net/virtio-user: check negotiated features before set
authorEric Zhang <eric.zhang@windriver.com>
Wed, 29 Aug 2018 15:55:21 +0000 (11:55 -0400)
committerFerruh Yigit <ferruh.yigit@intel.com>
Wed, 12 Sep 2018 17:10:09 +0000 (19:10 +0200)
This patch checks negotiated features to see if necessary to offload
before set the tap device offload capabilities. It also checks if kernel
support the TUNSETOFFLOAD operation.

Fixes: 5e97e4202563 ("net/virtio-user: enable offloading")
Cc: stable@dpdk.org
Signed-off-by: Eric Zhang <eric.zhang@windriver.com>
Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
drivers/net/virtio/virtio_user/vhost_kernel.c
drivers/net/virtio/virtio_user/vhost_kernel_tap.c
drivers/net/virtio/virtio_user/vhost_kernel_tap.h

index b244409..d1be821 100644 (file)
@@ -147,8 +147,8 @@ prepare_vhost_memory_kernel(void)
         (1ULL << VIRTIO_NET_F_HOST_TSO6) |     \
         (1ULL << VIRTIO_NET_F_CSUM))
 
-static int
-tap_supporte_mq(void)
+static unsigned int
+tap_support_features(void)
 {
        int tapfd;
        unsigned int tap_features;
@@ -167,7 +167,7 @@ tap_supporte_mq(void)
        }
 
        close(tapfd);
-       return tap_features & IFF_MULTI_QUEUE;
+       return tap_features;
 }
 
 static int
@@ -181,6 +181,7 @@ vhost_kernel_ioctl(struct virtio_user_dev *dev,
        struct vhost_memory_kernel *vm = NULL;
        int vhostfd;
        unsigned int queue_sel;
+       unsigned int features;
 
        PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
 
@@ -234,17 +235,20 @@ vhost_kernel_ioctl(struct virtio_user_dev *dev,
        }
 
        if (!ret && req_kernel == VHOST_GET_FEATURES) {
+               features = tap_support_features();
                /* with tap as the backend, all these features are supported
                 * but not claimed by vhost-net, so we add them back when
                 * reporting to upper layer.
                 */
-               *((uint64_t *)arg) |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
-               *((uint64_t *)arg) |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
+               if (features & IFF_VNET_HDR) {
+                       *((uint64_t *)arg) |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
+                       *((uint64_t *)arg) |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
+               }
 
                /* vhost_kernel will not declare this feature, but it does
                 * support multi-queue.
                 */
-               if (tap_supporte_mq())
+               if (features & IFF_MULTI_QUEUE)
                        *(uint64_t *)arg |= (1ull << VIRTIO_NET_F_MQ);
        }
 
@@ -339,7 +343,7 @@ vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
                hdr_size = sizeof(struct virtio_net_hdr);
 
        tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
-                        (char *)dev->mac_addr);
+                        (char *)dev->mac_addr, dev->features);
        if (tapfd < 0) {
                PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
                return -1;
index 9ea7ade..a3faf1d 100644 (file)
 
 #include "vhost_kernel_tap.h"
 #include "../virtio_logs.h"
+#include "../virtio_pci.h"
+
+static int
+vhost_kernel_tap_set_offload(int fd, uint64_t features)
+{
+       unsigned int offload = 0;
+
+       if (features & (1ULL << VIRTIO_NET_F_GUEST_CSUM)) {
+               offload |= TUN_F_CSUM;
+               if (features & (1ULL << VIRTIO_NET_F_GUEST_TSO4))
+                       offload |= TUN_F_TSO4;
+               if (features & (1ULL << VIRTIO_NET_F_GUEST_TSO6))
+                       offload |= TUN_F_TSO6;
+               if (features & ((1ULL << VIRTIO_NET_F_GUEST_TSO4) |
+                       (1ULL << VIRTIO_NET_F_GUEST_TSO6)) &&
+                       (features & (1ULL << VIRTIO_NET_F_GUEST_ECN)))
+                       offload |= TUN_F_TSO_ECN;
+               if (features & (1ULL << VIRTIO_NET_F_GUEST_UFO))
+                       offload |= TUN_F_UFO;
+       }
+
+       if (offload != 0) {
+               /* Check if our kernel supports TUNSETOFFLOAD */
+               if (ioctl(fd, TUNSETOFFLOAD, 0) != 0 && errno == EINVAL) {
+                       PMD_DRV_LOG(ERR, "Kernel does't support TUNSETOFFLOAD\n");
+                       return -ENOTSUP;
+               }
+
+               if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
+                       offload &= ~TUN_F_UFO;
+                       if (ioctl(fd, TUNSETOFFLOAD, offload) != 0) {
+                               PMD_DRV_LOG(ERR, "TUNSETOFFLOAD ioctl() failed: %s\n",
+                                       strerror(errno));
+                               return -1;
+                       }
+               }
+       }
+
+       return 0;
+}
 
 int
 vhost_kernel_open_tap(char **p_ifname, int hdr_size, int req_mq,
-                        const char *mac)
+                        const char *mac, uint64_t features)
 {
        unsigned int tap_features;
        int sndbuf = INT_MAX;
        struct ifreq ifr;
        int tapfd;
-       unsigned int offload =
-                       TUN_F_CSUM |
-                       TUN_F_TSO4 |
-                       TUN_F_TSO6 |
-                       TUN_F_TSO_ECN |
-                       TUN_F_UFO;
 
        /* TODO:
         * 1. verify we can get/set vnet_hdr_len, tap_probe_vnet_hdr_len
@@ -90,13 +124,7 @@ vhost_kernel_open_tap(char **p_ifname, int hdr_size, int req_mq,
                goto error;
        }
 
-       /* TODO: before set the offload capabilities, we'd better (1) check
-        * negotiated features to see if necessary to offload; (2) query tap
-        * to see if it supports the offload capabilities.
-        */
-       if (ioctl(tapfd, TUNSETOFFLOAD, offload) != 0)
-               PMD_DRV_LOG(ERR, "TUNSETOFFLOAD ioctl() failed: %s",
-                          strerror(errno));
+       vhost_kernel_tap_set_offload(tapfd, features);
 
        memset(&ifr, 0, sizeof(ifr));
        ifr.ifr_hwaddr.sa_family = ARPHRD_ETHER;
index 01a026f..e0e95b4 100644 (file)
@@ -36,4 +36,4 @@
 #define PATH_NET_TUN   "/dev/net/tun"
 
 int vhost_kernel_open_tap(char **p_ifname, int hdr_size, int req_mq,
-                        const char *mac);
+                        const char *mac, uint64_t features);