enable hotplug by default

Add link_down flag to network config

Original patch by Wolfgang, adopted for new hotplug implementation.

I do not verify link status, because that patch was rejected upstream.

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

Change check for better understanding!

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>

bump version to 3.3-10

Support additional e1000 variants for VM machines

It is required for OS X on qemu

Signed-off-by: Kamil Trzcinski <ayufan@ayufan.eu>

bump version to 3.3-9

pending api : fix parsing 0 value

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

fix test for ballon hotplug

set boot strict=on to prevent booting from not listed boot devices

Signed-off-by: Stefan Priebe <s.priebe@profihost.ag>

do not set romfile= if network is not in boot order

Signed-off-by: Stefan Priebe <s.priebe@profihost.ag>

bump version to 3.3-8

fix previous commit: forgot API definitions

update_vm api : add optionnal revert option

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

code cleanup

fix balloon pending update

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

don't try to hotplug|unplug ide && sata drive

We can't hotplug theses devices,
so we can avoid to throw always an error when adding them

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

add optionnal current param to config api

config api return by default the pending config, like this it don't break current behaviour

Add an optionnal current param to get to current running config

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

API: list new 'pending' method

pending : add disk hot-unplug

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

remove unused code

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vmconfig_update_net: do not call vm_deviceplug() if hotplug == 0

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vmconfig_hotplug_pending : add update_disk

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vm_deviceplug: always raise exception on error

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vm_deviceunplug: raise expection if something fail

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vmconfig_hotplug_pending : add update_net

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vmconfig_hotplug_pending: correctly skip values

Do not use $skip variable (simply raise an exception)

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

rename qemu_bridgeadd to qemu_add_pci_bridge

To make it obvious that we add a PCI device, and not a network bridge.

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

implement API/CLI to get pending changes

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vmconfig_hotplug_pending: improve hotplug error handling

Simplify code, and allow to partially apply pending changes using
a new $selection parameter.

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

code cleanup, delete trailing white space

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vmconfig_cleanup_pending: new method to clenup setting in [PENDING]

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vmconfig_hotplug_pending: implement tablet hotplug

Remove special case: now, we only hotplug if 'hotplug' is enabled.

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

implement trivial hotplug

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

fix balloon consistency check (consider pending changes)

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

vm_start: apply pending changes

I move related helper methods into PVE::QemuServer.

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

implement vmconfig_apply_pending for stopped VM

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

update_vm_api: always write into pending section

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

write_vm_config : write pending change

example:

$conf->{pending}->{virtio1}
$conf->{pending}->{delete} = "net0,net1"

[PENDING]
virtio1: ...
delete: net0,net1

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

parse_vm_config: only allow 'delete' inside [PENDING]

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

parse_vm_config: correctly handle $descr

Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

parse_vm_config : parse pending changes

example:

[PENDING]
virtio1:...
delete:net0,net1

$conf->{pending}->{virtio1}
$conf->{pending}->{del}->{net0}
$conf->{pending}->{del}->{net1}

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

bump version to 3.3-7

this commit:

commit 1c0c1c17b020710fd841c399c1e147ccc10007cd
Author: Wolfgang Link <wolfgang@linksystems.org>
Date:   Wed Nov 26 11:11:40 2014 +0100

    shutdown by Qemu Guest Agent if the agent flag in the config is set

    Important: "guest-shutdown" returns only by error a message.

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>
breaks live migration as it always tries to load the vm config - even in case of $nocheck. Also it double loads the config ($conf && $config)
Signed-off-by: Stefan Priebe <s.priebe@profihost.ag>

white space cleanups

Add qga freeze in vzdump in snapshot mode

it will freeze the filesystem to provide consistents.

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>

white space cleanups

add custom numa topology support

numaX: cpus=<id[-id],memory=<mb>[[,hostnodes=<id[-id]>][,policy=<preferred|bind|interleave>]]

example:
-------
sockets:4
cores:2
memory:4096
numa: 1
numa0: cpus=0-1,memory=1024,hostnodes=0-1,policy=interleave
numa1: cpus=2-3,memory=3072,hostnodes=2,policy=bind

qemu command line
-----------------
-object memory-backend-ram,size=1024M,policy=interleave,host-nodes=0-1,id=ram-node0
-numa node,nodeid=0,cpus=0-1,memdev=ram-node0

-object memory-backend-ram,size=3072M,policy=bind,host-nodes=2,id=ram-node1
-numa node,nodeid=1,cpus=2-3,memdev=ram-node1

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

add numa options v3

This enable numa support inside the guest, and share the memory and cores across the sockets numa nodes.

numa: 0|1

example:
-------
sockets:2
cores:2
memory:4096
numa: 1

qemu command line
-----------------
-object memory-backend-ram,size=2048,id=ram-node0
-numa node,nodeid=0,cpus=0-1,memdev=ram-node0
-object memory-backend-ram,size=2048,id=ram-node1
-numa node,nodeid=1,cpus=2-3,memdev=ram-node1

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.3-6

Fix in PVE::QemuServer::snapshot_create

remove the freezefs flag.
If Qemu Guest Agent flag is set in config the vm filesystem will always be frozen,
unless we save RAM.

also remove param freezefs in PVE::API2 snapshot,
because there is no use for it.

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

snapshot_create: fix bug from prev. commit, consider $freezefs parameter

snapshot_create: use guest-fsfreeze-freeze if possible

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

qmpclient: use guest-sync-delimited

bump version to 3.3-5

drive-mirror: avoid division by zero bug

qmpclient: fix mux_input regex

qmpclient: improve error handling

simplify code by using vm_qmp_command() directly

qmpclient: code cleanup, execute qga and qmp in parallel

shutdown by Qemu Guest Agent if the agent flag in the config is set

Important: "guest-shutdown" returns only by error a message.

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>

qmpclient: now if the QMP command starts with guest-+ , it will bind dynamicly to the VMID.qga socket

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>

qmpclient-qga : mux_input : parse qga result

result sample:

first json is guest-sync result, second json is command result

{ "return": 123456}\n{"return": {}}

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

qmpclient-qga : build qga command

example of command:

first json is guest-sync to sync and flush the client, second json is the command

{ "execute": "guest-sync", "arguments": { "id": 123456 } }{"execute":"guest-ping"}

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

qmpclient-qga : do not sent qmp_capabilities for qga

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

qmpclient-qga : cmdid : use integer instead string

qga client use only integer as id

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

qmpclient-qga : add qga param to qmp_socket

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

qmpclient-qga : add qga option at object creation

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

add vm_qga_command

and reuse vm_qmp_command && qmp_socket with $qga param

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

code cleanups

savevm-end : wait that savevm is finished

savevm-end is async, we need to wait that savevm is finished, before continue.

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

serial: allow to pass arbitrary device names

Add check if host has enough real CPUs for starting VM

To prevent a Qemu CPU emualtion!

Signed-off-by: Wolfgang Link <wolfgang@linksystems.org>

qemu_drive_mirror: cleanup, avoid code duplication

qemu_drive_mirror: raise exception if we cannot parse $dst_volid

bump version to 3.3-3

delete trailing whitespace

block-job-complete : retry if block job cannot be complete

Even if we check the busy flag, we can have sometime race condition if new write
are coming between the query-block-job and the block-job-complete.

block-job-complete throw an error "The active block job for device '%(name)' cannot be completed"

we just need to retry in this case.

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.3-2

cleanup: do not entry wait loop if block-job-cancel() fails.

drive-mirror : wait that busy eq false before block-job-complete

When the drive-mirror is at 100%, and write occurs, the busy flag can change from false->true

- 100% no new writes

transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: false
transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: false
transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: false

- 100% new writes

transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: true
transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: true
transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: true

- 100% no new writes

transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: false
transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: false
transferred: 1073741824 bytes remaining: 0 bytes total: 1073741824 bytes progression: 100.00 % busy: false

So, we need to check that busy is false before doing the block-job-complete.

Also, we force the vm to pause, if it's busy for more than 120s when drive-mirror is at 100%

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

block-job-cancel : wait that jobs is really finished

block-job-cancel is async, we need to check that job is really finished

before try to free the volume

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.3-1

enable write zeroes optimisations

This enable write zeroes optimisation inside guest.

(I have tested them with qcow2,raw (ext4|xfs) , zfs|iscsi and nfs too.

No impact if the block driver don't support it yet (like rbd)

https://lists.nongnu.org/archive/html/qemu-devel/2014-04/msg00009.html

> a) mkfs.ext4 -E lazy_itable_init=0,lazy_journal_init=0 /dev/vdX
>
> QCOW2         [off]     [on]     [unmap]
> -----
> runtime:       14secs    1.1secs  1.1secs
> filesize:      937M      18M      18M
>
> iSCSI         [off]     [on]     [unmap]
> ----
> runtime:       9.3s      0.9s     0.9s
>
> b) dd if=/dev/zero of=/dev/vdX bs=1M oflag=direct
>
> QCOW2         [off]     [on]     [unmap]
> -----
> runtime:       246secs   18secs   18secs
> filesize:      51G       192K     192K
> throughput:    203M/s    2.3G/s   2.3G/s
>
> iSCSI*        [off]     [on]     [unmap]
> ----
> runtime:       8mins     45secs   33secs
> throughput:    106M/s    1.2G/s   1.6G/s
> allocated:     100%      100%     0%

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

add iothread/dataplane support

new config option:

iothread: 1|0

This enable iothread/dataplane support, to improve io performance on fast storages

Currently block jobs don't work yet, it's planned for qemu 2.2.
So it's better to not expose yet this option in gui.

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.1-35

fix bug #542: return VMID as integer

bump version to 3.1-34

vm_devices_list : also list block devices

This allow scsi disk to be plug|unplug

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

Currently,if we don't have a "machine" option in running config, and we take a vmstate snapshot

the machine option is write in the snapshot (ok), but also in the running config (bad)

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>
Signed-off-by: Dietmar Maurer <dietmar@proxmox.com>

allow hotplug of virtio-scsi disks

It was an old protection, it's working fine now

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.1-33

clone_vm: auto generate new uuid

bump version to 3.1-32

add Broadwell cpu model

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.1-31

generate spice devices cmd line after pci bridge

we should push to $devices array instead $cmd array,

because pci bridges need to be create before spice devices

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.1-30

bump max hostpci to 4

user need to passthough 3 devices here
http://forum.proxmox.com/threads/19134-KVM-passtrought-PciExpress-3-card

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

vga=none if x-vga passthrough is enabled

we need to disable virtual vga card if we passthrough a physical gpu

Signed-off-by: Alexandre Derumier <aderumier@odiso.com>

bump version to 3.1-29