[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAGXu5jLAqki8oE7AF9BEwLxKzpTOOPsT1imjdSAG_U44uGXJdQ@mail.gmail.com>
Date: Fri, 16 Mar 2018 23:25:43 -0700
From: Kees Cook <keescook@...omium.org>
To: Andrew Morton <akpm@...ux-foundation.org>
Cc: LKML <linux-kernel@...r.kernel.org>,
Segher Boessenkool <segher@...nel.crashing.org>,
Thomas Gleixner <tglx@...utronix.de>,
Kernel Hardening <kernel-hardening@...ts.openwall.com>
Subject: Re: [PATCH v2] rslib: Remove VLAs by setting upper bound on nroots
On Fri, Mar 16, 2018 at 3:59 PM, Andrew Morton
<akpm@...ux-foundation.org> wrote:
> On Thu, 15 Mar 2018 15:59:19 -0700 Kees Cook <keescook@...omium.org> wrote:
>
>> Avoid stack VLAs[1] by always allocating the upper bound of stack space
>> needed. The existing users of rslib appear to max out at 24 roots[2],
>> so use that as the upper bound until we have a reason to change it.
>>
>> Alternative considered: make init_rs() a true caller-instance and
>> pre-allocate the workspaces. This would possibly need locking and
>> a refactoring of the returned structure.
>>
>> Using kmalloc in this path doesn't look great, especially since at
>> least one caller (pstore) is sensitive to allocations during rslib
>> usage (it expects to run it during an Oops, for example).
>
> Oh.
>
> Could we allocate the storage during init_rs(), attach it to `struct
> rs_control'?
No, because they're modified during decode, and struct rs_control is
shared between users. :(
Doing those changes is possible, but it requires a rather extensive
analysis of callers, etc.
Hence, the 24 ultimately.
-Kees
--
Kees Cook
Pixel Security
Powered by blists - more mailing lists