mpls: Netlink commands to add, remove, and dump routes
authorEric W. Biederman <ebiederm@xmission.com>
Wed, 4 Mar 2015 01:13:56 +0000 (19:13 -0600)
committerDavid S. Miller <davem@davemloft.net>
Wed, 4 Mar 2015 05:26:06 +0000 (00:26 -0500)
This change adds two new netlink routing attributes:
RTA_VIA and RTA_NEWDST.

RTA_VIA specifies the specifies the next machine to send a packet to
like RTA_GATEWAY.  RTA_VIA differs from RTA_GATEWAY in that it
includes the address family of the address of the next machine to send
a packet to.  Currently the MPLS code supports addresses in AF_INET,
AF_INET6 and AF_PACKET.  For AF_INET and AF_INET6 the destination mac
address is acquired from the neighbour table.  For AF_PACKET the
destination mac_address is specified in the netlink configuration.

I think raw destination mac address support with the family AF_PACKET
will prove useful.  There is MPLS-TP which is defined to operate
on machines that do not support internet packets of any flavor.  Further
seem to be corner cases where it can be useful.  At this point
I don't care much either way.

RTA_NEWDST specifies the destination address to forward the packet
with.  MPLS typically changes it's destination address at every hop.
For a swap operation RTA_NEWDST is specified with a length of one label.
For a push operation RTA_NEWDST is specified with two or more labels.
For a pop operation RTA_NEWDST is not specified or equivalently an emtpy
RTAN_NEWDST is specified.

Those new netlink attributes are used to implement handling of rt-netlink
RTM_NEWROUTE, RTM_DELROUTE, and RTM_GETROUTE messages, to maintain the
MPLS label table.

rtm_to_route_config parses a netlink RTM_NEWROUTE or RTM_DELROUTE message,
verify no unhandled attributes or unhandled values are present and sets
up the data structures for mpls_route_add and mpls_route_del.

I did my best to match up with the existing conventions with the caveats
that MPLS addresses are all destination-specific-addresses, and so
don't properly have a scope.

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/uapi/linux/rtnetlink.h
net/mpls/af_mpls.c

index 5cc5d66bf519f65cb4b29e041c3b06fdbe01c889..bad65550ae3e6729f66f7960bd05eca8c0e44838 100644 (file)
@@ -303,6 +303,8 @@ enum rtattr_type_t {
        RTA_TABLE,
        RTA_MARK,
        RTA_MFC_STATS,
+       RTA_VIA,
+       RTA_NEWDST,
        __RTA_MAX
 };
 
@@ -344,6 +346,12 @@ struct rtnexthop {
 #define RTNH_SPACE(len)        RTNH_ALIGN(RTNH_LENGTH(len))
 #define RTNH_DATA(rtnh)   ((struct rtattr*)(((char*)(rtnh)) + RTNH_LENGTH(0)))
 
+/* RTA_VIA */
+struct rtvia {
+       __kernel_sa_family_t    rtvia_family;
+       __u8                    rtvia_addr[0];
+};
+
 /* RTM_CACHEINFO */
 
 struct rta_cacheinfo {
index 2d6612a10e305671e68ed372cd4f19ab6640b3a6..b4d7cec398d236f14b6f3329e4530a7796aaa7ec 100644 (file)
@@ -212,6 +212,11 @@ static struct packet_type mpls_packet_type __read_mostly = {
        .func = mpls_forward,
 };
 
+const struct nla_policy rtm_mpls_policy[RTA_MAX+1] = {
+       [RTA_DST]               = { .type = NLA_U32 },
+       [RTA_OIF]               = { .type = NLA_U32 },
+};
+
 struct mpls_route_config {
        u32             rc_protocol;
        u32             rc_ifindex;
@@ -410,6 +415,22 @@ static struct notifier_block mpls_dev_notifier = {
        .notifier_call = mpls_dev_notify,
 };
 
+static int nla_put_via(struct sk_buff *skb,
+                      u16 family, const void *addr, int alen)
+{
+       struct nlattr *nla;
+       struct rtvia *via;
+
+       nla = nla_reserve(skb, RTA_VIA, alen + 2);
+       if (!nla)
+               return -EMSGSIZE;
+
+       via = nla_data(nla);
+       via->rtvia_family = family;
+       memcpy(via->rtvia_addr, addr, alen);
+       return 0;
+}
+
 int nla_put_labels(struct sk_buff *skb, int attrtype,
                   u8 labels, const u32 label[])
 {
@@ -467,6 +488,210 @@ int nla_get_labels(const struct nlattr *nla,
        return 0;
 }
 
+static int rtm_to_route_config(struct sk_buff *skb,  struct nlmsghdr *nlh,
+                              struct mpls_route_config *cfg)
+{
+       struct rtmsg *rtm;
+       struct nlattr *tb[RTA_MAX+1];
+       int index;
+       int err;
+
+       err = nlmsg_parse(nlh, sizeof(*rtm), tb, RTA_MAX, rtm_mpls_policy);
+       if (err < 0)
+               goto errout;
+
+       err = -EINVAL;
+       rtm = nlmsg_data(nlh);
+       memset(cfg, 0, sizeof(*cfg));
+
+       if (rtm->rtm_family != AF_MPLS)
+               goto errout;
+       if (rtm->rtm_dst_len != 20)
+               goto errout;
+       if (rtm->rtm_src_len != 0)
+               goto errout;
+       if (rtm->rtm_tos != 0)
+               goto errout;
+       if (rtm->rtm_table != RT_TABLE_MAIN)
+               goto errout;
+       /* Any value is acceptable for rtm_protocol */
+
+       /* As mpls uses destination specific addresses
+        * (or source specific address in the case of multicast)
+        * all addresses have universal scope.
+        */
+       if (rtm->rtm_scope != RT_SCOPE_UNIVERSE)
+               goto errout;
+       if (rtm->rtm_type != RTN_UNICAST)
+               goto errout;
+       if (rtm->rtm_flags != 0)
+               goto errout;
+
+       cfg->rc_label           = LABEL_NOT_SPECIFIED;
+       cfg->rc_protocol        = rtm->rtm_protocol;
+       cfg->rc_nlflags         = nlh->nlmsg_flags;
+       cfg->rc_nlinfo.portid   = NETLINK_CB(skb).portid;
+       cfg->rc_nlinfo.nlh      = nlh;
+       cfg->rc_nlinfo.nl_net   = sock_net(skb->sk);
+
+       for (index = 0; index <= RTA_MAX; index++) {
+               struct nlattr *nla = tb[index];
+               if (!nla)
+                       continue;
+
+               switch(index) {
+               case RTA_OIF:
+                       cfg->rc_ifindex = nla_get_u32(nla);
+                       break;
+               case RTA_NEWDST:
+                       if (nla_get_labels(nla, MAX_NEW_LABELS,
+                                          &cfg->rc_output_labels,
+                                          cfg->rc_output_label))
+                               goto errout;
+                       break;
+               case RTA_DST:
+               {
+                       u32 label_count;
+                       if (nla_get_labels(nla, 1, &label_count,
+                                          &cfg->rc_label))
+                               goto errout;
+
+                       /* The first 16 labels are reserved, and may not be set */
+                       if (cfg->rc_label < 16)
+                               goto errout;
+
+                       break;
+               }
+               case RTA_VIA:
+               {
+                       struct rtvia *via = nla_data(nla);
+                       cfg->rc_via_family = via->rtvia_family;
+                       cfg->rc_via_alen   = nla_len(nla) - 2;
+                       if (cfg->rc_via_alen > MAX_VIA_ALEN)
+                               goto errout;
+
+                       /* Validate the address family */
+                       switch(cfg->rc_via_family) {
+                       case AF_PACKET:
+                               break;
+                       case AF_INET:
+                               if (cfg->rc_via_alen != 4)
+                                       goto errout;
+                               break;
+                       case AF_INET6:
+                               if (cfg->rc_via_alen != 16)
+                                       goto errout;
+                               break;
+                       default:
+                               /* Unsupported address family */
+                               goto errout;
+                       }
+
+                       memcpy(cfg->rc_via, via->rtvia_addr, cfg->rc_via_alen);
+                       break;
+               }
+               default:
+                       /* Unsupported attribute */
+                       goto errout;
+               }
+       }
+
+       err = 0;
+errout:
+       return err;
+}
+
+static int mpls_rtm_delroute(struct sk_buff *skb, struct nlmsghdr *nlh)
+{
+       struct mpls_route_config cfg;
+       int err;
+
+       err = rtm_to_route_config(skb, nlh, &cfg);
+       if (err < 0)
+               return err;
+
+       return mpls_route_del(&cfg);
+}
+
+
+static int mpls_rtm_newroute(struct sk_buff *skb, struct nlmsghdr *nlh)
+{
+       struct mpls_route_config cfg;
+       int err;
+
+       err = rtm_to_route_config(skb, nlh, &cfg);
+       if (err < 0)
+               return err;
+
+       return mpls_route_add(&cfg);
+}
+
+static int mpls_dump_route(struct sk_buff *skb, u32 portid, u32 seq, int event,
+                          u32 label, struct mpls_route *rt, int flags)
+{
+       struct nlmsghdr *nlh;
+       struct rtmsg *rtm;
+
+       nlh = nlmsg_put(skb, portid, seq, event, sizeof(*rtm), flags);
+       if (nlh == NULL)
+               return -EMSGSIZE;
+
+       rtm = nlmsg_data(nlh);
+       rtm->rtm_family = AF_MPLS;
+       rtm->rtm_dst_len = 20;
+       rtm->rtm_src_len = 0;
+       rtm->rtm_tos = 0;
+       rtm->rtm_table = RT_TABLE_MAIN;
+       rtm->rtm_protocol = rt->rt_protocol;
+       rtm->rtm_scope = RT_SCOPE_UNIVERSE;
+       rtm->rtm_type = RTN_UNICAST;
+       rtm->rtm_flags = 0;
+
+       if (rt->rt_labels &&
+           nla_put_labels(skb, RTA_NEWDST, rt->rt_labels, rt->rt_label))
+               goto nla_put_failure;
+       if (nla_put_via(skb, rt->rt_via_family, rt->rt_via, rt->rt_via_alen))
+               goto nla_put_failure;
+       if (rt->rt_dev && nla_put_u32(skb, RTA_OIF, rt->rt_dev->ifindex))
+               goto nla_put_failure;
+       if (nla_put_labels(skb, RTA_DST, 1, &label))
+               goto nla_put_failure;
+
+       nlmsg_end(skb, nlh);
+       return 0;
+
+nla_put_failure:
+       nlmsg_cancel(skb, nlh);
+       return -EMSGSIZE;
+}
+
+static int mpls_dump_routes(struct sk_buff *skb, struct netlink_callback *cb)
+{
+       struct net *net = sock_net(skb->sk);
+       unsigned int index;
+
+       ASSERT_RTNL();
+
+       index = cb->args[0];
+       if (index < 16)
+               index = 16;
+
+       for (; index < net->mpls.platform_labels; index++) {
+               struct mpls_route *rt;
+               rt = net->mpls.platform_label[index];
+               if (!rt)
+                       continue;
+
+               if (mpls_dump_route(skb, NETLINK_CB(cb->skb).portid,
+                                   cb->nlh->nlmsg_seq, RTM_NEWROUTE,
+                                   index, rt, NLM_F_MULTI) < 0)
+                       break;
+       }
+       cb->args[0] = index;
+
+       return skb->len;
+}
+
 static int resize_platform_label_table(struct net *net, size_t limit)
 {
        size_t size = sizeof(struct mpls_route *) * limit;
@@ -662,6 +887,9 @@ static int __init mpls_init(void)
 
        dev_add_pack(&mpls_packet_type);
 
+       rtnl_register(PF_MPLS, RTM_NEWROUTE, mpls_rtm_newroute, NULL, NULL);
+       rtnl_register(PF_MPLS, RTM_DELROUTE, mpls_rtm_delroute, NULL, NULL);
+       rtnl_register(PF_MPLS, RTM_GETROUTE, NULL, mpls_dump_routes, NULL);
        err = 0;
 out:
        return err;
@@ -674,6 +902,7 @@ module_init(mpls_init);
 
 static void __exit mpls_exit(void)
 {
+       rtnl_unregister_all(PF_MPLS);
        dev_remove_pack(&mpls_packet_type);
        unregister_netdevice_notifier(&mpls_dev_notifier);
        unregister_pernet_subsys(&mpls_net_ops);