f2fs: fix to detect inconsistent nat entry during truncation

As Roman Smirnov reported as below:

"
There is a possible bug in f2fs_truncate_inode_blocks():

    if (err < 0 && err != -ENOENT)
    			goto fail;
        ...
        offset[1] = 0;
        offset[0]++;
        nofs += err;

If err = -ENOENT then nofs will sum with an error code,
which is strange behaviour. Also if nofs < ENOENT this will
cause an overflow. err will be equal to -ENOENT with the
following call stack:

truncate_nodes()
  f2fs_get_node_page()
     __get_node_page()
        read_node_page()
"

If nat is corrupted, truncate_nodes() may return -ENOENT, and
f2fs_truncate_inode_blocks() doesn't handle such error correctly,
fix it.

Reported-by: Roman Smirnov <r.smirnov@omp.ru>
Closes: https://lore.kernel.org/linux-f2fs-devel/085b27fd2b364a3c8c3a9ca77363e246@omp.ru
Signed-off-by: Chao Yu <chao@kernel.org>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>

diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
index b3de6d6cdb02199c13808bfaabbc8e742b52a067..bb57bbaff7b4f278da588e83cb4dbf9eae63832e 100644 (file)
--- a/fs/f2fs/node.c
+++ b/fs/f2fs/node.c
@@ -1187,7 +1187,17 @@ skip_partial:
                default:
                        BUG();
                }
-               if (err < 0 && err != -ENOENT)
+               if (err == -ENOENT) {
+                       set_sbi_flag(F2FS_P_SB(page), SBI_NEED_FSCK);
+                       f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
+                       f2fs_err_ratelimited(sbi,
+                               "truncate node fail, ino:%lu, nid:%u, "
+                               "offset[0]:%d, offset[1]:%d, nofs:%d",
+                               inode->i_ino, dn.nid, offset[0],
+                               offset[1], nofs);
+                       err = 0;
+               }
+               if (err < 0)
                        goto fail;
                if (offset[1] == 0 &&
                                ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {