xref: /openbmc/linux/fs/btrfs/verity.c (revision eafc0a02)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/init.h>
4 #include <linux/fs.h>
5 #include <linux/slab.h>
6 #include <linux/rwsem.h>
7 #include <linux/xattr.h>
8 #include <linux/security.h>
9 #include <linux/posix_acl_xattr.h>
10 #include <linux/iversion.h>
11 #include <linux/fsverity.h>
12 #include <linux/sched/mm.h>
13 #include "ctree.h"
14 #include "btrfs_inode.h"
15 #include "transaction.h"
16 #include "disk-io.h"
17 #include "locking.h"
18 
19 /*
20  * Implementation of the interface defined in struct fsverity_operations.
21  *
22  * The main question is how and where to store the verity descriptor and the
23  * Merkle tree. We store both in dedicated btree items in the filesystem tree,
24  * together with the rest of the inode metadata. This means we'll need to do
25  * extra work to encrypt them once encryption is supported in btrfs, but btrfs
26  * has a lot of careful code around i_size and it seems better to make a new key
27  * type than try and adjust all of our expectations for i_size.
28  *
29  * Note that this differs from the implementation in ext4 and f2fs, where
30  * this data is stored as if it were in the file, but past EOF. However, btrfs
31  * does not have a widespread mechanism for caching opaque metadata pages, so we
32  * do pretend that the Merkle tree pages themselves are past EOF for the
33  * purposes of caching them (as opposed to creating a virtual inode).
34  *
35  * fs verity items are stored under two different key types on disk.
36  * The descriptor items:
37  * [ inode objectid, BTRFS_VERITY_DESC_ITEM_KEY, offset ]
38  *
39  * At offset 0, we store a btrfs_verity_descriptor_item which tracks the
40  * size of the descriptor item and some extra data for encryption.
41  * Starting at offset 1, these hold the generic fs verity descriptor.
42  * The latter are opaque to btrfs, we just read and write them as a blob for
43  * the higher level verity code.  The most common descriptor size is 256 bytes.
44  *
45  * The merkle tree items:
46  * [ inode objectid, BTRFS_VERITY_MERKLE_ITEM_KEY, offset ]
47  *
48  * These also start at offset 0, and correspond to the merkle tree bytes.
49  * So when fsverity asks for page 0 of the merkle tree, we pull up one page
50  * starting at offset 0 for this key type.  These are also opaque to btrfs,
51  * we're blindly storing whatever fsverity sends down.
52  *
53  * Another important consideration is the fact that the Merkle tree data scales
54  * linearly with the size of the file (with 4K pages/blocks and SHA-256, it's
55  * ~1/127th the size) so for large files, writing the tree can be a lengthy
56  * operation. For that reason, we guard the whole enable verity operation
57  * (between begin_enable_verity and end_enable_verity) with an orphan item.
58  * Again, because the data can be pretty large, it's quite possible that we
59  * could run out of space writing it, so we try our best to handle errors by
60  * stopping and rolling back rather than aborting the victim transaction.
61  */
62 
63 #define MERKLE_START_ALIGN			65536
64 
65 /*
66  * Compute the logical file offset where we cache the Merkle tree.
67  *
68  * @inode:  inode of the verity file
69  *
70  * For the purposes of caching the Merkle tree pages, as required by
71  * fs-verity, it is convenient to do size computations in terms of a file
72  * offset, rather than in terms of page indices.
73  *
74  * Use 64K to be sure it's past the last page in the file, even with 64K pages.
75  * That rounding operation itself can overflow loff_t, so we do it in u64 and
76  * check.
77  *
78  * Returns the file offset on success, negative error code on failure.
79  */
80 static loff_t merkle_file_pos(const struct inode *inode)
81 {
82 	u64 sz = inode->i_size;
83 	u64 rounded = round_up(sz, MERKLE_START_ALIGN);
84 
85 	if (rounded > inode->i_sb->s_maxbytes)
86 		return -EFBIG;
87 
88 	return rounded;
89 }
90 
91 /*
92  * Drop all the items for this inode with this key_type.
93  *
94  * @inode:     inode to drop items for
95  * @key_type:  type of items to drop (BTRFS_VERITY_DESC_ITEM or
96  *             BTRFS_VERITY_MERKLE_ITEM)
97  *
98  * Before doing a verity enable we cleanup any existing verity items.
99  * This is also used to clean up if a verity enable failed half way through.
100  *
101  * Returns number of dropped items on success, negative error code on failure.
102  */
103 static int drop_verity_items(struct btrfs_inode *inode, u8 key_type)
104 {
105 	struct btrfs_trans_handle *trans;
106 	struct btrfs_root *root = inode->root;
107 	struct btrfs_path *path;
108 	struct btrfs_key key;
109 	int count = 0;
110 	int ret;
111 
112 	path = btrfs_alloc_path();
113 	if (!path)
114 		return -ENOMEM;
115 
116 	while (1) {
117 		/* 1 for the item being dropped */
118 		trans = btrfs_start_transaction(root, 1);
119 		if (IS_ERR(trans)) {
120 			ret = PTR_ERR(trans);
121 			goto out;
122 		}
123 
124 		/*
125 		 * Walk backwards through all the items until we find one that
126 		 * isn't from our key type or objectid
127 		 */
128 		key.objectid = btrfs_ino(inode);
129 		key.type = key_type;
130 		key.offset = (u64)-1;
131 
132 		ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
133 		if (ret > 0) {
134 			ret = 0;
135 			/* No more keys of this type, we're done */
136 			if (path->slots[0] == 0)
137 				break;
138 			path->slots[0]--;
139 		} else if (ret < 0) {
140 			btrfs_end_transaction(trans);
141 			goto out;
142 		}
143 
144 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
145 
146 		/* No more keys of this type, we're done */
147 		if (key.objectid != btrfs_ino(inode) || key.type != key_type)
148 			break;
149 
150 		/*
151 		 * This shouldn't be a performance sensitive function because
152 		 * it's not used as part of truncate.  If it ever becomes
153 		 * perf sensitive, change this to walk forward and bulk delete
154 		 * items
155 		 */
156 		ret = btrfs_del_items(trans, root, path, path->slots[0], 1);
157 		if (ret) {
158 			btrfs_end_transaction(trans);
159 			goto out;
160 		}
161 		count++;
162 		btrfs_release_path(path);
163 		btrfs_end_transaction(trans);
164 	}
165 	ret = count;
166 	btrfs_end_transaction(trans);
167 out:
168 	btrfs_free_path(path);
169 	return ret;
170 }
171 
172 /*
173  * Drop all verity items
174  *
175  * @inode:  inode to drop verity items for
176  *
177  * In most contexts where we are dropping verity items, we want to do it for all
178  * the types of verity items, not a particular one.
179  *
180  * Returns: 0 on success, negative error code on failure.
181  */
182 int btrfs_drop_verity_items(struct btrfs_inode *inode)
183 {
184 	int ret;
185 
186 	ret = drop_verity_items(inode, BTRFS_VERITY_DESC_ITEM_KEY);
187 	if (ret < 0)
188 		return ret;
189 	ret = drop_verity_items(inode, BTRFS_VERITY_MERKLE_ITEM_KEY);
190 	if (ret < 0)
191 		return ret;
192 
193 	return 0;
194 }
195 
196 /*
197  * Insert and write inode items with a given key type and offset.
198  *
199  * @inode:     inode to insert for
200  * @key_type:  key type to insert
201  * @offset:    item offset to insert at
202  * @src:       source data to write
203  * @len:       length of source data to write
204  *
205  * Write len bytes from src into items of up to 2K length.
206  * The inserted items will have key (ino, key_type, offset + off) where off is
207  * consecutively increasing from 0 up to the last item ending at offset + len.
208  *
209  * Returns 0 on success and a negative error code on failure.
210  */
211 static int write_key_bytes(struct btrfs_inode *inode, u8 key_type, u64 offset,
212 			   const char *src, u64 len)
213 {
214 	struct btrfs_trans_handle *trans;
215 	struct btrfs_path *path;
216 	struct btrfs_root *root = inode->root;
217 	struct extent_buffer *leaf;
218 	struct btrfs_key key;
219 	unsigned long copy_bytes;
220 	unsigned long src_offset = 0;
221 	void *data;
222 	int ret = 0;
223 
224 	path = btrfs_alloc_path();
225 	if (!path)
226 		return -ENOMEM;
227 
228 	while (len > 0) {
229 		/* 1 for the new item being inserted */
230 		trans = btrfs_start_transaction(root, 1);
231 		if (IS_ERR(trans)) {
232 			ret = PTR_ERR(trans);
233 			break;
234 		}
235 
236 		key.objectid = btrfs_ino(inode);
237 		key.type = key_type;
238 		key.offset = offset;
239 
240 		/*
241 		 * Insert 2K at a time mostly to be friendly for smaller leaf
242 		 * size filesystems
243 		 */
244 		copy_bytes = min_t(u64, len, 2048);
245 
246 		ret = btrfs_insert_empty_item(trans, root, path, &key, copy_bytes);
247 		if (ret) {
248 			btrfs_end_transaction(trans);
249 			break;
250 		}
251 
252 		leaf = path->nodes[0];
253 
254 		data = btrfs_item_ptr(leaf, path->slots[0], void);
255 		write_extent_buffer(leaf, src + src_offset,
256 				    (unsigned long)data, copy_bytes);
257 		offset += copy_bytes;
258 		src_offset += copy_bytes;
259 		len -= copy_bytes;
260 
261 		btrfs_release_path(path);
262 		btrfs_end_transaction(trans);
263 	}
264 
265 	btrfs_free_path(path);
266 	return ret;
267 }
268 
269 /*
270  * Read inode items of the given key type and offset from the btree.
271  *
272  * @inode:      inode to read items of
273  * @key_type:   key type to read
274  * @offset:     item offset to read from
275  * @dest:       Buffer to read into. This parameter has slightly tricky
276  *              semantics.  If it is NULL, the function will not do any copying
277  *              and will just return the size of all the items up to len bytes.
278  *              If dest_page is passed, then the function will kmap_local the
279  *              page and ignore dest, but it must still be non-NULL to avoid the
280  *              counting-only behavior.
281  * @len:        length in bytes to read
282  * @dest_page:  copy into this page instead of the dest buffer
283  *
284  * Helper function to read items from the btree.  This returns the number of
285  * bytes read or < 0 for errors.  We can return short reads if the items don't
286  * exist on disk or aren't big enough to fill the desired length.  Supports
287  * reading into a provided buffer (dest) or into the page cache
288  *
289  * Returns number of bytes read or a negative error code on failure.
290  */
291 static int read_key_bytes(struct btrfs_inode *inode, u8 key_type, u64 offset,
292 			  char *dest, u64 len, struct page *dest_page)
293 {
294 	struct btrfs_path *path;
295 	struct btrfs_root *root = inode->root;
296 	struct extent_buffer *leaf;
297 	struct btrfs_key key;
298 	u64 item_end;
299 	u64 copy_end;
300 	int copied = 0;
301 	u32 copy_offset;
302 	unsigned long copy_bytes;
303 	unsigned long dest_offset = 0;
304 	void *data;
305 	char *kaddr = dest;
306 	int ret;
307 
308 	path = btrfs_alloc_path();
309 	if (!path)
310 		return -ENOMEM;
311 
312 	if (dest_page)
313 		path->reada = READA_FORWARD;
314 
315 	key.objectid = btrfs_ino(inode);
316 	key.type = key_type;
317 	key.offset = offset;
318 
319 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
320 	if (ret < 0) {
321 		goto out;
322 	} else if (ret > 0) {
323 		ret = 0;
324 		if (path->slots[0] == 0)
325 			goto out;
326 		path->slots[0]--;
327 	}
328 
329 	while (len > 0) {
330 		leaf = path->nodes[0];
331 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
332 
333 		if (key.objectid != btrfs_ino(inode) || key.type != key_type)
334 			break;
335 
336 		item_end = btrfs_item_size(leaf, path->slots[0]) + key.offset;
337 
338 		if (copied > 0) {
339 			/*
340 			 * Once we've copied something, we want all of the items
341 			 * to be sequential
342 			 */
343 			if (key.offset != offset)
344 				break;
345 		} else {
346 			/*
347 			 * Our initial offset might be in the middle of an
348 			 * item.  Make sure it all makes sense.
349 			 */
350 			if (key.offset > offset)
351 				break;
352 			if (item_end <= offset)
353 				break;
354 		}
355 
356 		/* desc = NULL to just sum all the item lengths */
357 		if (!dest)
358 			copy_end = item_end;
359 		else
360 			copy_end = min(offset + len, item_end);
361 
362 		/* Number of bytes in this item we want to copy */
363 		copy_bytes = copy_end - offset;
364 
365 		/* Offset from the start of item for copying */
366 		copy_offset = offset - key.offset;
367 
368 		if (dest) {
369 			if (dest_page)
370 				kaddr = kmap_local_page(dest_page);
371 
372 			data = btrfs_item_ptr(leaf, path->slots[0], void);
373 			read_extent_buffer(leaf, kaddr + dest_offset,
374 					   (unsigned long)data + copy_offset,
375 					   copy_bytes);
376 
377 			if (dest_page)
378 				kunmap_local(kaddr);
379 		}
380 
381 		offset += copy_bytes;
382 		dest_offset += copy_bytes;
383 		len -= copy_bytes;
384 		copied += copy_bytes;
385 
386 		path->slots[0]++;
387 		if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
388 			/*
389 			 * We've reached the last slot in this leaf and we need
390 			 * to go to the next leaf.
391 			 */
392 			ret = btrfs_next_leaf(root, path);
393 			if (ret < 0) {
394 				break;
395 			} else if (ret > 0) {
396 				ret = 0;
397 				break;
398 			}
399 		}
400 	}
401 out:
402 	btrfs_free_path(path);
403 	if (!ret)
404 		ret = copied;
405 	return ret;
406 }
407 
408 /*
409  * Delete an fsverity orphan
410  *
411  * @trans:  transaction to do the delete in
412  * @inode:  inode to orphan
413  *
414  * Capture verity orphan specific logic that is repeated in the couple places
415  * we delete verity orphans. Specifically, handling ENOENT and ignoring inodes
416  * with 0 links.
417  *
418  * Returns zero on success or a negative error code on failure.
419  */
420 static int del_orphan(struct btrfs_trans_handle *trans, struct btrfs_inode *inode)
421 {
422 	struct btrfs_root *root = inode->root;
423 	int ret;
424 
425 	/*
426 	 * If the inode has no links, it is either already unlinked, or was
427 	 * created with O_TMPFILE. In either case, it should have an orphan from
428 	 * that other operation. Rather than reference count the orphans, we
429 	 * simply ignore them here, because we only invoke the verity path in
430 	 * the orphan logic when i_nlink is 1.
431 	 */
432 	if (!inode->vfs_inode.i_nlink)
433 		return 0;
434 
435 	ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
436 	if (ret == -ENOENT)
437 		ret = 0;
438 	return ret;
439 }
440 
441 /*
442  * Rollback in-progress verity if we encounter an error.
443  *
444  * @inode:  inode verity had an error for
445  *
446  * We try to handle recoverable errors while enabling verity by rolling it back
447  * and just failing the operation, rather than having an fs level error no
448  * matter what. However, any error in rollback is unrecoverable.
449  *
450  * Returns 0 on success, negative error code on failure.
451  */
452 static int rollback_verity(struct btrfs_inode *inode)
453 {
454 	struct btrfs_trans_handle *trans = NULL;
455 	struct btrfs_root *root = inode->root;
456 	int ret;
457 
458 	ASSERT(inode_is_locked(&inode->vfs_inode));
459 	truncate_inode_pages(inode->vfs_inode.i_mapping, inode->vfs_inode.i_size);
460 	clear_bit(BTRFS_INODE_VERITY_IN_PROGRESS, &inode->runtime_flags);
461 	ret = btrfs_drop_verity_items(inode);
462 	if (ret) {
463 		btrfs_handle_fs_error(root->fs_info, ret,
464 				"failed to drop verity items in rollback %llu",
465 				(u64)inode->vfs_inode.i_ino);
466 		goto out;
467 	}
468 
469 	/*
470 	 * 1 for updating the inode flag
471 	 * 1 for deleting the orphan
472 	 */
473 	trans = btrfs_start_transaction(root, 2);
474 	if (IS_ERR(trans)) {
475 		ret = PTR_ERR(trans);
476 		trans = NULL;
477 		btrfs_handle_fs_error(root->fs_info, ret,
478 			"failed to start transaction in verity rollback %llu",
479 			(u64)inode->vfs_inode.i_ino);
480 		goto out;
481 	}
482 	inode->ro_flags &= ~BTRFS_INODE_RO_VERITY;
483 	btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
484 	ret = btrfs_update_inode(trans, root, inode);
485 	if (ret) {
486 		btrfs_abort_transaction(trans, ret);
487 		goto out;
488 	}
489 	ret = del_orphan(trans, inode);
490 	if (ret) {
491 		btrfs_abort_transaction(trans, ret);
492 		goto out;
493 	}
494 out:
495 	if (trans)
496 		btrfs_end_transaction(trans);
497 	return ret;
498 }
499 
500 /*
501  * Finalize making the file a valid verity file
502  *
503  * @inode:      inode to be marked as verity
504  * @desc:       contents of the verity descriptor to write (not NULL)
505  * @desc_size:  size of the verity descriptor
506  *
507  * Do the actual work of finalizing verity after successfully writing the Merkle
508  * tree:
509  *
510  * - write out the descriptor items
511  * - mark the inode with the verity flag
512  * - delete the orphan item
513  * - mark the ro compat bit
514  * - clear the in progress bit
515  *
516  * Returns 0 on success, negative error code on failure.
517  */
518 static int finish_verity(struct btrfs_inode *inode, const void *desc,
519 			 size_t desc_size)
520 {
521 	struct btrfs_trans_handle *trans = NULL;
522 	struct btrfs_root *root = inode->root;
523 	struct btrfs_verity_descriptor_item item;
524 	int ret;
525 
526 	/* Write out the descriptor item */
527 	memset(&item, 0, sizeof(item));
528 	btrfs_set_stack_verity_descriptor_size(&item, desc_size);
529 	ret = write_key_bytes(inode, BTRFS_VERITY_DESC_ITEM_KEY, 0,
530 			      (const char *)&item, sizeof(item));
531 	if (ret)
532 		goto out;
533 
534 	/* Write out the descriptor itself */
535 	ret = write_key_bytes(inode, BTRFS_VERITY_DESC_ITEM_KEY, 1,
536 			      desc, desc_size);
537 	if (ret)
538 		goto out;
539 
540 	/*
541 	 * 1 for updating the inode flag
542 	 * 1 for deleting the orphan
543 	 */
544 	trans = btrfs_start_transaction(root, 2);
545 	if (IS_ERR(trans)) {
546 		ret = PTR_ERR(trans);
547 		goto out;
548 	}
549 	inode->ro_flags |= BTRFS_INODE_RO_VERITY;
550 	btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
551 	ret = btrfs_update_inode(trans, root, inode);
552 	if (ret)
553 		goto end_trans;
554 	ret = del_orphan(trans, inode);
555 	if (ret)
556 		goto end_trans;
557 	clear_bit(BTRFS_INODE_VERITY_IN_PROGRESS, &inode->runtime_flags);
558 	btrfs_set_fs_compat_ro(root->fs_info, VERITY);
559 end_trans:
560 	btrfs_end_transaction(trans);
561 out:
562 	return ret;
563 
564 }
565 
566 /*
567  * fsverity op that begins enabling verity.
568  *
569  * @filp:  file to enable verity on
570  *
571  * Begin enabling fsverity for the file. We drop any existing verity items, add
572  * an orphan and set the in progress bit.
573  *
574  * Returns 0 on success, negative error code on failure.
575  */
576 static int btrfs_begin_enable_verity(struct file *filp)
577 {
578 	struct btrfs_inode *inode = BTRFS_I(file_inode(filp));
579 	struct btrfs_root *root = inode->root;
580 	struct btrfs_trans_handle *trans;
581 	int ret;
582 
583 	ASSERT(inode_is_locked(file_inode(filp)));
584 
585 	if (test_bit(BTRFS_INODE_VERITY_IN_PROGRESS, &inode->runtime_flags))
586 		return -EBUSY;
587 
588 	/*
589 	 * This should almost never do anything, but theoretically, it's
590 	 * possible that we failed to enable verity on a file, then were
591 	 * interrupted or failed while rolling back, failed to cleanup the
592 	 * orphan, and finally attempt to enable verity again.
593 	 */
594 	ret = btrfs_drop_verity_items(inode);
595 	if (ret)
596 		return ret;
597 
598 	/* 1 for the orphan item */
599 	trans = btrfs_start_transaction(root, 1);
600 	if (IS_ERR(trans))
601 		return PTR_ERR(trans);
602 
603 	ret = btrfs_orphan_add(trans, inode);
604 	if (!ret)
605 		set_bit(BTRFS_INODE_VERITY_IN_PROGRESS, &inode->runtime_flags);
606 	btrfs_end_transaction(trans);
607 
608 	return 0;
609 }
610 
611 /*
612  * fsverity op that ends enabling verity.
613  *
614  * @filp:              file we are finishing enabling verity on
615  * @desc:              verity descriptor to write out (NULL in error conditions)
616  * @desc_size:         size of the verity descriptor (variable with signatures)
617  * @merkle_tree_size:  size of the merkle tree in bytes
618  *
619  * If desc is null, then VFS is signaling an error occurred during verity
620  * enable, and we should try to rollback. Otherwise, attempt to finish verity.
621  *
622  * Returns 0 on success, negative error code on error.
623  */
624 static int btrfs_end_enable_verity(struct file *filp, const void *desc,
625 				   size_t desc_size, u64 merkle_tree_size)
626 {
627 	struct btrfs_inode *inode = BTRFS_I(file_inode(filp));
628 	int ret = 0;
629 	int rollback_ret;
630 
631 	ASSERT(inode_is_locked(file_inode(filp)));
632 
633 	if (desc == NULL)
634 		goto rollback;
635 
636 	ret = finish_verity(inode, desc, desc_size);
637 	if (ret)
638 		goto rollback;
639 	return ret;
640 
641 rollback:
642 	rollback_ret = rollback_verity(inode);
643 	if (rollback_ret)
644 		btrfs_err(inode->root->fs_info,
645 			  "failed to rollback verity items: %d", rollback_ret);
646 	return ret;
647 }
648 
649 /*
650  * fsverity op that gets the struct fsverity_descriptor.
651  *
652  * @inode:     inode to get the descriptor of
653  * @buf:       output buffer for the descriptor contents
654  * @buf_size:  size of the output buffer. 0 to query the size
655  *
656  * fsverity does a two pass setup for reading the descriptor, in the first pass
657  * it calls with buf_size = 0 to query the size of the descriptor, and then in
658  * the second pass it actually reads the descriptor off disk.
659  *
660  * Returns the size on success or a negative error code on failure.
661  */
662 static int btrfs_get_verity_descriptor(struct inode *inode, void *buf,
663 				       size_t buf_size)
664 {
665 	u64 true_size;
666 	int ret = 0;
667 	struct btrfs_verity_descriptor_item item;
668 
669 	memset(&item, 0, sizeof(item));
670 	ret = read_key_bytes(BTRFS_I(inode), BTRFS_VERITY_DESC_ITEM_KEY, 0,
671 			     (char *)&item, sizeof(item), NULL);
672 	if (ret < 0)
673 		return ret;
674 
675 	if (item.reserved[0] != 0 || item.reserved[1] != 0)
676 		return -EUCLEAN;
677 
678 	true_size = btrfs_stack_verity_descriptor_size(&item);
679 	if (true_size > INT_MAX)
680 		return -EUCLEAN;
681 
682 	if (buf_size == 0)
683 		return true_size;
684 	if (buf_size < true_size)
685 		return -ERANGE;
686 
687 	ret = read_key_bytes(BTRFS_I(inode), BTRFS_VERITY_DESC_ITEM_KEY, 1,
688 			     buf, buf_size, NULL);
689 	if (ret < 0)
690 		return ret;
691 	if (ret != true_size)
692 		return -EIO;
693 
694 	return true_size;
695 }
696 
697 /*
698  * fsverity op that reads and caches a merkle tree page.
699  *
700  * @inode:         inode to read a merkle tree page for
701  * @index:         page index relative to the start of the merkle tree
702  * @num_ra_pages:  number of pages to readahead. Optional, we ignore it
703  *
704  * The Merkle tree is stored in the filesystem btree, but its pages are cached
705  * with a logical position past EOF in the inode's mapping.
706  *
707  * Returns the page we read, or an ERR_PTR on error.
708  */
709 static struct page *btrfs_read_merkle_tree_page(struct inode *inode,
710 						pgoff_t index,
711 						unsigned long num_ra_pages)
712 {
713 	struct page *page;
714 	u64 off = (u64)index << PAGE_SHIFT;
715 	loff_t merkle_pos = merkle_file_pos(inode);
716 	int ret;
717 
718 	if (merkle_pos < 0)
719 		return ERR_PTR(merkle_pos);
720 	if (merkle_pos > inode->i_sb->s_maxbytes - off - PAGE_SIZE)
721 		return ERR_PTR(-EFBIG);
722 	index += merkle_pos >> PAGE_SHIFT;
723 again:
724 	page = find_get_page_flags(inode->i_mapping, index, FGP_ACCESSED);
725 	if (page) {
726 		if (PageUptodate(page))
727 			return page;
728 
729 		lock_page(page);
730 		/*
731 		 * We only insert uptodate pages, so !Uptodate has to be
732 		 * an error
733 		 */
734 		if (!PageUptodate(page)) {
735 			unlock_page(page);
736 			put_page(page);
737 			return ERR_PTR(-EIO);
738 		}
739 		unlock_page(page);
740 		return page;
741 	}
742 
743 	page = __page_cache_alloc(mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS));
744 	if (!page)
745 		return ERR_PTR(-ENOMEM);
746 
747 	/*
748 	 * Merkle item keys are indexed from byte 0 in the merkle tree.
749 	 * They have the form:
750 	 *
751 	 * [ inode objectid, BTRFS_MERKLE_ITEM_KEY, offset in bytes ]
752 	 */
753 	ret = read_key_bytes(BTRFS_I(inode), BTRFS_VERITY_MERKLE_ITEM_KEY, off,
754 			     page_address(page), PAGE_SIZE, page);
755 	if (ret < 0) {
756 		put_page(page);
757 		return ERR_PTR(ret);
758 	}
759 	if (ret < PAGE_SIZE)
760 		memzero_page(page, ret, PAGE_SIZE - ret);
761 
762 	SetPageUptodate(page);
763 	ret = add_to_page_cache_lru(page, inode->i_mapping, index, GFP_NOFS);
764 
765 	if (!ret) {
766 		/* Inserted and ready for fsverity */
767 		unlock_page(page);
768 	} else {
769 		put_page(page);
770 		/* Did someone race us into inserting this page? */
771 		if (ret == -EEXIST)
772 			goto again;
773 		page = ERR_PTR(ret);
774 	}
775 	return page;
776 }
777 
778 /*
779  * fsverity op that writes a Merkle tree block into the btree.
780  *
781  * @inode:          inode to write a Merkle tree block for
782  * @buf:            Merkle tree data block to write
783  * @index:          index of the block in the Merkle tree
784  * @log_blocksize:  log base 2 of the Merkle tree block size
785  *
786  * Note that the block size could be different from the page size, so it is not
787  * safe to assume that index is a page index.
788  *
789  * Returns 0 on success or negative error code on failure
790  */
791 static int btrfs_write_merkle_tree_block(struct inode *inode, const void *buf,
792 					u64 index, int log_blocksize)
793 {
794 	u64 off = index << log_blocksize;
795 	u64 len = 1ULL << log_blocksize;
796 	loff_t merkle_pos = merkle_file_pos(inode);
797 
798 	if (merkle_pos < 0)
799 		return merkle_pos;
800 	if (merkle_pos > inode->i_sb->s_maxbytes - off - len)
801 		return -EFBIG;
802 
803 	return write_key_bytes(BTRFS_I(inode), BTRFS_VERITY_MERKLE_ITEM_KEY,
804 			       off, buf, len);
805 }
806 
807 const struct fsverity_operations btrfs_verityops = {
808 	.begin_enable_verity     = btrfs_begin_enable_verity,
809 	.end_enable_verity       = btrfs_end_enable_verity,
810 	.get_verity_descriptor   = btrfs_get_verity_descriptor,
811 	.read_merkle_tree_page   = btrfs_read_merkle_tree_page,
812 	.write_merkle_tree_block = btrfs_write_merkle_tree_block,
813 };
814