netsys_posix.mli

(* $Id$ *)

(** POSIX-specific system calls missing in the [Unix] module, and
    further API's from POSIX-style operating systems.
 *)

(** {1 Files, Processes, TTYs, Users, Groups} *)

val int_of_file_descr : Unix.file_descr -> int
  (** Return the file descriptor as integer. See also
      {!Netsys.int64_of_file_descr} which works for all OS.
   *)

val file_descr_of_int : int -> Unix.file_descr
  (** Make a file descriptor from an integer *)

external sysconf_open_max : unit -> int = "netsys_sysconf_open_max"
  (** Return the maximum number of open file descriptor per process.
   * It is also ensured that for every file descriptor [fd]:
   * [fd < sysconf_open_max()]
   *)

external get_nonblock : Unix.file_descr -> bool = "netsys_get_nonblock"
  (** Returns whether the nonblock flag is set *)

external fchdir : Unix.file_descr -> unit = "netsys_fchdir"
  (** Set the current directory to the directory referenced by the
      file descriptor
   *)

external fdopendir : Unix.file_descr -> Unix.dir_handle = "netsys_fdopendir"
  (** Make a directory handle from a file descriptor. The descriptor
      is then "owned" by the directory handle, and will be closed by
      [Unix.closedir].

      This function is useful in conjunction with {!Netsys_posix.openat}
      to read directories relative to a parent directory.

      This is a recent addition to the POSIX standard; be prepared to
      get [Invalid_argument] because it is unavailable.
   *)

external realpath : string -> string = "netsys_realpath"
  (** Returns a pathname pointing to the same filesystem object so that
      the pathname does not include "." or ".." or symbolic links.
   *)

(* Process groups, sessions, terminals *)

external getpgid : int -> int = "netsys_getpgid"
  (** Return the process group ID of the process with the passed PID.
   * For the number 0, the process group ID of the current process is
   * returned.
   *)

val getpgrp : unit -> int
  (** Same as [getpgid 0], i.e. returns the process group ID of the
   * current process.
   *)

external setpgid : int -> int -> unit = "netsys_setpgid"
  (** [setpgid pid pgid]: Set the process group ID of the process [pid]
   * to [pgid]. If [pid = 0], the process group ID of the current process
   * is changed. If [pgid = 0], as process group ID the process ID of the
   * process referenced by [pid] is used.
   *
   * It is only possible for a process to join a process group if both
   * belong to the same session.
   *)

val setpgrp : unit -> unit
  (** Same as [setpgid 0 0]: A new process group ID is created, and the
   * current process becomes its sole member.
   *)

external tcgetpgrp : Unix.file_descr -> int = "netsys_tcgetpgrp"
  (** Return the process group ID of the foreground process group of
   * the session associated with the file descriptor, which must be
   * a tty.
   *)

external tcsetpgrp : Unix.file_descr -> int -> unit = "netsys_tcsetpgrp"
  (** Sets the foreground process group ID of the session associated
   * with the file descriptor, which must be a tty.
   *)

external ctermid : unit -> string = "netsys_ctermid"
  (** Returns the name of the controlling tty of the current process 
   * as pathname to a device file
   *)

external ttyname : Unix.file_descr -> string = "netsys_ttyname"
  (** Returns the name of the controlling tty referred to by the
   * file descriptor.
   *)

external getsid : int -> int = "netsys_getsid"
  (** Returns the session ID of the process with the passed PID.
   * For the PID 0, the session ID of the current process is returned.
   *)

val with_tty : (Unix.file_descr -> unit) -> unit
  (** [with_tty f]: Runs [f fd] where [fd] is the terminal of the process.
      If the process does not have a terminal (because it is a daemon) 
      [with_tty] will fail.
   *)

val tty_read_password : ?tty:Unix.file_descr -> string -> string
  (** [tty_read_password prompt]: If [tty] is a terminal, the [prompt]
      is printed, and a password is read from the terminal (echo off).
      If [tty] is not a terminal, no [prompt] is printed, and just a
      line is read from the [tty] descriptor (non-interactive case).

      [tty] defaults to [Unix.stdin]. If this function is used in a
      program where stdin is not redirected, and the program is started
      in a terminal, it will read the password with prompt and 
      echo disabled. If stdin is redirected, it is assumed that the program is
      used in a script, and the password is piped into it.

      Use in conjunction with [with_tty] to ensure that [tty] is
      the terminal even if a redirection is in effect, e.g.
      {[ with_tty (fun tty -> tty_read_password ~tty prompt) ]}

      Raises [Sys.Break] if the user triggers SIGINT (i.e. presses
      CTRL-C) to abort the input of a password.
   *)

external posix_openpt : bool -> Unix.file_descr = "netsys_posix_openpt"
  (** [posix_openpt noctty]: Opens an unused PTY master.

      [noctty]: If true, the descriptor will not become the controlling
      terminal.

      If this function is not provided by the OS, an emulation is used.

      On some OS, System V style PTY's are unavailable (but they get
      rare).
   *)

external grantpt : Unix.file_descr -> unit = "netsys_grantpt"
  (** Grant access to this PTY *)

external unlockpt : Unix.file_descr -> unit = "netsys_unlockpt"
  (** Unlock a PTY master/slave pair *)

external ptsname : Unix.file_descr -> string = "netsys_ptsname"
  (** Get the name of the slave PTY *)

type node_type = 
  | S_IFREG 
  | S_IFCHR of int  (* major + minor *)
  | S_IFBLK of int  (* major + minor *)
  | S_IFIFO
  | S_IFSOCK
      
external mknod : string -> int -> node_type -> unit = "netsys_mknod"
    (** Creates the node with the given permissions and the given type *)


(* Users and groups *)

external setreuid : int -> int -> unit = "netsys_setreuid"
  (** Changes both the real and the effective user ID of the current
   * process.
   *)

external setregid : int -> int -> unit = "netsys_setregid"
  (** Changes both the real and the effective group ID of the current
   * process.
   *)

external initgroups : string -> int -> unit = "netsys_initgroups"
  (** See initgroups(3). This is a non-POSIX function but widely
      available.
   *)


(** {1 The "at" variants of system calls} *)

(** Note that a few "at" calls have been omitted because the same
    functionality can be achieved by first opening the file with
    [openat] and then by using a function that references the file
    by descriptor. An example for this is [fstatat]: After the
    [openat] call one can use [fstat] to get the stat record of the
    file.
 *)

val have_at : unit -> bool
  (** Whether the [*at] functions are available (they were only recently
      standardized and cannot be expected on all OS yet)
   *)

val at_fdcwd : Unix.file_descr
  (** Pseudo descriptor value to be used as first argument of [*at]
      functions
   *)

type at_flag = AT_EACCESS | AT_SYMLINK_NOFOLLOW | AT_SYMLINK_FOLLOW |
               AT_REMOVEDIR
  (** Flags one can pass to "at" functions. Not all functions support
      all flags
   *)

val openat : Unix.file_descr -> string -> Unix.open_flag list -> 
             Unix.file_perm -> 
                Unix.file_descr
  (** Same as [Unix.openfile] but open relative to the directory given
      by first argument
   *)

val faccessat : Unix.file_descr -> string -> Unix.access_permission list ->
                at_flag list ->
                  unit
  (** Same as [Unix.access] but the file is taken relative to the directory
      given by first argument
   *)

val mkdirat : Unix.file_descr -> string -> int -> unit
  (** Same as [Unix.mkdir] but the file is taken relative to the directory
      given by first argument
   *)
 
val renameat : Unix.file_descr -> string -> Unix.file_descr -> string -> unit
  (** [renameat olddirfd oldpath newdirfd newpath] *)

val linkat : Unix.file_descr -> string -> Unix.file_descr -> string ->
             at_flag list -> unit
  (** [linkat olddirfd oldpath newdirfd newpath flags] *)

val unlinkat : Unix.file_descr -> string -> at_flag list -> unit
  (** Same as [Unix.unlink] but unlink the file relative to the directory
      given by first argument
   *)

val symlinkat : string -> Unix.file_descr -> string -> unit
  (** [symlinkat oldpath newdirfd newpath flags] *)

val mkfifoat : Unix.file_descr -> string -> int -> unit
  (** [mkfifoat dirfd path mode]

      NB. MacOS 10.10 doesn't support mkfifoat although the other "at" functions
      are implemented. Be prepared to get [Invalid_argument].
   *)

val readlinkat : Unix.file_descr -> string -> string
  (** [readlinkat dirfd path] *)

(* TODO: futimens *)

(** {1 File descriptor polling} *)

type poll_array
  (** The array of [poll_cell] entries *)

type poll_req_events
type poll_act_events
  (** Poll events. [poll_req_events] is used to request that certain
      event types are observed. [poll_act_event] shows which 
      event types are actually possible
   *)

type poll_cell =
    { mutable poll_fd : Unix.file_descr;
      mutable poll_req_events : poll_req_events;
      mutable poll_act_events : poll_act_events;
    }
  (** The poll cell refers to the descriptor [poll_fd]. The [poll_req_events]
      are the events the descriptor is polled for. The [poll_act_events]
      are the actually reported events.
   *)

val have_poll : unit -> bool
  (** Whether there is a native [poll] implementation on this OS *)

val poll_req_events : bool -> bool -> bool -> poll_req_events
  (** [poll_req_events rd wr pri]: Create a set of in events consisting
      of the bits [rd], [wr], and [pri]. [rd] means to poll for 
      input data, [wr] to poll for output data, and [pri] to poll for urgent
      input data.
   *)

val poll_req_triple : poll_req_events -> bool * bool * bool
  (** Looks into a [poll_req_events] value, and returns the triple
      [(rd,wr,pri)].
   *)

val poll_null_events : unit -> poll_act_events
  (** Create an empty set of [poll_act_events], for initilization
      of poll cells.
   *)

val poll_result : poll_act_events -> bool
  (** Look whether there is any event in [poll_out_events] *)

val poll_rd_result : poll_act_events -> bool
val poll_wr_result : poll_act_events -> bool
val poll_pri_result : poll_act_events -> bool
val poll_err_result : poll_act_events -> bool
val poll_hup_result : poll_act_events -> bool
val poll_nval_result : poll_act_events -> bool
  (** Look for the bit in [poll_act_events] and return the status *)

val create_poll_array : int -> poll_array
  (** Create a poll array with the given size. The [poll_fd] member is
      initialized with [Unix.stdin], and the two event members are empty.
   *)

val set_poll_cell : poll_array -> int -> poll_cell -> unit
  (** [set_poll_cell a k c]: Sets the poll cell [k] to [c].
      The index [k] must be in the range from [0] to [N-1] when [N] is the
      length of the poll array.
   *)

val get_poll_cell : poll_array -> int -> poll_cell
  (** [get_poll_cell a k]: Returns the poll cell [k].
      The index [k] must be in the range from [0] to [N-1] when [N] is the
      length of the poll array.
   *)

val blit_poll_array : poll_array -> int -> poll_array -> int -> int -> unit
  (** [blit_poll_array a1 p1 a2 p2 len]: Copies the [len] cells at index [p1]
      from [a1] to [a2] at index [p2].
   *)

val poll_array_length : poll_array -> int
  (** Return the number of cells in the poll array *)

val poll : poll_array -> int -> float -> int
  (** [poll a n tmo]: Poll for the events of the cells 0 to [n-1] of 
      poll array [a], and set the [poll_act_events] member of all cells.
      Wait for at most [tmo] seconds (a negative value means there is
      no timeout). Returns the number of ready file descriptors.

      On platforms without native support for [poll] the function is
      emulated using [Unix.select]. Note, however, that there is a
      performance penalty for the emulation, and that the output
      flags [poll_error_result], [poll_hangup_result], and
      [poll_invalid_result] are not emulated.
   *)

val restarting_poll :
      poll_array -> int -> float -> int
  (** A wrapper around [poll] that handles the [EINTR] condition *)

val poll_single : Unix.file_descr -> bool -> bool -> bool -> float -> bool
  (** [poll_single fd rd wr pri tmo]: Polls a single descriptor for the
      events given by [rd], [wr], and [pri]. In [tmo] the timeout can be
      passed. Returns [true] if one of the requested events is indicated
      for the descriptor. The [EINTR] case is not handled.
   *)


(** Actually, [poll_req_events] and [poll_act_events] are integers that
    are bitmasks of some constants. The following functions allow access to
    this detail. 
 *)

val int_of_req_events : poll_req_events -> int
val int_of_act_events : poll_act_events -> int
val req_events_of_int : int -> poll_req_events
val act_events_of_int : int -> poll_act_events
val const_rd_event : int
val const_wr_event : int
val const_pri_event : int
val const_err_event : int
val const_hup_event : int
val const_nval_event : int



(** {1 Event aggregation} *)

(** Support for "high-speed" poll implementations. Currently, only
    [epoll] for Linux is supported.

    The model exhibited in this API is the smallest common denominator
    of Linux epoll, BSD kqueue, and Solaris ports. The [event_aggregator]
    represents the set of monitored event sources. There is, so far,
    only one source, namely file descriptors, i.e. one can check whether
    a descriptor is readable or writable (like [poll]). The source can
    be added to the [event_aggregator] to monitor the source.

    By calling [poll_event_sources] one can determine sources that
    are currently active (i.e. in signalling state). 

    It is undefined what happens when a file descriptor is closed while
    being member of the aggregator.
 *)

type event_aggregator
type event_source

val have_event_aggregation : unit -> bool
  (** Whether there is an implementation for this OS *)

val create_event_aggregator : bool -> event_aggregator
  (** [create_event_aggregator is_interruptible]: Creates a new aggregator,
      and allocates the required OS resources.

      If [is_interruptible], the aggregator can be interrupted from a
      different thread. See [interrupt_event_aggregator] below.
   *)

val destroy_event_aggregator : event_aggregator -> unit
  (** Frees all OS resources *)

val fd_event_source : Unix.file_descr -> poll_req_events -> event_source
  (** Wraps a file descriptor as event_source, and monitors the
      events in [poll_req_events].

      The [event_source] contains
      state about the relation to the aggregator, and because of this,
      the [event_source] should only be used together with one aggregator
      (at a time).
   *)

val modify_fd_event_source : event_source ->  poll_req_events -> unit
  (** Modifies the set of events monitored at this event source *)

val get_fd_of_event_source : event_source -> Unix.file_descr
  (** Get the file descriptor wrapped by this event source *)

val act_events_of_event_source : event_source -> poll_act_events
  (** Return the actual events of the source. This is updated when
      [poll_event_sources] returns the source.
   *)

val add_event_source : event_aggregator -> event_source ->  unit
  (** Adds the event source to the aggregator *)

val del_event_source : event_aggregator -> event_source ->  unit
  (** Removes the source from the aggregator *)

val interrupt_event_aggregator : event_aggregator -> unit
  (** If [create_event_aggregator] was called with [true] as argument, the
      aggregator is interruptible, and this function interrupts it. The
      effect is that a currently running [poll_event_sources], or, if
      it is not running, the next invocation of [poll_event_sources] 
      returns immediately.

      If the aggregator is not interruptible, this function is a NOP.
   *)

val push_event_updates : event_aggregator -> unit
  (** Pushes all modifications of the sources to the kernel *)

val poll_event_sources : event_aggregator -> float -> event_source list
  (** [poll_event_sources ea tmo]: First, all modifications are pushed
      to the kernel, and polling is set up to get events. If no events
      can currently be delivered, the function waits up to [tmo] seconds
      (or endlessly if negative) for events. The function returns only a
      limited number of events at a time. It is allowed that the function 
      returns fewer events than are currently in signalled state, even
      none.

      Call the function with [tmo=0.0] for non-blocking behavior.

      Note that this is the "level-triggered" behavior: If a source
      remains active it will be reported again by the next [poll_event_sources],
      just as [poll] would do.
   *)

val event_aggregator_fd : event_aggregator -> Unix.file_descr
  (** Returns the underlying file descriptor. It is implementation-defined
      whether this descriptor can also be polled for events. Generally,
      you should run [push_event_updates] before polling from the descriptor.
   *)


(* BSD: kqueue
   Solaris: ports (port_create, port_associate)
 *)


(** {1 Fork helpers} *)

(** Ocamlnet invokes [Unix.fork] at some places to create child processes
    for doing real work. The following functions
    allow it to register a handler that is run in the forked child
    process. Note that this is done by the O'caml code calling [fork],
    and not via the POSIX [atfork()] facility.

    The handler should release OS resources like file descriptors that
    are by default shared with the parent process.

    The handler are not invoked when the only purpose of the [fork] is
    to [exec] a different process.
 *)

(** A [post_fork_handler] is a named function [unit -> unit] *)
class type post_fork_handler =
object
  method name : string
  method run : unit -> unit
end

val register_post_fork_handler : post_fork_handler -> unit
  (** Registers a new post fork handler (MT-Safe) *)

val remove_post_fork_handler : post_fork_handler -> unit
  (** Removes a post fork handler from the registry (MT-Safe) *)

val run_post_fork_handlers : unit -> unit
  (** Runs all post fork handlers. Exceptions are caught and printed to
      stderr.
   *)


(** {1 Fork+exec} *)

(** The following function has some similarity with posix_spawn, but
    is extended to our needs, Only special (although frequent) cases
    are implemented with posix_spawn.
 *)

type wd_spec =
  | Wd_keep
      (** Keep the current working directory in the spawned process *)
  | Wd_chdir of string
      (** Change to this directory in the spawned process *)
  | Wd_fchdir of Unix.file_descr
      (** Change to the directory which has been previously been opened *)

type pg_spec =
  | Pg_keep
      (** The new process will be member of the same process group as
	  this process *)
  | Pg_new_bg_group
      (** A new background process group is created, and the spawned
	  process will be its single member
       *)
  | Pg_new_fg_group
      (** A new foreground process group is created, and the spawned
	  process will be its single member
       *)
  | Pg_join_group of int
      (** The spawned process will be member of this process group *)

type fd_action =
  | Fda_close of Unix.file_descr
      (** Close the descriptor *)
  | Fda_close_ignore of Unix.file_descr
      (** Close the descriptor but ignore [EBADF] errors *)
  | Fda_close_except of bool array
      (** Closes all descriptors except those for which
          [except.(k)] is true where [k = int_of_file_descr fd].
          Descriptors outside the array index range are closed.
       *)
  | Fda_dup2 of Unix.file_descr * Unix.file_descr
      (** Duplicate the first descriptor to the second as [dup2] does *)

type sig_action =
  | Sig_default of int
      (** Resets this signal to default behavior in the spawned process *)
  | Sig_ignore of int
      (** Ignores the signal in the spawned process *)
  | Sig_mask of int list
      (** Set the signal mask in the spawned process *)

val spawn : ?chdir:wd_spec ->
            ?pg:pg_spec ->
            ?fd_actions:fd_action list ->
            ?sig_actions:sig_action list ->
            ?env:string array ->
            ?no_posix_spawn:bool ->
            string -> string array ->
              int
  (** [spawn cmd args]: Fork the process and exec [cmd] which gets the
      arguments [args]. On success, the PID of the new process is returned.
      This function does not wait for the completion of the process; use
      [Unix.waitpid] for this purpose.

      - [chdir]: If set, the new process starts with this working directory
        (this is done before anything else)
      - [pg]: If set, the new process will be a member of this process group
      - [fd_actions]: If set, these descriptor actions are executed 
        sequentially
      - [sig_actions]: If set, these signal actions are executed sequentially
      - [env]: If set, the process gets this environment instead of the
        current one
      - [no_posix_spawn]: If set, the [posix_spawn] family of library
        functions is not used to spawn even if possible, and always a
        [fork/exec] approach is taken. This may be slower, but there is
        normally better error reporting.

      Any exceptions in the subprocess are detected, and reported. However,
      if [Fda_close_ignore] leads to [EBADF] for a descriptor, this error is
      ignored.

      If [pg=Pg_new_fg_group], one should include [Sig_ignore Sys.sigttou]
      in [sig_actions].

      There are two implementations for [spawn]: One calls [fork] and [exec]
      directly, and one uses the [posix_spawn] family of library functions.
      The latter is faster on certain conditions, but this is very OS-specific.
      A number of features are not supported by [posix_spawn] and will force
      that [fork/exec] is used: [Wd_chdir], [Wd_fchdir], [Pg_new_fg_group],
      and [Sig_ignore]. However, note some implementations of [posix_spawn]
      also fall back to [fork/exec] internally for some combinations of flags,
      and it is hard to predict which spawn calls can actually be accelerated.
      The tendency, though, is that recent OS have sped up [posix_spawn]
      so far possible (e.g. by using [vfork] internally, or even by making
      [posix_spawn] a system call).
   *)


(** {1 Notification via file descriptor events} *)

(** Often, it is advantageous to report asynchronous events via
    file descriptors. On Linux, this is available via the [eventfd]
    system call. On other platforms, pipes are used for emulation.

    A [not_event] can have two states: off and on. Initially, the
    [not_event] is off. By signalling it, the state changes to on,
    and the underlying real file descriptor becomes readable.
    By consuming the event, the state is switched back to off.

    Note that a similar API exists for Win32: See {!Netsys_win32.w32_event}.
 *)

type not_event

val create_event : unit -> not_event
  (** Creates a new event file descriptor. *)

external set_nonblock_event : not_event -> unit
  = "netsys_set_nonblock_not_event"
  (** Sets the event fd to non-blocking mode *)

val get_event_fd : not_event -> Unix.file_descr
  (** Returns a duplicate of the underlying file descriptor. This should only 
      be used for one thing: checking whether the desciptor becomes readable.
      As this is a duplicate, the caller has to close the descriptor.
   *)

external set_event : not_event -> unit = "netsys_set_not_event"
  (** Signals the event *)

external wait_event : not_event -> unit = "netsys_wait_not_event"
  (** If the event fd is not signalled, the function blocks until
      it gets signalled, even in non-blocking mode.
   *)

external consume_event : not_event -> unit = "netsys_consume_not_event"
  (** Consumes the event, and switches the event fd to off again.
      If the event fd is not signalled, the function blocks until
      it gets signalled (in blocking mode), or it raises [EAGAIN]
      or [EWOULDBLOCK] (in non-blocking mode).

      This is effectively an atomic "wait-and-reset" operation.
   *)

val destroy_event : not_event -> unit
  (** Releases the OS resources. Note that there can be a hidden second
      file descriptor, so closing the descriptor returned by [get_event_fd]
      is not sufficient.
   *)

val report_signal_as_event : not_event -> int -> unit
  (** [report_signal_as_event ev sig] Installs a new signal handler for
      signal [sig] so that [ev] is signalled when a signal arrives.
   *)

(** {1 Notification queues} *)

(** Unimplemented, but a spec exists. Notification queues are intended
    for forwarding events from C level to OCaml level. Possible uses:
    - POSIX timers
    - Realtime signals
    - Subprocess monitoring
    - AIO completion
 *)

(*
(** This is a helper data structure only. This type of queue is a FIFO
    implemented in C. When the queue is filled with data, a notification
    mechanism is triggered to inform user code. Note that the notification
    only happens when the first element is added to an empty queue, but not when
    more elements are added. Also note that there can only be one notification
    mechanism.

    Only C code can add new elements!
 *)
 *)

(*
type 'a not_queue

val create_nqueue : unit -> 'a not_queue
  (** create a new notification queue *)

val nqueue_length : 'a not_queue -> int
  (** returns the number of elements in the queue *)

val nqueue_take : 'a not_queue -> 'a
  (** takes the front element off the queue and returns it.
      Raises [Not_found] if the queue is empty
   *)

val nqueue_reset_notification : 'a not_queue -> unit
  (** Do not notify *)

val nqueue_notify_via_event : 'a not_queue -> not_event -> unit
  (** Arranges that the event is signalled when the first element is added
      to the queue
   *)

val nqueue_notify_via_condition : 'a not_queue -> Condition.t -> unit
  (** Arranges that the condition variable is signalled  when the first 
      element is added to the queue
   *)

(** Another notification mechanism is described in {!Netsys_posix.sem_not}. *)
 *)

(** {1 Subprocesses and signals} *)

(** Watching subprocesses requires that the right signal handler is
    installed: [install_subprocess_handler]
 *)

type watched_subprocess

val watch_subprocess : int -> int -> bool -> 
                          Unix.file_descr * watched_subprocess
  (** [let fd, ws = watch_subprocess pid pgid kill_flag]: 
      Enters the subprocess [pid]
      into the watch list. If [pgid > 0], the process group ID is
      [pgid] (for [killpg_subprocess] and [killpg_all_subprocesses]).
      The [kill_flag] controls the process selection of
      [kill_all_subprocesses] and [killpg_all_subprocesses].

      The returned descriptor [fd] is open for reading and
      will indicate EOF when the subprocess is terminated. Via [ws]
      it is possible to query information about the subprocess. The
      installed signal handler will [wait] for the subprocess and
      put the process status into [ws].

      The caller has to close [fd] after the termination is signaled.
   *)

val ignore_subprocess : watched_subprocess -> unit
  (** Changes the arrangement so that the termination of the subprocess
      is no longer reported by the file descriptor. The file descriptor
      indicates EOF immediately (and can be closed by the caller).
      Nevertheless, the signal handler still [wait]s for the subprocess
      to avoid zombies.

      Any further access to [ws] will fail.
   *)
     
val forget_subprocess : watched_subprocess -> unit
  (** Frees OS resources. Any further access to the [ws] will fail. *)

val get_subprocess_status : watched_subprocess -> Unix.process_status option
  (** If the subprocess is terminated, this function returns the status.
      Otherwise [None] is returned
   *)

val kill_subprocess : int -> watched_subprocess -> unit
  (** Sends this signal to the subprocess if this process still exists.
      Never throws an exception.
   *)

val killpg_subprocess : int -> watched_subprocess -> unit
  (** Sends this signal to the process group of the subprocess if there
      is still a watched subprocess belonging to this group.
      Never throws an exception.
   *)

val kill_all_subprocesses : int -> bool -> bool -> unit
  (** [kill_all_subprocess signal override nogroup]: 
      Sends a signal to potentially
      all subprocesses. The signal is sent to a watched process if the process
      still exists, and these two conditions hold both:
      - [not nogroup || pgid = 0]: Processes with [pgid > 0] are excluded
        if [nogroup] is set
      - [kill_flag || override]: A process needs to have
        [kill_flag] set, or [override] is specified

      Never throws an exception if the signal handler is installed.
   *)

val killpg_all_subprocesses : int -> bool -> unit
  (** [killpg_all_subprocess signal override]: Sends a signal to potentially
      all subprocesses belonging to a process group (i.e. [pgid>0]).
    . The signal is sent to a process group if there are still watched
      subprocesses
      belonging to the group, and if either the [kill_flag] of any of the 
      subprocesses process was set to [true], or [override] is [true].

      Never throws an exception if the signal handler is installed.
   *)


val install_subprocess_handler : unit -> unit
  (** Installs a SIGCHLD handler for watching subprocesses. Note that only
      processes are [wait]ed for that are registered with 
      [watch_subprocess].

      The handler works both in the single-threaded and the multi-threaded
      case. [install_subprocess_handler] can safely called several times.
      The handler is installed every time the function is called, but the
      required data structures are only initialized at the first call.
   *)

val register_subprocess_handler : unit -> unit
  (** Uses the {!Netsys_signal} framework to manage the installation of
      the SIGCHLD handler.

      This is the preferred method of installing the SIGCHLD handler.
   *)


  (** {b Further notes.} *)

  (** The subprocess handler and [fork()]: The subprocess handler uses
      pipes for notification, and because of this it is sensitive to
      unpredicted duplicates of the pipe descriptors. [fork()] duplicates
      these pipe descriptors. If nothing is done about this issue, it
      can happen that the notification does not work anymore as it relies
      on detecting closed pipes.

      If [fork()] is immediately followed by [exec()] (as it is done
      to run subcommands), the problem does not occur, because the relevant
      descriptors are closed at [exec()] time.

      If [fork()] is used to start worker processes, however, we have
      to be careful. The descriptors need to be closed, so that the
      parent can continue to monitor subprocesses, and to allow the worker
      processes to use this mechanism. This module defines post fork
      handlers (see above), and a handler is automatically added that
      cleans the descriptors up. All user code has to do is to call
      [run_post_fork_handlers] immediately after [fork()] has spawned
      the new child, from the new child. This completely resets
      everything.
   *)

  (** The subprocess handler and multi-threading: The handler has been
      carefully designed, and works even in multi-threaded programs.
      However, one should know that multi-threading and [fork()] do not
      interact well with each other. Again, the problems do not occur
      if [fork()] is followed by [exec()]. There is no solution for the
      case that worker processes are started with [fork()], though.
      The (very generic) problem is that the state of mutexes and other
      multi-threading primitives is not well-defined after a [fork()].
   *)


(** {1 Syslog} *)

type level = Netlog.level
  (*  [ `Emerg | `Alert | `Crit | `Err | `Warning | `Notice | `Info | `Debug ]
   *)
  (** The log levels *)

type syslog_facility =
    [ `Authpriv
    | `Cron
    | `Daemon
    | `Ftp
    | `Kern
    | `Local0
    | `Local1
    | `Local2
    | `Local3
    | `Local4
    | `Local5
    | `Local6
    | `Local7
    | `Lpr
    | `Mail
    | `News
    | `Syslog
    | `User
    | `Uucp
    | `Default
    ]
  (** The facilities. Only [`User] and [`Local0] to [`Local7] are
      standard POSIX. If a facility is unavailable it is silently
      substituted by [`Local0]. The value [`Default] leaves this unspecified.
   *)

type syslog_option =
    [ `Cons
    | `Ndelay
    | `Odelay
    | `Nowait
    | `Pid
    ]
 (** The syslog options:
     - [`Cons]: Fall back to console logging if syslog is unavailable
     - [`Ndelay]: Open the connection immediately
     - [`Odelay]: Open the connection at the first call [syslog] (default)
     - [`Nowait]: Do not wait until it is ensured that the message is
       sent
     - [`Pid]: Log the PID with every message
  *)

val openlog : string option -> syslog_option list -> syslog_facility -> unit
  (** [openlog ident options facility]: Opens a log stream. [ident] is
      prepended to every message if given (usually the program name).
      The [facility] is the default facility for [syslog] calls.
   *)

val syslog : syslog_facility -> level -> string -> unit
  (** [syslog facility level message]: Logs [message] at [level] for
      [facility]
   *)

val closelog : unit -> unit
  (** Closes the log stream *)

(** Usually, the log stream is redirected to syslog by either:
    - setting [Netlog.current_logger] to [syslog facility], e.g.
      {[ Netlog.current_logger := Netsys_posix.syslog `User ]}
    - using the Netplex class for sending message to syslog (XXX)
 *)

(** {1 Sync} *)

external fsync : Unix.file_descr -> unit = "netsys_fsync"
  (** Sync data and metadata to disk *)
  
external fdatasync : Unix.file_descr -> unit = "netsys_fdatasync"
  (** Syncs only data to disk. If this is not implemented, same effect
      as [fsync]
   *)

(** {1 Sending file descriptors over Unix domain sockets} *)

(** These functions can be used to send file descriptors from one process
    to another one. The descriptor [sock] must be a connected 
    Unix domain socket.

    The functionality backing this is non-standard but widely available.

    {b Not yet implemented, but spec exists.}
 *)

(*
val have_scm_rights : unit -> bool
  (** Whether this functionality is available *)

val send_fd : Unix.file_descr -> Unix.file_descr -> unit
  (** [send_fd sock fd]: Sends [fd] via [sock] as ancillary message.
      Also sends a single byte 'X' over the main message channel.
   *)

val receive_fd : Unix.file_descr -> Unix.file_descr
  (** [receive_fd sock]: Receives a single byte over the main message
      channel, and checks whether a file descriptor accompanies the
      byte. If so, it is returned. If not, the function will raise [Not_found].
   *)
 *)

(** {1 Optional POSIX functions} *)

external have_fadvise : unit -> bool = "netsys_have_posix_fadvise"
  (** Returns whether the OS supports the fadvise POSIX option *)

type advice =
  | POSIX_FADV_NORMAL
  | POSIX_FADV_SEQUENTIAL
  | POSIX_FADV_RANDOM
  | POSIX_FADV_NOREUSE
  | POSIX_FADV_WILLNEED
  | POSIX_FADV_DONTNEED
  | FADV_NORMAL
  | FADV_SEQUENTIAL
  | FADV_RANDOM
  | FADV_NOREUSE
  | FADV_WILLNEED
  | FADV_DONTNEED
  (** Possible advices for fadvise. The names starting with "POSIX_" and
      the ones lacking the prefix have the same meaning. In new code,
      the names starting with "POSIX_" should be preferred (for better
      compaibility with other libraries).
   *)

external fadvise : Unix.file_descr -> int64 -> int64 -> advice -> unit
                 = "netsys_fadvise"
  (** Advises to load pages into the page table from the file, or to remove
      such pages.
   *)

external have_fallocate : unit -> bool = "netsys_have_posix_fallocate"
  (** Returns whether the OS supports the fallocate POSIX option *)

external fallocate : Unix.file_descr -> int64 -> int64 -> unit
                   = "netsys_fallocate"
  (** Allocate space for the file and the specified file region *)


(** {1 POSIX Shared Memory} *)

external have_posix_shm : unit -> bool = "netsys_have_posix_shm"
  (** Returns whether the OS supports POSIX shared memory *)

type shm_open_flag =
  | SHM_O_RDONLY
  | SHM_O_RDWR
  | SHM_O_CREAT
  | SHM_O_EXCL
  | SHM_O_TRUNC

external shm_open : string -> shm_open_flag list -> int -> Unix.file_descr
  = "netsys_shm_open"
  (** Opens a shared memory object. The first arg is the name of the
    * object. The name must begin with a slash, but there must be no
    * further slash in it (e.g. "/sample"). The second arg are the
    * open flags. The third arg are the permission bits.
    *
    * The open flags are interpreted as follows:
    * - [SHM_O_RDONLY]: Open the object for read access
    * - [SHM_O_RDWR]: Open the object for read-write access
    * - [SHM_O_CREAT]: Create the object if it does not exist
    * - [SHM_O_EXCL]: If [SHM_O_CREAT] was also specified, and a an object
    *   with the given name already exists, return an error
    *   ([Unix.EEXIST]).
    * - [SHM_O_TRUNC]: If the object already exists, truncate it to 
    *   zero bytes
    *
    * One of [SHM_O_RDONLY] or [SHM_O_RDWR] must be given.
    *
    * On success, the function returns a file descriptor representing the
    * object. To access the object, one has to memory-map this file
    * use one of the [map_file] functions in the [Bigarray]
    * module, or in {!Netsys_mem}). Use [Unix.ftruncate] to resize the object.
    *
    * Note that it is unspecified whether this file pops up somewhere
    * in the file system, and if so, where.
    *
    * If a system error occurs, the function raises a [Unix.Unix_error]
    * exception.
   *)

external shm_unlink : string -> unit = "netsys_shm_unlink"
  (** Unlinks the name for a shared memory object *)

val shm_create : string -> int -> Unix.file_descr * string
  (** [let (fd,name) = shm_create prefix size]: Creates an shm object
      with a unique name. The name has the passed [prefix]. The [prefix]
      must start with "/" but must not contain any further "/". The object
      has a length of [size] bytes. The object has a permissions 0o600
      (independent of umask).
   *)


(** {1 POSIX semaphores} *)

val have_named_posix_semaphores : unit -> bool
  (** Returns [true] if named POSIX semaphores are supported on this system *)

val have_anon_posix_semaphores : unit -> bool
  (** Returns [true] if anonymous POSIX semaphores are supported on this 
      system *)

val have_posix_semaphores : unit -> bool
  (** Returns [true] if both kinds of semaphores are supported on this system *)

(** {b Constants.} *)

val sem_value_max : int
  (** The maximum value of a semaphore, but at most [max_int] *)

val sem_size : int
  (** The size of an anonymous semaphore in bytes ([sizeof(sem_t)]) *)

(** {b Types.} *)

type sem_kind = [ `Named | `Anonymous ]

type 'sem_kind semaphore

type named_semaphore = [ `Named ] semaphore
type anon_semaphore = [ `Anonymous ] semaphore

type sem_open_flag =
  | SEM_O_CREAT
  | SEM_O_EXCL

(** {b Named semaphores.} *)

val sem_open : 
  string -> sem_open_flag list -> int -> int -> named_semaphore
  (** [sem_open name flags mode init_value]: Opens a named semaphore
      which is optionally created. Sempahore names usually begin with
      a slash followed by a single name component (not containing a
      further slash).

      Interpretation of [flags]:
      - [SEM_O_CREAT]: The semaphore is created if not yet existing.
        The [mode] and [init_value] are interpreted if the creation
        actually occurs. [mode] is the permission of the semaphore.
        [init_value] is the (non-negative) initial value, up to
        [sem_value_max].
      - [SEM_O_EXCL]: The semaphore is only opened if the semaphore
        does not exist yet. Othwerwise an [EEXIST] error is returned
   *)

val sem_close : named_semaphore -> unit
  (** Closes a named semaphore. Semaphores are also automatically closed
      when the GC finds that the semaphore is unreachable.
   *)

val sem_unlink : string -> unit
  (** Unlinks the semaphore name *)

val sem_create : string -> int -> named_semaphore * string
  (** [let (sem,name) = sem_create prefix init_value]: Creates
      a new semaphore with a unique name. The name has the passed [prefix]. 
      The [prefix] must start with "/" but must not contain any further "/".
      The semaphore is initialized with [init_value]. The object has
      permissions 0o600 (modulo umask).
   *)


(** {b Anonymous semaphores.} *)

val sem_init : Netsys_types.memory -> int -> bool -> int -> 
                 anon_semaphore
  (** [sem_init mem pos pshared init_value]: Initializes the memory
      at position [pos] to [pos + sem_size() - 1] as anonymous semaphore.
      If [pshared] the semaphore is shared between processes. 
      [init_value] is the initial non-negative value (max is 
      [sem_value_max].
   *)

val sem_destroy : anon_semaphore -> unit
  (** Destroys the anonymous semaphore *)

val as_sem : Netsys_types.memory -> int -> anon_semaphore
  (** [as_sem mem pos]: Interprets the memory at position [pos]
      to [pos + sem_size() - 1] as anonymous semaphore.
      The memory region must already have been initialized.
   *)

(** {b Operations.} *)

val sem_getvalue : 'kind semaphore -> int
  (** Returns the value of the semaphore. If the value is bigger than
      what can be represented as [int], an [EINVAL] error is returned.

      The returned value is non-negative - if the underlying POSIX
      function reports a negative value zero is returned instead.

      Unavailable on MacOS.
   *)

val sem_post : 'kind semaphore -> unit
  (** Unlocks the semaphore (increases the value by 1) *)

type sem_wait_behavior =
  | SEM_WAIT_BLOCK
  | SEM_WAIT_NONBLOCK

val sem_wait : 'kind semaphore -> sem_wait_behavior -> unit
  (** Locks the semaphore (decreases the value by 1). If the semaphore
      value is already zero, and [SEM_WAIT_BLOCK] is given, the function
      waits until another process or thread posts. If [SEM_WAIT_NONBLOCK]
      the error [EAGAIN] is returned.

      [sem_wait] may be interrupted by signals.
   *)

(** {2:sem_not Semaphores and notification} *)

(** {b Not yet implemented.} *)

(*
val nqueue_notify_via_sem : 'a not_queue -> _ semaphore -> unit
  (** Arranges that the semaphore is signalled (posted)  when the first 
      element is added to the queue
   *)


val sem_wait_via_event : _ semaphore -> not_event -> unit
  (** Runs [sem_wait] in a thread, and arranges that the [not_event] is 
      signalled when 
      the blocking [sem_wait] succeeds. This function is only available on
      platforms with [pthread] support, because a helper thread is started.
   *)
 *)

(** {1 Locales} *)

type langinfo =
    { nl_CODESET : string;  (** from [LC_CTYPE]: codeset name *)
      nl_D_T_FMT : string;  (** from [LC_TIME]: string for formatting date and time *)
      nl_D_FMT : string;  (** from [LC_TIME]: date format string *)
      nl_T_FMT : string;  (** from [LC_TIME]: time format string *)
      nl_T_FMT_AMPM : string;  (** from [LC_TIME]: a.m. or p.m. time format string *)
      nl_AM_STR : string;  (** from [LC_TIME]: Ante Meridian affix *)
      nl_PM_STR : string;  (** from [LC_TIME]: Post Meridian affix *)
      nl_DAY_1 : string;  (** from [LC_TIME]: name of the first day of the week (for example, Sunday) *)
      nl_DAY_2 : string;  (** from [LC_TIME]: name of the second day of the week (for example, Monday) *)
      nl_DAY_3 : string;  (** from [LC_TIME]: name of the third day of the week (for example, Tuesday) *)
      nl_DAY_4 : string;  (** from [LC_TIME]: name of the fourth day of the week (for example, Wednesday) *)
      nl_DAY_5 : string;  (** from [LC_TIME]: name of the fifth day of the week (for example, Thursday) *)
      nl_DAY_6 : string;  (** from [LC_TIME]: name of the sixth day of the week (for example, Friday) *)
      nl_DAY_7 : string;  (** from [LC_TIME]: name of the seventh day of the week (for example, Saturday) *)
      nl_ABDAY_1 : string;  (** from [LC_TIME]: abbreviated name of the first day of the week *)
      nl_ABDAY_2 : string;  (** from [LC_TIME]: abbreviated name of the second day of the week *)
      nl_ABDAY_3 : string;  (** from [LC_TIME]: abbreviated name of the third day of the week *)
      nl_ABDAY_4 : string;  (** from [LC_TIME]: abbreviated name of the fourth day of the week *)
      nl_ABDAY_5 : string;  (** from [LC_TIME]: abbreviated name of the fifth day of the week *)
      nl_ABDAY_6 : string;  (** from [LC_TIME]: abbreviated name of the sixth day of the week *)
      nl_ABDAY_7 : string;  (** from [LC_TIME]: abbreviated name of the seventh day of the week *)
      nl_MON_1 : string;  (** from [LC_TIME]: name of the first month of the year *)
      nl_MON_2 : string;  (** from [LC_TIME]: name of the second month *)
      nl_MON_3 : string;  (** from [LC_TIME]: name of the third month *)
      nl_MON_4 : string;  (** from [LC_TIME]: name of the fourth month *)
      nl_MON_5 : string;  (** from [LC_TIME]: name of the fifth month *)
      nl_MON_6 : string;  (** from [LC_TIME]: name of the sixth month *)
      nl_MON_7 : string;  (** from [LC_TIME]: name of the seventh month *)
      nl_MON_8 : string;  (** from [LC_TIME]: name of the eighth month *)
      nl_MON_9 : string;  (** from [LC_TIME]: name of the ninth month *)
      nl_MON_10 : string;  (** from [LC_TIME]: name of the tenth month *)
      nl_MON_11 : string;  (** from [LC_TIME]: name of the eleventh month *)
      nl_MON_12 : string;  (** from [LC_TIME]: name of the twelfth month *)
      nl_ABMON_1 : string;  (** from [LC_TIME]: abbreviated name of the first month *)
      nl_ABMON_2 : string;  (** from [LC_TIME]: abbreviated name of the second month *)
      nl_ABMON_3 : string;  (** from [LC_TIME]: abbreviated name of the third month *)
      nl_ABMON_4 : string;  (** from [LC_TIME]: abbreviated name of the fourth month *)
      nl_ABMON_5 : string;  (** from [LC_TIME]: abbreviated name of the fifth month *)
      nl_ABMON_6 : string;  (** from [LC_TIME]: abbreviated name of the sixth month *)
      nl_ABMON_7 : string;  (** from [LC_TIME]: abbreviated name of the seventh month *)
      nl_ABMON_8 : string;  (** from [LC_TIME]: abbreviated name of the eighth month *)
      nl_ABMON_9 : string;  (** from [LC_TIME]: abbreviated name of the ninth month *)
      nl_ABMON_10 : string;  (** from [LC_TIME]: abbreviated name of the tenth month *)
      nl_ABMON_11 : string;  (** from [LC_TIME]: abbreviated name of the eleventh month *)
      nl_ABMON_12 : string;  (** from [LC_TIME]: abbreviated name of the twelfth month *)
      nl_ERA : string;  (** from [LC_TIME]: era description segments *)
      nl_ERA_D_FMT : string;  (** from [LC_TIME]: era date format string *)
      nl_ERA_D_T_FMT : string;  (** from [LC_TIME]: era date and time format string *)
      nl_ERA_T_FMT : string;  (** from [LC_TIME]: era time format string *)
      nl_ALT_DIGITS : string;  (** from [LC_TIME]: alternative symbols for digits *)
      nl_RADIXCHAR : string;  (** from [LC_NUMERIC]: radix character *)
      nl_THOUSEP : string;  (** from [LC_NUMERIC]: separator for thousands *)
      nl_YESEXPR : string;  (** from [LC_MESSAGES]: affirmative response expression *)
      nl_NOEXPR : string;  (** from [LC_MESSAGES]: negative response expression *)
      nl_CRNCYSTR : string;  (** from [LC_MONETARY]: currency  *)
    }

val query_langinfo : string -> langinfo
  (** [query_langinfo locale]: Temporarily sets the passed [locale] and
      determines the language attributes. After that the orignal locale is
      restored. Pass "" as [locale] to get the locale requested in the
      environment.

      The value for "" is cached.
   *)

(** {1 Clocks} *)

(** Support for clocks can be assumed to exist on all current POSIX
    systems.
 *)

type timespec = float * int
    (** [(t,t_nanos)]: Specifies a time by a base time [t] to which the
	nanoseconds [t_nanos] are added.

	If this pair is returned by a function [t] will always be integral.
	If a pair is passed to a function, it does not matter whether this
	is the case or not, but using integral values for [t] ensure
	maximum precision.
     *)

external nanosleep : timespec -> timespec ref -> unit = "netsys_nanosleep"
  (** [nanosleep t t_rem]: Sleeps for [t] seconds. The sleep can either be
      finished, or the sleep can be interrupted by a signal. In the
      latter case, the function will raise [EÍNTR], and write to [t_rem]
      the remaining seconds.
   *)

type clock_id

type clock =
  | CLOCK_REALTIME        (** A clock measuring wallclock time *)
  | CLOCK_MONOTONIC       (** A clock measuring kernel time (non-settable). Optional, i.e. not supported by all OS *)
  | CLOCK_ID of clock_id  (** A clock ID *)

external clock_gettime : clock -> timespec = "netsys_clock_gettime"
  (** Get the time of this clock *)

external clock_settime : clock -> timespec -> unit = "netsys_clock_settime"
  (** Set the time of this clock *)

external clock_getres : clock -> timespec = "netsys_clock_getres"
  (** Get the resolution of this clock *)

external clock_getcpuclockid : int -> clock_id = "netsys_clock_getcpuclockid"
  (** Return the ID of a clock that counts the CPU seconds of the given
      process. Pass the PID or 0 for the current process.

      This function is not supported on all OS.
   *)

(*
val clock_nanosleep : clock -> bool -> timespec -> timespec ref -> unit
  (** [clock_nanosleep c abstime t t_rem]: Uses the clock [c] to time
      a sleep. If [abstime], the function sleeps until [t]. If not
      [abstime], the function sleeps for [t] seconds.

      The sleep can either be
      finished, or the sleep can be interrupted by a signal. In the
      latter case, the function will raise [EÍNTR]. If also [abstime] is
      not specified, it writes to [t_rem] the remaining seconds.

      This function is not supported on all OS.
   *)
 *)

(** {1 POSIX timers} *)

type posix_timer

type timer_expiration =
  | TEXP_NONE
  | TEXP_EVENT of not_event
  | TEXP_EVENT_CREATE
  | TEXP_SIGNAL of int

(* Future: 
   TEXP_NQ of ??? not_queue
   TEXP_THREAD of ??? -> unit
   TEXP_EVENT_KQUEUE: Like TEXP_EVENT_CREATE, but backed by a kqueue (BSD),
      and with restrictions (only CLOCK_MONOTONIC, only milliseconds prevision,
      only oneshot timers)
 *)

val have_posix_timer : unit -> bool

val timer_create : clock -> timer_expiration -> posix_timer
  (** Create a new timer that will report expiration as given by the arg:
      - [TEXP_NONE]: no notification
      - [TEXP_EVENT e]: the [not_event] [e] is signalled
      - [TEXP_EVENT_CREATE]: a special [not_event] is created for the timer.
        (Get the event via [timer_event], see below.)
      - [TEXP_SIGNAL n]: the signal [n] is sent to the process

      Note that [TEXP_EVENT_CREATE] is much faster on Linux than
      [TEXP_EVENT], because it can be avoided to start a new thread
      whenever the timer expires. Instead, the timerfd machinery is used.

      [TEXP_EVENT] and [TEXP_EVENT_CREATE] are only supported on systems
      with pthreads.
   *)

val timer_settime : posix_timer -> bool -> timespec -> timespec -> unit
  (** [timer_settime tm abstime interval value]:

      If [value=(0.0,0)], the timer is stopped.

      If [value] is a positive time, the timer is started (or the timeout
      is changed if it is already started).  If [abstime], [value] is
      interpreted as the absolute point in time of the expiration.
      Otherwise [value] sets the number of seconds until the expiration.

      If [interval] is positive, the started timer will repeat to expire
      after this many seconds once it has expired for the first time.
      If [interval=(0.0,0)], the timer is a one-shot timer.
   *)

val timer_gettime : posix_timer -> timespec
  (** Returns the number of seconds until the expiration, or [(0.0,0)]
      if the timer is off
   *)

val timer_delete : posix_timer -> unit
  (** Deletes the timer *)

val timer_event : posix_timer -> not_event
  (** Returns the notification event for the timer styles [TEXP_EVENT] and
      [TEXP_EVENT_CREATE].

      Note that the latter type of event does not allow to call [set_event].
   *)

(** Intentionally there is no wrapper for [timer_getoverrun].
    Additional overruns can occur because of the further processing
    of the notifications: The OCaml runtime can merge signals,
    which would not be noticed by the kernel overrun counter,
    and events can also be merged. The workaround is to use one-shot timers
    with absolute expiration timestamps, and to check for overruns
    manually. Once we have [TEXP_NQ] the issue is solved.
 *)


(** {1 Linux I/O Priorities} *)

(** These system calls are only available on Linux since kernel 2.6.13,
    and not even on every architecture. i386, x86_64, ia64, and PPC are
    known to work. 

    Per-process I/O priorities are currently only supported by the
    CFQ I/O scheduler.
 *)

val have_ioprio : unit -> bool
  (** Returns [true] if the system call [ioprio_get] is supported *)

type ioprio_target =
  | Ioprio_process of int   (** A single process *)
  | Ioprio_pgrp of int      (** A process group *)
  | Ioprio_user of int      (** All processes owned by this user *)

type ioprio =
  | Noprio                  (** I/O prioritization is unsupported by block layer *)
  | Real_time of int        (** 0..7 (higest..lowest prio) *)
  | Best_effort of int      (** 0..7 (higest..lowest prio) *)
  | Idle


external ioprio_get : ioprio_target -> ioprio = "netsys_ioprio_get"
    (** Retrieve the priority of the target. If several processes match the
        target, the highest priority is returned. If no process matches,
        the unix error [ESRCH] will be raised.
     *)

external ioprio_set : ioprio_target -> ioprio -> unit = "netsys_ioprio_set"
    (** Sets the priority of the target processes. *)


(** {1 Debugging} *)

module Debug : sig
  val enable : bool ref
    (** Enables {!Netlog}-style debugging *)

end


(**/**)

external get_event_fd_nodup : not_event -> Unix.file_descr = "netsys_get_not_event_fd_nodup"