Bug #17658: DNS resolution failure with multiple named resolvers

• Status: Closed
• Assignee: -
• Target version: -
• ruby -v: ruby 2.7.2p137 (2020-10-01 revision 5445e04352) [x86_64-linux]
• Backport: 2.6: DONE, 2.7: DONE, 3.0: DONTNEED

[mcarpenter (Martin Carpenter)]

Description

I created a Resolv::DNS resolver with two nameservers described using their domain names. Calling #getresources on this for an A or NS (and possibly other resource types) of domain example.com returns an empty list (no results, no error). I expected it to return the corresponding resource record (IP of example.com).

Instead if the Resolv::DNS instance is created with one resolver domain name or two resolvers' IP addresses then correct results are obtained. The error occurs only when more than two nameservers [or possibly more] are specified using their domain names.

Testcases

I used two well-known public DNS servers to test (they have funky but legitimate TLDs):

• dns.google (8.8.4.4, 8.8.8.8)
• one.one.one.one (1.1.1.1, 1.0.0.1)

✗ Failing: two nameservers by name

require 'resolv'
Resolv::DNS.new({nameserver: ['dns.google', 'one.one.one.one']}).getresources('example.com', Resolv::DNS::Resource::IN::A)
=> []

✓ Successful: one nameserver by name

(either as a single-item list or a string)

Resolv::DNS.new({nameserver: 'dns.google'}).getresources('example.com', Resolv::DNS::Resource::IN::A)
=> [#<Resolv::DNS::Resource::IN::A:0x000055ee95df9ad8 @address=#<Resolv::IPv4 93.184.216.34>, @ttl=16840>]
Resolv::DNS.new({nameserver: ['one.one.one.one']}).getresources('example.com', Resolv::DNS::Resource::IN::A)
=> [#<Resolv::DNS::Resource::IN::A:0x000055ee95517c80 @address=#<Resolv::IPv4 93.184.216.34>, @ttl=77160>]

✓ Successful: one nameserver by IP

(either as a single-item list or a string)

Resolv::DNS.new({nameserver: ['8.8.4.4']}).getresources('example.com', Resolv::DNS::Resource::IN::A) 
=> [#<Resolv::DNS::Resource::IN::A:0x000055ee95e28d38 @address=#<Resolv::IPv4 93.184.216.34>, @ttl=20931>]
Resolv::DNS.new({nameserver: '1.1.1.1'}).getresources('example.com', Resolv::DNS::Resource::IN::A)
=> [#<Resolv::DNS::Resource::IN::A:0x000055ee95e41360 @address=#<Resolv::IPv4 93.184.216.34>, @ttl=66081>]

✓ Successful: two nameservers by IP

Resolv::DNS.new({nameserver: ['8.8.4.4', '1.1.1.1']}).getresources('example.com', Resolv::DNS::Resource::IN::A) 
=> [#<Resolv::DNS::Resource::IN::A:0x000055ee95d62ea8 @address=#<Resolv::IPv4 93.184.216.34>, @ttl=20894>]

A little analysis

I didn't get to the bottom of this but this may help:

• When only one nameserver is used then a Resolv::DNS::Requester::ConnectedUDP::Sender is used; when two nameservers are used an UnconnectedUDP::Sender is used.
• The code appears to be timing out and then retried even though correct requests are made and responses received.
• In particular, if I packet trace 53/udp then the conversation is exactly as I would expect and near identical in both failing and successful cases. That is:
  • Request A record for first nameserver from local resolver
  • Receive IP of first nameserver
  • Request A record of example.com from first nameserver
  • Receive IP of example.com
    In the successful case, the conversation ends here. In the failing case, the code round-robins between the two nameservers until it finally returns [].

Test environment

Tested ruby 2.5.0p0 on Red Hat 6 and rubies 2.7.1p83, 2.7.2p137 on Ubuntu 20 from two different networks with identical results.

[mcarpenter (Martin Carpenter)]

I tested ruby 3.0.0 and it works fine so I guess something got fixed:

Resolv::DNS.new({nameserver: ['dns.google', 'one.one.one.one']}).getresources('example.com', Resolv::DNS::Resource::IN::A) 
=> [#<Resolv::DNS::Resource::IN::A:0x00005630d981e268 @address=#<Resolv::IPv4 93.184.216.34>, @ttl=17981>]

(ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-linux])

[mcarpenter (Martin Carpenter)]

Fixing commit is below.

#12838 states that the problem is intermittent and due to duplicate responses. That's not what I have observed: I can consistently reproduce the problem. Can this one-line change + test please be back-ported to 2.x?

commit 9682db065158da5fa4ec8a3bc267da45b429b92c
Author: Jeremy Evans <[e-mail]>
Date:   Fri Sep 11 14:28:20 2020 -0700

    Remove sender/message_id pair after response received in resolv
    
    Once a response for a given DNS request has been received (which
    requires a matching message id), the [sender, message_id] pair
    should be removed from the list of valid senders.  This makes it
    so duplicate responses from the same sender are ignored.
    
    Fixes [Bug #12838]


diff --git a/lib/resolv.rb b/lib/resolv.rb
index d78531e174..d50940ad15 100644
--- a/lib/resolv.rb
+++ b/lib/resolv.rb
@@ -706,7 +706,7 @@ def request(sender, tout)
       end
 
       def sender_for(addr, msg)
-        @senders[[addr,msg.id]]
+        @senders.delete([addr,msg.id])
       end
 
       def close
[snip]

[jeremyevans0 (Jeremy Evans)]

• Status changed from Open to Closed
• Backport changed from 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN to 2.6: REQUIRED, 2.7: REQUIRED, 3.0: DONTNEED

[nagachika (Tomoyuki Chikanaga)]

• Backport changed from 2.6: REQUIRED, 2.7: REQUIRED, 3.0: DONTNEED to 2.6: REQUIRED, 2.7: DONE, 3.0: DONTNEED

ruby_2_7 70c3a195f39763dccdf9367d0c9b7e815431a41a merged revision(s) 9682db065158da5fa4ec8a3bc267da45b429b92c.

[usa (Usaku NAKAMURA)]

• Backport changed from 2.6: REQUIRED, 2.7: DONE, 3.0: DONTNEED to 2.6: DONE, 2.7: DONE, 3.0: DONTNEED

ruby_2_6 r67929 merged revision(s) 9682db065158da5fa4ec8a3bc267da45b429b92c.

[sam.saffron (Sam Saffron)]

This has caused a pretty giant regression:

https://bugs.ruby-lang.org/issues/17781

Requests are now being never cleaned up cause sender_for as a side effect causes a bypass on cleanup.