[CRIU] [PATCH] irmap: don't leak irmap objects in --irmap-scan-path

Pavel Emelyanov xemul at parallels.com
Tue Sep 22 12:25:14 PDT 2015


On 09/22/2015 10:13 PM, Tycho Andersen wrote:
> On Tue, Sep 22, 2015 at 09:47:07PM +0300, Pavel Emelyanov wrote:
>>
>>> @@ -268,9 +268,14 @@ char *irmap_lookup(unsigned int s_dev, unsigned long i_ino)
>>>  	list_for_each_entry(o, &opts.irmap_scan_paths, node) {
>>>  		struct irmap *ir;
>>>  
>>> -		ir = xzalloc(sizeof(*ir));
>>> -		if (!ir)
>>> -			goto out;
>>> +		if (!o->ir) {
>>> +			ir = xzalloc(sizeof(*ir));
>>> +			if (!ir)
>>> +				goto out;
>>> +			o->ir = ir;
>>> +		} else {
>>> +			ir = o->ir;
>>
>> OK, and since we keep the ir on o all the time, is there any reason why
>> we don't embed the struct irmap directly into the struct irmap_path_opt? :)
> 
> You mean just get rid of struct irmap_path_opt entirely? We could,
> although we'd have to add a list_head to irmap, or just use realloc to
> allocate the array or the argument list provided by --irmap-scan-path.
> I'm fine with either of those, let me know which you prefer.

Well, not get rid entirely :) I actually had smth like this in mind:

struct irmap_path_opt {
	struct ir;
	struct list_head list;
}

irmap_path_scan_add(char *path)
{
	irmap = xmalloc();
	irmap->ir.path = path;
	irmap->ir.nr_kids = -1;
	list_add(...);
}

and

irmap_lookup()
{
	...
	irmap_scan(&opt->ir, ...);
	...
}

--- Pavel


More information about the CRIU mailing list