2 * Copyright (c) 2007-2013 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/genetlink.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
60 #include "vport-internal_dev.h"
61 #include "vport-netdev.h"
63 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
65 int ovs_net_id __read_mostly;
67 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
68 struct genl_multicast_group *grp)
70 genl_notify(skb, genl_info_net(info), info->snd_portid,
71 grp->id, info->nlhdr, GFP_KERNEL);
77 * All writes e.g. Writes to device state (add/remove datapath, port, set
78 * operations on vports, etc.), Writes to other state (flow table
79 * modifications, set miscellaneous datapath parameters, etc.) are protected
82 * Reads are protected by RCU.
84 * There are a few special cases (mostly stats) that have their own
85 * synchronization but they nest under all of above and don't interact with
88 * The RTNL lock nests inside ovs_mutex.
91 static DEFINE_MUTEX(ovs_mutex);
95 mutex_lock(&ovs_mutex);
100 mutex_unlock(&ovs_mutex);
103 #ifdef CONFIG_LOCKDEP
104 int lockdep_ovsl_is_held(void)
107 return lockdep_is_held(&ovs_mutex);
113 static struct vport *new_vport(const struct vport_parms *);
114 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
115 const struct dp_upcall_info *);
116 static int queue_userspace_packet(struct net *, int dp_ifindex,
118 const struct dp_upcall_info *);
120 /* Must be called with rcu_read_lock or ovs_mutex. */
121 static struct datapath *get_dp(struct net *net, int dp_ifindex)
123 struct datapath *dp = NULL;
124 struct net_device *dev;
127 dev = dev_get_by_index_rcu(net, dp_ifindex);
129 struct vport *vport = ovs_internal_dev_get_vport(dev);
138 /* Must be called with rcu_read_lock or ovs_mutex. */
139 const char *ovs_dp_name(const struct datapath *dp)
141 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
142 return vport->ops->get_name(vport);
145 static int get_dpifindex(struct datapath *dp)
152 local = ovs_vport_rcu(dp, OVSP_LOCAL);
154 ifindex = netdev_vport_priv(local)->dev->ifindex;
163 static void destroy_dp_rcu(struct rcu_head *rcu)
165 struct datapath *dp = container_of(rcu, struct datapath, rcu);
167 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table, false);
168 free_percpu(dp->stats_percpu);
169 release_net(ovs_dp_get_net(dp));
174 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
177 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
180 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
183 struct hlist_head *head;
185 head = vport_hash_bucket(dp, port_no);
186 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
187 if (vport->port_no == port_no)
193 /* Called with ovs_mutex. */
194 static struct vport *new_vport(const struct vport_parms *parms)
198 vport = ovs_vport_add(parms);
199 if (!IS_ERR(vport)) {
200 struct datapath *dp = parms->dp;
201 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
203 hlist_add_head_rcu(&vport->dp_hash_node, head);
208 void ovs_dp_detach_port(struct vport *p)
212 /* First drop references to device. */
213 hlist_del_rcu(&p->dp_hash_node);
215 /* Then destroy it. */
219 /* Must be called with rcu_read_lock. */
220 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
222 struct datapath *dp = p->dp;
223 struct sw_flow *flow;
224 struct dp_stats_percpu *stats;
225 struct sw_flow_key key;
229 stats = this_cpu_ptr(dp->stats_percpu);
231 /* Extract flow from 'skb' into 'key'. */
232 error = ovs_flow_extract(skb, p->port_no, &key);
233 if (unlikely(error)) {
239 flow = ovs_flow_lookup(rcu_dereference(dp->table), &key);
240 if (unlikely(!flow)) {
241 struct dp_upcall_info upcall;
243 upcall.cmd = OVS_PACKET_CMD_MISS;
245 upcall.userdata = NULL;
246 upcall.portid = p->upcall_portid;
247 ovs_dp_upcall(dp, skb, &upcall);
249 stats_counter = &stats->n_missed;
253 OVS_CB(skb)->flow = flow;
254 OVS_CB(skb)->pkt_key = &key;
256 stats_counter = &stats->n_hit;
257 ovs_flow_used(OVS_CB(skb)->flow, skb);
258 ovs_execute_actions(dp, skb);
261 /* Update datapath statistics. */
262 u64_stats_update_begin(&stats->sync);
264 u64_stats_update_end(&stats->sync);
267 static struct genl_family dp_packet_genl_family = {
268 .id = GENL_ID_GENERATE,
269 .hdrsize = sizeof(struct ovs_header),
270 .name = OVS_PACKET_FAMILY,
271 .version = OVS_PACKET_VERSION,
272 .maxattr = OVS_PACKET_ATTR_MAX,
277 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
278 const struct dp_upcall_info *upcall_info)
280 struct dp_stats_percpu *stats;
284 if (upcall_info->portid == 0) {
289 dp_ifindex = get_dpifindex(dp);
295 if (!skb_is_gso(skb))
296 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
298 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
305 stats = this_cpu_ptr(dp->stats_percpu);
307 u64_stats_update_begin(&stats->sync);
309 u64_stats_update_end(&stats->sync);
314 static int queue_gso_packets(struct net *net, int dp_ifindex,
316 const struct dp_upcall_info *upcall_info)
318 unsigned short gso_type = skb_shinfo(skb)->gso_type;
319 struct dp_upcall_info later_info;
320 struct sw_flow_key later_key;
321 struct sk_buff *segs, *nskb;
324 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
326 return PTR_ERR(segs);
328 /* Queue all of the segments. */
331 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
335 if (skb == segs && gso_type & SKB_GSO_UDP) {
336 /* The initial flow key extracted by ovs_flow_extract()
337 * in this case is for a first fragment, so we need to
338 * properly mark later fragments.
340 later_key = *upcall_info->key;
341 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
343 later_info = *upcall_info;
344 later_info.key = &later_key;
345 upcall_info = &later_info;
347 } while ((skb = skb->next));
349 /* Free all of the segments. */
357 } while ((skb = nskb));
361 static size_t key_attr_size(void)
363 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
364 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
365 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
366 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
367 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
368 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
369 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
370 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
371 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
372 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
373 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
374 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
375 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
376 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
377 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
378 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
379 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
380 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
381 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
384 static size_t upcall_msg_size(const struct sk_buff *skb,
385 const struct nlattr *userdata)
387 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
388 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
389 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391 /* OVS_PACKET_ATTR_USERDATA */
393 size += NLA_ALIGN(userdata->nla_len);
398 static int queue_userspace_packet(struct net *net, int dp_ifindex,
400 const struct dp_upcall_info *upcall_info)
402 struct ovs_header *upcall;
403 struct sk_buff *nskb = NULL;
404 struct sk_buff *user_skb; /* to be queued to userspace */
408 if (vlan_tx_tag_present(skb)) {
409 nskb = skb_clone(skb, GFP_ATOMIC);
413 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
417 vlan_set_tci(nskb, 0);
422 if (nla_attr_size(skb->len) > USHRT_MAX) {
427 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
433 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
434 0, upcall_info->cmd);
435 upcall->dp_ifindex = dp_ifindex;
437 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
438 ovs_flow_to_nlattrs(upcall_info->key, upcall_info->key, user_skb);
439 nla_nest_end(user_skb, nla);
441 if (upcall_info->userdata)
442 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
443 nla_len(upcall_info->userdata),
444 nla_data(upcall_info->userdata));
446 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
448 skb_copy_and_csum_dev(skb, nla_data(nla));
450 genlmsg_end(user_skb, upcall);
451 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
458 /* Called with ovs_mutex. */
459 static int flush_flows(struct datapath *dp)
461 struct flow_table *old_table;
462 struct flow_table *new_table;
464 old_table = ovsl_dereference(dp->table);
465 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
469 rcu_assign_pointer(dp->table, new_table);
471 ovs_flow_tbl_destroy(old_table, true);
475 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
478 struct sw_flow_actions *acts;
480 int req_size = NLA_ALIGN(attr_len);
481 int next_offset = offsetof(struct sw_flow_actions, actions) +
484 if (req_size <= (ksize(*sfa) - next_offset))
487 new_acts_size = ksize(*sfa) * 2;
489 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
490 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
491 return ERR_PTR(-EMSGSIZE);
492 new_acts_size = MAX_ACTIONS_BUFSIZE;
495 acts = ovs_flow_actions_alloc(new_acts_size);
499 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
500 acts->actions_len = (*sfa)->actions_len;
505 (*sfa)->actions_len += req_size;
506 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
509 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
513 a = reserve_sfa_size(sfa, nla_attr_size(len));
517 a->nla_type = attrtype;
518 a->nla_len = nla_attr_size(len);
521 memcpy(nla_data(a), data, len);
522 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
527 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
529 int used = (*sfa)->actions_len;
532 err = add_action(sfa, attrtype, NULL, 0);
539 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
541 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
543 a->nla_len = sfa->actions_len - st_offset;
546 static int validate_and_copy_actions(const struct nlattr *attr,
547 const struct sw_flow_key *key, int depth,
548 struct sw_flow_actions **sfa);
550 static int validate_and_copy_sample(const struct nlattr *attr,
551 const struct sw_flow_key *key, int depth,
552 struct sw_flow_actions **sfa)
554 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
555 const struct nlattr *probability, *actions;
556 const struct nlattr *a;
557 int rem, start, err, st_acts;
559 memset(attrs, 0, sizeof(attrs));
560 nla_for_each_nested(a, attr, rem) {
561 int type = nla_type(a);
562 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
569 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
570 if (!probability || nla_len(probability) != sizeof(u32))
573 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
574 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
577 /* validation done, copy sample action. */
578 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
581 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
584 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
588 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
592 add_nested_action_end(*sfa, st_acts);
593 add_nested_action_end(*sfa, start);
598 static int validate_tp_port(const struct sw_flow_key *flow_key)
600 if (flow_key->eth.type == htons(ETH_P_IP)) {
601 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
603 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
604 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
611 static int validate_and_copy_set_tun(const struct nlattr *attr,
612 struct sw_flow_actions **sfa)
614 struct sw_flow_match match;
615 struct sw_flow_key key;
618 ovs_match_init(&match, &key, NULL);
619 err = ovs_ipv4_tun_from_nlattr(nla_data(attr), &match, false);
623 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
627 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
628 sizeof(match.key->tun_key));
629 add_nested_action_end(*sfa, start);
634 static int validate_set(const struct nlattr *a,
635 const struct sw_flow_key *flow_key,
636 struct sw_flow_actions **sfa,
639 const struct nlattr *ovs_key = nla_data(a);
640 int key_type = nla_type(ovs_key);
642 /* There can be only one key in a action */
643 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
646 if (key_type > OVS_KEY_ATTR_MAX ||
647 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
648 ovs_key_lens[key_type] != -1))
652 const struct ovs_key_ipv4 *ipv4_key;
653 const struct ovs_key_ipv6 *ipv6_key;
656 case OVS_KEY_ATTR_PRIORITY:
657 case OVS_KEY_ATTR_SKB_MARK:
658 case OVS_KEY_ATTR_ETHERNET:
661 case OVS_KEY_ATTR_TUNNEL:
663 err = validate_and_copy_set_tun(a, sfa);
668 case OVS_KEY_ATTR_IPV4:
669 if (flow_key->eth.type != htons(ETH_P_IP))
672 if (!flow_key->ip.proto)
675 ipv4_key = nla_data(ovs_key);
676 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
679 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
684 case OVS_KEY_ATTR_IPV6:
685 if (flow_key->eth.type != htons(ETH_P_IPV6))
688 if (!flow_key->ip.proto)
691 ipv6_key = nla_data(ovs_key);
692 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
695 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
698 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
703 case OVS_KEY_ATTR_TCP:
704 if (flow_key->ip.proto != IPPROTO_TCP)
707 return validate_tp_port(flow_key);
709 case OVS_KEY_ATTR_UDP:
710 if (flow_key->ip.proto != IPPROTO_UDP)
713 return validate_tp_port(flow_key);
715 case OVS_KEY_ATTR_SCTP:
716 if (flow_key->ip.proto != IPPROTO_SCTP)
719 return validate_tp_port(flow_key);
728 static int validate_userspace(const struct nlattr *attr)
730 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
731 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
732 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
734 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
737 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
738 attr, userspace_policy);
742 if (!a[OVS_USERSPACE_ATTR_PID] ||
743 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
749 static int copy_action(const struct nlattr *from,
750 struct sw_flow_actions **sfa)
752 int totlen = NLA_ALIGN(from->nla_len);
755 to = reserve_sfa_size(sfa, from->nla_len);
759 memcpy(to, from, totlen);
763 static int validate_and_copy_actions(const struct nlattr *attr,
764 const struct sw_flow_key *key,
766 struct sw_flow_actions **sfa)
768 const struct nlattr *a;
771 if (depth >= SAMPLE_ACTION_DEPTH)
774 nla_for_each_nested(a, attr, rem) {
775 /* Expected argument lengths, (u32)-1 for variable length. */
776 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
777 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
778 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
779 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
780 [OVS_ACTION_ATTR_POP_VLAN] = 0,
781 [OVS_ACTION_ATTR_SET] = (u32)-1,
782 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
784 const struct ovs_action_push_vlan *vlan;
785 int type = nla_type(a);
788 if (type > OVS_ACTION_ATTR_MAX ||
789 (action_lens[type] != nla_len(a) &&
790 action_lens[type] != (u32)-1))
795 case OVS_ACTION_ATTR_UNSPEC:
798 case OVS_ACTION_ATTR_USERSPACE:
799 err = validate_userspace(a);
804 case OVS_ACTION_ATTR_OUTPUT:
805 if (nla_get_u32(a) >= DP_MAX_PORTS)
810 case OVS_ACTION_ATTR_POP_VLAN:
813 case OVS_ACTION_ATTR_PUSH_VLAN:
815 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
817 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
821 case OVS_ACTION_ATTR_SET:
822 err = validate_set(a, key, sfa, &skip_copy);
827 case OVS_ACTION_ATTR_SAMPLE:
828 err = validate_and_copy_sample(a, key, depth, sfa);
838 err = copy_action(a, sfa);
850 static void clear_stats(struct sw_flow *flow)
854 flow->packet_count = 0;
855 flow->byte_count = 0;
858 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
860 struct ovs_header *ovs_header = info->userhdr;
861 struct nlattr **a = info->attrs;
862 struct sw_flow_actions *acts;
863 struct sk_buff *packet;
864 struct sw_flow *flow;
871 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
872 !a[OVS_PACKET_ATTR_ACTIONS])
875 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
876 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
880 skb_reserve(packet, NET_IP_ALIGN);
882 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
884 skb_reset_mac_header(packet);
885 eth = eth_hdr(packet);
887 /* Normally, setting the skb 'protocol' field would be handled by a
888 * call to eth_type_trans(), but it assumes there's a sending
889 * device, which we may not have. */
890 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
891 packet->protocol = eth->h_proto;
893 packet->protocol = htons(ETH_P_802_2);
895 /* Build an sw_flow for sending this packet. */
896 flow = ovs_flow_alloc();
901 err = ovs_flow_extract(packet, -1, &flow->key);
905 err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
908 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
913 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
914 rcu_assign_pointer(flow->sf_acts, acts);
918 OVS_CB(packet)->flow = flow;
919 OVS_CB(packet)->pkt_key = &flow->key;
920 packet->priority = flow->key.phy.priority;
921 packet->mark = flow->key.phy.skb_mark;
924 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
930 err = ovs_execute_actions(dp, packet);
934 ovs_flow_free(flow, false);
940 ovs_flow_free(flow, false);
947 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
948 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
949 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
950 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
953 static struct genl_ops dp_packet_genl_ops[] = {
954 { .cmd = OVS_PACKET_CMD_EXECUTE,
955 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
956 .policy = packet_policy,
957 .doit = ovs_packet_cmd_execute
961 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
963 struct flow_table *table;
966 table = rcu_dereference_check(dp->table, lockdep_ovsl_is_held());
967 stats->n_flows = ovs_flow_tbl_count(table);
969 stats->n_hit = stats->n_missed = stats->n_lost = 0;
970 for_each_possible_cpu(i) {
971 const struct dp_stats_percpu *percpu_stats;
972 struct dp_stats_percpu local_stats;
975 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
978 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
979 local_stats = *percpu_stats;
980 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
982 stats->n_hit += local_stats.n_hit;
983 stats->n_missed += local_stats.n_missed;
984 stats->n_lost += local_stats.n_lost;
988 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
989 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
990 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
991 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
994 static struct genl_family dp_flow_genl_family = {
995 .id = GENL_ID_GENERATE,
996 .hdrsize = sizeof(struct ovs_header),
997 .name = OVS_FLOW_FAMILY,
998 .version = OVS_FLOW_VERSION,
999 .maxattr = OVS_FLOW_ATTR_MAX,
1004 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1005 .name = OVS_FLOW_MCGROUP
1008 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1009 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1011 const struct nlattr *a;
1012 struct nlattr *start;
1015 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1019 nla_for_each_nested(a, attr, rem) {
1020 int type = nla_type(a);
1021 struct nlattr *st_sample;
1024 case OVS_SAMPLE_ATTR_PROBABILITY:
1025 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1028 case OVS_SAMPLE_ATTR_ACTIONS:
1029 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1032 err = actions_to_attr(nla_data(a), nla_len(a), skb);
1035 nla_nest_end(skb, st_sample);
1040 nla_nest_end(skb, start);
1044 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1046 const struct nlattr *ovs_key = nla_data(a);
1047 int key_type = nla_type(ovs_key);
1048 struct nlattr *start;
1052 case OVS_KEY_ATTR_IPV4_TUNNEL:
1053 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1057 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
1061 nla_nest_end(skb, start);
1064 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1072 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1074 const struct nlattr *a;
1077 nla_for_each_attr(a, attr, len, rem) {
1078 int type = nla_type(a);
1081 case OVS_ACTION_ATTR_SET:
1082 err = set_action_to_attr(a, skb);
1087 case OVS_ACTION_ATTR_SAMPLE:
1088 err = sample_action_to_attr(a, skb);
1093 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1102 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1104 return NLMSG_ALIGN(sizeof(struct ovs_header))
1105 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1106 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1107 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1108 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1109 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1110 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1113 /* Called with ovs_mutex. */
1114 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1115 struct sk_buff *skb, u32 portid,
1116 u32 seq, u32 flags, u8 cmd)
1118 const int skb_orig_len = skb->len;
1119 struct nlattr *start;
1120 struct ovs_flow_stats stats;
1121 struct ovs_header *ovs_header;
1127 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1131 ovs_header->dp_ifindex = get_dpifindex(dp);
1133 /* Fill flow key. */
1134 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1136 goto nla_put_failure;
1138 err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1139 &flow->unmasked_key, skb);
1142 nla_nest_end(skb, nla);
1144 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1146 goto nla_put_failure;
1148 err = ovs_flow_to_nlattrs(&flow->key, &flow->mask->key, skb);
1152 nla_nest_end(skb, nla);
1154 spin_lock_bh(&flow->lock);
1156 stats.n_packets = flow->packet_count;
1157 stats.n_bytes = flow->byte_count;
1158 tcp_flags = flow->tcp_flags;
1159 spin_unlock_bh(&flow->lock);
1162 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1163 goto nla_put_failure;
1165 if (stats.n_packets &&
1166 nla_put(skb, OVS_FLOW_ATTR_STATS,
1167 sizeof(struct ovs_flow_stats), &stats))
1168 goto nla_put_failure;
1171 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1172 goto nla_put_failure;
1174 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1175 * this is the first flow to be dumped into 'skb'. This is unusual for
1176 * Netlink but individual action lists can be longer than
1177 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1178 * The userspace caller can always fetch the actions separately if it
1179 * really wants them. (Most userspace callers in fact don't care.)
1181 * This can only fail for dump operations because the skb is always
1182 * properly sized for single flows.
1184 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1186 const struct sw_flow_actions *sf_acts;
1188 sf_acts = rcu_dereference_check(flow->sf_acts,
1189 lockdep_ovsl_is_held());
1191 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1193 nla_nest_end(skb, start);
1198 nla_nest_cancel(skb, start);
1200 } else if (skb_orig_len)
1201 goto nla_put_failure;
1203 return genlmsg_end(skb, ovs_header);
1208 genlmsg_cancel(skb, ovs_header);
1212 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1214 const struct sw_flow_actions *sf_acts;
1216 sf_acts = ovsl_dereference(flow->sf_acts);
1218 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1221 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1222 struct datapath *dp,
1223 u32 portid, u32 seq, u8 cmd)
1225 struct sk_buff *skb;
1228 skb = ovs_flow_cmd_alloc_info(flow);
1230 return ERR_PTR(-ENOMEM);
1232 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1237 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1239 struct nlattr **a = info->attrs;
1240 struct ovs_header *ovs_header = info->userhdr;
1241 struct sw_flow_key key, masked_key;
1242 struct sw_flow *flow = NULL;
1243 struct sw_flow_mask mask;
1244 struct sk_buff *reply;
1245 struct datapath *dp;
1246 struct flow_table *table;
1247 struct sw_flow_actions *acts = NULL;
1248 struct sw_flow_match match;
1253 if (!a[OVS_FLOW_ATTR_KEY])
1256 ovs_match_init(&match, &key, &mask);
1257 error = ovs_match_from_nlattrs(&match,
1258 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1262 /* Validate actions. */
1263 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1264 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1265 error = PTR_ERR(acts);
1269 ovs_flow_key_mask(&masked_key, &key, &mask);
1270 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1271 &masked_key, 0, &acts);
1273 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1276 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1282 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1285 goto err_unlock_ovs;
1287 table = ovsl_dereference(dp->table);
1289 /* Check if this is a duplicate flow */
1290 flow = ovs_flow_lookup(table, &key);
1292 struct flow_table *new_table = NULL;
1293 struct sw_flow_mask *mask_p;
1295 /* Bail out if we're not allowed to create a new flow. */
1297 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1298 goto err_unlock_ovs;
1300 /* Expand table, if necessary, to make room. */
1301 if (ovs_flow_tbl_need_to_expand(table))
1302 new_table = ovs_flow_tbl_expand(table);
1303 else if (time_after(jiffies, dp->last_rehash + REHASH_FLOW_INTERVAL))
1304 new_table = ovs_flow_tbl_rehash(table);
1306 if (new_table && !IS_ERR(new_table)) {
1307 rcu_assign_pointer(dp->table, new_table);
1308 ovs_flow_tbl_destroy(table, true);
1309 table = ovsl_dereference(dp->table);
1310 dp->last_rehash = jiffies;
1313 /* Allocate flow. */
1314 flow = ovs_flow_alloc();
1316 error = PTR_ERR(flow);
1317 goto err_unlock_ovs;
1321 flow->key = masked_key;
1322 flow->unmasked_key = key;
1324 /* Make sure mask is unique in the system */
1325 mask_p = ovs_sw_flow_mask_find(table, &mask);
1327 /* Allocate a new mask if none exsits. */
1328 mask_p = ovs_sw_flow_mask_alloc();
1331 mask_p->key = mask.key;
1332 mask_p->range = mask.range;
1333 ovs_sw_flow_mask_insert(table, mask_p);
1336 ovs_sw_flow_mask_add_ref(mask_p);
1337 flow->mask = mask_p;
1338 rcu_assign_pointer(flow->sf_acts, acts);
1340 /* Put flow in bucket. */
1341 ovs_flow_insert(table, flow);
1343 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1344 info->snd_seq, OVS_FLOW_CMD_NEW);
1346 /* We found a matching flow. */
1347 struct sw_flow_actions *old_acts;
1349 /* Bail out if we're not allowed to modify an existing flow.
1350 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1351 * because Generic Netlink treats the latter as a dump
1352 * request. We also accept NLM_F_EXCL in case that bug ever
1356 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1357 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1358 goto err_unlock_ovs;
1360 /* The unmasked key has to be the same for flow updates. */
1362 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1363 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1364 goto err_unlock_ovs;
1367 /* Update actions. */
1368 old_acts = ovsl_dereference(flow->sf_acts);
1369 rcu_assign_pointer(flow->sf_acts, acts);
1370 ovs_flow_deferred_free_acts(old_acts);
1372 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1373 info->snd_seq, OVS_FLOW_CMD_NEW);
1376 if (a[OVS_FLOW_ATTR_CLEAR]) {
1377 spin_lock_bh(&flow->lock);
1379 spin_unlock_bh(&flow->lock);
1385 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1387 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1388 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1392 ovs_flow_free(flow, false);
1401 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1403 struct nlattr **a = info->attrs;
1404 struct ovs_header *ovs_header = info->userhdr;
1405 struct sw_flow_key key;
1406 struct sk_buff *reply;
1407 struct sw_flow *flow;
1408 struct datapath *dp;
1409 struct flow_table *table;
1410 struct sw_flow_match match;
1413 if (!a[OVS_FLOW_ATTR_KEY]) {
1414 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1418 ovs_match_init(&match, &key, NULL);
1419 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1424 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1430 table = ovsl_dereference(dp->table);
1431 flow = ovs_flow_lookup_unmasked_key(table, &match);
1437 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1438 info->snd_seq, OVS_FLOW_CMD_NEW);
1439 if (IS_ERR(reply)) {
1440 err = PTR_ERR(reply);
1445 return genlmsg_reply(reply, info);
1451 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1453 struct nlattr **a = info->attrs;
1454 struct ovs_header *ovs_header = info->userhdr;
1455 struct sw_flow_key key;
1456 struct sk_buff *reply;
1457 struct sw_flow *flow;
1458 struct datapath *dp;
1459 struct flow_table *table;
1460 struct sw_flow_match match;
1464 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1470 if (!a[OVS_FLOW_ATTR_KEY]) {
1471 err = flush_flows(dp);
1475 ovs_match_init(&match, &key, NULL);
1476 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1480 table = ovsl_dereference(dp->table);
1481 flow = ovs_flow_lookup_unmasked_key(table, &match);
1487 reply = ovs_flow_cmd_alloc_info(flow);
1493 ovs_flow_remove(table, flow);
1495 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1496 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1499 ovs_flow_free(flow, true);
1502 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1509 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1511 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1512 struct datapath *dp;
1513 struct flow_table *table;
1516 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1522 table = rcu_dereference(dp->table);
1524 struct sw_flow *flow;
1527 bucket = cb->args[0];
1529 flow = ovs_flow_dump_next(table, &bucket, &obj);
1533 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1534 NETLINK_CB(cb->skb).portid,
1535 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1536 OVS_FLOW_CMD_NEW) < 0)
1539 cb->args[0] = bucket;
1546 static struct genl_ops dp_flow_genl_ops[] = {
1547 { .cmd = OVS_FLOW_CMD_NEW,
1548 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1549 .policy = flow_policy,
1550 .doit = ovs_flow_cmd_new_or_set
1552 { .cmd = OVS_FLOW_CMD_DEL,
1553 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1554 .policy = flow_policy,
1555 .doit = ovs_flow_cmd_del
1557 { .cmd = OVS_FLOW_CMD_GET,
1558 .flags = 0, /* OK for unprivileged users. */
1559 .policy = flow_policy,
1560 .doit = ovs_flow_cmd_get,
1561 .dumpit = ovs_flow_cmd_dump
1563 { .cmd = OVS_FLOW_CMD_SET,
1564 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1565 .policy = flow_policy,
1566 .doit = ovs_flow_cmd_new_or_set,
1570 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1571 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1572 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1575 static struct genl_family dp_datapath_genl_family = {
1576 .id = GENL_ID_GENERATE,
1577 .hdrsize = sizeof(struct ovs_header),
1578 .name = OVS_DATAPATH_FAMILY,
1579 .version = OVS_DATAPATH_VERSION,
1580 .maxattr = OVS_DP_ATTR_MAX,
1585 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1586 .name = OVS_DATAPATH_MCGROUP
1589 static size_t ovs_dp_cmd_msg_size(void)
1591 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1593 msgsize += nla_total_size(IFNAMSIZ);
1594 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1599 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1600 u32 portid, u32 seq, u32 flags, u8 cmd)
1602 struct ovs_header *ovs_header;
1603 struct ovs_dp_stats dp_stats;
1606 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1611 ovs_header->dp_ifindex = get_dpifindex(dp);
1614 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1617 goto nla_put_failure;
1619 get_dp_stats(dp, &dp_stats);
1620 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1621 goto nla_put_failure;
1623 return genlmsg_end(skb, ovs_header);
1626 genlmsg_cancel(skb, ovs_header);
1631 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1634 struct sk_buff *skb;
1637 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1639 return ERR_PTR(-ENOMEM);
1641 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1644 return ERR_PTR(retval);
1649 /* Called with ovs_mutex. */
1650 static struct datapath *lookup_datapath(struct net *net,
1651 struct ovs_header *ovs_header,
1652 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1654 struct datapath *dp;
1656 if (!a[OVS_DP_ATTR_NAME])
1657 dp = get_dp(net, ovs_header->dp_ifindex);
1659 struct vport *vport;
1662 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1663 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1666 return dp ? dp : ERR_PTR(-ENODEV);
1669 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1671 struct nlattr **a = info->attrs;
1672 struct vport_parms parms;
1673 struct sk_buff *reply;
1674 struct datapath *dp;
1675 struct vport *vport;
1676 struct ovs_net *ovs_net;
1680 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1686 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1688 goto err_unlock_ovs;
1690 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1692 /* Allocate table. */
1694 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1698 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1699 if (!dp->stats_percpu) {
1701 goto err_destroy_table;
1704 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1708 goto err_destroy_percpu;
1711 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1712 INIT_HLIST_HEAD(&dp->ports[i]);
1714 /* Set up our datapath device. */
1715 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1716 parms.type = OVS_VPORT_TYPE_INTERNAL;
1717 parms.options = NULL;
1719 parms.port_no = OVSP_LOCAL;
1720 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1722 vport = new_vport(&parms);
1723 if (IS_ERR(vport)) {
1724 err = PTR_ERR(vport);
1728 goto err_destroy_ports_array;
1731 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1732 info->snd_seq, OVS_DP_CMD_NEW);
1733 err = PTR_ERR(reply);
1735 goto err_destroy_local_port;
1737 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1738 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1742 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1745 err_destroy_local_port:
1746 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1747 err_destroy_ports_array:
1750 free_percpu(dp->stats_percpu);
1752 ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1754 release_net(ovs_dp_get_net(dp));
1762 /* Called with ovs_mutex. */
1763 static void __dp_destroy(struct datapath *dp)
1767 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1768 struct vport *vport;
1769 struct hlist_node *n;
1771 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1772 if (vport->port_no != OVSP_LOCAL)
1773 ovs_dp_detach_port(vport);
1776 list_del_rcu(&dp->list_node);
1778 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1779 * all port in datapath are destroyed first before freeing datapath.
1781 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1783 call_rcu(&dp->rcu, destroy_dp_rcu);
1786 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1788 struct sk_buff *reply;
1789 struct datapath *dp;
1793 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1798 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1799 info->snd_seq, OVS_DP_CMD_DEL);
1800 err = PTR_ERR(reply);
1807 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1815 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1817 struct sk_buff *reply;
1818 struct datapath *dp;
1822 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1827 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1828 info->snd_seq, OVS_DP_CMD_NEW);
1829 if (IS_ERR(reply)) {
1830 err = PTR_ERR(reply);
1831 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1832 ovs_dp_datapath_multicast_group.id, err);
1838 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1846 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1848 struct sk_buff *reply;
1849 struct datapath *dp;
1853 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1859 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1860 info->snd_seq, OVS_DP_CMD_NEW);
1861 if (IS_ERR(reply)) {
1862 err = PTR_ERR(reply);
1867 return genlmsg_reply(reply, info);
1874 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1876 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1877 struct datapath *dp;
1878 int skip = cb->args[0];
1882 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1884 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1885 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1886 OVS_DP_CMD_NEW) < 0)
1897 static struct genl_ops dp_datapath_genl_ops[] = {
1898 { .cmd = OVS_DP_CMD_NEW,
1899 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1900 .policy = datapath_policy,
1901 .doit = ovs_dp_cmd_new
1903 { .cmd = OVS_DP_CMD_DEL,
1904 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1905 .policy = datapath_policy,
1906 .doit = ovs_dp_cmd_del
1908 { .cmd = OVS_DP_CMD_GET,
1909 .flags = 0, /* OK for unprivileged users. */
1910 .policy = datapath_policy,
1911 .doit = ovs_dp_cmd_get,
1912 .dumpit = ovs_dp_cmd_dump
1914 { .cmd = OVS_DP_CMD_SET,
1915 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1916 .policy = datapath_policy,
1917 .doit = ovs_dp_cmd_set,
1921 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1922 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1923 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1924 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1925 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1926 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1927 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1930 static struct genl_family dp_vport_genl_family = {
1931 .id = GENL_ID_GENERATE,
1932 .hdrsize = sizeof(struct ovs_header),
1933 .name = OVS_VPORT_FAMILY,
1934 .version = OVS_VPORT_VERSION,
1935 .maxattr = OVS_VPORT_ATTR_MAX,
1940 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1941 .name = OVS_VPORT_MCGROUP
1944 /* Called with ovs_mutex or RCU read lock. */
1945 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1946 u32 portid, u32 seq, u32 flags, u8 cmd)
1948 struct ovs_header *ovs_header;
1949 struct ovs_vport_stats vport_stats;
1952 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1957 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1959 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1960 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1961 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1962 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1963 goto nla_put_failure;
1965 ovs_vport_get_stats(vport, &vport_stats);
1966 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1968 goto nla_put_failure;
1970 err = ovs_vport_get_options(vport, skb);
1971 if (err == -EMSGSIZE)
1974 return genlmsg_end(skb, ovs_header);
1979 genlmsg_cancel(skb, ovs_header);
1983 /* Called with ovs_mutex or RCU read lock. */
1984 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1987 struct sk_buff *skb;
1990 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1992 return ERR_PTR(-ENOMEM);
1994 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
2000 /* Called with ovs_mutex or RCU read lock. */
2001 static struct vport *lookup_vport(struct net *net,
2002 struct ovs_header *ovs_header,
2003 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2005 struct datapath *dp;
2006 struct vport *vport;
2008 if (a[OVS_VPORT_ATTR_NAME]) {
2009 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2011 return ERR_PTR(-ENODEV);
2012 if (ovs_header->dp_ifindex &&
2013 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2014 return ERR_PTR(-ENODEV);
2016 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2017 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2019 if (port_no >= DP_MAX_PORTS)
2020 return ERR_PTR(-EFBIG);
2022 dp = get_dp(net, ovs_header->dp_ifindex);
2024 return ERR_PTR(-ENODEV);
2026 vport = ovs_vport_ovsl_rcu(dp, port_no);
2028 return ERR_PTR(-ENODEV);
2031 return ERR_PTR(-EINVAL);
2034 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2036 struct nlattr **a = info->attrs;
2037 struct ovs_header *ovs_header = info->userhdr;
2038 struct vport_parms parms;
2039 struct sk_buff *reply;
2040 struct vport *vport;
2041 struct datapath *dp;
2046 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2047 !a[OVS_VPORT_ATTR_UPCALL_PID])
2051 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2056 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2057 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2060 if (port_no >= DP_MAX_PORTS)
2063 vport = ovs_vport_ovsl(dp, port_no);
2068 for (port_no = 1; ; port_no++) {
2069 if (port_no >= DP_MAX_PORTS) {
2073 vport = ovs_vport_ovsl(dp, port_no);
2079 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2080 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2081 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2083 parms.port_no = port_no;
2084 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2086 vport = new_vport(&parms);
2087 err = PTR_ERR(vport);
2092 if (a[OVS_VPORT_ATTR_STATS])
2093 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2095 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2097 if (IS_ERR(reply)) {
2098 err = PTR_ERR(reply);
2099 ovs_dp_detach_port(vport);
2103 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2111 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2113 struct nlattr **a = info->attrs;
2114 struct sk_buff *reply;
2115 struct vport *vport;
2119 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2120 err = PTR_ERR(vport);
2124 if (a[OVS_VPORT_ATTR_TYPE] &&
2125 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2130 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2136 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2137 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2142 if (a[OVS_VPORT_ATTR_STATS])
2143 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2145 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2146 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2148 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2149 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2153 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2163 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2165 struct nlattr **a = info->attrs;
2166 struct sk_buff *reply;
2167 struct vport *vport;
2171 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2172 err = PTR_ERR(vport);
2176 if (vport->port_no == OVSP_LOCAL) {
2181 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2182 info->snd_seq, OVS_VPORT_CMD_DEL);
2183 err = PTR_ERR(reply);
2188 ovs_dp_detach_port(vport);
2190 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2197 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2199 struct nlattr **a = info->attrs;
2200 struct ovs_header *ovs_header = info->userhdr;
2201 struct sk_buff *reply;
2202 struct vport *vport;
2206 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2207 err = PTR_ERR(vport);
2211 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2212 info->snd_seq, OVS_VPORT_CMD_NEW);
2213 err = PTR_ERR(reply);
2219 return genlmsg_reply(reply, info);
2226 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2228 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2229 struct datapath *dp;
2230 int bucket = cb->args[0], skip = cb->args[1];
2233 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2238 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2239 struct vport *vport;
2242 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2244 ovs_vport_cmd_fill_info(vport, skb,
2245 NETLINK_CB(cb->skb).portid,
2248 OVS_VPORT_CMD_NEW) < 0)
2264 static struct genl_ops dp_vport_genl_ops[] = {
2265 { .cmd = OVS_VPORT_CMD_NEW,
2266 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2267 .policy = vport_policy,
2268 .doit = ovs_vport_cmd_new
2270 { .cmd = OVS_VPORT_CMD_DEL,
2271 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2272 .policy = vport_policy,
2273 .doit = ovs_vport_cmd_del
2275 { .cmd = OVS_VPORT_CMD_GET,
2276 .flags = 0, /* OK for unprivileged users. */
2277 .policy = vport_policy,
2278 .doit = ovs_vport_cmd_get,
2279 .dumpit = ovs_vport_cmd_dump
2281 { .cmd = OVS_VPORT_CMD_SET,
2282 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2283 .policy = vport_policy,
2284 .doit = ovs_vport_cmd_set,
2288 struct genl_family_and_ops {
2289 struct genl_family *family;
2290 struct genl_ops *ops;
2292 struct genl_multicast_group *group;
2295 static const struct genl_family_and_ops dp_genl_families[] = {
2296 { &dp_datapath_genl_family,
2297 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2298 &ovs_dp_datapath_multicast_group },
2299 { &dp_vport_genl_family,
2300 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2301 &ovs_dp_vport_multicast_group },
2302 { &dp_flow_genl_family,
2303 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2304 &ovs_dp_flow_multicast_group },
2305 { &dp_packet_genl_family,
2306 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2310 static void dp_unregister_genl(int n_families)
2314 for (i = 0; i < n_families; i++)
2315 genl_unregister_family(dp_genl_families[i].family);
2318 static int dp_register_genl(void)
2325 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2326 const struct genl_family_and_ops *f = &dp_genl_families[i];
2328 err = genl_register_family_with_ops(f->family, f->ops,
2335 err = genl_register_mc_group(f->family, f->group);
2344 dp_unregister_genl(n_registered);
2348 static int __net_init ovs_init_net(struct net *net)
2350 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2352 INIT_LIST_HEAD(&ovs_net->dps);
2353 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2357 static void __net_exit ovs_exit_net(struct net *net)
2359 struct datapath *dp, *dp_next;
2360 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2363 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2367 cancel_work_sync(&ovs_net->dp_notify_work);
2370 static struct pernet_operations ovs_net_ops = {
2371 .init = ovs_init_net,
2372 .exit = ovs_exit_net,
2374 .size = sizeof(struct ovs_net),
2377 DEFINE_COMPAT_PNET_REG_FUNC(device);
2379 static int __init dp_init(void)
2383 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2385 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2388 err = ovs_flow_init();
2392 err = ovs_vport_init();
2394 goto error_flow_exit;
2396 err = register_pernet_device(&ovs_net_ops);
2398 goto error_vport_exit;
2400 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2402 goto error_netns_exit;
2404 err = dp_register_genl();
2406 goto error_unreg_notifier;
2410 error_unreg_notifier:
2411 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2413 unregister_pernet_device(&ovs_net_ops);
2422 static void dp_cleanup(void)
2424 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2425 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2426 unregister_pernet_device(&ovs_net_ops);
2432 module_init(dp_init);
2433 module_exit(dp_cleanup);
2435 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2436 MODULE_LICENSE("GPL");
2437 MODULE_VERSION(VERSION);