Btrfs: fix how we reserve space for deleting inodes
authorJosef Bacik <josef@redhat.com>
Fri, 5 Aug 2011 17:22:24 +0000 (13:22 -0400)
committerJosef Bacik <josef@redhat.com>
Wed, 19 Oct 2011 19:12:33 +0000 (15:12 -0400)
I converted btrfs_truncate to do sane reservations for truncate, but didn't
convert btrfs_evict_inode.  Basically we need to save the orphan_rsv for
deleting the orphan item, and do normal reservations for our truncate.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
fs/btrfs/inode.c

index 6402a41b902363286603ddc433b4566cf0d697b8..fe3891e240b34aa798cdf496f3def99cc7922603 100644 (file)
@@ -3527,6 +3527,8 @@ void btrfs_evict_inode(struct inode *inode)
 {
        struct btrfs_trans_handle *trans;
        struct btrfs_root *root = BTRFS_I(inode)->root;
+       struct btrfs_block_rsv *rsv;
+       u64 min_size = btrfs_calc_trans_metadata_size(root, 2);
        unsigned long nr;
        int ret;
 
@@ -3554,22 +3556,44 @@ void btrfs_evict_inode(struct inode *inode)
                goto no_delete;
        }
 
+       rsv = btrfs_alloc_block_rsv(root);
+       if (!rsv) {
+               btrfs_orphan_del(NULL, inode);
+               goto no_delete;
+       }
+
        btrfs_i_size_write(inode, 0);
 
+       /*
+        * This is a bit simpler than btrfs_truncate since
+        *
+        * 1) We've already reserved our space for our orphan item in the
+        *    unlink.
+        * 2) We're going to delete the inode item, so we don't need to update
+        *    it at all.
+        *
+        * So we just need to reserve some slack space in case we add bytes when
+        * doing the truncate.
+        */
        while (1) {
-               trans = btrfs_join_transaction(root);
-               BUG_ON(IS_ERR(trans));
-               trans->block_rsv = root->orphan_block_rsv;
-
-               ret = btrfs_block_rsv_check(trans, root,
-                                           root->orphan_block_rsv, 0, 5);
+               ret = btrfs_block_rsv_check(NULL, root, rsv, min_size, 0);
                if (ret) {
-                       BUG_ON(ret != -EAGAIN);
-                       ret = btrfs_commit_transaction(trans, root);
-                       BUG_ON(ret);
-                       continue;
+                       printk(KERN_WARNING "Could not get space for a "
+                              "delete, will truncate on mount\n");
+                       btrfs_orphan_del(NULL, inode);
+                       btrfs_free_block_rsv(root, rsv);
+                       goto no_delete;
+               }
+
+               trans = btrfs_start_transaction(root, 0);
+               if (IS_ERR(trans)) {
+                       btrfs_orphan_del(NULL, inode);
+                       btrfs_free_block_rsv(root, rsv);
+                       goto no_delete;
                }
 
+               trans->block_rsv = rsv;
+
                ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
                if (ret != -EAGAIN)
                        break;
@@ -3578,14 +3602,17 @@ void btrfs_evict_inode(struct inode *inode)
                btrfs_end_transaction(trans, root);
                trans = NULL;
                btrfs_btree_balance_dirty(root, nr);
-
        }
 
+       btrfs_free_block_rsv(root, rsv);
+
        if (ret == 0) {
+               trans->block_rsv = root->orphan_block_rsv;
                ret = btrfs_orphan_del(trans, inode);
                BUG_ON(ret);
        }
 
+       trans->block_rsv = &root->fs_info->trans_block_rsv;
        if (!(root == root->fs_info->tree_root ||
              root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
                btrfs_return_ino(root, btrfs_ino(inode));