jbd2: don't write superblock when unmounting an ro filesystem

PATCH In review Eric Sandeen 5 participants 8 messages Jul 24, 2012 – Aug 19, 2012 Journalling layer for block devices (JBD2)

lore

Share & tools

lore

Eric Sandeen

author

2012-07-24 22:05 UTC #0 lore raw

PATCH 1/1 jbd2: don't write superblock when unmounting an ro filesystem

This sequence:

# truncate --size=65536 fsfile
# losetup --offset 65536 /dev/loop0 fsfile 
# mkfs.ext4 /dev/loop0 
# losetup -d /dev/loop0
# mount -o loop,ro,offset=65536 fsfile  mnt/
# umount mnt
# dmesg | tail

results in an IO error when unmounting the RO filesystem:

[  312.386074] SELinux: initialized (dev loop1, type ext4), uses xattr
[  318.020828] Buffer I/O error on device loop1, logical block 196608
[  318.027024] lost page write due to I/O error on loop1
[  318.032088] JBD2: Error -5 detected when updating journal superblock for loop1-8.

This behavior changed with:

commit 24bcc89c7e7c64982e6192b4952a0a92379fc341

Author: Jan Kara <[email protected]>
Date:   Tue Mar 13 15:41:04 2012 -0400

jbd2: split updating of journal superblock and marking journal empty

which lost some of the magic in jbd2_journal_update_superblock() which used to test for a journal with no outstanding transactions.

I'm not sure if the following is quite the right approach, but it fixes it for me.

Signed-off-by Eric Sandeen <[email protected]>

---

p.s. no idea why this only happens if I use a loop device with an offset!

fs/jbd2/journal.c +5 -0

index e9a3c4c..987ec76 100644
@@ -1354,6 +1354,11 @@ static void jbd2_mark_journal_empty(journal_t *journal)
 
 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
 	read_lock(&journal->j_state_lock);
+	/* Is it already empty? */
+	if (sb->s_start == 0) {
+		read_unlock(&journal->j_state_lock);
+		return;
+	}
 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
 		  journal->j_tail_sequence);

Eric Sandeen

author

2012-07-24 22:18 UTC #1 lore raw

Quoted 28 lines · Eric Sandeen

This sequence:
# truncate --size=65536 fsfile
# losetup --offset 65536 /dev/loop0 fsfile 
# mkfs.ext4 /dev/loop0 
# losetup -d /dev/loop0
# mount -o loop,ro,offset=65536 fsfile  mnt/
# umount mnt
# dmesg | tail
results in an IO error when unmounting the RO filesystem:
[  312.386074] SELinux: initialized (dev loop1, type ext4), uses xattr
[  318.020828] Buffer I/O error on device loop1, logical block 196608
[  318.027024] lost page write due to I/O error on loop1
[  318.032088] JBD2: Error -5 detected when updating journal superblock for loop1-8.
This behavior changed with:
commit 24bcc89c7e7c64982e6192b4952a0a92379fc341
Author: Jan Kara <[email protected]>
Date:   Tue Mar 13 15:41:04 2012 -0400
jbd2: split updating of journal superblock and marking journal empty
which lost some of the magic in jbd2_journal_update_superblock() which used to test for a journal with no outstanding transactions.

Quoted 7 lines

I'm not sure if the following is quite the right approach, but it fixes it for me.
Signed-off-by Eric Sandeen <[email protected]>
---
p.s. no idea why this only happens if I use a loop device with an offset!

PEBKAC. no-offset ro loop mount does the same.

-Eric

I am having hard time understanding what the sequence should be doing. I actually can not reproduce it, because I can not create the file system on such loop device because it seems like the size of /dev/loop0 is zero, which makes sense considering that you set offset to 65536 in the 65536 sized file.

Also, is the use of lo device really needed ? We can create the file system on the file itself right ? Moreover, when I do that I can see that the journal is not created because the device (file) is too small, maybe it have something to do with the problem ?

Anyway, maybe I am missing something, but I do not understand what kind of problem are you reproducing there.

-Lukas

Quoted 49 lines

results in an IO error when unmounting the RO filesystem:
[  312.386074] SELinux: initialized (dev loop1, type ext4), uses xattr
[  318.020828] Buffer I/O error on device loop1, logical block 196608
[  318.027024] lost page write due to I/O error on loop1
[  318.032088] JBD2: Error -5 detected when updating journal superblock for loop1-8.
This behavior changed with:
commit 24bcc89c7e7c64982e6192b4952a0a92379fc341
Author: Jan Kara <[email protected]>
Date:   Tue Mar 13 15:41:04 2012 -0400
jbd2: split updating of journal superblock and marking journal empty
which lost some of the magic in jbd2_journal_update_superblock() which used to test for a journal with no outstanding transactions.
I'm not sure if the following is quite the right approach, but it fixes it for me.
Signed-off-by Eric Sandeen <[email protected]>
---
p.s. no idea why this only happens if I use a loop device with an offset!
fs/jbd2/journal.c +5 -1
index e9a3c4c..987ec76 100644
@@ -1354,6 +1354,11 @@ static void jbd2_mark_journal_empty(journal_t *journal)
 
 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
 	read_lock(&journal->j_state_lock);
+	/* Is it already empty? */
+	if (sb->s_start == 0) {
+		read_unlock(&journal->j_state_lock);
+		return;
+	}
 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
 		  journal->j_tail_sequence);
 
 
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Eric Sandeen

author

2012-07-30 15:49 UTC #3 lore raw

Quoted 18 lines · Lukáš Czerner

Quoted 8 lines · Eric Sandeen
Date: Tue, 24 Jul 2012 17:05:28 -0500
From: Eric Sandeen <[email protected]>
To: ext4 development <[email protected]>
Cc: Jan Kara <[email protected]>
Subject: [PATCH] jbd2: don't write superblock when unmounting an ro filesystem
This sequence:
Quoted 7 lines
# truncate --size=65536 fsfile
# losetup --offset 65536 /dev/loop0 fsfile 
# mkfs.ext4 /dev/loop0 
# losetup -d /dev/loop0
# mount -o loop,ro,offset=65536 fsfile  mnt/
# umount mnt
# dmesg | tail

Quoted 7 lines

Hi Eric,
I am having hard time understanding what the sequence should be doing. I actually can not reproduce it, because I can not create the file system on such loop device because it seems like the size of /dev/loop0 is zero, which makes sense considering that you set offset to 65536 in the 65536 sized file.

sorry, that must have been a cut & paste error. :(

The offset was leftover from trying to match the original reporter's case. This simpler test shows the same problem:

 1014  truncate --size=2g fsfile
 1015  mkfs.ext4 fsfile
 1016  mount -o loop,ro fsfile mnt
 1017  umount mnt
 1018  dmesg | tail

[  417.631105] EXT4-fs (loop0): mounted filesystem with ordered data mode. Opts: (null)
[  417.639183] SELinux: initialized (dev loop0, type ext4), uses xattr
[  420.903970] Buffer I/O error on device loop0, logical block 262144
[  420.910448] lost page write due to I/O error on loop0
[  420.915662] JBD2: Error -5 detected when updating journal superblock for loop0-8.

Quoted 54 lines

Also, is the use of lo device really needed ? We can create the file system on the file itself right ? Moreover, when I do that I can see that the journal is not created because the device (file) is too small, maybe it have something to do with the problem ?
Anyway, maybe I am missing something, but I do not understand what kind of problem are you reproducing there.
-Lukas
Quoted 44 lines
results in an IO error when unmounting the RO filesystem:
[  312.386074] SELinux: initialized (dev loop1, type ext4), uses xattr
[  318.020828] Buffer I/O error on device loop1, logical block 196608
[  318.027024] lost page write due to I/O error on loop1
[  318.032088] JBD2: Error -5 detected when updating journal superblock for loop1-8.
This behavior changed with:
commit 24bcc89c7e7c64982e6192b4952a0a92379fc341
Author: Jan Kara <[email protected]>
Date:   Tue Mar 13 15:41:04 2012 -0400
jbd2: split updating of journal superblock and marking journal empty
which lost some of the magic in jbd2_journal_update_superblock() which used to test for a journal with no outstanding transactions.
I'm not sure if the following is quite the right approach, but it fixes it for me.
Signed-off-by Eric Sandeen <[email protected]>
---
p.s. no idea why this only happens if I use a loop device with an offset!
fs/jbd2/journal.c +5 -0
index e9a3c4c..987ec76 100644
@@ -1354,6 +1354,11 @@ static void jbd2_mark_journal_empty(journal_t *journal)
 
 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
 	read_lock(&journal->j_state_lock);
+	/* Is it already empty? */
+	if (sb->s_start == 0) {
+		read_unlock(&journal->j_state_lock);
+		return;
+	}
 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
 		  journal->j_tail_sequence);
 

Quoted 5 lines

Quoted 5 lines
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Jan Kara

2012-08-15 12:14 UTC #4 lore raw

In reply to Eric Sandeen

Quoted 10 lines · Eric Sandeen

This sequence:

# truncate --size=65536 fsfile
# losetup --offset 65536 /dev/loop0 fsfile 
# mkfs.ext4 /dev/loop0 
# losetup -d /dev/loop0
# mount -o loop,ro,offset=65536 fsfile  mnt/
# umount mnt
# dmesg | tail

Quoted 6 lines

results in an IO error when unmounting the RO filesystem:

[  312.386074] SELinux: initialized (dev loop1, type ext4), uses xattr
[  318.020828] Buffer I/O error on device loop1, logical block 196608
[  318.027024] lost page write due to I/O error on loop1
[  318.032088] JBD2: Error -5 detected when updating journal superblock for loop1-8.

Ted, did this patch fall through cracks? I've ported the fix to JBD and
added it to my tree. I plan to send it to Linus in a few days. Also I've CC'd stable since this is a bit annoying regression.

Honza

Quoted 36 lines

This behavior changed with:
commit 24bcc89c7e7c64982e6192b4952a0a92379fc341
Author: Jan Kara <[email protected]>
Date:   Tue Mar 13 15:41:04 2012 -0400
jbd2: split updating of journal superblock and marking journal empty
which lost some of the magic in jbd2_journal_update_superblock() which used to test for a journal with no outstanding transactions.
I'm not sure if the following is quite the right approach, but it fixes it for me.
Signed-off-by Eric Sandeen <[email protected]>
---
p.s. no idea why this only happens if I use a loop device with an offset!
fs/jbd2/journal.c +5 -0
index e9a3c4c..987ec76 100644
@@ -1354,6 +1354,11 @@ static void jbd2_mark_journal_empty(journal_t *journal)
 
 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
 	read_lock(&journal->j_state_lock);
+	/* Is it already empty? */
+	if (sb->s_start == 0) {
+		read_unlock(&journal->j_state_lock);
+		return;
+	}
 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
 		  journal->j_tail_sequence);
 

Signature · 2 lines

Jan Kara <[email protected]>
SUSE Labs, CR

Eric Sandeen

2012-08-15 17:25 UTC #5 lore raw

Quoted 17 lines · Jan Kara

Quoted 10 lines · Eric Sandeen

This sequence:

# truncate --size=65536 fsfile
# losetup --offset 65536 /dev/loop0 fsfile
# mkfs.ext4 /dev/loop0
# losetup -d /dev/loop0
# mount -o loop,ro,offset=65536 fsfile  mnt/
# umount mnt
# dmesg | tail

Quoted 6 lines

results in an IO error when unmounting the RO filesystem:

[  312.386074] SELinux: initialized (dev loop1, type ext4), uses xattr
[  318.020828] Buffer I/O error on device loop1, logical block 196608
[  318.027024] lost page write due to I/O error on loop1
[  318.032088] JBD2: Error -5 detected when updating journal superblock for loop1-8.

Quoted 5 lines

Ted, did this patch fall through cracks? I've ported the fix to JBD and
added it to my tree. I plan to send it to Linus in a few days. Also I've CC'd stable since this is a bit annoying regression.
Honza

Thanks for doing both of those things, Jan.

BTW this happens on loop because mount -o,ro apparently sets up the loop device itself as readonly.

-Eric

Theodore Ts'o

maintainer

2012-08-18 18:45 UTC #6 lore raw

In reply to Eric Sandeen

Applied, thanks. I updated the reproduction instructions so that they wored correctly. (i.e., by removing the offset, and changing the size to 1g).

- Ted

Eric Sandeen

2012-08-19 18:54 UTC #7 lore raw

Quoted 4 lines · Theodore Ts'o

Applied, thanks. I updated the reproduction instructions so that they wored correctly. (i.e., by removing the offset, and changing the size to 1g).

Thanks, sorry about that really messed up commit message. :(

-Eric

Quoted 5 lines

- Ted
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html