2 * Copyright (c) 2007-2012 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <net/genetlink.h>
52 #include <net/net_namespace.h>
53 #include <net/netns/generic.h>
58 #include "genl_exec.h"
61 #include "vport-internal_dev.h"
63 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
64 LINUX_VERSION_CODE >= KERNEL_VERSION(3,7,0)
65 #error Kernels before 2.6.18 or after 3.6 are not supported by this version of Open vSwitch.
68 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
69 static void rehash_flow_table(struct work_struct *work);
70 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
72 int ovs_net_id __read_mostly;
74 int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
75 EXPORT_SYMBOL(ovs_dp_ioctl_hook);
80 * Writes to device state (add/remove datapath, port, set operations on vports,
81 * etc.) are protected by RTNL.
83 * Writes to other state (flow table modifications, set miscellaneous datapath
84 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
87 * Reads are protected by RCU.
89 * There are a few special cases (mostly stats) that have their own
90 * synchronization but they nest under all of above and don't interact with
94 static struct vport *new_vport(const struct vport_parms *);
95 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
96 const struct dp_upcall_info *);
97 static int queue_userspace_packet(struct net *, int dp_ifindex,
99 const struct dp_upcall_info *);
101 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
102 static struct datapath *get_dp(struct net *net, int dp_ifindex)
104 struct datapath *dp = NULL;
105 struct net_device *dev;
108 dev = dev_get_by_index_rcu(net, dp_ifindex);
110 struct vport *vport = ovs_internal_dev_get_vport(dev);
119 /* Must be called with rcu_read_lock or RTNL lock. */
120 const char *ovs_dp_name(const struct datapath *dp)
122 struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
123 return vport->ops->get_name(vport);
126 static int get_dpifindex(struct datapath *dp)
133 local = ovs_vport_rcu(dp, OVSP_LOCAL);
135 ifindex = local->ops->get_ifindex(local);
144 static size_t br_nlmsg_size(void)
146 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
147 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
148 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
149 + nla_total_size(4) /* IFLA_MASTER */
150 + nla_total_size(4) /* IFLA_MTU */
151 + nla_total_size(1); /* IFLA_OPERSTATE */
154 /* Caller must hold RTNL lock. */
155 static int dp_fill_ifinfo(struct sk_buff *skb,
156 const struct vport *port,
157 int event, unsigned int flags)
159 struct datapath *dp = port->dp;
160 struct ifinfomsg *hdr;
161 struct nlmsghdr *nlh;
163 if (!port->ops->get_ifindex)
166 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
170 hdr = nlmsg_data(nlh);
171 hdr->ifi_family = AF_BRIDGE;
173 hdr->ifi_type = ARPHRD_ETHER;
174 hdr->ifi_index = port->ops->get_ifindex(port);
175 hdr->ifi_flags = port->ops->get_dev_flags(port);
178 if (nla_put_string(skb, IFLA_IFNAME, port->ops->get_name(port)) ||
179 nla_put_u32(skb, IFLA_MASTER, get_dpifindex(dp)) ||
180 nla_put_u32(skb, IFLA_MTU, port->ops->get_mtu(port)) ||
181 #ifdef IFLA_OPERSTATE
182 nla_put_u8(skb, IFLA_OPERSTATE,
183 port->ops->is_running(port) ?
184 port->ops->get_operstate(port) :
187 nla_put(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port)))
188 goto nla_put_failure;
190 return nlmsg_end(skb, nlh);
193 nlmsg_cancel(skb, nlh);
197 /* Caller must hold RTNL lock. */
198 static void dp_ifinfo_notify(int event, struct vport *port)
203 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
209 err = dp_fill_ifinfo(skb, port, event, 0);
211 if (err == -ENODEV) {
214 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
215 WARN_ON(err == -EMSGSIZE);
220 rtnl_notify(skb, ovs_dp_get_net(port->dp), 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
224 rtnl_set_sk_err(ovs_dp_get_net(port->dp), RTNLGRP_LINK, err);
229 static void release_dp(struct kobject *kobj)
231 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
235 static struct kobj_type dp_ktype = {
236 .release = release_dp
239 static void destroy_dp_rcu(struct rcu_head *rcu)
241 struct datapath *dp = container_of(rcu, struct datapath, rcu);
243 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
244 free_percpu(dp->stats_percpu);
245 release_net(ovs_dp_get_net(dp));
247 kobject_put(&dp->ifobj);
250 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
253 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
256 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
259 struct hlist_node *n;
260 struct hlist_head *head;
262 head = vport_hash_bucket(dp, port_no);
263 hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
264 if (vport->port_no == port_no)
270 /* Called with RTNL lock and genl_lock. */
271 static struct vport *new_vport(const struct vport_parms *parms)
275 vport = ovs_vport_add(parms);
276 if (!IS_ERR(vport)) {
277 struct datapath *dp = parms->dp;
278 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
280 hlist_add_head_rcu(&vport->dp_hash_node, head);
281 dp_ifinfo_notify(RTM_NEWLINK, vport);
286 /* Called with RTNL lock. */
287 void ovs_dp_detach_port(struct vport *p)
291 if (p->port_no != OVSP_LOCAL)
292 ovs_dp_sysfs_del_if(p);
294 dp_ifinfo_notify(RTM_DELLINK, p);
296 /* First drop references to device. */
297 hlist_del_rcu(&p->dp_hash_node);
299 /* Then destroy it. */
303 /* Must be called with rcu_read_lock. */
304 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
306 struct datapath *dp = p->dp;
307 struct sw_flow *flow;
308 struct dp_stats_percpu *stats;
312 stats = this_cpu_ptr(dp->stats_percpu);
314 if (!OVS_CB(skb)->flow) {
315 struct sw_flow_key key;
318 /* Extract flow from 'skb' into 'key'. */
319 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
320 if (unlikely(error)) {
326 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
328 if (unlikely(!flow)) {
329 struct dp_upcall_info upcall;
331 upcall.cmd = OVS_PACKET_CMD_MISS;
333 upcall.userdata = NULL;
334 upcall.pid = p->upcall_pid;
335 ovs_dp_upcall(dp, skb, &upcall);
337 stats_counter = &stats->n_missed;
341 OVS_CB(skb)->flow = flow;
344 stats_counter = &stats->n_hit;
345 ovs_flow_used(OVS_CB(skb)->flow, skb);
346 ovs_execute_actions(dp, skb);
349 /* Update datapath statistics. */
350 u64_stats_update_begin(&stats->sync);
352 u64_stats_update_end(&stats->sync);
355 static struct genl_family dp_packet_genl_family = {
356 .id = GENL_ID_GENERATE,
357 .hdrsize = sizeof(struct ovs_header),
358 .name = OVS_PACKET_FAMILY,
359 .version = OVS_PACKET_VERSION,
360 .maxattr = OVS_PACKET_ATTR_MAX,
364 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
365 const struct dp_upcall_info *upcall_info)
367 struct dp_stats_percpu *stats;
371 if (upcall_info->pid == 0) {
376 dp_ifindex = get_dpifindex(dp);
382 forward_ip_summed(skb, true);
384 if (!skb_is_gso(skb))
385 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
387 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
394 stats = this_cpu_ptr(dp->stats_percpu);
396 u64_stats_update_begin(&stats->sync);
398 u64_stats_update_end(&stats->sync);
403 static int queue_gso_packets(struct net *net, int dp_ifindex,
405 const struct dp_upcall_info *upcall_info)
407 unsigned short gso_type = skb_shinfo(skb)->gso_type;
408 struct dp_upcall_info later_info;
409 struct sw_flow_key later_key;
410 struct sk_buff *segs, *nskb;
413 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
415 return PTR_ERR(segs);
417 /* Queue all of the segments. */
420 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
424 if (skb == segs && gso_type & SKB_GSO_UDP) {
425 /* The initial flow key extracted by ovs_flow_extract()
426 * in this case is for a first fragment, so we need to
427 * properly mark later fragments.
429 later_key = *upcall_info->key;
430 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
432 later_info = *upcall_info;
433 later_info.key = &later_key;
434 upcall_info = &later_info;
436 } while ((skb = skb->next));
438 /* Free all of the segments. */
446 } while ((skb = nskb));
450 static int queue_userspace_packet(struct net *net, int dp_ifindex,
452 const struct dp_upcall_info *upcall_info)
454 struct ovs_header *upcall;
455 struct sk_buff *nskb = NULL;
456 struct sk_buff *user_skb; /* to be queued to userspace */
461 if (vlan_tx_tag_present(skb)) {
462 nskb = skb_clone(skb, GFP_ATOMIC);
466 err = vlan_deaccel_tag(nskb);
473 if (nla_attr_size(skb->len) > USHRT_MAX) {
478 len = sizeof(struct ovs_header);
479 len += nla_total_size(skb->len);
480 len += nla_total_size(FLOW_BUFSIZE);
481 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
482 len += nla_total_size(8);
484 user_skb = genlmsg_new(len, GFP_ATOMIC);
490 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
491 0, upcall_info->cmd);
492 upcall->dp_ifindex = dp_ifindex;
494 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
495 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
496 nla_nest_end(user_skb, nla);
498 if (upcall_info->userdata)
499 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
500 nla_get_u64(upcall_info->userdata));
502 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
504 skb_copy_and_csum_dev(skb, nla_data(nla));
506 err = genlmsg_unicast(net, user_skb, upcall_info->pid);
513 /* Called with genl_mutex. */
514 static int flush_flows(struct datapath *dp)
516 struct flow_table *old_table;
517 struct flow_table *new_table;
519 old_table = genl_dereference(dp->table);
520 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
524 rcu_assign_pointer(dp->table, new_table);
526 ovs_flow_tbl_deferred_destroy(old_table);
530 static int validate_actions(const struct nlattr *attr,
531 const struct sw_flow_key *key, int depth);
533 static int validate_sample(const struct nlattr *attr,
534 const struct sw_flow_key *key, int depth)
536 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
537 const struct nlattr *probability, *actions;
538 const struct nlattr *a;
541 memset(attrs, 0, sizeof(attrs));
542 nla_for_each_nested(a, attr, rem) {
543 int type = nla_type(a);
544 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
551 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
552 if (!probability || nla_len(probability) != sizeof(u32))
555 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
556 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
558 return validate_actions(actions, key, depth + 1);
561 static int validate_tp_port(const struct sw_flow_key *flow_key)
563 if (flow_key->eth.type == htons(ETH_P_IP)) {
564 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
566 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
567 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
574 static int validate_set(const struct nlattr *a,
575 const struct sw_flow_key *flow_key)
577 const struct nlattr *ovs_key = nla_data(a);
578 int key_type = nla_type(ovs_key);
580 /* There can be only one key in a action */
581 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
584 if (key_type > OVS_KEY_ATTR_MAX ||
585 nla_len(ovs_key) != ovs_key_lens[key_type])
589 const struct ovs_key_ipv4 *ipv4_key;
590 const struct ovs_key_ipv4_tunnel *tun_key;
591 const struct ovs_key_ipv6 *ipv6_key;
593 case OVS_KEY_ATTR_PRIORITY:
594 case OVS_KEY_ATTR_TUN_ID:
595 case OVS_KEY_ATTR_ETHERNET:
598 case OVS_KEY_ATTR_IPV4_TUNNEL:
599 tun_key = nla_data(ovs_key);
600 if (!tun_key->ipv4_dst)
604 case OVS_KEY_ATTR_IPV4:
605 if (flow_key->eth.type != htons(ETH_P_IP))
608 if (!flow_key->ip.proto)
611 ipv4_key = nla_data(ovs_key);
612 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
615 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
620 case OVS_KEY_ATTR_IPV6:
621 if (flow_key->eth.type != htons(ETH_P_IPV6))
624 if (!flow_key->ip.proto)
627 ipv6_key = nla_data(ovs_key);
628 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
631 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
634 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
639 case OVS_KEY_ATTR_TCP:
640 if (flow_key->ip.proto != IPPROTO_TCP)
643 return validate_tp_port(flow_key);
645 case OVS_KEY_ATTR_UDP:
646 if (flow_key->ip.proto != IPPROTO_UDP)
649 return validate_tp_port(flow_key);
658 static int validate_userspace(const struct nlattr *attr)
660 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
661 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
662 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
664 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
667 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
668 attr, userspace_policy);
672 if (!a[OVS_USERSPACE_ATTR_PID] ||
673 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
679 static int validate_actions(const struct nlattr *attr,
680 const struct sw_flow_key *key, int depth)
682 const struct nlattr *a;
685 if (depth >= SAMPLE_ACTION_DEPTH)
688 nla_for_each_nested(a, attr, rem) {
689 /* Expected argument lengths, (u32)-1 for variable length. */
690 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
691 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
692 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
693 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
694 [OVS_ACTION_ATTR_POP_VLAN] = 0,
695 [OVS_ACTION_ATTR_SET] = (u32)-1,
696 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
698 const struct ovs_action_push_vlan *vlan;
699 int type = nla_type(a);
701 if (type > OVS_ACTION_ATTR_MAX ||
702 (action_lens[type] != nla_len(a) &&
703 action_lens[type] != (u32)-1))
707 case OVS_ACTION_ATTR_UNSPEC:
710 case OVS_ACTION_ATTR_USERSPACE:
711 err = validate_userspace(a);
716 case OVS_ACTION_ATTR_OUTPUT:
717 if (nla_get_u32(a) >= DP_MAX_PORTS)
722 case OVS_ACTION_ATTR_POP_VLAN:
725 case OVS_ACTION_ATTR_PUSH_VLAN:
727 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
729 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
733 case OVS_ACTION_ATTR_SET:
734 err = validate_set(a, key);
739 case OVS_ACTION_ATTR_SAMPLE:
740 err = validate_sample(a, key, depth);
756 static void clear_stats(struct sw_flow *flow)
760 flow->packet_count = 0;
761 flow->byte_count = 0;
764 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
766 struct ovs_header *ovs_header = info->userhdr;
767 struct nlattr **a = info->attrs;
768 struct sw_flow_actions *acts;
769 struct sk_buff *packet;
770 struct sw_flow *flow;
778 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
779 !a[OVS_PACKET_ATTR_ACTIONS] ||
780 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
783 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
784 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
788 skb_reserve(packet, NET_IP_ALIGN);
790 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
792 skb_reset_mac_header(packet);
793 eth = eth_hdr(packet);
795 /* Normally, setting the skb 'protocol' field would be handled by a
796 * call to eth_type_trans(), but it assumes there's a sending
797 * device, which we may not have. */
798 if (ntohs(eth->h_proto) >= 1536)
799 packet->protocol = eth->h_proto;
801 packet->protocol = htons(ETH_P_802_2);
803 /* Build an sw_flow for sending this packet. */
804 flow = ovs_flow_alloc();
809 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
813 err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
817 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
821 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
825 rcu_assign_pointer(flow->sf_acts, acts);
827 OVS_CB(packet)->flow = flow;
828 packet->priority = flow->key.phy.priority;
831 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
837 err = ovs_execute_actions(dp, packet);
854 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
855 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
856 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
857 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
860 static struct genl_ops dp_packet_genl_ops[] = {
861 { .cmd = OVS_PACKET_CMD_EXECUTE,
862 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
863 .policy = packet_policy,
864 .doit = ovs_packet_cmd_execute
868 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
871 struct flow_table *table = genl_dereference(dp->table);
873 stats->n_flows = ovs_flow_tbl_count(table);
875 stats->n_hit = stats->n_missed = stats->n_lost = 0;
876 for_each_possible_cpu(i) {
877 const struct dp_stats_percpu *percpu_stats;
878 struct dp_stats_percpu local_stats;
881 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
884 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
885 local_stats = *percpu_stats;
886 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
888 stats->n_hit += local_stats.n_hit;
889 stats->n_missed += local_stats.n_missed;
890 stats->n_lost += local_stats.n_lost;
894 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
895 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
896 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
897 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
900 static struct genl_family dp_flow_genl_family = {
901 .id = GENL_ID_GENERATE,
902 .hdrsize = sizeof(struct ovs_header),
903 .name = OVS_FLOW_FAMILY,
904 .version = OVS_FLOW_VERSION,
905 .maxattr = OVS_FLOW_ATTR_MAX,
909 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
910 .name = OVS_FLOW_MCGROUP
913 /* Called with genl_lock. */
914 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
915 struct sk_buff *skb, u32 pid,
916 u32 seq, u32 flags, u8 cmd)
918 const int skb_orig_len = skb->len;
919 const struct sw_flow_actions *sf_acts;
920 struct ovs_flow_stats stats;
921 struct ovs_header *ovs_header;
927 sf_acts = rcu_dereference_protected(flow->sf_acts,
928 lockdep_genl_is_held());
930 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
934 ovs_header->dp_ifindex = get_dpifindex(dp);
936 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
938 goto nla_put_failure;
939 err = ovs_flow_to_nlattrs(&flow->key, skb);
942 nla_nest_end(skb, nla);
944 spin_lock_bh(&flow->lock);
946 stats.n_packets = flow->packet_count;
947 stats.n_bytes = flow->byte_count;
948 tcp_flags = flow->tcp_flags;
949 spin_unlock_bh(&flow->lock);
952 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
953 goto nla_put_failure;
955 if (stats.n_packets &&
956 nla_put(skb, OVS_FLOW_ATTR_STATS,
957 sizeof(struct ovs_flow_stats), &stats))
958 goto nla_put_failure;
961 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
962 goto nla_put_failure;
964 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
965 * this is the first flow to be dumped into 'skb'. This is unusual for
966 * Netlink but individual action lists can be longer than
967 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
968 * The userspace caller can always fetch the actions separately if it
969 * really wants them. (Most userspace callers in fact don't care.)
971 * This can only fail for dump operations because the skb is always
972 * properly sized for single flows.
974 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
976 if (err < 0 && skb_orig_len)
979 return genlmsg_end(skb, ovs_header);
984 genlmsg_cancel(skb, ovs_header);
988 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
990 const struct sw_flow_actions *sf_acts;
993 sf_acts = rcu_dereference_protected(flow->sf_acts,
994 lockdep_genl_is_held());
996 /* OVS_FLOW_ATTR_KEY */
997 len = nla_total_size(FLOW_BUFSIZE);
998 /* OVS_FLOW_ATTR_ACTIONS */
999 len += nla_total_size(sf_acts->actions_len);
1000 /* OVS_FLOW_ATTR_STATS */
1001 len += nla_total_size(sizeof(struct ovs_flow_stats));
1002 /* OVS_FLOW_ATTR_TCP_FLAGS */
1003 len += nla_total_size(1);
1004 /* OVS_FLOW_ATTR_USED */
1005 len += nla_total_size(8);
1007 len += NLMSG_ALIGN(sizeof(struct ovs_header));
1009 return genlmsg_new(len, GFP_KERNEL);
1012 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1013 struct datapath *dp,
1014 u32 pid, u32 seq, u8 cmd)
1016 struct sk_buff *skb;
1019 skb = ovs_flow_cmd_alloc_info(flow);
1021 return ERR_PTR(-ENOMEM);
1023 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
1028 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1030 struct nlattr **a = info->attrs;
1031 struct ovs_header *ovs_header = info->userhdr;
1032 struct sw_flow_key key;
1033 struct sw_flow *flow;
1034 struct sk_buff *reply;
1035 struct datapath *dp;
1036 struct flow_table *table;
1042 if (!a[OVS_FLOW_ATTR_KEY])
1044 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1048 /* Validate actions. */
1049 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1050 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
1053 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1058 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1063 table = genl_dereference(dp->table);
1064 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1066 struct sw_flow_actions *acts;
1068 /* Bail out if we're not allowed to create a new flow. */
1070 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1073 /* Expand table, if necessary, to make room. */
1074 if (ovs_flow_tbl_need_to_expand(table)) {
1075 struct flow_table *new_table;
1077 new_table = ovs_flow_tbl_expand(table);
1078 if (!IS_ERR(new_table)) {
1079 rcu_assign_pointer(dp->table, new_table);
1080 ovs_flow_tbl_deferred_destroy(table);
1081 table = genl_dereference(dp->table);
1085 /* Allocate flow. */
1086 flow = ovs_flow_alloc();
1088 error = PTR_ERR(flow);
1093 /* Obtain actions. */
1094 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1095 error = PTR_ERR(acts);
1097 goto error_free_flow;
1098 rcu_assign_pointer(flow->sf_acts, acts);
1100 /* Put flow in bucket. */
1101 ovs_flow_tbl_insert(table, flow, &key, key_len);
1103 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1107 /* We found a matching flow. */
1108 struct sw_flow_actions *old_acts;
1109 struct nlattr *acts_attrs;
1111 /* Bail out if we're not allowed to modify an existing flow.
1112 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1113 * because Generic Netlink treats the latter as a dump
1114 * request. We also accept NLM_F_EXCL in case that bug ever
1118 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1119 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1122 /* Update actions. */
1123 old_acts = rcu_dereference_protected(flow->sf_acts,
1124 lockdep_genl_is_held());
1125 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1127 (old_acts->actions_len != nla_len(acts_attrs) ||
1128 memcmp(old_acts->actions, nla_data(acts_attrs),
1129 old_acts->actions_len))) {
1130 struct sw_flow_actions *new_acts;
1132 new_acts = ovs_flow_actions_alloc(acts_attrs);
1133 error = PTR_ERR(new_acts);
1134 if (IS_ERR(new_acts))
1137 rcu_assign_pointer(flow->sf_acts, new_acts);
1138 ovs_flow_deferred_free_acts(old_acts);
1141 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1142 info->snd_seq, OVS_FLOW_CMD_NEW);
1145 if (a[OVS_FLOW_ATTR_CLEAR]) {
1146 spin_lock_bh(&flow->lock);
1148 spin_unlock_bh(&flow->lock);
1153 genl_notify(reply, genl_info_net(info), info->snd_pid,
1154 ovs_dp_flow_multicast_group.id, info->nlhdr,
1157 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1158 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1162 ovs_flow_free(flow);
1167 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1169 struct nlattr **a = info->attrs;
1170 struct ovs_header *ovs_header = info->userhdr;
1171 struct sw_flow_key key;
1172 struct sk_buff *reply;
1173 struct sw_flow *flow;
1174 struct datapath *dp;
1175 struct flow_table *table;
1179 if (!a[OVS_FLOW_ATTR_KEY])
1181 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1185 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1189 table = genl_dereference(dp->table);
1190 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1194 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1195 info->snd_seq, OVS_FLOW_CMD_NEW);
1197 return PTR_ERR(reply);
1199 return genlmsg_reply(reply, info);
1202 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1204 struct nlattr **a = info->attrs;
1205 struct ovs_header *ovs_header = info->userhdr;
1206 struct sw_flow_key key;
1207 struct sk_buff *reply;
1208 struct sw_flow *flow;
1209 struct datapath *dp;
1210 struct flow_table *table;
1214 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1218 if (!a[OVS_FLOW_ATTR_KEY])
1219 return flush_flows(dp);
1221 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1225 table = genl_dereference(dp->table);
1226 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1230 reply = ovs_flow_cmd_alloc_info(flow);
1234 ovs_flow_tbl_remove(table, flow);
1236 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1237 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1240 ovs_flow_deferred_free(flow);
1242 genl_notify(reply, genl_info_net(info), info->snd_pid,
1243 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1247 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1249 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1250 struct datapath *dp;
1251 struct flow_table *table;
1253 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1257 table = genl_dereference(dp->table);
1260 struct sw_flow *flow;
1263 bucket = cb->args[0];
1265 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1269 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1270 NETLINK_CB(cb->skb).pid,
1271 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1272 OVS_FLOW_CMD_NEW) < 0)
1275 cb->args[0] = bucket;
1281 static struct genl_ops dp_flow_genl_ops[] = {
1282 { .cmd = OVS_FLOW_CMD_NEW,
1283 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1284 .policy = flow_policy,
1285 .doit = ovs_flow_cmd_new_or_set
1287 { .cmd = OVS_FLOW_CMD_DEL,
1288 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1289 .policy = flow_policy,
1290 .doit = ovs_flow_cmd_del
1292 { .cmd = OVS_FLOW_CMD_GET,
1293 .flags = 0, /* OK for unprivileged users. */
1294 .policy = flow_policy,
1295 .doit = ovs_flow_cmd_get,
1296 .dumpit = ovs_flow_cmd_dump
1298 { .cmd = OVS_FLOW_CMD_SET,
1299 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1300 .policy = flow_policy,
1301 .doit = ovs_flow_cmd_new_or_set,
1305 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1306 #ifdef HAVE_NLA_NUL_STRING
1307 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1309 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1312 static struct genl_family dp_datapath_genl_family = {
1313 .id = GENL_ID_GENERATE,
1314 .hdrsize = sizeof(struct ovs_header),
1315 .name = OVS_DATAPATH_FAMILY,
1316 .version = OVS_DATAPATH_VERSION,
1317 .maxattr = OVS_DP_ATTR_MAX,
1321 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1322 .name = OVS_DATAPATH_MCGROUP
1325 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1326 u32 pid, u32 seq, u32 flags, u8 cmd)
1328 struct ovs_header *ovs_header;
1329 struct ovs_dp_stats dp_stats;
1332 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1337 ovs_header->dp_ifindex = get_dpifindex(dp);
1340 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1343 goto nla_put_failure;
1345 get_dp_stats(dp, &dp_stats);
1346 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1347 goto nla_put_failure;
1349 return genlmsg_end(skb, ovs_header);
1352 genlmsg_cancel(skb, ovs_header);
1357 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1360 struct sk_buff *skb;
1363 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1365 return ERR_PTR(-ENOMEM);
1367 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1370 return ERR_PTR(retval);
1375 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1377 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1380 /* Called with genl_mutex and optionally with RTNL lock also. */
1381 static struct datapath *lookup_datapath(struct net *net,
1382 struct ovs_header *ovs_header,
1383 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1385 struct datapath *dp;
1387 if (!a[OVS_DP_ATTR_NAME])
1388 dp = get_dp(net, ovs_header->dp_ifindex);
1390 struct vport *vport;
1393 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1394 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1397 return dp ? dp : ERR_PTR(-ENODEV);
1400 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1402 struct nlattr **a = info->attrs;
1403 struct vport_parms parms;
1404 struct sk_buff *reply;
1405 struct datapath *dp;
1406 struct vport *vport;
1407 struct ovs_net *ovs_net;
1411 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1414 err = ovs_dp_cmd_validate(a);
1421 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1423 goto err_unlock_rtnl;
1425 /* Initialize kobject for bridge. This will be added as
1426 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1427 dp->ifobj.kset = NULL;
1428 kobject_init(&dp->ifobj, &dp_ktype);
1430 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1432 /* Allocate table. */
1434 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1438 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1439 if (!dp->stats_percpu) {
1441 goto err_destroy_table;
1444 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1448 goto err_destroy_percpu;
1451 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1452 INIT_HLIST_HEAD(&dp->ports[i]);
1454 /* Set up our datapath device. */
1455 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1456 parms.type = OVS_VPORT_TYPE_INTERNAL;
1457 parms.options = NULL;
1459 parms.port_no = OVSP_LOCAL;
1460 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1462 vport = new_vport(&parms);
1463 if (IS_ERR(vport)) {
1464 err = PTR_ERR(vport);
1468 goto err_destroy_ports_array;
1471 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1472 info->snd_seq, OVS_DP_CMD_NEW);
1473 err = PTR_ERR(reply);
1475 goto err_destroy_local_port;
1477 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1478 list_add_tail(&dp->list_node, &ovs_net->dps);
1479 ovs_dp_sysfs_add_dp(dp);
1483 genl_notify(reply, genl_info_net(info), info->snd_pid,
1484 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1488 err_destroy_local_port:
1489 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1490 err_destroy_ports_array:
1493 free_percpu(dp->stats_percpu);
1495 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1497 release_net(ovs_dp_get_net(dp));
1505 /* Called with genl_mutex. */
1506 static void __dp_destroy(struct datapath *dp)
1512 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1513 struct vport *vport;
1514 struct hlist_node *node, *n;
1516 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1517 if (vport->port_no != OVSP_LOCAL)
1518 ovs_dp_detach_port(vport);
1521 ovs_dp_sysfs_del_dp(dp);
1522 list_del(&dp->list_node);
1523 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1525 /* rtnl_unlock() will wait until all the references to devices that
1526 * are pending unregistration have been dropped. We do it here to
1527 * ensure that any internal devices (which contain DP pointers) are
1528 * fully destroyed before freeing the datapath.
1532 call_rcu(&dp->rcu, destroy_dp_rcu);
1535 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1537 struct sk_buff *reply;
1538 struct datapath *dp;
1541 err = ovs_dp_cmd_validate(info->attrs);
1545 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1550 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1551 info->snd_seq, OVS_DP_CMD_DEL);
1552 err = PTR_ERR(reply);
1558 genl_notify(reply, genl_info_net(info), info->snd_pid,
1559 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1565 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1567 struct sk_buff *reply;
1568 struct datapath *dp;
1571 err = ovs_dp_cmd_validate(info->attrs);
1575 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1579 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1580 info->snd_seq, OVS_DP_CMD_NEW);
1581 if (IS_ERR(reply)) {
1582 err = PTR_ERR(reply);
1583 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1584 ovs_dp_datapath_multicast_group.id, err);
1588 genl_notify(reply, genl_info_net(info), info->snd_pid,
1589 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1595 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1597 struct sk_buff *reply;
1598 struct datapath *dp;
1601 err = ovs_dp_cmd_validate(info->attrs);
1605 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1609 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1610 info->snd_seq, OVS_DP_CMD_NEW);
1612 return PTR_ERR(reply);
1614 return genlmsg_reply(reply, info);
1617 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1619 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1620 struct datapath *dp;
1621 int skip = cb->args[0];
1624 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1626 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1627 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1628 OVS_DP_CMD_NEW) < 0)
1638 static struct genl_ops dp_datapath_genl_ops[] = {
1639 { .cmd = OVS_DP_CMD_NEW,
1640 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1641 .policy = datapath_policy,
1642 .doit = ovs_dp_cmd_new
1644 { .cmd = OVS_DP_CMD_DEL,
1645 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1646 .policy = datapath_policy,
1647 .doit = ovs_dp_cmd_del
1649 { .cmd = OVS_DP_CMD_GET,
1650 .flags = 0, /* OK for unprivileged users. */
1651 .policy = datapath_policy,
1652 .doit = ovs_dp_cmd_get,
1653 .dumpit = ovs_dp_cmd_dump
1655 { .cmd = OVS_DP_CMD_SET,
1656 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1657 .policy = datapath_policy,
1658 .doit = ovs_dp_cmd_set,
1662 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1663 #ifdef HAVE_NLA_NUL_STRING
1664 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1665 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1666 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1668 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1669 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1671 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1672 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1673 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1674 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1677 static struct genl_family dp_vport_genl_family = {
1678 .id = GENL_ID_GENERATE,
1679 .hdrsize = sizeof(struct ovs_header),
1680 .name = OVS_VPORT_FAMILY,
1681 .version = OVS_VPORT_VERSION,
1682 .maxattr = OVS_VPORT_ATTR_MAX,
1686 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1687 .name = OVS_VPORT_MCGROUP
1690 /* Called with RTNL lock or RCU read lock. */
1691 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1692 u32 pid, u32 seq, u32 flags, u8 cmd)
1694 struct ovs_header *ovs_header;
1695 struct ovs_vport_stats vport_stats;
1698 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1703 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1705 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1706 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1707 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1708 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1709 goto nla_put_failure;
1711 ovs_vport_get_stats(vport, &vport_stats);
1712 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1714 goto nla_put_failure;
1716 if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1717 vport->ops->get_addr(vport)))
1718 goto nla_put_failure;
1720 err = ovs_vport_get_options(vport, skb);
1721 if (err == -EMSGSIZE)
1724 return genlmsg_end(skb, ovs_header);
1729 genlmsg_cancel(skb, ovs_header);
1733 /* Called with RTNL lock or RCU read lock. */
1734 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1737 struct sk_buff *skb;
1740 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1742 return ERR_PTR(-ENOMEM);
1744 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1747 return ERR_PTR(retval);
1752 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1754 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1757 /* Called with RTNL lock or RCU read lock. */
1758 static struct vport *lookup_vport(struct net *net,
1759 struct ovs_header *ovs_header,
1760 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1762 struct datapath *dp;
1763 struct vport *vport;
1765 if (a[OVS_VPORT_ATTR_NAME]) {
1766 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1768 return ERR_PTR(-ENODEV);
1769 if (ovs_header->dp_ifindex &&
1770 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1771 return ERR_PTR(-ENODEV);
1773 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1774 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1776 if (port_no >= DP_MAX_PORTS)
1777 return ERR_PTR(-EFBIG);
1779 dp = get_dp(net, ovs_header->dp_ifindex);
1781 return ERR_PTR(-ENODEV);
1783 vport = ovs_vport_rtnl_rcu(dp, port_no);
1785 return ERR_PTR(-ENOENT);
1788 return ERR_PTR(-EINVAL);
1791 /* Called with RTNL lock. */
1792 static int change_vport(struct vport *vport,
1793 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1797 if (a[OVS_VPORT_ATTR_STATS])
1798 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1800 if (a[OVS_VPORT_ATTR_ADDRESS])
1801 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1806 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1808 struct nlattr **a = info->attrs;
1809 struct ovs_header *ovs_header = info->userhdr;
1810 struct vport_parms parms;
1811 struct sk_buff *reply;
1812 struct vport *vport;
1813 struct datapath *dp;
1818 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1819 !a[OVS_VPORT_ATTR_UPCALL_PID])
1822 err = ovs_vport_cmd_validate(a);
1827 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1832 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1833 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1836 if (port_no >= DP_MAX_PORTS)
1839 vport = ovs_vport_rtnl(dp, port_no);
1844 for (port_no = 1; ; port_no++) {
1845 if (port_no >= DP_MAX_PORTS) {
1849 vport = ovs_vport_rtnl(dp, port_no);
1855 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1856 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1857 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1859 parms.port_no = port_no;
1860 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1862 vport = new_vport(&parms);
1863 err = PTR_ERR(vport);
1867 ovs_dp_sysfs_add_if(vport);
1869 err = change_vport(vport, a);
1871 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1875 err = PTR_ERR(reply);
1878 ovs_dp_detach_port(vport);
1881 genl_notify(reply, genl_info_net(info), info->snd_pid,
1882 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1890 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1892 struct nlattr **a = info->attrs;
1893 struct sk_buff *reply;
1894 struct vport *vport;
1897 err = ovs_vport_cmd_validate(a);
1902 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1903 err = PTR_ERR(vport);
1908 if (a[OVS_VPORT_ATTR_TYPE] &&
1909 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1912 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1913 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1915 err = change_vport(vport, a);
1918 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1919 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1921 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1923 if (IS_ERR(reply)) {
1924 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1925 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1929 genl_notify(reply, genl_info_net(info), info->snd_pid,
1930 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1938 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1940 struct nlattr **a = info->attrs;
1941 struct sk_buff *reply;
1942 struct vport *vport;
1945 err = ovs_vport_cmd_validate(a);
1950 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1951 err = PTR_ERR(vport);
1955 if (vport->port_no == OVSP_LOCAL) {
1960 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1962 err = PTR_ERR(reply);
1966 ovs_dp_detach_port(vport);
1968 genl_notify(reply, genl_info_net(info), info->snd_pid,
1969 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1977 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1979 struct nlattr **a = info->attrs;
1980 struct ovs_header *ovs_header = info->userhdr;
1981 struct sk_buff *reply;
1982 struct vport *vport;
1985 err = ovs_vport_cmd_validate(a);
1990 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1991 err = PTR_ERR(vport);
1995 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1997 err = PTR_ERR(reply);
2003 return genlmsg_reply(reply, info);
2011 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2013 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2014 struct datapath *dp;
2015 int bucket = cb->args[0], skip = cb->args[1];
2018 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2023 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2024 struct vport *vport;
2025 struct hlist_node *n;
2028 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2030 ovs_vport_cmd_fill_info(vport, skb,
2031 NETLINK_CB(cb->skb).pid,
2034 OVS_VPORT_CMD_NEW) < 0)
2050 static struct genl_ops dp_vport_genl_ops[] = {
2051 { .cmd = OVS_VPORT_CMD_NEW,
2052 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2053 .policy = vport_policy,
2054 .doit = ovs_vport_cmd_new
2056 { .cmd = OVS_VPORT_CMD_DEL,
2057 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2058 .policy = vport_policy,
2059 .doit = ovs_vport_cmd_del
2061 { .cmd = OVS_VPORT_CMD_GET,
2062 .flags = 0, /* OK for unprivileged users. */
2063 .policy = vport_policy,
2064 .doit = ovs_vport_cmd_get,
2065 .dumpit = ovs_vport_cmd_dump
2067 { .cmd = OVS_VPORT_CMD_SET,
2068 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2069 .policy = vport_policy,
2070 .doit = ovs_vport_cmd_set,
2074 struct genl_family_and_ops {
2075 struct genl_family *family;
2076 struct genl_ops *ops;
2078 struct genl_multicast_group *group;
2081 static const struct genl_family_and_ops dp_genl_families[] = {
2082 { &dp_datapath_genl_family,
2083 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2084 &ovs_dp_datapath_multicast_group },
2085 { &dp_vport_genl_family,
2086 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2087 &ovs_dp_vport_multicast_group },
2088 { &dp_flow_genl_family,
2089 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2090 &ovs_dp_flow_multicast_group },
2091 { &dp_packet_genl_family,
2092 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2096 static void dp_unregister_genl(int n_families)
2100 for (i = 0; i < n_families; i++)
2101 genl_unregister_family(dp_genl_families[i].family);
2104 static int dp_register_genl(void)
2111 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2112 const struct genl_family_and_ops *f = &dp_genl_families[i];
2114 err = genl_register_family_with_ops(f->family, f->ops,
2121 err = genl_register_mc_group(f->family, f->group);
2130 dp_unregister_genl(n_registered);
2134 static int __rehash_flow_table(void *dummy)
2136 struct datapath *dp;
2141 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2143 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2144 struct flow_table *old_table = genl_dereference(dp->table);
2145 struct flow_table *new_table;
2147 new_table = ovs_flow_tbl_rehash(old_table);
2148 if (!IS_ERR(new_table)) {
2149 rcu_assign_pointer(dp->table, new_table);
2150 ovs_flow_tbl_deferred_destroy(old_table);
2158 static void rehash_flow_table(struct work_struct *work)
2160 genl_exec(__rehash_flow_table, NULL);
2161 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2164 static int dp_destroy_all(void *data)
2166 struct datapath *dp, *dp_next;
2167 struct ovs_net *ovs_net = data;
2169 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2175 static int __net_init ovs_init_net(struct net *net)
2177 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2179 INIT_LIST_HEAD(&ovs_net->dps);
2183 static void __net_exit ovs_exit_net(struct net *net)
2185 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2187 genl_exec(dp_destroy_all, ovs_net);
2190 static struct pernet_operations ovs_net_ops = {
2191 .init = ovs_init_net,
2192 .exit = ovs_exit_net,
2194 .size = sizeof(struct ovs_net),
2197 static int __init dp_init(void)
2199 struct sk_buff *dummy_skb;
2202 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2204 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2207 err = genl_exec_init();
2211 err = ovs_workqueues_init();
2213 goto error_genl_exec;
2215 err = ovs_tnl_init();
2219 err = ovs_flow_init();
2221 goto error_tnl_exit;
2223 err = ovs_vport_init();
2225 goto error_flow_exit;
2227 err = register_pernet_device(&ovs_net_ops);
2229 goto error_vport_exit;
2231 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2233 goto error_netns_exit;
2235 err = dp_register_genl();
2237 goto error_unreg_notifier;
2239 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2243 error_unreg_notifier:
2244 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2246 unregister_pernet_device(&ovs_net_ops);
2254 ovs_workqueues_exit();
2261 static void dp_cleanup(void)
2263 cancel_delayed_work_sync(&rehash_flow_wq);
2264 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2265 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2266 unregister_pernet_device(&ovs_net_ops);
2271 ovs_workqueues_exit();
2275 module_init(dp_init);
2276 module_exit(dp_cleanup);
2278 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2279 MODULE_LICENSE("GPL");
2280 MODULE_VERSION(VERSION);