The Socket Interface

1
The Socket Interface

• Chapter 21

2
The Socket API

• Application Program Interface
• A set of operations performed by an application
  when interacting with protocol software
• BSD Sockets
• Integrated with I/O
• Reading from a socket is conceptually similar to
  reading from a file

3
Procedures That Implement the Socket API

• Socket - All
• Close - All
• Bind Servers
• Listen TCP Servers
• Accept TCP Servers
• Connect
• Clients
• Different function in TCP and UDP (that is a play
  on words)
• Send
• Recv

4
Socket(protofamily, type, protocol)

• Creates a socket and returns an integer
  descriptor, cf open()
• Protofamily is always PF_INET
• There are two types
• SOCK_STREAM (TCP)
• SOCK_DGRAM (UDP)
• Protocol will correspond to either TCP or to UDP
  (getprotobyname())

5
Close(socket)

• Just like closing a file.

6
Bind(socket, localaddr, addrlen)

• Socket has been created but not previously bound
• Localaddr is specified by a C struct
• Family, ie, address type
• Port number
• Ip address

7
Bind (continued)

• Specifying the IP address seems strange, and it
  is, except for multi-homed hosts, ie, hosts with
  more than one IP address, so we usually specify
  INADDR_ANY
• Bind essentially tells the OS to contact us by
  the socket when something arrives on the port to
  which the socket is bound

8
Listen(socket, queuesize)

• Listen registers interest in an event on the
  corresponding socket
• Queuesize specifies the number of requests we are
  willing to handle

9
Socket Accept(socket, caddress, caddresslen)

• Blocks until a client requests a connection
• Fills in the address of that client
• Returns a new socket to handle further
  communication

10
Connect(socket, saddress, saddresslen)

• Used by clients to contact servers
• Socket is a created socket
• Saddress is the IP address, port number you wish
  to contact
• Different with connection-oriented and
  connectionless protocols

11
Send(socket, data, length, flags)

• Like writing to a file whose descriptor is socket
• Data is the address of a buffer where you have
  stored what you wish to send
• Flags is usually zero
• Send assumes a connection

12
Variations on send()

• Sendto(socket, data, length, flags, destaddress,
  addresslen)
• Adds a struct sockaddr specifying where to send
  the data
• Sendmsg(socket, mgsstruct, flags)
• Fills in a struct mgsstruct with destination
  address and the address of the data

13
Recv(socket, buffer, length, flags)

• Just like reading a file
• Socket specifies a particular socket
• Buffer specifies where you want the data stored
• Length indicates the maximum number of bytes you
  are willing to receive
• Very important to prevent buffer overflow

14
Variations on recv

• Recvfrom corresponds to sendto
• Recvmsg corresponds to sendmsg

15
Read(socket, buffer, length)

• Must be used with connected sockets
• Really is just like reading a file
• Used for generality
• Today Im reading a socket
• Tomorrow Ill be reading a file
• Write is used similarly

16
Other Socket Procedures

• Gethostbyname is passed a string, eg, ursa and
  returns the IP address and other information
  about ursa
• Getprotobyname is passed a string, eg tcp and
  returns the corresponding number

17
Additional items of interest

• TCP clients and servers read data from a stream,
  hence in a loop
• UDP clients and servers use a single read
  statement that reads a datagram
• UDP servers
• Do not listen
• Do not accept
• Must use recvfrom and sendto
• UDP clients either
• Connect and use send
• Do not connect and use sendto

18
Two annoying things

• Clients do not bind!! They allow the OS to assign
  a free port for their communication.
• When filling in the destination information you
  usually use a struct sockaddr_in. The C functions
  almost uniformly expect a struct sockaddr. You
  end up doing a lot casts.

19
Start of New Material

• The following slides are taken from
  Internetworking with TCP/IP

20
Socket Inheritance and Termination

• There are OS mechanisms that can be used to
  create new processes or threads. Unix provides
  fork and exec.
• In most systems the new process inherits access
  to the sockets. Socket inheritance is used to
  allow concurrent copies of the server to handle
  multiple requests.

21
Inheritance (continued)

• The OS keeps a reference count associated with
  each socket so it knows how many application
  processes have access to it.
• It is the programmers responsibility to ensure
  that multiple processes use the shared socket
  meaningfully.
• Close doesnt exactly mean close.

22
Obtaining Local and Remoter Socket Addresses

• Sometimes a created process needs to determine
  the destination address to which a socket
  connects.
• A process may also wish to determine the local
  address of a socket.
• Two functions provide this information.

23
Getpeername

• getpeername(socket, destaddr, addrlen)
• socket
• destaddr is a pointer to a structure of type
  sockaddr that will receive the information
• addrlen is a pointer to an integer that will
  receive the length
• Only works with connected sockets

24
Getsockname

• getsockname(socket, localaddr, addrlen)

25
Obtaining and Setting Socket Options

• getsockopt
• setsockopt

26
getsockopt

• getsockopt(socket, level, optionid, optionval,
  length)
• level does the option refer to the socket itself
  or to the underlying protocols
• optionid which option
• optionval and length are pointers
• setsockopt is similar

27
How A Server Accepts Connections

• The bind function associates the socket with a
  well-known port but does not specify a foreign
  destination (which must specify a wildcard).
• The accept function blocks until a connection
  request arrives.

28
Server Connections (continued)

• After accept returns,
• Iterative approach handle the request and close
  the new socket, and calls accept again.
• Concurrent approach fork a process to handle the
  request. The process uses the socket, closes it,
  and terminates.

29
Server Connections (continued)

• Multiple processes will be using the same local
  protocol port number.
• In TCP a pair of endpoints determine a
  connection. It does not matter how many processes
  use a port number as long as they connect to
  different distinations.

30
Server Connections (continued)

• In the case of a concurrent server, there is one
  process per client and one additional process
  that accepts connections. The socket the master
  uses has a wildcard for the foreign destination.
  It will only honor TCP segments that request a
  new connection.

31
IP Address Manipulation Routines

• inet_aton forms a 32-bit IP address from a string
  representing an address in dotted decimal
  notation
• inet_network gets the network portion and zeros
  the host portion
• inet_ntoa reverses inet_aton
• internetaddr combines network and host (net and
  local reverse this op)

32
Accessing the Domain Name System

• res_init() reads a file that contains information
  such as the domain name server and stores the
  results in a global structure, res.
• Res_mkquery(op, dname, class, type, data,
  datalen, newrr, buffer, buflen) forms a domain
  name query and places it in a buffer