Matt Helsley wrote:
> On Fri, Oct 02, 2009 at 03:38:16PM -0400, Oren Laadan wrote:
>>>>> I should probably change the code elsewhere too, but this test
>>>>> is unnecessary because ckpt_obj_fetch() will complain anyway.
>>>> I don't think it will complain since I don't see anything in the read or hash
>>>> code that checks for negative objrefs. However moving this into
>>> What is "this" that you want to move ?
>>>
>>>> ckpt_obj_fetch() would get rid of alot of code much like passing NULL into
>>>> kfree() does, so I'll remove this test.
>>> ckpt_obj_fetch() won't complain about a negative value a-priori, but
>>> the search is certain to fail. Nevertheless, I'll tighten the restart
>>> related calls in objhash to ensure that.
>> I take it back: ckpt_obj_fetch() returns -EINVAL when an objref isn't
>> found, not that the original value was invalid.
>
> Right, so then the question is whether it's possible to insert a negative
> objref by modifying the checkpoint image before restart. As far as I can tell
> that will work. Are you saying we don't need to care about that?
You are correct, and I said I'll fix that ("Nevertheless..."). Fix
is queued already for ckpt-v18-dev.
I took back the incorrect suggestion that it's ok to _always_ let
ckpt_obj_fetch() do the work, as it depends on the context:
- if you use it to fetch an object you expect to already have been
inserted, then it's good enough.
- if you use it to fetch an object to decide if it's the first
encounter of that object, then you need to explicitly test for
invalid @objref before calling ckpt_obj_fetch().
Oren.
_______________________________________________
Containers mailing list
Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/containers
