Add an option to stop backups from running on metered networks

[ktosiek]

First part of #539 - a backup won't start on a metered connection, but won't stop if the connection status changes later.

[samuel-w]

Because you are not using the tristate properties of QCheckbox, I think you can use the inherited properties of QAbstractButton instead for simplicity.

self.dontRunOnMeteredNetworksCheckBox.setCheckState(QtCore.Qt.CheckState.Checked if profile.dont_run_on_metered_networks else QtCore.Qt.CheckState.Unchecked)
can become
self.dontRunOnMeteredNetworksCheckBox.setChecked(profile.dont_run_on_metered_networks)

Similarly,
profile.dont_run_on_metered_networks = state == QtCore.Qt.Checked
can become
profile.dont_run_on_metered_networks = state
as QtCore.Qt.PartiallyChecked and QtCore.Qt.Checked are True, and QtCore.Qt.Unchecked is False

Additionally, flatpak seems to block system dbus access. You need to add "--system-talk-name=org.freedesktop.NetworkManager", to flatpak/com.borgbase.Vorta.json

Overall it seems to work, the only change needed is the last one.

[Hofer-Julian]

I'll add the flatpak flag as soon as this PR is merged

[m3nu]

Mainly note to self: It would make sense to put macOS/Linux-specific network utility functions as separate modules. Like utils/{linux,darwin}/network{_status}.py.

For macOS we could fake it by looking for "AndroidAP" in the network name. Anyone know the default iOS SSID? Or how Linux detects it?

[ktosiek]

On Linux with NetworkManager users can explicitly mark network as metered, but most of the time NM just guesses. And that guessing involves a few heuristics:

1. It's a Bluetooth or GSM/CDMA connection.
2. The Wi-Fi declares it's metered by setting a Microsoft's Network cost information EID.
3. The DHCP server sends an ANDROID_METERED vendor-specific extension - an Android phone connected by USB will only set this if it's sharing the GSM connection, but not for unmetered Wi-Fi (Android also has a concept of metered Wi-Fi).

Links to NM code:

1. Most of the guessing logic lives in nm_device_update_metered
2. The Network cost information is extracted from Wi-Fi beacon
3. Extracting the ANDROID_METERED

[ktosiek]

I've fixed setCheckState and the missing flatpack flag.

[ktosiek]

There's a fix in my branch that I can't see in the PR - I think GitHub is not feeling well?
Edit: all commits are visible here.

[Hofer-Julian]

I currently have the exact same problem for a different project...

[ktosiek]

I've merged master in to trigger the CI, as it was stuck for a week. It's all green now.

[m3nu]

Can we make the feature more OS-agnostic and make the public functions in src/vorta/network_status.py less Network Manager-specific? Similar to the way we implement the keyring module.

That way the feature doesn't need to be hidden and is always available, but just implemented differently. Until someone looks at the details, it could just look for AndroidAP in the SSID.

[ktosiek]

An indirection layer like in the keyring module makes sense, I'll try adding that.

I'm not sure if we want to invent new concepts on OSes that don't have a "metered network" concept - it sounds confusing. I've just checked 3 different Android devices, and none of them uses AndroidAP as the default SSID.
I've looked into how NM guesses if a network is metered, and none of the rules depend on SSID (see #547 (comment) a few comments earlier).

[m3nu]

Thanks.

Did you see any other pattern in your Android hotspot names? Else macOS can just always return a hardcoded value until we figure something out or they implement it. I'm afraid for now we only have the SSID as data point.

[ktosiek]

From what I've seen the default SSID is a name of the device, series, or manufacturer (like "OnePlus", "Redmi", or "ALCATEL ONE TOUCH POP7"). So that's not very helpful :-C

[m3nu]

I was able to get the ANDROID_METERED DHCP option on macOS:

Normal router:

$ ipconfig getpacket en0
op = BOOTREPLY
htype = 1
flags = 0
hlen = 6
hops = 0
xid = 0x8dc8db4d
secs = 0
ciaddr = 0.0.0.0
yiaddr = 172.16.13.237
siaddr = 0.0.0.0
giaddr = 0.0.0.0
chaddr = 8c:85:90:ad:ee:a3
sname =
file =
options:
Options count is 9
dhcp_message_type (uint8): ACK 0x5
subnet_mask (ip): 255.255.252.0
router (ip_mult): {172.16.12.1}
domain_name_server (ip_mult): {172.16.12.1, 8.8.8.8}
domain_name (string): .
lease_time (uint32): 0xe10
interface_mtu (uint16): 0x5dc
server_identifier (ip): 172.16.12.1
end (none):

My phone (note the last line)

manu@nyx2: ~/Downloads $ ipconfig getpacket en0
op = BOOTREPLY
htype = 1
flags = 0
hlen = 6
hops = 0
xid = 0x8dc8db4e
secs = 0
ciaddr = 0.0.0.0
yiaddr = 192.168.43.223
siaddr = 192.168.43.242
giaddr = 0.0.0.0
chaddr = 8c:85:90:ad:ee:a3
sname =
file =
options:
Options count is 11
dhcp_message_type (uint8): ACK 0x5
server_identifier (ip): 192.168.43.242
lease_time (uint32): 0xe0f
renewal_t1_time_value (uint32): 0x707
rebinding_t2_time_value (uint32): 0xc4d
subnet_mask (ip): 255.255.255.0
broadcast_address (ip): 192.168.43.255
router (ip_mult): {192.168.43.242}
domain_name_server (ip_mult): {192.168.43.242}
vendor_specific (opaque):
0000  41 4e 44 52 4f 49 44 5f  4d 45 54 45 52 45 44     ANDROID_METERED

[m3nu]

• Wrap as OS-independent class/interface (similar to Keyring)
• Add macOS detection for ANDROID_METERED

[ktosiek]

I've moved getting network status into a separate classes: one for OSX, one for NetworkManager, and one dummy class when none of those is available. I've also moved listing WiFi networks into those classes, I hope to implement listing networks for NM shortly.

[ktosiek]

I'm not sure what the test failure is really about, I've seen it happen on ubuntu instead of macos for me: https://github.com/ktosiek/vorta/actions/runs/207355280.

After last changes both Darwin and NetworkManager implementations support guessing if network is metered and listing WiFi networks.

[m3nu]

Works as expected on macOS. Just fixed some issues with string type.

Sometimes tests fail due to some DB race conditions that aren't fully fixed yet. This time they time out, which is new. Will have a closer look.

[m3nu]

Excluded macOS/Darwin tests to not run on Linux. Is that OK? 😜

[m3nu]

Nice to have network functionality abstracted, even though it adds a fair bit of code.

[ktosiek]

The tests in test_darwin.py should be fully isolated from the environment, that's why I've moved calling commands to separate call_* functions. Running those tests on Linux was my only way of testing the new code, so they shouldn't need to be skipped :-)

Similarly with the NM tests - they should work as long as QtDBus is available, but shouldn't need access to NetworkManager, or even a working bus.

[m3nu]

OK. Unskipping.