[PATCH] x86_64, entry: Fix out of bounds read on sysenter

[Andy Lutomirski]

Rusty noticed a Really Bad Bug (tm) in my NT fix.  The entry code
reads out of bounds, causing the NT fix to be unreliable.  But, and
this is much, much worse, if your stack is somehow just below the
top of the direct map (or a hole), you read out of bounds and crash.

Excerpt from the crash:

[    1.129513] RSP: 0018:ffff88001da4bf88  EFLAGS: 00010296

  2b:*    f7 84 24 90 00 00 00     testl  $0x4000,0x90(%rsp)

That read is deterministically above the top of the stack.  I
thought I even single-stepped through this code when I wrote it to
check the offset, but I clearly screwed it up.

Fixes 8c7aa698baca x86_64, entry: Filter RFLAGS.NT on entry from userspace

Reported-by: Rusty Russell <rusty@xxxxxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Andy Lutomirski <luto@xxxxxxxxxxxxxx>
---

Linus, etc: this should probably go in pretty quickly before it hits -stable
too hard.  Fortunately it's unlikely to be a meaningful security problem,
but it's a nasty regression.  (It *is* a security problem in -next, but we
get a free pass on that one.)

 arch/x86/ia32/ia32entry.S | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S
index 8ffba18395c8..ffe71228fc10 100644
--- a/arch/x86/ia32/ia32entry.S
+++ b/arch/x86/ia32/ia32entry.S
@@ -157,7 +157,7 @@ ENTRY(ia32_sysenter_target)
 	 * ourselves.  To save a few cycles, we can check whether
 	 * NT was set instead of doing an unconditional popfq.
 	 */
-	testl $X86_EFLAGS_NT,EFLAGS(%rsp)	/* saved EFLAGS match cpu */
+	testl $X86_EFLAGS_NT,EFLAGS-ARGOFFSET(%rsp)
 	jnz sysenter_fix_flags
 sysenter_flags_fixed:
 
-- 
1.9.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/