On Thu, Jan 25, 2007 at 12:43:23AM +1100, Nick Piggin wrote: > Peter Zijlstra wrote: > >On Wed, 2007-01-24 at 09:37 +1100, David Chinner wrote: > > > >>With the recent changes to cancel_dirty_pages(), XFS will > >>dump warnings in the syslog because it can truncate_inode_pages() > >>on dirty mapped pages. > >> > >>I've determined that this is indeed correct behaviour for XFS > >>as this can happen in the case of races on mmap()d files with > >>direct I/O. In this case when we do a direct I/O read, we > >>flush the dirty pages to disk, then truncate them out of the > >>page cache. Unfortunately, between the flush and the truncate > >>the mmap could dirty the page again. At this point we toss a > >>dirty page that is mapped. > > > > > >This sounds iffy, why not just leave the page in the pagecache if its > >mapped anyway? > > And why not just leave it in the pagecache and be done with it?
because what is in cache is then not coherent with what is on disk, and a direct read is supposed to read the data that is present in the file at the time it is issued. > All you need is to do a writeout before a direct IO read, which is > what generic dio code does. No, that's not good enough - after writeout but before the direct I/O read is issued a process can fault the page and dirty it. If you do a direct read, followed by a buffered read you should get the same data. The only way to guarantee this is to chuck out any cached pages across the range of the direct I/O so they are fetched again from disk on the next buffered I/O. i.e. coherent at the time the direct I/O is issued. > I guess you'll say that direct writes still need to remove pages, Yup. > but in that case you'll either have to live with some racyness > (which is what the generic code does), or have a higher level > synchronisation to prevent buffered + direct IO writes I suppose? The XFS inode iolock - direct I/O writes take it shared, buffered writes takes it exclusive - so you can't do both at once. Buffered reads take is shared, which is another reason why we need to purge the cache on direct I/O writes - they can operate concurrently (and coherently) with buffered reads. Cheers, Dave. -- Dave Chinner Principal Engineer SGI Australian Software Group - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/