1FORK(2) Linux Programmer's Manual FORK(2)
2
6 fork - create a child process
7
9 #include <unistd.h>
10 pid_t fork(void);
12
14 fork() creates a new process by duplicating the calling process. The
15 new process is referred to as the child process. The calling process
16 is referred to as the parent process.
17 The child process and the parent process run in separate memory spaces.
19 At the time of fork() both memory spaces have the same content. Memory
20 writes, file mappings (mmap(2)), and unmappings (munmap(2)) performed
21 by one of the processes do not affect the other.
22 The child process is an exact duplicate of the parent process except
24 for the following points:
25 * The child has its own unique process ID, and this PID does not match
27 the ID of any existing process group (setpgid(2)) or session.
28 * The child's parent process ID is the same as the parent's process
30 ID.
31 * The child does not inherit its parent's memory locks (mlock(2),
33 mlockall(2)).
34 * Process resource utilizations (getrusage(2)) and CPU time counters
36 (times(2)) are reset to zero in the child.
37 * The child's set of pending signals is initially empty (sigpend‐
39 ing(2)).
40 * The child does not inherit semaphore adjustments from its parent
42 (semop(2)).
43 * The child does not inherit process-associated record locks from its
45 parent (fcntl(2)). (On the other hand, it does inherit fcntl(2)
46 open file description locks and flock(2) locks from its parent.)
47 * The child does not inherit timers from its parent (setitimer(2),
49 alarm(2), timer_create(2)).
50 * The child does not inherit outstanding asynchronous I/O operations
52 from its parent (aio_read(3), aio_write(3)), nor does it inherit any
53 asynchronous I/O contexts from its parent (see io_setup(2)).
54 The process attributes in the preceding list are all specified in
56 POSIX.1. The parent and child also differ with respect to the follow‐
57 ing Linux-specific process attributes:
58 * The child does not inherit directory change notifications (dnotify)
60 from its parent (see the description of F_NOTIFY in fcntl(2)).
61 * The prctl(2) PR_SET_PDEATHSIG setting is reset so that the child
63 does not receive a signal when its parent terminates.
64 * The default timer slack value is set to the parent's current timer
66 slack value. See the description of PR_SET_TIMERSLACK in prctl(2).
67 * Memory mappings that have been marked with the madvise(2) MADV_DONT‐
69 FORK flag are not inherited across a fork().
70 * Memory in address ranges that have been marked with the madvise(2)
72 MADV_WIPEONFORK flag is zeroed in the child after a fork(). (The
73 MADV_WIPEONFORK setting remains in place for those address ranges in
74 the child.)
75 * The termination signal of the child is always SIGCHLD (see
77 clone(2)).
78 * The port access permission bits set by ioperm(2) are not inherited
80 by the child; the child must turn on any bits that it requires using
81 ioperm(2).
82 Note the following further points:
84 * The child process is created with a single thread—the one that
86 called fork(). The entire virtual address space of the parent is
87 replicated in the child, including the states of mutexes, condition
88 variables, and other pthreads objects; the use of pthread_atfork(3)
89 may be helpful for dealing with problems that this can cause.
90 * After a fork() in a multithreaded program, the child can safely call
92 only async-signal-safe functions (see signal-safety(7)) until such
93 time as it calls execve(2).
94 * The child inherits copies of the parent's set of open file descrip‐
96 tors. Each file descriptor in the child refers to the same open
97 file description (see open(2)) as the corresponding file descriptor
98 in the parent. This means that the two file descriptors share open
99 file status flags, file offset, and signal-driven I/O attributes
100 (see the description of F_SETOWN and F_SETSIG in fcntl(2)).
101 * The child inherits copies of the parent's set of open message queue
103 descriptors (see mq_overview(7)). Each file descriptor in the child
104 refers to the same open message queue description as the correspond‐
105 ing file descriptor in the parent. This means that the two file de‐
106 scriptors share the same flags (mq_flags).
107 * The child inherits copies of the parent's set of open directory
109 streams (see opendir(3)). POSIX.1 says that the corresponding di‐
110 rectory streams in the parent and child may share the directory
111 stream positioning; on Linux/glibc they do not.
112
114 On success, the PID of the child process is returned in the parent, and
115 0 is returned in the child. On failure, -1 is returned in the parent,
116 no child process is created, and errno is set to indicate the error.
117
119 EAGAIN A system-imposed limit on the number of threads was encountered.
120 There are a number of limits that may trigger this error:
121 * the RLIMIT_NPROC soft resource limit (set via setrlimit(2)),
123 which limits the number of processes and threads for a real
124 user ID, was reached;
125 * the kernel's system-wide limit on the number of processes and
127 threads, /proc/sys/kernel/threads-max, was reached (see
128 proc(5));
129 * the maximum number of PIDs, /proc/sys/kernel/pid_max, was
131 reached (see proc(5)); or
132 * the PID limit (pids.max) imposed by the cgroup "process num‐
134 ber" (PIDs) controller was reached.
135 EAGAIN The caller is operating under the SCHED_DEADLINE scheduling pol‐
137 icy and does not have the reset-on-fork flag set. See sched(7).
138 ENOMEM fork() failed to allocate the necessary kernel structures be‐
140 cause memory is tight.
141 ENOMEM An attempt was made to create a child process in a PID namespace
143 whose "init" process has terminated. See pid_namespaces(7).
144 ENOSYS fork() is not supported on this platform (for example, hardware
146 without a Memory-Management Unit).
147 ERESTARTNOINTR (since Linux 2.6.17)
149 System call was interrupted by a signal and will be restarted.
150 (This can be seen only during a trace.)
151
153 POSIX.1-2001, POSIX.1-2008, SVr4, 4.3BSD.
154
156 Under Linux, fork() is implemented using copy-on-write pages, so the
157 only penalty that it incurs is the time and memory required to dupli‐
158 cate the parent's page tables, and to create a unique task structure
159 for the child.
160 C library/kernel differences
162 Since version 2.3.3, rather than invoking the kernel's fork() system
163 call, the glibc fork() wrapper that is provided as part of the NPTL
164 threading implementation invokes clone(2) with flags that provide the
165 same effect as the traditional system call. (A call to fork() is
166 equivalent to a call to clone(2) specifying flags as just SIGCHLD.)
167 The glibc wrapper invokes any fork handlers that have been established
168 using pthread_atfork(3).
169
171 See pipe(2) and wait(2).
172
174 clone(2), execve(2), exit(2), setrlimit(2), unshare(2), vfork(2),
175 wait(2), daemon(3), pthread_atfork(3), capabilities(7), credentials(7)
176
178 This page is part of release 5.13 of the Linux man-pages project. A
179 description of the project, information about reporting bugs, and the
180 latest version of this page, can be found at
181 https://www.kernel.org/doc/man-pages/.
182Linux 2021-03-22 FORK(2)