podman-container-tools/podman · error

cannot write to pipe: %m

Error message

cannot write to pipe: %m

What it means

In create_pause_process, after pause.pid was atomically installed via rename_noreplace, the intermediate child writes the one-byte '0' ack to pipe p[1] to release the parent (which blocks in read(p[0]) until the pid file exists). A failed write makes the child _exit(EXIT_FAILURE); the parent then sees r != 1 and returns -1, so rootless setup fails. Because the pipe has no other readers left when the parent dies, the usual errno is EPIPE/EBADF.

Source

Thrown at pkg/rootless/rootless_linux.c:1077

              fprintf (stderr, "cannot write to file descriptor: %m\n");
              kill (pid, SIGKILL);
              _exit (EXIT_FAILURE);
            }
          close (fd);

          /* There can be another process at this point trying to configure the user namespace and the pause
           process, do not override the pid file if it already exists. */
          if (rename_noreplace (AT_FDCWD, tmp_file_path, AT_FDCWD, pause_pid_file_path) < 0)
            {
              unlink (tmp_file_path);
              kill (pid, SIGKILL);
              _exit (EXIT_FAILURE);
            }

          r = TEMP_FAILURE_RETRY (write (p[1], "0", 1));
          if (r < 0)
            {
              fprintf (stderr, "cannot write to pipe: %m\n");
              _exit (EXIT_FAILURE);
            }
          close (p[1]);

          _exit (EXIT_SUCCESS);
        }
      else
        {
          int null;

          close (p[1]);

          null = open ("/dev/null", O_RDWR);
          if (null >= 0)
            {
              dup2 (null, 0);
              dup2 (null, 1);
              dup2 (null, 2);

View on GitHub (pinned to a2409076ef)

Solutions

1. Scroll up and fix the FIRST error printed before this one; this write failure is a downstream symptom
2. Re-run the podman command; a one-off parent-death race is transient
3. Check for OOM kills or manual kills of podman processes around the failure time: journalctl -k | grep -i oom

# no meaningful pre-check: this write only fails after the parent already died
# keep the parent healthy instead:
ulimit -v unlimited 2>/dev/null; cat /sys/fs/cgroup/memory.max 2>/dev/null

When it happens

Trigger: The parent side of create_pause_process already exited and closed p[0] (race when the parent is killed, or reexec_in_user_namespace_wait reaped it first), so write(p[1], "0", 1) fails with EPIPE; EIO on a broken pipe device is not expected. This message is almost always a symptom following another, earlier failure.

Common situations: System under memory pressure where the parent is OOM-killed; manual kill of the podman re-exec chain; usually appears together with another error message above it in the log.

Related errors

• cannot write to file descriptor: %m
• cannot save namespace handles: %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/fb7fdba32ba7b190. Report an issue: GitHub.