Data Communications and Networking Socket Programming Part I

Data Communications and Networking Socket Programming Part I: Preliminary Reading: Appendix C - Sockets: A Programmer’s Introduction Data and Computer Communications, 8 th edition By William Stallings Reference: Internetworking With TCP/IP, Volume III: Client-Server Programming And Applications, Windows Socket Version By Douglas E. Comer and David L. Stevens. 1

Learning Outcomes • Knowledge —Explain the principles of network programming —Explain the client-server model —Get familiar with Win. Sock APIs • Skills —Design and develop the client side of client-server network applications using socket programming 2

Outline • Client Server Model • Concept of Socket • Address Structures —Byte order • Win. Sock API • Client Software Design —TCP —UDP 3

Client Server Model • TCP/IP allows two application programs to pass data back and forth —The applications can execute on the same machine or on different machines • How to organize the application programs? —Client-server paradigm is the most popular model • The communication applications are divided into two categories: client and server • Client-server paradigm uses the direction of initiation to categorize whether a program is a client or server. 4

Client Server Model (Cont. ) • An application that initiates the communication is called a client — E. g. , Web browser, FTP client, Telnet client, Email client — The client contacts a server, sends a request, and awaits a response • An application that waits for incoming communication requests from clients is called a server — E. g. , Web server, FTP server, Telnet server, SMTP server — The server receives a client’s request, performs the necessary computation, and returns the result to the client — Remark: A server is usually designed to provide service to multiple clients • How ? ? ? 5

Client Server Model (Cont. ) Remark: Client Programs: Web browsers Because the Web servers and Web browsers are designed based on HTTP, different Web browsers can communicate with all kinds of Web servers. 6

Time Service: An Example • Time service: based on Time Protocol (RFC 868) — http: //www. faqs. org/rfcs/rfc 868. html — The Time service sends back to the originating source the time in seconds since midnight on January first 1900 — Can use TCP or UDP • Time protocol (when using UDP): — — — Server: Listen on port 37. Client: Send an empty datagram to port 37. Server: Receive the empty datagram. Server: Send a datagram containing the time as a 32 bit binary number. Client: Receive the time datagram. 7

Time Service: Server Program #include <stdio. h> #include <stdlib. h> #include <time. h> #include <string. h> #include <stdarg. h> #include <winsock 2. h> #define SA struct sockaddr #define MAXLINE 4096 #define S_PORT 37 #define WINEPOCH 2208988800 int main(int argc, char **argv) { WSADATA wsadata; struct sockaddr_in servaddr, cliaddr; char buff[MAXLINE]; time_t ticks; /* current time */ int n, len; SOCKET s; if (WSAStartup(MAKEWORD(2, 2), &wsadata) != 0) errexit("WSAStartup failedn"); /* Fill up the servaddr */ memset(&servaddr, 0, sizeof(servaddr)); servaddr. sin_family = AF_INET; servaddr. sin_addr. s_addr = INADDR_ANY; servaddr. sin_port = htons(S_PORT); s = socket(AF_INET, SOCK_DGRAM, 0); if (s == INVALID_SOCKET) errexit("cannot create socket: error number %dn", WSAGet. Last. Error()); /* Bind the servaddr to socket s */ if (bind(s, (SA *) &servaddr, sizeof(servaddr)) == SOCKET_ERROR) errexit("can't bind to port %d: error number %dn", S_PORT, WSAGet. Last. Error()); for ( ; ; ) { len = sizeof(SA); if (n = recvfrom(s, buff, sizeof(buff), 0, (SA *)&cliaddr, &len) == SOCKET_ERROR) errexit("recvfrom error: error number %dn", WSAGet. Last. Error()); time(&ticks); /* returns the number of seconds elapsed since midnight of January 1, 1970 */ ticks = htonl((u_long)(ticks + WINEPOCH)); sendto(s, (char *)&ticks, sizeof(ticks), 0, (SA *)&cliaddr, sizeof(cliaddr)); } /* end of for loop */ } 8

Time Service: Client Program #include <stdio. h> #include <stdlib. h> #include <time. h> #include <string. h> #include <stdarg. h> #include <winsock 2. h> #define SA struct sockaddr #define MAXLINE 4096 #define S_PORT 37 #define WINEPOCH 2208988800 int main(int argc, char **argv) { WSADATA wsadata; SOCKET s; int n, len; time_t now; struct sockaddr_in servaddr; /* Must initialize the socket DLL first */ if (WSAStartup(MAKEWORD(2, 2), &wsadata) != 0) errexit("WSAStartup failedn"); errexit("socket error: error number %dn", WSAGet. Last. Error()); memset(&servaddr, 0, sizeof(servaddr)); servaddr. sin_family = AF_INET; servaddr. sin_port = htons(S_PORT); if ( (servaddr. sin_addr. s_addr = inet_addr(argv[1])) == INADDR_NONE) errexit("inet_addr error: error number %dn", WSAGet. Last. Error()); sendto(s, NULL, 0, 0, (SA *)&servaddr, sizeof(servaddr)); len = sizeof(SA); n = recvfrom(s, (char *)&now, sizeof(now), 0, (SA *)&servaddr, &len); if (n == SOCKET_ERROR) errexit("recvfrom failed: error number %dn", WSAGet. Last. Error()); now = ntohl((u_long)now); now -= WINEPOCH; printf("%s", ctime(&now)); if (argc != 2) errexit("usage: daytimeudpcli <IPaddress>"); /* Create a socket, and check the return value */ if ( (s = socket(AF_INET, SOCK_DGRAM, 0)) == INVALID_SOCKET ) WSACleanup(); return 0; } 9

How to Write Network Applications? • TCP/IP is implemented inside the OS • OS exports the TCP/IP functionalities through SOCKET APIs —API: Application Program Interface —A set of functions • Programmers develop network applications by using SOCKET APIs • To learn SOCKET programming —Learn some data structures —Learn the socket APIs —Learn multi-threading 10

Socket Abstraction • “Socket” is an interface between applications and the network services provided by OS • An application sends and receives data through a socket Application SOCKET API TCP/IPv 4 TCP/IPv 6 UNIX 11

Socket Abstraction (Cont. ) Process B Process A data socket 12

Socket Abstraction (Cont. ) Browser A Web Server socket Browser B socket Browser C 13

History of Sockets • In early 1980 s, the Advanced Research Projects Agency (ARPA) funded a group at UC Berkeley to transport TCP/IP software to the UNIX operating system. • As part of the project, the designers created an interface that applications use for network communication. • The interface used an abstraction known as a socket, and the API became known as socket API. • Many computer vendors adopted the Berkeley UNIX operating system, and the socket interface became very popular. • Subsequently, Microsoft chose the socket interface as the primary network API for its operating systems, called Win. Sock. — Old version: Windows Sockets 1. 1 — Today’s version: Windows Sockets 2 • • http: //www. sockets. com/winsock 2. htm The important header file: winsock 2. h • The socket interface has become a de facto standard throughout the computer industry. 14

Specifying a Protocol Interface • The designers at Berkeley wanted to accommodate multiple sets of communication protocols. — TCP/IP was not the only choice at that time! • They allowed for multiple families of protocols, with TCP/IP represented as a single family. E. g. , — PF_INET or AF_INET (IPv 4) • Our focus — PF_INET 6 or AF_INET 6 (IPv 6) — PF_UNIX or AF_UNIX (Unix domain protocols) — PF_IPX or AF_IPX (IPX protocols) 15

Socket Descriptors • To perform file I/O, we use a file descriptor. • Similarly, to perform network I/O, we use a socket descriptor: — Each active socket is identified by its socket descriptor. — The data type of a socket descriptor is SOCKET. — Hack into the header file winsock 2. h: • typedef u_int SOCKET; /* u_int is defined as unsigned int */ — In UNIX systems, socket is just a special file, and socket descriptors are kept in the file descriptor table. — The Windows operating system keeps a separate table of socket descriptors (named socket descriptor table, or SDT) for each process. 16

How to Create a Socket? • The socket API contains a function socket() that can be called to create a socket. E. g. : #include <winsock 2. h> … SOCKET s; … s = socket(AF_INET, SOCK_DGRAM, 0) ; The socket is not usable yet. We need to specify more information before using it for data transmission. 17

Types of Sockets • Under protocol family AF_INET — Stream socket • Uses TCP for connection-oriented reliable communication • Identified by SOCK_STREAM • s = socket(AF_INET, SOCK_STREAM, 0) ; — Datagram socket • Uses UDP for connectionless communication • Identified by SOCK_DGRAM • s = socket(AF_INET, SOCK_DGRAM, 0) ; — RAW socket • Uses IP directly • Identified by SOCK_RAW • Advanced topic. Not covered by COMP 2330. 18

System Data Structures for Sockets • When an application process calls socket(), the operating system allocates a new data structure to hold the information needed for communication, and fills in a new entry in the process’s socket descriptor table (SDT) with a pointer to the data structure. — A process may use multiple sockets at the same time. The socket descriptor table is used to manage the sockets for this process. — Different processes use different SDTs. • The internal data structure for a socket contains many fields, but the system leaves most of them unfilled. The application must make additional procedure calls to fill in the socket data structure before the socket can be used. • The socket is used for data communication between two processes (which may locate at different machines). So the socket data structure should at least contain the address information, e. g. , IP addresses, port numbers, etc. 19

System Data Structures for Sockets (Cont. ) Fig 5. 2 from Comer. Conceptual operating system (Windows) data structures after five calls to socket() by a process. The system keeps a separate socket descriptor table for each process; threads in the process share the table. 20

C Review: Structure • C structures are collections of related variables under one name. struct employee { }; typedef struct employee Employee; char firstname[20]; Employee em; char lastname[20]; Employee *p. Em; /* pointer to structure */ int age; em. salary = 23456; /* dot operator */ float salary; p. Em = &em; /* assign address of em to p. Em */ p. Em->salary += 1000; /* struct pointer operator */ 21

memset() and memcpy() memset(): set buffer to a specified character void *memset(void *dest, int c, size_t count); — dest : Pointer to destination. — c : Character to set. — count : Number of characters memcpy(): copy characters between buffers void *memcpy(void *dest, const void *src, size_t count); — dest : New buffer. — src : Buffer to copy from. — count : Number of bytes to copy. 22

Addressing Issue • How to identify a communication? — Source IP Address, Source Port Number, Destination IP Address, Destination Port Number —TCP or UDP • IPv 4 address: 32 -bit • Port number: 16 -bit —http: //www. iana. org/assignments/port-numbers • Better to have a data structure that contains the addressing information —To simplify the socket APIs —Number of function parameters can be decreased 23

Specifying an Endpoint Address • When a socket is created, it doesn’t contain address information of either local machine or the remote machine. • Before an application uses a socket for sending/receiving data, it must specify one or both of these addresses for the socket. — This can simplify the APIs. • TCP/IP define a communication endpoint to consist of an IP address and a protocol port number. — when you create the socket, you already specified whether this socket is using TCP or UDP. • Several data structures are involved to represent the endpoint address — A general structure: struct sockaddr — A specific structure for IPv 4: struct sockaddr_in 24

Generic Address Structure • The socket system defines a generic endpoint address: — (address family, endpoint address in that family) — The purpose is to provide a common interface for different kinds of protocol families struct sockaddr { u_short sa_family; char sa_data[14]; }; /* address family */ /* up to 14 bytes of direct address */ • Unfortunately, not all address families define endpoints that fit into the sockaddr structure! • To keep program portable and maintainable, TCP/IP code should not use the sockaddr structure in declarations. • Instead, another structure sockaddr_in should be used to represent a socket address in AF_INET family. 25

TCP/IP Endpoint Address • TCP/IP (IPv 4) endpoint address is defined as: (AF_INET, 2 -byte port number, 4 -byte IP address, 8 -byte zero). struct sockaddr_in { short sin_family; u_short sin_port; struct in_addr sin_addr; char sin_zero[8]; }; /* type of address, always AF_INET */ /* protocol port number */ /* IP address */ /* unused */ The IP address is stored in structure sin_addr with type struct in_addr: struct in_addr { union { struct { u_char s_b 1, s_b 2, s_b 3, s_b 4; } S_un_b; struct { u_short s_w 1, s_w 2; } S_un_w; u_long S_addr; } S_un; #define s_addr S_un. S_addr … }; The 32 -bit IP address can be explained as 4 unsigned bytes, or 2 unsigned short, or an unsigned long. By using a union, you can manipulate the IP address conveniently. 26

sockaddr vs. sockaddr_in 2 bytes 14 bytes struct sockaddr_in sa_family sin_family 2 bytes sin_port 2 bytes sin_addr 4 bytes sin_zero 8 bytes sa_data A pointer to a struct sockaddr_in can be cast to a pointer to a struct sockaddr and vice-versa. 27

An Example • To specify an endpoint address 158. 182. 9. 1: 5678 #include <winsock 2. h> … There are several ways to specify an IP address: struct sockaddr_in addr; • Use a string – Good! … • “ 158. 182. 9. 1” • Use an integer – Bad! • 2662729985 addr. sin_family = AF_INET; addr. sin_port = htons(5678); addr. sin_addr. s_addr = htonl(2662729985); • Use DNS service – Good! Remark: 2662729985 = 10011110101101100000100100000001 1 182 158 9 28

Byte Ordering • Byte Ordering: • Memory is organized in bytes. • E. g, a 16 -bit integer takes 2 consecutive bytes; a 32 -bit integer takes 4 consecutive bytes. • Different machines use different host bye orderings: • little-endian: least significant byte located at lower memory • E. g. , Intel 80 x 86 • big-endian: least significant byte located at higher memory • E. g. , Motorola 68000 • These machines may communicate with one another over the network, and they may have different understandings on the same data. • Not a problem for data type char 29

Chaos of Byte Order Big-Endian Computer Send out an integer 169283078 Little-Endian Computer Receive an integer 68032266! 30

Solution: Network Byte Order • Let’s have some regulation! — Network byte order is defined as big-endian byte order — Integers should be sent out in network byte order • The members sin_port and sin_addr in struct sockaddr_in should be in network byte order — sin_port: TCP or UDP port number, put into the TCP/UDP header — sin_addr: IP address, put into the IP header • The member sin_family will not be sent out to the network. So it is not necessary to be in network byte order. • If your application needs to transfer integers between two machines, you should transfer them into network byte order. — You don’t know the type of CPU that executes your program! 31

Network Byte Order • When sending out an integer —Transfer it into network byte order by: —u_short htons (u_short host_short); —u_long htonl (u_long host_long); • After receiving an integer —Transfer it into (local) host byte order by: —u_short ntohs (u_short network_short); —u_long ntohl (u_long network_long); • Remark: —u_short: unsigned 16 -bit integer —u_long: unsigned 32 -bit integer h: host n: network s: short l: long h-to-n-s h-to-n-l n-to-h-s n-to-h-l 32

Win. Sock API Function Name Meaning WSAStartup Initialize the socket library (Windows only) WSACleanup Terminate use of socket library (Windows only) WSAGet. Last. Error Retrieve the specific error code following an unsuccessful socket function call (Windows only) socket Create a socket used for network communication connect Connect to a remote peer (client) closesocket Terminate communication and remove the socket bind Specify a local IP address and protocol port for a socket listen Place the socket in passive mode and set the number of incoming TCP connections the system will enqueue (server) accept Accept the next incoming connection (server) 33

Win. Sock API Function Name Meaning recv Acquire incoming data from a stream connection or the next incoming message (mainly for TCP) recvfrom Receive the next incoming datagram and record its source endpoint address (UDP) send Send outgoing data or a message (mainly for TCP) sendto Send an outgoing datagram to a specified endpoint address (UDP) select Wait until the first of a specified set of sockets becomes ready for I/O shutdown Terminate a TCP connection in one or both directions inet_addr Convert IP address in dotted decimal form to internal binary form inet_ntoa Convert IP address from internal binary form to a text string gethostbyname Map a host name to an IP address 34

A TCP Client-Server Example CLIENT SIDE SERVER SIDE WSAStartup() socket() bind() listen() connect() send() accept( ) recv() send() closesocket() WSACleanup() Remark: The server runs forever. It waits for a new connection on the well-known port, accepts the connection, interacts with the client, and then closes the connection. 35

A UDP Client-Server Example UDP Server UDP Client WSAStartup() socket() sendto() bind() data (request) recvfrom() process request recvfrom() closesocket() WSACleanup() data (reply) sendto() closesocket() WSACleanup() 36

Comparison Between TCP and UDP • Common function calls —WSAStartup(), WSACleanup() —socket(), closesocket() —bind() • Differences —TCP is connection-oriented, UDP is connectionless • TCP applications use listen(), accept(), connect() to setup TCP connections —Data transfer • TCP applications use send() and recv() • UDP applications use sendto() and recvfrom() 37

WSAStartup() Programs using Win. Sock must call WSAStartup() before using sockets. It is needed because the operating system uses dynamically linked libraries (DLLs). int WSAStartup ( WORD w. Version. Requested, LPWSADATA lp. WSAData ); • w. Version. Requested : The highest version of Windows Sockets support that the caller can use. The high order byte specifies the minor version (revision) number; the low-order byte specifies the major version number. • lp. WSAData : A pointer to the WSADATA data structure that is to receive details of the Windows Sockets implementation. WSAStartup() returns zero if successful. Example code: WSADATA wsadata; int err; err = WSAStartup(MAKEWORD(2, 2), &wsadata); if (err != 0) return; 38

WSACleanup() • Once an application finishes using and closing sockets, it calls WSACleanup() to deallocate all data structures and socket bindings. • A program usually calls WSACleanup() only when it is completely finished and ready to exit. int WSACleanup(void); The return value is zero if the operation was successful. Otherwise, the value SOCKET_ERROR is returned. 39

WSAGet. Last. Error() • An application calls the WSAGet. Last. Error function to retrieve the specific error code following an unsuccessful socket function call. • An application should call WSAGet. Last. Error immediately after a socket function returns an error indication. — So that you can know what the problem is. • Windows Sockets Error Codes — http: //msdn. microsoft. com/en-us/library/ms 740668. aspx int WSAGet. Last. Error(void); The return value indicates the error code for this thread's last Windows Sockets operation that failed. 40

socket() The Windows Sockets socket() function creates a socket. SOCKET socket ( int af, int type, int protocol ); • af: An address family specification. • type: A type specification for the new socket. The following are the only two type specifications supported for Windows Sockets 1. 1: SOCK_STREAM and SOCK_DGRAM. In Windows Sockets 2, some new socket types are introduced. • protocol: A particular protocol to be used with the socket that is specific to the indicated address family. Remark: we usually set it as 0. If no error occurs, socket() returns a descriptor referencing the new socket. Otherwise, a value of INVALID_SOCKET is returned. Example code: SOCKET s; s = socket(AF_INET, SOCK_STREAM, 0); if (s == INVALID_SOCKET) { printf(“Error code: %dn”, WSAGet. Last. Error()); return; } 41

closesocket() • • • Once a client or server finishes using a socket, it calls closesocket() to deallocate it. If only one process is using the socket, closesocket() immediately terminates the connection and deallocates the socket. If several processes share a socket, closesocket() decrements a reference count and deallocates the socket when the reference count reaches zero. int closesocket (SOCKET s); If no error occurs, closesocket() returns zero. Otherwise, a value of SOCKET_ERROR is returned. 42

Assigning Endpoint Address At the server side: At the client side: #define S_PORT 13 struct sockaddr_in servaddr; … … memset(&servaddr, 0, sizeof(servaddr)); servaddr. sin_family = AF_INET; servaddr. sin_addr. s_addr = htonl(INADDR_ANY); servaddr. sin_port = htons(S_PORT); servaddr. sin_addr. s_addr = inet_addr(argv[1]); Remark: INADDR_ANY represents a wildcard address that matches any of the computer’s IP address. inet_addr() takes an ASCII string that contains a dotted decimal address and returns the equivalent IP address in binary. In winsock 2. h: #define INADDR_ANY (u_long)0 x 0000 The server doesn’t know who will connect to it. It needs to define the local port number only. argv[1] represents the first argument to the command. The client must provide the IP address and port number of the server. 43

How to Get the Server Address? • The client software needs to know the server’s address. • You can: —Include the address into the source code • Not flexible. If the server’s address is changed, you have to recompile your source code. —Require the user to identify the server when invoking the program • the most popular way 44

inet_addr(): Parsing an Address • The arguments are stored in character strings • E. g. , when user input an IP address “ 128. 10. 2. 3”, your program usually receives just a string. You need to transfer this string into an IP address data structure. unsigned long inet_addr ( const char * cp ); If successful, inet_addr() returns an unsigned long that contains the IPv 4 address in binary form using network byte order. Otherwise, inet_addr() returns INADDR_NONE to indicate that the argument does not contain a valid dotted decimal address. Example code: void main (int argc, char * argv[]) { struct sockaddr_in servaddr; … servaddr. sin_addr. s_addr = inet_addr(argv[1]); } 45

inet_ntoa(): Output an IP Address • char * inet_ntoa(struct in_addr in); • The inet_ntoa() function takes an Internet address structure specified by the in parameter and returns an ASCII string representing the address in dot notation as in "a. b. c. d''. • Good for output! Example code: char *a; struct sockaddr_in servaddr; … a = inet_ntoa(servaddr. sin_addr); printf(“address is : %sn”, a); 46

Looking Up a Domain Name • Internet hosts can be identified by IP address or hostnames. • The Domain Name Service (DNS) provides mapping between hostname and IP address — E. g. : • www. hkbu. edu. hk 158. 182. 4. 1 • www. google. com 66. 249. 89. 104 • How can we use DNS service? — To enjoy DNS services, you must keep the IP address of a default DNS server (or more backups), configured by hand or by DHCP. • Check this out by command “ipconfig /all”. — Every time you need to map a domain name into its corresponding IP address, you call the function gethostbyname(). • It will trigger the OS to contact the DNS server to obtain the IP address. 47

gethostbyname() • • Sometimes one domain name are mapped to a list of IP addresses (e. g. , for a cluster server). Instead of returning a single IP address, it will return a complicated data structure “hostent” which can store a list of IP addresses. struct hostent { char * h_name; /* official host name */ char ** h_aliases; /* other aliases */ short h_addrtype; /* address type */ short h_length; /* length of each address in bytes */ char ** h_addr_list; /* list of addresses */ #define h_addr_list[0] }; struct hostent * gethostbyname( const char* name); Example code: char* server = “www. comp. hkbu. edu. hk”; 1. It will return a NULL pointer struct hostent *h; if error occurs. … 2. The returned IP address is in network byte order. if ( h = gethostbyname(server) ) memcpy(&servaddr. sin_addr, h->h_length); 48

getservbyname() Most client programs must look up the protocol port for the specific service they wish to invoke. To do so, the client invokes library function getservbyname() which takes two parameters: (1) a string that specifies the desired service; (2) a string that specifies the transport protocol being used (i. e. , tcp or udp). struct servent { char * s_name; char ** s_aliases; short s_port; char * s_proto; }; /* official service name */ /* other aliases */ /* port for this service */ /* protocol to use */ struct servent getservbyname ( const char * name, const char * proto ); Example code: struct servent *sptr; if ( sptr = getservbyname (“smtp”, “tcp”) ) { /* port number is now in sptr->s_port */ } else { /* error occurred – handle it */ } Another similar library function is getprotobyname(). Check it out by yourself. 49

bind() • When a socket is first created, it has no associated endpoint addresses. • Function bind() can specify the local endpoint address for a socket. — Usually, a server calls bind() to specify the well-known port at which they will await connections. — A client can skip bind() and let OS to decide. int bind(SOCKET s, const struct sockaddr * addr, int addrlen); • s: the socket descriptor • addr: a pointer to the address assigned to the socket • addrlen: the size (in bytes) of the real address structure If no error occurs, bind() returns 0. Otherwise, it returns SOCKET_ERROR. Example code: SOCKET s; struct sockaddr_in servaddr; /* fill in the fields of servaddr */ … if ( bind(s, (struct sockaddr*)&servaddr, sizeof(servaddr)) == SOCKET_ERROR ) { /* error handling */ } 50

Choosing Local Address • Local port number — A server operates at a well-known protocol port address, which all clients must know. — In general, the client does not care which port it uses as long as the port has no conflict and the port is not assigned to a well-known service. — For TCP connection, the local port is assigned by the OS during the call of connect() function. • Local IP address — It’s trivial for most machines that has only one IP address. — But for a machine with multiple IP addresses, it could be more complicated. • TCP client software usually leaves the local endpoint address unfilled, and allows TCP/IP module to select the correct local IP address and an unused local port number automatically. 51

Iterative TCP Server Algorithm 1. Create a socket and bind to the well-known address for the service being offered. • 2. Place the socket in passive mode, making it ready for use by a server. • 3. accept() Repeatedly receive a request from the client, formulate a response, and send a reply back to the client according to the application protocol. • 5. listen() Accept the next connection request from the socket, and obtain a new socket for the connection. • 4. socket() and bind() recv() and send() When finished with a particular client, close the connection and return to step 3 to accept a new connection. • closesocket() 52

listen() • TCP server calls listen() to place a socket in passive mode and makes it ready to accept incoming connections. — A socket in passive mode is used for acceptance of TCP connection request, not for data transfer. • To handle multiple clients for the server, listen() tells the operating system to queue the connection requests for the server’s socket. int listen(SOCKET s, int backlog); • s: the socket descriptor • backlog: the maximum length of the queue of pending connections If no error occurs, listen() returns 0. Otherwise, it returns SOCKET_ERROR. Example code: SOCKET listenfd, connfd; struct sockaddr_in servaddr; … bind(listenfd, (struct sockaddr*)&servaddr, sizeof(servaddr)); if ( listen(listenfd, 10) == SOCKET_ERROR) { /* error handling */ } 53

accept() • • After the server calls listen(), the server can then call accept() to extract the next incoming connection request from the listening queue. Function accept() creates a new socket for each new successful connection request, and returns the descriptor of the new socket to its caller. This new socket is used to transfer data. SOCKET accept(SOCKET s, struct sockaddr * addr, int *addrlen); • s: the socket descriptor • addr: an optional pointer to a buffer that receives the client’s address • addrlen: an optional pointer to an integer that contains the length of the client’s address structure Return value: If no error occurs, accept() returns the socket descriptor for the newly accepted socket. Otherwise, a value of INVALID_SOCKET is returned. 54

accept() Example code: SOCKET listenfd, connfd; struct sockaddr_in cliaddr; int clilen; … listenfd = socket(AF_INET, SOCK_STREAM, 0); … bind(listenfd, (struct sockaddr*)&servaddr, sizeof(servaddr)); listen(listenfd, 10); connfd = accept(listenfd, (struct sockaddr *)&cliaddr, &clilen); Remark: listenfd is used to accept client connection request. connfd is used for sending/receiving data. 55

Server: Two Kinds of Sockets Server Process Client Process conne ct() Listening socket Server Process Client Process data socket data A new socket returned by accept() 56

Server: Two Kinds of Sockets Browser A Web Server Browser B socket Remark: socket Browser C A single listening socket is used by the server to accept incoming connection requests. 57

TCP Client Algorithm 1. 2. Find the IP address and protocol port number of the server with which communication is desired; Allocate a socket; • 3. Specify that the connection needs an arbitrary, unused protocol port on the local machine, and allow TCP to choose one; • 4. connect() Communicate with the server using the application-level protocol; • 6. A side-effect of connect() Connect the socket to the server; • 5. socket() send() and recv() Close the connection. • closesocket() 58

connect() • A client calls connect() to establish an active connection to a remote server. int connect (SOCKET s, struct sockaddr * addr, int addrlen); • s: the socket descriptor • addr: a pointer to the server’s address • addrlen: the length of the server’s address structure If no error occurs, connect() returns 0. Otherwise, it returns SOCKET_ERROR. Example code: Remark: SOCKET s; servaddr is declared as type struct sockaddr_in servaddr; sockaddr_in. … It is cast into type struct sockaddr s = socket(AF_INET, SOCK_STREAM, 0); when calling connect(). /* fill in the structure servaddr */ … if ( connect(s, (struct sockaddr *)&servaddr, sizeof(servaddr)) == SOCKET_ERROR ) { /* error handling */ } 59 …

TCP Communications • When TCP is used, TCP module uses a sending buffer and a receiving buffer for reliable data transmission. — These buffers are inside the OS kernel and different from your application memory buffer! • For sending out data, send() is used. — Usually when send() returns, all the data have been put into the send buffer. We can assume that these data will be successfully delivered eventually. • For receiving data, recv() is used. — But when recv() returns, the number of received bytes could be different from what you expect (i. e. , different from what the other side has sent). — You must be careful when using recv(). You need to check how many bytes of data have been actually received and then decide what to do next. 60

TCP: Sending Data • For a TCP connection (SOCK_STREAM), both client and server call send() to send data and recv() to receive data. int send(SOCKET s, const char * buf, int len, int flags); • s: the socket descriptor • buf: a pointer to the message to be sent out • len: the length of the message in bytes If no error occurs, send() returns the total number of bytes sent. Otherwise, a value of SOCKET_ERROR is returned. • flags: can be 0 or MSG_OOB or MSG_DONTROUTE Remark: Each TCP socket has a send buffer in the OS. When we call send(), the OS copies all the data from the application buffer to the socket send buffer. And the TCP module in the OS kernel will handle the data transfer on the background. 1. If the send buffer has enough room, send() will return just after the data copying. 2. If the send buffer has no enough room, send() will block itself until all the data are copied into the send buffer. 61

TCP: Receiving Data int recv(SOCKET s, char * buff, int len, int flags); • s: the socket descriptor • buf: a pointer to the buffer to receive incoming data • len: the size of the buffer • flags: can be 0 or MSG_OOB or MSG_PEEK Each TCP socket also has a receive buffer in the OS to hold received data until it is read by the application (by calling the recv() function). So before calling recv(), an application buffer must be allocated by the user to store the incoming data. Function recv() will copy the incoming data from the OS kernel to the application buffer. 1. If no data has arrived, recv() will block until data arrives; 2. If the incoming data can fit into the application buffer, recv() extracts all the data and returns the number of bytes received; 3. If the incoming data cannot fit into the application buffer, recv() only extracts enough to fill the buffer. It may be necessary to call recv() several times in order to receive the entire message. 4. If the other side closed the socket, recv() will return 0. 62

Challenges • TCP allows reliable data transmission between two processes —Data are regarded as byte stream. —There is no data boundary. • Problems —The receiver doesn’t know how much data to receive. —The number of bytes returned by recv() cannot be known in advance. • Multiple calls of recv() is a common practice. But you don’t know how many times to call recv()! • Solutions? 63

Example 1: If Length is Known In this example, we assume the sender will send 20000 bytes to the receiver. But the receiving application may receive data in small chunks. /* receiver code */ #define BLEN 20000 char buf[BLEN]; int n, c=0; buf &buf[n] n += c for (n = 0; n < BLEN; n += c) { c = recv(s, &buf[n], BLEN-n, 0); } Question: in reality, how can the client know the length of the incoming data, i. e. , 20000 bytes? 64

Example 2: Connection Closed by Your Peer In this example, we assume the sender will send some bytes (no more than 20000) to the receiver and then close the socket. The receiver will keep on receiving data until the recv() returns 0. /* receiver code */ #define BLEN 20000 char buf[BLEN]; char *bptr; int n, buflen; bptr = buf; buflen = BLEN; n = recv(s, bptr, buflen, 0); while ((n != SOCKET_ERROR) && (n != 0)) { bptr += n; buflen -= n; n = recv(s, bptr, buflen, 0); } buf bptr n bptr 65

Example 3: A General Solution Data Encapsulation • The sender sends length + data to the receiver. — Usually it is enough to use a 32 -bit integer for length — Remember to use Network Byte Order for the integer! • The receiver extracts length, then allocates memory, then receives the data (see Example 1). /* sender code */ SOCKET sock; int len, m; char *buf; … len = 20000; buf = malloc(len); … /* fill the buf by your data */ m = htonl(len); send(sock, &m, sizeof(m), 0); send(sock, buf, len, 0); … /* receiver code */ SOCKET sock; int i, c, n, len; char *buf, *ptr; … ptr = &n; len = sizeof(n); for (i = 0; i < len; i += c) c = recv(sock, &ptr[i], len-i, 0); len = ntohl(n); buf = malloc(len); for (i = 0; i < len; i += c) c = recv(sock, &buf[i], len-i, 0); 66

UDP Client Algorithm 1. Find the IP address and protocol port number of the server with which communication is desired; 2. Allocate a socket; 3. Specify that the communication needs an arbitrary, unused protocol port on the local machine, and allow UDP to choose one; 4. Specify the server to which messages must be sent; 5. Communicate with the server using the application-level protocol; • This usually involves sending requests and awaiting replies 6. Close the socket. 67

UDP: Sending / Receiving Data • For connectionless service, i. e. , UDP (SOCK_DGRAM), sendto() and recvfrom() are normally used for data transfer. int sendto(SOCKET s, const char * buff, int buflen, int flags, const struct sockaddr * to, int tolen) sendto() requires the destination endpoint address as an argument. Why send() doesn’t? sendto() returns the number of bytes sent if successful, and SOCKET_ERROR to indicate that an error has occurred. int recvfrom(SOCKET s, char * buff, int buflen, int flags, struct sockaddr * from, int * fromlen) recvfrom() extracts the next message that arrives at a socket and records the sender’s address. recvfrom() returns the number of bytes in the message if successful, and SOCKET_ERROR to indicate that an error has occurred. The receiver doesn’t know who will send him data. Any machine can send data to this receiver if it knows the receiver’s IP and UDP port. Please study the example codes on Page 7 and 8. 68

“Connected” UDP • UDP is connectionless. — When call sendto() and recvfrom(), the peer’s address is included or returned as a parameter. — An unconnected UDP socket can receive packets from any machine. • Somehow, there is a “connected” UDP socket — You can call connect() to store the destination’s address in the OS kernel at the beginning — We can no longer specify the destination’s address — We can call send() to send a message and recv() to receive a response. Each call to recv() returns one complete message. — Connected UDP socket is still UDP. The only difference with unconnected UDP socket is the including of destination address information. With this information, the OS will help you to filter out the messages from unknown peers. 69

More about UDP Programming • Reliability issue — A client must implement reliability through timeout and retransmission, if reliability is desired. — Even if your application doesn’t require reliability, it is important to use timeout control because recvfrom() will block forever if the expected incoming packet is lost. • Flow control issue — UDP module has no flow control (but TCP has). — A high performance PC can send out tens of thousands of packets per second with a 100 Mbps Ethernet. But a very slow PC may be swallowed by so huge UDP traffic! If the slow PC’s CPU is too busy, the UDP receive buffer will be accumulated to full and the new incoming UDP packets will be dropped. 70

More References • Learn by practicing! — Some example codes are available on the course web site. — Use Wire. Shark to learn basic socket programming debugging skills. — Start with some toy programs. — Finish the course project individually. • References: — Douglas E. Comer and David L. Stevens, Internetworking with TCP/IP, Volume III: client-server programming and applications, 1997, Prentice-Hall. — W. Richard Stevens, Bill Fenner, and Andrew M. Rudoff, UNIX Network Programming, Volume I: The Sockets Networking API, 3 rd Edition, 2004, Addison-Wesley. — Winsock Programming FAQ: http: //www. tangentsoft. net/wskfaq/ 71