lkml.org 
[lkml]   [2008]   [Jan]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 2/2] updating ctime and mtime at syncing
From
Date

On Mon, 2008-01-14 at 14:35 +0100, Peter Zijlstra wrote:
> On Mon, 2008-01-14 at 14:14 +0100, Miklos Szeredi wrote:
> > > 2008/1/14, Miklos Szeredi <miklos@szeredi.hu>:
> > > > > http://bugzilla.kernel.org/show_bug.cgi?id=2645
> > > > >
> > > > > Changes for updating the ctime and mtime fields for memory-mapped files:
> > > > >
> > > > > 1) new flag triggering update of the inode data;
> > > > > 2) new function to update ctime and mtime for block device files;
> > > > > 3) new helper function to update ctime and mtime when needed;
> > > > > 4) updating time stamps for mapped files in sys_msync() and do_fsync();
> > > > > 5) implementing the feature of auto-updating ctime and mtime.
> > > >
> > > > How exactly is this done?
> > > >
> > > > Is this catering for this case:
> > > >
> > > > 1 page is dirtied through mapping
> > > > 2 app calls msync(MS_ASYNC)
> > > > 3 page is written again through mapping
> > > > 4 app calls msync(MS_ASYNC)
> > > > 5 ...
> > > > 6 page is written back
> > > >
> > > > What happens at 4? Do we care about this one at all?
> > >
> > > The POSIX standard requires updating the file times every time when msync()
> > > is called with MS_ASYNC. I.e. the time stamps should be updated even
> > > when no physical synchronization is being done immediately.
> >
> > Yes. However, on linux MS_ASYNC is basically a no-op, and without
> > doing _something_ with the dirty pages (which afaics your patch
> > doesn't do), it's impossible to observe later writes to the same page.
> >
> > I don't advocate full POSIX conformance anymore, because it's probably
> > too expensive to do (I've tried). Rather than that, we should
> > probably find some sane compromise, that just fixes the real life
> > issue. Here's a pointer to the thread about this:
> >
> > http://lkml.org/lkml/2007/3/27/55
> >
> > Your patch may be a good soultion, but you should describe in detail
> > what it does when pages are dirtied, and when msync/fsync are called,
> > and what happens with multiple msync calls that I've asked about.
> >
> > I suspect your patch is ignoring writes after the first msync, but
> > then why care about msync at all? What's so special about the _first_
> > msync? Is it just that most test programs only check this, and not
> > what happens if msync is called more than once? That would be a bug
> > in the test cases.
>
> I must agree, doing the mmap dirty, MS_ASYNC, mmap retouch, MS_ASYNC
> case correctly would need a lot more code which I doubt is worth the
> effort.
>
> It would require scanning the PTEs and marking them read-only again on
> MS_ASYNC, and some more logic in set_page_dirty() because that currently
> bails out early if the page in question is already dirty.

More fun, it would require marking them RO but leaving the dirty bit
set, because this ext3 fudge where we confuse the page dirty state - or
did that get fixed?



\
 
 \ /
  Last update: 2008-01-14 14:43    [W:0.051 / U:0.396 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site