Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 8 Mar 2016 14:38:12 -0700
From: Scotty Bauer <sbauer@....utah.edu>
To: One Thousand Gnomes <gnomes@...rguk.ukuu.org.uk>
Cc: linux-kernel@...r.kernel.org, kernel-hardening@...ts.openwall.com,
 x86@...nel.org, wmealing@...hat.com, ak@...ux.intel.com,
 luto@...capital.net, Abhiram Balasubramanian <abhiram@...utah.edu>
Subject: Re: [PATCH v3 2/3] x86: SROP mitigation: implement signal cookies



On 03/08/2016 02:03 PM, One Thousand Gnomes wrote:
>>  static int ia32_restore_sigcontext(struct pt_regs *regs,
>> -				   struct sigcontext_32 __user *sc)
>> +				   struct sigcontext_32 __user *sc,
>> +				   void __user **user_cookie)
>>  {
>>  	unsigned int tmpflags, err = 0;
>>  	void __user *buf;
>> @@ -105,6 +106,16 @@ static int ia32_restore_sigcontext(struct pt_regs *regs,
>>  		buf = compat_ptr(tmp);
>>  	} get_user_catch(err);
>>  
>> +	/*
>> +	 * If there is fp state get cookie from the top of the fp state,
>> +	 * else get it from the top of the sig frame.
>> +	 */
>> +
>> +	if (tmp != 0)
>> +		*user_cookie = compat_ptr(tmp + fpu__getsize(1));
>> +	else
>> +		*user_cookie = NULL;
> 
> user_cookie is is __user, so shouldn't just be poking at it without
> get/put_user ? It might fault if someone has engineered a bad stack frame.
> 
> Alan
> 

I guess I got a little carried away with the __user annotations. I will remove the __user annotation from this function
since what we're dereferencing isn't actually a __user pointer, its some stack memory sitting in the caller's stack frame. 

see patch 2/3:


	void __user *user_cookie;
                                                      vv-- Passed in here
	if (ia32_restore_sigcontext(regs, &frame->sc, &user_cookie))
		goto badframe;

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.