What is the logic behind 'Device or resource busy' ?

What is the logic behind 'Device or resource busy' ?

[Jure Pecar]

[-- Attachment #1: Type: text/plain, Size: 447 bytes --]


Hi all,


Consider the situation where a drive sdx gets kicked out of a mirror md9. If
i want to add it back with mdadm /dev/md9 -a /dev/sdx1, i get the 'device
busy' error.

However, if i stop the mirror and start it right after, i can easily add the
sdx1 back into md9. But this also requires me to stop any services using
this mirror and unmount the filesystem on it...


Any particular reason that this must work that way?



-- 

Jure Pecar

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: What is the logic behind 'Device or resource busy' ?

[Joe Pruett]

i don't know the mdadm equivalent, but i do raidhotremove followed by 
raidhotadd to get around that issue.  if you look at /proc/mdstat after 
the failure, you'll probably see the drive is still part of the array, but 
just marked as down.  raidhotremove tells the system it is gone and then 
you can add it back.

How do we handle the same but uptodated data to disk cache?

[bmoon]

Hello,

I want to write the data to the fixed sector with the same size frequently
just like MD super block, but I do not want to sync and flush the dev
on each write.

For example, to write md super block in /drivers/md/md.c

we do

        fsync_dev(dev);
        bh=getblk(dev, block, size);

        mark_buffer_uptodate(bh,1);
        mark_buffer_dirty(bh);
        ll_rw_block(WRITE,1, &bh);
        wait_on_buffer(bh);
        brelse(bh);
        fsync_dev(dev);

Instead, Can I do as follow

        bh=getblk(dev, block, size);
        if (buffer_uptodate(bh)) { /* still in cache, not flushed yet */
           bh->b_data = new SB data;
            brelse(bh);
        }
        else { /* it must be flushed */
            same as in above
       }

Please provide me your openion or suggestions.

Thanks in advance,

Bo

Re: How do we handle the same but uptodated data to disk cache?

[Neil Brown]

On Monday May 26, bo@anthologysolutions.com wrote:
> Hello,
> 
> I want to write the data to the fixed sector with the same size frequently
> just like MD super block, but I do not want to sync and flush the dev
> on each write.
> 
> For example, to write md super block in /drivers/md/md.c

I suggest you look at a newer md.c.  With 2.4.21pre or 2.5.* have
"sync_page_io" which does what you want more effectively.

NeilBrown

Re: What is the logic behind 'Device or resource busy' ?

[Jure Pecar]

[-- Attachment #1: Type: text/plain, Size: 511 bytes --]

On Mon, 26 May 2003 17:39:45 -0700 (PDT)
Joe Pruett <joey@q7.com> wrote:

> i don't know the mdadm equivalent, but i do raidhotremove followed by 
> raidhotadd to get around that issue.  if you look at /proc/mdstat after 
> the failure, you'll probably see the drive is still part of the array, but
> 
> just marked as down.  raidhotremove tells the system it is gone and then 
> you can add it back.


hm yes... a failed drive is still present and has to be removed. makes sense
now, thanks.


-- 

Jure Pecar

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: How do we handle the same but uptodated data to disk cache?

[bmoon]

Neil,

It improved a lot for the sync I/O.
However, I am looking for a solution to handle
the regular async I/O with a page data on the
same sector repeatedly like writing the MD SB
data to disk without flushing.

The follwing is my simple algorithm for it;

        bh=getblk(dev, block, size);
        if (buffer_uptodate(bh)) { /* still in cache, not flushed yet */
              bh->b_data = "updated data";
              brelse(bh);
        }
        else { /* all were flushed or new  */
            .
            .
             mark_buffer_uptodate(bh,1);
             mark_buffer_dirty(bh);
            generic_make_request(rw, &bh);
       }

 Please give me your suggestion or comment.


Bo
----- Original Message -----
From: "Neil Brown" <neilb@cse.unsw.edu.au>
To: "bmoon" <bo@anthologysolutions.com>
Cc: <linux-raid@vger.kernel.org>
Sent: Monday, May 26, 2003 7:47 PM
Subject: Re: How do we handle the same but uptodated data to disk cache?


> On Monday May 26, bo@anthologysolutions.com wrote:
> > Hello,
> >
> > I want to write the data to the fixed sector with the same size
frequently
> > just like MD super block, but I do not want to sync and flush the dev
> > on each write.
> >
> > For example, to write md super block in /drivers/md/md.c
>
> I suggest you look at a newer md.c.  With 2.4.21pre or 2.5.* have
> "sync_page_io" which does what you want more effectively.
>
> NeilBrown
>

Re: How do we handle the same but uptodated data to disk cache?

[bmoon]

Help!
There must be a function which will update the data in the disk cache.
From __make_request(), it does ordering(neighboring) the requests,
it does not deal with the same block request.

Please give me a light!

> 
> It improved a lot for the sync I/O.
> However, I am looking for a solution to handle
> the regular async I/O with a page data on the
> same sector repeatedly like writing the MD SB
> data to disk without flushing.
> 
> The follwing is my simple algorithm for it;
> 
>         bh=getblk(dev, block, size);
>         if (buffer_uptodate(bh)) { /* still in cache, not flushed yet */
>               bh->b_data = "updated data";
>               brelse(bh);
>         }
>         else { /* all were flushed or new  */
>             .
>             .
>              mark_buffer_uptodate(bh,1);
>              mark_buffer_dirty(bh);
>             generic_make_request(rw, &bh);
>        }
> 
>  Please give me your suggestion or comment.
> 

Bo

Re: How do we handle the same but uptodated data to disk cache?

[Neil Brown]

On Tuesday May 27, bo@anthologysolutions.com wrote:
> Neil,
> 
> It improved a lot for the sync I/O.
> However, I am looking for a solution to handle
> the regular async I/O with a page data on the
> same sector repeatedly like writing the MD SB
> data to disk without flushing.

This is a filesystem question.  Not a raid question.

I would just:
   getblk
   copy new data in
   mark update
   mark dirty
   release block

and it will eventually be flushed to disk.
If you want to written sooner, call
   submit_bh


NeilBrown

> 
> The follwing is my simple algorithm for it;
> 
>         bh=getblk(dev, block, size);
>         if (buffer_uptodate(bh)) { /* still in cache, not flushed yet */
>               bh->b_data = "updated data";
>               brelse(bh);
>         }
>         else { /* all were flushed or new  */
>             .
>             .
>              mark_buffer_uptodate(bh,1);
>              mark_buffer_dirty(bh);
>             generic_make_request(rw, &bh);
>        }
> 
>  Please give me your suggestion or comment.
> 
> 
> Bo
> ----- Original Message -----
> From: "Neil Brown" <neilb@cse.unsw.edu.au>
> To: "bmoon" <bo@anthologysolutions.com>
> Cc: <linux-raid@vger.kernel.org>
> Sent: Monday, May 26, 2003 7:47 PM
> Subject: Re: How do we handle the same but uptodated data to disk cache?
> 
> 
> > On Monday May 26, bo@anthologysolutions.com wrote:
> > > Hello,
> > >
> > > I want to write the data to the fixed sector with the same size
> frequently
> > > just like MD super block, but I do not want to sync and flush the dev
> > > on each write.
> > >
> > > For example, to write md super block in /drivers/md/md.c
> >
> > I suggest you look at a newer md.c.  With 2.4.21pre or 2.5.* have
> > "sync_page_io" which does what you want more effectively.
> >
> > NeilBrown
> >
> 
> -
> To unsubscribe from this list: send the line "unsubscribe linux-raid" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html