x86/ptrace: Fix the TIF_FORCED_TF logic in handle_signal()

When the TIF_SINGLESTEP tracee dequeues a signal,
handle_signal() clears TIF_FORCED_TF and X86_EFLAGS_TF but
leaves TIF_SINGLESTEP set.

If the tracer does PTRACE_SINGLESTEP again, enable_single_step()
sets X86_EFLAGS_TF but not TIF_FORCED_TF.  This means that the
subsequent PTRACE_CONT doesn't not clear X86_EFLAGS_TF, and the
tracee gets the wrong SIGTRAP.

Test-case (needs -O2 to avoid prologue insns in signal handler):

	#include <unistd.h>
	#include <stdio.h>
	#include <sys/ptrace.h>
	#include <sys/wait.h>
	#include <sys/user.h>
	#include <assert.h>
	#include <stddef.h>

	void handler(int n)
	{
		asm("nop");
	}

	int child(void)
	{
		assert(ptrace(PTRACE_TRACEME, 0,0,0) == 0);
		signal(SIGALRM, handler);
		kill(getpid(), SIGALRM);
		return 0x23;
	}

	void *getip(int pid)
	{
		return (void*)ptrace(PTRACE_PEEKUSER, pid,
					offsetof(struct user, regs.rip), 0);
	}

	int main(void)
	{
		int pid, status;

		pid = fork();
		if (!pid)
			return child();

		assert(wait(&status) == pid);
		assert(WIFSTOPPED(status) && WSTOPSIG(status) == SIGALRM);

		assert(ptrace(PTRACE_SINGLESTEP, pid, 0, SIGALRM) == 0);
		assert(wait(&status) == pid);
		assert(WIFSTOPPED(status) && WSTOPSIG(status) == SIGTRAP);
		assert((getip(pid) - (void*)handler) == 0);

		assert(ptrace(PTRACE_SINGLESTEP, pid, 0, SIGALRM) == 0);
		assert(wait(&status) == pid);
		assert(WIFSTOPPED(status) && WSTOPSIG(status) == SIGTRAP);
		assert((getip(pid) - (void*)handler) == 1);

		assert(ptrace(PTRACE_CONT, pid, 0,0) == 0);
		assert(wait(&status) == pid);
		assert(WIFEXITED(status) && WEXITSTATUS(status) == 0x23);

		return 0;
	}

The last assert() fails because PTRACE_CONT wrongly triggers
another single-step and X86_EFLAGS_TF can't be cleared by
debugger until the tracee does sys_rt_sigreturn().

Change handle_signal() to do user_disable_single_step() if
stepping, we do not need to preserve TIF_SINGLESTEP because we
are going to do ptrace_notify(), and it is simply wrong to leak
this bit.

While at it, change the comment to explain why we also need to
clear TF unconditionally after setup_rt_frame().

Note: in the longer term we should probably change
setup_sigcontext() to use get_flags() and then just remove this
user_disable_single_step().  And, the state of TIF_FORCED_TF can
be wrong after restore_sigcontext() which can set/clear TF, this
needs another fix.

This fix fixes the 'single_step_syscall_32' testcase in
the x86 testsuite:

Before:

	~/linux/tools/testing/selftests/x86> ./single_step_syscall_32
	[RUN]   Set TF and check nop
	[OK]    Survived with TF set and 9 traps
	[RUN]   Set TF and check int80
	[OK]    Survived with TF set and 9 traps
	[RUN]   Set TF and check a fast syscall
	[WARN]  Hit 10000 SIGTRAPs with si_addr 0xf7789cc0, ip 0xf7789cc0
	Trace/breakpoint trap (core dumped)

After:

	~/linux/linux/tools/testing/selftests/x86> ./single_step_syscall_32
	[RUN]   Set TF and check nop
	[OK]    Survived with TF set and 9 traps
	[RUN]   Set TF and check int80
	[OK]    Survived with TF set and 9 traps
	[RUN]   Set TF and check a fast syscall
	[OK]    Survived with TF set and 39 traps
	[RUN]   Fast syscall with TF cleared
	[OK]    Nothing unexpected happened

Reported-by: Evan Teran <eteran@alum.rit.edu>
Reported-by: Pedro Alves <palves@redhat.com>
Tested-by: Andres Freund <andres@anarazel.de>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
[ Added x86 self-test info. ]
Signed-off-by: Ingo Molnar <mingo@kernel.org>

1 files changed, 11 insertions, 11 deletions

diff --git a/arch/x86/kernel/signal.c b/arch/x86/kernel/signal.c
index 3e58186..d185bdd 100644
--- a/arch/x86/kernel/signal.c
+++ b/arch/x86/kernel/signal.c
@@ -630,7 +630,8 @@ setup_rt_frame(struct ksignal *ksig, struct pt_regs *regs)
 static void
 handle_signal(struct ksignal *ksig, struct pt_regs *regs)
 {
-	bool failed;
+	bool stepping, failed;
+
 	/* Are we from a system call? */
 	if (syscall_get_nr(current, regs) >= 0) {
 		/* If so, check system call restarting.. */
@@ -654,12 +655,13 @@ handle_signal(struct ksignal *ksig, struct pt_regs *regs)
 	}
 
 	/*
-	 * If TF is set due to a debugger (TIF_FORCED_TF), clear the TF
-	 * flag so that register information in the sigcontext is correct.
+	 * If TF is set due to a debugger (TIF_FORCED_TF), clear TF now
+	 * so that register information in the sigcontext is correct and
+	 * then notify the tracer before entering the signal handler.
 	 */
-	if (unlikely(regs->flags & X86_EFLAGS_TF) &&
-	    likely(test_and_clear_thread_flag(TIF_FORCED_TF)))
-		regs->flags &= ~X86_EFLAGS_TF;
+	stepping = test_thread_flag(TIF_SINGLESTEP);
+	if (stepping)
+		user_disable_single_step(current);
 
 	failed = (setup_rt_frame(ksig, regs) < 0);
 	if (!failed) {
@@ -670,10 +672,8 @@ handle_signal(struct ksignal *ksig, struct pt_regs *regs)
 		 * it might disable possible debug exception from the
 		 * signal handler.
 		 *
-		 * Clear TF when entering the signal handler, but
-		 * notify any tracer that was single-stepping it.
-		 * The tracer may want to single-step inside the
-		 * handler too.
+		 * Clear TF for the case when it wasn't set by debugger to
+		 * avoid the recursive send_sigtrap() in SIGTRAP handler.
 		 */
 		regs->flags &= ~(X86_EFLAGS_DF|X86_EFLAGS_RF|X86_EFLAGS_TF);
 		/*
@@ -682,7 +682,7 @@ handle_signal(struct ksignal *ksig, struct pt_regs *regs)
 		if (used_math())
 			fpu_reset_state(current);
 	}
-	signal_setup_done(failed, ksig, test_thread_flag(TIF_SINGLESTEP));
+	signal_setup_done(failed, ksig, stepping);
 }
 
 #ifdef CONFIG_X86_32