Handle UDP packet loss when fragmentation of large packets is broken.

CHANGELOG

@@ -109,6 +109,12 @@ version 2.73
 	    by quiet-dhcp6. Thanks to J. Pablo Abonia for 
 	    spotting the problem.
 
+	    Try and handle net connections with broken fragmentation 
+	    that lose large UDP packets. If a server times out, 
+            reduce the maximum UDP packet size field in the EDNS0
+	    header to 1280 bytes. If it then answers, make that
+	    change permanent.
+
 	
 version 2.72
             Add ra-advrouter mode, for RFC-3775 mobile IPv6 support.

src/config.h

@@ -19,6 +19,7 @@
 #define CHILD_LIFETIME 150 /* secs 'till terminated (RFC1035 suggests > 120s) */
 #define TCP_MAX_QUERIES 100 /* Maximum number of queries per incoming TCP connection */
 #define EDNS_PKTSZ 4096 /* default max EDNS.0 UDP packet from RFC5625 */
+#define SAFE_PKTSZ 1280 /* "go anywhere" UDP packet size */
 #define KEYBLOCK_LEN 40 /* choose to mininise fragmentation when storing DNSSEC keys */
 #define DNSSEC_WORK 50 /* Max number of queries to validate one question */
 #define TIMEOUT 10 /* drop UDP queries after TIMEOUT seconds */

src/dnsmasq.h

@@ -504,7 +504,7 @@ struct server {
   char interface[IF_NAMESIZE+1];
   struct serverfd *sfd; 
   char *domain; /* set if this server only handles a domain. */ 
-  int flags, tcpfd;
+  int flags, tcpfd, edns_pktsz;
   unsigned int queries, failed_queries;
 #ifdef HAVE_LOOP
   u32 uid;
@@ -594,6 +594,7 @@ struct hostsfile {
 #define FREC_DO_QUESTION       64
 #define FREC_ADDED_PHEADER    128
 #define FREC_CHECK_NOSIGN     256
+#define FREC_TEST_PKTSZ       512
 
 #ifdef HAVE_DNSSEC
 #define HASH_SIZE 20 /* SHA-1 digest size */
@@ -1148,7 +1149,7 @@ int in_zone(struct auth_zone *zone, char *name, char **cut);
 #endif
 
 /* dnssec.c */
-size_t dnssec_generate_query(struct dns_header *header, char *end, char *name, int class, int type, union mysockaddr *addr);
+size_t dnssec_generate_query(struct dns_header *header, char *end, char *name, int class, int type, union mysockaddr *addr, int edns_pktsz);
 int dnssec_validate_by_ds(time_t now, struct dns_header *header, size_t n, char *name, char *keyname, int class);
 int dnssec_validate_ds(time_t now, struct dns_header *header, size_t plen, char *name, char *keyname, int class);
 int dnssec_validate_reply(time_t now, struct dns_header *header, size_t plen, char *name, char *keyname, int *class, int *neganswer, int *nons);

src/dnssec.c

@@ -2162,10 +2162,12 @@ int dnskey_keytag(int alg, int flags, unsigned char *key, int keylen)
     }
 }
 
-size_t dnssec_generate_query(struct dns_header *header, char *end, char *name, int class, int type, union mysockaddr *addr)
+size_t dnssec_generate_query(struct dns_header *header, char *end, char *name, int class, 
+			     int type, union mysockaddr *addr, int edns_pktsz)
 {
   unsigned char *p;
   char *types = querystr("dnssec-query", type);
+  size_t ret;
 
   if (addr->sa.sa_family == AF_INET) 
     log_query(F_NOEXTRA | F_DNSSEC | F_IPV4, name, (struct all_addr *)&addr->in.sin_addr, types);
@@ -2194,7 +2196,12 @@ size_t dnssec_generate_query(struct dns_header *header, char *end, char *name, i
   PUTSHORT(type, p);
   PUTSHORT(class, p);
 
-  return add_do_bit(header, p - (unsigned char *)header, end);
+  ret = add_do_bit(header, p - (unsigned char *)header, end);
+
+  if (find_pseudoheader(header, ret, NULL, &p, NULL))
+    PUTSHORT(edns_pktsz, p);
+
+  return ret;
 }
 
 /* Go through a domain name, find "pointers" and fix them up based on how many bytes

src/forward.c

@@ -253,6 +253,7 @@ static int forward_query(int udpfd, union mysockaddr *udpaddr,
   void *hash = &crc;
 #endif
  unsigned int gotname = extract_request(header, plen, daemon->namebuff, NULL);
+ unsigned char *pheader;
 
  (void)do_bit;
 
@@ -261,19 +262,32 @@ static int forward_query(int udpfd, union mysockaddr *udpaddr,
     forward = NULL;
   else if (forward || (hash && (forward = lookup_frec_by_sender(ntohs(header->id), udpaddr, hash))))
     {
+      /* If we didn't get an answer advertising a maximal packet in EDNS,
+	 fall back to 1280, which should work everywhere on IPv6.
+	 If that generates an answer, it will become the new default
+	 for this server */
+      forward->flags |= FREC_TEST_PKTSZ;
+      
 #ifdef HAVE_DNSSEC
       /* If we've already got an answer to this query, but we're awaiting keys for validation,
 	 there's no point retrying the query, retry the key query instead...... */
       if (forward->blocking_query)
 	{
 	  int fd;
-
+	  
+	  forward->flags &= ~FREC_TEST_PKTSZ;
+	  
 	  while (forward->blocking_query)
 	    forward = forward->blocking_query;
+	   
+	  forward->flags |= FREC_TEST_PKTSZ;
 	  
 	  blockdata_retrieve(forward->stash, forward->stash_len, (void *)header);
 	  plen = forward->stash_len;
 	  
+	  if (find_pseudoheader(header, plen, NULL, &pheader, NULL))
+	    PUTSHORT((forward->flags & FREC_TEST_PKTSZ) ? SAFE_PKTSZ : forward->sentto->edns_pktsz, pheader);
+
 	  if (forward->sentto->addr.sa.sa_family == AF_INET) 
 	    log_query(F_NOEXTRA | F_DNSSEC | F_IPV4, "retry", (struct all_addr *)&forward->sentto->addr.in.sin_addr, "dnssec");
 #ifdef HAVE_IPV6
@@ -417,7 +431,7 @@ static int forward_query(int udpfd, union mysockaddr *udpaddr,
 	  plen = new_plen;
 	}
 #endif
-
+      
       while (1)
 	{ 
 	  /* only send to servers dealing with our domain.
@@ -464,6 +478,9 @@ static int forward_query(int udpfd, union mysockaddr *udpaddr,
 		    }
 #endif
 		}
+
+	      if (find_pseudoheader(header, plen, NULL, &pheader, NULL))
+		PUTSHORT((forward->flags & FREC_TEST_PKTSZ) ? SAFE_PKTSZ : start->edns_pktsz, pheader);
 	      
 	      if (retry_send(sendto(fd, (char *)header, plen, 0,
 				    &start->addr.sa,
@@ -760,7 +777,6 @@ void reply_query(int fd, int family, time_t now)
     }   
    
   server = forward->sentto;
-  
   if ((forward->sentto->flags & SERV_TYPE) == 0)
     {
       if (RCODE(header) == REFUSED)
@@ -781,7 +797,12 @@ void reply_query(int fd, int family, time_t now)
       if (!option_bool(OPT_ALL_SERVERS))
 	daemon->last_server = server;
     }
-
+ 
+  /* We tried resending to this server with a smaller maximum size and got an answer.
+     Make that permanent. */
+  if (server && (forward->flags & FREC_TEST_PKTSZ))
+    server->edns_pktsz = SAFE_PKTSZ;
+  
   /* If the answer is an error, keep the forward record in place in case
      we get a good reply from another server. Kill it when we've
      had replies from all to avoid filling the forwarding table when
@@ -890,7 +911,7 @@ void reply_query(int fd, int family, time_t now)
 		    {
 		      new->flags |= FREC_DNSKEY_QUERY; 
 		      nn = dnssec_generate_query(header, ((char *) header) + daemon->packet_buff_sz,
-						 daemon->keyname, forward->class, T_DNSKEY, &server->addr);
+						 daemon->keyname, forward->class, T_DNSKEY, &server->addr, server->edns_pktsz);
 		    }
 		  else 
 		    {
@@ -899,7 +920,7 @@ void reply_query(int fd, int family, time_t now)
 		      else
 			new->flags |= FREC_DS_QUERY;
 		      nn = dnssec_generate_query(header,((char *) header) + daemon->packet_buff_sz,
-						 daemon->keyname, forward->class, T_DS, &server->addr);
+						 daemon->keyname, forward->class, T_DS, &server->addr, server->edns_pktsz);
 		    }
 		  if ((hash = hash_questions(header, nn, daemon->namebuff)))
 		    memcpy(new->hash, hash, HASH_SIZE);
@@ -1526,7 +1547,7 @@ static int  tcp_check_for_unsigned_zone(time_t now, struct dns_header *header, s
       
       /* Can't find it in the cache, have to send a query */
 
-      m = dnssec_generate_query(header, ((char *) header) + 65536, name_start, class, T_DS, &server->addr);
+      m = dnssec_generate_query(header, ((char *) header) + 65536, name_start, class, T_DS, &server->addr, server->edns_pktsz);
       
       *length = htons(m);
       
@@ -1638,7 +1659,7 @@ static int tcp_key_recurse(time_t now, int status, struct dns_header *header, si
 
     another_tcp_key:
       m = dnssec_generate_query(new_header, ((char *) new_header) + 65536, keyname, class, 
-				new_status == STAT_NEED_KEY ? T_DNSKEY : T_DS, &server->addr);
+				new_status == STAT_NEED_KEY ? T_DNSKEY : T_DS, &server->addr, server->edns_pktsz);
       
       *length = htons(m);
       

src/network.c

@@ -1396,6 +1396,7 @@ void add_update_server(int flags,
       serv->domain = domain_str;
       serv->next = next;
       serv->queries = serv->failed_queries = 0;
+      serv->edns_pktsz = daemon->edns_pktsz;
 #ifdef HAVE_LOOP
       serv->uid = rand32();
 #endif      

src/option.c

@@ -4498,15 +4498,19 @@ void read_opts(int argc, char **argv, char *compile_opts)
     {
       struct server *tmp;
       for (tmp = daemon->servers; tmp; tmp = tmp->next)
-	if (!(tmp->flags & SERV_HAS_SOURCE))
+	{
-	  {
+	  tmp->edns_pktsz = daemon->edns_pktsz;
-	    if (tmp->source_addr.sa.sa_family == AF_INET)
+	 
-	      tmp->source_addr.in.sin_port = htons(daemon->query_port);
+	  if (!(tmp->flags & SERV_HAS_SOURCE))
+	    {
+	      if (tmp->source_addr.sa.sa_family == AF_INET)
+		tmp->source_addr.in.sin_port = htons(daemon->query_port);
 #ifdef HAVE_IPV6
-	    else if (tmp->source_addr.sa.sa_family == AF_INET6)
+	      else if (tmp->source_addr.sa.sa_family == AF_INET6)
-	      tmp->source_addr.in6.sin6_port = htons(daemon->query_port);
+		tmp->source_addr.in6.sin6_port = htons(daemon->query_port);
 #endif 
-	  } 
+	    }
+	} 
     }
   
   if (daemon->if_addrs)

src/rfc1035.c

@@ -552,7 +552,7 @@ static size_t add_pseudoheader(struct dns_header *header, size_t plen, unsigned
 	return plen;
       *p++ = 0; /* empty name */
       PUTSHORT(T_OPT, p);
-      PUTSHORT(daemon->edns_pktsz, p); /* max packet length */
+      PUTSHORT(SAFE_PKTSZ, p); /* max packet length, this will be overwritten */
       PUTSHORT(0, p);    /* extended RCODE and version */
       PUTSHORT(set_do ? 0x8000 : 0, p); /* DO flag */
       lenp = p;
@@ -1537,7 +1537,6 @@ size_t answer_request(struct dns_header *header, char *limit, size_t qlen,
   unsigned short flag;
   int q, ans, anscount = 0, addncount = 0;
   int dryrun = 0, sec_reqd = 0, have_pseudoheader = 0;
-  int is_sign;
   struct crec *crecp;
   int nxdomain = 0, auth = 1, trunc = 0, sec_data = 1;
   struct mx_srv_record *rec;
@@ -1557,28 +1556,19 @@ size_t answer_request(struct dns_header *header, char *limit, size_t qlen,
      forward rather than answering from the cache, which doesn't include
      security information, unless we're in DNSSEC validation mode. */
 
-  if (find_pseudoheader(header, qlen, NULL, &pheader, &is_sign))
+  if (find_pseudoheader(header, qlen, NULL, &pheader, NULL))
     { 
-      unsigned short udpsz, flags;
+      unsigned short flags;
-      unsigned char *psave = pheader;
+      
-
       have_pseudoheader = 1;
 
-      GETSHORT(udpsz, pheader);
+      pheader += 4; /* udp size, ext_rcode */
-      pheader += 2; /* ext_rcode */
       GETSHORT(flags, pheader);
       
       if ((sec_reqd = flags & 0x8000))
 	*do_bit = 1;/* do bit */ 
+
       *ad_reqd = 1;
-
-      /* If our client is advertising a larger UDP packet size
-	 than we allow, trim it so that we don't get an overlarge
-	 response from upstream */
-
-      if (!is_sign && (udpsz > daemon->edns_pktsz))
-	PUTSHORT(daemon->edns_pktsz, psave); 
-
       dryrun = 1;
     }