fs/ext4: enhance logic of ext4_nsec_timestamps_test

1. Check the big block devices at setup stage, umount it if it's mounted

2. Add "sleep 1" between mount and umount, to avoid umount failed due to device busy (accessed by system service like systemd-udev)

3. Add TFAIL if umount failed
4. Fix typo error

Signed-off-by Yixin Zhang <[email protected]>

---
 .../ext4-nsec-timestamps/ext4_nsec_timestamps_test.sh      | 14 +++++++++-----
 testcases/kernel/fs/ext4-new-features/ext4_funcs.sh        |  6 ++++++
 2 files changed, 15 insertions(+), 5 deletions(-)

testcases/kernel/fs/ext4-new-features/ext4-nsec-timestamps/ext4_nsec_timestamps_test.sh +9 -5

index c6ff7c2ba..5d561adf3 100755
@@ -76,8 +76,8 @@ ext4_test_nsec_timestamps()
 
 	mkfs.ext3 -I 256 $EXT4_DEV >/dev/null 2>&1
 	if [ $? -ne 0 ]; then
-		tst_resm TFAIL "failed to create ext4 filesystem"
-		return
+		tst_resm TFAIL "failed to create ext3 filesystem"
+	  return
 	fi
 	mount -t ext4 $EXT4_DEV mnt_point

@@ -105,8 +105,7 @@ ext4_test_nsec_timestamps()
 	nsec_ctime=`ext4_file_time mnt_point/tmp_file ctime nsec`
 
 	# Test nanosecond
-	if [ $nsec_atime -eq 0 -a $nsec_mtime -eq 0 -a $nsec_ctime -eq 0 ]
-	then
+	if [ $nsec_atime -eq 0 -a $nsec_mtime -eq 0 -a $nsec_ctime -eq 0 ]; then
 		tst_resm TFAIL "The timestamp is not nanosecond(nsec_atime: $nsec_atime, nsec_mtime: $nsec_mtime, nsec_ctime: $nsec_ctime)"
 		umount mnt_point
 		return

@@ -138,7 +137,12 @@ ext4_test_nsec_timestamps()
 		tst_resm TFAIL "failed to mount to ext3"
 		return
 	fi
+	sleep 1
 	umount mnt_point
+	if [ $? -ne 0 ]; then
+		tst_resm TFAIL "failed to umount ext3 filesystem"
+		return
+	fi
 
 	mount -t ext4 $EXT4_DEV mnt_point
 	if [ $? -ne 0 ]; then

@@ -148,7 +152,7 @@ ext4_test_nsec_timestamps()
 
 	nsec_atime2=`ext4_file_time mnt_point/tmp_file atime nsec`
 	nsec_mtime2=`ext4_file_time mnt_point/tmp_file mtime nsec`
-	nsec_ctime2=`ext4_file_time mnt_point/tmp_file mtime nsec`
+	nsec_ctime2=`ext4_file_time mnt_point/tmp_file ctime nsec`
 
 	if [ $nsec_atime -ne $nsec_atime2 -o $nsec_ctime -ne $nsec_ctime2 -o \
 	     $nsec_mtime -ne $nsec_mtime2 ]; then

testcases/kernel/fs/ext4-new-features/ext4_funcs.sh +6 -0

index a9eb54e8d..4de7899e4 100755
@@ -45,6 +45,12 @@ ext4_setup()
 		tst_brkm TCONF "tests need a big block device(5G-10G)"
 	else
 		export EXT4_DEV=$LTP_BIG_DEV
+		if mount | cut -d' ' -f1 | grep -q ^$EXT4_DEV$ ; then
+			umount $EXT4_DEV >/dev/null 2>&1
+			if [ $? -ne 0 ]; then
+				tst_brkm TCONF "umount $EXT4_DEV failed"
+			fi
+		fi
 	fi
 
 	tst_tmpdir

Signature · 5 lines

2.14.1


-- 
Mailing list info: https://lists.linux.it/listinfo/ltp

Quoted 2 lines

2. Add "sleep 1" between mount and umount, to avoid umount failed due to device busy (accessed by system service like systemd-udev)

NACK to this.

In a case that we need to wait for something like this we are supposed to retry in a loop with exponential backoff, which is usually faster than hardcoded sleep but also more robust.

There is a in-flight patch for adding TST_RETRY_FUNC function to the shell library "[PATCH] tst_test.sh: achieve TST_RETRY_FUNC function in shell" which should make this really easy but we would have to add it to the old shell test library as well so that we can reuse it here.

Quoted 5 lines

Quoted 2 lines

2. Add "sleep 1" between mount and umount, to avoid umount failed due to device busy (accessed by system service like systemd-udev)

NACK to this.

Quoted 8 lines

In a case that we need to wait for something like this we are supposed to retry in a loop with exponential backoff, which is usually faster than hardcoded sleep but also more robust.

There is a in-flight patch for adding TST_RETRY_FUNC function to the shell library "[PATCH] tst_test.sh: achieve TST_RETRY_FUNC function in shell" which should make this really easy but we would have to add it to the old shell test library as well so that we can reuse it here.

And I remebered that we already solved this by adding tst_umount function to the shell library. The systemd-udev is not first daemon to probe newly mounted filesystem, we used to have this kind of problems with gvfsd-trash as well.

So fixing this should be as easy as replacing the umount with tst_umount function.

Quoted 7 lines

And I remebered that we already solved this by adding tst_umount function to the shell library. The systemd-udev is not first daemon to probe newly mounted filesystem, we used to have this kind of problems with gvfsd-trash as well.

So fixing this should be as easy as replacing the umount with tst_umount function.

I'm using tst_umount in IMA tests and it works nice :).
BTW It's just that gvfsd-trash TINFO message is a bit too specific (in my case it was that some processes were still using the mounted filesystem - the same issue as Yixin has).

Hi!

Quoted 10 lines

Quoted 7 lines

And I remebered that we already solved this by adding tst_umount function to the shell library. The systemd-udev is not first daemon to probe newly mounted filesystem, we used to have this kind of problems with gvfsd-trash as well.

So fixing this should be as easy as replacing the umount with tst_umount function.

I'm using tst_umount in IMA tests and it works nice :).
BTW It's just that gvfsd-trash TINFO message is a bit too specific (in my case it was that some processes were still using the mounted filesystem - the same issue as Yixin has).

Yes, it is, we may want to reword it, but keep that after the release.

Signature · 2 lines

Cyril Hrubis
[email protected]