fix(volume): keep verity volumes out of the data disk search

CHANGELOG.md

@@ -7,6 +7,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed
+- guest: a fresh CVM with a verity volume could take the volume for its data disk and try to LUKS-format it: volumes labelled their data partition `dstack-data`, the data disk's label. `dstack verity` now labels it `dstack-volume-data`, and the guest skips disks that carry a `dstack-volume` partition, so volumes built by older releases are safe too
+
 ### Changed
 - vmm: `qmp_socket` now defaults to `true`. Each CVM's QEMU exposes `qmp.sock` in its work dir, so the host can change disk options or compact a bloated `hda.img` (`blockdev-mirror`) without stopping the CVM. Applies to CVMs started after the change; set `qmp_socket = false` to opt out
 

docs/verity-volumes.md

@@ -9,11 +9,16 @@ hash measured in `app-compose.json`.
 A built volume is a raw GPT image:
 
 ```text
-p1: DSTACK_VOLUME metadata envelope
-p2: filesystem data
-p3: dm-verity superblock and hash tree
+p1 dstack-volume:      DSTACK_VOLUME metadata envelope
+p2 dstack-volume-data: filesystem data
+p3 dstack-verity:      dm-verity superblock and hash tree
 ```
 
+The guest opens the partitions by number; the labels only keep a volume apart
+from the app data disk, whose partition is `dstack-data`. Older volumes label
+p2 `dstack-data` too; the guest skips any disk with a `dstack-volume`
+partition when it looks for the data disk.
+
 The envelope identifies a candidate disk. It is not trusted. The guest passes
 the root from the measured app compose to `veritysetup`, so a forged envelope or
 partition table cannot substitute different contents.

dstack/crates/dstack-volume/src/volume.rs

@@ -332,7 +332,7 @@ fn write_gpt(mut out: fs::File, layout: &GptLayout, root_hash: &str) -> Result<f
             first_lba: layout.data.first_lba,
             last_lba: layout.data.last_lba,
             flags: 0,
-            name: "dstack-data".to_string(),
+            name: "dstack-volume-data".to_string(),
         },
     );
     parts.insert(
@@ -455,7 +455,7 @@ mod tests {
         let data_partition = disk.partitions().get(&2).unwrap();
         let hash_partition = disk.partitions().get(&3).unwrap();
         assert_eq!(metadata.name, "dstack-volume");
-        assert_eq!(data_partition.name, "dstack-data");
+        assert_eq!(data_partition.name, "dstack-volume-data");
         assert_eq!(hash_partition.name, "dstack-verity");
         assert_eq!(
             data_partition.part_guid,

os/common/rootfs/dstack-prepare.sh

@@ -52,6 +52,36 @@ find_block_by_property() {
 	return 1
 }
 
+# A verity volume's first partition is labelled dstack-volume. Volumes built
+# before dstack-volume-data labelled their data partition dstack-data, the app
+# data disk's label: never take a partition on such a disk for the data disk.
+is_verity_volume_disk() {
+	local disk="$1" part
+	for part in /sys/class/block/"$disk"/"$disk"*; do
+		[ -e "$part/uevent" ] || continue
+		[ "$(read_uevent_property "$part/uevent" PARTNAME || true)" = dstack-volume ] && return 0
+	done
+	return 1
+}
+
+# Print the dstack-data partition. Returns 1 if there is none, 2 if there is
+# more than one.
+find_data_partition() {
+	local entry found=""
+	for entry in /sys/class/block/*; do
+		[ -e "$entry/partition" ] || continue
+		[ "$(read_uevent_property "$entry/uevent" PARTNAME || true)" = dstack-data ] || continue
+		is_verity_volume_disk "$(basename "$(dirname "$(readlink -f "$entry")")")" && continue
+		if [ -n "$found" ]; then
+			log "Error: more than one dstack-data partition: $found /dev/$(basename "$entry")"
+			return 2
+		fi
+		found="/dev/$(basename "$entry")"
+	done
+	[ -n "$found" ] || return 1
+	printf "%s" "$found"
+}
+
 resolve_block_spec() {
 	local spec="$1"
 	local device=""
@@ -203,8 +233,9 @@ choose_data_device() {
 	fi
 
 	# 2. Try to find partition with PARTLABEL=dstack-data
-	local data_disk
-	data_disk=$(resolve_block_spec "PARTLABEL=dstack-data" || true)
+	local data_disk rc=0
+	data_disk=$(find_data_partition) || rc=$?
+	[ "$rc" -ne 2 ] || return 1
 	if [ -n "$data_disk" ]; then
 		echo "$data_disk"
 		return 0