Since we have qemuSecurity wrappers over
virSecurityManagerSetHostdevLabel and
virSecurityManagerRestoreHostdevLabel we ought to use them
instead of calling secdriver APIs directly. Without those
wrappers the labelling won't be done in the correct namespace
and thus won't apply to the nodes seen by qemu itself.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
libvirt was able to set the host_mtu option when an MTU was explicitly
given in the interface config (with <mtu size='n'/>), set the MTU of a
libvirt network in the network config (with the same named
subelement), and would automatically set the MTU of any tap device to
the MTU of the network.
This patch ties that all together (for networks based on tap devices
and either Linux host bridges or OVS bridges) by learning the MTU of
the network (i.e. the bridge) during qemuInterfaceBridgeConnect(), and
returning that value so that it can then be passed to
qemuBuildNicDevStr(); qemuBuildNicDevStr() then sets host_mtu in the
interface's commandline options.
The result is that a higher MTU for all guests connecting to a
particular network will be plumbed top to bottom by simply changing
the MTU of the network (in libvirt's config for libvirt-managed
networks, or directly on the bridge device for simple host bridges or
OVS bridges managed outside of libvirt).
One question I have about this - it occurred to me that in the case of
migrating a guest from a host with an older libvirt to one with a
newer libvirt, the guest may have *not* had the host_mtu option on the
older machine, but *will* have it on the newer machine. I'm curious if
this could lead to incompatibilities between source and destination (I
guess it all depends on whether or not the setting of host_mtu has a
practical effect on a guest that is already running - Maxime?)
Likewise, we could run into problems when migrating from a newer
libvirt to older libvirt - The guest would have been told of the
higher MTU on the newer libvirt, then migrated to a host that didn't
understand <mtu size='blah'/>. (If this really is a problem, it would
be a problem with or without the current patch).
Example:
<network>
...
<mtu size='9000'/>
...
If mtu is unset, it's assumed that we want the default for whatever is
the underlying transport (usually this is 1500).
This setting isn't yet wired in, so it will have no effect.
This partially resolves: https://bugzilla.redhat.com/1224348
virNetDevTapCreateInBridgePort() has always set the new tap device to
the current MTU of the bridge it's being attached to. There is one
case where we will want to set the new tap device to a different
(usually larger) MTU - if that's done with the very first device added
to the bridge, the bridge's MTU will be set to the device's MTU. This
patch allows for that possibility by adding "int mtu" to the arg list
for virNetDevTapCreateInBridgePort(), but all callers are sending -1,
so it doesn't yet have any effect.
Since the requested MTU isn't necessarily what is used in the end (for
example, if there is no MTU requested, the tap device will be set to
the current MTU of the bridge), and the hypervisor may want to know
the actual MTU used, we also return the actual MTU to the caller (if
actualMTU is non-NULL).
In order for memory locking to work, the hard limit on memory
locking (and usage) has to be set appropriately by the user.
The documentation mentions the requirement already: with this
patch, it's going to be enforced by runtime checks as well,
by forbidding a non-compliant guest from being defined as well
as edited and started.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1316774
Like it usually happens, I fixed one thing and broke another:
in 803966c76 address allocation was fixed for SATA disks, but
broke that for virtio disks, because it dropped disk address
assignment completely. It's not needed for SATA disks anymore,
but still needed for the virtio ones.
Bring that back and add a couple of tests to make sure it won't
happen again.
Similarly to one of the previous commits, we need to deal
properly with symlinks in hotplug case too.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Imagine you have a disk with the following source set up:
/dev/disk/by-uuid/$uuid (symlink to) -> /dev/sda
After cbc45525cb the transitive end of the symlink chain is
created (/dev/sda), but we need to create any item in chain too.
Others might rely on that.
In this case, /dev/disk/by-uuid/$uuid comes from domain XML thus
it is this path that secdriver tries to relabel. Not the resolved
one.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
After previous commit this has become redundant step.
Also setting up devices in namespace and setting their label
later on are two different steps and should be not done at once.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
The idea is to move all the seclabel setting to security driver.
Having the relabel code spread all over the place looks very
messy.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Because of the nature of security driver transactions, it is
impossible to use them properly. The thing is, transactions enter
the domain namespace and commit all the seclabel changes.
However, in RestoreAllLabel() this is impossible - the qemu
process, the only process running in the namespace, is gone. And
thus is the namespace. Therefore we shouldn't use the transactions
as there is no namespace to enter.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
The current ordering is as follows:
1) set label
2) create the device in namespace
3) allow device in the cgroup
While this might work for now, it will definitely not work if the
security driver would use transactions as in that case there
would be no device to relabel in the domain namespace as the
device is created in the second step.
Swap steps 1) and 2) to allow security driver to use more
transactions.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
We will need to traverse the symlinks one step at the time.
Therefore we need to see where a symlink is pointing to.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
The comment to the function states that the errors from the child
process are reported. Well, the error buffer is filled with
possible error messages. But then it is thrown away. Among with
important error message from the child process.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
==11260== 1,006 bytes in 1 blocks are definitely lost in loss record 106 of 111
==11260== at 0x4C2AE5F: malloc (vg_replace_malloc.c:297)
==11260== by 0x4C2BDFF: realloc (vg_replace_malloc.c:693)
==11260== by 0x4EA430B: virReallocN (viralloc.c:245)
==11260== by 0x4EA7C52: virBufferGrow (virbuffer.c:130)
==11260== by 0x4EA7D28: virBufferAdd (virbuffer.c:165)
==11260== by 0x4EA8E10: virBufferStrcat (virbuffer.c:718)
==11260== by 0x42D263: xenFormatXLDiskSrcNet (xen_xl.c:960)
==11260== by 0x42D4EB: xenFormatXLDiskSrc (xen_xl.c:1015)
==11260== by 0x42D870: xenFormatXLDisk (xen_xl.c:1101)
==11260== by 0x42DA89: xenFormatXLDomainDisks (xen_xl.c:1148)
==11260== by 0x42EAF8: xenFormatXL (xen_xl.c:1558)
==11260== by 0x40E85F: testCompareParseXML (xlconfigtest.c:105)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Originally/discovered proposed by "Wang King <king.wang@huawei.com>"
When the virCloseCallbacksSet is first called, it increments the refcnt
on the domain object to ensure it doesn't get deleted before the callback
is called. The refcnt would be decremented in virCloseCallbacksUnset once
the entry is removed from the closeCallbacks has table.
When (mostly) normal shutdown occurs, the qemuProcessStop will end up
calling qemuProcessAutoDestroyRemove and will remove the callback from
the list and hash table normally and decrement the refcnt.
However, when qemuConnectClose calls virCloseCallbacksRun, it will scan
the (locked) closeCallbacks list for matching domain and callback function.
If an entry is found, it will be removed from the closeCallbacks list and
placed into a lookaside list to be processed when the closeCallbacks lock
is dropped. The callback function (e.g. qemuProcessAutoDestroy) is called
and will run qemuProcessStop. That code will fail to find the callback
in the list when qemuProcessAutoDestroyRemove is called and thus not decrement
the domain refcnt. Instead since the entry isn't found the code will just
return (mostly) harmlessly.
This patch will resolve the issue by taking another ref during the
search UUID process during virCloseCallackRun, decrementing the refcnt
taken by virCloseCallbacksSet, calling the callback routine and returning
overwriting the vm (since it could return NULL). Finally, it will call the
virDomainObjEndAPI to lower the refcnt and remove the lock taken during
the search UUID processing. This may cause the vm to be destroyed.
After deploying virtlogd by default we identified a number of
mistakes in the systemd unit file. virtlockd's relationship
to libvirtd is the same as virtlogd, so we must apply the
same unit file fixes to virtlockd
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Updating docs/news.xml in the same commit that performs the
documented change makes backports needlessly complicated,
both for mainteinance branches and downstream distributions,
because it introduces additional potential for merge
conflicts.
Document in the contributor guidelines that the release notes
should be updated in a separate commit instead, so that it's
easy to backport just the code change.
xen.git commit 57f8b13c changed several of the libxl memory
get/set functions to take 64 bit parameters. The libvirt
libxl driver still uses uint32_t variables for these various
parameters, which is particularly problematic for the
libxl_set_memory_target() function.
When dom0 autoballooning is enabled, libvirt (like xl) determines
the memory needed to start a domain and the memory available. If
memory available is less than memory needed, dom0 is ballooned
down by passing a negative value to libxl_set_memory_target()
'target_memkb' parameter. Prior to xen.git commit 57f8b13c,
'target_memkb' was an int32_t. Subtracting a larger uint32 from
a smaller uint32 and assigning it to int32 resulted in a negative
number. After commit 57f8b13c, the same subtraction is widened
to a int64, resulting in a large positive number. The simple
fix taken by this patch is to assign the difference of the
uint32 values to a temporary int32 variable, which is then
passed to 'target_memkb' parameter of libxl_set_memory_target().
Note that it is undesirable to change libvirt to use 64 bit
variables since it requires setting LIBXL_API_VERSION to 0x040800.
Currently libvirt supports LIBXL_API_VERSION >= 0x040400,
essentially Xen >= 4.4.
The test monitor should be freed separately so we need to remove the
pointer from the @vm object. This fixes a race condition crash in the
test introduced in commit a245abce43.
Now that we have a function for properly assigning the blockdeviotune
info, let's use it instead of dropping the group name on every
assignment. Otherwise it will not work with both --live and --config
options.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
This is necessary to be able to get statistics for venet0 or
"host-routed" adapter, which has -1 index and thus, its statistics
is shown as "net.nic4294967295".
Signed-off-by: Maxim Nestratov <mnestratov@virtuozzo.com>
Commit 815d98a started auto-adding one hub if there are more USB devices
than available USB ports.
This was a strange choice, since there might be even more devices.
Before USB address allocation was implemented in libvirt, QEMU
automatically added a new USB hub if the old one was full.
Adjust the logic to try adding as many hubs as will be needed
to plug in all the specified devices.
https://bugzilla.redhat.com/show_bug.cgi?id=1410188
As bhyve for a long time didn't have a notion of the explicit SATA
controller and created a controller for each drive, the bhyve driver
in libvirt acted in a similar way and didn't care about the SATA
controllers and assigned PCI addresses to drives directly, as
the generated command will look like this anyway:
2:0,ahci-hd,somedisk.img
This no longer makes sense because:
1. After commit c07d1c1c4f it's not possible to assign
PCI addresses to disks
2. Bhyve now supports multiple disk drives for a controller,
so it's going away from 1:1 controller:disk mapping, so
the controller object starts to make more sense now
So, this patch does the following:
- Assign PCI address to SATA controllers (previously we didn't do this)
- Assign disk addresses instead of PCI addresses for disks. Now, when
building a bhyve command, we take PCI address not from the disk
itself but from its controller
- Assign addresses at XML parsing time using the
assignAddressesCallback. This is done mainly for being able to
verify address allocation via xml2xml tests
- Adjust existing bhyvexml2{xml,argv} tests to chase the new
address allocation
This patch is largely based on work of Fabian Freyer.
Add virBhyveDriverCreateXMLConf, a simple wrapper around
virDomainXMLOptionNew that makes it easier to pass bhyveConnPtr
as a private data for parser. It will be used later for device
address allocation at parsing time.
Update consumers to use it instead of direct calls to
virDomainXMLOptionNew.
As we now have proper callbacks connected for the tests, update
test files accordingly to include the automatically generated
PCI root controller.
Introduce a BHYVE_CAP_AHCI32SLOT capability that shows
if 32 devices per SATA controller are supported, and
a bhyveProbeCapsAHCI32Slot function that probes it.
Because this is invalid xml for containers. This patch almost
reverts 7eda8369, but still skips converting vz sdk bootorder
for containers to libvirt bootorder because we use boot order
in containers for quite different purpurse.
==12618== 110 bytes in 10 blocks are definitely lost in loss record 269 of 295
==12618== at 0x4C2AE5F: malloc (vg_replace_malloc.c:297)
==12618== by 0x1CFC6DD7: vasprintf (vasprintf.c:73)
==12618== by 0x1912B2FC: virVasprintfInternal (virstring.c:551)
==12618== by 0x1912B411: virAsprintfInternal (virstring.c:572)
==12618== by 0x50B1FF: qemuAliasChardevFromDevAlias (qemu_alias.c:638)
==12618== by 0x518CCE: qemuBuildChrChardevStr (qemu_command.c:4973)
==12618== by 0x522DA0: qemuBuildShmemBackendChrStr (qemu_command.c:8674)
==12618== by 0x523209: qemuBuildShmemCommandLine (qemu_command.c:8789)
==12618== by 0x526135: qemuBuildCommandLine (qemu_command.c:9843)
==12618== by 0x48B4BA: qemuProcessCreatePretendCmd (qemu_process.c:5897)
==12618== by 0x4378C9: testCompareXMLToArgv (qemuxml2argvtest.c:498)
==12618== by 0x44D5A6: virTestRun (testutils.c:180)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
For example when both total_bytes_sec and total_bytes_sec_max are set,
but the former gets cleaned due to new call setting, let's say,
read_bytes_sec, we end up with this weird message for the command:
$ virsh blkdeviotune fedora vda --read-bytes-sec 3000
error: Unable to change block I/O throttle
error: unsupported configuration: value 'total_bytes_sec_max' cannot be set if 'total_bytes_sec' is not set
So let's make it more descriptive. This is how it looks after the change:
$ virsh blkdeviotune fedora vda --read-bytes-sec 3000
error: Unable to change block I/O throttle
error: unsupported configuration: cannot reset 'total_bytes_sec' when 'total_bytes_sec_max' is set
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1344897
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
