Skip to contentSkip to conversation

Lost JBD fix

PATCH Jan Kara 4 participants 5 messages Jul 9, 2007 – Jul 13, 2007

lore
Share & tools
lore

Jan Kara

author
#0 lore raw

Hi Andrew,

it seems we've accidentally lost one JBD fix (probably it was my mistake
when rediffing some checkpointing changes) as Kirill has noted. A transaction can currently be released when there are still some buffers on one of its checkpointing lists. Attached patch should fix it (it still seems to apply fine). Please apply it. Thanks.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SuSE CR Labs

We have to check that also the second checkpoint list is non-empty before dropping the transaction.

fs/jbd/commit.c +2 -1
diff -rupX /home/jack/.kerndiffexclude linux-2.6.16-rc1/fs/jbd/commit.c linux-2.6.16-rc1-1-checkpoint-fix/fs/jbd/commit.c
@@ -829,7 +829,8 @@ restart_loop:
 	journal->j_committing_transaction = NULL;
 	spin_unlock(&journal->j_state_lock);
 
-	if (commit_transaction->t_checkpoint_list == NULL) {
+	if (commit_transaction->t_checkpoint_list == NULL &&
+	    commit_transaction->t_checkpoint_io_list == NULL) {
 		__journal_drop_transaction(journal, commit_transaction);
 	} else {
 		if (journal->j_checkpoint_transactions == NULL) {
#1 lore raw
Quoted 22 lines · Jan Kara

Hi Andrew,

it seems we've accidentally lost one JBD fix (probably it was my mistake
when rediffing some checkpointing changes) as Kirill has noted. A transaction can currently be released when there are still some buffers on one of its checkpointing lists. Attached patch should fix it (it still seems to apply fine). Please apply it. Thanks.

Honza

------------------------------------------------------------------------

We have to check that also the second checkpoint list is non-empty before dropping the transaction.

diff -rupX /home/jack/.kerndiffexclude linux-2.6.16-rc1/fs/jbd/commit.c linux-2.6.16-rc1-1-checkpoint-fix/fs/jbd/commit.c

fs/jbd/commit.c restart_loop: · line 836

Quoted 9 lines
fs/jbd/commit.c +2 -1
@@ -829,7 +829,8 @@ restart_loop:
 	journal->j_committing_transaction = NULL;
 	spin_unlock(&journal->j_state_lock);
 
-	if (commit_transaction->t_checkpoint_list == NULL) {
+	if (commit_transaction->t_checkpoint_list == NULL &&
+	    commit_transaction->t_checkpoint_io_list == NULL) {
 		__journal_drop_transaction(journal, commit_transaction);
 	} else {
 		if (journal->j_checkpoint_transactions == NULL) {

Critical bugfix -- 2.6.22-stable (no earlier versions) should get this too, right?

If so, can we get this merged into 2.6.23-rc soon?

#2 lore raw
Quoted 26 lines · Chuck Ebbert
Quoted 25 lines · Jan Kara

Hi Andrew,

it seems we've accidentally lost one JBD fix (probably it was my mistake
when rediffing some checkpointing changes) as Kirill has noted. A transaction can currently be released when there are still some buffers on one of its checkpointing lists. Attached patch should fix it (it still seems to apply fine). Please apply it. Thanks.

Honza

------------------------------------------------------------------------

We have to check that also the second checkpoint list is non-empty before dropping the transaction.

diff -rupX /home/jack/.kerndiffexclude linux-2.6.16-rc1/fs/jbd/commit.c linux-2.6.16-rc1-1-checkpoint-fix/fs/jbd/commit.c

fs/jbd/commit.c +0 -0
@@ -829,7 +829,8 @@ restart_loop:
 	journal->j_committing_transaction = NULL;
 	spin_unlock(&journal->j_state_lock);
 
Quoted 9 lines

fs/jbd/commit.c restart_loop: · line 836

Quoted 6 lines
fs/jbd/commit.c +2 -1
-	if (commit_transaction->t_checkpoint_list == NULL) {
+	if (commit_transaction->t_checkpoint_list == NULL &&
+	    commit_transaction->t_checkpoint_io_list == NULL) {
 		__journal_drop_transaction(journal, commit_transaction);
 	} else {
 		if (journal->j_checkpoint_transactions == NULL) {

Critical bugfix -- 2.6.22-stable (no earlier versions) should get this too, right?

And don't forget that we have a jbd2/ clone as well, for the hugely succesful ext4 project as well.

Signature · 1 line
Jens Axboe
#3 lore raw
Quoted 10 lines · On Thu, 12 Jul 2007 08:50:11 +0200 Jens Axboe <[email protected]> wrote:
Quoted 9 lines

fs/jbd/commit.c restart_loop: · line 836

Quoted 6 lines
fs/jbd/commit.c +2 -1
-	if (commit_transaction->t_checkpoint_list == NULL) {
+	if (commit_transaction->t_checkpoint_list == NULL &&
+	    commit_transaction->t_checkpoint_io_list == NULL) {
 		__journal_drop_transaction(journal, commit_transaction);
 	} else {
 		if (journal->j_checkpoint_transactions == NULL) {

Critical bugfix -- 2.6.22-stable (no earlier versions) should get this too, right?

Quoted 2 lines

And don't forget that we have a jbd2/ clone as well, for the hugely succesful ext4 project as well.

Yeah, I did that. Again.

author
#4 lore raw
In reply to Chuck Ebbert
Quoted 26 lines · Chuck Ebbert
Quoted 25 lines · Jan Kara

Hi Andrew,

it seems we've accidentally lost one JBD fix (probably it was my mistake
when rediffing some checkpointing changes) as Kirill has noted. A transaction can currently be released when there are still some buffers on one of its checkpointing lists. Attached patch should fix it (it still seems to apply fine). Please apply it. Thanks.

Honza

------------------------------------------------------------------------

We have to check that also the second checkpoint list is non-empty before dropping the transaction.

diff -rupX /home/jack/.kerndiffexclude linux-2.6.16-rc1/fs/jbd/commit.c linux-2.6.16-rc1-1-checkpoint-fix/fs/jbd/commit.c

fs/jbd/commit.c +0 -0
@@ -829,7 +829,8 @@ restart_loop:
 	journal->j_committing_transaction = NULL;
 	spin_unlock(&journal->j_state_lock);
 
Quoted 9 lines

fs/jbd/commit.c restart_loop: · line 836

Quoted 6 lines
fs/jbd/commit.c +2 -1
-	if (commit_transaction->t_checkpoint_list == NULL) {
+	if (commit_transaction->t_checkpoint_list == NULL &&
+	    commit_transaction->t_checkpoint_io_list == NULL) {
 		__journal_drop_transaction(journal, commit_transaction);
 	} else {
 		if (journal->j_checkpoint_transactions == NULL) {

Critical bugfix -- 2.6.22-stable (no earlier versions) should get this too, right?

Yes, the bugfix is probably obvious enough to go into -stable.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SuSE CR Labs