podman-container-tools/podman · error

cannot chdir to %s: %m

Error message

cannot chdir to %s: %m

What it means

On the joined-namespaces shortcut path the preamble saved the cwd before joining, then after setresgid/setresuid(0,0,0) does chdir(cwd) to restore the working directory (pkg/rootless/rootless_linux.c:936-941). This message means that chdir failed in the joined user+mount namespace - commonly ENOENT (the directory was removed between getcwd and chdir), EACCES (the path is not searchable by the identity now in force, since uid 0 inside the userns maps to the original user), or a mount that is not visible in the joined mount namespace. The process _exits.

Source

Thrown at pkg/rootless/rootless_linux.c:938

      setenv ("_CONTAINERS_ROOTLESS_GID", gid_fmt, 1);

      /* We are in the user+mount namespace, these errors are not recoverable.  */

      if (syscall_setresgid (0, 0, 0) < 0)
        {
          fprintf (stderr, "cannot setresgid: %m\n");
          _exit (EXIT_FAILURE);
        }

      if (syscall_setresuid (0, 0, 0) < 0)
        {
          fprintf (stderr, "cannot setresuid: %m\n");
          _exit (EXIT_FAILURE);
        }

      if (chdir (cwd) < 0)
        {
          fprintf (stderr, "cannot chdir to %s: %m\n", cwd);
          _exit (EXIT_FAILURE);
        }

      rootless_uid_init = uid;
      rootless_gid_init = gid;
    }
}

static int
syscall_clone (unsigned long flags, void *child_stack)
{
#if defined(__s390__) || defined(__CRIS__)
  return (int) syscall (__NR_clone, child_stack, flags);
#else
  return (int) syscall (__NR_clone, flags, child_stack);
#endif
}

View on GitHub (pinned to a2409076ef)

Solutions

1. Re-run the command from a stable, accessible directory such as $HOME
2. If the directory was deleted, recreate it or change out of it before invoking podman
3. For FUSE mounts, ensure they are mounted with permissions allowing the podman process (allow_other, correct uid)
4. Report upstream if a permanently existing directory on a normal filesystem still triggers it

Example fix

# before
cd /tmp/job && rm -rf /tmp/job
podman ps   # cannot chdir to /tmp/job: No such file or directory

# after
WORK=$(pwd); cd "$HOME"; rm -rf "$WORK"
podman ps

# Use a directory that will exist and stay accessible across the namespace join
case "$PWD" in /tmp/*|/var/tmp/*) cd "$HOME" || exit 1;; esac
podman "$@"

When it happens

Trigger: The startup directory is deleted concurrently while podman joins the shared namespaces; the path to the cwd traverses directories whose permissions changed; the cwd sits on a mount (e.g. a FUSE mount with restrictive allow_other/uid settings) that is not accessible from the joined mount namespace.

Common situations: CI scripts removing the workspace then invoking podman; running podman from an automounted or FUSE directory; concurrent permission changes on home directory trees.

Related errors

• opendir %s: %m
• error getting current working directory: %m
• error getting current working directory: %m\n
• stat %s: %m
• error opening namespace handles: %m

AI-assisted analysis of podman-container-tools/podman@a2409076ef (2026-08-15). Data as JSON: /api/errors/fd6d0a5bedbab179. Report an issue: GitHub.