[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Mon, 15 Feb 2021 11:56:22 -0500
From: Rich Felker <dalias@...c.org>
To: Dominic Chen <d.c.ddcc@...il.com>
Cc: fweimer@...hat.com, musl@...ts.openwall.com
Subject: Re: Incorrect thread TID caching
On Wed, Feb 03, 2021 at 05:55:18PM -0500, Rich Felker wrote:
> On Wed, Feb 03, 2021 at 05:30:01PM -0500, Dominic Chen wrote:
> >
> > On 2/3/2021 4:01 PM, Rich Felker wrote:
> > >OK, raise should probably just be changed here to work even in vforked
> > >child since it seems plausible someone will use it there. It's not
> > >like saving the syscall actually matters here. But that's independent
> > >of the clone() issue.
> >
> > Sounds good, thanks!
>
> Hm, looking at how to do this now, and if clone is going to behave
> like _Fork, it needs to be able to run code in the child (to restore
> the signals is masked, etc.), which means it needs to wrap the child
> function passed to it. I think this is doable, but it's not entirely
> trivial.
Following up on this now, the code in _Fork is something I really
don't want to duplicate for clone() for risk of forgetting there's a
copy in the latter and letting it bitrot there. I'd rather refactor
things so the same logic can be shared.
In theory we don't need __clone and the above-mentioned callback
wrapper machinery for the non-CLONE_VM case (the only one that's safe
anyway) and can just use __syscall(SYS_clone, ...) from C, and then
the code from _Fork.c can just be generalized to take a function
pointer and arg pointer to call back to instead of directly making the
SYS_fork syscall. However, the fact that SYS_clone has arch-dependent
argument order makes this painful. The extended arguments whose order
vary are *mostly* invalid (you can't set a custom TLS pointer or exit
futex address, at least) so perhaps we could just EINVAL them in
clone().
The other option looks like it's to split the pre/post logic for _Fork
out into separate functions; then clone can call the pre logic in the
parent and the child function passed to __clone can call the post
logic for the child. This is uglier and more costly to _Fork, and I'd
like to avoid it, but not as much as I'd like to avoid duplicating
logic for per-arch argument order. Hopefully we can avoid both..
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
