[GIT PULL] uprobes: fix the handling of relative jmp/call's

From: Oleg Nesterov
Date: Thu Apr 17 2014 - 16:03:00 EST


Ingo, please pull from

git://git.kernel.org/pub/scm/linux/kernel/git/oleg/misc uprobes/core


To remind, we have a serious bug. Any probed jmp/call can kill the
application, see the changelog in 11/15.

I tried to test this as much as possible, seems to work. I also wrote
the test-case which explicitly checks every conditional jump with all
possible bits combinations in eflags:

#include <asm/processor-flags.h>
#include <stdio.h>

static int nojmp;

#define __MK_FUNC(opc, name) \
asm ( \
".text\n" \
".globl " #name "; " #name ":\n" \
".byte 0x" #opc "\n" \
".byte 0x0a \n" /* offs to 1f below */ \
"movl $1, nojmp(%rip)\n" \
"1: ret\n" \
); \

#define MK_FUNC(opc) __MK_FUNC(opc, probe_func_ ## opc)

MK_FUNC(70) MK_FUNC(71) MK_FUNC(72) MK_FUNC(73)
MK_FUNC(74) MK_FUNC(75) MK_FUNC(76) MK_FUNC(77)
MK_FUNC(78) MK_FUNC(79) MK_FUNC(7a) MK_FUNC(7b)
MK_FUNC(7c) MK_FUNC(7d) MK_FUNC(7e) MK_FUNC(7f)


#define __CALL(flags, func) \
asm volatile ("pushf; push %0; popf; call " #func "; popf" \
: : "m" (flags) : "memory");

#define CALL(opc) \
({ \
nojmp = 0; \
__CALL(flags, probe_func_ ## opc); \
!nojmp; \
})

long test_flags(unsigned long flags)
{
printf("%04lx %d %d %d %d %d %d %d %d %d %d %d %d %d %d %d %d\n",
flags,
CALL(70), CALL(71), CALL(72), CALL(73),
CALL(74), CALL(75), CALL(76), CALL(77),
CALL(78), CALL(79), CALL(7a), CALL(7b),
CALL(7c), CALL(7d), CALL(7e), CALL(7f));

return 0;
}

#define XF(xf) (X86_EFLAGS_ ## xf)
#define XF_MASK (XF(CF) | XF(OF) | XF(PF) | XF(SF) | XF(ZF))

int main(void)
{
unsigned int bits;
unsigned long __flags, flags;

asm volatile("pushf; pop %0" : "=rm" (__flags) : : "memory");

for (bits = 0; bits < (1 << 5); bits++) {
flags = __flags & ~XF_MASK;

#define CPY_BIT(nr, xf) \
if (bits & (1 << nr)) flags |= XF(xf)

CPY_BIT(0, CF);
CPY_BIT(1, OF);
CPY_BIT(2, PF);
CPY_BIT(3, SF);
CPY_BIT(4, ZF);

test_flags(flags);
}

return 0;
}

The output is the same with probe_func_70..probe_func_7f probed.

This series only fixes the problem. I'll send more changes to address
some of TODO's mentioned in the changelogs later. In particular, we
need to do something with "callw", see "Note: in 13/15.

Oleg Nesterov (15):
uprobes: Kill UPROBE_SKIP_SSTEP and can_skip_sstep()
uprobes/x86: Fold prepare_fixups() into arch_uprobe_analyze_insn()
uprobes/x86: Kill the "ia32_compat" check in handle_riprel_insn(), remove "mm" arg
uprobes/x86: Gather "riprel" functions together
uprobes/x86: move the UPROBE_FIX_{RIP,IP,CALL} code at the end of pre/post hooks
uprobes/x86: Introduce uprobe_xol_ops and arch_uprobe->ops
uprobes/x86: Conditionalize the usage of handle_riprel_insn()
uprobes/x86: Send SIGILL if arch_uprobe_post_xol() fails
uprobes/x86: Teach arch_uprobe_post_xol() to restart if possible
uprobes/x86: Introduce sizeof_long(), cleanup adjust_ret_addr() and arch_uretprobe_hijack_return_addr()
uprobes/x86: Emulate unconditional relative jmp's
uprobes/x86: Emulate nop's using ops->emulate()
uprobes/x86: Emulate relative call's
uprobes/x86: Emulate relative conditional "short" jmp's
uprobes/x86: Emulate relative conditional "near" jmp's

arch/x86/include/asm/uprobes.h | 16 +-
arch/x86/kernel/uprobes.c | 551 +++++++++++++++++++++++++---------------
kernel/events/uprobes.c | 31 +--
3 files changed, 372 insertions(+), 226 deletions(-)

--
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/