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 v2 0/4] open/accept directly into io_uring fixed file
 table
To:     Josh Triplett <josh@joshtriplett.org>
Cc:     Pavel Begunkov <asml.silence@gmail.com>, io-uring@vger.kernel.org,
        "David S . Miller" <davem@davemloft.net>,
        Jakub Kicinski <kuba@kernel.org>, linux-kernel@vger.kernel.org,
        netdev@vger.kernel.org, Stefan Metzmacher <metze@samba.org>
References: <cover.1628871893.git.asml.silence@gmail.com>
 <YRbBYCn29B+kgZcy@localhost> <bcb6f253-41d6-6e0f-5b4b-ea1e02a105bc@gmail.com>
 <5cf40313-d151-9d10-3ebd-967eb2f53b1f@kernel.dk> <YRiNGTL2Dp/7vNzt@localhost>
From:   Jens Axboe <axboe@kernel.dk>
Message-ID: <a2bd1600-5649-c4be-d2a9-79c89bae774a@kernel.dk>
Date:   Sun, 15 Aug 2021 09:05:55 -0600
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.10.0
MIME-Version: 1.0
In-Reply-To: <YRiNGTL2Dp/7vNzt@localhost>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 8/14/21 9:42 PM, Josh Triplett wrote:
> On Sat, Aug 14, 2021 at 05:03:44PM -0600, Jens Axboe wrote:
>> What's the plan in terms of limiting the amount of direct descriptors
>> (for lack of a better word)? That seems like an important aspect that
>> should get sorted out upfront.
> [...]
>> Maybe we have a way to size the direct table, which will consume entries
>> from the same pool that the regular file table does? That would then
>> work both ways, and could potentially just be done dynamically similarly
>> to how we expand the regular file table when we exceed its current size.
> 
> I think we'll want a way to size the direct table regardless, so that
> it's pre-allocated and doesn't need to be resized when an index is used.

But how do you size it then? I can see this being used into the hundreds
of thousands of fds easily, and right now the table is just an array
(though split into segments, avoiding huge allocs).

> Then, we could do one of two equally easy things, depending on what
> policy we want to set:
> 
> - Deduct the full size of the fixed-file table from the allowed number
>   of files the process can have open. So, if RLIMIT_NOFILE is 1048576,
>   and you pre-allocate 1000000 entries in the fixed-file table, you can
>   have no more than 48576 file descriptors open. Stricter, but
>   potentially problematic: a program *might* expect that it can
>   dup2(some_fd, nofile - 1) successfully.
> 
> - Use RLIMIT_NOFILE as the maximum size of the fixed-file table. There's
>   precedent for this: we already use RLIMIT_NOFILE as the maximum number
>   of file descriptors you can have in flight over UNIX sockets.
> 
> I personally would favor the latter; it seems simple and
> straightforward.

I strongly prefer the latter too, and hopefully that's palatable since
the default limits are quite low anyway. And, as you say, it already is
done for inflight fds as well.

-- 
Jens Axboe