s390/ptrace: race of single stepping vs signal delivery

The current single step code is racy in regard to concurrent delivery
of signals. If a signal is delivered after a PER program check occurred
but before the TIF_PER_TRAP bit has been checked in entry[64].S the code
clears TIF_PER_TRAP and then calls do_signal. This is wrong, if the
instruction completed (or has been suppressed) a SIGTRAP should be
delivered to the debugger in any case. Only if the instruction has been
nullified the SIGTRAP may not be send.

The new logic always sets TIF_PER_TRAP if the program check indicates PER
tracing but removes it again for all program checks that are nullifying.
The effect is that for each change in the PSW address we now get a
single SIGTRAP.

Reported-by: Andreas Arnez <arnez@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
diff --git a/arch/s390/kernel/entry.S b/arch/s390/kernel/entry.S
index ef46f66..f954b37 100644
--- a/arch/s390/kernel/entry.S
+++ b/arch/s390/kernel/entry.S
@@ -231,12 +231,12 @@
 	jo	sysc_mcck_pending
 	tm	__TI_flags+3(%r12),_TIF_NEED_RESCHED
 	jo	sysc_reschedule
+	tm	__TI_flags+3(%r12),_TIF_PER_TRAP
+	jo	sysc_singlestep
 	tm	__TI_flags+3(%r12),_TIF_SIGPENDING
 	jo	sysc_sigpending
 	tm	__TI_flags+3(%r12),_TIF_NOTIFY_RESUME
 	jo	sysc_notify_resume
-	tm	__TI_flags+3(%r12),_TIF_PER_TRAP
-	jo	sysc_singlestep
 	j	sysc_return		# beware of critical section cleanup
 
 #
@@ -259,7 +259,6 @@
 # _TIF_SIGPENDING is set, call do_signal
 #
 sysc_sigpending:
-	ni	__TI_flags+3(%r12),255-_TIF_PER_TRAP # clear TIF_PER_TRAP
 	lr	%r2,%r11		# pass pointer to pt_regs
 	l	%r1,BASED(.Ldo_signal)
 	basr	%r14,%r1		# call do_signal
@@ -286,7 +285,7 @@
 # _TIF_PER_TRAP is set, call do_per_trap
 #
 sysc_singlestep:
-	ni	__TI_flags+3(%r12),255-(_TIF_SYSCALL | _TIF_PER_TRAP)
+	ni	__TI_flags+3(%r12),255-_TIF_PER_TRAP
 	lr	%r2,%r11		# pass pointer to pt_regs
 	l	%r1,BASED(.Ldo_per_trap)
 	la	%r14,BASED(sysc_return)
diff --git a/arch/s390/kernel/entry64.S b/arch/s390/kernel/entry64.S
index e42842a..7a2d22d 100644
--- a/arch/s390/kernel/entry64.S
+++ b/arch/s390/kernel/entry64.S
@@ -262,12 +262,12 @@
 	jo	sysc_mcck_pending
 	tm	__TI_flags+7(%r12),_TIF_NEED_RESCHED
 	jo	sysc_reschedule
+	tm	__TI_flags+7(%r12),_TIF_PER_TRAP
+	jo	sysc_singlestep
 	tm	__TI_flags+7(%r12),_TIF_SIGPENDING
 	jo	sysc_sigpending
 	tm	__TI_flags+7(%r12),_TIF_NOTIFY_RESUME
 	jo	sysc_notify_resume
-	tm	__TI_flags+7(%r12),_TIF_PER_TRAP
-	jo	sysc_singlestep
 	j	sysc_return		# beware of critical section cleanup
 
 #
@@ -288,7 +288,6 @@
 # _TIF_SIGPENDING is set, call do_signal
 #
 sysc_sigpending:
-	ni	__TI_flags+7(%r12),255-_TIF_PER_TRAP # clear TIF_PER_TRAP
 	lgr	%r2,%r11		# pass pointer to pt_regs
 	brasl	%r14,do_signal
 	tm	__TI_flags+7(%r12),_TIF_SYSCALL
@@ -313,7 +312,7 @@
 # _TIF_PER_TRAP is set, call do_per_trap
 #
 sysc_singlestep:
-	ni	__TI_flags+7(%r12),255-(_TIF_SYSCALL | _TIF_PER_TRAP)
+	ni	__TI_flags+7(%r12),255-_TIF_PER_TRAP
 	lgr	%r2,%r11		# pass pointer to pt_regs
 	larl	%r14,sysc_return
 	jg	do_per_trap
diff --git a/arch/s390/kernel/signal.c b/arch/s390/kernel/signal.c
index d1259d8..c3ff70a 100644
--- a/arch/s390/kernel/signal.c
+++ b/arch/s390/kernel/signal.c
@@ -461,6 +461,8 @@
 			/* Restart system call with magic TIF bit. */
 			regs->gprs[2] = regs->orig_gpr2;
 			set_thread_flag(TIF_SYSCALL);
+			if (test_thread_flag(TIF_SINGLE_STEP))
+				set_thread_flag(TIF_PER_TRAP);
 			break;
 		}
 	}
diff --git a/arch/s390/mm/fault.c b/arch/s390/mm/fault.c
index 870a644..42601d6 100644
--- a/arch/s390/mm/fault.c
+++ b/arch/s390/mm/fault.c
@@ -277,10 +277,16 @@
 	unsigned int flags;
 	int fault;
 
+	tsk = current;
+	/*
+	 * The instruction that caused the program check has
+	 * been nullified. Don't signal single step via SIGTRAP.
+	 */
+	clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
+
 	if (notify_page_fault(regs))
 		return 0;
 
-	tsk = current;
 	mm = tsk->mm;
 	trans_exc_code = regs->int_parm_long;
 
@@ -376,11 +382,6 @@
 			goto retry;
 		}
 	}
-	/*
-	 * The instruction that caused the program check will
-	 * be repeated. Don't signal single step via SIGTRAP.
-	 */
-	clear_tsk_thread_flag(tsk, TIF_PER_TRAP);
 	fault = 0;
 out_up:
 	up_read(&mm->mmap_sem);
@@ -427,6 +428,12 @@
 	struct vm_area_struct *vma;
 	unsigned long trans_exc_code;
 
+	/*
+	 * The instruction that caused the program check has
+	 * been nullified. Don't signal single step via SIGTRAP.
+	 */
+	clear_tsk_thread_flag(current, TIF_PER_TRAP);
+
 	trans_exc_code = regs->int_parm_long;
 	if (unlikely(!user_space_fault(trans_exc_code) || in_atomic() || !mm))
 		goto no_context;