Advanced UDP Sockets

1
Advanced UDP Sockets

• Kwangwoon Univ,
• Department of Computer Science.
• M.Sc masters course.
• In-sik LEE (gohacker_at_kw.ac.kr)

2
Contents

• Receiving Flags, Destination IP address, and
  Interface index
• Datagram Truncation
• When to use UDP instead of TCP
• Adding Reliability to a UDP Application
• Binding Interface Addresses
• Concurrent UDP servers
• IPv6 Packet Information
• IPv6 Path MTU Control

3
Receiving Flags, Destination IP address, and
Interface Index(1)

• Historically sendmsg, recvmsg have been used.
• Will write a function recvfrom_flags.
• Returned msg_flags value.
• The destination address of the received datagram.
• The index of the interface on the datagram was
  received.
• Defined in unp.h
• Value of index is zero(0) -gt unknown value
• Destination address is zero (o.o.o.o) -gt unknown
  real value
• RFC 1122
• Valid only when a host booting , and does not yet
  known its IP address.

4
Receiving Flags, Destination IP address, and
Interface Index(2)
Recvform_flags functioncall recvmsg
5
Receiving Flags, Destination IP address, and
Interface Index(3)
Recvform_flags functioncall recvmsg
6
Receiving Flags, Destination IP address, and
Interface Index(4)
Recvform_flags functioncall recvmsg
7
Example (Destination IP and datagram truncated
Flag(1)
dg_echo function that call recvfrom_flag function
8
Example (Destination IP and datagram truncated
Flag(2)
dg_echo function that call recvfrom_flag function
9
Datagram Truncation

• When a UDP datagram arrives that is larger than
  the applications buffer, recvmsg sets the
  MSG_TRUNC flag in the msg_flags member of the
  msghdr structure.
• Three possible scenarios (other solution)
• Discard the excess bytes, and return the
  MSG_TRUNC flag to the application. Application
  call recvmsg to receive the flag
• BSD/OS
• Discard the excess bytes, do not tell the
  application.
• Solaris 2.5
• Keep the excess bytes and return them in
  subsequent read operations on the socket.

10
When to use UDP instead of TCP(1)

• Advantage of UDP
• Support broadcasting and multicasting.
• UDP must be used if the application uses
  broadcasting or multicasting.
• Require only two packets to exchange a request
  and a reply. (size of each is less than the
  minimum MTU between the two end-system, TCP
  require about 10 packets)
• Minimum transaction time for a UDP request-reply
  is RTT SPT (server processing time)
• TCP connection is used for the request-reply, the
  minimum transaction time is (2 x RTT) SPT,

11
When to use UDP instead of TCP(2)

• List of TCP that are not provided by UDP
• Acknowledgments, retransmission, duplicate
  detection, sequencing of packets reordered by the
  network.
• Windowed flow control
• Receiving TCP tell the sender hoe much buffer
  space, and the sender cannot exceed this.
• Slow start and congestion avoidance.
• There are exception to these rules.
• TFTP uses UDP for bulk data transfer.
• Simpler than TCP in bootstrap code (TCP4500,
  UDP800)
• NFS uses UDP for bulk data transfer.
• Some might claim it is really a request-reply
  application.

12
Adding reliability to a UDP application.(1)

• if use UDP for a request-reply application, then
  must add two features to our client.
• Timeout and retransmission to handle datagram
  that are discarded.
• Sequence number so the client can verify that a
  reply is for the appropriate request.
• Problem
• The RTT between a client and server can change
  rapidly with time, as network conditions change.
• Calculate the RTO (retransmission timeout)

13
Adding reliability to a UDP application.(2)

• Calculate RTO
• srtt smoothed RTT estimator
• rttvar smoothed mean deviation estimator
• delta difference between measured RTT and the
  current smoothed RTT estimator (srtt)
• g gain applied to the RTT estimator and equals
  1/8.
• h gain applied to the mean deviation estimator
  and equals ¼.
• When the retransmission timer expire, an
  exponential backoff (2,4,8,16)

14
Adding reliability to a UDP application.(3)

• Following three possible scenario when our
  retransmission timer expires.
• The request is lost
• The reply is lost
• The RTO is too small

15
Adding reliability to a UDP application.(4)
dg_send_recv function
16
Adding reliability to a UDP application.(5)
dg_send_recv function
17
Adding reliability to a UDP application.(6)
dg_send_recv function
18
Adding reliability to a UDP application.(7)
dg_send_recv function
19
Adding reliability to a UDP application.(8)
unprtt.h
20
Adding reliability to a UDP application.(9)
RTT_RTOCALC macro, rtt_minmax, rtt_init function
21
Adding reliability to a UDP application.(10)
rtt_ts, rtt_newpack, rtt_start function
22
Adding reliability to a UDP application.(11)
rtt_stop functionUpdate RTT estimators,
calculate new RTO
23
Adding reliability to a UDP application.(12)
rtt_timeout function exponential backoff
24
Binding Interface Addresses(1)
UDP server that binds all address
25
Binding Interface Addresses(2)
UDP server that binds all address
26
Binding Interface Addresses(3)
UDP server that binds all address
27
Binding Interface Addresses(4)
mydg_echo function
28
Binding Interface Addresses(4)
Execute udpserv03.c
29
Concurrent UDP Servers(1)

• The server waits for a client request, reads the
  request, processes the request, sends back the
  reply, and then waits for the next client
  request.
• TCP It is simple to just fork a new child (or
  create new thread) and let the child handle the
  new client.
• UDP Must deal with two different types of
  servers

30
Concurrent UDP Servers(2)
31
IPv6 Packet Information(1)

• IPv6 allows an application to specify up to five
  pieces of information for an outgoing datagram.
• Source IPv6 address
• Outgoing interface index
• Outgoing hop limit
• Next-hop address
• Outgoing traffic class
• Four similar pieces of information can be
  returned for a received packet (ancillary data
  with recvmsg)
• Destination IPv6 address
• Arriving interface index
• Arriving hop limit
• Arriving traffic class

32
IPv6 Packet Information(2)
33
IPv6 Packet Information(3)

• An in6_pktinfo structure contain either the
  source IPv6 address and outgoing interface index
  for an outgoing datagram or the destination IPv6
  address and arriving interface index for a
  received datagram

struct in6_pktinfo struct in6_addr
ipi6_addr / src/dst IPv6 address
/ int ipi6_ifindex
/ send/recv interface index /
34
IPv6 Packet Information(4)

• Outgoing and arriving interface
• No interface is ever assigned an index of 0.
• If the ipi6_ifindex value is 0, kernel will
  choose the outgoing interface.
• Source and destination IPv6 addresses
• The source IPv6 address is normally specified by
  calling bind. Source address together with the
  data may less overhead.
• When specifying the source IPv6 address ancillary
  data, if ipi6_addr member of the in6_pktinfo
  structure is IN6ADDR_ANY_INIT then,
• If and address is currently bound to the socket,
  Its use
• if no, kernel will choose the source address.

35
IPv6 Packet Information(5)

• Specifying the Next-Hop address
• The IPV6_NEXTHOP ancillary data specifies the
  next hop for the datagram as a socket address
  structure.
• In the cmsghdr structure containing this
  ancillary data.
• cmsg_level member is IPPROTO_IPV6, the cmsg_type
  member is IPV6_NEXTHOP, and the first byte of
  data is the first byte of the socket address
  structure.
• The node identified by that address must be a
  neighbor of the sending host.
• If that address equals the destination IPv6
  address of the datagram, then this is equivalent
  to the existing SO_DONTROUTE socket option.

36
IPv6 Packet Information(6)

• Specifying Receiving the traffic class
• IPV6_TCLASS ancillary data specifies the traffic
  class for the datagram.
• cmsg_type member will be IPV6_TCLASS. (diffserv)
• Include the ancillary data with that packet.
• Described in Section 27.7

37
IPv6 Path MTU Control(1)

• Purpose program may want to modify the path MTU
  discovery behavior. Follow four socket option.
• Sending with minimum MTU
• Packets are normally fragmented using the MTU of
  the outgoing interface or the path MTU, whichever
  is smaller.
• IPv6 defines a minimum MTU of 1,280 bytes.
• Dropped packets and delay while the MTU is being
  discovered.
• The use of the minimum MTU is controlled with the
  IPV6_USE_MIN_MTU socket option.
• Three define value
• -1 multicast (perform to unicast destination)
• O performs path MTU discovery to all
  destination
• 1 use the minimum MTU for all destination

38
IPv6 Path MTU Control(2)

• IPV6_USE_MIN_MTU can also be sent as ancillary
  data.
• cmsg_type is IPV6_USE_MIN_MTU.
• Receiving path MTU change indications
• Receive change notification is the path MTU, an
  application can enable the IPV6_RECVPATHMTU
  socket option.
• Anytime it change (in ancillary data cmsghdr
  structure)
• cmsg_type member will be IPV6_PATHMTU.

struct in6_mtuinfo struct sockaddr_in6
ip6m_addr / destination address /
unit32_t ipi6m_mtu /
path MTU in host byte order /
39
IPv6 Path MTU Control(3)

• Determining the Current path MTU
• If an application has not been keeping track with
  the IPV6_RECVPATHMTU option, it can determine the
  current path MTU of a connected socket with the
  IPV6_PATHMTU option.
• Returns an ipv6_mtuinfo structure containing the
  current path MTU.
• Avoiding fragmentation
• An application such as traceroute may not want
  this automatic fragmentation.
• The IPV6_DONTFRAG socket option is used to turn
  off automatic fragmentation.
• 0 permit
• 1 turn off

40
IPv6 Path MTU Control(4)

• IPV6_DONTFRAG can also be sent as ancillary data.
• cmsg_type member will be IPV6_DONTFRAG.