linux-kernel - Re: [PATCH] proc_sysctl: clamp sizes using table->maxlen

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Mon, 15 Feb 2021 19:49:31 -0500
From:   "Alex Xu (Hello71)" <alex_y_xu@...oo.ca>
To:     linux-kernel@...r.kernel.org, Luis Chamberlain <mcgrof@...nel.org>,
        Kees Cook <keescook@...omium.org>,
        Christoph Hellwig <hch@....de>, Andrey Ignatov <rdna@...com>,
        Al Viro <viro@...iv.linux.org.uk>,
        Iurii Zaikin <yzaikin@...gle.com>,
        Alexey Dobriyan <adobriyan@...il.com>,
        linux-fsdevel@...r.kernel.org
Subject: Re: [PATCH] proc_sysctl: clamp sizes using table->maxlen

Excerpts from Alex Xu (Hello71)'s message of February 15, 2021 9:53 am:
> This issue was discussed at [0] and following, and the solution was to
> clamp the size at KMALLOC_MAX_LEN. However, KMALLOC_MAX_LEN is a maximum
> allocation, and may be difficult to allocate in low memory conditions.
> 
> Since maxlen is already exposed, we can allocate approximately the right
> amount directly, fixing up those drivers which set a bogus maxlen. These
> drivers were located based on those which had copy_x_user replaced in
> 32927393dc1c, on the basis that other drivers either use builtin proc_*
> handlers, or do not access the data pointer. The latter is OK because
> maxlen only needs to be an upper limit.
> 
> [0] https://lore.kernel.org/lkml/1fc7ce08-26a7-59ff-e580-4e6c22554752@oracle.com/
> 
> Fixes: 32927393dc1c ("sysctl: pass kernel pointers to ->proc_handler")
> Signed-off-by: Alex Xu (Hello71) <alex_y_xu@...oo.ca>

Yeah, no, this doesn't work. A bunch of functions call proc_* but don't 
set maxlen, and it's annoying to check this statically. Also causes 
weird failures elsewhere. May need to think of a better solution here 
(kvzalloc?).