Unix system programming in OCaml
Xavier Leroy and Didier Rémy
Translation by Daniel C. Bünzli, Eric Cooper, Eliot Handelman, Priya Hattiangdi, Thad Meyer, Prashanth Mundkur, Richard Paradies, Till Varoquaux, Mark Wong-VanHaren
Proofread by David Allsopp, Erik de Castro Lopo, John Clements, Anil Madhavapeddy, Prashanth Mundkur
Translation coordination & layout by Daniel C. Bünzli.
Please send corrections to daniel.buenzl email@example.com.
Available as a monolithic file, by chapters, and in PDF — sources, darcs repository.
These course notes originate from a system programming course Xavier Leroy taught in 1994 to the first year students of the Master’s program in fundamental and applied mathematics and computer science at the École Normale Supérieure. This earliest version used the Caml-Light  language. For a Master’s course in computer science at the École Polytechnique taught from 2003 to 2006, Didier Rémy adapted the notes to use the OCaml language. During these years, Gilles Roussel, Fabrice Le Fessant and Maxence Guesdon helped to teach the course and also contributed to this document. The new version also brought additions and updates. In ten years, some orders of magnitude have shifted by a digit and the web has left its infancy. For instance, the http relay example, now commonplace, may have been a forerunner in 1994. But, most of all, the OCaml language gained maturity and was used to program real system applications like Unison .
Tradition dictates that Unix system programming must be done in C. For this course we found it more interesting to use a higher-level language, namely OCaml, to explain the fundamentals of Unix system programming.
The OCaml interface to Unix system calls is more abstract. Instead
of encoding everything in terms of integers and bit fields as in C,
OCaml uses the whole power of the ML type system to clearly
represent the arguments and return values of system calls. Hence, it
becomes easier to explain the semantics of the calls instead of losing
oneself explaining how the arguments and the results have to be
en/decoded. (See, for example, the presentation of the system call
wait, page ??.)
Furthermore, due to the static type system and the clarity of its primitives, it is safer to program in OCaml than in C. The experienced C programmer may see these benefits as useless luxury, however they are crucial for the inexperienced audience of this course.
A second goal of this exposition of system programming is to show OCaml performing in a domain out of its usual applications in theorem proving, compilation and symbolic computation. The outcome of the experiment is rather positive, thanks to OCaml’s solid imperative kernel and its other novel aspects like parametric polymorphism, higher-order functions and exceptions. It also shows that instead of applicative and imperative programming being mutually exclusive, their combination makes it possible to integrate in the same program complex symbolic computations and a good interface with the operating system.
These notes assume the reader is familiar with OCaml and Unix shell commands. For any question about the language, consult the OCaml System documentation  and for questions about Unix, read section 1 of the Unix manual or introductory books on Unix like [5, 6].
This document describes only the programmatic interface to the Unix system. It presents neither its implementation, neither its internal architecture. The internal architecture of bsd 4.3 is described in  and of System v in . Tanenbaum’s books [13, 14] give an overall view of network and operating system architecture.
The Unix interface presented in this document is part of the OCaml System available as free software at http://caml.inria.fr/ocaml/.
Functions that give access to the system from OCaml are grouped into two modules. The first module, Sys, contains those functions common to Unix and other operating systems under which OCaml runs. The second module, Unix, contains everything specific to Unix.
In what follows, we will refer to identifiers from the
Unix modules without specifying which modules they come from. That is, we
will suppose that we are within the scope of the directives
open Sys and
open Unix. In complete examples, we explicitly write
open, in order to be truly complete.
Unix modules can redefine certain
identifiers of the
Pervasives module, hiding previous
definitions. For example,
Pervasives.stdin is different from
Unix.stdin. The previous definitions can always be obtained
through a prefix.
To compile an OCaml program that uses the Unix library, do this:
where the program
prog is assumed to comprise of the three modules
mod3. The modules can also be compiled separately:
and linked with:
In both cases, the argument
unix.cma is the
written in OCaml. To use the native-code compiler rather than the
bytecode compiler, replace
If the compilation tool
ocamlbuild is used, simply add the
following line to the
The Unix system can also be accessed from the interactive system,
also known as the “toplevel”. If your platform supports dynamic
linking of C libraries, start an
ocaml toplevel and type in the
Otherwise, you will need to create an interactive system containing the pre-loaded system functions:
This toplevel can be started by:
When running a program from a shell (command interpreter), the shell passes arguments and an environment to the program. The arguments are words on the command line that follow the name of the command. The environment is a set of strings of the form variable=value, representing the global bindings of environment variables: bindings set with setenv var=val for the csh shell, or with var=val; export var for the sh shell.
The arguments passed to the program are in the string array
The environment of the program is obtained by the function
A more convenient way of looking up the environment is to use the
Sys.getenv v returns the value associated with the variable name
the environment, raising the exception
Not_found if this
variable is not bound.
As a first example, here is the
echo program, which prints a
list of its arguments, as does the Unix command of the same name.
A program can be terminated at any point with a call to
The argument is the return code to send back to the calling program. The
convention is to return 0 if all has gone well, and to return a
non-zero code to signal an error. In conditional constructions, the
sh shell interprets the return code 0 as the boolean
“true”, and all non-zero codes as the boolean “false”.
When a program terminates normally after executing all of the
expressions of which it is composed, it makes an implicit call to
exit 0. When a program terminates prematurely because an
exception was raised but not caught, it makes an implicit call to
exit always flushes the buffers of all channels open for
writing. The function
at_exit lets one register other actions
to be carried out when the program terminates.
The last function to be registered is called first. A function registered with
at_exit cannot be unregistered. However, this is not a
real restriction: we can easily get the same effect with a function
whose execution depends on a global variable.
Unless otherwise indicated, all functions in the
raise the exception
Unix_error in case of error.
The second argument of the
Unix_error exception is the name of
the system call that raised the error. The third argument identifies,
if possible, the object on which the error occurred; for example, in
the case of a system call taking a file name as an argument, this file name will be
in the third position in
Unix_error. Finally, the first argument
of the exception is an error code indicating the nature of the
error. It belongs to the variant type
Constructors of this type have the same names and meanings as those
used in the posix convention and certain errors from
unix98 and bsd. All other errors use the constructor
Given the semantics of exceptions, an error that is not specifically
foreseen and intercepted by a
try propagates up to the top of a
program and causes it to terminate prematurely. In small
applications, treating unforeseen errors as fatal is a good practice.
However, it is appropriate to display the error clearly. To do this,
Unix module supplies the
handle_unix_error f x applies function
f to the
x. If this raises the exception
message is displayed describing the error, and the program is
exit 2. A typical use is
where the function
prog : unit -> unit executes the body of the
program. For reference, here is how
Functions of the form
prerr_xxx are like the functions
print_xxx, except that they write on the error channel
stderr rather than on the standard output channel
The primitive error_message, of type
error -> string, returns a message describing the error given as an
argument (line 16). The argument number zero of the
Sys.argv.(0), contains the name of the command
that was used to invoke the program (line 6).
handle_unix_error handles fatal errors, i.e. errors
that stop the program. An advantage of OCaml is that it requires
all errors to be handled, if only at the highest level by
halting the program. Indeed, any error in a system call raises an
exception, and the execution thread in progress is interrupted up to
the level where the exception is explicitly caught and handled. This avoids
continuing the program in an inconsistent state.
Errors of type
Unix_error can, of course, be
selectively matched. We will often see the following
function later on:
which is used to execute a function and to restart it automatically when it executes a system call that is interrupted (see section 4.5).
As we will see throughout the examples, system programming often repeats the same patterns. To reduce the code of each application to its essentials, we will want to define library functions that factor out the common parts.
Whereas in a complete program one knows precisely which errors can be raised (and these are often fatal, resulting in the program being stopped), we generally do not know the execution context in the case of library functions. We cannot suppose that all errors are fatal. It is therefore necessary to let the error return to the caller, which will decide on a suitable course of action (e.g. stop the program, or handle or ignore the error). However, the library function in general will not allow the error to simply pass through, since it must maintain the system in a consistent state. For example, a library function that opens a file and then applies an operation to its file descriptor must take care to close the descriptor in all cases, including those where the processing of the file causes an error. This is in order to avoid a file descriptor leak, leading to the exhaustion of file descriptors.
Furthermore, the operation applied to a file may be defined by a function that was received as an argument, and we don’t know precisely when or how it can fail (but the caller in general will know). We are thus often led to protect the body of the processing with “finalization” code, which must be executed just before the function returns, whether normally or exceptionally.
There is no built-in finalize construct
the OCaml language, but it can be easily defined1:
This function takes the main body
f and the finalizer
finally, each in the form of a function, and two parameters
y, which are passed to their respective functions. The body
of the program
f x is executed first, and its result is kept
aside to be returned after the execution of the finalizer
finally. In case the program fails, i.e. raises an exception
the finalizer is run and the exception
exn is raised
again. If both the main function and the finalizer fail, the
finalizer’s exception is raised (one could choose to have the main
function’s exception raised instead).
In the rest of this course, we use an auxiliary library
which contains several useful functions like
try_finalize that are often
used in the examples. We will introduce them as they are needed. To
compile the examples of the course, the definitions of the
module need to be collected and compiled.
Misc module also contains certain functions, added for
illustration purposes, that will not be used in the course. These
simply enrich the
Unix library, sometimes by redefining the
behavior of certain functions. The
Misc module must thus take
precedence over the
The course provides numerous examples. They can be compiled with OCaml, version 4.00.1. Some programs will have to be slightly modified in order to work with older versions.
There are two kinds of examples: “library functions” (very general functions that can be reused) and small applications. It is important to distinguish between the two. In the case of library functions, we want their context of use to be as general as possible. We will thus carefully specify their interface and attentively treat all particular cases. In the case of small applications, an error is often fatal and causes the program to stop executing. It is sufficient to report the cause of an error, without needing to return to a consistent state, since the program is stopped immediately thereafter.
The term “file” in Unix covers several types of objects:
The file concept includes both the data contained in the file and information about the file itself (also called meta-data) like its type, its access rights, the latest access times, etc.
To a first approximation, the file system can be considered to be a tree. The root is
'/'. The branches are labeled by (file) names,
which are strings of any characters excluding
(but it is good practice to also avoid non-printing characters and
spaces). The non-terminal nodes are directories: these nodes
always contain two branches
.. which respectively
represent the directory itself and the directory’s parent. The other
nodes are sometimes called files, as opposed to directories,
but this is ambiguous, as we can also designate any node as a
“file”. To avoid all ambiguity we refer to them as
The nodes of the tree are addressed by paths. If the start of the path
is the root of the file hierarchy, the path is absolute, whereas if the
start is a directory it is relative. More precisely, a relative
path is a string of file names separated by the character
'/'. An absolute path is a relative path preceded by the
'/' (note the double use of this character both as
a separator and as the name of the root node).
The Filename module handles paths in a portable
manner. In particular, concat concatenates paths without
referring to the character
'/', allowing the code to function equally
well on other operating systems (for example, the path separator character
under Windows is
'\'). Similarly, the
provides the string values current_dir_name and
parent_dir_name to represent the branches
.. The functions basename and
dirname return the prefix
d and the suffix
b from a path
p such that the paths
d/b refer to the same file, where
d is the directory in
which the file is found and
b is the name of the file. The
functions defined in
Filename operate only on paths,
independently of their actual existence within the file hierarchy.
In fact, strictly speaking, the file hierarchy is not a tree. First
.. allow a directory to refer to
itself and to move up in the hierarchy to define paths leading from a
directory to itself. Moreover, non-directory files can have many
parents (we say that they have many hard links). Finally,
there are also symbolic links which can be seen as
non-directory files containing a path. Conceptually, this path can be
obtained by reading the contents of the symbolic link like an ordinary
file. Whenever a symbolic link occurs in the middle of a path we have
to follow its path transparently. If
s is a symbolic link whose
value is the path
l, then the path
p/s/q represents the file
l is an absolute path or the file
l is a relative path.
Figure 1 gives an example of a file hierarchy. The
11 corresponding to the path
path value is the relative path
../gnu, does not refer to any
existing file in the hierarchy (at the moment).
In general, a recursive traversal of the hierarchy will terminate if the following rules are respected:
But if symbolic links are followed we are traversing a graph and we need to keep track of the nodes we have already visited to avoid loops.
Each process has a current working directory. It is returned by the
function getcwd and can be changed with
chdir. It is also possible to constrict the
view of the file hierarchy by calling
p. This makes the node
should be a directory, the root of the restricted view of the
hierarchy. Absolute file paths are then
interpreted according to this new root
p (and of course
.. at the
new root is
There are two ways to access a file. The first is by its file
name (or path name) in the file system hierarchy. Due to
hard links, a file can have many different names. Names are values of
string. For example the system calls unlink,
link, symlink and rename all operate at
the file name level.
Their effect is as follows:
unlink ferases the file
flike the Unix command
rm -f f.
link f1 f2creates a hard link named
f2to the file
f1like the command
ln f1 f2.
symlink f1 f2creates a symbolic link named
f2to the file
f1like the command
ln -s f1 f2.
rename f1 f2renames the file
f2like the command
mv f1 f2.
The second way of accessing a file is by a file descriptor. A descriptor represents a pointer to a file along with other information like the current read/write position in the file, the access rights of the file (is it possible to read? write?) and flags which control the behavior of reads and writes (blocking or non-blocking, overwrite, append, etc.). File descriptors are values of the abstract type file_descr.
Access to a file via its descriptor is independent from the access via its name. In particular whenever we get a file descriptor, the file can be destroyed or renamed but the descriptor still points on the original file.
When a program is executed, three descriptors are allocated and
tied to the variables
stderr of the
They correspond, respectively, to the standard input, standard output and standard error of the process.
When a program is executed on the command line without any
redirections, the three descriptors refer to the terminal. But if,
for example, the input has been redirected using the shell expression
cmd < f, then the descriptor
stdin refers to the file named
during the execution of the command
cmd > f
cmd 2> f respectively bind the descriptors
stderr to the file named
f during the execution of the
The system calls stat, lstat and fstat return the meta-attributes of a file; that is, information about the node itself rather than its content. Among other things, this information contains the identity of the file, the type of file, the access rights, the time and date of last access and other information.
The system calls
lstat take a file name as an
fstat takes a previously opened descriptor and
returns information about the file it points to.
lstat differ on symbolic links :
lstat returns information
about the symbolic link itself, while
stat returns information
about the file that the link points to. The result of these three
calls is a record of type stats whose fields are
described in table 1.
|The id of the device on which the file is stored.|
|The id of the file (inode number) in its partition.
The pair |
|The file type. The type file_kind is an enumerated type
whose constructors are:
|Access rights for the file|
|For a directory: the number of entries in the directory. For others: the number of hard links to this file.|
|The id of the file’s user owner.|
|The id of the file’s group owner.|
|The id of the associated peripheral (for special files).|
|The file size, in bytes.|
|Last file content access date (in seconds from January 1st 1970, midnight, gmt).|
|Last file content modification date (idem).|
|Last file state modification date: either a
write to the file or a change in access rights, user or group owner,
or number of links.
A file is uniquely identified by the pair made of its device number
(typically the disk partition where it is located)
st_dev and its
A file has one user owner
st_uid and one group owner
st_gid. All the users and groups
on the machine are usually described in the
/etc/groups files. We can look up them by
name in a portable manner with the functions getpwnam and
getgrnam or by id with
getpwuid and getgrgid.
The name of the user of a running process and all the groups to which it belongs can be retrieved with the commands getlogin and getgroups.
The call chown changes the owner (second argument) and the group (third argument) of a file (first argument). If we have a file descriptor, fchown can be used instead. Only the super user can change this information arbitrarily.
Access rights are encoded as bits in an integer, and the type
file_perm is just an abbreviation for the type
int. They specify special bits and read, write and
execution rights for the user owner, the group owner and the other
users as vector of bits:
where in each of the user, group and other fields, the order of bits
indicates read (
r), write (
w) and execute (
The permissions on a file are the union of all these individual
rights, as shown in table 2.
|Bit (octal)||Notation ||Access right|
|executable by the user owner|
|writable by the user owner|
|readable by the user owner|
|executable by members of the group owner|
|writable by members of the group owner|
|readable by members of the group owner|
|executable by other users|
|writable by other users|
|readable by other users|
|the bit |
|the bit |
|the bit |
For files, the meaning of read, write and execute permissions is
obvious. For a directory, the execute permission means the right to
enter it (to
chdir to it) and read permission the right to list
its contents. Read permission on a directory is however not needed to
read its files or sub-directories (but we then need to know their
The special bits do not have meaning unless the
x bit is set (if
x set, they do not give additional rights). This
is why their representation is superimposed on the bit
T are used instead of
x is not set. The bit
t allows sub-directories to
inherit the permissions of the parent directory. On a directory,
s allows the use of the directory’s
than the user’s to create directories. For an executable file,
s allows the changing at execution time of the user’s
effective identity or group with the system calls setuid
The process also preserves its original identities unless
it has super user privileges, in which case
setgid change both its effective and original user and group
identities. The original identity is preserved to allow
the process to subsequently recover it as its effective identity
without needing further privileges. The system calls getuid and
getgid return the original identities and
geteuid and getegid return the effective identities.
A process also has a file creation mask encoded the same way file permissions are. As its name suggests, the mask specifies prohibitions (rights to remove): during file creation a bit set to 1 in the mask is set to 0 in the permissions of the created file. The mask can be consulted and changed with the system call umask:
Like many system calls that modify system variables, the modifying function returns the old value of the variable. Thus, to just look up the value we need to call the function twice. Once with an arbitrary value to get the mask and a second time to put it back. For example:
File access permissions can be modified with the system calls chmod and fchmod:
and they can be tested “dynamically” with the system call access:
where requested access rights to the file are specified by a list of
values of type access_permission whose meaning is
obvious except for
F_OK which just checks for the file’s
existence (without checking for the other rights). The function
raises an error if the access rights are not granted.
Note that the information inferred by
access may be more
restrictive than the information returned by
lstat because a file
system may be mounted with restricted rights — for example in
read-only mode. In that case
access will deny a write permission
on a file whose meta-attributes would allow it. This is why we
distinguish between “dynamic” (what a process can actually do)
and “static” (what the file system specifies) information.
Only the kernel can write in directories (when files are created). Thus opening a directory in write mode is prohibited. In certain versions of Unix a directory may be opened in read only mode and read with read, but other versions prohibit it. However, even if this is possible, it is preferable not to do so because the format of directory entries vary between Unix versions and is often complex. The following functions allow reading a directory sequentially in a portable manner:
The system call opendir returns a directory descriptor for a
directory. readdir reads the next entry of a descriptor, and
returns a file name relative to the directory or raises the exception
End_of_file if the end of the directory is
reached. rewinddir repositions the descriptor at the
beginning of the directory and closedir closes the directory
The following library function, in
Misc, iterates a
f over the entries of the directory
To create a directory or remove an empty directory, we have mkdir and rmdir:
The second argument of
mkdir determines the access rights of the
new directory. Note that we can only remove a directory that is
already empty. To remove a directory and its contents, it is thus
necessary to first recursively empty the contents of the directory and
then remove the directory.
The Unix command
find lists the files of a hierarchy matching
certain criteria (file name, type and permissions etc.). In this
section we develop a library function
implements these searches and a command
find that provides a version
of the Unix command
find that supports the options
We specify the following interface for
The function call
traverses the file hierarchy starting from the roots specified in the
roots (absolute or relative to the current directory of the
process when the call is made) up to a maximum depth
depth and following
symbolic links if the flag
follow is set. The paths found under
r as a prefix. Each found path
given to the function
action along with the data returned by
Unix.lstat p (or
Unix.stat p if
action returns a boolean indicating, for
directories, whether the search should continue for its contents (
or not (
handler function reports traversal errors of type
Unix_error. Whenever an error occurs the arguments of the
exception are given to the handler function and the traversal
continues. However when an exception is raised by the functions
handler themselves, we immediately stop the
traversal and let it propagate to the caller. To propagate an
Unix_error exception without catching it like a traversal error,
we wrap these exceptions in the
Hidden exception (see
A directory is identified by the
id pair (line 12)
made of its device and inode number. The list
track of the directories that have already been visited. In fact
this information is only needed if symbolic links are followed
It is now easy to program the
find command. The essential part of
the code parses the command line arguments with the Arg
find command is quite limited, the library
FindLib.find is far more general, as the following
which, starting from the current directory, recursively prints
files without printing or entering directories whose name is
getcwd is not a system call but is defined in the
Unix module. Give a “primitive” implementation of
getcwd. First describe the principle of your algorithm with words
and then implement it (you should avoid repeating the same system
Here are some hints. We move up from the current position towards the
root and construct backwards the path we are looking for. The root can
be detected as the only directory node whose parent is equal to itself
(relative to the root
.. are equal). To find the name
of a directory
r we need to list the contents of its parent
directory and detect the file that corresponds to
openfile function allows us to obtain a descriptor for
a file of a given name (the corresponding system call
is open, however
open is a keyword in OCaml).
The first argument is the name of the file to open. The second argument, a list of flags from the enumerated type open_flag, describes the mode in which the file should be opened and what to do if it does not exist. The third argument of type file_perm defines the file’s access rights, should the file be created. The result is a file descriptor for the given file name with the read/write position set to the beginning of the file.
The flag list must contain exactly one of the following flags:
|Open in read-only mode.|
|Open in write-only mode.|
|Open in read and write mode.|
These flags determine whether read or write calls can be done on the
descriptor. The call
openfile fails if a process requests an open
in write (resp. read) mode on a file on which it has no right to
write (resp. read). For this reason
O_RDWR should not be used
The flag list can also contain one or more of the following values:
|Open in append mode.|
|Create the file if it does not exist.|
|Truncate the file to zero if it already exists.|
|Fail if the file already exists.|
|Open in non-blocking mode.|
|Do not function in console mode.|
|Perform the writes in synchronous mode.|
|Perform the data writes in synchronous mode.|
|Perform the reads in synchronous mode.|
The first group defines the behavior to follow if the file exists or not. With:
O_APPEND, the read/write position will be set at the end of the file before each write. Consequently any written data will be added at the end of file. Without
O_APPEND, writes occur at the current read/write position (initially, the beginning of the file).
O_TRUNC, the file is truncated when it is opened. The length of the file is set to zero and the bytes contained in the file are lost, and writes start from an empty file. Without
O_TRUNC, the writes are made at the start of the file overwriting any data that may already be there.
O_CREAT, creates the file if it does not exist. The created file is empty and its access rights are specified by the third argument and the creation mask of the process (the mask can be retrieved and changed with umask).
openfilefails if the file already exists. This flag, used in conjunction with
O_CREATallows to use files as locks1. A process which wants to take the lock calls
openfileon the file with
O_CREAT. If the file already exists, this means that another process already holds the lock and
openfileraises an error. If the file does not exist
openfilereturns without error and the file is created, preventing other processes from taking the lock. To release the lock the process calls
unlinkon it. The creation of a file is an atomic operation: if two processes try to create the same file in parallel with the options
O_CREAT, at most one of them can succeed. The drawbacks of this technique is that a process must busy wait to acquire a lock that is currently held and the abnormal termination of a process holding a lock may never release it.
Most programs use
0o666 for the third argument
openfile. This means
rw-rw-rw- in symbolic notation.
With the default creation mask of
file is thus created with the permissions
rw-r--r--. With a more
lenient mask of
0o002, the file is created with the permissions
To read from a file:
The third argument can be anything as
O_CREAT is not specified, 0
is usually given.
To write to an empty a file without caring about any previous content:
If the file will contain executable code (e.g. files
ld, scripts, etc.), we create it with execution permissions:
If the file must be confidential (e.g. “mailbox” files where
To append data at the end of an existing file or create it if it doesn’t exist:
O_NONBLOCK flag guarantees that if the file is a named pipe
or a special file then the file opening and subsequent reads and
writes will be non-blocking.
O_NOCTYY flag guarantees that if the file is a control
terminal (keyboard, window, etc.), it won’t become the controlling
terminal of the calling process.
The last group of flags specifies how to synchronize read and write operations. By default these operations are not synchronized. With:
O_DSYNC, the data is written synchronously such that the process is blocked until all the writes have been done physically on the media (usually a disk).
O_SYNC, the file data and its meta-attributes are written synchronously.
O_DSYNCspecifies that the data reads are also synchronized: it is guaranteed that all current writes (requested but not necessarily performed) to the file are really written to the media before the next read. If
O_RSYNCis provided with
O_SYNCthe above also applies to meta-attributes changes.
The system calls read and write read and write
bytes in a file. For historical reasons, the system
write is provided in OCaml under the name
The two calls
single_write have the same
interface. The first argument is the file descriptor to act on. The
second argument is a string which will hold the read bytes (for
read) or the bytes to write (for
single_write). The third
argument is the position in the string of the first byte to be written
or read. The fourth argument is the number of the bytes to be read or
written. In fact the third and fourth argument define a sub-string of
the second argument (the sub-string should be valid,
single_write do not check this).
single_write return the number of bytes actually
read or written.
Reads and write calls are performed from the file descriptor’s current
read/write position (if the file was opened in
this position is set at the end of the file prior to any
write). After the system call, the current position is advanced by
the number of bytes read or written.
For writes, the number of bytes actually written is usually the number of bytes requested. However there are exceptions: (i) if it is not possible to write the bytes (e.g. if the disk is full) (ii) the descriptor is a pipe or a socket open in non-blocking mode (iii) due to OCaml, if the write is too large.
The reason for (iii) is that internally OCaml uses auxiliary
buffers whose size is bounded by a maximal value. If this value is
exceeded the write will be partial. To work around this problem
OCaml also provides the function write which
iterates the writes until all the data is written or an error occurs.
The problem is that in case of error there’s no way to know the number
of bytes that were actually written. Hence
single_write should be
preferred because it preserves the atomicity of writes (we know
exactly what was written) and it is more faithful to the original Unix
system call (note that the implementation of
described in section 5.7).
fd is a descriptor open in write-only mode.
writes the characters
"lo worl" in the corresponding file,
and returns 7.
For reads, it is possible that the number bytes actually read is
smaller than the number of requested bytes. For example when the end
of file is near, that is when the number of bytes between the current
position and the end of file is less than the number of requested
bytes. In particular, when the current position is at the end of file,
read returns zero. The convention “zero equals end of
file” also holds for special files, pipes and sockets. For example,
read on a terminal returns zero if we issue a
ctrl-D on the
Another example is when we read from a terminal. In that case,
read blocks until an entire line is available. If the line length
is smaller than the requested bytes
read returns immediately with
the line without waiting for more data to reach the number of
requested bytes. (This is the default behavior for terminals, but it
can be changed to read character-by-character instead of
line-by-line, see section 2.13 and the type
terminal_io for more details.)
The following expression reads at most 100 characters from standard input and returns them as a string.
really_read below has the same interface as
read, but makes additional read attempts to try to get
the number of requested bytes. It raises the exception
End_of_file if the end of file is reached while doing this.
The system call close closes a file descriptor.
Once a descriptor is closed, all attempts to read, write, or do
anything else with the descriptor will fail. Descriptors should be
closed when they are no longer needed; but it is not mandatory. In
particular, and in contrast to
channels, a file descriptor doesn’t need to be closed to ensure that
all pending writes have been performed as write requests made with
write are immediately transmitted to the kernel. On the other
hand, the number of descriptors allocated by a process is limited by
the kernel (from several hundreds to thousands). Doing a
an unused descriptor releases it, so that the process does not run out
We program a command
file_copy which, given two arguments
f2, copies to the file
f2 the bytes contained
The bulk of the work is performed by the the function
First we open a descriptor in read-only mode on the input file and
another in write-only mode on the output file.
If the output file already exists, it is truncated (option
O_TRUNC) and if it does not exist it is created (option
O_CREAT) with the permissions
rw-rw-rw- modified by the creation
mask. (This is unsatisfactory: if we copy an executable file, we would
like the copy to be also executable. We will see later how to give
a copy the same permissions as the original.)
copy_loop function we do the copy by blocks of
buffer_size bytes. We request
buffer_size bytes to read. If
read returns zero, we have reached the end of file and the copy
is over. Otherwise we write the
r bytes we have read in the
output file and start again.
Finally, we close the two descriptors. The main program
verifies that the command received two arguments and passes them to
Any error occurring during the copy results in a
caught and displayed by
handle_unix_error. Example of errors
include inability to open the input file because it does not
exist, failure to read because of restricted permissions, failure to
write because the disk is full, etc.
Add an option
-a to the program, such that
file_copy -a f1 f2 appends the contents of
f1 to the end of
Parsing the new option from the command line is left to the reader.
In the example
file_copy, reads were made in blocks of 8192
bytes. Why not read byte per by byte, or megabyte per by megabyte?
The reason is efficiency.
Figure 2 shows the copy speed of
bytes per second, against the size of blocks (the value
buffer_size). The amount of data transferred is the same
regardless of the size of the blocks.
For small block sizes, the copy speed is almost proportional to the
block size. Most of the time is spent not in data transfers but in the
execution of the loop
copy_loop and in the calls to
write. By profiling more carefully we can see that most of the
time is spent in the calls to
write. We conclude
that a system call, even if it has not much to do, takes a minimum of
about 4 micro-seconds (on the machine that was used for the test — a
2.8 GHz Pentium 4 ), let us say from 1 to 10 microseconds. For small
input/output blocks, the duration of the system call dominates.
For larger blocks, between 4KB and 1MB, the copy speed is constant and maximal. Here, the time spent in system calls and the loop is small relative to the time spent on the data transfer. Also, the buffer size becomes bigger than the cache sizes used by the system and the time spent by the system to make the transfer dominates the cost of a system call2.
Finally, for very large blocks (8MB and more) the speed is slightly under the maximum. Coming into play here is the time needed to allocate the block and assign memory pages to it as it fills up.
The moral of the story is that, a system call, even if it does very little work, costs dearly — much more than a normal function call: roughly, 2 to 20 microseconds for each system call, depending on the architecture. It is therefore important to minimize the number of system calls. In particular, read and write operations should be made in blocks of reasonable size and not character by character.
In examples like
file_copy, it is not difficult to do
input/output with large blocks. But other types of programs are more
naturally written with character by character input or output (e.g.
reading a line from a file, lexical analysis, displaying a number etc.).
To satisfy the needs of these programs, most systems provide
input/output libraries with an additional layer of software between
the application and the operating system. For example, in OCaml the
Pervasives module defines the abstract types
out_channel, similar to file descriptors, and
functions on these types like input_char,
output_string. This layer uses buffers to
group sequences of character by character reads or writes into a
single system call to read or write. This results in better
performance for programs that proceed character by character.
Moreover this additional layer makes programs more portable: we just
need to implement this layer with the system calls provided by another
operating system to port all the programs that use this library on
this new platform.
To illustrate the buffered input/output techniques, we implement a fragment
Pervasives library. Here is the interface:
We start with the “input” part. The abstract type
in_channel is defined as follows:
The character string of the
in_buffer field is, literally, the
buffer. The field
in_fd is a (Unix) file descriptor, opened on
the file to read. The field
in_pos is the current read position
in the buffer. The field
in_end is the number of valid
characters preloaded in the buffer.
in_end will be modified in place during
read operations; we therefore declare them as
When we open a file for reading, we create a buffer of reasonable size
(large enough so as not to make too many system calls; small enough so
as not to waste memory). We then initialize the field
a Unix file descriptor opened in read-only mode on the given file. The
buffer is initially empty (it does not contain any character from the
file); the field
in_end is therefore initialized to zero.
To read a character from an
in_channel, we do one of two
things. Either there is at least one unread character in the buffer;
that is to say, the field
in_pos is less than the field
in_end. We then return this character located at
in_pos. Or the buffer is empty and we call
refill the buffer. If
read returns zero, we have reached the end
of the file and we raise the exception
End_of_file. Otherwise, we
put the number of characters read in the field
in_end (we may
receive less characters than we requested, thus the buffer may be
only partially refilled) and we return the first character read.
in_channel just closes the underlying Unix file descriptor.
The “output” part is very similar to the “input” part. The only asymmetry is that the buffer now contains incomplete writes (characters that have already been buffered but not written to the file descriptor), and not reads in advance (characters that have buffered, but not yet read).
To write a character on an
out_channel, we do one of two things.
Either the buffer is not full and we just store the character in the
buffer at the position
out_pos and increment that value. Or the
buffer is full and we empty it with a call to
write and then
store the character at the beginning of the buffer.
When we close an
out_channel, we must not forget to write the
buffer contents (the characters from 0 to
out_pos - 1) to the
file otherwise the writes made on the channel since the last time
the buffer was emptied would be lost.
which behaves like a sequence of
output_char on each
character of the string, but is more efficient.
The idea is to copy the string to output into the buffer. We need to take into account the case where there is not enough space in the buffer (in that case the buffer needs to emptied), and also the case where the string is longer than the buffer (in that case it can be written directly). Here is a possible solution.
The system call lseek allows to set the current read/write position of a file descriptor.
The first argument is the file descriptor and the second one the desired position. The latter is interpreted according to the value of the third argument of type seek_command. This enumerated type specifies the kind of position:
|Absolute position. The second argument specifies the character number to point to. The first character of a file is at position zero.|
|Position relative to the current position. The second argument is an offset relative to the current position. A positive value moves forward and a negative value moves backwards.|
|Position relative to the end of file. The
second argument is an offset relative to the end of file.
As for |
The value returned by
lseek is the resulting absolute
An error is raised if a negative absolute position is
requested. The requested position can be located after the end
of file. In that case, a
read returns zero (end of
file reached) and a
write extends the file with zeros until
that position and then writes the supplied data.
To position the cursor on the 1000th character of a file:
To rewind by one character:
To find out the size of a file:
For descriptors opened in
O_APPEND mode, the read/write position
is automatically set at the end of the file before each write. Thus
lseek is useless to set the write position, it may however
be useful to set the read position.
The behavior of
lseek is undefined on certain type of files for
which absolute access is meaningless: communication devices (pipes,
sockets) but also many special files like the terminal.
In most Unix implementations a call to
lseek on these files is
simply ignored: the read/write position is set but read/write
operations ignore it. In some implementations,
lseek on a pipe or
a socket triggers an error.
tail displays the last n lines of a file.
How can it be implemented efficiently on regular files? What can we
do for the other kind of files? How can the option
A naive implementation of
tail is to read the file sequentially
from the beginning, keeping the last n lines read in a circular
buffer. When we reach the end of file, we display the buffer.
When the data comes from a pipe or a special file which
does not implement
lseek, there is no better way.
However if the data is coming from a normal file, it is better to read
the file from the end. With
lseek, we read the last 4096
characters. We scan them for the end of lines. If there are at least
n of them, we output and display the corresponding lines.
Otherwise, we start again by adding the next preceding 4096
To add the option
-f, we first proceed as above and then we go
back at the end of the file and try to
read from there. If
read returns data we display it immediately and start again. If it
0 we wait some time (
sleep 1) and try again.
In Unix, data communication is done via file descriptors representing either permanent files (files, peripherals) or volatile ones (pipes and sockets, see chapters 5 and 6). File descriptors provide a uniform and media-independent interface for data communication. Of course the actual implementation of the operations on a file descriptor depends on the underlying media.
However this uniformity breaks when we need to access all the features provided by a given media. General operations (opening, writing, reading, etc.) remain uniform on most descriptors but even, on certain special files, these may have an ad hoc behavior defined by the kind of peripheral and its parameters. There are also operations that work only with certain kind of media.
We can shorten a normal file with the system calls truncate and ftruncate.
The first argument is the file to truncate and the second the desired size. All the data after this position is lost.
Most operations on files “follow” symbolic links in the sense that they do not apply to the link itself but to the file on which the link points (for example openfile, stat, truncate, opendir, etc.).
The two system calls symlink and readlink operate specifically on symbolic links:
symlink f1 f2 creates the file
f2 as a symbolic
f1 (like the Unix command
ln -s f1 f2). The call
readlink returns the content of a symbolic link, i.e. the name of
the file to which the link points.
Special files can be of “character” or “block” type. The former are character streams: we can read or write characters only sequentially. These are the terminals, sound devices, printers, etc. The latter, typically disks, have a permanent medium: characters can be read by blocks and even seeked relative to the current position.
Among the special files, we may distinguish:
|This is the black hole which swallows
everything we put into and from which nothing comes out. This is
extremely useful for ignoring the results of a process: we redirect
its output to |
|These are the control terminals.|
|These are the pseudo-terminals: they are not real terminals but simulate them (they provide the same interface).|
|These are the disks.|
|Under Linux, system parameters organized as a file system. They allow reads and writes.|
The usual file system calls on special files can behave differently.
However, most special files (terminals, tape drives, disks, etc.)
write in the obvious manner (but
sometimes with restrictions on the number of bytes written or read),
but many ignore lseek.
In addition to the usual file system calls, special files which represent peripherals must be commanded and/or configured dynamically. For example, for a tape drive, rewind or fast forward the tape; for a terminal, choice of the line editing mode, behavior of special characters, serial connection parameters (speed, parity, etc.). These operations are made in Unix with the system call ioctl which group together all the particular cases. However, this system call is not provided by OCaml; it is ill-defined and cannot be treated in a uniform way.
Terminals and pseudo-terminals are special files of type character which can be configured from OCaml. The system call tcgetattr takes a file descriptor open on a special file and returns a structure of type terminal_io which describes the status of the terminal according to the posix standard.
This structure can be modified and given to the function tcsetattr to change the attributes of the peripheral.
The first argument is the file descriptor of the peripheral. The last
argument is a structure of type
terminal_io describing the
parameters of the peripheral as we want them. The second argument is a
value of the enumerated type setattr_when that
indicates when the change must be done: immediately (
after having transmitted all written data (
TCSADRAIN) or after
having read all the received data (
recommended for changing write parameters and
TCSAFLUSH for read
When a password is read, characters entered by the user should not be echoed if the standard input is connected to a terminal or a pseudo-terminal.
read_passwd function starts by getting the current settings
of the terminal connected to
stdin. Then it defines a modified
version of these in which characters are not echoed. If this fails the
standard input is not a control terminal and we just read a
line. Otherwise we display a message, change the terminal settings, read the
password and put the terminal back in its initial state. Care must be
taken to set the terminal back to its initial state even after a read
Sometimes a program needs to start another and connect its standard input
to a terminal (or pseudo-terminal). OCaml does not provide any
support for this3. To achieve that, we must manually look among the
pseudo-terminals (in general, they are files with names in the form of
/dev/tty[a-z][a-f0-9]) and find one that is not already open. We
can then open this file and start the program with this file on its
Four other functions control the stream of data of a terminal (flush waiting data, wait for the end of transmission and restart communication).
The function tcsendbreak sends an interrupt to the
peripheral. The second argument is the duration of the interrupt
0 is interpreted as the default value for the
The function tcdrain waits for all written data to be transmitted.
Depending on the value of the second argument, a call to the
function tcflush discards the data written but not yet
TCIFLUSH), or the data received but not yet read
TCOFLUSH) or both (
Depending on the value of the second argument, a call to the
function tcflow suspends the data transmission
TCOOFF), restarts the transmission (
TCOON), sends a control
character stop or start to request the
transmission to be suspended (
TCIOFF) or restarted (
The function setsid puts the process in a new session and detaches it from the terminal.
Two processes can modify the same file in parallel; however, their
writes may collide and result in inconsistent data. In some cases data
is always written at the end and opening the file with
prevents this. This is fine for
log files but it does not
work for files that store, for example, a database because writes are
performed at arbitrary positions. In that case processes using the
file must collaborate in order not to step on each others toes. A
lock on the whole file can be implemented with an auxiliary file (see
page ??) but the system call lockf allows
for finer synchronization patterns by locking only parts of a file.
We extend the function
file_copy (section 2.9) to
support symbolic links and directories in addition to normal files.
For directories, we recursively copy their contents.
To copy normal files we reuse the function
file_copy we already
set_infos below modifies the owner, the
access rights and the last dates of access/modification
of a file. We use it to preserve this information for copied files.
The system call utime modifies the dates of access and
modification. We use
chown to re-establish
the access rights and the owner. For normal users, there are
a certain number of cases where
chown will fail with a
“permission denied” error. We catch this error and ignore it.
Here’s the main recursive function.
We begin by reading the information of the
source file. If it is
a normal file, we copy its contents with
file_copy and its
set_infos. If it is a symbolic link, we read
where it points to and create a link pointing to the same object. If
it is a directory, we create a destination directory, then we read the
directory’s entries (ignoring the entries about the directory itself
or its parent) and recursively call
copy_rec for each entry. All
other file types are ignored, with a warning.
The main program is straightforward:
Copy hard links cleverly. As written above
copy_rec creates n
duplicates of the same file whenever a file occurs under n different
names in the hierarchy to copy. Try to detect this situation, copy
the file only once and make hard links in the destination hierarchy.
For the files that have already been copied we keep a map from their
(st_dev, st_ino) to their destination file name. Before
each copy we consult the map to see if a file with the same identity
was already copied. If that’s the case we do a hard link on the
destination file name instead of redoing the copy. To minimize the
size of the map we remember only the files which have more than one
name, i.e. those for which
st_nlink > 1.
tar file format (for
archive) can store a file
hierarchy into a single file. It can be seen as a mini file system.
In this section we define functions to read and write
files. We also program a command
readtar such that
displays the name of the files contained in the archive
readtar a f extracts the contents of the file
f contained in
a. Extracting the whole file hierarchy of an archive and
generating an archive for a file hierarchy is left as an exercise.
tar archive is a set of records. Each record represents a
file; it starts with a header which encodes the information
about the file (its name, type, size, owners, etc.) and is followed by
the contents of the file. The header is a block of 512 bytes structured as
shown in table 3.
|108||8||octal||Id of user owner|
|116||8||octal||Id of group owner|
|124||12||octal||File size (in bytes)|
|136||12||octal||Date of last modification|
|265||32||string||Name of user owner|
|297||32||string||Name of group owner|
|329||8||octal||Peripheral major number|
|337||8||octal||Peripheral minor number|
'\000'; except the fields
The file contents is stored right after the header, its size is rounded to a multiple of 512 bytes (the extra space is filled with zeros). Records are stored one after the other. If needed, the file is padded with empty blocks to reach at least 20 blocks.
Since tar archives are also designed to be written on brittle media
and reread many years later, the header contains a
field which allows to detect when the header is damaged. Its value is
the sum of all the bytes of the header (to compute that sum we assume
checksum field itself is made of zeros).
kind header field encodes the file type in a byte as follows4:
Most of the cases correspond to the values of the Unix file type
file_kind stored in the
st_kind field of the
LINK is for hard links which
must lead to another file already stored within the archive.
is for ordinary file, but stored in a contiguous area of memory (this
is a feature of some file systems, we can treat it like an ordinary
link header field stores the link when
LINK. The fields
minor contain the major
and minor numbers of the peripheral when
BLK. These three fields are not used in other cases.
The value of the
kind field is naturally represented by a
variant type and the header by a record:
Reading a header is not very interesting, but it cannot be ignored.
An archive ends either at the end of file where a new record would
start or on a complete, but empty, block. To read a header we thus try
to read a block which must be either empty or complete. For that we
really_read function defined earlier. The end of file
should not be reached when we try to read a block.
To perform an operation in an archive, we need to read the records sequentially until we find the target of the operation. Usually we just need to read the header of each record without its contents but sometimes we also need to get back to a previous one to read its contents. As such we keep, for each record, its header and its location in the archive:
We define a general iterator that reads and accumulates the records
of an archive (without their contents). To remain general, the
f is abstracted. This allows to use the
same iterator function to display records, destroy them, etc.
fold_aux starts from a position
offset with a
accu. It moves to
offset where a record
should start, reads a header, constructs the record
r and starts
again at the end of the record with the new (less partial) result
f r accu. It stops when there’s no header: the end of the archive
We just display the name of records without keeping them:
readtar a f must look for the file
f in the
archive and, if it is a regular file, display its contents. If
is a hard link on
g in the archive, we follow the link and
g since even though
g are represented
differently in the archive they represent the same file. The fact that
f is a link on the other or vice versa depends only on
the order in which the files were traversed when the archive was
created. For now we do not follow symbol links.
Hard link resolution is done by the following mutually recursive functions:
find_regular finds the regular file corresponding to
r is a regular file itself,
r is a hard link the function looks for the regular
file in the archive’s previous records stored in
list with the
find_file. In all other cases, the function aborts.
Once the record is found we just need to display its contents. After
positioning the descriptor at the start of the record’s contents this
operation is very similar to the
We now just need to combine these functions correctly.
We read the records in the archive (but not their contents) until we
find the record with the target name. We then call the function
find_regular to find the record that actually contains the file.
This second, backward, search must succeed if the archive is
well-formed. The first search may however fail if the target name is
not in the archive. In case of failure, the program takes care to
distinguish between these two cases.
Here is the main function which implements the command
Extend the command
readtar so that it follows symbolic links in
the sense that if the link points to a file of the archive that file’s
contents should be extracted.
Behind this apparently trivial requirement are hidden difficulties. Symbolic links are arbitrary paths, they can point on directories (which is not allowed for hard links) and they may not correspond to files contained in the archive.
A simple solution is to recreate, in memory, the file hierarchy contained in the archive.
Nodes of this in-memory file system are described by the
info field describes the file type, limited to ordinary
files, symbolic links and directories. Paths are represented by lists
of strings and directories by lists that associate a node to each file
name in the directory. The
record field stores the
record associated to the node. This field is optional because
intermediate directories are not always present in the archive; it is
mutable because a file may appear more than once in the archive and
the last occurrence takes precedence over the other.
As in Unix, each directory contains a link to itself and to its parent, except for the root directory (in contrast to Unix where it is its own parent). This allows us to detect and forbid any access outside the hierarchy contained in the archive.
find finds in the archive the node corresponding to
path by starting from the initial node
inode. If the search
result is a link, the flag
link indicates whether the link itself
should be returned (true) or the file pointed by the link (false).
mkpath traverses the path
path creating missing
nodes along the path.
explode parses a Unix path into a list of strings.
It removes the end “
/” of directory names which are allowed
add adds the record
the archive. The archive, represented by its root node, is modified by a
We end as before.
Write a command
untar such that
untar a extracts and creates
all the files in the archive
a (except special files)
restoring if possible the information about the files
(owners, permissions) as found in the archive.
The file hierarchy should be reconstructed in the current working
directory of the
untar command. If the archive tries to create
files outside a sub-directory of the current working directory this
should be detected and prohibited. Nonexistent directories not explicitly
mentioned in the archive should be created with the user’s default
One small difficulty is the management of permissions: we must create the archive’s directories with write permission and set them to their actual value only after all the files were extracted.
Let us first write an auxiliary function for
mkpath p m that
creates the missing directories along the path
p with permissions
m (and such that
p may be terminated by a
We also define a function
set_infos similar to the one
used to copy files (section 2.15):
The main function of the program is
processes a single record and accumulates directories explicitly
created by the archive:
The body of the program just iterates
the records and finally updates the directories with the correct access
Write a program
tar such that
tar -xvf a f1 f2 ...
constructs the archive
a containing the list of files
f2, etc. and their sub-directories.
We start with the function that writes a record header in a buffer. It’s a tedious function but it must be done with care as a single error in a header can corrupt the entire archive. In particular we must pay attention to the limits imposed by the file format. For example the size of paths is limited to 99 bytes (There are extensions to the format to handle longer path but it’s not the goal of this project).
The following function creates a record header for a file.
the file name,
infos is the stats information of
the file and
kind is the type of file.
To write a file in the archive, we define a variant of
which takes as an argument the number of bytes to copy and verifies
that the end of file corresponds to that size. Otherwise, an error is
raised: this handles the abnormal case where a file is modified during
the archival process. To limit the archive’s corruption to a single
file we do not write beyond that size.
We now tackle the creation of the archive. The files already written in the archive are stored in a hashtable with their path so that they are not copied more than once. We also store the directories that were already written so as not to copy them again: it can happen that the archival root is already contained in another and we don’t want to copy it again (even though that would be harmless).
The data needed to write an archive is a file descriptor pointing on
the file to write, the file and directory cache (see above) and a size
variable that remembers the current archive size (to pad it to a
minimal size if needed). The
archive type collects all this
information in a record:
Here is the main function that writes an entire hierarchy starting
file path given on the command line. This function is not
difficult but needs some care with pathological cases. In particular
we saw how to detect when a file is modified the archival. A sub case
of this when the archive is being archived itself…
We keep track of regular files that may have hard links in the
regfiles table. It’s not necessary for files that have a single
Here’s the main function. In case of error, it is better to remove the erroneous archive.
We end by parsing the command line arguments.
writesystem calls to make the complete transfer — see the discussion in section 5.7. But this limit is bigger than the size of system caches and it is not observable.
A process is a program executing on the operating system. It consists of a program (machine code) and a state of the program (current control point, variable values, call stack, open file descriptors, etc.).
This section presents the Unix system calls to create new processes and make them run other programs.
The system call fork creates a process.
The new child process is a nearly perfect clone of the
parent process which called
fork. Both processes execute
the same code, are initially at the same control point (the return
fork), attribute the same values to all variables, have
identical call stacks, and hold open the same file descriptors to
the same files. The only thing which distinguishes the two processes
is the return value from
fork: zero in the child process,
and a non-zero integer in the parent. By checking the return value
fork, a program can thus determine if it is in the parent
process or the child and behave accordingly:
The non-zero integer returned by
fork in the parent process
is the process id of the child. The process id is used by
the kernel to uniquely identify each process. A process can obtain
its process id by calling getpid.
The child process is initially in the same state as the parent process
(same variable values, same open file descriptors). This state is not
shared between the parent and the child, but merely duplicated at the
moment of the
fork. For example, if one variable is bound to a
reference before the
fork, a copy of that reference and its
current contents is made at the moment of the
fork; after the
fork, each process independently modifies its “own”
reference without affecting the other process.
Similarly, the open file descriptors are copied at the moment of the
fork: one may be closed and the other kept open. On the other
hand, the two descriptors designate the same entry in the file table
(residing in system memory) and share their current position: if one
reads and then the other, each will read a different part of the file;
likewise, changes in the read/write position by one process with
immediately visible to the other.
leave hhmm exits immediately, but
forks a background process which, at the time
hhmm, reports that
it is time to leave.
The program begins with a rudimentary parsing of the command line,
in order to extract the time provided. It then calculates the delay
in seconds (line 8). The time
call returns the current date, in seconds from the epoch (January 1st
1970, midnight). The function localtime splits
this duration into years, months, days, hours, minutes and seconds.
It then creates a new process using
fork. The parent process
(whose return value from
fork is a non-zero integer) terminates
immediately. The shell which launched
leave thereby returns
control to the user. The child process (whose return value from
fork is zero) continues executing. It does nothing during the
indicated time (the call to
sleep), then displays its message and
The system call
wait waits for one of the child processes created
fork to terminate and returns information about how it did.
It provides a parent-child synchronization mechanism and a very
rudimentary form of communication from the child to the parent.
The primitive system call is waitpid and the function
wait () is merely a shortcut for the expression
waitpid  (-1).
The behavior of
waitpid  p depends on the value of
p> 0, it awaits the termination of the child with id equal to
p= 0, it awaits any child with the same group id as the calling process.
p= −1, it awaits any process.
p<−1, it awaits a child process with group id equal to
The first component of the result is the process id of the child
wait. The second component of the result is a value of type
|The child process terminated normally via
|The child process was killed by a signal
|The child process was halted by the signal
If one of the child processes has already terminated by the time the
wait, the call returns immediately. Otherwise, the
parent process blocks until some child process terminates (a behavior
called “rendezvous”). To wait for n child processes, one must
wait n times.
waitpid accepts two optional flags for its first
argument: the flag
WNOHANG indicates not to wait if there is
a child that responds to the request but has not yet terminated.
In that case, the first result is
0 and the second undefined.
WUNTRACED returns the child processes that have been
halted by the signal
sigstop. The command raises the exception
ECHILD if no child processes match
p (in particular, if
-1 and the current process has no more children).
fork, the child process traverses the upper half of
the table, and exits with the return code 1 if it found an element
satisfying the predicate
cond, or 0 otherwise
(lines 16 and 17). The parent process
traverses the lower half of the table, then calls
sync with the child process (lines 21
and 22). If the child terminated normally, it combines
its return code with the boolean result of the search in the lower
half of the table. Otherwise, something horrible happened, and the
In addition to the synchronization between processes, the
call also ensures recovery of all resources used by the child
processes. When a process terminates, it moves into a “zombie”
state, where most, but not all, of its resources (memory, etc.) have
been freed. It continues to occupy a slot in the process table to
transmit its return value to the parent via the
Once the parent calls
wait, the zombie process is removed from
the process table. Since this table is of fixed size, it is important
wait on each forked process to avoid leaks.
If the parent process terminates before the child, the child is
given the process number 1 (usually
init) as parent. This
process contains an infinite loop of
wait calls, and will
therefore make the child process disappear once it finishes. This
leads to the useful “double fork” technique if you cannot
wait on each process you create (because you cannot
afford to block on termination of the child process,
The child terminates via
exit just after the second
The grandson becomes an orphan, and is adopted by
init. In this
way, it leaves no zombie processes. The parent immediately calls
wait to reap the child. This
wait will not block for long
since the child terminates very quickly.
The system calls execve, execv, and execvp launch a program within the current process. Except in case of error, these calls never return: they halt the progress of the current program and switch to the new program.
The first argument is the name of the file containing the program to
execute. In the case of
execvp, this name is looked for in the
directories of the search path (specified in the environment variable
The second argument is the array of command line arguments with which
to execute the program; this array will be the
of the executed program.
In the case of
execve, the third argument is the environment
given to the executed program;
give the current environment unchanged.
execvp never return a
result: either everything works without errors and the process starts
the requested program or an error occurs (file not found, etc.), and
the call raises the exception
Unix_error in the calling program.
The following three forms are equivalent:
Here is a “wrapper” around the command
adds the option
-i (to ignore case) to the list of arguments:
Here’s a “wrapper” around the command
changes the terminal type:
The process which calls
exec is the same one that executes the
new program. As a result, the new program inherits some features of
the execution environment of the program which called
The following program is a simplified command interpreter: it reads lines from standard input, breaks them into words, launches the corresponding command, and repeats until the end of file on the standard input. We begin with the function which splits a string into a list of words. Please, no comments on this horror.
We now move on to the main loop of the interpreter.
exec_command executes a command and handles errors.
The return code 255 indicates that the command could not be executed.
(This is not a standard convention; we just hope that few commands
terminate with a return code of 255.) The function
print_status decodes and prints the status information returned
by a process, ignoring the return code of 255.
Each time through the loop, we read a line from
stdin with the
input_line. This function raises the
exception when the end of file is reached, causing the loop to
exit. We split the line into words, and then call
child process uses
exec_command to execute the command. The
parent process calls
wait to wait for the command to finish and
prints the status information returned by
Add the ability to execute commands in the background if they are
If the command line ends with
&, we do not call
the parent process and immediately continue with the next iteration
of the loop. But there is one difficulty: the parent may now have
multiple children executing at the same time (the commands in the
background which haven’t terminated yet, plus the last synchronous
wait could synchronize with any of these children.
Thus, for synchronous command,
wait must be repeated until the
recovered child is the one actually executing that command.
Signals, or software interrupts, are external, asynchronous events used to alter the course of a program. These may occur at any time during the execution of a program. Because of this, they differ from other methods of inter-process communication, where two processes must be explicitly directed to wait for external messages; for example, by calling read on a pipe (see chapter 5).
The amount of information transmitted via a signal is minimal, just the type of signal, and although they were not originally intended for communication between processes, they do make it possible to transmit atomic information about the state of an external entity (e.g. the state of the system or another process).
When a process receives a signal, there are four possible outcomes:
There are several types of signals, each associated with a particular event. Table 4 lists some of them with their default behaviors.
|Hang-up (end of connection)||Termination|
|Strong interruption (||Term. & core dump|
|Arithmetic error (division by zero)||Term. & core dump|
|Very strong interruption (cannot be ignored)||Termination|
|Memory protection violation||Term. & core dump|
|Writing to a pipe without readers||Termination|
|Temporary halt (||Suspension|
|Resuming a stopped process||Ignored|
|A child process died or was stopped||Ignored |
The signals received by a process come from several possible sources:
ctrl-C, the console operator sends the
sigintsignal to the processes controlled by her terminal (that were not already put in the background). In the same way, the
sigquitsignal is sent by typing
ctrl-\1. When the terminal is closed (either through voluntary disconnection or owing to a disconnected network link), the
sighupsignal is sent.
kill. This makes it possible to send a specific signal to a specific process. For example,
kill -KILL 194sends the
sigkillsignal to the process with id 194, which causes the process to be killed.
kill(the preceding example being a specific case).
The system call kill makes it possible to send a signal to a process.
The first parameter is the process id of the destination process and
the second the signal number to send. An error occurs if we attempt to
send a signal to a process not owned by the user. A process may send
signals to itself. When the
kill system call returns, it is
guaranteed that the signal was delivered to the destination
process. If a process receives the same signal in rapid succession it
will execute the code associated with the signal only once. Therefore
a program cannot count the number of times it receives a signal,
rather only the number of times it responds to it.
The system call alarm makes it possible to schedule interruptions based on the system clock.
alarm s returns immediately but causes the
signal to be sent to the calling process at least
s seconds later
(note that there is no guarantee on the maximum wait time). The call returns
the number of seconds remaining to an alarm scheduled by a previous call.
0, the effect is simply to cancel an earlier alarm.
The system call signal makes it possible to modify the behavior of a process when it receives a signal of a certain type.
The first argument is the signal number and the second argument, a value of type signal_behavior, indicates the desired behavior for the signal. With:
|The signal is ignored.|
|The default behavior occurs.|
|The function |
Forking a process with the system call fork
preserves signal behavior: the initial definitions for the child are
those of the parent at the time when
fork was executed. The
execve system call sets all the behaviors to
Signal_default except that signals ignored before are still
Occasionally we want to log-off or end a session while allowing
background tasks (large calculations, “spyware” programs, etc.)
to continue to run. If this is desired, processes which normally
exit on receiving
sighup (sent at the time the user disconnects)
should be prevented from doing so. The Unix command
executes the command
cmd arg1 ... argn in a way unaffected by
sighup (certain shells execute
automatically for all processes launched as background tasks). Here’s how
to implement this in three lines:
The system call execvp preserves the fact that
sighup is ignored.
Carefully exiting when a program is misbehaving. For example,
a program like
tar can try to save important information
in a file or destroy the corrupted file before terminating. For this
it is possible to include the following lines at the beginning of the program:
where the function
quit is of the form:
Capturing user-initiated interruptions. Some interactive programs
need to return to a main control loop when a user
ctrl-C. For this we just need to raise an exception when the
sigint signal is received.
To carry out periodic tasks (animations, etc.) interleaved with the execution of the main program. For example, here is how to create “beep” sounds every 30 seconds, regardless of the activity of the main program (calculations or input/output).
Signals are useful for asynchronous communication — indeed, it is their raison d’être — but this asynchronous nature also makes them one of the major difficulties of system programming.
The signal handling function is executed asynchronously and thus pseudo-concurrently with the main program of the process. As signal handling functions cannot return a value, they usually modify global variables. This can result in race conditions between the signal handler and the main program if they try to modify the same variable. As explained in the next section one solution is to temporarily block signals when this variable is accessed by the main program.
In fact, OCaml does not treat signals in a strictly asynchronous fashion. On receiving a signal, OCaml records the receipt of the signal but the signal handling function will only be executed at certain checkpoints. These are frequent enough to provide the illusion of asynchronous execution. The checkpoints typically occur during allocations, loop controls, or interactions with the system (particularly system calls). OCaml guarantees that a program that does not loop, does not allocate, and does not interact with the system will not have its execution interleaved with that of a signal handler. In particular, storing an unallocated value (integer, boolean, etc. — but not a float!) in a reference cell cannot result in the race condition described above.
Signals may be blocked. Blocked signals are not ignored, but put on standby, generally to be delivered later. The sigprocmask system call makes it possible to change the mask for incoming signals:
sigprocmask cmd sigs changes the list of blocked signals and
returns the list of signals that were blocked before the execution of
the function. This makes it possible to later reset the mask to its
previous state. The argument
sigs is a list of signals and
a value of type sigprocmask_command which determines the
effect of the call:
|The signals |
|The signals |
|The signals |
A typical usage of
sigprocmask is to mask certain
Often, one has to guard against possible errors by using the following pattern:
Certain system calls can be interrupted by unignored signals. These
system calls are known as slow calls, which can take an
arbitrary amount of time (for example terminal i/o,
select, system, etc.). If
an interruption occurs, the system call is not completed and raises
EINTR. However file i/o is not interruptible:
although these operations can suspend the running process to execute
another process for disk i/o, when this occurs the interruption will
always be brief if the disk functions correctly. In particular, the
throughput of data depends only on the system, and not another user’s
Ignored signals are never delivered and a masked signal is not
delivered until unmasked. But in all other cases we must protect our
system calls against unwanted interruptions. A typical example is a
parent waiting for the termination of a child. In this case, the
parent executes waitpid
 pid where
pid is the
process id of the child. This is a blocking system call, it is thus
slow and could be interrupted by the arrival of a signal,
especially since the
sigchld signal is sent to the parent when a
child process dies.
Misc define the function
makes it possible to repeat a system call when it is interrupted by a
signal, i.e. when the
EINTR exception is raised.
To wait on a child correctly, call
restart_on_EINTR (waitpid flags) pid.
Children can also be recovered asynchronously in the signal handler of
sigchld, especially when their return value does not matter to
the parent. But when the process receives the
sigchld signal, it
is not possible to know the exact number of terminated processes, since if
the signal is received several times within a short period of time the
handler is invoked only once. This leads to the library function
Misc.free_children function to handle the
non-blocking mode (option
WNOHANG) to recover any dead children
and repeats until either there are only live children (zero is
returned instead of a child id) or there are no children (
Note that it is not important to guard against the
waitpid is non-blocking when called with the
system in the
Unix module is simply defined as:
The specification of the
system function in the C standard
library states that the parent ignores
sigquit signals and masks the
sigchld signal during the
command’s execution. This makes it possible to stop or kill the child
process without affecting the main program’s execution.
We prefer to define the function
system as a specialization of the
more general function
exec_as_system which does not necessarily
go through the shell.
Note that the signal changes must be made before the call
fork is executed because the parent could receive signals
sigchld if the child were to finish immediately) before it
proceeds. These changes are reset for the child on line 11
before executing the command. Indeed, all ignored signals
are preserved by
exec and their behavior is
exec system call uses the
default behavior of signals except if the calling process ignores a
signal in which case it also does.
Finally, the parent must also reset the changes immediately after the
call, even if an error occurs. This is why
try_finalize is used on line 15.
Since the earliest versions of Unix, time has been counted in seconds.
For compatibility reasons, therefore, one can always measure time in seconds.
The current time is defined as the number of seconds since January 1st, 1970
00:00:00 gmt. It is returned by the function:
The sleep system call can pause the execution of a program for the number of seconds specified in its argument:
However, this function is not primitive. It is programmable with
more elementary system calls using the function
sigsuspend l system call temporarily suspends the signals in the
l and then halts the execution of the program until the reception
of a signal which is not ignored or suspended (on return, the
signal mask is reset to its old value).
Now we may program the
Initially, the behavior of the
sigalrm signal does nothing. Note
that “doing nothing” is the same as ignoring the signal. To
ensure that the process will be awakened by the reception
of the signal, the
sigalrm signal is put in an unblocked
state. Then the process is put on standby by suspending all other
signals which were not already suspended (
old_mask). After the alarm
is signaled, the preceding modifications are erased. (Note that
line 9 could be placed immediately after
line 2 because the call to
the signal mask.)
In more modern versions of Unix, time can also be measured in microseconds.
In OCaml, time measured in microseconds is represented by a float.
The gettimeofday function is the equivalent of the
function for modern systems.
In present-day Unix each process is equipped with three timers, each measuring time from a different perspective. The timers are identified by a value of type interval_timer :
|Real time (|
|User time (|
|User time and system time (|
The state of a timer is described by the interval_timer_status
type which is a record with two fields (each a
it_intervalis the period of the timer.
it_valueis the current value of the timer; when it turns
sigvtalrmis sent and the timer is reset to the value in
A timer is therefore inactive when its two fields are
The timers can be queried or modified with the following functions:
The value returned by
setitimer is the old value of
the timer at the time of the modification.
new_timer k f should create a new timer of the
k starting the action
f, and inactive on creation;
set_timer t should set the value of the timer
(and return the old value).
Modern versions of Unix also provide functions to handle dates, see the structure tm which allows dates and times to be expressed according to a calendar (year, month, etc.) and the conversion functions: gmtime, localtime, mktime, etc.
Owing to their asynchronous nature, the use of signals for inter-process communication presents some limitations and difficulties:
Signals offer only a limited form of asynchronous communication but carry all the difficulties and problems associated with it. If possible, it is therefore better not to use them. For example, to wait for a small amount of time, select can be used instead of alarms. But in certain situations signals must be taken into account (for example in command line interpreters).
Signals are possibly the least useful concept in the Unix system. On
certain older versions of Unix (System V, in particular) the behavior
of a signal is automatically reset to
Signal_default when it is
received. The signal handling function can of course register itself
again. For example in the “beep” example on
page ??, it would be necessary to write:
However the problem is that the signals that are received between the
instant were the behavior is automatically reset to
Signal_default and the moment were
set_signal is invoked are
not treated correctly and depending on the type of the signal they may
be ignored or cause the process to die instead of invoking the signal
Other flavors of Unix (bsd or Linux) provide better support: the behavior associated with a signal is not altered when it is received, and during the handling of a signal other signals of the same type are put on hold.
So far, we have learned how to manage processes and how they can communicate with the environment by using files. In the remainder of the course we see how processes running in parallel can cooperate by communicating among themselves.
Regular files are not a satisfactory communication medium for processes
running in parallel. Take for example a reader/writer situation in
which one process writes data and the other reads them. If a file is used
as the communication medium, the reader can detect that the file
does not grow any more (
read returns zero), but it does not know
whether the writer is finished or simply busy computing
more data. Moreover, the file keeps track of all the data transmitted,
requiring needless disk space.
Pipes provide a mechanism suitable for this kind of communication. A pipe is made of two file descriptors. The first one represents the pipe’s output. The second one represents the pipe’s input. Pipes are created by the system call pipe:
The call returns a pair
(fd_in, fd_out) where
fd_in is a
file descriptor open in read mode on the pipe’s output and
fd_out is file descriptor open in write mode on the pipe’s
input. The pipe itself is an internal object of the kernel that can
only be accessed via these two descriptors. In particular, it has no
name in the file system.
A pipe behaves like a queue (first-in, first-out). The first
thing written to the pipe is the first thing read from the pipe.
Writes (calls to write on the pipe’s input
descriptor) fill the pipe and block when the pipe is full. They block
until another process reads enough data at the other end of the pipe
and return when all the data given to
write have been
transmitted. Reads (calls to read on the pipe’s output
descriptor) drain the pipe. If the pipe is empty, a call to
blocks until at least a byte is written at the other end. It then
returns immediately without waiting for the number of bytes requested
read to be available.
Pipes are useless if they are written and read by the same process (such a process will likely block forever on a substantial write or on a read on the empty pipe). Hence they are usually read and written by different processes. Since a pipe has no name, one of these processes must be created by forking the process that created the pipe. Indeed, the two file descriptors of the pipe, like any other file descriptors, are duplicated by the call to fork and thus refer to the same pipe in the parent and the child process.
The following snippet of code is typical.
fork there are two descriptors open on the pipe’s
input, one in the parent and the other in the child. The same
holds for the pipe’s output.
In this example the child becomes the writer and the parent the
reader. Consequently the child closes its descriptor
fd_in on the
pipe’s output (to save descriptors and to avoid programming
errors). This leaves the descriptor
fd_in of the parent unchanged
as descriptors are allocated in process memory and after the fork the
parent’s and child’s memory are disjoint. The pipe, allocated in system
memory, still lives as there’s still the descriptor
fd_in of the
parent open in read mode on the pipe’s output. Following the same
reasoning the parent closes its descriptor on the pipe’s input. The
result is as follows:
Data written by the child on
fd_out is transmitted to
When all the descriptors on a pipe’s input are closed and the pipe is
empty, a call to
read on its output returns zero:
end of file. And when all the descriptors on a pipe’s output are
closed, a call to
write on its input kills the writing
process. More precisely the kernel sends the signal
the process calling
write and the default handler of this signal
terminates the process. If the signal handler of
changed, the call to
write fails with an
This is a classic example of parallel programming. The task of the program is to enumerate the prime numbers and display them interactively as they are found. The idea of the algorithm is as follows. A process enumerates on its output the integers from 2 onwards. We connect this process to a “filter” process that reads an integer p on its input and displays it.
Therefore, the first filter process reads p=2. Then it creates a new filter process connected to its output and filters out the multiples of p it gets on its input; all numbers it reads that are not a multiple of p are rewritten on its output.
Hence the next process reads p=3, which it displays and then starts to filter multiples of 3, and so on.
This algorithm cannot be directly implemented in Unix because it creates too many processes (the number of primes already found plus one). Most Unix systems limit the number of process to a few dozens. Moreover, on a uniprocessor machine, too many processes active simultaneously can bring the system to its knees because of the high costs incurred by switching process contexts. In the following implementation each process first reads n primes p1, …, pn on its input before transforming itself in a filter that eliminate the multiples of p1, …, pn. In practice n = 1000 gives a reasonable slowdown on process creation.
We start with the process that enumerates integers from 2 to k.
To output and input the integers, the following functions are used:
output_binary_int from the standard library writes a
four-byte binary representation of an integer on an
out_channel. The integer can be read back by the function
input_binary_int on an
in_channel. Using these functions
from the standard library has two advantages: first, there is no need to
code the function converting integers to a bytewise
representation1; second, since
these functions use buffered i/o, fewer system calls are
performed, which results in better performance. The following functions
out_channel to buffer the
i/o on the given descriptor:
They allow a program to perform buffered i/o on descriptors acquired
indirectly or that are not the result of opening a file. These
functions are not here to mix buffered i/o with non-buffered
i/o; this is possible but very brittle and highly
discouraged — particularly for input. Note also that it is possible
but very risky to create more than one
in_channel (for example)
on the same descriptor.
We now continue with the filter process. It uses the auxiliary function
read_first_primes. A call to
read_first_primes input count
count prime numbers on
in_channel) and eliminates
multiples of the primes already read. These
count primes are
displayed as soon as they are read and we return them in a list.
And here is the concrete filter function:
The filter starts by calling
read_first_primes to read the first
1000 prime numbers on its input (the
input argument of type
in_channel). Then we create a pipe and clone the process with
fork. The child starts to filter the output of this pipe. The
parent reads numbers on its input and writes each one to the pipe if it
is not a multiple of one of the 1000 primes it initially read.
Finally, the main program just connects the integer generator to the
first filter process with a pipe. Invoking the program
enumerates the primes smaller than
k is omitted (or
not an integer), it defaults to
In this example we do not wait for the child before stopping the parent. The reason is that parent processes are generators for their children.
k is given, the parent will terminate first and close
the descriptor on the input of the pipe connected to its child. Since
OCaml empties the buffers of descriptors open in write mode when a
process stops, the child process will read the last integer provided
by the parent. After that the child also stops etc. Thus, in this
program children become orphaned and are temporarily attached to the
init before they die in turn.
k is not given, all processes continue indefinitely until one or
more are killed. The death of a process results in the death of its child
as described above. It also closes the output of the pipe connected to
its parent. This will in turn kill the parent at the next write on the
pipe (the parent will receive a
sigpipe signal whose default
handler terminates the process).
What needs to be changed so that the parent waits on the termination of its children? Answer.
Of course the parent must wait on its child. However, before that the
input of the pipe on which the child reads must be closed by the
parent, otherwise the child will wait indefinitely for new integers from the
parent. This leads to a deadlock (closing the channel empties the
buffer before closing the corresponding descriptor, therefore no data
is lost). Concretely, the line 10 of the
function needs to be replaced by:
Whenever a prime is found, the function
print_newline (). This performs a system call to empty the standard
output buffer and artificially limits the execution speed of the program.
print_newline () executes
flush Pervasives.stdout. What can happen if
print_newline () is replaced by
print_char '\n'? What needs
to be added to solve the problem?
Since the child process is an exact copy of the parent, i/o
buffers of the standard library are duplicated when
executed. If the buffers are not emptied after each write,
they must be emptied explicitly just before the call to
fork. All is needed is to add
flush Pervasives.stdout after the line 14 of the
On some Unix systems (System V, SunOS, Ultrix, Linux, bsd) pipes with a name in the file system can be created. These named pipes (also known as fifo) allow processes to communicate even if they are not in a parent/child relationship. This contrasts with regular pipes that limit communication between the pipe creator and its descendants.
The system call mkfifo creates a named pipe:
The first argument is the name of the pipe, and the second one represents the requested access permissions.
Named pipes are opened with a call to openfile like any
regular file. Reads and writes on a named pipe have the same semantics
as those on regular ones. Opening a named pipe in read-only mode
(resp. write-only mode) blocks until the pipe is opened by another
process for writing (resp. reading); if this has already happened,
there’s no blocking. Blocking can be avoided altogether by opening the
pipe with the flag
O_NONBLOCK, but in this case reads and writes
on the pipe won’t block either. After the
pipe is opened, the function
clear_nonblock will change this flag to make further
reads or writes on the pipe blocking. Alternatively,
set_nonblock will make reads and writes non-blocking.
So far, we still do not know how to connect the standard input and
output of processes with a pipe as the shell does to execute
cmd1 | cmd2. Indeed, the descriptors we get on the
ends of a pipe with a call to
pipe (or to
openfile on a
named pipe) are new descriptors, distinct from
To address this problem, Unix provides the system call dup2 (read: “duplicate a descriptor to another descriptor”) that gives one file descriptor another one’s meaning. This can be done because there is a level of indirection between a file descriptor (an object of type file_descr) and the object in the kernel called a file table entry that points to the actual file or pipe and maintains its current read/write position.
The effect of
dup2 fd1 fd2 is to update the descriptor
fd2 to refer to
the file table entry pointed to by
fd1. After the call, these two
descriptors refer to same file or pipe, at the same read/write
Standard input redirection.
After the call to
dup2, the descriptor
stdin points to the
foo. Any read on
stdin will read from the file
(so does any read on
fd; but since we won’t use it, we close it
immediately). This setting on
stdin is preserved by
and the program
bar will execute with its standard input
connected to the file
foo. This is the way the shell executes
bar < foo.
Standard output redirection.
After the call to
dup2, the descriptor
stdout points to
foo. Any write on
stdout will write to the file
foo (so does any write on
fd; but since we won’t use it we
close it immediately). This setting on
stdout is preserved by
execvp and the program
bar will execute with its standard output
connected to the file
foo. This is the way the shell executes
bar > foo.
Connecting the output of a program to the input of another.
cmd2 is executed with its standard input connected to
the output of the pipe. In parallel, the program
cmd1 is executed
with its standard output connected to the input of the pipe. Therefore
cmd1 writes on its standard output is read by
on its standard input.
What happens if
cmd1 terminates before
terminates, all its open descriptors are closed. This means that there’s no
open descriptor on the input of the pipe. When
cmd2 has read all
the data waiting in the pipe, the next read returns an end of file;
cmd2 will then do what it is assigned to do when it reaches the
end of its standard input — for example, terminate.
cmd2 terminates before
cmd1, the last descriptor on
the output of the pipe is closed and
cmd1 will get
a signal (which by default kills the process) the next time
it tries to write on its standard output.
>>, the answer is similar to the
>redirection, except that the file is opened with the flags
[O_WRONLY; O_APPEND; O_CREAT].
2>, the answer is similar to the
>redirection, except that
dup2 fd stderris executed instead of
dup2 fd stdout
2>1, we must call
dup2 stderr stdoutbefore executing the command.
<<, the shell
shmust create a temporary file in
/tmpcontaining the lines that follow
<<and execute the command with its standard input redirected from this file. Another solution is to connect the command’s standard input to the output of a pipe and let a child process write the lines following
<<on the input of that pipe.
Swapping two descriptors requires care. The naive sequence
dup2 fd1 fd2;
dup2 fd2 fd1 does not work. Indeed, the second
redirection has no effect since after the first one both descriptors
fd2 already point to the same file table entry. The
initial value pointed by
fd2 was lost. This is like swapping the
contents of two reference cells: a temporary variable is needed to
save one of the two values. Here we can save one of the
descriptors by copying it with the system call dup.
The call to
dup fd returns a new descriptor pointing on the same
file table entry as
fd. For example we can swap
After the swap, do not forget to close the temporary descriptor
tmp to prevent a descriptor leak.
We program a command
compose such that
behaves like the shell command:
The bulk of the work is done by the
for loop starting at
line 6. For each command except the last one, we
create a new pipe and a child process. The child connects the pipe’s
input to its standard output and executes the command. After the
fork it inherits the standard input of its parent. The main
process (the parent) connects the pipe’s output to its standard input
and continues the loop. Suppose (induction hypothesis) that at the
beginning of the ith iteration, the situation is as follows:
Rounded boxes represent processes. Their standard input is on the
left, their standard output on the right. The ellipses represent
the initial standard input and output of the
Just after the call to
fork we have:
When the parent calls
dup2, we get:
When the child calls
execv, we get:
and everything is ready for the next iteration.
The last command is forked after the loop because there’s no need to
create a new pipe: the process
compose already has the right
standard input (the output of the next to last command) and output
(the one initially given to the command
compose) for the
child. Hence it is sufficient to
exec. The parent then
waits for its children to terminate: it calls
until the error
ECHILD (no child to wait for) is raised. The
children’s return codes are combined with a bitwise “or”
lor operator) to create a meaningful return code for
compose : zero if all the children returned zero, different from
Note that we execute commands through the shell
prevents us from having to parse complex commands into tokens as
in the following invocation:
Adding this functionality to our program would complicate it needlessly.
In all the examples so far, processes communicate linearly: each process reads data coming from at most one other process. In this section we highlight and solve the problems occurring whenever a process needs to read data coming from many processes.
Consider the example of a multi-windowed terminal emulator. Suppose we have a computer, called the client, connected to a Unix machine by a serial port. We want to emulate, on the client, many terminal windows connected to different user processes on the Unix machine. For example, one window can be connected to a shell and another to a text editor. Outputs from the shell are displayed in the first window and those from the editor in the other. If the first window is active, keystrokes from the client’s keyboard are sent to the input of the shell and if the second window is active they are sent to the input of the editor.
Since there’s only a single physical link between the client and the Unix machine, we need to multiplex the virtual connections between windows and processes by interleaving the data transmissions. Here’s the protocol we are going to use. On the serial port, we send messages with the following structure:
On the Unix machine, user processes (shell, editor, etc.) are
connected by a pipe to one or more auxiliary processes that read and
write on the serial port and (de)multiplex the data. The serial port
is a special file (
/dev/ttya, for example), on which the
write to communicate with the
Demultiplexing (transmission from the client to the user processes) does not pose any particular problem. We just need a process that reads messages on the serial port and writes the extracted data on the pipe connected to the standard input of the receiving user process.
Multiplexing (transmission from user processes to the client) is more tricky. Let us try to mimic the demultiplexer: a process reads sequentially the output of the pipes connected to the standard output of the user processes and then writes the data it reads as message on the serial port by adding the receiving window number and the length of the data.
This does not work, because reading a pipe can block. For example, if we try to read the output of the shell but it has nothing to display at that moment, the multiplexer process will block, and waiting characters from the editor will be ignored. There’s no way to know in advance on which pipes there is data waiting to be displayed (in parallel algorithms, the situation where a process is perpetually denied access to a shared resource is called starvation).
Here is another approach: we associate with each user process a
repeater process. The repeater reads the output of the pipe
connected to the standard output of the user process, transforms the
data into messages and writes the result directly on the serial port
(each repeater process opens
/dev/ttya in write mode).
Since each user process has its output transmitted independently,
blocking problems are solved. However the protocol may not be
respected. Two repeaters may try to write a message at the same time
and the Unix kernel does not guarantee the atomicity of writes, i.e.
that they are performed in a single uninterruptible operation.
Thus the kernel may choose to write only a part of a message from a
/dev/ttya, then write a full message from another
repeater and finally write the remaining part of the first message.
This will utterly confuse the demultiplexer on the client:
it will interpret the second message as part of the data of the
first and then interpret the rest of the data as a new message header.
To avoid this, repeater processes must synchronize so that at anytime
at most one of them is writing on the serial port (in parallel
algorithms we say that we need to enforce the mutual exclusion of
repeaters on the access to the serial link). Technically, this can be
done with concepts we have already seen so far: repeaters can create a
specific file (the “lock”) with the
O_EXCL flag before
sending a message and destroy it after they are done writing to the
serial port. However this technique is not very efficient because the
lock creation and destruction costs are too high.
A better solution is to take the first approach (a single
multiplexer process) and set the output of the pipes connected to the
standard output of user processes in non-blocking mode with
set_nonblock. A read on an empty pipe will not block but return
immediately by raising the error
just ignore this error and try to read the output of the next user
process. This will prevent starvation and avoid any mutual exclusion
problem. However it is a very inefficient solution, the multiplexer
process performs what is called “busy waiting”: it uses
processing time even if no process is
sending data. This can be alleviated by introducing calls to
sleep in the reading loop; unfortunately, it is very difficult to find
the right frequency. Short
sleeps cause needless processor load when there
is little data, and long
sleeps introduce perceptible delays when there is a lot of data.
This is a serious problem. To solve it, the designers of bsd
Unix introduced a new system call,
select, which is now
available on most Unix variants. A call to
select allows a
process to wait (passively) on one or more input/output events.
An event can be:
The system call select has the following signature:
The first three arguments are sets of descriptors represented by
lists: the first argument is the set of descriptors to watch for read
events; the second argument is the set of descriptors to watch for
write events; the third argument is the set of descriptors to watch
for exceptional events. The fourth argument is a timeout in
seconds. If it is positive or zero, the call to
select will return
after that time, even if no event occurred. If it is negative, the call
select waits indefinitely until one of the requested events occurs.
select call returns a triplet of descriptor lists: the first
component is the list of descriptors ready for reading, the second
component those ready for writing and the third one those on which an
exceptional condition occurred. If the timeout expires before any
event occurs, the three lists are empty.
The code below watches read events on the descriptors
fd2 and returns after 0.5 seconds.
multiplex function is central to the
multiplexer/demultiplexer of the multi-windowed terminal emulator
To simplify, the multiplexer just sets the receiver of messages according to their provenance and the demultiplexer redirects data directly to the receiver number. In other words, we assume that either each sender talks to a receiver with the same number, or that the correspondence between them is magically established in the middle of the serial link by rewriting the receiver number.
multiplex function takes a descriptor open on the serial port
and two arrays of descriptors of the same size, one containing pipes
connected to the standard input of the user processes, the other
containing pipes connected to their standard output.
multiplex function starts by constructing a set of
input_fds) that contain the input descriptors
(those connected to the standard output of the user processes) and the
descriptor of the serial port. On each iteration of the
while loop we call
select to watch for pending reads in
input_fds. We do not watch for any write or exceptional event and
we do not limit the waiting time. When
select returns, we test whether
there is data waiting on an input descriptor or on the serial port.
If there is data on an input descriptor we
read this input into a
buffer, add a message header and write the result on the serial
read returns zero this indicates that the corresponding
pipe was closed. The terminal emulator on the client will receive a
message with zero bytes, signaling that the user process
with that number died; it can then close the corresponding window.
If there is data on the serial port, we read the two-byte message
header which gives us the number
i of the receiver and the number
n of bytes to read. We then read
n bytes on the channel and
write them on the output
i connected to the standard input of the
corresponding user process. However, if
n is 0, we close the
output i. The idea is that the terminal emulator at the other end
sends a message with
n = 0 to indicate an end of file on the
standard input of the receiving user process.
We get out of the loop when
really_read raises the exception
End_of_file, which indicates an end of file on the
write of the
Unix module iterates the system call
write until all the requested bytes are effectively written.
However, when the descriptor is a pipe (or a socket, see
chapter 6), writes may block and the system call
write may be interrupted by a signal. In this case the OCaml
Unix.write is interrupted and the error
raised. The problem is that some of the data may already have been
written by a previous system call to
write but the actual size
that was transferred is unknown and lost. This renders the function
write of the
Unix module useless in the presence of signals.
To address this problem, the
Unix module also provides the
“raw” system call
write under the name
single_write, if an error is raised it is guaranteed that no
data is written.
The rest of this section shows how to implement this
function. Fundamentally, it is just a matter of interfacing OCaml with
C (more information about this topic can be found in the relevant
section of the OCaml manual). The following code is written in the file
The first two lines include standard C headers. The following four
lines include C headers specific to OCaml installed by the
unixsupport.h header defines reusable C
functions of the OCaml Unix library.
The most important line is the call to
write. Since the call may
block (if the descriptor is a pipe or a socket) we need to release the
global lock on the OCaml runtime immediately before the call
(line 20) and reacquire it right after
(line 22). This makes the function compatible with the
Thread module (see chapter 7): it allows
other threads to execute during the blocking call.
During the system call OCaml may perform a garbage collection and
the address of the OCaml string
buf may move in memory. To
solve this problem we copy
buf into the C string
This has an additional cost, but only in the order of magnitude of
10% (and not 50% as one might think) because the overall cost of the
function is dominated by the system call. The size of this C string is
unix_support.h. If an error occurs during the system
call (indicated by a negative return value) it is propagated to
OCaml by the function
uerror, defined in the OCaml Unix library.
To access this code from OCaml, the file
But in practice we verify the arguments before calling the function:
This function has been available in the
Unix module since version
3.08. But if we had written the program above ourselves we would
need to compile it as follows to use it (assuming the OCaml code is
in the files
It is often more practical to build a library
both the C and the OCaml code:
write.cma can then be used like
The semantics of
single_write is as close as possible to the
write. The only remaining difference is when the
original string is very long (greater than
call may then not write all the data and must be iterated. The
write (guaranteed for regular files) is thus not
guaranteed for long writes. This difference is generally insignificant but one should
be aware of it.
On top of this function we can implement a higher-level function
really_write, analogous to the function
really_read of the
multiplexer example, that writes exactly the requested amount of data
(but not atomically).
Communication with pipes has some limitations. First, it is local to a machine: with named pipes, communicating processes must run on the same machine and what’s more, with anonymous pipes, they must share a common ancestor. Furthermore, pipes are not very suitable for a particularly useful model of communication: the client-server model. In this model, only one program, the server, has direct access to a shared resource. The other programs, the clients, access the resource by connecting to the server. The server serializes and controls the access to the shared resource. (Example: the x-window windowing system — the shared resources are the screen, the keyboard and the mouse.)
The client-server model is difficult to implement with pipes. The major difficulty is to establish the connection between a client and the server. With anonymous pipes, it is impossible: the server and the client would need a common ancestor that allocated an arbitrarily large number of pipes in advance. With named pipes, the server could read connection requests on a particular pipe. These requests would contain the name of another named pipe created and used by the client to communicate with the server. The problem is to ensure the mutual exclusion of simultaneous connection requests performed by multiple clients.
Sockets are a generalization of pipes addressing these issues. The client-server model is shown in figure 3.
In the model described above, the server U and the client A establish a private connection (3) to communicate without interference from other clients. For that reason, this mode of communication is referred to as the connection-oriented mode. If the transaction is short the server can handle the request directly (without forking) through the connection (3). In this case, the next client must wait for the server to be available, either because it is handling the connection (3), or because it explicitly manages several connections via multiplexing.
Sockets also allow a connectionless communication mode. In this mode, less frequently used, the server does not establish a private connection with the client, but responds directly to the client’s requests. We will briefly comment on this model in section 6.10 but in the remainder of this chapter, we mainly describe connection-oriented communication.
Sockets, an extension of pipes, were introduced in bsd 4.2. They are now found on all Unix machines connected to a network. Special system calls are provided to establish connections following the client-server model; they enable local and remote communication between processes in a (nearly) transparent way.
The communication domain of a socket limits the processes (and the format of their address) with which we can communicate on the socket. Different communication domains are available, for example:
18.104.22.168, for example) and a port number on that machine. Communication is possible between processes running on any two machines connected to Internet.1
The communication type of a socket indicates whether communication is reliable (no loss or duplication of data) and the way the data is sent and received (a stream of bytes, or a sequence of packets — small blocks of bytes). The communication type constrains the protocol used to transmit data. Different communication types are available, here are three of them with their properties:
The “stream” type is very similar to communication with pipes.
It is used most frequently, in particular to transmit unstructured
byte sequences (e.g.
rsh). The “segmented packets” type
transmits data as packets: each write delimits a packet, each read
returns at most a packet. It is well suited for message-oriented
communication. The “datagram” type is the closest to the
hardware properties of an Ethernet network: data is transmitted with
packets and there is no guarantee that they reach their
destination. It is the most economical type in terms of network
resources. Some programs use it to transmit data that is not of
crucial importance (e.g.
biff); others, to get more network
performance but with the burden of managing data losses manually.
The socket system creates a new socket:
The result is a file descriptor that represents the new
socket. Initially, this descriptor is “disconnected”,
it is not ready to accept any
The first argument is a value of type socket_domain, it specifies the socket’s communication domain:
|The Unix domain.|
|The Internet domain.|
The second argument, a value of type socket_type, specifies the desired communication type:
|Byte streams, reliable.|
|Direct access to the lower layers of the network.|
The third argument is the communication protocol to use. It is usually
0 which selects the default protocol for the given communication
domain and type (e.g. udp for
SOCK_STREAM). Other values allow to use special
protocols, for example icmp (Internet Control Message
Protocol) used by the
ping command to send packets which
return automatically to the sender. The numbers for these special protocols
are in the
/etc/protocols file or in the
protocols table of
the nis (Network Information Service) database, if
any. The system call getprotobyname returns information about a
protocol in a portable manner:
Given the name of a protocol the result is a record
of type protocol_entry. The
p_proto field of
this record has the protocol number.
Several socket operations use socket addresses, represented by the variant
ADDR_UNIX f is an address in the Unix domain,
the name of the corresponding file in the machine’s file
ADDR_INET (a,p) is an address in the Internet domain,
a is the Internet address of a machine and
p a port
number on this machine.
Internet addresses are represented by the abstract type
inet_addr. The following functions convert strings of the form
22.214.171.124 to values of type
inet_addr, and vice versa:
Another way to obtain internet addresses is to look them up by host name in
/etc/hosts table, the nis database or in domain name
servers. The system call gethostbyname does that. On
modern machines, the domain name servers are consulted first and
/etc/hosts is only used as a fallback but in general that may
depend on the machine’s configuration.
The argument is the host name to look for and the result a record of
type host_entry. The
h_addr_list field of this
record is an array of Internet addresses corresponding to the machine
(the same machine can be connected to multiple networks under
Regarding port numbers, the most common services are listed in the
/etc/services which can be read in a portable manner using
the getservbyname function:
The first argument is the service name (
"ftp" for ftp
"smtp" for email,
"nntp" for news servers,
"ntalk" for commands of that name, etc.) and the
second argument is the name of the protocol: usually
"tcp" if the
service is using the stream connection type or
"udp" for the
datagram type. The result of
getservbyname is a record of type
s_port field contains the
To obtain the address of the ftp server
The system call connect establishes a connection with a server on a socket.
The first argument is a socket descriptor and the second argument is the server’s address.
Once the connection is established, calls to write on the
socket descriptor send data to the server and calls to read
receive data from the server. Sockets behave like pipes for read and
write operations. First,
read blocks if no data is available and
can return fewer bytes than requested. Second, whenever the server
closes the connection
write sends a
sigpipe signal to the calling process.
connect binds the socket to a local address chosen by the
system. Sometimes, it is preferable to manually choose this address.
This can be done by calling the function
section 6.7) before
netstat Unix command lists the current connections on the
machine and their status.
There are two ways to disconnect a socket. The first is to call close on the socket. This closes the read and write sides of the connection, and deallocates the socket. But sometimes this is too brutal, for example we may want to close the connection from the client to the server to indicate an end of file but keep the connection open in the other direction to get remaining data from the server. The system call shutdown allows to close the connection gradually.
The first argument is the descriptor of the socket to close and the second a value of type shutdown_command indicating which direction to close:
|Closes the socket for reading; |
|Closes the socket for writing; |
|Closes the socket for reading and writing;
Note that disconnecting a socket can take some time whether done with
We program a
client command such that
client host port
establishes a connection on the port
port of the machine named
host, sends on the resulting socket the data it reads on its
standard input and writes the data it receives on its standard output.
For instance, the command
connects to the port
pauillac.inria.fr and sends an
http request for the web page
This command is a “universal” client application in the sense
that it factors out the code to establish a connection common to many
clients and delegates the implementation of the specific protocol to
the program that calls
The library function
Misc.retransmit fdin fdout reads data on the
fdin and writes it on
fdout. It terminates,
without closing the descriptors, when the end of file is reached on
the input descriptor. Note that
retransmit may be interrupted by
The serious matter starts here.
We start by determining the Internet address of the machine to which
we want to connect. It can be specified by a host name or in numerical
gethostbyname correctly handles both cases. Then, we create
a socket of type stream in the Internet domain with the default protocol
and connect it to the address of the machine.
The process is then cloned with
fork. The child process copies
the data from its standard input to the socket. Once the end of
standard input is reached it closes the connection in the sending
direction and terminates. The parent process copies the data it reads
on the socket to its standard output. Once the end of file is reached
on the socket, it closes the standard output, synchronizes with the
child process and terminates.
The connection is closed either by the client or by the server:
sigpipesignal the next time it tries to write on the socket. This does not however report that the connection was lost. If that is needed we can ignore the
sigpipesignal by inserting the following line after line 19:
and the write will raise an
EPIPE error instead.
Having seen how a client connects to a server, we now show how a server can provide a service for clients. First we need to associate a particular address to a socket to make it reachable from the network. The system call bind does this:
The first argument is the socket descriptor and the second the address to bind. The constant Internet address inet_addr_any can be used to bind all the Internet addresses that the machine has (it may be on multiple sub-networks).
We then indicate that the socket can accept connections with the system call listen:
The first argument is the socket descriptor and the second is the number of request that can be be put on hold while the server is busy (ranges from a few dozen to several hundreds for large servers). When the number of waiting clients exceeds this number, additional client connection requests fail.
Finally, connection requests on a socket descriptor are received via the system call accept:
When the call returns the socket given in argument is still free and
can accept more connection request. The first component of the result
is a new descriptor connected to the client, everything written
(resp. read) on that socket can be read (resp. is written) on the
socket the client gave to
connect. The second component of the
result is the address of the client. It can be used to check that the
client is authorized to connect (for example this is what the
x server does,
xhost can be used to add new
authorizations), or to establish a second connection from the server
to the client (as ftp does for each file transfer request).
The general structure of a tcp server is as follows.
The library function
Misc.install_tcp_server addr creates a
socket of type stream in the Internet domain with the default protocol
and prepares it to accept new connection requests on the address
listen. Given that this is a library
function, we close the socket in case of an error.
The library function
Misc.tcp_server creates a socket with
install_tcp_server and enters an infinite loop. At each iteration
of the loop it waits for a connection request with
treats it with the function
treat_connection. Since this is a
library function we restart the
accept call if it is
interrupted. We also ignore the signal
sigpipe so that unexpected
disconnection raise an
EPIPE exception that can be caught by
treat_connection rather than killing the server. Note that it in
any case it is
treat_connection’s duty to close the
descriptor at the end of the connection.
treat_connection is also given the descriptor of the
server so that if it
forks or or
double_forks it can be
closed by the child.
Now suppose we have the following, application specific,
function that talks to the client end ends by closing the connection:
The server itself can treat each connection sequentially. The
following library function, in
Misc, captures this pattern:
However as the server cannot handle any other requests while serving a client,
this scheme is only appropriate for quick services, where the
service function always runs in a short, bounded, amount of time (for
instance, a date server).
Most servers delegate the service to a child process:
fork is called
accept returns. The child process handles the
connection and the parent process immediately retries to
obtain the following library function in
Note that it is essential that the parent closes
otherwise the close made by the child will not terminate the
connection (besides the parent would also quickly run out of
descriptors). The descriptor is also closed if the fork fails, for the
server may eventually decide the error is not fatal and continue to
Similarly, the child immediately closes the
server descriptor on
which the connection request was received. First, it does not need
it. Second, the server may stop accepting new connections before
the child has terminated. The call to
exit 0 is important since
it ensures that the child terminates after the execution of the
service and that it does not start to execute the server loop.
So far we ignored the fact that children will become zombie processes
and that we need to recover them. There are two ways to to so. The
simple approach is to have a grandchild process handle the connection
using a double fork (see page ??). This gives the
following library function, also in
However with this approach the server loses all control on the
grandchild process. It is better to have the processes handling
services and the server in the same process group so that the whole
group can be killed at once to terminate the service. For this reason servers
usually keep the fork treatment but add children recovering code, for
example in the handler of the
sigchld signal (see the function
Misc.free_children on page ??).
Sockets have numerous internal parameters that can be tuned: the size of the transfer buffer, the size of the minimum transfer, the behavior on closing, etc.
These parameters have different types, for this reason there are as many getsockopt and setsockopt OCaml functions as there are types. Consult the OCaml documentation of the function getsockopt and its variants to get a detailed list of those options and the posix reference for getsockopt and setsockopt for their exact meaning.
The following two parameters apply only to sockets of type stream in the Internet domain.
In the tcp protocol, the disconnection of a socket is
negotiated and hence takes some time. Normally a call to
close returns immediately, and lets the system
negotiates the disconnection. The code below turns
close on the
sock into a blocking operation. It blocks either
until all the sent data has been transmitted or until 5 seconds have
SO_REUSEADDR option allows the bind system call
to allocate a new socket on a local address immediately after the
sock bound on that address is closed (there is however the
risk to get packets intended for the old connection). This option
allows to stop a server and restart it immediately, very useful for
We program a
server command such that:
receives connection requests on the port
port and, for each
cmd with the arguments
arg1 ... argn
and the socket connection as its standard input and output. For
example, if we execute:
pomerol machine and the universal client (see
section 6.6) on an other machine as follows:
the client displays the same result as if we had typed:
grep is executed on
pomerol, and not on the local machine.
This command is a “universal” server in the sense that it factors
out the code common to many server and delegates the implementation of
the specific service and communication protocol to the
cmd program it
The address given to
tcp_server contains the Internet address
of the machine running the program; the usual way to get it
(line 11) is by calling
gethostname. But in general many addresses are
referencing the same machine. For instance, the address of the
pauillac machine is
126.96.36.199, it can also be accessed
locally (provided we are already on the
pauillac machine) with
127.0.0.1. To provide a service on all the addresses
pointing to this machine, we can use the constant Internet address
The service is handled by a “double fork”. The
function, executed by the child, redirects standard input and the two
standard output on the connection socket and executes the requested
command (note that the handling of the service cannot be done
The connection is closed without any intervention of the
program. One of the following cases occurs:
exitwhen it is done. This closes the standard outputs which are the last descriptors open for writing on the connection and the client receives an end of file on its socket.
sigpipesignal the next time it tries to write data on the connection. This may kill the process but is perfectly acceptable since nothing is now reading the output of this command.
sigpipesignal (or an
EPIPEexception) when it tries to write on the connection.
Writing a server requires more care than writing a client. While the client usually knows the server to which it connects, the server knows nothing about its clients and particularly if the service is public, the client can be “hostile”. The server must therefore guard itself against all pathological cases.
A typical attack is to open connections and leave them open without transmitting requests. After accepting the connection the server is blocked on the socket as long as the client stays connected. An attacker can saturate the service by opening a lot of unused connections. The server must be robust against these attacks: it must only accept a limited number of simultaneous connections to avoid system resources exhaustion and it must terminate connections that remain inactive for too long.
A sequential server handling connections without forking is immediately exposed to this blocking issue. It will be unresponsive for further request even though it does nothing. A solution for a sequential server is to multiplex the connections, but it can be tricky to implement. The solution with a parallel server is more elegant, but it still needs a timeout, for example by programming an alarm (see section 4.2).
The tcp protocol used by most connections of type
SOCK_STREAM works only in connection-oriented mode. Conversely,
the udp protocol used by most connections of type
SOCK_DGRAM always works in connectionless mode, there is
no established connection between the two machines. For this type of
sockets, data is transmitted with the system calls recvfrom
Their interface is similar to
return the size of the transferred data. The
recvfrom also returns the address of the sending machine.
We can call connect on a socket of type
to obtain a pseudo-connection. This pseudo-connection is just an
illusion, the only effect is that the address passed in argument is
memorized by the socket and becomes the address used for sending and
receiving data (messages coming from other addresses are ignored). It
is possible to call
connect more than once to change the address
or disconnect the pseudo-connection by connecting to an invalid
0. In contrast, doing this with a socket of type stream
would generally issue an error.
The system calls recv and send respectively generalize read and write but they work only on socket descriptors.
Their interface is similar to
write but they add
a list of flags of type msg_flag whose semantics is:
|Process out-of-band data.|
|Short-circuit the default routing table.|
|Examines the data without reading it.|
These primitives can be used in connection-oriented mode instead of
write or in pseudo-connected mode instead of
Examples like the universal client-server are so frequent that the
Unix module provides higher-level functions to establish
or use network services.
open_connection function opens a connection to the given address
and creates a pair of
Pervasives input/output channels on the resulting
socket. Reads and writes on these channels communicate with the server
but since the output channel is buffered we must flush it to ensure
that a request has been really sent. The client can shutdown the
connection abruptly by closing either of the channels (this will close
the socket) or more “cleanly” by calling
shutdown_connection. If the server closes the connection, the
client receives an end of file on the input channel.
A service can be established with the
establish_server f addr establishes a service on the
addr and handles requests with the function
connection to the server creates a new socket and forks. The child
creates a pair of
Pervasives input/output channels on the socket
to communicate with the client and gives them to
f to provide the
f returns the child closes the socket and exits. If
the client closes the connection cleanly, the child gets and end of
file on the input channel and if it doesn’t it may receive a
sigpipe signal when
f writes on the output channel. As for the
parent, it has probably already handled another request! The
establish_server function never terminates, except in
case of error (e.g. of the OCaml runtime or the system during
the establishment of the service).
In simple cases (
rlogin, …), the data transmitted
between a client and a server is naturally represented by two streams
of bytes, one from the client to the server and the other in the
reverse direction. In other cases, the data to transmit is more
complex, and requires to be encoded and decoded to/from the streams of
bytes. The client and the server must then agree on a precise
transmission protocol, which specifies the format of requests and
responses exchanged on the connection. Most protocols used by Unix
commands are specified in documents called “rfc”
(request for comments): these documents start as proposals open
for discussion, and gradually become standards over time, as users
adopt the described protocol.2
Most binary protocols transmit data in a compact format, as close as possible to the in-memory representation, in order to minimize the encoding/decoding work needed for transmission and save network bandwidth. Typical examples of protocols of this type are the x-window protocol, which governs exchanges between the x server and x applications, and the nfs protocol (rfc 1094).
Binary protocols usually encode data as follows. An integer or floating point number is represented by its 1, 2, 4, or 8 bytes binary representation. A string by its length as an integer followed by its contents as bytes. A structured object (tuple, record) by the representation of its fields in order. A variable size structure (array, list) by its length as an integer followed by the representation of its elements. If the exact type of data being transmitted in known to a process it can easily recreate it in its memory. When different type of data is exchanged on a socket the data encoding can be preceded by an integer to identify the data that follows.
XFillPolygon call of the x library, which draws and
fills a polygon, sends a message with the following structure to the
With binary protocols we must pay attention to the computer architecture of the communicating machines. In particular for multi-byte integers, big-endian machines store the most significant byte first (that is, in memory, at the lower-address) and little-endian machines store the least significant byte first. For instance, the 16 bit integer 12345 = 48 × 256 + 57 is represented by the byte 48 at the address n and the byte 57 at the address n+1 on a big-endian machine, and by the byte 57 at the address n and the byte 48 at the address n+1 on a little-endian machine. Hence protocols must precisely specify which convention they use when multi-bytes integers are transmitted. Another option is to allow both and have it specified in the header of the transmitted message.
The OCaml system helps to encode and decode data structures (a procedure called marshalling, serialization or pickling in the literature) by providing two functions to convert an OCaml value into a sequence of bytes and vice versa:
These function are defined to save values to a disk file and get them back but they can also be used to transmit any value on a pipe or a socket. They handle any OCaml values except functions, preserve sharing and circularities inside values and work correctly between machines of different endianness. More information can be found in the Marshal module.
Note that semantically, the type of
input_value is incorrect. It
is too general, it is not true that the result of
'a for any type
'a. The value returned by
input_value belongs to a precise type, and not to all possible
types. But this type cannot be determined at compile time, it depends
on the content of the channel read at runtime. Type-checking
input_value correctly requires an extension to the ML language
known as dynamic objects: values are paired with a representation of
their type allowing to perform runtime type checks. Consult
 for a detailed presentation.
If the x-window protocol was written in OCaml, we would
define a variant type
request for requests sent to the
server and a
reply type for server responses:
The core of the server would be a loop that reads and decodes a request and responds by writing a reply:
The functions of the x library, linked with each application would have the following structure:
Another typical incarnation of binary protocols is remote procedure
calls (rpc). A user on machine A wants to call a function
f on a machine B. This is obviously not directly possible. It can
be programmed on a case by case basis using the system to open
a connection to the machine B, execute the call and send the
result back to the machine A.
But since this is a common situation, an rpc service can
handle that (see figure 4). An rpc server runs on
both machine A and B. A user on machine A requests the rpc
server on the machine to execute a function on the distant machine
B. The server on A relays the request to the rpc server on
machine B which executes the call to
f, sends the result back to
the server on A which gives the result to the user. The point is that
another user can call another function on B by going through the same
server on A. The connection work is shared by the rpc service
installed on the machines A and B and from the perspective of the
users, everything happens as if these calls were simple function calls
Network services where the efficiency of the protocol is not crucial are often “text” protocols. A “text” protocol is in fact a small command language. Requests are command lines, the first word identifies the request type and the possible remaining words the command’s arguments. Responses are also made of one or more lines of text, often starting with a numerical code to identify the kind of response. Here are some “text” protocols:
|smtp (Simple Mail Transfer Protocol)||rfc 821||Electronic mail|
|ftp (File Transfer Protocol)||rfc 959||File transfer|
|nttp (Network News Transfer Protocol)||rfc 977||News reading|
|http/1.0 (HyperText Transfer Protocol)||rfc 1945||Web navigation|
|http/1.1 (HyperText Transfer Protocol)||rfc 2068||Web navigation|
The great advantage of these protocols is that the exchanges between
the server and the client are human readable. For example we can just
telnet command to talk directly to the server. Invoke
telnet host service where
host is the host name on which the
server is running the service
nntp, etc.) and then type in the requests as a client would, the
server’s responses will be printed on standard output. This makes it easier to
understand the protocol. However coding and decoding requests and
responses is more involved than for binary protocols and the message
size also tends to be larger which is less efficient.
Here is an example of an interactive dialog, in the shell, to send an email on an smtp server. The lines preceded by >> go from the client to the server, and are typed in by the user. The lines preceded by << go from the server to the client.
RCPT respectively send to
the server: the name of the client machine, the address of the sender
and the address of the recipient. The
DATA command asks to send
the body of the email. The body of the message is then entered and
ended by a line containing the single character
'.' (would the body
of the email contain such a line, we just double the initial
that line, this additional period is then suppressed by the server).
The responses from the server are all made of a 3 digit numerical code
followed by a comment. Responses of the form
5xx indicate an
error and those with
2xx, that everything is fine. When the
client is a real program it only interprets the response code, the comment is
only to help the person who develops the mail system.
The http protocol (HyperText Transfer Protocol) is primarily used to read documents over the famous “world wide web”. This domain is a niche area of client-server examples: between the client that reads a page and the server that writes it there is a myriad of intermediary relays that act as virtual servers for the real client or delegated clients for the real server. These relay often provide additional service like caching, filtering, etc..
There are several versions of the http protocol. To allow us to focus on the essentials, namely the architecture of clients or relays, we use the simple protocol inherited from the very first versions of the protocol. Even if dust-covered it is still understood by most servers. At the end of the section we describe a more modern, but also more complex, version which is needed to make real tools to explore the web. We do however leave the translation of the examples to this new version as an exercise.
Version 1.0 of the http protocol specified in rfc 1945 defines simple requests of the form:
sp represents a space and
"\r\n" (“return” followed by
“linefeed”). The response to a simple request is also
simple: the content of the url is sent directly, without any
headers and the end of the request is signaled by the end of file,
which closes the connection. This form of request, inherited from
version 0.9 of the protocol, limits the connection to a single
We write a
geturl program that takes a single argument, a url,
retrieves the resource it denotes on the web and displays it.
The first task is to parse the url to extract the name of the
protocol (here, necessarily
"http"), the address of the server,
the optional port and the absolute path of the document on the
server. This is done with Str, OCaml’s regular
Sending a simple request is a trivial task, as the following function shows.
Note that the url can be complete, with the address and port of the server, or just contain the requested path on the server.
Reading the response is even easier, since only the document is
returned, without any additional information. If there’s an error in
the request, the error message returned by the server as an
html document. Thus we just print the response with the
Misc.retransmit without indicating whether this is an
error or the desired document. The rest of the program establishes the
connection with the server.
We conclude, as usual, by parsing the command line.
We program an http relay (or proxy), which is a server that redirects http requests from a client to another server (or relay…) and forwards responses from that server back to the client.
The role of a relay is shown in in figure 5. When a client uses a relay, it addresses its requests to the relay rather than to the individual http servers located around the world. A relay has multiple advantages. It can store the responses to the most recent or frequent requests and serve them without querying the remote server (e.g. to avoid network overload or if the server is down). It can filter the responses (e.g. to remove advertisements or image, etc.). It can also simplify the development of a program by making it see the whole world wide web through a single server.
proxy port command launches the server on the port
(or if omitted, on the default port for http). We reuse the code of
get_url function (we assume that the functions above are
available in a
Url module). It only remains to write the code to
analyze the requests and set up the server.
We establish the service with the
function, thus we just need to define the function to handle a
and the rest of the program just establishes the service:
Simple http requests need one connection per request. This is inefficient because most requests on a server are followed by others (e.g. if a client gets a web page with images, it will subsequently request the images) and the time to establish a connection can easily exceed the time spent in handling the request itself (chapter 7 show how we can reduce this by handling the requests with threads rather than processes). Version 1.1 of the http described in rfc 2068 uses complex requests that allow to make multiple requests on a single connection3.
In complex requests, the server precedes every response with a header describing the format of the response and possibly the size of the document transmitted. The end of the document is no longer indicated by an end of file, since we know its size. The connection can therefore stay open to handle more requests. Complex requests have the following form:
header part defines a list of key-value fields with
the following syntax:
Superfluous spaces are allowed around the
':' separator and any
space can always be replaced by a tab or a sequence of spaces. The
header fields can also span several lines: in this case, and in this
case only, the
crlf end of line lexeme is immediately
followed by a space
sp. Finally, uppercase and lowercase
letters are equivalent in the keyword of fields and in the values of
Mandatory and optional fields depend on the type of request. For
GET request must have a field indicating the
For this type of request, we may also request, using the optional field If-Modified, that the document be returned only if it has been modified since a given date.
The number of fields in the
header is not fixed in
advance but indicated by the end of the header: a line
containing only the characters
Here is a complete request (on each line an implicit
A response to a complex request is also a complex response. It contains a status line, a header, and the body of the response, if any.
The fields of a response header have a syntax similar to that of a request but the required and optional fields are different (they depend on type of request and the status of the response — see the full documentation of the protocol).
body of the response can be transmitted in a
single block, in chunks or be empty:
Content-Lengthfield specifying in decimal ascii notation the number of bytes in the body.
Transfer-Encodingfield with the value “
chunked”. The body is then a set of chunks and ends with an empty chunk. A chunk is of the form:
sizeis the size of the chunk in hexadecimal notation and
chunkis a chunk of the response body of the given size (the part between
]is optional and can safely be ignored). The last, empty, chunk is always of the following form:
Content-Lengthfield and that it is not chunked, the body is empty (for instance, a response to a request of type
HEADcontains only a header).
Here is an example of a single block response:
200 indicates that the request was successful. A
301 means the url was redirected to another url defined in
Location field of the response. The
indicate errors on the client side while
5XX errors on the server
Add a cache to the relay. Pages are saved on the hard drive and when a requested page is available in the cache, it is served unless too old. In that case the server is queried again and the cache updated.
Write a program
wget such that
wget u1 u2 ... un makes the
un and saves the responses in
pi are respectively the name of the server and the
absolute path of the request
ui. The program should take
advantage of the protocol to establish a single connection to a
m when it is the same for consecutive requests.
Permanent url redirections
should be followed. The following options can also be added:
|Do not download the url if the file |
|Download recursively all the urls embedded in the responses which are documents in html format.|
A thread, also called a lightweight process, is a flow of control that can execute in parallel with other threads in the same program.
This chapter describes the functions that allow a program to create
Thread module) and synchronize by means of locks
Mutex module), conditions (
Condition module), and
synchronous events (
The creation of a thread is very different from the fork operation that creates a copy of the current process (and therefore a copy of the program). After a fork, the address spaces of the parent and child are totally disjoint, and the two processes can communicate only through system calls (like reading or writing a file or a pipe).
In contrast, all the threads within a program share the same address space. The only information that is not shared, and differentiates one thread from another, is the thread’s identity and its execution stack (along with certain system information such as the signal mask, the state of locks and conditions, etc.) From this viewpoint, threads resemble coroutines. The threads within a given program are all treated in the same fashion, except for the initial thread that was created when the program started. When this thread terminates, so do all the other threads and therefore the program as a whole. (Whenever we speak of multiple threads, we will implicitly mean threads within a given program.)
But unlike coroutines, which pass control explicitly from one to another and cannot execute in parallel, threads can execute in parallel and can be scheduled preemptively by the system. From this viewpoint, threads resemble processes.
The common address space permits threads to communicate directly among themselves using shared memory. The fact that threads can execute in parallel means that they must synchronize their access to shared data, so that one finishes writing before the other begins reading. Although not necessary in principle, in practice this requires going through the operating system. Synchronization is often a difficult part of programming with threads. It can be done with locks and conditions, or in a higher-level fashion with events.
The advantages of threads over processes are the lower cost of creation and the ability to exchange large data structures simply by passing pointers rather than copying.
On the other hand, using threads incurs the cost of managing the synchronization among them, including the case of a fatal error in one of the threads. In particular, a thread must be careful to release its locks and preserve its invariant before stopping. Processes may also be preferable to threads when we cannot really benefit from the latter’s advantages.
To compile an application using native threads, use the following:
ocamlbuild tool is used, all that is needed is to add the
following to the
If your installation does not support native threads, you can refer to section 7.8 or the manual for instructions how to use simulated “vm-level” threads. The text and examples in this chapter assume native threads and do not apply, in general, to vm-level threads.
The functions described in this section are defined in the
The system call create
f v creates a new thread that
executes the function application
f v and returns a thread id that
the caller can use to control the newly-created thread.
The function application executes concurrently with the other threads in the program. The thread terminates when the application returns and its result is simply ignored. If the thread terminates with an uncaught exception, the exception is not propagated to any other thread: a message is printed on the standard error output and the exception is otherwise ignored. (The other threads have proceeded independently and would not be able to receive the exception.)
A thread can also terminate prematurely with the system call
exit of the
Thread module, not to be confused with
Pervasives.exit that terminates the
entire program, i.e. all its threads.
The initial thread of a program implicitly calls the
Pervasives.exit function when it terminates.
When another thread terminates before the initial thread, it is
deallocated immediately by the OCaml runtime library. It does not
become a zombie as in the case of a Unix process created by
The system call self returns the thread id of the calling thread.
We already know enough to propose an alternative to the preceding
model for the concurrent server that used “fork” (or
“double fork”) — by using a thread rather than a child process.
To establish such a server, we introduce a variant
Misc.co_treatment of the function
defined in section 6.7.
If the thread was successfully created, the treatment is handled
entirely by the
service function, including closing
client_descr. Otherwise, we close the
descriptor, the client is abandoned, and we let the main program
handle the error.
Note that all the difficulty of the co-server is hidden in the
service function, which must handle the connection robustly until
disconnection. In the case of a concurrent server where the service
is executed by another process, premature termination of the service
due to a fatal error produces by default the desired behavior — closing
the connection — because the system closes the file descriptors when a
process exits. But in the case where the service is executed by a
thread, the descriptors of the different threads are shared by default
and not closed at the termination of the thread. It is therefore up
to the thread to close its descriptors before exiting. In addition, a
thread cannot call
Pervasives.exit in the case of a fatal error
during the handling of a service, because it would stop not only the
service but also the entire server. Calling
Thread.exit is often
not a solution either, because we risk not having properly
deallocated the thread’s open resources, and in particular the
One solution consists of raising an exception to signify a fatal stop
(for example an
Exit exception), causing finalization code to be
executed as it is handled. For similar reasons, it is essential to
sigpipe signal during the handling of a service by a
thread, replacing the immediate termination of the thread by the
raising of an
The functions described in this section are defined in the
The system call join allows one thread to wait for another to finish.
The calling thread is suspended until the thread with the given thread id has terminated its execution. This function can also be used by the principal thread to wait for all the other threads to finish before terminating itself and the program. (The default behavior is to kill the other threads without waiting for them to terminate.)
Although this call is blocking and therefore “long”, it
is restarted automatically when a signal is received: it is
effectively interrupted by the signal, the handler is invoked, then the
call is restarted. The call therefore does not return until the
thread has really terminated, and the call never raises the
exception. From the viewpoint of the OCaml programmer, it behaves as
if the signal was received at the moment when the call returns.
A thread does not return, since it is executed asynchronously. But its action can be observed — luckily! — by its side effects. For example, one thread can place the result of a computation in a reference that another thread will consult after making sure that the calculation has finished. We illustrate this in the following example.
The system can suspend one thread in order to give control temporarily
to another, or because it is waiting for a resource being used by
another thread (locks and conditions, for example) or by another
process (file descriptors, for example). A thread can also suspend
itself voluntarily. The
yield function allows a thread to give
up control explicitly, without waiting for preemption by the system.
It is a hint for the thread scheduler, but it may have no effect, for example if no other thread can execute immediately, the system may give control back to the same thread.
Conversely, it is not necessary to execute
yield to permit other
threads to execute, because the system reserves the right to execute
yield command itself at any moment. In fact, it exercises
this right sufficiently often to permit other threads to execute and
to give the illusion that the threads are running in parallel, even on
a uniprocessor machine.
We can revisit the example of section 3.3 and modify it to use threads rather than processes.
psearch k f v searches with
k threads in
parallel for an array element satisfying the function
pcond allows the search to be interrupted when an
answer has been found. All the threads share the same reference
found: they can therefore access it concurrently. No critical
section is required, because if different threads write to this resource
in parallel, they write the same value. It is important that the
threads do not write the result of the search when it is false!
For example, replacing line 7 by
would be incorrect.
The parallel search is interesting even on a uniprocessor machine if the comparison of elements could be blocked temporarily (for example by disk accesses or network connections). In this case, the thread performing the search passes control to another and the machine can therefore continue the computation on another part of the array and return to the blocked thread when its resource is free.
Access to certain elements can have significant latency, on the order
of a second if information must be retrieved over the network. In
this case, the difference in behavior between a sequential search and
a parallel search becomes obvious.
Parallelize quicksort on arrays. Answer.
Quicksort lends itself well to parallelization, because the sorting is done recursively on independent sub-arrays. This can be delegated to threads whose only synchronization is to wait for all the threads to finish their sorting, in order for the sub-array to be sorted.
It would be correct, but uninteresting, to exchange lines 20 and 21. In effect, that would wait for the lower portion of the array to be sorted before starting the sort of the upper portion. We would thus obtain the behavior of a sequential program, with the additional cost of threads but without obtaining any benefit.
In practice, we should limit parallelization to a reasonable factor and continue sequentially after that.
The other forms of suspension are tied to operating system resources.
A thread can be suspended for a certain time by calling
s seconds elapse, it can be restarted.
This primitive is provided for portability with vm-level threads, but
delay s is simply an abbreviation for
ignore (Unix.select    s). This call, unlike
not restarted when it is interrupted by a signal.
To synchronize a thread with an external operation, we can use the
select command. Note that this will block only the
calling thread and not the entire program. (The
redefines this function, because in the case of simulated threads
calling the one in the
Unix module would block the whole program
and therefore all the threads. It is therefore necessary to use
select from the
Thread module and not
Unix.select, even if the two are equivalent in the case of native
However, we cannot expect any significant gain from this example, which uses few processes relative to computation time.
The functions in this section are defined in the
Mutex module (as in
We mentioned above a problem of concurrent access to mutable resources. In particular, the following scenario illustrates the problem of access to shared resources. Consider a counter c and two processes p and q, each incrementing the counter in parallel.
Assume the scenario described in figure 6. Thread p reads the value of counter c, then gives control to q. In its turn, q reads the value of c, then writes the value k+1 to c. The thread p resumes control and writes the value k+1 to c. The final value of c is therefore k+1 instead of k+2.
This classic problem can be resolved by using locks that prevent arbitrary interleaving of p and q.
Locks are shared objects that can be held by at most a single thread
within a program at a time. A lock is created by the function
This function returns a new lock, initially not held by any thread. To acquire an existing lock, it is necessary to use the system call lock with the lock as argument. If the lock is held by another thread, the caller is frozen until the lock is released. A lock must be released explicitly by the thread that holds it with the system call unlock.
lock call behaves like
Thread.join with respect to
signals: if the thread receives a signal while executing
the signal will be noted (i.e. the OCaml runtime will be notified
that the signal has arrived), but the thread will continue to wait so
lock effectively returns only when the lock has been
acquired, and never raises the
EINTR exception. The real
treatment of the signal by OCaml will happen only upon the return
We can also try to acquire a lock without blocking with the system call trylock
This function returns
true if the lock has been acquired and
false otherwise. In the latter case, execution is not suspended
since the lock is not acquired. The thread can therefore do something
else and eventually return and try its luck later.
Incrementing a global counter used by several threads poses a synchronization problem: the instants between reading the value of the counter and writing the incremented value are in a critical region, i.e. two threads cannot be in this region at the same time. The synchronization can easily be managed with a lock.
The sole read operation on the counter poses no problem. It can be performed in parallel with a modification of the counter: the result will simply be the value of the counter just before or just after the modification, both results being consistent.
A common pattern is to hold a lock temporarily during a function call. It is of course necessary to make sure to release the lock at the end of the call, whether the call succeeded or failed. We can abstract this behavior in a library function:
In the preceding example, we could also have written:
An alternative to the model of the server with threads is to start a number of threads in advance which handle requests in parallel.
tcp_farm_server function behaves like
takes an additional argument which is the number of threads to start,
each of which will become a server at the same address. The advantage
of a pool of threads is to reduce the time to handle each connection
by eliminating the cost of creating a thread for it, since they are
created once and for all.
The only precaution to take is to ensure mutual exclusion around the
accept so that only one of the threads accepts a connection at a
time. The idea is that the
treat_connection function performs a
sequential treatment, but it is not a requirement — we can
effectively combine a pool of threads with the creation of new
threads, which can be dynamically adjusted depending on the load.
Acquisition of a lock is an inexpensive operation when it succeeds without blocking. It is generally implemented with a single “test-and-set” instruction provided by all modern processors (plus other small costs that are involved, such as updating caches). However, when the lock is not available, the process must be suspended and rescheduled later, which involves a significant additional cost. We must therefore incur this penalty only for a real suspension of a process in order to give control to another, and not for its potential suspension during the acquisition of a lock. Consequently, we will almost always want to release a lock as soon as possible and take it back later if necessary, rather than simply holding onto the lock. Avoiding these two operations would have the effect of enlarging the critical region and therefore the frequency with which another thread finds itself effectively in competition for the lock and in need of suspension.
Locks reduce interleaving. In return, they increase the risk of deadlock. For example, there is a deadlock if a thread p waits for a lock v held by a thread q which itself waits for a lock u held by p. (In the worst case, a thread waits for a lock that it holds itself.) Concurrent programming is difficult, and guarding against deadlock is not always easy. A simple way of avoiding this situation that is often possible consists of defining a hierarchy among the locks and ensuring that the order in which the locks are acquired dynamically respects the hierarchy: a thread never acquires a lock unless that lock is dominated by all the other locks that the thread already holds.
We modify the http relay developed in section 6.14 so that it services requests using threads.
Intuitively, it suffices to replace the
that creates a process clone with a function that creates a
thread. We must however take certain precautions. The challenge
with threads is that they share the entire memory space. We must
therefore ensure that the threads are not “stepping on each
other’s toes” with one undoing what was just done by another.
That typically happens when two threads modify the same mutable
structure in parallel.
In the case of the http server, there are several changes to make.
Let us start by resolving problems with access to resources. The
proxy_service function, described in section 6.14,
handles the treatment of connections. Via the intermediary functions
parse_request, it calls the
regexp_match function which uses the
Str library. However,
this library is not re-entrant (the result of the last search is
stored in a global variable). This example shows that we must beware
of calls to innocent-looking functions that hide potential collisions.
In this case we will not rewrite the
Str library but simply
sequentialize its use. It suffices to protect calls to this library
with locks (and there is really no other choice). We must still take
the precaution of releasing the lock when returning from the function
abnormally due to an exception.
To modify the existing code as little as possible, we can just rename
the definition of
regexp_match in the
Url module as
unsafe_regexp_match and then define
regexp_match as a
protected version of
The change is rather minimal. It should be noted that the
regexp_match function includes both the expression matching and
the extraction of the matched groups. It would definitely have been
incorrect to protect the
Str.matched_group functions individually.
Another solution would be to rewrite the analysis functions without using
Str library. But there is no reason for such a choice, since
synchronizing the library primitives is easy to do and does not turn
out to be a source of inefficiency. Obviously, a better solution
would be for the
Str library to be re-entrant in the first place.
The other functions that are called are already re-entrant, in particular
Misc.retransmit function that allocates different buffers for
However, there are still some precautions to take regarding error
handling. The handling of a connection by a thread must be robust, as
explained above. In particular, in case of error, the other threads
must not be affected. In other words, the thread must terminate
“normally”, properly closing the connection in question and
going back to accepting other pending connections. We must first of
all replace the call to
handle_error because it is
essential not to kill the whole process. A call to
would not be correct either, because thread termination does not close
its (shared) descriptors, the way the system does for process
termination. An error in the handling of a connection would leave the
connection open. The solution consists of raising an
exception that allows the finalization code to do what is required.
We must now protect
treat_connection by catching all errors, in
Exit but also
EPIPE, which can be raised if the
client closes the connection prematurely. We will take care of this
by using a protected function.
The functions described in this section are defined in the Condition module.
Synchronization with locks is very simple, but it is not sufficient: locks allow waiting for shared data to be free, but do not allow waiting for the data to have a particular state. Let us replace the example of a counter by a (first-in/first-out) queue shared among several threads. Adding a value to the queue can be synchronized by using a lock as above, since no matter what the state of the queue, we can always add an element. But what about removing an element from the queue? What should be done when the queue is empty? We cannot hold the lock while waiting for the queue to be filled, because that would completely prevent another thread from filling the queue. So it must be released. But how can we know when the queue is no longer empty, except by testing it periodically? This solution, called “busy-waiting”, is definitely not satisfactory. Either it consumes computing cycles unnecessarily (period too short) or else it it is not reactive enough (period too long).
Conditions provide a solution to this problem. A thread that
holds a lock can wait for a condition object until another thread
sends a signal on that condition. As with locks, conditions are
passive structures that can be manipulated by synchronization
functions. They can be created by the
A process p that already holds a lock
v can wait on a
c and the lock
v with the system call wait.
The process p informs the system that it is waiting on the condition
c and the lock
v, then releases the lock
v and goes to
sleep. It will not be woken up by the system until another thread q
signals a change on the condition
c and the lock
available; the process p will then hold the lock
Note: it is an error to call
wait c v without holding the lock
v. The behavior of
wait c v with respect to signals is the
same as for
When a thread signals a change on a condition, it can either ask for all threads waiting on that condition to be woken up (system call broadcast), or else for just one of them to be woken up (system call signal).
Sending a signal or a broadcast on a condition does not require holding a lock (unlike waiting), in the sense that it will not trigger a “system” error. However, it can sometimes be a programming error.
The choice between waking up one thread or all the threads depends on the problem. To consider the example of the queue again, if a thread adds an element to an empty queue, there is no need to wake up all the others, since only one will effectively be able to remove that element. On the other hand, if it adds a number of elements that is either not statically known or very large, it must wake up all the threads. Note that if adding an element to a non-empty queue does not send a signal, then adding an element to an empty queue must send a broadcast, since it could be followed immediately by another addition (without a signal) and therefore behave like a multiple addition. In summary, either send a signal on every addition, or send a broadcast only when adding to an empty queue. The choice between these two strategies is a bet on whether the queue is usually empty (first solution) or usually non-empty (second solution).
Often, one thread knows only an approximation of the reason why another thread is waiting on a condition. It will therefore signal the condition whenever the situation might be what the other thread is waiting for. An awakened thread, therefore, cannot assume that the condition it was waiting is now satisfied. It must, in general, re-test the state of its shared data, and if necessary wait on the condition again. This does not constitute busy-waiting, because it only happens when another thread signals the condition.
Here is another justification for this approach: when a thread has
just produced a lot of some resource and wakes all the others using a
broadcast, nothing prevents the first one that wakes up from
being greedy and exhausting the entire resource. The second one to
wake up must go back to sleep, hoping to be luckier next time.
We can now give a concrete solution for shared queues. The queue
structure defined in the Queue module is extended with a lock and
Addition never blocks, but we must not forget to signal
non_empty condition when the list is empty beforehand,
because it is possible that someone is waiting on the condition.
Removal is a little more complicated: after acquiring the lock, we
must try to remove an element from the queue. If the queue is empty,
we must wait on the
non_empty condition. When awakened, we try
again, knowing that we already have the lock.
As explained above, the
broadcast q.non_empty signal
(line 9) is executed by a thread p already in
possession of the lock
This implies that a reader thread q executing the
cannot be between line 15 and 16
where it would have verified that the queue is empty but not yet have
gone to sleep. In this case, the signal sent by p would be
ineffective and ignored, since q has not gone to sleep yet; but q
would then go to sleep and not be woken up, because p has already
sent its signal.
The lock therefore guarantees that either q is already asleep or
else has not yet tested the state of the queue.
Implement a variant in which the queue is bounded: addition to the queue becomes blocking when the size of the queue reaches a fixed value. (In a concurrent world, we might need this scheme to avoid having a producer that produces endlessly while the consumer is blocked.) Answer.
Addition is a combination of the preceding versions of the
take functions above.
Removal is symmetric to addition (and must now signal
when the queue is full beforehand), and is left to the reader.
We get the behavior of unbounded queues by choosing
The functions described in this section are defined in the Event module.
Locks and conditions together allow all forms of synchronization to be expressed. However, their use is not always easy, as shown by the example of the initially simple queue whose synchronization code subsequently turned out to be subtle.
Event-based synchronous communication is a collection of higher-level
communication primitives that tend to facilitate concurrent
programming. The primitives in the
Event module were initially
developed by John Reppy as an extension of the Standard ML
language called Concurrent ML . In OCaml, these
primitives are located above the more elementary synchronization of
locks and conditions.
Communication occurs by sending events along channels.
Channels are like “lightweight pipes”: they allow communication
among threads in the same program and take care of synchronization
between producers and consumers. A channel carrying values of type
'a has the type
'a Event.channel. Channels are homogeneous
and therefore always carry values of the same type. A channel is
created with the
Sending or receiving a message is not done directly, but through the intermediary of an event. An elementary event is “sending a message” or “receiving a message”. They are constructed by means of the following primitives:
Construction of a message does not have an immediate effect: it just
creates a data structure describing the action to be done. To make an
event happen, the thread must synchronize with another thread wishing
to make the complementary event happen. The
allows a thread to wait for the occurrence of the event passed
Thus, to send a value
v on the channel
c, we can execute
sync (send c v). The thread is suspended until the event occurs,
that is to say until another thread is ready to receive a value on the
c. In a symmetric fashion, a thread can wait for a
message on channel
c by performing
sync (receive c).
There is a competition among all the producers on one hand and all the consumers on the other. For example, if several threads try to send a message on a channel but only one is ready to read it, it is clear that only one producer will make the event occur. The others will remain suspended, without even noticing that another was “served” ahead of them.
The competition can also occur within the same thread.
Multiple events can be combined by the
The resulting event is an offer, in parallel, of the events passed as
arguments, and occurs when exactly one of them occurs. We distinguish
between the offer of an event and its occurrence. The call
sync (choose [e1; e2]) synchronizes by offering a choice of two events
e2, but only one of the two events will effectively
occur (the offer of the other event will be simultaneously canceled).
wrap_abort primitive allows to handle an event being
wrap_abort e f creates an event that is equivalent to
e, but if it is not chosen during synchronization, then the
f is executed. (This is only interesting when it is
part of a complex event.)
A thread can try to synchronize on an event without blocking (somewhat
poll e offers the event
e but if it cannot occur
immediately, it cancels the offer rather than blocking and has no
effect (or more exactly, behaves as if the expression
poll e had
been replaced by the value
None). By contrast, if the event can
happen immediately, then it behaves as if the thread had done
sync e, except that the value
Some v is returned
In section 7.3 the example of the Sieve of Eratosthenes, the communication between different threads is done with pipes as in the original program, using system memory (the pipe) as intermediary. We may think that it is more efficient to communicate directly by using the memory of the process. A simple solution consists of replacing the pipe by a channel on which integers are sent.
Sending integers on the channel is not sufficient, because we must
also be able to detect the end of the stream. The simplest is
therefore to pass elements of the form
Some n and to terminate by
sending the value
None. To minimize the changes, we use the code
of the example in section 5.2. We simulate pipes and the
functions for reading and writing pipes by channels and functions for
reading and writing channels.
It is sufficient to take the previous version of the program and
change the input/output functions to ones that read and write a channel,
rather than an input/output buffer from the
For example, we can insert the following code at the beginning of the
program just after the
open Unix;; directive:
However, if we compare the efficiency of this version with the previous one, we find that it is twice as slow. Communication of each integer requires a synchronization between two threads and therefore several system calls for acquiring and releasing locks. On the other hand, communication via pipes uses buffered i/o that allows several thousand integers to be exchanged with each system call.
To be fair, one should also provide buffered communication on channels, using the channel only to exchange a packet of integers. The child can accumulate the results in a private queue, to which it can therefore write without synchronization. When the queue is full, or upon an explicit request, it is emptied by synchronizing on the channel. The parent has its own queue that it receives by synchronizing and empties gradually.
Here is a solution:
This version allows us to regain efficiency comparable to (but not better than) the version with pipes.
Compared to the original version with processes and pipes, there are two potential advantages. First, threads are more lightweight and less costly to launch. Second, communication on a channel merely passes a pointer, without copying. But these advantages are not noticeable here, because the number of threads created and the data exchanged are not big enough compared to the cost of system calls and compute time.
In conclusion, we can say that communication between threads has a cost of up to one system call (if the process must be suspended) and the cost can be significantly reduced by buffering communication and sending larger structures less often.
An http server can be subjected to a high, bursty load. To improve response time, we can refine the architecture of an http server by always keeping a dozen threads ready to handle new requests. This means that a thread does not handle only a single request, but a potentially infinite series of requests that it reads from a queue.
To avoid overloading the machine, we can limit the number of threads to a reasonable value beyond which the overhead of managing tasks exceeds the latency for servicing requests (time spent waiting for data on disk, etc.). After that, we can keep some connections waiting to be handled, and then finally we can refuse connections. When the load diminishes and the number of threads is above the “ideal” value, some of them are allowed to die and the others remain ready for the next requests.
Transform the example of section 7.5 into this architecture.
The Unix system was not originally designed to provide support for
threads. However, most modern Unix implementations now offer such
support. Nevertheless, threads remain an add-on that is sometimes
apparent. For example, when using threads it is strongly discouraged
to use fork except when doing
exec immediately afterward.
fork copies the current thread, which becomes a
crippled process that runs believing it has threads when in fact they
do not exist. The parent continues to run normally as before.
The special case of a call to
fork where the child immediately
launches another program does not cause the parent any problem.
Luckily, since that is the only way to start other programs!
Inversely, one can do
fork (not followed by
exec), and then launch
several threads in the child and the parent, without any problem.
When the underlying operating system has threads, OCaml can provide a native implementation of threads, leaving their management to the operating system as much as possible. Each thread then lives in a different Unix process but shares the same address space.
When the system does not provide support for threads, OCaml can emulate them. All the threads then execute in the same Unix process, and their management, including their scheduling, is handled by the OCaml runtime system. However, this implementation is only available when compiling to bytecode.
The OCaml system provides the same programming interface for the native and simulated versions of threads. The implementation of threads is therefore split: one implementation for the emulated version that includes its own task controller, and another implementation that is based on posix (1003.1c) threads and lifts the corresponding library functions to the level of the OCaml language. In the process, the OCaml language handles certain simple administrative tasks and ensures an interface identical to the emulated version. This guarantees that a program compilable on one Unix architecture remains compilable on another Unix architecture. However, whether threads are emulated or native can change the synchronization of calls to the C library, and therefore change, despite everything, the semantics of the program. It is therefore necessary to take certain precautions before believing that a program will behave the same way in these two versions. In this chapter, the discussion mainly concern these two implementations, but recall that by default, we have taken the viewpoint of a native implementation.
To use emulated threads, one must pass the
-threads to the
ocamlc compiler. This option is
not accepted by the
The implementation of threads in OCaml must face one of the peculiarities of the OCaml language: the automatic management of memory and its high consumption of allocated data. The solution adopted, which is the simplest and also generally the most efficient, is to sequentialize the execution of OCaml code in all threads: a lock in the runtime system prevents two threads from executing OCaml code simultaneously. This seems contrary to the whole idea of threads, but it is not, since the lock is released before blocking system calls and reacquired upon return. Other threads can therefore take control at that moment. A special case of such a system call is the call to sched_yield, performed at regular intervals to suspend the running thread and give control to another.
On a multiprocessor machine, the only source of true parallelism comes from the execution of C code and system calls. On a uniprocessor machine, the fact that the OCaml code is sequentialized is not really noticeable.
The programmer cannot rely on this sequentialization, because one thread can give control to another at almost any moment. With one exception, the sequentialization guarantees memory coherence: two threads always have the same view of memory, except perhaps when they execute C code. In effect, the passing of the lock implies a synchronization of the memory: a read operation by one thread occurring after a write operation to the same address by another thread will always return the freshly-written value, with no need for additional synchronization.
Generally speaking, using signals is already delicate with a single thread due to their asynchronous character. It is even more so in the presence of multiple threads because of the addition of new difficulties: which thread should a signal be sent to? To all, to the primary one, or to the one currently running? What happens if one thread sends a signal to another? In fact, threads were implemented before answering these questions, and different implementations can behave differently with respect to signals.
despite being long system calls, are not interruptible by a signal.
(They cannot therefore fail with the
EINTR error.) If a signal
is sent while waiting, it will be received and handled when the call
The posix standard specifies that the signal handler is shared among all the threads and in contrast the signal mask is private to each thread and inherited upon creation of a thread. But the behavior of threads with respect to signals remains largely underspecified and therefore non-portable.
It is therefore preferable to avoid as much as possible the use of
asynchronous signals (such as
sigchld, etc.) with threads. These can be blocked and examined
We can dedicate a thread to signal
handling and nothing else: it can wait for the reception of signals,
undertake the necessary actions, and update certain information
examined by other threads.
In addition, OCaml threads (since version 3.08) use the
sigvtalarm signal internally to implement preemption of threads.
This signal is therefore reserved and must not be used by the program
itself, since there is a risk of interference.
We have shown how OCaml’s Sys, Unix, and Threads modules can be used to program applications that interact with the operating system.
These modules allow the invocation of the most important Unix system calls directly from OCaml. Some of these calls were replaced by higher-level functions, either to facilitate programming or to maintain invariants needed by OCaml’s runtime system. In any case, this higher-level access to the Unix system streamlines the development of applications.
Not every feature of the Unix system is available through these modules, however it is still possible to access the missing ones by writing C bindings.
Another useful library is Cash  which focuses on writing
scripts directly in OCaml. This library completes the
module in two different ways. First, in addition to a few helper
functions to write scripts, it provides, on top of the
module, a few system call variations to assist the programmer in
process and pipe management. Second, it offers additional entry points
to the Unix system.
Links in bold refer to the description of a posix system call.