Fix SEGV in Dbus code.

A 'AddDhcpLease' DBus message which fails to allocate
a DHCP lease will cause dnsmasq to dereference a NULL pointer
and crash.

Allocation failure can happen if the particular
flavour (IPv4 pr IPv6) of DHCP is not configured in
this dnsmasq instance, or if the configured limit on
DHCP leases is reached or (unlikley) if dnsmasq cannot
allocate memory.

This patch provides error returns for all these situations.

This patch is inspired by a patch from
Stefan Hanreich <s.hanreich@proxmox.com> which was prompted by
a bug report from Lou Lécrivain.

CHANGELOG

@@ -70,6 +70,12 @@ version 2.92
 	is functional when client and server networks aren't mutually
 	route-able.
 
+	Fix failure to add client MAC address to queries in TCP mode.
+	The options which cause	dnsmasq	to decorate a DNS query	with the MAC
+	address	on the originating client can fail when the query is sent
+	using TCP. Thanks to Bruno Ravara for spotting and
+	characterising this bug.
+
 	
 version 2.91
 	Fix spurious "resource limit exceeded messages". Thanks to 

man/dnsmasq.8

@@ -1489,19 +1489,22 @@ prefix-delegation from relayed DHCP transactions. See
 .B --dhcp-script
 for details.
 .TP
-.B --dhcp-split-relay=<local address>,[<server address>[#<server port>]],<server-facing-interface>
+.B --dhcp-split-relay=<local address>,[<server address>[#<server port>]],<server-facing-interface>|<server-facing-address>
 A usefully enchanced version of DHCPv4 relay. IPv4 DHCP normally uses a single address
 for two functions; it is used by the DHCP server to determine which network to allocate
 an address on, and it is used as the address of the relay to which the server sends packets.
 
 This version of DHCP relay splits these functions. It uses the address of the server-facing relay
-interface as the address that the server talks to. The address of the client-facing interface
+interface or a directly-specified address as the address that the server talks to. The address of the client-facing interface
 (the first item in the config) is used as to determine the client's subnet. The
 local address is also used as server-ID override so that the client always sends requests
 via the relay. The effect of this is that server doesn't require
 a route to the client network and the clients don't require a route to the server.
 
-The interface parameter is mandatory and a cannot be a wildcard.
+The third parameter is mandatory. If it is an interface name it cannot be a wildcard and the same filtering as descibed in
+--dhcp-relay applies; answers from the server must arrve via the specified interface. If the third parameter
+is an IP address it must be an address of a local interface which is routable from the server; In this case no filtering
+is done, the reply packets can arrive via any route.
 
 If setting up a network where the client networks have limited routing, be careful
 about configuring the DHCP server. Dnsmasq, as DHCP server, will set the default route to the

src/arp.c

@@ -111,22 +111,26 @@ int find_mac(union mysockaddr *addr, unsigned char *mac, int lazy, time_t now)
 
  again:
   
-  /* If the database is less then INTERVAL old, look in there */
-  if (difftime(now, last) < INTERVAL)
+  /* If the database is less then INTERVAL old, look in there.
+     
+     If we're a child process, we always rely on the existing cache we
+     inherited from the parent, since we don't have a netlink socket.
+  */
+  if (difftime(now, last) < INTERVAL || daemon->pipe_to_parent != -1)
     {
       /* addr == NULL -> just make cache up-to-date */
       if (!addr)
 	return 0;
-
+	
       for (arp = arps; arp; arp = arp->next)
 	{
 	  if (addr->sa.sa_family != arp->family)
 	    continue;
-	    
+	  
 	  if (arp->family == AF_INET &&
 	      arp->addr.addr4.s_addr != addr->in.sin_addr.s_addr)
 	    continue;
-	    
+	  
 	  if (arp->family == AF_INET6 && 
 	      !IN6_ARE_ADDR_EQUAL(&arp->addr.addr6, &addr->in6.sin6_addr))
 	    continue;
@@ -141,6 +145,10 @@ int find_mac(union mysockaddr *addr, unsigned char *mac, int lazy, time_t now)
 	}
     }
 
+  /* Not in cache in child, no go. */
+  if (daemon->pipe_to_parent != -1)
+    return 0;
+  
   /* Not found, try the kernel */
   if (!updated)
      {

src/cache.c

@@ -938,15 +938,14 @@ int cache_recv_insert(time_t now, int fd)
 		if (newc)
 		  {
 		    newc->addr.cname.is_name_ptr = 0;
+		    newc->addr.cname.target.cache = crecp;
 		    
-		    if (!crecp)
-		      newc->addr.cname.target.cache = NULL;
-		    else
+		    if (crecp)
 		      {
 			next_uid(crecp);
-			newc->addr.cname.target.cache = crecp;
 			newc->addr.cname.uid = crecp->uid;
 		      }
+		    crecp = newc;
 		  }
 	      }
 	    else

src/dbus.c

@@ -530,8 +530,8 @@ static DBusMessage *dbus_add_lease(DBusMessage* message)
   union all_addr addr;
   time_t now = dnsmasq_time();
   unsigned char dhcp_chaddr[DHCP_CHADDR_MAX];
-
   DBusMessageIter iter, array_iter;
+
   if (!dbus_message_iter_init(message, &iter))
     return dbus_message_new_error(message, DBUS_ERROR_INVALID_ARGS,
 				  "Failed to initialize dbus message iter");
@@ -599,6 +599,10 @@ static DBusMessage *dbus_add_lease(DBusMessage* message)
 
   if (inet_pton(AF_INET, ipaddr, &addr.addr4))
     {
+      if (!daemon->dhcp)
+	return dbus_message_new_error(message, DBUS_ERROR_INVALID_ARGS,
+				      "DHCPv4 not configured");
+      
       if (ia_id != 0 || is_temporary)
 	return dbus_message_new_error(message, DBUS_ERROR_INVALID_ARGS,
 				      "ia_id and is_temporary must be zero for IPv4 lease");
@@ -609,16 +613,25 @@ static DBusMessage *dbus_add_lease(DBusMessage* message)
 #ifdef HAVE_DHCP6
   else if (inet_pton(AF_INET6, ipaddr, &addr.addr6))
     {
+      if (!daemon->doing_dhcp6)
+	return dbus_message_new_error(message, DBUS_ERROR_INVALID_ARGS,
+				      "DHCPv6 not configured");
+      
       if (!(lease = lease6_find_by_addr(&addr.addr6, 128, 0)))
 	lease = lease6_allocate(&addr.addr6,
 				is_temporary ? LEASE_TA : LEASE_NA);
-      lease_set_iaid(lease, ia_id);
+      if (lease)
+	lease_set_iaid(lease, ia_id);
     }
 #endif
   else
     return dbus_message_new_error_printf(message, DBUS_ERROR_INVALID_ARGS,
 					 "Invalid IP address '%s'", ipaddr);
-   
+
+  if (!lease) 
+    return dbus_message_new_error_printf(message, DBUS_ERROR_INVALID_ARGS,
+					 "unable to allocate lease for IP address '%s'", ipaddr);
+  
   hw_len = parse_hex((char*)hwaddr, dhcp_chaddr, DHCP_CHADDR_MAX, NULL, &hw_type);
   if (hw_len < 0)
     return dbus_message_new_error_printf(message, DBUS_ERROR_INVALID_ARGS,

src/dhcp-common.c

@@ -709,7 +709,8 @@ static const struct opttab_t {
   { "client-machine-id", 97, 0 },
   { "posix-timezone", 100, OT_NAME }, /* RFC 4833, Sec. 2 */
   { "tzdb-timezone", 101, OT_NAME }, /* RFC 4833, Sec. 2 */
-  { "ipv6-only", 108, 4 | OT_DEC },  /* RFC 8925 */ 
+  { "ipv6-only", 108, 4 | OT_DEC },  /* RFC 8925 */
+  { "captive-portal", 114, OT_NAME },  /* RFC 8910 */
   { "subnet-select", 118, OT_INTERNAL },
   { "domain-search", 119, OT_RFC1035_NAME },
   { "sip-server", 120, 0 },
@@ -751,6 +752,7 @@ static const struct opttab_t opttab6[] = {
   { "ntp-server", 56, 0 /* OT_ADDR_LIST | OT_RFC1035_NAME */ },
   { "bootfile-url", 59, OT_NAME },
   { "bootfile-param", 60, OT_CSTRING },
+  { "captive-portal", 103, OT_NAME },  /* RFC 8910 */
   { NULL, 0, 0 }
 };
 #endif

src/dhcp.c

@@ -142,7 +142,7 @@ void dhcp_packet(time_t now, int pxe_fd)
   struct iovec iov;
   ssize_t sz; 
   int iface_index = 0, unicast_dest = 0, is_inform = 0, loopback = 0;
-  int rcvd_iface_index;
+  int rcvd_iface_index, relay_index;
   struct in_addr iface_addr;
   struct iface_param parm;
   time_t recvtime = now;
@@ -302,10 +302,10 @@ void dhcp_packet(time_t now, int pxe_fd)
     unicast_dest = 1;
 #endif
   
-  if ((relay = relay_reply4((struct dhcp_packet *)daemon->dhcp_packet.iov_base, ifr.ifr_name)))
+  if ((relay_index = relay_reply4((struct dhcp_packet *)daemon->dhcp_packet.iov_base, (size_t)sz, ifr.ifr_name)))
     {
       /* Reply from server, using us as relay. */
-      rcvd_iface_index = relay->iface_index;
+      rcvd_iface_index = relay_index;
       if (!indextoname(daemon->dhcpfd, rcvd_iface_index, ifr.ifr_name))
 	return;
       is_relay_reply = 1; 
@@ -330,9 +330,12 @@ void dhcp_packet(time_t now, int pxe_fd)
 	if (tmp->name && (tmp->flags & INAME_4) && wildcard_match(tmp->name, ifr.ifr_name))
 	  return;
       
-      /* unlinked contexts/relays are marked by context->current == context */
+      /* unlinked contexts are marked by context->current == context */
       for (context = daemon->dhcp; context; context = context->next)
 	context->current = context;
+
+      for (relay = daemon->relay4; relay; relay = relay->next)
+	relay->matchcount = 0;
       
       parm.current = NULL;
       parm.ind = iface_index;
@@ -360,7 +363,7 @@ void dhcp_packet(time_t now, int pxe_fd)
       if (!iface_enumerate(AF_INET, &parm, (callback_t){.af_inet=complete_context}))
 	return;
 
-      relay_upstream4(iface_index, mess, (size_t)sz, unicast_dest);
+      relay_upstream4(iface_addr, iface_index, mess, (size_t)sz, unicast_dest);
        
       /* May have configured relay, but not DHCP server */
       if (!daemon->dhcp)
@@ -664,8 +667,19 @@ static int complete_context(struct in_addr local, int if_index, char *label,
     }
 
   for (relay = daemon->relay4; relay; relay = relay->next)
-    if (relay->local.addr4.s_addr == local.s_addr)
-      relay->iface_index = if_index;
+    if (!relay->split_mode && relay->local.addr4.s_addr == local.s_addr)
+      {
+	if (if_index == param->ind)
+	  relay->iface_index = if_index;
+	
+	/* More than one interface with the relay address breaks things. */
+	if (relay->matchcount++ == 1 && !relay->warned)
+	  {
+	    relay->warned = 1;
+	    inet_ntop(AF_INET, &local, daemon->addrbuff, ADDRSTRLEN);
+	    my_syslog(MS_DHCP | LOG_WARNING, _("DHCP relay address %s appears on more than one interface"), daemon->addrbuff);
+	  }
+      }
   
   return 1;
 }
@@ -682,7 +696,7 @@ struct dhcp_context *address_available(struct dhcp_context *context,
   struct dhcp_context *tmp;
 
   for (tmp = context; tmp; tmp = tmp->current)
-    if (taddr.s_addr == context->router.s_addr)
+    if (taddr.s_addr == tmp->router.s_addr)
       return NULL;
   
   for (tmp = context; tmp; tmp = tmp->current)

src/dhcp6.c

@@ -89,6 +89,7 @@ void dhcp6_init(void)
 void dhcp6_packet(time_t now)
 {
   struct dhcp_context *context;
+  struct dhcp_relay *relay;
   struct iface_param parm;
   struct cmsghdr *cmptr;
   struct msghdr msg;
@@ -231,7 +232,10 @@ void dhcp6_packet(time_t now)
 	    context->current = context;
 	    memset(&context->local6, 0, IN6ADDRSZ);
 	  }
-      
+
+      for (relay = daemon->relay6; relay; relay = relay->next)
+	relay->matchcount = 0;
+
       inet_pton(AF_INET6, ALL_RELAY_AGENTS_AND_SERVERS, &all_servers);
       if (IN6_ARE_ADDR_EQUAL(&dst_addr, &all_servers))
 	multicast_dest = 1;
@@ -452,7 +456,17 @@ static int complete_context6(struct in6_addr *local,  int prefix,
   if (match)
     for (relay = daemon->relay6; relay; relay = relay->next)
       if (IN6_ARE_ADDR_EQUAL(local, &relay->local.addr6))
-	relay->iface_index = if_index;
+	{
+	  relay->iface_index = if_index;
+
+	  /* More than one interface with the relay address breaks things. */
+	  if (relay->matchcount++ == 1 && !relay->warned)
+	    {
+	      relay->warned = 1;
+	      inet_ntop(AF_INET6, &local, daemon->addrbuff, ADDRSTRLEN);
+	      my_syslog(MS_DHCP | LOG_WARNING, _("DHCP relay address %s appears on more than one interface"), daemon->addrbuff);
+	    }
+	}
   
   return 1;
 }

src/dnsmasq.c

@@ -2033,9 +2033,24 @@ static void do_tcp_connection(struct listener *listener, time_t now, int slot)
   
   if (!option_bool(OPT_DEBUG))
     {
+      /* The code in edns0.c qthat decorates queries with the source MAC address depends
+	 on the code in arp.c, which populates a cache with the contents of the ARP table
+	 using netlink. Since the child process can't use netlink, we pre-populate
+	 the cache with the ARP table entry for our source here, including a negative entry
+	 if there is nothing for our address in the ARP table.
+	 
+	 When the edns0 code calls find_mac() in the child process, it will
+	 get the correct answer from the cache inherited from the parent
+	 without having to use netlink to consult the kernel ARP table.
+
+	 edns0_needs_mac() simply calls find_mac if any EDNS0 options
+	 which need a MAC address are enabled. */
+      
+      edns0_needs_mac(&tcp_addr, now);
+
       if (pipe(pipefd) == -1)
 	goto closeconandreturn; /* pipe failed */
-            
+
       if ((p = fork()) == -1)
 	{
 	  /* fork failed */

src/dnsmasq.h

@@ -1109,7 +1109,7 @@ struct ping_result {
 
 struct tftp_file {
   int refcount, fd;
-  off_t size;
+  off_t size, posn;
   dev_t dev;
   ino_t inode;
   char filename[];
@@ -1117,13 +1117,14 @@ struct tftp_file {
 
 struct tftp_transfer {
   int sockfd;
+  u16 block_hi, ackprev;
   time_t retransmit, start;
   unsigned int lastack, block, blocksize, windowsize, timeout, expansion;
   off_t offset;
   union mysockaddr peer;
   union all_addr source;
   int if_index;
-  unsigned char opt_blocksize, opt_transize, opt_windowsize, opt_timeout, netascii, carrylf, backoff;
+  unsigned char opt_blocksize, opt_transize, opt_windowsize, opt_timeout, netascii, carrylf, lastcarrylf, backoff;
   struct tftp_file *file;
   struct tftp_transfer *next;
 };
@@ -1144,11 +1145,12 @@ struct dhcp_relay {
   union {
     struct in_addr addr4;
     struct in6_addr addr6;
-  } local, server;
+  } local, server, uplink;
   char *interface; /* Allowable interface for replies from server, and dest for IPv6 multicast */
   int iface_index; /* working - interface in which requests arrived, for return */
   int port;        /* Port of relay we forward to. */
   int split_mode;  /* Split address allocation and relay address. */
+  int warned, matchcount;
 #ifdef HAVE_SCRIPT
   struct snoop_record {
     struct in6_addr client, prefix;
@@ -1276,7 +1278,7 @@ extern struct daemon {
   struct serverfd *sfds;
   struct irec *interfaces;
   struct listener *listeners;
-  struct server *srv_save; /* Used for resend on DoD */
+  void *srv_save;      /* Used for resend on DoD and tftp prefetch */
   size_t packet_len;       /*      "        "        */
   int    fd_save;          /*      "        "        */
   pid_t *tcp_pids;
@@ -1667,8 +1669,9 @@ size_t dhcp_reply(struct dhcp_context *context, char *iface_name, int int_index,
 		  time_t recvtime, struct in_addr leasequery_source);
 unsigned char *extended_hwaddr(int hwtype, int hwlen, unsigned char *hwaddr, 
 			       int clid_len, unsigned char *clid, int *len_out);
-void relay_upstream4(int iface_index, struct dhcp_packet *mess, size_t sz, int unicast);
-struct dhcp_relay *relay_reply4(struct dhcp_packet *mess, char *arrival_interface);
+void relay_upstream4(struct in_addr iface_addr, int iface_index,
+		     struct dhcp_packet *mess, size_t sz, int unicast);
+unsigned int relay_reply4(struct dhcp_packet *mess, size_t sz, char *arrival_interface);
 #endif
 
 /* dnsmasq.c */
@@ -1915,6 +1918,7 @@ unsigned char *find_pseudoheader(struct dns_header *header, size_t plen,
 size_t add_pseudoheader(struct dns_header *header, size_t plen, unsigned char *limit, 
 			int optno, unsigned char *opt, size_t optlen, int set_do, int replace);
 size_t add_do_bit(struct dns_header *header, size_t plen, unsigned char *limit);
+void edns0_needs_mac(union mysockaddr *addr, time_t now);
 size_t add_edns0_config(struct dns_header *header, size_t plen, unsigned char *limit, 
 			union mysockaddr *source, time_t now, int *cacheable);
 int check_source(struct dns_header *header, size_t plen, unsigned char *pseudoheader, union mysockaddr *peer);

src/edns0.c

@@ -265,6 +265,15 @@ static void encoder(unsigned char *in, char *out)
   out[3] = char64(in[2]);
 }
 
+/* This function needs to call find_mac if any option which requires a MAC address is enabled
+   and used below. If you add a new MAC consumer, modify this, otherwise your
+   new EDNS0 option won't work in TCP mode. */
+void edns0_needs_mac(union mysockaddr *addr, time_t now)
+{
+  if (option_bool(OPT_MAC_B64) || option_bool(OPT_MAC_HEX) || option_bool(OPT_ADD_MAC))
+    find_mac(addr, NULL, 0, now);
+}
+
 /* OPT_ADD_MAC = MAC is added (if available)
    OPT_ADD_MAC + OPT_STRIP_MAC = MAC is replaced, if not available, it is only removed
    OPT_STRIP_MAC = MAC is removed */
@@ -562,3 +571,4 @@ size_t add_edns0_config(struct dns_header *header, size_t plen, unsigned char *l
 
   return plen;
 }
+

src/forward.c

@@ -17,7 +17,7 @@
 #include "dnsmasq.h"
 
 static struct frec *get_new_frec(time_t now, struct server *serv, int force);
-static struct frec *lookup_frec(char *target, int class, int rrtype, int id, int flags, int flagmask);
+static struct frec *lookup_frec(time_t now, char *target, int class, int rrtype, int id, int flags, int flagmask);
 #ifdef HAVE_DNSSEC
 static int tcp_key_recurse(time_t now, int status, struct dns_header *header, size_t n, 
 			   int class, char *name, char *keyname, struct server *server, 
@@ -192,7 +192,7 @@ static void forward_query(int udpfd, union mysockaddr *udpaddr,
       old_reply = 1;
       fwd_flags = forward->flags;
     }
-  else if (gotname && (forward = lookup_frec(daemon->namebuff, (int)rrclass, (int)rrtype, -1, fwd_flags,
+  else if (gotname && (forward = lookup_frec(now, daemon->namebuff, (int)rrclass, (int)rrtype, -1, fwd_flags,
 					     FREC_CHECKING_DISABLED | FREC_AD_QUESTION | FREC_DO_QUESTION |
 					     FREC_HAS_PHEADER | FREC_DNSKEY_QUERY | FREC_DS_QUERY | FREC_NO_CACHE)))
     {
@@ -980,7 +980,7 @@ static void dnssec_validate(struct frec *forward, struct dns_header *header,
 	  unsigned int flags = STAT_ISEQUAL(status, STAT_NEED_KEY) ? FREC_DNSKEY_QUERY : FREC_DS_QUERY;
 	  struct frec *old;
 	  
-	  if ((old = lookup_frec(daemon->keyname, forward->class, -1, -1, flags, flags)))
+	  if ((old = lookup_frec(now, daemon->keyname, forward->class, -1, -1, flags, flags)))
 	    {
 	      /* This is tricky; it detects loops in the dependency
 		 graph for DNSSEC validation, say validating A requires DS B
@@ -1170,7 +1170,7 @@ void reply_query(int fd, time_t now)
   GETSHORT(rrtype, p); 
   GETSHORT(class, p);
 
-  if (!(forward = lookup_frec(daemon->namebuff, class, rrtype, ntohs(header->id), FREC_ANSWER, 0)))
+  if (!(forward = lookup_frec(now, daemon->namebuff, class, rrtype, ntohs(header->id), FREC_ANSWER, 0)))
     return;
 
   filter_servers(forward->sentto->arrayposn, F_SERVER, &first, &last);
@@ -3106,7 +3106,7 @@ static void free_frec(struct frec *f)
 /* Impose an absolute
    limit of 4*TIMEOUT before we wipe things (for random sockets).
    If force is set, always return a result, even if we have
-   to allocate above the limit, and don'y free any records.
+   to allocate above the limit, and don't free any records.
    This is set when allocating for DNSSEC to avoid cutting off
    the branch we are sitting on. */
 static struct frec *get_new_frec(time_t now, struct server *master, int force)
@@ -3128,36 +3128,40 @@ static struct frec *get_new_frec(time_t now, struct server *master, int force)
 	  /* Don't free DNSSEC sub-queries here, as we may end up with
 	     dangling references to them. They'll go when their "real" query 
 	     is freed. */
-	  if (!f->dependent && !force)
+	  if (!f->dependent)
 #endif
-	    {
-	      if (difftime(now, f->time) >= 4*TIMEOUT)
-		{
-		  daemon->metrics[METRIC_DNS_UNANSWERED_QUERY]++;
-		  free_frec(f);
-		  target = f;
-		}
-	      else if (!oldest || difftime(f->time, oldest->time) <= 0)
-		oldest = f;
-	    }
+	    if (!force)
+	      {
+		if (difftime(now, f->time) >= 4*TIMEOUT)
+		  {
+		    daemon->metrics[METRIC_DNS_UNANSWERED_QUERY]++;
+		    free_frec(f);
+		    target = f;
+		  }
+		else if (!oldest || difftime(f->time, oldest->time) <= 0)
+		  oldest = f;
+	      }
 	}
       
       if (f->sentto && ((int)difftime(now, f->time)) < TIMEOUT && server_samegroup(f->sentto, master))
 	count++;
     }
-
-  if (!force && count >= daemon->ftabsize)
-    {
-      query_full(now, master->domain);
-      return NULL;
-    }
   
-  if (!target && oldest && ((int)difftime(now, oldest->time)) >= TIMEOUT)
-    { 
-      /* can't find empty one, use oldest if there is one and it's older than timeout */
-      daemon->metrics[METRIC_DNS_UNANSWERED_QUERY]++;
-      free_frec(oldest);
-      target = oldest;
+  if (!force)
+    {
+      if (count >= daemon->ftabsize)
+	{
+	  query_full(now, master->domain);
+	  return NULL;
+	}
+      
+      if (!target && oldest && ((int)difftime(now, oldest->time)) >= TIMEOUT)
+	{ 
+	  /* can't find empty one, use oldest if there is one and it's older than timeout */
+	  daemon->metrics[METRIC_DNS_UNANSWERED_QUERY]++;
+	  free_frec(oldest);
+	  target = oldest;
+	}      
     }
   
   if (!target && (target = (struct frec *)whine_malloc(sizeof(struct frec))))
@@ -3192,7 +3196,7 @@ static void query_full(time_t now, char *domain)
     }
 }
 
-static struct frec *lookup_frec(char *target, int class, int rrtype, int id, int flags, int flagmask)
+static struct frec *lookup_frec(time_t now, char *target, int class, int rrtype, int id, int flags, int flagmask)
 {
   struct frec *f;
   struct dns_header *header;
@@ -3242,7 +3246,14 @@ static struct frec *lookup_frec(char *target, int class, int rrtype, int id, int
 	    
 	    continue;
 	  }
-		
+
+	/* frecs older than this will get garbage-collected in
+	   get_new_frec(), so don't return them here, so we have
+	   consistent behaviour from an idle dnsmasq which
+	   is not calling get_new_frec() often. */
+	if (difftime(now, f->time) >= 4*TIMEOUT)
+	  return NULL;
+	
 	return f;
       }
   

src/option.c

@@ -4756,6 +4756,9 @@ static int one_opt(int option, char *arg, char *errstr, char *gen_err, int comma
 		    else
 		      three = two;
 		  }
+		else if (new->split_mode && inet_pton(AF_INET, three, &new->uplink))
+		  /* Third arg in split mode can be an address. */
+		  three = NULL;
 		
 		new->next = daemon->relay4;
 		daemon->relay4 = new;

src/rfc1035.c

@@ -516,7 +516,7 @@ int do_doctor(struct dns_header *header, size_t qlen, char *namebuff)
    Cache said SOA and return the difference in length between name and the name of the 
    SOA RR so we can look it up again.
 */
-static int find_soa(struct dns_header *header, size_t qlen, char *name, int *substring, unsigned long *ttlp, int no_cache, time_t now)
+static int find_soa(struct dns_header *header, size_t qlen, char *name, int *substring, unsigned long *ttlp, int cache, time_t now)
 {
   unsigned char *p, *psave;
   int qtype, qclass, rdlen;
@@ -556,7 +556,7 @@ static int find_soa(struct dns_header *header, size_t qlen, char *name, int *sub
 	    {
 	      int prefix = name_len - soa_len;
 	      
-	      if (!no_cache)
+	      if (cache)
 		{
 		  if (!(addr.rrblock.rrdata = blockdata_alloc(NULL, 0)))
 		    return 0;
@@ -568,12 +568,12 @@ static int find_soa(struct dns_header *header, size_t qlen, char *name, int *sub
 		{
 		  if (!extract_name(header, qlen, &p, daemon->workspacename, EXTR_NAME_EXTRACT, 0))
 		    {
-		      if (!no_cache)
+		      if (cache)
 			blockdata_free(addr.rrblock.rrdata);
 		      return 0;
 		    }
 		  
-		  if (!no_cache)
+		  if (cache)
 		    {
 		      len = to_wire(daemon->workspacename);
 		      if (!blockdata_expand(addr.rrblock.rrdata, addr.rrblock.datalen, daemon->workspacename, len))
@@ -588,13 +588,13 @@ static int find_soa(struct dns_header *header, size_t qlen, char *name, int *sub
 
 	      if (!CHECK_LEN(header, p, qlen, 20))
 		{
-		  if (!no_cache)
+		  if (cache)
 		    blockdata_free(addr.rrblock.rrdata);
 		  return 0;
 		}
 	      
 	      /* rest of RR */
-	      if (!no_cache)
+	      if (cache)
 		{
 		  int secflag = 0;
 
@@ -1079,7 +1079,17 @@ int extract_addresses(struct dns_header *header, size_t qlen, char *name, time_t
 	  
 	  if (insert && !option_bool(OPT_NO_NEG))
 	    {
-	      int substring, have_soa = find_soa(header, qlen, name, &substring, &ttl, no_cache_dnssec, now);
+	      /* The order of records going into the cache matters (see  cache_recv_insert()).
+		 The target of a CNAME must immediately follow the CNAME.
+		 (CNAME has already gone into the cache at this point)
+		 Here we call find_soa to get the ttl and substring, but
+		 we DON'T LET IT INSERT the SOA into the cache if our negative record is a CNAME target
+		 so that the SOA doesn't come before the CNAME target.
+
+		 We call find_soa() again after inserting the CNAME target to insert the SOA
+		 if necessary. */
+
+	      int substring, have_soa = find_soa(header, qlen, name, &substring, &ttl, cpp == NULL, now);
 	      
 	      if (have_soa || daemon->neg_ttl)
 		{		  
@@ -1101,6 +1111,10 @@ int extract_addresses(struct dns_header *header, size_t qlen, char *name, time_t
 		      next_uid(newc);
 		      cpp->addr.cname.target.cache = newc;
 		      cpp->addr.cname.uid = newc->uid;
+		    
+		      /* we didn't insert the SOA before a CNAME target above, do it now. */
+		      if (have_soa)
+			find_soa(header, qlen, name, NULL, NULL, 1, now);
 		    }
 		}
 	    }

src/rfc2131.c

@@ -3055,7 +3055,7 @@ static void apply_delay(u32 xid, time_t recvtime, struct dhcp_netid *netid)
     }
 }
 
-void relay_upstream4(int iface_index, struct dhcp_packet *mess, size_t sz, int unicast)
+void relay_upstream4(struct in_addr iface_addr, int iface_index, struct dhcp_packet *mess, size_t sz, int unicast)
 {
   struct in_addr giaddr = mess->giaddr;
   u8 hops = mess->hops;
@@ -3063,141 +3063,159 @@ void relay_upstream4(int iface_index, struct dhcp_packet *mess, size_t sz, int u
   size_t orig_sz = sz;
   unsigned char *endopt = NULL;
     
-  if (mess->op != BOOTREQUEST)
+  if (mess->op != BOOTREQUEST || (mess->hops++) > 20)
     return;
-
+  
   for (relay = daemon->relay4; relay; relay = relay->next)
-    if (relay->iface_index != 0 && relay->iface_index == iface_index)
-      {
-	union mysockaddr to;
-	union all_addr from;
-	struct ifreq ifr;
-	
-	/* restore orig packet */
-	mess->hops = hops;
-	mess->giaddr = giaddr;
-	if (endopt)
-	  *endopt = OPTION_END;
-	sz = orig_sz;
-	
-	if ((mess->hops++) > 20)
-	  continue;
+    {
+      union mysockaddr to;
+      union all_addr from;
+      struct ifreq ifr;
 
-	if (relay->interface)
-	  {
-	    safe_strncpy(ifr.ifr_name, relay->interface, IF_NAMESIZE);
-	    ifr.ifr_addr.sa_family = AF_INET;
-	  }
-	
-	if (!relay->split_mode)
-	  {
-	    /* already gatewayed ? */
-	    if (giaddr.s_addr)
-	      {
-		/* if so check if by us, to stomp on loops. */
-		if (giaddr.s_addr == relay->local.addr4.s_addr)
-		  continue;
-	      }
+      /* restore orig packet */
+      mess->giaddr = giaddr;
+      if (endopt)
+	*endopt = OPTION_END;
+      sz = orig_sz;
 
-	    /* plug in our address */
-	    from.addr4 = mess->giaddr = relay->local.addr4;
-	  }
-	else
-	  {
-	    /* Split mode. We put our address on the server-facing interface
-	       into giaddr for the server to talk back to us on.
-
-	       Our address on client-facing interface goes into agent-id
-	       subnet-selector subopt, so that the server allocates the correct address. */
-	    
-	    /* get our address on the server-facing interface. */
-	    if (ioctl(daemon->dhcpfd, SIOCGIFADDR, &ifr) == -1)
-	      continue;
-	    
-	    /* already gatewayed ? */
-	    if (giaddr.s_addr)
-	      {
-		/* if so check if by us, to stomp on loops. */
-		if (giaddr.s_addr == ((struct sockaddr_in *) &ifr.ifr_addr)->sin_addr.s_addr)
-		  continue;
-	      }
-	    
-	    /* giaddr is our address on the outgoing interface in split mode. */
-	    from.addr4 = mess->giaddr = ((struct sockaddr_in *) &ifr.ifr_addr)->sin_addr;
-	    
-	    if (!endopt)
-	      {
-		/* Add an RFC3026 relay agent information option (2 bytes) at the very end of the options.
-		   Said option to contain a RFC 3527 link selection sub option (6 bytes) and
-		   RFC 5017 serverid-override option (6 bytes) and RFC5010 (3 bytes).
-		   New END option is a 18th byte, so we need 18 bytes free.
-		   We only need to do this once, and poke the address into the same place each time. */
-		
-		if (!(endopt = option_find1((&mess->options[0] + sizeof(u32)), ((unsigned char *)mess) + sz, OPTION_END, 0)) ||
-		    (endopt + 18 > (unsigned char *)(mess + 1)))
-		  continue;
-		
-		endopt[1] = 15; /* length */
-		endopt[2] = SUBOPT_SUBNET_SELECT;
-		endopt[3] = 4; /* length */
-		endopt[8] = SUBOPT_SERVER_OR;
-		endopt[9] = 4;
-		endopt[14] = SUBOPT_FLAGS;
-		endopt[15] = 1; /* length */
-		endopt[17] = OPTION_END;
-		sz = (endopt - (unsigned char *)mess) + 18;
-	      }
-	    
-	    /* IP address is already in network byte order */
-	    memcpy(&endopt[4], &relay->local.addr4.s_addr, INADDRSZ);
-	    memcpy(&endopt[10], &relay->local.addr4.s_addr, INADDRSZ);
-	    endopt[16] = unicast ? 0x80 : 0x00;
-	    endopt[0] = OPTION_AGENT_ID;
-	  }
-	
-	to.sa.sa_family = AF_INET;
-	to.in.sin_addr = relay->server.addr4;
-	to.in.sin_port = htons(relay->port);
-#ifdef HAVE_SOCKADDR_SA_LEN
-	to.in.sin_len = sizeof(struct sockaddr_in);
-#endif
-	
-	/* Broadcasting to server. */
-	if (relay->server.addr4.s_addr == 0)
-	  {
-	    if (!relay->interface || strchr(relay->interface, '*') ||
-		ioctl(daemon->dhcpfd, SIOCGIFBRDADDR, &ifr) == -1)
-	      {
-		my_syslog(MS_DHCP | LOG_ERR, _("Cannot broadcast DHCP relay via interface %s"), relay->interface);
-		continue;
-	      }
-	    
-	    to.in.sin_addr = ((struct sockaddr_in *) &ifr.ifr_addr)->sin_addr;
-	  }
-	
-#ifdef HAVE_DUMPFILE
+      if (relay->interface)
 	{
-	  union mysockaddr fromsock;
-	  fromsock.in.sin_port = htons(daemon->dhcp_server_port);
-	  fromsock.in.sin_addr = from.addr4;
-	  fromsock.sa.sa_family = AF_INET;
-
-	  dump_packet_udp(DUMP_DHCP, (void *)mess, sz, &fromsock, &to, -1);
+	  safe_strncpy(ifr.ifr_name, relay->interface, IF_NAMESIZE);
+	  ifr.ifr_addr.sa_family = AF_INET;
 	}
-#endif
+      
+      if (!relay->split_mode && relay->iface_index && relay->iface_index == iface_index)
+	{
+	  /* already gatewayed ? */
+	  if (giaddr.s_addr)
+	    {
+	      /* if so check if by us, to stomp on loops. */
+	      if (giaddr.s_addr == relay->local.addr4.s_addr)
+		continue;
+	    }
+	  else
+	    /* plug in our address */
+	    mess->giaddr = relay->local.addr4;
+	  
+	  from.addr4 = relay->local.addr4;
+	}
+      else if (relay->split_mode && relay->local.addr4.s_addr == iface_addr.s_addr)
+	{
+	  /* Split mode. We put our address on the server-facing interface
+	     or a directly specified third address into giaddr for the server to talk back to us on.
+	     
+	     Our address on client-facing interface goes into agent-id subnet-selector subopt,
+	     so that the server allocates the correct address. We also send a
+	     remote-id with the interface on which the request arrived,
+	     so that we can send the reply back the same way. */
+	  unsigned int net_index = htonl(iface_index);
+	  
+	  if (relay->interface)
+	    {
+	      /* get our address on the server-facing interface. */
+	      if (ioctl(daemon->dhcpfd, SIOCGIFADDR, &ifr) == -1)
+		{
+		  my_syslog(MS_DHCP | LOG_ERR, _("Cannot send to server via interface %s: %s"), relay->interface, strerror(errno));
+		  continue;
+		}
+	      
+	      relay->uplink.addr4 = ((struct sockaddr_in *) &ifr.ifr_addr)->sin_addr;
+	    }
+	  
+	  /* already gatewayed ? */
+	  if (giaddr.s_addr)
+	    {
+	      /* if so check if by us, to stomp on loops. */
+	      if (giaddr.s_addr == relay->uplink.addr4.s_addr)
+		continue;
+	    }
+	  else
+	    {
+	      /* giaddr is our address on the outgoing interface in split mode. */
+	      mess->giaddr = relay->uplink.addr4;
+	      
+	      if (!endopt)
+		{
+		  /* Add an RFC3026 relay agent information option (2 bytes) at the very end of the options.
+		     Said option to contain a RFC 3527 link selection sub option (6 bytes) and
+		     RFC 5017 serverid-override option (6 bytes) and RFC5010 flags (3 bytes) and
+		     an RFC3046 remote-id which holds an interface index (6 bytes)
+		     
+		     New END option is a 24th byte, so we need 24 bytes free.
+		     We only need to do this once, and poke the address/interface/flags into the same place each time. */
+		  
+		  if (!(endopt = option_find1((&mess->options[0] + sizeof(u32)), ((unsigned char *)mess) + sz, OPTION_END, 0)) ||
+		      (endopt + 24 > (unsigned char *)(mess + 1)))
+		    continue;
+		  
+		  endopt[1] = 21; /* length */
+		  endopt[2] = SUBOPT_SUBNET_SELECT;
+		  endopt[3] = 4; /* length */
+		  endopt[8] = SUBOPT_SERVER_OR;
+		  endopt[9] = 4;
+		  endopt[14] = SUBOPT_FLAGS;
+		  endopt[15] = 1; /* length */
+		  endopt[17] = SUBOPT_REMOTE_ID;
+		  endopt[18] = 4; /* length */
+		  endopt[23] = OPTION_END;
+		  sz = (endopt - (unsigned char *)mess) + 24;
+		}
+	      
+	      /* IP address is already in network byte order */
+	      memcpy(&endopt[4], &relay->local.addr4.s_addr, INADDRSZ);
+	      memcpy(&endopt[10], &relay->local.addr4.s_addr, INADDRSZ);
+	      endopt[16] = unicast ? 0x80 : 0x00;
+	      memcpy(&endopt[19], &net_index, 4);
+	      endopt[0] = OPTION_AGENT_ID;
+	    }
+	  
+	  from.addr4 = relay->uplink.addr4;
+	}
+      else
+	continue;
 	
-	 send_from(daemon->dhcpfd, 0, (char *)mess, sz, &to, &from, 0);
-	 
-	 if (option_bool(OPT_LOG_OPTS))
-	   {
-	     inet_ntop(AF_INET, &relay->local, daemon->addrbuff, ADDRSTRLEN);
-	     if (relay->server.addr4.s_addr == 0)
-	       snprintf(daemon->dhcp_buff2, DHCP_BUFF_SZ, _("broadcast via %s"), relay->interface);
-	     else
-	       inet_ntop(AF_INET, &relay->server.addr4, daemon->dhcp_buff2, DHCP_BUFF_SZ);
-	     my_syslog(MS_DHCP | LOG_INFO, _("DHCP relay at %s -> %s"), daemon->addrbuff, daemon->dhcp_buff2);
-	   }
+      to.sa.sa_family = AF_INET;
+      to.in.sin_addr = relay->server.addr4;
+      to.in.sin_port = htons(relay->port);
+#ifdef HAVE_SOCKADDR_SA_LEN
+      to.in.sin_len = sizeof(struct sockaddr_in);
+#endif
+      
+      /* Broadcasting to server. */
+      if (relay->server.addr4.s_addr == 0)
+	{
+	  if (ioctl(daemon->dhcpfd, SIOCGIFBRDADDR, &ifr) == -1)
+	    {
+	      my_syslog(MS_DHCP | LOG_ERR, _("Cannot broadcast DHCP relay via interface %s: %s"), relay->interface, strerror(errno));
+	      continue;
+	    }
+	  
+	  to.in.sin_addr = ((struct sockaddr_in *) &ifr.ifr_addr)->sin_addr;
+	}
+      
+#ifdef HAVE_DUMPFILE
+      {
+	union mysockaddr fromsock;
+	fromsock.in.sin_port = htons(daemon->dhcp_server_port);
+	fromsock.in.sin_addr = from.addr4;
+	fromsock.sa.sa_family = AF_INET;
+	
+	dump_packet_udp(DUMP_DHCP, (void *)mess, sz, &fromsock, &to, -1);
       }
+#endif
+      
+      send_from(daemon->dhcpfd, 0, (char *)mess, sz, &to, &from, 0);
+      
+      if (option_bool(OPT_LOG_OPTS))
+	{
+	  inet_ntop(AF_INET, &relay->local, daemon->addrbuff, ADDRSTRLEN);
+	  if (relay->server.addr4.s_addr == 0)
+	    snprintf(daemon->dhcp_buff2, DHCP_BUFF_SZ, _("broadcast via %s"), relay->interface);
+	  else
+	    inet_ntop(AF_INET, &relay->server.addr4, daemon->dhcp_buff2, DHCP_BUFF_SZ);
+	  my_syslog(MS_DHCP | LOG_INFO, _("DHCP relay at %s -> %s"), daemon->addrbuff, daemon->dhcp_buff2);
+	}
+    }
   
   /* restore in case of a local reply. */
   mess->hops = hops;
@@ -3206,35 +3224,41 @@ void relay_upstream4(int iface_index, struct dhcp_packet *mess, size_t sz, int u
     *endopt = OPTION_END;
 }
 
-struct dhcp_relay *relay_reply4(struct dhcp_packet *mess, char *arrival_interface)
+unsigned int relay_reply4(struct dhcp_packet *mess, size_t sz, char *arrival_interface)
 {
   struct dhcp_relay *relay;
-
+    
   if (mess->giaddr.s_addr == 0 || mess->op != BOOTREPLY)
-    return NULL;
+    return 0;
 
   for (relay = daemon->relay4; relay; relay = relay->next)
     {
+      unsigned int return_iface = 0;
+
       if (relay->split_mode)
 	{
-	  struct ifreq ifr;
-
-	  safe_strncpy(ifr.ifr_name, arrival_interface, IF_NAMESIZE);
-	  ifr.ifr_addr.sa_family = AF_INET;
-
+	  unsigned char *opt, *sopt;
+	  
 	  /* giaddr is our address on the returning interface in split mode. */
-	  if (ioctl(daemon->dhcpfd, SIOCGIFADDR, &ifr) == -1 ||
-	      mess->giaddr.s_addr != ((struct sockaddr_in *) &ifr.ifr_addr)->sin_addr.s_addr)
-	    continue;
+	  if (mess->giaddr.s_addr == relay->uplink.addr4.s_addr &&
+	      (opt = option_find(mess, sz, OPTION_AGENT_ID, 1)))
+	    {
+	      if ((sopt = option_find1(option_ptr(opt, 0), option_ptr(opt, option_len(opt)), SUBOPT_REMOTE_ID, sizeof(unsigned int))))
+		return_iface = option_uint(sopt, 0, sizeof(unsigned int));
+
+	      /* delete agent info before return RFC 3046 para 2.1 */
+	      *opt = OPTION_END;
+	      memset(opt + 1, 0, option_len(opt) + 2);
+	    }
 	}
-      else if (mess->giaddr.s_addr != relay->local.addr4.s_addr)
-	continue;
+      else if (mess->giaddr.s_addr == relay->local.addr4.s_addr)
+	return_iface = relay->iface_index;
       
-      if (!relay->interface || wildcard_match(relay->interface, arrival_interface))
-	return relay->iface_index != 0 ? relay : NULL;
+      if (return_iface && (!relay->interface || wildcard_match(relay->interface, arrival_interface)))
+	return return_iface;
     }
   
-  return NULL;	 
+  return 0;	 
 }     
 
 

src/tftp.c

@@ -18,12 +18,12 @@
 
 #ifdef HAVE_TFTP
 
-static void handle_tftp(time_t now, struct tftp_transfer *transfer, ssize_t len);
-static struct tftp_file *check_tftp_fileperm(ssize_t *len, char *prefix, char *client);
+static void handle_tftp(char *packet, time_t now, struct tftp_transfer *transfer, ssize_t len);
+static struct tftp_file *check_tftp_fileperm(char *packet, ssize_t *len, char *prefix, char *client);
 static void free_transfer(struct tftp_transfer *transfer);
 static ssize_t tftp_err(int err, char *packet, char *message, char *file, char *arg2);
 static ssize_t tftp_err_oops(char *packet, const char *file);
-static ssize_t get_block(char *packet, struct tftp_transfer *transfer);
+static ssize_t get_block(struct tftp_transfer *transfer);
 static char *next(char **p, char *end);
 static void sanitise(char *buf);
 
@@ -41,10 +41,9 @@ static void sanitise(char *buf);
 #define ERR_ILL    4
 #define ERR_TID    5
 
-static void tftp_request(struct listener *listen, time_t now)
+static void tftp_request(char *packet, ssize_t plen, struct listener *listen, time_t now)
 {
   ssize_t len;
-  char *packet = daemon->packet;
   char *filename, *mode, *p, *end;
   union mysockaddr addr, peer;
   struct msghdr msg;
@@ -87,12 +86,10 @@ static void tftp_request(struct listener *listen, time_t now)
   msg.msg_iov = &iov;
   msg.msg_iovlen = 1;
 
+  /* packet buff is DNS name workspace. */
   iov.iov_base = packet;
-  iov.iov_len = daemon->packet_buff_sz;
-
-  /* we overwrote the buffer... */
-  daemon->srv_save = NULL;
-
+  iov.iov_len = plen;
+  
   if ((len = recvmsg(listen->tftpfd, &msg, 0)) < 2)
     return;
 
@@ -267,7 +264,7 @@ static void tftp_request(struct listener *listen, time_t now)
 		}
 	      else
 		{
-		  handle_tftp(now, transfer, len);
+		  handle_tftp(packet, now, transfer, len);
 		  return;
 		}
 	    }
@@ -326,6 +323,8 @@ static void tftp_request(struct listener *listen, time_t now)
   transfer->start = now;
   transfer->backoff = 1;
   transfer->block = 1;
+  transfer->ackprev = 0;
+  transfer->block_hi = 0;
   transfer->blocksize = 512;
   transfer->windowsize = 1;
   
@@ -499,17 +498,21 @@ static void tftp_request(struct listener *listen, time_t now)
 	  strncat(daemon->namebuff, filename, (MAXDNAME-1) - strlen(daemon->namebuff));
 	  
 	  /* check permissions and open file */
-	  if ((transfer->file = check_tftp_fileperm(&len, prefix, daemon->addrbuff)))
+	  if ((transfer->file = check_tftp_fileperm(packet, &len, prefix, daemon->addrbuff)))
 	    {
 	      transfer->lastack = transfer->block;
 	      transfer->retransmit = now + transfer->timeout;
 	      /* This packet is may be the first data packet, but only if windowsize == 1
 		 To get windowsize greater then one requires an option negotiation,
 		 in which case this packet is the OACK. */
-	      if ((len = get_block(packet, transfer)) == -1)
+	      if ((len = get_block(transfer)) == -1)
 		len = tftp_err_oops(packet, daemon->namebuff);
 	      else
-		is_err = 0;
+		{
+		  is_err = 0;
+		  /* get_block put the packet to send in a different buffer. */
+		  packet = daemon->packet;
+		}
 	    }
 	}
     }
@@ -532,9 +535,9 @@ static void tftp_request(struct listener *listen, time_t now)
     }
 }
  
-static struct tftp_file *check_tftp_fileperm(ssize_t *len, char *prefix, char *client)
+static struct tftp_file *check_tftp_fileperm(char *packet, ssize_t *len, char *prefix, char *client)
 {
-  char *packet = daemon->packet, *namebuff = daemon->namebuff;
+  char *namebuff = daemon->namebuff;
   struct tftp_file *file;
   struct tftp_transfer *t;
   uid_t uid = geteuid();
@@ -597,6 +600,7 @@ static struct tftp_file *check_tftp_fileperm(ssize_t *len, char *prefix, char *c
   file->size = statbuf.st_size;
   file->dev = statbuf.st_dev;
   file->inode = statbuf.st_ino;
+  file->posn = 0;
   file->refcount = 1;
   strcpy(file->filename, namebuff);
   return file;
@@ -616,12 +620,15 @@ static struct tftp_file *check_tftp_fileperm(ssize_t *len, char *prefix, char *c
 
 void check_tftp_listeners(time_t now)
 {
+  /* Use workspace to receive (small) request/ACK, to avoid overwriting precomputed reply */
+  char *packet = daemon->workspacename;
+  ssize_t plen = MAXDNAME * 2;
   struct listener *listener;
   struct tftp_transfer *transfer, *tmp, **up;
   
   for (listener = daemon->listeners; listener; listener = listener->next)
     if (listener->tftpfd != -1 && poll_check(listener->tftpfd, POLLIN))
-      tftp_request(listener, now);
+      tftp_request(packet, plen, listener, now);
     
   /* In single port mode, all packets come via port 69 and tftp_request() */
   if (!option_bool(OPT_SINGLE_PORT))
@@ -632,26 +639,23 @@ void check_tftp_listeners(time_t now)
 	  socklen_t addr_len = sizeof(union mysockaddr);
 	  ssize_t len;
 	  
-	  /* we overwrote the buffer... */
-	  daemon->srv_save = NULL;
-
-	  if ((len = recvfrom(transfer->sockfd, daemon->packet, daemon->packet_buff_sz, 0, &peer.sa, &addr_len)) > 0)
+	  if ((len = recvfrom(transfer->sockfd, packet, plen, 0, &peer.sa, &addr_len)) > 0)
 	    {
 #ifdef HAVE_DUMPFILE
-	      dump_packet_udp(DUMP_TFTP, (void *)daemon->packet, len, (union mysockaddr *)&peer, NULL, transfer->sockfd);
+	      dump_packet_udp(DUMP_TFTP, (void *)packet, len, (union mysockaddr *)&peer, NULL, transfer->sockfd);
 #endif	      
 
 	      if (sockaddr_isequal(&peer, &transfer->peer)) 
-		handle_tftp(now, transfer, len);
+		handle_tftp(packet, now, transfer, len);
 	      else
 		{
 		  /* Wrong source address. See rfc1350 para 4. */
 		  prettyprint_addr(&peer, daemon->addrbuff);
-		  len = tftp_err(ERR_TID, daemon->packet, _("ignoring packet from %s (TID mismatch)"), daemon->addrbuff, NULL);
-		  while(retry_send(sendto(transfer->sockfd, daemon->packet, len, 0, &peer.sa, sa_len(&peer))));
+		  len = tftp_err(ERR_TID, packet, _("ignoring packet from %s (TID mismatch)"), daemon->addrbuff, NULL);
+		  while(retry_send(sendto(transfer->sockfd, packet, len, 0, &peer.sa, sa_len(&peer))));
 
 #ifdef HAVE_DUMPFILE
-		  dump_packet_udp(DUMP_TFTP, (void *)daemon->packet, len, NULL, (union mysockaddr *)&peer, transfer->sockfd);
+		  dump_packet_udp(DUMP_TFTP, (void *)packet, len, NULL, (union mysockaddr *)&peer, transfer->sockfd);
 #endif
 		}
 	    }
@@ -671,7 +675,7 @@ void check_tftp_listeners(time_t now)
 	  endcon = 1;
 	  /* don't complain about timeout when we're awaiting the last
 	     ACK, some clients never send it */
-	  if (get_block(daemon->packet, transfer) > 0)
+	  if (get_block(transfer) > 0)
 	    error = timeout = 1;
 	}
       else if (difftime(now, transfer->retransmit) >= 0.0)
@@ -680,41 +684,43 @@ void check_tftp_listeners(time_t now)
 	     bumps transfer->lastack and trips the retransmit timer so that we send the next block(s)
 	     here. */
 	  ssize_t len;
+	  unsigned int i, winsize;
 	  
 	  transfer->retransmit += transfer->timeout + (1<<(transfer->backoff/2));
 	  transfer->backoff++;
 	  transfer->block = transfer->lastack;
 	  
-	  if ((len = get_block(daemon->packet, transfer)) == 0)
-	    endcon = 1; /* got last ACK */
-	  else
+	  /* send a window'a worth of blocks unless we're retransmitting OACK */
+	  winsize = transfer->block ? transfer->windowsize : 1;
+	  
+	  for (i = 0; i < winsize; i++, transfer->block++)
 	    {
-	      /* send a window'a worth of blocks unless we're retransmitting OACK */
-	      unsigned int i, winsize = transfer->block ? transfer->windowsize : 1;
-	      
-	      for (i = 0; i < winsize && !endcon; i++, transfer->block++)
+	      if ((len = get_block(transfer)) == 0)
 		{
-		  if (i != 0)
-		    len = get_block(daemon->packet, transfer);
+		  if (i == 0)
+		    endcon = 1; /* got last ACK */
 
-		  if (len == 0)
-		    break;
-		  
-		  if (len == -1)
-		    {
-		      len = tftp_err_oops(daemon->packet, transfer->file->filename);
-		      endcon = error = 1;
-		    }
-		  
-		  send_from(transfer->sockfd, !option_bool(OPT_SINGLE_PORT), daemon->packet, len,
-			    &transfer->peer, &transfer->source, transfer->if_index);
-#ifdef HAVE_DUMPFILE
-		  dump_packet_udp(DUMP_TFTP, (void *)daemon->packet, len, NULL, (union mysockaddr *)&transfer->peer, transfer->sockfd);
-#endif
+		  break;
 		}
-	    }
-	}
 	      
+	      if (len == -1)
+		{
+		  len = tftp_err_oops(daemon->packet, transfer->file->filename);
+		  endcon = error = 1;
+		}
+	      
+	      send_from(transfer->sockfd, !option_bool(OPT_SINGLE_PORT), daemon->packet, len,
+			&transfer->peer, &transfer->source, transfer->if_index);
+#ifdef HAVE_DUMPFILE
+	      dump_packet_udp(DUMP_TFTP, (void *)daemon->packet, len, NULL, (union mysockaddr *)&transfer->peer, transfer->sockfd);
+#endif
+	    }
+	  
+	  /* prefetch the block we'll probably need when we get an ACK. */
+	  if (!endcon)
+	    get_block(transfer);
+	}
+		      
       if (endcon)
 	{
 	  strcpy(daemon->namebuff, transfer->file->filename);
@@ -743,23 +749,36 @@ void check_tftp_listeners(time_t now)
     }
 }
 
-/* packet in daemon->packet as this is called. */
-static void handle_tftp(time_t now, struct tftp_transfer *transfer, ssize_t len)
+static void handle_tftp(char *packet, time_t now, struct tftp_transfer *transfer, ssize_t len)
 {
   struct ack {
     unsigned short op, block;
-  } *mess = (struct ack *)daemon->packet;
+  } *mess = (struct ack *)packet;
   
   if (len >= (ssize_t)sizeof(struct ack))
     {
       if (ntohs(mess->op) == OP_ACK)
 	{
-	  /* try and handle 16-bit blockno wrap-around */
-	  unsigned int block = (unsigned short)ntohs(mess->block);
-	  if (block < transfer->lastack)
-	    block |= transfer->block & 0xffff0000;
-	  
-	  /* ignore duplicate ACKs and ACKs for blocks we've not yet sent. */
+	  /* Handle 16-bit block number wrap-around. */
+	  u16 new = ntohs(mess->block);
+	  u32 block;
+
+	  /* If the last ack received was in the top quarter of a 64k block
+	     and this one is in the bottom quarter, assume it has wrapped.
+	     
+	     Since this is UDP and an old packet can in theory wander in we may also
+	     need to drop back to a previous segment. Such an ACK is ignored below;
+	     here we're just getting the most likely 32 bit value from the
+	     16 bits that we have. */
+	  if (new <= 0x4000 && transfer->ackprev >= 0xc000)
+	    transfer->block_hi++;
+	  else if (new >= 0xc000 && transfer->ackprev <= 0x4000 && transfer->block_hi != 0)
+	    transfer->block_hi--;
+
+	  transfer->ackprev = new;
+	  block = (((u32)transfer->block_hi) << 16) + (u32)new;
+
+	  /* Ignore duplicate ACKs and ACKs for blocks we've not yet sent. */
 	  if (block >= transfer->lastack &&
 	      block <= transfer->block) 
 	    {
@@ -767,20 +786,23 @@ static void handle_tftp(time_t now, struct tftp_transfer *transfer, ssize_t len)
 	      transfer->retransmit = transfer->start = now;
 	      transfer->backoff = 0;
 	      transfer->lastack = block + 1;
-	      
+
 	      /* We have no easy function from block no. to file offset when
 		 expanding line breaks in netascii mode, so we update the offset here
 		 as each block is acknowledged. This explains why the window size must be
 		 one for a netascii transfer; to avoid  the block no. doing anything
 		 other than incrementing by one. */
 	      if (transfer->netascii && block != 0)
-		transfer->offset += transfer->blocksize - transfer->expansion;
+		{
+		  transfer->offset +=  (off_t)transfer->blocksize - (off_t)transfer->expansion;
+		  transfer->lastcarrylf = transfer->carrylf;
+		}
 	    }
 	}
       else if (ntohs(mess->op) == OP_ERR)
 	{
-	  char *p = daemon->packet + sizeof(struct ack);
-	  char *end = daemon->packet + len;
+	  char *p = packet + sizeof(struct ack);
+	  char *end = packet + len;
 	  char *err = next(&p, end);
 	  
 	  (void)prettyprint_addr(&transfer->peer, daemon->addrbuff);
@@ -857,10 +879,6 @@ static ssize_t tftp_err(int err, char *packet, char *message, char *file, char *
   } *mess = (struct errmess *)packet;
   ssize_t len, ret = 4;
 
-  /* we overwrote the buffer... */
-  daemon->srv_save = NULL;
-
-  memset(packet, 0, daemon->packet_buff_sz);
   if (file)
     sanitise(file);
   
@@ -884,12 +902,10 @@ static ssize_t tftp_err_oops(char *packet, const char *file)
 }
 
 /* return -1 for error, zero for done. */
-static ssize_t get_block(char *packet, struct tftp_transfer *transfer)
+static ssize_t get_block(struct tftp_transfer *transfer)
 {
-  memset(packet, 0, daemon->packet_buff_sz);
-
-  /* we overwrote the buffer... */
-  daemon->srv_save = NULL;
+  static off_t saved_offset = 0;
+  static ssize_t saved_len = 0;
 
   if (transfer->block == 0)
     {
@@ -898,8 +914,12 @@ static ssize_t get_block(char *packet, struct tftp_transfer *transfer)
       struct oackmess {
 	unsigned short op;
 	char data[];
-      } *mess = (struct oackmess *)packet;
-      
+      } *mess = (struct oackmess *)daemon->packet;
+
+      /* we overwrote the buffer... */
+      daemon->srv_save = NULL;
+      memset(daemon->packet, 0, daemon->packet_buff_sz);
+            
       p = mess->data;
       mess->op = htons(OP_OACK);
       if (transfer->opt_blocksize)
@@ -923,7 +943,7 @@ static ssize_t get_block(char *packet, struct tftp_transfer *transfer)
 	  p += (sprintf(p, "%u", (unsigned int)transfer->windowsize) + 1);
 	}
  
-      return p - packet;
+      return p - daemon->packet;
     }
   else
     {
@@ -931,44 +951,58 @@ static ssize_t get_block(char *packet, struct tftp_transfer *transfer)
       struct datamess {
 	unsigned short op, block;
 	unsigned char data[];
-      } *mess = (struct datamess *)packet;
+      } *mess = (struct datamess *)daemon->packet;
       
       size_t size;
       
       if (!transfer->netascii)
-	transfer->offset = (transfer->block - 1) * transfer->blocksize;
+	transfer->offset = (off_t)(transfer->block - 1) * (off_t)transfer->blocksize;
       
       if (transfer->offset > transfer->file->size)
 	return 0; /* finished */
-      
-      if ((size = transfer->file->size - transfer->offset) > transfer->blocksize)
-	size = transfer->blocksize;
+
+      /* We may have a prefetched block already in the buffer. */
+      if (daemon->srv_save == transfer && saved_offset == transfer->offset)
+	return saved_len;
+	
+      /* we overwrote the buffer... */
+      daemon->srv_save = NULL;
+
+      if ((size = transfer->file->size - transfer->offset) > (size_t)transfer->blocksize)
+	size = (size_t)transfer->blocksize;
       
       mess->op = htons(OP_DATA);
       mess->block = htons((unsigned short)(transfer->block));
 
-      if (size != 0 &&
-	  (lseek(transfer->file->fd, transfer->offset, SEEK_SET) == (off_t)-1 ||
-	   !read_write(transfer->file->fd, mess->data, size, RW_READ)))
-	return -1;
-      
+      if (size != 0)
+	{
+	  if (transfer->file->posn != transfer->offset &&
+	      lseek(transfer->file->fd, transfer->offset, SEEK_SET) == (off_t)-1)
+	    return -1;
+
+	  if (!read_write(transfer->file->fd, mess->data, size, RW_READ))
+	    return -1;
+
+	  transfer->file->posn = transfer->offset + size;
+	}
+
       /* Map '\n' to CR-LF in netascii mode */
       if (transfer->netascii)
 	{
 	  size_t i;
-	  int newcarrylf;
+	  	  
+	  /* Map '\n' to CR-LF in netascii mode */
+	  transfer->expansion = transfer->carrylf = 0;
 	  
-	  transfer->expansion = 0;
-	  
-	  for (i = 0, newcarrylf = 0; i < size; i++)
-	    if (mess->data[i] == '\n' && (i != 0 || !transfer->carrylf))
+	  for (i = 0; i < size; i++)
+	    if (mess->data[i] == '\n' && (i != 0 || !transfer->lastcarrylf))
 	      {
 		transfer->expansion++;
 
 		if (size != transfer->blocksize)
 		  size++; /* room in this block */
 		else  if (i == size - 1)
-		  newcarrylf = 1; /* don't expand LF again if it moves to the next block */
+		  transfer->carrylf = 1; /* don't expand LF again if it moves to the next block */
 		  
 		/* make space and insert CR */
 		memmove(&mess->data[i+1], &mess->data[i], size - (i + 1));
@@ -976,11 +1010,13 @@ static ssize_t get_block(char *packet, struct tftp_transfer *transfer)
 		
 		i++;
 	      }
-
-	  transfer->carrylf = newcarrylf;
 	}
 
-      return size + 4;
+      daemon->srv_save = transfer;
+      saved_offset = transfer->offset;
+      saved_len = size + 4;
+
+      return saved_len;
     }
 }