Re: [Qemu-devel] [PATCH 00/21] RFCv2: add Spice block device
On Fr, 2013-11-22 at 14:28 +0100, Marc-André Lureau wrote: Hi there, Even though there is no rush to review this series, it would be nice to get the first ~10 patches. It would make the next rebase easier, and the review smaller. Can you send the ready-to-go spice bits as separate patch series (after addressing alon's comment on patch #5)? Then I'll add them to the spice queue. thanks, Gerd
Re: [Qemu-devel] [PATCH 00/21] RFCv2: add Spice block device
Hi there, Even though there is no rush to review this series, it would be nice to get the first ~10 patches. It would make the next rebase easier, and the review smaller. cheers On Mon, Nov 18, 2013 at 1:25 PM, Marc-André Lureau marcandre.lur...@gmail.com wrote: Hi, The following patch series implement a Spice block device, which allows the client to redirect a block device using the NBD protocol, which greatly simplifies the Spice code by reusing an existing protocol, and allows sharing existing qemu NBD implementation. The backend only support read-only device atm (although it shouldn't be hard to add write support if necessary) Usage with a CDROM drive: -device ide-cd,drive=cd -drive if=none,id=cd,readonly,file=spicebd: The associated server and client bits are: http://lists.freedesktop.org/archives/spice-devel/2013-June/013608.html http://lists.freedesktop.org/archives/spice-devel/2013-November/015452.html http://lists.freedesktop.org/archives/spice-devel/2013-November/015431.html Caveats: This block device driver is a bit special, since it is successfully initialized with size 0, and once the client is connected (or want to change block device) it re-opens itself. For this to work, we allow a block driver to be open with an existing opaque data. We also save the associate device name in the block drivers. During migration, the source needs to be able to flush pending operations, so the Spice channel context must be in a running loop. A modification to the Spice server API allows to associate a particular channel with the AIO loop, and may be used in the future to associate channels with different context or athreads. However, the AIO context doesn't have timers yet. Since they aren't really needed for the NBD channel, it's not a problem. I have been told timers in AIO are on their way, so this could be updated later. Since the block driver state is not migrated, the destination needs to wait until the block driver is initialized before the VM can run. This is done with a simple hold count. It is also necessary to avoid extra media changed notifications, which is easily done by checking migration state. Marc-André Lureau (21): vscclient: do not add a socket watch if there is not data to send spice-char: remove unused field qmp_change_blockdev() remove unused has_format include: add missing config-host.h include char: add qemu_chr_fe_event() Split nbd block client code nbd: don't change socket block during negotiate nbd: pass export name as init argument nbd: make session_close() idempotent nbd: finish any pending coroutine nbd: avoid uninitialized warnings block: save the associated child name in BlockDriverState blockdev: add qmp_change_blockdev_int() block: extract make_snapshot() from bdrv_open() block: add snapshot.size option to avoid extra bdrv_open() block: learn to open a driver with a given opaque block: allow to call bdrv_open() with an opaque block: do not notify change during migration sysemu: add vm_start_hold/release spice-core: allow an interface to be in AIO context block: add spice block device backend block.c | 225 --- block/Makefile.objs | 3 +- block/nbd-client.c| 384 + block/nbd-client.h| 50 + block/nbd.c | 380 +++- block/spicebd.c | 536 ++ blockdev.c| 24 ++- hw/block/fdc.c| 8 +- hw/ide/core.c | 12 +- hw/scsi/scsi-disk.c | 11 +- hw/sd/sd.c| 6 +- include/block/block.h | 2 +- include/block/block_int.h | 1 + include/sysemu/blockdev.h | 5 +- include/sysemu/char.h | 10 + include/sysemu/sysemu.h | 2 + include/ui/qemu-spice.h | 4 +- libcacard/vscclient.c | 10 +- nbd.c | 1 - qemu-char.c | 9 +- qmp.c | 2 +- spice-qemu-char.c | 20 +- stubs/vm-stop.c | 5 + ui/spice-core.c | 62 +- vl.c | 17 ++ 25 files changed, 1320 insertions(+), 469 deletions(-) create mode 100644 block/nbd-client.c create mode 100644 block/nbd-client.h create mode 100644 block/spicebd.c -- 1.8.3.1 -- Marc-André Lureau
[Qemu-devel] [PATCH 00/21] RFCv2: add Spice block device
Hi, The following patch series implement a Spice block device, which allows the client to redirect a block device using the NBD protocol, which greatly simplifies the Spice code by reusing an existing protocol, and allows sharing existing qemu NBD implementation. The backend only support read-only device atm (although it shouldn't be hard to add write support if necessary) Usage with a CDROM drive: -device ide-cd,drive=cd -drive if=none,id=cd,readonly,file=spicebd: The associated server and client bits are: http://lists.freedesktop.org/archives/spice-devel/2013-June/013608.html http://lists.freedesktop.org/archives/spice-devel/2013-November/015452.html http://lists.freedesktop.org/archives/spice-devel/2013-November/015431.html Caveats: This block device driver is a bit special, since it is successfully initialized with size 0, and once the client is connected (or want to change block device) it re-opens itself. For this to work, we allow a block driver to be open with an existing opaque data. We also save the associate device name in the block drivers. During migration, the source needs to be able to flush pending operations, so the Spice channel context must be in a running loop. A modification to the Spice server API allows to associate a particular channel with the AIO loop, and may be used in the future to associate channels with different context or athreads. However, the AIO context doesn't have timers yet. Since they aren't really needed for the NBD channel, it's not a problem. I have been told timers in AIO are on their way, so this could be updated later. Since the block driver state is not migrated, the destination needs to wait until the block driver is initialized before the VM can run. This is done with a simple hold count. It is also necessary to avoid extra media changed notifications, which is easily done by checking migration state. Marc-André Lureau (21): vscclient: do not add a socket watch if there is not data to send spice-char: remove unused field qmp_change_blockdev() remove unused has_format include: add missing config-host.h include char: add qemu_chr_fe_event() Split nbd block client code nbd: don't change socket block during negotiate nbd: pass export name as init argument nbd: make session_close() idempotent nbd: finish any pending coroutine nbd: avoid uninitialized warnings block: save the associated child name in BlockDriverState blockdev: add qmp_change_blockdev_int() block: extract make_snapshot() from bdrv_open() block: add snapshot.size option to avoid extra bdrv_open() block: learn to open a driver with a given opaque block: allow to call bdrv_open() with an opaque block: do not notify change during migration sysemu: add vm_start_hold/release spice-core: allow an interface to be in AIO context block: add spice block device backend block.c | 225 --- block/Makefile.objs | 3 +- block/nbd-client.c| 384 + block/nbd-client.h| 50 + block/nbd.c | 380 +++- block/spicebd.c | 536 ++ blockdev.c| 24 ++- hw/block/fdc.c| 8 +- hw/ide/core.c | 12 +- hw/scsi/scsi-disk.c | 11 +- hw/sd/sd.c| 6 +- include/block/block.h | 2 +- include/block/block_int.h | 1 + include/sysemu/blockdev.h | 5 +- include/sysemu/char.h | 10 + include/sysemu/sysemu.h | 2 + include/ui/qemu-spice.h | 4 +- libcacard/vscclient.c | 10 +- nbd.c | 1 - qemu-char.c | 9 +- qmp.c | 2 +- spice-qemu-char.c | 20 +- stubs/vm-stop.c | 5 + ui/spice-core.c | 62 +- vl.c | 17 ++ 25 files changed, 1320 insertions(+), 469 deletions(-) create mode 100644 block/nbd-client.c create mode 100644 block/nbd-client.h create mode 100644 block/spicebd.c -- 1.8.3.1
