fix(e2e): resolve DRBD devices via drbdadm sh-dev in remaining cli-matrix cells

tests/e2e/cli-matrix/r-activate-deactivate-lifecycle.sh

@@ -50,7 +50,7 @@
 #
 #     2. IO during INACTIVE:
 #        - Promote N1 to Primary if not already
-#        - Write ~64 MiB random pattern to /dev/drbd/by-res/<RD>/0
+#        - Write ~64 MiB random pattern to the RD's /dev/drbdN device
 #        - Must complete without quorum block (single-replica quorum
 #          override). Capture md5 of pattern.
 #
@@ -153,9 +153,14 @@ wait_role "$RD" "$N1" "Primary" 30 \
 # written BEFORE the first deactivate so both replicas hold the
 # same bytes and the GI baseline is established.
 echo ">> seed initial ${IO_MIB} MiB pattern on $N1 (will be the md5 anchor)"
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so readlink-based resolution aborts on the stand.
+# Last-resort minor enumeration kept for stands where sh-dev fails.
+dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
 on_node "$N1" bash -c "
     set -e
-    dev=\$(readlink -f /dev/drbd/by-res/${RD}/0 2>/dev/null || true)
+    dev='$dev'
     if [ -z \"\$dev\" ]; then
         dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
     fi
@@ -289,9 +294,11 @@ for cycle in 1 2 3; do
     # the sole voter and DRBD's single-replica quorum override must
     # let writes through. A failure here = quorum frame is wrong on
     # INACTIVE peers.
+    # Same portable resolver as the seeding step (sh-dev, not by-res).
+    dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
     io_out=$(on_node "$N1" bash -c "
         set -e
-        dev=\$(readlink -f /dev/drbd/by-res/${RD}/0 2>/dev/null || true)
+        dev='$dev'
         if [ -z \"\$dev\" ]; then
             dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
         fi
@@ -317,8 +324,9 @@ for cycle in 1 2 3; do
     # — the deactivate path must not have corrupted it. (md5 of the
     # FIRST ${IO_MIB} MiB of the device, since the seed lives at
     # offset 0 and the new write was at offset ${IO_MIB} MiB.)
+    dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
     seed_md5_now=$(on_node "$N1" bash -c "
-        dev=\$(readlink -f /dev/drbd/by-res/${RD}/0 2>/dev/null || true)
+        dev='$dev'
         if [ -z \"\$dev\" ]; then
             dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
         fi
@@ -424,8 +432,9 @@ for cycle in 1 2 3; do
     # the pre-deact pattern. The partial-sync handshake must not
     # have written stale bytes onto the surviving replica's data.
     echo ">>   md5 of seed on $N1 still matches pre-deact pattern"
+    dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
     seed_md5_after=$(on_node "$N1" bash -c "
-        dev=\$(readlink -f /dev/drbd/by-res/${RD}/0 2>/dev/null || true)
+        dev='$dev'
         if [ -z \"\$dev\" ]; then
             dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
         fi

tests/e2e/cli-matrix/r-c-over-tiebreaker-skip-sync.sh

@@ -224,7 +224,11 @@ fi
 # Write 32 MiB; secondary will need to catch up. With skip-sync
 # the catch-up is fast but still passes through SyncSource on
 # the source side per upstream events2 semantics.
-on_node "$prim" bash -c "dd if=/dev/urandom of=/dev/drbd/by-res/$RD/0 bs=1M count=32 status=none oflag=direct 2>/dev/null" || true
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so the by-res dd silently no-ops on the stand.
+dev=$(resolve_drbd_device "$prim" "$RD" 0 2>/dev/null) || dev=""
+[ -n "$dev" ] && on_node "$prim" bash -c "dd if=/dev/urandom of=$dev bs=1M count=32 status=none oflag=direct 2>/dev/null" || true
 
 # Capture wire-shape for ~10s post-mutation.
 shape_ok=false

tests/e2e/cli-matrix/r-d-last-uptodate-midsync-rejected.sh

@@ -82,7 +82,11 @@ echo ">> Phase 2: write data on $N1 so the second replica must really resync"
 # Primary + dd a chunk to bump the GI; a fresh empty volume could
 # skip-sync and erase the SyncTarget window.
 on_node "$N1" bash -c "drbdadm primary --force $RD 2>/dev/null" || true
-on_node "$N1" bash -c "dd if=/dev/urandom of=/dev/drbd/by-res/$RD/0 bs=1M count=256 status=none oflag=direct 2>/dev/null" || true
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so the by-res dd silently no-ops on the stand.
+dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
+[ -n "$dev" ] && on_node "$N1" bash -c "dd if=/dev/urandom of=$dev bs=1M count=256 status=none oflag=direct 2>/dev/null" || true
 on_node "$N1" bash -c "drbdadm secondary $RD 2>/dev/null" || true
 
 echo ">> Phase 3: throttle resync (c-max-rate 1024 KiB/s) so the add stays SyncTarget"

tests/e2e/cli-matrix/snap-create-multiple-group-consistency.sh

@@ -114,9 +114,14 @@ on_node "$N1" drbdadm primary --force "$RD_B" 2>/dev/null || true
 # different counter values whenever the writer made progress
 # between the two per-RD snap calls.
 echo ">> start cross-RD correlated writer on $N1 (counter into rd-a + rd-b)"
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so readlink-based resolution aborts on the stand.
+dev_a=$(resolve_drbd_device "$N1" "$RD_A" 0 2>/dev/null) || dev_a=""
+dev_b=$(resolve_drbd_device "$N1" "$RD_B" 0 2>/dev/null) || dev_b=""
 on_node "$N1" bash -c "
-    dev_a=\$(readlink -f /dev/drbd/by-res/$RD_A/0 2>/dev/null || true)
-    dev_b=\$(readlink -f /dev/drbd/by-res/$RD_B/0 2>/dev/null || true)
+    dev_a='$dev_a'
+    dev_b='$dev_b'
     if [ -z \"\$dev_a\" ] || [ -z \"\$dev_b\" ]; then
         echo 'note: could not resolve drbd device paths'
         exit 0

tests/e2e/cli-matrix/snap-create-multiple-lifecycle.sh

@@ -119,10 +119,16 @@ done
 # Phase 2: cross-RD correlated writer (counter into bytes 0-7 of all 3)
 # =====================================================================
 echo ">> Phase 2: start cross-RD correlated writer on $N1"
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so readlink-based resolution aborts on the stand.
+dev_a=$(resolve_drbd_device "$N1" "$RD_A" 0 2>/dev/null) || dev_a=""
+dev_b=$(resolve_drbd_device "$N1" "$RD_B" 0 2>/dev/null) || dev_b=""
+dev_c=$(resolve_drbd_device "$N1" "$RD_C" 0 2>/dev/null) || dev_c=""
 on_node "$N1" bash -c "
-    dev_a=\$(readlink -f /dev/drbd/by-res/$RD_A/0 2>/dev/null || true)
-    dev_b=\$(readlink -f /dev/drbd/by-res/$RD_B/0 2>/dev/null || true)
-    dev_c=\$(readlink -f /dev/drbd/by-res/$RD_C/0 2>/dev/null || true)
+    dev_a='$dev_a'
+    dev_b='$dev_b'
+    dev_c='$dev_c'
     if [ -z \"\$dev_a\" ] || [ -z \"\$dev_b\" ] || [ -z \"\$dev_c\" ]; then
         echo 'note: could not resolve all 3 drbd device paths'
         exit 0

tests/e2e/cli-matrix/snap-cross-node-consistency.sh

@@ -91,9 +91,14 @@ on_node "$N1" drbdadm primary --force "$RD" 2>/dev/null || true
 # capture (one snap took data at byte N, the other at byte N+delta)
 # yields visibly different md5.
 echo ">> seed deterministic 256 MiB pattern on $N1's DRBD device"
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so readlink-based resolution aborts on the stand.
+# Last-resort minor enumeration kept for stands where sh-dev fails.
+dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
 on_node "$N1" bash -c "
     set -e
-    dev=\$(readlink -f /dev/drbd/by-res/$RD/0 2>/dev/null || true)
+    dev='$dev'
     if [ -z \"\$dev\" ]; then
         dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
     fi
@@ -111,8 +116,10 @@ wait_uptodate "$RD" "$N1" "$N2"
 # AFTER replica $N1 already finished its snapshot — and the two
 # resulting snapshots reflect that delta.
 echo ">> start continuous writer on $N1 (urandom → DRBD device)"
+dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
 on_node "$N1" bash -c "
-    dev=\$(readlink -f /dev/drbd/by-res/$RD/0 2>/dev/null || ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
+    dev='$dev'
+    [ -n \"\$dev\" ] || dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
     while true; do
         dd if=/dev/urandom of=\$dev bs=4K count=128 oflag=direct status=none 2>/dev/null || break
     done >/tmp/cli-matrix-snap-writer.log 2>&1 &

tests/e2e/cli-matrix/snap-full-lifecycle.sh

@@ -115,9 +115,14 @@ on_node "$N1" drbdadm primary --force "$RD" 2>/dev/null || true
 #   We therefore read the device's actual byte size and feed dd via
 #   count_bytes so the write stops exactly at the DRBD boundary.
 echo ">> Phase 2: seed deterministic random pattern on $N1 (DRBD-fit bytes), then start writer"
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so readlink-based resolution aborts on the stand.
+# Last-resort minor enumeration kept for stands where sh-dev fails.
+dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
 seed_out=$(on_node "$N1" bash -c "
     set -e
-    dev=\$(readlink -f /dev/drbd/by-res/$RD/0 2>/dev/null || true)
+    dev='$dev'
     if [ -z \"\$dev\" ]; then
         dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
     fi
@@ -136,8 +141,10 @@ echo "$seed_out"
 wait_uptodate "$RD" "$N1" "$N2"
 
 echo ">> Phase 2: start continuous writer on $N1"
+dev=$(resolve_drbd_device "$N1" "$RD" 0 2>/dev/null) || dev=""
 on_node "$N1" bash -c "
-    dev=\$(readlink -f /dev/drbd/by-res/$RD/0 2>/dev/null || ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
+    dev='$dev'
+    [ -n \"\$dev\" ] || dev=\$(ls -1 /dev/drbd* 2>/dev/null | grep -vE 'by-(res|disk)' | head -1)
     while true; do
         dd if=/dev/urandom of=\$dev bs=4K count=128 oflag=direct status=none 2>/dev/null || break
     done >/tmp/cli-matrix-snap-lifecycle-writer.log 2>&1 &

tests/e2e/cli-matrix/snap-r-rst-stamps-resources.sh

@@ -101,8 +101,12 @@ wait_uptodate "$SRC" "$N1" "$N2"
 # call and asserts Resource CRDs are stamped).
 echo ">> seed deterministic pattern on $N1 $SRC"
 on_node "$N1" drbdadm primary --force "$SRC" 2>/dev/null || true
+# Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+# /dev/drbd/by-res symlink is not reliably present in the satellite
+# mount namespace, so readlink-based resolution aborts on the stand.
+dev=$(resolve_drbd_device "$N1" "$SRC" 0 2>/dev/null) || dev=""
 on_node "$N1" bash -c "
-    dev=\$(readlink -f /dev/drbd/by-res/$SRC/0 2>/dev/null || true)
+    dev='$dev'
     if [ -n \"\$dev\" ]; then
         printf 'BLOCKSTOR-BUG354-MARKER' | dd of=\"\$dev\" bs=1 count=24 conv=fsync status=none
     fi
@@ -212,10 +216,13 @@ fi
 
 # ---- Bonus: read the marker on the restored replica ----------------------
 echo ">> bonus assert: marker bytes restored from snapshot on $N1 $TGT"
+# Same portable resolver as the seeding step: by-res symlinks are
+# not reliably present in the satellite mount namespace.
+dev=$(resolve_drbd_device "$N1" "$TGT" 0 2>/dev/null) || dev=""
 marker_read=$(on_node "$N1" bash -c "
     on_node_drbdadm() { drbdadm primary --force \$1 2>/dev/null; }
     on_node_drbdadm $TGT
-    dev=\$(readlink -f /dev/drbd/by-res/$TGT/0 2>/dev/null || true)
+    dev='$dev'
     if [ -n \"\$dev\" ]; then
         head -c 24 \"\$dev\" 2>/dev/null
     fi

tests/e2e/cli-matrix/snap-suspend-resume-isolation-u138-u52.sh

@@ -112,10 +112,14 @@ snap_ready() {
 # timeout. A still-suspended device makes dd block; `timeout` then kills
 # it and we FAIL — that is the U138 outage signal.
 write_survives() {
-    local rd=$1
+    local rd=$1 dev
     on_node "$N1" drbdadm primary --force "$rd" 2>/dev/null || true
+    # Resolve via `drbdadm sh-dev` (lib.sh resolve_drbd_device): the
+    # /dev/drbd/by-res symlink is not reliably present in the satellite
+    # mount namespace, so readlink-based resolution aborts on the stand.
+    dev=$(resolve_drbd_device "$N1" "$rd" 0 2>/dev/null) || dev=""
     if ! on_node "$N1" bash -c "
-        dev=\$(readlink -f /dev/drbd/by-res/$rd/0 2>/dev/null || true)
+        dev='$dev'
         [ -z \"\$dev\" ] && { echo 'no drbd device node for $rd' >&2; exit 2; }
         timeout 20 dd if=/dev/zero of=\"\$dev\" bs=4096 count=16 oflag=direct conv=fsync status=none
     "; then