doc/libpsx.3

.TH LIBPSX 3 "2021-01-31" "" "Linux Programmer's Manual"
.SH NAME
psx_syscall3, psx_syscall6 \- POSIX semantics for system calls
.SH SYNOPSIS
.nf
.B #include <sys/psx_syscall.h>
.sp
.BI "long int psx_syscall3(long int" " syscall_nr, " "long int" " arg1, " "long int" " arg2, " "long int" " arg3);"
.sp
.BI "long int psx_syscall6(long int" " syscall_nr, " "long int" " arg1, " "long int" " arg2, " "long int" " arg3, " "long int" " arg4, " "long int" " arg5, " "long int" " arg6);"
.sp
Link with one of these:
.sp
.I   ld ... \-lpsx \-lpthread \-\-wrap=pthread_create
.sp
.I   gcc ... \-lpsx \-lpthread \-Wl,\-wrap,pthread_create
.SH DESCRIPTION
The
.B libpsx
library attempts to fill a gap left by the
.BR pthreads (7)
implementation on Linux. To be compliant POSIX threads, via the
.BR nptl "(7) " setxid
mechanism glibc maintains consistent UID and GID credentials amongst
all of the threads associated with the current process. However, other
credential state is not supported by this abstraction. To support
these extended kernel managed security attributes,
.B libpsx
provides a more generic pair of wrapping system call functions:
.BR psx_syscall3 "() and " psx_syscall6 ().
Like the
.B setxid
mechanism, the coordination of thread state is mediated by a realtime
signal. Whereas the
.B nptl:setxid
mechanism uses signo=33 (which is hidden by glibc below a redefined
SIGRTMIN),
.B libpsx
inserts itself in the SIGSYS handler stack. It goes to great length to
be the first such handler but acts as a pass-through for other SIGSYS
uses.
.PP
A linker trick of
.I wrapping
the
.BR pthread_create ()
call with a psx thread registration function is used to ensure
.B libpsx
can keep track of all pthreads.
.PP
An inefficient macrology trick supports the
.BR psx_syscall ()
pseudo function which takes 1 to 7 arguments, depending on the needs
of the caller. The macrology pads out the call to actually use
.BR psx_syscall3 ()
or
.BR psx_syscall6 ()
with zeros filling the missing arguments. While using this in source
code will make it appear clean, the actual code footprint is
larger. You are encouraged to use the more explicit
.BR psx_syscall3 ()
and
.BR psx_syscall6 ()
functions as needed.
.SH RETURN VALUE
The return value for system call functions is generally the value
returned by the kernel, or \-1 in the case of an error. In such cases
.BR errno (3)
is set to the detailed error value. The
.BR psx_syscall3 "() and " psx_syscall6 ()
functions attempt a single threaded system call and return immediately
in the case of an error. Should this call succeed, then the same
system calls are executed from a signal handler on each of the other
threads of the process.
.SH CONFORMING TO
The needs of
.BR libcap (3)
for POSIX semantics of capability manipulation. You can read more
about why this is needed here:
.TP
https://sites.google.com/site/fullycapable/who-ordered-libpsx
.SH "REPORTING BUGS"
Please report bugs via:
.TP
https://bugzilla.kernel.org/buglist.cgi?component=libcap&list_id=1047723&product=Tools&resolution=---
.SH SEE ALSO
.BR libcap (3),
.BR pthreads "(7) and"
.BR nptl (7).