Re: [PATCH RFC net-next v4 1/8] vsock/dgram: generalize recvmsg and drop transport->dgram_dequeue

[Arseniy Krasnov <oxffffaa@gmail.com>]

On 22.06.2023 17:51, Stefano Garzarella wrote:
> On Sun, Jun 11, 2023 at 11:43:15PM +0300, Arseniy Krasnov wrote:
>> Hello Bobby! Thanks for this patchset! Small comment below:
>>
>> On 10.06.2023 03:58, Bobby Eshleman wrote:
>>> This commit drops the transport->dgram_dequeue callback and makes
>>> vsock_dgram_recvmsg() generic. It also adds additional transport
>>> callbacks for use by the generic vsock_dgram_recvmsg(), such as for
>>> parsing skbs for CID/port which vary in format per transport.
>>>
>>> Signed-off-by: Bobby Eshleman <bobby.eshleman@bytedance.com>
>>> ---
>>>  drivers/vhost/vsock.c                   |  4 +-
>>>  include/linux/virtio_vsock.h            |  3 ++
>>>  include/net/af_vsock.h                  | 13 ++++++-
>>>  net/vmw_vsock/af_vsock.c                | 51 ++++++++++++++++++++++++-
>>>  net/vmw_vsock/hyperv_transport.c        | 17 +++++++--
>>>  net/vmw_vsock/virtio_transport.c        |  4 +-
>>>  net/vmw_vsock/virtio_transport_common.c | 18 +++++++++
>>>  net/vmw_vsock/vmci_transport.c          | 68 +++++++++++++--------------------
>>>  net/vmw_vsock/vsock_loopback.c          |  4 +-
>>>  9 files changed, 132 insertions(+), 50 deletions(-)
>>>
>>> diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c
>>> index 6578db78f0ae..c8201c070b4b 100644
>>> --- a/drivers/vhost/vsock.c
>>> +++ b/drivers/vhost/vsock.c
>>> @@ -410,9 +410,11 @@ static struct virtio_transport vhost_transport = {
>>>          .cancel_pkt               = vhost_transport_cancel_pkt,
>>>
>>>          .dgram_enqueue            = virtio_transport_dgram_enqueue,
>>> -        .dgram_dequeue            = virtio_transport_dgram_dequeue,
>>>          .dgram_bind               = virtio_transport_dgram_bind,
>>>          .dgram_allow              = virtio_transport_dgram_allow,
>>> +        .dgram_get_cid          = virtio_transport_dgram_get_cid,
>>> +        .dgram_get_port          = virtio_transport_dgram_get_port,
>>> +        .dgram_get_length      = virtio_transport_dgram_get_length,
>>>
>>>          .stream_enqueue           = virtio_transport_stream_enqueue,
>>>          .stream_dequeue           = virtio_transport_stream_dequeue,
>>> diff --git a/include/linux/virtio_vsock.h b/include/linux/virtio_vsock.h
>>> index c58453699ee9..23521a318cf0 100644
>>> --- a/include/linux/virtio_vsock.h
>>> +++ b/include/linux/virtio_vsock.h
>>> @@ -219,6 +219,9 @@ bool virtio_transport_stream_allow(u32 cid, u32 port);
>>>  int virtio_transport_dgram_bind(struct vsock_sock *vsk,
>>>                  struct sockaddr_vm *addr);
>>>  bool virtio_transport_dgram_allow(u32 cid, u32 port);
>>> +int virtio_transport_dgram_get_cid(struct sk_buff *skb, unsigned int *cid);
>>> +int virtio_transport_dgram_get_port(struct sk_buff *skb, unsigned int *port);
>>> +int virtio_transport_dgram_get_length(struct sk_buff *skb, size_t *len);
>>>
>>>  int virtio_transport_connect(struct vsock_sock *vsk);
>>>
>>> diff --git a/include/net/af_vsock.h b/include/net/af_vsock.h
>>> index 0e7504a42925..7bedb9ee7e3e 100644
>>> --- a/include/net/af_vsock.h
>>> +++ b/include/net/af_vsock.h
>>> @@ -120,11 +120,20 @@ struct vsock_transport {
>>>
>>>      /* DGRAM. */
>>>      int (*dgram_bind)(struct vsock_sock *, struct sockaddr_vm *);
>>> -    int (*dgram_dequeue)(struct vsock_sock *vsk, struct msghdr *msg,
>>> -                 size_t len, int flags);
>>>      int (*dgram_enqueue)(struct vsock_sock *, struct sockaddr_vm *,
>>>                   struct msghdr *, size_t len);
>>>      bool (*dgram_allow)(u32 cid, u32 port);
>>> +    int (*dgram_get_cid)(struct sk_buff *skb, unsigned int *cid);
>>> +    int (*dgram_get_port)(struct sk_buff *skb, unsigned int *port);
>>> +    int (*dgram_get_length)(struct sk_buff *skb, size_t *length);
>>> +
>>> +    /* The number of bytes into the buffer at which the payload starts, as
>>> +     * first seen by the receiving socket layer. For example, if the
>>> +     * transport presets the skb pointers using skb_pull(sizeof(header))
>>> +     * than this would be zero, otherwise it would be the size of the
>>> +     * header.
>>> +     */
>>> +    const size_t dgram_payload_offset;
>>>
>>>      /* STREAM. */
>>>      /* TODO: stream_bind() */
>>> diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
>>> index efb8a0937a13..ffb4dd8b6ea7 100644
>>> --- a/net/vmw_vsock/af_vsock.c
>>> +++ b/net/vmw_vsock/af_vsock.c
>>> @@ -1271,11 +1271,15 @@ static int vsock_dgram_connect(struct socket *sock,
>>>  int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
>>>              size_t len, int flags)
>>>  {
>>> +    const struct vsock_transport *transport;
>>>  #ifdef CONFIG_BPF_SYSCALL
>>>      const struct proto *prot;
>>>  #endif
>>>      struct vsock_sock *vsk;
>>> +    struct sk_buff *skb;
>>> +    size_t payload_len;
>>>      struct sock *sk;
>>> +    int err;
>>>
>>>      sk = sock->sk;
>>>      vsk = vsock_sk(sk);
>>> @@ -1286,7 +1290,52 @@ int vsock_dgram_recvmsg(struct socket *sock, struct msghdr *msg,
>>>          return prot->recvmsg(sk, msg, len, flags, NULL);
>>>  #endif
>>>
>>> -    return vsk->transport->dgram_dequeue(vsk, msg, len, flags);
>>> +    if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
>>> +        return -EOPNOTSUPP;
>>> +
>>> +    transport = vsk->transport;
>>> +
>>> +    /* Retrieve the head sk_buff from the socket's receive queue. */
>>> +    err = 0;
>>> +    skb = skb_recv_datagram(sk_vsock(vsk), flags, &err);
>>> +    if (!skb)
>>> +        return err;
>>> +
>>> +    err = transport->dgram_get_length(skb, &payload_len);
> 
> What about ssize_t return value here?
> 
> Or maybe a single callback that return both length and offset?
> 
> .dgram_get_payload_info(skb, &payload_len, &payload_off)

Just architectural question:

May be we can avoid this callback for length? IIUC concept of skbuff is that
current level of network stack already have pointer to its data 'skb->data' and
length of the payload 'skb->len' (both are set by previous stack handler - transport in
this case), so here we can use just 'skb->len' and thats all. There is no need to ask
lower level of network stack for length of payload? I see that VMCI stores metadata
with payload in 'data' buffer, but may be it is more correctly to do 'skb_pull()'
in vmci before inserting skbuff to sockets queue? In this case field with dgram payload
offset could be removed from transport.

Thanks, Arseniy

> 
>>> +    if (err)
>>> +        goto out;
>>> +
>>> +    if (payload_len > len) {
>>> +        payload_len = len;
>>> +        msg->msg_flags |= MSG_TRUNC;
>>> +    }
>>> +
>>> +    /* Place the datagram payload in the user's iovec. */
>>> +    err = skb_copy_datagram_msg(skb, transport->dgram_payload_offset, msg, payload_len);
>>> +    if (err)
>>> +        goto out;
>>> +
>>> +    if (msg->msg_name) {
>>> +        /* Provide the address of the sender. */
>>> +        DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
>>> +        unsigned int cid, port;
>>> +
>>> +        err = transport->dgram_get_cid(skb, &cid);
>>> +        if (err)
>>> +            goto out;
>>> +
>>> +        err = transport->dgram_get_port(skb, &port);
>>> +        if (err)
>>> +            goto out;
>>
>> Maybe we can merge 'dgram_get_cid' and 'dgram_get_port' to a single callback? Because I see that this is
>> the only place where both are used (correct me if i'm wrong) and logically both operates with addresses:
>> CID and port. E.g. something like that: dgram_get_cid_n_port().
> 
> What about .dgram_addr_init(struct sk_buff *skb, struct sockaddr_vm *addr)
> and the transport can set cid and port?
> 
>>
>> Moreover, I'm not sure, but is it good "tradeoff" here: remove transport specific callback for dgram receive
>> where we already have 'msghdr' with both data buffer and buffer for 'sockaddr_vm' and instead of it add new
>> several fields (callbacks) to transports like dgram_get_cid(), dgram_get_port()? I agree, that in each transport
>> specific callback we will have same copying logic by calling 'skb_copy_datagram_msg()' and filling address
>> by using 'vsock_addr_init()', but in this case we don't need to update transports too much. For example HyperV
>> still unchanged as it does not support SOCK_DGRAM. For VMCI You just need to add 'vsock_addr_init()' logic
>> to it's dgram dequeue callback.
>>
>> What do You think?
> 
> Honestly, I'd rather avoid duplicate code than reduce changes in
> transports that don't support dgram.
> 
> One thing I do agree on though is minimizing the number of callbacks
> to call to reduce the number of indirection (more performance?).
> 
> Thanks,
> Stefano
> 
>>
>> Thanks, Arseniy
>>
>>> +
>>> +        vsock_addr_init(vm_addr, cid, port);
>>> +        msg->msg_namelen = sizeof(*vm_addr);
>>> +    }
>>> +    err = payload_len;
>>> +
>>> +out:
>>> +    skb_free_datagram(&vsk->sk, skb);
>>> +    return err;
>>>  }
>>>  EXPORT_SYMBOL_GPL(vsock_dgram_recvmsg);
>>>
>>> diff --git a/net/vmw_vsock/hyperv_transport.c b/net/vmw_vsock/hyperv_transport.c
>>> index 7cb1a9d2cdb4..ff6e87e25fa0 100644
>>> --- a/net/vmw_vsock/hyperv_transport.c
>>> +++ b/net/vmw_vsock/hyperv_transport.c
>>> @@ -556,8 +556,17 @@ static int hvs_dgram_bind(struct vsock_sock *vsk, struct sockaddr_vm *addr)
>>>      return -EOPNOTSUPP;
>>>  }
>>>
>>> -static int hvs_dgram_dequeue(struct vsock_sock *vsk, struct msghdr *msg,
>>> -                 size_t len, int flags)
>>> +static int hvs_dgram_get_cid(struct sk_buff *skb, unsigned int *cid)
>>> +{
>>> +    return -EOPNOTSUPP;
>>> +}
>>> +
>>> +static int hvs_dgram_get_port(struct sk_buff *skb, unsigned int *port)
>>> +{
>>> +    return -EOPNOTSUPP;
>>> +}
>>> +
>>> +static int hvs_dgram_get_length(struct sk_buff *skb, size_t *len)
>>>  {
>>>      return -EOPNOTSUPP;
>>>  }
>>> @@ -833,7 +842,9 @@ static struct vsock_transport hvs_transport = {
>>>      .shutdown                 = hvs_shutdown,
>>>
>>>      .dgram_bind               = hvs_dgram_bind,
>>> -    .dgram_dequeue            = hvs_dgram_dequeue,
>>> +    .dgram_get_cid          = hvs_dgram_get_cid,
>>> +    .dgram_get_port          = hvs_dgram_get_port,
>>> +    .dgram_get_length      = hvs_dgram_get_length,
>>>      .dgram_enqueue            = hvs_dgram_enqueue,
>>>      .dgram_allow              = hvs_dgram_allow,
>>>
>>> diff --git a/net/vmw_vsock/virtio_transport.c b/net/vmw_vsock/virtio_transport.c
>>> index e95df847176b..5763cdf13804 100644
>>> --- a/net/vmw_vsock/virtio_transport.c
>>> +++ b/net/vmw_vsock/virtio_transport.c
>>> @@ -429,9 +429,11 @@ static struct virtio_transport virtio_transport = {
>>>          .cancel_pkt               = virtio_transport_cancel_pkt,
>>>
>>>          .dgram_bind               = virtio_transport_dgram_bind,
>>> -        .dgram_dequeue            = virtio_transport_dgram_dequeue,
>>>          .dgram_enqueue            = virtio_transport_dgram_enqueue,
>>>          .dgram_allow              = virtio_transport_dgram_allow,
>>> +        .dgram_get_cid          = virtio_transport_dgram_get_cid,
>>> +        .dgram_get_port          = virtio_transport_dgram_get_port,
>>> +        .dgram_get_length      = virtio_transport_dgram_get_length,
>>>
>>>          .stream_dequeue           = virtio_transport_stream_dequeue,
>>>          .stream_enqueue           = virtio_transport_stream_enqueue,
>>> diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>>> index b769fc258931..e6903c719964 100644
>>> --- a/net/vmw_vsock/virtio_transport_common.c
>>> +++ b/net/vmw_vsock/virtio_transport_common.c
>>> @@ -797,6 +797,24 @@ int virtio_transport_dgram_bind(struct vsock_sock *vsk,
>>>  }
>>>  EXPORT_SYMBOL_GPL(virtio_transport_dgram_bind);
>>>
>>> +int virtio_transport_dgram_get_cid(struct sk_buff *skb, unsigned int *cid)
>>> +{
>>> +    return -EOPNOTSUPP;
>>> +}
>>> +EXPORT_SYMBOL_GPL(virtio_transport_dgram_get_cid);
>>> +
>>> +int virtio_transport_dgram_get_port(struct sk_buff *skb, unsigned int *port)
>>> +{
>>> +    return -EOPNOTSUPP;
>>> +}
>>> +EXPORT_SYMBOL_GPL(virtio_transport_dgram_get_port);
>>> +
>>> +int virtio_transport_dgram_get_length(struct sk_buff *skb, size_t *len)
>>> +{
>>> +    return -EOPNOTSUPP;
>>> +}
>>> +EXPORT_SYMBOL_GPL(virtio_transport_dgram_get_length);
>>> +
>>>  bool virtio_transport_dgram_allow(u32 cid, u32 port)
>>>  {
>>>      return false;
>>> diff --git a/net/vmw_vsock/vmci_transport.c b/net/vmw_vsock/vmci_transport.c
>>> index b370070194fa..bbc63826bf48 100644
>>> --- a/net/vmw_vsock/vmci_transport.c
>>> +++ b/net/vmw_vsock/vmci_transport.c
>>> @@ -1731,57 +1731,40 @@ static int vmci_transport_dgram_enqueue(
>>>      return err - sizeof(*dg);
>>>  }
>>>
>>> -static int vmci_transport_dgram_dequeue(struct vsock_sock *vsk,
>>> -                    struct msghdr *msg, size_t len,
>>> -                    int flags)
>>> +static int vmci_transport_dgram_get_cid(struct sk_buff *skb, unsigned int *cid)
>>>  {
>>> -    int err;
>>>      struct vmci_datagram *dg;
>>> -    size_t payload_len;
>>> -    struct sk_buff *skb;
>>>
>>> -    if (flags & MSG_OOB || flags & MSG_ERRQUEUE)
>>> -        return -EOPNOTSUPP;
>>> +    dg = (struct vmci_datagram *)skb->data;
>>> +    if (!dg)
>>> +        return -EINVAL;
>>>
>>> -    /* Retrieve the head sk_buff from the socket's receive queue. */
>>> -    err = 0;
>>> -    skb = skb_recv_datagram(&vsk->sk, flags, &err);
>>> -    if (!skb)
>>> -        return err;
>>> +    *cid = dg->src.context;
>>> +    return 0;
>>> +}
>>> +
>>> +static int vmci_transport_dgram_get_port(struct sk_buff *skb, unsigned int *port)
>>> +{
>>> +    struct vmci_datagram *dg;
>>>
>>>      dg = (struct vmci_datagram *)skb->data;
>>>      if (!dg)
>>> -        /* err is 0, meaning we read zero bytes. */
>>> -        goto out;
>>> -
>>> -    payload_len = dg->payload_size;
>>> -    /* Ensure the sk_buff matches the payload size claimed in the packet. */
>>> -    if (payload_len != skb->len - sizeof(*dg)) {
>>> -        err = -EINVAL;
>>> -        goto out;
>>> -    }
>>> +        return -EINVAL;
>>>
>>> -    if (payload_len > len) {
>>> -        payload_len = len;
>>> -        msg->msg_flags |= MSG_TRUNC;
>>> -    }
>>> +    *port = dg->src.resource;
>>> +    return 0;
>>> +}
>>>
>>> -    /* Place the datagram payload in the user's iovec. */
>>> -    err = skb_copy_datagram_msg(skb, sizeof(*dg), msg, payload_len);
>>> -    if (err)
>>> -        goto out;
>>> +static int vmci_transport_dgram_get_length(struct sk_buff *skb, size_t *len)
>>> +{
>>> +    struct vmci_datagram *dg;
>>>
>>> -    if (msg->msg_name) {
>>> -        /* Provide the address of the sender. */
>>> -        DECLARE_SOCKADDR(struct sockaddr_vm *, vm_addr, msg->msg_name);
>>> -        vsock_addr_init(vm_addr, dg->src.context, dg->src.resource);
>>> -        msg->msg_namelen = sizeof(*vm_addr);
>>> -    }
>>> -    err = payload_len;
>>> +    dg = (struct vmci_datagram *)skb->data;
>>> +    if (!dg)
>>> +        return -EINVAL;
>>>
>>> -out:
>>> -    skb_free_datagram(&vsk->sk, skb);
>>> -    return err;
>>> +    *len = dg->payload_size;
>>> +    return 0;
>>>  }
>>>
>>>  static bool vmci_transport_dgram_allow(u32 cid, u32 port)
>>> @@ -2040,9 +2023,12 @@ static struct vsock_transport vmci_transport = {
>>>      .release = vmci_transport_release,
>>>      .connect = vmci_transport_connect,
>>>      .dgram_bind = vmci_transport_dgram_bind,
>>> -    .dgram_dequeue = vmci_transport_dgram_dequeue,
>>>      .dgram_enqueue = vmci_transport_dgram_enqueue,
>>>      .dgram_allow = vmci_transport_dgram_allow,
>>> +    .dgram_get_cid = vmci_transport_dgram_get_cid,
>>> +    .dgram_get_port = vmci_transport_dgram_get_port,
>>> +    .dgram_get_length = vmci_transport_dgram_get_length,
>>> +    .dgram_payload_offset = sizeof(struct vmci_datagram),
>>>      .stream_dequeue = vmci_transport_stream_dequeue,
>>>      .stream_enqueue = vmci_transport_stream_enqueue,
>>>      .stream_has_data = vmci_transport_stream_has_data,
>>> diff --git a/net/vmw_vsock/vsock_loopback.c b/net/vmw_vsock/vsock_loopback.c
>>> index 5c6360df1f31..2f3cabc79ee5 100644
>>> --- a/net/vmw_vsock/vsock_loopback.c
>>> +++ b/net/vmw_vsock/vsock_loopback.c
>>> @@ -62,9 +62,11 @@ static struct virtio_transport loopback_transport = {
>>>          .cancel_pkt               = vsock_loopback_cancel_pkt,
>>>
>>>          .dgram_bind               = virtio_transport_dgram_bind,
>>> -        .dgram_dequeue            = virtio_transport_dgram_dequeue,
>>>          .dgram_enqueue            = virtio_transport_dgram_enqueue,
>>>          .dgram_allow              = virtio_transport_dgram_allow,
>>> +        .dgram_get_cid          = virtio_transport_dgram_get_cid,
>>> +        .dgram_get_port          = virtio_transport_dgram_get_port,
>>> +        .dgram_get_length      = virtio_transport_dgram_get_length,
>>>
>>>          .stream_dequeue           = virtio_transport_stream_dequeue,
>>>          .stream_enqueue           = virtio_transport_stream_enqueue,
>>>
>>
>