podman-container-tools/podman · error

cannot save namespace handles: %m

Error message

cannot save namespace handles: %m

What it means

In the reexec_userns_join child (same block exists in reexec_in_user_namespace), get_and_save_ns_handles_with_lock(state_dir) failed — it creates/locks state_dir/ns_handles, gets nsfs file handles for the user+mount namespaces and saves them atomically — AND errno was NOT one of the tolerated fallback values (EOPNOTSUPP, EPERM, ENOSYS, ENOENT). Because the failure is 'unexpected', podman does not fall back to the pause process and _exit(EXIT_FAILURE)s. Real errnos here: ENOSPC/EIO writing the ns_handles file, EACCES/EROFS on the state dir, ENAMETOOLONG, lock-acquisition failures.

Source

Thrown at pkg/rootless/rootless_linux.c:1249

      fprintf (stderr, "cannot chdir to %s: %m\n", cwd);
      _exit (EXIT_FAILURE);
    }

  if (state_dir && state_dir[0] != '\0')
    {
      /* Try to use namespace file handles instead of a pause process.  */
      if (get_and_save_ns_handles_with_lock (state_dir) < 0)
        {
          /* Fall back to pause process only if kernel doesn't support nsfs handles,
             if they are blocked (e.g. seccomp), or if the state directory doesn't exist yet.  */
          if (errno == EOPNOTSUPP || errno == EPERM || errno == ENOSYS || errno == ENOENT)
            {
              if (create_pause_process (state_dir, argv) < 0)
                _exit (EXIT_FAILURE);
            }
          else
            {
              fprintf (stderr, "cannot save namespace handles: %m\n");
              _exit (EXIT_FAILURE);
            }
        }
    }
  if (sigprocmask (SIG_SETMASK, &oldsigset, NULL) < 0)
    {
      fprintf (stderr, "cannot block signals: %m\n");
      _exit (EXIT_FAILURE);
    }

  execvp ("/proc/self/exe", argv);
  fprintf (stderr, "failed to reexec: %m\n");

  _exit (EXIT_FAILURE);
}

static void
check_proc_sys_userns_file (const char *path)

View on GitHub (pinned to a2409076ef)

Solutions

1. Fix the state dir: ls -ld "$HOME/.local/share/containers" (and $XDG_RUNTIME_DIR/containers) — it must be owned by you and writable; df -h to confirm space
2. Unset PODMAN_NO_PAUSE_PROCESS (or set it to 0) so rootless podman uses the pause-process fallback instead of ns handles, which does not hit this hard failure
3. Shorten the state dir path if it approaches PATH_MAX
4. Clear a corrupt ns_handles file: rm "$state_dir/ns_handles" and retry

Example fix

# before: ns-handles path fails hard
export PODMAN_NO_PAUSE_PROCESS=1
podman system service

# after: fall back to the pause process
unset PODMAN_NO_PAUSE_PROCESS   # default is the pause process
podman system service

d="$HOME/.local/share/containers"
[ -d "$d" ] && [ -w "$d" ] || { echo "state dir missing/not writable: $d"; exit 1; }
df --output=pcent "$d" | tail -1   # leave headroom; ns_handles writes fail hard on ENOSPC

# when ns-handle persistence keeps failing, switch to the pause-process mode
if ! podman info >/dev/null 2>&1; then
  unset PODMAN_NO_PAUSE_PROCESS   # pause process is the default fallback
  podman info
fi

When it happens

Trigger: PODMAN_NO_PAUSE_PROCESS=1 is set (ns-handles path active) and the state dir is unwritable, full, read-only, or its path exceeds PATH_MAX; corruption on the filesystem holding state_dir/ns_handles; flock contention failure inside acquire_ns_handles_lock.

Common situations: Disk-full or quota-exceeded home directories; state dir owned by another uid after uid changes; state dir on a read-only or failing mount; very long XDG paths.

Related errors

• cannot write to file descriptor: %m
• cannot write to pipe: %m
• cannot set %s namespace
• pause.pid file refers to PID %ld which is not a pause proces
• unable to print to string

AI-assisted analysis of podman-container-tools/podman@a2409076ef (2026-08-15). Data as JSON: /api/errors/d116ac680269a01b. Report an issue: GitHub.