summaryrefslogtreecommitdiffstats
path: root/net/netfilter/ipvs
diff options
context:
space:
mode:
authorDavid S. Miller <davem@davemloft.net>2018-05-06 21:51:37 -0400
committerDavid S. Miller <davem@davemloft.net>2018-05-06 21:51:37 -0400
commit90278871d4b0da39c84fc9aa4929b0809dc7cf3c (patch)
treef1f5f261ec4d3e7e13c05b2f27e704395c8a8e98 /net/netfilter/ipvs
parent8fb11a9a8d51df9a314a6d970436963c127ff1bd (diff)
parentb13468dc577498002cf4e62978359ff97ffcd187 (diff)
downloadblackbird-obmc-linux-90278871d4b0da39c84fc9aa4929b0809dc7cf3c.tar.gz
blackbird-obmc-linux-90278871d4b0da39c84fc9aa4929b0809dc7cf3c.zip
Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
Pablo Neira Ayuso says: ==================== Netfilter/IPVS updates for net-next The following patchset contains Netfilter/IPVS updates for your net-next tree, more relevant updates in this batch are: 1) Add Maglev support to IPVS. Moreover, store lastest server weight in IPVS since this is needed by maglev, patches from from Inju Song. 2) Preparation works to add iptables flowtable support, patches from Felix Fietkau. 3) Hand over flows back to conntrack slow path in case of TCP RST/FIN packet is seen via new teardown state, also from Felix. 4) Add support for extended netlink error reporting for nf_tables. 5) Support for larger timeouts that 23 days in nf_tables, patch from Florian Westphal. 6) Always set an upper limit to dynamic sets, also from Florian. 7) Allow number generator to make map lookups, from Laura Garcia. 8) Use hash_32() instead of opencode hashing in IPVS, from Vicent Bernat. 9) Extend ip6tables SRH match to support previous, next and last SID, from Ahmed Abdelsalam. 10) Move Passive OS fingerprint nf_osf.c, from Fernando Fernandez. 11) Expose nf_conntrack_max through ctnetlink, from Florent Fourcot. 12) Several housekeeping patches for xt_NFLOG, x_tables and ebtables, from Taehee Yoo. 13) Unify meta bridge with core nft_meta, then make nft_meta built-in. Make rt and exthdr built-in too, again from Florian. 14) Missing initialization of tbl->entries in IPVS, from Cong Wang. ==================== Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'net/netfilter/ipvs')
-rw-r--r--net/netfilter/ipvs/Kconfig37
-rw-r--r--net/netfilter/ipvs/Makefile1
-rw-r--r--net/netfilter/ipvs/ip_vs_ctl.c4
-rw-r--r--net/netfilter/ipvs/ip_vs_dh.c3
-rw-r--r--net/netfilter/ipvs/ip_vs_lblc.c4
-rw-r--r--net/netfilter/ipvs/ip_vs_lblcr.c4
-rw-r--r--net/netfilter/ipvs/ip_vs_mh.c540
-rw-r--r--net/netfilter/ipvs/ip_vs_proto_tcp.c4
-rw-r--r--net/netfilter/ipvs/ip_vs_sh.c3
9 files changed, 594 insertions, 6 deletions
diff --git a/net/netfilter/ipvs/Kconfig b/net/netfilter/ipvs/Kconfig
index b32fb0dbe237..05dc1b77e466 100644
--- a/net/netfilter/ipvs/Kconfig
+++ b/net/netfilter/ipvs/Kconfig
@@ -225,6 +225,25 @@ config IP_VS_SH
If you want to compile it in kernel, say Y. To compile it as a
module, choose M here. If unsure, say N.
+config IP_VS_MH
+ tristate "maglev hashing scheduling"
+ ---help---
+ The maglev consistent hashing scheduling algorithm provides the
+ Google's Maglev hashing algorithm as a IPVS scheduler. It assigns
+ network connections to the servers through looking up a statically
+ assigned special hash table called the lookup table. Maglev hashing
+ is to assign a preference list of all the lookup table positions
+ to each destination.
+
+ Through this operation, The maglev hashing gives an almost equal
+ share of the lookup table to each of the destinations and provides
+ minimal disruption by using the lookup table. When the set of
+ destinations changes, a connection will likely be sent to the same
+ destination as it was before.
+
+ If you want to compile it in kernel, say Y. To compile it as a
+ module, choose M here. If unsure, say N.
+
config IP_VS_SED
tristate "shortest expected delay scheduling"
---help---
@@ -266,6 +285,24 @@ config IP_VS_SH_TAB_BITS
needs to be large enough to effectively fit all the destinations
multiplied by their respective weights.
+comment 'IPVS MH scheduler'
+
+config IP_VS_MH_TAB_INDEX
+ int "IPVS maglev hashing table index of size (the prime numbers)"
+ range 8 17
+ default 12
+ ---help---
+ The maglev hashing scheduler maps source IPs to destinations
+ stored in a hash table. This table is assigned by a preference
+ list of the positions to each destination until all slots in
+ the table are filled. The index determines the prime for size of
+ the table as 251, 509, 1021, 2039, 4093, 8191, 16381, 32749,
+ 65521 or 131071. When using weights to allow destinations to
+ receive more connections, the table is assigned an amount
+ proportional to the weights specified. The table needs to be large
+ enough to effectively fit all the destinations multiplied by their
+ respective weights.
+
comment 'IPVS application helper'
config IP_VS_FTP
diff --git a/net/netfilter/ipvs/Makefile b/net/netfilter/ipvs/Makefile
index c552993fa4b9..bfce2677fda2 100644
--- a/net/netfilter/ipvs/Makefile
+++ b/net/netfilter/ipvs/Makefile
@@ -33,6 +33,7 @@ obj-$(CONFIG_IP_VS_LBLC) += ip_vs_lblc.o
obj-$(CONFIG_IP_VS_LBLCR) += ip_vs_lblcr.o
obj-$(CONFIG_IP_VS_DH) += ip_vs_dh.o
obj-$(CONFIG_IP_VS_SH) += ip_vs_sh.o
+obj-$(CONFIG_IP_VS_MH) += ip_vs_mh.o
obj-$(CONFIG_IP_VS_SED) += ip_vs_sed.o
obj-$(CONFIG_IP_VS_NQ) += ip_vs_nq.o
diff --git a/net/netfilter/ipvs/ip_vs_ctl.c b/net/netfilter/ipvs/ip_vs_ctl.c
index f36098887ad0..d4f68d0f7df7 100644
--- a/net/netfilter/ipvs/ip_vs_ctl.c
+++ b/net/netfilter/ipvs/ip_vs_ctl.c
@@ -821,6 +821,10 @@ __ip_vs_update_dest(struct ip_vs_service *svc, struct ip_vs_dest *dest,
if (add && udest->af != svc->af)
ipvs->mixed_address_family_dests++;
+ /* keep the last_weight with latest non-0 weight */
+ if (add || udest->weight != 0)
+ atomic_set(&dest->last_weight, udest->weight);
+
/* set the weight and the flags */
atomic_set(&dest->weight, udest->weight);
conn_flags = udest->conn_flags & IP_VS_CONN_F_DEST_MASK;
diff --git a/net/netfilter/ipvs/ip_vs_dh.c b/net/netfilter/ipvs/ip_vs_dh.c
index 75f798f8e83b..07459e71d907 100644
--- a/net/netfilter/ipvs/ip_vs_dh.c
+++ b/net/netfilter/ipvs/ip_vs_dh.c
@@ -43,6 +43,7 @@
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/skbuff.h>
+#include <linux/hash.h>
#include <net/ip_vs.h>
@@ -81,7 +82,7 @@ static inline unsigned int ip_vs_dh_hashkey(int af, const union nf_inet_addr *ad
addr_fold = addr->ip6[0]^addr->ip6[1]^
addr->ip6[2]^addr->ip6[3];
#endif
- return (ntohl(addr_fold)*2654435761UL) & IP_VS_DH_TAB_MASK;
+ return hash_32(ntohl(addr_fold), IP_VS_DH_TAB_BITS);
}
diff --git a/net/netfilter/ipvs/ip_vs_lblc.c b/net/netfilter/ipvs/ip_vs_lblc.c
index 3057e453bf31..b9f375e6dc93 100644
--- a/net/netfilter/ipvs/ip_vs_lblc.c
+++ b/net/netfilter/ipvs/ip_vs_lblc.c
@@ -48,6 +48,7 @@
#include <linux/kernel.h>
#include <linux/skbuff.h>
#include <linux/jiffies.h>
+#include <linux/hash.h>
/* for sysctl */
#include <linux/fs.h>
@@ -160,7 +161,7 @@ ip_vs_lblc_hashkey(int af, const union nf_inet_addr *addr)
addr_fold = addr->ip6[0]^addr->ip6[1]^
addr->ip6[2]^addr->ip6[3];
#endif
- return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLC_TAB_MASK;
+ return hash_32(ntohl(addr_fold), IP_VS_LBLC_TAB_BITS);
}
@@ -371,6 +372,7 @@ static int ip_vs_lblc_init_svc(struct ip_vs_service *svc)
tbl->counter = 1;
tbl->dead = false;
tbl->svc = svc;
+ atomic_set(&tbl->entries, 0);
/*
* Hook periodic timer for garbage collection
diff --git a/net/netfilter/ipvs/ip_vs_lblcr.c b/net/netfilter/ipvs/ip_vs_lblcr.c
index 92adc04557ed..542c4949937a 100644
--- a/net/netfilter/ipvs/ip_vs_lblcr.c
+++ b/net/netfilter/ipvs/ip_vs_lblcr.c
@@ -47,6 +47,7 @@
#include <linux/jiffies.h>
#include <linux/list.h>
#include <linux/slab.h>
+#include <linux/hash.h>
/* for sysctl */
#include <linux/fs.h>
@@ -323,7 +324,7 @@ ip_vs_lblcr_hashkey(int af, const union nf_inet_addr *addr)
addr_fold = addr->ip6[0]^addr->ip6[1]^
addr->ip6[2]^addr->ip6[3];
#endif
- return (ntohl(addr_fold)*2654435761UL) & IP_VS_LBLCR_TAB_MASK;
+ return hash_32(ntohl(addr_fold), IP_VS_LBLCR_TAB_BITS);
}
@@ -534,6 +535,7 @@ static int ip_vs_lblcr_init_svc(struct ip_vs_service *svc)
tbl->counter = 1;
tbl->dead = false;
tbl->svc = svc;
+ atomic_set(&tbl->entries, 0);
/*
* Hook periodic timer for garbage collection
diff --git a/net/netfilter/ipvs/ip_vs_mh.c b/net/netfilter/ipvs/ip_vs_mh.c
new file mode 100644
index 000000000000..0f795b186eb3
--- /dev/null
+++ b/net/netfilter/ipvs/ip_vs_mh.c
@@ -0,0 +1,540 @@
+// SPDX-License-Identifier: GPL-2.0
+/* IPVS: Maglev Hashing scheduling module
+ *
+ * Authors: Inju Song <inju.song@navercorp.com>
+ *
+ */
+
+/* The mh algorithm is to assign a preference list of all the lookup
+ * table positions to each destination and populate the table with
+ * the most-preferred position of destinations. Then it is to select
+ * destination with the hash key of source IP address through looking
+ * up a the lookup table.
+ *
+ * The algorithm is detailed in:
+ * [3.4 Consistent Hasing]
+https://www.usenix.org/system/files/conference/nsdi16/nsdi16-paper-eisenbud.pdf
+ *
+ */
+
+#define KMSG_COMPONENT "IPVS"
+#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
+
+#include <linux/ip.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/skbuff.h>
+
+#include <net/ip_vs.h>
+
+#include <linux/siphash.h>
+#include <linux/bitops.h>
+#include <linux/gcd.h>
+
+#define IP_VS_SVC_F_SCHED_MH_FALLBACK IP_VS_SVC_F_SCHED1 /* MH fallback */
+#define IP_VS_SVC_F_SCHED_MH_PORT IP_VS_SVC_F_SCHED2 /* MH use port */
+
+struct ip_vs_mh_lookup {
+ struct ip_vs_dest __rcu *dest; /* real server (cache) */
+};
+
+struct ip_vs_mh_dest_setup {
+ unsigned int offset; /* starting offset */
+ unsigned int skip; /* skip */
+ unsigned int perm; /* next_offset */
+ int turns; /* weight / gcd() and rshift */
+};
+
+/* Available prime numbers for MH table */
+static int primes[] = {251, 509, 1021, 2039, 4093,
+ 8191, 16381, 32749, 65521, 131071};
+
+/* For IPVS MH entry hash table */
+#ifndef CONFIG_IP_VS_MH_TAB_INDEX
+#define CONFIG_IP_VS_MH_TAB_INDEX 12
+#endif
+#define IP_VS_MH_TAB_BITS (CONFIG_IP_VS_MH_TAB_INDEX / 2)
+#define IP_VS_MH_TAB_INDEX (CONFIG_IP_VS_MH_TAB_INDEX - 8)
+#define IP_VS_MH_TAB_SIZE primes[IP_VS_MH_TAB_INDEX]
+
+struct ip_vs_mh_state {
+ struct rcu_head rcu_head;
+ struct ip_vs_mh_lookup *lookup;
+ struct ip_vs_mh_dest_setup *dest_setup;
+ hsiphash_key_t hash1, hash2;
+ int gcd;
+ int rshift;
+};
+
+static inline void generate_hash_secret(hsiphash_key_t *hash1,
+ hsiphash_key_t *hash2)
+{
+ hash1->key[0] = 2654435761UL;
+ hash1->key[1] = 2654435761UL;
+
+ hash2->key[0] = 2654446892UL;
+ hash2->key[1] = 2654446892UL;
+}
+
+/* Helper function to determine if server is unavailable */
+static inline bool is_unavailable(struct ip_vs_dest *dest)
+{
+ return atomic_read(&dest->weight) <= 0 ||
+ dest->flags & IP_VS_DEST_F_OVERLOAD;
+}
+
+/* Returns hash value for IPVS MH entry */
+static inline unsigned int
+ip_vs_mh_hashkey(int af, const union nf_inet_addr *addr,
+ __be16 port, hsiphash_key_t *key, unsigned int offset)
+{
+ unsigned int v;
+ __be32 addr_fold = addr->ip;
+
+#ifdef CONFIG_IP_VS_IPV6
+ if (af == AF_INET6)
+ addr_fold = addr->ip6[0] ^ addr->ip6[1] ^
+ addr->ip6[2] ^ addr->ip6[3];
+#endif
+ v = (offset + ntohs(port) + ntohl(addr_fold));
+ return hsiphash(&v, sizeof(v), key);
+}
+
+/* Reset all the hash buckets of the specified table. */
+static void ip_vs_mh_reset(struct ip_vs_mh_state *s)
+{
+ int i;
+ struct ip_vs_mh_lookup *l;
+ struct ip_vs_dest *dest;
+
+ l = &s->lookup[0];
+ for (i = 0; i < IP_VS_MH_TAB_SIZE; i++) {
+ dest = rcu_dereference_protected(l->dest, 1);
+ if (dest) {
+ ip_vs_dest_put(dest);
+ RCU_INIT_POINTER(l->dest, NULL);
+ }
+ l++;
+ }
+}
+
+static int ip_vs_mh_permutate(struct ip_vs_mh_state *s,
+ struct ip_vs_service *svc)
+{
+ struct list_head *p;
+ struct ip_vs_mh_dest_setup *ds;
+ struct ip_vs_dest *dest;
+ int lw;
+
+ /* If gcd is smaller then 1, number of dests or
+ * all last_weight of dests are zero. So, skip
+ * permutation for the dests.
+ */
+ if (s->gcd < 1)
+ return 0;
+
+ /* Set dest_setup for the dests permutation */
+ p = &svc->destinations;
+ ds = &s->dest_setup[0];
+ while ((p = p->next) != &svc->destinations) {
+ dest = list_entry(p, struct ip_vs_dest, n_list);
+
+ ds->offset = ip_vs_mh_hashkey(svc->af, &dest->addr,
+ dest->port, &s->hash1, 0) %
+ IP_VS_MH_TAB_SIZE;
+ ds->skip = ip_vs_mh_hashkey(svc->af, &dest->addr,
+ dest->port, &s->hash2, 0) %
+ (IP_VS_MH_TAB_SIZE - 1) + 1;
+ ds->perm = ds->offset;
+
+ lw = atomic_read(&dest->last_weight);
+ ds->turns = ((lw / s->gcd) >> s->rshift) ? : (lw != 0);
+ ds++;
+ }
+
+ return 0;
+}
+
+static int ip_vs_mh_populate(struct ip_vs_mh_state *s,
+ struct ip_vs_service *svc)
+{
+ int n, c, dt_count;
+ unsigned long *table;
+ struct list_head *p;
+ struct ip_vs_mh_dest_setup *ds;
+ struct ip_vs_dest *dest, *new_dest;
+
+ /* If gcd is smaller then 1, number of dests or
+ * all last_weight of dests are zero. So, skip
+ * the population for the dests and reset lookup table.
+ */
+ if (s->gcd < 1) {
+ ip_vs_mh_reset(s);
+ return 0;
+ }
+
+ table = kcalloc(BITS_TO_LONGS(IP_VS_MH_TAB_SIZE),
+ sizeof(unsigned long), GFP_KERNEL);
+ if (!table)
+ return -ENOMEM;
+
+ p = &svc->destinations;
+ n = 0;
+ dt_count = 0;
+ while (n < IP_VS_MH_TAB_SIZE) {
+ if (p == &svc->destinations)
+ p = p->next;
+
+ ds = &s->dest_setup[0];
+ while (p != &svc->destinations) {
+ /* Ignore added server with zero weight */
+ if (ds->turns < 1) {
+ p = p->next;
+ ds++;
+ continue;
+ }
+
+ c = ds->perm;
+ while (test_bit(c, table)) {
+ /* Add skip, mod IP_VS_MH_TAB_SIZE */
+ ds->perm += ds->skip;
+ if (ds->perm >= IP_VS_MH_TAB_SIZE)
+ ds->perm -= IP_VS_MH_TAB_SIZE;
+ c = ds->perm;
+ }
+
+ __set_bit(c, table);
+
+ dest = rcu_dereference_protected(s->lookup[c].dest, 1);
+ new_dest = list_entry(p, struct ip_vs_dest, n_list);
+ if (dest != new_dest) {
+ if (dest)
+ ip_vs_dest_put(dest);
+ ip_vs_dest_hold(new_dest);
+ RCU_INIT_POINTER(s->lookup[c].dest, new_dest);
+ }
+
+ if (++n == IP_VS_MH_TAB_SIZE)
+ goto out;
+
+ if (++dt_count >= ds->turns) {
+ dt_count = 0;
+ p = p->next;
+ ds++;
+ }
+ }
+ }
+
+out:
+ kfree(table);
+ return 0;
+}
+
+/* Get ip_vs_dest associated with supplied parameters. */
+static inline struct ip_vs_dest *
+ip_vs_mh_get(struct ip_vs_service *svc, struct ip_vs_mh_state *s,
+ const union nf_inet_addr *addr, __be16 port)
+{
+ unsigned int hash = ip_vs_mh_hashkey(svc->af, addr, port, &s->hash1, 0)
+ % IP_VS_MH_TAB_SIZE;
+ struct ip_vs_dest *dest = rcu_dereference(s->lookup[hash].dest);
+
+ return (!dest || is_unavailable(dest)) ? NULL : dest;
+}
+
+/* As ip_vs_mh_get, but with fallback if selected server is unavailable */
+static inline struct ip_vs_dest *
+ip_vs_mh_get_fallback(struct ip_vs_service *svc, struct ip_vs_mh_state *s,
+ const union nf_inet_addr *addr, __be16 port)
+{
+ unsigned int offset, roffset;
+ unsigned int hash, ihash;
+ struct ip_vs_dest *dest;
+
+ /* First try the dest it's supposed to go to */
+ ihash = ip_vs_mh_hashkey(svc->af, addr, port,
+ &s->hash1, 0) % IP_VS_MH_TAB_SIZE;
+ dest = rcu_dereference(s->lookup[ihash].dest);
+ if (!dest)
+ return NULL;
+ if (!is_unavailable(dest))
+ return dest;
+
+ IP_VS_DBG_BUF(6, "MH: selected unavailable server %s:%u, reselecting",
+ IP_VS_DBG_ADDR(dest->af, &dest->addr), ntohs(dest->port));
+
+ /* If the original dest is unavailable, loop around the table
+ * starting from ihash to find a new dest
+ */
+ for (offset = 0; offset < IP_VS_MH_TAB_SIZE; offset++) {
+ roffset = (offset + ihash) % IP_VS_MH_TAB_SIZE;
+ hash = ip_vs_mh_hashkey(svc->af, addr, port, &s->hash1,
+ roffset) % IP_VS_MH_TAB_SIZE;
+ dest = rcu_dereference(s->lookup[hash].dest);
+ if (!dest)
+ break;
+ if (!is_unavailable(dest))
+ return dest;
+ IP_VS_DBG_BUF(6,
+ "MH: selected unavailable server %s:%u (offset %u), reselecting",
+ IP_VS_DBG_ADDR(dest->af, &dest->addr),
+ ntohs(dest->port), roffset);
+ }
+
+ return NULL;
+}
+
+/* Assign all the hash buckets of the specified table with the service. */
+static int ip_vs_mh_reassign(struct ip_vs_mh_state *s,
+ struct ip_vs_service *svc)
+{
+ int ret;
+
+ if (svc->num_dests > IP_VS_MH_TAB_SIZE)
+ return -EINVAL;
+
+ if (svc->num_dests >= 1) {
+ s->dest_setup = kcalloc(svc->num_dests,
+ sizeof(struct ip_vs_mh_dest_setup),
+ GFP_KERNEL);
+ if (!s->dest_setup)
+ return -ENOMEM;
+ }
+
+ ip_vs_mh_permutate(s, svc);
+
+ ret = ip_vs_mh_populate(s, svc);
+ if (ret < 0)
+ goto out;
+
+ IP_VS_DBG_BUF(6, "MH: reassign lookup table of %s:%u\n",
+ IP_VS_DBG_ADDR(svc->af, &svc->addr),
+ ntohs(svc->port));
+
+out:
+ if (svc->num_dests >= 1) {
+ kfree(s->dest_setup);
+ s->dest_setup = NULL;
+ }
+ return ret;
+}
+
+static int ip_vs_mh_gcd_weight(struct ip_vs_service *svc)
+{
+ struct ip_vs_dest *dest;
+ int weight;
+ int g = 0;
+
+ list_for_each_entry(dest, &svc->destinations, n_list) {
+ weight = atomic_read(&dest->last_weight);
+ if (weight > 0) {
+ if (g > 0)
+ g = gcd(weight, g);
+ else
+ g = weight;
+ }
+ }
+ return g;
+}
+
+/* To avoid assigning huge weight for the MH table,
+ * calculate shift value with gcd.
+ */
+static int ip_vs_mh_shift_weight(struct ip_vs_service *svc, int gcd)
+{
+ struct ip_vs_dest *dest;
+ int new_weight, weight = 0;
+ int mw, shift;
+
+ /* If gcd is smaller then 1, number of dests or
+ * all last_weight of dests are zero. So, return
+ * shift value as zero.
+ */
+ if (gcd < 1)
+ return 0;
+
+ list_for_each_entry(dest, &svc->destinations, n_list) {
+ new_weight = atomic_read(&dest->last_weight);
+ if (new_weight > weight)
+ weight = new_weight;
+ }
+
+ /* Because gcd is greater than zero,
+ * the maximum weight and gcd are always greater than zero
+ */
+ mw = weight / gcd;
+
+ /* shift = occupied bits of weight/gcd - MH highest bits */
+ shift = fls(mw) - IP_VS_MH_TAB_BITS;
+ return (shift >= 0) ? shift : 0;
+}
+
+static void ip_vs_mh_state_free(struct rcu_head *head)
+{
+ struct ip_vs_mh_state *s;
+
+ s = container_of(head, struct ip_vs_mh_state, rcu_head);
+ kfree(s->lookup);
+ kfree(s);
+}
+
+static int ip_vs_mh_init_svc(struct ip_vs_service *svc)
+{
+ int ret;
+ struct ip_vs_mh_state *s;
+
+ /* Allocate the MH table for this service */
+ s = kzalloc(sizeof(*s), GFP_KERNEL);
+ if (!s)
+ return -ENOMEM;
+
+ s->lookup = kcalloc(IP_VS_MH_TAB_SIZE, sizeof(struct ip_vs_mh_lookup),
+ GFP_KERNEL);
+ if (!s->lookup) {
+ kfree(s);
+ return -ENOMEM;
+ }
+
+ generate_hash_secret(&s->hash1, &s->hash2);
+ s->gcd = ip_vs_mh_gcd_weight(svc);
+ s->rshift = ip_vs_mh_shift_weight(svc, s->gcd);
+
+ IP_VS_DBG(6,
+ "MH lookup table (memory=%zdbytes) allocated for current service\n",
+ sizeof(struct ip_vs_mh_lookup) * IP_VS_MH_TAB_SIZE);
+
+ /* Assign the lookup table with current dests */
+ ret = ip_vs_mh_reassign(s, svc);
+ if (ret < 0) {
+ ip_vs_mh_reset(s);
+ ip_vs_mh_state_free(&s->rcu_head);
+ return ret;
+ }
+
+ /* No more failures, attach state */
+ svc->sched_data = s;
+ return 0;
+}
+
+static void ip_vs_mh_done_svc(struct ip_vs_service *svc)
+{
+ struct ip_vs_mh_state *s = svc->sched_data;
+
+ /* Got to clean up lookup entry here */
+ ip_vs_mh_reset(s);
+
+ call_rcu(&s->rcu_head, ip_vs_mh_state_free);
+ IP_VS_DBG(6, "MH lookup table (memory=%zdbytes) released\n",
+ sizeof(struct ip_vs_mh_lookup) * IP_VS_MH_TAB_SIZE);
+}
+
+static int ip_vs_mh_dest_changed(struct ip_vs_service *svc,
+ struct ip_vs_dest *dest)
+{
+ struct ip_vs_mh_state *s = svc->sched_data;
+
+ s->gcd = ip_vs_mh_gcd_weight(svc);
+ s->rshift = ip_vs_mh_shift_weight(svc, s->gcd);
+
+ /* Assign the lookup table with the updated service */
+ return ip_vs_mh_reassign(s, svc);
+}
+
+/* Helper function to get port number */
+static inline __be16
+ip_vs_mh_get_port(const struct sk_buff *skb, struct ip_vs_iphdr *iph)
+{
+ __be16 _ports[2], *ports;
+
+ /* At this point we know that we have a valid packet of some kind.
+ * Because ICMP packets are only guaranteed to have the first 8
+ * bytes, let's just grab the ports. Fortunately they're in the
+ * same position for all three of the protocols we care about.
+ */
+ switch (iph->protocol) {
+ case IPPROTO_TCP:
+ case IPPROTO_UDP:
+ case IPPROTO_SCTP:
+ ports = skb_header_pointer(skb, iph->len, sizeof(_ports),
+ &_ports);
+ if (unlikely(!ports))
+ return 0;
+
+ if (likely(!ip_vs_iph_inverse(iph)))
+ return ports[0];
+ else
+ return ports[1];
+ default:
+ return 0;
+ }
+}
+
+/* Maglev Hashing scheduling */
+static struct ip_vs_dest *
+ip_vs_mh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb,
+ struct ip_vs_iphdr *iph)
+{
+ struct ip_vs_dest *dest;
+ struct ip_vs_mh_state *s;
+ __be16 port = 0;
+ const union nf_inet_addr *hash_addr;
+
+ hash_addr = ip_vs_iph_inverse(iph) ? &iph->daddr : &iph->saddr;
+
+ IP_VS_DBG(6, "%s : Scheduling...\n", __func__);
+
+ if (svc->flags & IP_VS_SVC_F_SCHED_MH_PORT)
+ port = ip_vs_mh_get_port(skb, iph);
+
+ s = (struct ip_vs_mh_state *)svc->sched_data;
+
+ if (svc->flags & IP_VS_SVC_F_SCHED_MH_FALLBACK)
+ dest = ip_vs_mh_get_fallback(svc, s, hash_addr, port);
+ else
+ dest = ip_vs_mh_get(svc, s, hash_addr, port);
+
+ if (!dest) {
+ ip_vs_scheduler_err(svc, "no destination available");
+ return NULL;
+ }
+
+ IP_VS_DBG_BUF(6, "MH: source IP address %s:%u --> server %s:%u\n",
+ IP_VS_DBG_ADDR(svc->af, hash_addr),
+ ntohs(port),
+ IP_VS_DBG_ADDR(dest->af, &dest->addr),
+ ntohs(dest->port));
+
+ return dest;
+}
+
+/* IPVS MH Scheduler structure */
+static struct ip_vs_scheduler ip_vs_mh_scheduler = {
+ .name = "mh",
+ .refcnt = ATOMIC_INIT(0),
+ .module = THIS_MODULE,
+ .n_list = LIST_HEAD_INIT(ip_vs_mh_scheduler.n_list),
+ .init_service = ip_vs_mh_init_svc,
+ .done_service = ip_vs_mh_done_svc,
+ .add_dest = ip_vs_mh_dest_changed,
+ .del_dest = ip_vs_mh_dest_changed,
+ .upd_dest = ip_vs_mh_dest_changed,
+ .schedule = ip_vs_mh_schedule,
+};
+
+static int __init ip_vs_mh_init(void)
+{
+ return register_ip_vs_scheduler(&ip_vs_mh_scheduler);
+}
+
+static void __exit ip_vs_mh_cleanup(void)
+{
+ unregister_ip_vs_scheduler(&ip_vs_mh_scheduler);
+ rcu_barrier();
+}
+
+module_init(ip_vs_mh_init);
+module_exit(ip_vs_mh_cleanup);
+MODULE_DESCRIPTION("Maglev hashing ipvs scheduler");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Inju Song <inju.song@navercorp.com>");
diff --git a/net/netfilter/ipvs/ip_vs_proto_tcp.c b/net/netfilter/ipvs/ip_vs_proto_tcp.c
index bcd9b7bde4ee..569631d2b2a1 100644
--- a/net/netfilter/ipvs/ip_vs_proto_tcp.c
+++ b/net/netfilter/ipvs/ip_vs_proto_tcp.c
@@ -436,7 +436,7 @@ static bool tcp_state_active(int state)
return tcp_state_active_table[state];
}
-static struct tcp_states_t tcp_states [] = {
+static struct tcp_states_t tcp_states[] = {
/* INPUT */
/* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
/*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSR }},
@@ -459,7 +459,7 @@ static struct tcp_states_t tcp_states [] = {
/*rst*/ {{sCL, sCL, sCL, sSR, sCL, sCL, sCL, sCL, sLA, sLI, sCL }},
};
-static struct tcp_states_t tcp_states_dos [] = {
+static struct tcp_states_t tcp_states_dos[] = {
/* INPUT */
/* sNO, sES, sSS, sSR, sFW, sTW, sCL, sCW, sLA, sLI, sSA */
/*syn*/ {{sSR, sES, sES, sSR, sSR, sSR, sSR, sSR, sSR, sSR, sSA }},
diff --git a/net/netfilter/ipvs/ip_vs_sh.c b/net/netfilter/ipvs/ip_vs_sh.c
index 16aaac6eedc9..1e01c782583a 100644
--- a/net/netfilter/ipvs/ip_vs_sh.c
+++ b/net/netfilter/ipvs/ip_vs_sh.c
@@ -96,7 +96,8 @@ ip_vs_sh_hashkey(int af, const union nf_inet_addr *addr,
addr_fold = addr->ip6[0]^addr->ip6[1]^
addr->ip6[2]^addr->ip6[3];
#endif
- return (offset + (ntohs(port) + ntohl(addr_fold))*2654435761UL) &
+ return (offset + hash_32(ntohs(port) + ntohl(addr_fold),
+ IP_VS_SH_TAB_BITS)) &
IP_VS_SH_TAB_MASK;
}
OpenPOWER on IntegriCloud