1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * fs/verity/verify.c: data verification functions, i.e. hooks for ->readpages()
  4 *
  5 * Copyright 2019 Google LLC
  6 */
  7
  8#include "fsverity_private.h"
  9
 10#include <crypto/hash.h>
 11#include <linux/bio.h>
 12#include <linux/ratelimit.h>
 13
 14static struct workqueue_struct *fsverity_read_workqueue;
 15
 16/**
 17 * hash_at_level() - compute the location of the block's hash at the given level
 18 *
 19 * @params:	(in) the Merkle tree parameters
 20 * @dindex:	(in) the index of the data block being verified
 21 * @level:	(in) the level of hash we want (0 is leaf level)
 22 * @hindex:	(out) the index of the hash block containing the wanted hash
 23 * @hoffset:	(out) the byte offset to the wanted hash within the hash block
 24 */
 25static void hash_at_level(const struct merkle_tree_params *params,
 26			  pgoff_t dindex, unsigned int level, pgoff_t *hindex,
 27			  unsigned int *hoffset)
 28{
 29	pgoff_t position;
 30
 31	/* Offset of the hash within the level's region, in hashes */
 32	position = dindex >> (level * params->log_arity);
 33
 34	/* Index of the hash block in the tree overall */
 35	*hindex = params->level_start[level] + (position >> params->log_arity);
 36
 37	/* Offset of the wanted hash (in bytes) within the hash block */
 38	*hoffset = (position & ((1 << params->log_arity) - 1)) <<
 39		   (params->log_blocksize - params->log_arity);
 40}
 41
 42/* Extract a hash from a hash page */
 43static void extract_hash(struct page *hpage, unsigned int hoffset,
 44			 unsigned int hsize, u8 *out)
 45{
 46	void *virt = kmap_atomic(hpage);
 47
 48	memcpy(out, virt + hoffset, hsize);
 49	kunmap_atomic(virt);
 50}
 51
 52static inline int cmp_hashes(const struct fsverity_info *vi,
 53			     const u8 *want_hash, const u8 *real_hash,
 54			     pgoff_t index, int level)
 55{
 56	const unsigned int hsize = vi->tree_params.digest_size;
 57
 58	if (memcmp(want_hash, real_hash, hsize) == 0)
 59		return 0;
 60
 61	fsverity_err(vi->inode,
 62		     "FILE CORRUPTED! index=%lu, level=%d, want_hash=%s:%*phN, real_hash=%s:%*phN",
 63		     index, level,
 64		     vi->tree_params.hash_alg->name, hsize, want_hash,
 65		     vi->tree_params.hash_alg->name, hsize, real_hash);
 66	return -EBADMSG;
 67}
 68
 69/*
 70 * Verify a single data page against the file's Merkle tree.
 71 *
 72 * In principle, we need to verify the entire path to the root node.  However,
 73 * for efficiency the filesystem may cache the hash pages.  Therefore we need
 74 * only ascend the tree until an already-verified page is seen, as indicated by
 75 * the PageChecked bit being set; then verify the path to that page.
 76 *
 77 * This code currently only supports the case where the verity block size is
 78 * equal to PAGE_SIZE.  Doing otherwise would be possible but tricky, since we
 79 * wouldn't be able to use the PageChecked bit.
 80 *
 81 * Note that multiple processes may race to verify a hash page and mark it
 82 * Checked, but it doesn't matter; the result will be the same either way.
 83 *
 84 * Return: true if the page is valid, else false.
 85 */
 86static bool verify_page(struct inode *inode, const struct fsverity_info *vi,
 87			struct ahash_request *req, struct page *data_page,
 88			unsigned long level0_ra_pages)
 89{
 90	const struct merkle_tree_params *params = &vi->tree_params;
 91	const unsigned int hsize = params->digest_size;
 92	const pgoff_t index = data_page->index;
 93	int level;
 94	u8 _want_hash[FS_VERITY_MAX_DIGEST_SIZE];
 95	const u8 *want_hash;
 96	u8 real_hash[FS_VERITY_MAX_DIGEST_SIZE];
 97	struct page *hpages[FS_VERITY_MAX_LEVELS];
 98	unsigned int hoffsets[FS_VERITY_MAX_LEVELS];
 99	int err;
100
101	if (WARN_ON_ONCE(!PageLocked(data_page) || PageUptodate(data_page)))
102		return false;
103
104	pr_debug_ratelimited("Verifying data page %lu...\n", index);
105
106	/*
107	 * Starting at the leaf level, ascend the tree saving hash pages along
108	 * the way until we find a verified hash page, indicated by PageChecked;
109	 * or until we reach the root.
110	 */
111	for (level = 0; level < params->num_levels; level++) {
112		pgoff_t hindex;
113		unsigned int hoffset;
114		struct page *hpage;
115
116		hash_at_level(params, index, level, &hindex, &hoffset);
117
118		pr_debug_ratelimited("Level %d: hindex=%lu, hoffset=%u\n",
119				     level, hindex, hoffset);
120
121		hpage = inode->i_sb->s_vop->read_merkle_tree_page(inode, hindex,
122				level == 0 ? level0_ra_pages : 0);
123		if (IS_ERR(hpage)) {
124			err = PTR_ERR(hpage);
125			fsverity_err(inode,
126				     "Error %d reading Merkle tree page %lu",
127				     err, hindex);
128			goto out;
129		}
130
131		if (PageChecked(hpage)) {
132			extract_hash(hpage, hoffset, hsize, _want_hash);
133			want_hash = _want_hash;
134			put_page(hpage);
135			pr_debug_ratelimited("Hash page already checked, want %s:%*phN\n",
136					     params->hash_alg->name,
137					     hsize, want_hash);
138			goto descend;
139		}
140		pr_debug_ratelimited("Hash page not yet checked\n");
141		hpages[level] = hpage;
142		hoffsets[level] = hoffset;
143	}
144
145	want_hash = vi->root_hash;
146	pr_debug("Want root hash: %s:%*phN\n",
147		 params->hash_alg->name, hsize, want_hash);
148descend:
149	/* Descend the tree verifying hash pages */
150	for (; level > 0; level--) {
151		struct page *hpage = hpages[level - 1];
152		unsigned int hoffset = hoffsets[level - 1];
153
154		err = fsverity_hash_page(params, inode, req, hpage, real_hash);
155		if (err)
156			goto out;
157		err = cmp_hashes(vi, want_hash, real_hash, index, level - 1);
158		if (err)
159			goto out;
160		SetPageChecked(hpage);
161		extract_hash(hpage, hoffset, hsize, _want_hash);
162		want_hash = _want_hash;
163		put_page(hpage);
164		pr_debug("Verified hash page at level %d, now want %s:%*phN\n",
165			 level - 1, params->hash_alg->name, hsize, want_hash);
166	}
167
168	/* Finally, verify the data page */
169	err = fsverity_hash_page(params, inode, req, data_page, real_hash);
170	if (err)
171		goto out;
172	err = cmp_hashes(vi, want_hash, real_hash, index, -1);
173out:
174	for (; level > 0; level--)
175		put_page(hpages[level - 1]);
176
177	return err == 0;
178}
179
180/**
181 * fsverity_verify_page() - verify a data page
182 * @page: the page to verity
183 *
184 * Verify a page that has just been read from a verity file.  The page must be a
185 * pagecache page that is still locked and not yet uptodate.
186 *
187 * Return: true if the page is valid, else false.
188 */
189bool fsverity_verify_page(struct page *page)
190{
191	struct inode *inode = page->mapping->host;
192	const struct fsverity_info *vi = inode->i_verity_info;
193	struct ahash_request *req;
194	bool valid;
195
196	/* This allocation never fails, since it's mempool-backed. */
197	req = fsverity_alloc_hash_request(vi->tree_params.hash_alg, GFP_NOFS);
198
199	valid = verify_page(inode, vi, req, page, 0);
200
201	fsverity_free_hash_request(vi->tree_params.hash_alg, req);
202
203	return valid;
204}
205EXPORT_SYMBOL_GPL(fsverity_verify_page);
206
207#ifdef CONFIG_BLOCK
208/**
209 * fsverity_verify_bio() - verify a 'read' bio that has just completed
210 * @bio: the bio to verify
211 *
212 * Verify a set of pages that have just been read from a verity file.  The pages
213 * must be pagecache pages that are still locked and not yet uptodate.  Pages
214 * that fail verification are set to the Error state.  Verification is skipped
215 * for pages already in the Error state, e.g. due to fscrypt decryption failure.
216 *
217 * This is a helper function for use by the ->readpages() method of filesystems
218 * that issue bios to read data directly into the page cache.  Filesystems that
219 * populate the page cache without issuing bios (e.g. non block-based
220 * filesystems) must instead call fsverity_verify_page() directly on each page.
221 * All filesystems must also call fsverity_verify_page() on holes.
222 */
223void fsverity_verify_bio(struct bio *bio)
224{
225	struct inode *inode = bio_first_page_all(bio)->mapping->host;
226	const struct fsverity_info *vi = inode->i_verity_info;
227	const struct merkle_tree_params *params = &vi->tree_params;
228	struct ahash_request *req;
229	struct bio_vec *bv;
230	struct bvec_iter_all iter_all;
231	unsigned long max_ra_pages = 0;
232
233	/* This allocation never fails, since it's mempool-backed. */
234	req = fsverity_alloc_hash_request(params->hash_alg, GFP_NOFS);
235
236	if (bio->bi_opf & REQ_RAHEAD) {
237		/*
238		 * If this bio is for data readahead, then we also do readahead
239		 * of the first (largest) level of the Merkle tree.  Namely,
240		 * when a Merkle tree page is read, we also try to piggy-back on
241		 * some additional pages -- up to 1/4 the number of data pages.
242		 *
243		 * This improves sequential read performance, as it greatly
244		 * reduces the number of I/O requests made to the Merkle tree.
245		 */
246		bio_for_each_segment_all(bv, bio, iter_all)
247			max_ra_pages++;
248		max_ra_pages /= 4;
249	}
250
251	bio_for_each_segment_all(bv, bio, iter_all) {
252		struct page *page = bv->bv_page;
253		unsigned long level0_index = page->index >> params->log_arity;
254		unsigned long level0_ra_pages =
255			min(max_ra_pages, params->level0_blocks - level0_index);
256
257		if (!PageError(page) &&
258		    !verify_page(inode, vi, req, page, level0_ra_pages))
259			SetPageError(page);
260	}
261
262	fsverity_free_hash_request(params->hash_alg, req);
263}
264EXPORT_SYMBOL_GPL(fsverity_verify_bio);
265#endif /* CONFIG_BLOCK */
266
267/**
268 * fsverity_enqueue_verify_work() - enqueue work on the fs-verity workqueue
269 * @work: the work to enqueue
270 *
271 * Enqueue verification work for asynchronous processing.
272 */
273void fsverity_enqueue_verify_work(struct work_struct *work)
274{
275	queue_work(fsverity_read_workqueue, work);
276}
277EXPORT_SYMBOL_GPL(fsverity_enqueue_verify_work);
278
279int __init fsverity_init_workqueue(void)
280{
281	/*
282	 * Use an unbound workqueue to allow bios to be verified in parallel
283	 * even when they happen to complete on the same CPU.  This sacrifices
284	 * locality, but it's worthwhile since hashing is CPU-intensive.
285	 *
286	 * Also use a high-priority workqueue to prioritize verification work,
287	 * which blocks reads from completing, over regular application tasks.
288	 */
289	fsverity_read_workqueue = alloc_workqueue("fsverity_read_queue",
290						  WQ_UNBOUND | WQ_HIGHPRI,
291						  num_online_cpus());
292	if (!fsverity_read_workqueue)
293		return -ENOMEM;
294	return 0;
295}
296
297void __init fsverity_exit_workqueue(void)
298{
299	destroy_workqueue(fsverity_read_workqueue);
300	fsverity_read_workqueue = NULL;
301}