xfs buffers

xfs buffers

[kanishk rastogi]

[-- Attachment #1.1: Type: text/plain, Size: 1205 bytes --]

hi all,
I was trying to understand the xfs_buf_t management in XFS.
I was looking at the xfs_bufhash_t which is used to keep the info of buffers
which have ongoing transactions.
I wanted to know if it uses it for caching these pages.

As far as i can see in xfs_buf_relse  calls ::

list_del_init(&bp->b_hash_list);

if the count drops down to 0 for the buffer...

if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) {
                if (bp->b_relse) { ----->set only in case of error??
                        atomic_inc(&bp->b_hold);
                        spin_unlock(&hash->bh_lock);
                        (*(bp->b_relse)) (bp);
                } else if (bp->b_flags & XBF_FS_MANAGED) { ----> mostly
unset......
                        spin_unlock(&hash->bh_lock);
                } else {
                        ASSERT(!(bp->b_flags & (XBF_DELWRI|_XBF_DELWRI_Q)));
                        *list_del_init(&bp->b_hash_list);*
                        spin_unlock(&hash->bh_lock);
                        xfs_buf_free(bp);
                }
        }

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

CMIIW:
So it doesnt keeps buffers if they are not in use....  ??

I dont think that should be the case...

-- 
regards
kanishk

[-- Attachment #1.2: Type: text/html, Size: 1425 bytes --]

[-- Attachment #2: Type: text/plain, Size: 121 bytes --]

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: xfs buffers

[Russell Cattelan]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

kanishk rastogi wrote:
> hi all, I was trying to understand the xfs_buf_t management in XFS.
>  I was looking at the xfs_bufhash_t which is used to keep the info
> of buffers which have ongoing transactions. I wanted to know if it
> uses it for caching these pages.
>
> As far as i can see in xfs_buf_relse  calls ::
>
> list_del_init(&bp->b_hash_list);
>
> if the count drops down to 0 for the buffer...
>
> if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) { if
> (bp->b_relse) { ----->set only in case of error??
> atomic_inc(&bp->b_hold); spin_unlock(&hash->bh_lock);
> (*(bp->b_relse)) (bp); } else if (bp->b_flags & XBF_FS_MANAGED) {
> ----> mostly unset......\
Indeed. that flag is used on exactly one buffer, the superblock.
Basically the superblock is pinned in memory and only flushed by the
filesystem.
> spin_unlock(&hash->bh_lock); } else { ASSERT(!(bp->b_flags &
> (XBF_DELWRI|_XBF_DELWRI_Q))); *list_del_init(&bp->b_hash_list);*
> spin_unlock(&hash->bh_lock); xfs_buf_free(bp); } }
>
> ---------------
>
> CMIIW: So it doesnt keeps buffers if they are not in use....  ??
buffers are released back to the buf_t  zone when they are done.

zone allocation / free is very fast so managing lists of "free"
buffers doesn't make a lot of sense.

>
> I dont think that should be the case...
>
> -- regards kanishk
>
> ----------------------------------------------------------------------
>
>
> _______________________________________________ xfs mailing list
> xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (Darwin)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFKhaWBNRmM+OaGhBgRAtqgAJ9CZfj0QXz8Yy7P4jYemgEPsvR9pQCfcSWM
a9OI8A4ckMdOuiXQ60o4PLE=
=OJAD
-----END PGP SIGNATURE-----

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: xfs buffers

[kanishk rastogi]

[-- Attachment #1.1: Type: text/plain, Size: 2224 bytes --]

inline...

On Fri, Aug 14, 2009 at 1:57 PM, Russell Cattelan <cattelan@thebarn.com>wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> kanishk rastogi wrote:
> > hi all, I was trying to understand the xfs_buf_t management in XFS.
> >  I was looking at the xfs_bufhash_t which is used to keep the info
> > of buffers which have ongoing transactions. I wanted to know if it
> > uses it for caching these pages.
> >
> > As far as i can see in xfs_buf_relse  calls ::
> >
> > list_del_init(&bp->b_hash_list);
> >
> > if the count drops down to 0 for the buffer...
> >
> > if (atomic_dec_and_lock(&bp->b_hold, &hash->bh_lock)) { if
> > (bp->b_relse) { ----->set only in case of error??
> > atomic_inc(&bp->b_hold); spin_unlock(&hash->bh_lock);
> > (*(bp->b_relse)) (bp); } else if (bp->b_flags & XBF_FS_MANAGED) {
> > ----> mostly unset......\
> Indeed. that flag is used on exactly one buffer, the superblock.ode of the
> Basically the superblock is pinned in memory and only flushed by the
> filesystem.
> > spin_unlock(&hash->bh_lock); } else { ASSERT(!(bp->b_flags &
> > (XBF_DELWRI|_XBF_DELWRI_Q))); *list_del_init(&bp->b_hash_list);*
> > spin_unlock(&hash->bh_lock); xfs_buf_free(bp); } }
>


>
> >
> > ---------------
> >
> > CMIIW: So it doesnt keeps buffers if they are not in use....  ??
> buffers are released back to the buf_t  zone when they are done.
>
> zone allocation / free is very fast so managing lists of "free"
> buffers doesn't make a lot of sense.
>

sorry my fault i was asking about the pages these buffers point to? ....
I see them being pickd from the device inode of the monted device...

thankx
kanishk

>
> >
> > I dont think that should be the case...
> >
> > -- regards kanishk
> >
> > ----------------------------------------------------------------------
> >
> >
> > _______________________________________________ xfs mailing list
> > xfs@oss.sgi.com http://oss.sgi.com/mailman/listinfo/xfs
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.7 (Darwin)
> Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
>
> iD8DBQFKhaWBNRmM+OaGhBgRAtqgAJ9CZfj0QXz8Yy7P4jYemgEPsvR9pQCfcSWM
> a9OI8A4ckMdOuiXQ60o4PLE=
> =OJAD
> -----END PGP SIGNATURE-----
>
>


-- 
regards
kanishk

[-- Attachment #1.2: Type: text/html, Size: 3472 bytes --]

[-- Attachment #2: Type: text/plain, Size: 121 bytes --]

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs