58		buf += n;
 59		pos += n;
 60		count -= n;
 61	}
 62	return 0;
 63}
 64
 65/*
 66 * Write some verity metadata to the inode for FS_IOC_ENABLE_VERITY.
 67 * kernel_write() can't be used because the file descriptor is readonly.
 68 */
 69static int pagecache_write(struct inode *inode, const void *buf, size_t count,
 70			   loff_t pos)
 71{
 72	struct address_space *mapping = inode->i_mapping;
 73	const struct address_space_operations *aops = mapping->a_ops;
 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 = NULL;
 83		int res;
 84
 85		res = aops->write_begin(NULL, mapping, pos, n, &page, &fsdata);
 86		if (res)
 87			return res;
 88
 89		memcpy_to_page(page, offset_in_page(pos), buf, n);
 90
 91		res = aops->write_end(NULL, mapping, pos, n, n, page, fsdata);
 92		if (res < 0)
 93			return res;
 94		if (res != n)
 95			return -EIO;
 96
 97		buf += n;
 98		pos += n;
 99		count -= n;
100	}
101	return 0;
102}
103
104static int ext4_begin_enable_verity(struct file *filp)
105{
106	struct inode *inode = file_inode(filp);
107	const int credits = 2; /* superblock and inode for ext4_orphan_add() */
108	handle_t *handle;
109	int err;
110
111	if (IS_DAX(inode) || ext4_test_inode_flag(inode, EXT4_INODE_DAX))
112		return -EINVAL;
113
114	if (ext4_verity_in_progress(inode))
115		return -EBUSY;
116
117	/*
118	 * Since the file was opened readonly, we have to initialize the jbd
119	 * inode and quotas here and not rely on ->open() doing it.  This must
120	 * be done before evicting the inline data.
121	 */
122
123	err = ext4_inode_attach_jinode(inode);
124	if (err)
125		return err;
126
127	err = dquot_initialize(inode);
128	if (err)
129		return err;
130
131	err = ext4_convert_inline_data(inode);
132	if (err)
133		return err;
134
135	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
136		ext4_warning_inode(inode,
137				   "verity is only allowed on extent-based files");
138		return -EOPNOTSUPP;
139	}
140
141	/*
142	 * ext4 uses the last allocated block to find the verity descriptor, so
143	 * we must remove any other blocks past EOF which might confuse things.
144	 */
145	err = ext4_truncate(inode);
146	if (err)
147		return err;
148
149	handle = ext4_journal_start(inode, EXT4_HT_INODE, credits);
150	if (IS_ERR(handle))
151		return PTR_ERR(handle);
152
153	err = ext4_orphan_add(handle, inode);
154	if (err == 0)
155		ext4_set_inode_state(inode, EXT4_STATE_VERITY_IN_PROGRESS);
156
157	ext4_journal_stop(handle);
158	return err;
159}
160
161/*
162 * ext4 stores the verity descriptor beginning on the next filesystem block
163 * boundary after the Merkle tree.  Then, the descriptor size is stored in the
164 * last 4 bytes of the last allocated filesystem block --- which is either the
165 * block in which the descriptor ends, or the next block after that if there
166 * weren't at least 4 bytes remaining.
167 *
168 * We can't simply store the descriptor in an xattr because it *must* be
169 * encrypted when ext4 encryption is used, but ext4 encryption doesn't encrypt
170 * xattrs.  Also, if the descriptor includes a large signature blob it may be
171 * too large to store in an xattr without the EA_INODE feature.
172 */
173static int ext4_write_verity_descriptor(struct inode *inode, const void *desc,
174					size_t desc_size, u64 merkle_tree_size)
175{
176	const u64 desc_pos = round_up(ext4_verity_metadata_pos(inode) +
177				      merkle_tree_size, i_blocksize(inode));
178	const u64 desc_end = desc_pos + desc_size;
179	const __le32 desc_size_disk = cpu_to_le32(desc_size);
180	const u64 desc_size_pos = round_up(desc_end + sizeof(desc_size_disk),
181					   i_blocksize(inode)) -
182				  sizeof(desc_size_disk);
183	int err;
184
185	err = pagecache_write(inode, desc, desc_size, desc_pos);
186	if (err)
187		return err;
188
189	return pagecache_write(inode, &desc_size_disk, sizeof(desc_size_disk),
190			       desc_size_pos);
191}
192
193static int ext4_end_enable_verity(struct file *filp, const void *desc,
194				  size_t desc_size, u64 merkle_tree_size)
195{
196	struct inode *inode = file_inode(filp);
197	const int credits = 2; /* superblock and inode for ext4_orphan_del() */
198	handle_t *handle;
199	struct ext4_iloc iloc;
200	int err = 0;
201
202	/*
203	 * If an error already occurred (which fs/verity/ signals by passing
204	 * desc == NULL), then only clean-up is needed.
205	 */
206	if (desc == NULL)
207		goto cleanup;
208
209	/* Append the verity descriptor. */
210	err = ext4_write_verity_descriptor(inode, desc, desc_size,
211					   merkle_tree_size);
212	if (err)
213		goto cleanup;
214
215	/*
216	 * Write all pages (both data and verity metadata).  Note that this must
217	 * happen before clearing EXT4_STATE_VERITY_IN_PROGRESS; otherwise pages
218	 * beyond i_size won't be written properly.  For crash consistency, this
219	 * also must happen before the verity inode flag gets persisted.
220	 */
221	err = filemap_write_and_wait(inode->i_mapping);
222	if (err)
223		goto cleanup;
224
225	/*
226	 * Finally, set the verity inode flag and remove the inode from the
227	 * orphan list (in a single transaction).
228	 */
229
230	handle = ext4_journal_start(inode, EXT4_HT_INODE, credits);
231	if (IS_ERR(handle)) {
232		err = PTR_ERR(handle);
233		goto cleanup;
234	}
235
236	err = ext4_orphan_del(handle, inode);
237	if (err)
238		goto stop_and_cleanup;
239
240	err = ext4_reserve_inode_write(handle, inode, &iloc);
241	if (err)
242		goto stop_and_cleanup;
243
244	ext4_set_inode_flag(inode, EXT4_INODE_VERITY);
245	ext4_set_inode_flags(inode, false);
246	err = ext4_mark_iloc_dirty(handle, inode, &iloc);
247	if (err)
248		goto stop_and_cleanup;
249
250	ext4_journal_stop(handle);
251
252	ext4_clear_inode_state(inode, EXT4_STATE_VERITY_IN_PROGRESS);
253	return 0;
254
255stop_and_cleanup:
256	ext4_journal_stop(handle);
257cleanup:
258	/*
259	 * Verity failed to be enabled, so clean up by truncating any verity
260	 * metadata that was written beyond i_size (both from cache and from
261	 * disk), removing the inode from the orphan list (if it wasn't done
262	 * already), and clearing EXT4_STATE_VERITY_IN_PROGRESS.
263	 */
264	truncate_inode_pages(inode->i_mapping, inode->i_size);
265	ext4_truncate(inode);
266	ext4_orphan_del(NULL, inode);
267	ext4_clear_inode_state(inode, EXT4_STATE_VERITY_IN_PROGRESS);
268	return err;
269}
270
271static int ext4_get_verity_descriptor_location(struct inode *inode,
272					       size_t *desc_size_ret,
273					       u64 *desc_pos_ret)
274{
275	struct ext4_ext_path *path;
276	struct ext4_extent *last_extent;
277	u32 end_lblk;
278	u64 desc_size_pos;
279	__le32 desc_size_disk;
280	u32 desc_size;
281	u64 desc_pos;
282	int err;
283
284	/*
285	 * Descriptor size is in last 4 bytes of last allocated block.
286	 * See ext4_write_verity_descriptor().
287	 */
288
289	if (!ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)) {
290		EXT4_ERROR_INODE(inode, "verity file doesn't use extents");
291		return -EFSCORRUPTED;
292	}
293
294	path = ext4_find_extent(inode, EXT_MAX_BLOCKS - 1, NULL, 0);
295	if (IS_ERR(path))
296		return PTR_ERR(path);
297
298	last_extent = path[path->p_depth].p_ext;
299	if (!last_extent) {
300		EXT4_ERROR_INODE(inode, "verity file has no extents");
301		ext4_free_ext_path(path);
302		return -EFSCORRUPTED;
303	}
304
305	end_lblk = le32_to_cpu(last_extent->ee_block) +
306		   ext4_ext_get_actual_len(last_extent);
307	desc_size_pos = (u64)end_lblk << inode->i_blkbits;
308	ext4_free_ext_path(path);
309
310	if (desc_size_pos < sizeof(desc_size_disk))
311		goto bad;
312	desc_size_pos -= sizeof(desc_size_disk);
313
314	err = pagecache_read(inode, &desc_size_disk, sizeof(desc_size_disk),
315			     desc_size_pos);
316	if (err)
317		return err;
318	desc_size = le32_to_cpu(desc_size_disk);
319
320	/*
321	 * The descriptor is stored just before the desc_size_disk, but starting
322	 * on a filesystem block boundary.
323	 */
324
325	if (desc_size > INT_MAX || desc_size > desc_size_pos)
326		goto bad;
327
328	desc_pos = round_down(desc_size_pos - desc_size, i_blocksize(inode));
329	if (desc_pos < ext4_verity_metadata_pos(inode))
330		goto bad;
331
332	*desc_size_ret = desc_size;
333	*desc_pos_ret = desc_pos;
334	return 0;
335
336bad:
337	EXT4_ERROR_INODE(inode, "verity file corrupted; can't find descriptor");
338	return -EFSCORRUPTED;
339}
340
341static int ext4_get_verity_descriptor(struct inode *inode, void *buf,
342				      size_t buf_size)
343{
344	size_t desc_size = 0;
345	u64 desc_pos = 0;
346	int err;
347
348	err = ext4_get_verity_descriptor_location(inode, &desc_size, &desc_pos);
349	if (err)
350		return err;
351
352	if (buf_size) {
353		if (desc_size > buf_size)
354			return -ERANGE;
355		err = pagecache_read(inode, buf, desc_size, desc_pos);
356		if (err)
357			return err;
358	}
359	return desc_size;
360}
361
362static struct page *ext4_read_merkle_tree_page(struct inode *inode,
363					       pgoff_t index,
364					       unsigned long num_ra_pages)
365{
366	struct page *page;
367
368	index += ext4_verity_metadata_pos(inode) >> PAGE_SHIFT;
369
370	page = find_get_page_flags(inode->i_mapping, index, FGP_ACCESSED);
371	if (!page || !PageUptodate(page)) {
372		DEFINE_READAHEAD(ractl, NULL, NULL, inode->i_mapping, index);
373
374		if (page)
375			put_page(page);
376		else if (num_ra_pages > 1)
377			page_cache_ra_unbounded(&ractl, num_ra_pages, 0);
378		page = read_mapping_page(inode->i_mapping, index, NULL);
379	}
380	return page;
381}
382
383static int ext4_write_merkle_tree_block(struct inode *inode, const void *buf,
384					u64 index, int log_blocksize)
385{
386	loff_t pos = ext4_verity_metadata_pos(inode) + (index << log_blocksize);
387
388	return pagecache_write(inode, buf, 1 << log_blocksize, pos);
389}
390
391const struct fsverity_operations ext4_verityops = {
392	.begin_enable_verity	= ext4_begin_enable_verity,
393	.end_enable_verity	= ext4_end_enable_verity,
394	.get_verity_descriptor	= ext4_get_verity_descriptor,
395	.read_merkle_tree_page	= ext4_read_merkle_tree_page,
396	.write_merkle_tree_block = ext4_write_merkle_tree_block,
397};