Merge branch 'bugfix/alpha.20' of github.com:Start9Labs/start-os into bugfix/alpha.20

build/image-recipe/build.sh

@@ -1,7 +1,6 @@
 #!/bin/bash
 set -e
 
-MAX_IMG_LEN=$((4 * 1024 * 1024 * 1024)) # 4GB
 
 echo "==== StartOS Image Build ===="
 
@@ -332,10 +331,10 @@ fi
 
 if [ "${IB_TARGET_PLATFORM}" = "raspberrypi" ]; then
     ln -sf /usr/bin/pi-beep /usr/local/bin/beep
-    sh /boot/config.sh > /boot/config.txt
+    sh /boot/firmware/config.sh > /boot/firmware/config.txt
     mkinitramfs -c gzip -o /boot/initrd.img-${RPI_KERNEL_VERSION}-rpi-v8 ${RPI_KERNEL_VERSION}-rpi-v8
     mkinitramfs -c gzip -o /boot/initrd.img-${RPI_KERNEL_VERSION}-rpi-2712 ${RPI_KERNEL_VERSION}-rpi-2712
-    cp /usr/lib/u-boot/rpi_arm64/u-boot.bin /boot/u-boot.bin
+    cp /usr/lib/u-boot/rpi_arm64/u-boot.bin /boot/firmware/u-boot.bin
 fi
 
 useradd --shell /bin/bash -G startos -m start9
@@ -411,7 +410,16 @@ elif [ "${IMAGE_TYPE}" = img ]; then
 	BOOT_LEN=$((2 * 1024 * 1024 * 1024))
 	BOOT_END=$((BOOT_START + BOOT_LEN - 1))
 	ROOT_START=$((BOOT_END + 1))
-	ROOT_LEN=$((MAX_IMG_LEN - ROOT_START))
+
+	# Size root partition to fit the squashfs + 256MB overhead for btrfs
+	# metadata and config overlay, avoiding the need for btrfs resize
+	SQUASHFS_SIZE=$(stat -c %s $prep_results_dir/binary/live/filesystem.squashfs)
+	ROOT_LEN=$(( SQUASHFS_SIZE + 256 * 1024 * 1024 ))
+	# Align to sector boundary
+	ROOT_LEN=$(( (ROOT_LEN + SECTOR_LEN - 1) / SECTOR_LEN * SECTOR_LEN ))
+
+	# Total image: partitions + GPT backup header (34 sectors)
+	IMG_LEN=$((ROOT_START + ROOT_LEN + 34 * SECTOR_LEN))
 
 	# Fixed GPT partition UUIDs (deterministic, based on old MBR disk ID cb15ae4d)
 	FW_UUID=cb15ae4d-0001-4000-8000-000000000001
@@ -420,7 +428,7 @@ elif [ "${IMAGE_TYPE}" = img ]; then
 	ROOT_UUID=cb15ae4d-0004-4000-8000-000000000004
 
 	TARGET_NAME=$prep_results_dir/${IMAGE_BASENAME}.img
-	truncate -s $MAX_IMG_LEN $TARGET_NAME
+	truncate -s $IMG_LEN $TARGET_NAME
 
 	sfdisk $TARGET_NAME <<-EOF
 		label: gpt
@@ -431,10 +439,23 @@ elif [ "${IMAGE_TYPE}" = img ]; then
 		${TARGET_NAME}4 : start=$((ROOT_START / SECTOR_LEN)), size=$((ROOT_LEN / SECTOR_LEN)), type=B921B045-1DF0-41C3-AF44-4C6F280D3FAE, uuid=${ROOT_UUID}, name="root"
 	EOF
 
-	FW_DEV=$(losetup --show -f --offset $FW_START --sizelimit $FW_LEN $TARGET_NAME)
+	# Create named loop device nodes (high minor numbers to avoid conflicts)
-	ESP_DEV=$(losetup --show -f --offset $ESP_START --sizelimit $ESP_LEN $TARGET_NAME)
+	# and detach any stale ones from previous failed builds
-	BOOT_DEV=$(losetup --show -f --offset $BOOT_START --sizelimit $BOOT_LEN $TARGET_NAME)
+	FW_DEV=/dev/startos-loop-fw
-	ROOT_DEV=$(losetup --show -f --offset $ROOT_START --sizelimit $ROOT_LEN $TARGET_NAME)
+	ESP_DEV=/dev/startos-loop-esp
+	BOOT_DEV=/dev/startos-loop-boot
+	ROOT_DEV=/dev/startos-loop-root
+	for dev in $FW_DEV:200 $ESP_DEV:201 $BOOT_DEV:202 $ROOT_DEV:203; do
+		name=${dev%:*}
+		minor=${dev#*:}
+		[ -e $name ] || mknod $name b 7 $minor
+		losetup -d $name 2>/dev/null || true
+	done
+
+	losetup $FW_DEV --offset $FW_START --sizelimit $FW_LEN $TARGET_NAME
+	losetup $ESP_DEV --offset $ESP_START --sizelimit $ESP_LEN $TARGET_NAME
+	losetup $BOOT_DEV --offset $BOOT_START --sizelimit $BOOT_LEN $TARGET_NAME
+	losetup $ROOT_DEV --offset $ROOT_START --sizelimit $ROOT_LEN $TARGET_NAME
 
 	mkfs.vfat -F32 -n firmware $FW_DEV
 	mkfs.vfat -F32 -n efi $ESP_DEV
@@ -447,18 +468,16 @@ elif [ "${IMAGE_TYPE}" = img ]; then
 	BOOT_STAGING=$(mktemp -d)
 	unsquashfs -n -f -d $BOOT_STAGING $prep_results_dir/binary/live/filesystem.squashfs boot
 
-	# Mount partitions
+	# Mount partitions (nested: firmware and efi inside boot)
-	mkdir -p $TMPDIR/firmware $TMPDIR/efi $TMPDIR/boot $TMPDIR/root
+	mkdir -p $TMPDIR/boot $TMPDIR/root
-	mount $FW_DEV $TMPDIR/firmware
-	mount $ESP_DEV $TMPDIR/efi
 	mount $BOOT_DEV $TMPDIR/boot
+	mkdir -p $TMPDIR/boot/firmware $TMPDIR/boot/efi
+	mount $FW_DEV $TMPDIR/boot/firmware
+	mount $ESP_DEV $TMPDIR/boot/efi
 	mount $ROOT_DEV $TMPDIR/root
 
-	# Split boot files: firmware to Part 1, kernels/initramfs to Part 3 (/boot)
+	# Copy boot files — nested mounts route firmware/* to the firmware partition
-	cp -a $BOOT_STAGING/boot/. $TMPDIR/firmware/
+	cp -a $BOOT_STAGING/boot/. $TMPDIR/boot/
-	for f in $TMPDIR/firmware/vmlinuz-* $TMPDIR/firmware/initrd.img-* $TMPDIR/firmware/System.map-* $TMPDIR/firmware/config-*; do
-		[ -e "$f" ] && mv "$f" $TMPDIR/boot/
-	done
 	rm -rf $BOOT_STAGING
 
 	mkdir $TMPDIR/root/images $TMPDIR/root/config
@@ -475,11 +494,9 @@ elif [ "${IMAGE_TYPE}" = img ]; then
 		rsync -a $SOURCE_DIR/raspberrypi/img/ $TMPDIR/next/
 
 		# Install GRUB: ESP at /boot/efi (Part 2), /boot (Part 3)
-		mkdir -p $TMPDIR/next/boot $TMPDIR/next/boot/efi $TMPDIR/next/boot/firmware \
+		mkdir -p $TMPDIR/next/boot \
 			$TMPDIR/next/dev $TMPDIR/next/proc $TMPDIR/next/sys $TMPDIR/next/media/startos/root
-		mount --bind $TMPDIR/boot $TMPDIR/next/boot
+		mount --rbind $TMPDIR/boot $TMPDIR/next/boot
-		mount --bind $TMPDIR/efi $TMPDIR/next/boot/efi
-		mount --bind $TMPDIR/firmware $TMPDIR/next/boot/firmware
 		mount --bind /dev $TMPDIR/next/dev
 		mount --bind /proc $TMPDIR/next/proc
 		mount --bind /sys $TMPDIR/next/sys
@@ -492,9 +509,7 @@ elif [ "${IMAGE_TYPE}" = img ]; then
 		umount $TMPDIR/next/sys
 		umount $TMPDIR/next/proc
 		umount $TMPDIR/next/dev
-		umount $TMPDIR/next/boot/firmware
+		umount -l $TMPDIR/next/boot
-		umount $TMPDIR/next/boot/efi
-		umount $TMPDIR/next/boot
 
 		# Fix root= in grub.cfg: update-grub sees loop devices, but the
 		# real device uses a fixed GPT PARTUUID for root (Part 4).
@@ -507,39 +522,16 @@ elif [ "${IMAGE_TYPE}" = img ]; then
 	umount $TMPDIR/next
 	umount $TMPDIR/lower
 
-	umount $TMPDIR/firmware
+	umount $TMPDIR/boot/firmware
-	umount $TMPDIR/efi
+	umount $TMPDIR/boot/efi
 	umount $TMPDIR/boot
 	umount $TMPDIR/root
 
-	# Shrink btrfs to minimum size
-	SHRINK_MNT=$(mktemp -d)
-	mount $ROOT_DEV $SHRINK_MNT
-	btrfs filesystem resize min $SHRINK_MNT
-	umount $SHRINK_MNT
-	rmdir $SHRINK_MNT
-	ROOT_LEN=$(btrfs inspect-internal dump-super $ROOT_DEV | awk '/^total_bytes/ {print $2}')
-
 	losetup -d $ROOT_DEV
 	losetup -d $BOOT_DEV
 	losetup -d $ESP_DEV
 	losetup -d $FW_DEV
 
-	# Recreate partition table with shrunk root
-	sfdisk $TARGET_NAME <<-EOF
-		label: gpt
-
-		${TARGET_NAME}1 : start=$((FW_START / SECTOR_LEN)), size=$((FW_LEN / SECTOR_LEN)), type=EBD0A0A2-B9E5-4433-87C0-68B6B72699C7, uuid=${FW_UUID}, name="firmware"
-		${TARGET_NAME}2 : start=$((ESP_START / SECTOR_LEN)), size=$((ESP_LEN / SECTOR_LEN)), type=C12A7328-F81F-11D2-BA4B-00A0C93EC93B, uuid=${ESP_UUID}, name="efi"
-		${TARGET_NAME}3 : start=$((BOOT_START / SECTOR_LEN)), size=$((BOOT_LEN / SECTOR_LEN)), type=0FC63DAF-8483-4772-8E79-3D69D8477DE4, uuid=${BOOT_UUID}, name="boot"
-		${TARGET_NAME}4 : start=$((ROOT_START / SECTOR_LEN)), size=$((ROOT_LEN / SECTOR_LEN)), type=B921B045-1DF0-41C3-AF44-4C6F280D3FAE, uuid=${ROOT_UUID}, name="root"
-	EOF
-
-	TARGET_SIZE=$((ROOT_START + ROOT_LEN))
-	truncate -s $TARGET_SIZE $TARGET_NAME
-	# Move backup GPT to new end of disk after truncation
-	sgdisk -e $TARGET_NAME
-
 	mv $TARGET_NAME $RESULTS_DIR/$IMAGE_BASENAME.img
 
 fi

build/lib/scripts/chroot-and-upgrade

@@ -34,7 +34,7 @@ set -- "${POSITIONAL_ARGS[@]}" # restore positional parameters
 
 if [ -z "$NO_SYNC" ]; then
     echo 'Syncing...'
-    umount -R /media/startos/next 2> /dev/null
+    umount -l /media/startos/next 2> /dev/null
     umount /media/startos/upper 2> /dev/null
     rm -rf /media/startos/upper /media/startos/next
     mkdir /media/startos/upper

build/lib/scripts/upgrade

@@ -24,7 +24,7 @@ fi
 
 unsquashfs -f -d / $1 boot
 
-umount -R /media/startos/next 2> /dev/null || true
+umount -l /media/startos/next 2> /dev/null || true
 umount /media/startos/upper 2> /dev/null || true
 umount /media/startos/lower 2> /dev/null || true
 
@@ -47,14 +47,9 @@ mount --bind /tmp /media/startos/next/tmp
 mount --bind /dev /media/startos/next/dev
 mount --bind /sys /media/startos/next/sys
 mount --bind /proc /media/startos/next/proc
-mount --bind /boot /media/startos/next/boot
+mount --rbind /boot /media/startos/next/boot
 mount --bind /media/startos/root /media/startos/next/media/startos/root
 
-if mountpoint /boot/efi 2>&1 > /dev/null; then
-    mkdir -p /media/startos/next/boot/efi
-    mount --bind /boot/efi /media/startos/next/boot/efi
-fi
-
 if mountpoint /sys/firmware/efi/efivars 2>&1 > /dev/null; then
     mount --bind /sys/firmware/efi/efivars /media/startos/next/sys/firmware/efi/efivars
 fi
@@ -79,7 +74,7 @@ SIGN_FILE="$(ls -1 /media/startos/next/usr/lib/linux-kbuild-*/scripts/sign-file
 
 sync
 
-umount -Rl /media/startos/next
+umount -l /media/startos/next
 umount /media/startos/upper
 umount /media/startos/lower
 

core/locales/i18n.yaml

@@ -1379,6 +1379,21 @@ net.tor.client-error:
   fr_FR: "Erreur du client Tor : %{error}"
   pl_PL: "Błąd klienta Tor: %{error}"
 
+# net/tunnel.rs
+net.tunnel.timeout-waiting-for-add:
+  en_US: "timed out waiting for gateway %{gateway} to appear in database"
+  de_DE: "Zeitüberschreitung beim Warten auf das Erscheinen von Gateway %{gateway} in der Datenbank"
+  es_ES: "se agotó el tiempo esperando que la puerta de enlace %{gateway} aparezca en la base de datos"
+  fr_FR: "délai d'attente dépassé pour l'apparition de la passerelle %{gateway} dans la base de données"
+  pl_PL: "upłynął limit czasu oczekiwania na pojawienie się bramy %{gateway} w bazie danych"
+
+net.tunnel.timeout-waiting-for-remove:
+  en_US: "timed out waiting for gateway %{gateway} to be removed from database"
+  de_DE: "Zeitüberschreitung beim Warten auf das Entfernen von Gateway %{gateway} aus der Datenbank"
+  es_ES: "se agotó el tiempo esperando que la puerta de enlace %{gateway} sea eliminada de la base de datos"
+  fr_FR: "délai d'attente dépassé pour la suppression de la passerelle %{gateway} de la base de données"
+  pl_PL: "upłynął limit czasu oczekiwania na usunięcie bramy %{gateway} z bazy danych"
+
 # net/wifi.rs
 net.wifi.ssid-no-special-characters:
   en_US: "SSID may not have special characters"

core/src/net/gateway.rs

@@ -1018,18 +1018,16 @@ async fn apply_policy_routing(
         })
         .copied();
 
-    // Flush and rebuild per-interface routing table.
+    // Rebuild per-interface routing table using `ip route replace` to avoid
-    // Clone all non-default routes from the main table so that LAN IPs on
+    // the connectivity gap that a flush+add cycle would create.  We replace
-    // other subnets remain reachable when the priority-75 catch-all overrides
+    // every desired route in-place (each replace is atomic in the kernel),
-    // default routing, then replace the default route with this interface's.
+    // then delete any stale routes that are no longer in the desired set.
-    Command::new("ip")
+
-        .arg("route")
+    // Collect the set of desired non-default route prefixes (the first
-        .arg("flush")
+    // whitespace-delimited token of each `ip route show` line is the
-        .arg("table")
+    // destination prefix, e.g. "192.168.1.0/24" or "10.0.0.0/8").
-        .arg(&table_str)
+    let mut desired_prefixes = BTreeSet::<String>::new();
-        .invoke(ErrorKind::Network)
+
-        .await
-        .log_err();
     if let Ok(main_routes) = Command::new("ip")
         .arg("route")
         .arg("show")
@@ -1044,11 +1042,14 @@ async fn apply_policy_routing(
             if line.is_empty() || line.starts_with("default") {
                 continue;
             }
+            if let Some(prefix) = line.split_whitespace().next() {
+                desired_prefixes.insert(prefix.to_owned());
+            }
             let mut cmd = Command::new("ip");
-            cmd.arg("route").arg("add");
+            cmd.arg("route").arg("replace");
             for part in line.split_whitespace() {
                 // Skip status flags that appear in route output but
-                // are not valid for `ip route add`.
+                // are not valid for `ip route replace`.
                 if part == "linkdown" || part == "dead" {
                     continue;
                 }
@@ -1058,10 +1059,11 @@ async fn apply_policy_routing(
             cmd.invoke(ErrorKind::Network).await.log_err();
         }
     }
-    // Add default route via this interface's gateway
+
+    // Replace the default route via this interface's gateway.
     {
         let mut cmd = Command::new("ip");
-        cmd.arg("route").arg("add").arg("default");
+        cmd.arg("route").arg("replace").arg("default");
         if let Some(gw) = ipv4_gateway {
             cmd.arg("via").arg(gw.to_string());
         }
@@ -1075,6 +1077,40 @@ async fn apply_policy_routing(
         cmd.invoke(ErrorKind::Network).await.log_err();
     }
 
+    // Delete stale routes: any non-default route in the per-interface table
+    // whose prefix is not in the desired set.
+    if let Ok(existing_routes) = Command::new("ip")
+        .arg("route")
+        .arg("show")
+        .arg("table")
+        .arg(&table_str)
+        .invoke(ErrorKind::Network)
+        .await
+        .and_then(|b| String::from_utf8(b).with_kind(ErrorKind::Utf8))
+    {
+        for line in existing_routes.lines() {
+            let line = line.trim();
+            if line.is_empty() || line.starts_with("default") {
+                continue;
+            }
+            let Some(prefix) = line.split_whitespace().next() else {
+                continue;
+            };
+            if desired_prefixes.contains(prefix) {
+                continue;
+            }
+            Command::new("ip")
+                .arg("route")
+                .arg("del")
+                .arg(prefix)
+                .arg("table")
+                .arg(&table_str)
+                .invoke(ErrorKind::Network)
+                .await
+                .log_err();
+        }
+    }
+
     // Ensure global CONNMARK restore rules in mangle PREROUTING (forwarded
     // packets) and OUTPUT (locally-generated replies). Both are needed:
     // PREROUTING handles DNAT-forwarded traffic, OUTPUT handles replies from

core/src/net/tunnel.rs

@@ -1,3 +1,5 @@
+use std::time::Duration;
+
 use clap::Parser;
 use imbl_value::InternedString;
 use patch_db::json_ptr::JsonPointer;
@@ -8,7 +10,9 @@ use ts_rs::TS;
 
 use crate::GatewayId;
 use crate::context::{CliContext, RpcContext};
-use crate::db::model::public::{GatewayType, NetworkInterfaceInfo, NetworkInterfaceType};
+use crate::db::model::public::{
+    GatewayType, NetworkInfo, NetworkInterfaceInfo, NetworkInterfaceType,
+};
 use crate::net::host::all_hosts;
 use crate::prelude::*;
 use crate::util::Invoke;
@@ -139,6 +143,34 @@ pub async fn add_tunnel(
             .result?;
     }
 
+    // Wait for the sync loop to fully commit gateway state (addresses, hosts)
+    // to the database, with a 15-second timeout.
+    if tokio::time::timeout(Duration::from_secs(15), async {
+        let mut watch = ctx
+            .db
+            .watch("/public/serverInfo/network".parse::<JsonPointer>().unwrap())
+            .await
+            .typed::<NetworkInfo>();
+        loop {
+            if watch
+                .peek()?
+                .as_gateways()
+                .as_idx(&iface)
+                .and_then(|g| g.as_ip_info().transpose_ref())
+                .is_some()
+            {
+                break;
+            }
+            watch.changed().await?;
+        }
+        Ok::<_, Error>(())
+    })
+    .await
+    .is_err()
+    {
+        tracing::warn!("{}", t!("net.tunnel.timeout-waiting-for-add", gateway = iface.as_str()));
+    }
+
     Ok(iface)
 }
 
@@ -224,5 +256,27 @@ pub async fn remove_tunnel(
         .await
         .result?;
 
+    // Wait for the sync loop to fully commit gateway removal to the database,
+    // with a 15-second timeout.
+    if tokio::time::timeout(Duration::from_secs(15), async {
+        let mut watch = ctx
+            .db
+            .watch("/public/serverInfo/network".parse::<JsonPointer>().unwrap())
+            .await
+            .typed::<NetworkInfo>();
+        loop {
+            if watch.peek()?.as_gateways().as_idx(&id).is_none() {
+                break;
+            }
+            watch.changed().await?;
+        }
+        Ok::<_, Error>(())
+    })
+    .await
+    .is_err()
+    {
+        tracing::warn!("{}", t!("net.tunnel.timeout-waiting-for-remove", gateway = id.as_str()));
+    }
+
     Ok(())
 }