[sheepdog] [PATCH 1/3] md: make vdisk management dynamic

MORITA Kazutaka morita.kazutaka at lab.ntt.co.jp
Tue Sep 3 10:05:48 CEST 2013


At Tue, 3 Sep 2013 15:46:36 +0800,
Liu Yuan wrote:
> 
> On Tue, Sep 03, 2013 at 04:34:04PM +0900, MORITA Kazutaka wrote:
> > At Tue,  3 Sep 2013 11:06:53 +0800,
> > Liu Yuan wrote:
> > > 
> > > diff --git a/sheep/md.c b/sheep/md.c
> > > index d51a7c6..f94fd29 100644
> > > --- a/sheep/md.c
> > > +++ b/sheep/md.c
> > > @@ -13,98 +13,116 @@
> > >  
> > >  #include "sheep_priv.h"
> > >  
> > > -#define MD_DEFAULT_VDISKS 128
> > > -#define MD_MAX_VDISK (MD_MAX_DISK * MD_DEFAULT_VDISKS)
> > > +#define MD_VDISK_SIZE ((uint64_t)1*1024*1024*1024) /* 1G */
> > > +
> > > +#define NONE_EXIST_PATH "/all/disks/are/broken/,ps/əʌo7/!"
> > 
> > The path may exist, no?  How about using "" as a non-existent path?
> 
> I think we can't use "", eg when it is used in "%s%lx" for oid, we'll
> oid for the working directory.

Ah, I see.

> 
> This path isn't magic enough?

Yes, it looks okay.

> 
> > > +static inline uint64_t vnode_hash(struct rb_node *node)
> > > +{
> > > +	struct vdisk *v;
> > >  
> > > -		nr++;
> > > -	}
> > > +	v = rb_entry(node, struct vdisk, rb);
> > > +	return v->hash;
> > > +}
> > >  
> > > -	return nr;
> > > +static inline struct vdisk *vnode_vdisk(struct rb_node *node)
> > > +{
> > > +	return rb_entry(node, struct vdisk, rb);
> > >  }
> > >  
> > > -static inline int disks_to_vdisks(const struct disk *ds, int nmds, struct vdisk *vds)
> > > +static struct vdisk *find_right_close(struct rb_node *node, uint64_t hash)
> > >  {
> > > -	int nr_vdisks = 0;
> > > +	struct vdisk *v;
> > >  
> > > -	for (int i = 0; i < nmds; i++)
> > > -		nr_vdisks += disk_to_vdisks(ds, i, vds + nr_vdisks);
> > > +	if (hash < vnode_hash(node)) {
> > > +		if (!node->rb_left)
> > > +			return vnode_vdisk(node);
> > >  
> > > -	xqsort(vds, nr_vdisks, vdisk_cmp);
> > > +		v = find_right_close(node->rb_left, hash);
> > > +		return v->hash - hash > vnode_hash(node) - hash ?
> > > +			vnode_vdisk(node) : v;
> > > +	} else if (hash > vnode_hash(node)) {
> > > +		if (!node->rb_right)
> > > +			return vnode_vdisk(rb_first(&md.vroot));
> > > +		return find_right_close(node->rb_right, hash);
> > > +	} else {
> > > +		return vnode_vdisk(node);
> > > +	}
> > > +}
> > >  
> > > -	return nr_vdisks;
> > > +/* If v1_hash < oid <= v2_hash, then oid is resident in v2 */
> > > +static struct vdisk *oid_to_vdisk(uint64_t oid)
> > > +{
> > > +	uint64_t hash = sd_hash_oid(oid);
> > > +
> > > +	return find_right_close(md.vroot.rb_node, hash);
> > >  }
> > 
> > I think we will implement a similar function when we change node
> > management from a fixed array to rbtree.  How about adding a helper
> > function in rbtree.h like as follows?  Then, the above code will be
> > much cleaner.
> > 
> > /*
> >  * Search for a value in the rbtree.  When the key is not found in the rbtree,
> >  * this returns the next greater node if one exists.
> >  */
> > #define rb_nsearch(root, key, member, compar)				\
> > ({									\
> > 	struct rb_node *__n = (root)->rb_node;				\
> > 	typeof(key) __ret = NULL, __data;				\
> > 									\
> > 	while (__n) {							\
> > 		__data = rb_entry(__n, typeof(*key), member);		\
> > 		int __cmp = compar(key, __data);			\
> > 									\
> > 		if (__cmp < 0) {					\
> > 			__ret = __data;					\
> > 			__n = __n->rb_left;				\
> > 		} else if (__cmp > 0)					\
> > 			__n = __n->rb_right;				\
> > 		else {							\
> > 			__ret = __data;					\
> > 			break;						\
> > 		}							\
> > 	}								\
> > 	__ret;								\
> > })
> > 
> 
> This macro can't return the right closest element, but yes, I can write a helper

Really?  I have no ideas about which part is wrong, but feel free to
update it.

Thanks,

Kazutaka



More information about the sheepdog mailing list