libvirt users - Apr 2018 - Re: error : virHashForEach:597 : Hash operation not allowed during iteration

❦  4 avril 2018 15:59 +0200, Michal Privoznik <mprivozn@redhat.com> :
>> I may have missed this suggestion. Maybe Dan only sent it to you?
>
> No, there is another thread where this issue is discussed:
>
> https://www.redhat.com/archives/libvir-list/2018-April/msg00190.html
>
>  In the
>> meantime, could I change the locks around virHashForEach() and similar
>> as read/write locks?
>> 
>
> You can do that locally, but as a patch it's very unlikely to be
> accepted upstream because we've introduced RW locks to be able to
access
> domain list from multiple threads.
Looking a bit more, the whole "iterating" bit is currently
unsafe. First, it is racy since it's "check then take". Two
threads may
check the value is false and start iterating both. Moreover, at some
places, it is not set, for example in virHashAddOrUpdateEntry(). So, it
would be possible to call this function and during the update to iterate
over the hash. This seems to be prevented only by callers using RW
lock. So, it seems we can just remove this "iterating" bool and just
check all users of these functions are using the appropriate lock.
-- 
Make sure all variables are initialised before use.
            - The Elements of Programming Style (Kernighan & Plauger)

❦  4 avril 2018 16:21 +0200, Vincent Bernat <bernat@luffy.cx> :
>> You can do that locally, but as a patch it's very unlikely to be
>> accepted upstream because we've introduced RW locks to be able to
access
>> domain list from multiple threads.
>
> Looking a bit more, the whole "iterating" bit is currently
> unsafe. First, it is racy since it's "check then take". Two
threads may
> check the value is false and start iterating both. Moreover, at some
> places, it is not set, for example in virHashAddOrUpdateEntry(). So, it
> would be possible to call this function and during the update to iterate
> over the hash. This seems to be prevented only by callers using RW
> lock. So, it seems we can just remove this "iterating" bool and
just
> check all users of these functions are using the appropriate lock.
My bad, I didn't see the use of table->current...
-- 
The only way to keep your health is to eat what you don't want, drink what
you don't like, and do what you'd rather not.
		-- Mark Twain

❦  4 avril 2018 17:00 +0200, Vincent Bernat <bernat@luffy.cx> :
>>> You can do that locally, but as a patch it's very unlikely to
be
>>> accepted upstream because we've introduced RW locks to be able
to access
>>> domain list from multiple threads.
>>
>> Looking a bit more, the whole "iterating" bit is currently
>> unsafe. First, it is racy since it's "check then take".
Two threads may
>> check the value is false and start iterating both. Moreover, at some
>> places, it is not set, for example in virHashAddOrUpdateEntry(). So, it
>> would be possible to call this function and during the update to
iterate
>> over the hash. This seems to be prevented only by callers using RW
>> lock. So, it seems we can just remove this "iterating" bool
and just
>> check all users of these functions are using the appropriate lock.
>
> My bad, I didn't see the use of table->current...
It seems it can also go away if caller correctly use a RW lock. I am
currently testing the attached patch on a few servers (patch for master,
but applied on our 3.6 version). After checking many locations, I came
to conclusion that locks are used correctly.




-- 
Use library functions.
            - The Elements of Programming Style (Kernighan & Plauger)