LKML Archive on
help / color / mirror / Atom feed
From: Dexuan Cui <>
To: Sunil Muthuswamy <>,
	KY Srinivasan <>,
	Haiyang Zhang <>,
	Stephen Hemminger <>,
	Sasha Levin <>,
	"David S. Miller" <>,
	Michael Kelley <>
Cc: "" <>,
	"" <>,
	"" <>
Subject: RE: [PATCH v2] hv_sock: Add support for delayed close
Date: Thu, 16 May 2019 17:16:37 +0000	[thread overview]
Message-ID: <PU1P153MB01693DB2206CD639AF356DBDBF0A0@PU1P153MB0169.APCP153.PROD.OUTLOOK.COM> (raw)
In-Reply-To: <PU1P153MB01698261307593C5D58AAF4FBF0A0@PU1P153MB0169.APCP153.PROD.OUTLOOK.COM>

> From:
> <> On Behalf Of Dexuan Cui
> Sent: Wednesday, May 15, 2019 9:34 PM
> ...

Hi Sunil,
To make it clear, your patch itself is good, and I was just talking about
the next change we're going to make. Once we make the next change,
IMO we need a further patch to schedule hvs_close_timeout() to the new
single-threaded workqueue rather than the global "system_wq".

> Next, we're going to remove the "channel->rescind" check in
> vmbus_hvsock_device_unregister() -- when doing that, IMO we need to
> fix a potential race revealed by the schedule_delayed_work() in this
> patch:
> When hvs_close_timeout() finishes, the "sk" struct has been freed, but
> vmbus_onoffer_rescind() -> channel->chn_rescind_callback(), i.e.
> hvs_close_connection(), may be still running and referencing the "chan"
> and "sk" structs (), which should no longer be referenced when
> hvs_close_timeout() finishes, i.e. "get_per_channel_state(chan)" is no
> longer safe. The problem is: currently there is no sync mechanism
> between vmbus_onoffer_rescind() and hvs_close_timeout().
> The race is a real issue only after we remove the "channel->rescind"
> in vmbus_hvsock_device_unregister().

A correction: IMO the race is real even for the current code, i.e. without
your patch: in vmbus_onoffer_rescind(), between we set channel->rescind
and we call channel->chn_rescind_callback(), the channel may have been
freed by vmbus_hvsock_device_unregister().

This race window is small and I guess that's why we never noticed it.

> I guess we need to introduce a new single-threaded workqueue in the
> vmbus driver, and offload both vmbus_onoffer_rescind() and
> hvs_close_timeout() onto the new workqueue.
-- Dexuan

  reply	other threads:[~2019-05-16 17:16 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-05-15  0:56 Sunil Muthuswamy
2019-05-16  4:34 ` Dexuan Cui
2019-05-16 17:16   ` Dexuan Cui [this message]
2019-05-16 18:11     ` Sunil Muthuswamy
2019-05-16 18:55       ` Dexuan Cui
2019-05-16 19:12 ` David Miller

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 \
    --in-reply-to=PU1P153MB01693DB2206CD639AF356DBDBF0A0@PU1P153MB0169.APCP153.PROD.OUTLOOK.COM \ \ \ \ \ \ \ \ \ \ \ \
    --subject='RE: [PATCH v2] hv_sock: Add support for delayed close' \

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

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).