fs/btrfs/root-tree.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * Copyright (C) 2007 Oracle.  All rights reserved.
0004  */
0005
0006 #include <linux/err.h>
0007 #include <linux/uuid.h>
0008 #include "ctree.h"
0009 #include "transaction.h"
0010 #include "disk-io.h"
0011 #include "print-tree.h"
0012 #include "qgroup.h"
0013 #include "space-info.h"
0014
0015 /*
0016  * Read a root item from the tree. In case we detect a root item smaller then
0017  * sizeof(root_item), we know it's an old version of the root structure and
0018  * initialize all new fields to zero. The same happens if we detect mismatching
0019  * generation numbers as then we know the root was once mounted with an older
0020  * kernel that was not aware of the root item structure change.
0021  */
0022 static void btrfs_read_root_item(struct extent_buffer *eb, int slot,
0023                 struct btrfs_root_item *item)
0024 {
0025     u32 len;
0026     int need_reset = 0;
0027
0028     len = btrfs_item_size(eb, slot);
0029     read_extent_buffer(eb, item, btrfs_item_ptr_offset(eb, slot),
0030                min_t(u32, len, sizeof(*item)));
0031     if (len < sizeof(*item))
0032         need_reset = 1;
0033     if (!need_reset && btrfs_root_generation(item)
0034         != btrfs_root_generation_v2(item)) {
0035         if (btrfs_root_generation_v2(item) != 0) {
0036             btrfs_warn(eb->fs_info,
0037                     "mismatching generation and generation_v2 found in root item. This root was probably mounted with an older kernel. Resetting all new fields.");
0038         }
0039         need_reset = 1;
0040     }
0041     if (need_reset) {
0042         /* Clear all members from generation_v2 onwards. */
0043         memset_startat(item, 0, generation_v2);
0044         generate_random_guid(item->uuid);
0045     }
0046 }
0047
0048 /*
0049  * btrfs_find_root - lookup the root by the key.
0050  * root: the root of the root tree
0051  * search_key: the key to search
0052  * path: the path we search
0053  * root_item: the root item of the tree we look for
0054  * root_key: the root key of the tree we look for
0055  *
0056  * If ->offset of 'search_key' is -1ULL, it means we are not sure the offset
0057  * of the search key, just lookup the root with the highest offset for a
0058  * given objectid.
0059  *
0060  * If we find something return 0, otherwise > 0, < 0 on error.
0061  */
0062 int btrfs_find_root(struct btrfs_root *root, const struct btrfs_key *search_key,
0063             struct btrfs_path *path, struct btrfs_root_item *root_item,
0064             struct btrfs_key *root_key)
0065 {
0066     struct btrfs_key found_key;
0067     struct extent_buffer *l;
0068     int ret;
0069     int slot;
0070
0071     ret = btrfs_search_slot(NULL, root, search_key, path, 0, 0);
0072     if (ret < 0)
0073         return ret;
0074
0075     if (search_key->offset != -1ULL) {  /* the search key is exact */
0076         if (ret > 0)
0077             goto out;
0078     } else {
0079         BUG_ON(ret == 0);       /* Logical error */
0080         if (path->slots[0] == 0)
0081             goto out;
0082         path->slots[0]--;
0083         ret = 0;
0084     }
0085
0086     l = path->nodes[0];
0087     slot = path->slots[0];
0088
0089     btrfs_item_key_to_cpu(l, &found_key, slot);
0090     if (found_key.objectid != search_key->objectid ||
0091         found_key.type != BTRFS_ROOT_ITEM_KEY) {
0092         ret = 1;
0093         goto out;
0094     }
0095
0096     if (root_item)
0097         btrfs_read_root_item(l, slot, root_item);
0098     if (root_key)
0099         memcpy(root_key, &found_key, sizeof(found_key));
0100 out:
0101     btrfs_release_path(path);
0102     return ret;
0103 }
0104
0105 void btrfs_set_root_node(struct btrfs_root_item *item,
0106              struct extent_buffer *node)
0107 {
0108     btrfs_set_root_bytenr(item, node->start);
0109     btrfs_set_root_level(item, btrfs_header_level(node));
0110     btrfs_set_root_generation(item, btrfs_header_generation(node));
0111 }
0112
0113 /*
0114  * copy the data in 'item' into the btree
0115  */
0116 int btrfs_update_root(struct btrfs_trans_handle *trans, struct btrfs_root
0117               *root, struct btrfs_key *key, struct btrfs_root_item
0118               *item)
0119 {
0120     struct btrfs_fs_info *fs_info = root->fs_info;
0121     struct btrfs_path *path;
0122     struct extent_buffer *l;
0123     int ret;
0124     int slot;
0125     unsigned long ptr;
0126     u32 old_len;
0127
0128     path = btrfs_alloc_path();
0129     if (!path)
0130         return -ENOMEM;
0131
0132     ret = btrfs_search_slot(trans, root, key, path, 0, 1);
0133     if (ret < 0)
0134         goto out;
0135
0136     if (ret > 0) {
0137         btrfs_crit(fs_info,
0138             "unable to find root key (%llu %u %llu) in tree %llu",
0139             key->objectid, key->type, key->offset,
0140             root->root_key.objectid);
0141         ret = -EUCLEAN;
0142         btrfs_abort_transaction(trans, ret);
0143         goto out;
0144     }
0145
0146     l = path->nodes[0];
0147     slot = path->slots[0];
0148     ptr = btrfs_item_ptr_offset(l, slot);
0149     old_len = btrfs_item_size(l, slot);
0150
0151     /*
0152      * If this is the first time we update the root item which originated
0153      * from an older kernel, we need to enlarge the item size to make room
0154      * for the added fields.
0155      */
0156     if (old_len < sizeof(*item)) {
0157         btrfs_release_path(path);
0158         ret = btrfs_search_slot(trans, root, key, path,
0159                 -1, 1);
0160         if (ret < 0) {
0161             btrfs_abort_transaction(trans, ret);
0162             goto out;
0163         }
0164
0165         ret = btrfs_del_item(trans, root, path);
0166         if (ret < 0) {
0167             btrfs_abort_transaction(trans, ret);
0168             goto out;
0169         }
0170         btrfs_release_path(path);
0171         ret = btrfs_insert_empty_item(trans, root, path,
0172                 key, sizeof(*item));
0173         if (ret < 0) {
0174             btrfs_abort_transaction(trans, ret);
0175             goto out;
0176         }
0177         l = path->nodes[0];
0178         slot = path->slots[0];
0179         ptr = btrfs_item_ptr_offset(l, slot);
0180     }
0181
0182     /*
0183      * Update generation_v2 so at the next mount we know the new root
0184      * fields are valid.
0185      */
0186     btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
0187
0188     write_extent_buffer(l, item, ptr, sizeof(*item));
0189     btrfs_mark_buffer_dirty(path->nodes[0]);
0190 out:
0191     btrfs_free_path(path);
0192     return ret;
0193 }
0194
0195 int btrfs_insert_root(struct btrfs_trans_handle *trans, struct btrfs_root *root,
0196               const struct btrfs_key *key, struct btrfs_root_item *item)
0197 {
0198     /*
0199      * Make sure generation v1 and v2 match. See update_root for details.
0200      */
0201     btrfs_set_root_generation_v2(item, btrfs_root_generation(item));
0202     return btrfs_insert_item(trans, root, key, item, sizeof(*item));
0203 }
0204
0205 int btrfs_find_orphan_roots(struct btrfs_fs_info *fs_info)
0206 {
0207     struct btrfs_root *tree_root = fs_info->tree_root;
0208     struct extent_buffer *leaf;
0209     struct btrfs_path *path;
0210     struct btrfs_key key;
0211     struct btrfs_root *root;
0212     int err = 0;
0213     int ret;
0214
0215     path = btrfs_alloc_path();
0216     if (!path)
0217         return -ENOMEM;
0218
0219     key.objectid = BTRFS_ORPHAN_OBJECTID;
0220     key.type = BTRFS_ORPHAN_ITEM_KEY;
0221     key.offset = 0;
0222
0223     while (1) {
0224         u64 root_objectid;
0225
0226         ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
0227         if (ret < 0) {
0228             err = ret;
0229             break;
0230         }
0231
0232         leaf = path->nodes[0];
0233         if (path->slots[0] >= btrfs_header_nritems(leaf)) {
0234             ret = btrfs_next_leaf(tree_root, path);
0235             if (ret < 0)
0236                 err = ret;
0237             if (ret != 0)
0238                 break;
0239             leaf = path->nodes[0];
0240         }
0241
0242         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
0243         btrfs_release_path(path);
0244
0245         if (key.objectid != BTRFS_ORPHAN_OBJECTID ||
0246             key.type != BTRFS_ORPHAN_ITEM_KEY)
0247             break;
0248
0249         root_objectid = key.offset;
0250         key.offset++;
0251
0252         root = btrfs_get_fs_root(fs_info, root_objectid, false);
0253         err = PTR_ERR_OR_ZERO(root);
0254         if (err && err != -ENOENT) {
0255             break;
0256         } else if (err == -ENOENT) {
0257             struct btrfs_trans_handle *trans;
0258
0259             btrfs_release_path(path);
0260
0261             trans = btrfs_join_transaction(tree_root);
0262             if (IS_ERR(trans)) {
0263                 err = PTR_ERR(trans);
0264                 btrfs_handle_fs_error(fs_info, err,
0265                         "Failed to start trans to delete orphan item");
0266                 break;
0267             }
0268             err = btrfs_del_orphan_item(trans, tree_root,
0269                             root_objectid);
0270             btrfs_end_transaction(trans);
0271             if (err) {
0272                 btrfs_handle_fs_error(fs_info, err,
0273                         "Failed to delete root orphan item");
0274                 break;
0275             }
0276             continue;
0277         }
0278
0279         WARN_ON(!test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state));
0280         if (btrfs_root_refs(&root->root_item) == 0) {
0281             struct btrfs_key drop_key;
0282
0283             btrfs_disk_key_to_cpu(&drop_key, &root->root_item.drop_progress);
0284             /*
0285              * If we have a non-zero drop_progress then we know we
0286              * made it partly through deleting this snapshot, and
0287              * thus we need to make sure we block any balance from
0288              * happening until this snapshot is completely dropped.
0289              */
0290             if (drop_key.objectid != 0 || drop_key.type != 0 ||
0291                 drop_key.offset != 0) {
0292                 set_bit(BTRFS_FS_UNFINISHED_DROPS, &fs_info->flags);
0293                 set_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state);
0294             }
0295
0296             set_bit(BTRFS_ROOT_DEAD_TREE, &root->state);
0297             btrfs_add_dead_root(root);
0298         }
0299         btrfs_put_root(root);
0300     }
0301
0302     btrfs_free_path(path);
0303     return err;
0304 }
0305
0306 /* drop the root item for 'key' from the tree root */
0307 int btrfs_del_root(struct btrfs_trans_handle *trans,
0308            const struct btrfs_key *key)
0309 {
0310     struct btrfs_root *root = trans->fs_info->tree_root;
0311     struct btrfs_path *path;
0312     int ret;
0313
0314     path = btrfs_alloc_path();
0315     if (!path)
0316         return -ENOMEM;
0317     ret = btrfs_search_slot(trans, root, key, path, -1, 1);
0318     if (ret < 0)
0319         goto out;
0320
0321     BUG_ON(ret != 0);
0322
0323     ret = btrfs_del_item(trans, root, path);
0324 out:
0325     btrfs_free_path(path);
0326     return ret;
0327 }
0328
0329 int btrfs_del_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
0330                u64 ref_id, u64 dirid, u64 *sequence, const char *name,
0331                int name_len)
0332
0333 {
0334     struct btrfs_root *tree_root = trans->fs_info->tree_root;
0335     struct btrfs_path *path;
0336     struct btrfs_root_ref *ref;
0337     struct extent_buffer *leaf;
0338     struct btrfs_key key;
0339     unsigned long ptr;
0340     int err = 0;
0341     int ret;
0342
0343     path = btrfs_alloc_path();
0344     if (!path)
0345         return -ENOMEM;
0346
0347     key.objectid = root_id;
0348     key.type = BTRFS_ROOT_BACKREF_KEY;
0349     key.offset = ref_id;
0350 again:
0351     ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
0352     if (ret < 0) {
0353         err = ret;
0354         goto out;
0355     } else if (ret == 0) {
0356         leaf = path->nodes[0];
0357         ref = btrfs_item_ptr(leaf, path->slots[0],
0358                      struct btrfs_root_ref);
0359         ptr = (unsigned long)(ref + 1);
0360         if ((btrfs_root_ref_dirid(leaf, ref) != dirid) ||
0361             (btrfs_root_ref_name_len(leaf, ref) != name_len) ||
0362             memcmp_extent_buffer(leaf, name, ptr, name_len)) {
0363             err = -ENOENT;
0364             goto out;
0365         }
0366         *sequence = btrfs_root_ref_sequence(leaf, ref);
0367
0368         ret = btrfs_del_item(trans, tree_root, path);
0369         if (ret) {
0370             err = ret;
0371             goto out;
0372         }
0373     } else
0374         err = -ENOENT;
0375
0376     if (key.type == BTRFS_ROOT_BACKREF_KEY) {
0377         btrfs_release_path(path);
0378         key.objectid = ref_id;
0379         key.type = BTRFS_ROOT_REF_KEY;
0380         key.offset = root_id;
0381         goto again;
0382     }
0383
0384 out:
0385     btrfs_free_path(path);
0386     return err;
0387 }
0388
0389 /*
0390  * add a btrfs_root_ref item.  type is either BTRFS_ROOT_REF_KEY
0391  * or BTRFS_ROOT_BACKREF_KEY.
0392  *
0393  * The dirid, sequence, name and name_len refer to the directory entry
0394  * that is referencing the root.
0395  *
0396  * For a forward ref, the root_id is the id of the tree referencing
0397  * the root and ref_id is the id of the subvol  or snapshot.
0398  *
0399  * For a back ref the root_id is the id of the subvol or snapshot and
0400  * ref_id is the id of the tree referencing it.
0401  *
0402  * Will return 0, -ENOMEM, or anything from the CoW path
0403  */
0404 int btrfs_add_root_ref(struct btrfs_trans_handle *trans, u64 root_id,
0405                u64 ref_id, u64 dirid, u64 sequence, const char *name,
0406                int name_len)
0407 {
0408     struct btrfs_root *tree_root = trans->fs_info->tree_root;
0409     struct btrfs_key key;
0410     int ret;
0411     struct btrfs_path *path;
0412     struct btrfs_root_ref *ref;
0413     struct extent_buffer *leaf;
0414     unsigned long ptr;
0415
0416     path = btrfs_alloc_path();
0417     if (!path)
0418         return -ENOMEM;
0419
0420     key.objectid = root_id;
0421     key.type = BTRFS_ROOT_BACKREF_KEY;
0422     key.offset = ref_id;
0423 again:
0424     ret = btrfs_insert_empty_item(trans, tree_root, path, &key,
0425                       sizeof(*ref) + name_len);
0426     if (ret) {
0427         btrfs_abort_transaction(trans, ret);
0428         btrfs_free_path(path);
0429         return ret;
0430     }
0431
0432     leaf = path->nodes[0];
0433     ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
0434     btrfs_set_root_ref_dirid(leaf, ref, dirid);
0435     btrfs_set_root_ref_sequence(leaf, ref, sequence);
0436     btrfs_set_root_ref_name_len(leaf, ref, name_len);
0437     ptr = (unsigned long)(ref + 1);
0438     write_extent_buffer(leaf, name, ptr, name_len);
0439     btrfs_mark_buffer_dirty(leaf);
0440
0441     if (key.type == BTRFS_ROOT_BACKREF_KEY) {
0442         btrfs_release_path(path);
0443         key.objectid = ref_id;
0444         key.type = BTRFS_ROOT_REF_KEY;
0445         key.offset = root_id;
0446         goto again;
0447     }
0448
0449     btrfs_free_path(path);
0450     return 0;
0451 }
0452
0453 /*
0454  * Old btrfs forgets to init root_item->flags and root_item->byte_limit
0455  * for subvolumes. To work around this problem, we steal a bit from
0456  * root_item->inode_item->flags, and use it to indicate if those fields
0457  * have been properly initialized.
0458  */
0459 void btrfs_check_and_init_root_item(struct btrfs_root_item *root_item)
0460 {
0461     u64 inode_flags = btrfs_stack_inode_flags(&root_item->inode);
0462
0463     if (!(inode_flags & BTRFS_INODE_ROOT_ITEM_INIT)) {
0464         inode_flags |= BTRFS_INODE_ROOT_ITEM_INIT;
0465         btrfs_set_stack_inode_flags(&root_item->inode, inode_flags);
0466         btrfs_set_root_flags(root_item, 0);
0467         btrfs_set_root_limit(root_item, 0);
0468     }
0469 }
0470
0471 void btrfs_update_root_times(struct btrfs_trans_handle *trans,
0472                  struct btrfs_root *root)
0473 {
0474     struct btrfs_root_item *item = &root->root_item;
0475     struct timespec64 ct;
0476
0477     ktime_get_real_ts64(&ct);
0478     spin_lock(&root->root_item_lock);
0479     btrfs_set_root_ctransid(item, trans->transid);
0480     btrfs_set_stack_timespec_sec(&item->ctime, ct.tv_sec);
0481     btrfs_set_stack_timespec_nsec(&item->ctime, ct.tv_nsec);
0482     spin_unlock(&root->root_item_lock);
0483 }
0484
0485 /*
0486  * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
0487  * root: the root of the parent directory
0488  * rsv: block reservation
0489  * items: the number of items that we need do reservation
0490  * use_global_rsv: allow fallback to the global block reservation
0491  *
0492  * This function is used to reserve the space for snapshot/subvolume
0493  * creation and deletion. Those operations are different with the
0494  * common file/directory operations, they change two fs/file trees
0495  * and root tree, the number of items that the qgroup reserves is
0496  * different with the free space reservation. So we can not use
0497  * the space reservation mechanism in start_transaction().
0498  */
0499 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
0500                      struct btrfs_block_rsv *rsv, int items,
0501                      bool use_global_rsv)
0502 {
0503     u64 qgroup_num_bytes = 0;
0504     u64 num_bytes;
0505     int ret;
0506     struct btrfs_fs_info *fs_info = root->fs_info;
0507     struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
0508
0509     if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
0510         /* One for parent inode, two for dir entries */
0511         qgroup_num_bytes = 3 * fs_info->nodesize;
0512         ret = btrfs_qgroup_reserve_meta_prealloc(root,
0513                              qgroup_num_bytes, true,
0514                              false);
0515         if (ret)
0516             return ret;
0517     }
0518
0519     num_bytes = btrfs_calc_insert_metadata_size(fs_info, items);
0520     rsv->space_info = btrfs_find_space_info(fs_info,
0521                         BTRFS_BLOCK_GROUP_METADATA);
0522     ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes,
0523                   BTRFS_RESERVE_FLUSH_ALL);
0524
0525     if (ret == -ENOSPC && use_global_rsv)
0526         ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
0527
0528     if (ret && qgroup_num_bytes)
0529         btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
0530
0531     if (!ret) {
0532         spin_lock(&rsv->lock);
0533         rsv->qgroup_rsv_reserved += qgroup_num_bytes;
0534         spin_unlock(&rsv->lock);
0535     }
0536     return ret;
0537 }
0538
0539 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
0540                       struct btrfs_block_rsv *rsv)
0541 {
0542     struct btrfs_fs_info *fs_info = root->fs_info;
0543     u64 qgroup_to_release;
0544
0545     btrfs_block_rsv_release(fs_info, rsv, (u64)-1, &qgroup_to_release);
0546     btrfs_qgroup_convert_reserved_meta(root, qgroup_to_release);
0547 }