2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/genetlink.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
59 #include "flow_table.h"
60 #include "flow_netlink.h"
62 #include "vport-internal_dev.h"
63 #include "vport-netdev.h"
65 int ovs_net_id __read_mostly;
67 static struct genl_family dp_packet_genl_family;
68 static struct genl_family dp_flow_genl_family;
69 static struct genl_family dp_datapath_genl_family;
71 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
72 .name = OVS_FLOW_MCGROUP
75 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
76 .name = OVS_DATAPATH_MCGROUP
79 struct genl_multicast_group ovs_dp_vport_multicast_group = {
80 .name = OVS_VPORT_MCGROUP
83 /* Check if need to build a reply message.
84 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
85 static bool ovs_must_notify(struct genl_info *info,
86 const struct genl_multicast_group *grp)
88 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
89 netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
92 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
93 struct sk_buff *skb, struct genl_info *info)
95 genl_notify(family, skb, genl_info_net(info),
96 info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
102 * All writes e.g. Writes to device state (add/remove datapath, port, set
103 * operations on vports, etc.), Writes to other state (flow table
104 * modifications, set miscellaneous datapath parameters, etc.) are protected
107 * Reads are protected by RCU.
109 * There are a few special cases (mostly stats) that have their own
110 * synchronization but they nest under all of above and don't interact with
113 * The RTNL lock nests inside ovs_mutex.
116 static DEFINE_MUTEX(ovs_mutex);
120 mutex_lock(&ovs_mutex);
123 void ovs_unlock(void)
125 mutex_unlock(&ovs_mutex);
128 #ifdef CONFIG_LOCKDEP
129 int lockdep_ovsl_is_held(void)
132 return lockdep_is_held(&ovs_mutex);
138 static struct vport *new_vport(const struct vport_parms *);
139 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
140 const struct dp_upcall_info *);
141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
142 const struct dp_upcall_info *);
144 /* Must be called with rcu_read_lock or ovs_mutex. */
145 static struct datapath *get_dp(struct net *net, int dp_ifindex)
147 struct datapath *dp = NULL;
148 struct net_device *dev;
151 dev = dev_get_by_index_rcu(net, dp_ifindex);
153 struct vport *vport = ovs_internal_dev_get_vport(dev);
162 /* Must be called with rcu_read_lock or ovs_mutex. */
163 const char *ovs_dp_name(const struct datapath *dp)
165 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
166 return vport->ops->get_name(vport);
169 static int get_dpifindex(struct datapath *dp)
176 local = ovs_vport_rcu(dp, OVSP_LOCAL);
178 ifindex = netdev_vport_priv(local)->dev->ifindex;
187 static void destroy_dp_rcu(struct rcu_head *rcu)
189 struct datapath *dp = container_of(rcu, struct datapath, rcu);
191 free_percpu(dp->stats_percpu);
192 release_net(ovs_dp_get_net(dp));
197 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
200 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
203 /* Called with ovs_mutex or RCU read lock. */
204 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
207 struct hlist_head *head;
209 head = vport_hash_bucket(dp, port_no);
210 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
211 if (vport->port_no == port_no)
217 /* Called with ovs_mutex. */
218 static struct vport *new_vport(const struct vport_parms *parms)
222 vport = ovs_vport_add(parms);
223 if (!IS_ERR(vport)) {
224 struct datapath *dp = parms->dp;
225 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
227 hlist_add_head_rcu(&vport->dp_hash_node, head);
232 void ovs_dp_detach_port(struct vport *p)
236 /* First drop references to device. */
237 hlist_del_rcu(&p->dp_hash_node);
239 /* Then destroy it. */
243 /* Must be called with rcu_read_lock. */
244 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
246 struct datapath *dp = p->dp;
247 struct sw_flow *flow;
248 struct dp_stats_percpu *stats;
249 struct sw_flow_key key;
254 stats = this_cpu_ptr(dp->stats_percpu);
256 /* Extract flow from 'skb' into 'key'. */
257 error = ovs_flow_extract(skb, p->port_no, &key);
258 if (unlikely(error)) {
264 flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
265 if (unlikely(!flow)) {
266 struct dp_upcall_info upcall;
268 upcall.cmd = OVS_PACKET_CMD_MISS;
270 upcall.userdata = NULL;
271 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
272 ovs_dp_upcall(dp, skb, &upcall);
274 stats_counter = &stats->n_missed;
278 OVS_CB(skb)->flow = flow;
279 OVS_CB(skb)->pkt_key = &key;
281 ovs_flow_stats_update(OVS_CB(skb)->flow, key.tp.flags, skb);
282 ovs_execute_actions(dp, skb);
283 stats_counter = &stats->n_hit;
286 /* Update datapath statistics. */
287 u64_stats_update_begin(&stats->sync);
289 stats->n_mask_hit += n_mask_hit;
290 u64_stats_update_end(&stats->sync);
293 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
294 const struct dp_upcall_info *upcall_info)
296 struct dp_stats_percpu *stats;
299 if (upcall_info->portid == 0) {
304 if (!skb_is_gso(skb))
305 err = queue_userspace_packet(dp, skb, upcall_info);
307 err = queue_gso_packets(dp, skb, upcall_info);
314 stats = this_cpu_ptr(dp->stats_percpu);
316 u64_stats_update_begin(&stats->sync);
318 u64_stats_update_end(&stats->sync);
323 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
324 const struct dp_upcall_info *upcall_info)
326 unsigned short gso_type = skb_shinfo(skb)->gso_type;
327 struct dp_upcall_info later_info;
328 struct sw_flow_key later_key;
329 struct sk_buff *segs, *nskb;
332 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
334 return PTR_ERR(segs);
336 /* Queue all of the segments. */
339 err = queue_userspace_packet(dp, skb, upcall_info);
343 if (skb == segs && gso_type & SKB_GSO_UDP) {
344 /* The initial flow key extracted by ovs_flow_extract()
345 * in this case is for a first fragment, so we need to
346 * properly mark later fragments.
348 later_key = *upcall_info->key;
349 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
351 later_info = *upcall_info;
352 later_info.key = &later_key;
353 upcall_info = &later_info;
355 } while ((skb = skb->next));
357 /* Free all of the segments. */
365 } while ((skb = nskb));
369 static size_t key_attr_size(void)
371 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
372 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
373 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
374 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
375 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
376 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
377 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
378 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
379 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
380 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
381 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
382 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
383 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
384 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
385 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
386 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
387 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
388 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
389 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
392 static size_t upcall_msg_size(const struct nlattr *userdata,
395 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
396 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
397 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
399 /* OVS_PACKET_ATTR_USERDATA */
401 size += NLA_ALIGN(userdata->nla_len);
406 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
407 const struct dp_upcall_info *upcall_info)
409 struct ovs_header *upcall;
410 struct sk_buff *nskb = NULL;
411 struct sk_buff *user_skb; /* to be queued to userspace */
413 struct genl_info info = {
414 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
415 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
417 .snd_portid = upcall_info->portid,
423 dp_ifindex = get_dpifindex(dp);
427 if (vlan_tx_tag_present(skb)) {
428 nskb = skb_clone(skb, GFP_ATOMIC);
432 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
436 vlan_set_tci(nskb, 0);
441 if (nla_attr_size(skb->len) > USHRT_MAX) {
446 /* Complete checksum if needed */
447 if (skb->ip_summed == CHECKSUM_PARTIAL &&
448 (err = skb_checksum_help(skb)))
451 /* Older versions of OVS user space enforce alignment of the last
452 * Netlink attribute to NLA_ALIGNTO which would require extensive
453 * padding logic. Only perform zerocopy if padding is not required.
455 if (dp->user_features & OVS_DP_F_UNALIGNED)
456 hlen = skb_zerocopy_headlen(skb);
460 len = upcall_msg_size(upcall_info->userdata, hlen);
461 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
467 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
468 0, upcall_info->cmd);
469 upcall->dp_ifindex = dp_ifindex;
471 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
472 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
473 nla_nest_end(user_skb, nla);
475 if (upcall_info->userdata)
476 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
477 nla_len(upcall_info->userdata),
478 nla_data(upcall_info->userdata));
480 /* Only reserve room for attribute header, packet data is added
481 * in skb_zerocopy() */
482 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
486 nla->nla_len = nla_attr_size(skb->len);
488 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
492 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
493 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
494 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
497 memset(skb_put(user_skb, plen), 0, plen);
500 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
502 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
510 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
512 struct ovs_header *ovs_header = info->userhdr;
513 struct nlattr **a = info->attrs;
514 struct sw_flow_actions *acts;
515 struct sk_buff *packet;
516 struct sw_flow *flow;
523 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
524 !a[OVS_PACKET_ATTR_ACTIONS])
527 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
528 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
532 skb_reserve(packet, NET_IP_ALIGN);
534 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
536 skb_reset_mac_header(packet);
537 eth = eth_hdr(packet);
539 /* Normally, setting the skb 'protocol' field would be handled by a
540 * call to eth_type_trans(), but it assumes there's a sending
541 * device, which we may not have. */
542 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
543 packet->protocol = eth->h_proto;
545 packet->protocol = htons(ETH_P_802_2);
547 /* Build an sw_flow for sending this packet. */
548 flow = ovs_flow_alloc();
553 err = ovs_flow_extract(packet, -1, &flow->key);
557 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
560 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
565 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
566 &flow->key, 0, &acts);
567 rcu_assign_pointer(flow->sf_acts, acts);
571 OVS_CB(packet)->flow = flow;
572 OVS_CB(packet)->pkt_key = &flow->key;
573 packet->priority = flow->key.phy.priority;
574 packet->mark = flow->key.phy.skb_mark;
577 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
583 err = ovs_execute_actions(dp, packet);
587 ovs_flow_free(flow, false);
593 ovs_flow_free(flow, false);
600 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
601 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
602 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
603 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
606 static struct genl_ops dp_packet_genl_ops[] = {
607 { .cmd = OVS_PACKET_CMD_EXECUTE,
608 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
609 .policy = packet_policy,
610 .doit = ovs_packet_cmd_execute
614 static struct genl_family dp_packet_genl_family = {
615 .id = GENL_ID_GENERATE,
616 .hdrsize = sizeof(struct ovs_header),
617 .name = OVS_PACKET_FAMILY,
618 .version = OVS_PACKET_VERSION,
619 .maxattr = OVS_PACKET_ATTR_MAX,
621 .parallel_ops = true,
622 .ops = dp_packet_genl_ops,
623 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
626 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
627 struct ovs_dp_megaflow_stats *mega_stats)
631 memset(mega_stats, 0, sizeof(*mega_stats));
633 stats->n_flows = ovs_flow_tbl_count(&dp->table);
634 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
636 stats->n_hit = stats->n_missed = stats->n_lost = 0;
638 for_each_possible_cpu(i) {
639 const struct dp_stats_percpu *percpu_stats;
640 struct dp_stats_percpu local_stats;
643 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
646 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
647 local_stats = *percpu_stats;
648 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
650 stats->n_hit += local_stats.n_hit;
651 stats->n_missed += local_stats.n_missed;
652 stats->n_lost += local_stats.n_lost;
653 mega_stats->n_mask_hit += local_stats.n_mask_hit;
657 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
659 return NLMSG_ALIGN(sizeof(struct ovs_header))
660 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
661 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
662 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
663 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
664 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
665 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
668 /* Called with ovs_mutex or RCU read lock. */
669 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
670 struct sk_buff *skb, u32 portid,
671 u32 seq, u32 flags, u8 cmd)
673 const int skb_orig_len = skb->len;
674 struct nlattr *start;
675 struct ovs_flow_stats stats;
678 struct ovs_header *ovs_header;
682 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
686 ovs_header->dp_ifindex = dp_ifindex;
689 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
691 goto nla_put_failure;
693 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
696 nla_nest_end(skb, nla);
698 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
700 goto nla_put_failure;
702 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
706 nla_nest_end(skb, nla);
708 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
711 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
712 goto nla_put_failure;
714 if (stats.n_packets &&
715 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
716 goto nla_put_failure;
718 if ((u8)ntohs(tcp_flags) &&
719 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
720 goto nla_put_failure;
722 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
723 * this is the first flow to be dumped into 'skb'. This is unusual for
724 * Netlink but individual action lists can be longer than
725 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
726 * The userspace caller can always fetch the actions separately if it
727 * really wants them. (Most userspace callers in fact don't care.)
729 * This can only fail for dump operations because the skb is always
730 * properly sized for single flows.
732 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
734 const struct sw_flow_actions *sf_acts;
736 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
737 err = ovs_nla_put_actions(sf_acts->actions,
738 sf_acts->actions_len, skb);
741 nla_nest_end(skb, start);
746 nla_nest_cancel(skb, start);
748 } else if (skb_orig_len)
749 goto nla_put_failure;
751 return genlmsg_end(skb, ovs_header);
756 genlmsg_cancel(skb, ovs_header);
760 /* May not be called with RCU read lock. */
761 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
762 struct genl_info *info,
767 if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
770 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
773 return ERR_PTR(-ENOMEM);
778 /* Called with ovs_mutex. */
779 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
781 struct genl_info *info, u8 cmd,
787 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
789 if (!skb || IS_ERR(skb))
792 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
793 info->snd_portid, info->snd_seq, 0,
799 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
801 struct nlattr **a = info->attrs;
802 struct ovs_header *ovs_header = info->userhdr;
803 struct sw_flow *flow, *new_flow;
804 struct sw_flow_mask mask;
805 struct sk_buff *reply;
807 struct sw_flow_actions *acts;
808 struct sw_flow_match match;
811 /* Must have key and actions. */
813 if (!a[OVS_FLOW_ATTR_KEY])
815 if (!a[OVS_FLOW_ATTR_ACTIONS])
818 /* Most of the time we need to allocate a new flow, do it before
820 new_flow = ovs_flow_alloc();
821 if (IS_ERR(new_flow)) {
822 error = PTR_ERR(new_flow);
827 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
828 error = ovs_nla_get_match(&match,
829 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
833 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
835 /* Validate actions. */
836 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
837 error = PTR_ERR(acts);
841 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
844 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
848 reply = ovs_flow_cmd_alloc_info(acts, info, false);
850 error = PTR_ERR(reply);
855 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
860 /* Check if this is a duplicate flow */
861 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
863 rcu_assign_pointer(new_flow->sf_acts, acts);
865 /* Put flow in bucket. */
866 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
867 if (unlikely(error)) {
872 if (unlikely(reply)) {
873 error = ovs_flow_cmd_fill_info(new_flow,
874 ovs_header->dp_ifindex,
875 reply, info->snd_portid,
882 struct sw_flow_actions *old_acts;
884 /* Bail out if we're not allowed to modify an existing flow.
885 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
886 * because Generic Netlink treats the latter as a dump
887 * request. We also accept NLM_F_EXCL in case that bug ever
890 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
895 /* The unmasked key has to be the same for flow updates. */
896 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
900 /* Update actions. */
901 old_acts = ovsl_dereference(flow->sf_acts);
902 rcu_assign_pointer(flow->sf_acts, acts);
904 if (unlikely(reply)) {
905 error = ovs_flow_cmd_fill_info(flow,
906 ovs_header->dp_ifindex,
907 reply, info->snd_portid,
914 ovs_nla_free_flow_actions(old_acts);
915 ovs_flow_free(new_flow, false);
919 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
928 ovs_flow_free(new_flow, false);
933 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
935 struct nlattr **a = info->attrs;
936 struct ovs_header *ovs_header = info->userhdr;
937 struct sw_flow_key key, masked_key;
938 struct sw_flow *flow;
939 struct sw_flow_mask mask;
940 struct sk_buff *reply = NULL;
942 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
943 struct sw_flow_match match;
948 if (!a[OVS_FLOW_ATTR_KEY])
951 ovs_match_init(&match, &key, &mask);
952 error = ovs_nla_get_match(&match,
953 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
957 /* Validate actions. */
958 if (a[OVS_FLOW_ATTR_ACTIONS]) {
959 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
960 error = PTR_ERR(acts);
964 ovs_flow_mask_key(&masked_key, &key, &mask);
965 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
966 &masked_key, 0, &acts);
968 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
973 /* Can allocate before locking if have acts. */
975 reply = ovs_flow_cmd_alloc_info(acts, info, false);
977 error = PTR_ERR(reply);
983 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
988 /* Check that the flow exists. */
989 flow = ovs_flow_tbl_lookup(&dp->table, &key);
990 if (unlikely(!flow)) {
994 /* The unmasked key has to be the same for flow updates. */
995 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
999 /* Update actions, if present. */
1001 old_acts = ovsl_dereference(flow->sf_acts);
1002 rcu_assign_pointer(flow->sf_acts, acts);
1004 if (unlikely(reply)) {
1005 error = ovs_flow_cmd_fill_info(flow,
1006 ovs_header->dp_ifindex,
1007 reply, info->snd_portid,
1013 /* Could not alloc without acts before locking. */
1014 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1015 info, OVS_FLOW_CMD_NEW, false);
1016 if (unlikely(IS_ERR(reply))) {
1017 error = PTR_ERR(reply);
1018 goto err_unlock_ovs;
1023 if (a[OVS_FLOW_ATTR_CLEAR])
1024 ovs_flow_stats_clear(flow);
1028 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1030 ovs_nla_free_flow_actions(old_acts);
1042 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1044 struct nlattr **a = info->attrs;
1045 struct ovs_header *ovs_header = info->userhdr;
1046 struct sw_flow_key key;
1047 struct sk_buff *reply;
1048 struct sw_flow *flow;
1049 struct datapath *dp;
1050 struct sw_flow_match match;
1053 if (!a[OVS_FLOW_ATTR_KEY]) {
1054 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1058 ovs_match_init(&match, &key, NULL);
1059 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1064 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1070 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1071 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1076 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1077 OVS_FLOW_CMD_NEW, true);
1078 if (IS_ERR(reply)) {
1079 err = PTR_ERR(reply);
1084 return genlmsg_reply(reply, info);
1090 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1092 struct nlattr **a = info->attrs;
1093 struct ovs_header *ovs_header = info->userhdr;
1094 struct sw_flow_key key;
1095 struct sk_buff *reply;
1096 struct sw_flow *flow;
1097 struct datapath *dp;
1098 struct sw_flow_match match;
1101 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1102 ovs_match_init(&match, &key, NULL);
1103 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1109 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1110 if (unlikely(!dp)) {
1114 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1115 err = ovs_flow_tbl_flush(&dp->table);
1118 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1119 if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1124 ovs_flow_tbl_remove(&dp->table, flow);
1127 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1130 if (likely(reply)) {
1131 if (likely(!IS_ERR(reply))) {
1132 rcu_read_lock(); /* Keep RCU checker happy. */
1133 err = ovs_flow_cmd_fill_info(flow,
1134 ovs_header->dp_ifindex,
1135 reply, info->snd_portid,
1140 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1142 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1143 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1148 ovs_flow_free(flow, true);
1155 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1157 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1158 struct table_instance *ti;
1159 struct datapath *dp;
1162 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1168 ti = rcu_dereference(dp->table.ti);
1170 struct sw_flow *flow;
1173 bucket = cb->args[0];
1175 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1179 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1180 NETLINK_CB(cb->skb).portid,
1181 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1182 OVS_FLOW_CMD_NEW) < 0)
1185 cb->args[0] = bucket;
1192 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1193 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1194 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1195 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1198 static struct genl_ops dp_flow_genl_ops[] = {
1199 { .cmd = OVS_FLOW_CMD_NEW,
1200 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1201 .policy = flow_policy,
1202 .doit = ovs_flow_cmd_new
1204 { .cmd = OVS_FLOW_CMD_DEL,
1205 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1206 .policy = flow_policy,
1207 .doit = ovs_flow_cmd_del
1209 { .cmd = OVS_FLOW_CMD_GET,
1210 .flags = 0, /* OK for unprivileged users. */
1211 .policy = flow_policy,
1212 .doit = ovs_flow_cmd_get,
1213 .dumpit = ovs_flow_cmd_dump
1215 { .cmd = OVS_FLOW_CMD_SET,
1216 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1217 .policy = flow_policy,
1218 .doit = ovs_flow_cmd_set,
1222 static struct genl_family dp_flow_genl_family = {
1223 .id = GENL_ID_GENERATE,
1224 .hdrsize = sizeof(struct ovs_header),
1225 .name = OVS_FLOW_FAMILY,
1226 .version = OVS_FLOW_VERSION,
1227 .maxattr = OVS_FLOW_ATTR_MAX,
1229 .parallel_ops = true,
1230 .ops = dp_flow_genl_ops,
1231 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1232 .mcgrps = &ovs_dp_flow_multicast_group,
1236 static size_t ovs_dp_cmd_msg_size(void)
1238 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1240 msgsize += nla_total_size(IFNAMSIZ);
1241 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1242 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1243 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1248 /* Called with ovs_mutex or RCU read lock. */
1249 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1250 u32 portid, u32 seq, u32 flags, u8 cmd)
1252 struct ovs_header *ovs_header;
1253 struct ovs_dp_stats dp_stats;
1254 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1257 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1262 ovs_header->dp_ifindex = get_dpifindex(dp);
1264 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1266 goto nla_put_failure;
1268 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1269 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1271 goto nla_put_failure;
1273 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1274 sizeof(struct ovs_dp_megaflow_stats),
1275 &dp_megaflow_stats))
1276 goto nla_put_failure;
1278 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1279 goto nla_put_failure;
1281 return genlmsg_end(skb, ovs_header);
1284 genlmsg_cancel(skb, ovs_header);
1289 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1291 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1294 /* Called with rcu_read_lock or ovs_mutex. */
1295 static struct datapath *lookup_datapath(struct net *net,
1296 struct ovs_header *ovs_header,
1297 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1299 struct datapath *dp;
1301 if (!a[OVS_DP_ATTR_NAME])
1302 dp = get_dp(net, ovs_header->dp_ifindex);
1304 struct vport *vport;
1306 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1307 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1309 return dp ? dp : ERR_PTR(-ENODEV);
1312 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1314 struct datapath *dp;
1316 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1320 WARN(dp->user_features, "Dropping previously announced user features\n");
1321 dp->user_features = 0;
1324 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1326 if (a[OVS_DP_ATTR_USER_FEATURES])
1327 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1330 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1332 struct nlattr **a = info->attrs;
1333 struct vport_parms parms;
1334 struct sk_buff *reply;
1335 struct datapath *dp;
1336 struct vport *vport;
1337 struct ovs_net *ovs_net;
1341 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1344 reply = ovs_dp_cmd_alloc_info(info);
1349 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1351 goto err_free_reply;
1353 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1355 /* Allocate table. */
1356 err = ovs_flow_tbl_init(&dp->table);
1360 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1361 if (!dp->stats_percpu) {
1363 goto err_destroy_table;
1366 for_each_possible_cpu(i) {
1367 struct dp_stats_percpu *dpath_stats;
1368 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1369 u64_stats_init(&dpath_stats->sync);
1372 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1376 goto err_destroy_percpu;
1379 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1380 INIT_HLIST_HEAD(&dp->ports[i]);
1382 /* Set up our datapath device. */
1383 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1384 parms.type = OVS_VPORT_TYPE_INTERNAL;
1385 parms.options = NULL;
1387 parms.port_no = OVSP_LOCAL;
1388 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1390 ovs_dp_change(dp, a);
1392 /* So far only local changes have been made, now need the lock. */
1395 vport = new_vport(&parms);
1396 if (IS_ERR(vport)) {
1397 err = PTR_ERR(vport);
1401 if (err == -EEXIST) {
1402 /* An outdated user space instance that does not understand
1403 * the concept of user_features has attempted to create a new
1404 * datapath and is likely to reuse it. Drop all user features.
1406 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1407 ovs_dp_reset_user_features(skb, info);
1410 goto err_destroy_ports_array;
1413 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1414 info->snd_seq, 0, OVS_DP_CMD_NEW);
1417 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1418 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1422 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1425 err_destroy_ports_array:
1429 free_percpu(dp->stats_percpu);
1431 ovs_flow_tbl_destroy(&dp->table, false);
1433 release_net(ovs_dp_get_net(dp));
1441 /* Called with ovs_mutex. */
1442 static void __dp_destroy(struct datapath *dp)
1446 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1447 struct vport *vport;
1448 struct hlist_node *n;
1450 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1451 if (vport->port_no != OVSP_LOCAL)
1452 ovs_dp_detach_port(vport);
1455 list_del_rcu(&dp->list_node);
1457 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1458 * all ports in datapath are destroyed first before freeing datapath.
1460 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1462 /* RCU destroy the flow table */
1463 ovs_flow_tbl_destroy(&dp->table, true);
1465 call_rcu(&dp->rcu, destroy_dp_rcu);
1468 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1470 struct sk_buff *reply;
1471 struct datapath *dp;
1474 reply = ovs_dp_cmd_alloc_info(info);
1479 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1482 goto err_unlock_free;
1484 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1485 info->snd_seq, 0, OVS_DP_CMD_DEL);
1491 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1500 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1502 struct sk_buff *reply;
1503 struct datapath *dp;
1506 reply = ovs_dp_cmd_alloc_info(info);
1511 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1514 goto err_unlock_free;
1516 ovs_dp_change(dp, info->attrs);
1518 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1519 info->snd_seq, 0, OVS_DP_CMD_NEW);
1523 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1532 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1534 struct sk_buff *reply;
1535 struct datapath *dp;
1538 reply = ovs_dp_cmd_alloc_info(info);
1543 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1546 goto err_unlock_free;
1548 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1549 info->snd_seq, 0, OVS_DP_CMD_NEW);
1553 return genlmsg_reply(reply, info);
1561 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1563 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1564 struct datapath *dp;
1565 int skip = cb->args[0];
1569 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1571 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1572 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1573 OVS_DP_CMD_NEW) < 0)
1584 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1585 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1586 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1587 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1590 static struct genl_ops dp_datapath_genl_ops[] = {
1591 { .cmd = OVS_DP_CMD_NEW,
1592 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1593 .policy = datapath_policy,
1594 .doit = ovs_dp_cmd_new
1596 { .cmd = OVS_DP_CMD_DEL,
1597 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1598 .policy = datapath_policy,
1599 .doit = ovs_dp_cmd_del
1601 { .cmd = OVS_DP_CMD_GET,
1602 .flags = 0, /* OK for unprivileged users. */
1603 .policy = datapath_policy,
1604 .doit = ovs_dp_cmd_get,
1605 .dumpit = ovs_dp_cmd_dump
1607 { .cmd = OVS_DP_CMD_SET,
1608 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1609 .policy = datapath_policy,
1610 .doit = ovs_dp_cmd_set,
1614 static struct genl_family dp_datapath_genl_family = {
1615 .id = GENL_ID_GENERATE,
1616 .hdrsize = sizeof(struct ovs_header),
1617 .name = OVS_DATAPATH_FAMILY,
1618 .version = OVS_DATAPATH_VERSION,
1619 .maxattr = OVS_DP_ATTR_MAX,
1621 .parallel_ops = true,
1622 .ops = dp_datapath_genl_ops,
1623 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1624 .mcgrps = &ovs_dp_datapath_multicast_group,
1628 /* Called with ovs_mutex or RCU read lock. */
1629 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1630 u32 portid, u32 seq, u32 flags, u8 cmd)
1632 struct ovs_header *ovs_header;
1633 struct ovs_vport_stats vport_stats;
1636 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1641 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1643 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1644 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1645 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1646 goto nla_put_failure;
1648 ovs_vport_get_stats(vport, &vport_stats);
1649 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1651 goto nla_put_failure;
1653 if (ovs_vport_get_upcall_portids(vport, skb))
1654 goto nla_put_failure;
1656 err = ovs_vport_get_options(vport, skb);
1657 if (err == -EMSGSIZE)
1660 return genlmsg_end(skb, ovs_header);
1665 genlmsg_cancel(skb, ovs_header);
1669 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1671 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1674 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1675 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1678 struct sk_buff *skb;
1681 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1683 return ERR_PTR(-ENOMEM);
1685 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1691 /* Called with ovs_mutex or RCU read lock. */
1692 static struct vport *lookup_vport(struct net *net,
1693 struct ovs_header *ovs_header,
1694 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1696 struct datapath *dp;
1697 struct vport *vport;
1699 if (a[OVS_VPORT_ATTR_NAME]) {
1700 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1702 return ERR_PTR(-ENODEV);
1703 if (ovs_header->dp_ifindex &&
1704 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1705 return ERR_PTR(-ENODEV);
1707 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1708 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1710 if (port_no >= DP_MAX_PORTS)
1711 return ERR_PTR(-EFBIG);
1713 dp = get_dp(net, ovs_header->dp_ifindex);
1715 return ERR_PTR(-ENODEV);
1717 vport = ovs_vport_ovsl_rcu(dp, port_no);
1719 return ERR_PTR(-ENODEV);
1722 return ERR_PTR(-EINVAL);
1725 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1727 struct nlattr **a = info->attrs;
1728 struct ovs_header *ovs_header = info->userhdr;
1729 struct vport_parms parms;
1730 struct sk_buff *reply;
1731 struct vport *vport;
1732 struct datapath *dp;
1736 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1737 !a[OVS_VPORT_ATTR_UPCALL_PID])
1740 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1741 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1742 if (port_no >= DP_MAX_PORTS)
1745 reply = ovs_vport_cmd_alloc_info();
1750 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1753 goto exit_unlock_free;
1756 vport = ovs_vport_ovsl(dp, port_no);
1759 goto exit_unlock_free;
1761 for (port_no = 1; ; port_no++) {
1762 if (port_no >= DP_MAX_PORTS) {
1764 goto exit_unlock_free;
1766 vport = ovs_vport_ovsl(dp, port_no);
1772 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1773 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1774 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1776 parms.port_no = port_no;
1777 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1779 vport = new_vport(&parms);
1780 err = PTR_ERR(vport);
1782 goto exit_unlock_free;
1785 if (a[OVS_VPORT_ATTR_STATS])
1786 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1788 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1789 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1793 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1802 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1804 struct nlattr **a = info->attrs;
1805 struct sk_buff *reply;
1806 struct vport *vport;
1809 reply = ovs_vport_cmd_alloc_info();
1814 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1815 err = PTR_ERR(vport);
1817 goto exit_unlock_free;
1819 if (a[OVS_VPORT_ATTR_TYPE] &&
1820 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1822 goto exit_unlock_free;
1825 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1826 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1828 goto exit_unlock_free;
1831 if (a[OVS_VPORT_ATTR_STATS])
1832 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1835 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1836 err = ovs_vport_set_upcall_portids(vport,
1837 a[OVS_VPORT_ATTR_UPCALL_PID]);
1839 goto exit_unlock_free;
1842 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1843 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1847 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1856 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1858 struct nlattr **a = info->attrs;
1859 struct sk_buff *reply;
1860 struct vport *vport;
1863 reply = ovs_vport_cmd_alloc_info();
1868 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1869 err = PTR_ERR(vport);
1871 goto exit_unlock_free;
1873 if (vport->port_no == OVSP_LOCAL) {
1875 goto exit_unlock_free;
1878 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1879 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1881 ovs_dp_detach_port(vport);
1884 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1893 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1895 struct nlattr **a = info->attrs;
1896 struct ovs_header *ovs_header = info->userhdr;
1897 struct sk_buff *reply;
1898 struct vport *vport;
1901 reply = ovs_vport_cmd_alloc_info();
1906 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1907 err = PTR_ERR(vport);
1909 goto exit_unlock_free;
1910 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1911 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1915 return genlmsg_reply(reply, info);
1923 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1925 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1926 struct datapath *dp;
1927 int bucket = cb->args[0], skip = cb->args[1];
1931 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1936 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1937 struct vport *vport;
1940 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1942 ovs_vport_cmd_fill_info(vport, skb,
1943 NETLINK_CB(cb->skb).portid,
1946 OVS_VPORT_CMD_NEW) < 0)
1962 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1963 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1964 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1965 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1966 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1967 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1968 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1971 static struct genl_ops dp_vport_genl_ops[] = {
1972 { .cmd = OVS_VPORT_CMD_NEW,
1973 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1974 .policy = vport_policy,
1975 .doit = ovs_vport_cmd_new
1977 { .cmd = OVS_VPORT_CMD_DEL,
1978 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1979 .policy = vport_policy,
1980 .doit = ovs_vport_cmd_del
1982 { .cmd = OVS_VPORT_CMD_GET,
1983 .flags = 0, /* OK for unprivileged users. */
1984 .policy = vport_policy,
1985 .doit = ovs_vport_cmd_get,
1986 .dumpit = ovs_vport_cmd_dump
1988 { .cmd = OVS_VPORT_CMD_SET,
1989 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1990 .policy = vport_policy,
1991 .doit = ovs_vport_cmd_set,
1995 struct genl_family dp_vport_genl_family = {
1996 .id = GENL_ID_GENERATE,
1997 .hdrsize = sizeof(struct ovs_header),
1998 .name = OVS_VPORT_FAMILY,
1999 .version = OVS_VPORT_VERSION,
2000 .maxattr = OVS_VPORT_ATTR_MAX,
2002 .parallel_ops = true,
2003 .ops = dp_vport_genl_ops,
2004 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2005 .mcgrps = &ovs_dp_vport_multicast_group,
2009 static struct genl_family *dp_genl_families[] = {
2010 &dp_datapath_genl_family,
2011 &dp_vport_genl_family,
2012 &dp_flow_genl_family,
2013 &dp_packet_genl_family,
2016 static void dp_unregister_genl(int n_families)
2020 for (i = 0; i < n_families; i++)
2021 genl_unregister_family(dp_genl_families[i]);
2024 static int dp_register_genl(void)
2029 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2031 err = genl_register_family(dp_genl_families[i]);
2039 dp_unregister_genl(i);
2043 static int __net_init ovs_init_net(struct net *net)
2045 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2047 INIT_LIST_HEAD(&ovs_net->dps);
2048 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2052 static void __net_exit ovs_exit_net(struct net *net)
2054 struct datapath *dp, *dp_next;
2055 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2058 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2062 cancel_work_sync(&ovs_net->dp_notify_work);
2065 static struct pernet_operations ovs_net_ops = {
2066 .init = ovs_init_net,
2067 .exit = ovs_exit_net,
2069 .size = sizeof(struct ovs_net),
2072 DEFINE_COMPAT_PNET_REG_FUNC(device);
2074 static int __init dp_init(void)
2078 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2080 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2083 err = ovs_flow_init();
2087 err = ovs_vport_init();
2089 goto error_flow_exit;
2091 err = register_pernet_device(&ovs_net_ops);
2093 goto error_vport_exit;
2095 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2097 goto error_netns_exit;
2099 err = dp_register_genl();
2101 goto error_unreg_notifier;
2105 error_unreg_notifier:
2106 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2108 unregister_pernet_device(&ovs_net_ops);
2117 static void dp_cleanup(void)
2119 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2120 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2121 unregister_pernet_device(&ovs_net_ops);
2127 module_init(dp_init);
2128 module_exit(dp_cleanup);
2130 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2131 MODULE_LICENSE("GPL");
2132 MODULE_VERSION(VERSION);