1PTHREAD_ATFORK(3P) POSIX Programmer's Manual PTHREAD_ATFORK(3P)
2
6 This manual page is part of the POSIX Programmer's Manual. The Linux
7 implementation of this interface may differ (consult the corresponding
8 Linux manual page for details of Linux behavior), or the interface may
9 not be implemented on Linux.
10
13 pthread_atfork — register fork handlers
14
16 #include <pthread.h>
17 int pthread_atfork(void (*prepare)(void), void (*parent)(void),
19 void (*child)(void));
20
22 The pthread_atfork() function shall declare fork handlers to be called
23 before and after fork(), in the context of the thread that called
24 fork(). The prepare fork handler shall be called before fork() pro‐
25 cessing commences. The parent fork handle shall be called after fork()
26 processing completes in the parent process. The child fork handler
27 shall be called after fork() processing completes in the child process.
28 If no handling is desired at one or more of these three points, the
29 corresponding fork handler address(es) may be set to NULL.
30 The order of calls to pthread_atfork() is significant. The parent and
32 child fork handlers shall be called in the order in which they were
33 established by calls to pthread_atfork(). The prepare fork handlers
34 shall be called in the opposite order.
35
37 Upon successful completion, pthread_atfork() shall return a value of
38 zero; otherwise, an error number shall be returned to indicate the
39 error.
40
42 The pthread_atfork() function shall fail if:
43 ENOMEM Insufficient table space exists to record the fork handler
45 addresses.
46 The pthread_atfork() function shall not return an error code of
48 [EINTR].
49 The following sections are informative.
51
53 None.
54
56 None.
57
59 There are at least two serious problems with the semantics of fork() in
60 a multi-threaded program. One problem has to do with state (for exam‐
61 ple, memory) covered by mutexes. Consider the case where one thread has
62 a mutex locked and the state covered by that mutex is inconsistent
63 while another thread calls fork(). In the child, the mutex is in the
64 locked state (locked by a nonexistent thread and thus can never be
65 unlocked). Having the child simply reinitialize the mutex is unsatis‐
66 factory since this approach does not resolve the question about how to
67 correct or otherwise deal with the inconsistent state in the child.
68 It is suggested that programs that use fork() call an exec function
70 very soon afterwards in the child process, thus resetting all states.
71 In the meantime, only a short list of async-signal-safe library rou‐
72 tines are promised to be available.
73 Unfortunately, this solution does not address the needs of multi-
75 threaded libraries. Application programs may not be aware that a multi-
76 threaded library is in use, and they feel free to call any number of
77 library routines between the fork() and exec calls, just as they always
78 have. Indeed, they may be extant single-threaded programs and cannot,
79 therefore, be expected to obey new restrictions imposed by the threads
80 library.
81 On the other hand, the multi-threaded library needs a way to protect
83 its internal state during fork() in case it is re-entered later in the
84 child process. The problem arises especially in multi-threaded I/O
85 libraries, which are almost sure to be invoked between the fork() and
86 exec calls to effect I/O redirection. The solution may require locking
87 mutex variables during fork(), or it may entail simply resetting the
88 state in the child after the fork() processing completes.
89 The pthread_atfork() function was intended to provide multi-threaded
91 libraries with a means to protect themselves from innocent application
92 programs that call fork(), and to provide multi-threaded application
93 programs with a standard mechanism for protecting themselves from
94 fork() calls in a library routine or the application itself.
95 The expected usage was that the prepare handler would acquire all mutex
97 locks and the other two fork handlers would release them.
98 For example, an application could have supplied a prepare routine that
100 acquires the necessary mutexes the library maintains and supplied child
101 and parent routines that release those mutexes, thus ensuring that the
102 child would have got a consistent snapshot of the state of the library
103 (and that no mutexes would have been left stranded). This is good in
104 theory, but in reality not practical. Each and every mutex and lock in
105 the process must be located and locked. Every component of a program
106 including third-party components must participate and they must agree
107 who is responsible for which mutex or lock. This is especially problem‐
108 atic for mutexes and locks in dynamically allocated memory. All mutexes
109 and locks internal to the implementation must be locked, too. This pos‐
110 sibly delays the thread calling fork() for a long time or even indefi‐
111 nitely since uses of these synchronization objects may not be under
112 control of the application. A final problem to mention here is the
113 problem of locking streams. At least the streams under control of the
114 system (like stdin, stdout, stderr) must be protected by locking the
115 stream with flockfile(). But the application itself could have done
116 that, possibly in the same thread calling fork(). In this case, the
117 process will deadlock.
118 Alternatively, some libraries might have been able to supply just a
120 child routine that reinitializes the mutexes in the library and all
121 associated states to some known value (for example, what it was when
122 the image was originally executed). This approach is not possible,
123 though, because implementations are allowed to fail *_init() and
124 *_destroy() calls for mutexes and locks if the mutex or lock is still
125 locked. In this case, the child routine is not able to reinitialize the
126 mutexes and locks.
127 When fork() is called, only the calling thread is duplicated in the
129 child process. Synchronization variables remain in the same state in
130 the child as they were in the parent at the time fork() was called.
131 Thus, for example, mutex locks may be held by threads that no longer
132 exist in the child process, and any associated states may be inconsis‐
133 tent. The intention was that the parent process could have avoided this
134 by explicit code that acquires and releases locks critical to the child
135 via pthread_atfork(). In addition, any critical threads would have
136 needed to be recreated and reinitialized to the proper state in the
137 child (also via pthread_atfork()).
138 A higher-level package may acquire locks on its own data structures
140 before invoking lower-level packages. Under this scenario, the order
141 specified for fork handler calls allows a simple rule of initialization
142 for avoiding package deadlock: a package initializes all packages on
143 which it depends before it calls the pthread_atfork() function for
144 itself.
145 As explained, there is no suitable solution for functionality which
147 requires non-atomic operations to be protected through mutexes and
148 locks. This is why the POSIX.1 standard since the 1996 release requires
149 that the child process after fork() in a multi-threaded process only
150 calls async-signal-safe interfaces.
151
153 None.
154
156 atexit(), exec, fork()
157 The Base Definitions volume of POSIX.1‐2008, <pthread.h>, <sys_types.h>
159
161 Portions of this text are reprinted and reproduced in electronic form
162 from IEEE Std 1003.1, 2013 Edition, Standard for Information Technology
163 -- Portable Operating System Interface (POSIX), The Open Group Base
164 Specifications Issue 7, Copyright (C) 2013 by the Institute of Electri‐
165 cal and Electronics Engineers, Inc and The Open Group. (This is
166 POSIX.1-2008 with the 2013 Technical Corrigendum 1 applied.) In the
167 event of any discrepancy between this version and the original IEEE and
168 The Open Group Standard, the original IEEE and The Open Group Standard
169 is the referee document. The original Standard can be obtained online
170 at http://www.unix.org/online.html .
171 Any typographical or formatting errors that appear in this page are
173 most likely to have been introduced during the conversion of the source
174 files to man page format. To report such errors, see https://www.ker‐
175 nel.org/doc/man-pages/reporting_bugs.html .
176IEEE/The Open Group 2013 PTHREAD_ATFORK(3P)