Fix double close on ubivol descriptor

[tvdstaaij]

The g_output_stream_close call was introduced at some point after the casync functionality for UBI was introduced (#501). This close call is fine for the copy_raw_image branch, but for the casync branch it causes the update to fail:

rauc[250]: Installation error: Failed updating slot rootfs.1: Error closing file descriptor: Bad file descriptor

This was also mentioned in #1200. However it was not investigated there but instead sidestepped by using a different update approach.

Investigation with strace shows that the desciptor is closed twice:

290   openat(AT_FDCWD, "/dev/ubi0_1", O_WRONLY|O_EXCL|O_LARGEFILE <unfinished ...>
290   <... openat resumed>)             = 8
[snip]
290   close(8 <unfinished ...>
290   <... close resumed>)              = 0
290   close(8 <unfinished ...>
290   <... close resumed>)              = -1 EBADF (Bad file descriptor)

This can be explained by the documentation of
g_subprocess_launcher_close():

Closes all the file descriptors previously passed to the object with
g_subprocess_launcher_take_fd(), g_subprocess_launcher_take_stderr_fd(), etc.
(...)
This function is called automatically when the #GSubprocessLauncher is disposed

[jluebbe]

The intention with the explicit close is to be able to catch more errors (some of which are only reported on close()). So perhaps a better way is to pass a dup()ed FD to g_subprocess_launcher_take_stdout_fd()?

Sadly this code is not covered by the CI, as we didn't want to build a custom version of casync systemd/casync#227.

[tvdstaaij]

The intention with the explicit close is to be able to catch more errors (some of which are only reported on close()). So perhaps a better way is to pass a dup()ed FD to g_subprocess_launcher_take_stdout_fd()?

Ah, I wasn't aware of the existence of dup(). I submitted a new PR.

[jluebbe]

Note that you can force-push to your branch to update this PR, which keeps the discussion in one place. :)