Re: [PATCH] isofs: Always advance to the next block in readdir and lookup

[Matthias Goergens <matthias.goergens@gmail.com>]

On Wed 23-09-26 18:55:11, Jan Kara wrote:
> So I agree what you do is easier to understand so I'll fold it into the
> previous commit. Just I'm not sure it is a realistic failure case - that
> would mean a directory would contain full block of zeros which I don't
> think can happen. Plus this infinite loop was there forever AFAICS and
> nobody complained...

Thanks for folding it in.

Maybe I'm missing something, but I thought the loop came in with
3c01d9263683.  Before it, both walkers advanced with
(pos + ISOFS_BLOCK_SIZE) & ~(ISOFS_BLOCK_SIZE - 1), which moves on even
when pos is already aligned, and mainline lists the reproducer images
fine; only for_next spins on them.  Is there an older path that loops
that I haven't found?

Agreed that a well-formed image with 2048-byte blocks won't have such a
block.  But only the first byte of the directory block has to be zero,
since the walk doesn't look past it: a crafted image needs one byte, and
faulty hardware or an incomplete copy that returns a zeroed sector does
it too.  The task then can't be killed.

Matthias