зарегистрировать обработчики вилок (register fork handlers)
Пролог (Prolog)
This manual page is part of the POSIX Programmer's Manual. The
Linux implementation of this interface may differ (consult the
corresponding Linux manual page for details of Linux behavior),
or the interface may not be implemented on Linux.
Имя (Name)
pthread_atfork — register fork handlers
Синопсис (Synopsis)
#include <pthread.h>
int pthread_atfork(void (*prepare)(void), void (*parent)(void),
void (*child)(void));
Описание (Description)
The pthread_atfork() function shall declare fork handlers to be
called before and after fork(), in the context of the thread that
called fork(). The prepare fork handler shall be called before
fork() processing commences. The parent fork handle shall be
called after fork() processing completes in the parent process.
The child fork handler shall be called after fork() processing
completes in the child process. If no handling is desired at one
or more of these three points, the corresponding fork handler
address(es) may be set to NULL.
If a fork() call in a multi-threaded process leads to a child
fork handler calling any function that is not async-signal-safe,
the behavior is undefined.
The order of calls to pthread_atfork() is significant. The parent
and child fork handlers shall be called in the order in which
they were established by calls to pthread_atfork(). The prepare
fork handlers shall be called in the opposite order.
Возвращаемое значение (Return value)
Upon successful completion, pthread_atfork() shall return a value
of zero; otherwise, an error number shall be returned to indicate
the error.
Ошибки (Error)
The pthread_atfork() function shall fail if:
ENOMEM
Insufficient table space exists to record the fork handler
addresses.
The pthread_atfork() function shall not return an error code of
[EINTR]
.
The following sections are informative.
Примеры (Examples)
None.
Использование в приложениях (Application usage)
The original usage pattern envisaged for pthread_atfork() was for
the prepare fork handler to lock mutexes and other locks, and for
the parent and child handlers to unlock them. However, since all
of the relevant unlocking functions, except sem_post(), are not
async-signal-safe, this usage results in undefined behavior in
the child process unless the only such unlocking function it
calls is sem_post().
Обоснование (Rationale)
There are at least two serious problems with the semantics of
fork() in a multi-threaded program. One problem has to do with
state (for example, memory) covered by mutexes. Consider the case
where one thread has a mutex locked and the state covered by that
mutex is inconsistent while another thread calls fork(). In the
child, the mutex is in the locked state (locked by a nonexistent
thread and thus can never be unlocked). Having the child simply
reinitialize the mutex is unsatisfactory since this approach does
not resolve the question about how to correct or otherwise deal
with the inconsistent state in the child.
It is suggested that programs that use fork() call an exec
function very soon afterwards in the child process, thus
resetting all states. In the meantime, only a short list of
async-signal-safe library routines are promised to be available.
Unfortunately, this solution does not address the needs of multi-
threaded libraries. Application programs may not be aware that a
multi-threaded library is in use, and they feel free to call any
number of library routines between the fork() and exec calls,
just as they always have. Indeed, they may be extant single-
threaded programs and cannot, therefore, be expected to obey new
restrictions imposed by the threads library.
On the other hand, the multi-threaded library needs a way to
protect its internal state during fork() in case it is re-entered
later in the child process. The problem arises especially in
multi-threaded I/O libraries, which are almost sure to be invoked
between the fork() and exec calls to effect I/O redirection. The
solution may require locking mutex variables during fork(), or it
may entail simply resetting the state in the child after the
fork() processing completes.
The pthread_atfork() function was intended to provide multi-
threaded libraries with a means to protect themselves from
innocent application programs that call fork(), and to provide
multi-threaded application programs with a standard mechanism for
protecting themselves from fork() calls in a library routine or
the application itself.
The expected usage was that the prepare handler would acquire all
mutex locks and the other two fork handlers would release them.
For example, an application could have supplied a prepare routine
that acquires the necessary mutexes the library maintains and
supplied child and parent routines that release those mutexes,
thus ensuring that the child would have got a consistent snapshot
of the state of the library (and that no mutexes would have been
left stranded). This is good in theory, but in reality not
practical. Each and every mutex and lock in the process must be
located and locked. Every component of a program including third-
party components must participate and they must agree who is
responsible for which mutex or lock. This is especially
problematic for mutexes and locks in dynamically allocated
memory. All mutexes and locks internal to the implementation must
be locked, too. This possibly delays the thread calling fork()
for a long time or even indefinitely since uses of these
synchronization objects may not be under control of the
application. A final problem to mention here is the problem of
locking streams. At least the streams under control of the system
(like stdin, stdout, stderr) must be protected by locking the
stream with flockfile(). But the application itself could have
done that, possibly in the same thread calling fork(). In this
case, the process will deadlock.
Alternatively, some libraries might have been able to supply just
a child routine that reinitializes the mutexes in the library and
all associated states to some known value (for example, what it
was when the image was originally executed). This approach is not
possible, though, because implementations are allowed to fail
*_init() and *_destroy() calls for mutexes and locks if the mutex
or lock is still locked. In this case, the child routine is not
able to reinitialize the mutexes and locks.
When fork() is called, only the calling thread is duplicated in
the child process. Synchronization variables remain in the same
state in the child as they were in the parent at the time fork()
was called. Thus, for example, mutex locks may be held by threads
that no longer exist in the child process, and any associated
states may be inconsistent. The intention was that the parent
process could have avoided this by explicit code that acquires
and releases locks critical to the child via pthread_atfork().
In addition, any critical threads would have needed to be
recreated and reinitialized to the proper state in the child
(also via pthread_atfork()).
A higher-level package may acquire locks on its own data
structures before invoking lower-level packages. Under this
scenario, the order specified for fork handler calls allows a
simple rule of initialization for avoiding package deadlock: a
package initializes all packages on which it depends before it
calls the pthread_atfork() function for itself.
As explained, there is no suitable solution for functionality
which requires non-atomic operations to be protected through
mutexes and locks. This is why the POSIX.1 standard since the
1996 release requires that the child process after fork() in a
multi-threaded process only calls async-signal-safe interfaces.
Будущие направления (Future directions)
The pthread_atfork() function may be formally deprecated (for
example, by shading it OB) in a future version of this standard.
Смотри также (See also)
atexit(3p), exec(1p), fork(3p)
The Base Definitions volume of POSIX.1‐2017, pthread.h(0p),
sys_types.h(0p)