linux-kernel - uprobes misses breakpoint insertion into VM

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-next>] [day] [month] [year] [list]

Message-ID: <670124481.11073.1521146881571.JavaMail.zimbra@efficios.com>
Date:   Thu, 15 Mar 2018 16:48:01 -0400 (EDT)
From:   Mathieu Desnoyers <mathieu.desnoyers@...icios.com>
To:     Oleg Nesterov <oleg@...hat.com>,
        Erica Bugden <ebugden@...icios.com>
Cc:     Srikar Dronamraju <srikar@...ux.vnet.ibm.com>,
        rostedt <rostedt@...dmis.org>, Ingo Molnar <mingo@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        linux-kernel <linux-kernel@...r.kernel.org>
Subject: uprobes misses breakpoint insertion into VM_WRITE mappings

Hi,

Erica has been working on extending test-cases for uprobes, and found
something unexpected:

Since commit e40cfce626a5 "uprobes: Restrict valid_vma(false) to skip VM_SHARED vmas"
uprobes does not insert breakpoints into mappings mprotect'd as writeable.

This issue can be reproduced by compiling a library without PIC (not using GOT),
and then concurrently:

A) Load the library (dynamic loader mprotect the code as writeable to do
   the relocations, and then mprotect as executable),

B) Enable a uprobe through perf.

(it is a race window between the two mprotect syscalls)

It appears that the following restriction in valid_vma() is responsible
for this behavior:

        if (is_register)
                flags |= VM_WRITE;

I don't figure a clear explanation for this flag based on the function
comment nor the commit changelog. Any idea on whether this is really
needed ?

Note that on uprobes unregister, it allows removing a breakpoint event
on a writeable mapping, so there is clearly a discrepancy between the
level of paranoia associated with registration and unregistration.

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com