1630093 - TRR canary check returns "disable_doh" if the DNS request fails

Status: RESOLVED FIXED

(Firefox :: Security, defect, P1)

Description

[Tim Smith (inactive) 👨‍🔬 [:tdsmith]]

The application DNS canary check in the DoH rollout addon returns disable_doh if the DNS request fails (for example, because the network is disconnected). These are probably inflating our disable rates.

I think the check should only return disable_doh if the check affirmatively fails according to the criteria in https://support.mozilla.org/en-US/kb/canary-domain-use-application-dnsnet.

STR:

1. Visit about:devtools-toolbox?type=extension&id=[e-mail] and run await globalCanary() in the console; result should be enable_doh.
2. Disconnect the network.
3. Repeat. Expected result to be enable_doh; result was disable_doh.

Comment 1

[Andy Grover [:grover]]

needinfo Nihanth and cc Dragana and Valentin.

Comment 2

[(inactive) Nihanth Subramanya [:nhnt11]]

Hmm, we aren't running heuristics until the network goes up. Is there any other situation in which this could result in a false-negative?

Comment 3

[Tim Smith (inactive) 👨‍🔬 [:tdsmith]]

I'm not sure. My concern is that a consistent baseline of 20-30% of heuristics evaluations on all networks report that the canary is disabling DoH, which seems unlikely to me, and this looked like a potential cause. This could be spurious but I'm suspicious of it.

Comment 4

[Johann Hofmann [:johannh]]

Can you set a severity? Thanks!

Comment 5

[(inactive) Nihanth Subramanya [:nhnt11]]

Attachment: Bug 1630093 - Don't run heuristics until internet connectivity has been established. r=valentin!

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/e29a263455f8
Don't run heuristics until internet connectivity has been established. r=valentin

Comment 7

[Razvan Maries]

https://hg.mozilla.org/mozilla-central/rev/e29a263455f8

Comment 8

[Daniel Cicas [:dcicas], Release QA]

Hello,

Im trying to verify this bug. On 79.0b9 if I repeat the steps (As i understand the steps I must be disconnected from the network when running await globalCanary() ) if that's the case I can still reproduce this issue.

Comment 9

[(inactive) Nihanth Subramanya [:nhnt11]]

Daniel, if that code is called when network is disconnected, the canary heuristic will still fail.

The patch in this bug simply tries to ensure that we don't call the code when the network is down.

This situation will be further improved in bug 1654520.

Comment 10

[Daniel Cicas [:dcicas], Release QA]

Hello,

Understood, will remove the qe+ flag from the bug.