On Mon, Aug 10, 2020 at 10:29:13AM +0200, Miklos Szeredi wrote:
> On Fri, Aug 7, 2020 at 9:55 PM Vivek Goyal <vgoyal@xxxxxxxxxx> wrote:
> >
> > fuse_file_put(sync) can be called with sync=true/false. If sync=true,
> > it waits for release request response and then calls iput() in the
> > caller's context. If sync=false, it does not wait for release request
> > response, frees the fuse_file struct immediately and req->end function
> > does the iput().
> >
> > iput() can be a problem with DAX if called in req->end context. If this
> > is last reference to inode (VFS has let go its reference already), then
> > iput() will clean DAX mappings as well and send REMOVEMAPPING requests
> > and wait for completion. (All the the worker thread context which is
> > processing fuse replies from daemon on the host).
> >
> > That means it blocks worker thread and it stops processing further
> > replies and system deadlocks.
>
> Is this reasoning specific to DAX? Seems to me this is a virtio-fs
> specific issue.
Hi Miklos,
I am looking at this patch closely and maybe we don't need it, given
current code.
Reason being that looks like for virtiofs now we set ctx->destroy=true
and that sets fc->destroy. And if that's set, we don't schedule async req
in fuse_release_common();
fuse_release_common() {
fuse_file_put(ff, ff->fc->destroy, isdir);
}
And if we don't schedule async file put, we don't run the risk of
blocking worker thread in evict_inode() with DAX enabled.
I ran blogbench multiple times with this patch reverted and I can't
reproduce the deadlock.
So we can drop this patch for now. If need be I will fix it as
post merge patch.
Thanks
Vivek
