Optimise TCP send.

In the DNS TCP code, there are a couple of places where
we have a buffer containing a message which we need to
send via TCP. The DNS protocol is that this is sent as
<16-bit length in network order><message>

Making two write calls, one for the length and one
for the message causes the TCP stack to send two packets,
one for each. A single packet containing both is
preferable from a performance POV.

Implement a scatter-gather version of our read_write()
wrapper and use it where necessary to send TCP DNS messages.

src/blockdata.c

@@ -202,11 +202,15 @@ void *blockdata_retrieve(struct blockdata *block, size_t len, void *data)
     {
       if (len > buff_len)
 	{
-	  if (!(new = whine_malloc(len)))
+	  blen = len + 1024;
+	  if (!(new = whine_malloc(blen)))
 	    return NULL;
+
 	  if (buff)
 	    free(buff);
+
 	  buff = new;
+	  buff_len = blen;
 	}
       data = buff;
     }

src/dnsmasq.c

@@ -864,6 +864,9 @@ int main (int argc, char **argv)
     }
 #endif
 
+   /* Don't start logging malloc before logging is set up. */
+  daemon->log_malloc = option_bool(OPT_LOG_MALLOC);
+  
   if (daemon->port == 0)
     my_syslog(LOG_INFO, _("started, version %s DNS disabled"), VERSION);
   else 
@@ -1977,11 +1980,11 @@ static void do_tcp_connection(struct listener *listener, time_t now, int slot)
   pid_t p;
   union mysockaddr tcp_addr;
   socklen_t tcp_len = sizeof(union mysockaddr);
-  unsigned char *buff;
   struct server *s; 
   int flags, auth_dns = 0;
   struct in_addr netmask;
   int pipefd[2];
+  struct iovec tcpbuff;
 #ifdef HAVE_LINUX_NETWORK
   unsigned char a = 0;
 #endif
@@ -2156,10 +2159,8 @@ static void do_tcp_connection(struct listener *listener, time_t now, int slot)
   if ((flags = fcntl(confd, F_GETFL, 0)) != -1)
     while(retry_send(fcntl(confd, F_SETFL, flags & ~O_NONBLOCK)));
 
-  buff = tcp_request(confd, now, &tcp_addr, netmask, auth_dns);
-	      
-  if (buff)
-    free(buff);
+  tcp_request(confd, now, &tcpbuff, &tcp_addr, netmask, auth_dns);
+  free(tcpbuff.iov_base);
   
   for (s = daemon->servers; s; s = s->next)
     if (s->tcpfd != -1)

src/dnsmasq.h

@@ -290,7 +290,8 @@ struct event_desc {
 #define OPT_AUTH_LOG       76
 #define OPT_LEASEQUERY     77
 #define OPT_LOG_ONLY_FAILED  78
-#define OPT_LAST           79
+#define OPT_LOG_MALLOC     79
+#define OPT_LAST           80
 
 #define OPTION_BITS (sizeof(unsigned int)*8)
 #define OPTION_SIZE ( (OPT_LAST/OPTION_BITS)+((OPT_LAST%OPTION_BITS)!=0) )
@@ -1207,6 +1208,7 @@ extern struct daemon {
   int log_fac; /* log facility */
   char *log_file; /* optional log file */
   int max_logs;  /* queue limit */
+  int log_malloc; /* log malloc/realloc/free */
   int randport_limit; /* Maximum number of source ports for query. */
   int cachesize, ftabsize;
   int port, query_port, min_port, max_port;
@@ -1495,8 +1497,12 @@ unsigned char *do_rfc1035_name(unsigned char *p, char *sval, char *limit);
 void *safe_malloc(size_t size);
 void safe_strncpy(char *dest, const char *src, size_t size);
 void safe_pipe(int *fd, int read_noblock);
-void *whine_malloc(size_t size);
-void *whine_realloc(void *ptr, size_t size);
+#define whine_malloc(x) whine_malloc_real(__func__, __LINE__, (x))
+#define whine_realloc(x, y) whine_realloc_real(__func__, __LINE__, (x), (y))
+#define free(x) free_real(__func__, __LINE__, (x))
+void free_real(const char *func, unsigned int line, void *ptr);
+void *whine_malloc_real(const char *func, unsigned int line, size_t size);
+void *whine_realloc_real(const char *func, unsigned int line, void *ptr, size_t size);
 int sa_len(union mysockaddr *addr);
 int sockaddr_isequal(const union mysockaddr *s1, const union mysockaddr *s2);
 int sockaddr_isnull(const union mysockaddr *s);
@@ -1521,6 +1527,7 @@ int memcmp_masked(unsigned char *a, unsigned char *b, int len,
 int expand_buf(struct iovec *iov, size_t size);
 char *print_mac(char *buff, unsigned char *mac, int len);
 int read_write(int fd, unsigned char *packet, int size, int rw);
+int read_writev(int fd, struct iovec *iov, int iovcnt, int rw);
 void close_fds(long max_fd, int spare1, int spare2, int spare3);
 int wildcard_match(const char* wildcard, const char* match);
 int wildcard_matchn(const char* wildcard, const char* match, int num);
@@ -1563,8 +1570,8 @@ int tcp_from_udp(time_t now, int status, struct dns_header *header, ssize_t *n,
 		 int class, char *name, struct server *server, 
 		 int *keycount, int *validatecount);
 #endif
-unsigned char *tcp_request(int confd, time_t now,
-			   union mysockaddr *local_addr, struct in_addr netmask, int auth_dns);
+void tcp_request(int confd, time_t now, struct iovec *bigbuff,
+		 union mysockaddr *local_addr, struct in_addr netmask, int auth_dns);
 void server_gone(struct server *server);
 int send_from(int fd, int nowild, char *packet, size_t len, 
 	       union mysockaddr *to, union all_addr *source,

src/forward.c

@@ -135,9 +135,9 @@ static void log_query_mysockaddr(unsigned int flags, char *name, union mysockadd
 }
 
 static void server_send(struct server *server, int fd,
-			const void *header, size_t plen, int flags)
+			const void *header, size_t plen)
 {
-  while (retry_send(sendto(fd, header, plen, flags,
+  while (retry_send(sendto(fd, header, plen, 0,
 			   &server->addr.sa,
 			   sa_len(&server->addr))));
 }
@@ -1074,7 +1074,7 @@ static void dnssec_validate(struct frec *forward, struct dns_header *header,
 		    set_outgoing_mark(orig, fd);
 #endif
 
-		  server_send(server, fd, header, nn, 0);
+		  server_send(server, fd, header, nn);
 		  server->queries++;
 #ifdef HAVE_DUMPFILE
 		  dump_packet_udp(DUMP_SEC_QUERY, (void *)header, (size_t)nn, NULL, &server->addr, fd);
@@ -2018,20 +2018,21 @@ void receive_query(struct listener *listen, time_t now)
 }
 
  
-/* Send query in packet, qsize to a server determined by first,last,start and
-   get the reply. return reply size. */
-static ssize_t tcp_talk(int first, int last, int start, unsigned char *packet,  size_t qsize,
-			int have_mark, unsigned int mark, struct server **servp)
+/* Send query in header, qsize to a server determined by first,last,start and
+   get the reply into the buffer passed by outbuff. Return reply size. */
+static ssize_t tcp_talk(int first, int last, int start, struct dns_header *header, size_t qsize,
+			struct iovec *recvbuff, int have_mark, unsigned int mark, struct server **servp)
 {
   int firstsendto = -1;
-  u16 *length = (u16 *)packet;
-  unsigned char *payload = &packet[2];
-  struct dns_header *header = (struct dns_header *)payload;
-  unsigned int rsize;
+  u16 length;
+  unsigned int rsize = 0;
   int class, rclass, type, rtype;
   unsigned char *p;
-  struct blockdata *saved_question;
   struct timeval tv;
+  struct iovec sendio[2];
+#ifdef MSG_FASTOPEN
+  struct msghdr msg;
+#endif
   
   (void)mark;
   (void)have_mark;
@@ -2043,10 +2044,12 @@ static ssize_t tcp_talk(int first, int last, int start, unsigned char *packet,
   GETSHORT(type, p); 
   GETSHORT(class, p);
 
-  /* Save question for retry. */
-  if (!(saved_question = blockdata_alloc((char *)header, (size_t)qsize)))
-    return 0;
-  
+  length = htons(qsize);
+  sendio[0].iov_base = &length;
+  sendio[0].iov_len = sizeof(length);
+  sendio[1].iov_base = header;
+  sendio[1].iov_len = qsize;
+
   while (1) 
     {
       int data_sent = 0, fatal = 0;
@@ -2068,10 +2071,6 @@ static ssize_t tcp_talk(int first, int last, int start, unsigned char *packet,
       *servp = serv = daemon->serverarray[start];
       
     retry:
-      blockdata_retrieve(saved_question, qsize, header);
-      
-      *length = htons(qsize);
-      
       if (serv->tcpfd == -1)
 	{
 	  if ((serv->tcpfd = socket(serv->addr.sa.sa_family, SOCK_STREAM, 0)) == -1)
@@ -2102,10 +2101,17 @@ static ssize_t tcp_talk(int first, int last, int start, unsigned char *packet,
 	  tv.tv_sec += TCP_TIMEOUT;
 	  setsockopt(serv->tcpfd, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv));
 #endif
-	  
 #ifdef MSG_FASTOPEN
-	  server_send(serv, serv->tcpfd, packet, qsize + sizeof(u16), MSG_FASTOPEN);
-	  
+	  msg.msg_name = &serv->addr.sa;
+	  msg.msg_namelen = sa_len(&serv->addr);
+	  msg.msg_iov = sendio;
+	  msg.msg_iovlen = 2;
+	  msg.msg_control = NULL;
+	  msg.msg_controllen = 0;
+	  msg.msg_flags = 0;
+
+	  while (retry_send(sendmsg(serv->tcpfd, &msg, MSG_FASTOPEN)));
+
 	  if (errno == 0)
 	    data_sent = 1;
 	  else if (errno == ETIMEDOUT || errno == EHOSTUNREACH || errno == EINPROGRESS || errno == ECONNREFUSED)
@@ -2131,12 +2137,13 @@ static ssize_t tcp_talk(int first, int last, int start, unsigned char *packet,
 	  serv->flags &= ~SERV_GOT_TCP;
 	}
       
-      /* We us the _ONCE veriant of read_write() here because we've set a timeout on the tcp socket
+      /* We use the _ONCE variant of read_write() here because we've set a timeout on the tcp socket
 	 and wish to abort if the whole data is not read/written within the timeout. */      
-      if ((!data_sent && !read_write(serv->tcpfd, (unsigned char *)packet, qsize + sizeof(u16), RW_WRITE_ONCE)) ||
-	  !read_write(serv->tcpfd, (unsigned char *)length, sizeof (*length), RW_READ_ONCE) ||
-	  !read_write(serv->tcpfd, payload, (rsize = ntohs(*length)), RW_READ_ONCE))
-	{
+      if ((!data_sent && !read_writev(serv->tcpfd, sendio, 2, RW_WRITE_ONCE)) ||
+	  !read_write(serv->tcpfd, (unsigned char *)&length, sizeof(length), RW_READ_ONCE) ||
+	  !expand_buf(recvbuff, (rsize = ntohs(length))) ||
+	  !read_write(serv->tcpfd, recvbuff->iov_base, rsize, RW_READ_ONCE))
+      {
 	  /* We get data then EOF, reopen connection to same server,
 	     else try next. This avoids DoS from a server which accepts
 	     connections and then closes them. */
@@ -2149,28 +2156,28 @@ static ssize_t tcp_talk(int first, int last, int start, unsigned char *packet,
 	  else
 	    goto failed;
 	}
+      else
+	{
+	  /* If the question section of the reply doesn't match the question we sent, then
+	     someone might be attempting to insert bogus values into the cache by 
+	     sending replies containing questions and bogus answers.
+	     Try another server, or give up */
+	  p = (unsigned char *)(((struct dns_header *)recvbuff->iov_base)+1);
+	  if (extract_name(((struct dns_header *)recvbuff->iov_base), rsize, &p, daemon->namebuff, EXTR_NAME_COMPARE, 4) != 1)
+	    continue;
+	  GETSHORT(rtype, p); 
+	  GETSHORT(rclass, p);
       
-      /* If the question section of the reply doesn't match the question we sent, then
-	 someone might be attempting to insert bogus values into the cache by 
-	 sending replies containing questions and bogus answers.
-	 Try another server, or give up */
-      p = (unsigned char *)(header+1);
-      if (extract_name(header, rsize, &p, daemon->namebuff, EXTR_NAME_COMPARE, 4) != 1)
-	continue;
-      GETSHORT(rtype, p); 
-      GETSHORT(rclass, p);
-      
-      if (type != rtype || class != rclass)
-	continue;
+	  if (type != rtype || class != rclass)
+	    continue;
+	}
       
       serv->flags |= SERV_GOT_TCP;
       
       *servp = serv;
-      blockdata_free(saved_question);
       return rsize;
     }
   
-  blockdata_free(saved_question);
   return 0;
 }
 		  
@@ -2182,19 +2189,16 @@ int tcp_from_udp(time_t now, int status, struct dns_header *header, ssize_t *ple
 		 int class, char *name, struct server *server, 
 		 int *keycount, int *validatecount)
 {
-  unsigned char *packet = whine_malloc(65536 + MAXDNAME + RRFIXEDSZ + sizeof(u16));
-  struct dns_header *new_header = (struct dns_header *)&packet[2];
   int start, first, last, new_status;
   ssize_t n = *plenp;
   int log_save = daemon->log_display_id;
+  struct iovec recvbuff;
+
+  recvbuff.iov_len = 0;
+  recvbuff.iov_base = NULL;
   
   *plenp = 0;
   
-  if (!packet)
-    return STAT_ABANDONED;
-
-  memcpy(new_header, header, n);
-
   /* Set TCP flag in logs. */
   daemon->log_display_id = -daemon->log_display_id;
 
@@ -2212,10 +2216,11 @@ int tcp_from_udp(time_t now, int status, struct dns_header *header, ssize_t *ple
 	log_query_mysockaddr(F_NOEXTRA | F_DNSSEC | F_SERVER, name, &server->addr,
 			     STAT_ISEQUAL(status, STAT_NEED_KEY) ? "dnssec-query[DNSKEY]" : "dnssec-query[DS]", 0);
 
-      if ((n = tcp_talk(first, last, start, packet, n, 0, 0, &server)) == 0)
+      if ((n = tcp_talk(first, last, start, header, n, &recvbuff, 0, 0, &server)) == 0)
 	new_status = STAT_ABANDONED;
       else
 	{
+	  struct dns_header *new_header = (struct dns_header *)recvbuff.iov_base;
 	  new_status = tcp_key_recurse(now, status, new_header, n, class, daemon->namebuff, daemon->keyname, server, 0, 0, keycount, validatecount);
 	  
 	  if (STAT_ISEQUAL(status, STAT_OK))
@@ -2252,7 +2257,7 @@ int tcp_from_udp(time_t now, int status, struct dns_header *header, ssize_t *ple
     }
   
   daemon->log_display_id = log_save;
-  free(packet);
+  free(recvbuff.iov_base);
   return new_status;
 }			    
  
@@ -2262,9 +2267,12 @@ static int tcp_key_recurse(time_t now, int status, struct dns_header *header, si
 			   int have_mark, unsigned int mark, int *keycount, int *validatecount)
 {
   int first, last, start, new_status;
-  unsigned char *packet = NULL;
-  struct dns_header *new_header = NULL;
+  struct iovec new_packet;
+  struct dns_header *query_header = NULL, *new_header;
 
+  new_packet.iov_base = NULL;
+  new_packet.iov_len = 0;
+  
   while (1)
     {
       size_t m;
@@ -2298,34 +2306,21 @@ static int tcp_key_recurse(time_t now, int status, struct dns_header *header, si
 	}
       
       /* Can't validate because we need a key/DS whose name now in keyname.
-	 Make query for same, and recurse to validate */
-      if (!packet)
-	{
-	  packet = whine_malloc(65536 + MAXDNAME + RRFIXEDSZ + sizeof(u16));
-	  new_header = (struct dns_header *)&packet[2];
-	}
-      
-      if (!packet)
-	{
-	  new_status = STAT_ABANDONED;
-	  break;
-	}
-      
-      m = dnssec_generate_query(new_header, ((unsigned char *) new_header) + 65536, keyname, class, 0,
+	 Make query for same in UDP packet buffer, recurse to validate*/
+      query_header = (struct dns_header *)daemon->packet;
+      daemon->srv_save = NULL;
+
+      m = dnssec_generate_query(query_header, ((unsigned char *)query_header) + daemon->edns_pktsz, keyname, class, 0,
 				STAT_ISEQUAL(new_status, STAT_NEED_KEY) ? T_DNSKEY : T_DS);
       
-      if ((start = dnssec_server(server, keyname, STAT_ISEQUAL(new_status, STAT_NEED_DS), &first, &last)) == -1)
+      if ((start = dnssec_server(server, keyname, STAT_ISEQUAL(new_status, STAT_NEED_DS), &first, &last)) == -1 ||
+	  (m = tcp_talk(first, last, start, query_header, m, &new_packet, have_mark, mark, &server)) == 0)
 	{
 	  new_status = STAT_ABANDONED;
 	  break;
 	}
       
-      if ((m = tcp_talk(first, last, start, packet, m, have_mark, mark, &server)) == 0)
-	{
-	  new_status = STAT_ABANDONED;
-	  break;
-	}
-
+      new_header = new_packet.iov_base;
       log_save = daemon->log_display_id;
       daemon->log_display_id = -(++daemon->log_id);
       
@@ -2342,8 +2337,7 @@ static int tcp_key_recurse(time_t now, int status, struct dns_header *header, si
 	break; 
     }
   
-  if (packet)
-    free(packet);
+  free(new_packet.iov_base);
   
   return new_status;
 }
@@ -2353,11 +2347,11 @@ static int tcp_key_recurse(time_t now, int status, struct dns_header *header, si
 /* The daemon forks before calling this: it should deal with one connection,
    blocking as necessary, and then return. Note, need to be a bit careful
    about resources for debug mode, when the fork is suppressed: that's
-   done by the caller. */
-unsigned char *tcp_request(int confd, time_t now,
-			   union mysockaddr *local_addr, struct in_addr netmask, int auth_dns)
+   done by the caller, which also frees bigbuff. */
+void tcp_request(int confd, time_t now, struct iovec *bigbuff, 
+		 union mysockaddr *local_addr, struct in_addr netmask, int auth_dns)
 {
-  size_t size = 0, saved_size = 0;
+  size_t size = 0;
   int norebind = 0;
 #ifdef HAVE_CONNTRACK
   int allowed = 1;
@@ -2366,16 +2360,10 @@ unsigned char *tcp_request(int confd, time_t now,
   int local_auth = 0;
 #endif
   int checking_disabled, do_bit = 0, ad_reqd = 0, have_pseudoheader = 0;
-  struct blockdata *saved_question = NULL;
   unsigned short qtype;
   unsigned int gotname = 0;
-  /* Max TCP packet + slop + size */
-  unsigned char *packet = whine_malloc(65536 + MAXDNAME + RRFIXEDSZ + sizeof(u16));
-  unsigned char *payload = &packet[2];
-  u16 tcp_len;
-  /* largest field in header is 16-bits, so this is still sufficiently aligned */
-  struct dns_header *header = (struct dns_header *)payload;
-  u16 *length = (u16 *)packet;
+  u16 tcp_len, out_len;
+  struct dns_header *header, *out_header;
   struct server *serv;
   struct in_addr dst_addr_4;
   union mysockaddr peer_addr;
@@ -2385,9 +2373,13 @@ unsigned char *tcp_request(int confd, time_t now,
   unsigned int mark = 0;
   int have_mark = 0;
   int first, last, filtered, do_stale = 0;
-      
-  if (!packet || getpeername(confd, (struct sockaddr *)&peer_addr, &peer_len) == -1)
-    return packet;
+  struct iovec out_iov[2];
+  
+  bigbuff->iov_base = NULL;
+  bigbuff->iov_len = 0;
+  
+  if (getpeername(confd, (struct sockaddr *)&peer_addr, &peer_len) == -1)
+    return;
 
 #ifdef HAVE_CONNTRACK
   /* Get connection mark of incoming query to set on outgoing connections. */
@@ -2408,7 +2400,7 @@ unsigned char *tcp_request(int confd, time_t now,
   if (option_bool(OPT_LOCAL_SERVICE))
     {
       struct addrlist *addr;
-
+      
       if (peer_addr.sa.sa_family == AF_INET6) 
 	{
 	  for (addr = daemon->interface_addrs; addr; addr = addr->next)
@@ -2431,7 +2423,7 @@ unsigned char *tcp_request(int confd, time_t now,
 	{
 	  prettyprint_addr(&peer_addr, daemon->addrbuff);
 	  my_syslog(LOG_WARNING, _("ignoring query from non-local network %s"), daemon->addrbuff);
-	  return packet;
+	  return;
 	}
     }
 
@@ -2449,18 +2441,26 @@ unsigned char *tcp_request(int confd, time_t now,
 	  if (query_count >= TCP_MAX_QUERIES)
 	    break;
 	  
+	  /* Now get the query into the normal UDP packet buffer.
+	     Ignore queries long than this. If we're answering locally,
+	     copy the query into the output buffer, but for forwarding, tcp_talk()
+	     wants the query in a a different buffer from the reply.
+	     Note that we overwrote any saved UDP query - this onlt matters in debug mode. */
+	  daemon->srv_save = NULL;
 	  if (!read_write(confd, (unsigned char *)&tcp_len, sizeof(tcp_len), RW_READ) ||
-	      !(size = ntohs(tcp_len)) ||
-	      !read_write(confd, payload, size, RW_READ))
+	      !(size = ntohs(tcp_len)) || size > (size_t)daemon->packet_buff_sz ||
+	      !read_write(confd, (unsigned char *)daemon->packet, size, RW_READ))
 	    break;
-	      
+	  
 	  if (size < (int)sizeof(struct dns_header))
 	    continue;
 	  
-	  /* Clear buffer beyond request to avoid risk of
-	     information disclosure. */
-	  memset(payload + size, 0, 65536 - size);
+	  /* Make sure we have a buffer big enough for the largest answer. */
+	  expand_buf(bigbuff, 65536 + MAXDNAME + RRFIXEDSZ);
+	  out_header = bigbuff->iov_base;
 	  
+	  /* header == query */
+	  header = (struct dns_header *)daemon->packet;
 	  query_count++;
 	  
 	  /* log_query gets called indirectly all over the place, so 
@@ -2479,9 +2479,6 @@ unsigned char *tcp_request(int confd, time_t now,
 	    ede = EDE_INVALID_DATA;
 	  else
 	    {
-	      if (saved_question)
-		blockdata_free(saved_question);
-
 	      do_bit = 0;
 	      
 	      if (find_pseudoheader(header, (size_t)size, NULL, &pheader, NULL, NULL))
@@ -2496,9 +2493,12 @@ unsigned char *tcp_request(int confd, time_t now,
 		    do_bit = 1; /* do bit */ 
 		}
 
-	      size = add_edns0_config(header, size, ((unsigned char *) header) + 65536, &peer_addr, now, &cacheable);
-	      saved_question = blockdata_alloc((char *)header, (size_t)size);
-	      saved_size = size;
+	      size = add_edns0_config(header, size, ((unsigned char *) header) + daemon->edns_pktsz, &peer_addr, now, &cacheable);
+
+	      /* Clear buffer to avoid risk of information disclosure. */
+	      memset(bigbuff->iov_base, 0, bigbuff->iov_len);
+	      /* Copy query into output buffer for local answering */
+	      memcpy(out_header, header, size);	    
 	      
 	      log_query_mysockaddr((auth_dns ? F_NOERR | F_AUTH : 0) | F_QUERY | F_FORWARD, daemon->namebuff,
 				   &peer_addr, NULL, qtype);
@@ -2557,10 +2557,10 @@ unsigned char *tcp_request(int confd, time_t now,
 #endif
 #ifdef HAVE_AUTH
 	      else if (auth_dns)
-		m = answer_auth(header, ((char *) header) + 65536, (size_t)size, now, &peer_addr, local_auth);
+		m = answer_auth(out_header, ((char *) out_header) + 65536, (size_t)size, now, &peer_addr, local_auth);
 #endif
 	      else
-		m = answer_request(header, ((char *) header) + 65536, (size_t)size, 
+		m = answer_request(out_header, ((char *) out_header) + 65536, (size_t)size, 
 				   dst_addr_4, netmask, now, ad_reqd, do_bit, !cacheable, &stale, &filtered);
 	    }
 	}
@@ -2568,15 +2568,12 @@ unsigned char *tcp_request(int confd, time_t now,
       /* Do this by steam now we're not in the select() loop */
       check_log_writer(1); 
       
-      if (m == 0 && ede == EDE_UNSET && saved_question)
+      if (m == 0 && ede == EDE_UNSET)
 	{
 	  struct server *master;
 	  int start;
 	  int no_cache_dnssec = 0, cache_secure = 0, bogusanswer = 0;
 
-	  blockdata_retrieve(saved_question, (size_t)saved_size, header);
-	  size = saved_size;
-
 	  /* save state of "cd" flag in query */
 	  checking_disabled = header->hb4 & HB4_CD;
 	  
@@ -2605,7 +2602,7 @@ unsigned char *tcp_request(int confd, time_t now,
 #ifdef HAVE_DNSSEC
 		  if (option_bool(OPT_DNSSEC_VALID))
 		    {
-		      size = add_do_bit(header, size, ((unsigned char *) header) + 65536);
+		      size = add_do_bit(header, size, ((unsigned char *) header) + daemon->edns_pktsz);
 		      
 		      /* For debugging, set Checking Disabled, otherwise, have the upstream check too,
 			 this allows it to select auth servers when one is returning bad data. */
@@ -2615,12 +2612,14 @@ unsigned char *tcp_request(int confd, time_t now,
 #endif
 		  
 		  /* Loop round available servers until we succeed in connecting to one. */
-		  if ((m = tcp_talk(first, last, start, packet, size, have_mark, mark, &serv)) == 0)
+		  if ((m = tcp_talk(first, last, start, header, size, bigbuff, have_mark, mark, &serv)) == 0)
 		    ede = EDE_NETERR;
 		  else
 		    {
+		      /* just in case tcp_talk() expanded buffer - should never happen */
+		      out_header = bigbuff->iov_base;
 		      /* get query name again for logging - may have been overwritten */
-		      if (!extract_name(header, (unsigned int)size, NULL, daemon->namebuff, EXTR_NAME_EXTRACT, 0))
+		      if (!extract_name(out_header, (unsigned int)size, NULL, daemon->namebuff, EXTR_NAME_EXTRACT, 0))
 			strcpy(daemon->namebuff, "query");
 		      log_query_mysockaddr(F_SERVER | F_FORWARD, daemon->namebuff, &serv->addr, NULL, 0);
 		      
@@ -2636,7 +2635,8 @@ unsigned char *tcp_request(int confd, time_t now,
 			    {
 			      int keycount = daemon->limit[LIMIT_WORK]; /* Limit to number of DNSSEC questions, to catch loops and avoid filling cache. */
 			      int validatecount = daemon->limit[LIMIT_CRYPTO]; 
-			      int status = tcp_key_recurse(now, STAT_OK, header, m, 0, daemon->namebuff, daemon->keyname, 
+			      /* tcp_key_recurse() may overwrite packetbuf, and thuse *header is now invalid */
+			      int status = tcp_key_recurse(now, STAT_OK, out_header, m, 0, daemon->namebuff, daemon->keyname, 
 							   serv, have_mark, mark, &keycount, &validatecount);
 			      char *result, *domain = "result";
 			      
@@ -2662,7 +2662,7 @@ unsigned char *tcp_request(int confd, time_t now,
 				  no_cache_dnssec = 1;
 				  bogusanswer = 1;
 				  
-				  if (extract_name(header, m, NULL, daemon->namebuff, EXTR_NAME_EXTRACT, 0))
+				  if (extract_name(out_header, m, NULL, daemon->namebuff, EXTR_NAME_EXTRACT, 0))
 				    domain = daemon->namebuff;
 				}
 			      
@@ -2686,18 +2686,18 @@ unsigned char *tcp_request(int confd, time_t now,
 		    
 		      /* restore CD bit to the value in the query */
 		      if (checking_disabled)
-			header->hb4 |= HB4_CD;
+			out_header->hb4 |= HB4_CD;
 		      else
-			header->hb4 &= ~HB4_CD;
+			out_header->hb4 &= ~HB4_CD;
 		      
 		      /* Never cache answers which are contingent on the source or MAC address EDSN0 option,
 			 since the cache is ignorant of such things. */
 		      if (!cacheable)
 			no_cache_dnssec = 1;
 		      
-		      m = process_reply(header, now, serv, (unsigned int)m, 
+		      m = process_reply(out_header, now, serv, (unsigned int)m, 
 					option_bool(OPT_NO_REBIND) && !norebind, no_cache_dnssec, cache_secure, bogusanswer,
-					ad_reqd, do_bit, !have_pseudoheader, &peer_addr, ((unsigned char *)header) + 65536, ede);
+					ad_reqd, do_bit, !have_pseudoheader, &peer_addr, ((unsigned char *)out_header) + 65536, ede);
 
 		      /* process_reply() adds pheader itself */
 		      have_pseudoheader = 0; 
@@ -2712,8 +2712,8 @@ unsigned char *tcp_request(int confd, time_t now,
       /* In case of local answer or no connections made. */
       if (m == 0)
 	{
-	  if (!(m = make_local_answer(flags, gotname, size, header, daemon->namebuff,
-				      ((char *) header) + 65536, first, last, ede)))
+	  if (!(m = make_local_answer(flags, gotname, size, out_header, daemon->namebuff,
+				      ((char *) out_header) + 65536, first, last, ede)))
 	    break;
 	}
       else if (ede == EDE_UNSET)
@@ -2729,14 +2729,12 @@ unsigned char *tcp_request(int confd, time_t now,
 	  u16 swap = htons((u16)ede);
 	  
 	  if (ede != EDE_UNSET)
-	    m = add_pseudoheader(header, m, ((unsigned char *) header) + 65536, EDNS0_OPTION_EDE, (unsigned char *)&swap, 2, do_bit, 0);
+	    m = add_pseudoheader(out_header, m, ((unsigned char *) out_header) + 65536, EDNS0_OPTION_EDE, (unsigned char *)&swap, 2, do_bit, 0);
 	  else
-	    m = add_pseudoheader(header, m, ((unsigned char *) header) + 65536, 0, NULL, 0, do_bit, 0);
+	    m = add_pseudoheader(out_header, m, ((unsigned char *) out_header) + 65536, 0, NULL, 0, do_bit, 0);
 	}
-		  
-      check_log_writer(1);
       
-      *length = htons(m);
+      check_log_writer(1);
       
 #if defined(HAVE_CONNTRACK) && defined(HAVE_UBUS)
 #ifdef HAVE_AUTH
@@ -2746,7 +2744,13 @@ unsigned char *tcp_request(int confd, time_t now,
 	  report_addresses(header, m, mark);
 #endif
       
-      if (!read_write(confd, packet, m + sizeof(u16), RW_WRITE))
+      /* use scatter-gather IO so that length doesn't end up in separate packet. */
+      out_len = htons(m);
+      out_iov[0].iov_len = sizeof(out_len);
+      out_iov[0].iov_base = &out_len;
+      out_iov[1].iov_len = m;
+      out_iov[1].iov_base = bigbuff->iov_base;
+      if (!read_writev(confd, out_iov, 2, RW_WRITE))
 	break;
       
       /* If we answered with stale data, this process will now try and get fresh data into
@@ -2762,18 +2766,15 @@ unsigned char *tcp_request(int confd, time_t now,
 	  daemon->log_source_addr = NULL;
 	}
     }
-
-  /* If we ran once to get fresh data, confd is already closed. */
+  
+/* If we ran once to get fresh data, confd is already closed. */
   if (!do_stale)
     {
       shutdown(confd, SHUT_RDWR);
       close(confd);
     }
-
-  blockdata_free(saved_question);
+  
   check_log_writer(1);
-
-  return packet;
 }
 
 /* return a UDP socket bound to a random port, have to cope with straying into
@@ -3274,7 +3275,7 @@ void resend_query(void)
 {
   if (daemon->srv_save)
     server_send(daemon->srv_save, daemon->fd_save,
-		daemon->packet, daemon->packet_len, 0);
+		daemon->packet, daemon->packet_len);
 }
 
 /* A server record is going away, remove references to it */

src/option.c

@@ -197,6 +197,7 @@ struct myoption {
 #define LOPT_DO_ENCODE     388
 #define LOPT_LEASEQUERY    389
 #define LOPT_SPLIT_RELAY   390
+#define LOPT_LOG_MALLOC    391
 
 #ifdef HAVE_GETOPT_LONG
 static const struct option opts[] =  
@@ -398,6 +399,7 @@ static const struct myoption opts[] =
     { "no-ident", 0, 0, LOPT_NO_IDENT },
     { "max-tcp-connections", 1, 0, LOPT_MAX_PROCS },
     { "leasequery", 2, 0, LOPT_LEASEQUERY },
+    { "log-malloc", 0, 0, LOPT_LOG_MALLOC },
     { NULL, 0, 0, 0 }
   };
 
@@ -606,6 +608,7 @@ static struct {
   { LOPT_NO_IDENT, OPT_NO_IDENT, NULL, gettext_noop("Do not add CHAOS TXT records."), NULL },
   { LOPT_CACHE_RR, ARG_DUP, "<RR-type>", gettext_noop("Cache this DNS resource record type."), NULL },
   { LOPT_MAX_PROCS, ARG_ONE, "<integer>", gettext_noop("Maximum number of concurrent tcp connections."), NULL },
+  { LOPT_LOG_MALLOC, OPT_LOG_MALLOC, NULL, gettext_noop("Log memory allocation for debugging."), NULL },
   { 0, 0, NULL, NULL, NULL }
 }; 
 

src/util.c

@@ -341,26 +341,6 @@ void safe_pipe(int *fd, int read_noblock)
     die(_("cannot create pipe: %s"), NULL, EC_MISC);
 }
 
-void *whine_malloc(size_t size)
-{
-  void *ret = calloc(1, size);
-
-  if (!ret)
-    my_syslog(LOG_ERR, _("failed to allocate %d bytes"), (int) size);
-  
-  return ret;
-}
-
-void *whine_realloc(void *ptr, size_t size)
-{
-  void *ret = realloc(ptr, size);
-
-  if (!ret)
-    my_syslog(LOG_ERR, _("failed to reallocate %d bytes"), (int) size);
-
-  return ret;
-}
-
 int sockaddr_isequal(const union mysockaddr *s1, const union mysockaddr *s2)
 {
   if (s1->sa.sa_family == s2->sa.sa_family)
@@ -777,43 +757,60 @@ int retry_send(ssize_t rc)
    "once" fails on EAGAIN, as this a timeout.
    This indicates a timeout of a TCP socket.
 */
-int read_write(int fd, unsigned char *packet, int size, int rw)
+int read_writev(int fd, struct iovec *iov, int iovcnt, int rw)
 {
-  ssize_t n, done;
-  
-  for (done = 0; done < size; done += n)
-    {
-      if (rw & 1)
-	n = read(fd, &packet[done], (size_t)(size - done));
-      else
-	n = write(fd, &packet[done], (size_t)(size - done));
-      
-      if (n == 0)
-	return 0;
+  int cur = 0;
+  ssize_t n, done = 0;
 
+  while (cur < iovcnt)
+    {
+      iov[cur].iov_len -= done;
+      iov[cur].iov_base =  ((char *)iov[cur].iov_base) + done;
+
+      if (rw & 1)
+	n = readv(fd, &iov[cur], iovcnt - cur);
+      else
+	n = writev(fd, &iov[cur], iovcnt - cur);
+
+      iov[cur].iov_len += done;
+      iov[cur].iov_base = ((char *)iov[cur].iov_base) - done;
+      
       if (n == -1)
 	{
-	  n = 0; /* don't mess with counter when we loop. */
-
 	  if (errno == EINTR || errno == ENOMEM || errno == ENOBUFS)
 	    continue;
-
-	  if (errno == EAGAIN || errno == EWOULDBLOCK)
-	    {
-	      /* "once" variant */
-	      if (rw & 2)
-		return 0;
-
-	      continue;
-	    }
+	  
+	  if (!(rw & 2) && (errno == EAGAIN || errno == EWOULDBLOCK))
+	    continue;
 
 	  return 0;
 	}
+      
+      if (n == 0 && (rw & 1))
+	return 0;
+      
+      done += n;
+      while ((size_t)done >= iov[cur].iov_len)
+	done -= iov[cur++].iov_len;
     }
           
   return 1;
 }
 
+int read_write(int fd, unsigned char *packet, int size, int rw)
+{
+  struct iovec iov;
+
+  /* size == 0 is not an error, just a NOOP. */
+  if (size == 0)
+    return 1;
+  
+  iov.iov_len = (size_t)size;
+  iov.iov_base = packet;
+
+  return read_writev(fd, &iov, 1, rw);
+}
+
 /* close all fds except STDIN, STDOUT and STDERR, spare1, spare2 and spare3 */
 void close_fds(long max_fd, int spare1, int spare2, int spare3) 
 {
@@ -949,3 +946,42 @@ int kernel_version(void)
   return version * 256 + (split ? atoi(split) : 0);
 }
 #endif
+
+#define hash_ptr(x) (((unsigned int)(((char *)(x)) - ((char *)NULL))) & 0xffffff)
+
+void *whine_malloc_real(const char *func, unsigned int line, size_t size)
+{
+  void *ret = calloc(1, size);
+  
+  if (!ret)
+    my_syslog(LOG_ERR, _("failed to allocate %d bytes"), (int) size);
+  else if (daemon->log_malloc)
+    my_syslog(LOG_INFO, _("malloc: %s:%u %zu bytes at %x"), func, line, size, hash_ptr(ret));
+
+  return ret;
+}
+
+void *whine_realloc_real(const char *func, unsigned int line, void *ptr, size_t size)
+{
+  unsigned int old = hash_ptr(ptr);
+  void *ret = realloc(ptr, size);
+  
+  if (!ret)
+    my_syslog(LOG_ERR, _("failed to reallocate %d bytes"), (int) size);
+  else if (daemon->log_malloc)
+    my_syslog(LOG_INFO, _("realloc: %s:%u %zu bytes from %x to %x"), func, line, size, old, hash_ptr(ret));
+  
+  return ret;
+}
+
+#undef free
+void free_real(const char *func, unsigned int line, void *ptr)
+{
+  if (ptr)
+    {
+      if (daemon->log_malloc)
+	my_syslog(LOG_INFO, _("free: %s:%u block at %x"), func, line, hash_ptr(ptr));
+  
+      free(ptr);
+    }
+}