From ef4a50be10c6dd0abffd957132cd146029c3d79d Mon Sep 17 00:00:00 2001
From: Ondrej Zajicek <santiago@crfreenet.org>
Date: Mon, 24 Jun 2013 16:37:30 +0200
Subject: [PATCH] Better packet priority and traffic class handling.

Implements support for IPv6 traffic class, sets higher priority for OSPF
and RIP outgoing packets by default and allows to configure ToS/DS/TClass
IP header field and the local priority of outgoing packets.
---
 doc/bird.sgml        | 48 +++++++++++++++++++++++++++++++++++++-------
 lib/ipv6.h           |  7 +------
 lib/socket.h         |  4 +++-
 nest/config.Y        |  8 ++++++--
 proto/ospf/config.Y  |  6 +++++-
 proto/ospf/iface.c   |  8 ++++++--
 proto/ospf/ospf.h    |  2 ++
 proto/rip/config.Y   |  6 +++++-
 proto/rip/rip.c      |  7 +++++--
 proto/rip/rip.h      |  2 ++
 sysdep/bsd/sysio.h   |  9 +++++++++
 sysdep/linux/sysio.h | 19 ++++++++++++++++++
 sysdep/unix/io.c     | 11 ++++++++--
 13 files changed, 113 insertions(+), 24 deletions(-)

diff --git a/doc/bird.sgml b/doc/bird.sgml
index 0681bd53..7277b2b9 100644
--- a/doc/bird.sgml
+++ b/doc/bird.sgml
@@ -567,6 +567,22 @@ to zero to disable it. An empty <cf><m/switch/</cf> is equivalent to <cf/on/
 	<cf>interface "eth*" 192.168.1.0/24;</cf> - start the protocol on all
 	ethernet interfaces that have address from 192.168.1.0/24.
 
+	<tag><label id="dsc-prio">tx class|dscp <m/num/</tag>
+        This option specifies the value of ToS/DS/Class field in IP
+        headers of the outgoing protocol packets. This may affect how the
+        protocol packets are processed by the network relative to the
+        other network traffic. With <cf/class/ keyword, the value
+        (0-255) is used for the whole ToS/Class octet (but two bits
+        reserved for ECN are ignored). With <cf/dscp/ keyword, the
+        value (0-63) is used just for the DS field in the
+        octet. Default value is 0xc0 (DSCP 0x30 - CS6).
+
+	<tag>tx priority <m/num/</tag>
+        This option specifies the local packet priority. This may
+        affect how the protocol packets are processed in the local TX
+        queues. This option is Linux specific. Default value is 7
+        (highest priority, privileged traffic).
+
 	<tag><label id="dsc-pass">password "<m/password/" [ { id <m/num/; generate from <m/time/; generate to <m/time/; accept from <m/time/; accept to <m/time/; } ]</tag>
 	Specifies a password that can be used by the protocol. Password option can
 	be used more times to specify more passwords. If more passwords are
@@ -2220,6 +2236,11 @@ protocol ospf &lt;name&gt; {
 	 prefix) is propagated. It is possible that some hardware
 	 drivers or platforms do not implement this feature. Default value is no.
 
+	<tag>tx class|dscp|priority <m/num/</tag>
+         These options specify the ToS/DiffServ/Traffic class/Priority
+         of the outgoing OSPF packets. See <ref id="dsc-prio" name="tx
+         class"> common option for detailed description.
+
 	<tag>ecmp weight <M>num</M></tag>
 	 When ECMP (multipath) routes are allowed, this value specifies
 	 a relative weight used for nexthops going through the iface.
@@ -2748,13 +2769,26 @@ makes it pretty much obsolete. (It is still usable on very small networks.)
 	  neighbors, that is not configurable. Default: never.
 </descrip>
 
-<p>There are two options that can be specified per-interface. First is <cf>metric</cf>, with
-default one.  Second is <cf>mode multicast|broadcast|quiet|nolisten|version1</cf>, it selects mode for
-rip to work in. If nothing is specified, rip runs in multicast mode. <cf>version1</cf> is
-currently equivalent to <cf>broadcast</cf>, and it makes RIP talk to a broadcast address even
-through multicast mode is possible. <cf>quiet</cf> option means that RIP will not transmit
-any periodic messages to this interface and <cf>nolisten</cf> means that RIP will send to this
-interface but not listen to it.
+<p>There are some options that can be specified per-interface:
+
+<descrip>
+	<tag>metric <m/num/</tag>
+	  This option specifies the metric of the interface. Valid
+
+	<tag>mode multicast|broadcast|quiet|nolisten|version1</tag>
+	  This option selects the mode for RIP to work in. If nothing is
+	  specified, RIP runs in multicast mode. <cf/version1/ is
+	  currently equivalent to <cf/broadcast/, and it makes RIP talk
+	  to a broadcast address even through multicast mode is
+	  possible. <cf/quiet/ option means that RIP will not transmit
+	  any periodic messages to this interface and <cf/nolisten/
+	  means that RIP will send to this interface butnot listen to it.
+
+	<tag>tx class|dscp|priority <m/num/</tag>
+          These options specify the ToS/DiffServ/Traffic class/Priority
+          of the outgoing RIP packets. See <ref id="dsc-prio" name="tx
+          class"> common option for detailed description.
+</descrip>
 
 <p>The following options generally override behavior specified in RFC. If you use any of these
 options, BIRD will no longer be RFC-compliant, which means it will not be able to talk to anything
diff --git a/lib/ipv6.h b/lib/ipv6.h
index 6f8e7b3c..2247d3fd 100644
--- a/lib/ipv6.h
+++ b/lib/ipv6.h
@@ -128,11 +128,6 @@ static inline byte * ipv6_put_addr(byte *buf, ip_addr a)
   return buf+16;
 }
 
-/*
- *  RFC 1883 defines packet precendece, but RFC 2460 replaces it
- *  by generic Traffic Class ID with no defined semantics. Better
- *  not use it yet.
- */
-#define IP_PREC_INTERNET_CONTROL -1
+#define IP_PREC_INTERNET_CONTROL 0xc0
 
 #endif
diff --git a/lib/socket.h b/lib/socket.h
index 0ee43b52..fbddfb4c 100644
--- a/lib/socket.h
+++ b/lib/socket.h
@@ -20,7 +20,8 @@ typedef struct birdsock {
   void *data;				/* User data */
   ip_addr saddr, daddr;			/* IPA_NONE = unspecified */
   unsigned sport, dport;		/* 0 = unspecified (for IP: protocol type) */
-  int tos;				/* TOS and priority, -1 = default */
+  int tos;				/* TOS / traffic class, -1 = default */
+  int priority;				/* Local socket priority, -1 = default */
   int ttl;				/* Time To Live, -1 = default */
   u32 flags;
   struct iface *iface;			/* Interface; specify this for broad/multicast sockets */
@@ -81,6 +82,7 @@ sk_send_buffer_empty(sock *sk)
 	return sk->tbuf == sk->tpos;
 }
 
+extern int sk_priority_control;	/* Suggested priority for control traffic, should be sysdep define */
 
 /* Socket flags */
 
diff --git a/nest/config.Y b/nest/config.Y
index 183059e8..b85a5733 100644
--- a/nest/config.Y
+++ b/nest/config.Y
@@ -48,7 +48,7 @@ CF_KEYWORDS(RECEIVE, LIMIT, ACTION, WARN, BLOCK, RESTART, DISABLE, KEEP, FILTERE
 CF_KEYWORDS(PASSWORD, FROM, PASSIVE, TO, ID, EVENTS, PACKETS, PROTOCOLS, INTERFACES)
 CF_KEYWORDS(PRIMARY, STATS, COUNT, FOR, COMMANDS, PREEXPORT, GENERATE, ROA, MAX, FLUSH)
 CF_KEYWORDS(LISTEN, BGP, V6ONLY, DUAL, ADDRESS, PORT, PASSWORDS, DESCRIPTION, SORTED)
-CF_KEYWORDS(RELOAD, IN, OUT, MRTDUMP, MESSAGES, RESTRICT, MEMORY, IGP_METRIC)
+CF_KEYWORDS(RELOAD, IN, OUT, MRTDUMP, MESSAGES, RESTRICT, MEMORY, IGP_METRIC, CLASS, DSCP)
 
 CF_ENUM(T_ENUM_RTS, RTS_, DUMMY, STATIC, INHERIT, DEVICE, STATIC_DEVICE, REDIRECT,
 	RIP, OSPF, OSPF_IA, OSPF_EXT1, OSPF_EXT2, BGP, PIPE)
@@ -65,7 +65,7 @@ CF_ENUM(T_ENUM_ROA, ROA_, UNKNOWN, VALID, INVALID)
 %type <ro> roa_args
 %type <rot> roa_table_arg
 %type <sd> sym_args
-%type <i> proto_start echo_mask echo_size debug_mask debug_list debug_flag mrtdump_mask mrtdump_list mrtdump_flag export_or_preexport roa_mode limit_action tab_sorted
+%type <i> proto_start echo_mask echo_size debug_mask debug_list debug_flag mrtdump_mask mrtdump_list mrtdump_flag export_or_preexport roa_mode limit_action tab_sorted tos
 %type <ps> proto_patt proto_patt2
 %type <g> limit_spec
 
@@ -277,6 +277,10 @@ iface_patt:
    iface_patt_init iface_patt_list
  ;
 
+tos:
+   CLASS expr { $$ = $2 & 0xfc;        if (($2 < 0) || ($2 > 255)) cf_error("TX class must be in range 0-255"); }
+ | DSCP expr  { $$ = ($2 & 0x3f) << 2; if (($2 < 0) || ($2 > 63))  cf_error("TX DSCP must be in range 0-63"); }
+ ;
 
 /* Direct device route protocol */
 
diff --git a/proto/ospf/config.Y b/proto/ospf/config.Y
index ba050d85..d9379a7c 100644
--- a/proto/ospf/config.Y
+++ b/proto/ospf/config.Y
@@ -131,7 +131,7 @@ CF_KEYWORDS(NONE, SIMPLE, AUTHENTICATION, STRICT, CRYPTOGRAPHIC)
 CF_KEYWORDS(ELIGIBLE, POLL, NETWORKS, HIDDEN, VIRTUAL, CHECK, LINK)
 CF_KEYWORDS(RX, BUFFER, LARGE, NORMAL, STUBNET, HIDDEN, SUMMARY, TAG, EXTERNAL)
 CF_KEYWORDS(WAIT, DELAY, LSADB, ECMP, LIMIT, WEIGHT, NSSA, TRANSLATOR, STABILITY)
-CF_KEYWORDS(GLOBAL, LSID, ROUTER, SELF, INSTANCE, REAL, NETMASK)
+CF_KEYWORDS(GLOBAL, LSID, ROUTER, SELF, INSTANCE, REAL, NETMASK, TX, PRIORITY)
 
 %type <t> opttext
 %type <ld> lsadb_args
@@ -305,6 +305,8 @@ ospf_iface_item:
  | RX BUFFER LARGE { OSPF_PATT->rxbuf = OSPF_RXBUF_LARGE ; } 
  | RX BUFFER NORMAL { OSPF_PATT->rxbuf = OSPF_RXBUF_NORMAL ; } 
  | RX BUFFER expr { OSPF_PATT->rxbuf = $3 ; if (($3 < OSPF_RXBUF_MINSIZE) || ($3 > OSPF_MAX_PKT_SIZE)) cf_error("Buffer size must be in range 256-65535"); } 
+ | TX tos { OSPF_PATT->tx_tos = $2; }
+ | TX PRIORITY expr { OSPF_PATT->tx_priority = $3; }
  | password_list
  ;
 
@@ -367,6 +369,8 @@ ospf_iface_start:
   init_list(&OSPF_PATT->nbma_list);
   OSPF_PATT->autype = OSPF_AUTH_NONE;
   OSPF_PATT->ptp_netmask = 2; /* not specified */
+  OSPF_PATT->tx_tos = IP_PREC_INTERNET_CONTROL;
+  OSPF_PATT->tx_priority = sk_priority_control;
   reset_passwords();
  }
 ;
diff --git a/proto/ospf/iface.c b/proto/ospf/iface.c
index 3da8f56c..bc3b1ef6 100644
--- a/proto/ospf/iface.c
+++ b/proto/ospf/iface.c
@@ -77,7 +77,8 @@ ospf_sk_open(struct ospf_iface *ifa)
   sk->dport = OSPF_PROTO;
   sk->saddr = IPA_NONE;
 
-  sk->tos = IP_PREC_INTERNET_CONTROL;
+  sk->tos = ifa->cf->tx_tos;
+  sk->priority = ifa->cf->tx_priority;
   sk->rx_hook = ospf_rx_hook;
   sk->tx_hook = ospf_tx_hook;
   sk->err_hook = ospf_err_hook;
@@ -659,7 +660,10 @@ ospf_iface_reconfigure(struct ospf_iface *ifa, struct ospf_iface_patt *new)
   if (ifa->stub != new_stub)
     return 0;
 
-  if (new->real_bcast != ifa->cf->real_bcast)
+  /* Change of these options would require to reset the iface socket */
+  if ((new->real_bcast != ifa->cf->real_bcast) ||
+      (new->tx_tos != ifa->cf->tx_tos) ||
+      (new->tx_priority != ifa->cf->tx_priority))
     return 0;
 
   ifa->cf = new;
diff --git a/proto/ospf/ospf.h b/proto/ospf/ospf.h
index 7608225f..56ebcd31 100644
--- a/proto/ospf/ospf.h
+++ b/proto/ospf/ospf.h
@@ -800,6 +800,8 @@ struct ospf_iface_patt
   u32 priority;
   u32 voa;
   u32 vid;
+  int tx_tos;
+  int tx_priority;
   u16 rxbuf;
 #define OSPF_RXBUF_NORMAL 0
 #define OSPF_RXBUF_LARGE 1
diff --git a/proto/rip/config.Y b/proto/rip/config.Y
index cd4f30e7..ec82aa3d 100644
--- a/proto/rip/config.Y
+++ b/proto/rip/config.Y
@@ -27,7 +27,7 @@ CF_DECLS
 CF_KEYWORDS(RIP, INFINITY, METRIC, PORT, PERIOD, GARBAGE, TIMEOUT,
 	    MODE, BROADCAST, MULTICAST, QUIET, NOLISTEN, VERSION1, 
 	    AUTHENTICATION, NONE, PLAINTEXT, MD5,
-	    HONOR, NEVER, NEIGHBOR, ALWAYS,
+	    HONOR, NEVER, NEIGHBOR, ALWAYS, TX, PRIORITY,
 	    RIP_METRIC, RIP_TAG)
 
 %type <i> rip_mode rip_auth
@@ -76,6 +76,8 @@ rip_mode:
 rip_iface_item:
  | METRIC expr { RIP_IPATT->metric = $2; }
  | MODE rip_mode { RIP_IPATT->mode |= $2; }
+ | TX tos { RIP_IPATT->tx_tos = $2; }
+ | TX PRIORITY expr { RIP_IPATT->tx_priority = $3; }
  ;
 
 rip_iface_opts: 
@@ -94,6 +96,8 @@ rip_iface_init:
      add_tail(&RIP_CFG->iface_list, NODE this_ipatt);
      init_list(&this_ipatt->ipn_list);
      RIP_IPATT->metric = 1;
+     RIP_IPATT->tx_tos = IP_PREC_INTERNET_CONTROL;
+     RIP_IPATT->tx_priority = sk_priority_control;
    }
  ;
 
diff --git a/proto/rip/rip.c b/proto/rip/rip.c
index 341df7eb..c09eae79 100644
--- a/proto/rip/rip.c
+++ b/proto/rip/rip.c
@@ -707,7 +707,8 @@ new_iface(struct proto *p, struct iface *new, unsigned long flags, struct iface_
   if (new)
     {
       rif->sock->ttl = 1;
-      rif->sock->tos = IP_PREC_INTERNET_CONTROL;
+      rif->sock->tos = PATT->tx_tos;
+      rif->sock->priority = PATT->tx_priority;
       rif->sock->flags = SKF_LADDR_RX;
     }
 
@@ -1007,7 +1008,9 @@ static int
 rip_pat_compare(struct rip_patt *a, struct rip_patt *b)
 {
   return ((a->metric == b->metric) &&
-	  (a->mode == b->mode));
+	  (a->mode == b->mode) &&
+	  (a->tx_tos == b->tx_tos) &&
+	  (a->tx_priority == b->tx_priority));
 }
 
 static int
diff --git a/proto/rip/rip.h b/proto/rip/rip.h
index e0816d0e..2cce8c81 100644
--- a/proto/rip/rip.h
+++ b/proto/rip/rip.h
@@ -128,6 +128,8 @@ struct rip_patt {
 #define IM_QUIET 4
 #define IM_NOLISTEN 8
 #define IM_VERSION1 16
+  int tx_tos;
+  int tx_priority;
 };
 
 struct rip_proto_config {
diff --git a/sysdep/bsd/sysio.h b/sysdep/bsd/sysio.h
index 4f91def5..085f16fa 100644
--- a/sysdep/bsd/sysio.h
+++ b/sysdep/bsd/sysio.h
@@ -284,3 +284,12 @@ sk_set_min_ttl6(sock *s, int ttl)
 
 #endif
 
+
+int sk_priority_control = -1;
+
+static int
+sk_set_priority(sock *s, int prio UNUSED)
+{
+  log(L_WARN "Socket priority not supported");
+  return -1;
+}
diff --git a/sysdep/linux/sysio.h b/sysdep/linux/sysio.h
index 90b3ebd9..41287e71 100644
--- a/sysdep/linux/sysio.h
+++ b/sysdep/linux/sysio.h
@@ -310,3 +310,22 @@ sk_set_min_ttl6(sock *s, int ttl)
 }
 
 #endif
+
+
+#ifndef IPV6_TCLASS
+#define IPV6_TCLASS 67
+#endif
+
+int sk_priority_control = 7;
+
+static int
+sk_set_priority(sock *s, int prio)
+{
+  if (setsockopt(s->fd, SOL_SOCKET, SO_PRIORITY, &prio, sizeof(prio)) < 0)
+  {
+    log(L_WARN "sk_set_priority: setsockopt: %m");
+    return -1;
+  }
+
+  return 0;
+}
diff --git a/sysdep/unix/io.c b/sysdep/unix/io.c
index 80914afe..434a05be 100644
--- a/sysdep/unix/io.c
+++ b/sysdep/unix/io.c
@@ -598,7 +598,7 @@ sock_new(pool *p)
   sock *s = ralloc(p, &sk_class);
   s->pool = p;
   // s->saddr = s->daddr = IPA_NONE;
-  s->tos = s->ttl = -1;
+  s->tos = s->priority = s->ttl = -1;
   s->fd = -1;
   return s;
 }
@@ -783,11 +783,18 @@ sk_setup(sock *s)
     ERR("fcntl(O_NONBLOCK)");
   if (s->type == SK_UNIX)
     return NULL;
-#ifndef IPV6
+
+#ifdef IPV6
+  if ((s->tos >= 0) && setsockopt(fd, SOL_IPV6, IPV6_TCLASS, &s->tos, sizeof(s->tos)) < 0)
+    WARN("IPV6_TCLASS");
+#else
   if ((s->tos >= 0) && setsockopt(fd, SOL_IP, IP_TOS, &s->tos, sizeof(s->tos)) < 0)
     WARN("IP_TOS");
 #endif
 
+  if (s->priority >= 0)
+    sk_set_priority(s, s->priority);
+
 #ifdef IPV6
   int v = 1;
   if ((s->flags & SKF_V6ONLY) && setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &v, sizeof(v)) < 0)