Adjust the config code so that it does not enforce that target memory
node is specified. To avoid breakage, adjust the qemu memory hotplug
config checker to disallow such config for now.
Since we already make sure before that the domain configuration is
valid we may execute it always at the cost of doing 0 iterations of the
for loop.
This patch will simplify later refactor as it will avoid whitespace
changes.
Make the function usable so that -1 can be passed to it as cell ID so
that we can later enable memory hotplug on non-NUMA guests for certain
architectures.
If mlock is required either due to use of VFIO hostdevs or due to the
fact that it's enabled it needs to be tweaked prior to adding new memory
or after removing a module. Add a helper to determine when it's
necessary and reuse it both on hotplug and hotunplug.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1273491
The code reported that a migration flag is unsupported but didn't jump
to the error label. Probably an oversight in commit f88af9dc that
introduced the flag checking.
Since the flag was not enabled when 'eating' the migration cookie,
libvirt reported a bogus error when memory hotplug was enabled:
unsupported migration cookie feature memory-hotplug
The error was ignored though due to a bug in the code so it slipped
through testing.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1278404
The condition checking whether --format was specified was incorrect.
virsh crashed if the following format was used:
virsh dump VM dump --format '' --memory-only
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1272301
qemuMigrationIsAllowed would disallow offline migration if the VM
contained host devices or memory modules. Since during offline migration
we don't transfer any state we can safely migrate VMs with such
configuration.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1265049
Use the migration @flags for checking various migration aspects rather
than picking them out as booleans. Document the new semantics in the
function header.
Now that qemuMigrationIsAllowed is always called with @vm, we can drop
the @def argument and simplify the control flow.
Additionally the comment is invalid so drop it.
Extract the hostdev check from qemuMigrationIsAllowed into a separate
function since that is the only part that needs to be done in the v2
migration protocol prepare phase on the destination. All other checks
were added when the v3 protocol existed so they don't need to be
extracted.
This change will allow to drop the @def argument for
qemuMigrationIsAllowed and further simplify the function.
Our docs state that subelements of <metadata> shall have a namespace
and the medatata APIs expect that too. To avoid inaccessible
<metadata> sub-elements, just remove those that don't conform to the
documentation.
Apart from adding the new condition this patch renames the function and
refactors the code flow to allow the changes.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1245525
Commit 792f81a40e caused a regression in the libssh2 host key
verification code by changing the variable type of 'i' to unsigned.
Since one of the loops used -1 as a special value if the asking
callback was found the conversion made a subsequent test always fail.
The bug was stealth enough to pass review, compilers and coverity.
Refactor the condition to avoid problems.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1047861
Since we'd disallow migration of a guest that would have possibly
invalid config but still be able to work, relax the WWN check to be
performed only on new starts of the VM.
Although 0 length block jobs aren't entirely useful, the output of virsh
blockjob is empty due to the condition that suppresses the output for
migration jobs that did not start. Since the only place that actually
uses the condition that suppresses the output is in migration, let's
move the check there and thus add support for 0 of 0 equaling to 100%.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1196711
Qemu unfortunately doesn't update internal state right after migration
and so the actual balloon size as returned by 'query-balloon' are
invalid for a while after the CPUs are started after migration. If we'd
refresh our internal state at this point we would report invalid current
memory size until the next balloon event would arrive.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1242940
My original implementation was based on a qemu version that still did
not have all the checks in place. Using sizes that would align to odd
megabyte increments will produce the following error:
qemu-kvm: -device pc-dimm,node=0,memdev=memdimm0,id=dimm0: backend memory size must be multiple of 0x200000
qemu-kvm: -device pc-dimm,node=0,memdev=memdimm0,id=dimm0: Device 'pc-dimm' could not be initialized
Introduce an alignment retrieval function for memory devices and use it
to align the devices separately and modify a test case to verify it.
For some machine types ppc64 machines now require that memory sizes are
aligned to 256MiB increments (due to the dynamically reconfigurable
memory). As now we treat existing configs reasonably in regards to
migration, we can round all the sizes unconditionally. The only drawback
will be that the memory size of a VM can potentially increase by
(256MiB - 1byte) * number_of_NUMA_nodes.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1249006
When we are starting a qemu process for an incomming migration or
snapshot reloading we should not modify the memory sizes in the domain
since we could potentially change the guest ABI that was tediously
checked before. Additionally the function now updates the initial memory
size according to the NUMA node size, which should not happen if we are
restoring state.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1252685
When implementing memory hotplug I've opted to recalculate the initial
memory size (contents of the <memory> element) as a sum of the sizes of
NUMA nodes when NUMA was enabled. This was based on an assumption that
qemu did not allow starting when the NUMA node size total didn't equal
to the initial memory size. Unfortunately the check was introduced to
qemu just lately.
This patch uses the new XML parser flag to decide whether it's safe to
update the memory size total from the NUMA cell sizes or not.
As an additional improvement we now report an error in case when the
size of hotplug memory would exceed the total memory size.
The rest of the changes assures that the function is called with correct
flags.
Add 'initial_memory' member to struct virDomainMemtune so that the
memory size can be pre-calculated once instead of inferring it always
again and again.
Separating of the fields will also allow finer granularity of decisions
in later patches where it will allow to keep the old initial memory
value in cases where we are handling incomming migration from older
versions that did not always update the size from NUMA as the code did
previously.
The change also requires modification of the qemu memory alignment
function since at the point where we are modifying the size of NUMA
nodes the total size needs to be recalculated too.
The refactoring done in this patch also fixes a crash in the hyperv
driver that did not properly initialize def->numa and thus
virDomainNumaGetMemorySize(def->numa) crashed.
In summary this patch should have no functional impact at this point.
The post parse func is growing rather large. Since later patches will
introduce more logic in the memory post parse code, split it into a
separate handler.
Add a new parser flag that will mark code paths that parse XML files
wich will not be used with existing VM state so that post parse
callbacks can possibly do ABI incompatible changes if needed.
The flag was used only for formatting the XML and once the parser and
formatter flags were split in 0ecd685109
it doesn't make sense any more to have it.
Extract the size determination into a separate function and reuse it
across the memory device alignment functions. Since later we will need
to decide the alignment size according to architecture let's pass def to
the functions.
Commit 8125113c added code that should remove the disk backend if the
fronted hotplug failed for any reason. The code had a bug though as it
used the disk string for unplug rather than the backend alias. Fix the
code by pre-creating an alias string and using it instead of the disk
string. In cases where qemu does not support QEMU_CAPS_DEVICE, we ignore
the unplug of the backend since we can't really create an alias in that
case.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1262399
If the current live definition does not have memory hotplug enabled, but
the persistent one does libvirt would reject migration if the
destination does not support memory hotplug even if the user didn't want
to persist the VM at the destination and thus the XML containing the
memory hotplug definition would not be used. To fix this corner case the
code will check for memory hotplug in the newDef only if
VIR_MIGRATE_PERSIST_DEST was used.
Commit 6983d6d2 tried to improve parseRateStr but broke the build
instead for compilers that were not able to properly introspect the for
loop indexed by the enum resulting into the following error:
virsh-domain.c: In function 'parseRateStr':
virsh-domain.c:916:13: error: 'field_name' may be used uninitialized in this function [-Werror=maybe-uninitialized]
vshError(ctl, _("malformed %s field"), field_name);
^
virsh-domain.c:915:13: error: 'tmp' may be used uninitialized in this function [-Werror=maybe-uninitialized]
if (virStrToLong_ullp(token, NULL, 10, tmp) < 0) {
^
Rather than trying to fix the code, refactor the function again by
reusing virStringSplit.
Qemu reports physical size 0 for block devices. As 15fa84acbb
changed the behavior of qemuDomainGetBlockInfo to just query the monitor
this created a regression since we didn't report the size correctly any
more.
This patch adds code to refresh the physical size of a block device by
opening it and seeking to the end and uses it both in
qemuDomainGetBlockInfo and also in qemuDomainGetStatsOneBlock that was
broken since it was introduced in this respect.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1250982
Libvirt doesn't reliably know the location of the backing chain when
pre-creating images for non-shared migration. This isn't a problem for
full copy, but incremental copy requires the information.
Forbid pre-creating the image in cases where incremental migration is
required. This limitation can perhaps be lifted once libvirt will fully
support loading of backing chain information from the XML.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1249587
If destination libvirt doesn't support memory hotplug since all the
support was introduced by adding new elements the destination would
attempt to start qemu with an invalid configuration. The worse part is
that qemu might hang in such situation.
Fix this by sending a required migration feature called 'memory-hotplug'
to the destination. If the destination doesn't recognize it it will fail
the migration.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1248350
virDomainDeleteConfig is meant to delete the persistent config and thus
it resets vm->autostart. Copy parts of qemuProcessRemoveDomainStatus to
a new helper to avoid using the incorrect function.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1230071
The scope name, even according to our docs is
"machine-$DRIVER\x2d$VMNAME.scope" virSystemdMakeScopeName would use the
resource partition name instead of "machine-" if it was specified thus
creating invalid scope paths.
This makes libvirt drop cgroups for a VM that uses custom resource
partition upon reconnecting since the detected scope name would not
match the expected name generated by virSystemdMakeScopeName.
The error is exposed by the following log entry:
debug : virCgroupValidateMachineGroup:302 : Name 'machine-qemu\x2dtestvm.scope' for controller 'cpu' does not match 'testvm', 'testvm.libvirt-qemu' or 'machine-test-qemu\x2dtestvm.scope'
for a "/machine/test" resource and "testvm" vm.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1238570
Reuse the vshBlockJobWait infrastructure to refactor cmdBlockCommit to
use the common code. This additionally fixes a bug when working with
new qemus, where when doing an active commit with --pivot the pivoting
would fail, since qemu reaches 100% completion but the job doesn't
switch to synchronized phase right away.
Introduce helper function that will provide logic for waiting for block
job completion so the 3 open coded places can be unified and improved.
This patch introduces the whole logic and uses it to fix
cmdBlockJobPull. The vshBlockJobWait function provides common logic for
block job waiting that should be robust enough to work across all
previous versions of libvirt. Since virsh allows passing user-provided
strings as paths of block devices we can't reliably use block job events
for detection of block job states so the function contains a great deal
of fallback logic.
Few parts of the code looked at the current progress of and assumed that
a two phase blockjob is in the _READY state as soon as the progress
reached 100% (info.cur == info.end). In current versions of qemu this
assumption is invalid and qemu exposes a new flag 'ready' in the
query-block-jobs output that is set to true if the job is actually
finished.
This patch adds internal data handling for reading the 'ready' flag and
acting appropriately as long as the flag is present.
While this still doesn't fix the virsh client problem with two phase
block jobs and the --pivot option, it at least improves the error
message:
$ virsh blockcommit --wait --verbose vm vda --base vda[1] --active --pivot
Block commit: [100 %]error: failed to pivot job for disk vda
error: internal error: unable to execute QEMU command 'block-job-complete': The active block job for device 'drive-virtio-disk0' cannot be completed
to
$ virsh blockcommit --wait --verbose VM vda --base vda[1] --active --pivot
Block commit: [100 %]error: failed to pivot job for disk vda
error: block copy still active: disk 'vda' not ready for pivot yet
Use the VSH_EXCLUSIVE_OPTIONS to exclude combinations of --pivot and
--keep-overlay and refactor the enforcing of the --wait option and other
flags that imply --wait.
Use the VSH_EXCLUSIVE_OPTIONS_VAR to interlock incompatible options.
Since a variable named 'abort' would conflict with older compilers use
VSH_EXCLUSIVE_OPTIONS for the --abort option.
In commit 641a145d73 I've added code that
resets the balloon memory value to full size prior to resuming the vCPUs
since the size certainly was not reduced at that point.
Since qemuProcessStart is used also in code paths with already booted
up guests (migration, save/restore) the assumption is not entirely true
since the guest might already been running before.
This patch adds a function that queries the monitor rather than using
the full size since a balloon event would not be reissued in case we are
recovering a saved migration state.
Additionally the new function is used also when reconnecting to a VM
after libvirtd restart since we might have missed a few balloon events
while libvirtd was not running.
In commit 714b38cb23 I tried to avoid
having two disks with the same WWN in a VM. I forgot to check the
hotplug paths though which make it possible bypass that check. Reinforce
the fix by checking the wwn when attaching the disk.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1208009
Optimize the virBitmap to array-of-char bitmap conversion by skipping
trailing zero bytes.
This also fixes a regression when requesting iothread information from a
live VM since after commit 825df8c315 the
bitmap returned from virProcessGetAffinity is too big to be formatted
properly via RPC. A user would get the following error:
error: Unable to get domain IOThreads information
error: Unable to encode message payload
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1238589
Using a custom device tree image may cause unexpected behavior in
architectures that use this approach to detect platform devices. Since
usually the device tree is generated by qemu and thus it's not normally
used let's taint VMs using it to make it obvious as a possible source of
problems.
Since the balloon driver does not guarantee that it returns memory to
the host, using the value in the audit message is not a good idea.
This patch removes auditing from updating the balloon size and reports
the total physical size at startup.
While CPU0 was made unpluggable in Linux a while ago it's not desirable
to unplug it since some parts of the kernel (suspend-to-ram) still
depend on it.
This patch fixes the vCPU selection code in libvirt so that it will not
be disabled.
Spice events have mostly similar information present in the event JSON
but they differ in the name of the element containing the port.
The JSON event also provides connection ID which might be useful in the
future.
This patch splits up the event parser code into two functions and the
SPICE reimplements the event parsing with correct names and drops the
VNC only stuff.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1236585
Instead of using qemuMonitorJSONDevGetBlockExtent (which I plan to
remove later) extract the data in place.
Additionally add a flag that will be set when the wr_highest_offset was
extracted correctly so that callers can act according to that.
The test case addition should help make sure that everything works.
Function prlsdkGetStatsParam was missing a prototype or the static
keyword. I went with static since it built successfully.
Pushed as a build breaker fix.
Drop internal data structures and use the proper fields in virDomainDef.
This allows to greatly simplify the code and allows to remove the
private data structure that was holding just redundant data.
This patch also fixes the bogous output where we'd report that a fresh
VM without vCPU pinning would not run on all vcpus.
Only self-locking APIs are used and the pointer is immutable so there's
no need to lock the driver to access the domain list.
This patch removes locking partially for everything that will not be
converted to testDomObjFromDomain in the next patch.
Make testObjectEventQueue tolerant to NULL @event and move it so that it
does not require a prototype. Additionally we are now able to remove
locking when accessing driver->eventState, since it's using self-locking
APIs and the pointer is immutable.
When the block job would fail while watching it using the "--wait"
option for blockcopy, virsh would rather unhelpfully report:
$ virsh blockcopy vm hdc /tmp/raw.img --granularity 4096 --verbose --wait
Now in mirroring phase
Add a special case when the block job vanishes while waiting for it to
finish to improve the message:
$ virsh blockcopy vm hdc /tmp/raw.img --granularity 8192 --verbose --wait
error: Block Copy unexpectedly failed
Peter Krempa