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>
62 #include "vport-internal_dev.h"
63 #include "vport-netdev.h"
65 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
66 LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0)
67 #error Kernels before 2.6.18 or after 3.8 are not supported by this version of Open vSwitch.
70 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
71 static void rehash_flow_table(struct work_struct *work);
72 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
74 int ovs_net_id __read_mostly;
76 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
77 struct genl_multicast_group *grp)
79 genl_notify(skb, genl_info_net(info), info->snd_portid,
80 grp->id, info->nlhdr, GFP_KERNEL);
86 * All writes e.g. Writes to device state (add/remove datapath, port, set
87 * operations on vports, etc.), Writes to other state (flow table
88 * modifications, set miscellaneous datapath parameters, etc.) are protected
91 * Reads are protected by RCU.
93 * There are a few special cases (mostly stats) that have their own
94 * synchronization but they nest under all of above and don't interact with
97 * The RTNL lock nests inside ovs_mutex.
100 static DEFINE_MUTEX(ovs_mutex);
104 mutex_lock(&ovs_mutex);
107 void ovs_unlock(void)
109 mutex_unlock(&ovs_mutex);
112 #ifdef CONFIG_LOCKDEP
113 int lockdep_ovsl_is_held(void)
116 return lockdep_is_held(&ovs_mutex);
122 static struct vport *new_vport(const struct vport_parms *);
123 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
124 const struct dp_upcall_info *);
125 static int queue_userspace_packet(struct net *, int dp_ifindex,
127 const struct dp_upcall_info *);
129 /* Must be called with rcu_read_lock or ovs_mutex. */
130 static struct datapath *get_dp(struct net *net, int dp_ifindex)
132 struct datapath *dp = NULL;
133 struct net_device *dev;
136 dev = dev_get_by_index_rcu(net, dp_ifindex);
138 struct vport *vport = ovs_internal_dev_get_vport(dev);
147 /* Must be called with rcu_read_lock or ovs_mutex. */
148 const char *ovs_dp_name(const struct datapath *dp)
150 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
151 return vport->ops->get_name(vport);
154 static int get_dpifindex(struct datapath *dp)
161 local = ovs_vport_rcu(dp, OVSP_LOCAL);
163 ifindex = netdev_vport_priv(local)->dev->ifindex;
172 static void destroy_dp_rcu(struct rcu_head *rcu)
174 struct datapath *dp = container_of(rcu, struct datapath, rcu);
176 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table, false);
177 free_percpu(dp->stats_percpu);
178 release_net(ovs_dp_get_net(dp));
183 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
186 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
192 struct hlist_head *head;
194 head = vport_hash_bucket(dp, port_no);
195 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
196 if (vport->port_no == port_no)
202 /* Called with ovs_mutex. */
203 static struct vport *new_vport(const struct vport_parms *parms)
207 vport = ovs_vport_add(parms);
208 if (!IS_ERR(vport)) {
209 struct datapath *dp = parms->dp;
210 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
212 hlist_add_head_rcu(&vport->dp_hash_node, head);
217 void ovs_dp_detach_port(struct vport *p)
221 /* First drop references to device. */
222 hlist_del_rcu(&p->dp_hash_node);
224 /* Then destroy it. */
228 /* Must be called with rcu_read_lock. */
229 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
231 struct datapath *dp = p->dp;
232 struct sw_flow *flow;
233 struct dp_stats_percpu *stats;
234 struct sw_flow_key key;
238 stats = this_cpu_ptr(dp->stats_percpu);
240 /* Extract flow from 'skb' into 'key'. */
241 error = ovs_flow_extract(skb, p->port_no, &key);
242 if (unlikely(error)) {
248 flow = ovs_flow_lookup(rcu_dereference(dp->table), &key);
249 if (unlikely(!flow)) {
250 struct dp_upcall_info upcall;
252 upcall.cmd = OVS_PACKET_CMD_MISS;
254 upcall.userdata = NULL;
255 upcall.portid = p->upcall_portid;
256 ovs_dp_upcall(dp, skb, &upcall);
258 stats_counter = &stats->n_missed;
262 OVS_CB(skb)->flow = flow;
264 stats_counter = &stats->n_hit;
265 ovs_flow_used(OVS_CB(skb)->flow, skb);
266 ovs_execute_actions(dp, skb);
269 /* Update datapath statistics. */
270 u64_stats_update_begin(&stats->sync);
272 u64_stats_update_end(&stats->sync);
275 static struct genl_family dp_packet_genl_family = {
276 .id = GENL_ID_GENERATE,
277 .hdrsize = sizeof(struct ovs_header),
278 .name = OVS_PACKET_FAMILY,
279 .version = OVS_PACKET_VERSION,
280 .maxattr = OVS_PACKET_ATTR_MAX,
284 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
285 const struct dp_upcall_info *upcall_info)
287 struct dp_stats_percpu *stats;
291 if (upcall_info->portid == 0) {
296 dp_ifindex = get_dpifindex(dp);
302 forward_ip_summed(skb, true);
304 if (!skb_is_gso(skb))
305 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
307 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
314 stats = this_cpu_ptr(dp->stats_percpu);
316 u64_stats_update_begin(&stats->sync);
318 u64_stats_update_end(&stats->sync);
323 static int queue_gso_packets(struct net *net, int dp_ifindex,
325 const struct dp_upcall_info *upcall_info)
327 unsigned short gso_type = skb_shinfo(skb)->gso_type;
328 struct dp_upcall_info later_info;
329 struct sw_flow_key later_key;
330 struct sk_buff *segs, *nskb;
333 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
335 return PTR_ERR(segs);
337 /* Queue all of the segments. */
340 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
344 if (skb == segs && gso_type & SKB_GSO_UDP) {
345 /* The initial flow key extracted by ovs_flow_extract()
346 * in this case is for a first fragment, so we need to
347 * properly mark later fragments.
349 later_key = *upcall_info->key;
350 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
352 later_info = *upcall_info;
353 later_info.key = &later_key;
354 upcall_info = &later_info;
356 } while ((skb = skb->next));
358 /* Free all of the segments. */
366 } while ((skb = nskb));
370 static size_t key_attr_size(void)
372 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
373 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
374 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
375 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
376 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
377 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
378 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
379 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
380 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
381 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
382 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
383 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
384 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
385 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
386 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
387 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
388 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
389 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
390 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
393 static size_t upcall_msg_size(const struct sk_buff *skb,
394 const struct nlattr *userdata)
396 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
397 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
398 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
400 /* OVS_PACKET_ATTR_USERDATA */
402 size += NLA_ALIGN(userdata->nla_len);
407 static int queue_userspace_packet(struct net *net, int dp_ifindex,
409 const struct dp_upcall_info *upcall_info)
411 struct ovs_header *upcall;
412 struct sk_buff *nskb = NULL;
413 struct sk_buff *user_skb; /* to be queued to userspace */
417 if (vlan_tx_tag_present(skb)) {
418 nskb = skb_clone(skb, GFP_ATOMIC);
422 err = vlan_deaccel_tag(nskb);
429 if (nla_attr_size(skb->len) > USHRT_MAX) {
434 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
440 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
441 0, upcall_info->cmd);
442 upcall->dp_ifindex = dp_ifindex;
444 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
445 ovs_flow_to_nlattrs(upcall_info->key, upcall_info->key, user_skb);
446 nla_nest_end(user_skb, nla);
448 if (upcall_info->userdata)
449 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
450 nla_len(upcall_info->userdata),
451 nla_data(upcall_info->userdata));
453 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
455 skb_copy_and_csum_dev(skb, nla_data(nla));
457 genlmsg_end(user_skb, upcall);
458 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
465 /* Called with ovs_mutex. */
466 static int flush_flows(struct datapath *dp)
468 struct flow_table *old_table;
469 struct flow_table *new_table;
471 old_table = ovsl_dereference(dp->table);
472 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
476 rcu_assign_pointer(dp->table, new_table);
478 ovs_flow_tbl_destroy(old_table, true);
482 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
485 struct sw_flow_actions *acts;
487 int req_size = NLA_ALIGN(attr_len);
488 int next_offset = offsetof(struct sw_flow_actions, actions) +
491 if (req_size <= (ksize(*sfa) - next_offset))
494 new_acts_size = ksize(*sfa) * 2;
496 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
497 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
498 return ERR_PTR(-EMSGSIZE);
499 new_acts_size = MAX_ACTIONS_BUFSIZE;
502 acts = ovs_flow_actions_alloc(new_acts_size);
506 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
507 acts->actions_len = (*sfa)->actions_len;
512 (*sfa)->actions_len += req_size;
513 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
516 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
520 a = reserve_sfa_size(sfa, nla_attr_size(len));
524 a->nla_type = attrtype;
525 a->nla_len = nla_attr_size(len);
528 memcpy(nla_data(a), data, len);
529 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
534 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
536 int used = (*sfa)->actions_len;
539 err = add_action(sfa, attrtype, NULL, 0);
546 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
548 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
550 a->nla_len = sfa->actions_len - st_offset;
553 static int validate_and_copy_actions(const struct nlattr *attr,
554 const struct sw_flow_key *key, int depth,
555 struct sw_flow_actions **sfa);
557 static int validate_and_copy_sample(const struct nlattr *attr,
558 const struct sw_flow_key *key, int depth,
559 struct sw_flow_actions **sfa)
561 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
562 const struct nlattr *probability, *actions;
563 const struct nlattr *a;
564 int rem, start, err, st_acts;
566 memset(attrs, 0, sizeof(attrs));
567 nla_for_each_nested(a, attr, rem) {
568 int type = nla_type(a);
569 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
576 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
577 if (!probability || nla_len(probability) != sizeof(u32))
580 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
581 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
584 /* validation done, copy sample action. */
585 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
588 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
591 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
595 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
599 add_nested_action_end(*sfa, st_acts);
600 add_nested_action_end(*sfa, start);
605 static int validate_tp_port(const struct sw_flow_key *flow_key)
607 if (flow_key->eth.type == htons(ETH_P_IP)) {
608 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
610 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
611 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
618 static int validate_and_copy_set_tun(const struct nlattr *attr,
619 struct sw_flow_actions **sfa)
621 struct sw_flow_match match;
622 struct sw_flow_key key;
625 ovs_match_init(&match, &key, NULL);
626 err = ipv4_tun_from_nlattr(nla_data(attr), &match, false);
630 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
634 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &match.key->tun_key,
635 sizeof(match.key->tun_key));
636 add_nested_action_end(*sfa, start);
641 static int validate_set(const struct nlattr *a,
642 const struct sw_flow_key *flow_key,
643 struct sw_flow_actions **sfa,
646 const struct nlattr *ovs_key = nla_data(a);
647 int key_type = nla_type(ovs_key);
649 /* There can be only one key in a action */
650 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
653 if (key_type > OVS_KEY_ATTR_MAX ||
654 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
655 ovs_key_lens[key_type] != -1))
659 const struct ovs_key_ipv4 *ipv4_key;
660 const struct ovs_key_ipv6 *ipv6_key;
663 case OVS_KEY_ATTR_PRIORITY:
664 case OVS_KEY_ATTR_ETHERNET:
667 case OVS_KEY_ATTR_SKB_MARK:
668 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
669 if (nla_get_u32(ovs_key) != 0)
674 case OVS_KEY_ATTR_TUNNEL:
676 err = validate_and_copy_set_tun(a, sfa);
681 case OVS_KEY_ATTR_IPV4:
682 if (flow_key->eth.type != htons(ETH_P_IP))
685 if (!flow_key->ip.proto)
688 ipv4_key = nla_data(ovs_key);
689 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
692 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
697 case OVS_KEY_ATTR_IPV6:
698 if (flow_key->eth.type != htons(ETH_P_IPV6))
701 if (!flow_key->ip.proto)
704 ipv6_key = nla_data(ovs_key);
705 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
708 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
711 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
716 case OVS_KEY_ATTR_TCP:
717 if (flow_key->ip.proto != IPPROTO_TCP)
720 return validate_tp_port(flow_key);
722 case OVS_KEY_ATTR_UDP:
723 if (flow_key->ip.proto != IPPROTO_UDP)
726 return validate_tp_port(flow_key);
735 static int validate_userspace(const struct nlattr *attr)
737 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
738 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
739 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
741 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
744 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
745 attr, userspace_policy);
749 if (!a[OVS_USERSPACE_ATTR_PID] ||
750 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
756 static int copy_action(const struct nlattr *from,
757 struct sw_flow_actions **sfa)
759 int totlen = NLA_ALIGN(from->nla_len);
762 to = reserve_sfa_size(sfa, from->nla_len);
766 memcpy(to, from, totlen);
770 static int validate_and_copy_actions(const struct nlattr *attr,
771 const struct sw_flow_key *key,
773 struct sw_flow_actions **sfa)
775 const struct nlattr *a;
778 if (depth >= SAMPLE_ACTION_DEPTH)
781 nla_for_each_nested(a, attr, rem) {
782 /* Expected argument lengths, (u32)-1 for variable length. */
783 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
784 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
785 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
786 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
787 [OVS_ACTION_ATTR_POP_VLAN] = 0,
788 [OVS_ACTION_ATTR_SET] = (u32)-1,
789 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
791 const struct ovs_action_push_vlan *vlan;
792 int type = nla_type(a);
795 if (type > OVS_ACTION_ATTR_MAX ||
796 (action_lens[type] != nla_len(a) &&
797 action_lens[type] != (u32)-1))
802 case OVS_ACTION_ATTR_UNSPEC:
805 case OVS_ACTION_ATTR_USERSPACE:
806 err = validate_userspace(a);
811 case OVS_ACTION_ATTR_OUTPUT:
812 if (nla_get_u32(a) >= DP_MAX_PORTS)
817 case OVS_ACTION_ATTR_POP_VLAN:
820 case OVS_ACTION_ATTR_PUSH_VLAN:
822 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
824 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
828 case OVS_ACTION_ATTR_SET:
829 err = validate_set(a, key, sfa, &skip_copy);
834 case OVS_ACTION_ATTR_SAMPLE:
835 err = validate_and_copy_sample(a, key, depth, sfa);
845 err = copy_action(a, sfa);
857 static void clear_stats(struct sw_flow *flow)
861 flow->packet_count = 0;
862 flow->byte_count = 0;
865 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
867 struct ovs_header *ovs_header = info->userhdr;
868 struct nlattr **a = info->attrs;
869 struct sw_flow_actions *acts;
870 struct sk_buff *packet;
871 struct sw_flow *flow;
878 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
879 !a[OVS_PACKET_ATTR_ACTIONS])
882 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
883 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
887 skb_reserve(packet, NET_IP_ALIGN);
889 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
891 skb_reset_mac_header(packet);
892 eth = eth_hdr(packet);
894 /* Normally, setting the skb 'protocol' field would be handled by a
895 * call to eth_type_trans(), but it assumes there's a sending
896 * device, which we may not have. */
897 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
898 packet->protocol = eth->h_proto;
900 packet->protocol = htons(ETH_P_802_2);
902 /* Build an sw_flow for sending this packet. */
903 flow = ovs_flow_alloc();
908 err = ovs_flow_extract(packet, -1, &flow->key);
912 err = ovs_flow_metadata_from_nlattrs(flow, a[OVS_PACKET_ATTR_KEY]);
915 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
920 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
921 rcu_assign_pointer(flow->sf_acts, acts);
925 OVS_CB(packet)->flow = flow;
926 packet->priority = flow->key.phy.priority;
927 skb_set_mark(packet, flow->key.phy.skb_mark);
930 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
936 err = ovs_execute_actions(dp, packet);
940 ovs_flow_free(flow, false);
946 ovs_flow_free(flow, false);
953 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
954 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
955 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
957 [OVS_PACKET_ATTR_PACKET] = { .minlen = ETH_HLEN },
959 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
960 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
963 static struct genl_ops dp_packet_genl_ops[] = {
964 { .cmd = OVS_PACKET_CMD_EXECUTE,
965 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
966 .policy = packet_policy,
967 .doit = ovs_packet_cmd_execute
971 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
974 struct flow_table *table = ovsl_dereference(dp->table);
976 stats->n_flows = ovs_flow_tbl_count(table);
978 stats->n_hit = stats->n_missed = stats->n_lost = 0;
979 for_each_possible_cpu(i) {
980 const struct dp_stats_percpu *percpu_stats;
981 struct dp_stats_percpu local_stats;
984 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
987 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
988 local_stats = *percpu_stats;
989 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
991 stats->n_hit += local_stats.n_hit;
992 stats->n_missed += local_stats.n_missed;
993 stats->n_lost += local_stats.n_lost;
997 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
998 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
999 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1000 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1003 static struct genl_family dp_flow_genl_family = {
1004 .id = GENL_ID_GENERATE,
1005 .hdrsize = sizeof(struct ovs_header),
1006 .name = OVS_FLOW_FAMILY,
1007 .version = OVS_FLOW_VERSION,
1008 .maxattr = OVS_FLOW_ATTR_MAX,
1012 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1013 .name = OVS_FLOW_MCGROUP
1016 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1017 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1019 const struct nlattr *a;
1020 struct nlattr *start;
1023 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1027 nla_for_each_nested(a, attr, rem) {
1028 int type = nla_type(a);
1029 struct nlattr *st_sample;
1032 case OVS_SAMPLE_ATTR_PROBABILITY:
1033 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1036 case OVS_SAMPLE_ATTR_ACTIONS:
1037 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1040 err = actions_to_attr(nla_data(a), nla_len(a), skb);
1043 nla_nest_end(skb, st_sample);
1048 nla_nest_end(skb, start);
1052 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1054 const struct nlattr *ovs_key = nla_data(a);
1055 int key_type = nla_type(ovs_key);
1056 struct nlattr *start;
1060 case OVS_KEY_ATTR_IPV4_TUNNEL:
1061 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1065 err = ipv4_tun_to_nlattr(skb,
1066 nla_data(ovs_key), nla_data(ovs_key));
1069 nla_nest_end(skb, start);
1072 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1080 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1082 const struct nlattr *a;
1085 nla_for_each_attr(a, attr, len, rem) {
1086 int type = nla_type(a);
1089 case OVS_ACTION_ATTR_SET:
1090 err = set_action_to_attr(a, skb);
1095 case OVS_ACTION_ATTR_SAMPLE:
1096 err = sample_action_to_attr(a, skb);
1101 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1110 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
1112 return NLMSG_ALIGN(sizeof(struct ovs_header))
1113 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
1114 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
1115 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
1116 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
1117 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
1118 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
1121 /* Called with ovs_mutex. */
1122 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1123 struct sk_buff *skb, u32 portid,
1124 u32 seq, u32 flags, u8 cmd)
1126 const int skb_orig_len = skb->len;
1127 const struct sw_flow_actions *sf_acts;
1128 struct nlattr *start;
1129 struct ovs_flow_stats stats;
1130 struct ovs_header *ovs_header;
1136 sf_acts = ovsl_dereference(flow->sf_acts);
1138 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1142 ovs_header->dp_ifindex = get_dpifindex(dp);
1144 /* Fill flow key. */
1145 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1147 goto nla_put_failure;
1149 err = ovs_flow_to_nlattrs(&flow->unmasked_key,
1150 &flow->unmasked_key, skb);
1153 nla_nest_end(skb, nla);
1155 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
1157 goto nla_put_failure;
1159 err = ovs_flow_to_nlattrs(&flow->key,
1160 &ovsl_dereference(flow->mask)->key, skb);
1164 nla_nest_end(skb, nla);
1166 spin_lock_bh(&flow->lock);
1168 stats.n_packets = flow->packet_count;
1169 stats.n_bytes = flow->byte_count;
1170 tcp_flags = flow->tcp_flags;
1171 spin_unlock_bh(&flow->lock);
1174 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1175 goto nla_put_failure;
1177 if (stats.n_packets &&
1178 nla_put(skb, OVS_FLOW_ATTR_STATS,
1179 sizeof(struct ovs_flow_stats), &stats))
1180 goto nla_put_failure;
1183 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1184 goto nla_put_failure;
1186 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1187 * this is the first flow to be dumped into 'skb'. This is unusual for
1188 * Netlink but individual action lists can be longer than
1189 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1190 * The userspace caller can always fetch the actions separately if it
1191 * really wants them. (Most userspace callers in fact don't care.)
1193 * This can only fail for dump operations because the skb is always
1194 * properly sized for single flows.
1196 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1198 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1200 nla_nest_end(skb, start);
1205 nla_nest_cancel(skb, start);
1207 } else if (skb_orig_len)
1208 goto nla_put_failure;
1210 return genlmsg_end(skb, ovs_header);
1215 genlmsg_cancel(skb, ovs_header);
1219 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1221 const struct sw_flow_actions *sf_acts;
1223 sf_acts = ovsl_dereference(flow->sf_acts);
1225 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
1228 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1229 struct datapath *dp,
1230 u32 portid, u32 seq, u8 cmd)
1232 struct sk_buff *skb;
1235 skb = ovs_flow_cmd_alloc_info(flow);
1237 return ERR_PTR(-ENOMEM);
1239 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1244 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1246 struct nlattr **a = info->attrs;
1247 struct ovs_header *ovs_header = info->userhdr;
1248 struct sw_flow_key key;
1249 struct sw_flow *flow = NULL;
1250 struct sw_flow_mask mask;
1251 struct sk_buff *reply;
1252 struct datapath *dp;
1253 struct flow_table *table;
1254 struct sw_flow_actions *acts = NULL;
1255 struct sw_flow_match match;
1260 if (!a[OVS_FLOW_ATTR_KEY])
1263 ovs_match_init(&match, &key, &mask);
1264 error = ovs_match_from_nlattrs(&match,
1265 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1269 /* Validate actions. */
1270 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1271 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1272 error = PTR_ERR(acts);
1276 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0, &acts);
1279 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1285 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1288 goto err_unlock_ovs;
1290 table = ovsl_dereference(dp->table);
1292 /* Check if this is a duplicate flow */
1293 flow = ovs_flow_lookup(table, &key);
1295 struct sw_flow_mask *mask_p;
1296 /* Bail out if we're not allowed to create a new flow. */
1298 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1299 goto err_unlock_ovs;
1301 /* Expand table, if necessary, to make room. */
1302 if (ovs_flow_tbl_need_to_expand(table)) {
1303 struct flow_table *new_table;
1305 new_table = ovs_flow_tbl_expand(table);
1306 if (!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);
1313 /* Allocate flow. */
1314 flow = ovs_flow_alloc();
1316 error = PTR_ERR(flow);
1317 goto err_unlock_ovs;
1321 /* Make sure mask is unique in the system */
1322 mask_p = ovs_sw_flow_mask_find(table, &mask);
1324 /* Allocate a new mask if none exsits. */
1325 mask_p = ovs_sw_flow_mask_alloc();
1328 mask_p->key = mask.key;
1329 mask_p->range = mask.range;
1330 ovs_sw_flow_mask_insert(table, mask_p);
1333 ovs_sw_flow_mask_add_ref(mask_p);
1334 rcu_assign_pointer(flow->mask, mask_p);
1335 rcu_assign_pointer(flow->sf_acts, acts);
1337 /* Put flow in bucket. */
1338 ovs_flow_insert(table, flow, &key, match.range.end);
1340 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1341 info->snd_seq, OVS_FLOW_CMD_NEW);
1343 /* We found a matching flow. */
1344 struct sw_flow_actions *old_acts;
1346 /* Make sure the it has the same unmasked key. */
1347 if (!ovs_flow_cmp_unmasked_key(flow, &key, match.range.end)) {
1349 goto err_unlock_ovs;
1352 /* Bail out if we're not allowed to modify an existing flow.
1353 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1354 * because Generic Netlink treats the latter as a dump
1355 * request. We also accept NLM_F_EXCL in case that bug ever
1359 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1360 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1361 goto err_unlock_ovs;
1363 /* Update actions. */
1364 old_acts = ovsl_dereference(flow->sf_acts);
1365 rcu_assign_pointer(flow->sf_acts, acts);
1366 ovs_flow_deferred_free_acts(old_acts);
1368 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1369 info->snd_seq, OVS_FLOW_CMD_NEW);
1372 if (a[OVS_FLOW_ATTR_CLEAR]) {
1373 spin_lock_bh(&flow->lock);
1375 spin_unlock_bh(&flow->lock);
1381 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1383 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1384 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1388 ovs_flow_free(flow, false);
1397 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1399 struct nlattr **a = info->attrs;
1400 struct ovs_header *ovs_header = info->userhdr;
1401 struct sw_flow_key key;
1402 struct sk_buff *reply;
1403 struct sw_flow *flow;
1404 struct datapath *dp;
1405 struct flow_table *table;
1406 struct sw_flow_match match;
1409 if (!a[OVS_FLOW_ATTR_KEY])
1412 ovs_match_init(&match, &key, NULL);
1413 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1418 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1424 table = ovsl_dereference(dp->table);
1425 flow = ovs_flow_lookup_unmasked_key(table, &match);
1431 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1432 info->snd_seq, OVS_FLOW_CMD_NEW);
1433 if (IS_ERR(reply)) {
1434 err = PTR_ERR(reply);
1439 return genlmsg_reply(reply, info);
1445 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1447 struct nlattr **a = info->attrs;
1448 struct ovs_header *ovs_header = info->userhdr;
1449 struct sw_flow_key key;
1450 struct sk_buff *reply;
1451 struct sw_flow *flow;
1452 struct datapath *dp;
1453 struct flow_table *table;
1454 struct sw_flow_match match;
1458 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1464 if (!a[OVS_FLOW_ATTR_KEY]) {
1465 err = flush_flows(dp);
1469 ovs_match_init(&match, &key, NULL);
1470 err = ovs_match_from_nlattrs(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1474 table = ovsl_dereference(dp->table);
1475 flow = ovs_flow_lookup_unmasked_key(table, &match);
1481 reply = ovs_flow_cmd_alloc_info(flow);
1487 ovs_flow_remove(table, flow);
1489 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1490 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1493 ovs_flow_free(flow, true);
1496 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1503 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1505 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1506 struct datapath *dp;
1507 struct flow_table *table;
1510 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1516 table = ovsl_dereference(dp->table);
1519 struct sw_flow *flow;
1522 bucket = cb->args[0];
1524 flow = ovs_flow_dump_next(table, &bucket, &obj);
1528 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1529 NETLINK_CB(cb->skb).portid,
1530 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1531 OVS_FLOW_CMD_NEW) < 0)
1534 cb->args[0] = bucket;
1541 static struct genl_ops dp_flow_genl_ops[] = {
1542 { .cmd = OVS_FLOW_CMD_NEW,
1543 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1544 .policy = flow_policy,
1545 .doit = ovs_flow_cmd_new_or_set
1547 { .cmd = OVS_FLOW_CMD_DEL,
1548 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1549 .policy = flow_policy,
1550 .doit = ovs_flow_cmd_del
1552 { .cmd = OVS_FLOW_CMD_GET,
1553 .flags = 0, /* OK for unprivileged users. */
1554 .policy = flow_policy,
1555 .doit = ovs_flow_cmd_get,
1556 .dumpit = ovs_flow_cmd_dump
1558 { .cmd = OVS_FLOW_CMD_SET,
1559 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1560 .policy = flow_policy,
1561 .doit = ovs_flow_cmd_new_or_set,
1565 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1566 #ifdef HAVE_NLA_NUL_STRING
1567 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1569 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1572 static struct genl_family dp_datapath_genl_family = {
1573 .id = GENL_ID_GENERATE,
1574 .hdrsize = sizeof(struct ovs_header),
1575 .name = OVS_DATAPATH_FAMILY,
1576 .version = OVS_DATAPATH_VERSION,
1577 .maxattr = OVS_DP_ATTR_MAX,
1581 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1582 .name = OVS_DATAPATH_MCGROUP
1585 static size_t ovs_dp_cmd_msg_size(void)
1587 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1589 msgsize += nla_total_size(IFNAMSIZ);
1590 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1595 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1596 u32 portid, u32 seq, u32 flags, u8 cmd)
1598 struct ovs_header *ovs_header;
1599 struct ovs_dp_stats dp_stats;
1602 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1607 ovs_header->dp_ifindex = get_dpifindex(dp);
1610 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1613 goto nla_put_failure;
1615 get_dp_stats(dp, &dp_stats);
1616 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1617 goto nla_put_failure;
1619 return genlmsg_end(skb, ovs_header);
1622 genlmsg_cancel(skb, ovs_header);
1627 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1630 struct sk_buff *skb;
1633 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1635 return ERR_PTR(-ENOMEM);
1637 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1640 return ERR_PTR(retval);
1645 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1647 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1650 /* Called with ovs_mutex. */
1651 static struct datapath *lookup_datapath(struct net *net,
1652 struct ovs_header *ovs_header,
1653 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1655 struct datapath *dp;
1657 if (!a[OVS_DP_ATTR_NAME])
1658 dp = get_dp(net, ovs_header->dp_ifindex);
1660 struct vport *vport;
1663 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1664 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1667 return dp ? dp : ERR_PTR(-ENODEV);
1670 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1672 struct nlattr **a = info->attrs;
1673 struct vport_parms parms;
1674 struct sk_buff *reply;
1675 struct datapath *dp;
1676 struct vport *vport;
1677 struct ovs_net *ovs_net;
1681 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1684 err = ovs_dp_cmd_validate(a);
1691 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1693 goto err_unlock_ovs;
1695 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1697 /* Allocate table. */
1699 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1703 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1704 if (!dp->stats_percpu) {
1706 goto err_destroy_table;
1709 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1713 goto err_destroy_percpu;
1716 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1717 INIT_HLIST_HEAD(&dp->ports[i]);
1719 /* Set up our datapath device. */
1720 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1721 parms.type = OVS_VPORT_TYPE_INTERNAL;
1722 parms.options = NULL;
1724 parms.port_no = OVSP_LOCAL;
1725 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1727 vport = new_vport(&parms);
1728 if (IS_ERR(vport)) {
1729 err = PTR_ERR(vport);
1733 goto err_destroy_ports_array;
1736 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1737 info->snd_seq, OVS_DP_CMD_NEW);
1738 err = PTR_ERR(reply);
1740 goto err_destroy_local_port;
1742 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1743 list_add_tail(&dp->list_node, &ovs_net->dps);
1747 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1750 err_destroy_local_port:
1751 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1752 err_destroy_ports_array:
1755 free_percpu(dp->stats_percpu);
1757 ovs_flow_tbl_destroy(ovsl_dereference(dp->table), false);
1759 release_net(ovs_dp_get_net(dp));
1767 /* Called with ovs_mutex. */
1768 static void __dp_destroy(struct datapath *dp)
1772 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1773 struct vport *vport;
1774 struct hlist_node *n;
1776 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1777 if (vport->port_no != OVSP_LOCAL)
1778 ovs_dp_detach_port(vport);
1781 list_del(&dp->list_node);
1783 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1784 * all port in datapath are destroyed first before freeing datapath.
1786 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1788 call_rcu(&dp->rcu, destroy_dp_rcu);
1791 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1793 struct sk_buff *reply;
1794 struct datapath *dp;
1797 err = ovs_dp_cmd_validate(info->attrs);
1802 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1807 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1808 info->snd_seq, OVS_DP_CMD_DEL);
1809 err = PTR_ERR(reply);
1816 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1824 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1826 struct sk_buff *reply;
1827 struct datapath *dp;
1830 err = ovs_dp_cmd_validate(info->attrs);
1835 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1840 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1841 info->snd_seq, OVS_DP_CMD_NEW);
1842 if (IS_ERR(reply)) {
1843 err = PTR_ERR(reply);
1844 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1845 ovs_dp_datapath_multicast_group.id, err);
1851 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1859 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1861 struct sk_buff *reply;
1862 struct datapath *dp;
1865 err = ovs_dp_cmd_validate(info->attrs);
1870 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1876 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1877 info->snd_seq, OVS_DP_CMD_NEW);
1878 if (IS_ERR(reply)) {
1879 err = PTR_ERR(reply);
1884 return genlmsg_reply(reply, info);
1891 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1893 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1894 struct datapath *dp;
1895 int skip = cb->args[0];
1899 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1901 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1902 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1903 OVS_DP_CMD_NEW) < 0)
1914 static struct genl_ops dp_datapath_genl_ops[] = {
1915 { .cmd = OVS_DP_CMD_NEW,
1916 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1917 .policy = datapath_policy,
1918 .doit = ovs_dp_cmd_new
1920 { .cmd = OVS_DP_CMD_DEL,
1921 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1922 .policy = datapath_policy,
1923 .doit = ovs_dp_cmd_del
1925 { .cmd = OVS_DP_CMD_GET,
1926 .flags = 0, /* OK for unprivileged users. */
1927 .policy = datapath_policy,
1928 .doit = ovs_dp_cmd_get,
1929 .dumpit = ovs_dp_cmd_dump
1931 { .cmd = OVS_DP_CMD_SET,
1932 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1933 .policy = datapath_policy,
1934 .doit = ovs_dp_cmd_set,
1938 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1939 #ifdef HAVE_NLA_NUL_STRING
1940 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1941 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1943 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1945 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1946 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1947 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1948 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1951 static struct genl_family dp_vport_genl_family = {
1952 .id = GENL_ID_GENERATE,
1953 .hdrsize = sizeof(struct ovs_header),
1954 .name = OVS_VPORT_FAMILY,
1955 .version = OVS_VPORT_VERSION,
1956 .maxattr = OVS_VPORT_ATTR_MAX,
1960 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1961 .name = OVS_VPORT_MCGROUP
1964 /* Called with ovs_mutex or RCU read lock. */
1965 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1966 u32 portid, u32 seq, u32 flags, u8 cmd)
1968 struct ovs_header *ovs_header;
1969 struct ovs_vport_stats vport_stats;
1972 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1977 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1979 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1980 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1981 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1982 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1983 goto nla_put_failure;
1985 ovs_vport_get_stats(vport, &vport_stats);
1986 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1988 goto nla_put_failure;
1990 err = ovs_vport_get_options(vport, skb);
1991 if (err == -EMSGSIZE)
1994 return genlmsg_end(skb, ovs_header);
1999 genlmsg_cancel(skb, ovs_header);
2003 /* Called with ovs_mutex or RCU read lock. */
2004 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
2007 struct sk_buff *skb;
2010 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
2012 return ERR_PTR(-ENOMEM);
2014 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
2020 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2022 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
2025 /* Called with ovs_mutex or RCU read lock. */
2026 static struct vport *lookup_vport(struct net *net,
2027 struct ovs_header *ovs_header,
2028 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2030 struct datapath *dp;
2031 struct vport *vport;
2033 if (a[OVS_VPORT_ATTR_NAME]) {
2034 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2036 return ERR_PTR(-ENODEV);
2037 if (ovs_header->dp_ifindex &&
2038 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2039 return ERR_PTR(-ENODEV);
2041 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2042 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2044 if (port_no >= DP_MAX_PORTS)
2045 return ERR_PTR(-EFBIG);
2047 dp = get_dp(net, ovs_header->dp_ifindex);
2049 return ERR_PTR(-ENODEV);
2051 vport = ovs_vport_ovsl_rcu(dp, port_no);
2053 return ERR_PTR(-ENODEV);
2056 return ERR_PTR(-EINVAL);
2059 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2061 struct nlattr **a = info->attrs;
2062 struct ovs_header *ovs_header = info->userhdr;
2063 struct vport_parms parms;
2064 struct sk_buff *reply;
2065 struct vport *vport;
2066 struct datapath *dp;
2071 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2072 !a[OVS_VPORT_ATTR_UPCALL_PID])
2075 err = ovs_vport_cmd_validate(a);
2080 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2085 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2086 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2089 if (port_no >= DP_MAX_PORTS)
2092 vport = ovs_vport_ovsl(dp, port_no);
2097 for (port_no = 1; ; port_no++) {
2098 if (port_no >= DP_MAX_PORTS) {
2102 vport = ovs_vport_ovsl(dp, port_no);
2108 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2109 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2110 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2112 parms.port_no = port_no;
2113 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2115 vport = new_vport(&parms);
2116 err = PTR_ERR(vport);
2121 if (a[OVS_VPORT_ATTR_STATS])
2122 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2124 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
2126 if (IS_ERR(reply)) {
2127 err = PTR_ERR(reply);
2128 ovs_dp_detach_port(vport);
2132 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2140 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2142 struct nlattr **a = info->attrs;
2143 struct sk_buff *reply;
2144 struct vport *vport;
2147 err = ovs_vport_cmd_validate(a);
2152 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2153 err = PTR_ERR(vport);
2157 if (a[OVS_VPORT_ATTR_TYPE] &&
2158 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2163 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2169 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2170 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2175 if (a[OVS_VPORT_ATTR_STATS])
2176 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2178 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2179 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2181 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2182 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2186 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2197 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2199 struct nlattr **a = info->attrs;
2200 struct sk_buff *reply;
2201 struct vport *vport;
2204 err = ovs_vport_cmd_validate(a);
2209 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2210 err = PTR_ERR(vport);
2214 if (vport->port_no == OVSP_LOCAL) {
2219 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2220 info->snd_seq, OVS_VPORT_CMD_DEL);
2221 err = PTR_ERR(reply);
2226 ovs_dp_detach_port(vport);
2228 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
2236 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2238 struct nlattr **a = info->attrs;
2239 struct ovs_header *ovs_header = info->userhdr;
2240 struct sk_buff *reply;
2241 struct vport *vport;
2244 err = ovs_vport_cmd_validate(a);
2249 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2250 err = PTR_ERR(vport);
2254 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2255 info->snd_seq, OVS_VPORT_CMD_NEW);
2256 err = PTR_ERR(reply);
2262 return genlmsg_reply(reply, info);
2270 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2272 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2273 struct datapath *dp;
2274 int bucket = cb->args[0], skip = cb->args[1];
2277 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2282 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2283 struct vport *vport;
2286 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2288 ovs_vport_cmd_fill_info(vport, skb,
2289 NETLINK_CB(cb->skb).portid,
2292 OVS_VPORT_CMD_NEW) < 0)
2308 static struct genl_ops dp_vport_genl_ops[] = {
2309 { .cmd = OVS_VPORT_CMD_NEW,
2310 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2311 .policy = vport_policy,
2312 .doit = ovs_vport_cmd_new
2314 { .cmd = OVS_VPORT_CMD_DEL,
2315 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2316 .policy = vport_policy,
2317 .doit = ovs_vport_cmd_del
2319 { .cmd = OVS_VPORT_CMD_GET,
2320 .flags = 0, /* OK for unprivileged users. */
2321 .policy = vport_policy,
2322 .doit = ovs_vport_cmd_get,
2323 .dumpit = ovs_vport_cmd_dump
2325 { .cmd = OVS_VPORT_CMD_SET,
2326 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2327 .policy = vport_policy,
2328 .doit = ovs_vport_cmd_set,
2332 struct genl_family_and_ops {
2333 struct genl_family *family;
2334 struct genl_ops *ops;
2336 struct genl_multicast_group *group;
2339 static const struct genl_family_and_ops dp_genl_families[] = {
2340 { &dp_datapath_genl_family,
2341 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2342 &ovs_dp_datapath_multicast_group },
2343 { &dp_vport_genl_family,
2344 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2345 &ovs_dp_vport_multicast_group },
2346 { &dp_flow_genl_family,
2347 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2348 &ovs_dp_flow_multicast_group },
2349 { &dp_packet_genl_family,
2350 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2354 static void dp_unregister_genl(int n_families)
2358 for (i = 0; i < n_families; i++)
2359 genl_unregister_family(dp_genl_families[i].family);
2362 static int dp_register_genl(void)
2369 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2370 const struct genl_family_and_ops *f = &dp_genl_families[i];
2372 err = genl_register_family_with_ops(f->family, f->ops,
2379 err = genl_register_mc_group(f->family, f->group);
2388 dp_unregister_genl(n_registered);
2392 static void rehash_flow_table(struct work_struct *work)
2394 struct datapath *dp;
2400 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2402 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2403 struct flow_table *old_table = ovsl_dereference(dp->table);
2404 struct flow_table *new_table;
2406 new_table = ovs_flow_tbl_rehash(old_table);
2407 if (!IS_ERR(new_table)) {
2408 rcu_assign_pointer(dp->table, new_table);
2409 ovs_flow_tbl_destroy(old_table, true);
2415 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2418 static int __net_init ovs_init_net(struct net *net)
2420 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2422 INIT_LIST_HEAD(&ovs_net->dps);
2423 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2427 static void __net_exit ovs_exit_net(struct net *net)
2429 struct datapath *dp, *dp_next;
2430 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2433 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2437 cancel_work_sync(&ovs_net->dp_notify_work);
2440 static struct pernet_operations ovs_net_ops = {
2441 .init = ovs_init_net,
2442 .exit = ovs_exit_net,
2444 .size = sizeof(struct ovs_net),
2447 static int __init dp_init(void)
2451 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2453 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2456 err = ovs_workqueues_init();
2460 err = ovs_flow_init();
2464 err = ovs_vport_init();
2466 goto error_flow_exit;
2468 err = register_pernet_device(&ovs_net_ops);
2470 goto error_vport_exit;
2472 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2474 goto error_netns_exit;
2476 err = dp_register_genl();
2478 goto error_unreg_notifier;
2480 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2484 error_unreg_notifier:
2485 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2487 unregister_pernet_device(&ovs_net_ops);
2493 ovs_workqueues_exit();
2498 static void dp_cleanup(void)
2500 cancel_delayed_work_sync(&rehash_flow_wq);
2501 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2502 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2503 unregister_pernet_device(&ovs_net_ops);
2507 ovs_workqueues_exit();
2510 module_init(dp_init);
2511 module_exit(dp_cleanup);
2513 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2514 MODULE_LICENSE("GPL");
2515 MODULE_VERSION(VERSION);