Transcript Remote Procedure Call Implementations

IP: putting it all together
Part 2
G53ACC
Chris Greenhalgh
1
Contents
Fragmentation
 Error reporting (ICMP)
 Auto-configuration
 Network Address Translation

2
Fragmentation
IP allows datagram sizes up to 64Kbytes
 Physical networks often only support
smaller frame types (Maximum
Transmission Unit, MTU):

– E.g. Ethernet 1500bytes, dialup PPP ~256bytes

 Single IP datagram may need to be
divided into “fragments” for transmission…
3
IP fragmentation

Each fragment is a (new) IP packet
–
–
–
–
Has IP header, original source & destination
Identification field same for each fragment
Fragment offset identifies what bit it is
“More Fragments” flag set in all but last
fragment
4
Fragmenting packets
May be done by sending host
 May be done by intermediate router:


May be prevented with IP “Do not
fragment” flag
–  ICMP fragmentation required response if a
router would have needed to fragment it
– Used by TCP to learn path MTU and avoid
fragmentation
5
Reassembling fragments

Done ONLY by the ultimate destination of
the packet
– After checking header checksum and
destination, but before any more processing

Maintains a pool of fragments
– Discarded after a time-out
– If all fragments of a datagram received the
datagram is reassembled and handled as before
6
Fragmentation and
reassembly issues

Lose one segment and you lose the whole
message
– Bad if segment loss is likely or
number of segments is large

E.g. NFS v.2 used UDP, v.3 uses TCP
– because block size 8K -> 32K
– many more segments!
=> higher effective packet loss rate with UDP
and more wasted bandwidth
7
Error reporting


IP includes Internet Control Message Protocol
(ICMP) RFC 792
ICMP messages sent in IP packets
– (i.e. same protocol level as UDP or TCP)
– IP protocol number 2

Not seen by applications - between hosts or
routers OSs only
– Error messages
– Informational messages (mostly superceded by DHCP)

NOTE: some may be dropped by firewalls to
avoid possible attacks e.g. denial of service (but
makes diagnosis of problems harder)
8
ICMP message types
9
ICMP Error messages (i)

Source Quech
– router to host, please slow down (buffer
overflow)

Time exceeded
– datagram discarded due to TTL=0 or lost
fragment
can be used to trace a route by gradually increasing
TTL and seeing which router it gets to before timing
out
 See commands: tracert (windows), traceroute (unix)10

ICMP error messages (ii)

Destination unreachable
– datagram discarded by router because host or network
not reachable
– Datagram discarded by host because UDP/TCP port not
in use

Redirect
– datagram sent to wrong next hop (gives alternative)

Fragmentation required
– if fragmentation not allowed but necessary

can be used to determine path MTU (maximum transmission
unit)
11
ICMP informational messages

Echo Request/Reply
– ICMP software sends Reply when receives Request


test computer accessible (e.g. ping)
Address mask request/reply
– allow host on booting to query local router for netmask
(see DHCP, later)

Gateway discovery
– allow host on booting to find default router (see DHCP)
12
Auto-configuration - low-level

ICMP address mask request/reply
– => netmask

Reverse ARP (RARP) RFC 903
– send Ethernet address and a server returns your
IP address

ICMP gateway discovery
– => default route
13
Auto-configuration - higherlevel (i)

Bootstrap Protocol (BOOTP)
RFC 951 and RFC 1542
– single BOOTP request
– BOOTP server replies with IP address, Router
IP address, server information
– requires server configuration for each machine
14
Auto-configuration - higher
level (ii)

Dynamic Host Configuration Protocol
(DHCP) RFC 1541
–
–
–
–
conceptually an extension of BOOTP
server can maintain pool of IP addresses
no configuration for a new machine
but IP address (and therefore domain names)
may change each time a machine is booted
15
Network Address Translation:
motivations

IP requires every machine to have a unique
IP address
– But there are not enough IPv4 addresses to go
round so…
– Allow sites to have their own internal private
addresses
– And share just a few global IP addresses
between all of their machines
16
Network Address Translation
– NAT device at boundary between private
network and Internet

translates to and from internal private addresses…
17
Simple NAT

Maps between an internal private IP address
and an external global IP address
– E.g. for a server machine
– NAT device is configured (by hand?!) with the
address mapping
– Re-writes IP packet headers when forwarding:
18
Network Address and Port
Translation (NAPT)

Allows a single external IP to be shared by
many private IPs
– By changing port numbers as well as IP
addresses:
19
Configuring NAPT

Can be statically configured
– E.g. for a web server


External IP, port 80  Internal server IP, port 80
Can be dynamically configured by outgoing
connections/packets
– For normal clients, e.g. accessing external
servers…
– NB. Does NOT allow external hosts to initiate
connections to internal hosts (good security )
20
NAPT dynamic configuration
example

Internal IP IA, port PA sends a packet to external IP
IB, port PB…
– IP header has IPs, UDP/TCP header has ports

NAT device sees outgoing packet
– Chooses a currently unused port number PC
– for its own global IP address, IC
– Creates a new translation mapping

IA, PA  IC,PC (leaves external IP/port)
– Discards mapping if unused for some time
(configurable)
21
NAT/NAPT deployment

Most ISPs
– Hence need to apply specifically for “static”
(globally routable) IP addresses

Many home/small office firewalls and
broadband routers
22
Additional NAT/NAPT issues



Internet server sees NAT device’s IP address and
translated port number (if NAPT)
Private network client only knows its private IP
address and local port
 Client IP address not transferable (correct or
useful) outside the NAT device
– E.g. RMI references passed from client to server will
contain private IP and so won’t work for server
– The client and server will disagree about what they
consider the client’s IP address to be (security issue?!)
23