netdev - Re: [PATCH v2 3/5] io_uring/zcrx: share an ifq between rings

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: <a8c453b4-acc8-4ec7-a064-3c3e470c5669@kernel.dk>
Date: Sun, 26 Oct 2025 07:43:45 -0600
From: Jens Axboe <axboe@...nel.dk>
To: David Wei <dw@...idwei.uk>, io-uring@...r.kernel.org,
 netdev@...r.kernel.org
Cc: Pavel Begunkov <asml.silence@...il.com>
Subject: Re: [PATCH v2 3/5] io_uring/zcrx: share an ifq between rings

On 10/26/25 7:16 AM, Jens Axboe wrote:
> On 10/25/25 10:12 PM, David Wei wrote:
>> Sorry I missed this during the splitting. Will include in v3.
>>
>>>
>>>> +    ifq->proxy = src_ifq;
>>>
>>> For this, since the ifq is shared and reference counted, why don't they
>>> just point at the same memory here? Would avoid having this ->proxy
>>> thing and just skipping to that in other spots where the actual
>>> io_zcrx_ifq is required?
>>>
>>
>> I wanted a way to separate src and dst rings, while also decrementing
>> refcounts once and only once. I used separate ifq objects to do this,
>> but having learnt about xarray marks, I think I can use that instead.
> 
> I'm confused, why do you even need that? You already have
> ifq->proxy which is just a "link" to the shared queue, why aren't both
> rings just using the same ifq structure? You already increment the
> refcount when you add proxy, why can't the new ring just store the same
> ifq?

And just to follow up - if this isn't directly feasible, then I think
the ifq bits need to be refactored a bit first to facilitate having it
be shared. Having a dummy ifq that only uses ->proxy is not super clean
to look at. From a quick look, ->ctx is the odd one out there, that part
is obviously not shared between the two rings.

-- 
Jens Axboe