On Thu, Jun 18, 2020 at 02:46:03PM +0200, Andreas Gruenbacher wrote:

On Wed, Jun 17, 2020 at 4:22 AM Matthew Wilcox [off-list ref] wrote:

quoted

On Wed, Jun 17, 2020 at 02:57:14AM +0200, Andreas Grünbacher wrote:

quoted

Right, the approach from the following thread might fix this:

https://lore.kernel.org/linux-fsdevel/20191122235324.17245-1-agruenba@redhat.com/T/#t (local)

In general, I think this is a sound approach.

Specifically, I think FAULT_FLAG_CACHED can go away.  map_pages()
will bring in the pages which are in the page cache, so when we get to
gfs2_fault(), we know there's a reason to acquire the glock.

We'd still be grabbing a glock while holding a dependent page lock.
Another process could be holding the glock and could try to grab the
same page lock (i.e., a concurrent writer), leading to the same kind
of deadlock.

What I'm saying is that gfs2_fault should just be:

+static vm_fault_t gfs2_fault(struct vm_fault *vmf)
+{
+	struct inode *inode = file_inode(vmf->vma->vm_file);
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_holder gh;
+	vm_fault_t ret;
+	int err;
+
+	gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh);
+	err = gfs2_glock_nq(&gh);
+	if (err) {
+		ret = block_page_mkwrite_return(err);
+		goto out_uninit;
+	}
+	ret = filemap_fault(vmf);
+	gfs2_glock_dq(&gh);
+out_uninit:
+	gfs2_holder_uninit(&gh);
+	return ret;
+}

because by the time gfs2_fault() is called, map_pages() has already been
called and has failed to insert the necessary page, so we should just
acquire the glock now instead of trying again to look for the page in
the page cache.