buffer->store_block calls trigger->push_block when it receives the
first block. The trigger user fires directly , and if the device
accepts the data immediately, our data_done() will retr_block from the
buffer.  This happens with zio-zero, cset 1, and leads to a deadlock.

To allow immediate success of push_block down to device level, release
the lock before calling push_block. The code here also handles
pathologic situations like two processes storing at the same time.

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
Acked-by: Federico Vaga <federico.vaga@gmail.com>