On Wed, Sep 02, 2020 at 01:56:53AM +0000, linmiaohe wrote:
> Eric Biggers <ebiggers@xxxxxxxxxx> wrote:
> >On Tue, Sep 01, 2020 at 07:59:21AM -0400, Miaohe Lin wrote:
> >> mempool_alloc() may return NULL if __GFP_DIRECT_RECLAIM is not set in
> >> gfp_mask under memory pressure. So we should check the return value of
> >> mempool_alloc() against NULL before dereference.
> >>
> >> Fixes: a892c8d52c02 ("block: Inline encryption support for blk-mq")
> >> Signed-off-by: Miaohe Lin <linmiaohe@xxxxxxxxxx>
> >
> >It's intended that __GFP_DIRECT_RECLAIM always be set here.
> >Do you have an example where it isn't set here?
>
> map_request() only pass GFP_ATOMIC to gfp_mask, though bio crypt is not used yet.
>
> >Also, if this can indeed happen, then we need to make __bio_crypt_clone() (and bio_crypt_clone()) return a bool (or an error code) to indicate whether it succeeded or failed. We can't just ignore the allocation failure.
> >
> >- Eric
>
> IMO, just the allocation failure is ok or we would break KABI.
> Many thanks.
>
Ignoring the allocation failure isn't okay, since it would cause encrypted I/O
to fall back to unencrypted I/O, which would cause data corruption.
Also, upstream doesn't have a stable KABI.
I sent out a patch with what I have in mind; can you take a look?
https://lkml.kernel.org/r/20200902051511.79821-1-ebiggers@xxxxxxxxxx
- Eric
