In order to learn libvirt multiqueue several things must be done:
1) The '/dev/net/tun' device needs to be opened multiple times with
IFF_MULTI_QUEUE flag passed to ioctl(fd, TUNSETIFF, &ifr);
2) Similarly, '/dev/vhost-net' must be opened as many times as in 1)
in order to keep 1:1 ratio recommended by qemu and kernel folks.
3) The command line construction code needs to switch from 'fd=X' to
'fds=X:Y:...:Z' and from 'vhostfd=X' to 'vhostfds=X:Y:...:Z'.
4) The monitor handling code needs to learn to pass multiple FDs.
In 84c59ffa I've tried to fix changing ejectable media process. The
process should go like this:
1) we need to call 'eject' on the monitor
2) we should wait for 'DEVICE_TRAY_MOVED' event
3) now we can issue 'change' command
However, while waiting in step 2) the domain monitor was locked. So
even if qemu reported the desired event, the proper callback was not
called immediately. The monitor handling code needs to lock the
monitor in order to read the event. So that's the first lock we must
not hold while waiting. The second one is the domain lock. When
monitor handling code reads an event, the appropriate callback is
called then. The first thing that each callback does is locking the
corresponding domain as a domain or its device is about to change
state. So we need to unlock both monitor and VM lock. Well, holding
any lock while sleep()-ing is not the best thing to do anyway.
Since 0d70656afded, it starts to access the sysfs files to build
the qemu command line (by virSCSIDeviceGetSgName, which is to find
out the scsi generic device name by adpater🚌target:unit), there
is no way to work around, qemu wants to see the scsi generic device
like "/dev/sg6" anyway.
And there might be other places which need to access sysfs files
when building qemu command line in future.
Instead of increasing the arguments of qemuBuildCommandLine, this
introduces a new callback for qemuBuildCommandLine, and thus tests
can register their own callbacks for sysfs test input files accessing.
* src/qemu/qemu_command.h: (New callback struct
qemuBuildCommandLineCallbacks;
extern buildCommandLineCallbacks)
* src/qemu/qemu_command.c: (wire up the callback struct)
* src/qemu/qemu_driver.c: (Use the new syntax of qemuBuildCommandLine)
* src/qemu/qemu_hotplug.c: Likewise
* src/qemu/qemu_process.c: Likewise
* tests/testutilsqemu.[ch]: (Helper testSCSIDeviceGetSgName;
callback struct testCallbacks;)
* tests/qemuxml2argvtest.c: (Use testCallbacks)
* src/tests/qemuxmlnstest.c: (Like above)
This adds both attachment and detachment support for scsi host
device.
Signed-off-by: Han Cheng <hanc.fnst@cn.fujitsu.com>
Signed-off-by: Osier Yang <jyang@redhat>
It's better to put the usb related codes into qemuDomainAttachHostUsbDevice
instead of qemuDomainAttachHostDevice.
And in the old qemuDomainAttachHostDevice, just stealing the "usb" from
driver->activeUsbHostdevs leaks the memory.
The source code base needs to be adapted as well. Some files
include virutil.h just for the string related functions (here,
the include is substituted to match the new file), some include
virutil.h without any need (here, the include is removed), and
some require both.
The USB-specific cgroup setup had been inserted inline in
qemuDomainAttachHostUsbDevice and qemuSetupCgroup, but now there is a
common cgroup setup function called for all hostdevs, so it makes sens
to put the usb-specific setup there and just rely on that function
being called.
The one thing I'm uncertain of here (and a reason for not pushing
until after release) is that previously hostdev->missing was checked
only when starting a domain (and cgroup setup for the device skipped
if missing was true), but with this consolidation, it is now checked
in the case of hotplug as well. I don't know if this will have any
practical effect (does it make sense to hotplug a "missing" usb
device?)
PCIO device assignment using VFIO requires read/write access by the
qemu process to /dev/vfio/vfio, and /dev/vfio/nn, where "nn" is the
VFIO group number that the assigned device belongs to (and can be
found with the function virPCIDeviceGetVFIOGroupDev)
/dev/vfio/vfio can be accessible to any guest without danger
(according to vfio developers), so it is added to the static ACL.
The group device must be dynamically added to the cgroup ACL for each
vfio hostdev in two places:
1) for any devices in the persistent config when the domain is started
(done during qemuSetupCgroup())
2) at device attach time for any hotplug devices (done in
qemuDomainAttachHostDevice)
The group device must be removed from the ACL when a device it
"hot-unplugged" (in qemuDomainDetachHostDevice())
Note that USB devices are already doing their own cgroup setup and
teardown in the hostdev-usb specific function. I chose to make the new
functions generic and call them in a common location though. We can
then move the USB-specific code (which is duplicated in two locations)
to this single location. I'll be posting a followup patch to do that.
This isn't strictly speaking a bugfix, but I realized I'd gotten a bit
too verbose when I chose the names for
VIR_DOMAIN_HOSTDEV_PCI_BACKEND_TYPE_*. This shortens them all a bit.
<source type='bridge'> uses a helper application to do the necessary
TUN/TAP setup to use an existing network bridge, thus letting
unprivileged users use TUN/TAP interfaces.
However, libvirt should be preventing QEMU from running any setuid
programs at all, which would include this helper program. From
a security POV, any setuid helper needs to be run by libvirtd itself,
not QEMU.
This is what this patch does. libvirt now invokes the setuid helper,
gets the TAP fd and then passes it to QEMU in the normal manner.
The path to the helper is specified in qemu.conf.
As a small advantage, this adds a <target dev='tap0'/> element to the
XML of an active domain using <interface type='bridge'>.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
VFIO requires all of the guest's memory and IO space to be lockable in
RAM. The domain's max_balloon is the maximum amount of memory the
domain can have (in KiB). We add a generous 1GiB to that for IO space
(still much better than KVM device assignment, where the KVM module
actually *ignores* the process limits and locks everything anyway),
and convert from KiB to bytes.
In the case of hotplug, we are changing the limit for the already
existing qemu process (prlimit() is used under the hood), and for
regular commandline additions of vfio devices, we schedule a call to
setrlimit() that will happen after the qemu process is forked.
The device option for vfio-pci is nearly identical to that for
pci-assign - only the configfd parameter isn't supported (or needed).
Checking for presence of the bootindex parameter is done separately
from constructing the commandline, similar to how it is done for
pci-assign.
This patch contains tests to check for proper commandline
construction. It also includes tests for parser-formatter-parser
roundtrips (xml2xml), because those tests use the same data files, and
would have failed had they been included before now.
qemu: xml/args tests for VFIO hostdev and <interface type='hostdev'/>
These should be squashed in with the patch that adds commandline
handling of vfio (they would fail at any earlier time).
There will soon be other items related to pci hostdevs that need to be
in the same part of the hostdevsubsys union as the pci address (which
is currently a single member called "pci". This patch replaces the
single member named pci with a struct named pci that contains a single
member named "addr".
Instead of calling virCgroupForDomain every time we need
the virCgrouPtr instance, just do it once at Vm startup
and cache a reference to the object in qemuDomainObjPrivatePtr
until shutdown of the VM. Removing the virCgroupPtr from
the QEMU driver state also means we don't have stale mount
info, if someone mounts the cgroups filesystem after libvirtd
has been started
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
f946462e14ac036357b7c11ce5c23f94a3ee4e49 changed behavior by settings
VIR_DOMAIN_DEVICE_ADDRESS_TYPE_PCI upfront. If we do so before invoking
qemuDomainPCIAddressEnsureAddr we merely try to set the PCI slot via
qemuDomainPCIAddressReserveSlot instead reserving a new address via
qemuDomainPCIAddressSetNextAddr which fails with
$ ~/run-tck-test domain/200-disk-hotplug.t
./scripts/domain/200-disk-hotplug.t .. # Creating a new transient domain
./scripts/domain/200-disk-hotplug.t .. 1/5 # Attaching the new disk /var/lib/jenkins/jobs/libvirt-tck-build/workspace/scratchdir/200-disk-hotplug/extra.img
# Failed test 'disk has been attached'
# at ./scripts/domain/200-disk-hotplug.t line 67.
# died: Sys::Virt::Error (libvirt error code: 1, message: internal error unable to reserve PCI address 0:0:0.0
# )
The VIR_ERR_NO_SUPPORT error code is reserved for cases where an
API is not implemented in a driver. It definitely should not be
used when an API execution fails due to unsupported operation.
We didn't yet expose the virtio device attach and detach functionality
for s390 domains as the device hotplug was very limited with the old
virtio-s390 bus. With the CCW bus there's full hotplug support for
virtio devices in QEMU, so we are adding this to libvirt too.
Since the virtio hotplug isn't limited to PCI anymore, we change the
function names from xxxPCIyyy to xxxVirtioyyy, where we handle all
three virtio bus types.
Signed-off-by: J.B. Joret <jb@linux.vnet.ibm.com>
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.vnet.ibm.com>
For both AttachDevice and UpdateDevice APIs, if the disk device
is 'cdrom' or 'floppy', the operations could be ejecting, updating,
and inserting. For either ejecting or updating, the shared disk
entry of the original disk src has to be removed, because it's
not useful anymore.
And since the original disk def will be changed, new disk def passed
as argument will be free'ed in qemuDomainChangeEjectableMedia, so
we need to copy the orignal disk def before
qemuDomainChangeEjectableMedia, to use it for qemuRemoveSharedDisk.
Some functions were using virDomainDeviceInfo where virDevicePCIAddress
would suffice. Some were only using integers for slots and functions,
assuming the bus numbers are always 0.
Switch from virDomainDeviceInfoPtr to virDevicePCIAddressPtr:
qemuPCIAddressAsString
qemuDomainPCIAddressCheckSlot
qemuDomainPCIAddressReserveAddr
qemuDomainPCIAddressReleaseAddr
Switch from int slot to virDevicePCIAddressPtr:
qemuDomainPCIAddressReserveSlot
qemuDomainPCIAddressReleaseSlot
qemuDomainPCIAddressGetNextSlot
Deleted functions (they would take the same parameters
as ReserveAddr/ReleaseAddr do now.)
qemuDomainPCIAddressReserveFunction
qemuDomainPCIAddressReleaseFunction
With the majority of fields in the virQEMUDriverPtr struct
now immutable or self-locking, there is no need for practically
any methods to be using the QEMU driver lock. Only a handful
of helper APIs in qemu_conf.c now need it
From qemu's point of view these are still just tap devices, so there's
no reason they shouldn't work with vhost-net; as a matter of fact,
Raja Sivaramakrishnan <srajag00@yahoo.com> verified on libvir-list
that at least the qemu_command.c part of this patch works:
https://www.redhat.com/archives/libvir-list/2012-December/msg01314.html
(the hotplug case is extrapolation on my part).
To avoid confusion between 'virCapsPtr' and 'qemuCapsPtr'
do some renaming of various fucntions/variables. All
instances of 'qemuCapsPtr' are renamed to 'qemuCaps'. To
avoid that clashing with the 'qemuCaps' typedef though,
rename the latter to virQEMUCaps.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Currently the activePciHostdevs, inactivePciHostdevsd and
activeUsbHostdevs lists are all implicitly protected by the
QEMU driver lock. Now that the lists all inherit from the
virObjectLockable, we can make the locking explicit, removing
the dependency on the QEMU driver lock for correctness.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
To allow modifications to the lists to be synchronized, convert
virPCIDeviceList and virUSBDeviceList into virObjectLockable
classes. The locking, however, will not be self-contained. The
users of these classes will have to call virObjectLock/Unlock
in the critical regions.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Currently the virQEMUDriverPtr struct contains an wide variety
of data with varying access needs. Move all the static config
data into a dedicated virQEMUDriverConfigPtr object. The only
locking requirement is to hold the driver lock, while obtaining
an instance of virQEMUDriverConfigPtr. Once a reference is held
on the config object, it can be used completely lockless since
it is immutable.
NB, not all APIs correctly hold the driver lock while getting
a reference to the config object in this patch. This is safe
for now since the config is never updated on the fly. Later
patches will address this fully.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=892289
It seems like with new udev within guest OS, the tray is locked,
so we need to:
- 'eject'
- wait for tray to open
- 'change'
Moreover, even when doing bare 'eject', we should check for
'tray_open' as guest may have locked the tray. However, the
waiting phase shouldn't be unbounded, so I've chosen 10 retries
maximum, each per 500ms. This should give enough time for guest
to eject a media and open the tray.
This avoids "Event negative_returns: A negative constant "-1" is passed as
an argument to a parameter that cannot be negative.". The called function
uses -1 to determine whether it needs to traverse all the hostdevs.
When LXC labels USB devices during hotplug, it is running in
host context, so it needs to pass in a vroot path to the
container root.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
When a network device's bridge connection is changed by
virDomainUpdateDevice, libvirt first removes the netdev's tap from its
old bridge, then adds it to the new bridge. Sometimes, due to a
network being destroyed while a guest device is still attached, the
tap may already be "removed" from the old bridge (or the old bridge
may not even exist any more); the existing code was needlessly failing
the update when this happened, making it impossible to recover from
the situation without completely detaching (i.e. removing) the netdev
from the guest and re-attaching.
Instead of failing the entire operation when removal of the tap from
the old bridge fails, this patch changes qemuDomainChangeNetBridge to
just log a warning and continue, allowing a reasonable recover from
the situation.
(you'll appreciate this change if you ever accidentally destroy a
network while your guests are still using it).
This fixes a problem that showed up during testing of:
https://bugzilla.redhat.com/show_bug.cgi?id=881480
Due to a logic error in the function that gets the name of the bridge
an interface connects to, any time a bridge was specified directly
(type='bridge') rather than indirectly (type='network'), An error
would be logged (although the operation would then complete
successfully):
Network type 6 is not supported
The final virReportError() in the function
qemuDomainNetGetBridgeName() was apparently avoided in the past with a
"goto cleanup" at the end of each case, but the case of bridge somehow
no longer has that final goto cleanup.
The proper solution is anyway to not rely on goto's, but put the error
log inside an else {} clause, so that it's executed only if the type
is neither bridge nor network (in reality, this function should only
ever be called for those two types, that's why this is an internal
error).
While making this change, the error message was also tuned to be more
correct (since it's not really the type of the network, but the type
of the interface, and it *is* otherwise supported, it's just that the
interface type in question doesn't *have* a bridge device associated
with it, or at least we don't know how to get it).
Since we can't (currently) rely on the ability to provide blanket
support for all possible network changes by calling the toplevel
netdev hostside disconnect/connect functions (due to qemu only
supporting a lockstep between initialization of host side and guest
side of devices), in order to support live change of an interface's
nwfilter we need to make a special purpose function to only call the
nwfilter teardown and setup functions if the filter for an interface
(or its parameters) changes. The pattern is nearly identical to that
used to change the bridge that an interface is connected to.
This patch was inspired by a request from Guido Winkelmann
<guido@sagersystems.de>, who tested an earlier version.
https://bugzilla.redhat.com/show_bug.cgi?id=876828
Commit 38c4a9cc introduced a regression in hot unplugging of disks
from qemu, where cgroup device ACLs were no longer being revoked
(thankfully not a security hole: cgroup ACLs only prevent open()
of the disk; so reverting the ACL prevents future abuse but doesn't
stop abuse from an fd that was already opened before the ACL change).
Commit 1b2ebf95 overlooked that there were two spots affected.
* src/qemu/qemu_hotplug.c (qemuDomainDetachDiskDevice):
Transfer backing chain before deletion.
* src/qemu/qemu_driver.c (qemuDomainDetachDeviceDiskLive): Fix
spacing (partly to ensure a different-looking patch).
Remove the obsolete 'qemud' naming prefix and underscore
based type name. Introduce virQEMUDriverPtr as the replacement,
in common with LXC driver naming style
https://bugzilla.redhat.com/show_bug.cgi?id=876828
Commit 38c4a9cc introduced a regression in hot unplugging of disks
from qemu, where cgroup device ACLs were no longer being revoked
(thankfully not a security hole: cgroup ACLs only prevent open()
of the disk; so reverting the ACL prevents future abuse but doesn't
stop abuse from an fd that was already opened before the ACL change).
The actual regression is due to a latent bug. The hot unplug code
was computing the set of files needing cgroup ACL revocation based
on the XML passed in by the user, rather than based on the domain's
details on which disk was being deleted. As long as the revoke
path was always recomputing the backing chain, this didn't really
matter; but now that we want to compute the chain exactly once and
remember that computation, we need to hang on to the backing chain
until after the revoke has happened.
* src/qemu/qemu_hotplug.c (qemuDomainDetachPciDiskDevice):
Transfer backing chain before deletion.
The libvirt coding standard is to use 'function(...args...)'
instead of 'function (...args...)'. A non-trivial number of
places did not follow this rule and are fixed in this patch.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
For now, disk migration via block copy job is not implemented in
libvirt. But when we do implement it, we have to deal with the
fact that qemu does not yet provide an easy way to re-start a qemu
process with mirroring still intact. Paolo has proposed an idea
for a persistent dirty bitmap that might make this possible, but
until that design is complete, it's hard to say what changes
libvirt would need. Even something like 'virDomainSave' becomes
hairy, if you realize the implications that 'virDomainRestore'
would be stuck with recreating the same mirror layout.
But if we step back and look at the bigger picture, we realize that
the initial client of live storage migration via disk mirroring is
oVirt, which always uses transient domains, and that if a transient
domain is destroyed while a mirror exists, oVirt can easily restart
the storage migration by creating a new domain that visits just the
source storage, with no loss in data.
We can make life a lot easier by being cowards for now, forbidding
certain operations on a domain. This patch guarantees that we
never get in a state where we would have to restart a domain with
a mirroring block copy, by preventing saves, snapshots, migration,
hot unplug of a disk in use, and conversion to a persistent domain
(thankfully, it is still relatively easy to 'virsh undefine' a
running domain to temporarily make it transient, run tests on
'virsh blockcopy', then 'virsh define' to restore the persistence).
Later, if the qemu design is enhanced, we can relax our code.
The change to qemudDomainDefine looks a bit odd for undoing an
assignment, rather than probing up front to avoid the assignment,
but this is because of how virDomainAssignDef combines both a
lookup and assignment into a single function call.
* src/conf/domain_conf.h (virDomainHasDiskMirror): New prototype.
* src/conf/domain_conf.c (virDomainHasDiskMirror): New function.
* src/libvirt_private.syms (domain_conf.h): Export it.
* src/qemu/qemu_driver.c (qemuDomainSaveInternal)
(qemuDomainSnapshotCreateXML, qemuDomainRevertToSnapshot)
(qemuDomainBlockJobImpl, qemudDomainDefine): Prevent dangerous
actions while block copy is already in action.
* src/qemu/qemu_hotplug.c (qemuDomainDetachDiskDevice): Likewise.
* src/qemu/qemu_migration.c (qemuMigrationIsAllowed): Likewise.
