Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Reply-To: efremov@linux.com
To:     Steven Rostedt <rostedt@goodmis.org>
Cc:     Gaurav Kohli <gkohli@codeaurora.org>, linux-kernel@vger.kernel.org,
        linux-arm-msm@vger.kernel.org, stable@vger.kernel.org,
        Julia Lawall <julia.lawall@inria.fr>
References: <1601976833-24377-1-git-send-email-gkohli@codeaurora.org>
 <f06efd7b-c7b5-85c9-1a0e-6bb865111ede@linux.com>
 <20210121140951.2a554a5e@gandalf.local.home>
From:   Denis Efremov <efremov@linux.com>
Subject: Re: [PATCH v1] trace: Fix race in trace_open and buffer resize call
Message-ID: <021b1b38-47ce-bc8b-3867-99160cc85523@linux.com>
Date:   Thu, 21 Jan 2021 23:15:22 +0300
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.6.0
MIME-Version: 1.0
In-Reply-To: <20210121140951.2a554a5e@gandalf.local.home>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 1/21/21 10:09 PM, Steven Rostedt wrote:
> On Thu, 21 Jan 2021 17:30:40 +0300
> Denis Efremov <efremov@linux.com> wrote:
> 
>> Hi,
>>
>> This patch (CVE-2020-27825) was tagged with
>> Fixes: b23d7a5f4a07a ("ring-buffer: speed up buffer resets by avoiding synchronize_rcu for each CPU")
>>
>> I'm not an expert here but it seems like b23d7a5f4a07a only refactored
>> ring_buffer_reset_cpu() by introducing reset_disabled_cpu_buffer() without
>> significant changes. Hence, mutex_lock(&buffer->mutex)/mutex_unlock(&buffer->mutex)
>> can be backported further than b23d7a5f4a07a~ and to all LTS kernels. Is
>> b23d7a5f4a07a the actual cause of the bug?
>>
> 
> Ug, that looks to be a mistake. Looking back at the thread about this:
> 
>   https://lore.kernel.org/linux-arm-msm/20200915141304.41fa7c30@gandalf.local.home/

I see from the link that it was planned to backport the patch to LTS kernels:

> Actually we are seeing issue in older kernel like 4.19/4.14/5.4 and there below patch was not 
> present in stable branches:
> Commit b23d7a5f4a07 ("ring-buffer: speed up buffer resets by avoiding synchronize_rcu for each CPU")

The point is that it's not backported yet. Maybe because of Fixes tag. I've discovered
this while trying to formalize CVE-2020-27825 bug in cvehound
https://github.com/evdenis/cvehound/blob/master/cvehound/cve/CVE-2020-27825.cocci

I think that the backport to the 4.4+ should be something like:

diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 547a3a5ac57b..2171b377bbc1 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -4295,6 +4295,8 @@ void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
 	if (!cpumask_test_cpu(cpu, buffer->cpumask))
 		return;
 
+	mutex_lock(&buffer->mutex);
+
 	atomic_inc(&buffer->resize_disabled);
 	atomic_inc(&cpu_buffer->record_disabled);
 
@@ -4317,6 +4319,8 @@ void ring_buffer_reset_cpu(struct ring_buffer *buffer, int cpu)
 
 	atomic_dec(&cpu_buffer->record_disabled);
 	atomic_dec(&buffer->resize_disabled);
+
+	mutex_unlock(&buffer->mutex);
 }
 EXPORT_SYMBOL_GPL(ring_buffer_reset_cpu);
 

Thanks,
Denis