2009-04-23 13:16:44

by Alessio Igor Bogani

Subject: [PATCH -tip] remove the BKL: remove "BKL auto-drop" assumption from ext3_remount()

Fix ext3_remount()'s "schedule() drops the BKL automatically" assumption,
when ext3_mark_recovery_complete(), which can sleep, does not do that it can
lock up.

Signed-off-by: Alessio Igor Bogani <[email protected]>
---
fs/ext3/super.c | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/fs/ext3/super.c b/fs/ext3/super.c
index dc905f9..3fa457a 100644
--- a/fs/ext3/super.c
+++ b/fs/ext3/super.c
@@ -2561,9 +2561,11 @@ static int ext3_remount (struct super_block * sb, int * flags, char * data)
* We have to unlock super so that we can wait for
* transactions.
*/
+ unlock_kernel();
unlock_super(sb);
ext3_mark_recovery_complete(sb, es);
lock_super(sb);
+ lock_kernel();
} else {
__le32 ret;
if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
--
1.6.0.4

2009-04-26 16:39:49

by Ingo Molnar

Subject: Re: [PATCH -tip] remove the BKL: remove "BKL auto-drop" assumption from ext3_remount()

* Alessio Igor Bogani <[email protected]> wrote:

> Fix ext3_remount()'s "schedule() drops the BKL automatically" assumption,
> when ext3_mark_recovery_complete(), which can sleep, does not do that it can
> lock up.
>
> Signed-off-by: Alessio Igor Bogani <[email protected]>
> ---
> fs/ext3/super.c | 2 ++
> 1 files changed, 2 insertions(+), 0 deletions(-)
>
> diff --git a/fs/ext3/super.c b/fs/ext3/super.c
> index dc905f9..3fa457a 100644
> --- a/fs/ext3/super.c
> +++ b/fs/ext3/super.c
> @@ -2561,9 +2561,11 @@ static int ext3_remount (struct super_block * sb, int * flags, char * data)
> * We have to unlock super so that we can wait for
> * transactions.
> */
> + unlock_kernel();
> unlock_super(sb);
> ext3_mark_recovery_complete(sb, es);
> lock_super(sb);
> + lock_kernel();

Shouldnt this be a conditional-unlock, so that if the remount moves
away from under the BKL we dont break in this place?

Ingo

2009-04-27 20:21:51

by Alessio Igor Bogani

Subject: Re: [PATCH -tip] remove the BKL: remove "BKL auto-drop" assumption from ext3_remount()

Dear Sir Molnar,

2009/4/26 Ingo Molnar <[email protected]>:
[...]
>> + ? ? ? ? ? ? ? ? ? ? unlock_kernel();
>> ? ? ? ? ? ? ? ? ? ? ? unlock_super(sb);
>> ? ? ? ? ? ? ? ? ? ? ? ext3_mark_recovery_complete(sb, es);
>> ? ? ? ? ? ? ? ? ? ? ? lock_super(sb);
>> + ? ? ? ? ? ? ? ? ? ? lock_kernel();
>
> Shouldnt this be a conditional-unlock, so that if the remount moves
> away from under the BKL we dont break in this place?
[...]

Please excuse me I'm only a stupid howler monkey. :-)

Patch with conditional-unlock/lock follow in a second.

Thank you!

Ciao,
Alessio

2009-04-27 20:22:46

by Alessio Igor Bogani

Subject: [PATCH -tip:kill-the-BKL] remove the BKL: remove "BKL auto-drop" assumption from ext3_remount()

Fix ext3_remount()'s "schedule() drops the BKL automatically" assumption,
when ext3_mark_recovery_complete(), which can sleep, does not do that it can
lock up.

Signed-off-by: Alessio Igor Bogani <[email protected]>
---
fs/ext3/super.c | 5 +++++
1 files changed, 5 insertions(+), 0 deletions(-)

diff --git a/fs/ext3/super.c b/fs/ext3/super.c
index dc905f9..5d61a7f 100644
--- a/fs/ext3/super.c
+++ b/fs/ext3/super.c
@@ -2503,6 +2503,7 @@ static int ext3_remount (struct super_block * sb, int * flags, char * data)
#ifdef CONFIG_QUOTA
int i;
#endif
+ int bkl = kernel_locked();

/* Store the original options */
old_sb_flags = sb->s_flags;
@@ -2561,9 +2562,13 @@ static int ext3_remount (struct super_block * sb, int * flags, char * data)
* We have to unlock super so that we can wait for
* transactions.
*/
+ if (bkl)
+ unlock_kernel();
unlock_super(sb);
ext3_mark_recovery_complete(sb, es);
lock_super(sb);
+ if (bkl)
+ lock_kernel();
} else {
__le32 ret;
if ((ret = EXT3_HAS_RO_COMPAT_FEATURE(sb,
--
1.6.0.4