1 files changed, 53 insertions, 0 deletions
diff --git a/qa/workunits/rbd/krbd_watch_errors.sh b/qa/workunits/rbd/krbd_watch_errors.sh
new file mode 100755
index 000000000..f650d2a74
--- /dev/null
+++ b/qa/workunits/rbd/krbd_watch_errors.sh
@@ -0,0 +1,53 @@
+#!/usr/bin/env bash
+
+set -ex
+set -o pipefail
+
+function refresh_loop() {
+    local dev_id="$1"
+
+    set +x
+
+    local i
+    for ((i = 1; ; i++)); do
+        echo 1 | sudo tee "${SYSFS_DIR}/${dev_id}/refresh" > /dev/null
+        if ((i % 100 == 0)); then
+            echo "Refreshed ${i} times"
+        fi
+    done
+}
+
+readonly SYSFS_DIR="/sys/bus/rbd/devices"
+readonly IMAGE_NAME="watch-errors-test"
+
+rbd create -s 1G --image-feature exclusive-lock "${IMAGE_NAME}"
+
+# induce a watch error every 30 seconds
+dev="$(sudo rbd device map -o osdkeepalive=60 "${IMAGE_NAME}")"
+dev_id="${dev#/dev/rbd}"
+
+# constantly refresh, not just on watch errors
+refresh_loop "${dev_id}" &
+refresh_pid=$!
+
+sudo dmesg -C
+
+# test that none of the above triggers a deadlock with a workload
+fio --name test --filename="${dev}" --ioengine=libaio --direct=1 \
+    --rw=randwrite --norandommap --randrepeat=0 --bs=512 --iodepth=128 \
+    --time_based --runtime=1h --eta=never
+
+num_errors="$(dmesg | grep -c "rbd${dev_id}: encountered watch error")"
+echo "Recorded ${num_errors} watch errors"
+
+kill "${refresh_pid}"
+wait
+
+sudo rbd device unmap "${dev}"
+
+if ((num_errors < 60)); then
+    echo "Too few watch errors"
+    exit 1
+fi
+
+echo OK