Skip to content

nbd: Try for cleaner TLS shutdown

Eric Blake qemu-devel 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest3 patches 10 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/3crypto: Add qcrypto_tls_shutdown()crypto/tlssession.c include/crypto/tlssession.h qapi/crypto.json+66 -03 replies
2/3io: Support shutdown of TLS channelio/channel-tls.c+26 -14 replies
3/3nbd: Use shutdown(SHUT_WR) after last item sentblock/nbd.c nbd/client.c nbd/server.c+7 -13 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [PATCH 0/3] nbd: Try for cleaner TLS shutdown

I'm currently trying to debug a hang in 'nbdkit nbd' as a client when TLS is in use, when dealing with three processes:
nbdkit example1[1] <=tls=> nbdkit nbd[2] <=plain=> qemu-nbd --list[3]

In theory, when process [3] exits, the server side of process [2] is supposed to disconnect as client from process [1]. But we were hitting sporadic hangs where process [2] was stuck in a poll() while in gnutls cleanup code, which appears to be because it did not recognize a clean shutdown from process [3].

I do not yet have a strong reproducible test case (which might include adding strategic sleep()s in one-off compilations) to prove that the problem is an unclean TLS shutdown, but at the same time, the gnutls documentation is clear that without proper use of gnutls_bye(), a client cannot distinguish between clean end of communications and a malicious interruption in service. Shutting down the write side of a bi-directional socket as soon as possible makes it easier for a client to begin its own shutdown actions, and can help in avoiding deadlocks where both sides of a connection are waiting for the other side to close() the socket first. If I can come up with a reliable test case, this series may be a candidate for 5.0; but for now, I'm only proposing it for 5.1 (we've had more than one release where qemu was not exploiting the full power of gnutls_bye(), and even if that triggers poor interaction with other endpoints such as nbdkit, going one more release with the same behavior isn't making things worse).

Eric Blake (3):
crypto: Add qcrypto_tls_shutdown()
io: Support shutdown of TLS channel
nbd: Use shutdown(SHUT_WR) after last item sent

 qapi/crypto.json            | 15 +++++++++++++++
 include/crypto/tlssession.h | 24 ++++++++++++++++++++++++
 block/nbd.c                 |  1 +
 crypto/tlssession.c         | 27 +++++++++++++++++++++++++++
 io/channel-tls.c            | 27 ++++++++++++++++++++++++++-
 nbd/client.c                |  3 ++-
 nbd/server.c                |  4 ++++
 7 files changed, 99 insertions(+), 2 deletions(-)
Signature · 1 line
2.26.0.rc2