Re: [PATCH] ftrace: enable trampoline when rec count decrement to one
From: chengjian (D)
Date: Wed May 08 2019 - 11:06:38 EST
On 2019/5/6 3:34, Steven Rostedt wrote:
Thanks for the patch. There was some race condition that prevented me
from doing this in the first place, but unfortunately, I don't remember
what that was :-/
I'll have to think about this before applying this patch.
Maybe there isn't a race condition, and I was just playing it safe, as
there was a race condition between switching from regs caller back to
non regs caller.
function tracer uses ftrace_caller() and livepatch uses
I can modify my testcase to trigger this race condition.
unshare_files (1) R I tramp: 0xffffffffc0008000 (klp_ftrace_handler+0x0/0xa0) ->ftrace_ops_assist_func+0x0/0xf0
[NOW, the rec caller is ftracer_regs_caller TRAMPOLINE]
echo unshare_files > /sys/kernel/debug/tracing/set_ftrace_filter
echo function > /sys/kernel/debug/tracing/current_tracer
unshare_files (2) R I ->ftrace_ops_list_func+0x0/0x170
[NOW, the rec caller is ftracer_regs_caller]
# disable livepatch
echo 0 > /sys/kernel/livepatch/klp_unshare_files/enabled
unshare_files (1) tramp: 0xffffffffc0005000 (function_trace_call+0x0/0x120) ->function_trace_call+0x0/0x120
[NOW, the rec caller is ftrace_caller TRAMPOLINE]
So, the caller switch from regs caller back to non regs caller
when disable the livepatch. Could this testcase cause the race
condition ? BUT, Nothing happened here.
What will happen when the race triggers ? What can I do.