[PATCH] utilization of kprobe_mutex is incorrect on x86_64

[PATCH] utilization of kprobe_mutex is incorrect on x86_64

[Zhang, Yanmin]

[-- Attachment #1: Type: text/plain, Size: 312 bytes --]

 <<kprobe_incorrect_kprobe_mutex_2.6.14-rc2_x86_64.patch>> I found it
when reading the source codes. Basically, the bug could break
kprobe_insn_pages under multi-thread environment. PPC arch also has the
problem.
Here is the patch against x86_64.

Signed-off-by: Zhang Yanmin <Yanmin.zhang@intel.com>



[-- Attachment #2: kprobe_incorrect_kprobe_mutex_2.6.14-rc2_x86_64.patch --]
[-- Type: application/octet-stream, Size: 943 bytes --]

diff -Nraup linux-2.6.14-rc2/arch/x86_64/kernel/kprobes.c linux-2.6.14-rc2_fix/arch/x86_64/kernel/kprobes.c
--- linux-2.6.14-rc2/arch/x86_64/kernel/kprobes.c	2005-09-29 08:35:46.000000000 +0800
+++ linux-2.6.14-rc2_fix/arch/x86_64/kernel/kprobes.c	2005-09-29 08:36:27.000000000 +0800
@@ -77,9 +77,9 @@ static inline int is_IF_modifier(kprobe_
 int __kprobes arch_prepare_kprobe(struct kprobe *p)
 {
 	/* insn: must be on special executable page on x86_64. */
-	up(&kprobe_mutex);
-	p->ainsn.insn = get_insn_slot();
 	down(&kprobe_mutex);
+	p->ainsn.insn = get_insn_slot();
+	up(&kprobe_mutex);
 	if (!p->ainsn.insn) {
 		return -ENOMEM;
 	}
@@ -231,9 +231,9 @@ void __kprobes arch_disarm_kprobe(struct
 
 void __kprobes arch_remove_kprobe(struct kprobe *p)
 {
-	up(&kprobe_mutex);
-	free_insn_slot(p->ainsn.insn);
 	down(&kprobe_mutex);
+	free_insn_slot(p->ainsn.insn);
+	up(&kprobe_mutex);
 }
 
 static inline void save_previous_kprobe(void)

Re: [discuss] [PATCH] utilization of kprobe_mutex is incorrect on x86_64

[Andi Kleen]

On Thu, Sep 29, 2005 at 08:43:44AM +0800, Zhang, Yanmin wrote:
>  <<kprobe_incorrect_kprobe_mutex_2.6.14-rc2_x86_64.patch>> I found it
> when reading the source codes. Basically, the bug could break
> kprobe_insn_pages under multi-thread environment. PPC arch also has the
> problem.

Can you describe what the problem actually is? 

-Andi

> Here is the patch against x86_64.
> 
> Signed-off-by: Zhang Yanmin <Yanmin.zhang@intel.com>
> 
>