[PATCH v2 1/3] arm64: Prohibit instrumentation on arch_stack_walk()

[Mark Rutland]

On Fri, Dec 02, 2022 at 11:18:33AM +0900, Masami Hiramatsu (Google) wrote:
> From: Masami Hiramatsu (Google) <mhiramat at kernel.org>
> 
> Mark arch_stack_walk() as noinstr instead of notrace and inline functions
> called from arch_stack_walk() as __always_inline so that user does not
> put any instrumentations on it, because this function can be used from
> return_address() which is used by lockdep.

Hmmm... since arch_stack_walk is marked as notrace, that will be prohibited by
default unless the kernel was built with CONFIG_KPROBE_EVENTS_ON_NOTRACE=y,
and the commit message for that says:

│ This option disables such protection and allows you to put kprobe                                                                                                                                   │
│ events on ftrace functions for debugging ftrace by itself.                                                                                                                                          │
│ Note that this might let you shoot yourself in the foot.

... and IIUC we generally don't expect people to set that, and that might be
worth calling out in the commit message.

> Without this, if the kernel built with CONFIG_LOCKDEP=y, just probing
> arch_stack_walk() via <tracefs>/kprobe_events will crash the kernel on
> arm64.
> 
>  # echo p arch_stack_walk >> ${TRACEFS}/kprobe_events
>  # echo 1 > ${TRACEFS}/events/kprobes/enable

I had a go at testing this patch, and it fixes the crash with the reproducer
above, but there are plenty of other instances in stacktrace.c that lead to the
same sort of crash, e.g.

# echo p stackinfo_get_task >> ${TRACEFS}/kprobe_events
# echo 1 > ${TRACEFS}/events/kprobes/enable

... so I think there's more that we need to do to fix this generally.

Note: I found interesting functions to try tracing by looking at the output of:

  aarch64-linux-objdump -t arch/arm64/kernel/stacktrace.o | grep -w '.text'

That all said, I think this patch is nice-to-have, and that we can address the
other cases as a follow-up, so for this patch as-is (with or without some
wording regarding CONFIG_KPROBE_EVENTS_ON_NOTRACE):

Acked-by: Mark Rutland <mark.rutland at arm.com>

Mark.

>   kprobes: Failed to recover from reentered kprobes.
>   kprobes: Dump kprobe:
>   .symbol_name = arch_stack_walk, .offset = 0, .addr = arch_stack_walk+0x0/0x1c0
>   ------------[ cut here ]------------
>   kernel BUG at arch/arm64/kernel/probes/kprobes.c:241!
>   kprobes: Failed to recover from reentered kprobes.
>   kprobes: Dump kprobe:
>   .symbol_name = arch_stack_walk, .offset = 0, .addr = arch_stack_walk+0x0/0x1c0
>   ------------[ cut here ]------------
>   kernel BUG at arch/arm64/kernel/probes/kprobes.c:241!
>   PREEMPT SMP
>   Modules linked in:
>   CPU: 0 PID: 17 Comm: migration/0 Tainted: G                 N 6.1.0-rc5+ #6
>   Hardware name: linux,dummy-virt (DT)
>   Stopper: 0x0 <- 0x0
>   pstate: 600003c5 (nZCv DAIF -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
>   pc : kprobe_breakpoint_handler+0x178/0x17c
>   lr : kprobe_breakpoint_handler+0x178/0x17c
>   sp : ffff8000080d3090
>   x29: ffff8000080d3090 x28: ffff0df5845798c0 x27: ffffc4f59057a774
>   x26: ffff0df5ffbba770 x25: ffff0df58f420f18 x24: ffff49006f641000
>   x23: ffffc4f590579768 x22: ffff0df58f420f18 x21: ffff8000080d31c0
>   x20: ffffc4f590579768 x19: ffffc4f590579770 x18: 0000000000000006
>   x17: 5f6b636174735f68 x16: 637261203d207264 x15: 64612e202c30203d
>   x14: 2074657366666f2e x13: 30633178302f3078 x12: 302b6b6c61775f6b
>   x11: 636174735f686372 x10: ffffc4f590dc5bd8 x9 : ffffc4f58eb31958
>   x8 : 00000000ffffefff x7 : ffffc4f590dc5bd8 x6 : 80000000fffff000
>   x5 : 000000000000bff4 x4 : 0000000000000000 x3 : 0000000000000000
>   x2 : 0000000000000000 x1 : ffff0df5845798c0 x0 : 0000000000000064
>   Call trace:
>   kprobes: Failed to recover from reentered kprobes.
>   kprobes: Dump kprobe:
>   .symbol_name = arch_stack_walk, .offset = 0, .addr = arch_stack_walk+0x0/0x1c0
>   ------------[ cut here ]------------
>   kernel BUG at arch/arm64/kernel/probes/kprobes.c:241!
> 
> Fixes: 39ef362d2d45 ("arm64: Make return_address() use arch_stack_walk()")
> Cc: stable at vger.kernel.org
> Signed-off-by: Masami Hiramatsu (Google) <mhiramat at kernel.org>
> ---
>  Changes in v2:
>   - Use noinstr instead of NOKPROBE_SYMBOL()
>   - Use __always_inline because nokprobe_inline will be changed if
>     CONFIG_KPROBES=n.
>   - Fix indentation.
> ---
>  arch/arm64/kernel/stacktrace.c |   10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/arch/arm64/kernel/stacktrace.c b/arch/arm64/kernel/stacktrace.c
> index 634279b3b03d..117e2c180f3c 100644
> --- a/arch/arm64/kernel/stacktrace.c
> +++ b/arch/arm64/kernel/stacktrace.c
> @@ -23,8 +23,8 @@
>   *
>   * The regs must be on a stack currently owned by the calling task.
>   */
> -static inline void unwind_init_from_regs(struct unwind_state *state,
> -					 struct pt_regs *regs)
> +static __always_inline void unwind_init_from_regs(struct unwind_state *state,
> +						  struct pt_regs *regs)
>  {
>  	unwind_init_common(state, current);
>  
> @@ -58,8 +58,8 @@ static __always_inline void unwind_init_from_caller(struct unwind_state *state)
>   * duration of the unwind, or the unwind will be bogus. It is never valid to
>   * call this for the current task.
>   */
> -static inline void unwind_init_from_task(struct unwind_state *state,
> -					 struct task_struct *task)
> +static __always_inline void unwind_init_from_task(struct unwind_state *state,
> +						  struct task_struct *task)
>  {
>  	unwind_init_common(state, task);
>  
> @@ -186,7 +186,7 @@ void show_stack(struct task_struct *tsk, unsigned long *sp, const char *loglvl)
>  			: stackinfo_get_unknown();		\
>  	})
>  
> -noinline notrace void arch_stack_walk(stack_trace_consume_fn consume_entry,
> +noinline noinstr void arch_stack_walk(stack_trace_consume_fn consume_entry,
>  			      void *cookie, struct task_struct *task,
>  			      struct pt_regs *regs)
>  {
>