Re: [PATCH v3] mm: process_mrelease: introduce PROCESS_MRELEASE_REAP_KILL flag

[Linus Torvalds]

On Fri, 15 May 2026 at 16:45, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> Yes, yes, we'd need some sequence counter for when the mm changes due
> to execve, but *that* should be absolutely trivial.

I guess we already  have it in the form of "self_exec_id". So the
tuple <mm,self_exec_id> should basically guarantee it's the same mm as
it was at open time.

Except it's not updated at quite the right time. If it was done inside
the task lock when task->mm was actually set in exec_mmap(), it would
work as-is.

This isn't exactly what that 'self_exec_id' was designed for, but it
*is* a sequence number for the mm of a task. So if a 'mm' pointer
value gets reused, you can tell that it's not the same mm as it used
to be by looking at the exec_id.

And I think the only reason for it being done outside the task lock is
that the current use is all synchronous to the task itself, so locking
or location simply didn't matter.

But there's some mqueue signal handling thing that looks like it's
actually violating that situation and would race with execve().

            Linus