Make 'zfs rollback' work even if the previous 'zfs receive -s' was in…

…terrupted.

Signed-off-by: Wolfgang Hoschek <wolfgang.hoschek@mac.com>

bzfs/bzfs.py

@@ -2189,7 +2189,12 @@ def latest_common_snapshot(snapshots_with_guids: List[str], intersect_guids: Set
                 cmd = p.split_args(
                     f"{dst.sudo} {p.zfs_program} rollback -r {p.force_unmount} {p.force_hard}", latest_common_dst_snapshot
                 )
-                self.run_ssh_command(dst, log_debug, is_dry=p.dry_run, print_stdout=True, cmd=cmd)
+                try:
+                    self.run_ssh_command(dst, log_debug, is_dry=p.dry_run, print_stdout=True, cmd=cmd)
+                except (subprocess.CalledProcessError, subprocess.TimeoutExpired, UnicodeDecodeError) as e:
+                    no_sleep = self.clear_resumable_recv_state_if_necessary(dst_dataset, e.stderr)
+                    # op isn't idempotent so retries regather current state from the start
+                    raise RetryableError("Subprocess failed", no_sleep=no_sleep) from e
 
             if latest_src_snapshot and latest_src_snapshot == latest_common_src_snapshot:
                 log.info(f"{tid} Already up-to-date: %s", dst_dataset)
@@ -2568,6 +2573,16 @@ def clear_resumable_recv_state() -> bool:
         if all(marker in stderr for marker in markers):
             return clear_resumable_recv_state()
 
+        # Same cause as above, except that this error can occur during 'zfs rollback'
+        markers = [
+            "cannot rollback to",
+            "clones of previous snapshots exist",
+            "use '-R' to force deletion of the following clones and dependents",
+            f"\n{dst_dataset}/%recv\n",
+        ]
+        if all(marker in stderr for marker in markers):
+            return clear_resumable_recv_state()
+
         return False
 
     def _recv_resume_token(self, dst_dataset: str, retry_count: int) -> Tuple[Optional[str], List[str], List[str]]: