The dirty bitmap functions are called from the block I/O processing
code. Make them visible to block_int.h users so they can be used
outside block.c.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For "dirty-bitmap" sync mode, the block job will iterate through the
given dirty bitmap to decide if a sector needs backup (backup all the
dirty clusters and skip clean ones), just as allocation conditions of
"top" sync mode.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-11-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We treat this field with a variety of different types everywhere
in the code. Now it's just uint32_t.
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1429314609-29776-4-git-send-email-jsnow@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add driver functions for geometry and blocksize detection
Signed-off-by: Ekaterina Tumanova <tumanova@linux.vnet.ibm.com>
Reviewed-by: Thomas Huth <thuth@linux.vnet.ibm.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1424087278-49393-2-git-send-email-tumanova@linux.vnet.ibm.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add a creation option to qcow2 for setting the refcount order of images
to be created, and respect that option's value.
This breaks some test outputs, fix them.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Previously, qemu block driver of sheepdog used hard-coded VDI object size.
This patch enables users to handle VDI object size.
When you start qemu, you don't need to specify additional command option.
But when you create the VDI which doesn't have default object size
with qemu-img command, you specify object_size option.
If you want to create a VDI of 8MB object size,
you need to specify following command option.
# qemu-img create -o object_size=8M sheepdog:test1 100M
In addition, when you don't specify qemu-img command option,
a default value of sheepdog cluster is used for creating VDI.
# qemu-img create sheepdog:test2 100M
Signed-off-by: Teruaki Ishizaki <ishizaki.teruaki@lab.ntt.co.jp>
Acked-by: Hitoshi Mitake <mitake.hitoshi@lab.ntt.co.jp>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that request clamping is done in the BlockBackend, the "growable"
field can be removed from the BlockDriverState. All BDSs are now treated
as being "growable" (that is, they are allowed to grow; they are not
necessarily actually able to).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1423162705-32065-16-git-send-email-mreitz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Managing applications, like oVirt (http://www.ovirt.org), make extensive
use of thin-provisioned disk images.
To let the guest run smoothly and be not unnecessarily paused, oVirt sets
a disk usage threshold (so called 'high water mark') based on the occupation
of the device, and automatically extends the image once the threshold
is reached or exceeded.
In order to detect the crossing of the threshold, oVirt has no choice but
aggressively polling the QEMU monitor using the query-blockstats command.
This lead to unnecessary system load, and is made even worse under scale:
deployments with hundreds of VMs are no longer rare.
To fix this, this patch adds:
* A new monitor command `block-set-write-threshold', to set a mark for
a given block device.
* A new event `BLOCK_WRITE_THRESHOLD', to report if a block device
usage exceeds the threshold.
* A new `write_threshold' field into the `BlockDeviceInfo' structure,
to report the configured threshold.
This will allow the managing application to use smarter and more
efficient monitoring, greatly reducing the need of polling.
[Updated qemu-iotests 067 output to add the new 'write_threshold'
property. --Stefan]
[Changed g_assert_false() to !g_assert() to fix the build on older glib
versions. --Kevin]
Signed-off-by: Francesco Romani <fromani@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 1421068273-692-1-git-send-email-fromani@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The string field entries 'filename', 'backing_file', and
'exact_filename' in the BlockDriverState struct are defined as 1024
bytes.
However, many places that use these values accept a maximum of PATH_MAX
bytes, so we have a mixture of 1024 byte and PATH_MAX byte allocations.
This patch makes the BlockDriverStruct field string sizes match usage.
This patch also does a few fixes related to the size that needs to
happen now:
* the block qapi driver is updated to use PATH_MAX bytes
* the qcow and qcow2 drivers have an additional safety check
* the block vvfat driver is updated to use PATH_MAX bytes
for the size of backing_file, for systems where PATH_MAX is < 1024
bytes.
* qemu-img uses PATH_MAX rather than 1024. These instances were not
changed to be dynamically allocated, however, as the extra
temporary 3K in stack usage for qemu-img does not seem worrisome.
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
There are some block drivers which are essential to QEMU and may not be
removed: These are raw, file and qcow2 (as the default non-raw format).
Make their BlockDriver objects public so they can be directly referenced
throughout the block layer without needing to call bdrv_find_format()
and having to deal with an error at runtime, while the real problem
occurred during linking (where raw, file or qcow2 were not linked into
qemu).
Cc: qemu-stable@nongnu.org
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the user neglects to specify the image format, QEMU probes the
image to guess it automatically, for convenience.
Relying on format probing is insecure for raw images (CVE-2008-2004).
If the guest writes a suitable header to the device, the next probe
will recognize a format chosen by the guest. A malicious guest can
abuse this to gain access to host files, e.g. by crafting a QCOW2
header with backing file /etc/shadow.
Commit 1e72d3b (April 2008) provided -drive parameter format to let
users disable probing. Commit f965509 (March 2009) extended QCOW2 to
optionally store the backing file format, to let users disable backing
file probing. QED has had a flag to suppress probing since the
beginning (2010), set whenever a raw backing file is assigned.
All of these additions that allow to avoid format probing have to be
specified explicitly. The default still allows the attack.
In order to fix this, commit 79368c8 (July 2010) put probed raw images
in a restricted mode, in which they wouldn't be able to overwrite the
first few bytes of the image so that they would identify as a different
image. If a write to the first sector would write one of the signatures
of another driver, qemu would instead zero out the first four bytes.
This patch was later reverted in commit 8b33d9e (September 2010) because
it didn't get the handling of unaligned qiov members right.
Today's block layer that is based on coroutines and has qiov utility
functions makes it much easier to get this functionality right, so this
patch implements it.
The other differences of this patch to the old one are that it doesn't
silently write something different than the guest requested by zeroing
out some bytes (it fails the request instead) and that it doesn't
maintain a list of signatures in the raw driver (it calls the usual
probe function instead).
Note that this change doesn't introduce new breakage for false positive
cases where the guest legitimately writes data into the first sector
that matches the signatures of an image format (e.g. for nested virt):
These cases were broken before, only the failure mode changes from
corruption after the next restart (when the wrong format is probed) to
failing the problematic write request.
Also note that like in the original patch, the restrictions only apply
if the image format has been guessed by probing. Explicitly specifying a
format allows guests to write anything they like.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 1416497234-29880-8-git-send-email-kwolf@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The only image format driver that even potentially accesses anything
after 512 bytes in its bdrv_probe() implementation is VMDK, which reads
a plain-text descriptor file. In practice, the field it's looking for
seems to come first and will be well within the first 512 bytes, too.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 1416497234-29880-7-git-send-email-kwolf@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Depending on the changed options and the image format,
bdrv_amend_options() may take a significant amount of time. In these
cases, a way to be informed about the operation's status is desirable.
Since the operation is rather complex and may fundamentally change the
image, implementing it as AIO or a coroutine does not seem feasible. On
the other hand, implementing it as a block job would be significantly
more difficult than a simple callback and would not add benefits other
than progress report to the amending operation, because it should not
actually be run as a block job at all.
A callback may not be very pretty, but it's very easy to implement and
perfectly fits its purpose here.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Benoît Canet <benoit.canet@nodalink.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 1414404776-4919-2-git-send-email-mreitz@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Move device model attachment / detachment and the BlockDevOps device
model callbacks and their wrappers from BlockDriverState to
BlockBackend.
Wrapper calls in block.c change from
bdrv_dev_FOO_cb(bs, ...)
to
if (bs->blk) {
bdrv_dev_FOO_cb(bs->blk, ...);
}
No change, because both bdrv_dev_change_media_cb() and
bdrv_dev_resize_cb() do nothing when no device model is attached, and
a device model can be attached only when bs->blk.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
I'll use it with block backends shortly, and the name is going to fit
badly there. It's a block layer thing anyway, not just a block driver
thing.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
I'll use BlockDriverAIOCB with block backends shortly, and the name is
going to fit badly there. It's a block layer thing anyway, not just a
block driver thing.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
device_name[] can become non-empty only in bdrv_new_root() and
bdrv_move_feature_fields(). The latter is used only to undo damage
done by bdrv_swap(). The former is called only by blk_new_with_bs().
Therefore, when a BlockDriverState's device_name[] is non-empty, then
it's been created with a BlockBackend, and vice versa. Furthermore,
blk_new_with_bs() keeps the two names equal.
Therefore, device_name[] is redundant. Eliminate it.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Convenience function blk_new_with_bs() creates a BlockBackend with its
BlockDriverState. Callers have to unref both. The commit after next
will relieve them of the need to unref the BlockDriverState.
Complication: due to the silly way drive_del works, we need a way to
hide a BlockBackend, just like bdrv_make_anon(). To emphasize its
"special" status, give the function a suitably off-putting name:
blk_hide_on_behalf_of_do_drive_del(). Unfortunately, hiding turns the
BlockBackend's name into the empty string. Can't avoid that without
breaking the blk->bs->device_name equals blk->name invariant.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The plan is to add new accounting metrics (latency, invalid requests, failed
requests, queue depth) and block.c is overpopulated so it will be better to work
in a separate module.
Moreover the long term plan is to have statistics in each of the BDS of the graph
for metrology purpose; this means that the device model statistics must move from
the topmost BDS to the device model.
So we need to decouple the statistic code from BlockDriverState.
This is another argument for the extraction of the code in a separate module.
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
CC: Max Reitz <mreitz@redhat.com>
CC: Eric Blake <eblake@redhat.com>
CC: Benoit Canet <benoit@irqsave.net>
CC: Fam Zheng <famz@redhat.com>
CC: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
CC: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Benoît Canet <benoit.canet@nodalink.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Extract the block accounting statistics into a structure so the block device
models can hold them in the future.
CC: Kevin Wolf <kwolf@redhat.com>
CC: Stefan Hajnoczi <stefanha@redhat.com>
CC: Max Reitz <mreitz@redhat.com>
CC: Eric Blake <eblake@redhat.com>
Signed-off-by: Benoît Canet <benoit.canet@nodalink.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If a long-running operation on a BDS wants to always remain in the same
AIO context, it somehow needs to keep track of the BDS changing its
context. This adds a function for registering callbacks on a BDS which
are called whenever the BDS is attached or detached from an AIO context.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Some block devices may not have a filename in their BDS; and for some,
there may not even be a normal filename at all. To work around this, add
a function which tries to construct a valid filename for the
BDS.filename field.
If a filename exists or a block driver is able to reconstruct a valid
filename (which is placed in BDS.exact_filename), this can directly be
used.
If no filename can be constructed, we can still construct an options
QDict which is then converted to a JSON object and prefixed with the
"json:" pseudo protocol prefix. The QDict is placed in
BDS.full_open_options.
For most block drivers, this process can be done automatically; those
that need special handling may define a .bdrv_refresh_filename() method
to fill BDS.exact_filename and BDS.full_open_options themselves.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch introduces three APIs so that following
patches can support queuing I/O requests and submitting them
as a batch for improving I/O performance.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Ming Lei <ming.lei@canonical.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
On some image chains, QEMU may not always be able to resolve the
filenames properly, when updating the backing file of an image
after a block commit.
For instance, certain relative pathnames may fail, or drives may
have been specified originally by file descriptor (e.g. /dev/fd/???),
or a relative protocol pathname may have been used.
In these instances, QEMU may lack the information to be able to make
the correct choice, but the user or management layer most likely does
have that knowledge.
With this extension to the block-commit api, the user is able to change
the backing file of the overlay image as part of the block-commit
operation.
This allows the change to be 'safe', in the sense that if the attempt
to write the overlay image metadata fails, then the block-commit
operation returns failure, without disrupting the guest.
If the commit top is the active layer, then specifying the backing
file string will be treated as an error (there is no overlay image
to modify in that case).
If a backing file string is not specified in the command, the backing
file string to use is determined in the same manner as it was
previously.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Add 'nocow' option so that users could have a chance to set NOCOW flag to
newly created files. It's useful on btrfs file system to enhance performance.
Btrfs has low performance when hosting VM images, even more when the guest
in those VM are also using btrfs as file system. One way to mitigate this bad
performance is to turn off COW attributes on VM files. Generally, there are
two ways to turn off NOCOW on btrfs: a) by mounting fs with nodatacow, then
all newly created files will be NOCOW. b) per file. Add the NOCOW file
attribute. It could only be done to empty or new files.
This patch tries the second way, according to the option, it could add NOCOW
per file.
For most block drivers, since the create file step is in raw-posix.c, so we
can do setting NOCOW flag ioctl in raw-posix.c only.
But there are some exceptions, like block/vpc.c and block/vdi.c, they are
creating file by calling qemu_open directly. For them, do the same setting
NOCOW flag ioctl work in them separately.
[Fixed up 082.out due to the new 'nocow' creation option
--Stefan]
Signed-off-by: Chunyan Liu <cyliu@suse.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
drive-mirror will bdrv_swap the new BDS named node-name with the one
pointed by replaces when the mirroring is finished.
Signed-off-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Since we parse backing.* options to add a backing file from the command
line when the driver didn't assign one, it has been possible to have a
backing file for e.g. raw images (it just was never accessed).
This is obvious nonsense and should be rejected.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Now that all backend drivers are using QemuOpts, remove all
QEMUOptionParameter related codes.
Signed-off-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
Signed-off-by: Chunyan Liu <cyliu@suse.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Change block layer to support both QemuOpts and QEMUOptionParameter.
After this patch, it will change backend drivers one by one. At the end,
QEMUOptionParameter will be removed and only QemuOpts is kept.
Signed-off-by: Dong Xu Wang <wdongxu@linux.vnet.ibm.com>
Signed-off-by: Chunyan Liu <cyliu@suse.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
block_int.h is for block layer and block drivers, other code shouldn't
include it. But similar to bdrv_set_aio_context, bdrv_get_aio_context
should also be accessible from outside of block layer.
Move it.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Up until now all BlockDriverState instances have used the QEMU main loop
for fd handlers, timers, and BHs. This is not scalable on SMP guests
and hosts so we need to move to a model with multiple event loops on
different host CPUs.
bdrv_set_aio_context() assigns the AioContext event loop to use for a
particular BlockDriverState. It first detaches the entire
BlockDriverState graph from the current AioContext and then attaches to
the new AioContext.
This function will be used by virtio-blk data-plane to assign a
BlockDriverState to its IOThread AioContext. Make
bdrv_aio_set_context() public since data-plane should not include
block_int.h.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This makes use of op_blocker and blocks all the operations except for
commit target, on each BlockDriverState->backing_hd.
The asserts for op_blocker in bdrv_swap are removed because with this
change, the target of block commit has at least the backing blocker of
its child, so the assertion is not true. Callers should do their check.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This drops BlockDriverState.in_use with op_blockers:
- Call bdrv_op_block_all in place of bdrv_set_in_use(bs, 1).
- Call bdrv_op_unblock_all in place of bdrv_set_in_use(bs, 0).
- Check bdrv_op_is_blocked() in place of bdrv_in_use(bs).
The specific types are used, e.g. in place of starting block backup,
bdrv_op_is_blocked(bs, BLOCK_OP_TYPE_BACKUP, ...).
There is one exception in block_job_create, where
bdrv_op_blocker_is_empty() is used, because we don't know the operation
type here. This doesn't matter because in a few commits away we will drop
the check and move it to callers that _do_ know the type.
- Check bdrv_op_blocker_is_empty() in place of assert(!bs->in_use).
Note: there is only bdrv_op_block_all and bdrv_op_unblock_all callers at
this moment. So although the checks are specific to op types, this
changes can still be seen as identical logic with previously with
in_use. The difference is error message are improved because of blocker
error info.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
BlockDriverState.op_blockers is an array of lists with BLOCK_OP_TYPE_MAX
elements. Each list is a list of blockers of an operation type
(BlockOpType), that marks this BDS as currently blocked for a certain
type of operation with reason errors stored in the list. The rule of
usage is:
* BDS user who wants to take an operation should check if there's any
blocker of the type with bdrv_op_is_blocked().
* BDS user who wants to block certain types of operation, should call
bdrv_op_block (or bdrv_op_block_all to block all types of operations,
which is similar to the existing bdrv_set_in_use()).
* A blocker is only referenced by op_blockers, so the lifecycle is
managed by caller, and shouldn't be lost until unblock, so typically
a caller does these:
- Allocate a blocker with error_setg or similar, call bdrv_op_block()
to block some operations.
- Hold the blocker, do his job.
- Unblock operations that it blocked, with the same reason pointer
passed to bdrv_op_unblock().
- Release the blocker with error_free().
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
this patch tries to optimize zero write requests
by automatically using bdrv_write_zeroes if it is
supported by the format.
This significantly speeds up file system initialization and
should speed zero write test used to test backend storage
performance.
I ran the following 2 tests on my internal SSD with a
50G QCOW2 container and on an attached iSCSI storage.
a) mkfs.ext4 -E lazy_itable_init=0,lazy_journal_init=0 /dev/vdX
QCOW2 [off] [on] [unmap]
-----
runtime: 14secs 1.1secs 1.1secs
filesize: 937M 18M 18M
iSCSI [off] [on] [unmap]
----
runtime: 9.3s 0.9s 0.9s
b) dd if=/dev/zero of=/dev/vdX bs=1M oflag=direct
QCOW2 [off] [on] [unmap]
-----
runtime: 246secs 18secs 18secs
filesize: 51G 192K 192K
throughput: 203M/s 2.3G/s 2.3G/s
iSCSI* [off] [on] [unmap]
----
runtime: 8mins 45secs 33secs
throughput: 106M/s 1.2G/s 1.6G/s
allocated: 100% 100% 0%
* The storage was connected via an 1Gbit interface.
It seems to internally handle writing zeroes
via WRITESAME16 very fast.
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of having unlink() calls in the generic block layer, where we
aren't even guarateed to have a file name, move them to those block
drivers that are actually used and that always have a filename. Gets us
rid of some #ifdefs as well.
The patch also converts bs->is_temporary to a new BDRV_O_TEMPORARY open
flag so that it is inherited in the protocol layer and the raw-posix and
raw-win32 drivers can unlink the file.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
If it returns an error, the migrated VM will not be started, but qemu
exits with an error message.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
This patch keep the recursive way of doing things but simplify it by giving
two responsabilities to all block filters implementors.
They will need to do two things:
-Set the is_filter field of their block driver to true.
-Implement the bdrv_recurse_is_first_non_filter method of their block driver like
it is done on the Quorum block driver. (block/quorum.c)
[Paolo Bonzini <pbonzini@redhat.com> pointed out that this patch changes
the semantics of blkverify, which now recurses down both bs->file and
s->test_file.
-- Stefan]
Reported-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
We can only have a single wait_serialising_requests() call per request
because otherwise we can run into deadlocks where requests are waiting
for each other. The same is true when wait_serialising_requests() is not
at the very beginning of a request, so that other requests can be issued
between the start of the tracking and wait_serialising_requests().
Fix this by changing wait_serialising_requests() to ignore requests that
are already (directly or indirectly) waiting for the calling request.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Copy on Read wants to serialise with all requests touching the same
cluster, so wait_serialising_requests() rounded to cluster boundaries.
Other users like alignment RMW will have different requirements, though
(requests touching the same sector), so make it dynamic.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Change the API so that specific requests can be marked serialising. Only
these requests are checked for overlaps then.
This means that during a Copy on Read operation, not all requests
overlapping other requests are serialised any more, but only those that
actually overlap with the specific COR request.
Also remove COR from function and variable names because this
functionality can be useful in other contexts.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Add a bs->request_alignment field that contains the required
offset/length alignment for I/O requests and fill it in the raw block
drivers. Use ioctls if possible, else see what alignment it takes for
O_DIRECT to succeed.
While at it, also expose the memory alignment requirements, which may be
(and in practice are) different from the disk alignment requirements.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
The alignment field is now set to the value that is promised to the
guest, rather than required by the host. The next patches will make
QEMU aware of the host-provided values, so make this clear.
The alignment is also not about memory buffers, but about the sectors on
the disk, change the documentation of the field.
At this point, the field is set by the device emulation, but completely
ignored by the block layer.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
bs->buffer_alignment is set by the device emulation and contains the
logical block size of the guest device. This isn't something that the
block layer should know, and even less something to use for determining
the right alignment of buffers to be used for the host.
The new BlockLimits field opt_mem_alignment tells the qemu block layer
the optimal alignment to be used so that no bounce buffer must be used
in the driver.
This patch may change the buffer alignment from 4k to 512 for all
callers that used qemu_blockalign() with the top-level image format
BlockDriverState. The value was never propagated to other levels in the
tree, so in particular raw-posix never required anything else than 512.
While on disks with 4k sectors direct I/O requires a 4k alignment,
memory may still be okay when aligned to 512 byte boundaries. This is
what must have happened in practice, because otherwise this would
already have failed earlier. Therefore I don't expect regressions even
with this intermediate state. Later, raw-posix can implement the hook
and expose a different memory alignment requirement.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
This function separates filling the BlockLimits from bdrv_open(), which
allows it to call it from other operations which may change the limits
(e.g. modifications to the backing file chain or bdrv_reopen)
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Benoit Canet <benoit@irqsave.net>
Add the minimum of code to prepare for the following patches.
Signed-off-by: Benoit Canet <benoit@irqsave.net>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
commit_active_start is implemented in block/mirror.c, It will create a
job with "commit" type and designated base in block-commit command. This
will be used for committing active layer of device.
Sync mode is removed from MirrorBlockJob because there's no proper type
for commit. The used information is is_none_mode.
The common part of mirror_start and commit_active_start is moved to
mirror_start_job().
Fix the comment wording for commit_start.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Since later this function will be used so improve it. The only caller of it
now is qemu-img, and it is not impacted by introduce function
bdrv_snapshot_load_tmp_by_id_or_name() that call bdrv_snapshot_load_tmp()
twice to keep old search logic. bdrv_snapshot_load_tmp_by_id_or_name() return
int to let caller know the errno, and errno will be used later.
Also fix a typo in comments of bdrv_snapshot_delete().
Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This adds "remove_break" command which is the reverse of blkdebug
command "break": it removes all breakpoints with given tag and resumes
all the requests.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Sheepdog support two kinds of redundancy, full replication and erasure coding.
# create a fully replicated vdi with x copies
-o redundancy=x (1 <= x <= SD_MAX_COPIES)
# create a erasure coded vdi with x data strips and y parity strips
-o redundancy=x:y (x must be one of {2,4,8,16} and 1 <= y < SD_EC_MAX_STRIP)
E.g, to convert a vdi into sheepdog vdi 'test' with 8:3 erasure coding scheme
$ qemu-img convert -o redundancy=8:3 linux-0.2.img sheepdog:test
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Liu Yuan <namei.unix@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Previously a BlockDriverState has only one dirty bitmap, so only one
caller (e.g. a block job) can keep track of writing. This changes the
dirty bitmap to a list and creates a BdrvDirtyBitmap for each caller, the
lifecycle is managed with these new functions:
bdrv_create_dirty_bitmap
bdrv_release_dirty_bitmap
Where BdrvDirtyBitmap is a linked list wrapper structure of HBitmap.
In place of bdrv_set_dirty_tracking, a BdrvDirtyBitmap pointer argument
is added to these functions, since each caller has its own dirty bitmap:
bdrv_get_dirty
bdrv_dirty_iter_init
bdrv_get_dirty_count
bdrv_set_dirty and bdrv_reset_dirty prototypes are unchanged but will
internally walk the list of all dirty bitmaps and set them one by one.
Signed-off-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
this patch adds BlockLimits which introduces discard and write_zeroes
limits and alignment information to the BlockDriverState.
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The block layer generally keeps the size of an image cached in
bs->total_sectors so that it doesn't have to perform expensive
operations to get the size whenever it needs it.
This doesn't work however when using a backend that can change its size
without qemu being aware of it, i.e. passthrough of removable media like
CD-ROMs or floppy disks. For this reason, the caching is disabled when a
removable device is used.
It is obvious that checking whether the _guest_ device has removable
media isn't the right thing to do when we want to know whether the size
of the host backend can change. To make things worse, non-top-level
BlockDriverStates never have any device attached, which makes qemu
assume they are removable, so drv->bdrv_getlength() is always called on
the protocol layer. In the case of raw-posix, this causes unnecessary
lseek() system calls, which turned out to be rather expensive.
This patch completely changes the logic and disables bs->total_sectors
caching only for certain block driver types, for which a size change is
expected: host_cdrom and host_floppy on POSIX, host_device on win32; also
the raw format in case it sits on top of one of these protocols, but in
the common case the nested bdrv_getlength() call on the protocol driver
will use the cache again and avoid an expensive drv->bdrv_getlength()
call.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
This field is used by blkverify to disable external snapshots creation.
It will also be used by block filters like quorum to disable external
snapshot creation.
Signed-off-by: Benoit Canet <benoit@irqsave.net>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add a function for retrieving an ImageInfoSpecific object from a block
driver.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Some drivers will have driver specifics options but no filename.
This new bool allow the block layer to treat them correctly.
The .bdrv_needs_filename is set in drivers not having .bdrv_parse_filename and
not having .bdrv_open.
The first exception to this rule will be the quorum driver.
Signed-off-by: Benoit Canet <benoit@irqsave.net>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add an Error ** parameter to BlockDriver.bdrv_open and
BlockDriver.bdrv_file_open to allow more specific error messages.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Snapshot creation actually already distinguish id and name since it take
a structured parameter *sn, but delete can't. Later an accurate delete
is needed in qmp_transaction abort and blockdev-snapshot-delete-sync,
so change its prototype. Also *errp is added to tip error, but return
value is kepted to let caller check what kind of error happens. Existing
caller for it are savevm, delvm and qemu-img, they are not impacted by
introducing a new function bdrv_snapshot_delete_by_id_or_name(), which
check the return value and do the operation again.
Before this patch:
For qcow2, it search id first then name to find the one to delete.
For rbd, it search name.
For sheepdog, it does nothing.
After this patch:
For qcow2, logic is the same by call it twice in caller.
For rbd, it always fails in delete with id, but still search for name
in second try, no change to user.
Some code for *errp is based on Pavel's patch.
Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch adds the "amend" option to qemu-img which allows changing
image options on existing image files. It also adds the generic bdrv
implementation which is basically just a wrapper for the image format
specific function.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For now, bdrv_get_block_status is just another name for bdrv_is_allocated.
The next patches will add more flags.
This also touches all block drivers with a mostly mechanical rename. The
sole exception is cow; because it calls cow_co_is_allocated from the read
code, we keep that function and make cow_co_get_block_status a wrapper.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Introduce bdrv_ref/bdrv_unref to manage the lifecycle of
BlockDriverState. They are unused for now but will used to replace
bdrv_delete() later.
Signed-off-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
include/qemu/timer.h has no need to include main-loop.h and
doing so causes an issue for the next patch. Unfortunately
various files assume including timers.h will pull in main-loop.h.
Untangle this mess.
Signed-off-by: Alex Bligh <alex@alex.org.uk>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
In 4146b46c42e0989cb5842e04d88ab6ccb1713a48 (block: Produce zeros when
protocols reading beyond end of file), we break qemu-iotests ./check
-qcow2 022. This happens because qcow2 temporarily sets ->growable = 1
for vmstate accesses (which are stored beyond the end of regular image
data).
We introduce the bs->zero_beyond_eof to allow qcow2_load_vmstate() to
disable ->zero_beyond_eof temporarily in addition to enable ->growable.
[Since the broken patch "block: Produce zeros when protocols reading
beyond end of file" has not been merged yet, I have applied this fix
*first* and will then apply the next patch to keep the tree bisectable.
-- Stefan]
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Asias He <asias@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch adds sync-modes to the drive-backup interface and
implements the FULL, NONE and TOP modes of synchronization.
FULL performs as before copying the entire contents of the drive
while preserving the point-in-time using CoW.
NONE only copies new writes to the target drive.
TOP copies changes to the topmost drive image and preserves the
point-in-time using CoW.
For sync mode TOP are creating a new target image using the same backing
file as the original disk image. Then any new data that has been laid
on top of it since creation is copied in the main backup_run() loop.
There is an extra check in the 'TOP' case so that we don't bother to copy
all the data of the backing file as it already exists in the target.
This is where the bdrv_co_is_allocated() is used to determine if the
data exists in the topmost layer or below.
Also any new data being written is intercepted via the write_notifier
hook which ends up calling backup_do_cow() to copy old data out before
it gets overwritten.
For mode 'NONE' we create the new target image and only copy in the
original data from the disk image starting from the time the call was
made. This preserves the point in time data by only copying the parts
that are *going to change* to the target image. This way we can
reconstruct the final image by checking to see if the given block exists
in the new target image first, and if it does not, you can get it from
the original image. This is basically an optimization allowing you to
do point-in-time snapshots with low overhead vs the 'FULL' version.
Since there is no old data to copy out the loop in backup_run() for the
NONE case just calls qemu_coroutine_yield() which only wakes up after
an event (usually cancel in this case). The rest is handled by the
before_write notifier which again calls backup_do_cow() to write out
the old data so it can be preserved.
Signed-off-by: Ian Main <imain@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
backup_start() creates a block job that copies a point-in-time snapshot
of a block device to a target block device.
We call backup_do_cow() for each write during backup. That function
reads the original data from the block device before it gets
overwritten. The data is then written to the target device.
Currently backup cluster size is hardcoded to 65536 bytes.
[I made a number of changes to Dietmar's original patch and folded them
in to make code review easy. Here is the full list:
* Drop BackupDumpFunc interface in favor of a target block device
* Detect zero clusters with buffer_is_zero() and use bdrv_co_write_zeroes()
* Use 0 delay instead of 1us, like other block jobs
* Unify creation/start functions into backup_start()
* Simplify cleanup, free bitmap in backup_run() instead of cb
* function
* Use HBitmap to avoid duplicating bitmap code
* Use bdrv_getlength() instead of accessing ->total_sectors
* directly
* Delete the backup.h header file, it is no longer necessary
* Move ./backup.c to block/backup.c
* Remove #ifdefed out code
* Coding style and whitespace cleanups
* Use bdrv_add_before_write_notifier() instead of blockjob-specific hooks
* Keep our own in-flight CowRequest list instead of using block.c
tracked requests. This means a little code duplication but is much
simpler than trying to share the tracked requests list and use the
backup block size.
* Add on_source_error and on_target_error error handling.
* Use trace events instead of DPRINTF()
-- stefanha]
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The bdrv_add_before_write_notifier() function installs a callback that
is invoked before a write request is processed. This will be used to
implement copy-on-write point-in-time snapshots where we need to copy
out old data before overwriting it.
Note that BdrvTrackedRequest is moved to block_int.h since it is passed
to .notify() functions.
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch is a pure code move patch, except following modification:
1 get_human_readable_size() is changed to static function.
2 dump_human_image_info() is renamed to bdrv_image_info_dump().
3 in qmp_query_block() and qmp_query_blockstats, use bdrv_next(bs)
instead of direct traverse of global array 'bdrv_states'.
4 collect_snapshots() and collect_image_info() are renamed, unused parameter
*fmt in collect_image_info() is removed.
5 code style fix.
To avoid conflict and tip better, macro in header file is BLOCK_QAPI_H
instead of QAPI_H. Now block.h and snapshot.h are at the same level in
include path, block_int.h and qapi.h will both include them.
Signed-off-by: Wenchao Xia <xiawenc@linux.vnet.ibm.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It is not necessary to adjust the slice time at runtime. We already
extend the current slice in order to carry over accounting into the next
slice. Changing the actual slice time value introduces oscillations.
The guest may experience large changes in throughput or IOPS from one
moment to the next when slice times are adjusted.
Reported-by: Benoît Canet <benoit@irqsave.net>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-By: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
I/O throttling relies on bdrv_acct_done() which is called when a request
completes. This leaves a blind spot since we only charge for completed
requests, not submitted requests.
For example, if there is 1 operation remaining in this time slice the
guest could submit 3 operations and they will all be submitted
successfully since they don't actually get accounted for until they
complete.
Originally we probably thought this is okay since the requests will be
accounted when the time slice is extended. In practice it causes
fluctuations since the guest can exceed its I/O limit and it will be
punished for this later on.
Account for I/O upon submission so that I/O limits are enforced
properly.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Tested-By: Benoit Canet <benoit@irqsave.net>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
After this patch, using -drive with an empty file name continues to open
the file if driver-specific options are used. If no driver-specific
options are specified, the semantics stay as it was: It defines a drive
without an inserted medium.
In order to achieve this, bdrv_open() must be made safe to work with a
NULL filename parameter. The assumption that is made is that only block
drivers which implement bdrv_parse_filename() support using driver
specific options and could therefore work without a filename. These
drivers must make sure to cope with NULL in their implementation of
.bdrv_open() (this is only NBD for now). For all other drivers, the
block layer code will make sure to error out before calling into their
code - they can't possibly work without a filename.
Now an NBD connection can be opened like this:
qemu-system-x86_64 -drive file.driver=nbd,file.port=1234,file.host=::1
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
If a driver needs structured data and not just a string, it can provide
a .bdrv_parse_filename callback now that parses the command line string
into separate options. Keeping this separate from .bdrv_open_filename
ensures that the preferred way of directly specifying the options always
works as well if parsing the string works.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
For now bdrv_get_aio_context() is just a stub that calls
qemu_aio_get_context() since the block layer is currently tied to the
main loop AioContext.
Add the stub now so that the block layer can begin accessing its
AioContext.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
It doesn't do anything yet except storing the options QDict in the
BlockDriverState.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Introduce a new option "adapter_type" when converting to vmdk images.
It can be one of the following: ide (default), buslogic, lsilogic
or legacyESX (according to the vmdk spec from vmware).
In case of a non-ide adapter, heads is set to 255 instead of the 16.
The latter is used for "ide".
Also see LP#545089
Signed-off-by: Othmar Pasteka <pasteka@kabsi.at>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This makes sense when the next commit starts using the extra buffer space
to perform many I/O operations asynchronously.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The desired granularity may be very different depending on the kind of
operation (e.g. continuous replication vs. collapse-to-raw) and whether
the VM is expected to perform lots of I/O while mirroring is in progress.
Allow the user to customize it, while providing a sane default so that
in general there will be no extra allocated space in the target compared
to the source.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This actually uses the dirty bitmap in the block layer, and converts
mirroring to use an HBitmapIter.
Reviewed-by: Laszlo Ersek <lersek@redhat.com> (except block/mirror.c parts)
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
