MIPS has 128 signals, the highest of which has the number 128. The
following command causes get_signal_to_deliver() to pass this signal
number straight through to do_group_exit() as the exit code:
strace sleep 10 & sleep 1 && kill -128 `pidof sleep`
However do_group_exit() checks for the core dump bit (0x80) in the exit
code which matches in this particular case and the kernel panics:
BUG_ON(exit_code & 0x80); /* core dumps don't get here */
This is worked around by changing get_signal_to_deliver() to pass
min(info->si_signo, 127) instead of info->si_signo, so that this highest
of signal numbers get rounded down to 127. This makes the exit code
technically incorrect, but it's better than killing the whole kernel.
Signed-off-by: James Hogan <[email protected]>
Cc: Ralf Baechle <[email protected]>
Cc: Al Viro <[email protected]>
Cc: Andrew Morton <[email protected]>
Cc: Oleg Nesterov <[email protected]>
Cc: Kees Cook <[email protected]>
---
This is based on v3.10-rc3.
It's a little hacky, but aside from reducing the number of signals to
127 to avoid this case (which isn't backwards compatible) I'm not sure
what else can be done. Any comments?
kernel/signal.c | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/kernel/signal.c b/kernel/signal.c
index 113411b..69bc00f 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2366,8 +2366,12 @@ relock:
/*
* Death signals, no core dump.
+ *
+ * MIPS has a signal number 128 which clashes with the core dump
+ * bit. If this was the signal we still want to report a valid
+ * exit code, so round it down to 127.
*/
- do_group_exit(info->si_signo);
+ do_group_exit(min(info->si_signo, 127));
/* NOTREACHED */
}
spin_unlock_irq(&sighand->siglock);
--
1.8.1.2
On 05/29/2013 10:01 AM, James Hogan wrote:
> MIPS has 128 signals, the highest of which has the number 128. The
I wonder if we should change the ABI and reduce the number of signals to
127 instead of this patch.
David Daney
> following command causes get_signal_to_deliver() to pass this signal
> number straight through to do_group_exit() as the exit code:
>
> strace sleep 10 & sleep 1 && kill -128 `pidof sleep`
>
> However do_group_exit() checks for the core dump bit (0x80) in the exit
> code which matches in this particular case and the kernel panics:
>
> BUG_ON(exit_code & 0x80); /* core dumps don't get here */
>
> This is worked around by changing get_signal_to_deliver() to pass
> min(info->si_signo, 127) instead of info->si_signo, so that this highest
> of signal numbers get rounded down to 127. This makes the exit code
> technically incorrect, but it's better than killing the whole kernel.
>
> Signed-off-by: James Hogan <[email protected]>
> Cc: Ralf Baechle <[email protected]>
> Cc: Al Viro <[email protected]>
> Cc: Andrew Morton <[email protected]>
> Cc: Oleg Nesterov <[email protected]>
> Cc: Kees Cook <[email protected]>
> ---
>
> This is based on v3.10-rc3.
>
> It's a little hacky, but aside from reducing the number of signals to
> 127 to avoid this case (which isn't backwards compatible) I'm not sure
> what else can be done. Any comments?
>
> kernel/signal.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/signal.c b/kernel/signal.c
> index 113411b..69bc00f 100644
> --- a/kernel/signal.c
> +++ b/kernel/signal.c
> @@ -2366,8 +2366,12 @@ relock:
>
> /*
> * Death signals, no core dump.
> + *
> + * MIPS has a signal number 128 which clashes with the core dump
> + * bit. If this was the signal we still want to report a valid
> + * exit code, so round it down to 127.
> */
> - do_group_exit(info->si_signo);
> + do_group_exit(min(info->si_signo, 127));
> /* NOTREACHED */
> }
> spin_unlock_irq(&sighand->siglock);
>
On 05/29, David Daney wrote:
>
> On 05/29/2013 10:01 AM, James Hogan wrote:
>> MIPS has 128 signals, the highest of which has the number 128. The
>
> I wonder if we should change the ABI and reduce the number of signals to
> 127 instead of this patch.
Same thoughts...
>> @@ -2366,8 +2366,12 @@ relock:
>>
>> /*
>> * Death signals, no core dump.
>> + *
>> + * MIPS has a signal number 128 which clashes with the core dump
>> + * bit. If this was the signal we still want to report a valid
>> + * exit code, so round it down to 127.
>> */
>> - do_group_exit(info->si_signo);
>> + do_group_exit(min(info->si_signo, 127));
This avoids BUG_ON() but obviously fools WIFSIGNALED(), doesn't look
very nice.
Oleg.
On 29 May 2013 18:36, Oleg Nesterov <[email protected]> wrote:
> On 05/29, David Daney wrote:
>>
>> On 05/29/2013 10:01 AM, James Hogan wrote:
>>> MIPS has 128 signals, the highest of which has the number 128. The
>>
>> I wonder if we should change the ABI and reduce the number of signals to
>> 127 instead of this patch.
>
> Same thoughts...
I'll give it a try. I wouldn't have thought it'd break anything, but
you never know. glibc (incorrectly) sets [__]SIGRTMAX to 127 already.
On the other hand uClibc sets it to 128, so anything built against
uClibc that uses signals SIGRTMAX-n (where n may be 0) or uses an
excessive number of rt signals starting from SIGRTMIN (sounds
unlikely) could well need an updated uClibc (or a full rebuild if it's
crazy enough to use __SIGRTMAX).
>>> @@ -2366,8 +2366,12 @@ relock:
>>>
>>> /*
>>> * Death signals, no core dump.
>>> + *
>>> + * MIPS has a signal number 128 which clashes with the core dump
>>> + * bit. If this was the signal we still want to report a valid
>>> + * exit code, so round it down to 127.
>>> */
>>> - do_group_exit(info->si_signo);
>>> + do_group_exit(min(info->si_signo, 127));
>
> This avoids BUG_ON() but obviously fools WIFSIGNALED(), doesn't look
> very nice.
Agreed.
Cheers
James
On Wednesday 29 May 2013 23:56, James Hogan wrote:
> On 29 May 2013 18:36, Oleg Nesterov <[email protected]> wrote:
> > On 05/29, David Daney wrote:
> >>
> >> On 05/29/2013 10:01 AM, James Hogan wrote:
> >>> MIPS has 128 signals, the highest of which has the number 128. The
> >>
> >> I wonder if we should change the ABI and reduce the number of signals to
> >> 127 instead of this patch.
> >
> > Same thoughts...
>
> I'll give it a try. I wouldn't have thought it'd break anything, but
> you never know. glibc (incorrectly) sets [__]SIGRTMAX to 127 already.
> On the other hand uClibc sets it to 128, so anything built against
> uClibc that uses signals SIGRTMAX-n (where n may be 0) or uses an
> excessive number of rt signals starting from SIGRTMIN (sounds
> unlikely) could well need an updated uClibc (or a full rebuild if it's
> crazy enough to use __SIGRTMAX).
Fixed in uclibc git: _NSIG is 128, __SIGRTMAX is 127
(_NSIG in libc is not the same as in kernel, but +1).
While at it, added extensive comment why it is so.