sendfile ?pos ?len ~fd sock sends mmap-able data from file descriptor fd to socket sock using offset pos and length len. Returns the number of characters actually written.
NOTE: If the returned value is unequal to what was requested (= the initial size of the data by default), the system call may have been interrupted by a signal, the source file may have been truncated during operation, or a timeout may have occurred on the socket during sending. It is currently impossible to find out which of these events actually happened. Calling sendfile several times on the same descriptor that only partially accepted data due to a timeout will eventually lead to the Unix error EAGAIN.
Type for status of SO_BINDTODEVICE socket option. The socket may either restrict the traffic to a given (by name, e.g. "eth0") interface, or do no restriction at all.
Non-portable TCP functionality
type tcp_bool_option =
| TCP_CORK
(*
(Since Linux 2.2) If set, don’t send out partial frames. All queued partial frames are sent when the option is cleared again. This is useful for prepending headers before calling sendfile(2), or for throughput optimization. As currently implemented, there is a 200ms ceiling on the time for which output is corked by TCP_CORK. If this ceiling is reached, queued data is automatically transmitted.
This option should not be used in code intended to be portable.
*)
| TCP_QUICKACK
(*
(Since Linux 2.4.4) Quick ack solves an unfortunate interaction between the delayed acks and the Nagle algorithm (TCP_NODELAY). On fast LANs, the Linux TCP stack quickly reaches a CWND (congestion window) of 1 (Linux interprets this as "1 unacknowledged packet", BSD/Windows and others consider it "1 unacknowledged segment of data").
If Linux determines a connection to be bidirectional, it will delay sending acks, hoping to bundle them with other outgoing data. This can lead to serious connection stalls on, say, a TCP market data connection with one second heartbeats. TCP_QUICKACK can be used to prevent entering this delayed ack state.
This option should not be used in code intended to be portable.
send_nonblocking_no_sigpipe sock ?pos ?len buf tries to do a nonblocking send on socket sock given buffer buf, offset pos and length len. Prevents SIGPIPE, i.e., raises a Unix-error in that case immediately. Returns Some
bytes_written or None if the operation would have blocked.
Raises Invalid_argument if the designated buffer range is invalid. Raises Unix_error on Unix-errors.
send_no_sigpipe sock ?pos ?len buf tries to do a blocking send on socket sock given buffer buf, offset pos and length len. Prevents SIGPIPE, i.e., raises a Unix-error in that case immediately. Returns the number of bytes written.
Raises Invalid_argument if the designated buffer range is invalid. Raises Unix_error on Unix-errors.
sendmsg_nonblocking_no_sigpipe sock ?count iovecs tries to do a nonblocking send on socket sock using count I/O-vectors iovecs. Prevents SIGPIPE, i.e., raises a Unix-error in that case immediately. Returns Some bytes_written or None if the operation would have blocked.
Raises Invalid_argument if the designated ranges are invalid. Raises Unix_error on Unix-errors.
peer_credential fd takes a file descriptor of a unix socket. It returns the pid and real ids of the process on the other side, as described in man 7 socket entry for SO_PEERCRED. This is useful in particular in the presence of pid namespace, as the returned pid will be a pid in the current namespace, not the namespace of the other process.
Raises Unix_error if something goes wrong (file descriptor doesn't satisfy the conditions above, no process on the other side of the socket, etc.).
pr_set_pdeathsig s sets the signal s to be sent to the executing process when its parent dies. NOTE: the parent may have died before or while executing this system call. To make sure that you do not miss this event, you should call getppid to get the parent process id after this system call. If the parent has died, the returned parent PID will be 1, i.e., the init process will have adopted the child. You should then either send the signal to yourself using Unix.kill, or execute an appropriate handler.
Setting the CPU affinity causes a process to only run on the cores chosen. You can find out how many cores a system has in /proc/cpuinfo. This can be useful in two ways: first, it limits a process to a core so that it won't interfere with processes on other cores. Second, you save time by not moving the process back and forth between CPUs, which sometimes invalidates their cache. See man sched_setaffinity for details.
get_terminal_size term returns (rows, cols), the number of rows and columns of the controlling terminal (raises if no controlling terminal), or of the specified file descriptor (useful when writing to stdout, because stdout doesn't have to be the controlling terminal).
Priority.t is what is usually referred to as the "nice" value of a process. It is also known as the "dynamic" priority. It is used with normal (as opposed to real-time) processes that have static priority zero. See Unix.Scheduler.set for setting the static priority.
bind_to_interface fd (Only "eth0") restricts packets from being received/sent on the given file descriptor fd on any interface other than "eth0". Use bind_to_interface fd Any to allow traffic on any interface. The bindings are not cumulative; you may only select one interface, or Any.
Not to be confused with a traditional BSD sockets API bind() call, this Linux-specific socket option (SO_BINDTODEVICE) is used for applications on multi-homed machines with specific security concerns. For similar functionality when using multicast, see Core_unix.mcast_set_ifname.
get_bind_to_interface fd returns the current interface the socket is bound to. It uses getsockopt() with Linux-specific SO_BINDTODEVICE option. Empty string means it is not bound to any specific interface. See man 7 socket for more information.
epoll(): a Linux I/O multiplexer of the same family as select() or poll(). Its main differences are support for Edge- or Level-triggered notifications (we're using Level-triggered to emulate "select") and much better scaling with the number of file descriptors.
See the man pages for a full description of the epoll facility.
Extended attributes are name:value pairs associated with inodes (files, directories, symlinks, etc). They are extensions to the normal attributes which are associated with all inodes in the system (i.e. the 'man 2 stat' data). A complete overview of extended attributes concepts can be found in 'man 5 attr'.