2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/genetlink.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
59 #include "flow_table.h"
60 #include "flow_netlink.h"
62 #include "vport-internal_dev.h"
63 #include "vport-netdev.h"
65 int ovs_net_id __read_mostly;
67 static struct genl_family dp_packet_genl_family;
68 static struct genl_family dp_flow_genl_family;
69 static struct genl_family dp_datapath_genl_family;
71 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
72 .name = OVS_FLOW_MCGROUP
75 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
76 .name = OVS_DATAPATH_MCGROUP
79 struct genl_multicast_group ovs_dp_vport_multicast_group = {
80 .name = OVS_VPORT_MCGROUP
83 /* Check if need to build a reply message.
84 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
85 static bool ovs_must_notify(struct genl_info *info,
86 const struct genl_multicast_group *grp)
88 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
89 netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
92 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
93 struct sk_buff *skb, struct genl_info *info)
95 genl_notify(family, skb, genl_info_net(info),
96 info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
102 * All writes e.g. Writes to device state (add/remove datapath, port, set
103 * operations on vports, etc.), Writes to other state (flow table
104 * modifications, set miscellaneous datapath parameters, etc.) are protected
107 * Reads are protected by RCU.
109 * There are a few special cases (mostly stats) that have their own
110 * synchronization but they nest under all of above and don't interact with
113 * The RTNL lock nests inside ovs_mutex.
116 static DEFINE_MUTEX(ovs_mutex);
120 mutex_lock(&ovs_mutex);
123 void ovs_unlock(void)
125 mutex_unlock(&ovs_mutex);
128 #ifdef CONFIG_LOCKDEP
129 int lockdep_ovsl_is_held(void)
132 return lockdep_is_held(&ovs_mutex);
138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 const struct dp_upcall_info *);
140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 const struct dp_upcall_info *);
143 /* Must be called with rcu_read_lock or ovs_mutex. */
144 static struct datapath *get_dp(struct net *net, int dp_ifindex)
146 struct datapath *dp = NULL;
147 struct net_device *dev;
150 dev = dev_get_by_index_rcu(net, dp_ifindex);
152 struct vport *vport = ovs_internal_dev_get_vport(dev);
161 /* Must be called with rcu_read_lock or ovs_mutex. */
162 const char *ovs_dp_name(const struct datapath *dp)
164 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
165 return vport->ops->get_name(vport);
168 static int get_dpifindex(struct datapath *dp)
175 local = ovs_vport_rcu(dp, OVSP_LOCAL);
177 ifindex = netdev_vport_priv(local)->dev->ifindex;
186 static void destroy_dp_rcu(struct rcu_head *rcu)
188 struct datapath *dp = container_of(rcu, struct datapath, rcu);
190 ovs_flow_tbl_destroy(&dp->table);
191 free_percpu(dp->stats_percpu);
192 release_net(ovs_dp_get_net(dp));
197 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
200 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
203 /* Called with ovs_mutex or RCU read lock. */
204 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
207 struct hlist_head *head;
209 head = vport_hash_bucket(dp, port_no);
210 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
211 if (vport->port_no == port_no)
217 /* Called with ovs_mutex. */
218 static struct vport *new_vport(const struct vport_parms *parms)
222 vport = ovs_vport_add(parms);
223 if (!IS_ERR(vport)) {
224 struct datapath *dp = parms->dp;
225 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
227 hlist_add_head_rcu(&vport->dp_hash_node, head);
232 void ovs_dp_detach_port(struct vport *p)
236 /* First drop references to device. */
237 hlist_del_rcu(&p->dp_hash_node);
239 /* Then destroy it. */
243 void ovs_dp_process_packet_with_key(struct sk_buff *skb,
244 struct sw_flow_key *pkt_key,
247 const struct vport *p = OVS_CB(skb)->input_vport;
248 struct datapath *dp = p->dp;
249 struct sw_flow *flow;
250 struct dp_stats_percpu *stats;
254 stats = this_cpu_ptr(dp->stats_percpu);
257 flow = ovs_flow_tbl_lookup_stats(&dp->table, pkt_key, skb_get_hash(skb),
259 if (unlikely(!flow)) {
260 struct dp_upcall_info upcall;
262 upcall.cmd = OVS_PACKET_CMD_MISS;
263 upcall.key = pkt_key;
264 upcall.userdata = NULL;
265 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
266 ovs_dp_upcall(dp, skb, &upcall);
268 stats_counter = &stats->n_missed;
272 OVS_CB(skb)->pkt_key = pkt_key;
273 OVS_CB(skb)->flow = flow;
275 ovs_flow_stats_update(OVS_CB(skb)->flow, pkt_key->tp.flags, skb);
276 ovs_execute_actions(dp, skb, recirc);
277 stats_counter = &stats->n_hit;
280 /* Update datapath statistics. */
281 u64_stats_update_begin(&stats->sync);
283 stats->n_mask_hit += n_mask_hit;
284 u64_stats_update_end(&stats->sync);
287 /* Must be called with rcu_read_lock. */
288 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
291 struct sw_flow_key key;
293 OVS_CB(skb)->input_vport = p;
295 /* Extract flow from 'skb' into 'key'. */
296 error = ovs_flow_extract(skb, p->port_no, &key);
297 if (unlikely(error)) {
302 ovs_dp_process_packet_with_key(skb, &key, false);
305 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
306 const struct dp_upcall_info *upcall_info)
308 struct dp_stats_percpu *stats;
311 if (upcall_info->portid == 0) {
316 if (!skb_is_gso(skb))
317 err = queue_userspace_packet(dp, skb, upcall_info);
319 err = queue_gso_packets(dp, skb, upcall_info);
326 stats = this_cpu_ptr(dp->stats_percpu);
328 u64_stats_update_begin(&stats->sync);
330 u64_stats_update_end(&stats->sync);
335 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
336 const struct dp_upcall_info *upcall_info)
338 unsigned short gso_type = skb_shinfo(skb)->gso_type;
339 struct dp_upcall_info later_info;
340 struct sw_flow_key later_key;
341 struct sk_buff *segs, *nskb;
344 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
346 return PTR_ERR(segs);
348 /* Queue all of the segments. */
351 err = queue_userspace_packet(dp, skb, upcall_info);
355 if (skb == segs && gso_type & SKB_GSO_UDP) {
356 /* The initial flow key extracted by ovs_flow_extract()
357 * in this case is for a first fragment, so we need to
358 * properly mark later fragments.
360 later_key = *upcall_info->key;
361 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
363 later_info = *upcall_info;
364 later_info.key = &later_key;
365 upcall_info = &later_info;
367 } while ((skb = skb->next));
369 /* Free all of the segments. */
377 } while ((skb = nskb));
381 static size_t key_attr_size(void)
383 /* Whenever adding new OVS_KEY_ FIELDS, we should consider
384 * updating this function. */
385 BUILD_BUG_ON(OVS_KEY_ATTR_IPV4_TUNNEL != 21);
387 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
388 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
389 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
390 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
391 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
392 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
393 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
394 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
395 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
396 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
397 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
398 + nla_total_size(4) /* OVS_KEY_ATTR_DP_HASH */
399 + nla_total_size(4) /* OVS_KEY_ATTR_RECIRC_ID */
400 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
401 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
402 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
403 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
404 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
405 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
406 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
407 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
410 static size_t upcall_msg_size(const struct nlattr *userdata,
413 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
414 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
415 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
417 /* OVS_PACKET_ATTR_USERDATA */
419 size += NLA_ALIGN(userdata->nla_len);
424 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
425 const struct dp_upcall_info *upcall_info)
427 struct ovs_header *upcall;
428 struct sk_buff *nskb = NULL;
429 struct sk_buff *user_skb; /* to be queued to userspace */
431 struct genl_info info = {
432 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
433 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
435 .snd_portid = upcall_info->portid,
441 dp_ifindex = get_dpifindex(dp);
445 if (vlan_tx_tag_present(skb)) {
446 nskb = skb_clone(skb, GFP_ATOMIC);
450 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
454 vlan_set_tci(nskb, 0);
459 if (nla_attr_size(skb->len) > USHRT_MAX) {
464 /* Complete checksum if needed */
465 if (skb->ip_summed == CHECKSUM_PARTIAL &&
466 (err = skb_checksum_help(skb)))
469 /* Older versions of OVS user space enforce alignment of the last
470 * Netlink attribute to NLA_ALIGNTO which would require extensive
471 * padding logic. Only perform zerocopy if padding is not required.
473 if (dp->user_features & OVS_DP_F_UNALIGNED)
474 hlen = skb_zerocopy_headlen(skb);
478 len = upcall_msg_size(upcall_info->userdata, hlen);
479 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
485 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
486 0, upcall_info->cmd);
487 upcall->dp_ifindex = dp_ifindex;
489 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
490 err = ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
492 nla_nest_end(user_skb, nla);
494 if (upcall_info->userdata)
495 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
496 nla_len(upcall_info->userdata),
497 nla_data(upcall_info->userdata));
499 /* Only reserve room for attribute header, packet data is added
500 * in skb_zerocopy() */
501 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
505 nla->nla_len = nla_attr_size(skb->len);
507 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
511 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
512 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
513 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
516 memset(skb_put(user_skb, plen), 0, plen);
519 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
521 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
529 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
531 struct ovs_header *ovs_header = info->userhdr;
532 struct nlattr **a = info->attrs;
533 struct sw_flow_actions *acts;
534 struct sk_buff *packet;
535 struct sw_flow *flow;
538 struct vport *input_vport;
543 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
544 !a[OVS_PACKET_ATTR_ACTIONS])
547 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
548 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
552 skb_reserve(packet, NET_IP_ALIGN);
554 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
556 skb_reset_mac_header(packet);
557 eth = eth_hdr(packet);
559 /* Normally, setting the skb 'protocol' field would be handled by a
560 * call to eth_type_trans(), but it assumes there's a sending
561 * device, which we may not have. */
562 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
563 packet->protocol = eth->h_proto;
565 packet->protocol = htons(ETH_P_802_2);
567 /* Build an sw_flow for sending this packet. */
568 flow = ovs_flow_alloc();
573 err = ovs_flow_extract(packet, -1, &flow->key);
577 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
580 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
585 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
586 &flow->key, 0, &acts);
587 rcu_assign_pointer(flow->sf_acts, acts);
591 OVS_CB(packet)->flow = flow;
592 OVS_CB(packet)->pkt_key = &flow->key;
593 packet->priority = flow->key.phy.priority;
594 packet->mark = flow->key.phy.skb_mark;
597 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
602 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
604 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
609 OVS_CB(packet)->input_vport = input_vport;
612 err = ovs_execute_actions(dp, packet, false);
616 ovs_flow_free(flow, false);
622 ovs_flow_free(flow, false);
629 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
630 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
631 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
632 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
635 static struct genl_ops dp_packet_genl_ops[] = {
636 { .cmd = OVS_PACKET_CMD_EXECUTE,
637 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
638 .policy = packet_policy,
639 .doit = ovs_packet_cmd_execute
643 static struct genl_family dp_packet_genl_family = {
644 .id = GENL_ID_GENERATE,
645 .hdrsize = sizeof(struct ovs_header),
646 .name = OVS_PACKET_FAMILY,
647 .version = OVS_PACKET_VERSION,
648 .maxattr = OVS_PACKET_ATTR_MAX,
650 .parallel_ops = true,
651 .ops = dp_packet_genl_ops,
652 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
655 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
656 struct ovs_dp_megaflow_stats *mega_stats)
660 memset(mega_stats, 0, sizeof(*mega_stats));
662 stats->n_flows = ovs_flow_tbl_count(&dp->table);
663 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
665 stats->n_hit = stats->n_missed = stats->n_lost = 0;
667 for_each_possible_cpu(i) {
668 const struct dp_stats_percpu *percpu_stats;
669 struct dp_stats_percpu local_stats;
672 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
675 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
676 local_stats = *percpu_stats;
677 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
679 stats->n_hit += local_stats.n_hit;
680 stats->n_missed += local_stats.n_missed;
681 stats->n_lost += local_stats.n_lost;
682 mega_stats->n_mask_hit += local_stats.n_mask_hit;
686 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
688 return NLMSG_ALIGN(sizeof(struct ovs_header))
689 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
690 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
691 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
692 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
693 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
694 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
697 /* Called with ovs_mutex or RCU read lock. */
698 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
699 struct sk_buff *skb, u32 portid,
700 u32 seq, u32 flags, u8 cmd)
702 const int skb_orig_len = skb->len;
703 struct nlattr *start;
704 struct ovs_flow_stats stats;
707 struct ovs_header *ovs_header;
711 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
715 ovs_header->dp_ifindex = dp_ifindex;
718 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
720 goto nla_put_failure;
722 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
725 nla_nest_end(skb, nla);
727 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
729 goto nla_put_failure;
731 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
735 nla_nest_end(skb, nla);
737 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
740 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
741 goto nla_put_failure;
743 if (stats.n_packets &&
744 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
745 goto nla_put_failure;
747 if ((u8)ntohs(tcp_flags) &&
748 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
749 goto nla_put_failure;
751 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
752 * this is the first flow to be dumped into 'skb'. This is unusual for
753 * Netlink but individual action lists can be longer than
754 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
755 * The userspace caller can always fetch the actions separately if it
756 * really wants them. (Most userspace callers in fact don't care.)
758 * This can only fail for dump operations because the skb is always
759 * properly sized for single flows.
761 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
763 const struct sw_flow_actions *sf_acts;
765 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
766 err = ovs_nla_put_actions(sf_acts->actions,
767 sf_acts->actions_len, skb);
770 nla_nest_end(skb, start);
775 nla_nest_cancel(skb, start);
777 } else if (skb_orig_len)
778 goto nla_put_failure;
780 return genlmsg_end(skb, ovs_header);
785 genlmsg_cancel(skb, ovs_header);
789 /* May not be called with RCU read lock. */
790 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
791 struct genl_info *info,
796 if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
799 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
802 return ERR_PTR(-ENOMEM);
807 /* Called with ovs_mutex. */
808 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
810 struct genl_info *info, u8 cmd,
816 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
818 if (!skb || IS_ERR(skb))
821 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
822 info->snd_portid, info->snd_seq, 0,
828 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
830 struct nlattr **a = info->attrs;
831 struct ovs_header *ovs_header = info->userhdr;
832 struct sw_flow *flow, *new_flow;
833 struct sw_flow_mask mask;
834 struct sk_buff *reply;
836 struct sw_flow_actions *acts;
837 struct sw_flow_match match;
840 /* Must have key and actions. */
842 if (!a[OVS_FLOW_ATTR_KEY])
844 if (!a[OVS_FLOW_ATTR_ACTIONS])
847 /* Most of the time we need to allocate a new flow, do it before
849 new_flow = ovs_flow_alloc();
850 if (IS_ERR(new_flow)) {
851 error = PTR_ERR(new_flow);
856 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
857 error = ovs_nla_get_match(&match,
858 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
862 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
864 /* Validate actions. */
865 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
866 error = PTR_ERR(acts);
870 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
873 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
877 reply = ovs_flow_cmd_alloc_info(acts, info, false);
879 error = PTR_ERR(reply);
884 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
889 /* Check if this is a duplicate flow */
890 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
892 rcu_assign_pointer(new_flow->sf_acts, acts);
894 /* Put flow in bucket. */
895 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
896 if (unlikely(error)) {
901 if (unlikely(reply)) {
902 error = ovs_flow_cmd_fill_info(new_flow,
903 ovs_header->dp_ifindex,
904 reply, info->snd_portid,
911 struct sw_flow_actions *old_acts;
913 /* Bail out if we're not allowed to modify an existing flow.
914 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
915 * because Generic Netlink treats the latter as a dump
916 * request. We also accept NLM_F_EXCL in case that bug ever
919 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
924 /* The unmasked key has to be the same for flow updates. */
925 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
929 /* Update actions. */
930 old_acts = ovsl_dereference(flow->sf_acts);
931 rcu_assign_pointer(flow->sf_acts, acts);
933 if (unlikely(reply)) {
934 error = ovs_flow_cmd_fill_info(flow,
935 ovs_header->dp_ifindex,
936 reply, info->snd_portid,
943 ovs_nla_free_flow_actions(old_acts);
944 ovs_flow_free(new_flow, false);
948 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
957 ovs_flow_free(new_flow, false);
962 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
964 struct nlattr **a = info->attrs;
965 struct ovs_header *ovs_header = info->userhdr;
966 struct sw_flow_key key, masked_key;
967 struct sw_flow *flow;
968 struct sw_flow_mask mask;
969 struct sk_buff *reply = NULL;
971 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
972 struct sw_flow_match match;
977 if (!a[OVS_FLOW_ATTR_KEY])
980 ovs_match_init(&match, &key, &mask);
981 error = ovs_nla_get_match(&match,
982 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
986 /* Validate actions. */
987 if (a[OVS_FLOW_ATTR_ACTIONS]) {
988 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
989 error = PTR_ERR(acts);
993 ovs_flow_mask_key(&masked_key, &key, &mask);
994 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
995 &masked_key, 0, &acts);
997 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1002 /* Can allocate before locking if have acts. */
1004 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1005 if (IS_ERR(reply)) {
1006 error = PTR_ERR(reply);
1007 goto err_kfree_acts;
1012 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1013 if (unlikely(!dp)) {
1015 goto err_unlock_ovs;
1017 /* Check that the flow exists. */
1018 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1019 if (unlikely(!flow)) {
1021 goto err_unlock_ovs;
1023 /* The unmasked key has to be the same for flow updates. */
1024 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
1026 goto err_unlock_ovs;
1028 /* Update actions, if present. */
1030 old_acts = ovsl_dereference(flow->sf_acts);
1031 rcu_assign_pointer(flow->sf_acts, acts);
1033 if (unlikely(reply)) {
1034 error = ovs_flow_cmd_fill_info(flow,
1035 ovs_header->dp_ifindex,
1036 reply, info->snd_portid,
1042 /* Could not alloc without acts before locking. */
1043 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1044 info, OVS_FLOW_CMD_NEW, false);
1045 if (unlikely(IS_ERR(reply))) {
1046 error = PTR_ERR(reply);
1047 goto err_unlock_ovs;
1052 if (a[OVS_FLOW_ATTR_CLEAR])
1053 ovs_flow_stats_clear(flow);
1057 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1059 ovs_nla_free_flow_actions(old_acts);
1071 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1073 struct nlattr **a = info->attrs;
1074 struct ovs_header *ovs_header = info->userhdr;
1075 struct sw_flow_key key;
1076 struct sk_buff *reply;
1077 struct sw_flow *flow;
1078 struct datapath *dp;
1079 struct sw_flow_match match;
1082 if (!a[OVS_FLOW_ATTR_KEY]) {
1083 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1087 ovs_match_init(&match, &key, NULL);
1088 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1093 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1099 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1100 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1105 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1106 OVS_FLOW_CMD_NEW, true);
1107 if (IS_ERR(reply)) {
1108 err = PTR_ERR(reply);
1113 return genlmsg_reply(reply, info);
1119 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1121 struct nlattr **a = info->attrs;
1122 struct ovs_header *ovs_header = info->userhdr;
1123 struct sw_flow_key key;
1124 struct sk_buff *reply;
1125 struct sw_flow *flow;
1126 struct datapath *dp;
1127 struct sw_flow_match match;
1130 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1131 ovs_match_init(&match, &key, NULL);
1132 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1138 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1139 if (unlikely(!dp)) {
1143 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1144 err = ovs_flow_tbl_flush(&dp->table);
1147 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1148 if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1153 ovs_flow_tbl_remove(&dp->table, flow);
1156 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1159 if (likely(reply)) {
1160 if (likely(!IS_ERR(reply))) {
1161 rcu_read_lock(); /* Keep RCU checker happy. */
1162 err = ovs_flow_cmd_fill_info(flow,
1163 ovs_header->dp_ifindex,
1164 reply, info->snd_portid,
1169 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1171 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1172 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1177 ovs_flow_free(flow, true);
1184 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1186 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1187 struct table_instance *ti;
1188 struct datapath *dp;
1191 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1197 ti = rcu_dereference(dp->table.ti);
1199 struct sw_flow *flow;
1202 bucket = cb->args[0];
1204 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1208 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1209 NETLINK_CB(cb->skb).portid,
1210 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1211 OVS_FLOW_CMD_NEW) < 0)
1214 cb->args[0] = bucket;
1221 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1222 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1223 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1224 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1227 static struct genl_ops dp_flow_genl_ops[] = {
1228 { .cmd = OVS_FLOW_CMD_NEW,
1229 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1230 .policy = flow_policy,
1231 .doit = ovs_flow_cmd_new
1233 { .cmd = OVS_FLOW_CMD_DEL,
1234 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1235 .policy = flow_policy,
1236 .doit = ovs_flow_cmd_del
1238 { .cmd = OVS_FLOW_CMD_GET,
1239 .flags = 0, /* OK for unprivileged users. */
1240 .policy = flow_policy,
1241 .doit = ovs_flow_cmd_get,
1242 .dumpit = ovs_flow_cmd_dump
1244 { .cmd = OVS_FLOW_CMD_SET,
1245 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1246 .policy = flow_policy,
1247 .doit = ovs_flow_cmd_set,
1251 static struct genl_family dp_flow_genl_family = {
1252 .id = GENL_ID_GENERATE,
1253 .hdrsize = sizeof(struct ovs_header),
1254 .name = OVS_FLOW_FAMILY,
1255 .version = OVS_FLOW_VERSION,
1256 .maxattr = OVS_FLOW_ATTR_MAX,
1258 .parallel_ops = true,
1259 .ops = dp_flow_genl_ops,
1260 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1261 .mcgrps = &ovs_dp_flow_multicast_group,
1265 static size_t ovs_dp_cmd_msg_size(void)
1267 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1269 msgsize += nla_total_size(IFNAMSIZ);
1270 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1271 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1272 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1277 /* Called with ovs_mutex or RCU read lock. */
1278 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1279 u32 portid, u32 seq, u32 flags, u8 cmd)
1281 struct ovs_header *ovs_header;
1282 struct ovs_dp_stats dp_stats;
1283 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1286 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1291 ovs_header->dp_ifindex = get_dpifindex(dp);
1293 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1295 goto nla_put_failure;
1297 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1298 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1300 goto nla_put_failure;
1302 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1303 sizeof(struct ovs_dp_megaflow_stats),
1304 &dp_megaflow_stats))
1305 goto nla_put_failure;
1307 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1308 goto nla_put_failure;
1310 return genlmsg_end(skb, ovs_header);
1313 genlmsg_cancel(skb, ovs_header);
1318 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1320 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1323 /* Called with rcu_read_lock or ovs_mutex. */
1324 static struct datapath *lookup_datapath(struct net *net,
1325 struct ovs_header *ovs_header,
1326 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1328 struct datapath *dp;
1330 if (!a[OVS_DP_ATTR_NAME])
1331 dp = get_dp(net, ovs_header->dp_ifindex);
1333 struct vport *vport;
1335 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1336 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1338 return dp ? dp : ERR_PTR(-ENODEV);
1341 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1343 struct datapath *dp;
1345 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1349 WARN(dp->user_features, "Dropping previously announced user features\n");
1350 dp->user_features = 0;
1353 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1355 if (a[OVS_DP_ATTR_USER_FEATURES])
1356 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1359 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1361 struct nlattr **a = info->attrs;
1362 struct vport_parms parms;
1363 struct sk_buff *reply;
1364 struct datapath *dp;
1365 struct vport *vport;
1366 struct ovs_net *ovs_net;
1370 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1373 reply = ovs_dp_cmd_alloc_info(info);
1378 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1380 goto err_free_reply;
1382 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1384 /* Allocate table. */
1385 err = ovs_flow_tbl_init(&dp->table);
1389 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1390 if (!dp->stats_percpu) {
1392 goto err_destroy_table;
1395 for_each_possible_cpu(i) {
1396 struct dp_stats_percpu *dpath_stats;
1397 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1398 u64_stats_init(&dpath_stats->sync);
1401 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1405 goto err_destroy_percpu;
1408 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1409 INIT_HLIST_HEAD(&dp->ports[i]);
1411 /* Set up our datapath device. */
1412 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1413 parms.type = OVS_VPORT_TYPE_INTERNAL;
1414 parms.options = NULL;
1416 parms.port_no = OVSP_LOCAL;
1417 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1419 ovs_dp_change(dp, a);
1421 /* So far only local changes have been made, now need the lock. */
1424 vport = new_vport(&parms);
1425 if (IS_ERR(vport)) {
1426 err = PTR_ERR(vport);
1430 if (err == -EEXIST) {
1431 /* An outdated user space instance that does not understand
1432 * the concept of user_features has attempted to create a new
1433 * datapath and is likely to reuse it. Drop all user features.
1435 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1436 ovs_dp_reset_user_features(skb, info);
1439 goto err_destroy_ports_array;
1442 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1443 info->snd_seq, 0, OVS_DP_CMD_NEW);
1446 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1447 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1451 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1454 err_destroy_ports_array:
1458 free_percpu(dp->stats_percpu);
1460 ovs_flow_tbl_destroy(&dp->table);
1462 release_net(ovs_dp_get_net(dp));
1470 /* Called with ovs_mutex. */
1471 static void __dp_destroy(struct datapath *dp)
1475 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1476 struct vport *vport;
1477 struct hlist_node *n;
1479 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1480 if (vport->port_no != OVSP_LOCAL)
1481 ovs_dp_detach_port(vport);
1484 list_del_rcu(&dp->list_node);
1486 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1487 * all ports in datapath are destroyed first before freeing datapath.
1489 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1491 /* RCU destroy the flow table */
1492 call_rcu(&dp->rcu, destroy_dp_rcu);
1495 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1497 struct sk_buff *reply;
1498 struct datapath *dp;
1501 reply = ovs_dp_cmd_alloc_info(info);
1506 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1509 goto err_unlock_free;
1511 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1512 info->snd_seq, 0, OVS_DP_CMD_DEL);
1518 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1527 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1529 struct sk_buff *reply;
1530 struct datapath *dp;
1533 reply = ovs_dp_cmd_alloc_info(info);
1538 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1541 goto err_unlock_free;
1543 ovs_dp_change(dp, info->attrs);
1545 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1546 info->snd_seq, 0, OVS_DP_CMD_NEW);
1550 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1559 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1561 struct sk_buff *reply;
1562 struct datapath *dp;
1565 reply = ovs_dp_cmd_alloc_info(info);
1570 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1573 goto err_unlock_free;
1575 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1576 info->snd_seq, 0, OVS_DP_CMD_NEW);
1580 return genlmsg_reply(reply, info);
1588 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1590 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1591 struct datapath *dp;
1592 int skip = cb->args[0];
1596 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1598 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1599 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1600 OVS_DP_CMD_NEW) < 0)
1611 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1612 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1613 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1614 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1617 static struct genl_ops dp_datapath_genl_ops[] = {
1618 { .cmd = OVS_DP_CMD_NEW,
1619 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1620 .policy = datapath_policy,
1621 .doit = ovs_dp_cmd_new
1623 { .cmd = OVS_DP_CMD_DEL,
1624 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1625 .policy = datapath_policy,
1626 .doit = ovs_dp_cmd_del
1628 { .cmd = OVS_DP_CMD_GET,
1629 .flags = 0, /* OK for unprivileged users. */
1630 .policy = datapath_policy,
1631 .doit = ovs_dp_cmd_get,
1632 .dumpit = ovs_dp_cmd_dump
1634 { .cmd = OVS_DP_CMD_SET,
1635 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1636 .policy = datapath_policy,
1637 .doit = ovs_dp_cmd_set,
1641 static struct genl_family dp_datapath_genl_family = {
1642 .id = GENL_ID_GENERATE,
1643 .hdrsize = sizeof(struct ovs_header),
1644 .name = OVS_DATAPATH_FAMILY,
1645 .version = OVS_DATAPATH_VERSION,
1646 .maxattr = OVS_DP_ATTR_MAX,
1648 .parallel_ops = true,
1649 .ops = dp_datapath_genl_ops,
1650 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1651 .mcgrps = &ovs_dp_datapath_multicast_group,
1655 /* Called with ovs_mutex or RCU read lock. */
1656 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1657 u32 portid, u32 seq, u32 flags, u8 cmd)
1659 struct ovs_header *ovs_header;
1660 struct ovs_vport_stats vport_stats;
1663 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1668 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1670 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1671 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1672 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1673 goto nla_put_failure;
1675 ovs_vport_get_stats(vport, &vport_stats);
1676 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1678 goto nla_put_failure;
1680 if (ovs_vport_get_upcall_portids(vport, skb))
1681 goto nla_put_failure;
1683 err = ovs_vport_get_options(vport, skb);
1684 if (err == -EMSGSIZE)
1687 return genlmsg_end(skb, ovs_header);
1692 genlmsg_cancel(skb, ovs_header);
1696 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1698 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1701 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1702 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1705 struct sk_buff *skb;
1708 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1710 return ERR_PTR(-ENOMEM);
1712 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1718 /* Called with ovs_mutex or RCU read lock. */
1719 static struct vport *lookup_vport(struct net *net,
1720 struct ovs_header *ovs_header,
1721 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1723 struct datapath *dp;
1724 struct vport *vport;
1726 if (a[OVS_VPORT_ATTR_NAME]) {
1727 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1729 return ERR_PTR(-ENODEV);
1730 if (ovs_header->dp_ifindex &&
1731 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1732 return ERR_PTR(-ENODEV);
1734 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1735 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1737 if (port_no >= DP_MAX_PORTS)
1738 return ERR_PTR(-EFBIG);
1740 dp = get_dp(net, ovs_header->dp_ifindex);
1742 return ERR_PTR(-ENODEV);
1744 vport = ovs_vport_ovsl_rcu(dp, port_no);
1746 return ERR_PTR(-ENODEV);
1749 return ERR_PTR(-EINVAL);
1752 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1754 struct nlattr **a = info->attrs;
1755 struct ovs_header *ovs_header = info->userhdr;
1756 struct vport_parms parms;
1757 struct sk_buff *reply;
1758 struct vport *vport;
1759 struct datapath *dp;
1763 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1764 !a[OVS_VPORT_ATTR_UPCALL_PID])
1767 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1768 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1769 if (port_no >= DP_MAX_PORTS)
1772 reply = ovs_vport_cmd_alloc_info();
1777 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1780 goto exit_unlock_free;
1783 vport = ovs_vport_ovsl(dp, port_no);
1786 goto exit_unlock_free;
1788 for (port_no = 1; ; port_no++) {
1789 if (port_no >= DP_MAX_PORTS) {
1791 goto exit_unlock_free;
1793 vport = ovs_vport_ovsl(dp, port_no);
1799 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1800 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1801 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1803 parms.port_no = port_no;
1804 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1806 vport = new_vport(&parms);
1807 err = PTR_ERR(vport);
1809 goto exit_unlock_free;
1812 if (a[OVS_VPORT_ATTR_STATS])
1813 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1815 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1816 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1820 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1829 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1831 struct nlattr **a = info->attrs;
1832 struct sk_buff *reply;
1833 struct vport *vport;
1836 reply = ovs_vport_cmd_alloc_info();
1841 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1842 err = PTR_ERR(vport);
1844 goto exit_unlock_free;
1846 if (a[OVS_VPORT_ATTR_TYPE] &&
1847 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1849 goto exit_unlock_free;
1852 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1853 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1855 goto exit_unlock_free;
1858 if (a[OVS_VPORT_ATTR_STATS])
1859 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1862 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1863 err = ovs_vport_set_upcall_portids(vport,
1864 a[OVS_VPORT_ATTR_UPCALL_PID]);
1866 goto exit_unlock_free;
1869 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1870 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1874 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1883 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1885 struct nlattr **a = info->attrs;
1886 struct sk_buff *reply;
1887 struct vport *vport;
1890 reply = ovs_vport_cmd_alloc_info();
1895 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1896 err = PTR_ERR(vport);
1898 goto exit_unlock_free;
1900 if (vport->port_no == OVSP_LOCAL) {
1902 goto exit_unlock_free;
1905 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1906 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1908 ovs_dp_detach_port(vport);
1911 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1920 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1922 struct nlattr **a = info->attrs;
1923 struct ovs_header *ovs_header = info->userhdr;
1924 struct sk_buff *reply;
1925 struct vport *vport;
1928 reply = ovs_vport_cmd_alloc_info();
1933 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1934 err = PTR_ERR(vport);
1936 goto exit_unlock_free;
1937 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1938 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1942 return genlmsg_reply(reply, info);
1950 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1952 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1953 struct datapath *dp;
1954 int bucket = cb->args[0], skip = cb->args[1];
1958 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1963 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1964 struct vport *vport;
1967 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1969 ovs_vport_cmd_fill_info(vport, skb,
1970 NETLINK_CB(cb->skb).portid,
1973 OVS_VPORT_CMD_NEW) < 0)
1989 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1990 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1991 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1992 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1993 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1994 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1995 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1998 static struct genl_ops dp_vport_genl_ops[] = {
1999 { .cmd = OVS_VPORT_CMD_NEW,
2000 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2001 .policy = vport_policy,
2002 .doit = ovs_vport_cmd_new
2004 { .cmd = OVS_VPORT_CMD_DEL,
2005 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2006 .policy = vport_policy,
2007 .doit = ovs_vport_cmd_del
2009 { .cmd = OVS_VPORT_CMD_GET,
2010 .flags = 0, /* OK for unprivileged users. */
2011 .policy = vport_policy,
2012 .doit = ovs_vport_cmd_get,
2013 .dumpit = ovs_vport_cmd_dump
2015 { .cmd = OVS_VPORT_CMD_SET,
2016 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2017 .policy = vport_policy,
2018 .doit = ovs_vport_cmd_set,
2022 struct genl_family dp_vport_genl_family = {
2023 .id = GENL_ID_GENERATE,
2024 .hdrsize = sizeof(struct ovs_header),
2025 .name = OVS_VPORT_FAMILY,
2026 .version = OVS_VPORT_VERSION,
2027 .maxattr = OVS_VPORT_ATTR_MAX,
2029 .parallel_ops = true,
2030 .ops = dp_vport_genl_ops,
2031 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2032 .mcgrps = &ovs_dp_vport_multicast_group,
2036 static struct genl_family *dp_genl_families[] = {
2037 &dp_datapath_genl_family,
2038 &dp_vport_genl_family,
2039 &dp_flow_genl_family,
2040 &dp_packet_genl_family,
2043 static void dp_unregister_genl(int n_families)
2047 for (i = 0; i < n_families; i++)
2048 genl_unregister_family(dp_genl_families[i]);
2051 static int dp_register_genl(void)
2056 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2058 err = genl_register_family(dp_genl_families[i]);
2066 dp_unregister_genl(i);
2070 static int __net_init ovs_init_net(struct net *net)
2072 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2074 INIT_LIST_HEAD(&ovs_net->dps);
2075 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2079 static void __net_exit ovs_exit_net(struct net *net)
2081 struct datapath *dp, *dp_next;
2082 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2085 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2089 cancel_work_sync(&ovs_net->dp_notify_work);
2092 static struct pernet_operations ovs_net_ops = {
2093 .init = ovs_init_net,
2094 .exit = ovs_exit_net,
2096 .size = sizeof(struct ovs_net),
2099 DEFINE_COMPAT_PNET_REG_FUNC(device);
2101 static int __init dp_init(void)
2105 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2107 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2110 err = ovs_flow_init();
2114 err = ovs_vport_init();
2116 goto error_flow_exit;
2118 err = register_pernet_device(&ovs_net_ops);
2120 goto error_vport_exit;
2122 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2124 goto error_netns_exit;
2126 err = dp_register_genl();
2128 goto error_unreg_notifier;
2132 error_unreg_notifier:
2133 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2135 unregister_pernet_device(&ovs_net_ops);
2144 static void dp_cleanup(void)
2146 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2147 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2148 unregister_pernet_device(&ovs_net_ops);
2154 module_init(dp_init);
2155 module_exit(dp_cleanup);
2157 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2158 MODULE_LICENSE("GPL");
2159 MODULE_VERSION(VERSION);