1PIDFD_OPEN(2) Linux Programmer's Manual PIDFD_OPEN(2)
2
6 pidfd_open - obtain a file descriptor that refers to a process
7
9 #include <sys/syscall.h> /* Definition of SYS_* constants */
10 #include <unistd.h>
11 int syscall(SYS_pidfd_open, pid_t pid, unsigned int flags);
13 Note: glibc provides no wrapper for pidfd_open(), necessitating the use
15 of syscall(2).
16
18 The pidfd_open() system call creates a file descriptor that refers to
19 the process whose PID is specified in pid. The file descriptor is re‐
20 turned as the function result; the close-on-exec flag is set on the
21 file descriptor.
22 The flags argument is reserved for future use; currently, this argument
24 must be specified as 0.
25
27 On success, pidfd_open() returns a file descriptor (a nonnegative inte‐
28 ger). On error, -1 is returned and errno is set to indicate the error.
29
31 EINVAL flags is not 0.
32 EINVAL pid is not valid.
34 EMFILE The per-process limit on the number of open file descriptors has
36 been reached (see the description of RLIMIT_NOFILE in getr‐
37 limit(2)).
38 ENFILE The system-wide limit on the total number of open files has been
40 reached.
41 ENODEV The anonymous inode filesystem is not available in this kernel.
43 ENOMEM Insufficient kernel memory was available.
45 ESRCH The process specified by pid does not exist.
47
49 pidfd_open() first appeared in Linux 5.3.
50
52 pidfd_open() is Linux specific.
53
55 The following code sequence can be used to obtain a file descriptor for
56 the child of fork(2):
57 pid = fork();
59 if (pid > 0) { /* If parent */
60 pidfd = pidfd_open(pid, 0);
61 ...
62 }
63 Even if the child has already terminated by the time of the
65 pidfd_open() call, its PID will not have been recycled and the returned
66 file descriptor will refer to the resulting zombie process. Note, how‐
67 ever, that this is guaranteed only if the following conditions hold
68 true:
69 • the disposition of SIGCHLD has not been explicitly set to SIG_IGN
71 (see sigaction(2));
72 • the SA_NOCLDWAIT flag was not specified while establishing a handler
74 for SIGCHLD or while setting the disposition of that signal to
75 SIG_DFL (see sigaction(2)); and
76 • the zombie process was not reaped elsewhere in the program (e.g., ei‐
78 ther by an asynchronously executed signal handler or by wait(2) or
79 similar in another thread).
80 If any of these conditions does not hold, then the child process (along
82 with a PID file descriptor that refers to it) should instead be created
83 using clone(2) with the CLONE_PIDFD flag.
84 Use cases for PID file descriptors
86 A PID file descriptor returned by pidfd_open() (or by clone(2) with the
87 CLONE_PID flag) can be used for the following purposes:
88 • The pidfd_send_signal(2) system call can be used to send a signal to
90 the process referred to by a PID file descriptor.
91 • A PID file descriptor can be monitored using poll(2), select(2), and
93 epoll(7). When the process that it refers to terminates, these in‐
94 terfaces indicate the file descriptor as readable. Note, however,
95 that in the current implementation, nothing can be read from the file
96 descriptor (read(2) on the file descriptor fails with the error EIN‐
97 VAL).
98 • If the PID file descriptor refers to a child of the calling process,
100 then it can be waited on using waitid(2).
101 • The pidfd_getfd(2) system call can be used to obtain a duplicate of a
103 file descriptor of another process referred to by a PID file descrip‐
104 tor.
105 • A PID file descriptor can be used as the argument of setns(2) in or‐
107 der to move into one or more of the same namespaces as the process
108 referred to by the file descriptor.
109 • A PID file descriptor can be used as the argument of process_mad‐
111 vise(2) in order to provide advice on the memory usage patterns of
112 the process referred to by the file descriptor.
113 The pidfd_open() system call is the preferred way of obtaining a PID
115 file descriptor for an already existing process. The alternative is to
116 obtain a file descriptor by opening a /proc/[pid] directory. However,
117 the latter technique is possible only if the proc(5) filesystem is
118 mounted; furthermore, the file descriptor obtained in this way is not
119 pollable and can't be waited on with waitid(2).
120
122 The program below opens a PID file descriptor for the process whose PID
123 is specified as its command-line argument. It then uses poll(2) to
124 monitor the file descriptor for process exit, as indicated by an
125 EPOLLIN event.
126 Program source
128 #define _GNU_SOURCE
130 #include <sys/types.h>
131 #include <sys/syscall.h>
132 #include <unistd.h>
133 #include <poll.h>
134 #include <stdlib.h>
135 #include <stdio.h>
136 #ifndef __NR_pidfd_open
138 #define __NR_pidfd_open 434 /* System call # on most architectures */
139 #endif
140 static int
142 pidfd_open(pid_t pid, unsigned int flags)
143 {
144 return syscall(__NR_pidfd_open, pid, flags);
145 }
146 int
148 main(int argc, char *argv[])
149 {
150 struct pollfd pollfd;
151 int pidfd, ready;
152 if (argc != 2) {
154 fprintf(stderr, "Usage: %s <pid>\n", argv[0]);
155 exit(EXIT_SUCCESS);
156 }
157 pidfd = pidfd_open(atoi(argv[1]), 0);
159 if (pidfd == -1) {
160 perror("pidfd_open");
161 exit(EXIT_FAILURE);
162 }
163 pollfd.fd = pidfd;
165 pollfd.events = POLLIN;
166 ready = poll(&pollfd, 1, -1);
168 if (ready == -1) {
169 perror("poll");
170 exit(EXIT_FAILURE);
171 }
172 printf("Events (%#x): POLLIN is %sset\n", pollfd.revents,
174 (pollfd.revents & POLLIN) ? "" : "not ");
175 close(pidfd);
177 exit(EXIT_SUCCESS);
178 }
179
181 clone(2), kill(2), pidfd_getfd(2), pidfd_send_signal(2), poll(2),
182 process_madvise(2), select(2), setns(2), waitid(2), epoll(7)
183
185 This page is part of release 5.12 of the Linux man-pages project. A
186 description of the project, information about reporting bugs, and the
187 latest version of this page, can be found at
188 https://www.kernel.org/doc/man-pages/.
189Linux 2021-03-22 PIDFD_OPEN(2)