Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 11 Mar 2012 16:42:37 -0700
From: Linus Torvalds <>
To: Djalal Harouni <>
Cc: Oleg Nesterov <>,,, 
	Andrew Morton <>, Al Viro <>, 
	Alexey Dobriyan <>, "Eric W. Biederman" <>, 
	Vasiliy Kulikov <>, Kees Cook <>, 
	Solar Designer <>, WANG Cong <>, 
	James Morris <>,,, Alan Cox <>, 
	Greg KH <>, Ingo Molnar <>, 
	Stephen Wilson <>, "Jason A. Donenfeld" <>
Subject: Re: [PATCH 1/9] exec: add a global execve counter

On Sun, Mar 11, 2012 at 4:32 PM, Djalal Harouni <> wrote:
>> Just increment the mm_count for the thing, and hold a reference to it,
>> and now you're all done.
> Please Linus have you checked the:
> [PATCH 9/9] proc: improve and clean up /proc/<pid>/mem protection
> That keeping the mm struct wont work, since it will eat memory and the
> OOM-killer will kill some innocent processes, and the abuse can only be
> catched by the VFS.

That's the point. I made the mistake of using mm_users initially, but
ysing mm_count - which is what I said to use (and what Oleg fixed
things to in commit 6d08f2c71397) should *not* have that problem. It
just keeps the 'struct mm_struct' itself around.

> What's your opinion on it ?

What's the advantage? You replace it with *another* allocation, and a
64-bit thing that is much less useful.

The size of the patch also speaks for itself:

 fs/proc/base.c |   99 ++++++++++++++++++++++++++++++++++++++++++++------------

and it's more complex and uses more memory on average (the refcount
thing is *free* for usual cases).

I do agree that it would be nicer if mm_struct was a bit smaller, but
at the same time, I really don't see the advantage of replacing it
with another allocation entirely that makes the code just more


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.