If a disconnection fails (because stop_network() failed), base-bearer
flips the state back to CONNECTED. Oops. At that point something is
clearly messed up, but it seems correct to assume the bearer is
connected.
Nevertheless, we will have already have unhooked the unsolicited events
reporting. A subsequent attempt to disconnect the bearer will trip the
assertion:
#0 0x00007ffff75f2eb5 in __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
#1 0x00007ffff75dd895 in __GI_abort () at abort.c:79
#2 0x00007ffff77beb53 in g_assertion_message
(domain=<optimized out>, file=<optimized out>, line=<optimized out>, func=0x5088e0 <__FUNCTION__.56253> "cleanup_event_report_unsolicited_events", message=<optimized out>) at ../glib/gtestutils.c:2878
#3 0x00007ffff781a96f in g_assertion_message_expr
(domain=domain@entry=0x0, file=file@entry=0x507aad "mm-bearer-qmi.c", line=line@entry=1138, func=func@entry=0x5088e0 <__FUNCTION__.56253> "cleanup_event_report_unsolicited_events", expr=expr@entry=0x507ae5 "*indication_id != 0") at ../glib/gtestutils.c:2904
#4 0x00000000004a0c49 in cleanup_event_report_unsolicited_events (client=<optimized out>, indication_id=0x5bb30c, self=<optimized out>)
at mm-bearer-qmi.c:1138
#5 0x00000000004a0c49 in cleanup_event_report_unsolicited_events
(client=<optimized out>, indication_id=indication_id@entry=0x5bb30c, self=0x5bb420 [MMBearerQmi]) at mm-bearer-qmi.c:1132
#6 0x00000000004a0ee3 in disconnect_context_step (task=0x7fffe8012100 [GTask]) at mm-bearer-qmi.c:1854
#7 0x000000000046e889 in disconnect_auth_ready (self=<optimized out>, res=<optimized out>, ctx=ctx@entry=0x654630)
at mm-iface-modem-simple.c:865
#8 0x00007ffff79cfa9a in g_task_return_now (task=0x7fffe8012640 [GTask]) at ../gio/gtask.c:1209
...
Add checks for indication_id to calls to
cleanup_event_report_unsolicited_events() on
DISCONNECT_STEP_STOP_NETWORK_IPV4 or DISCONNECT_STEP_STOP_NETWORK_IPV6, as
is done elsewhere.
If we cancel connect_cancellable, the connect task will return a
G_IO_ERROR_CANCELLED. We may need to do some cleanup (like, disconnect
if the connection went far enough to actually start the network
connection), thus it's important we check for cancellation correctly.
GVfs' libgvfsdbus.so GIo module, when automatically loaded, attempts to
discover if there's a session D-Bus instance to use. It tries real hard
to get the socket name it would use -- in absence of XDG_RUNTIME_DIR it
decides to make it up with user's home directory. When HOME is unset
too, it just tries to figure it out by looking into /etc/passwd. Which
upsets SELinux that would better not see us looking into it.
We trigger the load of the GIo modules, by using the GFile API to access
the the ports in /dev. They're utterly uesless to us, but there doesn't
seem to be a way to disable their load. Oh well.
For now, let's just ensure HOME is set and the problematic path in glib
is not taken.
Keeps build with GCC 8 happy.
mm-base-call.c:758:18: warning: variable 'response' set but not used [-Wunused-but-set-variable]
mm-base-call.c:822:18: warning: variable 'response' set but not used [-Wunused-but-set-variable]
mm-base-sms.c:908:18: warning: variable 'response' set but not used [-Wunused-but-set-variable]
mm-sms-list.c:331:25: warning: variable 'ctx' set but not used [-Wunused-but-set-variable]
mm-iface-modem-messaging.c:1210:21: warning: variable 'storage_ctx' set but not used [-Wunused-but-set-variable]
huawei/mm-plugin-huawei.c:183:18: warning: variable 'response' set but not used [-Wunused-but-set-variable]
ublox/mm-plugin-ublox.c:161:24: warning: variable 'response' set but not used [-Wunused-but-set-variable]
ublox/mm-plugin-ublox.c:159:24: warning: variable 'ctx' set but not used [-Wunused-but-set-variable]
icera/mm-modem-helpers-icera.c:218:25: warning: variable 'first_free' set but not used [-Wunused-but-set-variable]
novatel/mm-common-novatel.c:50:18: warning: variable 'response' set but not used [-Wunused-but-set-variable]
What's left enabled:
* Access to /dev -- obviously
* CAP_SYS_ADMIN -- this is needed by TIOCSSERIAL only. Too bad this also
allows TIOCSTI, which allows for code injection unless something else
(SELinux) disallows access to ttys with shells.
Maybe kernel should use CAP_SYS_TTY_CONFIG for this.
* socket(AF_NETLINK) -- udev & kernel device changes
* socket(AF_UNIX) -- D-Bus
Otherwise we can accidentally use something that we ought not. Also,
there's deprecation warnings for stuff that's not deprecated in the
version we support yet:
mm-modem-simple.c: In function 'mm_modem_simple_connect_finish':
mm-modem-simple.c:117:5: warning: 'g_simple_async_result_propagate_error' is deprecated [-Wdeprecated-declarations]
if (g_simple_async_result_propagate_error (G_SIMPLE_ASYNC_RESULT (res), error))
^~
We now don't break the strict aliasing rules.
Also, having a compiler flag that disables a warning among the flags that are
meant to add extra sanity checking is not correct either:
--enable-extra-warnings=no would generate a bad aliasin warning while
the --enable-extra-warnings=yes would not.
We shouldn't be accessiing u_int8_t * as u_int16_t *. Let's construct
the 16-bit value by or-ing the 8-bit halves directly; avoiding the
endianness conversion too.
Bluetooth serial devices are weird. They begin life being bound with
RFCOMMCREATEDEV ioctl() and then move around the sysfs tree when they are
connected and disconnected. The connection is estabilished upon the first
open() and torn down upon last close(), their first user virtually being
"owner" of the connection. We don't want to be that process, we're only
interested in actually connected modems. However, currently we have no
knowledge of that and therefore we connect and disconnect multiple times while
probing.
This patch marks unconnected RFCOMM devices as uninteresting to us.
The actual connection and disconnection will be handled by NetworkManager.
For certain devices the name changes with their status. Notably, RFCOMM
devices move from /devices/virtual/ to underneath the HCI that is used
for the connection as the session is estabilished, and return back when
it's torn down.
Lubomir Rintel