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;
77 * Writes to device state (add/remove datapath, port, set operations on vports,
78 * etc.) are protected by RTNL.
80 * Writes to other state (flow table modifications, set miscellaneous datapath
81 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
84 * Reads are protected by RCU.
86 * There are a few special cases (mostly stats) that have their own
87 * synchronization but they nest under all of above and don't interact with
91 static struct vport *new_vport(const struct vport_parms *);
92 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
93 const struct dp_upcall_info *);
94 static int queue_userspace_packet(struct net *, int dp_ifindex,
96 const struct dp_upcall_info *);
98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
99 static struct datapath *get_dp(struct net *net, int dp_ifindex)
101 struct datapath *dp = NULL;
102 struct net_device *dev;
105 dev = dev_get_by_index_rcu(net, dp_ifindex);
107 struct vport *vport = ovs_internal_dev_get_vport(dev);
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 const char *ovs_dp_name(const struct datapath *dp)
119 struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
120 return vport->ops->get_name(vport);
123 static int get_dpifindex(struct datapath *dp)
130 local = ovs_vport_rcu(dp, OVSP_LOCAL);
132 ifindex = local->ops->get_ifindex(local);
141 static void destroy_dp_rcu(struct rcu_head *rcu)
143 struct datapath *dp = container_of(rcu, struct datapath, rcu);
145 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
146 free_percpu(dp->stats_percpu);
147 release_net(ovs_dp_get_net(dp));
152 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
155 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
158 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
161 struct hlist_node *n;
162 struct hlist_head *head;
164 head = vport_hash_bucket(dp, port_no);
165 hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
166 if (vport->port_no == port_no)
172 /* Called with RTNL lock and genl_lock. */
173 static struct vport *new_vport(const struct vport_parms *parms)
177 vport = ovs_vport_add(parms);
178 if (!IS_ERR(vport)) {
179 struct datapath *dp = parms->dp;
180 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
182 hlist_add_head_rcu(&vport->dp_hash_node, head);
187 /* Called with RTNL lock. */
188 void ovs_dp_detach_port(struct vport *p)
192 /* First drop references to device. */
193 hlist_del_rcu(&p->dp_hash_node);
195 /* Then destroy it. */
199 /* Must be called with rcu_read_lock. */
200 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
202 struct datapath *dp = p->dp;
203 struct sw_flow *flow;
204 struct dp_stats_percpu *stats;
208 stats = this_cpu_ptr(dp->stats_percpu);
210 if (!OVS_CB(skb)->flow) {
211 struct sw_flow_key key;
214 /* Extract flow from 'skb' into 'key'. */
215 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
216 if (unlikely(error)) {
222 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
224 if (unlikely(!flow)) {
225 struct dp_upcall_info upcall;
227 upcall.cmd = OVS_PACKET_CMD_MISS;
229 upcall.userdata = NULL;
230 upcall.portid = p->upcall_portid;
231 ovs_dp_upcall(dp, skb, &upcall);
233 stats_counter = &stats->n_missed;
237 OVS_CB(skb)->flow = flow;
240 stats_counter = &stats->n_hit;
241 ovs_flow_used(OVS_CB(skb)->flow, skb);
242 ovs_execute_actions(dp, skb);
245 /* Update datapath statistics. */
246 u64_stats_update_begin(&stats->sync);
248 u64_stats_update_end(&stats->sync);
251 static struct genl_family dp_packet_genl_family = {
252 .id = GENL_ID_GENERATE,
253 .hdrsize = sizeof(struct ovs_header),
254 .name = OVS_PACKET_FAMILY,
255 .version = OVS_PACKET_VERSION,
256 .maxattr = OVS_PACKET_ATTR_MAX,
260 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
261 const struct dp_upcall_info *upcall_info)
263 struct dp_stats_percpu *stats;
267 if (upcall_info->portid == 0) {
272 dp_ifindex = get_dpifindex(dp);
278 forward_ip_summed(skb, true);
280 if (!skb_is_gso(skb))
281 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
283 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
290 stats = this_cpu_ptr(dp->stats_percpu);
292 u64_stats_update_begin(&stats->sync);
294 u64_stats_update_end(&stats->sync);
299 static int queue_gso_packets(struct net *net, int dp_ifindex,
301 const struct dp_upcall_info *upcall_info)
303 unsigned short gso_type = skb_shinfo(skb)->gso_type;
304 struct dp_upcall_info later_info;
305 struct sw_flow_key later_key;
306 struct sk_buff *segs, *nskb;
309 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
311 return PTR_ERR(segs);
313 /* Queue all of the segments. */
316 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
320 if (skb == segs && gso_type & SKB_GSO_UDP) {
321 /* The initial flow key extracted by ovs_flow_extract()
322 * in this case is for a first fragment, so we need to
323 * properly mark later fragments.
325 later_key = *upcall_info->key;
326 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
328 later_info = *upcall_info;
329 later_info.key = &later_key;
330 upcall_info = &later_info;
332 } while ((skb = skb->next));
334 /* Free all of the segments. */
342 } while ((skb = nskb));
346 static int queue_userspace_packet(struct net *net, int dp_ifindex,
348 const struct dp_upcall_info *upcall_info)
350 struct ovs_header *upcall;
351 struct sk_buff *nskb = NULL;
352 struct sk_buff *user_skb; /* to be queued to userspace */
357 if (vlan_tx_tag_present(skb)) {
358 nskb = skb_clone(skb, GFP_ATOMIC);
362 err = vlan_deaccel_tag(nskb);
369 if (nla_attr_size(skb->len) > USHRT_MAX) {
374 len = sizeof(struct ovs_header);
375 len += nla_total_size(skb->len);
376 len += nla_total_size(FLOW_BUFSIZE);
377 if (upcall_info->userdata)
378 len += NLA_ALIGN(upcall_info->userdata->nla_len);
380 user_skb = genlmsg_new(len, GFP_ATOMIC);
386 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
387 0, upcall_info->cmd);
388 upcall->dp_ifindex = dp_ifindex;
390 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
391 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
392 nla_nest_end(user_skb, nla);
394 if (upcall_info->userdata)
395 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
396 nla_len(upcall_info->userdata),
397 nla_data(upcall_info->userdata));
399 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
401 skb_copy_and_csum_dev(skb, nla_data(nla));
403 genlmsg_end(user_skb, upcall);
404 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
411 /* Called with genl_mutex. */
412 static int flush_flows(struct datapath *dp)
414 struct flow_table *old_table;
415 struct flow_table *new_table;
417 old_table = genl_dereference(dp->table);
418 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
422 rcu_assign_pointer(dp->table, new_table);
424 ovs_flow_tbl_deferred_destroy(old_table);
428 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
431 struct sw_flow_actions *acts;
433 int req_size = NLA_ALIGN(attr_len);
434 int next_offset = offsetof(struct sw_flow_actions, actions) +
437 if (req_size <= (ksize(*sfa) - next_offset))
440 new_acts_size = ksize(*sfa) * 2;
442 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
443 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
444 return ERR_PTR(-EMSGSIZE);
445 new_acts_size = MAX_ACTIONS_BUFSIZE;
448 acts = ovs_flow_actions_alloc(new_acts_size);
452 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
453 acts->actions_len = (*sfa)->actions_len;
458 (*sfa)->actions_len += req_size;
459 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
462 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
466 a = reserve_sfa_size(sfa, nla_attr_size(len));
470 a->nla_type = attrtype;
471 a->nla_len = nla_attr_size(len);
474 memcpy(nla_data(a), data, len);
475 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
480 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
482 int used = (*sfa)->actions_len;
485 err = add_action(sfa, attrtype, NULL, 0);
492 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
494 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
496 a->nla_len = sfa->actions_len - st_offset;
499 static int validate_and_copy_actions(const struct nlattr *attr,
500 const struct sw_flow_key *key, int depth,
501 struct sw_flow_actions **sfa);
503 static int validate_and_copy_sample(const struct nlattr *attr,
504 const struct sw_flow_key *key, int depth,
505 struct sw_flow_actions **sfa)
507 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
508 const struct nlattr *probability, *actions;
509 const struct nlattr *a;
510 int rem, start, err, st_acts;
512 memset(attrs, 0, sizeof(attrs));
513 nla_for_each_nested(a, attr, rem) {
514 int type = nla_type(a);
515 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
522 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
523 if (!probability || nla_len(probability) != sizeof(u32))
526 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
527 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
530 /* validation done, copy sample action. */
531 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
534 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
537 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
541 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
545 add_nested_action_end(*sfa, st_acts);
546 add_nested_action_end(*sfa, start);
551 static int validate_tp_port(const struct sw_flow_key *flow_key)
553 if (flow_key->eth.type == htons(ETH_P_IP)) {
554 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
556 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
557 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
564 static int validate_and_copy_set_tun(const struct nlattr *attr,
565 struct sw_flow_actions **sfa)
567 struct ovs_key_ipv4_tunnel tun_key;
570 err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
574 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
578 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
579 add_nested_action_end(*sfa, start);
584 static int validate_set(const struct nlattr *a,
585 const struct sw_flow_key *flow_key,
586 struct sw_flow_actions **sfa,
589 const struct nlattr *ovs_key = nla_data(a);
590 int key_type = nla_type(ovs_key);
592 /* There can be only one key in a action */
593 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
596 if (key_type > OVS_KEY_ATTR_MAX ||
597 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
598 ovs_key_lens[key_type] != -1))
602 const struct ovs_key_ipv4 *ipv4_key;
603 const struct ovs_key_ipv6 *ipv6_key;
606 case OVS_KEY_ATTR_PRIORITY:
607 case OVS_KEY_ATTR_ETHERNET:
610 case OVS_KEY_ATTR_SKB_MARK:
611 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
612 if (nla_get_u32(ovs_key) != 0)
617 case OVS_KEY_ATTR_TUNNEL:
619 err = validate_and_copy_set_tun(a, sfa);
624 case OVS_KEY_ATTR_IPV4:
625 if (flow_key->eth.type != htons(ETH_P_IP))
628 if (!flow_key->ip.proto)
631 ipv4_key = nla_data(ovs_key);
632 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
635 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
640 case OVS_KEY_ATTR_IPV6:
641 if (flow_key->eth.type != htons(ETH_P_IPV6))
644 if (!flow_key->ip.proto)
647 ipv6_key = nla_data(ovs_key);
648 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
651 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
654 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
659 case OVS_KEY_ATTR_TCP:
660 if (flow_key->ip.proto != IPPROTO_TCP)
663 return validate_tp_port(flow_key);
665 case OVS_KEY_ATTR_UDP:
666 if (flow_key->ip.proto != IPPROTO_UDP)
669 return validate_tp_port(flow_key);
678 static int validate_userspace(const struct nlattr *attr)
680 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
681 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
682 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
684 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
687 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
688 attr, userspace_policy);
692 if (!a[OVS_USERSPACE_ATTR_PID] ||
693 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
699 static int copy_action(const struct nlattr *from,
700 struct sw_flow_actions **sfa)
702 int totlen = NLA_ALIGN(from->nla_len);
705 to = reserve_sfa_size(sfa, from->nla_len);
709 memcpy(to, from, totlen);
713 static int validate_and_copy_actions(const struct nlattr *attr,
714 const struct sw_flow_key *key,
716 struct sw_flow_actions **sfa)
718 const struct nlattr *a;
721 if (depth >= SAMPLE_ACTION_DEPTH)
724 nla_for_each_nested(a, attr, rem) {
725 /* Expected argument lengths, (u32)-1 for variable length. */
726 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
727 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
728 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
729 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
730 [OVS_ACTION_ATTR_POP_VLAN] = 0,
731 [OVS_ACTION_ATTR_SET] = (u32)-1,
732 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
734 const struct ovs_action_push_vlan *vlan;
735 int type = nla_type(a);
738 if (type > OVS_ACTION_ATTR_MAX ||
739 (action_lens[type] != nla_len(a) &&
740 action_lens[type] != (u32)-1))
745 case OVS_ACTION_ATTR_UNSPEC:
748 case OVS_ACTION_ATTR_USERSPACE:
749 err = validate_userspace(a);
754 case OVS_ACTION_ATTR_OUTPUT:
755 if (nla_get_u32(a) >= DP_MAX_PORTS)
760 case OVS_ACTION_ATTR_POP_VLAN:
763 case OVS_ACTION_ATTR_PUSH_VLAN:
765 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
767 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
771 case OVS_ACTION_ATTR_SET:
772 err = validate_set(a, key, sfa, &skip_copy);
777 case OVS_ACTION_ATTR_SAMPLE:
778 err = validate_and_copy_sample(a, key, depth, sfa);
788 err = copy_action(a, sfa);
800 static void clear_stats(struct sw_flow *flow)
804 flow->packet_count = 0;
805 flow->byte_count = 0;
808 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
810 struct ovs_header *ovs_header = info->userhdr;
811 struct nlattr **a = info->attrs;
812 struct sw_flow_actions *acts;
813 struct sk_buff *packet;
814 struct sw_flow *flow;
822 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
823 !a[OVS_PACKET_ATTR_ACTIONS] ||
824 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
827 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
828 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
832 skb_reserve(packet, NET_IP_ALIGN);
834 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
836 skb_reset_mac_header(packet);
837 eth = eth_hdr(packet);
839 /* Normally, setting the skb 'protocol' field would be handled by a
840 * call to eth_type_trans(), but it assumes there's a sending
841 * device, which we may not have. */
842 if (ntohs(eth->h_proto) >= 1536)
843 packet->protocol = eth->h_proto;
845 packet->protocol = htons(ETH_P_802_2);
847 /* Build an sw_flow for sending this packet. */
848 flow = ovs_flow_alloc();
853 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
857 err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
860 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
865 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
866 rcu_assign_pointer(flow->sf_acts, acts);
870 OVS_CB(packet)->flow = flow;
871 packet->priority = flow->key.phy.priority;
872 skb_set_mark(packet, flow->key.phy.skb_mark);
875 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
881 err = ovs_execute_actions(dp, packet);
898 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
899 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
900 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
901 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
904 static struct genl_ops dp_packet_genl_ops[] = {
905 { .cmd = OVS_PACKET_CMD_EXECUTE,
906 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
907 .policy = packet_policy,
908 .doit = ovs_packet_cmd_execute
912 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
915 struct flow_table *table = genl_dereference(dp->table);
917 stats->n_flows = ovs_flow_tbl_count(table);
919 stats->n_hit = stats->n_missed = stats->n_lost = 0;
920 for_each_possible_cpu(i) {
921 const struct dp_stats_percpu *percpu_stats;
922 struct dp_stats_percpu local_stats;
925 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
928 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
929 local_stats = *percpu_stats;
930 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
932 stats->n_hit += local_stats.n_hit;
933 stats->n_missed += local_stats.n_missed;
934 stats->n_lost += local_stats.n_lost;
938 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
939 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
940 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
941 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
944 static struct genl_family dp_flow_genl_family = {
945 .id = GENL_ID_GENERATE,
946 .hdrsize = sizeof(struct ovs_header),
947 .name = OVS_FLOW_FAMILY,
948 .version = OVS_FLOW_VERSION,
949 .maxattr = OVS_FLOW_ATTR_MAX,
953 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
954 .name = OVS_FLOW_MCGROUP
957 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
958 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
960 const struct nlattr *a;
961 struct nlattr *start;
964 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
968 nla_for_each_nested(a, attr, rem) {
969 int type = nla_type(a);
970 struct nlattr *st_sample;
973 case OVS_SAMPLE_ATTR_PROBABILITY:
974 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
977 case OVS_SAMPLE_ATTR_ACTIONS:
978 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
981 err = actions_to_attr(nla_data(a), nla_len(a), skb);
984 nla_nest_end(skb, st_sample);
989 nla_nest_end(skb, start);
993 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
995 const struct nlattr *ovs_key = nla_data(a);
996 int key_type = nla_type(ovs_key);
997 struct nlattr *start;
1001 case OVS_KEY_ATTR_IPV4_TUNNEL:
1002 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1006 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1009 nla_nest_end(skb, start);
1012 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1020 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1022 const struct nlattr *a;
1025 nla_for_each_attr(a, attr, len, rem) {
1026 int type = nla_type(a);
1029 case OVS_ACTION_ATTR_SET:
1030 err = set_action_to_attr(a, skb);
1035 case OVS_ACTION_ATTR_SAMPLE:
1036 err = sample_action_to_attr(a, skb);
1041 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1050 /* Called with genl_lock. */
1051 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1052 struct sk_buff *skb, u32 portid,
1053 u32 seq, u32 flags, u8 cmd)
1055 const int skb_orig_len = skb->len;
1056 const struct sw_flow_actions *sf_acts;
1057 struct nlattr *start;
1058 struct ovs_flow_stats stats;
1059 struct ovs_header *ovs_header;
1065 sf_acts = rcu_dereference_protected(flow->sf_acts,
1066 lockdep_genl_is_held());
1068 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1072 ovs_header->dp_ifindex = get_dpifindex(dp);
1074 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1076 goto nla_put_failure;
1077 err = ovs_flow_to_nlattrs(&flow->key, skb);
1080 nla_nest_end(skb, nla);
1082 spin_lock_bh(&flow->lock);
1084 stats.n_packets = flow->packet_count;
1085 stats.n_bytes = flow->byte_count;
1086 tcp_flags = flow->tcp_flags;
1087 spin_unlock_bh(&flow->lock);
1090 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1091 goto nla_put_failure;
1093 if (stats.n_packets &&
1094 nla_put(skb, OVS_FLOW_ATTR_STATS,
1095 sizeof(struct ovs_flow_stats), &stats))
1096 goto nla_put_failure;
1099 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1100 goto nla_put_failure;
1102 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1103 * this is the first flow to be dumped into 'skb'. This is unusual for
1104 * Netlink but individual action lists can be longer than
1105 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1106 * The userspace caller can always fetch the actions separately if it
1107 * really wants them. (Most userspace callers in fact don't care.)
1109 * This can only fail for dump operations because the skb is always
1110 * properly sized for single flows.
1112 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1114 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1116 nla_nest_end(skb, start);
1121 nla_nest_cancel(skb, start);
1123 } else if (skb_orig_len)
1124 goto nla_put_failure;
1126 return genlmsg_end(skb, ovs_header);
1131 genlmsg_cancel(skb, ovs_header);
1135 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1137 const struct sw_flow_actions *sf_acts;
1140 sf_acts = rcu_dereference_protected(flow->sf_acts,
1141 lockdep_genl_is_held());
1143 /* OVS_FLOW_ATTR_KEY */
1144 len = nla_total_size(FLOW_BUFSIZE);
1145 /* OVS_FLOW_ATTR_ACTIONS */
1146 len += nla_total_size(sf_acts->actions_len);
1147 /* OVS_FLOW_ATTR_STATS */
1148 len += nla_total_size(sizeof(struct ovs_flow_stats));
1149 /* OVS_FLOW_ATTR_TCP_FLAGS */
1150 len += nla_total_size(1);
1151 /* OVS_FLOW_ATTR_USED */
1152 len += nla_total_size(8);
1154 len += NLMSG_ALIGN(sizeof(struct ovs_header));
1156 return genlmsg_new(len, GFP_KERNEL);
1159 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1160 struct datapath *dp,
1161 u32 portid, u32 seq, u8 cmd)
1163 struct sk_buff *skb;
1166 skb = ovs_flow_cmd_alloc_info(flow);
1168 return ERR_PTR(-ENOMEM);
1170 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1175 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1177 struct nlattr **a = info->attrs;
1178 struct ovs_header *ovs_header = info->userhdr;
1179 struct sw_flow_key key;
1180 struct sw_flow *flow;
1181 struct sk_buff *reply;
1182 struct datapath *dp;
1183 struct flow_table *table;
1184 struct sw_flow_actions *acts = NULL;
1190 if (!a[OVS_FLOW_ATTR_KEY])
1192 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1196 /* Validate actions. */
1197 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1198 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1199 error = PTR_ERR(acts);
1203 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0, &acts);
1206 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1211 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1216 table = genl_dereference(dp->table);
1217 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1219 /* Bail out if we're not allowed to create a new flow. */
1221 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1224 /* Expand table, if necessary, to make room. */
1225 if (ovs_flow_tbl_need_to_expand(table)) {
1226 struct flow_table *new_table;
1228 new_table = ovs_flow_tbl_expand(table);
1229 if (!IS_ERR(new_table)) {
1230 rcu_assign_pointer(dp->table, new_table);
1231 ovs_flow_tbl_deferred_destroy(table);
1232 table = genl_dereference(dp->table);
1236 /* Allocate flow. */
1237 flow = ovs_flow_alloc();
1239 error = PTR_ERR(flow);
1244 rcu_assign_pointer(flow->sf_acts, acts);
1246 /* Put flow in bucket. */
1247 ovs_flow_tbl_insert(table, flow, &key, key_len);
1249 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1253 /* We found a matching flow. */
1254 struct sw_flow_actions *old_acts;
1256 /* Bail out if we're not allowed to modify an existing flow.
1257 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1258 * because Generic Netlink treats the latter as a dump
1259 * request. We also accept NLM_F_EXCL in case that bug ever
1263 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1264 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1267 /* Update actions. */
1268 old_acts = rcu_dereference_protected(flow->sf_acts,
1269 lockdep_genl_is_held());
1270 rcu_assign_pointer(flow->sf_acts, acts);
1271 ovs_flow_deferred_free_acts(old_acts);
1273 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1274 info->snd_seq, OVS_FLOW_CMD_NEW);
1277 if (a[OVS_FLOW_ATTR_CLEAR]) {
1278 spin_lock_bh(&flow->lock);
1280 spin_unlock_bh(&flow->lock);
1285 genl_notify(reply, genl_info_net(info), info->snd_portid,
1286 ovs_dp_flow_multicast_group.id, info->nlhdr,
1289 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1290 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1299 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1301 struct nlattr **a = info->attrs;
1302 struct ovs_header *ovs_header = info->userhdr;
1303 struct sw_flow_key key;
1304 struct sk_buff *reply;
1305 struct sw_flow *flow;
1306 struct datapath *dp;
1307 struct flow_table *table;
1311 if (!a[OVS_FLOW_ATTR_KEY])
1313 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1317 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);
1326 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1327 info->snd_seq, OVS_FLOW_CMD_NEW);
1329 return PTR_ERR(reply);
1331 return genlmsg_reply(reply, info);
1334 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1336 struct nlattr **a = info->attrs;
1337 struct ovs_header *ovs_header = info->userhdr;
1338 struct sw_flow_key key;
1339 struct sk_buff *reply;
1340 struct sw_flow *flow;
1341 struct datapath *dp;
1342 struct flow_table *table;
1346 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1350 if (!a[OVS_FLOW_ATTR_KEY])
1351 return flush_flows(dp);
1353 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1357 table = genl_dereference(dp->table);
1358 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1362 reply = ovs_flow_cmd_alloc_info(flow);
1366 ovs_flow_tbl_remove(table, flow);
1368 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1369 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1372 ovs_flow_deferred_free(flow);
1374 genl_notify(reply, genl_info_net(info), info->snd_portid,
1375 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1379 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1381 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1382 struct datapath *dp;
1383 struct flow_table *table;
1385 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1389 table = genl_dereference(dp->table);
1392 struct sw_flow *flow;
1395 bucket = cb->args[0];
1397 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1401 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1402 NETLINK_CB(cb->skb).portid,
1403 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1404 OVS_FLOW_CMD_NEW) < 0)
1407 cb->args[0] = bucket;
1413 static struct genl_ops dp_flow_genl_ops[] = {
1414 { .cmd = OVS_FLOW_CMD_NEW,
1415 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1416 .policy = flow_policy,
1417 .doit = ovs_flow_cmd_new_or_set
1419 { .cmd = OVS_FLOW_CMD_DEL,
1420 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1421 .policy = flow_policy,
1422 .doit = ovs_flow_cmd_del
1424 { .cmd = OVS_FLOW_CMD_GET,
1425 .flags = 0, /* OK for unprivileged users. */
1426 .policy = flow_policy,
1427 .doit = ovs_flow_cmd_get,
1428 .dumpit = ovs_flow_cmd_dump
1430 { .cmd = OVS_FLOW_CMD_SET,
1431 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1432 .policy = flow_policy,
1433 .doit = ovs_flow_cmd_new_or_set,
1437 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1438 #ifdef HAVE_NLA_NUL_STRING
1439 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1441 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1444 static struct genl_family dp_datapath_genl_family = {
1445 .id = GENL_ID_GENERATE,
1446 .hdrsize = sizeof(struct ovs_header),
1447 .name = OVS_DATAPATH_FAMILY,
1448 .version = OVS_DATAPATH_VERSION,
1449 .maxattr = OVS_DP_ATTR_MAX,
1453 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1454 .name = OVS_DATAPATH_MCGROUP
1457 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1458 u32 portid, u32 seq, u32 flags, u8 cmd)
1460 struct ovs_header *ovs_header;
1461 struct ovs_dp_stats dp_stats;
1464 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1469 ovs_header->dp_ifindex = get_dpifindex(dp);
1472 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1475 goto nla_put_failure;
1477 get_dp_stats(dp, &dp_stats);
1478 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1479 goto nla_put_failure;
1481 return genlmsg_end(skb, ovs_header);
1484 genlmsg_cancel(skb, ovs_header);
1489 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1492 struct sk_buff *skb;
1495 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1497 return ERR_PTR(-ENOMEM);
1499 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1502 return ERR_PTR(retval);
1507 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1509 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1512 /* Called with genl_mutex and optionally with RTNL lock also. */
1513 static struct datapath *lookup_datapath(struct net *net,
1514 struct ovs_header *ovs_header,
1515 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1517 struct datapath *dp;
1519 if (!a[OVS_DP_ATTR_NAME])
1520 dp = get_dp(net, ovs_header->dp_ifindex);
1522 struct vport *vport;
1525 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1526 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1529 return dp ? dp : ERR_PTR(-ENODEV);
1532 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1534 struct nlattr **a = info->attrs;
1535 struct vport_parms parms;
1536 struct sk_buff *reply;
1537 struct datapath *dp;
1538 struct vport *vport;
1539 struct ovs_net *ovs_net;
1543 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1546 err = ovs_dp_cmd_validate(a);
1553 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1555 goto err_unlock_rtnl;
1557 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1559 /* Allocate table. */
1561 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1565 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1566 if (!dp->stats_percpu) {
1568 goto err_destroy_table;
1571 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1575 goto err_destroy_percpu;
1578 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1579 INIT_HLIST_HEAD(&dp->ports[i]);
1581 /* Set up our datapath device. */
1582 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1583 parms.type = OVS_VPORT_TYPE_INTERNAL;
1584 parms.options = NULL;
1586 parms.port_no = OVSP_LOCAL;
1587 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1589 vport = new_vport(&parms);
1590 if (IS_ERR(vport)) {
1591 err = PTR_ERR(vport);
1595 goto err_destroy_ports_array;
1598 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1599 info->snd_seq, OVS_DP_CMD_NEW);
1600 err = PTR_ERR(reply);
1602 goto err_destroy_local_port;
1604 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1605 list_add_tail(&dp->list_node, &ovs_net->dps);
1609 genl_notify(reply, genl_info_net(info), info->snd_portid,
1610 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1614 err_destroy_local_port:
1615 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1616 err_destroy_ports_array:
1619 free_percpu(dp->stats_percpu);
1621 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1623 release_net(ovs_dp_get_net(dp));
1631 /* Called with genl_mutex. */
1632 static void __dp_destroy(struct datapath *dp)
1638 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1639 struct vport *vport;
1640 struct hlist_node *node, *n;
1642 hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1643 if (vport->port_no != OVSP_LOCAL)
1644 ovs_dp_detach_port(vport);
1647 list_del(&dp->list_node);
1648 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1650 /* rtnl_unlock() will wait until all the references to devices that
1651 * are pending unregistration have been dropped. We do it here to
1652 * ensure that any internal devices (which contain DP pointers) are
1653 * fully destroyed before freeing the datapath.
1657 call_rcu(&dp->rcu, destroy_dp_rcu);
1660 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1662 struct sk_buff *reply;
1663 struct datapath *dp;
1666 err = ovs_dp_cmd_validate(info->attrs);
1670 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1675 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1676 info->snd_seq, OVS_DP_CMD_DEL);
1677 err = PTR_ERR(reply);
1683 genl_notify(reply, genl_info_net(info), info->snd_portid,
1684 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1690 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1692 struct sk_buff *reply;
1693 struct datapath *dp;
1696 err = ovs_dp_cmd_validate(info->attrs);
1700 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1704 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1705 info->snd_seq, OVS_DP_CMD_NEW);
1706 if (IS_ERR(reply)) {
1707 err = PTR_ERR(reply);
1708 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1709 ovs_dp_datapath_multicast_group.id, err);
1713 genl_notify(reply, genl_info_net(info), info->snd_portid,
1714 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1720 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1722 struct sk_buff *reply;
1723 struct datapath *dp;
1726 err = ovs_dp_cmd_validate(info->attrs);
1730 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1734 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1735 info->snd_seq, OVS_DP_CMD_NEW);
1737 return PTR_ERR(reply);
1739 return genlmsg_reply(reply, info);
1742 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1744 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1745 struct datapath *dp;
1746 int skip = cb->args[0];
1749 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1751 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1752 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1753 OVS_DP_CMD_NEW) < 0)
1763 static struct genl_ops dp_datapath_genl_ops[] = {
1764 { .cmd = OVS_DP_CMD_NEW,
1765 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1766 .policy = datapath_policy,
1767 .doit = ovs_dp_cmd_new
1769 { .cmd = OVS_DP_CMD_DEL,
1770 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1771 .policy = datapath_policy,
1772 .doit = ovs_dp_cmd_del
1774 { .cmd = OVS_DP_CMD_GET,
1775 .flags = 0, /* OK for unprivileged users. */
1776 .policy = datapath_policy,
1777 .doit = ovs_dp_cmd_get,
1778 .dumpit = ovs_dp_cmd_dump
1780 { .cmd = OVS_DP_CMD_SET,
1781 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1782 .policy = datapath_policy,
1783 .doit = ovs_dp_cmd_set,
1787 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1788 #ifdef HAVE_NLA_NUL_STRING
1789 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1790 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1792 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1794 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1795 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1796 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1797 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1800 static struct genl_family dp_vport_genl_family = {
1801 .id = GENL_ID_GENERATE,
1802 .hdrsize = sizeof(struct ovs_header),
1803 .name = OVS_VPORT_FAMILY,
1804 .version = OVS_VPORT_VERSION,
1805 .maxattr = OVS_VPORT_ATTR_MAX,
1809 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1810 .name = OVS_VPORT_MCGROUP
1813 /* Called with RTNL lock or RCU read lock. */
1814 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1815 u32 portid, u32 seq, u32 flags, u8 cmd)
1817 struct ovs_header *ovs_header;
1818 struct ovs_vport_stats vport_stats;
1821 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1826 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1828 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1829 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1830 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1831 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1832 goto nla_put_failure;
1834 ovs_vport_get_stats(vport, &vport_stats);
1835 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1837 goto nla_put_failure;
1839 err = ovs_vport_get_options(vport, skb);
1840 if (err == -EMSGSIZE)
1843 return genlmsg_end(skb, ovs_header);
1848 genlmsg_cancel(skb, ovs_header);
1852 /* Called with RTNL lock or RCU read lock. */
1853 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1856 struct sk_buff *skb;
1859 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1861 return ERR_PTR(-ENOMEM);
1863 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1866 return ERR_PTR(retval);
1871 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1873 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1876 /* Called with RTNL lock or RCU read lock. */
1877 static struct vport *lookup_vport(struct net *net,
1878 struct ovs_header *ovs_header,
1879 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1881 struct datapath *dp;
1882 struct vport *vport;
1884 if (a[OVS_VPORT_ATTR_NAME]) {
1885 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1887 return ERR_PTR(-ENODEV);
1888 if (ovs_header->dp_ifindex &&
1889 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1890 return ERR_PTR(-ENODEV);
1892 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1893 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1895 if (port_no >= DP_MAX_PORTS)
1896 return ERR_PTR(-EFBIG);
1898 dp = get_dp(net, ovs_header->dp_ifindex);
1900 return ERR_PTR(-ENODEV);
1902 vport = ovs_vport_rtnl_rcu(dp, port_no);
1904 return ERR_PTR(-ENODEV);
1907 return ERR_PTR(-EINVAL);
1910 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1912 struct nlattr **a = info->attrs;
1913 struct ovs_header *ovs_header = info->userhdr;
1914 struct vport_parms parms;
1915 struct sk_buff *reply;
1916 struct vport *vport;
1917 struct datapath *dp;
1922 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1923 !a[OVS_VPORT_ATTR_UPCALL_PID])
1926 err = ovs_vport_cmd_validate(a);
1931 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1936 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1937 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1940 if (port_no >= DP_MAX_PORTS)
1943 vport = ovs_vport_rtnl(dp, port_no);
1948 for (port_no = 1; ; port_no++) {
1949 if (port_no >= DP_MAX_PORTS) {
1953 vport = ovs_vport_rtnl(dp, port_no);
1959 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1960 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1961 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1963 parms.port_no = port_no;
1964 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1966 vport = new_vport(&parms);
1967 err = PTR_ERR(vport);
1972 if (a[OVS_VPORT_ATTR_STATS])
1973 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1975 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1977 if (IS_ERR(reply)) {
1978 err = PTR_ERR(reply);
1979 ovs_dp_detach_port(vport);
1982 genl_notify(reply, genl_info_net(info), info->snd_portid,
1983 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1991 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1993 struct nlattr **a = info->attrs;
1994 struct sk_buff *reply;
1995 struct vport *vport;
1998 err = ovs_vport_cmd_validate(a);
2003 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2004 err = PTR_ERR(vport);
2009 if (a[OVS_VPORT_ATTR_TYPE] &&
2010 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2013 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2014 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2018 if (a[OVS_VPORT_ATTR_STATS])
2019 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2021 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2022 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2024 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2025 info->snd_seq, OVS_VPORT_CMD_NEW);
2026 if (IS_ERR(reply)) {
2027 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
2028 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
2032 genl_notify(reply, genl_info_net(info), info->snd_portid,
2033 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2041 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2043 struct nlattr **a = info->attrs;
2044 struct sk_buff *reply;
2045 struct vport *vport;
2048 err = ovs_vport_cmd_validate(a);
2053 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2054 err = PTR_ERR(vport);
2058 if (vport->port_no == OVSP_LOCAL) {
2063 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2064 info->snd_seq, OVS_VPORT_CMD_DEL);
2065 err = PTR_ERR(reply);
2070 ovs_dp_detach_port(vport);
2072 genl_notify(reply, genl_info_net(info), info->snd_portid,
2073 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2081 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2083 struct nlattr **a = info->attrs;
2084 struct ovs_header *ovs_header = info->userhdr;
2085 struct sk_buff *reply;
2086 struct vport *vport;
2089 err = ovs_vport_cmd_validate(a);
2094 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2095 err = PTR_ERR(vport);
2099 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2100 info->snd_seq, OVS_VPORT_CMD_NEW);
2101 err = PTR_ERR(reply);
2107 return genlmsg_reply(reply, info);
2115 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2117 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2118 struct datapath *dp;
2119 int bucket = cb->args[0], skip = cb->args[1];
2122 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2127 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2128 struct vport *vport;
2129 struct hlist_node *n;
2132 hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
2134 ovs_vport_cmd_fill_info(vport, skb,
2135 NETLINK_CB(cb->skb).portid,
2138 OVS_VPORT_CMD_NEW) < 0)
2154 static struct genl_ops dp_vport_genl_ops[] = {
2155 { .cmd = OVS_VPORT_CMD_NEW,
2156 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2157 .policy = vport_policy,
2158 .doit = ovs_vport_cmd_new
2160 { .cmd = OVS_VPORT_CMD_DEL,
2161 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2162 .policy = vport_policy,
2163 .doit = ovs_vport_cmd_del
2165 { .cmd = OVS_VPORT_CMD_GET,
2166 .flags = 0, /* OK for unprivileged users. */
2167 .policy = vport_policy,
2168 .doit = ovs_vport_cmd_get,
2169 .dumpit = ovs_vport_cmd_dump
2171 { .cmd = OVS_VPORT_CMD_SET,
2172 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2173 .policy = vport_policy,
2174 .doit = ovs_vport_cmd_set,
2178 struct genl_family_and_ops {
2179 struct genl_family *family;
2180 struct genl_ops *ops;
2182 struct genl_multicast_group *group;
2185 static const struct genl_family_and_ops dp_genl_families[] = {
2186 { &dp_datapath_genl_family,
2187 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2188 &ovs_dp_datapath_multicast_group },
2189 { &dp_vport_genl_family,
2190 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2191 &ovs_dp_vport_multicast_group },
2192 { &dp_flow_genl_family,
2193 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2194 &ovs_dp_flow_multicast_group },
2195 { &dp_packet_genl_family,
2196 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2200 static void dp_unregister_genl(int n_families)
2204 for (i = 0; i < n_families; i++)
2205 genl_unregister_family(dp_genl_families[i].family);
2208 static int dp_register_genl(void)
2215 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2216 const struct genl_family_and_ops *f = &dp_genl_families[i];
2218 err = genl_register_family_with_ops(f->family, f->ops,
2225 err = genl_register_mc_group(f->family, f->group);
2234 dp_unregister_genl(n_registered);
2238 static int __rehash_flow_table(void *dummy)
2240 struct datapath *dp;
2245 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2247 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2248 struct flow_table *old_table = genl_dereference(dp->table);
2249 struct flow_table *new_table;
2251 new_table = ovs_flow_tbl_rehash(old_table);
2252 if (!IS_ERR(new_table)) {
2253 rcu_assign_pointer(dp->table, new_table);
2254 ovs_flow_tbl_deferred_destroy(old_table);
2262 static void rehash_flow_table(struct work_struct *work)
2264 genl_exec(__rehash_flow_table, NULL);
2265 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2268 static int dp_destroy_all(void *data)
2270 struct datapath *dp, *dp_next;
2271 struct ovs_net *ovs_net = data;
2273 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2279 static int __net_init ovs_init_net(struct net *net)
2281 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2283 INIT_LIST_HEAD(&ovs_net->dps);
2287 static void __net_exit ovs_exit_net(struct net *net)
2289 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2291 genl_exec(dp_destroy_all, ovs_net);
2294 static struct pernet_operations ovs_net_ops = {
2295 .init = ovs_init_net,
2296 .exit = ovs_exit_net,
2298 .size = sizeof(struct ovs_net),
2301 static int __init dp_init(void)
2305 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2307 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2310 err = genl_exec_init();
2314 err = ovs_workqueues_init();
2316 goto error_genl_exec;
2318 err = ovs_flow_init();
2322 err = ovs_vport_init();
2324 goto error_flow_exit;
2326 err = register_pernet_device(&ovs_net_ops);
2328 goto error_vport_exit;
2330 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2332 goto error_netns_exit;
2334 err = dp_register_genl();
2336 goto error_unreg_notifier;
2338 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2342 error_unreg_notifier:
2343 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2345 unregister_pernet_device(&ovs_net_ops);
2351 ovs_workqueues_exit();
2358 static void dp_cleanup(void)
2360 cancel_delayed_work_sync(&rehash_flow_wq);
2361 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2362 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2363 unregister_pernet_device(&ovs_net_ops);
2367 ovs_workqueues_exit();
2371 module_init(dp_init);
2372 module_exit(dp_cleanup);
2374 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2375 MODULE_LICENSE("GPL");
2376 MODULE_VERSION(VERSION);