[pve-devel] [PATCH many] add cluster-wide hardware device mapping

[DERUMIER, Alexandre]

many thanks for this patch series.

I had a student at the previous training needing this feature.

I'll be usefull in the future to get mdev vgpu live migration (It's not 
yet working with qemu 6.2)


I'll try to test it next week when I'll back from holiday.



Le 19/07/22 à 13:46, Dominik Csapak a écrit :
> this series aims to add a cluster-wide device mapping for pci and usb devices.
> so that an admin can configure a device to be availble for migration and
> configuring for uses that are non-root
>
> built-in are some additional safety checks in contrast to current
> passthrough, e.g. if pci addresses shift, with the mapping
> we can detect that and prevent a vm to boot with the wrong device
> (in most cases, there are some edge cases when one has multiple
> of the same device, e.g. the same gpu, that we cannot detect)
>
> a few pain points that are probably worth discussing/thinking about:
> (i did not really get feedback on my last RFC on this)
> * the config format
>      i changed to a json backed config, since it makes handling it much
>      easier (since we have a id -> nodenames -> mapping relation that
>      we cannot easily represent with a section config). some
>      (small) parts are written from scratch (update/createSchema for
>      instance) but we would have to do that anyway
>
>      if wanted i can make the section config work, but it makes the
>      handling quite a big uglier (for example, we have name the usb/pci
>      properties differently because the section config cannnot have
>      different formats for different sections)
>
> * getting the cluster wide info
>      the configuring of mappings is all done via node specific api paths,
>      but i created a cluster wide api path that returns the overall
>      structure for easy consumption from the gui. to get the remaining
>      data from the other nodes, i let the gui make an api call
>      for each node.
>
>      alternatively we could distribute the necessary info via pmxcfs,
>      but we'd have to broadcast basically the whole pci listing for all
>      nodes in a relatively short interval, only for it to be extremly
>      seldomly used (when looking at the cluster wide hardware
>      mappings...)
>
> * some minor things that can be improved are how the gui looks/behaves:
>      - 'add new' and 'add mapping' are probably to similar, but i did
>        not come up with really better alternatives
>      - i find the tree of entry -> node-mappings nice, but there may be
>        an even better representation?
>      - position in cluster menu is probably not optimal
>        (but where to put it?)
>
> changes from the rfc:
> * new cluster wide gui instead of node-local one (removed that, since
>    it's not necessary when we have a cluster-wide one)
> * uses json instead of a section config
> * api is quite different overall, i split the type into its own level
>    for configuring, similar to what we do in pbs
>    (e.g. /nodes/NODENAME/hardware/mapping/usb/)
> * fixed quite some bugs the rfc had
> * added patch for handling the gui with limited permissions better
> * added a 'comment' field for mappings
>
> dependencies are ofc:
>
>      manager depends on qemu-server,pve-access-control,pve-common
>      qemu-server depends on pve-access-control,pve-common
>      pve-common depends on pve-cluster
>
> pve-cluster:
>
> Dominik Csapak (1):
>    add nodes/hardware-map.conf
>
>   data/PVE/Cluster.pm | 1 +
>   data/src/status.c   | 1 +
>   2 files changed, 2 insertions(+)
>
> pve-access-control:
>
> Dominik Csapak (2):
>    PVE/AccessControl: add Hardware.* privileges and /hardware/ paths
>    PVE/RPCEnvironment: add helper for checking hw permissions
>
>   src/PVE/AccessControl.pm  | 13 +++++++++++++
>   src/PVE/RPCEnvironment.pm |  9 +++++++++
>   2 files changed, 22 insertions(+)
>
> pve-common:
>
> Dominik Csapak (1):
>    add PVE/HardwareMap
>
>   src/Makefile           |   1 +
>   src/PVE/HardwareMap.pm | 363 +++++++++++++++++++++++++++++++++++++++++
>   2 files changed, 364 insertions(+)
>   create mode 100644 src/PVE/HardwareMap.pm
>
> qemu-server:
>
> Dominik Csapak (7):
>    PVE/QemuServer: allow mapped usb devices in config
>    PVE/QemuServer: allow mapped pci deviced in config
>    PVE/API2/Qemu: add permission checks for mapped usb devices
>    PVE/API2/Qemu: add permission checks for mapped pci devices
>    PVE/QemuServer: extend 'check_local_resources' for mapped resources
>    PVE/API2/Qemu: migrate preconditions: use new check_local_resources
>      info
>    PVE/QemuMigrate: check for mapped resources on migration
>
>   PVE/API2/Qemu.pm      | 108 ++++++++++++++++++++++++++++++++++++++----
>   PVE/QemuMigrate.pm    |  13 ++++-
>   PVE/QemuServer.pm     |  38 ++++++++++++++-
>   PVE/QemuServer/PCI.pm |  20 +++++++-
>   PVE/QemuServer/USB.pm |  21 +++++++-
>   5 files changed, 185 insertions(+), 15 deletions(-)
>
> pve-manager:
>
> Dominik Csapak (12):
>    PVE/API2/Hardware: add Mapping.pm
>    PVE/API2/Cluster: add Hardware mapping list api call
>    ui: form/USBSelector: make it more flexible with nodename
>    ui: form: add PCIMapSelector
>    ui: form: add USBMapSelector
>    ui: qemu/PCIEdit: rework panel to add a mapped configuration
>    ui: qemu/USBEdit: add 'mapped' device case
>    ui: add window/PCIEdit: edit window for pci mappings
>    ui: add window/USBEdit: edit window for usb mappings
>    ui: add dc/HardwareView: a CRUD interface for hardware mapping
>    ui: window/Migrate: allow mapped devices
>    ui: improve permission handling for hardware
>
>   PVE/API2/Cluster.pm                 |   8 +
>   PVE/API2/Cluster/Hardware.pm        | 117 +++++
>   PVE/API2/Cluster/Makefile           |   1 +
>   PVE/API2/Hardware.pm                |   6 +
>   PVE/API2/Hardware/Makefile          |   1 +
>   PVE/API2/Hardware/Mapping.pm        | 708 ++++++++++++++++++++++++++++
>   www/manager6/Makefile               |   5 +
>   www/manager6/data/PermPathStore.js  |   1 +
>   www/manager6/dc/Config.js           |  18 +-
>   www/manager6/dc/HardwareView.js     | 314 ++++++++++++
>   www/manager6/form/PCIMapSelector.js |  95 ++++
>   www/manager6/form/PCISelector.js    |  18 +-
>   www/manager6/form/USBMapSelector.js |  73 +++
>   www/manager6/form/USBSelector.js    |  33 +-
>   www/manager6/qemu/HardwareView.js   |  17 +-
>   www/manager6/qemu/PCIEdit.js        | 231 ++++++---
>   www/manager6/qemu/USBEdit.js        |  34 +-
>   www/manager6/window/Migrate.js      |  37 +-
>   www/manager6/window/PCIEdit.js      | 323 +++++++++++++
>   www/manager6/window/USBEdit.js      | 248 ++++++++++
>   20 files changed, 2185 insertions(+), 103 deletions(-)
>   create mode 100644 PVE/API2/Cluster/Hardware.pm
>   create mode 100644 PVE/API2/Hardware/Mapping.pm
>   create mode 100644 www/manager6/dc/HardwareView.js
>   create mode 100644 www/manager6/form/PCIMapSelector.js
>   create mode 100644 www/manager6/form/USBMapSelector.js
>   create mode 100644 www/manager6/window/PCIEdit.js
>   create mode 100644 www/manager6/window/USBEdit.js
>