Permalink
Browse files

runtime, syscall: workaround for bug in Linux's execve

Linux's execve has (at the time of writing, and since v2.6.30) a bug when it ran
concurrently with clone, in that it would fail to set up some datastructures if
the thread count before and after some steps differed. This is described better
and in more detail by Colin King in Launchpad¹ and kernel² bugs. When a program
written in Go runtime.Exec's a setuid binary, this issue may cause the resulting
process to not have the expected uid. This patch works around the issue by using
a mutex to serialize exec and clone.

1. https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1672819
2. https://bugzilla.kernel.org/show_bug.cgi?id=195453

Fixes #19546

Change-Id: I126e87d1d9ce3be5ea4ec9c7ffe13f92e087903d
Reviewed-on: https://go-review.googlesource.com/43713
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
  • Loading branch information...
chipaca authored and ianlancetaylor committed May 20, 2017
1 parent 3ca8ee1 commit 91139b87f776a553524b022753981e7909386777
Showing with 28 additions and 0 deletions.
  1. +22 −0 src/runtime/proc.go
  2. +6 −0 src/syscall/exec_unix.go
View
@@ -1615,6 +1615,12 @@ func unlockextra(mp *m) {
atomic.Storeuintptr(&extram, uintptr(unsafe.Pointer(mp)))
}
// execLock serializes exec and clone to avoid bugs or unspecified behaviour
// around exec'ing while creating/destroying threads. See issue #19546.
//
// TODO: look into using a rwmutex, to avoid serializing thread creation.
var execLock mutex
// Create a new m. It will start off with a call to fn, or else the scheduler.
// fn needs to be static and not a heap allocated closure.
// May run with m.p==nil, so write barriers are not allowed.
@@ -1634,10 +1640,14 @@ func newm(fn func(), _p_ *p) {
if msanenabled {
msanwrite(unsafe.Pointer(&ts), unsafe.Sizeof(ts))
}
lock(&execLock)
asmcgocall(_cgo_thread_start, unsafe.Pointer(&ts))
unlock(&execLock)
return
}
lock(&execLock)
newosproc(mp, unsafe.Pointer(mp.g0.stack.hi))
unlock(&execLock)
}
// Stops execution of the current m until new work is available.
@@ -2857,6 +2867,18 @@ func syscall_runtime_AfterForkInChild() {
msigrestore(getg().m.sigmask)
}
// Called from syscall package before Exec.
//go:linkname syscall_runtime_BeforeExec syscall.runtime_BeforeExec
func syscall_runtime_BeforeExec() {
lock(&execLock)
}
// Called from syscall package after Exec.
//go:linkname syscall_runtime_AfterExec syscall.runtime_AfterExec
func syscall_runtime_AfterExec() {
unlock(&execLock)
}
// Allocate a new g, with a stack big enough for stacksize bytes.
func malg(stacksize int32) *g {
newg := new(g)
View
@@ -242,6 +242,10 @@ func StartProcess(argv0 string, argv []string, attr *ProcAttr) (pid int, handle
return pid, 0, err
}
// Implemented in runtime package.
func runtime_BeforeExec()
func runtime_AfterExec()
// Exec invokes the execve(2) system call.
func Exec(argv0 string, argv []string, envv []string) (err error) {
argv0p, err := BytePtrFromString(argv0)
@@ -256,9 +260,11 @@ func Exec(argv0 string, argv []string, envv []string) (err error) {
if err != nil {
return err
}
runtime_BeforeExec()
_, _, err1 := RawSyscall(SYS_EXECVE,
uintptr(unsafe.Pointer(argv0p)),
uintptr(unsafe.Pointer(&argvp[0])),
uintptr(unsafe.Pointer(&envvp[0])))
runtime_AfterExec()
return Errno(err1)
}

0 comments on commit 91139b8

Please sign in to comment.