On Sat, Aug 23, 2014 at 07:52:44PM +0200, Jilles Tjoelker wrote: > On Fri, Aug 22, 2014 at 06:31:39PM +0300, Konstantin Belousov wrote: > > On Fri, Aug 22, 2014 at 03:43:53PM +0200, Jilles Tjoelker wrote: > > > This is good and necessary for SA_SIGINFO (because of the type of the > > > SIG_DFL and SIG_IGN constants, and because POSIX says so in the > > > description of SA_RESETHAND in the sigaction() page). However, there > > > seems no reason to clear the other flags, which have no effect when the > > > disposition is SIG_DFL or SIG_IGN but have historically always been > > > preserved. (Slight exception: if kernel code erroneously loops on > > > ERESTART, it will eat CPU time iff SA_RESTART is set, independent of the > > > signal's disposition.) > > Well, I already committed the patch with several bugs fixed comparing > > with what was mailed, before your feedback arrived. > > > Do you consider it is important enough to revert the resetting of other > > flags ? In particular, your note about the traditional historic > > behaviour makes me wonder. > > I consider it important enough. Clearing the other flags is not > POSIX-compliant and might break applications. For example, I can imagine > an application modifying a struct sigaction with sa_handler == SIG_DFL > from a sigaction() call. This feels somewhat strange to me. E.g., I can easily imagine an implementation which relies on some code executing in the process user context for default action on some signal. Having the flags, like SA_ONSTACK or SA_NODEFER to influence the handler is weird. Such implementation is not unix, but I think it is quite possible that cygwin or interix do core dumping in userspace.
> > > I do not see why SA_SIGINFO is so special that it must be reset, > > while other flags are not. The absence of the cases where the > > default/ignored disposition is affected by the flags seems rather > > arbitrary. > > The difference is that SA_SIGINFO changes the disposition field from > sa_handler to sa_sigaction, and it is not unambiguously clear how > SIG_DFL and SIG_IGN are represented in sa_sigaction. Note that > sa_handler and sa_sigaction may or may not share storage, and > implementations may or may not support (void (*)(int, siginfo_t *, void > *))SIG_DFL. > > For example, when I wrote system() using posix_spawn() in > https://lists.freebsd.org/pipermail/freebsd-hackers/2012-July/040065.html > I needed to know whether SIGINT and SIGQUIT were ignored or not. I wrote > > ] if ((intact.sa_flags & SA_SIGINFO) != 0 || > ] intact.sa_handler != SIG_IGN) > ] (void)sigaddset(&defmask, SIGINT); > ] if ((quitact.sa_flags & SA_SIGINFO) != 0 || > ] quitact.sa_handler != SIG_IGN) > ] (void)sigaddset(&defmask, SIGQUIT); > > in the assumption that there is always an actual handler if SA_SIGINFO > is set. I did not really like this code and eventually system() was > changed to use vfork() directly instead of posix_spawn(), but I think it > is correct. If the implementation provides separate storage for sa_handler and sa_sigaction, then isn't it more correct to assume that sa_handler is NULL when sa_sigaction is valid ? In other words, simple sa.sa_handler != SIG_IGN test would be more reasonable. I see that the SUSv4 is worded in a way that SA_SIGINFO always accompanies sa_sigaction. Are there any implementations which separate the storage for sa_handler and sa_sigaction ? > > Note that this means that it is sometimes necessary to install a handler > function that will never be called. Per POSIX, SA_SIGINFO must be set > for sigwaitinfo() to be guaranteed to return siginfo_t data, and the > only way to do this is to specify a handler function, even though it > will never be called because the signal is masked. (A never-called > handler function also needs to be specified when using sigwait-like > functions with signals that default to ignore.) > > The other flags do not affect the representation of the disposition, and > can therefore remain set without problem. Anyway, below is the patch with reverts the behaviour WRT flags other than SA_SIGINFO. I like the compactness of sigact_flag_test() calls, so I kept the function, despite it is only used in non-trivial ways for SA_SIGINFO flag test in kern_sigaction(). diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c index 8810bf3..f73c801 100644 --- a/sys/kern/kern_sig.c +++ b/sys/kern/kern_sig.c @@ -625,9 +625,14 @@ static bool sigact_flag_test(struct sigaction *act, int flag) { - return ((act->sa_flags & flag) != 0 && - (__sighandler_t *)act->sa_sigaction != SIG_IGN && - (__sighandler_t *)act->sa_sigaction != SIG_DFL); + /* + * SA_SIGINFO is reset when signal disposition is set to + * ignore or default. Other flags are kept according to user + * settings. + */ + return ((act->sa_flags & flag) != 0 && (flag != SA_SIGINFO || + ((__sighandler_t *)act->sa_sigaction != SIG_IGN && + (__sighandler_t *)act->sa_sigaction != SIG_DFL))); } /* @@ -916,7 +921,6 @@ siginit(p) for (i = 1; i <= NSIG; i++) { if (sigprop(i) & SA_IGNORE && i != SIGCONT) { SIGADDSET(ps->ps_sigignore, i); - SIGADDSET(ps->ps_sigintr, i); } } mtx_unlock(&ps->ps_mtx); @@ -936,10 +940,6 @@ sigdflt(struct sigacts *ps, int sig) SIGADDSET(ps->ps_sigignore, sig); ps->ps_sigact[_SIG_IDX(sig)] = SIG_DFL; SIGDELSET(ps->ps_siginfo, sig); - SIGADDSET(ps->ps_sigintr, sig); - SIGDELSET(ps->ps_sigonstack, sig); - SIGDELSET(ps->ps_sigreset, sig); - SIGDELSET(ps->ps_signodefer, sig); } /*
pgpG14F8sBAjY.pgp
Description: PGP signature