compel: set TRACESYSGOOD to distinguish breakpoints from syscalls

When delivering system call traps, set bit 7 in the  signal  number  (i.e.,
deliver SIGTRAP|0x80).  This makes it easy for the tracer  to  distinguish
normal traps from those caused by a system call.

Signed-off-by: Andrei Vagin <avagin@gmail.com>

5 files changed, 27 insertions, 21 deletions

diff --git a/compel/include/ptrace.h b/compel/include/ptrace.h
index bf2701e63..00013f937 100644
--- a/compel/include/ptrace.h
+++ b/compel/include/ptrace.h
@@ -5,6 +5,8 @@
 #include <compel/asm/infect-types.h>
 #include <compel/ptrace.h>
 
+#define PTRACE_SYSCALL_TRAP 0x80
+
 #define PTRACE_SI_EVENT(_si_code) (((_si_code)&0xFFFF) >> 8)
 
 extern int ptrace_get_regs(pid_t pid, user_regs_struct_t *regs);
diff --git a/compel/include/uapi/infect.h b/compel/include/uapi/infect.h
index 7073f343f..19d4da2b1 100644
--- a/compel/include/uapi/infect.h
+++ b/compel/include/uapi/infect.h
@@ -80,9 +80,9 @@ enum trace_flags {
 	TRACE_EXIT,
 };
 
-extern int __must_check compel_stop_on_syscall(int tasks, int sys_nr, int sys_nr_compat, enum trace_flags trace);
+extern int __must_check compel_stop_on_syscall(int tasks, int sys_nr, int sys_nr_compat);
 
-extern int __must_check compel_stop_pie(pid_t pid, void *addr, enum trace_flags *tf, bool no_bp);
+extern int __must_check compel_stop_pie(pid_t pid, void *addr, bool no_bp);
 
 extern int __must_check compel_unmap(struct parasite_ctl *ctl, unsigned long addr);
 
diff --git a/compel/src/lib/infect.c b/compel/src/lib/infect.c
index b99f23b36..7d7865480 100644
--- a/compel/src/lib/infect.c
+++ b/compel/src/lib/infect.c
@@ -304,6 +304,11 @@ try_again:
 		goto try_again;
 	}
 
+	if (ptrace(PTRACE_SETOPTIONS, pid, NULL, PTRACE_O_TRACESYSGOOD)) {
+		pr_perror("Unable to set PTRACE_O_TRACESYSGOOD for %d", pid);
+		return -1;
+	}
+
 	if (ss->seccomp_mode != SECCOMP_MODE_DISABLED && ptrace_suspend_seccomp(pid) < 0)
 		goto err;
 
@@ -1366,7 +1371,6 @@ static int parasite_fini_seized(struct parasite_ctl *ctl)
 	pid_t pid = ctl->rpid;
 	user_regs_struct_t regs;
 	int status, ret = 0;
-	enum trace_flags flag;
 
 	/* stop getting chld from parasite -- we're about to step-by-step it */
 	if (restore_child_handler(ctl))
@@ -1407,11 +1411,11 @@ static int parasite_fini_seized(struct parasite_ctl *ctl)
 		return -1;
 
 	/* Go to sigreturn as closer as we can */
-	ret = compel_stop_pie(pid, ctl->sigreturn_addr, &flag, ctl->ictx.flags & INFECT_NO_BREAKPOINTS);
+	ret = compel_stop_pie(pid, ctl->sigreturn_addr, ctl->ictx.flags & INFECT_NO_BREAKPOINTS);
 	if (ret < 0)
 		return ret;
 
-	if (compel_stop_on_syscall(1, __NR(rt_sigreturn, 0), __NR(rt_sigreturn, 1), flag))
+	if (compel_stop_on_syscall(1, __NR(rt_sigreturn, 0), __NR(rt_sigreturn, 1)))
 		return -1;
 
 	if (ptrace_flush_breakpoints(pid))
@@ -1546,7 +1550,7 @@ int compel_unmap(struct parasite_ctl *ctl, unsigned long addr)
 	if (ret)
 		goto err;
 
-	ret = compel_stop_on_syscall(1, __NR(munmap, 0), __NR(munmap, 1), TRACE_ENTER);
+	ret = compel_stop_on_syscall(1, __NR(munmap, 0), __NR(munmap, 1));
 
 	/*
 	 * Don't touch extended registers here: they were restored
@@ -1558,7 +1562,7 @@ err:
 	return ret;
 }
 
-int compel_stop_pie(pid_t pid, void *addr, enum trace_flags *tf, bool no_bp)
+int compel_stop_pie(pid_t pid, void *addr, bool no_bp)
 {
 	int ret;
 
@@ -1575,7 +1579,6 @@ int compel_stop_pie(pid_t pid, void *addr, enum trace_flags *tf, bool no_bp)
 		 * PIE will stop on a breakpoint, next
 		 * stop after that will be syscall enter.
 		 */
-		*tf = TRACE_EXIT;
 		return 0;
 	}
 
@@ -1588,14 +1591,12 @@ int compel_stop_pie(pid_t pid, void *addr, enum trace_flags *tf, bool no_bp)
 		pr_perror("Unable to restart the %d process", pid);
 		return -1;
 	}
-
-	*tf = TRACE_ENTER;
 	return 0;
 }
 
 static bool task_is_trapped(int status, pid_t pid)
 {
-	if (WIFSTOPPED(status) && WSTOPSIG(status) == SIGTRAP)
+	if (WIFSTOPPED(status) && (WSTOPSIG(status) & ~PTRACE_SYSCALL_TRAP) == SIGTRAP)
 		return true;
 
 	pr_err("Task %d is in unexpected state: %x\n", pid, status);
@@ -1629,15 +1630,13 @@ static inline int is_required_syscall(user_regs_struct_t *regs, pid_t pid, const
  * sys_nr - the required syscall number
  * sys_nr_compat - the required compatible syscall number
  */
-int compel_stop_on_syscall(int tasks, const int sys_nr, const int sys_nr_compat, enum trace_flags trace)
+int compel_stop_on_syscall(int tasks, const int sys_nr, const int sys_nr_compat)
 {
+	enum trace_flags trace = tasks > 1 ? TRACE_ALL : TRACE_ENTER;
 	user_regs_struct_t regs;
 	int status, ret;
 	pid_t pid;
 
-	if (tasks > 1)
-		trace = TRACE_ALL;
-
 	/* Stop all threads on the enter point in sys_rt_sigreturn */
 	while (tasks) {
 		pid = wait4(-1, &status, __WALL, NULL);
@@ -1651,6 +1650,8 @@ int compel_stop_on_syscall(int tasks, const int sys_nr, const int sys_nr_compat,
 
 		pr_debug("%d was trapped\n", pid);
 
+		if ((WSTOPSIG(status) & PTRACE_SYSCALL_TRAP) == 0)
+			goto goon;
 		if (trace == TRACE_EXIT) {
 			trace = TRACE_ENTER;
 			pr_debug("`- Expecting exit\n");
diff --git a/compel/src/lib/ptrace.c b/compel/src/lib/ptrace.c
index 49b685d70..717ee2839 100644
--- a/compel/src/lib/ptrace.c
+++ b/compel/src/lib/ptrace.c
@@ -23,7 +23,7 @@
 
 int ptrace_suspend_seccomp(pid_t pid)
 {
-	if (ptrace(PTRACE_SETOPTIONS, pid, NULL, PTRACE_O_SUSPEND_SECCOMP) < 0) {
+	if (ptrace(PTRACE_SETOPTIONS, pid, NULL, PTRACE_O_SUSPEND_SECCOMP | PTRACE_O_TRACESYSGOOD) < 0) {
 		pr_perror("suspending seccomp failed");
 		return -1;
 	}
diff --git a/criu/cr-restore.c b/criu/cr-restore.c
index 919d10ab5..9a1b23999 100644
--- a/criu/cr-restore.c
+++ b/criu/cr-restore.c
@@ -1964,6 +1964,10 @@ static int attach_to_tasks(bool root_seized)
 				return -1;
 			}
 
+			if (ptrace(PTRACE_SETOPTIONS, pid, NULL, PTRACE_O_TRACESYSGOOD)) {
+				pr_perror("Unable to set PTRACE_O_TRACESYSGOOD for %d", pid);
+				return -1;
+			}
 			/*
 			 * Suspend seccomp if necessary. We need to do this because
 			 * although seccomp is restored at the very end of the
@@ -2028,7 +2032,7 @@ static int restore_rseq_cs(void)
 	return 0;
 }
 
-static int catch_tasks(bool root_seized, enum trace_flags *flag)
+static int catch_tasks(bool root_seized)
 {
 	struct pstree_item *item;
 
@@ -2058,7 +2062,7 @@ static int catch_tasks(bool root_seized, enum trace_flags *flag)
 				return -1;
 			}
 
-			ret = compel_stop_pie(pid, rsti(item)->breakpoint, flag, fault_injected(FI_NO_BREAKPOINTS));
+			ret = compel_stop_pie(pid, rsti(item)->breakpoint, fault_injected(FI_NO_BREAKPOINTS));
 			if (ret < 0)
 				return -1;
 		}
@@ -2225,7 +2229,6 @@ static void reap_zombies(void)
 
 static int restore_root_task(struct pstree_item *init)
 {
-	enum trace_flags flag = TRACE_ALL;
 	int ret, fd, mnt_ns_fd = -1;
 	int root_seized = 0;
 	struct pstree_item *item;
@@ -2440,7 +2443,7 @@ skip_ns_bouncing:
 
 	timing_stop(TIME_RESTORE);
 
-	if (catch_tasks(root_seized, &flag)) {
+	if (catch_tasks(root_seized)) {
 		pr_err("Can't catch all tasks\n");
 		goto out_kill_network_unlocked;
 	}
@@ -2450,7 +2453,7 @@ skip_ns_bouncing:
 
 	__restore_switch_stage(CR_STATE_COMPLETE);
 
-	ret = compel_stop_on_syscall(task_entries->nr_threads, __NR(rt_sigreturn, 0), __NR(rt_sigreturn, 1), flag);
+	ret = compel_stop_on_syscall(task_entries->nr_threads, __NR(rt_sigreturn, 0), __NR(rt_sigreturn, 1));
 	if (ret) {
 		pr_err("Can't stop all tasks on rt_sigreturn\n");
 		goto out_kill_network_unlocked;