Re: [PATCH] dcache: Translating dentry into pathname without taking rename_lock

[Waiman Long <waiman.long@hp.com>]

On 09/04/2013 03:11 PM, Al Viro wrote:
> On Wed, Sep 04, 2013 at 03:05:23PM -0400, Waiman Long wrote:
>>
>>   static int prepend_name(char **buffer, int *buflen, struct qstr *name)
>>   {
>> -	return prepend(buffer, buflen, name->name, name->len);
>> +	/*
>> +	 * With RCU path tracing, it may race with rename. Use
>> +	 * ACCESS_ONCE() to make sure that it is either the old or
>> +	 * the new name pointer. The length does not really matter as
>> +	 * the sequence number check will eventually catch any ongoing
>> +	 * rename operation.
>> +	 */
>> +	const char *dname = ACCESS_ONCE(name->name);
>> +	int   dlen = name->len;
>> +
>> +	if (unlikely(!dname || !dlen))
>> +		return -EINVAL;
>> +	return prepend(buffer, buflen, dname, dlen);
> NAK.  A race with d_move() can very well leave you with dname pointing into
> an object of length smaller than dlen.  You *can* copy it byte-by-byte
> and rely on NUL-termination, but you can't rely on length being accurate -
> not without having excluded d_move().

I have thought about that. But if a d_move() is going on, the string in 
the buffer will be discarded as the sequence number will change. So 
whether or not it have embedded null byte shouldn't matter. That is why 
I didn't add code to do byte-by-byte copy at this first patch. I can add 
code to do that if you think it is safer to do so.

Regards,
Longman