linux-kernel - Re: [GIT PULL] tracing: Prevent trace_marker being bigger than unsigned short

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, 4 Mar 2024 19:43:43 -0500
From: Steven Rostedt <rostedt@...dmis.org>
To: Linus Torvalds <torvalds@...ux-foundation.org>
Cc: LKML <linux-kernel@...r.kernel.org>, Masami Hiramatsu
 <mhiramat@...nel.org>, Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
 Sachin Sant <sachinp@...ux.ibm.com>
Subject: Re: [GIT PULL] tracing: Prevent trace_marker being bigger than
 unsigned short

On Mon, 4 Mar 2024 16:17:15 -0800
Linus Torvalds <torvalds@...ux-foundation.org> wrote:

> On Mon, 4 Mar 2024 at 15:50, Steven Rostedt <rostedt@...dmis.org> wrote:
> >
> > But this still isn't fixing anything. It's just adding a limit.  
> 
> Limiting things to a common maximum size is a good thing. The kernel
> limits much  more important things for very good reasons.
> 
> The kernel really shouldn't have big strings. EVER.  And it literally
> shows in our kernel infrastructure. It showed in that vsnprintf
> precision thing. It shows in our implementation choices, where we tend
> to have simplistic implementations because doing things a byte at a
> time is simple and cheap when the strings are limited in size (and we
> don't want fancy and can't use vector state anyway).
> 
> If something as core as a pathname can be limited to 4kB, then
> something as unimportant as a trace string had better be limited too.
> Because we simply DO NOT WANT to have to deal with longer strings in
> the kernel.
> 

So I made three patches that do basically what you want. And as a bonus,
it's not really an arbitrary limit but based on trace_seq size.

The first patch will be removing the precision check, as that's not needed.

The second patch is to remove the dependency between trace_seq and
PAGE_SIZE, as its size really can just be 8K for all architectures. Which
has the side effect of limiting the size of trace_marker, as its size is
limited by the trace_seq size.

Finally, because the trace_seq defines the max output that a trace_event
can write (for all its fields), the extra data of a print event could
possibly overflow that, which will cause the event not to print, and just
an "OVERFLOW" output would show in the trace buffer. So I used the
TRACE_SEQ_SIZE / 2 as the max size that trace_marker can read, which
happens to be 4K.

-- Steve