On 01.12.2015 14:16, Kevin Wolf wrote:
> Am 30.11.2015 um 18:22 hat Max Reitz geschrieben:
>> On 30.11.2015 16:36, Kevin Wolf wrote:
>>> Am 09.11.2015 um 23:39 hat Max Reitz geschrieben:
>>>> The NBD code uses the BDS close notifier to determine when a medium is
>>>> ejected. However, now it should use the BB's BDS removal notifier for
>>>> that instead of the BDS's close notifier.
>>>>
>>>> Signed-off-by: Max Reitz <mreitz@redhat.com>
>>>> ---
>>>>  blockdev-nbd.c | 37 +------------------------------------
>>>>  nbd.c          | 13 +++++++++++++
>>>>  2 files changed, 14 insertions(+), 36 deletions(-)
>>>>
>>>> diff --git a/blockdev-nbd.c b/blockdev-nbd.c
>>>> index bcdd18b..b28a55b 100644
>>>> --- a/blockdev-nbd.c
>>>> +++ b/blockdev-nbd.c
>>>> @@ -46,37 +46,11 @@ void qmp_nbd_server_start(SocketAddress *addr, Error **errp)
>>>>      }
>>>>  }
>>>>  
>>>> -/*
>>>> - * Hook into the BlockBackend notifiers to close the export when the
>>>> - * backend is closed.
>>>> - */
>>>> -typedef struct NBDCloseNotifier {
>>>> -    Notifier n;
>>>> -    NBDExport *exp;
>>>> -    QTAILQ_ENTRY(NBDCloseNotifier) next;
>>>> -} NBDCloseNotifier;
>>>> -
>>>> -static QTAILQ_HEAD(, NBDCloseNotifier) close_notifiers =
>>>> -    QTAILQ_HEAD_INITIALIZER(close_notifiers);
>>>> -
>>>> -static void nbd_close_notifier(Notifier *n, void *data)
>>>> -{
>>>> -    NBDCloseNotifier *cn = DO_UPCAST(NBDCloseNotifier, n, n);
>>>> -
>>>> -    notifier_remove(&cn->n);
>>>> -    QTAILQ_REMOVE(&close_notifiers, cn, next);
>>>> -
>>>> -    nbd_export_close(cn->exp);
>>>> -    nbd_export_put(cn->exp);
>>>
>>> Here you remove a close/put pair, but in the new code you only add the
>>> close call. Why is this correct?
>>
>> Thanks for putting so much unfounded faith in me. :-)
>>
>> After having put quite some time into looking into it, first I have to
>> say that it's a very good question.
>>
>> First off, it's wrong. This is because I thought every export would have
>> a refcount of one. Turns out this is wrong, they have a refcount of two:
>> It is created with a refcount of one, and then it gains another by
>> giving it a name and entering it into the export list.
>>
>> I did know about the second but I completely forgot about the former.
>> And indeed I do think it is wrong. qmp_nbd_server_add() does not keep
>> the reference to the export, once the function returns, it is gone.
>> Therefore, it should call nbd_export_put() before returning.
>>
>> So in my opinion the current code is wrong and I didn't fix it enough.
>> *cough*
>>
>> So, with the nbd_export_put() added to qmp_nbd_server_add(), every
>> export will have a refcount of 1 + |clients|. The eject notifier will
>> call nbd_close_notifier(), which first disconnects the clients, thus
>> reducing the refcount by |clients|, and then sets the name to NULL, thus
>> dropping the final refcount and destroying the export.
>>
>> In the old code, we needed another nbd_export_put() because of the
>> superfluous reference from nbd_export_new(), which doesn't actually
>> exist for blockdev-nbd (it does for qemu-nbd, though).
> 
> Okay, that makes sense to me. So first a patch that moves the existing
> nbd_export_put() call from nbd_close_notifier() to qmp_nbd_server_add()
> and then this one on top?

I would have put both in the same patch, but this does make more sense.
I'll do so.

Max