UNIX Network Programming

with TCP/IP

UNIX Network Programming

with TCP/IP

UNIX Network Programming

with TCP/IP

UNIX Network
Programming with TCP/IP UNIX
Network Programming
with TCP/IP

Short Course Notes

Alan Dix © 1996 UNIX

Network Programming
with TCP/IP

http://www.hiraeth.com/alan/tutorials
 Course
UNIX
Network Programming
with TCP/IP
Outline
Alan Dix
http://www.hcibook.com/alan

Session 1 Internet Basics

Session 2 First Code

Session 3 Standard Applications

Session 4 Building Clients

Session 5 Servers I

Session 6 Servers II

Session 7 Security

Three interrelated aspects:

❍ TCP/IP protocol suite

❍ standard Internet applications

❍ coding using UNIX sockets API

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 1
UNIX Network Programming
with TCP/IP Reading
Books:
1. W. Richard Stevens, "TCP/IP Illustrated. Vol. 1: The protocols", Addison
Wesley, 1994, (ISBN 0-201-63346-9).
Explains the protocols using network monitoring tools without programming.

2. Douglas E. Comer and David L. Stevens, "Internetworking with TCP/IP.

Vol.3: Client-server programming and applications BSD socket version",
Prentice Hall, 1993, (ISBN 0-13-020272-X).
Good book about principles of client/server design. Assumes you have some
knowledge or at least some other reference for actual programming.

3. Michael Santifaller , translated by Stephen S. Wilson, "TCP/IP and ONC/NFS

internetworking in a UNIX environment", 2nd Edition, Addison Wesley, 1994,
(ISBN 0-201-42275-1).
Covers more ground less deeply. Translation from German seems good.

4. W. Richard Stevens, "UNIX Network Programming", Prentice Hall, 1990,

(ISBN 0-13-949876-1).
A programming book. I'm waiting for a copy, but Stevens is a good writer and
this book is recommended by other authors.

See also:
• your local manual pages (man 2)
• RFCs

Requests for comments (RFCs)

• these are the definition of the Internet protocols
• obtain via anonymous ftp from sun.doc.ic.ac.uk (193.63.255.1)
login as anonymous
give your email address as password
cd to rfc

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2
UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

Session 1 UNIX
Network Programming
with TCP/IP

Internet Basics UNIX

Network Programming
with TCP/IP
 Session 1
UNIX
Network Programming
with TCP/IP Alan Dix
http://www.hcibook.com/alan

• origins
• internets and the Internet
• protocol layers
• addressing
• common applications
☞ using them
• TCP and UDP
• port numbers
• APIs
☞ information calls

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 1
 Origins

Development of Internet & TCP/IP

1968 First proposal for ARPANET – military & gov’t research
Contracted to Bolt, Beranek & Newman
1971 ARPANET enters regular use
1973/4 redesign of lower level protocols
leads to TCP/IP
1983 Berkeley TCP/IP implementation for 4.2BSD
public domain code
1980s rapid growth of NSFNET – broad academic use
1990s WWW and public access to the Internet

The Internet Now

• growing commercialisation of the Internet
• 50,000 networks
• 6 million hosts
• 30 million users
• WWW dominating Internet growth

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2
 internets and the Internet

an internet is
a collection of
• interconnected networks
• (possibly different)
e.g. X25, AppleTalk

the Internet is
a particular internet which
• uses the TCP/IP protocols
• is global
• is hardware and network independent
• is non-proprietary
in addition
• supports commonly used applications
• publicly available standards (RFCs)

the Internet is not (just) the web !

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3
 Characteristics of the Internet

To communicate you need:

• continuous connection
• common language
• means of addressing

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4
 Global Connectivity

ethernet

routers
token ring

sub-network

PPP
star
network router

lots of networks:
• ethernet, FDDI, token ring
• AppleTalk (itself an internet!)
• etc. etc. etc.
connected (possibly indirectly)
• to each other
• to the central ‘ARPAnet’ backbone in the US
protocols can be used in isolation
? but is it the Internet

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5
 Protocols – the Language of the
Internet

TCP/IP OSI
electrical signals Physical

low-level networks Link

(e.g. ethernet)

IP layer (end-to-end)
routers Network
ICMP (control and routing)

TCP/UDP layer Transport

end-points application protocols Session,

(e.g. FTP, telnet, http) Presentation,
Application
application user interfaces
(e.g. Fetch, mosaic)

Standardisation:
• RFC (request for comments) and DoD MIL
RFCs also include (defined but not required):
• PPP, ethernet packaging, etc.
• FTP and other protocols

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6
 Addressing

J. C. T. Jennings,
Linbury Court School,
Dunhambury,
Sussex,
England,
Europe,
Eastern Hemisphere,
Earth,
near Moon,
Solar System,
Space,
near More Space†

Without addresses can only broadcast

Four types of address:
① location independent e.g. personal names
② physical location e.g. letter addresses
③ logical location e.g. organisational hierarchy
④ route based e.g. old email addresses

Two kinds of Internet address:

IP address – type ② (sort of)
e.g. 161.12.188.167
domain name – type ③
e.g. zeus.hud.ac.uk

† extract from Jennings Goes to School, Anthony Buckeridge, Collins, 1950.

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7
 IP addresses
• 32 bit integer – 2701966503
• Often represented as 4 octets – 161.12.188.167

• General structure:
net id { sub-net id } host id
• N.B. octets do not map simply onto components

Five classes of IP address:

Class A 0 netid subnet/hostid

7 bits 24 bits

Class B 1 0 netid subnet/hostid

14 bits 16 bits

Class C 1 1 0 netid hostid

21 bits 8 bits

Class D & Class E – experimental

• hostids may divided using subnet mask

❍ different for each major network (netid)
⇒ needs to be set for each machine on network

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 8
 Domain names

• human readable names

..... or at least ASCII !
• Heirarchical (roughly organisational)
zeus.hud.ac.uk
uk – United Kingdom
ac – academic
hud – huddersfield
zeus – local machine
N.B. USA is implicit – cs.washington.edu

• Decentralised administration
• Mapping
from name to IP address
– domain name servers
also reverse mapping

• C API :
gethostbyname – name → IP address
gethostbyaddr – IP address → name

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 9
 Common applications

• FTP (file transfer protocol)

• SMTP (simple mail transfer protocol)

• telnet (remote logins)

• rlogin (simple remote login between UNIX machines)

• World Wide Web (built on http)

• NFS (network filing system – originally for SUNs)

• TFTP (trivial file transfer protocol – used for booting)

• SNMP (simple network management protocol)

❈ In each case protocols are defined

❈ User interfaces depend on platform

(where relevant)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 10
 Hands on

☞ connect to zeus using telenet:

% telnet zeus.hud.ac.uk
login: c5
... etc.

☞ what happens if you just say “telnet zeus”?

☞ what is zeus’ IP address?

☞ try “telnet aa.bb.cc.dd”

(where ‘aa.bb.cc.dd’ is zeus’ IP address)

☞ connect to zeus using ftp:

% ftp zeus.hud.ac.uk
connect as yourself and then as anonymous

Read between the lines

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 11
 Network communications

Communication can be:

• Connectionless
❍ address every message

❈ like letters

• Connection based
❍ use address to establish a fixed link
❍ send each message using the link

❈ like telephone

N.B. both need an address

⇒ some sort of system address book
or, publicly known addresses

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 12
 Network communications – 2
Other issues:

• Reliability
Do all messages arrive?
Do they arrive in the right order?

• Buffering
effects responsiveness
hides potential deadlock

• Messages or byte-stream
sent:
write 1 (len=26): “abcde....vwxyz”
write 2 (len=10): “0123456789”
received:
read 1 (len=20): “abcde....qrst”
read 2 (len=16): “uvwxyz012...89”

⇒ fixed length messages or prefix with length

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 13
 IP – the fundamental Internet protocol

point to point
❍ between machines
❍ addressed using IP address

message (packet) based

unreliable
❍ network failures
❍ router buffers fill up

dynamic routing
⇒ order may be lost

heterogeneous intermediate networks

⇒ fragmentation

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 14
 TCP & UDP

Both
• built on top of IP
• addressed using port numbers
process to process
(on UNIX platforms)

TCP
• connection based
• reliable
• byte stream
used in: FTP, telnet, http, SMTP

UDP
• connectionless
• unreliable
• datagram (packet based)
used in: NFS, TFTP

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 15
 Port numbers

• 16 bit integers
• unique within a machine
• to connect need IP address + port no

TCP
• connection defined by
IP address & port of server
+ IP address & port of client

UNIX
• port < 1023 – root only
• used for authentication
(e.g. rlogin)

How do you find them?

✔ well known port numbers

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 16
 Well known port numbers

Service Port no Protocol

echo 7 UDP/TCP sends back what it receives

discard 9 UDP/TCP throws away input

daytime 13 UDP/TCP returns ASCII time

chargen 19 UDP/TCP returns characters

ftp 21 TCP file transfer

telnet 23 TCP remote login

smtp 25 TCP email

daytime 37 UDP/TCP returns binary time

tftp 69 UDP trivial file transfer

finger 79 TCP info on users

http 80 TCP World Wide Web

login 513 TCP remote login

who 513 UDP different info on users

Xserver 6000 TCP X windows (N.B. >1023)

N.B. different ‘name’ spaces for TCP & UDP

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 17
 API – the language of the programmer

Application Programmer Interfaces

Not part of the Internet standard – but very important!

A story about DOS

TCP/IP stacks supplied by different vendors
⇒ different device drivers
different APIs
chaos

APIs depend on platform:

UNIX – sockets (original Berkley system calls)
– TLI (transport layer interface)
Apple Mac – MacTCP
MS Windows – WinSock (similar to sockets)

• UNIX TCP/IP API are kernel system calls

• Mac & Windows are extensions/drivers (+DLL)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 18
 Hands on

☞ copy skeleton.c from tcp directory

☞ edit to make two programs:

getid.c – returns IP address of machine
getname.c – returns name of machine

☞ use the following C calls:

gethostid()
returns (lon unsigned) integer result

gethostname(buff,len)
returns error code
puts name into buff (maximum len bytes)

☞ if you have time, play with telnet on different ports

% telnet zeus.hud.ac.uk port_no

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 19
UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

Session 2 UNIX
Network Programming
with TCP/IP

First Code UNIX

Network Programming
with TCP/IP
 Session 2
UNIX
Network Programming
with TCP/IP Alan Dix
http://www.hcibook.com/alan

• features of sockets API

• establishing TCP connections
• simple client/server program
☞ use it
• read & write with sockets
• wrapper functions
• what they do
☞ an echo server

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/1
 Sockets
• developed for Berkeley UNIX
❍ recall early Berkeley TCP/IP implementation
❍ first delivered with BSD 2.1

• central features
❍ central abstraction - the socket - an end-point
like an electrical connector
❍ not TCP/IP specific (e.g. UNIX named pipes)
❍ uses normal read/write system calls
❍ sockets associated with UNIX file descriptors
but some not for normal I/O
❍ some extra system calls

• sits more comfortably with TCP than with UDP

because of byte-stream nature of UNIX I/O

• special UDP functions

e.g., recv(...) – accepts a UDP datagram

• additional non-socket functions

e.g., gethostbyname(...) – domain name server

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/2
 Establishing a TCP Connection
Initial State

Internet

different different different

processes processes processes
151.100.17.25 161.112.192.5 161.112.4.3

• TCP is connection based

... establishing it is a complex multistage process

• initially all machines are the same

• no special ‘server’ machines

• the difference is all in the software

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/3
 Establishing a TCP Connection
Passive Open

21

151.100.17.25
server

161.112.192.5

• server process does a ‘passive’ open on a port

• it waits for a client to connect

• at this stage there is no Internet network traffic

• tells the TCP layer which process to connect to

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/4
 Establishing a TCP Connection
Active Open

161.112.192.5 : 21

21
2397

server

client 161.112.192.5

151.100.17.25

• client process usually on a different machine

• performs an ‘active’ open on the port
• port number at the client end is needed
usually automatic (e.g., 2397)
but can be chosen

• network message → server machine

requests connection

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/5
 Establishing a TCP Connection
Rendezvous

21

client server

151.100.17.25 161.112.192.5

• server side accepts and TCP connection established

• a bi-directional reliable byte-stream

• connection identified by both host/port numbers

e.g. <151.10017.25:2397/161.112.192.5:21>

• server port is not consumed

can stay ‘passive’ open for more connections

• like telephone call desk: one number many lines

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/6
 Establishing a TCP Connection
and more ...

21

client server client

151.100.17.25 161.112.192.5 161.112.4.3

• other clients can connect to the same port

• state for connections in the client/server only

• no information needed in the network

not like old style relay-based exchanges

• server can restrict access to specified host or port

• server can find out connected host/port

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/7
 Passive & Active Open

passive – patient but lazy

active – industrious but impatient

passive active
waits for request for sends out request for
connection connection
waits for ever times out

• normally server does passive open

– waiting for client

• but not always (e.g. ftp)

• active opens can rendezvous ...

... but may miss due to time-outs

• either can specify local port

but if not specified, allocated automatically

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/8
 Simple client/server ‘talk’

• uses simplified calls

• server handles only one client

• strict turntaking

user 1 user 2
zeus: s i m p l e - s e r v e r
start up complete
io: s i m p l e - c l i e n t - h o s t z e u s
You can send now
speak: h i t h e r e
client says: hi there
speak: n i c e d a y i s n ' t i t
server says: nice day isn't it
speak: b i t c o l d h e r e
client says: bit cold here
speak: ^D (EOF) bye bye
zeus:
server finished the conversation
io:

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/9
 Server Code
establish port
port_sk = tcp_passive_open(port)
/* only done once */

wait for client to connect

client_sk = tcp_accept(port_sk)
/* repeated for multiple clients */

then talk to client

for(;;) {

/* wait for client’s message */

len = read(client_sk,buff,buf_len);
buff[len] = '\0';
printf("client says: %s\n",buff);

/* now it’s our turn */

printf("speak: ");
gets(buff);
write(client_sk,buff,strlen(buff));
}

N.B. strict turn taking: client–server–client–server ...

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/10
 Client Code
request connection to server
serv_sk = tcp_active_open(host,port)
/* waits for server to accept */
/* returns negative on failure */
/* host is server’s machine */

then talk to server

for(;;) {

/* our turn first */

printf("speak: ");
gets(buff);
write(serv_sk,buff,strlen(buff));

/* wait for server’s message */

len = read(serv_sk,buff,buf_len);
buff[len] = '\0';
printf("server says: %s\n",buff);
}

N.B. ➀ opposite turn order

➁ no error checking!

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/11
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

☞ copy simple-client.c from tcp/session2 directory

• simple-client.c
• simple-server.c
• makefile

☞ compile and run the programs:

• make simple – compiles them both
• on one machine type:
simple-server
• on another type:
simple-client machine-name
where machine-name is the name of the first

☞ what happens if you re-run the server straight after

it finishes?

☞ use the -port option

z e u s : simple-server -port 3865
io: simple-client -host zeus -port 3865

☞ try a port less than 1024!

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/12
 read & write
Reminder:
ret = read(fd,buff,len)
int fd – a file descriptor (int), open for reading
char *buff – buffer in which to put chars
int len – maximum number of bytes to read
int ret – returns actual number read

• ret is 0 at end of file, negative for error

• buff is not NULL terminated
leave room if you need to add ‘\0’!

ret = write(fd,buff,len)
int fd – a file descriptor (int), open for writing
char *buff – buffer from which to get chars
int len – number of bytes to write
int ret – returns actual number written

• is negative for error, 0 means “end of file”

ret
may be less than len e.g. if OS buffers full
ret
* should really check and repeat until all gone *
• buff need not be NULL terminated
if buff is a C string, use strlen to get its length

N.B. Both may return negative after interrupt (signal)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/13
 read & write with sockets
• similar to normal UNIX pipes
• bi-directional byte stream
❍ read and write to same file descriptor
✘ difficult to close one direction
✔ special socket call shutdown(sock,dir)

• reading may block

❍ reading from a file either:
(i) succeeds
(ii) gets end of file (ret = 0)
❍ reading from a socket waits until
(i) network data received (ret > 0)
(ii) connection closed (ret = 0)
(iii) network error (ret < 0)

• writing may block

❍ writing to a socket may
(i) send to the network (ret > 0)
(ii) find connection is closed (ret = 0)
(iii) network error (ret < 0)
❍ it may return instantly
❍ but may block if buffers are full

✘ BEWARE – may work during testing

then fail in use

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/14
 Wrapper Functions (1)

• not real socket functions

• simplified versions for examples

ret = parse_network_args( &argc, argv,

&host, &port, &errmess )
scan command arguments for network options

port_sk = tcp_passive_open(port)
server performs passive open

serv_sk = tcp_active_open(host,port)
client performs active open

client_sk = tcp_accept(port_sk)
server accepts client connection

❍ parse_network_args does not use socket calls

❍ the rest package one or more socket calls

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/15
 Wrapper Functions (2)
ret = parse_network_args( &argc, argv,
&host, &port, &errmess )

• scans and edits argument list

• looks for options: -host name -port nos
• removes them from argument list
• sets the arguments host and port if options found
• set either host or port to NULL to disable options
• returns 0 for success
non-zero failed – errmess set to appropriate message

port_sk = tcp_passive_open(port)
int port – port number to use
int port_sk – file descriptor of socket

① creates Internet TCP socket

port_sk = socket( AF_INET, SOCK_STREAM, 0 );

② ‘binds’ socket with right port and address 0.0.0.0

(special address means “this machine”)
bind( port_sk, &bind_addr, addr_len );

N.B. port_sk is not used for normal reading and writing

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/16
 Wrapper Functions (3)
serv_sk = tcp_active_open(hostname,port)
char *hostname – name of server’s machine
int port – port number to use
int serv_sk – file descriptor of socket

① finds IP address of host

hostIP = gethostbyname(hostname);

② creates Internet TCP socket

serv_sk = socket( AF_INET, SOCK_STREAM, 0 );

③ ‘connects’ socket to appropriate port and host

connect( serv_sk, &bind_addr, addr_len );

• rendezvous with the server happens at ③

socket serv_sk can then be used to talk to the server

client_sk = tcp_accept(port_sk)
int port_sk – file descriptor of socket

① performs raw accept call

client_sk = accept(port_sk, &bind_addr, &len);

• waits for rendezvous at ①

when it returns client_sk can be used to talk to client

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/17
 Special IP addresses

• bind call in tcp_passive_open uses IP address 0.0.0.0

One of several special IP addresses

0.0.0.0
• source only
• default IP address – ‘local machine’
• filled in by socket API call

127.0.0.0
• loopback address,
• also means ‘the local machine’
• usually used as recipient for local server
• doesn’t normally hit network
• N.B. can also connect to own IP address

255.255.255.255
• limited broadcast (doesn’t pass routers)

any netid – subnetid/hostid = –1

any netid & any subnetid –hostid = –1
• broadcast to specified net or subnet
• N.B. need to know subnet mask

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/18
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

build an echo server

☞ copy simple-server.c and call it echo-server.c

☞ alter code so that instead of asking the user for

input (gets) it simply uses the last message from the
client (in buff)

☞ you will need to add to the makefile:

echo-server: echo-server.o $(MYLIBS)
cc $(CFLAGS) -o echo-server echo-server.o $(MYLIBS)

N.B. this must be a tab

☞ compile and run your code

☞ does your server echo everything once or twice to

its terminal?

☞ the server exits after it has finished echoing

make it continue to wait for additional clients
(don’t try for two at once!)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 2/19
UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

Session 3 UNIX
Network Programming
with TCP/IP

Application UNIX
Network Programming

Protocols
with TCP/IP
 Session 3
UNIX
Network Programming
with TCP/IP Alan Dix
http://www.hcibook.com/alan

Standard Applications
• trusted login – rlogin
• negotiating options – telnet
• world wide web– http
☞ peeking
• file transfer – ftp
• standard response codes
• electronic mail – SMTP
☞ drive it by hand
• argc , argv & makefiles
☞ build your own mail client

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/1
 Types of Protocol
user character stream
❍ used by remote terminal/login applications
( rlogin & telnet )
❍ most of the traffic is uninterpretted data
❍ some embedded control sequences

ascii turn-taking protocols

❍ includes ftp, SMTP, http
❍ human readable client & server messages
❍ usually line oriented
❍ turn-taking typically: client command
server response
···
but roles may reverse
❍ bulk data may be embedded (SMTP, http)
or use separate connection (ftp)

binary protocols
❍ used for low level protocols:
TCP/IP itself!
SNMP – simple network management protocol
NFS (built on top of RPC – remote procedure call)
❍ issues such as byte order important

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/2
 Remote Terminal Access:
rlogin and telnet
• one of the earliest Internet application areas
• the client end – interacts with the user
• the server end – shell or command interpreter
Internet
➅
client
➀

➁ ➄

➂
shell server
➃

basic pattern:
① user types characters
② the client sends them to the server
③ the server passes them on to the shell
④ shell generates output
⑤ server passes output to client
⑥ client puts output on user’s screen

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/3
 Remote Terminals – Issues

• initialisation and authentication

① how does the server know who you are?
② how do you know the server is official?
answer to ②:
❍ the server is on a reserved port (<1024)
N.B. only works for UNIX servers!

• how to deal with special characters

... including end-of-line !

• which end performs different things:

❍ user flow control (crtl-S, ctrl-Q)
❍ line editing
❍ echoing

• how do the client and server communicate:

❍ user interrupts
❍ window size changes
❍ who does what

• if embedded control characters are used

what happens if the user types them?

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/4
 rlogin
• simple stable protocol

• designed for UNIX–UNIX logins

⇒ can make more assumptions
( terminal handling, interrupts, etc. )

• authentication by ‘trusted’ hosts

❍ no password required if:
client uses port <1024
and
client host is in ‘.rhosts’ file
❍ means that client must be setuid to root

• responsibility
❍ echoing – server
❍ flow-control – client on server request

• client–server communication
❍ client→server initialisation string
❍ client→server window size change:
ctrl chars – 2 bytes of 255
followed by window size in 2 bytes
no protection against user typing it!
❍ server→client requests:
special characters (bytes x02,x10,x20,x80)
marked by URG (urgent) pointer

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/5
 Urgent Data
• sometimes called out-of-band data
. . . but it’s not!

• data sent in normal TCP stream

• special URG pointer set
❍ officially to the last byte of urgent data
❍ BSD set it one beyond!
o m e t e x t 0x80 m o r

URG pointer
Berkeley URG pointer!

• client should:
① read until urgent data reached
② if necessary discard intervening data
(e.g. if insufficient buffer space to store it)
problem with ①
❍ URG pointer says where it ends . . .
. . . but how do you know where it starts?
❍ have to have special codes again

• with UNIX sockets

❍ send urgent data with ‘send’ system call
❍ recipient gets a SIGURG signal

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/6
 telnet

• cross platform ⇒ more complex

• many downward-compatible options

• can be used to connect to non-login services

• client authentication
❍ not in protocol – application specific
e.g. getty

• responsibility
❍ client may handle echoing, line editing etc.
subject to option negotiation

• NVT character set

❍ needed because cross-platform
❍ 7 bit US ASCII
❍ end-of-line sent as “\r\n” (carriage return, line feed)
❍ carriage return sent as “\r\0”
❍ also used by SMTP, ftp, finger etc.

✔ high bit free for control characters!

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/7
 telnet – 2

control codes
• introduced by byte 255
❍ called: IAC – interpret as command
• following byte is actual control code
examples:
255 – the actual byte 255 (needed for binary mode)
236 – end of file
241 – no op
243 – break
option negotiation control codes:
251 – WILL
252 – WONT
253 – DO
254 – DONT
250 – sub-option begin
240 – sub-option end

option negotiation
• many different options:
❍ echoing ❍ line editing,
❍ flow control ❍ window size information
• client and server play “will you/wont you”
to determine common protocol
• just like fax machines and modems

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/8
 http

• the World Wide Web protocol

• protocol:
❍ ASCII control messages
❍ standard data formats for pages/images

• uses single step transactions

① establish TCP connection
② client sends request
③ server sends reply + page
④ connection closed

• why transaction based?

❍ client end – many different servers
(hypertext links to different sites)
❍ server end – many clients
❍ load time < interaction time (ideally!)

• why use TCP?

✘ high cost of establishing connection
❍ wide area, large messages & simple clients ⇒
✔ reliable communication needed

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/9
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

peeking
☞ use the program proxy in tcp/session3
☞ it sits between client and server
☞ use it to see how http works:
① run: proxy www.hud.ac.uk 80 -port 8800
② start up Netscape using background menu
③ go to the url:
http://www.hud.ac.uk/schools/comp+maths/private/alan/alandix.html
④ now edit the host name in the url field
if your machine is io
change //www.hud.ac.uk to //io.hud.ac.uk:8800
the 8800 is to set the port number used by proxy
⑤ hit return and watch the proxy window

☞ you can do the same with telnet:

① run: proxy zeus.hud.ac.uk 23 -port 2300
① then: telnet io 2300

N.B. cannot be used for protected ports (ftp, mail etc.)

☞ try using the -v option of ftp

type:
ftp -v prometheus.hud.ac.uk

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/10
 File Transfer Protocol
FTP
• used to transfer files and list directory contents

• uses two types of connection:

control – for commands and responses
data – for files and listings

• protocol for control is ascii turn-taking

client command, server response, ...

• client commands nearly user level, including:

USER user name for connection

often ‘anonymous’ is accepted
PASS password, email address for anonymous
GET receive a file from remote machine
PUT send file to remote machine
CWD change remote directory
LIST change remote directory
PORT tell server what data port to use
HELP info about commands supported
QUIT finish session

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/11
 FTP - 2
control and data
control connection
❍ server waits (passive open) on port 21
❍ client establishes connection (active open)
❍ client sends ascii commands – one per line
❍ server responds: single or multi-line response
❍ when required a data connection is established

data connection
❍ client performs a passive open on some port
(may leave OS to determine port number)
❍ client tells server using control connection
PORT 161.112.192.5.9.93
port 2397 (=9*256+93) on host 161.112.192.5
when data transfer is required
❍ client sends appropriate command
e.g. GET simple-client.c
then waits listening for connection
❍ server performs an active open on port
then sends data
❍ server tells client when transfer is complete
e.g. 226 Transfer complete.
then both sides (usually) close the data port

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/12
 standard response codes
• ftp server replies with lines such as:
200 PORT command successful

• SMTP and some other protocols use similar codes

• three digit codes – type given by first digit:

1yz – expect further reply from server
2yz – all OK
3yz – more required from client
4yz – temporary failure (try again)
5yz – error or permanent failure

• single-line response general format

999 a text message
space here

• multi-line response
either:
hyphen means ‘more to come’
999-first line
999-one or more further lines
999 the last line
space here on last line
or
999-first line
lots of lines all starting with
at least one space
999 the last line

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/13
 Simple Mail Transfer Protocol
SMTP
• allows:
❍ mail client (user interface) to send via server
❍ servers to talk to one another
(one server takes ‘client’ role)
• note:
• not used by user interface for receipt
• sendmail is common SMTP server under UNIX

• client commands:

HELO client tells server who it is

MAIL initiates message and sets sender
RCPT sets one of the recipients
DATA says actual message content follows
VRFY check that recipient exists (no mail sent)
EXPN expand mail alias (no mail sent)
RSET start from scratch
EHLO see if server handles advanced features
QUIT finish session

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/14
 SMTP – 2

• authentication, servers typically:

❍ do not trust HELO
use reverse name mapping instead
❍ do trust sender name (From:)
how could they verify it?

• SMTP specifies delivery not content

• other standards used for content:

❍ non-ASCII characters in headers
=?ISO-8859-1?Q?Alan=20Dix?=
❍ MIME for multi-part mixed content messages

• simple mail message is just:

❍ header
From: [email protected]
To: [email protected]
Subject: HCI book 2E

❍ blank line
❍ body
Russell,
have you heard from Prentice Hall
yet concerning the web pages?
Alan

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/15
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

see what it does

☞ we want to send a mail message using raw SMTP!

☞ first of all see how ‘mail’ does it

cannot use proxy as SMTP is at port 25 (protected)

☞ instead try the -v option of mail

type:
mail -v c3 – or whoever you want to send mail to!
see the messages from the server and the client
N.B. not all messages are shown

☞ when does mail establish the connection?

why?

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/16
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

drive it by hand
☞ use telnet to send a message
type:
telnet zeus.hud.ac.uk 25

☞ you are connected to the SMTP server on zeus

☞ say hello! which machine you are on

HELO walt.disney.com
did it believe you?
how does it know?
☞ now say who the message is from and who it is to
MAIL From:<Donald_Duck>
RCPT To:<[email protected]>

☞ next send the message

DATA
first line of message
..dotty
shear quackery
.

☞ finally say goodbye

QUIT

☞ run mail to see if any celebrity has sent you any

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/17
 argc & argv
• recall: int main( int argc, char **argv ) ...
or: int main( int argc, char *argv[ ] ) ...

• one of the ways to get information into a C program

• in UNIX you type:

myprog a "b c" d
the program gets:
argc = 4 – length of argv
argv[0] = "myprog" – program name
argv[1] = "a"
argv[2] = "b c" – single second argument
argv[3] = "d"
argv[4] = NULL – terminator

N.B. ❍ DOS is identical (except argv[0] is NULL early versions)

❍ argc is one less than the number of arguments!

• other ways to get information in (UNIX & DOS):

❍ configuration file (known name)
❍ standard input
❍ environment variables using getenv()
or (UNIX only) third arg to main:
main(int argc, char **argv, char **envp)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/18
 Make
‘make’ is a UNIX† command which:
• automates program construction and linking
• tracks dependencies
• keeps things up-to-date after changes

to use it:
❍ construct a file with rules in it
you can call it anything, but ‘makefile’ is the default
❍ run ‘make’ itself
make target
– (uses the default makefile)
make -f myfile target
– (uses the rule file myfile)
either rebuilds the program ‘target’ if necessary

• each makefile consists of:

❍ definitions
❍ rules

• rules say how one thing depends on another

they are either:
❍ specific – e.g. to make mail-client do this ...
❍ generic – e.g. to make any ‘.o’ from its ‘.c’ ...

† make is also available in many other programming environments

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/19
 Makefile format
Definitions
• general form:
variable = value
• example:
SDIR = tcp
MYLIBS = $(SDIR)/lib
N.B. one variable used in another's definition
• make variables are referred to later using $
e.g. $(SDIR), $(MYLIBS)
• expanded like #defines or shell variables
(some versions of make will expand shell variables also)

Rules (just specific rules)

• general form:
target : dependent1 dependent2 ...
command-line
N.B. this must be a tab
• example:
myprog: myprog.o another.o
cc -o myprog myprog.o another.o $(MYLIBS)

this says:
to make myprog you need myprog.o and another.o
if either of them is newer than myprog rebuild it using the
then rebuild it using the command: “cc -o myprog ...”

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/20
 Helper Functions
standard response lines

• to make life easier!

• my own helper functions
❍ to read standard response lines
#include "protocol.h"
❍ to interact with SMTP server
#include "mail-helper.h"

int get_response_fd( int server_fd, int echo_fd,

char *buff, int len );

• reads from server_fd

• parses a single or multi-line response
• returns the response code (of last line)
• echoes full response to echo_fd
• also copies it into buff if non-NULL

int get_response_fp( FILE *server_fp, FILE *echo_fp,

char *buff, int len );

• similar with stdio files

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/21
 Helper Functions – 2
for sending mail
int do_mail_init(int serv_fd);

• awaits first response and does ‘HELO’

• checks response and returns 0 if OK

int do_mail_from(int serv_fd, char *from);

int do_mail_to(int serv_fd, char *to);

• sends ‘MAIL’ and ‘RCPT’ commands respectively

• sender (from) and recipient (to) are C strings

int do_mail_data_fp(int serv_fd, FILE *user_fp);

int do_mail_data_buff(int serv_fd, char *buff, int len);

• send ‘DATA’ command and send message

copied from user_fp or buff respectively

int do_mail_quit(int serv_fd);

• does ‘QUIT’ command

All optionally echo all exchanges to a file (or terminal) set by:
FILE *do_mail_set_echo_fp(FILE *new_echo_fp)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/22
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

build your own mail client

☞ copy simple-client.c and call it mail-client.c

☞ copy the following from tcp/session3:

mail-helper.c
make3
the makefile is ready to compile your mail client
you can type (when ready!):
make -f make3 mail-client

N.B. ① SMTP obeys strict turn-taking:

server–client–server–client–server
② server starts with a return code
③ but client ‘in control’

☞ modify the client code

① set default host (zeus) and port (25)
② to and from addresses:
either read in or use argv
③ message: initially read a single line
④ ‘unwrap’ loop to give fixed turns

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/23
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

mail client – 2
☞ resulting program structure:
(a) read (parse) to/from addresses from user
(b) read message from user (gets or scanf)
(c) open tcp connection to mail server on correct port
(d) wait for server response line(s)
(e) say hello to server
(f) wait for server response line(s)
(g) say who the mail is from
(h) wait for server response line(s)
(i) say who the mail is to
(j) wait for server response line(s)
(k) say that data is coming
(l) wait for server response line(s)
(m) send one line message
(n) send line with just full stop
(o) wait for server response line(s)
(p) say goodbye
(q) wait for server response line(s)
(r) close connection

☞ compile and run your code!

☞ if you have time modify it to send longer messages

either: change step (b) and (m) to accept long messages
or: remove step (b) and
make (m) read from user before sending each line
or: whatever you like ...

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 3/24
UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

Session 4 UNIX
Network Programming
with TCP/IP

Concurrent UNIX
Network Programming

Clients
with TCP/IP
 Session 4
UNIX
Network Programming
with TCP/IP Alan Dix
http://www.hcibook.com/alan

• sequential and concurrent clients

• techniques for concurrency
• call-backs
• knowing what you’re doing
• callback–based client
☞ using it

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/1
 Sequential Clients
e.g. FTP
1. client waits for user input
2. user types “DIR”
3. client performs passive open on data port (2397)
4. client sends “PORT 161.112.192.5.9.93” to server
5. client waits for standard ‘200’ reply line
6. if not OK then fail
7. client sends “LIST” to server
8. client waits for standard ‘150’ reply line
9. if not OK then fail
10. client reads from data port
11. client waits for standard ‘226’ reply line
12. if not OK then fail
13. report success to user

• client is in control

• next client action depends on:

❍ what happened last
e.g. what commend the user types
❍ NOT on when it happens

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/2
 Naturally Concurrent Clients

e.g. telnet

• at any moment either

user may type something
or
output may come from server end

• client must respond whichever happens

• program a bit like:

❍ when user types
then send to the server
❍ when server sends message
then print on terminal

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/3
 Concurrency for Usability

e.g. Netscape – WWW client

• basic protocol transaction based

✘. but response can be slow

✓ interaction allowed during transaction

➥ scrolling
➥ ‘STOP’ button

⇒ client has to listen to

server – more data
user – mouse and keyboard

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/4
 Programming Concurrency

Problem
• doing more than one thing at once
listening user terminal & TCP server port

Solutions
• polling
❍ use non-blocking I/O
✘ keeps processor busy

• threads
❍ needs built-in support (language or OS)
❍ program written as several sequential parts
❍ all executed at the same time
❍ communicate using shared data
(also semaphores etc.)

• event driven programming

❍ low-level – e.g. UNIX select
❍ event-loop – e.g., raw X and Mac
❍ program paradigm – e.g. Visual Basic, HyperCard
❍ call-backs – e.g., Windows, X Motif

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/5
 Event Loop

Typical program structure

for (; ; ) { /* loop forever */

struct event_st event;
read_event(&event);
if ( event.type == BUTTON
&& event.target = quit_button)
return OK;
else if ( event.type == KEYPRESS )
insert_char(event.char);

else if ( event.type == INPUT_READY )

do_network_task(event.buff);
. . .
}

✓ programmer in control

✘ related code gets spread out in if/case statements

• often written with sub-loops e.g. for dialogue boxes

⇒ unforeseen events (e.g. network I/O)

may be delayed or even ignored!

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/6
 Event–Based Languages

program = collection of event handlers

e.g. HyperCard
on mouseUp
set cursor to watch
put getServerAddress() into serverAddr
put getUserName() into userName
put cd fld "ToOrFrom" into toName
put cd fld "Message" into theMess
send "toServerSendMail" ¬
&& quote & toName & quote & comma ¬
&& quote & userName & quote & comma ¬
&& quote & theMess & quote ¬
to program serverAddr
end mouseUp

on AppleEvent class, id, sender

answer "AppleEvent" && class && “from” && sender
-- dialogue box for user
end AppleEvent

✓ concurrency naturally part of language

✘ network I/O not always treated uniformly

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/7
 Call-backs

used in many toolkits and window mgrs:

e.g.:
❍ WinSock (TCP/IP under Windows)
❍ X Motif

General pattern
Program
① define a function
② tell toolkit to attach it to event
③ give control to the toolkit

Toolkit
❊ when event happens
call user defined function

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/8
 Example – X Motif Call-backs

XtAddCallback( widget, callback-type, func, my-data )

widget – a widget such as a button

type – a callback resource name:
which type of event to respond to
e.g., XmNactivateCallback
func – pointer to C function defined by you
e.g., quit_func
my-data – an integer or pointer to your data
passed on to your callback

The callback function definition:

void quit_func( widget, my-data, event-data )

widget – where the event occurred

my-data – the integer or pointer passed in
the call to XtAddCallback
event-data – the X event structure which caused
the callback

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/9
 What’s going on?

Sequential Programs
for ( ; ; ) { /* N.B. pseudo-C !!! */
gets(command);
if ( . . . )
if ( command is "quit" ) {
char response[MAX_LINE_SIZE+1]; ← ②
write(serv_sd,"QUIT\n",5);
①☞ read(serv_fd,response,MAX_LINE_SIZE);
if ( response[0] != '2' ) . . .
printf("session complete\n");
exit(0);
}
if ( . . . )
}

features for free

① program counter (☞)
– what you are doing
② local variables
– what you are doing it to

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/10
 What’s going on? - 2

sequential concurrent
implicit explicit

• local variables
global variables
or dynamic data structures
e.g. partial line of user input

• program counter
mode variable
or finite state machines!
e.g. TELNET command sequences
server output modes:
① normal echoing
② waiting for command
③ waiting for option
not byte 255
253 – DO
byte 255 254 – DONT
1 2 3 any
other bytes option

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/11
 Callback based client – 1

➀ Initialisation
main(...) {
/* request connection to server */
sd = tcp_active_open(host,port)
/* set-up callback for server */
inform_input(sd,read_socket,NULL);
/* set-up call-backs for interface */
...
/* give control to toolkit */
inform_loop();
}

➁ When server sends a message ...

... read_socket is called
read_socket( int sd, ... ) {
/* read server’s message */
len = read(sd,buff,buf_len);
/* process message */
/* probably update interface */
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/12
 Callback based client – 2

➂ When user does something ...

... appropriate function is called
term_line( int fd, void *id, char *buff ) {
/* process interface event */
mess("sending {%s}\n",buff);
/* possibly send message to server */
write(sd,buff,strlen(buff));
}

step ➀ once at initialisation

steps ➁ & ➂ any number of times in any order

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/13
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

an electronic conference
☞ copy the following from tcp/session4:
client.c
server.c
make4

the makefile is ready to compile, type :

make -f make4 conf

☞ one person run the server:

io: server

☞ two or more others run the client:

other: client -host io

N.B. you cannot participate from the server

to join in launch a client in another
window of the server’s machine

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 4/14
UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

Session 5 UNIX
Network Programming
with TCP/IP

Server Design UNIX

Network Programming
with TCP/IP
 Session 5
UNIX
Network Programming
with TCP/IP Alan Dix
http://www.hcibook.com/alan

• types of server
• handling server concurrency
• server state
• stateless servers
• when things go wrong!
• survival – the 3 Rs
• callback–based server
☞ modify server

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/1
 Servers

Kinds of server
① transaction based
e.g. database: 1 query → 1 result
② strict turn-taking
e.g. ftp
③ inherent concurrency
e.g. electronic conferences, MUDs

for lots of clients either:

• serve one at a time in turn
⇒ ① may be slow
② may take forever!
• serve several at the same time
⇒ both require concurrency

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/2
 Server Concurrency

• similar solutions to client

❍ polling
✓ acceptable if machine dedicated to server
❍ threads
❍ UNIX select
❍ event driven
✘ less likely to run in event-based system
✓ some web based servers do

• in addition:
❍ when no intrinsic concurrency
❍ can use UNIX fork
✓ launch separate process to serve each client
so each is simpler
✓ uses standard UNIX process concurrency
✘ can be expensive (process creation)
especially with lots of small transactions

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/3
 Server State

• concurrent server needs to remember

❍ how many clients
❍ state of their connection
❍ state of each transaction/protocol
etc. etc. etc.

✘. many clients ⇒ large state

✘. disaster scenarios
❍ client establishes connection
❍ client crashes
❍ client restarts
❍ client establishes a new connection
❍ it crashes again ...

✓ solution – no state

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/4
 Stateless Servers

stateless = no per client state

• for transaction based services

❍ client makes request
❍ server performs action
❍ server returns result

• really only possible with UDP

e.g. http – transaction based, but uses TCP
⇒ may need several reads for request
need to store partially filled buffer ...
N.B. in general, buffers part of the per client state

✘ not all plain sailing ...

❍ clients have to maintain more state
❍ requests more complex (no context)
❍ unreliable protocol ⇒
transactions must be idempotent
time-outs for lots transactions ...

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/5
 When things go wrong

PC crash ⇒ one sad user

server crash ⇒ lots of angry users

• take special care with servers!

• probability of failure:
clients – prob. of failure = p
server – prob. of failure = q
n clients and only 1 server, so:
probability of some failure ≈ np+q

• good news!
❍ server failure less likely (or is it?)

• bad news!
❍ servers are more complex (q > p)
❍ what if client brings server down?

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/6
 Causes of failure

➀ hardware failures
➁ programming errors
➂ unforeseen sequences of events
➃ system does not scale

Large number of components

⇒ ➀ more frequent
Complexity of algorithms
⇒ ➁ more likely
Interleaving and delays
⇒ ➂ difficult to debug
Limited testing conditions
⇒ ➃ unexercised

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/7
 Survival

Network or server failure

standard solutions

Client fails — three Rs for server

• robust
server should survive
• never wait for response from client
• non-blocking network I/O

• reconfigure
detect and respond to failure
• time-out or failure of I/O operations
• reset internal data structures
• inform other clients

• resynchronise
catch up when client restarts
• similar to new client
• N.B. client may not know (network)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/8
 Software faults

Defensive programming
• inconsistent client/server data structures

Use simple algorithms

• fixed sized structures – but check bounds!
• may conflict with scaleability – document

Verify
• close hand checks
• for production code – formal methods

Unforeseen sequences of events

• deadlock – never use blocking I/O
• never assume particular orders of events
• back-to-back messages
network packet ≠ logical message

Debugging and testing

• logging – to reproduce failure
• random data – at interface or network
• ask your friends

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/9
 Callback based server – 1

➀ Initialisation

main(...) {
/* establish port */
pd = tcp_passive_open(port)
/* set-up callback for port */
inform_input(pd,accept_client,NULL);
/* give control to notifier */
inform_loop();
}

➁ When client requests connection ...

... notifier calls accept_client

accept_client(...) {
/* accept client’s connection */
fd = tcp_accept(port_fd);
/* record connection details */
client_fd[count] = fd;
/* set-up callback for client */
inform_input(fd,read_client,count);
/* keep track of number of clients */
count = count+1;
/* probably tell other clients also */
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/10
 Callback based server – 2

➂ When client sends message ...

... notifier calls read_client
read_client( c_fd, id ) {
/* read client’s message */
len = read(c_fd,buff,buf_len);
/* broadcast to other clients */
for( c=0; c<client_count; c++) {
if ( client_fd == c_fd ) {
/* special reply for sender */
}
else {
/* relay message to other clients */
}
}

N.B. step ➀ performed once at initialisation

steps ➁ & ➂ happen any number of times ...
... in any order

• similar to client code, but with extra ‘accept’ stage.

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/11
 My window-less callbacks – 1

• so you can experience the pain of

callbacks without the added pain of
windows ...

# i nc l ude " i nform . h"

int inform_input( int fd, inform_fun f,

inform_id id );

• function f is your callback

• f is called when a buffer can be read from fd
... without blocking
• the identifier id is also passed to f

int inform_output( int fd, inform_fun f,

inform_id id );

• similar to inform_input but for output

• f is called when a buffer can be written to fd

int inform_loop( );

• gives control to the 'notifier' which performs

callbacks for you

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/12
 My window-less callbacks – 2

# i nc l ude " l i ne_by_ l i ne . h"

int inform_line_by_line( int fd, line_fun line_f,

eof_fun eof_f, id_type id );

• the file fd is monitored by notifier

• two callbacks: line_f and eof_f
• line_f is called when a complete line is read
• eof_f is called when the end of file is reached

# i nc l ude "mon i t or . h"

struct mon_tab_struct monitor_tab[] = {

{ 0, "command", callback, "description" },
{ 0, 0, 0, 0 }
};

int perform_line( char *buff );

• helper for simple command interface

• you make monitor_tab with suitable functions
• the first word in buff is regarded as a command
• it is looked up in monitor_tab
. . . and the relevant callback is run

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/13
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

❉ the conference server is not very friendly

it refers to everyone by number
you are going to make this better!

☞ copy server.c call it new-server.c

☞ edit the makefile make4 so that you can

compile new-server.c by using:
make -f make4 new-conf

☞ locate the place where the server first establishes

contact with the client.

☞ make the server wait for a line (or buffer) of input

from the client (the clients name)

☞ modify the notification message it sends to all the

clients to make it name the user

☞ compile and run (use the same client)

run several clients, do you notice delays?

❉ Harder bits
☞ add the user name to the per-client data structure
☞ alter the server so that all messages use the name
rather than client number

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 5/14
UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

Session 6 UNIX
Network Programming
with TCP/IP

Forking Servers UNIX

Network Programming
with TCP/IP

& more TCP/IP

 Session 6
UNIX
Network Programming
with TCP/IP Alan Dix
http://www.hcibook.com/alan

Forking Servers
& TCP/IP behaviour
• UNIX processes and fork
• forking servers
• fork system call
• example code
• dup, exec and wait
☞ remote shell
• inet demon and remote login
☞ another echo server
• IP fragmentation
• TCP flow control

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/1
 Loosely coupled services
• closely coupled:
strong client interaction
e.g. electronic conference

• loosely coupled:
little or no client interaction
e.g. WWW

• no interaction at all ⇒
separate process to serve each client
• weak interaction ⇒
need locking, database server etc.
i.e. some central point of control
client client client client

server server server server

process process process process

database server file locking demon

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/2
 A UNIX process
UNIX process:
• identified by process id (pid)
• process includes:
❍ program code
❍ application data
❍ system data
❋ including file descriptors

pid = 597

code

data

system data
e.g. file descriptors

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/3
 Forking

UNIX 'fork' duplicates process:

• copies complete process state:
❍ program data + system data
❍ including file descriptors
• code immutable – shared

597 632

code code

data data

system data system data

$ echo $$
597
$ (echo $$)
632
$

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/4
 Forking – 2

• old process called the parent

• new process called the child

• process ids
❍ allocated sequentially
❍ so effectively unique
(but do wrap after a very long time)

• finding process ids

❍ at the shell prompt:
use 'ps'
❍ in a C program:
use ' i n t p = ge t p i d( ) ;'
❍ in a shell script:
use '$$'
N.B. useful for naming temporary files:
t mpf i l e = " / t mp / myf i l e$$"

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/5
 Use in servers
① the server passive opens a port
and waits for a client

client server

② the client performs an active open

a connection is established

client server

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/6
 Use in servers – 2

③ the server forks a child

fork
client child server

• child is a copy of the server

• both socket connections are duplicated

⇒ server waiting on port . . .
. . . and child waiting on port
⇒ child connected to client . . .
. . . and server connected to client

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/7
 Use in servers – 3
④ server closes the connection
child closes the passive port

client child server

⑤ server waits for further connections

child talks to client

client child server

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/8
 Fork system call

pid_t p = fork();

( pid_t ≈ int )

• if successful
❍ process
❍ successful fork returns:
0 – to child process
child pid – to parent process

⇒ parent and child are different!

• negative result on failure

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/9
 Execution – 1
• parent forks
597
int i = 3, c_pid = -1;
➡ c_pid = fork();
if ( c_pid == 0 )
printf("child\n");
else if ( c_pid > 0 )
printf("parent\n");
else
printf("failed\n");
DATA i = 3
c_pid = -1

• after fork parent and child identical

597 632

int i = 3, c_pid = -1; int i = 3, c_pid = -1;

c_pid = fork(); c_pid = fork();
➡ if ( c_pid == 0 ) ➡ if ( c_pid == 0 )
printf("child\n"); printf("child\n");
else if ( c_pid > 0 ) else if ( c_pid > 0 )
printf("parent\n"); printf("parent\n");
else else
printf("failed\n"); printf("failed\n");
DATA i = 3 DATA i = 3
c_p i d = 632 c_p i d = 0

• except for the return value of fork

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/10
 Execution – 2
• because data are different
597 632

int i = 3, c_pid = -1; int i = 3, c_pid = -1;

c_pid = fork(); c_pid = fork();
➡ if ( c_pid == 0 ) ➡ if ( c_pid == 0 )
printf("child\n"); printf("child\n");
else if ( c_pid > 0 ) else if ( c_pid > 0 )
printf("parent\n"); printf("parent\n");
else else
printf("failed\n"); printf("failed\n");
DATA i = 3 DATA i = 3
c_p i d = 632 c_p i d = 0

• program execution differs

597 632

int i = 3, c_pid = -1; int i = 3, c_pid = -1;

c_pid = fork(); c_pid = fork();
if ( c_pid == 0 ) if ( c_pid == 0 )
printf("child\n"); ➡ printf("child\n");
else if ( c_pid > 0 ) else if ( c_pid > 0 )
➡ printf("parent\n"); printf("parent\n");
else else
printf("failed\n"); printf("failed\n");
DATA i = 3 DATA i = 3
c_p i d = 632 c_p i d = 0

• so parent and child behaviour diverge

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/11
 fork based shell server – 1

Basic structure:
• establish port
• loop forever
• on each loop:
❍ accept a single client connection
❍ fork a child to manage client
• child execs a copy of the shell

N.B. no login – very insecure !

➀ Main loop

main(...) {
/* open port */
port_sk = tcp_passive_open(port)
/* loop forever accepting clients */
while ( accept_one(port_sk) > 0 );
/* on error close and exit */
close(port_sk);
exit(0);
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/12
 fork based shell server – 2

➁ Process each client in turn

accept_one( int port_sk ) {

/* accept a single connection */
client_sk = tcp_accept(port_sk);
/* perform fork */
child_pid = fork();

• child gets zero return from fork

if ( child_pid == 0 ) {
/* child closes passive port */
close(port_sk);
/* then starts its own behaviour */
exec_a_shell(client_sk);
}

• parent gets child process id returned from fork

else if ( child_pid > 0 ) {
/* parent closes client socket */
close(client_sk);
/* N.B. child has open descriptor */
/* so client is not cut off */
/* returns child pid to main loop */
return child_pid;
}

• negative result on failure

else return 0;
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/13
 fork based shell server – 3

➂ Child execs a copy of the shell

N.B. only the child process calls this function

int exec_a_shell(int fd) /* doesn't return */

{
int tty_fd;;

• shell will expect I/O from standard file descriptors

use 'dup2' system call to link them to fd
dup2(fd,0); /* standard input from fd */
dup2(fd,1); /* standard output to fd */
dup2(fd,2); /* standard error to fd */
close(fd);
execv("/bin/sh",argv);

• exec only returns if it fails

• standard error has been closed
so need to open /dev/tty explicitly
tty_fd = open("/dev/tty",1);
write(tty_fd,exec_fail_mess);
_exit(1);
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/14
 dup2 system call

int res = dup2(old_fd, new_fd);

• makes new_fd point to same file/stream as old_fd

• new_fd is closed if already open

• most often used with standard I/O descriptors:

dup2(fd,0);
– standard input reads from fd

• can close the old descriptor

. . . but new descriptor still works
dup2(fd,0);
close(fd):
n = read(0,buff,buff_len);

• negative return on failure

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/15
 exec system call

execv(char *prog, char **argv);

• replaces the current process with prog

• never returns except on failure

• argvis passed to the 'main' of prog

N.B. needs at least argv[0] set to program name

• new process:
❍ code – replaced by prog
❍ data – reinitialised
❍ system data – partly retained
❇ file descriptors still open

• several variants (execl, execvp, . . . )

• often used after fork to spawn a fresh program

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/16
 exec vs. fork
• fork duplicates process
• exec replaces process

597 493

code code

data data

system system

fork exec

597 632 493

code code code

data data data

system system system

• fork child shares open file descriptors

• exec-ed process retains open fds

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/17
 death of a forked process

• when parent dies

❍ children become orphans !
❍ system init process 'adopts' them

• when child dies

❍ parent (or init) informed by signal
( SIGCHLD )
❍ child process partly destroyed
❍ rump retained until parent 'reaps'
– using wait or wait3 system call
❍ until then child is 'zombie'
– ps says <exiting> or <defunct>

N.B. zombie state necessary so parent

can discover which child died

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/18
 SIGCHLD & wait3

• if parent does not reap children

. . . they stay zombies forever
. . . system resources may run out

① first catch your signal

signal(my_reaper,SIGCHLD);

• function 'my_reaper' called when signal arrives

② then reap a child

int my_reaper()
{
union wait status;
while( wait3(&status,WNOHANG,NULL) >= 0 );
}

• use WNOHANG so that wait3 doesn't block

• loop to reap multiple children

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/19
 fork and I/O

low-level I/O
• open file descriptors shared so:
❍ output is merged
❍ input goes to first read
– accept similar
❍ close down may be delayed
until all processes close fd
⇒ close all unwanted fds
or use ioctl to set close-on-exec

high-level I/O
• C stdio is buffered:
❍ duplicated at fork
❍ may get flushed after fork
⇒ duplicate writes
✓ stderr OK – unbuffered
⇒ careful with stdio
use stderr or setbuff(fd,NULL)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/20
☞ ☞ ☞ Hands on ✍ ✍ ✍

☞ copy the following from tcp/session6:

knife.c
make6

☞ compile knife.c :
make -f make6 knife

☞ launch the knife server: knife.c :

io 3% knife -port 2345

☞ connect to it from a different machine or window

klah 7% telnet io 2345

☞ do you get a shell prompt?

☞ try something simple like

echo hello

☞ then try ps

☞ what happens?

☞ try typing a # at the end of each line

echo hello#
ps #

☞ what is happening?

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/21
 inet demon

• there are many Internet services:

ftp, telnet, rlogin, echo, etc.

• a server for each is expensive

• inetd is a multi-service server

• it does a passive open on lots of ports:

21 – ftp, 25 – SMTP, etc.

• when a client connects

it forks the appropriate service

• remote logins somewhat complicated

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/22
 remote login

First solution . . .
. . . simply fork a shell or getty
✘ no translation of codes
e.g. end of line sequence

✘ no terminal driver at server end

⇒ no tty control by application
e.g. editors need tty raw mode

Actual solution . . .
. . . intermediate process
• server-end process
between client and shell/getty
✓ can perform translation
✓ pseudo-tty between it and shell
⇒ server-end tty control

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/23
 remote login – 2
① remote login client connects to server

client server

② server forks child to handle login

fork
client child server

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/24
 remote login – 3

③ child then forks another process

server

fork and
client child exec

/dev shell

④ the new process connects to the child

using a pseudo-terminal

⑤ and finally execs a shell (or getty etc.)

❇ user is now connected to shell

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/25
 remote login – 4

• client and server-side child similar

❍ both connected to network
❍ both connected to (pseudo)terminal

• general algorithm:
❍ echo terminal input to network
❍ echo network input to terminal

N.B. both concurrent

• difference in use of terminal:

❍ where
client – application end of tty
child – 'user' end of pseudo-tty
❍ how
client – tty always in raw mode
child – pseudo-tty mode set by shell
⇒ only one layer of tty processing

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/26
☞ ☞ ☞ Hands on ✍ ✍ ✍

echo server
☞ modify knife.c to make a forking echo server
your previous echo server (session 2) only dealt
with one client – this one will deal with any number

☞ copy knife.c into echo-all

☞ locate the sub-routine where the shell is exec-ed

☞ replace the code duplicating file descriptors and

exec-ing the shell – simply have a loop which reads
from the socket and writes back to it

☞ compile and run echo-all

io 15% make -f make6 echo-all
io 16% echo-all -port 2345

☞ an connect to it:
klah 23% telnet io 2345

☞ there is an alternative solution which only involves

replacing 2 characters of knife.c

☞ hint: the answer doesn't involve any dogs

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/27
 MTUs

• the Internet is heterogeneous

❍ heterogeneous transport layers
⇒ different packet sizes
❍ dynamic routing
⇒ hops on different layers
⇒ unpredictable packet size

• transport layer limit called MTU:

– maximum transmission unit

transport layer MTU in bytes

Hyperchannel 65535
16Mbps IBM token ring 17914
4Mbps IEEE 802.5 token ring 4464
FDDI 4352
Ethernet 1500
IEEE 802.3/802.2 1492
X.25 576
PPP (performance limit) 296
(from RFC 1191)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/28
 IP fragmentation
• what happens when size is too small?
• fragmentation
❍ any intermediate router detects problem
❍ IP datagram broken into pieces
❍ each sent separately (possibly different routes)
❍ reconstructed at further router or destination

• real limit is recipient's buffer size

❍ 576 bytes IP datagram guaranteed
... but this includes headers
❍ UDP limit = 512 bytes user data
❍ TCP divides data up for you
limit is UNIX read/write buffers

• only end points matter

⇒ in a controlled environment . . .
. . . larger datagrams possible
e.g. NFS = 8192 bytes

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/29
fragmentation considered harmful

• fragmentation
⇒ IP transparent to underlying link layer MTU
. . . well almost . . .

• IP is not reliable
⇒ some packets (fragments) may be lost

• no re-transmission
❍ IP handles reconstruction . . .
. . . but not fragment retransmission
❍ fragment lost
⇒ whole IP datagram lost
❍ probability one fragment lost = p
n fragments
⇒ probability IP datagram lost ≈ n p

• avoiding fragmentation
❍ UDP – most protocols ≤ 512 bytes
❍ TCP – uses local (end-point) MTU
+ path MTU discovery algorithm

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/30
 TCP reliability
• underlying IP unreliable
⇒ TCP must handshake

• stream protocol
❍ sender: this is bytes n–m of the data
❍ recipient: ack m – last byte received

• retransmission
❍ recipient: out of order receipt → repeat ack
❍ timeout or several repeat acks → retransmit

• too many acks

❍ avoid lots of little acknowledgement packets
❍ ack of last packet ⇒ previous packets arrived
❍ piggyback A→B ack on B→A message
❍ delay acks to allow piggyback
❍ turn off delay for some protocols (e.g. X)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/31
 TCP flow control
Cannot send without limits:
● network capacity → packet loss
❍ exponential backoff
rapid resend → nightmare scenario
⇒ long delay before failure (2-9 mins)
❍ slow-start algorithm
● link-layer buffer
❍ MSS announcement

● TCP buffer
❍ window size announcement
only send to last ack + window size

window size

last ack last byte sent

known to must be held
be received at sender end
sent but not may be sent
acknowleged before next ack

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 6/32
UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

UNIX
Network Programming
with TCP/IP

Session 7 UNIX
Network Programming
with TCP/IP

Select and UNIX

Network Programming
with TCP/IP

Security
 Session 7
UNIX
Network Programming
with TCP/IP Alan Dix
http://www.hcibook.com/alan

Select and
Security
• UNIX events
• select system call
• proxy server
☞ raw client
• security, secrecy and privacy
• under attack: viruses & worm
• the Internet worm
• levels of security
• encryption and authentication

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/1
 UNIX Events

Computational programs:
• busy most of the time
• read/write when they are ready

Interactive programs:
• servers & clients
• idle most of the time
• respond to events

UNIX processes – 4 types of event

① signal (interrupt)
② time (alarm)
③ input ready
read will not block
④ output can accept (more) data
write will not block

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/2
 Responding to events

Events:
① signal (interrupt)
② time (alarm)
③ input (read) ready
④ output (write) ready

Responding
• interrupt handler – ①&②
use signal system call
use setitimer to send SIGALRM
• turntaking – ②,③&④
call read/write when ready
use sleep for delays
• polling – ②,③&④
use non-blocking read/write
use time to do things at specific times
• wait for several events
use select system call
timeout or SIGALRM

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/3
 polling in UNIX

#include <sys/filio.h>
ioctl(fd,FIONBIO,1);

• call to ioctl tells system:

don’t block on read/write

• polling therefore possible

• structure of polling telnet-like client:

ioctl(tty_fd,FNBIO,1);
ioctl(net_fd,FNBIO,1);

for(;;) {
/* any terminal input? */
n = read(tty_fd,buff,buff_len);
if ( n > 0 ) { /* yes! do something */ }
/* any network input? */
n = read(net_fd,buff,buff_len);
if ( n > 0 ) { /* yes! do something */ }
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/4
 read & write

read:
• waits on one file descriptor
• returns when input data is ready
• and reads the data into a buffer

read(0,buff,len)

write:
• waits on one file descriptor
• returns when output is possible
• and writes the data from the buffer

write(1,buff,len)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/5
 select

select:
• waits on many file descriptor
• returns when input or output ready
• but does no actual I/O
+ also allows timeout

select(width,&in_fds,&out_fds,&err_fds,&timeout)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/6
 select system call – 2

int ret =
select(size,&in_fds,&out_fds,&err_fds,&timeout);

• in_fds, out_fds:
– bitmaps of file descriptors
❍ in_fds – wait for input
i.e. read will not block
❍ out_fds – wait for output
i.e. write will not block

• size: – size of in_fds, out_fds, err_fds

• timeout: – when to timeout

in seconds and milliseconds

Returns when:
• input ready on one of in_fds (ret > 0)
• output ready on one of out_fds (ret > 0)
• error occurs on one of err_fds (ret > 0)
• timeout expires (ret == 0)
• signal has been caught (ret < 0)
• some other error occurs (ret < 0)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/7
 select and I/O

#include <sys/types.h>
fd_set in_fds, out_fds, err_fds

• modified by call:
call – bit set = wait for file desc
return – bit set = file desc ready
return value from select
= number ready

• long integer in early UNIX systems

in_fds = in_fds || ( 1<<fd );

⇒ limit of 32 file descriptors

. . . but some systems allow more
• now a special fd_set structure
actually an array of integers!
❍ setting:

FD_ZERO( &in_fds );
FD_SET( fd, &in_fds );
FD_CLR( fd, &in_fds );
❍ testing:
if ( FD_ISSET(fd,&in_fds) ) ...

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/8
 select and I/O – 2

• input
❍ terminal/socket
– read will not block
❍ passive socket
– accept will not block

• output
❍ terminal/socket
– write ‘ready’
❍ write relies on system resources
❍ change between select and write?
⇒ write may block
❄ use non-blocking write

• can ‘get away’ without select on write

. . . but dangerous!

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/9
 select and timeouts

#include <sys/time.h>
struct timeval timeout;

• timeout.tv_secs
timeout.tv_ms
– maximum time to wait in seconds and ms

• if no I/O ready and no signals in time limit

then select returns with zero result
N.B. in_fds, out_fds, err_fds all zero also

• modified by call?
❍ ideally should return time remaining
❍ doesn’t now . . .
. . . but may do one day
⇒ don’t rely on timeout not being changed
reset for each call to select

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/10
 select and signals

• signal occurs during system call:

read, write, or select

• signal not caught . . .

. . . process aborts!

• signal caught . . .
① relevant handler called
② systems call returns with ‘error’

• how do you know?

❍ negative return value
❍ errno set to EINTR

• negative return & errno ≠ EINTR

⇒ really an error!

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/11
 care with signals

• signal handlers can run at any time

int i = 0

int my_handler()
{
i = i + 1
}

main()
{
signal(my_handler,SIGINTR);
for(;;)
if ( i > 0 ) {
do_something();
i = i - 1;
}
}

• intention:
execute do_something once per interrupt

• what actually happens:

① interupt processed (i=1)
② do_something executes
③ main calculates i-1 gets result 0
④ before it stores the result . . .
. . . another interupt (i=2)
⑤ main stores result (i=0)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/12
 when to use select

• servers:
❍ where concurrency essential
❍ possibly ftp server
– listen to control & data
❍ telnet server
– listen to user over network
+ listen to shell/application

• clients
❍ not with most window managers
– instead use callback
❍ some event stream WMs
– single fd for WM events
– listen to WM and network
❍ terminal based clients
– not needed for turn-taking
– e.g. telnet/rlogin clients

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/13
 proxy server

• proxy server used in session 3

• structure of code
① passive open on own port
② wait for client connection
③ active open on remote server
④ loop forever
waiting for client or server input:
❍ when client data ready
read it
send to server
echo it to terminal
❍ when server data ready
read it
send to client
echo it to terminal

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/14
 proxy code – 1

➀ Main loop

main(...) {
/* establish port */
port_sk = tcp_passive_open(port);
/* wait for client to connect */
client_sk = tcp_accept(port_sk);

/* only want one client, */

/* so close port_sk */
close(port_sk);

/* now connect to remote server */

serv_sk = tcp_active_open(rem_host,rem_port);

ret = do_proxy( client_sk, serv_sk );

exit(0);
}

• when do_proxy is called both network sockets open

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/15
 proxy code – 2
② perform proxy loop

int do_proxy( int client_sk, int serv_sk )

{

• first declare and initialise fd bitmaps

fd_set read_fds, write_fds, ex_fds;
FD_ZERO(&read_fds); FD_ZERO(&write_fds);
FD_ZERO(&ex_fds);
FD_SET(client_sk,&read_fds);
FD_SET(serv_sk ,&read_fds);

• then loop forever

for(;;) {
int num, len;

• copy bitmaps because select modifies them

fd_set read_copy = read_fds;
fd_set write_copy = write_fds;
fd_set ex_copy = ex_fds;
static struct timeval timeout = {0,0};

• then call select

num = select(MAX_FD, &read_copy, &write_copy,
&ex_copy, &timeout);

➥ check return – ③, ④ & ⑤ at this point

}
return 0;
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/16
 proxy code – 3

③ check for signals, errors and timeout

• first check for signals:

in this case, we are not expecting any so return
in general, we may need to do some processing
following the interrupt
it is usually better for the interrupt to set some
flag and let the main loop do most of the work
this reduces the risk of stacked interrupts and
mistakes in concurrent access to data structures

if (num < 0 && errno == EINTR ) {

/* stopped by signal */
perror("EINTR"); return 1;
}

• if there has been no signal num < 0 is an error

if (num < 0 ) { /* not stopped by signal */
perror("select"); return 1;
}

• if num is zero then a timeout has occurred

again, in this case no processing
but in general this is the opportunity for animation
or other periodic activity

if ( num == 0 ) continue; /* timeout */

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/17
 proxy code – 4
④ check for client input
client ready if bit is set in read_copy
if ( FD_ISSET(client_sk,&read_copy) ) {
int len = read( client_sk, buff, buf_len );

• on end of file or error exit the loop

if ( len <= 0 ) { /* error or close */

close(serv_sk); return len;
}

• if there is some input data, write it to the server and log it

else {
write(serv_sk,buff,len);
log_from_client( buff, len );
}
}

⑤ server input similar

if ( FD_ISSET(serv_sk ,&read_copy) ) {
int len = read( serv_sk , buff, buf_len );
if ( len <= 0 ) { /* error or close */
close(client_sk);
return len;
}
else {
write(client_sk,buff,len);
log_from_server( buff, len );
}
}

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/18
 ☞ ☞ ☞ ☞ Hands on ✍ ✍ ✍ ✍

❉ the proxy server is a bit similar to a telnet client

both open a connection to a remote server
both echo from the user to the server . . .
. . . and from the server to the user
the major difference is that the proxy server
operates on the ‘other end’ of a network connection
☞ you are going make a simple telnet-like client

☞ copy proxy.c and make7 from tcp/session7

copy proxy.c and call it raw-client.c
❉ proxy.creads and writes the client socket
you want to read from standard input (0)
and write to standard output (1)

☞ proceed as follows:
① remove the code to open the client connection
(passive open and accept)
② remove the parameter to do_proxy which
corresponds to the client socket
③ modify the FD_SET calls so that select waits
for standard input (0) rather than the client
④ change all read calls from the client so that
they read from standard input (0)
⑤ change all write calls to the client so that
they write to standard output (1)

☞ now compile and run your raw client, e.g.:

raw-client hades 25
(send mail as in session 3 page 3/17)

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/19
 Security

• types of security:
❍ information:
– secrecy
– privacy
❍ resources:
– destructive access
– virus infection

• linked
❍ information → resources
e.g. password → login
❍ resources → information
e.g. modify /etc/passwd

• chain reaction
❍ small breach → complete loss
e.g. root password!
❍ N.B. special problem for computers

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/20
 who are you afraid of?
• internal
❍ selling your secrets
❍ personal data
– payroll, debtor files etc
❍ using resources
– surfing, doom!
❍ downloading material
– indecent, possibly illegal
❍ backdoors
client_sk = tcp_accept(port_sk);
n= read(client_sk,buff,buff_len);
buff(len) = '\0';
if ( strcmp(buff,"Alan's secret way in") == 0 ) {
/* connect client_sk to a root shell */
}
/* normal operation * /

• external
❍ hackers
❍ accidental release
– e.g. forgotten portable on the train
❍ industrial espionage
❍ viruses

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/21
 under attack

• viruses a real risk?

✔ heterogeneous
⇒ cross-infection more difficult
✘ lots of machines just like yours
? interpreted languages?
– can be made secure (e.g. JAVA)

• types of attack
❍ virus
– embeds itself in another program
❍ Trojan horse
– masquerades as another program
❍ worm
– independent self-replicating program
N.B. names and definitions differ

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/22
 viruses on the web?
• explicit download of code
❍ helpers – machine specific code
❍ general software
✘ both risk infection

• implicit download
❍ semi-compiled – JAVA
❍ interpreted – JAVA script
❍ embedded in HTML
✘ you may never know!

✔ the good news

❍ JAVA & JAVA script ‘safe’
❍ cannot read or write to local disk

✘ the bad news

❍ JAVA script can connect remotely
❍ send details of browsing patterns
❍ minor breach of privacy
? the only breach possible?

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/23
 The Internet Worm
• for 2 days in 1988, the Internet was
under siege

November 2nd, 1988

17:00 worm launched from Cornell University
21:00 worm detected at Stanford
22:04 worm detected at Berkeley
23:40 Berkeley discover one means of attack (sendmail)
23:45 infects Dartmouth and Army Ballistics Res. Lab.

November 3rd, 1988

00:21 Princeton University main machine crashes due to load
02:38 email from Berkeley: “We are under attack”
03:15 anonymous warning from [email protected]
05:54 patches to sendmail distributed
06:45 National Computer Security Centre (NCSC) informed
11:30 Milnet severs itself from Arpanet to prevent infection
16:00 inoculation method found (directory sh in /usr/tmp)
21:30 Berkeley start to decompile ‘captured’ worm

November 4th, 1988

05:00 MIT finish decompiling worm
11:00 Milnet rejoins Arpanet
17:20 final set of preventative patches mailed
21:30 worm’s author identified – named in the next day’s
newspaper as Robert T. Morris son of the NCSC’s chief
scientist Robert Morris!

• infections still noted as late as December 1988

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/24
 What went wrong?

• several means of attack

• between machines:
❍ debug mode in sendmail
❍ buffer overflow in fingerd
❍ once broken into a user on a machine
– rlogin/rsh to other hosts

• within a machine:
❍ simple password attacks
– permutations of user’s own name
– internal list of 432 common passwords
– system dictionary

• attempted to prevent repeat infection

❍ didn’t always work
❍ main damage was excessive load due
to repeat infections (often 100s)
❍ also how it was detected

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/25
 sendmail attack

• sendmail had a debug mode

❍ worm connects to sendmail
❍ worm sends ‘debug’ command
❍ sendmail will then execute
any command!
❍ should have been disabled
but sendmail is complex!

• similar attacks still possible

❍ system engineer accounts
❍ remote vendor maintenance

• any debug modes on your system?

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/26
 fingerd attack

• fingerd uses gets – buffer overflow

❍ worm connects to fingerd
❍ worm sends 536 byte line
❍ overflows fingerd’s buffer (512 bytes)
. . . and corrupts stack
❍ extra 24 bytes executed as code!

• lessons:
❍ never use gets!
– at best may crash
- at worst is a loophole
❍ always be careful of buffer lengths

• never again?
❍ a popular WWW browser . . .
– corrected in later versions

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/27
 physical security

• physical security:
❍ are the machines secure
can someone reboot, substitute disks etc.?
❍ is the network secure
can someone link-in their own computer?

• local or global?
① local network and machines
② backbone and routers
③ remote network and machines
• secure?
① possible
② reasonable for non-critical data
③ no way!
N.B. ‘listening in’ easy on many networks
e.g. ethernet

• never trust transport layer

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/28
 logical security
• secrecy:
❍ TCP/IP packets not secure
e.g. credit card by email
❍ use encryption
e.g. Netscape secure sockets layer for WWW

• authentication:
❍ who am I talking to?
❍ is it the real server?
✔ rely on correct routing and protected ports
✘ impostor machine, non-UNIX server host
❍ is it an acceptable client?
✔ user passwords
✘ often sent as plain text! – e.g. telnet
• audit:
❍ risk of detection deters
❍ keeping logs
❍ relies on authentication
✔ SMTP reverse name lookup
✘ can’t check FROM field – e.g. worm warning

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/29
 low-level protection – firewalls

• simple measures
❍ isolation
– don’t connect to the global Internet
... but lose the benefits too
❍ anonymity
– don’t publish domain machine names
... but IP addresses still valid

• firewalls
❍ application independent
❍ act at router/gateway
❍ can only look at IP or TCP headers

• what is possible
❍ only allow friendly IP addresses
– N.B. impostors
❍ limited internal routing
– protect sensitive machines/data
❍ restrict incoming TCP packets
– only allow connection to protected ports
... but difficult for ftp

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/30
high-level protection – ring fences

• rlogin
❍ beware external root logins!
❍ passwords:
– if reasonable no ‘equiv’ hosts
– certainly no root ‘equiv’ hosts
? means lots of duplicate password files?

• servers
❍ never run as root?
– impossible! e.g. inetd, rshd
❍ never unnecessarily run as root?
– special login e.g. user ‘ftp’
– run as user ‘nobody’

• the rest of the system – normal measures

❍ backups – damage limitation
❍ permissions – restrict ‘other’ access
❍ setuid – dangerous, no write perm!
❍ /etc/passwd – encrypt or restricted read
– may cause problems

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/31
 encryption

• one way function:

cypher = f(input) – easy
input = ?(cypher) – hard
❍ used in /etc/passwd
❍ brute force attack:
for each possible input inp
if f(inp) is cypher – got it!

• single key
cypher = code(key,input)
input = decode(key,cypher)
❍ in DES – code = decode

• public key encryption

cypher = code(key1,input)
input = decode(key2,cypher)
❍ key1 – given to everyone – public
key2 – kept by you – private
❍ anyone can send a message
only you can decrypt it

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/32
 session keys and authentication

• public keys good, but:

❍ expensive
❍ the more you use a key
the easier it is to break
• use public keys to exchange single key
① machine A generates session key KS
② A encrypts it using B’s public key
KSB = code(KB1 , KS )
③ A sends KSB to B
④ B decrypts KSB to obtain KS
KS = decode(KB2 , KSB )
⑤ B generates value X
⑤ B encrypts X and Ks using A’s public key
KXA = code(KA1 , X.KS )
⑥ B sends KXA to A
⑦ A decrypts KXA
X.KS = decode(KA2 , KSA )
⑧ A encrypts X using B’s public key
XB = code(KB1 , X )
⑨ and sends it to B

• result:
❍ A and B share a secret key
❍ A and B sure of each other’s identity

• discard key after session or fixed time

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/33
 authentication servers

• how do you find out B’s public key?

• answers:
① B tells you
② someone else, C, tells you
③ use physical means (post, hand)
• if ① or ②: how do you know it is B/C?
• if ②: why should you believe C?
⇒ ③?
✘ no good for broad distribution

✔ use an authentication server

❍ trusted machine
❍ everyone tells it their public key
(using its public key or physical)
❍ ask it for other’s public keys
❍ or ask it for session keys

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/34
 don’t panic!

• how secure is a fax?

• credit card number by phone

• hacker ≈ burglar
– if they want in, you won’t stop them

• main differences
– rate of loss (Mbytes/sec)
– hidden loss (electronic copies)
– automatic attack

• ease of use ≈ ease of access

– where do you draw the line

UNIX
TCP/IP Short Course Notes Alan Dix © 1996 7/35