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 <net/genetlink.h>
52 #include <net/net_namespace.h>
53 #include <net/netns/generic.h>
57 #include "flow_table.h"
58 #include "flow_netlink.h"
60 #include "vport-internal_dev.h"
61 #include "vport-netdev.h"
63 int ovs_net_id __read_mostly;
65 static struct genl_family dp_packet_genl_family;
66 static struct genl_family dp_flow_genl_family;
67 static struct genl_family dp_datapath_genl_family;
69 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
70 .name = OVS_FLOW_MCGROUP
73 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
74 .name = OVS_DATAPATH_MCGROUP
77 struct genl_multicast_group ovs_dp_vport_multicast_group = {
78 .name = OVS_VPORT_MCGROUP
81 /* Check if need to build a reply message.
82 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
86 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
87 genl_has_listeners(family, genl_info_net(info)->genl_sock,
91 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
92 struct sk_buff *skb, struct genl_info *info)
94 genl_notify(family, skb, genl_info_net(info),
95 info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
101 * All writes e.g. Writes to device state (add/remove datapath, port, set
102 * operations on vports, etc.), Writes to other state (flow table
103 * modifications, set miscellaneous datapath parameters, etc.) are protected
106 * Reads are protected by RCU.
108 * There are a few special cases (mostly stats) that have their own
109 * synchronization but they nest under all of above and don't interact with
112 * The RTNL lock nests inside ovs_mutex.
115 static DEFINE_MUTEX(ovs_mutex);
119 mutex_lock(&ovs_mutex);
122 void ovs_unlock(void)
124 mutex_unlock(&ovs_mutex);
127 #ifdef CONFIG_LOCKDEP
128 int lockdep_ovsl_is_held(void)
131 return lockdep_is_held(&ovs_mutex);
137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
138 const struct sw_flow_key *,
139 const struct dp_upcall_info *);
140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 const struct sw_flow_key *key,
142 const struct dp_upcall_info *);
144 /* Must be called with rcu_read_lock. */
145 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
147 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
150 struct vport *vport = ovs_internal_dev_get_vport(dev);
158 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
159 * returned dp pointer valid. */
160 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
164 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
166 dp = get_dp_rcu(net, dp_ifindex);
172 /* Must be called with rcu_read_lock or ovs_mutex. */
173 const char *ovs_dp_name(const struct datapath *dp)
175 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
176 return vport->ops->get_name(vport);
179 static int get_dpifindex(const struct datapath *dp)
186 local = ovs_vport_rcu(dp, OVSP_LOCAL);
188 ifindex = netdev_vport_priv(local)->dev->ifindex;
197 static void destroy_dp_rcu(struct rcu_head *rcu)
199 struct datapath *dp = container_of(rcu, struct datapath, rcu);
201 ovs_flow_tbl_destroy(&dp->table);
202 free_percpu(dp->stats_percpu);
203 release_net(ovs_dp_get_net(dp));
208 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
211 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
214 /* Called with ovs_mutex or RCU read lock. */
215 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
218 struct hlist_head *head;
220 head = vport_hash_bucket(dp, port_no);
221 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
222 if (vport->port_no == port_no)
228 /* Called with ovs_mutex. */
229 static struct vport *new_vport(const struct vport_parms *parms)
233 vport = ovs_vport_add(parms);
234 if (!IS_ERR(vport)) {
235 struct datapath *dp = parms->dp;
236 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
238 hlist_add_head_rcu(&vport->dp_hash_node, head);
243 void ovs_dp_detach_port(struct vport *p)
247 /* First drop references to device. */
248 hlist_del_rcu(&p->dp_hash_node);
250 /* Then destroy it. */
254 /* Must be called with rcu_read_lock. */
255 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
257 const struct vport *p = OVS_CB(skb)->input_vport;
258 struct datapath *dp = p->dp;
259 struct sw_flow *flow;
260 struct sw_flow_actions *sf_acts;
261 struct dp_stats_percpu *stats;
265 stats = this_cpu_ptr(dp->stats_percpu);
268 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
270 if (unlikely(!flow)) {
271 struct dp_upcall_info upcall;
274 upcall.cmd = OVS_PACKET_CMD_MISS;
275 upcall.userdata = NULL;
276 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 upcall.egress_tun_info = NULL;
279 error = ovs_dp_upcall(dp, skb, key, &upcall);
285 stats_counter = &stats->n_missed;
289 ovs_flow_stats_update(flow, key->tp.flags, skb);
291 sf_acts = rcu_dereference(flow->sf_acts);
292 ovs_execute_actions(dp, skb, key, sf_acts);
293 stats_counter = &stats->n_hit;
296 /* Update datapath statistics. */
297 u64_stats_update_begin(&stats->syncp);
299 stats->n_mask_hit += n_mask_hit;
300 u64_stats_update_end(&stats->syncp);
303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
304 const struct sw_flow_key *key,
305 const struct dp_upcall_info *upcall_info)
307 struct dp_stats_percpu *stats;
310 if (upcall_info->portid == 0) {
315 if (!skb_is_gso(skb))
316 err = queue_userspace_packet(dp, skb, key, upcall_info);
318 err = queue_gso_packets(dp, skb, key, upcall_info);
325 stats = this_cpu_ptr(dp->stats_percpu);
327 u64_stats_update_begin(&stats->syncp);
329 u64_stats_update_end(&stats->syncp);
334 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
335 const struct sw_flow_key *key,
336 const struct dp_upcall_info *upcall_info)
338 unsigned short gso_type = skb_shinfo(skb)->gso_type;
339 struct sw_flow_key later_key;
340 struct sk_buff *segs, *nskb;
341 struct ovs_skb_cb ovs_cb;
344 ovs_cb = *OVS_CB(skb);
345 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
346 *OVS_CB(skb) = ovs_cb;
348 return PTR_ERR(segs);
350 if (gso_type & SKB_GSO_UDP) {
351 /* The initial flow key extracted by ovs_flow_key_extract()
352 * in this case is for a first fragment, so we need to
353 * properly mark later fragments.
356 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
359 /* Queue all of the segments. */
362 *OVS_CB(skb) = ovs_cb;
363 if (gso_type & SKB_GSO_UDP && skb != segs)
366 err = queue_userspace_packet(dp, skb, key, upcall_info);
370 } while ((skb = skb->next));
372 /* Free all of the segments. */
380 } while ((skb = nskb));
384 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
387 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
388 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
389 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391 /* OVS_PACKET_ATTR_USERDATA */
392 if (upcall_info->userdata)
393 size += NLA_ALIGN(upcall_info->userdata->nla_len);
395 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
396 if (upcall_info->egress_tun_info)
397 size += nla_total_size(ovs_tun_key_attr_size());
402 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
403 const struct sw_flow_key *key,
404 const struct dp_upcall_info *upcall_info)
406 struct ovs_header *upcall;
407 struct sk_buff *nskb = NULL;
408 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
410 struct genl_info info = {
411 #ifdef HAVE_GENLMSG_NEW_UNICAST
412 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
414 .snd_portid = upcall_info->portid,
420 dp_ifindex = get_dpifindex(dp);
424 if (vlan_tx_tag_present(skb)) {
425 nskb = skb_clone(skb, GFP_ATOMIC);
429 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
433 vlan_set_tci(nskb, 0);
438 if (nla_attr_size(skb->len) > USHRT_MAX) {
443 /* Complete checksum if needed */
444 if (skb->ip_summed == CHECKSUM_PARTIAL &&
445 (err = skb_checksum_help(skb)))
448 /* Older versions of OVS user space enforce alignment of the last
449 * Netlink attribute to NLA_ALIGNTO which would require extensive
450 * padding logic. Only perform zerocopy if padding is not required.
452 if (dp->user_features & OVS_DP_F_UNALIGNED)
453 hlen = skb_zerocopy_headlen(skb);
457 len = upcall_msg_size(upcall_info, hlen);
458 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
464 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
465 0, upcall_info->cmd);
466 upcall->dp_ifindex = dp_ifindex;
468 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
469 err = ovs_nla_put_flow(key, key, user_skb);
471 nla_nest_end(user_skb, nla);
473 if (upcall_info->userdata)
474 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
475 nla_len(upcall_info->userdata),
476 nla_data(upcall_info->userdata));
478 if (upcall_info->egress_tun_info) {
479 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
480 err = ovs_nla_put_egress_tunnel_key(user_skb,
481 upcall_info->egress_tun_info);
483 nla_nest_end(user_skb, nla);
486 /* Only reserve room for attribute header, packet data is added
487 * in skb_zerocopy() */
488 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
492 nla->nla_len = nla_attr_size(skb->len);
494 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
498 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
499 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
500 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
503 memset(skb_put(user_skb, plen), 0, plen);
506 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
508 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
519 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
521 struct ovs_header *ovs_header = info->userhdr;
522 struct nlattr **a = info->attrs;
523 struct sw_flow_actions *acts;
524 struct sk_buff *packet;
525 struct sw_flow *flow;
526 struct sw_flow_actions *sf_acts;
529 struct vport *input_vport;
532 bool log = !a[OVS_FLOW_ATTR_PROBE];
535 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
536 !a[OVS_PACKET_ATTR_ACTIONS])
539 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
540 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
544 skb_reserve(packet, NET_IP_ALIGN);
546 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
548 skb_reset_mac_header(packet);
549 eth = eth_hdr(packet);
551 /* Normally, setting the skb 'protocol' field would be handled by a
552 * call to eth_type_trans(), but it assumes there's a sending
553 * device, which we may not have. */
554 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
555 packet->protocol = eth->h_proto;
557 packet->protocol = htons(ETH_P_802_2);
559 /* Build an sw_flow for sending this packet. */
560 flow = ovs_flow_alloc();
565 err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
570 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
571 &flow->key, &acts, log);
575 rcu_assign_pointer(flow->sf_acts, acts);
576 OVS_CB(packet)->egress_tun_info = NULL;
577 packet->priority = flow->key.phy.priority;
578 packet->mark = flow->key.phy.skb_mark;
581 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
586 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
588 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
593 OVS_CB(packet)->input_vport = input_vport;
594 sf_acts = rcu_dereference(flow->sf_acts);
597 err = ovs_execute_actions(dp, packet, &flow->key, sf_acts);
601 ovs_flow_free(flow, false);
607 ovs_flow_free(flow, false);
614 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
615 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
616 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
617 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
620 static struct genl_ops dp_packet_genl_ops[] = {
621 { .cmd = OVS_PACKET_CMD_EXECUTE,
622 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
623 .policy = packet_policy,
624 .doit = ovs_packet_cmd_execute
628 static struct genl_family dp_packet_genl_family = {
629 .id = GENL_ID_GENERATE,
630 .hdrsize = sizeof(struct ovs_header),
631 .name = OVS_PACKET_FAMILY,
632 .version = OVS_PACKET_VERSION,
633 .maxattr = OVS_PACKET_ATTR_MAX,
635 .parallel_ops = true,
636 .ops = dp_packet_genl_ops,
637 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
640 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
641 struct ovs_dp_megaflow_stats *mega_stats)
645 memset(mega_stats, 0, sizeof(*mega_stats));
647 stats->n_flows = ovs_flow_tbl_count(&dp->table);
648 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
650 stats->n_hit = stats->n_missed = stats->n_lost = 0;
652 for_each_possible_cpu(i) {
653 const struct dp_stats_percpu *percpu_stats;
654 struct dp_stats_percpu local_stats;
657 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
660 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
661 local_stats = *percpu_stats;
662 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
664 stats->n_hit += local_stats.n_hit;
665 stats->n_missed += local_stats.n_missed;
666 stats->n_lost += local_stats.n_lost;
667 mega_stats->n_mask_hit += local_stats.n_mask_hit;
671 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
673 return NLMSG_ALIGN(sizeof(struct ovs_header))
674 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
675 + nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
676 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
677 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
678 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
679 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
682 /* Called with ovs_mutex or RCU read lock. */
683 static int ovs_flow_cmd_fill_match(const struct sw_flow *flow,
690 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
694 err = ovs_nla_put_flow(&flow->unmasked_key,
695 &flow->unmasked_key, skb);
698 nla_nest_end(skb, nla);
700 /* Fill flow mask. */
701 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
705 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
708 nla_nest_end(skb, nla);
713 /* Called with ovs_mutex or RCU read lock. */
714 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
717 struct ovs_flow_stats stats;
721 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
724 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
727 if (stats.n_packets &&
728 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
731 if ((u8)ntohs(tcp_flags) &&
732 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
738 /* Called with ovs_mutex or RCU read lock. */
739 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
740 struct sk_buff *skb, int skb_orig_len)
742 struct nlattr *start;
745 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
746 * this is the first flow to be dumped into 'skb'. This is unusual for
747 * Netlink but individual action lists can be longer than
748 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
749 * The userspace caller can always fetch the actions separately if it
750 * really wants them. (Most userspace callers in fact don't care.)
752 * This can only fail for dump operations because the skb is always
753 * properly sized for single flows.
755 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
757 const struct sw_flow_actions *sf_acts;
759 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
760 err = ovs_nla_put_actions(sf_acts->actions,
761 sf_acts->actions_len, skb);
764 nla_nest_end(skb, start);
769 nla_nest_cancel(skb, start);
771 } else if (skb_orig_len) {
778 /* Called with ovs_mutex or RCU read lock. */
779 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
780 struct sk_buff *skb, u32 portid,
781 u32 seq, u32 flags, u8 cmd)
783 const int skb_orig_len = skb->len;
784 struct ovs_header *ovs_header;
787 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
790 ovs_header->dp_ifindex = dp_ifindex;
792 err = ovs_flow_cmd_fill_match(flow, skb);
796 err = ovs_flow_cmd_fill_stats(flow, skb);
800 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
804 return genlmsg_end(skb, ovs_header);
807 genlmsg_cancel(skb, ovs_header);
811 /* May not be called with RCU read lock. */
812 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
813 struct genl_info *info,
818 if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
819 GROUP_ID(&ovs_dp_flow_multicast_group)))
822 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
825 return ERR_PTR(-ENOMEM);
830 /* Called with ovs_mutex. */
831 static struct sk_buff *ovs_flow_cmd_build_info(struct datapath *dp,
832 const struct sw_flow *flow,
834 struct genl_info *info, u8 cmd,
840 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
842 if (IS_ERR_OR_NULL(skb))
845 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
846 info->snd_portid, info->snd_seq, 0,
852 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
854 struct nlattr **a = info->attrs;
855 struct ovs_header *ovs_header = info->userhdr;
856 struct sw_flow *flow, *new_flow;
857 struct sw_flow_mask mask;
858 struct sk_buff *reply;
860 struct sw_flow_actions *acts;
861 struct sw_flow_match match;
863 bool log = !a[OVS_FLOW_ATTR_PROBE];
865 /* Must have key and actions. */
867 if (!a[OVS_FLOW_ATTR_KEY]) {
868 OVS_NLERR(log, "Flow key attribute not present in new flow.");
871 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
873 "Flow actions attribute not present in new flow.");
877 /* Most of the time we need to allocate a new flow, do it before
879 new_flow = ovs_flow_alloc();
880 if (IS_ERR(new_flow)) {
881 error = PTR_ERR(new_flow);
886 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
887 error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
888 a[OVS_FLOW_ATTR_MASK], log);
892 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
894 /* Validate actions. */
895 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
900 "Flow actions may not be safe on all matching packets."
905 reply = ovs_flow_cmd_alloc_info(acts, info, false);
907 error = PTR_ERR(reply);
912 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
917 /* Check if this is a duplicate flow */
918 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
920 rcu_assign_pointer(new_flow->sf_acts, acts);
922 /* Put flow in bucket. */
923 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
924 if (unlikely(error)) {
929 if (unlikely(reply)) {
930 error = ovs_flow_cmd_fill_info(new_flow,
931 ovs_header->dp_ifindex,
932 reply, info->snd_portid,
939 struct sw_flow_actions *old_acts;
941 /* Bail out if we're not allowed to modify an existing flow.
942 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
943 * because Generic Netlink treats the latter as a dump
944 * request. We also accept NLM_F_EXCL in case that bug ever
947 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
952 /* The unmasked key has to be the same for flow updates. */
953 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
954 /* Look for any overlapping flow. */
955 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
961 /* Update actions. */
962 old_acts = ovsl_dereference(flow->sf_acts);
963 rcu_assign_pointer(flow->sf_acts, acts);
965 if (unlikely(reply)) {
966 error = ovs_flow_cmd_fill_info(flow,
967 ovs_header->dp_ifindex,
968 reply, info->snd_portid,
975 ovs_nla_free_flow_actions(old_acts);
976 ovs_flow_free(new_flow, false);
980 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
989 ovs_flow_free(new_flow, false);
994 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
995 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
996 const struct sw_flow_key *key,
997 const struct sw_flow_mask *mask,
1000 struct sw_flow_actions *acts;
1001 struct sw_flow_key masked_key;
1004 ovs_flow_mask_key(&masked_key, key, mask);
1005 error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1008 "Actions may not be safe on all matching packets.");
1009 return ERR_PTR(error);
1015 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1017 struct nlattr **a = info->attrs;
1018 struct ovs_header *ovs_header = info->userhdr;
1019 struct sw_flow_key key;
1020 struct sw_flow *flow;
1021 struct sw_flow_mask mask;
1022 struct sk_buff *reply = NULL;
1023 struct datapath *dp;
1024 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1025 struct sw_flow_match match;
1027 bool log = !a[OVS_FLOW_ATTR_PROBE];
1031 if (!a[OVS_FLOW_ATTR_KEY]) {
1032 OVS_NLERR(log, "Flow key attribute not present in set flow.");
1036 ovs_match_init(&match, &key, &mask);
1037 error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1038 a[OVS_FLOW_ATTR_MASK], log);
1042 /* Validate actions. */
1043 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1044 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1047 error = PTR_ERR(acts);
1051 /* Can allocate before locking if have acts. */
1052 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1053 if (IS_ERR(reply)) {
1054 error = PTR_ERR(reply);
1055 goto err_kfree_acts;
1060 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1061 if (unlikely(!dp)) {
1063 goto err_unlock_ovs;
1065 /* Check that the flow exists. */
1066 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1067 if (unlikely(!flow)) {
1069 goto err_unlock_ovs;
1072 /* Update actions, if present. */
1074 old_acts = ovsl_dereference(flow->sf_acts);
1075 rcu_assign_pointer(flow->sf_acts, acts);
1077 if (unlikely(reply)) {
1078 error = ovs_flow_cmd_fill_info(flow,
1079 ovs_header->dp_ifindex,
1080 reply, info->snd_portid,
1086 /* Could not alloc without acts before locking. */
1087 reply = ovs_flow_cmd_build_info(dp, flow,
1088 ovs_header->dp_ifindex,
1089 info, OVS_FLOW_CMD_NEW, false);
1090 if (unlikely(IS_ERR(reply))) {
1091 error = PTR_ERR(reply);
1092 goto err_unlock_ovs;
1097 if (a[OVS_FLOW_ATTR_CLEAR])
1098 ovs_flow_stats_clear(flow);
1102 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1104 ovs_nla_free_flow_actions(old_acts);
1116 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1118 struct nlattr **a = info->attrs;
1119 struct ovs_header *ovs_header = info->userhdr;
1120 struct sw_flow_key key;
1121 struct sk_buff *reply;
1122 struct sw_flow *flow;
1123 struct datapath *dp;
1124 struct sw_flow_match match;
1126 bool log = !a[OVS_FLOW_ATTR_PROBE];
1128 if (!a[OVS_FLOW_ATTR_KEY]) {
1130 "Flow get message rejected, Key attribute missing.");
1134 ovs_match_init(&match, &key, NULL);
1135 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log);
1140 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1146 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1152 reply = ovs_flow_cmd_build_info(dp, flow, ovs_header->dp_ifindex, info,
1153 OVS_FLOW_CMD_NEW, true);
1154 if (IS_ERR(reply)) {
1155 err = PTR_ERR(reply);
1160 return genlmsg_reply(reply, info);
1166 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1168 struct nlattr **a = info->attrs;
1169 struct ovs_header *ovs_header = info->userhdr;
1170 struct sw_flow_key key;
1171 struct sk_buff *reply;
1172 struct sw_flow *flow;
1173 struct datapath *dp;
1174 struct sw_flow_match match;
1176 bool log = !a[OVS_FLOW_ATTR_PROBE];
1178 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1179 ovs_match_init(&match, &key, NULL);
1180 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1187 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1188 if (unlikely(!dp)) {
1192 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1193 err = ovs_flow_tbl_flush(&dp->table);
1196 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1197 if (unlikely(!flow)) {
1202 ovs_flow_tbl_remove(&dp->table, flow);
1205 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1208 if (likely(reply)) {
1209 if (likely(!IS_ERR(reply))) {
1210 rcu_read_lock(); /* Keep RCU checker happy. */
1211 err = ovs_flow_cmd_fill_info(flow,
1212 ovs_header->dp_ifindex,
1213 reply, info->snd_portid,
1218 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1220 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1221 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1226 ovs_flow_free(flow, true);
1233 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1235 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1236 struct table_instance *ti;
1237 struct datapath *dp;
1240 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1246 ti = rcu_dereference(dp->table.ti);
1248 struct sw_flow *flow;
1251 bucket = cb->args[0];
1253 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1257 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1258 NETLINK_CB(cb->skb).portid,
1259 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1260 OVS_FLOW_CMD_NEW) < 0)
1263 cb->args[0] = bucket;
1270 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1271 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1272 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1273 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1274 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1275 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1278 static struct genl_ops dp_flow_genl_ops[] = {
1279 { .cmd = OVS_FLOW_CMD_NEW,
1280 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1281 .policy = flow_policy,
1282 .doit = ovs_flow_cmd_new
1284 { .cmd = OVS_FLOW_CMD_DEL,
1285 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1286 .policy = flow_policy,
1287 .doit = ovs_flow_cmd_del
1289 { .cmd = OVS_FLOW_CMD_GET,
1290 .flags = 0, /* OK for unprivileged users. */
1291 .policy = flow_policy,
1292 .doit = ovs_flow_cmd_get,
1293 .dumpit = ovs_flow_cmd_dump
1295 { .cmd = OVS_FLOW_CMD_SET,
1296 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1297 .policy = flow_policy,
1298 .doit = ovs_flow_cmd_set,
1302 static struct genl_family dp_flow_genl_family = {
1303 .id = GENL_ID_GENERATE,
1304 .hdrsize = sizeof(struct ovs_header),
1305 .name = OVS_FLOW_FAMILY,
1306 .version = OVS_FLOW_VERSION,
1307 .maxattr = OVS_FLOW_ATTR_MAX,
1309 .parallel_ops = true,
1310 .ops = dp_flow_genl_ops,
1311 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1312 .mcgrps = &ovs_dp_flow_multicast_group,
1316 static size_t ovs_dp_cmd_msg_size(void)
1318 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1320 msgsize += nla_total_size(IFNAMSIZ);
1321 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1322 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1323 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1328 /* Called with ovs_mutex or RCU read lock. */
1329 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1330 u32 portid, u32 seq, u32 flags, u8 cmd)
1332 struct ovs_header *ovs_header;
1333 struct ovs_dp_stats dp_stats;
1334 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1337 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1342 ovs_header->dp_ifindex = get_dpifindex(dp);
1344 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1346 goto nla_put_failure;
1348 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1349 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1351 goto nla_put_failure;
1353 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1354 sizeof(struct ovs_dp_megaflow_stats),
1355 &dp_megaflow_stats))
1356 goto nla_put_failure;
1358 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1359 goto nla_put_failure;
1361 return genlmsg_end(skb, ovs_header);
1364 genlmsg_cancel(skb, ovs_header);
1369 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1371 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1374 /* Called with rcu_read_lock or ovs_mutex. */
1375 static struct datapath *lookup_datapath(struct net *net,
1376 const struct ovs_header *ovs_header,
1377 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1379 struct datapath *dp;
1381 if (!a[OVS_DP_ATTR_NAME])
1382 dp = get_dp(net, ovs_header->dp_ifindex);
1384 struct vport *vport;
1386 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1387 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1389 return dp ? dp : ERR_PTR(-ENODEV);
1392 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1394 struct datapath *dp;
1396 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1400 WARN(dp->user_features, "Dropping previously announced user features\n");
1401 dp->user_features = 0;
1404 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1406 if (a[OVS_DP_ATTR_USER_FEATURES])
1407 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1410 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1412 struct nlattr **a = info->attrs;
1413 struct vport_parms parms;
1414 struct sk_buff *reply;
1415 struct datapath *dp;
1416 struct vport *vport;
1417 struct ovs_net *ovs_net;
1421 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1424 reply = ovs_dp_cmd_alloc_info(info);
1429 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1431 goto err_free_reply;
1433 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1435 /* Allocate table. */
1436 err = ovs_flow_tbl_init(&dp->table);
1440 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1441 if (!dp->stats_percpu) {
1443 goto err_destroy_table;
1446 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1450 goto err_destroy_percpu;
1453 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1454 INIT_HLIST_HEAD(&dp->ports[i]);
1456 /* Set up our datapath device. */
1457 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1458 parms.type = OVS_VPORT_TYPE_INTERNAL;
1459 parms.options = NULL;
1461 parms.port_no = OVSP_LOCAL;
1462 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1464 ovs_dp_change(dp, a);
1466 /* So far only local changes have been made, now need the lock. */
1469 vport = new_vport(&parms);
1470 if (IS_ERR(vport)) {
1471 err = PTR_ERR(vport);
1475 if (err == -EEXIST) {
1476 /* An outdated user space instance that does not understand
1477 * the concept of user_features has attempted to create a new
1478 * datapath and is likely to reuse it. Drop all user features.
1480 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1481 ovs_dp_reset_user_features(skb, info);
1484 goto err_destroy_ports_array;
1487 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1488 info->snd_seq, 0, OVS_DP_CMD_NEW);
1491 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1492 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1496 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1499 err_destroy_ports_array:
1503 free_percpu(dp->stats_percpu);
1505 ovs_flow_tbl_destroy(&dp->table);
1507 release_net(ovs_dp_get_net(dp));
1515 /* Called with ovs_mutex. */
1516 static void __dp_destroy(struct datapath *dp)
1520 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1521 struct vport *vport;
1522 struct hlist_node *n;
1524 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1525 if (vport->port_no != OVSP_LOCAL)
1526 ovs_dp_detach_port(vport);
1529 list_del_rcu(&dp->list_node);
1531 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1532 * all ports in datapath are destroyed first before freeing datapath.
1534 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1536 /* RCU destroy the flow table */
1537 call_rcu(&dp->rcu, destroy_dp_rcu);
1540 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1542 struct sk_buff *reply;
1543 struct datapath *dp;
1546 reply = ovs_dp_cmd_alloc_info(info);
1551 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1554 goto err_unlock_free;
1556 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1557 info->snd_seq, 0, OVS_DP_CMD_DEL);
1563 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1572 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1574 struct sk_buff *reply;
1575 struct datapath *dp;
1578 reply = ovs_dp_cmd_alloc_info(info);
1583 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1586 goto err_unlock_free;
1588 ovs_dp_change(dp, info->attrs);
1590 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1591 info->snd_seq, 0, OVS_DP_CMD_NEW);
1595 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1604 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1606 struct sk_buff *reply;
1607 struct datapath *dp;
1610 reply = ovs_dp_cmd_alloc_info(info);
1615 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1618 goto err_unlock_free;
1620 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1621 info->snd_seq, 0, OVS_DP_CMD_NEW);
1625 return genlmsg_reply(reply, info);
1633 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1635 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1636 struct datapath *dp;
1637 int skip = cb->args[0];
1641 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1643 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1644 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1645 OVS_DP_CMD_NEW) < 0)
1656 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1657 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1658 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1659 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1662 static struct genl_ops dp_datapath_genl_ops[] = {
1663 { .cmd = OVS_DP_CMD_NEW,
1664 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1665 .policy = datapath_policy,
1666 .doit = ovs_dp_cmd_new
1668 { .cmd = OVS_DP_CMD_DEL,
1669 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1670 .policy = datapath_policy,
1671 .doit = ovs_dp_cmd_del
1673 { .cmd = OVS_DP_CMD_GET,
1674 .flags = 0, /* OK for unprivileged users. */
1675 .policy = datapath_policy,
1676 .doit = ovs_dp_cmd_get,
1677 .dumpit = ovs_dp_cmd_dump
1679 { .cmd = OVS_DP_CMD_SET,
1680 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1681 .policy = datapath_policy,
1682 .doit = ovs_dp_cmd_set,
1686 static struct genl_family dp_datapath_genl_family = {
1687 .id = GENL_ID_GENERATE,
1688 .hdrsize = sizeof(struct ovs_header),
1689 .name = OVS_DATAPATH_FAMILY,
1690 .version = OVS_DATAPATH_VERSION,
1691 .maxattr = OVS_DP_ATTR_MAX,
1693 .parallel_ops = true,
1694 .ops = dp_datapath_genl_ops,
1695 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1696 .mcgrps = &ovs_dp_datapath_multicast_group,
1700 /* Called with ovs_mutex or RCU read lock. */
1701 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1702 u32 portid, u32 seq, u32 flags, u8 cmd)
1704 struct ovs_header *ovs_header;
1705 struct ovs_vport_stats vport_stats;
1708 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1713 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1715 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1716 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1717 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1718 goto nla_put_failure;
1720 ovs_vport_get_stats(vport, &vport_stats);
1721 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1723 goto nla_put_failure;
1725 if (ovs_vport_get_upcall_portids(vport, skb))
1726 goto nla_put_failure;
1728 err = ovs_vport_get_options(vport, skb);
1729 if (err == -EMSGSIZE)
1732 return genlmsg_end(skb, ovs_header);
1737 genlmsg_cancel(skb, ovs_header);
1741 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1743 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1746 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1747 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1750 struct sk_buff *skb;
1753 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1755 return ERR_PTR(-ENOMEM);
1757 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1763 /* Called with ovs_mutex or RCU read lock. */
1764 static struct vport *lookup_vport(struct net *net,
1765 const struct ovs_header *ovs_header,
1766 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1768 struct datapath *dp;
1769 struct vport *vport;
1771 if (a[OVS_VPORT_ATTR_NAME]) {
1772 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1774 return ERR_PTR(-ENODEV);
1775 if (ovs_header->dp_ifindex &&
1776 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1777 return ERR_PTR(-ENODEV);
1779 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1780 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1782 if (port_no >= DP_MAX_PORTS)
1783 return ERR_PTR(-EFBIG);
1785 dp = get_dp(net, ovs_header->dp_ifindex);
1787 return ERR_PTR(-ENODEV);
1789 vport = ovs_vport_ovsl_rcu(dp, port_no);
1791 return ERR_PTR(-ENODEV);
1794 return ERR_PTR(-EINVAL);
1797 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1799 struct nlattr **a = info->attrs;
1800 struct ovs_header *ovs_header = info->userhdr;
1801 struct vport_parms parms;
1802 struct sk_buff *reply;
1803 struct vport *vport;
1804 struct datapath *dp;
1808 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1809 !a[OVS_VPORT_ATTR_UPCALL_PID])
1812 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1813 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1814 if (port_no >= DP_MAX_PORTS)
1817 reply = ovs_vport_cmd_alloc_info();
1822 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1825 goto exit_unlock_free;
1828 vport = ovs_vport_ovsl(dp, port_no);
1831 goto exit_unlock_free;
1833 for (port_no = 1; ; port_no++) {
1834 if (port_no >= DP_MAX_PORTS) {
1836 goto exit_unlock_free;
1838 vport = ovs_vport_ovsl(dp, port_no);
1844 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1845 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1846 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1848 parms.port_no = port_no;
1849 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1851 vport = new_vport(&parms);
1852 err = PTR_ERR(vport);
1854 goto exit_unlock_free;
1856 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1857 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1861 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1870 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1872 struct nlattr **a = info->attrs;
1873 struct sk_buff *reply;
1874 struct vport *vport;
1877 reply = ovs_vport_cmd_alloc_info();
1882 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1883 err = PTR_ERR(vport);
1885 goto exit_unlock_free;
1887 if (a[OVS_VPORT_ATTR_TYPE] &&
1888 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1890 goto exit_unlock_free;
1893 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1894 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1896 goto exit_unlock_free;
1899 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1900 err = ovs_vport_set_upcall_portids(vport,
1901 a[OVS_VPORT_ATTR_UPCALL_PID]);
1903 goto exit_unlock_free;
1906 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1907 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1911 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1920 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1922 struct nlattr **a = info->attrs;
1923 struct sk_buff *reply;
1924 struct vport *vport;
1927 reply = ovs_vport_cmd_alloc_info();
1932 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1933 err = PTR_ERR(vport);
1935 goto exit_unlock_free;
1937 if (vport->port_no == OVSP_LOCAL) {
1939 goto exit_unlock_free;
1942 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1943 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1945 ovs_dp_detach_port(vport);
1948 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1957 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1959 struct nlattr **a = info->attrs;
1960 struct ovs_header *ovs_header = info->userhdr;
1961 struct sk_buff *reply;
1962 struct vport *vport;
1965 reply = ovs_vport_cmd_alloc_info();
1970 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1971 err = PTR_ERR(vport);
1973 goto exit_unlock_free;
1974 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1975 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1979 return genlmsg_reply(reply, info);
1987 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1989 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1990 struct datapath *dp;
1991 int bucket = cb->args[0], skip = cb->args[1];
1995 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2000 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2001 struct vport *vport;
2004 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2006 ovs_vport_cmd_fill_info(vport, skb,
2007 NETLINK_CB(cb->skb).portid,
2010 OVS_VPORT_CMD_NEW) < 0)
2026 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2027 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2028 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2029 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2030 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2031 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2032 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2035 static struct genl_ops dp_vport_genl_ops[] = {
2036 { .cmd = OVS_VPORT_CMD_NEW,
2037 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2038 .policy = vport_policy,
2039 .doit = ovs_vport_cmd_new
2041 { .cmd = OVS_VPORT_CMD_DEL,
2042 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2043 .policy = vport_policy,
2044 .doit = ovs_vport_cmd_del
2046 { .cmd = OVS_VPORT_CMD_GET,
2047 .flags = 0, /* OK for unprivileged users. */
2048 .policy = vport_policy,
2049 .doit = ovs_vport_cmd_get,
2050 .dumpit = ovs_vport_cmd_dump
2052 { .cmd = OVS_VPORT_CMD_SET,
2053 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2054 .policy = vport_policy,
2055 .doit = ovs_vport_cmd_set,
2059 struct genl_family dp_vport_genl_family = {
2060 .id = GENL_ID_GENERATE,
2061 .hdrsize = sizeof(struct ovs_header),
2062 .name = OVS_VPORT_FAMILY,
2063 .version = OVS_VPORT_VERSION,
2064 .maxattr = OVS_VPORT_ATTR_MAX,
2066 .parallel_ops = true,
2067 .ops = dp_vport_genl_ops,
2068 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2069 .mcgrps = &ovs_dp_vport_multicast_group,
2073 static struct genl_family *dp_genl_families[] = {
2074 &dp_datapath_genl_family,
2075 &dp_vport_genl_family,
2076 &dp_flow_genl_family,
2077 &dp_packet_genl_family,
2080 static void dp_unregister_genl(int n_families)
2084 for (i = 0; i < n_families; i++)
2085 genl_unregister_family(dp_genl_families[i]);
2088 static int dp_register_genl(void)
2093 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2095 err = genl_register_family(dp_genl_families[i]);
2103 dp_unregister_genl(i);
2107 static int __net_init ovs_init_net(struct net *net)
2109 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2111 INIT_LIST_HEAD(&ovs_net->dps);
2112 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2116 static void __net_exit ovs_exit_net(struct net *net)
2118 struct datapath *dp, *dp_next;
2119 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2122 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2126 cancel_work_sync(&ovs_net->dp_notify_work);
2129 static struct pernet_operations ovs_net_ops = {
2130 .init = ovs_init_net,
2131 .exit = ovs_exit_net,
2133 .size = sizeof(struct ovs_net),
2136 DEFINE_COMPAT_PNET_REG_FUNC(device);
2138 static int __init dp_init(void)
2142 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2144 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2147 err = action_fifos_init();
2151 err = ovs_flow_init();
2153 goto error_action_fifos_exit;
2155 err = ovs_vport_init();
2157 goto error_flow_exit;
2159 err = register_pernet_device(&ovs_net_ops);
2161 goto error_vport_exit;
2163 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2165 goto error_netns_exit;
2167 err = dp_register_genl();
2169 goto error_unreg_notifier;
2173 error_unreg_notifier:
2174 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2176 unregister_pernet_device(&ovs_net_ops);
2181 error_action_fifos_exit:
2182 action_fifos_exit();
2187 static void dp_cleanup(void)
2189 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2190 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2191 unregister_pernet_device(&ovs_net_ops);
2195 action_fifos_exit();
2198 module_init(dp_init);
2199 module_exit(dp_cleanup);
2201 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2202 MODULE_LICENSE("GPL");
2203 MODULE_VERSION(VERSION);