The vCPUs are passed as devices with specific id only when CPU
hot-plug is enable at cold start.
So, we can't enable/disable allow-hotplug online as then vCPU hotplug
API will thrown errors not finding core id.
Not enforcing this could also lead to migration failure, as the QEMU
command line for the target VM could be made different than the one it
was actually running with, causing a crash of the target as Fiona
observed [0].
[0]: https://lists.proxmox.com/pipermail/pve-devel/2023-October/059434.html
Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
[ TL: Reflowed & expanded commit message ]
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Fix races with ACPI-suspended VMs which could wake up during migration
or during a suspend-mode backup.
Revert prevention, of ACPI-suspended VMs automatically resuming after
migration, introduced by 7ba974a682. The commit introduced a
potential problem that causes a suspended VM that wakes up during
migration to remain paused after the migration finishes.
This can be fixed once QEMU preserves the 'suspended' runstate during
migration (current patch on the qemu-devel list [0]) by checking for
the 'suspended' runstate on the target after migration.
Furthermore the commit increased the race window during the
preparation of a suspend-mode backup, when a suspended VM wakes up
between the vm_is_paused check in PVE::VZDump::QemuServer::prepare and
PVE::VZDump::QemuServer::qga_fs_freeze. This causes the code to skip
fs-freeze even if the VM has woken up, potentially leaving the file
system in an inconsistent state.
To prevent this, do not treat the suspended runstate as paused when
migrating or archiving a VM.
[0]: https://lists.nongnu.org/archive/html/qemu-devel/2023-08/msg05260.html
Signed-off-by: Filip Schauer <f.schauer@proxmox.com>
Reviewed-by: Fiona Ebner <f.ebner@proxmox.com>
[ TL: massage in Fiona's extra info into commit message ]
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Add checks for "suspended" and "prelaunch" runstates when checking
whether a VM is paused.
This fixes the following issues:
* ACPI-suspended VMs automatically resuming after migration
* Shutdown and reboot commands timing out instead of failing
immediately on suspended VMs
Signed-off-by: Filip Schauer <f.schauer@proxmox.com>
In preparation to add more properties to the memory configuration like
maximum hotpluggable memory and whether virtio-mem devices should be
used.
This also allows to get rid of the cyclic include of PVE::QemuServer
in the memory module.
Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
[FE: also convert new usage in get_derived_property
remove cyclic include of PVE::QemuServer
add commit message]
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
moving qemu_{device,object}{add,del} helpers there for now.
In preparation to remove the cyclic include of PVE::QemuServer in the
memory module and generally for better modularity in the future.
No functional change intended.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
which is the only user of the parse_numa() helper. While at it, avoid
the duplication of MAX_NUMA.
In preparation to remove the cyclic include of PVE::QemuServer in the
memory module.
No functional change intended.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
10 minutes is not long enough when disks are large and/or network
storages are used when preallocation is not disabled. The default is
metadata preallocation for qcow2, so there are still reports of the
issue [0][1]. If allocation really does not finish like the comment
describing the timeout feared, just let the user cancel it.
Also note that when restoring a PBS backup, there is no timeout for
disk allocation, and there don't seem to be any user complaints yet.
The 5 second timeout for receiving the config from vma is kept,
because certain corruptions in the VMA header can lead to the
operation hanging there.
There is no need for the $tmp variable before setting back the old
timeout, because that is at least one second, so we'll always be able
to set the $oldtimeout variable to undef in time in practice.
Currently, there shouldn't even be an outer timeout in the first
place, because the only call path leading to here is via the create
API (also used by qmrestore), both of which don't set a timeout.
[0]: https://forum.proxmox.com/threads/126825/
[1]: https://forum.proxmox.com/threads/128093/
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Skip the software TPM startup when starting a template VM for performing
a backup. This fixes an error that occurs when the TPM state disk is
write-protected.
Signed-off-by: Filip Schauer <f.schauer@proxmox.com>
Quoting from QEMU commit 4271f40383 ("virtio-net: correctly report
maximum tx_queue_size value"):
> Maximum value for tx_queue_size depends on the backend type.
> 1024 for vDPA/vhost-user, 256 for all the others.
> So the parameter is silently ignored and ethtool reports a different
> value than the one provided by the user.
Indeed, for a non-vDPA/vhost-user netdev, the guest will see TX: 256
instead of the specified 1024 here. With the mentioned QEMU commit (in
master and will be part of 8.1), using 1024 will be a hard error:
> Invalid tx_queue_size (= 1024), must be a power of 2 between 256 and 256
Since neither vhost-user, nor vhost-vdpa netdev types are exposed by
Proxmox VE, just changing the limit to the correct 256 should be fine.
No obvious issue during live-migration found.
Fixes: 620d6b32 ("virtio-net: increase defaults rx|tx-queue-size to 1024")
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
While the comment sated
> # order of precedence, filtered by whether storage supports it:
> # 1. explicit requested format
> # 2. format of current volume
> # 3. default format of storage
the code did not fall back to the default format in the case of remote
migration, because the format was already set and the code used
> $format //= $defFormat;
This made remote migration from dir with qcow2 to e.g. LVM-thin fail.
Move extracting the format from the volume name to the call side for
local migration. This allows the logic here to be much simpler.
Reported-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Suggested-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Previously, qemu_img_format() was called with the target storage's
$scfg and the source storage's volume name.
This mismatch should only be relevant for certain special kinds of
storage plugins:
- no path, but does support an additional QEMU image format besides
'raw', in short NPAF.
- no path, volume name can match QEMU_FORMAT_RE, in short NPVM.
Note that all integrated plugins are neither NPAF nor NPVM.
Note that for NPAF plugins, qemu_img_format() already always returns
'raw' because there is no path. It's a bit unlikely such a plugin
exists, because there were no bug reports about qemu_img_format()
misbehaving there yet.
Let's go through the cases:
- If source and target storage both have or don't have a path,
qemu_img_format($scfg, $volname) returns the same for both $scfg's.
- If source storage has a path, but target storage does not, the
format hint was previously 'raw', but can only be more correct now
(being what the source image actually is):
- For non-NPAF targets, since we know there is no path, it follows
that 'raw' is the only supported QEMU image format.
- For NPAF targets, the format will be preserved now (if actually
supported).
- If source storage does not have a path, but target storage does, the
format hint will be 'raw' now.
- For non-NPVM sources, QEMU_FORMAT_RE didn't match when
qemu_img_format() was called with the target storage's $scfg, so
the hint also was 'raw' before this commit.
- For NPVM sources, qemu_img_format() might've guessed a format from
the source volume name when called with the target's $scfg before
this commit. If the target storage supports the previously guessed
format, it was preserved before this commit, but will not be
anymore. In theory, the guess might've also been wrong, and in
this case, this commit avoids the wrong guess.
To summarize, there is only one edge case with an exotic kind of third
party storage plugin where format preservation would be lost and in
another edge case, format preservation is gained.
In preparation to simplify the format fallback logic implementation.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
The new ciupgrade option was missing in $cloudinitoptions in
PVE::API2::Qemu, so $check_vm_modify_config_perm defaulted to
requiring root@pam for modifying the option. To fix this, add
ciupgrade to $cloudinitoptions. This also fixes an issue where
ciupgrade was missing in the output of `qm cloudinit pending`,
as it also relies on $cloudinitoptions.
This issue was originally reported in the forum [0].
Also add a comment to avoid similar issues when adding new options in
the future.
[0]: https://forum.proxmox.com/threads/131043/
Signed-off-by: Friedrich Weber <f.weber@proxmox.com>
Only unit 0 for IDE is supported with machine type q35. Currently,
QEMU will fail startup with machine type q35 with an error like
> Can't create IDE unit 1, bus supports only 1 units
when ide1 or ide3 is configured.
Make sure to keep backwards compat form migration by leaving ide0 and
ide2 fixed. Since starting with ide1 or ide3 never worked, they can be
moved to a controller with a higher ID without issue.
Reported in the community forum:
https://forum.proxmox.com/threads/124615/post-543127https://forum.proxmox.com/threads/130815/
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Since we don't scan all storages for matching disk images anymore for a
migration we don't have any images found via storage alone. They will be
referenced in the config somewhere.
Therefore, there is no need for the 'storage' ref.
The 'referenced_in_config' is not really needed and can apply to both,
attached and unused disk images.
Therefore the QemuServer::foreach_volid() will change the
'referenced_in_config' attribute to an 'is_attached' one that only
applies to disk images that are in the _main_ config part and are not
unused.
In QemuMigrate::scan_local_volumes() we can then quite easily map the
refs to each state, attached, unused, referenced_in_{pending,snapshot}.
The refs are mostly used for informational use to print out in the logs
why a disk image is part of the migration. Except for the 'attached' case.
In the future the extra step of the refs in QemuMigrate could probably
be streamlined even more.
Signed-off-by: Aaron Lauterer <a.lauterer@proxmox.com>
All calling sites except for QemuConfig.pm::get_replicatable_volumes()
already enabled it. Making it the non-configurable default results in a
change in the VM replication. Now a disk image only referenced in the
pending section will also be replicated.
Signed-off-by: Aaron Lauterer <a.lauterer@proxmox.com>
Make it possible to optionally iterate over disks in the pending section
of VMs, similar as to how snapshots are handled already.
This is for example useful in the migration if we don't want to rely on
the scanning of all storages.
All calling sites are adapted and enable it, except for
QemuConfig::get_replicatable_volumes as that would cause a change for
the replication if pending disks would be included.
The following lists the calling sites and if they should be fine with
the change (source [0]):
1. QemuMigrate: scan_local_volumes(): needed to include pending disk
images
2. API2/Qemu.pm: check_vm_disks_local() for migration precondition:
related to migration, so more consistent with pending
3. QemuConfig.pm: get_replicatable_volumes(): would change the behavior
of the replication, will not use it for now.
4. QemuServer.pm: get_vm_volumes(): is used multiple times by:
4a. vm_stop_cleanup() to deactivate/unmap: should also be fine with
including pending
4b. QemuMigrate.pm: in prepare(): part of migration, so more consistent
with pending
4c. QemuMigrate.pm: in phase3_cleanup() for deactivation: part of
migration, so more consistent with pending
[0] https://lists.proxmox.com/pipermail/pve-devel/2023-May/056868.html
Signed-off-by: Aaron Lauterer <a.lauterer@proxmox.com>
Commit efa3355d ("fix #3428: cloudinit: add parameter for upgrade on
boot") changed the default, but this is a breaking change. The bug
report was only about making the option configurable.
The commit doesn't give an explicit reason for why, and arguably,
doing the upgrade is not an issue for most users. It also leads to a
different cloud-init instance ID, because of the different setting,
which in turn leads to ssh host key regeneration within the VM.
Reported-by: Friedrich Weber <f.weber@proxmox.com>
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
The scope can get into failed state for some issues like OOM kills of
the whole scope, in that case a user cannot re-start the VM until
they manually reset it.
Do this for now inline to avoid a pve-common bump as done in [0]
(location was suggested by me thinking we could maybe do it over
dbus, but as we have a stop command here already it probably doesn't
matters)
[0]: https://lists.proxmox.com/pipermail/pve-devel/2023-June/057770.html
Originally-by: Fiona Ebner <f.ebner@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
to allow early checking of the merged config, if the backup archive
passed in is a proper volume where extraction is possible.
Signed-off-by: Fabian Grünbichler <f.gruenbichler@proxmox.com>
by adding them to their own list, saving the nodes where they are not
allowed, and return those on 'wantarray' so we don't break existing
callers that don't expect it.
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
Tested-By: Markus Frank <m.frank@proxmox.com>
this patch allows configuring pci devices that are mapped via cluster
resource mapping when the user has 'Resource.Use' on the ACL path
'/mapping/pci/{ID}' (in addition to the usual required vm config
privileges)
When given multiple mappings in the config, we use them as alternatives
for the passthrough, and will select the first free one on startup.
It is using our regular pci reservation mechanism for regular devices and
we introduce a selection mechanism for mediated devices.
A few changes to the inner workings were required to make this work well:
* parse_hostpci now returns a different structure where we have a list
of lists (first level is for the different alternatives and second
level is for the different devices that should be passed through
together)
* factor out the 'parse_hostpci_devices' which parses each device from
the config and does some precondition checks
* reserve_pci_usage now behaves slightly different when trying to
reserve an device with the same VMID that's already reserved for,
since for checking which alternative we can use, we already must
reserve one (this means that qm showcmd can actually reserve devices,
albeit only for up to 10 seconds)
* configuring a mediated device on a multifunction device is not
supported anymore, and results in failure to start (previously, it
just chose the first device to do it). This is a breaking change
* configuring a single pci device twice on different hostpci slots now
fails during commandline generation instead on qemu start, so we had
to adapt one test where this occurred (it could never have worked
anyway)
* we now check permissions during clone/restore, meaning raw/real
devices can only be cloned/restored by root@pam from now on.
this is a breaking change.
Fixes#3574: Improve SR-IOV usability
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
Tested-By: Markus Frank <m.frank@proxmox.com>
this patch allows configuring usb devices that are mapped via
cluster resource mapping when the user has 'Mapping.Use' on the ACL
path '/mapping/usb/{ID}' (in addition to the usual required vm config
privileges)
for now, this is only valid if there is exactly one mapping for the
host, since we don't track passed through usb devices yet
This now also checks permissions on clone/restore, meaning a
'non-mapped' device can only be cloned/restored as root@pam user.
That is a breaking change.
Refactor the checks for restoring into a sub, so we have central place
where we can add such checks
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
Tested-By: Markus Frank <m.frank@proxmox.com>
similar to how we handle the PCI module and format. This makes the
'verify_usb_device' method and format unnecessary since
we simply check the format with a regex.
while doing tihs, i noticed that we don't correctly check for the
case-insensitive variant for 'spice' during hotplug, so fix that too
With this we can also remove some parameters from the get_usb_devices
and get_usb_controllers functions
while were at it, refactor the permission checks for the usb config too
and use the new 'my sub' style for the functions
also make print_usbdevice_full parse the device itself, so we don't have
to do it in multiple places (especially in places where we don't see
that this is needed)
No functional change intended
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
Tested-By: Markus Frank <m.frank@proxmox.com>
Using the word 'agent' is highly confusing here as there is no QMP
agent and thus wrongly suggests that the value is related to the
guest agent[0].
[0]: https://forum.proxmox.com/threads/123590/post-537716
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
This was not only rather inefficient (getting the config from the
archive twice) but also wrong, as we can override options on restore,
so we can do the check only when the backed-up config and override
config got merged.
If this is to late from POV of volume deletion or the like, then the
issue is that those things happen to early, as we can only know what
to do with the actual target config, so destructive actions that
happen before that are wrong by design.
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
for convenience. These options do not influence the QEMU instance
directly, but are only used for migration, so no need to keep them in
pending.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
instead use a recent example that served as a workaround in #4625.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
like the deprecation message printed by QEMU suggests.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
like the deprecation message printed by QEMU suggests.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Commit 7246e8f9 ("Set zero $size and continue if volume_resize()
returns false") mentions that this is needed for "some storages with
backing block devices to do online resize" and since this patch came
together [0] with pve-storage commit a4aee43 ("Fix RBD resize with
krbd option enabled."), it's safe to assume that RBD with krbd is
meant. But it should be the same situation for any external plugin
relying on the same behavior.
Other storages backed by block devices like LVM(-thin) and ZFS return
1 and the new size respectively, and the code is older than the above
mentioned commits. So really, the RBD plugin just should have returned
a positive value to be in-line with those and there should be no need
to pass 0 to the block_resize QMP command either.
Actually, it's a hack, because the block_resize QMP command does not
actually do special handling for the value 0. It's just that in the
case of a block device, QEMU won't try to resize it (and not fail for
shrinkage). But the size in the raw driver's BlockDriverState is
temporarily set to 0 (which is not nice), until the sector count is
refreshed, where raw_co_getlength is called, which queries the new
size and sets the size in the raw driver's BlockDriverState again as a
side effect. It's not known to cause any issues, but bdrv_getlength is
a coroutine wrapper starting from QEMU 8.0.0, and it's just better to
avoid setting a completely wrong value even temporarily. Just pass the
actually requested size like is done for LVM(thin) and ZFS.
[0]: https://lists.proxmox.com/pipermail/pve-devel/2017-January/025060.html
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Handle and warn about network interfaces which are not attached to
any bridge because the user actively removed it from the VM config.
Signed-off-by: Christian Ebner <c.ebner@proxmox.com>
With the recent pve-storage commit d70d814 ("api: fix get content call response
type for RBD/ZFS/iSCSI volumes"), the volume_size_info call for RBD in
list context is much slower than before (from a quick test, about twice as long
without snapshots, even longer with snapshots and untested, but when using an
external cluster with image not having the fast-diff feature, it should be worse
still) and thus increases the likelihood to run into timeouts here.
None of the callers here actually need the more expensive call, so just
avoid calling in list context.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
in some nvidia grid drivers (e.g. 14.4 and 15.x), their kernel module
tries to clean up the mdev device when the vm is shutdown and if it
cannot do that (e.g. becaues we already cleaned it up), their removal
process cancels with an error such that the vgpu does still exist inside
their book-keeping, but can't be used/recreated/freed until a reboot.
since there seems no obvious way to detect if thats the case besides
either parsing dmesg (which is racy), or the nvidia kernel module
version(which i'd rather not do), we simply test the pci device vendor
for nvidia and add a 10s sleep. that should give the driver enough time
to clean up and we will not find the path anymore and skip the cleanup.
This way, it works with both the newer and older versions of the driver
(some of the older drivers are LTS releases, so they're still
supported).
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
instead of using the mdev uuid. The nvidia driver does not actually care
that it's the same as the mdev, and in qemu the uuid parameter
overwrites the smbios1 uuid internally, so we should have been reusing
that in the first place.
Signed-off-by: Dominik Csapak <d.csapak@proxmox.com>
Respecting bandwidth limit for offline clone was implemented by commit
56d16f16 ("fix #4249: make image clone or conversion respect bandwidth
limit"). It's still not respected for EFI disks, but those are small,
so just ignore it.
Signed-off-by: Fiona Ebner <f.ebner@proxmox.com>
Previously, cloning a stopped VM didn't respect bwlimit. Passing the -r
(ratelimit) parameter to qemu-img convert fixes this issue.
Signed-off-by: Leo Nunner <l.nunner@proxmox.com>
[ T: reword subject line slightly ]
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Avoid pretending that a MTU change on a existing network device gets
applied live to a running VM.
Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
[ T: reworded and expanded commit message slightly ]
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
Required because there's one single efi for ARM, and the 2m/4m
difference doesn't seem to apply.
Signed-off-by: Matthias Heiserer <m.heiserer@proxmox.com>
[ T: move description to format and reword subject ]
Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>
