Skip to contentSkip to conversation

parisc/entry: fix trace test in syscall exit path

PATCH Superseded Sven Schnelle 2 participants 5 messages Nov 11, 2021 – Nov 13, 2021 PA-RISC architecture

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 parisc/entry: fix trace test in syscall exit path

commit 8779e05ba8aa ("parisc: Fix ptrace check on syscall return") fixed testing of TI_FLAGS. This uncovered a bug in the test mask. syscall_restore_rfi is only used when the kernel needs to exit to usespace with single stepping via recovery counter enabled. The test however used _TIF_SYSCALL_TRACE_MASK, which includes a lot of bits that shouldn't be tested here.

Fix this by using TIF_SINGLESTEP and TIF_BLOCKSTEP directly and remove those bits from TIF_SYSCALL_TRACE_MASK.

I encountered this bug by enabling syscall tracepoints. Both in qemu and on real hardware. As soon as i enabled the tracepoint (sys_exit_read, but i guess it doesn't really matter which one), i got random page faults in userspace almost immediately.

---
 arch/parisc/include/asm/thread_info.h | 3 +--
 arch/parisc/kernel/entry.S            | 2 +-
 2 files changed, 2 insertions(+), 3 deletions(-)
arch/parisc/include/asm/thread_info.h +1 -2
index 1a58795f785c..a3ba8c518292 100644
@@ -68,8 +68,7 @@ struct thread_info {
 
 #define _TIF_USER_WORK_MASK     (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | \
                                  _TIF_NEED_RESCHED | _TIF_NOTIFY_SIGNAL)
-#define _TIF_SYSCALL_TRACE_MASK (_TIF_SYSCALL_TRACE | _TIF_SINGLESTEP |	\
-				 _TIF_BLOCKSTEP | _TIF_SYSCALL_AUDIT | \
+#define _TIF_SYSCALL_TRACE_MASK (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
 				 _TIF_SECCOMP | _TIF_SYSCALL_TRACEPOINT)
 
 #ifdef CONFIG_64BIT
arch/parisc/kernel/entry.S +1 -1
index 57944d6f9ebb..88c188a965d8 100644
@@ -1805,7 +1805,7 @@ syscall_restore:
 
 	/* Are we being ptraced? */
 	LDREG	TASK_TI_FLAGS(%r1),%r19
-	ldi	_TIF_SYSCALL_TRACE_MASK,%r2
+	ldi	_TIF_SINGLESTEP|_TIF_BLOCKSTEP,%r2
 	and,COND(=)	%r19,%r2,%r0
 	b,n	syscall_restore_rfi
 
Signature · 1 line
2.33.1
#1 lore raw
Quoted 40 lines · Sven Schnelle

commit 8779e05ba8aa ("parisc: Fix ptrace check on syscall return") fixed testing of TI_FLAGS. This uncovered a bug in the test mask. syscall_restore_rfi is only used when the kernel needs to exit to usespace with single stepping via recovery counter enabled. The test however used _TIF_SYSCALL_TRACE_MASK, which includes a lot of bits that shouldn't be tested here.

Fix this by using TIF_SINGLESTEP and TIF_BLOCKSTEP directly and remove those bits from TIF_SYSCALL_TRACE_MASK.

I encountered this bug by enabling syscall tracepoints. Both in qemu and on real hardware. As soon as i enabled the tracepoint (sys_exit_read, but i guess it doesn't really matter which one), i got random page faults in userspace almost immediately.

---
  arch/parisc/include/asm/thread_info.h | 3 +--
  arch/parisc/kernel/entry.S            | 2 +-
  2 files changed, 2 insertions(+), 3 deletions(-)
arch/parisc/include/asm/thread_info.h +1 -2
index 1a58795f785c..a3ba8c518292 100644
@@ -68,8 +68,7 @@ struct thread_info {
 
  #define _TIF_USER_WORK_MASK     (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | \
                                   _TIF_NEED_RESCHED | _TIF_NOTIFY_SIGNAL)
-#define _TIF_SYSCALL_TRACE_MASK (_TIF_SYSCALL_TRACE | _TIF_SINGLESTEP |	\
-				 _TIF_BLOCKSTEP | _TIF_SYSCALL_AUDIT | \
+#define _TIF_SYSCALL_TRACE_MASK (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
  				 _TIF_SECCOMP | _TIF_SYSCALL_TRACEPOINT)
 
  #ifdef CONFIG_64BIT
arch/parisc/kernel/entry.S +0 -0
index 57944d6f9ebb..88c188a965d8 100644

arch/parisc/kernel/entry.S syscall_restore: · line 1808

Quoted 5 lines
arch/parisc/kernel/entry.S +1 -1
@@ -1805,7 +1805,7 @@ syscall_restore:
 
  	/* Are we being ptraced? */
  	LDREG	TASK_TI_FLAGS(%r1),%r19
-	ldi	_TIF_SYSCALL_TRACE_MASK,%r2
+	ldi	_TIF_SINGLESTEP|_TIF_BLOCKSTEP,%r2

This change seems applied to the old code and not 8779e05ba8aa.

Dave

Signature · 1 line
John David Anglin  [email protected]
#2 lore raw
Quoted 5 lines · John David Anglin
Quoted 4 lines

      /* Are we being ptraced? */
      LDREG    TASK_TI_FLAGS(%r1),%r19
-    ldi    _TIF_SYSCALL_TRACE_MASK,%r2
+    ldi    _TIF_SINGLESTEP|_TIF_BLOCKSTEP,%r2

This change seems applied to the old code and not 8779e05ba8aa.

Forget this.  TASK_TI_FLAGS was introduced in "parisc: Move thread_info into task struct".

Dave

Signature · 1 line
John David Anglin  [email protected]
#3 lore raw
In reply to Sven Schnelle
Quoted 9 lines · Sven Schnelle

commit 8779e05ba8aa ("parisc: Fix ptrace check on syscall return") fixed testing of TI_FLAGS. This uncovered a bug in the test mask. syscall_restore_rfi is only used when the kernel needs to exit to usespace with single stepping via recovery counter enabled. The test however used _TIF_SYSCALL_TRACE_MASK, which includes a lot of bits that shouldn't be tested here.

Fix this by using TIF_SINGLESTEP and TIF_BLOCKSTEP directly and remove those bits from TIF_SYSCALL_TRACE_MASK.

I think we need to have TIF_SINGLESTEP and TIF_BLOCKSTEP in TIF_SYSCALL_TRACE_MASK otherwise do_syscall_trace_exit() isn't called when leaving to userspace. I'll read the code a bit more during the weekend and prepare a v2.

#4 lore raw
Quoted 9 lines · Sven Schnelle
Quoted 7 lines · Sven Schnelle

commit 8779e05ba8aa ("parisc: Fix ptrace check on syscall return") fixed testing of TI_FLAGS. This uncovered a bug in the test mask. syscall_restore_rfi is only used when the kernel needs to exit to usespace with single stepping via recovery counter enabled. The test however used _TIF_SYSCALL_TRACE_MASK, which includes a lot of bits that shouldn't be tested here.

Quoted 6 lines
Quoted 2 lines

Fix this by using TIF_SINGLESTEP and TIF_BLOCKSTEP directly and remove those bits from TIF_SYSCALL_TRACE_MASK.

I think we need to have TIF_SINGLESTEP and TIF_BLOCKSTEP in TIF_SYSCALL_TRACE_MASK otherwise do_syscall_trace_exit() isn't called when leaving to userspace. I'll read the code a bit more during the weekend and prepare a v2.

Signal delivery is broken in 5.16.x.  This causes a number of glibc and ada test regressions.

Signature · 1 line
John David Anglin  [email protected]