I think it should NULL out bufp in the beginning in case it errors out
earlier.  Otherwise looks fine to me.

Ethan

On Wed, Jul 20, 2011 at 13:16, Ben Pfaff <[email protected]> wrote:
> Until now, each attempt to receive a message from a Netlink socket has
> taken at least two system calls, one to check the size of the message to
> be received and a second one to delete the message from the socket buffer.
> This commit switches to a new strategy that requires only one system call
> per message received.
>
> In my testing this increases the maximum flow setups per second by a little
> over 10%.
> ---
>  lib/netlink-socket.c |  110 +++++++++++++++++++++++--------------------------
>  1 files changed, 52 insertions(+), 58 deletions(-)
>
> diff --git a/lib/netlink-socket.c b/lib/netlink-socket.c
> index 95b2401..7ad5ed9 100644
> --- a/lib/netlink-socket.c
> +++ b/lib/netlink-socket.c
> @@ -38,7 +38,7 @@ VLOG_DEFINE_THIS_MODULE(netlink_socket);
>
>  COVERAGE_DEFINE(netlink_overflow);
>  COVERAGE_DEFINE(netlink_received);
> -COVERAGE_DEFINE(netlink_recv_retry);
> +COVERAGE_DEFINE(netlink_recv_jumbo);
>  COVERAGE_DEFINE(netlink_send);
>  COVERAGE_DEFINE(netlink_sent);
>
> @@ -257,73 +257,67 @@ STRESS_OPTION(
>  static int
>  nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
>  {
> -    uint8_t tmp;
> -    ssize_t bufsize = 2048;
> -    ssize_t nbytes, nbytes2;
> -    struct ofpbuf *buf;
> +    /* We can't accurately predict the size of the data to be received.  Most
> +     * received data will fit in a 2 kB buffer, so we allocate that much 
> space.
> +     * In case the data is actually bigger than that, we make available 
> enough
> +     * additional space to allow Netlink messages to be up to 64 kB long (a
> +     * reasonable figure since that's the maximum length of a Netlink
> +     * attribute). */
> +    enum { MAX_SIZE = 65536 };
> +    enum { HEAD_SIZE = 2048 };
> +    enum { TAIL_SIZE = MAX_SIZE - HEAD_SIZE };
> +
>     struct nlmsghdr *nlmsghdr;
> -    struct iovec iov;
> -    struct msghdr msg = {
> -        .msg_name = NULL,
> -        .msg_namelen = 0,
> -        .msg_iov = &iov,
> -        .msg_iovlen = 1,
> -        .msg_control = NULL,
> -        .msg_controllen = 0,
> -        .msg_flags = 0
> -    };
> +    uint8_t tail[TAIL_SIZE];
> +    struct iovec iov[2];
> +    struct ofpbuf *buf;
> +    struct msghdr msg;
> +    ssize_t retval;
> +
> +    buf = ofpbuf_new(HEAD_SIZE);
> +
> +    iov[0].iov_base = buf->data;
> +    iov[0].iov_len = HEAD_SIZE;
> +    iov[1].iov_base = tail;
> +    iov[1].iov_len = TAIL_SIZE;
> +
> +    memset(&msg, 0, sizeof msg);
> +    msg.msg_iov = iov;
> +    msg.msg_iovlen = 2;
>
> -    buf = ofpbuf_new(bufsize);
> -    *bufp = NULL;
> -
> -try_again:
> -    /* Attempt to read the message.  We don't know the size of the data
> -     * yet, so we take a guess at 2048.  If we're wrong, we keep trying
> -     * and doubling the buffer size each time.
> -     */
> -    nlmsghdr = ofpbuf_put_uninit(buf, bufsize);
> -    iov.iov_base = nlmsghdr;
> -    iov.iov_len = bufsize;
>     do {
> -        nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | 
> MSG_PEEK);
> -    } while (nbytes < 0 && errno == EINTR);
> -    if (nbytes < 0) {
> +        retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
> +    } while (retval < 0 && errno == EINTR);
> +
> +    if (retval < 0) {
> +        int error = errno;
> +        if (error == ENOBUFS) {
> +            /* Socket receive buffer overflow dropped one or more messages 
> that
> +             * the kernel tried to send to us. */
> +            COVERAGE_INC(netlink_overflow);
> +        }
>         ofpbuf_delete(buf);
> -        return errno;
> +        return error;
>     }
> +
>     if (msg.msg_flags & MSG_TRUNC) {
> -        COVERAGE_INC(netlink_recv_retry);
> -        bufsize *= 2;
> -        ofpbuf_reinit(buf, bufsize);
> -        goto try_again;
> +        VLOG_ERR_RL(&rl, "truncated message (longer than %d bytes)", 
> MAX_SIZE);
> +        ofpbuf_delete(buf);
> +        return E2BIG;
>     }
> -    buf->size = nbytes;
>
> -    /* We successfully read the message, so recv again to clear the queue */
> -    iov.iov_base = &tmp;
> -    iov.iov_len = 1;
> -    do {
> -        nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT);
> -    } while (nbytes2 < 0 && errno == EINTR);
> -    if (nbytes2 < 0) {
> -        if (errno == ENOBUFS) {
> -            /* The kernel is notifying us that a message it tried to send to 
> us
> -             * was dropped.  We have to pass this along to the caller in case
> -             * it wants to retry a request.  So kill the buffer, which we can
> -             * re-read next time. */
> -            COVERAGE_INC(netlink_overflow);
> -            ofpbuf_delete(buf);
> -            return ENOBUFS;
> -        } else {
> -            VLOG_ERR_RL(&rl, "failed to remove nlmsg from socket: %s\n",
> -                        strerror(errno));
> -        }
> +    ofpbuf_put_uninit(buf, MIN(retval, HEAD_SIZE));
> +    if (retval > HEAD_SIZE) {
> +        COVERAGE_INC(netlink_recv_jumbo);
> +        ofpbuf_put(buf, tail, retval - HEAD_SIZE);
>     }
> -    if (nbytes < sizeof *nlmsghdr
> +
> +    nlmsghdr = buf->data;
> +    if (retval < sizeof *nlmsghdr
>         || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
> -        || nlmsghdr->nlmsg_len > nbytes) {
> +        || nlmsghdr->nlmsg_len > retval) {
>         VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)",
> -                    bufsize, NLMSG_HDRLEN);
> +                    retval, NLMSG_HDRLEN);
>         ofpbuf_delete(buf);
>         return EPROTO;
>     }
> --
> 1.7.4.4
>
> _______________________________________________
> dev mailing list
> [email protected]
> http://openvswitch.org/mailman/listinfo/dev
>
_______________________________________________
dev mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/dev

Reply via email to