IO-Uring Archive on
 help / color / Atom feed
From: Jens Axboe <>
To: Xiaoguang Wang <>,
Subject: Re: [PATCH] io_uring: refacor file register/unregister/update based on sequence
Date: Mon, 23 Mar 2020 10:49:08 -0600
Message-ID: <> (raw)
In-Reply-To: <>

On 3/23/20 10:45 AM, Xiaoguang Wang wrote:
> hi,
>> On 3/23/20 5:50 AM, Xiaoguang Wang wrote:
>>> While diving into iouring fileset resigster/unregister/update codes,
>>> we found one bug in fileset update codes. Iouring fileset update codes
>>> use a percpu_ref variable to check whether can put previous registered
>>> file, only when the refcnt of the perfcpu_ref variable reachs zero, can
>>> we safely put these files, but this do not work well. If applications
>>> always issue requests continually, this perfcpu_ref will never have an
>>> chance to reach zero, and it'll always be in atomic mode, also will
>>> defeat the gains introduced by fileset register/unresiger/update feature,
>>> which are used to reduce the atomic operation overhead of fput/fget.
>>> To fix this issue, we remove the percpu_ref related codes, and add two new
>>> counter: sq_seq and cq_seq to struct io_ring_ctx:
>>>      sq_seq: the most recent issued requset sequence number, which is
>>>              protected uring_lock.
>>>      cq_seq: the most recent completed request sequence number, which is
>>>              protected completion_lock.
>>> When we update fileset(under uring_lock), we record the current sq_seq,
>>> and when cq_seq is greater or equal to recorded sq_seq, we know we can
>>> put previous registered file safely.
>> Maybe I'm misunderstanding the idea here, but what if you have the
>> following:
>> - sq_seq 200, cq_seq 100
>> We have 100 inflight, and an unregister request comes in. I then
>> issue 100 nops, which complete. cq_seq is now 200, but none of the
>> original requests that used the file have completed.
>> What am I missing?
> No, you're right. I had thought requests will be completed in the order
> they are issued, thanks for pointing this.
> As for not using per percpu_ref per registered file, I also worry about
> the memory consume, because the max allowed registered files are 32768.

Yeah, I think we have to be a bit creative here with the solution...
Please continue to think about it, would be great to have a better
solution for this!

Jens Axboe

      reply index

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-03-23 11:50 Xiaoguang Wang
2020-03-23 12:02 ` Xiaoguang Wang
2020-03-23 16:05 ` Jens Axboe
2020-03-23 16:45   ` Xiaoguang Wang
2020-03-23 16:49     ` Jens Axboe [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \ \ \ \ \ \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

IO-Uring Archive on

Archives are clonable:
	git clone --mirror io-uring/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 io-uring io-uring/ \
	public-inbox-index io-uring

Example config snippet for mirrors

Newsgroup available over NNTP:

AGPL code for this site: git clone