platform/ptrace: workaround a kernel ptrace issue on ARM64

On ARM64, when ptrace stops on a system call, it uses the x7 register to
indicate whether the stop has been signalled from syscall entry or syscall
exit. This means that we can't get a value of this register and we can't change
it. More details are in the comment for tracehook_report_syscall in
arch/arm64/kernel/ptrace.c.

This happens only if we stop on a system call, so let's queue a signal, resume
a stub thread and catch it on a signal handling.

Fixes: #5238
PiperOrigin-RevId: 352668695

3 files changed, 47 insertions, 4 deletions

diff --git a/pkg/sentry/platform/ptrace/subprocess.go b/pkg/sentry/platform/ptrace/subprocess.go
index aacd7ce70..17fb0a0d8 100644
--- a/pkg/sentry/platform/ptrace/subprocess.go
+++ b/pkg/sentry/platform/ptrace/subprocess.go
@@ -550,6 +550,12 @@ func (s *subprocess) switchToApp(c *context, ac arch.Context) bool {
 		// Wait for the syscall-enter stop.
 		sig := t.wait(stopped)
 
+		if sig == syscall.SIGSTOP {
+			// SIGSTOP was delivered to another thread in the same thread
+			// group, which initiated another group stop. Just ignore it.
+			continue
+		}
+
 		// Refresh all registers.
 		if err := t.getRegs(regs); err != nil {
 			panic(fmt.Sprintf("ptrace get regs failed: %v", err))
@@ -566,13 +572,11 @@ func (s *subprocess) switchToApp(c *context, ac arch.Context) bool {
 
 		// Is it a system call?
 		if sig == (syscallEvent | syscall.SIGTRAP) {
+			s.arm64SyscallWorkaround(t, regs)
+
 			// Ensure registers are sane.
 			updateSyscallRegs(regs)
 			return true
-		} else if sig == syscall.SIGSTOP {
-			// SIGSTOP was delivered to another thread in the same thread
-			// group, which initiated another group stop. Just ignore it.
-			continue
 		}
 
 		// Grab signal information.
diff --git a/pkg/sentry/platform/ptrace/subprocess_amd64.go b/pkg/sentry/platform/ptrace/subprocess_amd64.go
index 020bbda79..04815282b 100644
--- a/pkg/sentry/platform/ptrace/subprocess_amd64.go
+++ b/pkg/sentry/platform/ptrace/subprocess_amd64.go
@@ -257,3 +257,6 @@ func probeSeccomp() bool {
 		}
 	}
 }
+
+func (s *subprocess) arm64SyscallWorkaround(t *thread, regs *arch.Registers) {
+}
diff --git a/pkg/sentry/platform/ptrace/subprocess_arm64.go b/pkg/sentry/platform/ptrace/subprocess_arm64.go
index bd618fae8..416132967 100644
--- a/pkg/sentry/platform/ptrace/subprocess_arm64.go
+++ b/pkg/sentry/platform/ptrace/subprocess_arm64.go
@@ -21,6 +21,7 @@ import (
 	"strings"
 	"syscall"
 
+	"golang.org/x/sys/unix"
 	"gvisor.dev/gvisor/pkg/abi/linux"
 	"gvisor.dev/gvisor/pkg/seccomp"
 	"gvisor.dev/gvisor/pkg/sentry/arch"
@@ -172,3 +173,38 @@ func appendArchSeccompRules(rules []seccomp.RuleSet, defaultAction linux.BPFActi
 func probeSeccomp() bool {
 	return true
 }
+
+func (s *subprocess) arm64SyscallWorkaround(t *thread, regs *arch.Registers) {
+	// On ARM64, when ptrace stops on a system call, it uses the x7
+	// register to indicate whether the stop has been signalled from
+	// syscall entry or syscall exit. This means that we can't get a value
+	// of this register and we can't change it. More details are in the
+	// comment for tracehook_report_syscall in arch/arm64/kernel/ptrace.c.
+	//
+	// This happens only if we stop on a system call, so let's queue a
+	// signal, resume a stub thread and catch it on a signal handling.
+	t.NotifyInterrupt()
+	for {
+		if _, _, errno := syscall.RawSyscall6(
+			syscall.SYS_PTRACE,
+			unix.PTRACE_SYSEMU,
+			uintptr(t.tid), 0, 0, 0, 0); errno != 0 {
+			panic(fmt.Sprintf("ptrace sysemu failed: %v", errno))
+		}
+
+		// Wait for the syscall-enter stop.
+		sig := t.wait(stopped)
+		if sig == syscall.SIGSTOP {
+			// SIGSTOP was delivered to another thread in the same thread
+			// group, which initiated another group stop. Just ignore it.
+			continue
+		}
+		if sig == (syscallEvent | syscall.SIGTRAP) {
+			t.dumpAndPanic(fmt.Sprintf("unexpected syscall event"))
+		}
+		break
+	}
+	if err := t.getRegs(regs); err != nil {
+		panic(fmt.Sprintf("ptrace get regs failed: %v", err))
+	}
+}