Re: [PATCH 03/17] fs: don't allow to complete sync iocbs through aio_complete

Re: [PATCH 03/17] fs: don't allow to complete sync iocbs through aio_complete

[Tadeusz Struk]

On 04/11/2015 02:18 PM, Al Viro wrote:
> @@ -766,8 +765,7 @@ int sock_recvmsg(struct socket *sock, struct msghdr *msg,
>  
>  	init_sync_kiocb(&iocb, NULL);
>  	ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
> -	if (-EIOCBQUEUED == ret)
> -		ret = wait_on_sync_kiocb(&iocb);
> +	BUG_ON(ret == -EIOCBQUEUED);

NACK - PF_ALG type sockets support async operations and return -EIOCBQUEUED
See skcipher_recvmsg_async() in crypto/algif_skcipher.c in net-next

Re: [PATCH 03/17] fs: don't allow to complete sync iocbs through aio_complete

[Al Viro]

On Tue, Apr 14, 2015 at 11:00:49AM -0700, Tadeusz Struk wrote:
> On 04/11/2015 02:18 PM, Al Viro wrote:
> > @@ -766,8 +765,7 @@ int sock_recvmsg(struct socket *sock, struct msghdr *msg,
> >  
> >  	init_sync_kiocb(&iocb, NULL);
> >  	ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
> > -	if (-EIOCBQUEUED == ret)
> > -		ret = wait_on_sync_kiocb(&iocb);
> > +	BUG_ON(ret == -EIOCBQUEUED);
> 
> NACK - PF_ALG type sockets support async operations and return -EIOCBQUEUED
> See skcipher_recvmsg_async() in crypto/algif_skcipher.c in net-next

Its only caller is

static int skcipher_recvmsg(struct socket *sock, struct msghdr *msg,
                            size_t ignored, int flags)
{
        return (msg->msg_iocb && !is_sync_kiocb(msg->msg_iocb)) ?
                skcipher_recvmsg_async(sock, msg, flags) :
                skcipher_recvmsg_sync(sock, msg, flags);
}

Note that !is_sync_kiocb() in there.  Compare with init_sync_kiocb() in
sock_recvmsg()...

Re: [PATCH 03/17] fs: don't allow to complete sync iocbs through aio_complete

[Tadeusz Struk]

On 04/14/2015 11:26 AM, Al Viro wrote:
>>> @@ -766,8 +765,7 @@ int sock_recvmsg(struct socket *sock, struct msghdr *msg,
>>> > >  
>>> > >  	init_sync_kiocb(&iocb, NULL);
>>> > >  	ret = __sock_recvmsg(&iocb, sock, msg, size, flags);
>>> > > -	if (-EIOCBQUEUED == ret)
>>> > > -		ret = wait_on_sync_kiocb(&iocb);
>>> > > +	BUG_ON(ret == -EIOCBQUEUED);
>> > 
>> > NACK - PF_ALG type sockets support async operations and return -EIOCBQUEUED
>> > See skcipher_recvmsg_async() in crypto/algif_skcipher.c in net-next
> Its only caller is
> 
> static int skcipher_recvmsg(struct socket *sock, struct msghdr *msg,
>                             size_t ignored, int flags)
> {
>         return (msg->msg_iocb && !is_sync_kiocb(msg->msg_iocb)) ?
>                 skcipher_recvmsg_async(sock, msg, flags) :
>                 skcipher_recvmsg_sync(sock, msg, flags);
> }
> 
> Note that !is_sync_kiocb() in there.  Compare with init_sync_kiocb() in
> sock_recvmsg()...

The only problem is it calls init_sync_kiocb(&iocb, NULL) on a different iocb.
The one that isn't even passed to skcipher_recvmsg()
skcipher_recvmsg() checks is_sync_kiocb(msg->msg_iocb).
I just want to make sure that after the merging window is closed I can still
trigger aio_read on PF_ALG socket.
thanks

Re: [PATCH 03/17] fs: don't allow to complete sync iocbs through aio_complete

[Al Viro]

On Tue, Apr 14, 2015 at 11:37:01AM -0700, Tadeusz Struk wrote:

> The only problem is it calls init_sync_kiocb(&iocb, NULL) on a different iocb.
> The one that isn't even passed to skcipher_recvmsg()
> skcipher_recvmsg() checks is_sync_kiocb(msg->msg_iocb).
> I just want to make sure that after the merging window is closed I can still
> trigger aio_read on PF_ALG socket.
> thanks

... and after the merge with net-next#master that BUG_ON() is gone, so
you are still just fine.

FWIW, in the current for-davem (or in net-next#master, now that Dave has
pulled it) the functions in question are:
static inline int sock_recvmsg_nosec(struct socket *sock, struct msghdr *msg,
                                     size_t size, int flags)
{
        return sock->ops->recvmsg(sock, msg, size, flags);
}

int sock_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
                 int flags) 
{
        int err = security_socket_recvmsg(sock, msg, size, flags);

        return err ?: sock_recvmsg_nosec(sock, msg, size, flags);
}
EXPORT_SYMBOL(sock_recvmsg);
static ssize_t sock_read_iter(struct kiocb *iocb, struct iov_iter *to)
{
        struct file *file = iocb->ki_filp;
        struct socket *sock = file->private_data;
        struct msghdr msg = {.msg_iter = *to,
                             .msg_iocb = iocb};
        ssize_t res;

        if (file->f_flags & O_NONBLOCK)
                msg.msg_flags = MSG_DONTWAIT;

        if (iocb->ki_pos != 0)
                return -ESPIPE; 

        if (!iov_iter_count(to))        /* Match SYS5 behaviour */
                return 0;

        res = sock_recvmsg(sock, &msg, iov_iter_count(to), msg.msg_flags);
        *to = msg.msg_iter;
        return res;
}

AFAICS, everything looks fine.  Now, _another_ BUG_ON() might be worth
eventual removal (one in sock_sendmsg_nosec()), but right now we don't
have async ->sendmsg() instances, so it's not urgent.