socket.txt

- int listen(int sockfd, int backlog);

  Linux: The  backlog  argument defines the maximum length to which the
  queue of pending connections for sockfd may grow.  If a connection
  request arrives when the queue is full, the client may receive an error
  with an indication of ECONNREFUSED or, if the underlying protocol
  supports retransmission, the request may be ignored so that a later
  reattempt at connection succeeds.

  backlog specifies the queue length for completely established sockets
  waiting to be accepted, instead of the number of incomplete connection
  requests. The maximum length of the queue for incomplete sockets can
  be set using /proc/sys/net/ipv4/tcp_max_syn_backlog.

  If the backlog argument is greater than the value in /proc/sys/net/core/somaxconn,
  then it is silently truncated to that value; the default value in this
  file is 128.  In kernels before 2.4.25, this limit was a hard coded value,
  SOMAXCONN, with the value 128.

  BSD: The backlog argument defines the maximum length the queue of pending
  connections may grow to.  The real maximum queue length will be 1.5 times
  more than the value specified in the backlog argument.  A subsequent
  listen() system call on the listening socket allows the caller to change
  the maximum queue length using a new backlog argument.  If a connection
  request arrives with the queue full the client may receive an error with
  an indication of ECONNREFUSED, or, in the case of TCP, the connection
  will be silently dropped.

  The listen() system call appeared in 4.2BSD.  The ability to configure
  the maximum backlog at run-time, and to use a negative backlog to request
  the maximum allowable value, was introduced in FreeBSD 2.2.

- SO_LINGER (linger) socket option

  This option specifies what should happen when the socket of a type that
  promises reliable delivery still has untransmitted messages when it is
  closed

  struct linger {
      int l_onoff;  /* nonzero to linger on close  */
      int l_linger; /* time to linger (in secs)  */
  };

  l_onoff = 0 (default), then l_linger value is ignored and close returns
  immediately. But if there is any data still remaining in the socket send
  buffer, the system will try to deliver the data to the peer

  l_onoff = nonzero, then close blocks until data is transmitted or the
  l_linger timeout period expires
  a) l_linger = 0, TCP aborts connection, discards any data still remaining
     in the socket send buffer and sends RST to peer. This avoids the
     TCP's TIME_WAIT state
  b) l_linger = nonzero, then kernel will linger when socket is closed. If
     there is any pending data in the socket send buffer, the kernel waits
     until all the data is sent and acknowledged by peer TCP, or the
     linger time expires

  If a socket is set as nonblocking, it will not wait for close to complete
  even if linger time is nonzero

- TIME_WAIT state

  The end that performs active close i.e. the end that sends the first FIN
  goes into TIME_WAIT state. After a FIN packet is sent to the peer and
  after that peers FIN/ACK arrvies and is ACKed, we go into a TIME_WAIT
  state. The duration that the end point remains in this state is 2 x MSL
  (maximum segment lifetime). The reason that the duration of the TIME_WAIT
  state is 2 x MSL is because the maximum amount of time a packet can wander
  around a network is assumed to be MSL seconds. The factor of 2 is for the
  round-trip. The recommended value for MSL is 120 seconds, but Berkeley
  derived implementations normally use 30 seconds instead. This means a
  TIME_WAIT delay is between 1 and 4 minutes.

  For Linux, the TIME_WAIT state duration is 1 minute (net/tcp.h):
  #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
                                    * state, about 60 seconds */

  TIME_WAIT state on client, combined with limited number of ephermeral ports
  available for TCP connections severely limits the rate at which new
  connections to the server can be created. On Linux, by default ephemeral
  ports are in the range of 32768 to 61000:

  $ cat /proc/sys/net/ipv4/ip_local_port_range
  32768 61000

  So with a TIME_WAIT state duration of 1 minute, the maximum sustained rate
  for any client is ~470 new connections per second

- TCP keepalive

  TCP keepalive packet (TCP packet with no data and the ACK flag turned on)
  is used to assert that connection is still up and running. This is useful
  because if the remote peer goes away without closing their connection, the
  keepalive probe will detect this and notice that the connection is broken
  even if there is no traffic on it.

  Imagine, the following scenario: You have a valid TCP connection established
  between two endpoints A and B. B terminates abnormally (think kernel panic
  or unplugging of network cable) without sending anything over the network
  to notify A that connection is broken. A, from its side, is ready to
  receive data, and has no idea that B has gone away. Now B comes back up
  again, and while A knows about a connection with B and still thinks that it
  active, B has no such idea. A tries to send data to B over a dead
  connection, and B replies with an RST packet, causing A to finally close
  the connection. So, without a keepalive probe A would never close the
  connection if it never sent data over it.

- There are four socket functions that pass a socket address structure from
  the process to the kernel - bind, connect, sendmsg and sendto. These
  function are also responsible for passing the length of the sockaddr that
  they are passing (socklen_t).
  There are five socket functions that pass a socket from the kernel to the
  process - accept, recvfrom, recvmsg, getpeername, getsockname. The kernel
  is also responsible for returning the length of the sockaddr struct that
  it returns back to the userspace

  Different sockaddr structs:
  1. sockaddr_in
  2. sockaddr_in6
  3. sockaddr_un

  Special types of in_addr_t
  /* Address to accept any incoming messages */
  #define INADDR_ANY ((in_addr_t) 0x00000000)

  /* Address to send to all hosts */
  #define INADDR_BROADCAST ((in_addr_t) 0xffffffff)

  /* Address indicating an error return */
  #define INADDR_NONE ((in_addr_t) 0xffffffff)