Responsibilities for the semantics

The semantics for the asyncio function are split between the PFS and the LFS. Some of the features whose support might be ambiguous are discussed here. Refer to aio_suspend (BPX1ASP, BPX4ASP) — Wait for an asynchronous I/O request in z/OS UNIX System Services Programming: Assembler Callable Services Reference while reading this topic.

The LFS must handle the following:

The aiocb structure. The interface to the PFS is through the regular vnode operations, such as vn_rdwr and vn_sndrcv.
The returned information. The PFS should return 0 for a successful Part 1, and the normal functional values from Part 2. In particular, the LFS handles the EINPROGRESS return_code.
Scheduling the SRB and calling the I/O completion notification. This includes calling the user exit, posting an ECB, and sending a signal.
AioSync. This appears to the PFS as a normal synchronous operation (osi_asy1=osi_asy2=OFF).
AioOk2CompImd, for accept and connect. The osi_ok2compimd bit is always on in the PFS for vn_accept and vn_connect, so the PFS can always complete these operations immediately without calling osi_upda or osi_sched. osi_compimd should be turned on if the PFS does happen to complete these operations immediately.
The select and poll functions, which are already asynchronous with respect to the PFS. The PFS continues to call osi_selpost for the vfs_batsel and vn_select operations.

The PFS must handle or contribute to the support of:

AioOk2CompImd, for reads (including accept_and_recv) and writes, through support for osi_ok2compimd. Even when the PFS is able to complete a read or write type of operation immediately, it must still call osi_sched whenever osi_ok2compimd=off. See Asynchronous I/O flow details for details.
AioCallB4 and deferred buffer allocation, by not requiring the presence of the user's data buffers during Part 1, unless osi_ok2compimd=on; and by passing the length of data that is available to be received to osi_sched.
The ECANCELED Return_code, by failing a request with that return code when the request has been removed from a waiting queue because of vn_cancel. The race condition between vn_cancel and data arrival can only be resolved by the PFS.