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])
821 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
822 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
826 skb_reserve(packet, NET_IP_ALIGN);
828 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
830 skb_reset_mac_header(packet);
831 eth = eth_hdr(packet);
833 /* Normally, setting the skb 'protocol' field would be handled by a
834 * call to eth_type_trans(), but it assumes there's a sending
835 * device, which we may not have. */
836 if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
837 packet->protocol = eth->h_proto;
839 packet->protocol = htons(ETH_P_802_2);
841 /* Build an sw_flow for sending this packet. */
842 flow = ovs_flow_alloc();
847 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
851 err = ovs_flow_metadata_from_nlattrs(flow, key_len, a[OVS_PACKET_ATTR_KEY]);
854 acts = ovs_flow_actions_alloc(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
859 err = validate_and_copy_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0, &acts);
860 rcu_assign_pointer(flow->sf_acts, acts);
864 OVS_CB(packet)->flow = flow;
865 packet->priority = flow->key.phy.priority;
866 skb_set_mark(packet, flow->key.phy.skb_mark);
869 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
875 err = ovs_execute_actions(dp, packet);
892 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
893 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,18)
894 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
896 [OVS_PACKET_ATTR_PACKET] = { .minlen = ETH_HLEN },
898 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
899 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
902 static struct genl_ops dp_packet_genl_ops[] = {
903 { .cmd = OVS_PACKET_CMD_EXECUTE,
904 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
905 .policy = packet_policy,
906 .doit = ovs_packet_cmd_execute
910 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
913 struct flow_table *table = genl_dereference(dp->table);
915 stats->n_flows = ovs_flow_tbl_count(table);
917 stats->n_hit = stats->n_missed = stats->n_lost = 0;
918 for_each_possible_cpu(i) {
919 const struct dp_stats_percpu *percpu_stats;
920 struct dp_stats_percpu local_stats;
923 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
926 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
927 local_stats = *percpu_stats;
928 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
930 stats->n_hit += local_stats.n_hit;
931 stats->n_missed += local_stats.n_missed;
932 stats->n_lost += local_stats.n_lost;
936 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
937 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
938 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
939 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
942 static struct genl_family dp_flow_genl_family = {
943 .id = GENL_ID_GENERATE,
944 .hdrsize = sizeof(struct ovs_header),
945 .name = OVS_FLOW_FAMILY,
946 .version = OVS_FLOW_VERSION,
947 .maxattr = OVS_FLOW_ATTR_MAX,
951 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
952 .name = OVS_FLOW_MCGROUP
955 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb);
956 static int sample_action_to_attr(const struct nlattr *attr, struct sk_buff *skb)
958 const struct nlattr *a;
959 struct nlattr *start;
962 start = nla_nest_start(skb, OVS_ACTION_ATTR_SAMPLE);
966 nla_for_each_nested(a, attr, rem) {
967 int type = nla_type(a);
968 struct nlattr *st_sample;
971 case OVS_SAMPLE_ATTR_PROBABILITY:
972 if (nla_put(skb, OVS_SAMPLE_ATTR_PROBABILITY, sizeof(u32), nla_data(a)))
975 case OVS_SAMPLE_ATTR_ACTIONS:
976 st_sample = nla_nest_start(skb, OVS_SAMPLE_ATTR_ACTIONS);
979 err = actions_to_attr(nla_data(a), nla_len(a), skb);
982 nla_nest_end(skb, st_sample);
987 nla_nest_end(skb, start);
991 static int set_action_to_attr(const struct nlattr *a, struct sk_buff *skb)
993 const struct nlattr *ovs_key = nla_data(a);
994 int key_type = nla_type(ovs_key);
995 struct nlattr *start;
999 case OVS_KEY_ATTR_IPV4_TUNNEL:
1000 start = nla_nest_start(skb, OVS_ACTION_ATTR_SET);
1004 err = ipv4_tun_to_nlattr(skb, nla_data(ovs_key));
1007 nla_nest_end(skb, start);
1010 if (nla_put(skb, OVS_ACTION_ATTR_SET, nla_len(a), ovs_key))
1018 static int actions_to_attr(const struct nlattr *attr, int len, struct sk_buff *skb)
1020 const struct nlattr *a;
1023 nla_for_each_attr(a, attr, len, rem) {
1024 int type = nla_type(a);
1027 case OVS_ACTION_ATTR_SET:
1028 err = set_action_to_attr(a, skb);
1033 case OVS_ACTION_ATTR_SAMPLE:
1034 err = sample_action_to_attr(a, skb);
1039 if (nla_put(skb, type, nla_len(a), nla_data(a)))
1048 /* Called with genl_lock. */
1049 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
1050 struct sk_buff *skb, u32 portid,
1051 u32 seq, u32 flags, u8 cmd)
1053 const int skb_orig_len = skb->len;
1054 const struct sw_flow_actions *sf_acts;
1055 struct nlattr *start;
1056 struct ovs_flow_stats stats;
1057 struct ovs_header *ovs_header;
1063 sf_acts = rcu_dereference_protected(flow->sf_acts,
1064 lockdep_genl_is_held());
1066 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
1070 ovs_header->dp_ifindex = get_dpifindex(dp);
1072 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
1074 goto nla_put_failure;
1075 err = ovs_flow_to_nlattrs(&flow->key, skb);
1078 nla_nest_end(skb, nla);
1080 spin_lock_bh(&flow->lock);
1082 stats.n_packets = flow->packet_count;
1083 stats.n_bytes = flow->byte_count;
1084 tcp_flags = flow->tcp_flags;
1085 spin_unlock_bh(&flow->lock);
1088 nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
1089 goto nla_put_failure;
1091 if (stats.n_packets &&
1092 nla_put(skb, OVS_FLOW_ATTR_STATS,
1093 sizeof(struct ovs_flow_stats), &stats))
1094 goto nla_put_failure;
1097 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
1098 goto nla_put_failure;
1100 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
1101 * this is the first flow to be dumped into 'skb'. This is unusual for
1102 * Netlink but individual action lists can be longer than
1103 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
1104 * The userspace caller can always fetch the actions separately if it
1105 * really wants them. (Most userspace callers in fact don't care.)
1107 * This can only fail for dump operations because the skb is always
1108 * properly sized for single flows.
1110 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
1112 err = actions_to_attr(sf_acts->actions, sf_acts->actions_len, skb);
1114 nla_nest_end(skb, start);
1119 nla_nest_cancel(skb, start);
1121 } else if (skb_orig_len)
1122 goto nla_put_failure;
1124 return genlmsg_end(skb, ovs_header);
1129 genlmsg_cancel(skb, ovs_header);
1133 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
1135 const struct sw_flow_actions *sf_acts;
1138 sf_acts = rcu_dereference_protected(flow->sf_acts,
1139 lockdep_genl_is_held());
1141 /* OVS_FLOW_ATTR_KEY */
1142 len = nla_total_size(FLOW_BUFSIZE);
1143 /* OVS_FLOW_ATTR_ACTIONS */
1144 len += nla_total_size(sf_acts->actions_len);
1145 /* OVS_FLOW_ATTR_STATS */
1146 len += nla_total_size(sizeof(struct ovs_flow_stats));
1147 /* OVS_FLOW_ATTR_TCP_FLAGS */
1148 len += nla_total_size(1);
1149 /* OVS_FLOW_ATTR_USED */
1150 len += nla_total_size(8);
1152 len += NLMSG_ALIGN(sizeof(struct ovs_header));
1154 return genlmsg_new(len, GFP_KERNEL);
1157 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
1158 struct datapath *dp,
1159 u32 portid, u32 seq, u8 cmd)
1161 struct sk_buff *skb;
1164 skb = ovs_flow_cmd_alloc_info(flow);
1166 return ERR_PTR(-ENOMEM);
1168 retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
1173 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
1175 struct nlattr **a = info->attrs;
1176 struct ovs_header *ovs_header = info->userhdr;
1177 struct sw_flow_key key;
1178 struct sw_flow *flow;
1179 struct sk_buff *reply;
1180 struct datapath *dp;
1181 struct flow_table *table;
1182 struct sw_flow_actions *acts = NULL;
1188 if (!a[OVS_FLOW_ATTR_KEY])
1190 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1194 /* Validate actions. */
1195 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1196 acts = ovs_flow_actions_alloc(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
1197 error = PTR_ERR(acts);
1201 error = validate_and_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0, &acts);
1204 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1209 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1214 table = genl_dereference(dp->table);
1215 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1217 /* Bail out if we're not allowed to create a new flow. */
1219 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1222 /* Expand table, if necessary, to make room. */
1223 if (ovs_flow_tbl_need_to_expand(table)) {
1224 struct flow_table *new_table;
1226 new_table = ovs_flow_tbl_expand(table);
1227 if (!IS_ERR(new_table)) {
1228 rcu_assign_pointer(dp->table, new_table);
1229 ovs_flow_tbl_deferred_destroy(table);
1230 table = genl_dereference(dp->table);
1234 /* Allocate flow. */
1235 flow = ovs_flow_alloc();
1237 error = PTR_ERR(flow);
1242 rcu_assign_pointer(flow->sf_acts, acts);
1244 /* Put flow in bucket. */
1245 ovs_flow_tbl_insert(table, flow, &key, key_len);
1247 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1251 /* We found a matching flow. */
1252 struct sw_flow_actions *old_acts;
1254 /* Bail out if we're not allowed to modify an existing flow.
1255 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1256 * because Generic Netlink treats the latter as a dump
1257 * request. We also accept NLM_F_EXCL in case that bug ever
1261 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1262 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1265 /* Update actions. */
1266 old_acts = rcu_dereference_protected(flow->sf_acts,
1267 lockdep_genl_is_held());
1268 rcu_assign_pointer(flow->sf_acts, acts);
1269 ovs_flow_deferred_free_acts(old_acts);
1271 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1272 info->snd_seq, OVS_FLOW_CMD_NEW);
1275 if (a[OVS_FLOW_ATTR_CLEAR]) {
1276 spin_lock_bh(&flow->lock);
1278 spin_unlock_bh(&flow->lock);
1283 genl_notify(reply, genl_info_net(info), info->snd_portid,
1284 ovs_dp_flow_multicast_group.id, info->nlhdr,
1287 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1288 ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1297 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1299 struct nlattr **a = info->attrs;
1300 struct ovs_header *ovs_header = info->userhdr;
1301 struct sw_flow_key key;
1302 struct sk_buff *reply;
1303 struct sw_flow *flow;
1304 struct datapath *dp;
1305 struct flow_table *table;
1309 if (!a[OVS_FLOW_ATTR_KEY])
1311 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1315 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1319 table = genl_dereference(dp->table);
1320 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1324 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1325 info->snd_seq, OVS_FLOW_CMD_NEW);
1327 return PTR_ERR(reply);
1329 return genlmsg_reply(reply, info);
1332 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1334 struct nlattr **a = info->attrs;
1335 struct ovs_header *ovs_header = info->userhdr;
1336 struct sw_flow_key key;
1337 struct sk_buff *reply;
1338 struct sw_flow *flow;
1339 struct datapath *dp;
1340 struct flow_table *table;
1344 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1348 if (!a[OVS_FLOW_ATTR_KEY])
1349 return flush_flows(dp);
1351 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1355 table = genl_dereference(dp->table);
1356 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1360 reply = ovs_flow_cmd_alloc_info(flow);
1364 ovs_flow_tbl_remove(table, flow);
1366 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1367 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1370 ovs_flow_deferred_free(flow);
1372 genl_notify(reply, genl_info_net(info), info->snd_portid,
1373 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1377 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1379 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1380 struct datapath *dp;
1381 struct flow_table *table;
1383 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1387 table = genl_dereference(dp->table);
1390 struct sw_flow *flow;
1393 bucket = cb->args[0];
1395 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1399 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1400 NETLINK_CB(cb->skb).portid,
1401 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1402 OVS_FLOW_CMD_NEW) < 0)
1405 cb->args[0] = bucket;
1411 static struct genl_ops dp_flow_genl_ops[] = {
1412 { .cmd = OVS_FLOW_CMD_NEW,
1413 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414 .policy = flow_policy,
1415 .doit = ovs_flow_cmd_new_or_set
1417 { .cmd = OVS_FLOW_CMD_DEL,
1418 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1419 .policy = flow_policy,
1420 .doit = ovs_flow_cmd_del
1422 { .cmd = OVS_FLOW_CMD_GET,
1423 .flags = 0, /* OK for unprivileged users. */
1424 .policy = flow_policy,
1425 .doit = ovs_flow_cmd_get,
1426 .dumpit = ovs_flow_cmd_dump
1428 { .cmd = OVS_FLOW_CMD_SET,
1429 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1430 .policy = flow_policy,
1431 .doit = ovs_flow_cmd_new_or_set,
1435 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1436 #ifdef HAVE_NLA_NUL_STRING
1437 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1439 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1442 static struct genl_family dp_datapath_genl_family = {
1443 .id = GENL_ID_GENERATE,
1444 .hdrsize = sizeof(struct ovs_header),
1445 .name = OVS_DATAPATH_FAMILY,
1446 .version = OVS_DATAPATH_VERSION,
1447 .maxattr = OVS_DP_ATTR_MAX,
1451 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1452 .name = OVS_DATAPATH_MCGROUP
1455 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1456 u32 portid, u32 seq, u32 flags, u8 cmd)
1458 struct ovs_header *ovs_header;
1459 struct ovs_dp_stats dp_stats;
1462 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1467 ovs_header->dp_ifindex = get_dpifindex(dp);
1470 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1473 goto nla_put_failure;
1475 get_dp_stats(dp, &dp_stats);
1476 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1477 goto nla_put_failure;
1479 return genlmsg_end(skb, ovs_header);
1482 genlmsg_cancel(skb, ovs_header);
1487 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1490 struct sk_buff *skb;
1493 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1495 return ERR_PTR(-ENOMEM);
1497 retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1500 return ERR_PTR(retval);
1505 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1507 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1510 /* Called with genl_mutex and optionally with RTNL lock also. */
1511 static struct datapath *lookup_datapath(struct net *net,
1512 struct ovs_header *ovs_header,
1513 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1515 struct datapath *dp;
1517 if (!a[OVS_DP_ATTR_NAME])
1518 dp = get_dp(net, ovs_header->dp_ifindex);
1520 struct vport *vport;
1523 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1524 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1527 return dp ? dp : ERR_PTR(-ENODEV);
1530 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1532 struct nlattr **a = info->attrs;
1533 struct vport_parms parms;
1534 struct sk_buff *reply;
1535 struct datapath *dp;
1536 struct vport *vport;
1537 struct ovs_net *ovs_net;
1541 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1544 err = ovs_dp_cmd_validate(a);
1551 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1553 goto err_unlock_rtnl;
1555 ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1557 /* Allocate table. */
1559 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1563 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1564 if (!dp->stats_percpu) {
1566 goto err_destroy_table;
1569 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1573 goto err_destroy_percpu;
1576 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1577 INIT_HLIST_HEAD(&dp->ports[i]);
1579 /* Set up our datapath device. */
1580 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1581 parms.type = OVS_VPORT_TYPE_INTERNAL;
1582 parms.options = NULL;
1584 parms.port_no = OVSP_LOCAL;
1585 parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1587 vport = new_vport(&parms);
1588 if (IS_ERR(vport)) {
1589 err = PTR_ERR(vport);
1593 goto err_destroy_ports_array;
1596 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1597 info->snd_seq, OVS_DP_CMD_NEW);
1598 err = PTR_ERR(reply);
1600 goto err_destroy_local_port;
1602 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1603 list_add_tail(&dp->list_node, &ovs_net->dps);
1607 genl_notify(reply, genl_info_net(info), info->snd_portid,
1608 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1612 err_destroy_local_port:
1613 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1614 err_destroy_ports_array:
1617 free_percpu(dp->stats_percpu);
1619 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1621 release_net(ovs_dp_get_net(dp));
1629 /* Called with genl_mutex. */
1630 static void __dp_destroy(struct datapath *dp)
1636 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1637 struct vport *vport;
1638 struct hlist_node *n;
1640 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1641 if (vport->port_no != OVSP_LOCAL)
1642 ovs_dp_detach_port(vport);
1645 list_del(&dp->list_node);
1646 ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1648 /* rtnl_unlock() will wait until all the references to devices that
1649 * are pending unregistration have been dropped. We do it here to
1650 * ensure that any internal devices (which contain DP pointers) are
1651 * fully destroyed before freeing the datapath.
1655 call_rcu(&dp->rcu, destroy_dp_rcu);
1658 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1660 struct sk_buff *reply;
1661 struct datapath *dp;
1664 err = ovs_dp_cmd_validate(info->attrs);
1668 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1673 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1674 info->snd_seq, OVS_DP_CMD_DEL);
1675 err = PTR_ERR(reply);
1681 genl_notify(reply, genl_info_net(info), info->snd_portid,
1682 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1688 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1690 struct sk_buff *reply;
1691 struct datapath *dp;
1694 err = ovs_dp_cmd_validate(info->attrs);
1698 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1702 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1703 info->snd_seq, OVS_DP_CMD_NEW);
1704 if (IS_ERR(reply)) {
1705 err = PTR_ERR(reply);
1706 netlink_set_err(GENL_SOCK(sock_net(skb->sk)), 0,
1707 ovs_dp_datapath_multicast_group.id, err);
1711 genl_notify(reply, genl_info_net(info), info->snd_portid,
1712 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1718 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1720 struct sk_buff *reply;
1721 struct datapath *dp;
1724 err = ovs_dp_cmd_validate(info->attrs);
1728 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1732 reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1733 info->snd_seq, OVS_DP_CMD_NEW);
1735 return PTR_ERR(reply);
1737 return genlmsg_reply(reply, info);
1740 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1742 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1743 struct datapath *dp;
1744 int skip = cb->args[0];
1747 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1749 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1750 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1751 OVS_DP_CMD_NEW) < 0)
1761 static struct genl_ops dp_datapath_genl_ops[] = {
1762 { .cmd = OVS_DP_CMD_NEW,
1763 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1764 .policy = datapath_policy,
1765 .doit = ovs_dp_cmd_new
1767 { .cmd = OVS_DP_CMD_DEL,
1768 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1769 .policy = datapath_policy,
1770 .doit = ovs_dp_cmd_del
1772 { .cmd = OVS_DP_CMD_GET,
1773 .flags = 0, /* OK for unprivileged users. */
1774 .policy = datapath_policy,
1775 .doit = ovs_dp_cmd_get,
1776 .dumpit = ovs_dp_cmd_dump
1778 { .cmd = OVS_DP_CMD_SET,
1779 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1780 .policy = datapath_policy,
1781 .doit = ovs_dp_cmd_set,
1785 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1786 #ifdef HAVE_NLA_NUL_STRING
1787 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1788 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1790 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1792 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1793 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1794 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1795 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1798 static struct genl_family dp_vport_genl_family = {
1799 .id = GENL_ID_GENERATE,
1800 .hdrsize = sizeof(struct ovs_header),
1801 .name = OVS_VPORT_FAMILY,
1802 .version = OVS_VPORT_VERSION,
1803 .maxattr = OVS_VPORT_ATTR_MAX,
1807 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1808 .name = OVS_VPORT_MCGROUP
1811 /* Called with RTNL lock or RCU read lock. */
1812 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1813 u32 portid, u32 seq, u32 flags, u8 cmd)
1815 struct ovs_header *ovs_header;
1816 struct ovs_vport_stats vport_stats;
1819 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1824 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1826 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1827 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1828 nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1829 nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1830 goto nla_put_failure;
1832 ovs_vport_get_stats(vport, &vport_stats);
1833 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1835 goto nla_put_failure;
1837 err = ovs_vport_get_options(vport, skb);
1838 if (err == -EMSGSIZE)
1841 return genlmsg_end(skb, ovs_header);
1846 genlmsg_cancel(skb, ovs_header);
1850 /* Called with RTNL lock or RCU read lock. */
1851 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1854 struct sk_buff *skb;
1857 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1859 return ERR_PTR(-ENOMEM);
1861 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1867 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1869 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1872 /* Called with RTNL lock or RCU read lock. */
1873 static struct vport *lookup_vport(struct net *net,
1874 struct ovs_header *ovs_header,
1875 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1877 struct datapath *dp;
1878 struct vport *vport;
1880 if (a[OVS_VPORT_ATTR_NAME]) {
1881 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1883 return ERR_PTR(-ENODEV);
1884 if (ovs_header->dp_ifindex &&
1885 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1886 return ERR_PTR(-ENODEV);
1888 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1889 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1891 if (port_no >= DP_MAX_PORTS)
1892 return ERR_PTR(-EFBIG);
1894 dp = get_dp(net, ovs_header->dp_ifindex);
1896 return ERR_PTR(-ENODEV);
1898 vport = ovs_vport_rtnl_rcu(dp, port_no);
1900 return ERR_PTR(-ENODEV);
1903 return ERR_PTR(-EINVAL);
1906 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1908 struct nlattr **a = info->attrs;
1909 struct ovs_header *ovs_header = info->userhdr;
1910 struct vport_parms parms;
1911 struct sk_buff *reply;
1912 struct vport *vport;
1913 struct datapath *dp;
1918 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1919 !a[OVS_VPORT_ATTR_UPCALL_PID])
1922 err = ovs_vport_cmd_validate(a);
1927 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1932 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1933 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1936 if (port_no >= DP_MAX_PORTS)
1939 vport = ovs_vport_rtnl(dp, port_no);
1944 for (port_no = 1; ; port_no++) {
1945 if (port_no >= DP_MAX_PORTS) {
1949 vport = ovs_vport_rtnl(dp, port_no);
1955 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1956 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1957 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1959 parms.port_no = port_no;
1960 parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1962 vport = new_vport(&parms);
1963 err = PTR_ERR(vport);
1968 if (a[OVS_VPORT_ATTR_STATS])
1969 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1971 reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1973 if (IS_ERR(reply)) {
1974 err = PTR_ERR(reply);
1975 ovs_dp_detach_port(vport);
1978 genl_notify(reply, genl_info_net(info), info->snd_portid,
1979 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1987 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1989 struct nlattr **a = info->attrs;
1990 struct sk_buff *reply;
1991 struct vport *vport;
1994 err = ovs_vport_cmd_validate(a);
1999 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2000 err = PTR_ERR(vport);
2005 if (a[OVS_VPORT_ATTR_TYPE] &&
2006 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
2009 reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2015 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
2016 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2020 if (a[OVS_VPORT_ATTR_STATS])
2021 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
2023 if (a[OVS_VPORT_ATTR_UPCALL_PID])
2024 vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
2026 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2027 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2030 genl_notify(reply, genl_info_net(info), info->snd_portid,
2031 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2044 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2046 struct nlattr **a = info->attrs;
2047 struct sk_buff *reply;
2048 struct vport *vport;
2051 err = ovs_vport_cmd_validate(a);
2056 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2057 err = PTR_ERR(vport);
2061 if (vport->port_no == OVSP_LOCAL) {
2066 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2067 info->snd_seq, OVS_VPORT_CMD_DEL);
2068 err = PTR_ERR(reply);
2073 ovs_dp_detach_port(vport);
2075 genl_notify(reply, genl_info_net(info), info->snd_portid,
2076 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
2084 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2086 struct nlattr **a = info->attrs;
2087 struct ovs_header *ovs_header = info->userhdr;
2088 struct sk_buff *reply;
2089 struct vport *vport;
2092 err = ovs_vport_cmd_validate(a);
2097 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2098 err = PTR_ERR(vport);
2102 reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
2103 info->snd_seq, OVS_VPORT_CMD_NEW);
2104 err = PTR_ERR(reply);
2110 return genlmsg_reply(reply, info);
2118 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2120 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2121 struct datapath *dp;
2122 int bucket = cb->args[0], skip = cb->args[1];
2125 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2130 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2131 struct vport *vport;
2134 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2136 ovs_vport_cmd_fill_info(vport, skb,
2137 NETLINK_CB(cb->skb).portid,
2140 OVS_VPORT_CMD_NEW) < 0)
2156 static struct genl_ops dp_vport_genl_ops[] = {
2157 { .cmd = OVS_VPORT_CMD_NEW,
2158 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2159 .policy = vport_policy,
2160 .doit = ovs_vport_cmd_new
2162 { .cmd = OVS_VPORT_CMD_DEL,
2163 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2164 .policy = vport_policy,
2165 .doit = ovs_vport_cmd_del
2167 { .cmd = OVS_VPORT_CMD_GET,
2168 .flags = 0, /* OK for unprivileged users. */
2169 .policy = vport_policy,
2170 .doit = ovs_vport_cmd_get,
2171 .dumpit = ovs_vport_cmd_dump
2173 { .cmd = OVS_VPORT_CMD_SET,
2174 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2175 .policy = vport_policy,
2176 .doit = ovs_vport_cmd_set,
2180 struct genl_family_and_ops {
2181 struct genl_family *family;
2182 struct genl_ops *ops;
2184 struct genl_multicast_group *group;
2187 static const struct genl_family_and_ops dp_genl_families[] = {
2188 { &dp_datapath_genl_family,
2189 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2190 &ovs_dp_datapath_multicast_group },
2191 { &dp_vport_genl_family,
2192 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2193 &ovs_dp_vport_multicast_group },
2194 { &dp_flow_genl_family,
2195 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2196 &ovs_dp_flow_multicast_group },
2197 { &dp_packet_genl_family,
2198 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2202 static void dp_unregister_genl(int n_families)
2206 for (i = 0; i < n_families; i++)
2207 genl_unregister_family(dp_genl_families[i].family);
2210 static int dp_register_genl(void)
2217 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2218 const struct genl_family_and_ops *f = &dp_genl_families[i];
2220 err = genl_register_family_with_ops(f->family, f->ops,
2227 err = genl_register_mc_group(f->family, f->group);
2236 dp_unregister_genl(n_registered);
2240 static int __rehash_flow_table(void *dummy)
2242 struct datapath *dp;
2247 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2249 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2250 struct flow_table *old_table = genl_dereference(dp->table);
2251 struct flow_table *new_table;
2253 new_table = ovs_flow_tbl_rehash(old_table);
2254 if (!IS_ERR(new_table)) {
2255 rcu_assign_pointer(dp->table, new_table);
2256 ovs_flow_tbl_deferred_destroy(old_table);
2264 static void rehash_flow_table(struct work_struct *work)
2266 genl_exec(__rehash_flow_table, NULL);
2267 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2270 static int dp_destroy_all(void *data)
2272 struct datapath *dp, *dp_next;
2273 struct ovs_net *ovs_net = data;
2275 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2281 static int __net_init ovs_init_net(struct net *net)
2283 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2285 INIT_LIST_HEAD(&ovs_net->dps);
2289 static void __net_exit ovs_exit_net(struct net *net)
2291 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2293 genl_exec(dp_destroy_all, ovs_net);
2296 static struct pernet_operations ovs_net_ops = {
2297 .init = ovs_init_net,
2298 .exit = ovs_exit_net,
2300 .size = sizeof(struct ovs_net),
2303 static int __init dp_init(void)
2307 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2309 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2312 err = genl_exec_init();
2316 err = ovs_workqueues_init();
2318 goto error_genl_exec;
2320 err = ovs_flow_init();
2324 err = ovs_vport_init();
2326 goto error_flow_exit;
2328 err = register_pernet_device(&ovs_net_ops);
2330 goto error_vport_exit;
2332 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2334 goto error_netns_exit;
2336 err = dp_register_genl();
2338 goto error_unreg_notifier;
2340 schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2344 error_unreg_notifier:
2345 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2347 unregister_pernet_device(&ovs_net_ops);
2353 ovs_workqueues_exit();
2360 static void dp_cleanup(void)
2362 cancel_delayed_work_sync(&rehash_flow_wq);
2363 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2364 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2365 unregister_pernet_device(&ovs_net_ops);
2369 ovs_workqueues_exit();
2373 module_init(dp_init);
2374 module_exit(dp_cleanup);
2376 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2377 MODULE_LICENSE("GPL");
2378 MODULE_VERSION(VERSION);