The virConnectPtr is passed around loads of nwfilter code in
order to provide it as a parameter to the callback registered
by the virt drivers. None of the virt drivers use this param
though, so it serves no purpose.
Avoiding the need to pass a virConnectPtr means that the
nwfilterStateReload method no longer needs to open a bogus
QEMU driver connection. This addresses a race condition that
can lead to a crash on startup.
The nwfilter driver starts before the QEMU driver and registers
some callbacks with DBus to detect firewalld reload. If the
firewalld reload happens while the QEMU driver is still starting
up though, the nwfilterStateReload method will open a connection
to the partially initialized QEMU driver and cause a crash.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
(cherry picked from commit 999d72fbd59ea712128ae294b69b6a54039d757b)
https://bugzilla.redhat.com/show_bug.cgi?id=1058839
Commit f9f56340 for CVE-2014-0028 almost had the right idea - we
need to check the ACL rules to filter which events to send. But
it overlooked one thing: the event dispatch queue is running in
the main loop thread, and therefore does not normally have a
current virIdentityPtr. But filter checks can be based on current
identity, so when libvirtd.conf contains access_drivers=["polkit"],
we ended up rejecting access for EVERY event due to failure to
look up the current identity, even if it should have been allowed.
Furthermore, even for events that are triggered by API calls, it
is important to remember that the point of events is that they can
be copied across multiple connections, which may have separate
identities and permissions. So even if events were dispatched
from a context where we have an identity, we must change to the
correct identity of the connection that will be receiving the
event, rather than basing a decision on the context that triggered
the event, when deciding whether to filter an event to a
particular connection.
If there were an easy way to get from virConnectPtr to the
appropriate virIdentityPtr, then object_event.c could adjust the
identity prior to checking whether to dispatch an event. But
setting up that back-reference is a bit invasive. Instead, it
is easier to delay the filtering check until lower down the
stack, at the point where we have direct access to the RPC
client object that owns an identity. As such, this patch ends
up reverting a large portion of the framework of commit f9f56340.
We also have to teach 'make check' to special-case the fact that
the event registration filtering is done at the point of dispatch,
rather than the point of registration. Note that even though we
don't actually use virConnectDomainEventRegisterCheckACL (because
the RegisterAny variant is sufficient), we still generate the
function for the purposes of documenting that the filtering
takes place.
Also note that I did not entirely delete the notion of a filter
from object_event.c; I still plan on using that for my upcoming
patch series for qemu monitor events in libvirt-qemu.so. In
other words, while this patch changes ACL filtering to live in
remote.c and therefore we have no current client of the filtering
in object_event.c, the notion of filtering in object_event.c is
still useful down the road.
* src/check-aclrules.pl: Exempt event registration from having to
pass checkACL filter down call stack.
* daemon/remote.c (remoteRelayDomainEventCheckACL)
(remoteRelayNetworkEventCheckACL): New functions.
(remoteRelay*Event*): Use new functions.
* src/conf/domain_event.h (virDomainEventStateRegister)
(virDomainEventStateRegisterID): Drop unused parameter.
* src/conf/network_event.h (virNetworkEventStateRegisterID):
Likewise.
* src/conf/domain_event.c (virDomainEventFilter): Delete unused
function.
* src/conf/network_event.c (virNetworkEventFilter): Likewise.
* src/libxl/libxl_driver.c: Adjust caller.
* src/lxc/lxc_driver.c: Likewise.
* src/network/bridge_driver.c: Likewise.
* src/qemu/qemu_driver.c: Likewise.
* src/remote/remote_driver.c: Likewise.
* src/test/test_driver.c: Likewise.
* src/uml/uml_driver.c: Likewise.
* src/vbox/vbox_tmpl.c: Likewise.
* src/xen/xen_driver.c: Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
(cherry picked from commit 11f20e43f1388d5f8f8c0bfac8c9cda6160a106b)
Conflicts:
daemon/remote.c - not backporting network events
src/conf/network_event.c - likewise
src/conf/network_event.h - likewise
src/network/bridge_driver.c - likewise
src/conf/domain_event.c - revert back to pre-CVE state
src/conf/domain_event.h - likewise
src/libxl/libxl_driver.c - likewise
src/lxc/lxc_driver.c - likewise
src/remote/remote_driver.c - likewise
src/test/test_driver.c - likewise
src/uml/uml_driver.c - likewise
src/xen/xen_driver.c - likewise
Ever since ACL filtering was added in commit 7639736 (v1.1.1), a
user could still use event registration to obtain access to a
domain that they could not normally access via virDomainLookup*
or virConnectListAllDomains and friends. We already have the
framework in the RPC generator for creating the filter, and
previous cleanup patches got us to the point that we can now
wire the filter through the entire object event stack.
Furthermore, whether or not domain:getattr is honored, use of
global events is a form of obtaining a list of networks, which
is covered by connect:search_domains added in a93cd08 (v1.1.0).
Ideally, we'd have a way to enforce connect:search_domains when
doing global registrations while omitting that check on a
per-domain registration. But this patch just unconditionally
requires connect:search_domains, even when no list could be
obtained, based on the following observations:
1. Administrators are unlikely to grant domain:getattr for one
or all domains while still denying connect:search_domains - a
user that is able to manage domains will want to be able to
manage them efficiently, but efficient management includes being
able to list the domains they can access. The idea of denying
connect:search_domains while still granting access to individual
domains is therefore not adding any real security, but just
serves as a layer of obscurity to annoy the end user.
2. In the current implementation, domain events are filtered
on the client; the server has no idea if a domain filter was
requested, and must therefore assume that all domain event
requests are global. Even if we fix the RPC protocol to
allow for server-side filtering for newer client/server combos,
making the connect:serach_domains ACL check conditional on
whether the domain argument was NULL won't benefit older clients.
Therefore, we choose to document that connect:search_domains
is a pre-requisite to any domain event management.
Network events need the same treatment, with the obvious
change of using connect:search_networks and network:getattr.
* src/access/viraccessperm.h
(VIR_ACCESS_PERM_CONNECT_SEARCH_DOMAINS)
(VIR_ACCESS_PERM_CONNECT_SEARCH_NETWORKS): Document additional
effect of the permission.
* src/conf/domain_event.h (virDomainEventStateRegister)
(virDomainEventStateRegisterID): Add new parameter.
* src/conf/network_event.h (virNetworkEventStateRegisterID):
Likewise.
* src/conf/object_event_private.h (virObjectEventStateRegisterID):
Likewise.
* src/conf/object_event.c (_virObjectEventCallback): Track a filter.
(virObjectEventDispatchMatchCallback): Use filter.
(virObjectEventCallbackListAddID): Register filter.
* src/conf/domain_event.c (virDomainEventFilter): New function.
(virDomainEventStateRegister, virDomainEventStateRegisterID):
Adjust callers.
* src/conf/network_event.c (virNetworkEventFilter): New function.
(virNetworkEventStateRegisterID): Adjust caller.
* src/remote/remote_protocol.x
(REMOTE_PROC_CONNECT_DOMAIN_EVENT_REGISTER)
(REMOTE_PROC_CONNECT_DOMAIN_EVENT_REGISTER_ANY)
(REMOTE_PROC_CONNECT_NETWORK_EVENT_REGISTER_ANY): Generate a
filter, and require connect:search_domains instead of weaker
connect:read.
* src/test/test_driver.c (testConnectDomainEventRegister)
(testConnectDomainEventRegisterAny)
(testConnectNetworkEventRegisterAny): Update callers.
* src/remote/remote_driver.c (remoteConnectDomainEventRegister)
(remoteConnectDomainEventRegisterAny): Likewise.
* src/xen/xen_driver.c (xenUnifiedConnectDomainEventRegister)
(xenUnifiedConnectDomainEventRegisterAny): Likewise.
* src/vbox/vbox_tmpl.c (vboxDomainGetXMLDesc): Likewise.
* src/libxl/libxl_driver.c (libxlConnectDomainEventRegister)
(libxlConnectDomainEventRegisterAny): Likewise.
* src/qemu/qemu_driver.c (qemuConnectDomainEventRegister)
(qemuConnectDomainEventRegisterAny): Likewise.
* src/uml/uml_driver.c (umlConnectDomainEventRegister)
(umlConnectDomainEventRegisterAny): Likewise.
* src/network/bridge_driver.c
(networkConnectNetworkEventRegisterAny): Likewise.
* src/lxc/lxc_driver.c (lxcConnectDomainEventRegister)
(lxcConnectDomainEventRegisterAny): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
(cherry picked from commit f9f56340539d609cdc2e9d4ab812b9f146c3f100)
Conflicts:
src/conf/object_event.c - not backporting event refactoring
src/conf/object_event_private.h - likewise
src/conf/network_event.c - not backporting network events
src/conf/network_event.h - likewise
src/network/bridge_driver.c - likewise
src/access/viraccessperm.h - likewise
src/remote/remote_protocol.x - likewise
src/conf/domain_event.c - includes code that upstream has in object_event
src/conf/domain_event.h - context
src/libxl/libxl_driver.c - context
src/lxc/lxc_driver.c - context
src/remote/remote_driver.c - context, not backporting network events
src/test/test_driver.c - context, not backporting network events
src/uml/uml_driver.c - context
src/xen/xen_driver.c - context
CVE-2013-6458
Every API that is going to begin a job should do that before fetching
data from vm->def.
(cherry picked from commit 3b56425938e2f97208d5918263efa0d6439e4ecd)
Every API that is going to begin a job should do that before fetching
data from vm->def.
(cherry picked from commit ff5f30b6bfa317f2a4c33f69289baf4e887eb048)
CVE-2013-6458
Every API that is going to begin a job should do that before fetching
data from vm->def.
(cherry picked from commit f93d2caa070f6197ab50d372d286018b0ba6bbd8)
CVE-2013-6458
Generally, every API that is going to begin a job should do that before
fetching data from vm->def. However, qemuDomainGetBlockInfo does not
know whether it will have to start a job or not before checking vm->def.
To avoid using disk alias that might have been freed while we were
waiting for a job, we use its copy. In case the disk was removed in the
meantime, we will fail with "cannot find statistics for device '...'"
error message.
(cherry picked from commit b799259583bd65c0b2f5042e6c3ff19637ade881)
CVE-2013-6458
https://bugzilla.redhat.com/show_bug.cgi?id=1043069
When virDomainDetachDeviceFlags is called concurrently to
virDomainBlockStats: libvirtd may crash because qemuDomainBlockStats
finds a disk in vm->def before getting a job on a domain and uses the
disk pointer after getting the job. However, the domain in unlocked
while waiting on a job condition and thus data behind the disk pointer
may disappear. This happens when thread 1 runs
virDomainDetachDeviceFlags and enters monitor to actually remove the
disk. Then another thread starts running virDomainBlockStats, finds the
disk in vm->def, and while it's waiting on the job condition (owned by
the first thread), the first thread finishes the disk removal. When the
second thread gets the job, the memory pointed to be the disk pointer is
already gone.
That said, every API that is going to begin a job should do that before
fetching data from vm->def.
(cherry picked from commit db86da5ca2109e4006c286a09b6c75bfe10676ad)
Since 86d90b3a (yes, my patch; again) we are supporting NBD storage
migration. However, on error recovery path we got the steps reversed.
The correct order is: return NBD port to the virPortAllocator and then
either unlock the vm or remove it from the driver. Not vice versa.
==11192== Invalid write of size 4
==11192== at 0x11488559: qemuMigrationPrepareAny (qemu_migration.c:2459)
==11192== by 0x11488EA6: qemuMigrationPrepareDirect (qemu_migration.c:2652)
==11192== by 0x114D1509: qemuDomainMigratePrepare3Params (qemu_driver.c:10332)
==11192== by 0x519075D: virDomainMigratePrepare3Params (libvirt.c:7290)
==11192== by 0x1502DA: remoteDispatchDomainMigratePrepare3Params (remote.c:4798)
==11192== by 0x12DECA: remoteDispatchDomainMigratePrepare3ParamsHelper (remote_dispatch.h:5741)
==11192== by 0x5212127: virNetServerProgramDispatchCall (virnetserverprogram.c:435)
==11192== by 0x5211C86: virNetServerProgramDispatch (virnetserverprogram.c:305)
==11192== by 0x520A8FD: virNetServerProcessMsg (virnetserver.c:165)
==11192== by 0x520A9E1: virNetServerHandleJob (virnetserver.c:186)
==11192== by 0x50DA78F: virThreadPoolWorker (virthreadpool.c:144)
==11192== by 0x50DA11C: virThreadHelper (virthreadpthread.c:161)
==11192== Address 0x1368baa0 is 576 bytes inside a block of size 688 free'd
==11192== at 0x4A07F5C: free (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==11192== by 0x5079A2F: virFree (viralloc.c:580)
==11192== by 0x11456C34: qemuDomainObjPrivateFree (qemu_domain.c:267)
==11192== by 0x50F41B4: virDomainObjDispose (domain_conf.c:2034)
==11192== by 0x50C2991: virObjectUnref (virobject.c:262)
==11192== by 0x50F4CFC: virDomainObjListRemove (domain_conf.c:2361)
==11192== by 0x1145C125: qemuDomainRemoveInactive (qemu_domain.c:2087)
==11192== by 0x11488520: qemuMigrationPrepareAny (qemu_migration.c:2456)
==11192== by 0x11488EA6: qemuMigrationPrepareDirect (qemu_migration.c:2652)
==11192== by 0x114D1509: qemuDomainMigratePrepare3Params (qemu_driver.c:10332)
==11192== by 0x519075D: virDomainMigratePrepare3Params (libvirt.c:7290)
==11192== by 0x1502DA: remoteDispatchDomainMigratePrepare3Params (remote.c:4798)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
(cherry picked from commit 1f2f879ed17a784559f67b0fa2408d2436b731fd)
Most of the usage of getuid()/getgid() is in cases where we are
considering what privileges we have. As such the code should be
using the effective IDs, not real IDs.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
(cherry picked from commit 9b0af09240618184fea5884952941217e65b824f)
When running setuid, we must be careful about what env vars
we allow commands to inherit from us. Replace the
virCommandAddEnvPass function with two new ones which do
filtering
virCommandAddEnvPassAllowSUID
virCommandAddEnvPassBlockSUID
And make virCommandAddEnvPassCommon use the appropriate
ones
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
(cherry picked from commit 9b8f307c6ad002a17a0510513883d06395636793)
Conflicts:
src/qemu/qemu_command.c
After successful @cmd construction the memory where @keys points to is
part of @cmd. Avoid double freeing it.
(cherry picked from commit 3e8343e1510741623aa5bc1dfb74ec39fde868dd)
Since the wait is done during migration (still inside
QEMU_ASYNC_JOB_MIGRATION_OUT), the code should enter the monitor as such
in order to prohibit all other jobs from interfering in the meantime.
This patch fixes bug #1009886 in which qemuDomainGetBlockInfo was
waiting on the monitor condition and after GetSpiceMigrationStatus
mangled its internal data, the daemon crashed.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1009886
(cherry picked from commit 484cc3217b73b865f00bf42a9c12187b37200699)
vhost only works in KVM mode at the moment, and is infact compiled
out if the emulator is built for non-native architecture. While it
may work at some point in the future for plain qemu, for now it's
just noise on the command line (and which contributes to arm cli
breakage).
When using a <interface type="network"> that points to a network with
hostdev forwarding mode a hostdev alias is created for the network. This
allias is inserted into the hostdev list, but is backed with a part of
the network object that it is connected to.
When a VM is being stopped qemuProcessStop() calls
networkReleaseActualDevice() which eventually frees the memory for the
hostdev object. Afterwards when the domain definition is being freed by
virDomainDefFree() an invalid pointer is accessed by
virDomainHostdevDefFree() and may cause a crash of the daemon.
This patch removes the entry in the hostdev list before freeing the
depending memory to avoid this issue.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1000973
QEMU commit 3984890 introduced the "pci-hole64-size" property,
to i440FX-pcihost and q35-pcihost with a default setting of 2 GB.
Translate <pcihole64>x<pcihole64/> to:
-global q35-pcihost.pci-hole64-size=x for q35 machines and
-global i440FX-pcihost.pci-hole64-size=x for i440FX-based machines.
Error out on other machine types or if the size was specified
but the pcihost device lacks 'pci-hole64-size' property.
https://bugzilla.redhat.com/show_bug.cgi?id=990418
The ftp protocol is already recognized by qemu/KVM so add this support to
libvirt as well.
The xml should be as following:
<disk type='network' device='cdrom'>
<source protocol='ftp' name='/url/path'>
<host name='host.name' port='21'/>
</source>
</disk>
Signed-off-by: Aline Manera <alinefm@br.ibm.com>
QEMU/KVM already allows a HTTP URL for the cdrom ISO image so add this support
to libvirt as well.
The xml should be as following:
<disk type='network' device='cdrom'>
<source protocol='http' name='/url/path'>
<host name='host.name' port='80'/>
</source>
</disk>
Signed-off-by: Aline Manera <alinefm@br.ibm.com>
If there's no hard_limit set and domain uses VFIO we still must lock
the guest memory (prerequisite from qemu). Hence, we should compute
the amount to be locked from max_balloon.
When cpu hotplug fails without reporting an error, we would fail the
command but update the count of vCPUs anyways.
Commit 761fc481365703b861429d73a341bde352ba8d41 fixed the case when CPU
hot-unplug failed silently, but forgot to fix up the value in this case.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1000357
The virDomainOpenGraphics method accepts a UNIX socket FD from
the client app. It must set the label on this FD otherwise QEMU
will be prevented from receiving it with recvmsg.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If user requested multiqueue networking, beside multiple /dev/tap and
/dev/vhost-net openings, we forgot to pass mq=on onto the -device
virtio-net-pci command line. This is advised at:
http://www.linux-kvm.org/page/Multiqueue#Enable_MQ_feature
Re-arrange the code so that the returned bitmap is always initialized to
NULL even on early failures and return an error message as some callers
are already expecting it. Fix up the rest not to shadow the error.
https://bugzilla.redhat.com/show_bug.cgi?id=822052
When doing a live migration, if the destination fails for any
reason after the point in which files should be labeled, then
the cleanup of the destination would restore the labels to their
defaults, even though the source is still trying to continue
running with the image open. Bug 822052 mentioned one source
of live migration failure - a mismatch in SELinux virt_use_nfs
settings (on for source, off for destination); but I found other
situations that would also trigger it (for example, having a
graphics device tied to port 5999 on the source, and a different
domain on the destination already using that port, so that the
destination cannot reuse the port).
In short, just as cleanup of the source on a successful migration
must not relabel files (because the destination would be crippled
by the relabel), cleanup of the destination on a failed migration
must not relabel files (because the source would be crippled).
* src/qemu/qemu_process.c (qemuProcessStart): Set flag to avoid
label restoration when cleaning up on failed migration.
Signed-off-by: Eric Blake <eblake@redhat.com>
Each of the modules handled reporting error messages from the secret fetching
slightly differently with respect to the error. Provide a similar message
for each error case and provide as much data as possible.
Following XML would fail :
<disk type='network' device='lun'>
<driver name='qemu' type='raw'/>
<source protocol='iscsi' name='iqn.2013-07.com.example:iscsi/1'>
<host name='example.com' port='3260'/>
</source>
<target dev='sda' bus='scsi'/>
</disk>
With the message:
error: Failed to start domain iscsilun
error: Unable to get device ID 'iqn.2013-07.com.example:iscsi/1': No such fi
Cause was commit id '1f49b05a' which added 'virDomainDiskSourceIsBlockType'
If there's no hard_limit set and domain uses VFIO we still must lock the
guest memory (prerequisite from qemu). Hence, we should compute the
amount to be locked from max_balloon.
Since 16bcb3 we have a regression. The hard_limit is set
unconditionally. By default the limit is zero. Hence, if user hasn't
configured any, we set the zero in cgroup subsystem making the kernel
kill the corresponding qemu process immediately. The proper fix is to
set hard_limit iff user has configured any.
This function is to guess the correct limit for maximal memory
usage by qemu for given domain. This can never be guessed
correctly, not to mention all the pains and sleepless nights this
code has caused. Once somebody discovers algorithm to solve the
Halting Problem, we can compute the limit algorithmically. But
till then, this code should never see the light of the release
again.
Currently the virConnectBaselineCPU API does not expose the CPU features
that are part of the CPU's model. This patch adds a new flag,
VIR_CONNECT_BASELINE_CPU_EXPAND_FEATURES, that causes the API to explicitly
list all features that are part of that model.
Signed-off-by: Don Dugger <donald.d.dugger@intel.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
Hotplugging a single SCSI device works, but adding additional ones
result in an error from QEMU:
[root@gpok197 ~]# virsh attach-device guest01 blah.xml
Device attached successfully
[root@gpok197 ~]# virsh attach-device guest01 blah2.xml
error: Failed to attach device from blah2.xml
error: internal error unable to execute QEMU command 'device_add': Duplicate ID 'hostdev0' for device
The hostdev ID that is created is always set to zero, regardless
of the contents of the XML. Changing the index in the hotplug case
to a negative one so the next available index is used.
Signed-off-by: Eric Farman <farman@linux.vnet.ibm.com>
Reviewed-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
Go through disks of guest, if one disk doesn't exist or its backing
chain is broken, with 'optional' startupPolicy, for CDROM and Floppy
we only discard its source path definition in xml, for disks we drop
it from disk list and free it.
This patch addresses two concerns with the error reporting when an
incompatible PCI address is specified for a device:
1) It wasn't always apparent which device had the problem. With this
patch applied, any error about an incompatible address will always
contain the full address as given in the config, so it will be easier
to determine which device's config aused the problem.
2) In some cases when the problem came from bad config, the error
message was erroneously classified as VIR_ERR_INTERNAL_ERROR. With
this patch applied, the same error message will be changed to indicate
either "internal" or "xml" error depending on whether the address came
from the config, or was automatically generated by libvirt.
Note that in the case of "internal" (due to bad auto-generation)
errors, the PCI address won't be of much use in finding the location
in config to change (because it was automatically generated). Of
course that makes perfect sense, but still the address could provide a
clue about a bug in libvirt attempting to use a type of pci bus that
doesn't have its flags set correctly (or something similar). In other
words, it's not perfect, but it is definitely better.
q35 machines have an implicit ahci (sata) controller at 00:1F.2 which
has no "id" associated with it. For this reason, we can't refer to it
as "ahci0". Instead, we don't give an id on the commandline, which
qemu interprets as "use the first ahci controller". We then need to
specify the unit with "unit=%d" rather than adding it onto the bus
arg.
https://bugzilla.redhat.com/show_bug.cgi?id=979477
Since 1.0.3 we are using the new way to copy non shared storage during
migration (the NBD way). However, whether the new or old way is used is
not controllable by user but unconditionally turned on if both sides of
migration support it. Moreover, the implementation is not complete: the
combination for VIR_MIGRATE_TUNNELLED flag is missing (as we need to
open new port on the destination) in which case we just error out. This
is a deadly combination: not letting users choose their destiny and
erroring out. We should not do that but VIR_WARN and turn the NBD off
instead.
We had been setting the device alias in the devinceinfo for pci
controllers to "pci%u", but then hardcoding "pci.%u" when creating the
device address for other devices using that pci bus. This all worked
just fine until we encountered the built-in "pcie.0" bus (the PCIe
root complex) in Q35 machines.
In order to create the correct commandline for this one case, this
patch:
1) sets the alias for PCI controllers correctly, to "pci.%u" (or
"pcie.%u" for the pcie-root controller)
2) eliminates the hardcoded "pci.%u" for pci controllers when
generatuing device address strings, and instead uses the controller's
alias.
3) plumbs a pointer to the virDomainDef all the way down to
qemuBuildDeviceAddressStr. This was necessary in order to make the
aliase of the controller *used by a device* available (previously
qemuBuildDeviceAddressStr only had the deviceinfo of the device
itself, *not* of the controller it was connecting to). This made for a
larger than desired diff, but at least in the future we won't have to
do it again, since all the information we could possibly ever need for
future enhancements is in the virDomainDef. (right?)
This should be done for *all* controllers, but for now we just do it
in the case of PCI controllers, to reduce the likelyhood of
regression.
This patch adds in special handling for a few devices that need to be
treated differently for q35 domains:
usb - there is no implicit/default usb controller for the q35
machinetype. This is done because normally the default usb controller
is added to a domain by just adding "-usb" to the qemu commandline,
and it's assumed that this will add a single piix3 usb1 controller at
slot 1 function 2. That's not what happens when the machinetype is
q35, though. Instead, adding -usb to the commandline adds 3 usb
(version 2) controllers to the domain at slot 0x1D.{1,2,7}. Rather
than having
<controller type='usb' index='0'/>
translate into 3 separate devices on the PCI bus, it's cleaner to not
automatically add a default usb device; one can always be added
explicitly if desired. Or we may decide that on q35 machines, 3 usb
controllers will be automatically added when none is given. But for
this initial commit, at least we aren't locking ourselves into
something we later won't want.
video - qemu always initializes the primary video device immediately
after any integrated devices for the machinetype. Unless instructed
otherwise (by using "-device vga..." instead of "-vga" which libvirt
uses in many cases to work around deficiencies and bugs in various
qemu versions) qemu will always pick the first unused slot. In the
case of the "pc" machinetype and its derivatives, this is always slot
2, but on q35 machinetypes, the first free slot is slot 1 (since the
q35's integrated peripheral devices are placed in other slots,
e.g. slot 0x1f). In order to make the PCI address of the video device
predictable, that slot (1 or 2, depending on machinetype) is reserved
even when no video device has been specified.
sata - a q35 machine always has a sata controller implicitly added at
slot 0x1F, function 2. There is no way to avoid this controller, so we
always add it. Note that the xml2xml tests for the pcie-root and q35
cases were changed to use DO_TEST_DIFFERENT() so that we can check for
the sata controller being automatically added. This is especially
important because we can't check for it in the xml2argv output (it has
no effect on that output since it's an implicit device).
ide - q35 has no ide controllers.
isa and smbus controllers - these two are always present in a q35 (at
slot 0x1F functions 0 and 3) but we have no way of modelling them in
our config. We do need to reserve those functions so that the user
doesn't attempt to put anything else there though. (note that the "pc"
machine type also has an ISA controller, which we also ignore).
This PCI controller, named "dmi-to-pci-bridge" in the libvirt config,
and implemented with qemu's "i82801b11-bridge" device, connects to a
PCI Express slot (e.g. one of the slots provided by the pcie-root
controller, aka "pcie.0" on the qemu commandline), and provides 31
*non-hot-pluggable* PCI (*not* PCIe) slots, numbered 1-31.
Any time a machine is defined which has a pcie-root controller
(i.e. any q35-based machinetype), libvirt will automatically add a
dmi-to-pci-bridge controller if one doesn't exist, and also add a
pci-bridge controller. The reasoning here is that any useful domain
will have either an immediate (startup time) or eventual (subsequent
hot-plug) need for a standard PCI slot; since the pcie-root controller
only provides PCIe slots, we need to connect a dmi-to-pci-bridge
controller to it in order to get a non-hot-plug PCI slot that we can
then use to connect a pci-bridge - the slots provided by the
pci-bridge will be both standard PCI and hot-pluggable.
Since pci-bridge devices themselves can not be hot-plugged into a
running system (although you can hot-plug other devices into a
pci-bridge's slots), any new pci-bridge controller that is added can
(and will) be plugged into the dmi-to-pci-bridge as long as it has
empty slots available.
This patch is also changing the qemuxml2xml-pcie test from a "DO_TEST"
to a "DO_DIFFERENT_TEST". This is so that the "before" xml can omit
the automatically added dmi-to-pci-bridge and pci-bridge devices, and
the "after" xml can include it - this way we are testing if libvirt is
properly adding these devices.
