net: unit test failure

[alberts]

What steps will reproduce the problem?
1. run go tests

What do you see instead?

panic: interface conversion: interface is nil, not *net.TCPAddr

goroutine 215 [running]:
net.selfConnect(0xf8401691b0, 0x3, 0x0, 0x0)
go/workspace/go/src/pkg/net/tcpsock_posix.go:263 +0x5f
net.DialTCP(0x54f26c, 0xf800000003, 0x0, 0xf8400c1d00, 0x539f01, ...)
go/workspace/go/src/pkg/net/tcpsock_posix.go:250 +0x24e
net.dialAddr(0x54f26c, 0x70637400000003, 0xf8400b1e10, 0xf, 0xf840046cc0, ...)
go/workspace/go/src/pkg/net/dial.go:102 +0x20b
net._func_001(0xf8400b1e20, 0xf8400b1e30, 0xf84007ad98, 0xf84007ada0, 0x0, ...)
go/workspace/go/src/pkg/net/dial.go:140 +0x177
created by net.DialTimeout
go/workspace/go/src/pkg/net/dial.go:142 +0x17c

Which compiler are you using (5g, 6g, 8g, gccgo)?

6g

Which operating system are you using?

linux

Which revision are you using?  (hg identify)

tip

Please provide any additional information below.

Log attached.

Attachments:

1. net.txt (41016 bytes)

[dsymonds]

Comment 1:

Tracing the problem:
- selfConnect is getting a *netFD with a nil raddr field.
- that fd is coming from internetSocket, which must be returning no error.
From the other end:
- DialTCP checks for raddr==nil, so raddr.toAddr() must be a non-nil sockaddr.
- internetSocket must call raddr.sockaddr (*TCPAddr.sockaddr), which only returns a
non-nil syscall.Sockaddr with nil error (and so ra is non-nil).
- This means that the problem must be socket returning an fd with a nil raddr even if
its ra argument is non-nil.
- That, in turn, means the toAddr func passed to socket (and internetSocket) must be
doing that.
- This fingers sockaddrToTCP, which will return nil if the syscall.Sockaddr is not one
of {*syscall.SockaddrInet4,*syscall.SockaddrInet6}.
- However, ipToSockaddr never returns a different type than those two, except with an
error.
Here the trail runs cold. I can't see what is going wrong without a way to reproduce
this and scattering some logging throughout. All I can think of is that it slipped in
with 002f08814913 (Russ' "net: avoid TCP self-connect"), so assigning it to him for
further diagnosis.

Labels changed: added priority-later, removed priority-triage.

Owner changed to @rsc.

[rsc]

Comment 2:

Labels changed: added priority-go1, removed priority-later.

[alberts]

Comment 3:

It took a long time, but running the test repeatedly eventually reproduced this.
#!/bin/bash
set -xe
rm -f out err net.test
go test -c net
while true; do
for procs in 1 2 8 10 50 100; do
GOMAXPROCS=$procs time ./net.test -test.v >out 2>err || exit 1
echo $?
done
done

[robpike]

Comment 4:

I notice that the Linux ARM builder is reproducibly giving the same failure, every time
it seems. for instance,
http://build.golang.org/log/5fd6fdde31381ed018fbaa55af7132cc72fbda8b
goroutine 236 [running]:
net.selfConnect(0x107ab9a0, 0x3, 0x0)
    /mnt/work2/tmp/gobuilder/linux-arm-minux-2c15f43065c1/go/src/pkg/net/tcpsock_posix.go:263 +0x84
net.DialTCP(0x15ad68, 0x3, 0x0, 0x1079c6d0, 0x10732801, ...)
    /mnt/work2/tmp/gobuilder/linux-arm-minux-2c15f43065c1/go/src/pkg/net/tcpsock_posix.go:250 +0x218
net.dialAddr(0x15ad68, 0x3, 0x107805f0, 0xf, 0x10732860, ...)
    /mnt/work2/tmp/gobuilder/linux-arm-minux-2c15f43065c1/go/src/pkg/net/dial.go:102 +0x1fc
net._func_001(0x107838a8, 0x107838b0, 0x107838b8, 0x107838c0, 0x0, ...)
    /mnt/work2/tmp/gobuilder/linux-arm-minux-2c15f43065c1/go/src/pkg/net/dial.go:140 +0x1ac
created by net.DialTimeout
    /mnt/work2/tmp/gobuilder/linux-arm-minux-2c15f43065c1/go/src/pkg/net/dial.go:142 +0x194

[robpike]

Comment 5:

Labels changed: added go1-must.

[robpike]

Comment 6:

This issue was closed by revision 152d806.

Status changed to Fixed.