.\" Copyright (C) 2012, Cyrill Gorcunov <gorcunov@openvz.org>
.\" and Copyright (C) 2012, 2016, Michael Kerrisk <mtk.manpages@gmail.com>
.\"
.\" SPDX-License-Identifier: Linux-man-pages-copyleft
.\"
.\" Kernel commit d97b46a64674a267bc41c9e16132ee2a98c3347d
.\"
.TH kcmp 2 2023-03-30 "Linux man-pages 6.04"
.SH NAME
kcmp \- compare two processes to determine if they share a kernel resource
.SH LIBRARY
Standard C library
.RI ( libc ", " \-lc )
.SH SYNOPSIS
.nf
.BR "#include <linux/kcmp.h>" "       /* Definition of " KCMP_* " constants */"
.BR "#include <sys/syscall.h>" "      /* Definition of " SYS_* " constants */"
.B #include <unistd.h>
.PP
.BI "int syscall(SYS_kcmp, pid_t " pid1 ", pid_t " pid2 ", int " type ,
.BI "            unsigned long " idx1 ", unsigned long "  idx2 );
.fi
.PP
.IR Note :
glibc provides no wrapper for
.BR kcmp (),
necessitating the use of
.BR syscall (2).
.SH DESCRIPTION
The
.BR kcmp ()
system call can be used to check whether the two processes identified by
.I pid1
and
.I pid2
share a kernel resource such as virtual memory, file descriptors,
and so on.
.PP
Permission to employ
.BR kcmp ()
is governed by ptrace access mode
.B PTRACE_MODE_READ_REALCREDS
checks against both
.I pid1
and
.IR pid2 ;
see
.BR ptrace (2).
.PP
The
.I type
argument specifies which resource is to be compared in the two processes.
It has one of the following values:
.TP
.B KCMP_FILE
Check whether a file descriptor
.I idx1
in the process
.I pid1
refers to the same open file description (see
.BR open (2))
as file descriptor
.I idx2
in the process
.IR pid2 .
The existence of two file descriptors that refer to the same
open file description can occur as a result of
.BR dup (2)
(and similar)
.BR fork (2),
or passing file descriptors via a domain socket (see
.BR unix (7)).
.TP
.B KCMP_FILES
Check whether the processes share the same set of open file descriptors.
The arguments
.I idx1
and
.I idx2
are ignored.
See the discussion of the
.B CLONE_FILES
flag in
.BR clone (2).
.TP
.B KCMP_FS
Check whether the processes share the same filesystem information
(i.e., file mode creation mask, working directory, and filesystem root).
The arguments
.I idx1
and
.I idx2
are ignored.
See the discussion of the
.B CLONE_FS
flag in
.BR clone (2).
.TP
.B KCMP_IO
Check whether the processes share I/O context.
The arguments
.I idx1
and
.I idx2
are ignored.
See the discussion of the
.B CLONE_IO
flag in
.BR clone (2).
.TP
.B KCMP_SIGHAND
Check whether the processes share the same table of signal dispositions.
The arguments
.I idx1
and
.I idx2
are ignored.
See the discussion of the
.B CLONE_SIGHAND
flag in
.BR clone (2).
.TP
.B KCMP_SYSVSEM
Check whether the processes share the same
list of System\ V semaphore undo operations.
The arguments
.I idx1
and
.I idx2
are ignored.
See the discussion of the
.B CLONE_SYSVSEM
flag in
.BR clone (2).
.TP
.B KCMP_VM
Check whether the processes share the same address space.
The arguments
.I idx1
and
.I idx2
are ignored.
See the discussion of the
.B CLONE_VM
flag in
.BR clone (2).
.TP
.BR KCMP_EPOLL_TFD " (since Linux 4.13)"
.\" commit 0791e3644e5ef21646fe565b9061788d05ec71d4
Check whether the file descriptor
.I idx1
of the process
.I pid1
is present in the
.BR epoll (7)
instance described by
.I idx2
of the process
.IR pid2 .
The argument
.I idx2
is a pointer to a structure where the target file is described.
This structure has the form:
.PP
.in +4n
.EX
struct kcmp_epoll_slot {
    __u32 efd;
    __u32 tfd;
    __u64 toff;
};
.EE
.in
.PP
Within this structure,
.I efd
is an epoll file descriptor returned from
.BR epoll_create (2),
.I tfd
is a target file descriptor number, and
.I toff
is a target file offset counted from zero.
Several different targets may be registered with
the same file descriptor number and setting a specific
offset helps to investigate each of them.
.PP
Note the
.BR kcmp ()
is not protected against false positives which may occur if
the processes are currently running.
One should stop the processes by sending
.B SIGSTOP
(see
.BR signal (7))
prior to inspection with this system call to obtain meaningful results.
.SH RETURN VALUE
The return value of a successful call to
.BR kcmp ()
is simply the result of arithmetic comparison
of kernel pointers (when the kernel compares resources, it uses their
memory addresses).
.PP
The easiest way to explain is to consider an example.
Suppose that
.I v1
and
.I v2
are the addresses of appropriate resources, then the return value
is one of the following:
.RS
.TP
.B 0
.I v1
is equal to
.IR v2 ;
in other words, the two processes share the resource.
.TP
.B 1
.I v1
is less than
.IR v2 .
.TP
.B 2
.I v1
is greater than
.IR v2 .
.TP
.B 3
.I v1
is not equal to
.IR v2 ,
but ordering information is unavailable.
.RE
.PP
On error, \-1 is returned, and
.I errno
is set to indicate the error.
.PP
.BR kcmp ()
was designed to return values suitable for sorting.
This is particularly handy if one needs to compare
a large number of file descriptors.
.SH ERRORS
.TP
.B EBADF
.I type
is
.B KCMP_FILE
and
.I fd1
or
.I fd2
is not an open file descriptor.
.TP
.B EFAULT
The epoll slot addressed by
.I idx2
is outside of the user's address space.
.TP
.B EINVAL
.I type
is invalid.
.TP
.B ENOENT
The target file is not present in
.BR epoll (7)
instance.
.TP
.B EPERM
Insufficient permission to inspect process resources.
The
.B CAP_SYS_PTRACE
capability is required to inspect processes that you do not own.
Other ptrace limitations may also apply, such as
.BR CONFIG_SECURITY_YAMA ,
which, when
.I /proc/sys/kernel/yama/ptrace_scope
is 2, limits
.BR kcmp ()
to child processes;
see
.BR ptrace (2).
.TP
.B ESRCH
Process
.I pid1
or
.I pid2
does not exist.
.SH STANDARDS
Linux.
.SH HISTORY
Linux 3.5.
.PP
Before Linux 5.12,
this system call is available only if the kernel is configured with
.BR CONFIG_CHECKPOINT_RESTORE ,
since the original purpose of the system call was for the
checkpoint/restore in user space (CRIU) feature.
(The alternative to this system call would have been to expose suitable
process information via the
.BR proc (5)
filesystem; this was deemed to be unsuitable for security reasons.)
Since Linux 5.12,
this system call is also available if the kernel is configured with
.BR CONFIG_KCMP .
.SH NOTES
See
.BR clone (2)
for some background information on the shared resources
referred to on this page.
.SH EXAMPLES
The program below uses
.BR kcmp ()
to test whether pairs of file descriptors refer to
the same open file description.
The program tests different cases for the file descriptor pairs,
as described in the program output.
An example run of the program is as follows:
.PP
.in +4n
.EX
$ \fB./a.out\fP
Parent PID is 1144
Parent opened file on FD 3

PID of child of fork() is 1145
	Compare duplicate FDs from different processes:
		kcmp(1145, 1144, KCMP_FILE, 3, 3) ==> same
Child opened file on FD 4
	Compare FDs from distinct open()s in same process:
		kcmp(1145, 1145, KCMP_FILE, 3, 4) ==> different
Child duplicated FD 3 to create FD 5
	Compare duplicated FDs in same process:
		kcmp(1145, 1145, KCMP_FILE, 3, 5) ==> same
.EE
.in
.SS Program source
\&
.\" SRC BEGIN (kcmp.c)
.EX
#define _GNU_SOURCE
#include <err.h>
#include <fcntl.h>
#include <linux/kcmp.h>
#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/syscall.h>
#include <sys/wait.h>
#include <unistd.h>

static int
kcmp(pid_t pid1, pid_t pid2, int type,
     unsigned long idx1, unsigned long idx2)
{
    return syscall(SYS_kcmp, pid1, pid2, type, idx1, idx2);
}

static void
test_kcmp(char *msg, pid_t pid1, pid_t pid2, int fd_a, int fd_b)
{
    printf("\et%s\en", msg);
    printf("\et\etkcmp(%jd, %jd, KCMP_FILE, %d, %d) ==> %s\en",
           (intmax_t) pid1, (intmax_t) pid2, fd_a, fd_b,
           (kcmp(pid1, pid2, KCMP_FILE, fd_a, fd_b) == 0) ?
                        "same" : "different");
}

int
main(void)
{
    int                fd1, fd2, fd3;
    static const char  pathname[] = "/tmp/kcmp.test";

    fd1 = open(pathname, O_CREAT | O_RDWR, 0600);
    if (fd1 == \-1)
        err(EXIT_FAILURE, "open");

    printf("Parent PID is %jd\en", (intmax_t) getpid());
    printf("Parent opened file on FD %d\en\en", fd1);

    switch (fork()) {
    case \-1:
        err(EXIT_FAILURE, "fork");

    case 0:
        printf("PID of child of fork() is %jd\en", (intmax_t) getpid());

        test_kcmp("Compare duplicate FDs from different processes:",
                  getpid(), getppid(), fd1, fd1);

        fd2 = open(pathname, O_CREAT | O_RDWR, 0600);
        if (fd2 == \-1)
            err(EXIT_FAILURE, "open");
        printf("Child opened file on FD %d\en", fd2);

        test_kcmp("Compare FDs from distinct open()s in same process:",
                  getpid(), getpid(), fd1, fd2);

        fd3 = dup(fd1);
        if (fd3 == \-1)
            err(EXIT_FAILURE, "dup");
        printf("Child duplicated FD %d to create FD %d\en", fd1, fd3);

        test_kcmp("Compare duplicated FDs in same process:",
                  getpid(), getpid(), fd1, fd3);
        break;

    default:
        wait(NULL);
    }

    exit(EXIT_SUCCESS);
}
.EE
.\" SRC END
.SH SEE ALSO
.BR clone (2),
.BR unshare (2)