runtime: always keep global reference to mp until mexit completes

Ms are allocated via standard heap allocation (`new(m)`), which means we
must keep them alive (i.e., reachable by the GC) until we are completely
done using them.

Ms are primarily reachable through runtime.allm. However, runtime.mexit
drops the M from allm fairly early, long before it is done using the M
structure. If that was the last reference to the M, it is now at risk of
being freed by the GC and used for some other allocation, leading to
memory corruption.

Ms with a Go-allocated stack coincidentally already keep a reference to
the M in sched.freem, so that the stack can be freed lazily. This
reference has the side effect of keeping this Ms reachable. However, Ms
with an OS stack skip this and are at risk of corruption.

Fix this lifetime by extending sched.freem use to all Ms, with the value
of mp.freeWait determining whether the stack needs to be freed or not.

Fixes #56243.

Change-Id: Ic0c01684775f5646970df507111c9abaac0ba52e
Reviewed-on: https://go-review.googlesource.com/c/go/+/443716
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Michael Pratt <mpratt@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>

src/runtime/os3_solaris.go

@@ -7,6 +7,7 @@ package runtime
 import (
 	"internal/abi"
 	"internal/goarch"
+	"runtime/internal/atomic"
 	"unsafe"
 )
 
@@ -182,7 +183,7 @@ func newosproc(mp *m) {
 	}
 }
 
-func exitThread(wait *uint32) {
+func exitThread(wait *atomic.Uint32) {
 	// We should never reach exitThread on Solaris because we let
 	// libc clean up threads.
 	throw("exitThread")

src/runtime/os_aix.go

@@ -8,6 +8,7 @@ package runtime
 
 import (
 	"internal/abi"
+	"runtime/internal/atomic"
 	"unsafe"
 )
 
@@ -233,7 +234,7 @@ func newosproc(mp *m) {
 
 }
 
-func exitThread(wait *uint32) {
+func exitThread(wait *atomic.Uint32) {
 	// We should never reach exitThread on AIX because we let
 	// libc clean up threads.
 	throw("exitThread")

src/runtime/os_js.go

@@ -7,6 +7,7 @@
 package runtime
 
 import (
+	"runtime/internal/atomic"
 	"unsafe"
 )
 
@@ -35,7 +36,7 @@ func usleep_no_g(usec uint32) {
 	usleep(usec)
 }
 
-func exitThread(wait *uint32)
+func exitThread(wait *atomic.Uint32)
 
 type mOS struct{}
 

src/runtime/os_openbsd_syscall2.go

@@ -7,6 +7,7 @@
 package runtime
 
 import (
+	"runtime/internal/atomic"
 	"unsafe"
 )
 
@@ -49,11 +50,11 @@ func open(name *byte, mode, perm int32) int32
 // return value is only set on linux to be used in osinit()
 func madvise(addr unsafe.Pointer, n uintptr, flags int32) int32
 
-// exitThread terminates the current thread, writing *wait = 0 when
+// exitThread terminates the current thread, writing *wait = freeMStack when
 // the stack is safe to reclaim.
 //
 //go:noescape
-func exitThread(wait *uint32)
+func exitThread(wait *atomic.Uint32)
 
 //go:noescape
 func obsdsigprocmask(how int32, new sigset) sigset

src/runtime/os_plan9.go

@@ -468,7 +468,7 @@ func newosproc(mp *m) {
 	}
 }
 
-func exitThread(wait *uint32) {
+func exitThread(wait *atomic.Uint32) {
 	// We should never reach exitThread on Plan 9 because we let
 	// the OS clean up threads.
 	throw("exitThread")

src/runtime/os_windows.go

@@ -941,7 +941,7 @@ func newosproc0(mp *m, stk unsafe.Pointer) {
 	throw("bad newosproc0")
 }
 
-func exitThread(wait *uint32) {
+func exitThread(wait *atomic.Uint32) {
 	// We should never reach exitThread on Windows because we let
 	// the OS clean up threads.
 	throw("exitThread")

src/runtime/proc.go

@@ -1582,19 +1582,18 @@ func mexit(osStack bool) {
 	}
 	throw("m not found in allm")
 found:
-	if !osStack {
+	// Delay reaping m until it's done with the stack.
-		// Delay reaping m until it's done with the stack.
+	//
-		//
+	// Put mp on the free list, though it will not be reaped while freeWait
-		// If this is using an OS stack, the OS will free it
+	// is freeMWait. mp is no longer reachable via allm, so even if it is
-		// so there's no need for reaping.
+	// on an OS stack, we must keep a reference to mp alive so that the GC
-		atomic.Store(&mp.freeWait, 1)
+	// doesn't free mp while we are still using it.
-		// Put m on the free list, though it will not be reaped until
+	//
-		// freeWait is 0. Note that the free list must not be linked
+	// Note that the free list must not be linked through alllink because
-		// through alllink because some functions walk allm without
+	// some functions walk allm without locking, so may be using alllink.
-		// locking, so may be using alllink.
+	mp.freeWait.Store(freeMWait)
-		mp.freelink = sched.freem
+	mp.freelink = sched.freem
-		sched.freem = mp
+	sched.freem = mp
-	}
 	unlock(&sched.lock)
 
 	atomic.Xadd64(&ncgocall, int64(mp.ncgocall))
@@ -1624,6 +1623,9 @@ found:
 	mdestroy(mp)
 
 	if osStack {
+		// No more uses of mp, so it is safe to drop the reference.
+		mp.freeWait.Store(freeMRef)
+
 		// Return from mstart and let the system thread
 		// library free the g0 stack and terminate the thread.
 		return
@@ -1795,19 +1797,25 @@ func allocm(pp *p, fn func(), id int64) *m {
 		lock(&sched.lock)
 		var newList *m
 		for freem := sched.freem; freem != nil; {
-			if freem.freeWait != 0 {
+			wait := freem.freeWait.Load()
+			if wait == freeMWait {
 				next := freem.freelink
 				freem.freelink = newList
 				newList = freem
 				freem = next
 				continue
 			}
-			// stackfree must be on the system stack, but allocm is
+			// Free the stack if needed. For freeMRef, there is
-			// reachable off the system stack transitively from
+			// nothing to do except drop freem from the sched.freem
-			// startm.
+			// list.
-			systemstack(func() {
+			if wait == freeMStack {
-				stackfree(freem.g0.stack)
+				// stackfree must be on the system stack, but allocm is
-			})
+				// reachable off the system stack transitively from
+				// startm.
+				systemstack(func() {
+					stackfree(freem.g0.stack)
+				})
+			}
 			freem = freem.freelink
 		}
 		sched.freem = newList

src/runtime/runtime2.go

@@ -516,6 +516,13 @@ const (
 	tlsSize  = tlsSlots * goarch.PtrSize
 )
 
+// Values for m.freeWait.
+const (
+	freeMStack = 0  // M done, free stack and reference.
+	freeMRef   = 1  // M done, free reference.
+	freeMWait  = 2  // M still in use.
+)
+
 type m struct {
 	g0      *g     // goroutine with scheduling stack
 	morebuf gobuf  // gobuf arg to morestack
@@ -547,7 +554,7 @@ type m struct {
 	printlock     int8
 	incgo         bool   // m is executing a cgo call
 	isextra       bool   // m is an extra m
-	freeWait      uint32 // if == 0, safe to free g0 and delete m (atomic)
+	freeWait      atomic.Uint32 // Whether it is safe to free g0 and delete m (one of freeMRef, freeMStack, freeMWait)
 	fastrand      uint64
 	needextram    bool
 	traceback     uint8

src/runtime/stubs2.go

@@ -6,7 +6,10 @@
 
 package runtime
 
-import "unsafe"
+import (
+	"runtime/internal/atomic"
+	"unsafe"
+)
 
 // read calls the read system call.
 // It returns a non-negative number of bytes written or a negative errno value.
@@ -34,8 +37,8 @@ func open(name *byte, mode, perm int32) int32
 // return value is only set on linux to be used in osinit()
 func madvise(addr unsafe.Pointer, n uintptr, flags int32) int32
 
-// exitThread terminates the current thread, writing *wait = 0 when
+// exitThread terminates the current thread, writing *wait = freeMStack when
 // the stack is safe to reclaim.
 //
 //go:noescape
-func exitThread(wait *uint32)
+func exitThread(wait *atomic.Uint32)

src/runtime/sys_darwin.go

@@ -6,6 +6,7 @@ package runtime
 
 import (
 	"internal/abi"
+	"runtime/internal/atomic"
 	"unsafe"
 )
 
@@ -474,7 +475,7 @@ func pthread_cond_signal(c *pthreadcond) int32 {
 func pthread_cond_signal_trampoline()
 
 // Not used on Darwin, but must be defined.
-func exitThread(wait *uint32) {
+func exitThread(wait *atomic.Uint32) {
 }
 
 //go:nosplit

src/runtime/sys_dragonfly_amd64.s

@@ -65,7 +65,7 @@ TEXT runtime·exit(SB),NOSPLIT,$-8
 	MOVL	$0xf1, 0xf1  // crash
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-8
 	MOVQ	wait+0(FP), AX
 	// We're done using the stack.

src/runtime/sys_freebsd_386.s

@@ -99,7 +99,7 @@ GLOBL exitStack<>(SB),RODATA,$8
 DATA exitStack<>+0x00(SB)/4, $0
 DATA exitStack<>+0x04(SB)/4, $0
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-4
 	MOVL	wait+0(FP), AX
 	// We're done using the stack.

src/runtime/sys_freebsd_amd64.s

@@ -96,7 +96,7 @@ TEXT runtime·exit(SB),NOSPLIT,$-8
 	MOVL	$0xf1, 0xf1  // crash
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-8
 	MOVQ	wait+0(FP), AX
 	// We're done using the stack.

src/runtime/sys_freebsd_arm.s

@@ -85,7 +85,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0
 	MOVW.CS R8, (R8)
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-4
 	MOVW	wait+0(FP), R0
 	// We're done using the stack.

src/runtime/sys_freebsd_arm64.s

@@ -99,7 +99,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	MOVD	$0, R0
 	MOVD	R0, (R0)
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOVD	wait+0(FP), R0
 	// We're done using the stack.

src/runtime/sys_freebsd_riscv64.s

@@ -96,7 +96,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	ECALL
 	WORD	$0	// crash
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOV	wait+0(FP), A0
 	// We're done using the stack.

src/runtime/sys_linux_386.s

@@ -72,7 +72,7 @@ TEXT exit1<>(SB),NOSPLIT,$0
 	INT $3	// not reached
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-4
 	MOVL	wait+0(FP), AX
 	// We're done using the stack.

src/runtime/sys_linux_amd64.s

@@ -54,7 +54,7 @@ TEXT runtime·exit(SB),NOSPLIT,$0-4
 	SYSCALL
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-8
 	MOVQ	wait+0(FP), AX
 	// We're done using the stack.

src/runtime/sys_linux_arm.s

@@ -117,7 +117,7 @@ TEXT exit1<>(SB),NOSPLIT|NOFRAME,$0
 	MOVW	$1003, R1
 	MOVW	R0, (R1)	// fail hard
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-4
 	MOVW	wait+0(FP), R0
 	// We're done using the stack.

src/runtime/sys_linux_arm64.s

@@ -56,7 +56,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	SVC
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOVD	wait+0(FP), R0
 	// We're done using the stack.

src/runtime/sys_linux_loong64.s

@@ -49,7 +49,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	SYSCALL
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOVV	wait+0(FP), R19
 	// We're done using the stack.

src/runtime/sys_linux_mips64x.s

@@ -51,7 +51,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	SYSCALL
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOVV	wait+0(FP), R1
 	// We're done using the stack.

src/runtime/sys_linux_mipsx.s

@@ -50,7 +50,7 @@ TEXT runtime·exit(SB),NOSPLIT,$0-4
 	UNDEF
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-4
 	MOVW	wait+0(FP), R1
 	// We're done using the stack.

src/runtime/sys_linux_ppc64x.s

@@ -49,7 +49,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	SYSCALL	$SYS_exit_group
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOVD	wait+0(FP), R1
 	// We're done using the stack.

src/runtime/sys_linux_riscv64.s

@@ -57,7 +57,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	ECALL
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOV	wait+0(FP), A0
 	// We're done using the stack.

src/runtime/sys_linux_s390x.s

@@ -46,7 +46,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0-4
 	SYSCALL
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT|NOFRAME,$0-8
 	MOVD	wait+0(FP), R1
 	// We're done using the stack.

src/runtime/sys_netbsd_386.s

@@ -53,7 +53,7 @@ TEXT runtime·exit(SB),NOSPLIT,$-4
 	MOVL	$0xf1, 0xf1		// crash
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-4
 	MOVL	wait+0(FP), AX
 	// We're done using the stack.

src/runtime/sys_netbsd_amd64.s

@@ -122,7 +122,7 @@ TEXT runtime·exit(SB),NOSPLIT,$-8
 	MOVL	$0xf1, 0xf1		// crash
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-8
 	MOVQ	wait+0(FP), AX
 	// We're done using the stack.

src/runtime/sys_netbsd_arm.s

@@ -56,7 +56,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0
 	MOVW.CS R8, (R8)
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-4
 	MOVW wait+0(FP), R0
 	// We're done using the stack.

src/runtime/sys_netbsd_arm64.s

@@ -115,7 +115,7 @@ TEXT runtime·exit(SB),NOSPLIT,$-8
 	MOVD	$0, R0			// If we're still running,
 	MOVD	R0, (R0)		// crash
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0-8
 	MOVD	wait+0(FP), R0
 	// We're done using the stack.

src/runtime/sys_openbsd2.go

@@ -8,6 +8,7 @@ package runtime
 
 import (
 	"internal/abi"
+	"runtime/internal/atomic"
 	"unsafe"
 )
 
@@ -248,7 +249,7 @@ func sigaltstack(new *stackt, old *stackt) {
 func sigaltstack_trampoline()
 
 // Not used on OpenBSD, but must be defined.
-func exitThread(wait *uint32) {
+func exitThread(wait *atomic.Uint32) {
 }
 
 //go:nosplit

src/runtime/sys_openbsd_mips64.s

@@ -24,7 +24,7 @@ TEXT runtime·exit(SB),NOSPLIT|NOFRAME,$0
 	MOVV	R2, (R2)
 	RET
 
-// func exitThread(wait *uint32)
+// func exitThread(wait *atomic.Uint32)
 TEXT runtime·exitThread(SB),NOSPLIT,$0
 	MOVV	wait+0(FP), R4		// arg 1 - notdead
 	MOVV	$302, R2		// sys___threxit