Skip to contentSkip to conversation

rseq: fix using an uninitialized stack variable in rseq_exit_user_update

PATCH Superseded Qing Wang 8 participants 12 messages Jun 1 – Jun 25

lore
Share & tools
lore

Qing Wang

author
#0 lore raw

PATCH 1/1 rseq: fix using an uninitialized stack variable in rseq_exit_user_update

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```

According to the C standard, the evaluation order of expressions in an initializer list is indeterminately sequenced. The compiler (Clang, in this KMSAN build) evaluates cpu_to_node(ids.cpu_id) *before* ids.cpu_id is initialized with task_cpu(t).

This is fixed by moving the assignment of ids.node_id outside the structure initialization.

Reported-by [email protected]
Fixes 82f572449cfe ("rseq: Implement read only ABI enforcement for optimized RSEQ V2 mode")
---
 include/linux/rseq_entry.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/rseq_entry.h +1 -1
index 63bc72086e75..e05f4c18f39e 100644
@@ -638,8 +638,8 @@ static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t
 	struct rseq_ids ids = {
 		.cpu_id	 = task_cpu(t),
 		.mm_cid	 = task_mm_cid(t),
-		.node_id = cpu_to_node(ids.cpu_id),
 	};
+	ids.node_id = cpu_to_node(ids.cpu_id);
 
 	return rseq_update_usr(t, regs, &ids);
 efault:
Signature · 1 line
2.34.1
#1 lore raw
Quoted 34 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```

According to the C standard, the evaluation order of expressions in an initializer list is indeterminately sequenced. The compiler (Clang, in this KMSAN build) evaluates cpu_to_node(ids.cpu_id) *before* ids.cpu_id is initialized with task_cpu(t).

This is fixed by moving the assignment of ids.node_id outside the structure initialization.

Reported-by [email protected]
Fixes 82f572449cfe ("rseq: Implement read only ABI enforcement for optimized RSEQ V2 mode")
---
  include/linux/rseq_entry.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/rseq_entry.h +0 -0
index 63bc72086e75..e05f4c18f39e 100644

include/linux/rseq_entry.h static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t · line 642

Quoted 6 lines
include/linux/rseq_entry.h +1 -1
@@ -638,8 +638,8 @@ static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t
  	struct rseq_ids ids = {
  		.cpu_id	 = task_cpu(t),
  		.mm_cid	 = task_mm_cid(t),
-		.node_id = cpu_to_node(ids.cpu_id),
  	};
+	ids.node_id = cpu_to_node(ids.cpu_id);

I find the fix odd. How about:

int cpu = task_cpu(t);

struct rseq_ids ids = {
	.cpu_id = cpu,
	.mm_cid = task_mm_cid(t),
	.node_id = cpu_to_node(cpu),
};

instead ?

Thanks,

Mathieu

Quoted 3 lines
include/linux/rseq_entry.h +0 -0
  	return rseq_update_usr(t, regs, &ids);
  efault:
Signature · 3 lines
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com
#2 lore raw
Quoted 6 lines · Mathieu Desnoyers
Quoted 5 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
Quoted 12 lines
Quoted 8 lines

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```
Quoted 4 lines

According to the C standard, the evaluation order of expressions in an initializer list is indeterminately sequenced. The compiler (Clang, in this KMSAN build) evaluates cpu_to_node(ids.cpu_id) *before* ids.cpu_id is initialized with task_cpu(t).

W.T.F. would they (C committee) do that :-(, and since this is apparently a known thing, why aren't the compilers issuing WARNs about this pattern?

Clearly any non address-of self reference in a struct initializer is buggered.

And ideally it would actually observe the data-dependency and respect it, irrespective of the lunacy of the C standard.

Quoted 25 lines
Quoted 16 lines

This is fixed by moving the assignment of ids.node_id outside the structure initialization.

Reported-by [email protected]
Fixes 82f572449cfe ("rseq: Implement read only ABI enforcement for optimized RSEQ V2 mode")
---
  include/linux/rseq_entry.h | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/rseq_entry.h +0 -0
index 63bc72086e75..e05f4c18f39e 100644

include/linux/rseq_entry.h static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t · line 642

Quoted 6 lines
include/linux/rseq_entry.h +1 -1
@@ -638,8 +638,8 @@ static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t
  	struct rseq_ids ids = {
  		.cpu_id	 = task_cpu(t),
  		.mm_cid	 = task_mm_cid(t),
-		.node_id = cpu_to_node(ids.cpu_id),
  	};
+	ids.node_id = cpu_to_node(ids.cpu_id);

I find the fix odd. How about:

Quoted 8 lines

int cpu = task_cpu(t);

struct rseq_ids ids = {
	.cpu_id = cpu,
	.mm_cid = task_mm_cid(t),
	.node_id = cpu_to_node(cpu),
};

instead ?

Yes, much saner.

author
#3 lore raw

PATCH v2 1/1 rseq: fix using an uninitialized stack variable in rseq_exit_user_update

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```

According to the C standard, the evaluation order of expressions in an initializer list is indeterminately sequenced. The compiler (Clang, in this KMSAN build) evaluates cpu_to_node(ids.cpu_id) *before* ids.cpu_id is initialized with task_cpu(t).

This is fixed by moving the assignment of ids.node_id outside the structure initialization.

Reported-by [email protected]
Fixes 82f572449cfe ("rseq: Implement read only ABI enforcement for optimized RSEQ V2 mode")
---
 include/linux/rseq_entry.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
include/linux/rseq_entry.h +3 -2
index 63bc72086e75..ed9da6e41a2a 100644
@@ -635,10 +635,11 @@ static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t
 		return true;
 	}
 
+	int cpu = task_cpu(t);
 	struct rseq_ids ids = {
-		.cpu_id	 = task_cpu(t),
+		.cpu_id	 = cpu,
 		.mm_cid	 = task_mm_cid(t),
-		.node_id = cpu_to_node(ids.cpu_id),
+		.node_id = cpu_to_node(cpu),
 	};
 
 	return rseq_update_usr(t, regs, &ids);
Signature · 1 line
2.34.1
Acked-by on v2 1/1
#4 lore raw
Quoted 18 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```

According to the C standard, the evaluation order of expressions in an initializer list is indeterminately sequenced. The compiler (Clang, in this KMSAN build) evaluates cpu_to_node(ids.cpu_id) *before* ids.cpu_id is initialized with task_cpu(t).

Quoted 7 lines

This is fixed by moving the assignment of ids.node_id outside the structure initialization.

Reported-by [email protected]
Fixes 82f572449cfe ("rseq: Implement read only ABI enforcement for optimized RSEQ V2 mode")

FWIW, this looks sane to me, so:

Mark.

Quoted 25 lines
---
 include/linux/rseq_entry.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
include/linux/rseq_entry.h +3 -2
index 63bc72086e75..ed9da6e41a2a 100644
@@ -635,10 +635,11 @@ static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t
 		return true;
 	}
 
+	int cpu = task_cpu(t);
 	struct rseq_ids ids = {
-		.cpu_id	 = task_cpu(t),
+		.cpu_id	 = cpu,
 		.mm_cid	 = task_mm_cid(t),
-		.node_id = cpu_to_node(ids.cpu_id),
+		.node_id = cpu_to_node(cpu),
 	};
 
 	return rseq_update_usr(t, regs, &ids);
Signature · 1 line
2.34.1
#5 lore raw
In reply to Qing Wang
Quoted 5 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
Quoted 8 lines

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```

FWIW, I've no idea what that ``` nonsense is, but it does not belong in Changelogs. I've removed it.

#6 lore raw
Quoted 15 lines · Peter Zijlstra
Quoted 5 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
Quoted 8 lines

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```
Quoted 3 lines

FWIW, I've no idea what that ``` nonsense is, but it does not belong in Changelogs. I've removed it.

It seems that this problem is still happening after
commit 6d99479799c6 ("rseq: Fix using an uninitialized stack variable in rseq_exit_user_update()") was applied. Please check.

#7 lore raw
Quoted 16 lines · Tetsuo Handa
Quoted 15 lines · Peter Zijlstra
Quoted 13 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```
Quoted 7 lines
Quoted 3 lines

FWIW, I've no idea what that ``` nonsense is, but it does not belong in Changelogs. I've removed it.

It seems that this problem is still happening after
commit 6d99479799c6 ("rseq: Fix using an uninitialized stack variable in rseq_exit_user_update()") was applied. Please check.

It seems is not really helpful. If you observe the problem can you please provide the full debug splat?

#8 lore raw
Quoted 21 lines · Thomas Gleixner
Quoted 20 lines · Tetsuo Handa
Quoted 18 lines · Peter Zijlstra
Quoted 5 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
Quoted 8 lines

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```

FWIW, I've no idea what that ``` nonsense is, but it does not belong in Changelogs. I've removed it.

Quoted 6 lines
Quoted 3 lines

It seems that this problem is still happening after
commit 6d99479799c6 ("rseq: Fix using an uninitialized stack variable in rseq_exit_user_update()") was applied. Please check.

It seems is not really helpful. If you observe the problem can you please provide the full debug splat?

Please fetch the full debug splat from https://syzkaller.appspot.com/bug?extid=185a631927096f9da2fc . This problem is still happening as of commit 9ecfb2f7287a which includes commit 6d99479799c6.

#9 lore raw
Quoted 26 lines · Tetsuo Handa
Quoted 25 lines · Thomas Gleixner
Quoted 23 lines · Tetsuo Handa
Quoted 15 lines · Peter Zijlstra
Quoted 5 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
Quoted 8 lines

The local variable:

```c
	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};
```
Quoted 3 lines

FWIW, I've no idea what that ``` nonsense is, but it does not belong in Changelogs. I've removed it.

It seems that this problem is still happening after
commit 6d99479799c6 ("rseq: Fix using an uninitialized stack variable in rseq_exit_user_update()") was applied. Please check.

Quoted 5 lines
Quoted 2 lines

It seems is not really helpful. If you observe the problem can you please provide the full debug splat?

Please fetch the full debug splat from https://syzkaller.appspot.com/bug?extid=185a631927096f9da2fc . This problem is still happening as of commit 9ecfb2f7287a which includes commit 6d99479799c6.

That's not a code problem. That's a KMSAN/compiler issue.

Quoted 2 lines

=====================================================

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

So it claims that the copy to user in rseq_set_ids_get_csaddr() is leaking kernel info through an unitialized variable.

The ids struct is on stack and was initialized correctly ...

Quoted 2 lines
BUG: KMSAN: kernel-infoleak in rseq_update_usr include/linux/rseq_entry.h:536 [inline]
BUG: KMSAN: kernel-infoleak in rseq_exit_user_update include/linux/rseq_entry.h:645 [inline]

here in rseq_exit_user_update().

ffffffff90ff6433:       48 89 df                mov    %rbx,%rdi
ffffffff90ff6436:       e8 e5 3f db f0          call   ffffffff81daa420 <task_cpu>
ffffffff90ff643b:       89 c7                   mov    %eax,%edi
ffffffff90ff643d:       89 45 98                mov    %eax,-0x68(%rbp)         ids.cpu_id
ffffffff90ff6440:       8b 8b 94 0b 00 00       mov    0xb94(%rbx),%ecx
ffffffff90ff6446:       b8 ff ff ff 9f          mov    $0x9fffffff,%eax
ffffffff90ff644b:       21 c1                   and    %eax,%ecx
ffffffff90ff644d:       89 4d 90                mov    %ecx,-0x70(%rbp)
ffffffff90ff6450:       89 4d 9c                mov    %ecx,-0x64(%rbp)         ids.mm_cid
ffffffff90ff6453:       89 7d ac                mov    %edi,-0x54(%rbp)
ffffffff90ff6456:       e8 25 40 db f0          call   ffffffff81daa480 <cpu_to_node>
ffffffff90ff645b:       89 45 94                mov    %eax,-0x6c(%rbp) 
ffffffff90ff645e:       89 45 a0                mov    %eax,-0x60(%rbp)         ids.node_id
ffffffff90ff6461:       c7 45 a4 00 00 00 00    movl   $0x0,-0x5c(%rbp)
ffffffff90ff6468:       48 8b bb 50 0b 00 00    mov    0xb50(%rbx),%rdi
ffffffff90ff646f:       e8 9c 3e db f0          call   ffffffff81daa310 <mask_user_address>
Quoted 11 lines
BUG: KMSAN: kernel-infoleak in __rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:674 [inline]
BUG: KMSAN: kernel-infoleak in rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:703 [inline]
BUG: KMSAN: kernel-infoleak in exit_to_user_mode_loop kernel/entry/common.c:103 [inline]
BUG: KMSAN: kernel-infoleak in __exit_to_user_mode_prepare include/linux/irq-entry-common.h:207 [inline]
BUG: KMSAN: kernel-infoleak in irqentry_exit_to_user_mode_prepare include/linux/irq-entry-common.h:244 [inline]
BUG: KMSAN: kernel-infoleak in irqentry_exit_to_user_mode include/linux/irq-entry-common.h:315 [inline]
BUG: KMSAN: kernel-infoleak in irqentry_exit+0x4a6/0xa40 kernel/entry/common.c:165
 rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
 rseq_update_usr include/linux/rseq_entry.h:536 [inline]
 rseq_exit_user_update include/linux/rseq_entry.h:645 [inline]
 __rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:674 [inline]
Quoted 8 lines
 rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:703 [inline]
 exit_to_user_mode_loop kernel/entry/common.c:103 [inline]
 __exit_to_user_mode_prepare include/linux/irq-entry-common.h:207 [inline]
 irqentry_exit_to_user_mode_prepare include/linux/irq-entry-common.h:244 [inline]
 irqentry_exit_to_user_mode include/linux/irq-entry-common.h:315 [inline]
 irqentry_exit+0x4a6/0xa40 kernel/entry/common.c:165
 exc_page_fault+0x7e/0xb0 arch/x86/mm/fault.c:1539
 asm_exc_page_fault+0x2b/0x30 arch/x86/include/asm/idtentry.h:595

And then it claims that the local variable was created in a completely unrelated piece of code:

Quoted 5 lines

Local variable st.i.i created at:

 __do_sys_statfs fs/statfs.c:193 [inline]
 __se_sys_statfs fs/statfs.c:191 [inline]
 __x64_sys_statfs+0x73/0x200 fs/statfs.c:191
 x64_sys_call+0x334c/0x3ea0 arch/x86/include/generated/asm/syscalls_64.h:138

The above backtrace clearly tells that the CPU cannot have been in that code because it's in a user mode page fault exception.

Quoted 3 lines

Bytes 0-3 of 4 are uninitialized
Memory access of size 4 starts at ffff8881192c3e78 Data copied to user address 00007f05e438a680

The code is correct and does the right thing, so something is wrong in KMSAN land. Alexander?

Thanks,

tglx

#10 lore raw
Quoted 98 lines · Thomas Gleixner
Quoted 26 lines · Tetsuo Handa
Quoted 25 lines · Thomas Gleixner
Quoted 23 lines · Tetsuo Handa
Quoted 15 lines · Peter Zijlstra
Quoted 5 lines · Qing Wang

There is an bug which is an uninitialized stack variable use in rseq_exit_user_update() reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
Quoted 8 lines

The local variable:

```c
  struct rseq_ids ids = {
          .cpu_id  = task_cpu(t),
          .mm_cid  = task_mm_cid(t),
          .node_id = cpu_to_node(ids.cpu_id),
  };
```
Quoted 3 lines

FWIW, I've no idea what that ``` nonsense is, but it does not belong in Changelogs. I've removed it.

It seems that this problem is still happening after
commit 6d99479799c6 ("rseq: Fix using an uninitialized stack variable in rseq_exit_user_update()") was applied. Please check.

Quoted 5 lines
Quoted 2 lines

It seems is not really helpful. If you observe the problem can you please provide the full debug splat?

Please fetch the full debug splat from https://syzkaller.appspot.com/bug?extid=185a631927096f9da2fc . This problem is still happening as of commit 9ecfb2f7287a which includes commit 6d99479799c6.

That's not a code problem. That's a KMSAN/compiler issue.

Quoted 2 lines

=====================================================

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

So it claims that the copy to user in rseq_set_ids_get_csaddr() is leaking kernel info through an unitialized variable.

The ids struct is on stack and was initialized correctly ...

Quoted 2 lines
BUG: KMSAN: kernel-infoleak in rseq_update_usr include/linux/rseq_entry.h:536 [inline]
BUG: KMSAN: kernel-infoleak in rseq_exit_user_update include/linux/rseq_entry.h:645 [inline]

here in rseq_exit_user_update().

ffffffff90ff6433:       48 89 df                mov    %rbx,%rdi
ffffffff90ff6436:       e8 e5 3f db f0          call   ffffffff81daa420 <task_cpu>
ffffffff90ff643b:       89 c7                   mov    %eax,%edi
ffffffff90ff643d:       89 45 98                mov    %eax,-0x68(%rbp)         ids.cpu_id
ffffffff90ff6440:       8b 8b 94 0b 00 00       mov    0xb94(%rbx),%ecx
ffffffff90ff6446:       b8 ff ff ff 9f          mov    $0x9fffffff,%eax
ffffffff90ff644b:       21 c1                   and    %eax,%ecx
ffffffff90ff644d:       89 4d 90                mov    %ecx,-0x70(%rbp)
ffffffff90ff6450:       89 4d 9c                mov    %ecx,-0x64(%rbp)         ids.mm_cid
ffffffff90ff6453:       89 7d ac                mov    %edi,-0x54(%rbp)
ffffffff90ff6456:       e8 25 40 db f0          call   ffffffff81daa480 <cpu_to_node>
ffffffff90ff645b:       89 45 94                mov    %eax,-0x6c(%rbp)
ffffffff90ff645e:       89 45 a0                mov    %eax,-0x60(%rbp)         ids.node_id
ffffffff90ff6461:       c7 45 a4 00 00 00 00    movl   $0x0,-0x5c(%rbp)
ffffffff90ff6468:       48 8b bb 50 0b 00 00    mov    0xb50(%rbx),%rdi
ffffffff90ff646f:       e8 9c 3e db f0          call   ffffffff81daa310 <mask_user_address>
Quoted 11 lines
BUG: KMSAN: kernel-infoleak in __rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:674 [inline]
BUG: KMSAN: kernel-infoleak in rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:703 [inline]
BUG: KMSAN: kernel-infoleak in exit_to_user_mode_loop kernel/entry/common.c:103 [inline]
BUG: KMSAN: kernel-infoleak in __exit_to_user_mode_prepare include/linux/irq-entry-common.h:207 [inline]
BUG: KMSAN: kernel-infoleak in irqentry_exit_to_user_mode_prepare include/linux/irq-entry-common.h:244 [inline]
BUG: KMSAN: kernel-infoleak in irqentry_exit_to_user_mode include/linux/irq-entry-common.h:315 [inline]
BUG: KMSAN: kernel-infoleak in irqentry_exit+0x4a6/0xa40 kernel/entry/common.c:165
 rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]
 rseq_update_usr include/linux/rseq_entry.h:536 [inline]
 rseq_exit_user_update include/linux/rseq_entry.h:645 [inline]
 __rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:674 [inline]
Quoted 8 lines
 rseq_exit_to_user_mode_restart include/linux/rseq_entry.h:703 [inline]
 exit_to_user_mode_loop kernel/entry/common.c:103 [inline]
 __exit_to_user_mode_prepare include/linux/irq-entry-common.h:207 [inline]
 irqentry_exit_to_user_mode_prepare include/linux/irq-entry-common.h:244 [inline]
 irqentry_exit_to_user_mode include/linux/irq-entry-common.h:315 [inline]
 irqentry_exit+0x4a6/0xa40 kernel/entry/common.c:165
 exc_page_fault+0x7e/0xb0 arch/x86/mm/fault.c:1539
 asm_exc_page_fault+0x2b/0x30 arch/x86/include/asm/idtentry.h:595

And then it claims that the local variable was created in a completely unrelated piece of code:

Quoted 5 lines

Local variable st.i.i created at:

 __do_sys_statfs fs/statfs.c:193 [inline]
 __se_sys_statfs fs/statfs.c:191 [inline]
 __x64_sys_statfs+0x73/0x200 fs/statfs.c:191
 x64_sys_call+0x334c/0x3ea0 arch/x86/include/generated/asm/syscalls_64.h:138

The above backtrace clearly tells that the CPU cannot have been in that code because it's in a user mode page fault exception.

Quoted 6 lines
Quoted 3 lines

Bytes 0-3 of 4 are uninitialized
Memory access of size 4 starts at ffff8881192c3e78 Data copied to user address 00007f05e438a680

The code is correct and does the right thing, so something is wrong in KMSAN land. Alexander?

Thanks for the heads-up.
I am currently reworking both the compiler pass and the exception handling, which should hopefully address these problems.

merge bot Applied
#11 lore raw

[tip: sched/urgent] rseq: Fix using an uninitialized stack variable in rseq_exit_user_update()

In reply to Qing Wang
The following commit has been merged into the sched/urgent branch of tip:

The following commit has been merged into the sched/urgent branch of tip:

Commit-ID:     6d99479799c69c3cb588fcda19c81d8f61d64ecd
Gitweb:        https://git.kernel.org/tip/6d99479799c69c3cb588fcda19c81d8f61d64ecd
Author:        Qing Wang <[email protected]>
AuthorDate:    Tue, 02 Jun 2026 11:08:54 +08:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Tue, 02 Jun 2026 12:41:15 +02:00

rseq: Fix using an uninitialized stack variable in rseq_exit_user_update()

There is an bug in which an uninitialized stack variable is used in rseq_exit_user_update() as reported by syzbot:

BUG: KMSAN: kernel-infoleak in rseq_set_ids_get_csaddr include/linux/rseq_entry.h:502 [inline]

The local variable:

	struct rseq_ids ids = {
		.cpu_id	 = task_cpu(t),
		.mm_cid	 = task_mm_cid(t),
		.node_id = cpu_to_node(ids.cpu_id),
	};

According to the C standard, the evaluation order of expressions in an initializer list is indeterminately sequenced. The compiler (Clang, in this KMSAN build) evaluates cpu_to_node(ids.cpu_id) *before* ids.cpu_id is initialized with task_cpu(t).

This is fixed by moving the assignment of ids.node_id outside the structure initialization.

---
 include/linux/rseq_entry.h | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
include/linux/rseq_entry.h +3 -2
index 63bc720..ed9da6e 100644
@@ -635,10 +635,11 @@ static __always_inline bool rseq_exit_user_update(struct pt_regs *regs, struct t
 		return true;
 	}
 
+	int cpu = task_cpu(t);
 	struct rseq_ids ids = {
-		.cpu_id	 = task_cpu(t),
+		.cpu_id	 = cpu,
 		.mm_cid	 = task_mm_cid(t),
-		.node_id = cpu_to_node(ids.cpu_id),
+		.node_id = cpu_to_node(cpu),
 	};
 
 	return rseq_update_usr(t, regs, &ids);