From: Fred Isaman <faisaman4@gmail.com>
To: Jeff Layton <jlayton@redhat.com>
Cc: trond.myklebust@netapp.com, linux-nfs@vger.kernel.org, eguan@redhat.com
Subject: Re: non-aligned DIO reads on NFS are corrupting memory in 3.7.0
Date: Wed, 12 Dec 2012 11:20:11 -0500 [thread overview]
Message-ID: <CADnza44_ZTHuuFHbQds9TBcSAdHD-FBoSJLwym7ETcC-HeFjZw@mail.gmail.com> (raw)
In-Reply-To: <20121212094654.5ce53d89@tlielax.poochiereds.net>
On Wed, Dec 12, 2012 at 9:46 AM, Jeff Layton <jlayton@redhat.com> wrote:
> One of our QA folks found that the attached testcase would segfault
> when run on a recent rhel6 kernel that has a backport of the pnfs dio
> code. I get the same segfault when I run it on a 3.7.0 kernel as well.
>
> I think the problem is that because the buffer we're reading into is on
> the stack, the kernel is scribbling over the rest of the page after the
> read and corrupting it.
>
> The problem, I think is this block in nfs_direct_read_completion():
>
> -----------------------[snip]-----------------------
> if (test_bit(NFS_IOHDR_EOF, &hdr->flags)) {
> if (bytes > hdr->good_bytes)
> zero_user(page, 0, PAGE_SIZE);
> else if (hdr->good_bytes - bytes < PAGE_SIZE)
> zero_user_segment(page,
> hdr->good_bytes & ~PAGE_MASK,
> PAGE_SIZE);
> }
> -----------------------[snip]-----------------------
>
> If I comment that out, then the test passes and it doesn't scribble
> over memory. I'm not clear on what that block is trying to accomplish.
> If we get a short read in the DIO codepath, I don't think we ought to
> be zeroing out the rest of the page. We should just return the number
> of bytes read and be done with it.
>
I would say the problem is not zeroing memory, but that the code isn't
taking into account the offsets into the page.
> I'm also suspicious of the "if (!PageCompound(page))" check in that
> function as well. It doesn't seem like we ought to be marking pages
> dirty in the DIO codepaths, should we?
>
> --
> Jeff Layton <jlayton@redhat.com>
I'm not sure if we should, but code to do so has been around forever.
The exception for PageCompound is from commit 566dd6064e89b "NFS: Make
directIO aware of compound pages", almost 7 years ago.
Fred
next prev parent reply other threads:[~2012-12-12 16:20 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-12-12 14:46 non-aligned DIO reads on NFS are corrupting memory in 3.7.0 Jeff Layton
2012-12-12 16:20 ` Fred Isaman [this message]
2012-12-12 16:30 ` Myklebust, Trond
2012-12-12 16:31 ` Jeff Layton
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:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CADnza44_ZTHuuFHbQds9TBcSAdHD-FBoSJLwym7ETcC-HeFjZw@mail.gmail.com \
--to=faisaman4@gmail.com \
--cc=eguan@redhat.com \
--cc=jlayton@redhat.com \
--cc=linux-nfs@vger.kernel.org \
--cc=trond.myklebust@netapp.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).