Re: [RFC PATCH v8 19/19] af_vsock: serialize writes to shared socket

From: Arseny Krasnov <arseny.krasnov@kaspersky.com>
To: Stefano Garzarella <sgarzare@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>,
	"Michael S. Tsirkin" <mst@redhat.com>,
	Jason Wang <jasowang@redhat.com>,
	"David S. Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>,
	Jorgen Hansen <jhansen@vmware.com>,
	Colin Ian King <colin.king@canonical.com>,
	Norbert Slusarek <nslusarek@gmx.net>,
	Andra Paraschiv <andraprs@amazon.com>,
	Jeff Vander Stoep <jeffv@google.com>,
	Alexander Popov <alex.popov@linux.com>,
	"kvm@vger.kernel.org" <kvm@vger.kernel.org>,
	"virtualization@lists.linux-foundation.org" 
	<virtualization@lists.linux-foundation.org>,
	"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"stsp2@yandex.ru" <stsp2@yandex.ru>,
	"oxffffaa@gmail.com" <oxffffaa@gmail.com>
Subject: Re: [RFC PATCH v8 19/19] af_vsock: serialize writes to shared socket
Date: Wed, 21 Apr 2021 18:02:31 +0300	[thread overview]
Message-ID: <99ada1d4-3d03-3d0b-877f-176f7b83bc76@kaspersky.com> (raw)
In-Reply-To: <20210421093851.36yazy5vp4uwimd6@steredhat>

On 21.04.2021 12:38, Stefano Garzarella wrote:
> On Wed, Apr 14, 2021 at 01:51:17PM +0300, Arseny Krasnov wrote:
>> On 13.04.2021 15:47, Arseny Krasnov wrote:
>>> This add logic, that serializes write access to single socket
>>> by multiple threads. It is implemented be adding field with TID
>>> of current writer. When writer tries to send something, it checks
>>> that field is -1(free), else it sleep in the same way as waiting
>>> for free space at peers' side.
>>>
>>> This implementation is PoC and not related to SEQPACKET close, so
>>> i've placed it after whole patchset.
>>>
>>> Signed-off-by: Arseny Krasnov <arseny.krasnov@kaspersky.com>
>>> ---
>>>  include/net/af_vsock.h   |  1 +
>>>  net/vmw_vsock/af_vsock.c | 10 +++++++++-
>>>  2 files changed, 10 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/include/net/af_vsock.h b/include/net/af_vsock.h
>>> index 53d3f33dbdbf..786df80b9fc3 100644
>>> --- a/include/net/af_vsock.h
>>> +++ b/include/net/af_vsock.h
>>> @@ -69,6 +69,7 @@ struct vsock_sock {
>>>  	u64 buffer_size;
>>>  	u64 buffer_min_size;
>>>  	u64 buffer_max_size;
>>> +	pid_t tid_owner;
>>>
>>>  	/* Private to transport. */
>>>  	void *trans;
>>> diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
>>> index 54bee7e643f4..d00f8c07a9d3 100644
>>> --- a/net/vmw_vsock/af_vsock.c
>>> +++ b/net/vmw_vsock/af_vsock.c
>>> @@ -1765,7 +1765,9 @@ static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
>>>  		ssize_t written;
>>>
>>>  		add_wait_queue(sk_sleep(sk), &wait);
>>> -		while (vsock_stream_has_space(vsk) == 0 &&
>>> +		while ((vsock_stream_has_space(vsk) == 0 ||
>>> +			(vsk->tid_owner != current->pid &&
>>> +			 vsk->tid_owner != -1)) &&
>>>  		       sk->sk_err == 0 &&
>>>  		       !(sk->sk_shutdown & SEND_SHUTDOWN) &&
>>>  		       !(vsk->peer_shutdown & RCV_SHUTDOWN)) {
>>> @@ -1796,6 +1798,8 @@ static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
>>>  				goto out_err;
>>>  			}
>>>  		}
>>> +
>>> +		vsk->tid_owner = current->pid;
>>>  		remove_wait_queue(sk_sleep(sk), &wait);
>>>
>>>  		/* These checks occur both as part of and after the loop
>>> @@ -1852,7 +1856,10 @@ static int vsock_connectible_sendmsg(struct socket *sock, struct msghdr *msg,
>>>  			err = total_written;
>>>  	}
>>>  out:
>>> +	vsk->tid_owner = -1;
>>>  	release_sock(sk);
>>> +	sk->sk_write_space(sk);
>>> +
>>>  	return err;
>>>  }
>>>
>>> @@ -2199,6 +2206,7 @@ static int vsock_create(struct net *net, struct socket *sock,
>>>  		return -ENOMEM;
>>>
>>>  	vsk = vsock_sk(sk);
>>> +	vsk->tid_owner = -1;
>> This must be moved to '__vsock_create()'
> Okay, I'll review the next version.
>
> In order to backport this fix to stable branches I think is better to 
> move at the beginning of this series or even out as a separate patch.

Ok, i'll put it as first patch of patchset. I don't like to prepare it as

separate patch, because SEQPACKET will use this fix.

>
> Thanks,
> Stefano
>
>>>  	if (sock->type == SOCK_DGRAM) {
>>>  		ret = vsock_assign_transport(vsk, NULL);
>