syscall: Clearenv is racy on Plan9

[fhs]

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

$ go version
go version devel +master Tue Oct 22 13:09:23 EDT 2019 plan9/386

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

go env Output

$ go env
GO111MODULE='on'
GOARCH='386'
GOBIN=''
GOCACHE='/usr/fhs/lib/cache/go-build'
GOENV='/usr/fhs/lib/go/env'
GOEXE=''
GOFLAGS=''
GOHOSTARCH='386'
GOHOSTOS='plan9'
GONOPROXY=''
GONOSUMDB=''
GOOS='plan9'
GOPATH='/home/fhs/go'
GOPRIVATE=''
GOPROXY='https://proxy.golang.org,direct'
GOROOT='/home/big/go'
GOSUMDB='sum.golang.org'
GOTMPDIR=''
GOTOOLDIR='/home/big/go/pkg/tool/plan9_386'
GCCGO='gccgo'
GO386='sse2'
AR='ar'
CC='8c'
CXX='g++'
CGO_ENABLED='0'
GOMOD='/home/big/go/src/go.mod'
CGO_CFLAGS='-g -O2'
CGO_CPPFLAGS=''
CGO_CXXFLAGS='-g -O2'
CGO_FFLAGS='-g -O2'
CGO_LDFLAGS='-g -O2'
PKG_CONFIG='pkg-config'
GOGCCFLAGS='-fPIC -m32 -fno-caret-diagnostics -Qunused-arguments -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build454216994=/tmp/go-build -gno-record-gcc-switches'

What did you do?

cpu% cd src/os
cpu% go test -run Clearenv -count 100000

What did you expect to see?

No test failure

What did you see instead?

--- FAIL: TestClearenv (0.02s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.02s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.02s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.02s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.01s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.01s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:149: Clearenv() didn't clear $GO_TEST_CLEARENV, remained with value "1"
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
--- FAIL: TestClearenv (0.00s)
    env_test.go:145: Setenv("GO_TEST_CLEARENV", "1") didn't set $GO_TEST_CLEARENV
FAIL
exit status: 'os.test 25721: 1'
FAIL	os	145.298s

Existence of a race in env tests was originally reported in issue #25234, but it incorrectly assumed it was due to shared /env. Instead, the race seems to be in the Plan 9 kernel, between rfork(RFCENVG) and access to /env.

I propose replacing rfork(RFCENVG) (which creates a new env group) with removal of all environment variables in the current env group. I'll send a CL soon.

@gopherbot Please add label OS-Plan9, NeedsFix

[gopherbot]

Change https://golang.org/cl/202657 mentions this issue: syscall: fix Clearenv race on Plan 9