Changed DisableTechnology to do a full device disable.

Changed DisableTechnology to do a full device disable.
This causes a disconnect to occur on all devices with the specified technology, including teardown of ip state. This fixes issue 1821. I added a call to set_carrier() in __connman_device_disable. Testing included the following scenarios, running test-manager and route -n after each one to check that things were in the expected state. Note that there is a separate, pre-existing bug in which a technology can be shown in the list of "connected technologies" even when the technology is disabled. I will fix that in a separate CL.

- Start up connman with wifi disabled. Enable wifi, connect to GoogleGuest, then disable wifi.
- Start up connman with wifi enabled. Connect to GoogleGuest. Disable wifi, then re-enable it and connect to GoogleGuest.
- Do the above two tests with 3G (except that the service connected to is not GoogleGuest).
- With wifi connected, establish a 3G connection. Then disable wifi and make sure that all the expected routing table entries exist. Enable wifi and connect to GoogleGuest, and make sure everything goes back to the state it was in before wifi was disabled.
- The above test, but with Ethernet taking the place of wifi, and wifi taking the place of 3G.
- With wifi and 3G connected, plug in the Ethernet. Then disable wifi. Then unplug the Ethernet. Check that routing is set up for 3G, and load a web page.

[Sam Leffler, 2010-03-16 19:25:46]

Please also indicate the testing done.

http://codereview.chromium.org/992006/diff/1/2
File files/src/device.c (right):

http://codereview.chromium.org/992006/diff/1/2#newcode1289
files/src/device.c:1289: return set_powered(device, FALSE);
I think we want to use set_powered in __connman_device_disable instead of
calling it here.  But we need to review all the work done as a result (e.g.
__connman_profile_remove_device in set_carrier).

[Eric Shienbrood, 2010-03-17 22:26:54]

The testing I did is described in the revised CL description.

http://codereview.chromium.org/992006/diff/1/2
File files/src/device.c (right):

http://codereview.chromium.org/992006/diff/1/2#newcode1289
files/src/device.c:1289: return set_powered(device, FALSE);
On 2010/03/16 19:25:46, Sam Leffler wrote:
> I think we want to use set_powered in __connman_device_disable instead of
> calling it here.  But we need to review all the work done as a result (e.g.
> __connman_profile_remove_device in set_carrier).
I moved it to __connman_device_disable, as that was the right thing to do. But I
also changed it to call set_carrier() instead of set_powered(), since
device_disable() does mostly the same work as set_powered(), except for the
set_carrier() part (which is what does the IP teardown). As far as I can tell,
we do want all the work that set_carrier() does, including
__connman_profile_remove_device(), since we are now treating DisableTechnology
the same as turning off power to the devices.

[Sam Leffler, 2010-03-17 23:47:34]

LGTM after you do one more test for me: disable and reboot and verify the state
is recorded in the profile (i.e. so the technology is still disabled)--and also
verify enable works across reboot.

Also my original comment about set_powered was intended to mean you can replace
most of the body of the _disable method with this so we consolidate all the
logic in one place.  But since you've already done all the testing w/ the
existing change we should defer that to a separate CL so we can get this fix in.

[Eric Shienbrood, 2010-03-18 00:19:16]

OK, I ran the two tests you requested, and everything checks out. You're right
that we should be looking for opportunities to eliminate duplicate code. For
this bug, I was just trying to make a minimally disruptive change to fix the
problem.