Hi, we have a custom block driver downstream, which currently calls bdrv_get_info() (for its file child) in the bdrv_refresh_limits() callback. However, with graph locking, this doesn't work anymore. AFAICT, the reason is the following:
The block driver has a backing file option. During initialization, in bdrv_set_backing_hd(), the graph lock is acquired exclusively. Then the bdrv_refresh_limits() callback is invoked. Now bdrv_get_info() is called, which is a generated co-wrapper. The bdrv_co_get_info_entry() function tries to acquire the graph lock for reading, sees that has_writer is true and so the coroutine will be put to wait, leading to a deadlock. For my specific case, I can move the bdrv_get_info() call to bdrv_open() as a workaround. But I wanted to ask if there is a way to make generated co-wrappers inside an exclusively locked section work? And if not, could we introduce/extend the annotations, so the compiler can catch this kind of issue, i.e. calling a generated co-wrapper while in an exclusively locked section? Best Regards, Fiona
