Consider the case when the whole buffer is zero and end is unaligned.
If i <= tail, we return 1 and do one unaligned WRITE, RMW happens.
If i > tail, we do on aligned WRITE_ZERO (or skip if target is zeroed)
and again one unaligned WRITE, RMW happens.
Let's do better: don't fragment the whole-zero buffer and report it as
ZERO: in case of zeroed target we just do nothing and avoid RMW. If
target is not zeroes, one unaligned WRITE_ZERO should not be much worse
than one unaligned WRITE.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20211217164654.1184218-3-vsementsov@virtuozzo.com>
Tested-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This demonstrates what happens when the block status changes in
sub-min_sparse granularity, but all of the parts are zeroed out. The
alignment logic in is_allocated_sectors() prevents that the target image
remains fully sparse as expected, but turns it into a data cluster of
explicit zeros.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20211217164654.1184218-2-vsementsov@virtuozzo.com>
Tested-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The recently-added NBD context qemu:allocation-depth is able to
distinguish between locally-present data (even when that data is
sparse) [shown as depth 1 over NBD], and data that could not be found
anywhere in the backing chain [shown as depth 0]; and the libnbd
project was recently patched to give the human-readable name "absent"
to an allocation-depth of 0. But qemu-img map --output=json predates
that addition, and has the unfortunate behavior that all portions of
the backing chain that resolve without finding a hit in any backing
layer report the same depth as the final backing layer. This makes it
harder to reconstruct a qcow2 backing chain using just 'qemu-img map'
output, especially when using "backing":null to artificially limit a
backing chain, because it is impossible to distinguish between a
QCOW2_CLUSTER_UNALLOCATED (which defers to a [missing] backing file)
and a QCOW2_CLUSTER_ZERO_PLAIN cluster (which would override any
backing file), since both types of clusters otherwise show as
"data":false,"zero":true" (but note that we can distinguish a
QCOW2_CLUSTER_ZERO_ALLOCATED, which would also have an "offset":
listing).
The task of reconstructing a qcow2 chain was made harder in commit
0da9856851 (nbd: server: Report holes for raw images), because prior
to that point, it was possible to abuse NBD's block status command to
see which portions of a qcow2 file resulted in BDRV_BLOCK_ALLOCATED
(showing up as NBD_STATE_ZERO in isolation) vs. missing from the chain
(showing up as NBD_STATE_ZERO|NBD_STATE_HOLE); but now qemu reports
more accurate sparseness information over NBD.
An obvious solution is to make 'qemu-img map --output=json' add an
additional "present":false designation to any cluster lacking an
allocation anywhere in the chain, without any change to the "depth"
parameter to avoid breaking existing clients. The iotests have
several examples where this distinction demonstrates the additional
accuracy.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20210701190655.2131223-3-eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
[eblake: fix more iotest fallout]
Signed-off-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200721135520.72355-3-kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
There are many existing qcow2 images that specify a backing file but
no format. This has been the source of CVEs in the past, but has
become more prominent of a problem now that libvirt has switched to
-blockdev. With older -drive, at least the probing was always done by
qemu (so the only risk of a changed format between successive boots of
a guest was if qemu was upgraded and probed differently). But with
newer -blockdev, libvirt must specify a format; if libvirt guesses raw
where the image was formatted, this results in data corruption visible
to the guest; conversely, if libvirt guesses qcow2 where qemu was
using raw, this can result in potential security holes, so modern
libvirt instead refuses to use images without explicit backing format.
The change in libvirt to reject images without explicit backing format
has pointed out that a number of tools have been far too reliant on
probing in the past. It's time to set a better example in our own
iotests of properly setting this parameter.
iotest calls to create, rebase, and convert are all impacted to some
degree. It's a bit annoying that we are inconsistent on command line
- while all of those accept -o backing_file=...,backing_fmt=..., the
shortcuts are different: create and rebase have -b and -F, while
convert has -B but no -F. (amend has no shortcuts, but the previous
patch just deprecated the use of amend to change backing chains).
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200706203954.341758-9-eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It's been two releases since we started warning; time to make the
combination an error as promised. There was no iotest coverage, so
add some.
While touching the documentation, tweak another section heading for
consistent style.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200706203954.341758-3-eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This must not crash.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200121155915.98232-3-mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
We currently don't enforce that the sparse segments we detect during convert are
aligned. This leads to unnecessary and costly read-modify-write cycles either
internally in Qemu or in the background on the storage device as nearly all
modern filesystems or hardware have a 4k alignment internally.
This patch modifies is_allocated_sectors so that its *pnum result will always
end at an alignment boundary. This way all requests will end at an alignment
boundary. The start of all requests will also be aligned as long as the results
of get_block_status do not lead to an unaligned offset.
The number of RMW cycles when converting an example image [1] to a raw device that
has 4k sector size is about 4600 4k read requests to perform a total of about 15000
write requests. With this path the additional 4600 read requests are eliminated while
the number of total write requests stays constant.
[1] https://cloud-images.ubuntu.com/releases/16.04/release/ubuntu-16.04-server-cloudimg-amd64-disk1.vmdk
Signed-off-by: Peter Lieven <pl@kamp.de>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This adds a test case to 122 for what happens when you convert to a
target with a backing file that is shorter than the target, and the
image format does not support efficient zero writes (as is the case with
qcow2 v2).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180501165750.19242-3-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Commit abd3622cc0 added a case to 122
regarding how the qcow2 driver handles an incorrect compressed data
length value. This does not really fit into 122, as that file is
supposed to contain qemu-img convert test cases, which this case is not.
So this patch splits it off into its own file; maybe we will even get
more qcow2-only compression tests in the future.
Also, that test case does not work with refcount_bits=1, so mark that
option as unsupported.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180406164108.26118-1-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
L2 entries for compressed clusters have a field that indicates the
number of sectors used to store the data in the image.
That's however not the size of the compressed data itself, just the
number of sectors where that data is located. The actual data size is
usually not a multiple of the sector size, and therefore cannot be
represented with this field.
The way it works is that QEMU reads all the specified sectors and
starts decompressing the data until there's enough to recover the
original uncompressed cluster. If there are any bytes left that
haven't been decompressed they are simply ignored.
One consequence of this is that even if the size field is larger than
it needs to be QEMU can handle it just fine: it will read more data
from disk but it will ignore the extra bytes.
This test creates an image with two compressed clusters that use 5
sectors (2.5 KB) each, increases the size field to the maximum (8192
sectors, or 4 MB) and verifies that the data can be read without
problems.
This test is important because while the decompressed data takes
exactly one cluster, the maximum value allowed in the compressed size
field is twice the cluster size. So although QEMU won't produce images
with such large values we need to make sure that it can handle them.
Another effect of increasing the size field is that it can make
it include data from the following host cluster(s). In this case
'qemu-img check' will detect that the refcounts are not correct, and
we'll need to rebuild them.
Additionally, this patch also tests that decreasing the size corrupts
the image since the original data can no longer be recovered. In this
case QEMU returns an error when trying to read the compressed data,
but 'qemu-img check' doesn't see anything wrong if the refcounts are
consistent.
One possible task for the future is to make 'qemu-img check' verify
the sizes of the compressed clusters, by trying to decompress the data
and checking that the size stored in the L2 entry is correct.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180329120745.11154-1-berto@igalia.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Although _filter_qemu_img_map documents that it scrubs offsets, it
was only doing so for human mode. Of the existing tests using the
filter (97, 122, 150, 154, 176), two of them are affected, but it
does not hurt the validity of the tests to not require particular
mappings (another test, 66, uses offsets but intentionally does not
pass through _filter_qemu_img_map, because it checks that offsets
are unchanged before and after an operation).
Another justification for this patch is that it will allow a future
patch to utilize 'qemu-img map --output=json' to check the status of
preallocated zero clusters without regards to the mapping (since
the qcow2 mapping can be very sensitive to the chosen cluster size,
when preallocation is not in use).
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20170507000552.20847-9-eblake@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
It does not make much sense to use a backing image for the target when
you concatenate multiple images (because then there is no correspondence
between the source images' backing files and the target's); but it was
still possible to give one by using -o backing_file=X instead of -B X.
Fix this by moving the check.
(Also, change the error message because -B is not the only way to
specify the backing file, evidently.)
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When passing -S 0 to qemu-img convert, the target image is supposed to
be fully allocated. Right now, this is not the case if the source image
contains areas which bdrv_get_block_status() reports as being zero.
This patch changes a zeroed area's status from BLK_ZERO to BLK_DATA
before invoking convert_write() if -S 0 has been specified. In addition,
the check whether convert_read() actually needs to do anything
(basically only if the current area is a BLK_DATA area) is pulled out of
that function to the caller.
If -S 0 has been specified, zeroed areas need to be written as data to
the output, thus they then have to be accounted when calculating the
progress made.
This patch changes the reference output for iotest 122; contrary to what
it assumed, -S 0 really should allocate everything in the output, not
just areas that are filled with zeros (as opposed to being zeroed).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This will let us print options in a format that the user would actually
write it on the command line (foo=bar,baz=asd,etc=def), without
prepending a spurious comma at the beginning of the list, or quoting
values unnecessarily. This patch provides the following changes:
* write and id=, if the option has an id
* do not print separator before the first element
* do not quote string arguments
* properly escape commas (,) for QEMU
Signed-off-by: Kővágó, Zoltán <DirtY.iCE.hu@gmail.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
This adds a regression test for some problems that the qemu-img convert
rewrite just fixed.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Vladimir Sementsov-Ogievskiy
Kevin Wolf
Eric Blake
Max Reitz
Peter Lieven
Alberto Garcia
Kővágó, Zoltán