OSDN Git Service

e2fsck: correctly deallocate invalid extent-mapped symlinks
authorTheodore Ts'o <tytso@mit.edu>
Mon, 29 Jul 2013 01:49:38 +0000 (21:49 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Mon, 29 Jul 2013 02:03:01 +0000 (22:03 -0400)
commitc8ec2bad18fdaa842f786f3b37c9320a3411aea3
tree4c13783b8bfb32e94da5a419361e07a3d4f76997
parentdd50ef87432284d93b9f39178c4fab242d76654e
e2fsck: correctly deallocate invalid extent-mapped symlinks

The function deallocate_inode() in e2fsck/pass2.c was buggy in that it
would clear out the inode's mode and flags fields before trying to
deallocate any blocks which might belong to the inode.

The good news is that deallocate_inode() is mostly used to free inodes
which do not have blocks: device inodes, FIFO's, Unix-domain sockets.

The bad news is that if deallocate_inode() tried to free an invalid
extent-mapped inode, it would try to interpret the root of the extent
node as block numbers, and would therefore mark various file system
metadata blocks (the superblock, block group descriptors, the root
directory, etc.) as free and available for allocation.  This was
unfortunate.

(Try running an older e2fsck against the test file system image in the
new test f_invalid_extent_symlink, and then run e2fsck a second time
on the fs image, and weep.)

Fortunately, this kind of file system image corruption appears to be
fairly rare in actual practice, since it would require a very unlucky
set of bits to be flipped, or a buggy file system implementation.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
e2fsck/pass2.c
tests/f_invalid_extent_symlink/expect.1 [new file with mode: 0644]
tests/f_invalid_extent_symlink/expect.2 [new file with mode: 0644]
tests/f_invalid_extent_symlink/image.gz [new file with mode: 0644]
tests/f_invalid_extent_symlink/name [new file with mode: 0644]