COS-3126: overlay.d: replace 25rhcos-azure-udev

manifest-el9-shared.yaml

@@ -8,6 +8,13 @@ remove-from-packages:
   - - zram-generator
     - "/usr/lib/systemd/zram-generator.conf"
 
+# The azure-vm-utils package provides these udev rules, but it wont be added
+# until EL10[1]. This can be dropped when moving to EL10, provided the
+# package is included at that time.
+# [1]:https://issues.redhat.com/browse/RHEL-73904
+ostree-layers:
+  - overlay/25rhcos-azure-udev
+
 postprocess:
   - |
     #!/usr/bin/bash

overlay.d/25rhcos-azure-udev/usr/lib/dracut/modules.d/25rhcos-azure-udev/module-setup.sh

@@ -2,17 +2,17 @@
 # -*- mode: shell-script; indent-tabs-mode: nil; sh-basic-offset: 4; -*-
 # ex: ts=8 sw=4 sts=4 et filetype=sh
 
-# We want to provide Azure udev rules as part of the initrd, so that Ignition
-# is able to detect disks and act on them.
+# These udev rules support certain VM types that present managed disks as
+# NVMe devices instead of traditional SCSI devices (e.g. Standard_M16bds_v3,
+# Standard_M16bs_v3, Standard_L8s_v4). The rules allow the Azure Disk CSI
+# driver to perform LUN-based disk detection when mounting these NVMe disks.
+# The azure-vm-utils package provides these udev rules[1], but it wont be added
+# until EL10 [2]. This can be dropped when moving to EL10, provided the
+# package is included at that time.
 #
-# The WALinuxAgent-udev has been changed to install udev rules into
-# the initramfs [1], but that change isn't in el8 yet. This can be
-# dropped when moving to el9.
-#
-# [1] https://src.fedoraproject.org/rpms/WALinuxAgent/c/521b67bc8575f53a30b4b2c4e63292e67483a4e1?branch=rawhide
+# [1] https://github.com/Azure/azure-vm-utils/blob/9c596916b6774f24420dac0ee7a72a6c9ddb5060/udev/80-azure-disk.rules
+# [2] https://issues.redhat.com/browse/RHEL-73904
 
 install() {
-    inst_multiple \
-        /usr/lib/udev/rules.d/66-azure-storage.rules \
-        /usr/lib/udev/rules.d/99-azure-product-uuid.rules
+    inst_rules /usr/lib/udev/rules.d/80-azure-disk.rules
 }

overlay.d/25rhcos-azure-udev/usr/lib/udev/rules.d/80-azure-disk.rules

@@ -0,0 +1,54 @@
+ACTION!="add|change", GOTO="azure_disk_end"
+SUBSYSTEM!="block", GOTO="azure_disk_end"
+
+KERNEL=="nvme*", ATTRS{nsid}=="?*", ENV{ID_MODEL}=="Microsoft NVMe Direct Disk", GOTO="azure_disk_nvme_direct_v1"
+KERNEL=="nvme*", ATTRS{nsid}=="?*", ENV{ID_MODEL}=="Microsoft NVMe Direct Disk v2", GOTO="azure_disk_nvme_direct_v2"
+KERNEL=="nvme*", ATTRS{nsid}=="?*", ENV{ID_MODEL}=="MSFT NVMe Accelerator v1.0", GOTO="azure_disk_nvme_remote_v1"
+ENV{ID_VENDOR}=="Msft", ENV{ID_MODEL}=="Virtual_Disk", GOTO="azure_disk_scsi"
+GOTO="azure_disk_end"
+
+LABEL="azure_disk_scsi"
+ATTRS{device_id}=="?00000000-0000-*", ENV{AZURE_DISK_TYPE}="os", GOTO="azure_disk_symlink"
+ENV{DEVTYPE}=="partition", PROGRAM="/bin/sh -c 'readlink /sys/class/block/%k/../device|cut -d: -f4'", ENV{AZURE_DISK_LUN}="$result"
+ENV{DEVTYPE}=="disk", PROGRAM="/bin/sh -c 'readlink /sys/class/block/%k/device|cut -d: -f4'", ENV{AZURE_DISK_LUN}="$result"
+ATTRS{device_id}=="{f8b3781a-1e82-4818-a1c3-63d806ec15bb}", ENV{AZURE_DISK_LUN}=="0", ENV{AZURE_DISK_TYPE}="os", ENV{AZURE_DISK_LUN}="", GOTO="azure_disk_symlink"
+ATTRS{device_id}=="{f8b3781b-1e82-4818-a1c3-63d806ec15bb}", ENV{AZURE_DISK_TYPE}="data", GOTO="azure_disk_symlink"
+ATTRS{device_id}=="{f8b3781c-1e82-4818-a1c3-63d806ec15bb}", ENV{AZURE_DISK_TYPE}="data", GOTO="azure_disk_symlink"
+ATTRS{device_id}=="{f8b3781d-1e82-4818-a1c3-63d806ec15bb}", ENV{AZURE_DISK_TYPE}="data", GOTO="azure_disk_symlink"
+
+# Use "resource" type for local SCSI because some VM skus offer NVMe local disks in addition to a SCSI resource disk, e.g. LSv3 family.
+# This logic is already in walinuxagent rules but we duplicate it here to avoid an unnecessary dependency for anyone requiring it.
+ATTRS{device_id}=="?00000000-0001-*", ENV{AZURE_DISK_TYPE}="resource", ENV{AZURE_DISK_LUN}="", GOTO="azure_disk_symlink"
+ATTRS{device_id}=="{f8b3781a-1e82-4818-a1c3-63d806ec15bb}", ENV{AZURE_DISK_LUN}=="1", ENV{AZURE_DISK_TYPE}="resource", ENV{AZURE_DISK_LUN}="", GOTO="azure_disk_symlink"
+GOTO="azure_disk_end"
+
+LABEL="azure_disk_nvme_direct_v1"
+LABEL="azure_disk_nvme_direct_v2"
+ATTRS{nsid}=="?*", ENV{AZURE_DISK_TYPE}="local", ENV{AZURE_DISK_SERIAL}="$env{ID_SERIAL_SHORT}"
+GOTO="azure_disk_nvme_id"
+
+LABEL="azure_disk_nvme_remote_v1"
+ATTRS{nsid}=="1", ENV{AZURE_DISK_TYPE}="os", GOTO="azure_disk_nvme_id"
+ATTRS{nsid}=="?*", ENV{AZURE_DISK_TYPE}="data", PROGRAM="/bin/sh -ec 'echo $$((%s{nsid}-2))'", ENV{AZURE_DISK_LUN}="$result"
+
+LABEL="azure_disk_nvme_id"
+# This rule is commented out because the 'azure-nvme-id' binary is only
+# available in the 'azure-vm-utils' package, which will not be included
+# until RHEL 10. The LABEL is retained to preserve the execution flow
+# for NVMe-related rules, effectively making this a no-op section.
+#ATTRS{nsid}=="?*", IMPORT{program}="/usr/sbin/azure-nvme-id --udev"
+
+LABEL="azure_disk_symlink"
+# systemd v254 ships an updated 60-persistent-storage.rules that would allow
+# these to be deduplicated using $env{.PART_SUFFIX}
+ENV{DEVTYPE}=="disk", ENV{AZURE_DISK_TYPE}=="os|resource|root", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}"
+ENV{DEVTYPE}=="disk", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_INDEX}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-index/$env{AZURE_DISK_INDEX}"
+ENV{DEVTYPE}=="disk", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_LUN}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-lun/$env{AZURE_DISK_LUN}"
+ENV{DEVTYPE}=="disk", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_NAME}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-name/$env{AZURE_DISK_NAME}"
+ENV{DEVTYPE}=="disk", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_SERIAL}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-serial/$env{AZURE_DISK_SERIAL}"
+ENV{DEVTYPE}=="partition", ENV{AZURE_DISK_TYPE}=="os|resource|root", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}-part%n"
+ENV{DEVTYPE}=="partition", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_INDEX}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-index/$env{AZURE_DISK_INDEX}-part%n"
+ENV{DEVTYPE}=="partition", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_LUN}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-lun/$env{AZURE_DISK_LUN}-part%n"
+ENV{DEVTYPE}=="partition", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_NAME}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-name/$env{AZURE_DISK_NAME}-part%n"
+ENV{DEVTYPE}=="partition", ENV{AZURE_DISK_TYPE}=="?*", ENV{AZURE_DISK_SERIAL}=="?*", SYMLINK+="disk/azure/$env{AZURE_DISK_TYPE}/by-serial/$env{AZURE_DISK_SERIAL}-part%n"
+LABEL="azure_disk_end"

overlay.d/README.md

@@ -41,10 +41,16 @@ https://listengine.tuxfamily.org/chrony.tuxfamily.org/chrony-dev/2020/05/msg0002
 25rhcos-azure-udev
 ------------------
 
-We want to provide Azure udev rules as part of the initrd, so that Ignition
-is able to detect disks and act on them. The WALinuxAgent-udev has been
-changed to install udev rules into the initramfs, but that change isn't
-in el8 yet. This can be dropped when moving to el9.
+These udev rules support certain VM types that present managed disks as
+NVMe devices instead of traditional SCSI devices (e.g. Standard_M16bds_v3,
+Standard_M16bs_v3, Standard_L8s_v4). The rules allow the Azure Disk CSI
+driver to perform LUN-based disk detection when mounting these NVMe disks.
+The azure-vm-utils package provides these udev rules[1], but it wont be added
+until EL10 [1]. This can be dropped when moving to EL10, provided the
+package is included at that time.
+
+[1] https://github.com/Azure/azure-vm-utils/blob/9c596916b6774f24420dac0ee7a72a6c9ddb5060/udev/80-azure-disk.rules
+[2] https://issues.redhat.com/browse/RHEL-73904
 
 30rhcos-nvme-compat-udev
 ------------------------