kcmp(2)

Compare whether two resources in different processes refer to the same kernel object.

Since: Linux 3.5check your kernel →
glibc: none (use syscall(2))
System: Linux
Man page: kcmp(2)

Synopsis

#include <linux/kcmp.h>
#include <sys/syscall.h>
#include <unistd.h>

int syscall(SYS_kcmp, pid_t pid1, pid_t pid2, int type,
            unsigned long idx1, unsigned long idx2);

Rod's take

No glibc wrapper; use syscall(SYS_kcmp). Return value is strcmp-like: zero means same object, non-zero means different but ordered by kernel address (useful for CRIU sorting, meaningless otherwise). Requires ptrace access to both processes. KCMP_FILE is the common use case: it distinguishes a dup'd fd from an independently opened one, which /proc inode comparison cannot do without a race.

Flags and arguments that matter

Name	What it does
KCMP_FILE	idx1 and idx2 are fd numbers in pid1 and pid2. Returns zero if they share the same open file description (same struct file). Duplicated fds and SCM_RIGHTS-passed fds share a description; independently opened fds do not.
KCMP_VM	idx1/idx2 ignored. Returns zero if pid1 and pid2 share the same virtual memory space (CLONE_VM threads, or before exec).
KCMP_FS	Returns zero if pid1 and pid2 share the same filesystem context: root directory, cwd, and umask. Threads created with CLONE_FS share this.
KCMP_FILES	Returns zero if pid1 and pid2 share the same open file descriptor table entirely. Threads created with CLONE_FILES share the table.
KCMP_SIGHAND	Returns zero if pid1 and pid2 share the same signal handler table. Threads created with CLONE_SIGHAND share it.
KCMP_IO	Returns zero if pid1 and pid2 share the same I/O context (block-layer accounting).
KCMP_SYSVSEM	Returns zero if pid1 and pid2 share the same System V semaphore undo list.
KCMP_EPOLL_TFD	idx1 is an epoll fd in pid1, idx2 is a target fd in pid2. Returns zero if the target fd is registered with that epoll instance. Used by CRIU to reconstruct epoll state.

Errors you will actually hit

errno	Meaning
EPERM	Caller lacks ptrace access to both target processes. Requires same uid or CAP_SYS_PTRACE; also blocked by Yama ptrace_scope >= 1 for non-descendant processes.
ESRCH	pid1 or pid2 does not exist or is a zombie.
EINVAL	type is not a valid KCMP_* value.
EBADF	idx1 or idx2 is not an open fd in the respective process (KCMP_FILE / KCMP_EPOLL_TFD).

Example

#define _GNU_SOURCE
#include <linux/kcmp.h>
#include <sys/syscall.h>
#include <unistd.h>

/* are fd 3 in pid1 and fd 7 in pid2 the same open file description? */
int same = (syscall(SYS_kcmp, pid1, pid2, KCMP_FILE, 3, 7) == 0);

Trimmed from the full walkthrough in kcmp(2): comparing kernel objects across processes.

BSD equivalent

No equivalent. FreeBSD and OpenBSD have no kcmp. Closest approach is parsing /proc/<pid>/fd symlinks, which is racy.

The long version, from the console

kcmp(2): comparing kernel objects across processes

kcmp(2) answers a question you cannot answer any other way without a race: do two file descriptors in different processes point to the same underlying kernel object? Not the same path. The same struct file. Linux 3.5, designed for CRIU, available to any process with ptrace access. Here is what it does and when to reach for it.

2026-10-28

Sources

kcmp(2), man7.org

Version numbers are read off the cited man pages, not memory. Spot one that's wrong? Tell Rod.