1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
|
/*
Copyright (C) 2020- The University of Notre Dame
This software is distributed under the GNU General Public License.
See the file COPYING for details.
*/
#ifndef PPOLL_COMPAT_H
#define PPOLL_COMPAT_H
#include <poll.h>
/** Wait for some event on file descriptors.
*
* While Linux provides ppoll() natively, other platforms like
* like OSX don't, so this compatibility shim is necessary.
* The interface differs somewhat: stoptime is given as an
* int (for a 5 second timeout, pass time(NULL) + 5). In addition,
* the signal set is assumed to be the empty set, i.e. all signals
* are unblocked for the duration of the call. To break out on
* receiving a signal, be sure it's not set to SIG_IGN. The normal
* way to set up signals to work with ppoll is
* 1) Block the signal.
* 2) Install a signal handler (may be a no-op function).
* 3) Call ppoll_compat(), which will unblock the signal only
* during the call.
*
* If ppoll() is detected at compile time, it will be used.
* Otherwise, its behavior is emulated using regular poll().
* Note that this fallback suffers from the race condition
* described in select(2). It is therefore important not to rely
* on signals for correctness (it is OK to use them as a fast path,
* and in most cases the race condition probably won't appear).
*/
int ppoll_compat(struct pollfd fds[], nfds_t nfds, int stoptime);
/** Set up signal handling to ensure that SIGCHLD will interrupt ppoll_compat()
*/
void ppoll_compat_set_up_sigchld(void);
#endif
|