From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:47567) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1YyHuw-0004vs-LZ for qemu-devel@nongnu.org; Fri, 29 May 2015 06:53:35 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1YyHuv-0006ta-5X for qemu-devel@nongnu.org; Fri, 29 May 2015 06:53:34 -0400 From: Fam Zheng Date: Fri, 29 May 2015 18:53:13 +0800 Message-Id: <1432896805-23867-1-git-send-email-famz@redhat.com> Subject: [Qemu-devel] [RFC PATCH 00/12] block: Protect block jobs with lock / unlock API List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Cc: Kevin Wolf , fam@euphon.net, qemu-block@nongnu.org, Jeff Cody , Stefan Hajnoczi , Paolo Bonzini This is the partial work to introduce bdrv_lock / bdrv_unlock and use them in block jobs where exclusive access to a BDS is necessary. It address the same category of problems as [1] with a different API, as the idea proposed by Paolo and Kevin. What's implemented in this series is also very close to [1], i.e. pausing ioeventfd and NBD server, with a notifier list. The important missing pieces are converting bdrv_lock/unlock callers to coroutine, so that they can wait in the bs->lock_queue. In other words, this API is not complete without that, because we can't handle the case where two non-coroutine callers have contention, which is now asserted in bdrv_lock as practically impossible. [1]: https://lists.gnu.org/archive/html/qemu-block/2015-05/msg00800.html Thanks, Fam Fam Zheng (12): block: Use bdrv_drain to replace uncessary bdrv_drain_all block: Introduce bdrv_lock and bdrv_unlock API blockdev: Lock BDS during internal snapshot transaction blockdev: Lock BDS during external snapshot transaction blockdev: Lock BDS during drive-backup transaction blockdev: Lock BDS during blockdev-backup transaction block-backend: Add blk_add_lock_unlock_notifier virtio-blk: Move complete_request to 'ops' structure virtio-blk: Don't handle output when backend is locked virtio-scsi-dataplane: Add backend lock listener nbd-server: Clear "can_read" when backend is locked mirror: Protect source between bdrv_drain and bdrv_swap block.c | 16 +++++++-- block/block-backend.c | 6 ++++ block/io.c | 69 ++++++++++++++++++++++++++++++++++++ block/mirror.c | 18 ++++++++-- block/snapshot.c | 2 +- blockdev.c | 17 +++++++-- hw/block/dataplane/virtio-blk.c | 36 ++++++++++++++++--- hw/block/virtio-blk.c | 63 +++++++++++++++++++++++++++++++-- hw/scsi/virtio-scsi-dataplane.c | 78 ++++++++++++++++++++++++++++++----------- hw/scsi/virtio-scsi.c | 3 ++ include/block/block.h | 39 +++++++++++++++++++++ include/block/block_int.h | 5 +++ include/hw/virtio/virtio-blk.h | 17 +++++++-- include/hw/virtio/virtio-scsi.h | 3 ++ include/sysemu/block-backend.h | 1 + migration/block.c | 2 +- nbd.c | 21 +++++++++++ 17 files changed, 356 insertions(+), 40 deletions(-) -- 2.4.2