From c61d322d5f9e3fcffa4c523892af432dca030c12 Mon Sep 17 00:00:00 2001 From: Russ Cox Date: Thu, 22 Dec 2022 09:44:56 -0500 Subject: [PATCH] runtime: call __fork instead of fork on darwin Issues #33565 and #56784 were caused by hangs in the child process after fork, while it ran atfork handlers that ran into slow paths that didn't work in the child. CL 451735 worked around those two issues by calling a couple functions at startup to try to warm up those child paths. That mostly worked, but it broke programs using cgo with certain macOS frameworks (#57263). CL 459175 reverted CL 451735. This CL introduces a different fix: bypass the atfork child handlers entirely. For a general fork call where the child and parent are both meant to keep executing the original program, atfork handlers can be necessary to fix any state that would otherwise be tied to the parent process. But Go only uses fork as preparation for exec, and it takes care to limit what it attempts to do in the child between the fork and exec. In particular it doesn't use any of the things that the macOS atfork handlers are trying to fix up (malloc, xpc, others). So we can use the low-level fork system call (__fork) instead of the atfork-wrapped one. The full list of functions that can be called in a child after fork in exec_libc2.go is: - ptrace - setsid - setpgid - getpid - ioctl - chroot - setgroups - setgid - setuid - chdir - dup2 - fcntl - close - execve - write - exit I disassembled all of these while attached to a hung exec.test binary and confirmed that nearly all of them are making direct kernel calls, not using anything that the atfork handler needs to fix up. The exceptions are ioctl, fcntl, and exit. The ioctl and fcntl implementations do some extra work around the kernel call but don't call any other functions, so they should still be OK. (If not, we could use __ioctl and __fcntl instead, but without a good reason, we should keep using the standard entry points.) The exit implementation calls atexit handlers. That is almost certainly inappropriate in a failed fork child, so this CL changes that call to __exit on darwin. To avoid making unnecessary changes at this point in the release cycle, this CL leaves OpenBSD calling plain exit, even though that is probably a bug in the OpenBSD port (filed #57446). Fixes #33565. Fixes #56784. Fixes #57263. Change-Id: I26812c26a72bdd7fcf72ec41899ba11cf6b9c4ab Reviewed-on: https://go-review.googlesource.com/c/go/+/459176 Reviewed-by: David Chase Reviewed-by: Cherry Mui TryBot-Result: Gopher Robot Run-TryBot: Russ Cox --- src/syscall/exec_libc2.go | 4 ++-- src/syscall/syscall_darwin.go | 32 ++++++++++++++++++++++++++-- src/syscall/syscall_openbsd_libc.go | 6 +++++- src/syscall/zsyscall_darwin_amd64.go | 22 +++++++++++++++---- src/syscall/zsyscall_darwin_amd64.s | 6 ++++-- src/syscall/zsyscall_darwin_arm64.go | 22 +++++++++++++++---- src/syscall/zsyscall_darwin_arm64.s | 6 ++++-- 7 files changed, 81 insertions(+), 17 deletions(-) diff --git a/src/syscall/exec_libc2.go b/src/syscall/exec_libc2.go index 41bc79a721e..6e3c2bf9d76 100644 --- a/src/syscall/exec_libc2.go +++ b/src/syscall/exec_libc2.go @@ -78,7 +78,7 @@ func forkAndExecInChild(argv0 *byte, argv, envv []*byte, chroot, dir *byte, attr // About to call fork. // No more allocation or calls of non-assembly functions. runtime_BeforeFork() - r1, _, err1 = rawSyscall(abi.FuncPCABI0(libc_fork_trampoline), 0, 0, 0) + r1, _, err1 = rawSyscall(forkTrampoline, 0, 0, 0) if err1 != 0 { runtime_AfterFork() return 0, err1 @@ -276,6 +276,6 @@ childerror: // send error code on pipe rawSyscall(abi.FuncPCABI0(libc_write_trampoline), uintptr(pipe), uintptr(unsafe.Pointer(&err1)), unsafe.Sizeof(err1)) for { - rawSyscall(abi.FuncPCABI0(libc_exit_trampoline), 253, 0, 0) + rawSyscall(exitTrampoline, 253, 0, 0) } } diff --git a/src/syscall/syscall_darwin.go b/src/syscall/syscall_darwin.go index a39e99dc633..5ec311962a5 100644 --- a/src/syscall/syscall_darwin.go +++ b/src/syscall/syscall_darwin.go @@ -22,7 +22,34 @@ func Syscall6(trap, a1, a2, a3, a4, a5, a6 uintptr) (r1, r2 uintptr, err Errno) func RawSyscall(trap, a1, a2, a3 uintptr) (r1, r2 uintptr, err Errno) func RawSyscall6(trap, a1, a2, a3, a4, a5, a6 uintptr) (r1, r2 uintptr, err Errno) -var dupTrampoline = abi.FuncPCABI0(libc_dup2_trampoline) +// These are called from exec_libc2.go in the child of fork. +// The names differ between macOS and OpenBSD, so we need +// to declare the specific ones used here to keep the exec_libc2.go +// code portable. +// +// We use __fork and __exit, not fork and exit, to avoid the libc atfork +// and atexit handlers. The atfork handlers have caused fork child +// hangs in the past (see #33565, #56784). The atexit handlers have +// not, but the non-libc ports all invoke the system call, so doing +// the same here makes sense. In general we wouldn't expect +// atexit handlers to work terribly well in a fork child anyway. +// (Also, perhaps the atfork handlers clear the atexit handlers, +// in which case we definitely need to avoid calling the libc exit +// if we bypass the libc fork.) +// +// Other calls that are made in the child after the fork are +// ptrace, setsid, setpgid, getpid, ioctl, chroot, setgroups, +// setgid, setuid, chdir, dup2, fcntl, close, execve, and write. +// Those are all simple kernel wrappers that should be safe +// to be called directly. The fcntl and ioctl functions do run +// some code around the kernel call, but they don't call any +// other functions, so for now we keep using them instead of +// calling the lower-level __fcntl and __ioctl functions. +var ( + dupTrampoline = abi.FuncPCABI0(libc_dup2_trampoline) + exitTrampoline = abi.FuncPCABI0(libc___exit_trampoline) + forkTrampoline = abi.FuncPCABI0(libc___fork_trampoline) +) type SockaddrDatalink struct { Len uint8 @@ -210,11 +237,12 @@ func Kill(pid int, signum Signal) (err error) { return kill(pid, int(signum), 1) //sys writev(fd int, iovecs []Iovec) (cnt uintptr, err error) //sys mmap(addr uintptr, length uintptr, prot int, flag int, fd int, pos int64) (ret uintptr, err error) //sys munmap(addr uintptr, length uintptr) (err error) -//sysnb fork() (pid int, err error) +//sysnb __fork() (pid int, err error) //sysnb ioctl(fd int, req int, arg int) (err error) //sysnb ioctlPtr(fd int, req uint, arg unsafe.Pointer) (err error) = SYS_ioctl //sysnb execve(path *byte, argv **byte, envp **byte) (err error) //sysnb exit(res int) (err error) +//sysnb __exit(res int) (err error) //sys sysctl(mib []_C_int, old *byte, oldlen *uintptr, new *byte, newlen uintptr) (err error) //sys fcntlPtr(fd int, cmd int, arg unsafe.Pointer) (val int, err error) = SYS_fcntl //sys unlinkat(fd int, path string, flags int) (err error) diff --git a/src/syscall/syscall_openbsd_libc.go b/src/syscall/syscall_openbsd_libc.go index 516d02975c5..6358a9a390a 100644 --- a/src/syscall/syscall_openbsd_libc.go +++ b/src/syscall/syscall_openbsd_libc.go @@ -10,7 +10,11 @@ import ( "internal/abi" ) -var dupTrampoline = abi.FuncPCABI0(libc_dup3_trampoline) +var ( + dupTrampoline = abi.FuncPCABI0(libc_dup3_trampoline) + exitTrampoline = abi.FuncPCABI0(libc_exit_trampoline) + forkTrampoline = abi.FuncPCABI0(libc_fork_trampoline) +) func init() { execveOpenBSD = execve diff --git a/src/syscall/zsyscall_darwin_amd64.go b/src/syscall/zsyscall_darwin_amd64.go index 6b3fff3f372..5e3f6cccf00 100644 --- a/src/syscall/zsyscall_darwin_amd64.go +++ b/src/syscall/zsyscall_darwin_amd64.go @@ -1734,8 +1734,8 @@ func libc_munmap_trampoline() // THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT -func fork() (pid int, err error) { - r0, _, e1 := rawSyscall(abi.FuncPCABI0(libc_fork_trampoline), 0, 0, 0) +func __fork() (pid int, err error) { + r0, _, e1 := rawSyscall(abi.FuncPCABI0(libc___fork_trampoline), 0, 0, 0) pid = int(r0) if e1 != 0 { err = errnoErr(e1) @@ -1743,9 +1743,9 @@ func fork() (pid int, err error) { return } -func libc_fork_trampoline() +func libc___fork_trampoline() -//go:cgo_import_dynamic libc_fork fork "/usr/lib/libSystem.B.dylib" +//go:cgo_import_dynamic libc___fork __fork "/usr/lib/libSystem.B.dylib" // THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT @@ -1801,6 +1801,20 @@ func libc_exit_trampoline() // THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT +func __exit(res int) (err error) { + _, _, e1 := rawSyscall(abi.FuncPCABI0(libc___exit_trampoline), uintptr(res), 0, 0) + if e1 != 0 { + err = errnoErr(e1) + } + return +} + +func libc___exit_trampoline() + +//go:cgo_import_dynamic libc___exit __exit "/usr/lib/libSystem.B.dylib" + +// THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT + func sysctl(mib []_C_int, old *byte, oldlen *uintptr, new *byte, newlen uintptr) (err error) { var _p0 unsafe.Pointer if len(mib) > 0 { diff --git a/src/syscall/zsyscall_darwin_amd64.s b/src/syscall/zsyscall_darwin_amd64.s index 90e51fb9a4f..cbb4496a50e 100644 --- a/src/syscall/zsyscall_darwin_amd64.s +++ b/src/syscall/zsyscall_darwin_amd64.s @@ -221,14 +221,16 @@ TEXT ·libc_mmap_trampoline(SB),NOSPLIT,$0-0 JMP libc_mmap(SB) TEXT ·libc_munmap_trampoline(SB),NOSPLIT,$0-0 JMP libc_munmap(SB) -TEXT ·libc_fork_trampoline(SB),NOSPLIT,$0-0 - JMP libc_fork(SB) +TEXT ·libc___fork_trampoline(SB),NOSPLIT,$0-0 + JMP libc___fork(SB) TEXT ·libc_ioctl_trampoline(SB),NOSPLIT,$0-0 JMP libc_ioctl(SB) TEXT ·libc_execve_trampoline(SB),NOSPLIT,$0-0 JMP libc_execve(SB) TEXT ·libc_exit_trampoline(SB),NOSPLIT,$0-0 JMP libc_exit(SB) +TEXT ·libc___exit_trampoline(SB),NOSPLIT,$0-0 + JMP libc___exit(SB) TEXT ·libc_sysctl_trampoline(SB),NOSPLIT,$0-0 JMP libc_sysctl(SB) TEXT ·libc_unlinkat_trampoline(SB),NOSPLIT,$0-0 diff --git a/src/syscall/zsyscall_darwin_arm64.go b/src/syscall/zsyscall_darwin_arm64.go index 61601449a04..d4c56be8188 100644 --- a/src/syscall/zsyscall_darwin_arm64.go +++ b/src/syscall/zsyscall_darwin_arm64.go @@ -1734,8 +1734,8 @@ func libc_munmap_trampoline() // THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT -func fork() (pid int, err error) { - r0, _, e1 := rawSyscall(abi.FuncPCABI0(libc_fork_trampoline), 0, 0, 0) +func __fork() (pid int, err error) { + r0, _, e1 := rawSyscall(abi.FuncPCABI0(libc___fork_trampoline), 0, 0, 0) pid = int(r0) if e1 != 0 { err = errnoErr(e1) @@ -1743,9 +1743,9 @@ func fork() (pid int, err error) { return } -func libc_fork_trampoline() +func libc___fork_trampoline() -//go:cgo_import_dynamic libc_fork fork "/usr/lib/libSystem.B.dylib" +//go:cgo_import_dynamic libc___fork __fork "/usr/lib/libSystem.B.dylib" // THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT @@ -1801,6 +1801,20 @@ func libc_exit_trampoline() // THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT +func __exit(res int) (err error) { + _, _, e1 := rawSyscall(abi.FuncPCABI0(libc___exit_trampoline), uintptr(res), 0, 0) + if e1 != 0 { + err = errnoErr(e1) + } + return +} + +func libc___exit_trampoline() + +//go:cgo_import_dynamic libc___exit __exit "/usr/lib/libSystem.B.dylib" + +// THIS FILE IS GENERATED BY THE COMMAND AT THE TOP; DO NOT EDIT + func sysctl(mib []_C_int, old *byte, oldlen *uintptr, new *byte, newlen uintptr) (err error) { var _p0 unsafe.Pointer if len(mib) > 0 { diff --git a/src/syscall/zsyscall_darwin_arm64.s b/src/syscall/zsyscall_darwin_arm64.s index f00747939ef..ce1850e0aef 100644 --- a/src/syscall/zsyscall_darwin_arm64.s +++ b/src/syscall/zsyscall_darwin_arm64.s @@ -221,14 +221,16 @@ TEXT ·libc_mmap_trampoline(SB),NOSPLIT,$0-0 JMP libc_mmap(SB) TEXT ·libc_munmap_trampoline(SB),NOSPLIT,$0-0 JMP libc_munmap(SB) -TEXT ·libc_fork_trampoline(SB),NOSPLIT,$0-0 - JMP libc_fork(SB) +TEXT ·libc___fork_trampoline(SB),NOSPLIT,$0-0 + JMP libc___fork(SB) TEXT ·libc_ioctl_trampoline(SB),NOSPLIT,$0-0 JMP libc_ioctl(SB) TEXT ·libc_execve_trampoline(SB),NOSPLIT,$0-0 JMP libc_execve(SB) TEXT ·libc_exit_trampoline(SB),NOSPLIT,$0-0 JMP libc_exit(SB) +TEXT ·libc___exit_trampoline(SB),NOSPLIT,$0-0 + JMP libc___exit(SB) TEXT ·libc_sysctl_trampoline(SB),NOSPLIT,$0-0 JMP libc_sysctl(SB) TEXT ·libc_unlinkat_trampoline(SB),NOSPLIT,$0-0