Merge pull request #1243 from jlebon/pr/udev-nvme-whitespace-symlink

overlay: add 30rhcos-nvme-compat-udev

common.yaml

@@ -22,6 +22,7 @@ ostree-layers:
   - overlay/20platform-chrony
   - overlay/21dhcp-chrony
   - overlay/25azure-udev-rules
+  - overlay/30rhcos-nvme-compat-udev
 
 arch-include:
   x86_64:

overlay.d/30rhcos-nvme-compat-udev/statoverride

@@ -0,0 +1,2 @@
+# Config file for overriding permission bits on overlay files/dirs
+# Format: =<file mode in decimal> <absolute path to a file or directory>

overlay.d/30rhcos-nvme-compat-udev/usr/lib/dracut/modules.d/30rhcos-nvme-compat-udev/module-setup.sh

@@ -0,0 +1,13 @@
+#!/bin/bash
+
+# NVMe by-id/ symlinks changed wrt leading spaces from RHEL8 to RHEL9:
+# https://issues.redhat.com/browse/OCPBUGS-11375
+# https://github.com/systemd/systemd/issues/27155
+
+# This rule adds back the previous symlinks for backwards compatibility. We want
+# it in the initramfs in case there are Ignition configs which referenced the
+# old symlinks.
+
+install() {
+    inst_multiple /usr/lib/udev/rules.d/61-persistent-storage-nvme-compat.rules
+}

overlay.d/30rhcos-nvme-compat-udev/usr/lib/udev/rules.d/61-persistent-storage-nvme-compat.rules

@@ -0,0 +1,20 @@
+# NVMe by-id/ symlinks changed wrt leading spaces from RHEL8 to RHEL9:
+# https://issues.redhat.com/browse/OCPBUGS-11375
+# https://github.com/systemd/systemd/issues/27155
+
+# This rule adds back the previous symlinks for backwards compatibility.
+
+# They're the same as the regular rules in 60-persistent-storage.rules, except
+# they don't include the `OPTIONS="string_escape=replace"` directive.
+
+ACTION=="remove", GOTO="persistent_storage_nvme_compat_end"
+ENV{UDEV_DISABLE_PERSISTENT_STORAGE_RULES_FLAG}=="1", GOTO="persistent_storage_nvme_compat_end"
+SUBSYSTEM!="block", GOTO="persistent_storage_nvme_compat_end"
+
+KERNEL=="nvme*[0-9]n*[0-9]", ENV{DEVTYPE}=="disk", ENV{ID_MODEL}=="?*", ENV{ID_SERIAL_SHORT}=="?*", \
+  ENV{ID_SERIAL}="$env{ID_MODEL}_$env{ID_SERIAL_SHORT}", SYMLINK+="disk/by-id/nvme-$env{ID_SERIAL}"
+
+KERNEL=="nvme*[0-9]n*[0-9]p*[0-9]", ENV{DEVTYPE}=="partition", ENV{ID_MODEL}=="?*", ENV{ID_SERIAL_SHORT}=="?*", \
+  ENV{ID_SERIAL}="$env{ID_MODEL}_$env{ID_SERIAL_SHORT}", SYMLINK+="disk/by-id/nvme-$env{ID_SERIAL}-part%n"
+
+LABEL="persistent_storage_nvme_compat_end"

overlay.d/README.md

@@ -39,9 +39,23 @@ lands in downstream packages. See upstream thread:
 https://listengine.tuxfamily.org/chrony.tuxfamily.org/chrony-dev/2020/05/msg00022.html
 
 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.
+
+30rhcos-nvme-compat-udev
+------------------------
+
+NVMe by-id/ symlinks changed wrt leading spaces from RHEL8 to RHEL9:
+https://issues.redhat.com/browse/OCPBUGS-11375
+https://github.com/systemd/systemd/issues/27155
+
+This overlay ships a rule that adds back the previous symlinks for backwards
+compatibility. TBD when we can drop this, e.g. by having layered software use
+the `eui` links instead or pivoting to GPT partition UUIDs. Customers may have
+also manually typed the old symlink in their Ignition configs and other k8s
+resources though. Those would require some communication before we can rip this
+out.

tests/kola/disk/nvme-compat-symlink

@@ -0,0 +1,27 @@
+#!/bin/bash
+## kola:
+##   exclusive: true
+##   additionalDisks: ["1G:channel=nvme,serial=    foobar"]
+##   platforms: qemu
+#
+# By default coreos-generate-iscsi-initiatorname.service is active on RHCOS,
+# inactive on FCOS
+# https://github.com/openshift/os/pull/453
+
+set -xeuo pipefail
+
+. $KOLA_EXT_DATA/commonlib.sh
+
+# the canonical symlink
+s=/dev/disk/by-id/nvme-QEMU_NVMe_Ctrl_____foobar
+if ! test -L "${s}"; then
+    fatal "missing canonical symlink ${s}"
+fi
+
+# the compat symlink
+s=/dev/disk/by-id/nvme-QEMU_NVMe_Ctrl__foobar
+if ! test -L "${s}"; then
+    fatal "missing compat symlink ${s}"
+fi
+
+ok "nvme compat symlink"