btrfs: qgroup: Don't copy extent buffer to do qgroup rescan
authorQu Wenruo <quwenruo@cn.fujitsu.com>
Mon, 26 Oct 2015 01:19:43 +0000 (09:19 +0800)
committerChris Mason <clm@fb.com>
Tue, 27 Oct 2015 02:42:30 +0000 (19:42 -0700)
Ancient qgroup code call memcpy() on a extent buffer and use it for leaf
iteration.

As extent buffer contains lock, pointers to pages, it's never sane to do
such copy.

The following bug may be caused by this insane operation:
[92098.841309] general protection fault: 0000 [#1] SMP
[92098.841338] Modules linked in: ...
[92098.841814] CPU: 1 PID: 24655 Comm: kworker/u4:12 Not tainted
4.3.0-rc1 #1
[92098.841868] Workqueue: btrfs-qgroup-rescan btrfs_qgroup_rescan_helper
[btrfs]
[92098.842261] Call Trace:
[92098.842277]  [<ffffffffc035a5d8>] ? read_extent_buffer+0xb8/0x110
[btrfs]
[92098.842304]  [<ffffffffc0396d00>] ? btrfs_find_all_roots+0x60/0x70
[btrfs]
[92098.842329]  [<ffffffffc039af3d>]
btrfs_qgroup_rescan_worker+0x28d/0x5a0 [btrfs]

Where btrfs_qgroup_rescan_worker+0x28d is btrfs_disk_key_to_cpu(),
called in reading key from the copied extent_buffer.

This patch will use btrfs_clone_extent_buffer() to a better copy of
extent buffer to deal such case.

Reported-by: Stephane Lesimple <stephane_btrfs@lesimple.fr>
Suggested-by: Filipe Manana <fdmanana@kernel.org>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: Filipe Manana <fdmanana@suse.com>
Signed-off-by: Chris Mason <clm@fb.com>
fs/btrfs/qgroup.c

index 158633c..31d1934 100644 (file)
@@ -2192,10 +2192,10 @@ void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
  */
 static int
 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
-                  struct btrfs_trans_handle *trans,
-                  struct extent_buffer *scratch_leaf)
+                  struct btrfs_trans_handle *trans)
 {
        struct btrfs_key found;
+       struct extent_buffer *scratch_leaf = NULL;
        struct ulist *roots = NULL;
        struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
        u64 num_bytes;
@@ -2233,7 +2233,15 @@ qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
        fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
 
        btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
-       memcpy(scratch_leaf, path->nodes[0], sizeof(*scratch_leaf));
+       scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
+       if (!scratch_leaf) {
+               ret = -ENOMEM;
+               mutex_unlock(&fs_info->qgroup_rescan_lock);
+               goto out;
+       }
+       extent_buffer_get(scratch_leaf);
+       btrfs_tree_read_lock(scratch_leaf);
+       btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
        slot = path->slots[0];
        btrfs_release_path(path);
        mutex_unlock(&fs_info->qgroup_rescan_lock);
@@ -2259,6 +2267,10 @@ qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
                        goto out;
        }
 out:
+       if (scratch_leaf) {
+               btrfs_tree_read_unlock_blocking(scratch_leaf);
+               free_extent_buffer(scratch_leaf);
+       }
        btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
 
        return ret;
@@ -2270,16 +2282,12 @@ static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
                                                     qgroup_rescan_work);
        struct btrfs_path *path;
        struct btrfs_trans_handle *trans = NULL;
-       struct extent_buffer *scratch_leaf = NULL;
        int err = -ENOMEM;
        int ret = 0;
 
        path = btrfs_alloc_path();
        if (!path)
                goto out;
-       scratch_leaf = kmalloc(sizeof(*scratch_leaf), GFP_NOFS);
-       if (!scratch_leaf)
-               goto out;
 
        err = 0;
        while (!err) {
@@ -2291,8 +2299,7 @@ static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
                if (!fs_info->quota_enabled) {
                        err = -EINTR;
                } else {
-                       err = qgroup_rescan_leaf(fs_info, path, trans,
-                                                scratch_leaf);
+                       err = qgroup_rescan_leaf(fs_info, path, trans);
                }
                if (err > 0)
                        btrfs_commit_transaction(trans, fs_info->fs_root);
@@ -2301,7 +2308,6 @@ static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
        }
 
 out:
-       kfree(scratch_leaf);
        btrfs_free_path(path);
 
        mutex_lock(&fs_info->qgroup_rescan_lock);