Linux-Fsdevel Archive on
 help / color / Atom feed
From: Christoph Hellwig <>
To: "Darrick J. Wong" <>
Cc: Christoph Hellwig <>,,,
Subject: aio completions vs file_accessed race, was: Re: [PATCH 7/8] xfs: split direct I/O and DAX path
Date: Thu, 29 Sep 2016 10:38:24 +0200
Message-ID: <> (raw)
In-Reply-To: <>

On Wed, Sep 28, 2016 at 07:53:52PM -0700, Darrick J. Wong wrote:
> So I noticed that generic/323 starts crashing in file_accessed -> touch_atime
> because iocb->ki_filp->f_path.dentry == NULL.  For a while I thought it was
> some weird reflink bug, but I finally had time to go build a vanilla 4.8-rc8
> kernel and that blew up here too.  I'm not sure why this line got inserted
> here, since it wasn't there prior to this patch, AFAICT.

This line was there before near the end of xfs_file_dio_aio_read already,
e.g. line 376 just before the above commit, but it only got introduced
a bit earlier in "xfs: stop using generic_file_read_iter for direct I/O",
which copied it over from generic_file_read_iter.  І think any new
issues in these commits could just be a minor timing change, as
we're not changing struct file refcounting in any way here.

generic/323 reproduces the last struct file reference being dropped
by aio completions, so it seems like we have an issue here, which
I suspect is something in the common code.  I can't reproduce it
locally, but looking at the aio_complete -> kiocb_free callchain
and the lack of other struct file refcounting in aio.c it seems
inherently unsafe to reference struct file once the completion
may have run, that is after (__)blkdev_direct_IO returned.

I'll see if I can come up with a solution for that, most likely
that would involve moving the file_accessed call into __blkdev_direct_IO
before we drop the final reference on the dio structure.

  reply index

Thread overview: 27+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-06-22 15:27 xfs: untangle the direct I/O and DAX path, fix DAX locking Christoph Hellwig
2016-06-22 15:27 ` [PATCH 1/8] xfs: don't pass ioflags around in the ioctl path Christoph Hellwig
2016-06-22 15:27 ` [PATCH 2/8] xfs: kill ioflags Christoph Hellwig
2016-06-22 15:27 ` [PATCH 3/8] xfs: remove s_maxbytes enforcement in xfs_file_read_iter Christoph Hellwig
2016-06-22 15:27 ` [PATCH 4/8] xfs: split xfs_file_read_iter into buffered and direct I/O helpers Christoph Hellwig
2016-06-22 15:27 ` [PATCH 5/8] xfs: stop using generic_file_read_iter for direct I/O Christoph Hellwig
2016-06-22 15:27 ` [PATCH 6/8] xfs: direct calls in the direct I/O path Christoph Hellwig
2016-06-22 15:27 ` [PATCH 7/8] xfs: split direct I/O and DAX path Christoph Hellwig
2016-09-29  2:53   ` Darrick J. Wong
2016-09-29  8:38     ` Christoph Hellwig [this message]
2016-09-29 20:18       ` aio completions vs file_accessed race, was: " Christoph Hellwig
2016-09-29 20:18         ` Christoph Hellwig
2016-09-29 20:33           ` Darrick J. Wong
2016-06-22 15:27 ` [PATCH 8/8] xfs: fix locking for DAX writes Christoph Hellwig
2016-06-23 14:22   ` Boaz Harrosh
2016-06-23 23:24 ` xfs: untangle the direct I/O and DAX path, fix DAX locking Dave Chinner
2016-06-24  1:14   ` Dan Williams
2016-06-24  7:13     ` Dave Chinner
2016-06-24  7:31       ` Christoph Hellwig
2016-06-24  7:26   ` Christoph Hellwig
2016-06-24 23:00     ` Dave Chinner
2016-06-28 13:10       ` Christoph Hellwig
2016-06-28 13:27         ` Boaz Harrosh
2016-06-28 13:39           ` Christoph Hellwig
2016-06-28 13:56             ` Boaz Harrosh
2016-06-28 15:39               ` Christoph Hellwig
2016-06-29 12:23                 ` Boaz Harrosh

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

Linux-Fsdevel Archive on

Archives are clonable:
	git clone --mirror linux-fsdevel/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 linux-fsdevel linux-fsdevel/ \
	public-inbox-index linux-fsdevel

Example config snippet for mirrors

Newsgroup available over NNTP:

AGPL code for this site: git clone