Date: Fri, 17 Mar 2017 14:44:37 +0200 From: Konstantin Belousov <kostikbel@gmail.com> To: Steven Hartland <killing@multiplay.co.uk> Cc: "K. Macy" <kmacy@freebsd.org>, "freebsd-hackers@freebsd.org" <freebsd-hackers@freebsd.org> Subject: Re: Help needed to identify golang fork / memory corruption issue on FreeBSD Message-ID: <20170317124437.GR16105@kib.kiev.ua> In-Reply-To: <180a601b-5481-bb41-f7fc-67976aabe451@multiplay.co.uk> References: <27e1a828-5cd9-0755-50ca-d7143e7df117@multiplay.co.uk> <20161206125919.GQ54029@kib.kiev.ua> <8b502580-4d2d-1e1f-9e05-61d46d5ac3b1@multiplay.co.uk> <20161206143532.GR54029@kib.kiev.ua> <e160381c-9935-6edf-04a9-1ff78e95d818@multiplay.co.uk> <CAHM0Q_Mg662u9D0KJ9knEWWqi9Ydy38qKDnjLt6XaS0ks%2B9-iw@mail.gmail.com> <18b40a69-4460-faf2-c0ce-7491eca92782@multiplay.co.uk> <20170317082333.GP16105@kib.kiev.ua> <180a601b-5481-bb41-f7fc-67976aabe451@multiplay.co.uk>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Mar 17, 2017 at 11:27:52AM +0000, Steven Hartland wrote: > On 17/03/2017 08:23, Konstantin Belousov wrote: > > On Fri, Mar 17, 2017 at 06:30:49AM +0000, Steven Hartland wrote: > >> Ok I think I've identified the cause. > >> > >> If an alternative signal stack is applied to a non-main thread and that > >> thread calls execve then the signal stack is not cleared. > >> > >> This results in all sorts of badness. > >> > >> Full details, including a small C reproduction case can be found here: > >> https://github.com/golang/go/issues/15658#issuecomment-287276856 > >> > >> So looks like its kernel bug. If anyone has an ideas about that before I > >> look tomorrow that would be appreciated. > > Yes, there is definitely a kernel bug, which should be fixed by the patch > > below. > > > > Still, what I saw when I looked at the issue, is not quite resembling > > potential consequences of the bug. Using wrong memory for signal stack > > would result either in much more significant memory corruption if the > > alt stack range is mapped and used for something unrelated, or in killed > > process on signal delivery, if the range is not mapped. While I saw a > > systematic 'off by 0x10' in some gc structures. > > > > Anyway, patch for the issue you identified: > > > > diff --git a/sys/kern/kern_sig.c b/sys/kern/kern_sig.c > > index 29d5dd4b132..9bf3ba66f5c 100644 > > --- a/sys/kern/kern_sig.c > > +++ b/sys/kern/kern_sig.c > > @@ -976,7 +976,6 @@ execsigs(struct proc *p) > > * and are now ignored by default). > > */ > > PROC_LOCK_ASSERT(p, MA_OWNED); > > - td = FIRST_THREAD_IN_PROC(p); > > ps = p->p_sigacts; > > mtx_lock(&ps->ps_mtx); > > while (SIGNOTEMPTY(ps->ps_sigcatch)) { > > @@ -1007,6 +1006,8 @@ execsigs(struct proc *p) > > * Reset stack state to the user stack. > > * Clear set of signals caught on the signal stack. > > */ > > + td = curthread; > > + MPASS(td->td_proc == p); > > td->td_sigstk.ss_flags = SS_DISABLE; > > td->td_sigstk.ss_size = 0; > > td->td_sigstk.ss_sp = 0; > Thanks Kostik, pretty obvious now looking at :) > > Testing here we've seen all sorts of corruption looking things, mainly > around random signals from SIGILL to SIGSEGV but also random kernel > messages including: > pid 4603 (test): sigreturn copying xfpustate failed > pid 5013 (test): sigreturn xfpusave_len = 0x44d9bb > > I'm currently running a test, but its looking good as the test case > usually crashes in a matter of seconds. > > Would you mind if I committed it? I am capable of committing the patches. > > I'm guessing given its nature this is something we'd want MFC'ed and > Errata's issued for all supported versions? MFC will be done for sure. I am not so sure about EN, this is a routine bugfix. For some reasons 10.3 errata might be indeed the only way to get this for 10.x users, but I do not see why bother re/so with 11.0.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20170317124437.GR16105>