Network sockets

Quick overview of Linux network sockets API

1. Sockets types Raw socket Stream socket (SOCK_STREAM) Datagram socket (SOCK_DGRAM)

2. Data types struct addrinfo struct sockaddr struct sockaddr_in struct in_addr struct sockaddr_storage

3. Helper functions inet_pton() - convert address from str to int inet_ntop() - convert address from int to str

4. Address binding Local addr.port Remote addr.port Description *.port *.* All local interfaces laddr.lport *.* Specific local interface laddr.lport raddr.rport One client

5. Server init getaddrinfo () - convert hostname to ip address socket () - create a socket bind () - associate with a port to listen on listen () - marks the socket referred as a passive socket and setups the number of pending connections accept () - block waiting on a new connection getnameinfo() - get host and port as strings Note: Use setsockopt with SO_REUSEADDR to steal the port

6. Client init getaddrinfo() - convert hostname to ip address socket() - create a socket connect() - connect to a remote port (optionally)

7. Communication over TCP sockets send() - returns number of sent bytes or -1 recv() - returns number of received bytes or 0 (closed conn on a remote side) or -1

8. Communication over UDP sockets sendto() - returns number of sent bytes or -1 recvfrom() - returns number of received bytes or -1 OR , if connect() was called, use: send() - returns number of sent bytes or -1 recv() - returns number of received bytes or -1

9. Closing socket close() - close and free shutdown() - close not free with option: SHUT_RD SHUT_RDWR SHUT_WR

10. Get info getpeername() - remote side address gethostname() - local hostname

11. Socket settings Blocking mode - fcntl(O_NONBLOCK)

12. Multiplexing server getaddrinfo () - convert hostname to ip address socket () - create a socket bind () - associate with a port to listen on listen () - marks the socket referred as a passive socket and setups the number of pending connections select/poll/epoll/kqueue - blocks waiting on multiple connections events getnameinfo() - get host and port as strings

13. Multiplexing using select select(num, readfds, writefds, exceptfds, timeout) Manipulate bitmaps of max size 1024 bits using macros FD_SET, FD_CLR, FD_ISSET, FD_ZERO Select returns modifies bitmaps thus applications should refill the interest sets for every call. Thus both application and kernel have to scan entire bitmaps on each call to figure out interest sets and the result sets. Moreover kernel iterates over interest set to find out which file descriptor is ready. Conclusion : Using select is slow

14. Multiplexing using poll Cons App and kernel still have to scan to figure out interest and result fd sets Kernel still have to check which fd of interest are ready as far as there is no state saved. API poll(struct pollfd *fds, int nfds, int timeout) Pros Relies on array of file descriptors instead of bitmaps. Thus no 1024 limit. Interest and result sets are separated, so app does not need to update it each time

15. Multiplexing using epoll in Linux API int epoll_create(int size) - Create interest set int epoll_ctl(int epfd, int op, int fd, struct epoll_event *event) - Add fd int epoll_wait(int epfd, struct epoll_event *events, int maxevents, int timeout) - Wait Two modes: level triggered and edge triggered Pros Stateful, no scan on each time is required Multiple interest sets in one process Cons Can not add multiple descriptors at once

16. Multiplexing using kqueue in FreeBSD API int kqueue(void); int kevent(int kq, const struct kevent *changelist, int nchanges, struct kevent *eventlist, int nevents, const struct timespec *timeout); Pros Stateful, no scan on each time is required Multiple interest sets in one process Can add multiple descriptors at once Can work not only for sockets, but for files, timers, signals, ...

17. Broadcast setsockopt (SO_BROADCAST)

18. References Using network sockets Scalable Event Multiplexing: epoll vs. kqueue The C10K problem Re: Linux's implementation of poll() not scalable? How to use epoll? A complete example in C

19. My blog Learning Network Programming