2 * Copyright (c) 2007-2013 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/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/lockdep.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/workqueue.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
58 #include "flow_netlink.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
62 int ovs_net_id __read_mostly;
64 static void ovs_notify(struct sk_buff *skb, struct genl_info *info,
65 struct genl_multicast_group *grp)
67 genl_notify(skb, genl_info_net(info), info->snd_portid,
68 grp->id, info->nlhdr, GFP_KERNEL);
74 * All writes e.g. Writes to device state (add/remove datapath, port, set
75 * operations on vports, etc.), Writes to other state (flow table
76 * modifications, set miscellaneous datapath parameters, etc.) are protected
79 * Reads are protected by RCU.
81 * There are a few special cases (mostly stats) that have their own
82 * synchronization but they nest under all of above and don't interact with
85 * The RTNL lock nests inside ovs_mutex.
88 static DEFINE_MUTEX(ovs_mutex);
92 mutex_lock(&ovs_mutex);
97 mutex_unlock(&ovs_mutex);
100 #ifdef CONFIG_LOCKDEP
101 int lockdep_ovsl_is_held(void)
104 return lockdep_is_held(&ovs_mutex);
110 static struct vport *new_vport(const struct vport_parms *);
111 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
112 const struct dp_upcall_info *);
113 static int queue_userspace_packet(struct net *, int dp_ifindex,
115 const struct dp_upcall_info *);
117 /* Must be called with rcu_read_lock or ovs_mutex. */
118 static struct datapath *get_dp(struct net *net, int dp_ifindex)
120 struct datapath *dp = NULL;
121 struct net_device *dev;
124 dev = dev_get_by_index_rcu(net, dp_ifindex);
126 struct vport *vport = ovs_internal_dev_get_vport(dev);
135 /* Must be called with rcu_read_lock or ovs_mutex. */
136 const char *ovs_dp_name(const struct datapath *dp)
138 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
139 return vport->ops->get_name(vport);
142 static int get_dpifindex(struct datapath *dp)
149 local = ovs_vport_rcu(dp, OVSP_LOCAL);
151 ifindex = netdev_vport_priv(local)->dev->ifindex;
160 static void destroy_dp_rcu(struct rcu_head *rcu)
162 struct datapath *dp = container_of(rcu, struct datapath, rcu);
164 ovs_flow_tbl_destroy(&dp->table);
165 free_percpu(dp->stats_percpu);
166 release_net(ovs_dp_get_net(dp));
171 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
174 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
177 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
180 struct hlist_head *head;
182 head = vport_hash_bucket(dp, port_no);
183 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
184 if (vport->port_no == port_no)
190 /* Called with ovs_mutex. */
191 static struct vport *new_vport(const struct vport_parms *parms)
195 vport = ovs_vport_add(parms);
196 if (!IS_ERR(vport)) {
197 struct datapath *dp = parms->dp;
198 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
200 hlist_add_head_rcu(&vport->dp_hash_node, head);
205 void ovs_dp_detach_port(struct vport *p)
209 /* First drop references to device. */
210 hlist_del_rcu(&p->dp_hash_node);
212 /* Then destroy it. */
216 /* Must be called with rcu_read_lock. */
217 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
219 struct datapath *dp = p->dp;
220 struct sw_flow *flow;
221 struct dp_stats_percpu *stats;
222 struct sw_flow_key key;
227 stats = this_cpu_ptr(dp->stats_percpu);
229 /* Extract flow from 'skb' into 'key'. */
230 error = ovs_flow_extract(skb, p->port_no, &key);
231 if (unlikely(error)) {
237 flow = ovs_flow_tbl_lookup(&dp->table, &key, &n_mask_hit);
238 if (unlikely(!flow)) {
239 struct dp_upcall_info upcall;
241 upcall.cmd = OVS_PACKET_CMD_MISS;
243 upcall.userdata = NULL;
244 upcall.portid = p->upcall_portid;
245 ovs_dp_upcall(dp, skb, &upcall);
247 stats_counter = &stats->n_missed;
251 OVS_CB(skb)->flow = flow;
252 OVS_CB(skb)->pkt_key = &key;
254 stats_counter = &stats->n_hit;
255 ovs_flow_used(OVS_CB(skb)->flow, skb);
256 ovs_execute_actions(dp, skb);
259 /* Update datapath statistics. */
260 u64_stats_update_begin(&stats->sync);
262 stats->n_mask_hit += n_mask_hit;
263 u64_stats_update_end(&stats->sync);
266 static struct genl_family dp_packet_genl_family = {
267 .id = GENL_ID_GENERATE,
268 .hdrsize = sizeof(struct ovs_header),
269 .name = OVS_PACKET_FAMILY,
270 .version = OVS_PACKET_VERSION,
271 .maxattr = OVS_PACKET_ATTR_MAX,
273 .parallel_ops = true,
276 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
277 const struct dp_upcall_info *upcall_info)
279 struct dp_stats_percpu *stats;
283 if (upcall_info->portid == 0) {
288 dp_ifindex = get_dpifindex(dp);
294 if (!skb_is_gso(skb))
295 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
297 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
304 stats = this_cpu_ptr(dp->stats_percpu);
306 u64_stats_update_begin(&stats->sync);
308 u64_stats_update_end(&stats->sync);
313 static int queue_gso_packets(struct net *net, int dp_ifindex,
315 const struct dp_upcall_info *upcall_info)
317 unsigned short gso_type = skb_shinfo(skb)->gso_type;
318 struct dp_upcall_info later_info;
319 struct sw_flow_key later_key;
320 struct sk_buff *segs, *nskb;
323 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
325 return PTR_ERR(segs);
327 /* Queue all of the segments. */
330 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
334 if (skb == segs && gso_type & SKB_GSO_UDP) {
335 /* The initial flow key extracted by ovs_flow_extract()
336 * in this case is for a first fragment, so we need to
337 * properly mark later fragments.
339 later_key = *upcall_info->key;
340 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
342 later_info = *upcall_info;
343 later_info.key = &later_key;
344 upcall_info = &later_info;
346 } while ((skb = skb->next));
348 /* Free all of the segments. */
356 } while ((skb = nskb));
360 static size_t key_attr_size(void)
362 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
363 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
364 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
365 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
366 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
367 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
368 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
369 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
370 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
371 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
372 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
373 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
374 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
375 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
376 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
377 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
378 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
379 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
380 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
383 static size_t upcall_msg_size(const struct sk_buff *skb,
384 const struct nlattr *userdata)
386 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
387 + nla_total_size(skb->len) /* OVS_PACKET_ATTR_PACKET */
388 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
390 /* OVS_PACKET_ATTR_USERDATA */
392 size += NLA_ALIGN(userdata->nla_len);
397 static int queue_userspace_packet(struct net *net, int dp_ifindex,
399 const struct dp_upcall_info *upcall_info)
401 struct ovs_header *upcall;
402 struct sk_buff *nskb = NULL;
403 struct sk_buff *user_skb; /* to be queued to userspace */
407 if (vlan_tx_tag_present(skb)) {
408 nskb = skb_clone(skb, GFP_ATOMIC);
412 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
420 if (nla_attr_size(skb->len) > USHRT_MAX) {
425 user_skb = genlmsg_new(upcall_msg_size(skb, upcall_info->userdata), GFP_ATOMIC);
431 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
432 0, upcall_info->cmd);
433 upcall->dp_ifindex = dp_ifindex;
435 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
436 ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
437 nla_nest_end(user_skb, nla);
439 if (upcall_info->userdata)
440 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
441 nla_len(upcall_info->userdata),
442 nla_data(upcall_info->userdata));
444 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
446 skb_copy_and_csum_dev(skb, nla_data(nla));
448 genlmsg_end(user_skb, upcall);
449 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
456 static void clear_stats(struct sw_flow *flow)
460 flow->packet_count = 0;
461 flow->byte_count = 0;
464 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
466 struct ovs_header *ovs_header = info->userhdr;
467 struct nlattr **a = info->attrs;
468 struct sw_flow_actions *acts;
469 struct sk_buff *packet;
470 struct sw_flow *flow;
477 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
478 !a[OVS_PACKET_ATTR_ACTIONS])
481 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
482 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
486 skb_reserve(packet, NET_IP_ALIGN);
488 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
490 skb_reset_mac_header(packet);
491 eth = eth_hdr(packet);
493 /* Normally, setting the skb 'protocol' field would be handled by a
494 * call to eth_type_trans(), but it assumes there's a sending
495 * device, which we may not have. */
496 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
497 packet->protocol = eth->h_proto;
499 packet->protocol = htons(ETH_P_802_2);
501 /* Build an sw_flow for sending this packet. */
502 flow = ovs_flow_alloc();
507 err = ovs_flow_extract(packet, -1, &flow->key);
511 err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
514 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
519 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
520 &flow->key, 0, &acts);
521 rcu_assign_pointer(flow->sf_acts, acts);
525 OVS_CB(packet)->flow = flow;
526 OVS_CB(packet)->pkt_key = &flow->key;
527 packet->priority = flow->key.phy.priority;
528 packet->mark = flow->key.phy.skb_mark;
531 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
537 err = ovs_execute_actions(dp, packet);
541 ovs_flow_free(flow, false);
547 ovs_flow_free(flow, false);
554 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
555 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
556 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
557 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
560 static struct genl_ops dp_packet_genl_ops[] = {
561 { .cmd = OVS_PACKET_CMD_EXECUTE,
562 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
563 .policy = packet_policy,
564 .doit = ovs_packet_cmd_execute
568 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
569 struct ovs_dp_megaflow_stats *mega_stats)
573 memset(mega_stats, 0, sizeof(*mega_stats));
575 stats->n_flows = ovs_flow_tbl_count(&dp->table);
576 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
578 stats->n_hit = stats->n_missed = stats->n_lost = 0;
580 for_each_possible_cpu(i) {
581 const struct dp_stats_percpu *percpu_stats;
582 struct dp_stats_percpu local_stats;
585 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
588 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
589 local_stats = *percpu_stats;
590 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
592 stats->n_hit += local_stats.n_hit;
593 stats->n_missed += local_stats.n_missed;
594 stats->n_lost += local_stats.n_lost;
595 mega_stats->n_mask_hit += local_stats.n_mask_hit;
599 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
600 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
601 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
602 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
605 static struct genl_family dp_flow_genl_family = {
606 .id = GENL_ID_GENERATE,
607 .hdrsize = sizeof(struct ovs_header),
608 .name = OVS_FLOW_FAMILY,
609 .version = OVS_FLOW_VERSION,
610 .maxattr = OVS_FLOW_ATTR_MAX,
612 .parallel_ops = true,
615 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
616 .name = OVS_FLOW_MCGROUP
619 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
621 return NLMSG_ALIGN(sizeof(struct ovs_header))
622 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
623 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
624 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
625 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
626 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
627 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
630 /* Called with ovs_mutex. */
631 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
632 struct sk_buff *skb, u32 portid,
633 u32 seq, u32 flags, u8 cmd)
635 const int skb_orig_len = skb->len;
636 struct nlattr *start;
637 struct ovs_flow_stats stats;
638 struct ovs_header *ovs_header;
644 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
648 ovs_header->dp_ifindex = get_dpifindex(dp);
651 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
653 goto nla_put_failure;
655 err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
658 nla_nest_end(skb, nla);
660 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
662 goto nla_put_failure;
664 err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
668 nla_nest_end(skb, nla);
670 spin_lock_bh(&flow->lock);
672 stats.n_packets = flow->packet_count;
673 stats.n_bytes = flow->byte_count;
674 tcp_flags = (u8)ntohs(flow->tcp_flags);
675 spin_unlock_bh(&flow->lock);
678 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
679 goto nla_put_failure;
681 if (stats.n_packets &&
682 nla_put(skb, OVS_FLOW_ATTR_STATS,
683 sizeof(struct ovs_flow_stats), &stats))
684 goto nla_put_failure;
687 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
688 goto nla_put_failure;
690 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
691 * this is the first flow to be dumped into 'skb'. This is unusual for
692 * Netlink but individual action lists can be longer than
693 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
694 * The userspace caller can always fetch the actions separately if it
695 * really wants them. (Most userspace callers in fact don't care.)
697 * This can only fail for dump operations because the skb is always
698 * properly sized for single flows.
700 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
702 const struct sw_flow_actions *sf_acts;
704 sf_acts = rcu_dereference_check(flow->sf_acts,
705 lockdep_ovsl_is_held());
707 err = ovs_nla_put_actions(sf_acts->actions,
708 sf_acts->actions_len, skb);
710 nla_nest_end(skb, start);
715 nla_nest_cancel(skb, start);
717 } else if (skb_orig_len)
718 goto nla_put_failure;
720 return genlmsg_end(skb, ovs_header);
725 genlmsg_cancel(skb, ovs_header);
729 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
731 const struct sw_flow_actions *sf_acts;
733 sf_acts = ovsl_dereference(flow->sf_acts);
735 return genlmsg_new(ovs_flow_cmd_msg_size(sf_acts), GFP_KERNEL);
738 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
740 u32 portid, u32 seq, u8 cmd)
745 skb = ovs_flow_cmd_alloc_info(flow);
747 return ERR_PTR(-ENOMEM);
749 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
754 static struct sw_flow *__ovs_flow_tbl_lookup(struct flow_table *tbl,
755 const struct sw_flow_key *key)
757 u32 __always_unused n_mask_hit;
759 return ovs_flow_tbl_lookup(tbl, key, &n_mask_hit);
762 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
764 struct nlattr **a = info->attrs;
765 struct ovs_header *ovs_header = info->userhdr;
766 struct sw_flow_key key, masked_key;
767 struct sw_flow *flow = NULL;
768 struct sw_flow_mask mask;
769 struct sk_buff *reply;
771 struct sw_flow_actions *acts = NULL;
772 struct sw_flow_match match;
777 if (!a[OVS_FLOW_ATTR_KEY])
780 ovs_match_init(&match, &key, &mask);
781 error = ovs_nla_get_match(&match,
782 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
786 /* Validate actions. */
787 if (a[OVS_FLOW_ATTR_ACTIONS]) {
788 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
789 error = PTR_ERR(acts);
793 ovs_flow_mask_key(&masked_key, &key, &mask);
794 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
795 &masked_key, 0, &acts);
797 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
800 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
806 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
811 /* Check if this is a duplicate flow */
812 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
814 /* Bail out if we're not allowed to create a new flow. */
816 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
820 flow = ovs_flow_alloc();
822 error = PTR_ERR(flow);
827 flow->key = masked_key;
828 flow->unmasked_key = key;
829 rcu_assign_pointer(flow->sf_acts, acts);
831 /* Put flow in bucket. */
832 error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
838 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
839 info->snd_seq, OVS_FLOW_CMD_NEW);
841 /* We found a matching flow. */
842 struct sw_flow_actions *old_acts;
844 /* Bail out if we're not allowed to modify an existing flow.
845 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
846 * because Generic Netlink treats the latter as a dump
847 * request. We also accept NLM_F_EXCL in case that bug ever
851 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
852 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
855 /* The unmasked key has to be the same for flow updates. */
857 if (!ovs_flow_cmp_unmasked_key(flow, &match)) {
858 OVS_NLERR("Flow modification message rejected, unmasked key does not match.\n");
862 /* Update actions. */
863 old_acts = ovsl_dereference(flow->sf_acts);
864 rcu_assign_pointer(flow->sf_acts, acts);
865 ovs_nla_free_flow_actions(old_acts);
867 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
868 info->snd_seq, OVS_FLOW_CMD_NEW);
871 if (a[OVS_FLOW_ATTR_CLEAR]) {
872 spin_lock_bh(&flow->lock);
874 spin_unlock_bh(&flow->lock);
880 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
882 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
883 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
887 ovs_flow_free(flow, false);
896 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
898 struct nlattr **a = info->attrs;
899 struct ovs_header *ovs_header = info->userhdr;
900 struct sw_flow_key key;
901 struct sk_buff *reply;
902 struct sw_flow *flow;
904 struct sw_flow_match match;
907 if (!a[OVS_FLOW_ATTR_KEY]) {
908 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
912 ovs_match_init(&match, &key, NULL);
913 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
918 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
924 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
925 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
930 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
931 info->snd_seq, OVS_FLOW_CMD_NEW);
933 err = PTR_ERR(reply);
938 return genlmsg_reply(reply, info);
944 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
946 struct nlattr **a = info->attrs;
947 struct ovs_header *ovs_header = info->userhdr;
948 struct sw_flow_key key;
949 struct sk_buff *reply;
950 struct sw_flow *flow;
952 struct sw_flow_match match;
956 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
962 if (!a[OVS_FLOW_ATTR_KEY]) {
963 err = ovs_flow_tbl_flush(&dp->table);
967 ovs_match_init(&match, &key, NULL);
968 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
972 flow = __ovs_flow_tbl_lookup(&dp->table, &key);
973 if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
978 reply = ovs_flow_cmd_alloc_info(flow);
984 ovs_flow_tbl_remove(&dp->table, flow);
986 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
987 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
990 ovs_flow_free(flow, true);
993 ovs_notify(reply, info, &ovs_dp_flow_multicast_group);
1000 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1002 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1003 struct table_instance *ti;
1004 struct datapath *dp;
1007 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1013 ti = rcu_dereference(dp->table.ti);
1015 struct sw_flow *flow;
1018 bucket = cb->args[0];
1020 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1024 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1025 NETLINK_CB(cb->skb).portid,
1026 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1027 OVS_FLOW_CMD_NEW) < 0)
1030 cb->args[0] = bucket;
1037 static struct genl_ops dp_flow_genl_ops[] = {
1038 { .cmd = OVS_FLOW_CMD_NEW,
1039 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1040 .policy = flow_policy,
1041 .doit = ovs_flow_cmd_new_or_set
1043 { .cmd = OVS_FLOW_CMD_DEL,
1044 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1045 .policy = flow_policy,
1046 .doit = ovs_flow_cmd_del
1048 { .cmd = OVS_FLOW_CMD_GET,
1049 .flags = 0, /* OK for unprivileged users. */
1050 .policy = flow_policy,
1051 .doit = ovs_flow_cmd_get,
1052 .dumpit = ovs_flow_cmd_dump
1054 { .cmd = OVS_FLOW_CMD_SET,
1055 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1056 .policy = flow_policy,
1057 .doit = ovs_flow_cmd_new_or_set,
1061 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1062 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1063 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1066 static struct genl_family dp_datapath_genl_family = {
1067 .id = GENL_ID_GENERATE,
1068 .hdrsize = sizeof(struct ovs_header),
1069 .name = OVS_DATAPATH_FAMILY,
1070 .version = OVS_DATAPATH_VERSION,
1071 .maxattr = OVS_DP_ATTR_MAX,
1073 .parallel_ops = true,
1076 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1077 .name = OVS_DATAPATH_MCGROUP
1080 static size_t ovs_dp_cmd_msg_size(void)
1082 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1084 msgsize += nla_total_size(IFNAMSIZ);
1085 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1086 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1091 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1092 u32 portid, u32 seq, u32 flags, u8 cmd)
1094 struct ovs_header *ovs_header;
1095 struct ovs_dp_stats dp_stats;
1096 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1099 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1104 ovs_header->dp_ifindex = get_dpifindex(dp);
1107 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1110 goto nla_put_failure;
1112 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1113 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1115 goto nla_put_failure;
1117 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1118 sizeof(struct ovs_dp_megaflow_stats),
1119 &dp_megaflow_stats))
1120 goto nla_put_failure;
1122 return genlmsg_end(skb, ovs_header);
1125 genlmsg_cancel(skb, ovs_header);
1130 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1133 struct sk_buff *skb;
1136 skb = genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1138 return ERR_PTR(-ENOMEM);
1140 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1143 return ERR_PTR(retval);
1148 /* Called with ovs_mutex. */
1149 static struct datapath *lookup_datapath(struct net *net,
1150 struct ovs_header *ovs_header,
1151 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1153 struct datapath *dp;
1155 if (!a[OVS_DP_ATTR_NAME])
1156 dp = get_dp(net, ovs_header->dp_ifindex);
1158 struct vport *vport;
1161 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1162 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1165 return dp ? dp : ERR_PTR(-ENODEV);
1168 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1170 struct nlattr **a = info->attrs;
1171 struct vport_parms parms;
1172 struct sk_buff *reply;
1173 struct datapath *dp;
1174 struct vport *vport;
1175 struct ovs_net *ovs_net;
1179 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1185 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1187 goto err_unlock_ovs;
1189 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1191 /* Allocate table. */
1192 err = ovs_flow_tbl_init(&dp->table);
1196 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1197 if (!dp->stats_percpu) {
1199 goto err_destroy_table;
1202 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1206 goto err_destroy_percpu;
1209 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1210 INIT_HLIST_HEAD(&dp->ports[i]);
1212 /* Set up our datapath device. */
1213 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1214 parms.type = OVS_VPORT_TYPE_INTERNAL;
1215 parms.options = NULL;
1217 parms.port_no = OVSP_LOCAL;
1218 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1220 vport = new_vport(&parms);
1221 if (IS_ERR(vport)) {
1222 err = PTR_ERR(vport);
1226 goto err_destroy_ports_array;
1229 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1230 info->snd_seq, OVS_DP_CMD_NEW);
1231 err = PTR_ERR(reply);
1233 goto err_destroy_local_port;
1235 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1236 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1240 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1243 err_destroy_local_port:
1244 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1245 err_destroy_ports_array:
1248 free_percpu(dp->stats_percpu);
1250 ovs_flow_tbl_destroy(&dp->table);
1252 release_net(ovs_dp_get_net(dp));
1260 /* Called with ovs_mutex. */
1261 static void __dp_destroy(struct datapath *dp)
1265 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1266 struct vport *vport;
1267 struct hlist_node *n;
1269 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1270 if (vport->port_no != OVSP_LOCAL)
1271 ovs_dp_detach_port(vport);
1274 list_del_rcu(&dp->list_node);
1276 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1277 * all port in datapath are destroyed first before freeing datapath.
1279 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1281 call_rcu(&dp->rcu, destroy_dp_rcu);
1284 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1286 struct sk_buff *reply;
1287 struct datapath *dp;
1291 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1296 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1297 info->snd_seq, OVS_DP_CMD_DEL);
1298 err = PTR_ERR(reply);
1305 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1313 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1315 struct sk_buff *reply;
1316 struct datapath *dp;
1320 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1325 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1326 info->snd_seq, OVS_DP_CMD_NEW);
1327 if (IS_ERR(reply)) {
1328 err = PTR_ERR(reply);
1329 netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1330 ovs_dp_datapath_multicast_group.id, err);
1336 ovs_notify(reply, info, &ovs_dp_datapath_multicast_group);
1344 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1346 struct sk_buff *reply;
1347 struct datapath *dp;
1351 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1357 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1358 info->snd_seq, OVS_DP_CMD_NEW);
1359 if (IS_ERR(reply)) {
1360 err = PTR_ERR(reply);
1365 return genlmsg_reply(reply, info);
1372 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1374 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1375 struct datapath *dp;
1376 int skip = cb->args[0];
1380 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1382 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1383 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1384 OVS_DP_CMD_NEW) < 0)
1395 static struct genl_ops dp_datapath_genl_ops[] = {
1396 { .cmd = OVS_DP_CMD_NEW,
1397 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1398 .policy = datapath_policy,
1399 .doit = ovs_dp_cmd_new
1401 { .cmd = OVS_DP_CMD_DEL,
1402 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1403 .policy = datapath_policy,
1404 .doit = ovs_dp_cmd_del
1406 { .cmd = OVS_DP_CMD_GET,
1407 .flags = 0, /* OK for unprivileged users. */
1408 .policy = datapath_policy,
1409 .doit = ovs_dp_cmd_get,
1410 .dumpit = ovs_dp_cmd_dump
1412 { .cmd = OVS_DP_CMD_SET,
1413 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414 .policy = datapath_policy,
1415 .doit = ovs_dp_cmd_set,
1419 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1420 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1421 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1422 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1423 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1424 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1425 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1428 static struct genl_family dp_vport_genl_family = {
1429 .id = GENL_ID_GENERATE,
1430 .hdrsize = sizeof(struct ovs_header),
1431 .name = OVS_VPORT_FAMILY,
1432 .version = OVS_VPORT_VERSION,
1433 .maxattr = OVS_VPORT_ATTR_MAX,
1435 .parallel_ops = true,
1438 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1439 .name = OVS_VPORT_MCGROUP
1442 /* Called with ovs_mutex or RCU read lock. */
1443 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1444 u32 portid, u32 seq, u32 flags, u8 cmd)
1446 struct ovs_header *ovs_header;
1447 struct ovs_vport_stats vport_stats;
1450 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1455 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1457 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1458 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1459 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1460 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1461 goto nla_put_failure;
1463 ovs_vport_get_stats(vport, &vport_stats);
1464 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1466 goto nla_put_failure;
1468 err = ovs_vport_get_options(vport, skb);
1469 if (err == -EMSGSIZE)
1472 return genlmsg_end(skb, ovs_header);
1477 genlmsg_cancel(skb, ovs_header);
1481 /* Called with ovs_mutex or RCU read lock. */
1482 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1485 struct sk_buff *skb;
1488 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1490 return ERR_PTR(-ENOMEM);
1492 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1498 /* Called with ovs_mutex or RCU read lock. */
1499 static struct vport *lookup_vport(struct net *net,
1500 struct ovs_header *ovs_header,
1501 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1503 struct datapath *dp;
1504 struct vport *vport;
1506 if (a[OVS_VPORT_ATTR_NAME]) {
1507 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1509 return ERR_PTR(-ENODEV);
1510 if (ovs_header->dp_ifindex &&
1511 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1512 return ERR_PTR(-ENODEV);
1514 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1515 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1517 if (port_no >= DP_MAX_PORTS)
1518 return ERR_PTR(-EFBIG);
1520 dp = get_dp(net, ovs_header->dp_ifindex);
1522 return ERR_PTR(-ENODEV);
1524 vport = ovs_vport_ovsl_rcu(dp, port_no);
1526 return ERR_PTR(-ENODEV);
1529 return ERR_PTR(-EINVAL);
1532 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1534 struct nlattr **a = info->attrs;
1535 struct ovs_header *ovs_header = info->userhdr;
1536 struct vport_parms parms;
1537 struct sk_buff *reply;
1538 struct vport *vport;
1539 struct datapath *dp;
1544 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1545 !a[OVS_VPORT_ATTR_UPCALL_PID])
1549 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1554 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1555 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1558 if (port_no >= DP_MAX_PORTS)
1561 vport = ovs_vport_ovsl(dp, port_no);
1566 for (port_no = 1; ; port_no++) {
1567 if (port_no >= DP_MAX_PORTS) {
1571 vport = ovs_vport_ovsl(dp, port_no);
1577 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1578 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1579 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1581 parms.port_no = port_no;
1582 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1584 vport = new_vport(&parms);
1585 err = PTR_ERR(vport);
1590 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1592 if (IS_ERR(reply)) {
1593 err = PTR_ERR(reply);
1594 ovs_dp_detach_port(vport);
1598 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1606 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1608 struct nlattr **a = info->attrs;
1609 struct sk_buff *reply;
1610 struct vport *vport;
1614 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1615 err = PTR_ERR(vport);
1619 if (a[OVS_VPORT_ATTR_TYPE] &&
1620 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1625 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1631 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1632 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1637 if (a[OVS_VPORT_ATTR_UPCALL_PID])
1638 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1640 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1641 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1645 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1655 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1657 struct nlattr **a = info->attrs;
1658 struct sk_buff *reply;
1659 struct vport *vport;
1663 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1664 err = PTR_ERR(vport);
1668 if (vport->port_no == OVSP_LOCAL) {
1673 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1674 info->snd_seq, OVS_VPORT_CMD_DEL);
1675 err = PTR_ERR(reply);
1680 ovs_dp_detach_port(vport);
1682 ovs_notify(reply, info, &ovs_dp_vport_multicast_group);
1689 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1691 struct nlattr **a = info->attrs;
1692 struct ovs_header *ovs_header = info->userhdr;
1693 struct sk_buff *reply;
1694 struct vport *vport;
1698 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1699 err = PTR_ERR(vport);
1703 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1704 info->snd_seq, OVS_VPORT_CMD_NEW);
1705 err = PTR_ERR(reply);
1711 return genlmsg_reply(reply, info);
1718 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1720 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1721 struct datapath *dp;
1722 int bucket = cb->args[0], skip = cb->args[1];
1725 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1730 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1731 struct vport *vport;
1734 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1736 ovs_vport_cmd_fill_info(vport, skb,
1737 NETLINK_CB(cb->skb).portid,
1740 OVS_VPORT_CMD_NEW) < 0)
1756 static struct genl_ops dp_vport_genl_ops[] = {
1757 { .cmd = OVS_VPORT_CMD_NEW,
1758 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1759 .policy = vport_policy,
1760 .doit = ovs_vport_cmd_new
1762 { .cmd = OVS_VPORT_CMD_DEL,
1763 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1764 .policy = vport_policy,
1765 .doit = ovs_vport_cmd_del
1767 { .cmd = OVS_VPORT_CMD_GET,
1768 .flags = 0, /* OK for unprivileged users. */
1769 .policy = vport_policy,
1770 .doit = ovs_vport_cmd_get,
1771 .dumpit = ovs_vport_cmd_dump
1773 { .cmd = OVS_VPORT_CMD_SET,
1774 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1775 .policy = vport_policy,
1776 .doit = ovs_vport_cmd_set,
1780 struct genl_family_and_ops {
1781 struct genl_family *family;
1782 struct genl_ops *ops;
1784 struct genl_multicast_group *group;
1787 static const struct genl_family_and_ops dp_genl_families[] = {
1788 { &dp_datapath_genl_family,
1789 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1790 &ovs_dp_datapath_multicast_group },
1791 { &dp_vport_genl_family,
1792 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1793 &ovs_dp_vport_multicast_group },
1794 { &dp_flow_genl_family,
1795 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1796 &ovs_dp_flow_multicast_group },
1797 { &dp_packet_genl_family,
1798 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1802 static void dp_unregister_genl(int n_families)
1806 for (i = 0; i < n_families; i++)
1807 genl_unregister_family(dp_genl_families[i].family);
1810 static int dp_register_genl(void)
1817 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1818 const struct genl_family_and_ops *f = &dp_genl_families[i];
1820 err = genl_register_family_with_ops(f->family, f->ops,
1827 err = genl_register_mc_group(f->family, f->group);
1836 dp_unregister_genl(n_registered);
1840 static int __net_init ovs_init_net(struct net *net)
1842 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1844 INIT_LIST_HEAD(&ovs_net->dps);
1845 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1849 static void __net_exit ovs_exit_net(struct net *net)
1851 struct datapath *dp, *dp_next;
1852 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1855 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1859 cancel_work_sync(&ovs_net->dp_notify_work);
1862 static struct pernet_operations ovs_net_ops = {
1863 .init = ovs_init_net,
1864 .exit = ovs_exit_net,
1866 .size = sizeof(struct ovs_net),
1869 static int __init dp_init(void)
1873 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1875 pr_info("Open vSwitch switching datapath\n");
1877 err = ovs_flow_init();
1881 err = ovs_vport_init();
1883 goto error_flow_exit;
1885 err = register_pernet_device(&ovs_net_ops);
1887 goto error_vport_exit;
1889 err = register_netdevice_notifier(&ovs_dp_device_notifier);
1891 goto error_netns_exit;
1893 err = dp_register_genl();
1895 goto error_unreg_notifier;
1899 error_unreg_notifier:
1900 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1902 unregister_pernet_device(&ovs_net_ops);
1911 static void dp_cleanup(void)
1913 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1914 unregister_netdevice_notifier(&ovs_dp_device_notifier);
1915 unregister_pernet_device(&ovs_net_ops);
1921 module_init(dp_init);
1922 module_exit(dp_cleanup);
1924 MODULE_DESCRIPTION("Open vSwitch switching datapath");
1925 MODULE_LICENSE("GPL");