Message ID | 1455721691-14802-1-git-send-email-den@openvz.org |
---|---|
State | New |
Headers | show |
On 17.02.2016 16:08, Denis V. Lunev wrote: > There is VM with 8 GB QCOW2 storage. Real size of the image file is 2 GB. > 'drive-mirror' redirected to NBD target creates 8 GB image at destination. > The situation is even worse as zeroes are sent through the channel. > > The patch simply adds .bdrv_co_write_zeroes callback to NBD block driver > which works though NBD_TRIM to avoid transfer of zeroes. The specification[1] says the following about TRIM: > After issuing this command, a client MUST NOT make any assumptions > about the contents of the export affected by this command, until > overwriting it again with NBD_CMD_WRITE. So I don't think this is correct. The correct solution would probably to introduce a specific command to write zeroes, or at least to use detect-zeroes on the NBD server side (but this will still lead to the zeroes being sent over the line). Max [1] https://github.com/yoe/nbd/blob/master/doc/proto.md > Signed-off-by: Denis V. Lunev <den@openvz.org> > CC: Paolo Bonzini <pbonzini@redhat.com> > --- > block/nbd-client.c | 10 ++++++++++ > block/nbd-client.h | 2 ++ > block/nbd.c | 9 +++++++++ > 3 files changed, 21 insertions(+)
On 02/17/2016 06:21 PM, Max Reitz wrote: > On 17.02.2016 16:08, Denis V. Lunev wrote: >> There is VM with 8 GB QCOW2 storage. Real size of the image file is 2 GB. >> 'drive-mirror' redirected to NBD target creates 8 GB image at destination. >> The situation is even worse as zeroes are sent through the channel. >> >> The patch simply adds .bdrv_co_write_zeroes callback to NBD block driver >> which works though NBD_TRIM to avoid transfer of zeroes. > The specification[1] says the following about TRIM: > >> After issuing this command, a client MUST NOT make any assumptions >> about the contents of the export affected by this command, until >> overwriting it again with NBD_CMD_WRITE. > So I don't think this is correct. > > The correct solution would probably to introduce a specific command to > write zeroes, or at least to use detect-zeroes on the NBD server side > (but this will still lead to the zeroes being sent over the line). > > Max > > [1] https://github.com/yoe/nbd/blob/master/doc/proto.md > >> Signed-off-by: Denis V. Lunev <den@openvz.org> >> CC: Paolo Bonzini <pbonzini@redhat.com> >> --- >> block/nbd-client.c | 10 ++++++++++ >> block/nbd-client.h | 2 ++ >> block/nbd.c | 9 +++++++++ >> 3 files changed, 21 insertions(+) > This sounds not that good. OK. At least there is correct lengthy way for this. White-outs must be supported by the protocol. Den
diff --git a/block/nbd-client.c b/block/nbd-client.c index 568c56c..ef0c5e8 100644 --- a/block/nbd-client.c +++ b/block/nbd-client.c @@ -344,7 +344,17 @@ int nbd_client_co_discard(BlockDriverState *bs, int64_t sector_num, } nbd_coroutine_end(client, &request); return -reply.error; +} + +int nbd_client_co_write_zeroes(BlockDriverState *bs, int64_t sector_num, + int nb_sectors, BdrvRequestFlags flags) +{ + NbdClientSession *client = nbd_get_client_session(bs); + if (!(client->nbdflags & NBD_FLAG_SEND_TRIM)) { + return -ENOTSUP; + } + return nbd_client_co_discard(bs, sector_num, nb_sectors); } void nbd_client_detach_aio_context(BlockDriverState *bs) diff --git a/block/nbd-client.h b/block/nbd-client.h index e841340..aaef18d 100644 --- a/block/nbd-client.h +++ b/block/nbd-client.h @@ -40,6 +40,8 @@ void nbd_client_close(BlockDriverState *bs); int nbd_client_co_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors); +int nbd_client_co_write_zeroes(BlockDriverState *bs, int64_t sector_num, + int nb_sectors, BdrvRequestFlags flags); int nbd_client_co_flush(BlockDriverState *bs); int nbd_client_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors, QEMUIOVector *qiov); diff --git a/block/nbd.c b/block/nbd.c index 1a90bc7..8b05480 100644 --- a/block/nbd.c +++ b/block/nbd.c @@ -318,6 +318,13 @@ static int nbd_co_discard(BlockDriverState *bs, int64_t sector_num, return nbd_client_co_discard(bs, sector_num, nb_sectors); } +static coroutine_fn int nbd_co_write_zeroes(BlockDriverState *bs, + int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) +{ + return nbd_client_co_write_zeroes(bs, sector_num, nb_sectors, flags); +} + + static void nbd_close(BlockDriverState *bs) { nbd_client_close(bs); @@ -415,6 +422,7 @@ static BlockDriver bdrv_nbd_tcp = { .bdrv_close = nbd_close, .bdrv_co_flush_to_os = nbd_co_flush, .bdrv_co_discard = nbd_co_discard, + .bdrv_co_write_zeroes = nbd_co_write_zeroes, .bdrv_refresh_limits = nbd_refresh_limits, .bdrv_getlength = nbd_getlength, .bdrv_detach_aio_context = nbd_detach_aio_context, @@ -433,6 +441,7 @@ static BlockDriver bdrv_nbd_unix = { .bdrv_close = nbd_close, .bdrv_co_flush_to_os = nbd_co_flush, .bdrv_co_discard = nbd_co_discard, + .bdrv_co_write_zeroes = nbd_co_write_zeroes, .bdrv_refresh_limits = nbd_refresh_limits, .bdrv_getlength = nbd_getlength, .bdrv_detach_aio_context = nbd_detach_aio_context,
There is VM with 8 GB QCOW2 storage. Real size of the image file is 2 GB. 'drive-mirror' redirected to NBD target creates 8 GB image at destination. The situation is even worse as zeroes are sent through the channel. The patch simply adds .bdrv_co_write_zeroes callback to NBD block driver which works though NBD_TRIM to avoid transfer of zeroes. Signed-off-by: Denis V. Lunev <den@openvz.org> CC: Paolo Bonzini <pbonzini@redhat.com> --- block/nbd-client.c | 10 ++++++++++ block/nbd-client.h | 2 ++ block/nbd.c | 9 +++++++++ 3 files changed, 21 insertions(+)