| Message ID | 20220727085251.1474340-3-shinichiro.kawasaki@wdc.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fix module check issues | expand |
On Wed, Jul 27, 2022 at 05:52:47PM +0900, Shin'ichiro Kawasaki wrote: > After the commit "common/rc: avoid module load in _have_driver()", > _have_driver() no longer loads specified module. However, nbd test cases > and _have_nbd_netlink() function assume that the module is loaded by > calling _have_driver(). This causes test case failures and unexpected > skips. To fix them, load and unload modules explicitly in functions > _start_nbd_server*(), _stop_nbd_server*() and _have_nbd_netlink(). Did you test this with built-in nbd?
On Jul 28, 2022 / 16:46, Christoph Hellwig wrote: > On Wed, Jul 27, 2022 at 05:52:47PM +0900, Shin'ichiro Kawasaki wrote: > > After the commit "common/rc: avoid module load in _have_driver()", > > _have_driver() no longer loads specified module. However, nbd test cases > > and _have_nbd_netlink() function assume that the module is loaded by > > calling _have_driver(). This causes test case failures and unexpected > > skips. To fix them, load and unload modules explicitly in functions > > _start_nbd_server*(), _stop_nbd_server*() and _have_nbd_netlink(). > > Did you test this with built-in nbd? Yes. I confirmed all test cases in nbd test group pass with built-in nbd. The modprobe command with -q option does nothing for built-in drivers. When I confirmed that, I found even nbd/004 passes, which should be skipped. Next patch addresses this issue.
diff --git a/tests/nbd/rc b/tests/nbd/rc index 9c1c15b..32eea45 100644 --- a/tests/nbd/rc +++ b/tests/nbd/rc @@ -28,17 +28,21 @@ _have_nbd() { } _have_nbd_netlink() { + local ret=0 + if ! _have_nbd; then return 1 fi if ! _have_program genl-ctrl-list; then return 1 fi + modprobe -q nbd if ! genl-ctrl-list | grep -q nbd; then SKIP_REASONS+=("nbd does not support netlink") - return 1 + ret=1 fi - return 0 + modprobe -qr nbd + return $ret } _wait_for_nbd_connect() { @@ -62,6 +66,7 @@ _wait_for_nbd_disconnect() { } _start_nbd_server() { + modprobe -q nbd truncate -s 10G "${TMPDIR}/export" cat > "${TMPDIR}/nbd.conf" << EOF [generic] @@ -73,17 +78,20 @@ EOF _stop_nbd_server() { kill -SIGTERM "$(cat "${TMPDIR}/nbd.pid")" + modprobe -qr nbd rm -f "${TMPDIR}/nbd.pid" rm -f "${TMPDIR}/export" } _start_nbd_server_netlink() { + modprobe -q nbd truncate -s 10G "${TMPDIR}/export" nbd-server 8000 "${TMPDIR}/export" >/dev/null 2>&1 } _stop_nbd_server_netlink() { killall -SIGTERM nbd-server + modprobe -qr nbd rm -f "${TMPDIR}/export" }
After the commit "common/rc: avoid module load in _have_driver()", _have_driver() no longer loads specified module. However, nbd test cases and _have_nbd_netlink() function assume that the module is loaded by calling _have_driver(). This causes test case failures and unexpected skips. To fix them, load and unload modules explicitly in functions _start_nbd_server*(), _stop_nbd_server*() and _have_nbd_netlink(). Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com> --- tests/nbd/rc | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-)