[XFRM]: Speed up xfrm_policy and xfrm_state walking

Change xfrm_policy and xfrm_state walking algorithm from O(n^2) to O(n).
This is achieved adding the entries to one more list which is used
solely for walking the entries.

This also fixes some races where the dump can have duplicate or missing
entries when the SPD/SADB is modified during an ongoing dump.

Dumping SADB with 20000 entries using "time ip xfrm state" the sys
time dropped from 1.012s to 0.080s.

Signed-off-by: Timo Teras <timo.teras@iki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 125 insertions, 78 deletions

diff --git a/net/xfrm/xfrm_policy.c b/net/xfrm/xfrm_policy.c
index 9fc4c315f6cd..bae94a8031a2 100644
--- a/net/xfrm/xfrm_policy.c
+++ b/net/xfrm/xfrm_policy.c
@@ -46,6 +46,7 @@ EXPORT_SYMBOL(xfrm_cfg_mutex);
 
 static DEFINE_RWLOCK(xfrm_policy_lock);
 
+static struct list_head xfrm_policy_bytype[XFRM_POLICY_TYPE_MAX];
 unsigned int xfrm_policy_count[XFRM_POLICY_MAX*2];
 EXPORT_SYMBOL(xfrm_policy_count);
 
@@ -208,6 +209,7 @@ struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
 	policy = kzalloc(sizeof(struct xfrm_policy), gfp);
 
 	if (policy) {
+		INIT_LIST_HEAD(&policy->bytype);
 		INIT_HLIST_NODE(&policy->bydst);
 		INIT_HLIST_NODE(&policy->byidx);
 		rwlock_init(&policy->lock);
@@ -230,6 +232,10 @@ void xfrm_policy_destroy(struct xfrm_policy *policy)
 	if (del_timer(&policy->timer))
 		BUG();
 
+	write_lock_bh(&xfrm_policy_lock);
+	list_del(&policy->bytype);
+	write_unlock_bh(&xfrm_policy_lock);
+
 	security_xfrm_policy_free(policy);
 	kfree(policy);
 }
@@ -584,6 +590,7 @@ int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
 	policy->curlft.use_time = 0;
 	if (!mod_timer(&policy->timer, jiffies + HZ))
 		xfrm_pol_hold(policy);
+	list_add_tail(&policy->bytype, &xfrm_policy_bytype[policy->type]);
 	write_unlock_bh(&xfrm_policy_lock);
 
 	if (delpol)
@@ -822,57 +829,60 @@ out:
 }
 EXPORT_SYMBOL(xfrm_policy_flush);
 
-int xfrm_policy_walk(u8 type, int (*func)(struct xfrm_policy *, int, int, void*),
+int xfrm_policy_walk(struct xfrm_policy_walk *walk,
+		     int (*func)(struct xfrm_policy *, int, int, void*),
 		     void *data)
 {
-	struct xfrm_policy *pol, *last = NULL;
-	struct hlist_node *entry;
-	int dir, last_dir = 0, count, error;
+	struct xfrm_policy *old, *pol, *last = NULL;
+	int error = 0;
+
+	if (walk->type >= XFRM_POLICY_TYPE_MAX &&
+	    walk->type != XFRM_POLICY_TYPE_ANY)
+		return -EINVAL;
 
+	if (walk->policy == NULL && walk->count != 0)
+		return 0;
+
+	old = pol = walk->policy;
+	walk->policy = NULL;
 	read_lock_bh(&xfrm_policy_lock);
-	count = 0;
 
-	for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
-		struct hlist_head *table = xfrm_policy_bydst[dir].table;
-		int i;
+	for (; walk->cur_type < XFRM_POLICY_TYPE_MAX; walk->cur_type++) {
+		if (walk->type != walk->cur_type &&
+		    walk->type != XFRM_POLICY_TYPE_ANY)
+			continue;
 
-		hlist_for_each_entry(pol, entry,
-				     &xfrm_policy_inexact[dir], bydst) {
-			if (pol->type != type)
+		if (pol == NULL) {
+			pol = list_first_entry(&xfrm_policy_bytype[walk->cur_type],
+					       struct xfrm_policy, bytype);
+		}
+		list_for_each_entry_from(pol, &xfrm_policy_bytype[walk->cur_type], bytype) {
+			if (pol->dead)
 				continue;
 			if (last) {
-				error = func(last, last_dir % XFRM_POLICY_MAX,
-					     count, data);
-				if (error)
+				error = func(last, xfrm_policy_id2dir(last->index),
+					     walk->count, data);
+				if (error) {
+					xfrm_pol_hold(last);
+					walk->policy = last;
 					goto out;
-			}
-			last = pol;
-			last_dir = dir;
-			count++;
-		}
-		for (i = xfrm_policy_bydst[dir].hmask; i >= 0; i--) {
-			hlist_for_each_entry(pol, entry, table + i, bydst) {
-				if (pol->type != type)
-					continue;
-				if (last) {
-					error = func(last, last_dir % XFRM_POLICY_MAX,
-						     count, data);
-					if (error)
-						goto out;
 				}
-				last = pol;
-				last_dir = dir;
-				count++;
 			}
+			last = pol;
+			walk->count++;
 		}
+		pol = NULL;
 	}
-	if (count == 0) {
+	if (walk->count == 0) {
 		error = -ENOENT;
 		goto out;
 	}
-	error = func(last, last_dir % XFRM_POLICY_MAX, 0, data);
+	if (last)
+		error = func(last, xfrm_policy_id2dir(last->index), 0, data);
 out:
 	read_unlock_bh(&xfrm_policy_lock);
+	if (old != NULL)
+		xfrm_pol_put(old);
 	return error;
 }
 EXPORT_SYMBOL(xfrm_policy_walk);
@@ -2365,6 +2375,9 @@ static void __init xfrm_policy_init(void)
 			panic("XFRM: failed to allocate bydst hash\n");
 	}
 
+	for (dir = 0; dir < XFRM_POLICY_TYPE_MAX; dir++)
+		INIT_LIST_HEAD(&xfrm_policy_bytype[dir]);
+
 	INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task);
 	register_netdevice_notifier(&xfrm_dev_notifier);
 }
diff --git a/net/xfrm/xfrm_state.c b/net/xfrm/xfrm_state.c
index 7ba65e82941c..9880b792e6a5 100644
--- a/net/xfrm/xfrm_state.c
+++ b/net/xfrm/xfrm_state.c
@@ -50,6 +50,7 @@ static DEFINE_SPINLOCK(xfrm_state_lock);
  * Main use is finding SA after policy selected tunnel or transport mode.
  * Also, it can be used by ah/esp icmp error handler to find offending SA.
  */
+static LIST_HEAD(xfrm_state_all);
 static struct hlist_head *xfrm_state_bydst __read_mostly;
 static struct hlist_head *xfrm_state_bysrc __read_mostly;
 static struct hlist_head *xfrm_state_byspi __read_mostly;
@@ -510,6 +511,7 @@ struct xfrm_state *xfrm_state_alloc(void)
 	if (x) {
 		atomic_set(&x->refcnt, 1);
 		atomic_set(&x->tunnel_users, 0);
+		INIT_LIST_HEAD(&x->all);
 		INIT_HLIST_NODE(&x->bydst);
 		INIT_HLIST_NODE(&x->bysrc);
 		INIT_HLIST_NODE(&x->byspi);
@@ -533,6 +535,10 @@ void __xfrm_state_destroy(struct xfrm_state *x)
 {
 	BUG_TRAP(x->km.state == XFRM_STATE_DEAD);
 
+	spin_lock_bh(&xfrm_state_lock);
+	list_del(&x->all);
+	spin_unlock_bh(&xfrm_state_lock);
+
 	spin_lock_bh(&xfrm_state_gc_lock);
 	hlist_add_head(&x->bydst, &xfrm_state_gc_list);
 	spin_unlock_bh(&xfrm_state_gc_lock);
@@ -909,6 +915,8 @@ static void __xfrm_state_insert(struct xfrm_state *x)
 
 	x->genid = ++xfrm_state_genid;
 
+	list_add_tail(&x->all, &xfrm_state_all);
+
 	h = xfrm_dst_hash(&x->id.daddr, &x->props.saddr,
 			  x->props.reqid, x->props.family);
 	hlist_add_head(&x->bydst, xfrm_state_bydst+h);
@@ -1518,36 +1526,47 @@ unlock:
 }
 EXPORT_SYMBOL(xfrm_alloc_spi);
 
-int xfrm_state_walk(u8 proto, int (*func)(struct xfrm_state *, int, void*),
+int xfrm_state_walk(struct xfrm_state_walk *walk,
+		    int (*func)(struct xfrm_state *, int, void*),
 		    void *data)
 {
-	int i;
-	struct xfrm_state *x, *last = NULL;
-	struct hlist_node *entry;
-	int count = 0;
+	struct xfrm_state *old, *x, *last = NULL;
 	int err = 0;
 
+	if (walk->state == NULL && walk->count != 0)
+		return 0;
+
+	old = x = walk->state;
+	walk->state = NULL;
 	spin_lock_bh(&xfrm_state_lock);
-	for (i = 0; i <= xfrm_state_hmask; i++) {
-		hlist_for_each_entry(x, entry, xfrm_state_bydst+i, bydst) {
-			if (!xfrm_id_proto_match(x->id.proto, proto))
-				continue;
-			if (last) {
-				err = func(last, count, data);
-				if (err)
-					goto out;
+	if (x == NULL)
+		x = list_first_entry(&xfrm_state_all, struct xfrm_state, all);
+	list_for_each_entry_from(x, &xfrm_state_all, all) {
+		if (x->km.state == XFRM_STATE_DEAD)
+			continue;
+		if (!xfrm_id_proto_match(x->id.proto, walk->proto))
+			continue;
+		if (last) {
+			err = func(last, walk->count, data);
+			if (err) {
+				xfrm_state_hold(last);
+				walk->state = last;
+				goto out;
 			}
-			last = x;
-			count++;
 		}
+		last = x;
+		walk->count++;
 	}
-	if (count == 0) {
+	if (walk->count == 0) {
 		err = -ENOENT;
 		goto out;
 	}
-	err = func(last, 0, data);
+	if (last)
+		err = func(last, 0, data);
 out:
 	spin_unlock_bh(&xfrm_state_lock);
+	if (old != NULL)
+		xfrm_state_put(old);
 	return err;
 }
 EXPORT_SYMBOL(xfrm_state_walk);
diff --git a/net/xfrm/xfrm_user.c b/net/xfrm/xfrm_user.c
index f971ca5645f8..f5fd5b3147cc 100644
--- a/net/xfrm/xfrm_user.c
+++ b/net/xfrm/xfrm_user.c
@@ -532,8 +532,6 @@ struct xfrm_dump_info {
 	struct sk_buff *out_skb;
 	u32 nlmsg_seq;
 	u16 nlmsg_flags;
-	int start_idx;
-	int this_idx;
 };
 
 static int copy_sec_ctx(struct xfrm_sec_ctx *s, struct sk_buff *skb)
@@ -600,9 +598,6 @@ static int dump_one_state(struct xfrm_state *x, int count, void *ptr)
 	struct nlmsghdr *nlh;
 	int err;
 
-	if (sp->this_idx < sp->start_idx)
-		goto out;
-
 	nlh = nlmsg_put(skb, NETLINK_CB(in_skb).pid, sp->nlmsg_seq,
 			XFRM_MSG_NEWSA, sizeof(*p), sp->nlmsg_flags);
 	if (nlh == NULL)
@@ -615,8 +610,6 @@ static int dump_one_state(struct xfrm_state *x, int count, void *ptr)
 		goto nla_put_failure;
 
 	nlmsg_end(skb, nlh);
-out:
-	sp->this_idx++;
 	return 0;
 
 nla_put_failure:
@@ -624,18 +617,32 @@ nla_put_failure:
 	return err;
 }
 
+static int xfrm_dump_sa_done(struct netlink_callback *cb)
+{
+	struct xfrm_state_walk *walk = (struct xfrm_state_walk *) &cb->args[1];
+	xfrm_state_walk_done(walk);
+	return 0;
+}
+
 static int xfrm_dump_sa(struct sk_buff *skb, struct netlink_callback *cb)
 {
+	struct xfrm_state_walk *walk = (struct xfrm_state_walk *) &cb->args[1];
 	struct xfrm_dump_info info;
 
+	BUILD_BUG_ON(sizeof(struct xfrm_state_walk) >
+		     sizeof(cb->args) - sizeof(cb->args[0]));
+
 	info.in_skb = cb->skb;
 	info.out_skb = skb;
 	info.nlmsg_seq = cb->nlh->nlmsg_seq;
 	info.nlmsg_flags = NLM_F_MULTI;
-	info.this_idx = 0;
-	info.start_idx = cb->args[0];
-	(void) xfrm_state_walk(0, dump_one_state, &info);
-	cb->args[0] = info.this_idx;
+
+	if (!cb->args[0]) {
+		cb->args[0] = 1;
+		xfrm_state_walk_init(walk, 0);
+	}
+
+	(void) xfrm_state_walk(walk, dump_one_state, &info);
 
 	return skb->len;
 }
@@ -654,7 +661,6 @@ static struct sk_buff *xfrm_state_netlink(struct sk_buff *in_skb,
 	info.out_skb = skb;
 	info.nlmsg_seq = seq;
 	info.nlmsg_flags = 0;
-	info.this_idx = info.start_idx = 0;
 
 	if (dump_one_state(x, 0, &info)) {
 		kfree_skb(skb);
@@ -1232,9 +1238,6 @@ static int dump_one_policy(struct xfrm_policy *xp, int dir, int count, void *ptr
 	struct sk_buff *skb = sp->out_skb;
 	struct nlmsghdr *nlh;
 
-	if (sp->this_idx < sp->start_idx)
-		goto out;
-
 	nlh = nlmsg_put(skb, NETLINK_CB(in_skb).pid, sp->nlmsg_seq,
 			XFRM_MSG_NEWPOLICY, sizeof(*p), sp->nlmsg_flags);
 	if (nlh == NULL)
@@ -1250,8 +1253,6 @@ static int dump_one_policy(struct xfrm_policy *xp, int dir, int count, void *ptr
 		goto nlmsg_failure;
 
 	nlmsg_end(skb, nlh);
-out:
-	sp->this_idx++;
 	return 0;
 
 nlmsg_failure:
@@ -1259,21 +1260,33 @@ nlmsg_failure:
 	return -EMSGSIZE;
 }
 
+static int xfrm_dump_policy_done(struct netlink_callback *cb)
+{
+	struct xfrm_policy_walk *walk = (struct xfrm_policy_walk *) &cb->args[1];
+
+	xfrm_policy_walk_done(walk);
+	return 0;
+}
+
 static int xfrm_dump_policy(struct sk_buff *skb, struct netlink_callback *cb)
 {
+	struct xfrm_policy_walk *walk = (struct xfrm_policy_walk *) &cb->args[1];
 	struct xfrm_dump_info info;
 
+	BUILD_BUG_ON(sizeof(struct xfrm_policy_walk) >
+		     sizeof(cb->args) - sizeof(cb->args[0]));
+
 	info.in_skb = cb->skb;
 	info.out_skb = skb;
 	info.nlmsg_seq = cb->nlh->nlmsg_seq;
 	info.nlmsg_flags = NLM_F_MULTI;
-	info.this_idx = 0;
-	info.start_idx = cb->args[0];
-	(void) xfrm_policy_walk(XFRM_POLICY_TYPE_MAIN, dump_one_policy, &info);
-#ifdef CONFIG_XFRM_SUB_POLICY
-	(void) xfrm_policy_walk(XFRM_POLICY_TYPE_SUB, dump_one_policy, &info);
-#endif
-	cb->args[0] = info.this_idx;
+
+	if (!cb->args[0]) {
+		cb->args[0] = 1;
+		xfrm_policy_walk_init(walk, XFRM_POLICY_TYPE_ANY);
+	}
+
+	(void) xfrm_policy_walk(walk, dump_one_policy, &info);
 
 	return skb->len;
 }
@@ -1293,7 +1306,6 @@ static struct sk_buff *xfrm_policy_netlink(struct sk_buff *in_skb,
 	info.out_skb = skb;
 	info.nlmsg_seq = seq;
 	info.nlmsg_flags = 0;
-	info.this_idx = info.start_idx = 0;
 
 	if (dump_one_policy(xp, dir, 0, &info) < 0) {
 		kfree_skb(skb);
@@ -1891,15 +1903,18 @@ static const struct nla_policy xfrma_policy[XFRMA_MAX+1] = {
 static struct xfrm_link {
 	int (*doit)(struct sk_buff *, struct nlmsghdr *, struct nlattr **);
 	int (*dump)(struct sk_buff *, struct netlink_callback *);
+	int (*done)(struct netlink_callback *);
 } xfrm_dispatch[XFRM_NR_MSGTYPES] = {
 	[XFRM_MSG_NEWSA       - XFRM_MSG_BASE] = { .doit = xfrm_add_sa        },
 	[XFRM_MSG_DELSA       - XFRM_MSG_BASE] = { .doit = xfrm_del_sa        },
 	[XFRM_MSG_GETSA       - XFRM_MSG_BASE] = { .doit = xfrm_get_sa,
-						   .dump = xfrm_dump_sa       },
+						   .dump = xfrm_dump_sa,
+						   .done = xfrm_dump_sa_done  },
 	[XFRM_MSG_NEWPOLICY   - XFRM_MSG_BASE] = { .doit = xfrm_add_policy    },
 	[XFRM_MSG_DELPOLICY   - XFRM_MSG_BASE] = { .doit = xfrm_get_policy    },
 	[XFRM_MSG_GETPOLICY   - XFRM_MSG_BASE] = { .doit = xfrm_get_policy,
-						   .dump = xfrm_dump_policy   },
+						   .dump = xfrm_dump_policy,
+						   .done = xfrm_dump_policy_done },
 	[XFRM_MSG_ALLOCSPI    - XFRM_MSG_BASE] = { .doit = xfrm_alloc_userspi },
 	[XFRM_MSG_ACQUIRE     - XFRM_MSG_BASE] = { .doit = xfrm_add_acquire   },
 	[XFRM_MSG_EXPIRE      - XFRM_MSG_BASE] = { .doit = xfrm_add_sa_expire },
@@ -1938,7 +1953,7 @@ static int xfrm_user_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
 		if (link->dump == NULL)
 			return -EINVAL;
 
-		return netlink_dump_start(xfrm_nl, skb, nlh, link->dump, NULL);
+		return netlink_dump_start(xfrm_nl, skb, nlh, link->dump, link->done);
 	}
 
 	err = nlmsg_parse(nlh, xfrm_msg_min[type], attrs, XFRMA_MAX,