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 /* Queue all of the segments. */
420 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
424 if (skb == segs && gso_type & SKB_GSO_UDP) {
425 /* The initial flow key extracted by ovs_flow_extract()
426 * in this case is for a first fragment, so we need to
427 * properly mark later fragments.
429 later_key = *upcall_info->key;
430 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
432 later_info = *upcall_info;
433 later_info.key = &later_key;
434 upcall_info = &later_info;
436 } while ((skb = skb->next));
438 /* Free all of the segments. */
446 } while ((skb = nskb));
450 static int queue_userspace_packet(struct net *net, int dp_ifindex,
452 const struct dp_upcall_info *upcall_info)
454 struct ovs_header *upcall;
455 struct sk_buff *nskb = NULL;
456 struct sk_buff *user_skb; /* to be queued to userspace */
461 if (vlan_tx_tag_present(skb)) {
462 nskb = skb_clone(skb, GFP_ATOMIC);
466 err = vlan_deaccel_tag(nskb);
473 if (nla_attr_size(skb->len) > USHRT_MAX) {
478 len = sizeof(struct ovs_header);
479 len += nla_total_size(skb->len);
480 len += nla_total_size(FLOW_BUFSIZE);
481 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
482 len += nla_total_size(8);
484 user_skb = genlmsg_new(len, GFP_ATOMIC);
490 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
491 0, upcall_info->cmd);
492 upcall->dp_ifindex = dp_ifindex;
494 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
495 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
496 nla_nest_end(user_skb, nla);
498 if (upcall_info->userdata)
499 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
500 nla_get_u64(upcall_info->userdata));
502 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
504 skb_copy_and_csum_dev(skb, nla_data(nla));
506 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
513 /* Called with genl_mutex. */
514 static int flush_flows(struct datapath *dp)
516 struct flow_table *old_table;
517 struct flow_table *new_table;
519 old_table = genl_dereference(dp->table);
520 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
524 rcu_assign_pointer(dp->table, new_table);
526 ovs_flow_tbl_deferred_destroy(old_table);
530 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
533 struct sw_flow_actions *acts;
535 int req_size = NLA_ALIGN(attr_len);
536 int next_offset = offsetof(struct sw_flow_actions, actions) +
539 if (req_size <= (ksize(*sfa) - next_offset))
542 new_acts_size = ksize(*sfa) * 2;
544 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
545 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
546 return ERR_PTR(-EMSGSIZE);
547 new_acts_size = MAX_ACTIONS_BUFSIZE;
550 acts = ovs_flow_actions_alloc(new_acts_size);
554 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
555 acts->actions_len = (*sfa)->actions_len;
560 (*sfa)->actions_len += req_size;
561 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
564 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
568 a = reserve_sfa_size(sfa, nla_attr_size(len));
572 a->nla_type = attrtype;
573 a->nla_len = nla_attr_size(len);
576 memcpy(nla_data(a), data, len);
577 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
582 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
584 int used = (*sfa)->actions_len;
587 err = add_action(sfa, attrtype, NULL, 0);
594 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
596 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
598 a->nla_len = sfa->actions_len - st_offset;
601 static int validate_and_copy_actions(const struct nlattr *attr,
602 const struct sw_flow_key *key, int depth,
603 struct sw_flow_actions **sfa);
605 static int validate_and_copy_sample(const struct nlattr *attr,
606 const struct sw_flow_key *key, int depth,
607 struct sw_flow_actions **sfa)
609 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
610 const struct nlattr *probability, *actions;
611 const struct nlattr *a;
612 int rem, start, err, st_acts;
614 memset(attrs, 0, sizeof(attrs));
615 nla_for_each_nested(a, attr, rem) {
616 int type = nla_type(a);
617 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
624 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
625 if (!probability || nla_len(probability) != sizeof(u32))
628 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
629 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
632 /* validation done, copy sample action. */
633 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
636 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
639 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
643 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
647 add_nested_action_end(*sfa, st_acts);
648 add_nested_action_end(*sfa, start);
653 static int validate_tp_port(const struct sw_flow_key *flow_key)
655 if (flow_key->eth.type == htons(ETH_P_IP)) {
656 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
658 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
659 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
666 static int validate_and_copy_set_tun(const struct nlattr *attr,
667 struct sw_flow_actions **sfa)
669 struct ovs_key_ipv4_tunnel tun_key;
672 err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
676 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
680 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
681 add_nested_action_end(*sfa, start);
686 static int validate_set(const struct nlattr *a,
687 const struct sw_flow_key *flow_key,
688 struct sw_flow_actions **sfa,
691 const struct nlattr *ovs_key = nla_data(a);
692 int key_type = nla_type(ovs_key);
694 /* There can be only one key in a action */
695 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
698 if (key_type > OVS_KEY_ATTR_MAX ||
699 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
700 ovs_key_lens[key_type] != -1))
704 const struct ovs_key_ipv4 *ipv4_key;
705 const struct ovs_key_ipv6 *ipv6_key;
708 case OVS_KEY_ATTR_PRIORITY:
709 case OVS_KEY_ATTR_TUN_ID:
710 case OVS_KEY_ATTR_ETHERNET:
713 case OVS_KEY_ATTR_SKB_MARK:
714 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
715 if (nla_get_u32(ovs_key) != 0)
720 case OVS_KEY_ATTR_TUNNEL:
722 err = validate_and_copy_set_tun(a, sfa);
727 case OVS_KEY_ATTR_IPV4:
728 if (flow_key->eth.type != htons(ETH_P_IP))
731 if (!flow_key->ip.proto)
734 ipv4_key = nla_data(ovs_key);
735 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
738 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
743 case OVS_KEY_ATTR_IPV6:
744 if (flow_key->eth.type != htons(ETH_P_IPV6))
747 if (!flow_key->ip.proto)
750 ipv6_key = nla_data(ovs_key);
751 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
754 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
757 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
762 case OVS_KEY_ATTR_TCP:
763 if (flow_key->ip.proto != IPPROTO_TCP)
766 return validate_tp_port(flow_key);
768 case OVS_KEY_ATTR_UDP:
769 if (flow_key->ip.proto != IPPROTO_UDP)
772 return validate_tp_port(flow_key);
781 static int validate_userspace(const struct nlattr *attr)
783 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
784 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
785 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
787 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
790 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
791 attr, userspace_policy);
795 if (!a[OVS_USERSPACE_ATTR_PID] ||
796 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
802 static int copy_action(const struct nlattr *from,
803 struct sw_flow_actions **sfa)
805 int totlen = NLA_ALIGN(from->nla_len);
808 to = reserve_sfa_size(sfa, from->nla_len);
812 memcpy(to, from, totlen);
816 static int validate_and_copy_actions(const struct nlattr *attr,
817 const struct sw_flow_key *key,
819 struct sw_flow_actions **sfa)
821 const struct nlattr *a;
824 if (depth >= SAMPLE_ACTION_DEPTH)
827 nla_for_each_nested(a, attr, rem) {
828 /* Expected argument lengths, (u32)-1 for variable length. */
829 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
830 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
831 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
832 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
833 [OVS_ACTION_ATTR_POP_VLAN] = 0,
834 [OVS_ACTION_ATTR_SET] = (u32)-1,
835 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
837 const struct ovs_action_push_vlan *vlan;
838 int type = nla_type(a);
841 if (type > OVS_ACTION_ATTR_MAX ||
842 (action_lens[type] != nla_len(a) &&
843 action_lens[type] != (u32)-1))
848 case OVS_ACTION_ATTR_UNSPEC:
851 case OVS_ACTION_ATTR_USERSPACE:
852 err = validate_userspace(a);
857 case OVS_ACTION_ATTR_OUTPUT:
858 if (nla_get_u32(a) >= DP_MAX_PORTS)
863 case OVS_ACTION_ATTR_POP_VLAN:
866 case OVS_ACTION_ATTR_PUSH_VLAN:
868 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
870 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
874 case OVS_ACTION_ATTR_SET:
875 err = validate_set(a, key, sfa, &skip_copy);
880 case OVS_ACTION_ATTR_SAMPLE:
881 err = validate_and_copy_sample(a, key, depth, sfa);
891 err = copy_action(a, sfa);
903 static void clear_stats(struct sw_flow *flow)
907 flow->packet_count = 0;
908 flow->byte_count = 0;
911 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
913 struct ovs_header *ovs_header = info->userhdr;
914 struct nlattr **a = info->attrs;
915 struct sw_flow_actions *acts;
916 struct sk_buff *packet;
917 struct sw_flow *flow;
925 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
926 !a[OVS_PACKET_ATTR_ACTIONS] ||
927 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
930 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
931 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
935 skb_reserve(packet, NET_IP_ALIGN);
937 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
939 skb_reset_mac_header(packet);
940 eth = eth_hdr(packet);
942 /* Normally, setting the skb 'protocol' field would be handled by a
943 * call to eth_type_trans(), but it assumes there's a sending
944 * device, which we may not have. */
945 if (ntohs(eth->h_proto) >= 1536)
946 packet->protocol = eth->h_proto;
948 packet->protocol = htons(ETH_P_802_2);
950 /* Build an sw_flow for sending this packet. */
951 flow = ovs_flow_alloc();
956 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
960 err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
963 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
968 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
969 rcu_assign_pointer(flow->sf_acts, acts);
973 OVS_CB(packet)->flow = flow;
974 packet->priority = flow->key.phy.priority;
975 skb_set_mark(packet, flow->key.phy.skb_mark);
978 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
984 err = ovs_execute_actions(dp, packet);
1001 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
1002 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
1003 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
1004 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
1007 static struct genl_ops dp_packet_genl_ops[] = {
1008 { .cmd = OVS_PACKET_CMD_EXECUTE,
1009 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1010 .policy = packet_policy,
1011 .doit = ovs_packet_cmd_execute
1015 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
1018 struct flow_table *table = genl_dereference(dp->table);
1020 stats->n_flows = ovs_flow_tbl_count(table);
1022 stats->n_hit = stats->n_missed = stats->n_lost = 0;
1023 for_each_possible_cpu(i) {
1024 const struct dp_stats_percpu *percpu_stats;
1025 struct dp_stats_percpu local_stats;
1028 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
1031 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
1032 local_stats = *percpu_stats;
1033 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
1035 stats->n_hit += local_stats.n_hit;
1036 stats->n_missed += local_stats.n_missed;
1037 stats->n_lost += local_stats.n_lost;
1041 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1042 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1043 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1044 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1047 static struct genl_family dp_flow_genl_family = {
1048 .id = GENL_ID_GENERATE,
1049 .hdrsize = sizeof(struct ovs_header),
1050 .name = OVS_FLOW_FAMILY,
1051 .version = OVS_FLOW_VERSION,
1052 .maxattr = OVS_FLOW_ATTR_MAX,
1056 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
1057 .name = OVS_FLOW_MCGROUP
1060 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
1061 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
1063 const struct nlattr *a;
1064 struct nlattr *start;
1067 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
1071 nla_for_each_nested(a, attr, rem) {
1072 int type = nla_type(a);
1073 struct nlattr *st_sample;
1076 case OVS_SAMPLE_ATTR_PROBABILITY:
1077 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
1080 case OVS_SAMPLE_ATTR_ACTIONS:
1081 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
1084 err = actions_to_attr(nla_data(a), nla_len(a), skb);
1087 nla_nest_end(skb, st_sample);
1092 nla_nest_end(skb, start);
1096 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
1098 const struct nlattr *ovs_key = nla_data(a);
1099 int key_type = nla_type(ovs_key);
1100 struct nlattr *start;
1104 case OVS_KEY_ATTR_IPV4_TUNNEL:
1105 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1109 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1112 nla_nest_end(skb, start);
1115 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1123 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1125 const struct nlattr *a;
1128 nla_for_each_attr(a, attr, len, rem) {
1129 int type = nla_type(a);
1132 case OVS_ACTION_ATTR_SET:
1133 err = set_action_to_attr(a, skb);
1138 case OVS_ACTION_ATTR_SAMPLE:
1139 err = sample_action_to_attr(a, skb);
1144 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1153 /* Called with genl_lock. */
1154 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1155 struct sk_buff *skb, u32 portid,
1156 u32 seq, u32 flags, u8 cmd)
1158 const int skb_orig_len = skb->len;
1159 const struct sw_flow_actions *sf_acts;
1160 struct nlattr *start;
1161 struct ovs_flow_stats stats;
1162 struct ovs_header *ovs_header;
1168 sf_acts = rcu_dereference_protected(flow->sf_acts,
1169 lockdep_genl_is_held());
1171 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1175 ovs_header->dp_ifindex = get_dpifindex(dp);
1177 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1179 goto nla_put_failure;
1180 err = ovs_flow_to_nlattrs(&flow->key, skb);
1183 nla_nest_end(skb, nla);
1185 spin_lock_bh(&flow->lock);
1187 stats.n_packets = flow->packet_count;
1188 stats.n_bytes = flow->byte_count;
1189 tcp_flags = flow->tcp_flags;
1190 spin_unlock_bh(&flow->lock);
1193 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1194 goto nla_put_failure;
1196 if (stats.n_packets &&
1197 nla_put(skb, OVS_FLOW_ATTR_STATS,
1198 sizeof(struct ovs_flow_stats), &stats))
1199 goto nla_put_failure;
1202 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1203 goto nla_put_failure;
1205 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1206 * this is the first flow to be dumped into 'skb'. This is unusual for
1207 * Netlink but individual action lists can be longer than
1208 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1209 * The userspace caller can always fetch the actions separately if it
1210 * really wants them. (Most userspace callers in fact don't care.)
1212 * This can only fail for dump operations because the skb is always
1213 * properly sized for single flows.
1215 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1217 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1219 nla_nest_end(skb, start);
1224 nla_nest_cancel(skb, start);
1226 } else if (skb_orig_len) {
1231 return genlmsg_end(skb, ovs_header);
1236 genlmsg_cancel(skb, ovs_header);
1240 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1242 const struct sw_flow_actions *sf_acts;
1245 sf_acts = rcu_dereference_protected(flow->sf_acts,
1246 lockdep_genl_is_held());
1248 /* OVS_FLOW_ATTR_KEY */
1249 len = nla_total_size(FLOW_BUFSIZE);
1250 /* OVS_FLOW_ATTR_ACTIONS */
1251 len += nla_total_size(sf_acts->actions_len);
1252 /* OVS_FLOW_ATTR_STATS */
1253 len += nla_total_size(sizeof(struct ovs_flow_stats));
1254 /* OVS_FLOW_ATTR_TCP_FLAGS */
1255 len += nla_total_size(1);
1256 /* OVS_FLOW_ATTR_USED */
1257 len += nla_total_size(8);
1259 len += NLMSG_ALIGN(sizeof(struct ovs_header));
1261 return genlmsg_new(len, GFP_KERNEL);
1264 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1265 struct datapath *dp,
1266 u32 portid, u32 seq, u8 cmd)
1268 struct sk_buff *skb;
1271 skb = ovs_flow_cmd_alloc_info(flow);
1273 return ERR_PTR(-ENOMEM);
1275 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1280 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1282 struct nlattr **a = info->attrs;
1283 struct ovs_header *ovs_header = info->userhdr;
1284 struct sw_flow_key key;
1285 struct sw_flow *flow;
1286 struct sk_buff *reply;
1287 struct datapath *dp;
1288 struct flow_table *table;
1289 struct sw_flow_actions *acts = NULL;
1295 if (!a[OVS_FLOW_ATTR_KEY])
1297 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1301 /* Validate actions. */
1302 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1303 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1304 error = PTR_ERR(acts);
1308 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0, &acts);
1311 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1316 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1321 table = genl_dereference(dp->table);
1322 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1324 /* Bail out if we're not allowed to create a new flow. */
1326 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1329 /* Expand table, if necessary, to make room. */
1330 if (ovs_flow_tbl_need_to_expand(table)) {
1331 struct flow_table *new_table;
1333 new_table = ovs_flow_tbl_expand(table);
1334 if (!IS_ERR(new_table)) {
1335 rcu_assign_pointer(dp->table, new_table);
1336 ovs_flow_tbl_deferred_destroy(table);
1337 table = genl_dereference(dp->table);
1341 /* Allocate flow. */
1342 flow = ovs_flow_alloc();
1344 error = PTR_ERR(flow);
1349 rcu_assign_pointer(flow->sf_acts, acts);
1351 /* Put flow in bucket. */
1352 ovs_flow_tbl_insert(table, flow, &key, key_len);
1354 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1358 /* We found a matching flow. */
1359 struct sw_flow_actions *old_acts;
1361 /* Bail out if we're not allowed to modify an existing flow.
1362 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1363 * because Generic Netlink treats the latter as a dump
1364 * request. We also accept NLM_F_EXCL in case that bug ever
1368 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1369 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1372 /* Update actions. */
1373 old_acts = rcu_dereference_protected(flow->sf_acts,
1374 lockdep_genl_is_held());
1375 rcu_assign_pointer(flow->sf_acts, acts);
1376 ovs_flow_deferred_free_acts(old_acts);
1378 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1379 info->snd_seq, OVS_FLOW_CMD_NEW);
1382 if (a[OVS_FLOW_ATTR_CLEAR]) {
1383 spin_lock_bh(&flow->lock);
1385 spin_unlock_bh(&flow->lock);
1390 genl_notify(reply, genl_info_net(info), info->snd_portid,
1391 ovs_dp_flow_multicast_group.id, info->nlhdr,
1394 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1395 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1404 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1406 struct nlattr **a = info->attrs;
1407 struct ovs_header *ovs_header = info->userhdr;
1408 struct sw_flow_key key;
1409 struct sk_buff *reply;
1410 struct sw_flow *flow;
1411 struct datapath *dp;
1412 struct flow_table *table;
1416 if (!a[OVS_FLOW_ATTR_KEY])
1418 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1422 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1426 table = genl_dereference(dp->table);
1427 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1431 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1432 info->snd_seq, OVS_FLOW_CMD_NEW);
1434 return PTR_ERR(reply);
1436 return genlmsg_reply(reply, info);
1439 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1441 struct nlattr **a = info->attrs;
1442 struct ovs_header *ovs_header = info->userhdr;
1443 struct sw_flow_key key;
1444 struct sk_buff *reply;
1445 struct sw_flow *flow;
1446 struct datapath *dp;
1447 struct flow_table *table;
1451 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1455 if (!a[OVS_FLOW_ATTR_KEY])
1456 return flush_flows(dp);
1458 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1462 table = genl_dereference(dp->table);
1463 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1467 reply = ovs_flow_cmd_alloc_info(flow);
1471 ovs_flow_tbl_remove(table, flow);
1473 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1474 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1477 ovs_flow_deferred_free(flow);
1479 genl_notify(reply, genl_info_net(info), info->snd_portid,
1480 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1484 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1486 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1487 struct datapath *dp;
1488 struct flow_table *table;
1490 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1494 table = genl_dereference(dp->table);
1497 struct sw_flow *flow;
1500 bucket = cb->args[0];
1502 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1506 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1507 NETLINK_CB(cb->skb).portid,
1508 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1509 OVS_FLOW_CMD_NEW) < 0)
1512 cb->args[0] = bucket;
1518 static struct genl_ops dp_flow_genl_ops[] = {
1519 { .cmd = OVS_FLOW_CMD_NEW,
1520 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1521 .policy = flow_policy,
1522 .doit = ovs_flow_cmd_new_or_set
1524 { .cmd = OVS_FLOW_CMD_DEL,
1525 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1526 .policy = flow_policy,
1527 .doit = ovs_flow_cmd_del
1529 { .cmd = OVS_FLOW_CMD_GET,
1530 .flags = 0, /* OK for unprivileged users. */
1531 .policy = flow_policy,
1532 .doit = ovs_flow_cmd_get,
1533 .dumpit = ovs_flow_cmd_dump
1535 { .cmd = OVS_FLOW_CMD_SET,
1536 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1537 .policy = flow_policy,
1538 .doit = ovs_flow_cmd_new_or_set,
1542 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1543 #ifdef HAVE_NLA_NUL_STRING
1544 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1546 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1549 static struct genl_family dp_datapath_genl_family = {
1550 .id = GENL_ID_GENERATE,
1551 .hdrsize = sizeof(struct ovs_header),
1552 .name = OVS_DATAPATH_FAMILY,
1553 .version = OVS_DATAPATH_VERSION,
1554 .maxattr = OVS_DP_ATTR_MAX,
1558 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1559 .name = OVS_DATAPATH_MCGROUP
1562 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1563 u32 portid, u32 seq, u32 flags, u8 cmd)
1565 struct ovs_header *ovs_header;
1566 struct ovs_dp_stats dp_stats;
1569 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1574 ovs_header->dp_ifindex = get_dpifindex(dp);
1577 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1580 goto nla_put_failure;
1582 get_dp_stats(dp, &dp_stats);
1583 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1584 goto nla_put_failure;
1586 return genlmsg_end(skb, ovs_header);
1589 genlmsg_cancel(skb, ovs_header);
1594 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1597 struct sk_buff *skb;
1600 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1602 return ERR_PTR(-ENOMEM);
1604 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1607 return ERR_PTR(retval);
1612 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1614 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1617 /* Called with genl_mutex and optionally with RTNL lock also. */
1618 static struct datapath *lookup_datapath(struct net *net,
1619 struct ovs_header *ovs_header,
1620 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1622 struct datapath *dp;
1624 if (!a[OVS_DP_ATTR_NAME])
1625 dp = get_dp(net, ovs_header->dp_ifindex);
1627 struct vport *vport;
1630 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1631 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1634 return dp ? dp : ERR_PTR(-ENODEV);
1637 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1639 struct nlattr **a = info->attrs;
1640 struct vport_parms parms;
1641 struct sk_buff *reply;
1642 struct datapath *dp;
1643 struct vport *vport;
1644 struct ovs_net *ovs_net;
1648 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1651 err = ovs_dp_cmd_validate(a);
1658 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1660 goto err_unlock_rtnl;
1662 /* Initialize kobject for bridge. This will be added as
1663 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1664 dp->ifobj.kset = NULL;
1665 kobject_init(&dp->ifobj, &dp_ktype);
1667 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1669 /* Allocate table. */
1671 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1675 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1676 if (!dp->stats_percpu) {
1678 goto err_destroy_table;
1681 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1685 goto err_destroy_percpu;
1688 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1689 INIT_HLIST_HEAD(&dp->ports[i]);
1691 /* Set up our datapath device. */
1692 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1693 parms.type = OVS_VPORT_TYPE_INTERNAL;
1694 parms.options = NULL;
1696 parms.port_no = OVSP_LOCAL;
1697 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1699 vport = new_vport(&parms);
1700 if (IS_ERR(vport)) {
1701 err = PTR_ERR(vport);
1705 goto err_destroy_ports_array;
1708 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1709 info->snd_seq, OVS_DP_CMD_NEW);
1710 err = PTR_ERR(reply);
1712 goto err_destroy_local_port;
1714 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1715 list_add_tail(&dp->list_node, &ovs_net->dps);
1716 ovs_dp_sysfs_add_dp(dp);
1720 genl_notify(reply, genl_info_net(info), info->snd_portid,
1721 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1725 err_destroy_local_port:
1726 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1727 err_destroy_ports_array:
1730 free_percpu(dp->stats_percpu);
1732 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1734 release_net(ovs_dp_get_net(dp));
1742 /* Called with genl_mutex. */
1743 static void __dp_destroy(struct datapath *dp)
1749 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1750 struct vport *vport;
1751 struct hlist_node *node, *n;
1753 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1754 if (vport->port_no != OVSP_LOCAL)
1755 ovs_dp_detach_port(vport);
1758 ovs_dp_sysfs_del_dp(dp);
1759 list_del(&dp->list_node);
1760 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1762 /* rtnl_unlock() will wait until all the references to devices that
1763 * are pending unregistration have been dropped. We do it here to
1764 * ensure that any internal devices (which contain DP pointers) are
1765 * fully destroyed before freeing the datapath.
1769 call_rcu(&dp->rcu, destroy_dp_rcu);
1772 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1774 struct sk_buff *reply;
1775 struct datapath *dp;
1778 err = ovs_dp_cmd_validate(info->attrs);
1782 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1787 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1788 info->snd_seq, OVS_DP_CMD_DEL);
1789 err = PTR_ERR(reply);
1795 genl_notify(reply, genl_info_net(info), info->snd_portid,
1796 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1802 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1804 struct sk_buff *reply;
1805 struct datapath *dp;
1808 err = ovs_dp_cmd_validate(info->attrs);
1812 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1816 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1817 info->snd_seq, OVS_DP_CMD_NEW);
1818 if (IS_ERR(reply)) {
1819 err = PTR_ERR(reply);
1820 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1821 ovs_dp_datapath_multicast_group.id, err);
1825 genl_notify(reply, genl_info_net(info), info->snd_portid,
1826 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1832 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1834 struct sk_buff *reply;
1835 struct datapath *dp;
1838 err = ovs_dp_cmd_validate(info->attrs);
1842 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1846 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1847 info->snd_seq, OVS_DP_CMD_NEW);
1849 return PTR_ERR(reply);
1851 return genlmsg_reply(reply, info);
1854 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1856 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1857 struct datapath *dp;
1858 int skip = cb->args[0];
1861 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1863 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1864 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1865 OVS_DP_CMD_NEW) < 0)
1875 static struct genl_ops dp_datapath_genl_ops[] = {
1876 { .cmd = OVS_DP_CMD_NEW,
1877 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1878 .policy = datapath_policy,
1879 .doit = ovs_dp_cmd_new
1881 { .cmd = OVS_DP_CMD_DEL,
1882 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1883 .policy = datapath_policy,
1884 .doit = ovs_dp_cmd_del
1886 { .cmd = OVS_DP_CMD_GET,
1887 .flags = 0, /* OK for unprivileged users. */
1888 .policy = datapath_policy,
1889 .doit = ovs_dp_cmd_get,
1890 .dumpit = ovs_dp_cmd_dump
1892 { .cmd = OVS_DP_CMD_SET,
1893 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1894 .policy = datapath_policy,
1895 .doit = ovs_dp_cmd_set,
1899 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1900 #ifdef HAVE_NLA_NUL_STRING
1901 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1902 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1903 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1905 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1906 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1908 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1909 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1910 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1911 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1914 static struct genl_family dp_vport_genl_family = {
1915 .id = GENL_ID_GENERATE,
1916 .hdrsize = sizeof(struct ovs_header),
1917 .name = OVS_VPORT_FAMILY,
1918 .version = OVS_VPORT_VERSION,
1919 .maxattr = OVS_VPORT_ATTR_MAX,
1923 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1924 .name = OVS_VPORT_MCGROUP
1927 /* Called with RTNL lock or RCU read lock. */
1928 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1929 u32 portid, u32 seq, u32 flags, u8 cmd)
1931 struct ovs_header *ovs_header;
1932 struct ovs_vport_stats vport_stats;
1935 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1940 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1942 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1943 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1944 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1945 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1946 goto nla_put_failure;
1948 ovs_vport_get_stats(vport, &vport_stats);
1949 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1951 goto nla_put_failure;
1953 if (nla_put(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1954 vport->ops->get_addr(vport)))
1955 goto nla_put_failure;
1957 err = ovs_vport_get_options(vport, skb);
1958 if (err == -EMSGSIZE)
1961 return genlmsg_end(skb, ovs_header);
1966 genlmsg_cancel(skb, ovs_header);
1970 /* Called with RTNL lock or RCU read lock. */
1971 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1974 struct sk_buff *skb;
1977 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1979 return ERR_PTR(-ENOMEM);
1981 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1984 return ERR_PTR(retval);
1989 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1991 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1994 /* Called with RTNL lock or RCU read lock. */
1995 static struct vport *lookup_vport(struct net *net,
1996 struct ovs_header *ovs_header,
1997 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1999 struct datapath *dp;
2000 struct vport *vport;
2002 if (a[OVS_VPORT_ATTR_NAME]) {
2003 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2005 return ERR_PTR(-ENODEV);
2006 if (ovs_header->dp_ifindex &&
2007 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2008 return ERR_PTR(-ENODEV);
2010 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2011 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2013 if (port_no >= DP_MAX_PORTS)
2014 return ERR_PTR(-EFBIG);
2016 dp = get_dp(net, ovs_header->dp_ifindex);
2018 return ERR_PTR(-ENODEV);
2020 vport = ovs_vport_rtnl_rcu(dp, port_no);
2022 return ERR_PTR(-ENOENT);
2025 return ERR_PTR(-EINVAL);
2028 /* Called with RTNL lock. */
2029 static int change_vport(struct vport *vport,
2030 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2034 if (a[OVS_VPORT_ATTR_STATS])
2035 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2037 if (a[OVS_VPORT_ATTR_ADDRESS])
2038 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
2043 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2045 struct nlattr **a = info->attrs;
2046 struct ovs_header *ovs_header = info->userhdr;
2047 struct vport_parms parms;
2048 struct sk_buff *reply;
2049 struct vport *vport;
2050 struct datapath *dp;
2055 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2056 !a[OVS_VPORT_ATTR_UPCALL_PID])
2059 err = ovs_vport_cmd_validate(a);
2064 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2069 if (a[OVS_VPORT_ATTR_PORT_NO]) {
2070 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2073 if (port_no >= DP_MAX_PORTS)
2076 vport = ovs_vport_rtnl(dp, port_no);
2081 for (port_no = 1; ; port_no++) {
2082 if (port_no >= DP_MAX_PORTS) {
2086 vport = ovs_vport_rtnl(dp, port_no);
2092 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2093 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2094 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2096 parms.port_no = port_no;
2097 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2099 vport = new_vport(&parms);
2100 err = PTR_ERR(vport);
2104 ovs_dp_sysfs_add_if(vport);
2106 err = change_vport(vport, a);
2108 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2112 err = PTR_ERR(reply);
2115 ovs_dp_detach_port(vport);
2118 genl_notify(reply, genl_info_net(info), info->snd_portid,
2119 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2127 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2129 struct nlattr **a = info->attrs;
2130 struct sk_buff *reply;
2131 struct vport *vport;
2134 err = ovs_vport_cmd_validate(a);
2139 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2140 err = PTR_ERR(vport);
2145 if (a[OVS_VPORT_ATTR_TYPE] &&
2146 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2149 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2150 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2152 err = change_vport(vport, a);
2155 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
2156 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2158 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2159 info->snd_seq, OVS_VPORT_CMD_NEW);
2160 if (IS_ERR(reply)) {
2161 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
2162 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
2166 genl_notify(reply, genl_info_net(info), info->snd_portid,
2167 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2175 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2177 struct nlattr **a = info->attrs;
2178 struct sk_buff *reply;
2179 struct vport *vport;
2182 err = ovs_vport_cmd_validate(a);
2187 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2188 err = PTR_ERR(vport);
2192 if (vport->port_no == OVSP_LOCAL) {
2197 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2198 info->snd_seq, OVS_VPORT_CMD_DEL);
2199 err = PTR_ERR(reply);
2203 ovs_dp_detach_port(vport);
2205 genl_notify(reply, genl_info_net(info), info->snd_portid,
2206 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2214 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2216 struct nlattr **a = info->attrs;
2217 struct ovs_header *ovs_header = info->userhdr;
2218 struct sk_buff *reply;
2219 struct vport *vport;
2222 err = ovs_vport_cmd_validate(a);
2227 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2228 err = PTR_ERR(vport);
2232 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2233 info->snd_seq, OVS_VPORT_CMD_NEW);
2234 err = PTR_ERR(reply);
2240 return genlmsg_reply(reply, info);
2248 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2250 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2251 struct datapath *dp;
2252 int bucket = cb->args[0], skip = cb->args[1];
2255 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2260 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2261 struct vport *vport;
2262 struct hlist_node *n;
2265 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2267 ovs_vport_cmd_fill_info(vport, skb,
2268 NETLINK_CB(cb->skb).portid,
2271 OVS_VPORT_CMD_NEW) < 0)
2287 static struct genl_ops dp_vport_genl_ops[] = {
2288 { .cmd = OVS_VPORT_CMD_NEW,
2289 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2290 .policy = vport_policy,
2291 .doit = ovs_vport_cmd_new
2293 { .cmd = OVS_VPORT_CMD_DEL,
2294 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2295 .policy = vport_policy,
2296 .doit = ovs_vport_cmd_del
2298 { .cmd = OVS_VPORT_CMD_GET,
2299 .flags = 0, /* OK for unprivileged users. */
2300 .policy = vport_policy,
2301 .doit = ovs_vport_cmd_get,
2302 .dumpit = ovs_vport_cmd_dump
2304 { .cmd = OVS_VPORT_CMD_SET,
2305 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2306 .policy = vport_policy,
2307 .doit = ovs_vport_cmd_set,
2311 struct genl_family_and_ops {
2312 struct genl_family *family;
2313 struct genl_ops *ops;
2315 struct genl_multicast_group *group;
2318 static const struct genl_family_and_ops dp_genl_families[] = {
2319 { &dp_datapath_genl_family,
2320 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2321 &ovs_dp_datapath_multicast_group },
2322 { &dp_vport_genl_family,
2323 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2324 &ovs_dp_vport_multicast_group },
2325 { &dp_flow_genl_family,
2326 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2327 &ovs_dp_flow_multicast_group },
2328 { &dp_packet_genl_family,
2329 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2333 static void dp_unregister_genl(int n_families)
2337 for (i = 0; i < n_families; i++)
2338 genl_unregister_family(dp_genl_families[i].family);
2341 static int dp_register_genl(void)
2348 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2349 const struct genl_family_and_ops *f = &dp_genl_families[i];
2351 err = genl_register_family_with_ops(f->family, f->ops,
2358 err = genl_register_mc_group(f->family, f->group);
2367 dp_unregister_genl(n_registered);
2371 static int __rehash_flow_table(void *dummy)
2373 struct datapath *dp;
2378 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2380 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2381 struct flow_table *old_table = genl_dereference(dp->table);
2382 struct flow_table *new_table;
2384 new_table = ovs_flow_tbl_rehash(old_table);
2385 if (!IS_ERR(new_table)) {
2386 rcu_assign_pointer(dp->table, new_table);
2387 ovs_flow_tbl_deferred_destroy(old_table);
2395 static void rehash_flow_table(struct work_struct *work)
2397 genl_exec(__rehash_flow_table, NULL);
2398 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2401 static int dp_destroy_all(void *data)
2403 struct datapath *dp, *dp_next;
2404 struct ovs_net *ovs_net = data;
2406 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2412 static int __net_init ovs_init_net(struct net *net)
2414 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2416 INIT_LIST_HEAD(&ovs_net->dps);
2420 static void __net_exit ovs_exit_net(struct net *net)
2422 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2424 genl_exec(dp_destroy_all, ovs_net);
2427 static struct pernet_operations ovs_net_ops = {
2428 .init = ovs_init_net,
2429 .exit = ovs_exit_net,
2431 .size = sizeof(struct ovs_net),
2434 static int __init dp_init(void)
2436 struct sk_buff *dummy_skb;
2439 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2441 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2444 err = genl_exec_init();
2448 err = ovs_workqueues_init();
2450 goto error_genl_exec;
2452 err = ovs_tnl_init();
2456 err = ovs_flow_init();
2458 goto error_tnl_exit;
2460 err = ovs_vport_init();
2462 goto error_flow_exit;
2464 err = register_pernet_device(&ovs_net_ops);
2466 goto error_vport_exit;
2468 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2470 goto error_netns_exit;
2472 err = dp_register_genl();
2474 goto error_unreg_notifier;
2476 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2480 error_unreg_notifier:
2481 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2483 unregister_pernet_device(&ovs_net_ops);
2491 ovs_workqueues_exit();
2498 static void dp_cleanup(void)
2500 cancel_delayed_work_sync(&rehash_flow_wq);
2501 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2502 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2503 unregister_pernet_device(&ovs_net_ops);
2508 ovs_workqueues_exit();
2512 module_init(dp_init);
2513 module_exit(dp_cleanup);
2515 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2516 MODULE_LICENSE("GPL");
2517 MODULE_VERSION(VERSION);