Skip to contentSkip to conversation

signal: always clear sa_restorer on execve

PATCH Superseded Kees Cook 3 participants 6 messages Mar 11, 2013

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Kees Cook

author
#0 lore raw

PATCH 1/1 signal: always clear sa_restorer on execve

When the new signal handlers are set up for a fork, the location of sa_restorer is not cleared, leaking a parent process's address space location to children. This allows for a potential bypass of the parent's ASLR by examining the sa_restorer value returned when calling sigaction().

$ cat /proc/$$/maps
...
7fb9f3083000-7fb9f3238000 r-xp 00000000 fd:01 404469 .../libc-2.15.so
...
$ ./leak
...
7f278bc74000-7f278be29000 r-xp 00000000 fd:01 404469 .../libc-2.15.so
...
1 0 (nil) 0x7fb9f30b94a0
2 4000000 (nil) 0x7f278bcaa4a0
3 4000000 (nil) 0x7f278bcaa4a0
4 0 (nil) 0x7fb9f30b94a0
...
---
 kernel/signal.c |    3 +++
 1 file changed, 3 insertions(+)
kernel/signal.c +3 -0
index 2ec870a..8c8e3ca 100644
@@ -485,6 +485,9 @@ flush_signal_handlers(struct task_struct *t, int force_default)
 		if (force_default || ka->sa.sa_handler != SIG_IGN)
 			ka->sa.sa_handler = SIG_DFL;
 		ka->sa.sa_flags = 0;
+#ifdef __ARCH_HAS_SA_RESTORER
+		ka->sa.sa_restorer = NULL;
+#endif
 		sigemptyset(&ka->sa.sa_mask);
 		ka++;
 	}
Signature · 6 lines
1.7.9.5


-- 
Kees Cook
Chrome OS Security
#1 lore raw
Quoted 5 lines · Kees Cook

When the new signal handlers are set up for a fork, the location of sa_restorer is not cleared, leaking a parent process's address space location to children. This allows for a potential bypass of the parent's ASLR by examining the sa_restorer value returned when calling sigaction().

I don't understand.

fork() should not change restorer/etc, and the child has the same address space anyway. There is no any leak and the patch can't make any difference in this case because flush_signal_handlers() is not called by fork().

kernel/signal.c flush_signal_handlers(struct task_struct *t, int force_default) · line 490

Quoted 7 lines
kernel/signal.c +3 -0
@@ -485,6 +485,9 @@ flush_signal_handlers(struct task_struct *t, int force_default)
 		if (force_default || ka->sa.sa_handler != SIG_IGN)
 			ka->sa.sa_handler = SIG_DFL;
 		ka->sa.sa_flags = 0;
+#ifdef __ARCH_HAS_SA_RESTORER
+		ka->sa.sa_restorer = NULL;
+#endif

However, exec sets SIG_DFL but keeps ->sa_restorer, so probably this patch makes sense anyway.

Oleg.

author
#2 lore raw
Quoted 7 lines · Oleg Nesterov
Quoted 5 lines · Kees Cook

When the new signal handlers are set up for a fork, the location of sa_restorer is not cleared, leaking a parent process's address space location to children. This allows for a potential bypass of the parent's ASLR by examining the sa_restorer value returned when calling sigaction().

Quoted 5 lines

I don't understand.

fork() should not change restorer/etc, and the child has the same address space anyway. There is no any leak and the patch can't make any difference in this case because flush_signal_handlers() is not called by fork().

I probably failed to explain this correctly. From the perspective of what should be considered "secret", it only matters across the exec, not the fork (since the VMAs haven't changed until the exec). But the info leak is easy to see, and this patch fixes it. As you say, since other things were reset, so should sa_restorer.

Quoted 12 lines

kernel/signal.c flush_signal_handlers(struct task_struct *t, int force_default) · line 490

Quoted 8 lines
kernel/signal.c +3 -0
@@ -485,6 +485,9 @@ flush_signal_handlers(struct task_struct *t, int force_default)
              if (force_default || ka->sa.sa_handler != SIG_IGN)
                      ka->sa.sa_handler = SIG_DFL;
              ka->sa.sa_flags = 0;
+#ifdef __ARCH_HAS_SA_RESTORER
+             ka->sa.sa_restorer = NULL;
+#endif

However, exec sets SIG_DFL but keeps ->sa_restorer, so probably this patch makes sense anyway.

Quoted 2 lines

Oleg.

-Kees

Signature · 2 lines
Kees Cook
Chrome OS Security
#3 lore raw
Quoted 10 lines · Kees Cook
Quoted 9 lines · Oleg Nesterov
Quoted 5 lines · Kees Cook

When the new signal handlers are set up for a fork, the location of sa_restorer is not cleared, leaking a parent process's address space location to children. This allows for a potential bypass of the parent's ASLR by examining the sa_restorer value returned when calling sigaction().

I don't understand.

Quoted 7 lines
Quoted 3 lines

fork() should not change restorer/etc, and the child has the same address space anyway. There is no any leak and the patch can't make any difference in this case because flush_signal_handlers() is not called by fork().

I probably failed to explain this correctly. From the perspective of what should be considered "secret", it only matters across the exec, not the fork (since the VMAs haven't changed until the exec).

Exactly. So may be you should update the changelog? It really looks as if the parent should "hide" something from the forked child.

Quoted 3 lines

But the
info leak is easy to see, and this patch fixes it. As you say, since other things were reset, so should sa_restorer.

Yes, I didn't argue with the patch itself.

Oleg.

#4 lore raw
In reply to Kees Cook
Quoted 14 lines · Kees Cook
Quoted 7 lines · Oleg Nesterov
Quoted 5 lines · Kees Cook

When the new signal handlers are set up for a fork, the location of sa_restorer is not cleared, leaking a parent process's address space location to children. This allows for a potential bypass of the parent's ASLR by examining the sa_restorer value returned when calling sigaction().

Quoted 5 lines

I don't understand.

fork() should not change restorer/etc, and the child has the same address space anyway. There is no any leak and the patch can't make any difference in this case because flush_signal_handlers() is not called by fork().

Quoted 5 lines

I probably failed to explain this correctly. From the perspective of what should be considered "secret", it only matters across the exec, not the fork (since the VMAs haven't changed until the exec). But the info leak is easy to see, and this patch fixes it. As you say, since other things were reset, so should sa_restorer.

At the very least please correct the explanation in your patch description.

Too often I have had seen a confused patch description, indicate confusion elsewhere in the patch. Let's make it easy for reviewers and future bisectors to understand what is intended.

Quoted 7 lines
Quoted 7 lines

kernel/signal.c flush_signal_handlers(struct task_struct *t, int force_default) · line 490

Quoted 7 lines
kernel/signal.c +3 -0
@@ -485,6 +485,9 @@ flush_signal_handlers(struct task_struct *t, int force_default)
              if (force_default || ka->sa.sa_handler != SIG_IGN)
                      ka->sa.sa_handler = SIG_DFL;
              ka->sa.sa_flags = 0;
+#ifdef __ARCH_HAS_SA_RESTORER
+             ka->sa.sa_restorer = NULL;
+#endif

Also I am inclined to suggest this should be an inline function in a header.

		clear_sa_restorer(ka);

Just so we don't litter the code with #ifdefs.

Eric

author
#5 lore raw
Quoted 22 lines · Eric W. Biederman
Quoted 14 lines · Kees Cook
Quoted 7 lines · Oleg Nesterov
Quoted 5 lines · Kees Cook

When the new signal handlers are set up for a fork, the location of sa_restorer is not cleared, leaking a parent process's address space location to children. This allows for a potential bypass of the parent's ASLR by examining the sa_restorer value returned when calling sigaction().

Quoted 5 lines

I don't understand.

fork() should not change restorer/etc, and the child has the same address space anyway. There is no any leak and the patch can't make any difference in this case because flush_signal_handlers() is not called by fork().

Quoted 5 lines

I probably failed to explain this correctly. From the perspective of what should be considered "secret", it only matters across the exec, not the fork (since the VMAs haven't changed until the exec). But the info leak is easy to see, and this patch fixes it. As you say, since other things were reset, so should sa_restorer.

Quoted 6 lines

At the very least please correct the explanation in your patch description.

Too often I have had seen a confused patch description, indicate confusion elsewhere in the patch. Let's make it easy for reviewers and future bisectors to understand what is intended.

Sounds good. I'll re-attempt and send a v2.

Quoted 10 lines
Quoted 7 lines
Quoted 7 lines

kernel/signal.c flush_signal_handlers(struct task_struct *t, int force_default) · line 490

Quoted 7 lines
kernel/signal.c +3 -0
@@ -485,6 +485,9 @@ flush_signal_handlers(struct task_struct *t, int force_default)
              if (force_default || ka->sa.sa_handler != SIG_IGN)
                      ka->sa.sa_handler = SIG_DFL;
              ka->sa.sa_flags = 0;
+#ifdef __ARCH_HAS_SA_RESTORER
+             ka->sa.sa_restorer = NULL;
+#endif

Also I am inclined to suggest this should be an inline function in a header.

Quoted 3 lines
                clear_sa_restorer(ka);

Just so we don't litter the code with #ifdefs.

Yeah, I took a look, and the code was already using the __ARCH_HAS... ifdef, so I went with that. I can respin with some kind of set_sa_restorer() for the other places sa_restorer is used.

-Kees

Signature · 2 lines
Kees Cook
Chrome OS Security