linux-kernel - Re: [PATCH v2] brd: fix sleeping function called from invalid context in brd_insert

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]

Message-ID: <c76c61ac-9335-b116-31dd-5ecfb32dd7dd@huaweicloud.com>
Date: Tue, 1 Jul 2025 15:38:16 +0800
From: Yu Kuai <yukuai1@...weicloud.com>
To: Jens Axboe <axboe@...nel.dk>, Yu Kuai <yukuai1@...weicloud.com>,
 hch@....de
Cc: penguin-kernel@...ove.SAKURA.ne.jp, linux-block@...r.kernel.org,
 linux-kernel@...r.kernel.org, yi.zhang@...wei.com, yangerkun@...wei.com,
 johnny.chenyi@...wei.com, "yukuai (C)" <yukuai3@...wei.com>
Subject: Re: [PATCH v2] brd: fix sleeping function called from invalid context
 in brd_insert_page()

Hi,

在 2025/07/01 11:00, Jens Axboe 写道:
> On 6/30/25 7:28 PM, Yu Kuai wrote:
>> Hi,
>>
>> ? 2025/06/30 23:28, Jens Axboe ??:
>>> On 6/30/25 9:24 AM, Jens Axboe wrote:
>>>> On 6/30/25 5:28 AM, Yu Kuai wrote:
>>>>> From: Yu Kuai <yukuai3@...wei.com>
>>>>>
>>>>> __xa_cmpxchg() is called with rcu_read_lock(), and it will allocate
>>>>> memory if necessary.
>>>>>
>>>>> Fix the problem by moving rcu_read_lock() after __xa_cmpxchg(), meanwhile,
>>>>> it still should be held before xa_unlock(), prevent returned page to be
>>>>> freed by concurrent discard.
>>>>
>>>> The rcu locking in there is a bit of a mess, imho. What _exactly_ is the
>>>> rcu read side locking protecting? Is it only needed around the lookup
>>>> and insert? We even hold it over the kmap and copy, which seems very
>>>> heavy handed.
>>>
>>> Gah it's holding the page alive too. Can't we just grab a ref to the
>>> page when inserting it, and drop that at free time? It would be a lot
>>> better to have only the lookup be RCU protected, having the full
>>> copies under it seems kind of crazy.
>>
>> In this case, we must grab a ref to the page for each read/write as
>> well, I choose RCU because I think it has less performance overhead than
>> page ref, which is atomic. BTW, I thought copy at most one page is
>> lightweight, if this is not true, I agree page ref is better.
> 
> Right, you'd need to grab a ref. I do think that is (by far) the better
> solution. Yes if you microbenchmark I'm sure the current approach will
> look fine, but it's a heavy section inside an rcu read lock and will
> hold off the grace period.
> 
> So yeah, I do think it'd be a lot better to do proper page references on
> lookup+free, and have just the lookup be behind rcu.
> 

Ok, and just to be sure, since the rcu is introduced before the fixed
tag, do you think it's better to do cleanups after this patch, I prefer
this way, or fix this problem directly by page ref?

Thanks,
Kuai