2 * Copyright (c) 2007-2012 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
58 #include "vport-internal_dev.h"
61 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
62 static void rehash_flow_table(struct work_struct *work);
63 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
65 int ovs_net_id __read_mostly;
67 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
68 struct genl_multicast_group *grp)
70 genl_notify(skb, genl_info_net(info), info->snd_portid,
71 grp->id, info->nlhdr, GFP_KERNEL);
77 * All writes e.g. Writes to device state (add/remove datapath, port, set
78 * operations on vports, etc.), Writes to other state (flow table
79 * modifications, set miscellaneous datapath parameters, etc.) are protected
82 * Reads are protected by RCU.
84 * There are a few special cases (mostly stats) that have their own
85 * synchronization but they nest under all of above and don't interact with
88 * The RTNL lock nests inside ovs_mutex.
91 static DEFINE_MUTEX(ovs_mutex);
95 mutex_lock(&ovs_mutex);
100 mutex_unlock(&ovs_mutex);
103 #ifdef CONFIG_LOCKDEP
104 int lockdep_ovsl_is_held(void)
107 return lockdep_is_held(&ovs_mutex);
113 static struct vport *new_vport(const struct vport_parms *);
114 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
115 const struct dp_upcall_info *);
116 static int queue_userspace_packet(struct net *, int dp_ifindex,
118 const struct dp_upcall_info *);
120 /* Must be called with rcu_read_lock or ovs_mutex. */
121 static struct datapath *get_dp(struct net *net, int dp_ifindex)
123 struct datapath *dp = NULL;
124 struct net_device *dev;
127 dev = dev_get_by_index_rcu(net, dp_ifindex);
129 struct vport *vport = ovs_internal_dev_get_vport(dev);
138 /* Must be called with rcu_read_lock or ovs_mutex. */
139 const char *ovs_dp_name(const struct datapath *dp)
141 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
142 return vport->ops->get_name(vport);
145 static int get_dpifindex(struct datapath *dp)
152 local = ovs_vport_rcu(dp, OVSP_LOCAL);
154 ifindex = local->ops->get_ifindex(local);
163 static void destroy_dp_rcu(struct rcu_head *rcu)
165 struct datapath *dp = container_of(rcu, struct datapath, rcu);
167 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
168 free_percpu(dp->stats_percpu);
169 release_net(ovs_dp_get_net(dp));
174 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
177 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
180 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
183 struct hlist_head *head;
185 head = vport_hash_bucket(dp, port_no);
186 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
187 if (vport->port_no == port_no)
193 /* Called with ovs_mutex. */
194 static struct vport *new_vport(const struct vport_parms *parms)
198 vport = ovs_vport_add(parms);
199 if (!IS_ERR(vport)) {
200 struct datapath *dp = parms->dp;
201 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
203 hlist_add_head_rcu(&vport->dp_hash_node, head);
208 void ovs_dp_detach_port(struct vport *p)
212 /* First drop references to device. */
213 hlist_del_rcu(&p->dp_hash_node);
215 /* Then destroy it. */
219 /* Must be called with rcu_read_lock. */
220 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
222 struct datapath *dp = p->dp;
223 struct sw_flow *flow;
224 struct dp_stats_percpu *stats;
225 struct sw_flow_key key;
230 stats = this_cpu_ptr(dp->stats_percpu);
232 /* Extract flow from 'skb' into 'key'. */
233 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
234 if (unlikely(error)) {
240 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
241 if (unlikely(!flow)) {
242 struct dp_upcall_info upcall;
244 upcall.cmd = OVS_PACKET_CMD_MISS;
246 upcall.userdata = NULL;
247 upcall.portid = p->upcall_portid;
248 ovs_dp_upcall(dp, skb, &upcall);
250 stats_counter = &stats->n_missed;
254 OVS_CB(skb)->flow = flow;
256 stats_counter = &stats->n_hit;
257 ovs_flow_used(OVS_CB(skb)->flow, skb);
258 ovs_execute_actions(dp, skb);
261 /* Update datapath statistics. */
262 u64_stats_update_begin(&stats->sync);
264 u64_stats_update_end(&stats->sync);
267 static struct genl_family dp_packet_genl_family = {
268 .id = GENL_ID_GENERATE,
269 .hdrsize = sizeof(struct ovs_header),
270 .name = OVS_PACKET_FAMILY,
271 .version = OVS_PACKET_VERSION,
272 .maxattr = OVS_PACKET_ATTR_MAX,
276 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
277 const struct dp_upcall_info *upcall_info)
279 struct dp_stats_percpu *stats;
283 if (upcall_info->portid == 0) {
288 dp_ifindex = get_dpifindex(dp);
294 if (!skb_is_gso(skb))
295 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
297 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
304 stats = this_cpu_ptr(dp->stats_percpu);
306 u64_stats_update_begin(&stats->sync);
308 u64_stats_update_end(&stats->sync);
313 static int queue_gso_packets(struct net *net, int dp_ifindex,
315 const struct dp_upcall_info *upcall_info)
317 unsigned short gso_type = skb_shinfo(skb)->gso_type;
318 struct dp_upcall_info later_info;
319 struct sw_flow_key later_key;
320 struct sk_buff *segs, *nskb;
323 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
325 return PTR_ERR(segs);
327 /* Queue all of the segments. */
330 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
334 if (skb == segs && gso_type & SKB_GSO_UDP) {
335 /* The initial flow key extracted by ovs_flow_extract()
336 * in this case is for a first fragment, so we need to
337 * properly mark later fragments.
339 later_key = *upcall_info->key;
340 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
342 later_info = *upcall_info;
343 later_info.key = &later_key;
344 upcall_info = &later_info;
346 } while ((skb = skb->next));
348 /* Free all of the segments. */
356 } while ((skb = nskb));
360 static size_t key_attr_size(void)
362 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
363 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
364 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
365 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
366 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
367 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
368 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
369 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
370 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
371 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
372 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
375 static size_t upcall_msg_size(const struct sk_buff *skb,
376 const struct nlattr *userdata)
378 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
379 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
380 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
382 /* OVS_PACKET_ATTR_USERDATA */
384 size += NLA_ALIGN(userdata->nla_len);
389 static int queue_userspace_packet(struct net *net, int dp_ifindex,
391 const struct dp_upcall_info *upcall_info)
393 struct ovs_header *upcall;
394 struct sk_buff *nskb = NULL;
395 struct sk_buff *user_skb; /* to be queued to userspace */
399 if (vlan_tx_tag_present(skb)) {
400 nskb = skb_clone(skb, GFP_ATOMIC);
404 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
412 if (nla_attr_size(skb->len) > USHRT_MAX) {
417 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
423 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
424 0, upcall_info->cmd);
425 upcall->dp_ifindex = dp_ifindex;
427 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
428 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
429 nla_nest_end(user_skb, nla);
431 if (upcall_info->userdata)
432 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
433 nla_len(upcall_info->userdata),
434 nla_data(upcall_info->userdata));
436 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
438 skb_copy_and_csum_dev(skb, nla_data(nla));
440 genlmsg_end(user_skb, upcall);
441 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
448 /* Called with ovs_mutex. */
449 static int flush_flows(struct datapath *dp)
451 struct flow_table *old_table;
452 struct flow_table *new_table;
454 old_table = ovsl_dereference(dp->table);
455 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
459 rcu_assign_pointer(dp->table, new_table);
461 ovs_flow_tbl_deferred_destroy(old_table);
465 static int validate_actions(const struct nlattr *attr,
466 const struct sw_flow_key *key, int depth);
468 static int validate_sample(const struct nlattr *attr,
469 const struct sw_flow_key *key, int depth)
471 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
472 const struct nlattr *probability, *actions;
473 const struct nlattr *a;
476 memset(attrs, 0, sizeof(attrs));
477 nla_for_each_nested(a, attr, rem) {
478 int type = nla_type(a);
479 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
486 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
487 if (!probability || nla_len(probability) != sizeof(u32))
490 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
491 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
493 return validate_actions(actions, key, depth + 1);
496 static int validate_tp_port(const struct sw_flow_key *flow_key)
498 if (flow_key->eth.type == htons(ETH_P_IP)) {
499 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
501 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
502 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
509 static int validate_set(const struct nlattr *a,
510 const struct sw_flow_key *flow_key)
512 const struct nlattr *ovs_key = nla_data(a);
513 int key_type = nla_type(ovs_key);
515 /* There can be only one key in a action */
516 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
519 if (key_type > OVS_KEY_ATTR_MAX ||
520 nla_len(ovs_key) != ovs_key_lens[key_type])
524 const struct ovs_key_ipv4 *ipv4_key;
525 const struct ovs_key_ipv6 *ipv6_key;
527 case OVS_KEY_ATTR_PRIORITY:
528 case OVS_KEY_ATTR_SKB_MARK:
529 case OVS_KEY_ATTR_ETHERNET:
532 case OVS_KEY_ATTR_IPV4:
533 if (flow_key->eth.type != htons(ETH_P_IP))
536 if (!flow_key->ip.proto)
539 ipv4_key = nla_data(ovs_key);
540 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
543 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
548 case OVS_KEY_ATTR_IPV6:
549 if (flow_key->eth.type != htons(ETH_P_IPV6))
552 if (!flow_key->ip.proto)
555 ipv6_key = nla_data(ovs_key);
556 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
559 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
562 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
567 case OVS_KEY_ATTR_TCP:
568 if (flow_key->ip.proto != IPPROTO_TCP)
571 return validate_tp_port(flow_key);
573 case OVS_KEY_ATTR_UDP:
574 if (flow_key->ip.proto != IPPROTO_UDP)
577 return validate_tp_port(flow_key);
586 static int validate_userspace(const struct nlattr *attr)
588 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
589 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
590 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
592 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
595 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
596 attr, userspace_policy);
600 if (!a[OVS_USERSPACE_ATTR_PID] ||
601 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
607 static int validate_actions(const struct nlattr *attr,
608 const struct sw_flow_key *key, int depth)
610 const struct nlattr *a;
613 if (depth >= SAMPLE_ACTION_DEPTH)
616 nla_for_each_nested(a, attr, rem) {
617 /* Expected argument lengths, (u32)-1 for variable length. */
618 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
619 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
620 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
621 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
622 [OVS_ACTION_ATTR_POP_VLAN] = 0,
623 [OVS_ACTION_ATTR_SET] = (u32)-1,
624 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
626 const struct ovs_action_push_vlan *vlan;
627 int type = nla_type(a);
629 if (type > OVS_ACTION_ATTR_MAX ||
630 (action_lens[type] != nla_len(a) &&
631 action_lens[type] != (u32)-1))
635 case OVS_ACTION_ATTR_UNSPEC:
638 case OVS_ACTION_ATTR_USERSPACE:
639 err = validate_userspace(a);
644 case OVS_ACTION_ATTR_OUTPUT:
645 if (nla_get_u32(a) >= DP_MAX_PORTS)
650 case OVS_ACTION_ATTR_POP_VLAN:
653 case OVS_ACTION_ATTR_PUSH_VLAN:
655 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
657 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
661 case OVS_ACTION_ATTR_SET:
662 err = validate_set(a, key);
667 case OVS_ACTION_ATTR_SAMPLE:
668 err = validate_sample(a, key, depth);
684 static void clear_stats(struct sw_flow *flow)
688 flow->packet_count = 0;
689 flow->byte_count = 0;
692 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
694 struct ovs_header *ovs_header = info->userhdr;
695 struct nlattr **a = info->attrs;
696 struct sw_flow_actions *acts;
697 struct sk_buff *packet;
698 struct sw_flow *flow;
706 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
707 !a[OVS_PACKET_ATTR_ACTIONS])
710 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
711 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
715 skb_reserve(packet, NET_IP_ALIGN);
717 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
719 skb_reset_mac_header(packet);
720 eth = eth_hdr(packet);
722 /* Normally, setting the skb 'protocol' field would be handled by a
723 * call to eth_type_trans(), but it assumes there's a sending
724 * device, which we may not have. */
725 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
726 packet->protocol = eth->h_proto;
728 packet->protocol = htons(ETH_P_802_2);
730 /* Build an sw_flow for sending this packet. */
731 flow = ovs_flow_alloc();
736 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
740 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
741 &flow->key.phy.skb_mark,
742 &flow->key.phy.in_port,
743 a[OVS_PACKET_ATTR_KEY]);
747 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
751 flow->hash = ovs_flow_hash(&flow->key, key_len);
753 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
757 rcu_assign_pointer(flow->sf_acts, acts);
759 OVS_CB(packet)->flow = flow;
760 packet->priority = flow->key.phy.priority;
761 packet->mark = flow->key.phy.skb_mark;
764 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
770 err = ovs_execute_actions(dp, packet);
787 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
788 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
789 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
790 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
793 static struct genl_ops dp_packet_genl_ops[] = {
794 { .cmd = OVS_PACKET_CMD_EXECUTE,
795 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
796 .policy = packet_policy,
797 .doit = ovs_packet_cmd_execute
801 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
804 struct flow_table *table = ovsl_dereference(dp->table);
806 stats->n_flows = ovs_flow_tbl_count(table);
808 stats->n_hit = stats->n_missed = stats->n_lost = 0;
809 for_each_possible_cpu(i) {
810 const struct dp_stats_percpu *percpu_stats;
811 struct dp_stats_percpu local_stats;
814 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
817 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
818 local_stats = *percpu_stats;
819 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
821 stats->n_hit += local_stats.n_hit;
822 stats->n_missed += local_stats.n_missed;
823 stats->n_lost += local_stats.n_lost;
827 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
828 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
829 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
830 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
833 static struct genl_family dp_flow_genl_family = {
834 .id = GENL_ID_GENERATE,
835 .hdrsize = sizeof(struct ovs_header),
836 .name = OVS_FLOW_FAMILY,
837 .version = OVS_FLOW_VERSION,
838 .maxattr = OVS_FLOW_ATTR_MAX,
842 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
843 .name = OVS_FLOW_MCGROUP
846 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
848 return NLMSG_ALIGN(sizeof(struct ovs_header))
849 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
850 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
851 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
852 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
853 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
856 /* Called with ovs_mutex. */
857 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
858 struct sk_buff *skb, u32 portid,
859 u32 seq, u32 flags, u8 cmd)
861 const int skb_orig_len = skb->len;
862 const struct sw_flow_actions *sf_acts;
863 struct ovs_flow_stats stats;
864 struct ovs_header *ovs_header;
870 sf_acts = ovsl_dereference(flow->sf_acts);
872 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
876 ovs_header->dp_ifindex = get_dpifindex(dp);
878 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
880 goto nla_put_failure;
881 err = ovs_flow_to_nlattrs(&flow->key, skb);
884 nla_nest_end(skb, nla);
886 spin_lock_bh(&flow->lock);
888 stats.n_packets = flow->packet_count;
889 stats.n_bytes = flow->byte_count;
890 tcp_flags = flow->tcp_flags;
891 spin_unlock_bh(&flow->lock);
894 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
895 goto nla_put_failure;
897 if (stats.n_packets &&
898 nla_put(skb, OVS_FLOW_ATTR_STATS,
899 sizeof(struct ovs_flow_stats), &stats))
900 goto nla_put_failure;
903 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
904 goto nla_put_failure;
906 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
907 * this is the first flow to be dumped into 'skb'. This is unusual for
908 * Netlink but individual action lists can be longer than
909 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
910 * The userspace caller can always fetch the actions separately if it
911 * really wants them. (Most userspace callers in fact don't care.)
913 * This can only fail for dump operations because the skb is always
914 * properly sized for single flows.
916 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
918 if (err < 0 && skb_orig_len)
921 return genlmsg_end(skb, ovs_header);
926 genlmsg_cancel(skb, ovs_header);
930 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
932 const struct sw_flow_actions *sf_acts;
934 sf_acts = ovsl_dereference(flow->sf_acts);
936 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
939 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
941 u32 portid, u32 seq, u8 cmd)
946 skb = ovs_flow_cmd_alloc_info(flow);
948 return ERR_PTR(-ENOMEM);
950 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
955 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
957 struct nlattr **a = info->attrs;
958 struct ovs_header *ovs_header = info->userhdr;
959 struct sw_flow_key key;
960 struct sw_flow *flow;
961 struct sk_buff *reply;
963 struct flow_table *table;
969 if (!a[OVS_FLOW_ATTR_KEY])
971 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
975 /* Validate actions. */
976 if (a[OVS_FLOW_ATTR_ACTIONS]) {
977 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
980 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
986 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
991 table = ovsl_dereference(dp->table);
992 flow = ovs_flow_tbl_lookup(table, &key, key_len);
994 struct sw_flow_actions *acts;
996 /* Bail out if we're not allowed to create a new flow. */
998 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1001 /* Expand table, if necessary, to make room. */
1002 if (ovs_flow_tbl_need_to_expand(table)) {
1003 struct flow_table *new_table;
1005 new_table = ovs_flow_tbl_expand(table);
1006 if (!IS_ERR(new_table)) {
1007 rcu_assign_pointer(dp->table, new_table);
1008 ovs_flow_tbl_deferred_destroy(table);
1009 table = ovsl_dereference(dp->table);
1013 /* Allocate flow. */
1014 flow = ovs_flow_alloc();
1016 error = PTR_ERR(flow);
1017 goto err_unlock_ovs;
1022 /* Obtain actions. */
1023 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1024 error = PTR_ERR(acts);
1026 goto error_free_flow;
1027 rcu_assign_pointer(flow->sf_acts, acts);
1029 /* Put flow in bucket. */
1030 flow->hash = ovs_flow_hash(&key, key_len);
1031 ovs_flow_tbl_insert(table, flow);
1033 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1037 /* We found a matching flow. */
1038 struct sw_flow_actions *old_acts;
1039 struct nlattr *acts_attrs;
1041 /* Bail out if we're not allowed to modify an existing flow.
1042 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1043 * because Generic Netlink treats the latter as a dump
1044 * request. We also accept NLM_F_EXCL in case that bug ever
1048 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1049 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1050 goto err_unlock_ovs;
1052 /* Update actions. */
1053 old_acts = ovsl_dereference(flow->sf_acts);
1054 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1056 (old_acts->actions_len != nla_len(acts_attrs) ||
1057 memcmp(old_acts->actions, nla_data(acts_attrs),
1058 old_acts->actions_len))) {
1059 struct sw_flow_actions *new_acts;
1061 new_acts = ovs_flow_actions_alloc(acts_attrs);
1062 error = PTR_ERR(new_acts);
1063 if (IS_ERR(new_acts))
1064 goto err_unlock_ovs;
1066 rcu_assign_pointer(flow->sf_acts, new_acts);
1067 ovs_flow_deferred_free_acts(old_acts);
1070 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1071 info->snd_seq, OVS_FLOW_CMD_NEW);
1074 if (a[OVS_FLOW_ATTR_CLEAR]) {
1075 spin_lock_bh(&flow->lock);
1077 spin_unlock_bh(&flow->lock);
1083 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1085 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1086 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1090 ovs_flow_free(flow);
1097 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1099 struct nlattr **a = info->attrs;
1100 struct ovs_header *ovs_header = info->userhdr;
1101 struct sw_flow_key key;
1102 struct sk_buff *reply;
1103 struct sw_flow *flow;
1104 struct datapath *dp;
1105 struct flow_table *table;
1109 if (!a[OVS_FLOW_ATTR_KEY])
1111 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1116 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1122 table = ovsl_dereference(dp->table);
1123 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1129 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1130 info->snd_seq, OVS_FLOW_CMD_NEW);
1131 if (IS_ERR(reply)) {
1132 err = PTR_ERR(reply);
1137 return genlmsg_reply(reply, info);
1143 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1145 struct nlattr **a = info->attrs;
1146 struct ovs_header *ovs_header = info->userhdr;
1147 struct sw_flow_key key;
1148 struct sk_buff *reply;
1149 struct sw_flow *flow;
1150 struct datapath *dp;
1151 struct flow_table *table;
1156 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1162 if (!a[OVS_FLOW_ATTR_KEY]) {
1163 err = flush_flows(dp);
1166 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1170 table = ovsl_dereference(dp->table);
1171 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1177 reply = ovs_flow_cmd_alloc_info(flow);
1183 ovs_flow_tbl_remove(table, flow);
1185 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1186 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1189 ovs_flow_deferred_free(flow);
1192 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1199 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1201 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1202 struct datapath *dp;
1203 struct flow_table *table;
1206 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1212 table = ovsl_dereference(dp->table);
1215 struct sw_flow *flow;
1218 bucket = cb->args[0];
1220 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1224 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1225 NETLINK_CB(cb->skb).portid,
1226 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1227 OVS_FLOW_CMD_NEW) < 0)
1230 cb->args[0] = bucket;
1237 static struct genl_ops dp_flow_genl_ops[] = {
1238 { .cmd = OVS_FLOW_CMD_NEW,
1239 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1240 .policy = flow_policy,
1241 .doit = ovs_flow_cmd_new_or_set
1243 { .cmd = OVS_FLOW_CMD_DEL,
1244 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1245 .policy = flow_policy,
1246 .doit = ovs_flow_cmd_del
1248 { .cmd = OVS_FLOW_CMD_GET,
1249 .flags = 0, /* OK for unprivileged users. */
1250 .policy = flow_policy,
1251 .doit = ovs_flow_cmd_get,
1252 .dumpit = ovs_flow_cmd_dump
1254 { .cmd = OVS_FLOW_CMD_SET,
1255 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1256 .policy = flow_policy,
1257 .doit = ovs_flow_cmd_new_or_set,
1261 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1262 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1263 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1266 static struct genl_family dp_datapath_genl_family = {
1267 .id = GENL_ID_GENERATE,
1268 .hdrsize = sizeof(struct ovs_header),
1269 .name = OVS_DATAPATH_FAMILY,
1270 .version = OVS_DATAPATH_VERSION,
1271 .maxattr = OVS_DP_ATTR_MAX,
1275 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1276 .name = OVS_DATAPATH_MCGROUP
1279 static size_t ovs_dp_cmd_msg_size(void)
1281 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1283 msgsize += nla_total_size(IFNAMSIZ);
1284 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1289 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1290 u32 portid, u32 seq, u32 flags, u8 cmd)
1292 struct ovs_header *ovs_header;
1293 struct ovs_dp_stats dp_stats;
1296 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1301 ovs_header->dp_ifindex = get_dpifindex(dp);
1304 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1307 goto nla_put_failure;
1309 get_dp_stats(dp, &dp_stats);
1310 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1311 goto nla_put_failure;
1313 return genlmsg_end(skb, ovs_header);
1316 genlmsg_cancel(skb, ovs_header);
1321 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1324 struct sk_buff *skb;
1327 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1329 return ERR_PTR(-ENOMEM);
1331 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1334 return ERR_PTR(retval);
1339 /* Called with ovs_mutex. */
1340 static struct datapath *lookup_datapath(struct net *net,
1341 struct ovs_header *ovs_header,
1342 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1344 struct datapath *dp;
1346 if (!a[OVS_DP_ATTR_NAME])
1347 dp = get_dp(net, ovs_header->dp_ifindex);
1349 struct vport *vport;
1352 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1353 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1356 return dp ? dp : ERR_PTR(-ENODEV);
1359 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1361 struct nlattr **a = info->attrs;
1362 struct vport_parms parms;
1363 struct sk_buff *reply;
1364 struct datapath *dp;
1365 struct vport *vport;
1366 struct ovs_net *ovs_net;
1370 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1376 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1378 goto err_unlock_ovs;
1380 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1382 /* Allocate table. */
1384 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1388 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1389 if (!dp->stats_percpu) {
1391 goto err_destroy_table;
1394 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1398 goto err_destroy_percpu;
1401 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1402 INIT_HLIST_HEAD(&dp->ports[i]);
1404 /* Set up our datapath device. */
1405 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1406 parms.type = OVS_VPORT_TYPE_INTERNAL;
1407 parms.options = NULL;
1409 parms.port_no = OVSP_LOCAL;
1410 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1412 vport = new_vport(&parms);
1413 if (IS_ERR(vport)) {
1414 err = PTR_ERR(vport);
1418 goto err_destroy_ports_array;
1421 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1422 info->snd_seq, OVS_DP_CMD_NEW);
1423 err = PTR_ERR(reply);
1425 goto err_destroy_local_port;
1427 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1428 list_add_tail(&dp->list_node, &ovs_net->dps);
1432 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1435 err_destroy_local_port:
1436 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1437 err_destroy_ports_array:
1440 free_percpu(dp->stats_percpu);
1442 ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1444 release_net(ovs_dp_get_net(dp));
1452 /* Called with ovs_mutex. */
1453 static void __dp_destroy(struct datapath *dp)
1457 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1458 struct vport *vport;
1459 struct hlist_node *n;
1461 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1462 if (vport->port_no != OVSP_LOCAL)
1463 ovs_dp_detach_port(vport);
1466 list_del(&dp->list_node);
1468 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1469 * all port in datapath are destroyed first before freeing datapath.
1471 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1473 call_rcu(&dp->rcu, destroy_dp_rcu);
1476 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1478 struct sk_buff *reply;
1479 struct datapath *dp;
1483 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1488 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1489 info->snd_seq, OVS_DP_CMD_DEL);
1490 err = PTR_ERR(reply);
1497 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1505 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1507 struct sk_buff *reply;
1508 struct datapath *dp;
1512 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1517 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1518 info->snd_seq, OVS_DP_CMD_NEW);
1519 if (IS_ERR(reply)) {
1520 err = PTR_ERR(reply);
1521 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1522 ovs_dp_datapath_multicast_group.id, err);
1528 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1536 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1538 struct sk_buff *reply;
1539 struct datapath *dp;
1543 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1549 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1550 info->snd_seq, OVS_DP_CMD_NEW);
1551 if (IS_ERR(reply)) {
1552 err = PTR_ERR(reply);
1557 return genlmsg_reply(reply, info);
1564 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1566 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1567 struct datapath *dp;
1568 int skip = cb->args[0];
1572 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1574 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1575 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1576 OVS_DP_CMD_NEW) < 0)
1587 static struct genl_ops dp_datapath_genl_ops[] = {
1588 { .cmd = OVS_DP_CMD_NEW,
1589 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1590 .policy = datapath_policy,
1591 .doit = ovs_dp_cmd_new
1593 { .cmd = OVS_DP_CMD_DEL,
1594 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1595 .policy = datapath_policy,
1596 .doit = ovs_dp_cmd_del
1598 { .cmd = OVS_DP_CMD_GET,
1599 .flags = 0, /* OK for unprivileged users. */
1600 .policy = datapath_policy,
1601 .doit = ovs_dp_cmd_get,
1602 .dumpit = ovs_dp_cmd_dump
1604 { .cmd = OVS_DP_CMD_SET,
1605 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1606 .policy = datapath_policy,
1607 .doit = ovs_dp_cmd_set,
1611 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1612 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1613 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1614 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1615 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1616 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1617 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1620 static struct genl_family dp_vport_genl_family = {
1621 .id = GENL_ID_GENERATE,
1622 .hdrsize = sizeof(struct ovs_header),
1623 .name = OVS_VPORT_FAMILY,
1624 .version = OVS_VPORT_VERSION,
1625 .maxattr = OVS_VPORT_ATTR_MAX,
1629 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1630 .name = OVS_VPORT_MCGROUP
1633 /* Called with ovs_mutex or RCU read lock. */
1634 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1635 u32 portid, u32 seq, u32 flags, u8 cmd)
1637 struct ovs_header *ovs_header;
1638 struct ovs_vport_stats vport_stats;
1641 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1646 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1648 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1649 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1650 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1651 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1652 goto nla_put_failure;
1654 ovs_vport_get_stats(vport, &vport_stats);
1655 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1657 goto nla_put_failure;
1659 err = ovs_vport_get_options(vport, skb);
1660 if (err == -EMSGSIZE)
1663 return genlmsg_end(skb, ovs_header);
1668 genlmsg_cancel(skb, ovs_header);
1672 /* Called with ovs_mutex or RCU read lock. */
1673 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1676 struct sk_buff *skb;
1679 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1681 return ERR_PTR(-ENOMEM);
1683 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1686 return ERR_PTR(retval);
1691 /* Called with ovs_mutex or RCU read lock. */
1692 static struct vport *lookup_vport(struct net *net,
1693 struct ovs_header *ovs_header,
1694 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1696 struct datapath *dp;
1697 struct vport *vport;
1699 if (a[OVS_VPORT_ATTR_NAME]) {
1700 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1702 return ERR_PTR(-ENODEV);
1703 if (ovs_header->dp_ifindex &&
1704 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1705 return ERR_PTR(-ENODEV);
1707 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1708 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1710 if (port_no >= DP_MAX_PORTS)
1711 return ERR_PTR(-EFBIG);
1713 dp = get_dp(net, ovs_header->dp_ifindex);
1715 return ERR_PTR(-ENODEV);
1717 vport = ovs_vport_ovsl_rcu(dp, port_no);
1719 return ERR_PTR(-ENODEV);
1722 return ERR_PTR(-EINVAL);
1725 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1727 struct nlattr **a = info->attrs;
1728 struct ovs_header *ovs_header = info->userhdr;
1729 struct vport_parms parms;
1730 struct sk_buff *reply;
1731 struct vport *vport;
1732 struct datapath *dp;
1737 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1738 !a[OVS_VPORT_ATTR_UPCALL_PID])
1742 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1747 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1748 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1751 if (port_no >= DP_MAX_PORTS)
1754 vport = ovs_vport_ovsl(dp, port_no);
1759 for (port_no = 1; ; port_no++) {
1760 if (port_no >= DP_MAX_PORTS) {
1764 vport = ovs_vport_ovsl(dp, port_no);
1770 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1771 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1772 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1774 parms.port_no = port_no;
1775 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1777 vport = new_vport(&parms);
1778 err = PTR_ERR(vport);
1783 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1785 if (IS_ERR(reply)) {
1786 err = PTR_ERR(reply);
1787 ovs_dp_detach_port(vport);
1791 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1799 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1801 struct nlattr **a = info->attrs;
1802 struct sk_buff *reply;
1803 struct vport *vport;
1807 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1808 err = PTR_ERR(vport);
1813 if (a[OVS_VPORT_ATTR_TYPE] &&
1814 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1817 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1818 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1821 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1822 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1824 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1826 if (IS_ERR(reply)) {
1827 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1828 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1833 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1841 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1843 struct nlattr **a = info->attrs;
1844 struct sk_buff *reply;
1845 struct vport *vport;
1849 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1850 err = PTR_ERR(vport);
1854 if (vport->port_no == OVSP_LOCAL) {
1859 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1861 err = PTR_ERR(reply);
1866 ovs_dp_detach_port(vport);
1868 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1875 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1877 struct nlattr **a = info->attrs;
1878 struct ovs_header *ovs_header = info->userhdr;
1879 struct sk_buff *reply;
1880 struct vport *vport;
1884 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1885 err = PTR_ERR(vport);
1889 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1891 err = PTR_ERR(reply);
1897 return genlmsg_reply(reply, info);
1904 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1906 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1907 struct datapath *dp;
1908 int bucket = cb->args[0], skip = cb->args[1];
1911 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1916 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1917 struct vport *vport;
1920 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1922 ovs_vport_cmd_fill_info(vport, skb,
1923 NETLINK_CB(cb->skb).portid,
1926 OVS_VPORT_CMD_NEW) < 0)
1942 static struct genl_ops dp_vport_genl_ops[] = {
1943 { .cmd = OVS_VPORT_CMD_NEW,
1944 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1945 .policy = vport_policy,
1946 .doit = ovs_vport_cmd_new
1948 { .cmd = OVS_VPORT_CMD_DEL,
1949 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1950 .policy = vport_policy,
1951 .doit = ovs_vport_cmd_del
1953 { .cmd = OVS_VPORT_CMD_GET,
1954 .flags = 0, /* OK for unprivileged users. */
1955 .policy = vport_policy,
1956 .doit = ovs_vport_cmd_get,
1957 .dumpit = ovs_vport_cmd_dump
1959 { .cmd = OVS_VPORT_CMD_SET,
1960 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1961 .policy = vport_policy,
1962 .doit = ovs_vport_cmd_set,
1966 struct genl_family_and_ops {
1967 struct genl_family *family;
1968 struct genl_ops *ops;
1970 struct genl_multicast_group *group;
1973 static const struct genl_family_and_ops dp_genl_families[] = {
1974 { &dp_datapath_genl_family,
1975 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1976 &ovs_dp_datapath_multicast_group },
1977 { &dp_vport_genl_family,
1978 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1979 &ovs_dp_vport_multicast_group },
1980 { &dp_flow_genl_family,
1981 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1982 &ovs_dp_flow_multicast_group },
1983 { &dp_packet_genl_family,
1984 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1988 static void dp_unregister_genl(int n_families)
1992 for (i = 0; i < n_families; i++)
1993 genl_unregister_family(dp_genl_families[i].family);
1996 static int dp_register_genl(void)
2003 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2004 const struct genl_family_and_ops *f = &dp_genl_families[i];
2006 err = genl_register_family_with_ops(f->family, f->ops,
2013 err = genl_register_mc_group(f->family, f->group);
2022 dp_unregister_genl(n_registered);
2026 static void rehash_flow_table(struct work_struct *work)
2028 struct datapath *dp;
2034 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2036 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2037 struct flow_table *old_table = ovsl_dereference(dp->table);
2038 struct flow_table *new_table;
2040 new_table = ovs_flow_tbl_rehash(old_table);
2041 if (!IS_ERR(new_table)) {
2042 rcu_assign_pointer(dp->table, new_table);
2043 ovs_flow_tbl_deferred_destroy(old_table);
2049 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2052 static int __net_init ovs_init_net(struct net *net)
2054 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2056 INIT_LIST_HEAD(&ovs_net->dps);
2057 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2061 static void __net_exit ovs_exit_net(struct net *net)
2063 struct datapath *dp, *dp_next;
2064 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2067 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2071 cancel_work_sync(&ovs_net->dp_notify_work);
2074 static struct pernet_operations ovs_net_ops = {
2075 .init = ovs_init_net,
2076 .exit = ovs_exit_net,
2078 .size = sizeof(struct ovs_net),
2081 static int __init dp_init(void)
2085 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2087 pr_info("Open vSwitch switching datapath\n");
2089 err = ovs_flow_init();
2093 err = ovs_vport_init();
2095 goto error_flow_exit;
2097 err = register_pernet_device(&ovs_net_ops);
2099 goto error_vport_exit;
2101 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2103 goto error_netns_exit;
2105 err = dp_register_genl();
2107 goto error_unreg_notifier;
2109 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2113 error_unreg_notifier:
2114 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2116 unregister_pernet_device(&ovs_net_ops);
2125 static void dp_cleanup(void)
2127 cancel_delayed_work_sync(&rehash_flow_wq);
2128 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2129 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2130 unregister_pernet_device(&ovs_net_ops);
2136 module_init(dp_init);
2137 module_exit(dp_cleanup);
2139 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2140 MODULE_LICENSE("GPL");