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 free_percpu(dp->stats_percpu);
191 release_net(ovs_dp_get_net(dp));
196 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
199 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
202 /* Called with ovs_mutex or RCU read lock. */
203 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
206 struct hlist_head *head;
208 head = vport_hash_bucket(dp, port_no);
209 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
210 if (vport->port_no == port_no)
216 /* Called with ovs_mutex. */
217 static struct vport *new_vport(const struct vport_parms *parms)
221 vport = ovs_vport_add(parms);
222 if (!IS_ERR(vport)) {
223 struct datapath *dp = parms->dp;
224 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
226 hlist_add_head_rcu(&vport->dp_hash_node, head);
231 void ovs_dp_detach_port(struct vport *p)
235 /* First drop references to device. */
236 hlist_del_rcu(&p->dp_hash_node);
238 /* Then destroy it. */
242 /* Must be called with rcu_read_lock. */
243 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
245 struct datapath *dp = p->dp;
246 struct sw_flow *flow;
247 struct dp_stats_percpu *stats;
248 struct sw_flow_key key;
253 stats = this_cpu_ptr(dp->stats_percpu);
255 /* Extract flow from 'skb' into 'key'. */
256 error = ovs_flow_extract(skb, p->port_no, &key);
257 if (unlikely(error)) {
263 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
264 if (unlikely(!flow)) {
265 struct dp_upcall_info upcall;
267 upcall.cmd = OVS_PACKET_CMD_MISS;
269 upcall.userdata = NULL;
270 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
271 ovs_dp_upcall(dp, skb, &upcall);
273 stats_counter = &stats->n_missed;
277 OVS_CB(skb)->flow = flow;
278 OVS_CB(skb)->pkt_key = &key;
280 ovs_flow_stats_update(OVS_CB(skb)->flow, key.tp.flags, skb);
281 ovs_execute_actions(dp, skb);
282 stats_counter = &stats->n_hit;
285 /* Update datapath statistics. */
286 u64_stats_update_begin(&stats->sync);
288 stats->n_mask_hit += n_mask_hit;
289 u64_stats_update_end(&stats->sync);
292 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
293 const struct dp_upcall_info *upcall_info)
295 struct dp_stats_percpu *stats;
298 if (upcall_info->portid == 0) {
303 if (!skb_is_gso(skb))
304 err = queue_userspace_packet(dp, skb, upcall_info);
306 err = queue_gso_packets(dp, skb, upcall_info);
313 stats = this_cpu_ptr(dp->stats_percpu);
315 u64_stats_update_begin(&stats->sync);
317 u64_stats_update_end(&stats->sync);
322 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
323 const struct dp_upcall_info *upcall_info)
325 unsigned short gso_type = skb_shinfo(skb)->gso_type;
326 struct dp_upcall_info later_info;
327 struct sw_flow_key later_key;
328 struct sk_buff *segs, *nskb;
331 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
333 return PTR_ERR(segs);
335 /* Queue all of the segments. */
338 err = queue_userspace_packet(dp, skb, upcall_info);
342 if (skb == segs && gso_type & SKB_GSO_UDP) {
343 /* The initial flow key extracted by ovs_flow_extract()
344 * in this case is for a first fragment, so we need to
345 * properly mark later fragments.
347 later_key = *upcall_info->key;
348 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
350 later_info = *upcall_info;
351 later_info.key = &later_key;
352 upcall_info = &later_info;
354 } while ((skb = skb->next));
356 /* Free all of the segments. */
364 } while ((skb = nskb));
368 static size_t key_attr_size(void)
370 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
371 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
372 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
373 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
374 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
375 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
376 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
377 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
378 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
379 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
380 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
381 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
382 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
383 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
384 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
385 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
386 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
387 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
388 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
391 static size_t upcall_msg_size(const struct nlattr *userdata,
394 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
395 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
396 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
398 /* OVS_PACKET_ATTR_USERDATA */
400 size += NLA_ALIGN(userdata->nla_len);
405 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
406 const struct dp_upcall_info *upcall_info)
408 struct ovs_header *upcall;
409 struct sk_buff *nskb = NULL;
410 struct sk_buff *user_skb; /* to be queued to userspace */
412 struct genl_info info = {
413 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
414 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
416 .snd_portid = upcall_info->portid,
422 dp_ifindex = get_dpifindex(dp);
426 if (vlan_tx_tag_present(skb)) {
427 nskb = skb_clone(skb, GFP_ATOMIC);
431 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
435 vlan_set_tci(nskb, 0);
440 if (nla_attr_size(skb->len) > USHRT_MAX) {
445 /* Complete checksum if needed */
446 if (skb->ip_summed == CHECKSUM_PARTIAL &&
447 (err = skb_checksum_help(skb)))
450 /* Older versions of OVS user space enforce alignment of the last
451 * Netlink attribute to NLA_ALIGNTO which would require extensive
452 * padding logic. Only perform zerocopy if padding is not required.
454 if (dp->user_features & OVS_DP_F_UNALIGNED)
455 hlen = skb_zerocopy_headlen(skb);
459 len = upcall_msg_size(upcall_info->userdata, hlen);
460 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
466 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
467 0, upcall_info->cmd);
468 upcall->dp_ifindex = dp_ifindex;
470 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
471 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
472 nla_nest_end(user_skb, nla);
474 if (upcall_info->userdata)
475 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
476 nla_len(upcall_info->userdata),
477 nla_data(upcall_info->userdata));
479 /* Only reserve room for attribute header, packet data is added
480 * in skb_zerocopy() */
481 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
485 nla->nla_len = nla_attr_size(skb->len);
487 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
491 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
492 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
493 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
496 memset(skb_put(user_skb, plen), 0, plen);
499 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
501 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
509 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
511 struct ovs_header *ovs_header = info->userhdr;
512 struct nlattr **a = info->attrs;
513 struct sw_flow_actions *acts;
514 struct sk_buff *packet;
515 struct sw_flow *flow;
522 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
523 !a[OVS_PACKET_ATTR_ACTIONS])
526 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
527 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
531 skb_reserve(packet, NET_IP_ALIGN);
533 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
535 skb_reset_mac_header(packet);
536 eth = eth_hdr(packet);
538 /* Normally, setting the skb 'protocol' field would be handled by a
539 * call to eth_type_trans(), but it assumes there's a sending
540 * device, which we may not have. */
541 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
542 packet->protocol = eth->h_proto;
544 packet->protocol = htons(ETH_P_802_2);
546 /* Build an sw_flow for sending this packet. */
547 flow = ovs_flow_alloc();
552 err = ovs_flow_extract(packet, -1, &flow->key);
556 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
559 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
564 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
565 &flow->key, 0, &acts);
566 rcu_assign_pointer(flow->sf_acts, acts);
570 OVS_CB(packet)->flow = flow;
571 OVS_CB(packet)->pkt_key = &flow->key;
572 packet->priority = flow->key.phy.priority;
573 packet->mark = flow->key.phy.skb_mark;
576 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
582 err = ovs_execute_actions(dp, packet);
586 ovs_flow_free(flow, false);
592 ovs_flow_free(flow, false);
599 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
600 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
601 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
602 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
605 static struct genl_ops dp_packet_genl_ops[] = {
606 { .cmd = OVS_PACKET_CMD_EXECUTE,
607 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
608 .policy = packet_policy,
609 .doit = ovs_packet_cmd_execute
613 static struct genl_family dp_packet_genl_family = {
614 .id = GENL_ID_GENERATE,
615 .hdrsize = sizeof(struct ovs_header),
616 .name = OVS_PACKET_FAMILY,
617 .version = OVS_PACKET_VERSION,
618 .maxattr = OVS_PACKET_ATTR_MAX,
620 .parallel_ops = true,
621 .ops = dp_packet_genl_ops,
622 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
625 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
626 struct ovs_dp_megaflow_stats *mega_stats)
630 memset(mega_stats, 0, sizeof(*mega_stats));
632 stats->n_flows = ovs_flow_tbl_count(&dp->table);
633 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
635 stats->n_hit = stats->n_missed = stats->n_lost = 0;
637 for_each_possible_cpu(i) {
638 const struct dp_stats_percpu *percpu_stats;
639 struct dp_stats_percpu local_stats;
642 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
645 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
646 local_stats = *percpu_stats;
647 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
649 stats->n_hit += local_stats.n_hit;
650 stats->n_missed += local_stats.n_missed;
651 stats->n_lost += local_stats.n_lost;
652 mega_stats->n_mask_hit += local_stats.n_mask_hit;
656 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
658 return NLMSG_ALIGN(sizeof(struct ovs_header))
659 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
660 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
661 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
662 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
663 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
664 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
667 /* Called with ovs_mutex or RCU read lock. */
668 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
669 struct sk_buff *skb, u32 portid,
670 u32 seq, u32 flags, u8 cmd)
672 const int skb_orig_len = skb->len;
673 struct nlattr *start;
674 struct ovs_flow_stats stats;
677 struct ovs_header *ovs_header;
681 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
685 ovs_header->dp_ifindex = dp_ifindex;
688 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
690 goto nla_put_failure;
692 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
695 nla_nest_end(skb, nla);
697 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
699 goto nla_put_failure;
701 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
705 nla_nest_end(skb, nla);
707 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
710 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
711 goto nla_put_failure;
713 if (stats.n_packets &&
714 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
715 goto nla_put_failure;
717 if ((u8)ntohs(tcp_flags) &&
718 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
719 goto nla_put_failure;
721 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
722 * this is the first flow to be dumped into 'skb'. This is unusual for
723 * Netlink but individual action lists can be longer than
724 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
725 * The userspace caller can always fetch the actions separately if it
726 * really wants them. (Most userspace callers in fact don't care.)
728 * This can only fail for dump operations because the skb is always
729 * properly sized for single flows.
731 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
733 const struct sw_flow_actions *sf_acts;
735 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
736 err = ovs_nla_put_actions(sf_acts->actions,
737 sf_acts->actions_len, skb);
740 nla_nest_end(skb, start);
745 nla_nest_cancel(skb, start);
747 } else if (skb_orig_len)
748 goto nla_put_failure;
750 return genlmsg_end(skb, ovs_header);
755 genlmsg_cancel(skb, ovs_header);
759 /* May not be called with RCU read lock. */
760 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
761 struct genl_info *info,
766 if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
769 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
772 return ERR_PTR(-ENOMEM);
777 /* Called with ovs_mutex. */
778 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
780 struct genl_info *info, u8 cmd,
786 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
788 if (!skb || IS_ERR(skb))
791 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
792 info->snd_portid, info->snd_seq, 0,
798 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
800 struct nlattr **a = info->attrs;
801 struct ovs_header *ovs_header = info->userhdr;
802 struct sw_flow *flow, *new_flow;
803 struct sw_flow_mask mask;
804 struct sk_buff *reply;
806 struct sw_flow_actions *acts;
807 struct sw_flow_match match;
810 /* Must have key and actions. */
812 if (!a[OVS_FLOW_ATTR_KEY])
814 if (!a[OVS_FLOW_ATTR_ACTIONS])
817 /* Most of the time we need to allocate a new flow, do it before
819 new_flow = ovs_flow_alloc();
820 if (IS_ERR(new_flow)) {
821 error = PTR_ERR(new_flow);
826 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
827 error = ovs_nla_get_match(&match,
828 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
832 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
834 /* Validate actions. */
835 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
836 error = PTR_ERR(acts);
840 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
843 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
847 reply = ovs_flow_cmd_alloc_info(acts, info, false);
849 error = PTR_ERR(reply);
854 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
859 /* Check if this is a duplicate flow */
860 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
862 rcu_assign_pointer(new_flow->sf_acts, acts);
864 /* Put flow in bucket. */
865 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
866 if (unlikely(error)) {
871 if (unlikely(reply)) {
872 error = ovs_flow_cmd_fill_info(new_flow,
873 ovs_header->dp_ifindex,
874 reply, info->snd_portid,
881 struct sw_flow_actions *old_acts;
883 /* Bail out if we're not allowed to modify an existing flow.
884 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
885 * because Generic Netlink treats the latter as a dump
886 * request. We also accept NLM_F_EXCL in case that bug ever
889 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
894 /* The unmasked key has to be the same for flow updates. */
895 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
899 /* Update actions. */
900 old_acts = ovsl_dereference(flow->sf_acts);
901 rcu_assign_pointer(flow->sf_acts, acts);
903 if (unlikely(reply)) {
904 error = ovs_flow_cmd_fill_info(flow,
905 ovs_header->dp_ifindex,
906 reply, info->snd_portid,
913 ovs_nla_free_flow_actions(old_acts);
914 ovs_flow_free(new_flow, false);
918 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
927 ovs_flow_free(new_flow, false);
932 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
934 struct nlattr **a = info->attrs;
935 struct ovs_header *ovs_header = info->userhdr;
936 struct sw_flow_key key, masked_key;
937 struct sw_flow *flow;
938 struct sw_flow_mask mask;
939 struct sk_buff *reply = NULL;
941 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
942 struct sw_flow_match match;
947 if (!a[OVS_FLOW_ATTR_KEY])
950 ovs_match_init(&match, &key, &mask);
951 error = ovs_nla_get_match(&match,
952 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
956 /* Validate actions. */
957 if (a[OVS_FLOW_ATTR_ACTIONS]) {
958 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
959 error = PTR_ERR(acts);
963 ovs_flow_mask_key(&masked_key, &key, &mask);
964 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
965 &masked_key, 0, &acts);
967 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
972 /* Can allocate before locking if have acts. */
974 reply = ovs_flow_cmd_alloc_info(acts, info, false);
976 error = PTR_ERR(reply);
982 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
987 /* Check that the flow exists. */
988 flow = ovs_flow_tbl_lookup(&dp->table, &key);
989 if (unlikely(!flow)) {
993 /* The unmasked key has to be the same for flow updates. */
994 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
998 /* Update actions, if present. */
1000 old_acts = ovsl_dereference(flow->sf_acts);
1001 rcu_assign_pointer(flow->sf_acts, acts);
1003 if (unlikely(reply)) {
1004 error = ovs_flow_cmd_fill_info(flow,
1005 ovs_header->dp_ifindex,
1006 reply, info->snd_portid,
1012 /* Could not alloc without acts before locking. */
1013 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1014 info, OVS_FLOW_CMD_NEW, false);
1015 if (unlikely(IS_ERR(reply))) {
1016 error = PTR_ERR(reply);
1017 goto err_unlock_ovs;
1022 if (a[OVS_FLOW_ATTR_CLEAR])
1023 ovs_flow_stats_clear(flow);
1027 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1029 ovs_nla_free_flow_actions(old_acts);
1041 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1043 struct nlattr **a = info->attrs;
1044 struct ovs_header *ovs_header = info->userhdr;
1045 struct sw_flow_key key;
1046 struct sk_buff *reply;
1047 struct sw_flow *flow;
1048 struct datapath *dp;
1049 struct sw_flow_match match;
1052 if (!a[OVS_FLOW_ATTR_KEY]) {
1053 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1057 ovs_match_init(&match, &key, NULL);
1058 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1063 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1069 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1070 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1075 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1076 OVS_FLOW_CMD_NEW, true);
1077 if (IS_ERR(reply)) {
1078 err = PTR_ERR(reply);
1083 return genlmsg_reply(reply, info);
1089 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1091 struct nlattr **a = info->attrs;
1092 struct ovs_header *ovs_header = info->userhdr;
1093 struct sw_flow_key key;
1094 struct sk_buff *reply;
1095 struct sw_flow *flow;
1096 struct datapath *dp;
1097 struct sw_flow_match match;
1100 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1101 ovs_match_init(&match, &key, NULL);
1102 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1108 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1109 if (unlikely(!dp)) {
1113 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1114 err = ovs_flow_tbl_flush(&dp->table);
1117 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1118 if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1123 ovs_flow_tbl_remove(&dp->table, flow);
1126 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1129 if (likely(reply)) {
1130 if (likely(!IS_ERR(reply))) {
1131 rcu_read_lock(); /* Keep RCU checker happy. */
1132 err = ovs_flow_cmd_fill_info(flow,
1133 ovs_header->dp_ifindex,
1134 reply, info->snd_portid,
1139 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1141 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1142 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1147 ovs_flow_free(flow, true);
1154 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1156 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1157 struct table_instance *ti;
1158 struct datapath *dp;
1161 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1167 ti = rcu_dereference(dp->table.ti);
1169 struct sw_flow *flow;
1172 bucket = cb->args[0];
1174 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1178 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1179 NETLINK_CB(cb->skb).portid,
1180 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1181 OVS_FLOW_CMD_NEW) < 0)
1184 cb->args[0] = bucket;
1191 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1192 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1193 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1194 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1197 static struct genl_ops dp_flow_genl_ops[] = {
1198 { .cmd = OVS_FLOW_CMD_NEW,
1199 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1200 .policy = flow_policy,
1201 .doit = ovs_flow_cmd_new
1203 { .cmd = OVS_FLOW_CMD_DEL,
1204 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1205 .policy = flow_policy,
1206 .doit = ovs_flow_cmd_del
1208 { .cmd = OVS_FLOW_CMD_GET,
1209 .flags = 0, /* OK for unprivileged users. */
1210 .policy = flow_policy,
1211 .doit = ovs_flow_cmd_get,
1212 .dumpit = ovs_flow_cmd_dump
1214 { .cmd = OVS_FLOW_CMD_SET,
1215 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1216 .policy = flow_policy,
1217 .doit = ovs_flow_cmd_set,
1221 static struct genl_family dp_flow_genl_family = {
1222 .id = GENL_ID_GENERATE,
1223 .hdrsize = sizeof(struct ovs_header),
1224 .name = OVS_FLOW_FAMILY,
1225 .version = OVS_FLOW_VERSION,
1226 .maxattr = OVS_FLOW_ATTR_MAX,
1228 .parallel_ops = true,
1229 .ops = dp_flow_genl_ops,
1230 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1231 .mcgrps = &ovs_dp_flow_multicast_group,
1235 static size_t ovs_dp_cmd_msg_size(void)
1237 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1239 msgsize += nla_total_size(IFNAMSIZ);
1240 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1241 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1242 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1247 /* Called with ovs_mutex or RCU read lock. */
1248 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1249 u32 portid, u32 seq, u32 flags, u8 cmd)
1251 struct ovs_header *ovs_header;
1252 struct ovs_dp_stats dp_stats;
1253 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1256 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1261 ovs_header->dp_ifindex = get_dpifindex(dp);
1263 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1265 goto nla_put_failure;
1267 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1268 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1270 goto nla_put_failure;
1272 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1273 sizeof(struct ovs_dp_megaflow_stats),
1274 &dp_megaflow_stats))
1275 goto nla_put_failure;
1277 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1278 goto nla_put_failure;
1280 return genlmsg_end(skb, ovs_header);
1283 genlmsg_cancel(skb, ovs_header);
1288 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1290 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1293 /* Called with rcu_read_lock or ovs_mutex. */
1294 static struct datapath *lookup_datapath(struct net *net,
1295 struct ovs_header *ovs_header,
1296 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1298 struct datapath *dp;
1300 if (!a[OVS_DP_ATTR_NAME])
1301 dp = get_dp(net, ovs_header->dp_ifindex);
1303 struct vport *vport;
1305 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1306 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1308 return dp ? dp : ERR_PTR(-ENODEV);
1311 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1313 struct datapath *dp;
1315 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1319 WARN(dp->user_features, "Dropping previously announced user features\n");
1320 dp->user_features = 0;
1323 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1325 if (a[OVS_DP_ATTR_USER_FEATURES])
1326 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1329 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1331 struct nlattr **a = info->attrs;
1332 struct vport_parms parms;
1333 struct sk_buff *reply;
1334 struct datapath *dp;
1335 struct vport *vport;
1336 struct ovs_net *ovs_net;
1340 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1343 reply = ovs_dp_cmd_alloc_info(info);
1348 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1350 goto err_free_reply;
1352 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1354 /* Allocate table. */
1355 err = ovs_flow_tbl_init(&dp->table);
1359 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1360 if (!dp->stats_percpu) {
1362 goto err_destroy_table;
1365 for_each_possible_cpu(i) {
1366 struct dp_stats_percpu *dpath_stats;
1367 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1368 u64_stats_init(&dpath_stats->sync);
1371 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1375 goto err_destroy_percpu;
1378 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1379 INIT_HLIST_HEAD(&dp->ports[i]);
1381 /* Set up our datapath device. */
1382 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1383 parms.type = OVS_VPORT_TYPE_INTERNAL;
1384 parms.options = NULL;
1386 parms.port_no = OVSP_LOCAL;
1387 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1389 ovs_dp_change(dp, a);
1391 /* So far only local changes have been made, now need the lock. */
1394 vport = new_vport(&parms);
1395 if (IS_ERR(vport)) {
1396 err = PTR_ERR(vport);
1400 if (err == -EEXIST) {
1401 /* An outdated user space instance that does not understand
1402 * the concept of user_features has attempted to create a new
1403 * datapath and is likely to reuse it. Drop all user features.
1405 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1406 ovs_dp_reset_user_features(skb, info);
1409 goto err_destroy_ports_array;
1412 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1413 info->snd_seq, 0, OVS_DP_CMD_NEW);
1416 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1417 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1421 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1424 err_destroy_ports_array:
1428 free_percpu(dp->stats_percpu);
1430 ovs_flow_tbl_destroy(&dp->table, false);
1432 release_net(ovs_dp_get_net(dp));
1440 /* Called with ovs_mutex. */
1441 static void __dp_destroy(struct datapath *dp)
1445 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1446 struct vport *vport;
1447 struct hlist_node *n;
1449 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1450 if (vport->port_no != OVSP_LOCAL)
1451 ovs_dp_detach_port(vport);
1454 list_del_rcu(&dp->list_node);
1456 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1457 * all ports in datapath are destroyed first before freeing datapath.
1459 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1461 /* RCU destroy the flow table */
1462 ovs_flow_tbl_destroy(&dp->table, true);
1464 call_rcu(&dp->rcu, destroy_dp_rcu);
1467 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1469 struct sk_buff *reply;
1470 struct datapath *dp;
1473 reply = ovs_dp_cmd_alloc_info(info);
1478 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1481 goto err_unlock_free;
1483 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1484 info->snd_seq, 0, OVS_DP_CMD_DEL);
1490 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1499 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1501 struct sk_buff *reply;
1502 struct datapath *dp;
1505 reply = ovs_dp_cmd_alloc_info(info);
1510 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1513 goto err_unlock_free;
1515 ovs_dp_change(dp, info->attrs);
1517 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1518 info->snd_seq, 0, OVS_DP_CMD_NEW);
1522 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1531 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1533 struct sk_buff *reply;
1534 struct datapath *dp;
1537 reply = ovs_dp_cmd_alloc_info(info);
1542 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1545 goto err_unlock_free;
1547 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1548 info->snd_seq, 0, OVS_DP_CMD_NEW);
1552 return genlmsg_reply(reply, info);
1560 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1562 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1563 struct datapath *dp;
1564 int skip = cb->args[0];
1568 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1570 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1571 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1572 OVS_DP_CMD_NEW) < 0)
1583 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1584 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1585 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1586 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1589 static struct genl_ops dp_datapath_genl_ops[] = {
1590 { .cmd = OVS_DP_CMD_NEW,
1591 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1592 .policy = datapath_policy,
1593 .doit = ovs_dp_cmd_new
1595 { .cmd = OVS_DP_CMD_DEL,
1596 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1597 .policy = datapath_policy,
1598 .doit = ovs_dp_cmd_del
1600 { .cmd = OVS_DP_CMD_GET,
1601 .flags = 0, /* OK for unprivileged users. */
1602 .policy = datapath_policy,
1603 .doit = ovs_dp_cmd_get,
1604 .dumpit = ovs_dp_cmd_dump
1606 { .cmd = OVS_DP_CMD_SET,
1607 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1608 .policy = datapath_policy,
1609 .doit = ovs_dp_cmd_set,
1613 static struct genl_family dp_datapath_genl_family = {
1614 .id = GENL_ID_GENERATE,
1615 .hdrsize = sizeof(struct ovs_header),
1616 .name = OVS_DATAPATH_FAMILY,
1617 .version = OVS_DATAPATH_VERSION,
1618 .maxattr = OVS_DP_ATTR_MAX,
1620 .parallel_ops = true,
1621 .ops = dp_datapath_genl_ops,
1622 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1623 .mcgrps = &ovs_dp_datapath_multicast_group,
1627 /* Called with ovs_mutex or RCU read lock. */
1628 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1629 u32 portid, u32 seq, u32 flags, u8 cmd)
1631 struct ovs_header *ovs_header;
1632 struct ovs_vport_stats vport_stats;
1635 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1640 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1642 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1643 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1644 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1645 goto nla_put_failure;
1647 ovs_vport_get_stats(vport, &vport_stats);
1648 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1650 goto nla_put_failure;
1652 if (ovs_vport_get_upcall_portids(vport, skb))
1653 goto nla_put_failure;
1655 err = ovs_vport_get_options(vport, skb);
1656 if (err == -EMSGSIZE)
1659 return genlmsg_end(skb, ovs_header);
1664 genlmsg_cancel(skb, ovs_header);
1668 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1670 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1673 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1674 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1677 struct sk_buff *skb;
1680 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1682 return ERR_PTR(-ENOMEM);
1684 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1690 /* Called with ovs_mutex or RCU read lock. */
1691 static struct vport *lookup_vport(struct net *net,
1692 struct ovs_header *ovs_header,
1693 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1695 struct datapath *dp;
1696 struct vport *vport;
1698 if (a[OVS_VPORT_ATTR_NAME]) {
1699 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1701 return ERR_PTR(-ENODEV);
1702 if (ovs_header->dp_ifindex &&
1703 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1704 return ERR_PTR(-ENODEV);
1706 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1707 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1709 if (port_no >= DP_MAX_PORTS)
1710 return ERR_PTR(-EFBIG);
1712 dp = get_dp(net, ovs_header->dp_ifindex);
1714 return ERR_PTR(-ENODEV);
1716 vport = ovs_vport_ovsl_rcu(dp, port_no);
1718 return ERR_PTR(-ENODEV);
1721 return ERR_PTR(-EINVAL);
1724 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1726 struct nlattr **a = info->attrs;
1727 struct ovs_header *ovs_header = info->userhdr;
1728 struct vport_parms parms;
1729 struct sk_buff *reply;
1730 struct vport *vport;
1731 struct datapath *dp;
1735 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1736 !a[OVS_VPORT_ATTR_UPCALL_PID])
1739 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1740 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1741 if (port_no >= DP_MAX_PORTS)
1744 reply = ovs_vport_cmd_alloc_info();
1749 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1752 goto exit_unlock_free;
1755 vport = ovs_vport_ovsl(dp, port_no);
1758 goto exit_unlock_free;
1760 for (port_no = 1; ; port_no++) {
1761 if (port_no >= DP_MAX_PORTS) {
1763 goto exit_unlock_free;
1765 vport = ovs_vport_ovsl(dp, port_no);
1771 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1772 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1773 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1775 parms.port_no = port_no;
1776 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1778 vport = new_vport(&parms);
1779 err = PTR_ERR(vport);
1781 goto exit_unlock_free;
1784 if (a[OVS_VPORT_ATTR_STATS])
1785 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1787 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1788 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1792 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1801 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1803 struct nlattr **a = info->attrs;
1804 struct sk_buff *reply;
1805 struct vport *vport;
1808 reply = ovs_vport_cmd_alloc_info();
1813 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1814 err = PTR_ERR(vport);
1816 goto exit_unlock_free;
1818 if (a[OVS_VPORT_ATTR_TYPE] &&
1819 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1821 goto exit_unlock_free;
1824 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1825 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1827 goto exit_unlock_free;
1830 if (a[OVS_VPORT_ATTR_STATS])
1831 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1834 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1835 err = ovs_vport_set_upcall_portids(vport,
1836 a[OVS_VPORT_ATTR_UPCALL_PID]);
1838 goto exit_unlock_free;
1841 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1842 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1846 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1855 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1857 struct nlattr **a = info->attrs;
1858 struct sk_buff *reply;
1859 struct vport *vport;
1862 reply = ovs_vport_cmd_alloc_info();
1867 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1868 err = PTR_ERR(vport);
1870 goto exit_unlock_free;
1872 if (vport->port_no == OVSP_LOCAL) {
1874 goto exit_unlock_free;
1877 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1878 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1880 ovs_dp_detach_port(vport);
1883 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1892 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1894 struct nlattr **a = info->attrs;
1895 struct ovs_header *ovs_header = info->userhdr;
1896 struct sk_buff *reply;
1897 struct vport *vport;
1900 reply = ovs_vport_cmd_alloc_info();
1905 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1906 err = PTR_ERR(vport);
1908 goto exit_unlock_free;
1909 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1910 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1914 return genlmsg_reply(reply, info);
1922 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1924 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1925 struct datapath *dp;
1926 int bucket = cb->args[0], skip = cb->args[1];
1930 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1935 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1936 struct vport *vport;
1939 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1941 ovs_vport_cmd_fill_info(vport, skb,
1942 NETLINK_CB(cb->skb).portid,
1945 OVS_VPORT_CMD_NEW) < 0)
1961 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1962 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1963 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1964 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1965 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1966 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1967 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1970 static struct genl_ops dp_vport_genl_ops[] = {
1971 { .cmd = OVS_VPORT_CMD_NEW,
1972 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1973 .policy = vport_policy,
1974 .doit = ovs_vport_cmd_new
1976 { .cmd = OVS_VPORT_CMD_DEL,
1977 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1978 .policy = vport_policy,
1979 .doit = ovs_vport_cmd_del
1981 { .cmd = OVS_VPORT_CMD_GET,
1982 .flags = 0, /* OK for unprivileged users. */
1983 .policy = vport_policy,
1984 .doit = ovs_vport_cmd_get,
1985 .dumpit = ovs_vport_cmd_dump
1987 { .cmd = OVS_VPORT_CMD_SET,
1988 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1989 .policy = vport_policy,
1990 .doit = ovs_vport_cmd_set,
1994 struct genl_family dp_vport_genl_family = {
1995 .id = GENL_ID_GENERATE,
1996 .hdrsize = sizeof(struct ovs_header),
1997 .name = OVS_VPORT_FAMILY,
1998 .version = OVS_VPORT_VERSION,
1999 .maxattr = OVS_VPORT_ATTR_MAX,
2001 .parallel_ops = true,
2002 .ops = dp_vport_genl_ops,
2003 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2004 .mcgrps = &ovs_dp_vport_multicast_group,
2008 static struct genl_family *dp_genl_families[] = {
2009 &dp_datapath_genl_family,
2010 &dp_vport_genl_family,
2011 &dp_flow_genl_family,
2012 &dp_packet_genl_family,
2015 static void dp_unregister_genl(int n_families)
2019 for (i = 0; i < n_families; i++)
2020 genl_unregister_family(dp_genl_families[i]);
2023 static int dp_register_genl(void)
2028 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2030 err = genl_register_family(dp_genl_families[i]);
2038 dp_unregister_genl(i);
2042 static int __net_init ovs_init_net(struct net *net)
2044 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2046 INIT_LIST_HEAD(&ovs_net->dps);
2047 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2051 static void __net_exit ovs_exit_net(struct net *net)
2053 struct datapath *dp, *dp_next;
2054 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2057 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2061 cancel_work_sync(&ovs_net->dp_notify_work);
2064 static struct pernet_operations ovs_net_ops = {
2065 .init = ovs_init_net,
2066 .exit = ovs_exit_net,
2068 .size = sizeof(struct ovs_net),
2071 DEFINE_COMPAT_PNET_REG_FUNC(device);
2073 static int __init dp_init(void)
2077 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2079 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2082 err = ovs_flow_init();
2086 err = ovs_vport_init();
2088 goto error_flow_exit;
2090 err = register_pernet_device(&ovs_net_ops);
2092 goto error_vport_exit;
2094 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2096 goto error_netns_exit;
2098 err = dp_register_genl();
2100 goto error_unreg_notifier;
2104 error_unreg_notifier:
2105 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2107 unregister_pernet_device(&ovs_net_ops);
2116 static void dp_cleanup(void)
2118 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2119 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2120 unregister_pernet_device(&ovs_net_ops);
2126 module_init(dp_init);
2127 module_exit(dp_cleanup);
2129 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2130 MODULE_LICENSE("GPL");
2131 MODULE_VERSION(VERSION);