inetpeer: Abstract out the tree root accesses.
authorDavid S. Miller <davem@davemloft.net>
Tue, 30 Nov 2010 19:41:59 +0000 (11:41 -0800)
committerDavid S. Miller <davem@davemloft.net>
Tue, 30 Nov 2010 19:41:59 +0000 (11:41 -0800)
Instead of directly accessing "peer", change to code to
operate using a "struct inet_peer_base *" pointer.

This will facilitate the addition of a seperate tree for
ipv6 peer entries.

Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv4/inetpeer.c

index 9e94d7c..f944008 100644 (file)
@@ -79,13 +79,13 @@ static const struct inet_peer peer_fake_node = {
        .avl_height     = 0
 };
 
-static struct {
+static struct inet_peer_base {
        struct inet_peer __rcu *root;
        spinlock_t      lock;
        int             total;
-} peers = {
+} v4_peers = {
        .root           = peer_avl_empty_rcu,
-       .lock           = __SPIN_LOCK_UNLOCKED(peers.lock),
+       .lock           = __SPIN_LOCK_UNLOCKED(v4_peers.lock),
        .total          = 0,
 };
 #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
@@ -155,15 +155,15 @@ static void unlink_from_unused(struct inet_peer *p)
 /*
  * Called with local BH disabled and the pool lock held.
  */
-#define lookup(_daddr, _stack)                                         \
+#define lookup(_daddr, _stack, _base)                          \
 ({                                                             \
        struct inet_peer *u;                                    \
        struct inet_peer __rcu **v;                             \
                                                                \
        stackptr = _stack;                                      \
-       *stackptr++ = &peers.root;                              \
-       for (u = rcu_dereference_protected(peers.root,          \
-                       lockdep_is_held(&peers.lock));          \
+       *stackptr++ = &_base->root;                             \
+       for (u = rcu_dereference_protected(_base->root,         \
+                       lockdep_is_held(&_base->lock));         \
             u != peer_avl_empty; ) {                           \
                if (_daddr == u->v4daddr)                       \
                        break;                                  \
@@ -173,7 +173,7 @@ static void unlink_from_unused(struct inet_peer *p)
                        v = &u->avl_right;                      \
                *stackptr++ = v;                                \
                u = rcu_dereference_protected(*v,               \
-                       lockdep_is_held(&peers.lock));          \
+                       lockdep_is_held(&_base->lock));         \
        }                                                       \
        u;                                                      \
 })
@@ -185,9 +185,9 @@ static void unlink_from_unused(struct inet_peer *p)
  * But every pointer we follow is guaranteed to be valid thanks to RCU.
  * We exit from this function if number of links exceeds PEER_MAXDEPTH
  */
-static struct inet_peer *lookup_rcu_bh(__be32 daddr)
+static struct inet_peer *lookup_rcu_bh(__be32 daddr, struct inet_peer_base *base)
 {
-       struct inet_peer *u = rcu_dereference_bh(peers.root);
+       struct inet_peer *u = rcu_dereference_bh(base->root);
        int count = 0;
 
        while (u != peer_avl_empty) {
@@ -212,19 +212,19 @@ static struct inet_peer *lookup_rcu_bh(__be32 daddr)
 }
 
 /* Called with local BH disabled and the pool lock held. */
-#define lookup_rightempty(start)                               \
+#define lookup_rightempty(start, base)                         \
 ({                                                             \
        struct inet_peer *u;                                    \
        struct inet_peer __rcu **v;                             \
        *stackptr++ = &start->avl_left;                         \
        v = &start->avl_left;                                   \
        for (u = rcu_dereference_protected(*v,                  \
-                       lockdep_is_held(&peers.lock));          \
+                       lockdep_is_held(&base->lock));          \
             u->avl_right != peer_avl_empty_rcu; ) {            \
                v = &u->avl_right;                              \
                *stackptr++ = v;                                \
                u = rcu_dereference_protected(*v,               \
-                       lockdep_is_held(&peers.lock));          \
+                       lockdep_is_held(&base->lock));          \
        }                                                       \
        u;                                                      \
 })
@@ -234,7 +234,8 @@ static struct inet_peer *lookup_rcu_bh(__be32 daddr)
  * Look into mm/map_avl.c for more detail description of the ideas.
  */
 static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
-               struct inet_peer __rcu ***stackend)
+                              struct inet_peer __rcu ***stackend,
+                              struct inet_peer_base *base)
 {
        struct inet_peer __rcu **nodep;
        struct inet_peer *node, *l, *r;
@@ -243,20 +244,20 @@ static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
        while (stackend > stack) {
                nodep = *--stackend;
                node = rcu_dereference_protected(*nodep,
-                               lockdep_is_held(&peers.lock));
+                               lockdep_is_held(&base->lock));
                l = rcu_dereference_protected(node->avl_left,
-                               lockdep_is_held(&peers.lock));
+                               lockdep_is_held(&base->lock));
                r = rcu_dereference_protected(node->avl_right,
-                               lockdep_is_held(&peers.lock));
+                               lockdep_is_held(&base->lock));
                lh = node_height(l);
                rh = node_height(r);
                if (lh > rh + 1) { /* l: RH+2 */
                        struct inet_peer *ll, *lr, *lrl, *lrr;
                        int lrh;
                        ll = rcu_dereference_protected(l->avl_left,
-                               lockdep_is_held(&peers.lock));
+                               lockdep_is_held(&base->lock));
                        lr = rcu_dereference_protected(l->avl_right,
-                               lockdep_is_held(&peers.lock));
+                               lockdep_is_held(&base->lock));
                        lrh = node_height(lr);
                        if (lrh <= node_height(ll)) {   /* ll: RH+1 */
                                RCU_INIT_POINTER(node->avl_left, lr);   /* lr: RH or RH+1 */
@@ -268,9 +269,9 @@ static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
                                RCU_INIT_POINTER(*nodep, l);
                        } else { /* ll: RH, lr: RH+1 */
                                lrl = rcu_dereference_protected(lr->avl_left,
-                                       lockdep_is_held(&peers.lock));  /* lrl: RH or RH-1 */
+                                       lockdep_is_held(&base->lock));  /* lrl: RH or RH-1 */
                                lrr = rcu_dereference_protected(lr->avl_right,
-                                       lockdep_is_held(&peers.lock));  /* lrr: RH or RH-1 */
+                                       lockdep_is_held(&base->lock));  /* lrr: RH or RH-1 */
                                RCU_INIT_POINTER(node->avl_left, lrr);  /* lrr: RH or RH-1 */
                                RCU_INIT_POINTER(node->avl_right, r);   /* r: RH */
                                node->avl_height = rh + 1; /* node: RH+1 */
@@ -286,9 +287,9 @@ static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
                        struct inet_peer *rr, *rl, *rlr, *rll;
                        int rlh;
                        rr = rcu_dereference_protected(r->avl_right,
-                               lockdep_is_held(&peers.lock));
+                               lockdep_is_held(&base->lock));
                        rl = rcu_dereference_protected(r->avl_left,
-                               lockdep_is_held(&peers.lock));
+                               lockdep_is_held(&base->lock));
                        rlh = node_height(rl);
                        if (rlh <= node_height(rr)) {   /* rr: LH+1 */
                                RCU_INIT_POINTER(node->avl_right, rl);  /* rl: LH or LH+1 */
@@ -300,9 +301,9 @@ static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
                                RCU_INIT_POINTER(*nodep, r);
                        } else { /* rr: RH, rl: RH+1 */
                                rlr = rcu_dereference_protected(rl->avl_right,
-                                       lockdep_is_held(&peers.lock));  /* rlr: LH or LH-1 */
+                                       lockdep_is_held(&base->lock));  /* rlr: LH or LH-1 */
                                rll = rcu_dereference_protected(rl->avl_left,
-                                       lockdep_is_held(&peers.lock));  /* rll: LH or LH-1 */
+                                       lockdep_is_held(&base->lock));  /* rll: LH or LH-1 */
                                RCU_INIT_POINTER(node->avl_right, rll); /* rll: LH or LH-1 */
                                RCU_INIT_POINTER(node->avl_left, l);    /* l: LH */
                                node->avl_height = lh + 1; /* node: LH+1 */
@@ -321,14 +322,14 @@ static void peer_avl_rebalance(struct inet_peer __rcu **stack[],
 }
 
 /* Called with local BH disabled and the pool lock held. */
-#define link_to_pool(n)                                                \
+#define link_to_pool(n, base)                                  \
 do {                                                           \
        n->avl_height = 1;                                      \
        n->avl_left = peer_avl_empty_rcu;                       \
        n->avl_right = peer_avl_empty_rcu;                      \
        /* lockless readers can catch us now */                 \
        rcu_assign_pointer(**--stackptr, n);                    \
-       peer_avl_rebalance(stack, stackptr);                    \
+       peer_avl_rebalance(stack, stackptr, base);              \
 } while (0)
 
 static void inetpeer_free_rcu(struct rcu_head *head)
@@ -337,13 +338,13 @@ static void inetpeer_free_rcu(struct rcu_head *head)
 }
 
 /* May be called with local BH enabled. */
-static void unlink_from_pool(struct inet_peer *p)
+static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base)
 {
        int do_free;
 
        do_free = 0;
 
-       spin_lock_bh(&peers.lock);
+       spin_lock_bh(&base->lock);
        /* Check the reference counter.  It was artificially incremented by 1
         * in cleanup() function to prevent sudden disappearing.  If we can
         * atomically (because of lockless readers) take this last reference,
@@ -353,7 +354,7 @@ static void unlink_from_pool(struct inet_peer *p)
        if (atomic_cmpxchg(&p->refcnt, 1, -1) == 1) {
                struct inet_peer __rcu **stack[PEER_MAXDEPTH];
                struct inet_peer __rcu ***stackptr, ***delp;
-               if (lookup(p->v4daddr, stack) != p)
+               if (lookup(p->v4daddr, stack, base) != p)
                        BUG();
                delp = stackptr - 1; /* *delp[0] == p */
                if (p->avl_left == peer_avl_empty_rcu) {
@@ -362,9 +363,9 @@ static void unlink_from_pool(struct inet_peer *p)
                } else {
                        /* look for a node to insert instead of p */
                        struct inet_peer *t;
-                       t = lookup_rightempty(p);
+                       t = lookup_rightempty(p, base);
                        BUG_ON(rcu_dereference_protected(*stackptr[-1],
-                                       lockdep_is_held(&peers.lock)) != t);
+                                       lockdep_is_held(&base->lock)) != t);
                        **--stackptr = t->avl_left;
                        /* t is removed, t->v4daddr > x->v4daddr for any
                         * x in p->avl_left subtree.
@@ -376,11 +377,11 @@ static void unlink_from_pool(struct inet_peer *p)
                        BUG_ON(delp[1] != &p->avl_left);
                        delp[1] = &t->avl_left; /* was &p->avl_left */
                }
-               peer_avl_rebalance(stack, stackptr);
-               peers.total--;
+               peer_avl_rebalance(stack, stackptr, base);
+               base->total--;
                do_free = 1;
        }
-       spin_unlock_bh(&peers.lock);
+       spin_unlock_bh(&base->lock);
 
        if (do_free)
                call_rcu_bh(&p->rcu, inetpeer_free_rcu);
@@ -395,6 +396,11 @@ static void unlink_from_pool(struct inet_peer *p)
                inet_putpeer(p);
 }
 
+static struct inet_peer_base *peer_to_base(struct inet_peer *p)
+{
+       return &v4_peers;
+}
+
 /* May be called with local BH enabled. */
 static int cleanup_once(unsigned long ttl)
 {
@@ -428,21 +434,27 @@ static int cleanup_once(unsigned long ttl)
                 * happen because of entry limits in route cache. */
                return -1;
 
-       unlink_from_pool(p);
+       unlink_from_pool(p, peer_to_base(p));
        return 0;
 }
 
+static struct inet_peer_base *family_to_base(int family)
+{
+       return &v4_peers;
+}
+
 /* Called with or without local BH being disabled. */
 struct inet_peer *inet_getpeer(__be32 daddr, int create)
 {
-       struct inet_peer *p;
        struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
+       struct inet_peer_base *base = family_to_base(AF_INET);
+       struct inet_peer *p;
 
        /* Look up for the address quickly, lockless.
         * Because of a concurrent writer, we might not find an existing entry.
         */
        rcu_read_lock_bh();
-       p = lookup_rcu_bh(daddr);
+       p = lookup_rcu_bh(daddr, base);
        rcu_read_unlock_bh();
 
        if (p) {
@@ -456,11 +468,11 @@ struct inet_peer *inet_getpeer(__be32 daddr, int create)
        /* retry an exact lookup, taking the lock before.
         * At least, nodes should be hot in our cache.
         */
-       spin_lock_bh(&peers.lock);
-       p = lookup(daddr, stack);
+       spin_lock_bh(&base->lock);
+       p = lookup(daddr, stack, base);
        if (p != peer_avl_empty) {
                atomic_inc(&p->refcnt);
-               spin_unlock_bh(&peers.lock);
+               spin_unlock_bh(&base->lock);
                /* Remove the entry from unused list if it was there. */
                unlink_from_unused(p);
                return p;
@@ -476,30 +488,36 @@ struct inet_peer *inet_getpeer(__be32 daddr, int create)
 
 
                /* Link the node. */
-               link_to_pool(p);
-               peers.total++;
+               link_to_pool(p, base);
+               base->total++;
        }
-       spin_unlock_bh(&peers.lock);
+       spin_unlock_bh(&base->lock);
 
-       if (peers.total >= inet_peer_threshold)
+       if (base->total >= inet_peer_threshold)
                /* Remove one less-recently-used entry. */
                cleanup_once(0);
 
        return p;
 }
 
+static int compute_total(void)
+{
+       return v4_peers.total;
+}
+
 /* Called with local BH disabled. */
 static void peer_check_expire(unsigned long dummy)
 {
        unsigned long now = jiffies;
-       int ttl;
+       int ttl, total;
 
-       if (peers.total >= inet_peer_threshold)
+       total = compute_total();
+       if (total >= inet_peer_threshold)
                ttl = inet_peer_minttl;
        else
                ttl = inet_peer_maxttl
                                - (inet_peer_maxttl - inet_peer_minttl) / HZ *
-                                       peers.total / inet_peer_threshold * HZ;
+                                       total / inet_peer_threshold * HZ;
        while (!cleanup_once(ttl)) {
                if (jiffies != now)
                        break;
@@ -508,13 +526,14 @@ static void peer_check_expire(unsigned long dummy)
        /* Trigger the timer after inet_peer_gc_mintime .. inet_peer_gc_maxtime
         * interval depending on the total number of entries (more entries,
         * less interval). */
-       if (peers.total >= inet_peer_threshold)
+       total = compute_total();
+       if (total >= inet_peer_threshold)
                peer_periodic_timer.expires = jiffies + inet_peer_gc_mintime;
        else
                peer_periodic_timer.expires = jiffies
                        + inet_peer_gc_maxtime
                        - (inet_peer_gc_maxtime - inet_peer_gc_mintime) / HZ *
-                               peers.total / inet_peer_threshold * HZ;
+                               total / inet_peer_threshold * HZ;
        add_timer(&peer_periodic_timer);
 }