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>
61 #include "vport-internal_dev.h"
62 #include "vport-netdev.h"
64 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
65 static void rehash_flow_table(struct work_struct *work);
66 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
68 int ovs_net_id __read_mostly;
70 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
71 struct genl_multicast_group *grp)
73 genl_notify(skb, genl_info_net(info), info->snd_portid,
74 grp->id, info->nlhdr, GFP_KERNEL);
80 * All writes e.g. Writes to device state (add/remove datapath, port, set
81 * operations on vports, etc.), Writes to other state (flow table
82 * modifications, set miscellaneous datapath parameters, etc.) are protected
85 * Reads are protected by RCU.
87 * There are a few special cases (mostly stats) that have their own
88 * synchronization but they nest under all of above and don't interact with
91 * The RTNL lock nests inside ovs_mutex.
94 static DEFINE_MUTEX(ovs_mutex);
98 mutex_lock(&ovs_mutex);
101 void ovs_unlock(void)
103 mutex_unlock(&ovs_mutex);
106 #ifdef CONFIG_LOCKDEP
107 int lockdep_ovsl_is_held(void)
110 return lockdep_is_held(&ovs_mutex);
116 static struct vport *new_vport(const struct vport_parms *);
117 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
118 const struct dp_upcall_info *);
119 static int queue_userspace_packet(struct net *, int dp_ifindex,
121 const struct dp_upcall_info *);
123 /* Must be called with rcu_read_lock or ovs_mutex. */
124 static struct datapath *get_dp(struct net *net, int dp_ifindex)
126 struct datapath *dp = NULL;
127 struct net_device *dev;
130 dev = dev_get_by_index_rcu(net, dp_ifindex);
132 struct vport *vport = ovs_internal_dev_get_vport(dev);
141 /* Must be called with rcu_read_lock or ovs_mutex. */
142 const char *ovs_dp_name(const struct datapath *dp)
144 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
145 return vport->ops->get_name(vport);
148 static int get_dpifindex(struct datapath *dp)
155 local = ovs_vport_rcu(dp, OVSP_LOCAL);
157 ifindex = netdev_vport_priv(local)->dev->ifindex;
166 static void destroy_dp_rcu(struct rcu_head *rcu)
168 struct datapath *dp = container_of(rcu, struct datapath, rcu);
170 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table, false);
171 free_percpu(dp->stats_percpu);
172 release_net(ovs_dp_get_net(dp));
177 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
180 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
183 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
186 struct hlist_head *head;
188 head = vport_hash_bucket(dp, port_no);
189 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
190 if (vport->port_no == port_no)
196 /* Called with ovs_mutex. */
197 static struct vport *new_vport(const struct vport_parms *parms)
201 vport = ovs_vport_add(parms);
202 if (!IS_ERR(vport)) {
203 struct datapath *dp = parms->dp;
204 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
206 hlist_add_head_rcu(&vport->dp_hash_node, head);
211 void ovs_dp_detach_port(struct vport *p)
215 /* First drop references to device. */
216 hlist_del_rcu(&p->dp_hash_node);
218 /* Then destroy it. */
222 /* Must be called with rcu_read_lock. */
223 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
225 struct datapath *dp = p->dp;
226 struct sw_flow *flow;
227 struct dp_stats_percpu *stats;
228 struct sw_flow_key key;
232 stats = this_cpu_ptr(dp->stats_percpu);
234 /* Extract flow from 'skb' into 'key'. */
235 error = ovs_flow_extract(skb, p->port_no, &key);
236 if (unlikely(error)) {
242 flow = ovs_flow_lookup(rcu_dereference(dp->table), &key);
243 if (unlikely(!flow)) {
244 struct dp_upcall_info upcall;
246 upcall.cmd = OVS_PACKET_CMD_MISS;
248 upcall.userdata = NULL;
249 upcall.portid = p->upcall_portid;
250 ovs_dp_upcall(dp, skb, &upcall);
252 stats_counter = &stats->n_missed;
256 OVS_CB(skb)->flow = flow;
257 OVS_CB(skb)->pkt_key = &key;
259 stats_counter = &stats->n_hit;
260 ovs_flow_used(OVS_CB(skb)->flow, skb);
261 ovs_execute_actions(dp, skb);
264 /* Update datapath statistics. */
265 u64_stats_update_begin(&stats->sync);
267 u64_stats_update_end(&stats->sync);
270 static struct genl_family dp_packet_genl_family = {
271 .id = GENL_ID_GENERATE,
272 .hdrsize = sizeof(struct ovs_header),
273 .name = OVS_PACKET_FAMILY,
274 .version = OVS_PACKET_VERSION,
275 .maxattr = OVS_PACKET_ATTR_MAX,
280 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
281 const struct dp_upcall_info *upcall_info)
283 struct dp_stats_percpu *stats;
287 if (upcall_info->portid == 0) {
292 dp_ifindex = get_dpifindex(dp);
298 forward_ip_summed(skb, true);
300 if (!skb_is_gso(skb))
301 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
303 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
310 stats = this_cpu_ptr(dp->stats_percpu);
312 u64_stats_update_begin(&stats->sync);
314 u64_stats_update_end(&stats->sync);
319 static int queue_gso_packets(struct net *net, int dp_ifindex,
321 const struct dp_upcall_info *upcall_info)
323 unsigned short gso_type = skb_shinfo(skb)->gso_type;
324 struct dp_upcall_info later_info;
325 struct sw_flow_key later_key;
326 struct sk_buff *segs, *nskb;
329 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
331 return PTR_ERR(segs);
333 /* Queue all of the segments. */
336 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
340 if (skb == segs && gso_type & SKB_GSO_UDP) {
341 /* The initial flow key extracted by ovs_flow_extract()
342 * in this case is for a first fragment, so we need to
343 * properly mark later fragments.
345 later_key = *upcall_info->key;
346 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
348 later_info = *upcall_info;
349 later_info.key = &later_key;
350 upcall_info = &later_info;
352 } while ((skb = skb->next));
354 /* Free all of the segments. */
362 } while ((skb = nskb));
366 static size_t key_attr_size(void)
368 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
369 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
370 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
371 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
372 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
373 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
374 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
375 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
376 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
377 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
378 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
379 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
380 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
381 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
382 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
383 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
384 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
385 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
386 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
389 static size_t upcall_msg_size(const struct sk_buff *skb,
390 const struct nlattr *userdata)
392 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
393 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
394 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
396 /* OVS_PACKET_ATTR_USERDATA */
398 size += NLA_ALIGN(userdata->nla_len);
403 static int queue_userspace_packet(struct net *net, int dp_ifindex,
405 const struct dp_upcall_info *upcall_info)
407 struct ovs_header *upcall;
408 struct sk_buff *nskb = NULL;
409 struct sk_buff *user_skb; /* to be queued to userspace */
413 if (vlan_tx_tag_present(skb)) {
414 nskb = skb_clone(skb, GFP_ATOMIC);
418 err = vlan_deaccel_tag(nskb);
425 if (nla_attr_size(skb->len) > USHRT_MAX) {
430 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
436 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
437 0, upcall_info->cmd);
438 upcall->dp_ifindex = dp_ifindex;
440 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
441 ovs_flow_to_nlattrs(upcall_info->key, upcall_info->key, user_skb);
442 nla_nest_end(user_skb, nla);
444 if (upcall_info->userdata)
445 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
446 nla_len(upcall_info->userdata),
447 nla_data(upcall_info->userdata));
449 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
451 skb_copy_and_csum_dev(skb, nla_data(nla));
453 genlmsg_end(user_skb, upcall);
454 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
461 /* Called with ovs_mutex. */
462 static int flush_flows(struct datapath *dp)
464 struct flow_table *old_table;
465 struct flow_table *new_table;
467 old_table = ovsl_dereference(dp->table);
468 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
472 rcu_assign_pointer(dp->table, new_table);
474 ovs_flow_tbl_destroy(old_table, true);
478 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
481 struct sw_flow_actions *acts;
483 int req_size = NLA_ALIGN(attr_len);
484 int next_offset = offsetof(struct sw_flow_actions, actions) +
487 if (req_size <= (ksize(*sfa) - next_offset))
490 new_acts_size = ksize(*sfa) * 2;
492 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
493 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
494 return ERR_PTR(-EMSGSIZE);
495 new_acts_size = MAX_ACTIONS_BUFSIZE;
498 acts = ovs_flow_actions_alloc(new_acts_size);
502 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
503 acts->actions_len = (*sfa)->actions_len;
508 (*sfa)->actions_len += req_size;
509 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
512 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
516 a = reserve_sfa_size(sfa, nla_attr_size(len));
520 a->nla_type = attrtype;
521 a->nla_len = nla_attr_size(len);
524 memcpy(nla_data(a), data, len);
525 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
530 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
532 int used = (*sfa)->actions_len;
535 err = add_action(sfa, attrtype, NULL, 0);
542 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
544 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
546 a->nla_len = sfa->actions_len - st_offset;
549 static int validate_and_copy_actions(const struct nlattr *attr,
550 const struct sw_flow_key *key, int depth,
551 struct sw_flow_actions **sfa);
553 static int validate_and_copy_sample(const struct nlattr *attr,
554 const struct sw_flow_key *key, int depth,
555 struct sw_flow_actions **sfa)
557 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
558 const struct nlattr *probability, *actions;
559 const struct nlattr *a;
560 int rem, start, err, st_acts;
562 memset(attrs, 0, sizeof(attrs));
563 nla_for_each_nested(a, attr, rem) {
564 int type = nla_type(a);
565 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
572 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
573 if (!probability || nla_len(probability) != sizeof(u32))
576 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
577 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
580 /* validation done, copy sample action. */
581 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
584 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
587 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
591 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
595 add_nested_action_end(*sfa, st_acts);
596 add_nested_action_end(*sfa, start);
601 static int validate_tp_port(const struct sw_flow_key *flow_key)
603 if (flow_key->eth.type == htons(ETH_P_IP)) {
604 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
606 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
607 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
614 static int validate_and_copy_set_tun(const struct nlattr *attr,
615 struct sw_flow_actions **sfa)
617 struct sw_flow_match match;
618 struct sw_flow_key key;
621 ovs_match_init(&match, &key, NULL);
622 err = ovs_ipv4_tun_from_nlattr(nla_data(attr), &match, false);
626 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
630 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
631 sizeof(match.key->tun_key));
632 add_nested_action_end(*sfa, start);
637 static int validate_set(const struct nlattr *a,
638 const struct sw_flow_key *flow_key,
639 struct sw_flow_actions **sfa,
642 const struct nlattr *ovs_key = nla_data(a);
643 int key_type = nla_type(ovs_key);
645 /* There can be only one key in a action */
646 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
649 if (key_type > OVS_KEY_ATTR_MAX ||
650 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
651 ovs_key_lens[key_type] != -1))
655 const struct ovs_key_ipv4 *ipv4_key;
656 const struct ovs_key_ipv6 *ipv6_key;
659 case OVS_KEY_ATTR_PRIORITY:
660 case OVS_KEY_ATTR_SKB_MARK:
661 case OVS_KEY_ATTR_ETHERNET:
664 case OVS_KEY_ATTR_TUNNEL:
666 err = validate_and_copy_set_tun(a, sfa);
671 case OVS_KEY_ATTR_IPV4:
672 if (flow_key->eth.type != htons(ETH_P_IP))
675 if (!flow_key->ip.proto)
678 ipv4_key = nla_data(ovs_key);
679 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
682 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
687 case OVS_KEY_ATTR_IPV6:
688 if (flow_key->eth.type != htons(ETH_P_IPV6))
691 if (!flow_key->ip.proto)
694 ipv6_key = nla_data(ovs_key);
695 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
698 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
701 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
706 case OVS_KEY_ATTR_TCP:
707 if (flow_key->ip.proto != IPPROTO_TCP)
710 return validate_tp_port(flow_key);
712 case OVS_KEY_ATTR_UDP:
713 if (flow_key->ip.proto != IPPROTO_UDP)
716 return validate_tp_port(flow_key);
718 case OVS_KEY_ATTR_SCTP:
719 if (flow_key->ip.proto != IPPROTO_SCTP)
722 return validate_tp_port(flow_key);
731 static int validate_userspace(const struct nlattr *attr)
733 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
734 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
735 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
737 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
740 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
741 attr, userspace_policy);
745 if (!a[OVS_USERSPACE_ATTR_PID] ||
746 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
752 static int copy_action(const struct nlattr *from,
753 struct sw_flow_actions **sfa)
755 int totlen = NLA_ALIGN(from->nla_len);
758 to = reserve_sfa_size(sfa, from->nla_len);
762 memcpy(to, from, totlen);
766 static int validate_and_copy_actions(const struct nlattr *attr,
767 const struct sw_flow_key *key,
769 struct sw_flow_actions **sfa)
771 const struct nlattr *a;
774 if (depth >= SAMPLE_ACTION_DEPTH)
777 nla_for_each_nested(a, attr, rem) {
778 /* Expected argument lengths, (u32)-1 for variable length. */
779 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
780 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
781 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
782 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
783 [OVS_ACTION_ATTR_POP_VLAN] = 0,
784 [OVS_ACTION_ATTR_SET] = (u32)-1,
785 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
787 const struct ovs_action_push_vlan *vlan;
788 int type = nla_type(a);
791 if (type > OVS_ACTION_ATTR_MAX ||
792 (action_lens[type] != nla_len(a) &&
793 action_lens[type] != (u32)-1))
798 case OVS_ACTION_ATTR_UNSPEC:
801 case OVS_ACTION_ATTR_USERSPACE:
802 err = validate_userspace(a);
807 case OVS_ACTION_ATTR_OUTPUT:
808 if (nla_get_u32(a) >= DP_MAX_PORTS)
813 case OVS_ACTION_ATTR_POP_VLAN:
816 case OVS_ACTION_ATTR_PUSH_VLAN:
818 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
820 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
824 case OVS_ACTION_ATTR_SET:
825 err = validate_set(a, key, sfa, &skip_copy);
830 case OVS_ACTION_ATTR_SAMPLE:
831 err = validate_and_copy_sample(a, key, depth, sfa);
841 err = copy_action(a, sfa);
853 static void clear_stats(struct sw_flow *flow)
857 flow->packet_count = 0;
858 flow->byte_count = 0;
861 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
863 struct ovs_header *ovs_header = info->userhdr;
864 struct nlattr **a = info->attrs;
865 struct sw_flow_actions *acts;
866 struct sk_buff *packet;
867 struct sw_flow *flow;
874 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
875 !a[OVS_PACKET_ATTR_ACTIONS])
878 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
879 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
883 skb_reserve(packet, NET_IP_ALIGN);
885 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
887 skb_reset_mac_header(packet);
888 eth = eth_hdr(packet);
890 /* Normally, setting the skb 'protocol' field would be handled by a
891 * call to eth_type_trans(), but it assumes there's a sending
892 * device, which we may not have. */
893 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
894 packet->protocol = eth->h_proto;
896 packet->protocol = htons(ETH_P_802_2);
898 /* Build an sw_flow for sending this packet. */
899 flow = ovs_flow_alloc();
904 err = ovs_flow_extract(packet, -1, &flow->key);
908 err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
911 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
916 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
917 rcu_assign_pointer(flow->sf_acts, acts);
921 OVS_CB(packet)->flow = flow;
922 OVS_CB(packet)->pkt_key = &flow->key;
923 packet->priority = flow->key.phy.priority;
924 packet->mark = flow->key.phy.skb_mark;
927 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
933 err = ovs_execute_actions(dp, packet);
937 ovs_flow_free(flow, false);
943 ovs_flow_free(flow, false);
950 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
951 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
952 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
954 [OVS_PACKET_ATTR_PACKET] = { .minlen = ETH_HLEN },
956 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
957 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
960 static struct genl_ops dp_packet_genl_ops[] = {
961 { .cmd = OVS_PACKET_CMD_EXECUTE,
962 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
963 .policy = packet_policy,
964 .doit = ovs_packet_cmd_execute
968 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
970 struct flow_table *table;
973 table = rcu_dereference_check(dp->table, lockdep_ovsl_is_held());
974 stats->n_flows = ovs_flow_tbl_count(table);
976 stats->n_hit = stats->n_missed = stats->n_lost = 0;
977 for_each_possible_cpu(i) {
978 const struct dp_stats_percpu *percpu_stats;
979 struct dp_stats_percpu local_stats;
982 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
985 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
986 local_stats = *percpu_stats;
987 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
989 stats->n_hit += local_stats.n_hit;
990 stats->n_missed += local_stats.n_missed;
991 stats->n_lost += local_stats.n_lost;
995 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
996 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
997 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
998 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1001 static struct genl_family dp_flow_genl_family = {
1002 .id = GENL_ID_GENERATE,
1003 .hdrsize = sizeof(struct ovs_header),
1004 .name = OVS_FLOW_FAMILY,
1005 .version = OVS_FLOW_VERSION,
1006 .maxattr = OVS_FLOW_ATTR_MAX,
1011 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1012 .name = OVS_FLOW_MCGROUP
1015 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1016 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1018 const struct nlattr *a;
1019 struct nlattr *start;
1022 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1026 nla_for_each_nested(a, attr, rem) {
1027 int type = nla_type(a);
1028 struct nlattr *st_sample;
1031 case OVS_SAMPLE_ATTR_PROBABILITY:
1032 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1035 case OVS_SAMPLE_ATTR_ACTIONS:
1036 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1039 err = actions_to_attr(nla_data(a), nla_len(a), skb);
1042 nla_nest_end(skb, st_sample);
1047 nla_nest_end(skb, start);
1051 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1053 const struct nlattr *ovs_key = nla_data(a);
1054 int key_type = nla_type(ovs_key);
1055 struct nlattr *start;
1059 case OVS_KEY_ATTR_IPV4_TUNNEL:
1060 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1064 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
1068 nla_nest_end(skb, start);
1071 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1079 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1081 const struct nlattr *a;
1084 nla_for_each_attr(a, attr, len, rem) {
1085 int type = nla_type(a);
1088 case OVS_ACTION_ATTR_SET:
1089 err = set_action_to_attr(a, skb);
1094 case OVS_ACTION_ATTR_SAMPLE:
1095 err = sample_action_to_attr(a, skb);
1100 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1109 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1111 return NLMSG_ALIGN(sizeof(struct ovs_header))
1112 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1113 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1114 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1115 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1116 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1117 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1120 /* Called with ovs_mutex. */
1121 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1122 struct sk_buff *skb, u32 portid,
1123 u32 seq, u32 flags, u8 cmd)
1125 const int skb_orig_len = skb->len;
1126 struct nlattr *start;
1127 struct ovs_flow_stats stats;
1128 struct ovs_header *ovs_header;
1134 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1138 ovs_header->dp_ifindex = get_dpifindex(dp);
1140 /* Fill flow key. */
1141 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1143 goto nla_put_failure;
1145 err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1146 &flow->unmasked_key, skb);
1149 nla_nest_end(skb, nla);
1151 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1153 goto nla_put_failure;
1155 err = ovs_flow_to_nlattrs(&flow->key, &flow->mask->key, skb);
1159 nla_nest_end(skb, nla);
1161 spin_lock_bh(&flow->lock);
1163 stats.n_packets = flow->packet_count;
1164 stats.n_bytes = flow->byte_count;
1165 tcp_flags = flow->tcp_flags;
1166 spin_unlock_bh(&flow->lock);
1169 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1170 goto nla_put_failure;
1172 if (stats.n_packets &&
1173 nla_put(skb, OVS_FLOW_ATTR_STATS,
1174 sizeof(struct ovs_flow_stats), &stats))
1175 goto nla_put_failure;
1178 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1179 goto nla_put_failure;
1181 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1182 * this is the first flow to be dumped into 'skb'. This is unusual for
1183 * Netlink but individual action lists can be longer than
1184 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1185 * The userspace caller can always fetch the actions separately if it
1186 * really wants them. (Most userspace callers in fact don't care.)
1188 * This can only fail for dump operations because the skb is always
1189 * properly sized for single flows.
1191 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1193 const struct sw_flow_actions *sf_acts;
1195 sf_acts = rcu_dereference_check(flow->sf_acts,
1196 lockdep_ovsl_is_held());
1198 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1200 nla_nest_end(skb, start);
1205 nla_nest_cancel(skb, start);
1207 } else if (skb_orig_len)
1208 goto nla_put_failure;
1210 return genlmsg_end(skb, ovs_header);
1215 genlmsg_cancel(skb, ovs_header);
1219 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1221 const struct sw_flow_actions *sf_acts;
1223 sf_acts = ovsl_dereference(flow->sf_acts);
1225 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1228 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1229 struct datapath *dp,
1230 u32 portid, u32 seq, u8 cmd)
1232 struct sk_buff *skb;
1235 skb = ovs_flow_cmd_alloc_info(flow);
1237 return ERR_PTR(-ENOMEM);
1239 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1244 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1246 struct nlattr **a = info->attrs;
1247 struct ovs_header *ovs_header = info->userhdr;
1248 struct sw_flow_key key, masked_key;
1249 struct sw_flow *flow = NULL;
1250 struct sw_flow_mask mask;
1251 struct sk_buff *reply;
1252 struct datapath *dp;
1253 struct flow_table *table;
1254 struct sw_flow_actions *acts = NULL;
1255 struct sw_flow_match match;
1260 if (!a[OVS_FLOW_ATTR_KEY])
1263 ovs_match_init(&match, &key, &mask);
1264 error = ovs_match_from_nlattrs(&match,
1265 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1269 /* Validate actions. */
1270 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1271 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1272 error = PTR_ERR(acts);
1276 ovs_flow_key_mask(&masked_key, &key, &mask);
1277 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1278 &masked_key, 0, &acts);
1280 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1283 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1289 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1292 goto err_unlock_ovs;
1294 table = ovsl_dereference(dp->table);
1296 /* Check if this is a duplicate flow */
1297 flow = ovs_flow_lookup(table, &key);
1299 struct sw_flow_mask *mask_p;
1300 /* Bail out if we're not allowed to create a new flow. */
1302 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1303 goto err_unlock_ovs;
1305 /* Expand table, if necessary, to make room. */
1306 if (ovs_flow_tbl_need_to_expand(table)) {
1307 struct flow_table *new_table;
1309 new_table = ovs_flow_tbl_expand(table);
1310 if (!IS_ERR(new_table)) {
1311 rcu_assign_pointer(dp->table, new_table);
1312 ovs_flow_tbl_destroy(table, true);
1313 table = ovsl_dereference(dp->table);
1317 /* Allocate flow. */
1318 flow = ovs_flow_alloc();
1320 error = PTR_ERR(flow);
1321 goto err_unlock_ovs;
1325 flow->key = masked_key;
1326 flow->unmasked_key = key;
1328 /* Make sure mask is unique in the system */
1329 mask_p = ovs_sw_flow_mask_find(table, &mask);
1331 /* Allocate a new mask if none exsits. */
1332 mask_p = ovs_sw_flow_mask_alloc();
1335 mask_p->key = mask.key;
1336 mask_p->range = mask.range;
1337 ovs_sw_flow_mask_insert(table, mask_p);
1340 ovs_sw_flow_mask_add_ref(mask_p);
1341 flow->mask = mask_p;
1342 rcu_assign_pointer(flow->sf_acts, acts);
1344 /* Put flow in bucket. */
1345 ovs_flow_insert(table, flow);
1347 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1348 info->snd_seq, OVS_FLOW_CMD_NEW);
1350 /* We found a matching flow. */
1351 struct sw_flow_actions *old_acts;
1353 /* Bail out if we're not allowed to modify an existing flow.
1354 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1355 * because Generic Netlink treats the latter as a dump
1356 * request. We also accept NLM_F_EXCL in case that bug ever
1360 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1361 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1362 goto err_unlock_ovs;
1364 /* The unmasked key has to be the same for flow updates. */
1366 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1367 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1368 goto err_unlock_ovs;
1371 /* Update actions. */
1372 old_acts = ovsl_dereference(flow->sf_acts);
1373 rcu_assign_pointer(flow->sf_acts, acts);
1374 ovs_flow_deferred_free_acts(old_acts);
1376 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1377 info->snd_seq, OVS_FLOW_CMD_NEW);
1380 if (a[OVS_FLOW_ATTR_CLEAR]) {
1381 spin_lock_bh(&flow->lock);
1383 spin_unlock_bh(&flow->lock);
1389 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1391 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1392 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1396 ovs_flow_free(flow, false);
1405 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1407 struct nlattr **a = info->attrs;
1408 struct ovs_header *ovs_header = info->userhdr;
1409 struct sw_flow_key key;
1410 struct sk_buff *reply;
1411 struct sw_flow *flow;
1412 struct datapath *dp;
1413 struct flow_table *table;
1414 struct sw_flow_match match;
1417 if (!a[OVS_FLOW_ATTR_KEY]) {
1418 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1422 ovs_match_init(&match, &key, NULL);
1423 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1428 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1434 table = ovsl_dereference(dp->table);
1435 flow = ovs_flow_lookup_unmasked_key(table, &match);
1441 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1442 info->snd_seq, OVS_FLOW_CMD_NEW);
1443 if (IS_ERR(reply)) {
1444 err = PTR_ERR(reply);
1449 return genlmsg_reply(reply, info);
1455 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1457 struct nlattr **a = info->attrs;
1458 struct ovs_header *ovs_header = info->userhdr;
1459 struct sw_flow_key key;
1460 struct sk_buff *reply;
1461 struct sw_flow *flow;
1462 struct datapath *dp;
1463 struct flow_table *table;
1464 struct sw_flow_match match;
1468 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1474 if (!a[OVS_FLOW_ATTR_KEY]) {
1475 err = flush_flows(dp);
1479 ovs_match_init(&match, &key, NULL);
1480 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1484 table = ovsl_dereference(dp->table);
1485 flow = ovs_flow_lookup_unmasked_key(table, &match);
1491 reply = ovs_flow_cmd_alloc_info(flow);
1497 ovs_flow_remove(table, flow);
1499 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1500 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1503 ovs_flow_free(flow, true);
1506 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1513 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1515 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1516 struct datapath *dp;
1517 struct flow_table *table;
1520 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1526 table = rcu_dereference(dp->table);
1528 struct sw_flow *flow;
1531 bucket = cb->args[0];
1533 flow = ovs_flow_dump_next(table, &bucket, &obj);
1537 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1538 NETLINK_CB(cb->skb).portid,
1539 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1540 OVS_FLOW_CMD_NEW) < 0)
1543 cb->args[0] = bucket;
1550 static struct genl_ops dp_flow_genl_ops[] = {
1551 { .cmd = OVS_FLOW_CMD_NEW,
1552 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1553 .policy = flow_policy,
1554 .doit = ovs_flow_cmd_new_or_set
1556 { .cmd = OVS_FLOW_CMD_DEL,
1557 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1558 .policy = flow_policy,
1559 .doit = ovs_flow_cmd_del
1561 { .cmd = OVS_FLOW_CMD_GET,
1562 .flags = 0, /* OK for unprivileged users. */
1563 .policy = flow_policy,
1564 .doit = ovs_flow_cmd_get,
1565 .dumpit = ovs_flow_cmd_dump
1567 { .cmd = OVS_FLOW_CMD_SET,
1568 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1569 .policy = flow_policy,
1570 .doit = ovs_flow_cmd_new_or_set,
1574 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1575 #ifdef HAVE_NLA_NUL_STRING
1576 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1578 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1581 static struct genl_family dp_datapath_genl_family = {
1582 .id = GENL_ID_GENERATE,
1583 .hdrsize = sizeof(struct ovs_header),
1584 .name = OVS_DATAPATH_FAMILY,
1585 .version = OVS_DATAPATH_VERSION,
1586 .maxattr = OVS_DP_ATTR_MAX,
1591 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1592 .name = OVS_DATAPATH_MCGROUP
1595 static size_t ovs_dp_cmd_msg_size(void)
1597 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1599 msgsize += nla_total_size(IFNAMSIZ);
1600 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1605 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1606 u32 portid, u32 seq, u32 flags, u8 cmd)
1608 struct ovs_header *ovs_header;
1609 struct ovs_dp_stats dp_stats;
1612 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1617 ovs_header->dp_ifindex = get_dpifindex(dp);
1620 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1623 goto nla_put_failure;
1625 get_dp_stats(dp, &dp_stats);
1626 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1627 goto nla_put_failure;
1629 return genlmsg_end(skb, ovs_header);
1632 genlmsg_cancel(skb, ovs_header);
1637 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1640 struct sk_buff *skb;
1643 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1645 return ERR_PTR(-ENOMEM);
1647 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1650 return ERR_PTR(retval);
1655 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1657 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1660 /* Called with ovs_mutex. */
1661 static struct datapath *lookup_datapath(struct net *net,
1662 struct ovs_header *ovs_header,
1663 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1665 struct datapath *dp;
1667 if (!a[OVS_DP_ATTR_NAME])
1668 dp = get_dp(net, ovs_header->dp_ifindex);
1670 struct vport *vport;
1673 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1674 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1677 return dp ? dp : ERR_PTR(-ENODEV);
1680 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1682 struct nlattr **a = info->attrs;
1683 struct vport_parms parms;
1684 struct sk_buff *reply;
1685 struct datapath *dp;
1686 struct vport *vport;
1687 struct ovs_net *ovs_net;
1691 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1694 err = ovs_dp_cmd_validate(a);
1701 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1703 goto err_unlock_ovs;
1705 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1707 /* Allocate table. */
1709 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1713 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1714 if (!dp->stats_percpu) {
1716 goto err_destroy_table;
1719 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1723 goto err_destroy_percpu;
1726 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1727 INIT_HLIST_HEAD(&dp->ports[i]);
1729 /* Set up our datapath device. */
1730 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1731 parms.type = OVS_VPORT_TYPE_INTERNAL;
1732 parms.options = NULL;
1734 parms.port_no = OVSP_LOCAL;
1735 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1737 vport = new_vport(&parms);
1738 if (IS_ERR(vport)) {
1739 err = PTR_ERR(vport);
1743 goto err_destroy_ports_array;
1746 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1747 info->snd_seq, OVS_DP_CMD_NEW);
1748 err = PTR_ERR(reply);
1750 goto err_destroy_local_port;
1752 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1753 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1757 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1760 err_destroy_local_port:
1761 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1762 err_destroy_ports_array:
1765 free_percpu(dp->stats_percpu);
1767 ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1769 release_net(ovs_dp_get_net(dp));
1777 /* Called with ovs_mutex. */
1778 static void __dp_destroy(struct datapath *dp)
1782 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1783 struct vport *vport;
1784 struct hlist_node *n;
1786 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1787 if (vport->port_no != OVSP_LOCAL)
1788 ovs_dp_detach_port(vport);
1791 list_del_rcu(&dp->list_node);
1793 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1794 * all port in datapath are destroyed first before freeing datapath.
1796 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1798 call_rcu(&dp->rcu, destroy_dp_rcu);
1801 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1803 struct sk_buff *reply;
1804 struct datapath *dp;
1807 err = ovs_dp_cmd_validate(info->attrs);
1812 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1817 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1818 info->snd_seq, OVS_DP_CMD_DEL);
1819 err = PTR_ERR(reply);
1826 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1834 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1836 struct sk_buff *reply;
1837 struct datapath *dp;
1840 err = ovs_dp_cmd_validate(info->attrs);
1845 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1850 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1851 info->snd_seq, OVS_DP_CMD_NEW);
1852 if (IS_ERR(reply)) {
1853 err = PTR_ERR(reply);
1854 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1855 ovs_dp_datapath_multicast_group.id, err);
1861 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1869 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1871 struct sk_buff *reply;
1872 struct datapath *dp;
1875 err = ovs_dp_cmd_validate(info->attrs);
1880 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1886 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1887 info->snd_seq, OVS_DP_CMD_NEW);
1888 if (IS_ERR(reply)) {
1889 err = PTR_ERR(reply);
1894 return genlmsg_reply(reply, info);
1901 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1903 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1904 struct datapath *dp;
1905 int skip = cb->args[0];
1909 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1911 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1912 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1913 OVS_DP_CMD_NEW) < 0)
1924 static struct genl_ops dp_datapath_genl_ops[] = {
1925 { .cmd = OVS_DP_CMD_NEW,
1926 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1927 .policy = datapath_policy,
1928 .doit = ovs_dp_cmd_new
1930 { .cmd = OVS_DP_CMD_DEL,
1931 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1932 .policy = datapath_policy,
1933 .doit = ovs_dp_cmd_del
1935 { .cmd = OVS_DP_CMD_GET,
1936 .flags = 0, /* OK for unprivileged users. */
1937 .policy = datapath_policy,
1938 .doit = ovs_dp_cmd_get,
1939 .dumpit = ovs_dp_cmd_dump
1941 { .cmd = OVS_DP_CMD_SET,
1942 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1943 .policy = datapath_policy,
1944 .doit = ovs_dp_cmd_set,
1948 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1949 #ifdef HAVE_NLA_NUL_STRING
1950 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1951 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1953 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1955 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1956 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1957 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1958 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1961 static struct genl_family dp_vport_genl_family = {
1962 .id = GENL_ID_GENERATE,
1963 .hdrsize = sizeof(struct ovs_header),
1964 .name = OVS_VPORT_FAMILY,
1965 .version = OVS_VPORT_VERSION,
1966 .maxattr = OVS_VPORT_ATTR_MAX,
1971 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1972 .name = OVS_VPORT_MCGROUP
1975 /* Called with ovs_mutex or RCU read lock. */
1976 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1977 u32 portid, u32 seq, u32 flags, u8 cmd)
1979 struct ovs_header *ovs_header;
1980 struct ovs_vport_stats vport_stats;
1983 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1988 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1990 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1991 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1992 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1993 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1994 goto nla_put_failure;
1996 ovs_vport_get_stats(vport, &vport_stats);
1997 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1999 goto nla_put_failure;
2001 err = ovs_vport_get_options(vport, skb);
2002 if (err == -EMSGSIZE)
2005 return genlmsg_end(skb, ovs_header);
2010 genlmsg_cancel(skb, ovs_header);
2014 /* Called with ovs_mutex or RCU read lock. */
2015 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
2018 struct sk_buff *skb;
2021 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2023 return ERR_PTR(-ENOMEM);
2025 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
2031 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2033 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
2036 /* Called with ovs_mutex or RCU read lock. */
2037 static struct vport *lookup_vport(struct net *net,
2038 struct ovs_header *ovs_header,
2039 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2041 struct datapath *dp;
2042 struct vport *vport;
2044 if (a[OVS_VPORT_ATTR_NAME]) {
2045 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2047 return ERR_PTR(-ENODEV);
2048 if (ovs_header->dp_ifindex &&
2049 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2050 return ERR_PTR(-ENODEV);
2052 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2053 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2055 if (port_no >= DP_MAX_PORTS)
2056 return ERR_PTR(-EFBIG);
2058 dp = get_dp(net, ovs_header->dp_ifindex);
2060 return ERR_PTR(-ENODEV);
2062 vport = ovs_vport_ovsl_rcu(dp, port_no);
2064 return ERR_PTR(-ENODEV);
2067 return ERR_PTR(-EINVAL);
2070 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2072 struct nlattr **a = info->attrs;
2073 struct ovs_header *ovs_header = info->userhdr;
2074 struct vport_parms parms;
2075 struct sk_buff *reply;
2076 struct vport *vport;
2077 struct datapath *dp;
2082 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2083 !a[OVS_VPORT_ATTR_UPCALL_PID])
2086 err = ovs_vport_cmd_validate(a);
2091 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2096 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2097 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2100 if (port_no >= DP_MAX_PORTS)
2103 vport = ovs_vport_ovsl(dp, port_no);
2108 for (port_no = 1; ; port_no++) {
2109 if (port_no >= DP_MAX_PORTS) {
2113 vport = ovs_vport_ovsl(dp, port_no);
2119 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2120 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2121 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2123 parms.port_no = port_no;
2124 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2126 vport = new_vport(&parms);
2127 err = PTR_ERR(vport);
2132 if (a[OVS_VPORT_ATTR_STATS])
2133 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2135 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2137 if (IS_ERR(reply)) {
2138 err = PTR_ERR(reply);
2139 ovs_dp_detach_port(vport);
2143 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2151 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2153 struct nlattr **a = info->attrs;
2154 struct sk_buff *reply;
2155 struct vport *vport;
2158 err = ovs_vport_cmd_validate(a);
2163 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2164 err = PTR_ERR(vport);
2168 if (a[OVS_VPORT_ATTR_TYPE] &&
2169 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2174 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2180 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2181 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2186 if (a[OVS_VPORT_ATTR_STATS])
2187 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2189 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2190 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2192 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2193 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2197 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2208 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2210 struct nlattr **a = info->attrs;
2211 struct sk_buff *reply;
2212 struct vport *vport;
2215 err = ovs_vport_cmd_validate(a);
2220 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2221 err = PTR_ERR(vport);
2225 if (vport->port_no == OVSP_LOCAL) {
2230 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2231 info->snd_seq, OVS_VPORT_CMD_DEL);
2232 err = PTR_ERR(reply);
2237 ovs_dp_detach_port(vport);
2239 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2247 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2249 struct nlattr **a = info->attrs;
2250 struct ovs_header *ovs_header = info->userhdr;
2251 struct sk_buff *reply;
2252 struct vport *vport;
2255 err = ovs_vport_cmd_validate(a);
2260 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2261 err = PTR_ERR(vport);
2265 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2266 info->snd_seq, OVS_VPORT_CMD_NEW);
2267 err = PTR_ERR(reply);
2273 return genlmsg_reply(reply, info);
2281 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2283 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2284 struct datapath *dp;
2285 int bucket = cb->args[0], skip = cb->args[1];
2288 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2293 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2294 struct vport *vport;
2297 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2299 ovs_vport_cmd_fill_info(vport, skb,
2300 NETLINK_CB(cb->skb).portid,
2303 OVS_VPORT_CMD_NEW) < 0)
2319 static struct genl_ops dp_vport_genl_ops[] = {
2320 { .cmd = OVS_VPORT_CMD_NEW,
2321 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2322 .policy = vport_policy,
2323 .doit = ovs_vport_cmd_new
2325 { .cmd = OVS_VPORT_CMD_DEL,
2326 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2327 .policy = vport_policy,
2328 .doit = ovs_vport_cmd_del
2330 { .cmd = OVS_VPORT_CMD_GET,
2331 .flags = 0, /* OK for unprivileged users. */
2332 .policy = vport_policy,
2333 .doit = ovs_vport_cmd_get,
2334 .dumpit = ovs_vport_cmd_dump
2336 { .cmd = OVS_VPORT_CMD_SET,
2337 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2338 .policy = vport_policy,
2339 .doit = ovs_vport_cmd_set,
2343 struct genl_family_and_ops {
2344 struct genl_family *family;
2345 struct genl_ops *ops;
2347 struct genl_multicast_group *group;
2350 static const struct genl_family_and_ops dp_genl_families[] = {
2351 { &dp_datapath_genl_family,
2352 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2353 &ovs_dp_datapath_multicast_group },
2354 { &dp_vport_genl_family,
2355 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2356 &ovs_dp_vport_multicast_group },
2357 { &dp_flow_genl_family,
2358 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2359 &ovs_dp_flow_multicast_group },
2360 { &dp_packet_genl_family,
2361 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2365 static void dp_unregister_genl(int n_families)
2369 for (i = 0; i < n_families; i++)
2370 genl_unregister_family(dp_genl_families[i].family);
2373 static int dp_register_genl(void)
2380 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2381 const struct genl_family_and_ops *f = &dp_genl_families[i];
2383 err = genl_register_family_with_ops(f->family, f->ops,
2390 err = genl_register_mc_group(f->family, f->group);
2399 dp_unregister_genl(n_registered);
2403 static void rehash_flow_table(struct work_struct *work)
2405 struct datapath *dp;
2411 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2413 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2414 struct flow_table *old_table = ovsl_dereference(dp->table);
2415 struct flow_table *new_table;
2417 new_table = ovs_flow_tbl_rehash(old_table);
2418 if (!IS_ERR(new_table)) {
2419 rcu_assign_pointer(dp->table, new_table);
2420 ovs_flow_tbl_destroy(old_table, true);
2426 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2429 static int __net_init ovs_init_net(struct net *net)
2431 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2433 INIT_LIST_HEAD(&ovs_net->dps);
2434 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2438 static void __net_exit ovs_exit_net(struct net *net)
2440 struct datapath *dp, *dp_next;
2441 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2444 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2448 cancel_work_sync(&ovs_net->dp_notify_work);
2451 static struct pernet_operations ovs_net_ops = {
2452 .init = ovs_init_net,
2453 .exit = ovs_exit_net,
2455 .size = sizeof(struct ovs_net),
2458 DEFINE_COMPAT_PNET_REG_FUNC(device);
2460 static int __init dp_init(void)
2464 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2466 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2469 err = ovs_workqueues_init();
2473 err = ovs_flow_init();
2477 err = ovs_vport_init();
2479 goto error_flow_exit;
2481 err = register_pernet_device(&ovs_net_ops);
2483 goto error_vport_exit;
2485 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2487 goto error_netns_exit;
2489 err = dp_register_genl();
2491 goto error_unreg_notifier;
2493 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2497 error_unreg_notifier:
2498 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2500 unregister_pernet_device(&ovs_net_ops);
2506 ovs_workqueues_exit();
2511 static void dp_cleanup(void)
2513 cancel_delayed_work_sync(&rehash_flow_wq);
2514 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2515 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2516 unregister_pernet_device(&ovs_net_ops);
2520 ovs_workqueues_exit();
2523 module_init(dp_init);
2524 module_exit(dp_cleanup);
2526 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2527 MODULE_LICENSE("GPL");
2528 MODULE_VERSION(VERSION);