2 * Copyright (c) 2007-2012 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/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>
62 #include "vport-internal_dev.h"
63 #include "vport-netdev.h"
65 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
66 LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0)
67 #error Kernels before 2.6.18 or after 3.8 are not supported by this version of Open vSwitch.
70 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
71 static void rehash_flow_table(struct work_struct *work);
72 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
74 int ovs_net_id __read_mostly;
76 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
77 struct genl_multicast_group *grp)
79 genl_notify(skb, genl_info_net(info), info->snd_portid,
80 grp->id, info->nlhdr, GFP_KERNEL);
86 * All writes e.g. Writes to device state (add/remove datapath, port, set
87 * operations on vports, etc.), Writes to other state (flow table
88 * modifications, set miscellaneous datapath parameters, etc.) are protected
91 * Reads are protected by RCU.
93 * There are a few special cases (mostly stats) that have their own
94 * synchronization but they nest under all of above and don't interact with
97 * The RTNL lock nests inside ovs_mutex.
100 static DEFINE_MUTEX(ovs_mutex);
104 mutex_lock(&ovs_mutex);
107 void ovs_unlock(void)
109 mutex_unlock(&ovs_mutex);
112 #ifdef CONFIG_LOCKDEP
113 int lockdep_ovsl_is_held(void)
116 return lockdep_is_held(&ovs_mutex);
122 static struct vport *new_vport(const struct vport_parms *);
123 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
124 const struct dp_upcall_info *);
125 static int queue_userspace_packet(struct net *, int dp_ifindex,
127 const struct dp_upcall_info *);
129 /* Must be called with rcu_read_lock or ovs_mutex. */
130 static struct datapath *get_dp(struct net *net, int dp_ifindex)
132 struct datapath *dp = NULL;
133 struct net_device *dev;
136 dev = dev_get_by_index_rcu(net, dp_ifindex);
138 struct vport *vport = ovs_internal_dev_get_vport(dev);
147 /* Must be called with rcu_read_lock or ovs_mutex. */
148 const char *ovs_dp_name(const struct datapath *dp)
150 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
151 return vport->ops->get_name(vport);
154 static int get_dpifindex(struct datapath *dp)
161 local = ovs_vport_rcu(dp, OVSP_LOCAL);
163 ifindex = netdev_vport_priv(local)->dev->ifindex;
172 static void destroy_dp_rcu(struct rcu_head *rcu)
174 struct datapath *dp = container_of(rcu, struct datapath, rcu);
176 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
177 free_percpu(dp->stats_percpu);
178 release_net(ovs_dp_get_net(dp));
183 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
186 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
192 struct hlist_head *head;
194 head = vport_hash_bucket(dp, port_no);
195 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
196 if (vport->port_no == port_no)
202 /* Called with ovs_mutex. */
203 static struct vport *new_vport(const struct vport_parms *parms)
207 vport = ovs_vport_add(parms);
208 if (!IS_ERR(vport)) {
209 struct datapath *dp = parms->dp;
210 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
212 hlist_add_head_rcu(&vport->dp_hash_node, head);
217 void ovs_dp_detach_port(struct vport *p)
221 /* First drop references to device. */
222 hlist_del_rcu(&p->dp_hash_node);
224 /* Then destroy it. */
228 /* Must be called with rcu_read_lock. */
229 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
231 struct datapath *dp = p->dp;
232 struct sw_flow *flow;
233 struct dp_stats_percpu *stats;
234 struct sw_flow_key key;
239 stats = this_cpu_ptr(dp->stats_percpu);
241 /* Extract flow from 'skb' into 'key'. */
242 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
243 if (unlikely(error)) {
249 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
250 if (unlikely(!flow)) {
251 struct dp_upcall_info upcall;
253 upcall.cmd = OVS_PACKET_CMD_MISS;
255 upcall.userdata = NULL;
256 upcall.portid = p->upcall_portid;
257 ovs_dp_upcall(dp, skb, &upcall);
259 stats_counter = &stats->n_missed;
263 OVS_CB(skb)->flow = flow;
265 stats_counter = &stats->n_hit;
266 ovs_flow_used(OVS_CB(skb)->flow, skb);
267 ovs_execute_actions(dp, skb);
270 /* Update datapath statistics. */
271 u64_stats_update_begin(&stats->sync);
273 u64_stats_update_end(&stats->sync);
276 static struct genl_family dp_packet_genl_family = {
277 .id = GENL_ID_GENERATE,
278 .hdrsize = sizeof(struct ovs_header),
279 .name = OVS_PACKET_FAMILY,
280 .version = OVS_PACKET_VERSION,
281 .maxattr = OVS_PACKET_ATTR_MAX,
285 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
286 const struct dp_upcall_info *upcall_info)
288 struct dp_stats_percpu *stats;
292 if (upcall_info->portid == 0) {
297 dp_ifindex = get_dpifindex(dp);
303 forward_ip_summed(skb, true);
305 if (!skb_is_gso(skb))
306 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
308 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
315 stats = this_cpu_ptr(dp->stats_percpu);
317 u64_stats_update_begin(&stats->sync);
319 u64_stats_update_end(&stats->sync);
324 static int queue_gso_packets(struct net *net, int dp_ifindex,
326 const struct dp_upcall_info *upcall_info)
328 unsigned short gso_type = skb_shinfo(skb)->gso_type;
329 struct dp_upcall_info later_info;
330 struct sw_flow_key later_key;
331 struct sk_buff *segs, *nskb;
334 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
336 return PTR_ERR(segs);
338 /* Queue all of the segments. */
341 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
345 if (skb == segs && gso_type & SKB_GSO_UDP) {
346 /* The initial flow key extracted by ovs_flow_extract()
347 * in this case is for a first fragment, so we need to
348 * properly mark later fragments.
350 later_key = *upcall_info->key;
351 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
353 later_info = *upcall_info;
354 later_info.key = &later_key;
355 upcall_info = &later_info;
357 } while ((skb = skb->next));
359 /* Free all of the segments. */
367 } while ((skb = nskb));
371 static size_t key_attr_size(void)
373 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
374 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
375 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
376 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
377 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
378 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
379 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
380 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
381 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
382 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
383 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
384 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
385 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
386 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
387 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
388 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
389 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
390 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
391 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
394 static size_t upcall_msg_size(const struct sk_buff *skb,
395 const struct nlattr *userdata)
397 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
398 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
399 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
401 /* OVS_PACKET_ATTR_USERDATA */
403 size += NLA_ALIGN(userdata->nla_len);
408 static int queue_userspace_packet(struct net *net, int dp_ifindex,
410 const struct dp_upcall_info *upcall_info)
412 struct ovs_header *upcall;
413 struct sk_buff *nskb = NULL;
414 struct sk_buff *user_skb; /* to be queued to userspace */
418 if (vlan_tx_tag_present(skb)) {
419 nskb = skb_clone(skb, GFP_ATOMIC);
423 err = vlan_deaccel_tag(nskb);
430 if (nla_attr_size(skb->len) > USHRT_MAX) {
435 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
441 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
442 0, upcall_info->cmd);
443 upcall->dp_ifindex = dp_ifindex;
445 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
446 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
447 nla_nest_end(user_skb, nla);
449 if (upcall_info->userdata)
450 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
451 nla_len(upcall_info->userdata),
452 nla_data(upcall_info->userdata));
454 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
456 skb_copy_and_csum_dev(skb, nla_data(nla));
458 genlmsg_end(user_skb, upcall);
459 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
466 /* Called with ovs_mutex. */
467 static int flush_flows(struct datapath *dp)
469 struct flow_table *old_table;
470 struct flow_table *new_table;
472 old_table = ovsl_dereference(dp->table);
473 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
477 rcu_assign_pointer(dp->table, new_table);
479 ovs_flow_tbl_deferred_destroy(old_table);
483 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
486 struct sw_flow_actions *acts;
488 int req_size = NLA_ALIGN(attr_len);
489 int next_offset = offsetof(struct sw_flow_actions, actions) +
492 if (req_size <= (ksize(*sfa) - next_offset))
495 new_acts_size = ksize(*sfa) * 2;
497 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
498 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
499 return ERR_PTR(-EMSGSIZE);
500 new_acts_size = MAX_ACTIONS_BUFSIZE;
503 acts = ovs_flow_actions_alloc(new_acts_size);
507 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
508 acts->actions_len = (*sfa)->actions_len;
513 (*sfa)->actions_len += req_size;
514 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
517 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
521 a = reserve_sfa_size(sfa, nla_attr_size(len));
525 a->nla_type = attrtype;
526 a->nla_len = nla_attr_size(len);
529 memcpy(nla_data(a), data, len);
530 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
535 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
537 int used = (*sfa)->actions_len;
540 err = add_action(sfa, attrtype, NULL, 0);
547 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
549 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
551 a->nla_len = sfa->actions_len - st_offset;
554 static int validate_and_copy_actions(const struct nlattr *attr,
555 const struct sw_flow_key *key, int depth,
556 struct sw_flow_actions **sfa);
558 static int validate_and_copy_sample(const struct nlattr *attr,
559 const struct sw_flow_key *key, int depth,
560 struct sw_flow_actions **sfa)
562 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
563 const struct nlattr *probability, *actions;
564 const struct nlattr *a;
565 int rem, start, err, st_acts;
567 memset(attrs, 0, sizeof(attrs));
568 nla_for_each_nested(a, attr, rem) {
569 int type = nla_type(a);
570 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
577 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
578 if (!probability || nla_len(probability) != sizeof(u32))
581 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
582 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
585 /* validation done, copy sample action. */
586 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
589 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
592 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
596 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
600 add_nested_action_end(*sfa, st_acts);
601 add_nested_action_end(*sfa, start);
606 static int validate_tp_port(const struct sw_flow_key *flow_key)
608 if (flow_key->eth.type == htons(ETH_P_IP)) {
609 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
611 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
612 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
619 static int validate_and_copy_set_tun(const struct nlattr *attr,
620 struct sw_flow_actions **sfa)
622 struct ovs_key_ipv4_tunnel tun_key;
625 err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
629 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
633 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
634 add_nested_action_end(*sfa, start);
639 static int validate_set(const struct nlattr *a,
640 const struct sw_flow_key *flow_key,
641 struct sw_flow_actions **sfa,
644 const struct nlattr *ovs_key = nla_data(a);
645 int key_type = nla_type(ovs_key);
647 /* There can be only one key in a action */
648 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
651 if (key_type > OVS_KEY_ATTR_MAX ||
652 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
653 ovs_key_lens[key_type] != -1))
657 const struct ovs_key_ipv4 *ipv4_key;
658 const struct ovs_key_ipv6 *ipv6_key;
661 case OVS_KEY_ATTR_PRIORITY:
662 case OVS_KEY_ATTR_ETHERNET:
665 case OVS_KEY_ATTR_SKB_MARK:
666 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
667 if (nla_get_u32(ovs_key) != 0)
672 case OVS_KEY_ATTR_TUNNEL:
674 err = validate_and_copy_set_tun(a, sfa);
679 case OVS_KEY_ATTR_IPV4:
680 if (flow_key->eth.type != htons(ETH_P_IP))
683 if (!flow_key->ip.proto)
686 ipv4_key = nla_data(ovs_key);
687 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
690 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
695 case OVS_KEY_ATTR_IPV6:
696 if (flow_key->eth.type != htons(ETH_P_IPV6))
699 if (!flow_key->ip.proto)
702 ipv6_key = nla_data(ovs_key);
703 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
706 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
709 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
714 case OVS_KEY_ATTR_TCP:
715 if (flow_key->ip.proto != IPPROTO_TCP)
718 return validate_tp_port(flow_key);
720 case OVS_KEY_ATTR_UDP:
721 if (flow_key->ip.proto != IPPROTO_UDP)
724 return validate_tp_port(flow_key);
733 static int validate_userspace(const struct nlattr *attr)
735 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
736 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
737 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
739 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
742 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
743 attr, userspace_policy);
747 if (!a[OVS_USERSPACE_ATTR_PID] ||
748 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
754 static int copy_action(const struct nlattr *from,
755 struct sw_flow_actions **sfa)
757 int totlen = NLA_ALIGN(from->nla_len);
760 to = reserve_sfa_size(sfa, from->nla_len);
764 memcpy(to, from, totlen);
768 static int validate_and_copy_actions(const struct nlattr *attr,
769 const struct sw_flow_key *key,
771 struct sw_flow_actions **sfa)
773 const struct nlattr *a;
776 if (depth >= SAMPLE_ACTION_DEPTH)
779 nla_for_each_nested(a, attr, rem) {
780 /* Expected argument lengths, (u32)-1 for variable length. */
781 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
782 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
783 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
784 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
785 [OVS_ACTION_ATTR_POP_VLAN] = 0,
786 [OVS_ACTION_ATTR_SET] = (u32)-1,
787 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
789 const struct ovs_action_push_vlan *vlan;
790 int type = nla_type(a);
793 if (type > OVS_ACTION_ATTR_MAX ||
794 (action_lens[type] != nla_len(a) &&
795 action_lens[type] != (u32)-1))
800 case OVS_ACTION_ATTR_UNSPEC:
803 case OVS_ACTION_ATTR_USERSPACE:
804 err = validate_userspace(a);
809 case OVS_ACTION_ATTR_OUTPUT:
810 if (nla_get_u32(a) >= DP_MAX_PORTS)
815 case OVS_ACTION_ATTR_POP_VLAN:
818 case OVS_ACTION_ATTR_PUSH_VLAN:
820 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
822 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
826 case OVS_ACTION_ATTR_SET:
827 err = validate_set(a, key, sfa, &skip_copy);
832 case OVS_ACTION_ATTR_SAMPLE:
833 err = validate_and_copy_sample(a, key, depth, sfa);
843 err = copy_action(a, sfa);
855 static void clear_stats(struct sw_flow *flow)
859 flow->packet_count = 0;
860 flow->byte_count = 0;
863 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
865 struct ovs_header *ovs_header = info->userhdr;
866 struct nlattr **a = info->attrs;
867 struct sw_flow_actions *acts;
868 struct sk_buff *packet;
869 struct sw_flow *flow;
877 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
878 !a[OVS_PACKET_ATTR_ACTIONS])
881 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
882 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
886 skb_reserve(packet, NET_IP_ALIGN);
888 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
890 skb_reset_mac_header(packet);
891 eth = eth_hdr(packet);
893 /* Normally, setting the skb 'protocol' field would be handled by a
894 * call to eth_type_trans(), but it assumes there's a sending
895 * device, which we may not have. */
896 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
897 packet->protocol = eth->h_proto;
899 packet->protocol = htons(ETH_P_802_2);
901 /* Build an sw_flow for sending this packet. */
902 flow = ovs_flow_alloc();
907 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
911 err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
914 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
919 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
920 rcu_assign_pointer(flow->sf_acts, acts);
924 OVS_CB(packet)->flow = flow;
925 packet->priority = flow->key.phy.priority;
926 skb_set_mark(packet, flow->key.phy.skb_mark);
929 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
935 err = ovs_execute_actions(dp, packet);
952 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
953 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
954 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
956 [OVS_PACKET_ATTR_PACKET] = { .minlen = ETH_HLEN },
958 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
959 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
962 static struct genl_ops dp_packet_genl_ops[] = {
963 { .cmd = OVS_PACKET_CMD_EXECUTE,
964 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
965 .policy = packet_policy,
966 .doit = ovs_packet_cmd_execute
970 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
973 struct flow_table *table = ovsl_dereference(dp->table);
975 stats->n_flows = ovs_flow_tbl_count(table);
977 stats->n_hit = stats->n_missed = stats->n_lost = 0;
978 for_each_possible_cpu(i) {
979 const struct dp_stats_percpu *percpu_stats;
980 struct dp_stats_percpu local_stats;
983 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
986 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
987 local_stats = *percpu_stats;
988 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
990 stats->n_hit += local_stats.n_hit;
991 stats->n_missed += local_stats.n_missed;
992 stats->n_lost += local_stats.n_lost;
996 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
997 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
998 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
999 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1002 static struct genl_family dp_flow_genl_family = {
1003 .id = GENL_ID_GENERATE,
1004 .hdrsize = sizeof(struct ovs_header),
1005 .name = OVS_FLOW_FAMILY,
1006 .version = OVS_FLOW_VERSION,
1007 .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 = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1067 nla_nest_end(skb, start);
1070 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1078 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1080 const struct nlattr *a;
1083 nla_for_each_attr(a, attr, len, rem) {
1084 int type = nla_type(a);
1087 case OVS_ACTION_ATTR_SET:
1088 err = set_action_to_attr(a, skb);
1093 case OVS_ACTION_ATTR_SAMPLE:
1094 err = sample_action_to_attr(a, skb);
1099 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1108 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1110 return NLMSG_ALIGN(sizeof(struct ovs_header))
1111 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1112 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1113 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1114 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1115 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1118 /* Called with ovs_mutex. */
1119 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1120 struct sk_buff *skb, u32 portid,
1121 u32 seq, u32 flags, u8 cmd)
1123 const int skb_orig_len = skb->len;
1124 const struct sw_flow_actions *sf_acts;
1125 struct nlattr *start;
1126 struct ovs_flow_stats stats;
1127 struct ovs_header *ovs_header;
1133 sf_acts = ovsl_dereference(flow->sf_acts);
1135 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1139 ovs_header->dp_ifindex = get_dpifindex(dp);
1141 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1143 goto nla_put_failure;
1144 err = ovs_flow_to_nlattrs(&flow->key, skb);
1147 nla_nest_end(skb, nla);
1149 spin_lock_bh(&flow->lock);
1151 stats.n_packets = flow->packet_count;
1152 stats.n_bytes = flow->byte_count;
1153 tcp_flags = flow->tcp_flags;
1154 spin_unlock_bh(&flow->lock);
1157 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1158 goto nla_put_failure;
1160 if (stats.n_packets &&
1161 nla_put(skb, OVS_FLOW_ATTR_STATS,
1162 sizeof(struct ovs_flow_stats), &stats))
1163 goto nla_put_failure;
1166 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1167 goto nla_put_failure;
1169 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1170 * this is the first flow to be dumped into 'skb'. This is unusual for
1171 * Netlink but individual action lists can be longer than
1172 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1173 * The userspace caller can always fetch the actions separately if it
1174 * really wants them. (Most userspace callers in fact don't care.)
1176 * This can only fail for dump operations because the skb is always
1177 * properly sized for single flows.
1179 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1181 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1183 nla_nest_end(skb, start);
1188 nla_nest_cancel(skb, start);
1190 } else if (skb_orig_len)
1191 goto nla_put_failure;
1193 return genlmsg_end(skb, ovs_header);
1198 genlmsg_cancel(skb, ovs_header);
1202 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1204 const struct sw_flow_actions *sf_acts;
1206 sf_acts = ovsl_dereference(flow->sf_acts);
1208 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1211 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1212 struct datapath *dp,
1213 u32 portid, u32 seq, u8 cmd)
1215 struct sk_buff *skb;
1218 skb = ovs_flow_cmd_alloc_info(flow);
1220 return ERR_PTR(-ENOMEM);
1222 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1227 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1229 struct nlattr **a = info->attrs;
1230 struct ovs_header *ovs_header = info->userhdr;
1231 struct sw_flow_key key;
1232 struct sw_flow *flow;
1233 struct sk_buff *reply;
1234 struct datapath *dp;
1235 struct flow_table *table;
1236 struct sw_flow_actions *acts = NULL;
1242 if (!a[OVS_FLOW_ATTR_KEY])
1244 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1248 /* Validate actions. */
1249 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1250 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1251 error = PTR_ERR(acts);
1255 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0, &acts);
1258 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1264 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1267 goto err_unlock_ovs;
1269 table = ovsl_dereference(dp->table);
1270 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1272 /* Bail out if we're not allowed to create a new flow. */
1274 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1275 goto err_unlock_ovs;
1277 /* Expand table, if necessary, to make room. */
1278 if (ovs_flow_tbl_need_to_expand(table)) {
1279 struct flow_table *new_table;
1281 new_table = ovs_flow_tbl_expand(table);
1282 if (!IS_ERR(new_table)) {
1283 rcu_assign_pointer(dp->table, new_table);
1284 ovs_flow_tbl_deferred_destroy(table);
1285 table = ovsl_dereference(dp->table);
1289 /* Allocate flow. */
1290 flow = ovs_flow_alloc();
1292 error = PTR_ERR(flow);
1293 goto err_unlock_ovs;
1297 rcu_assign_pointer(flow->sf_acts, acts);
1299 /* Put flow in bucket. */
1300 ovs_flow_tbl_insert(table, flow, &key, key_len);
1302 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1306 /* We found a matching flow. */
1307 struct sw_flow_actions *old_acts;
1309 /* Bail out if we're not allowed to modify an existing flow.
1310 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1311 * because Generic Netlink treats the latter as a dump
1312 * request. We also accept NLM_F_EXCL in case that bug ever
1316 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1317 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1318 goto err_unlock_ovs;
1320 /* Update actions. */
1321 old_acts = ovsl_dereference(flow->sf_acts);
1322 rcu_assign_pointer(flow->sf_acts, acts);
1323 ovs_flow_deferred_free_acts(old_acts);
1325 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1326 info->snd_seq, OVS_FLOW_CMD_NEW);
1329 if (a[OVS_FLOW_ATTR_CLEAR]) {
1330 spin_lock_bh(&flow->lock);
1332 spin_unlock_bh(&flow->lock);
1338 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1340 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1341 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1352 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1354 struct nlattr **a = info->attrs;
1355 struct ovs_header *ovs_header = info->userhdr;
1356 struct sw_flow_key key;
1357 struct sk_buff *reply;
1358 struct sw_flow *flow;
1359 struct datapath *dp;
1360 struct flow_table *table;
1364 if (!a[OVS_FLOW_ATTR_KEY])
1366 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1371 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1377 table = ovsl_dereference(dp->table);
1378 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1384 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1385 info->snd_seq, OVS_FLOW_CMD_NEW);
1386 if (IS_ERR(reply)) {
1387 err = PTR_ERR(reply);
1392 return genlmsg_reply(reply, info);
1398 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1400 struct nlattr **a = info->attrs;
1401 struct ovs_header *ovs_header = info->userhdr;
1402 struct sw_flow_key key;
1403 struct sk_buff *reply;
1404 struct sw_flow *flow;
1405 struct datapath *dp;
1406 struct flow_table *table;
1411 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1417 if (!a[OVS_FLOW_ATTR_KEY]) {
1418 err = flush_flows(dp);
1421 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1425 table = ovsl_dereference(dp->table);
1426 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1432 reply = ovs_flow_cmd_alloc_info(flow);
1438 ovs_flow_tbl_remove(table, flow);
1440 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1441 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1444 ovs_flow_deferred_free(flow);
1447 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1454 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1456 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1457 struct datapath *dp;
1458 struct flow_table *table;
1461 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1467 table = ovsl_dereference(dp->table);
1470 struct sw_flow *flow;
1473 bucket = cb->args[0];
1475 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1479 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1480 NETLINK_CB(cb->skb).portid,
1481 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1482 OVS_FLOW_CMD_NEW) < 0)
1485 cb->args[0] = bucket;
1492 static struct genl_ops dp_flow_genl_ops[] = {
1493 { .cmd = OVS_FLOW_CMD_NEW,
1494 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1495 .policy = flow_policy,
1496 .doit = ovs_flow_cmd_new_or_set
1498 { .cmd = OVS_FLOW_CMD_DEL,
1499 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1500 .policy = flow_policy,
1501 .doit = ovs_flow_cmd_del
1503 { .cmd = OVS_FLOW_CMD_GET,
1504 .flags = 0, /* OK for unprivileged users. */
1505 .policy = flow_policy,
1506 .doit = ovs_flow_cmd_get,
1507 .dumpit = ovs_flow_cmd_dump
1509 { .cmd = OVS_FLOW_CMD_SET,
1510 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1511 .policy = flow_policy,
1512 .doit = ovs_flow_cmd_new_or_set,
1516 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1517 #ifdef HAVE_NLA_NUL_STRING
1518 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1520 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1523 static struct genl_family dp_datapath_genl_family = {
1524 .id = GENL_ID_GENERATE,
1525 .hdrsize = sizeof(struct ovs_header),
1526 .name = OVS_DATAPATH_FAMILY,
1527 .version = OVS_DATAPATH_VERSION,
1528 .maxattr = OVS_DP_ATTR_MAX,
1532 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1533 .name = OVS_DATAPATH_MCGROUP
1536 static size_t ovs_dp_cmd_msg_size(void)
1538 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1540 msgsize += nla_total_size(IFNAMSIZ);
1541 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1546 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1547 u32 portid, u32 seq, u32 flags, u8 cmd)
1549 struct ovs_header *ovs_header;
1550 struct ovs_dp_stats dp_stats;
1553 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1558 ovs_header->dp_ifindex = get_dpifindex(dp);
1561 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1564 goto nla_put_failure;
1566 get_dp_stats(dp, &dp_stats);
1567 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1568 goto nla_put_failure;
1570 return genlmsg_end(skb, ovs_header);
1573 genlmsg_cancel(skb, ovs_header);
1578 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1581 struct sk_buff *skb;
1584 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1586 return ERR_PTR(-ENOMEM);
1588 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1591 return ERR_PTR(retval);
1596 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1598 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1601 /* Called with ovs_mutex. */
1602 static struct datapath *lookup_datapath(struct net *net,
1603 struct ovs_header *ovs_header,
1604 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1606 struct datapath *dp;
1608 if (!a[OVS_DP_ATTR_NAME])
1609 dp = get_dp(net, ovs_header->dp_ifindex);
1611 struct vport *vport;
1614 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1615 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1618 return dp ? dp : ERR_PTR(-ENODEV);
1621 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1623 struct nlattr **a = info->attrs;
1624 struct vport_parms parms;
1625 struct sk_buff *reply;
1626 struct datapath *dp;
1627 struct vport *vport;
1628 struct ovs_net *ovs_net;
1632 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1635 err = ovs_dp_cmd_validate(a);
1642 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1644 goto err_unlock_ovs;
1646 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1648 /* Allocate table. */
1650 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1654 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1655 if (!dp->stats_percpu) {
1657 goto err_destroy_table;
1660 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1664 goto err_destroy_percpu;
1667 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1668 INIT_HLIST_HEAD(&dp->ports[i]);
1670 /* Set up our datapath device. */
1671 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1672 parms.type = OVS_VPORT_TYPE_INTERNAL;
1673 parms.options = NULL;
1675 parms.port_no = OVSP_LOCAL;
1676 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1678 vport = new_vport(&parms);
1679 if (IS_ERR(vport)) {
1680 err = PTR_ERR(vport);
1684 goto err_destroy_ports_array;
1687 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1688 info->snd_seq, OVS_DP_CMD_NEW);
1689 err = PTR_ERR(reply);
1691 goto err_destroy_local_port;
1693 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1694 list_add_tail(&dp->list_node, &ovs_net->dps);
1698 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1701 err_destroy_local_port:
1702 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1703 err_destroy_ports_array:
1706 free_percpu(dp->stats_percpu);
1708 ovs_flow_tbl_destroy(ovsl_dereference(dp->table));
1710 release_net(ovs_dp_get_net(dp));
1718 /* Called with ovs_mutex. */
1719 static void __dp_destroy(struct datapath *dp)
1723 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1724 struct vport *vport;
1725 struct hlist_node *n;
1727 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1728 if (vport->port_no != OVSP_LOCAL)
1729 ovs_dp_detach_port(vport);
1732 list_del(&dp->list_node);
1734 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1735 * all port in datapath are destroyed first before freeing datapath.
1737 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1739 call_rcu(&dp->rcu, destroy_dp_rcu);
1742 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1744 struct sk_buff *reply;
1745 struct datapath *dp;
1748 err = ovs_dp_cmd_validate(info->attrs);
1753 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1758 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1759 info->snd_seq, OVS_DP_CMD_DEL);
1760 err = PTR_ERR(reply);
1767 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1775 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1777 struct sk_buff *reply;
1778 struct datapath *dp;
1781 err = ovs_dp_cmd_validate(info->attrs);
1786 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1791 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1792 info->snd_seq, OVS_DP_CMD_NEW);
1793 if (IS_ERR(reply)) {
1794 err = PTR_ERR(reply);
1795 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1796 ovs_dp_datapath_multicast_group.id, err);
1802 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1810 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1812 struct sk_buff *reply;
1813 struct datapath *dp;
1816 err = ovs_dp_cmd_validate(info->attrs);
1821 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);
1835 return genlmsg_reply(reply, info);
1842 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1844 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1845 struct datapath *dp;
1846 int skip = cb->args[0];
1850 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1852 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1853 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1854 OVS_DP_CMD_NEW) < 0)
1865 static struct genl_ops dp_datapath_genl_ops[] = {
1866 { .cmd = OVS_DP_CMD_NEW,
1867 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1868 .policy = datapath_policy,
1869 .doit = ovs_dp_cmd_new
1871 { .cmd = OVS_DP_CMD_DEL,
1872 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1873 .policy = datapath_policy,
1874 .doit = ovs_dp_cmd_del
1876 { .cmd = OVS_DP_CMD_GET,
1877 .flags = 0, /* OK for unprivileged users. */
1878 .policy = datapath_policy,
1879 .doit = ovs_dp_cmd_get,
1880 .dumpit = ovs_dp_cmd_dump
1882 { .cmd = OVS_DP_CMD_SET,
1883 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1884 .policy = datapath_policy,
1885 .doit = ovs_dp_cmd_set,
1889 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1890 #ifdef HAVE_NLA_NUL_STRING
1891 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1892 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1894 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1896 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1897 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1898 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1899 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1902 static struct genl_family dp_vport_genl_family = {
1903 .id = GENL_ID_GENERATE,
1904 .hdrsize = sizeof(struct ovs_header),
1905 .name = OVS_VPORT_FAMILY,
1906 .version = OVS_VPORT_VERSION,
1907 .maxattr = OVS_VPORT_ATTR_MAX,
1911 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1912 .name = OVS_VPORT_MCGROUP
1915 /* Called with ovs_mutex or RCU read lock. */
1916 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1917 u32 portid, u32 seq, u32 flags, u8 cmd)
1919 struct ovs_header *ovs_header;
1920 struct ovs_vport_stats vport_stats;
1923 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1928 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1930 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1931 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1932 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1933 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1934 goto nla_put_failure;
1936 ovs_vport_get_stats(vport, &vport_stats);
1937 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1939 goto nla_put_failure;
1941 err = ovs_vport_get_options(vport, skb);
1942 if (err == -EMSGSIZE)
1945 return genlmsg_end(skb, ovs_header);
1950 genlmsg_cancel(skb, ovs_header);
1954 /* Called with ovs_mutex or RCU read lock. */
1955 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1958 struct sk_buff *skb;
1961 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1963 return ERR_PTR(-ENOMEM);
1965 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1971 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1973 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1976 /* Called with ovs_mutex or RCU read lock. */
1977 static struct vport *lookup_vport(struct net *net,
1978 struct ovs_header *ovs_header,
1979 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1981 struct datapath *dp;
1982 struct vport *vport;
1984 if (a[OVS_VPORT_ATTR_NAME]) {
1985 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1987 return ERR_PTR(-ENODEV);
1988 if (ovs_header->dp_ifindex &&
1989 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1990 return ERR_PTR(-ENODEV);
1992 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1993 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1995 if (port_no >= DP_MAX_PORTS)
1996 return ERR_PTR(-EFBIG);
1998 dp = get_dp(net, ovs_header->dp_ifindex);
2000 return ERR_PTR(-ENODEV);
2002 vport = ovs_vport_ovsl_rcu(dp, port_no);
2004 return ERR_PTR(-ENODEV);
2007 return ERR_PTR(-EINVAL);
2010 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2012 struct nlattr **a = info->attrs;
2013 struct ovs_header *ovs_header = info->userhdr;
2014 struct vport_parms parms;
2015 struct sk_buff *reply;
2016 struct vport *vport;
2017 struct datapath *dp;
2022 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2023 !a[OVS_VPORT_ATTR_UPCALL_PID])
2026 err = ovs_vport_cmd_validate(a);
2031 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2036 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2037 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2040 if (port_no >= DP_MAX_PORTS)
2043 vport = ovs_vport_ovsl(dp, port_no);
2048 for (port_no = 1; ; port_no++) {
2049 if (port_no >= DP_MAX_PORTS) {
2053 vport = ovs_vport_ovsl(dp, port_no);
2059 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2060 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2061 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2063 parms.port_no = port_no;
2064 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2066 vport = new_vport(&parms);
2067 err = PTR_ERR(vport);
2072 if (a[OVS_VPORT_ATTR_STATS])
2073 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2075 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2077 if (IS_ERR(reply)) {
2078 err = PTR_ERR(reply);
2079 ovs_dp_detach_port(vport);
2083 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2091 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2093 struct nlattr **a = info->attrs;
2094 struct sk_buff *reply;
2095 struct vport *vport;
2098 err = ovs_vport_cmd_validate(a);
2103 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2104 err = PTR_ERR(vport);
2108 if (a[OVS_VPORT_ATTR_TYPE] &&
2109 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2114 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2120 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2121 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2126 if (a[OVS_VPORT_ATTR_STATS])
2127 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2129 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2130 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2132 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2133 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2137 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2148 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2150 struct nlattr **a = info->attrs;
2151 struct sk_buff *reply;
2152 struct vport *vport;
2155 err = ovs_vport_cmd_validate(a);
2160 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2161 err = PTR_ERR(vport);
2165 if (vport->port_no == OVSP_LOCAL) {
2170 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2171 info->snd_seq, OVS_VPORT_CMD_DEL);
2172 err = PTR_ERR(reply);
2177 ovs_dp_detach_port(vport);
2179 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2187 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2189 struct nlattr **a = info->attrs;
2190 struct ovs_header *ovs_header = info->userhdr;
2191 struct sk_buff *reply;
2192 struct vport *vport;
2195 err = ovs_vport_cmd_validate(a);
2200 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2201 err = PTR_ERR(vport);
2205 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2206 info->snd_seq, OVS_VPORT_CMD_NEW);
2207 err = PTR_ERR(reply);
2213 return genlmsg_reply(reply, info);
2221 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2223 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2224 struct datapath *dp;
2225 int bucket = cb->args[0], skip = cb->args[1];
2228 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2233 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2234 struct vport *vport;
2237 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2239 ovs_vport_cmd_fill_info(vport, skb,
2240 NETLINK_CB(cb->skb).portid,
2243 OVS_VPORT_CMD_NEW) < 0)
2259 static struct genl_ops dp_vport_genl_ops[] = {
2260 { .cmd = OVS_VPORT_CMD_NEW,
2261 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2262 .policy = vport_policy,
2263 .doit = ovs_vport_cmd_new
2265 { .cmd = OVS_VPORT_CMD_DEL,
2266 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2267 .policy = vport_policy,
2268 .doit = ovs_vport_cmd_del
2270 { .cmd = OVS_VPORT_CMD_GET,
2271 .flags = 0, /* OK for unprivileged users. */
2272 .policy = vport_policy,
2273 .doit = ovs_vport_cmd_get,
2274 .dumpit = ovs_vport_cmd_dump
2276 { .cmd = OVS_VPORT_CMD_SET,
2277 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2278 .policy = vport_policy,
2279 .doit = ovs_vport_cmd_set,
2283 struct genl_family_and_ops {
2284 struct genl_family *family;
2285 struct genl_ops *ops;
2287 struct genl_multicast_group *group;
2290 static const struct genl_family_and_ops dp_genl_families[] = {
2291 { &dp_datapath_genl_family,
2292 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2293 &ovs_dp_datapath_multicast_group },
2294 { &dp_vport_genl_family,
2295 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2296 &ovs_dp_vport_multicast_group },
2297 { &dp_flow_genl_family,
2298 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2299 &ovs_dp_flow_multicast_group },
2300 { &dp_packet_genl_family,
2301 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2305 static void dp_unregister_genl(int n_families)
2309 for (i = 0; i < n_families; i++)
2310 genl_unregister_family(dp_genl_families[i].family);
2313 static int dp_register_genl(void)
2320 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2321 const struct genl_family_and_ops *f = &dp_genl_families[i];
2323 err = genl_register_family_with_ops(f->family, f->ops,
2330 err = genl_register_mc_group(f->family, f->group);
2339 dp_unregister_genl(n_registered);
2343 static void rehash_flow_table(struct work_struct *work)
2345 struct datapath *dp;
2351 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2353 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2354 struct flow_table *old_table = ovsl_dereference(dp->table);
2355 struct flow_table *new_table;
2357 new_table = ovs_flow_tbl_rehash(old_table);
2358 if (!IS_ERR(new_table)) {
2359 rcu_assign_pointer(dp->table, new_table);
2360 ovs_flow_tbl_deferred_destroy(old_table);
2366 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2369 static int __net_init ovs_init_net(struct net *net)
2371 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2373 INIT_LIST_HEAD(&ovs_net->dps);
2374 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2378 static void __net_exit ovs_exit_net(struct net *net)
2380 struct datapath *dp, *dp_next;
2381 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2384 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2388 cancel_work_sync(&ovs_net->dp_notify_work);
2391 static struct pernet_operations ovs_net_ops = {
2392 .init = ovs_init_net,
2393 .exit = ovs_exit_net,
2395 .size = sizeof(struct ovs_net),
2398 static int __init dp_init(void)
2402 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2404 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2407 err = ovs_workqueues_init();
2411 err = ovs_flow_init();
2415 err = ovs_vport_init();
2417 goto error_flow_exit;
2419 err = register_pernet_device(&ovs_net_ops);
2421 goto error_vport_exit;
2423 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2425 goto error_netns_exit;
2427 err = dp_register_genl();
2429 goto error_unreg_notifier;
2431 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2435 error_unreg_notifier:
2436 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2438 unregister_pernet_device(&ovs_net_ops);
2444 ovs_workqueues_exit();
2449 static void dp_cleanup(void)
2451 cancel_delayed_work_sync(&rehash_flow_wq);
2452 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2453 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2454 unregister_pernet_device(&ovs_net_ops);
2458 ovs_workqueues_exit();
2461 module_init(dp_init);
2462 module_exit(dp_cleanup);
2464 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2465 MODULE_LICENSE("GPL");
2466 MODULE_VERSION(VERSION);