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,
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 static const struct nla_policy sample_policy[OVS_SAMPLE_ATTR_MAX + 1] =
517 [OVS_SAMPLE_ATTR_PROBABILITY] = {.type = NLA_U32 },
518 [OVS_SAMPLE_ATTR_ACTIONS] = {.type = NLA_UNSPEC },
520 struct nlattr *a[OVS_SAMPLE_ATTR_MAX + 1];
523 error = nla_parse_nested(a, OVS_SAMPLE_ATTR_MAX, attr, sample_policy);
527 if (!a[OVS_SAMPLE_ATTR_PROBABILITY])
529 if (!a[OVS_SAMPLE_ATTR_ACTIONS])
532 return validate_actions(a[OVS_SAMPLE_ATTR_ACTIONS], key, (depth + 1));
535 static int validate_action_key(const struct nlattr *a,
536 const struct sw_flow_key *flow_key)
538 int act_type = nla_type(a);
539 const struct nlattr *ovs_key = nla_data(a);
540 int key_type = nla_type(ovs_key);
542 /* There can be only one key in a action */
543 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
546 if (key_type > OVS_KEY_ATTR_MAX ||
547 nla_len(ovs_key) != ovs_key_lens[key_type])
550 #define ACTION(act, key) ((act << 8) | key)
552 switch(ACTION(act_type, key_type)) {
553 const struct ovs_key_ipv4 *ipv4_key;
554 const struct ovs_key_8021q *q_key;
556 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TUN_ID):
557 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_ETHERNET):
560 case ACTION(OVS_ACTION_ATTR_PUSH, OVS_KEY_ATTR_8021Q):
561 q_key = nla_data(ovs_key);
562 if (q_key->q_tpid != htons(ETH_P_8021Q))
565 if (q_key->q_tci & htons(VLAN_TAG_PRESENT))
569 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_IPV4):
570 if (flow_key->eth.type != htons(ETH_P_IP))
573 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
576 ipv4_key = nla_data(ovs_key);
577 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
580 if (ipv4_key->ipv4_tos & INET_ECN_MASK)
583 if (ipv4_key->ipv4_frag !=
584 (flow_key->ip.tos_frag & OVS_FRAG_TYPE_MASK))
589 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_TCP):
590 if (flow_key->ip.proto != IPPROTO_TCP)
593 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
598 case ACTION(OVS_ACTION_ATTR_SET, OVS_KEY_ATTR_UDP):
599 if (flow_key->ip.proto != IPPROTO_UDP)
602 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
613 static int validate_userspace(const struct nlattr *attr)
615 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =
617 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
618 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
620 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
623 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX, attr, userspace_policy);
627 if (!a[OVS_USERSPACE_ATTR_PID] || !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
633 static int validate_actions(const struct nlattr *attr,
634 const struct sw_flow_key *key, int depth)
636 const struct nlattr *a;
639 if (depth >= SAMPLE_ACTION_DEPTH)
642 nla_for_each_nested(a, attr, rem) {
643 /* Expected argument lengths, (u32)-1 for variable length. */
644 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
645 [OVS_ACTION_ATTR_OUTPUT] = 4,
646 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
647 [OVS_ACTION_ATTR_PUSH] = (u32)-1,
648 [OVS_ACTION_ATTR_POP] = 2,
649 [OVS_ACTION_ATTR_SET] = (u32)-1,
650 [OVS_ACTION_ATTR_SET_PRIORITY] = 4,
651 [OVS_ACTION_ATTR_POP_PRIORITY] = 0,
652 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
654 int type = nla_type(a);
656 if (type > OVS_ACTION_ATTR_MAX ||
657 (action_lens[type] != nla_len(a) &&
658 action_lens[type] != (u32)-1))
662 case OVS_ACTION_ATTR_UNSPEC:
665 case OVS_ACTION_ATTR_SET_PRIORITY:
666 case OVS_ACTION_ATTR_POP_PRIORITY:
667 /* No validation needed. */
670 case OVS_ACTION_ATTR_USERSPACE:
671 err = validate_userspace(a);
676 case OVS_ACTION_ATTR_OUTPUT:
677 if (nla_get_u32(a) >= DP_MAX_PORTS)
682 case OVS_ACTION_ATTR_POP:
683 if (nla_get_u16(a) != OVS_KEY_ATTR_8021Q)
687 case OVS_ACTION_ATTR_SET:
688 case OVS_ACTION_ATTR_PUSH:
689 err = validate_action_key(a, key);
694 case OVS_ACTION_ATTR_SAMPLE:
695 err = validate_sample(a, key, depth);
711 static void clear_stats(struct sw_flow *flow)
715 flow->packet_count = 0;
716 flow->byte_count = 0;
719 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
721 struct ovs_header *ovs_header = info->userhdr;
722 struct nlattr **a = info->attrs;
723 struct sw_flow_actions *acts;
724 struct sk_buff *packet;
725 struct sw_flow *flow;
733 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
734 !a[OVS_PACKET_ATTR_ACTIONS] ||
735 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
738 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
739 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
743 skb_reserve(packet, NET_IP_ALIGN);
745 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
747 skb_reset_mac_header(packet);
748 eth = eth_hdr(packet);
750 /* Normally, setting the skb 'protocol' field would be handled by a
751 * call to eth_type_trans(), but it assumes there's a sending
752 * device, which we may not have. */
753 if (ntohs(eth->h_proto) >= 1536)
754 packet->protocol = eth->h_proto;
756 packet->protocol = htons(ETH_P_802_2);
758 /* Build an sw_flow for sending this packet. */
764 err = flow_extract(packet, -1, &flow->key, &key_len);
768 err = flow_metadata_from_nlattrs(&flow->key.eth.in_port,
769 &flow->key.eth.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;
789 dp = get_dp(ovs_header->dp_ifindex);
794 if (flow->key.eth.in_port < DP_MAX_PORTS)
795 OVS_CB(packet)->vport = get_vport_protected(dp,
796 flow->key.eth.in_port);
799 err = execute_actions(dp, packet);
816 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
817 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
818 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
819 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
822 static struct genl_ops dp_packet_genl_ops[] = {
823 { .cmd = OVS_PACKET_CMD_EXECUTE,
824 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
825 .policy = packet_policy,
826 .doit = ovs_packet_cmd_execute
830 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
833 struct flow_table *table = get_table_protected(dp);
835 stats->n_flows = flow_tbl_count(table);
837 stats->n_hit = stats->n_missed = stats->n_lost = 0;
838 for_each_possible_cpu(i) {
839 const struct dp_stats_percpu *percpu_stats;
840 struct dp_stats_percpu local_stats;
843 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
846 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
847 local_stats = *percpu_stats;
848 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
850 stats->n_hit += local_stats.n_hit;
851 stats->n_missed += local_stats.n_missed;
852 stats->n_lost += local_stats.n_lost;
856 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
857 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
858 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
859 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
862 static struct genl_family dp_flow_genl_family = {
863 .id = GENL_ID_GENERATE,
864 .hdrsize = sizeof(struct ovs_header),
865 .name = OVS_FLOW_FAMILY,
867 .maxattr = OVS_FLOW_ATTR_MAX
870 static struct genl_multicast_group dp_flow_multicast_group = {
871 .name = OVS_FLOW_MCGROUP
874 /* Called with genl_lock. */
875 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
876 struct sk_buff *skb, u32 pid, u32 seq, u32 flags, u8 cmd)
878 const int skb_orig_len = skb->len;
879 const struct sw_flow_actions *sf_acts;
880 struct ovs_flow_stats stats;
881 struct ovs_header *ovs_header;
887 sf_acts = rcu_dereference_protected(flow->sf_acts,
888 lockdep_genl_is_held());
890 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
894 ovs_header->dp_ifindex = get_dpifindex(dp);
896 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
898 goto nla_put_failure;
899 err = flow_to_nlattrs(&flow->key, skb);
902 nla_nest_end(skb, nla);
904 spin_lock_bh(&flow->lock);
906 stats.n_packets = flow->packet_count;
907 stats.n_bytes = flow->byte_count;
908 tcp_flags = flow->tcp_flags;
909 spin_unlock_bh(&flow->lock);
912 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
915 NLA_PUT(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats);
918 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
920 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
921 * this is the first flow to be dumped into 'skb'. This is unusual for
922 * Netlink but individual action lists can be longer than
923 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
924 * The userspace caller can always fetch the actions separately if it
925 * really wants them. (Most userspace callers in fact don't care.)
927 * This can only fail for dump operations because the skb is always
928 * properly sized for single flows.
930 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
932 if (err < 0 && skb_orig_len)
935 return genlmsg_end(skb, ovs_header);
940 genlmsg_cancel(skb, ovs_header);
944 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
946 const struct sw_flow_actions *sf_acts;
949 sf_acts = rcu_dereference_protected(flow->sf_acts,
950 lockdep_genl_is_held());
952 len = nla_total_size(FLOW_BUFSIZE); /* OVS_FLOW_ATTR_KEY */
953 len += nla_total_size(sf_acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
954 len += nla_total_size(sizeof(struct ovs_flow_stats)); /* OVS_FLOW_ATTR_STATS */
955 len += nla_total_size(1); /* OVS_FLOW_ATTR_TCP_FLAGS */
956 len += nla_total_size(8); /* OVS_FLOW_ATTR_USED */
957 return genlmsg_new(NLMSG_ALIGN(sizeof(struct ovs_header)) + len, GFP_KERNEL);
960 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp,
961 u32 pid, u32 seq, u8 cmd)
966 skb = ovs_flow_cmd_alloc_info(flow);
968 return ERR_PTR(-ENOMEM);
970 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
975 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
977 struct nlattr **a = info->attrs;
978 struct ovs_header *ovs_header = info->userhdr;
979 struct sw_flow_key key;
980 struct sw_flow *flow;
981 struct sk_buff *reply;
983 struct flow_table *table;
989 if (!a[OVS_FLOW_ATTR_KEY])
991 error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
995 /* Validate actions. */
996 if (a[OVS_FLOW_ATTR_ACTIONS]) {
997 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
1000 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1005 dp = get_dp(ovs_header->dp_ifindex);
1010 table = get_table_protected(dp);
1011 flow = flow_tbl_lookup(table, &key, key_len);
1013 struct sw_flow_actions *acts;
1015 /* Bail out if we're not allowed to create a new flow. */
1017 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1020 /* Expand table, if necessary, to make room. */
1021 if (flow_tbl_need_to_expand(table)) {
1022 struct flow_table *new_table;
1024 new_table = flow_tbl_expand(table);
1025 if (!IS_ERR(new_table)) {
1026 rcu_assign_pointer(dp->table, new_table);
1027 flow_tbl_deferred_destroy(table);
1028 table = get_table_protected(dp);
1032 /* Allocate flow. */
1033 flow = flow_alloc();
1035 error = PTR_ERR(flow);
1041 /* Obtain actions. */
1042 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1043 error = PTR_ERR(acts);
1045 goto error_free_flow;
1046 rcu_assign_pointer(flow->sf_acts, acts);
1048 /* Put flow in bucket. */
1049 flow->hash = flow_hash(&key, key_len);
1050 flow_tbl_insert(table, flow);
1052 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1053 info->snd_seq, OVS_FLOW_CMD_NEW);
1055 /* We found a matching flow. */
1056 struct sw_flow_actions *old_acts;
1058 /* Bail out if we're not allowed to modify an existing flow.
1059 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1060 * because Generic Netlink treats the latter as a dump
1061 * request. We also accept NLM_F_EXCL in case that bug ever
1065 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1066 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1069 /* Update actions. */
1070 old_acts = rcu_dereference_protected(flow->sf_acts,
1071 lockdep_genl_is_held());
1072 if (a[OVS_FLOW_ATTR_ACTIONS] &&
1073 (old_acts->actions_len != nla_len(a[OVS_FLOW_ATTR_ACTIONS]) ||
1074 memcmp(old_acts->actions, nla_data(a[OVS_FLOW_ATTR_ACTIONS]),
1075 old_acts->actions_len))) {
1076 struct sw_flow_actions *new_acts;
1078 new_acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1079 error = PTR_ERR(new_acts);
1080 if (IS_ERR(new_acts))
1083 rcu_assign_pointer(flow->sf_acts, new_acts);
1084 flow_deferred_free_acts(old_acts);
1087 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1088 info->snd_seq, OVS_FLOW_CMD_NEW);
1091 if (a[OVS_FLOW_ATTR_CLEAR]) {
1092 spin_lock_bh(&flow->lock);
1094 spin_unlock_bh(&flow->lock);
1099 genl_notify(reply, genl_info_net(info), info->snd_pid,
1100 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1102 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1103 dp_flow_multicast_group.id, PTR_ERR(reply));
1112 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1114 struct nlattr **a = info->attrs;
1115 struct ovs_header *ovs_header = info->userhdr;
1116 struct sw_flow_key key;
1117 struct sk_buff *reply;
1118 struct sw_flow *flow;
1119 struct datapath *dp;
1120 struct flow_table *table;
1124 if (!a[OVS_FLOW_ATTR_KEY])
1126 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1130 dp = get_dp(ovs_header->dp_ifindex);
1134 table = get_table_protected(dp);
1135 flow = flow_tbl_lookup(table, &key, key_len);
1139 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid, info->snd_seq, OVS_FLOW_CMD_NEW);
1141 return PTR_ERR(reply);
1143 return genlmsg_reply(reply, info);
1146 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1148 struct nlattr **a = info->attrs;
1149 struct ovs_header *ovs_header = info->userhdr;
1150 struct sw_flow_key key;
1151 struct sk_buff *reply;
1152 struct sw_flow *flow;
1153 struct datapath *dp;
1154 struct flow_table *table;
1158 if (!a[OVS_FLOW_ATTR_KEY])
1159 return flush_flows(ovs_header->dp_ifindex);
1160 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1164 dp = get_dp(ovs_header->dp_ifindex);
1168 table = get_table_protected(dp);
1169 flow = flow_tbl_lookup(table, &key, key_len);
1173 reply = ovs_flow_cmd_alloc_info(flow);
1177 flow_tbl_remove(table, flow);
1179 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1180 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1183 flow_deferred_free(flow);
1185 genl_notify(reply, genl_info_net(info), info->snd_pid,
1186 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1190 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1192 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1193 struct datapath *dp;
1195 dp = get_dp(ovs_header->dp_ifindex);
1200 struct sw_flow *flow;
1203 bucket = cb->args[0];
1205 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1209 if (ovs_flow_cmd_fill_info(flow, dp, skb, NETLINK_CB(cb->skb).pid,
1210 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1211 OVS_FLOW_CMD_NEW) < 0)
1214 cb->args[0] = bucket;
1220 static struct genl_ops dp_flow_genl_ops[] = {
1221 { .cmd = OVS_FLOW_CMD_NEW,
1222 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1223 .policy = flow_policy,
1224 .doit = ovs_flow_cmd_new_or_set
1226 { .cmd = OVS_FLOW_CMD_DEL,
1227 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1228 .policy = flow_policy,
1229 .doit = ovs_flow_cmd_del
1231 { .cmd = OVS_FLOW_CMD_GET,
1232 .flags = 0, /* OK for unprivileged users. */
1233 .policy = flow_policy,
1234 .doit = ovs_flow_cmd_get,
1235 .dumpit = ovs_flow_cmd_dump
1237 { .cmd = OVS_FLOW_CMD_SET,
1238 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1239 .policy = flow_policy,
1240 .doit = ovs_flow_cmd_new_or_set,
1244 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1245 #ifdef HAVE_NLA_NUL_STRING
1246 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1248 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1251 static struct genl_family dp_datapath_genl_family = {
1252 .id = GENL_ID_GENERATE,
1253 .hdrsize = sizeof(struct ovs_header),
1254 .name = OVS_DATAPATH_FAMILY,
1256 .maxattr = OVS_DP_ATTR_MAX
1259 static struct genl_multicast_group dp_datapath_multicast_group = {
1260 .name = OVS_DATAPATH_MCGROUP
1263 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1264 u32 pid, u32 seq, u32 flags, u8 cmd)
1266 struct ovs_header *ovs_header;
1270 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1275 ovs_header->dp_ifindex = get_dpifindex(dp);
1278 err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1281 goto nla_put_failure;
1283 nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1285 goto nla_put_failure;
1286 get_dp_stats(dp, nla_data(nla));
1288 return genlmsg_end(skb, ovs_header);
1291 genlmsg_cancel(skb, ovs_header);
1296 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1299 struct sk_buff *skb;
1302 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1304 return ERR_PTR(-ENOMEM);
1306 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1309 return ERR_PTR(retval);
1314 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1316 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1319 /* Called with genl_mutex and optionally with RTNL lock also. */
1320 static struct datapath *lookup_datapath(struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1322 struct datapath *dp;
1324 if (!a[OVS_DP_ATTR_NAME])
1325 dp = get_dp(ovs_header->dp_ifindex);
1327 struct vport *vport;
1330 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1331 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1334 return dp ? dp : ERR_PTR(-ENODEV);
1337 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1339 struct nlattr **a = info->attrs;
1340 struct vport_parms parms;
1341 struct sk_buff *reply;
1342 struct datapath *dp;
1343 struct vport *vport;
1347 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1350 err = ovs_dp_cmd_validate(a);
1356 if (!try_module_get(THIS_MODULE))
1357 goto err_unlock_rtnl;
1360 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1362 goto err_put_module;
1363 INIT_LIST_HEAD(&dp->port_list);
1365 /* Initialize kobject for bridge. This will be added as
1366 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1367 dp->ifobj.kset = NULL;
1368 kobject_init(&dp->ifobj, &dp_ktype);
1370 /* Allocate table. */
1372 rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1376 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1377 if (!dp->stats_percpu) {
1379 goto err_destroy_table;
1382 /* Set up our datapath device. */
1383 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1384 parms.type = OVS_VPORT_TYPE_INTERNAL;
1385 parms.options = NULL;
1387 parms.port_no = OVSP_LOCAL;
1388 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1390 vport = new_vport(&parms);
1391 if (IS_ERR(vport)) {
1392 err = PTR_ERR(vport);
1396 goto err_destroy_percpu;
1399 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1400 err = PTR_ERR(reply);
1402 goto err_destroy_local_port;
1404 list_add_tail(&dp->list_node, &dps);
1405 dp_sysfs_add_dp(dp);
1409 genl_notify(reply, genl_info_net(info), info->snd_pid,
1410 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1413 err_destroy_local_port:
1414 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1416 free_percpu(dp->stats_percpu);
1418 flow_tbl_destroy(get_table_protected(dp));
1422 module_put(THIS_MODULE);
1429 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1431 struct vport *vport, *next_vport;
1432 struct sk_buff *reply;
1433 struct datapath *dp;
1436 err = ovs_dp_cmd_validate(info->attrs);
1441 dp = lookup_datapath(info->userhdr, info->attrs);
1446 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_DEL);
1447 err = PTR_ERR(reply);
1451 list_for_each_entry_safe (vport, next_vport, &dp->port_list, node)
1452 if (vport->port_no != OVSP_LOCAL)
1453 dp_detach_port(vport);
1455 dp_sysfs_del_dp(dp);
1456 list_del(&dp->list_node);
1457 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1459 /* rtnl_unlock() will wait until all the references to devices that
1460 * are pending unregistration have been dropped. We do it here to
1461 * ensure that any internal devices (which contain DP pointers) are
1462 * fully destroyed before freeing the datapath.
1466 call_rcu(&dp->rcu, destroy_dp_rcu);
1467 module_put(THIS_MODULE);
1469 genl_notify(reply, genl_info_net(info), info->snd_pid,
1470 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1480 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1482 struct sk_buff *reply;
1483 struct datapath *dp;
1486 err = ovs_dp_cmd_validate(info->attrs);
1490 dp = lookup_datapath(info->userhdr, info->attrs);
1494 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1495 if (IS_ERR(reply)) {
1496 err = PTR_ERR(reply);
1497 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1498 dp_datapath_multicast_group.id, err);
1502 genl_notify(reply, genl_info_net(info), info->snd_pid,
1503 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1507 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1509 struct sk_buff *reply;
1510 struct datapath *dp;
1513 err = ovs_dp_cmd_validate(info->attrs);
1517 dp = lookup_datapath(info->userhdr, info->attrs);
1521 reply = ovs_dp_cmd_build_info(dp, info->snd_pid, info->snd_seq, OVS_DP_CMD_NEW);
1523 return PTR_ERR(reply);
1525 return genlmsg_reply(reply, info);
1528 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1530 struct datapath *dp;
1531 int skip = cb->args[0];
1534 list_for_each_entry (dp, &dps, list_node) {
1537 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1538 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1539 OVS_DP_CMD_NEW) < 0)
1549 static struct genl_ops dp_datapath_genl_ops[] = {
1550 { .cmd = OVS_DP_CMD_NEW,
1551 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1552 .policy = datapath_policy,
1553 .doit = ovs_dp_cmd_new
1555 { .cmd = OVS_DP_CMD_DEL,
1556 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1557 .policy = datapath_policy,
1558 .doit = ovs_dp_cmd_del
1560 { .cmd = OVS_DP_CMD_GET,
1561 .flags = 0, /* OK for unprivileged users. */
1562 .policy = datapath_policy,
1563 .doit = ovs_dp_cmd_get,
1564 .dumpit = ovs_dp_cmd_dump
1566 { .cmd = OVS_DP_CMD_SET,
1567 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1568 .policy = datapath_policy,
1569 .doit = ovs_dp_cmd_set,
1573 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1574 #ifdef HAVE_NLA_NUL_STRING
1575 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1576 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1577 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1579 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1580 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1582 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1583 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1584 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1585 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1588 static struct genl_family dp_vport_genl_family = {
1589 .id = GENL_ID_GENERATE,
1590 .hdrsize = sizeof(struct ovs_header),
1591 .name = OVS_VPORT_FAMILY,
1593 .maxattr = OVS_VPORT_ATTR_MAX
1596 struct genl_multicast_group dp_vport_multicast_group = {
1597 .name = OVS_VPORT_MCGROUP
1600 /* Called with RTNL lock or RCU read lock. */
1601 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1602 u32 pid, u32 seq, u32 flags, u8 cmd)
1604 struct ovs_header *ovs_header;
1608 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1613 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1615 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1616 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1617 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1618 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1620 nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats));
1622 goto nla_put_failure;
1624 vport_get_stats(vport, nla_data(nla));
1626 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1628 err = vport_get_options(vport, skb);
1629 if (err == -EMSGSIZE)
1632 return genlmsg_end(skb, ovs_header);
1637 genlmsg_cancel(skb, ovs_header);
1641 /* Called with RTNL lock or RCU read lock. */
1642 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1645 struct sk_buff *skb;
1648 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1650 return ERR_PTR(-ENOMEM);
1652 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1655 return ERR_PTR(retval);
1660 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1662 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1665 /* Called with RTNL lock or RCU read lock. */
1666 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1667 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1669 struct datapath *dp;
1670 struct vport *vport;
1672 if (a[OVS_VPORT_ATTR_NAME]) {
1673 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1675 return ERR_PTR(-ENODEV);
1677 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1678 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1680 if (port_no >= DP_MAX_PORTS)
1681 return ERR_PTR(-EFBIG);
1683 dp = get_dp(ovs_header->dp_ifindex);
1685 return ERR_PTR(-ENODEV);
1687 vport = get_vport_protected(dp, port_no);
1689 return ERR_PTR(-ENOENT);
1692 return ERR_PTR(-EINVAL);
1695 /* Called with RTNL lock. */
1696 static int change_vport(struct vport *vport, struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1700 if (a[OVS_VPORT_ATTR_STATS])
1701 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1703 if (a[OVS_VPORT_ATTR_ADDRESS])
1704 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1709 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1711 struct nlattr **a = info->attrs;
1712 struct ovs_header *ovs_header = info->userhdr;
1713 struct vport_parms parms;
1714 struct sk_buff *reply;
1715 struct vport *vport;
1716 struct datapath *dp;
1721 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1722 !a[OVS_VPORT_ATTR_UPCALL_PID])
1725 err = ovs_vport_cmd_validate(a);
1730 dp = get_dp(ovs_header->dp_ifindex);
1735 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1736 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1739 if (port_no >= DP_MAX_PORTS)
1742 vport = get_vport_protected(dp, port_no);
1747 for (port_no = 1; ; port_no++) {
1748 if (port_no >= DP_MAX_PORTS) {
1752 vport = get_vport_protected(dp, port_no);
1758 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1759 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1760 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1762 parms.port_no = port_no;
1763 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1765 vport = new_vport(&parms);
1766 err = PTR_ERR(vport);
1770 dp_sysfs_add_if(vport);
1772 err = change_vport(vport, a);
1774 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1775 info->snd_seq, OVS_VPORT_CMD_NEW);
1777 err = PTR_ERR(reply);
1780 dp_detach_port(vport);
1783 genl_notify(reply, genl_info_net(info), info->snd_pid,
1784 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1793 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1795 struct nlattr **a = info->attrs;
1796 struct sk_buff *reply;
1797 struct vport *vport;
1800 err = ovs_vport_cmd_validate(a);
1805 vport = lookup_vport(info->userhdr, a);
1806 err = PTR_ERR(vport);
1811 if (a[OVS_VPORT_ATTR_TYPE] && nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport))
1813 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1814 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1816 err = change_vport(vport, a);
1817 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1818 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1820 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1822 if (IS_ERR(reply)) {
1823 err = PTR_ERR(reply);
1824 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1825 dp_vport_multicast_group.id, err);
1829 genl_notify(reply, genl_info_net(info), info->snd_pid,
1830 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1838 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1840 struct nlattr **a = info->attrs;
1841 struct sk_buff *reply;
1842 struct vport *vport;
1845 err = ovs_vport_cmd_validate(a);
1850 vport = lookup_vport(info->userhdr, a);
1851 err = PTR_ERR(vport);
1855 if (vport->port_no == OVSP_LOCAL) {
1860 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1862 err = PTR_ERR(reply);
1866 dp_detach_port(vport);
1868 genl_notify(reply, genl_info_net(info), info->snd_pid,
1869 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1877 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1879 struct nlattr **a = info->attrs;
1880 struct ovs_header *ovs_header = info->userhdr;
1881 struct sk_buff *reply;
1882 struct vport *vport;
1885 err = ovs_vport_cmd_validate(a);
1890 vport = lookup_vport(ovs_header, a);
1891 err = PTR_ERR(vport);
1895 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1897 err = PTR_ERR(reply);
1903 return genlmsg_reply(reply, info);
1911 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1913 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1914 struct datapath *dp;
1918 dp = get_dp(ovs_header->dp_ifindex);
1923 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1924 struct vport *vport;
1926 vport = get_vport_protected(dp, port_no);
1930 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1931 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1932 OVS_VPORT_CMD_NEW) < 0)
1937 cb->args[0] = port_no;
1943 static struct genl_ops dp_vport_genl_ops[] = {
1944 { .cmd = OVS_VPORT_CMD_NEW,
1945 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1946 .policy = vport_policy,
1947 .doit = ovs_vport_cmd_new
1949 { .cmd = OVS_VPORT_CMD_DEL,
1950 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1951 .policy = vport_policy,
1952 .doit = ovs_vport_cmd_del
1954 { .cmd = OVS_VPORT_CMD_GET,
1955 .flags = 0, /* OK for unprivileged users. */
1956 .policy = vport_policy,
1957 .doit = ovs_vport_cmd_get,
1958 .dumpit = ovs_vport_cmd_dump
1960 { .cmd = OVS_VPORT_CMD_SET,
1961 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1962 .policy = vport_policy,
1963 .doit = ovs_vport_cmd_set,
1967 struct genl_family_and_ops {
1968 struct genl_family *family;
1969 struct genl_ops *ops;
1971 struct genl_multicast_group *group;
1974 static const struct genl_family_and_ops dp_genl_families[] = {
1975 { &dp_datapath_genl_family,
1976 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1977 &dp_datapath_multicast_group },
1978 { &dp_vport_genl_family,
1979 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1980 &dp_vport_multicast_group },
1981 { &dp_flow_genl_family,
1982 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1983 &dp_flow_multicast_group },
1984 { &dp_packet_genl_family,
1985 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1989 static void dp_unregister_genl(int n_families)
1993 for (i = 0; i < n_families; i++)
1994 genl_unregister_family(dp_genl_families[i].family);
1997 static int dp_register_genl(void)
2004 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2005 const struct genl_family_and_ops *f = &dp_genl_families[i];
2007 err = genl_register_family_with_ops(f->family, f->ops,
2014 err = genl_register_mc_group(f->family, f->group);
2023 dp_unregister_genl(n_registered);
2027 static int __init dp_init(void)
2029 struct sk_buff *dummy_skb;
2032 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2034 printk("Open vSwitch %s, built "__DATE__" "__TIME__"\n", VERSION BUILDNR);
2042 goto error_tnl_exit;
2046 goto error_flow_exit;
2048 err = register_netdevice_notifier(&dp_device_notifier);
2050 goto error_vport_exit;
2052 err = dp_register_genl();
2054 goto error_unreg_notifier;
2058 error_unreg_notifier:
2059 unregister_netdevice_notifier(&dp_device_notifier);
2070 static void dp_cleanup(void)
2073 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2074 unregister_netdevice_notifier(&dp_device_notifier);
2080 module_init(dp_init);
2081 module_exit(dp_cleanup);
2083 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2084 MODULE_LICENSE("GPL");