1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2007-2014 Nicira, Inc.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/init.h>
9 #include <linux/module.h>
10 #include <linux/if_arp.h>
11 #include <linux/if_vlan.h>
14 #include <linux/jhash.h>
15 #include <linux/delay.h>
16 #include <linux/time.h>
17 #include <linux/etherdevice.h>
18 #include <linux/genetlink.h>
19 #include <linux/kernel.h>
20 #include <linux/kthread.h>
21 #include <linux/mutex.h>
22 #include <linux/percpu.h>
23 #include <linux/rcupdate.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/ethtool.h>
27 #include <linux/wait.h>
28 #include <asm/div64.h>
29 #include <linux/highmem.h>
30 #include <linux/netfilter_bridge.h>
31 #include <linux/netfilter_ipv4.h>
32 #include <linux/inetdevice.h>
33 #include <linux/list.h>
34 #include <linux/openvswitch.h>
35 #include <linux/rculist.h>
36 #include <linux/dmi.h>
37 #include <net/genetlink.h>
38 #include <net/net_namespace.h>
39 #include <net/netns/generic.h>
43 #include "flow_table.h"
44 #include "flow_netlink.h"
46 #include "vport-internal_dev.h"
47 #include "vport-netdev.h"
49 unsigned int ovs_net_id __read_mostly;
51 static struct genl_family dp_packet_genl_family;
52 static struct genl_family dp_flow_genl_family;
53 static struct genl_family dp_datapath_genl_family;
55 static const struct nla_policy flow_policy[];
57 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
58 .name = OVS_FLOW_MCGROUP,
61 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
62 .name = OVS_DATAPATH_MCGROUP,
65 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
66 .name = OVS_VPORT_MCGROUP,
69 /* Check if need to build a reply message.
70 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
71 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
74 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
75 genl_has_listeners(family, genl_info_net(info), group);
78 static void ovs_notify(struct genl_family *family,
79 struct sk_buff *skb, struct genl_info *info)
81 genl_notify(family, skb, info, 0, GFP_KERNEL);
87 * All writes e.g. Writes to device state (add/remove datapath, port, set
88 * operations on vports, etc.), Writes to other state (flow table
89 * modifications, set miscellaneous datapath parameters, etc.) are protected
92 * Reads are protected by RCU.
94 * There are a few special cases (mostly stats) that have their own
95 * synchronization but they nest under all of above and don't interact with
98 * The RTNL lock nests inside ovs_mutex.
101 static DEFINE_MUTEX(ovs_mutex);
105 mutex_lock(&ovs_mutex);
108 void ovs_unlock(void)
110 mutex_unlock(&ovs_mutex);
113 #ifdef CONFIG_LOCKDEP
114 int lockdep_ovsl_is_held(void)
117 return lockdep_is_held(&ovs_mutex);
123 static struct vport *new_vport(const struct vport_parms *);
124 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
125 const struct sw_flow_key *,
126 const struct dp_upcall_info *,
128 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
129 const struct sw_flow_key *,
130 const struct dp_upcall_info *,
133 /* Must be called with rcu_read_lock or ovs_mutex. */
134 const char *ovs_dp_name(const struct datapath *dp)
136 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
137 return ovs_vport_name(vport);
140 static int get_dpifindex(const struct datapath *dp)
147 local = ovs_vport_rcu(dp, OVSP_LOCAL);
149 ifindex = local->dev->ifindex;
158 static void destroy_dp_rcu(struct rcu_head *rcu)
160 struct datapath *dp = container_of(rcu, struct datapath, rcu);
162 ovs_flow_tbl_destroy(&dp->table);
163 free_percpu(dp->stats_percpu);
169 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
172 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
175 /* Called with ovs_mutex or RCU read lock. */
176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
179 struct hlist_head *head;
181 head = vport_hash_bucket(dp, port_no);
182 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
183 if (vport->port_no == port_no)
189 /* Called with ovs_mutex. */
190 static struct vport *new_vport(const struct vport_parms *parms)
194 vport = ovs_vport_add(parms);
195 if (!IS_ERR(vport)) {
196 struct datapath *dp = parms->dp;
197 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
199 hlist_add_head_rcu(&vport->dp_hash_node, head);
204 void ovs_dp_detach_port(struct vport *p)
208 /* First drop references to device. */
209 hlist_del_rcu(&p->dp_hash_node);
211 /* Then destroy it. */
215 /* Must be called with rcu_read_lock. */
216 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
218 const struct vport *p = OVS_CB(skb)->input_vport;
219 struct datapath *dp = p->dp;
220 struct sw_flow *flow;
221 struct sw_flow_actions *sf_acts;
222 struct dp_stats_percpu *stats;
227 stats = this_cpu_ptr(dp->stats_percpu);
230 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
232 if (unlikely(!flow)) {
233 struct dp_upcall_info upcall;
235 memset(&upcall, 0, sizeof(upcall));
236 upcall.cmd = OVS_PACKET_CMD_MISS;
237 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
238 upcall.mru = OVS_CB(skb)->mru;
239 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
244 stats_counter = &stats->n_missed;
248 ovs_flow_stats_update(flow, key->tp.flags, skb);
249 sf_acts = rcu_dereference(flow->sf_acts);
250 error = ovs_execute_actions(dp, skb, sf_acts, key);
252 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
253 ovs_dp_name(dp), error);
255 stats_counter = &stats->n_hit;
258 /* Update datapath statistics. */
259 u64_stats_update_begin(&stats->syncp);
261 stats->n_mask_hit += n_mask_hit;
262 u64_stats_update_end(&stats->syncp);
265 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
266 const struct sw_flow_key *key,
267 const struct dp_upcall_info *upcall_info,
270 struct dp_stats_percpu *stats;
273 if (upcall_info->portid == 0) {
278 if (!skb_is_gso(skb))
279 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
281 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
288 stats = this_cpu_ptr(dp->stats_percpu);
290 u64_stats_update_begin(&stats->syncp);
292 u64_stats_update_end(&stats->syncp);
297 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
298 const struct sw_flow_key *key,
299 const struct dp_upcall_info *upcall_info,
302 unsigned int gso_type = skb_shinfo(skb)->gso_type;
303 struct sw_flow_key later_key;
304 struct sk_buff *segs, *nskb;
307 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
308 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
310 return PTR_ERR(segs);
314 if (gso_type & SKB_GSO_UDP) {
315 /* The initial flow key extracted by ovs_flow_key_extract()
316 * in this case is for a first fragment, so we need to
317 * properly mark later fragments.
320 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
323 /* Queue all of the segments. */
324 skb_list_walk_safe(segs, skb, nskb) {
325 if (gso_type & SKB_GSO_UDP && skb != segs)
328 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
334 /* Free all of the segments. */
335 skb_list_walk_safe(segs, skb, nskb) {
344 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
345 unsigned int hdrlen, int actions_attrlen)
347 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
348 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
349 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
350 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
351 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
353 /* OVS_PACKET_ATTR_USERDATA */
354 if (upcall_info->userdata)
355 size += NLA_ALIGN(upcall_info->userdata->nla_len);
357 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
358 if (upcall_info->egress_tun_info)
359 size += nla_total_size(ovs_tun_key_attr_size());
361 /* OVS_PACKET_ATTR_ACTIONS */
362 if (upcall_info->actions_len)
363 size += nla_total_size(actions_attrlen);
365 /* OVS_PACKET_ATTR_MRU */
366 if (upcall_info->mru)
367 size += nla_total_size(sizeof(upcall_info->mru));
372 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
374 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
375 size_t plen = NLA_ALIGN(skb->len) - skb->len;
378 skb_put_zero(skb, plen);
382 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
383 const struct sw_flow_key *key,
384 const struct dp_upcall_info *upcall_info,
387 struct ovs_header *upcall;
388 struct sk_buff *nskb = NULL;
389 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
396 dp_ifindex = get_dpifindex(dp);
400 if (skb_vlan_tag_present(skb)) {
401 nskb = skb_clone(skb, GFP_ATOMIC);
405 nskb = __vlan_hwaccel_push_inside(nskb);
412 if (nla_attr_size(skb->len) > USHRT_MAX) {
417 /* Complete checksum if needed */
418 if (skb->ip_summed == CHECKSUM_PARTIAL &&
419 (err = skb_csum_hwoffload_help(skb, 0)))
422 /* Older versions of OVS user space enforce alignment of the last
423 * Netlink attribute to NLA_ALIGNTO which would require extensive
424 * padding logic. Only perform zerocopy if padding is not required.
426 if (dp->user_features & OVS_DP_F_UNALIGNED)
427 hlen = skb_zerocopy_headlen(skb);
431 len = upcall_msg_size(upcall_info, hlen - cutlen,
432 OVS_CB(skb)->acts_origlen);
433 user_skb = genlmsg_new(len, GFP_ATOMIC);
439 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
440 0, upcall_info->cmd);
445 upcall->dp_ifindex = dp_ifindex;
447 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
451 if (upcall_info->userdata)
452 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
453 nla_len(upcall_info->userdata),
454 nla_data(upcall_info->userdata));
456 if (upcall_info->egress_tun_info) {
457 nla = nla_nest_start_noflag(user_skb,
458 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
463 err = ovs_nla_put_tunnel_info(user_skb,
464 upcall_info->egress_tun_info);
468 nla_nest_end(user_skb, nla);
471 if (upcall_info->actions_len) {
472 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
477 err = ovs_nla_put_actions(upcall_info->actions,
478 upcall_info->actions_len,
481 nla_nest_end(user_skb, nla);
483 nla_nest_cancel(user_skb, nla);
486 /* Add OVS_PACKET_ATTR_MRU */
487 if (upcall_info->mru &&
488 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
493 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
495 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
500 /* Add OVS_PACKET_ATTR_HASH */
501 hash = skb_get_hash_raw(skb);
503 hash |= OVS_PACKET_HASH_SW_BIT;
506 hash |= OVS_PACKET_HASH_L4_BIT;
508 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
513 /* Only reserve room for attribute header, packet data is added
514 * in skb_zerocopy() */
515 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
519 nla->nla_len = nla_attr_size(skb->len - cutlen);
521 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
525 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
526 pad_packet(dp, user_skb);
528 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
530 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
540 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
542 struct ovs_header *ovs_header = info->userhdr;
543 struct net *net = sock_net(skb->sk);
544 struct nlattr **a = info->attrs;
545 struct sw_flow_actions *acts;
546 struct sk_buff *packet;
547 struct sw_flow *flow;
548 struct sw_flow_actions *sf_acts;
550 struct vport *input_vport;
555 bool log = !a[OVS_PACKET_ATTR_PROBE];
558 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
559 !a[OVS_PACKET_ATTR_ACTIONS])
562 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
563 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
567 skb_reserve(packet, NET_IP_ALIGN);
569 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
571 /* Set packet's mru */
572 if (a[OVS_PACKET_ATTR_MRU]) {
573 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
574 packet->ignore_df = 1;
576 OVS_CB(packet)->mru = mru;
578 if (a[OVS_PACKET_ATTR_HASH]) {
579 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
581 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
582 !!(hash & OVS_PACKET_HASH_SW_BIT),
583 !!(hash & OVS_PACKET_HASH_L4_BIT));
586 /* Build an sw_flow for sending this packet. */
587 flow = ovs_flow_alloc();
592 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
593 packet, &flow->key, log);
597 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
598 &flow->key, &acts, log);
602 rcu_assign_pointer(flow->sf_acts, acts);
603 packet->priority = flow->key.phy.priority;
604 packet->mark = flow->key.phy.skb_mark;
607 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
612 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
614 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
619 packet->dev = input_vport->dev;
620 OVS_CB(packet)->input_vport = input_vport;
621 sf_acts = rcu_dereference(flow->sf_acts);
624 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
628 ovs_flow_free(flow, false);
634 ovs_flow_free(flow, false);
641 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
642 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
643 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
644 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
645 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
646 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
649 static const struct genl_ops dp_packet_genl_ops[] = {
650 { .cmd = OVS_PACKET_CMD_EXECUTE,
651 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
652 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
653 .doit = ovs_packet_cmd_execute
657 static struct genl_family dp_packet_genl_family __ro_after_init = {
658 .hdrsize = sizeof(struct ovs_header),
659 .name = OVS_PACKET_FAMILY,
660 .version = OVS_PACKET_VERSION,
661 .maxattr = OVS_PACKET_ATTR_MAX,
662 .policy = packet_policy,
664 .parallel_ops = true,
665 .ops = dp_packet_genl_ops,
666 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
667 .module = THIS_MODULE,
670 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
671 struct ovs_dp_megaflow_stats *mega_stats)
675 memset(mega_stats, 0, sizeof(*mega_stats));
677 stats->n_flows = ovs_flow_tbl_count(&dp->table);
678 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
680 stats->n_hit = stats->n_missed = stats->n_lost = 0;
682 for_each_possible_cpu(i) {
683 const struct dp_stats_percpu *percpu_stats;
684 struct dp_stats_percpu local_stats;
687 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
690 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
691 local_stats = *percpu_stats;
692 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
694 stats->n_hit += local_stats.n_hit;
695 stats->n_missed += local_stats.n_missed;
696 stats->n_lost += local_stats.n_lost;
697 mega_stats->n_mask_hit += local_stats.n_mask_hit;
701 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
703 return ovs_identifier_is_ufid(sfid) &&
704 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
707 static bool should_fill_mask(uint32_t ufid_flags)
709 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
712 static bool should_fill_actions(uint32_t ufid_flags)
714 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
717 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
718 const struct sw_flow_id *sfid,
721 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
723 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
724 * see ovs_nla_put_identifier()
726 if (sfid && ovs_identifier_is_ufid(sfid))
727 len += nla_total_size(sfid->ufid_len);
729 len += nla_total_size(ovs_key_attr_size());
731 /* OVS_FLOW_ATTR_KEY */
732 if (!sfid || should_fill_key(sfid, ufid_flags))
733 len += nla_total_size(ovs_key_attr_size());
735 /* OVS_FLOW_ATTR_MASK */
736 if (should_fill_mask(ufid_flags))
737 len += nla_total_size(ovs_key_attr_size());
739 /* OVS_FLOW_ATTR_ACTIONS */
740 if (should_fill_actions(ufid_flags))
741 len += nla_total_size(acts->orig_len);
744 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
745 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
746 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
749 /* Called with ovs_mutex or RCU read lock. */
750 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
753 struct ovs_flow_stats stats;
757 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
760 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
764 if (stats.n_packets &&
765 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
766 sizeof(struct ovs_flow_stats), &stats,
770 if ((u8)ntohs(tcp_flags) &&
771 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
777 /* Called with ovs_mutex or RCU read lock. */
778 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
779 struct sk_buff *skb, int skb_orig_len)
781 struct nlattr *start;
784 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
785 * this is the first flow to be dumped into 'skb'. This is unusual for
786 * Netlink but individual action lists can be longer than
787 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
788 * The userspace caller can always fetch the actions separately if it
789 * really wants them. (Most userspace callers in fact don't care.)
791 * This can only fail for dump operations because the skb is always
792 * properly sized for single flows.
794 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
796 const struct sw_flow_actions *sf_acts;
798 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
799 err = ovs_nla_put_actions(sf_acts->actions,
800 sf_acts->actions_len, skb);
803 nla_nest_end(skb, start);
808 nla_nest_cancel(skb, start);
810 } else if (skb_orig_len) {
817 /* Called with ovs_mutex or RCU read lock. */
818 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
819 struct sk_buff *skb, u32 portid,
820 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
822 const int skb_orig_len = skb->len;
823 struct ovs_header *ovs_header;
826 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
831 ovs_header->dp_ifindex = dp_ifindex;
833 err = ovs_nla_put_identifier(flow, skb);
837 if (should_fill_key(&flow->id, ufid_flags)) {
838 err = ovs_nla_put_masked_key(flow, skb);
843 if (should_fill_mask(ufid_flags)) {
844 err = ovs_nla_put_mask(flow, skb);
849 err = ovs_flow_cmd_fill_stats(flow, skb);
853 if (should_fill_actions(ufid_flags)) {
854 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
859 genlmsg_end(skb, ovs_header);
863 genlmsg_cancel(skb, ovs_header);
867 /* May not be called with RCU read lock. */
868 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
869 const struct sw_flow_id *sfid,
870 struct genl_info *info,
877 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
880 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
881 skb = genlmsg_new(len, GFP_KERNEL);
883 return ERR_PTR(-ENOMEM);
888 /* Called with ovs_mutex. */
889 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
891 struct genl_info *info, u8 cmd,
892 bool always, u32 ufid_flags)
897 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
898 &flow->id, info, always, ufid_flags);
899 if (IS_ERR_OR_NULL(skb))
902 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
903 info->snd_portid, info->snd_seq, 0,
905 if (WARN_ON_ONCE(retval < 0)) {
907 skb = ERR_PTR(retval);
912 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
914 struct net *net = sock_net(skb->sk);
915 struct nlattr **a = info->attrs;
916 struct ovs_header *ovs_header = info->userhdr;
917 struct sw_flow *flow = NULL, *new_flow;
918 struct sw_flow_mask mask;
919 struct sk_buff *reply;
921 struct sw_flow_actions *acts;
922 struct sw_flow_match match;
923 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
925 bool log = !a[OVS_FLOW_ATTR_PROBE];
927 /* Must have key and actions. */
929 if (!a[OVS_FLOW_ATTR_KEY]) {
930 OVS_NLERR(log, "Flow key attr not present in new flow.");
933 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
934 OVS_NLERR(log, "Flow actions attr not present in new flow.");
938 /* Most of the time we need to allocate a new flow, do it before
941 new_flow = ovs_flow_alloc();
942 if (IS_ERR(new_flow)) {
943 error = PTR_ERR(new_flow);
948 ovs_match_init(&match, &new_flow->key, false, &mask);
949 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
950 a[OVS_FLOW_ATTR_MASK], log);
954 /* Extract flow identifier. */
955 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
956 &new_flow->key, log);
960 /* unmasked key is needed to match when ufid is not used. */
961 if (ovs_identifier_is_key(&new_flow->id))
962 match.key = new_flow->id.unmasked_key;
964 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
966 /* Validate actions. */
967 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
968 &new_flow->key, &acts, log);
970 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
974 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
977 error = PTR_ERR(reply);
982 dp = get_dp(net, ovs_header->dp_ifindex);
988 /* Check if this is a duplicate flow */
989 if (ovs_identifier_is_ufid(&new_flow->id))
990 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
992 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
994 rcu_assign_pointer(new_flow->sf_acts, acts);
996 /* Put flow in bucket. */
997 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
998 if (unlikely(error)) {
1000 goto err_unlock_ovs;
1003 if (unlikely(reply)) {
1004 error = ovs_flow_cmd_fill_info(new_flow,
1005 ovs_header->dp_ifindex,
1006 reply, info->snd_portid,
1014 struct sw_flow_actions *old_acts;
1016 /* Bail out if we're not allowed to modify an existing flow.
1017 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1018 * because Generic Netlink treats the latter as a dump
1019 * request. We also accept NLM_F_EXCL in case that bug ever
1022 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1025 goto err_unlock_ovs;
1027 /* The flow identifier has to be the same for flow updates.
1028 * Look for any overlapping flow.
1030 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1031 if (ovs_identifier_is_key(&flow->id))
1032 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1034 else /* UFID matches but key is different */
1038 goto err_unlock_ovs;
1041 /* Update actions. */
1042 old_acts = ovsl_dereference(flow->sf_acts);
1043 rcu_assign_pointer(flow->sf_acts, acts);
1045 if (unlikely(reply)) {
1046 error = ovs_flow_cmd_fill_info(flow,
1047 ovs_header->dp_ifindex,
1048 reply, info->snd_portid,
1056 ovs_nla_free_flow_actions_rcu(old_acts);
1057 ovs_flow_free(new_flow, false);
1061 ovs_notify(&dp_flow_genl_family, reply, info);
1068 ovs_nla_free_flow_actions(acts);
1070 ovs_flow_free(new_flow, false);
1075 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1076 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1077 const struct nlattr *a,
1078 const struct sw_flow_key *key,
1079 const struct sw_flow_mask *mask,
1082 struct sw_flow_actions *acts;
1083 struct sw_flow_key masked_key;
1086 ovs_flow_mask_key(&masked_key, key, true, mask);
1087 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1090 "Actions may not be safe on all matching packets");
1091 return ERR_PTR(error);
1097 /* Factor out match-init and action-copy to avoid
1098 * "Wframe-larger-than=1024" warning. Because mask is only
1099 * used to get actions, we new a function to save some
1102 * If there are not key and action attrs, we return 0
1103 * directly. In the case, the caller will also not use the
1104 * match as before. If there is action attr, we try to get
1105 * actions and save them to *acts. Before returning from
1106 * the function, we reset the match->mask pointer. Because
1107 * we should not to return match object with dangling reference
1110 static noinline_for_stack int
1111 ovs_nla_init_match_and_action(struct net *net,
1112 struct sw_flow_match *match,
1113 struct sw_flow_key *key,
1115 struct sw_flow_actions **acts,
1118 struct sw_flow_mask mask;
1121 if (a[OVS_FLOW_ATTR_KEY]) {
1122 ovs_match_init(match, key, true, &mask);
1123 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1124 a[OVS_FLOW_ATTR_MASK], log);
1129 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1130 if (!a[OVS_FLOW_ATTR_KEY]) {
1132 "Flow key attribute not present in set flow.");
1137 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1139 if (IS_ERR(*acts)) {
1140 error = PTR_ERR(*acts);
1145 /* On success, error is 0. */
1151 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1153 struct net *net = sock_net(skb->sk);
1154 struct nlattr **a = info->attrs;
1155 struct ovs_header *ovs_header = info->userhdr;
1156 struct sw_flow_key key;
1157 struct sw_flow *flow;
1158 struct sk_buff *reply = NULL;
1159 struct datapath *dp;
1160 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1161 struct sw_flow_match match;
1162 struct sw_flow_id sfid;
1163 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1165 bool log = !a[OVS_FLOW_ATTR_PROBE];
1168 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1169 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1171 "Flow set message rejected, Key attribute missing.");
1175 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1181 /* Can allocate before locking if have acts. */
1182 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1184 if (IS_ERR(reply)) {
1185 error = PTR_ERR(reply);
1186 goto err_kfree_acts;
1191 dp = get_dp(net, ovs_header->dp_ifindex);
1192 if (unlikely(!dp)) {
1194 goto err_unlock_ovs;
1196 /* Check that the flow exists. */
1198 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1200 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1201 if (unlikely(!flow)) {
1203 goto err_unlock_ovs;
1206 /* Update actions, if present. */
1208 old_acts = ovsl_dereference(flow->sf_acts);
1209 rcu_assign_pointer(flow->sf_acts, acts);
1211 if (unlikely(reply)) {
1212 error = ovs_flow_cmd_fill_info(flow,
1213 ovs_header->dp_ifindex,
1214 reply, info->snd_portid,
1221 /* Could not alloc without acts before locking. */
1222 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1223 info, OVS_FLOW_CMD_SET, false,
1226 if (IS_ERR(reply)) {
1227 error = PTR_ERR(reply);
1228 goto err_unlock_ovs;
1233 if (a[OVS_FLOW_ATTR_CLEAR])
1234 ovs_flow_stats_clear(flow);
1238 ovs_notify(&dp_flow_genl_family, reply, info);
1240 ovs_nla_free_flow_actions_rcu(old_acts);
1248 ovs_nla_free_flow_actions(acts);
1253 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1255 struct nlattr **a = info->attrs;
1256 struct ovs_header *ovs_header = info->userhdr;
1257 struct net *net = sock_net(skb->sk);
1258 struct sw_flow_key key;
1259 struct sk_buff *reply;
1260 struct sw_flow *flow;
1261 struct datapath *dp;
1262 struct sw_flow_match match;
1263 struct sw_flow_id ufid;
1264 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1266 bool log = !a[OVS_FLOW_ATTR_PROBE];
1269 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1270 if (a[OVS_FLOW_ATTR_KEY]) {
1271 ovs_match_init(&match, &key, true, NULL);
1272 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1274 } else if (!ufid_present) {
1276 "Flow get message rejected, Key attribute missing.");
1283 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1290 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1292 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1298 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1299 OVS_FLOW_CMD_GET, true, ufid_flags);
1300 if (IS_ERR(reply)) {
1301 err = PTR_ERR(reply);
1306 return genlmsg_reply(reply, info);
1312 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1314 struct nlattr **a = info->attrs;
1315 struct ovs_header *ovs_header = info->userhdr;
1316 struct net *net = sock_net(skb->sk);
1317 struct sw_flow_key key;
1318 struct sk_buff *reply;
1319 struct sw_flow *flow = NULL;
1320 struct datapath *dp;
1321 struct sw_flow_match match;
1322 struct sw_flow_id ufid;
1323 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1325 bool log = !a[OVS_FLOW_ATTR_PROBE];
1328 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1329 if (a[OVS_FLOW_ATTR_KEY]) {
1330 ovs_match_init(&match, &key, true, NULL);
1331 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1338 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1339 if (unlikely(!dp)) {
1344 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1345 err = ovs_flow_tbl_flush(&dp->table);
1350 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1352 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1353 if (unlikely(!flow)) {
1358 ovs_flow_tbl_remove(&dp->table, flow);
1361 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1362 &flow->id, info, false, ufid_flags);
1363 if (likely(reply)) {
1364 if (!IS_ERR(reply)) {
1365 rcu_read_lock(); /*To keep RCU checker happy. */
1366 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1367 reply, info->snd_portid,
1372 if (WARN_ON_ONCE(err < 0)) {
1377 ovs_notify(&dp_flow_genl_family, reply, info);
1379 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1384 ovs_flow_free(flow, true);
1391 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1393 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1394 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1395 struct table_instance *ti;
1396 struct datapath *dp;
1400 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1401 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1404 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1407 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1413 ti = rcu_dereference(dp->table.ti);
1415 struct sw_flow *flow;
1418 bucket = cb->args[0];
1420 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1424 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1425 NETLINK_CB(cb->skb).portid,
1426 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1427 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1430 cb->args[0] = bucket;
1437 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1438 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1439 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1440 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1441 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1442 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1443 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1444 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1447 static const struct genl_ops dp_flow_genl_ops[] = {
1448 { .cmd = OVS_FLOW_CMD_NEW,
1449 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1450 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1451 .doit = ovs_flow_cmd_new
1453 { .cmd = OVS_FLOW_CMD_DEL,
1454 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1455 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1456 .doit = ovs_flow_cmd_del
1458 { .cmd = OVS_FLOW_CMD_GET,
1459 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1460 .flags = 0, /* OK for unprivileged users. */
1461 .doit = ovs_flow_cmd_get,
1462 .dumpit = ovs_flow_cmd_dump
1464 { .cmd = OVS_FLOW_CMD_SET,
1465 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1466 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1467 .doit = ovs_flow_cmd_set,
1471 static struct genl_family dp_flow_genl_family __ro_after_init = {
1472 .hdrsize = sizeof(struct ovs_header),
1473 .name = OVS_FLOW_FAMILY,
1474 .version = OVS_FLOW_VERSION,
1475 .maxattr = OVS_FLOW_ATTR_MAX,
1476 .policy = flow_policy,
1478 .parallel_ops = true,
1479 .ops = dp_flow_genl_ops,
1480 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1481 .mcgrps = &ovs_dp_flow_multicast_group,
1483 .module = THIS_MODULE,
1486 static size_t ovs_dp_cmd_msg_size(void)
1488 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1490 msgsize += nla_total_size(IFNAMSIZ);
1491 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1492 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1493 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1498 /* Called with ovs_mutex. */
1499 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1500 u32 portid, u32 seq, u32 flags, u8 cmd)
1502 struct ovs_header *ovs_header;
1503 struct ovs_dp_stats dp_stats;
1504 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1507 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1512 ovs_header->dp_ifindex = get_dpifindex(dp);
1514 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1516 goto nla_put_failure;
1518 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1519 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1520 &dp_stats, OVS_DP_ATTR_PAD))
1521 goto nla_put_failure;
1523 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1524 sizeof(struct ovs_dp_megaflow_stats),
1525 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1526 goto nla_put_failure;
1528 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1529 goto nla_put_failure;
1531 genlmsg_end(skb, ovs_header);
1535 genlmsg_cancel(skb, ovs_header);
1540 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1542 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1545 /* Called with rcu_read_lock or ovs_mutex. */
1546 static struct datapath *lookup_datapath(struct net *net,
1547 const struct ovs_header *ovs_header,
1548 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1550 struct datapath *dp;
1552 if (!a[OVS_DP_ATTR_NAME])
1553 dp = get_dp(net, ovs_header->dp_ifindex);
1555 struct vport *vport;
1557 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1558 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1560 return dp ? dp : ERR_PTR(-ENODEV);
1563 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1565 struct datapath *dp;
1567 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1571 WARN(dp->user_features, "Dropping previously announced user features\n");
1572 dp->user_features = 0;
1575 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1577 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1579 u32 user_features = 0;
1581 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1582 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1584 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1585 OVS_DP_F_UNALIGNED |
1586 OVS_DP_F_TC_RECIRC_SHARING))
1589 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1590 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1595 dp->user_features = user_features;
1597 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1598 static_branch_enable(&tc_recirc_sharing_support);
1600 static_branch_disable(&tc_recirc_sharing_support);
1605 static int ovs_dp_stats_init(struct datapath *dp)
1607 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1608 if (!dp->stats_percpu)
1614 static int ovs_dp_vport_init(struct datapath *dp)
1618 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1619 sizeof(struct hlist_head),
1624 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1625 INIT_HLIST_HEAD(&dp->ports[i]);
1630 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1632 struct nlattr **a = info->attrs;
1633 struct vport_parms parms;
1634 struct sk_buff *reply;
1635 struct datapath *dp;
1636 struct vport *vport;
1637 struct ovs_net *ovs_net;
1641 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1644 reply = ovs_dp_cmd_alloc_info();
1649 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1651 goto err_destroy_reply;
1653 ovs_dp_set_net(dp, sock_net(skb->sk));
1655 /* Allocate table. */
1656 err = ovs_flow_tbl_init(&dp->table);
1658 goto err_destroy_dp;
1660 err = ovs_dp_stats_init(dp);
1662 goto err_destroy_table;
1664 err = ovs_dp_vport_init(dp);
1666 goto err_destroy_stats;
1668 err = ovs_meters_init(dp);
1670 goto err_destroy_ports;
1672 /* Set up our datapath device. */
1673 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1674 parms.type = OVS_VPORT_TYPE_INTERNAL;
1675 parms.options = NULL;
1677 parms.port_no = OVSP_LOCAL;
1678 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1680 err = ovs_dp_change(dp, a);
1682 goto err_destroy_meters;
1684 /* So far only local changes have been made, now need the lock. */
1687 vport = new_vport(&parms);
1688 if (IS_ERR(vport)) {
1689 err = PTR_ERR(vport);
1693 if (err == -EEXIST) {
1694 /* An outdated user space instance that does not understand
1695 * the concept of user_features has attempted to create a new
1696 * datapath and is likely to reuse it. Drop all user features.
1698 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1699 ovs_dp_reset_user_features(skb, info);
1703 goto err_destroy_meters;
1706 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1707 info->snd_seq, 0, OVS_DP_CMD_NEW);
1710 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1711 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1715 ovs_notify(&dp_datapath_genl_family, reply, info);
1719 ovs_meters_exit(dp);
1723 free_percpu(dp->stats_percpu);
1725 ovs_flow_tbl_destroy(&dp->table);
1734 /* Called with ovs_mutex. */
1735 static void __dp_destroy(struct datapath *dp)
1739 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1740 struct vport *vport;
1741 struct hlist_node *n;
1743 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1744 if (vport->port_no != OVSP_LOCAL)
1745 ovs_dp_detach_port(vport);
1748 list_del_rcu(&dp->list_node);
1750 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1751 * all ports in datapath are destroyed first before freeing datapath.
1753 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1755 /* RCU destroy the flow table */
1756 call_rcu(&dp->rcu, destroy_dp_rcu);
1759 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1761 struct sk_buff *reply;
1762 struct datapath *dp;
1765 reply = ovs_dp_cmd_alloc_info();
1770 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1773 goto err_unlock_free;
1775 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1776 info->snd_seq, 0, OVS_DP_CMD_DEL);
1782 ovs_notify(&dp_datapath_genl_family, reply, info);
1792 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1794 struct sk_buff *reply;
1795 struct datapath *dp;
1798 reply = ovs_dp_cmd_alloc_info();
1803 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1806 goto err_unlock_free;
1808 err = ovs_dp_change(dp, info->attrs);
1810 goto err_unlock_free;
1812 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1813 info->snd_seq, 0, OVS_DP_CMD_SET);
1817 ovs_notify(&dp_datapath_genl_family, reply, info);
1827 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1829 struct sk_buff *reply;
1830 struct datapath *dp;
1833 reply = ovs_dp_cmd_alloc_info();
1838 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1841 goto err_unlock_free;
1843 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1844 info->snd_seq, 0, OVS_DP_CMD_GET);
1848 return genlmsg_reply(reply, info);
1856 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1858 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1859 struct datapath *dp;
1860 int skip = cb->args[0];
1864 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1866 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1867 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1868 OVS_DP_CMD_GET) < 0)
1879 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1880 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1881 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1882 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1885 static const struct genl_ops dp_datapath_genl_ops[] = {
1886 { .cmd = OVS_DP_CMD_NEW,
1887 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1888 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1889 .doit = ovs_dp_cmd_new
1891 { .cmd = OVS_DP_CMD_DEL,
1892 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1893 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1894 .doit = ovs_dp_cmd_del
1896 { .cmd = OVS_DP_CMD_GET,
1897 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1898 .flags = 0, /* OK for unprivileged users. */
1899 .doit = ovs_dp_cmd_get,
1900 .dumpit = ovs_dp_cmd_dump
1902 { .cmd = OVS_DP_CMD_SET,
1903 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1904 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1905 .doit = ovs_dp_cmd_set,
1909 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1910 .hdrsize = sizeof(struct ovs_header),
1911 .name = OVS_DATAPATH_FAMILY,
1912 .version = OVS_DATAPATH_VERSION,
1913 .maxattr = OVS_DP_ATTR_MAX,
1914 .policy = datapath_policy,
1916 .parallel_ops = true,
1917 .ops = dp_datapath_genl_ops,
1918 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1919 .mcgrps = &ovs_dp_datapath_multicast_group,
1921 .module = THIS_MODULE,
1924 /* Called with ovs_mutex or RCU read lock. */
1925 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1926 struct net *net, u32 portid, u32 seq,
1927 u32 flags, u8 cmd, gfp_t gfp)
1929 struct ovs_header *ovs_header;
1930 struct ovs_vport_stats vport_stats;
1933 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1938 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1940 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1941 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1942 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1943 ovs_vport_name(vport)) ||
1944 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1945 goto nla_put_failure;
1947 if (!net_eq(net, dev_net(vport->dev))) {
1948 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1950 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1951 goto nla_put_failure;
1954 ovs_vport_get_stats(vport, &vport_stats);
1955 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1956 sizeof(struct ovs_vport_stats), &vport_stats,
1957 OVS_VPORT_ATTR_PAD))
1958 goto nla_put_failure;
1960 if (ovs_vport_get_upcall_portids(vport, skb))
1961 goto nla_put_failure;
1963 err = ovs_vport_get_options(vport, skb);
1964 if (err == -EMSGSIZE)
1967 genlmsg_end(skb, ovs_header);
1973 genlmsg_cancel(skb, ovs_header);
1977 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1979 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1982 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1983 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1984 u32 portid, u32 seq, u8 cmd)
1986 struct sk_buff *skb;
1989 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1991 return ERR_PTR(-ENOMEM);
1993 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2000 /* Called with ovs_mutex or RCU read lock. */
2001 static struct vport *lookup_vport(struct net *net,
2002 const struct ovs_header *ovs_header,
2003 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2005 struct datapath *dp;
2006 struct vport *vport;
2008 if (a[OVS_VPORT_ATTR_IFINDEX])
2009 return ERR_PTR(-EOPNOTSUPP);
2010 if (a[OVS_VPORT_ATTR_NAME]) {
2011 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2013 return ERR_PTR(-ENODEV);
2014 if (ovs_header->dp_ifindex &&
2015 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2016 return ERR_PTR(-ENODEV);
2018 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2019 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2021 if (port_no >= DP_MAX_PORTS)
2022 return ERR_PTR(-EFBIG);
2024 dp = get_dp(net, ovs_header->dp_ifindex);
2026 return ERR_PTR(-ENODEV);
2028 vport = ovs_vport_ovsl_rcu(dp, port_no);
2030 return ERR_PTR(-ENODEV);
2033 return ERR_PTR(-EINVAL);
2037 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2039 unsigned int dev_headroom, max_headroom = 0;
2040 struct net_device *dev;
2041 struct vport *vport;
2044 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2045 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2047 dev_headroom = netdev_get_fwd_headroom(dev);
2048 if (dev_headroom > max_headroom)
2049 max_headroom = dev_headroom;
2053 return max_headroom;
2056 /* Called with ovs_mutex */
2057 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2059 struct vport *vport;
2062 dp->max_headroom = new_headroom;
2063 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2064 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2065 netdev_set_rx_headroom(vport->dev, new_headroom);
2068 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2070 struct nlattr **a = info->attrs;
2071 struct ovs_header *ovs_header = info->userhdr;
2072 struct vport_parms parms;
2073 struct sk_buff *reply;
2074 struct vport *vport;
2075 struct datapath *dp;
2076 unsigned int new_headroom;
2080 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2081 !a[OVS_VPORT_ATTR_UPCALL_PID])
2083 if (a[OVS_VPORT_ATTR_IFINDEX])
2086 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2087 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2088 if (port_no >= DP_MAX_PORTS)
2091 reply = ovs_vport_cmd_alloc_info();
2097 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2100 goto exit_unlock_free;
2103 vport = ovs_vport_ovsl(dp, port_no);
2106 goto exit_unlock_free;
2108 for (port_no = 1; ; port_no++) {
2109 if (port_no >= DP_MAX_PORTS) {
2111 goto exit_unlock_free;
2113 vport = ovs_vport_ovsl(dp, port_no);
2119 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2120 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2121 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2123 parms.port_no = port_no;
2124 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2126 vport = new_vport(&parms);
2127 err = PTR_ERR(vport);
2128 if (IS_ERR(vport)) {
2131 goto exit_unlock_free;
2134 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2135 info->snd_portid, info->snd_seq, 0,
2136 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2138 new_headroom = netdev_get_fwd_headroom(vport->dev);
2140 if (new_headroom > dp->max_headroom)
2141 ovs_update_headroom(dp, new_headroom);
2143 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2148 ovs_notify(&dp_vport_genl_family, reply, info);
2157 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2159 struct nlattr **a = info->attrs;
2160 struct sk_buff *reply;
2161 struct vport *vport;
2164 reply = ovs_vport_cmd_alloc_info();
2169 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2170 err = PTR_ERR(vport);
2172 goto exit_unlock_free;
2174 if (a[OVS_VPORT_ATTR_TYPE] &&
2175 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2177 goto exit_unlock_free;
2180 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2181 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2183 goto exit_unlock_free;
2187 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2188 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2190 err = ovs_vport_set_upcall_portids(vport, ids);
2192 goto exit_unlock_free;
2195 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2196 info->snd_portid, info->snd_seq, 0,
2197 OVS_VPORT_CMD_SET, GFP_KERNEL);
2201 ovs_notify(&dp_vport_genl_family, reply, info);
2210 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2212 bool update_headroom = false;
2213 struct nlattr **a = info->attrs;
2214 struct sk_buff *reply;
2215 struct datapath *dp;
2216 struct vport *vport;
2217 unsigned int new_headroom;
2220 reply = ovs_vport_cmd_alloc_info();
2225 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2226 err = PTR_ERR(vport);
2228 goto exit_unlock_free;
2230 if (vport->port_no == OVSP_LOCAL) {
2232 goto exit_unlock_free;
2235 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2236 info->snd_portid, info->snd_seq, 0,
2237 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2240 /* the vport deletion may trigger dp headroom update */
2242 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2243 update_headroom = true;
2245 netdev_reset_rx_headroom(vport->dev);
2246 ovs_dp_detach_port(vport);
2248 if (update_headroom) {
2249 new_headroom = ovs_get_max_headroom(dp);
2251 if (new_headroom < dp->max_headroom)
2252 ovs_update_headroom(dp, new_headroom);
2256 ovs_notify(&dp_vport_genl_family, reply, info);
2265 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2267 struct nlattr **a = info->attrs;
2268 struct ovs_header *ovs_header = info->userhdr;
2269 struct sk_buff *reply;
2270 struct vport *vport;
2273 reply = ovs_vport_cmd_alloc_info();
2278 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2279 err = PTR_ERR(vport);
2281 goto exit_unlock_free;
2282 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2283 info->snd_portid, info->snd_seq, 0,
2284 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2288 return genlmsg_reply(reply, info);
2296 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2298 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2299 struct datapath *dp;
2300 int bucket = cb->args[0], skip = cb->args[1];
2304 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2309 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2310 struct vport *vport;
2313 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2315 ovs_vport_cmd_fill_info(vport, skb,
2317 NETLINK_CB(cb->skb).portid,
2337 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2338 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2339 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2340 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2341 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2342 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2343 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2344 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2345 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2348 static const struct genl_ops dp_vport_genl_ops[] = {
2349 { .cmd = OVS_VPORT_CMD_NEW,
2350 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2351 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2352 .doit = ovs_vport_cmd_new
2354 { .cmd = OVS_VPORT_CMD_DEL,
2355 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2356 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2357 .doit = ovs_vport_cmd_del
2359 { .cmd = OVS_VPORT_CMD_GET,
2360 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2361 .flags = 0, /* OK for unprivileged users. */
2362 .doit = ovs_vport_cmd_get,
2363 .dumpit = ovs_vport_cmd_dump
2365 { .cmd = OVS_VPORT_CMD_SET,
2366 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2367 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2368 .doit = ovs_vport_cmd_set,
2372 struct genl_family dp_vport_genl_family __ro_after_init = {
2373 .hdrsize = sizeof(struct ovs_header),
2374 .name = OVS_VPORT_FAMILY,
2375 .version = OVS_VPORT_VERSION,
2376 .maxattr = OVS_VPORT_ATTR_MAX,
2377 .policy = vport_policy,
2379 .parallel_ops = true,
2380 .ops = dp_vport_genl_ops,
2381 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2382 .mcgrps = &ovs_dp_vport_multicast_group,
2384 .module = THIS_MODULE,
2387 static struct genl_family * const dp_genl_families[] = {
2388 &dp_datapath_genl_family,
2389 &dp_vport_genl_family,
2390 &dp_flow_genl_family,
2391 &dp_packet_genl_family,
2392 &dp_meter_genl_family,
2393 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2394 &dp_ct_limit_genl_family,
2398 static void dp_unregister_genl(int n_families)
2402 for (i = 0; i < n_families; i++)
2403 genl_unregister_family(dp_genl_families[i]);
2406 static int __init dp_register_genl(void)
2411 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2413 err = genl_register_family(dp_genl_families[i]);
2421 dp_unregister_genl(i);
2425 static int __net_init ovs_init_net(struct net *net)
2427 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2429 INIT_LIST_HEAD(&ovs_net->dps);
2430 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2431 return ovs_ct_init(net);
2434 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2435 struct list_head *head)
2437 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2438 struct datapath *dp;
2440 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2443 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2444 struct vport *vport;
2446 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2447 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2450 if (dev_net(vport->dev) == dnet)
2451 list_add(&vport->detach_list, head);
2457 static void __net_exit ovs_exit_net(struct net *dnet)
2459 struct datapath *dp, *dp_next;
2460 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2461 struct vport *vport, *vport_next;
2467 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2470 down_read(&net_rwsem);
2472 list_vports_from_net(net, dnet, &head);
2473 up_read(&net_rwsem);
2475 /* Detach all vports from given namespace. */
2476 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2477 list_del(&vport->detach_list);
2478 ovs_dp_detach_port(vport);
2483 cancel_work_sync(&ovs_net->dp_notify_work);
2486 static struct pernet_operations ovs_net_ops = {
2487 .init = ovs_init_net,
2488 .exit = ovs_exit_net,
2490 .size = sizeof(struct ovs_net),
2493 static int __init dp_init(void)
2497 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof_field(struct sk_buff, cb));
2499 pr_info("Open vSwitch switching datapath\n");
2501 err = action_fifos_init();
2505 err = ovs_internal_dev_rtnl_link_register();
2507 goto error_action_fifos_exit;
2509 err = ovs_flow_init();
2511 goto error_unreg_rtnl_link;
2513 err = ovs_vport_init();
2515 goto error_flow_exit;
2517 err = register_pernet_device(&ovs_net_ops);
2519 goto error_vport_exit;
2521 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2523 goto error_netns_exit;
2525 err = ovs_netdev_init();
2527 goto error_unreg_notifier;
2529 err = dp_register_genl();
2531 goto error_unreg_netdev;
2537 error_unreg_notifier:
2538 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2540 unregister_pernet_device(&ovs_net_ops);
2545 error_unreg_rtnl_link:
2546 ovs_internal_dev_rtnl_link_unregister();
2547 error_action_fifos_exit:
2548 action_fifos_exit();
2553 static void dp_cleanup(void)
2555 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2557 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2558 unregister_pernet_device(&ovs_net_ops);
2562 ovs_internal_dev_rtnl_link_unregister();
2563 action_fifos_exit();
2566 module_init(dp_init);
2567 module_exit(dp_cleanup);
2569 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2570 MODULE_LICENSE("GPL");
2571 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2572 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2573 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2574 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2575 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2576 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);