Introduction

• Computer Network: hosts, routers, communication channels
• Hosts run applications
• Routers forward information
• Packets: sequence of bytes contain control information
  e.g. destination host
• Protocolis an agreement: meaning of packets structure and size of packets
  e.g. Hypertext Transfer Protocol (HTTP)

Protocol Families -TCP/IP

• Several protocols for different problems: Protocol Suites(套件) or Protocol Families: TCP/IP

• TCP/IP provides end-to-end connectivity specifying how data should be formatted, addressed, transmitted, routed, and received at the destination

• can be used in the internet and in stand-alone private networks(标准的私人网络)

• it is organized into layers

Internet Protocol (IP)

• provides a datagram service
• packets are handled and delivered independently
• best-effort protocol: may loose, reorder(重排序) or duplicate(复制) packets
• each packet must contain an IP address of its destination

Addresses -IPv4

The 32bits of an IPv4 address are broken into 4 octets(字节), or 8 bit fields (0-255 value in decimal notation(10进制标记法)).

For networks of different size, the first one (for large networks) to three (for small networks)octets can be used to identify the network, while the rest of the octets can be used to identify the node on the network.

TCP vs UDP

• Both use port numbers
  1. application-specific construct serving as a communication endpoint
  2. 16-bit unsigned integer, thus ranging from 0 to 65535
  3. to provide end-to-end transport
• UDP: User Datagram Protocol
  1. no acknowledgements(不确认的)
  2. no retransmissions(不重传)
  3. out of order, duplicates possible(无序的,可能存在复制)
  4. connectionless(无连接的), i.e., app indicates destination for each packet(在每个包中显示目的地)
• TCP: Transmission Control Protocol
  1. reliable byte-stream channel(in order, all arrive, no duplicates) similar to file I/O
  2. flow control
  3. connection-oriented
  4. bidirectional(双向的)
• TCP is used for services with a large data capacity(容量), and a persistent(持续的) connection.
• UDP is more commonly used for quick lookups, and single use query-reply actions.
• Some common examples of TCP and UDP with their default ports:

Berkley Sockets

Universally known as Sockets(众所周知): It is an abstraction through which an application may send and receive data. Provide generic accessto interprocesscommunication(进程间通信) services. e.g. IPX/SPX, Appletalk, TCP/IP

Standard API for networking

Sockets

• Uniquely identified by
  1. an internet address
  2. an end-to-end protocol (e.g. TCP or UDP)
  3. a port number
• Two types of (TCP/IP) sockets
  1. Stream sockets (e.g. uses TCP) provide reliable byte-stream service
  2. Datagram sockets (e.g. uses UDP) provide best-effort datagram service messages up to 65.500 bytes
• Socket extend the convectional UNIX I/O facilities
  1. file descriptors for network communication
  2. extended the read and write system calls

Socket Programming

Client-Server communication

• Server
  1. passively waits for and responds to clients
  2. passivesocket
• Client
  1. initiates the communication
  2. must know the address and the port of the server
  3. activesocket

Sockets Procesures

Client-Server Communication - Unix

Socket creation in C: socket()

int sockid= socket(family, type, protocol);

• sockid: socket descriptor, an integer (like a file-handle)
• family: integer, communication domain, e.g.,PF_INET, IPv4 protocols, Internet addresses (typically used); PF_UNIX, Local communication, File addresses
• type: communication type
  1. SOCK_STREAM -reliable, 2-way, connection-based service
  2. SOCK_DGRAM -unreliable, connectionless, messages of maximum length
• protocol: specifies protocol
  1. IPPROTO_TCP IPPROTO_UDP
  2. usually set to 0 (i.e., use default protocol)
• upon failure returns -1

NOTE: socket call does not specify where data will be coming from, nor where it will be going to –it just creates the interface!

Socket close in C: close()

When finished using a socket, the socket should be closed

status= close(sockid);

• sockid: the file descriptor (socket being closed)
• status: 0 if successful, -1 if error

Closing a socket. closes a connection (for stream socket). frees up the port used by the socket

Specifying Addresses

1. Socket API defines a genericdata type for addresses:

struct sockaddr{
unsigned short sa_family; /* Address family (e.g. AF_INET) */
char sa_data[14]; /* Family-specific address information */
}

2. Particular form of the sockaddr used for TCP/IPaddresses:

struct in_addr {
unsigned long s_addr; /* Internet address (32 bits) */
}
struct sockaddr_in{
unsigned short sin_family; /* Internet protocol (AF_INET) */
unsigned short sin_port; /* Address port (16 bits) */
struct in_addrsin_addr; /* Internet address (32 bits) */
char sin_zero[8]; /* Not used */
}

Important: sockaddr_in can be casted to a sockaddr

Assign address to socket: bind()

associates and reserves a port for use by the socket

int status = bind(sockid, &addrport, size);

• sockid: integer, socket descriptor
• addrport: struct sockaddr, the (IP) address and port of the machine: for TCP/IP server, internet address is usually set to INADDR_ANY, i.e., chooses any incoming interface
• size: the size (in bytes) of the addrport structure
• status: upon failure -1 is returned

int sockid;
struct sockaddr_in addrport;

sockid= socket(PF_INET, SOCK_STREAM, 0);

addrport.sin_family= AF_INET;
addrport.sin_port= htons(5100);
addrport.sin_addr.s_addr = htonl(INADDR_ANY);

if(bind(sockid, (struct sockaddr *) &addrport, sizeof(addrport))!= -1){
//…
}

Skipping the bind()

bind can be skipped for both types of sockets:

• Datagram socket:
  1. if only sending, no need to bind. The OS finds a port each time the socket sends a packet
  2. if receiving, need to bind
• Stream socket:
  1. destination determined during connection setup
  2. don’t need to know port sending from (during connection setup, receiving end is informed of port)

Assign address to socket: bind()

Instructs(指示) TCP protocol implementation to listen for connections

int status = listen(sockid, queueLimit);

• sockid: integer, socket descriptor
• queuelen: integer, # of active participants that can “wait”for a connection
• status: 0 if listening, -1 if error

listen()is non-blocking: returns immediately

The listening socket (sockid)
1. is never used for sending and receiving
2. is used by the server only as a way to get new sockets

Establish Connection: connect()

The client establishes a connection with the server by calling connect()

int status = connect(sockid, &foreignAddr, addrlen);

• sockid: integer, socket to be used in connection
• foreignAddr: struct sockaddr: address of the passive participant
• addrlen: integer, sizeof(name)
• status: 0 if successful connect, -1 otherwise

connect()is blocking

Incoming Connection: accept()

The server gets a socket for an incoming client connection by calling accept()

int s= accept(sockid, &clientAddr, &addrLen);

• s: integer, the new socket (used for data-transfer)
• sockid: integer, the orig. socket (being listened on)
• clientAddr: struct sockaddr, address of the active participant;filled in upon return
• addrLen: sizeof(clientAddr): value/result parameter;must be set appropriately(适当地) before call;adjusted upon return

accept():is blocking:
1. waits for connection before returning;
2. dequeues the next connection on the queue for socket (sockid);

Exchanging data with stream socket

int count = send(sockid, msg, msgLen, flags);

• msg: const void[], message to be transmitted
• msgLen: integer, length of message (in bytes) to transmit
• flags: integer, special options, usually just 0
• count: # bytes transmitted (-1 if error)

int count = recv(sockid, recvBuf, bufLen, flags);

• recvBuf: void[], stores received bytes
• bufLen: # bytes received
• flags: integer, special options, usually just 0
• count: # bytes received (-1 if error)

Calls are blocking: returns only after data is sent / received

Exchanging data with datagram socket

int count = sendto(sockid, msg, msgLen, flags, &foreignAddr, addrlen);

msg, msgLen, flags, count: same with send()

• foreignAddr: struct sockaddr, address of the destination
• addrLen: sizeof(foreignAddr)

- int count = recvfrom(sockid, recvBuf, bufLen, flags,&clientAddr, addrlen);

recvBuf, bufLen, flags, count: same with recv()

• clientAddr: struct sockaddr, address of the client
• addrLen: sizeof(clientAddr)

Calls are blocking:returns only after data is sent / received

Example -Echo

• A client communicates with an “echo”server
• The server simply echoes whatever it receives back to the client

Example -Echo using stream socket

The server starts by getting ready to receive client connections…

Server Create a TCP socket

/* Create socket for incoming connections */
if ((servSock= socket(PF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
    DieWithError("socket() failed");

Server Assign a port to socket

echoServAddr.sin_family= AF_INET; /* Internet address family */
echoServAddr.sin_addr.s_addr= htonl(INADDR_ANY); /* Any incoming interface */
echoServAddr.sin_port= htons(echoServPort); /* Local port */

if (bind(servSock, (struct sockaddr *) &echoServAddr, sizeof(echoServAddr)) < 0)
    DieWithError("bind() failed");

Server Set socket to listen

/* Mark the socket so it will listen for incoming connections */
if (listen(servSock, MAXPENDING) < 0)
    DieWithError("listen() failed");

Server Accept new connection

for (;;) /* Run forever */{
    clntLen= sizeof(echoClntAddr);
    if ((clientSock=accept(servSock,(structsockaddr *)&echoClntAddr,&clntLen))<0)
        DieWithError("accept() failed");
        ...

A client decides to talk to the server

Client Create a TCP socket

/* Create a reliable, stream socket using TCP */
if ((clientSock= socket(PF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
    DieWithError("socket() failed");

Client Establish connection

echoServAddr.sin_family= AF_INET; /* Internet address family */
echoServAddr.sin_addr.s_addr= inet_addr(echoservIP); /* Server IP address*/
echoServAddr.sin_port= htons(echoServPort); /* Server port */
if (connect(clientSock, (struct sockaddr *) &echoServAddr, sizeof(echoServAddr)) < 0)
    DieWithError("connect() failed");

Server’s accept procedure in now unblocked and returns client’s socket

for (;;) /* Run forever */{
    clntLen= sizeof(echoClntAddr);
    if ((clientSock=accept(servSock,(structsockaddr *)&echoClntAddr,&clntLen))<0)
        DieWithError("accept() failed");
    ...

Client Communicate

echoStringLen= strlen(echoString); /* Determine input length */
/* Send the string to the server */
if (send(clientSock, echoString, echoStringLen, 0) != echoStringLen)
    DieWithError("send() sent a different number of bytes than expected");

Server using stream socket to Communicate

/* Receive message from client */
if ((recvMsgSize= recv(clntSocket, echoBuffer, RCVBUFSIZE, 0)) < 0)
    DieWithError("recv() failed");

/* Send received string and receive again until end of transmission */
while (recvMsgSize> 0) { /* zero indicates end of transmission */
    if (send(clientSocket, echobuffer, recvMsgSize, 0) != recvMsgSize)
        DieWithError(“send() failed”);

    if ((recvMsgSize= recv(clientSocket, echoBuffer, RECVBUFSIZE, 0)) < 0)
        DieWithError(“recv() failed”);
}

Similarly, the client receives the data from the server

Close communicate

close(clientSocket); /* Client close socket */
close(ClientSocket); /* Server close clientSocket */

Server is now blocked waiting for connection from a client …

Echo using datagram socket

Client and Server Create UDP socket

/* Create socket for sending/receiving datagrams*/
if ((servSock= socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP)) < 0)
    DieWithError("socket() failed");

/* Create a datagram/UDP socket */
if ((clientSock= socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP)) < 0)
    DieWithError("socket() failed");

Assign a port to socket

echoServAddr.sin_family= AF_INET; /* Internet address family */
echoServAddr.sin_addr.s_addr= htonl(INADDR_ANY); /* Any incoming interface */
echoServAddr.sin_port= htons(echoServPort); /* Local port */
if (bind(servSock, (struct sockaddr *) &echoServAddr, sizeof(echoServAddr)) < 0)
    DieWithError("bind() failed");

echoClientAddr.sin_family= AF_INET; /* Internet address family */
echoClientAddr.sin_addr.s_addr= htonl(INADDR_ANY); /* Any incoming interface */
echoClientAddr.sin_port= htons(echoClientPort); /* Local port */
if(bind(clientSock,(structsockaddr *)&echoClientAddr,sizeof(echoClientAddr))<0)
    DieWithError("connect() failed");

Client Sendto Server

echoServAddr.sin_family= AF_INET; /* Internet address family */
echoServAddr.sin_addr.s_addr= inet_addr(echoservIP); /* Server IP address*/
echoServAddr.sin_port= htons(echoServPort); /* Server port */
echoStringLen= strlen(echoString); /* Determine input length */
/* Send the string to the server */
if (sendto( clientSock, echoString, echoStringLen, 0, (struct sockaddr *) &echoServAddr, sizeof(echoServAddr)) != echoStringLen)
    DieWithError("send() sent a different number of bytes than expected");

Server recive and send to Client

for (;;) /* Run forever */{
    clientAddrLen= sizeof(echoClientAddr) /* Set the size of the in-out parameter */
    /*Block until receive message from client*/
    if ((recvMsgSize= recvfrom(servSock, echoBuffer, ECHOMAX, 0),(struct sockaddr *) &echoClientAddr, sizeof(echoClientAddr))) < 0)
        DieWithError("ecvfrom() failed");
        
    if (sendto(servSock, echobuffer, recvMsgSize, 0,(struct sockaddr *) &echoClientAddr, sizeof(echoClientAddr)) != recvMsgSize)
        DieWithError(“send() failed”);
}

Similarly, the client receives the data from the server

Client close socket

close(clientsocket);

Constructing Messages -Encoding Data

Client wants to send two integers x and y to server

1. Solution: Character Encoding
   e.g. ASCII
   • the same representation is used to print or display them to screen
   • allows sending arbitrarily(任意的) large numbers (at least in principle)
     e.g. x = 17,998,720 and y = 47,034,615
     49 55 57 57 56 55 50 48 32 52 55 48 51 52 54 49 53 32
     1 7 9 9 8 7 2 0 _ 4 7 0 3 4 6 1 5 _

     sprintf(msgBuffer, “%d %d ”, x, y);
     send(clientSocket, strlen(msgBuffer), 0);
     

   • Pitfalls(陷阱)
     1. the second delimiter is required
        otherwise the server will not be able to separate it from whatever it follows
     2. msgBuffermust be large enough
     3. strlencounts only the bytes of the message, not the null at the end of the string
   • This solution is not efficient
     1. each digit can be represented using 4 bits, instead of one byte
     2. it is inconvenient to manipulate numbers
2. Solution: Sending the valuesof x and y
   • pitfall: native integer format,a protocolis used
     1. how many bits are used for each integer
     2. what type of encoding is used (e.g. two’s complement(补码), sign(标记)/magnitude(大小), unsigned)

     typedef struct {
     int x,y;
     } msgStruct;
     //…1 implementation
     msgStruct.x = x;
     msgStruct.y = y;
     send(clientSock, &msgStruct, sizeof(msgStruct), 0);
     // or 2 implementation
     send(clientSock, &x, sizeof(x)), 0);
     send(clientSock, &y, sizeof(y)), 0);
     

Address and port are stored as integers

u_short sin_port; (16 bit)
in_addr sin_addr; (32 bit)

Problem

different machines / OS’s use different word orderings

• little-endian: lower bytes first
• big-endian: higher bytes first

these machines may communicate with one another over the network

• Big-Endian
  

• Little-Endian
  

Byte Ordering -Solution: Network Byte Ordering

Host Byte-Ordering: the byte ordering used by a host (big or little)

Network Byte-Ordering: the byte ordering used by the network –always big-endian

u_long htonl(u_longx);
u_short htons(u_shortx);

u_long ntohl(u_longx);
u_short ntohs(u_shortx);

On big-endian machines, these routines do nothing;
On little-endian machines, they reverse the byte order;

Example

Client

unsigned short clientPort, message;
unsigned int messageLenth;

servPort= 1111;
message = htons(clientPort);
messageLength= sizeof(message);

if (sendto( clientSock, message, messageLength, 0, (struct sockaddr *) &echoServAddr, sizeof(echoServAddr)) != messageLength)
    DieWithError("send() sent a different number of bytes than expected");

Server

unsigned short clientPort, rcvBuffer;
unsigned int recvMsgSize;

if(recvfrom(servSock, &rcvBuffer, sizeof(unsignedint), 0),(struct sockaddr*) &echoClientAddr, sizeof(echoClientAddr)) < 0)
    DieWithError("ecvfrom() failed");

clientPort= ntohs(rcvBuffer);
printf (“Client’s port: %d”, clientPort);

Constructing Messages -Alignment and Padding

consider the following 12 bytestructure

typedef struct {
    int x;
    short x2;
    int y;
    short y2;
} msgStruct;

After compilation it will be a 14 bytestructure!

Why? Alignment!

Remember the following rules:data structures are maximally aligned, according to the size of the largest native integer;other multibytefields are aligned to their size, e.g., a four-byte integer’s address will be divisible by four

x x2 y y2 ==> x x2 [pad] y y2
4 bytes 2 bytes 4 bytes 2 bytes ==> 4 bytes 2 bytes 2bytes 4 bytes 2 bytes

This can be avoided
1. include padding to data structure
2. reorder fields

typedef struct {
int x;
short x2;
char pad[2];
int y;
short y2;
} msgStruct;

or

typedef struct {
int x;
int y;
short x2;
short y2;
} msgStruct;

Constructing Messages -Framing and Parsing

Framingis the problem of formatting the information so that the receiver can parse messages.

Parse means to locate the beginning and the end of message.

• This is easy if the fields have fixed sizes, e.g., msgStruct
• For text-string representations is harder
  • Solution: use of appropriate delimiters(使用适当的分隔符)
  • caution(慎重) is needed since a call of recvmay return the messages sent by multiple calls of send

Socket Options

getsockopt and setsockopt allow socket options values to be queried and set, respectively(分别的).

int getsockopt (sockid, level, optName, optVal, optLen);

• sockid: integer, socket descriptor
• level: integer, the layers of the protocol stack (socket, TCP, IP)
• optName: integer, option
• optVal: pointer to a buffer; upon return it contains the value of the specified option
• optLen: integer, in-out parameter
• it returns -1 if an error occured

int setsockopt(sockid, level, optName, optVal, optLen);

• optLen is now only an input parameter

Socket Options Table

Example

Fetch and then double the current number of bytes in the socket’s receive buffer

int rcvBufferSize;
int sockOptSize;
//…
/* Retrieve and print the default buffer size */
sockOptSize= sizeof(recvBuffSize);

if (getsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvBufferSize, &sockOptSize) < 0)
    DieWithError(“getsockopt() failed”);

printf(“InitialReceive Buffer Size: %d\n”, rcvBufferSize);

/* Double the buffer size */
recvBufferSize*= 2;

/* Set the buffer size to new value */
if (setsockopt(sock, SOL_SOCKET, SO_RCVBUF, &rcvBufferSize, sizeof(rcvBufferSize)) < 0)
    DieWithError(“getsockopt() failed”);

Dealing with blocking calls

Many of the functions we saw block (by default) until a certain event

• accept: until a connection comes in
• connect: until the connection is established
• recv, recvfrom: until a packet (of data) is received
  what if a packet is lost (in datagram socket)?
• send: until data are pushed into socket’s buffer
• sendto: until data are given to the network subsystem

For simple programs, blocking is convenient; What about more complex programs?
- multiple connections
- simultaneous(同时的) sends and receives
- simultaneously doing non-networking processing

Non-blocking Sockets

If an operation can be completed immediately, success is returned; otherwise, a failure is returned (usually -1)

errnois properly set, to distinguish this (blocking) failure from other -(EINPROGRESSfor connect, EWOULDBLOCKfor the other)

1. Solution:int fcntl (sockid, command, argument);
   • sockid: integer, socket descriptor
   • command: integer, the operation to be performed (F_GETFL, F_SETFL)
   • argument: long, e.g. O_NONBLOCK
     fcntl (sockid, F_SETFL, O_NONBLOCK);
2. Solution: flags parameter of send, recv, sendto, recvfrom
   • MSG_DONTWAIT
   • not supported by all implementations

Signals

Provide a mechanism for operating system to notify processes that certain events occur e.g., the user typed the “interrupt”character, or a timer expired;

signals are delivered asynchronously;

upon signal delivery to program
- it may be ignored, the process is never aware of it
- the program is forcefully terminatedby the OS
- a signal-handling routine, specified by the program, is executed
this happens in a different thread
- the signal is blocked, until the program takes action to allow its delivery
each process (or thread) has a corresponding mask

Each signal has a default behavior

e.g. SIGINT (i.e., Ctrl+C) causes termination, it can be changed using sigaction(); Signals can be nested(嵌套)(i.e., while one is being handled another is delivered)

int sigaction(whichSignal, &newAction, &oldAction);

• whichSignal: integer
• newAction: struct sigaction, defines the new behavior
• oldAction: struct sigaction, if not NULL, then previous behavior is copied
• it returns 0 on success, -1 otherwise

struct sigaction {
void (*sa_handler)(int); /* Signal handler */
sigset_tsa_mask; /* Signals to be blocked during handler execution */
int sa_flags; /* Flags to modify default behavior */
};

• sa_handler determines which of the first three possibilities occurs when signal is delivered, i.e., it is not masked
  SIG_IGN, SIG_DFL, address of a function
• sa_masks pecifies the signals to be blocked while handling whichSignal
  whichSignal is always blocked
  it is implemented as a set of boolean flags

int sigemptyset(sigset_t*set); /* unset all the flags */
int sigfullset(sigset_t*set); /* set all the flags */
int sigaddset(sigset_t*set, int whichSignal); /* set individual flag */
int sigdelset(sigset_t*set, int whichSignal); /* unset individual flag */

Signal Example

#include <stdio.h>
#include <signal.h>
#include <unistd.h>

void DieWithError(char *errorMessage);
voidInterruptSignalHandler(int signalType);

int main (int argc, char *argv[]) {
    struct sigaction handler;/* Signal handler specification structure */

    handler.sa_handler =InterruptSignalHandler; /* Set handler function */

    if (sigfillset(&handler.sa_mask) < 0)/* Create mask that masks all signals */
        DieWithError (“sigfillset() failed”);

    handler.sa_flags = 0;
    if (sigaction(SIGINT,&handler, 0) < 0)/* Set signal handling for interrupt signals */
        DieWithError (“sigaction() failed”);

    for(;;)
        pause();/* Suspend program until signal received */

    exit(0);
}

voidInterruptHandler(intsignalType) {
    printf (“Interrupt received. Exiting program.\n);
    exit(1);
}

Asynchronous I/O

Non-blocking sockets require “polling(查询)”;

With asynchronous I/O the operating system informs the program when a socket call is completed.the SIGIO signal is delivered to the process, when some I/O-related event occurs on the socket;

Three steps:

/* i. inform the system of the desired dispositionof(所需配置) the signal */
struct sigaction handler;
handler.sa_handler = SIGIOHandler;
if (sigfillset(&handler.sa_mask) < 0) 
    DiewithError(“…”);

handler.sa_flags = 0;
if (sigaction(SIGIO, &handler, 0) < 0) 
    DieWithError(“…”);

/* ii. ensure that signals related to the socket will be delivered to this process*/
if (fcntl(sock, F_SETOWN, getpid()) < 0) 
    DieWithError();

/* iii. mark the socket as being primed for asynchronous I/O*/
if (fcntl(sock, F_SETFL, O_NONBLOCK | FASYNC) < 0) 
    DieWithError();

Asynchronous I/O Example

/* Inform the system of the desired dispositionof the signal */
struct sigaction myAction;
myAction.sa_handler= CatchAlarm;

if (sigfillset(&myAction.sa_mask) < 0) 
    DiewithError(“…”);

myAction.sa_flags= 0;
if (sigaction(SIGALARM, &handler, 0) < 0) 
    DieWithError(“…”);

/* Set alarm */
alarm(TIMEOUT_SECS);

/* Call blocking receive */
if (recvfrom(sock, echoBuffer, ECHOMAX, 0, …) < 0) {
    if (errno = EINTR)
        …/*Alarm went off */
    else 
        DieWithError(“recvfrom() failed”);
}

Timeouts

Using asynchronous I/O the operating system informs the program for the occurrence of an I/O related event

what happens if a UPD packet is lost?

We may need to know if something doesn’t happen after some time

unsigned int alarm (unsigned int secs);

starts a timer that expires after the specified number of seconds (secs)

• returns the number of seconds remaining until any previously scheduled alarm was due to be delivered, or zero if there was no previously scheduled alarm

process receives SIGALARM signal when timer expires and errnois set to EINTR;

Iterative Stream Socket Server

• Handles one client at a time
• Additional clients can connect while one is being served
  1. connections are established
  2. they are able to send requests
     but, the server will respond after it finishes with the first client
• Works well if each client required a small, bounded amount of work by the server
• otherwise, the clients experience long delays

Iterative Server-Example: echo using stream socket

#include <stdio.h> /* for printf() and fprintf() */
#include <sys/socket.h> /* for socket(), bind(), connect(), recv() and send() */
#include <arpa/inet.h> /* for sockaddr_inand inet_ntoa() */
#include <stdlib.h> /* for atoi() and exit() */
#include <string.h> /* for memset() */
#include <unistd.h> /* for close() */

#define MAXPENDING 5 /* Maximum outstanding connection requests */

void DieWithError(char *errorMessage); /* Error handling function */
void HandleTCPClient(intclntSocket); /* TCP client handling function */

int main(intargc, char *argv[]) 
{
    int servSock; /* Socket descriptor for server */
    int clntSock; /* Socket descriptor for client */
    struct sockaddr_inechoServAddr; /* Local address */
    struct sockaddr_inechoClntAddr; /* Client address */
    unsigned short echoServPort; /* Server port */
    unsigned int clntLen; /* Length of client address data structure */

    if (argc != 2) { /* Test for correct number of arguments*/
        fprintf(stderr, "Usage: %s <Server Port>\n", argv[0]);
        exit(1);
    }

    echoServPort = atoi(argv[1]); /* First arg: local port */
    /* Create socket for incoming connections */
    if ((servSock = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP)) < 0)
        DieWithError("socket() failed");

    /* Construct local address structure */
    memset(&echoServAddr, 0, sizeof(echoServAddr)); /* Zero out structure */
    echoServAddr.sin_family= AF_INET; /* Internet address family */
    echoServAddr.sin_addr.s_addr= htonl(INADDR_ANY); /* Any incoming interface */
    echoServAddr.sin_port= htons(echoServPort); /* Local port */

    /* Bind to the local address */
    if (bind(servSock, (struct sockaddr*) &echoServAddr, sizeof(echoServAddr)) < 0)
        DieWithError("bind() failed");

    /* Mark the socket so it will listen for incoming connections */
    if (listen(servSock, MAXPENDING) < 0)
        DieWithError("listen() failed");

    for (;;) /* Run forever */
    {
        /* Set the size of the in-out parameter */
        clntLen= sizeof(echoClntAddr);

        /* Wait for a client to connect */
        if ((clntSock= accept(servSock, (struct sockaddr*) &echoClntAddr, &clntLen)) < 0)
            DieWithError("accept() failed");

        /* clntSock is connected to a client! */
        printf("Handlingclient %s\n", inet_ntoa(echoClntAddr.sin_addr));
        HandleTCPClient(clntSock);
    }
    /* NOT REACHED */
}

#define RCVBUFSIZE 32 /* Size of receive buffer */
void HandleTCPClient(intclntSocket)
{
    char echoBuffer[RCVBUFSIZE]; /* Buffer for echo string */
    int recvMsgSize; /* Size of received message */

    /* Receive message from client */
    if ((recvMsgSize= recv(clntSocket, echoBuffer, RCVBUFSIZE, 0)) < 0)
        DieWithError("recv() failed");

    /* Send received string and receive again until end of transmission */
    while (recvMsgSize> 0) /* zero indicates end of transmission */
    {
        /* Echo message back to client */
        if (send(clntSocket, echoBuffer, recvMsgSize, 0) != recvMsgSize)
            DieWithError("send() failed");

        /* See if there is more data to receive */
        if ((recvMsgSize= recv(clntSocket, echoBuffer, RCVBUFSIZE, 0)) < 0)
            DieWithError("recv() failed");
    }
    close(clntSocket); /* Close client socket */
}

Multitasking -Per-Client Process

For each client connection request, a new process is created to handle the communication.

int fork();

1. a new process is created, identical to the calling process, except for its process ID and the return value it receives from fork()
2. returns 0 to childprocess, and the process ID of the new child to parent

aution:(慎重)

1. when a child process terminates, it does not automatically disappears
2. use waitpid()to parent in order to “harvest” zombies

Multitasking-Per-Client Process -Example: echo using stream socket

#include <sys/wait.h>/* for waitpid() */

int main(intargc, char *argv[]) 
{
    int servSock; /* Socket descriptor for server */
    int clntSock; /* Socket descriptor for client */
    unsigned short echoServPort; /* Server port */
    pid_t processID;/* Process ID from fork()*/
    unsigned int childProcCount= 0; /* Number of child processes */

    if (argc != 2) { /* Test for correct number of arguments */
        fprintf(stderr, "Usage: %s <Server Port>\n", argv[0]);
        exit(1);
    }

    echoServPort = atoi(argv[1]); /* First arg: local port */
    servSock= CreateTCPServerSocket(echoServPort);

    for (;;) { /* Run forever */
        clntSock= AcceptTCPConnection(servSock);
        if ((processID= fork()) < 0) 
            DieWithError (“fork() failed”); /* Fork child process */
        else if (processID= 0) { /* This is the child process */
            close(servSock); /* child closes listening socket */

            HandleTCPClient(clntSock);
            exit(0); /* child process terminates */
        }

        close(clntSock);/* parent closes child socket */
        childProcCount++;/* Increment number of outstanding child processes */

        while (childProcCount) {/* Clean up all zombies */
            processID= waitpid((pid_t) -1, NULL, WHOANG); /* Non-blocking wait */
            if (processID< 0) 
                DieWithError (“...”);
            else if (processID== 0) 
                break;/* No zombie to wait */
            else
                childProcCount--;/* Cleaned up after a child */
        }
    }
    /* NOT REACHED */
}

Multitasking -Per-Client Thread

• Forking a new process is expensive
  duplicate the entire state (memory, stack, file/socket descriptors, …)
• Threads decrease this cost by allowing multitasking within the same process
  threads share the same address space (code and data)

Multitasking -Per-Client Thread -Example: echo using stream socket

#include <pthread.h>/* for POSIX threads */

void *ThreadMain(void*arg)/* Main program of a thread */

struct ThreadArgs{/* Structure of arguments to pass to client thread */
int clntSock;/* socket descriptor for client */
};

int main(intargc, char *argv[]) 
{
    int servSock; /* Socket descriptor for server */
    int clntSock; /* Socket descriptor for client */
    unsigned short echoServPort; /* Server port */
    pthread_t threadID;/* Thread ID from pthread_create()*/
    struct ThreadArgs*threadArgs; /* Pointer to argument structure for thread */
    if (argc != 2) { /* Test for correct number of arguments */
        fprintf(stderr, "Usage: %s <Server Port>\n", argv[0]);
        exit(1);
    }

    echoServPort = atoi(argv[1]); /* First arg: local port */
    servSock= CreateTCPServerSocket(echoServPort);
    for (;;) { /* Run forever */
        clntSock= AcceptTCPConnection(servSock);
        /* Create separate memory for client argument */
        if ((threadArgs= (struct ThreadArgs*) malloc(sizeof(structThreadArgs)))) == NULL) 
            DieWithError(“…”);

        threadArgs-> clntSock = clntSock;
        /* Create client thread */
        if (pthread_create (&threadID, NULL, ThreadMain, (void *) threadArgs) != 0) 
            DieWithError(“…”);
    }
    /* NOT REACHED */
}

void *ThreadMain(void*threadArgs)
{
    int clntSock; /* Socket descriptor for client connection */
    pthread_detach(pthread_self()); /* Guarantees that thread resources are deallocatedupon return */
    /* Extract socket file descriptor from argument */
    clntSock= ((struct ThreadArgs*) threadArgs) -> clntSock;
    free(threadArgs); /* Deallocatememory for argument */
    HandleTCPClient(clntSock);
    return (NULL);
}

Multitasking -Constrained

• Both process and thread incurs(承受) overhead(开销): creation, scheduling and context switching
• As their numbers increases
  1. this overhead increases
  2. after some point it would be better if a client was blocked

Solution: Constrained multitasking. The server:
1. begins, creating, binding and listening to a socket
2. creates a number of processes, each loops forever and accept connections from the same socket
3. when a connection is established
- the client socket descriptor is returned to only one process
- the other remain blocked

Multitasking -Constrained -Example: echo using stream socket

void ProcessMain(int servSock); /* Main program of process */

int main(intargc, char *argv[]) 
{
    int servSock; /* Socket descriptor for server*/
    unsigned short echoServPort; /* Server port */
    pid_t processID; /* Process ID */
    unsigned int processLimit; /* Number of child processes to create */
    unsigned int processCt; /* Process counter */

    if (argc != 3) { /* Test for correct number of arguments*/
        fprintf(stderr,"Usage: %s <SERVER PORT> <FORK LIMIT>\n", argv[0]);
        exit(1);
    }

    echoServPort = atoi(argv[1]); /* First arg: local port */
    processLimit= atoi(argv[2]); /* Second arg: number of child processes */
    servSock = CreateTCPServerSocket(echoServPort);
    for (processCt=0; processCt< processLimit; processCt++)
        if ((processID= fork()) < 0) 
            DieWithError("fork() failed"); /* Fork child process */
        else if (processID== 0)
            ProcessMain(servSock); /* If this is the child process */

    exit(0); /* The children will carry on */
}

void ProcessMain(int servSock) 
{
    int clntSock; /* Socket descriptor for client connection */
    for (;;) { /* Run forever */
        clntSock = AcceptTCPConnection(servSock);
        printf("withchild process: %d\n", (unsigned int) getpid());
        HandleTCPClient(clntSock);
    }
}

Multiplexing

So far, we have dealt with a single I/O channel;

• We may need to cope with multiple I/O channels;
  e.g., supporting the echo service over multiple ports

• Problem: from which socket the server should accept connections or receive messages?
  it can be solved using non-blocking sockets, but it requires polling

• Solution: select()

  • specifies a list of descriptors to check for pending I/O operations
  • blocks until one of the descriptors is ready
  • returns which descriptors are ready

int select (maxDescPlus1, &readDescs, &writeDescs, &exceptionDescs, &timeout);

• maxDescsPlus1: integer, hint of the maximum number of descriptors
• readDescs: fd_set, checked for immediate input availability
• writeDescs: fd_set, checked for the ability to immediately write data
• exceptionDescs: fd_set, checked for pending exceptions
• timeout: struct timeval, how long it blocks (NULL forever)
• returns the total number of ready descriptors, -1 in case of error
• changes the descriptor lists so that only the corresponding positions are set

int FD_ZERO (fd_set*descriptorVector); /* removes all descriptors from vector */
int FD_CLR (int descriptor, fd_set*descriptorVector); /* remove descriptor from vector */
int FD_SET (int descriptor, fd_set*descriptorVector); /* add descriptor to vector */
int FD_ISSET (int descriptor, fd_set*descriptorVector); /* vector membership check */

struct timeval{
time_ttv_sec;/* seconds */
time_ttv_usec;/* microseconds */
};

Multiplexing -Example: echo using stream socket

#include <sys/time.h> /* for struct timeval{} */
int main(intargc, char *argv[])
{
    int *servSock; /* Socket descriptors for server */
    int maxDescriptor; /* Maximum socket descriptor value */
    fd_setsockSet; /* Set of socket descriptors for select() */
    long timeout; /* Timeout value given on command-line */
    struct timevalselTimeout; /* Timeout for select() */
    int running = 1; /* 1 if server should be running; 0 otherwise */
    int noPorts; /* Number of port specified on command-line */
    int port; /* Looping variable for ports */
    unsigned short portNo; /* Actual port number */

    if (argc < 3) { /* Test for correct number of arguments */
        fprintf(stderr, "Usage: %s <Timeout (secs.)> <Port 1> ...\n", argv[0]);
        exit(1);
    }

    timeout= atol(argv[1]); /* First arg: Timeout */
    noPorts= argc -2; /* Number of ports is argument count minus 2 */
    servSock = (int *) malloc(noPorts* sizeof(int)); /* Allocate list of sockets for incoming connections */
    maxDescriptor= -1; /* Initialize maxDescriptorfor use by select() */

    for (port = 0; port < noPorts; port++) { /* Create list of ports and sockets to handle ports */
        portNo= atoi(argv[port+ 2]); /* Add port to port list. Skip first two arguments */
        servSock[port] = CreateTCPServerSocket(portNo); /* Create port socket */
        if (servSock[port] > maxDescriptor)/* Determine if new descriptor is the largest */
        maxDescriptor= servSock[port];
    }

    printf("Startingserver: Hit return to shutdown\n");
    while (running) {
    /* Zero socket descriptor vector and set for server sockets */
    /* This must be reset every time select() is called */
    FD_ZERO(&sockSet);
    FD_SET(STDIN_FILENO, &sockSet); /* Add keyboard to descriptor vector */
    for (port = 0; port < noPorts; port++) 
        FD_SET(servSock[port], &sockSet);
        /* Timeout specification */

        /* This must be reset every time select() is called */
        selTimeout.tv_sec= timeout; /* timeout (secs.) */
        selTimeout.tv_usec= 0; /* 0 microseconds */

        /* Suspend program until descriptor is ready or timeout */
        if (select(maxDescriptor+ 1, &sockSet, NULL, NULL, &selTimeout) == 0)
            printf("Noecho requests for %ld secs...Server still alive\n", timeout);
        else {
            if (FD_ISSET(0, &sockSet)) { /* Check keyboard */
                printf("Shuttingdown server\n");

            getchar();
            running = 0;
        }

        for (port = 0; port < noPorts; port++)
            if (FD_ISSET(servSock[port], &sockSet)) {
                printf("Requeston port %d: ", port);
                HandleTCPClient(AcceptTCPConnection(servSock[port]));
            }
        }
    }

    for (port = 0; port < noPorts; port++)
        close(servSock[port]); /* Close sockets */

    free(servSock);/* Free list of sockets */
    exit(0);
}

Multiple Recipients(接收者)

So far, all sockets have dealt with unicast(单播) communication i.e., an one-to-one communication, where one copy (“uni”) of the data is sent (“cast”).

what if we want to send data to multiple recipients?

1. Solution: unicasta(单播) copy of the data to each recipient; inefficient(低效率), e.g.,
   • consider we are connected to the internet through a 3Mbps line
   • a video server sends 1-Mbps streams
   • then, server can support only three clients simultaneously
2. Solution: using network support
   • broadcast, all the hosts of the network receive the message
   • multicast, a message is sent to some subset of the host
   • for IP: only UDP socketsare allowed to broadcast and multicast

Multiple Recipients -Broadcast

Only the IP address changes

Localbroadcast: to address 255.255.255.255
- send the message to every host on the same broadcast network
- not forwarded by the routers(不被路由转发)

Directed broadcast:
- for network identifier 169.125(i.e., with subnet mask 255.255.0.0)
- the directed broadcast address is 169.125.255.255

No network-wide broadcast address is available, why?

• In order to use broadcast the options of socket must change:

int broadcastPermission = 1;
setsockopt(sock, SOL_SOCKET, SO_BROADCAST, (void*) &broadcastPermission, sizeof(broadcastPermission));

• Using class Daddresses
  range from 224.0.0.0to 239.255.255.255
• hosts send multicast requestsfor specific addresses
• a multicast groupis formed

we need to set TTL (time-to-live), to limit the number of hops -using sockopt()

no need to change the options of socket

Usefull Functions

int atoi(constchar *nptr);

converts the initial portion of the string pointed to by nptrto int

int inet_aton(constchar *cp, struct in_addr*inp);

• converts the Internet host address cpfrom the IPv4 numbers-and-dots notation into binary form (in network byte order)
• stores it in the structure that inppoints to.
• it returns nonzero if the address is valid, and 0 if not

char *inet_ntoa(structin_addrin);

converts the Internet host address in, given in network byte order, to a string in IPv4 dotted-decimal notation

typedef uint32_t in_addr_t;
struct in_addr{
in_addr_t s_addr;
};

int getpeername(intsockfd, struct sockaddr*addr, socklen_t*addrlen);

• returns the address (IP and port) of the peer connected to the socket sockfd, in the buffer pointed to by addr
• 0 is returned on success; -1 otherwise

int getsockname(intsockfd, struct sockaddr*addr, socklen_t*addrlen);

• returns the current address to which the socket sockfdis bound, in the buffer pointed to by addr
• 0 is returned on success; -1 otherwise

Domain Name Service

struct hostent*gethostbyname(constchar *name);

• returns a structure of type hostentfor the given host name
• name is a hostname, or an IPv4 address in standard dot notatione.g. gethostbyname(“www.csd.uoc.gr”);

struct hostent*gethostbyaddr(constvoid *addr, socklen_tlen, int type);

returns a structure of type hostentfor the given host address addrof length lenand address type type

struct hostent{
    char *h_name; /* official name of host */
    char **h_aliases; /* alias list (strings) */
    int h_addrtype; /* host address type (AF_INET) */
    int h_length; /* length of address */
    char **h_addr_list; /* list of addresses (binary in network byte order) */
}

#define h_addrh_addr_list[0] /* for backward compatibility */

struct servent*getservbyname(constchar *name, const char *proto);

• returns a servent structure for the entry from the database that matches the service name using protocol proto.
• if protois NULL, any protocol will be matched.e.g. getservbyname(“echo”, “tcp”);

struct servent *getservbyport(int port, const char *proto);

• returns a servent structure for the entry from the database that matches the service name using port port

struct servent{
    char *s_name; /* official service name */
    char **s_aliases; /* list of alternate names (strings)*/
    int s_port; /* service port number */
    char *s_proto; /* protocol to use (“tcp”or “udp”)*/
}

Compiling and Executing

milo:~/CS556/sockets> gcc-o TCPEchoServerTCPEchoServer.cDieWithError.cHandleTCPClient.c
milo:~/CS556/sockets> gcc-o TCPEchoClientTCPEchoClient.cDieWithError.c
milo:~/CS556/sockets> TCPEchoServer3451 &
[1] 6273
milo:~/CS556/sockets> TCPEchoClient0.0.0.0 hello! 3451
Handling client 127.0.0.1
Received: hello!
milo:~/CS556/sockets> ps
PID TTY TIME CMD
5128 pts/9 00:00:00 tcsh
6273 pts/9 00:00:00 TCPEchoServer
6279 pts/9 00:00:00 ps
milo:~/CS556/sockets> kill 6273
milo:~/CS556/sockets>
[1] Terminated TCPEchoServer3451
milo:~/CS556/sockets>