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>
36 #include <linux/highmem.h>
37 #include <linux/netfilter_bridge.h>
38 #include <linux/netfilter_ipv4.h>
39 #include <linux/inetdevice.h>
40 #include <linux/list.h>
41 #include <linux/openvswitch.h>
42 #include <linux/rculist.h>
43 #include <linux/dmi.h>
44 #include <net/inet_ecn.h>
45 #include <net/genetlink.h>
53 #include "vport-internal_dev.h"
55 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
56 LINUX_VERSION_CODE >= KERNEL_VERSION(3,2,0)
57 #error Kernels before 2.6.18 or after 3.1 are not supported by this version of Open vSwitch.
60 int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
61 EXPORT_SYMBOL(dp_ioctl_hook);
66 * Writes to device state (add/remove datapath, port, set operations on vports,
67 * etc.) are protected by RTNL.
69 * Writes to other state (flow table modifications, set miscellaneous datapath
70 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
73 * Reads are protected by RCU.
75 * There are a few special cases (mostly stats) that have their own
76 * synchronization but they nest under all of above and don't interact with
80 /* Global list of datapaths to enable dumping them all out.
81 * Protected by genl_mutex.
83 static LIST_HEAD(dps);
85 static struct vport *new_vport(const struct vport_parms *);
86 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
87 const struct dp_upcall_info *);
88 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
89 const struct dp_upcall_info *);
91 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
92 struct datapath *get_dp(int dp_ifindex)
94 struct datapath *dp = NULL;
95 struct net_device *dev;
98 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
100 struct vport *vport = internal_dev_get_vport(dev);
108 EXPORT_SYMBOL_GPL(get_dp);
110 /* Must be called with genl_mutex. */
111 static struct flow_table *get_table_protected(struct datapath *dp)
113 return rcu_dereference_protected(dp->table, lockdep_genl_is_held());
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 static struct vport *get_vport_protected(struct datapath *dp, u16 port_no)
119 return rcu_dereference_rtnl(dp->ports[port_no]);
122 /* Must be called with rcu_read_lock or RTNL lock. */
123 const char *dp_name(const struct datapath *dp)
125 return vport_get_name(rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]));
128 static int get_dpifindex(struct datapath *dp)
135 local = get_vport_protected(dp, OVSP_LOCAL);
137 ifindex = vport_get_ifindex(local);
146 static size_t br_nlmsg_size(void)
148 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
149 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
150 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
151 + nla_total_size(4) /* IFLA_MASTER */
152 + nla_total_size(4) /* IFLA_MTU */
153 + nla_total_size(1); /* IFLA_OPERSTATE */
156 /* Caller must hold RTNL lock. */
157 static int dp_fill_ifinfo(struct sk_buff *skb,
158 const struct vport *port,
159 int event, unsigned int flags)
161 struct datapath *dp = port->dp;
162 int ifindex = vport_get_ifindex(port);
163 struct ifinfomsg *hdr;
164 struct nlmsghdr *nlh;
169 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
173 hdr = nlmsg_data(nlh);
174 hdr->ifi_family = AF_BRIDGE;
176 hdr->ifi_type = ARPHRD_ETHER;
177 hdr->ifi_index = ifindex;
178 hdr->ifi_flags = vport_get_flags(port);
181 NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port));
182 NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
183 NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port));
184 #ifdef IFLA_OPERSTATE
185 NLA_PUT_U8(skb, IFLA_OPERSTATE,
186 vport_is_running(port)
187 ? vport_get_operstate(port)
191 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, vport_get_addr(port));
193 return nlmsg_end(skb, nlh);
196 nlmsg_cancel(skb, nlh);
200 /* Caller must hold RTNL lock. */
201 static void dp_ifinfo_notify(int event, struct vport *port)
206 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
210 err = dp_fill_ifinfo(skb, port, event, 0);
212 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
213 WARN_ON(err == -EMSGSIZE);
217 rtnl_notify(skb, &init_net, 0, RTNLGRP_LINK, NULL, GFP_KERNEL);
221 rtnl_set_sk_err(&init_net, RTNLGRP_LINK, err);
224 static void release_dp(struct kobject *kobj)
226 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
230 static struct kobj_type dp_ktype = {
231 .release = release_dp
234 static void destroy_dp_rcu(struct rcu_head *rcu)
236 struct datapath *dp = container_of(rcu, struct datapath, rcu);
238 flow_tbl_destroy(dp->table);
239 free_percpu(dp->stats_percpu);
240 kobject_put(&dp->ifobj);
243 /* Called with RTNL lock and genl_lock. */
244 static struct vport *new_vport(const struct vport_parms *parms)
248 vport = vport_add(parms);
249 if (!IS_ERR(vport)) {
250 struct datapath *dp = parms->dp;
252 rcu_assign_pointer(dp->ports[parms->port_no], vport);
253 list_add(&vport->node, &dp->port_list);
255 dp_ifinfo_notify(RTM_NEWLINK, vport);
261 /* Called with RTNL lock. */
262 void dp_detach_port(struct vport *p)
266 if (p->port_no != OVSP_LOCAL)
268 dp_ifinfo_notify(RTM_DELLINK, p);
270 /* First drop references to device. */
272 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
274 /* Then destroy it. */
278 /* Must be called with rcu_read_lock. */
279 void dp_process_received_packet(struct vport *p, struct sk_buff *skb)
281 struct datapath *dp = p->dp;
282 struct sw_flow *flow;
283 struct dp_stats_percpu *stats;
287 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
288 OVS_CB(skb)->vport = p;
290 if (!OVS_CB(skb)->flow) {
291 struct sw_flow_key key;
294 /* Extract flow from 'skb' into 'key'. */
295 error = flow_extract(skb, p->port_no, &key, &key_len);
296 if (unlikely(error)) {
302 flow = flow_tbl_lookup(rcu_dereference(dp->table),
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] = {
622 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
623 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
625 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
628 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
629 attr, userspace_policy);
633 if (!a[OVS_USERSPACE_ATTR_PID] ||
634 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
640 static int validate_actions(const struct nlattr *attr,
641 const struct sw_flow_key *key, int depth)
643 const struct nlattr *a;
646 if (depth >= SAMPLE_ACTION_DEPTH)
649 nla_for_each_nested(a, attr, rem) {
650 /* Expected argument lengths, (u32)-1 for variable length. */
651 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
652 [OVS_ACTION_ATTR_OUTPUT] = 4,
653 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
654 [OVS_ACTION_ATTR_PUSH] = (u32)-1,
655 [OVS_ACTION_ATTR_POP] = 2,
656 [OVS_ACTION_ATTR_SET] = (u32)-1,
657 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
659 int type = nla_type(a);
661 if (type > OVS_ACTION_ATTR_MAX ||
662 (action_lens[type] != nla_len(a) &&
663 action_lens[type] != (u32)-1))
667 case OVS_ACTION_ATTR_UNSPEC:
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.phy.priority,
769 &flow->key.phy.in_port,
770 &flow->key.phy.tun_id,
771 a[OVS_PACKET_ATTR_KEY]);
775 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
779 flow->hash = flow_hash(&flow->key, key_len);
781 acts = flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
785 rcu_assign_pointer(flow->sf_acts, acts);
787 OVS_CB(packet)->flow = flow;
788 packet->priority = flow->key.phy.priority;
791 dp = get_dp(ovs_header->dp_ifindex);
796 if (flow->key.phy.in_port < DP_MAX_PORTS)
797 OVS_CB(packet)->vport = get_vport_protected(dp,
798 flow->key.phy.in_port);
801 err = execute_actions(dp, packet);
818 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
819 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
820 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
821 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
824 static struct genl_ops dp_packet_genl_ops[] = {
825 { .cmd = OVS_PACKET_CMD_EXECUTE,
826 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
827 .policy = packet_policy,
828 .doit = ovs_packet_cmd_execute
832 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
835 struct flow_table *table = get_table_protected(dp);
837 stats->n_flows = flow_tbl_count(table);
839 stats->n_hit = stats->n_missed = stats->n_lost = 0;
840 for_each_possible_cpu(i) {
841 const struct dp_stats_percpu *percpu_stats;
842 struct dp_stats_percpu local_stats;
845 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
848 seqcount = read_seqcount_begin(&percpu_stats->seqlock);
849 local_stats = *percpu_stats;
850 } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount));
852 stats->n_hit += local_stats.n_hit;
853 stats->n_missed += local_stats.n_missed;
854 stats->n_lost += local_stats.n_lost;
858 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
859 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
860 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
861 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
864 static struct genl_family dp_flow_genl_family = {
865 .id = GENL_ID_GENERATE,
866 .hdrsize = sizeof(struct ovs_header),
867 .name = OVS_FLOW_FAMILY,
868 .version = OVS_FLOW_VERSION,
869 .maxattr = OVS_FLOW_ATTR_MAX
872 static struct genl_multicast_group dp_flow_multicast_group = {
873 .name = OVS_FLOW_MCGROUP
876 /* Called with genl_lock. */
877 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
878 struct sk_buff *skb, u32 pid,
879 u32 seq, u32 flags, u8 cmd)
881 const int skb_orig_len = skb->len;
882 const struct sw_flow_actions *sf_acts;
883 struct ovs_flow_stats stats;
884 struct ovs_header *ovs_header;
890 sf_acts = rcu_dereference_protected(flow->sf_acts,
891 lockdep_genl_is_held());
893 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
897 ovs_header->dp_ifindex = get_dpifindex(dp);
899 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
901 goto nla_put_failure;
902 err = flow_to_nlattrs(&flow->key, skb);
905 nla_nest_end(skb, nla);
907 spin_lock_bh(&flow->lock);
909 stats.n_packets = flow->packet_count;
910 stats.n_bytes = flow->byte_count;
911 tcp_flags = flow->tcp_flags;
912 spin_unlock_bh(&flow->lock);
915 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, flow_used_time(used));
918 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
919 sizeof(struct ovs_flow_stats), &stats);
922 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
924 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
925 * this is the first flow to be dumped into 'skb'. This is unusual for
926 * Netlink but individual action lists can be longer than
927 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
928 * The userspace caller can always fetch the actions separately if it
929 * really wants them. (Most userspace callers in fact don't care.)
931 * This can only fail for dump operations because the skb is always
932 * properly sized for single flows.
934 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
936 if (err < 0 && skb_orig_len)
939 return genlmsg_end(skb, ovs_header);
944 genlmsg_cancel(skb, ovs_header);
948 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
950 const struct sw_flow_actions *sf_acts;
953 sf_acts = rcu_dereference_protected(flow->sf_acts,
954 lockdep_genl_is_held());
956 /* OVS_FLOW_ATTR_KEY */
957 len = nla_total_size(FLOW_BUFSIZE);
958 /* OVS_FLOW_ATTR_ACTIONS */
959 len += nla_total_size(sf_acts->actions_len);
960 /* OVS_FLOW_ATTR_STATS */
961 len += nla_total_size(sizeof(struct ovs_flow_stats));
962 /* OVS_FLOW_ATTR_TCP_FLAGS */
963 len += nla_total_size(1);
964 /* OVS_FLOW_ATTR_USED */
965 len += nla_total_size(8);
967 len += NLMSG_ALIGN(sizeof(struct ovs_header));
969 return genlmsg_new(len, GFP_KERNEL);
972 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
974 u32 pid, u32 seq, u8 cmd)
979 skb = ovs_flow_cmd_alloc_info(flow);
981 return ERR_PTR(-ENOMEM);
983 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
988 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
990 struct nlattr **a = info->attrs;
991 struct ovs_header *ovs_header = info->userhdr;
992 struct sw_flow_key key;
993 struct sw_flow *flow;
994 struct sk_buff *reply;
996 struct flow_table *table;
1002 if (!a[OVS_FLOW_ATTR_KEY])
1004 error = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1008 /* Validate actions. */
1009 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1010 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
1013 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1018 dp = get_dp(ovs_header->dp_ifindex);
1023 table = get_table_protected(dp);
1024 flow = flow_tbl_lookup(table, &key, key_len);
1026 struct sw_flow_actions *acts;
1028 /* Bail out if we're not allowed to create a new flow. */
1030 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1033 /* Expand table, if necessary, to make room. */
1034 if (flow_tbl_need_to_expand(table)) {
1035 struct flow_table *new_table;
1037 new_table = flow_tbl_expand(table);
1038 if (!IS_ERR(new_table)) {
1039 rcu_assign_pointer(dp->table, new_table);
1040 flow_tbl_deferred_destroy(table);
1041 table = get_table_protected(dp);
1045 /* Allocate flow. */
1046 flow = flow_alloc();
1048 error = PTR_ERR(flow);
1054 /* Obtain actions. */
1055 acts = flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1056 error = PTR_ERR(acts);
1058 goto error_free_flow;
1059 rcu_assign_pointer(flow->sf_acts, acts);
1061 /* Put flow in bucket. */
1062 flow->hash = flow_hash(&key, key_len);
1063 flow_tbl_insert(table, flow);
1065 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1069 /* We found a matching flow. */
1070 struct sw_flow_actions *old_acts;
1071 struct nlattr *acts_attrs;
1073 /* Bail out if we're not allowed to modify an existing flow.
1074 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1075 * because Generic Netlink treats the latter as a dump
1076 * request. We also accept NLM_F_EXCL in case that bug ever
1080 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1081 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1084 /* Update actions. */
1085 old_acts = rcu_dereference_protected(flow->sf_acts,
1086 lockdep_genl_is_held());
1087 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1089 (old_acts->actions_len != nla_len(acts_attrs) ||
1090 memcmp(old_acts->actions, nla_data(acts_attrs),
1091 old_acts->actions_len))) {
1092 struct sw_flow_actions *new_acts;
1094 new_acts = flow_actions_alloc(acts_attrs);
1095 error = PTR_ERR(new_acts);
1096 if (IS_ERR(new_acts))
1099 rcu_assign_pointer(flow->sf_acts, new_acts);
1100 flow_deferred_free_acts(old_acts);
1103 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1104 info->snd_seq, OVS_FLOW_CMD_NEW);
1107 if (a[OVS_FLOW_ATTR_CLEAR]) {
1108 spin_lock_bh(&flow->lock);
1110 spin_unlock_bh(&flow->lock);
1115 genl_notify(reply, genl_info_net(info), info->snd_pid,
1116 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1118 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1119 dp_flow_multicast_group.id, PTR_ERR(reply));
1128 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1130 struct nlattr **a = info->attrs;
1131 struct ovs_header *ovs_header = info->userhdr;
1132 struct sw_flow_key key;
1133 struct sk_buff *reply;
1134 struct sw_flow *flow;
1135 struct datapath *dp;
1136 struct flow_table *table;
1140 if (!a[OVS_FLOW_ATTR_KEY])
1142 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1146 dp = get_dp(ovs_header->dp_ifindex);
1150 table = get_table_protected(dp);
1151 flow = flow_tbl_lookup(table, &key, key_len);
1155 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1156 info->snd_seq, OVS_FLOW_CMD_NEW);
1158 return PTR_ERR(reply);
1160 return genlmsg_reply(reply, info);
1163 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1165 struct nlattr **a = info->attrs;
1166 struct ovs_header *ovs_header = info->userhdr;
1167 struct sw_flow_key key;
1168 struct sk_buff *reply;
1169 struct sw_flow *flow;
1170 struct datapath *dp;
1171 struct flow_table *table;
1175 if (!a[OVS_FLOW_ATTR_KEY])
1176 return flush_flows(ovs_header->dp_ifindex);
1177 err = flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1181 dp = get_dp(ovs_header->dp_ifindex);
1185 table = get_table_protected(dp);
1186 flow = flow_tbl_lookup(table, &key, key_len);
1190 reply = ovs_flow_cmd_alloc_info(flow);
1194 flow_tbl_remove(table, flow);
1196 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1197 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1200 flow_deferred_free(flow);
1202 genl_notify(reply, genl_info_net(info), info->snd_pid,
1203 dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1207 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1209 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1210 struct datapath *dp;
1212 dp = get_dp(ovs_header->dp_ifindex);
1217 struct sw_flow *flow;
1220 bucket = cb->args[0];
1222 flow = flow_tbl_next(get_table_protected(dp), &bucket, &obj);
1226 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1227 NETLINK_CB(cb->skb).pid,
1228 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1229 OVS_FLOW_CMD_NEW) < 0)
1232 cb->args[0] = bucket;
1238 static struct genl_ops dp_flow_genl_ops[] = {
1239 { .cmd = OVS_FLOW_CMD_NEW,
1240 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1241 .policy = flow_policy,
1242 .doit = ovs_flow_cmd_new_or_set
1244 { .cmd = OVS_FLOW_CMD_DEL,
1245 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1246 .policy = flow_policy,
1247 .doit = ovs_flow_cmd_del
1249 { .cmd = OVS_FLOW_CMD_GET,
1250 .flags = 0, /* OK for unprivileged users. */
1251 .policy = flow_policy,
1252 .doit = ovs_flow_cmd_get,
1253 .dumpit = ovs_flow_cmd_dump
1255 { .cmd = OVS_FLOW_CMD_SET,
1256 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1257 .policy = flow_policy,
1258 .doit = ovs_flow_cmd_new_or_set,
1262 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1263 #ifdef HAVE_NLA_NUL_STRING
1264 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1266 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1269 static struct genl_family dp_datapath_genl_family = {
1270 .id = GENL_ID_GENERATE,
1271 .hdrsize = sizeof(struct ovs_header),
1272 .name = OVS_DATAPATH_FAMILY,
1273 .version = OVS_DATAPATH_VERSION,
1274 .maxattr = OVS_DP_ATTR_MAX
1277 static struct genl_multicast_group dp_datapath_multicast_group = {
1278 .name = OVS_DATAPATH_MCGROUP
1281 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1282 u32 pid, u32 seq, u32 flags, u8 cmd)
1284 struct ovs_header *ovs_header;
1288 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1293 ovs_header->dp_ifindex = get_dpifindex(dp);
1296 err = nla_put_string(skb, OVS_DP_ATTR_NAME, dp_name(dp));
1299 goto nla_put_failure;
1301 nla = nla_reserve(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats));
1303 goto nla_put_failure;
1304 get_dp_stats(dp, nla_data(nla));
1306 return genlmsg_end(skb, ovs_header);
1309 genlmsg_cancel(skb, ovs_header);
1314 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1317 struct sk_buff *skb;
1320 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1322 return ERR_PTR(-ENOMEM);
1324 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1327 return ERR_PTR(retval);
1332 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1334 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1337 /* Called with genl_mutex and optionally with RTNL lock also. */
1338 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1339 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1341 struct datapath *dp;
1343 if (!a[OVS_DP_ATTR_NAME])
1344 dp = get_dp(ovs_header->dp_ifindex);
1346 struct vport *vport;
1349 vport = vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1350 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1353 return dp ? dp : ERR_PTR(-ENODEV);
1356 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1358 struct nlattr **a = info->attrs;
1359 struct vport_parms parms;
1360 struct sk_buff *reply;
1361 struct datapath *dp;
1362 struct vport *vport;
1366 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1369 err = ovs_dp_cmd_validate(a);
1375 if (!try_module_get(THIS_MODULE))
1376 goto err_unlock_rtnl;
1379 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1381 goto err_put_module;
1382 INIT_LIST_HEAD(&dp->port_list);
1384 /* Initialize kobject for bridge. This will be added as
1385 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1386 dp->ifobj.kset = NULL;
1387 kobject_init(&dp->ifobj, &dp_ktype);
1389 /* Allocate table. */
1391 rcu_assign_pointer(dp->table, flow_tbl_alloc(TBL_MIN_BUCKETS));
1395 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1396 if (!dp->stats_percpu) {
1398 goto err_destroy_table;
1401 /* Set up our datapath device. */
1402 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1403 parms.type = OVS_VPORT_TYPE_INTERNAL;
1404 parms.options = NULL;
1406 parms.port_no = OVSP_LOCAL;
1407 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1409 vport = new_vport(&parms);
1410 if (IS_ERR(vport)) {
1411 err = PTR_ERR(vport);
1415 goto err_destroy_percpu;
1418 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1419 info->snd_seq, OVS_DP_CMD_NEW);
1420 err = PTR_ERR(reply);
1422 goto err_destroy_local_port;
1424 list_add_tail(&dp->list_node, &dps);
1425 dp_sysfs_add_dp(dp);
1429 genl_notify(reply, genl_info_net(info), info->snd_pid,
1430 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1433 err_destroy_local_port:
1434 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1436 free_percpu(dp->stats_percpu);
1438 flow_tbl_destroy(get_table_protected(dp));
1442 module_put(THIS_MODULE);
1449 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1451 struct vport *vport, *next_vport;
1452 struct sk_buff *reply;
1453 struct datapath *dp;
1456 err = ovs_dp_cmd_validate(info->attrs);
1461 dp = lookup_datapath(info->userhdr, info->attrs);
1466 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1467 info->snd_seq, OVS_DP_CMD_DEL);
1468 err = PTR_ERR(reply);
1472 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1473 if (vport->port_no != OVSP_LOCAL)
1474 dp_detach_port(vport);
1476 dp_sysfs_del_dp(dp);
1477 list_del(&dp->list_node);
1478 dp_detach_port(get_vport_protected(dp, OVSP_LOCAL));
1480 /* rtnl_unlock() will wait until all the references to devices that
1481 * are pending unregistration have been dropped. We do it here to
1482 * ensure that any internal devices (which contain DP pointers) are
1483 * fully destroyed before freeing the datapath.
1487 call_rcu(&dp->rcu, destroy_dp_rcu);
1488 module_put(THIS_MODULE);
1490 genl_notify(reply, genl_info_net(info), info->snd_pid,
1491 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1501 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1503 struct sk_buff *reply;
1504 struct datapath *dp;
1507 err = ovs_dp_cmd_validate(info->attrs);
1511 dp = lookup_datapath(info->userhdr, info->attrs);
1515 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1516 info->snd_seq, OVS_DP_CMD_NEW);
1517 if (IS_ERR(reply)) {
1518 err = PTR_ERR(reply);
1519 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1520 dp_datapath_multicast_group.id, err);
1524 genl_notify(reply, genl_info_net(info), info->snd_pid,
1525 dp_datapath_multicast_group.id, info->nlhdr, GFP_KERNEL);
1529 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1531 struct sk_buff *reply;
1532 struct datapath *dp;
1535 err = ovs_dp_cmd_validate(info->attrs);
1539 dp = lookup_datapath(info->userhdr, info->attrs);
1543 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1544 info->snd_seq, OVS_DP_CMD_NEW);
1546 return PTR_ERR(reply);
1548 return genlmsg_reply(reply, info);
1551 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1553 struct datapath *dp;
1554 int skip = cb->args[0];
1557 list_for_each_entry(dp, &dps, list_node) {
1560 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1561 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1562 OVS_DP_CMD_NEW) < 0)
1572 static struct genl_ops dp_datapath_genl_ops[] = {
1573 { .cmd = OVS_DP_CMD_NEW,
1574 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1575 .policy = datapath_policy,
1576 .doit = ovs_dp_cmd_new
1578 { .cmd = OVS_DP_CMD_DEL,
1579 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1580 .policy = datapath_policy,
1581 .doit = ovs_dp_cmd_del
1583 { .cmd = OVS_DP_CMD_GET,
1584 .flags = 0, /* OK for unprivileged users. */
1585 .policy = datapath_policy,
1586 .doit = ovs_dp_cmd_get,
1587 .dumpit = ovs_dp_cmd_dump
1589 { .cmd = OVS_DP_CMD_SET,
1590 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1591 .policy = datapath_policy,
1592 .doit = ovs_dp_cmd_set,
1596 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1597 #ifdef HAVE_NLA_NUL_STRING
1598 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1599 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1600 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1602 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1603 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1605 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1606 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1607 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1608 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1611 static struct genl_family dp_vport_genl_family = {
1612 .id = GENL_ID_GENERATE,
1613 .hdrsize = sizeof(struct ovs_header),
1614 .name = OVS_VPORT_FAMILY,
1615 .version = OVS_VPORT_VERSION,
1616 .maxattr = OVS_VPORT_ATTR_MAX
1619 struct genl_multicast_group dp_vport_multicast_group = {
1620 .name = OVS_VPORT_MCGROUP
1623 /* Called with RTNL lock or RCU read lock. */
1624 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1625 u32 pid, u32 seq, u32 flags, u8 cmd)
1627 struct ovs_header *ovs_header;
1631 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1636 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1638 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1639 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport_get_type(vport));
1640 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport_get_name(vport));
1641 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1643 nla = nla_reserve(skb, OVS_VPORT_ATTR_STATS,
1644 sizeof(struct ovs_vport_stats));
1646 goto nla_put_failure;
1648 vport_get_stats(vport, nla_data(nla));
1650 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN, vport_get_addr(vport));
1652 err = vport_get_options(vport, skb);
1653 if (err == -EMSGSIZE)
1656 return genlmsg_end(skb, ovs_header);
1661 genlmsg_cancel(skb, ovs_header);
1665 /* Called with RTNL lock or RCU read lock. */
1666 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1669 struct sk_buff *skb;
1672 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1674 return ERR_PTR(-ENOMEM);
1676 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1679 return ERR_PTR(retval);
1684 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1686 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1689 /* Called with RTNL lock or RCU read lock. */
1690 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1691 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1693 struct datapath *dp;
1694 struct vport *vport;
1696 if (a[OVS_VPORT_ATTR_NAME]) {
1697 vport = vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1699 return ERR_PTR(-ENODEV);
1701 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1702 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1704 if (port_no >= DP_MAX_PORTS)
1705 return ERR_PTR(-EFBIG);
1707 dp = get_dp(ovs_header->dp_ifindex);
1709 return ERR_PTR(-ENODEV);
1711 vport = get_vport_protected(dp, port_no);
1713 return ERR_PTR(-ENOENT);
1716 return ERR_PTR(-EINVAL);
1719 /* Called with RTNL lock. */
1720 static int change_vport(struct vport *vport,
1721 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1725 if (a[OVS_VPORT_ATTR_STATS])
1726 vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1728 if (a[OVS_VPORT_ATTR_ADDRESS])
1729 err = vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1734 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1736 struct nlattr **a = info->attrs;
1737 struct ovs_header *ovs_header = info->userhdr;
1738 struct vport_parms parms;
1739 struct sk_buff *reply;
1740 struct vport *vport;
1741 struct datapath *dp;
1746 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1747 !a[OVS_VPORT_ATTR_UPCALL_PID])
1750 err = ovs_vport_cmd_validate(a);
1755 dp = get_dp(ovs_header->dp_ifindex);
1760 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1761 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1764 if (port_no >= DP_MAX_PORTS)
1767 vport = get_vport_protected(dp, port_no);
1772 for (port_no = 1; ; port_no++) {
1773 if (port_no >= DP_MAX_PORTS) {
1777 vport = get_vport_protected(dp, port_no);
1783 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1784 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1785 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1787 parms.port_no = port_no;
1788 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1790 vport = new_vport(&parms);
1791 err = PTR_ERR(vport);
1795 dp_sysfs_add_if(vport);
1797 err = change_vport(vport, a);
1799 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1803 err = PTR_ERR(reply);
1806 dp_detach_port(vport);
1809 genl_notify(reply, genl_info_net(info), info->snd_pid,
1810 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1819 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1821 struct nlattr **a = info->attrs;
1822 struct sk_buff *reply;
1823 struct vport *vport;
1826 err = ovs_vport_cmd_validate(a);
1831 vport = lookup_vport(info->userhdr, a);
1832 err = PTR_ERR(vport);
1837 if (a[OVS_VPORT_ATTR_TYPE] &&
1838 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport_get_type(vport))
1841 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1842 err = vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1844 err = change_vport(vport, a);
1845 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1846 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1848 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1850 if (IS_ERR(reply)) {
1851 err = PTR_ERR(reply);
1852 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1853 dp_vport_multicast_group.id, err);
1857 genl_notify(reply, genl_info_net(info), info->snd_pid,
1858 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1866 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1868 struct nlattr **a = info->attrs;
1869 struct sk_buff *reply;
1870 struct vport *vport;
1873 err = ovs_vport_cmd_validate(a);
1878 vport = lookup_vport(info->userhdr, a);
1879 err = PTR_ERR(vport);
1883 if (vport->port_no == OVSP_LOCAL) {
1888 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1890 err = PTR_ERR(reply);
1894 dp_detach_port(vport);
1896 genl_notify(reply, genl_info_net(info), info->snd_pid,
1897 dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1905 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1907 struct nlattr **a = info->attrs;
1908 struct ovs_header *ovs_header = info->userhdr;
1909 struct sk_buff *reply;
1910 struct vport *vport;
1913 err = ovs_vport_cmd_validate(a);
1918 vport = lookup_vport(ovs_header, a);
1919 err = PTR_ERR(vport);
1923 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1925 err = PTR_ERR(reply);
1931 return genlmsg_reply(reply, info);
1939 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1941 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1942 struct datapath *dp;
1946 dp = get_dp(ovs_header->dp_ifindex);
1951 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1952 struct vport *vport;
1954 vport = get_vport_protected(dp, port_no);
1958 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1959 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1960 OVS_VPORT_CMD_NEW) < 0)
1965 cb->args[0] = port_no;
1971 static struct genl_ops dp_vport_genl_ops[] = {
1972 { .cmd = OVS_VPORT_CMD_NEW,
1973 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1974 .policy = vport_policy,
1975 .doit = ovs_vport_cmd_new
1977 { .cmd = OVS_VPORT_CMD_DEL,
1978 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1979 .policy = vport_policy,
1980 .doit = ovs_vport_cmd_del
1982 { .cmd = OVS_VPORT_CMD_GET,
1983 .flags = 0, /* OK for unprivileged users. */
1984 .policy = vport_policy,
1985 .doit = ovs_vport_cmd_get,
1986 .dumpit = ovs_vport_cmd_dump
1988 { .cmd = OVS_VPORT_CMD_SET,
1989 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1990 .policy = vport_policy,
1991 .doit = ovs_vport_cmd_set,
1995 struct genl_family_and_ops {
1996 struct genl_family *family;
1997 struct genl_ops *ops;
1999 struct genl_multicast_group *group;
2002 static const struct genl_family_and_ops dp_genl_families[] = {
2003 { &dp_datapath_genl_family,
2004 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
2005 &dp_datapath_multicast_group },
2006 { &dp_vport_genl_family,
2007 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
2008 &dp_vport_multicast_group },
2009 { &dp_flow_genl_family,
2010 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
2011 &dp_flow_multicast_group },
2012 { &dp_packet_genl_family,
2013 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2017 static void dp_unregister_genl(int n_families)
2021 for (i = 0; i < n_families; i++)
2022 genl_unregister_family(dp_genl_families[i].family);
2025 static int dp_register_genl(void)
2032 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2033 const struct genl_family_and_ops *f = &dp_genl_families[i];
2035 err = genl_register_family_with_ops(f->family, f->ops,
2042 err = genl_register_mc_group(f->family, f->group);
2051 dp_unregister_genl(n_registered);
2055 static int __init dp_init(void)
2057 struct sk_buff *dummy_skb;
2060 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2062 pr_info("Open vSwitch %s, built "__DATE__" "__TIME__"\n",
2071 goto error_tnl_exit;
2075 goto error_flow_exit;
2077 err = register_netdevice_notifier(&dp_device_notifier);
2079 goto error_vport_exit;
2081 err = dp_register_genl();
2083 goto error_unreg_notifier;
2087 error_unreg_notifier:
2088 unregister_netdevice_notifier(&dp_device_notifier);
2099 static void dp_cleanup(void)
2102 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2103 unregister_netdevice_notifier(&dp_device_notifier);
2109 module_init(dp_init);
2110 module_exit(dp_cleanup);
2112 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2113 MODULE_LICENSE("GPL");