2019-07-03 20:31:55

by Eric Biggers

[permalink] [raw]
Subject: Re: [PATCH] crypto: user - prevent operating on larval algorithms

Hi Michal,

On Wed, Jul 03, 2019 at 10:21:08PM +0200, Michal Such?nek wrote:
> On Wed, 3 Jul 2019 22:30:57 +0800
> Herbert Xu <[email protected]> wrote:
>
> > On Tue, Jul 02, 2019 at 02:17:00PM -0700, Eric Biggers wrote:
> > > From: Eric Biggers <[email protected]>
> > >
> > > Michal Suchanek reported [1] that running the pcrypt_aead01 test from
> > > LTP [2] in a loop and holding Ctrl-C causes a NULL dereference of
> > > alg->cra_users.next in crypto_remove_spawns(), via crypto_del_alg().
> > > The test repeatedly uses CRYPTO_MSG_NEWALG and CRYPTO_MSG_DELALG.
> > >
> > > The crash occurs when the instance that CRYPTO_MSG_DELALG is trying to
> > > unregister isn't a real registered algorithm, but rather is a "test
> > > larval", which is a special "algorithm" added to the algorithms list
> > > while the real algorithm is still being tested. Larvals don't have
> > > initialized cra_users, so that causes the crash. Normally pcrypt_aead01
> > > doesn't trigger this because CRYPTO_MSG_NEWALG waits for the algorithm
> > > to be tested; however, CRYPTO_MSG_NEWALG returns early when interrupted.
> > >
>
> Do you have some way to reproduce this reliably?
>
> I suppose you would have to send a signal to the process for the call
> to get interrupted, right?
>

It reproduced pretty reliably for me with what you suggested. Just typing in
terminal:

while true; do pcrypt_aead01; done

and then holding Ctrl-C.

If I have time I'll try writing an LTP test that specifically reproduces it.
Yes, it would involve sending a signal to a thread or process that's executing
CRYPTO_MSG_NEWALG (unless I find a better way).

- Eric


2019-07-03 21:10:41

by Michal Suchánek

[permalink] [raw]
Subject: Re: [PATCH] crypto: user - prevent operating on larval algorithms

On Wed, 3 Jul 2019 13:31:29 -0700
Eric Biggers <[email protected]> wrote:

> Hi Michal,
>
> On Wed, Jul 03, 2019 at 10:21:08PM +0200, Michal Suchánek wrote:
> > On Wed, 3 Jul 2019 22:30:57 +0800
> > Herbert Xu <[email protected]> wrote:
> >
> > > On Tue, Jul 02, 2019 at 02:17:00PM -0700, Eric Biggers wrote:
> > > > From: Eric Biggers <[email protected]>
> > > >
> > > > Michal Suchanek reported [1] that running the pcrypt_aead01 test from
> > > > LTP [2] in a loop and holding Ctrl-C causes a NULL dereference of
> > > > alg->cra_users.next in crypto_remove_spawns(), via crypto_del_alg().
> > > > The test repeatedly uses CRYPTO_MSG_NEWALG and CRYPTO_MSG_DELALG.
> > > >
> > > > The crash occurs when the instance that CRYPTO_MSG_DELALG is trying to
> > > > unregister isn't a real registered algorithm, but rather is a "test
> > > > larval", which is a special "algorithm" added to the algorithms list
> > > > while the real algorithm is still being tested. Larvals don't have
> > > > initialized cra_users, so that causes the crash. Normally pcrypt_aead01
> > > > doesn't trigger this because CRYPTO_MSG_NEWALG waits for the algorithm
> > > > to be tested; however, CRYPTO_MSG_NEWALG returns early when interrupted.
> > > >
> >
> > Do you have some way to reproduce this reliably?
> >
> > I suppose you would have to send a signal to the process for the call
> > to get interrupted, right?
> >
>
> It reproduced pretty reliably for me with what you suggested. Just typing in
> terminal:
>
> while true; do pcrypt_aead01; done
>
> and then holding Ctrl-C.
>
> If I have time I'll try writing an LTP test that specifically reproduces it.
> Yes, it would involve sending a signal to a thread or process that's executing
> CRYPTO_MSG_NEWALG (unless I find a better way).

Maybe it is possible to just send the remove message without waiting
for the ack on add.

Thanks

Michal