Skip to contentSkip to conversation

generic/258 questions (mount issue)...

PATCH New Michael L. Semon 4 participants 11 messages May 22, 2013 – May 23, 2013

lore
Share & tools
lore
#0 lore raw

Hi! When using xfstests generic/258 with along with $TEST_RTDEV $TEST_LOGDEV, it tends to scream bloody murder about corrupted partitions and such. In fact, the commands in the test seem to do the right thing when executed by hand. So once again, I grasped for straws and came up with this:

tests/generic/258 renamed from tests/generic/258.orig +1 -1
@@ -62,7 +62,7 @@
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"

My questions are these:

1) Was there a better way to do this?

2) Not knowing the policy on umounting $TEST_DEV, could this have been a test for $SCRATCH_DEV?

This was tested only on XFS with rtdev and logdev partitions enabled.

Thanks!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#1 lore raw
Quoted 18 lines · Michael L. Semon

Hi! When using xfstests generic/258 with along with $TEST_RTDEV $TEST_LOGDEV, it tends to scream bloody murder about corrupted partitions and such. In fact, the commands in the test seem to do the right thing when executed by hand. So once again, I grasped for straws and came up with this:

tests/generic/258 renamed from tests/generic/258.orig +1 -1
@@ -62,7 +62,7 @@
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"
Quoted 3 lines

My questions are these:

1) Was there a better way to do this?

No, your change is correct. Can you clean up the description of the problem you had and add a Signed-off-by?

Quoted 2 lines

2) Not knowing the policy on umounting $TEST_DEV, could this have been a test for $SCRATCH_DEV?

There are a handful of other tests that also unmount the TEST_DEV. Perhaps adding a _test_umount() wrapper to common/rc (similar to _scratch_umount) would be best. At least shared/243 needs the same _test_mount treatment as this test.

Cheers,

Dave.

Signature · 7 lines
Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#2 lore raw
Quoted 29 lines · Dave Chinner
Quoted 18 lines · Michael L. Semon

Hi! When using xfstests generic/258 with along with $TEST_RTDEV $TEST_LOGDEV, it tends to scream bloody murder about corrupted partitions and such. In fact, the commands in the test seem to do the right thing when executed by hand. So once again, I grasped for straws and came up with this:

tests/generic/258 renamed from tests/generic/258.orig +1 -1
@@ -62,7 +62,7 @@
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"
Quoted 3 lines

My questions are these:

1) Was there a better way to do this?

No, your change is correct. Can you clean up the description of the problem you had and add a Signed-off-by?

Quoted 2 lines

2) Not knowing the policy on umounting $TEST_DEV, could this have been a test for $SCRATCH_DEV?

Quoted 9 lines

There are a handful of other tests that also unmount the TEST_DEV. Perhaps adding a _test_umount() wrapper to common/rc (similar to _scratch_umount) would be best. At least shared/243 needs the same _test_mount treatment as this test.

Cheers,

Dave.

This is my first attempt at doing a git format-patch --stdout. Hopefully, this is somewhat more presentable to git. Comments and brow-beating are welcome.

Thanks!

Michael

Quoted 1 line

From 7dc0667e1f8cd1c98f15ebf412dd0f038b55306e Mon Sep 17 00:00:00 2001

From: "Michael L. Semon" <[email protected]>
Date: Tue, 21 May 2013 22:50:59 -0400
Subject: [PATCH] Change mount method for generic/258

Use the built-in _test_mount function from xfstests so it will use the correct mount options for xfstests. The script used a simple umount-and-mount sequence, which caused a test failure on an XFS filesystem that used both realtime and external log devices.

---
 tests/generic/258 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/generic/258 +1 -1
index fb091ae..3aeb5ce 100755
@@ -62,7 +62,7 @@ fi
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"
Signature · 7 lines
1.8.2


_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Reviewed-by
#3 lore raw

On 5/21/13 10:03 PM, Michael L. Semon wrote:

...

Quoted 11 lines

From 7dc0667e1f8cd1c98f15ebf412dd0f038b55306e Mon Sep 17 00:00:00 2001
From: "Michael L. Semon" <[email protected]>
Date: Tue, 21 May 2013 22:50:59 -0400
Subject: [PATCH] Change mount method for generic/258

Use the built-in _test_mount function from xfstests so it will use the correct mount options for xfstests. The script used a simple umount-and-mount sequence, which caused a test failure on an XFS filesystem that used both realtime and external log devices.

Thanks, this was my fault :(

Want to consider dave's other suggestions in another patch?

Quoted 4 lines

There are a handful of other tests that also unmount the TEST_DEV. Perhaps adding a _test_umount() wrapper to common/rc (similar to _scratch_umount) would be best. At least shared/243 needs the same _test_mount treatment as this test.

Eternal git-log fame would be yours. ;)

-Eric

Quoted 10 lines
---
 tests/generic/258 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/generic/258 +0 -0
index fb091ae..3aeb5ce 100755

tests/generic/258 fi · line 68

Quoted 8 lines
tests/generic/258 +1 -1
@@ -62,7 +62,7 @@ fi
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#4 lore raw
Quoted 22 lines · Eric Sandeen

On 5/21/13 10:03 PM, Michael L. Semon wrote:

...

Quoted 11 lines

From 7dc0667e1f8cd1c98f15ebf412dd0f038b55306e Mon Sep 17 00:00:00 2001
From: "Michael L. Semon" <[email protected]>
Date: Tue, 21 May 2013 22:50:59 -0400
Subject: [PATCH] Change mount method for generic/258

Use the built-in _test_mount function from xfstests so it will use the correct mount options for xfstests. The script used a simple umount-and-mount sequence, which caused a test failure on an XFS filesystem that used both realtime and external log devices.

Thanks, this was my fault :(

Want to consider dave's other suggestions in another patch?

Quoted 8 lines
Quoted 4 lines

There are a handful of other tests that also unmount the TEST_DEV. Perhaps adding a _test_umount() wrapper to common/rc (similar to _scratch_umount) would be best. At least shared/243 needs the same _test_mount treatment as this test.

Eternal git-log fame would be yours. ;)

-Eric

Will do. Let me get my feet back under me. The initial grep -l _test_mount [0-9][0-9][0-9] returned 014, 097, 192; so the thread was started in the spirit of "there's no way this can possibly be the correct function to use."

I'll look at shared/243.

One day, I would like to earn git-log fame by incorporating better support for JFS and especially NILFS2 into xfstests. However, that's a topic for another day. I know that I have uses for both filesystems, but that doesn't mean anybody else does. Their mailing lists don't give much hint of user community or progress.

Thanks!

Michael

Quoted 15 lines
Quoted 15 lines
---
 tests/generic/258 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/generic/258 +1 -1
index fb091ae..3aeb5ce 100755
@@ -62,7 +62,7 @@ fi
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
Quoted 4 lines

tests/generic/258 fi · line 68

Quoted 3 lines
tests/generic/258 +0 -0
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#5 lore raw
Quoted 37 lines · Michael L. Semon
Quoted 22 lines · Eric Sandeen

On 5/21/13 10:03 PM, Michael L. Semon wrote:

...

Quoted 11 lines

From 7dc0667e1f8cd1c98f15ebf412dd0f038b55306e Mon Sep 17 00:00:00 2001
From: "Michael L. Semon" <[email protected]>
Date: Tue, 21 May 2013 22:50:59 -0400
Subject: [PATCH] Change mount method for generic/258

Use the built-in _test_mount function from xfstests so it will use the correct mount options for xfstests. The script used a simple umount-and-mount sequence, which caused a test failure on an XFS filesystem that used both realtime and external log devices.

Thanks, this was my fault :(

Want to consider dave's other suggestions in another patch?

Quoted 8 lines
Quoted 4 lines

There are a handful of other tests that also unmount the TEST_DEV. Perhaps adding a _test_umount() wrapper to common/rc (similar to _scratch_umount) would be best. At least shared/243 needs the same _test_mount treatment as this test.

Eternal git-log fame would be yours. ;)

-Eric

Will do. Let me get my feet back under me. The initial grep -l _test_mount [0-9][0-9][0-9] returned 014, 097, 192; so the thread was started in the spirit of "there's no way this can possibly be the correct function to use."

Quoted 7 lines

I'll look at shared/243.

One day, I would like to earn git-log fame by incorporating better support for JFS and especially NILFS2 into xfstests. However, that's a topic for another day. I know that I have uses for both filesystems, but that doesn't mean anybody else does. Their mailing lists don't give much hint of user community or progress.

The generic tests hopefully work; if not, it should be a fairly simple fixup. And you're free to add fs-specific tests :)

(I'm not sure if we fall back by default to mkfs.$FSTYP and fsck.$FSTYP; if so, it might just work)

-Eric

Quoted 24 lines

Thanks!

Michael

Quoted 19 lines
Quoted 18 lines
---
 tests/generic/258 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/generic/258 +1 -1
index fb091ae..3aeb5ce 100755
@@ -62,7 +62,7 @@ fi
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#6 lore raw
Quoted 10 lines · Eric Sandeen
Quoted 8 lines · Michael L. Semon
Quoted 1 line · Eric Sandeen

On 5/21/13 10:03 PM, Michael L. Semon wrote:

One day, I would like to earn git-log fame by incorporating better support for JFS and especially NILFS2 into xfstests. However, that's a topic for another day. I know that I have uses for both filesystems, but that doesn't mean anybody else does. Their mailing lists don't give much hint of user community or progress.

Quoted 7 lines

The generic tests hopefully work; if not, it should be a fairly simple fixup. And you're free to add fs-specific tests :)

(I'm not sure if we fall back by default to mkfs.$FSTYP and fsck.$FSTYP; if so, it might just work)

-Eric

They work fine--JFS currently survives all of generic/* intact--but...

JFS: The syntax for a lot of things is very close to that of ReiserFS, ext3, and ext4 filesystems. However, before just inserting "jfs" into the common/rc case switches, test results need to be audited for JFS. Additionally, I don't know if the current tests use external journals for non-XFS filesystems: My first attempt stepped all over the XFS journal creation script, looked messy, and was probably buggy as well.

NILFS2: fsstress/fsx seems to slowly rip NILFS2 to bits, and I have to come up with a narrower test and report bugs to those guys. common/rc could be revised to better target filesystems with no fsck whatsoever, such as NILFS2 and F2FS. I've been getting by with a simple ln -s /bin/true /sbin/fsck.nilfs2, and really, that is the best long-term solution from an administrative standpoint.

So you see how this might be an easy fix but still take up hours in debugging...

Thanks again!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#7 lore raw
Quoted 19 lines · Michael L. Semon
Quoted 10 lines · Eric Sandeen
Quoted 8 lines · Michael L. Semon
Quoted 1 line · Eric Sandeen

On 5/21/13 10:03 PM, Michael L. Semon wrote:

One day, I would like to earn git-log fame by incorporating better support for JFS and especially NILFS2 into xfstests. However, that's a topic for another day. I know that I have uses for both filesystems, but that doesn't mean anybody else does. Their mailing lists don't give much hint of user community or progress.

Quoted 7 lines

The generic tests hopefully work; if not, it should be a fairly simple fixup. And you're free to add fs-specific tests :)

(I'm not sure if we fall back by default to mkfs.$FSTYP and fsck.$FSTYP; if so, it might just work)

-Eric

Quoted 8 lines

They work fine--JFS currently survives all of generic/* intact--but...

JFS: The syntax for a lot of things is very close to that of ReiserFS, ext3, and ext4 filesystems. However, before just inserting "jfs" into the common/rc case switches, test results need to be audited for JFS. Additionally, I don't know if the current tests use external journals for non-XFS filesystems: My first attempt stepped all over the XFS journal creation script, looked messy, and was probably buggy as well.

xfs doesn't use external journal by default, it needs to have an extra couple env. vars set (I think README covers this)

To test JFS that way it'd need more tweaking I suppose. It could be done, certainly.

Quoted 2 lines

NILFS2: fsstress/fsx seems to slowly rip NILFS2 to bits, and I have to come up with a narrower test and report bugs to those guys.

:) not too surprising. (first time I looked at ecryptfs vs. fsx, fsx won the fight in about 3 ops. Much better now.)

I'd just use fsx directly to narrow down failures to the smallest set of operations.

Quoted 4 lines

common/rc could be revised to better target filesystems with no fsck whatsoever, such as NILFS2 and F2FS. I've been getting by with a simple ln -s /bin/true /sbin/fsck.nilfs2, and really, that is the best long-term solution from an administrative standpoint.

Yeah I'd suggest that maybe nilfs2-utils should just ship something like that.

Quoted 2 lines

So you see how this might be an easy fix but still take up hours in debugging...

Yup.

-Eric

Quoted 4 lines

Thanks again!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#8 lore raw
In reply to Michael L. Semon
Quoted 8 lines · Michael L. Semon

This is my first attempt at doing a git format-patch --stdout. Hopefully, this is somewhat more presentable to git. Comments and brow-beating are welcome.

Thanks!

Michael

Thanks for the patch Michael, it has been committed:

commit f4ba26dcdc64a206d73d47cf054ecb2742867c55

Author: Michael L. Semon <[email protected]>
Date:   Wed May 22 13:50:59 2013 -0500

xfstests: Change mount method for generic/258

The only other suggestion would be to make your email subject match the patch name and starting with which part of xfs it is submitted for (xfs:, xfstests:, xfsprogs:, xfsdump:, xfsdmapi:. etc) i.e.

Subject: [PATCH] xfstests: Change mount method for generic/258

--Rich

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#9 lore raw
In reply to Dave Chinner
Quoted 29 lines · Dave Chinner
Quoted 18 lines · Michael L. Semon

Hi! When using xfstests generic/258 with along with $TEST_RTDEV $TEST_LOGDEV, it tends to scream bloody murder about corrupted partitions and such. In fact, the commands in the test seem to do the right thing when executed by hand. So once again, I grasped for straws and came up with this:

tests/generic/258 renamed from tests/generic/258.orig +1 -1
@@ -62,7 +62,7 @@
 # unmount, remount, and check the timestamp
 echo "Remounting to flush cache"
 umount $TEST_DEV
-mount $TEST_DEV $TEST_DIR
+_test_mount
 
 # Should yield -315593940 (prior to epoch)
 echo "Testing for negative seconds since epoch"
Quoted 3 lines

My questions are these:

1) Was there a better way to do this?

No, your change is correct. Can you clean up the description of the problem you had and add a Signed-off-by?

Quoted 2 lines

2) Not knowing the policy on umounting $TEST_DEV, could this have been a test for $SCRATCH_DEV?

Quoted 9 lines

There are a handful of other tests that also unmount the TEST_DEV. Perhaps adding a _test_umount() wrapper to common/rc (similar to _scratch_umount) would be best. At least shared/243 needs the same _test_mount treatment as this test.

Cheers,

Dave.

I have a simple _test_mount change in shared/243, tested and ready to post. However, it may work the same either way. AFAIK, the test seems to go like this:

xfs_io this;
if ( ext4 )
{
umount;
do some unmounted ext4 stuff;
mount -t ${FSTYP} ...;
}
xfs_io that;
if ( ext4 )
{
...
}
...

So if it isn't ext4 being tested, $TEST_DEV is not umounted at all. Until I got a closer look at this, it was a surprise to see the XFS+rtdev+logdev pass without changes.

The rest of your comments will be followed when I get home. To use the external logdev means a 20%-30% improvement in some test results on XFS. To get the other file systems to do this will certainly have me looking at the $TEST_DEV mount and umount code anyway.

Thanks!

Michael

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#10 lore raw

PATCH 1/1 xfstests: Change mount method for shared/243

In reply to Dave Chinner

Hi! This patch updates the mount method in xfstests shared/243 without seeming to change the behavior of the test. Should this patch get a NACK for not being useful, no worries! There will be other opportunities.

The patch was tested with ext4, with and without external journals. It was also tested on XFS with an external logdev. The MOUNT_OPTIONS for the ext4 tests included '-o barrier=1' and '-o barrier=0'. The test results varied on ext4 with an external journal, but the results shown by /proc/mounts was OK. [barrier=1 causes a failed test, barrier=0 succeeds, both in the original and patched test.]

Quoted 1 line

From 8762f621a5fb9556fa96ec289bceffc4fcc844f3 Mon Sep 17 00:00:00 2001

From: "Michael L. Semon" <[email protected]>
Date: Wed, 22 May 2013 20:57:56 -0400
Subject: [PATCH] xfstests: Change mount method for shared/243

Change the mount invocations in shared/243 to use the built-in _test_mount function of xfstests. This was done as part of a clean-up recommendation. Currently, only ext4 file systems are affected by this change.

---
 tests/shared/243 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/shared/243 +1 -1
index e0f2db2..b3eb025 100755
@@ -108,7 +108,7 @@ _check_ext4_eof_flag()
     fi
 
     # Mount the ${TEST_DEV}
-    mount ${TEST_DEV} -t ${FSTYP} ${TEST_DIR}
+    _test_mount
   fi
 }
 
Signature · 7 lines
1.8.2


_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs