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_head *head;
163 head = vport_hash_bucket(dp, port_no);
164 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
165 if (vport->port_no == port_no)
171 /* Called with RTNL lock and genl_lock. */
172 static struct vport *new_vport(const struct vport_parms *parms)
176 vport = ovs_vport_add(parms);
177 if (!IS_ERR(vport)) {
178 struct datapath *dp = parms->dp;
179 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
181 hlist_add_head_rcu(&vport->dp_hash_node, head);
186 /* Called with RTNL lock. */
187 void ovs_dp_detach_port(struct vport *p)
191 /* First drop references to device. */
192 hlist_del_rcu(&p->dp_hash_node);
194 /* Then destroy it. */
198 /* Must be called with rcu_read_lock. */
199 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
201 struct datapath *dp = p->dp;
202 struct sw_flow *flow;
203 struct dp_stats_percpu *stats;
204 struct sw_flow_key key;
209 stats = this_cpu_ptr(dp->stats_percpu);
211 /* Extract flow from 'skb' into 'key'. */
212 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
213 if (unlikely(error)) {
219 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
220 if (unlikely(!flow)) {
221 struct dp_upcall_info upcall;
223 upcall.cmd = OVS_PACKET_CMD_MISS;
225 upcall.userdata = NULL;
226 upcall.portid = p->upcall_portid;
227 ovs_dp_upcall(dp, skb, &upcall);
229 stats_counter = &stats->n_missed;
233 OVS_CB(skb)->flow = flow;
235 stats_counter = &stats->n_hit;
236 ovs_flow_used(OVS_CB(skb)->flow, skb);
237 ovs_execute_actions(dp, skb);
240 /* Update datapath statistics. */
241 u64_stats_update_begin(&stats->sync);
243 u64_stats_update_end(&stats->sync);
246 static struct genl_family dp_packet_genl_family = {
247 .id = GENL_ID_GENERATE,
248 .hdrsize = sizeof(struct ovs_header),
249 .name = OVS_PACKET_FAMILY,
250 .version = OVS_PACKET_VERSION,
251 .maxattr = OVS_PACKET_ATTR_MAX,
255 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
256 const struct dp_upcall_info *upcall_info)
258 struct dp_stats_percpu *stats;
262 if (upcall_info->portid == 0) {
267 dp_ifindex = get_dpifindex(dp);
273 forward_ip_summed(skb, true);
275 if (!skb_is_gso(skb))
276 err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
278 err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
285 stats = this_cpu_ptr(dp->stats_percpu);
287 u64_stats_update_begin(&stats->sync);
289 u64_stats_update_end(&stats->sync);
294 static int queue_gso_packets(struct net *net, int dp_ifindex,
296 const struct dp_upcall_info *upcall_info)
298 unsigned short gso_type = skb_shinfo(skb)->gso_type;
299 struct dp_upcall_info later_info;
300 struct sw_flow_key later_key;
301 struct sk_buff *segs, *nskb;
304 segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
306 return PTR_ERR(segs);
308 /* Queue all of the segments. */
311 err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
315 if (skb == segs && gso_type & SKB_GSO_UDP) {
316 /* The initial flow key extracted by ovs_flow_extract()
317 * in this case is for a first fragment, so we need to
318 * properly mark later fragments.
320 later_key = *upcall_info->key;
321 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
323 later_info = *upcall_info;
324 later_info.key = &later_key;
325 upcall_info = &later_info;
327 } while ((skb = skb->next));
329 /* Free all of the segments. */
337 } while ((skb = nskb));
341 static int queue_userspace_packet(struct net *net, int dp_ifindex,
343 const struct dp_upcall_info *upcall_info)
345 struct ovs_header *upcall;
346 struct sk_buff *nskb = NULL;
347 struct sk_buff *user_skb; /* to be queued to userspace */
352 if (vlan_tx_tag_present(skb)) {
353 nskb = skb_clone(skb, GFP_ATOMIC);
357 err = vlan_deaccel_tag(nskb);
364 if (nla_attr_size(skb->len) > USHRT_MAX) {
369 len = sizeof(struct ovs_header);
370 len += nla_total_size(skb->len);
371 len += nla_total_size(FLOW_BUFSIZE);
372 if (upcall_info->userdata)
373 len += NLA_ALIGN(upcall_info->userdata->nla_len);
375 user_skb = genlmsg_new(len, GFP_ATOMIC);
381 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
382 0, upcall_info->cmd);
383 upcall->dp_ifindex = dp_ifindex;
385 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
386 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
387 nla_nest_end(user_skb, nla);
389 if (upcall_info->userdata)
390 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
391 nla_len(upcall_info->userdata),
392 nla_data(upcall_info->userdata));
394 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
396 skb_copy_and_csum_dev(skb, nla_data(nla));
398 genlmsg_end(user_skb, upcall);
399 err = genlmsg_unicast(net, user_skb, upcall_info->portid);
406 /* Called with genl_mutex. */
407 static int flush_flows(struct datapath *dp)
409 struct flow_table *old_table;
410 struct flow_table *new_table;
412 old_table = genl_dereference(dp->table);
413 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
417 rcu_assign_pointer(dp->table, new_table);
419 ovs_flow_tbl_deferred_destroy(old_table);
423 static struct nlattr *reserve_sfa_size(struct sw_flow_actions **sfa, int attr_len)
426 struct sw_flow_actions *acts;
428 int req_size = NLA_ALIGN(attr_len);
429 int next_offset = offsetof(struct sw_flow_actions, actions) +
432 if (req_size <= (ksize(*sfa) - next_offset))
435 new_acts_size = ksize(*sfa) * 2;
437 if (new_acts_size > MAX_ACTIONS_BUFSIZE) {
438 if ((MAX_ACTIONS_BUFSIZE - next_offset) < req_size)
439 return ERR_PTR(-EMSGSIZE);
440 new_acts_size = MAX_ACTIONS_BUFSIZE;
443 acts = ovs_flow_actions_alloc(new_acts_size);
447 memcpy(acts->actions, (*sfa)->actions, (*sfa)->actions_len);
448 acts->actions_len = (*sfa)->actions_len;
453 (*sfa)->actions_len += req_size;
454 return (struct nlattr *) ((unsigned char *)(*sfa) + next_offset);
457 static int add_action(struct sw_flow_actions **sfa, int attrtype, void *data, int len)
461 a = reserve_sfa_size(sfa, nla_attr_size(len));
465 a->nla_type = attrtype;
466 a->nla_len = nla_attr_size(len);
469 memcpy(nla_data(a), data, len);
470 memset((unsigned char *) a + a->nla_len, 0, nla_padlen(len));
475 static inline int add_nested_action_start(struct sw_flow_actions **sfa, int attrtype)
477 int used = (*sfa)->actions_len;
480 err = add_action(sfa, attrtype, NULL, 0);
487 static inline void add_nested_action_end(struct sw_flow_actions *sfa, int st_offset)
489 struct nlattr *a = (struct nlattr *) ((unsigned char *)sfa->actions + st_offset);
491 a->nla_len = sfa->actions_len - st_offset;
494 static int validate_and_copy_actions(const struct nlattr *attr,
495 const struct sw_flow_key *key, int depth,
496 struct sw_flow_actions **sfa);
498 static int validate_and_copy_sample(const struct nlattr *attr,
499 const struct sw_flow_key *key, int depth,
500 struct sw_flow_actions **sfa)
502 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
503 const struct nlattr *probability, *actions;
504 const struct nlattr *a;
505 int rem, start, err, st_acts;
507 memset(attrs, 0, sizeof(attrs));
508 nla_for_each_nested(a, attr, rem) {
509 int type = nla_type(a);
510 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
517 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
518 if (!probability || nla_len(probability) != sizeof(u32))
521 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
522 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
525 /* validation done, copy sample action. */
526 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SAMPLE);
529 err = add_action(sfa, OVS_SAMPLE_ATTR_PROBABILITY, nla_data(probability), sizeof(u32));
532 st_acts = add_nested_action_start(sfa, OVS_SAMPLE_ATTR_ACTIONS);
536 err = validate_and_copy_actions(actions, key, depth + 1, sfa);
540 add_nested_action_end(*sfa, st_acts);
541 add_nested_action_end(*sfa, start);
546 static int validate_tp_port(const struct sw_flow_key *flow_key)
548 if (flow_key->eth.type == htons(ETH_P_IP)) {
549 if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
551 } else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
552 if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
559 static int validate_and_copy_set_tun(const struct nlattr *attr,
560 struct sw_flow_actions **sfa)
562 struct ovs_key_ipv4_tunnel tun_key;
565 err = ipv4_tun_from_nlattr(nla_data(attr), &tun_key);
569 start = add_nested_action_start(sfa, OVS_ACTION_ATTR_SET);
573 err = add_action(sfa, OVS_KEY_ATTR_IPV4_TUNNEL, &tun_key, sizeof(tun_key));
574 add_nested_action_end(*sfa, start);
579 static int validate_set(const struct nlattr *a,
580 const struct sw_flow_key *flow_key,
581 struct sw_flow_actions **sfa,
584 const struct nlattr *ovs_key = nla_data(a);
585 int key_type = nla_type(ovs_key);
587 /* There can be only one key in a action */
588 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
591 if (key_type > OVS_KEY_ATTR_MAX ||
592 (ovs_key_lens[key_type] != nla_len(ovs_key) &&
593 ovs_key_lens[key_type] != -1))
597 const struct ovs_key_ipv4 *ipv4_key;
598 const struct ovs_key_ipv6 *ipv6_key;
601 case OVS_KEY_ATTR_PRIORITY:
602 case OVS_KEY_ATTR_ETHERNET:
605 case OVS_KEY_ATTR_SKB_MARK:
606 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,20) && !defined(CONFIG_NETFILTER)
607 if (nla_get_u32(ovs_key) != 0)
612 case OVS_KEY_ATTR_TUNNEL:
614 err = validate_and_copy_set_tun(a, sfa);
619 case OVS_KEY_ATTR_IPV4:
620 if (flow_key->eth.type != htons(ETH_P_IP))
623 if (!flow_key->ip.proto)
626 ipv4_key = nla_data(ovs_key);
627 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
630 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
635 case OVS_KEY_ATTR_IPV6:
636 if (flow_key->eth.type != htons(ETH_P_IPV6))
639 if (!flow_key->ip.proto)
642 ipv6_key = nla_data(ovs_key);
643 if (ipv6_key->ipv6_proto != flow_key->ip.proto)
646 if (ipv6_key->ipv6_frag != flow_key->ip.frag)
649 if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
654 case OVS_KEY_ATTR_TCP:
655 if (flow_key->ip.proto != IPPROTO_TCP)
658 return validate_tp_port(flow_key);
660 case OVS_KEY_ATTR_UDP:
661 if (flow_key->ip.proto != IPPROTO_UDP)
664 return validate_tp_port(flow_key);
673 static int validate_userspace(const struct nlattr *attr)
675 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
676 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
677 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
679 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
682 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
683 attr, userspace_policy);
687 if (!a[OVS_USERSPACE_ATTR_PID] ||
688 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
694 static int copy_action(const struct nlattr *from,
695 struct sw_flow_actions **sfa)
697 int totlen = NLA_ALIGN(from->nla_len);
700 to = reserve_sfa_size(sfa, from->nla_len);
704 memcpy(to, from, totlen);
708 static int validate_and_copy_actions(const struct nlattr *attr,
709 const struct sw_flow_key *key,
711 struct sw_flow_actions **sfa)
713 const struct nlattr *a;
716 if (depth >= SAMPLE_ACTION_DEPTH)
719 nla_for_each_nested(a, attr, rem) {
720 /* Expected argument lengths, (u32)-1 for variable length. */
721 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
722 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
723 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
724 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
725 [OVS_ACTION_ATTR_POP_VLAN] = 0,
726 [OVS_ACTION_ATTR_SET] = (u32)-1,
727 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
729 const struct ovs_action_push_vlan *vlan;
730 int type = nla_type(a);
733 if (type > OVS_ACTION_ATTR_MAX ||
734 (action_lens[type] != nla_len(a) &&
735 action_lens[type] != (u32)-1))
740 case OVS_ACTION_ATTR_UNSPEC:
743 case OVS_ACTION_ATTR_USERSPACE:
744 err = validate_userspace(a);
749 case OVS_ACTION_ATTR_OUTPUT:
750 if (nla_get_u32(a) >= DP_MAX_PORTS)
755 case OVS_ACTION_ATTR_POP_VLAN:
758 case OVS_ACTION_ATTR_PUSH_VLAN:
760 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
762 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
766 case OVS_ACTION_ATTR_SET:
767 err = validate_set(a, key, sfa, &skip_copy);
772 case OVS_ACTION_ATTR_SAMPLE:
773 err = validate_and_copy_sample(a, key, depth, sfa);
783 err = copy_action(a, sfa);
795 static void clear_stats(struct sw_flow *flow)
799 flow->packet_count = 0;
800 flow->byte_count = 0;
803 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
805 struct ovs_header *ovs_header = info->userhdr;
806 struct nlattr **a = info->attrs;
807 struct sw_flow_actions *acts;
808 struct sk_buff *packet;
809 struct sw_flow *flow;
817 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
818 !a[OVS_PACKET_ATTR_ACTIONS] ||
819 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
822 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
823 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
827 skb_reserve(packet, NET_IP_ALIGN);
829 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
831 skb_reset_mac_header(packet);
832 eth = eth_hdr(packet);
834 /* Normally, setting the skb 'protocol' field would be handled by a
835 * call to eth_type_trans(), but it assumes there's a sending
836 * device, which we may not have. */
837 if (ntohs(eth->h_proto) >= 1536)
838 packet->protocol = eth->h_proto;
840 packet->protocol = htons(ETH_P_802_2);
842 /* Build an sw_flow for sending this packet. */
843 flow = ovs_flow_alloc();
848 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
852 err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
855 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
860 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
861 rcu_assign_pointer(flow->sf_acts, acts);
865 OVS_CB(packet)->flow = flow;
866 packet->priority = flow->key.phy.priority;
867 skb_set_mark(packet, flow->key.phy.skb_mark);
870 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
876 err = ovs_execute_actions(dp, packet);
893 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
894 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
895 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
896 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
899 static struct genl_ops dp_packet_genl_ops[] = {
900 { .cmd = OVS_PACKET_CMD_EXECUTE,
901 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
902 .policy = packet_policy,
903 .doit = ovs_packet_cmd_execute
907 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
910 struct flow_table *table = genl_dereference(dp->table);
912 stats->n_flows = ovs_flow_tbl_count(table);
914 stats->n_hit = stats->n_missed = stats->n_lost = 0;
915 for_each_possible_cpu(i) {
916 const struct dp_stats_percpu *percpu_stats;
917 struct dp_stats_percpu local_stats;
920 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
923 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
924 local_stats = *percpu_stats;
925 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
927 stats->n_hit += local_stats.n_hit;
928 stats->n_missed += local_stats.n_missed;
929 stats->n_lost += local_stats.n_lost;
933 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
934 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
935 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
936 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
939 static struct genl_family dp_flow_genl_family = {
940 .id = GENL_ID_GENERATE,
941 .hdrsize = sizeof(struct ovs_header),
942 .name = OVS_FLOW_FAMILY,
943 .version = OVS_FLOW_VERSION,
944 .maxattr = OVS_FLOW_ATTR_MAX,
948 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
949 .name = OVS_FLOW_MCGROUP
952 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
953 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
955 const struct nlattr *a;
956 struct nlattr *start;
959 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
963 nla_for_each_nested(a, attr, rem) {
964 int type = nla_type(a);
965 struct nlattr *st_sample;
968 case OVS_SAMPLE_ATTR_PROBABILITY:
969 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
972 case OVS_SAMPLE_ATTR_ACTIONS:
973 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
976 err = actions_to_attr(nla_data(a), nla_len(a), skb);
979 nla_nest_end(skb, st_sample);
984 nla_nest_end(skb, start);
988 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
990 const struct nlattr *ovs_key = nla_data(a);
991 int key_type = nla_type(ovs_key);
992 struct nlattr *start;
996 case OVS_KEY_ATTR_IPV4_TUNNEL:
997 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1001 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1004 nla_nest_end(skb, start);
1007 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1015 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1017 const struct nlattr *a;
1020 nla_for_each_attr(a, attr, len, rem) {
1021 int type = nla_type(a);
1024 case OVS_ACTION_ATTR_SET:
1025 err = set_action_to_attr(a, skb);
1030 case OVS_ACTION_ATTR_SAMPLE:
1031 err = sample_action_to_attr(a, skb);
1036 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1045 /* Called with genl_lock. */
1046 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1047 struct sk_buff *skb, u32 portid,
1048 u32 seq, u32 flags, u8 cmd)
1050 const int skb_orig_len = skb->len;
1051 const struct sw_flow_actions *sf_acts;
1052 struct nlattr *start;
1053 struct ovs_flow_stats stats;
1054 struct ovs_header *ovs_header;
1060 sf_acts = rcu_dereference_protected(flow->sf_acts,
1061 lockdep_genl_is_held());
1063 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1067 ovs_header->dp_ifindex = get_dpifindex(dp);
1069 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1071 goto nla_put_failure;
1072 err = ovs_flow_to_nlattrs(&flow->key, skb);
1075 nla_nest_end(skb, nla);
1077 spin_lock_bh(&flow->lock);
1079 stats.n_packets = flow->packet_count;
1080 stats.n_bytes = flow->byte_count;
1081 tcp_flags = flow->tcp_flags;
1082 spin_unlock_bh(&flow->lock);
1085 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1086 goto nla_put_failure;
1088 if (stats.n_packets &&
1089 nla_put(skb, OVS_FLOW_ATTR_STATS,
1090 sizeof(struct ovs_flow_stats), &stats))
1091 goto nla_put_failure;
1094 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1095 goto nla_put_failure;
1097 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1098 * this is the first flow to be dumped into 'skb'. This is unusual for
1099 * Netlink but individual action lists can be longer than
1100 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1101 * The userspace caller can always fetch the actions separately if it
1102 * really wants them. (Most userspace callers in fact don't care.)
1104 * This can only fail for dump operations because the skb is always
1105 * properly sized for single flows.
1107 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1109 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1111 nla_nest_end(skb, start);
1116 nla_nest_cancel(skb, start);
1118 } else if (skb_orig_len)
1119 goto nla_put_failure;
1121 return genlmsg_end(skb, ovs_header);
1126 genlmsg_cancel(skb, ovs_header);
1130 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1132 const struct sw_flow_actions *sf_acts;
1135 sf_acts = rcu_dereference_protected(flow->sf_acts,
1136 lockdep_genl_is_held());
1138 /* OVS_FLOW_ATTR_KEY */
1139 len = nla_total_size(FLOW_BUFSIZE);
1140 /* OVS_FLOW_ATTR_ACTIONS */
1141 len += nla_total_size(sf_acts->actions_len);
1142 /* OVS_FLOW_ATTR_STATS */
1143 len += nla_total_size(sizeof(struct ovs_flow_stats));
1144 /* OVS_FLOW_ATTR_TCP_FLAGS */
1145 len += nla_total_size(1);
1146 /* OVS_FLOW_ATTR_USED */
1147 len += nla_total_size(8);
1149 len += NLMSG_ALIGN(sizeof(struct ovs_header));
1151 return genlmsg_new(len, GFP_KERNEL);
1154 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1155 struct datapath *dp,
1156 u32 portid, u32 seq, u8 cmd)
1158 struct sk_buff *skb;
1161 skb = ovs_flow_cmd_alloc_info(flow);
1163 return ERR_PTR(-ENOMEM);
1165 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1170 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1172 struct nlattr **a = info->attrs;
1173 struct ovs_header *ovs_header = info->userhdr;
1174 struct sw_flow_key key;
1175 struct sw_flow *flow;
1176 struct sk_buff *reply;
1177 struct datapath *dp;
1178 struct flow_table *table;
1179 struct sw_flow_actions *acts = NULL;
1185 if (!a[OVS_FLOW_ATTR_KEY])
1187 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1191 /* Validate actions. */
1192 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1193 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1194 error = PTR_ERR(acts);
1198 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0, &acts);
1201 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1206 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1211 table = genl_dereference(dp->table);
1212 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1214 /* Bail out if we're not allowed to create a new flow. */
1216 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1219 /* Expand table, if necessary, to make room. */
1220 if (ovs_flow_tbl_need_to_expand(table)) {
1221 struct flow_table *new_table;
1223 new_table = ovs_flow_tbl_expand(table);
1224 if (!IS_ERR(new_table)) {
1225 rcu_assign_pointer(dp->table, new_table);
1226 ovs_flow_tbl_deferred_destroy(table);
1227 table = genl_dereference(dp->table);
1231 /* Allocate flow. */
1232 flow = ovs_flow_alloc();
1234 error = PTR_ERR(flow);
1239 rcu_assign_pointer(flow->sf_acts, acts);
1241 /* Put flow in bucket. */
1242 ovs_flow_tbl_insert(table, flow, &key, key_len);
1244 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1248 /* We found a matching flow. */
1249 struct sw_flow_actions *old_acts;
1251 /* Bail out if we're not allowed to modify an existing flow.
1252 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1253 * because Generic Netlink treats the latter as a dump
1254 * request. We also accept NLM_F_EXCL in case that bug ever
1258 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1259 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1262 /* Update actions. */
1263 old_acts = rcu_dereference_protected(flow->sf_acts,
1264 lockdep_genl_is_held());
1265 rcu_assign_pointer(flow->sf_acts, acts);
1266 ovs_flow_deferred_free_acts(old_acts);
1268 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1269 info->snd_seq, OVS_FLOW_CMD_NEW);
1272 if (a[OVS_FLOW_ATTR_CLEAR]) {
1273 spin_lock_bh(&flow->lock);
1275 spin_unlock_bh(&flow->lock);
1280 genl_notify(reply, genl_info_net(info), info->snd_portid,
1281 ovs_dp_flow_multicast_group.id, info->nlhdr,
1284 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1285 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1294 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1296 struct nlattr **a = info->attrs;
1297 struct ovs_header *ovs_header = info->userhdr;
1298 struct sw_flow_key key;
1299 struct sk_buff *reply;
1300 struct sw_flow *flow;
1301 struct datapath *dp;
1302 struct flow_table *table;
1306 if (!a[OVS_FLOW_ATTR_KEY])
1308 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1312 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1316 table = genl_dereference(dp->table);
1317 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1321 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1322 info->snd_seq, OVS_FLOW_CMD_NEW);
1324 return PTR_ERR(reply);
1326 return genlmsg_reply(reply, info);
1329 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1331 struct nlattr **a = info->attrs;
1332 struct ovs_header *ovs_header = info->userhdr;
1333 struct sw_flow_key key;
1334 struct sk_buff *reply;
1335 struct sw_flow *flow;
1336 struct datapath *dp;
1337 struct flow_table *table;
1341 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1345 if (!a[OVS_FLOW_ATTR_KEY])
1346 return flush_flows(dp);
1348 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1352 table = genl_dereference(dp->table);
1353 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1357 reply = ovs_flow_cmd_alloc_info(flow);
1361 ovs_flow_tbl_remove(table, flow);
1363 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1364 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1367 ovs_flow_deferred_free(flow);
1369 genl_notify(reply, genl_info_net(info), info->snd_portid,
1370 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1374 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1376 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1377 struct datapath *dp;
1378 struct flow_table *table;
1380 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1384 table = genl_dereference(dp->table);
1387 struct sw_flow *flow;
1390 bucket = cb->args[0];
1392 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1396 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1397 NETLINK_CB(cb->skb).portid,
1398 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1399 OVS_FLOW_CMD_NEW) < 0)
1402 cb->args[0] = bucket;
1408 static struct genl_ops dp_flow_genl_ops[] = {
1409 { .cmd = OVS_FLOW_CMD_NEW,
1410 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1411 .policy = flow_policy,
1412 .doit = ovs_flow_cmd_new_or_set
1414 { .cmd = OVS_FLOW_CMD_DEL,
1415 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1416 .policy = flow_policy,
1417 .doit = ovs_flow_cmd_del
1419 { .cmd = OVS_FLOW_CMD_GET,
1420 .flags = 0, /* OK for unprivileged users. */
1421 .policy = flow_policy,
1422 .doit = ovs_flow_cmd_get,
1423 .dumpit = ovs_flow_cmd_dump
1425 { .cmd = OVS_FLOW_CMD_SET,
1426 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1427 .policy = flow_policy,
1428 .doit = ovs_flow_cmd_new_or_set,
1432 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1433 #ifdef HAVE_NLA_NUL_STRING
1434 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1436 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1439 static struct genl_family dp_datapath_genl_family = {
1440 .id = GENL_ID_GENERATE,
1441 .hdrsize = sizeof(struct ovs_header),
1442 .name = OVS_DATAPATH_FAMILY,
1443 .version = OVS_DATAPATH_VERSION,
1444 .maxattr = OVS_DP_ATTR_MAX,
1448 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1449 .name = OVS_DATAPATH_MCGROUP
1452 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1453 u32 portid, u32 seq, u32 flags, u8 cmd)
1455 struct ovs_header *ovs_header;
1456 struct ovs_dp_stats dp_stats;
1459 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1464 ovs_header->dp_ifindex = get_dpifindex(dp);
1467 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1470 goto nla_put_failure;
1472 get_dp_stats(dp, &dp_stats);
1473 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1474 goto nla_put_failure;
1476 return genlmsg_end(skb, ovs_header);
1479 genlmsg_cancel(skb, ovs_header);
1484 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1487 struct sk_buff *skb;
1490 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1492 return ERR_PTR(-ENOMEM);
1494 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1497 return ERR_PTR(retval);
1502 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1504 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1507 /* Called with genl_mutex and optionally with RTNL lock also. */
1508 static struct datapath *lookup_datapath(struct net *net,
1509 struct ovs_header *ovs_header,
1510 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1512 struct datapath *dp;
1514 if (!a[OVS_DP_ATTR_NAME])
1515 dp = get_dp(net, ovs_header->dp_ifindex);
1517 struct vport *vport;
1520 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1521 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1524 return dp ? dp : ERR_PTR(-ENODEV);
1527 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1529 struct nlattr **a = info->attrs;
1530 struct vport_parms parms;
1531 struct sk_buff *reply;
1532 struct datapath *dp;
1533 struct vport *vport;
1534 struct ovs_net *ovs_net;
1538 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1541 err = ovs_dp_cmd_validate(a);
1548 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1550 goto err_unlock_rtnl;
1552 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1554 /* Allocate table. */
1556 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1560 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1561 if (!dp->stats_percpu) {
1563 goto err_destroy_table;
1566 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1570 goto err_destroy_percpu;
1573 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1574 INIT_HLIST_HEAD(&dp->ports[i]);
1576 /* Set up our datapath device. */
1577 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1578 parms.type = OVS_VPORT_TYPE_INTERNAL;
1579 parms.options = NULL;
1581 parms.port_no = OVSP_LOCAL;
1582 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1584 vport = new_vport(&parms);
1585 if (IS_ERR(vport)) {
1586 err = PTR_ERR(vport);
1590 goto err_destroy_ports_array;
1593 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1594 info->snd_seq, OVS_DP_CMD_NEW);
1595 err = PTR_ERR(reply);
1597 goto err_destroy_local_port;
1599 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1600 list_add_tail(&dp->list_node, &ovs_net->dps);
1604 genl_notify(reply, genl_info_net(info), info->snd_portid,
1605 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1609 err_destroy_local_port:
1610 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1611 err_destroy_ports_array:
1614 free_percpu(dp->stats_percpu);
1616 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1618 release_net(ovs_dp_get_net(dp));
1626 /* Called with genl_mutex. */
1627 static void __dp_destroy(struct datapath *dp)
1633 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1634 struct vport *vport;
1635 struct hlist_node *n;
1637 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1638 if (vport->port_no != OVSP_LOCAL)
1639 ovs_dp_detach_port(vport);
1642 list_del(&dp->list_node);
1643 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1645 /* rtnl_unlock() will wait until all the references to devices that
1646 * are pending unregistration have been dropped. We do it here to
1647 * ensure that any internal devices (which contain DP pointers) are
1648 * fully destroyed before freeing the datapath.
1652 call_rcu(&dp->rcu, destroy_dp_rcu);
1655 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1657 struct sk_buff *reply;
1658 struct datapath *dp;
1661 err = ovs_dp_cmd_validate(info->attrs);
1665 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1670 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1671 info->snd_seq, OVS_DP_CMD_DEL);
1672 err = PTR_ERR(reply);
1678 genl_notify(reply, genl_info_net(info), info->snd_portid,
1679 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1685 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1687 struct sk_buff *reply;
1688 struct datapath *dp;
1691 err = ovs_dp_cmd_validate(info->attrs);
1695 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1699 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1700 info->snd_seq, OVS_DP_CMD_NEW);
1701 if (IS_ERR(reply)) {
1702 err = PTR_ERR(reply);
1703 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1704 ovs_dp_datapath_multicast_group.id, err);
1708 genl_notify(reply, genl_info_net(info), info->snd_portid,
1709 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1715 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1717 struct sk_buff *reply;
1718 struct datapath *dp;
1721 err = ovs_dp_cmd_validate(info->attrs);
1725 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1729 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1730 info->snd_seq, OVS_DP_CMD_NEW);
1732 return PTR_ERR(reply);
1734 return genlmsg_reply(reply, info);
1737 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1739 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1740 struct datapath *dp;
1741 int skip = cb->args[0];
1744 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1746 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1747 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1748 OVS_DP_CMD_NEW) < 0)
1758 static struct genl_ops dp_datapath_genl_ops[] = {
1759 { .cmd = OVS_DP_CMD_NEW,
1760 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1761 .policy = datapath_policy,
1762 .doit = ovs_dp_cmd_new
1764 { .cmd = OVS_DP_CMD_DEL,
1765 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1766 .policy = datapath_policy,
1767 .doit = ovs_dp_cmd_del
1769 { .cmd = OVS_DP_CMD_GET,
1770 .flags = 0, /* OK for unprivileged users. */
1771 .policy = datapath_policy,
1772 .doit = ovs_dp_cmd_get,
1773 .dumpit = ovs_dp_cmd_dump
1775 { .cmd = OVS_DP_CMD_SET,
1776 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1777 .policy = datapath_policy,
1778 .doit = ovs_dp_cmd_set,
1782 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1783 #ifdef HAVE_NLA_NUL_STRING
1784 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1785 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1787 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1789 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1790 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1791 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1792 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1795 static struct genl_family dp_vport_genl_family = {
1796 .id = GENL_ID_GENERATE,
1797 .hdrsize = sizeof(struct ovs_header),
1798 .name = OVS_VPORT_FAMILY,
1799 .version = OVS_VPORT_VERSION,
1800 .maxattr = OVS_VPORT_ATTR_MAX,
1804 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1805 .name = OVS_VPORT_MCGROUP
1808 /* Called with RTNL lock or RCU read lock. */
1809 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1810 u32 portid, u32 seq, u32 flags, u8 cmd)
1812 struct ovs_header *ovs_header;
1813 struct ovs_vport_stats vport_stats;
1816 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1821 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1823 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1824 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1825 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1826 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1827 goto nla_put_failure;
1829 ovs_vport_get_stats(vport, &vport_stats);
1830 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1832 goto nla_put_failure;
1834 err = ovs_vport_get_options(vport, skb);
1835 if (err == -EMSGSIZE)
1838 return genlmsg_end(skb, ovs_header);
1843 genlmsg_cancel(skb, ovs_header);
1847 /* Called with RTNL lock or RCU read lock. */
1848 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1851 struct sk_buff *skb;
1854 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1856 return ERR_PTR(-ENOMEM);
1858 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1861 return ERR_PTR(retval);
1866 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1868 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1871 /* Called with RTNL lock or RCU read lock. */
1872 static struct vport *lookup_vport(struct net *net,
1873 struct ovs_header *ovs_header,
1874 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1876 struct datapath *dp;
1877 struct vport *vport;
1879 if (a[OVS_VPORT_ATTR_NAME]) {
1880 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1882 return ERR_PTR(-ENODEV);
1883 if (ovs_header->dp_ifindex &&
1884 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1885 return ERR_PTR(-ENODEV);
1887 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1888 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1890 if (port_no >= DP_MAX_PORTS)
1891 return ERR_PTR(-EFBIG);
1893 dp = get_dp(net, ovs_header->dp_ifindex);
1895 return ERR_PTR(-ENODEV);
1897 vport = ovs_vport_rtnl_rcu(dp, port_no);
1899 return ERR_PTR(-ENODEV);
1902 return ERR_PTR(-EINVAL);
1905 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1907 struct nlattr **a = info->attrs;
1908 struct ovs_header *ovs_header = info->userhdr;
1909 struct vport_parms parms;
1910 struct sk_buff *reply;
1911 struct vport *vport;
1912 struct datapath *dp;
1917 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1918 !a[OVS_VPORT_ATTR_UPCALL_PID])
1921 err = ovs_vport_cmd_validate(a);
1926 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1931 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1932 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1935 if (port_no >= DP_MAX_PORTS)
1938 vport = ovs_vport_rtnl(dp, port_no);
1943 for (port_no = 1; ; port_no++) {
1944 if (port_no >= DP_MAX_PORTS) {
1948 vport = ovs_vport_rtnl(dp, port_no);
1954 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1955 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1956 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1958 parms.port_no = port_no;
1959 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1961 vport = new_vport(&parms);
1962 err = PTR_ERR(vport);
1967 if (a[OVS_VPORT_ATTR_STATS])
1968 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1970 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1972 if (IS_ERR(reply)) {
1973 err = PTR_ERR(reply);
1974 ovs_dp_detach_port(vport);
1977 genl_notify(reply, genl_info_net(info), info->snd_portid,
1978 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1986 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1988 struct nlattr **a = info->attrs;
1989 struct sk_buff *reply;
1990 struct vport *vport;
1993 err = ovs_vport_cmd_validate(a);
1998 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1999 err = PTR_ERR(vport);
2004 if (a[OVS_VPORT_ATTR_TYPE] &&
2005 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2008 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2009 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2013 if (a[OVS_VPORT_ATTR_STATS])
2014 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2016 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2017 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2019 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2020 info->snd_seq, OVS_VPORT_CMD_NEW);
2021 if (IS_ERR(reply)) {
2022 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
2023 ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
2027 genl_notify(reply, genl_info_net(info), info->snd_portid,
2028 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2036 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2038 struct nlattr **a = info->attrs;
2039 struct sk_buff *reply;
2040 struct vport *vport;
2043 err = ovs_vport_cmd_validate(a);
2048 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2049 err = PTR_ERR(vport);
2053 if (vport->port_no == OVSP_LOCAL) {
2058 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2059 info->snd_seq, OVS_VPORT_CMD_DEL);
2060 err = PTR_ERR(reply);
2065 ovs_dp_detach_port(vport);
2067 genl_notify(reply, genl_info_net(info), info->snd_portid,
2068 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2076 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2078 struct nlattr **a = info->attrs;
2079 struct ovs_header *ovs_header = info->userhdr;
2080 struct sk_buff *reply;
2081 struct vport *vport;
2084 err = ovs_vport_cmd_validate(a);
2089 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2090 err = PTR_ERR(vport);
2094 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2095 info->snd_seq, OVS_VPORT_CMD_NEW);
2096 err = PTR_ERR(reply);
2102 return genlmsg_reply(reply, info);
2110 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2112 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2113 struct datapath *dp;
2114 int bucket = cb->args[0], skip = cb->args[1];
2117 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2122 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2123 struct vport *vport;
2126 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2128 ovs_vport_cmd_fill_info(vport, skb,
2129 NETLINK_CB(cb->skb).portid,
2132 OVS_VPORT_CMD_NEW) < 0)
2148 static struct genl_ops dp_vport_genl_ops[] = {
2149 { .cmd = OVS_VPORT_CMD_NEW,
2150 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2151 .policy = vport_policy,
2152 .doit = ovs_vport_cmd_new
2154 { .cmd = OVS_VPORT_CMD_DEL,
2155 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2156 .policy = vport_policy,
2157 .doit = ovs_vport_cmd_del
2159 { .cmd = OVS_VPORT_CMD_GET,
2160 .flags = 0, /* OK for unprivileged users. */
2161 .policy = vport_policy,
2162 .doit = ovs_vport_cmd_get,
2163 .dumpit = ovs_vport_cmd_dump
2165 { .cmd = OVS_VPORT_CMD_SET,
2166 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2167 .policy = vport_policy,
2168 .doit = ovs_vport_cmd_set,
2172 struct genl_family_and_ops {
2173 struct genl_family *family;
2174 struct genl_ops *ops;
2176 struct genl_multicast_group *group;
2179 static const struct genl_family_and_ops dp_genl_families[] = {
2180 { &dp_datapath_genl_family,
2181 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2182 &ovs_dp_datapath_multicast_group },
2183 { &dp_vport_genl_family,
2184 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2185 &ovs_dp_vport_multicast_group },
2186 { &dp_flow_genl_family,
2187 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2188 &ovs_dp_flow_multicast_group },
2189 { &dp_packet_genl_family,
2190 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2194 static void dp_unregister_genl(int n_families)
2198 for (i = 0; i < n_families; i++)
2199 genl_unregister_family(dp_genl_families[i].family);
2202 static int dp_register_genl(void)
2209 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2210 const struct genl_family_and_ops *f = &dp_genl_families[i];
2212 err = genl_register_family_with_ops(f->family, f->ops,
2219 err = genl_register_mc_group(f->family, f->group);
2228 dp_unregister_genl(n_registered);
2232 static int __rehash_flow_table(void *dummy)
2234 struct datapath *dp;
2239 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2241 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2242 struct flow_table *old_table = genl_dereference(dp->table);
2243 struct flow_table *new_table;
2245 new_table = ovs_flow_tbl_rehash(old_table);
2246 if (!IS_ERR(new_table)) {
2247 rcu_assign_pointer(dp->table, new_table);
2248 ovs_flow_tbl_deferred_destroy(old_table);
2256 static void rehash_flow_table(struct work_struct *work)
2258 genl_exec(__rehash_flow_table, NULL);
2259 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2262 static int dp_destroy_all(void *data)
2264 struct datapath *dp, *dp_next;
2265 struct ovs_net *ovs_net = data;
2267 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2273 static int __net_init ovs_init_net(struct net *net)
2275 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2277 INIT_LIST_HEAD(&ovs_net->dps);
2281 static void __net_exit ovs_exit_net(struct net *net)
2283 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2285 genl_exec(dp_destroy_all, ovs_net);
2288 static struct pernet_operations ovs_net_ops = {
2289 .init = ovs_init_net,
2290 .exit = ovs_exit_net,
2292 .size = sizeof(struct ovs_net),
2295 static int __init dp_init(void)
2299 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2301 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2304 err = genl_exec_init();
2308 err = ovs_workqueues_init();
2310 goto error_genl_exec;
2312 err = ovs_flow_init();
2316 err = ovs_vport_init();
2318 goto error_flow_exit;
2320 err = register_pernet_device(&ovs_net_ops);
2322 goto error_vport_exit;
2324 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2326 goto error_netns_exit;
2328 err = dp_register_genl();
2330 goto error_unreg_notifier;
2332 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2336 error_unreg_notifier:
2337 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2339 unregister_pernet_device(&ovs_net_ops);
2345 ovs_workqueues_exit();
2352 static void dp_cleanup(void)
2354 cancel_delayed_work_sync(&rehash_flow_wq);
2355 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2356 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2357 unregister_pernet_device(&ovs_net_ops);
2361 ovs_workqueues_exit();
2365 module_init(dp_init);
2366 module_exit(dp_cleanup);
2368 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2369 MODULE_LICENSE("GPL");
2370 MODULE_VERSION(VERSION);