[Bug 1804487] Re: systemd-resolved has issues when the answer is over 512 bytes with EDNS disabled

[Dan Streetman]

** Bug watch added: github.com/systemd/systemd/issues #10816
   https://github.com/systemd/systemd/issues/10816

** Also affects: systemd via
   https://github.com/systemd/systemd/issues/10816
   Importance: Unknown
       Status: Unknown

-- 
You received this bug notification because you are a member of Ubuntu
Foundations Bugs, which is subscribed to systemd in Ubuntu.
https://bugs.launchpad.net/bugs/1804487

Title:
  systemd-resolved has issues when the answer is over 512 bytes with
  EDNS disabled

Status in systemd:
  Unknown
Status in systemd package in Ubuntu:
  In Progress
Status in systemd source package in Xenial:
  Invalid
Status in systemd source package in Bionic:
  In Progress
Status in systemd source package in Cosmic:
  In Progress
Status in systemd source package in Disco:
  In Progress

Bug description:
  [Impact]

  TCP stub is cutting down the payload to 512 bytes when EDNS is
  disabled. This makes non-EDNS clients (nslookup) receive a "shortened"
  answer even when UDP returns a truncated reply for a new TCP query.
  For instance,

  - If the client supports EDNS:

  $ dig +noall +answer testing.irongiantdesign.com @127.0.0.53 | wc -l
  30

  - If the client does not support EDNS:

  $ dig +noedns +noall +answer testing.irongiantdesign.com @127.0.0.53 | wc -l
  29

  In the second case, no-EDNS, TCP should provide the complete answer,
  but it's capped at UDP's size.

  [Test Case]

  Query systemd-resolved with a domain name that resolves to multiple
  (lots.. 30+) A records. A client with EDNS support (dig) will receive
  all of them, a client without support (nslookup or dig +noedns) will
  have a truncated list. Using the example above:

  EDNS: dig +noall +answer testing.irongiantdesign.com @127.0.0.53 | wc -l
  non-EDNS: dig +noedns +noall +answer testing.irongiantdesign.com @127.0.0.53 | wc -l

  [Regression potential]

  Minimal. This change only affects TCP requests, and the new size is
  already used in the code for other requests.

  [Other Info]

  Upstream bug: https://github.com/systemd/systemd/issues/10816
  Fixed upstream with commit: https://github.com/systemd/systemd/commit/e6eed9445956cfa496e1db933bfd3530db23bfce

  [Original Description]

  Querying a domain name that has >512 bytes in records (e.g. 30+ A
  records), the number of results depends on the DNS client used:

  - If the client supports EDNS:

  $ dig +noall +answer testing.irongiantdesign.com @127.0.0.53 | wc -l
  30

  - If the client does not support EDNS:

  $ dig +noedns +noall +answer testing.irongiantdesign.com @127.0.0.53 | wc -l
  29

  Normally a client that doesn't support EDNS would receive a truncated
  reply from the initial UDP connection (limited by the spec to 512
  bytes) and a second query would be established via TCP to receive the
  complete results. In this case, the number of results is the same
  regardless of the protocol used (29).

  Upstream bug: https://github.com/systemd/systemd/issues/10816

To manage notifications about this bug go to:
https://bugs.launchpad.net/systemd/+bug/1804487/+subscriptions