root-tree.c

  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 */
 26static 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 */
 67int 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 (unlikely(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));
112out:
113	btrfs_release_path(path);
114	return ret;
115}
116
117void 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 */
128int 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	BTRFS_PATH_AUTO_FREE(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		return ret;
147
148	if (unlikely(ret > 0)) {
149		btrfs_crit(fs_info,
150			   "unable to find root key " BTRFS_KEY_FMT " in tree %llu",
151			   BTRFS_KEY_FMT_VALUE(key), btrfs_root_id(root));
152		ret = -EUCLEAN;
153		btrfs_abort_transaction(trans, ret);
154		return ret;
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 (unlikely(ret < 0)) {
172			btrfs_abort_transaction(trans, ret);
173			return ret;
174		}
175
176		ret = btrfs_del_item(trans, root, path);
177		if (unlikely(ret < 0)) {
178			btrfs_abort_transaction(trans, ret);
179			return ret;
180		}
181		btrfs_release_path(path);
182		ret = btrfs_insert_empty_item(trans, root, path,
183				key, sizeof(*item));
184		if (unlikely(ret < 0)) {
185			btrfs_abort_transaction(trans, ret);
186			return ret;
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	return ret;
201}
202
203int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
204		      const struct btrfs_key *key, struct btrfs_root_item *item)
205{
206	/*
207	 * Make sure generation v1 and v2 match. See update_root for details.
208	 */
209	btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
210	return btrfs_insert_item(trans, root, key, item, sizeof(*item));
211}
212
213int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
214{
215	struct btrfs_root *tree_root = fs_info->tree_root;
216	struct extent_buffer *leaf;
217	BTRFS_PATH_AUTO_FREE(path);
218	struct btrfs_key key;
219	struct btrfs_root *root;
220	int err = 0;
221	int ret;
222
223	path = btrfs_alloc_path();
224	if (!path)
225		return -ENOMEM;
226
227	key.objectid = BTRFS_ORPHAN_OBJECTID;
228	key.type = BTRFS_ORPHAN_ITEM_KEY;
229	key.offset = 0;
230
231	while (1) {
232		u64 root_objectid;
233
234		ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
235		if (ret < 0) {
236			err = ret;
237			break;
238		}
239
240		leaf = path->nodes[0];
241		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
242			ret = btrfs_next_leaf(tree_root, path);
243			if (ret < 0)
244				err = ret;
245			if (ret != 0)
246				break;
247			leaf = path->nodes[0];
248		}
249
250		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
251		btrfs_release_path(path);
252
253		if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
254		    key.type != BTRFS_ORPHAN_ITEM_KEY)
255			break;
256
257		root_objectid = key.offset;
258		key.offset++;
259
260		root = btrfs_get_fs_root(fs_info, root_objectid, false);
261		err = PTR_ERR_OR_ZERO(root);
262		if (err && err != -ENOENT) {
263			break;
264		} else if (err == -ENOENT) {
265			struct btrfs_trans_handle *trans;
266
267			btrfs_release_path(path);
268
269			trans = btrfs_join_transaction(tree_root);
270			if (IS_ERR(trans)) {
271				err = PTR_ERR(trans);
272				btrfs_handle_fs_error(fs_info, err,
273					    "Failed to start trans to delete orphan item");
274				break;
275			}
276			err = btrfs_del_orphan_item(trans, tree_root,
277						    root_objectid);
278			btrfs_end_transaction(trans);
279			if (err) {
280				btrfs_handle_fs_error(fs_info, err,
281					    "Failed to delete root orphan item");
282				break;
283			}
284			continue;
285		}
286
287		WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
288		if (btrfs_root_refs(&root->root_item) == 0) {
289			struct btrfs_key drop_key;
290
291			btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
292			/*
293			 * If we have a non-zero drop_progress then we know we
294			 * made it partly through deleting this snapshot, and
295			 * thus we need to make sure we block any balance from
296			 * happening until this snapshot is completely dropped.
297			 */
298			if (drop_key.objectid != 0 || drop_key.type != 0 ||
299			    drop_key.offset != 0) {
300				set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
301				set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
302			}
303
304			set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
305			btrfs_add_dead_root(root);
306		}
307		btrfs_put_root(root);
308	}
309
310	return err;
311}
312
313/* drop the root item for 'key' from the tree root */
314int btrfs_del_root(struct btrfs_trans_handle *trans,
315		   const struct btrfs_key *key)
316{
317	struct btrfs_root *root = trans->fs_info->tree_root;
318	BTRFS_PATH_AUTO_FREE(path);
319	int ret;
320
321	path = btrfs_alloc_path();
322	if (!path)
323		return -ENOMEM;
324	ret = btrfs_search_slot(trans, root, key, path, -1, 1);
325	if (ret < 0)
326		return ret;
327	if (unlikely(ret > 0))
328		/* The root must exist but we did not find it by the key. */
329		return -EUCLEAN;
330
331	return btrfs_del_item(trans, root, path);
332}
333
334int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
335		       u64 ref_id, u64 dirid, u64 *sequence,
336		       const struct fscrypt_str *name)
337{
338	struct btrfs_root *tree_root = trans->fs_info->tree_root;
339	BTRFS_PATH_AUTO_FREE(path);
340	struct btrfs_root_ref *ref;
341	struct extent_buffer *leaf;
342	struct btrfs_key key;
343	unsigned long ptr;
344	int ret;
345
346	path = btrfs_alloc_path();
347	if (!path)
348		return -ENOMEM;
349
350	key.objectid = root_id;
351	key.type = BTRFS_ROOT_BACKREF_KEY;
352	key.offset = ref_id;
353again:
354	ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
355	if (ret < 0) {
356		return ret;
357	} else if (ret == 0) {
358		leaf = path->nodes[0];
359		ref = btrfs_item_ptr(leaf, path->slots[0],
360				     struct btrfs_root_ref);
361		ptr = (unsigned long)(ref + 1);
362		if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
363		    (btrfs_root_ref_name_len(leaf, ref) != name->len) ||
364		    memcmp_extent_buffer(leaf, name->name, ptr, name->len))
365			return -ENOENT;
366
367		*sequence = btrfs_root_ref_sequence(leaf, ref);
368
369		ret = btrfs_del_item(trans, tree_root, path);
370		if (ret)
371			return ret;
372	} else {
373		return -ENOENT;
374	}
375
376	if (key.type == BTRFS_ROOT_BACKREF_KEY) {
377		btrfs_release_path(path);
378		key.objectid = ref_id;
379		key.type = BTRFS_ROOT_REF_KEY;
380		key.offset = root_id;
381		goto again;
382	}
383
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 */
402int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
403		       u64 ref_id, u64 dirid, u64 sequence,
404		       const struct fscrypt_str *name)
405{
406	struct btrfs_root *tree_root = trans->fs_info->tree_root;
407	struct btrfs_key key;
408	int ret;
409	BTRFS_PATH_AUTO_FREE(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;
421again:
422	ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
423				      sizeof(*ref) + name->len);
424	if (unlikely(ret)) {
425		btrfs_abort_transaction(trans, ret);
426		return ret;
427	}
428
429	leaf = path->nodes[0];
430	ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
431	btrfs_set_root_ref_dirid(leaf, ref, dirid);
432	btrfs_set_root_ref_sequence(leaf, ref, sequence);
433	btrfs_set_root_ref_name_len(leaf, ref, name->len);
434	ptr = (unsigned long)(ref + 1);
435	write_extent_buffer(leaf, name->name, ptr, name->len);
436
437	if (key.type == BTRFS_ROOT_BACKREF_KEY) {
438		btrfs_release_path(path);
439		key.objectid = ref_id;
440		key.type = BTRFS_ROOT_REF_KEY;
441		key.offset = root_id;
442		goto again;
443	}
444
445	return 0;
446}
447
448/*
449 * Old btrfs forgets to init root_item->flags and root_item->byte_limit
450 * for subvolumes. To work around this problem, we steal a bit from
451 * root_item->inode_item->flags, and use it to indicate if those fields
452 * have been properly initialized.
453 */
454void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
455{
456	u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
457
458	if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
459		inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
460		btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
461		btrfs_set_root_flags(root_item, 0);
462		btrfs_set_root_limit(root_item, 0);
463	}
464}
465
466void btrfs_update_root_times(struct btrfs_trans_handle *trans,
467			     struct btrfs_root *root)
468{
469	struct btrfs_root_item *item = &root->root_item;
470	struct timespec64 ct;
471
472	ktime_get_real_ts64(&ct);
473	spin_lock(&root->root_item_lock);
474	btrfs_set_root_ctransid(item, trans->transid);
475	btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
476	btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
477	spin_unlock(&root->root_item_lock);
478}
479
480/*
481 * Reserve space for subvolume operation.
482 *
483 * root: the root of the parent directory
484 * rsv: block reservation
485 * items: the number of items that we need do reservation
486 * use_global_rsv: allow fallback to the global block reservation
487 *
488 * This function is used to reserve the space for snapshot/subvolume
489 * creation and deletion. Those operations are different with the
490 * common file/directory operations, they change two fs/file trees
491 * and root tree, the number of items that the qgroup reserves is
492 * different with the free space reservation. So we can not use
493 * the space reservation mechanism in start_transaction().
494 */
495int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
496				     struct btrfs_block_rsv *rsv, int items,
497				     bool use_global_rsv)
498{
499	u64 qgroup_num_bytes = 0;
500	u64 num_bytes;
501	int ret;
502	struct btrfs_fs_info *fs_info = root->fs_info;
503	struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
504
505	if (btrfs_qgroup_enabled(fs_info)) {
506		/* One for parent inode, two for dir entries */
507		qgroup_num_bytes = 3 * fs_info->nodesize;
508		ret = btrfs_qgroup_reserve_meta_prealloc(root,
509							 qgroup_num_bytes, true,
510							 false);
511		if (ret)
512			return ret;
513	}
514
515	num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
516	rsv->space_info = btrfs_find_space_info(fs_info,
517					    BTRFS_BLOCK_GROUP_METADATA);
518	ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes,
519				  BTRFS_RESERVE_FLUSH_ALL);
520
521	if (ret == -ENOSPC && use_global_rsv)
522		ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
523
524	if (ret && qgroup_num_bytes)
525		btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
526
527	if (!ret) {
528		spin_lock(&rsv->lock);
529		rsv->qgroup_rsv_reserved += qgroup_num_bytes;
530		spin_unlock(&rsv->lock);
531	}
532	return ret;
533}