MIPS: Fix potential DOS by untrusted user app.

From: Vlad Malov <[email protected]>

Fix potential DOS by untrusted user app.

Version 3: Fix stupid typo I introduced. This version has received extensive testing with the LTP where it fixes kernel crashes.

On a 64 bit kernel if an o32 syscall was made with a syscall number less than 4000, we would read the function from outside of the bounds of the syscall table. This led to non-deterministic behavior including system crashes.

While we were at it we reworked the 32 bit version as well to use fewer instructions. Both 32 and 64 bit versions are use the same code now.

This version two should address the concerns Maciej raised. gas seems to expand this instruction ordering with no nops in the delay slots.

---
arch/mips/kernel/scall32-o32.S | 5 +----
arch/mips/kernel/scall64-o32.S | 12 +++++-------

2 files changed, 6 insertions(+), 11 deletions(-)

index 759f680..34a4dbd 100644
@@ -262,14 +262,11 @@ bad_alignment:

	LEAF(sys_syscall)
	subu	t0, a0, __NR_O32_Linux	# check syscall number
	sltiu	v0, t0, __NR_O32_Linux_syscalls + 1
+	beqz	t0, einval		# do not recurse
	sll	t1, t0, 3
	beqz	v0, einval

-
lw t2, sys_call_table(t1) # syscall routine

- li v1, 4000 - __NR_O32_Linux # index of sys_syscall
- beq t0, v1, einval # do not recurse
-
/* Some syscalls like execve get their arguments from struct pt_regs
and claim zero arguments in the syscall table. Thus we have to assume the worst case and shuffle around all potential arguments.

index 6c7ef83..facb41a 100644
@@ -174,14 +174,12 @@ not_o32_scall:

END(handle_sys)

LEAF(sys32_syscall)
- sltu v0, a0, __NR_O32_Linux + __NR_O32_Linux_syscalls + 1
+ subu t0, a0, __NR_O32_Linux # check syscall number
+ sltiu v0, t0, __NR_O32_Linux_syscalls + 1
+ beqz t0, einval # do not recurse
+ dsll t1, t0, 3
beqz v0, einval
-
- dsll v0, a0, 3
- ld t2, (sys_call_table - (__NR_O32_Linux * 8))(v0)
-
- li v1, 4000 # indirect syscall number
- beq a0, v1, einval # do not recurse
+ ld t2, sys_call_table(t1) # syscall routine

	move	a0, a1			# shift argument registers
	move	a1, a2

On Tue, Nov 18, 2008 at 03:05:46PM -0800, David Daney wrote:

Thanks, applied.

Ralf