| Message ID | 20250102035321.391540-1-sunjunchao2870@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [1/2] ocfs2: Add a sanity check for corrupted file system. | expand |
On 2025/1/2 11:53, Julian Sun wrote: > Hi, > > Recently syzbot reported a use-after-free issue[1]. > > The root cause of the problem is that the journal > inode recorded in this file system image is corrupted. > The value of "di->id2.i_list.l_next_free_rec" is 8193, > which is greater than the value of "di->id2.i_list.l_count" (19). > > To solve this problem, an additional check should be added > within ocfs2_get_clusters_nocache(). If the check fails, an error will > be returned and the file system will be set to read-only. > > [1]: https://lore.kernel.org/all/67577778.050a0220.a30f1.01bc.GAE@google.com/T/ > > Reported-and-tested-by: syzbot+2313dda4dc4885c93578@syzkaller.appspotmail.com > Closes: https://syzkaller.appspot.com/bug?extid=2313dda4dc4885c93578 > Signed-off-by: Julian Sun <sunjunchao2870@gmail.com> > --- > fs/ocfs2/extent_map.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/fs/ocfs2/extent_map.c b/fs/ocfs2/extent_map.c > index f7672472fa82..7fdbecb9c031 100644 > --- a/fs/ocfs2/extent_map.c > +++ b/fs/ocfs2/extent_map.c > @@ -414,6 +414,16 @@ static int ocfs2_get_clusters_nocache(struct inode *inode, > el = &di->id2.i_list; > tree_height = le16_to_cpu(el->l_tree_depth); Move this down close to tree_height check. > > + if (le16_to_cpu(di->id2.i_list.l_next_free_rec) > > + le16_to_cpu(di->id2.i_list.l_count)) { Use le16_to_cpu(el->l_next_free_rec) and le16_to_cpu(el->l_count). Thanks, Joseph > + ret = ocfs2_error(inode->i_sb, "Invalid dinode #%llu: next_free_rec is %u, count is %u\n", > + (unsigned long long)di_bh->b_blocknr, > + le16_to_cpu(di->id2.i_list.l_next_free_rec), > + le16_to_cpu(di->id2.i_list.l_count)); > + > + return ret; > + } > + > if (tree_height > 0) { > ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster, > &eb_bh);
diff --git a/fs/ocfs2/extent_map.c b/fs/ocfs2/extent_map.c index f7672472fa82..7fdbecb9c031 100644 --- a/fs/ocfs2/extent_map.c +++ b/fs/ocfs2/extent_map.c @@ -414,6 +414,16 @@ static int ocfs2_get_clusters_nocache(struct inode *inode, el = &di->id2.i_list; tree_height = le16_to_cpu(el->l_tree_depth); + if (le16_to_cpu(di->id2.i_list.l_next_free_rec) > + le16_to_cpu(di->id2.i_list.l_count)) { + ret = ocfs2_error(inode->i_sb, "Invalid dinode #%llu: next_free_rec is %u, count is %u\n", + (unsigned long long)di_bh->b_blocknr, + le16_to_cpu(di->id2.i_list.l_next_free_rec), + le16_to_cpu(di->id2.i_list.l_count)); + + return ret; + } + if (tree_height > 0) { ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster, &eb_bh);
Hi, Recently syzbot reported a use-after-free issue[1]. The root cause of the problem is that the journal inode recorded in this file system image is corrupted. The value of "di->id2.i_list.l_next_free_rec" is 8193, which is greater than the value of "di->id2.i_list.l_count" (19). To solve this problem, an additional check should be added within ocfs2_get_clusters_nocache(). If the check fails, an error will be returned and the file system will be set to read-only. [1]: https://lore.kernel.org/all/67577778.050a0220.a30f1.01bc.GAE@google.com/T/ Reported-and-tested-by: syzbot+2313dda4dc4885c93578@syzkaller.appspotmail.com Closes: https://syzkaller.appspot.com/bug?extid=2313dda4dc4885c93578 Signed-off-by: Julian Sun <sunjunchao2870@gmail.com> --- fs/ocfs2/extent_map.c | 10 ++++++++++ 1 file changed, 10 insertions(+)