The default cache-clean-interval is set to 10 minutes, in order to lower
the overhead of the qcow2 caches (before the default was 0, i.e.
disabled).
* For non-Linux platforms the default is kept at 0, because
cache-clean-interval is not supported there yet.
Signed-off-by: Leonid Bloch <lbloch@janustech.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The upper limit on the L2 cache size is increased from 1 MB to 32 MB
on Linux platforms, and to 8 MB on other platforms (this difference is
caused by the ability to set intervals for cache cleaning on Linux
platforms only).
This is done in order to allow default full coverage with the L2 cache
for images of up to 256 GB in size (was 8 GB). Note, that only the
needed amount to cover the full image is allocated. The value which is
changed here is just the upper limit on the L2 cache size, beyond which
it will not grow, even if the size of the image will require it to.
Signed-off-by: Leonid Bloch <lbloch@janustech.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Sufficient L2 cache can noticeably improve the performance when using
large images with frequent I/O.
Previously, unless 'cache-size' was specified and was large enough, the
L2 cache was set to a certain size without taking the virtual image size
into account.
Now, the L2 cache assignment is aware of the virtual size of the image,
and will cover the entire image, unless the cache size needed for that is
larger than a certain maximum. This maximum is set to 1 MB by default
(enough to cover an 8 GB image with the default cluster size) but can
be increased or decreased using the 'l2-cache-size' option. This option
was previously documented as the *maximum* L2 cache size, and this patch
makes it behave as such, instead of as a constant size. Also, the
existing option 'cache-size' can limit the sum of both L2 and refcount
caches, as previously.
Signed-off-by: Leonid Bloch <lbloch@janustech.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Leonid Bloch <lbloch@janustech.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Do data compression in separate threads. This significantly improve
performance for qemu-img convert with -W (allow async writes) and -c
(compressed) options.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If we managed to allocate the clusters, but then failed to write the
data, there's a good chance that we'll still be able to free the
clusters again in order to avoid cluster leaks (the refcounts are
cached, so even if we can't write them out right now, we may be able to
do so when the VM is resumed after a werror=stop/enospc pause).
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Tested-by: Eric Blake <eblake@redhat.com>
The L2 and refcount caches have default sizes that can be overridden
using the l2-cache-size and refcount-cache-size (an additional
parameter named cache-size sets the combined size of both caches).
Unless forced by one of the aforementioned parameters, QEMU will set
the unspecified sizes so that the L2 cache is 4 times larger than the
refcount cache.
This is based on the premise that the refcount metadata needs to be
only a fourth of the L2 metadata to cover the same amount of disk
space. This is incorrect for two reasons:
a) The amount of disk covered by an L2 table depends solely on the
cluster size, but in the case of a refcount block it depends on
the cluster size *and* the width of each refcount entry.
The 4/1 ratio is only valid with 16-bit entries (the default).
b) When we talk about disk space and L2 tables we are talking about
guest space (L2 tables map guest clusters to host clusters),
whereas refcount blocks are used for host clusters (including
L1/L2 tables and the refcount blocks themselves). On a fully
populated (and uncompressed) qcow2 file, image size > virtual size
so there are more refcount entries than L2 entries.
Problem (a) could be fixed by adjusting the algorithm to take into
account the refcount entry width. Problem (b) could be fixed by
increasing a bit the refcount cache size to account for the clusters
used for qcow2 metadata.
However this patch takes a completely different approach and instead
of keeping a ratio between both cache sizes it assigns as much as
possible to the L2 cache and the remainder to the refcount cache.
The reason is that L2 tables are used for every single I/O request
from the guest and the effect of increasing the cache is significant
and clearly measurable. Refcount blocks are however only used for
cluster allocation and internal snapshots and in practice are accessed
sequentially in most cases, so the effect of increasing the cache is
negligible (even when doing random writes from the guest).
So, make the refcount cache as small as possible unless the user
explicitly asks for a larger one.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 9695182c2eb11b77cb319689a1ebaa4e7c9d6591.1523968389.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Checking reopen by existence of some bitmaps is wrong, as it may be
some other bitmaps, or on the other hand, user may remove bitmaps. This
criteria is bad. To simplify things and make behavior more predictable
let's just add a flag to remember, that we've already tried to load
bitmaps on open and do not want do it again.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20180411122606.367301-2-vsementsov@virtuozzo.com
[mreitz: Changed comment wording according to Eric Blake's suggestion]
Signed-off-by: Max Reitz <mreitz@redhat.com>
Add version of qcow2_reopen_bitmaps_rw, which do the same work but
also return a hint about was header updated or not. This will be
used in the following fix for bitmaps reloading after migration.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180320170521.32152-2-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function checks that the offset and size of a table are valid.
While the offset checks are fine, the size check is too generic, since
it only verifies that the total size in bytes fits in a 64-bit
integer. In practice all tables used in qcow2 have much smaller size
limits, so the size needs to be checked again for each table using its
actual limit.
This patch generalizes this function by allowing the caller to specify
the maximum size for that table. In addition to that it allows passing
an Error variable.
The function is also renamed and made public since we're going to use
it in other parts of the code.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
They will be used to avoid recursively taking s->lock during
bdrv_open or bdrv_check.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <1516279431-30424-7-git-send-email-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The align_offset() function is equivalent to the ROUND_UP() macro so
there's no need to use the former. The ROUND_UP() name is also a bit
more explicit.
This patch uses ROUND_UP() instead of the slower QEMU_ALIGN_UP()
because align_offset() already requires that the second parameter is a
power of two.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-id: 20180215131008.5153-1-berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Now that the code is ready to handle L2 slices we can finally add an
option to allow configuring their size.
An L2 slice is the portion of an L2 table that is read by the qcow2
cache. Until now the cache was always reading full L2 tables, and
since the L2 table size is equal to the cluster size this was not very
efficient with large clusters. Here's a more detailed explanation of
why it makes sense to have smaller cache entries in order to load L2
data:
https://lists.gnu.org/archive/html/qemu-block/2017-09/msg00635.html
This patch introduces a new command-line option to the qcow2 driver
named l2-cache-entry-size (cf. l2-cache-size). The cache entry size
has the same restrictions as the cluster size: it must be a power of
two and it has the same range of allowed values, with the additional
requirement that it must not be larger than the cluster size.
The L2 cache entry size (L2 slice size) remains equal to the cluster
size for now by default, so this feature must be explicitly enabled.
Although my tests show that 4KB slices consistently improve
performance and give the best results, let's wait and make more tests
with different cluster sizes before deciding on an optimal default.
Now that the cache entry size is not necessarily equal to the cluster
size we need to reflect that in the MIN_L2_CACHE_SIZE documentation.
That minimum value is a requirement of the COW algorithm: we need to
read two L2 slices (and not two L2 tables) in order to do COW, see
l2_allocate() for the actual code.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: c73e5611ff4a9ec5d20de68a6c289553a13d2354.1517840877.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Similar to offset_to_l2_index(), this function takes a guest offset
and returns the index in the L2 slice that contains its L2 entry.
An L2 slice has currently the same size as an L2 table (one cluster),
so both functions return the same value for now.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: a1c45c5c5a76146dd1712d8d1e7b409ad539c718.1517840877.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
The BDRVQcow2State structure contains an l2_size field, which stores
the number of 64-bit entries in an L2 table.
For efficiency reasons we want to be able to load slices instead of
full L2 tables, so we need to know how many entries an L2 slice can
hold.
An L2 slice is the portion of an L2 table that is loaded by the qcow2
cache. At the moment that cache can only load complete tables,
therefore an L2 slice has the same size as an L2 table (one cluster)
and l2_size == l2_slice_size.
Later we'll allow smaller slices, but until then we have to use this
new l2_slice_size field to make the rest of the code ready for that.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: adb048595f9fb5dfb110c802a8b3c3be3b937f37.1517840877.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Similar to offset_to_l2_index(), this function returns the index in
the L1 table for a given guest offset. This is only used in a couple
of places and it's not a particularly complex calculation, but it
makes the code a bit more readable.
Although in the qcow2_get_cluster_offset() case the old code was
taking advantage of the l1_bits variable, we're going to get rid of
the other uses of l1_bits in a later patch anyway, so it doesn't make
sense to keep it just for this.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: a5f626fed526b7459a0425fad06d823d18df8522.1517840877.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function was only using the BlockDriverState parameter to pass it
to qcow2_cache_get_table_addr(). This is no longer necessary so this
parameter can be removed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: eb0ed90affcf302e5a954bafb5931b5215483d3a.1517840877.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function was only using the BlockDriverState parameter to pass it
to qcow2_cache_get_table_idx() and qcow2_cache_table_release(). This
is no longer necessary so this parameter can be removed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 9724f7e38e763ad3be32627c6b7fe8df9edb1476.1517840877.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function was only using the BlockDriverState parameter to pass it
to qcow2_cache_table_release(). This is no longer necessary so this
parameter can be removed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: b74f17591af52f201de0ea3a3b2dd0a81932334d.1517840876.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function was never using the BlockDriverState parameter so it can
be safely removed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 49c74fe8b3aead9056e61a85b145ce787d06262b.1517840876.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function was only using the BlockDriverState parameter to pass it
to qcow2_cache_get_table_idx(). This is no longer necessary so this
parameter can be removed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 6f98155489054a457563da77cdad1a66ebb3e896.1517840876.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function was only using the BlockDriverState parameter to pass it
to qcow2_cache_get_table_idx(). This is no longer necessary so this
parameter can be removed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 5c40516a91782b083c1428b7b6a41bb9e2679bfb.1517840876.git.berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
To maintain load/store disabled bitmap there is new approach:
- deprecate @autoload flag of block-dirty-bitmap-add, make it ignored
- store enabled bitmaps as "auto" to qcow2
- store disabled bitmaps without "auto" flag to qcow2
- on qcow2 open load "auto" bitmaps as enabled and others
as disabled (except in_use bitmaps)
Also, adjust iotests 165 and 176 appropriately.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20180202160752.143796-1-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Since bdrv_co_preadv does all neccessary checks including
reading after the end of the backing file, avoid duplication
of verification before bdrv_co_preadv call.
Signed-off-by: Edgar Kaziakhmedov <edgar.kaziakhmedov@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reported-by: R. Nageswara Sastry <nasastry@in.ibm.com>
Buglink: https://bugs.launchpad.net/qemu/+bug/1728661
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20171110203111.7666-5-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Now after shrinking the image, at the end of the image file, there might be a
tail that probably will never be used. So we can find the last used cluster and
cut the tail.
Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20170929121613.25997-3-pbutsykin@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This patch add shrinking of the image file for qcow2. As a result, this allows
us to reduce the virtual image size and free up space on the disk without
copying the image. Image can be fragmented and shrink is done by punching holes
in the image file.
Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20170918124230.8152-4-pbutsykin@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Whenever l2/refcount table clusters are discarded from the file we can
automatically drop unnecessary content of the cache tables. This reduces
the chance of eviction useful cache data and eliminates inconsistent data
in the cache with the data in the file.
Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20170918124230.8152-3-pbutsykin@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Implement the preallocation modes falloc and full for growing qcow2
images.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20170613202107.10125-15-mreitz@redhat.com
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function creates a collection of self-describing refcount
structures (including a new refcount table) at the end of a qcow2 image
file. Optionally, these structures can also describe a number of
additional clusters beyond themselves; this will be important for
preallocated truncation, which will place the data clusters and L2
tables there.
For now, we can use this function to replace the part of
alloc_refcount_block() that grows the refcount table (from which it is
actually derived).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20170613202107.10125-13-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Store bitmaps and mark them read-only on reopening image as read-only.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20170628120530.31251-21-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Store persistent dirty bitmaps in qcow2 image.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20170628120530.31251-20-vsementsov@virtuozzo.com
[mreitz: Always assign ret in store_bitmap() in case of an error]
Signed-off-by: Max Reitz <mreitz@redhat.com>
Auto loading bitmaps are bitmaps in Qcow2, with the AUTO flag set. They
are loaded when the image is opened and become BdrvDirtyBitmaps for the
corresponding drive.
Extra data in bitmaps is not supported for now.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20170628120530.31251-12-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Add bitmap extension as specified in docs/specs/qcow2.txt.
For now, just mirror extension header into Qcow2 state and check
constraints. Also, calculate refcounts for qcow2 bitmaps, to not break
qemu-img check.
For now, disable image resize if it has bitmaps. It will be fixed later.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20170628120530.31251-9-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This is needed for the following patch, which will introduce refcounts
checking for qcow2 bitmaps.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20170628120530.31251-8-vsementsov@virtuozzo.com
[mreitz: s/inc_refcounts/qcow2_inc_refcounts_imrt/ in one more (new)
place]
Signed-off-by: Max Reitz <mreitz@redhat.com>
This adds support for using LUKS as an encryption format
with the qcow2 file, using the new encrypt.format parameter
to request "luks" format. e.g.
# qemu-img create --object secret,data=123456,id=sec0 \
-f qcow2 -o encrypt.format=luks,encrypt.key-secret=sec0 \
test.qcow2 10G
The legacy "encryption=on" parameter still results in
creation of the old qcow2 AES format (and is equivalent
to the new 'encryption-format=aes'). e.g. the following are
equivalent:
# qemu-img create --object secret,data=123456,id=sec0 \
-f qcow2 -o encryption=on,encrypt.key-secret=sec0 \
test.qcow2 10G
# qemu-img create --object secret,data=123456,id=sec0 \
-f qcow2 -o encryption-format=aes,encrypt.key-secret=sec0 \
test.qcow2 10G
With the LUKS format it is necessary to store the LUKS
partition header and key material in the QCow2 file. This
data can be many MB in size, so cannot go into the QCow2
header region directly. Thus the spec defines a FDE
(Full Disk Encryption) header extension that specifies
the offset of a set of clusters to hold the FDE headers,
as well as the length of that region. The LUKS header is
thus stored in these extra allocated clusters before the
main image payload.
Aside from all the cryptographic differences implied by
use of the LUKS format, there is one further key difference
between the use of legacy AES and LUKS encryption in qcow2.
For LUKS, the initialiazation vectors are generated using
the host physical sector as the input, rather than the
guest virtual sector. This guarantees unique initialization
vectors for all sectors when qcow2 internal snapshots are
used, thus giving stronger protection against watermarking
attacks.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-id: 20170623162419.26068-14-berrange@redhat.com
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
This converts the qcow2 driver to make use of the QCryptoBlock
APIs for encrypting image content, using the legacy QCow2 AES
scheme.
With this change it is now required to use the QCryptoSecret
object for providing passwords, instead of the current block
password APIs / interactive prompting.
$QEMU \
-object secret,id=sec0,file=/home/berrange/encrypted.pw \
-drive file=/home/berrange/encrypted.qcow2,encrypt.key-secret=sec0
The test 087 could be simplified since there is no longer a
difference in behaviour when using blockdev_add with encrypted
images for the running vs stopped CPU state.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-id: 20170623162419.26068-12-berrange@redhat.com
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Instead of requiring separate input/output buffers for
encrypting data, change qcow2_encrypt_sectors() to assume
use of a single buffer, encrypting in place. The current
callers all used the same buffer for input/output already.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Message-id: 20170623162419.26068-11-berrange@redhat.com
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
If the guest tries to write data that results on the allocation of a
new cluster, instead of writing the guest data first and then the data
from the COW regions, write everything together using one single I/O
operation.
This can improve the write performance by 25% or more, depending on
several factors such as the media type, the cluster size and the I/O
request size.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Qcow2COWRegion has two attributes:
- The offset of the COW region from the start of the first cluster
touched by the I/O request. Since it's always going to be positive
and the maximum request size is at most INT_MAX, we can use a
regular unsigned int to store this offset.
- The size of the COW region in bytes. This is guaranteed to be >= 0,
so we should use an unsigned type instead.
In x86_64 this reduces the size of Qcow2COWRegion from 16 to 8 bytes.
It will also help keep some assertions simpler now that we know that
there are no negative numbers.
The prototype of do_perform_cow() is also updated to reflect these
changes.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Passing a byte offset, but sector count, when we ultimately
want to operate on cluster granularity, is madness. Clean up
the external interfaces to take both offset and count as bytes,
while still keeping the assertion added previously that the
caller must align the values to a cluster. Then rename things
to make sure backports don't get confused by changed units:
instead of qcow2_discard_clusters() and qcow2_zero_clusters(),
we now have qcow2_cluster_discard() and qcow2_cluster_zeroize().
The internal functions still operate on clusters at a time, and
return an int for number of cleared clusters; but on an image
with 2M clusters, a single L2 table holds 256k entries that each
represent a 2M cluster, totalling well over INT_MAX bytes if we
ever had a request for that many bytes at once. All our callers
currently limit themselves to 32-bit bytes (and therefore fewer
clusters), but by making this function 64-bit clean, we have one
less place to clean up if we later improve the block layer to
support 64-bit bytes through all operations (with the block layer
auto-fragmenting on behalf of more-limited drivers), rather than
the current state where some interfaces are artificially limited
to INT_MAX at a time.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20170507000552.20847-13-eblake@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Treat plain zero clusters differently from allocated ones, so that
we can simplify the logic of checking whether an offset is present.
Do this by splitting QCOW2_CLUSTER_ZERO into two new enums,
QCOW2_CLUSTER_ZERO_PLAIN and QCOW2_CLUSTER_ZERO_ALLOC.
I tried to arrange the enum so that we could use
'ret <= QCOW2_CLUSTER_ZERO_PLAIN' for all unallocated types, and
'ret >= QCOW2_CLUSTER_ZERO_ALLOC' for allocated types, although
I didn't actually end up taking advantage of the layout.
In many cases, this leads to simpler code, by properly combining
cases (sometimes, both zero types pair together, other times,
plain zero is more like unallocated while allocated zero is more
like normal).
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 20170507000552.20847-7-eblake@redhat.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Although it doesn't add all that much type safety (this is C, after
all), it does add a bit of legibility to use the name QCow2ClusterType
instead of a plain int.
In particular, qcow2_get_cluster_offset() has an overloaded return
type; a QCow2ClusterType on success, and -errno on failure; keeping
the cluster type in a separate variable makes it slightly easier for
the next patch to make further computations based on the type.
Suggested-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-id: 20170507000552.20847-6-eblake@redhat.com
[mreitz: Use the new type in two more places (one of them pulled from
the next patch)]
Signed-off-by: Max Reitz <mreitz@redhat.com>
Instead of just freeing preallocated zero clusters and completely
allocating them from scratch, reuse them.
We cannot do this in handle_copied(), however, since this is a COW
operation. Therefore, we have to add the new logic to handle_alloc() and
simply return the existing offset if it exists. The only catch is that
we have to convince qcow2_alloc_cluster_link_l2() not to free the old
clusters (because we have reused them).
Reported-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The metadata overlap checks introduced in a40f1c2add help detect
corruption in the qcow2 image by verifying that data writes don't
overlap with existing metadata sections.
The 'refcount-block' check in particular iterates over the refcount
table in order to get the addresses of all refcount blocks and check
that none of them overlap with the region where we want to write.
The problem with the refcount table is that since it always occupies
complete clusters its size is usually very big. With the default
values of cluster_size=64KB and refcount_bits=16 this table holds 8192
entries, each one of them enough to map 2GB worth of host clusters.
So unless we're using images with several TB of allocated data this
table is going to be mostly empty, and iterating over it is a waste of
CPU. If the storage backend is fast enough this can have an effect on
I/O performance.
This patch keeps the index of the last used (i.e. non-zero) entry in
the refcount table and updates it every time the table changes. The
refcount-block overlap check then uses that index instead of reading
the whole table.
In my tests with a 4GB qcow2 file stored in RAM this doubles the
amount of write IOPS.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 20170201123828.4815-1-berto@igalia.com
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
It was from the time when none of the global functions had a qcow2_
prefix.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
