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 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
491 nla_nest_end(user_skb, nla);
493 if (upcall_info->userdata)
494 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
495 nla_len(upcall_info->userdata),
496 nla_data(upcall_info->userdata));
498 /* Only reserve room for attribute header, packet data is added
499 * in skb_zerocopy() */
500 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
504 nla->nla_len = nla_attr_size(skb->len);
506 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
510 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
511 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
512 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
515 memset(skb_put(user_skb, plen), 0, plen);
518 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
520 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
528 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
530 struct ovs_header *ovs_header = info->userhdr;
531 struct nlattr **a = info->attrs;
532 struct sw_flow_actions *acts;
533 struct sk_buff *packet;
534 struct sw_flow *flow;
537 struct vport *input_vport;
542 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
543 !a[OVS_PACKET_ATTR_ACTIONS])
546 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
547 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
551 skb_reserve(packet, NET_IP_ALIGN);
553 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
555 skb_reset_mac_header(packet);
556 eth = eth_hdr(packet);
558 /* Normally, setting the skb 'protocol' field would be handled by a
559 * call to eth_type_trans(), but it assumes there's a sending
560 * device, which we may not have. */
561 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
562 packet->protocol = eth->h_proto;
564 packet->protocol = htons(ETH_P_802_2);
566 /* Build an sw_flow for sending this packet. */
567 flow = ovs_flow_alloc();
572 err = ovs_flow_extract(packet, -1, &flow->key);
576 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
579 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
584 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
585 &flow->key, 0, &acts);
586 rcu_assign_pointer(flow->sf_acts, acts);
590 OVS_CB(packet)->flow = flow;
591 OVS_CB(packet)->pkt_key = &flow->key;
592 packet->priority = flow->key.phy.priority;
593 packet->mark = flow->key.phy.skb_mark;
596 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
601 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
603 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
608 OVS_CB(packet)->input_vport = input_vport;
611 err = ovs_execute_actions(dp, packet, false);
615 ovs_flow_free(flow, false);
621 ovs_flow_free(flow, false);
628 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
629 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
630 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
631 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
634 static struct genl_ops dp_packet_genl_ops[] = {
635 { .cmd = OVS_PACKET_CMD_EXECUTE,
636 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
637 .policy = packet_policy,
638 .doit = ovs_packet_cmd_execute
642 static struct genl_family dp_packet_genl_family = {
643 .id = GENL_ID_GENERATE,
644 .hdrsize = sizeof(struct ovs_header),
645 .name = OVS_PACKET_FAMILY,
646 .version = OVS_PACKET_VERSION,
647 .maxattr = OVS_PACKET_ATTR_MAX,
649 .parallel_ops = true,
650 .ops = dp_packet_genl_ops,
651 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
654 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
655 struct ovs_dp_megaflow_stats *mega_stats)
659 memset(mega_stats, 0, sizeof(*mega_stats));
661 stats->n_flows = ovs_flow_tbl_count(&dp->table);
662 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
664 stats->n_hit = stats->n_missed = stats->n_lost = 0;
666 for_each_possible_cpu(i) {
667 const struct dp_stats_percpu *percpu_stats;
668 struct dp_stats_percpu local_stats;
671 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
674 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
675 local_stats = *percpu_stats;
676 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
678 stats->n_hit += local_stats.n_hit;
679 stats->n_missed += local_stats.n_missed;
680 stats->n_lost += local_stats.n_lost;
681 mega_stats->n_mask_hit += local_stats.n_mask_hit;
685 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
687 return NLMSG_ALIGN(sizeof(struct ovs_header))
688 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
689 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
690 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
691 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
692 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
693 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
696 /* Called with ovs_mutex or RCU read lock. */
697 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
698 struct sk_buff *skb, u32 portid,
699 u32 seq, u32 flags, u8 cmd)
701 const int skb_orig_len = skb->len;
702 struct nlattr *start;
703 struct ovs_flow_stats stats;
706 struct ovs_header *ovs_header;
710 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
714 ovs_header->dp_ifindex = dp_ifindex;
717 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
719 goto nla_put_failure;
721 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
724 nla_nest_end(skb, nla);
726 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
728 goto nla_put_failure;
730 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
734 nla_nest_end(skb, nla);
736 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
739 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
740 goto nla_put_failure;
742 if (stats.n_packets &&
743 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
744 goto nla_put_failure;
746 if ((u8)ntohs(tcp_flags) &&
747 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
748 goto nla_put_failure;
750 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
751 * this is the first flow to be dumped into 'skb'. This is unusual for
752 * Netlink but individual action lists can be longer than
753 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
754 * The userspace caller can always fetch the actions separately if it
755 * really wants them. (Most userspace callers in fact don't care.)
757 * This can only fail for dump operations because the skb is always
758 * properly sized for single flows.
760 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
762 const struct sw_flow_actions *sf_acts;
764 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
765 err = ovs_nla_put_actions(sf_acts->actions,
766 sf_acts->actions_len, skb);
769 nla_nest_end(skb, start);
774 nla_nest_cancel(skb, start);
776 } else if (skb_orig_len)
777 goto nla_put_failure;
779 return genlmsg_end(skb, ovs_header);
784 genlmsg_cancel(skb, ovs_header);
788 /* May not be called with RCU read lock. */
789 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
790 struct genl_info *info,
795 if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
798 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
801 return ERR_PTR(-ENOMEM);
806 /* Called with ovs_mutex. */
807 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
809 struct genl_info *info, u8 cmd,
815 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
817 if (!skb || IS_ERR(skb))
820 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
821 info->snd_portid, info->snd_seq, 0,
827 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
829 struct nlattr **a = info->attrs;
830 struct ovs_header *ovs_header = info->userhdr;
831 struct sw_flow *flow, *new_flow;
832 struct sw_flow_mask mask;
833 struct sk_buff *reply;
835 struct sw_flow_actions *acts;
836 struct sw_flow_match match;
839 /* Must have key and actions. */
841 if (!a[OVS_FLOW_ATTR_KEY])
843 if (!a[OVS_FLOW_ATTR_ACTIONS])
846 /* Most of the time we need to allocate a new flow, do it before
848 new_flow = ovs_flow_alloc();
849 if (IS_ERR(new_flow)) {
850 error = PTR_ERR(new_flow);
855 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
856 error = ovs_nla_get_match(&match,
857 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
861 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
863 /* Validate actions. */
864 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
865 error = PTR_ERR(acts);
869 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
872 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
876 reply = ovs_flow_cmd_alloc_info(acts, info, false);
878 error = PTR_ERR(reply);
883 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
888 /* Check if this is a duplicate flow */
889 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
891 rcu_assign_pointer(new_flow->sf_acts, acts);
893 /* Put flow in bucket. */
894 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
895 if (unlikely(error)) {
900 if (unlikely(reply)) {
901 error = ovs_flow_cmd_fill_info(new_flow,
902 ovs_header->dp_ifindex,
903 reply, info->snd_portid,
910 struct sw_flow_actions *old_acts;
912 /* Bail out if we're not allowed to modify an existing flow.
913 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
914 * because Generic Netlink treats the latter as a dump
915 * request. We also accept NLM_F_EXCL in case that bug ever
918 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
923 /* The unmasked key has to be the same for flow updates. */
924 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
928 /* Update actions. */
929 old_acts = ovsl_dereference(flow->sf_acts);
930 rcu_assign_pointer(flow->sf_acts, acts);
932 if (unlikely(reply)) {
933 error = ovs_flow_cmd_fill_info(flow,
934 ovs_header->dp_ifindex,
935 reply, info->snd_portid,
942 ovs_nla_free_flow_actions(old_acts);
943 ovs_flow_free(new_flow, false);
947 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
956 ovs_flow_free(new_flow, false);
961 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
963 struct nlattr **a = info->attrs;
964 struct ovs_header *ovs_header = info->userhdr;
965 struct sw_flow_key key, masked_key;
966 struct sw_flow *flow;
967 struct sw_flow_mask mask;
968 struct sk_buff *reply = NULL;
970 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
971 struct sw_flow_match match;
976 if (!a[OVS_FLOW_ATTR_KEY])
979 ovs_match_init(&match, &key, &mask);
980 error = ovs_nla_get_match(&match,
981 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
985 /* Validate actions. */
986 if (a[OVS_FLOW_ATTR_ACTIONS]) {
987 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
988 error = PTR_ERR(acts);
992 ovs_flow_mask_key(&masked_key, &key, &mask);
993 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
994 &masked_key, 0, &acts);
996 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1001 /* Can allocate before locking if have acts. */
1003 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1004 if (IS_ERR(reply)) {
1005 error = PTR_ERR(reply);
1006 goto err_kfree_acts;
1011 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1012 if (unlikely(!dp)) {
1014 goto err_unlock_ovs;
1016 /* Check that the flow exists. */
1017 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1018 if (unlikely(!flow)) {
1020 goto err_unlock_ovs;
1022 /* The unmasked key has to be the same for flow updates. */
1023 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
1025 goto err_unlock_ovs;
1027 /* Update actions, if present. */
1029 old_acts = ovsl_dereference(flow->sf_acts);
1030 rcu_assign_pointer(flow->sf_acts, acts);
1032 if (unlikely(reply)) {
1033 error = ovs_flow_cmd_fill_info(flow,
1034 ovs_header->dp_ifindex,
1035 reply, info->snd_portid,
1041 /* Could not alloc without acts before locking. */
1042 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1043 info, OVS_FLOW_CMD_NEW, false);
1044 if (unlikely(IS_ERR(reply))) {
1045 error = PTR_ERR(reply);
1046 goto err_unlock_ovs;
1051 if (a[OVS_FLOW_ATTR_CLEAR])
1052 ovs_flow_stats_clear(flow);
1056 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1058 ovs_nla_free_flow_actions(old_acts);
1070 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1072 struct nlattr **a = info->attrs;
1073 struct ovs_header *ovs_header = info->userhdr;
1074 struct sw_flow_key key;
1075 struct sk_buff *reply;
1076 struct sw_flow *flow;
1077 struct datapath *dp;
1078 struct sw_flow_match match;
1081 if (!a[OVS_FLOW_ATTR_KEY]) {
1082 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1086 ovs_match_init(&match, &key, NULL);
1087 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1092 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1098 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1099 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1104 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1105 OVS_FLOW_CMD_NEW, true);
1106 if (IS_ERR(reply)) {
1107 err = PTR_ERR(reply);
1112 return genlmsg_reply(reply, info);
1118 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1120 struct nlattr **a = info->attrs;
1121 struct ovs_header *ovs_header = info->userhdr;
1122 struct sw_flow_key key;
1123 struct sk_buff *reply;
1124 struct sw_flow *flow;
1125 struct datapath *dp;
1126 struct sw_flow_match match;
1129 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1130 ovs_match_init(&match, &key, NULL);
1131 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1137 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1138 if (unlikely(!dp)) {
1142 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1143 err = ovs_flow_tbl_flush(&dp->table);
1146 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1147 if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1152 ovs_flow_tbl_remove(&dp->table, flow);
1155 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1158 if (likely(reply)) {
1159 if (likely(!IS_ERR(reply))) {
1160 rcu_read_lock(); /* Keep RCU checker happy. */
1161 err = ovs_flow_cmd_fill_info(flow,
1162 ovs_header->dp_ifindex,
1163 reply, info->snd_portid,
1168 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1170 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1171 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1176 ovs_flow_free(flow, true);
1183 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1185 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1186 struct table_instance *ti;
1187 struct datapath *dp;
1190 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1196 ti = rcu_dereference(dp->table.ti);
1198 struct sw_flow *flow;
1201 bucket = cb->args[0];
1203 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1207 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1208 NETLINK_CB(cb->skb).portid,
1209 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1210 OVS_FLOW_CMD_NEW) < 0)
1213 cb->args[0] = bucket;
1220 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1221 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1222 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1223 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1226 static struct genl_ops dp_flow_genl_ops[] = {
1227 { .cmd = OVS_FLOW_CMD_NEW,
1228 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1229 .policy = flow_policy,
1230 .doit = ovs_flow_cmd_new
1232 { .cmd = OVS_FLOW_CMD_DEL,
1233 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1234 .policy = flow_policy,
1235 .doit = ovs_flow_cmd_del
1237 { .cmd = OVS_FLOW_CMD_GET,
1238 .flags = 0, /* OK for unprivileged users. */
1239 .policy = flow_policy,
1240 .doit = ovs_flow_cmd_get,
1241 .dumpit = ovs_flow_cmd_dump
1243 { .cmd = OVS_FLOW_CMD_SET,
1244 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1245 .policy = flow_policy,
1246 .doit = ovs_flow_cmd_set,
1250 static struct genl_family dp_flow_genl_family = {
1251 .id = GENL_ID_GENERATE,
1252 .hdrsize = sizeof(struct ovs_header),
1253 .name = OVS_FLOW_FAMILY,
1254 .version = OVS_FLOW_VERSION,
1255 .maxattr = OVS_FLOW_ATTR_MAX,
1257 .parallel_ops = true,
1258 .ops = dp_flow_genl_ops,
1259 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1260 .mcgrps = &ovs_dp_flow_multicast_group,
1264 static size_t ovs_dp_cmd_msg_size(void)
1266 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1268 msgsize += nla_total_size(IFNAMSIZ);
1269 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1270 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1271 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1276 /* Called with ovs_mutex or RCU read lock. */
1277 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1278 u32 portid, u32 seq, u32 flags, u8 cmd)
1280 struct ovs_header *ovs_header;
1281 struct ovs_dp_stats dp_stats;
1282 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1285 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1290 ovs_header->dp_ifindex = get_dpifindex(dp);
1292 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1294 goto nla_put_failure;
1296 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1297 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1299 goto nla_put_failure;
1301 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1302 sizeof(struct ovs_dp_megaflow_stats),
1303 &dp_megaflow_stats))
1304 goto nla_put_failure;
1306 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1307 goto nla_put_failure;
1309 return genlmsg_end(skb, ovs_header);
1312 genlmsg_cancel(skb, ovs_header);
1317 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1319 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1322 /* Called with rcu_read_lock or ovs_mutex. */
1323 static struct datapath *lookup_datapath(struct net *net,
1324 struct ovs_header *ovs_header,
1325 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1327 struct datapath *dp;
1329 if (!a[OVS_DP_ATTR_NAME])
1330 dp = get_dp(net, ovs_header->dp_ifindex);
1332 struct vport *vport;
1334 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1335 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1337 return dp ? dp : ERR_PTR(-ENODEV);
1340 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1342 struct datapath *dp;
1344 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1348 WARN(dp->user_features, "Dropping previously announced user features\n");
1349 dp->user_features = 0;
1352 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1354 if (a[OVS_DP_ATTR_USER_FEATURES])
1355 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1358 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1360 struct nlattr **a = info->attrs;
1361 struct vport_parms parms;
1362 struct sk_buff *reply;
1363 struct datapath *dp;
1364 struct vport *vport;
1365 struct ovs_net *ovs_net;
1369 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1372 reply = ovs_dp_cmd_alloc_info(info);
1377 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1379 goto err_free_reply;
1381 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1383 /* Allocate table. */
1384 err = ovs_flow_tbl_init(&dp->table);
1388 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1389 if (!dp->stats_percpu) {
1391 goto err_destroy_table;
1394 for_each_possible_cpu(i) {
1395 struct dp_stats_percpu *dpath_stats;
1396 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1397 u64_stats_init(&dpath_stats->sync);
1400 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1404 goto err_destroy_percpu;
1407 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1408 INIT_HLIST_HEAD(&dp->ports[i]);
1410 /* Set up our datapath device. */
1411 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1412 parms.type = OVS_VPORT_TYPE_INTERNAL;
1413 parms.options = NULL;
1415 parms.port_no = OVSP_LOCAL;
1416 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1418 ovs_dp_change(dp, a);
1420 /* So far only local changes have been made, now need the lock. */
1423 vport = new_vport(&parms);
1424 if (IS_ERR(vport)) {
1425 err = PTR_ERR(vport);
1429 if (err == -EEXIST) {
1430 /* An outdated user space instance that does not understand
1431 * the concept of user_features has attempted to create a new
1432 * datapath and is likely to reuse it. Drop all user features.
1434 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1435 ovs_dp_reset_user_features(skb, info);
1438 goto err_destroy_ports_array;
1441 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1442 info->snd_seq, 0, OVS_DP_CMD_NEW);
1445 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1446 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1450 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1453 err_destroy_ports_array:
1457 free_percpu(dp->stats_percpu);
1459 ovs_flow_tbl_destroy(&dp->table);
1461 release_net(ovs_dp_get_net(dp));
1469 /* Called with ovs_mutex. */
1470 static void __dp_destroy(struct datapath *dp)
1474 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1475 struct vport *vport;
1476 struct hlist_node *n;
1478 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1479 if (vport->port_no != OVSP_LOCAL)
1480 ovs_dp_detach_port(vport);
1483 list_del_rcu(&dp->list_node);
1485 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1486 * all ports in datapath are destroyed first before freeing datapath.
1488 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1490 /* RCU destroy the flow table */
1491 call_rcu(&dp->rcu, destroy_dp_rcu);
1494 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1496 struct sk_buff *reply;
1497 struct datapath *dp;
1500 reply = ovs_dp_cmd_alloc_info(info);
1505 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1508 goto err_unlock_free;
1510 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1511 info->snd_seq, 0, OVS_DP_CMD_DEL);
1517 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1526 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1528 struct sk_buff *reply;
1529 struct datapath *dp;
1532 reply = ovs_dp_cmd_alloc_info(info);
1537 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1540 goto err_unlock_free;
1542 ovs_dp_change(dp, info->attrs);
1544 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1545 info->snd_seq, 0, OVS_DP_CMD_NEW);
1549 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1558 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1560 struct sk_buff *reply;
1561 struct datapath *dp;
1564 reply = ovs_dp_cmd_alloc_info(info);
1569 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1572 goto err_unlock_free;
1574 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1575 info->snd_seq, 0, OVS_DP_CMD_NEW);
1579 return genlmsg_reply(reply, info);
1587 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1589 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1590 struct datapath *dp;
1591 int skip = cb->args[0];
1595 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1597 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1598 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1599 OVS_DP_CMD_NEW) < 0)
1610 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1611 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1612 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1613 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1616 static struct genl_ops dp_datapath_genl_ops[] = {
1617 { .cmd = OVS_DP_CMD_NEW,
1618 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1619 .policy = datapath_policy,
1620 .doit = ovs_dp_cmd_new
1622 { .cmd = OVS_DP_CMD_DEL,
1623 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1624 .policy = datapath_policy,
1625 .doit = ovs_dp_cmd_del
1627 { .cmd = OVS_DP_CMD_GET,
1628 .flags = 0, /* OK for unprivileged users. */
1629 .policy = datapath_policy,
1630 .doit = ovs_dp_cmd_get,
1631 .dumpit = ovs_dp_cmd_dump
1633 { .cmd = OVS_DP_CMD_SET,
1634 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1635 .policy = datapath_policy,
1636 .doit = ovs_dp_cmd_set,
1640 static struct genl_family dp_datapath_genl_family = {
1641 .id = GENL_ID_GENERATE,
1642 .hdrsize = sizeof(struct ovs_header),
1643 .name = OVS_DATAPATH_FAMILY,
1644 .version = OVS_DATAPATH_VERSION,
1645 .maxattr = OVS_DP_ATTR_MAX,
1647 .parallel_ops = true,
1648 .ops = dp_datapath_genl_ops,
1649 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1650 .mcgrps = &ovs_dp_datapath_multicast_group,
1654 /* Called with ovs_mutex or RCU read lock. */
1655 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1656 u32 portid, u32 seq, u32 flags, u8 cmd)
1658 struct ovs_header *ovs_header;
1659 struct ovs_vport_stats vport_stats;
1662 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1667 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1669 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1670 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1671 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1672 goto nla_put_failure;
1674 ovs_vport_get_stats(vport, &vport_stats);
1675 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1677 goto nla_put_failure;
1679 if (ovs_vport_get_upcall_portids(vport, skb))
1680 goto nla_put_failure;
1682 err = ovs_vport_get_options(vport, skb);
1683 if (err == -EMSGSIZE)
1686 return genlmsg_end(skb, ovs_header);
1691 genlmsg_cancel(skb, ovs_header);
1695 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1697 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1700 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1701 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1704 struct sk_buff *skb;
1707 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1709 return ERR_PTR(-ENOMEM);
1711 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1717 /* Called with ovs_mutex or RCU read lock. */
1718 static struct vport *lookup_vport(struct net *net,
1719 struct ovs_header *ovs_header,
1720 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1722 struct datapath *dp;
1723 struct vport *vport;
1725 if (a[OVS_VPORT_ATTR_NAME]) {
1726 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1728 return ERR_PTR(-ENODEV);
1729 if (ovs_header->dp_ifindex &&
1730 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1731 return ERR_PTR(-ENODEV);
1733 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1734 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1736 if (port_no >= DP_MAX_PORTS)
1737 return ERR_PTR(-EFBIG);
1739 dp = get_dp(net, ovs_header->dp_ifindex);
1741 return ERR_PTR(-ENODEV);
1743 vport = ovs_vport_ovsl_rcu(dp, port_no);
1745 return ERR_PTR(-ENODEV);
1748 return ERR_PTR(-EINVAL);
1751 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1753 struct nlattr **a = info->attrs;
1754 struct ovs_header *ovs_header = info->userhdr;
1755 struct vport_parms parms;
1756 struct sk_buff *reply;
1757 struct vport *vport;
1758 struct datapath *dp;
1762 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1763 !a[OVS_VPORT_ATTR_UPCALL_PID])
1766 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1767 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1768 if (port_no >= DP_MAX_PORTS)
1771 reply = ovs_vport_cmd_alloc_info();
1776 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1779 goto exit_unlock_free;
1782 vport = ovs_vport_ovsl(dp, port_no);
1785 goto exit_unlock_free;
1787 for (port_no = 1; ; port_no++) {
1788 if (port_no >= DP_MAX_PORTS) {
1790 goto exit_unlock_free;
1792 vport = ovs_vport_ovsl(dp, port_no);
1798 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1799 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1800 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1802 parms.port_no = port_no;
1803 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1805 vport = new_vport(&parms);
1806 err = PTR_ERR(vport);
1808 goto exit_unlock_free;
1811 if (a[OVS_VPORT_ATTR_STATS])
1812 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1814 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1815 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1819 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1828 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1830 struct nlattr **a = info->attrs;
1831 struct sk_buff *reply;
1832 struct vport *vport;
1835 reply = ovs_vport_cmd_alloc_info();
1840 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1841 err = PTR_ERR(vport);
1843 goto exit_unlock_free;
1845 if (a[OVS_VPORT_ATTR_TYPE] &&
1846 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1848 goto exit_unlock_free;
1851 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1852 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1854 goto exit_unlock_free;
1857 if (a[OVS_VPORT_ATTR_STATS])
1858 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1861 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1862 err = ovs_vport_set_upcall_portids(vport,
1863 a[OVS_VPORT_ATTR_UPCALL_PID]);
1865 goto exit_unlock_free;
1868 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1869 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1873 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1882 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1884 struct nlattr **a = info->attrs;
1885 struct sk_buff *reply;
1886 struct vport *vport;
1889 reply = ovs_vport_cmd_alloc_info();
1894 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1895 err = PTR_ERR(vport);
1897 goto exit_unlock_free;
1899 if (vport->port_no == OVSP_LOCAL) {
1901 goto exit_unlock_free;
1904 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1905 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1907 ovs_dp_detach_port(vport);
1910 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1919 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1921 struct nlattr **a = info->attrs;
1922 struct ovs_header *ovs_header = info->userhdr;
1923 struct sk_buff *reply;
1924 struct vport *vport;
1927 reply = ovs_vport_cmd_alloc_info();
1932 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1933 err = PTR_ERR(vport);
1935 goto exit_unlock_free;
1936 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1937 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1941 return genlmsg_reply(reply, info);
1949 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1951 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1952 struct datapath *dp;
1953 int bucket = cb->args[0], skip = cb->args[1];
1957 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1962 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1963 struct vport *vport;
1966 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1968 ovs_vport_cmd_fill_info(vport, skb,
1969 NETLINK_CB(cb->skb).portid,
1972 OVS_VPORT_CMD_NEW) < 0)
1988 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1989 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1990 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1991 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1992 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1993 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1994 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1997 static struct genl_ops dp_vport_genl_ops[] = {
1998 { .cmd = OVS_VPORT_CMD_NEW,
1999 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2000 .policy = vport_policy,
2001 .doit = ovs_vport_cmd_new
2003 { .cmd = OVS_VPORT_CMD_DEL,
2004 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2005 .policy = vport_policy,
2006 .doit = ovs_vport_cmd_del
2008 { .cmd = OVS_VPORT_CMD_GET,
2009 .flags = 0, /* OK for unprivileged users. */
2010 .policy = vport_policy,
2011 .doit = ovs_vport_cmd_get,
2012 .dumpit = ovs_vport_cmd_dump
2014 { .cmd = OVS_VPORT_CMD_SET,
2015 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2016 .policy = vport_policy,
2017 .doit = ovs_vport_cmd_set,
2021 struct genl_family dp_vport_genl_family = {
2022 .id = GENL_ID_GENERATE,
2023 .hdrsize = sizeof(struct ovs_header),
2024 .name = OVS_VPORT_FAMILY,
2025 .version = OVS_VPORT_VERSION,
2026 .maxattr = OVS_VPORT_ATTR_MAX,
2028 .parallel_ops = true,
2029 .ops = dp_vport_genl_ops,
2030 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2031 .mcgrps = &ovs_dp_vport_multicast_group,
2035 static struct genl_family *dp_genl_families[] = {
2036 &dp_datapath_genl_family,
2037 &dp_vport_genl_family,
2038 &dp_flow_genl_family,
2039 &dp_packet_genl_family,
2042 static void dp_unregister_genl(int n_families)
2046 for (i = 0; i < n_families; i++)
2047 genl_unregister_family(dp_genl_families[i]);
2050 static int dp_register_genl(void)
2055 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2057 err = genl_register_family(dp_genl_families[i]);
2065 dp_unregister_genl(i);
2069 static int __net_init ovs_init_net(struct net *net)
2071 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2073 INIT_LIST_HEAD(&ovs_net->dps);
2074 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2078 static void __net_exit ovs_exit_net(struct net *net)
2080 struct datapath *dp, *dp_next;
2081 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2084 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2088 cancel_work_sync(&ovs_net->dp_notify_work);
2091 static struct pernet_operations ovs_net_ops = {
2092 .init = ovs_init_net,
2093 .exit = ovs_exit_net,
2095 .size = sizeof(struct ovs_net),
2098 DEFINE_COMPAT_PNET_REG_FUNC(device);
2100 static int __init dp_init(void)
2104 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2106 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2109 err = ovs_flow_init();
2113 err = ovs_vport_init();
2115 goto error_flow_exit;
2117 err = register_pernet_device(&ovs_net_ops);
2119 goto error_vport_exit;
2121 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2123 goto error_netns_exit;
2125 err = dp_register_genl();
2127 goto error_unreg_notifier;
2131 error_unreg_notifier:
2132 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2134 unregister_pernet_device(&ovs_net_ops);
2143 static void dp_cleanup(void)
2145 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2146 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2147 unregister_pernet_device(&ovs_net_ops);
2153 module_init(dp_init);
2154 module_exit(dp_cleanup);
2156 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2157 MODULE_LICENSE("GPL");
2158 MODULE_VERSION(VERSION);