v3:
- Add aio_wait_kick() into nbd_read_eof() to avoid race with nbd_drained_poll()
[Kevin]
v2:
- Drop useless if (client->quiesing) check [Kevin]
- run qio_channel_read_wake() in export AioContext to avoid race [Kevin]
- Introduce nbd_client_put_nonzero() optimization [Paolo]
- Reach goto label disconnect with client->lock taken [Paolo]
- Add doc comment explaining nbd_co_receive_request() loop in nbd_trip() [Kevin]
Kevin merged several of my outstanding multi-queue block layer patch series and
found that qemu-iotests -qcow2 was broken. This patch series fixes the block branch.
Most of the fixes are easy but the NBD server required deeper debugging and
thread-safety fixes. The NBD server patches can be inserted before "aio: make
aio_context_acquire()/aio_context_release() a no-op" to preserve bisectability.
The other patches are fixups that can be squashed into the original patches.
Stefan Hajnoczi (6):
fixup block-coroutine-wrapper: use qemu_get_current_aio_context()
fixup block: remove AioContext locking
fixup scsi: only access SCSIDevice->requests from one thread
nbd/server: avoid per-NBDRequest nbd_client_get/put()
nbd/server: only traverse NBDExport->clients from main loop thread
nbd/server: introduce NBDClient->lock to protect fields
hw/scsi/scsi-bus.c | 3 +-
migration/block.c | 7 ++
nbd/server.c | 208 ++++++++++++++++++++++++++--------
tests/qemu-iotests/051.pc.out | 4 +-
4 files changed, 173 insertions(+), 49 deletions(-)
--
2.43.0