Re: [PATCH] kprobes: disable preempt for module_text_address()

[Lai Jiangshan <laijs@cn.fujitsu.com>]

Ananth N Mavinakayanahalli wrote:
> On Tue, Nov 04, 2008 at 01:56:21PM +0800, Lai Jiangshan wrote:
>> __register_kprobe() may be preempted after module_text_address()
>> but before try_module_get(), and in this interval the module may be
>> unloaded and try_module_get(probed_mod) will access to invalid address.
>> this patch uses preempt_disable() to protect it.
> 
> Looking at other users of try_module_get, I don't see this as a usage
> model being followed elsewhere. Also, in case such a preemption does
> happen, module_is_live() will fail and we should still be ok.

when preemption happen, and mod is freed, module_is_live() will access to
invalid address. So it's NOT OK.

Other users of try_module_get() are correct. most are like this:

void func(XXX, XXXX)
{
	try_module_get(XXX->owner)
}

Because we have had a reference to the module before calling try_module_get().
this means the module is still in the kernel when try_module_get() called.
so we do not need any protection for using try_module_get().
<in other word, caller of func() has made sure the module will not be unloaded>

In this function __register_kprobe(), probed_mod is the return value of
module_text_address(), probed_mod will go in any time before try_module_get().

> 
> I don't see a reason for this patch unless there is a clear failure case
> (register_kprobe failing 'cos of a module unload is perfectly ok).
> 
> Ananth
>