x/sys/unix: MmapPtr upsets checkptr

[Jille]

Go version

go version go1.23.2 linux/amd64

Output of go env in your module/workspace:

GO111MODULE=''
GOARCH='amd64'
GOBIN=''
GOCACHE='/home/jille/.cache/go-build'
GOENV='/home/jille/.config/go/env'
GOEXE=''
GOEXPERIMENT=''
GOFLAGS=''
GOHOSTARCH='amd64'
GOHOSTOS='linux'
GOINSECURE=''
GOMODCACHE='/home/jille/go/pkg/mod'
GONOPROXY='src.hexon.nl'
GONOSUMDB='src.hexon.nl'
GOOS='linux'
GOPATH='/home/jille/go'
GOPRIVATE='src.hexon.nl'
GOPROXY='https://proxy.golang.org,direct'
GOROOT='/usr/lib/go-1.23'
GOSUMDB='sum.golang.org'
GOTMPDIR=''
GOTOOLCHAIN='auto'
GOTOOLDIR='/usr/lib/go-1.23/pkg/tool/linux_amd64'
GOVCS=''
GOVERSION='go1.23.2'
GODEBUG=''
GOTELEMETRY='local'
GOTELEMETRYDIR='/home/jille/.config/go/telemetry'
GCCGO='gccgo'
GOAMD64='v1'
AR='ar'
CC='gcc'
CXX='g++'
CGO_ENABLED='1'
GOMOD='/home/jille/src/ilsa/go.mod'
GOWORK=''
CGO_CFLAGS='-O2 -g'
CGO_CPPFLAGS=''
CGO_CXXFLAGS='-O2 -g'
CGO_FFLAGS='-O2 -g'
CGO_LDFLAGS='-O2 -g'
PKG_CONFIG='pkg-config'
GOGCCFLAGS='-fPIC -m64 -pthread -Wl,--no-gc-sections -fmessage-length=0 -ffile-prefix-map=/tmp/go-build4165437972=/tmp/go-build -gno-record-gcc-switches'

What did you do?

I'm using unix.MmapPtr() with unix.MAP_FIXED, which triggers checkptr (for example when run with -race).

https://go.dev/play/p/3raMoMpk-YA (build with -race to trigger the error)

What did you see happen?

fatal error: checkptr: pointer arithmetic result points to invalid allocation

goroutine 1 gp=0xc0000061c0 m=0 mp=0x5d2f80 [running]:
runtime.throw({0x50bb48?, 0xc0000da000?})
	/usr/lib/go-1.23/src/runtime/panic.go:1067 +0x48 fp=0xc0000a3eb8 sp=0xc0000a3e88 pc=0x49e608
runtime.checkptrArithmetic(0xc0000da000?, {0x0, 0x0, 0x32?})
	/usr/lib/go-1.23/src/runtime/checkptr.go:69 +0x9c fp=0xc0000a3ee8 sp=0xc0000a3eb8 pc=0x43e0fc
golang.org/x/sys/unix.MmapPtr(...)
	/home/jille/go/pkg/mod/golang.org/x/sys@v0.26.0/unix/syscall_unix.go:159
main.main()
	/home/jille/src/ilsa/testmmap/testmmap.go:14 +0x125 fp=0xc0000a3f50 sp=0xc0000a3ee8 pc=0x4e0a65
runtime.main()
	/usr/lib/go-1.23/src/runtime/proc.go:272 +0x28b fp=0xc0000a3fe0 sp=0xc0000a3f50 pc=0x46ca0b
runtime.goexit({})
	/usr/lib/go-1.23/src/runtime/asm_amd64.s:1700 +0x1 fp=0xc0000a3fe8 sp=0xc0000a3fe0 pc=0x4a5221

What did you expect to see?

This is the implementation of MmapPtr. The compiler can't prove this conversion from uinptr to unsafe.Pointer is safe.

func MmapPtr(fd int, offset int64, addr unsafe.Pointer, length uintptr, prot int, flags int) (ret unsafe.Pointer, err error) {
	xaddr, err := mapper.mmap(uintptr(addr), length, prot, flags, fd, offset)
	return unsafe.Pointer(xaddr), err
}

[gabyhelp]

Related Issues

• runtime: cgoCheckPointer fatal error: can't happen #70016 (closed)
• cmd/compile: fatal error: checkptr: pointer arithmetic computed bad pointer value #64467
• cmd/compile: -d=checkptr should not reject unaligned pointers to non-pointer data #37298 (closed)
• go build use -asan it panic. #69225 (closed)
• runtime/checkptr: fatal error: checkptr: pointer arithmetic result points to invalid allocation #68415 (closed)
• x/sys/unix: TestMmap crashes on openbsd/386 #21962 (closed)
• runtime: invalid pointer found on stack regression in Go1.15 triggered by mapassign in init #41816 (closed)
• proposal: cmd/go: consider adding unsafeptr to go vet during test list #57794 (closed)
• runtime: checkptr incorrectly -race flagging when using &^ arithmetic #40917 (closed)
• cmd/compile: -race does not obey go:nocheckptr #42880

_{(Emoji vote if this was helpful or unhelpful; more detailed feedback welcome in this discussion.)}

[randall77]

This is a correct report. You cannot remap the Go stack or heap.

You'd have to either pass nil as the addr or something previously obtained from one of the mmap functions.

[Jille]

Here is the full scope of my crime: https://pkg.go.dev/github.com/Jille/gcmmap

I agree that people shouldn't remap the Go heap or stack and expect the Go compatibility promise to apply to them.

However, if someone got here it seems unlikely they did it by accident. If they're doing something wrong with the unsafe.Pointer they're passing in, checkptr should catch it before the call. Once mmap returns I think people likely are getting what they asked for.

[randall77]

If they're doing something wrong with the unsafe.Pointer they're passing in, checkptr should catch it before the call.

I'm not sure that's possible. It is ok by the unsafe.Pointer rules to make one with, e.g., unsafe.Pointer(&b[0]). The only thing not allowed here is to mmap it. How would we catch that before the mmap call?

Once mmap returns I think people likely are getting what they asked for.

But mmap hasn't returned. The checkptr error happens inside mmap.

[gopherbot]

Timed out in state WaitingForInfo. Closing.

(I am just a bot, though. Please speak up if this is a mistake or you have the requested information.)