BUG dcache.c:613 during autofs unmounting in 2.6.19rc2

BUG dcache.c:613 during autofs unmounting in 2.6.19rc2

[Andi Kleen]

While unmounting autofs on shutdown my workstation got a dcache.c:613 BUG 
with 2.6.19rc2.

Only jpegs available unfortunately:

http://one.firstfloor.org/~andi/autofs-oops1.jpg
http://one.firstfloor.org/~andi/autofs-oops2.jpg

I think it was autofs3 instead of autofs4 - at least I got both compiled in.
The autofs user land was autofs-4.1.4 (-6 suse rpm) 

-Andi

Re: BUG dcache.c:613 during autofs unmounting in 2.6.19rc2

[Ian Kent]

On Mon, 2006-10-16 at 16:58 +0200, Andi Kleen wrote:
> While unmounting autofs on shutdown my workstation got a dcache.c:613 BUG 
> with 2.6.19rc2.
> 
> Only jpegs available unfortunately:
> 
> http://one.firstfloor.org/~andi/autofs-oops1.jpg
> http://one.firstfloor.org/~andi/autofs-oops2.jpg
> 
> I think it was autofs3 instead of autofs4 - at least I got both compiled in.
> The autofs user land was autofs-4.1.4 (-6 suse rpm) 

Don't think compiling both in is a good idea.
They both register as "autofs" so you really should choose one and
disable the other.

For my part I have to recommend autofs4 (personally I'd like to see the
autofs v3 module deprecated) and autofs4 is really needed if your using
autofs version 4 or above.

Ian

Re: BUG dcache.c:613 during autofs unmounting in 2.6.19rc2

[Andi Kleen]

On Tuesday 17 October 2006 06:15, Ian Kent wrote:
> On Mon, 2006-10-16 at 16:58 +0200, Andi Kleen wrote:
> > While unmounting autofs on shutdown my workstation got a dcache.c:613 BUG 
> > with 2.6.19rc2.
> > 
> > Only jpegs available unfortunately:
> > 
> > http://one.firstfloor.org/~andi/autofs-oops1.jpg
> > http://one.firstfloor.org/~andi/autofs-oops2.jpg
> > 
> > I think it was autofs3 instead of autofs4 - at least I got both compiled in.
> > The autofs user land was autofs-4.1.4 (-6 suse rpm) 
> 
> Don't think compiling both in is a good idea.
> They both register as "autofs" so you really should choose one and
> disable the other.
> 
> For my part I have to recommend autofs4 (personally I'd like to see the
> autofs v3 module deprecated) and autofs4 is really needed if your using
> autofs version 4 or above.

Well it always worked this way in earlier kernels and even if the
wrong module was suddenly used for some reason it shouldn't BUG.
So something is broken.

-Andi

Re: BUG dcache.c:613 during autofs unmounting in 2.6.19rc2

[Ian Kent]

On Tue, 2006-10-17 at 12:50 +0200, Andi Kleen wrote:
> On Tuesday 17 October 2006 06:15, Ian Kent wrote:
> > On Mon, 2006-10-16 at 16:58 +0200, Andi Kleen wrote:
> > > While unmounting autofs on shutdown my workstation got a dcache.c:613 BUG 
> > > with 2.6.19rc2.
> > > 
> > > Only jpegs available unfortunately:
> > > 
> > > http://one.firstfloor.org/~andi/autofs-oops1.jpg
> > > http://one.firstfloor.org/~andi/autofs-oops2.jpg
> > > 
> > > I think it was autofs3 instead of autofs4 - at least I got both compiled in.
> > > The autofs user land was autofs-4.1.4 (-6 suse rpm) 
> > 
> > Don't think compiling both in is a good idea.
> > They both register as "autofs" so you really should choose one and
> > disable the other.
> > 
> > For my part I have to recommend autofs4 (personally I'd like to see the
> > autofs v3 module deprecated) and autofs4 is really needed if your using
> > autofs version 4 or above.
> 
> Well it always worked this way in earlier kernels and even if the
> wrong module was suddenly used for some reason it shouldn't BUG.
> So something is broken.

True.

There have been some changes in this area (David Howells made some
changes which affected autofs4) and I'm not sure that the autofs module
was reviewed. I didn't look closely at it at the time, I guess I should
have. Sorry.

It will take a while longer to work out if the autofs if open to the
same issue resulting from Davids change.

Ian

Re: BUG dcache.c:613 during autofs unmounting in 2.6.19rc2

[Ian Kent]

On Tue, 2006-10-17 at 21:55 +0800, Ian Kent wrote:
> > Well it always worked this way in earlier kernels and even if the
> > wrong module was suddenly used for some reason it shouldn't BUG.
> > So something is broken.
> 
> True.
> 
> There have been some changes in this area (David Howells made some
> changes which affected autofs4) and I'm not sure that the autofs module
> was reviewed. I didn't look closely at it at the time, I guess I should
> have. Sorry.
> 
> It will take a while longer to work out if the autofs if open to the
> same issue resulting from Davids change.

Oh forgot .. I know it may be hard to do but could you try and confirm
whether it was autofs or autofs4. They are very different and I don't
want to stare at code trying to work out what's wrong if it's not
broken.

> 
> Ian
> 

Re: BUG dcache.c:613 during autofs unmounting in 2.6.19rc2

[David Howells]

Ian Kent <raven@themaw.net> wrote:

> There have been some changes in this area (David Howells made some
> changes which affected autofs4) and I'm not sure that the autofs module
> was reviewed. I didn't look closely at it at the time, I guess I should
> have. Sorry.

It looks like my fixes for autofs4 need applying to autofs also.

David

AUTOFS3: Make sure all dentries refs are released before calling kill_anon_super()

[David Howells]

Make sure all dentries refs are released before calling kill_anon_super() so
that the assumption that generic_shutdown_super() can completely destroy the
dentry tree for there will be no external references holds true.

What was being done in the put_super() superblock op, is now done in the
kill_sb() filesystem op instead, prior to calling kill_anon_super().

The call to shrink_dcache_sb() is removed as it is redundant since
shrink_dcache_for_umount() will now be called after the cleanup routine.

Signed-off-by: David Howells <dhowells@redhat.com>
---

 fs/autofs/autofs_i.h |    1 +
 fs/autofs/dirhash.c  |    1 -
 fs/autofs/init.c     |    2 +-
 fs/autofs/inode.c    |    4 ++--
 4 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h
index c7700d9..906ba5c 100644
--- a/fs/autofs/autofs_i.h
+++ b/fs/autofs/autofs_i.h
@@ -149,6 +149,7 @@ extern const struct file_operations auto
 /* Initializing function */
 
 int autofs_fill_super(struct super_block *, void *, int);
+void autofs_kill_sb(struct super_block *sb);
 
 /* Queue management functions */
 
diff --git a/fs/autofs/dirhash.c b/fs/autofs/dirhash.c
index 3fded38..bf8c8af 100644
--- a/fs/autofs/dirhash.c
+++ b/fs/autofs/dirhash.c
@@ -246,5 +246,4 @@ void autofs_hash_nuke(struct autofs_sb_i
 			kfree(ent);
 		}
 	}
-	shrink_dcache_sb(sbi->sb);
 }
diff --git a/fs/autofs/init.c b/fs/autofs/init.c
index aca1237..cea5219 100644
--- a/fs/autofs/init.c
+++ b/fs/autofs/init.c
@@ -24,7 +24,7 @@ static struct file_system_type autofs_fs
 	.owner		= THIS_MODULE,
 	.name		= "autofs",
 	.get_sb		= autofs_get_sb,
-	.kill_sb	= kill_anon_super,
+	.kill_sb	= autofs_kill_sb,
 };
 
 static int __init init_autofs_fs(void)
diff --git a/fs/autofs/inode.c b/fs/autofs/inode.c
index 2c9759b..54c518c 100644
--- a/fs/autofs/inode.c
+++ b/fs/autofs/inode.c
@@ -20,7 +20,7 @@ #include <linux/magic.h>
 #include "autofs_i.h"
 #include <linux/module.h>
 
-static void autofs_put_super(struct super_block *sb)
+void autofs_kill_sb(struct super_block *sb)
 {
 	struct autofs_sb_info *sbi = autofs_sbi(sb);
 	unsigned int n;
@@ -37,13 +37,13 @@ static void autofs_put_super(struct supe
 	kfree(sb->s_fs_info);
 
 	DPRINTK(("autofs: shutting down\n"));
+	kill_anon_super(sb);
 }
 
 static void autofs_read_inode(struct inode *inode);
 
 static struct super_operations autofs_sops = {
 	.read_inode	= autofs_read_inode,
-	.put_super	= autofs_put_super,
 	.statfs		= simple_statfs,
 };
 

Re: AUTOFS3: Make sure all dentries refs are released before calling kill_anon_super()

[Ian Kent]

On Tue, 2006-10-17 at 17:18 +0100, David Howells wrote:
> Make sure all dentries refs are released before calling kill_anon_super() so
> that the assumption that generic_shutdown_super() can completely destroy the
> dentry tree for there will be no external references holds true.
> 
> What was being done in the put_super() superblock op, is now done in the
> kill_sb() filesystem op instead, prior to calling kill_anon_super().
> 
> The call to shrink_dcache_sb() is removed as it is redundant since
> shrink_dcache_for_umount() will now be called after the cleanup routine.
> 
> Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Ian Kent <raven@themaw.net>
> ---
> 
>  fs/autofs/autofs_i.h |    1 +
>  fs/autofs/dirhash.c  |    1 -
>  fs/autofs/init.c     |    2 +-
>  fs/autofs/inode.c    |    4 ++--
>  4 files changed, 4 insertions(+), 4 deletions(-)
> 
> diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h
> index c7700d9..906ba5c 100644
> --- a/fs/autofs/autofs_i.h
> +++ b/fs/autofs/autofs_i.h
> @@ -149,6 +149,7 @@ extern const struct file_operations auto
>  /* Initializing function */
>  
>  int autofs_fill_super(struct super_block *, void *, int);
> +void autofs_kill_sb(struct super_block *sb);
>  
>  /* Queue management functions */
>  
> diff --git a/fs/autofs/dirhash.c b/fs/autofs/dirhash.c
> index 3fded38..bf8c8af 100644
> --- a/fs/autofs/dirhash.c
> +++ b/fs/autofs/dirhash.c
> @@ -246,5 +246,4 @@ void autofs_hash_nuke(struct autofs_sb_i
>  			kfree(ent);
>  		}
>  	}
> -	shrink_dcache_sb(sbi->sb);
>  }
> diff --git a/fs/autofs/init.c b/fs/autofs/init.c
> index aca1237..cea5219 100644
> --- a/fs/autofs/init.c
> +++ b/fs/autofs/init.c
> @@ -24,7 +24,7 @@ static struct file_system_type autofs_fs
>  	.owner		= THIS_MODULE,
>  	.name		= "autofs",
>  	.get_sb		= autofs_get_sb,
> -	.kill_sb	= kill_anon_super,
> +	.kill_sb	= autofs_kill_sb,
>  };
>  
>  static int __init init_autofs_fs(void)
> diff --git a/fs/autofs/inode.c b/fs/autofs/inode.c
> index 2c9759b..54c518c 100644
> --- a/fs/autofs/inode.c
> +++ b/fs/autofs/inode.c
> @@ -20,7 +20,7 @@ #include <linux/magic.h>
>  #include "autofs_i.h"
>  #include <linux/module.h>
>  
> -static void autofs_put_super(struct super_block *sb)
> +void autofs_kill_sb(struct super_block *sb)
>  {
>  	struct autofs_sb_info *sbi = autofs_sbi(sb);
>  	unsigned int n;
> @@ -37,13 +37,13 @@ static void autofs_put_super(struct supe
>  	kfree(sb->s_fs_info);
>  
>  	DPRINTK(("autofs: shutting down\n"));
> +	kill_anon_super(sb);
>  }
>  
>  static void autofs_read_inode(struct inode *inode);
>  
>  static struct super_operations autofs_sops = {
>  	.read_inode	= autofs_read_inode,
> -	.put_super	= autofs_put_super,
>  	.statfs		= simple_statfs,
>  };
>