linux-mm.kvack.org archive mirror
 help / color / mirror / Atom feed
From: Nick Piggin <nickpiggin@yahoo.com.au>
To: Peter Zijlstra <peterz@infradead.org>
Cc: linux-mm@kvack.org, "Dickins, Hugh" <hugh@veritas.com>
Subject: Re: [rfc][patch] mm: dirty page accounting hole
Date: Tue, 12 Aug 2008 17:06:26 +1000	[thread overview]
Message-ID: <200808121706.26686.nickpiggin@yahoo.com.au> (raw)
In-Reply-To: <1218523840.10800.129.camel@twins>

On Tuesday 12 August 2008 16:50, Peter Zijlstra wrote:
> On Tue, 2008-08-12 at 15:58 +1000, Nick Piggin wrote:
> > Hi,
> >
> > I think I'm running into a hole in dirty page accounting...
> >
> > What seems to be happening is that a page gets written to via a
> > VM_SHARED vma. We then set the pte dirty, then mark the page dirty.
> > Next, mprotect changes the vma so it is no longer writeable so it
> > is no longer VM_SHARED. The pte is still dirty.
> >
> > Then clear_page_dirty_for_io is called and leaves that pte dirty
> > and cleans the page. It never gets cleaned until munmap, so msync
> > and writeout accounting are broken.
> >
> > I have a fix which just scans VM_SHARED to VM_MAYSHARE. The other
> > way I tried is to clear the dirty and write bits and set the page
> > dirty in mprotect. The problem with that for me is that I'm trying
> > to rework the vm/fs layer so we never have to allocate data to
> > write out dirty pages (using page_mkwrite and dirty accounting),
>
> Ooh, nice!

Thanks for confirming.


> > and so this still leaves me with a window where the vma flags are
> > changed but before the pte is marked clean, in which time the page
> > is still dirty but it may have its metadata freed because it
> > doesn't look dirty.
> >
> > There are several other problems I've also run into, including a
> > fundamentally indadequate page_mkwrite locking scheme, which was
> > naturally ignored when I brought it up during reviewing those
> > patches. I digress...
>
> Yes, I remember you bringing that up, and later too when you did those
> fault patches. I assumed you were 'working' on it.

Oh that wasn't aimed at you, sorry ;) It is our general... "process"
of development. I mean, there is apparently some impending apocolypse
due to our scarcity of review bandwidth, so ignoring any review is
by definition insignificant.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

  reply	other threads:[~2008-08-12  7:06 UTC|newest]

Thread overview: 7+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2008-08-12  5:58 Nick Piggin
2008-08-12  6:50 ` Peter Zijlstra
2008-08-12  7:06   ` Nick Piggin [this message]
2008-08-12 11:15 ` Hugh Dickins
2008-08-12 11:30   ` Peter Zijlstra
2008-08-12 11:53   ` Nick Piggin
2008-08-12 13:17     ` Nick Piggin

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=200808121706.26686.nickpiggin@yahoo.com.au \
    --to=nickpiggin@yahoo.com.au \
    --cc=hugh@veritas.com \
    --cc=linux-mm@kvack.org \
    --cc=peterz@infradead.org \
    /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