On Fri 06-09-19 23:48:31, Andreas Gruenbacher wrote: > On Fri, Sep 6, 2019 at 11:28 PM Dave Chinner <david@xxxxxxxxxxxxx> wrote: > > On Fri, Sep 06, 2019 at 10:52:41PM +0200, Andreas Gruenbacher wrote: > > > Hi, > > > > > > I've just fixed a mmap write vs. truncate consistency issue on gfs on > > > filesystems with a block size smaller that the page size [1]. > > > > > > It turns out that the same problem exists between mmap write and hole > > > punching, and since xfstests doesn't seem to cover that, > > > > AFAIA, fsx exercises it pretty often. Certainly it's found problems > > with XFS in the past w.r.t. these operations. > > > > > I've written a > > > new test [2]. > > > > I suspect that what we really want is a test that runs > > _test_generic_punch using mmap rather than pwrite... > > > > > Ext4 and xfs both pass that test; they both apparently > > > mark the pages that have a hole punched in them as read-only so that > > > page_mkwrite is called before those pages can be written to again. > > > > XFS invalidates the range being hole punched (see > > xfs_flush_unmap_range() under XFS_MMAPLOCK_EXCL, which means any > > attempt to fault that page back in will block on the MMAPLOCK until > > the hole punch finishes. > > This isn't about writes during the hole punching, this is about writes > once the hole is punched. For example, the test case I've posted > creates the following file layout with 1k blocksize: > > DDDD DDDD DDDD > > Then it punches a hole like this: > > DDHH HHHH HHDD > > Then it fills the hole again with mwrite: > > DDDD DDDD DDDD > > As far as I can tell, that needs to trigger page faults on all three > pages. I did get these on ext4; judging from the fact that xfs works, > the also seem to occur there; but on gfs2, page_mkwrite isn't called > for the two partially mapped pages, only for the page in the middle > that's entirely within the hole. And I don't see where those pages are > marked read-only; it appears like pagecache_isize_extended isn't > called on ext4 or xfs. So how does this work there? The trick ext4 & xfs use is that they writeout the range being punched first (see e.g. ext4_punch_hole() calling filemap_write_and_wait_range() or xfs_flush_unmap_range() called from xfs_free_file_space()). This writeout also has the effect that all the page mappings for that range get write-protected. Another related issue is what Dave points out: Even if you use writeout to writeprotect pages, GFS2 still seems to have a race where page fault can come while you are freeing blocks and if you allow that you usually get into a problematic state. Effects depend on fs implementation details but usually it can result in stale data exposure or fs corruption. Honza -- Jan Kara <jack@xxxxxxxx> SUSE Labs, CR
