On Tue, Oct 11, 2022 at 08:57:07PM -0400, Theodore Ts'o wrote:
> On Tue, Oct 11, 2022 at 04:57:45PM +0100, Luís Henriques wrote:
> > diff --git a/fs/ext4/namei.c b/fs/ext4/namei.c
> > index 3a31b662f661..06803292e394 100644
> > --- a/fs/ext4/namei.c
> > +++ b/fs/ext4/namei.c
> > @@ -2254,8 +2254,18 @@ static int make_indexed_dir(handle_t *handle, struct ext4_filename *fname,
> > memset(de, 0, len); /* wipe old data */
> > de = (struct ext4_dir_entry_2 *) data2;
> > top = data2 + len;
> > - while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
> > + while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top) {
> > + if (de->rec_len & 3) {
>
> As the kernel test bot as flaged, de->rec_len needs to be byte swapped
> on big endian machines. Also, for block sizes larger than 64k the low
> 2 bits are used to encode rec_len sizes 256k-4. All of this is
> encoded in ext4_rec_len_from_disk().
>
> However, I think a better thing to do is instead of doing this one
> check on rec len, that instead we call ext4_check_dir_entry(), which
> will do this check, and many more besides. It will also avoid some
> code duplication, since it will take care of calling EXT4_ERROR_INODE
> with the appropriate explanatory message.
Awesome, thanks for the explanation, Ted. I'll work on a v2 of the patch
that'll use ext4_check_dir_entry() and send it after running some tests
with it. Thanks for the suggestion!
Cheers,
--
Luís
