net: dnsclient_unix_test.go will fail on a machine with 'test' in /etc/hosts and 'files' first in nsswitch.conf

[siebenmann]

What version of Go are you using (go version)?

This issue reproduces when building the current git tip version of go, b9f6b22. Because it blocks finishing the build, I cannot show you literal 'go version' output on the affected machine. Testing says that this was introduced in commit bfc164c, 'net: add Resolver.StrictErrors', intended to fix issue #17448.

What operating system and processor architecture are you using (go env)?

GOARCH="amd64"
GOBIN=""
GOEXE=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOOS="linux"

This also reproduces on FreeBSD.

What did you do?

Configure a machine with an /etc/hosts entry for the name test, such as:

199.199.199.199   test

Then set /etc/nsswitch.conf to have hosts: files dns, so that /etc/hosts is checked before DNS. Then attempt to build go from source.

What did you expect to see?

Build succeeds.

What did you see instead?

The build fails in dnsclient_unix_test.go where a whole bunch of DNS tests expect to get various errors or results when looking up the name 'test' but instead get the test entry from /etc/hosts.

--- FAIL: TestStrictErrorsLookupIP (0.00s)
        dnsclient_unix_test.go:1112: #0 (No errors) strict=true: got ips map[199.199.199.199:{}]; want map[2001:db8::1:{} 192.0.2.1:{}]
        dnsclient_unix_test.go:1112: #0 (No errors) strict=false: got ips map[199.199.199.199:{}]; want map[192.0.2.1:{} 2001:db8::1:{}]
        dnsclient_unix_test.go:1098: #1 (searchX error fails in strict mode) strict=true: got err <nil>; want &net.DNSError{Err:"i/o timeout", Name:"test", Server:"192.0.2.53:53", IsTimeout:true, IsTemporary:false}
        dnsclient_unix_test.go:1112: #1 (searchX error fails in strict mode) strict=true: got ips map[199.199.199.199:{}]; want map[]
        dnsclient_unix_test.go:1112: #1 (searchX error fails in strict mode) strict=false: got ips map[199.199.199.199:{}]; want map[192.0.2.1:{} 2001:db8::1:{}]
        dnsclient_unix_test.go:1098: #2 (searchX IPv4-only timeout fails in strict mode) strict=true: got err <nil>; want &net.DNSError{Err:"i/o timeout", Name:"test", Server:"192.0.2.53:53", IsTimeout:true, IsTemporary:false}
        dnsclient_unix_test.go:1112: #2 (searchX IPv4-only timeout fails in strict mode) strict=true: got ips map[199.199.199.199:{}]; want map[]
        dnsclient_unix_test.go:1112: #2 (searchX IPv4-only timeout fails in strict mode) strict=false: got ips map[199.199.199.199:{}]; want map[192.0.2.1:{} 2001:db8::1:{}]
        dnsclient_unix_test.go:1098: #3 (searchX IPv6-only servfail fails in strict mode) strict=true: got err <nil>; want &net.DNSError{Err:"server misbehaving", Name:"test", Server:"192.0.2.53:53", IsTimeout:false, IsTemporary:true}
        dnsclient_unix_test.go:1112: #3 (searchX IPv6-only servfail fails in strict mode) strict=true: got ips map[199.199.199.199:{}]; want map[]
        dnsclient_unix_test.go:1112: #3 (searchX IPv6-only servfail fails in strict mode) strict=false: got ips map[199.199.199.199:{}]; want map[192.0.2.1:{} 2001:db8::1:{}]
        dnsclient_unix_test.go:1098: #4 (searchY error always fails) strict=true: got err <nil>; want &net.DNSError{Err:"i/o timeout", Name:"test", Server:"192.0.2.53:53", IsTimeout:true, IsTemporary:false}
        dnsclient_unix_test.go:1112: #4 (searchY error always fails) strict=true: got ips map[199.199.199.199:{}]; want map[]
        dnsclient_unix_test.go:1098: #4 (searchY error always fails) strict=false: got err <nil>; want &net.DNSError{Err:"no such host", Name:"test", Server:"192.0.2.53:53", IsTimeout:false, IsTemporary:false}
        dnsclient_unix_test.go:1112: #4 (searchY error always fails) strict=false: got ips map[199.199.199.199:{}]; want map[]
        dnsclient_unix_test.go:1098: #5 (searchY IPv4-only socket error fails in strict mode) strict=true: got err <nil>; want &net.DNSError{Err:"write: socket on fire", Name:"test", Server:"192.0.2.53:53", IsTimeout:false, IsTemporary:true}
        dnsclient_unix_test.go:1112: #5 (searchY IPv4-only socket error fails in strict mode) strict=true: got ips map[199.199.199.199:{}]; want map[]
        dnsclient_unix_test.go:1112: #5 (searchY IPv4-only socket error fails in strict mode) strict=false: got ips map[199.199.199.199:{}]; want map[2001:db8::1:{}]
        dnsclient_unix_test.go:1098: #6 (searchY IPv6-only timeout fails in strict mode) strict=true: got err <nil>; want &net.DNSError{Err:"i/o timeout", Name:"test", Server:"192.0.2.53:53", IsTimeout:true, IsTemporary:false}
        dnsclient_unix_test.go:1112: #6 (searchY IPv6-only timeout fails in strict mode) strict=true: got ips map[199.199.199.199:{}]; want map[]
        dnsclient_unix_test.go:1112: #6 (searchY IPv6-only timeout fails in strict mode) strict=false: got ips map[199.199.199.199:{}]; want map[192.0.2.1:{}]
FAIL
FAIL    net     1.533s

[mdempsky]

Out of curiosity, why do you have a test entry in your /etc/hosts? Is that just a personal config you have? Is there some software you're using that uses that? Were you just wanting to demonstrate an issue with the package net tests?

[siebenmann]

I originally encountered this issue on a (FreeBSD) machine run by another group here at the university. Their habit is to have a full /etc/hosts where entries for hosts in their subdomain(s) have an entry for host's bare hostname (eg IP machine.subdomain machine). And they have a machine called test.<subdomain>, so they wind up with an /etc/hosts entry of IP test.subdomain test.

I tested this on my Linux machine with an artificial /etc/hosts entry to make sure I understood the issue and could confidently reproduce it in an environment where I could make changes to the system to test any questions or proposed fixes (I don't have root access to that other group's machines).

[mdempsky]

@siebenmann Can you confirm whether the fix in https://go-review.googlesource.com/c/38300/ works for you?

[gopherbot]

CL https://golang.org/cl/38300 mentions this issue.

[siebenmann]

@mdempsky I hand-applied the change to my source and this appears to completely fix the issue; I get no build failure here on the FreeBSD server with the original problem.