Merge branch 'master' into int-new

1 files changed, 205 insertions, 76 deletions

diff --git a/sysdep/linux/netlink.c b/sysdep/linux/netlink.c
index 6d99b0e8..e56863cc 100644
--- a/sysdep/linux/netlink.c
+++ b/sysdep/linux/netlink.c
@@ -25,6 +25,7 @@
 #include "lib/krt.h"
 #include "lib/socket.h"
 #include "lib/string.h"
+#include "lib/hash.h"
 #include "conf/conf.h"
 
 #include <asm/types.h>
@@ -32,6 +33,7 @@
 #include <linux/netlink.h>
 #include <linux/rtnetlink.h>
 
+
 #ifndef MSG_TRUNC			/* Hack: Several versions of glibc miss this one :( */
 #define MSG_TRUNC 0x20
 #endif
@@ -40,6 +42,11 @@
 #define IFF_LOWER_UP 0x10000
 #endif
 
+#ifndef RTA_TABLE
+#define RTA_TABLE  15
+#endif
+
+
 /*
  *	Synchronous Netlink interface
  */
@@ -219,27 +226,101 @@ nl_checkin(struct nlmsghdr *h, int lsize)
   return NLMSG_DATA(h);
 }
 
+struct nl_want_attrs {
+  u8 defined:1;
+  u8 checksize:1;
+  u8 size;
+};
+
+
+#define BIRD_IFLA_MAX (IFLA_WIRELESS+1)
+
+static struct nl_want_attrs ifla_attr_want[BIRD_IFLA_MAX] = {
+  [IFLA_IFNAME]	  = { 1, 0, 0 },
+  [IFLA_MTU]	  = { 1, 1, sizeof(u32) },
+  [IFLA_WIRELESS] = { 1, 0, 0 },
+};
+
+
+#define BIRD_IFA_MAX  (IFA_ANYCAST+1)
+
+#ifndef IPV6
+static struct nl_want_attrs ifa_attr_want4[BIRD_IFA_MAX] = {
+  [IFA_ADDRESS]	  = { 1, 1, sizeof(ip4_addr) },
+  [IFA_LOCAL]	  = { 1, 1, sizeof(ip4_addr) },
+  [IFA_BROADCAST] = { 1, 1, sizeof(ip4_addr) },
+};
+#else
+static struct nl_want_attrs ifa_attr_want6[BIRD_IFA_MAX] = {
+  [IFA_ADDRESS]	  = { 1, 1, sizeof(ip6_addr) },
+  [IFA_LOCAL]	  = { 1, 1, sizeof(ip6_addr) },
+};
+#endif
+
+
+#define BIRD_RTA_MAX  (RTA_TABLE+1)
+
+static struct nl_want_attrs mpnh_attr_want4[BIRD_RTA_MAX] = {
+  [RTA_GATEWAY]	  = { 1, 1, sizeof(ip4_addr) },
+};
+
+#ifndef IPV6
+static struct nl_want_attrs rtm_attr_want4[BIRD_RTA_MAX] = {
+  [RTA_DST]	  = { 1, 1, sizeof(ip4_addr) },
+  [RTA_OIF]	  = { 1, 1, sizeof(u32) },
+  [RTA_GATEWAY]	  = { 1, 1, sizeof(ip4_addr) },
+  [RTA_PRIORITY]  = { 1, 1, sizeof(u32) },
+  [RTA_PREFSRC]	  = { 1, 1, sizeof(ip4_addr) },
+  [RTA_METRICS]	  = { 1, 0, 0 },
+  [RTA_MULTIPATH] = { 1, 0, 0 },
+  [RTA_FLOW]	  = { 1, 1, sizeof(u32) },
+  [RTA_TABLE]	  = { 1, 1, sizeof(u32) },
+};
+#else
+static struct nl_want_attrs rtm_attr_want6[BIRD_RTA_MAX] = {
+  [RTA_DST]	  = { 1, 1, sizeof(ip6_addr) },
+  [RTA_IIF]	  = { 1, 1, sizeof(u32) },
+  [RTA_OIF]	  = { 1, 1, sizeof(u32) },
+  [RTA_GATEWAY]	  = { 1, 1, sizeof(ip6_addr) },
+  [RTA_PRIORITY]  = { 1, 1, sizeof(u32) },
+  [RTA_PREFSRC]	  = { 1, 1, sizeof(ip6_addr) },
+  [RTA_METRICS]	  = { 1, 0, 0 },
+  [RTA_FLOW]	  = { 1, 1, sizeof(u32) },
+  [RTA_TABLE]	  = { 1, 1, sizeof(u32) },
+};
+#endif
+
+
 static int
-nl_parse_attrs(struct rtattr *a, struct rtattr **k, int ksize)
+nl_parse_attrs(struct rtattr *a, struct nl_want_attrs *want, struct rtattr **k, int ksize)
 {
   int max = ksize / sizeof(struct rtattr *);
   bzero(k, ksize);
-  while (RTA_OK(a, nl_attr_len))
+
+  for ( ; RTA_OK(a, nl_attr_len); a = RTA_NEXT(a, nl_attr_len))
     {
-      if (a->rta_type < max)
-	k[a->rta_type] = a;
-      a = RTA_NEXT(a, nl_attr_len);
+      if ((a->rta_type >= max) || !want[a->rta_type].defined)
+	continue;
+
+      if (want[a->rta_type].checksize && (RTA_PAYLOAD(a) != want[a->rta_type].size))
+	{
+	  log(L_ERR "nl_parse_attrs: Malformed message received");
+	  return 0;
+	}
+
+      k[a->rta_type] = a;
     }
+
   if (nl_attr_len)
     {
       log(L_ERR "nl_parse_attrs: remnant of size %d", nl_attr_len);
       return 0;
     }
-  else
-    return 1;
+
+  return 1;
 }
 
-static inline ip4_addr rta_get_u32(struct rtattr *a)
+static inline u32 rta_get_u32(struct rtattr *a)
 { return *(u32 *) RTA_DATA(a); }
 
 static inline ip4_addr rta_get_ip4(struct rtattr *a)
@@ -343,7 +424,7 @@ nl_parse_multipath(struct krt_proto *p, struct rtattr *ra)
   static int nh_buf_size;	/* in number of structures */
   static int nh_buf_used;
 
-  struct rtattr *a[RTA_CACHEINFO+1];
+  struct rtattr *a[BIRD_RTA_MAX];
   struct rtnexthop *nh = RTA_DATA(ra);
   struct mpnh *rv, *first, **last;
   int len = RTA_PAYLOAD(ra);
@@ -374,12 +455,9 @@ nl_parse_multipath(struct krt_proto *p, struct rtattr *ra)
 
       /* Nonexistent RTNH_PAYLOAD ?? */
       nl_attr_len = nh->rtnh_len - RTNH_LENGTH(0);
-      nl_parse_attrs(RTNH_DATA(nh), a, sizeof(a));
+      nl_parse_attrs(RTNH_DATA(nh), mpnh_attr_want4, a, sizeof(a));
       if (a[RTA_GATEWAY])
 	{
-	  if (RTA_PAYLOAD(a[RTA_GATEWAY]) != sizeof(ip_addr))
-	    return NULL;
-
 	  memcpy(&rv->gw, RTA_DATA(a[RTA_GATEWAY]), sizeof(ip_addr));
 	  ipa_ntoh(rv->gw);
 
@@ -448,7 +526,7 @@ static void
 nl_parse_link(struct nlmsghdr *h, int scan)
 {
   struct ifinfomsg *i;
-  struct rtattr *a[IFLA_WIRELESS+1];
+  struct rtattr *a[BIRD_IFLA_MAX];
   int new = h->nlmsg_type == RTM_NEWLINK;
   struct iface f = {};
   struct iface *ifi;
@@ -456,15 +534,23 @@ nl_parse_link(struct nlmsghdr *h, int scan)
   u32 mtu;
   uint fl;
 
-  if (!(i = nl_checkin(h, sizeof(*i))) || !nl_parse_attrs(IFLA_RTA(i), a, sizeof(a)))
+  if (!(i = nl_checkin(h, sizeof(*i))) || !nl_parse_attrs(IFLA_RTA(i), ifla_attr_want, a, sizeof(a)))
     return;
-  if (!a[IFLA_IFNAME] || RTA_PAYLOAD(a[IFLA_IFNAME]) < 2 ||
-      !a[IFLA_MTU] || RTA_PAYLOAD(a[IFLA_MTU]) != 4)
+  if (!a[IFLA_IFNAME] || (RTA_PAYLOAD(a[IFLA_IFNAME]) < 2) || !a[IFLA_MTU])
     {
-      if (scan || !a[IFLA_WIRELESS])
-        log(L_ERR "nl_parse_link: Malformed message received");
+      /*
+       * IFLA_IFNAME and IFLA_MTU are required, in fact, but there may also come
+       * a message with IFLA_WIRELESS set, where (e.g.) no IFLA_IFNAME exists.
+       * We simply ignore all such messages with IFLA_WIRELESS without notice.
+       */
+
+      if (a[IFLA_WIRELESS])
+	return;
+
+      log(L_ERR "KIF: Malformed message received");
       return;
     }
+
   name = RTA_DATA(a[IFLA_IFNAME]);
   mtu = rta_get_u32(a[IFLA_MTU]);
 
@@ -515,26 +601,40 @@ static void
 nl_parse_addr(struct nlmsghdr *h, int scan)
 {
   struct ifaddrmsg *i;
-  struct rtattr *a[IFA_ANYCAST+1];
+  struct rtattr *a[BIRD_IFA_MAX];
   int new = h->nlmsg_type == RTM_NEWADDR;
   struct ifa ifa;
   struct iface *ifi;
   int scope;
 
-  if (!(i = nl_checkin(h, sizeof(*i))) || !nl_parse_attrs(IFA_RTA(i), a, sizeof(a)))
-    return;
-  if (i->ifa_family != BIRD_AF)
+  if (!(i = nl_checkin(h, sizeof(*i))))
     return;
-  if (!a[IFA_ADDRESS] || RTA_PAYLOAD(a[IFA_ADDRESS]) != sizeof(ip_addr)
-#ifdef IPV6
-      || a[IFA_LOCAL] && RTA_PAYLOAD(a[IFA_LOCAL]) != sizeof(ip_addr)
+
+  switch (i->ifa_family)
+    {
+#ifndef IPV6
+      case AF_INET:
+	if (!nl_parse_attrs(IFA_RTA(i), ifa_attr_want4, a, sizeof(a)))
+	  return;
+	if (!a[IFA_LOCAL])
+	  {
+	    log(L_ERR "KIF: Malformed message received (missing IFA_LOCAL)");
+	    return;
+	  }
+	break;
 #else
-      || !a[IFA_LOCAL] || RTA_PAYLOAD(a[IFA_LOCAL]) != sizeof(ip_addr)
-      || (a[IFA_BROADCAST] && RTA_PAYLOAD(a[IFA_BROADCAST]) != sizeof(ip_addr))
+      case AF_INET6:
+	if (!nl_parse_attrs(IFA_RTA(i), ifa_attr_want6, a, sizeof(a)))
+	  return;
+	break;
 #endif
-      )
+      default:
+	return;
+    }
+
+  if (!a[IFA_ADDRESS])
     {
-      log(L_ERR "nl_parse_addr: Malformed message received");
+      log(L_ERR "KIF: Malformed message received (missing IFA_ADDRESS)");
       return;
     }
 
@@ -650,7 +750,23 @@ kif_do_scan(struct kif_proto *p UNUSED)
  *	Routes
  */
 
-static struct krt_proto *nl_table_map[NL_NUM_TABLES];
+static inline u32
+krt_table_id(struct krt_proto *p)
+{
+  return KRT_CF->sys.table_id;
+}
+
+static HASH(struct krt_proto) nl_table_map;
+
+#define RTH_FN(k)	u32_hash(k)
+#define RTH_EQ(k1,k2)	k1 == k2
+#define RTH_KEY(p)	krt_table_id(p)
+#define RTH_NEXT(p)	p->sys.hash_next
+
+#define RTH_REHASH		rth_rehash
+#define RTH_PARAMS		/8, *2, 2, 2, 6, 20
+
+HASH_DEFINE_REHASH_FN(RTH, struct krt_proto)
 
 int
 krt_capable(rte *e)
@@ -708,12 +824,15 @@ nl_send_route(struct krt_proto *p, rte *e, struct ea_list *eattrs, int new)
 
   r.r.rtm_family = BIRD_AF;
   r.r.rtm_dst_len = net->n.pxlen;
-  r.r.rtm_tos = 0;
-  r.r.rtm_table = KRT_CF->sys.table_id;
   r.r.rtm_protocol = RTPROT_BIRD;
   r.r.rtm_scope = RT_SCOPE_UNIVERSE;
   nl_add_attr_ipa(&r.h, sizeof(r), RTA_DST, net->n.prefix);
 
+  if (krt_table_id(p) < 256)
+    r.r.rtm_table = krt_table_id(p);
+  else
+    nl_add_attr_u32(&r.h, sizeof(r), RTA_TABLE, krt_table_id(p));
+
   /* For route delete, we do not specify route attributes */
   if (!new)
     return nl_exchange(&r.h);
@@ -809,31 +928,35 @@ nl_parse_route(struct nlmsghdr *h, int scan)
 {
   struct krt_proto *p;
   struct rtmsg *i;
-  struct rtattr *a[RTA_CACHEINFO+1];
+  struct rtattr *a[BIRD_RTA_MAX];
   int new = h->nlmsg_type == RTM_NEWROUTE;
 
   ip_addr dst = IPA_NONE;
   u32 oif = ~0;
+  u32 table;
   int src;
 
-  if (!(i = nl_checkin(h, sizeof(*i))) || !nl_parse_attrs(RTM_RTA(i), a, sizeof(a)))
-    return;
-  if (i->rtm_family != BIRD_AF)
+  if (!(i = nl_checkin(h, sizeof(*i))))
     return;
-  if ((a[RTA_DST] && RTA_PAYLOAD(a[RTA_DST]) != sizeof(ip_addr)) ||
-#ifdef IPV6
-      (a[RTA_IIF] && RTA_PAYLOAD(a[RTA_IIF]) != 4) ||
-#endif
-      (a[RTA_OIF] && RTA_PAYLOAD(a[RTA_OIF]) != 4) ||
-      (a[RTA_GATEWAY] && RTA_PAYLOAD(a[RTA_GATEWAY]) != sizeof(ip_addr)) ||
-      (a[RTA_PRIORITY] && RTA_PAYLOAD(a[RTA_PRIORITY]) != 4) ||
-      (a[RTA_PREFSRC] && RTA_PAYLOAD(a[RTA_PREFSRC]) != sizeof(ip_addr)) ||
-      (a[RTA_FLOW] && RTA_PAYLOAD(a[RTA_FLOW]) != 4))
+
+  switch (i->rtm_family)
     {
-      log(L_ERR "KRT: Malformed message received");
-      return;
+#ifndef IPV6
+      case AF_INET:
+	if (!nl_parse_attrs(RTM_RTA(i), rtm_attr_want4, a, sizeof(a)))
+	  return;
+	break;
+#else
+      case AF_INET6:
+	if (!nl_parse_attrs(RTM_RTA(i), rtm_attr_want6, a, sizeof(a)))
+	  return;
+	break;
+#endif
+      default:
+	return;
     }
 
+
   if (a[RTA_DST])
     {
       memcpy(&dst, RTA_DATA(a[RTA_DST]), sizeof(dst));
@@ -843,10 +966,15 @@ nl_parse_route(struct nlmsghdr *h, int scan)
   if (a[RTA_OIF])
     oif = rta_get_u32(a[RTA_OIF]);
 
-  p = nl_table_map[i->rtm_table];	/* Do we know this table? */
-  DBG("KRT: Got %I/%d, type=%d, oif=%d, table=%d, prid=%d, proto=%s\n", dst, i->rtm_dst_len, i->rtm_type, oif, i->rtm_table, i->rtm_protocol, p ? p->p.name : "(none)");
+  if (a[RTA_TABLE])
+    table = rta_get_u32(a[RTA_TABLE]);
+  else
+    table = i->rtm_table;
+
+  p = HASH_FIND(nl_table_map, RTH, table); /* Do we know this table? */
+  DBG("KRT: Got %I/%d, type=%d, oif=%d, table=%d, prid=%d, proto=%s\n", dst, i->rtm_dst_len, i->rtm_type, oif, table, i->rtm_protocol, p ? p->p.name : "(none)");
   if (!p)
-    SKIP("unknown table %d\n", i->rtm_table);
+    SKIP("unknown table %d\n", table);
 
 
 #ifdef IPV6
@@ -905,7 +1033,7 @@ nl_parse_route(struct nlmsghdr *h, int scan)
     {
     case RTN_UNICAST:
 
-      if (a[RTA_MULTIPATH])
+      if (a[RTA_MULTIPATH] && (i->rtm_family == AF_INET))
 	{
 	  ra.dest = RTD_MULTIPATH;
 	  ra.nexthops = nl_parse_multipath(p, a[RTA_MULTIPATH]);
@@ -1079,12 +1207,14 @@ nl_async_msg(struct nlmsghdr *h)
     case RTM_NEWLINK:
     case RTM_DELLINK:
       DBG("KRT: Received async link notification (%d)\n", h->nlmsg_type);
-      nl_parse_link(h, 0);
+      if (kif_proto)
+	nl_parse_link(h, 0);
       break;
     case RTM_NEWADDR:
     case RTM_DELADDR:
       DBG("KRT: Received async address notification (%d)\n", h->nlmsg_type);
-      nl_parse_addr(h, 0);
+      if (kif_proto)
+	nl_parse_addr(h, 0);
       break;
     default:
       DBG("KRT: Received unknown async notification (%d)\n", h->nlmsg_type);
@@ -1182,25 +1312,41 @@ nl_open_async(void)
     bug("Netlink: sk_open failed");
 }
 
+
 /*
  *	Interface to the UNIX krt module
  */
 
-static u8 nl_cf_table[(NL_NUM_TABLES+7) / 8];
-
 void
+krt_sys_io_init(void)
+{
+  HASH_INIT(nl_table_map, krt_pool, 6);
+}
+
+int
 krt_sys_start(struct krt_proto *p)
 {
-  nl_table_map[KRT_CF->sys.table_id] = p;
+  struct krt_proto *old = HASH_FIND(nl_table_map, RTH, krt_table_id(p));
+
+  if (old)
+    {
+      log(L_ERR "%s: Kernel table %u already registered by %s",
+	  p->p.name, krt_table_id(p), old->p.name);
+      return 0;
+    }
+
+  HASH_INSERT2(nl_table_map, RTH, krt_pool, p);
 
   nl_open();
   nl_open_async();
+
+  return 1;
 }
 
 void
-krt_sys_shutdown(struct krt_proto *p UNUSED)
+krt_sys_shutdown(struct krt_proto *p)
 {
-  nl_table_map[KRT_CF->sys.table_id] = NULL;
+  HASH_REMOVE2(nl_table_map, RTH, krt_pool, p);
 }
 
 int
@@ -1209,23 +1355,6 @@ krt_sys_reconfigure(struct krt_proto *p UNUSED, struct krt_config *n, struct krt
   return n->sys.table_id == o->sys.table_id;
 }
 
-
-void
-krt_sys_preconfig(struct config *c UNUSED)
-{
-  bzero(&nl_cf_table, sizeof(nl_cf_table));
-}
-
-void
-krt_sys_postconfig(struct krt_config *x)
-{
-  int id = x->sys.table_id;
-
-  if (nl_cf_table[id/8] & (1 << (id%8)))
-    cf_error("Multiple kernel syncers defined for table #%d", id);
-  nl_cf_table[id/8] |= (1 << (id%8));
-}
-
 void
 krt_sys_init_config(struct krt_config *cf)
 {