new data logging patches available

new data logging patches available

[Chris Mason]

Hello all,

New data logging patches are available for download at:

ftp.suse.com/pub/people/mason/patches/data-logging/2.4.21

It might take the suse internal ftp server an hour to see the new
files.  Check for 03-relocation-8.diff to make sure you are grabbing the
most recent stuff.

These include Oleg Drokin's merge with 2.4.21-rc1 (thanks Oleg!), and
they no longer change fs/super.c:sync_supers().  I believe things are
now clean enough for me to start submitting for 2.4.22-pre once it
starts.

03-relocation-8.diff is also updated, I believe it will fix the bug
where md reports the device is busy during reboot.  It has not been
tested on a box with a software raid root device, so please handle with
care.

-chris

Re: new data logging patches available

[Soeren Sonnenburg]

On Thu, 2003-05-08 at 15:22, Chris Mason wrote:

> 03-relocation-8.diff is also updated, I believe it will fix the bug
> where md reports the device is busy during reboot.  It has not been
> tested on a box with a software raid root device, so please handle with
> care.

errhm... I would pretty much like to have data logging on my root
sw-raid... unfortunately I cannot risk a dataloss atm...
so could you please tell what kind of bugs you expect to happen such
that I know the risks a bit better ?

Soeren.

Re: new data logging patches available

[Chris Mason]

On Thu, 2003-05-08 at 09:32, Soeren Sonnenburg wrote:
> On Thu, 2003-05-08 at 15:22, Chris Mason wrote:
> 
> > 03-relocation-8.diff is also updated, I believe it will fix the bug
> > where md reports the device is busy during reboot.  It has not been
> > tested on a box with a software raid root device, so please handle with
> > care.
> 
> errhm... I would pretty much like to have data logging on my root
> sw-raid... unfortunately I cannot risk a dataloss atm...
> so could you please tell what kind of bugs you expect to happen such
> that I know the risks a bit better ?

Basically the change is to avoid adding an extra reference count when
the log is not on an external device.  It should be safe, but if I've
screwed something up it will cause oopsen and other problems.

I would suggest watching the list for success and problem reports for a
few days.

-chris

Re: new data logging patches available

[Dieter Nützel]

Am Donnerstag, 8. Mai 2003 15:22 schrieb Chris Mason:
> Hello all,
>
> New data logging patches are available for download at:
>
> ftp.suse.com/pub/people/mason/patches/data-logging/2.4.21
>
> It might take the suse internal ftp server an hour to see the new
> files.  Check for 03-relocation-8.diff to make sure you are grabbing the
> most recent stuff.
>
> These include Oleg Drokin's merge with 2.4.21-rc1 (thanks Oleg!), and
> they no longer change fs/super.c:sync_supers().  I believe things are
> now clean enough for me to start submitting for 2.4.22-pre once it
> starts.
>
> 03-relocation-8.diff is also updated, I believe it will fix the bug
> where md reports the device is busy during reboot.  It has not been
> tested on a box with a software raid root device, so please handle with
> care.

Up and running under 2.4.21-rc1-jam1 (-aa) ;-)

06-write_times.diff
07-quota-v2-2.4.21.diff
Not needed.

05-data-logging-37.diff
need some fixing in fs/reiserfs/inode.c

Ignore the first:
***************
*** 2027,2033 ****

  fail:
      if (nr) {
-         submit_bh_for_writepage(arr, nr) ;
      } else {
          UnlockPage(page) ;
      }
--- 2191,2197 ----

  fail:
      if (nr) {
+         submit_bh_for_writepage(page, arr, nr) ;
      } else {
          UnlockPage(page) ;
      }
***************

But change it to this some lines above:

    if (nr) {
        submit_bh_for_writepage(page, arr, nr) ;
	wakeup_page_waiters(page);
    } else {


Second:
***************
*** 2143,2148 ****
                                struct kiobuf *iobuf, unsigned long blocknr,
                              int blocksize)
  {
      return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize,
                               reiserfs_get_block_direct_io) ;
  }
--- 2530,2539 ----
                                struct kiobuf *iobuf, unsigned long blocknr,
                              int blocksize)
  {
+
+     if (reiserfs_data_ordered(inode->i_sb) || reiserfs_file_data_log(inode)) 
{
+       return -EINVAL;
+     }
      return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize,
                               reiserfs_get_block_direct_io) ;
  }
***************

Shouldn't it be this?:

static int reiserfs_direct_io(int rw, struct file * filp,
                              struct kiobuf *iobuf, unsigned long blocknr,
			      int blocksize) 
{
    struct inode * inode = filp->f_dentry->d_inode->i_mapping->host;

    if (reiserfs_data_ordered(inode->i_sb) || reiserfs_file_data_log(inode)) {
	return -EINVAL;
    }
    return generic_direct_IO(rw, inode, iobuf, blocknr, blocksize,
                             reiserfs_get_block_direct_io) ;
}

Regards,
	Dieter

Re: new data logging patches available

[Christian Mayrhuber]

Am Donnerstag, 8. Mai 2003 15:22 schrieb Chris Mason:

> These include Oleg Drokin's merge with 2.4.21-rc1 (thanks Oleg!), and
> they no longer change fs/super.c:sync_supers().  I believe things are
> now clean enough for me to start submitting for 2.4.22-pre once it
> starts.
This is great news!

>
> 03-relocation-8.diff is also updated, I believe it will fix the bug
> where md reports the device is busy during reboot.  It has not been
> tested on a box with a software raid root device, so please handle with
> care.
Yes, relocation-8 fixes the MD root device issue. /dev/md0 can switch to
read only mode and do a shutdown afterwards.

I can't tell if there are some data issues which it, because I've only
booted this kernel once. For the moment it looks good.

PS: I've applied all patches from 02- to 10-
-- 
lg, Chris

Re: new data logging patches available

[Chris Mason]

On Fri, 2003-05-09 at 02:37, Christian Mayrhuber wrote:

> > 03-relocation-8.diff is also updated, I believe it will fix the bug
> > where md reports the device is busy during reboot.  It has not been
> > tested on a box with a software raid root device, so please handle with
> > care.
> Yes, relocation-8 fixes the MD root device issue. /dev/md0 can switch to
> read only mode and do a shutdown afterwards.

Thanks for testing, and for your help in tracking down the problem.

-chris

Re: new data logging patches available

[Oleg Drokin]

Hello!

On Thu, May 08, 2003 at 09:22:19AM -0400, Chris Mason wrote:

> These include Oleg Drokin's merge with 2.4.21-rc1 (thanks Oleg!), and
> they no longer change fs/super.c:sync_supers().  I believe things are
> now clean enough for me to start submitting for 2.4.22-pre once it
> starts.

BTW, what have happened to your work on porting this stuff to 2.5 and the whole idea
of feeding this stuff into 2.5 first?

Bye,
    Oleg

Re: new data logging patches available

[Chris Mason]

On Mon, 2003-05-12 at 07:33, Oleg Drokin wrote:
> Hello!
> 
> On Thu, May 08, 2003 at 09:22:19AM -0400, Chris Mason wrote:
> 
> > These include Oleg Drokin's merge with 2.4.21-rc1 (thanks Oleg!), and
> > they no longer change fs/super.c:sync_supers().  I believe things are
> > now clean enough for me to start submitting for 2.4.22-pre once it
> > starts.
> 
> BTW, what have happened to your work on porting this stuff to 2.5 and the whole idea
> of feeding this stuff into 2.5 first?

The part I wanted in 2.5 first was the sync supers change.  For now I'll
be happy if I can get the whole thing (without sync supers) in 2.4, so I
can concentrate more on 2.5 code.

-chris

Re: new data logging patches available

[Dieter Nützel]

Am Dienstag, 20. Mai 2003 17:00 schrieb Chris Mason:
> On Mon, 2003-05-12 at 07:33, Oleg Drokin wrote:
> > Hello!
> >
> > On Thu, May 08, 2003 at 09:22:19AM -0400, Chris Mason wrote:
> > > These include Oleg Drokin's merge with 2.4.21-rc1 (thanks Oleg!), and
> > > they no longer change fs/super.c:sync_supers().  I believe things are
> > > now clean enough for me to start submitting for 2.4.22-pre once it
> > > starts.
> >
> > BTW, what have happened to your work on porting this stuff to 2.5 and the
> > whole idea of feeding this stuff into 2.5 first?
>
> The part I wanted in 2.5 first was the sync supers change.  For now I'll
> be happy if I can get the whole thing (without sync supers) in 2.4, so I
> can concentrate more on 2.5 code.

Sync super worked fine for ages, here...;-)

Regards,
	Dieter