summaryrefslogtreecommitdiffstats
path: root/qa/workunits/rbd/krbd_udev_netlink_enobufs.sh
diff options
context:
space:
mode:
Diffstat (limited to 'qa/workunits/rbd/krbd_udev_netlink_enobufs.sh')
-rwxr-xr-xqa/workunits/rbd/krbd_udev_netlink_enobufs.sh24
1 files changed, 24 insertions, 0 deletions
diff --git a/qa/workunits/rbd/krbd_udev_netlink_enobufs.sh b/qa/workunits/rbd/krbd_udev_netlink_enobufs.sh
new file mode 100755
index 000000000..7c9c53a2f
--- /dev/null
+++ b/qa/workunits/rbd/krbd_udev_netlink_enobufs.sh
@@ -0,0 +1,24 @@
+#!/usr/bin/env bash
+
+# This is a test for https://tracker.ceph.com/issues/41404, verifying that udev
+# events are properly reaped while the image is being (un)mapped in the kernel.
+# UDEV_BUF_SIZE is 1M (giving us a 2M socket receive buffer), but modprobe +
+# modprobe -r generate ~28M worth of "block" events.
+
+set -ex
+
+rbd create --size 1 img
+
+ceph osd pause
+sudo rbd map img &
+PID=$!
+sudo modprobe scsi_debug max_luns=16 add_host=16 num_parts=1 num_tgts=16
+sudo udevadm settle
+sudo modprobe -r scsi_debug
+[[ $(rbd showmapped | wc -l) -eq 0 ]]
+ceph osd unpause
+wait $PID
+[[ $(rbd showmapped | wc -l) -eq 2 ]]
+sudo rbd unmap img
+
+echo OK