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 /* Check if need to build a reply message.
68 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
69 static bool ovs_must_notify(struct genl_info *info,
70 const struct genl_multicast_group *grp)
72 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
73 netlink_has_listeners(genl_info_net(info)->genl_sock, grp->id);
76 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
77 struct genl_multicast_group *grp)
79 genl_notify(skb, genl_info_net(info), info->snd_portid,
80 grp->id, info->nlhdr, GFP_KERNEL);
86 * All writes e.g. Writes to device state (add/remove datapath, port, set
87 * operations on vports, etc.), Writes to other state (flow table
88 * modifications, set miscellaneous datapath parameters, etc.) are protected
91 * Reads are protected by RCU.
93 * There are a few special cases (mostly stats) that have their own
94 * synchronization but they nest under all of above and don't interact with
97 * The RTNL lock nests inside ovs_mutex.
100 static DEFINE_MUTEX(ovs_mutex);
104 mutex_lock(&ovs_mutex);
107 void ovs_unlock(void)
109 mutex_unlock(&ovs_mutex);
112 #ifdef CONFIG_LOCKDEP
113 int lockdep_ovsl_is_held(void)
116 return lockdep_is_held(&ovs_mutex);
122 static struct vport *new_vport(const struct vport_parms *);
123 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
124 const struct dp_upcall_info *);
125 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
126 const struct dp_upcall_info *);
128 /* Must be called with rcu_read_lock or ovs_mutex. */
129 static struct datapath *get_dp(struct net *net, int dp_ifindex)
131 struct datapath *dp = NULL;
132 struct net_device *dev;
135 dev = dev_get_by_index_rcu(net, dp_ifindex);
137 struct vport *vport = ovs_internal_dev_get_vport(dev);
146 /* Must be called with rcu_read_lock or ovs_mutex. */
147 const char *ovs_dp_name(const struct datapath *dp)
149 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
150 return vport->ops->get_name(vport);
153 static int get_dpifindex(struct datapath *dp)
160 local = ovs_vport_rcu(dp, OVSP_LOCAL);
162 ifindex = netdev_vport_priv(local)->dev->ifindex;
171 static void destroy_dp_rcu(struct rcu_head *rcu)
173 struct datapath *dp = container_of(rcu, struct datapath, rcu);
175 free_percpu(dp->stats_percpu);
176 release_net(ovs_dp_get_net(dp));
181 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
184 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
187 /* Called with ovs_mutex or RCU read lock. */
188 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
191 struct hlist_head *head;
193 head = vport_hash_bucket(dp, port_no);
194 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
195 if (vport->port_no == port_no)
201 /* Called with ovs_mutex. */
202 static struct vport *new_vport(const struct vport_parms *parms)
206 vport = ovs_vport_add(parms);
207 if (!IS_ERR(vport)) {
208 struct datapath *dp = parms->dp;
209 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
211 hlist_add_head_rcu(&vport->dp_hash_node, head);
216 void ovs_dp_detach_port(struct vport *p)
220 /* First drop references to device. */
221 hlist_del_rcu(&p->dp_hash_node);
223 /* Then destroy it. */
227 /* Must be called with rcu_read_lock. */
228 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
230 struct datapath *dp = p->dp;
231 struct sw_flow *flow;
232 struct dp_stats_percpu *stats;
233 struct sw_flow_key key;
238 stats = this_cpu_ptr(dp->stats_percpu);
240 /* Extract flow from 'skb' into 'key'. */
241 error = ovs_flow_extract(skb, p->port_no, &key);
242 if (unlikely(error)) {
248 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
249 if (unlikely(!flow)) {
250 struct dp_upcall_info upcall;
252 upcall.cmd = OVS_PACKET_CMD_MISS;
254 upcall.userdata = NULL;
255 upcall.portid = p->upcall_portid;
256 ovs_dp_upcall(dp, skb, &upcall);
258 stats_counter = &stats->n_missed;
262 OVS_CB(skb)->flow = flow;
263 OVS_CB(skb)->pkt_key = &key;
265 ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
266 ovs_execute_actions(dp, skb);
267 stats_counter = &stats->n_hit;
270 /* Update datapath statistics. */
271 u64_stats_update_begin(&stats->sync);
273 stats->n_mask_hit += n_mask_hit;
274 u64_stats_update_end(&stats->sync);
277 static struct genl_family dp_packet_genl_family = {
278 .id = GENL_ID_GENERATE,
279 .hdrsize = sizeof(struct ovs_header),
280 .name = OVS_PACKET_FAMILY,
281 .version = OVS_PACKET_VERSION,
282 .maxattr = OVS_PACKET_ATTR_MAX,
287 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
288 const struct dp_upcall_info *upcall_info)
290 struct dp_stats_percpu *stats;
293 if (upcall_info->portid == 0) {
298 if (!skb_is_gso(skb))
299 err = queue_userspace_packet(dp, skb, upcall_info);
301 err = queue_gso_packets(dp, skb, upcall_info);
308 stats = this_cpu_ptr(dp->stats_percpu);
310 u64_stats_update_begin(&stats->sync);
312 u64_stats_update_end(&stats->sync);
317 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
318 const struct dp_upcall_info *upcall_info)
320 unsigned short gso_type = skb_shinfo(skb)->gso_type;
321 struct dp_upcall_info later_info;
322 struct sw_flow_key later_key;
323 struct sk_buff *segs, *nskb;
326 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
328 return PTR_ERR(segs);
330 /* Queue all of the segments. */
333 err = queue_userspace_packet(dp, skb, upcall_info);
337 if (skb == segs && gso_type & SKB_GSO_UDP) {
338 /* The initial flow key extracted by ovs_flow_extract()
339 * in this case is for a first fragment, so we need to
340 * properly mark later fragments.
342 later_key = *upcall_info->key;
343 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
345 later_info = *upcall_info;
346 later_info.key = &later_key;
347 upcall_info = &later_info;
349 } while ((skb = skb->next));
351 /* Free all of the segments. */
359 } while ((skb = nskb));
363 static size_t key_attr_size(void)
365 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
366 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
367 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
368 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
369 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
370 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
371 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
372 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
373 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
374 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
375 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
376 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
377 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
378 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
379 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
380 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
381 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
382 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
383 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
386 static size_t upcall_msg_size(const struct nlattr *userdata,
389 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
390 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
391 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
393 /* OVS_PACKET_ATTR_USERDATA */
395 size += NLA_ALIGN(userdata->nla_len);
400 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
401 const struct dp_upcall_info *upcall_info)
403 struct ovs_header *upcall;
404 struct sk_buff *nskb = NULL;
405 struct sk_buff *user_skb; /* to be queued to userspace */
407 struct genl_info info = {
408 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
409 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
411 .snd_portid = upcall_info->portid,
417 dp_ifindex = get_dpifindex(dp);
421 if (vlan_tx_tag_present(skb)) {
422 nskb = skb_clone(skb, GFP_ATOMIC);
426 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
430 vlan_set_tci(nskb, 0);
435 if (nla_attr_size(skb->len) > USHRT_MAX) {
440 /* Complete checksum if needed */
441 if (skb->ip_summed == CHECKSUM_PARTIAL &&
442 (err = skb_checksum_help(skb)))
445 /* Older versions of OVS user space enforce alignment of the last
446 * Netlink attribute to NLA_ALIGNTO which would require extensive
447 * padding logic. Only perform zerocopy if padding is not required.
449 if (dp->user_features & OVS_DP_F_UNALIGNED)
450 hlen = skb_zerocopy_headlen(skb);
454 len = upcall_msg_size(upcall_info->userdata, hlen);
455 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
461 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
462 0, upcall_info->cmd);
463 upcall->dp_ifindex = dp_ifindex;
465 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
466 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
467 nla_nest_end(user_skb, nla);
469 if (upcall_info->userdata)
470 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
471 nla_len(upcall_info->userdata),
472 nla_data(upcall_info->userdata));
474 /* Only reserve room for attribute header, packet data is added
475 * in skb_zerocopy() */
476 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
480 nla->nla_len = nla_attr_size(skb->len);
482 skb_zerocopy(user_skb, skb, skb->len, hlen);
484 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
485 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
486 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
489 memset(skb_put(user_skb, plen), 0, plen);
492 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
494 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
500 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
502 struct ovs_header *ovs_header = info->userhdr;
503 struct nlattr **a = info->attrs;
504 struct sw_flow_actions *acts;
505 struct sk_buff *packet;
506 struct sw_flow *flow;
513 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
514 !a[OVS_PACKET_ATTR_ACTIONS])
517 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
518 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
522 skb_reserve(packet, NET_IP_ALIGN);
524 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
526 skb_reset_mac_header(packet);
527 eth = eth_hdr(packet);
529 /* Normally, setting the skb 'protocol' field would be handled by a
530 * call to eth_type_trans(), but it assumes there's a sending
531 * device, which we may not have. */
532 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
533 packet->protocol = eth->h_proto;
535 packet->protocol = htons(ETH_P_802_2);
537 /* Build an sw_flow for sending this packet. */
538 flow = ovs_flow_alloc();
543 err = ovs_flow_extract(packet, -1, &flow->key);
547 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
550 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
555 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
556 &flow->key, 0, &acts);
557 rcu_assign_pointer(flow->sf_acts, acts);
561 OVS_CB(packet)->flow = flow;
562 OVS_CB(packet)->pkt_key = &flow->key;
563 packet->priority = flow->key.phy.priority;
564 packet->mark = flow->key.phy.skb_mark;
567 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
573 err = ovs_execute_actions(dp, packet);
577 ovs_flow_free(flow, false);
583 ovs_flow_free(flow, false);
590 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
591 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
592 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
593 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
596 static struct genl_ops dp_packet_genl_ops[] = {
597 { .cmd = OVS_PACKET_CMD_EXECUTE,
598 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
599 .policy = packet_policy,
600 .doit = ovs_packet_cmd_execute
604 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
605 struct ovs_dp_megaflow_stats *mega_stats)
609 memset(mega_stats, 0, sizeof(*mega_stats));
611 stats->n_flows = ovs_flow_tbl_count(&dp->table);
612 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
614 stats->n_hit = stats->n_missed = stats->n_lost = 0;
616 for_each_possible_cpu(i) {
617 const struct dp_stats_percpu *percpu_stats;
618 struct dp_stats_percpu local_stats;
621 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
624 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
625 local_stats = *percpu_stats;
626 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
628 stats->n_hit += local_stats.n_hit;
629 stats->n_missed += local_stats.n_missed;
630 stats->n_lost += local_stats.n_lost;
631 mega_stats->n_mask_hit += local_stats.n_mask_hit;
635 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
636 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
637 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
638 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
641 static struct genl_family dp_flow_genl_family = {
642 .id = GENL_ID_GENERATE,
643 .hdrsize = sizeof(struct ovs_header),
644 .name = OVS_FLOW_FAMILY,
645 .version = OVS_FLOW_VERSION,
646 .maxattr = OVS_FLOW_ATTR_MAX,
651 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
652 .name = OVS_FLOW_MCGROUP
655 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
657 return NLMSG_ALIGN(sizeof(struct ovs_header))
658 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
659 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
660 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
661 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
662 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
663 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
666 /* Called with ovs_mutex or RCU read lock. */
667 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
668 struct sk_buff *skb, u32 portid,
669 u32 seq, u32 flags, u8 cmd)
671 const int skb_orig_len = skb->len;
672 struct nlattr *start;
673 struct ovs_flow_stats stats;
676 struct ovs_header *ovs_header;
680 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
684 ovs_header->dp_ifindex = get_dpifindex(dp);
687 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
689 goto nla_put_failure;
691 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
694 nla_nest_end(skb, nla);
696 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
698 goto nla_put_failure;
700 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
704 nla_nest_end(skb, nla);
706 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
708 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
709 goto nla_put_failure;
711 if (stats.n_packets &&
712 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
713 goto nla_put_failure;
715 if ((u8)ntohs(tcp_flags) &&
716 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
717 goto nla_put_failure;
719 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
720 * this is the first flow to be dumped into 'skb'. This is unusual for
721 * Netlink but individual action lists can be longer than
722 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
723 * The userspace caller can always fetch the actions separately if it
724 * really wants them. (Most userspace callers in fact don't care.)
726 * This can only fail for dump operations because the skb is always
727 * properly sized for single flows.
729 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
731 const struct sw_flow_actions *sf_acts;
733 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
735 err = ovs_nla_put_actions(sf_acts->actions,
736 sf_acts->actions_len, skb);
738 nla_nest_end(skb, start);
743 nla_nest_cancel(skb, start);
745 } else if (skb_orig_len)
746 goto nla_put_failure;
748 return genlmsg_end(skb, ovs_header);
753 genlmsg_cancel(skb, ovs_header);
757 /* Must be called with ovs_mutex. */
758 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
759 struct genl_info *info,
765 if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
768 len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
770 skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
773 return ERR_PTR(-ENOMEM);
778 /* Must be called with ovs_mutex. */
779 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
781 struct genl_info *info,
787 skb = ovs_flow_cmd_alloc_info(flow, info, always);
788 if (!skb || IS_ERR(skb))
791 retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
792 info->snd_seq, 0, cmd);
797 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
799 struct nlattr **a = info->attrs;
800 struct ovs_header *ovs_header = info->userhdr;
801 struct sw_flow_key key, masked_key;
802 struct sw_flow *flow = NULL;
803 struct sw_flow_mask mask;
804 struct sk_buff *reply;
806 struct sw_flow_actions *acts = NULL;
807 struct sw_flow_match match;
812 if (!a[OVS_FLOW_ATTR_KEY])
815 ovs_match_init(&match, &key, &mask);
816 error = ovs_nla_get_match(&match,
817 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
821 /* Validate actions. */
822 if (a[OVS_FLOW_ATTR_ACTIONS]) {
823 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
824 error = PTR_ERR(acts);
828 ovs_flow_mask_key(&masked_key, &key, &mask);
829 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
830 &masked_key, 0, &acts);
832 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
835 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
836 /* OVS_FLOW_CMD_NEW must have actions. */
842 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
847 /* Check if this is a duplicate flow */
848 flow = ovs_flow_tbl_lookup(&dp->table, &key);
850 /* Bail out if we're not allowed to create a new flow. */
852 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
856 flow = ovs_flow_alloc();
858 error = PTR_ERR(flow);
862 flow->key = masked_key;
863 flow->unmasked_key = key;
864 rcu_assign_pointer(flow->sf_acts, acts);
866 /* Put flow in bucket. */
867 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
873 reply = ovs_flow_cmd_build_info(flow, dp, info,
874 OVS_FLOW_CMD_NEW, false);
876 /* We found a matching flow. */
877 /* Bail out if we're not allowed to modify an existing flow.
878 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
879 * because Generic Netlink treats the latter as a dump
880 * request. We also accept NLM_F_EXCL in case that bug ever
884 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
885 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
888 /* The unmasked key has to be the same for flow updates. */
889 if (!ovs_flow_cmp_unmasked_key(flow, &match))
892 /* Update actions, if present. */
894 struct sw_flow_actions *old_acts;
896 old_acts = ovsl_dereference(flow->sf_acts);
897 rcu_assign_pointer(flow->sf_acts, acts);
898 ovs_nla_free_flow_actions(old_acts);
900 reply = ovs_flow_cmd_build_info(flow, dp, info,
901 OVS_FLOW_CMD_NEW, false);
904 if (a[OVS_FLOW_ATTR_CLEAR])
905 ovs_flow_stats_clear(flow);
911 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
913 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
914 ovs_dp_flow_multicast_group.id,
920 ovs_flow_free(flow, false);
929 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
931 struct nlattr **a = info->attrs;
932 struct ovs_header *ovs_header = info->userhdr;
933 struct sw_flow_key key;
934 struct sk_buff *reply;
935 struct sw_flow *flow;
937 struct sw_flow_match match;
940 if (!a[OVS_FLOW_ATTR_KEY]) {
941 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
945 ovs_match_init(&match, &key, NULL);
946 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
951 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
957 flow = ovs_flow_tbl_lookup(&dp->table, &key);
958 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
963 reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW, true);
965 err = PTR_ERR(reply);
970 return genlmsg_reply(reply, info);
976 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
978 struct nlattr **a = info->attrs;
979 struct ovs_header *ovs_header = info->userhdr;
980 struct sw_flow_key key;
981 struct sk_buff *reply;
982 struct sw_flow *flow;
984 struct sw_flow_match match;
988 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
994 if (!a[OVS_FLOW_ATTR_KEY]) {
995 err = ovs_flow_tbl_flush(&dp->table);
999 ovs_match_init(&match, &key, NULL);
1000 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1004 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1005 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1010 reply = ovs_flow_cmd_alloc_info(flow, info, false);
1011 if (IS_ERR(reply)) {
1012 err = PTR_ERR(reply);
1016 ovs_flow_tbl_remove(&dp->table, flow);
1019 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1025 ovs_flow_free(flow, true);
1029 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1036 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1038 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1039 struct table_instance *ti;
1040 struct datapath *dp;
1043 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1049 ti = rcu_dereference(dp->table.ti);
1051 struct sw_flow *flow;
1054 bucket = cb->args[0];
1056 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1060 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1061 NETLINK_CB(cb->skb).portid,
1062 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1063 OVS_FLOW_CMD_NEW) < 0)
1066 cb->args[0] = bucket;
1073 static struct genl_ops dp_flow_genl_ops[] = {
1074 { .cmd = OVS_FLOW_CMD_NEW,
1075 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1076 .policy = flow_policy,
1077 .doit = ovs_flow_cmd_new_or_set
1079 { .cmd = OVS_FLOW_CMD_DEL,
1080 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1081 .policy = flow_policy,
1082 .doit = ovs_flow_cmd_del
1084 { .cmd = OVS_FLOW_CMD_GET,
1085 .flags = 0, /* OK for unprivileged users. */
1086 .policy = flow_policy,
1087 .doit = ovs_flow_cmd_get,
1088 .dumpit = ovs_flow_cmd_dump
1090 { .cmd = OVS_FLOW_CMD_SET,
1091 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1092 .policy = flow_policy,
1093 .doit = ovs_flow_cmd_new_or_set,
1097 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1098 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1099 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1100 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1103 static struct genl_family dp_datapath_genl_family = {
1104 .id = GENL_ID_GENERATE,
1105 .hdrsize = sizeof(struct ovs_header),
1106 .name = OVS_DATAPATH_FAMILY,
1107 .version = OVS_DATAPATH_VERSION,
1108 .maxattr = OVS_DP_ATTR_MAX,
1113 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1114 .name = OVS_DATAPATH_MCGROUP
1117 static size_t ovs_dp_cmd_msg_size(void)
1119 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1121 msgsize += nla_total_size(IFNAMSIZ);
1122 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1123 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1124 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1129 /* Called with ovs_mutex or RCU read lock. */
1130 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1131 u32 portid, u32 seq, u32 flags, u8 cmd)
1133 struct ovs_header *ovs_header;
1134 struct ovs_dp_stats dp_stats;
1135 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1138 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1143 ovs_header->dp_ifindex = get_dpifindex(dp);
1145 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1147 goto nla_put_failure;
1149 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1150 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1152 goto nla_put_failure;
1154 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1155 sizeof(struct ovs_dp_megaflow_stats),
1156 &dp_megaflow_stats))
1157 goto nla_put_failure;
1159 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1160 goto nla_put_failure;
1162 return genlmsg_end(skb, ovs_header);
1165 genlmsg_cancel(skb, ovs_header);
1170 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1172 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1175 /* Called with rcu_read_lock or ovs_mutex. */
1176 static struct datapath *lookup_datapath(struct net *net,
1177 struct ovs_header *ovs_header,
1178 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1180 struct datapath *dp;
1182 if (!a[OVS_DP_ATTR_NAME])
1183 dp = get_dp(net, ovs_header->dp_ifindex);
1185 struct vport *vport;
1187 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1188 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1190 return dp ? dp : ERR_PTR(-ENODEV);
1193 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1195 struct datapath *dp;
1197 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1201 WARN(dp->user_features, "Dropping previously announced user features\n");
1202 dp->user_features = 0;
1205 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1207 if (a[OVS_DP_ATTR_USER_FEATURES])
1208 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1211 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1213 struct nlattr **a = info->attrs;
1214 struct vport_parms parms;
1215 struct sk_buff *reply;
1216 struct datapath *dp;
1217 struct vport *vport;
1218 struct ovs_net *ovs_net;
1222 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1225 reply = ovs_dp_cmd_alloc_info(info);
1230 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1232 goto err_free_reply;
1234 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1236 /* Allocate table. */
1237 err = ovs_flow_tbl_init(&dp->table);
1241 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1242 if (!dp->stats_percpu) {
1244 goto err_destroy_table;
1247 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1251 goto err_destroy_percpu;
1254 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1255 INIT_HLIST_HEAD(&dp->ports[i]);
1257 /* Set up our datapath device. */
1258 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1259 parms.type = OVS_VPORT_TYPE_INTERNAL;
1260 parms.options = NULL;
1262 parms.port_no = OVSP_LOCAL;
1263 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1265 ovs_dp_change(dp, a);
1267 /* So far only local changes have been made, now need the lock. */
1270 vport = new_vport(&parms);
1271 if (IS_ERR(vport)) {
1272 err = PTR_ERR(vport);
1276 if (err == -EEXIST) {
1277 /* An outdated user space instance that does not understand
1278 * the concept of user_features has attempted to create a new
1279 * datapath and is likely to reuse it. Drop all user features.
1281 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1282 ovs_dp_reset_user_features(skb, info);
1285 goto err_destroy_ports_array;
1288 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1289 info->snd_seq, 0, OVS_DP_CMD_NEW);
1292 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1293 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1297 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1300 err_destroy_ports_array:
1304 free_percpu(dp->stats_percpu);
1306 ovs_flow_tbl_destroy(&dp->table, false);
1308 release_net(ovs_dp_get_net(dp));
1316 /* Called with ovs_mutex. */
1317 static void __dp_destroy(struct datapath *dp)
1321 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1322 struct vport *vport;
1323 struct hlist_node *n;
1325 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1326 if (vport->port_no != OVSP_LOCAL)
1327 ovs_dp_detach_port(vport);
1330 list_del_rcu(&dp->list_node);
1332 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1333 * all ports in datapath are destroyed first before freeing datapath.
1335 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1337 /* RCU destroy the flow table */
1338 ovs_flow_tbl_destroy(&dp->table, true);
1340 call_rcu(&dp->rcu, destroy_dp_rcu);
1343 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1345 struct sk_buff *reply;
1346 struct datapath *dp;
1349 reply = ovs_dp_cmd_alloc_info(info);
1354 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1357 goto err_unlock_free;
1359 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1360 info->snd_seq, 0, OVS_DP_CMD_DEL);
1366 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1375 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1377 struct sk_buff *reply;
1378 struct datapath *dp;
1381 reply = ovs_dp_cmd_alloc_info(info);
1386 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1389 goto err_unlock_free;
1391 ovs_dp_change(dp, info->attrs);
1393 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1394 info->snd_seq, 0, OVS_DP_CMD_NEW);
1398 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1407 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1409 struct sk_buff *reply;
1410 struct datapath *dp;
1413 reply = ovs_dp_cmd_alloc_info(info);
1418 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1421 goto err_unlock_free;
1423 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1424 info->snd_seq, 0, OVS_DP_CMD_NEW);
1428 return genlmsg_reply(reply, info);
1436 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1438 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1439 struct datapath *dp;
1440 int skip = cb->args[0];
1444 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1446 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1447 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1448 OVS_DP_CMD_NEW) < 0)
1459 static struct genl_ops dp_datapath_genl_ops[] = {
1460 { .cmd = OVS_DP_CMD_NEW,
1461 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1462 .policy = datapath_policy,
1463 .doit = ovs_dp_cmd_new
1465 { .cmd = OVS_DP_CMD_DEL,
1466 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1467 .policy = datapath_policy,
1468 .doit = ovs_dp_cmd_del
1470 { .cmd = OVS_DP_CMD_GET,
1471 .flags = 0, /* OK for unprivileged users. */
1472 .policy = datapath_policy,
1473 .doit = ovs_dp_cmd_get,
1474 .dumpit = ovs_dp_cmd_dump
1476 { .cmd = OVS_DP_CMD_SET,
1477 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1478 .policy = datapath_policy,
1479 .doit = ovs_dp_cmd_set,
1483 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1484 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1485 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1486 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1487 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1488 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1489 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1492 static struct genl_family dp_vport_genl_family = {
1493 .id = GENL_ID_GENERATE,
1494 .hdrsize = sizeof(struct ovs_header),
1495 .name = OVS_VPORT_FAMILY,
1496 .version = OVS_VPORT_VERSION,
1497 .maxattr = OVS_VPORT_ATTR_MAX,
1502 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1503 .name = OVS_VPORT_MCGROUP
1506 /* Called with ovs_mutex or RCU read lock. */
1507 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1508 u32 portid, u32 seq, u32 flags, u8 cmd)
1510 struct ovs_header *ovs_header;
1511 struct ovs_vport_stats vport_stats;
1514 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1519 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1521 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1522 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1523 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1524 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1525 goto nla_put_failure;
1527 ovs_vport_get_stats(vport, &vport_stats);
1528 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1530 goto nla_put_failure;
1532 err = ovs_vport_get_options(vport, skb);
1533 if (err == -EMSGSIZE)
1536 return genlmsg_end(skb, ovs_header);
1541 genlmsg_cancel(skb, ovs_header);
1545 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1547 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1550 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1551 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1554 struct sk_buff *skb;
1557 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1559 return ERR_PTR(-ENOMEM);
1561 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1567 /* Called with ovs_mutex or RCU read lock. */
1568 static struct vport *lookup_vport(struct net *net,
1569 struct ovs_header *ovs_header,
1570 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1572 struct datapath *dp;
1573 struct vport *vport;
1575 if (a[OVS_VPORT_ATTR_NAME]) {
1576 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1578 return ERR_PTR(-ENODEV);
1579 if (ovs_header->dp_ifindex &&
1580 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1581 return ERR_PTR(-ENODEV);
1583 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1584 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1586 if (port_no >= DP_MAX_PORTS)
1587 return ERR_PTR(-EFBIG);
1589 dp = get_dp(net, ovs_header->dp_ifindex);
1591 return ERR_PTR(-ENODEV);
1593 vport = ovs_vport_ovsl_rcu(dp, port_no);
1595 return ERR_PTR(-ENODEV);
1598 return ERR_PTR(-EINVAL);
1601 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1603 struct nlattr **a = info->attrs;
1604 struct ovs_header *ovs_header = info->userhdr;
1605 struct vport_parms parms;
1606 struct sk_buff *reply;
1607 struct vport *vport;
1608 struct datapath *dp;
1612 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1613 !a[OVS_VPORT_ATTR_UPCALL_PID])
1616 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1617 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1618 if (port_no >= DP_MAX_PORTS)
1621 reply = ovs_vport_cmd_alloc_info();
1626 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1629 goto exit_unlock_free;
1632 vport = ovs_vport_ovsl(dp, port_no);
1635 goto exit_unlock_free;
1637 for (port_no = 1; ; port_no++) {
1638 if (port_no >= DP_MAX_PORTS) {
1640 goto exit_unlock_free;
1642 vport = ovs_vport_ovsl(dp, port_no);
1648 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1649 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1650 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1652 parms.port_no = port_no;
1653 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1655 vport = new_vport(&parms);
1656 err = PTR_ERR(vport);
1658 goto exit_unlock_free;
1661 if (a[OVS_VPORT_ATTR_STATS])
1662 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1664 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1665 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1669 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1678 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1680 struct nlattr **a = info->attrs;
1681 struct sk_buff *reply;
1682 struct vport *vport;
1685 reply = ovs_vport_cmd_alloc_info();
1690 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1691 err = PTR_ERR(vport);
1693 goto exit_unlock_free;
1695 if (a[OVS_VPORT_ATTR_TYPE] &&
1696 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1698 goto exit_unlock_free;
1701 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1702 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1704 goto exit_unlock_free;
1707 if (a[OVS_VPORT_ATTR_STATS])
1708 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1710 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1711 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1713 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1714 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1718 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1727 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1729 struct nlattr **a = info->attrs;
1730 struct sk_buff *reply;
1731 struct vport *vport;
1734 reply = ovs_vport_cmd_alloc_info();
1739 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1740 err = PTR_ERR(vport);
1742 goto exit_unlock_free;
1744 if (vport->port_no == OVSP_LOCAL) {
1746 goto exit_unlock_free;
1749 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1750 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1752 ovs_dp_detach_port(vport);
1755 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1764 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1766 struct nlattr **a = info->attrs;
1767 struct ovs_header *ovs_header = info->userhdr;
1768 struct sk_buff *reply;
1769 struct vport *vport;
1772 reply = ovs_vport_cmd_alloc_info();
1777 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1778 err = PTR_ERR(vport);
1780 goto exit_unlock_free;
1781 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1782 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1786 return genlmsg_reply(reply, info);
1794 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1796 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1797 struct datapath *dp;
1798 int bucket = cb->args[0], skip = cb->args[1];
1802 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1807 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1808 struct vport *vport;
1811 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1813 ovs_vport_cmd_fill_info(vport, skb,
1814 NETLINK_CB(cb->skb).portid,
1817 OVS_VPORT_CMD_NEW) < 0)
1833 static struct genl_ops dp_vport_genl_ops[] = {
1834 { .cmd = OVS_VPORT_CMD_NEW,
1835 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1836 .policy = vport_policy,
1837 .doit = ovs_vport_cmd_new
1839 { .cmd = OVS_VPORT_CMD_DEL,
1840 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1841 .policy = vport_policy,
1842 .doit = ovs_vport_cmd_del
1844 { .cmd = OVS_VPORT_CMD_GET,
1845 .flags = 0, /* OK for unprivileged users. */
1846 .policy = vport_policy,
1847 .doit = ovs_vport_cmd_get,
1848 .dumpit = ovs_vport_cmd_dump
1850 { .cmd = OVS_VPORT_CMD_SET,
1851 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1852 .policy = vport_policy,
1853 .doit = ovs_vport_cmd_set,
1857 struct genl_family_and_ops {
1858 struct genl_family *family;
1859 struct genl_ops *ops;
1861 struct genl_multicast_group *group;
1864 static const struct genl_family_and_ops dp_genl_families[] = {
1865 { &dp_datapath_genl_family,
1866 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1867 &ovs_dp_datapath_multicast_group },
1868 { &dp_vport_genl_family,
1869 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1870 &ovs_dp_vport_multicast_group },
1871 { &dp_flow_genl_family,
1872 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1873 &ovs_dp_flow_multicast_group },
1874 { &dp_packet_genl_family,
1875 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1879 static void dp_unregister_genl(int n_families)
1883 for (i = 0; i < n_families; i++)
1884 genl_unregister_family(dp_genl_families[i].family);
1887 static int dp_register_genl(void)
1894 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1895 const struct genl_family_and_ops *f = &dp_genl_families[i];
1897 err = genl_register_family_with_ops(f->family, f->ops,
1904 err = genl_register_mc_group(f->family, f->group);
1913 dp_unregister_genl(n_registered);
1917 static int __net_init ovs_init_net(struct net *net)
1919 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1921 INIT_LIST_HEAD(&ovs_net->dps);
1922 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1926 static void __net_exit ovs_exit_net(struct net *net)
1928 struct datapath *dp, *dp_next;
1929 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1932 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1936 cancel_work_sync(&ovs_net->dp_notify_work);
1939 static struct pernet_operations ovs_net_ops = {
1940 .init = ovs_init_net,
1941 .exit = ovs_exit_net,
1943 .size = sizeof(struct ovs_net),
1946 DEFINE_COMPAT_PNET_REG_FUNC(device);
1948 static int __init dp_init(void)
1952 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1954 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
1957 err = ovs_flow_init();
1961 err = ovs_vport_init();
1963 goto error_flow_exit;
1965 err = register_pernet_device(&ovs_net_ops);
1967 goto error_vport_exit;
1969 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1971 goto error_netns_exit;
1973 err = dp_register_genl();
1975 goto error_unreg_notifier;
1979 error_unreg_notifier:
1980 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1982 unregister_pernet_device(&ovs_net_ops);
1991 static void dp_cleanup(void)
1993 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1994 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1995 unregister_pernet_device(&ovs_net_ops);
2001 module_init(dp_init);
2002 module_exit(dp_cleanup);
2004 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2005 MODULE_LICENSE("GPL");
2006 MODULE_VERSION(VERSION);