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 static void ovs_notify(struct genl_family *family,
66 struct sk_buff *skb, struct genl_info *info)
68 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
69 0, info->nlhdr, GFP_KERNEL);
75 * All writes e.g. Writes to device state (add/remove datapath, port, set
76 * operations on vports, etc.), Writes to other state (flow table
77 * modifications, set miscellaneous datapath parameters, etc.) are protected
80 * Reads are protected by RCU.
82 * There are a few special cases (mostly stats) that have their own
83 * synchronization but they nest under all of above and don't interact with
86 * The RTNL lock nests inside ovs_mutex.
89 static DEFINE_MUTEX(ovs_mutex);
93 mutex_lock(&ovs_mutex);
98 mutex_unlock(&ovs_mutex);
101 #ifdef CONFIG_LOCKDEP
102 int lockdep_ovsl_is_held(void)
105 return lockdep_is_held(&ovs_mutex);
111 static struct vport *new_vport(const struct vport_parms *);
112 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
113 const struct dp_upcall_info *);
114 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
115 const struct dp_upcall_info *);
117 /* Must be called with rcu_read_lock or ovs_mutex. */
118 static struct datapath *get_dp(struct net *net, int dp_ifindex)
120 struct datapath *dp = NULL;
121 struct net_device *dev;
124 dev = dev_get_by_index_rcu(net, dp_ifindex);
126 struct vport *vport = ovs_internal_dev_get_vport(dev);
135 /* Must be called with rcu_read_lock or ovs_mutex. */
136 static const char *ovs_dp_name(const struct datapath *dp)
138 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
139 return vport->ops->get_name(vport);
142 static int get_dpifindex(struct datapath *dp)
149 local = ovs_vport_rcu(dp, OVSP_LOCAL);
151 ifindex = netdev_vport_priv(local)->dev->ifindex;
160 static void destroy_dp_rcu(struct rcu_head *rcu)
162 struct datapath *dp = container_of(rcu, struct datapath, rcu);
164 free_percpu(dp->stats_percpu);
165 release_net(ovs_dp_get_net(dp));
170 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
173 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
179 struct hlist_head *head;
181 head = vport_hash_bucket(dp, port_no);
182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
183 if (vport->port_no == port_no)
189 /* Called with ovs_mutex. */
190 static struct vport *new_vport(const struct vport_parms *parms)
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
199 hlist_add_head_rcu(&vport->dp_hash_node, head);
204 void ovs_dp_detach_port(struct vport *p)
208 /* First drop references to device. */
209 hlist_del_rcu(&p->dp_hash_node);
211 /* Then destroy it. */
215 /* Must be called with rcu_read_lock. */
216 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
218 struct datapath *dp = p->dp;
219 struct sw_flow *flow;
220 struct dp_stats_percpu *stats;
221 struct sw_flow_key key;
226 stats = this_cpu_ptr(dp->stats_percpu);
228 /* Extract flow from 'skb' into 'key'. */
229 error = ovs_flow_extract(skb, p->port_no, &key);
230 if (unlikely(error)) {
236 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
237 if (unlikely(!flow)) {
238 struct dp_upcall_info upcall;
240 upcall.cmd = OVS_PACKET_CMD_MISS;
242 upcall.userdata = NULL;
243 upcall.portid = p->upcall_portid;
244 ovs_dp_upcall(dp, skb, &upcall);
246 stats_counter = &stats->n_missed;
250 OVS_CB(skb)->flow = flow;
251 OVS_CB(skb)->pkt_key = &key;
253 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
254 ovs_execute_actions(dp, skb);
255 stats_counter = &stats->n_hit;
258 /* Update datapath statistics. */
259 u64_stats_update_begin(&stats->syncp);
261 stats->n_mask_hit += n_mask_hit;
262 u64_stats_update_end(&stats->syncp);
265 static struct genl_family dp_packet_genl_family = {
266 .id = GENL_ID_GENERATE,
267 .hdrsize = sizeof(struct ovs_header),
268 .name = OVS_PACKET_FAMILY,
269 .version = OVS_PACKET_VERSION,
270 .maxattr = OVS_PACKET_ATTR_MAX,
272 .parallel_ops = true,
275 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
276 const struct dp_upcall_info *upcall_info)
278 struct dp_stats_percpu *stats;
281 if (upcall_info->portid == 0) {
286 if (!skb_is_gso(skb))
287 err = queue_userspace_packet(dp, skb, upcall_info);
289 err = queue_gso_packets(dp, skb, upcall_info);
296 stats = this_cpu_ptr(dp->stats_percpu);
298 u64_stats_update_begin(&stats->syncp);
300 u64_stats_update_end(&stats->syncp);
305 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
306 const struct dp_upcall_info *upcall_info)
308 unsigned short gso_type = skb_shinfo(skb)->gso_type;
309 struct dp_upcall_info later_info;
310 struct sw_flow_key later_key;
311 struct sk_buff *segs, *nskb;
314 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
316 return PTR_ERR(segs);
318 /* Queue all of the segments. */
321 err = queue_userspace_packet(dp, skb, upcall_info);
325 if (skb == segs && gso_type & SKB_GSO_UDP) {
326 /* The initial flow key extracted by ovs_flow_extract()
327 * in this case is for a first fragment, so we need to
328 * properly mark later fragments.
330 later_key = *upcall_info->key;
331 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
333 later_info = *upcall_info;
334 later_info.key = &later_key;
335 upcall_info = &later_info;
337 } while ((skb = skb->next));
339 /* Free all of the segments. */
347 } while ((skb = nskb));
351 static size_t key_attr_size(void)
353 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
354 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
355 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
356 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
357 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
358 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
359 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
360 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
361 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
362 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
363 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
364 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
365 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
366 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
367 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
368 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
369 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
370 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
371 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
374 static size_t upcall_msg_size(const struct nlattr *userdata,
377 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
378 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
379 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
381 /* OVS_PACKET_ATTR_USERDATA */
383 size += NLA_ALIGN(userdata->nla_len);
388 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
389 const struct dp_upcall_info *upcall_info)
391 struct ovs_header *upcall;
392 struct sk_buff *nskb = NULL;
393 struct sk_buff *user_skb; /* to be queued to userspace */
395 struct genl_info info = {
396 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
397 .snd_portid = upcall_info->portid,
403 dp_ifindex = get_dpifindex(dp);
407 if (vlan_tx_tag_present(skb)) {
408 nskb = skb_clone(skb, GFP_ATOMIC);
412 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
420 if (nla_attr_size(skb->len) > USHRT_MAX) {
425 /* Complete checksum if needed */
426 if (skb->ip_summed == CHECKSUM_PARTIAL &&
427 (err = skb_checksum_help(skb)))
430 /* Older versions of OVS user space enforce alignment of the last
431 * Netlink attribute to NLA_ALIGNTO which would require extensive
432 * padding logic. Only perform zerocopy if padding is not required.
434 if (dp->user_features & OVS_DP_F_UNALIGNED)
435 hlen = skb_zerocopy_headlen(skb);
439 len = upcall_msg_size(upcall_info->userdata, hlen);
440 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
446 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
447 0, upcall_info->cmd);
448 upcall->dp_ifindex = dp_ifindex;
450 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
451 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
452 nla_nest_end(user_skb, nla);
454 if (upcall_info->userdata)
455 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
456 nla_len(upcall_info->userdata),
457 nla_data(upcall_info->userdata));
459 /* Only reserve room for attribute header, packet data is added
460 * in skb_zerocopy() */
461 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
465 nla->nla_len = nla_attr_size(skb->len);
467 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
471 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
472 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
473 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
476 memset(skb_put(user_skb, plen), 0, plen);
479 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
481 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
489 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
491 struct ovs_header *ovs_header = info->userhdr;
492 struct nlattr **a = info->attrs;
493 struct sw_flow_actions *acts;
494 struct sk_buff *packet;
495 struct sw_flow *flow;
502 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
503 !a[OVS_PACKET_ATTR_ACTIONS])
506 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
507 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
511 skb_reserve(packet, NET_IP_ALIGN);
513 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
515 skb_reset_mac_header(packet);
516 eth = eth_hdr(packet);
518 /* Normally, setting the skb 'protocol' field would be handled by a
519 * call to eth_type_trans(), but it assumes there's a sending
520 * device, which we may not have. */
521 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
522 packet->protocol = eth->h_proto;
524 packet->protocol = htons(ETH_P_802_2);
526 /* Build an sw_flow for sending this packet. */
527 flow = ovs_flow_alloc(false);
532 err = ovs_flow_extract(packet, -1, &flow->key);
536 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
539 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
544 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
545 &flow->key, 0, &acts);
546 rcu_assign_pointer(flow->sf_acts, acts);
550 OVS_CB(packet)->flow = flow;
551 OVS_CB(packet)->pkt_key = &flow->key;
552 packet->priority = flow->key.phy.priority;
553 packet->mark = flow->key.phy.skb_mark;
556 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
562 err = ovs_execute_actions(dp, packet);
566 ovs_flow_free(flow, false);
572 ovs_flow_free(flow, false);
579 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
580 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
581 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
582 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
585 static const struct genl_ops dp_packet_genl_ops[] = {
586 { .cmd = OVS_PACKET_CMD_EXECUTE,
587 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
588 .policy = packet_policy,
589 .doit = ovs_packet_cmd_execute
593 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
594 struct ovs_dp_megaflow_stats *mega_stats)
598 memset(mega_stats, 0, sizeof(*mega_stats));
600 stats->n_flows = ovs_flow_tbl_count(&dp->table);
601 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
603 stats->n_hit = stats->n_missed = stats->n_lost = 0;
605 for_each_possible_cpu(i) {
606 const struct dp_stats_percpu *percpu_stats;
607 struct dp_stats_percpu local_stats;
610 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
613 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
614 local_stats = *percpu_stats;
615 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
617 stats->n_hit += local_stats.n_hit;
618 stats->n_missed += local_stats.n_missed;
619 stats->n_lost += local_stats.n_lost;
620 mega_stats->n_mask_hit += local_stats.n_mask_hit;
624 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
625 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
626 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
627 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
630 static struct genl_family dp_flow_genl_family = {
631 .id = GENL_ID_GENERATE,
632 .hdrsize = sizeof(struct ovs_header),
633 .name = OVS_FLOW_FAMILY,
634 .version = OVS_FLOW_VERSION,
635 .maxattr = OVS_FLOW_ATTR_MAX,
637 .parallel_ops = true,
640 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
641 .name = OVS_FLOW_MCGROUP
644 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
646 return NLMSG_ALIGN(sizeof(struct ovs_header))
647 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
648 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
649 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
650 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
651 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
652 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
655 /* Called with ovs_mutex. */
656 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
657 struct sk_buff *skb, u32 portid,
658 u32 seq, u32 flags, u8 cmd)
660 const int skb_orig_len = skb->len;
661 struct nlattr *start;
662 struct ovs_flow_stats stats;
665 struct ovs_header *ovs_header;
669 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
673 ovs_header->dp_ifindex = get_dpifindex(dp);
676 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
678 goto nla_put_failure;
680 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
683 nla_nest_end(skb, nla);
685 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
687 goto nla_put_failure;
689 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
693 nla_nest_end(skb, nla);
695 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
697 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
698 goto nla_put_failure;
700 if (stats.n_packets &&
701 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
702 goto nla_put_failure;
704 if ((u8)ntohs(tcp_flags) &&
705 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
706 goto nla_put_failure;
708 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
709 * this is the first flow to be dumped into 'skb'. This is unusual for
710 * Netlink but individual action lists can be longer than
711 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
712 * The userspace caller can always fetch the actions separately if it
713 * really wants them. (Most userspace callers in fact don't care.)
715 * This can only fail for dump operations because the skb is always
716 * properly sized for single flows.
718 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
720 const struct sw_flow_actions *sf_acts;
722 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
724 err = ovs_nla_put_actions(sf_acts->actions,
725 sf_acts->actions_len, skb);
727 nla_nest_end(skb, start);
732 nla_nest_cancel(skb, start);
734 } else if (skb_orig_len)
735 goto nla_put_failure;
737 return genlmsg_end(skb, ovs_header);
742 genlmsg_cancel(skb, ovs_header);
746 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
747 struct genl_info *info)
751 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
753 return genlmsg_new_unicast(len, info, GFP_KERNEL);
756 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
758 struct genl_info *info,
764 skb = ovs_flow_cmd_alloc_info(flow, info);
766 return ERR_PTR(-ENOMEM);
768 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
769 info->snd_seq, 0, cmd);
774 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
776 struct nlattr **a = info->attrs;
777 struct ovs_header *ovs_header = info->userhdr;
778 struct sw_flow_key key, masked_key;
779 struct sw_flow *flow = NULL;
780 struct sw_flow_mask mask;
781 struct sk_buff *reply;
783 struct sw_flow_actions *acts = NULL;
784 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, &exact_5tuple,
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) {
819 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
824 /* Check if this is a duplicate flow */
825 flow = ovs_flow_tbl_lookup(&dp->table, &key);
827 /* Bail out if we're not allowed to create a new flow. */
829 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
833 flow = ovs_flow_alloc(!exact_5tuple);
835 error = PTR_ERR(flow);
839 flow->key = masked_key;
840 flow->unmasked_key = key;
841 rcu_assign_pointer(flow->sf_acts, acts);
843 /* Put flow in bucket. */
844 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
850 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
852 /* We found a matching flow. */
853 struct sw_flow_actions *old_acts;
855 /* Bail out if we're not allowed to modify an existing flow.
856 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
857 * because Generic Netlink treats the latter as a dump
858 * request. We also accept NLM_F_EXCL in case that bug ever
862 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
863 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
866 /* The unmasked key has to be the same for flow updates. */
867 if (!ovs_flow_cmp_unmasked_key(flow, &match))
870 /* Update actions. */
871 old_acts = ovsl_dereference(flow->sf_acts);
872 rcu_assign_pointer(flow->sf_acts, acts);
873 ovs_nla_free_flow_actions(old_acts);
875 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
878 if (a[OVS_FLOW_ATTR_CLEAR])
879 ovs_flow_stats_clear(flow);
884 ovs_notify(&dp_flow_genl_family, reply, info);
886 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
891 ovs_flow_free(flow, false);
900 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
902 struct nlattr **a = info->attrs;
903 struct ovs_header *ovs_header = info->userhdr;
904 struct sw_flow_key key;
905 struct sk_buff *reply;
906 struct sw_flow *flow;
908 struct sw_flow_match match;
911 if (!a[OVS_FLOW_ATTR_KEY]) {
912 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
916 ovs_match_init(&match, &key, NULL);
917 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
922 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
928 flow = ovs_flow_tbl_lookup(&dp->table, &key);
929 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
934 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
936 err = PTR_ERR(reply);
941 return genlmsg_reply(reply, info);
947 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
949 struct nlattr **a = info->attrs;
950 struct ovs_header *ovs_header = info->userhdr;
951 struct sw_flow_key key;
952 struct sk_buff *reply;
953 struct sw_flow *flow;
955 struct sw_flow_match match;
959 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
965 if (!a[OVS_FLOW_ATTR_KEY]) {
966 err = ovs_flow_tbl_flush(&dp->table);
970 ovs_match_init(&match, &key, NULL);
971 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
975 flow = ovs_flow_tbl_lookup(&dp->table, &key);
976 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
981 reply = ovs_flow_cmd_alloc_info(flow, info);
987 ovs_flow_tbl_remove(&dp->table, flow);
989 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
990 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
993 ovs_flow_free(flow, true);
996 ovs_notify(&dp_flow_genl_family, reply, info);
1003 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1005 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1006 struct table_instance *ti;
1007 struct datapath *dp;
1010 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1016 ti = rcu_dereference(dp->table.ti);
1018 struct sw_flow *flow;
1021 bucket = cb->args[0];
1023 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1027 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1028 NETLINK_CB(cb->skb).portid,
1029 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1030 OVS_FLOW_CMD_NEW) < 0)
1033 cb->args[0] = bucket;
1040 static const struct genl_ops dp_flow_genl_ops[] = {
1041 { .cmd = OVS_FLOW_CMD_NEW,
1042 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1043 .policy = flow_policy,
1044 .doit = ovs_flow_cmd_new_or_set
1046 { .cmd = OVS_FLOW_CMD_DEL,
1047 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1048 .policy = flow_policy,
1049 .doit = ovs_flow_cmd_del
1051 { .cmd = OVS_FLOW_CMD_GET,
1052 .flags = 0, /* OK for unprivileged users. */
1053 .policy = flow_policy,
1054 .doit = ovs_flow_cmd_get,
1055 .dumpit = ovs_flow_cmd_dump
1057 { .cmd = OVS_FLOW_CMD_SET,
1058 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1059 .policy = flow_policy,
1060 .doit = ovs_flow_cmd_new_or_set,
1064 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1065 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1066 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1067 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1070 static struct genl_family dp_datapath_genl_family = {
1071 .id = GENL_ID_GENERATE,
1072 .hdrsize = sizeof(struct ovs_header),
1073 .name = OVS_DATAPATH_FAMILY,
1074 .version = OVS_DATAPATH_VERSION,
1075 .maxattr = OVS_DP_ATTR_MAX,
1077 .parallel_ops = true,
1080 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1081 .name = OVS_DATAPATH_MCGROUP
1084 static size_t ovs_dp_cmd_msg_size(void)
1086 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1088 msgsize += nla_total_size(IFNAMSIZ);
1089 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1090 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1091 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1096 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1097 u32 portid, u32 seq, u32 flags, u8 cmd)
1099 struct ovs_header *ovs_header;
1100 struct ovs_dp_stats dp_stats;
1101 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1104 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1109 ovs_header->dp_ifindex = get_dpifindex(dp);
1112 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1115 goto nla_put_failure;
1117 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1118 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1120 goto nla_put_failure;
1122 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1123 sizeof(struct ovs_dp_megaflow_stats),
1124 &dp_megaflow_stats))
1125 goto nla_put_failure;
1127 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1128 goto nla_put_failure;
1130 return genlmsg_end(skb, ovs_header);
1133 genlmsg_cancel(skb, ovs_header);
1138 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1139 struct genl_info *info, u8 cmd)
1141 struct sk_buff *skb;
1144 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1146 return ERR_PTR(-ENOMEM);
1148 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1151 return ERR_PTR(retval);
1156 /* Called with ovs_mutex. */
1157 static struct datapath *lookup_datapath(struct net *net,
1158 struct ovs_header *ovs_header,
1159 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1161 struct datapath *dp;
1163 if (!a[OVS_DP_ATTR_NAME])
1164 dp = get_dp(net, ovs_header->dp_ifindex);
1166 struct vport *vport;
1169 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1170 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1173 return dp ? dp : ERR_PTR(-ENODEV);
1176 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1178 struct datapath *dp;
1180 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1184 WARN(dp->user_features, "Dropping previously announced user features\n");
1185 dp->user_features = 0;
1188 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1190 if (a[OVS_DP_ATTR_USER_FEATURES])
1191 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1194 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1196 struct nlattr **a = info->attrs;
1197 struct vport_parms parms;
1198 struct sk_buff *reply;
1199 struct datapath *dp;
1200 struct vport *vport;
1201 struct ovs_net *ovs_net;
1205 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1211 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1213 goto err_unlock_ovs;
1215 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1217 /* Allocate table. */
1218 err = ovs_flow_tbl_init(&dp->table);
1222 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1223 if (!dp->stats_percpu) {
1225 goto err_destroy_table;
1228 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1232 goto err_destroy_percpu;
1235 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1236 INIT_HLIST_HEAD(&dp->ports[i]);
1238 /* Set up our datapath device. */
1239 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1240 parms.type = OVS_VPORT_TYPE_INTERNAL;
1241 parms.options = NULL;
1243 parms.port_no = OVSP_LOCAL;
1244 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1246 ovs_dp_change(dp, a);
1248 vport = new_vport(&parms);
1249 if (IS_ERR(vport)) {
1250 err = PTR_ERR(vport);
1254 if (err == -EEXIST) {
1255 /* An outdated user space instance that does not understand
1256 * the concept of user_features has attempted to create a new
1257 * datapath and is likely to reuse it. Drop all user features.
1259 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1260 ovs_dp_reset_user_features(skb, info);
1263 goto err_destroy_ports_array;
1266 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1267 err = PTR_ERR(reply);
1269 goto err_destroy_local_port;
1271 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1272 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1276 ovs_notify(&dp_datapath_genl_family, reply, info);
1279 err_destroy_local_port:
1280 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1281 err_destroy_ports_array:
1284 free_percpu(dp->stats_percpu);
1286 ovs_flow_tbl_destroy(&dp->table, false);
1288 release_net(ovs_dp_get_net(dp));
1296 /* Called with ovs_mutex. */
1297 static void __dp_destroy(struct datapath *dp)
1301 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1302 struct vport *vport;
1303 struct hlist_node *n;
1305 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1306 if (vport->port_no != OVSP_LOCAL)
1307 ovs_dp_detach_port(vport);
1310 list_del_rcu(&dp->list_node);
1312 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1313 * all ports in datapath are destroyed first before freeing datapath.
1315 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1317 /* RCU destroy the flow table */
1318 ovs_flow_tbl_destroy(&dp->table, true);
1320 call_rcu(&dp->rcu, destroy_dp_rcu);
1323 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1325 struct sk_buff *reply;
1326 struct datapath *dp;
1330 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1335 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1336 err = PTR_ERR(reply);
1343 ovs_notify(&dp_datapath_genl_family, reply, info);
1351 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1353 struct sk_buff *reply;
1354 struct datapath *dp;
1358 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1363 ovs_dp_change(dp, info->attrs);
1365 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1366 if (IS_ERR(reply)) {
1367 err = PTR_ERR(reply);
1368 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1375 ovs_notify(&dp_datapath_genl_family, reply, info);
1383 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1385 struct sk_buff *reply;
1386 struct datapath *dp;
1390 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1396 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1397 if (IS_ERR(reply)) {
1398 err = PTR_ERR(reply);
1403 return genlmsg_reply(reply, info);
1410 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1412 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1413 struct datapath *dp;
1414 int skip = cb->args[0];
1418 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1420 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1421 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1422 OVS_DP_CMD_NEW) < 0)
1433 static const struct genl_ops dp_datapath_genl_ops[] = {
1434 { .cmd = OVS_DP_CMD_NEW,
1435 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1436 .policy = datapath_policy,
1437 .doit = ovs_dp_cmd_new
1439 { .cmd = OVS_DP_CMD_DEL,
1440 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1441 .policy = datapath_policy,
1442 .doit = ovs_dp_cmd_del
1444 { .cmd = OVS_DP_CMD_GET,
1445 .flags = 0, /* OK for unprivileged users. */
1446 .policy = datapath_policy,
1447 .doit = ovs_dp_cmd_get,
1448 .dumpit = ovs_dp_cmd_dump
1450 { .cmd = OVS_DP_CMD_SET,
1451 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1452 .policy = datapath_policy,
1453 .doit = ovs_dp_cmd_set,
1457 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1458 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1459 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1460 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1461 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1462 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1463 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1466 struct genl_family dp_vport_genl_family = {
1467 .id = GENL_ID_GENERATE,
1468 .hdrsize = sizeof(struct ovs_header),
1469 .name = OVS_VPORT_FAMILY,
1470 .version = OVS_VPORT_VERSION,
1471 .maxattr = OVS_VPORT_ATTR_MAX,
1473 .parallel_ops = true,
1476 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1477 .name = OVS_VPORT_MCGROUP
1480 /* Called with ovs_mutex or RCU read lock. */
1481 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1482 u32 portid, u32 seq, u32 flags, u8 cmd)
1484 struct ovs_header *ovs_header;
1485 struct ovs_vport_stats vport_stats;
1488 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1493 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1495 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1496 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1497 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1498 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1499 goto nla_put_failure;
1501 ovs_vport_get_stats(vport, &vport_stats);
1502 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1504 goto nla_put_failure;
1506 err = ovs_vport_get_options(vport, skb);
1507 if (err == -EMSGSIZE)
1510 return genlmsg_end(skb, ovs_header);
1515 genlmsg_cancel(skb, ovs_header);
1519 /* Called with ovs_mutex or RCU read lock. */
1520 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1523 struct sk_buff *skb;
1526 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1528 return ERR_PTR(-ENOMEM);
1530 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1536 /* Called with ovs_mutex or RCU read lock. */
1537 static struct vport *lookup_vport(struct net *net,
1538 struct ovs_header *ovs_header,
1539 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1541 struct datapath *dp;
1542 struct vport *vport;
1544 if (a[OVS_VPORT_ATTR_NAME]) {
1545 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1547 return ERR_PTR(-ENODEV);
1548 if (ovs_header->dp_ifindex &&
1549 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1550 return ERR_PTR(-ENODEV);
1552 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1553 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1555 if (port_no >= DP_MAX_PORTS)
1556 return ERR_PTR(-EFBIG);
1558 dp = get_dp(net, ovs_header->dp_ifindex);
1560 return ERR_PTR(-ENODEV);
1562 vport = ovs_vport_ovsl_rcu(dp, port_no);
1564 return ERR_PTR(-ENODEV);
1567 return ERR_PTR(-EINVAL);
1570 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1572 struct nlattr **a = info->attrs;
1573 struct ovs_header *ovs_header = info->userhdr;
1574 struct vport_parms parms;
1575 struct sk_buff *reply;
1576 struct vport *vport;
1577 struct datapath *dp;
1582 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1583 !a[OVS_VPORT_ATTR_UPCALL_PID])
1587 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1592 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1593 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1596 if (port_no >= DP_MAX_PORTS)
1599 vport = ovs_vport_ovsl(dp, port_no);
1604 for (port_no = 1; ; port_no++) {
1605 if (port_no >= DP_MAX_PORTS) {
1609 vport = ovs_vport_ovsl(dp, port_no);
1615 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1616 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1617 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1619 parms.port_no = port_no;
1620 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1622 vport = new_vport(&parms);
1623 err = PTR_ERR(vport);
1628 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1630 if (IS_ERR(reply)) {
1631 err = PTR_ERR(reply);
1632 ovs_dp_detach_port(vport);
1636 ovs_notify(&dp_vport_genl_family, reply, info);
1644 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1646 struct nlattr **a = info->attrs;
1647 struct sk_buff *reply;
1648 struct vport *vport;
1652 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1653 err = PTR_ERR(vport);
1657 if (a[OVS_VPORT_ATTR_TYPE] &&
1658 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1663 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1669 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1670 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1675 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1676 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1678 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1679 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1683 ovs_notify(&dp_vport_genl_family, reply, info);
1693 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1695 struct nlattr **a = info->attrs;
1696 struct sk_buff *reply;
1697 struct vport *vport;
1701 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1702 err = PTR_ERR(vport);
1706 if (vport->port_no == OVSP_LOCAL) {
1711 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1712 info->snd_seq, OVS_VPORT_CMD_DEL);
1713 err = PTR_ERR(reply);
1718 ovs_dp_detach_port(vport);
1720 ovs_notify(&dp_vport_genl_family, reply, info);
1727 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1729 struct nlattr **a = info->attrs;
1730 struct ovs_header *ovs_header = info->userhdr;
1731 struct sk_buff *reply;
1732 struct vport *vport;
1736 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1737 err = PTR_ERR(vport);
1741 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1742 info->snd_seq, OVS_VPORT_CMD_NEW);
1743 err = PTR_ERR(reply);
1749 return genlmsg_reply(reply, info);
1756 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1758 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1759 struct datapath *dp;
1760 int bucket = cb->args[0], skip = cb->args[1];
1764 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1769 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1770 struct vport *vport;
1773 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1775 ovs_vport_cmd_fill_info(vport, skb,
1776 NETLINK_CB(cb->skb).portid,
1779 OVS_VPORT_CMD_NEW) < 0)
1795 static const struct genl_ops dp_vport_genl_ops[] = {
1796 { .cmd = OVS_VPORT_CMD_NEW,
1797 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1798 .policy = vport_policy,
1799 .doit = ovs_vport_cmd_new
1801 { .cmd = OVS_VPORT_CMD_DEL,
1802 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1803 .policy = vport_policy,
1804 .doit = ovs_vport_cmd_del
1806 { .cmd = OVS_VPORT_CMD_GET,
1807 .flags = 0, /* OK for unprivileged users. */
1808 .policy = vport_policy,
1809 .doit = ovs_vport_cmd_get,
1810 .dumpit = ovs_vport_cmd_dump
1812 { .cmd = OVS_VPORT_CMD_SET,
1813 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1814 .policy = vport_policy,
1815 .doit = ovs_vport_cmd_set,
1819 struct genl_family_and_ops {
1820 struct genl_family *family;
1821 const struct genl_ops *ops;
1823 const struct genl_multicast_group *group;
1826 static const struct genl_family_and_ops dp_genl_families[] = {
1827 { &dp_datapath_genl_family,
1828 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1829 &ovs_dp_datapath_multicast_group },
1830 { &dp_vport_genl_family,
1831 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1832 &ovs_dp_vport_multicast_group },
1833 { &dp_flow_genl_family,
1834 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1835 &ovs_dp_flow_multicast_group },
1836 { &dp_packet_genl_family,
1837 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1841 static void dp_unregister_genl(int n_families)
1845 for (i = 0; i < n_families; i++)
1846 genl_unregister_family(dp_genl_families[i].family);
1849 static int dp_register_genl(void)
1856 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1857 const struct genl_family_and_ops *f = &dp_genl_families[i];
1859 f->family->ops = f->ops;
1860 f->family->n_ops = f->n_ops;
1861 f->family->mcgrps = f->group;
1862 f->family->n_mcgrps = f->group ? 1 : 0;
1863 err = genl_register_family(f->family);
1872 dp_unregister_genl(n_registered);
1876 static int __net_init ovs_init_net(struct net *net)
1878 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1880 INIT_LIST_HEAD(&ovs_net->dps);
1881 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1885 static void __net_exit ovs_exit_net(struct net *net)
1887 struct datapath *dp, *dp_next;
1888 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1891 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1895 cancel_work_sync(&ovs_net->dp_notify_work);
1898 static struct pernet_operations ovs_net_ops = {
1899 .init = ovs_init_net,
1900 .exit = ovs_exit_net,
1902 .size = sizeof(struct ovs_net),
1905 static int __init dp_init(void)
1909 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1911 pr_info("Open vSwitch switching datapath\n");
1913 err = ovs_flow_init();
1917 err = ovs_vport_init();
1919 goto error_flow_exit;
1921 err = register_pernet_device(&ovs_net_ops);
1923 goto error_vport_exit;
1925 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1927 goto error_netns_exit;
1929 err = dp_register_genl();
1931 goto error_unreg_notifier;
1935 error_unreg_notifier:
1936 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1938 unregister_pernet_device(&ovs_net_ops);
1947 static void dp_cleanup(void)
1949 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1950 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1951 unregister_pernet_device(&ovs_net_ops);
1957 module_init(dp_init);
1958 module_exit(dp_cleanup);
1960 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1961 MODULE_LICENSE("GPL");