Prior to commit 55ce6564 (first in libvirt 4.6.0), the XML sent to
virDomainAttachDeviceFlags() was parsed only once, and the results of
that parse were inserted into both the live object of the running
domain and into the persistent config. Thus, if MAC address was
omitted from in XML for a network device (<interface>), both the live
and config object would have the same MAC address.

Commit 55ce6564 changed the code to parse the incoming XML twice -
once for live and once for config. This does eliminate the problem of
PCI (/scsi/sata) address conflicts caused by allocating an address
based on existing devices in live object, but then inserting the
result into the config (which may already have a device using that
address), BUT it also means that when the MAC address of a network
device hasn't been specified in the XML, each copy will get a
different auto-generated MAC address.

This results in the MAC address of the device changing the next time
the domain is shutdown and restarted, which creates havoc with the
guest OS's network config.

There have been several discussions about this in the last > 1 year,
attempting to find the ideal solution to this problem that makes MAC
addresses consistent and accounts for all sorts of corner cases with
PCI/scsi/sata addresses. All of these discussions fizzled out because
every proposal was either too difficult to implement or failed to fix
some esoteric case someone thought up.

So, in the interest of solving the MAC address problem while not
making the "other address" situation any worse than before, this patch
simply adds a qemuDomainAttachDeviceLiveAndConfigHomogenize() function
that (for now) copies the MAC address from the config object to the
live object (if the original xml had <mac address='blah'/> then this
will be an effective NOP (as the macs already match)).

Any downstream libvirt containing upstream commit
55ce6564 should have this patch as well.

https://bugzilla.redhat.com/1783411Signed-off-by: NLaine Stump <laine@redhat.com>
Reviewed-by: NMichal Privoznik <mprivozn@redhat.com>

The intent of get_nonnull_domain() is not to validate virDomain
as sent by the client but just to construct the virDomain
structure. The validation is then done in each API when looking
up the domain in our internal hash tables.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

There are some functions which pass virConnectPtr around for one
reason and one reason only: to obtain virLXCDriverPtr in the end.
Might replace the argument and pass a pointer to the driver right
from the start.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

If we use glib alloc functions, we can drop the 'cleanup' label
and @rv variable and also simplify the code a bit.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Some variables are not used outside of the for() loop. Move their
declaration to clean up the code a bit.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

When using the monolithic daemon, then dom->conn has all driver
tables filled in properly and thus it's safe to call an API other
than virDomain*(). However, when using split daemons then
dom->conn has only hypervisor driver table set
(dom->conn->driver) and the rest is NULL. Therefore, if we want
to call a non-domain API (virNetworkLookupByName() in this case),
we have obtain the cached connection object accessible via
virGetConnectNetwork().
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

If we use glib alloc functions, we can drop the 'cleanup' label
and @rv variable and also simplify the code a bit.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Some variables are not used outside of the for() loop. Move their
declaration to clean up the code a bit.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

When using the monolithic daemon, then dom->conn has all driver
tables filled in properly and thus it's safe to call an API other
than virDomain*(). However, when using split daemons then
dom->conn has only hypervisor driver table set
(dom->conn->driver) and the rest is NULL. Therefore, if we want
to call a non-domain API (virNetworkLookupByName() in this case),
we have obtain the cached connection object accessible via
virGetConnectNetwork().
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

If we place qemuDomainInterfaceAddresses() a few lines below the
two functions its using then we can drop forward declarations of
those functions.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

While at bugfixing, convert the whole function to the new-style memory
allocation handling.
Reviewed-by: NCole Robinson <crobinso@redhat.com>
Signed-off-by: NPavel Mores <pmores@redhat.com>

Pick 256k as the limit.

While -Wno-frame-larger-than would make more sense for usage
in our test suite, the -Wno version seems to have no effect
if -Wframe-larger-than was already specified.

Use an (un)reasonably large value instead.

Fixes the build with clang:
../../tests/cputest.c:964:1: error: stack frame size of 33176 bytes
in function 'mymain' [-Werror,-Wframe-larger-than=]
mymain(void)
^
1 error generated.
Signed-off-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

My commit e73889b6
split the -Wframe-larger-than warning setting into
two different variables - STRICT_FRAME_LIMIT_CFLAGS
for the library code and RELAXED_FRAME_LIMIT_CFLAGS
which was needed for tests.

Use the strict limit by default and specify the warning
flag twice for the parts that require a larger stack
frame, relying on the fact that the compiler will pick
up the latter value.
Signed-off-by: NJán Tomko <jtomko@redhat.com>
Reviewed-by: NDaniel Henrique Barboza <danielhb413@gmail.com>

This is slightly more complicated because NVMe disk source is not
a simple attribute to <source/> element. The format in which the
PCI address and namespace ID are printed is the same as QEMU
accepts them:

  nvme://XXXX:XX:XX.X/X
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

With NVMe disks, one can start a blockjob with a NVMe disk
that is not visible in domain XML (at least right away). Usually,
it's fairly easy to override this limitation of
qemuDomainGetMemLockLimitBytes() - for instance for hostdevs we
temporarily add the device to domain def, let the function
calculate the limit and then remove the device. But it's not so
easy with virStorageSourcePtr - in some cases they don't
necessarily are attached to a disk. And even if they are it's
done later in the process and frankly, I find it too complicated
to be able to use the simple trick we use with hostdevs.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

At the very beginning of the attach function the
qemuDomainStorageSourceChainAccessAllow() is called which
modifies CGroups, locks and seclabels for new disk and its
backing chain. This must be followed by a counterpart which
reverts back all the changes if something goes wrong. This boils
down to calling qemuDomainStorageSourceChainAccessRevoke() which
is done under 'error' label. But not all failure branches jump
there. They just jump onto 'cleanup' label where no revoke is
done. Such mistake is easy to do because 'cleanup' label does
exist. Therefore, dissolve 'error' block in 'cleanup' and have
everything jump onto 'cleanup' label.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Because this is a HMP we're dealing with, there is nothing like
class of reply message, so we have to do some string comparison
to guess if the command fails. Well, with NVMe disks whole new
class of errors comes to play because qemu needs to initialize
IOMMU and VFIO for them. You can see all the messages it may
produce in qemu_vfio_init_pci().
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Now, that we have everything prepared, we can generate command
line for NVMe disks.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

This capability tracks if qemu is capable of:

  -drive file.driver=nvme

The feature was added in QEMU's commit of v2.12.0-rc0~104^2~2.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

This function is currently not called for any type of storage
source that is not considered 'local' (as defined by
virStorageSourceIsLocalStorage()). Well, NVMe disks are not
'local' from that point of view and therefore we will need to
call this function more frequently.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

If a domain has an NVMe disk configured, then we need to allow it
on devices CGroup so that qemu can access it. There is one caveat
though - if an NVMe disk is read only we need CGroup to allow
write too. This is because when opening the device, qemu does
couple of ioctl()-s which are considered as write.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

There are couple of places where a domain with a VFIO device gets
special treatment: in CGroups when enabling/disabling access to
/dev/vfio/vfio, and when creating/removing nodes in domain mount
namespace. Well, a NVMe disk is a VFIO device too. Fortunately,
we have this qemuDomainNeedsVFIO() function which is the only
place that needs adjustment.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

If a domain has an NVMe disk configured, then we need to create
/dev/vfio/* paths in domain's namespace so that qemu can open
them.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

We have this beautiful function that does crystal ball
divination. The function is named
qemuDomainGetMemLockLimitBytes() and it calculates the upper
limit of how much locked memory is given guest going to need. The
function bases its guess on devices defined for a domain. For
instance, if there is a VFIO hostdev defined then it adds 1GiB to
the guessed maximum. Since NVMe disks are pretty much VFIO
hostdevs (but not quite), we have to do the same sorcery.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

The qemu driver has its own wrappers around virHostdev module (so
that some arguments are filled in automatically). Extend these to
include NVMe devices too.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

The device configs (which are actually the same one config)
come from a NVMe disk of mine.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Now that we have virNVMeDevice module (introduced in previous
commit), let's use it int virHostdev to track which NVMe devices
are free to be used by a domain and which are taken.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

This module will be used by virHostdevManager and it's inspired
by virPCIDevice module. They are very similar except instead of
what makes a NVMe device: PCI address AND namespace ID. This
means that a NVMe device can appear in a domain multiple times,
each time with a different namespace.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

This function will return true if any of disks (or their backing
chain) for given domain contains an NVMe disk.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

This function will return true if there's a storage source of
type VIR_STORAGE_TYPE_NVME, or false otherwise.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

To simplify implementation, some restrictions are added. For
instance, an NVMe disk can't go to any bus but virtio and has to
be type of 'disk' and can't have startupPolicy set.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

There is this class of PCI devices that act like disks: NVMe.
Therefore, they are both PCI devices and disks. While we already
have <hostdev/> (and can assign a NVMe device to a domain
successfully) we don't have disk representation. There are three
problems with PCI assignment in case of a NVMe device:

1) domains with <hostdev/> can't be migrated

2) NVMe device is assigned whole, there's no way to assign only a
   namespace

3) Because hypervisors see <hostdev/> they don't put block layer
   on top of it - users don't get all the fancy features like
   snapshots

NVMe namespaces are way of splitting one continuous NVDIMM memory
into smaller ones, effectively creating smaller NVMe-s (which can
then be partitioned, LVMed, etc.)

Because of all of this the following XML was chosen to model a
NVMe device:

  <disk type='nvme' device='disk'>
    <driver name='qemu' type='raw'/>
    <source type='pci' managed='yes' namespace='1'>
      <address domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>
    </source>
    <target dev='vda' bus='virtio'/>
  </disk>
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

There are going to be more disk types that are considered unsafe
with respect to migration. Therefore, move the error reporting
call outside of if() body and rework if-else combo to switch().
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

This helper is cleaner than plain memcpy() because one doesn't
have to look into virPCIDeviceAddress struct to see if it
contains any strings / pointers.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

In near future we will have a list of PCI devices we want to
re-attach to the host (held in virPCIDeviceListPtr) but we don't
have virDomainHostdevDefPtr. That's okay because
virHostdevReAttachPCIDevices() works with virPCIDeviceListPtr
mostly anyway. And in very few places where it needs
virDomainHostdevDefPtr are not interesting for our case.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

In near future we will have a list of PCI devices we want to
detach (held in virPCIDeviceListPtr) but we don't have
virDomainHostdevDefPtr. That's okay because
virHostdevPreparePCIDevices() works with virPCIDeviceListPtr
mostly anyway. And in very few places where it needs
virDomainHostdevDefPtr are not interesting for our case.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
ACKed-by: NPeter Krempa <pkrempa@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Sometimes, we have a PCI address and not fully allocated
virPCIDevice and yet we still want to know its /dev/vfio/N path.
Introduce virPCIDeviceAddressGetIOMMUGroupDev() function exactly
for that.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>

Previous patches rendered some of 'cleanup' labels needless.
Drop them.
Signed-off-by: NMichal Privoznik <mprivozn@redhat.com>
Reviewed-by: NCole Robinson <crobinso@redhat.com>