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;
Subject: Re: [PATCH RT] signal: Prevent double-free of user struct
To:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        Matt Fleming <matt@codeblueprint.co.uk>
Cc:     linux-rt@vger.kernel.org, linux-kernel@vger.kernel.org
References: <20200407095413.30039-1-matt@codeblueprint.co.uk>
 <20200415164628.2dgrj4ghvtev45sy@linutronix.de>
From:   Daniel Wagner <wagi@monom.org>
Message-ID: <f8152ef5-f71d-b8b6-fe9e-6a8678930775@monom.org>
Date:   Wed, 15 Apr 2020 21:05:14 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.6.0
MIME-Version: 1.0
In-Reply-To: <20200415164628.2dgrj4ghvtev45sy@linutronix.de>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Sebastian,

On 15.04.20 18:46, Sebastian Andrzej Siewior wrote:
> On 2020-04-07 10:54:13 [+0100], Matt Fleming wrote:
>> The way user struct reference counting works changed significantly with,
>>
>>    fda31c50292a ("signal: avoid double atomic counter increments for user accounting")
>>
>> Now user structs are only freed once the last pending signal is
>> dequeued. Make sigqueue_free_current() follow this new convention to
>> avoid freeing the user struct multiple times and triggering this
>> warning:
>>
>>   refcount_t: underflow; use-after-free.
>>   WARNING: CPU: 0 PID: 6794 at lib/refcount.c:288 refcount_dec_not_one+0x45/0x50
>>   Call Trace:
>>    refcount_dec_and_lock_irqsave+0x16/0x60
>>    free_uid+0x31/0xa0
>>    ? schedule_hrtimeout_range_clock+0x104/0x110
>>    __dequeue_signal+0x17c/0x190
>>    dequeue_signal+0x5a/0x1b0
>>    do_sigtimedwait+0x208/0x250
>>    __x64_sys_rt_sigtimedwait+0x6f/0xd0
>>    do_syscall_64+0x72/0x200
>>    entry_SYSCALL_64_after_hwframe+0x49/0xbe
> 
> While all this sounds reasonable, may I ask what did you do to trigger
> this?

This can be triggered by running sigwaittest.

  # sigwaittest -t -a -p 98

a few seconds should be enough to get the splat.

> This is v5.6 only, correct?

I've seen this also with a frankstein version of 5.2-rt...v5-4-rt :)

Thanks,
Daniel