2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <linux/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/genetlink.h>
54 #include <net/net_namespace.h>
55 #include <net/netns/generic.h>
59 #include "flow_table.h"
60 #include "flow_netlink.h"
62 #include "vport-internal_dev.h"
63 #include "vport-netdev.h"
65 int ovs_net_id __read_mostly;
67 static struct genl_family dp_packet_genl_family;
68 static struct genl_family dp_flow_genl_family;
69 static struct genl_family dp_datapath_genl_family;
71 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
72 .name = OVS_FLOW_MCGROUP
75 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
76 .name = OVS_DATAPATH_MCGROUP
79 struct genl_multicast_group ovs_dp_vport_multicast_group = {
80 .name = OVS_VPORT_MCGROUP
83 /* Check if need to build a reply message.
84 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
85 static bool ovs_must_notify(struct genl_info *info,
86 const struct genl_multicast_group *grp)
88 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
89 netlink_has_listeners(genl_info_net(info)->genl_sock, GROUP_ID(grp));
92 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
93 struct sk_buff *skb, struct genl_info *info)
95 genl_notify(family, skb, genl_info_net(info),
96 info->snd_portid, GROUP_ID(grp), info->nlhdr, GFP_KERNEL);
102 * All writes e.g. Writes to device state (add/remove datapath, port, set
103 * operations on vports, etc.), Writes to other state (flow table
104 * modifications, set miscellaneous datapath parameters, etc.) are protected
107 * Reads are protected by RCU.
109 * There are a few special cases (mostly stats) that have their own
110 * synchronization but they nest under all of above and don't interact with
113 * The RTNL lock nests inside ovs_mutex.
116 static DEFINE_MUTEX(ovs_mutex);
120 mutex_lock(&ovs_mutex);
123 void ovs_unlock(void)
125 mutex_unlock(&ovs_mutex);
128 #ifdef CONFIG_LOCKDEP
129 int lockdep_ovsl_is_held(void)
132 return lockdep_is_held(&ovs_mutex);
138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 const struct dp_upcall_info *);
140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 const struct dp_upcall_info *);
143 /* Must be called with rcu_read_lock. */
144 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
146 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
149 struct vport *vport = ovs_internal_dev_get_vport(dev);
157 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
158 * returned dp pointer valid. */
159 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
163 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
165 dp = get_dp_rcu(net, dp_ifindex);
171 /* Must be called with rcu_read_lock or ovs_mutex. */
172 const char *ovs_dp_name(const struct datapath *dp)
174 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
175 return vport->ops->get_name(vport);
178 static int get_dpifindex(struct datapath *dp)
185 local = ovs_vport_rcu(dp, OVSP_LOCAL);
187 ifindex = netdev_vport_priv(local)->dev->ifindex;
196 static void destroy_dp_rcu(struct rcu_head *rcu)
198 struct datapath *dp = container_of(rcu, struct datapath, rcu);
200 ovs_flow_tbl_destroy(&dp->table);
201 free_percpu(dp->stats_percpu);
202 release_net(ovs_dp_get_net(dp));
207 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
210 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
213 /* Called with ovs_mutex or RCU read lock. */
214 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
217 struct hlist_head *head;
219 head = vport_hash_bucket(dp, port_no);
220 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
221 if (vport->port_no == port_no)
227 /* Called with ovs_mutex. */
228 static struct vport *new_vport(const struct vport_parms *parms)
232 vport = ovs_vport_add(parms);
233 if (!IS_ERR(vport)) {
234 struct datapath *dp = parms->dp;
235 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
237 hlist_add_head_rcu(&vport->dp_hash_node, head);
242 void ovs_dp_detach_port(struct vport *p)
246 /* First drop references to device. */
247 hlist_del_rcu(&p->dp_hash_node);
249 /* Then destroy it. */
253 void ovs_dp_process_packet_with_key(struct sk_buff *skb,
254 struct sw_flow_key *pkt_key,
257 const struct vport *p = OVS_CB(skb)->input_vport;
258 struct datapath *dp = p->dp;
259 struct sw_flow *flow;
260 struct dp_stats_percpu *stats;
264 stats = this_cpu_ptr(dp->stats_percpu);
265 OVS_CB(skb)->pkt_key = pkt_key;
268 flow = ovs_flow_tbl_lookup_stats(&dp->table, pkt_key, skb_get_hash(skb),
270 if (unlikely(!flow)) {
271 struct dp_upcall_info upcall;
273 upcall.cmd = OVS_PACKET_CMD_MISS;
274 upcall.userdata = NULL;
275 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
276 ovs_dp_upcall(dp, skb, &upcall);
278 stats_counter = &stats->n_missed;
282 OVS_CB(skb)->flow = flow;
284 ovs_flow_stats_update(OVS_CB(skb)->flow, pkt_key->tp.flags, skb);
285 ovs_execute_actions(dp, skb, recirc);
286 stats_counter = &stats->n_hit;
289 /* Update datapath statistics. */
290 u64_stats_update_begin(&stats->sync);
292 stats->n_mask_hit += n_mask_hit;
293 u64_stats_update_end(&stats->sync);
296 /* Must be called with rcu_read_lock. */
297 void ovs_dp_process_received_packet(struct sk_buff *skb)
300 struct sw_flow_key key;
302 /* Extract flow from 'skb' into 'key'. */
303 error = ovs_flow_key_extract(skb, &key);
304 if (unlikely(error)) {
309 ovs_dp_process_packet_with_key(skb, &key, false);
312 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
313 const struct dp_upcall_info *upcall_info)
315 struct dp_stats_percpu *stats;
318 BUG_ON(!OVS_CB(skb)->pkt_key);
320 if (upcall_info->portid == 0) {
325 if (!skb_is_gso(skb))
326 err = queue_userspace_packet(dp, skb, upcall_info);
328 err = queue_gso_packets(dp, skb, upcall_info);
335 stats = this_cpu_ptr(dp->stats_percpu);
337 u64_stats_update_begin(&stats->sync);
339 u64_stats_update_end(&stats->sync);
344 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
345 const struct dp_upcall_info *upcall_info)
347 unsigned short gso_type = skb_shinfo(skb)->gso_type;
348 struct sw_flow_key later_key;
349 struct sk_buff *segs, *nskb;
352 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
354 return PTR_ERR(segs);
356 if (gso_type & SKB_GSO_UDP) {
357 /* The initial flow key extracted by ovs_flow_key_extract()
358 * in this case is for a first fragment, so we need to
359 * properly mark later fragments.
361 later_key = *OVS_CB(skb)->pkt_key;
362 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
365 /* Queue all of the segments. */
368 if (gso_type & SKB_GSO_UDP && skb != segs)
369 OVS_CB(skb)->pkt_key = &later_key;
371 err = queue_userspace_packet(dp, skb, upcall_info);
375 } while ((skb = skb->next));
377 /* Free all of the segments. */
385 } while ((skb = nskb));
389 static size_t key_attr_size(void)
391 /* Whenever adding new OVS_KEY_ FIELDS, we should consider
392 * updating this function. */
393 BUILD_BUG_ON(OVS_KEY_ATTR_TUNNEL_INFO != 22);
395 return nla_total_size(4) /* OVS_KEY_ATTR_PRIORITY */
396 + nla_total_size(0) /* OVS_KEY_ATTR_TUNNEL */
397 + nla_total_size(8) /* OVS_TUNNEL_KEY_ATTR_ID */
398 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
399 + nla_total_size(4) /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
400 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TOS */
401 + nla_total_size(1) /* OVS_TUNNEL_KEY_ATTR_TTL */
402 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
403 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_CSUM */
404 + nla_total_size(0) /* OVS_TUNNEL_KEY_ATTR_OAM */
405 + nla_total_size(256) /* OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS */
406 + nla_total_size(4) /* OVS_KEY_ATTR_IN_PORT */
407 + nla_total_size(4) /* OVS_KEY_ATTR_SKB_MARK */
408 + nla_total_size(4) /* OVS_KEY_ATTR_DP_HASH */
409 + nla_total_size(4) /* OVS_KEY_ATTR_RECIRC_ID */
410 + nla_total_size(12) /* OVS_KEY_ATTR_ETHERNET */
411 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
412 + nla_total_size(4) /* OVS_KEY_ATTR_8021Q */
413 + nla_total_size(0) /* OVS_KEY_ATTR_ENCAP */
414 + nla_total_size(2) /* OVS_KEY_ATTR_ETHERTYPE */
415 + nla_total_size(40) /* OVS_KEY_ATTR_IPV6 */
416 + nla_total_size(2) /* OVS_KEY_ATTR_ICMPV6 */
417 + nla_total_size(28); /* OVS_KEY_ATTR_ND */
420 static size_t upcall_msg_size(const struct nlattr *userdata,
423 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
424 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
425 + nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
427 /* OVS_PACKET_ATTR_USERDATA */
429 size += NLA_ALIGN(userdata->nla_len);
434 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
435 const struct dp_upcall_info *upcall_info)
437 struct ovs_header *upcall;
438 struct sk_buff *nskb = NULL;
439 struct sk_buff *user_skb; /* to be queued to userspace */
440 struct sw_flow_key *pkt_key = OVS_CB(skb)->pkt_key;
442 struct genl_info info = {
443 #if LINUX_VERSION_CODE >= KERNEL_VERSION(3,14,0)
444 .dst_sk = ovs_dp_get_net(dp)->genl_sock,
446 .snd_portid = upcall_info->portid,
452 dp_ifindex = get_dpifindex(dp);
456 if (vlan_tx_tag_present(skb)) {
457 nskb = skb_clone(skb, GFP_ATOMIC);
461 nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
465 vlan_set_tci(nskb, 0);
470 if (nla_attr_size(skb->len) > USHRT_MAX) {
475 /* Complete checksum if needed */
476 if (skb->ip_summed == CHECKSUM_PARTIAL &&
477 (err = skb_checksum_help(skb)))
480 /* Older versions of OVS user space enforce alignment of the last
481 * Netlink attribute to NLA_ALIGNTO which would require extensive
482 * padding logic. Only perform zerocopy if padding is not required.
484 if (dp->user_features & OVS_DP_F_UNALIGNED)
485 hlen = skb_zerocopy_headlen(skb);
489 len = upcall_msg_size(upcall_info->userdata, hlen);
490 user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
496 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
497 0, upcall_info->cmd);
498 upcall->dp_ifindex = dp_ifindex;
500 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
501 err = ovs_nla_put_flow(dp, pkt_key, pkt_key, user_skb);
503 nla_nest_end(user_skb, nla);
505 if (upcall_info->userdata)
506 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
507 nla_len(upcall_info->userdata),
508 nla_data(upcall_info->userdata));
510 /* Only reserve room for attribute header, packet data is added
511 * in skb_zerocopy() */
512 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
516 nla->nla_len = nla_attr_size(skb->len);
518 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
522 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
523 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
524 size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
527 memset(skb_put(user_skb, plen), 0, plen);
530 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
532 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
540 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
542 struct ovs_header *ovs_header = info->userhdr;
543 struct nlattr **a = info->attrs;
544 struct sw_flow_actions *acts;
545 struct sk_buff *packet;
546 struct sw_flow *flow;
549 struct vport *input_vport;
554 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
555 !a[OVS_PACKET_ATTR_ACTIONS])
558 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
559 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
563 skb_reserve(packet, NET_IP_ALIGN);
565 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
567 skb_reset_mac_header(packet);
568 eth = eth_hdr(packet);
570 /* Normally, setting the skb 'protocol' field would be handled by a
571 * call to eth_type_trans(), but it assumes there's a sending
572 * device, which we may not have. */
573 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
574 packet->protocol = eth->h_proto;
576 packet->protocol = htons(ETH_P_802_2);
578 /* Build an sw_flow for sending this packet. */
579 flow = ovs_flow_alloc();
584 err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
589 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
594 err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
596 rcu_assign_pointer(flow->sf_acts, acts);
600 OVS_CB(packet)->flow = flow;
601 OVS_CB(packet)->pkt_key = &flow->key;
602 packet->priority = flow->key.phy.priority;
603 packet->mark = flow->key.phy.skb_mark;
606 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
611 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
613 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
618 OVS_CB(packet)->input_vport = input_vport;
621 err = ovs_execute_actions(dp, packet, false);
625 ovs_flow_free(flow, false);
631 ovs_flow_free(flow, false);
638 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
639 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
640 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
641 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
644 static struct genl_ops dp_packet_genl_ops[] = {
645 { .cmd = OVS_PACKET_CMD_EXECUTE,
646 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
647 .policy = packet_policy,
648 .doit = ovs_packet_cmd_execute
652 static struct genl_family dp_packet_genl_family = {
653 .id = GENL_ID_GENERATE,
654 .hdrsize = sizeof(struct ovs_header),
655 .name = OVS_PACKET_FAMILY,
656 .version = OVS_PACKET_VERSION,
657 .maxattr = OVS_PACKET_ATTR_MAX,
659 .parallel_ops = true,
660 .ops = dp_packet_genl_ops,
661 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
664 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
665 struct ovs_dp_megaflow_stats *mega_stats)
669 memset(mega_stats, 0, sizeof(*mega_stats));
671 stats->n_flows = ovs_flow_tbl_count(&dp->table);
672 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
674 stats->n_hit = stats->n_missed = stats->n_lost = 0;
676 for_each_possible_cpu(i) {
677 const struct dp_stats_percpu *percpu_stats;
678 struct dp_stats_percpu local_stats;
681 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
684 start = u64_stats_fetch_begin_irq(&percpu_stats->sync);
685 local_stats = *percpu_stats;
686 } while (u64_stats_fetch_retry_irq(&percpu_stats->sync, start));
688 stats->n_hit += local_stats.n_hit;
689 stats->n_missed += local_stats.n_missed;
690 stats->n_lost += local_stats.n_lost;
691 mega_stats->n_mask_hit += local_stats.n_mask_hit;
695 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
697 return NLMSG_ALIGN(sizeof(struct ovs_header))
698 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
699 + nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
700 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
701 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
702 + nla_total_size(8) /* OVS_FLOW_ATTR_USED */
703 + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
706 /* Called with ovs_mutex or RCU read lock. */
707 static int ovs_flow_cmd_fill_match(struct datapath *dp,
708 const struct sw_flow *flow,
715 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
719 err = ovs_nla_put_flow(dp, &flow->unmasked_key,
720 &flow->unmasked_key, skb);
723 nla_nest_end(skb, nla);
725 /* Fill flow mask. */
726 nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
730 err = ovs_nla_put_flow(dp, &flow->key, &flow->mask->key, skb);
733 nla_nest_end(skb, nla);
738 /* Called with ovs_mutex or RCU read lock. */
739 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
742 struct ovs_flow_stats stats;
746 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
749 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
752 if (stats.n_packets &&
753 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
756 if ((u8)ntohs(tcp_flags) &&
757 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
763 /* Called with ovs_mutex or RCU read lock. */
764 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
765 struct sk_buff *skb, int skb_orig_len)
767 struct nlattr *start;
770 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
771 * this is the first flow to be dumped into 'skb'. This is unusual for
772 * Netlink but individual action lists can be longer than
773 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
774 * The userspace caller can always fetch the actions separately if it
775 * really wants them. (Most userspace callers in fact don't care.)
777 * This can only fail for dump operations because the skb is always
778 * properly sized for single flows.
780 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
782 const struct sw_flow_actions *sf_acts;
784 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
785 err = ovs_nla_put_actions(sf_acts->actions,
786 sf_acts->actions_len, skb);
789 nla_nest_end(skb, start);
794 nla_nest_cancel(skb, start);
796 } else if (skb_orig_len) {
803 /* Called with ovs_mutex or RCU read lock. */
804 static int ovs_flow_cmd_fill_info(struct datapath *dp,
805 const struct sw_flow *flow, int dp_ifindex,
806 struct sk_buff *skb, u32 portid,
807 u32 seq, u32 flags, u8 cmd)
809 const int skb_orig_len = skb->len;
810 struct ovs_header *ovs_header;
813 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
816 ovs_header->dp_ifindex = dp_ifindex;
818 err = ovs_flow_cmd_fill_match(dp, flow, skb);
822 err = ovs_flow_cmd_fill_stats(flow, skb);
826 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
830 return genlmsg_end(skb, ovs_header);
833 genlmsg_cancel(skb, ovs_header);
837 /* May not be called with RCU read lock. */
838 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
839 struct genl_info *info,
844 if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
847 skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
850 return ERR_PTR(-ENOMEM);
855 /* Called with ovs_mutex. */
856 static struct sk_buff *ovs_flow_cmd_build_info(struct datapath *dp,
857 const struct sw_flow *flow,
859 struct genl_info *info, u8 cmd,
865 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
867 if (!skb || IS_ERR(skb))
870 retval = ovs_flow_cmd_fill_info(dp, flow, dp_ifindex, skb,
871 info->snd_portid, info->snd_seq, 0,
877 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
879 struct nlattr **a = info->attrs;
880 struct ovs_header *ovs_header = info->userhdr;
881 struct sw_flow *flow, *new_flow;
882 struct sw_flow_mask mask;
883 struct sk_buff *reply;
885 struct sw_flow_actions *acts;
886 struct sw_flow_match match;
889 /* Must have key and actions. */
891 if (!a[OVS_FLOW_ATTR_KEY]) {
892 OVS_NLERR("Flow key attribute not present in new flow.\n");
895 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
896 OVS_NLERR("Flow actions attribute not present in new flow.\n");
900 /* Most of the time we need to allocate a new flow, do it before
902 new_flow = ovs_flow_alloc();
903 if (IS_ERR(new_flow)) {
904 error = PTR_ERR(new_flow);
909 ovs_match_init(&match, &new_flow->unmasked_key, &mask);
910 error = ovs_nla_get_match(&match,
911 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
915 ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
917 /* Validate actions. */
918 acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
919 error = PTR_ERR(acts);
923 error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
926 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
930 reply = ovs_flow_cmd_alloc_info(acts, info, false);
932 error = PTR_ERR(reply);
937 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
942 /* Check if this is a duplicate flow */
943 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
945 rcu_assign_pointer(new_flow->sf_acts, acts);
947 /* Put flow in bucket. */
948 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
949 if (unlikely(error)) {
954 if (unlikely(reply)) {
955 error = ovs_flow_cmd_fill_info(dp, new_flow,
956 ovs_header->dp_ifindex,
957 reply, info->snd_portid,
964 struct sw_flow_actions *old_acts;
966 /* Bail out if we're not allowed to modify an existing flow.
967 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
968 * because Generic Netlink treats the latter as a dump
969 * request. We also accept NLM_F_EXCL in case that bug ever
972 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
977 /* The unmasked key has to be the same for flow updates. */
978 if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
979 /* Look for any overlapping flow. */
980 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
986 /* Update actions. */
987 old_acts = ovsl_dereference(flow->sf_acts);
988 rcu_assign_pointer(flow->sf_acts, acts);
990 if (unlikely(reply)) {
991 error = ovs_flow_cmd_fill_info(dp, flow,
992 ovs_header->dp_ifindex,
993 reply, info->snd_portid,
1000 ovs_nla_free_flow_actions(old_acts);
1001 ovs_flow_free(new_flow, false);
1005 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1014 ovs_flow_free(new_flow, false);
1019 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
1020 const struct sw_flow_key *key,
1021 const struct sw_flow_mask *mask)
1023 struct sw_flow_actions *acts;
1024 struct sw_flow_key masked_key;
1027 acts = ovs_nla_alloc_flow_actions(nla_len(a));
1031 ovs_flow_mask_key(&masked_key, key, mask);
1032 error = ovs_nla_copy_actions(a, &masked_key, &acts);
1034 OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
1036 return ERR_PTR(error);
1042 static int ovs_flow_cmd_set(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 sw_flow *flow;
1048 struct sw_flow_mask mask;
1049 struct sk_buff *reply = NULL;
1050 struct datapath *dp;
1051 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1052 struct sw_flow_match match;
1057 if (!a[OVS_FLOW_ATTR_KEY]) {
1058 OVS_NLERR("Flow key attribute not present in set flow.\n");
1062 ovs_match_init(&match, &key, &mask);
1063 error = ovs_nla_get_match(&match,
1064 a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
1068 /* Validate actions. */
1069 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1070 acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask);
1072 error = PTR_ERR(acts);
1077 /* Can allocate before locking if have acts. */
1079 reply = ovs_flow_cmd_alloc_info(acts, info, false);
1080 if (IS_ERR(reply)) {
1081 error = PTR_ERR(reply);
1082 goto err_kfree_acts;
1087 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1088 if (unlikely(!dp)) {
1090 goto err_unlock_ovs;
1092 /* Check that the flow exists. */
1093 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1094 if (unlikely(!flow)) {
1096 goto err_unlock_ovs;
1099 /* Update actions, if present. */
1101 old_acts = ovsl_dereference(flow->sf_acts);
1102 rcu_assign_pointer(flow->sf_acts, acts);
1104 if (unlikely(reply)) {
1105 error = ovs_flow_cmd_fill_info(dp, flow,
1106 ovs_header->dp_ifindex,
1107 reply, info->snd_portid,
1113 /* Could not alloc without acts before locking. */
1114 reply = ovs_flow_cmd_build_info(dp, flow,
1115 ovs_header->dp_ifindex,
1116 info, OVS_FLOW_CMD_NEW, false);
1117 if (unlikely(IS_ERR(reply))) {
1118 error = PTR_ERR(reply);
1119 goto err_unlock_ovs;
1124 if (a[OVS_FLOW_ATTR_CLEAR])
1125 ovs_flow_stats_clear(flow);
1129 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1131 ovs_nla_free_flow_actions(old_acts);
1143 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1145 struct nlattr **a = info->attrs;
1146 struct ovs_header *ovs_header = info->userhdr;
1147 struct sw_flow_key key;
1148 struct sk_buff *reply;
1149 struct sw_flow *flow;
1150 struct datapath *dp;
1151 struct sw_flow_match match;
1154 if (!a[OVS_FLOW_ATTR_KEY]) {
1155 OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1159 ovs_match_init(&match, &key, NULL);
1160 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1165 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1171 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1177 reply = ovs_flow_cmd_build_info(dp, flow, ovs_header->dp_ifindex, info,
1178 OVS_FLOW_CMD_NEW, true);
1179 if (IS_ERR(reply)) {
1180 err = PTR_ERR(reply);
1185 return genlmsg_reply(reply, info);
1191 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1193 struct nlattr **a = info->attrs;
1194 struct ovs_header *ovs_header = info->userhdr;
1195 struct sw_flow_key key;
1196 struct sk_buff *reply;
1197 struct sw_flow *flow;
1198 struct datapath *dp;
1199 struct sw_flow_match match;
1202 if (likely(a[OVS_FLOW_ATTR_KEY])) {
1203 ovs_match_init(&match, &key, NULL);
1204 err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1210 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1211 if (unlikely(!dp)) {
1215 if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1216 err = ovs_flow_tbl_flush(&dp->table);
1219 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1220 if (unlikely(!flow)) {
1225 ovs_flow_tbl_remove(&dp->table, flow);
1228 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *)flow->sf_acts,
1231 if (likely(reply)) {
1232 if (likely(!IS_ERR(reply))) {
1233 rcu_read_lock(); /* Keep RCU checker happy. */
1234 err = ovs_flow_cmd_fill_info(dp, flow,
1235 ovs_header->dp_ifindex,
1236 reply, info->snd_portid,
1241 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1243 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1244 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1249 ovs_flow_free(flow, true);
1256 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1258 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1259 struct table_instance *ti;
1260 struct datapath *dp;
1263 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1269 ti = rcu_dereference(dp->table.ti);
1271 struct sw_flow *flow;
1274 bucket = cb->args[0];
1276 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1280 if (ovs_flow_cmd_fill_info(dp, flow, ovs_header->dp_ifindex, skb,
1281 NETLINK_CB(cb->skb).portid,
1282 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1283 OVS_FLOW_CMD_NEW) < 0)
1286 cb->args[0] = bucket;
1293 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1294 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1295 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1296 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1299 static struct genl_ops dp_flow_genl_ops[] = {
1300 { .cmd = OVS_FLOW_CMD_NEW,
1301 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1302 .policy = flow_policy,
1303 .doit = ovs_flow_cmd_new
1305 { .cmd = OVS_FLOW_CMD_DEL,
1306 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1307 .policy = flow_policy,
1308 .doit = ovs_flow_cmd_del
1310 { .cmd = OVS_FLOW_CMD_GET,
1311 .flags = 0, /* OK for unprivileged users. */
1312 .policy = flow_policy,
1313 .doit = ovs_flow_cmd_get,
1314 .dumpit = ovs_flow_cmd_dump
1316 { .cmd = OVS_FLOW_CMD_SET,
1317 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1318 .policy = flow_policy,
1319 .doit = ovs_flow_cmd_set,
1323 static struct genl_family dp_flow_genl_family = {
1324 .id = GENL_ID_GENERATE,
1325 .hdrsize = sizeof(struct ovs_header),
1326 .name = OVS_FLOW_FAMILY,
1327 .version = OVS_FLOW_VERSION,
1328 .maxattr = OVS_FLOW_ATTR_MAX,
1330 .parallel_ops = true,
1331 .ops = dp_flow_genl_ops,
1332 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1333 .mcgrps = &ovs_dp_flow_multicast_group,
1337 static size_t ovs_dp_cmd_msg_size(void)
1339 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1341 msgsize += nla_total_size(IFNAMSIZ);
1342 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1343 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1344 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1349 /* Called with ovs_mutex or RCU read lock. */
1350 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1351 u32 portid, u32 seq, u32 flags, u8 cmd)
1353 struct ovs_header *ovs_header;
1354 struct ovs_dp_stats dp_stats;
1355 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1358 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1363 ovs_header->dp_ifindex = get_dpifindex(dp);
1365 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1367 goto nla_put_failure;
1369 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1370 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1372 goto nla_put_failure;
1374 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1375 sizeof(struct ovs_dp_megaflow_stats),
1376 &dp_megaflow_stats))
1377 goto nla_put_failure;
1379 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1380 goto nla_put_failure;
1382 return genlmsg_end(skb, ovs_header);
1385 genlmsg_cancel(skb, ovs_header);
1390 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1392 return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1395 /* Called with rcu_read_lock or ovs_mutex. */
1396 static struct datapath *lookup_datapath(struct net *net,
1397 struct ovs_header *ovs_header,
1398 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1400 struct datapath *dp;
1402 if (!a[OVS_DP_ATTR_NAME])
1403 dp = get_dp(net, ovs_header->dp_ifindex);
1405 struct vport *vport;
1407 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1408 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1410 return dp ? dp : ERR_PTR(-ENODEV);
1413 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1415 struct datapath *dp;
1417 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1421 WARN(dp->user_features, "Dropping previously announced user features\n");
1422 dp->user_features = 0;
1425 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1427 if (a[OVS_DP_ATTR_USER_FEATURES])
1428 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1431 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1433 struct nlattr **a = info->attrs;
1434 struct vport_parms parms;
1435 struct sk_buff *reply;
1436 struct datapath *dp;
1437 struct vport *vport;
1438 struct ovs_net *ovs_net;
1442 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1445 reply = ovs_dp_cmd_alloc_info(info);
1450 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1452 goto err_free_reply;
1454 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1456 /* Allocate table. */
1457 err = ovs_flow_tbl_init(&dp->table);
1461 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1462 if (!dp->stats_percpu) {
1464 goto err_destroy_table;
1467 for_each_possible_cpu(i) {
1468 struct dp_stats_percpu *dpath_stats;
1469 dpath_stats = per_cpu_ptr(dp->stats_percpu, i);
1470 u64_stats_init(&dpath_stats->sync);
1473 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1477 goto err_destroy_percpu;
1480 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1481 INIT_HLIST_HEAD(&dp->ports[i]);
1483 /* Set up our datapath device. */
1484 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1485 parms.type = OVS_VPORT_TYPE_INTERNAL;
1486 parms.options = NULL;
1488 parms.port_no = OVSP_LOCAL;
1489 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1491 ovs_dp_change(dp, a);
1493 /* So far only local changes have been made, now need the lock. */
1496 vport = new_vport(&parms);
1497 if (IS_ERR(vport)) {
1498 err = PTR_ERR(vport);
1502 if (err == -EEXIST) {
1503 /* An outdated user space instance that does not understand
1504 * the concept of user_features has attempted to create a new
1505 * datapath and is likely to reuse it. Drop all user features.
1507 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1508 ovs_dp_reset_user_features(skb, info);
1511 goto err_destroy_ports_array;
1514 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1515 info->snd_seq, 0, OVS_DP_CMD_NEW);
1518 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1519 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1523 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1526 err_destroy_ports_array:
1530 free_percpu(dp->stats_percpu);
1532 ovs_flow_tbl_destroy(&dp->table);
1534 release_net(ovs_dp_get_net(dp));
1542 /* Called with ovs_mutex. */
1543 static void __dp_destroy(struct datapath *dp)
1547 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1548 struct vport *vport;
1549 struct hlist_node *n;
1551 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1552 if (vport->port_no != OVSP_LOCAL)
1553 ovs_dp_detach_port(vport);
1556 list_del_rcu(&dp->list_node);
1558 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1559 * all ports in datapath are destroyed first before freeing datapath.
1561 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1563 /* RCU destroy the flow table */
1564 call_rcu(&dp->rcu, destroy_dp_rcu);
1567 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1569 struct sk_buff *reply;
1570 struct datapath *dp;
1573 reply = ovs_dp_cmd_alloc_info(info);
1578 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1581 goto err_unlock_free;
1583 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1584 info->snd_seq, 0, OVS_DP_CMD_DEL);
1590 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1599 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1601 struct sk_buff *reply;
1602 struct datapath *dp;
1605 reply = ovs_dp_cmd_alloc_info(info);
1610 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1613 goto err_unlock_free;
1615 ovs_dp_change(dp, info->attrs);
1617 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1618 info->snd_seq, 0, OVS_DP_CMD_NEW);
1622 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1631 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1633 struct sk_buff *reply;
1634 struct datapath *dp;
1637 reply = ovs_dp_cmd_alloc_info(info);
1642 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1645 goto err_unlock_free;
1647 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1648 info->snd_seq, 0, OVS_DP_CMD_NEW);
1652 return genlmsg_reply(reply, info);
1660 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1662 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1663 struct datapath *dp;
1664 int skip = cb->args[0];
1668 list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1670 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1671 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1672 OVS_DP_CMD_NEW) < 0)
1683 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1684 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1685 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1686 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1689 static struct genl_ops dp_datapath_genl_ops[] = {
1690 { .cmd = OVS_DP_CMD_NEW,
1691 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1692 .policy = datapath_policy,
1693 .doit = ovs_dp_cmd_new
1695 { .cmd = OVS_DP_CMD_DEL,
1696 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1697 .policy = datapath_policy,
1698 .doit = ovs_dp_cmd_del
1700 { .cmd = OVS_DP_CMD_GET,
1701 .flags = 0, /* OK for unprivileged users. */
1702 .policy = datapath_policy,
1703 .doit = ovs_dp_cmd_get,
1704 .dumpit = ovs_dp_cmd_dump
1706 { .cmd = OVS_DP_CMD_SET,
1707 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1708 .policy = datapath_policy,
1709 .doit = ovs_dp_cmd_set,
1713 static struct genl_family dp_datapath_genl_family = {
1714 .id = GENL_ID_GENERATE,
1715 .hdrsize = sizeof(struct ovs_header),
1716 .name = OVS_DATAPATH_FAMILY,
1717 .version = OVS_DATAPATH_VERSION,
1718 .maxattr = OVS_DP_ATTR_MAX,
1720 .parallel_ops = true,
1721 .ops = dp_datapath_genl_ops,
1722 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1723 .mcgrps = &ovs_dp_datapath_multicast_group,
1727 /* Called with ovs_mutex or RCU read lock. */
1728 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1729 u32 portid, u32 seq, u32 flags, u8 cmd)
1731 struct ovs_header *ovs_header;
1732 struct ovs_vport_stats vport_stats;
1735 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1740 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1742 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1743 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1744 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)))
1745 goto nla_put_failure;
1747 ovs_vport_get_stats(vport, &vport_stats);
1748 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1750 goto nla_put_failure;
1752 if (ovs_vport_get_upcall_portids(vport, skb))
1753 goto nla_put_failure;
1755 err = ovs_vport_get_options(vport, skb);
1756 if (err == -EMSGSIZE)
1759 return genlmsg_end(skb, ovs_header);
1764 genlmsg_cancel(skb, ovs_header);
1768 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1770 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1773 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1774 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1777 struct sk_buff *skb;
1780 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1782 return ERR_PTR(-ENOMEM);
1784 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1790 /* Called with ovs_mutex or RCU read lock. */
1791 static struct vport *lookup_vport(struct net *net,
1792 struct ovs_header *ovs_header,
1793 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1795 struct datapath *dp;
1796 struct vport *vport;
1798 if (a[OVS_VPORT_ATTR_NAME]) {
1799 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1801 return ERR_PTR(-ENODEV);
1802 if (ovs_header->dp_ifindex &&
1803 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1804 return ERR_PTR(-ENODEV);
1806 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1807 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1809 if (port_no >= DP_MAX_PORTS)
1810 return ERR_PTR(-EFBIG);
1812 dp = get_dp(net, ovs_header->dp_ifindex);
1814 return ERR_PTR(-ENODEV);
1816 vport = ovs_vport_ovsl_rcu(dp, port_no);
1818 return ERR_PTR(-ENODEV);
1821 return ERR_PTR(-EINVAL);
1824 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1826 struct nlattr **a = info->attrs;
1827 struct ovs_header *ovs_header = info->userhdr;
1828 struct vport_parms parms;
1829 struct sk_buff *reply;
1830 struct vport *vport;
1831 struct datapath *dp;
1835 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1836 !a[OVS_VPORT_ATTR_UPCALL_PID])
1839 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1840 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1841 if (port_no >= DP_MAX_PORTS)
1844 reply = ovs_vport_cmd_alloc_info();
1849 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1852 goto exit_unlock_free;
1855 vport = ovs_vport_ovsl(dp, port_no);
1858 goto exit_unlock_free;
1860 for (port_no = 1; ; port_no++) {
1861 if (port_no >= DP_MAX_PORTS) {
1863 goto exit_unlock_free;
1865 vport = ovs_vport_ovsl(dp, port_no);
1871 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1872 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1873 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1875 parms.port_no = port_no;
1876 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1878 vport = new_vport(&parms);
1879 err = PTR_ERR(vport);
1881 goto exit_unlock_free;
1884 if (a[OVS_VPORT_ATTR_STATS])
1885 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1887 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1888 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1892 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1901 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1903 struct nlattr **a = info->attrs;
1904 struct sk_buff *reply;
1905 struct vport *vport;
1908 reply = ovs_vport_cmd_alloc_info();
1913 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1914 err = PTR_ERR(vport);
1916 goto exit_unlock_free;
1918 if (a[OVS_VPORT_ATTR_TYPE] &&
1919 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1921 goto exit_unlock_free;
1924 if (a[OVS_VPORT_ATTR_OPTIONS]) {
1925 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1927 goto exit_unlock_free;
1930 if (a[OVS_VPORT_ATTR_STATS])
1931 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1934 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1935 err = ovs_vport_set_upcall_portids(vport,
1936 a[OVS_VPORT_ATTR_UPCALL_PID]);
1938 goto exit_unlock_free;
1941 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1942 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1946 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1955 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1957 struct nlattr **a = info->attrs;
1958 struct sk_buff *reply;
1959 struct vport *vport;
1962 reply = ovs_vport_cmd_alloc_info();
1967 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1968 err = PTR_ERR(vport);
1970 goto exit_unlock_free;
1972 if (vport->port_no == OVSP_LOCAL) {
1974 goto exit_unlock_free;
1977 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1978 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1980 ovs_dp_detach_port(vport);
1983 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
1992 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1994 struct nlattr **a = info->attrs;
1995 struct ovs_header *ovs_header = info->userhdr;
1996 struct sk_buff *reply;
1997 struct vport *vport;
2000 reply = ovs_vport_cmd_alloc_info();
2005 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2006 err = PTR_ERR(vport);
2008 goto exit_unlock_free;
2009 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2010 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2014 return genlmsg_reply(reply, info);
2022 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2024 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2025 struct datapath *dp;
2026 int bucket = cb->args[0], skip = cb->args[1];
2030 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2035 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2036 struct vport *vport;
2039 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2041 ovs_vport_cmd_fill_info(vport, skb,
2042 NETLINK_CB(cb->skb).portid,
2045 OVS_VPORT_CMD_NEW) < 0)
2061 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2062 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2063 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2064 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2065 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2066 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2067 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2070 static struct genl_ops dp_vport_genl_ops[] = {
2071 { .cmd = OVS_VPORT_CMD_NEW,
2072 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2073 .policy = vport_policy,
2074 .doit = ovs_vport_cmd_new
2076 { .cmd = OVS_VPORT_CMD_DEL,
2077 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2078 .policy = vport_policy,
2079 .doit = ovs_vport_cmd_del
2081 { .cmd = OVS_VPORT_CMD_GET,
2082 .flags = 0, /* OK for unprivileged users. */
2083 .policy = vport_policy,
2084 .doit = ovs_vport_cmd_get,
2085 .dumpit = ovs_vport_cmd_dump
2087 { .cmd = OVS_VPORT_CMD_SET,
2088 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2089 .policy = vport_policy,
2090 .doit = ovs_vport_cmd_set,
2094 struct genl_family dp_vport_genl_family = {
2095 .id = GENL_ID_GENERATE,
2096 .hdrsize = sizeof(struct ovs_header),
2097 .name = OVS_VPORT_FAMILY,
2098 .version = OVS_VPORT_VERSION,
2099 .maxattr = OVS_VPORT_ATTR_MAX,
2101 .parallel_ops = true,
2102 .ops = dp_vport_genl_ops,
2103 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2104 .mcgrps = &ovs_dp_vport_multicast_group,
2108 static struct genl_family *dp_genl_families[] = {
2109 &dp_datapath_genl_family,
2110 &dp_vport_genl_family,
2111 &dp_flow_genl_family,
2112 &dp_packet_genl_family,
2115 static void dp_unregister_genl(int n_families)
2119 for (i = 0; i < n_families; i++)
2120 genl_unregister_family(dp_genl_families[i]);
2123 static int dp_register_genl(void)
2128 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2130 err = genl_register_family(dp_genl_families[i]);
2138 dp_unregister_genl(i);
2142 static int __net_init ovs_init_net(struct net *net)
2144 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2146 INIT_LIST_HEAD(&ovs_net->dps);
2147 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2151 static void __net_exit ovs_exit_net(struct net *net)
2153 struct datapath *dp, *dp_next;
2154 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2157 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2161 cancel_work_sync(&ovs_net->dp_notify_work);
2164 static struct pernet_operations ovs_net_ops = {
2165 .init = ovs_init_net,
2166 .exit = ovs_exit_net,
2168 .size = sizeof(struct ovs_net),
2171 DEFINE_COMPAT_PNET_REG_FUNC(device);
2173 static int __init dp_init(void)
2177 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2179 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2182 err = ovs_flow_init();
2186 err = ovs_vport_init();
2188 goto error_flow_exit;
2190 err = register_pernet_device(&ovs_net_ops);
2192 goto error_vport_exit;
2194 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2196 goto error_netns_exit;
2198 err = dp_register_genl();
2200 goto error_unreg_notifier;
2204 error_unreg_notifier:
2205 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2207 unregister_pernet_device(&ovs_net_ops);
2216 static void dp_cleanup(void)
2218 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2219 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2220 unregister_pernet_device(&ovs_net_ops);
2226 module_init(dp_init);
2227 module_exit(dp_cleanup);
2229 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2230 MODULE_LICENSE("GPL");
2231 MODULE_VERSION(VERSION);