xref: /linux/fs/btrfs/root-tree.c (revision e094f480)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2007 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/err.h>
7 #include <linux/uuid.h>
8 #include "ctree.h"
9 #include "fs.h"
10 #include "messages.h"
11 #include "transaction.h"
12 #include "disk-io.h"
13 #include "qgroup.h"
14 #include "space-info.h"
15 #include "accessors.h"
16 #include "root-tree.h"
17 #include "orphan.h"
18 
19 /*
20  * Read a root item from the tree. In case we detect a root item smaller then
21  * sizeof(root_item), we know it's an old version of the root structure and
22  * initialize all new fields to zero. The same happens if we detect mismatching
23  * generation numbers as then we know the root was once mounted with an older
24  * kernel that was not aware of the root item structure change.
25  */
btrfs_read_root_item(struct extent_buffer * eb,int slot,struct btrfs_root_item * item)26 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
27 				struct btrfs_root_item *item)
28 {
29 	u32 len;
30 	int need_reset = 0;
31 
32 	len = btrfs_item_size(eb, slot);
33 	read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
34 			   min_t(u32, len, sizeof(*item)));
35 	if (len < sizeof(*item))
36 		need_reset = 1;
37 	if (!need_reset && btrfs_root_generation(item)
38 		!= btrfs_root_generation_v2(item)) {
39 		if (btrfs_root_generation_v2(item) != 0) {
40 			btrfs_warn(eb->fs_info,
41 					"mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
42 		}
43 		need_reset = 1;
44 	}
45 	if (need_reset) {
46 		/* Clear all members from generation_v2 onwards. */
47 		memset_startat(item, 0, generation_v2);
48 		generate_random_guid(item->uuid);
49 	}
50 }
51 
52 /*
53  * Lookup the root by the key.
54  *
55  * root: the root of the root tree
56  * search_key: the key to search
57  * path: the path we search
58  * root_item: the root item of the tree we look for
59  * root_key: the root key of the tree we look for
60  *
61  * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
62  * of the search key, just lookup the root with the highest offset for a
63  * given objectid.
64  *
65  * If we find something return 0, otherwise > 0, < 0 on error.
66  */
btrfs_find_root(struct btrfs_root * root,const struct btrfs_key * search_key,struct btrfs_path * path,struct btrfs_root_item * root_item,struct btrfs_key * root_key)67 int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
68 		    struct btrfs_path *path, struct btrfs_root_item *root_item,
69 		    struct btrfs_key *root_key)
70 {
71 	struct btrfs_key found_key;
72 	struct extent_buffer *l;
73 	int ret;
74 	int slot;
75 
76 	ret = btrfs_search_slot(NULL, root, search_key, path, 0, 0);
77 	if (ret < 0)
78 		return ret;
79 
80 	if (search_key->offset != -1ULL) {	/* the search key is exact */
81 		if (ret > 0)
82 			goto out;
83 	} else {
84 		/*
85 		 * Key with offset -1 found, there would have to exist a root
86 		 * with such id, but this is out of the valid range.
87 		 */
88 		if (ret == 0) {
89 			ret = -EUCLEAN;
90 			goto out;
91 		}
92 		if (path->slots[0] == 0)
93 			goto out;
94 		path->slots[0]--;
95 		ret = 0;
96 	}
97 
98 	l = path->nodes[0];
99 	slot = path->slots[0];
100 
101 	btrfs_item_key_to_cpu(l, &found_key, slot);
102 	if (found_key.objectid != search_key->objectid ||
103 	    found_key.type != BTRFS_ROOT_ITEM_KEY) {
104 		ret = 1;
105 		goto out;
106 	}
107 
108 	if (root_item)
109 		btrfs_read_root_item(l, slot, root_item);
110 	if (root_key)
111 		memcpy(root_key, &found_key, sizeof(found_key));
112 out:
113 	btrfs_release_path(path);
114 	return ret;
115 }
116 
btrfs_set_root_node(struct btrfs_root_item * item,struct extent_buffer * node)117 void btrfs_set_root_node(struct btrfs_root_item *item,
118 			 struct extent_buffer *node)
119 {
120 	btrfs_set_root_bytenr(item, node->start);
121 	btrfs_set_root_level(item, btrfs_header_level(node));
122 	btrfs_set_root_generation(item, btrfs_header_generation(node));
123 }
124 
125 /*
126  * copy the data in 'item' into the btree
127  */
btrfs_update_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,struct btrfs_key * key,struct btrfs_root_item * item)128 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
129 		      *root, struct btrfs_key *key, struct btrfs_root_item
130 		      *item)
131 {
132 	struct btrfs_fs_info *fs_info = root->fs_info;
133 	struct btrfs_path *path;
134 	struct extent_buffer *l;
135 	int ret;
136 	int slot;
137 	unsigned long ptr;
138 	u32 old_len;
139 
140 	path = btrfs_alloc_path();
141 	if (!path)
142 		return -ENOMEM;
143 
144 	ret = btrfs_search_slot(trans, root, key, path, 0, 1);
145 	if (ret < 0)
146 		goto out;
147 
148 	if (ret > 0) {
149 		btrfs_crit(fs_info,
150 			"unable to find root key (%llu %u %llu) in tree %llu",
151 			key->objectid, key->type, key->offset, btrfs_root_id(root));
152 		ret = -EUCLEAN;
153 		btrfs_abort_transaction(trans, ret);
154 		goto out;
155 	}
156 
157 	l = path->nodes[0];
158 	slot = path->slots[0];
159 	ptr = btrfs_item_ptr_offset(l, slot);
160 	old_len = btrfs_item_size(l, slot);
161 
162 	/*
163 	 * If this is the first time we update the root item which originated
164 	 * from an older kernel, we need to enlarge the item size to make room
165 	 * for the added fields.
166 	 */
167 	if (old_len < sizeof(*item)) {
168 		btrfs_release_path(path);
169 		ret = btrfs_search_slot(trans, root, key, path,
170 				-1, 1);
171 		if (ret < 0) {
172 			btrfs_abort_transaction(trans, ret);
173 			goto out;
174 		}
175 
176 		ret = btrfs_del_item(trans, root, path);
177 		if (ret < 0) {
178 			btrfs_abort_transaction(trans, ret);
179 			goto out;
180 		}
181 		btrfs_release_path(path);
182 		ret = btrfs_insert_empty_item(trans, root, path,
183 				key, sizeof(*item));
184 		if (ret < 0) {
185 			btrfs_abort_transaction(trans, ret);
186 			goto out;
187 		}
188 		l = path->nodes[0];
189 		slot = path->slots[0];
190 		ptr = btrfs_item_ptr_offset(l, slot);
191 	}
192 
193 	/*
194 	 * Update generation_v2 so at the next mount we know the new root
195 	 * fields are valid.
196 	 */
197 	btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
198 
199 	write_extent_buffer(l, item, ptr, sizeof(*item));
200 	btrfs_mark_buffer_dirty(trans, path->nodes[0]);
201 out:
202 	btrfs_free_path(path);
203 	return ret;
204 }
205 
btrfs_insert_root(struct btrfs_trans_handle * trans,struct btrfs_root * root,const struct btrfs_key * key,struct btrfs_root_item * item)206 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
207 		      const struct btrfs_key *key, struct btrfs_root_item *item)
208 {
209 	/*
210 	 * Make sure generation v1 and v2 match. See update_root for details.
211 	 */
212 	btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
213 	return btrfs_insert_item(trans, root, key, item, sizeof(*item));
214 }
215 
btrfs_find_orphan_roots(struct btrfs_fs_info * fs_info)216 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
217 {
218 	struct btrfs_root *tree_root = fs_info->tree_root;
219 	struct extent_buffer *leaf;
220 	struct btrfs_path *path;
221 	struct btrfs_key key;
222 	struct btrfs_root *root;
223 	int err = 0;
224 	int ret;
225 
226 	path = btrfs_alloc_path();
227 	if (!path)
228 		return -ENOMEM;
229 
230 	key.objectid = BTRFS_ORPHAN_OBJECTID;
231 	key.type = BTRFS_ORPHAN_ITEM_KEY;
232 	key.offset = 0;
233 
234 	while (1) {
235 		u64 root_objectid;
236 
237 		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
238 		if (ret < 0) {
239 			err = ret;
240 			break;
241 		}
242 
243 		leaf = path->nodes[0];
244 		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
245 			ret = btrfs_next_leaf(tree_root, path);
246 			if (ret < 0)
247 				err = ret;
248 			if (ret != 0)
249 				break;
250 			leaf = path->nodes[0];
251 		}
252 
253 		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
254 		btrfs_release_path(path);
255 
256 		if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
257 		    key.type != BTRFS_ORPHAN_ITEM_KEY)
258 			break;
259 
260 		root_objectid = key.offset;
261 		key.offset++;
262 
263 		root = btrfs_get_fs_root(fs_info, root_objectid, false);
264 		err = PTR_ERR_OR_ZERO(root);
265 		if (err && err != -ENOENT) {
266 			break;
267 		} else if (err == -ENOENT) {
268 			struct btrfs_trans_handle *trans;
269 
270 			btrfs_release_path(path);
271 
272 			trans = btrfs_join_transaction(tree_root);
273 			if (IS_ERR(trans)) {
274 				err = PTR_ERR(trans);
275 				btrfs_handle_fs_error(fs_info, err,
276 					    "Failed to start trans to delete orphan item");
277 				break;
278 			}
279 			err = btrfs_del_orphan_item(trans, tree_root,
280 						    root_objectid);
281 			btrfs_end_transaction(trans);
282 			if (err) {
283 				btrfs_handle_fs_error(fs_info, err,
284 					    "Failed to delete root orphan item");
285 				break;
286 			}
287 			continue;
288 		}
289 
290 		WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
291 		if (btrfs_root_refs(&root->root_item) == 0) {
292 			struct btrfs_key drop_key;
293 
294 			btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
295 			/*
296 			 * If we have a non-zero drop_progress then we know we
297 			 * made it partly through deleting this snapshot, and
298 			 * thus we need to make sure we block any balance from
299 			 * happening until this snapshot is completely dropped.
300 			 */
301 			if (drop_key.objectid != 0 || drop_key.type != 0 ||
302 			    drop_key.offset != 0) {
303 				set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
304 				set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
305 			}
306 
307 			set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
308 			btrfs_add_dead_root(root);
309 		}
310 		btrfs_put_root(root);
311 	}
312 
313 	btrfs_free_path(path);
314 	return err;
315 }
316 
317 /* drop the root item for 'key' from the tree root */
btrfs_del_root(struct btrfs_trans_handle * trans,const struct btrfs_key * key)318 int btrfs_del_root(struct btrfs_trans_handle *trans,
319 		   const struct btrfs_key *key)
320 {
321 	struct btrfs_root *root = trans->fs_info->tree_root;
322 	struct btrfs_path *path;
323 	int ret;
324 
325 	path = btrfs_alloc_path();
326 	if (!path)
327 		return -ENOMEM;
328 	ret = btrfs_search_slot(trans, root, key, path, -1, 1);
329 	if (ret < 0)
330 		goto out;
331 	if (ret != 0) {
332 		/* The root must exist but we did not find it by the key. */
333 		ret = -EUCLEAN;
334 		goto out;
335 	}
336 
337 	ret = btrfs_del_item(trans, root, path);
338 out:
339 	btrfs_free_path(path);
340 	return ret;
341 }
342 
btrfs_del_root_ref(struct btrfs_trans_handle * trans,u64 root_id,u64 ref_id,u64 dirid,u64 * sequence,const struct fscrypt_str * name)343 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
344 		       u64 ref_id, u64 dirid, u64 *sequence,
345 		       const struct fscrypt_str *name)
346 {
347 	struct btrfs_root *tree_root = trans->fs_info->tree_root;
348 	struct btrfs_path *path;
349 	struct btrfs_root_ref *ref;
350 	struct extent_buffer *leaf;
351 	struct btrfs_key key;
352 	unsigned long ptr;
353 	int ret;
354 
355 	path = btrfs_alloc_path();
356 	if (!path)
357 		return -ENOMEM;
358 
359 	key.objectid = root_id;
360 	key.type = BTRFS_ROOT_BACKREF_KEY;
361 	key.offset = ref_id;
362 again:
363 	ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
364 	if (ret < 0) {
365 		goto out;
366 	} else if (ret == 0) {
367 		leaf = path->nodes[0];
368 		ref = btrfs_item_ptr(leaf, path->slots[0],
369 				     struct btrfs_root_ref);
370 		ptr = (unsigned long)(ref + 1);
371 		if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
372 		    (btrfs_root_ref_name_len(leaf, ref) != name->len) ||
373 		    memcmp_extent_buffer(leaf, name->name, ptr, name->len)) {
374 			ret = -ENOENT;
375 			goto out;
376 		}
377 		*sequence = btrfs_root_ref_sequence(leaf, ref);
378 
379 		ret = btrfs_del_item(trans, tree_root, path);
380 		if (ret)
381 			goto out;
382 	} else {
383 		ret = -ENOENT;
384 		goto out;
385 	}
386 
387 	if (key.type == BTRFS_ROOT_BACKREF_KEY) {
388 		btrfs_release_path(path);
389 		key.objectid = ref_id;
390 		key.type = BTRFS_ROOT_REF_KEY;
391 		key.offset = root_id;
392 		goto again;
393 	}
394 
395 out:
396 	btrfs_free_path(path);
397 	return ret;
398 }
399 
400 /*
401  * add a btrfs_root_ref item.  type is either BTRFS_ROOT_REF_KEY
402  * or BTRFS_ROOT_BACKREF_KEY.
403  *
404  * The dirid, sequence, name and name_len refer to the directory entry
405  * that is referencing the root.
406  *
407  * For a forward ref, the root_id is the id of the tree referencing
408  * the root and ref_id is the id of the subvol  or snapshot.
409  *
410  * For a back ref the root_id is the id of the subvol or snapshot and
411  * ref_id is the id of the tree referencing it.
412  *
413  * Will return 0, -ENOMEM, or anything from the CoW path
414  */
btrfs_add_root_ref(struct btrfs_trans_handle * trans,u64 root_id,u64 ref_id,u64 dirid,u64 sequence,const struct fscrypt_str * name)415 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
416 		       u64 ref_id, u64 dirid, u64 sequence,
417 		       const struct fscrypt_str *name)
418 {
419 	struct btrfs_root *tree_root = trans->fs_info->tree_root;
420 	struct btrfs_key key;
421 	int ret;
422 	struct btrfs_path *path;
423 	struct btrfs_root_ref *ref;
424 	struct extent_buffer *leaf;
425 	unsigned long ptr;
426 
427 	path = btrfs_alloc_path();
428 	if (!path)
429 		return -ENOMEM;
430 
431 	key.objectid = root_id;
432 	key.type = BTRFS_ROOT_BACKREF_KEY;
433 	key.offset = ref_id;
434 again:
435 	ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
436 				      sizeof(*ref) + name->len);
437 	if (ret) {
438 		btrfs_abort_transaction(trans, ret);
439 		btrfs_free_path(path);
440 		return ret;
441 	}
442 
443 	leaf = path->nodes[0];
444 	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
445 	btrfs_set_root_ref_dirid(leaf, ref, dirid);
446 	btrfs_set_root_ref_sequence(leaf, ref, sequence);
447 	btrfs_set_root_ref_name_len(leaf, ref, name->len);
448 	ptr = (unsigned long)(ref + 1);
449 	write_extent_buffer(leaf, name->name, ptr, name->len);
450 	btrfs_mark_buffer_dirty(trans, leaf);
451 
452 	if (key.type == BTRFS_ROOT_BACKREF_KEY) {
453 		btrfs_release_path(path);
454 		key.objectid = ref_id;
455 		key.type = BTRFS_ROOT_REF_KEY;
456 		key.offset = root_id;
457 		goto again;
458 	}
459 
460 	btrfs_free_path(path);
461 	return 0;
462 }
463 
464 /*
465  * Old btrfs forgets to init root_item->flags and root_item->byte_limit
466  * for subvolumes. To work around this problem, we steal a bit from
467  * root_item->inode_item->flags, and use it to indicate if those fields
468  * have been properly initialized.
469  */
btrfs_check_and_init_root_item(struct btrfs_root_item * root_item)470 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
471 {
472 	u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
473 
474 	if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
475 		inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
476 		btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
477 		btrfs_set_root_flags(root_item, 0);
478 		btrfs_set_root_limit(root_item, 0);
479 	}
480 }
481 
btrfs_update_root_times(struct btrfs_trans_handle * trans,struct btrfs_root * root)482 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
483 			     struct btrfs_root *root)
484 {
485 	struct btrfs_root_item *item = &root->root_item;
486 	struct timespec64 ct;
487 
488 	ktime_get_real_ts64(&ct);
489 	spin_lock(&root->root_item_lock);
490 	btrfs_set_root_ctransid(item, trans->transid);
491 	btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
492 	btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
493 	spin_unlock(&root->root_item_lock);
494 }
495 
496 /*
497  * Reserve space for subvolume operation.
498  *
499  * root: the root of the parent directory
500  * rsv: block reservation
501  * items: the number of items that we need do reservation
502  * use_global_rsv: allow fallback to the global block reservation
503  *
504  * This function is used to reserve the space for snapshot/subvolume
505  * creation and deletion. Those operations are different with the
506  * common file/directory operations, they change two fs/file trees
507  * and root tree, the number of items that the qgroup reserves is
508  * different with the free space reservation. So we can not use
509  * the space reservation mechanism in start_transaction().
510  */
btrfs_subvolume_reserve_metadata(struct btrfs_root * root,struct btrfs_block_rsv * rsv,int items,bool use_global_rsv)511 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
512 				     struct btrfs_block_rsv *rsv, int items,
513 				     bool use_global_rsv)
514 {
515 	u64 qgroup_num_bytes = 0;
516 	u64 num_bytes;
517 	int ret;
518 	struct btrfs_fs_info *fs_info = root->fs_info;
519 	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
520 
521 	if (btrfs_qgroup_enabled(fs_info)) {
522 		/* One for parent inode, two for dir entries */
523 		qgroup_num_bytes = 3 * fs_info->nodesize;
524 		ret = btrfs_qgroup_reserve_meta_prealloc(root,
525 							 qgroup_num_bytes, true,
526 							 false);
527 		if (ret)
528 			return ret;
529 	}
530 
531 	num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
532 	rsv->space_info = btrfs_find_space_info(fs_info,
533 					    BTRFS_BLOCK_GROUP_METADATA);
534 	ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes,
535 				  BTRFS_RESERVE_FLUSH_ALL);
536 
537 	if (ret == -ENOSPC && use_global_rsv)
538 		ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
539 
540 	if (ret && qgroup_num_bytes)
541 		btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
542 
543 	if (!ret) {
544 		spin_lock(&rsv->lock);
545 		rsv->qgroup_rsv_reserved += qgroup_num_bytes;
546 		spin_unlock(&rsv->lock);
547 	}
548 	return ret;
549 }
550