Date: Thu, 25 Sep 1997 23:11:44 +1000 From: Bruce Evans <bde@zeta.org.au> To: kato@migmatite.eps.nagoya-u.ac.jp, phk@critter.freebsd.dk Cc: current@FreeBSD.ORG Subject: Re: Daily SNAPshots at current.freebsd.org shut down for now. Message-ID: <199709251311.XAA16478@godzilla.zeta.org.au>
next in thread | raw e-mail | index | archive | help
>> I would suggest you make a mount option for it so you can decide for
>> other filesystems if you want clustering or not.
>
>I don't have enough time, so what I can do now is to add vnconfig
>option to disable clustering. (I also think mount option is good
>mechanism and I need it to use union fs safely.)
There's always time to do it again? :-)
I think this has a major problem - it disables clustering in the wrong
layer. I/O to the vn file must be clustered for efficiency.
The mount option would automatically apply to the correct layer.
>diff -arcN sys.old/gnu/ext2fs/ext2_readwrite.c sys/gnu/ext2fs/ext2_readwrite.c
>*** sys.old/gnu/ext2fs/ext2_readwrite.c Wed Sep 24 20:27:46 1997
>--- sys/gnu/ext2fs/ext2_readwrite.c Wed Sep 24 22:44:01 1997
>***************
>*** 112,118 ****
>
> if (lblktosize(fs, nextlbn) >= ip->i_size)
> error = bread(vp, lbn, size, NOCRED, &bp);
>! else if (doclusterread)
> error = cluster_read(vp,
> ip->i_size, lbn, size, NOCRED, &bp);
> else if (lbn - 1 == vp->v_lastr) {
>--- 112,118 ----
>
> if (lblktosize(fs, nextlbn) >= ip->i_size)
> error = bread(vp, lbn, size, NOCRED, &bp);
>! else if (doclusterread && !(vp->v_flag & VNOCLUSTER))
> error = cluster_read(vp,
> ip->i_size, lbn, size, NOCRED, &bp);
> else if (lbn - 1 == vp->v_lastr) {
For the current version, I think the test needs to be something like
else if (doclusterread &&
!(vp->v_mount->mnt_vnodecovered & VNOCLUSTER))
Perhaps the mount-option version should eliminate doclusterread and
doclusterwrite (this requires a mount option for each) and move the
tests into cluster_read() and cluster_write().
>diff -arcN sys.old/sys/vnode.h sys/sys/vnode.h
>*** sys.old/sys/vnode.h Wed Sep 24 20:26:36 1997
>--- sys/sys/vnode.h Wed Sep 24 22:37:44 1997
>***************
>*** 111,129 ****
> /*
> * Vnode flags.
> */
>! #define VROOT 0x0001 /* root of its file system */
>! #define VTEXT 0x0002 /* vnode is a pure text prototype */
>! #define VSYSTEM 0x0004 /* vnode being used by kernel */
>! #define VOLOCK 0x0008 /* vnode is locked waiting for an object */
>! #define VOWANT 0x0010 /* a process is waiting for VOLOCK */
>! #define VXLOCK 0x0100 /* vnode is locked to change underlying type */
>! #define VXWANT 0x0200 /* process is waiting for vnode */
>! #define VBWAIT 0x0400 /* waiting for output to complete */
>! #define VALIASED 0x0800 /* vnode has an alias */
>! #define VDIROP 0x1000 /* LFS: vnode is involved in a directory op */
>! #define VVMIO 0x2000 /* VMIO flag */
>! #define VNINACT 0x4000 /* LFS: skip ufs_inactive() in lfs_vunref */
>! #define VAGE 0x8000 /* Insert vnode at head of free list */
>
> /*
> * Vnode attributes. A field value of VNOVAL represents a field whose value
>--- 111,130 ----
> /*
> * Vnode flags.
> */
>! #define VROOT 0x00001 /* root of its file system */
>! #define VTEXT 0x00002 /* vnode is a pure text prototype */
>! #define VSYSTEM 0x00004 /* vnode being used by kernel */
>! #define VOLOCK 0x00008 /* vnode is locked waiting for an object */
>! #define VOWANT 0x00010 /* a process is waiting for VOLOCK */
>! #define VXLOCK 0x00100 /* vnode is locked to change underlying type */
>! #define VXWANT 0x00200 /* process is waiting for vnode */
>! #define VBWAIT 0x00400 /* waiting for output to complete */
>! #define VALIASED 0x00800 /* vnode has an alias */
>! #define VDIROP 0x01000 /* LFS: vnode is involved in a directory op */
>! #define VVMIO 0x02000 /* VMIO flag */
>! #define VNINACT 0x04000 /* LFS: skip ufs_inactive() in lfs_vunref */
>! #define VAGE 0x08000 /* Insert vnode at head of free list */
>! #define VNOCLUSTER 0x10000 /* disable cluster read / write */
>
> /*
> * Vnode attributes. A field value of VNOVAL represents a field whose value
I don't like large essentially-null changes just to keep 0x numbers
lined up. Rev.1.49 has already added another digit to all the numbers.
You would need to change them all again.
Bruce
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199709251311.XAA16478>