diff mbox series

[RFC,net-next,08/14] ipvs: use resizable hash table for services

Message ID 20230815173031.168344-9-ja@ssi.bg (mailing list archive)
State New, archived
Headers show
Series ipvs: per-net tables and optimizations | expand

Commit Message

Julian Anastasov Aug. 15, 2023, 5:30 p.m. UTC
Make the hash table for services resizable in the bit range of 4-20.
Table is attached only while services are present. Resizing is done
by delayed work based on load (the number of hashed services).
Table grows when load increases 2+ times (above 12.5% with factor=3)
and shrinks 8+ times when load decreases 16+ times (below 0.78%).

Switch to jhash hashing to reduce the collisions for multiple
services.

Add a hash_key field into the service that includes table ID and
bucket ID. This helps the lookup and delete operations.

Signed-off-by: Julian Anastasov <ja@ssi.bg>
---
 include/net/ip_vs.h            |  49 ++-
 net/netfilter/ipvs/ip_vs_ctl.c | 676 +++++++++++++++++++++++++++------
 2 files changed, 594 insertions(+), 131 deletions(-)
diff mbox series

Patch

diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index 51adba5f1bb9..1fe04b91d4a7 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -35,12 +35,10 @@ 
 
 #define IP_VS_HDR_INVERSE	1
 #define IP_VS_HDR_ICMP		2
-/*
- *	Hash table: for virtual service lookups
- */
-#define IP_VS_SVC_TAB_BITS 8
-#define IP_VS_SVC_TAB_SIZE BIT(IP_VS_SVC_TAB_BITS)
-#define IP_VS_SVC_TAB_MASK (IP_VS_SVC_TAB_SIZE - 1)
+
+/* svc_table limits */
+#define IP_VS_SVC_TAB_MIN_BITS	4
+#define IP_VS_SVC_TAB_MAX_BITS	20
 
 /* Generic access of ipvs struct */
 static inline struct netns_ipvs *net_ipvs(struct net* net)
@@ -51,8 +49,6 @@  static inline struct netns_ipvs *net_ipvs(struct net* net)
 /* Connections' size value needed by ip_vs_ctl.c */
 extern int ip_vs_conn_tab_size;
 
-extern struct mutex __ip_vs_mutex;
-
 struct ip_vs_iphdr {
 	int hdr_flags;	/* ipvs flags */
 	__u32 off;	/* Where IP or IPv4 header starts */
@@ -289,6 +285,12 @@  static inline int ip_vs_af_index(int af)
 	return af == AF_INET6 ? IP_VS_AF_INET6 : IP_VS_AF_INET;
 }
 
+/* work_flags */
+enum {
+	IP_VS_WORK_SVC_RESIZE,		/* Schedule svc_resize_work */
+	IP_VS_WORK_SVC_NORESIZE,	/* Stopping svc_resize_work */
+};
+
 /* The port number of FTP service (in network order). */
 #define FTPPORT  cpu_to_be16(21)
 #define FTPDATA  cpu_to_be16(20)
@@ -888,14 +890,15 @@  struct ip_vs_dest_user_kern {
  * forwarding entries.
  */
 struct ip_vs_service {
-	struct hlist_node	s_list;   /* node in service table */
-	atomic_t		refcnt;   /* reference counter */
-
+	struct hlist_bl_node	s_list;   /* node in service table */
+	u32			hash_key; /* Key for the hash table */
 	u16			af;       /* address family */
 	__u16			protocol; /* which protocol (TCP/UDP) */
+
 	union nf_inet_addr	addr;	  /* IP address for virtual service */
-	__be16			port;	  /* port number for the service */
 	__u32                   fwmark;   /* firewall mark of the service */
+	atomic_t		refcnt;   /* reference counter */
+	__be16			port;	  /* port number for the service */
 	unsigned int		flags;	  /* service status flags */
 	unsigned int		timeout;  /* persistent timeout in ticks */
 	__be32			netmask;  /* grouping granularity, mask/plen */
@@ -1153,6 +1156,10 @@  struct netns_ipvs {
 	struct list_head	dest_trash;
 	spinlock_t		dest_trash_lock;
 	struct timer_list	dest_trash_timer; /* expiration timer */
+	struct mutex		service_mutex;    /* service reconfig */
+	struct rw_semaphore	svc_resize_sem;   /* svc_table resizing */
+	struct delayed_work	svc_resize_work;  /* resize svc_table */
+	atomic_t		svc_table_changes;/* ++ on new table */
 	/* Service counters */
 	atomic_t		num_services[IP_VS_AF_MAX];   /* Services */
 	atomic_t		fwm_services[IP_VS_AF_MAX];   /* Services */
@@ -1216,6 +1223,7 @@  struct netns_ipvs {
 	int			sysctl_est_nice;	/* kthread nice */
 	int			est_stopped;		/* stop tasks */
 #endif
+	int			sysctl_svc_lfactor;
 
 	/* ip_vs_lblc */
 	int			sysctl_lblc_expiration;
@@ -1225,6 +1233,7 @@  struct netns_ipvs {
 	int			sysctl_lblcr_expiration;
 	struct ctl_table_header	*lblcr_ctl_header;
 	struct ctl_table	*lblcr_ctl_table;
+	unsigned long		work_flags;	/* IP_VS_WORK_* flags */
 	/* ip_vs_est */
 	struct delayed_work	est_reload_work;/* Reload kthread tasks */
 	struct mutex		est_mutex;	/* protect kthread tasks */
@@ -1256,9 +1265,7 @@  struct netns_ipvs {
 	unsigned int		mixed_address_family_dests;
 	unsigned int		hooks_afmask;	/* &1=AF_INET, &2=AF_INET6 */
 
-	/* the service mutex that protect svc_table and svc_fwm_table */
-	struct mutex service_mutex;
-	struct hlist_head svc_table[IP_VS_SVC_TAB_SIZE];	/* Services */
+	struct ip_vs_rht __rcu	*svc_table;	/* Services */
 };
 
 #define DEFAULT_SYNC_THRESHOLD	3
@@ -1495,6 +1502,18 @@  static inline int sysctl_est_nice(struct netns_ipvs *ipvs)
 
 #endif
 
+/* Get load factor to map num_services/u_thresh to t->size
+ * Large value decreases u_thresh to reduce collisions but increases
+ * the table size
+ * Returns factor where:
+ * - non-negative: u_thresh = size >> factor, eg. lfactor 2 = 25% load
+ * - negative: u_thresh = size << factor, eg. lfactor -1 = 200% load
+ */
+static inline int sysctl_svc_lfactor(struct netns_ipvs *ipvs)
+{
+	return READ_ONCE(ipvs->sysctl_svc_lfactor);
+}
+
 /* IPVS core functions
  * (from ip_vs_core.c)
  */
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index e44dfb97566a..dcb60ed726f2 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -294,47 +294,59 @@  ip_vs_use_count_dec(void)
 }
 
 
-
+/* Service hashing:
+ * Operation			Locking order
+ * ---------------------------------------------------------------------------
+ * add table			service_mutex, svc_resize_sem(W)
+ * del table			service_mutex
+ * move between tables		svc_resize_sem(W), seqcount_t(W), bit lock
+ * add/del service		service_mutex, bit lock
+ * find service			RCU, seqcount_t(R)
+ * walk services(blocking)	service_mutex, svc_resize_sem(R)
+ * walk services(non-blocking)	RCU, seqcount_t(R)
+ *
+ * - new tables are linked/unlinked under service_mutex and svc_resize_sem
+ * - new table is linked on resizing and all operations can run in parallel
+ * in 2 tables until the new table is registered as current one
+ * - two contexts can modify buckets: config and table resize, both in
+ * process context
+ * - only table resizer can move entries, so we do not protect t->seqc[]
+ * items with t->lock[]
+ * - lookups occur under RCU lock and seqcount reader lock to detect if
+ * services are moved to new table
+ * - move operations may disturb readers: find operation will not miss entries
+ * but walkers may see same entry twice if they are forced to retry chains
+ * - walkers using cond_resched_rcu() on !PREEMPT_RCU may need to hold
+ * service_mutex to disallow new tables to be installed or to check
+ * svc_table_changes and repeat the RCU read section if new table is installed
+ */
 
 /*
  *	Returns hash value for virtual service
  */
-static inline unsigned int
-ip_vs_svc_hashkey(struct netns_ipvs *ipvs, int af, unsigned int proto,
+static inline u32
+ip_vs_svc_hashval(struct ip_vs_rht *t, int af, unsigned int proto,
 		  const union nf_inet_addr *addr, __be16 port)
 {
-	unsigned int porth = ntohs(port);
-	__be32 addr_fold = addr->ip;
-	__u32 ahash;
-
-#ifdef CONFIG_IP_VS_IPV6
-	if (af == AF_INET6)
-		addr_fold = addr->ip6[0]^addr->ip6[1]^
-			    addr->ip6[2]^addr->ip6[3];
-#endif
-	ahash = ntohl(addr_fold);
-	ahash ^= ((size_t) ipvs >> 8);
-
-	return (proto ^ ahash ^ (porth >> IP_VS_SVC_TAB_BITS) ^ porth) &
-	       IP_VS_SVC_TAB_MASK;
+	return ip_vs_rht_hash_linfo(t, af, addr, ntohs(port), proto);
 }
 
 /*
  *	Returns hash value of fwmark for virtual service lookup
  */
-static inline unsigned int ip_vs_svc_fwm_hashkey(struct netns_ipvs *ipvs, __u32 fwmark)
+static inline u32 ip_vs_svc_fwm_hashval(struct ip_vs_rht *t, int af,
+					__u32 fwmark)
 {
-	return (((size_t)ipvs>>8) ^ fwmark) & IP_VS_SVC_TAB_MASK;
+	return jhash_2words(fwmark, af, (u32)t->hash_key.key[0]);
 }
 
-/*
- *	Hashes a service in the svc_table by <netns,proto,addr,port>
- *	or by fwmark.
- *	Should be called with locked tables.
- */
+/* Hashes a service in the svc_table by <proto,addr,port> or by fwmark */
 static int ip_vs_svc_hash(struct ip_vs_service *svc)
 {
-	unsigned int hash;
+	struct netns_ipvs *ipvs = svc->ipvs;
+	struct hlist_bl_head *head;
+	struct ip_vs_rht *t;
+	u32 hash;
 
 	if (svc->flags & IP_VS_SVC_F_HASHED) {
 		pr_err("%s(): request for already hashed, called from %pS\n",
@@ -342,23 +354,32 @@  static int ip_vs_svc_hash(struct ip_vs_service *svc)
 		return 0;
 	}
 
+	/* increase its refcnt because it is referenced by the svc table */
+	atomic_inc(&svc->refcnt);
+
+	/* New entries go into recent table */
+	t = rcu_dereference_protected(ipvs->svc_table, 1);
+	t = rcu_dereference_protected(t->new_tbl, 1);
+
 	if (svc->fwmark == 0) {
 		/*
-		 *  Hash it by <netns,protocol,addr,port>
+		 *  Hash it by <protocol,addr,port>
 		 */
-		hash = ip_vs_svc_hashkey(svc->ipvs, svc->af, svc->protocol,
+		hash = ip_vs_svc_hashval(t, svc->af, svc->protocol,
 					 &svc->addr, svc->port);
 	} else {
 		/*
 		 *  Hash it by fwmark
 		 */
-		hash = ip_vs_svc_fwm_hashkey(svc->ipvs, svc->fwmark);
+		hash = ip_vs_svc_fwm_hashval(t, svc->af, svc->fwmark);
 	}
-	hlist_add_head_rcu(&svc->s_list, &svc->ipvs->svc_table[hash]);
-
+	head = t->buckets + (hash & t->mask);
+	hlist_bl_lock(head);
+	WRITE_ONCE(svc->hash_key, ip_vs_rht_build_hash_key(t, hash));
 	svc->flags |= IP_VS_SVC_F_HASHED;
-	/* increase its refcnt because it is referenced by the svc table */
-	atomic_inc(&svc->refcnt);
+	hlist_bl_add_head_rcu(&svc->s_list, head);
+	hlist_bl_unlock(head);
+
 	return 1;
 }
 
@@ -369,17 +390,45 @@  static int ip_vs_svc_hash(struct ip_vs_service *svc)
  */
 static int ip_vs_svc_unhash(struct ip_vs_service *svc)
 {
+	struct netns_ipvs *ipvs = svc->ipvs;
+	struct hlist_bl_head *head;
+	struct ip_vs_rht *t;
+	u32 hash_key2;
+	u32 hash_key;
+
 	if (!(svc->flags & IP_VS_SVC_F_HASHED)) {
 		pr_err("%s(): request for unhash flagged, called from %pS\n",
 		       __func__, __builtin_return_address(0));
 		return 0;
 	}
 
+	t = rcu_dereference_protected(ipvs->svc_table, 1);
+	hash_key = READ_ONCE(svc->hash_key);
+	/* We need to lock the bucket in the right table */
+	if (ip_vs_rht_same_table(t, hash_key)) {
+		head = t->buckets + (hash_key & t->mask);
+		hlist_bl_lock(head);
+		/* Ensure hash_key is read under lock */
+		hash_key2 = READ_ONCE(svc->hash_key);
+		/* Moved to new table ? */
+		if (hash_key != hash_key2) {
+			hlist_bl_unlock(head);
+			t = rcu_dereference_protected(t->new_tbl, 1);
+			head = t->buckets + (hash_key2 & t->mask);
+			hlist_bl_lock(head);
+		}
+	} else {
+		/* It is already moved to new table */
+		t = rcu_dereference_protected(t->new_tbl, 1);
+		head = t->buckets + (hash_key & t->mask);
+		hlist_bl_lock(head);
+	}
 	/* Remove it from svc_table */
-	hlist_del_rcu(&svc->s_list);
+	hlist_bl_del_rcu(&svc->s_list);
 
 	svc->flags &= ~IP_VS_SVC_F_HASHED;
 	atomic_dec(&svc->refcnt);
+	hlist_bl_unlock(head);
 	return 1;
 }
 
@@ -391,18 +440,29 @@  static inline struct ip_vs_service *
 __ip_vs_service_find(struct netns_ipvs *ipvs, int af, __u16 protocol,
 		     const union nf_inet_addr *vaddr, __be16 vport)
 {
-	unsigned int hash;
+	DECLARE_IP_VS_RHT_WALK_BUCKET_RCU();
+	struct hlist_bl_head *head;
 	struct ip_vs_service *svc;
-
-	/* Check for "full" addressed entries */
-	hash = ip_vs_svc_hashkey(ipvs, af, protocol, vaddr, vport);
-
-	hlist_for_each_entry_rcu(svc, &ipvs->svc_table[hash], s_list) {
-		if (svc->af == af && ip_vs_addr_equal(af, &svc->addr, vaddr) &&
-		    svc->port == vport && svc->protocol == protocol &&
-		    !svc->fwmark) {
-			/* HIT */
-			return svc;
+	struct ip_vs_rht *t, *p;
+	struct hlist_bl_node *e;
+	u32 hash, hash_key;
+
+	ip_vs_rht_for_each_table_rcu(ipvs->svc_table, t, p) {
+		/* Check for "full" addressed entries */
+		hash = ip_vs_svc_hashval(t, af, protocol, vaddr, vport);
+
+		hash_key = ip_vs_rht_build_hash_key(t, hash);
+		ip_vs_rht_walk_bucket_rcu(t, hash_key, head) {
+			hlist_bl_for_each_entry_rcu(svc, e, head, s_list) {
+				if (READ_ONCE(svc->hash_key) == hash_key &&
+				    svc->af == af &&
+				    ip_vs_addr_equal(af, &svc->addr, vaddr) &&
+				    svc->port == vport &&
+				    svc->protocol == protocol && !svc->fwmark) {
+					/* HIT */
+					return svc;
+				}
+			}
 		}
 	}
 
@@ -416,16 +476,26 @@  __ip_vs_service_find(struct netns_ipvs *ipvs, int af, __u16 protocol,
 static inline struct ip_vs_service *
 __ip_vs_svc_fwm_find(struct netns_ipvs *ipvs, int af, __u32 fwmark)
 {
-	unsigned int hash;
+	DECLARE_IP_VS_RHT_WALK_BUCKET_RCU();
+	struct hlist_bl_head *head;
 	struct ip_vs_service *svc;
-
-	/* Check for fwmark addressed entries */
-	hash = ip_vs_svc_fwm_hashkey(ipvs, fwmark);
-
-	hlist_for_each_entry_rcu(svc, &ipvs->svc_table[hash], s_list) {
-		if (svc->fwmark == fwmark && svc->af == af) {
-			/* HIT */
-			return svc;
+	struct ip_vs_rht *t, *p;
+	struct hlist_bl_node *e;
+	u32 hash, hash_key;
+
+	ip_vs_rht_for_each_table_rcu(ipvs->svc_table, t, p) {
+		/* Check for fwmark addressed entries */
+		hash = ip_vs_svc_fwm_hashval(t, af, fwmark);
+
+		hash_key = ip_vs_rht_build_hash_key(t, hash);
+		ip_vs_rht_walk_bucket_rcu(t, hash_key, head) {
+			hlist_bl_for_each_entry_rcu(svc, e, head, s_list) {
+				if (READ_ONCE(svc->hash_key) == hash_key &&
+				    svc->fwmark == fwmark && svc->af == af) {
+					/* HIT */
+					return svc;
+				}
+			}
 		}
 	}
 
@@ -488,6 +558,220 @@  ip_vs_service_find(struct netns_ipvs *ipvs, int af, __u32 fwmark, __u16 protocol
 	return svc;
 }
 
+/* Return the number of registered services */
+static int ip_vs_get_num_services(struct netns_ipvs *ipvs)
+{
+	int ns = 0, ni = IP_VS_AF_MAX;
+
+	while (--ni >= 0)
+		ns += atomic_read(&ipvs->num_services[ni]);
+	return ns;
+}
+
+/* Get default load factor to map num_services/u_thresh to t->size */
+static int ip_vs_svc_default_load_factor(struct netns_ipvs *ipvs)
+{
+	int factor;
+
+	if (net_eq(ipvs->net, &init_net))
+		factor = 3;	/* grow if load is above 12.5% */
+	else
+		factor = 2;	/* grow if load is above 25% */
+	return factor;
+}
+
+/* Get the desired svc_table size */
+static int ip_vs_svc_desired_size(struct netns_ipvs *ipvs, struct ip_vs_rht *t,
+				  int lfactor)
+{
+	return ip_vs_rht_desired_size(ipvs, t, ip_vs_get_num_services(ipvs),
+				      lfactor, IP_VS_SVC_TAB_MIN_BITS,
+				      IP_VS_SVC_TAB_MAX_BITS);
+}
+
+/* Allocate svc_table */
+static struct ip_vs_rht *ip_vs_svc_table_alloc(struct netns_ipvs *ipvs,
+					       int buckets, int lfactor)
+{
+	struct ip_vs_rht *t;
+	int scounts, locks;
+
+	/* No frequent lookups to race with resizing, so use max of 64
+	 * seqcounts. Only resizer moves entries, so use 0 locks.
+	 */
+	scounts = clamp(buckets >> 4, 1, 64);
+	locks = 0;
+
+	t = ip_vs_rht_alloc(buckets, scounts, locks);
+	if (!t)
+		return NULL;
+	t->lfactor = lfactor;
+	ip_vs_rht_set_thresholds(t, t->size, lfactor, IP_VS_SVC_TAB_MIN_BITS,
+				 IP_VS_SVC_TAB_MAX_BITS);
+	return t;
+}
+
+/* svc_table resizer work */
+static void svc_resize_work_handler(struct work_struct *work)
+{
+	struct hlist_bl_head *head, *head2;
+	struct ip_vs_rht *t_free = NULL;
+	unsigned int resched_score = 0;
+	struct hlist_bl_node *cn, *nn;
+	struct ip_vs_rht *t, *t_new;
+	struct ip_vs_service *svc;
+	struct netns_ipvs *ipvs;
+	bool more_work = true;
+	seqcount_t *sc;
+	int limit = 0;
+	int new_size;
+	int lfactor;
+	u32 bucket;
+
+	ipvs = container_of(work, struct netns_ipvs, svc_resize_work.work);
+
+	if (!down_write_trylock(&ipvs->svc_resize_sem))
+		goto out;
+	if (!mutex_trylock(&ipvs->service_mutex))
+		goto unlock_sem;
+	more_work = false;
+	clear_bit(IP_VS_WORK_SVC_RESIZE, &ipvs->work_flags);
+	if (!ipvs->enable ||
+	    test_bit(IP_VS_WORK_SVC_NORESIZE, &ipvs->work_flags))
+		goto unlock_m;
+	t = rcu_dereference_protected(ipvs->svc_table, 1);
+	/* Do nothing if table is removed */
+	if (!t)
+		goto unlock_m;
+	/* New table needs to be registered? BUG! */
+	if (t != rcu_dereference_protected(t->new_tbl, 1))
+		goto unlock_m;
+
+	lfactor = sysctl_svc_lfactor(ipvs);
+	/* Should we resize ? */
+	new_size = ip_vs_svc_desired_size(ipvs, t, lfactor);
+	if (new_size == t->size && lfactor == t->lfactor)
+		goto unlock_m;
+
+	t_new = ip_vs_svc_table_alloc(ipvs, new_size, lfactor);
+	if (!t_new) {
+		more_work = true;
+		goto unlock_m;
+	}
+	/* Flip the table_id */
+	t_new->table_id = t->table_id ^ IP_VS_RHT_TABLE_ID_MASK;
+
+	rcu_assign_pointer(t->new_tbl, t_new);
+	/* Allow add/del to new_tbl while moving from old table */
+	mutex_unlock(&ipvs->service_mutex);
+
+	ip_vs_rht_for_each_bucket(t, bucket, head) {
+same_bucket:
+		if (++limit >= 16) {
+			if (!ipvs->enable ||
+			    test_bit(IP_VS_WORK_SVC_NORESIZE,
+				     &ipvs->work_flags))
+				goto unlock_sem;
+			if (resched_score >= 100) {
+				resched_score = 0;
+				cond_resched();
+			}
+			limit = 0;
+		}
+		if (hlist_bl_empty(head)) {
+			resched_score++;
+			continue;
+		}
+		/* Preemption calls ahead... */
+		resched_score = 0;
+
+		sc = &t->seqc[bucket & t->seqc_mask];
+		/* seqcount_t usage considering PREEMPT_RT rules:
+		 * - we are the only writer => preemption can be allowed
+		 * - readers (SoftIRQ) => disable BHs
+		 * - readers (processes) => preemption should be disabled
+		 */
+		local_bh_disable();
+		preempt_disable_nested();
+		write_seqcount_begin(sc);
+		hlist_bl_lock(head);
+
+		hlist_bl_for_each_entry_safe(svc, cn, nn, head, s_list) {
+			u32 hash;
+
+			/* New hash for the new table */
+			if (svc->fwmark == 0) {
+				/*  Hash it by <protocol,addr,port> */
+				hash = ip_vs_svc_hashval(t_new, svc->af,
+							 svc->protocol,
+							 &svc->addr, svc->port);
+			} else {
+				/* Hash it by fwmark */
+				hash = ip_vs_svc_fwm_hashval(t_new, svc->af,
+							     svc->fwmark);
+			}
+			hlist_bl_del_rcu(&svc->s_list);
+			head2 = t_new->buckets + (hash & t_new->mask);
+
+			hlist_bl_lock(head2);
+			WRITE_ONCE(svc->hash_key,
+				   ip_vs_rht_build_hash_key(t_new, hash));
+			/* t_new->seqc are not used at this stage, we race
+			 * only with add/del, so only lock the bucket.
+			 */
+			hlist_bl_add_head_rcu(&svc->s_list, head2);
+			hlist_bl_unlock(head2);
+			/* Too long chain? Do it in steps */
+			if (++limit >= 64)
+				break;
+		}
+
+		hlist_bl_unlock(head);
+		write_seqcount_end(sc);
+		preempt_enable_nested();
+		local_bh_enable();
+		if (limit >= 64)
+			goto same_bucket;
+	}
+
+	/* Tables can be switched only under service_mutex */
+	while (!mutex_trylock(&ipvs->service_mutex)) {
+		cond_resched();
+		if (!ipvs->enable ||
+		    test_bit(IP_VS_WORK_SVC_NORESIZE, &ipvs->work_flags))
+			goto unlock_sem;
+	}
+	if (!ipvs->enable ||
+	    test_bit(IP_VS_WORK_SVC_NORESIZE, &ipvs->work_flags))
+		goto unlock_sem;
+
+	rcu_assign_pointer(ipvs->svc_table, t_new);
+	/* Inform readers that new table is installed */
+	smp_mb__before_atomic();
+	atomic_inc(&ipvs->svc_table_changes);
+	t_free = t;
+
+unlock_m:
+	mutex_unlock(&ipvs->service_mutex);
+
+unlock_sem:
+	up_write(&ipvs->svc_resize_sem);
+
+	if (t_free) {
+		/* RCU readers should not see more than two tables in chain.
+		 * To prevent new table to be attached wait here instead of
+		 * freeing the old table in RCU callback.
+		 */
+		synchronize_rcu();
+		ip_vs_rht_free(t_free);
+	}
+
+out:
+	if (!ipvs->enable || !more_work ||
+	    test_bit(IP_VS_WORK_SVC_NORESIZE, &ipvs->work_flags))
+		return;
+	queue_delayed_work(system_unbound_wq, &ipvs->svc_resize_work, 1);
+}
 
 static inline void
 __ip_vs_bind_svc(struct ip_vs_dest *dest, struct ip_vs_service *svc)
@@ -1357,12 +1641,13 @@  static int
 ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 		  struct ip_vs_service **svc_p)
 {
-	int ret = 0;
 	struct ip_vs_scheduler *sched = NULL;
+	struct ip_vs_rht *t, *t_new = NULL;
 	int af_id = ip_vs_af_index(u->af);
-	struct ip_vs_pe *pe = NULL;
 	struct ip_vs_service *svc = NULL;
+	struct ip_vs_pe *pe = NULL;
 	int ret_hooks = -1;
+	int ret = 0;
 
 	/* increase the module use count */
 	if (!ip_vs_use_count_inc())
@@ -1404,6 +1689,18 @@  ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 	}
 #endif
 
+	t = rcu_dereference_protected(ipvs->svc_table, 1);
+	if (!t) {
+		int lfactor = sysctl_svc_lfactor(ipvs);
+		int new_size = ip_vs_svc_desired_size(ipvs, NULL, lfactor);
+
+		t_new = ip_vs_svc_table_alloc(ipvs, new_size, lfactor);
+		if (!t_new) {
+			ret = -ENOMEM;
+			goto out_err;
+		}
+	}
+
 	if (!atomic_read(&ipvs->num_services[af_id])) {
 		ret = ip_vs_register_hooks(ipvs, u->af);
 		if (ret < 0)
@@ -1453,6 +1750,15 @@  ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 	RCU_INIT_POINTER(svc->pe, pe);
 	pe = NULL;
 
+	if (t_new) {
+		clear_bit(IP_VS_WORK_SVC_NORESIZE, &ipvs->work_flags);
+		rcu_assign_pointer(ipvs->svc_table, t_new);
+		t_new = NULL;
+	}
+
+	/* Hash the service into the service table */
+	ip_vs_svc_hash(svc);
+
 	/* Update the virtual service counters */
 	if (svc->port == FTPPORT)
 		atomic_inc(&ipvs->ftpsvc_counter[af_id]);
@@ -1467,8 +1773,11 @@  ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 		atomic_inc(&ipvs->nonfwm_services[af_id]);
 	atomic_inc(&ipvs->num_services[af_id]);
 
-	/* Hash the service into the service table */
-	ip_vs_svc_hash(svc);
+	/* Schedule resize work */
+	if (t && ip_vs_get_num_services(ipvs) > t->u_thresh &&
+	    !test_and_set_bit(IP_VS_WORK_SVC_RESIZE, &ipvs->work_flags))
+		queue_delayed_work(system_unbound_wq, &ipvs->svc_resize_work,
+				   1);
 
 	*svc_p = svc;
 
@@ -1484,6 +1793,8 @@  ip_vs_add_service(struct netns_ipvs *ipvs, struct ip_vs_service_user_kern *u,
 
 
  out_err:
+	if (t_new)
+		ip_vs_rht_free(t_new);
 	if (ret_hooks >= 0)
 		ip_vs_unregister_hooks(ipvs, u->af);
 	if (svc != NULL) {
@@ -1671,10 +1982,38 @@  static void ip_vs_unlink_service(struct ip_vs_service *svc, bool cleanup)
  */
 static int ip_vs_del_service(struct ip_vs_service *svc)
 {
+	struct netns_ipvs *ipvs;
+	struct ip_vs_rht *t, *p;
+	int ns;
+
 	if (svc == NULL)
 		return -EEXIST;
+	ipvs = svc->ipvs;
 	ip_vs_unlink_service(svc, false);
-
+	t = rcu_dereference_protected(ipvs->svc_table, 1);
+
+	/* Drop the table if no more services */
+	ns = ip_vs_get_num_services(ipvs);
+	if (!ns) {
+		/* Stop the resizer and drop the tables */
+		set_bit(IP_VS_WORK_SVC_NORESIZE, &ipvs->work_flags);
+		cancel_delayed_work_sync(&ipvs->svc_resize_work);
+		if (t) {
+			rcu_assign_pointer(ipvs->svc_table, NULL);
+			while (1) {
+				p = rcu_dereference_protected(t->new_tbl, 1);
+				call_rcu(&t->rcu_head, ip_vs_rht_rcu_free);
+				if (p == t)
+					break;
+				t = p;
+			}
+		}
+	} else if (ns <= t->l_thresh &&
+		   !test_and_set_bit(IP_VS_WORK_SVC_RESIZE,
+				     &ipvs->work_flags)) {
+		queue_delayed_work(system_unbound_wq, &ipvs->svc_resize_work,
+				   1);
+	}
 	return 0;
 }
 
@@ -1684,14 +2023,36 @@  static int ip_vs_del_service(struct ip_vs_service *svc)
  */
 static int ip_vs_flush(struct netns_ipvs *ipvs, bool cleanup)
 {
-	int idx;
+	DECLARE_IP_VS_RHT_WALK_BUCKETS();
+	struct hlist_bl_head *head;
 	struct ip_vs_service *svc;
-	struct hlist_node *n;
+	struct hlist_bl_node *ne;
+	struct hlist_bl_node *e;
+	struct ip_vs_rht *t, *p;
+
+	/* Stop the resizer and drop the tables */
+	if (!test_and_set_bit(IP_VS_WORK_SVC_NORESIZE, &ipvs->work_flags))
+		cancel_delayed_work_sync(&ipvs->svc_resize_work);
+	/* No resizer, so now we have exclusive write access */
+
+	if (ip_vs_get_num_services(ipvs)) {
+		ip_vs_rht_walk_buckets(ipvs->svc_table, head) {
+			hlist_bl_for_each_entry_safe(svc, e, ne, head, s_list)
+				ip_vs_unlink_service(svc, cleanup);
+		}
+	}
 
-	for(idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
-		hlist_for_each_entry_safe(svc, n, &ipvs->svc_table[idx],
-					  s_list)
-			ip_vs_unlink_service(svc, cleanup);
+	/* Unregister the hash table and release it after RCU grace period */
+	t = rcu_dereference_protected(ipvs->svc_table, 1);
+	if (t) {
+		rcu_assign_pointer(ipvs->svc_table, NULL);
+		while (1) {
+			p = rcu_dereference_protected(t->new_tbl, 1);
+			call_rcu(&t->rcu_head, ip_vs_rht_rcu_free);
+			if (p == t)
+				break;
+			t = p;
+		}
 	}
 	return 0;
 }
@@ -1742,19 +2103,44 @@  static int ip_vs_dst_event(struct notifier_block *this, unsigned long event,
 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 	struct net *net = dev_net(dev);
 	struct netns_ipvs *ipvs = net_ipvs(net);
+	DECLARE_IP_VS_RHT_WALK_BUCKETS_RCU();
+	unsigned int resched_score = 0;
+	struct hlist_bl_head *head;
 	struct ip_vs_service *svc;
+	struct hlist_bl_node *e;
 	struct ip_vs_dest *dest;
-	unsigned int idx;
+	int old_gen, new_gen;
 
 	if (event != NETDEV_DOWN || !ipvs)
 		return NOTIFY_DONE;
 	IP_VS_DBG(3, "%s() dev=%s\n", __func__, dev->name);
+
+	old_gen = atomic_read(&ipvs->svc_table_changes);
+
 	rcu_read_lock();
-	for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
-		hlist_for_each_entry_rcu(svc, &ipvs->svc_table[idx], s_list)
+
+repeat:
+	smp_rmb(); /* ipvs->svc_table and svc_table_changes */
+	ip_vs_rht_walk_buckets_rcu(ipvs->svc_table, head) {
+		hlist_bl_for_each_entry_rcu(svc, e, head, s_list) {
 			list_for_each_entry_rcu(dest, &svc->destinations,
-						n_list)
+						n_list) {
 				ip_vs_forget_dev(dest, dev);
+				resched_score += 10;
+			}
+			resched_score++;
+		}
+		resched_score++;
+		if (resched_score >= 100) {
+			resched_score = 0;
+			cond_resched_rcu();
+			new_gen = atomic_read(&ipvs->svc_table_changes);
+			/* New table installed ? */
+			if (old_gen != new_gen) {
+				old_gen = new_gen;
+				goto repeat;
+			}
+		}
 	}
 	rcu_read_unlock();
 
@@ -1777,14 +2163,28 @@  static int ip_vs_zero_service(struct ip_vs_service *svc)
 
 static int ip_vs_zero_all(struct netns_ipvs *ipvs)
 {
-	int idx;
+	DECLARE_IP_VS_RHT_WALK_BUCKETS_RCU();
+	unsigned int resched_score = 0;
+	struct hlist_bl_head *head;
 	struct ip_vs_service *svc;
+	struct hlist_bl_node *e;
+
+	rcu_read_lock();
 
-	for(idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
-		hlist_for_each_entry(svc, &ipvs->svc_table[idx], s_list)
+	ip_vs_rht_walk_buckets_rcu(ipvs->svc_table, head) {
+		hlist_bl_for_each_entry_rcu(svc, e, head, s_list) {
 			ip_vs_zero_service(svc);
+			resched_score += 10;
+		}
+		resched_score++;
+		if (resched_score >= 100) {
+			resched_score = 0;
+			cond_resched_rcu();
+		}
 	}
 
+	rcu_read_unlock();
+
 	ip_vs_zero_stats(&ipvs->tot_stats->s);
 	return 0;
 }
@@ -2215,7 +2615,8 @@  static struct ctl_table vs_vars[] = {
 
 struct ip_vs_iter {
 	struct seq_net_private p;  /* Do not move this, netns depends upon it*/
-	int bucket;
+	struct ip_vs_rht *t;
+	u32 bucket;
 };
 
 /*
@@ -2236,17 +2637,23 @@  static inline const char *ip_vs_fwd_name(unsigned int flags)
 	}
 }
 
-
+/* Do not expect consistent view during add, del and move(table resize).
+ * We may miss entries and even show duplicates.
+ */
 static struct ip_vs_service *ip_vs_info_array(struct seq_file *seq, loff_t pos)
 {
-	struct net *net = seq_file_net(seq);
-	struct netns_ipvs *ipvs = net_ipvs(net);
 	struct ip_vs_iter *iter = seq->private;
-	int idx;
+	struct ip_vs_rht *t = iter->t;
 	struct ip_vs_service *svc;
+	struct hlist_bl_node *e;
+	int idx;
 
-	for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
-		hlist_for_each_entry_rcu(svc, &ipvs->svc_table[idx], s_list) {
+	if (!t)
+		return NULL;
+	for (idx = 0; idx < t->size; idx++) {
+		hlist_bl_for_each_entry_rcu(svc, e, &t->buckets[idx], s_list) {
+			if (!ip_vs_rht_same_table(t, READ_ONCE(svc->hash_key)))
+				break;
 			if (pos-- == 0) {
 				iter->bucket = idx;
 				return svc;
@@ -2259,18 +2666,22 @@  static struct ip_vs_service *ip_vs_info_array(struct seq_file *seq, loff_t pos)
 static void *ip_vs_info_seq_start(struct seq_file *seq, loff_t *pos)
 	__acquires(RCU)
 {
+	struct ip_vs_iter *iter = seq->private;
+	struct net *net = seq_file_net(seq);
+	struct netns_ipvs *ipvs = net_ipvs(net);
+
 	rcu_read_lock();
+	iter->t = rcu_dereference(ipvs->svc_table);
 	return *pos ? ip_vs_info_array(seq, *pos - 1) : SEQ_START_TOKEN;
 }
 
 
 static void *ip_vs_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 {
-	struct hlist_node *e;
-	struct ip_vs_iter *iter;
 	struct ip_vs_service *svc;
-	struct net *net = seq_file_net(seq);
-	struct netns_ipvs *ipvs = net_ipvs(net);
+	struct ip_vs_iter *iter;
+	struct hlist_bl_node *e;
+	struct ip_vs_rht *t;
 
 	++*pos;
 	if (v == SEQ_START_TOKEN)
@@ -2278,15 +2689,22 @@  static void *ip_vs_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 
 	svc = v;
 	iter = seq->private;
+	t = iter->t;
+	if (!t)
+		return NULL;
 
-	e = rcu_dereference(hlist_next_rcu(&svc->s_list));
-	if (e)
-		return hlist_entry(e, struct ip_vs_service, s_list);
+	hlist_bl_for_each_entry_continue_rcu(svc, e, s_list) {
+		/* Our cursor was moved to new table ? */
+		if (!ip_vs_rht_same_table(t, READ_ONCE(svc->hash_key)))
+			break;
+		return svc;
+	}
 
-	while (++iter->bucket < IP_VS_SVC_TAB_SIZE) {
-		hlist_for_each_entry_rcu(svc,
-					 &ipvs->svc_table[iter->bucket],
-					 s_list) {
+	while (++iter->bucket < t->size) {
+		hlist_bl_for_each_entry_rcu(svc, e, &t->buckets[iter->bucket],
+					    s_list) {
+			if (!ip_vs_rht_same_table(t, READ_ONCE(svc->hash_key)))
+				break;
 			return svc;
 		}
 	}
@@ -2767,13 +3185,18 @@  __ip_vs_get_service_entries(struct netns_ipvs *ipvs,
 			    const struct ip_vs_get_services *get,
 			    struct ip_vs_get_services __user *uptr)
 {
-	int idx, count=0;
-	struct ip_vs_service *svc;
 	struct ip_vs_service_entry entry;
+	DECLARE_IP_VS_RHT_WALK_BUCKETS();
+	struct hlist_bl_head *head;
+	struct ip_vs_service *svc;
+	struct hlist_bl_node *e;
+	int count = 0;
 	int ret = 0;
 
-	for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++) {
-		hlist_for_each_entry(svc, &ipvs->svc_table[idx], s_list) {
+	lockdep_assert_held(&ipvs->svc_resize_sem);
+	/* All service modifications are disabled, go ahead */
+	ip_vs_rht_walk_buckets(ipvs->svc_table, head) {
+		hlist_bl_for_each_entry(svc, e, head, s_list) {
 			/* Only expose IPv4 entries to old interface */
 			if (svc->af != AF_INET)
 				continue;
@@ -2945,6 +3368,35 @@  do_ip_vs_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
 		return ret;
 	}
 
+	if (cmd == IP_VS_SO_GET_SERVICES) {
+		struct ip_vs_get_services *get;
+		int size;
+
+		get = (struct ip_vs_get_services *)arg;
+		size = struct_size(get, entrytable, get->num_services);
+		if (*len != size) {
+			pr_err("length: %u != %u\n", *len, size);
+			return -EINVAL;
+		}
+		/* Protect against table resizer moving the entries.
+		 * Try reverse locking, so that we do not hold the mutex
+		 * while waiting for semaphore.
+		 */
+		while (1) {
+			ret = down_read_killable(&ipvs->svc_resize_sem);
+			if (ret < 0)
+				return ret;
+			if (mutex_trylock(&ipvs->service_mutex))
+				break;
+			up_read(&ipvs->svc_resize_sem);
+			cond_resched();
+		}
+		ret = __ip_vs_get_service_entries(ipvs, get, user);
+		up_read(&ipvs->svc_resize_sem);
+		mutex_unlock(&ipvs->service_mutex);
+		return ret;
+	}
+
 	mutex_lock(&ipvs->service_mutex);
 	switch (cmd) {
 	case IP_VS_SO_GET_VERSION:
@@ -2973,22 +3425,6 @@  do_ip_vs_get_ctl(struct sock *sk, int cmd, void __user *user, int *len)
 	}
 	break;
 
-	case IP_VS_SO_GET_SERVICES:
-	{
-		struct ip_vs_get_services *get;
-		int size;
-
-		get = (struct ip_vs_get_services *)arg;
-		size = struct_size(get, entrytable, get->num_services);
-		if (*len != size) {
-			pr_err("length: %u != %u\n", *len, size);
-			ret = -EINVAL;
-			goto out;
-		}
-		ret = __ip_vs_get_service_entries(ipvs, get, user);
-	}
-	break;
-
 	case IP_VS_SO_GET_SERVICE:
 	{
 		struct ip_vs_service_entry *entry;
@@ -3274,15 +3710,19 @@  static int ip_vs_genl_dump_service(struct sk_buff *skb,
 static int ip_vs_genl_dump_services(struct sk_buff *skb,
 				    struct netlink_callback *cb)
 {
-	int idx = 0, i;
-	int start = cb->args[0];
-	struct ip_vs_service *svc;
+	DECLARE_IP_VS_RHT_WALK_BUCKETS_SAFE_RCU();
 	struct net *net = sock_net(skb->sk);
 	struct netns_ipvs *ipvs = net_ipvs(net);
+	struct hlist_bl_head *head;
+	struct ip_vs_service *svc;
+	struct hlist_bl_node *e;
+	int start = cb->args[0];
+	int idx = 0;
 
+	down_read(&ipvs->svc_resize_sem);
 	rcu_read_lock();
-	for (i = 0; i < IP_VS_SVC_TAB_SIZE; i++) {
-		hlist_for_each_entry_rcu(svc, &ipvs->svc_table[i], s_list) {
+	ip_vs_rht_walk_buckets_safe_rcu(ipvs->svc_table, head) {
+		hlist_bl_for_each_entry_rcu(svc, e, head, s_list) {
 			if (++idx <= start)
 				continue;
 			if (ip_vs_genl_dump_service(skb, svc, cb) < 0) {
@@ -3294,6 +3734,7 @@  static int ip_vs_genl_dump_services(struct sk_buff *skb,
 
 nla_put_failure:
 	rcu_read_unlock();
+	up_read(&ipvs->svc_resize_sem);
 	cb->args[0] = idx;
 
 	return skb->len;
@@ -4292,8 +4733,10 @@  int __net_init ip_vs_control_net_init(struct netns_ipvs *ipvs)
 
 	/* Initialize service_mutex, svc_table per netns */
 	__mutex_init(&ipvs->service_mutex, "ipvs->service_mutex", &__ipvs_service_key);
-	for (idx = 0; idx < IP_VS_SVC_TAB_SIZE; idx++)
-		INIT_HLIST_HEAD(&ipvs->svc_table[idx]);
+	init_rwsem(&ipvs->svc_resize_sem);
+	INIT_DELAYED_WORK(&ipvs->svc_resize_work, svc_resize_work_handler);
+	atomic_set(&ipvs->svc_table_changes, 0);
+	RCU_INIT_POINTER(ipvs->svc_table, NULL);
 
 	/* Initialize rs_table */
 	for (idx = 0; idx < IP_VS_RTAB_SIZE; idx++)
@@ -4312,6 +4755,7 @@  int __net_init ip_vs_control_net_init(struct netns_ipvs *ipvs)
 	}
 
 	INIT_DELAYED_WORK(&ipvs->est_reload_work, est_reload_work_handler);
+	ipvs->sysctl_svc_lfactor = ip_vs_svc_default_load_factor(ipvs);
 
 	/* procfs stats */
 	ipvs->tot_stats = kzalloc(sizeof(*ipvs->tot_stats), GFP_KERNEL);