Commit ff181ed8 authored by Alexander Duyck's avatar Alexander Duyck Committed by David S. Miller

fib_trie: Push assignment of child to parent down into inflate/halve

This change makes it so that the assignment of the tnode to the parent is
handled directly within whatever function is currently handling the node be
it inflate, halve, or resize.  By doing this we can avoid some of the need
to set NULL pointers in the tree while we are resizing the subnodes.
Signed-off-by: default avatarAlexander Duyck <alexander.h.duyck@redhat.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent f05a4819
...@@ -146,9 +146,7 @@ struct trie { ...@@ -146,9 +146,7 @@ struct trie {
#endif #endif
}; };
static void tnode_put_child_reorg(struct tnode *tn, unsigned long i, static void resize(struct trie *t, struct tnode *tn);
struct tnode *n, int wasfull);
static struct tnode *resize(struct trie *t, struct tnode *tn);
/* tnodes to free after resize(); protected by RTNL */ /* tnodes to free after resize(); protected by RTNL */
static struct callback_head *tnode_free_head; static struct callback_head *tnode_free_head;
static size_t tnode_free_size; static size_t tnode_free_size;
...@@ -396,22 +394,13 @@ static inline int tnode_full(const struct tnode *tn, const struct tnode *n) ...@@ -396,22 +394,13 @@ static inline int tnode_full(const struct tnode *tn, const struct tnode *n)
return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n); return n && ((n->pos + n->bits) == tn->pos) && IS_TNODE(n);
} }
static inline void put_child(struct tnode *tn, unsigned long i, /* Add a child at position i overwriting the old value.
struct tnode *n) * Update the value of full_children and empty_children.
{ */
tnode_put_child_reorg(tn, i, n, -1); static void put_child(struct tnode *tn, unsigned long i, struct tnode *n)
}
/*
* Add a child at position i overwriting the old value.
* Update the value of full_children and empty_children.
*/
static void tnode_put_child_reorg(struct tnode *tn, unsigned long i,
struct tnode *n, int wasfull)
{ {
struct tnode *chi = rtnl_dereference(tn->child[i]); struct tnode *chi = rtnl_dereference(tn->child[i]);
int isfull; int isfull, wasfull;
BUG_ON(i >= tnode_child_length(tn)); BUG_ON(i >= tnode_child_length(tn));
...@@ -422,10 +411,9 @@ static void tnode_put_child_reorg(struct tnode *tn, unsigned long i, ...@@ -422,10 +411,9 @@ static void tnode_put_child_reorg(struct tnode *tn, unsigned long i,
tn->empty_children--; tn->empty_children--;
/* update fullChildren */ /* update fullChildren */
if (wasfull == -1) wasfull = tnode_full(tn, chi);
wasfull = tnode_full(tn, chi);
isfull = tnode_full(tn, n); isfull = tnode_full(tn, n);
if (wasfull && !isfull) if (wasfull && !isfull)
tn->full_children--; tn->full_children--;
else if (!wasfull && isfull) else if (!wasfull && isfull)
...@@ -458,9 +446,10 @@ static void tnode_clean_free(struct tnode *tn) ...@@ -458,9 +446,10 @@ static void tnode_clean_free(struct tnode *tn)
node_free(tn); node_free(tn);
} }
static struct tnode *inflate(struct trie *t, struct tnode *oldtnode) static int inflate(struct trie *t, struct tnode *oldtnode)
{ {
unsigned long olen = tnode_child_length(oldtnode); unsigned long olen = tnode_child_length(oldtnode);
struct tnode *tp = node_parent(oldtnode);
struct tnode *tn; struct tnode *tn;
unsigned long i; unsigned long i;
t_key m; t_key m;
...@@ -468,9 +457,8 @@ static struct tnode *inflate(struct trie *t, struct tnode *oldtnode) ...@@ -468,9 +457,8 @@ static struct tnode *inflate(struct trie *t, struct tnode *oldtnode)
pr_debug("In inflate\n"); pr_debug("In inflate\n");
tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1); tn = tnode_new(oldtnode->key, oldtnode->pos - 1, oldtnode->bits + 1);
if (!tn) if (!tn)
return ERR_PTR(-ENOMEM); return -ENOMEM;
/* /*
* Preallocate and store tnodes before the actual work so we * Preallocate and store tnodes before the actual work so we
...@@ -564,30 +552,36 @@ static struct tnode *inflate(struct trie *t, struct tnode *oldtnode) ...@@ -564,30 +552,36 @@ static struct tnode *inflate(struct trie *t, struct tnode *oldtnode)
put_child(left, j, rtnl_dereference(inode->child[j])); put_child(left, j, rtnl_dereference(inode->child[j]));
put_child(right, j, rtnl_dereference(inode->child[j + size])); put_child(right, j, rtnl_dereference(inode->child[j + size]));
} }
put_child(tn, 2*i, resize(t, left));
put_child(tn, 2*i+1, resize(t, right)); put_child(tn, 2 * i, left);
put_child(tn, 2 * i + 1, right);
tnode_free_safe(inode); tnode_free_safe(inode);
resize(t, left);
resize(t, right);
} }
put_child_root(tp, t, tn->key, tn);
tnode_free_safe(oldtnode); tnode_free_safe(oldtnode);
return tn; return 0;
nomem: nomem:
tnode_clean_free(tn); tnode_clean_free(tn);
return ERR_PTR(-ENOMEM); return -ENOMEM;
} }
static struct tnode *halve(struct trie *t, struct tnode *oldtnode) static int halve(struct trie *t, struct tnode *oldtnode)
{ {
unsigned long olen = tnode_child_length(oldtnode); unsigned long olen = tnode_child_length(oldtnode);
struct tnode *tp = node_parent(oldtnode);
struct tnode *tn, *left, *right; struct tnode *tn, *left, *right;
int i; int i;
pr_debug("In halve\n"); pr_debug("In halve\n");
tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1); tn = tnode_new(oldtnode->key, oldtnode->pos + 1, oldtnode->bits - 1);
if (!tn) if (!tn)
return ERR_PTR(-ENOMEM); return -ENOMEM;
/* /*
* Preallocate and store tnodes before the actual work so we * Preallocate and store tnodes before the actual work so we
...@@ -606,8 +600,10 @@ static struct tnode *halve(struct trie *t, struct tnode *oldtnode) ...@@ -606,8 +600,10 @@ static struct tnode *halve(struct trie *t, struct tnode *oldtnode)
newn = tnode_new(left->key, oldtnode->pos, 1); newn = tnode_new(left->key, oldtnode->pos, 1);
if (!newn) if (!newn) {
goto nomem; tnode_clean_free(tn);
return -ENOMEM;
}
put_child(tn, i/2, newn); put_child(tn, i/2, newn);
} }
...@@ -635,16 +631,18 @@ static struct tnode *halve(struct trie *t, struct tnode *oldtnode) ...@@ -635,16 +631,18 @@ static struct tnode *halve(struct trie *t, struct tnode *oldtnode)
/* Two nonempty children */ /* Two nonempty children */
newBinNode = tnode_get_child(tn, i/2); newBinNode = tnode_get_child(tn, i/2);
put_child(tn, i/2, NULL);
put_child(newBinNode, 0, left); put_child(newBinNode, 0, left);
put_child(newBinNode, 1, right); put_child(newBinNode, 1, right);
put_child(tn, i/2, resize(t, newBinNode));
put_child(tn, i / 2, newBinNode);
resize(t, newBinNode);
} }
put_child_root(tp, t, tn->key, tn);
tnode_free_safe(oldtnode); tnode_free_safe(oldtnode);
return tn;
nomem: return 0;
tnode_clean_free(tn);
return ERR_PTR(-ENOMEM);
} }
/* From "Implementing a dynamic compressed trie" by Stefan Nilsson of /* From "Implementing a dynamic compressed trie" by Stefan Nilsson of
...@@ -704,45 +702,48 @@ static struct tnode *halve(struct trie *t, struct tnode *oldtnode) ...@@ -704,45 +702,48 @@ static struct tnode *halve(struct trie *t, struct tnode *oldtnode)
* tnode_child_length(tn) * tnode_child_length(tn)
* *
*/ */
static bool should_inflate(const struct tnode *tn) static bool should_inflate(const struct tnode *tp, const struct tnode *tn)
{ {
unsigned long used = tnode_child_length(tn); unsigned long used = tnode_child_length(tn);
unsigned long threshold = used; unsigned long threshold = used;
/* Keep root node larger */ /* Keep root node larger */
threshold *= node_parent(tn) ? inflate_threshold : threshold *= tp ? inflate_threshold : inflate_threshold_root;
inflate_threshold_root;
used += tn->full_children; used += tn->full_children;
used -= tn->empty_children; used -= tn->empty_children;
return tn->pos && ((50 * used) >= threshold); return tn->pos && ((50 * used) >= threshold);
} }
static bool should_halve(const struct tnode *tn) static bool should_halve(const struct tnode *tp, const struct tnode *tn)
{ {
unsigned long used = tnode_child_length(tn); unsigned long used = tnode_child_length(tn);
unsigned long threshold = used; unsigned long threshold = used;
/* Keep root node larger */ /* Keep root node larger */
threshold *= node_parent(tn) ? halve_threshold : threshold *= tp ? halve_threshold : halve_threshold_root;
halve_threshold_root;
used -= tn->empty_children; used -= tn->empty_children;
return (tn->bits > 1) && ((100 * used) < threshold); return (tn->bits > 1) && ((100 * used) < threshold);
} }
#define MAX_WORK 10 #define MAX_WORK 10
static struct tnode *resize(struct trie *t, struct tnode *tn) static void resize(struct trie *t, struct tnode *tn)
{ {
struct tnode *old_tn, *n = NULL; struct tnode *tp = node_parent(tn), *n = NULL;
struct tnode __rcu **cptr;
int max_work; int max_work;
if (!tn)
return NULL;
pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n", pr_debug("In tnode_resize %p inflate_threshold=%d threshold=%d\n",
tn, inflate_threshold, halve_threshold); tn, inflate_threshold, halve_threshold);
/* track the tnode via the pointer from the parent instead of
* doing it ourselves. This way we can let RCU fully do its
* thing without us interfering
*/
cptr = tp ? &tp->child[get_index(tn->key, tp)] : &t->trie;
BUG_ON(tn != rtnl_dereference(*cptr));
/* No children */ /* No children */
if (tn->empty_children > (tnode_child_length(tn) - 1)) if (tn->empty_children > (tnode_child_length(tn) - 1))
goto no_children; goto no_children;
...@@ -755,39 +756,35 @@ static struct tnode *resize(struct trie *t, struct tnode *tn) ...@@ -755,39 +756,35 @@ static struct tnode *resize(struct trie *t, struct tnode *tn)
* nonempty nodes that are above the threshold. * nonempty nodes that are above the threshold.
*/ */
max_work = MAX_WORK; max_work = MAX_WORK;
while (should_inflate(tn) && max_work--) { while (should_inflate(tp, tn) && max_work--) {
old_tn = tn; if (inflate(t, tn)) {
tn = inflate(t, tn);
if (IS_ERR(tn)) {
tn = old_tn;
#ifdef CONFIG_IP_FIB_TRIE_STATS #ifdef CONFIG_IP_FIB_TRIE_STATS
this_cpu_inc(t->stats->resize_node_skipped); this_cpu_inc(t->stats->resize_node_skipped);
#endif #endif
break; break;
} }
tn = rtnl_dereference(*cptr);
} }
/* Return if at least one inflate is run */ /* Return if at least one inflate is run */
if (max_work != MAX_WORK) if (max_work != MAX_WORK)
return tn; return;
/* Halve as long as the number of empty children in this /* Halve as long as the number of empty children in this
* node is above threshold. * node is above threshold.
*/ */
max_work = MAX_WORK; max_work = MAX_WORK;
while (should_halve(tn) && max_work--) { while (should_halve(tp, tn) && max_work--) {
old_tn = tn; if (halve(t, tn)) {
tn = halve(t, tn);
if (IS_ERR(tn)) {
tn = old_tn;
#ifdef CONFIG_IP_FIB_TRIE_STATS #ifdef CONFIG_IP_FIB_TRIE_STATS
this_cpu_inc(t->stats->resize_node_skipped); this_cpu_inc(t->stats->resize_node_skipped);
#endif #endif
break; break;
} }
}
tn = rtnl_dereference(*cptr);
}
/* Only one child remains */ /* Only one child remains */
if (tn->empty_children == (tnode_child_length(tn) - 1)) { if (tn->empty_children == (tnode_child_length(tn) - 1)) {
...@@ -797,11 +794,12 @@ static struct tnode *resize(struct trie *t, struct tnode *tn) ...@@ -797,11 +794,12 @@ static struct tnode *resize(struct trie *t, struct tnode *tn)
n = tnode_get_child(tn, --i); n = tnode_get_child(tn, --i);
no_children: no_children:
/* compress one level */ /* compress one level */
node_set_parent(n, NULL); put_child_root(tp, t, tn->key, n);
node_set_parent(n, tp);
/* drop dead node */
tnode_free_safe(tn); tnode_free_safe(tn);
return n;
} }
return tn;
} }
/* readside must use rcu_read_lock currently dump routines /* readside must use rcu_read_lock currently dump routines
...@@ -882,34 +880,19 @@ static struct tnode *fib_find_node(struct trie *t, u32 key) ...@@ -882,34 +880,19 @@ static struct tnode *fib_find_node(struct trie *t, u32 key)
static void trie_rebalance(struct trie *t, struct tnode *tn) static void trie_rebalance(struct trie *t, struct tnode *tn)
{ {
int wasfull;
t_key cindex, key;
struct tnode *tp; struct tnode *tp;
key = tn->key; while ((tp = node_parent(tn)) != NULL) {
resize(t, tn);
while (tn != NULL && (tp = node_parent(tn)) != NULL) {
cindex = get_index(key, tp);
wasfull = tnode_full(tp, tnode_get_child(tp, cindex));
tn = resize(t, tn);
tnode_put_child_reorg(tp, cindex, tn, wasfull);
tp = node_parent(tn);
if (!tp)
rcu_assign_pointer(t->trie, tn);
tnode_free_flush(); tnode_free_flush();
if (!tp)
break;
tn = tp; tn = tp;
} }
/* Handle last (top) tnode */ /* Handle last (top) tnode */
if (IS_TNODE(tn)) if (IS_TNODE(tn))
tn = resize(t, tn); resize(t, tn);
rcu_assign_pointer(t->trie, tn);
tnode_free_flush(); tnode_free_flush();
} }
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment