Re: [RFC PATCH v2 0/9] vsock: updates for SO_RCVLOWAT handling

From: Arseniy Krasnov
Date: Thu Jul 28 2022 - 02:09:06 EST


On 27.07.2022 15:37, Stefano Garzarella wrote:
> Hi Arseniy,
>
> On Mon, Jul 25, 2022 at 07:54:05AM +0000, Arseniy Krasnov wrote:
>> Hello,
>>
>> This patchset includes some updates for SO_RCVLOWAT:
>>
>> 1) af_vsock:
>>   During my experiments with zerocopy receive, i found, that in some
>>   cases, poll() implementation violates POSIX: when socket has non-
>>   default SO_RCVLOWAT(e.g. not 1), poll() will always set POLLIN and
>>   POLLRDNORM bits in 'revents' even number of bytes available to read
>>   on socket is smaller than SO_RCVLOWAT value. In this case,user sees
>>   POLLIN flag and then tries to read data(for example using  'read()'
>>   call), but read call will be blocked, because  SO_RCVLOWAT logic is
>>   supported in dequeue loop in af_vsock.c. But the same time,  POSIX
>>   requires that:
>>
>>   "POLLIN     Data other than high-priority data may be read without
>>               blocking.
>>    POLLRDNORM Normal data may be read without blocking."
>>
>>   See https://www.open-std.org/jtc1/sc22/open/n4217.pdf, page 293.
>>
>>   So, we have, that poll() syscall returns POLLIN, but read call will
>>   be blocked.
>>
>>   Also in man page socket(7) i found that:
>>
>>   "Since Linux 2.6.28, select(2), poll(2), and epoll(7) indicate a
>>   socket as readable only if at least SO_RCVLOWAT bytes are available."
>>
>>   I checked TCP callback for poll()(net/ipv4/tcp.c, tcp_poll()), it
>>   uses SO_RCVLOWAT value to set POLLIN bit, also i've tested TCP with
>>   this case for TCP socket, it works as POSIX required.
>>
>>   I've added some fixes to af_vsock.c and virtio_transport_common.c,
>>   test is also implemented.
>>
>> 2) virtio/vsock:
>>   It adds some optimization to wake ups, when new data arrived. Now,
>>   SO_RCVLOWAT is considered before wake up sleepers who wait new data.
>>   There is no sense, to kick waiter, when number of available bytes
>>   in socket's queue < SO_RCVLOWAT, because if we wake up reader in
>>   this case, it will wait for SO_RCVLOWAT data anyway during dequeue,
>>   or in poll() case, POLLIN/POLLRDNORM bits won't be set, so such
>>   exit from poll() will be "spurious". This logic is also used in TCP
>>   sockets.
>
> Nice, it looks good!
Thank You!
>
>>
>> 3) vmci/vsock:
>>   Same as 2), but i'm not sure about this changes. Will be very good,
>>   to get comments from someone who knows this code.
>
> I CCed VMCI maintainers to the patch and also to this cover, maybe better to keep them in the loop for next versions.
>
> (Jorgen's and Rajesh's emails bounced back, so I'm CCing here only Bryan, Vishnu, and pv-drivers@xxxxxxxxxx)
Ok, i'll CC them in the next version
>
>>
>> 4) Hyper-V:
>>   As Dexuan Cui mentioned, for Hyper-V transport it is difficult to
>>   support SO_RCVLOWAT, so he suggested to disable this feature for
>>   Hyper-V.
>
> I left a couple of comments in some patches, but it seems to me to be in a good state :-)
>
> I would just suggest a bit of a re-organization of the series (the patches are fine, just the order):
>   - introduce vsock_set_rcvlowat()
>   - disabling it for hv_sock
>   - use 'target' in virtio transports
>   - use 'target' in vmci transports
>   - use sock_rcvlowat in vsock_poll()
>     I think is better to pass sock_rcvlowat() as 'target' when the
>     transports are already able to use it
>   - add vsock_data_ready()
>   - use vsock_data_ready() in virtio transports
>   - use vsock_data_ready() in vmci transports
>   - tests
>
> What do you think?
No problem! I think i can wait for reply from VMWare guys before preparing v3
>
> Thanks,
> Stefano
>