]> www.infradead.org Git - users/dwmw2/linux.git/commit
ext4: fix race conditions in ->d_compare() and ->d_hash()
authorEric Biggers <ebiggers@google.com>
Fri, 24 Jan 2020 04:12:34 +0000 (20:12 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 11 Feb 2020 12:37:02 +0000 (04:37 -0800)
commitd30f7b3fa0dc6f3272886f7dbc55ede1ab3b136e
tree378b3394dfbcc37cd0110e8ebc405723b30bc077
parentd32f686fe12f7fd069e6db3c6222a245b9a42507
ext4: fix race conditions in ->d_compare() and ->d_hash()

commit ec772f01307a2c06ebf6cdd221e6b518a71ddae7 upstream.

Since ->d_compare() and ->d_hash() can be called in RCU-walk mode,
->d_parent and ->d_inode can be concurrently modified, and in
particular, ->d_inode may be changed to NULL.  For ext4_d_hash() this
resulted in a reproducible NULL dereference if a lookup is done in a
directory being deleted, e.g. with:

int main()
{
if (fork()) {
for (;;) {
mkdir("subdir", 0700);
rmdir("subdir");
}
} else {
for (;;)
access("subdir/file", 0);
}
}

... or by running the 't_encrypted_d_revalidate' program from xfstests.
Both repros work in any directory on a filesystem with the encoding
feature, even if the directory doesn't actually have the casefold flag.

I couldn't reproduce a crash in ext4_d_compare(), but it appears that a
similar crash is possible there.

Fix these bugs by reading ->d_parent and ->d_inode using READ_ONCE() and
falling back to the case sensitive behavior if the inode is NULL.

Reported-by: Al Viro <viro@zeniv.linux.org.uk>
Fixes: b886ee3e778e ("ext4: Support case-insensitive file name lookups")
Cc: <stable@vger.kernel.org> # v5.2+
Signed-off-by: Eric Biggers <ebiggers@google.com>
Link: https://lore.kernel.org/r/20200124041234.159740-1-ebiggers@kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
fs/ext4/dir.c