On Tue, 04 Jun 2024 15:06:32 -0400, Gabriel Krisman Bertazi wrote:
> >> > When mounting the ext4 filesystem, if the hash version and casefolded are not
> >> > consistent, exit the mounting.
> >> >
> >> > Reported-by: syzbot+340581ba9dceb7e06fb3@xxxxxxxxxxxxxxxxxxxxxxxxx
> >> > Signed-off-by: Lizhi Xu <lizhi.xu@xxxxxxxxxxxxx>
> >> > ---
> >> > fs/ext4/super.c | 5 +++++
> >> > 1 file changed, 5 insertions(+)
> >> >
> >> > diff --git a/fs/ext4/super.c b/fs/ext4/super.c
> >> > index c682fb927b64..0ad326504c50 100644
> >> > --- a/fs/ext4/super.c
> >> > +++ b/fs/ext4/super.c
> >> > @@ -5262,6 +5262,11 @@ static int __ext4_fill_super(struct fs_context *fc, struct super_block *sb)
> >> > goto failed_mount;
> >> >
> >> > ext4_hash_info_init(sb);
> >> > + if (es->s_def_hash_version == DX_HASH_SIPHASH &&
> >> > + !ext4_has_feature_casefold(sb)) {
> >>
> >> Can we ever have DX_HASH_SIPHASH set up in the super block? I thought
> >> it was used solely for directories where ext4_hash_in_dirent(inode) is
> >> true.
> > The value of s'def_hash_version is obtained by reading the super block from the
> > buffer cache of the block device in ext4_load_super().
>
> Yes, I know. My point is whether this check should just be:
Based on the existing information, it cannot be confirmed that it is incorrect
to separately determine the value of s_def_hash_version as DX_HASH_SIPHASH.
Additionally, I have come up with a better solution, and I will issue the next
fixed version in a while.
>
> if (es->s_def_hash_version == DX_HASH_SIPHASH)
> goto failed_mount;
>
> Since, IIUC, DX_HASH_SIPHASH is done per-directory and not written to
> the sb.
>
> >> If this is only for the case of a superblock corruption, perhaps we
> >> should always reject the mount, whether casefold is enabled or not?
> > Based on the existing information, it cannot be confirmed whether the superblock
> > is corrupt, but one thing is clear: if the default hash version of the superblock
> > is set to DX_HASH_SIPHASH, but the casefold feature is not set at the same time,
> > it is definitely an error.
Lizhi
