Refactor build.func: Simplify GPU passthrough, add APT cacher check, remove var_ctid/var_ipv6_static from defaults
Some checks failed
Bump build.func Revision / bump-revision (push) Has been cancelled
Some checks failed
Bump build.func Revision / bump-revision (push) Has been cancelled
This commit is contained in:
CanbiZ
parent
6cb374c542
commit
cb33e40562
133
misc/build.func
133
misc/build.func
@ -307,6 +307,19 @@ base_settings() {
|
||||
GATE=${var_gateway:-""}
|
||||
APT_CACHER=${var_apt_cacher:-""}
|
||||
APT_CACHER_IP=${var_apt_cacher_ip:-""}
|
||||
|
||||
# Runtime check: Verify APT cacher is reachable if configured
|
||||
if [[ -n "$APT_CACHER_IP" && "$APT_CACHER" == "yes" ]]; then
|
||||
if ! curl -s --connect-timeout 2 "http://${APT_CACHER_IP}:3142" >/dev/null 2>&1; then
|
||||
msg_warn "APT Cacher configured but not reachable at ${APT_CACHER_IP}:3142"
|
||||
msg_info "Disabling APT Cacher for this installation"
|
||||
APT_CACHER=""
|
||||
APT_CACHER_IP=""
|
||||
else
|
||||
msg_ok "APT Cacher verified at ${APT_CACHER_IP}:3142"
|
||||
fi
|
||||
fi
|
||||
|
||||
MTU=${var_mtu:-""}
|
||||
SD=${var_storage:-""}
|
||||
NS=${var_ns:-""}
|
||||
@ -981,9 +994,10 @@ EOF
|
||||
# ------------------------------------------------------------------------------
|
||||
default_var_settings() {
|
||||
# Allowed var_* keys (alphabetically sorted)
|
||||
# Note: Removed var_ctid (can only exist once), var_ipv6_static (static IPs are unique)
|
||||
local VAR_WHITELIST=(
|
||||
var_apt_cacher var_apt_cacher_ip var_brg var_cpu var_ctid var_disk var_fuse
|
||||
var_gateway var_hostname var_ipv6_method var_ipv6_static var_mac var_mtu
|
||||
var_apt_cacher var_apt_cacher_ip var_brg var_cpu var_disk var_fuse
|
||||
var_gateway var_hostname var_ipv6_method var_mac var_mtu
|
||||
var_net var_ns var_pw var_ram var_tags var_tun var_unprivileged
|
||||
var_verbose var_vlan var_ssh var_ssh_authorized_key var_container_storage var_template_storage
|
||||
)
|
||||
@ -1046,7 +1060,6 @@ var_brg=vmbr0
|
||||
var_net=dhcp
|
||||
var_ipv6_method=none
|
||||
# var_gateway=
|
||||
# var_ipv6_static=
|
||||
# var_vlan=
|
||||
# var_mtu=
|
||||
# var_mac=
|
||||
@ -1184,9 +1197,10 @@ get_app_defaults_path() {
|
||||
# - Extracts raw values from flags like ",gw=..." ",mtu=..." etc.
|
||||
# ------------------------------------------------------------------------------
|
||||
if ! declare -p VAR_WHITELIST >/dev/null 2>&1; then
|
||||
# Note: Removed var_ctid (can only exist once), var_ipv6_static (static IPs are unique)
|
||||
declare -ag VAR_WHITELIST=(
|
||||
var_apt_cacher var_apt_cacher_ip var_brg var_cpu var_ctid var_disk var_fuse
|
||||
var_gateway var_hostname var_ipv6_method var_ipv6_static var_mac var_mtu
|
||||
var_apt_cacher var_apt_cacher_ip var_brg var_cpu var_disk var_fuse
|
||||
var_gateway var_hostname var_ipv6_method var_mac var_mtu
|
||||
var_net var_ns var_pw var_ram var_tags var_tun var_unprivileged
|
||||
var_verbose var_vlan var_ssh var_ssh_authorized_key var_container_storage var_template_storage
|
||||
)
|
||||
@ -1361,7 +1375,7 @@ _build_current_app_vars_tmp() {
|
||||
[ -n "$_ns" ] && echo "var_ns=$(_sanitize_value "$_ns")"
|
||||
|
||||
[ -n "$_ipv6_method" ] && echo "var_ipv6_method=$(_sanitize_value "$_ipv6_method")"
|
||||
[ -n "$_ipv6_static" ] && echo "var_ipv6_static=$(_sanitize_value "$_ipv6_static")"
|
||||
# var_ipv6_static removed - static IPs are unique, can't be default
|
||||
|
||||
[ -n "$_ssh" ] && echo "var_ssh=$(_sanitize_value "$_ssh")"
|
||||
[ -n "$_ssh_auth" ] && echo "var_ssh_authorized_key=$(_sanitize_value "$_ssh_auth")"
|
||||
@ -2183,42 +2197,17 @@ build_container() {
|
||||
# Check for NVIDIA GPU - look for NVIDIA vendor ID [10de]
|
||||
if echo "$pci_vga_info" | grep -q "\[10de:"; then
|
||||
msg_info "Detected NVIDIA GPU"
|
||||
if ! check_nvidia_host_setup; then
|
||||
msg_error "NVIDIA host setup incomplete. Skipping GPU passthrough."
|
||||
msg_info "Fix NVIDIA drivers on host, then recreate container or passthrough manually."
|
||||
return 0
|
||||
fi
|
||||
|
||||
for d in /dev/nvidia* /dev/nvidiactl /dev/nvidia-modeset; do
|
||||
# Simple passthrough - just bind /dev/nvidia* devices if they exist
|
||||
for d in /dev/nvidia* /dev/nvidiactl /dev/nvidia-modeset /dev/nvidia-uvm /dev/nvidia-uvm-tools; do
|
||||
[[ -e "$d" ]] && NVIDIA_DEVICES+=("$d")
|
||||
done
|
||||
|
||||
if [[ ${#NVIDIA_DEVICES[@]} -eq 0 ]]; then
|
||||
msg_warn "NVIDIA GPU detected but no /dev/nvidia* devices found"
|
||||
msg_warn "Please install NVIDIA drivers on host: apt install nvidia-driver"
|
||||
if [[ ${#NVIDIA_DEVICES[@]} -gt 0 ]]; then
|
||||
msg_info "Found ${#NVIDIA_DEVICES[@]} NVIDIA device(s) for passthrough"
|
||||
else
|
||||
if [[ "$CT_TYPE" == "0" ]]; then
|
||||
cat <<EOF >>"$LXC_CONFIG"
|
||||
# NVIDIA GPU Passthrough (privileged)
|
||||
lxc.cgroup2.devices.allow: c 195:* rwm
|
||||
lxc.cgroup2.devices.allow: c 243:* rwm
|
||||
lxc.mount.entry: /dev/nvidia0 dev/nvidia0 none bind,optional,create=file
|
||||
lxc.mount.entry: /dev/nvidiactl dev/nvidiactl none bind,optional,create=file
|
||||
lxc.mount.entry: /dev/nvidia-uvm dev/nvidia-uvm none bind,optional,create=file
|
||||
lxc.mount.entry: /dev/nvidia-uvm-tools dev/nvidia-uvm-tools none bind,optional,create=file
|
||||
EOF
|
||||
|
||||
if [[ -e /dev/dri/renderD128 ]]; then
|
||||
echo "lxc.mount.entry: /dev/dri/renderD128 dev/dri/renderD128 none bind,optional,create=file" >>"$LXC_CONFIG"
|
||||
fi
|
||||
|
||||
export GPU_TYPE="NVIDIA"
|
||||
export NVIDIA_DRIVER_VERSION=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -n1)
|
||||
msg_ok "NVIDIA GPU passthrough configured (driver: ${NVIDIA_DRIVER_VERSION})"
|
||||
else
|
||||
msg_warn "NVIDIA passthrough only supported for privileged containers"
|
||||
return 0
|
||||
fi
|
||||
msg_warn "NVIDIA GPU detected via PCI but no /dev/nvidia* devices found"
|
||||
msg_info "Skipping NVIDIA passthrough (host drivers may not be loaded)"
|
||||
fi
|
||||
fi
|
||||
|
||||
@ -2319,19 +2308,12 @@ EOF
|
||||
[[ "$selected_gpu" == "INTEL" ]] && devices=("${INTEL_DEVICES[@]}")
|
||||
[[ "$selected_gpu" == "AMD" ]] && devices=("${AMD_DEVICES[@]}")
|
||||
|
||||
# For Proxmox WebUI visibility, add as dev0, dev1 etc.
|
||||
# Add lxc.mount.entry for each device
|
||||
for dev in "${devices[@]}"; do
|
||||
if [[ "$CT_TYPE" == "0" ]]; then
|
||||
# Privileged container - use dev entries for WebUI visibility
|
||||
# Use initial GID 104 (render) for renderD*, 44 (video) for card*
|
||||
if [[ "$dev" =~ renderD ]]; then
|
||||
echo "dev${dev_idx}: $dev,gid=104" >>"$LXC_CONFIG"
|
||||
else
|
||||
echo "dev${dev_idx}: $dev,gid=44" >>"$LXC_CONFIG"
|
||||
fi
|
||||
dev_idx=$((dev_idx + 1))
|
||||
echo "lxc.mount.entry: $dev $dev none bind,optional,create=file" >>"$LXC_CONFIG"
|
||||
|
||||
# Also add cgroup allows for privileged containers
|
||||
if [[ "$CT_TYPE" == "0" ]]; then
|
||||
# Privileged container - also add cgroup allows
|
||||
local major minor
|
||||
major=$(stat -c '%t' "$dev" 2>/dev/null || echo "0")
|
||||
minor=$(stat -c '%T' "$dev" 2>/dev/null || echo "0")
|
||||
@ -2339,33 +2321,25 @@ EOF
|
||||
if [[ "$major" != "0" && "$minor" != "0" ]]; then
|
||||
echo "lxc.cgroup2.devices.allow: c $((0x$major)):$((0x$minor)) rwm" >>"$LXC_CONFIG"
|
||||
fi
|
||||
else
|
||||
# Unprivileged container
|
||||
if [[ "$dev" =~ renderD ]]; then
|
||||
echo "dev${dev_idx}: $dev,uid=0,gid=104" >>"$LXC_CONFIG"
|
||||
else
|
||||
echo "dev${dev_idx}: $dev,uid=0,gid=44" >>"$LXC_CONFIG"
|
||||
fi
|
||||
dev_idx=$((dev_idx + 1))
|
||||
fi
|
||||
done
|
||||
|
||||
export GPU_TYPE="$selected_gpu"
|
||||
msg_ok "${selected_gpu} GPU passthrough configured (${dev_idx} devices)"
|
||||
msg_ok "${selected_gpu} GPU passthrough configured (${#devices[@]} devices)"
|
||||
;;
|
||||
|
||||
NVIDIA)
|
||||
if [[ ${#NVIDIA_DEVICES[@]} -eq 0 ]]; then
|
||||
msg_error "NVIDIA drivers not installed on host. Please install: apt install nvidia-driver"
|
||||
return 1
|
||||
msg_warn "No NVIDIA devices available for passthrough"
|
||||
return 0
|
||||
fi
|
||||
|
||||
# Add lxc.mount.entry for each NVIDIA device
|
||||
for dev in "${NVIDIA_DEVICES[@]}"; do
|
||||
# NVIDIA devices typically need different handling
|
||||
echo "dev${dev_idx}: $dev,uid=0,gid=44" >>"$LXC_CONFIG"
|
||||
dev_idx=$((dev_idx + 1))
|
||||
echo "lxc.mount.entry: $dev $dev none bind,optional,create=file" >>"$LXC_CONFIG"
|
||||
|
||||
if [[ "$CT_TYPE" == "0" ]]; then
|
||||
# Privileged container - also add cgroup allows
|
||||
local major minor
|
||||
major=$(stat -c '%t' "$dev" 2>/dev/null || echo "0")
|
||||
minor=$(stat -c '%T' "$dev" 2>/dev/null || echo "0")
|
||||
@ -2377,7 +2351,7 @@ EOF
|
||||
done
|
||||
|
||||
export GPU_TYPE="NVIDIA"
|
||||
msg_ok "NVIDIA GPU passthrough configured (${dev_idx} devices)"
|
||||
msg_ok "NVIDIA GPU passthrough configured (${#NVIDIA_DEVICES[@]} devices) - install drivers in container if needed"
|
||||
;;
|
||||
esac
|
||||
}
|
||||
@ -2511,19 +2485,6 @@ EOF'
|
||||
|
||||
msg_ok "Customized LXC Container"
|
||||
|
||||
# Verify GPU access if enabled
|
||||
if [[ "${ENABLE_VAAPI:-0}" == "1" ]] && [ "$var_os" != "alpine" ]; then
|
||||
pct exec "$CTID" -- bash -c "vainfo >/dev/null 2>&1" &&
|
||||
msg_ok "VAAPI verified working" ||
|
||||
msg_warn "VAAPI verification failed - may need additional configuration"
|
||||
fi
|
||||
|
||||
if [[ "${ENABLE_NVIDIA:-0}" == "1" ]] && [ "$var_os" != "alpine" ]; then
|
||||
pct exec "$CTID" -- bash -c "nvidia-smi >/dev/null 2>&1" &&
|
||||
msg_ok "NVIDIA verified working" ||
|
||||
msg_warn "NVIDIA verification failed - may need additional configuration"
|
||||
fi
|
||||
|
||||
# Install SSH keys
|
||||
install_ssh_keys_into_ct
|
||||
|
||||
@ -2701,26 +2662,6 @@ fix_gpu_gids() {
|
||||
fi
|
||||
}
|
||||
|
||||
# NVIDIA-spezific check on host
|
||||
check_nvidia_host_setup() {
|
||||
if ! command -v nvidia-smi >/dev/null 2>&1; then
|
||||
msg_warn "NVIDIA GPU detected but nvidia-smi not found on host"
|
||||
msg_warn "Please install NVIDIA drivers on host first."
|
||||
#echo " 1. Download driver: wget https://us.download.nvidia.com/XFree86/Linux-x86_64/550.127.05/NVIDIA-Linux-x86_64-550.127.05.run"
|
||||
#echo " 2. Install: ./NVIDIA-Linux-x86_64-550.127.05.run --dkms"
|
||||
#echo " 3. Verify: nvidia-smi"
|
||||
return 1
|
||||
fi
|
||||
|
||||
# check if nvidia-smi works
|
||||
if ! nvidia-smi >/dev/null 2>&1; then
|
||||
msg_warn "nvidia-smi installed but not working. Driver issue?"
|
||||
return 1
|
||||
fi
|
||||
|
||||
return 0
|
||||
}
|
||||
|
||||
check_storage_support() {
|
||||
local CONTENT="$1" VALID=0
|
||||
while IFS= read -r line; do
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user