Thanks for working on it. LGTM.

Could you please add inline for function netdev_linux_tap_batch_send?  That
makes it looks like function netdev_linux_sock_batch_send.
Or eliminate inline for function netdev_linux_sock_batch_send.

Thanks
Zhenyu Gao

2017-08-03 5:58 GMT+08:00 Ben Pfaff <[email protected]>:

> From: Zhenyu Gao <[email protected]>
>
> Sendmmsg can reduce cpu cycles in sending packets to kernel.
> Replace sendmsg with sendmmsg in function netdev_linux_send to send
> batch packets if sendmmsg is available.
>
> If kernel side doesn't support sendmmsg, will fallback to sendmsg.
>
>     netserver
> |------------|
> |            |
> |  container |
> |----veth----|
>           |
>           |        |------------|
>           |---veth-|   dpdk-ovs |      netperf
>                    |            |  |--------------|
>                    |----dpdk----|  | bare-metal   |
>                          |         |--------------|
>                          |              |
>                          |              |
>                         pnic-----------pnic
>
> Netperf was consumed to test the performance:
>
> 1)cmd:netperf -H remote-container -t UDP_STREAM -l 60 -- -m 1400
> result: netserver received 2383.21Mb(sendmsg)/2551.64Mb(sendmmsg)
>
> 2)cmd:netperf -H remote-container -t UDP_STREAM -l 60 -- -m 60
> result: netserver received 109.72Mb(sendmsg)/115.18Mb(sendmmsg)
>
> Sendmmsg show about 6% improvement in netperf UDP testing.
>
> Signed-off-by: Zhenyu Gao <[email protected]>
> Signed-off-by: Ben Pfaff <[email protected]>
> ---
>  lib/netdev-linux.c | 163 +++++++++++++++++++++++++++++-
> -----------------------
>  1 file changed, 89 insertions(+), 74 deletions(-)
>
> diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c
> index 5b4c643e4eec..19572a5bc949 100644
> --- a/lib/netdev-linux.c
> +++ b/lib/netdev-linux.c
> @@ -1182,13 +1182,89 @@ netdev_linux_rxq_drain(struct netdev_rxq *rxq_)
>      }
>  }
>
> -/* Sends 'buffer' on 'netdev'.  Returns 0 if successful, otherwise a
> positive
> +static inline int
> +netdev_linux_sock_batch_send(int sock, int ifindex,
> +                             struct dp_packet_batch *batch)
> +{
> +    /* We don't bother setting most fields in sockaddr_ll because the
> +     * kernel ignores them for SOCK_RAW. */
> +    struct sockaddr_ll sll = { .sll_family = AF_PACKET,
> +                               .sll_ifindex = ifindex };
> +
> +    struct mmsghdr *mmsg = xmalloc(sizeof(*mmsg) * batch->count);
> +    struct iovec *iov = xmalloc(sizeof(*iov) * batch->count);
> +
> +    for (int i = 0; i < batch->count; i++) {
> +        struct dp_packet *packet = batch->packets[i];
> +        iov[i].iov_base = dp_packet_data(packet);
> +        iov[i].iov_len = dp_packet_get_send_len(packet);
> +        mmsg[i].msg_hdr = (struct msghdr) { .msg_name = &sll,
> +                                            .msg_namelen = sizeof sll,
> +                                            .msg_iov = &iov[i],
> +                                            .msg_iovlen = 1 };
> +    }
> +
> +    int error = 0;
> +    for (uint32_t ofs = 0; ofs < batch->count; ) {
> +        ssize_t retval;
> +        do {
> +            retval = sendmmsg(sock, mmsg + ofs, batch->count - ofs, 0);
> +            error = retval < 0 ? errno : 0;
> +        } while (error == EINTR);
> +        if (error) {
> +            break;
> +        }
> +        ofs += retval;
> +    }
> +
> +    free(mmsg);
> +    free(iov);
> +    return error;
> +}
> +
> +/* Use the tap fd to send 'batch' to tap device 'netdev'.  Using the tap
> fd is
> + * essential, because packets sent to a tap device with an AF_PACKET
> socket
> + * will loop back to be *received* again on the tap device.  This doesn't
> occur
> + * on other interface types because we attach a socket filter to the rx
> + * socket. */
> +static int
> +netdev_linux_tap_batch_send(struct netdev *netdev_,
> +                            struct dp_packet_batch *batch)
> +{
> +    struct netdev_linux *netdev = netdev_linux_cast(netdev_);
> +    for (int i = 0; i < batch->count; i++) {
> +        struct dp_packet *packet = batch->packets[i];
> +        size_t size = dp_packet_get_send_len(packet);
> +        ssize_t retval;
> +        int error;
> +
> +        do {
> +            retval = write(netdev->tap_fd, dp_packet_data(packet), size);
> +            error = retval < 0 ? errno : 0;
> +        } while (error == EINTR);
> +
> +        if (error) {
> +            /* The Linux tap driver returns EIO if the device is not up.
> From
> +             * the OVS side this is not an error, so we ignore it;
> otherwise,
> +             * return the erro. */
> +            if (error != EIO) {
> +                return error;
> +            }
> +        } else if (retval != size) {
> +            VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE" "
> +                         "bytes of %"PRIuSIZE") on %s",
> +                         retval, size, netdev_get_name(netdev_));
> +            return EMSGSIZE;
> +        }
> +    }
> +    return 0;
> +}
> +
> +/* Sends 'batch' on 'netdev'.  Returns 0 if successful, otherwise a
> positive
>   * errno value.  Returns EAGAIN without blocking if the packet cannot be
> queued
>   * immediately.  Returns EMSGSIZE if a partial packet was transmitted or
> if
>   * the packet is too big or too small to transmit on the device.
>   *
> - * The caller retains ownership of 'buffer' in all cases.
> - *
>   * The kernel maintains a packet transmission queue, so the caller is not
>   * expected to do additional queuing of packets. */
>  static int
> @@ -1199,8 +1275,6 @@ netdev_linux_send(struct netdev *netdev_, int qid
> OVS_UNUSED,
>      int error = 0;
>      int sock = 0;
>
> -    struct sockaddr_ll sll;
> -    struct msghdr msg;
>      if (!is_tap_netdev(netdev_)) {
>          sock = af_packet_sock();
>          if (sock < 0) {
> @@ -1214,84 +1288,25 @@ netdev_linux_send(struct netdev *netdev_, int qid
> OVS_UNUSED,
>              goto free_batch;
>          }
>
> -        /* We don't bother setting most fields in sockaddr_ll because the
> -         * kernel ignores them for SOCK_RAW. */
> -        memset(&sll, 0, sizeof sll);
> -        sll.sll_family = AF_PACKET;
> -        sll.sll_ifindex = ifindex;
> -
> -        msg.msg_name = &sll;
> -        msg.msg_namelen = sizeof sll;
> -        msg.msg_iovlen = 1;
> -        msg.msg_control = NULL;
> -        msg.msg_controllen = 0;
> -        msg.msg_flags = 0;
> +        error = netdev_linux_sock_batch_send(sock, ifindex, batch);
> +    } else {
> +        error = netdev_linux_tap_batch_send(netdev_, batch);
>      }
> -
> -    /* 'i' is incremented only if there's no error */
> -    for (int i = 0; i < batch->count; ) {
> -        const void *data = dp_packet_data(batch->packets[i]);
> -        size_t size = dp_packet_get_send_len(batch->packets[i]);
> -        ssize_t retval;
> -
> -        if (!is_tap_netdev(netdev_)) {
> -            /* Use our AF_PACKET socket to send to this device. */
> -            struct iovec iov;
> -
> -            iov.iov_base = CONST_CAST(void *, data);
> -            iov.iov_len = size;
> -
> -            msg.msg_iov = &iov;
> -
> -            retval = sendmsg(sock, &msg, 0);
> +    if (error) {
> +        if (error == ENOBUFS) {
> +            /* The Linux AF_PACKET implementation never blocks waiting
> +             * for room for packets, instead returning ENOBUFS.
> +             * Translate this into EAGAIN for the caller. */
> +            error = EAGAIN;
>          } else {
> -            /* Use the tap fd to send to this device.  This is essential
> for
> -             * tap devices, because packets sent to a tap device with an
> -             * AF_PACKET socket will loop back to be *received* again on
> the
> -             * tap device.  This doesn't occur on other interface types
> -             * because we attach a socket filter to the rx socket. */
> -            struct netdev_linux *netdev = netdev_linux_cast(netdev_);
> -
> -            retval = write(netdev->tap_fd, data, size);
> -        }
> -
> -        if (retval < 0) {
> -            if (errno == EINTR) {
> -                /* The send was interrupted by a signal.  Retry the
> packet by
> -                 * continuing without incrementing 'i'.*/
> -                continue;
> -            } else if (errno == EIO && is_tap_netdev(netdev_)) {
> -                /* The Linux tap driver returns EIO if the device is not
> up.
> -                 * From the OVS side this is not an error, so ignore it.
> */
> -            } else {
> -                /* The Linux AF_PACKET implementation never blocks
> waiting for
> -                 * room for packets, instead returning ENOBUFS.
> Translate this
> -                 * into EAGAIN for the caller. */
> -                error = errno == ENOBUFS ? EAGAIN : errno;
> -                break;
> -            }
> -        } else if (retval != size) {
> -            VLOG_WARN_RL(&rl, "sent partial Ethernet packet (%"PRIuSIZE"
> bytes"
> -                              " of %"PRIuSIZE") on %s", retval, size,
> -                         netdev_get_name(netdev_));
> -            error = EMSGSIZE;
> -            break;
> -        }
> -
> -        /* Process the next packet in the batch */
> -        i++;
> -    }
> -
> -    if (error && error != EAGAIN) {
>              VLOG_WARN_RL(&rl, "error sending Ethernet packet on %s: %s",
>                           netdev_get_name(netdev_), ovs_strerror(error));
> +        }
>      }
>
>  free_batch:
>      dp_packet_delete_batch(batch, may_steal);
> -
>      return error;
> -
>  }
>
>  /* Registers with the poll loop to wake up from the next call to
> poll_block()
> --
> 2.10.2
>
>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to