fs/btrfs/tree-defrag.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * Copyright (C) 2007 Oracle.  All rights reserved.
0004  */
0005
0006 #include <linux/sched.h>
0007 #include "ctree.h"
0008 #include "disk-io.h"
0009 #include "print-tree.h"
0010 #include "transaction.h"
0011 #include "locking.h"
0012
0013 /*
0014  * Defrag all the leaves in a given btree.
0015  * Read all the leaves and try to get key order to
0016  * better reflect disk order
0017  */
0018
0019 int btrfs_defrag_leaves(struct btrfs_trans_handle *trans,
0020             struct btrfs_root *root)
0021 {
0022     struct btrfs_path *path = NULL;
0023     struct btrfs_key key;
0024     int ret = 0;
0025     int wret;
0026     int level;
0027     int next_key_ret = 0;
0028     u64 last_ret = 0;
0029
0030     if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
0031         goto out;
0032
0033     path = btrfs_alloc_path();
0034     if (!path)
0035         return -ENOMEM;
0036
0037     level = btrfs_header_level(root->node);
0038
0039     if (level == 0)
0040         goto out;
0041
0042     if (root->defrag_progress.objectid == 0) {
0043         struct extent_buffer *root_node;
0044         u32 nritems;
0045
0046         root_node = btrfs_lock_root_node(root);
0047         nritems = btrfs_header_nritems(root_node);
0048         root->defrag_max.objectid = 0;
0049         /* from above we know this is not a leaf */
0050         btrfs_node_key_to_cpu(root_node, &root->defrag_max,
0051                       nritems - 1);
0052         btrfs_tree_unlock(root_node);
0053         free_extent_buffer(root_node);
0054         memset(&key, 0, sizeof(key));
0055     } else {
0056         memcpy(&key, &root->defrag_progress, sizeof(key));
0057     }
0058
0059     path->keep_locks = 1;
0060
0061     ret = btrfs_search_forward(root, &key, path, BTRFS_OLDEST_GENERATION);
0062     if (ret < 0)
0063         goto out;
0064     if (ret > 0) {
0065         ret = 0;
0066         goto out;
0067     }
0068     btrfs_release_path(path);
0069     /*
0070      * We don't need a lock on a leaf. btrfs_realloc_node() will lock all
0071      * leafs from path->nodes[1], so set lowest_level to 1 to avoid later
0072      * a deadlock (attempting to write lock an already write locked leaf).
0073      */
0074     path->lowest_level = 1;
0075     wret = btrfs_search_slot(trans, root, &key, path, 0, 1);
0076
0077     if (wret < 0) {
0078         ret = wret;
0079         goto out;
0080     }
0081     if (!path->nodes[1]) {
0082         ret = 0;
0083         goto out;
0084     }
0085     /*
0086      * The node at level 1 must always be locked when our path has
0087      * keep_locks set and lowest_level is 1, regardless of the value of
0088      * path->slots[1].
0089      */
0090     BUG_ON(path->locks[1] == 0);
0091     ret = btrfs_realloc_node(trans, root,
0092                  path->nodes[1], 0,
0093                  &last_ret,
0094                  &root->defrag_progress);
0095     if (ret) {
0096         WARN_ON(ret == -EAGAIN);
0097         goto out;
0098     }
0099     /*
0100      * Now that we reallocated the node we can find the next key. Note that
0101      * btrfs_find_next_key() can release our path and do another search
0102      * without COWing, this is because even with path->keep_locks = 1,
0103      * btrfs_search_slot() / ctree.c:unlock_up() does not keeps a lock on a
0104      * node when path->slots[node_level - 1] does not point to the last
0105      * item or a slot beyond the last item (ctree.c:unlock_up()). Therefore
0106      * we search for the next key after reallocating our node.
0107      */
0108     path->slots[1] = btrfs_header_nritems(path->nodes[1]);
0109     next_key_ret = btrfs_find_next_key(root, path, &key, 1,
0110                        BTRFS_OLDEST_GENERATION);
0111     if (next_key_ret == 0) {
0112         memcpy(&root->defrag_progress, &key, sizeof(key));
0113         ret = -EAGAIN;
0114     }
0115 out:
0116     btrfs_free_path(path);
0117     if (ret == -EAGAIN) {
0118         if (root->defrag_max.objectid > root->defrag_progress.objectid)
0119             goto done;
0120         if (root->defrag_max.type > root->defrag_progress.type)
0121             goto done;
0122         if (root->defrag_max.offset > root->defrag_progress.offset)
0123             goto done;
0124         ret = 0;
0125     }
0126 done:
0127     if (ret != -EAGAIN)
0128         memset(&root->defrag_progress, 0,
0129                sizeof(root->defrag_progress));
0130
0131     return ret;
0132 }