snapshot restore fails spectacularly on btrfs

[JohnPeacockMessageSystems]

I created a snapshot on a btrfs filesystem and lxc-snapshot correctly
created a btrfs subvolume snapshot. However when restoring the
snapshot, I got this instead:

$ sudo lxc-snapshot -n rh65-dev -r snap0
rsync: writefd_unbuffered failed to write 4 bytes to socket [generator]: Broken pipe (32)
rsync error: error in rsync protocol data stream (code 12) at io.c(1532) [generator=3.0.9]
lxc_container: rsyncing /var/lib/lxc/containersnaps/rh65-dev/snap0/rootfs to /var/lib/lxc/container/rh65-dev/rootfs

lxc_container: Error copying storage
lxc_container: _recursive_rmdir_onedev: failed to delete /var/lib/lxc/container/rh65-dev
lxc_container: Error destroying container directory for rh65-dev
lxc_container: Error restoring snapshot snap0

There are a couple of problems with that:

1. lxc-snapshot restore is using rsync and not the btrfs snapshot tool, e.g.

btrfs subvolume delete ORIG.ROOTFS
btrfs subvolume snap SNAP.ROOTFS ORIG.ROOTFS
btrfs subvolume delete SNAP.ROOTFS (optional)

2. lxc-snapshot has now messed up the original subvolume filesystem in such a way that I can no longer restore to the same path:

# rm -r /var/lib/lxc/container/rh65-dev/rootfs
 rm: cannot remove ‘/var/lib/lxc/container/rh65-dev/rootfs’: Operation not permitted

Even though the subvolume was no longer listed by

btrfs subvolume list

(presumably because the rsync had messed it up), but it was apparently
still marked as such in the filesystem metadata (even after a reboot).
I was able to clean up the mess via:

btrfs subvolume delete /var/lib/lxc/container/rh65-dev/rootfs
rm -rf /var/lib/lxc/container/rh65-dev/rootfs

and move the snapshot back in place to restore access to my original
instance.

[hallyn]

I will not be able to rework this in the next few weeks. I believe on the mailing list it was agreed that you would attempt a redesign, and I'll look at any patch you send. Otherwise I hope to look at this in february.

[caglar10ur]

Something like following seems to work fine for me. I'm guessing it's not going to work with unprivileged container case but on the other hand I'm not sure whether we support btrfs in that case at all. Pasting here for reference (aka until I find some time to test more and submit to the list)

diff --git a/src/lxc/bdev.c b/src/lxc/bdev.c
index 627c09a..64d0fe0 100644
--- a/src/lxc/bdev.c
+++ b/src/lxc/bdev.c
@@ -2598,6 +2598,30 @@ struct bdev *bdev_copy(struct lxc_container *c0, const char *cname,
        if (snap)
                return new;

+       /*
+        * https://github.com/lxc/lxc/issues/131 - Use btrfs snapshot feature instead of rsync
+        */
+       if (bdevtype && strcmp(orig->type, "btrfs") == 0 && strcmp(new->type, "btrfs") == 0) {
+               if (btrfs_destroy(new) < 0) {
+                       bdev_put(orig);
+                       bdev_put(new);
+                       ERROR("Error destroying %s", new->dest);
+               }
+               if (mkdir_p(new->dest, 0755) < 0) {
+                       ERROR("Error creating %s", new->dest);
+                       bdev_put(orig);
+                       bdev_put(new);
+                       return NULL;
+               }
+               if (btrfs_snapshot(orig->dest, new->dest) < 0) {
+                       bdev_put(orig);
+                       bdev_put(new);
+                       ERROR("Error restoring %s to %s", orig->dest, new->dest);
+               }
+               bdev_put(orig);
+               return new;
+       }
+
        pid = fork();
        if (pid < 0) {
                SYSERROR("fork");