reliably detect writes to a file: mmap, mtime, ...

reliably detect writes to a file: mmap, mtime, ...

[Martin Uecker]

Hi all,

I want to reliably detect changes to a file even when 
written to using mmap. Surprisingly, there seems to be
no API which would make this possible. Or at least I 
haven't found a way to do it... 


I looked at:

- mtime. What is missing here is an API which would
force mtime to be updated if there are dirty PTEs
in some mapping (which need to be cleared/transferred 
to struct page at this point). This would allow to 
reliably detect changes to the file. If I understand it 
correctly, there was patch from Andy Lutomirski which
made msync(ASYNC) do exactly this:

http://oss.sgi.com/archives/xfs/2013-08/msg00748.html

But it seems this never got in. The other problem with
this is that mtime has limited granularity.
(but maybe that could be worked around by having some
kind of counter + API which tells how often mtime has
been updated without changing its nominal value)



- I also looked at soft-dirty bits, but this API seems
to have several limitations:  1.) it tracks writes
through a specific mapping 2.) it can only have
a single user at the same time 3.) who has to have 
special privileges 4.) and it seems impossible read
and clear the soft-dirty bits at the same time (so you
might miss writes).


But maybe there are other ways... I am missing
something?


Martin



--
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>

Re: reliably detect writes to a file: mmap, mtime, ...

[Andy Lutomirski]

On Dec 30, 2015 4:24 AM, "Martin Uecker" <muecker@gwdg.de> wrote:
>
>
>
> Hi all,
>
> I want to reliably detect changes to a file even when
> written to using mmap. Surprisingly, there seems to be
> no API which would make this possible. Or at least I
> haven't found a way to do it...
>
>
> I looked at:
>
> - mtime. What is missing here is an API which would
> force mtime to be updated if there are dirty PTEs
> in some mapping (which need to be cleared/transferred
> to struct page at this point). This would allow to
> reliably detect changes to the file. If I understand it
> correctly, there was patch from Andy Lutomirski which
> made msync(ASYNC) do exactly this:
>
> http://oss.sgi.com/archives/xfs/2013-08/msg00748.html
>
> But it seems this never got in. The other problem with
> this is that mtime has limited granularity.
> (but maybe that could be worked around by having some
> kind of counter + API which tells how often mtime has
> been updated without changing its nominal value)

Those patches plus nanosecond granularity should do it, I think.  I
keep meaning to dust them off.  You could do it :)

--Andy

--
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>