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_netlink.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
62 int ovs_net_id __read_mostly;
64 static void ovs_notify(struct genl_family *family,
65 struct sk_buff *skb, struct genl_info *info)
67 genl_notify(family, skb, genl_info_net(info), info->snd_portid,
68 0, info->nlhdr, GFP_KERNEL);
74 * All writes e.g. Writes to device state (add/remove datapath, port, set
75 * operations on vports, etc.), Writes to other state (flow table
76 * modifications, set miscellaneous datapath parameters, etc.) are protected
79 * Reads are protected by RCU.
81 * There are a few special cases (mostly stats) that have their own
82 * synchronization but they nest under all of above and don't interact with
85 * The RTNL lock nests inside ovs_mutex.
88 static DEFINE_MUTEX(ovs_mutex);
92 mutex_lock(&ovs_mutex);
97 mutex_unlock(&ovs_mutex);
100 #ifdef CONFIG_LOCKDEP
101 int lockdep_ovsl_is_held(void)
104 return lockdep_is_held(&ovs_mutex);
110 static struct vport *new_vport(const struct vport_parms *);
111 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
112 const struct dp_upcall_info *);
113 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
114 const struct dp_upcall_info *);
116 /* Must be called with rcu_read_lock or ovs_mutex. */
117 static struct datapath *get_dp(struct net *net, int dp_ifindex)
119 struct datapath *dp = NULL;
120 struct net_device *dev;
123 dev = dev_get_by_index_rcu(net, dp_ifindex);
125 struct vport *vport = ovs_internal_dev_get_vport(dev);
134 /* Must be called with rcu_read_lock or ovs_mutex. */
135 static const char *ovs_dp_name(const struct datapath *dp)
137 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
138 return vport->ops->get_name(vport);
141 static int get_dpifindex(struct datapath *dp)
148 local = ovs_vport_rcu(dp, OVSP_LOCAL);
150 ifindex = netdev_vport_priv(local)->dev->ifindex;
159 static void destroy_dp_rcu(struct rcu_head *rcu)
161 struct datapath *dp = container_of(rcu, struct datapath, rcu);
163 ovs_flow_tbl_destroy(&dp->table);
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->sync);
261 stats->n_mask_hit += n_mask_hit;
262 u64_stats_update_end(&stats->sync);
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->sync);
300 u64_stats_update_end(&stats->sync);
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 skb_zerocopy(user_skb, skb, skb->len, hlen);
469 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
471 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
477 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
479 struct ovs_header *ovs_header = info->userhdr;
480 struct nlattr **a = info->attrs;
481 struct sw_flow_actions *acts;
482 struct sk_buff *packet;
483 struct sw_flow *flow;
490 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
491 !a[OVS_PACKET_ATTR_ACTIONS])
494 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
495 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
499 skb_reserve(packet, NET_IP_ALIGN);
501 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
503 skb_reset_mac_header(packet);
504 eth = eth_hdr(packet);
506 /* Normally, setting the skb 'protocol' field would be handled by a
507 * call to eth_type_trans(), but it assumes there's a sending
508 * device, which we may not have. */
509 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
510 packet->protocol = eth->h_proto;
512 packet->protocol = htons(ETH_P_802_2);
514 /* Build an sw_flow for sending this packet. */
515 flow = ovs_flow_alloc(false);
520 err = ovs_flow_extract(packet, -1, &flow->key);
524 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
527 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
532 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
533 &flow->key, 0, &acts);
534 rcu_assign_pointer(flow->sf_acts, acts);
538 OVS_CB(packet)->flow = flow;
539 OVS_CB(packet)->pkt_key = &flow->key;
540 packet->priority = flow->key.phy.priority;
541 packet->mark = flow->key.phy.skb_mark;
544 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
550 err = ovs_execute_actions(dp, packet);
554 ovs_flow_free(flow, false);
560 ovs_flow_free(flow, false);
567 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
568 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
569 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
570 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
573 static const struct genl_ops dp_packet_genl_ops[] = {
574 { .cmd = OVS_PACKET_CMD_EXECUTE,
575 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
576 .policy = packet_policy,
577 .doit = ovs_packet_cmd_execute
581 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
582 struct ovs_dp_megaflow_stats *mega_stats)
586 memset(mega_stats, 0, sizeof(*mega_stats));
588 stats->n_flows = ovs_flow_tbl_count(&dp->table);
589 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
591 stats->n_hit = stats->n_missed = stats->n_lost = 0;
593 for_each_possible_cpu(i) {
594 const struct dp_stats_percpu *percpu_stats;
595 struct dp_stats_percpu local_stats;
598 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
601 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
602 local_stats = *percpu_stats;
603 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
605 stats->n_hit += local_stats.n_hit;
606 stats->n_missed += local_stats.n_missed;
607 stats->n_lost += local_stats.n_lost;
608 mega_stats->n_mask_hit += local_stats.n_mask_hit;
612 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
613 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
614 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
615 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
618 static struct genl_family dp_flow_genl_family = {
619 .id = GENL_ID_GENERATE,
620 .hdrsize = sizeof(struct ovs_header),
621 .name = OVS_FLOW_FAMILY,
622 .version = OVS_FLOW_VERSION,
623 .maxattr = OVS_FLOW_ATTR_MAX,
625 .parallel_ops = true,
628 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
629 .name = OVS_FLOW_MCGROUP
632 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
634 return NLMSG_ALIGN(sizeof(struct ovs_header))
635 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
636 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
637 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
638 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
639 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
640 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
643 /* Called with ovs_mutex. */
644 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
645 struct sk_buff *skb, u32 portid,
646 u32 seq, u32 flags, u8 cmd)
648 const int skb_orig_len = skb->len;
649 struct nlattr *start;
650 struct ovs_flow_stats stats;
653 struct ovs_header *ovs_header;
657 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
661 ovs_header->dp_ifindex = get_dpifindex(dp);
664 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
666 goto nla_put_failure;
668 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
671 nla_nest_end(skb, nla);
673 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
675 goto nla_put_failure;
677 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
681 nla_nest_end(skb, nla);
683 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
685 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
686 goto nla_put_failure;
688 if (stats.n_packets &&
689 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
690 goto nla_put_failure;
692 if ((u8)ntohs(tcp_flags) &&
693 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
694 goto nla_put_failure;
696 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
697 * this is the first flow to be dumped into 'skb'. This is unusual for
698 * Netlink but individual action lists can be longer than
699 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
700 * The userspace caller can always fetch the actions separately if it
701 * really wants them. (Most userspace callers in fact don't care.)
703 * This can only fail for dump operations because the skb is always
704 * properly sized for single flows.
706 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
708 const struct sw_flow_actions *sf_acts;
710 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
712 err = ovs_nla_put_actions(sf_acts->actions,
713 sf_acts->actions_len, skb);
715 nla_nest_end(skb, start);
720 nla_nest_cancel(skb, start);
722 } else if (skb_orig_len)
723 goto nla_put_failure;
725 return genlmsg_end(skb, ovs_header);
730 genlmsg_cancel(skb, ovs_header);
734 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
735 struct genl_info *info)
739 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
741 return genlmsg_new_unicast(len, info, GFP_KERNEL);
744 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
746 struct genl_info *info,
752 skb = ovs_flow_cmd_alloc_info(flow, info);
754 return ERR_PTR(-ENOMEM);
756 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
757 info->snd_seq, 0, cmd);
762 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
764 struct nlattr **a = info->attrs;
765 struct ovs_header *ovs_header = info->userhdr;
766 struct sw_flow_key key, masked_key;
767 struct sw_flow *flow = NULL;
768 struct sw_flow_mask mask;
769 struct sk_buff *reply;
771 struct sw_flow_actions *acts = NULL;
772 struct sw_flow_match match;
778 if (!a[OVS_FLOW_ATTR_KEY])
781 ovs_match_init(&match, &key, &mask);
782 error = ovs_nla_get_match(&match, &exact_5tuple,
783 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
787 /* Validate actions. */
788 if (a[OVS_FLOW_ATTR_ACTIONS]) {
789 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
790 error = PTR_ERR(acts);
794 ovs_flow_mask_key(&masked_key, &key, &mask);
795 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
796 &masked_key, 0, &acts);
798 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
801 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
807 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
812 /* Check if this is a duplicate flow */
813 flow = ovs_flow_tbl_lookup(&dp->table, &key);
815 /* Bail out if we're not allowed to create a new flow. */
817 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
821 flow = ovs_flow_alloc(!exact_5tuple);
823 error = PTR_ERR(flow);
827 flow->key = masked_key;
828 flow->unmasked_key = key;
829 rcu_assign_pointer(flow->sf_acts, acts);
831 /* Put flow in bucket. */
832 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
838 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
840 /* We found a matching flow. */
841 struct sw_flow_actions *old_acts;
843 /* Bail out if we're not allowed to modify an existing flow.
844 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
845 * because Generic Netlink treats the latter as a dump
846 * request. We also accept NLM_F_EXCL in case that bug ever
850 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
851 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
854 /* The unmasked key has to be the same for flow updates. */
856 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
857 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
861 /* Update actions. */
862 old_acts = ovsl_dereference(flow->sf_acts);
863 rcu_assign_pointer(flow->sf_acts, acts);
864 ovs_nla_free_flow_actions(old_acts);
866 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
869 if (a[OVS_FLOW_ATTR_CLEAR])
870 ovs_flow_stats_clear(flow);
875 ovs_notify(&dp_flow_genl_family, reply, info);
877 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
882 ovs_flow_free(flow, false);
891 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
893 struct nlattr **a = info->attrs;
894 struct ovs_header *ovs_header = info->userhdr;
895 struct sw_flow_key key;
896 struct sk_buff *reply;
897 struct sw_flow *flow;
899 struct sw_flow_match match;
902 if (!a[OVS_FLOW_ATTR_KEY]) {
903 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
907 ovs_match_init(&match, &key, NULL);
908 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
913 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
919 flow = ovs_flow_tbl_lookup(&dp->table, &key);
920 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
925 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
927 err = PTR_ERR(reply);
932 return genlmsg_reply(reply, info);
938 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
940 struct nlattr **a = info->attrs;
941 struct ovs_header *ovs_header = info->userhdr;
942 struct sw_flow_key key;
943 struct sk_buff *reply;
944 struct sw_flow *flow;
946 struct sw_flow_match match;
950 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
956 if (!a[OVS_FLOW_ATTR_KEY]) {
957 err = ovs_flow_tbl_flush(&dp->table);
961 ovs_match_init(&match, &key, NULL);
962 err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL);
966 flow = ovs_flow_tbl_lookup(&dp->table, &key);
967 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
972 reply = ovs_flow_cmd_alloc_info(flow, info);
978 ovs_flow_tbl_remove(&dp->table, flow);
980 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
981 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
984 ovs_flow_free(flow, true);
987 ovs_notify(&dp_flow_genl_family, reply, info);
994 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
996 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
997 struct table_instance *ti;
1001 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1007 ti = rcu_dereference(dp->table.ti);
1009 struct sw_flow *flow;
1012 bucket = cb->args[0];
1014 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1018 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1019 NETLINK_CB(cb->skb).portid,
1020 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1021 OVS_FLOW_CMD_NEW) < 0)
1024 cb->args[0] = bucket;
1031 static const struct genl_ops dp_flow_genl_ops[] = {
1032 { .cmd = OVS_FLOW_CMD_NEW,
1033 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1034 .policy = flow_policy,
1035 .doit = ovs_flow_cmd_new_or_set
1037 { .cmd = OVS_FLOW_CMD_DEL,
1038 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1039 .policy = flow_policy,
1040 .doit = ovs_flow_cmd_del
1042 { .cmd = OVS_FLOW_CMD_GET,
1043 .flags = 0, /* OK for unprivileged users. */
1044 .policy = flow_policy,
1045 .doit = ovs_flow_cmd_get,
1046 .dumpit = ovs_flow_cmd_dump
1048 { .cmd = OVS_FLOW_CMD_SET,
1049 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1050 .policy = flow_policy,
1051 .doit = ovs_flow_cmd_new_or_set,
1055 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1056 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1057 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1058 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1061 static struct genl_family dp_datapath_genl_family = {
1062 .id = GENL_ID_GENERATE,
1063 .hdrsize = sizeof(struct ovs_header),
1064 .name = OVS_DATAPATH_FAMILY,
1065 .version = OVS_DATAPATH_VERSION,
1066 .maxattr = OVS_DP_ATTR_MAX,
1068 .parallel_ops = true,
1071 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1072 .name = OVS_DATAPATH_MCGROUP
1075 static size_t ovs_dp_cmd_msg_size(void)
1077 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1079 msgsize += nla_total_size(IFNAMSIZ);
1080 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1081 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1086 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1087 u32 portid, u32 seq, u32 flags, u8 cmd)
1089 struct ovs_header *ovs_header;
1090 struct ovs_dp_stats dp_stats;
1091 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1094 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1099 ovs_header->dp_ifindex = get_dpifindex(dp);
1102 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1105 goto nla_put_failure;
1107 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1108 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1110 goto nla_put_failure;
1112 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1113 sizeof(struct ovs_dp_megaflow_stats),
1114 &dp_megaflow_stats))
1115 goto nla_put_failure;
1117 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1118 goto nla_put_failure;
1120 return genlmsg_end(skb, ovs_header);
1123 genlmsg_cancel(skb, ovs_header);
1128 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1129 struct genl_info *info, u8 cmd)
1131 struct sk_buff *skb;
1134 skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1136 return ERR_PTR(-ENOMEM);
1138 retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1141 return ERR_PTR(retval);
1146 /* Called with ovs_mutex. */
1147 static struct datapath *lookup_datapath(struct net *net,
1148 struct ovs_header *ovs_header,
1149 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1151 struct datapath *dp;
1153 if (!a[OVS_DP_ATTR_NAME])
1154 dp = get_dp(net, ovs_header->dp_ifindex);
1156 struct vport *vport;
1159 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1160 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1163 return dp ? dp : ERR_PTR(-ENODEV);
1166 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1168 struct datapath *dp;
1170 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1174 WARN(dp->user_features, "Dropping previously announced user features\n");
1175 dp->user_features = 0;
1178 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1180 if (a[OVS_DP_ATTR_USER_FEATURES])
1181 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1184 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1186 struct nlattr **a = info->attrs;
1187 struct vport_parms parms;
1188 struct sk_buff *reply;
1189 struct datapath *dp;
1190 struct vport *vport;
1191 struct ovs_net *ovs_net;
1195 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1201 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1203 goto err_unlock_ovs;
1205 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1207 /* Allocate table. */
1208 err = ovs_flow_tbl_init(&dp->table);
1212 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1213 if (!dp->stats_percpu) {
1215 goto err_destroy_table;
1218 for_each_possible_cpu(i) {
1219 struct dp_stats_percpu *dpath_stats;
1220 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1221 u64_stats_init(&dpath_stats->sync);
1224 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1228 goto err_destroy_percpu;
1231 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1232 INIT_HLIST_HEAD(&dp->ports[i]);
1234 /* Set up our datapath device. */
1235 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1236 parms.type = OVS_VPORT_TYPE_INTERNAL;
1237 parms.options = NULL;
1239 parms.port_no = OVSP_LOCAL;
1240 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1242 ovs_dp_change(dp, a);
1244 vport = new_vport(&parms);
1245 if (IS_ERR(vport)) {
1246 err = PTR_ERR(vport);
1250 if (err == -EEXIST) {
1251 /* An outdated user space instance that does not understand
1252 * the concept of user_features has attempted to create a new
1253 * datapath and is likely to reuse it. Drop all user features.
1255 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1256 ovs_dp_reset_user_features(skb, info);
1259 goto err_destroy_ports_array;
1262 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1263 err = PTR_ERR(reply);
1265 goto err_destroy_local_port;
1267 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1268 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1272 ovs_notify(&dp_datapath_genl_family, reply, info);
1275 err_destroy_local_port:
1276 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1277 err_destroy_ports_array:
1280 free_percpu(dp->stats_percpu);
1282 ovs_flow_tbl_destroy(&dp->table);
1284 release_net(ovs_dp_get_net(dp));
1292 /* Called with ovs_mutex. */
1293 static void __dp_destroy(struct datapath *dp)
1297 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1298 struct vport *vport;
1299 struct hlist_node *n;
1301 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1302 if (vport->port_no != OVSP_LOCAL)
1303 ovs_dp_detach_port(vport);
1306 list_del_rcu(&dp->list_node);
1308 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1309 * all port in datapath are destroyed first before freeing datapath.
1311 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1313 call_rcu(&dp->rcu, destroy_dp_rcu);
1316 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1318 struct sk_buff *reply;
1319 struct datapath *dp;
1323 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1328 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1329 err = PTR_ERR(reply);
1336 ovs_notify(&dp_datapath_genl_family, reply, info);
1344 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1346 struct sk_buff *reply;
1347 struct datapath *dp;
1351 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1356 ovs_dp_change(dp, info->attrs);
1358 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1359 if (IS_ERR(reply)) {
1360 err = PTR_ERR(reply);
1361 genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1368 ovs_notify(&dp_datapath_genl_family, reply, info);
1376 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1378 struct sk_buff *reply;
1379 struct datapath *dp;
1383 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1389 reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1390 if (IS_ERR(reply)) {
1391 err = PTR_ERR(reply);
1396 return genlmsg_reply(reply, info);
1403 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1405 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1406 struct datapath *dp;
1407 int skip = cb->args[0];
1411 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1413 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1414 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1415 OVS_DP_CMD_NEW) < 0)
1426 static const struct genl_ops dp_datapath_genl_ops[] = {
1427 { .cmd = OVS_DP_CMD_NEW,
1428 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429 .policy = datapath_policy,
1430 .doit = ovs_dp_cmd_new
1432 { .cmd = OVS_DP_CMD_DEL,
1433 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1434 .policy = datapath_policy,
1435 .doit = ovs_dp_cmd_del
1437 { .cmd = OVS_DP_CMD_GET,
1438 .flags = 0, /* OK for unprivileged users. */
1439 .policy = datapath_policy,
1440 .doit = ovs_dp_cmd_get,
1441 .dumpit = ovs_dp_cmd_dump
1443 { .cmd = OVS_DP_CMD_SET,
1444 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1445 .policy = datapath_policy,
1446 .doit = ovs_dp_cmd_set,
1450 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1451 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1452 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1453 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1454 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1455 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1456 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1459 struct genl_family dp_vport_genl_family = {
1460 .id = GENL_ID_GENERATE,
1461 .hdrsize = sizeof(struct ovs_header),
1462 .name = OVS_VPORT_FAMILY,
1463 .version = OVS_VPORT_VERSION,
1464 .maxattr = OVS_VPORT_ATTR_MAX,
1466 .parallel_ops = true,
1469 static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1470 .name = OVS_VPORT_MCGROUP
1473 /* Called with ovs_mutex or RCU read lock. */
1474 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1475 u32 portid, u32 seq, u32 flags, u8 cmd)
1477 struct ovs_header *ovs_header;
1478 struct ovs_vport_stats vport_stats;
1481 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1486 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1488 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1489 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1490 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1491 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1492 goto nla_put_failure;
1494 ovs_vport_get_stats(vport, &vport_stats);
1495 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1497 goto nla_put_failure;
1499 err = ovs_vport_get_options(vport, skb);
1500 if (err == -EMSGSIZE)
1503 return genlmsg_end(skb, ovs_header);
1508 genlmsg_cancel(skb, ovs_header);
1512 /* Called with ovs_mutex or RCU read lock. */
1513 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1516 struct sk_buff *skb;
1519 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1521 return ERR_PTR(-ENOMEM);
1523 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1529 /* Called with ovs_mutex or RCU read lock. */
1530 static struct vport *lookup_vport(struct net *net,
1531 struct ovs_header *ovs_header,
1532 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1534 struct datapath *dp;
1535 struct vport *vport;
1537 if (a[OVS_VPORT_ATTR_NAME]) {
1538 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1540 return ERR_PTR(-ENODEV);
1541 if (ovs_header->dp_ifindex &&
1542 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1543 return ERR_PTR(-ENODEV);
1545 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1546 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1548 if (port_no >= DP_MAX_PORTS)
1549 return ERR_PTR(-EFBIG);
1551 dp = get_dp(net, ovs_header->dp_ifindex);
1553 return ERR_PTR(-ENODEV);
1555 vport = ovs_vport_ovsl_rcu(dp, port_no);
1557 return ERR_PTR(-ENODEV);
1560 return ERR_PTR(-EINVAL);
1563 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1565 struct nlattr **a = info->attrs;
1566 struct ovs_header *ovs_header = info->userhdr;
1567 struct vport_parms parms;
1568 struct sk_buff *reply;
1569 struct vport *vport;
1570 struct datapath *dp;
1575 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1576 !a[OVS_VPORT_ATTR_UPCALL_PID])
1580 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1585 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1586 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1589 if (port_no >= DP_MAX_PORTS)
1592 vport = ovs_vport_ovsl(dp, port_no);
1597 for (port_no = 1; ; port_no++) {
1598 if (port_no >= DP_MAX_PORTS) {
1602 vport = ovs_vport_ovsl(dp, port_no);
1608 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1609 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1610 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1612 parms.port_no = port_no;
1613 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1615 vport = new_vport(&parms);
1616 err = PTR_ERR(vport);
1621 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1623 if (IS_ERR(reply)) {
1624 err = PTR_ERR(reply);
1625 ovs_dp_detach_port(vport);
1629 ovs_notify(&dp_vport_genl_family, reply, info);
1637 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1639 struct nlattr **a = info->attrs;
1640 struct sk_buff *reply;
1641 struct vport *vport;
1645 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1646 err = PTR_ERR(vport);
1650 if (a[OVS_VPORT_ATTR_TYPE] &&
1651 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1656 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1662 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1663 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1668 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1669 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1671 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1672 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1676 ovs_notify(&dp_vport_genl_family, reply, info);
1686 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1688 struct nlattr **a = info->attrs;
1689 struct sk_buff *reply;
1690 struct vport *vport;
1694 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1695 err = PTR_ERR(vport);
1699 if (vport->port_no == OVSP_LOCAL) {
1704 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1705 info->snd_seq, OVS_VPORT_CMD_DEL);
1706 err = PTR_ERR(reply);
1711 ovs_dp_detach_port(vport);
1713 ovs_notify(&dp_vport_genl_family, reply, info);
1720 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1722 struct nlattr **a = info->attrs;
1723 struct ovs_header *ovs_header = info->userhdr;
1724 struct sk_buff *reply;
1725 struct vport *vport;
1729 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1730 err = PTR_ERR(vport);
1734 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1735 info->snd_seq, OVS_VPORT_CMD_NEW);
1736 err = PTR_ERR(reply);
1742 return genlmsg_reply(reply, info);
1749 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1751 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1752 struct datapath *dp;
1753 int bucket = cb->args[0], skip = cb->args[1];
1756 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1761 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1762 struct vport *vport;
1765 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1767 ovs_vport_cmd_fill_info(vport, skb,
1768 NETLINK_CB(cb->skb).portid,
1771 OVS_VPORT_CMD_NEW) < 0)
1787 static const struct genl_ops dp_vport_genl_ops[] = {
1788 { .cmd = OVS_VPORT_CMD_NEW,
1789 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1790 .policy = vport_policy,
1791 .doit = ovs_vport_cmd_new
1793 { .cmd = OVS_VPORT_CMD_DEL,
1794 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1795 .policy = vport_policy,
1796 .doit = ovs_vport_cmd_del
1798 { .cmd = OVS_VPORT_CMD_GET,
1799 .flags = 0, /* OK for unprivileged users. */
1800 .policy = vport_policy,
1801 .doit = ovs_vport_cmd_get,
1802 .dumpit = ovs_vport_cmd_dump
1804 { .cmd = OVS_VPORT_CMD_SET,
1805 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1806 .policy = vport_policy,
1807 .doit = ovs_vport_cmd_set,
1811 struct genl_family_and_ops {
1812 struct genl_family *family;
1813 const struct genl_ops *ops;
1815 const struct genl_multicast_group *group;
1818 static const struct genl_family_and_ops dp_genl_families[] = {
1819 { &dp_datapath_genl_family,
1820 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1821 &ovs_dp_datapath_multicast_group },
1822 { &dp_vport_genl_family,
1823 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1824 &ovs_dp_vport_multicast_group },
1825 { &dp_flow_genl_family,
1826 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1827 &ovs_dp_flow_multicast_group },
1828 { &dp_packet_genl_family,
1829 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1833 static void dp_unregister_genl(int n_families)
1837 for (i = 0; i < n_families; i++)
1838 genl_unregister_family(dp_genl_families[i].family);
1841 static int dp_register_genl(void)
1848 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1849 const struct genl_family_and_ops *f = &dp_genl_families[i];
1851 f->family->ops = f->ops;
1852 f->family->n_ops = f->n_ops;
1853 f->family->mcgrps = f->group;
1854 f->family->n_mcgrps = f->group ? 1 : 0;
1855 err = genl_register_family(f->family);
1864 dp_unregister_genl(n_registered);
1868 static int __net_init ovs_init_net(struct net *net)
1870 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1872 INIT_LIST_HEAD(&ovs_net->dps);
1873 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1877 static void __net_exit ovs_exit_net(struct net *net)
1879 struct datapath *dp, *dp_next;
1880 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1883 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1887 cancel_work_sync(&ovs_net->dp_notify_work);
1890 static struct pernet_operations ovs_net_ops = {
1891 .init = ovs_init_net,
1892 .exit = ovs_exit_net,
1894 .size = sizeof(struct ovs_net),
1897 static int __init dp_init(void)
1901 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1903 pr_info("Open vSwitch switching datapath\n");
1905 err = ovs_flow_init();
1909 err = ovs_vport_init();
1911 goto error_flow_exit;
1913 err = register_pernet_device(&ovs_net_ops);
1915 goto error_vport_exit;
1917 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1919 goto error_netns_exit;
1921 err = dp_register_genl();
1923 goto error_unreg_notifier;
1927 error_unreg_notifier:
1928 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1930 unregister_pernet_device(&ovs_net_ops);
1939 static void dp_cleanup(void)
1941 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1942 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1943 unregister_pernet_device(&ovs_net_ops);
1949 module_init(dp_init);
1950 module_exit(dp_cleanup);
1952 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1953 MODULE_LICENSE("GPL");