For managed save we can choose between various compression
methods. I randomly tested the 'xz' program on a 8 GB guest
and was surprised to have to wait > 50 minutes for it to
finish compressing, with 'xz' burning 100% cpu for the
entire time. Despite the impressive compression, this is
completely useless in the real world as it is far too long
to wait to save the VM.
The 'xz' binary defaults to '-6' optimization level which
aims for high compression, with moderate memory usage,
at the expense of speed.
This change switches it to use the '-3' optimization level
which is documented as being the one that optimizes speed
at expense of compression. Even with this, it will still
outperform all the other options in terms of compression
level. It is a little less than x4 faster than '-6' which
means it starts to be a viable choice to use 'xz' for
people who really want best compression.
The test results on a 1 GB, fairly freshly booted VM are
as follows
format | save | restore size
=======+=======+=============
raw | 05s | 1s | 428 MB
lzop | 05s | 3s | 160 MB
gzip | 29s | 5s | 118 MB
bz2 | 54s | 22s | 114 MB
xz | 4m37s | 13s | 86 MB
xz -3 | 1m20s | 12s | 95 MB
Based on this we can say
* For moderate compression with no noticable loss in speed
=> use lzop
* For high compression with moderate loss in speed
=> use gzip
* For best compression with significant loss in speed
=> use xz
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Commit v5.7.0-248-g03449e2504 removed "cd tests" without updating the
patch to test-suite.log.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
This is a very simple and straightforward implementation of the opposite
what buildPool does for the disk backend.
The background for this change comes from an existing test case in TCK
which does use the delete method for a pool of type disk, but it
truly could not have ever worked since the implementation simply
wasn't there for the pool of type disk.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The news schema requires two digits for both month and day in the date
attribute. s/2/02/ in the day value of date to fix the following
'make check' failure
2165) Checking ../docs/news.xml against ../news.rng ...
libvirt: XML Util error : XML document failed to validate against schema:
Unable to validate doc against /home/jfehlig/virt/upstream/libvirt/build/../docs/schemas/../news.rng
Element release failed to validate attributes
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Commit 4b58fdf280 which enabled block copy also for network
destinations needed to limit when the 'mirror' storage source is
initialized in cases when we e.g. don't have an appropriate backend.
Limiting it just to virStorageFileSupportsCreate is too restrictive as
for example we can't precreate block devices and thus wouldn't
initialize the 'mirror' but since it's a local source we'd try to
examine it. This would fail since it wouldn't be initialized.
Fix it by introducing a more granular check whether certain operations
are supported and fix the check interlocks.
https://bugzilla.redhat.com/show_bug.cgi?id=1778058
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
We tolerate image format detection during block copy in very specific
circumstances, but the code didn't error out on failure of the format
detection.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The API XML files are generated files, so live in the build dir not the
source dir.
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Gluster 6.0 is not built on i686 for RHEL-8, which prevents libvirt from
building. Let's just disable gluster there as all we need are client
libraries anyway.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Fabiano Fidêncio <fidencio@redhat.com>
In v5.9.0-273-g8ecab214de I've tried to fix a lock ordering
problem, but introduced a crasher. Problem is that because the
client lock is unlocked (in order to honour lock ordering) the
stream we are currently checking in daemonStreamFilter() might be
freed and thus stream->priv might not even exist when the control
get to virMutexLock() call.
To resolve this, grab an extra reference to the stream and handle
its cleanup should the refcounter reach zero after the deref.
If that's the case and we are the only ones holding a reference
to the stream, we MUST return a positive value to make
virNetServerClientDispatchRead() break its loop where it iterates
over filters. The problem is, if we did not do so, then
"filter = filter->next" line will read from a memory that was
just freed (freeing a stream also unregisters its filter).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In commit 2ccb5335dc I've refactored how we fill the typed parameters
for domain statistics. The commit introduced a regression in the
formating of stats for IOthreads by using the array index to label the
entries as it's common for all other types of statistics rather than
the iothread IDs used for iothreads.
Since only the design of iothread deviates from the common approach used
in all other statistic types this was not caught.
https://bugzilla.redhat.com/show_bug.cgi?id=1778014
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The original implementation used QEMU_ADD_COUNT_PARAM which added the
'count' suffix, but 'cnt' was documented. Fix the documentation to
conform with the original implementation.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Before we rewrote nss plugin so that it doesn't use libvirt's
internal functions it used virLeaseReadCustomLeaseFile() to parse
.status files. After the rewrite it's using read() + yajl_parse()
+ yajl_complete_parse(). There's one catch though,
virLeaseReadCustomLeaseFile() skipped over empty files.
An empty .status file is created when a network is started. This
is because we configure dnsmasq to use our leasehelper. So the
first thing it does it calls it as follows:
DNSMASQ_INTERFACE=virbr0 /usr/libexec/libvirt_leaseshelper init
which causes the leasehelper to create empty virbr0.status file.
If there is only one libvirt network then that is no problem -
there are no other .status files to parse anyway. But if there
are two or more networks then the first empty .status file causes
whole parsing process and subsequently the whole name lookup
process to fail.
Fixes: v5.7.0-rc1~343
Reported-by: Pavel Hrdina <phrdina@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
After generating the API HTML files we run xmllint in docs/html/*.html
to validate the correctness. Since
commit 0aa8536f14
Author: Daniel P. Berrangé <berrange@redhat.com>
Date: Wed Nov 20 14:49:26 2019 +0000
docs: generate API reference pages for admin, qemu & lxc libraries
we have many rules generating files into docs/html/. The xmllint
calls for each rule are picking up files which are part-generated by
other parallel build rules resulting in transient errors like:
GEN html/index.html
GEN html/index-admin.html
GEN html/index-qemu.html
GEN html/index-lxc.html
GEN hvsupport.html.in
html/index-lxc.html:1: parser error : Document is empty
^
make[4]: *** [Makefile:2407: html/index-qemu.html] Error 1
The easiest solution is to move the xmllint rules to the 'make check'
phase of the build.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
On Fedora 31 with GCC 9.2.1, compiling qemuxml2argvtest takes
about 36 seconds since
commit 30c6d99209
Author: Jiri Denemark <jdenemar@redhat.com>
Date: Thu Oct 24 17:51:42 2019 +0200
qemuxml2argvtest: Update host arch for DO_TEST*ARCH* tests
The optimizer is hitting some pathological performance behaviour due to
the high number of branches in the mymain() method.
Pushing the branch tests down into the testCompareXMLToArgv method
brings the compile time down to 3 seconds.
This likely related to this GCC bug:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58479
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The virTypedParamsFilter function doesn't mind params == NULL if nparams
is zero. And there's no need to check for params == NULL && nparams > 0
because this is checked higher in the stack.
In fact all the virCheckNonNull* checks in virTypedParamsFilter are
useless.
https://bugzilla.redhat.com/show_bug.cgi?id=1777094
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
The default macOS image in travis is broken, throwing python
exceptions when trying to install glib. Explicitly ask for the
newer 10.3 image which works correctly.
We now need to also point to the homebrew installed libxml2
rather than the OS distro provided one, since the OS distro
one has a pkg-config file present, but no actual header
files.
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Now that we have a separate job type which will not trigger normal code
paths for terminating job we can remove the ad-hoc handling.
This possibly fixes the issue of a broken job inheriting the disk and
then finishing in which case we'd not detach the backing chain.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
To better track jobs we couldn't parse let's introduce a new job type
which will clarify semantics internally in few places.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
We will need to clear per-job type data when we will be marking a
blockjob as broken in the new way. Extract the code for future reuse.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
Both failure to refresh and to dismiss the job are very unlikely but if
they happen there's not much we can do about the blockjob.
The concluded job handlers treat it as if the job failed if we don't
update the state to 'QEMU_BLOCKJOB_STATE_COMPLETED' which is probably
the safest thing to do here.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
Otherwise it would get dropped later on as untracked despite us knowing
about it. Additionally since we cancelled it we must wait to dismiss it
which would not be possible if we unregister it. This also opened a
window for a race condition since the job state change event of the
just-cancelled job might be delivered prior to us unregistering the job
in which case everything would work properly.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
Since we don't know what happened to the job we can't do much about it
but we can at least log that this happened.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
We must exit the monitor prior to refusing other work, otherwise the VM
object will become unusable.
This bug was introduced in commit v5.5.0-244-gc412383796 but thankfully
the code path was not excercised without QEMU_CAPS_BLOCKDEV.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
Block jobs may be members of async jobs so it makes more sense to
refresh block job state after we do steps for async job recovery.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
qemu returns an error message in the job statistics even if the job was
cancelled to emphasize it was not successful. Libvirt didn't properly
transform it into QEMU_BLOCKJOB_STATE_CANCELLED though.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
Commit ed56851f1b didn't wire up fetching of the statistics for the
job which are reported by 'query-jobs'.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
The magic number is taken from the coreutils stat.c file since
there is no constant for it in normal system headers.
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Fabiano Fidêncio <fidencio@redhat.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
This reverts commit 421c9550f5
qemuDomainBlockPullCommon calls virDomainObjEndAPI internally so the
original commit made us shed two references of @vm instead of one
getting us into a premature free of @vm.
This is not a straight revert as qemuDomainBlockPull was modified
meanwhile. I've also added a warning comment that @vm is consumed.
https://bugzilla.redhat.com/show_bug.cgi?id=1777230
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
There are two daemons that wait for acquiring their pid files:
virtnetworkd and virtstoraged. This is undesirable as the idea
is to quit early if unable to acquire the pid file.
Fixes: v5.6.0-rc1~207.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Commit d30a1ad044 translated the symbol file checker from perl to
python by doing a literal translation in most cases. Unfortunately one
string formatting operation was not really translated into python
leaving users with non-helpful error:
'Symbol $1 is listed twice'
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
In the past the network driver was (mistakenly) being called for all
interfaces, not just those of type='network', and so it had a chance
to validate all interface configs after the actual type of the
interface was known.
But since the network driver has been more completely/properly
separated from qemu, the network driver isn't called during the
startup of any interfaces except those with type='network', so this
validation no longer takes place for, e.g. <interface type='bridge'>
(or direct, etc). This in turn meant that a config could erroneously
specify a vlan tag, or bandwidth settings, for a type of interface
that didn't support it, and the domain would start without complaint,
just silently ignoring those settings.
This patch moves those validation checks out of the network driver,
and into virDomainActualNetDefValidate() so they will be done for all
interfaces, not just type='network'.
https://bugzilla.redhat.com/1741121
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
<interface> devices (virDomainNetDef) are a bit different from other
types of devices in that their actual type may come from a network (in
the form of a port connection), and that doesn't happen until the
domain is started. This means that any validation of an <interface> at
parse time needs to be a bit liberal in what it accepts - when
type='network', you could think that something is/isn't allowed, but
once the domain is started and a port is created by the configured
network, the opposite might be true.
To solve this problem hypervisor drivers need to do an extra
validation step when the domain is being started. I recently (commit
3cff23f7, libvirt 5.7.0) added a function to peform such validation
for all interfaces to the QEMU driver -
qemuDomainValidateActualNetDef() - but while that function is a good
single point to call for the multiple places that need to "start" an
interface (domain startup, device hotplug, device update), it can't be
called by the other hypervisor drivers, since 1) it's in the QEMU
driver, and 2) it contains some checks specific to QEMU. For
validation that applies to network devices on *all* hypervisors, we
need yet another interface validation function that can be called by
any hypervisor driver (not just QEMU) right after its network port has
been created during domain startup or hotplug. This patch adds that
function - virDomainActualNetDefValidate(), in the conf directory,
and calls it in appropriate places in the QEMU, lxc, and libxl
drivers.
This new function is the place to put all network device validation
that 1) is hypervisor agnostic, and 2) can't be done until we know the
"actual type" of an interface.
There is no framework for validation at domain startup as there is for
post-parse validation, but I don't want to create a whole elaborate
system that will only be used by one type of device. For that reason,
I just made a single function that should be called directly from the
hypervisors, when they are initializing interfaces to start a domain,
right after conditionally allocating the network port (and regardless
of whether or not that was actually needed). In the case of the QEMU
driver, qemuDomainValidateActualNetDef() is already called in all the
appropriate places, so we can just call the new function from
there. In the case of the other hypervisors, we search for
virDomainNetAllocateActualDevice() (which is the hypervisor-agnostic
function that calls virNetworkPortCreateXML()), and add the call to our
new function right after that.
The new function itself could be plunked down into many places in the
code, but we already have 3 validation functions for network devices
in 2 different places (not counting any basic validation done in
virDomainNetDefParseXML() itself):
1) post-parse hypervisor-agnostic
(virDomainNetDefValidate() - domain_conf.c:6145)
2) post-parse hypervisor-specific
(qemuDomainDeviceDefValidateNetwork() - qemu_domain.c:5498)
3) domain-start hypervisor-specific
(qemuDomainValidateActualNetDef() - qemu_domain.c:5390)
I placed (3) right next to (2) when I added it, specifically to avoid
spreading validation all over the code. For the same reason, I decided
to put this new function right next to (1) - this way if someone needs
to add validation specific to qemu, they go to one location, and if
they need to add validation applying to everyone, they go to the
other. It looks a bit strange to have a public function in between a
bunch of statics, but I think it's better than the alternative of
further fragmentation. (I'm open to other ideas though, of course.)
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
These all just return a scalar value, so there's no daisy-chained
fallout from changing them, and they can easily be combined in a
single patch.
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
This also isn't required (due to the vportprofile being stored in the
NetDef as a pointer rather than being directly contained), but it
seemed dishonest to not mark it as const (and thus permit users to
modify its contents)
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
In this case, the virNetDevBandwidthPtr that is returned is not to a
region within the virDomainNetDef arg, but points elsewhere (the
NetDef has the pointer, not the entire object), so technically it's
not necessary to make the return value a const, but it's a bit
disingenuous to *not* do it.
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
This is needed if we want to call the function when the
virDomainNetDef* we have is a const.
Since virDomainNetGetActualVlan returns a pointer to memory that is
within the virDomainNetDefPtr arg, the returned pointer must also be
made const. This leads to a cascade of other virNetDevVlanPtr's that
must be changed to "const virNetDevVlan *".
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
This makes it easier to understand which interface's config caused the
error.
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Cole Robinson <crobinso@redhat.com>
The cpuModels member of _virQEMUCapsAccel struct is not a
virObject but regular struct with a free function defined:
qemuMonitorCPUDefsFree(). Use that when clearing parent structure
instead of virObjectUnref() to avoid a memleak:
==212322== 57,275 (48 direct, 57,227 indirect) bytes in 3 blocks are definitely lost in loss record 623 of 627
==212322== at 0x4838B86: calloc (vg_replace_malloc.c:762)
==212322== by 0x554A158: g_malloc0 (in /usr/lib64/libglib-2.0.so.0.6000.6)
==212322== by 0x17B14BF5: qemuMonitorCPUDefsNew (qemu_monitor.c:3587)
==212322== by 0x17B27BA7: qemuMonitorJSONGetCPUDefinitions (qemu_monitor_json.c:5616)
==212322== by 0x17B14B0B: qemuMonitorGetCPUDefinitions (qemu_monitor.c:3559)
==212322== by 0x17A6AFBB: virQEMUCapsFetchCPUDefinitions (qemu_capabilities.c:2571)
==212322== by 0x17A6B2CC: virQEMUCapsProbeQMPCPUDefinitions (qemu_capabilities.c:2629)
==212322== by 0x17A70C00: virQEMUCapsInitQMPMonitorTCG (qemu_capabilities.c:4769)
==212322== by 0x17A70DDF: virQEMUCapsInitQMPSingle (qemu_capabilities.c:4820)
==212322== by 0x17A70E99: virQEMUCapsInitQMP (qemu_capabilities.c:4848)
==212322== by 0x17A71044: virQEMUCapsNewForBinaryInternal (qemu_capabilities.c:4891)
==212322== by 0x17A7119C: virQEMUCapsNewData (qemu_capabilities.c:4923)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
On s390 machines host-passthrough and host-model CPUs result in the same
guest ABI (with QEMU new enough to be able to tell us what "host" CPU is
expanded to, which was implemented around 2.9.0). So instead of using
host-passthrough CPU when there's no CPU specified in a domain XML we
can safely use host-model and benefit from CPU compatibility checks
during migration, snapshot restore and similar operations.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The match attribute is only relevant for custom mode CPUs. Reporting
failure when match == 'minimum' regardless on CPU mode can cause
unexpected failures. We should only report the error for custom CPUs. In
fact, calling virCPUs390Update on a custom mode CPU should always report
an error as optional features are not supported on s390 either.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Most likely for historical reasons our CPU def formatting code is
happily adding useless <model fallback='allow'/> for host-model CPUs. We
can just drop it.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Commit v0.8.4-66-g95ff6b18ec (9 years ago) changed the default value for
the cpu/@match attribute to 'exact' in a rather complicated way. It did
so only if <model> subelement was present and set -1 otherwise (which is
not expected to ever happen). Thus the following two equivalent XML
elements:
<cpu mode='host-model'/>
and
<cpu mode='host-model'>
<model/>
</cpu>
would be parsed differently. The former would end up with match == -1
while the latter would have match == 1 ('exact'). This is not a big deal
since the match attribute is ignored for host-model CPUs, but we can
simplify the code and make it a little bit saner anyway.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The test case for x86_64 and neither cirrus nor vga capability is of the
xml2argv type because it actually fails to parse the XML at all [*] which
is something that xml2xml tests don't seem to handle. xml2argv test fails
to produce a qemu argv for this case which xml2argv tests can handle.
[*] This is a consequence of the decision not to have a fallback if the
obvious choices (cirrus and vga) aren't viable due to missing QEMU caps.
Reviewed-by: Cole Robinson <crobinso@redhat.com>
Signed-off-by: Pavel Mores <pmores@redhat.com>
