We were creating the source, attaching it to the main context, and
removing our reference right away. That meant that the ownership of
the source was fully given to the main context, and so we shouldn't
have attempted to remove it from the main context later on, as it
could have already been fired and removed behind our back.

Solve this, by making sure we keep a valid source reference of our
own, and in this way we can safely destroy the source (it's fine to
destroy an already destroyed source, as long as the reference is
valid) and then fully unref it.

  ModemManager[55085]: transaction 0x6 aborted, but message is not abortable
  ModemManager[55085]: [/dev/mhi_MBIM] MBIM error: Cannot write message: Broken pipe

  (ModemManager:55085): GLib-CRITICAL **: 10:04:58.605: g_source_destroy: assertion 'g_atomic_int_get (&source->ref_count) > 0' failed

  Thread 1 "ModemManager" received signal SIGTRAP, Trace/breakpoint trap.
  0x00007ffff775ef23 in g_logv () from /usr/lib/libglib-2.0.so.0
  (gdb) bt
  #0  0x00007ffff775ef23 in g_logv () at /usr/lib/libglib-2.0.so.0
  #1  0x00007ffff775f1a0 in g_log () at /usr/lib/libglib-2.0.so.0
  #2  0x00007ffff7a8f564 in transaction_context_free (ctx=0x5555558254a0) at mbim-device.c:162
  #3  0x00007ffff79240ba in  () at /usr/lib/libgio-2.0.so.0
  #4  0x00007ffff78537b1 in g_object_unref () at /usr/lib/libgobject-2.0.so.0
  #5  0x00007ffff77512b3 in  () at /usr/lib/libglib-2.0.so.0
  #6  0x00007ffff775384a in  () at /usr/lib/libglib-2.0.so.0
  #7  0x00007ffff7756b90 in g_main_context_dispatch () at /usr/lib/libglib-2.0.so.0
  #8  0x00007ffff77a81f9 in  () at /usr/lib/libglib-2.0.so.0
  #9  0x00007ffff7756163 in g_main_loop_run () at /usr/lib/libglib-2.0.so.0
  #10 0x00005555555a0b15 in main (argc=2, argv=0x7fffffffeac8) at main.c:206

(cherry picked from commit 2f870ec4)

All fragments must have the same transaction id.

Signed-off-by: Bjørn Mork <bjorn@mork.no>

If we're using the proxy, we should not try to handle the status of
the MBIM channel or attempt to keep the flow of transaction ids in the
mbim-network program, because that is implicitly done by the proxy
itself.

Because if the context was never created, we would be creating it at
this point, and it wouldn't make sense to do so; e.g.:

  [Debug] [/dev/cdc-wdm1] unexpected port hangup!
  [Debug] [/dev/cdc-wdm1] channel destroyed
  [Debug] Initial device subscribe list...
  [Debug] [service 0] a289cc33-bcbb-8b4f-b6b0-133ec2aae6df (basic-connect)
  [Debug] [service 0] 11 CIDs enabled
  [Debug] [service 0] [cid 0] 2 (subscriber-ready-status)
  [Debug] [service 0] [cid 1] 3 (radio-state)

    [09 ago 2020, 10:15:51] [Debug] [/dev/cdc-wdm1] Received message (translated)...
    >>>>>> Header:
    >>>>>>   length      = 16
    >>>>>>   type        = function-error (0x80000004)
    >>>>>>   transaction = 11
    >>>>>> Contents:
    >>>>>>   error = 'NotOpened' (0x00000005)

    error: couldn't close device: MBIM protocol error: NotOpened[09 ago 2020, 10:15:51] [Debug] [/dev/cdc-wdm1] channel destroyed

We were doing the subscribe list merge correctly, but not reporting it
as done because the 'updated' pointer is set to NULL before returning
it.

This was breaking QMI over MBIM indications, as the registration
message to request them was never sent to the device.

We were printing the 'updated' variable after its contents had already
been stolen.

    [28 Jun 2020, 03:06:27] [Debug] Merged service subscribe list built for device '/dev/cdc-wdm0'
    Thread 1 "mbim-proxy" received signal SIGSEGV, Segmentation fault.
    _mbim_proxy_helper_service_subscribe_list_debug (list=list@entry=0x0, list_size=6) at mbim-proxy-helpers.c:104
    104	        const MbimEventEntry *entry = list[i];
    (gdb) p list
    $1 = (const MbimEventEntry * const *) 0x0
    (gdb) p list_size
    $2 = 6

The segfault was only triggered when the mbim-proxy was run with
--verbose.

The Cinterion mPLS62-w may end up sending messages which are clearly
not well formatted MBIM messages, and upon receiving one of them, the
mbim-proxy would get completely stuck (e.g. attempting to read a
message of >800MB because of the reported length in the first
malformed messae).

Try to avoid this, by doing an initial filter by message type, which
at least solves the seen problem. Not a fully robust fix yet, though.

E.g. new logs:

    [23 jun 2020, 10:43:49] [Debug] [/dev/cdc-wdm1] Sent message (translated)...
    <<<<<< Header:
    <<<<<<   length      = 48
    <<<<<<   type        = command (0x00000003)
    <<<<<<   transaction = 20
    <<<<<< Fragment header:
    <<<<<<   total   = 1
    <<<<<<   current = 0
    <<<<<< Contents:
    <<<<<<   service = 'atds' (5967bdcc-7fd2-49a2-9f5c-b2e70e527db3)
    <<<<<<   cid     = 'location' (0x00000002)
    <<<<<<   type    = 'query' (0x00000000)

    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 293 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 200 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 168 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 311 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 311 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 311 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 335 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 277 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 312 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 314 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 312 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 280 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 312 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 312 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 326 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 260 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 244 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 326 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 326 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 326 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 306 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 334 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 324 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 356 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 356 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 191 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 201 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 356 bytes in MBIM stream as message type validation fails
    [23 jun 2020, 10:43:49] -Warning ** [/dev/cdc-wdm1] discarding 356 bytes in MBIM stream as message type validation fails

Don't add in introspection the methods reading GObject properties.

Equivalent to the glib-mkenums released in 2.66, we just include it in
the repo to be able to use it earlier than requiring 2.66 in the whole
project.

E.g.:
  "since" : "1.0"

Or per-message-type version tags with 'since-ex', e.g.:
  "since-ex" : { "query"        : "1.0",
                 "response"     : "1.0",
                 "notification" : "1.8" },

Fixes bf9a5286