os/signal: Reset doesn't reset ignored signals

[perillo]

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

$ go version
go version go1.16.4 linux/amd64

Does this issue reproduce with the latest release?

Yes, including go1.17 devel

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

go env Output

$ go env
GO111MODULE="on"
GOARCH="amd64"
GOBIN="/home/manlio/.local/bin"
GOCACHE="/home/manlio/.cache/go-build"
GOENV="/home/manlio/.config/go/env"
GOEXE=""
GOFLAGS=""
GOHOSTARCH="amd64"
GOHOSTOS="linux"
GOINSECURE="*.local"
GOMODCACHE="/home/manlio/.local/lib/go/pkg/mod"
GONOPROXY=""
GONOSUMDB="*.local"
GOOS="linux"
GOPATH="/home/manlio/.local/lib/go:/home/manlio/src/go"
GOPRIVATE=""
GOPROXY="https://proxy.golang.org,direct"
GOROOT="/usr/lib/go"
GOSUMDB="sum.golang.org"
GOTMPDIR=""
GOTOOLDIR="/usr/lib/go/pkg/tool/linux_amd64"
GOVCS=""
GOVERSION="go1.16.4"
GCCGO="gccgo"
AR="ar"
CC="gcc"
CXX="g++"
CGO_ENABLED="1"
GOMOD="/dev/null"
CGO_CFLAGS="-g -O2"
CGO_CPPFLAGS=""
CGO_CXXFLAGS="-g -O2"
CGO_FFLAGS="-g -O2"
CGO_LDFLAGS="-g -O2"
PKG_CONFIG="pkg-config"
GOGCCFLAGS="-fPIC -m64 -pthread -fmessage-length=0 -fdebug-prefix-map=/tmp/go-build3881295983=/tmp/go-build -gno-record-gcc-switches"
GOROOT/bin/go version: go version go1.16.4 linux/amd64
GOROOT/bin/go tool compile -V: compile version go1.16.4
uname -sr: Linux 5.12.5-arch1-1
/usr/lib/libc.so.6: GNU C Library (GNU libc) release release version 2.33.
gdb --version: GNU gdb (GDB) 10.2

What did you do?

https://play.golang.org/p/vYlTzluIuBL

What did you expect to see?

The last line of the output being false

What did you see instead?

The last line is true.

See https://groups.google.com/g/golang-nuts/c/W5NNNC0h5Zc for a discussion.
Thanks to Brian Candler for the simplified example.

[perillo]

https://golang.org/cl/3580 described signal.Reset as "ability to restore initial signal handlers".

[mknyszek]

CC @ianlancetaylor via https://dev.golang.org/owners

To be clear, Reset says it specifically undoes the effects of Notify (https://golang.org/pkg/os/signal/#Reset) but it does not mention Ignore. But it does also say "all signal handlers are reset," so I see your point. I worry that maybe this has been working this way long enough that we can't actually change it at this point.

[ianlancetaylor]

At a quick look Reset already undoes a call to Ignore, except that it fails to adjust the data checked by Ignored. If I'm right that just seems like a bug, with no serious compatibility issues.

[bluec0re]

According to my tests (and a real life bug) this is not just with syscall.Ignored:

https://play.golang.org/p/NQlbXjitE17

The last run of sleepAndKill should have a similar output to the first run. I can't use time.Now in the playground, otherwise a measured delay would show ~2s for the last 2 executions instead of only for the one SIGINT is ignored.

EDIT: this is particular bad as ignored signals are inherited by child processes, so unmasking them before forking the child doesn't work. This can lead to quite unexpected behavior, especially if default signals like SIGCHLD are masked.

[gopherbot]

Change https://go.dev/cl/649155 mentions this issue: os/signal: Let Reset reset ignored signals

[fmeum]

I've been bitten by this a number of times now, which motivated me to look into it more. As @bluec0re mentioned, there appears to be now way to recover from a signal.Ignore(syscall.SIGCHLD) to the point where exec.Cmd will no longer be able to start subprocesses. Other signals, once ignored, will also be forwarded to child processes as ignored with no way to recover from that state (excluding cgo).

Would a fix for this be accepted, perhaps with an associated GODEBUG flag?

[ianlancetaylor]

Patches are always welcome. I don't see why this needs a GODEBUG flag. Thanks.

I see that you sent a patch which is currently marked as "work in progress". You can send it to me when it is ready for review. Thanks.