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(3,4,0)
65 #error Kernels before 2.6.18 or after 3.3 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 = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
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 = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
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 struct dp_upcall_info later_info;
408 struct sw_flow_key later_key;
409 struct sk_buff *segs, *nskb;
412 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
414 return PTR_ERR(segs);
416 /* Queue all of the segments. */
419 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
423 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
424 /* The initial flow key extracted by ovs_flow_extract()
425 * in this case is for a first fragment, so we need to
426 * properly mark later fragments.
428 later_key = *upcall_info->key;
429 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
431 later_info = *upcall_info;
432 later_info.key = &later_key;
433 upcall_info = &later_info;
435 } while ((skb = skb->next));
437 /* Free all of the segments. */
445 } while ((skb = nskb));
449 static int queue_userspace_packet(struct net *net, int dp_ifindex,
451 const struct dp_upcall_info *upcall_info)
453 struct ovs_header *upcall;
454 struct sk_buff *nskb = NULL;
455 struct sk_buff *user_skb; /* to be queued to userspace */
460 if (vlan_tx_tag_present(skb)) {
461 nskb = skb_clone(skb, GFP_ATOMIC);
465 err = vlan_deaccel_tag(nskb);
472 if (nla_attr_size(skb->len) > USHRT_MAX) {
477 len = sizeof(struct ovs_header);
478 len += nla_total_size(skb->len);
479 len += nla_total_size(FLOW_BUFSIZE);
480 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
481 len += nla_total_size(8);
483 user_skb = genlmsg_new(len, GFP_ATOMIC);
489 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
490 0, upcall_info->cmd);
491 upcall->dp_ifindex = dp_ifindex;
493 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
494 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
495 nla_nest_end(user_skb, nla);
497 if (upcall_info->userdata)
498 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
499 nla_get_u64(upcall_info->userdata));
501 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
503 skb_copy_and_csum_dev(skb, nla_data(nla));
505 err = genlmsg_unicast(net, user_skb, upcall_info->pid);
512 /* Called with genl_mutex. */
513 static int flush_flows(struct datapath *dp)
515 struct flow_table *old_table;
516 struct flow_table *new_table;
518 old_table = genl_dereference(dp->table);
519 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
523 rcu_assign_pointer(dp->table, new_table);
525 ovs_flow_tbl_deferred_destroy(old_table);
529 static int validate_actions(const struct nlattr *attr,
530 const struct sw_flow_key *key, int depth);
532 static int validate_sample(const struct nlattr *attr,
533 const struct sw_flow_key *key, int depth)
535 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
536 const struct nlattr *probability, *actions;
537 const struct nlattr *a;
540 memset(attrs, 0, sizeof(attrs));
541 nla_for_each_nested(a, attr, rem) {
542 int type = nla_type(a);
543 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
550 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
551 if (!probability || nla_len(probability) != sizeof(u32))
554 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
555 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
557 return validate_actions(actions, key, depth + 1);
560 static int validate_tp_port(const struct sw_flow_key *flow_key)
562 if (flow_key->eth.type == htons(ETH_P_IP)) {
563 if (flow_key->ipv4.tp.src && flow_key->ipv4.tp.dst)
565 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
566 if (flow_key->ipv6.tp.src && flow_key->ipv6.tp.dst)
573 static int validate_set(const struct nlattr *a,
574 const struct sw_flow_key *flow_key)
576 const struct nlattr *ovs_key = nla_data(a);
577 int key_type = nla_type(ovs_key);
579 /* There can be only one key in a action */
580 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
583 if (key_type > OVS_KEY_ATTR_MAX ||
584 nla_len(ovs_key) != ovs_key_lens[key_type])
588 const struct ovs_key_ipv4 *ipv4_key;
590 case OVS_KEY_ATTR_PRIORITY:
591 case OVS_KEY_ATTR_TUN_ID:
592 case OVS_KEY_ATTR_ETHERNET:
595 case OVS_KEY_ATTR_IPV4:
596 if (flow_key->eth.type != htons(ETH_P_IP))
599 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
602 ipv4_key = nla_data(ovs_key);
603 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
606 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
611 case OVS_KEY_ATTR_TCP:
612 if (flow_key->ip.proto != IPPROTO_TCP)
615 return validate_tp_port(flow_key);
617 case OVS_KEY_ATTR_UDP:
618 if (flow_key->ip.proto != IPPROTO_UDP)
621 return validate_tp_port(flow_key);
630 static int validate_userspace(const struct nlattr *attr)
632 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
633 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
634 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
636 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
639 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
640 attr, userspace_policy);
644 if (!a[OVS_USERSPACE_ATTR_PID] ||
645 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
651 static int validate_actions(const struct nlattr *attr,
652 const struct sw_flow_key *key, int depth)
654 const struct nlattr *a;
657 if (depth >= SAMPLE_ACTION_DEPTH)
660 nla_for_each_nested(a, attr, rem) {
661 /* Expected argument lengths, (u32)-1 for variable length. */
662 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
663 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
664 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
665 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
666 [OVS_ACTION_ATTR_POP_VLAN] = 0,
667 [OVS_ACTION_ATTR_SET] = (u32)-1,
668 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
670 const struct ovs_action_push_vlan *vlan;
671 int type = nla_type(a);
673 if (type > OVS_ACTION_ATTR_MAX ||
674 (action_lens[type] != nla_len(a) &&
675 action_lens[type] != (u32)-1))
679 case OVS_ACTION_ATTR_UNSPEC:
682 case OVS_ACTION_ATTR_USERSPACE:
683 err = validate_userspace(a);
688 case OVS_ACTION_ATTR_OUTPUT:
689 if (nla_get_u32(a) >= DP_MAX_PORTS)
694 case OVS_ACTION_ATTR_POP_VLAN:
697 case OVS_ACTION_ATTR_PUSH_VLAN:
699 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
701 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
705 case OVS_ACTION_ATTR_SET:
706 err = validate_set(a, key);
711 case OVS_ACTION_ATTR_SAMPLE:
712 err = validate_sample(a, key, depth);
728 static void clear_stats(struct sw_flow *flow)
732 flow->packet_count = 0;
733 flow->byte_count = 0;
736 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
738 struct ovs_header *ovs_header = info->userhdr;
739 struct nlattr **a = info->attrs;
740 struct sw_flow_actions *acts;
741 struct sk_buff *packet;
742 struct sw_flow *flow;
750 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
751 !a[OVS_PACKET_ATTR_ACTIONS] ||
752 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
755 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
756 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
760 skb_reserve(packet, NET_IP_ALIGN);
762 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
764 skb_reset_mac_header(packet);
765 eth = eth_hdr(packet);
767 /* Normally, setting the skb 'protocol' field would be handled by a
768 * call to eth_type_trans(), but it assumes there's a sending
769 * device, which we may not have. */
770 if (ntohs(eth->h_proto) >= 1536)
771 packet->protocol = eth->h_proto;
773 packet->protocol = htons(ETH_P_802_2);
775 /* Build an sw_flow for sending this packet. */
776 flow = ovs_flow_alloc();
781 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
785 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
786 &flow->key.phy.in_port,
787 &flow->key.phy.tun_id,
788 a[OVS_PACKET_ATTR_KEY]);
792 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
796 flow->hash = ovs_flow_hash(&flow->key, key_len);
798 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
802 rcu_assign_pointer(flow->sf_acts, acts);
804 OVS_CB(packet)->flow = flow;
805 packet->priority = flow->key.phy.priority;
808 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
814 err = ovs_execute_actions(dp, packet);
831 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
832 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
833 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
834 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
837 static struct genl_ops dp_packet_genl_ops[] = {
838 { .cmd = OVS_PACKET_CMD_EXECUTE,
839 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
840 .policy = packet_policy,
841 .doit = ovs_packet_cmd_execute
845 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
848 struct flow_table *table = genl_dereference(dp->table);
850 stats->n_flows = ovs_flow_tbl_count(table);
852 stats->n_hit = stats->n_missed = stats->n_lost = 0;
853 for_each_possible_cpu(i) {
854 const struct dp_stats_percpu *percpu_stats;
855 struct dp_stats_percpu local_stats;
858 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
861 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
862 local_stats = *percpu_stats;
863 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
865 stats->n_hit += local_stats.n_hit;
866 stats->n_missed += local_stats.n_missed;
867 stats->n_lost += local_stats.n_lost;
871 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
872 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
873 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
874 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
877 static struct genl_family dp_flow_genl_family = {
878 .id = GENL_ID_GENERATE,
879 .hdrsize = sizeof(struct ovs_header),
880 .name = OVS_FLOW_FAMILY,
881 .version = OVS_FLOW_VERSION,
882 .maxattr = OVS_FLOW_ATTR_MAX,
886 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
887 .name = OVS_FLOW_MCGROUP
890 /* Called with genl_lock. */
891 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
892 struct sk_buff *skb, u32 pid,
893 u32 seq, u32 flags, u8 cmd)
895 const int skb_orig_len = skb->len;
896 const struct sw_flow_actions *sf_acts;
897 struct ovs_flow_stats stats;
898 struct ovs_header *ovs_header;
904 sf_acts = rcu_dereference_protected(flow->sf_acts,
905 lockdep_genl_is_held());
907 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
911 ovs_header->dp_ifindex = get_dpifindex(dp);
913 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
915 goto nla_put_failure;
916 err = ovs_flow_to_nlattrs(&flow->key, skb);
919 nla_nest_end(skb, nla);
921 spin_lock_bh(&flow->lock);
923 stats.n_packets = flow->packet_count;
924 stats.n_bytes = flow->byte_count;
925 tcp_flags = flow->tcp_flags;
926 spin_unlock_bh(&flow->lock);
929 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
930 goto nla_put_failure;
932 if (stats.n_packets &&
933 nla_put(skb, OVS_FLOW_ATTR_STATS,
934 sizeof(struct ovs_flow_stats), &stats))
935 goto nla_put_failure;
938 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
939 goto nla_put_failure;
941 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
942 * this is the first flow to be dumped into 'skb'. This is unusual for
943 * Netlink but individual action lists can be longer than
944 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
945 * The userspace caller can always fetch the actions separately if it
946 * really wants them. (Most userspace callers in fact don't care.)
948 * This can only fail for dump operations because the skb is always
949 * properly sized for single flows.
951 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
953 if (err < 0 && skb_orig_len)
956 return genlmsg_end(skb, ovs_header);
961 genlmsg_cancel(skb, ovs_header);
965 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
967 const struct sw_flow_actions *sf_acts;
970 sf_acts = rcu_dereference_protected(flow->sf_acts,
971 lockdep_genl_is_held());
973 /* OVS_FLOW_ATTR_KEY */
974 len = nla_total_size(FLOW_BUFSIZE);
975 /* OVS_FLOW_ATTR_ACTIONS */
976 len += nla_total_size(sf_acts->actions_len);
977 /* OVS_FLOW_ATTR_STATS */
978 len += nla_total_size(sizeof(struct ovs_flow_stats));
979 /* OVS_FLOW_ATTR_TCP_FLAGS */
980 len += nla_total_size(1);
981 /* OVS_FLOW_ATTR_USED */
982 len += nla_total_size(8);
984 len += NLMSG_ALIGN(sizeof(struct ovs_header));
986 return genlmsg_new(len, GFP_KERNEL);
989 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
991 u32 pid, u32 seq, u8 cmd)
996 skb = ovs_flow_cmd_alloc_info(flow);
998 return ERR_PTR(-ENOMEM);
1000 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
1005 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1007 struct nlattr **a = info->attrs;
1008 struct ovs_header *ovs_header = info->userhdr;
1009 struct sw_flow_key key;
1010 struct sw_flow *flow;
1011 struct sk_buff *reply;
1012 struct datapath *dp;
1013 struct flow_table *table;
1019 if (!a[OVS_FLOW_ATTR_KEY])
1021 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1025 /* Validate actions. */
1026 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1027 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
1030 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1035 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1040 table = genl_dereference(dp->table);
1041 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1043 struct sw_flow_actions *acts;
1045 /* Bail out if we're not allowed to create a new flow. */
1047 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1050 /* Expand table, if necessary, to make room. */
1051 if (ovs_flow_tbl_need_to_expand(table)) {
1052 struct flow_table *new_table;
1054 new_table = ovs_flow_tbl_expand(table);
1055 if (!IS_ERR(new_table)) {
1056 rcu_assign_pointer(dp->table, new_table);
1057 ovs_flow_tbl_deferred_destroy(table);
1058 table = genl_dereference(dp->table);
1062 /* Allocate flow. */
1063 flow = ovs_flow_alloc();
1065 error = PTR_ERR(flow);
1071 /* Obtain actions. */
1072 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1073 error = PTR_ERR(acts);
1075 goto error_free_flow;
1076 rcu_assign_pointer(flow->sf_acts, acts);
1078 /* Put flow in bucket. */
1079 flow->hash = ovs_flow_hash(&key, key_len);
1080 ovs_flow_tbl_insert(table, flow);
1082 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1086 /* We found a matching flow. */
1087 struct sw_flow_actions *old_acts;
1088 struct nlattr *acts_attrs;
1090 /* Bail out if we're not allowed to modify an existing flow.
1091 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1092 * because Generic Netlink treats the latter as a dump
1093 * request. We also accept NLM_F_EXCL in case that bug ever
1097 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1098 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1101 /* Update actions. */
1102 old_acts = rcu_dereference_protected(flow->sf_acts,
1103 lockdep_genl_is_held());
1104 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1106 (old_acts->actions_len != nla_len(acts_attrs) ||
1107 memcmp(old_acts->actions, nla_data(acts_attrs),
1108 old_acts->actions_len))) {
1109 struct sw_flow_actions *new_acts;
1111 new_acts = ovs_flow_actions_alloc(acts_attrs);
1112 error = PTR_ERR(new_acts);
1113 if (IS_ERR(new_acts))
1116 rcu_assign_pointer(flow->sf_acts, new_acts);
1117 ovs_flow_deferred_free_acts(old_acts);
1120 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1121 info->snd_seq, OVS_FLOW_CMD_NEW);
1124 if (a[OVS_FLOW_ATTR_CLEAR]) {
1125 spin_lock_bh(&flow->lock);
1127 spin_unlock_bh(&flow->lock);
1132 genl_notify(reply, genl_info_net(info), info->snd_pid,
1133 ovs_dp_flow_multicast_group.id, info->nlhdr,
1136 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1137 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1146 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1148 struct nlattr **a = info->attrs;
1149 struct ovs_header *ovs_header = info->userhdr;
1150 struct sw_flow_key key;
1151 struct sk_buff *reply;
1152 struct sw_flow *flow;
1153 struct datapath *dp;
1154 struct flow_table *table;
1158 if (!a[OVS_FLOW_ATTR_KEY])
1160 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1164 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1168 table = genl_dereference(dp->table);
1169 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1173 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1174 info->snd_seq, OVS_FLOW_CMD_NEW);
1176 return PTR_ERR(reply);
1178 return genlmsg_reply(reply, info);
1181 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1183 struct nlattr **a = info->attrs;
1184 struct ovs_header *ovs_header = info->userhdr;
1185 struct sw_flow_key key;
1186 struct sk_buff *reply;
1187 struct sw_flow *flow;
1188 struct datapath *dp;
1189 struct flow_table *table;
1193 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1197 if (!a[OVS_FLOW_ATTR_KEY])
1198 return flush_flows(dp);
1200 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1204 table = genl_dereference(dp->table);
1205 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1209 reply = ovs_flow_cmd_alloc_info(flow);
1213 ovs_flow_tbl_remove(table, flow);
1215 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1216 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1219 ovs_flow_deferred_free(flow);
1221 genl_notify(reply, genl_info_net(info), info->snd_pid,
1222 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1226 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1228 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1229 struct datapath *dp;
1230 struct flow_table *table;
1232 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1236 table = genl_dereference(dp->table);
1239 struct sw_flow *flow;
1242 bucket = cb->args[0];
1244 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1248 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1249 NETLINK_CB(cb->skb).pid,
1250 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1251 OVS_FLOW_CMD_NEW) < 0)
1254 cb->args[0] = bucket;
1260 static struct genl_ops dp_flow_genl_ops[] = {
1261 { .cmd = OVS_FLOW_CMD_NEW,
1262 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1263 .policy = flow_policy,
1264 .doit = ovs_flow_cmd_new_or_set
1266 { .cmd = OVS_FLOW_CMD_DEL,
1267 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1268 .policy = flow_policy,
1269 .doit = ovs_flow_cmd_del
1271 { .cmd = OVS_FLOW_CMD_GET,
1272 .flags = 0, /* OK for unprivileged users. */
1273 .policy = flow_policy,
1274 .doit = ovs_flow_cmd_get,
1275 .dumpit = ovs_flow_cmd_dump
1277 { .cmd = OVS_FLOW_CMD_SET,
1278 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1279 .policy = flow_policy,
1280 .doit = ovs_flow_cmd_new_or_set,
1284 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1285 #ifdef HAVE_NLA_NUL_STRING
1286 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1288 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1291 static struct genl_family dp_datapath_genl_family = {
1292 .id = GENL_ID_GENERATE,
1293 .hdrsize = sizeof(struct ovs_header),
1294 .name = OVS_DATAPATH_FAMILY,
1295 .version = OVS_DATAPATH_VERSION,
1296 .maxattr = OVS_DP_ATTR_MAX,
1300 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1301 .name = OVS_DATAPATH_MCGROUP
1304 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1305 u32 pid, u32 seq, u32 flags, u8 cmd)
1307 struct ovs_header *ovs_header;
1308 struct ovs_dp_stats dp_stats;
1311 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1316 ovs_header->dp_ifindex = get_dpifindex(dp);
1319 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1322 goto nla_put_failure;
1324 get_dp_stats(dp, &dp_stats);
1325 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1326 goto nla_put_failure;
1328 return genlmsg_end(skb, ovs_header);
1331 genlmsg_cancel(skb, ovs_header);
1336 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1339 struct sk_buff *skb;
1342 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1344 return ERR_PTR(-ENOMEM);
1346 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1349 return ERR_PTR(retval);
1354 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1356 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1359 /* Called with genl_mutex and optionally with RTNL lock also. */
1360 static struct datapath *lookup_datapath(struct net *net,
1361 struct ovs_header *ovs_header,
1362 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1364 struct datapath *dp;
1366 if (!a[OVS_DP_ATTR_NAME])
1367 dp = get_dp(net, ovs_header->dp_ifindex);
1369 struct vport *vport;
1372 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1373 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1376 return dp ? dp : ERR_PTR(-ENODEV);
1379 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1381 struct nlattr **a = info->attrs;
1382 struct vport_parms parms;
1383 struct sk_buff *reply;
1384 struct datapath *dp;
1385 struct vport *vport;
1386 struct ovs_net *ovs_net;
1390 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1393 err = ovs_dp_cmd_validate(a);
1400 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1402 goto err_unlock_rtnl;
1404 /* Initialize kobject for bridge. This will be added as
1405 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1406 dp->ifobj.kset = NULL;
1407 kobject_init(&dp->ifobj, &dp_ktype);
1409 /* Allocate table. */
1411 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1415 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1416 if (!dp->stats_percpu) {
1418 goto err_destroy_table;
1420 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1422 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1426 goto err_destroy_percpu;
1429 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1430 INIT_HLIST_HEAD(&dp->ports[i]);
1432 /* Set up our datapath device. */
1433 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1434 parms.type = OVS_VPORT_TYPE_INTERNAL;
1435 parms.options = NULL;
1437 parms.port_no = OVSP_LOCAL;
1438 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1440 vport = new_vport(&parms);
1441 if (IS_ERR(vport)) {
1442 err = PTR_ERR(vport);
1446 goto err_destroy_ports_array;
1449 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1450 info->snd_seq, OVS_DP_CMD_NEW);
1451 err = PTR_ERR(reply);
1453 goto err_destroy_local_port;
1455 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1456 list_add_tail(&dp->list_node, &ovs_net->dps);
1457 ovs_dp_sysfs_add_dp(dp);
1461 genl_notify(reply, genl_info_net(info), info->snd_pid,
1462 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1466 err_destroy_local_port:
1467 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1468 err_destroy_ports_array:
1471 free_percpu(dp->stats_percpu);
1473 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1482 /* Called with genl_mutex. */
1483 static void __dp_destroy(struct datapath *dp)
1489 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1490 struct vport *vport;
1491 struct hlist_node *node, *n;
1493 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1494 if (vport->port_no != OVSP_LOCAL)
1495 ovs_dp_detach_port(vport);
1498 ovs_dp_sysfs_del_dp(dp);
1499 list_del(&dp->list_node);
1500 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1502 /* rtnl_unlock() will wait until all the references to devices that
1503 * are pending unregistration have been dropped. We do it here to
1504 * ensure that any internal devices (which contain DP pointers) are
1505 * fully destroyed before freeing the datapath.
1509 call_rcu(&dp->rcu, destroy_dp_rcu);
1512 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1514 struct sk_buff *reply;
1515 struct datapath *dp;
1518 err = ovs_dp_cmd_validate(info->attrs);
1522 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1527 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1528 info->snd_seq, OVS_DP_CMD_DEL);
1529 err = PTR_ERR(reply);
1535 genl_notify(reply, genl_info_net(info), info->snd_pid,
1536 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1542 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1544 struct sk_buff *reply;
1545 struct datapath *dp;
1548 err = ovs_dp_cmd_validate(info->attrs);
1552 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1556 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1557 info->snd_seq, OVS_DP_CMD_NEW);
1558 if (IS_ERR(reply)) {
1559 err = PTR_ERR(reply);
1560 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1561 ovs_dp_datapath_multicast_group.id, err);
1565 genl_notify(reply, genl_info_net(info), info->snd_pid,
1566 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1572 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1574 struct sk_buff *reply;
1575 struct datapath *dp;
1578 err = ovs_dp_cmd_validate(info->attrs);
1582 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1586 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1587 info->snd_seq, OVS_DP_CMD_NEW);
1589 return PTR_ERR(reply);
1591 return genlmsg_reply(reply, info);
1594 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1596 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1597 struct datapath *dp;
1598 int skip = cb->args[0];
1601 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1603 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1604 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1605 OVS_DP_CMD_NEW) < 0)
1615 static struct genl_ops dp_datapath_genl_ops[] = {
1616 { .cmd = OVS_DP_CMD_NEW,
1617 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1618 .policy = datapath_policy,
1619 .doit = ovs_dp_cmd_new
1621 { .cmd = OVS_DP_CMD_DEL,
1622 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1623 .policy = datapath_policy,
1624 .doit = ovs_dp_cmd_del
1626 { .cmd = OVS_DP_CMD_GET,
1627 .flags = 0, /* OK for unprivileged users. */
1628 .policy = datapath_policy,
1629 .doit = ovs_dp_cmd_get,
1630 .dumpit = ovs_dp_cmd_dump
1632 { .cmd = OVS_DP_CMD_SET,
1633 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1634 .policy = datapath_policy,
1635 .doit = ovs_dp_cmd_set,
1639 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1640 #ifdef HAVE_NLA_NUL_STRING
1641 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1642 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1643 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1645 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1646 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1648 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1649 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1650 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1651 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1654 static struct genl_family dp_vport_genl_family = {
1655 .id = GENL_ID_GENERATE,
1656 .hdrsize = sizeof(struct ovs_header),
1657 .name = OVS_VPORT_FAMILY,
1658 .version = OVS_VPORT_VERSION,
1659 .maxattr = OVS_VPORT_ATTR_MAX,
1663 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1664 .name = OVS_VPORT_MCGROUP
1667 /* Called with RTNL lock or RCU read lock. */
1668 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1669 u32 pid, u32 seq, u32 flags, u8 cmd)
1671 struct ovs_header *ovs_header;
1672 struct ovs_vport_stats vport_stats;
1675 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1680 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1682 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1683 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1684 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1685 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid))
1686 goto nla_put_failure;
1688 ovs_vport_get_stats(vport, &vport_stats);
1689 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1691 goto nla_put_failure;
1693 if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1694 vport->ops->get_addr(vport)))
1695 goto nla_put_failure;
1697 err = ovs_vport_get_options(vport, skb);
1698 if (err == -EMSGSIZE)
1701 return genlmsg_end(skb, ovs_header);
1706 genlmsg_cancel(skb, ovs_header);
1710 /* Called with RTNL lock or RCU read lock. */
1711 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1714 struct sk_buff *skb;
1717 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1719 return ERR_PTR(-ENOMEM);
1721 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1724 return ERR_PTR(retval);
1729 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1731 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1734 /* Called with RTNL lock or RCU read lock. */
1735 static struct vport *lookup_vport(struct net *net,
1736 struct ovs_header *ovs_header,
1737 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1739 struct datapath *dp;
1740 struct vport *vport;
1742 if (a[OVS_VPORT_ATTR_NAME]) {
1743 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1745 return ERR_PTR(-ENODEV);
1746 if (ovs_header->dp_ifindex &&
1747 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1748 return ERR_PTR(-ENODEV);
1750 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1751 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1753 if (port_no >= DP_MAX_PORTS)
1754 return ERR_PTR(-EFBIG);
1756 dp = get_dp(net, ovs_header->dp_ifindex);
1758 return ERR_PTR(-ENODEV);
1760 vport = ovs_vport_rtnl_rcu(dp, port_no);
1762 return ERR_PTR(-ENOENT);
1765 return ERR_PTR(-EINVAL);
1768 /* Called with RTNL lock. */
1769 static int change_vport(struct vport *vport,
1770 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1774 if (a[OVS_VPORT_ATTR_STATS])
1775 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1777 if (a[OVS_VPORT_ATTR_ADDRESS])
1778 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1783 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1785 struct nlattr **a = info->attrs;
1786 struct ovs_header *ovs_header = info->userhdr;
1787 struct vport_parms parms;
1788 struct sk_buff *reply;
1789 struct vport *vport;
1790 struct datapath *dp;
1795 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1796 !a[OVS_VPORT_ATTR_UPCALL_PID])
1799 err = ovs_vport_cmd_validate(a);
1804 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1809 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1810 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1813 if (port_no >= DP_MAX_PORTS)
1816 vport = ovs_vport_rtnl(dp, port_no);
1821 for (port_no = 1; ; port_no++) {
1822 if (port_no >= DP_MAX_PORTS) {
1826 vport = ovs_vport_rtnl(dp, port_no);
1832 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1833 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1834 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1836 parms.port_no = port_no;
1837 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1839 vport = new_vport(&parms);
1840 err = PTR_ERR(vport);
1844 ovs_dp_sysfs_add_if(vport);
1846 err = change_vport(vport, a);
1848 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1852 err = PTR_ERR(reply);
1855 ovs_dp_detach_port(vport);
1858 genl_notify(reply, genl_info_net(info), info->snd_pid,
1859 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1867 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1869 struct nlattr **a = info->attrs;
1870 struct sk_buff *reply;
1871 struct vport *vport;
1874 err = ovs_vport_cmd_validate(a);
1879 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1880 err = PTR_ERR(vport);
1885 if (a[OVS_VPORT_ATTR_TYPE] &&
1886 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1889 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1890 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1892 err = change_vport(vport, a);
1895 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1896 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1898 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1900 if (IS_ERR(reply)) {
1901 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1902 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1906 genl_notify(reply, genl_info_net(info), info->snd_pid,
1907 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1915 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1917 struct nlattr **a = info->attrs;
1918 struct sk_buff *reply;
1919 struct vport *vport;
1922 err = ovs_vport_cmd_validate(a);
1927 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1928 err = PTR_ERR(vport);
1932 if (vport->port_no == OVSP_LOCAL) {
1937 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1939 err = PTR_ERR(reply);
1943 ovs_dp_detach_port(vport);
1945 genl_notify(reply, genl_info_net(info), info->snd_pid,
1946 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1954 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1956 struct nlattr **a = info->attrs;
1957 struct ovs_header *ovs_header = info->userhdr;
1958 struct sk_buff *reply;
1959 struct vport *vport;
1962 err = ovs_vport_cmd_validate(a);
1967 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1968 err = PTR_ERR(vport);
1972 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1974 err = PTR_ERR(reply);
1980 return genlmsg_reply(reply, info);
1988 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1990 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1991 struct datapath *dp;
1992 int bucket = cb->args[0], skip = cb->args[1];
1995 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2000 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2001 struct vport *vport;
2002 struct hlist_node *n;
2005 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2007 ovs_vport_cmd_fill_info(vport, skb,
2008 NETLINK_CB(cb->skb).pid,
2011 OVS_VPORT_CMD_NEW) < 0)
2027 static struct genl_ops dp_vport_genl_ops[] = {
2028 { .cmd = OVS_VPORT_CMD_NEW,
2029 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2030 .policy = vport_policy,
2031 .doit = ovs_vport_cmd_new
2033 { .cmd = OVS_VPORT_CMD_DEL,
2034 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2035 .policy = vport_policy,
2036 .doit = ovs_vport_cmd_del
2038 { .cmd = OVS_VPORT_CMD_GET,
2039 .flags = 0, /* OK for unprivileged users. */
2040 .policy = vport_policy,
2041 .doit = ovs_vport_cmd_get,
2042 .dumpit = ovs_vport_cmd_dump
2044 { .cmd = OVS_VPORT_CMD_SET,
2045 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2046 .policy = vport_policy,
2047 .doit = ovs_vport_cmd_set,
2051 struct genl_family_and_ops {
2052 struct genl_family *family;
2053 struct genl_ops *ops;
2055 struct genl_multicast_group *group;
2058 static const struct genl_family_and_ops dp_genl_families[] = {
2059 { &dp_datapath_genl_family,
2060 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2061 &ovs_dp_datapath_multicast_group },
2062 { &dp_vport_genl_family,
2063 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2064 &ovs_dp_vport_multicast_group },
2065 { &dp_flow_genl_family,
2066 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2067 &ovs_dp_flow_multicast_group },
2068 { &dp_packet_genl_family,
2069 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2073 static void dp_unregister_genl(int n_families)
2077 for (i = 0; i < n_families; i++)
2078 genl_unregister_family(dp_genl_families[i].family);
2081 static int dp_register_genl(void)
2088 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2089 const struct genl_family_and_ops *f = &dp_genl_families[i];
2091 err = genl_register_family_with_ops(f->family, f->ops,
2098 err = genl_register_mc_group(f->family, f->group);
2107 dp_unregister_genl(n_registered);
2111 static int __rehash_flow_table(void *dummy)
2113 struct datapath *dp;
2118 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2120 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2121 struct flow_table *old_table = genl_dereference(dp->table);
2122 struct flow_table *new_table;
2124 new_table = ovs_flow_tbl_rehash(old_table);
2125 if (!IS_ERR(new_table)) {
2126 rcu_assign_pointer(dp->table, new_table);
2127 ovs_flow_tbl_deferred_destroy(old_table);
2135 static void rehash_flow_table(struct work_struct *work)
2137 genl_exec(__rehash_flow_table, NULL);
2138 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2141 static int dp_destroy_all(void *data)
2143 struct datapath *dp, *dp_next;
2144 struct ovs_net *ovs_net = data;
2146 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2152 static int __net_init ovs_init_net(struct net *net)
2154 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2156 INIT_LIST_HEAD(&ovs_net->dps);
2160 static void __net_exit ovs_exit_net(struct net *net)
2162 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2164 genl_exec(dp_destroy_all, ovs_net);
2167 static struct pernet_operations ovs_net_ops = {
2168 .init = ovs_init_net,
2169 .exit = ovs_exit_net,
2171 .size = sizeof(struct ovs_net),
2174 static int __init dp_init(void)
2176 struct sk_buff *dummy_skb;
2179 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2181 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2184 err = genl_exec_init();
2188 err = ovs_workqueues_init();
2190 goto error_genl_exec;
2192 err = ovs_tnl_init();
2196 err = ovs_flow_init();
2198 goto error_tnl_exit;
2200 err = ovs_vport_init();
2202 goto error_flow_exit;
2204 err = register_pernet_device(&ovs_net_ops);
2206 goto error_vport_exit;
2208 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2210 goto error_netns_exit;
2212 err = dp_register_genl();
2214 goto error_unreg_notifier;
2216 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2220 error_unreg_notifier:
2221 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2223 unregister_pernet_device(&ovs_net_ops);
2231 ovs_workqueues_exit();
2238 static void dp_cleanup(void)
2240 cancel_delayed_work_sync(&rehash_flow_wq);
2241 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2242 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2243 unregister_pernet_device(&ovs_net_ops);
2248 ovs_workqueues_exit();
2252 module_init(dp_init);
2253 module_exit(dp_cleanup);
2255 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2256 MODULE_LICENSE("GPL");