On 1. August 2014 15:51:45 GMT+01:00, Thomas Graf [off-list ref] wrote:

On 08/01/14 at 03:12pm, Patrick McHardy wrote:

quoted

On 1. August 2014 14:49:01 GMT+01:00, Thomas Graf [off-list ref]

wrote:

quoted

quoted

On 08/01/14 at 02:21pm, Nikolay Aleksandrov wrote:

quoted

Sorry for the late response but I just got around to check the 3rd

patch now,

quoted

so my question about this fragment is: before nft_hash_remove()

would

quoted

quoted

free the

quoted

element after removal, but after this change I don't see where and

when the

quoted

removed element would get freed ?

You are right. Looking at this closer we were leaking references

even

quoted

quoted

before the change because it would not call nft_data_uninit() to
release
the data objects. So it should really just call
nft_hash_elem_destroy().

No, the data uninit is performed by the set core. Just freeing it is

fine.

OK, I can see the uninit on the data in nft_del_setelem() for maps
prior to the actual commit. The location is slightly unexpected since
the element is still findable in the hash table at that point.

That's most likely a bug from the transaction API.

Pablo?

I'll wait a short while to see if there is more feedback and resubmit
afterwards.