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(3,9,0)
65 #error Kernels before 2.6.18 or after 3.8 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.portid = p->upcall_portid;
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->portid == 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 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 return queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
419 /* Queue all of the segments. */
422 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
426 if (skb == segs && gso_type & SKB_GSO_UDP) {
427 /* The initial flow key extracted by ovs_flow_extract()
428 * in this case is for a first fragment, so we need to
429 * properly mark later fragments.
431 later_key = *upcall_info->key;
432 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
434 later_info = *upcall_info;
435 later_info.key = &later_key;
436 upcall_info = &later_info;
438 } while ((skb = skb->next));
440 /* Free all of the segments. */
448 } while ((skb = nskb));
452 static int queue_userspace_packet(struct net *net, int dp_ifindex,
454 const struct dp_upcall_info *upcall_info)
456 struct ovs_header *upcall;
457 struct sk_buff *nskb = NULL;
458 struct sk_buff *user_skb; /* to be queued to userspace */
463 if (vlan_tx_tag_present(skb)) {
464 nskb = skb_clone(skb, GFP_ATOMIC);
468 err = vlan_deaccel_tag(nskb);
475 if (nla_attr_size(skb->len) > USHRT_MAX) {
480 len = sizeof(struct ovs_header);
481 len += nla_total_size(skb->len);
482 len += nla_total_size(FLOW_BUFSIZE);
483 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
484 len += nla_total_size(8);
486 user_skb = genlmsg_new(len, GFP_ATOMIC);
492 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
493 0, upcall_info->cmd);
494 upcall->dp_ifindex = dp_ifindex;
496 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
497 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
498 nla_nest_end(user_skb, nla);
500 if (upcall_info->userdata)
501 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
502 nla_get_u64(upcall_info->userdata));
504 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
506 skb_copy_and_csum_dev(skb, nla_data(nla));
508 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
515 /* Called with genl_mutex. */
516 static int flush_flows(struct datapath *dp)
518 struct flow_table *old_table;
519 struct flow_table *new_table;
521 old_table = genl_dereference(dp->table);
522 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
526 rcu_assign_pointer(dp->table, new_table);
528 ovs_flow_tbl_deferred_destroy(old_table);
532 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
535 struct sw_flow_actions *acts;
537 int req_size = NLA_ALIGN(attr_len);
538 int next_offset = offsetof(struct sw_flow_actions, actions) +
541 if (req_size <= (ksize(*sfa) - next_offset))
544 new_acts_size = ksize(*sfa) * 2;
546 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
547 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
548 return ERR_PTR(-EMSGSIZE);
549 new_acts_size = MAX_ACTIONS_BUFSIZE;
552 acts = ovs_flow_actions_alloc(new_acts_size);
556 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
557 acts->actions_len = (*sfa)->actions_len;
562 (*sfa)->actions_len += req_size;
563 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
566 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
570 a = reserve_sfa_size(sfa, nla_attr_size(len));
574 a->nla_type = attrtype;
575 a->nla_len = nla_attr_size(len);
578 memcpy(nla_data(a), data, len);
579 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
584 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
586 int used = (*sfa)->actions_len;
589 err = add_action(sfa, attrtype, NULL, 0);
596 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
598 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
600 a->nla_len = sfa->actions_len - st_offset;
603 static int validate_and_copy_actions(const struct nlattr *attr,
604 const struct sw_flow_key *key, int depth,
605 struct sw_flow_actions **sfa);
607 static int validate_and_copy_sample(const struct nlattr *attr,
608 const struct sw_flow_key *key, int depth,
609 struct sw_flow_actions **sfa)
611 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
612 const struct nlattr *probability, *actions;
613 const struct nlattr *a;
614 int rem, start, err, st_acts;
616 memset(attrs, 0, sizeof(attrs));
617 nla_for_each_nested(a, attr, rem) {
618 int type = nla_type(a);
619 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
626 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
627 if (!probability || nla_len(probability) != sizeof(u32))
630 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
631 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
634 /* validation done, copy sample action. */
635 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
638 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
641 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
645 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
649 add_nested_action_end(*sfa, st_acts);
650 add_nested_action_end(*sfa, start);
655 static int validate_tp_port(const struct sw_flow_key *flow_key)
657 if (flow_key->eth.type == htons(ETH_P_IP)) {
658 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
660 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
661 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
668 static int validate_and_copy_set_tun(const struct nlattr *attr,
669 struct sw_flow_actions **sfa)
671 struct ovs_key_ipv4_tunnel tun_key;
674 err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
678 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
682 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
683 add_nested_action_end(*sfa, start);
688 static int validate_set(const struct nlattr *a,
689 const struct sw_flow_key *flow_key,
690 struct sw_flow_actions **sfa,
693 const struct nlattr *ovs_key = nla_data(a);
694 int key_type = nla_type(ovs_key);
696 /* There can be only one key in a action */
697 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
700 if (key_type > OVS_KEY_ATTR_MAX ||
701 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
702 ovs_key_lens[key_type] != -1))
706 const struct ovs_key_ipv4 *ipv4_key;
707 const struct ovs_key_ipv6 *ipv6_key;
710 case OVS_KEY_ATTR_PRIORITY:
711 case OVS_KEY_ATTR_TUN_ID:
712 case OVS_KEY_ATTR_ETHERNET:
715 case OVS_KEY_ATTR_SKB_MARK:
716 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
717 if (nla_get_u32(ovs_key) != 0)
722 case OVS_KEY_ATTR_TUNNEL:
724 err = validate_and_copy_set_tun(a, sfa);
729 case OVS_KEY_ATTR_IPV4:
730 if (flow_key->eth.type != htons(ETH_P_IP))
733 if (!flow_key->ip.proto)
736 ipv4_key = nla_data(ovs_key);
737 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
740 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
745 case OVS_KEY_ATTR_IPV6:
746 if (flow_key->eth.type != htons(ETH_P_IPV6))
749 if (!flow_key->ip.proto)
752 ipv6_key = nla_data(ovs_key);
753 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
756 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
759 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
764 case OVS_KEY_ATTR_TCP:
765 if (flow_key->ip.proto != IPPROTO_TCP)
768 return validate_tp_port(flow_key);
770 case OVS_KEY_ATTR_UDP:
771 if (flow_key->ip.proto != IPPROTO_UDP)
774 return validate_tp_port(flow_key);
783 static int validate_userspace(const struct nlattr *attr)
785 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
786 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
787 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
789 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
792 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
793 attr, userspace_policy);
797 if (!a[OVS_USERSPACE_ATTR_PID] ||
798 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
804 static int copy_action(const struct nlattr *from,
805 struct sw_flow_actions **sfa)
807 int totlen = NLA_ALIGN(from->nla_len);
810 to = reserve_sfa_size(sfa, from->nla_len);
814 memcpy(to, from, totlen);
818 static int validate_and_copy_actions(const struct nlattr *attr,
819 const struct sw_flow_key *key,
821 struct sw_flow_actions **sfa)
823 const struct nlattr *a;
826 if (depth >= SAMPLE_ACTION_DEPTH)
829 nla_for_each_nested(a, attr, rem) {
830 /* Expected argument lengths, (u32)-1 for variable length. */
831 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
832 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
833 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
834 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
835 [OVS_ACTION_ATTR_POP_VLAN] = 0,
836 [OVS_ACTION_ATTR_SET] = (u32)-1,
837 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
839 const struct ovs_action_push_vlan *vlan;
840 int type = nla_type(a);
843 if (type > OVS_ACTION_ATTR_MAX ||
844 (action_lens[type] != nla_len(a) &&
845 action_lens[type] != (u32)-1))
850 case OVS_ACTION_ATTR_UNSPEC:
853 case OVS_ACTION_ATTR_USERSPACE:
854 err = validate_userspace(a);
859 case OVS_ACTION_ATTR_OUTPUT:
860 if (nla_get_u32(a) >= DP_MAX_PORTS)
865 case OVS_ACTION_ATTR_POP_VLAN:
868 case OVS_ACTION_ATTR_PUSH_VLAN:
870 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
872 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
876 case OVS_ACTION_ATTR_SET:
877 err = validate_set(a, key, sfa, &skip_copy);
882 case OVS_ACTION_ATTR_SAMPLE:
883 err = validate_and_copy_sample(a, key, depth, sfa);
893 err = copy_action(a, sfa);
905 static void clear_stats(struct sw_flow *flow)
909 flow->packet_count = 0;
910 flow->byte_count = 0;
913 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
915 struct ovs_header *ovs_header = info->userhdr;
916 struct nlattr **a = info->attrs;
917 struct sw_flow_actions *acts;
918 struct sk_buff *packet;
919 struct sw_flow *flow;
927 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
928 !a[OVS_PACKET_ATTR_ACTIONS] ||
929 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
932 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
933 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
937 skb_reserve(packet, NET_IP_ALIGN);
939 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
941 skb_reset_mac_header(packet);
942 eth = eth_hdr(packet);
944 /* Normally, setting the skb 'protocol' field would be handled by a
945 * call to eth_type_trans(), but it assumes there's a sending
946 * device, which we may not have. */
947 if (ntohs(eth->h_proto) >= 1536)
948 packet->protocol = eth->h_proto;
950 packet->protocol = htons(ETH_P_802_2);
952 /* Build an sw_flow for sending this packet. */
953 flow = ovs_flow_alloc();
958 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
962 err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
965 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
970 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
971 rcu_assign_pointer(flow->sf_acts, acts);
975 OVS_CB(packet)->flow = flow;
976 packet->priority = flow->key.phy.priority;
977 skb_set_mark(packet, flow->key.phy.skb_mark);
980 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
986 err = ovs_execute_actions(dp, packet);
1003 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
1004 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
1005 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
1006 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
1009 static struct genl_ops dp_packet_genl_ops[] = {
1010 { .cmd = OVS_PACKET_CMD_EXECUTE,
1011 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1012 .policy = packet_policy,
1013 .doit = ovs_packet_cmd_execute
1017 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
1020 struct flow_table *table = genl_dereference(dp->table);
1022 stats->n_flows = ovs_flow_tbl_count(table);
1024 stats->n_hit = stats->n_missed = stats->n_lost = 0;
1025 for_each_possible_cpu(i) {
1026 const struct dp_stats_percpu *percpu_stats;
1027 struct dp_stats_percpu local_stats;
1030 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1033 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
1034 local_stats = *percpu_stats;
1035 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
1037 stats->n_hit += local_stats.n_hit;
1038 stats->n_missed += local_stats.n_missed;
1039 stats->n_lost += local_stats.n_lost;
1043 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1044 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1045 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1046 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1049 static struct genl_family dp_flow_genl_family = {
1050 .id = GENL_ID_GENERATE,
1051 .hdrsize = sizeof(struct ovs_header),
1052 .name = OVS_FLOW_FAMILY,
1053 .version = OVS_FLOW_VERSION,
1054 .maxattr = OVS_FLOW_ATTR_MAX,
1058 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1059 .name = OVS_FLOW_MCGROUP
1062 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1063 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1065 const struct nlattr *a;
1066 struct nlattr *start;
1069 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1073 nla_for_each_nested(a, attr, rem) {
1074 int type = nla_type(a);
1075 struct nlattr *st_sample;
1078 case OVS_SAMPLE_ATTR_PROBABILITY:
1079 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1082 case OVS_SAMPLE_ATTR_ACTIONS:
1083 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1086 err = actions_to_attr(nla_data(a), nla_len(a), skb);
1089 nla_nest_end(skb, st_sample);
1094 nla_nest_end(skb, start);
1098 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1100 const struct nlattr *ovs_key = nla_data(a);
1101 int key_type = nla_type(ovs_key);
1102 struct nlattr *start;
1106 case OVS_KEY_ATTR_IPV4_TUNNEL:
1107 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1111 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1114 nla_nest_end(skb, start);
1117 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1125 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1127 const struct nlattr *a;
1130 nla_for_each_attr(a, attr, len, rem) {
1131 int type = nla_type(a);
1134 case OVS_ACTION_ATTR_SET:
1135 err = set_action_to_attr(a, skb);
1140 case OVS_ACTION_ATTR_SAMPLE:
1141 err = sample_action_to_attr(a, skb);
1146 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1155 /* Called with genl_lock. */
1156 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1157 struct sk_buff *skb, u32 portid,
1158 u32 seq, u32 flags, u8 cmd)
1160 const int skb_orig_len = skb->len;
1161 const struct sw_flow_actions *sf_acts;
1162 struct nlattr *start;
1163 struct ovs_flow_stats stats;
1164 struct ovs_header *ovs_header;
1170 sf_acts = rcu_dereference_protected(flow->sf_acts,
1171 lockdep_genl_is_held());
1173 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1177 ovs_header->dp_ifindex = get_dpifindex(dp);
1179 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1181 goto nla_put_failure;
1182 err = ovs_flow_to_nlattrs(&flow->key, skb);
1185 nla_nest_end(skb, nla);
1187 spin_lock_bh(&flow->lock);
1189 stats.n_packets = flow->packet_count;
1190 stats.n_bytes = flow->byte_count;
1191 tcp_flags = flow->tcp_flags;
1192 spin_unlock_bh(&flow->lock);
1195 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1196 goto nla_put_failure;
1198 if (stats.n_packets &&
1199 nla_put(skb, OVS_FLOW_ATTR_STATS,
1200 sizeof(struct ovs_flow_stats), &stats))
1201 goto nla_put_failure;
1204 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1205 goto nla_put_failure;
1207 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1208 * this is the first flow to be dumped into 'skb'. This is unusual for
1209 * Netlink but individual action lists can be longer than
1210 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1211 * The userspace caller can always fetch the actions separately if it
1212 * really wants them. (Most userspace callers in fact don't care.)
1214 * This can only fail for dump operations because the skb is always
1215 * properly sized for single flows.
1217 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1219 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1220 if (err < 0 && skb_orig_len)
1222 nla_nest_end(skb, start);
1223 } else if (skb_orig_len) {
1228 return genlmsg_end(skb, ovs_header);
1233 genlmsg_cancel(skb, ovs_header);
1237 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1239 const struct sw_flow_actions *sf_acts;
1242 sf_acts = rcu_dereference_protected(flow->sf_acts,
1243 lockdep_genl_is_held());
1245 /* OVS_FLOW_ATTR_KEY */
1246 len = nla_total_size(FLOW_BUFSIZE);
1247 /* OVS_FLOW_ATTR_ACTIONS */
1248 len += nla_total_size(sf_acts->actions_len);
1249 /* OVS_FLOW_ATTR_STATS */
1250 len += nla_total_size(sizeof(struct ovs_flow_stats));
1251 /* OVS_FLOW_ATTR_TCP_FLAGS */
1252 len += nla_total_size(1);
1253 /* OVS_FLOW_ATTR_USED */
1254 len += nla_total_size(8);
1256 len += NLMSG_ALIGN(sizeof(struct ovs_header));
1258 return genlmsg_new(len, GFP_KERNEL);
1261 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1262 struct datapath *dp,
1263 u32 portid, u32 seq, u8 cmd)
1265 struct sk_buff *skb;
1268 skb = ovs_flow_cmd_alloc_info(flow);
1270 return ERR_PTR(-ENOMEM);
1272 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1277 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1279 struct nlattr **a = info->attrs;
1280 struct ovs_header *ovs_header = info->userhdr;
1281 struct sw_flow_key key;
1282 struct sw_flow *flow;
1283 struct sk_buff *reply;
1284 struct datapath *dp;
1285 struct flow_table *table;
1286 struct sw_flow_actions *acts = NULL;
1292 if (!a[OVS_FLOW_ATTR_KEY])
1294 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1298 /* Validate actions. */
1299 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1300 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1301 error = PTR_ERR(acts);
1305 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0, &acts);
1308 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1313 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1318 table = genl_dereference(dp->table);
1319 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1321 /* Bail out if we're not allowed to create a new flow. */
1323 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1326 /* Expand table, if necessary, to make room. */
1327 if (ovs_flow_tbl_need_to_expand(table)) {
1328 struct flow_table *new_table;
1330 new_table = ovs_flow_tbl_expand(table);
1331 if (!IS_ERR(new_table)) {
1332 rcu_assign_pointer(dp->table, new_table);
1333 ovs_flow_tbl_deferred_destroy(table);
1334 table = genl_dereference(dp->table);
1338 /* Allocate flow. */
1339 flow = ovs_flow_alloc();
1341 error = PTR_ERR(flow);
1346 rcu_assign_pointer(flow->sf_acts, acts);
1348 /* Put flow in bucket. */
1349 ovs_flow_tbl_insert(table, flow, &key, key_len);
1351 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1355 /* We found a matching flow. */
1356 struct sw_flow_actions *old_acts;
1358 /* Bail out if we're not allowed to modify an existing flow.
1359 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1360 * because Generic Netlink treats the latter as a dump
1361 * request. We also accept NLM_F_EXCL in case that bug ever
1365 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1366 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1369 /* Update actions. */
1370 old_acts = rcu_dereference_protected(flow->sf_acts,
1371 lockdep_genl_is_held());
1372 rcu_assign_pointer(flow->sf_acts, acts);
1373 ovs_flow_deferred_free_acts(old_acts);
1375 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1376 info->snd_seq, OVS_FLOW_CMD_NEW);
1379 if (a[OVS_FLOW_ATTR_CLEAR]) {
1380 spin_lock_bh(&flow->lock);
1382 spin_unlock_bh(&flow->lock);
1387 genl_notify(reply, genl_info_net(info), info->snd_portid,
1388 ovs_dp_flow_multicast_group.id, info->nlhdr,
1391 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1392 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1401 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1403 struct nlattr **a = info->attrs;
1404 struct ovs_header *ovs_header = info->userhdr;
1405 struct sw_flow_key key;
1406 struct sk_buff *reply;
1407 struct sw_flow *flow;
1408 struct datapath *dp;
1409 struct flow_table *table;
1413 if (!a[OVS_FLOW_ATTR_KEY])
1415 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1419 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1423 table = genl_dereference(dp->table);
1424 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1428 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1429 info->snd_seq, OVS_FLOW_CMD_NEW);
1431 return PTR_ERR(reply);
1433 return genlmsg_reply(reply, info);
1436 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1438 struct nlattr **a = info->attrs;
1439 struct ovs_header *ovs_header = info->userhdr;
1440 struct sw_flow_key key;
1441 struct sk_buff *reply;
1442 struct sw_flow *flow;
1443 struct datapath *dp;
1444 struct flow_table *table;
1448 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1452 if (!a[OVS_FLOW_ATTR_KEY])
1453 return flush_flows(dp);
1455 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1459 table = genl_dereference(dp->table);
1460 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1464 reply = ovs_flow_cmd_alloc_info(flow);
1468 ovs_flow_tbl_remove(table, flow);
1470 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1471 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1474 ovs_flow_deferred_free(flow);
1476 genl_notify(reply, genl_info_net(info), info->snd_portid,
1477 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1481 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1483 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1484 struct datapath *dp;
1485 struct flow_table *table;
1487 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1491 table = genl_dereference(dp->table);
1494 struct sw_flow *flow;
1497 bucket = cb->args[0];
1499 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1503 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1504 NETLINK_CB(cb->skb).portid,
1505 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1506 OVS_FLOW_CMD_NEW) < 0)
1509 cb->args[0] = bucket;
1515 static struct genl_ops dp_flow_genl_ops[] = {
1516 { .cmd = OVS_FLOW_CMD_NEW,
1517 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1518 .policy = flow_policy,
1519 .doit = ovs_flow_cmd_new_or_set
1521 { .cmd = OVS_FLOW_CMD_DEL,
1522 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1523 .policy = flow_policy,
1524 .doit = ovs_flow_cmd_del
1526 { .cmd = OVS_FLOW_CMD_GET,
1527 .flags = 0, /* OK for unprivileged users. */
1528 .policy = flow_policy,
1529 .doit = ovs_flow_cmd_get,
1530 .dumpit = ovs_flow_cmd_dump
1532 { .cmd = OVS_FLOW_CMD_SET,
1533 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1534 .policy = flow_policy,
1535 .doit = ovs_flow_cmd_new_or_set,
1539 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1540 #ifdef HAVE_NLA_NUL_STRING
1541 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1543 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1546 static struct genl_family dp_datapath_genl_family = {
1547 .id = GENL_ID_GENERATE,
1548 .hdrsize = sizeof(struct ovs_header),
1549 .name = OVS_DATAPATH_FAMILY,
1550 .version = OVS_DATAPATH_VERSION,
1551 .maxattr = OVS_DP_ATTR_MAX,
1555 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1556 .name = OVS_DATAPATH_MCGROUP
1559 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1560 u32 portid, u32 seq, u32 flags, u8 cmd)
1562 struct ovs_header *ovs_header;
1563 struct ovs_dp_stats dp_stats;
1566 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1571 ovs_header->dp_ifindex = get_dpifindex(dp);
1574 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1577 goto nla_put_failure;
1579 get_dp_stats(dp, &dp_stats);
1580 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1581 goto nla_put_failure;
1583 return genlmsg_end(skb, ovs_header);
1586 genlmsg_cancel(skb, ovs_header);
1591 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1594 struct sk_buff *skb;
1597 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1599 return ERR_PTR(-ENOMEM);
1601 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1604 return ERR_PTR(retval);
1609 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1611 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1614 /* Called with genl_mutex and optionally with RTNL lock also. */
1615 static struct datapath *lookup_datapath(struct net *net,
1616 struct ovs_header *ovs_header,
1617 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1619 struct datapath *dp;
1621 if (!a[OVS_DP_ATTR_NAME])
1622 dp = get_dp(net, ovs_header->dp_ifindex);
1624 struct vport *vport;
1627 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1628 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1631 return dp ? dp : ERR_PTR(-ENODEV);
1634 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1636 struct nlattr **a = info->attrs;
1637 struct vport_parms parms;
1638 struct sk_buff *reply;
1639 struct datapath *dp;
1640 struct vport *vport;
1641 struct ovs_net *ovs_net;
1645 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1648 err = ovs_dp_cmd_validate(a);
1655 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1657 goto err_unlock_rtnl;
1659 /* Initialize kobject for bridge. This will be added as
1660 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1661 dp->ifobj.kset = NULL;
1662 kobject_init(&dp->ifobj, &dp_ktype);
1664 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1666 /* Allocate table. */
1668 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1672 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1673 if (!dp->stats_percpu) {
1675 goto err_destroy_table;
1678 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1682 goto err_destroy_percpu;
1685 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1686 INIT_HLIST_HEAD(&dp->ports[i]);
1688 /* Set up our datapath device. */
1689 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1690 parms.type = OVS_VPORT_TYPE_INTERNAL;
1691 parms.options = NULL;
1693 parms.port_no = OVSP_LOCAL;
1694 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1696 vport = new_vport(&parms);
1697 if (IS_ERR(vport)) {
1698 err = PTR_ERR(vport);
1702 goto err_destroy_ports_array;
1705 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1706 info->snd_seq, OVS_DP_CMD_NEW);
1707 err = PTR_ERR(reply);
1709 goto err_destroy_local_port;
1711 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1712 list_add_tail(&dp->list_node, &ovs_net->dps);
1713 ovs_dp_sysfs_add_dp(dp);
1717 genl_notify(reply, genl_info_net(info), info->snd_portid,
1718 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1722 err_destroy_local_port:
1723 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1724 err_destroy_ports_array:
1727 free_percpu(dp->stats_percpu);
1729 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1731 release_net(ovs_dp_get_net(dp));
1739 /* Called with genl_mutex. */
1740 static void __dp_destroy(struct datapath *dp)
1746 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1747 struct vport *vport;
1748 struct hlist_node *node, *n;
1750 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1751 if (vport->port_no != OVSP_LOCAL)
1752 ovs_dp_detach_port(vport);
1755 ovs_dp_sysfs_del_dp(dp);
1756 list_del(&dp->list_node);
1757 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1759 /* rtnl_unlock() will wait until all the references to devices that
1760 * are pending unregistration have been dropped. We do it here to
1761 * ensure that any internal devices (which contain DP pointers) are
1762 * fully destroyed before freeing the datapath.
1766 call_rcu(&dp->rcu, destroy_dp_rcu);
1769 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1771 struct sk_buff *reply;
1772 struct datapath *dp;
1775 err = ovs_dp_cmd_validate(info->attrs);
1779 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1784 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1785 info->snd_seq, OVS_DP_CMD_DEL);
1786 err = PTR_ERR(reply);
1792 genl_notify(reply, genl_info_net(info), info->snd_portid,
1793 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1799 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1801 struct sk_buff *reply;
1802 struct datapath *dp;
1805 err = ovs_dp_cmd_validate(info->attrs);
1809 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1813 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1814 info->snd_seq, OVS_DP_CMD_NEW);
1815 if (IS_ERR(reply)) {
1816 err = PTR_ERR(reply);
1817 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1818 ovs_dp_datapath_multicast_group.id, err);
1822 genl_notify(reply, genl_info_net(info), info->snd_portid,
1823 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1829 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1831 struct sk_buff *reply;
1832 struct datapath *dp;
1835 err = ovs_dp_cmd_validate(info->attrs);
1839 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1843 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1844 info->snd_seq, OVS_DP_CMD_NEW);
1846 return PTR_ERR(reply);
1848 return genlmsg_reply(reply, info);
1851 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1853 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1854 struct datapath *dp;
1855 int skip = cb->args[0];
1858 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1860 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1861 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1862 OVS_DP_CMD_NEW) < 0)
1872 static struct genl_ops dp_datapath_genl_ops[] = {
1873 { .cmd = OVS_DP_CMD_NEW,
1874 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1875 .policy = datapath_policy,
1876 .doit = ovs_dp_cmd_new
1878 { .cmd = OVS_DP_CMD_DEL,
1879 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1880 .policy = datapath_policy,
1881 .doit = ovs_dp_cmd_del
1883 { .cmd = OVS_DP_CMD_GET,
1884 .flags = 0, /* OK for unprivileged users. */
1885 .policy = datapath_policy,
1886 .doit = ovs_dp_cmd_get,
1887 .dumpit = ovs_dp_cmd_dump
1889 { .cmd = OVS_DP_CMD_SET,
1890 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1891 .policy = datapath_policy,
1892 .doit = ovs_dp_cmd_set,
1896 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1897 #ifdef HAVE_NLA_NUL_STRING
1898 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1899 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1900 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1902 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1903 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1905 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1906 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1907 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1908 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1911 static struct genl_family dp_vport_genl_family = {
1912 .id = GENL_ID_GENERATE,
1913 .hdrsize = sizeof(struct ovs_header),
1914 .name = OVS_VPORT_FAMILY,
1915 .version = OVS_VPORT_VERSION,
1916 .maxattr = OVS_VPORT_ATTR_MAX,
1920 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1921 .name = OVS_VPORT_MCGROUP
1924 /* Called with RTNL lock or RCU read lock. */
1925 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1926 u32 portid, u32 seq, u32 flags, u8 cmd)
1928 struct ovs_header *ovs_header;
1929 struct ovs_vport_stats vport_stats;
1932 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1937 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1939 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1940 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1941 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1942 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1943 goto nla_put_failure;
1945 ovs_vport_get_stats(vport, &vport_stats);
1946 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1948 goto nla_put_failure;
1950 if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1951 vport->ops->get_addr(vport)))
1952 goto nla_put_failure;
1954 err = ovs_vport_get_options(vport, skb);
1955 if (err == -EMSGSIZE)
1958 return genlmsg_end(skb, ovs_header);
1963 genlmsg_cancel(skb, ovs_header);
1967 /* Called with RTNL lock or RCU read lock. */
1968 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1971 struct sk_buff *skb;
1974 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1976 return ERR_PTR(-ENOMEM);
1978 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1981 return ERR_PTR(retval);
1986 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1988 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1991 /* Called with RTNL lock or RCU read lock. */
1992 static struct vport *lookup_vport(struct net *net,
1993 struct ovs_header *ovs_header,
1994 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1996 struct datapath *dp;
1997 struct vport *vport;
1999 if (a[OVS_VPORT_ATTR_NAME]) {
2000 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2002 return ERR_PTR(-ENODEV);
2003 if (ovs_header->dp_ifindex &&
2004 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2005 return ERR_PTR(-ENODEV);
2007 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2008 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2010 if (port_no >= DP_MAX_PORTS)
2011 return ERR_PTR(-EFBIG);
2013 dp = get_dp(net, ovs_header->dp_ifindex);
2015 return ERR_PTR(-ENODEV);
2017 vport = ovs_vport_rtnl_rcu(dp, port_no);
2019 return ERR_PTR(-ENOENT);
2022 return ERR_PTR(-EINVAL);
2025 /* Called with RTNL lock. */
2026 static int change_vport(struct vport *vport,
2027 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2031 if (a[OVS_VPORT_ATTR_STATS])
2032 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2034 if (a[OVS_VPORT_ATTR_ADDRESS])
2035 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
2040 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2042 struct nlattr **a = info->attrs;
2043 struct ovs_header *ovs_header = info->userhdr;
2044 struct vport_parms parms;
2045 struct sk_buff *reply;
2046 struct vport *vport;
2047 struct datapath *dp;
2052 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2053 !a[OVS_VPORT_ATTR_UPCALL_PID])
2056 err = ovs_vport_cmd_validate(a);
2061 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2066 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2067 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2070 if (port_no >= DP_MAX_PORTS)
2073 vport = ovs_vport_rtnl(dp, port_no);
2078 for (port_no = 1; ; port_no++) {
2079 if (port_no >= DP_MAX_PORTS) {
2083 vport = ovs_vport_rtnl(dp, port_no);
2089 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2090 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2091 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2093 parms.port_no = port_no;
2094 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2096 vport = new_vport(&parms);
2097 err = PTR_ERR(vport);
2101 ovs_dp_sysfs_add_if(vport);
2103 err = change_vport(vport, a);
2105 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2109 err = PTR_ERR(reply);
2112 ovs_dp_detach_port(vport);
2115 genl_notify(reply, genl_info_net(info), info->snd_portid,
2116 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2124 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2126 struct nlattr **a = info->attrs;
2127 struct sk_buff *reply;
2128 struct vport *vport;
2131 err = ovs_vport_cmd_validate(a);
2136 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2137 err = PTR_ERR(vport);
2142 if (a[OVS_VPORT_ATTR_TYPE] &&
2143 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2146 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2147 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2149 err = change_vport(vport, a);
2152 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
2153 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2155 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2156 info->snd_seq, OVS_VPORT_CMD_NEW);
2157 if (IS_ERR(reply)) {
2158 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
2159 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
2163 genl_notify(reply, genl_info_net(info), info->snd_portid,
2164 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2172 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2174 struct nlattr **a = info->attrs;
2175 struct sk_buff *reply;
2176 struct vport *vport;
2179 err = ovs_vport_cmd_validate(a);
2184 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2185 err = PTR_ERR(vport);
2189 if (vport->port_no == OVSP_LOCAL) {
2194 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2195 info->snd_seq, OVS_VPORT_CMD_DEL);
2196 err = PTR_ERR(reply);
2200 ovs_dp_detach_port(vport);
2202 genl_notify(reply, genl_info_net(info), info->snd_portid,
2203 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2211 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2213 struct nlattr **a = info->attrs;
2214 struct ovs_header *ovs_header = info->userhdr;
2215 struct sk_buff *reply;
2216 struct vport *vport;
2219 err = ovs_vport_cmd_validate(a);
2224 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2225 err = PTR_ERR(vport);
2229 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2230 info->snd_seq, OVS_VPORT_CMD_NEW);
2231 err = PTR_ERR(reply);
2237 return genlmsg_reply(reply, info);
2245 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2247 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2248 struct datapath *dp;
2249 int bucket = cb->args[0], skip = cb->args[1];
2252 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2257 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2258 struct vport *vport;
2259 struct hlist_node *n;
2262 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2264 ovs_vport_cmd_fill_info(vport, skb,
2265 NETLINK_CB(cb->skb).portid,
2268 OVS_VPORT_CMD_NEW) < 0)
2284 static struct genl_ops dp_vport_genl_ops[] = {
2285 { .cmd = OVS_VPORT_CMD_NEW,
2286 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2287 .policy = vport_policy,
2288 .doit = ovs_vport_cmd_new
2290 { .cmd = OVS_VPORT_CMD_DEL,
2291 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2292 .policy = vport_policy,
2293 .doit = ovs_vport_cmd_del
2295 { .cmd = OVS_VPORT_CMD_GET,
2296 .flags = 0, /* OK for unprivileged users. */
2297 .policy = vport_policy,
2298 .doit = ovs_vport_cmd_get,
2299 .dumpit = ovs_vport_cmd_dump
2301 { .cmd = OVS_VPORT_CMD_SET,
2302 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2303 .policy = vport_policy,
2304 .doit = ovs_vport_cmd_set,
2308 struct genl_family_and_ops {
2309 struct genl_family *family;
2310 struct genl_ops *ops;
2312 struct genl_multicast_group *group;
2315 static const struct genl_family_and_ops dp_genl_families[] = {
2316 { &dp_datapath_genl_family,
2317 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2318 &ovs_dp_datapath_multicast_group },
2319 { &dp_vport_genl_family,
2320 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2321 &ovs_dp_vport_multicast_group },
2322 { &dp_flow_genl_family,
2323 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2324 &ovs_dp_flow_multicast_group },
2325 { &dp_packet_genl_family,
2326 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2330 static void dp_unregister_genl(int n_families)
2334 for (i = 0; i < n_families; i++)
2335 genl_unregister_family(dp_genl_families[i].family);
2338 static int dp_register_genl(void)
2345 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2346 const struct genl_family_and_ops *f = &dp_genl_families[i];
2348 err = genl_register_family_with_ops(f->family, f->ops,
2355 err = genl_register_mc_group(f->family, f->group);
2364 dp_unregister_genl(n_registered);
2368 static int __rehash_flow_table(void *dummy)
2370 struct datapath *dp;
2375 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2377 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2378 struct flow_table *old_table = genl_dereference(dp->table);
2379 struct flow_table *new_table;
2381 new_table = ovs_flow_tbl_rehash(old_table);
2382 if (!IS_ERR(new_table)) {
2383 rcu_assign_pointer(dp->table, new_table);
2384 ovs_flow_tbl_deferred_destroy(old_table);
2392 static void rehash_flow_table(struct work_struct *work)
2394 genl_exec(__rehash_flow_table, NULL);
2395 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2398 static int dp_destroy_all(void *data)
2400 struct datapath *dp, *dp_next;
2401 struct ovs_net *ovs_net = data;
2403 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2409 static int __net_init ovs_init_net(struct net *net)
2411 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2413 INIT_LIST_HEAD(&ovs_net->dps);
2417 static void __net_exit ovs_exit_net(struct net *net)
2419 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2421 genl_exec(dp_destroy_all, ovs_net);
2424 static struct pernet_operations ovs_net_ops = {
2425 .init = ovs_init_net,
2426 .exit = ovs_exit_net,
2428 .size = sizeof(struct ovs_net),
2431 static int __init dp_init(void)
2433 struct sk_buff *dummy_skb;
2436 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2438 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2441 err = genl_exec_init();
2445 err = ovs_workqueues_init();
2447 goto error_genl_exec;
2449 err = ovs_tnl_init();
2453 err = ovs_flow_init();
2455 goto error_tnl_exit;
2457 err = ovs_vport_init();
2459 goto error_flow_exit;
2461 err = register_pernet_device(&ovs_net_ops);
2463 goto error_vport_exit;
2465 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2467 goto error_netns_exit;
2469 err = dp_register_genl();
2471 goto error_unreg_notifier;
2473 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2477 error_unreg_notifier:
2478 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2480 unregister_pernet_device(&ovs_net_ops);
2488 ovs_workqueues_exit();
2495 static void dp_cleanup(void)
2497 cancel_delayed_work_sync(&rehash_flow_wq);
2498 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2499 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2500 unregister_pernet_device(&ovs_net_ops);
2505 ovs_workqueues_exit();
2509 module_init(dp_init);
2510 module_exit(dp_cleanup);
2512 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2513 MODULE_LICENSE("GPL");
2514 MODULE_VERSION(VERSION);