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 "flow_table.h"
59 #include "flow_netlink.h"
60 #include "vport-internal_dev.h"
61 #include "vport-netdev.h"
63 int ovs_net_id __read_mostly;
65 /* Check if need to build a reply message.
66 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
67 static bool ovs_must_notify(struct genl_info *info,
68 const struct genl_multicast_group *grp)
70 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
71 netlink_has_listeners(genl_info_net(info)->genl_sock, 0);
74 static void ovs_notify(struct genl_family *family,
75 struct sk_buff *skb, struct genl_info *info)
77 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
78 0, info->nlhdr, GFP_KERNEL);
84 * All writes e.g. Writes to device state (add/remove datapath, port, set
85 * operations on vports, etc.), Writes to other state (flow table
86 * modifications, set miscellaneous datapath parameters, etc.) are protected
89 * Reads are protected by RCU.
91 * There are a few special cases (mostly stats) that have their own
92 * synchronization but they nest under all of above and don't interact with
95 * The RTNL lock nests inside ovs_mutex.
98 static DEFINE_MUTEX(ovs_mutex);
102 mutex_lock(&ovs_mutex);
105 void ovs_unlock(void)
107 mutex_unlock(&ovs_mutex);
110 #ifdef CONFIG_LOCKDEP
111 int lockdep_ovsl_is_held(void)
114 return lockdep_is_held(&ovs_mutex);
120 static struct vport *new_vport(const struct vport_parms *);
121 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
122 const struct dp_upcall_info *);
123 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
124 const struct dp_upcall_info *);
126 /* Must be called with rcu_read_lock or ovs_mutex. */
127 static struct datapath *get_dp(struct net *net, int dp_ifindex)
129 struct datapath *dp = NULL;
130 struct net_device *dev;
133 dev = dev_get_by_index_rcu(net, dp_ifindex);
135 struct vport *vport = ovs_internal_dev_get_vport(dev);
144 /* Must be called with rcu_read_lock or ovs_mutex. */
145 static const char *ovs_dp_name(const struct datapath *dp)
147 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
148 return vport->ops->get_name(vport);
151 static int get_dpifindex(struct datapath *dp)
158 local = ovs_vport_rcu(dp, OVSP_LOCAL);
160 ifindex = netdev_vport_priv(local)->dev->ifindex;
169 static void destroy_dp_rcu(struct rcu_head *rcu)
171 struct datapath *dp = container_of(rcu, struct datapath, rcu);
173 free_percpu(dp->stats_percpu);
174 release_net(ovs_dp_get_net(dp));
179 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
182 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
185 /* Called with ovs_mutex or RCU read lock. */
186 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
189 struct hlist_head *head;
191 head = vport_hash_bucket(dp, port_no);
192 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
193 if (vport->port_no == port_no)
199 /* Called with ovs_mutex. */
200 static struct vport *new_vport(const struct vport_parms *parms)
204 vport = ovs_vport_add(parms);
205 if (!IS_ERR(vport)) {
206 struct datapath *dp = parms->dp;
207 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
209 hlist_add_head_rcu(&vport->dp_hash_node, head);
214 void ovs_dp_detach_port(struct vport *p)
218 /* First drop references to device. */
219 hlist_del_rcu(&p->dp_hash_node);
221 /* Then destroy it. */
225 /* Must be called with rcu_read_lock. */
226 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
228 struct datapath *dp = p->dp;
229 struct sw_flow *flow;
230 struct dp_stats_percpu *stats;
231 struct sw_flow_key key;
236 stats = this_cpu_ptr(dp->stats_percpu);
238 /* Extract flow from 'skb' into 'key'. */
239 error = ovs_flow_extract(skb, p->port_no, &key);
240 if (unlikely(error)) {
246 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
247 if (unlikely(!flow)) {
248 struct dp_upcall_info upcall;
250 upcall.cmd = OVS_PACKET_CMD_MISS;
252 upcall.userdata = NULL;
253 upcall.portid = p->upcall_portid;
254 ovs_dp_upcall(dp, skb, &upcall);
256 stats_counter = &stats->n_missed;
260 OVS_CB(skb)->flow = flow;
261 OVS_CB(skb)->pkt_key = &key;
263 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
264 ovs_execute_actions(dp, skb);
265 stats_counter = &stats->n_hit;
268 /* Update datapath statistics. */
269 u64_stats_update_begin(&stats->syncp);
271 stats->n_mask_hit += n_mask_hit;
272 u64_stats_update_end(&stats->syncp);
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,
282 .parallel_ops = true,
285 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
286 const struct dp_upcall_info *upcall_info)
288 struct dp_stats_percpu *stats;
291 if (upcall_info->portid == 0) {
296 if (!skb_is_gso(skb))
297 err = queue_userspace_packet(dp, skb, upcall_info);
299 err = queue_gso_packets(dp, skb, upcall_info);
306 stats = this_cpu_ptr(dp->stats_percpu);
308 u64_stats_update_begin(&stats->syncp);
310 u64_stats_update_end(&stats->syncp);
315 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
316 const struct dp_upcall_info *upcall_info)
318 unsigned short gso_type = skb_shinfo(skb)->gso_type;
319 struct dp_upcall_info later_info;
320 struct sw_flow_key later_key;
321 struct sk_buff *segs, *nskb;
324 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
326 return PTR_ERR(segs);
328 /* Queue all of the segments. */
331 err = queue_userspace_packet(dp, skb, upcall_info);
335 if (skb == segs && gso_type & SKB_GSO_UDP) {
336 /* The initial flow key extracted by ovs_flow_extract()
337 * in this case is for a first fragment, so we need to
338 * properly mark later fragments.
340 later_key = *upcall_info->key;
341 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
343 later_info = *upcall_info;
344 later_info.key = &later_key;
345 upcall_info = &later_info;
347 } while ((skb = skb->next));
349 /* Free all of the segments. */
357 } while ((skb = nskb));
361 static size_t key_attr_size(void)
363 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
364 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
365 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
366 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
367 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
368 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
369 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
370 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
371 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
372 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
373 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
374 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
375 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
376 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
377 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
378 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
379 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
380 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
381 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
384 static size_t upcall_msg_size(const struct nlattr *userdata,
387 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
388 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
389 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391 /* OVS_PACKET_ATTR_USERDATA */
393 size += NLA_ALIGN(userdata->nla_len);
398 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
399 const struct dp_upcall_info *upcall_info)
401 struct ovs_header *upcall;
402 struct sk_buff *nskb = NULL;
403 struct sk_buff *user_skb; /* to be queued to userspace */
405 struct genl_info info = {
406 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
407 .snd_portid = upcall_info->portid,
413 dp_ifindex = get_dpifindex(dp);
417 if (vlan_tx_tag_present(skb)) {
418 nskb = skb_clone(skb, GFP_ATOMIC);
422 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
430 if (nla_attr_size(skb->len) > USHRT_MAX) {
435 /* Complete checksum if needed */
436 if (skb->ip_summed == CHECKSUM_PARTIAL &&
437 (err = skb_checksum_help(skb)))
440 /* Older versions of OVS user space enforce alignment of the last
441 * Netlink attribute to NLA_ALIGNTO which would require extensive
442 * padding logic. Only perform zerocopy if padding is not required.
444 if (dp->user_features & OVS_DP_F_UNALIGNED)
445 hlen = skb_zerocopy_headlen(skb);
449 len = upcall_msg_size(upcall_info->userdata, hlen);
450 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
456 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
457 0, upcall_info->cmd);
458 upcall->dp_ifindex = dp_ifindex;
460 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
461 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
462 nla_nest_end(user_skb, nla);
464 if (upcall_info->userdata)
465 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
466 nla_len(upcall_info->userdata),
467 nla_data(upcall_info->userdata));
469 /* Only reserve room for attribute header, packet data is added
470 * in skb_zerocopy() */
471 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
475 nla->nla_len = nla_attr_size(skb->len);
477 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
481 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
482 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
483 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
486 memset(skb_put(user_skb, plen), 0, plen);
489 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
491 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
499 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
501 struct ovs_header *ovs_header = info->userhdr;
502 struct nlattr **a = info->attrs;
503 struct sw_flow_actions *acts;
504 struct sk_buff *packet;
505 struct sw_flow *flow;
512 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
513 !a[OVS_PACKET_ATTR_ACTIONS])
516 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
517 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
521 skb_reserve(packet, NET_IP_ALIGN);
523 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
525 skb_reset_mac_header(packet);
526 eth = eth_hdr(packet);
528 /* Normally, setting the skb 'protocol' field would be handled by a
529 * call to eth_type_trans(), but it assumes there's a sending
530 * device, which we may not have. */
531 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
532 packet->protocol = eth->h_proto;
534 packet->protocol = htons(ETH_P_802_2);
536 /* Build an sw_flow for sending this packet. */
537 flow = ovs_flow_alloc();
542 err = ovs_flow_extract(packet, -1, &flow->key);
546 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
549 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
554 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
555 &flow->key, 0, &acts);
556 rcu_assign_pointer(flow->sf_acts, acts);
560 OVS_CB(packet)->flow = flow;
561 OVS_CB(packet)->pkt_key = &flow->key;
562 packet->priority = flow->key.phy.priority;
563 packet->mark = flow->key.phy.skb_mark;
566 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
572 err = ovs_execute_actions(dp, packet);
576 ovs_flow_free(flow, false);
582 ovs_flow_free(flow, false);
589 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
590 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
591 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
592 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
595 static const struct genl_ops dp_packet_genl_ops[] = {
596 { .cmd = OVS_PACKET_CMD_EXECUTE,
597 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
598 .policy = packet_policy,
599 .doit = ovs_packet_cmd_execute
603 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
604 struct ovs_dp_megaflow_stats *mega_stats)
608 memset(mega_stats, 0, sizeof(*mega_stats));
610 stats->n_flows = ovs_flow_tbl_count(&dp->table);
611 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
613 stats->n_hit = stats->n_missed = stats->n_lost = 0;
615 for_each_possible_cpu(i) {
616 const struct dp_stats_percpu *percpu_stats;
617 struct dp_stats_percpu local_stats;
620 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
623 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
624 local_stats = *percpu_stats;
625 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
627 stats->n_hit += local_stats.n_hit;
628 stats->n_missed += local_stats.n_missed;
629 stats->n_lost += local_stats.n_lost;
630 mega_stats->n_mask_hit += local_stats.n_mask_hit;
634 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
635 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
636 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
637 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
640 static struct genl_family dp_flow_genl_family = {
641 .id = GENL_ID_GENERATE,
642 .hdrsize = sizeof(struct ovs_header),
643 .name = OVS_FLOW_FAMILY,
644 .version = OVS_FLOW_VERSION,
645 .maxattr = OVS_FLOW_ATTR_MAX,
647 .parallel_ops = true,
650 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
651 .name = OVS_FLOW_MCGROUP
654 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
656 return NLMSG_ALIGN(sizeof(struct ovs_header))
657 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
658 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
659 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
660 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
661 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
662 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
665 /* Called with ovs_mutex or RCU read lock. */
666 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
667 struct sk_buff *skb, u32 portid,
668 u32 seq, u32 flags, u8 cmd)
670 const int skb_orig_len = skb->len;
671 struct nlattr *start;
672 struct ovs_flow_stats stats;
675 struct ovs_header *ovs_header;
679 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
683 ovs_header->dp_ifindex = get_dpifindex(dp);
686 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
688 goto nla_put_failure;
690 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
693 nla_nest_end(skb, nla);
695 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
697 goto nla_put_failure;
699 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
703 nla_nest_end(skb, nla);
705 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
707 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
708 goto nla_put_failure;
710 if (stats.n_packets &&
711 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
712 goto nla_put_failure;
714 if ((u8)ntohs(tcp_flags) &&
715 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
716 goto nla_put_failure;
718 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
719 * this is the first flow to be dumped into 'skb'. This is unusual for
720 * Netlink but individual action lists can be longer than
721 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
722 * The userspace caller can always fetch the actions separately if it
723 * really wants them. (Most userspace callers in fact don't care.)
725 * This can only fail for dump operations because the skb is always
726 * properly sized for single flows.
728 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
730 const struct sw_flow_actions *sf_acts;
732 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
734 err = ovs_nla_put_actions(sf_acts->actions,
735 sf_acts->actions_len, skb);
737 nla_nest_end(skb, start);
742 nla_nest_cancel(skb, start);
744 } else if (skb_orig_len)
745 goto nla_put_failure;
747 return genlmsg_end(skb, ovs_header);
752 genlmsg_cancel(skb, ovs_header);
756 /* Must be called with ovs_mutex. */
757 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
758 struct genl_info *info,
764 if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
767 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
769 skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
771 return ERR_PTR(-ENOMEM);
776 /* Must be called with ovs_mutex. */
777 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
779 struct genl_info *info,
785 skb = ovs_flow_cmd_alloc_info(flow, info, always);
786 if (!skb || IS_ERR(skb))
789 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
790 info->snd_seq, 0, cmd);
795 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
797 struct nlattr **a = info->attrs;
798 struct ovs_header *ovs_header = info->userhdr;
799 struct sw_flow_key key, masked_key;
800 struct sw_flow *flow = NULL;
801 struct sw_flow_mask mask;
802 struct sk_buff *reply;
804 struct sw_flow_actions *acts = NULL;
805 struct sw_flow_match match;
810 if (!a[OVS_FLOW_ATTR_KEY])
813 ovs_match_init(&match, &key, &mask);
814 error = ovs_nla_get_match(&match,
815 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
819 /* Validate actions. */
820 if (a[OVS_FLOW_ATTR_ACTIONS]) {
821 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
822 error = PTR_ERR(acts);
826 ovs_flow_mask_key(&masked_key, &key, &mask);
827 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
828 &masked_key, 0, &acts);
830 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
833 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
834 /* OVS_FLOW_CMD_NEW must have actions. */
840 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
845 /* Check if this is a duplicate flow */
846 flow = ovs_flow_tbl_lookup(&dp->table, &key);
848 /* Bail out if we're not allowed to create a new flow. */
850 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
854 flow = ovs_flow_alloc();
856 error = PTR_ERR(flow);
860 flow->key = masked_key;
861 flow->unmasked_key = key;
862 rcu_assign_pointer(flow->sf_acts, acts);
864 /* Put flow in bucket. */
865 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
871 reply = ovs_flow_cmd_build_info(flow, dp, info,
872 OVS_FLOW_CMD_NEW, false);
874 /* We found a matching flow. */
875 /* Bail out if we're not allowed to modify an existing flow.
876 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
877 * because Generic Netlink treats the latter as a dump
878 * request. We also accept NLM_F_EXCL in case that bug ever
882 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
883 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
886 /* The unmasked key has to be the same for flow updates. */
887 if (!ovs_flow_cmp_unmasked_key(flow, &match))
890 /* Update actions, if present. */
892 struct sw_flow_actions *old_acts;
894 old_acts = ovsl_dereference(flow->sf_acts);
895 rcu_assign_pointer(flow->sf_acts, acts);
896 ovs_nla_free_flow_actions(old_acts);
898 reply = ovs_flow_cmd_build_info(flow, dp, info,
899 OVS_FLOW_CMD_NEW, false);
902 if (a[OVS_FLOW_ATTR_CLEAR])
903 ovs_flow_stats_clear(flow);
909 ovs_notify(&dp_flow_genl_family, reply, info);
911 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
918 ovs_flow_free(flow, false);
927 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
929 struct nlattr **a = info->attrs;
930 struct ovs_header *ovs_header = info->userhdr;
931 struct sw_flow_key key;
932 struct sk_buff *reply;
933 struct sw_flow *flow;
935 struct sw_flow_match match;
938 if (!a[OVS_FLOW_ATTR_KEY]) {
939 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
943 ovs_match_init(&match, &key, NULL);
944 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
949 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
955 flow = ovs_flow_tbl_lookup(&dp->table, &key);
956 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
961 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW, true);
963 err = PTR_ERR(reply);
968 return genlmsg_reply(reply, info);
974 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
976 struct nlattr **a = info->attrs;
977 struct ovs_header *ovs_header = info->userhdr;
978 struct sw_flow_key key;
979 struct sk_buff *reply;
980 struct sw_flow *flow;
982 struct sw_flow_match match;
986 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
992 if (!a[OVS_FLOW_ATTR_KEY]) {
993 err = ovs_flow_tbl_flush(&dp->table);
997 ovs_match_init(&match, &key, NULL);
998 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1002 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1003 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1008 reply = ovs_flow_cmd_alloc_info(flow, info, false);
1009 if (IS_ERR(reply)) {
1010 err = PTR_ERR(reply);
1014 ovs_flow_tbl_remove(&dp->table, flow);
1017 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1022 ovs_flow_free(flow, true);
1026 ovs_notify(&dp_flow_genl_family, reply, info);
1033 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1035 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1036 struct table_instance *ti;
1037 struct datapath *dp;
1040 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1046 ti = rcu_dereference(dp->table.ti);
1048 struct sw_flow *flow;
1051 bucket = cb->args[0];
1053 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1057 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1058 NETLINK_CB(cb->skb).portid,
1059 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1060 OVS_FLOW_CMD_NEW) < 0)
1063 cb->args[0] = bucket;
1070 static const struct genl_ops dp_flow_genl_ops[] = {
1071 { .cmd = OVS_FLOW_CMD_NEW,
1072 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1073 .policy = flow_policy,
1074 .doit = ovs_flow_cmd_new_or_set
1076 { .cmd = OVS_FLOW_CMD_DEL,
1077 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1078 .policy = flow_policy,
1079 .doit = ovs_flow_cmd_del
1081 { .cmd = OVS_FLOW_CMD_GET,
1082 .flags = 0, /* OK for unprivileged users. */
1083 .policy = flow_policy,
1084 .doit = ovs_flow_cmd_get,
1085 .dumpit = ovs_flow_cmd_dump
1087 { .cmd = OVS_FLOW_CMD_SET,
1088 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1089 .policy = flow_policy,
1090 .doit = ovs_flow_cmd_new_or_set,
1094 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1095 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1096 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1097 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1100 static struct genl_family dp_datapath_genl_family = {
1101 .id = GENL_ID_GENERATE,
1102 .hdrsize = sizeof(struct ovs_header),
1103 .name = OVS_DATAPATH_FAMILY,
1104 .version = OVS_DATAPATH_VERSION,
1105 .maxattr = OVS_DP_ATTR_MAX,
1107 .parallel_ops = true,
1110 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1111 .name = OVS_DATAPATH_MCGROUP
1114 static size_t ovs_dp_cmd_msg_size(void)
1116 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1118 msgsize += nla_total_size(IFNAMSIZ);
1119 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1120 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1121 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1126 /* Called with ovs_mutex or RCU read lock. */
1127 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1128 u32 portid, u32 seq, u32 flags, u8 cmd)
1130 struct ovs_header *ovs_header;
1131 struct ovs_dp_stats dp_stats;
1132 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1135 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1140 ovs_header->dp_ifindex = get_dpifindex(dp);
1142 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1144 goto nla_put_failure;
1146 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1147 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1149 goto nla_put_failure;
1151 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1152 sizeof(struct ovs_dp_megaflow_stats),
1153 &dp_megaflow_stats))
1154 goto nla_put_failure;
1156 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1157 goto nla_put_failure;
1159 return genlmsg_end(skb, ovs_header);
1162 genlmsg_cancel(skb, ovs_header);
1167 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1169 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1172 /* Called with rcu_read_lock or ovs_mutex. */
1173 static struct datapath *lookup_datapath(struct net *net,
1174 struct ovs_header *ovs_header,
1175 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1177 struct datapath *dp;
1179 if (!a[OVS_DP_ATTR_NAME])
1180 dp = get_dp(net, ovs_header->dp_ifindex);
1182 struct vport *vport;
1184 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1185 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1187 return dp ? dp : ERR_PTR(-ENODEV);
1190 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1192 struct datapath *dp;
1194 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1198 WARN(dp->user_features, "Dropping previously announced user features\n");
1199 dp->user_features = 0;
1202 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1204 if (a[OVS_DP_ATTR_USER_FEATURES])
1205 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1208 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1210 struct nlattr **a = info->attrs;
1211 struct vport_parms parms;
1212 struct sk_buff *reply;
1213 struct datapath *dp;
1214 struct vport *vport;
1215 struct ovs_net *ovs_net;
1219 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1222 reply = ovs_dp_cmd_alloc_info(info);
1227 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1229 goto err_free_reply;
1231 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1233 /* Allocate table. */
1234 err = ovs_flow_tbl_init(&dp->table);
1238 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1239 if (!dp->stats_percpu) {
1241 goto err_destroy_table;
1244 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1248 goto err_destroy_percpu;
1251 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1252 INIT_HLIST_HEAD(&dp->ports[i]);
1254 /* Set up our datapath device. */
1255 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1256 parms.type = OVS_VPORT_TYPE_INTERNAL;
1257 parms.options = NULL;
1259 parms.port_no = OVSP_LOCAL;
1260 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1262 ovs_dp_change(dp, a);
1264 /* So far only local changes have been made, now need the lock. */
1267 vport = new_vport(&parms);
1268 if (IS_ERR(vport)) {
1269 err = PTR_ERR(vport);
1273 if (err == -EEXIST) {
1274 /* An outdated user space instance that does not understand
1275 * the concept of user_features has attempted to create a new
1276 * datapath and is likely to reuse it. Drop all user features.
1278 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1279 ovs_dp_reset_user_features(skb, info);
1282 goto err_destroy_ports_array;
1285 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1286 info->snd_seq, 0, OVS_DP_CMD_NEW);
1289 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1290 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1294 ovs_notify(&dp_datapath_genl_family, reply, info);
1297 err_destroy_ports_array:
1301 free_percpu(dp->stats_percpu);
1303 ovs_flow_tbl_destroy(&dp->table, false);
1305 release_net(ovs_dp_get_net(dp));
1313 /* Called with ovs_mutex. */
1314 static void __dp_destroy(struct datapath *dp)
1318 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1319 struct vport *vport;
1320 struct hlist_node *n;
1322 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1323 if (vport->port_no != OVSP_LOCAL)
1324 ovs_dp_detach_port(vport);
1327 list_del_rcu(&dp->list_node);
1329 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1330 * all ports in datapath are destroyed first before freeing datapath.
1332 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1334 /* RCU destroy the flow table */
1335 ovs_flow_tbl_destroy(&dp->table, true);
1337 call_rcu(&dp->rcu, destroy_dp_rcu);
1340 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1342 struct sk_buff *reply;
1343 struct datapath *dp;
1346 reply = ovs_dp_cmd_alloc_info(info);
1351 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1354 goto err_unlock_free;
1356 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1357 info->snd_seq, 0, OVS_DP_CMD_DEL);
1363 ovs_notify(&dp_datapath_genl_family, reply, info);
1373 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1375 struct sk_buff *reply;
1376 struct datapath *dp;
1379 reply = ovs_dp_cmd_alloc_info(info);
1384 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1387 goto err_unlock_free;
1389 ovs_dp_change(dp, info->attrs);
1391 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1392 info->snd_seq, 0, OVS_DP_CMD_NEW);
1396 ovs_notify(&dp_datapath_genl_family, reply, info);
1406 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1408 struct sk_buff *reply;
1409 struct datapath *dp;
1412 reply = ovs_dp_cmd_alloc_info(info);
1417 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1420 goto err_unlock_free;
1422 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1423 info->snd_seq, 0, OVS_DP_CMD_NEW);
1427 return genlmsg_reply(reply, info);
1435 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1437 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1438 struct datapath *dp;
1439 int skip = cb->args[0];
1443 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1445 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1446 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1447 OVS_DP_CMD_NEW) < 0)
1458 static const struct genl_ops dp_datapath_genl_ops[] = {
1459 { .cmd = OVS_DP_CMD_NEW,
1460 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1461 .policy = datapath_policy,
1462 .doit = ovs_dp_cmd_new
1464 { .cmd = OVS_DP_CMD_DEL,
1465 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1466 .policy = datapath_policy,
1467 .doit = ovs_dp_cmd_del
1469 { .cmd = OVS_DP_CMD_GET,
1470 .flags = 0, /* OK for unprivileged users. */
1471 .policy = datapath_policy,
1472 .doit = ovs_dp_cmd_get,
1473 .dumpit = ovs_dp_cmd_dump
1475 { .cmd = OVS_DP_CMD_SET,
1476 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1477 .policy = datapath_policy,
1478 .doit = ovs_dp_cmd_set,
1482 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1483 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1484 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1485 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1486 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1487 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1488 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1491 struct genl_family dp_vport_genl_family = {
1492 .id = GENL_ID_GENERATE,
1493 .hdrsize = sizeof(struct ovs_header),
1494 .name = OVS_VPORT_FAMILY,
1495 .version = OVS_VPORT_VERSION,
1496 .maxattr = OVS_VPORT_ATTR_MAX,
1498 .parallel_ops = true,
1501 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1502 .name = OVS_VPORT_MCGROUP
1505 /* Called with ovs_mutex or RCU read lock. */
1506 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1507 u32 portid, u32 seq, u32 flags, u8 cmd)
1509 struct ovs_header *ovs_header;
1510 struct ovs_vport_stats vport_stats;
1513 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1518 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1520 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1521 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1522 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1523 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1524 goto nla_put_failure;
1526 ovs_vport_get_stats(vport, &vport_stats);
1527 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1529 goto nla_put_failure;
1531 err = ovs_vport_get_options(vport, skb);
1532 if (err == -EMSGSIZE)
1535 return genlmsg_end(skb, ovs_header);
1540 genlmsg_cancel(skb, ovs_header);
1544 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1546 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1549 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1550 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1553 struct sk_buff *skb;
1556 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1558 return ERR_PTR(-ENOMEM);
1560 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1566 /* Called with ovs_mutex or RCU read lock. */
1567 static struct vport *lookup_vport(struct net *net,
1568 struct ovs_header *ovs_header,
1569 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1571 struct datapath *dp;
1572 struct vport *vport;
1574 if (a[OVS_VPORT_ATTR_NAME]) {
1575 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1577 return ERR_PTR(-ENODEV);
1578 if (ovs_header->dp_ifindex &&
1579 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1580 return ERR_PTR(-ENODEV);
1582 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1583 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1585 if (port_no >= DP_MAX_PORTS)
1586 return ERR_PTR(-EFBIG);
1588 dp = get_dp(net, ovs_header->dp_ifindex);
1590 return ERR_PTR(-ENODEV);
1592 vport = ovs_vport_ovsl_rcu(dp, port_no);
1594 return ERR_PTR(-ENODEV);
1597 return ERR_PTR(-EINVAL);
1600 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1602 struct nlattr **a = info->attrs;
1603 struct ovs_header *ovs_header = info->userhdr;
1604 struct vport_parms parms;
1605 struct sk_buff *reply;
1606 struct vport *vport;
1607 struct datapath *dp;
1611 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1612 !a[OVS_VPORT_ATTR_UPCALL_PID])
1615 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1616 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1617 if (port_no >= DP_MAX_PORTS)
1620 reply = ovs_vport_cmd_alloc_info();
1625 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1628 goto exit_unlock_free;
1631 vport = ovs_vport_ovsl(dp, port_no);
1634 goto exit_unlock_free;
1636 for (port_no = 1; ; port_no++) {
1637 if (port_no >= DP_MAX_PORTS) {
1639 goto exit_unlock_free;
1641 vport = ovs_vport_ovsl(dp, port_no);
1647 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1648 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1649 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1651 parms.port_no = port_no;
1652 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1654 vport = new_vport(&parms);
1655 err = PTR_ERR(vport);
1657 goto exit_unlock_free;
1659 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1660 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1664 ovs_notify(&dp_vport_genl_family, reply, info);
1673 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1675 struct nlattr **a = info->attrs;
1676 struct sk_buff *reply;
1677 struct vport *vport;
1680 reply = ovs_vport_cmd_alloc_info();
1685 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1686 err = PTR_ERR(vport);
1688 goto exit_unlock_free;
1690 if (a[OVS_VPORT_ATTR_TYPE] &&
1691 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1693 goto exit_unlock_free;
1696 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1697 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1699 goto exit_unlock_free;
1702 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1703 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1705 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1706 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1710 ovs_notify(&dp_vport_genl_family, reply, info);
1719 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1721 struct nlattr **a = info->attrs;
1722 struct sk_buff *reply;
1723 struct vport *vport;
1726 reply = ovs_vport_cmd_alloc_info();
1731 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1732 err = PTR_ERR(vport);
1734 goto exit_unlock_free;
1736 if (vport->port_no == OVSP_LOCAL) {
1738 goto exit_unlock_free;
1741 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1742 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1744 ovs_dp_detach_port(vport);
1747 ovs_notify(&dp_vport_genl_family, reply, info);
1756 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1758 struct nlattr **a = info->attrs;
1759 struct ovs_header *ovs_header = info->userhdr;
1760 struct sk_buff *reply;
1761 struct vport *vport;
1764 reply = ovs_vport_cmd_alloc_info();
1769 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1770 err = PTR_ERR(vport);
1772 goto exit_unlock_free;
1773 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1774 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1778 return genlmsg_reply(reply, info);
1786 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1788 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1789 struct datapath *dp;
1790 int bucket = cb->args[0], skip = cb->args[1];
1794 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1799 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1800 struct vport *vport;
1803 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1805 ovs_vport_cmd_fill_info(vport, skb,
1806 NETLINK_CB(cb->skb).portid,
1809 OVS_VPORT_CMD_NEW) < 0)
1825 static const struct genl_ops dp_vport_genl_ops[] = {
1826 { .cmd = OVS_VPORT_CMD_NEW,
1827 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1828 .policy = vport_policy,
1829 .doit = ovs_vport_cmd_new
1831 { .cmd = OVS_VPORT_CMD_DEL,
1832 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1833 .policy = vport_policy,
1834 .doit = ovs_vport_cmd_del
1836 { .cmd = OVS_VPORT_CMD_GET,
1837 .flags = 0, /* OK for unprivileged users. */
1838 .policy = vport_policy,
1839 .doit = ovs_vport_cmd_get,
1840 .dumpit = ovs_vport_cmd_dump
1842 { .cmd = OVS_VPORT_CMD_SET,
1843 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1844 .policy = vport_policy,
1845 .doit = ovs_vport_cmd_set,
1849 struct genl_family_and_ops {
1850 struct genl_family *family;
1851 const struct genl_ops *ops;
1853 const struct genl_multicast_group *group;
1856 static const struct genl_family_and_ops dp_genl_families[] = {
1857 { &dp_datapath_genl_family,
1858 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1859 &ovs_dp_datapath_multicast_group },
1860 { &dp_vport_genl_family,
1861 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1862 &ovs_dp_vport_multicast_group },
1863 { &dp_flow_genl_family,
1864 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1865 &ovs_dp_flow_multicast_group },
1866 { &dp_packet_genl_family,
1867 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1871 static void dp_unregister_genl(int n_families)
1875 for (i = 0; i < n_families; i++)
1876 genl_unregister_family(dp_genl_families[i].family);
1879 static int dp_register_genl(void)
1886 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1887 const struct genl_family_and_ops *f = &dp_genl_families[i];
1889 f->family->ops = f->ops;
1890 f->family->n_ops = f->n_ops;
1891 f->family->mcgrps = f->group;
1892 f->family->n_mcgrps = f->group ? 1 : 0;
1893 err = genl_register_family(f->family);
1902 dp_unregister_genl(n_registered);
1906 static int __net_init ovs_init_net(struct net *net)
1908 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1910 INIT_LIST_HEAD(&ovs_net->dps);
1911 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1915 static void __net_exit ovs_exit_net(struct net *net)
1917 struct datapath *dp, *dp_next;
1918 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1921 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1925 cancel_work_sync(&ovs_net->dp_notify_work);
1928 static struct pernet_operations ovs_net_ops = {
1929 .init = ovs_init_net,
1930 .exit = ovs_exit_net,
1932 .size = sizeof(struct ovs_net),
1935 static int __init dp_init(void)
1939 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1941 pr_info("Open vSwitch switching datapath\n");
1943 err = ovs_flow_init();
1947 err = ovs_vport_init();
1949 goto error_flow_exit;
1951 err = register_pernet_device(&ovs_net_ops);
1953 goto error_vport_exit;
1955 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1957 goto error_netns_exit;
1959 err = dp_register_genl();
1961 goto error_unreg_notifier;
1965 error_unreg_notifier:
1966 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1968 unregister_pernet_device(&ovs_net_ops);
1977 static void dp_cleanup(void)
1979 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1980 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1981 unregister_pernet_device(&ovs_net_ops);
1987 module_init(dp_init);
1988 module_exit(dp_cleanup);
1990 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1991 MODULE_LICENSE("GPL");