Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
From:   "Eric W. Biederman" <ebiederm@xmission.com>
To:     Miaohe Lin <linmiaohe@huawei.com>
Cc:     <akpm@linux-foundation.org>, <hughd@google.com>,
        <linux-mm@kvack.org>, <linux-kernel@vger.kernel.org>,
        Alexey Gladkov <legion@kernel.org>
References: <20220314064039.62972-1-linmiaohe@huawei.com>
        <87h78036hl.fsf@email.froward.int.ebiederm.org>
        <82cf5aa8-a721-3ff3-7b09-54a66da0d506@huawei.com>
        <87lexbyslf.fsf@email.froward.int.ebiederm.org>
        <4803adf1-ba98-badc-6820-0948871b0742@huawei.com>
Date:   Wed, 16 Mar 2022 09:11:29 -0500
In-Reply-To: <4803adf1-ba98-badc-6820-0948871b0742@huawei.com> (Miaohe Lin's
        message of "Wed, 16 Mar 2022 14:55:15 +0800")
Message-ID: <87sfri3s32.fsf@email.froward.int.ebiederm.org>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/27.1 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain
Subject: Re: [PATCH v2] mm/mlock: fix potential imbalanced rlimit ucounts
 adjustment
Precedence: bulk

Miaohe Lin <linmiaohe@huawei.com> writes:

> On 2022/3/16 2:32, Eric W. Biederman wrote:
>> Miaohe Lin <linmiaohe@huawei.com> writes:
>> 
>>> On 2022/3/14 23:21, Eric W. Biederman wrote:
>>>> Miaohe Lin <linmiaohe@huawei.com> writes:
>>>>
>>>>> user_shm_lock forgets to set allowed to 0 when get_ucounts fails. So
>>>>> the later user_shm_unlock might do the extra dec_rlimit_ucounts. Fix
>>>>> this by resetting allowed to 0.
>>>>
>>>> This fix looks correct.  But the ability for people to follow and read
>>>> the code seems questionable.  I saw in v1 of this patch Hugh originally
>>>> misread the logic.
>>>>
>>>> Could we instead change the code to leave lock_limit at ULONG_MAX aka
>>>> RLIM_INFINITY, leave initialized to 0, and not even need a special case
>>>> of RLIM_INFINITY as nothing can be greater that ULONG_MAX?
>>>>
>>>
>>> Many thanks for your advice. This looks good but it seems this results in different
>>> behavior: When (memlock == LONG_MAX) && !capable(CAP_IPC_LOCK), we would fail now
>>> while it will always success without this change. We should avoid this difference.
>>> Or am I miss something? Maybe the origin patch is more suitable and
>>> simple?
>> 
>> Interesting.  I think that is an unintended and necessary bug fix.
>> 
>> When memlock == LONG_MAX that means inc_rlimit_ucounts failed.
>> 
>> It either failed because at another level the limit was exceeded or
>> because the counter wrapped.  In either case it is not appropriate to
>> succeed if inc_rlimit_ucounts detects a failure.
>> 
>> Which is a long way of saying I think we really want the simplification
>> because it found and fixed another bug as well.
>> 
>> Without the simplification I don't think I will be confident the code is
>> correct.
>
> Agree with you. This is a potential bug and you just catch it with the
> code simplification. :)
>
> Am I supposed to do this altogether or will you do this simplification part?
> Many thanks.

If you can that would be great, and you can have the credit.

Otherwise I will make my proposed changes into a proper patch.  At this
point we just need to dot the i's and cross the t's and get this fix in.

Eric

>>>> Something like this?
>>>>
>>>> diff --git a/mm/mlock.c b/mm/mlock.c
>>>> index 8f584eddd305..e7eabf5193ab 100644
>>>> --- a/mm/mlock.c
>>>> +++ b/mm/mlock.c
>>>> @@ -827,13 +827,12 @@ int user_shm_lock(size_t size, struct ucounts *ucounts)
>>>>  
>>>>  	locked = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
>>>>  	lock_limit = rlimit(RLIMIT_MEMLOCK);
>>>> -	if (lock_limit == RLIM_INFINITY)
>>>> -		allowed = 1;
>>>> -	lock_limit >>= PAGE_SHIFT;
>>>> +	if (lock_limit != RLIM_INFINITY)
>>>> +		lock_limit >>= PAGE_SHIFT;
>>>>  	spin_lock(&shmlock_user_lock);
>>>>  	memlock = inc_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
>>>>  
>>>> -	if (!allowed && (memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
>>>> +	if ((memlock == LONG_MAX || memlock > lock_limit) && !capable(CAP_IPC_LOCK)) {
>>>>  		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
>>>>  		goto out;
>>>>  	}
>>>>
>>>>>
>>>>> Fixes: d7c9e99aee48 ("Reimplement RLIMIT_MEMLOCK on top of ucounts")
>>>>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
>>>>> Acked-by: Hugh Dickins <hughd@google.com>
>>>>> ---
>>>>> v1->v2:
>>>>>   correct Fixes tag and collect Acked-by tag
>>>>>   Thanks Hugh for review!
>>>>> ---
>>>>>  mm/mlock.c | 1 +
>>>>>  1 file changed, 1 insertion(+)
>>>>>
>>>>> diff --git a/mm/mlock.c b/mm/mlock.c
>>>>> index 29372c0eebe5..efd2dd2943de 100644
>>>>> --- a/mm/mlock.c
>>>>> +++ b/mm/mlock.c
>>>>> @@ -733,6 +733,7 @@ int user_shm_lock(size_t size, struct ucounts *ucounts)
>>>>>  	}
>>>>>  	if (!get_ucounts(ucounts)) {
>>>>>  		dec_rlimit_ucounts(ucounts, UCOUNT_RLIMIT_MEMLOCK, locked);
>>>>> +		allowed = 0;
>>>>>  		goto out;
>>>>>  	}
>>>>>  	allowed = 1;
>>>>
>>>> Eric
>>>> .
>>>>
>> .
>>