Bad file access on the rise

[Steve Grubb]

On Saturday, June 08, 2013 06:36:38 AM Matthew Garrett wrote:
> On Fri, Jun 07, 2013 at 05:24:30PM -0400, Steve Grubb wrote:
> > Hmm...sounds like kernel change. But in the meantime, most of the
> > offenders I see seem to have something to do with loading icons
>
> Sounds like code that doesn't differentiate between files that are in
> user-local directories and system-global directories. That's something
> that could presumably be fixed, but it seems like a bunch of effort.

>From the code I saw previously, it seems like just changing the function being 
called to the variant without noatime in its name. The comments in the code 
Colin pointed to say this:

 * gs_file_read_noatime:
 * @file: a #GFile
 * @cancellable: a #GCancellable
 * @error: a #GError
 *
 * Like g_file_read(), but try to avoid updating the file's
 * access time.  This should be used by background scanning
 * components such as search indexers, antivirus programs, etc.

And evince, firefox, or openoffice are not any of those  ^^.   :-)


> Other than a heuristic based on whether a path is in the user's home
> directory or not, the only way to avoid this is to stat before opening -
> and that's obviously prone to failure.

Does opening with noatime really make a measurable difference (assuming it 
worked)? I suspect not since what we have now is 2 syscalls. It would probably 
be faster to load icons without trying to set NOATIME.

Thanks,
-Steve