kprobe: sync issue's on ftraced-kprobe.
PATCH 1/1 kprobe: sync issue's on ftraced-kprobe.
In kprobe_ftrace_handler, it accesses get kporbe without kprobe_mutex held.
This makes some of synchronizing issue when we use kprobe API in kernel-module.
Below is what i experienced:
CPU 0 CPU 1
<...> <In module code>kprobe_ftrace_handler
get_kprobe
__this_cpu_write
unregister_kprobe unload_module
< kprobe memory gone>
p->pre_handler <access invalid memory>
page_fault
kprobe_fault_handler
(In here, kprobe memory gone,
double page fault is happening inifinie).
--- arch/x86/kernel/kprobes/ftrace.c | 3 +++ include/linux/kprobes.h | 2 ++ kernel/kprobes.c | 2 +- 3 files changed, 6 insertions(+), 1 deletion(-)
arch/x86/kernel/kprobes/ftrace.c +3 -0
@@ -25,6 +25,7 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,if (bit <0 )return ; +mutex_lock (&kprobe_mutex); p =get_kprobe ((kprobe_opcode_t *)ip);if (unlikely (!p) ||kprobe_disabled (p)) goto out;
@@ -57,7 +58,9 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,*/ __this_cpu_write (current_kprobe,NULL ); } + out: +mutex_unlock (&kprobe_mutex);ftrace_test_recursion_unlock (bit); }NOKPROBE_SYMBOL (kprobe_ftrace_handler);
include/linux/kprobes.h +2 -0
@@ -191,6 +191,8 @@ struct kprobe_blacklist_entry { DECLARE_PER_CPU(struct kprobe *, current_kprobe); DECLARE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); +extern struct mutex kprobe_mutex; +extern void kprobe_busy_begin (void );extern void kprobe_busy_end (void );
kernel/kprobes.c +1 -1
@@ -64,7 +64,7 @@ static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];static bool kprobes_all_disarmed;/* This protects 'kprobe_table' and 'optimizing_list' */ -static DEFINE_MUTEX (kprobe_mutex); +DEFINE_MUTEX (kprobe_mutex); static DEFINE_PER_CPU(struct kprobe *, kprobe_instance); kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,
Signature · 1 line
2.35.1
Masami Hiramatsu
maintainerQuoted 5 lines · On Mon, 2 May 2022 13:51:02 +0900 Levi Yun <[email protected]> wrote:
In
kprobe_ftrace_handler, it accesses get kporbe withoutkprobe_mutexheld.This makes some of synchronizing issue when we use kprobe API in kernel-module.
NAK this, because get_kprobes() doesn't require the kprobe_mutex in the preempt-disabled context. Please read the comment of get_kprobe().
/*
* This routine is called either:
* - under the 'kprobe_mutex' - during kprobe_[un]register().
* OR
* - with preemption disabled - from architecture specific code.
*/
struct kprobe *get_kprobe(void *addr)
Moreover, we can not use mutex inside kprobe handler because it runs in the interrupt context.
Thank you,
Quoted 75 lines
Below is what i experienced:
CPU 0 CPU 1
<...> <In module code>kprobe_ftrace_handler
get_kprobe
__this_cpu_write
unregister_kprobeunload_module
< kprobe memory gone>
p->pre_handler<access invalid memory>
page_fault
kprobe_fault_handler
(In here, kprobe memory gone,
double page fault is happening inifinie).Signed-off-by Levi Yun <[email protected]>--- arch/x86/kernel/kprobes/ftrace.c | 3 +++ include/linux/kprobes.h | 2 ++ kernel/kprobes.c | 2 +- 3 files changed, 6 insertions(+), 1 deletion(-)arch/x86/kernel/kprobes/ftrace.c +3
-0@@ -25,6 +25,7 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,if (bit <0 )return ; +mutex_lock (&kprobe_mutex); p =get_kprobe ((kprobe_opcode_t *)ip);if (unlikely (!p) ||kprobe_disabled (p)) goto out;@@ -57,7 +58,9 @@ void kprobe_ftrace_handler(unsigned long ip, unsigned long parent_ip,*/ __this_cpu_write (current_kprobe,NULL ); } + out: +mutex_unlock (&kprobe_mutex);ftrace_test_recursion_unlock (bit); }NOKPROBE_SYMBOL (kprobe_ftrace_handler);include/linux/kprobes.h +2
-0@@ -191,6 +191,8 @@ struct kprobe_blacklist_entry { DECLARE_PER_CPU(struct kprobe *, current_kprobe); DECLARE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk); +extern struct mutex kprobe_mutex; +extern void kprobe_busy_begin (void );extern void kprobe_busy_end (void );kernel/kprobes.c +1
-1@@ -64,7 +64,7 @@ static struct hlist_head kprobe_table[KPROBE_TABLE_SIZE];static bool kprobes_all_disarmed;/* This protects 'kprobe_table' and 'optimizing_list' */ -static DEFINE_MUTEX (kprobe_mutex); +DEFINE_MUTEX (kprobe_mutex); static DEFINE_PER_CPU(struct kprobe *, kprobe_instance); kprobe_opcode_t * __weak kprobe_lookup_name(const char *name,Signature · 1 line
2.35.1
Signature · 1 line
Masami Hiramatsu <[email protected]>
Quoted 11 lines
NAK this, because
get_kprobes()doesn't require thekprobe_mutexin the preempt-disabled context. Please read the comment ofget_kprobe()./*
* This routine is called either:
* - under the 'kprobe_mutex' - during kprobe_[un]register().
* OR
* - with preemption disabled - from architecture specific code.
*/
struct kprobe *get_kprobe(void *addr)
Quoted 2 lines
Moreover, we can not use mutex inside kprobe handler because it runs in the interrupt context.
Actually, I think it's in the preemption disabled in situation. What I mentioned is actually it seems inside the kprobe handler which installed via ftrace isn't interrupt context.
#ifdef CONFIG_KPROBES_ON_FTRACE
static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
Quoted 2 lines
---.func =
kprobe_ftrace_handler,
---.flags =FTRACE_OPS_FL_SAVE_REGS,
};
In case of optimized kprobe it disables preemption when it enter "optimized callback"
And the trampoline isn't the interrupted context.
It's also the same to install kprobe via ftraced .
When I see the ftraced patched code, I couldn't find any interrupt and preemption disable code.
But it seems to allow preemption in ftarce handler.
(In the ftrace_reg_caller, ftace_reg_end, and other ftrace trampoline code also)
What I understand is break point based kprobe is only interrupted disabled status
But optimized kprobe and ftarced kprobe isn't interrupted disabled status. For this, optimized kprobe disable disable preemption in "optimized_callback"
But ftrace kprobe isn't preemption disabled nor interrupt-disabled context.
That's why I sent patches and I experienced the problem above.
Am I missing something?
Thx.