[PATCH V2 3/5] powerpc: Fix SPR leak across exec() syscalls

[Cyril Bur]

Currently start_thread() doesn't sanitise TAR nor the VRSAVE.

The TAR SPR can be set and branched to, not sanitising it presents an
information leak to the new executable.

The VRSAVE SPR can be used by both application and operating system. The
PowerISA states (in a programming note) that an application could use VRSAVE as a
hint to the operating system as to which vector registers are being used and
therefore which ones to save and restore. Linux doesn't care and simply
saves/restores all vector registers, however, it should also be sanitising
VRSAVE as applications are free to do with it what they wish.

Other SPR registers such as the Performance registers used by perf (and are
managed entirely by perf) as well as the Event Based Branch (EBB) registers are
left alone by design as these fall into the same category as leaving file
descriptors open across exec(), it is up the parent thread to sanitise what it
deems necessary.

Signed-off-by: Cyril Bur <cyrilbur at gmail.com>
---
 arch/powerpc/kernel/process.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c
index d7a9df5..56444a6 100644
--- a/arch/powerpc/kernel/process.c
+++ b/arch/powerpc/kernel/process.c
@@ -1577,6 +1577,8 @@ void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 	current->thread.vr_state.vscr.u[3] = 0x00010000; /* Java mode disabled */
 	current->thread.vr_save_area = NULL;
 	current->thread.vrsave = 0;
+	if (cpu_has_feature(CPU_FTR_ALTIVEC))
+		mtspr(SPRN_VRSAVE, 0);
 	current->thread.used_vr = 0;
 #endif /* CONFIG_ALTIVEC */
 #ifdef CONFIG_SPE
@@ -1592,6 +1594,18 @@ void start_thread(struct pt_regs *regs, unsigned long start, unsigned long sp)
 	current->thread.tm_texasr = 0;
 	current->thread.tm_tfiar = 0;
 #endif /* CONFIG_PPC_TRANSACTIONAL_MEM */
+#ifdef CONFIG_PPC_BOOK3S_64
+	/*
+	 * Zero out the SPRs.
+	 * Don't touch the ones use by perf, it controls them.
+	 * Don't touch the EBB regs. This falls into the same category of
+	 *   responsibly as open file descriptors across exec(), the parent should
+	 *   sanitise if it feels it would be a problem
+	 */
+	current->thread.tar = 0;
+	if (cpu_has_feature(CPU_FTR_ARCH_206))
+		mtspr(SPRN_TAR, 0);
+#endif /* CONFIG_PPC_BOOK3S_64 */
 }
 EXPORT_SYMBOL(start_thread);
 
-- 
2.7.4