Michael S. Tsirkin
2021-Sep-05 15:55 UTC
[PATCH net-next v5 0/6] virtio/vsock: introduce MSG_EOR flag for SEQPACKET
On Fri, Sep 03, 2021 at 03:30:13PM +0300, Arseny Krasnov wrote:> This patchset implements support of MSG_EOR bit for SEQPACKET > AF_VSOCK sockets over virtio transport. > First we need to define 'messages' and 'records' like this: > Message is result of sending calls: 'write()', 'send()', 'sendmsg()' > etc. It has fixed maximum length, and it bounds are visible using > return from receive calls: 'read()', 'recv()', 'recvmsg()' etc. > Current implementation based on message definition above. > Record has unlimited length, it consists of multiple message, > and bounds of record are visible via MSG_EOR flag returned from > 'recvmsg()' call. Sender passes MSG_EOR to sending system call and > receiver will see MSG_EOR when corresponding message will be processed. > Idea of patchset comes from POSIX: it says that SEQPACKET > supports record boundaries which are visible for receiver using > MSG_EOR bit. So, it looks like MSG_EOR is enough thing for SEQPACKET > and we don't need to maintain boundaries of corresponding send - > receive system calls. But, for 'sendXXX()' and 'recXXX()' POSIX says, > that all these calls operates with messages, e.g. 'sendXXX()' sends > message, while 'recXXX()' reads messages and for SEQPACKET, 'recXXX()' > must read one entire message from socket, dropping all out of size > bytes. Thus, both message boundaries and MSG_EOR bit must be supported > to follow POSIX rules. > To support MSG_EOR new bit was added along with existing > 'VIRTIO_VSOCK_SEQ_EOR': 'VIRTIO_VSOCK_SEQ_EOM'(end-of-message) - now it > works in the same way as 'VIRTIO_VSOCK_SEQ_EOR'. But 'VIRTIO_VSOCK_SEQ_EOR' > is used to mark 'MSG_EOR' bit passed from userspace. > This patchset includes simple test for MSG_EOR.I'm prepared to merge this for this window, but I'm not sure who's supposed to ack the net/vmw_vsock/af_vsock.c bits. It's a harmless variable renaming so maybe it does not matter. The rest is virtio stuff so I guess my tree is ok. Objections, anyone?> Arseny Krasnov(6): > virtio/vsock: rename 'EOR' to 'EOM' bit. > virtio/vsock: add 'VIRTIO_VSOCK_SEQ_EOR' bit. > vhost/vsock: support MSG_EOR bit processing > virtio/vsock: support MSG_EOR bit processing > af_vsock: rename variables in receive loop > vsock_test: update message bounds test for MSG_EOR > > drivers/vhost/vsock.c | 28 +++++++++++++---------- > include/uapi/linux/virtio_vsock.h | 3 ++- > net/vmw_vsock/af_vsock.c | 10 ++++---- > net/vmw_vsock/virtio_transport_common.c | 23 ++++++++++++------- > tools/testing/vsock/vsock_test.c | 8 ++++++- > 5 files changed, 45 insertions(+), 27 deletions(-) > > v4 -> v5: > - Move bitwise and out of le32_to_cpu() in 0003. > > v3 -> v4: > - 'sendXXX()' renamed to 'send*()' in 0002- commit msg. > - Comment about bit restore updated in 0003-. > - 'same' renamed to 'similar' in 0003- commit msg. > - u32 used instead of uint32_t in 0003-. > > v2 -> v3: > - 'virtio/vsock: rename 'EOR' to 'EOM' bit.' - commit message updated. > - 'VIRTIO_VSOCK_SEQ_EOR' bit add moved to separate patch. > - 'vhost/vsock: support MSG_EOR bit processing' - commit message > updated. > - 'vhost/vsock: support MSG_EOR bit processing' - removed unneeded > 'le32_to_cpu()', because input argument was already in CPU > endianness. > > v1 -> v2: > - 'VIRTIO_VSOCK_SEQ_EOR' is renamed to 'VIRTIO_VSOCK_SEQ_EOM', to > support backward compatibility. > - use bitmask of flags to restore in vhost.c, instead of separated > bool variable for each flag. > - test for EAGAIN removed, as logically it is not part of this > patchset(will be sent separately). > - cover letter updated(added part with POSIX description). > > Signed-off-by: Arseny Krasnov <arseny.krasnov at kaspersky.com> > -- > 2.25.1