f7f82b81d2
Walking backref tree and btrfs quota rely on ulist very much. This patch tries to use rb_tree to speed up search time. The original code always checks whether an element exists before adding a new element, however it costs O(n). I try to add a rb_tree in the ulist,this is only used to speed up search. I also do some measurements with quota enabled. fsstress -p 4 -n 10000 Without this path: real 0m51.058s 2m4.745s 1m28.222s 1m5.137s user 0m0.035s 0m0.041s 0m0.105s 0m0.100s sys 0m12.009s 0m11.246s 0m10.901s 0m10.999s 0m11.287s With this path: real 0m55.295s 0m50.960s 1m2.214s 0m48.273s user 0m0.053s 0m0.095s 0m0.135s 0m0.107s sys 0m7.766s 0m6.013s 0m6.319s 0m6.030s 0m6.532s After applying the patch,the execute time is down by ~42%.(11.287s->6.532s) Signed-off-by: Wang Shilong <wangsl-fnst@cn.fujitsu.com> Reviewed-by: Miao Xie <miaox@cn.fujitsu.com> Reviewed-by: Jan Schmidt <list.btrfs@jan-o-sch.net> Signed-off-by: Josef Bacik <jbacik@fusionio.com>
264 lines
6.8 KiB
C
264 lines
6.8 KiB
C
/*
|
|
* Copyright (C) 2011 STRATO AG
|
|
* written by Arne Jansen <sensille@gmx.net>
|
|
* Distributed under the GNU GPL license version 2.
|
|
*/
|
|
|
|
#include <linux/slab.h>
|
|
#include <linux/export.h>
|
|
#include "ulist.h"
|
|
|
|
/*
|
|
* ulist is a generic data structure to hold a collection of unique u64
|
|
* values. The only operations it supports is adding to the list and
|
|
* enumerating it.
|
|
* It is possible to store an auxiliary value along with the key.
|
|
*
|
|
* The implementation is preliminary and can probably be sped up
|
|
* significantly. A first step would be to store the values in an rbtree
|
|
* as soon as ULIST_SIZE is exceeded.
|
|
*
|
|
* A sample usage for ulists is the enumeration of directed graphs without
|
|
* visiting a node twice. The pseudo-code could look like this:
|
|
*
|
|
* ulist = ulist_alloc();
|
|
* ulist_add(ulist, root);
|
|
* ULIST_ITER_INIT(&uiter);
|
|
*
|
|
* while ((elem = ulist_next(ulist, &uiter)) {
|
|
* for (all child nodes n in elem)
|
|
* ulist_add(ulist, n);
|
|
* do something useful with the node;
|
|
* }
|
|
* ulist_free(ulist);
|
|
*
|
|
* This assumes the graph nodes are adressable by u64. This stems from the
|
|
* usage for tree enumeration in btrfs, where the logical addresses are
|
|
* 64 bit.
|
|
*
|
|
* It is also useful for tree enumeration which could be done elegantly
|
|
* recursively, but is not possible due to kernel stack limitations. The
|
|
* loop would be similar to the above.
|
|
*/
|
|
|
|
/**
|
|
* ulist_init - freshly initialize a ulist
|
|
* @ulist: the ulist to initialize
|
|
*
|
|
* Note: don't use this function to init an already used ulist, use
|
|
* ulist_reinit instead.
|
|
*/
|
|
void ulist_init(struct ulist *ulist)
|
|
{
|
|
ulist->nnodes = 0;
|
|
ulist->nodes = ulist->int_nodes;
|
|
ulist->nodes_alloced = ULIST_SIZE;
|
|
ulist->root = RB_ROOT;
|
|
}
|
|
EXPORT_SYMBOL(ulist_init);
|
|
|
|
/**
|
|
* ulist_fini - free up additionally allocated memory for the ulist
|
|
* @ulist: the ulist from which to free the additional memory
|
|
*
|
|
* This is useful in cases where the base 'struct ulist' has been statically
|
|
* allocated.
|
|
*/
|
|
void ulist_fini(struct ulist *ulist)
|
|
{
|
|
/*
|
|
* The first ULIST_SIZE elements are stored inline in struct ulist.
|
|
* Only if more elements are alocated they need to be freed.
|
|
*/
|
|
if (ulist->nodes_alloced > ULIST_SIZE)
|
|
kfree(ulist->nodes);
|
|
ulist->nodes_alloced = 0; /* in case ulist_fini is called twice */
|
|
ulist->root = RB_ROOT;
|
|
}
|
|
EXPORT_SYMBOL(ulist_fini);
|
|
|
|
/**
|
|
* ulist_reinit - prepare a ulist for reuse
|
|
* @ulist: ulist to be reused
|
|
*
|
|
* Free up all additional memory allocated for the list elements and reinit
|
|
* the ulist.
|
|
*/
|
|
void ulist_reinit(struct ulist *ulist)
|
|
{
|
|
ulist_fini(ulist);
|
|
ulist_init(ulist);
|
|
}
|
|
EXPORT_SYMBOL(ulist_reinit);
|
|
|
|
/**
|
|
* ulist_alloc - dynamically allocate a ulist
|
|
* @gfp_mask: allocation flags to for base allocation
|
|
*
|
|
* The allocated ulist will be returned in an initialized state.
|
|
*/
|
|
struct ulist *ulist_alloc(gfp_t gfp_mask)
|
|
{
|
|
struct ulist *ulist = kmalloc(sizeof(*ulist), gfp_mask);
|
|
|
|
if (!ulist)
|
|
return NULL;
|
|
|
|
ulist_init(ulist);
|
|
|
|
return ulist;
|
|
}
|
|
EXPORT_SYMBOL(ulist_alloc);
|
|
|
|
/**
|
|
* ulist_free - free dynamically allocated ulist
|
|
* @ulist: ulist to free
|
|
*
|
|
* It is not necessary to call ulist_fini before.
|
|
*/
|
|
void ulist_free(struct ulist *ulist)
|
|
{
|
|
if (!ulist)
|
|
return;
|
|
ulist_fini(ulist);
|
|
kfree(ulist);
|
|
}
|
|
EXPORT_SYMBOL(ulist_free);
|
|
|
|
static struct ulist_node *ulist_rbtree_search(struct ulist *ulist, u64 val)
|
|
{
|
|
struct rb_node *n = ulist->root.rb_node;
|
|
struct ulist_node *u = NULL;
|
|
|
|
while (n) {
|
|
u = rb_entry(n, struct ulist_node, rb_node);
|
|
if (u->val < val)
|
|
n = n->rb_right;
|
|
else if (u->val > val)
|
|
n = n->rb_left;
|
|
else
|
|
return u;
|
|
}
|
|
return NULL;
|
|
}
|
|
|
|
static int ulist_rbtree_insert(struct ulist *ulist, struct ulist_node *ins)
|
|
{
|
|
struct rb_node **p = &ulist->root.rb_node;
|
|
struct rb_node *parent = NULL;
|
|
struct ulist_node *cur = NULL;
|
|
|
|
while (*p) {
|
|
parent = *p;
|
|
cur = rb_entry(parent, struct ulist_node, rb_node);
|
|
|
|
if (cur->val < ins->val)
|
|
p = &(*p)->rb_right;
|
|
else if (cur->val > ins->val)
|
|
p = &(*p)->rb_left;
|
|
else
|
|
return -EEXIST;
|
|
}
|
|
rb_link_node(&ins->rb_node, parent, p);
|
|
rb_insert_color(&ins->rb_node, &ulist->root);
|
|
return 0;
|
|
}
|
|
|
|
/**
|
|
* ulist_add - add an element to the ulist
|
|
* @ulist: ulist to add the element to
|
|
* @val: value to add to ulist
|
|
* @aux: auxiliary value to store along with val
|
|
* @gfp_mask: flags to use for allocation
|
|
*
|
|
* Note: locking must be provided by the caller. In case of rwlocks write
|
|
* locking is needed
|
|
*
|
|
* Add an element to a ulist. The @val will only be added if it doesn't
|
|
* already exist. If it is added, the auxiliary value @aux is stored along with
|
|
* it. In case @val already exists in the ulist, @aux is ignored, even if
|
|
* it differs from the already stored value.
|
|
*
|
|
* ulist_add returns 0 if @val already exists in ulist and 1 if @val has been
|
|
* inserted.
|
|
* In case of allocation failure -ENOMEM is returned and the ulist stays
|
|
* unaltered.
|
|
*/
|
|
int ulist_add(struct ulist *ulist, u64 val, u64 aux, gfp_t gfp_mask)
|
|
{
|
|
return ulist_add_merge(ulist, val, aux, NULL, gfp_mask);
|
|
}
|
|
|
|
int ulist_add_merge(struct ulist *ulist, u64 val, u64 aux,
|
|
u64 *old_aux, gfp_t gfp_mask)
|
|
{
|
|
int ret = 0;
|
|
struct ulist_node *node = NULL;
|
|
node = ulist_rbtree_search(ulist, val);
|
|
if (node) {
|
|
if (old_aux)
|
|
*old_aux = node->aux;
|
|
return 0;
|
|
}
|
|
|
|
if (ulist->nnodes >= ulist->nodes_alloced) {
|
|
u64 new_alloced = ulist->nodes_alloced + 128;
|
|
struct ulist_node *new_nodes;
|
|
void *old = NULL;
|
|
|
|
/*
|
|
* if nodes_alloced == ULIST_SIZE no memory has been allocated
|
|
* yet, so pass NULL to krealloc
|
|
*/
|
|
if (ulist->nodes_alloced > ULIST_SIZE)
|
|
old = ulist->nodes;
|
|
|
|
new_nodes = krealloc(old, sizeof(*new_nodes) * new_alloced,
|
|
gfp_mask);
|
|
if (!new_nodes)
|
|
return -ENOMEM;
|
|
|
|
if (!old)
|
|
memcpy(new_nodes, ulist->int_nodes,
|
|
sizeof(ulist->int_nodes));
|
|
|
|
ulist->nodes = new_nodes;
|
|
ulist->nodes_alloced = new_alloced;
|
|
}
|
|
ulist->nodes[ulist->nnodes].val = val;
|
|
ulist->nodes[ulist->nnodes].aux = aux;
|
|
ret = ulist_rbtree_insert(ulist, &ulist->nodes[ulist->nnodes]);
|
|
BUG_ON(ret);
|
|
++ulist->nnodes;
|
|
|
|
return 1;
|
|
}
|
|
EXPORT_SYMBOL(ulist_add);
|
|
|
|
/**
|
|
* ulist_next - iterate ulist
|
|
* @ulist: ulist to iterate
|
|
* @uiter: iterator variable, initialized with ULIST_ITER_INIT(&iterator)
|
|
*
|
|
* Note: locking must be provided by the caller. In case of rwlocks only read
|
|
* locking is needed
|
|
*
|
|
* This function is used to iterate an ulist.
|
|
* It returns the next element from the ulist or %NULL when the
|
|
* end is reached. No guarantee is made with respect to the order in which
|
|
* the elements are returned. They might neither be returned in order of
|
|
* addition nor in ascending order.
|
|
* It is allowed to call ulist_add during an enumeration. Newly added items
|
|
* are guaranteed to show up in the running enumeration.
|
|
*/
|
|
struct ulist_node *ulist_next(struct ulist *ulist, struct ulist_iterator *uiter)
|
|
{
|
|
if (ulist->nnodes == 0)
|
|
return NULL;
|
|
if (uiter->i < 0 || uiter->i >= ulist->nnodes)
|
|
return NULL;
|
|
|
|
return &ulist->nodes[uiter->i++];
|
|
}
|
|
EXPORT_SYMBOL(ulist_next);
|