lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20220613085046.ee7cb2ye5yq5cbfo@sgarzare-redhat>
Date:   Mon, 13 Jun 2022 10:50:46 +0200
From:   Stefano Garzarella <sgarzare@...hat.com>
To:     Arseniy Krasnov <AVKrasnov@...rdevices.ru>
Cc:     Stefan Hajnoczi <stefanha@...hat.com>,
        "Michael S. Tsirkin" <mst@...hat.com>,
        Jason Wang <jasowang@...hat.com>,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "kvm@...r.kernel.org" <kvm@...r.kernel.org>,
        "virtualization@...ts.linux-foundation.org" 
        <virtualization@...ts.linux-foundation.org>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        kernel <kernel@...rdevices.ru>,
        Krasnov Arseniy <oxffffaa@...il.com>
Subject: Re: [RFC PATCH v2 3/8] af_vsock: add zerocopy receive logic

On Thu, Jun 09, 2022 at 12:20:22PM +0000, Arseniy Krasnov wrote:
>On 09.06.2022 11:39, Stefano Garzarella wrote:
>> On Fri, Jun 03, 2022 at 05:35:48AM +0000, Arseniy Krasnov wrote:
>>> This:
>>> 1) Adds callback for 'mmap()' call on socket. It checks vm
>>>   area flags and sets vm area ops.
>>> 2) Adds special 'getsockopt()' case which calls transport
>>>   zerocopy callback. Input argument is vm area address.
>>> 3) Adds 'getsockopt()/setsockopt()' for switching on/off rx
>>>   zerocopy mode.
>>>
>>> Signed-off-by: Arseniy Krasnov <AVKrasnov@...rdevices.ru>
>>> ---
>>> include/net/af_vsock.h          |   7 +++
>>> include/uapi/linux/vm_sockets.h |   3 +
>>> net/vmw_vsock/af_vsock.c        | 100 ++++++++++++++++++++++++++++++++
>>> 3 files changed, 110 insertions(+)
>>>
>>> diff --git a/include/net/af_vsock.h b/include/net/af_vsock.h
>>> index f742e50207fb..f15f84c648ff 100644
>>> --- a/include/net/af_vsock.h
>>> +++ b/include/net/af_vsock.h
>>> @@ -135,6 +135,13 @@ struct vsock_transport {
>>>     bool (*stream_is_active)(struct vsock_sock *);
>>>     bool (*stream_allow)(u32 cid, u32 port);
>>>
>>> +    int (*rx_zerocopy_set)(struct vsock_sock *vsk,
>>> +                   bool enable);
>>> +    int (*rx_zerocopy_get)(struct vsock_sock *vsk);
>>> +    int (*zerocopy_dequeue)(struct vsock_sock *vsk,
>>> +                struct vm_area_struct *vma,
>>> +                unsigned long addr);
>>> +
>>>     /* SEQ_PACKET. */
>>>     ssize_t (*seqpacket_dequeue)(struct vsock_sock *vsk, struct msghdr *msg,
>>>                      int flags);
>>> diff --git a/include/uapi/linux/vm_sockets.h b/include/uapi/linux/vm_sockets.h
>>> index c60ca33eac59..d1f792bed1a7 100644
>>> --- a/include/uapi/linux/vm_sockets.h
>>> +++ b/include/uapi/linux/vm_sockets.h
>>> @@ -83,6 +83,9 @@
>>>
>>> #define SO_VM_SOCKETS_CONNECT_TIMEOUT_NEW 8
>>>
>>> +#define SO_VM_SOCKETS_MAP_RX 9
>>> +#define SO_VM_SOCKETS_ZEROCOPY 10
>>> +
>>> #if !defined(__KERNEL__)
>>> #if __BITS_PER_LONG == 64 || (defined(__x86_64__) && defined(__ILP32__))
>>> #define SO_VM_SOCKETS_CONNECT_TIMEOUT SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD
>>> diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
>>> index f04abf662ec6..10061ef21730 100644
>>> --- a/net/vmw_vsock/af_vsock.c
>>> +++ b/net/vmw_vsock/af_vsock.c
>>> @@ -1644,6 +1644,17 @@ static int vsock_connectible_setsockopt(struct socket *sock,
>>>         }
>>>         break;
>>>     }
>>> +    case SO_VM_SOCKETS_ZEROCOPY: {
>>> +        if (!transport || !transport->rx_zerocopy_set) {
>>> +            err = -EOPNOTSUPP;
>>> +        } else {
>>> +            COPY_IN(val);
>>> +
>>> +            if (transport->rx_zerocopy_set(vsk, val))
>>> +                err = -EINVAL;
>>> +        }
>>> +        break;
>>> +    }
>>>
>>>     default:
>>>         err = -ENOPROTOOPT;
>>> @@ -1657,6 +1668,48 @@ static int vsock_connectible_setsockopt(struct socket *sock,
>>>     return err;
>>> }
>>>
>>> +static const struct vm_operations_struct afvsock_vm_ops = {
>>> +};
>>> +
>>> +static int vsock_recv_zerocopy(struct socket *sock,
>>> +                   unsigned long address)
>>> +{
>>> +    struct sock *sk = sock->sk;
>>> +    struct vsock_sock *vsk = vsock_sk(sk);
>>> +    struct vm_area_struct *vma;
>>> +    const struct vsock_transport *transport;
>>> +    int res;
>>> +
>>> +    transport = vsk->transport;
>>> +
>>> +    if (!transport->rx_zerocopy_get)
>>> +        return -EOPNOTSUPP;
>>> +
>>> +    if (!transport->rx_zerocopy_get(vsk))
>>> +        return -EOPNOTSUPP;
>>
>> Maybe we can merge in
>>         if (!transport->rx_zerocopy_get ||
>>             !transport->rx_zerocopy_get(vsk)}
>>                 return -EOPNOTSUPP;
>>
>>> +
>>> +    if (!transport->zerocopy_dequeue)
>>> +        return -EOPNOTSUPP;
>>> +
>>> +    lock_sock(sk);
>>> +    mmap_write_lock(current->mm);
>>
>> So, multiple threads using different sockets are serialized if they use zero-copy?
>>
>> IIUC this is necessary because the callback calls vm_insert_page().
>>
>> At this point I think it's better not to do this in every transport, but have the callback return an array of pages to map and we map them here trying to limit as much as possible the critical section to protect with mmap_write_lock().
>
>Yes, it will be easy to return array of pages by transport callback,
>
>>
>>> +
>>> +    vma = vma_lookup(current->mm, address);
>>> +
>>> +    if (!vma || vma->vm_ops != &afvsock_vm_ops) {
>>> +        mmap_write_unlock(current->mm);
>>> +        release_sock(sk);
>>> +        return -EINVAL;
>>> +    }
>>> +
>>> +    res = transport->zerocopy_dequeue(vsk, vma, address);
>>> +
>>> +    mmap_write_unlock(current->mm);
>>> +    release_sock(sk);
>>> +
>>> +    return res;
>>> +}
>>> +
>>> static int vsock_connectible_getsockopt(struct socket *sock,
>>>                     int level, int optname,
>>>                     char __user *optval,
>>> @@ -1701,6 +1754,39 @@ static int vsock_connectible_getsockopt(struct socket *sock,
>>>         lv = sock_get_timeout(vsk->connect_timeout, &v,
>>>                       optname == SO_VM_SOCKETS_CONNECT_TIMEOUT_OLD);
>>>         break;
>>> +    case SO_VM_SOCKETS_ZEROCOPY: {
>>> +        const struct vsock_transport *transport;
>>> +        int res;
>>> +
>>> +        transport = vsk->transport;
>>> +
>>> +        if (!transport->rx_zerocopy_get)
>>> +            return -EOPNOTSUPP;
>>> +
>>> +        lock_sock(sk);
>>
>> I think we should call lock_sock() before reading the transport to avoid races and we should check if it is assigned.
>>
>> At that point I think is better to store this info in vsock_sock and not in the transport.
>You mean to store flag that zerocopy is enabled in 'vsock_sock', just reading it here, without touching transport?

Yep.

Thanks,
Stefano

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ