On 4/2/2022 23:00, Laurent Bercot wrote:
So, SIGTERM does nothing because posix_spawn() is lying to s6-svscan, pretending to have succeeded when it doesn't know it yet (and is going to fail), and goading it into not doing anything. And it's lying because something in qemu is messing with the semantics of CLONE_VFORK.
Amazing. Thank you, Laurent! It probably would have taken me a while to turn my suspicions on such a crucial library function.
For any future s6 users who happen to stumble upon this thread in an archive search, the upstream qemu bug appears to be here:
https://gitlab.com/qemu-project/qemu/-/issues/140#note_567553059
I'm sorry, but I don't have an easy solution for you
No problem at all. Given the proximate cause, I can avoid the fault without too much trouble.
Sometimes, the little devil on my shoulder yelling 'that is not right' makes it hard to ignore even the most practically insignificant bugs. (Not that posix_spawn() is, in any way, insignificant, but we do not use qemu in production.) Your comprehensive reply help put my mind at ease.
Thanks again -- for this, and for s6.
