While we need to know the difference between the total memory stored in
<memory> and the actual size not included in the possible memory modules
we can't pre-calculate it reliably. This is due to the fact that
libvirt's XML is copied via formatting and parsing the XML and the
initial memory size can be reliably calculated only when certain
conditions are met due to backwards compatibility.
This patch removes the storage of 'initial_memory' and fixes the helpers
to recalculate the initial memory size all the time from the total
memory size. This conversion is possible when we also make sure that
memory hotplug accounts properly for the update of the total memory size
and thus the helpers for inserting and removing memory devices need to
be tweaked too.
This fixes a bug where a cold-plug and cold-remove of a memory device
would increase the size reported in <memory> in the XML by the size of
the memory device. This would happen as the persistent definition is
copied before attaching the device and this would lead to the loss of
data in 'initial_memory'.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1344892
Until now, a Q35 domain (or arm/virt, or any other domain that has a
pcie-root bus) would always have a pci-bridge added, so that there
would be a hotpluggable standard PCI slot available to plug in any PCI
devices that might be added. This patch removes the explicit add,
instead relying on the pci-bridge being auto-added during PCI address
assignment (it will add a pci-bridge if there are no free slots).
This doesn't eliminate the dmi-to-pci-bridge controller that is
explicitly added whether or not a standard PCI slot is required (and
that is almost never used as anything other than a converter between
pcie.0's PCIe slots and standard PCI). That will be done separately.
Previously there was no way to have a Q35 domain that didn't have
these two controllers. This patch skips their creation as long as
there are some other kinds of pci controllers at index 1 and 2
(e.g. some pcie-root-port controllers).
I'm hoping that soon we won't add them at all, plugging all devices
into auto-added pcie-*-port ports instead, but in the meantime this
makes it easier to experiment with alternative bus hierarchies.
The other two DomainHasBlockJob usage error messages don't contain
'an', so unify things to save translators some effort. Dropping
the 'an' is closer to the sentence structure in the errors from
qemuDomainDiskBlockJobIsActive as well
If the domain is not running, but for example the CPUs are stopped, the
ACPI event gets queued and resume of the domain will just shut it off.
https://bugzilla.redhat.com/show_bug.cgi?id=1216281
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Since obtaining a job can wait for another job to finish, the state
might change in the meantime. And checking it more than once is
pointless.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
The virQEMUDriverConfig object contains lists of
loader:nvram pairs to advertise firmwares supported by
by the driver, and qemu_conf.c contains code to populate
the lists, all of which is useful for other drivers too.
To avoid code duplication, introduce a virFirmware object
to encapsulate firmware details and switch the qemu driver
to use it.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1298070
We have the code for attaching redirdevs for ages now.
Unfortunately, our monitor code that handles talking to the qemu
process was missing a little piece of code that actually enabled
the feature.
BTW: it really is called "type" on the monitor, even though it's
called "name" on the cmd line. Don't ask.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Move all APIs with a virHostMEM name prefix out into new
util/virhostmem.h & util/virhostmem.c files
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Move all APIs with a virHostCPU name prefix out into new
util/virhostcpu.h & util/virhostcpu.c files
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
In preparation for moving all the CPU related APIs out of
the nodeinfo file, give them a virHostCPU name prefix.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
In preparation for moving all the memory related APIs out of
the nodeinfo file, give them a virHostMem name prefix.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Nearly all the methods in the nodeinfo file are given a
'const char *sysfs_prefix' parameter to override the
default sysfs path (/sys/devices/system). Every single
caller passes in NULL for this, except one use in the
unit tests. Furthermore this parameter is totally
Linux-specific, when the APIs are intended to be cross
platform portable.
This removes the sysfs_prefix parameter and instead gives
a new method linuxNodeInfoSetSysFSSystemPath for use by
the test suite.
For two of the methods this hardcodes use of the constant
SYSFS_SYSTEM_PATH, since the test suite does not need to
override the path for thos methods.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If you want to set block device I/O tuning values that end with '_max'
and there is nothing else set, libvirt emits an error. In particular:
error: internal error: Unexpected error
That's an unknown error. That is because *_max values depend on their
respective non-_max values. QEMU even says that in the error message
sent as a response to the monitor command:
"error": {"class": "GenericError", "desc": "bps_max/iops_max require
corresponding bps/iops values"}
the problem was that we didn't know that and there was no check for it.
Adding such check makes sure that there will be less confused users.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
This new listen type is currently supported only by spice graphics.
It's introduced to make it easier and clearer specify to not listen
anywhere in order to start a guest with OpenGL support.
The old way to do this was set spice graphics autoport='no' and don't
specify any ports. The new way is to use <listen type='none'/>. In
order to be able to migrate to old libvirt the migratable XML will be
generated without the listen element and with autoport='no'. Also the
old configuration will be automatically converted to the this listen
type.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
VNC graphics already supports sockets but only via 'socket' attribute.
This patch coverts that attribute into listen type 'socket'.
For backward compatibility we need to handle listen type 'socket' and 'socket'
attribute properly to support old XMLs and new XMLs. If both are provided they
have to match, if only one of them is provided we need to be able to parse that
configuration too.
To not break migration back to old libvirt if the socket is provided by user we
need to generate migratable XML without the listen element and use only 'socket'
attribute.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
This moves the socket generation if "vnc_auto_unix_socket" is set.
It also fixes a bug with this config option that we should auto-generate
socket path only if listen type is address and there is no address
specified.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Even though it's auto-generated it's based on qemu.conf option and listen type
address already uses "fromConfig" to carry this information. Following commits
will convert the socket to listen element so this rename is required because
there will be also an option to get socket auto-generated independently on the
qemu.conf option.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Since commit 7140807917, qemu agent
channel cannot be plugged in because we won't generate its path
automatically. Let's not only fix that, but also add tests for it so
next time it's checked for.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1322210
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Put it into separate function called qemuDomainPrepareChannel() and call
it from the new qemuProcessPrepareDomain().
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
CPUID instruction normally takes its parameter from EAX, but sometimes
ECX is used as an additional parameter. This patch prepares the x86 CPU
driver code for the new 'ecx_in' CPUID parameter.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
virCPUData, virCPUx86Feature, and virCPUx86Model all contained a pointer
to virCPUx86Data, which was not very nice since the real CPUID data were
accessible by yet another pointer from virCPUx86Data. Moreover, using
virCPUx86Data directly will make static definitions of internal CPU
features a bit easier.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
This patch splits qemuMonitorJSONGetCPUx86Data in three functions:
- qemuMonitorJSONCheckCPUx86 checks if QEMU supports reporting CPUID
features for a guest CPU
- qemuMonitorJSONParseCPUx86Features parses CPUID features from a JSON
array
- qemuMonitorJSONGetCPUx86Data gets the requested guest CPU property
from QOM and uses qemuMonitorJSONParseCPUx86Features to parse it
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
CPUID instruction normally takes its parameter from EAX, but sometimes
ECX is used as an additional parameter. Let's rename 'function' to
'eax_in' in preparation for adding 'ecx_in'.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
The virConnectOpenInternal method opens the libvirt client
config file and uses it to resolve things like URI aliases.
There may be driver specific things that are useful to
store in the config file too, so rather than have them
re-parse the same file, pass the virConfPtr down to the
drivers.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
In commit 1e38ef72 the disk startup policy check was moved prior to the
call to virDomainObjSetDefTransient which dropped the disk from the
config rather than the def to be started which is a bug.
Additionally we'd not report the disk change event for this since the
disk aliases were not set at that point.
Finally 'volume' based disks would not work with startup policy too.
Fix it by moving it back after the definition is copied, aliases are
assigned and disk sources are translated.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1341415
qemuProcessStart does not unset the infrastructure that retrieves errors
from the qemu log file in case of migration. As this wasn't handled
properly in qemuDomainSaveImageStartVM we kept the logging context/fd
open for the lifetime of the VM rather than closing it after it's not
needed.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1325080
Use qemuDomainLogAppendMessage rather than attempting to open a new
logging context with file descriptors. The new approach allows to log
the message even if qemu is still running at that point which appens
during migration finish phase where qemuProcessStop is killing qemu.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1312188
Along with the virtlogd addition of the log file appending API implement
a helper for logging one-shot entries to the log file including the
fallback approach of using direct file access.
This will be used for noting the shutdown of the qemu proces and
possibly other actions such as VM migration and other critical VM
lifecycle events.
Since it will not be called from outside of conf we can unexport it too
if we move it to the appropriate place.
Test suite change is necessary since the error will be reported sooner
now.
Validation of qemu process startup requires to know whether the process
is used for a fresh VM or whether it's reloaded from a
snapshot/migration. Pass this information in via a flag rather than
calculating it from a bunch of bools.
Similarly to the domain definition validator add a device validator. The
change to the prototype of the domain validator is necessary as
virDomainDeviceInfoIterateInternal requires a non-const pointer.
Until now we weren't able to add checks that would reject configuration
once accepted by the parser. This patch adds a new callback and
infrastructure to add such checks. In this patch all the places where
rejecting a now-invalid configuration wouldn't be a good idea are marked
with a new parser flag.
Historically, we added heads=1 to videos, but for example for qxl, we
did not reflect that on the command line.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1283207
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Move the module from qemu_command.c to a new module virqemu.c and
rename the API to virQEMUBuildObjectCommandline.
This API will then be shareable with qemu-img and the need to build
a security object for luks support.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Remove the live attribute and mark the definition as transient
whether the domain is runing or not.
There were only two callers left calling with live=false:
* testDomainStartState, where the domain already is active
because we assigned vm->def->id just a few lines above the call
* virDomainObjGetPersistentDef, which now only calls
virDomainObjSetDefTransient for an active domain
Commit b4a5fd95 introduced vram64 attribute for QXL video device but
there were two issues. Only function
qemuMonitorJSONUpdateVideoVram64Size should update the vram64 attribute
and also the value is in MiB, not in B.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
There's a bug in the function. We expect the following format for
the data we are parsing here:
key: value
So we use strchr() to find ':' and then see if it is followed by
space. But the check that does just that is slightly incorrect.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Commit ff2126225d changed the error message to be more
detailed about the failure at hand; however, while the new
error message claims that "bus must be <= index", the error
message is displayed if "idx <= addr->bus", ie. when bus
is larger than or *equal to* index.
Change the error message to report the correct constraint,
and format it in a way that mirrors the check exactly to
make it clearer to people reading the code. The new error
message reads "index must be larger than bus".
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1339900
Hand-entering indexes for 20 PCI controllers is not as tedious as
manually determining and entering their PCI addresses, but it's still
annoying, and the algorithm for determining the proper index is
incredibly simple (in all cases except one) - just pick the lowest
unused index.
The one exception is USB2 controllers because multiple controllers in
the same group have the same index. For these we look to see if 1) the
most recently added USB controller is also a USB2 controller, and 2)
the group *that* controller belongs to doesn't yet have a controller
of the exact model we're just now adding - if both are true, the new
controller gets the same index, but in all other cases we just assign
the lowest unused index.
With this patch in place and combined with the automatic PCI address
assignment, we can define a PCIe switch with several ports like this:
<controller type='pci' model='pcie-root-port'/>
<controller type='pci' model='pcie-switch-upstream-port'/>
<controller type='pci' model='pcie-switch-downstream-port'/>
<controller type='pci' model='pcie-switch-downstream-port'/>
<controller type='pci' model='pcie-switch-downstream-port'/>
<controller type='pci' model='pcie-switch-downstream-port'/>
<controller type='pci' model='pcie-switch-downstream-port'/>
...
These will each get a unique index, and PCI addresses that connect
them together appropriately with no pesky numbers required.
IS_USB2_CONTROLLER() is useful in more places aside from just when
assigning PCI addresses in QEMU, and is checking for enum values that
are all defined in conf/domain_conf.h anyway, so define it there
instead.
<os>
<acpi>
<table type="slic">/path/to/acpi/table/file</table>
</acpi>
</os>
will result in:
-acpitable sig=SLIC,file=/path/to/acpi/table/file
This option was introduced by QEMU commit 8a92ea2 in 2009.
https://bugzilla.redhat.com/show_bug.cgi?id=1327537
The libvirt internal bits can be changed for disks that don't otherwise
support changing media. Remove the switch statement and allow changes of
non-source data for all disks.
qemuDomainChangeDiskLive rolled back few changes to the disk definition
if changing of the media failed. This can be avoided by moving some code
around.
Based on some digital archaeology performed by jtomko, it's been determined
that the persistentAddrs variable is no longer necessary...
The variable was added by:
commit 141dea6bc7
CommitDate: 2010-02-12 17:25:52 +0000
Add persistence of PCI addresses to QEMU
Where it was set to 0 on domain startup if qemu did not support the
QEMUD_CMD_FLAG_DEVICE capability, to clear the addresses at shutdown,
because QEMU might make up different ones next time.
As of commit f5dd58a608
CommitDate: 2012-07-11 11:19:05 +0200
qemu: Extended qemuDomainAssignAddresses to be callable from
everywhere.
this was broken, when the persistentAddrs = 0 assignment was moved
inside qemuDomainAssignPCIAddresses and while it pretends to check
for !QEMU_CAPS_DEVICE, its parent qemuDomainAssignAddresses is only
called if QEMU_CAPS_DEVICE is present.
Since commit id '20a0fa8e' removed the QEMU_CAPS_DEVICE, Coverity notes
that it's no longer possible to have 'addrs' be NULL when checking for
a live domain since qemuDomainPCIAddressSetCreate would have jumped to
cleanup if addrs was NULL.
Use the detected tray presence flag to trigger the tray waiting code
only if the given storage device in qemu reports to have a tray.
This is necessary as the floppy device lost it's tray as of qemu commit:
commit abb3e55b5b718d6392441f56ba0729a62105ac56
Author: Max Reitz <mreitz@redhat.com>
Date: Fri Jan 29 20:49:12 2016 +0100
Revert "hw/block/fdc: Implement tray status"
Commit 1fad65d49a used a really big hammer
and overwrote the error message that might be reported by qemu if the
tray is locked. Fix it by reporting the error only if no error is
currently set.
Error after commit mentioned above:
error: internal error: timed out waiting for disk tray status update
New error:
error: internal error: unable to execute QEMU command 'eject': Tray of
device 'drive-ide0-0-0' is not open
Extract information for all disks and update tray state and source only
for removable drives. Additionally store whether a drive is removable
and whether it has a tray.
Extract whether a given drive has a tray and whether there is no image
inserted.
Negative logic for the image insertion is chosen so that the flag is set
only if we are certain of the fact.
Rather than only assigning a PCI address when no address is given at
all, also do it when the config says that the address type is 'pci',
but it gives no address (virDeviceInfoPCIAddressWanted()).
There are also several places after parsing but prior to address
assignment where code previously expected that any info with address
type='pci' would have a *valid* PCI address, which isn't always the
case - now we check not only for type='pci', but also for a valid
address (virDeviceInfoPCIAddressPresent()).
The test case added in this patch was directly copied from Cole's patch titled:
qemu: Wire up address type=pci auto_allocate
https://bugzilla.redhat.com/show_bug.cgi?id=1182074
If they're available and we need to pass secrets to qemu, then use the
qemu domain secret object in order to pass the secrets for RBD volumes
instead of passing the base64 encoded secret on the command line.
The goal is to make AES secrets the default and have no user interaction
required in order to allow using the AES mechanism. If the mechanism
is not available, then fall back to the current plain mechanism using
a base64 encoded secret.
New APIs:
qemu_domain.c:
qemuDomainGetSecretAESAlias:
Generate/return the secret object alias for an AES Secret Info type.
This will be called from qemuDomainSecretAESSetup.
qemuDomainSecretAESSetup: (private)
This API handles the details of the generation of the AES secret
and saves the pieces that need to be passed to qemu in order for
the secret to be decrypted. The encrypted secret based upon the
domain master key, an initialization vector (16 byte random value),
and the stored secret. Finally, the requirement from qemu is the IV
and encrypted secret are to be base64 encoded.
qemu_command.c:
qemuBuildSecretInfoProps: (private)
Generate/return a JSON properties object for the AES secret to
be used by both the command building and eventually the hotplug
code in order to add the secret object. Code was designed so that
in the future perhaps hotplug could use it if it made sense.
qemuBuildObjectSecretCommandLine (private)
Generate and add to the command line the -object secret for the
secret. This will be required for the subsequent RBD reference
to the object.
qemuBuildDiskSecinfoCommandLine (private)
Handle adding the AES secret object.
Adjustments:
qemu_domain.c:
The qemuDomainSecretSetup was altered to call either the AES or Plain
Setup functions based upon whether AES secrets are possible (we have
the encryption API) or not, we have secrets, and of course if the
protocol source is RBD.
qemu_command.c:
Adjust the qemuBuildRBDSecinfoURI API's in order to generate the
specific command options for an AES secret, such as:
-object secret,id=$alias,keyid=$masterKey,data=$base64encodedencrypted,
format=base64
-drive file=rbd:pool/image:id=myname:auth_supported=cephx\;none:\
mon_host=mon1.example.org\:6321,password-secret=$alias,...
where the 'id=' value is the secret object alias generated by
concatenating the disk alias and "-aesKey0". The 'keyid= $masterKey'
is the master key shared with qemu, and the -drive syntax will
reference that alias as the 'password-secret'. For the -drive
syntax, the 'id=myname' is kept to define the username, while the
'key=$base64 encoded secret' is removed.
While according to the syntax described for qemu commit '60390a21'
or as seen in the email archive:
https://lists.gnu.org/archive/html/qemu-devel/2016-01/msg04083.html
it is possible to pass a plaintext password via a file, the qemu
commit 'ac1d8878' describes the more feature rich 'keyid=' option
based upon the shared masterKey.
Add tests for checking/comparing output.
NB: For hotplug, since the hotplug code doesn't add command line
arguments, passing the encoded secret directly to the monitor
will suffice.
Move the logic from qemuDomainGenerateRandomKey into this new
function, altering the comments, variable names, and error messages
to keep things more generic.
NB: Although perhaps more reasonable to add soemthing to virrandom.c.
The virrandom.c was included in the setuid_rpc_client, so I chose
placement in vircrypto.
According to QEMU docs, the '-m' option for specifying RAM is by default
in MiB, and a suffix of "M" or "G" may be passed for values in MiB and
GiB respectively. This commit adds support and a test for the same.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=812295
Signed-off-by: Nishith Shah <nishithshah.2211@gmail.com>
Both VNC and SPICE requires the same code to resolve address for listen
type network. Remove code duplication and create a new function that
will be used in qemuProcessSetupGraphics().
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
For some disk types (SD), we want to emit the syntax
we used for disks before -device was available even
if QEMU supports -device.
Use the qemuDiskBusNeedsDeviceArg helper to figure out
whether to use the old or new syntax.
If the stats for a block device can't be acquired from qemu we've
fallen back to loading them from the file on the disk in libvirt.
If qemu is not cooperating due to being stuck on an inaccessible NFS
share we would then attempt to read the files and get stuck too with
the VM object locked. All other APIs would eventually get stuck waiting
on the VM lock.
Avoid this problem by skipping the block stats if the VM is online but
the monitor did not provide any stats.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1337073
Some Intel processor families (e.g. the Intel Xeon processor E5 v3
family) introduced some RDT (Resource Director Technology) features
to monitor or control shared resource. Among these features, MBM
(Memory Bandwidth Monitoring), which is build on the CMT (Cache
Monitoring Technology) infrastructure, provides OS/VMM a way to
monitor bandwidth from one level of cache to another.
With current perf framework, this patch adds support to perf event
for MBM.
Signed-off-by: Qiaowei Ren <qiaowei.ren@intel.com>
QEMU needs access to the /dev/dri/render* device for
virgl to work.
Allow access to all /dev/dri/* devices for domains with
<video>
<model type='virtio' heads='1' primary='yes'>
<acceleration accel3d='yes'/>
</model>
</video>
https://bugzilla.redhat.com/show_bug.cgi?id=1337290
All qemu versions we support have QEMU_CAPS_DEVICE, so checking
for it is redundant. Remove the usage.
The code diff isn't clear, but all that code is just inindented
with no other change.
Test cases that hit qemuDomainAssignAddresses but don't have
infrastructure for specifying qemuCaps values see lots of
churn, since now PCI addresses are in the XML output.
hotplug APIs with the AFFECT_CONFIG flag are essentially replicating
'insert <device> into XML document, and redefine XML'. Thinking of
it this way, it's natural that we call virDomainDefPostParse after
manually editing the XML here.
Not only does doing so allow us to drop a bunch of open coded calls
to qemuDomainAssignAddresses, but it also means we are going through
the standard channels for XML validation and potentially catching
errors in user submitted XML.
This wires up qemuDomainAssignAddresses into the new
virDomainDefAssignAddressesCallback, so it's always triggered
via virDomainDefPostParse. We are essentially doing this already
with open coded calls sprinkled about.
qemu argv parse output changes slightly since previously it wasn't
hitting qemuDomainAssignAddresses.
When the <gic/> element in not present in the domain XML, use the
domain capabilities to figure out what GIC version is usable and
choose that one automatically.
This allows guests to be created on hardware that only supports
GIC v3 without having to update virt-manager and similar tools.
Keep using the default GIC version if the <gic/> element has been
added to the domain XML but no version has been specified, as not
to break existing guests.
We support omitting listen attribute of graphics element so we should
also support omitting address attribute of listen element. This patch
also updates libvirt to always add a listen element into domain XML
except for VNC graphics if socket attribute is specified.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
The only QEMU versions that don't have such capability are <0.12,
which we no longer support anyway.
Additionally, this solves the issue of some QEMU binaries being
reported as not having such capability just because they lacked
the {kvm-}pci-assign QMP object.
Recent adjustments to the code produced a litany of coverity false
positives, but only because the "standard" procedure of setting a
variable to NULL after it was assigned to something else and keeping
the *Free/*FREE call in the cleanup path wasn't kept. So this patch
makes those adjustments (assign variable to NULL and remove the if
'ret < 0' condition to clean it up).
Signed-off-by: John Ferlan <jferlan@redhat.com>
Rather than returning a "char *" indicating perhaps some sized set of
characters that is NUL terminated, alter the function to return 0 or -1
for success/failure and add two parameters to handle returning the
buffer and it's size.
The function no longer encodes the returned secret, rather it returns
the unencoded secret forcing callers to make the necessary adjustments.
Alter the callers to handle the adjusted model.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Call the internal driver callbacks rather than the public APIs to avoid
calling unnecessarily the error dispatching code and don't overwrite
the error messages provided by the APIs. They are good enough to
describe which secret is missing either by UUID or the usage (basically
name).
When -cpu host is supported by a QEMU binary, a user can use
<cpu mode='host-passthrough'/> in domain XML even when libvirtd failed
to find a matching model for the host CPU. Let's make it obvious by
advertising <cpuselection/> guest capability whenever -cpu host is
supported.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Further followup discussions in list on commit 192a53e concluded
that we should be leaving out the USB controller only for
i440fx machines as default USB can be used by someone on q35
at random slots.
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
Move filling out the default video (v)ram to DeviceDefPostParse.
This means it can be removed from virDomainVideoDefParseXML
and qemuParseCommandLine. Also, we no longer need to special case
VIR_DOMAIN_VIRT_XEN, since the per-driver callback gets called
before the generic one.
libvirt may automatically add a pci-root or pcie-root controller to a
domain, depending on the arch/machinetype, and it hopefully always
makes the right decision about which to add (since in all cases these
controllers are an implicit part of the virtual machine).
But it's always possible that someone will create a config that
explicitly supplies the wrong type of PCI controller for the selected
machinetype. In the past that would lead to an error later when
libvirt was trying to assign addresses to other devices, for example:
XML error: PCI bus is not compatible with the device at
0000:00:02.0. Device requires a PCI Express slot, which is not
provided by bus 0000:00
(that's the error message that appears if you replace the pcie-root
controller in a Q35 domain with a pci-root controller).
This patch adds a check at the same place that the implicit
controllers are added (to ensure that the same logic is used to check
which type of pci root is correct). If a pci controller with index='0'
is already present, we verify that it is of the model that we would
have otherwise added automatically; if not, an error is logged:
The PCI controller with index='0' must be " model='pcie-root' for
this machine type, " but model='pci-root' was found instead.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1004602
Remove the possibility that a NULL hostdev->privateData or a
disk->privateData could crash libvirtd by checking for NULL
before dereferencing for the secinfo structure in the
qemuDomainSecret{Disk|Hostdev}Destroy functions. The hostdevPriv
could be NULL if qemuProcessNetworkPrepareDevices adds a new
hostdev during virDomainNetGetActualHostdev that then gets
inserted via virDomainHostdevInsert. The hostdevPriv was added
by commit id '27726d8' and is currently only used by scsi hostdev.
SRIOV VFs used in macvtap passthrough mode can take advantage of the
SRIOV card's transparent vlan tagging. All the code was there to set
the vlan tag, and it has been used for SRIOV VFs used for hostdev
interfaces for several years, but for some reason, the vlan tag for
macvtap passthrough devices was stubbed out with a -1.
This patch moves a bit of common validation down to a lower level
(virNetDevReplaceNetConfig()) so it is shared by hostdev and macvtap
modes, and updates the macvtap caller to actually send the vlan config
instead of -1.
Requires adding the plumbing for <device><video>
The value is <enum name='modelType'> to match the associated domain
XML of <video><model type='XXX'/>
Wire it up for qemu too
qemuDomainCheckDiskPresence has short-circuit code to skip the
determination of the disk backing chain for storage formats that can't
have backing volumes. The code treats VIR_STORAGE_FILE_NONE as not
having backing chain and skips the call to qemuDomainDetermineDiskChain.
This is wrong as qemuDomainDetermineDiskChain is responsible for storage
format detection and has logic to determine the default type if format
detection is disabled.
This allows to storage passed via <disk type="volume"> to circumvent the
enforcement to have correct storage format or that we shall default to
format='raw', since we don't set the default type via the post parse
callback for "volume" backed disks as the translation code could come up
with a better guess.
This resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1328003
Extract the relevant parts of the existing checker and reuse them for
blockcopy since copying to a non-block device creates an invalid
configuration.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1209802
In qemuCheckDiskConfig would now use virDomainDiskSourceIsBlockType just
as a glorified version of virStorageSourceIsBlockLocal that reports
error messages. Replace it with the latter including the message for
clarity.
Commit c820fbff9f added support for iSCSI
disk as backing for <disk device='lun'>. We would not use it for a disk
type="volume" with direct access mode which basically maps to direct
iSCSI usage. Fix it by adding the storage source type accessor that
resolves the volume type.
For disks sources described by a libvirt volume we don't need to do a
complicated check since virStorageTranslateDiskSourcePool already
correctly determines the actual disk type.
Replace the checks using a new accessor that does not open-code the
whole logic.
In 7884d089d2 I've started to refactor qemu_monitor_json.c.
Thing is, it's current structure is nothing like the rest of our
code. The @ret variable is rewritten all the time, if()-s are
nested instead of using goto and so on.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Move adding the config listen type=address if there is none in
qemuProcessPrepareDomain and move check for multiple listens to
qemuProcessStartValidate.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Add the data structure and infrastructure to support an initialization
vector (IV) secrets. The IV secret generation will need to have access
to the domain private master key, so let's make sure the prepare disk
and hostdev functions can accept that now.
Anywhere that needs to make a decision over which secret type to use
in order to fill in or use the IV secret has a switch added.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Create helper API's in order to build the network URI as shortly we will
be adding a new SecretInfo type
Signed-off-by: John Ferlan <jferlan@redhat.com>
Rather than need to call qemuDomainSecretDestroy after any call to
qemuProcessLaunch, let's do the destroy in qemuProcessLaunch since
that's where command line is eventually generated and processed. Once
it's generated, we can clear out the secrets.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Commit id '40d8e2ba3' added the function to qemuProcessStart because
in order to set up some secrets in the future we will need the master
key. However, since the previous patch split the master key creation
into two parts (create just the key and create the file), we can now
call qemuDomainSecretPrepare from qemuProcessPrepareDomain since the
file is not necessary.
Signed-off-by: John Ferlan <jferlan@redhat.com>
A recent review of related changes noted that we should split the creation
(or generation) of the master key into the qemuProcessPrepareDomain and leave
the writing of the master key for qemuProcessPrepareHost.
Made the adjustment and modified some comments to functions that have
changed calling parameters, but didn't change the intro doc.
Signed-off-by: John Ferlan <jferlan@redhat.com>
From a review after push, add the "_TYPE" into the name.
Also use qemuDomainSecretInfoType in the struct rather than int
with the comment field containing the struct name
Signed-off-by: John Ferlan <jferlan@redhat.com>
virQEMUCapsNewForBinary unconditionally loads data from cache and probes
using both QMP and -help parsing, which is suboptimal when we want to
use it in tests.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1286709
Now that we have all the pieces in place, we can add the 'iothread=#' to
the command line for the (two) controllers that support it (virtio-scsi-pci
and virtio-scsi-ccw). Add the tests as well...
Rather than an if statement, use a switch.
The switch will also catch the illegal usage of 'iothread' with some other
kind of unsupported bus configuration.
An iothread for virtio-scsi is a property of the controller. Add a lookup
of the 'virtio-scsi-pci' and 'virtio-scsi-ccw' device properties and parse
the output. For both, support for the iothread was added in qemu 2.4
while support for virtio-scsi in general was added in qemu 1.4.
Modify the various mock capabilities replies (by hand) to reflect the
when virtio-scsi was supported and then specifically when the iothread
property was added. For versions prior to 1.4, use the no device error
return for virtio-scsi. For versions 1.4 to before 2.4, add some data
for virtio-scsi-pci even though it isn't complete we're not looking for
anything specific there anyway. For 2.4 to 2.6, add a more complete reply.
Signed-off-by: John Ferlan <jferlan@redhat.com>
In majority of our functions we have this variable @ret that is
overwritten a lot. In other areas of the code we use 'goto
cleanup;' just so that this wouldn't happen. But here.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
This adds a ports= attribute to usb controller XML, like
<controller type='usb' model='nec-xhci' ports='8'/>
This maps to:
qemu -device nec-usb-xhci,p2=8,p3=8
Meaning, 8 ports that support both usb2 and usb3 devices. Gerd
suggested to just expose them as one knob.
https://bugzilla.redhat.com/show_bug.cgi?id=1271408
In these functions I'm fixing here, we do call
qemuMonitorJSONCheckError() followed by another check if qemu
reply contains 'return' object. If it wouldn't, the former
CheckError() function would error out and the flow would not even
get to the latter.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Usually, the flow in this area of the code is as follows:
qemuMonitorJSONMakeCommand()
qemuMonitorJSONCommand()
qemuMonitorJSONCheckError()
parseReply()
But in this function, for some reasons, the last two steps were
swapped. This makes no sense.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
In qemuDomainDefAddDefaultDevices we check for a non-NULL
def->os.machine for x86 archs, but not the others.
Moreover, the only caller - qemuDomainDefPostParse
already checks for it and even then it can happen only
if /etc/libvirt contains an XML without a machine type.
We do not need to propagate the exact return values
and the only possible ones are 0 and -1 anyway.
Remove the temporary variable and use the usual pattern:
if (f() < 0)
return -1;
https://bugzilla.redhat.com/show_bug.cgi?id=1139766
Thing is, for some reasons you can have your domain's RTC to be
in something different than UTC. More weirdly, it's not only time
zone what you can shift it of, but an arbitrary value. So, if
domain is configured that way, libvirt will correctly put it onto
qemu cmd line and moreover track it as this offset changes during
domain's life time (e.g. because guest OS decides the best thing
to do is set new time to RTC). Anyway, they way in which this
tracking is implemented is events. But we've got a problem if
change in guest's RTC occurs and the daemon is not running. The
event is lost and we end up reporting invalid value in domain
XML. Therefore, when the daemon is starting up again and it is
reconnecting to all running domains, re-fetch their RTC so the
correct offset value can be computed.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
If a panic device is being defined without a model in a domain
the default value is always overwritten with model ISA. An ISA
bus does not exist on S390 and therefore specifying a panic device
results in an unsupported configuration.
Since the S390 architecture inherently provides a crash detection
capability the panic device should be defined in the domain xml.
This patch adds an s390 panic device model and prevents setting a
device address on it.
Signed-off-by: Boris Fiuczynski <fiuczy@linux.vnet.ibm.com>
Reviewed-by: Cornelia Huck <cornelia.huck@de.ibm.com>
The default USB controller is not sent to destination as the older versions
of libvirt(0.9.4 or earlier as I see in commit log of 409b5f54) didn't
support them. For some archs where the support started much later can
safely send the USB controllers without this worry. So, send the controller
to destination for all archs except x86. Moreover this is not very applicable
to x86 as the USB controller has model ich9_ehci1 on q35 and for pc-i440fx,
there cant be any slots before USB as it is fixed on slot 1.
The patch fixes a bug that, if the USB controller happens to occupy
a slot after disks/interfaces and one of them is hot-unplugged, then
the default USB controller added on destination takes the smallest slot
number and that would lead to savestate mismatch and migration
failure. Seen and verified on PPC64.
Signed-off-by: Shivaprasad G Bhat <sbhat@linux.vnet.ibm.com>
We had both and the only difference was that the latter also included
information about multifunction setting. The problem with that was that
we couldn't use functions made for only one of the structs (e.g.
parsing). To consolidate those two structs, use the one in virpci.h,
include that in domain_conf.h and add the multifunction member in it.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Rather than take username and password as parameters, now take
a qemuDomainSecretInfoPtr and decode within the function.
NB: Having secinfo implies having the username for a plain type
from a successful virSecretGetSecretString
Signed-off-by: John Ferlan <jferlan@redhat.com>
Similar to the qemuDomainSecretDiskPrepare, generate the secret
for the Hostdev's prior to call qemuProcessLaunch which calls
qemuBuildCommandLine. Additionally, since the secret is not longer
added as part of building the command, the hotplug code will need
to make the call to add the secret in the hostdevPriv.
Since this then is the last requirement to pass a virConnectPtr
to qemuBuildCommandLine, we now can remove that as part of these
changes. That removal has cascading effects through various callers.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Modeled after the qemuDomainDiskPrivatePtr logic, create a privateData
pointer in the _virDomainHostdevDef to allow storage of private data
for a hypervisor in order to at least temporarily store auth/secrets
data for usage during qemuBuildCommandLine.
NB: Since the qemu_parse_command (qemuParseCommandLine) code is not
expecting to restore the auth/secret data, there's no need to add
code to handle this new structure there.
Updated copyrights for modules touched. Some didn't have updates in a
couple years even though changes have been made.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Rather than needing to pass the conn parameter to various command
line building API's, add qemuDomainSecretPrepare just prior to the
qemuProcessLaunch which calls qemuBuilCommandLine. The function
must be called after qemuProcessPrepareHost since it's expected
to eventually need the domain masterKey generated during the prepare
host call. Additionally, future patches may require device aliases
(assigned during the prepare domain call) in order to associate
the secret objects.
The qemuDomainSecretDestroy is called after the qemuProcessLaunch
finishes in order to clear and free memory used by the secrets
that were recently prepared, so they are not kept around in memory
too long.
Placing the setup here is beneficial for future patches which will
need the domain masterKey in order to generate an encrypted secret
along with an initialization vector to be saved and passed (since
the masterKey shouldn't be passed around).
Finally, since the secret is not added during command line build,
the hotplug code will need to get the secret into the private disk data.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Introduce a new private structure to hold qemu domain auth/secret data.
This will be stored in the qemuDomainDiskPrivate as a means to store the
auth and fetched secret data rather than generating during building of
the command line.
The initial changes will handle the current username and secret values
for rbd and iscsi disks (in their various forms). The rbd secret is
stored as a base64 encoded value, while the iscsi secret is stored as
a plain text value. Future changes will store encoded/encrypted secret
data as well as an initialization vector needed to be given to qemu
in order to decrypt the encoded password along with the domain masterKey.
The inital assumption will be that VIR_DOMAIN_SECRET_INFO_PLAIN is
being used.
Although it's expected that the cleanup of the secret data will be
done immediately after command line generation, reintroduce the object
dispose function qemuDomainDiskPrivateDispose to handle removing
memory associated with the structure for "normal" cleanup paths.
Signed-off-by: John Ferlan <jferlan@redhat.com>
After killing one of the conditionals it's now guaranteed to have
@drivealias populated when calling the monitor, so the code attempting
to cleanup can be simplified.
For strange reasons if a perf event type was not supported or failed to
be enabled at VM start libvirt would ignore the failure.
On the other hand on restart if the event could not be re-enabled
libvirt would fail to reconnect to the VM and kill it.
Both don't make really sense. Fix it by failing to start the VM if the
event is not supported and change the event to disabled if it can't be
reconnected (unlikely).
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1329045
Both disk->src->shared and disk->src->readonly can't be modified when
changing disk source for floppy and cdrom drives since both arguments
are passed as arguments of the disk rather than the image in qemu.
Historically these fields have only two possible values since they are
represented as XML thus we need to ignore if user did not provide them
and thus we are treating them as false.
If qemu doesn't support DEVICE_TRAY_MOVED event the code that attempts
to change media would attempt to re-eject the tray even if it wouldn't
be notified when the tray opened. Add a capability bit and skip retrying
for old qemus.
Empty floppy drives start with tray in "open" state and libvirt did not
refresh it after startup. The code that inserts media into the tray then
waited until the tray was open before inserting the media and thus
floppies could not be inserted.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1326660
This reverts commit 6e244c659f, which
added support to qemu for the "peer" attribute in domain interface <ip>
elements.
It's being removed temporarily for the release of libvirt 1.3.4
because the feature doesn't work, and there are concerns that it may
need to be modified in an externally visible manner which could create
backward compatibility problems.
Conflicts:
tests/qemuxml2argvmock.c - a mock of virNetDevSetOnline() was added
which may be assumed by other tests added since the original commit,
so it isn't being reverted.
Similarly to what commit 7140807917 did with some internal paths,
clear vnc socket paths that were generated by us. Having such path in
the definition can cause trouble when restoring the domain. The path is
generated to the per-domain directory that contains the domain ID.
However, that ID will be different upon restoration, so qemu won't be
able to create that socket because the directory will not be prepared.
To be able to migrate to older libvirt, skip formatting the socket path
in migratable XML if it was autogenerated. And mark it as autogenerated
if it already exists and we're parsing live XML.
Best viewed with '-C'.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1326270
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
When the domain definition describes a machine with NUMA, setting the
maximum vCPU count via the API might lead to an invalid config.
Add a check that will forbid this until we add more advanced cpu config
capabilities.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1327499
Instead of setting the default qemu stdio logging approach in
virQEMUDriverConfigLoadFile set it in virQEMUDriverConfigNew so that
it's properly set even when the config is not present.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1325075
If the domain name is long enough, the timestamp can prolong the
filename for automatic coredump to more than the filesystem's limit.
Simply shorten it like we do in other places. The timestamp helps with
the unification, but having the ID in the name won't hurt.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1289363
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Add virDomainObjGetShortName() and use it. For now that's used in one
place, but we should expose it so that future patches can use it.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
This function - in contrast with qemuBuildCommandLine - merely
constructs our internal command representation of a domain. This
is then later compared against expected output. Or, this function
is used also in virConnectDomainXMLToNative(). But due to a copy
paste error this function, just like its image - has @forceFips
argument that if enabled forces FIPS, otherwise mimics FIPS state
in the host. If FIPS is enabled or forced the generated command
line is different to state in which FIPS is disabled. Problem is,
while this could be desired in the virConnectDomainXMLToNative()
case, this is undesirable in the test suite as it will produce
unpredicted results.
Solution to this is to rename argument to @enableFips to
specifically tell whether we expect command line to be build in
either of fashions and make virConnectDomainXMLToNative()
implementation fetch FIPS state and pass it to
qemuProcessCreatePretendCmd().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
This error message was too specific, based on the incorrect assumption
that any error was cause by auto-added bridges:
failed to create PCI bridge on bus 2: too many devices
with fixed addresses
In practice you can't know if a bridge with an index <= the bus it's
connecting to was added automatically, or if it was a mistake in
explicit config, and the auto-add problem is going to be dealt with in
a different way in an upcoming patch. The new message is this:
PCI Controller at index 1 (0x01) has "
bus='0x02', but bus must be <= index
(note that index is given in both decimal and hex because it is
formatted as decimal in the XML, but bus is formatted as hex, and
displaying the hex value of index makes it easier to see the problem
when index > 9 (which will often be the case with PCIe, since most
controllers only have a single port, not 32 slots as with standard
PCI)).
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1004593
The values are currently limited to LLONG_MAX which causes some
problems. QEMU conveniently changed their maximum to 1e15 (1 PB) which
is enough for some time and we need to adapt to that so that we don't
throw "Unknown error" messages. Strictly limiting these values actually
fixes some corner case values (off-by-one checks in QEMU probably).
Since values out of the new specified range do not overflow anything,
change the type of error as well.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1317531
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
This an ubuntu/debian packaging convention. At one point it may have
been an actually different binary, but at least as of ubuntu precise
(the oldest supported ubuntu distro, released april 2012) kvm-img is
just a symlink to qemu-img for back compat.
I think it's safe to drop support for it
QEMU introduced the query-gic-capabilities QMP command
with commit 4468d4e0f383: use the command, if available,
to probe available GIC capabilities.
The information obtained is stored in a virQEMUCaps
instance, and will be later used to fill in a
virDomainCaps instance.
The struct contains a single boolean field, 'supported':
the meaning of this field is too generic to be limited to
devices only, and in fact it's already being used for
other things like loaders and OSs.
Instead of trying to come up with a more generic name just
get rid of the struct altogether.
The common idiom in the driver API implementations is roughly:
- ACL check
- BeginJob (if needed)
- AgentAvailable (if needed)
- !IsActive
A few calls had an extra !IsActive before BeginJob, which doesn't
seem to serve much use. Drop them
Migration API allows to specify a destination domain configuration.
Offline domain has only inactive XML and it is replaced by configuration
specified using VIR_MIGRATE_PARAM_DEST_XML param. In case of live
migration VIR_MIGRATE_PARAM_DEST_XML param is applied for active XML.
This commit introduces the new VIR_MIGRATE_PARAM_PERSIST_XML param
that can be used within live migration to replace persistent/inactive
configuration.
Required for: https://bugzilla.redhat.com/show_bug.cgi?id=835300
Commit id '4b75237f' seems to have triggered Coverity into finding
at least one memory leak in xen_xl.c for error path for cleanup where
the listenAddr would be leaked. Reviewing other callers, it seems that
qemu_parse_command.c would have the same issue, so just it too.
When creating the master key, we used mode 0600 (which we should) but
because we were creating it as root, the file is not readable by any
qemu running as non-root. Fortunately, it's just a matter of labelling
the file. We are generating the file path few times already, so let's
label it in the same function that has access to the path already.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
In a few places in libvirt we busy-wait for events, for example qemu
creating a monitor socket. This is problematic because:
- We need to choose a sufficiently small polling period so that
libvirt doesn't add unnecessary delays.
- We need to choose a sufficiently large polling period so that
the effect of busy-waiting doesn't affect the system.
The solution to this conflict is to use an exponential backoff.
This patch adds two functions to hide the details, and modifies a few
places where we currently busy-wait.
Signed-off-by: Richard W.M. Jones <rjones@redhat.com>
Rather than trying some magic calculations on our side query the monitor
for the current size of the memory balloon both on hotplug and
hotunplug.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1220702
Now that there is just one format of the memory balloon command line
used the code can be merged into a single function.
Additionally with some tweaks to the control flow the code is easier to
read.
The change that made qemu not add the memballoon by default happened
prior to 0.12.0. Additionaly the comment was misleading due to the code
that was added below. Since we always need to add a balloon on the
commandline drop the comment.
The only caller of this code is:
for (i = 0; i < dom->def->ngraphics; i++) {
if (dom->def->graphics[i]->type == VIR_DOMAIN_GRAPHICS_TYPE_SPICE) {
if (!(mig->graphics =
qemuMigrationCookieGraphicsAlloc(driver, dom->def->graphics[i])))
return -1;
mig->flags |= QEMU_MIGRATION_COOKIE_GRAPHICS;
break;
}
}
So this is never triggered for VNC, and in fact VNC has no support for
seamless migration anyways so that seems correct. Drop the dead VNC
handling.
This is backed by the qemu device pxb-pcie, which will be available in
qemu 2.6.0.
As with pci-expander-bus (which uses qemu's pxb device), the busNr
attribute and <node> subelement of <target> are used to set the bus_nr
and numa_node options.
During post-parse we validate that the domain's machinetype is
q35-based (since the device shows up for 440fx-based machinetypes, but
is unusable), as well as checking that <node> specifies a node that is
actually configured on the guest.
This controller provides a single PCIe port on a new root. It is
similar to pci-expander-bus, intended to provide a bus that can be
associated with a guest-identifiable NUMA node, but is for
machinetypes with PCIe rather than PCI (e.g. q35-based machinetypes).
Aside from PCIe vs. PCI, the other main difference is that a
pci-expander-bus has a companion pci-bridge that is automatically
attached along with it, but pcie-expander-bus has only a single port,
and that port will only connect to a pcie-root-port, or to a
pcie-switch-upstream-port. In order for the bus to be of any use in
the guest, it must have either a pcie-root-port or a
pcie-switch-upstream-port attached (and one or more
pcie-switch-downstream-ports attached to the
pcie-switch-upstream-port).
The pxb device is a PCIe expander bus that can be added to any
Q35-based machinetype. A single PCIe port (*not* hotpluggable) is
provided; if more than one device is desired, or if hotplug
support is needed, either a pcie-root-port, or some combination of
pcie-switch-upstream-port and pcie-swith-downstream-ports must be
added to it. It can have a NUMA node number associated with it, as
well as a bus number.
This is backed by the qemu device "pxb".
The pxb device always includes a pci-bridge that is at the bus number
of the pxb + 1.
busNr and <node> from the <target> subelement are used to set the
bus_nr and numa_node options for pxb.
During post-parse we validate that the domain's machinetype is
440fx-based (since the pxb device only works on 440fx-based machines),
and <node> also gets a sanity check to assure that the NUMA node
specified for the pxb (if any - it's optional) actually exists on the
guest.
This is a standard PCI root bus (not a bridge) that can be added to a
440fx-based domain. Although it uses a PCI slot, this is *not* how it
is connected into the PCI bus hierarchy, but is only used for
control. Each pci-expander-bus provides 32 slots (0-31) that can
accept hotplug of standard PCI devices.
The usefulness of pci-expander-bus relative to a pci-bridge is that
the NUMA node of the bus can be specified with the <node> subelement
of <target>. This gives guest-side visibility to the NUMA node of
attached devices (presuming that management apps only assign a device
to a bus that has a NUMA node number matching the node number of the
device on the host).
Each pci-expander-bus also has a "busNr" attribute. The expander-bus
itself will take the busNr specified, and all buses that are connected
to this bus (including the pci-bridge that is automatically added to
any expander bus of model "pxb" (see the next commit)) will use
busNr+1, busNr+2, etc, and the pci-root (or the expander-bus with next
lower busNr) will use bus numbers lower than busNr.
The pxb device is a PCI expander bus that can be added to any
440fx-based machinetype. The PCI bus that is created has 32 standard
PCI slots (hotpluggable). It can have a NUMA node number associated
with it, as well as a bus number.
There are two places in qemu_domain_address.c where we have a switch
statement to convert PCI controller models
(VIR_DOMAIN_CONTROLLER_MODEL_PCI*) into the connection type flag that
is matched when looking for an upstream connection for that model of
controller (VIR_PCI_CONNECT_TYPE_*). This patch makes a utility
function in conf/domain_addr.c to do that, so that when a new PCI
controller is added, we only need to add the new model-->connect-type
in a single place.
The flags used to determine which devices could be plugged into which
controllers were quite confusing, as they tried to create classes of
connections, then put particular devices into possibly multiple
classes, while sometimes setting multiple flags for the controllers
themselves. The attempt to have a single flag indicate, e.g. that a
root-port or a switch-downstream-port could connect was not only
confusing, it was leading to a situation where it would be impossible
to specify exactly the right combinations for a new controller.
The solution is for the VIR_PCI_CONNECT_TYPE_* flags to have a 1:1
correspondence with each type of PCI controller, plus a flag for a PCI
endpoint device and another for a PCIe endpoint device (the only
exception to this is that pci-bridge and pcie-expander-bus controllers
have their upstream connection classified as
VIR_PCI_CONNECT_TYPE_PCI_DEVICE since they can be plugged into
*exactly* the same ports as any endpoint device). Each device then
has a single flag for connect type (plus the HOTPLUG flag if that
device can e hotplugged), and each controller sets the CONNECT bits
for all controllers that can be plugged into it, as well as for either
type of endpoint device that can be plugged in (and the HOTPLUG flag
if it can accept hotplugged devices).
With this change, it is *slightly* easier to understand the matching
of connections (as long as you remember that the flag for a
device/upstream-facing connection of a controller is the same as that
device's type, while the flags for a controller's downstream
connections is the OR of all device types that can be plugged into
that controller). More importantly, it will be possible to correctly
specify what can be plugged into a pcie-switch-expander-bus, when
support for it is added.
The watchdog cli refactoring in 4666b762 dropped the temporary variable
we use to convert to action=dump to action=pause for the qemu cli, and
stored the converted value in the domain structure. Our other watchdog
handling code then treated it as though the user requested action=pause,
which broke action=dump handling.
Revive the temporary variable to fix things.
I tried compiling libvirt with older gcc and probably because I used
different configure options I got some shadowed declarations.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
GCC in RHEL-6 complains about listen:
../../src/conf/domain_conf.c:23718: error: declaration of 'listen' shadows a global declaration [-Wshadow]
/usr/include/sys/socket.h:204: error: shadowed declaration is here [-Wshadow]
This renames all the listen to gListen.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Since qemu is now able to notify us that the guest rejected the memory
unplug operation we can relay this to the user and make the API fail
right away.
Additionally document the possible values from the ACPI docs for future
reference.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1320447
The event is emitted on ACPI OSPM Status Indication events.
ACPI standard documentation describes the method as:
This object is an optional control method that is invoked by OSPM to
indicate processing status to the platform. During device ejection,
device hot add, or other event processing, OSPM may need to perform
specific handshaking with the platform. OSPM may also need to indicate
to the platform its inability to complete a requested operation; for
example, when a user presses an ejection button for a device that is
currently in use or is otherwise currently incapable of being ejected.
In this case, the processing of the ACPI Eject Request notification by
OSPM fails. OSPM may indicate this failure to the platform through the
invocation of the _OST control method. As a result of the status
notification indicating ejection failure, the platform may take certain
action including reissuing the notification or perhaps turning on an
appropriate indicator light to signal the failure to the user.
Similarly to the DEVICE_DELETED event we will be able to tell when
unplug of certain device types will be rejected by the guest OS. Wire up
the device deletion signalling code to allow handling this.
Neither of the callers cares whether the DEVICE_DELETED event isn't
supported or the event was received. Simplify the code and callers by
unifying the two values and changing the return value constants so that
a temporary variable can be omitted.
Callers ignore if this function returns -1 and continue as though the
DEVICE_DELETED event was not received. Since we can't be sure that the
event was not received we should behave as if the event was not
supported and remove the device definition right away. The error
fortunately won't really happen here.
The address assigning code might add new pci bridges.
We need them to have an alias when building the command line.
In real word usage, this is not a problem because all the code
paths already call qemuDomainAssignAddresses. However moving
this call lets us remove one extra call from qemuxml2argvtest.
Essentially revert commit 3a6204c which added these to allow the test
suite to pass without depending on the host system state.
Since commit 4b527c1 we already mock virSCSIDeviceGetSgName, so these
callbacks are useless.
This effectively removes virDomainGraphicsListenSetAddress which was
used only to change the address of listen structure and possible change
the listen type. The new function will auto-expand the listens array
and append a new listen.
The old function was used on pre-allocated array of listens and in most
cases it only "add" a new listen. The two remaining uses can access the
listen structure directly.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Replace the nonsensical debug statement by adding the expected event
code into the existing debug statement.
Since the monitor code always notifies the agent on guest
reboot/shutdown even if that was not initiated by the agent the warning
emitted later is bogus and pollutes the logs in such cases. Delete it
and keep just the original debug message where this info can be
inferred.
Move including of gnutls/gnutls.h in qemu/qemu_domain.c under the
"ifdef WITH_GNUTLS" check because otherwise it fails like this:
CC qemu/libvirt_driver_qemu_impl_la-qemu_domain.lo
qemu/qemu_domain.c:50:10: fatal error: 'gnutls/gnutls.h' file not found
in case if gnutls is not installed on the system.
Some places already check for "virt-" prefix as well as plain "virt".
virQEMUCapsHasPCIMultiBus did not, resulting in multiple PCI devices
having assigned the same unnumbered "pci" alias.
Add a test for the "virt-2.6" machine type which also omits the
<model type='virtio'/> in <interface>, to check if
qemuDomainDefaultNetModel works too.
https://bugzilla.redhat.com/show_bug.cgi?id=1325085
Our use of gnutls_rnd(), introduced with commit ad7520e8, is
conditional to the availability of the <gnutls/crypto.h> header
file.
Such check, however, turns out not to be strict enough, as there
are some versions of GnuTLS (eg. 2.8.5 from CentOS 6) that provide
the header file, but not the function itself, which was introduced
only in GnuTLS 2.12.0.
Introduce an explicit check for the function.
The new perf code didn't bother to clear a pointer in 'priv' causing a
double free or other memory corruption goodness if a VM failed to start.
Clear the pointer after freeing the memory.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1324757
For device hotplug, the new alias ID needs to be checked in the list
rather than using the count of devices. Unplugging a device that is not
last in the array will make further hotplug impossible due to alias
collision.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1324551
For device hotplug, the new alias ID needs to be checked in the list
rather than using the count of devices. Unplugging a device that is not
last in the array will make further hotplug impossible due to alias
collision.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1324551
Commit id 'fb2bd208' essentially copied the qemuGetSecretString
creating an libxlGetSecretString. Rather than have multiple copies
of the same code, create src/secret/secret_util.{c,h} files and
place the common function in there.
Modify the the build in order to build the module as a library
which is then pulled in by both the qemu and libxl drivers for
usage from both qemu_command.c and libxl_conf.c
If the -object secret capability exists, then get the path to the
masterKey file and provide that to qemu. Checking for the existence
of the file before passing to qemu could be done, but causes issues
in mock test environment.
Since the qemuDomainObjPrivate is not available when building the
command line, the qemuBuildHasMasterKey API will have to suffice
as the primary arbiter for whether the capability exists in order
to find/return the path to the master key for usage.
Created the qemuDomainGetMasterKeyAlias API which will be used by
later patches to define the 'keyid' (eg, masterKey) to be used by
other secrets to provide the id to qemu for the master key.
Add a masterKey and masterKeyLen to _qemuDomainObjPrivate to store a
random domain master key and its length in order to support the ability
to encrypt/decrypt sensitive data shared between libvirt and qemu. The
key will be base64 encoded and written to a file to be used by the
command line building code to share with qemu.
New API's from this patch:
qemuDomainGetMasterKeyFilePath:
Return a path to where the key is located
qemuDomainWriteMasterKeyFile: (private)
Open (create/trunc) the masterKey path and write the masterKey
qemuDomainMasterKeyReadFile:
Using the master key path, open/read the file, and store the
masterKey and masterKeyLen. Expected use only from qemuProcessReconnect
qemuDomainGenerateRandomKey: (private)
Generate a random key using available algorithms
The key is generated either from the gnutls_rnd function if it
exists or a less cryptographically strong mechanism using
virGenerateRandomBytes
qemuDomainMasterKeyRemove:
Remove traces of the master key, remove the *KeyFilePath
qemuDomainMasterKeyCreate:
Generate the domain master key and save the key in the location
returned by qemuDomainGetMasterKeyFilePath.
This API will first ensure the QEMU_CAPS_OBJECT_SECRET is set
in the capabilities. If not, then there's no need to generate
the secret or file.
The creation of the key will be attempted from qemuProcessPrepareHost
once the libDir directory structure exists.
The removal of the key will handled from qemuProcessStop just prior
to deleting the libDir tree.
Since the key will not be written out to the domain object XML file,
the qemuProcessReconnect will read the saved file and restore the
masterKey and masterKeyLen.
Add a capability bit for the qemu secret object.
Adjust the 2.6.0-1 caps/replies to add the secret object. For the
.replies it's take from the '{"execute":"qom-list-types"}' output.
When a hostdev is attached to the guest (and removed from the host),
the order of operations is call qemuHostdevPreparePCIDevices to remove
the device from the host, call qemuSetupHostdevCgroup to setup the cgroups,
and virSecurityManagerSetHostdevLabel to set the labels.
When the device is removed from the guest, the code didn't use the
reverse order leading to possible issues (especially if the path to
the device no longer exists). This patch will move the call to
qemuTeardownHostdevCgroup to prior to reattaching the device to
the host.
When a hostdev is attached to the guest (and removed from the host),
the order of operations is call qemuHostdevPreparePCIDevices to remove
the device from the host, call qemuSetupHostdevCgroup to setup the cgroups,
and virSecurityManagerSetHostdevLabel to set the labels.
When the device is removed from the guest, the code didn't use the
reverse order leading to possible issues (especially if the path to
the device no longer exists). This patch will move the call to
virSecurityManagerRestoreHostdevLabel to prior to reattaching the
device to the host.
Starting with commit f8e712fe, if you start a domain that has an
<interface type='hostdev' (or that has <interface type='network'>
where the network is a pool of devices for hostdev assignment), when
you later try to add *another* interface (of any kind) with hotplug,
the function qemuAssignDeviceNetAlias() fails as soon as it sees a
"hostdevN" alias in the list of interfaces), causing the attach to
fail.
This is because (starting with f8e712fe) the device alias names are
assigned during the new function qemuProcessPrepareDomain(), which is
called *before* networkAllocateActualDevice() (which is called from
qemuProcessPrepareHost(), which is called from
qemuProcessLaunch()). Prior to that commit,
networkAllocateActualDevice() was called first.
The problem with this is that the alias for interfaces that are really
a hostdev (<interface type='hostdev'>) is of the form "hostdevN" (just
like other hostdevs), while other interfaces are "netN". But if you
don't know that the interface is going to be a hostdev at the time you
assign the alias name, you can't name it differently. (As far as I've
seen so far, the change in name by itself wouldn't have been a problem
(other than just an outwardly noticeable change in behavior) except
for the abovementioned failure to attach/detach new interfaces.
Rather than take the chance that there may be other not-yet-revealed
problems associated with changing the alias name, this patch changes
the way that aliases are assigned to restore the old behavior.
Old: In the past, assigning an alias to an interface was skipped if it
was seen that the interface was type='hostdev' - we knew that the
hostdev part of the interface was also in the list of hostdevs (that's
part of what happens in networkAllocateActualDevice()) and it would be
assigned when all the other hostdev aliases were assigned.
New: When assigning an alias to an interface, we haven't yet called
networkAllocateActualDevice() to construct the hostdev part of the
interface, so we can't just wait for the loop that creates aliases for
all the hostdevs (there's nothing on that list for this device
yet!). Instead we handle it immediately in the loop creating interface
aliases, by calling the new function networkGetActualType() to
determine if it is going to be hostdev, and if so calling
qemuAssignDeviceHostdevAlias() instead.
Some adjustments have to be made to both
qemuAssignDeviceHostdevAlias() and to qemuAssignDeviceNetAlias() to
accommodate this. In both of them, an error return from
qemuDomainDeviceAliasIndex() is no longer considered an error; instead
it's just ignored (because it almost certainly means that the alias
string for the device was "net" when we expected "hostdev" or vice
versa). in qemuAssignDeviceHostdevAlias() we have to look at all
interface aliases for hostdevN in addition to looking at all hostdev
aliases (this wasn't necessary in the past, because both the interface
entry and the hostdev entry for the device already pointed at the
device info; no longer the case since the hostdev entry hasn't yet
been setup).
Fortunately the buggy behavior hasn't yet been in any official release
of libvirt.
In certain cases, we need to assign a hostdevN-style alias in a case
when we don't have a virDomainHostdevDefPtr (instead we have a
virDomainNetDefPtr). Since qemuAssignDeviceHostdevAlias() doesn't use
anything in the virDomainHostdevDef except the alias string itself
anyway, this patch just changes the arguments to pass a pointer to the
alias pointer instead.
The paths have the domain ID in them. Without cleaning them, they would
contain the same ID even after multiple restarts. That could cause
various problems, e.g. with access.
Add function qemuDomainClearPrivatePaths() for this as a counterpart of
qemuDomainSetPrivatePaths().
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Commit 08cc14f moved the conversion of MiB/s to B/s out of the
qemuMonitor APIs, but forgot to adjust the qemuMigrationDriveMirror
caller.
This patch will convert the migrate_speed value from MiB/s to its
mirror_speed equivalent in bytes/s.
Signed-off-by: Rudy Zhang <rudyflyzhang@gmail.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
@flags have a valid modification impact only after calling
virDomainObjUpdateModificationImpact. virDomainObjGetOneDef calls it but
doesn't update them in the caller.
We use _LAST items in enums to mark the last position in given
enum. Now, if and enum is passed to switch(), compiler checks
that all the values from enum occur in 'case' enumeration.
Including _LAST. But coverity spots it's a dead code. And it
really is. So to resolve this, we tend to put a comment just
above 'case ..._LAST' notifying coverity that we know this is a
dead code but we want to have it that way.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Even though we have the machine locked throughout whole APIs we
are querying/modifying domain internal state. We should grab a
job whilst doing that.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Now that we have @flags we can support changing perf events just
in active or inactive configuration regardless of the other.
Previously, calling virDomainSetPerfEvents set events in both
active and inactive configuration at once. Even though we allow
users to set perf events that are to be enabled once domain is
started up. The virDomainGetPerfEvents API was flawed too. It
returned just runtime info.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
I've noticed that these APIs are missing @flags argument. Even
though we don't have a use for them, it's our policy that every
new API must have @flags.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
They recently were extracted to a separate function. They don't belong
together though. Since -numa formatting is pretty compact, move it to
the main function and rename qemuBuildNumaCommandLine to
qemuBuildMemoryDeviceCommandLine.
When starting up a VM libvirtd asks numad to place the VM in case of
automatic nodeset. The nodeset would not be passed to the memory device
formatter and the user would get an error.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1269715
Commit 7068b56c introduced several hyperv features. Not all hyperv
features are supported by old enough kernels and we shouldn't allow to
start a guest if kernel doesn't support any of the hyperv feature.
There is one exception, for backward compatibility we cannot error out
if one of the RELAXED, VAPIC or SPINLOCKS isn't supported, for the same
reason we ignore invtsc, to not break restoring saved domains with older
libvirt.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
This check is there to allow restore saved domain with older libvirt
where we included invtsc by default for host-passthrough model. Don't
skip the whole function, but only the part that checks for invtsc.
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Since commit v1.3.2-119-g1e34a8f which enabled debug-threads in QEMU
qemuGetProcessInfo would fail to parse stats for any thread with a space
in its name.
https://bugzilla.redhat.com/show_bug.cgi?id=1316803
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
