2 * Copyright (c) 2007-2013 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/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/genetlink.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
60 #include "vport-internal_dev.h"
61 #include "vport-netdev.h"
63 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
64 static void rehash_flow_table(struct work_struct *work);
65 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
67 int ovs_net_id __read_mostly;
69 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
70 struct genl_multicast_group *grp)
72 genl_notify(skb, genl_info_net(info), info->snd_portid,
73 grp->id, info->nlhdr, GFP_KERNEL);
79 * All writes e.g. Writes to device state (add/remove datapath, port, set
80 * operations on vports, etc.), Writes to other state (flow table
81 * modifications, set miscellaneous datapath parameters, etc.) are protected
84 * Reads are protected by RCU.
86 * There are a few special cases (mostly stats) that have their own
87 * synchronization but they nest under all of above and don't interact with
90 * The RTNL lock nests inside ovs_mutex.
93 static DEFINE_MUTEX(ovs_mutex);
97 mutex_lock(&ovs_mutex);
100 void ovs_unlock(void)
102 mutex_unlock(&ovs_mutex);
105 #ifdef CONFIG_LOCKDEP
106 int lockdep_ovsl_is_held(void)
109 return lockdep_is_held(&ovs_mutex);
115 static struct vport *new_vport(const struct vport_parms *);
116 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
117 const struct dp_upcall_info *);
118 static int queue_userspace_packet(struct net *, int dp_ifindex,
120 const struct dp_upcall_info *);
122 /* Must be called with rcu_read_lock or ovs_mutex. */
123 static struct datapath *get_dp(struct net *net, int dp_ifindex)
125 struct datapath *dp = NULL;
126 struct net_device *dev;
129 dev = dev_get_by_index_rcu(net, dp_ifindex);
131 struct vport *vport = ovs_internal_dev_get_vport(dev);
140 /* Must be called with rcu_read_lock or ovs_mutex. */
141 const char *ovs_dp_name(const struct datapath *dp)
143 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
144 return vport->ops->get_name(vport);
147 static int get_dpifindex(struct datapath *dp)
154 local = ovs_vport_rcu(dp, OVSP_LOCAL);
156 ifindex = netdev_vport_priv(local)->dev->ifindex;
165 static void destroy_dp_rcu(struct rcu_head *rcu)
167 struct datapath *dp = container_of(rcu, struct datapath, rcu);
169 free_percpu(dp->stats_percpu);
170 release_net(ovs_dp_get_net(dp));
175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
178 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
181 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
184 struct hlist_head *head;
186 head = vport_hash_bucket(dp, port_no);
187 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
188 if (vport->port_no == port_no)
194 /* Called with ovs_mutex. */
195 static struct vport *new_vport(const struct vport_parms *parms)
199 vport = ovs_vport_add(parms);
200 if (!IS_ERR(vport)) {
201 struct datapath *dp = parms->dp;
202 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
204 hlist_add_head_rcu(&vport->dp_hash_node, head);
209 void ovs_dp_detach_port(struct vport *p)
213 /* First drop references to device. */
214 hlist_del_rcu(&p->dp_hash_node);
216 /* Then destroy it. */
220 /* Must be called with rcu_read_lock. */
221 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
223 struct datapath *dp = p->dp;
224 struct sw_flow *flow;
225 struct dp_stats_percpu *stats;
226 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);
234 if (unlikely(error)) {
240 flow = ovs_flow_lookup(rcu_dereference(dp->table), &key);
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;
255 OVS_CB(skb)->pkt_key = &key;
257 stats_counter = &stats->n_hit;
258 ovs_flow_used(OVS_CB(skb)->flow, skb);
259 ovs_execute_actions(dp, skb);
262 /* Update datapath statistics. */
263 u64_stats_update_begin(&stats->sync);
265 u64_stats_update_end(&stats->sync);
268 static struct genl_family dp_packet_genl_family = {
269 .id = GENL_ID_GENERATE,
270 .hdrsize = sizeof(struct ovs_header),
271 .name = OVS_PACKET_FAMILY,
272 .version = OVS_PACKET_VERSION,
273 .maxattr = OVS_PACKET_ATTR_MAX,
278 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
279 const struct dp_upcall_info *upcall_info)
281 struct dp_stats_percpu *stats;
285 if (upcall_info->portid == 0) {
290 dp_ifindex = get_dpifindex(dp);
296 if (!skb_is_gso(skb))
297 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
299 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
306 stats = this_cpu_ptr(dp->stats_percpu);
308 u64_stats_update_begin(&stats->sync);
310 u64_stats_update_end(&stats->sync);
315 static int queue_gso_packets(struct net *net, int dp_ifindex,
317 const struct dp_upcall_info *upcall_info)
319 unsigned short gso_type = skb_shinfo(skb)->gso_type;
320 struct dp_upcall_info later_info;
321 struct sw_flow_key later_key;
322 struct sk_buff *segs, *nskb;
325 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
327 return PTR_ERR(segs);
329 /* Queue all of the segments. */
332 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
336 if (skb == segs && gso_type & SKB_GSO_UDP) {
337 /* The initial flow key extracted by ovs_flow_extract()
338 * in this case is for a first fragment, so we need to
339 * properly mark later fragments.
341 later_key = *upcall_info->key;
342 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
344 later_info = *upcall_info;
345 later_info.key = &later_key;
346 upcall_info = &later_info;
348 } while ((skb = skb->next));
350 /* Free all of the segments. */
358 } while ((skb = nskb));
362 static size_t key_attr_size(void)
364 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
365 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
366 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
367 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
368 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
369 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
370 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
371 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
372 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
373 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
374 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
375 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
376 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
377 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
378 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
379 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
380 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
381 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
382 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
385 static size_t upcall_msg_size(const struct sk_buff *skb,
386 const struct nlattr *userdata)
388 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
389 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
390 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
392 /* OVS_PACKET_ATTR_USERDATA */
394 size += NLA_ALIGN(userdata->nla_len);
399 static int queue_userspace_packet(struct net *net, int dp_ifindex,
401 const struct dp_upcall_info *upcall_info)
403 struct ovs_header *upcall;
404 struct sk_buff *nskb = NULL;
405 struct sk_buff *user_skb; /* to be queued to userspace */
409 if (vlan_tx_tag_present(skb)) {
410 nskb = skb_clone(skb, GFP_ATOMIC);
414 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
418 vlan_set_tci(nskb, 0);
423 if (nla_attr_size(skb->len) > USHRT_MAX) {
428 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
434 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
435 0, upcall_info->cmd);
436 upcall->dp_ifindex = dp_ifindex;
438 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
439 ovs_flow_to_nlattrs(upcall_info->key, upcall_info->key, user_skb);
440 nla_nest_end(user_skb, nla);
442 if (upcall_info->userdata)
443 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
444 nla_len(upcall_info->userdata),
445 nla_data(upcall_info->userdata));
447 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
449 skb_copy_and_csum_dev(skb, nla_data(nla));
451 genlmsg_end(user_skb, upcall);
452 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
459 /* Called with ovs_mutex. */
460 static int flush_flows(struct datapath *dp)
462 struct flow_table *old_table;
463 struct flow_table *new_table;
465 old_table = ovsl_dereference(dp->table);
466 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
470 rcu_assign_pointer(dp->table, new_table);
472 ovs_flow_tbl_destroy(old_table, true);
476 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
479 struct sw_flow_actions *acts;
481 int req_size = NLA_ALIGN(attr_len);
482 int next_offset = offsetof(struct sw_flow_actions, actions) +
485 if (req_size <= (ksize(*sfa) - next_offset))
488 new_acts_size = ksize(*sfa) * 2;
490 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
491 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
492 return ERR_PTR(-EMSGSIZE);
493 new_acts_size = MAX_ACTIONS_BUFSIZE;
496 acts = ovs_flow_actions_alloc(new_acts_size);
500 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
501 acts->actions_len = (*sfa)->actions_len;
506 (*sfa)->actions_len += req_size;
507 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
510 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
514 a = reserve_sfa_size(sfa, nla_attr_size(len));
518 a->nla_type = attrtype;
519 a->nla_len = nla_attr_size(len);
522 memcpy(nla_data(a), data, len);
523 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
528 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
530 int used = (*sfa)->actions_len;
533 err = add_action(sfa, attrtype, NULL, 0);
540 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
542 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
544 a->nla_len = sfa->actions_len - st_offset;
547 static int validate_and_copy_actions(const struct nlattr *attr,
548 const struct sw_flow_key *key, int depth,
549 struct sw_flow_actions **sfa);
551 static int validate_and_copy_sample(const struct nlattr *attr,
552 const struct sw_flow_key *key, int depth,
553 struct sw_flow_actions **sfa)
555 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
556 const struct nlattr *probability, *actions;
557 const struct nlattr *a;
558 int rem, start, err, st_acts;
560 memset(attrs, 0, sizeof(attrs));
561 nla_for_each_nested(a, attr, rem) {
562 int type = nla_type(a);
563 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
570 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
571 if (!probability || nla_len(probability) != sizeof(u32))
574 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
575 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
578 /* validation done, copy sample action. */
579 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
582 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
585 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
589 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
593 add_nested_action_end(*sfa, st_acts);
594 add_nested_action_end(*sfa, start);
599 static int validate_tp_port(const struct sw_flow_key *flow_key)
601 if (flow_key->eth.type == htons(ETH_P_IP)) {
602 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
604 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
605 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
612 static int validate_and_copy_set_tun(const struct nlattr *attr,
613 struct sw_flow_actions **sfa)
615 struct sw_flow_match match;
616 struct sw_flow_key key;
619 ovs_match_init(&match, &key, NULL);
620 err = ovs_ipv4_tun_from_nlattr(nla_data(attr), &match, false);
624 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
628 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
629 sizeof(match.key->tun_key));
630 add_nested_action_end(*sfa, start);
635 static int validate_set(const struct nlattr *a,
636 const struct sw_flow_key *flow_key,
637 struct sw_flow_actions **sfa,
640 const struct nlattr *ovs_key = nla_data(a);
641 int key_type = nla_type(ovs_key);
643 /* There can be only one key in a action */
644 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
647 if (key_type > OVS_KEY_ATTR_MAX ||
648 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
649 ovs_key_lens[key_type] != -1))
653 const struct ovs_key_ipv4 *ipv4_key;
654 const struct ovs_key_ipv6 *ipv6_key;
657 case OVS_KEY_ATTR_PRIORITY:
658 case OVS_KEY_ATTR_SKB_MARK:
659 case OVS_KEY_ATTR_ETHERNET:
662 case OVS_KEY_ATTR_TUNNEL:
664 err = validate_and_copy_set_tun(a, sfa);
669 case OVS_KEY_ATTR_IPV4:
670 if (flow_key->eth.type != htons(ETH_P_IP))
673 if (!flow_key->ip.proto)
676 ipv4_key = nla_data(ovs_key);
677 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
680 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
685 case OVS_KEY_ATTR_IPV6:
686 if (flow_key->eth.type != htons(ETH_P_IPV6))
689 if (!flow_key->ip.proto)
692 ipv6_key = nla_data(ovs_key);
693 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
696 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
699 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
704 case OVS_KEY_ATTR_TCP:
705 if (flow_key->ip.proto != IPPROTO_TCP)
708 return validate_tp_port(flow_key);
710 case OVS_KEY_ATTR_UDP:
711 if (flow_key->ip.proto != IPPROTO_UDP)
714 return validate_tp_port(flow_key);
716 case OVS_KEY_ATTR_SCTP:
717 if (flow_key->ip.proto != IPPROTO_SCTP)
720 return validate_tp_port(flow_key);
729 static int validate_userspace(const struct nlattr *attr)
731 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
732 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
733 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
735 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
738 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
739 attr, userspace_policy);
743 if (!a[OVS_USERSPACE_ATTR_PID] ||
744 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
750 static int copy_action(const struct nlattr *from,
751 struct sw_flow_actions **sfa)
753 int totlen = NLA_ALIGN(from->nla_len);
756 to = reserve_sfa_size(sfa, from->nla_len);
760 memcpy(to, from, totlen);
764 static int validate_and_copy_actions(const struct nlattr *attr,
765 const struct sw_flow_key *key,
767 struct sw_flow_actions **sfa)
769 const struct nlattr *a;
772 if (depth >= SAMPLE_ACTION_DEPTH)
775 nla_for_each_nested(a, attr, rem) {
776 /* Expected argument lengths, (u32)-1 for variable length. */
777 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
778 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
779 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
780 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
781 [OVS_ACTION_ATTR_POP_VLAN] = 0,
782 [OVS_ACTION_ATTR_SET] = (u32)-1,
783 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
785 const struct ovs_action_push_vlan *vlan;
786 int type = nla_type(a);
789 if (type > OVS_ACTION_ATTR_MAX ||
790 (action_lens[type] != nla_len(a) &&
791 action_lens[type] != (u32)-1))
796 case OVS_ACTION_ATTR_UNSPEC:
799 case OVS_ACTION_ATTR_USERSPACE:
800 err = validate_userspace(a);
805 case OVS_ACTION_ATTR_OUTPUT:
806 if (nla_get_u32(a) >= DP_MAX_PORTS)
811 case OVS_ACTION_ATTR_POP_VLAN:
814 case OVS_ACTION_ATTR_PUSH_VLAN:
816 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
818 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
822 case OVS_ACTION_ATTR_SET:
823 err = validate_set(a, key, sfa, &skip_copy);
828 case OVS_ACTION_ATTR_SAMPLE:
829 err = validate_and_copy_sample(a, key, depth, sfa);
839 err = copy_action(a, sfa);
851 static void clear_stats(struct sw_flow *flow)
855 flow->packet_count = 0;
856 flow->byte_count = 0;
859 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
861 struct ovs_header *ovs_header = info->userhdr;
862 struct nlattr **a = info->attrs;
863 struct sw_flow_actions *acts;
864 struct sk_buff *packet;
865 struct sw_flow *flow;
872 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
873 !a[OVS_PACKET_ATTR_ACTIONS])
876 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
877 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
881 skb_reserve(packet, NET_IP_ALIGN);
883 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
885 skb_reset_mac_header(packet);
886 eth = eth_hdr(packet);
888 /* Normally, setting the skb 'protocol' field would be handled by a
889 * call to eth_type_trans(), but it assumes there's a sending
890 * device, which we may not have. */
891 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
892 packet->protocol = eth->h_proto;
894 packet->protocol = htons(ETH_P_802_2);
896 /* Build an sw_flow for sending this packet. */
897 flow = ovs_flow_alloc();
902 err = ovs_flow_extract(packet, -1, &flow->key);
906 err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
909 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
914 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
915 rcu_assign_pointer(flow->sf_acts, acts);
919 OVS_CB(packet)->flow = flow;
920 OVS_CB(packet)->pkt_key = &flow->key;
921 packet->priority = flow->key.phy.priority;
922 packet->mark = flow->key.phy.skb_mark;
925 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
931 err = ovs_execute_actions(dp, packet);
935 ovs_flow_free(flow, false);
941 ovs_flow_free(flow, false);
948 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
949 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
950 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
951 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
954 static struct genl_ops dp_packet_genl_ops[] = {
955 { .cmd = OVS_PACKET_CMD_EXECUTE,
956 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
957 .policy = packet_policy,
958 .doit = ovs_packet_cmd_execute
962 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
964 struct flow_table *table;
967 table = rcu_dereference_check(dp->table, lockdep_ovsl_is_held());
968 stats->n_flows = ovs_flow_tbl_count(table);
970 stats->n_hit = stats->n_missed = stats->n_lost = 0;
971 for_each_possible_cpu(i) {
972 const struct dp_stats_percpu *percpu_stats;
973 struct dp_stats_percpu local_stats;
976 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
979 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
980 local_stats = *percpu_stats;
981 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
983 stats->n_hit += local_stats.n_hit;
984 stats->n_missed += local_stats.n_missed;
985 stats->n_lost += local_stats.n_lost;
989 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
990 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
991 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
992 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
995 static struct genl_family dp_flow_genl_family = {
996 .id = GENL_ID_GENERATE,
997 .hdrsize = sizeof(struct ovs_header),
998 .name = OVS_FLOW_FAMILY,
999 .version = OVS_FLOW_VERSION,
1000 .maxattr = OVS_FLOW_ATTR_MAX,
1005 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1006 .name = OVS_FLOW_MCGROUP
1009 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1010 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1012 const struct nlattr *a;
1013 struct nlattr *start;
1016 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1020 nla_for_each_nested(a, attr, rem) {
1021 int type = nla_type(a);
1022 struct nlattr *st_sample;
1025 case OVS_SAMPLE_ATTR_PROBABILITY:
1026 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1029 case OVS_SAMPLE_ATTR_ACTIONS:
1030 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1033 err = actions_to_attr(nla_data(a), nla_len(a), skb);
1036 nla_nest_end(skb, st_sample);
1041 nla_nest_end(skb, start);
1045 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1047 const struct nlattr *ovs_key = nla_data(a);
1048 int key_type = nla_type(ovs_key);
1049 struct nlattr *start;
1053 case OVS_KEY_ATTR_IPV4_TUNNEL:
1054 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1058 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
1062 nla_nest_end(skb, start);
1065 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1073 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1075 const struct nlattr *a;
1078 nla_for_each_attr(a, attr, len, rem) {
1079 int type = nla_type(a);
1082 case OVS_ACTION_ATTR_SET:
1083 err = set_action_to_attr(a, skb);
1088 case OVS_ACTION_ATTR_SAMPLE:
1089 err = sample_action_to_attr(a, skb);
1094 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1103 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1105 return NLMSG_ALIGN(sizeof(struct ovs_header))
1106 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1107 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1108 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1109 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1110 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1111 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1114 /* Called with ovs_mutex. */
1115 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1116 struct sk_buff *skb, u32 portid,
1117 u32 seq, u32 flags, u8 cmd)
1119 const int skb_orig_len = skb->len;
1120 struct nlattr *start;
1121 struct ovs_flow_stats stats;
1122 struct ovs_header *ovs_header;
1128 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1132 ovs_header->dp_ifindex = get_dpifindex(dp);
1134 /* Fill flow key. */
1135 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1137 goto nla_put_failure;
1139 err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1140 &flow->unmasked_key, skb);
1143 nla_nest_end(skb, nla);
1145 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1147 goto nla_put_failure;
1149 err = ovs_flow_to_nlattrs(&flow->key, &flow->mask->key, skb);
1153 nla_nest_end(skb, nla);
1155 spin_lock_bh(&flow->lock);
1157 stats.n_packets = flow->packet_count;
1158 stats.n_bytes = flow->byte_count;
1159 tcp_flags = flow->tcp_flags;
1160 spin_unlock_bh(&flow->lock);
1163 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1164 goto nla_put_failure;
1166 if (stats.n_packets &&
1167 nla_put(skb, OVS_FLOW_ATTR_STATS,
1168 sizeof(struct ovs_flow_stats), &stats))
1169 goto nla_put_failure;
1172 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1173 goto nla_put_failure;
1175 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1176 * this is the first flow to be dumped into 'skb'. This is unusual for
1177 * Netlink but individual action lists can be longer than
1178 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1179 * The userspace caller can always fetch the actions separately if it
1180 * really wants them. (Most userspace callers in fact don't care.)
1182 * This can only fail for dump operations because the skb is always
1183 * properly sized for single flows.
1185 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1187 const struct sw_flow_actions *sf_acts;
1189 sf_acts = rcu_dereference_check(flow->sf_acts,
1190 lockdep_ovsl_is_held());
1192 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1194 nla_nest_end(skb, start);
1199 nla_nest_cancel(skb, start);
1201 } else if (skb_orig_len)
1202 goto nla_put_failure;
1204 return genlmsg_end(skb, ovs_header);
1209 genlmsg_cancel(skb, ovs_header);
1213 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1215 const struct sw_flow_actions *sf_acts;
1217 sf_acts = ovsl_dereference(flow->sf_acts);
1219 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1222 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1223 struct datapath *dp,
1224 u32 portid, u32 seq, u8 cmd)
1226 struct sk_buff *skb;
1229 skb = ovs_flow_cmd_alloc_info(flow);
1231 return ERR_PTR(-ENOMEM);
1233 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1238 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1240 struct nlattr **a = info->attrs;
1241 struct ovs_header *ovs_header = info->userhdr;
1242 struct sw_flow_key key, masked_key;
1243 struct sw_flow *flow = NULL;
1244 struct sw_flow_mask mask;
1245 struct sk_buff *reply;
1246 struct datapath *dp;
1247 struct flow_table *table;
1248 struct sw_flow_actions *acts = NULL;
1249 struct sw_flow_match match;
1254 if (!a[OVS_FLOW_ATTR_KEY])
1257 ovs_match_init(&match, &key, &mask);
1258 error = ovs_match_from_nlattrs(&match,
1259 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1263 /* Validate actions. */
1264 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1265 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1266 error = PTR_ERR(acts);
1270 ovs_flow_key_mask(&masked_key, &key, &mask);
1271 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1272 &masked_key, 0, &acts);
1274 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1277 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1283 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1286 goto err_unlock_ovs;
1288 table = ovsl_dereference(dp->table);
1290 /* Check if this is a duplicate flow */
1291 flow = ovs_flow_lookup(table, &key);
1293 struct sw_flow_mask *mask_p;
1294 /* Bail out if we're not allowed to create a new flow. */
1296 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1297 goto err_unlock_ovs;
1299 /* Expand table, if necessary, to make room. */
1300 if (ovs_flow_tbl_need_to_expand(table)) {
1301 struct flow_table *new_table;
1303 new_table = ovs_flow_tbl_expand(table);
1304 if (!IS_ERR(new_table)) {
1305 rcu_assign_pointer(dp->table, new_table);
1306 ovs_flow_tbl_destroy(table, true);
1307 table = ovsl_dereference(dp->table);
1311 /* Allocate flow. */
1312 flow = ovs_flow_alloc();
1314 error = PTR_ERR(flow);
1315 goto err_unlock_ovs;
1319 flow->key = masked_key;
1320 flow->unmasked_key = key;
1322 /* Make sure mask is unique in the system */
1323 mask_p = ovs_sw_flow_mask_find(table, &mask);
1325 /* Allocate a new mask if none exsits. */
1326 mask_p = ovs_sw_flow_mask_alloc();
1329 mask_p->key = mask.key;
1330 mask_p->range = mask.range;
1331 ovs_sw_flow_mask_insert(table, mask_p);
1334 ovs_sw_flow_mask_add_ref(mask_p);
1335 flow->mask = mask_p;
1336 rcu_assign_pointer(flow->sf_acts, acts);
1338 /* Put flow in bucket. */
1339 ovs_flow_insert(table, flow);
1341 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1342 info->snd_seq, OVS_FLOW_CMD_NEW);
1344 /* We found a matching flow. */
1345 struct sw_flow_actions *old_acts;
1347 /* Bail out if we're not allowed to modify an existing flow.
1348 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1349 * because Generic Netlink treats the latter as a dump
1350 * request. We also accept NLM_F_EXCL in case that bug ever
1354 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1355 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1356 goto err_unlock_ovs;
1358 /* The unmasked key has to be the same for flow updates. */
1360 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1361 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1362 goto err_unlock_ovs;
1365 /* Update actions. */
1366 old_acts = ovsl_dereference(flow->sf_acts);
1367 rcu_assign_pointer(flow->sf_acts, acts);
1368 ovs_flow_deferred_free_acts(old_acts);
1370 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1371 info->snd_seq, OVS_FLOW_CMD_NEW);
1374 if (a[OVS_FLOW_ATTR_CLEAR]) {
1375 spin_lock_bh(&flow->lock);
1377 spin_unlock_bh(&flow->lock);
1383 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1385 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1386 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1390 ovs_flow_free(flow, false);
1399 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1401 struct nlattr **a = info->attrs;
1402 struct ovs_header *ovs_header = info->userhdr;
1403 struct sw_flow_key key;
1404 struct sk_buff *reply;
1405 struct sw_flow *flow;
1406 struct datapath *dp;
1407 struct flow_table *table;
1408 struct sw_flow_match match;
1411 if (!a[OVS_FLOW_ATTR_KEY]) {
1412 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1416 ovs_match_init(&match, &key, NULL);
1417 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1422 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1428 table = ovsl_dereference(dp->table);
1429 flow = ovs_flow_lookup_unmasked_key(table, &match);
1435 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1436 info->snd_seq, OVS_FLOW_CMD_NEW);
1437 if (IS_ERR(reply)) {
1438 err = PTR_ERR(reply);
1443 return genlmsg_reply(reply, info);
1449 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1451 struct nlattr **a = info->attrs;
1452 struct ovs_header *ovs_header = info->userhdr;
1453 struct sw_flow_key key;
1454 struct sk_buff *reply;
1455 struct sw_flow *flow;
1456 struct datapath *dp;
1457 struct flow_table *table;
1458 struct sw_flow_match match;
1462 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1468 if (!a[OVS_FLOW_ATTR_KEY]) {
1469 err = flush_flows(dp);
1473 ovs_match_init(&match, &key, NULL);
1474 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1478 table = ovsl_dereference(dp->table);
1479 flow = ovs_flow_lookup_unmasked_key(table, &match);
1485 reply = ovs_flow_cmd_alloc_info(flow);
1491 ovs_flow_remove(table, flow);
1493 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1494 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1497 ovs_flow_free(flow, true);
1500 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1507 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1509 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1510 struct datapath *dp;
1511 struct flow_table *table;
1514 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1520 table = rcu_dereference(dp->table);
1522 struct sw_flow *flow;
1525 bucket = cb->args[0];
1527 flow = ovs_flow_dump_next(table, &bucket, &obj);
1531 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1532 NETLINK_CB(cb->skb).portid,
1533 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1534 OVS_FLOW_CMD_NEW) < 0)
1537 cb->args[0] = bucket;
1544 static struct genl_ops dp_flow_genl_ops[] = {
1545 { .cmd = OVS_FLOW_CMD_NEW,
1546 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1547 .policy = flow_policy,
1548 .doit = ovs_flow_cmd_new_or_set
1550 { .cmd = OVS_FLOW_CMD_DEL,
1551 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1552 .policy = flow_policy,
1553 .doit = ovs_flow_cmd_del
1555 { .cmd = OVS_FLOW_CMD_GET,
1556 .flags = 0, /* OK for unprivileged users. */
1557 .policy = flow_policy,
1558 .doit = ovs_flow_cmd_get,
1559 .dumpit = ovs_flow_cmd_dump
1561 { .cmd = OVS_FLOW_CMD_SET,
1562 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1563 .policy = flow_policy,
1564 .doit = ovs_flow_cmd_new_or_set,
1568 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1569 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1570 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1573 static struct genl_family dp_datapath_genl_family = {
1574 .id = GENL_ID_GENERATE,
1575 .hdrsize = sizeof(struct ovs_header),
1576 .name = OVS_DATAPATH_FAMILY,
1577 .version = OVS_DATAPATH_VERSION,
1578 .maxattr = OVS_DP_ATTR_MAX,
1583 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1584 .name = OVS_DATAPATH_MCGROUP
1587 static size_t ovs_dp_cmd_msg_size(void)
1589 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1591 msgsize += nla_total_size(IFNAMSIZ);
1592 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1597 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1598 u32 portid, u32 seq, u32 flags, u8 cmd)
1600 struct ovs_header *ovs_header;
1601 struct ovs_dp_stats dp_stats;
1604 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1609 ovs_header->dp_ifindex = get_dpifindex(dp);
1612 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1615 goto nla_put_failure;
1617 get_dp_stats(dp, &dp_stats);
1618 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1619 goto nla_put_failure;
1621 return genlmsg_end(skb, ovs_header);
1624 genlmsg_cancel(skb, ovs_header);
1629 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1632 struct sk_buff *skb;
1635 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1637 return ERR_PTR(-ENOMEM);
1639 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1642 return ERR_PTR(retval);
1647 /* Called with ovs_mutex. */
1648 static struct datapath *lookup_datapath(struct net *net,
1649 struct ovs_header *ovs_header,
1650 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1652 struct datapath *dp;
1654 if (!a[OVS_DP_ATTR_NAME])
1655 dp = get_dp(net, ovs_header->dp_ifindex);
1657 struct vport *vport;
1660 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1661 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1664 return dp ? dp : ERR_PTR(-ENODEV);
1667 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1669 struct nlattr **a = info->attrs;
1670 struct vport_parms parms;
1671 struct sk_buff *reply;
1672 struct datapath *dp;
1673 struct vport *vport;
1674 struct ovs_net *ovs_net;
1678 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1684 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1686 goto err_unlock_ovs;
1688 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1690 /* Allocate table. */
1692 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1696 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1697 if (!dp->stats_percpu) {
1699 goto err_destroy_table;
1702 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1706 goto err_destroy_percpu;
1709 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1710 INIT_HLIST_HEAD(&dp->ports[i]);
1712 /* Set up our datapath device. */
1713 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1714 parms.type = OVS_VPORT_TYPE_INTERNAL;
1715 parms.options = NULL;
1717 parms.port_no = OVSP_LOCAL;
1718 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1720 vport = new_vport(&parms);
1721 if (IS_ERR(vport)) {
1722 err = PTR_ERR(vport);
1726 goto err_destroy_ports_array;
1729 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1730 info->snd_seq, OVS_DP_CMD_NEW);
1731 err = PTR_ERR(reply);
1733 goto err_destroy_local_port;
1735 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1736 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1740 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1743 err_destroy_local_port:
1744 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1745 err_destroy_ports_array:
1748 free_percpu(dp->stats_percpu);
1750 ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1752 release_net(ovs_dp_get_net(dp));
1760 /* Called with ovs_mutex. */
1761 static void __dp_destroy(struct datapath *dp)
1765 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1766 struct vport *vport;
1767 struct hlist_node *n;
1769 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1770 if (vport->port_no != OVSP_LOCAL)
1771 ovs_dp_detach_port(vport);
1774 list_del_rcu(&dp->list_node);
1776 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1777 * all port in datapath are destroyed first before freeing datapath.
1779 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1781 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table, true);
1783 call_rcu(&dp->rcu, destroy_dp_rcu);
1786 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1788 struct sk_buff *reply;
1789 struct datapath *dp;
1793 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1798 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1799 info->snd_seq, OVS_DP_CMD_DEL);
1800 err = PTR_ERR(reply);
1807 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1815 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1817 struct sk_buff *reply;
1818 struct datapath *dp;
1822 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1827 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1828 info->snd_seq, OVS_DP_CMD_NEW);
1829 if (IS_ERR(reply)) {
1830 err = PTR_ERR(reply);
1831 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1832 ovs_dp_datapath_multicast_group.id, err);
1838 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1846 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1848 struct sk_buff *reply;
1849 struct datapath *dp;
1853 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1859 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1860 info->snd_seq, OVS_DP_CMD_NEW);
1861 if (IS_ERR(reply)) {
1862 err = PTR_ERR(reply);
1867 return genlmsg_reply(reply, info);
1874 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1876 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1877 struct datapath *dp;
1878 int skip = cb->args[0];
1882 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1884 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1885 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1886 OVS_DP_CMD_NEW) < 0)
1897 static struct genl_ops dp_datapath_genl_ops[] = {
1898 { .cmd = OVS_DP_CMD_NEW,
1899 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1900 .policy = datapath_policy,
1901 .doit = ovs_dp_cmd_new
1903 { .cmd = OVS_DP_CMD_DEL,
1904 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1905 .policy = datapath_policy,
1906 .doit = ovs_dp_cmd_del
1908 { .cmd = OVS_DP_CMD_GET,
1909 .flags = 0, /* OK for unprivileged users. */
1910 .policy = datapath_policy,
1911 .doit = ovs_dp_cmd_get,
1912 .dumpit = ovs_dp_cmd_dump
1914 { .cmd = OVS_DP_CMD_SET,
1915 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1916 .policy = datapath_policy,
1917 .doit = ovs_dp_cmd_set,
1921 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1922 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1923 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1924 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1925 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1926 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1927 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1930 static struct genl_family dp_vport_genl_family = {
1931 .id = GENL_ID_GENERATE,
1932 .hdrsize = sizeof(struct ovs_header),
1933 .name = OVS_VPORT_FAMILY,
1934 .version = OVS_VPORT_VERSION,
1935 .maxattr = OVS_VPORT_ATTR_MAX,
1940 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1941 .name = OVS_VPORT_MCGROUP
1944 /* Called with ovs_mutex or RCU read lock. */
1945 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1946 u32 portid, u32 seq, u32 flags, u8 cmd)
1948 struct ovs_header *ovs_header;
1949 struct ovs_vport_stats vport_stats;
1952 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1957 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1959 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1960 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1961 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1962 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1963 goto nla_put_failure;
1965 ovs_vport_get_stats(vport, &vport_stats);
1966 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1968 goto nla_put_failure;
1970 err = ovs_vport_get_options(vport, skb);
1971 if (err == -EMSGSIZE)
1974 return genlmsg_end(skb, ovs_header);
1979 genlmsg_cancel(skb, ovs_header);
1983 /* Called with ovs_mutex or RCU read lock. */
1984 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1987 struct sk_buff *skb;
1990 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1992 return ERR_PTR(-ENOMEM);
1994 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
2000 /* Called with ovs_mutex or RCU read lock. */
2001 static struct vport *lookup_vport(struct net *net,
2002 struct ovs_header *ovs_header,
2003 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2005 struct datapath *dp;
2006 struct vport *vport;
2008 if (a[OVS_VPORT_ATTR_NAME]) {
2009 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2011 return ERR_PTR(-ENODEV);
2012 if (ovs_header->dp_ifindex &&
2013 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2014 return ERR_PTR(-ENODEV);
2016 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2017 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2019 if (port_no >= DP_MAX_PORTS)
2020 return ERR_PTR(-EFBIG);
2022 dp = get_dp(net, ovs_header->dp_ifindex);
2024 return ERR_PTR(-ENODEV);
2026 vport = ovs_vport_ovsl_rcu(dp, port_no);
2028 return ERR_PTR(-ENODEV);
2031 return ERR_PTR(-EINVAL);
2034 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2036 struct nlattr **a = info->attrs;
2037 struct ovs_header *ovs_header = info->userhdr;
2038 struct vport_parms parms;
2039 struct sk_buff *reply;
2040 struct vport *vport;
2041 struct datapath *dp;
2046 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2047 !a[OVS_VPORT_ATTR_UPCALL_PID])
2051 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2056 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2057 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2060 if (port_no >= DP_MAX_PORTS)
2063 vport = ovs_vport_ovsl(dp, port_no);
2068 for (port_no = 1; ; port_no++) {
2069 if (port_no >= DP_MAX_PORTS) {
2073 vport = ovs_vport_ovsl(dp, port_no);
2079 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2080 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2081 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2083 parms.port_no = port_no;
2084 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2086 vport = new_vport(&parms);
2087 err = PTR_ERR(vport);
2092 if (a[OVS_VPORT_ATTR_STATS])
2093 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2095 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2097 if (IS_ERR(reply)) {
2098 err = PTR_ERR(reply);
2099 ovs_dp_detach_port(vport);
2103 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2111 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2113 struct nlattr **a = info->attrs;
2114 struct sk_buff *reply;
2115 struct vport *vport;
2119 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2120 err = PTR_ERR(vport);
2124 if (a[OVS_VPORT_ATTR_TYPE] &&
2125 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2130 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2136 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2137 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2142 if (a[OVS_VPORT_ATTR_STATS])
2143 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2145 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2146 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2148 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2149 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2153 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2163 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2165 struct nlattr **a = info->attrs;
2166 struct sk_buff *reply;
2167 struct vport *vport;
2171 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2172 err = PTR_ERR(vport);
2176 if (vport->port_no == OVSP_LOCAL) {
2181 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2182 info->snd_seq, OVS_VPORT_CMD_DEL);
2183 err = PTR_ERR(reply);
2188 ovs_dp_detach_port(vport);
2190 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2197 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2199 struct nlattr **a = info->attrs;
2200 struct ovs_header *ovs_header = info->userhdr;
2201 struct sk_buff *reply;
2202 struct vport *vport;
2206 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2207 err = PTR_ERR(vport);
2211 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2212 info->snd_seq, OVS_VPORT_CMD_NEW);
2213 err = PTR_ERR(reply);
2219 return genlmsg_reply(reply, info);
2226 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2228 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2229 struct datapath *dp;
2230 int bucket = cb->args[0], skip = cb->args[1];
2234 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2239 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2240 struct vport *vport;
2243 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2245 ovs_vport_cmd_fill_info(vport, skb,
2246 NETLINK_CB(cb->skb).portid,
2249 OVS_VPORT_CMD_NEW) < 0)
2265 static struct genl_ops dp_vport_genl_ops[] = {
2266 { .cmd = OVS_VPORT_CMD_NEW,
2267 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2268 .policy = vport_policy,
2269 .doit = ovs_vport_cmd_new
2271 { .cmd = OVS_VPORT_CMD_DEL,
2272 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2273 .policy = vport_policy,
2274 .doit = ovs_vport_cmd_del
2276 { .cmd = OVS_VPORT_CMD_GET,
2277 .flags = 0, /* OK for unprivileged users. */
2278 .policy = vport_policy,
2279 .doit = ovs_vport_cmd_get,
2280 .dumpit = ovs_vport_cmd_dump
2282 { .cmd = OVS_VPORT_CMD_SET,
2283 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2284 .policy = vport_policy,
2285 .doit = ovs_vport_cmd_set,
2289 struct genl_family_and_ops {
2290 struct genl_family *family;
2291 struct genl_ops *ops;
2293 struct genl_multicast_group *group;
2296 static const struct genl_family_and_ops dp_genl_families[] = {
2297 { &dp_datapath_genl_family,
2298 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2299 &ovs_dp_datapath_multicast_group },
2300 { &dp_vport_genl_family,
2301 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2302 &ovs_dp_vport_multicast_group },
2303 { &dp_flow_genl_family,
2304 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2305 &ovs_dp_flow_multicast_group },
2306 { &dp_packet_genl_family,
2307 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2311 static void dp_unregister_genl(int n_families)
2315 for (i = 0; i < n_families; i++)
2316 genl_unregister_family(dp_genl_families[i].family);
2319 static int dp_register_genl(void)
2326 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2327 const struct genl_family_and_ops *f = &dp_genl_families[i];
2329 err = genl_register_family_with_ops(f->family, f->ops,
2336 err = genl_register_mc_group(f->family, f->group);
2345 dp_unregister_genl(n_registered);
2349 static void rehash_flow_table(struct work_struct *work)
2351 struct datapath *dp;
2357 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2359 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2360 struct flow_table *old_table = ovsl_dereference(dp->table);
2361 struct flow_table *new_table;
2363 new_table = ovs_flow_tbl_rehash(old_table);
2364 if (!IS_ERR(new_table)) {
2365 rcu_assign_pointer(dp->table, new_table);
2366 ovs_flow_tbl_destroy(old_table, true);
2372 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2375 static int __net_init ovs_init_net(struct net *net)
2377 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2379 INIT_LIST_HEAD(&ovs_net->dps);
2380 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2384 static void __net_exit ovs_exit_net(struct net *net)
2386 struct datapath *dp, *dp_next;
2387 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2390 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2394 cancel_work_sync(&ovs_net->dp_notify_work);
2397 static struct pernet_operations ovs_net_ops = {
2398 .init = ovs_init_net,
2399 .exit = ovs_exit_net,
2401 .size = sizeof(struct ovs_net),
2404 DEFINE_COMPAT_PNET_REG_FUNC(device);
2406 static int __init dp_init(void)
2410 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2412 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2415 err = ovs_workqueues_init();
2419 err = ovs_flow_init();
2423 err = ovs_vport_init();
2425 goto error_flow_exit;
2427 err = register_pernet_device(&ovs_net_ops);
2429 goto error_vport_exit;
2431 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2433 goto error_netns_exit;
2435 err = dp_register_genl();
2437 goto error_unreg_notifier;
2439 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2443 error_unreg_notifier:
2444 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2446 unregister_pernet_device(&ovs_net_ops);
2452 ovs_workqueues_exit();
2457 static void dp_cleanup(void)
2459 cancel_delayed_work_sync(&rehash_flow_wq);
2460 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2461 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2462 unregister_pernet_device(&ovs_net_ops);
2466 ovs_workqueues_exit();
2469 module_init(dp_init);
2470 module_exit(dp_cleanup);
2472 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2473 MODULE_LICENSE("GPL");
2474 MODULE_VERSION(VERSION);