On 3/23/22 2:52 PM, Pavel Begunkov wrote: > On 3/23/22 20:45, Constantine Gavrilov wrote: >> On Wed, Mar 23, 2022 at 10:14 PM Pavel Begunkov <asml.silence@xxxxxxxxx> wrote: >>> >>> On 3/23/22 15:39, Jens Axboe wrote: >>>> We currently don't attempt to get the full asked for length even if >>>> MSG_WAITALL is set, if we get a partial receive. If we do see a partial >>>> receive, then just note how many bytes we did and return -EAGAIN to >>>> get it retried. >>>> >>>> The iov is advanced appropriately for the vector based case, and we >>>> manually bump the buffer and remainder for the non-vector case. >>> >>> How datagrams work with MSG_WAITALL? I highly doubt it coalesces 2+ >>> packets to satisfy the length requirement (e.g. because it may move >>> the address back into the userspace). I'm mainly afraid about >>> breaking io_uring users who are using the flag just to fail links >>> when there is not enough data in a packet. >>> >>> -- >>> Pavel Begunkov >> >> Pavel: >> >> Datagrams have message boundaries and the MSG_WAITALL flag does not >> make sense there. I believe it is ignored by receive code on daragram >> sockets. MSG_WAITALL makes sends only on stream sockets, like TCP. The >> manual page says "This flag has no effect for datagram sockets.". > > Missed the line this in mans, thanks, and it's exactly as expected. > The problem is on the io_uring side where with the patch it might > blindly do a second call into the network stack consuming 2+ packets. Right, it should not be applied for datagrams. -- Jens Axboe