T8329: Fix interface naming for Azure VF interfaces with Accelerated Networking enabled

ROOT-CAUSE:
Some Azure VF interfaces miss to get renamed leading to errors in the downstream rules and mess up with the interface names.
Two main problematic scenarios were found which prohibited the renaming of a VF interface to vf_ethN:
1. Missing udev add event, when change event is received directly
2. A VF interface registering during rootfs stage

FIX:
Rule 63:
-Azure VF naming is now handled by a dedicated helper - vyos_vf_name to provide collision-free names for VF interfaces. The helper vyos_vf_name would be packaged into initramfs so the same behavior works in early boot and normal boot (a separate PR).
-Rule 63 is now guarded to prevent recursive renaming of VF interfaces.

Rule 65:
-A fallback VF rename path has been added for any leftover VF interfaces that were missed to be renamed due to some unexpected situation. Those VF interfaces are renamed prior to running persistent renaming of synthetic interfaces. This prevents VF interfaces from being considered as synthetic interfaces which may lead to errors in the flow of execution.
-Rule 65 is now guarded so generic persistent naming does not override VF names.

Author: ritika0313

src/etc/udev/rules.d/63-hyperv-vf-net.rules

@@ -1,5 +1,9 @@
 ATTR{[dmi/id]sys_vendor}!="Microsoft Corporation", GOTO="end_hyperv_nic"
 
-ACTION=="add", SUBSYSTEM=="net", DRIVERS=="hv_pci", NAME="vf_%k"
+# Primary rename path for Azure VF.
+# Use helper-based naming so the VFs that reuse eth0,eth1,.. names do not
+# collide with already assigned vf_ethN names. Also, prevent renaming of
+# the interfaces already renamed to vf_ethN
+ACTION=="add", SUBSYSTEM=="net", DRIVERS=="hv_pci", KERNEL=="eth*", PROGRAM="vyos_vf_name %k", NAME="%c"
 
 LABEL="end_hyperv_nic"

src/etc/udev/rules.d/65-vyos-net.rules

@@ -4,6 +4,21 @@
 ACTION!="add", 				GOTO="vyos_net_end"
 SUBSYSTEM!="net",			GOTO="vyos_net_end"
 
+# Fallback path for Azure VFs: if any Azure Mellanox VF still appears as plain
+# ethN, rename it to vf_ethX before persistent synthetic interface naming runs
+# to prevent it from incorrectly being considered as a synthetic interface.
+ATTR{[dmi/id]sys_vendor}=="Microsoft Corporation", DRIVERS=="mlx*_core", KERNEL=="eth*", PROGRAM="vyos_vf_name %k", NAME="%c", GOTO="vyos_net_end"
+
+# VF interfaces are not handled by vyos_net_name, so skip them.
+# KERNEL=="vf_*" catches already-renamed VFs.
+# DRIVERS=="mlx*_core" on Azure catches VFs that reached here in an unexpected
+# intermediate state, are still named ethN/eN and must not be remapped
+# by vyos_net_name. This guard is intentionally scoped to Microsoft Azure
+# to avoid excluding bare-metal Mellanox NICs on physical hardware, which do
+# need persistent naming via vyos_net_name.
+KERNEL=="vf_*", 							GOTO="vyos_net_end"
+ATTR{[dmi/id]sys_vendor}=="Microsoft Corporation", DRIVERS=="mlx*_core",	GOTO="vyos_net_end"
+
 # Do name change for ethernet and wireless devices only
 KERNEL!="eth*|wlan*|e*", 			GOTO="vyos_net_end"
 

src/udev/vyos_vf_name

@@ -0,0 +1,88 @@
+#!/bin/sh
+# vyos_vf_name - pick a non-colliding VF interface name.
+# Called by udev rules for VF devices that still have plain ethN names.
+#
+# Written as a POSIX shell script so it works in both initramfs (no Python3)
+# and the normal rootfs udev context.
+#
+# Usage: vyos_vf_name <current_ifname>
+# Prints the target vf_ethN name to stdout.
+
+set -e
+
+if [ -z "$1" ]; then
+    exit 1
+fi
+
+IFNAME="$1"
+LOCK_DIR="/run/udev"
+LOCK_PATH="${LOCK_DIR}/vyos_vf_name.lock"
+
+# Already normalized - return as-is.
+case "$IFNAME" in
+    vf_*)
+        echo "$IFNAME"
+        exit 0
+        ;;
+esac
+
+mkdir -p "$LOCK_DIR"
+
+# Acquire a lock to avoid race conditions when multiple VF devices are being
+# renamed concurrently.
+LOCK_ACQUIRED=0
+I=0
+while [ "$I" -lt 200 ]; do
+    if mkdir "$LOCK_PATH" 2>/dev/null; then
+        LOCK_ACQUIRED=1
+        break
+    fi
+    I=$((I + 1))
+    sleep 0.01
+done
+
+[ "$LOCK_ACQUIRED" -eq 1 ] || exit 1
+
+cleanup() {
+    rmdir "$LOCK_PATH" 2>/dev/null || true
+}
+trap cleanup EXIT INT TERM
+
+# Preferred: preserve the original ethN index when vf_ethN is not yet taken.
+case "$IFNAME" in
+    eth*)
+        PREFERRED="${IFNAME#eth}"
+        # Validate it is a pure integer.
+        case "$PREFERRED" in
+            *[!0-9]*)
+                ;;
+            *)
+                if ! [ -d "/sys/class/net/vf_eth${PREFERRED}" ]; then
+                    echo "vf_eth${PREFERRED}"
+                    exit 0
+                fi
+                ;;
+        esac
+        ;;
+esac
+
+# Fallback: allocate just above the highest ethN synthetic index present to
+# avoid collisions.
+MAX=-1
+for D in /sys/class/net/eth*; do
+    [ -d "$D" ] || continue
+    N="${D##*/eth}"
+    # Skip if not a pure integer (e.g. no match expands to literal path).
+    case "$N" in *[!0-9]*) continue;; esac
+    [ "$N" -gt "$MAX" ] && MAX="$N"
+done
+
+IDX=$((MAX + 1))
+[ "$IDX" -lt 0 ] && IDX=0
+
+# Skip any indices already occupied by existing vf_ethN interfaces.
+while [ -d "/sys/class/net/vf_eth${IDX}" ]; do
+    IDX=$((IDX + 1))
+done
+
+echo "vf_eth${IDX}"