On Wed, Sep 04, 2019 at 11:32:29AM +0800, Sun Ke wrote:
Add the test case to check nbd devices.This test case catches regressions
fixed by commit 92b5c8f0063e4 "nbd: replace kill_bdev() with
__invalidate_device() again".
Establish the nbd connection.Run two processes.One do mount and umount,
anther one do clear_sock ioctl.
Signed-off-by: Sun Ke <sunke32@xxxxxxxxxx>
Thanks for the test! A few comments below.
---
src/Makefile | 3 +-
src/nbdmount.c | 89 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
tests/nbd/003 | 47 +++++++++++++++++++++++++++++
tests/nbd/003.out | 1 +
tests/nbd/rc | 22 ++++++++++++++
5 files changed, 161 insertions(+), 1 deletion(-)
create mode 100644 src/nbdmount.c
create mode 100644 tests/nbd/003
create mode 100644 tests/nbd/003.out
diff --git a/src/Makefile b/src/Makefile
index 917d6f4..f48f264 100644
--- a/src/Makefile
+++ b/src/Makefile
@@ -10,7 +10,8 @@ C_TARGETS := \
sg/syzkaller1 \
nbdsetsize \
loop_change_fd \
- zbdioctl
+ zbdioctl \
+ nbdmount
This could use a less generic name.
CXX_TARGETS := \
discontiguous-io
diff --git a/src/nbdmount.c b/src/nbdmount.c
new file mode 100644
index 0000000..bd77c32
--- /dev/null
+++ b/src/nbdmount.c
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: GPL-3.0+
+// Copyright (C) 2019 Sun Ke
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+
+#include <linux/nbd.h>
+#include <assert.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <string.h>
+#include <sys/ioctl.h>
+#include <sys/mount.h>
+#include <linux/fs.h>
+
+struct mount_para
+{
+ char *mp;
+ char *dev;
+ char *fs;
+};
+
+struct mount_para para;
Get rid of this structure and global variable, just pass the fields to
ops_nbd().
+static int nbd_fd = -1;
Same here, this doesn't really need to be global.
+void setup_nbd(char *dev)
+{
+ nbd_fd = open(dev, O_RDWR);
+ if (nbd_fd < 0 ) {
+ printf("open the nbd failed\n");
Instead of printf(), do:
perror("open");
+ }
+}
+
+void teardown_nbd(void)
+{
+ close(nbd_fd);
+}
Fold these two trival functions (setup_nbd() and teardown_nbd()) into
main().
+void clear_sock(void)
+{
+ int err;
+
+ err = ioctl(nbd_fd, NBD_CLEAR_SOCK, 0);
+ assert(!err);
+}
perror("ioctl")
+void mount_nbd(char *dev, char *mp, char *fs)
+{
+ mount(dev, mp, fs, 2 | 16, 0);
What are these flags? ^^^^^^
+ umount(mp);
+}
+
+void ops_nbd(void)
+{
+ int i;
+
+ fflush(stdout);
What is this fflush() here for?
+ for (i=0; i < 2; i++) {
+ if (fork() == 0) {
+ if (i == 0) {
+ mount_nbd(para.dev, para.mp, para.fs);
+ exit(0);
+ }
+ if (i == 1) {
+ clear_sock();
+ exit(0);
+ }
+ }
+ }
This shouldn't be a loop. You can do:
if (fork() == 0) {
mount_nbd(dev, mp, fs);
exit(0);
}
if (fork() == 0) {
clear_sock(fd);
exit(0);
}
+ while(wait(NULL) > 0)
+ continue;
+}
+
+int main(int argc, char **argv)
+{
Check argc and print a usage message here, please:
if (argc != 4) {
fprintf(stderr, "usage: $0 MOUNTPOINT DEV FS");
return EXIT_FAILURE;
}
+ para.mp = argv[1];
+ para.dev = argv[2];
+ para.fs = argv[3];
+
+ setup_nbd(para.dev);
+ ops_nbd();
+ teardown_nbd();
+
+ return 0;
+}
diff --git a/tests/nbd/003 b/tests/nbd/003
new file mode 100644
index 0000000..1c2dcea
--- /dev/null
+++ b/tests/nbd/003
@@ -0,0 +1,47 @@
+#!/bin/bash
+
+# SPDX-License-Identifier: GPL-3.0+
+# Copyright (C) 2019 Sun Ke
+#
+# Test nbd device resizing. Regression test for patch
+#
+# 2b5c8f0063e4 ("nbd: replace kill_bdev() with __invalidate_device() again")
+
+. tests/nbd/rc
+
+DESCRIPTION="resize a connected nbd device"
+QUICK=1
+
+fs_type=ext4
+disk_capacity=256M
+run_cnt=1
+
+requires() {
+ _have_nbd && _have_src_program nbdmount
+}
+
+test() {
+ echo "Running ${TEST_NAME}"
+ for i in $(seq 0 15)
As documented in the new test template:
Use bash for loops instead of seq. E.g., for ((i = 0; i < 10; i++)), not
for i in $(seq 0 9).
+ do
+ _start_nbd_mount_server $disk_capacity $fs_type
+ nbd-client localhost 800$i /dev/nbd$i >> "$FULL" 2>&1
So the test creates 15 nbd connections, runs the test on all of them,
then tears them down? Would it work to only use one connection and just
repeat the test 15 times? That way you can keep track of the nbd-client
process to kill instead of having to use pkill later down.
+ if [ "$?" -ne "0" ]; then
Also from the new test template:
Use the bash [[ ]] form of tests instead of [ ].
+ echo "nbd$i connnect failed"
+ fi
+ done
+
+ for j in $(seq 0 $run_cnt)
+ do
+ for i in $(seq 0 15)
+ do
+ src/nbdmount "${TMPDIR}/$i" /dev/nbd$i $fs_type
Also from the new test template:
Always quote variable expansions unless the variable is a number or inside of
a [[ ]] test.
+ done
+ done
+
+ for i in $(seq 0 15)
+ do
+ nbd-client -d /dev/nbd$i
+ _stop_nbd_mount_server
+ done
+}
diff --git a/tests/nbd/003.out b/tests/nbd/003.out
new file mode 100644
index 0000000..aa340db
--- /dev/null
+++ b/tests/nbd/003.out
@@ -0,0 +1 @@
+Running nbd/003
diff --git a/tests/nbd/rc b/tests/nbd/rc
index 9d0e3d1..eb7ff24 100644
--- a/tests/nbd/rc
+++ b/tests/nbd/rc
@@ -76,3 +76,25 @@ _stop_nbd_server() {
rm -f "${TMPDIR}/nbd.pid"
rm -f "${TMPDIR}/export"
}
+
+
+_start_nbd_mount_server() {
+
+ fallocate -l $1 "${TMPDIR}/mnt_$i"
+
+ if [ "$2"x = "ext4"x ];
+ then
+ mkfs.ext4 "${TMPDIR}/mnt_$i" >> "$FULL" 2>&1
+ else
+ mkdosfs "${TMPDIR}/mnt_$i" >> "$FULL" 2>&1
+ fi
+ nbd-server 800$i "${TMPDIR}/mnt_$i" >> "$FULL" 2>&1
+
+ mkdir -p "${TMPDIR}/$i"
+}
+
+_stop_nbd_mount_server() {
+ pkill -9 -f 800$i
+ rm -f "${TMPDIR}/mnt_$i"
+ rm -rf "${TMPDIR}/$i"
+}
These helpers are using a variable ($i) that happens to be set by the
calling function. That's really icky. Just fold these into your test
case, we can make them generic helpers if someone else needs them.
.
