Feature #3835
closedResolv::DNS: Retry via TCP on truncated UDP response
Description
=begin
Resolv::DNS has code implementing DNS queries via TCP, but it is not currently used at all. Furthermore, any truncated responses received via UDP are parsed and intact RRs are used, silently discarding any truncated RRs. E.g.:
$ dig amazon.com TXT
;; Truncated, retrying in TCP mode.
; <<>> DiG 9.6.1-P3 <<>> amazon.com TXT
;; global options: +cmd
;; Got answer:
;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 1197
;; flags: qr rd ra; QUERY: 1, ANSWER: 2, AUTHORITY: 0, ADDITIONAL: 0
;; QUESTION SECTION:
;amazon.com. IN TXT
;; ANSWER SECTION:
amazon.com. 3592 IN TXT "spf2.0/pra ip4:207.171.160.0/19 ip4:87.238.80.0/21 ip4:72.21.193.0/24
ip4:72.21.196.0/22 ip4:72.21.208.0/24 ip4:72.21.205.0/24
ip4:72.21.209.0/24 ip4:194.154.193.200/28 ip4:194.7.41.152/28
ip4:212.123.28.40/32 ip4:203.81.17.0/24 ~all"
amazon.com. 3592 IN TXT "v=spf1 ip4:207.171.160.0/19 ip4:87.238.80.0/21 ip4:72.21.193.0/24
ip4:72.21.196.0/22 ip4:72.21.208.0/24 ip4:72.21.205.0/24
ip4:72.21.209.0/24 ip4:194.154.193.200/28 ip4:194.7.41.152/28
ip4:212.123.28.40/32 ip4:203.81.17.0/24 ~all"
;; Query time: 0 msec
;; SERVER: 127.0.0.1#53(127.0.0.1)
;; WHEN: Wed Sep 15 20:21:49 2010
;; MSG SIZE rcvd: 516
With an unpatched 1.9.2 Resolv::DNS:
>> pp Resolv::DNS.new.getresources('amazon.com', Resolv::DNS::Resource::IN::TXT); nil
[#<Resolv::DNS::Resource::IN::TXT:0x0000000143b0b0
@strings=
["spf2.0/pra ip4:207.171.160.0/19 ip4:87.238.80.0/21 ip4:72.21.193.0/24 ip4:72.21.196.0/22
ip4:72.21.208.0/24 ip4:72.21.205.0/24 ip4:72.21.209.0/24 ip4:194.154.193.200/28 ip4:194.7.41.152/28
ip4:212.123.28.40/32 ip4:203.81.17.0/24 ~all"],
@ttl=2185>]
=> nil
The attached patch changes Resolv::DNS#each_resource to check UDP responses for truncation and retries via TCP, which is the proper behavior per RFC 1123, section 6.1.3.2 http://tools.ietf.org/html/rfc1123#page-75 and implemented by most resolver libraries.
Effect:
>> pp Resolv::DNS.new.getresources('amazon.com', Resolv::DNS::Resource::IN::TXT); nil
[#<Resolv::DNS::Resource::IN::TXT:0x0000000214adf0
@strings=
["spf2.0/pra ip4:207.171.160.0/19 ip4:87.238.80.0/21 ip4:72.21.193.0/24 ip4:72.21.196.0/22
ip4:72.21.208.0/24 ip4:72.21.205.0/24 ip4:72.21.209.0/24 ip4:194.154.193.200/28 ip4:194.7.41.152/28
ip4:212.123.28.40/32 ip4:203.81.17.0/24 ~all"],
@ttl=2237>,
#<Resolv::DNS::Resource::IN::TXT:0x0000000214a580
@strings=
["v=spf1 ip4:207.171.160.0/19 ip4:87.238.80.0/21 ip4:72.21.193.0/24 ip4:72.21.196.0/22
ip4:72.21.208.0/24 ip4:72.21.205.0/24 ip4:72.21.209.0/24 ip4:194.154.193.200/28 ip4:194.7.41.152/28
ip4:212.123.28.40/32 ip4:203.81.17.0/24 ~all"],
@ttl=2237>]
=> nil
I'm also attaching a patch for 1.8.7 in the hope that it will be merged into 1.8.7, too.
Of course it would be nice to also support EDNS0, but that's beyond what I can deliver right now.
=end
Files
Updated by julian.mehnle (Julian Mehnle) over 14 years ago
- File resolv-1.9.2-tcp-fallback2.diff resolv-1.9.2-tcp-fallback2.diff added
- File resolv-1.8.7-tcp-fallback2.diff resolv-1.8.7-tcp-fallback2.diff added
=begin
I've discovered a bug in my patch: Resolv::DNS::Message#tc returns a numerical value (0 or 1), not a boolean. Also, let's play it safe and #close the UDP requester before replacing it with the TCP requester.
I've successfully tested these revised patches in a heavy-load production environment that does hundreds of thousands of DNS lookups every day.
=end
Updated by akr (Akira Tanaka) about 14 years ago
- Status changed from Open to Closed
- % Done changed from 0 to 100
=begin
This issue was solved with changeset r29631.
Julian, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.
=end
Updated by julian.mehnle (Julian Mehnle) about 14 years ago
=begin
Oh, thanks for reminding me! I recently discovered another issue with my patch. It fails if the system has more than one resolver configured in /etc/resolv.conf:
$ cat /etc/resolv.conf
192.168.0.123 does not exist, 127.0.0.1 does:¶
nameserver 192.168.0.123
nameserver 127.0.0.1
search mehnle.net
$ irb
require 'resolv'
=> true
require 'patch/resolv'
=> true
dns = Resolv::DNS.new
=> #<Resolv::DNS:0x7fa2e9fe01a8 @config=#<Resolv::DNS::Config:0x7fa2e9fe0130 @initialized=nil, @config_info=nil, @mutex=#Mutex:0x7fa2e9fe00b8>, @initialized=nil, @mutex=#Mutex:0x7fa2e9fe0158>
dns.getresources('amazon.com', Resolv::DNS::Resource::IN::TXT)
Errno::EHOSTUNREACH: No route to host - connect(2)
from /usr/lib/ruby/1.8/resolv.rb:762:ininitialize' from /usr/lib/ruby/1.8/resolv.rb:762:in
new'
from /usr/lib/ruby/1.8/resolv.rb:762:ininitialize' from patch/resolv.rb:61:in
new'
from patch/resolv.rb:61:inmake_tcp_requester' from patch/resolv.rb:29:in
each_resource'
from patch/resolv.rb:131:inresolv' from patch/resolv.rb:129:in
each'
from patch/resolv.rb:129:inresolv' from patch/resolv.rb:128:in
each'
from patch/resolv.rb:128:inresolv' from patch/resolv.rb:126:in
each'
from patch/resolv.rb:126:inresolv' from patch/resolv.rb:15:in
each_resource'
from /usr/lib/ruby/1.8/resolv.rb:468:in `getresources'
from (irb):2
We worked around this by reducing our resolv.conf files to a single nameserver, but obviously this is not a generally acceptable solution.
I'll try to come up with a fix today.
=end