2 * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
3 * Distributed under the terms of the GNU GPL version 2.
5 * Significant portions of this file may be copied from parts of the Linux
6 * kernel, by Linus Torvalds and others.
9 /* Functions for managing the dp interface/device. */
11 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/init.h>
14 #include <linux/module.h>
15 #include <linux/if_arp.h>
16 #include <linux/if_vlan.h>
19 #include <linux/jhash.h>
20 #include <linux/delay.h>
21 #include <linux/time.h>
22 #include <linux/etherdevice.h>
23 #include <linux/genetlink.h>
24 #include <linux/kernel.h>
25 #include <linux/kthread.h>
26 #include <linux/mutex.h>
27 #include <linux/percpu.h>
28 #include <linux/rcupdate.h>
29 #include <linux/tcp.h>
30 #include <linux/udp.h>
31 #include <linux/version.h>
32 #include <linux/ethtool.h>
33 #include <linux/wait.h>
34 #include <asm/system.h>
35 #include <asm/div64.h>
37 #include <linux/highmem.h>
38 #include <linux/netfilter_bridge.h>
39 #include <linux/netfilter_ipv4.h>
40 #include <linux/inetdevice.h>
41 #include <linux/list.h>
42 #include <linux/openvswitch.h>
43 #include <linux/rculist.h>
44 #include <linux/dmi.h>
45 #include <net/inet_ecn.h>
46 #include <net/genetlink.h>
54 #include "vport-internal_dev.h"
56 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
57 LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0)
58 #error Kernels before 2.6.18 or after 3.1 are not supported by this version of Open vSwitch.
61 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
62 EXPORT_SYMBOL(dp_ioctl_hook);
67 * Writes to device state (add/remove datapath, port, set operations on vports,
68 * etc.) are protected by RTNL.
70 * Writes to other state (flow table modifications, set miscellaneous datapath
71 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
74 * Reads are protected by RCU.
76 * There are a few special cases (mostly stats) that have their own
77 * synchronization but they nest under all of above and don't interact with
81 /* Global list of datapaths to enable dumping them all out.
82 * Protected by genl_mutex.
84 static LIST_HEAD(dps);
86 static struct vport *new_vport(const struct vport_parms *);
87 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
88 const struct dp_upcall_info *);
89 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
90 const struct dp_upcall_info *);
92 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
93 struct datapath *get_dp(int dp_ifindex)
95 struct datapath *dp = NULL;
96 struct net_device *dev;
99 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
101 struct vport *vport = internal_dev_get_vport(dev);
109 EXPORT_SYMBOL_GPL(get_dp);
111 /* Must be called with genl_mutex. */
112 static struct flow_table *get_table_protected(struct datapath *dp)
114 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
117 /* Must be called with rcu_read_lock or RTNL lock. */
118 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
120 return rcu_dereference_rtnl(dp->ports[port_no]);
123 /* Must be called with rcu_read_lock or RTNL lock. */
124 const char *dp_name(const struct datapath *dp)
126 return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
129 static int get_dpifindex(struct datapath *dp)
136 local = get_vport_protected(dp, OVSP_LOCAL);
138 ifindex = vport_get_ifindex(local);
147 static inline size_t br_nlmsg_size(void)
149 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
150 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
151 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
152 + nla_total_size(4) /* IFLA_MASTER */
153 + nla_total_size(4) /* IFLA_MTU */
154 + nla_total_size(1); /* IFLA_OPERSTATE */
157 /* Caller must hold RTNL lock. */
158 static int dp_fill_ifinfo(struct sk_buff *skb,
159 const struct vport *port,
160 int event, unsigned int flags)
162 struct datapath *dp = port->dp;
163 int ifindex = vport_get_ifindex(port);
164 struct ifinfomsg *hdr;
165 struct nlmsghdr *nlh;
170 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
174 hdr = nlmsg_data(nlh);
175 hdr->ifi_family = AF_BRIDGE;
177 hdr->ifi_type = ARPHRD_ETHER;
178 hdr->ifi_index = ifindex;
179 hdr->ifi_flags = vport_get_flags(port);
182 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
183 NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
184 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
185 #ifdef IFLA_OPERSTATE
186 NLA_PUT_U8(skb, IFLA_OPERSTATE,
187 vport_is_running(port)
188 ? vport_get_operstate(port)
192 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
194 return nlmsg_end(skb, nlh);
197 nlmsg_cancel(skb, nlh);
201 /* Caller must hold RTNL lock. */
202 static void dp_ifinfo_notify(int event, struct vport *port)
207 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
211 err = dp_fill_ifinfo(skb, port, event, 0);
213 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
214 WARN_ON(err == -EMSGSIZE);
218 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
222 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
225 static void release_dp(struct kobject *kobj)
227 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
231 static struct kobj_type dp_ktype = {
232 .release = release_dp
235 static void destroy_dp_rcu(struct rcu_head *rcu)
237 struct datapath *dp = container_of(rcu, struct datapath, rcu);
239 flow_tbl_destroy(dp->table);
240 free_percpu(dp->stats_percpu);
241 kobject_put(&dp->ifobj);
244 /* Called with RTNL lock and genl_lock. */
245 static struct vport *new_vport(const struct vport_parms *parms)
249 vport = vport_add(parms);
250 if (!IS_ERR(vport)) {
251 struct datapath *dp = parms->dp;
253 rcu_assign_pointer(dp->ports[parms->port_no], vport);
254 list_add(&vport->node, &dp->port_list);
256 dp_ifinfo_notify(RTM_NEWLINK, vport);
262 /* Called with RTNL lock. */
263 void dp_detach_port(struct vport *p)
267 if (p->port_no != OVSP_LOCAL)
269 dp_ifinfo_notify(RTM_DELLINK, p);
271 /* First drop references to device. */
273 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
275 /* Then destroy it. */
279 /* Must be called with rcu_read_lock. */
280 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
282 struct datapath *dp = p->dp;
283 struct sw_flow *flow;
284 struct dp_stats_percpu *stats;
288 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
289 OVS_CB(skb)->vport = p;
291 if (!OVS_CB(skb)->flow) {
292 struct sw_flow_key key;
295 /* Extract flow from 'skb' into 'key'. */
296 error = flow_extract(skb, p->port_no, &key, &key_len);
297 if (unlikely(error)) {
303 flow = flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
304 if (unlikely(!flow)) {
305 struct dp_upcall_info upcall;
307 upcall.cmd = OVS_PACKET_CMD_MISS;
309 upcall.userdata = NULL;
310 upcall.pid = p->upcall_pid;
311 dp_upcall(dp, skb, &upcall);
313 stats_counter = &stats->n_missed;
317 OVS_CB(skb)->flow = flow;
320 stats_counter = &stats->n_hit;
321 flow_used(OVS_CB(skb)->flow, skb);
322 execute_actions(dp, skb);
325 /* Update datapath statistics. */
327 write_seqcount_begin(&stats->seqlock);
329 write_seqcount_end(&stats->seqlock);
332 static void copy_and_csum_skb(struct sk_buff *skb, void *to)
334 u16 csum_start, csum_offset;
337 get_skb_csum_pointers(skb, &csum_start, &csum_offset);
338 csum_start -= skb_headroom(skb);
340 skb_copy_bits(skb, 0, to, csum_start);
342 csum = skb_copy_and_csum_bits(skb, csum_start, to + csum_start,
343 skb->len - csum_start, 0);
344 *(__sum16 *)(to + csum_start + csum_offset) = csum_fold(csum);
347 static struct genl_family dp_packet_genl_family = {
348 .id = GENL_ID_GENERATE,
349 .hdrsize = sizeof(struct ovs_header),
350 .name = OVS_PACKET_FAMILY,
351 .version = OVS_PACKET_VERSION,
352 .maxattr = OVS_PACKET_ATTR_MAX
355 int dp_upcall(struct datapath *dp, struct sk_buff *skb,
356 const struct dp_upcall_info *upcall_info)
358 struct dp_stats_percpu *stats;
362 if (upcall_info->pid == 0) {
367 dp_ifindex = get_dpifindex(dp);
373 forward_ip_summed(skb, true);
375 if (!skb_is_gso(skb))
376 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
378 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
385 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
387 write_seqcount_begin(&stats->seqlock);
389 write_seqcount_end(&stats->seqlock);
394 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
395 const struct dp_upcall_info *upcall_info)
397 struct dp_upcall_info later_info;
398 struct sw_flow_key later_key;
399 struct sk_buff *segs, *nskb;
402 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
406 /* Queue all of the segments. */
409 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
413 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
414 /* The initial flow key extracted by flow_extract() in
415 * this case is for a first fragment, so we need to
416 * properly mark later fragments.
418 later_key = *upcall_info->key;
419 later_key.ip.tos_frag &= ~OVS_FRAG_TYPE_MASK;
420 later_key.ip.tos_frag |= OVS_FRAG_TYPE_LATER;
422 later_info = *upcall_info;
423 later_info.key = &later_key;
424 upcall_info = &later_info;
426 } while ((skb = skb->next));
428 /* Free all of the segments. */
436 } while ((skb = nskb));
440 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
441 const struct dp_upcall_info *upcall_info)
443 struct ovs_header *upcall;
444 struct sk_buff *user_skb; /* to be queued to userspace */
449 err = vlan_deaccel_tag(skb);
453 if (nla_attr_size(skb->len) > USHRT_MAX)
456 len = sizeof(struct ovs_header);
457 len += nla_total_size(skb->len);
458 len += nla_total_size(FLOW_BUFSIZE);
459 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
460 len += nla_total_size(8);
462 user_skb = genlmsg_new(len, GFP_ATOMIC);
466 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
467 0, upcall_info->cmd);
468 upcall->dp_ifindex = dp_ifindex;
470 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
471 flow_to_nlattrs(upcall_info->key, user_skb);
472 nla_nest_end(user_skb, nla);
474 if (upcall_info->userdata)
475 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
476 nla_get_u64(upcall_info->userdata));
478 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
479 if (skb->ip_summed == CHECKSUM_PARTIAL)
480 copy_and_csum_skb(skb, nla_data(nla));
482 skb_copy_bits(skb, 0, nla_data(nla), skb->len);
484 return genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
487 /* Called with genl_mutex. */
488 static int flush_flows(int dp_ifindex)
490 struct flow_table *old_table;
491 struct flow_table *new_table;
494 dp = get_dp(dp_ifindex);
498 old_table = get_table_protected(dp);
499 new_table = flow_tbl_alloc(TBL_MIN_BUCKETS);
503 rcu_assign_pointer(dp->table, new_table);
505 flow_tbl_deferred_destroy(old_table);
509 static int validate_actions(const struct nlattr *attr,
510 const struct sw_flow_key *key, int depth);
512 static int validate_sample(const struct nlattr *attr,
513 const struct sw_flow_key *key, int depth)
515 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
516 const struct nlattr *probability, *actions;
517 const struct nlattr *a;
520 memset(attrs, 0, sizeof(attrs));
521 nla_for_each_nested (a, attr, rem) {
522 int type = nla_type(a);
523 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
530 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
531 if (!probability || nla_len(probability) != sizeof(u32))
534 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
535 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
537 return validate_actions(actions, key, depth + 1);
540 static int validate_action_key(const struct nlattr *a,
541 const struct sw_flow_key *flow_key)
543 int act_type = nla_type(a);
544 const struct nlattr *ovs_key = nla_data(a);
545 int key_type = nla_type(ovs_key);
547 /* There can be only one key in a action */
548 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
551 if (key_type > OVS_KEY_ATTR_MAX ||
552 nla_len(ovs_key) != ovs_key_lens[key_type])
555 #define ACTION(act, key) (((act) << 8) | (key))
557 switch(ACTION(act_type, key_type)) {
558 const struct ovs_key_ipv4 *ipv4_key;
559 const struct ovs_key_8021q *q_key;
561 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_PRIORITY):
562 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TUN_ID):
563 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_ETHERNET):
566 case ACTION(OVS_ACTION_ATTR_PUSH, OVS_KEY_ATTR_8021Q):
567 q_key = nla_data(ovs_key);
568 if (q_key->q_tpid != htons(ETH_P_8021Q))
571 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
575 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_IPV4):
576 if (flow_key->eth.type != htons(ETH_P_IP))
579 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
582 ipv4_key = nla_data(ovs_key);
583 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
586 if (ipv4_key->ipv4_tos & INET_ECN_MASK)
589 if (ipv4_key->ipv4_frag !=
590 (flow_key->ip.tos_frag & OVS_FRAG_TYPE_MASK))
595 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TCP):
596 if (flow_key->ip.proto != IPPROTO_TCP)
599 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
604 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_UDP):
605 if (flow_key->ip.proto != IPPROTO_UDP)
608 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
619 static int validate_userspace(const struct nlattr *attr)
621 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =
623 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
624 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
626 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
629 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr, userspace_policy);
633 if (!a[OVS_USERSPACE_ATTR_PID] || !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
639 static int validate_actions(const struct nlattr *attr,
640 const struct sw_flow_key *key, int depth)
642 const struct nlattr *a;
645 if (depth >= SAMPLE_ACTION_DEPTH)
648 nla_for_each_nested(a, attr, rem) {
649 /* Expected argument lengths, (u32)-1 for variable length. */
650 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
651 [OVS_ACTION_ATTR_OUTPUT] = 4,
652 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
653 [OVS_ACTION_ATTR_PUSH] = (u32)-1,
654 [OVS_ACTION_ATTR_POP] = 2,
655 [OVS_ACTION_ATTR_SET] = (u32)-1,
656 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
658 int type = nla_type(a);
660 if (type > OVS_ACTION_ATTR_MAX ||
661 (action_lens[type] != nla_len(a) &&
662 action_lens[type] != (u32)-1))
666 case OVS_ACTION_ATTR_UNSPEC:
669 case OVS_ACTION_ATTR_USERSPACE:
670 err = validate_userspace(a);
675 case OVS_ACTION_ATTR_OUTPUT:
676 if (nla_get_u32(a) >= DP_MAX_PORTS)
681 case OVS_ACTION_ATTR_POP:
682 if (nla_get_u16(a) != OVS_KEY_ATTR_8021Q)
686 case OVS_ACTION_ATTR_SET:
687 case OVS_ACTION_ATTR_PUSH:
688 err = validate_action_key(a, key);
693 case OVS_ACTION_ATTR_SAMPLE:
694 err = validate_sample(a, key, depth);
710 static void clear_stats(struct sw_flow *flow)
714 flow->packet_count = 0;
715 flow->byte_count = 0;
718 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
720 struct ovs_header *ovs_header = info->userhdr;
721 struct nlattr **a = info->attrs;
722 struct sw_flow_actions *acts;
723 struct sk_buff *packet;
724 struct sw_flow *flow;
732 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
733 !a[OVS_PACKET_ATTR_ACTIONS] ||
734 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
737 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
738 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
742 skb_reserve(packet, NET_IP_ALIGN);
744 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
746 skb_reset_mac_header(packet);
747 eth = eth_hdr(packet);
749 /* Normally, setting the skb 'protocol' field would be handled by a
750 * call to eth_type_trans(), but it assumes there's a sending
751 * device, which we may not have. */
752 if (ntohs(eth->h_proto) >= 1536)
753 packet->protocol = eth->h_proto;
755 packet->protocol = htons(ETH_P_802_2);
757 /* Build an sw_flow for sending this packet. */
763 err = flow_extract(packet, -1, &flow->key, &key_len);
767 err = flow_metadata_from_nlattrs(&flow->key.phy.priority,
768 &flow->key.phy.in_port,
769 &flow->key.phy.tun_id,
770 a[OVS_PACKET_ATTR_KEY]);
774 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
778 flow->hash = flow_hash(&flow->key, key_len);
780 acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
784 rcu_assign_pointer(flow->sf_acts, acts);
786 OVS_CB(packet)->flow = flow;
787 packet->priority = flow->key.phy.priority;
790 dp = get_dp(ovs_header->dp_ifindex);
795 if (flow->key.phy.in_port < DP_MAX_PORTS)
796 OVS_CB(packet)->vport = get_vport_protected(dp,
797 flow->key.phy.in_port);
800 err = execute_actions(dp, packet);
817 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
818 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
819 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
820 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
823 static struct genl_ops dp_packet_genl_ops[] = {
824 { .cmd = OVS_PACKET_CMD_EXECUTE,
825 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
826 .policy = packet_policy,
827 .doit = ovs_packet_cmd_execute
831 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
834 struct flow_table *table = get_table_protected(dp);
836 stats->n_flows = flow_tbl_count(table);
838 stats->n_hit = stats->n_missed = stats->n_lost = 0;
839 for_each_possible_cpu(i) {
840 const struct dp_stats_percpu *percpu_stats;
841 struct dp_stats_percpu local_stats;
844 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
847 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
848 local_stats = *percpu_stats;
849 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
851 stats->n_hit += local_stats.n_hit;
852 stats->n_missed += local_stats.n_missed;
853 stats->n_lost += local_stats.n_lost;
857 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
858 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
859 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
860 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
863 static struct genl_family dp_flow_genl_family = {
864 .id = GENL_ID_GENERATE,
865 .hdrsize = sizeof(struct ovs_header),
866 .name = OVS_FLOW_FAMILY,
867 .version = OVS_FLOW_VERSION,
868 .maxattr = OVS_FLOW_ATTR_MAX
871 static struct genl_multicast_group dp_flow_multicast_group = {
872 .name = OVS_FLOW_MCGROUP
875 /* Called with genl_lock. */
876 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
877 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
879 const int skb_orig_len = skb->len;
880 const struct sw_flow_actions *sf_acts;
881 struct ovs_flow_stats stats;
882 struct ovs_header *ovs_header;
888 sf_acts = rcu_dereference_protected(flow->sf_acts,
889 lockdep_genl_is_held());
891 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
895 ovs_header->dp_ifindex = get_dpifindex(dp);
897 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
899 goto nla_put_failure;
900 err = flow_to_nlattrs(&flow->key, skb);
903 nla_nest_end(skb, nla);
905 spin_lock_bh(&flow->lock);
907 stats.n_packets = flow->packet_count;
908 stats.n_bytes = flow->byte_count;
909 tcp_flags = flow->tcp_flags;
910 spin_unlock_bh(&flow->lock);
913 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
916 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
919 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
921 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
922 * this is the first flow to be dumped into 'skb'. This is unusual for
923 * Netlink but individual action lists can be longer than
924 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
925 * The userspace caller can always fetch the actions separately if it
926 * really wants them. (Most userspace callers in fact don't care.)
928 * This can only fail for dump operations because the skb is always
929 * properly sized for single flows.
931 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
933 if (err < 0 && skb_orig_len)
936 return genlmsg_end(skb, ovs_header);
941 genlmsg_cancel(skb, ovs_header);
945 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
947 const struct sw_flow_actions *sf_acts;
950 sf_acts = rcu_dereference_protected(flow->sf_acts,
951 lockdep_genl_is_held());
953 len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
954 len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
955 len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
956 len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
957 len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
958 return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
961 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
962 u32 pid, u32 seq, u8 cmd)
967 skb = ovs_flow_cmd_alloc_info(flow);
969 return ERR_PTR(-ENOMEM);
971 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
976 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
978 struct nlattr **a = info->attrs;
979 struct ovs_header *ovs_header = info->userhdr;
980 struct sw_flow_key key;
981 struct sw_flow *flow;
982 struct sk_buff *reply;
984 struct flow_table *table;
990 if (!a[OVS_FLOW_ATTR_KEY])
992 error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
996 /* Validate actions. */
997 if (a[OVS_FLOW_ATTR_ACTIONS]) {
998 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
1001 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1006 dp = get_dp(ovs_header->dp_ifindex);
1011 table = get_table_protected(dp);
1012 flow = flow_tbl_lookup(table, &key, key_len);
1014 struct sw_flow_actions *acts;
1016 /* Bail out if we're not allowed to create a new flow. */
1018 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1021 /* Expand table, if necessary, to make room. */
1022 if (flow_tbl_need_to_expand(table)) {
1023 struct flow_table *new_table;
1025 new_table = flow_tbl_expand(table);
1026 if (!IS_ERR(new_table)) {
1027 rcu_assign_pointer(dp->table, new_table);
1028 flow_tbl_deferred_destroy(table);
1029 table = get_table_protected(dp);
1033 /* Allocate flow. */
1034 flow = flow_alloc();
1036 error = PTR_ERR(flow);
1042 /* Obtain actions. */
1043 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1044 error = PTR_ERR(acts);
1046 goto error_free_flow;
1047 rcu_assign_pointer(flow->sf_acts, acts);
1049 /* Put flow in bucket. */
1050 flow->hash = flow_hash(&key, key_len);
1051 flow_tbl_insert(table, flow);
1053 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1054 info->snd_seq, OVS_FLOW_CMD_NEW);
1056 /* We found a matching flow. */
1057 struct sw_flow_actions *old_acts;
1059 /* Bail out if we're not allowed to modify an existing flow.
1060 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1061 * because Generic Netlink treats the latter as a dump
1062 * request. We also accept NLM_F_EXCL in case that bug ever
1066 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1067 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1070 /* Update actions. */
1071 old_acts = rcu_dereference_protected(flow->sf_acts,
1072 lockdep_genl_is_held());
1073 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1074 (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1075 memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1076 old_acts->actions_len))) {
1077 struct sw_flow_actions *new_acts;
1079 new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1080 error = PTR_ERR(new_acts);
1081 if (IS_ERR(new_acts))
1084 rcu_assign_pointer(flow->sf_acts, new_acts);
1085 flow_deferred_free_acts(old_acts);
1088 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1089 info->snd_seq, OVS_FLOW_CMD_NEW);
1092 if (a[OVS_FLOW_ATTR_CLEAR]) {
1093 spin_lock_bh(&flow->lock);
1095 spin_unlock_bh(&flow->lock);
1100 genl_notify(reply, genl_info_net(info), info->snd_pid,
1101 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1103 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1104 dp_flow_multicast_group.id, PTR_ERR(reply));
1113 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1115 struct nlattr **a = info->attrs;
1116 struct ovs_header *ovs_header = info->userhdr;
1117 struct sw_flow_key key;
1118 struct sk_buff *reply;
1119 struct sw_flow *flow;
1120 struct datapath *dp;
1121 struct flow_table *table;
1125 if (!a[OVS_FLOW_ATTR_KEY])
1127 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1131 dp = get_dp(ovs_header->dp_ifindex);
1135 table = get_table_protected(dp);
1136 flow = flow_tbl_lookup(table, &key, key_len);
1140 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1142 return PTR_ERR(reply);
1144 return genlmsg_reply(reply, info);
1147 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1149 struct nlattr **a = info->attrs;
1150 struct ovs_header *ovs_header = info->userhdr;
1151 struct sw_flow_key key;
1152 struct sk_buff *reply;
1153 struct sw_flow *flow;
1154 struct datapath *dp;
1155 struct flow_table *table;
1159 if (!a[OVS_FLOW_ATTR_KEY])
1160 return flush_flows(ovs_header->dp_ifindex);
1161 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1165 dp = get_dp(ovs_header->dp_ifindex);
1169 table = get_table_protected(dp);
1170 flow = flow_tbl_lookup(table, &key, key_len);
1174 reply = ovs_flow_cmd_alloc_info(flow);
1178 flow_tbl_remove(table, flow);
1180 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1181 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1184 flow_deferred_free(flow);
1186 genl_notify(reply, genl_info_net(info), info->snd_pid,
1187 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1191 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1193 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1194 struct datapath *dp;
1196 dp = get_dp(ovs_header->dp_ifindex);
1201 struct sw_flow *flow;
1204 bucket = cb->args[0];
1206 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1210 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1211 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1212 OVS_FLOW_CMD_NEW) < 0)
1215 cb->args[0] = bucket;
1221 static struct genl_ops dp_flow_genl_ops[] = {
1222 { .cmd = OVS_FLOW_CMD_NEW,
1223 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1224 .policy = flow_policy,
1225 .doit = ovs_flow_cmd_new_or_set
1227 { .cmd = OVS_FLOW_CMD_DEL,
1228 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1229 .policy = flow_policy,
1230 .doit = ovs_flow_cmd_del
1232 { .cmd = OVS_FLOW_CMD_GET,
1233 .flags = 0, /* OK for unprivileged users. */
1234 .policy = flow_policy,
1235 .doit = ovs_flow_cmd_get,
1236 .dumpit = ovs_flow_cmd_dump
1238 { .cmd = OVS_FLOW_CMD_SET,
1239 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1240 .policy = flow_policy,
1241 .doit = ovs_flow_cmd_new_or_set,
1245 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1246 #ifdef HAVE_NLA_NUL_STRING
1247 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1249 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1252 static struct genl_family dp_datapath_genl_family = {
1253 .id = GENL_ID_GENERATE,
1254 .hdrsize = sizeof(struct ovs_header),
1255 .name = OVS_DATAPATH_FAMILY,
1256 .version = OVS_DATAPATH_VERSION,
1257 .maxattr = OVS_DP_ATTR_MAX
1260 static struct genl_multicast_group dp_datapath_multicast_group = {
1261 .name = OVS_DATAPATH_MCGROUP
1264 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1265 u32 pid, u32 seq, u32 flags, u8 cmd)
1267 struct ovs_header *ovs_header;
1271 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1276 ovs_header->dp_ifindex = get_dpifindex(dp);
1279 err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1282 goto nla_put_failure;
1284 nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1286 goto nla_put_failure;
1287 get_dp_stats(dp, nla_data(nla));
1289 return genlmsg_end(skb, ovs_header);
1292 genlmsg_cancel(skb, ovs_header);
1297 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1300 struct sk_buff *skb;
1303 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1305 return ERR_PTR(-ENOMEM);
1307 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1310 return ERR_PTR(retval);
1315 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1317 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1320 /* Called with genl_mutex and optionally with RTNL lock also. */
1321 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1323 struct datapath *dp;
1325 if (!a[OVS_DP_ATTR_NAME])
1326 dp = get_dp(ovs_header->dp_ifindex);
1328 struct vport *vport;
1331 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1332 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1335 return dp ? dp : ERR_PTR(-ENODEV);
1338 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1340 struct nlattr **a = info->attrs;
1341 struct vport_parms parms;
1342 struct sk_buff *reply;
1343 struct datapath *dp;
1344 struct vport *vport;
1348 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1351 err = ovs_dp_cmd_validate(a);
1357 if (!try_module_get(THIS_MODULE))
1358 goto err_unlock_rtnl;
1361 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1363 goto err_put_module;
1364 INIT_LIST_HEAD(&dp->port_list);
1366 /* Initialize kobject for bridge. This will be added as
1367 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1368 dp->ifobj.kset = NULL;
1369 kobject_init(&dp->ifobj, &dp_ktype);
1371 /* Allocate table. */
1373 rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1377 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1378 if (!dp->stats_percpu) {
1380 goto err_destroy_table;
1383 /* Set up our datapath device. */
1384 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1385 parms.type = OVS_VPORT_TYPE_INTERNAL;
1386 parms.options = NULL;
1388 parms.port_no = OVSP_LOCAL;
1389 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1391 vport = new_vport(&parms);
1392 if (IS_ERR(vport)) {
1393 err = PTR_ERR(vport);
1397 goto err_destroy_percpu;
1400 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1401 err = PTR_ERR(reply);
1403 goto err_destroy_local_port;
1405 list_add_tail(&dp->list_node, &dps);
1406 dp_sysfs_add_dp(dp);
1410 genl_notify(reply, genl_info_net(info), info->snd_pid,
1411 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1414 err_destroy_local_port:
1415 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1417 free_percpu(dp->stats_percpu);
1419 flow_tbl_destroy(get_table_protected(dp));
1423 module_put(THIS_MODULE);
1430 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1432 struct vport *vport, *next_vport;
1433 struct sk_buff *reply;
1434 struct datapath *dp;
1437 err = ovs_dp_cmd_validate(info->attrs);
1442 dp = lookup_datapath(info->userhdr, info->attrs);
1447 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1448 err = PTR_ERR(reply);
1452 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1453 if (vport->port_no != OVSP_LOCAL)
1454 dp_detach_port(vport);
1456 dp_sysfs_del_dp(dp);
1457 list_del(&dp->list_node);
1458 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1460 /* rtnl_unlock() will wait until all the references to devices that
1461 * are pending unregistration have been dropped. We do it here to
1462 * ensure that any internal devices (which contain DP pointers) are
1463 * fully destroyed before freeing the datapath.
1467 call_rcu(&dp->rcu, destroy_dp_rcu);
1468 module_put(THIS_MODULE);
1470 genl_notify(reply, genl_info_net(info), info->snd_pid,
1471 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1481 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1483 struct sk_buff *reply;
1484 struct datapath *dp;
1487 err = ovs_dp_cmd_validate(info->attrs);
1491 dp = lookup_datapath(info->userhdr, info->attrs);
1495 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1496 if (IS_ERR(reply)) {
1497 err = PTR_ERR(reply);
1498 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1499 dp_datapath_multicast_group.id, err);
1503 genl_notify(reply, genl_info_net(info), info->snd_pid,
1504 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1508 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1510 struct sk_buff *reply;
1511 struct datapath *dp;
1514 err = ovs_dp_cmd_validate(info->attrs);
1518 dp = lookup_datapath(info->userhdr, info->attrs);
1522 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1524 return PTR_ERR(reply);
1526 return genlmsg_reply(reply, info);
1529 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1531 struct datapath *dp;
1532 int skip = cb->args[0];
1535 list_for_each_entry (dp, &dps, list_node) {
1538 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1539 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1540 OVS_DP_CMD_NEW) < 0)
1550 static struct genl_ops dp_datapath_genl_ops[] = {
1551 { .cmd = OVS_DP_CMD_NEW,
1552 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1553 .policy = datapath_policy,
1554 .doit = ovs_dp_cmd_new
1556 { .cmd = OVS_DP_CMD_DEL,
1557 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1558 .policy = datapath_policy,
1559 .doit = ovs_dp_cmd_del
1561 { .cmd = OVS_DP_CMD_GET,
1562 .flags = 0, /* OK for unprivileged users. */
1563 .policy = datapath_policy,
1564 .doit = ovs_dp_cmd_get,
1565 .dumpit = ovs_dp_cmd_dump
1567 { .cmd = OVS_DP_CMD_SET,
1568 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1569 .policy = datapath_policy,
1570 .doit = ovs_dp_cmd_set,
1574 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1575 #ifdef HAVE_NLA_NUL_STRING
1576 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1577 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1578 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1580 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1581 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1583 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1584 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1585 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1586 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1589 static struct genl_family dp_vport_genl_family = {
1590 .id = GENL_ID_GENERATE,
1591 .hdrsize = sizeof(struct ovs_header),
1592 .name = OVS_VPORT_FAMILY,
1593 .version = OVS_VPORT_VERSION,
1594 .maxattr = OVS_VPORT_ATTR_MAX
1597 struct genl_multicast_group dp_vport_multicast_group = {
1598 .name = OVS_VPORT_MCGROUP
1601 /* Called with RTNL lock or RCU read lock. */
1602 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1603 u32 pid, u32 seq, u32 flags, u8 cmd)
1605 struct ovs_header *ovs_header;
1609 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1614 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1616 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1617 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1618 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1619 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1621 nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1623 goto nla_put_failure;
1625 vport_get_stats(vport, nla_data(nla));
1627 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1629 err = vport_get_options(vport, skb);
1630 if (err == -EMSGSIZE)
1633 return genlmsg_end(skb, ovs_header);
1638 genlmsg_cancel(skb, ovs_header);
1642 /* Called with RTNL lock or RCU read lock. */
1643 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1646 struct sk_buff *skb;
1649 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1651 return ERR_PTR(-ENOMEM);
1653 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1656 return ERR_PTR(retval);
1661 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1663 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1666 /* Called with RTNL lock or RCU read lock. */
1667 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1668 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1670 struct datapath *dp;
1671 struct vport *vport;
1673 if (a[OVS_VPORT_ATTR_NAME]) {
1674 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1676 return ERR_PTR(-ENODEV);
1678 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1679 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1681 if (port_no >= DP_MAX_PORTS)
1682 return ERR_PTR(-EFBIG);
1684 dp = get_dp(ovs_header->dp_ifindex);
1686 return ERR_PTR(-ENODEV);
1688 vport = get_vport_protected(dp, port_no);
1690 return ERR_PTR(-ENOENT);
1693 return ERR_PTR(-EINVAL);
1696 /* Called with RTNL lock. */
1697 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1701 if (a[OVS_VPORT_ATTR_STATS])
1702 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1704 if (a[OVS_VPORT_ATTR_ADDRESS])
1705 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1710 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1712 struct nlattr **a = info->attrs;
1713 struct ovs_header *ovs_header = info->userhdr;
1714 struct vport_parms parms;
1715 struct sk_buff *reply;
1716 struct vport *vport;
1717 struct datapath *dp;
1722 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1723 !a[OVS_VPORT_ATTR_UPCALL_PID])
1726 err = ovs_vport_cmd_validate(a);
1731 dp = get_dp(ovs_header->dp_ifindex);
1736 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1737 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1740 if (port_no >= DP_MAX_PORTS)
1743 vport = get_vport_protected(dp, port_no);
1748 for (port_no = 1; ; port_no++) {
1749 if (port_no >= DP_MAX_PORTS) {
1753 vport = get_vport_protected(dp, port_no);
1759 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1760 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1761 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1763 parms.port_no = port_no;
1764 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1766 vport = new_vport(&parms);
1767 err = PTR_ERR(vport);
1771 dp_sysfs_add_if(vport);
1773 err = change_vport(vport, a);
1775 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1776 info->snd_seq, OVS_VPORT_CMD_NEW);
1778 err = PTR_ERR(reply);
1781 dp_detach_port(vport);
1784 genl_notify(reply, genl_info_net(info), info->snd_pid,
1785 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1794 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1796 struct nlattr **a = info->attrs;
1797 struct sk_buff *reply;
1798 struct vport *vport;
1801 err = ovs_vport_cmd_validate(a);
1806 vport = lookup_vport(info->userhdr, a);
1807 err = PTR_ERR(vport);
1812 if (a[OVS_VPORT_ATTR_TYPE] && nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport))
1814 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1815 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1817 err = change_vport(vport, a);
1818 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1819 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1821 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1823 if (IS_ERR(reply)) {
1824 err = PTR_ERR(reply);
1825 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1826 dp_vport_multicast_group.id, err);
1830 genl_notify(reply, genl_info_net(info), info->snd_pid,
1831 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1839 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1841 struct nlattr **a = info->attrs;
1842 struct sk_buff *reply;
1843 struct vport *vport;
1846 err = ovs_vport_cmd_validate(a);
1851 vport = lookup_vport(info->userhdr, a);
1852 err = PTR_ERR(vport);
1856 if (vport->port_no == OVSP_LOCAL) {
1861 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1863 err = PTR_ERR(reply);
1867 dp_detach_port(vport);
1869 genl_notify(reply, genl_info_net(info), info->snd_pid,
1870 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1878 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1880 struct nlattr **a = info->attrs;
1881 struct ovs_header *ovs_header = info->userhdr;
1882 struct sk_buff *reply;
1883 struct vport *vport;
1886 err = ovs_vport_cmd_validate(a);
1891 vport = lookup_vport(ovs_header, a);
1892 err = PTR_ERR(vport);
1896 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1898 err = PTR_ERR(reply);
1904 return genlmsg_reply(reply, info);
1912 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1914 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1915 struct datapath *dp;
1919 dp = get_dp(ovs_header->dp_ifindex);
1924 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1925 struct vport *vport;
1927 vport = get_vport_protected(dp, port_no);
1931 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1932 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1933 OVS_VPORT_CMD_NEW) < 0)
1938 cb->args[0] = port_no;
1944 static struct genl_ops dp_vport_genl_ops[] = {
1945 { .cmd = OVS_VPORT_CMD_NEW,
1946 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1947 .policy = vport_policy,
1948 .doit = ovs_vport_cmd_new
1950 { .cmd = OVS_VPORT_CMD_DEL,
1951 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1952 .policy = vport_policy,
1953 .doit = ovs_vport_cmd_del
1955 { .cmd = OVS_VPORT_CMD_GET,
1956 .flags = 0, /* OK for unprivileged users. */
1957 .policy = vport_policy,
1958 .doit = ovs_vport_cmd_get,
1959 .dumpit = ovs_vport_cmd_dump
1961 { .cmd = OVS_VPORT_CMD_SET,
1962 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1963 .policy = vport_policy,
1964 .doit = ovs_vport_cmd_set,
1968 struct genl_family_and_ops {
1969 struct genl_family *family;
1970 struct genl_ops *ops;
1972 struct genl_multicast_group *group;
1975 static const struct genl_family_and_ops dp_genl_families[] = {
1976 { &dp_datapath_genl_family,
1977 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1978 &dp_datapath_multicast_group },
1979 { &dp_vport_genl_family,
1980 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1981 &dp_vport_multicast_group },
1982 { &dp_flow_genl_family,
1983 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1984 &dp_flow_multicast_group },
1985 { &dp_packet_genl_family,
1986 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1990 static void dp_unregister_genl(int n_families)
1994 for (i = 0; i < n_families; i++)
1995 genl_unregister_family(dp_genl_families[i].family);
1998 static int dp_register_genl(void)
2005 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2006 const struct genl_family_and_ops *f = &dp_genl_families[i];
2008 err = genl_register_family_with_ops(f->family, f->ops,
2015 err = genl_register_mc_group(f->family, f->group);
2024 dp_unregister_genl(n_registered);
2028 static int __init dp_init(void)
2030 struct sk_buff *dummy_skb;
2033 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2035 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2043 goto error_tnl_exit;
2047 goto error_flow_exit;
2049 err = register_netdevice_notifier(&dp_device_notifier);
2051 goto error_vport_exit;
2053 err = dp_register_genl();
2055 goto error_unreg_notifier;
2059 error_unreg_notifier:
2060 unregister_netdevice_notifier(&dp_device_notifier);
2071 static void dp_cleanup(void)
2074 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2075 unregister_netdevice_notifier(&dp_device_notifier);
2081 module_init(dp_init);
2082 module_exit(dp_cleanup);
2084 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2085 MODULE_LICENSE("GPL");