1 files changed, 193 insertions, 237 deletions

diff --git a/net/ipv4/inetpeer.c b/net/ipv4/inetpeer.c
index dd1b20eca1a..bd5f5928167 100644
--- a/net/ipv4/inetpeer.c
+++ b/net/ipv4/inetpeer.c
@@ -17,27 +17,16 @@
 #include <linux/kernel.h>
 #include <linux/mm.h>
 #include <linux/net.h>
+#include <linux/workqueue.h>
 #include <net/ip.h>
 #include <net/inetpeer.h>
+#include <net/secure_seq.h>
 
 /*
  *  Theory of operations.
  *  We keep one entry for each peer IP address.  The nodes contains long-living
  *  information about the peer which doesn't depend on routes.
- *  At this moment this information consists only of ID field for the next
- *  outgoing IP packet.  This field is incremented with each packet as encoded
- *  in inet_getid() function (include/net/inetpeer.h).
- *  At the moment of writing this notes identifier of IP packets is generated
- *  to be unpredictable using this code only for packets subjected
- *  (actually or potentially) to defragmentation.  I.e. DF packets less than
- *  PMTU in size uses a constant ID and do not use this code (see
- *  ip_select_ident() in include/net/ip.h).
  *
- *  Route cache entries hold references to our nodes.
- *  New cache entries get references via lookup by destination IP address in
- *  the avl tree.  The reference is grabbed only when it's needed i.e. only
- *  when we try to output IP packet which needs an unpredictable ID (see
- *  __ip_select_ident() in net/ipv4/route.c).
  *  Nodes are removed only when reference counter goes to 0.
  *  When it's happened the node may be removed when a sufficient amount of
  *  time has been passed since its last use.  The less-recently-used entry can
@@ -54,21 +43,21 @@
  *  1.  Nodes may appear in the tree only with the pool lock held.
  *  2.  Nodes may disappear from the tree only with the pool lock held
  *      AND reference count being 0.
- *  3.  Nodes appears and disappears from unused node list only under
- *      "inet_peer_unused_lock".
- *  4.  Global variable peer_total is modified under the pool lock.
- *  5.  struct inet_peer fields modification:
+ *  3.  Global variable peer_total is modified under the pool lock.
+ *  4.  struct inet_peer fields modification:
  *		avl_left, avl_right, avl_parent, avl_height: pool lock
- *		unused: unused node list lock
  *		refcnt: atomically against modifications on other CPU;
  *		   usually under some other lock to prevent node disappearing
- *		dtime: unused node list lock
  *		daddr: unchangeable
- *		ip_id_count: atomic value (no lock needed)
  */
 
 static struct kmem_cache *peer_cachep __read_mostly;
 
+static LIST_HEAD(gc_list);
+static const int gc_delay = 60 * HZ;
+static struct delayed_work gc_work;
+static DEFINE_SPINLOCK(gc_lock);
+
 #define node_height(x) x->avl_height
 
 #define peer_avl_empty ((struct inet_peer *)&peer_fake_node)
@@ -79,23 +68,32 @@ static const struct inet_peer peer_fake_node = {
 	.avl_height	= 0
 };
 
-struct inet_peer_base {
-	struct inet_peer __rcu *root;
-	seqlock_t	lock;
-	int		total;
-};
+void inet_peer_base_init(struct inet_peer_base *bp)
+{
+	bp->root = peer_avl_empty_rcu;
+	seqlock_init(&bp->lock);
+	bp->flush_seq = ~0U;
+	bp->total = 0;
+}
+EXPORT_SYMBOL_GPL(inet_peer_base_init);
 
-static struct inet_peer_base v4_peers = {
-	.root		= peer_avl_empty_rcu,
-	.lock		= __SEQLOCK_UNLOCKED(v4_peers.lock),
-	.total		= 0,
-};
+static atomic_t v4_seq = ATOMIC_INIT(0);
+static atomic_t v6_seq = ATOMIC_INIT(0);
 
-static struct inet_peer_base v6_peers = {
-	.root		= peer_avl_empty_rcu,
-	.lock		= __SEQLOCK_UNLOCKED(v6_peers.lock),
-	.total		= 0,
-};
+static atomic_t *inetpeer_seq_ptr(int family)
+{
+	return (family == AF_INET ? &v4_seq : &v6_seq);
+}
+
+static inline void flush_check(struct inet_peer_base *base, int family)
+{
+	atomic_t *fp = inetpeer_seq_ptr(family);
+
+	if (unlikely(base->flush_seq != atomic_read(fp))) {
+		inetpeer_invalidate_tree(base);
+		base->flush_seq = atomic_read(fp);
+	}
+}
 
 #define PEER_MAXDEPTH 40 /* sufficient for about 2^27 nodes */
 
@@ -104,20 +102,53 @@ int inet_peer_threshold __read_mostly = 65536 + 128;	/* start to throw entries m
 					 * aggressively at this stage */
 int inet_peer_minttl __read_mostly = 120 * HZ;	/* TTL under high load: 120 sec */
 int inet_peer_maxttl __read_mostly = 10 * 60 * HZ;	/* usual time to live: 10 min */
-int inet_peer_gc_mintime __read_mostly = 10 * HZ;
-int inet_peer_gc_maxtime __read_mostly = 120 * HZ;
-
-static struct {
-	struct list_head	list;
-	spinlock_t		lock;
-} unused_peers = {
-	.list			= LIST_HEAD_INIT(unused_peers.list),
-	.lock			= __SPIN_LOCK_UNLOCKED(unused_peers.lock),
-};
 
-static void peer_check_expire(unsigned long dummy);
-static DEFINE_TIMER(peer_periodic_timer, peer_check_expire, 0, 0);
+static void inetpeer_gc_worker(struct work_struct *work)
+{
+	struct inet_peer *p, *n, *c;
+	struct list_head list;
+
+	spin_lock_bh(&gc_lock);
+	list_replace_init(&gc_list, &list);
+	spin_unlock_bh(&gc_lock);
+
+	if (list_empty(&list))
+		return;
+
+	list_for_each_entry_safe(p, n, &list, gc_list) {
+
+		if (need_resched())
+			cond_resched();
+
+		c = rcu_dereference_protected(p->avl_left, 1);
+		if (c != peer_avl_empty) {
+			list_add_tail(&c->gc_list, &list);
+			p->avl_left = peer_avl_empty_rcu;
+		}
+
+		c = rcu_dereference_protected(p->avl_right, 1);
+		if (c != peer_avl_empty) {
+			list_add_tail(&c->gc_list, &list);
+			p->avl_right = peer_avl_empty_rcu;
+		}
+
+		n = list_entry(p->gc_list.next, struct inet_peer, gc_list);
+
+		if (!atomic_read(&p->refcnt)) {
+			list_del(&p->gc_list);
+			kmem_cache_free(peer_cachep, p);
+		}
+	}
 
+	if (list_empty(&list))
+		return;
+
+	spin_lock_bh(&gc_lock);
+	list_splice(&list, &gc_list);
+	spin_unlock_bh(&gc_lock);
+
+	schedule_delayed_work(&gc_work, gc_delay);
+}
 
 /* Called from ip_output.c:ip_init  */
 void __init inet_initpeers(void)
@@ -142,23 +173,7 @@ void __init inet_initpeers(void)
 			0, SLAB_HWCACHE_ALIGN | SLAB_PANIC,
 			NULL);
 
-	/* All the timers, started at system startup tend
-	   to synchronize. Perturb it a bit.
-	 */
-	peer_periodic_timer.expires = jiffies
-		+ net_random() % inet_peer_gc_maxtime
-		+ inet_peer_gc_maxtime;
-	add_timer(&peer_periodic_timer);
-}
-
-/* Called with or without local BH being disabled. */
-static void unlink_from_unused(struct inet_peer *p)
-{
-	if (!list_empty(&p->unused)) {
-		spin_lock_bh(&unused_peers.lock);
-		list_del_init(&p->unused);
-		spin_unlock_bh(&unused_peers.lock);
-	}
+	INIT_DEFERRABLE_WORK(&gc_work, inetpeer_gc_worker);
 }
 
 static int addr_compare(const struct inetpeer_addr *a,
@@ -169,7 +184,7 @@ static int addr_compare(const struct inetpeer_addr *a,
 	for (i = 0; i < n; i++) {
 		if (a->addr.a6[i] == b->addr.a6[i])
 			continue;
-		if (a->addr.a6[i] < b->addr.a6[i])
+		if ((__force u32)a->addr.a6[i] < (__force u32)b->addr.a6[i])
 			return -1;
 		return 1;
 	}
@@ -191,7 +206,7 @@ static int addr_compare(const struct inetpeer_addr *a,
 	stackptr = _stack;					\
 	*stackptr++ = &_base->root;				\
 	for (u = rcu_deref_locked(_base->root, _base);		\
-	     u != peer_avl_empty; ) {				\
+	     u != peer_avl_empty;) {				\
 		int cmp = addr_compare(_daddr, &u->daddr);	\
 		if (cmp == 0)					\
 			break;					\
@@ -222,11 +237,9 @@ static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
 		int cmp = addr_compare(daddr, &u->daddr);
 		if (cmp == 0) {
 			/* Before taking a reference, check if this entry was
-			 * deleted, unlink_from_pool() sets refcnt=-1 to make
-			 * distinction between an unused entry (refcnt=0) and
-			 * a freed one.
+			 * deleted (refcnt=-1)
 			 */
-			if (unlikely(!atomic_add_unless(&u->refcnt, 1, -1)))
+			if (!atomic_add_unless(&u->refcnt, 1, -1))
 				u = NULL;
 			return u;
 		}
@@ -248,7 +261,7 @@ static struct inet_peer *lookup_rcu(const struct inetpeer_addr *daddr,
 	*stackptr++ = &start->avl_left;				\
 	v = &start->avl_left;					\
 	for (u = rcu_deref_locked(*v, base);			\
-	     u->avl_right != peer_avl_empty_rcu; ) {		\
+	     u->avl_right != peer_avl_empty_rcu;) {		\
 		v = &u->avl_right;				\
 		*stackptr++ = v;				\
 		u = rcu_deref_locked(*v, base);			\
@@ -353,121 +366,87 @@ static void inetpeer_free_rcu(struct rcu_head *head)
 	kmem_cache_free(peer_cachep, container_of(head, struct inet_peer, rcu));
 }
 
-/* May be called with local BH enabled. */
-static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base)
+static void unlink_from_pool(struct inet_peer *p, struct inet_peer_base *base,
+			     struct inet_peer __rcu **stack[PEER_MAXDEPTH])
 {
-	int do_free;
-
-	do_free = 0;
-
-	write_seqlock_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,
-	 * it's safe to remove the node and free it later.
-	 * We use refcnt=-1 to alert lockless readers this entry is deleted.
-	 */
-	if (atomic_cmpxchg(&p->refcnt, 1, -1) == 1) {
-		struct inet_peer __rcu **stack[PEER_MAXDEPTH];
-		struct inet_peer __rcu ***stackptr, ***delp;
-		if (lookup(&p->daddr, stack, base) != p)
-			BUG();
-		delp = stackptr - 1; /* *delp[0] == p */
-		if (p->avl_left == peer_avl_empty_rcu) {
-			*delp[0] = p->avl_right;
-			--stackptr;
-		} else {
-			/* look for a node to insert instead of p */
-			struct inet_peer *t;
-			t = lookup_rightempty(p, base);
-			BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
-			**--stackptr = t->avl_left;
-			/* t is removed, t->daddr > x->daddr for any
-			 * x in p->avl_left subtree.
-			 * Put t in the old place of p. */
-			RCU_INIT_POINTER(*delp[0], t);
-			t->avl_left = p->avl_left;
-			t->avl_right = p->avl_right;
-			t->avl_height = p->avl_height;
-			BUG_ON(delp[1] != &p->avl_left);
-			delp[1] = &t->avl_left; /* was &p->avl_left */
-		}
-		peer_avl_rebalance(stack, stackptr, base);
-		base->total--;
-		do_free = 1;
+	struct inet_peer __rcu ***stackptr, ***delp;
+
+	if (lookup(&p->daddr, stack, base) != p)
+		BUG();
+	delp = stackptr - 1; /* *delp[0] == p */
+	if (p->avl_left == peer_avl_empty_rcu) {
+		*delp[0] = p->avl_right;
+		--stackptr;
+	} else {
+		/* look for a node to insert instead of p */
+		struct inet_peer *t;
+		t = lookup_rightempty(p, base);
+		BUG_ON(rcu_deref_locked(*stackptr[-1], base) != t);
+		**--stackptr = t->avl_left;
+		/* t is removed, t->daddr > x->daddr for any
+		 * x in p->avl_left subtree.
+		 * Put t in the old place of p. */
+		RCU_INIT_POINTER(*delp[0], t);
+		t->avl_left = p->avl_left;
+		t->avl_right = p->avl_right;
+		t->avl_height = p->avl_height;
+		BUG_ON(delp[1] != &p->avl_left);
+		delp[1] = &t->avl_left; /* was &p->avl_left */
 	}
-	write_sequnlock_bh(&base->lock);
-
-	if (do_free)
-		call_rcu(&p->rcu, inetpeer_free_rcu);
-	else
-		/* The node is used again.  Decrease the reference counter
-		 * back.  The loop "cleanup -> unlink_from_unused
-		 *   -> unlink_from_pool -> putpeer -> link_to_unused
-		 *   -> cleanup (for the same node)"
-		 * doesn't really exist because the entry will have a
-		 * recent deletion time and will not be cleaned again soon.
-		 */
-		inet_putpeer(p);
+	peer_avl_rebalance(stack, stackptr, base);
+	base->total--;
+	call_rcu(&p->rcu, inetpeer_free_rcu);
 }
 
-static struct inet_peer_base *family_to_base(int family)
+/* perform garbage collect on all items stacked during a lookup */
+static int inet_peer_gc(struct inet_peer_base *base,
+			struct inet_peer __rcu **stack[PEER_MAXDEPTH],
+			struct inet_peer __rcu ***stackptr)
 {
-	return (family == AF_INET ? &v4_peers : &v6_peers);
-}
-
-static struct inet_peer_base *peer_to_base(struct inet_peer *p)
-{
-	return family_to_base(p->daddr.family);
-}
-
-/* May be called with local BH enabled. */
-static int cleanup_once(unsigned long ttl)
-{
-	struct inet_peer *p = NULL;
-
-	/* Remove the first entry from the list of unused nodes. */
-	spin_lock_bh(&unused_peers.lock);
-	if (!list_empty(&unused_peers.list)) {
-		__u32 delta;
-
-		p = list_first_entry(&unused_peers.list, struct inet_peer, unused);
-		delta = (__u32)jiffies - p->dtime;
+	struct inet_peer *p, *gchead = NULL;
+	__u32 delta, ttl;
+	int cnt = 0;
 
-		if (delta < ttl) {
-			/* Do not prune fresh entries. */
-			spin_unlock_bh(&unused_peers.lock);
-			return -1;
+	if (base->total >= inet_peer_threshold)
+		ttl = 0; /* be aggressive */
+	else
+		ttl = inet_peer_maxttl
+				- (inet_peer_maxttl - inet_peer_minttl) / HZ *
+					base->total / inet_peer_threshold * HZ;
+	stackptr--; /* last stack slot is peer_avl_empty */
+	while (stackptr > stack) {
+		stackptr--;
+		p = rcu_deref_locked(**stackptr, base);
+		if (atomic_read(&p->refcnt) == 0) {
+			smp_rmb();
+			delta = (__u32)jiffies - p->dtime;
+			if (delta >= ttl &&
+			    atomic_cmpxchg(&p->refcnt, 0, -1) == 0) {
+				p->gc_next = gchead;
+				gchead = p;
+			}
 		}
-
-		list_del_init(&p->unused);
-
-		/* Grab an extra reference to prevent node disappearing
-		 * before unlink_from_pool() call. */
-		atomic_inc(&p->refcnt);
 	}
-	spin_unlock_bh(&unused_peers.lock);
-
-	if (p == NULL)
-		/* It means that the total number of USED entries has
-		 * grown over inet_peer_threshold.  It shouldn't really
-		 * happen because of entry limits in route cache. */
-		return -1;
-
-	unlink_from_pool(p, peer_to_base(p));
-	return 0;
+	while ((p = gchead) != NULL) {
+		gchead = p->gc_next;
+		cnt++;
+		unlink_from_pool(p, base, stack);
+	}
+	return cnt;
 }
 
-/* Called with or without local BH being disabled. */
-struct inet_peer *inet_getpeer(struct inetpeer_addr *daddr, int create)
+struct inet_peer *inet_getpeer(struct inet_peer_base *base,
+			       const struct inetpeer_addr *daddr,
+			       int create)
 {
 	struct inet_peer __rcu **stack[PEER_MAXDEPTH], ***stackptr;
-	struct inet_peer_base *base = family_to_base(daddr->family);
 	struct inet_peer *p;
 	unsigned int sequence;
-	int invalidated;
+	int invalidated, gccnt = 0;
+
+	flush_check(base, daddr->family);
 
-	/* Look up for the address quickly, lockless.
+	/* Attempt a lockless lookup first.
 	 * Because of a concurrent writer, we might not find an existing entry.
 	 */
 	rcu_read_lock();
@@ -476,13 +455,8 @@ struct inet_peer *inet_getpeer(struct inetpeer_addr *daddr, int create)
 	invalidated = read_seqretry(&base->lock, sequence);
 	rcu_read_unlock();
 
-	if (p) {
-		/* The existing node has been found.
-		 * Remove the entry from unused list if it was there.
-		 */
-		unlink_from_unused(p);
+	if (p)
 		return p;
-	}
 
 	/* If no writer did a change during our lookup, we can return early. */
 	if (!create && !invalidated)
@@ -492,29 +466,30 @@ struct inet_peer *inet_getpeer(struct inetpeer_addr *daddr, int create)
 	 * At least, nodes should be hot in our cache.
 	 */
 	write_seqlock_bh(&base->lock);
+relookup:
 	p = lookup(daddr, stack, base);
 	if (p != peer_avl_empty) {
 		atomic_inc(&p->refcnt);
 		write_sequnlock_bh(&base->lock);
-		/* Remove the entry from unused list if it was there. */
-		unlink_from_unused(p);
 		return p;
 	}
+	if (!gccnt) {
+		gccnt = inet_peer_gc(base, stack, stackptr);
+		if (gccnt && create)
+			goto relookup;
+	}
 	p = create ? kmem_cache_alloc(peer_cachep, GFP_ATOMIC) : NULL;
 	if (p) {
 		p->daddr = *daddr;
 		atomic_set(&p->refcnt, 1);
 		atomic_set(&p->rid, 0);
-		atomic_set(&p->ip_id_count, secure_ip_id(daddr->addr.a4));
-		p->tcp_ts_stamp = 0;
 		p->metrics[RTAX_LOCK-1] = INETPEER_METRICS_NEW;
 		p->rate_tokens = 0;
-		p->rate_last = 0;
-		p->pmtu_expires = 0;
-		p->pmtu_orig = 0;
-		memset(&p->redirect_learned, 0, sizeof(p->redirect_learned));
-		INIT_LIST_HEAD(&p->unused);
-
+		/* 60*HZ is arbitrary, but chosen enough high so that the first
+		 * calculation of tokens is at its maximum.
+		 */
+		p->rate_last = jiffies - 60*HZ;
+		INIT_LIST_HEAD(&p->gc_list);
 
 		/* Link the node. */
 		link_to_pool(p, base);
@@ -522,62 +497,15 @@ struct inet_peer *inet_getpeer(struct inetpeer_addr *daddr, int create)
 	}
 	write_sequnlock_bh(&base->lock);
 
-	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 + v6_peers.total;
-}
 EXPORT_SYMBOL_GPL(inet_getpeer);
 
-/* Called with local BH disabled. */
-static void peer_check_expire(unsigned long dummy)
-{
-	unsigned long now = jiffies;
-	int ttl, total;
-
-	total = compute_total();
-	if (total >= inet_peer_threshold)
-		ttl = inet_peer_minttl;
-	else
-		ttl = inet_peer_maxttl
-				- (inet_peer_maxttl - inet_peer_minttl) / HZ *
-					total / inet_peer_threshold * HZ;
-	while (!cleanup_once(ttl)) {
-		if (jiffies != now)
-			break;
-	}
-
-	/* Trigger the timer after inet_peer_gc_mintime .. inet_peer_gc_maxtime
-	 * interval depending on the total number of entries (more entries,
-	 * less interval). */
-	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 *
-				total / inet_peer_threshold * HZ;
-	add_timer(&peer_periodic_timer);
-}
-
 void inet_putpeer(struct inet_peer *p)
 {
-	local_bh_disable();
-
-	if (atomic_dec_and_lock(&p->refcnt, &unused_peers.lock)) {
-		list_add_tail(&p->unused, &unused_peers.list);
-		p->dtime = (__u32)jiffies;
-		spin_unlock(&unused_peers.lock);
-	}
-
-	local_bh_enable();
+	p->dtime = (__u32)jiffies;
+	smp_mb__before_atomic();
+	atomic_dec(&p->refcnt);
 }
 EXPORT_SYMBOL_GPL(inet_putpeer);
 
@@ -621,3 +549,31 @@ bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
 	return rc;
 }
 EXPORT_SYMBOL(inet_peer_xrlim_allow);
+
+static void inetpeer_inval_rcu(struct rcu_head *head)
+{
+	struct inet_peer *p = container_of(head, struct inet_peer, gc_rcu);
+
+	spin_lock_bh(&gc_lock);
+	list_add_tail(&p->gc_list, &gc_list);
+	spin_unlock_bh(&gc_lock);
+
+	schedule_delayed_work(&gc_work, gc_delay);
+}
+
+void inetpeer_invalidate_tree(struct inet_peer_base *base)
+{
+	struct inet_peer *root;
+
+	write_seqlock_bh(&base->lock);
+
+	root = rcu_deref_locked(base->root, base);
+	if (root != peer_avl_empty) {
+		base->root = peer_avl_empty_rcu;
+		base->total = 0;
+		call_rcu(&root->gc_rcu, inetpeer_inval_rcu);
+	}
+
+	write_sequnlock_bh(&base->lock);
+}
+EXPORT_SYMBOL(inetpeer_invalidate_tree);