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