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>
59 #include "flow_table.h"
60 #include "flow_netlink.h"
62 #include "vport-internal_dev.h"
63 #include "vport-netdev.h"
65 int ovs_net_id __read_mostly;
67 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
68 struct genl_multicast_group *grp)
70 genl_notify(skb, genl_info_net(info), info->snd_portid,
71 grp->id, info->nlhdr, GFP_KERNEL);
77 * All writes e.g. Writes to device state (add/remove datapath, port, set
78 * operations on vports, etc.), Writes to other state (flow table
79 * modifications, set miscellaneous datapath parameters, etc.) are protected
82 * Reads are protected by RCU.
84 * There are a few special cases (mostly stats) that have their own
85 * synchronization but they nest under all of above and don't interact with
88 * The RTNL lock nests inside ovs_mutex.
91 static DEFINE_MUTEX(ovs_mutex);
95 mutex_lock(&ovs_mutex);
100 mutex_unlock(&ovs_mutex);
103 #ifdef CONFIG_LOCKDEP
104 int lockdep_ovsl_is_held(void)
107 return lockdep_is_held(&ovs_mutex);
113 static struct vport *new_vport(const struct vport_parms *);
114 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
115 const struct dp_upcall_info *);
116 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
117 const struct dp_upcall_info *);
119 /* Must be called with rcu_read_lock or ovs_mutex. */
120 static struct datapath *get_dp(struct net *net, int dp_ifindex)
122 struct datapath *dp = NULL;
123 struct net_device *dev;
126 dev = dev_get_by_index_rcu(net, dp_ifindex);
128 struct vport *vport = ovs_internal_dev_get_vport(dev);
137 /* Must be called with rcu_read_lock or ovs_mutex. */
138 const char *ovs_dp_name(const struct datapath *dp)
140 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
141 return vport->ops->get_name(vport);
144 static int get_dpifindex(struct datapath *dp)
151 local = ovs_vport_rcu(dp, OVSP_LOCAL);
153 ifindex = netdev_vport_priv(local)->dev->ifindex;
162 static void destroy_dp_rcu(struct rcu_head *rcu)
164 struct datapath *dp = container_of(rcu, struct datapath, rcu);
166 free_percpu(dp->stats_percpu);
167 release_net(ovs_dp_get_net(dp));
172 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
175 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
178 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
181 struct hlist_head *head;
183 head = vport_hash_bucket(dp, port_no);
184 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
185 if (vport->port_no == port_no)
191 /* Called with ovs_mutex. */
192 static struct vport *new_vport(const struct vport_parms *parms)
196 vport = ovs_vport_add(parms);
197 if (!IS_ERR(vport)) {
198 struct datapath *dp = parms->dp;
199 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
201 hlist_add_head_rcu(&vport->dp_hash_node, head);
206 void ovs_dp_detach_port(struct vport *p)
210 /* First drop references to device. */
211 hlist_del_rcu(&p->dp_hash_node);
213 /* Then destroy it. */
217 /* Must be called with rcu_read_lock. */
218 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
220 struct datapath *dp = p->dp;
221 struct sw_flow *flow;
222 struct dp_stats_percpu *stats;
223 struct sw_flow_key key;
228 stats = this_cpu_ptr(dp->stats_percpu);
230 /* Extract flow from 'skb' into 'key'. */
231 error = ovs_flow_extract(skb, p->port_no, &key);
232 if (unlikely(error)) {
238 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
239 if (unlikely(!flow)) {
240 struct dp_upcall_info upcall;
242 upcall.cmd = OVS_PACKET_CMD_MISS;
244 upcall.userdata = NULL;
245 upcall.portid = p->upcall_portid;
246 ovs_dp_upcall(dp, skb, &upcall);
248 stats_counter = &stats->n_missed;
252 OVS_CB(skb)->flow = flow;
253 OVS_CB(skb)->pkt_key = &key;
255 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
256 ovs_execute_actions(dp, skb);
257 stats_counter = &stats->n_hit;
260 /* Update datapath statistics. */
261 u64_stats_update_begin(&stats->sync);
263 stats->n_mask_hit += n_mask_hit;
264 u64_stats_update_end(&stats->sync);
267 static struct genl_family dp_packet_genl_family = {
268 .id = GENL_ID_GENERATE,
269 .hdrsize = sizeof(struct ovs_header),
270 .name = OVS_PACKET_FAMILY,
271 .version = OVS_PACKET_VERSION,
272 .maxattr = OVS_PACKET_ATTR_MAX,
277 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
278 const struct dp_upcall_info *upcall_info)
280 struct dp_stats_percpu *stats;
283 if (upcall_info->portid == 0) {
288 if (!skb_is_gso(skb))
289 err = queue_userspace_packet(dp, skb, upcall_info);
291 err = queue_gso_packets(dp, skb, upcall_info);
298 stats = this_cpu_ptr(dp->stats_percpu);
300 u64_stats_update_begin(&stats->sync);
302 u64_stats_update_end(&stats->sync);
307 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
308 const struct dp_upcall_info *upcall_info)
310 unsigned short gso_type = skb_shinfo(skb)->gso_type;
311 struct dp_upcall_info later_info;
312 struct sw_flow_key later_key;
313 struct sk_buff *segs, *nskb;
316 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
318 return PTR_ERR(segs);
320 /* Queue all of the segments. */
323 err = queue_userspace_packet(dp, skb, upcall_info);
327 if (skb == segs && gso_type & SKB_GSO_UDP) {
328 /* The initial flow key extracted by ovs_flow_extract()
329 * in this case is for a first fragment, so we need to
330 * properly mark later fragments.
332 later_key = *upcall_info->key;
333 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
335 later_info = *upcall_info;
336 later_info.key = &later_key;
337 upcall_info = &later_info;
339 } while ((skb = skb->next));
341 /* Free all of the segments. */
349 } while ((skb = nskb));
353 static size_t key_attr_size(void)
355 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
356 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
357 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
358 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
359 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
360 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
361 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
362 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
363 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
364 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
365 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
366 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
367 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
368 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
369 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
370 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
371 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
372 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
373 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
376 static size_t upcall_msg_size(const struct nlattr *userdata,
379 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
380 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
381 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
383 /* OVS_PACKET_ATTR_USERDATA */
385 size += NLA_ALIGN(userdata->nla_len);
390 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
391 const struct dp_upcall_info *upcall_info)
393 struct ovs_header *upcall;
394 struct sk_buff *nskb = NULL;
395 struct sk_buff *user_skb; /* to be queued to userspace */
397 struct genl_info info = {
398 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
399 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
401 .snd_portid = upcall_info->portid,
407 dp_ifindex = get_dpifindex(dp);
411 if (vlan_tx_tag_present(skb)) {
412 nskb = skb_clone(skb, GFP_ATOMIC);
416 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
420 vlan_set_tci(nskb, 0);
425 if (nla_attr_size(skb->len) > USHRT_MAX) {
430 /* Complete checksum if needed */
431 if (skb->ip_summed == CHECKSUM_PARTIAL &&
432 (err = skb_checksum_help(skb)))
435 /* Older versions of OVS user space enforce alignment of the last
436 * Netlink attribute to NLA_ALIGNTO which would require extensive
437 * padding logic. Only perform zerocopy if padding is not required.
439 if (dp->user_features & OVS_DP_F_UNALIGNED)
440 hlen = skb_zerocopy_headlen(skb);
444 len = upcall_msg_size(upcall_info->userdata, hlen);
445 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
451 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
452 0, upcall_info->cmd);
453 upcall->dp_ifindex = dp_ifindex;
455 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
456 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
457 nla_nest_end(user_skb, nla);
459 if (upcall_info->userdata)
460 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
461 nla_len(upcall_info->userdata),
462 nla_data(upcall_info->userdata));
464 /* Only reserve room for attribute header, packet data is added
465 * in skb_zerocopy() */
466 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
470 nla->nla_len = nla_attr_size(skb->len);
472 skb_zerocopy(user_skb, skb, skb->len, hlen);
474 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
475 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
476 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
479 memset(skb_put(user_skb, plen), 0, plen);
482 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
484 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
490 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
492 struct ovs_header *ovs_header = info->userhdr;
493 struct nlattr **a = info->attrs;
494 struct sw_flow_actions *acts;
495 struct sk_buff *packet;
496 struct sw_flow *flow;
503 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
504 !a[OVS_PACKET_ATTR_ACTIONS])
507 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
508 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
512 skb_reserve(packet, NET_IP_ALIGN);
514 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
516 skb_reset_mac_header(packet);
517 eth = eth_hdr(packet);
519 /* Normally, setting the skb 'protocol' field would be handled by a
520 * call to eth_type_trans(), but it assumes there's a sending
521 * device, which we may not have. */
522 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
523 packet->protocol = eth->h_proto;
525 packet->protocol = htons(ETH_P_802_2);
527 /* Build an sw_flow for sending this packet. */
528 flow = ovs_flow_alloc();
533 err = ovs_flow_extract(packet, -1, &flow->key);
537 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
540 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
545 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
546 &flow->key, 0, &acts);
547 rcu_assign_pointer(flow->sf_acts, acts);
551 OVS_CB(packet)->flow = flow;
552 OVS_CB(packet)->pkt_key = &flow->key;
553 packet->priority = flow->key.phy.priority;
554 packet->mark = flow->key.phy.skb_mark;
557 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
563 err = ovs_execute_actions(dp, packet);
567 ovs_flow_free(flow, false);
573 ovs_flow_free(flow, false);
580 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
581 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
582 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
583 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
586 static struct genl_ops dp_packet_genl_ops[] = {
587 { .cmd = OVS_PACKET_CMD_EXECUTE,
588 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
589 .policy = packet_policy,
590 .doit = ovs_packet_cmd_execute
594 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
595 struct ovs_dp_megaflow_stats *mega_stats)
599 memset(mega_stats, 0, sizeof(*mega_stats));
601 stats->n_flows = ovs_flow_tbl_count(&dp->table);
602 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
604 stats->n_hit = stats->n_missed = stats->n_lost = 0;
606 for_each_possible_cpu(i) {
607 const struct dp_stats_percpu *percpu_stats;
608 struct dp_stats_percpu local_stats;
611 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
614 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
615 local_stats = *percpu_stats;
616 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
618 stats->n_hit += local_stats.n_hit;
619 stats->n_missed += local_stats.n_missed;
620 stats->n_lost += local_stats.n_lost;
621 mega_stats->n_mask_hit += local_stats.n_mask_hit;
625 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
626 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
627 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
628 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
631 static struct genl_family dp_flow_genl_family = {
632 .id = GENL_ID_GENERATE,
633 .hdrsize = sizeof(struct ovs_header),
634 .name = OVS_FLOW_FAMILY,
635 .version = OVS_FLOW_VERSION,
636 .maxattr = OVS_FLOW_ATTR_MAX,
641 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
642 .name = OVS_FLOW_MCGROUP
645 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
647 return NLMSG_ALIGN(sizeof(struct ovs_header))
648 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
649 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
650 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
651 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
652 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
653 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
656 /* Called with ovs_mutex. */
657 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
658 struct sk_buff *skb, u32 portid,
659 u32 seq, u32 flags, u8 cmd)
661 const int skb_orig_len = skb->len;
662 struct nlattr *start;
663 struct ovs_flow_stats stats;
666 struct ovs_header *ovs_header;
670 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
674 ovs_header->dp_ifindex = get_dpifindex(dp);
677 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
679 goto nla_put_failure;
681 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
684 nla_nest_end(skb, nla);
686 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
688 goto nla_put_failure;
690 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
694 nla_nest_end(skb, nla);
696 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
698 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
699 goto nla_put_failure;
701 if (stats.n_packets &&
702 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
703 goto nla_put_failure;
705 if ((u8)ntohs(tcp_flags) &&
706 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
707 goto nla_put_failure;
709 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
710 * this is the first flow to be dumped into 'skb'. This is unusual for
711 * Netlink but individual action lists can be longer than
712 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
713 * The userspace caller can always fetch the actions separately if it
714 * really wants them. (Most userspace callers in fact don't care.)
716 * This can only fail for dump operations because the skb is always
717 * properly sized for single flows.
719 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
721 const struct sw_flow_actions *sf_acts;
723 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
725 err = ovs_nla_put_actions(sf_acts->actions,
726 sf_acts->actions_len, skb);
728 nla_nest_end(skb, start);
733 nla_nest_cancel(skb, start);
735 } else if (skb_orig_len)
736 goto nla_put_failure;
738 return genlmsg_end(skb, ovs_header);
743 genlmsg_cancel(skb, ovs_header);
747 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
748 struct genl_info *info)
752 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
754 return genlmsg_new_unicast(len, info, GFP_KERNEL);
757 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
759 struct genl_info *info,
765 skb = ovs_flow_cmd_alloc_info(flow, info);
767 return ERR_PTR(-ENOMEM);
769 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
770 info->snd_seq, 0, cmd);
775 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
777 struct nlattr **a = info->attrs;
778 struct ovs_header *ovs_header = info->userhdr;
779 struct sw_flow_key key, masked_key;
780 struct sw_flow *flow = NULL;
781 struct sw_flow_mask mask;
782 struct sk_buff *reply;
784 struct sw_flow_actions *acts = NULL;
785 struct sw_flow_match match;
790 if (!a[OVS_FLOW_ATTR_KEY])
793 ovs_match_init(&match, &key, &mask);
794 error = ovs_nla_get_match(&match,
795 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
799 /* Validate actions. */
800 if (a[OVS_FLOW_ATTR_ACTIONS]) {
801 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
802 error = PTR_ERR(acts);
806 ovs_flow_mask_key(&masked_key, &key, &mask);
807 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
808 &masked_key, 0, &acts);
810 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
813 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
814 /* OVS_FLOW_CMD_NEW must have actions. */
820 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
825 /* Check if this is a duplicate flow */
826 flow = ovs_flow_tbl_lookup(&dp->table, &key);
828 /* Bail out if we're not allowed to create a new flow. */
830 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
834 flow = ovs_flow_alloc();
836 error = PTR_ERR(flow);
840 flow->key = masked_key;
841 flow->unmasked_key = key;
842 rcu_assign_pointer(flow->sf_acts, acts);
844 /* Put flow in bucket. */
845 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
851 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
853 /* We found a matching flow. */
854 /* Bail out if we're not allowed to modify an existing flow.
855 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
856 * because Generic Netlink treats the latter as a dump
857 * request. We also accept NLM_F_EXCL in case that bug ever
861 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
862 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
865 /* The unmasked key has to be the same for flow updates. */
866 if (!ovs_flow_cmp_unmasked_key(flow, &match))
869 /* Update actions, if present. */
871 struct sw_flow_actions *old_acts;
873 old_acts = ovsl_dereference(flow->sf_acts);
874 rcu_assign_pointer(flow->sf_acts, acts);
875 ovs_nla_free_flow_actions(old_acts);
877 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
880 if (a[OVS_FLOW_ATTR_CLEAR])
881 ovs_flow_stats_clear(flow);
886 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
888 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
889 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
893 ovs_flow_free(flow, false);
902 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
904 struct nlattr **a = info->attrs;
905 struct ovs_header *ovs_header = info->userhdr;
906 struct sw_flow_key key;
907 struct sk_buff *reply;
908 struct sw_flow *flow;
910 struct sw_flow_match match;
913 if (!a[OVS_FLOW_ATTR_KEY]) {
914 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
918 ovs_match_init(&match, &key, NULL);
919 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
924 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
930 flow = ovs_flow_tbl_lookup(&dp->table, &key);
931 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
936 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
938 err = PTR_ERR(reply);
943 return genlmsg_reply(reply, info);
949 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
951 struct nlattr **a = info->attrs;
952 struct ovs_header *ovs_header = info->userhdr;
953 struct sw_flow_key key;
954 struct sk_buff *reply;
955 struct sw_flow *flow;
957 struct sw_flow_match match;
961 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
967 if (!a[OVS_FLOW_ATTR_KEY]) {
968 err = ovs_flow_tbl_flush(&dp->table);
972 ovs_match_init(&match, &key, NULL);
973 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
977 flow = ovs_flow_tbl_lookup(&dp->table, &key);
978 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
983 reply = ovs_flow_cmd_alloc_info(flow, info);
989 ovs_flow_tbl_remove(&dp->table, flow);
991 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
992 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
995 ovs_flow_free(flow, true);
998 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1005 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1007 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1008 struct table_instance *ti;
1009 struct datapath *dp;
1012 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1018 ti = rcu_dereference(dp->table.ti);
1020 struct sw_flow *flow;
1023 bucket = cb->args[0];
1025 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1029 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1030 NETLINK_CB(cb->skb).portid,
1031 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1032 OVS_FLOW_CMD_NEW) < 0)
1035 cb->args[0] = bucket;
1042 static struct genl_ops dp_flow_genl_ops[] = {
1043 { .cmd = OVS_FLOW_CMD_NEW,
1044 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1045 .policy = flow_policy,
1046 .doit = ovs_flow_cmd_new_or_set
1048 { .cmd = OVS_FLOW_CMD_DEL,
1049 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1050 .policy = flow_policy,
1051 .doit = ovs_flow_cmd_del
1053 { .cmd = OVS_FLOW_CMD_GET,
1054 .flags = 0, /* OK for unprivileged users. */
1055 .policy = flow_policy,
1056 .doit = ovs_flow_cmd_get,
1057 .dumpit = ovs_flow_cmd_dump
1059 { .cmd = OVS_FLOW_CMD_SET,
1060 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1061 .policy = flow_policy,
1062 .doit = ovs_flow_cmd_new_or_set,
1066 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1067 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1068 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1069 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1072 static struct genl_family dp_datapath_genl_family = {
1073 .id = GENL_ID_GENERATE,
1074 .hdrsize = sizeof(struct ovs_header),
1075 .name = OVS_DATAPATH_FAMILY,
1076 .version = OVS_DATAPATH_VERSION,
1077 .maxattr = OVS_DP_ATTR_MAX,
1082 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1083 .name = OVS_DATAPATH_MCGROUP
1086 static size_t ovs_dp_cmd_msg_size(void)
1088 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1090 msgsize += nla_total_size(IFNAMSIZ);
1091 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1092 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1093 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1098 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1099 u32 portid, u32 seq, u32 flags, u8 cmd)
1101 struct ovs_header *ovs_header;
1102 struct ovs_dp_stats dp_stats;
1103 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1106 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1111 ovs_header->dp_ifindex = get_dpifindex(dp);
1114 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1117 goto nla_put_failure;
1119 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1120 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1122 goto nla_put_failure;
1124 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1125 sizeof(struct ovs_dp_megaflow_stats),
1126 &dp_megaflow_stats))
1127 goto nla_put_failure;
1129 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1130 goto nla_put_failure;
1132 return genlmsg_end(skb, ovs_header);
1135 genlmsg_cancel(skb, ovs_header);
1140 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1141 struct genl_info *info, u8 cmd)
1143 struct sk_buff *skb;
1146 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1148 return ERR_PTR(-ENOMEM);
1150 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1153 return ERR_PTR(retval);
1158 /* Called with ovs_mutex. */
1159 static struct datapath *lookup_datapath(struct net *net,
1160 struct ovs_header *ovs_header,
1161 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1163 struct datapath *dp;
1165 if (!a[OVS_DP_ATTR_NAME])
1166 dp = get_dp(net, ovs_header->dp_ifindex);
1168 struct vport *vport;
1171 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1172 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1175 return dp ? dp : ERR_PTR(-ENODEV);
1178 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1180 struct datapath *dp;
1182 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1186 WARN(dp->user_features, "Dropping previously announced user features\n");
1187 dp->user_features = 0;
1190 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1192 if (a[OVS_DP_ATTR_USER_FEATURES])
1193 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1196 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1198 struct nlattr **a = info->attrs;
1199 struct vport_parms parms;
1200 struct sk_buff *reply;
1201 struct datapath *dp;
1202 struct vport *vport;
1203 struct ovs_net *ovs_net;
1207 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1213 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1215 goto err_unlock_ovs;
1217 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1219 /* Allocate table. */
1220 err = ovs_flow_tbl_init(&dp->table);
1224 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1225 if (!dp->stats_percpu) {
1227 goto err_destroy_table;
1230 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1234 goto err_destroy_percpu;
1237 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1238 INIT_HLIST_HEAD(&dp->ports[i]);
1240 /* Set up our datapath device. */
1241 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1242 parms.type = OVS_VPORT_TYPE_INTERNAL;
1243 parms.options = NULL;
1245 parms.port_no = OVSP_LOCAL;
1246 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1248 ovs_dp_change(dp, a);
1250 vport = new_vport(&parms);
1251 if (IS_ERR(vport)) {
1252 err = PTR_ERR(vport);
1256 if (err == -EEXIST) {
1257 /* An outdated user space instance that does not understand
1258 * the concept of user_features has attempted to create a new
1259 * datapath and is likely to reuse it. Drop all user features.
1261 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1262 ovs_dp_reset_user_features(skb, info);
1265 goto err_destroy_ports_array;
1268 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1269 err = PTR_ERR(reply);
1271 goto err_destroy_local_port;
1273 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1274 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1278 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1281 err_destroy_local_port:
1282 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1283 err_destroy_ports_array:
1286 free_percpu(dp->stats_percpu);
1288 ovs_flow_tbl_destroy(&dp->table, false);
1290 release_net(ovs_dp_get_net(dp));
1298 /* Called with ovs_mutex. */
1299 static void __dp_destroy(struct datapath *dp)
1303 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1304 struct vport *vport;
1305 struct hlist_node *n;
1307 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1308 if (vport->port_no != OVSP_LOCAL)
1309 ovs_dp_detach_port(vport);
1312 list_del_rcu(&dp->list_node);
1314 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1315 * all ports in datapath are destroyed first before freeing datapath.
1317 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1319 /* RCU destroy the flow table */
1320 ovs_flow_tbl_destroy(&dp->table, true);
1322 call_rcu(&dp->rcu, destroy_dp_rcu);
1325 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1327 struct sk_buff *reply;
1328 struct datapath *dp;
1332 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1337 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1338 err = PTR_ERR(reply);
1345 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1353 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1355 struct sk_buff *reply;
1356 struct datapath *dp;
1360 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1365 ovs_dp_change(dp, info->attrs);
1367 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1368 if (IS_ERR(reply)) {
1369 err = PTR_ERR(reply);
1370 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1371 ovs_dp_datapath_multicast_group.id, err);
1377 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1385 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1387 struct sk_buff *reply;
1388 struct datapath *dp;
1392 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1398 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1399 if (IS_ERR(reply)) {
1400 err = PTR_ERR(reply);
1405 return genlmsg_reply(reply, info);
1412 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1414 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1415 struct datapath *dp;
1416 int skip = cb->args[0];
1420 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1422 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1423 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1424 OVS_DP_CMD_NEW) < 0)
1435 static struct genl_ops dp_datapath_genl_ops[] = {
1436 { .cmd = OVS_DP_CMD_NEW,
1437 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1438 .policy = datapath_policy,
1439 .doit = ovs_dp_cmd_new
1441 { .cmd = OVS_DP_CMD_DEL,
1442 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1443 .policy = datapath_policy,
1444 .doit = ovs_dp_cmd_del
1446 { .cmd = OVS_DP_CMD_GET,
1447 .flags = 0, /* OK for unprivileged users. */
1448 .policy = datapath_policy,
1449 .doit = ovs_dp_cmd_get,
1450 .dumpit = ovs_dp_cmd_dump
1452 { .cmd = OVS_DP_CMD_SET,
1453 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1454 .policy = datapath_policy,
1455 .doit = ovs_dp_cmd_set,
1459 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1460 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1461 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1462 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1463 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1464 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1465 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1468 static struct genl_family dp_vport_genl_family = {
1469 .id = GENL_ID_GENERATE,
1470 .hdrsize = sizeof(struct ovs_header),
1471 .name = OVS_VPORT_FAMILY,
1472 .version = OVS_VPORT_VERSION,
1473 .maxattr = OVS_VPORT_ATTR_MAX,
1478 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1479 .name = OVS_VPORT_MCGROUP
1482 /* Called with ovs_mutex or RCU read lock. */
1483 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1484 u32 portid, u32 seq, u32 flags, u8 cmd)
1486 struct ovs_header *ovs_header;
1487 struct ovs_vport_stats vport_stats;
1490 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1495 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1497 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1498 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1499 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1500 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1501 goto nla_put_failure;
1503 ovs_vport_get_stats(vport, &vport_stats);
1504 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1506 goto nla_put_failure;
1508 err = ovs_vport_get_options(vport, skb);
1509 if (err == -EMSGSIZE)
1512 return genlmsg_end(skb, ovs_header);
1517 genlmsg_cancel(skb, ovs_header);
1521 /* Called with ovs_mutex or RCU read lock. */
1522 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1525 struct sk_buff *skb;
1528 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1530 return ERR_PTR(-ENOMEM);
1532 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1538 /* Called with ovs_mutex or RCU read lock. */
1539 static struct vport *lookup_vport(struct net *net,
1540 struct ovs_header *ovs_header,
1541 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1543 struct datapath *dp;
1544 struct vport *vport;
1546 if (a[OVS_VPORT_ATTR_NAME]) {
1547 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1549 return ERR_PTR(-ENODEV);
1550 if (ovs_header->dp_ifindex &&
1551 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1552 return ERR_PTR(-ENODEV);
1554 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1555 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1557 if (port_no >= DP_MAX_PORTS)
1558 return ERR_PTR(-EFBIG);
1560 dp = get_dp(net, ovs_header->dp_ifindex);
1562 return ERR_PTR(-ENODEV);
1564 vport = ovs_vport_ovsl_rcu(dp, port_no);
1566 return ERR_PTR(-ENODEV);
1569 return ERR_PTR(-EINVAL);
1572 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1574 struct nlattr **a = info->attrs;
1575 struct ovs_header *ovs_header = info->userhdr;
1576 struct vport_parms parms;
1577 struct sk_buff *reply;
1578 struct vport *vport;
1579 struct datapath *dp;
1584 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1585 !a[OVS_VPORT_ATTR_UPCALL_PID])
1589 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1594 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1595 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1598 if (port_no >= DP_MAX_PORTS)
1601 vport = ovs_vport_ovsl(dp, port_no);
1606 for (port_no = 1; ; port_no++) {
1607 if (port_no >= DP_MAX_PORTS) {
1611 vport = ovs_vport_ovsl(dp, port_no);
1617 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1618 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1619 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1621 parms.port_no = port_no;
1622 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1624 vport = new_vport(&parms);
1625 err = PTR_ERR(vport);
1630 if (a[OVS_VPORT_ATTR_STATS])
1631 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1633 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1635 if (IS_ERR(reply)) {
1636 err = PTR_ERR(reply);
1637 ovs_dp_detach_port(vport);
1641 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1649 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1651 struct nlattr **a = info->attrs;
1652 struct sk_buff *reply;
1653 struct vport *vport;
1657 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1658 err = PTR_ERR(vport);
1662 if (a[OVS_VPORT_ATTR_TYPE] &&
1663 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1668 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1674 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1675 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1680 if (a[OVS_VPORT_ATTR_STATS])
1681 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1683 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1684 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1686 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1687 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1691 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1701 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1703 struct nlattr **a = info->attrs;
1704 struct sk_buff *reply;
1705 struct vport *vport;
1709 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1710 err = PTR_ERR(vport);
1714 if (vport->port_no == OVSP_LOCAL) {
1719 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1720 info->snd_seq, OVS_VPORT_CMD_DEL);
1721 err = PTR_ERR(reply);
1726 ovs_dp_detach_port(vport);
1728 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1735 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1737 struct nlattr **a = info->attrs;
1738 struct ovs_header *ovs_header = info->userhdr;
1739 struct sk_buff *reply;
1740 struct vport *vport;
1744 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1745 err = PTR_ERR(vport);
1749 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1750 info->snd_seq, OVS_VPORT_CMD_NEW);
1751 err = PTR_ERR(reply);
1757 return genlmsg_reply(reply, info);
1764 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1766 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1767 struct datapath *dp;
1768 int bucket = cb->args[0], skip = cb->args[1];
1772 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1777 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1778 struct vport *vport;
1781 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1783 ovs_vport_cmd_fill_info(vport, skb,
1784 NETLINK_CB(cb->skb).portid,
1787 OVS_VPORT_CMD_NEW) < 0)
1803 static struct genl_ops dp_vport_genl_ops[] = {
1804 { .cmd = OVS_VPORT_CMD_NEW,
1805 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1806 .policy = vport_policy,
1807 .doit = ovs_vport_cmd_new
1809 { .cmd = OVS_VPORT_CMD_DEL,
1810 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1811 .policy = vport_policy,
1812 .doit = ovs_vport_cmd_del
1814 { .cmd = OVS_VPORT_CMD_GET,
1815 .flags = 0, /* OK for unprivileged users. */
1816 .policy = vport_policy,
1817 .doit = ovs_vport_cmd_get,
1818 .dumpit = ovs_vport_cmd_dump
1820 { .cmd = OVS_VPORT_CMD_SET,
1821 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1822 .policy = vport_policy,
1823 .doit = ovs_vport_cmd_set,
1827 struct genl_family_and_ops {
1828 struct genl_family *family;
1829 struct genl_ops *ops;
1831 struct genl_multicast_group *group;
1834 static const struct genl_family_and_ops dp_genl_families[] = {
1835 { &dp_datapath_genl_family,
1836 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1837 &ovs_dp_datapath_multicast_group },
1838 { &dp_vport_genl_family,
1839 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1840 &ovs_dp_vport_multicast_group },
1841 { &dp_flow_genl_family,
1842 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1843 &ovs_dp_flow_multicast_group },
1844 { &dp_packet_genl_family,
1845 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1849 static void dp_unregister_genl(int n_families)
1853 for (i = 0; i < n_families; i++)
1854 genl_unregister_family(dp_genl_families[i].family);
1857 static int dp_register_genl(void)
1864 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1865 const struct genl_family_and_ops *f = &dp_genl_families[i];
1867 err = genl_register_family_with_ops(f->family, f->ops,
1874 err = genl_register_mc_group(f->family, f->group);
1883 dp_unregister_genl(n_registered);
1887 static int __net_init ovs_init_net(struct net *net)
1889 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1891 INIT_LIST_HEAD(&ovs_net->dps);
1892 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1896 static void __net_exit ovs_exit_net(struct net *net)
1898 struct datapath *dp, *dp_next;
1899 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1902 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1906 cancel_work_sync(&ovs_net->dp_notify_work);
1909 static struct pernet_operations ovs_net_ops = {
1910 .init = ovs_init_net,
1911 .exit = ovs_exit_net,
1913 .size = sizeof(struct ovs_net),
1916 DEFINE_COMPAT_PNET_REG_FUNC(device);
1918 static int __init dp_init(void)
1922 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1924 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1927 err = ovs_flow_init();
1931 err = ovs_vport_init();
1933 goto error_flow_exit;
1935 err = register_pernet_device(&ovs_net_ops);
1937 goto error_vport_exit;
1939 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1941 goto error_netns_exit;
1943 err = dp_register_genl();
1945 goto error_unreg_notifier;
1949 error_unreg_notifier:
1950 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1952 unregister_pernet_device(&ovs_net_ops);
1961 static void dp_cleanup(void)
1963 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1964 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1965 unregister_pernet_device(&ovs_net_ops);
1971 module_init(dp_init);
1972 module_exit(dp_cleanup);
1974 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1975 MODULE_LICENSE("GPL");
1976 MODULE_VERSION(VERSION);