Help needed to identify golang fork / memory corruption issue on FreeBSD
Konstantin Belousov
kostikbel at gmail.com
Tue Mar 28 07:59:05 UTC 2017
On Tue, Mar 28, 2017 at 12:32:37AM +0100, Steven Hartland wrote:
> On 27/03/2017 17:49, Konstantin Belousov wrote:
> > On Mon, Mar 27, 2017 at 05:33:49PM +0100, Steven Hartland wrote:
> >> On 27/03/2017 17:18, Konstantin Belousov wrote:
> >>> On Mon, Mar 27, 2017 at 12:47:11PM +0100, Steven Hartland wrote:
> >>>> OK now the similar but unrelated issue with signal stacks is solved I've
> >>>> moved back to the initial issue.
> >>>>
> >>>> I've made some progress with a reproduction case as detailed here:
> >>>> https://github.com/golang/go/issues/15658#issuecomment-288747812
> >>>>
> >>>> In short it seems that having a running child, while the parent runs GC,
> >>>> is some how responsible for memory corruption in the parent.
> >>>>
> >>>> The reason I believe this is if I run the same GC in the parent after
> >>>> the child exits instead of while its running, I've been unable to
> >>>> reproduce the issue.
> >>>>
> >>>> As the memory segments are COW then the issue might be in VM subsystem.
> >>> Well, it might be, but it is a strange corruption mode to believe.
> >> Indeed, but would you agree the evidence seems to indicate that this may
> >> be the case, as otherwise I would have expected that running the GC
> >> after the child process has exited would have zero impact on the issue.
> >>>> In order to confirm / deny this I was wondering if there was a way to
> >>>> force a full copy of all segments for the child instead of using the COW
> >>>> optimisation.
> >>> No, there is no. By design, copying only occurs on faults, when VM
> >>> detects that the map entry needs copying. Doing the actual copy at fork
> >>> time would require writing a lot of new code.
> >> I noticed in vm_map_copy_entry the following:
> >> /*
> >> * We don't want to make writeable wired pages
> >> copy-on-write.
> >> * Immediately copy these pages into the new map by
> >> simulating
> >> * page faults. The new pages are pageable.
> >> */
> >> vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
> >> fork_charge);
> >>
> >> I wondered if I could use vm_fault_copy_entry to force the copy on fork?
> > No, the vm_fault_copy_entry() only works with wired entries, e.g. it cannot
> > page in not yet touched page, and the result is also wired.
> >
> >>> Does go have FreeBSD/i386 port ? If yes, is the issue reproducable there ?
> >> Yes it does, I don't currently have i386 machine to test with, I'm
> >> assuming testing i386 on amd64 kernel, would likely not have any effect.
> > Only if the bug is in kernel and not in the go runtime. I am still not
> > convinced that the kernel is the culprit.
> >
> >>> Another blind experiment to try is to comment out call to
> >>> vm_object_collapse() in sys/vm/vm_map.c:vm_map_copy_entry() and see if
> >>> it changes anything.
> >> I'll do that shortly.
> >>> What could be quite interesting is to look at the parent and possibly
> >>> child address map after the error occured, using procstat -v. At
> >>> least for parent, this should be relatively easy to set up, just make
> >>> go runtime spin or pause on panic, instead of exiting, and then use
> >>> procstat.
> >>
> Here's both parent and child after a failure in the parent, which I
> obtained by putting the child in a nanosleep loop and only after
> successful GC call I send SIGTERM the child and reap it.
>
> procstat -v 53832 61121
> PID START END PRT RES PRES REF SHD FLAG
> TP PATH
> 53832 0x400000 0x70e000 r-x 308 601 5 1 CN-- vn
> /root/golang/src/test5/test5
> 53832 0x70e000 0x951000 r-- 261 601 5 1 CN-- vn
> /root/golang/src/test5/test5
> 53832 0x951000 0x988000 rw- 31 0 1 0 C--- vn
> /root/golang/src/test5/test5
> 53832 0x988000 0x9ab000 rw- 18 0 1 0 C--- df
> 53832 0x800951000 0x800b51000 rw- 41 0 1 0 C--- df
> 53832 0x800b51000 0x800c21000 rw- 26 0 1 0 C--- df
> 53832 0x800c21000 0x800c71000 rw- 18 0 1 0 C--- df
> 53832 0x800c71000 0x800cb1000 rw- 1 0 1 0 C--- df
> 53832 0x800cb1000 0x800cf1000 rw- 2 0 1 0 C--- df
> 53832 0x800cf1000 0x800d71000 rw- 3 0 1 0 C--- df
> 53832 0x800d71000 0x800db1000 rw- 1 0 1 0 C--- df
> 53832 0x800db1000 0x800e71000 rw- 3 0 1 0 C--- df
> 53832 0x800e71000 0x800eb1000 rw- 1 1 1 0 ---- df
> 53832 0xc000000000 0xc000001000 rw- 1 1 2 0 CN-- df
> 53832 0xc41fff0000 0xc41fff8000 rw- 3 3 2 0 CN-- df
> 53832 0xc41fff8000 0xc420200000 rw- 251 0 1 0 C--- df
> 53832 0x7ffffffdf000 0x7ffffffff000 rwx 2 0 1 0 C--D df
> 53832 0x7ffffffff000 0x800000000000 r-x 1 1 28 0 ---- ph
> 61121 0x400000 0x70e000 r-x 308 601 5 1 CN-- vn
> /root/golang/src/test5/test5
> 61121 0x70e000 0x951000 r-- 261 601 5 1 CN-- vn
> /root/golang/src/test5/test5
> 61121 0x951000 0x988000 rw- 31 0 2 1 CN-- vn
> /root/golang/src/test5/test5
> 61121 0x988000 0x9ab000 rw- 18 18 2 1 CN-- df
> 61121 0x800951000 0x800b51000 rw- 41 41 2 1 CN-- df
> 61121 0x800b51000 0x800c21000 rw- 26 26 2 1 CN-- df
> 61121 0x800c21000 0x800c71000 rw- 18 18 2 1 CN-- df
> 61121 0x800c71000 0x800cb1000 rw- 1 1 2 1 CN-- df
> 61121 0x800cb1000 0x800cf1000 rw- 2 2 2 1 CN-- df
> 61121 0x800cf1000 0x800d71000 rw- 3 3 2 1 CN-- df
> 61121 0x800d71000 0x800db1000 rw- 1 1 2 1 CN-- df
> 61121 0x800db1000 0x800e71000 rw- 3 3 2 1 CN-- df
> 61121 0xc000000000 0xc000001000 rw- 1 1 2 0 CN-- df
> 61121 0xc41fff0000 0xc41fff8000 rw- 3 3 2 0 CN-- df
> 61121 0xc41fff8000 0xc420200000 rw- 251 0 1 0 C--- df
> 61121 0x7ffffffdf000 0x7ffffffff000 rwx 2 2 2 1 CN-D df
> 61121 0x7ffffffff000 0x800000000000 r-x 1 1 28 0 ---- ph
>
> Should the parent have lost the COW flag to the region starting at
> 0x800e71000?
Note that the region is absent in the child, so most likely this is a new
mmaping occured in the parent after fork. Again, what is the address
where an invalid value was detected ?
More information about the freebsd-hackers
mailing list