datapath: net: make skb_gso_segment error handling more robust
[cascardo/ovs.git] / datapath / datapath.c
index ed9d7bd..31db569 100644 (file)
@@ -80,11 +80,12 @@ struct genl_multicast_group ovs_dp_vport_multicast_group = {
 
 /* Check if need to build a reply message.
  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
-static bool ovs_must_notify(struct genl_info *info,
-                           const struct genl_multicast_group *grp)
+static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
+                           unsigned int group)
 {
        return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
-               netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
+              genl_has_listeners(family, genl_info_net(info)->genl_sock,
+                                 group);
 }
 
 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
@@ -134,8 +135,10 @@ int lockdep_ovsl_is_held(void)
 #endif
 
 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
+                            const struct sw_flow_key *,
                             const struct dp_upcall_info *);
 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
+                                 const struct sw_flow_key *key,
                                  const struct dp_upcall_info *);
 
 /* Must be called with rcu_read_lock. */
@@ -173,7 +176,7 @@ const char *ovs_dp_name(const struct datapath *dp)
        return vport->ops->get_name(vport);
 }
 
-static int get_dpifindex(struct datapath *dp)
+static int get_dpifindex(const struct datapath *dp)
 {
        struct vport *local;
        int ifindex;
@@ -249,10 +252,9 @@ void ovs_dp_detach_port(struct vport *p)
 }
 
 /* Must be called with rcu_read_lock. */
-void ovs_dp_process_packet(struct sk_buff *skb)
+void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
 {
        const struct vport *p = OVS_CB(skb)->input_vport;
-       struct sw_flow_key *pkt_key = OVS_CB(skb)->pkt_key;
        struct datapath *dp = p->dp;
        struct sw_flow *flow;
        struct sw_flow_actions *sf_acts;
@@ -263,7 +265,7 @@ void ovs_dp_process_packet(struct sk_buff *skb)
        stats = this_cpu_ptr(dp->stats_percpu);
 
        /* Look up flow. */
-       flow = ovs_flow_tbl_lookup_stats(&dp->table, pkt_key, skb_get_hash(skb),
+       flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
                                         &n_mask_hit);
        if (unlikely(!flow)) {
                struct dp_upcall_info upcall;
@@ -274,7 +276,7 @@ void ovs_dp_process_packet(struct sk_buff *skb)
                upcall.portid = ovs_vport_find_upcall_portid(p, skb);
                upcall.egress_tun_info = NULL;
 
-               error = ovs_dp_upcall(dp, skb, &upcall);
+               error = ovs_dp_upcall(dp, skb, key, &upcall);
                if (unlikely(error))
                        kfree_skb(skb);
                else
@@ -284,10 +286,10 @@ void ovs_dp_process_packet(struct sk_buff *skb)
                goto out;
        }
 
-       ovs_flow_stats_update(flow, pkt_key->tp.flags, skb);
+       ovs_flow_stats_update(flow, key->tp.flags, skb);
 
        sf_acts = rcu_dereference(flow->sf_acts);
-       ovs_execute_actions(dp, skb, sf_acts);
+       ovs_execute_actions(dp, skb, key, sf_acts);
        stats_counter = &stats->n_hit;
 
 out:
@@ -299,22 +301,21 @@ out:
 }
 
 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
+                 const struct sw_flow_key *key,
                  const struct dp_upcall_info *upcall_info)
 {
        struct dp_stats_percpu *stats;
        int err;
 
-       BUG_ON(!OVS_CB(skb)->pkt_key);
-
        if (upcall_info->portid == 0) {
                err = -ENOTCONN;
                goto err;
        }
 
        if (!skb_is_gso(skb))
-               err = queue_userspace_packet(dp, skb, upcall_info);
+               err = queue_userspace_packet(dp, skb, key, upcall_info);
        else
-               err = queue_gso_packets(dp, skb, upcall_info);
+               err = queue_gso_packets(dp, skb, key, upcall_info);
        if (err)
                goto err;
 
@@ -331,33 +332,40 @@ err:
 }
 
 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
+                            const struct sw_flow_key *key,
                             const struct dp_upcall_info *upcall_info)
 {
        unsigned short gso_type = skb_shinfo(skb)->gso_type;
        struct sw_flow_key later_key;
        struct sk_buff *segs, *nskb;
+       struct ovs_skb_cb ovs_cb;
        int err;
 
+       ovs_cb = *OVS_CB(skb);
        segs = __skb_gso_segment(skb, NETIF_F_SG, false);
+       *OVS_CB(skb) = ovs_cb;
        if (IS_ERR(segs))
                return PTR_ERR(segs);
+       if (segs == NULL)
+               return -EINVAL;
 
        if (gso_type & SKB_GSO_UDP) {
                /* The initial flow key extracted by ovs_flow_key_extract()
                 * in this case is for a first fragment, so we need to
                 * properly mark later fragments.
                 */
-               later_key = *OVS_CB(skb)->pkt_key;
+               later_key = *key;
                later_key.ip.frag = OVS_FRAG_TYPE_LATER;
        }
 
        /* Queue all of the segments. */
        skb = segs;
        do {
+               *OVS_CB(skb) = ovs_cb;
                if (gso_type & SKB_GSO_UDP && skb != segs)
-                       OVS_CB(skb)->pkt_key = &later_key;
+                       key = &later_key;
 
-               err = queue_userspace_packet(dp, skb, upcall_info);
+               err = queue_userspace_packet(dp, skb, key, upcall_info);
                if (err)
                        break;
 
@@ -394,15 +402,15 @@ static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
 }
 
 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
+                                 const struct sw_flow_key *key,
                                  const struct dp_upcall_info *upcall_info)
 {
        struct ovs_header *upcall;
        struct sk_buff *nskb = NULL;
        struct sk_buff *user_skb = NULL; /* to be queued to userspace */
-       struct sw_flow_key *pkt_key = OVS_CB(skb)->pkt_key;
        struct nlattr *nla;
        struct genl_info info = {
-#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
+#ifdef HAVE_GENLMSG_NEW_UNICAST
                .dst_sk = ovs_dp_get_net(dp)->genl_sock,
 #endif
                .snd_portid = upcall_info->portid,
@@ -460,7 +468,7 @@ static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
        upcall->dp_ifindex = dp_ifindex;
 
        nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
-       err = ovs_nla_put_flow(pkt_key, pkt_key, user_skb);
+       err = ovs_nla_put_flow(key, key, user_skb);
        BUG_ON(err);
        nla_nest_end(user_skb, nla);
 
@@ -523,6 +531,7 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
        struct vport *input_vport;
        int len;
        int err;
+       bool log = !a[OVS_FLOW_ATTR_PROBE];
 
        err = -EINVAL;
        if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
@@ -556,17 +565,16 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
                goto err_kfree_skb;
 
        err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
-                                            &flow->key);
+                                            &flow->key, log);
        if (err)
                goto err_flow_free;
 
        err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
-                                  &flow->key, &acts);
+                                  &flow->key, &acts, log);
        if (err)
                goto err_flow_free;
 
        rcu_assign_pointer(flow->sf_acts, acts);
-       OVS_CB(packet)->pkt_key = &flow->key;
        OVS_CB(packet)->egress_tun_info = NULL;
        packet->priority = flow->key.phy.priority;
        packet->mark = flow->key.phy.skb_mark;
@@ -588,7 +596,7 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
        sf_acts = rcu_dereference(flow->sf_acts);
 
        local_bh_disable();
-       err = ovs_execute_actions(dp, packet, sf_acts);
+       err = ovs_execute_actions(dp, packet, &flow->key, sf_acts);
        local_bh_enable();
        rcu_read_unlock();
 
@@ -631,7 +639,7 @@ static struct genl_family dp_packet_genl_family = {
        .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
 };
 
-static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
+static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
                         struct ovs_dp_megaflow_stats *mega_stats)
 {
        int i;
@@ -809,7 +817,8 @@ static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *act
 {
        struct sk_buff *skb;
 
-       if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
+       if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
+                                       GROUP_ID(&ovs_dp_flow_multicast_group)))
                return NULL;
 
        skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
@@ -853,15 +862,17 @@ static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
        struct sw_flow_actions *acts;
        struct sw_flow_match match;
        int error;
+       bool log = !a[OVS_FLOW_ATTR_PROBE];
 
        /* Must have key and actions. */
        error = -EINVAL;
        if (!a[OVS_FLOW_ATTR_KEY]) {
-               OVS_NLERR("Flow key attribute not present in new flow.\n");
+               OVS_NLERR(log, "Flow key attribute not present in new flow.");
                goto error;
        }
        if (!a[OVS_FLOW_ATTR_ACTIONS]) {
-               OVS_NLERR("Flow actions attribute not present in new flow.\n");
+               OVS_NLERR(log,
+                         "Flow actions attribute not present in new flow.");
                goto error;
        }
 
@@ -875,8 +886,8 @@ static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 
        /* Extract key. */
        ovs_match_init(&match, &new_flow->unmasked_key, &mask);
-       error = ovs_nla_get_match(&match,
-                                 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
+       error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
+                                 a[OVS_FLOW_ATTR_MASK], log);
        if (error)
                goto err_kfree_flow;
 
@@ -884,10 +895,13 @@ static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
 
        /* Validate actions. */
        error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
-                                    &acts);
+                                    &acts, log);
        if (error) {
-               OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
-               goto err_kfree_acts;
+               OVS_NLERR(
+                       log,
+                       "Flow actions may not be safe on all matching packets."
+                       );
+               goto err_kfree_flow;
        }
 
        reply = ovs_flow_cmd_alloc_info(acts, info, false);
@@ -982,16 +996,18 @@ error:
 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
                                                const struct sw_flow_key *key,
-                                               const struct sw_flow_mask *mask)
+                                               const struct sw_flow_mask *mask,
+                                               bool log)
 {
        struct sw_flow_actions *acts;
        struct sw_flow_key masked_key;
        int error;
 
        ovs_flow_mask_key(&masked_key, key, mask);
-       error = ovs_nla_copy_actions(a, &masked_key, &acts);
+       error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
        if (error) {
-               OVS_NLERR("Actions may not be safe on all matching packets.\n");
+               OVS_NLERR(log,
+                         "Actions may not be safe on all matching packets.");
                return ERR_PTR(error);
        }
 
@@ -1010,23 +1026,25 @@ static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
        struct sw_flow_actions *old_acts = NULL, *acts = NULL;
        struct sw_flow_match match;
        int error;
+       bool log = !a[OVS_FLOW_ATTR_PROBE];
 
        /* Extract key. */
        error = -EINVAL;
        if (!a[OVS_FLOW_ATTR_KEY]) {
-               OVS_NLERR("Flow key attribute not present in set flow.\n");
+               OVS_NLERR(log, "Flow key attribute not present in set flow.");
                goto error;
        }
 
        ovs_match_init(&match, &key, &mask);
-       error = ovs_nla_get_match(&match,
-                                 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
+       error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
+                                 a[OVS_FLOW_ATTR_MASK], log);
        if (error)
                goto error;
 
        /* Validate actions. */
        if (a[OVS_FLOW_ATTR_ACTIONS]) {
-               acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask);
+               acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
+                                       log);
                if (IS_ERR(acts)) {
                        error = PTR_ERR(acts);
                        goto error;
@@ -1107,14 +1125,16 @@ static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
        struct datapath *dp;
        struct sw_flow_match match;
        int err;
+       bool log = !a[OVS_FLOW_ATTR_PROBE];
 
        if (!a[OVS_FLOW_ATTR_KEY]) {
-               OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
+               OVS_NLERR(log,
+                         "Flow get message rejected, Key attribute missing.");
                return -EINVAL;
        }
 
        ovs_match_init(&match, &key, NULL);
-       err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
+       err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log);
        if (err)
                return err;
 
@@ -1155,10 +1175,12 @@ static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
        struct datapath *dp;
        struct sw_flow_match match;
        int err;
+       bool log = !a[OVS_FLOW_ATTR_PROBE];
 
        if (likely(a[OVS_FLOW_ATTR_KEY])) {
                ovs_match_init(&match, &key, NULL);
-               err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
+               err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
+                                       log);
                if (unlikely(err))
                        return err;
        }
@@ -1249,8 +1271,10 @@ static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
 
 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
        [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
+       [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
        [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
        [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
+       [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
 };
 
 static struct genl_ops dp_flow_genl_ops[] = {
@@ -1351,7 +1375,7 @@ static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
 
 /* Called with rcu_read_lock or ovs_mutex. */
 static struct datapath *lookup_datapath(struct net *net,
-                                       struct ovs_header *ovs_header,
+                                       const struct ovs_header *ovs_header,
                                        struct nlattr *a[OVS_DP_ATTR_MAX + 1])
 {
        struct datapath *dp;
@@ -1379,7 +1403,7 @@ static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *in
        dp->user_features = 0;
 }
 
-static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
+static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
 {
        if (a[OVS_DP_ATTR_USER_FEATURES])
                dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
@@ -1740,7 +1764,7 @@ struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
 
 /* Called with ovs_mutex or RCU read lock. */
 static struct vport *lookup_vport(struct net *net,
-                                 struct ovs_header *ovs_header,
+                                 const struct ovs_header *ovs_header,
                                  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
 {
        struct datapath *dp;
@@ -1831,10 +1855,6 @@ static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
        if (IS_ERR(vport))
                goto exit_unlock_free;
 
-       err = 0;
-       if (a[OVS_VPORT_ATTR_STATS])
-               ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
-
        err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
                                      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
        BUG_ON(err < 0);
@@ -1878,10 +1898,6 @@ static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
                        goto exit_unlock_free;
        }
 
-       if (a[OVS_VPORT_ATTR_STATS])
-               ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
-
-
        if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
                err = ovs_vport_set_upcall_portids(vport,
                                                   a[OVS_VPORT_ATTR_UPCALL_PID]);