16 Advanced Linux Programming

16.1 Writing Your Own Bindings

• the basic Ada data types are equivalent to their C counterparts: an Ada integer array is a C integer array
• in parameters are the same as pass by copy parameters in C
• in out parameters are the same as passing a pointer as a parameter in C
• Ada string parameters ending in ASCII.NUL are the same as a C string
• Ada procedures are the same as C void functions

There are rare cases when these assumptions don't hold (e.g. certain cases when null pointer parameters are not allowed by Ada), but, generally speaking, these assumptions are valid under Linux. Gnat has general purpose interfacing pragmas and support for C types in the Interfaces.C package. Use these if you want maximum portability.

Because of these assumptions, most C library calls are easily represented in Ada. For example, we check the man page for gettime and discover it returns the current time as a long integer. To call this from Ada, we use

Since there is no Ada body for the gettime function, we use pragma import to let gnat know gettime is a C function. When we link, we need to specify the C library that function is in. In the case for the GNU C library, this is unnecessary since it's automatically linked. We can now call the C function gettime as if we wrote it ourselves.In C, it's possible to call a function and discard the result by not assigning it to anything. You can call C functions from Ada this way by declaring them a procedure. For example:

In this case, it's not particularly useful to call gettime and throw away the time it gives you. In general, you should avoid discarding the result because you may find it useful at some time in the future. However, there are certain C function where the result is provided only for flexibility, such as functions that return a pointer in a parameter and return the same pointer as the function result as well. These can safely be discarded by treating the function as a procedure.If we wanted to export an integer variable called TotalTimeEstimate to C, we'd use

A C function that returns void corresponds to an Ada procedure.

When importing or exporting to C, gnat converts the variable to lower case because C is a case-sensitive language. TotalTimeEstimate would be called totaltimeestimate in a C program. You can override this by providing a specific C name to link to. For example,

Import and Export don't require the name be the same at all. However, using entirely different names in C and Ada will make your program hard to understand.

If you want to import functions from libraries other than the standard C library, you will have to explicitly link them in. For example, to use the C math library, libm.a, would have to be explicitly linked using -lm. In C, functions can have parameters that change value, while in Ada this kind of function is not allowed because functions can only have "in" parameters. To get around this problem, gnat defines an import_valued_procedure pragma. Suppose you have a C function like this:

Normally, there is no way to represent this kind of function in an Ada program. However, we can import it by treating it as a procedure using the import_valued_procedure pragma:

The import_valued_procedure pragma tells gnat that this procedure corresponds to a C function: the first parameter is the result of the C function, and the remaining parameters correspond to the parameters of the C function. The first import pragma is not strictly required, but ACT recommends using it.

You can't import identifiers created by the #define statement since they only exist before a C program is compiled. You also can't import types (except for C++ classes) since types have no address in memory. [KB-true?]

There is one case where these tricks fail: when the C function returns a pointer to a C variable that it declared. In this case, the function is returning a new C pointer. Luckily, Ada provides a package called Address_To_Access_Conversions to convert between C pointers and Ada access types. You instantiate the package with the type you want to convert between, and the package creates an access type that can be converted to and from an address (which is a C pointer).The following program demonstrates conversions to and from C pointer types.

with Ada.Text_IO, System.Address_To_Access_Conversions;
use Ada.Text_IO;
procedure pointers is

package IntPtrs is
  new System.Address_To_Access_Conversions( integer );
  -- Instantiate a package to convert access types to/from addresses.
  -- This creates an integer access type called Object_Pointer.

  five : aliased integer := 5;
  -- Five is aliased because we will be using access types on it

  int_pointer : IntPtrs.Object_Pointer;
  -- This is an Ada access all type

  int_address : System.Address;
  -- This is an address in memory, a C pointer

begin
  int_pointer := five'unchecked_access;
  -- Unchecked_access needed because five is local to main program.
  -- If it was global, we could use 'access.

  int_address := five'address;
  -- Addresses can be found with the 'address attribute.
  -- This is the equivalent of a C pointer.

  int_pointer := IntPtrs.To_Pointer( int_address );
  int_address := IntPtrs.To_Address( int_pointer);
  -- Convert between Ada and C pointer types.

end pointers;

For example, the standard C library function get_current_dir_name returns a pointer to a C string which it declares. To use get_current_dir_name, we have to instantiate Address_To_Access_Conversions for an array of characters (a C string), and convert the address to an access type using something like

There is no other way in Ada to access the array that get_current_dir_name points to.

To import a C union, you'll have to use pragma unchecked_union to disable Ada features that are incompatible with C (namely, that the discriminant is not saved).

KB-If your main program is a C program, you need to call adainit before any Ada code.

16.2 Linux Errors and Errno

Most standard C library errors are returned in an integer variable called "errno". You can examine errno in your Ada programs by importing it.

Errno contains a number for the error returned by the last library function.

In Multithreading programs, be aware errno may not be not "thread safe" because it can be shared between threads. [KB: document how to do it with threads]

Linux provides two functions for working with errno error numbers.

type string255is new string(1..255);
type strptr is access string255;
-- error messages are no longer than 255 characters

The following example program makes a deliberate error with the link function and prints the error message using perror and stderror.

with ada.text_io, ada.strings.fixed;
use ada.text_io, ada.strings.fixed;
procedure perr is
-- an example of perror and strerror error messages

  procedure perror( message : string );
  pragma import( C, perror );
  -- print a standard error description with a leading message

  type string255 is new string(1..255);
  type strptr is access string255;
  -- error messages are no longer than 255 characters

  function strerror( error_number : integer )  return strptr;
  pragma import( C, strerror);
  -- get a standard error description

  errno : integer;
  pragma import( C, errno );
  -- last error number

  function link( path1, path2 : string ) return integer;
  pragma import( C, link);
  -- we'll use the link function to create an error

  LinkResult      : integer; -- value returned by link
  ErrorMessagePtr : strptr;  -- pointer to stderror message
  NullLocation    : integer; -- location of NUL in stderror message

begin

  Put_Line( "This is an example of perror and strerror");
  New_Line;

  -- make a deliberate error and print it with perror

  Put_Line( "Trying to link a non-existent file to itself.." );
  LinkResult := Link( "blahblah", "blahblah" );
  if LinkResult = -1 then
    perror( "Link failed" );
  end if;
  New_Line;

  -- Retrieve the last error message with strerror.
  -- Because strerror returns a C string, only print the
  -- string up to the first NUL character.

  ErrorMessagePtr := StrError( Errno );
  NullLocation := Index( string( ErrorMessagePtr.all ), "" & ASCII.NUL );
  Put( "The last error message was '" );
  Put( Head( string( ErrorMessagePtr.all ), NullLocation-1 ) );
  Put_Line( "'." );

end perr;

This is an example of perror and strerror
Trying to link a non-existent file to itself.
Link failed: No such file or directory
The last error message was 'No such file or directory'.

A table of error numbers is in the appendix.

16.3 The Linux Clock

The Ada.Calendar package is the standard method of working with time in Ada programs. If you need to interface with C programs, you may need to use Linux's time features.

The Linux clock functions are either kernel calls or are a part of the standard C library, and they don't need to be linked in with the -lc option.

16.3.1 Basic time functions

The basic Linux time functions work with the number of seconds since January 1, 1970. This is referred to as the epoch in the Linux man pages. Because of the limits of a long integer value, the Linux clock will stop working properly around the year 2038.

The basic functions use a long_integer for the time:

type time_t is new long_integer;

function difftime( time1, time2 : time_t ) return long_float;
pragma import( C, difftime );
Returns the number of seconds between two times (as a long_float).

16.3.2 Timeval Calls - Microsecond Accuracy

The timeval kernel calls return (or set) the current time with microsecond accuracy using a timeval record.

type timeval is record
  tv_sec  : time_t; -- number of seconds (since epoch)
  tv_usec : long_integer; -- number of microseconds
end record;

type timezone is record
  tz_minuteswest : integer; -- minutes west of Greenwich
  tz_dsttime     : integer  -- unsupported in Linux
end record;

procedure gettimeofday( result : out integer; tv : in out timeval, tz : in out timezone );
pragma import(C, gettimeofday );
pragma import_valued_procedure( gettimeofday );
Get the current time as the number of microseconds since January 1, 1970.  Returns 0 for success.  ftime() is an obsolete version of this function

procedure settimeofday( result : out integer; tv : in out timeval; tz : in out timezone );
pragma import( C, settimeofday );
pragma import_valued_procedure( settimeofday );
Set the current time as the number of microseconds since January 1, 1970.  Returns 0 for success.

procedure tzset;
pragma import( C, tzset );
Create the TZ environment variable, if it doesn't exist, and sets it to the current timezone as specified in /etc/localtime or /usr/lib/zoneinfo/localtime.  This is automatically invoked by the standard C library time functions whenever necessary.

procedure adjtimex( result : out integer; buf : inout timex );
pragma import( C, adjtimex );
Tunes the kernel's clock for specific time parameters

16.3.3 Functions using the tm record

Besides the number of seconds elapsed since 1970, Linux can also work with records containing the time broken down into common measurements. These functions use a tm record. These functions are all a part of the standard C library.

type tm is record
   sec   : integer; -- seconds on the clock (0-59)
   min   : integer; -- minutes on the clock (0-59)
   hour  : integer; -- hour on the clock (0-23)
   mday  : integer; -- day of the month (1-31)
   mon   : integer; -- month (0-11)
   year  : integer; -- year
   wday  : integer; -- day of the week (0-6)
   yday  : integer; -- day of the year (0-365)
   isdst : integer; -- >0 is daylight savings time, 0=not, <0 unknown
end record;

You will also need the Address_To_Access_Conversions package to convert C pointers to tm record into Ada access type pointers.

package TmPtrs is
  new System.Address_To_Access_Conversions( tm );

To get the current time in tm format,

  seconds_since_1970 : long_integer;
  tm_rec : tm;
  ...
  time( seconds_since_1970 );
  tm = TmPtrs.To_Pointer( localtime( seconds_since_1970'address ) ).all;

16.3.4 Time as a String

procedure strftime( result: size_t; datestr : in out stringtype; max : size_t; format : string; tm : in out tmrec );
pragma import( C, strftime );
Like asctime(), converts a tm time into text.  strftime() uses formatting codes to determine the appearance of the text, similar to the shell date command.  Returns the length of the date string (including the ending ASCII.NUL). See the man page for complete details.

Example:
datestring : string(1..80);
...
statftime( datestringsize, datestring, datestring'size/8, "%I:%M" & ASCII.NUL, tm );
Ada.Text_IO.Put_Line( "The time is " & datestring( 1..datestringsize-1 ) );
 

16.3.5 Timer Functions

Timer functions use the timeval structure

function timerclear( tv : timeval );
function timerisset( tv : timeval );
function timercmp( t0, t1 : timeval; operator : ? );

16.4 Process Information

The Linux process functions are part of the standard C library, and do not need to be linked in with -lc.

16.4.1 Ownership

The owner of a program is referred to as the UID (user identification number). Under Linux, there are actually three owners to any given program: the effective UID, the real UID and the saved UID. Normally, these three are all the same login. The real and saved uids are provided for programs that must temporarily pretend to be somebody else, like a daemon that needs special login for a short period of time, or setuid/setgid programs that must temporarily switch between owners. These special functions are not covered here.

The GID (group identification number) is the group the program belongs to. In Linux, there's a main, effective group number, and any number of secondary groups that a program can belong to. There is also real and saved GIDs, just like UIDs.

Linux also allows you to arrange processes into groups for easier management of multiple processes at once. Each process group as a, no surprise, a process group identification number (PGID).

Every program and process group also belongs to a session (as in a login session). When you log off the computer, Linux automatically stops all programs that were running in your session. The session leader is the top process in a session, such as your login shell. However, if you want to create a new session for some reason, you can use the following function:

Signal handling, in general, is easier through Ada.Interrupts than through Linux kernel calls because of their heavy reliance on C macros.--KB

16.5 Environment Variables

Environment variables can easily be set and read with Ada.Command_Line.Environment package. You can also set them directly through the standard C library.

16.6 Multitasking

Multitasking creates child processes to do tasks for your main program. On multiprocessor machines, different processes can be assigned to different processors allowing work to be done simultaneously. On single processor machines, the processor switches several times a second between processes and does a little work on each.

The function to create a new child process is called fork. When Linux creates a new process, it doesn't start by creating a blank process. Instead, it makes a copy of the original process so there are effectively two copies of your program running. The fork function returns a value to tells your program if it is the original program or the new copy.

If you want to run a different program, you'll have to use one of the exec family of functions to load and run the new program. The exec functions destroy the old program and run a new program in its place. The above section, Using System and OSLib.Spawn, has an example C function called CrunIt that uses fork to start a new process and run a new program.

Wait3 and Wait4 are BSD UNIX variations which perform the same functions as wait and waitpid but with slightly different parameters.

When multitasking, if a child process stops, it's retained in memory so that the parent can use wait to find out the reason it stopped running. These children are called "zombies". If the parent process doesn't use wait, the zombies will remain indefinitely, using up system resources. For any large multitasking program, make sure you handle SIGCHLD signals: these are created when a child stops running. The SIGCHLD handler only needs to call wait and Linux will then remove the child process from memory.

The following is a simple multitasking example that multitasks two Put_Line statements.

-- a simple example of multitasking that multitasks
-- two put_line statements

with ada.text_io;
use ada.text_io;
procedure multitask  is
type pid_t is new integer;

  function fork return pid_t;
  pragma import( C, fork);
  -- create a new process

  errno : integer;
  pragma import( C, errno);
  -- the last error code

  procedure wait( pid : out pid_t; status : in out integer);
  pragma import( C, wait);
  pragma import_valued_procedure( wait);
  -- wait until all child processes are finished

  myPID : pid_t;
  wait_pid : pid_t;
  wait_status : integer;

begin

  Put_Line( "Welcome to this multitasking example" );
  Put_Line( "This is the original process." );
  New_Line;

  -- the fork function duplicates this program into
  -- two identical processes.

  Put_Line( "Splitting into two identical processes..." );
  Put_Line( "-----------------------------------------" );
  myPID := fork; -- split in two!

  -- This program is now the original process or the
  -- new child process. myPID tells you which process
  -- you are.

  if myPID < 0 then
    Put_Line( Standard_Error, "Fork has failed.  Error code " & errno'img );
  elsif myPID = 0 then
    Put_Line( "This is the child process" );
  else
    Put_Line( "This is the original process." );
    -- wait until child is finished
    wait( wait_pid, wait_status);
    if wait_pid < 0  then
      Put_Line( Standard_Error, "Wait error: wait returned PID " & wait_pid'img
        & " and error number " & errno'img);
    end if;
  end if;

end multitask;

Welcome to this multitasking example
This is the original process.
Splitting into two identical processes...
-----------------------------------------
This is the original process.
This is the child process

16.7 Linux File Operations

The Linux file operations are part of the standard C library, and don't need to be linked in with the -lc option. The C calls are defined in the "fcntl.h" header file.

[Explain Linux files here]

Linux never shortens files. If your file gets smaller, you must shorten it yourself using truncate.

The following bindings assume these types have been defined.

type file_id is new integer;
-- file ID number are discussed below

type mode_t is new integer;
type gid_t is new integer;
type uid_t is new integer;
type size_t is new long_integer;

Other low-level file operations are all done with the fcntl (file control) function. There are three variations to fcntl: it may have an operation code, an operation code and a long integer argument, or an operation code and a locking record argument.

The operation numbers are defined in /usr/src/linux/asm-i386/fnctl.h:

F_DUPFD  : constant integer :=  0;
F_GETFD  : constant integer :=  1;
F_SETFD  : constant integer :=  2;
F_GETFL  : constant integer :=  3;
F_SETFL  : constant integer :=  4;
F_GETLK  : constant integer :=  5;
F_SETLK  : constant integer :=  6;
F_SETLKW : constant integer :=  7;
F_SETOWN : constant integer :=  8;
F_GETOWN : constant integer :=  9;
F_SETSIG : constant integer := 10;
F_GETSIG : constant integer := 11;

Additional information about locks are found in /usr/src/linux/Documentation/locks.txt

type aLock is new short_integer;
F_RDLCK : constant aLock := 0; -- read lock
F_WRLCK : constant aLock := 1; -- write lock
F_UNLCK : constant aLock := 2; -- unlock (remove a lock)
F_EXLCK : constant aLock := 3; -- exclusive lock
F_SHLCK : constant aLock := 4; -- shared lock

type aWhenceMode is new short_integer;
SEEK_SET : constant aWhenceMode := 0; -- absolute position
SEEK_CUR : constant aWhenceMode := 1; -- offset from current position
SEEK_END : constant aWhenceMode := 2; -- offset from end of file

type lockstruct is record
  l_type   : aLock;         -- type of lock
  l_whence : short_integer; -- how to interpret l_start
  l_start  : integer;       -- offset or position
  l_len    : integer;       -- number of bytes to lock (0 for all)
  l_pid    : integer;       -- with GETLK, process ID owning lock
end record;

To lock a file, create a lockstruct record and fill in the details about the kind of lock you want.

A read lock (F_RDLCK) makes the part of the file you specify read-only. No one can write to that part of the file.

A write lock prevents any other program from reading or writing to the part of the file you specify. Your program may change that part of the file without being concerned that another process will try to read it before you're finished.

If your program stops prematurely, the locks will be released.

  -- lock file
  myLockStruct : lockStruct;
  result : integer;
  ...
  myLockStruct.l_type := F_WRLCK;
  myLockStruct.l_whence := 0;
  myLockStruct.l_start := 0;
  myLockStruct.l_end := 0;
  fcntl( result, fd, F_SETLKW, myLockStruct );
  if result = -1 then
     put_line( standard_error, "fcntl failed" );
  end if;
  -- file is now locked
  ...
  -- unlock file
  myLockStruct.l_type := F_UNLCK;
  myLockStruct.l_whence := 0;
  fcntl( result, fd, F_SETLKW, myLockStruct );
  if result = -1 then
     put_line( standard_error, "fcntl failed" );
  end if;

[Double check off_t size for l_start, l_len--KB]

16.8 Opening and Closing Files

The standard Ada packages Text_IO, Sequential_IO and Direct_IO are suitable for simple projects, but they were never intended as a complete solution for large-scale applications. If you want to do efficient file manipulation, you'll have to write your own routines based on kernel calls or the standard C library.

gnat's OSLIB package contains low-level commands to work with UNIX files. However, you can always create your own.

The following bindings assume these types have been defined.

type file_id is new integer;
type mode_t is new integer;
type off_t is new long_integer;
type size_t is new long_integer;
subype ssize_t is size_t;

O_RDONLY   : constant integer :=      8#00#; -- open for reading only
O_WRONLY   : constant integer :=      8#01#; -- open for writing only
O_RDWR     : constant integer :=      8#02#; -- open for reading and writing
O_CREAT    : constant integer :=    8#0100#; -- no file? create it
O_EXCL     : constant integer :=    8#0200#; -- lock file (see below)
O_NOCTTY   : constant integer :=    8#0400#; -- if tty, don't acquire it
O_TRUNC    : constant integer :=   8#01000#; -- file exists? truncate it
O_APPEND   : constant integer :=   8#02000#; -- file exists? move to end
O_NONBLOCK : constant integer :=   8#04000#; -- if pipe, don't wait for data
O_SYNC     : constant integer :=  8#010000#; -- don't cache writes
O_ASYNC    : constant integer :=  8#020000#; -- async. IO via SIGIO
O_DIRECT   : constant integer :=  8#040000#; -- direct disk access
O_LARGEFILE: constant integer := 8#0100000#; -- not implemented in Linux (yet)
O_DIRECTORY: constant integer := 8#0200000#; -- error if file isn't a dir
O_NOFOLLOW : constant integer := 8#0400000#; -- if sym link, open link itself

SEEK_SET : constant integer := 0; -- from start of file
SEEK_CUR : constant integer := 1; -- offset from current position
SEEK_END : constant integer := 2; -- from end of file

File input/output is naturally suited to generic packages. You can use the generic package to hide the low-level details of the standard C library. In following example, SeqIO is a generic package for reading and writing a sequential file of some type, using the kernel calls mentioned above.

-- SeqIO
--
-- A simple sequential IO package using standard C functions

generic
  type AFileElement is private;

package SeqIO is

  type AFileID is new short_integer;
  seqio_error : exception;

  function Open( path : string; read : boolean := true ) return AFileID;
  -- open a new file for read or write

  procedure Close( fid : AFileID );
  -- close a file

  procedure Read( fid : AFileID; data : in out AFileElement);
  -- read one data item from the file. seqio_error is raised
  -- if the data couldn't be read

  procedure Write( fid : AFileID; data : AFileElement );
  -- write one data item to the file.  seqio_error is raised
  -- if the data couldn't be written

end SeqIO;

package body SeqIO is
pragma optimize( space);

  -- Import C file handling functions
  type mode_t is new integer; -- C mode_t type
  type size_t is new integer; -- C size_t type
  subtype ssize_t  is size_t; -- C ssize_t type

  -- The C file functions we'll be using
  -- (denoted with a C_ prefix for clarity )

  function C_Open( path : string; flags : integer; mode : mode_t)
    return AFileID;
  pragma import( C, C_Open, "open");

  function C_Close( file : AFileID )  return integer;
  pragma import( C, C_Close, "close" );

  procedure C_Read( size : out ssize_t;
    file : AFileID;
    data : in out AFileElement;
    count: size_t);
  pragma import( C, C_Read, "read");
  pragma import_valued_procedure( C_Read);
  -- Using an "in out" parameter is the easiest way to pass
  -- the address of the data element. Because Ada doesn't
  -- allow in out parameters in functions, we'll use gnat's
  -- valued procedure pragma to pretend read is a procedure

  procedure C_Write( size : out ssize_t;
    file : AFileID;
    data : in out AFileElement;
    count: size_t);
  pragma import( C, C_Write, "write");
  pragma import_valued_procedure( C_Write);
  -- Using an "in out" parameter is the easiest way to pass
  -- the address of the data element. Because Ada doesn't
  -- allow in out parameters in functions, we'll use gnat's
  -- valued procedure pragma to pretend write is a procedure
  -- Our Ada subprograms

function Open( path : string; read : boolean := true ) return AFileID  is
  -- open a new file for read or write
  flags : integer;
begin

  -- the flag values are listed in fcntlbits.h and man 2 open
  if read then
    flags := 0; -- read only, existing file
  else
    flags := 1000 + 100 + 1; -- write only, create or truncate
  end if;
  -- octal 640 => usr=read/write, group=read, others=no access

  return C_Open( path & ASCII.NUL, flags, 8#640# );
end Open;

procedure Close( fid : AFileID ) is
  -- close a file
  Result : integer; -- we'll ignore it
begin
  Result := C_Close( fid);
end Close;

procedure Read( fid : AFileID; data : in out AFileElement ) is
  -- read one data item from the file
  BytesRead : ssize_t;
begin
  -- 'size returns the size of the type in bits, so we
  -- divide by 8 for number of bytes to read
  C_Read( BytesRead, fid, data, AFileElement'size / 8 );
  if BytesRead /= AFileElement'size / 8  then
    raise seqio_error;
  end if;
end Read;

procedure Write( fid : AFileID; data : AFileElement ) is
  -- write one data item to the file
  BytesWritten : ssize_t;
  data2write : AFileElement;
begin
  -- can't use data directly because it's an "in" parameter
  data2write := data;
  -- 'size returns the size of the type in bits, so we
  -- divide by 8 for number of bytes to write
  C_Write( BytesWritten, fid, data2write, AFileElement'size / 8);
  if BytesWritten /= AFileElement'size / 8 then
    raise seqio_error;
  end if;
end Write;

end SeqIO;

You can test SeqIO with the following program:

with SeqIO;
with Ada.Text_IO;
use Ada.Text_IO;

procedure SeqIOtest is
-- program to test SeqIO

package IntIO is new SeqIO( integer);
-- IntIO is a SeqIO for integer numbers

  id : IntIO.AFileID;
  int2read : integer;

begin

  Put_Line( "Testing SeqIO package..." );
  New_Line;

  -- Part #1: Write numbers to a file

  Put_Line( "Writing numbers 1 to 10 to a file...");
  id := IntIO.Open( "int_list.txt", read => false );
  for i in 1..10 loop
    IntIO.Write( id, i)
  end loop;
  IntIO.Close( id);

  -- Part #2: Read the numbers back from the same file

  Put_Line( "Reading numbers back..." );
  id := IntIO.Open( "int_list.txt", read => true);
  for i in 1..10  loop
    IntIO.Read( id, int2read);
    Put_Line( "Number" & i'img & " =" & int2read'img );
  end loop;
  IntIO.Close( id );

exception when IntIO.seqio_error =>
  Put_Line( "Oh, oh! seqio_error!");
end SeqIOtest;

Note: This should be rewritten because a failure to write all the bytes is not necessarily an error--Linux has a buffer limit on how much it writes at one time--KB

Writing numbers 1 to 10 to a file...
Reading numbers back...
Number 1 = 1
Number 2 = 2
Number 3 = 3
Number 4 = 4
Number 5 = 5
Number 6 = 6
Number 7 = 7
Number 8 = 8
Number 9 = 9
Number 10 = 10

These kernel calls help programs that have to monitor several file descriptors at once for activity.

type pollfd is record
  fd : integer;
  events : short_integer;
  revents : short_integer;
end record;

Poll Events
POLLIN := 16#1#;
POLLPRI := 16#2#;
POLLOUT := 16#4#;
POLLERR := 16#8#;
POLLHUP := 16#10#;
POLLNVAL := 16#20#;

These are defined in asm/poll.h.

16.9 Directories

Directories are "folders" containing collections of files and other directories. In Linux, a directory is a special kind of file. Some of the standard file operations work on directories and some other file operations are specific to directories.

The top-most directory is /, or the root directory. All files on a system are located under the root directory. Disk drives do not have separate designations as in MS-DOS.

A period (.) represents the current directory, and a double period (..) represents the parent directory of the current directory. All directories have . and .. defined. The root directory, of course, doesn't have a parent: it's .. entry points to itself.

Many of the kernel calls and standard C library functions dealing with directories use functions that return C pointers. As mentioned in the bindings section, the only way to convert these kind of functions to Ada is by declaring the C pointers as a System.Address type and changing the C pointers to Ada access types using the Address_To_Access_Conversions package.

There is also a scandir function that reads a directory and sorts the entries, but this is difficult to use directly from Ada.

The following program demonstrates some of the directory subprograms in Linux.

with Ada.Text_IO, Interfaces.C, Ada.Strings.Fixed;
use Ada.Text_IO, Interfaces.C, Ada.Strings.Fixed;
with System.Address_To_Access_Conversions;

procedure direct is

   -- Working with directories

  subtype size_t is Interfaces.C.size_t;
  -- renaming size_t to save some typing

  package CStringPtrs is  new
    System.Address_To_Access_Conversions( string );
  use CStringPtrs;
-- Convert between C and Ada pointers to a string

  subtype DirInfoCPtr  is System.Address;
  subtype DirEntCPtr is System.Address;
  -- two C pointers (System.Address types), renamed for
  -- clarity below

  type DirEnt is record
    inode : long_integer; -- inode number
    offset : integer; -- system dependent
    offset2: unsigned_char; -- system dependent
    reclen : unsigned_short; -- system dependent
    name : string( 1..257 ); -- name of file
  end record;
  pragma pack( dirent);
  -- dirent is defined in /usr/src/linux../linux/dirent.h

  package DirEntPtrs is new
    System.Address_To_Access_Conversions( DirEnt );
  use DirEntPtrs;
    -- Convert between C and Ada pointers to a directory entry

  procedure getcwd( buffer : out string; maxsize : size_t );
  pragma import( C, getcwd );

  function get_current_dir_name return System.Address;
  pragma import( C, get_current_dir_name);

  function mkdir( path : string; mode : size_t ) return integer;
  pragma import( C, mkdir );

  function rmdir( path : string ) return integer;
  pragma import( C, rmdir );

  function opendir( path : string ) return DirInfoCPtr;
  pragma import( C, opendir );

  function closedir( info : DirInfoCPtr ) return integer;
  pragma import( C, closedir );

  function readdir( info : DirInfoCPtr ) return DirEntCPtr;
  pragma import( C, readdir );

  function rewinddir( info : DirInfoCPtr ) return integer;
  pragma import( C, rewinddir );

  function telldir( info : DirInfoCPtr ) return integer;
  pragma import( C, telldir );

  function seekdir( info : DirInfoCPtr; position : integer ) return integer;
  pragma import( C, seekdir );

  -- scandir hard to use from Ada

  s: string(1..80);
  csop: CStringPtrs.Object_Pointer;
  Result: integer;
  DirInfo: DirInfoCPtr;
  direntop : DirEntPtrs.Object_Pointer;
  Position : integer;
  LastPosition : integer;

begin

  Put_Line( "This program demonstrates Linux's directory functions" );
  New_Line;

  -- getcwd example

  getcwd( s, s'length );
  Put( "The current path (simplified) is " );
  Put_Line( Head( s, Index( s, ASCII.NUL & "" )-1 ));

  -- Index for fixed strings takes a string as the second parameter
  -- We'll make a string containing an ASCII.NUL with &

  -- get_current_dir_name example

  csop := To_Pointer( get_current_dir_name );
  Put( "The current path is " );
  Put_Line( Head( csop.all, Index( csop.all, ASCII.NUL & "" )-1 ) );

  -- mkdir example: create a directory named "temp"

  Result := mkdir( "temp" & ASCII.NUL, 755 );
  if Result /= 0 then
    Put_Line( "mkdir error" );
  else
    Put_Line( "temp directory created" );
  end if;

  -- rmdir example: remove the directory we just made

  Result := rmdir( "temp" & ASCII.NUL );
  if Result /= 0 then
    Put_Line( "rmdir error" );
  else
    Put_Line( "temp directory removed" );
  end if;
  New_Line;

  -- directory reading

  DirInfo := OpenDir( "/home/ken/ada" & ASCII.NUL);
  Put_Line( "Directory /home/ken/ada contains these files:");
  loop
    direntop := To_Pointer( ReadDir( DirInfo ) );
  exit when direntop = null;

    -- TellDir returns the position in the directory
    -- LastPosition will hold the position of the last entry read

    LastPosition := Position;
    Position := TellDir( DirInfo );
    Put_Line( Head( Direntop.name, Index( Direntop.name, ASCII.NUL & "" )-1 ) );
  end loop;
  New_Line;

  -- SeekDir: move to last position in directory
  Result := SeekDir( DirInfo, LastPosition );
  Put( "The last position is " );
  direntop := To_Pointer( ReadDir( DirInfo ) );
  Put_Line( Head( Direntop.name, Index( Direntop.name, ASCII.NUL & "" )-1 ) );
  New_Line;

  -- RewindDir: Start reading again

  Result := RewindDir( DirInfo );
  Put( "The first position is " );
  direntop := To_Pointer( ReadDir( DirInfo ) );
  Put_Line( Head( Direntop.name, Index( Direntop.name, ASCII.NUL & "" )-1 ) );
  New_Line;

  -- close the directory

  Result := CloseDir( DirInfo );
end direct;

This program demonstrates Linux's directory functions
The current path (simplified) is /home/ken/ada/trials
The current path is /home/ken/ada/trial
temp directory created
temp directory removed
Directory /home/ken/ada contains these files:
.
..
temp
all.zip
README
posix.zip
sm
posix
cgi
tia
x
rcsinfo.txt
text_only
original
lintel
texttools
smbeta2.zip
trials
plugins
texttools.zip

The last position is texttools.zip

The first position is .

16.10 Stdio Files

C has a library called stdio, or STanDard IO, which contains standard operations for text files. Loosely, stdio is the C equivalent of Ada's Text_IO package.The standard gnat package cstreams(?) provides a thin binding to many of the stdio functions. In this section, we'll looking at using stdio directly.

Some of the stdio functions can't be used from Ada because of differences in the languages. For example, printf, the standard command for writing to the screen, has a variable number of parameters. Because there's no way to express a variable number of parameters in Ada, printf can't be imported into Ada.

with System;
type AStdioFileID is new System.Address;

Operation: LOCK_SH (1) - shared lock
           LOCK_EX (2) - exclusive lock
           LOCK_NB (4) - no block flag (may be added to others)
           LOCK_UN (8) - unlock

16.11 Stdio Pipes

Pipes are the equivalent of shell command pipes formed by the '|' character. You can open a pipe to or from a shell command, depending if the pipe is for writing or reading respectively.

These single direction pipe commands are a part of the standard C library.

The following program prints to a printer by opening a pipe to the lpr command.

with Ada.Text_IO, System, SeqIO;
use Ada.Text_IO;

procedure printer2 is
  -- a program for simple printing

  ---> Pipe Stuff -------------------------------------

  type AStdioFileID is new System.Address;
  -- a pointer to a C standard IO (stdio) file id

  function popen( command, mode : string )  return AStdioFileID;
  pragma import( C, popen, "popen" );
  -- opens a pipe to command

  procedure pclose( result : out integer; fid : AStdioFileID );
  pragma import( C, pclose, "pclose" );
  pragma import_valued_procedure( pclose);
  -- closes a pipe

  function fputc( c : integer; fid : AStdioFileID ) return integer;
  pragma import( C, fputc, "fputc" );
  -- part of standard C library.Writes one charctera to a file.

  function fputs( s : string; fid : AStdioFileID ) return integer;
  pragma import( C, fputs, "fputs" );
  -- part of standard C library.Writes a string to a file.

  PipeID : AStdioFileID; -- File ID for lpr pipe

  procedure BeginPrinting is
    -- open a pipe to lpr
  begin
    Put_Line( "Opening pipe to lpr ..." );
    PipeID := popen( "lpr" & ASCII.NUL, "w"& ASCII.NUL);
  end BeginPrinting;

  procedure EndPrinting is
    -- close the pipe.Result doesn't matter.
    -- Linux normally will not eject a page when
    -- printing is done, so we'll use a form feed.
    Result : integer;
  begin
    Result := fputc( character'pos( ASCII.FF ), PipeID);
    pclose( Result, PipeID );
  end EndPrinting;

  --> Input/Output Stuff --------------------------------

  procedure Print( s : string ) is
    -- print a string to the pipe, with a carriage
    -- return and line feed.
    Result : integer;
  begin
    Result := fputs( s & ASCII.CR & ASCII.LF & ASCII.NUL, PipeID );
  end Print;

begin

  -- Open the pipe to the lpr command


  Put_Line( "Starting to print..." );
  BeginPrinting;
  Print( "Sales Report" );
  Print( "------------" );
  Print( "" );

  Print( "Sales were good" );

  -- Now, close the pipe.

  EndPrinting;

  Put_Line( "Program done...check the printer" );

end printer2;

16.12 Memory Management

The amount of virtual memory for a process depends on the processor. For Intel x86 processors, your program and data must be 3 Gigabytes or less. (An additional 1 Gigabyte per process is reserved for the kernel, accounting for the full 32-bits of addressing space.)

[not finished--KB]

16.13 The Virtual Consoles

The virtual consoles are controlled by the ioctl() function.

[not finished--KB]

with Ada.Interrupts.Names;
useAda.Interrupts;

package sigwinch is

protected SignalHandler is

  procedure SizeChangeHandler;
  pragma Attach_Handler( SizeChangeHandler, Names.SIGWINCH );
  -- this handler will catch SIGWINCH signals, a window size
  -- change

end SignalHandler;

end sigwinch;

with Ada.Text_IO;
useAda.Text_IO;

package body sigwinch  is

  -- imported C functions

  TIOGWINSIZ : constant integer := 16#5413#;
  -- get window size ioctl request

  type WindowSizeInfo is record
    row, column, unused1, unused2 : short_integer;
  end record;
  pragma pack( WindowSizeInfo );

  -- the window size information returned by ioctl

  type AFileID is new integer;
  -- a file descriptor, a new integer for safety

  procedure ioctl_winsz( Result : out integer; fid : AFileID; request : integer;
  info : in out WindowSizeInfo );
  pragma import( C, ioctl_winsz, "ioctl");
  pragma import_valued_procedure( ioctl_winsz, "ioctl" );
  -- get the size of the window

  function open( path : string; flags : integer ) return AFileID;
  pragma import( C, open, "open" );
  -- open a file (in this case, the tty)

  procedure close( fid : AFileID );
  pragma import( C, close, "close" );
  -- close a file

  -- The Signal Handler

  protected body SignalHandler is

    procedure SizeChangeHandler is
    -- handle a window size change, SIGWINCH

      fid: AFileID;-- open's file ID

      Result : integer; -- function result of ioctl

      Info: WindowSizeInfo; -- window size returned by ioctl

    begin

      fid := Open( "/dev/tty" & ASCII.NUL, 0 );

      ioctl_winsz( Result, fid, TIOGWINSIZ, Info );

      if Result = 0 then
        Put_Line( "Window is now " & info.column'img & " x " & info.row'img);
      else
        Put_Line( "ioctl reports an error" );
      end if;

      Close( fid );

    end SizeChangeHandler;

  end SignalHandler;

end sigwinch;

with Ada.Text_IO, sigwinch;
useAda.Text_IO, sigwinch;

procedure winch is
begin
  Put_Line( "This program catches SIGWINCH signals");
  New_Line;

  Put_Line( "It will stop running is 60 seconds. If you are using" );
  Put_Line( "X Windows, move the window to send signals.");
  New_Line;

  delay 60.0; -- run for 60 seconds

end winch;

16.14. Making Database Queries

16.14.1 mySQL

mySQL (pronounced "my ess que ell" ) is a free, high-performance database from T.c.X. It's available for a number of platform, including Linux. The mySQL home page is http://www.mysql.org.

mySQL comes with a C library called "mysqlclient". If an Ada program links in this library with "-lmysqlclient", it can access mySQL databases. You issue commands to the database called queries using the database language SQL (pronounced "sequel").

Connecting to a mySQL database is a six step process:

1. Open a new connect using mysql_init.
2. Login using mysql_real_connect.
3. Perform database queries with mysql_query or mysql_real_query. real_query allows binary data in the query.
4. Retrieve the results using mysql_store_result or mysql_use_result.
5. Free any memory using mysql_free_result.
6. Close your connection with mysql_close.

Usually, a null point or non-zero integer result indicates an error. mysql_errono returns the error.

Complete documentation is available from the mySQL web site.

16.14.2 PostgreSQL

Not finished--KB

16.15 Dynamic Loading

Not finished--KB

16.16 Writing Linux Modules

If you are writing kernel components, you'll have to use "GNORT" (pragma no_run_time). Without the run time components, you won't be able to use some Ada packages and features, including exceptions and the Text_IO library (which uses exceptions). You'll have to import the appropriate C functions to do I/O. Without the run time library, you're source code will also be much smaller, comparable with C.

If you have the time, you can always copy some of the standard Ada packages to a separate directory and compile them into your GNORT project, effectively creating your own small, custom run-time system.

The Gnat run-time system is described at http://www.iuma.ulpgc.es/users/jmiranda/.

For details on how to program for the Linux kernel, read Linux Kernel Module Programming Guide. To create a Linux kernel module in Ada, you will also need to register a license. In C, this is accomplised with the MODULE_LICENSE macro. The Linux kernel will expect init_module and cleanup_module subprograms to exist.

with Interfaces.C;

package SomeModule is

  -- Module Variables

  type Aliased_String is array (Positive range <>)
  of aliased Character;
  pragma Convention (C, Aliased_String);

  Kernel_Version: constant Aliased_String:="2.4.18-5" & Character'Val(0);
  pragma Export (C, Kernel_Version, "kernel_version");

  Mod_Use_Count: Integer;
  Pragma Export (C, Mod_Use_Count, "mod_use_count_");

  -- Kernel Calls

  procedure Printk( s : string );
  pragma import( C, printk, "printk" );

  -- Our Module Functions

  function Init_Module return Interfaces.C.int;
  pragma Export (C, Init_Module, "init_module");

  procedure Cleanup_Module;
  pragma Export (C, Cleanup_Module, "cleanup_module");
end SomeModule;

package body SomeModule is
  -- sample module layout

  function Init_Module return Interfaces.C.int is
  begin
    Printk("Hello,World!" & Character'val(10) & character'val(0));
    return 0;
  end Init_Module;

  procedure Cleanup_Module is
  begin
    Printk("Goodbye , World!" & Character'val(10) & character'val(0));
  end Cleanup_Module;
 end SomeModule;

Multiple object files must be combined into a single loadable module object file using the ld command. Never use -fPIC when working with the kernel. You may need the '-a' flag and '-s' flags so that Gnat will recompile the system package and related low-level Gnat files, or copy them by hand and compile them yourself (use the -gnatg switch).

Use insmod, lsmod and rmmod to install, test and uninstall your module.

$ insmod hello.o
   Hello, World!

For embedded systems, there is RTEMS, an embedded run time, and supports a more complete Ada runtime (with tasking) on some targets. See http://www.rtems.com/.

with Ada.Text_IO;
use Ada.Text_IO;

procedure nrt2 is
  -- Simple Program
begin
  put_line( "Hello World" );
end nrt2;

pragma no_run_time;

procedure nrt is
  -- Same as nrt2 but using no run time

  type file_id is new integer;

  -- No Ada.Text_IO possible, so we'll write our own
  -- that talks directly to the Linux kernel

  procedure write_char( amount_written : out long_integer;
	id : file_id;
  	buffer : in out character;
  	amount2write : long_integer );
  pragma import( C, write_char, "write" );
  pragma import_valued_procedure( write_char, "write" );

  procedure put( c : character ) is
    result : long_integer;
    buf    : character := c;
  begin
    write_char( result, 1, buf, 1 );
  end put;

  procedure new_line is
  begin
    put( character'val( 10 ) );
  end new_line;

  procedure put_line( s : string ) is
    pragma suppress( index_check, s );
    -- s(i) won't throw a range error, but Gnat checks for it
    -- by default.  Exceptions are a part of the run time.
  begin
    for i in s'range loop
        put( s(i) );
    end loop;
    new_line;
  end put_line;

begin
  put_line( "Hello World" );
end nrt;

16.17 Linux Sound

The Linux sound capabilities, called OSS, were developed by 4front technologies.You can find more advanced documentation at their website http://www.opensound.com. This section describes only the basic functions.

The newest Linux sound standard is ALSA.

Most distributions have OSS in the kernel by default, but there's no reason that OSS must be present--it can always be turned off for computers without a sound card.

16.17.1 Detecting a Sound Card

Open the file /dev/sndstatus. If there is no error, the computer has a sound card.

16.17.2 Playing Sound Samples

There are no C libraries or kernel calls to play sound samples. Instead, there is a device file called /dev/audio which plays sound samples in the .au sound format.

The .au sound format consists of a header describing the sound followed by the actual sound data. The header looks like this:

type AAUHeader is record
  Magic : integer;        -- a unique number denoting a .au file,
                          -- as used with the magic file, SND_MAGIC
                          -- Hex 646E732E (bytes 2E, 73, 6E, 64)
  dataLocation : integer; -- offset or pointer to the sound data
  dataSize: integer;      -- number of bytes of sound data
  dataFormat: integer;    -- the data format code
  samplingRate : integer; -- the sampling rate
  channelCount : integer; -- the number of channels
  info1, info2, info3, info4 : character;-- name of sound
end record;

dataLocation is an offset to the first byte of the sound data. If there's no sound name, it's 28, the size of the header. It can a pointer to the data, depending on the dataFormat code, but that doesn't apply if you're playing a .au file.

dataSize is the size of the sound data in bytes, not including the header.

dataFormat describes how the sound data is to be interpreted. Here is a table of some common values.

SamplingRate is the playback rate in hertz.CD quality samples are 44100.

channelCount is 1 for mono, 2 for stereo.

The info characters are a C null-terminated string giving a name for the sound. It's always at least 4 characters long, even if unused.

In order to play a sound, treat /dev/audio as if it were a device attached to your computer for playing .au sounds.Write a program to open /dev/audio for writing and write the .au sound to it.

Playing sounds is a natural candidate for multithreading because you don't want your entire program to stop while a sound is being played.

The following program uses the seqio generic package we developed above to play an .au sound through /dev/audio.

with seqio;
with Ada.Text_IO;
use Ada.Text_IO;

procedure playsnd is

  -- simple program to play an .au sound file

  package byteio is new seqio( short_short_integer );
  -- sequential files of bytes

  au_filename : constant string := "beep.au";
  -- sound file to play.  supply the name of the .au file to play

  au_file: byteio.AFileID; -- the sound file
  dev_audio: byteio.AFileID; -- /dev/audio device

  soundbyte : short_short_integer;

begin

  Put_Line( "Playing " & au_filename & "...");

  -- open the files

  au_file := byteio.Open( au_filename, read => true);
  dev_audio := byteio.Open( "/dev/audio", read => false);

  -- read until we run out of bytes, send all bytes to
  -- /dev/audio.The end of file will cause a seqio_error

  begin
    -- nested block to catch the exception

    loop
      byteio.Read( au_file, soundbyte );
      byteio.Write( dev_audio, soundbyte );
    end loop;

  exception when byteio.seqio_error =>
    null; -- just leave block
  end;

  -- close files

  byteio.Close( au_file );
  byteio.Close( dev_audio );

  Put_Line( "All done" );

exception when others =>
  Put_Line( "Oh, oh! An exception occurred!" );
  byteio.Close( au_file );
  byteio.Close( dev_audio );
  raise;

end playsnd;

16.17.3 Using the Mixer

You control the mixer chip, if your sound card has one, by using the ioctl() kernel call. If there is no mixer, the ioctl() function returns -1. Mixer Functions Table

Reading or writing values have a special bit set [Ken check using soundcard.h].

Ioctl calls return an integer value. Volume levels can be 0 to 100, but many sound cards do not offer 100 levels of volume. /dev/mixer will set the volume to setting nearest to your requested volume.

[Not complete--KB]

Sound_mixer_volume : constant integer := 0;
Sound_Mixer_Read : constant integer := ?;
Sound_Mixer_Write : constant integer := ?;

oldVolume : integer;

ioctlResult := Ioctl( mixer_file_id, sound_mixer_read + sound_mixer_volume, oldVolume );

-- master volume now in oldVolume

if ioctlResult = -1  then
  Put_Line( "No mixer installed " );
end if;

newVolume := 75;

ioctlResult := ioctl( mixer_file_id, sound_mixer_write + sound_mixer_volume, newVolume );

-- master volume is 75%

16.17.4 Recording Sound Samples

Recording sounds works is the reverse process of playing sounds. Open /dev/dsp for reading, and it returns sound data. However, before you can begin recording from /dev/dsp, you need to describe how you want the recording done. You need to select the input source, sample format (sometimes called as number of bits), number of channels (mono/stereo) , and the sampling rate (speed). These are set by using the ioctl function on the /dev/dsp file.

To select the input source, you'll need to use /dev/mixer.

[Not finished--KB

Sound_Mixer_Recsrc : constant integer := ?;
Sound_Mixer_Read : constant integer := ?;
Sound_Mixer_Write : constant integer := ?;

newInputSource := Sound_Mixer_Mic;

ioctlResult := ioctl( mixer_file_id, sound_mixer_write + sound_mixer_recsrc, newInputSource
);

Common formats

/* Audio data formats (Note! U8=8 and S16_LE=16 for compatibility) */

sndctl_dsp_setfmt : constant integer := ?;

newFormat : integer;

newFormat := 16#0000010#;

ioctlResult := ioctl( dsp_id, sndctl_dsp_setfmt, newFormat );
-- recording format now 16 bit signed little endian

if newFormat /= 16#00000010  then
  Put_Line( "Sound card doesn't support recording format" );
end if;

Selecting mono or stereo recording is a matter of 0 or 1 respectively.

sndctl_dsp_stereo : constant integer := ?;

stereo : integer;

stereo := 1;

...

ioctlResult := ioctl( dsp_id, sndctl_dsp_stereo, stereo );
-- recording format now stereo

if stereo /= 1 then
  Put_Line( "Sound card doesn't support stereo" );
end if;

Finally, select a sampling rate.

sndctl_dsp_speed : constant integer := ?;

newSpeed : integer;

newSpeed:= 44100;

ioctlResult := ioctl( dsp_id, sndctl_dsp_speed, newFormat );
-- recording now CD quality sampling speed

if newSpeed /= 44100  then
  Put_Line( "Sound card doesn't support sampling speed" );
end if;

Now read /dev/dsp for the raw sound data. If you want to save the sound as an .au file, you'll have to create the .au header information to attach to the sound data.

16.18 Audio CDs

16.19 Kernel Pipes

16.20 Shared Memory

Shared Memory Flags

IPC_PRIVATE indicates no key is supplied.

16.21 Message Queues

Message queues are one of three IPC (Inter Process Communication) ways of System V. The others are shared memory and semaphores. Message queues are linked lists of messages maintained by the kernel. A process can set one up, disappear, and the queue still remains.

Are they usefull?
Yes, they provide a fairly simple way of passing messages between processes. They are also very fast.

We'll look at a simple case where two processes will pass a message between each other.

First we'll need a System V IPC key. Ftok generates a, almost always uniqe, key by gathering some info from a user provided file. This key is needed when we create a queue.

Ftok proberbly means "File To Key".

Although Ada integers and C "int" types are identical, we'll use the Interfaces.C package for maximum portability.

  -- C is key_t ftok(const char *pathname, int proj_id);

  package C renames Interfaces.C;
  type Key_t is new C.Int;
  pragma Convention (C, Key_t);
  function C_Ftok(Pathname : in String; Proj : in C.Int) return Key_t;
  pragma Import(C,C_Ftok,"ftok");

By calling C_Ftok, with Proj greater than 0, we get a key, or -1 for error. We now wants to create a message queue, with this key.

Mesget can be called with a lot of options, but we'll go for getting an id for a queue, and if it does not exists, we create it.

  int msgget(key_t key, int msgflg);

  IPC_CREAT       : constant C.Int :=  512 ;
  IPC_PERMISSIONS : constant C.Int := 8#660#;
  function C_Msgget(Key : Key_t; Msgflg : C.Int) return C.Int;
  pragma Import (C, C_Msgget, "msgget");

By calling msgget with IPC_CREAT + IPC_PERMISSIONS, and the generated key, we get the identity of a message queue, that either exists, or is newly created with the corresponding permissons. The execute flag has no meaning for message queues.

Now we want to do something, but first, have a look at 'ipcs -q'. This command lists message queues in the system.

We send a record that looks like this

      struct msgbuf {
        long    mtype;   /* message type, must be > 0 */
        char    mtext[1];        /* message data */
       };
  

    type Message_Type is record
      M_Type         : C.Long := 100;
      An_Integer     : Integer;
      Anther_Integer : Integer;
    end record;

    function C_Send(Queue_Identity : in C.Int;
                    Message        : in Message_Type) return C.Int is

    type Message_Pointer_Type is access all Message_Type;

    Tmp_Msg     : aliased Message_Type := Message;
    Tmp_Msg_Ptr : Message_Pointer_Type := Tmp_Msg'Access;
    -- All 'size are in bits. There are System.Storage_Unit bits in a byte
    Local_Size : C.Int := C.Int((C.Long'Size + 2 * Integer'Size)/System.Storage_Unit);

    function C_Msgsnd(Msqid : C.Int; Msgp   : Message_Pointer_Type;
                      Msgsz : C.Int; Msgflg : C.Int) return C.Int;
    pragma Import (C, C_Msgsnd, "msgsnd");
  begin
    return C_Msgsnd(Queue_Identity,Tmp_Msg_Ptr,Local_Size,0);
  end C_Send;
  

msgrec

When receiving from a queue, we can use fifo-order or just look at messages of a certain kind. This is determined by the Msg_Type parameter. 0 means fifo, greater than 0 means first message of that type, less than 0 means message with lowest type less than or equal to the absolute value of Msg_Type.

    ssize_t msgrcv(int msqid, struct msgbuf *msgp, size_t msgsz, long msgtyp, int msgflg);

    function C_Receive(Queue_Identity : in C.Int;
                       Msg_Type       : in C.Long;
                       Msg_Flag       : in C.Int := 0) return Message_Type is

      type Message_Pointer_Type is access all Message_Type;
      Receive_Failure : exception;
      function C_Msgrcv(Msqid   : C.Int; Msgpointer : Message_Pointer_Type;
                        Msgsize : C.Int; Msgtype    : C.Long;
                        Msgflag : C.Int)                 return C.Int;
      pragma Import (C, C_Msgrcv, "msgrcv");

      Tmp_Msg     : aliased Message_Type;
      Tmp_Msg_Ptr : Message_Pointer_Type := Tmp_Msg'Access;
      Result      : C.Int                := C.Int'First;
    begin
      Result := C_Msgrcv(Queue_Identity, Tmp_Msg_Ptr,
                         C.Int((Message_type'Size + C.Long'Size)/System.Storage_Unit),
                         Msg_Type, Msg_Flag);
      if (Result >= 0) then
        return Tmp_Msg_Ptr.all;
      else
        raise Receive_Failure;
      end if;
    end C_Receive;
  

msgctl

With msgctl, you can examine and remove existing message queues.

Putting it all together

We'll do 2 simple programs, the first sends a message to a message queue and creates the queue if it does not exists. Then the program exits. The second program retrieves a message from a queue. If the queue does not exists, it creates the queue, and blocks until a message arrives. It then prints out the sum of the fields in the record. The programs was tested with Mandrake 9 and Gnat 3.15p. 4 files are involved:

-- test_def.ads
--
-- a package to provide a simple message queue binding

with Interfaces.C;
package Test_Def is
  package C renames Interfaces.C;

  type Key_t is new C.Int;
  pragma Convention (C, Key_t);

  IPC_CREAT       : constant C.Int :=  512 ;
  IPC_PERMISSIONS : constant C.Int := 8#660#;

  type Message_Type is record
    M_Type         : C.Long := 100;
    An_Integer     : Integer;
    Another_Integer : Integer;
  end record;

  Receive_Failure : exception;

  function C_Ftok(Pathname : in String; Proj : in C.Int) return Key_t;
  pragma Import(C,C_Ftok,"ftok");

  function C_Msgget(Key : Key_t; Msgflg : C.Int) return C.Int;
  pragma Import (C, C_Msgget, "msgget");

  function C_Send(Queue_Identity : in C.Int;
                  Message        : in Message_Type) return C.Int ;
  function C_Receive(Queue_Identity : in C.Int;
                     Msg_Type       : in C.Long;
                     Msg_Flag       : in C.Int := 0) return Message_Type;
end Test_Def;
  

-- test_def.adb package body
--
-- a package to provide a simple message queue binding

with Ada.Text_IO;
with System;
  package body Test_Def is
  use C;

  function C_Send(Queue_Identity : in C.Int;
                  Message        : in Message_Type) return C.Int is
    -- Send a message through the message queue.  Wrapper function for msgsnd.

    type Message_Pointer_Type is new System.Address;

    Tmp_Msg     : aliased Message_Type := Message;
    Tmp_Msg_Ptr : Message_Pointer_Type := Tmp_Msg'Address;
    -- All 'size are in bits. There are System.Storage_Unit bits in a byte
    Local_Size : C.Int := C.Int((C.Long'Size + 2 * Integer'Size)/System.Storage_Unit);

    function C_Msgsnd(Msqid : C.Int; Msgp   : Message_Pointer_Type;
                      Msgsz : C.Int; Msgflg : C.Int) return C.Int;
    pragma Import (C, C_Msgsnd, "msgsnd");

  begin
    return C_Msgsnd(Queue_Identity,Tmp_Msg_Ptr,Local_Size,0);
  end C_Send;

  function C_Receive(Queue_Identity : in C.Int;
                     Msg_Type       : in C.Long;
                     Msg_Flag       : in C.Int := 0) return Message_Type is
    -- Receive a message from a message queue.  Wrapper function for msgrcv.

    type Message_Pointer_Type is new System.Address;

    function C_Msgrcv(Msqid   : C.Int; Msgpointer : Message_Pointer_Type;
                      Msgsize : C.Int; Msgtype    : C.Long;
                      Msgflag : C.Int)                return C.Int;
    pragma Import (C, C_Msgrcv, "msgrcv");

    Tmp_Msg     : aliased Message_Type;
    Tmp_Msg_Ptr : Message_Pointer_Type := Tmp_Msg'Address;
    Result      : C.Int                := C.Int'First;

  begin
    Result := C_Msgrcv(Queue_Identity, Tmp_Msg_Ptr,
                     C.Int((Message_type'Size + C.Long'Size)/System.Storage_Unit),
                     Msg_Type, Msg_Flag);
    Ada.Text_Io.Put_Line("Lenght of message:" & C.Int'Image(Result));
    if Result >= 0 then
      return Tmp_Msg_Ptr.all;
    else
      raise Receive_Failure;
    end if;
  end C_Receive;

end Test_Def;
  

-- test_send.adb
--
-- a program to test our message queue package

with Ada.Text_IO;
with Test_Def; use Test_Def;
procedure Test_Send is
  use C;
  Key  : Key_T ;
  Q_Id : C.Int ;
  Message : Message_Type;
  Result : C.Int;
begin
  Key  := C_Ftok("/etc/profile" & Ascii.NUL,1);
  Q_Id := C_Msgget(Key, IPC_CREAT + IPC_PERMISSIONS);

  Message.An_Integer := 40;
  Message.Another_Integer := 2;
  Result := C_Send(Q_Id,Message);
  Ada.Text_Io.Put_Line("Created/got hold of Key:" & Key_T'Image(Key) &
  " Q-id:" & C.Int'Image(Q_Id) & "Result sending:" & C.Int'Image(Result));

  Ada.Text_Io.Put_Line("Check with 'ipcs -q'");
end Test_Send;
  

-- test_receive.adb
--
-- another program to test our message queue package

with Ada.Text_IO;
with Test_Def; use Test_Def;
procedure Test_Receive is
  use C;
  Key  : Key_T ;
  Q_Id : C.Int ;
  Message : Message_Type;
begin
  Key  := C_Ftok("/etc/profile" & Ascii.NUL,1);
  Ada.Text_Io.Put_Line("Received with  Key:" & Key_T'Image(Key) );

  Q_Id := C_Msgget(Key, IPC_CREAT + IPC_PERMISSIONS);
  Ada.Text_Io.Put_Line("Q-id:" & C.Int'Image(Q_Id));

  Message := C_Receive(Q_Id, Message.M_Type);
  Ada.Text_Io.Put_Line("Received with  Key:" & Key_T'Image(Key) &
   " Q-id:" & C.Int'Image(Q_Id) & " The sum of the two fields are:" &
   Integer'Image(Message.An_Integer + Message.Another_Integer));

  Ada.Text_Io.Put_Line("Check with 'ipcs -q'");
end Test_Receive;
  

Compile with gnatmake test_send.adb and gnatmake test_receive.adb. Run test_send first, check with 'ipcs -q' and then run test_receive. remove the queue with 'ipcrm'

16.22 Semaphores

16.23 Sockets

Send (sentto and sendmsg) are supersets of write. When you use write on a socket, it's actually implemented using the send family.

Write will not work on UDP because it's connectionless. Use send to specify an address everytime.

Protocol Families
PF_INET Internet (IPv4)
PF_INET6 Internet (IPv6)
PF_IPX Novell
PF_NETLINK Kernel user interface device
PF_X25 ITU-T X.25 / ISO-8208
PF_AX25 Amateur radio AX.25
PF_ATMPVC Access to raw ATM PVCs
PF_APPLETALK Appletalk
PF_PACKET Low-level packet interface

Socket Types
SOCK_STREAM Two-way reliable connection, with possible out-of-band transmission (eg. TCP/IP)
SOCK_DGRAM (Datagram) Connectionless, unreliable messages (eg. UDP/IP)
SOCK_SEQPACKET Sequenced, reliable datagram connection.
SOCK_RAQ Raw network protocol access.
SOCK_RDM Reliable, unordered datagrams.

function socket( domain, soctype, protocol : integer ) return integer;
pragma import( C, socket );
Creates a network socket for protocol family domain, connection type soctype, and a protocol (0 uses the default protocol).  Returns -1 on an error, or else a kernel file descriptor for the socket.
Example: mySocket := socket( PF_INET, SOCK_STREAM, 0 ); -- open a standard Internet socket

procedure connect( result : out integer; socket : integer; addr : in out socketaddr; addrlen : integer );
pragma import( C, connect );
pragma import_valued_procedure( connect );
Connects to a server on the network.  socket is the socket to use; addr is the machine and service to connect to; addrlen is the length of the addr record.  Returns -1 on an error, 0 for success.
Example: connect( result, webserver, webserver'size/8 ); -- connect to the web server described by the webserver record

This section ends with a demonstration of how to get a web page off the Internet.

with Ada.Text_IO, Interfaces.C, System.Address_To_Access_Conversions;
use  Ada.Text_IO, Interfaces.C;

procedure websocket is

  -- A program to fetch a web page from a server


  -- Socket related definitions
  --
  -- These are the kernel calls and types we need to create
  -- and use a basic Internet socket.

  type aSocketFD is new int;

  -- a socket file descriptor is an integer -- man socket
  -- make this a new integer for strong typing purposes

  type aProtocolFamily is new unsigned_short;
  AF_INET : constant aProtocolFamily := 2;

  -- Internet protocol PF_Net defined as 2 in
  -- /usr/src/linux/include/linux/socket.h
  -- Make this a new integer for strong typing purposes

  type aSocketType is new int;
  SOCK_STREAM : constant aSocketType := 1;

  -- this is for a steady connection.  Defined as 1 in
  -- /usr/src/linux/include/linux/socket.h
  -- Make this a new integer for strong typing purposes

  type aNetProtocol is new int;
  IPPROTO_TCP : constant aNetProtocol := 6;

  -- The number of the TCP/IP protocol
  -- TCP protocol defined as 6 in /etc/protocols
  -- See man 5 protocols
  -- Make this a new integer for strong typing purposes

  type aNetDomain is new integer;
  PF_INET : constant aNetDomain := 2;

  -- The number of the Internet domain
  -- Make this a new integer for strong typing purposes

  type aInAddr is record
       addr : unsigned := 0;
  end record;
  for aInAddr'size use 96;
  -- A sockaddr_in record is defined as 16 bytes long (or 96 bits)
  -- Request Ada to use 16 bytes to represent this record

  type aSocketAddr is record
       family : aProtocolFamily := AF_INET; -- protocol (AF_INET for TCP/IP)
       port   : unsigned_short := 0;  -- the port number (eg 80 for web)
       ip     : aInAddr;              -- IP number
  end record;
  -- an Internet socket address
  -- defined in /usr/src/linux/include/linux/socket.h
  -- and /usr/src/linux/include/linux/in.h

  function socket( domain   : aNetDomain;
                   stype    : aSocketType;
                   protocol : aNetProtocol )
    return aSocketFD;
  pragma import( C, socket );
  -- initialize a communication socket.  -1 if error

  procedure bind( result : out int; sockfd : aSocketFD;
    sa : in out aSocketAddr; addrlen : int );
  pragma import( C, bind );
  pragma import_valued_procedure( bind );
  -- give socket a name. 0 if successful

  procedure Connect( result : out int; socket : aSocketFD;
    sa : in out aSocketAddr; addrlen : int );
  pragma import( C, connect );
  pragma import_valued_procedure( connect );
  -- connect to a (Internet) server.  0 if successful

  procedure Close( fd : aSocketFD );
  pragma import( C, close );
  -- close the socket, discard the integer result

  procedure Read( result : out integer; from : aSocketFD; buffer : in out string;
    buffersize : integer );
  pragma import( C, read );
  pragma import_valued_procedure( read );
  -- read from a socket

  procedure Write( result : out integer; from : aSocketFD;
    buffer : system.address; buffersize : integer );
  pragma import( C, write );
  pragma import_valued_procedure( write );
  -- write to a socket

  package addrListPtrs is new System.Address_To_Access_Conversions( System.Address );
  -- We need to use C pointers with the address list because this is
  -- a pointer to a pointer in C.  This will allow us to dereference
  -- the C pointers in Ada.

  subtype addrListPtr is System.Address;
  -- easier to read than System.Address

  type aHostEnt is record
       h_name      : System.Address;    -- pointer to offical name of host
       h_aliases   : System.Address;    -- pointer to alias list
       h_addrtype  : int     := 0;      -- host address type (PF_INET)
       h_length    : int     := 0;      -- length of address
       h_addr_list : addrListPtr;       -- pointer to list IP addresses
                                        -- we only want first one
  end record;
  -- defined in man gethostbyname

  package HEptrs is new System.Address_To_Access_Conversions( aHostEnt );
  -- Again, we need to work with C pointers here
  subtype aHEptr is System.Address;
  -- and this is easier to read
  use HEptrs;
  -- use makes = (equals) visible

  function getHostByName( cname : string ) return aHEptr;
  pragma import( C, getHostByName );
  -- look up a host by it's name, returning the IP number

  function htons( s : unsigned_short ) return unsigned_short;
  pragma import( C, htons );
  -- acronym: host to network short -- on Intel x86 platforms,
  -- switches the byte order on a short integer to the network
  -- Most Significant Byte first standard of the Internet

  procedure memcpy( dest, src : System.Address; numbytes : int );
  pragma import( C, memcpy);
  -- Copies bytes from one C pointer to another.  We could probably
  -- use unchecked_conversion, but the C examples use this.

  errno : int;
  pragma import( C, errno );
  -- last error number

  procedure perror( s : string );
  pragma import( C, perror );
  -- print the last kernel error and a leading C string

  procedure PutIPNum( ia : aInAddr ) is
    -- divide an IP number into bytes and display it
    IP : unsigned := ia.addr;
    Byte1, Byte2, Byte3, Byte4 : unsigned;
  begin
    Byte4 := IP mod 256;
    IP := IP / 256;
    Byte3 := IP mod 256;
    IP := IP / 256;
    Byte2 := IP mod 256;
    Byte1 := IP / 256;
    Put( Byte4'img );
    Put( "." );
    Put( Byte3'img );
    Put( "." );
    Put( Byte2'img );
    Put( "." );
    Put( Byte1'img );
  end PutIPNum;

  procedure SendHTTPCommand( soc : aSocketFD; cmd : string ) is
  -- Write a HTTP command string to the socket
    amountWritten : integer := 0;
    totalWritten  : integer := 0;
    position      : integer := cmd'first;
  begin
    loop
          Write( amountWritten, soc, cmd( position )'address,
            cmd'length - integer( totalWritten ) );
          if amountWritten < 0 then
             Put_Line( Standard_Error, "Write to socket failed" );
             return;
          end if;
          Put_Line( "Sent" & amountWritten'img & " bytes" );
          totalWritten := totalWritten + amountWritten;
          position := position + amountWritten;
    exit when totalWritten = cmd'length;
    end loop;
  end SendHTTPCommand;

  procedure ShowWebPage( soc : aSocketFD ) is
  -- Read the web server's response and display it to the screen
    amountRead : integer := 1;
    buffer : string( 1..80 );
  begin
    -- continue reading until an error or no more data read
    -- up to 80 bytes at a time
    while amountRead > 0 loop
       Read( amountRead, soc, buffer, buffer'length );
       if amountRead > 0 then
          Put( buffer( 1..amountRead ) );
       end if;
    end loop;
  end ShowWebPage;

  ServerName: string := "www.adapower.com";

  mySocket    : aSocketFD;     -- the socket
  myAddress   : aSocketAddr;   -- where it goes
  myServer    : aHEptr;        -- IP number of server
  myServerPtr : HEptrs.Object_Pointer;
  addrList    : addrListPtrs.Object_Pointer;
  Result      : int;

begin

  Put_Line( "Socket Demonstration" );
  New_Line;
  Put_Line( "This program opens a socket to a web server" );
  Put_Line( "and retrieves the server's home page" );
  New_Line;

  -- initialize a new TCP/IP socket
  -- 0 for the third param lets the kernel decide

  Put_Line( "Initializing a TCP/IP socket" );
  Put_Line( "Socket( " & PF_INET'img & ',' & SOCK_STREAM'img &
     ", 0 );" );

  mySocket := Socket( PF_INET, SOCK_STREAM, 0 );
  if mySocket = -1 then
     perror( "Error making socket" & ASCII.NUL );
     return;
  end if;
  New_Line;

  -- Lookup the IP number for the server

  Put_Line( "Looking for information on " & ServerName );
  Put_Line( "GetHostByName( " & ServerName & ");" );

  myServer := GetHostByName( ServerName & ASCII.NUL );
  myServerPtr := HEptrs.To_Pointer( myServer );
  if myServerPtr = null then
     Put_Line( Standard_Error, "Error looking up server" );
     return;
  end if;

  Put_Line( "IP number is" & myServerPtr.h_length'img & " bytes long" );
  addrList := addrlistPtrs.To_Pointer( myServerPtr.h_addr_list );
  New_Line;

  -- Create the IP, port and protocol information

  Put_Line( "Preparing connection destination information" );
  myAddress.family := AF_INET;
  myAddress.port   := htons( 80 );
  memcpy( myAddress.ip'address, addrlist.all, myServerPtr.h_length );
  New_Line;

  -- Open a connection to the server

  Put_Line( "Connect( Result, Socket, Family/Address rec, F/A rec size )" );

  Connect( Result, mySocket, myAddress, myAddress'size/8 );

  Put( "Connect( " & Result'img & "," );
  Put(  myAddress.family'img & "/" );
  PutIPNum( myAddress.ip );
  Put(  "," & integer'image( myAddress'size / 8 ) & ")" );
  if Result /= 0 then
     perror( "Error connecting to server" & ASCII.NUL );
     return;
  end if;
  New_Line;

  -- Write the request
  --
  -- "GET" returns a web page from a web server
  -- Also send minimal HTTP header using User-Agent
  -- Followed with a blank line to indicate end of command

  Put_Line( "Transmitting HTTP command..." );
  SendHTTPCommand( mySocket,
     "GET /index.html HTTP/1.0" & ASCII.CR & ASCII.LF &
     "User-Agent: WebSocket/1.0 (BigBookLinuxAda Example)" & ASCII.CR & ASCII.LF
     & ASCII.CR & ASCII.LF );
  New_Line;

  -- read web page back

  Put_Line( "---Web Page / Server Results-------------------------" );
  ShowWebPage( mySocket );
  Put_Line( "-----------------------------------------------------" );

  -- close the connection

  Put_Line( "Closing the connection" );
  close( mySocket );

  Put_Line( "Demonstration finished - have a nice day" );
  
end websocket;

Socket Demonstration

This program opens a socket to a web server
and retrieves the server's home page

Initializing a TCP/IP socket
Socket(  2, 1, 0 );

Looking for information on www.adapower.com
GetHostByName( www.adapower.com);
IP number is 4 bytes long

Preparing connection destination information

Connect( Result, Socket, Family/Address rec, F/A rec size )
Connect(  0, 2/ 216. 92. 66. 46, 16)
Transmitting HTTP command...
Sent 81 bytes

---Web Page / Server Results-------------------------
HTTP/1.0 200 OK
Date: Wed, 29 Mar 2000 02:32:56 GMT
Server: Apache/1.3.3
Last-Modified: Thu, 11 Nov 1999 02:03:14 GMT
Etag: "1f31-406-382a23e2"
Accept-Ranges: Bytes
Content-Length: 1030
Content-Type: text/html
Age: 39
Via: HTTP/1.0 csmc2 (Traffic-Server/3.0.3 [uScHs f p eN:t cCHi p s ])

<!DOCTYPE HTML PUBLIC "-//IETF//DTD HTML//EN">
<HTML>
<HEAD>
   <META NAME="author" CONTENT="David Botton">>
   <META NAME="keywords" CONTENT="Ada AdaPower power source code free treasury repository">
   <META NAME="description" CONTENT="The Ada Source Code Treasurey contains components, procedures, algorithms and articles for Ada developers.">
   <META http-equiv="Page-Enter" CONTENT="revealtrans(duration=2.0, transition=3)">
   <TITLE>AdaPower.com</TITLE>
   <LINK href="mailto:David@Botton.com" rev="made">
</HEAD>
<FRAMESET COLS="120,*" FRAMEBORDER=0 FRAMESPACING=0 BORDER=0>
	<FRAME SRC="buttons.html" name="menu" frameborder=0 marginheight=0 marginwidth=0 noresize scrolling=auto border=0>
	<FRAME SRC="http://www.adapower.com/body.html" name="body" frameborder=0 marginheight=5 marginwidth=0 noresize scrolling=auto border=0>
<NOFRAMES>
<meta HTTP-EQUIV="REFRESH" CONTENT="0; URL="body.html">
<body bgcolor="#ffffff" text="#000000">
<a href="body.html">Click here</a>
</BODY>
</BODY>
</NOFRAMES>
</FRAMESET>

</HTML>
-----------------------------------------------------
Closing the connection
Demonstration finished - have a nice day

16.24 Memory Management

Other mmap flags:

  PROT_NONE     : constant aProtection := 0;     -- shorthand for no access
  PROT_READ     : constant aProtection := 1;     -- read access allowed
  PROT_WRITE    : constant aProtection := 2;     -- write access allowed
  PROT_EXEC     : constant aProtection := 4;     -- execute access allowed

  MAP_SHARED    : constant aMapFlag := 16#01#;   -- share changes with child processes
                                                 -- (write changes to the file, if any)
  MAP_PRIVATE   : constant aMapFlag := 16#02#;   -- separate copy for child processes
                                                 -- (changes kept in memory, if any)
  MAP_FIXED     : constant aMapFlag := 16#10#;   -- use specified address
  MAP_ANON      : constant aMapFlag := 16#20#;   -- just alloc memory, no related file
  MAP_ANONYMOUS : constant aMapFlag := MAP_ANON; -- another name for MAP_ANON
  MAP_GROWSDOWN : constant aMapFlag := 16#0100#; -- stack-line usage
  MAP_DENYWRITE : constant aMapFlag := 16#0800#; -- write lock the file
  MAP_EXECUTABLE: constant aMapFlag := 16#1000#; -- mark as executable
  MAP_LOCKED    : constant aMapFlag := 16#2000#; -- don't swap out memory
  MAP_NORESERVE : constant aMapFlag := 16#4000#; -- don't check for reservations

function msync( start : system.address; size : length;
flags : aMSyncFlag ) return integer;
pragma import( C, msync );
Updates the file associated with the memory allocated by mmap. Returns -1 on an error.

  MS_ASYNC     : constant aSyncFlag := 1;       -- request memory to be saved soon
  MS_INVALIDATE: constant aSyncFlag := 2;       -- mark cache as needing updating
  MS_SYNC      : constant aSyncFlag := 4;       -- save memory to file immediately

function mlock( start : system.address; size : long_integer ) return integer;
pragma import( C, mlock );
Deny page swapping on this block of memory allocated by mmap. Only a superuser process may lock pages. Returns -1 on an error.

function munlock( start : system.address; size : long_integer) return integer;
pragma import( C, munlock );
Allow page swapping on this block of memory allocated by mmap. Returns -1 on an error.

function mlockall( flags : aLockFlag ) return integer;
pragma import( C, mlockall );
Deny swapping on all memory for this process. Only a superuser process can lock memory. Returns -1 on an error.

function munlockall return integer;
pragma import( C, mlockall );
Allow swapping on all memory for this process.  Returns -1 on an error.

  MCL_CURRENT : constant aLockFlag := 1; -- lock current blocks
  MCL_FUTURE  : constant aLockFlag := 2; -- lock subsequent blocks

function brk( end_data_segment : system.address ) return integer;
pragma import( C, brk );
Resize the (Intel) data segment to the specified ending address. Returns -1 on an error.

procedure sbrk( increment : long_integer );
pragma import( C, sbrk );
Increase the (Intel) data segment by the specified number of bytes.

16.25 Exit Procedures