[PATCH 4/5] linux: Optimize posix_spawn spurious sigaction calls

[Adhemerval Zanella]

On 02/09/2019 10:14, Florian Weimer wrote:
> * Adhemerval Zanella:
> 
>> The problem is in fact false negatives, where posix_spawn will get a mask 
>> *without* the bit set, but with a set signal disposition.
> 
> Hmm.  Right.  Incidentally, the Go routine should be fine with that:
> 
> | // When using cgo, call the C library for sigaction, so that we call into
> | // any sanitizer interceptors. This supports using the memory
> | // sanitizer with Go programs. The memory sanitizer only applies to
> | // C/C++ code; this permits that code to see the Go runtime's existing signal
> | // handlers when registering new signal handlers for the process.
> | 
> | //go:cgo_import_static x_cgo_sigaction
> | //go:linkname x_cgo_sigaction x_cgo_sigaction
> | //go:linkname _cgo_sigaction _cgo_sigaction
> | var x_cgo_sigaction byte
> | var _cgo_sigaction = &x_cgo_sigaction
> 
> libjsig also keeps calling to glibc.
> 
> Is there anything else we should check?

No idea, my take on that is once you start to calling syscall directly
where libbc provide a wrapper you are in your own. We had a similar
discussing with clone usage by some container applications and their
expectation regarding libc internal state afterwards. 

> 
>> In fact I think due the syscall, even relaxed operations would work
>> (since the syscall acts a strong memory barrier).
> 
> Only as a signal fence, not a thread fence.  Some architectures can even
> keep cache inconsistency across fork system calls.
> 
> I find it a bit counter-intuitive that calling sigaction or signal
> directly without the glibc wrappers could lead to data corruption, even
> when done for standard signals such as SIGINT.  But that's what's going
> to happen with this change, unfortunately.

What is counter-intuitive imho is to rely on libc to keep its internal
consistency by bypassing it. This might be even worse if glibc start to
wrapper the signal handler as a way to implement BZ#19702, for instance.

One thing we may do it to make it clean on manual that an application is
*not* expect to call sigaction using syscall().

> 
>>>>> I wonder if we can get kernel support for this in the new clone system
>>>>> call with more flags.  Then we don't have to complicate the sigaction
>>>>> implementation.
>>>>
>>>> Maybe a CLONE_RESET_SIGNALS where the cloned process sets its signal
>>>> disposition to default SIG_IGN/SIG_DFL values may help us here.  However
>>>> afaik clone now is out of space on 'flags' for newer ones (it already
>>>> defines 24 flags plus it reserve 8 bits for signal to be sent at process
>>>> exit) and it would take time to use this feature on glibc.
>>>
>>> Christian Brauner has been working on fixing this.
>>
>> Which strategy he is proposing? Even with proper kernel support, it would
>> take time to enable glibc to use it.
> 
> Lots of flag arguments, with the reset of the arguments located
> indirectly via a pointer argument.
> 
> For a pure optimization, I think it's not too bad to require kernel
> backports of system calls.
> 
> Thanks,
> Florian
>