[PATCH v2v 0/5] convert: Find out if Windows guest is expecting BIOS localtime or UTC
[Alice: See patch 2] [This patch is a bit rough, it could do with better commit messages and some tests. Please test it to see if it solves the Windows conversion issue described in the thread below.] We currently do not set any <clock/> field in guest output. Most Windows guests expect the BIOS to be set to localtime, whereas almost all Linux guests would expect it to be set to UTC. It is also possible to configure a Windows guest to expect BIOS set to UTC. The default is usually BIOS set to UTC, so for many Windows guests this would be wrong. This specifically may cause problems when scheduling qemu-ga installation, see the thread here: https://listman.redhat.com/archives/libguestfs/2023-September/thread.html#32... but could cause other general issues with time in the guest. One way to implement this would be to copy the source hypervisor information across; however I'm not confident this information is read correctly. A better way is to read out what the guest is expecting from the Windows registry. (For Linux we just assume BIOS is always UTC, since that's the default for almost any Linux guest which hasn't been dual-booted with Windows, which for VMs would be incredibly rare.) Rich.
This change simply adds the flag to guestcaps, and sets it to always true, so there is no change to the output. --- lib/types.mli | 5 +++++ convert/convert_linux.ml | 1 + convert/convert_windows.ml | 1 + lib/types.ml | 3 +++ 4 files changed, 10 insertions(+) diff --git a/lib/types.mli b/lib/types.mli index 65ef2e35cf..44b35faeb0 100644 --- a/lib/types.mli +++ b/lib/types.mli @@ -277,6 +277,11 @@ type guestcaps = { gcaps_virtio_1_0 : bool; (** The guest supports the virtio devices that it does at the virtio-1.0 protocol level. *) + + gcaps_bios_utc : bool; + (** Is the BIOS set to UTC ([true]) or localtime ([false])? For + Linux guests this is always true. For Windows we find out + what the guest is expecting by looking at the registry. *) } (** Guest capabilities after conversion. eg. Was virtio found or installed? *) diff --git a/convert/convert_linux.ml b/convert/convert_linux.ml index 8d7020848f..a70d43224c 100644 --- a/convert/convert_linux.ml +++ b/convert/convert_linux.ml @@ -220,6 +220,7 @@ let convert (g : G.guestfs) source inspect i_firmware _ keep_serial_console _ = gcaps_arch = Utils.kvm_arch inspect.i_arch; gcaps_arch_min_version = arch_min_version; gcaps_virtio_1_0 = virtio_1_0; + gcaps_bios_utc = true; (* almost all Linux expect BIOS to be UTC *) } in guestcaps diff --git a/convert/convert_windows.ml b/convert/convert_windows.ml index 122d95469d..f6e039be7e 100644 --- a/convert/convert_windows.ml +++ b/convert/convert_windows.ml @@ -275,6 +275,7 @@ let convert (g : G.guestfs) _ inspect i_firmware block_driver _ static_ips = gcaps_arch = Utils.kvm_arch inspect.i_arch; gcaps_arch_min_version = 0; gcaps_virtio_1_0 = virtio_win_installed.Inject_virtio_win.virtio_1_0; + gcaps_bios_utc = true; } in guestcaps diff --git a/lib/types.ml b/lib/types.ml index 75c14fd4f6..0f659ebd9a 100644 --- a/lib/types.ml +++ b/lib/types.ml @@ -399,6 +399,7 @@ type guestcaps = { gcaps_arch : string; gcaps_arch_min_version : int; gcaps_virtio_1_0 : bool; + gcaps_bios_utc : bool; } and guestcaps_block_type = Virtio_blk | Virtio_SCSI | IDE and guestcaps_net_type = Virtio_net | E1000 | RTL8139 @@ -429,6 +430,7 @@ let string_of_guestcaps gcaps = gcaps_arch = %s\n\ gcaps_arch_min_version = %d\n\ gcaps_virtio_1_0 = %b\n\ + gcaps_bios_utc = %b\n\ " (string_of_block_type gcaps.gcaps_block_bus) (string_of_net_type gcaps.gcaps_net_bus) @@ -440,6 +442,7 @@ let string_of_guestcaps gcaps = gcaps.gcaps_arch gcaps.gcaps_arch_min_version gcaps.gcaps_virtio_1_0 + gcaps.gcaps_bios_utc type target_buses = { target_virtio_blk_bus : target_bus_slot array; -- 2.41.0
--- output/create_kubevirt_yaml.ml | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/output/create_kubevirt_yaml.ml b/output/create_kubevirt_yaml.ml index 689555e4dc..b96c742832 100644 --- a/output/create_kubevirt_yaml.ml +++ b/output/create_kubevirt_yaml.ml @@ -54,6 +54,11 @@ let create_kubevirt_yaml source inspect "pit", Assoc [ "tickPolicy", String "delay" ]; "rtc", Assoc [ "tickPolicy", String "catchup" ]; ]; + (* XXX Note that we may need to set "localtime" here + * depending on guestcaps.gcaps_bios_utc. However that + * requires the following PR to be merged in Kubevirt: + * https://github.com/kubevirt/kubevirt/pull/9587 + *) "utc", List [] ] ) -- 2.41.0
--- output/create_libvirt_xml.ml | 10 ++++++++-- tests/test-v2v-i-ova.xml | 1 + 2 files changed, 9 insertions(+), 2 deletions(-) diff --git a/output/create_libvirt_xml.ml b/output/create_libvirt_xml.ml index 964acd25fd..f97272ca31 100644 --- a/output/create_libvirt_xml.ml +++ b/output/create_libvirt_xml.ml @@ -292,10 +292,16 @@ let create_libvirt_xml ?pool source inspect e "nvram" ["template", vars_template] [] ] in List.push_back_list os loader; - !os in + e "os" [] !os in + + (* The <clock> section. *) + let clock_section = + let offset = if guestcaps.gcaps_bios_utc then "utc" else "localtime" in + e "clock" [ "offset", offset ] [] in List.push_back_list body [ - e "os" [] os_section; + os_section; + clock_section; e "on_poweroff" [] [PCData "destroy"]; e "on_reboot" [] [PCData "restart"]; diff --git a/tests/test-v2v-i-ova.xml b/tests/test-v2v-i-ova.xml index e5907ea1cc..a41827bfd5 100644 --- a/tests/test-v2v-i-ova.xml +++ b/tests/test-v2v-i-ova.xml @@ -18,6 +18,7 @@ <os> <type arch='x86_64' machine='q35'>hvm</type> </os> + <clock offset='utc'/> <on_poweroff>destroy</on_poweroff> <on_reboot>restart</on_reboot> <on_crash>restart</on_crash> -- 2.41.0
On 9/25/23 16:04, Richard W.M. Jones wrote:
--- output/create_libvirt_xml.ml | 10 ++++++++-- tests/test-v2v-i-ova.xml | 1 + 2 files changed, 9 insertions(+), 2 deletions(-)
diff --git a/output/create_libvirt_xml.ml b/output/create_libvirt_xml.ml index 964acd25fd..f97272ca31 100644 --- a/output/create_libvirt_xml.ml +++ b/output/create_libvirt_xml.ml @@ -292,10 +292,16 @@ let create_libvirt_xml ?pool source inspect e "nvram" ["template", vars_template] [] ] in
List.push_back_list os loader; - !os in + e "os" [] !os in + + (* The <clock> section. *) + let clock_section = + let offset = if guestcaps.gcaps_bios_utc then "utc" else "localtime" in + e "clock" [ "offset", offset ] [] in
List.push_back_list body [ - e "os" [] os_section; + os_section; + clock_section;
e "on_poweroff" [] [PCData "destroy"]; e "on_reboot" [] [PCData "restart"];
I don't think we need to change what "os_section" is bound to, here; I understand the idea is to increase consistency, but for me it makes the patch harder to read -- ultimately it is a refactoring, and then adding in clock_section is the new thing. If you can split these apart, that's optimal; if not, this is viable too IMO. Thanks Laszlo
diff --git a/tests/test-v2v-i-ova.xml b/tests/test-v2v-i-ova.xml index e5907ea1cc..a41827bfd5 100644 --- a/tests/test-v2v-i-ova.xml +++ b/tests/test-v2v-i-ova.xml @@ -18,6 +18,7 @@ <os> <type arch='x86_64' machine='q35'>hvm</type> </os> + <clock offset='utc'/> <on_poweroff>destroy</on_poweroff> <on_reboot>restart</on_reboot> <on_crash>restart</on_crash>
On Mon, Sep 25, 2023 at 05:47:26PM +0200, Laszlo Ersek wrote:
On 9/25/23 16:04, Richard W.M. Jones wrote:
--- output/create_libvirt_xml.ml | 10 ++++++++-- tests/test-v2v-i-ova.xml | 1 + 2 files changed, 9 insertions(+), 2 deletions(-)
diff --git a/output/create_libvirt_xml.ml b/output/create_libvirt_xml.ml index 964acd25fd..f97272ca31 100644 --- a/output/create_libvirt_xml.ml +++ b/output/create_libvirt_xml.ml @@ -292,10 +292,16 @@ let create_libvirt_xml ?pool source inspect e "nvram" ["template", vars_template] [] ] in
List.push_back_list os loader; - !os in + e "os" [] !os in + + (* The <clock> section. *) + let clock_section = + let offset = if guestcaps.gcaps_bios_utc then "utc" else "localtime" in + e "clock" [ "offset", offset ] [] in
List.push_back_list body [ - e "os" [] os_section; + os_section; + clock_section;
e "on_poweroff" [] [PCData "destroy"]; e "on_reboot" [] [PCData "restart"];
I don't think we need to change what "os_section" is bound to, here; I understand the idea is to increase consistency, but for me it makes the patch harder to read -- ultimately it is a refactoring, and then adding in clock_section is the new thing.
If you can split these apart, that's optimal; if not, this is viable too IMO.
Yup, that is a separate clean-up, will split. Rich.
Thanks Laszlo
diff --git a/tests/test-v2v-i-ova.xml b/tests/test-v2v-i-ova.xml index e5907ea1cc..a41827bfd5 100644 --- a/tests/test-v2v-i-ova.xml +++ b/tests/test-v2v-i-ova.xml @@ -18,6 +18,7 @@ <os> <type arch='x86_64' machine='q35'>hvm</type> </os> + <clock offset='utc'/> <on_poweroff>destroy</on_poweroff> <on_reboot>restart</on_reboot> <on_crash>restart</on_crash>
-- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com Fedora Windows cross-compiler. Compile Windows programs, test, and build Windows installers. Over 100 libraries supported. http://fedoraproject.org/wiki/MinGW
I didn't set the -rtc flag in the normal (UTC) case as that is the default. --- output/output_qemu.ml | 2 ++ 1 file changed, 2 insertions(+) diff --git a/output/output_qemu.ml b/output/output_qemu.ml index 57b1e58da2..a9d7b37071 100644 --- a/output/output_qemu.ml +++ b/output/output_qemu.ml @@ -158,6 +158,8 @@ module QEMU = struct arg_list "-device" ["vmgenid"; sprintf "guid=%s" genid; "id=vmgenid0"] ); + if not guestcaps.gcaps_bios_utc then arg "-rtc" "base=localtime"; + arg_list "-machine" (machine_str :: (if smm then ["smm=on"] else []) @ ["accel=kvm:tcg"]); -- 2.41.0
Read HKLM\SYSTEM\CurrentControlSet\Control\TimeZoneInformation key "RealTimeIsUniversal" to see if the Windows guest is expecting BIOS set to localtime (not present) or UTC (present and set to 1). See: https://wiki.archlinux.org/title/System_time#UTC_in_Microsoft_Windows See: https://listman.redhat.com/archives/libguestfs/2023-September/thread.html#32... Reported-by: Lee Garrett --- convert/convert_windows.ml | 38 +++++++++++++++++++++++++++++++++++++- tests/test-v2v-i-ova.xml | 2 +- 2 files changed, 38 insertions(+), 2 deletions(-) diff --git a/convert/convert_windows.ml b/convert/convert_windows.ml index f6e039be7e..84e8f7b7d3 100644 --- a/convert/convert_windows.ml +++ b/convert/convert_windows.ml @@ -103,6 +103,42 @@ let convert (g : G.guestfs) _ inspect i_firmware block_driver _ static_ips = (* If the Windows guest has AV installed. *) let has_antivirus = Windows.detect_antivirus inspect in + (* Does the guest expect the BIOS to be set to UTC or localtime? + * See https://wiki.archlinux.org/title/System_time#UTC_in_Microsoft_Windows + * Note this might be a QWORD on 64 bit Windows instances. + *) + let bios_utc = + Registry.with_hive_readonly g inspect.i_windows_system_hive + (fun reg -> + try + let key_path = [ "Control"; "TimeZoneInformation" ] in + let path = inspect.i_windows_current_control_set :: key_path in + let node = + match Registry.get_node reg path with + | None -> raise Not_found + | Some node -> node in + let valueh = g#hivex_node_get_value node "RealTimeIsUniversal" in + if valueh = 0L then raise Not_found; + let t = g#hivex_value_type valueh in + let data = g#hivex_value_value valueh in + let is_utc = + match t with + | 0_L (* REG_NONE *) -> false (* localtime *) + | 4_L (* REG_DWORD *) -> data = "\001\000\000\000" + | 11_L (* REG_QWORD *) -> data = "\001\000\000\000\000\000\000\000" + | _ (* who knows ... *) -> + warning (f_"unknown CurrentControlSet\\Control\\\ + TimeZoneInformation key RealTimeIsUniversal \ + type 0x%Lx, assuming BIOS set to UTC") t; + true in + is_utc + with Not_found -> + (* If the key is not found then by default we assume + * that Windows is expecting the BIOS to be set to localtime. + *) + false + ) in + (* Open the software hive (readonly) and find the Xen PV uninstaller, * if it exists. *) @@ -275,7 +311,7 @@ let convert (g : G.guestfs) _ inspect i_firmware block_driver _ static_ips = gcaps_arch = Utils.kvm_arch inspect.i_arch; gcaps_arch_min_version = 0; gcaps_virtio_1_0 = virtio_win_installed.Inject_virtio_win.virtio_1_0; - gcaps_bios_utc = true; + gcaps_bios_utc = bios_utc; } in guestcaps diff --git a/tests/test-v2v-i-ova.xml b/tests/test-v2v-i-ova.xml index a41827bfd5..fa7b4dbfc5 100644 --- a/tests/test-v2v-i-ova.xml +++ b/tests/test-v2v-i-ova.xml @@ -18,7 +18,7 @@ <os> <type arch='x86_64' machine='q35'>hvm</type> </os> - <clock offset='utc'/> + <clock offset='localtime'/> <on_poweroff>destroy</on_poweroff> <on_reboot>restart</on_reboot> <on_crash>restart</on_crash> -- 2.41.0
On 9/25/23 16:04, Richard W.M. Jones wrote:
Read HKLM\SYSTEM\CurrentControlSet\Control\TimeZoneInformation key "RealTimeIsUniversal" to see if the Windows guest is expecting BIOS set to localtime (not present) or UTC (present and set to 1).
See: https://wiki.archlinux.org/title/System_time#UTC_in_Microsoft_Windows See: https://listman.redhat.com/archives/libguestfs/2023-September/thread.html#32... Reported-by: Lee Garrett --- convert/convert_windows.ml | 38 +++++++++++++++++++++++++++++++++++++- tests/test-v2v-i-ova.xml | 2 +- 2 files changed, 38 insertions(+), 2 deletions(-)
diff --git a/convert/convert_windows.ml b/convert/convert_windows.ml index f6e039be7e..84e8f7b7d3 100644 --- a/convert/convert_windows.ml +++ b/convert/convert_windows.ml @@ -103,6 +103,42 @@ let convert (g : G.guestfs) _ inspect i_firmware block_driver _ static_ips = (* If the Windows guest has AV installed. *) let has_antivirus = Windows.detect_antivirus inspect in
+ (* Does the guest expect the BIOS to be set to UTC or localtime? + * See https://wiki.archlinux.org/title/System_time#UTC_in_Microsoft_Windows + * Note this might be a QWORD on 64 bit Windows instances. + *) + let bios_utc = + Registry.with_hive_readonly g inspect.i_windows_system_hive + (fun reg -> + try + let key_path = [ "Control"; "TimeZoneInformation" ] in + let path = inspect.i_windows_current_control_set :: key_path in + let node = + match Registry.get_node reg path with + | None -> raise Not_found + | Some node -> node in + let valueh = g#hivex_node_get_value node "RealTimeIsUniversal" in + if valueh = 0L then raise Not_found; + let t = g#hivex_value_type valueh in + let data = g#hivex_value_value valueh in + let is_utc = + match t with + | 0_L (* REG_NONE *) -> false (* localtime *) + | 4_L (* REG_DWORD *) -> data = "\001\000\000\000" + | 11_L (* REG_QWORD *) -> data = "\001\000\000\000\000\000\000\000" + | _ (* who knows ... *) -> + warning (f_"unknown CurrentControlSet\\Control\\\ + TimeZoneInformation key RealTimeIsUniversal \ + type 0x%Lx, assuming BIOS set to UTC") t; + true in
[*]
+ is_utc + with Not_found -> + (* If the key is not found then by default we assume + * that Windows is expecting the BIOS to be set to localtime. + *) + false + ) in + (* Open the software hive (readonly) and find the Xen PV uninstaller, * if it exists. *)
[*] Just guessing, but I'm not sure if assuming UTC is right when the key is found but its type is unrecognized, in light of determining localtime when the key is absent or has type REG_NONE. Either way, the warning is honest about it!
@@ -275,7 +311,7 @@ let convert (g : G.guestfs) _ inspect i_firmware block_driver _ static_ips = gcaps_arch = Utils.kvm_arch inspect.i_arch; gcaps_arch_min_version = 0; gcaps_virtio_1_0 = virtio_win_installed.Inject_virtio_win.virtio_1_0; - gcaps_bios_utc = true; + gcaps_bios_utc = bios_utc; } in
guestcaps diff --git a/tests/test-v2v-i-ova.xml b/tests/test-v2v-i-ova.xml index a41827bfd5..fa7b4dbfc5 100644 --- a/tests/test-v2v-i-ova.xml +++ b/tests/test-v2v-i-ova.xml @@ -18,7 +18,7 @@ <os> <type arch='x86_64' machine='q35'>hvm</type> </os> - <clock offset='utc'/> + <clock offset='localtime'/> <on_poweroff>destroy</on_poweroff> <on_reboot>restart</on_reboot> <on_crash>restart</on_crash>
With the BIOS -> RTC typo (?) fixed everywhere: series Reviewed-by: Laszlo Ersek <lersek@redhat.com>
On Mon, Sep 25, 2023 at 05:58:00PM +0200, Laszlo Ersek wrote:
On 9/25/23 16:04, Richard W.M. Jones wrote:
Read HKLM\SYSTEM\CurrentControlSet\Control\TimeZoneInformation key "RealTimeIsUniversal" to see if the Windows guest is expecting BIOS set to localtime (not present) or UTC (present and set to 1).
See: https://wiki.archlinux.org/title/System_time#UTC_in_Microsoft_Windows See: https://listman.redhat.com/archives/libguestfs/2023-September/thread.html#32... Reported-by: Lee Garrett --- convert/convert_windows.ml | 38 +++++++++++++++++++++++++++++++++++++- tests/test-v2v-i-ova.xml | 2 +- 2 files changed, 38 insertions(+), 2 deletions(-)
diff --git a/convert/convert_windows.ml b/convert/convert_windows.ml index f6e039be7e..84e8f7b7d3 100644 --- a/convert/convert_windows.ml +++ b/convert/convert_windows.ml @@ -103,6 +103,42 @@ let convert (g : G.guestfs) _ inspect i_firmware block_driver _ static_ips = (* If the Windows guest has AV installed. *) let has_antivirus = Windows.detect_antivirus inspect in
+ (* Does the guest expect the BIOS to be set to UTC or localtime? + * See https://wiki.archlinux.org/title/System_time#UTC_in_Microsoft_Windows + * Note this might be a QWORD on 64 bit Windows instances. + *) + let bios_utc = + Registry.with_hive_readonly g inspect.i_windows_system_hive + (fun reg -> + try + let key_path = [ "Control"; "TimeZoneInformation" ] in + let path = inspect.i_windows_current_control_set :: key_path in + let node = + match Registry.get_node reg path with + | None -> raise Not_found + | Some node -> node in + let valueh = g#hivex_node_get_value node "RealTimeIsUniversal" in + if valueh = 0L then raise Not_found; + let t = g#hivex_value_type valueh in + let data = g#hivex_value_value valueh in + let is_utc = + match t with + | 0_L (* REG_NONE *) -> false (* localtime *) + | 4_L (* REG_DWORD *) -> data = "\001\000\000\000" + | 11_L (* REG_QWORD *) -> data = "\001\000\000\000\000\000\000\000" + | _ (* who knows ... *) -> + warning (f_"unknown CurrentControlSet\\Control\\\ + TimeZoneInformation key RealTimeIsUniversal \ + type 0x%Lx, assuming BIOS set to UTC") t; + true in
[*]
+ is_utc + with Not_found -> + (* If the key is not found then by default we assume + * that Windows is expecting the BIOS to be set to localtime. + *) + false + ) in + (* Open the software hive (readonly) and find the Xen PV uninstaller, * if it exists. *)
[*] Just guessing, but I'm not sure if assuming UTC is right when the key is found but its type is unrecognized, in light of determining localtime when the key is absent or has type REG_NONE.
Either way, the warning is honest about it!
I don't know - I would need to examine lots of Windows machines to find out what they really do. But I expect that this field only ever contains dword:1 or hex(b):1 [qword:1], or is absent.
@@ -275,7 +311,7 @@ let convert (g : G.guestfs) _ inspect i_firmware block_driver _ static_ips = gcaps_arch = Utils.kvm_arch inspect.i_arch; gcaps_arch_min_version = 0; gcaps_virtio_1_0 = virtio_win_installed.Inject_virtio_win.virtio_1_0; - gcaps_bios_utc = true; + gcaps_bios_utc = bios_utc; } in
guestcaps diff --git a/tests/test-v2v-i-ova.xml b/tests/test-v2v-i-ova.xml index a41827bfd5..fa7b4dbfc5 100644 --- a/tests/test-v2v-i-ova.xml +++ b/tests/test-v2v-i-ova.xml @@ -18,7 +18,7 @@ <os> <type arch='x86_64' machine='q35'>hvm</type> </os> - <clock offset='utc'/> + <clock offset='localtime'/> <on_poweroff>destroy</on_poweroff> <on_reboot>restart</on_reboot> <on_crash>restart</on_crash>
With the BIOS -> RTC typo (?) fixed everywhere:
series Reviewed-by: Laszlo Ersek <lersek@redhat.com>
Thanks, I'll post v2 anyway since I've written it now. Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org
Hi Rich, On Mon, Sep 25, 2023 at 4:10 PM Richard W.M. Jones <rjones@redhat.com> wrote:
[Alice: See patch 2]
I'm not 100% sure about the source of this work. However, we had in KubeVirt people interested in using localtime with Windows [1]. Yes, I see that you pointed to that PR in patch 2. The problem with that PR is the migration. What happens if we migrate the VM to a host that is in another timezone? Is it going to break the application running inside the VM? IMO, that was the major blocker of that PR. I initially suggested that we either avoid migration as a first step or try to migrate only on nodes at the same timezone, Do you know if anyone has experience on this particular problem? [1] https://github.com/kubevirt/kubevirt/pull/9587 Many thanks, Alice
On Mon, Sep 25, 2023 at 04:26:59PM +0200, Alice Frosi wrote:
Hi Rich,
On Mon, Sep 25, 2023 at 4:10 PM Richard W.M. Jones <rjones@redhat.com> wrote:
[Alice: See patch 2]
I'm not 100% sure about the source of this work. However, we had in KubeVirt people interested in using localtime with Windows [1]. Yes, I see that you pointed to that PR in patch 2. The problem with that PR is the migration. What happens if we migrate the VM to a host that is in another timezone? Is it going to break the application running inside the VM?
Does every Windows VM running in Kubevirt have its registry adjusted so it expects the BIOS to be set to UTC? (ie: https://wiki.archlinux.org/title/System_time#UTC_in_Microsoft_Windows). This is a non-standard setting, but I can't understand how Windows VMs would have correct clocks if this were not the case. BTW you could find out by running: $ virt-win-reg window.img 'HKLM\SYSTEM' on some Kubevirt Windows VMs, and looking for: [\ControlSet001\Control\TimeZoneInformation] in the output. See if there is a key called "RealTimeIsUniversal" and what it is set to. If it is the case that this is always set for existing Kubevirt guests, then virt-v2v would have to be changed so it also sets this registry entry when the output is Kubevirt. I agree that migration across timezones is highly problematic for the BIOS = localtime situation. I don't see any solution for that except disallowing it.
IMO, that was the major blocker of that PR. I initially suggested that we either avoid migration as a first step or try to migrate only on nodes at the same timezone, Do you know if anyone has experience on this particular problem?
I guess RHV must have encountered this before, but I don't know any specifics. Thanks, Rich.
[1] https://github.com/kubevirt/kubevirt/pull/9587
Many thanks, Alice
-- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com virt-top is 'top' for virtual machines. Tiny program with many powerful monitoring features, net stats, disk stats, logging, etc. http://people.redhat.com/~rjones/virt-top
On Mon, Sep 25, 2023 at 04:26:59PM +0200, Alice Frosi wrote:
Hi Rich,
On Mon, Sep 25, 2023 at 4:10 PM Richard W.M. Jones <rjones@redhat.com> wrote:
[Alice: See patch 2]
I'm not 100% sure about the source of this work. However, we had in KubeVirt people interested in using localtime with Windows [1]. Yes, I see that you pointed to that PR in patch 2. The problem with that PR is the migration. What happens if we migrate the VM to a host that is in another timezone? Is it going to break the application running inside the VM?
I would generally expect all hosts to be configured in the same timezone, for any given single cloud deployment. If you really are concerned about this problem though, libvirt lets you set this explicitly eg <clock offset='localtime' timezone='Europe/London'/> With regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
On Mon, Sep 25, 2023 at 5:19 PM Daniel P. Berrangé <berrange@redhat.com> wrote:
On Mon, Sep 25, 2023 at 04:26:59PM +0200, Alice Frosi wrote:
Hi Rich,
On Mon, Sep 25, 2023 at 4:10 PM Richard W.M. Jones <rjones@redhat.com> wrote:
[Alice: See patch 2]
I'm not 100% sure about the source of this work. However, we had in KubeVirt people interested in using localtime with Windows [1]. Yes, I see that you pointed to that PR in patch 2. The problem with that PR is the migration. What happens if we migrate the VM to a host that is in another timezone? Is it going to break the application running inside the VM?
I would generally expect all hosts to be configured in the same timezone, for any given single cloud deployment.
If you really are concerned about this problem though, libvirt lets you set this explicitly eg
<clock offset='localtime' timezone='Europe/London'/>
I don't think we support this option in KubeVirt yet. It might be useful to extend it. However, I'm not sure if it is the user's responsibility to choose the TZ or if it should be decided by KubeVirt and not exposed to the API. Generally, I don't think we can always assume that all the node in the cluster are in the same time zone. This could be signaled with some labels and allow the VM to be scheduled only on the node with that label.
With regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
On 9/25/23 16:04, Richard W.M. Jones wrote:
[Alice: See patch 2]
[This patch is a bit rough, it could do with better commit messages and some tests. Please test it to see if it solves the Windows conversion issue described in the thread below.]
We currently do not set any <clock/> field in guest output. Most Windows guests expect the BIOS to be set to localtime, whereas almost all Linux guests would expect it to be set to UTC. It is also possible to configure a Windows guest to expect BIOS set to UTC.
The default is usually BIOS set to UTC, so for many Windows guests this would be wrong. This specifically may cause problems when scheduling qemu-ga installation, see the thread here:
https://listman.redhat.com/archives/libguestfs/2023-September/thread.html#32...
but could cause other general issues with time in the guest.
One way to implement this would be to copy the source hypervisor information across; however I'm not confident this information is read correctly. A better way is to read out what the guest is expecting from the Windows registry. (For Linux we just assume BIOS is always UTC, since that's the default for almost any Linux guest which hasn't been dual-booted with Windows, which for VMs would be incredibly rare.)
I think the word "BIOS" is incorrectly used all over the series; I'd rather say "RTC" / "real time clock". Laszlo
On 9/25/23 17:48, Laszlo Ersek wrote:
On 9/25/23 16:04, Richard W.M. Jones wrote:
[Alice: See patch 2]
[This patch is a bit rough, it could do with better commit messages and some tests. Please test it to see if it solves the Windows conversion issue described in the thread below.]
We currently do not set any <clock/> field in guest output. Most Windows guests expect the BIOS to be set to localtime, whereas almost all Linux guests would expect it to be set to UTC. It is also possible to configure a Windows guest to expect BIOS set to UTC.
The default is usually BIOS set to UTC, so for many Windows guests this would be wrong. This specifically may cause problems when scheduling qemu-ga installation, see the thread here:
https://listman.redhat.com/archives/libguestfs/2023-September/thread.html#32...
but could cause other general issues with time in the guest.
One way to implement this would be to copy the source hypervisor information across; however I'm not confident this information is read correctly. A better way is to read out what the guest is expecting from the Windows registry. (For Linux we just assume BIOS is always UTC, since that's the default for almost any Linux guest which hasn't been dual-booted with Windows, which for VMs would be incredibly rare.)
I think the word "BIOS" is incorrectly used all over the series; I'd rather say "RTC" / "real time clock".
... meaning patch subject lines, commit message bodies, and code.
On Mon, Sep 25, 2023 at 05:49:18PM +0200, Laszlo Ersek wrote:
On 9/25/23 17:48, Laszlo Ersek wrote:
On 9/25/23 16:04, Richard W.M. Jones wrote:
[Alice: See patch 2]
[This patch is a bit rough, it could do with better commit messages and some tests. Please test it to see if it solves the Windows conversion issue described in the thread below.]
We currently do not set any <clock/> field in guest output. Most Windows guests expect the BIOS to be set to localtime, whereas almost all Linux guests would expect it to be set to UTC. It is also possible to configure a Windows guest to expect BIOS set to UTC.
The default is usually BIOS set to UTC, so for many Windows guests this would be wrong. This specifically may cause problems when scheduling qemu-ga installation, see the thread here:
https://listman.redhat.com/archives/libguestfs/2023-September/thread.html#32...
but could cause other general issues with time in the guest.
One way to implement this would be to copy the source hypervisor information across; however I'm not confident this information is read correctly. A better way is to read out what the guest is expecting from the Windows registry. (For Linux we just assume BIOS is always UTC, since that's the default for almost any Linux guest which hasn't been dual-booted with Windows, which for VMs would be incredibly rare.)
I think the word "BIOS" is incorrectly used all over the series; I'd rather say "RTC" / "real time clock".
... meaning patch subject lines, commit message bodies, and code.
OK :-) Rich. -- Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones Read my programming and virtualization blog: http://rwmj.wordpress.com libguestfs lets you edit virtual machines. Supports shell scripting, bindings from many languages. http://libguestfs.org
participants (4)
-
Alice Frosi -
Daniel P. Berrangé -
Laszlo Ersek -
Richard W.M. Jones