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/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
62 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
63 static void rehash_flow_table(struct work_struct *work);
64 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
66 int ovs_net_id __read_mostly;
68 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
69 struct genl_multicast_group *grp)
71 genl_notify(skb, genl_info_net(info), info->snd_portid,
72 grp->id, info->nlhdr, GFP_KERNEL);
78 * All writes e.g. Writes to device state (add/remove datapath, port, set
79 * operations on vports, etc.), Writes to other state (flow table
80 * modifications, set miscellaneous datapath parameters, etc.) are protected
83 * Reads are protected by RCU.
85 * There are a few special cases (mostly stats) that have their own
86 * synchronization but they nest under all of above and don't interact with
89 * The RTNL lock nests inside ovs_mutex.
92 static DEFINE_MUTEX(ovs_mutex);
96 mutex_lock(&ovs_mutex);
101 mutex_unlock(&ovs_mutex);
104 #ifdef CONFIG_LOCKDEP
105 int lockdep_ovsl_is_held(void)
108 return lockdep_is_held(&ovs_mutex);
114 static struct vport *new_vport(const struct vport_parms *);
115 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
116 const struct dp_upcall_info *);
117 static int queue_userspace_packet(struct net *, int dp_ifindex,
119 const struct dp_upcall_info *);
121 /* Must be called with rcu_read_lock or ovs_mutex. */
122 static struct datapath *get_dp(struct net *net, int dp_ifindex)
124 struct datapath *dp = NULL;
125 struct net_device *dev;
128 dev = dev_get_by_index_rcu(net, dp_ifindex);
130 struct vport *vport = ovs_internal_dev_get_vport(dev);
139 /* Must be called with rcu_read_lock or ovs_mutex. */
140 const char *ovs_dp_name(const struct datapath *dp)
142 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
143 return vport->ops->get_name(vport);
146 static int get_dpifindex(struct datapath *dp)
153 local = ovs_vport_rcu(dp, OVSP_LOCAL);
155 ifindex = netdev_vport_priv(local)->dev->ifindex;
164 static void destroy_dp_rcu(struct rcu_head *rcu)
166 struct datapath *dp = container_of(rcu, struct datapath, rcu);
168 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table, false);
169 free_percpu(dp->stats_percpu);
170 release_net(ovs_dp_get_net(dp));
175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
178 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
181 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
184 struct hlist_head *head;
186 head = vport_hash_bucket(dp, port_no);
187 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
188 if (vport->port_no == port_no)
194 /* Called with ovs_mutex. */
195 static struct vport *new_vport(const struct vport_parms *parms)
199 vport = ovs_vport_add(parms);
200 if (!IS_ERR(vport)) {
201 struct datapath *dp = parms->dp;
202 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
204 hlist_add_head_rcu(&vport->dp_hash_node, head);
209 void ovs_dp_detach_port(struct vport *p)
213 /* First drop references to device. */
214 hlist_del_rcu(&p->dp_hash_node);
216 /* Then destroy it. */
220 /* Must be called with rcu_read_lock. */
221 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
223 struct datapath *dp = p->dp;
224 struct sw_flow *flow;
225 struct dp_stats_percpu *stats;
226 struct sw_flow_key key;
230 stats = this_cpu_ptr(dp->stats_percpu);
232 /* Extract flow from 'skb' into 'key'. */
233 error = ovs_flow_extract(skb, p->port_no, &key);
234 if (unlikely(error)) {
240 flow = ovs_flow_lookup(rcu_dereference(dp->table), &key);
241 if (unlikely(!flow)) {
242 struct dp_upcall_info upcall;
244 upcall.cmd = OVS_PACKET_CMD_MISS;
246 upcall.userdata = NULL;
247 upcall.portid = p->upcall_portid;
248 ovs_dp_upcall(dp, skb, &upcall);
250 stats_counter = &stats->n_missed;
254 OVS_CB(skb)->flow = flow;
255 OVS_CB(skb)->pkt_key = &key;
257 stats_counter = &stats->n_hit;
258 ovs_flow_used(OVS_CB(skb)->flow, skb);
259 ovs_execute_actions(dp, skb);
262 /* Update datapath statistics. */
263 u64_stats_update_begin(&stats->sync);
265 u64_stats_update_end(&stats->sync);
268 static struct genl_family dp_packet_genl_family = {
269 .id = GENL_ID_GENERATE,
270 .hdrsize = sizeof(struct ovs_header),
271 .name = OVS_PACKET_FAMILY,
272 .version = OVS_PACKET_VERSION,
273 .maxattr = OVS_PACKET_ATTR_MAX,
275 .parallel_ops = true,
278 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
279 const struct dp_upcall_info *upcall_info)
281 struct dp_stats_percpu *stats;
285 if (upcall_info->portid == 0) {
290 dp_ifindex = get_dpifindex(dp);
296 if (!skb_is_gso(skb))
297 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
299 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
306 stats = this_cpu_ptr(dp->stats_percpu);
308 u64_stats_update_begin(&stats->sync);
310 u64_stats_update_end(&stats->sync);
315 static int queue_gso_packets(struct net *net, int dp_ifindex,
317 const struct dp_upcall_info *upcall_info)
319 unsigned short gso_type = skb_shinfo(skb)->gso_type;
320 struct dp_upcall_info later_info;
321 struct sw_flow_key later_key;
322 struct sk_buff *segs, *nskb;
325 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
327 return PTR_ERR(segs);
329 /* Queue all of the segments. */
332 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
336 if (skb == segs && gso_type & SKB_GSO_UDP) {
337 /* The initial flow key extracted by ovs_flow_extract()
338 * in this case is for a first fragment, so we need to
339 * properly mark later fragments.
341 later_key = *upcall_info->key;
342 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
344 later_info = *upcall_info;
345 later_info.key = &later_key;
346 upcall_info = &later_info;
348 } while ((skb = skb->next));
350 /* Free all of the segments. */
358 } while ((skb = nskb));
362 static size_t key_attr_size(void)
364 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
365 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
366 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
367 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
368 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
369 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
370 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
371 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
372 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
373 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
374 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
375 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
376 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
377 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
378 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
379 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
380 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
381 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
382 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
385 static size_t upcall_msg_size(const struct sk_buff *skb,
386 const struct nlattr *userdata)
388 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
389 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
390 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
392 /* OVS_PACKET_ATTR_USERDATA */
394 size += NLA_ALIGN(userdata->nla_len);
399 static int queue_userspace_packet(struct net *net, int dp_ifindex,
401 const struct dp_upcall_info *upcall_info)
403 struct ovs_header *upcall;
404 struct sk_buff *nskb = NULL;
405 struct sk_buff *user_skb; /* to be queued to userspace */
409 if (vlan_tx_tag_present(skb)) {
410 nskb = skb_clone(skb, GFP_ATOMIC);
414 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
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);
722 static int validate_userspace(const struct nlattr *attr)
724 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
725 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
726 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
728 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
731 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
732 attr, userspace_policy);
736 if (!a[OVS_USERSPACE_ATTR_PID] ||
737 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
743 static int copy_action(const struct nlattr *from,
744 struct sw_flow_actions **sfa)
746 int totlen = NLA_ALIGN(from->nla_len);
749 to = reserve_sfa_size(sfa, from->nla_len);
753 memcpy(to, from, totlen);
757 static int validate_and_copy_actions(const struct nlattr *attr,
758 const struct sw_flow_key *key,
760 struct sw_flow_actions **sfa)
762 const struct nlattr *a;
765 if (depth >= SAMPLE_ACTION_DEPTH)
768 nla_for_each_nested(a, attr, rem) {
769 /* Expected argument lengths, (u32)-1 for variable length. */
770 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
771 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
772 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
773 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
774 [OVS_ACTION_ATTR_POP_VLAN] = 0,
775 [OVS_ACTION_ATTR_SET] = (u32)-1,
776 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
778 const struct ovs_action_push_vlan *vlan;
779 int type = nla_type(a);
782 if (type > OVS_ACTION_ATTR_MAX ||
783 (action_lens[type] != nla_len(a) &&
784 action_lens[type] != (u32)-1))
789 case OVS_ACTION_ATTR_UNSPEC:
792 case OVS_ACTION_ATTR_USERSPACE:
793 err = validate_userspace(a);
798 case OVS_ACTION_ATTR_OUTPUT:
799 if (nla_get_u32(a) >= DP_MAX_PORTS)
804 case OVS_ACTION_ATTR_POP_VLAN:
807 case OVS_ACTION_ATTR_PUSH_VLAN:
809 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
811 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
815 case OVS_ACTION_ATTR_SET:
816 err = validate_set(a, key, sfa, &skip_copy);
821 case OVS_ACTION_ATTR_SAMPLE:
822 err = validate_and_copy_sample(a, key, depth, sfa);
832 err = copy_action(a, sfa);
844 static void clear_stats(struct sw_flow *flow)
848 flow->packet_count = 0;
849 flow->byte_count = 0;
852 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
854 struct ovs_header *ovs_header = info->userhdr;
855 struct nlattr **a = info->attrs;
856 struct sw_flow_actions *acts;
857 struct sk_buff *packet;
858 struct sw_flow *flow;
865 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
866 !a[OVS_PACKET_ATTR_ACTIONS])
869 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
870 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
874 skb_reserve(packet, NET_IP_ALIGN);
876 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
878 skb_reset_mac_header(packet);
879 eth = eth_hdr(packet);
881 /* Normally, setting the skb 'protocol' field would be handled by a
882 * call to eth_type_trans(), but it assumes there's a sending
883 * device, which we may not have. */
884 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
885 packet->protocol = eth->h_proto;
887 packet->protocol = htons(ETH_P_802_2);
889 /* Build an sw_flow for sending this packet. */
890 flow = ovs_flow_alloc();
895 err = ovs_flow_extract(packet, -1, &flow->key);
899 err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
902 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
907 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
908 rcu_assign_pointer(flow->sf_acts, acts);
912 OVS_CB(packet)->flow = flow;
913 OVS_CB(packet)->pkt_key = &flow->key;
914 packet->priority = flow->key.phy.priority;
915 packet->mark = flow->key.phy.skb_mark;
918 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
924 err = ovs_execute_actions(dp, packet);
928 ovs_flow_free(flow, false);
934 ovs_flow_free(flow, false);
941 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
942 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
943 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
944 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
947 static struct genl_ops dp_packet_genl_ops[] = {
948 { .cmd = OVS_PACKET_CMD_EXECUTE,
949 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
950 .policy = packet_policy,
951 .doit = ovs_packet_cmd_execute
955 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
957 struct flow_table *table;
960 table = rcu_dereference_check(dp->table, lockdep_ovsl_is_held());
961 stats->n_flows = ovs_flow_tbl_count(table);
963 stats->n_hit = stats->n_missed = stats->n_lost = 0;
964 for_each_possible_cpu(i) {
965 const struct dp_stats_percpu *percpu_stats;
966 struct dp_stats_percpu local_stats;
969 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
972 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
973 local_stats = *percpu_stats;
974 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
976 stats->n_hit += local_stats.n_hit;
977 stats->n_missed += local_stats.n_missed;
978 stats->n_lost += local_stats.n_lost;
982 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
983 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
984 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
985 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
988 static struct genl_family dp_flow_genl_family = {
989 .id = GENL_ID_GENERATE,
990 .hdrsize = sizeof(struct ovs_header),
991 .name = OVS_FLOW_FAMILY,
992 .version = OVS_FLOW_VERSION,
993 .maxattr = OVS_FLOW_ATTR_MAX,
995 .parallel_ops = true,
998 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
999 .name = OVS_FLOW_MCGROUP
1002 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1003 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1005 const struct nlattr *a;
1006 struct nlattr *start;
1009 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1013 nla_for_each_nested(a, attr, rem) {
1014 int type = nla_type(a);
1015 struct nlattr *st_sample;
1018 case OVS_SAMPLE_ATTR_PROBABILITY:
1019 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1022 case OVS_SAMPLE_ATTR_ACTIONS:
1023 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1026 err = actions_to_attr(nla_data(a), nla_len(a), skb);
1029 nla_nest_end(skb, st_sample);
1034 nla_nest_end(skb, start);
1038 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1040 const struct nlattr *ovs_key = nla_data(a);
1041 int key_type = nla_type(ovs_key);
1042 struct nlattr *start;
1046 case OVS_KEY_ATTR_IPV4_TUNNEL:
1047 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1051 err = ovs_ipv4_tun_to_nlattr(skb, nla_data(ovs_key),
1055 nla_nest_end(skb, start);
1058 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1066 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1068 const struct nlattr *a;
1071 nla_for_each_attr(a, attr, len, rem) {
1072 int type = nla_type(a);
1075 case OVS_ACTION_ATTR_SET:
1076 err = set_action_to_attr(a, skb);
1081 case OVS_ACTION_ATTR_SAMPLE:
1082 err = sample_action_to_attr(a, skb);
1087 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1096 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1098 return NLMSG_ALIGN(sizeof(struct ovs_header))
1099 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1100 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1101 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1102 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1103 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1104 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1107 /* Called with ovs_mutex. */
1108 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1109 struct sk_buff *skb, u32 portid,
1110 u32 seq, u32 flags, u8 cmd)
1112 const int skb_orig_len = skb->len;
1113 struct nlattr *start;
1114 struct ovs_flow_stats stats;
1115 struct ovs_header *ovs_header;
1121 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1125 ovs_header->dp_ifindex = get_dpifindex(dp);
1127 /* Fill flow key. */
1128 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1130 goto nla_put_failure;
1132 err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1133 &flow->unmasked_key, skb);
1136 nla_nest_end(skb, nla);
1138 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1140 goto nla_put_failure;
1142 err = ovs_flow_to_nlattrs(&flow->key, &flow->mask->key, skb);
1146 nla_nest_end(skb, nla);
1148 spin_lock_bh(&flow->lock);
1150 stats.n_packets = flow->packet_count;
1151 stats.n_bytes = flow->byte_count;
1152 tcp_flags = flow->tcp_flags;
1153 spin_unlock_bh(&flow->lock);
1156 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1157 goto nla_put_failure;
1159 if (stats.n_packets &&
1160 nla_put(skb, OVS_FLOW_ATTR_STATS,
1161 sizeof(struct ovs_flow_stats), &stats))
1162 goto nla_put_failure;
1165 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1166 goto nla_put_failure;
1168 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1169 * this is the first flow to be dumped into 'skb'. This is unusual for
1170 * Netlink but individual action lists can be longer than
1171 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1172 * The userspace caller can always fetch the actions separately if it
1173 * really wants them. (Most userspace callers in fact don't care.)
1175 * This can only fail for dump operations because the skb is always
1176 * properly sized for single flows.
1178 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1180 const struct sw_flow_actions *sf_acts;
1182 sf_acts = rcu_dereference_check(flow->sf_acts,
1183 lockdep_ovsl_is_held());
1185 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1187 nla_nest_end(skb, start);
1192 nla_nest_cancel(skb, start);
1194 } else if (skb_orig_len)
1195 goto nla_put_failure;
1197 return genlmsg_end(skb, ovs_header);
1202 genlmsg_cancel(skb, ovs_header);
1206 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1208 const struct sw_flow_actions *sf_acts;
1210 sf_acts = ovsl_dereference(flow->sf_acts);
1212 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1215 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1216 struct datapath *dp,
1217 u32 portid, u32 seq, u8 cmd)
1219 struct sk_buff *skb;
1222 skb = ovs_flow_cmd_alloc_info(flow);
1224 return ERR_PTR(-ENOMEM);
1226 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1231 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1233 struct nlattr **a = info->attrs;
1234 struct ovs_header *ovs_header = info->userhdr;
1235 struct sw_flow_key key, masked_key;
1236 struct sw_flow *flow = NULL;
1237 struct sw_flow_mask mask;
1238 struct sk_buff *reply;
1239 struct datapath *dp;
1240 struct flow_table *table;
1241 struct sw_flow_actions *acts = NULL;
1242 struct sw_flow_match match;
1247 if (!a[OVS_FLOW_ATTR_KEY])
1250 ovs_match_init(&match, &key, &mask);
1251 error = ovs_match_from_nlattrs(&match,
1252 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1256 /* Validate actions. */
1257 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1258 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1259 error = PTR_ERR(acts);
1263 ovs_flow_key_mask(&masked_key, &key, &mask);
1264 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
1265 &masked_key, 0, &acts);
1267 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1270 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1276 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1279 goto err_unlock_ovs;
1281 table = ovsl_dereference(dp->table);
1283 /* Check if this is a duplicate flow */
1284 flow = ovs_flow_lookup(table, &key);
1286 struct sw_flow_mask *mask_p;
1287 /* Bail out if we're not allowed to create a new flow. */
1289 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1290 goto err_unlock_ovs;
1292 /* Expand table, if necessary, to make room. */
1293 if (ovs_flow_tbl_need_to_expand(table)) {
1294 struct flow_table *new_table;
1296 new_table = ovs_flow_tbl_expand(table);
1297 if (!IS_ERR(new_table)) {
1298 rcu_assign_pointer(dp->table, new_table);
1299 ovs_flow_tbl_destroy(table, true);
1300 table = ovsl_dereference(dp->table);
1304 /* Allocate flow. */
1305 flow = ovs_flow_alloc();
1307 error = PTR_ERR(flow);
1308 goto err_unlock_ovs;
1312 flow->key = masked_key;
1313 flow->unmasked_key = key;
1315 /* Make sure mask is unique in the system */
1316 mask_p = ovs_sw_flow_mask_find(table, &mask);
1318 /* Allocate a new mask if none exsits. */
1319 mask_p = ovs_sw_flow_mask_alloc();
1322 mask_p->key = mask.key;
1323 mask_p->range = mask.range;
1324 ovs_sw_flow_mask_insert(table, mask_p);
1327 ovs_sw_flow_mask_add_ref(mask_p);
1328 flow->mask = mask_p;
1329 rcu_assign_pointer(flow->sf_acts, acts);
1331 /* Put flow in bucket. */
1332 ovs_flow_insert(table, flow);
1334 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1335 info->snd_seq, OVS_FLOW_CMD_NEW);
1337 /* We found a matching flow. */
1338 struct sw_flow_actions *old_acts;
1340 /* Bail out if we're not allowed to modify an existing flow.
1341 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1342 * because Generic Netlink treats the latter as a dump
1343 * request. We also accept NLM_F_EXCL in case that bug ever
1347 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1348 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1349 goto err_unlock_ovs;
1351 /* The unmasked key has to be the same for flow updates. */
1353 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1354 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
1355 goto err_unlock_ovs;
1358 /* Update actions. */
1359 old_acts = ovsl_dereference(flow->sf_acts);
1360 rcu_assign_pointer(flow->sf_acts, acts);
1361 ovs_flow_deferred_free_acts(old_acts);
1363 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1364 info->snd_seq, OVS_FLOW_CMD_NEW);
1367 if (a[OVS_FLOW_ATTR_CLEAR]) {
1368 spin_lock_bh(&flow->lock);
1370 spin_unlock_bh(&flow->lock);
1376 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1378 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1379 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1383 ovs_flow_free(flow, false);
1392 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1394 struct nlattr **a = info->attrs;
1395 struct ovs_header *ovs_header = info->userhdr;
1396 struct sw_flow_key key;
1397 struct sk_buff *reply;
1398 struct sw_flow *flow;
1399 struct datapath *dp;
1400 struct flow_table *table;
1401 struct sw_flow_match match;
1404 if (!a[OVS_FLOW_ATTR_KEY]) {
1405 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1409 ovs_match_init(&match, &key, NULL);
1410 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1415 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1421 table = ovsl_dereference(dp->table);
1422 flow = ovs_flow_lookup_unmasked_key(table, &match);
1428 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1429 info->snd_seq, OVS_FLOW_CMD_NEW);
1430 if (IS_ERR(reply)) {
1431 err = PTR_ERR(reply);
1436 return genlmsg_reply(reply, info);
1442 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1444 struct nlattr **a = info->attrs;
1445 struct ovs_header *ovs_header = info->userhdr;
1446 struct sw_flow_key key;
1447 struct sk_buff *reply;
1448 struct sw_flow *flow;
1449 struct datapath *dp;
1450 struct flow_table *table;
1451 struct sw_flow_match match;
1455 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1461 if (!a[OVS_FLOW_ATTR_KEY]) {
1462 err = flush_flows(dp);
1466 ovs_match_init(&match, &key, NULL);
1467 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1471 table = ovsl_dereference(dp->table);
1472 flow = ovs_flow_lookup_unmasked_key(table, &match);
1478 reply = ovs_flow_cmd_alloc_info(flow);
1484 ovs_flow_remove(table, flow);
1486 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1487 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1490 ovs_flow_free(flow, true);
1493 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1500 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1502 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1503 struct datapath *dp;
1504 struct flow_table *table;
1507 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1513 table = rcu_dereference(dp->table);
1515 struct sw_flow *flow;
1518 bucket = cb->args[0];
1520 flow = ovs_flow_dump_next(table, &bucket, &obj);
1524 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1525 NETLINK_CB(cb->skb).portid,
1526 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1527 OVS_FLOW_CMD_NEW) < 0)
1530 cb->args[0] = bucket;
1537 static struct genl_ops dp_flow_genl_ops[] = {
1538 { .cmd = OVS_FLOW_CMD_NEW,
1539 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1540 .policy = flow_policy,
1541 .doit = ovs_flow_cmd_new_or_set
1543 { .cmd = OVS_FLOW_CMD_DEL,
1544 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1545 .policy = flow_policy,
1546 .doit = ovs_flow_cmd_del
1548 { .cmd = OVS_FLOW_CMD_GET,
1549 .flags = 0, /* OK for unprivileged users. */
1550 .policy = flow_policy,
1551 .doit = ovs_flow_cmd_get,
1552 .dumpit = ovs_flow_cmd_dump
1554 { .cmd = OVS_FLOW_CMD_SET,
1555 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1556 .policy = flow_policy,
1557 .doit = ovs_flow_cmd_new_or_set,
1561 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1562 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1563 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1566 static struct genl_family dp_datapath_genl_family = {
1567 .id = GENL_ID_GENERATE,
1568 .hdrsize = sizeof(struct ovs_header),
1569 .name = OVS_DATAPATH_FAMILY,
1570 .version = OVS_DATAPATH_VERSION,
1571 .maxattr = OVS_DP_ATTR_MAX,
1573 .parallel_ops = true,
1576 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1577 .name = OVS_DATAPATH_MCGROUP
1580 static size_t ovs_dp_cmd_msg_size(void)
1582 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1584 msgsize += nla_total_size(IFNAMSIZ);
1585 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1590 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1591 u32 portid, u32 seq, u32 flags, u8 cmd)
1593 struct ovs_header *ovs_header;
1594 struct ovs_dp_stats dp_stats;
1597 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1602 ovs_header->dp_ifindex = get_dpifindex(dp);
1605 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1608 goto nla_put_failure;
1610 get_dp_stats(dp, &dp_stats);
1611 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1612 goto nla_put_failure;
1614 return genlmsg_end(skb, ovs_header);
1617 genlmsg_cancel(skb, ovs_header);
1622 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1625 struct sk_buff *skb;
1628 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1630 return ERR_PTR(-ENOMEM);
1632 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1635 return ERR_PTR(retval);
1640 /* Called with ovs_mutex. */
1641 static struct datapath *lookup_datapath(struct net *net,
1642 struct ovs_header *ovs_header,
1643 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1645 struct datapath *dp;
1647 if (!a[OVS_DP_ATTR_NAME])
1648 dp = get_dp(net, ovs_header->dp_ifindex);
1650 struct vport *vport;
1653 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1654 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1657 return dp ? dp : ERR_PTR(-ENODEV);
1660 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1662 struct nlattr **a = info->attrs;
1663 struct vport_parms parms;
1664 struct sk_buff *reply;
1665 struct datapath *dp;
1666 struct vport *vport;
1667 struct ovs_net *ovs_net;
1671 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1677 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1679 goto err_unlock_ovs;
1681 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1683 /* Allocate table. */
1685 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1689 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1690 if (!dp->stats_percpu) {
1692 goto err_destroy_table;
1695 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1699 goto err_destroy_percpu;
1702 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1703 INIT_HLIST_HEAD(&dp->ports[i]);
1705 /* Set up our datapath device. */
1706 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1707 parms.type = OVS_VPORT_TYPE_INTERNAL;
1708 parms.options = NULL;
1710 parms.port_no = OVSP_LOCAL;
1711 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1713 vport = new_vport(&parms);
1714 if (IS_ERR(vport)) {
1715 err = PTR_ERR(vport);
1719 goto err_destroy_ports_array;
1722 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1723 info->snd_seq, OVS_DP_CMD_NEW);
1724 err = PTR_ERR(reply);
1726 goto err_destroy_local_port;
1728 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1729 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1733 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1736 err_destroy_local_port:
1737 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1738 err_destroy_ports_array:
1741 free_percpu(dp->stats_percpu);
1743 ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1745 release_net(ovs_dp_get_net(dp));
1753 /* Called with ovs_mutex. */
1754 static void __dp_destroy(struct datapath *dp)
1758 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1759 struct vport *vport;
1760 struct hlist_node *n;
1762 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1763 if (vport->port_no != OVSP_LOCAL)
1764 ovs_dp_detach_port(vport);
1767 list_del_rcu(&dp->list_node);
1769 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1770 * all port in datapath are destroyed first before freeing datapath.
1772 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1774 call_rcu(&dp->rcu, destroy_dp_rcu);
1777 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1779 struct sk_buff *reply;
1780 struct datapath *dp;
1784 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1789 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1790 info->snd_seq, OVS_DP_CMD_DEL);
1791 err = PTR_ERR(reply);
1798 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1806 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1808 struct sk_buff *reply;
1809 struct datapath *dp;
1813 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1818 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1819 info->snd_seq, OVS_DP_CMD_NEW);
1820 if (IS_ERR(reply)) {
1821 err = PTR_ERR(reply);
1822 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1823 ovs_dp_datapath_multicast_group.id, err);
1829 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1837 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1839 struct sk_buff *reply;
1840 struct datapath *dp;
1844 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1850 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1851 info->snd_seq, OVS_DP_CMD_NEW);
1852 if (IS_ERR(reply)) {
1853 err = PTR_ERR(reply);
1858 return genlmsg_reply(reply, info);
1865 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1867 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1868 struct datapath *dp;
1869 int skip = cb->args[0];
1873 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1875 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1876 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1877 OVS_DP_CMD_NEW) < 0)
1888 static struct genl_ops dp_datapath_genl_ops[] = {
1889 { .cmd = OVS_DP_CMD_NEW,
1890 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1891 .policy = datapath_policy,
1892 .doit = ovs_dp_cmd_new
1894 { .cmd = OVS_DP_CMD_DEL,
1895 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1896 .policy = datapath_policy,
1897 .doit = ovs_dp_cmd_del
1899 { .cmd = OVS_DP_CMD_GET,
1900 .flags = 0, /* OK for unprivileged users. */
1901 .policy = datapath_policy,
1902 .doit = ovs_dp_cmd_get,
1903 .dumpit = ovs_dp_cmd_dump
1905 { .cmd = OVS_DP_CMD_SET,
1906 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1907 .policy = datapath_policy,
1908 .doit = ovs_dp_cmd_set,
1912 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1913 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1914 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1915 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1916 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1917 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1918 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1921 static struct genl_family dp_vport_genl_family = {
1922 .id = GENL_ID_GENERATE,
1923 .hdrsize = sizeof(struct ovs_header),
1924 .name = OVS_VPORT_FAMILY,
1925 .version = OVS_VPORT_VERSION,
1926 .maxattr = OVS_VPORT_ATTR_MAX,
1928 .parallel_ops = true,
1931 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1932 .name = OVS_VPORT_MCGROUP
1935 /* Called with ovs_mutex or RCU read lock. */
1936 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1937 u32 portid, u32 seq, u32 flags, u8 cmd)
1939 struct ovs_header *ovs_header;
1940 struct ovs_vport_stats vport_stats;
1943 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1948 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1950 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1951 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1952 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1953 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1954 goto nla_put_failure;
1956 ovs_vport_get_stats(vport, &vport_stats);
1957 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1959 goto nla_put_failure;
1961 err = ovs_vport_get_options(vport, skb);
1962 if (err == -EMSGSIZE)
1965 return genlmsg_end(skb, ovs_header);
1970 genlmsg_cancel(skb, ovs_header);
1974 /* Called with ovs_mutex or RCU read lock. */
1975 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1978 struct sk_buff *skb;
1981 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1983 return ERR_PTR(-ENOMEM);
1985 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1991 /* Called with ovs_mutex or RCU read lock. */
1992 static struct vport *lookup_vport(struct net *net,
1993 struct ovs_header *ovs_header,
1994 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1996 struct datapath *dp;
1997 struct vport *vport;
1999 if (a[OVS_VPORT_ATTR_NAME]) {
2000 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2002 return ERR_PTR(-ENODEV);
2003 if (ovs_header->dp_ifindex &&
2004 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2005 return ERR_PTR(-ENODEV);
2007 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2008 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2010 if (port_no >= DP_MAX_PORTS)
2011 return ERR_PTR(-EFBIG);
2013 dp = get_dp(net, ovs_header->dp_ifindex);
2015 return ERR_PTR(-ENODEV);
2017 vport = ovs_vport_ovsl_rcu(dp, port_no);
2019 return ERR_PTR(-ENODEV);
2022 return ERR_PTR(-EINVAL);
2025 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2027 struct nlattr **a = info->attrs;
2028 struct ovs_header *ovs_header = info->userhdr;
2029 struct vport_parms parms;
2030 struct sk_buff *reply;
2031 struct vport *vport;
2032 struct datapath *dp;
2037 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2038 !a[OVS_VPORT_ATTR_UPCALL_PID])
2042 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2047 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2048 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2051 if (port_no >= DP_MAX_PORTS)
2054 vport = ovs_vport_ovsl(dp, port_no);
2059 for (port_no = 1; ; port_no++) {
2060 if (port_no >= DP_MAX_PORTS) {
2064 vport = ovs_vport_ovsl(dp, port_no);
2070 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2071 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2072 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2074 parms.port_no = port_no;
2075 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2077 vport = new_vport(&parms);
2078 err = PTR_ERR(vport);
2083 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2085 if (IS_ERR(reply)) {
2086 err = PTR_ERR(reply);
2087 ovs_dp_detach_port(vport);
2091 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2099 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2101 struct nlattr **a = info->attrs;
2102 struct sk_buff *reply;
2103 struct vport *vport;
2107 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2108 err = PTR_ERR(vport);
2112 if (a[OVS_VPORT_ATTR_TYPE] &&
2113 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2118 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2124 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2125 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2130 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2131 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2133 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2134 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2138 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;
2156 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2157 err = PTR_ERR(vport);
2161 if (vport->port_no == OVSP_LOCAL) {
2166 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2167 info->snd_seq, OVS_VPORT_CMD_DEL);
2168 err = PTR_ERR(reply);
2173 ovs_dp_detach_port(vport);
2175 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2182 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2184 struct nlattr **a = info->attrs;
2185 struct ovs_header *ovs_header = info->userhdr;
2186 struct sk_buff *reply;
2187 struct vport *vport;
2191 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2192 err = PTR_ERR(vport);
2196 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2197 info->snd_seq, OVS_VPORT_CMD_NEW);
2198 err = PTR_ERR(reply);
2204 return genlmsg_reply(reply, info);
2211 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2213 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2214 struct datapath *dp;
2215 int bucket = cb->args[0], skip = cb->args[1];
2218 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2223 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2224 struct vport *vport;
2227 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2229 ovs_vport_cmd_fill_info(vport, skb,
2230 NETLINK_CB(cb->skb).portid,
2233 OVS_VPORT_CMD_NEW) < 0)
2249 static struct genl_ops dp_vport_genl_ops[] = {
2250 { .cmd = OVS_VPORT_CMD_NEW,
2251 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2252 .policy = vport_policy,
2253 .doit = ovs_vport_cmd_new
2255 { .cmd = OVS_VPORT_CMD_DEL,
2256 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2257 .policy = vport_policy,
2258 .doit = ovs_vport_cmd_del
2260 { .cmd = OVS_VPORT_CMD_GET,
2261 .flags = 0, /* OK for unprivileged users. */
2262 .policy = vport_policy,
2263 .doit = ovs_vport_cmd_get,
2264 .dumpit = ovs_vport_cmd_dump
2266 { .cmd = OVS_VPORT_CMD_SET,
2267 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2268 .policy = vport_policy,
2269 .doit = ovs_vport_cmd_set,
2273 struct genl_family_and_ops {
2274 struct genl_family *family;
2275 struct genl_ops *ops;
2277 struct genl_multicast_group *group;
2280 static const struct genl_family_and_ops dp_genl_families[] = {
2281 { &dp_datapath_genl_family,
2282 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2283 &ovs_dp_datapath_multicast_group },
2284 { &dp_vport_genl_family,
2285 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2286 &ovs_dp_vport_multicast_group },
2287 { &dp_flow_genl_family,
2288 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2289 &ovs_dp_flow_multicast_group },
2290 { &dp_packet_genl_family,
2291 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2295 static void dp_unregister_genl(int n_families)
2299 for (i = 0; i < n_families; i++)
2300 genl_unregister_family(dp_genl_families[i].family);
2303 static int dp_register_genl(void)
2310 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2311 const struct genl_family_and_ops *f = &dp_genl_families[i];
2313 err = genl_register_family_with_ops(f->family, f->ops,
2320 err = genl_register_mc_group(f->family, f->group);
2329 dp_unregister_genl(n_registered);
2333 static void rehash_flow_table(struct work_struct *work)
2335 struct datapath *dp;
2341 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2343 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2344 struct flow_table *old_table = ovsl_dereference(dp->table);
2345 struct flow_table *new_table;
2347 new_table = ovs_flow_tbl_rehash(old_table);
2348 if (!IS_ERR(new_table)) {
2349 rcu_assign_pointer(dp->table, new_table);
2350 ovs_flow_tbl_destroy(old_table, true);
2356 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2359 static int __net_init ovs_init_net(struct net *net)
2361 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2363 INIT_LIST_HEAD(&ovs_net->dps);
2364 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2368 static void __net_exit ovs_exit_net(struct net *net)
2370 struct datapath *dp, *dp_next;
2371 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2374 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2378 cancel_work_sync(&ovs_net->dp_notify_work);
2381 static struct pernet_operations ovs_net_ops = {
2382 .init = ovs_init_net,
2383 .exit = ovs_exit_net,
2385 .size = sizeof(struct ovs_net),
2388 static int __init dp_init(void)
2392 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2394 pr_info("Open vSwitch switching datapath\n");
2396 err = ovs_flow_init();
2400 err = ovs_vport_init();
2402 goto error_flow_exit;
2404 err = register_pernet_device(&ovs_net_ops);
2406 goto error_vport_exit;
2408 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2410 goto error_netns_exit;
2412 err = dp_register_genl();
2414 goto error_unreg_notifier;
2416 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2420 error_unreg_notifier:
2421 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2423 unregister_pernet_device(&ovs_net_ops);
2432 static void dp_cleanup(void)
2434 cancel_delayed_work_sync(&rehash_flow_wq);
2435 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2436 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2437 unregister_pernet_device(&ovs_net_ops);
2443 module_init(dp_init);
2444 module_exit(dp_cleanup);
2446 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2447 MODULE_LICENSE("GPL");