On 2015-10-01, Michael McConville <mmcco...@sccs.swarthmore.edu> wrote:
> Atanas Vladimirov wrote:
>> Snapshot from sep 30 bgpd didn't startup:
>> Oct 1 08:32:28 ns /bsd: bgpd(28055): syscall 105
>> Oct 1 08:32:28 ns bgpd[29697]: handle_pollfd: poll fd: Undefined error: 0
>> Oct 1 08:32:28 ns bgpd[29697]: RDE: Lost connection to SE
>> Oct 1 08:32:28 ns bgpd[27739]: handle_pollfd: poll fd: No such file or
>> directory
>> Oct 1 08:32:28 ns bgpd[29697]: handle_pollfd: poll fd: Undefined error: 0
>> Oct 1 08:32:28 ns bgpd[29697]: RDE: Lost connection to SE control
>> Oct 1 08:32:28 ns bgpd[27739]: main: Lost connection to SE
>> Oct 1 08:32:28 ns bgpd[27739]: Lost child: session engine terminated;
>> signal 9
>
> This looks like a result of the new tame(2)ing. Below are the tame calls
> that were just added to bgpd, according to Theo's diff.
>
> Syscall 105 is setsockopt(2). Both "unix" and "inet" allow it. However,
> the man page notes that "inet" restricts setsockopt significantly.
> Because this error looks like it's happening within a setsockopt call,
> maybe that's the issue. Changing "inet" to "unix" could potentially fix
> it, as could refactoring the bgpd code.
I think this should be ok under a newer kernel. But if not, please run
bgpd under ktrace ("ktrace -i bgpd" or similar), then run kdump and
include the last page or so of output.
Same for any other crashes which show up in dmesg like 'program(12345): syscall
XXX'.
