@@@ man wip

[mLib] / sys / mdup.3

.\" -*-nroff-*-
.de VS
.sp 1
.RS
.nf
.ft B
..
.de VE
.ft R
.fi
.RE
.sp 1
..
.de hP
.IP
.ft B
\h'-\w'\\$1\ 'u'\\$1\ \c
.ft P
..
.ie t \{\
.  ds o \(bu
.  de VP
.    sp .4v
..
\}
.el \{\
.  ds o o
.  de VP
.    sp
..
\}
.TH mdup 3 "4 January" "Straylight/Edgeware" "mLib utilities library"
.SH NAME
mdup \- renumber file descriptors
.SH SYNOPSIS
.nf
.B "#include <mLib/mdup.h>"
.PP
.ta 2n
.B "typedef struct {"
.B "	int cur;"
.B "	int want;"
.B "} mdup_fd;"
.PP
.BI "int mdup(mdup_fd *" v ", size_t " n ");"
.fi
.SH DESCRIPTION
The
.B mdup
function renumbers file descriptors, using the
.BR dup (2)
and
.BR dup2 (2)
system calls.  Its arguments are a pointer
.I v
to a vector of
.B mdup_fd
structures, and the length
.I n
of this vector, in elements.  Each `slot' (element) in the vector
.I v
represents a file.  The slot's
.B cur
member names the current file descriptor for this file; the
.B want
member is the file descriptor to move it to.  In order to keep a file
alive when you don't care which descriptor it ends up with, set
.I want
= \-1.  Several slots may specify the same
.B cur
descriptor; but they all have to declare different
.BR want s
(except that several slots may have
.I want
= \-1.
.PP
On successful exit, the function will have rearranged the file
descriptors as requested.  To reflect this, the
.B cur
members will all be set to match the
.B want
members (except where the latter are \-1).
.PP
If there is a failure, then some rearrangement may have been performed
and some not; the
.B cur
members are set to reflect which file descriptors are to be used.
.PP
The old file descriptors are
.IR closed .
This is different from usual
.BR dup (2)
behaviour, of course, but essential for reliable error handling.  If you
want to keep a particular source file descriptor open as well as make a
new copy then specify two slots with the same
.BR cur ,
one with
.B want " = " cur
and one with the desired output descriptor.
.PP
The
.B mdup
function is capable of arbitrary file descriptor remappings.  In
particular, it works correctly even if the desired remappings contain
cycles.
.SS "Background: the problem that mdup solves"
The
.B mdup
function is intended to be used to adjust file descriptors prior to
invoking one of the
.B exec
system calls.  The standard use of
.BR dup (2)
to establish the child process's standard input/output/error files is
prone to errors in the case where the newly opened file in fact already
has one of the relevant file descriptors.
.PP
Consider the case where we want to run a process with separate pipes
attached to each of the standard descriptors.  Typical code looks like
this.
.VS
#define P_INIT { \-1, \-1 }
int p_in[2] = P_INIT, p_out[2] = P_INIT, p_err[2] = P_INIT;
pid_t kid = -1;
int i;
.VP
if (pipe(p_in) || pipe(p_out) || pipe(p_err)) goto error;
if ((kid = fork()) < 0) goto error;
if (!kid) {
.ta 2n 4n 2n+\w'\fBif ('u
	if (dup2(p_in[0], STDIN_FILENO) < 0 ||
			dup2(p_out[1], STDOUT_FILENO) < 0 ||
			dup2(p_err[2], STDERR_FILENO) < 0 ||
			close(p_in[0]) || close(p_out[0]) || close(p_err[0]) ||
			close(p_in[1]) || close(p_out[1]) || close(p_err[1]))
		_exit(127);
	execvp("/bin/sh", "sh", "-c", "...", (char *)0);
}
\&...
.VE
Now suppose that, in the parent process, the standard input, output and
error descriptors are all initially closed.  After the calls to
.BR pipe (2),
descriptors 0, 1, and 2 refer to
.BR p_in[0] ,
.BR p_in[1] ,
and
.B p_out[0]
respectively.  In the child process, the calls to
.BR dup2 (2)
rearrange these.  But then the
.BR close (2)
calls will immediately close all three descriptors, before
.BR exec ing
the child.
.PP
Here's how to rewrite the above function using
.BR mdup .
.VS
.ta 2n 4n 2n+\w'\fBmd[0].cur = p_out[1]; 'u
#define P_INIT { \-1, \-1 }
int p_in[2] = P_INIT, p_out[2] = P_INIT, p_err[2] = P_INIT;
pid_t kid = -1;
mdup_fd md[3];
int i;
.VP
if (pipe(p_in) || pipe(p_out) || pipe(p_err)) goto error;
if ((kid = fork()) < 0) goto error;
if (!kid) {
	if (close(p_in[1] || close(p_out[0]) || close(p_err[0]))
		goto _exit(127);
	md[0].cur = p_in[0];		md[0].want = STDIN_FILENO;
	md[1].cur = p_out[1];		md[1].want = STDOUT_FILENO;
	md[2].cur = p_err[1];		md[2].want = STDERR_FILENO;
	if (mdup(md, 3)) _exit(127);
	execvp("/bin/sh", "sh", "-c", "...", (char *)0);
}
\&...
.VE
One can see that, not only is the resulting program more correct, it's
also simpler.  Note that we close the unwanted ends of the pipes
.I before
invoking
.BR mdup .
Closing them afterwards risks interfering with the newly assigned
descriptors which are meant to be passed to the child process.  Note
also that
.B mdup
has taken responsibility for closing the other descriptors for the
wanted ends of the pipes.
.SH "SEE ALSO"
.BR dup (2),
.BR dup2 (2),
.BR mLib (3).
.SH AUTHOR
Mark Wooding, <mdw@distorted.org.uk>