cryptoapi highmem bug

[Christophe Saout]

Hi,

someone noticed strange corruptions with dm-crypt and highmem. After I
found out that I could force my machine to use highmem even though it
only has 256MB, I finally found the problem after some debugging:

The problem is in cbc_process (well, partly):

>      const int need_stack = (src == dst);
>      u8 stack[need_stack ? crypto_tfm_alg_blocksize(tfm) : 0];
>      u8 *buf = need_stack ? stack : dst;

src == dst fails if the page was in highmem because crypto_kmap will
assign two different virtual addresses for the same page.

The result is data corruption.

How could this be fixed?

scapperwalk_map could check if this page was already mapped (walk_in)
and reuse the virtual address if so. So a single page is only mapped
once and the check in cbc_process will work.

I can really use the src == dst case because I would need to allocate
unnecessary buffers (at least 512 bytes at a time and per cpu).

(I just hacked dm-crypt to allocate 512 bytes on the stack and use it
temporarily and kmap around myself to copy it back and the problem is
gone. Ugly.)


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/