2 * Copyright (c) 2007-2011 Nicira Networks.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/system.h>
43 #include <asm/div64.h>
44 #include <linux/highmem.h>
45 #include <linux/netfilter_bridge.h>
46 #include <linux/netfilter_ipv4.h>
47 #include <linux/inetdevice.h>
48 #include <linux/list.h>
49 #include <linux/openvswitch.h>
50 #include <linux/rculist.h>
51 #include <linux/dmi.h>
52 #include <net/genetlink.h>
57 #include "genl_exec.h"
60 #include "vport-internal_dev.h"
62 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18) || \
63 LINUX_VERSION_CODE > KERNEL_VERSION(3,2,0)
64 #error Kernels before 2.6.18 or after 3.2 are not supported by this version of Open vSwitch.
67 int (*ovs_dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
68 EXPORT_SYMBOL(ovs_dp_ioctl_hook);
73 * Writes to device state (add/remove datapath, port, set operations on vports,
74 * etc.) are protected by RTNL.
76 * Writes to other state (flow table modifications, set miscellaneous datapath
77 * parameters, etc.) are protected by genl_mutex. The RTNL lock nests inside
80 * Reads are protected by RCU.
82 * There are a few special cases (mostly stats) that have their own
83 * synchronization but they nest under all of above and don't interact with
87 /* Global list of datapaths to enable dumping them all out.
88 * Protected by genl_mutex.
90 static LIST_HEAD(dps);
92 static struct vport *new_vport(const struct vport_parms *);
93 static int queue_gso_packets(int dp_ifindex, struct sk_buff *,
94 const struct dp_upcall_info *);
95 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *,
96 const struct dp_upcall_info *);
98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
99 static struct datapath *get_dp(int dp_ifindex)
101 struct datapath *dp = NULL;
102 struct net_device *dev;
105 dev = dev_get_by_index_rcu(&init_net, dp_ifindex);
107 struct vport *vport = ovs_internal_dev_get_vport(dev);
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 const char *ovs_dp_name(const struct datapath *dp)
119 struct vport *vport = rcu_dereference_rtnl(dp->ports[OVSP_LOCAL]);
120 return vport->ops->get_name(vport);
123 static int get_dpifindex(struct datapath *dp)
130 local = rcu_dereference(dp->ports[OVSP_LOCAL]);
132 ifindex = local->ops->get_ifindex(local);
141 static size_t br_nlmsg_size(void)
143 return NLMSG_ALIGN(sizeof(struct ifinfomsg))
144 + nla_total_size(IFNAMSIZ) /* IFLA_IFNAME */
145 + nla_total_size(MAX_ADDR_LEN) /* IFLA_ADDRESS */
146 + nla_total_size(4) /* IFLA_MASTER */
147 + nla_total_size(4) /* IFLA_MTU */
148 + nla_total_size(1); /* IFLA_OPERSTATE */
151 /* Caller must hold RTNL lock. */
152 static int dp_fill_ifinfo(struct sk_buff *skb,
153 const struct vport *port,
154 int event, unsigned int flags)
156 struct datapath *dp = port->dp;
157 struct ifinfomsg *hdr;
158 struct nlmsghdr *nlh;
160 if (!port->ops->get_ifindex)
163 nlh = nlmsg_put(skb, 0, 0, event, sizeof(*hdr), flags);
167 hdr = nlmsg_data(nlh);
168 hdr->ifi_family = AF_BRIDGE;
170 hdr->ifi_type = ARPHRD_ETHER;
171 hdr->ifi_index = port->ops->get_ifindex(port);
172 hdr->ifi_flags = port->ops->get_dev_flags(port);
175 NLA_PUT_STRING(skb, IFLA_IFNAME, port->ops->get_name(port));
176 NLA_PUT_U32(skb, IFLA_MASTER, get_dpifindex(dp));
177 NLA_PUT_U32(skb, IFLA_MTU, port->ops->get_mtu(port));
178 #ifdef IFLA_OPERSTATE
179 NLA_PUT_U8(skb, IFLA_OPERSTATE,
180 port->ops->is_running(port)
181 ? port->ops->get_operstate(port)
185 NLA_PUT(skb, IFLA_ADDRESS, ETH_ALEN, port->ops->get_addr(port));
187 return nlmsg_end(skb, nlh);
190 nlmsg_cancel(skb, nlh);
194 /* Caller must hold RTNL lock. */
195 static void dp_ifinfo_notify(int event, struct vport *port)
200 skb = nlmsg_new(br_nlmsg_size(), GFP_KERNEL);
206 err = dp_fill_ifinfo(skb, port, event, 0);
208 if (err == -ENODEV) {
211 /* -EMSGSIZE implies BUG in br_nlmsg_size() */
212 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);
226 static void release_dp(struct kobject *kobj)
228 struct datapath *dp = container_of(kobj, struct datapath, ifobj);
232 static struct kobj_type dp_ktype = {
233 .release = release_dp
236 static void destroy_dp_rcu(struct rcu_head *rcu)
238 struct datapath *dp = container_of(rcu, struct datapath, rcu);
240 ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
241 free_percpu(dp->stats_percpu);
242 kobject_put(&dp->ifobj);
245 /* Called with RTNL lock and genl_lock. */
246 static struct vport *new_vport(const struct vport_parms *parms)
250 vport = ovs_vport_add(parms);
251 if (!IS_ERR(vport)) {
252 struct datapath *dp = parms->dp;
254 rcu_assign_pointer(dp->ports[parms->port_no], vport);
255 list_add(&vport->node, &dp->port_list);
257 dp_ifinfo_notify(RTM_NEWLINK, vport);
263 /* Called with RTNL lock. */
264 void ovs_dp_detach_port(struct vport *p)
268 if (p->port_no != OVSP_LOCAL)
269 ovs_dp_sysfs_del_if(p);
270 dp_ifinfo_notify(RTM_DELLINK, p);
272 /* First drop references to device. */
274 rcu_assign_pointer(p->dp->ports[p->port_no], NULL);
276 /* Then destroy it. */
280 /* Must be called with rcu_read_lock. */
281 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
283 struct datapath *dp = p->dp;
284 struct sw_flow *flow;
285 struct dp_stats_percpu *stats;
289 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
291 if (!OVS_CB(skb)->flow) {
292 struct sw_flow_key key;
295 /* Extract flow from 'skb' into 'key'. */
296 error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
297 if (unlikely(error)) {
303 flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table),
305 if (unlikely(!flow)) {
306 struct dp_upcall_info upcall;
308 upcall.cmd = OVS_PACKET_CMD_MISS;
310 upcall.userdata = NULL;
311 upcall.pid = p->upcall_pid;
312 ovs_dp_upcall(dp, skb, &upcall);
314 stats_counter = &stats->n_missed;
318 OVS_CB(skb)->flow = flow;
321 stats_counter = &stats->n_hit;
322 ovs_flow_used(OVS_CB(skb)->flow, skb);
323 ovs_execute_actions(dp, skb);
326 /* Update datapath statistics. */
327 u64_stats_update_begin(&stats->sync);
329 u64_stats_update_end(&stats->sync);
332 static struct genl_family dp_packet_genl_family = {
333 .id = GENL_ID_GENERATE,
334 .hdrsize = sizeof(struct ovs_header),
335 .name = OVS_PACKET_FAMILY,
336 .version = OVS_PACKET_VERSION,
337 .maxattr = OVS_PACKET_ATTR_MAX
340 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
341 const struct dp_upcall_info *upcall_info)
343 struct dp_stats_percpu *stats;
347 if (upcall_info->pid == 0) {
352 dp_ifindex = get_dpifindex(dp);
358 forward_ip_summed(skb, true);
360 if (!skb_is_gso(skb))
361 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
363 err = queue_gso_packets(dp_ifindex, skb, upcall_info);
370 stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id());
372 u64_stats_update_begin(&stats->sync);
374 u64_stats_update_end(&stats->sync);
379 static int queue_gso_packets(int dp_ifindex, struct sk_buff *skb,
380 const struct dp_upcall_info *upcall_info)
382 struct dp_upcall_info later_info;
383 struct sw_flow_key later_key;
384 struct sk_buff *segs, *nskb;
387 segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
391 /* Queue all of the segments. */
394 err = queue_userspace_packet(dp_ifindex, skb, upcall_info);
398 if (skb == segs && skb_shinfo(skb)->gso_type & SKB_GSO_UDP) {
399 /* The initial flow key extracted by ovs_flow_extract()
400 * in this case is for a first fragment, so we need to
401 * properly mark later fragments.
403 later_key = *upcall_info->key;
404 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
406 later_info = *upcall_info;
407 later_info.key = &later_key;
408 upcall_info = &later_info;
410 } while ((skb = skb->next));
412 /* Free all of the segments. */
420 } while ((skb = nskb));
424 static int queue_userspace_packet(int dp_ifindex, struct sk_buff *skb,
425 const struct dp_upcall_info *upcall_info)
427 struct ovs_header *upcall;
428 struct sk_buff *nskb = NULL;
429 struct sk_buff *user_skb; /* to be queued to userspace */
434 if (vlan_tx_tag_present(skb)) {
435 nskb = skb_clone(skb, GFP_ATOMIC);
439 err = vlan_deaccel_tag(nskb);
446 if (nla_attr_size(skb->len) > USHRT_MAX) {
451 len = sizeof(struct ovs_header);
452 len += nla_total_size(skb->len);
453 len += nla_total_size(FLOW_BUFSIZE);
454 if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
455 len += nla_total_size(8);
457 user_skb = genlmsg_new(len, GFP_ATOMIC);
463 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
464 0, upcall_info->cmd);
465 upcall->dp_ifindex = dp_ifindex;
467 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
468 ovs_flow_to_nlattrs(upcall_info->key, user_skb);
469 nla_nest_end(user_skb, nla);
471 if (upcall_info->userdata)
472 nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
473 nla_get_u64(upcall_info->userdata));
475 nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
477 skb_copy_and_csum_dev(skb, nla_data(nla));
479 err = genlmsg_unicast(&init_net, user_skb, upcall_info->pid);
486 /* Called with genl_mutex. */
487 static int flush_flows(int dp_ifindex)
489 struct flow_table *old_table;
490 struct flow_table *new_table;
493 dp = get_dp(dp_ifindex);
497 old_table = genl_dereference(dp->table);
498 new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
502 rcu_assign_pointer(dp->table, new_table);
504 ovs_flow_tbl_deferred_destroy(old_table);
508 static int validate_actions(const struct nlattr *attr,
509 const struct sw_flow_key *key, int depth);
511 static int validate_sample(const struct nlattr *attr,
512 const struct sw_flow_key *key, int depth)
514 const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
515 const struct nlattr *probability, *actions;
516 const struct nlattr *a;
519 memset(attrs, 0, sizeof(attrs));
520 nla_for_each_nested(a, attr, rem) {
521 int type = nla_type(a);
522 if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
529 probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
530 if (!probability || nla_len(probability) != sizeof(u32))
533 actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
534 if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
536 return validate_actions(actions, key, depth + 1);
539 static int validate_set(const struct nlattr *a,
540 const struct sw_flow_key *flow_key)
542 const struct nlattr *ovs_key = nla_data(a);
543 int key_type = nla_type(ovs_key);
545 /* There can be only one key in a action */
546 if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
549 if (key_type > OVS_KEY_ATTR_MAX ||
550 nla_len(ovs_key) != ovs_key_lens[key_type])
554 const struct ovs_key_ipv4 *ipv4_key;
556 case OVS_KEY_ATTR_PRIORITY:
557 case OVS_KEY_ATTR_TUN_ID:
558 case OVS_KEY_ATTR_ETHERNET:
561 case OVS_KEY_ATTR_IPV4:
562 if (flow_key->eth.type != htons(ETH_P_IP))
565 if (!flow_key->ipv4.addr.src || !flow_key->ipv4.addr.dst)
568 ipv4_key = nla_data(ovs_key);
569 if (ipv4_key->ipv4_proto != flow_key->ip.proto)
572 if (ipv4_key->ipv4_frag != flow_key->ip.frag)
577 case OVS_KEY_ATTR_TCP:
578 if (flow_key->ip.proto != IPPROTO_TCP)
581 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
586 case OVS_KEY_ATTR_UDP:
587 if (flow_key->ip.proto != IPPROTO_UDP)
590 if (!flow_key->ipv4.tp.src || !flow_key->ipv4.tp.dst)
601 static int validate_userspace(const struct nlattr *attr)
603 static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] = {
604 [OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
605 [OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
607 struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
610 error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
611 attr, userspace_policy);
615 if (!a[OVS_USERSPACE_ATTR_PID] ||
616 !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
622 static int validate_actions(const struct nlattr *attr,
623 const struct sw_flow_key *key, int depth)
625 const struct nlattr *a;
628 if (depth >= SAMPLE_ACTION_DEPTH)
631 nla_for_each_nested(a, attr, rem) {
632 /* Expected argument lengths, (u32)-1 for variable length. */
633 static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
634 [OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
635 [OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
636 [OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
637 [OVS_ACTION_ATTR_POP_VLAN] = 0,
638 [OVS_ACTION_ATTR_SET] = (u32)-1,
639 [OVS_ACTION_ATTR_SAMPLE] = (u32)-1
641 const struct ovs_action_push_vlan *vlan;
642 int type = nla_type(a);
644 if (type > OVS_ACTION_ATTR_MAX ||
645 (action_lens[type] != nla_len(a) &&
646 action_lens[type] != (u32)-1))
650 case OVS_ACTION_ATTR_UNSPEC:
653 case OVS_ACTION_ATTR_USERSPACE:
654 err = validate_userspace(a);
659 case OVS_ACTION_ATTR_OUTPUT:
660 if (nla_get_u32(a) >= DP_MAX_PORTS)
665 case OVS_ACTION_ATTR_POP_VLAN:
668 case OVS_ACTION_ATTR_PUSH_VLAN:
670 if (vlan->vlan_tpid != htons(ETH_P_8021Q))
672 if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
676 case OVS_ACTION_ATTR_SET:
677 err = validate_set(a, key);
682 case OVS_ACTION_ATTR_SAMPLE:
683 err = validate_sample(a, key, depth);
699 static void clear_stats(struct sw_flow *flow)
703 flow->packet_count = 0;
704 flow->byte_count = 0;
707 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
709 struct ovs_header *ovs_header = info->userhdr;
710 struct nlattr **a = info->attrs;
711 struct sw_flow_actions *acts;
712 struct sk_buff *packet;
713 struct sw_flow *flow;
721 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
722 !a[OVS_PACKET_ATTR_ACTIONS] ||
723 nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
726 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
727 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
731 skb_reserve(packet, NET_IP_ALIGN);
733 memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
735 skb_reset_mac_header(packet);
736 eth = eth_hdr(packet);
738 /* Normally, setting the skb 'protocol' field would be handled by a
739 * call to eth_type_trans(), but it assumes there's a sending
740 * device, which we may not have. */
741 if (ntohs(eth->h_proto) >= 1536)
742 packet->protocol = eth->h_proto;
744 packet->protocol = htons(ETH_P_802_2);
746 /* Build an sw_flow for sending this packet. */
747 flow = ovs_flow_alloc();
752 err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
756 err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
757 &flow->key.phy.in_port,
758 &flow->key.phy.tun_id,
759 a[OVS_PACKET_ATTR_KEY]);
763 err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
767 flow->hash = ovs_flow_hash(&flow->key, key_len);
769 acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
773 rcu_assign_pointer(flow->sf_acts, acts);
775 OVS_CB(packet)->flow = flow;
776 packet->priority = flow->key.phy.priority;
779 dp = get_dp(ovs_header->dp_ifindex);
785 err = ovs_execute_actions(dp, packet);
802 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
803 [OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
804 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
805 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
808 static struct genl_ops dp_packet_genl_ops[] = {
809 { .cmd = OVS_PACKET_CMD_EXECUTE,
810 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
811 .policy = packet_policy,
812 .doit = ovs_packet_cmd_execute
816 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
819 struct flow_table *table = genl_dereference(dp->table);
821 stats->n_flows = ovs_flow_tbl_count(table);
823 stats->n_hit = stats->n_missed = stats->n_lost = 0;
824 for_each_possible_cpu(i) {
825 const struct dp_stats_percpu *percpu_stats;
826 struct dp_stats_percpu local_stats;
829 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
832 start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
833 local_stats = *percpu_stats;
834 } while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
836 stats->n_hit += local_stats.n_hit;
837 stats->n_missed += local_stats.n_missed;
838 stats->n_lost += local_stats.n_lost;
842 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
843 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
844 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
845 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
848 static struct genl_family dp_flow_genl_family = {
849 .id = GENL_ID_GENERATE,
850 .hdrsize = sizeof(struct ovs_header),
851 .name = OVS_FLOW_FAMILY,
852 .version = OVS_FLOW_VERSION,
853 .maxattr = OVS_FLOW_ATTR_MAX
856 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
857 .name = OVS_FLOW_MCGROUP
860 /* Called with genl_lock. */
861 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
862 struct sk_buff *skb, u32 pid,
863 u32 seq, u32 flags, u8 cmd)
865 const int skb_orig_len = skb->len;
866 const struct sw_flow_actions *sf_acts;
867 struct ovs_flow_stats stats;
868 struct ovs_header *ovs_header;
874 sf_acts = rcu_dereference_protected(flow->sf_acts,
875 lockdep_genl_is_held());
877 ovs_header = genlmsg_put(skb, pid, seq, &dp_flow_genl_family, flags, cmd);
881 ovs_header->dp_ifindex = get_dpifindex(dp);
883 nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
885 goto nla_put_failure;
886 err = ovs_flow_to_nlattrs(&flow->key, skb);
889 nla_nest_end(skb, nla);
891 spin_lock_bh(&flow->lock);
893 stats.n_packets = flow->packet_count;
894 stats.n_bytes = flow->byte_count;
895 tcp_flags = flow->tcp_flags;
896 spin_unlock_bh(&flow->lock);
899 NLA_PUT_U64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used));
902 NLA_PUT(skb, OVS_FLOW_ATTR_STATS,
903 sizeof(struct ovs_flow_stats), &stats);
906 NLA_PUT_U8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags);
908 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
909 * this is the first flow to be dumped into 'skb'. This is unusual for
910 * Netlink but individual action lists can be longer than
911 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
912 * The userspace caller can always fetch the actions separately if it
913 * really wants them. (Most userspace callers in fact don't care.)
915 * This can only fail for dump operations because the skb is always
916 * properly sized for single flows.
918 err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
920 if (err < 0 && skb_orig_len)
923 return genlmsg_end(skb, ovs_header);
928 genlmsg_cancel(skb, ovs_header);
932 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
934 const struct sw_flow_actions *sf_acts;
937 sf_acts = rcu_dereference_protected(flow->sf_acts,
938 lockdep_genl_is_held());
940 /* OVS_FLOW_ATTR_KEY */
941 len = nla_total_size(FLOW_BUFSIZE);
942 /* OVS_FLOW_ATTR_ACTIONS */
943 len += nla_total_size(sf_acts->actions_len);
944 /* OVS_FLOW_ATTR_STATS */
945 len += nla_total_size(sizeof(struct ovs_flow_stats));
946 /* OVS_FLOW_ATTR_TCP_FLAGS */
947 len += nla_total_size(1);
948 /* OVS_FLOW_ATTR_USED */
949 len += nla_total_size(8);
951 len += NLMSG_ALIGN(sizeof(struct ovs_header));
953 return genlmsg_new(len, GFP_KERNEL);
956 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
958 u32 pid, u32 seq, u8 cmd)
963 skb = ovs_flow_cmd_alloc_info(flow);
965 return ERR_PTR(-ENOMEM);
967 retval = ovs_flow_cmd_fill_info(flow, dp, skb, pid, seq, 0, cmd);
972 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
974 struct nlattr **a = info->attrs;
975 struct ovs_header *ovs_header = info->userhdr;
976 struct sw_flow_key key;
977 struct sw_flow *flow;
978 struct sk_buff *reply;
980 struct flow_table *table;
986 if (!a[OVS_FLOW_ATTR_KEY])
988 error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
992 /* Validate actions. */
993 if (a[OVS_FLOW_ATTR_ACTIONS]) {
994 error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, 0);
997 } else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
1002 dp = get_dp(ovs_header->dp_ifindex);
1007 table = genl_dereference(dp->table);
1008 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1010 struct sw_flow_actions *acts;
1012 /* Bail out if we're not allowed to create a new flow. */
1014 if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
1017 /* Expand table, if necessary, to make room. */
1018 if (ovs_flow_tbl_need_to_expand(table)) {
1019 struct flow_table *new_table;
1021 new_table = ovs_flow_tbl_expand(table);
1022 if (!IS_ERR(new_table)) {
1023 rcu_assign_pointer(dp->table, new_table);
1024 ovs_flow_tbl_deferred_destroy(table);
1025 table = genl_dereference(dp->table);
1029 /* Allocate flow. */
1030 flow = ovs_flow_alloc();
1032 error = PTR_ERR(flow);
1038 /* Obtain actions. */
1039 acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
1040 error = PTR_ERR(acts);
1042 goto error_free_flow;
1043 rcu_assign_pointer(flow->sf_acts, acts);
1045 /* Put flow in bucket. */
1046 flow->hash = ovs_flow_hash(&key, key_len);
1047 ovs_flow_tbl_insert(table, flow);
1049 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1053 /* We found a matching flow. */
1054 struct sw_flow_actions *old_acts;
1055 struct nlattr *acts_attrs;
1057 /* Bail out if we're not allowed to modify an existing flow.
1058 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1059 * because Generic Netlink treats the latter as a dump
1060 * request. We also accept NLM_F_EXCL in case that bug ever
1064 if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1065 info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1068 /* Update actions. */
1069 old_acts = rcu_dereference_protected(flow->sf_acts,
1070 lockdep_genl_is_held());
1071 acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1073 (old_acts->actions_len != nla_len(acts_attrs) ||
1074 memcmp(old_acts->actions, nla_data(acts_attrs),
1075 old_acts->actions_len))) {
1076 struct sw_flow_actions *new_acts;
1078 new_acts = ovs_flow_actions_alloc(acts_attrs);
1079 error = PTR_ERR(new_acts);
1080 if (IS_ERR(new_acts))
1083 rcu_assign_pointer(flow->sf_acts, new_acts);
1084 ovs_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 ovs_dp_flow_multicast_group.id, info->nlhdr,
1103 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1104 ovs_dp_flow_multicast_group.id,
1114 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1116 struct nlattr **a = info->attrs;
1117 struct ovs_header *ovs_header = info->userhdr;
1118 struct sw_flow_key key;
1119 struct sk_buff *reply;
1120 struct sw_flow *flow;
1121 struct datapath *dp;
1122 struct flow_table *table;
1126 if (!a[OVS_FLOW_ATTR_KEY])
1128 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1132 dp = get_dp(ovs_header->dp_ifindex);
1136 table = genl_dereference(dp->table);
1137 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1141 reply = ovs_flow_cmd_build_info(flow, dp, info->snd_pid,
1142 info->snd_seq, OVS_FLOW_CMD_NEW);
1144 return PTR_ERR(reply);
1146 return genlmsg_reply(reply, info);
1149 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1151 struct nlattr **a = info->attrs;
1152 struct ovs_header *ovs_header = info->userhdr;
1153 struct sw_flow_key key;
1154 struct sk_buff *reply;
1155 struct sw_flow *flow;
1156 struct datapath *dp;
1157 struct flow_table *table;
1161 if (!a[OVS_FLOW_ATTR_KEY])
1162 return flush_flows(ovs_header->dp_ifindex);
1163 err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1167 dp = get_dp(ovs_header->dp_ifindex);
1171 table = genl_dereference(dp->table);
1172 flow = ovs_flow_tbl_lookup(table, &key, key_len);
1176 reply = ovs_flow_cmd_alloc_info(flow);
1180 ovs_flow_tbl_remove(table, flow);
1182 err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_pid,
1183 info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1186 ovs_flow_deferred_free(flow);
1188 genl_notify(reply, genl_info_net(info), info->snd_pid,
1189 ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1193 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1195 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1196 struct datapath *dp;
1197 struct flow_table *table;
1199 dp = get_dp(ovs_header->dp_ifindex);
1203 table = genl_dereference(dp->table);
1206 struct sw_flow *flow;
1209 bucket = cb->args[0];
1211 flow = ovs_flow_tbl_next(table, &bucket, &obj);
1215 if (ovs_flow_cmd_fill_info(flow, dp, skb,
1216 NETLINK_CB(cb->skb).pid,
1217 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1218 OVS_FLOW_CMD_NEW) < 0)
1221 cb->args[0] = bucket;
1227 static struct genl_ops dp_flow_genl_ops[] = {
1228 { .cmd = OVS_FLOW_CMD_NEW,
1229 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1230 .policy = flow_policy,
1231 .doit = ovs_flow_cmd_new_or_set
1233 { .cmd = OVS_FLOW_CMD_DEL,
1234 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1235 .policy = flow_policy,
1236 .doit = ovs_flow_cmd_del
1238 { .cmd = OVS_FLOW_CMD_GET,
1239 .flags = 0, /* OK for unprivileged users. */
1240 .policy = flow_policy,
1241 .doit = ovs_flow_cmd_get,
1242 .dumpit = ovs_flow_cmd_dump
1244 { .cmd = OVS_FLOW_CMD_SET,
1245 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1246 .policy = flow_policy,
1247 .doit = ovs_flow_cmd_new_or_set,
1251 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1252 #ifdef HAVE_NLA_NUL_STRING
1253 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1255 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1258 static struct genl_family dp_datapath_genl_family = {
1259 .id = GENL_ID_GENERATE,
1260 .hdrsize = sizeof(struct ovs_header),
1261 .name = OVS_DATAPATH_FAMILY,
1262 .version = OVS_DATAPATH_VERSION,
1263 .maxattr = OVS_DP_ATTR_MAX
1266 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1267 .name = OVS_DATAPATH_MCGROUP
1270 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1271 u32 pid, u32 seq, u32 flags, u8 cmd)
1273 struct ovs_header *ovs_header;
1274 struct ovs_dp_stats dp_stats;
1277 ovs_header = genlmsg_put(skb, pid, seq, &dp_datapath_genl_family,
1282 ovs_header->dp_ifindex = get_dpifindex(dp);
1285 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1288 goto nla_put_failure;
1290 get_dp_stats(dp, &dp_stats);
1291 NLA_PUT(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats);
1293 return genlmsg_end(skb, ovs_header);
1296 genlmsg_cancel(skb, ovs_header);
1301 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 pid,
1304 struct sk_buff *skb;
1307 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1309 return ERR_PTR(-ENOMEM);
1311 retval = ovs_dp_cmd_fill_info(dp, skb, pid, seq, 0, cmd);
1314 return ERR_PTR(retval);
1319 static int ovs_dp_cmd_validate(struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1321 return CHECK_NUL_STRING(a[OVS_DP_ATTR_NAME], IFNAMSIZ - 1);
1324 /* Called with genl_mutex and optionally with RTNL lock also. */
1325 static struct datapath *lookup_datapath(struct ovs_header *ovs_header,
1326 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1328 struct datapath *dp;
1330 if (!a[OVS_DP_ATTR_NAME])
1331 dp = get_dp(ovs_header->dp_ifindex);
1333 struct vport *vport;
1336 vport = ovs_vport_locate(nla_data(a[OVS_DP_ATTR_NAME]));
1337 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1340 return dp ? dp : ERR_PTR(-ENODEV);
1343 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1345 struct nlattr **a = info->attrs;
1346 struct vport_parms parms;
1347 struct sk_buff *reply;
1348 struct datapath *dp;
1349 struct vport *vport;
1353 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1356 err = ovs_dp_cmd_validate(a);
1362 if (!try_module_get(THIS_MODULE))
1363 goto err_unlock_rtnl;
1366 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1368 goto err_put_module;
1369 INIT_LIST_HEAD(&dp->port_list);
1371 /* Initialize kobject for bridge. This will be added as
1372 * /sys/class/net/<devname>/brif later, if sysfs is enabled. */
1373 dp->ifobj.kset = NULL;
1374 kobject_init(&dp->ifobj, &dp_ktype);
1376 /* Allocate table. */
1378 rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1382 dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1383 if (!dp->stats_percpu) {
1385 goto err_destroy_table;
1388 /* Set up our datapath device. */
1389 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1390 parms.type = OVS_VPORT_TYPE_INTERNAL;
1391 parms.options = NULL;
1393 parms.port_no = OVSP_LOCAL;
1394 parms.upcall_pid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1396 vport = new_vport(&parms);
1397 if (IS_ERR(vport)) {
1398 err = PTR_ERR(vport);
1402 goto err_destroy_percpu;
1405 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1406 info->snd_seq, OVS_DP_CMD_NEW);
1407 err = PTR_ERR(reply);
1409 goto err_destroy_local_port;
1411 list_add_tail(&dp->list_node, &dps);
1412 ovs_dp_sysfs_add_dp(dp);
1416 genl_notify(reply, genl_info_net(info), info->snd_pid,
1417 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1421 err_destroy_local_port:
1422 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1424 free_percpu(dp->stats_percpu);
1426 ovs_flow_tbl_destroy(genl_dereference(dp->table));
1430 module_put(THIS_MODULE);
1437 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1439 struct vport *vport, *next_vport;
1440 struct sk_buff *reply;
1441 struct datapath *dp;
1444 err = ovs_dp_cmd_validate(info->attrs);
1449 dp = lookup_datapath(info->userhdr, info->attrs);
1454 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1455 info->snd_seq, OVS_DP_CMD_DEL);
1456 err = PTR_ERR(reply);
1460 list_for_each_entry_safe(vport, next_vport, &dp->port_list, node)
1461 if (vport->port_no != OVSP_LOCAL)
1462 ovs_dp_detach_port(vport);
1464 ovs_dp_sysfs_del_dp(dp);
1465 list_del(&dp->list_node);
1466 ovs_dp_detach_port(rtnl_dereference(dp->ports[OVSP_LOCAL]));
1468 /* rtnl_unlock() will wait until all the references to devices that
1469 * are pending unregistration have been dropped. We do it here to
1470 * ensure that any internal devices (which contain DP pointers) are
1471 * fully destroyed before freeing the datapath.
1475 call_rcu(&dp->rcu, destroy_dp_rcu);
1476 module_put(THIS_MODULE);
1478 genl_notify(reply, genl_info_net(info), info->snd_pid,
1479 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1490 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1492 struct sk_buff *reply;
1493 struct datapath *dp;
1496 err = ovs_dp_cmd_validate(info->attrs);
1500 dp = lookup_datapath(info->userhdr, info->attrs);
1504 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1505 info->snd_seq, OVS_DP_CMD_NEW);
1506 if (IS_ERR(reply)) {
1507 err = PTR_ERR(reply);
1508 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1509 ovs_dp_datapath_multicast_group.id, err);
1513 genl_notify(reply, genl_info_net(info), info->snd_pid,
1514 ovs_dp_datapath_multicast_group.id, info->nlhdr,
1520 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1522 struct sk_buff *reply;
1523 struct datapath *dp;
1526 err = ovs_dp_cmd_validate(info->attrs);
1530 dp = lookup_datapath(info->userhdr, info->attrs);
1534 reply = ovs_dp_cmd_build_info(dp, info->snd_pid,
1535 info->snd_seq, OVS_DP_CMD_NEW);
1537 return PTR_ERR(reply);
1539 return genlmsg_reply(reply, info);
1542 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1544 struct datapath *dp;
1545 int skip = cb->args[0];
1548 list_for_each_entry(dp, &dps, list_node) {
1551 if (ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).pid,
1552 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1553 OVS_DP_CMD_NEW) < 0)
1563 static struct genl_ops dp_datapath_genl_ops[] = {
1564 { .cmd = OVS_DP_CMD_NEW,
1565 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1566 .policy = datapath_policy,
1567 .doit = ovs_dp_cmd_new
1569 { .cmd = OVS_DP_CMD_DEL,
1570 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1571 .policy = datapath_policy,
1572 .doit = ovs_dp_cmd_del
1574 { .cmd = OVS_DP_CMD_GET,
1575 .flags = 0, /* OK for unprivileged users. */
1576 .policy = datapath_policy,
1577 .doit = ovs_dp_cmd_get,
1578 .dumpit = ovs_dp_cmd_dump
1580 { .cmd = OVS_DP_CMD_SET,
1581 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1582 .policy = datapath_policy,
1583 .doit = ovs_dp_cmd_set,
1587 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1588 #ifdef HAVE_NLA_NUL_STRING
1589 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1590 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1591 [OVS_VPORT_ATTR_ADDRESS] = { .len = ETH_ALEN },
1593 [OVS_VPORT_ATTR_STATS] = { .minlen = sizeof(struct ovs_vport_stats) },
1594 [OVS_VPORT_ATTR_ADDRESS] = { .minlen = ETH_ALEN },
1596 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1597 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1598 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1599 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1602 static struct genl_family dp_vport_genl_family = {
1603 .id = GENL_ID_GENERATE,
1604 .hdrsize = sizeof(struct ovs_header),
1605 .name = OVS_VPORT_FAMILY,
1606 .version = OVS_VPORT_VERSION,
1607 .maxattr = OVS_VPORT_ATTR_MAX
1610 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1611 .name = OVS_VPORT_MCGROUP
1614 /* Called with RTNL lock or RCU read lock. */
1615 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1616 u32 pid, u32 seq, u32 flags, u8 cmd)
1618 struct ovs_header *ovs_header;
1619 struct ovs_vport_stats vport_stats;
1622 ovs_header = genlmsg_put(skb, pid, seq, &dp_vport_genl_family,
1627 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1629 NLA_PUT_U32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no);
1630 NLA_PUT_U32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type);
1631 NLA_PUT_STRING(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport));
1632 NLA_PUT_U32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_pid);
1634 ovs_vport_get_stats(vport, &vport_stats);
1635 NLA_PUT(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1638 NLA_PUT(skb, OVS_VPORT_ATTR_ADDRESS, ETH_ALEN,
1639 vport->ops->get_addr(vport));
1641 err = ovs_vport_get_options(vport, skb);
1642 if (err == -EMSGSIZE)
1645 return genlmsg_end(skb, ovs_header);
1650 genlmsg_cancel(skb, ovs_header);
1654 /* Called with RTNL lock or RCU read lock. */
1655 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 pid,
1658 struct sk_buff *skb;
1661 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1663 return ERR_PTR(-ENOMEM);
1665 retval = ovs_vport_cmd_fill_info(vport, skb, pid, seq, 0, cmd);
1668 return ERR_PTR(retval);
1673 static int ovs_vport_cmd_validate(struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1675 return CHECK_NUL_STRING(a[OVS_VPORT_ATTR_NAME], IFNAMSIZ - 1);
1678 /* Called with RTNL lock or RCU read lock. */
1679 static struct vport *lookup_vport(struct ovs_header *ovs_header,
1680 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1682 struct datapath *dp;
1683 struct vport *vport;
1685 if (a[OVS_VPORT_ATTR_NAME]) {
1686 vport = ovs_vport_locate(nla_data(a[OVS_VPORT_ATTR_NAME]));
1688 return ERR_PTR(-ENODEV);
1690 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1691 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1693 if (port_no >= DP_MAX_PORTS)
1694 return ERR_PTR(-EFBIG);
1696 dp = get_dp(ovs_header->dp_ifindex);
1698 return ERR_PTR(-ENODEV);
1700 vport = rcu_dereference_rtnl(dp->ports[port_no]);
1702 return ERR_PTR(-ENOENT);
1705 return ERR_PTR(-EINVAL);
1708 /* Called with RTNL lock. */
1709 static int change_vport(struct vport *vport,
1710 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1714 if (a[OVS_VPORT_ATTR_STATS])
1715 ovs_vport_set_stats(vport, nla_data(a[OVS_VPORT_ATTR_STATS]));
1717 if (a[OVS_VPORT_ATTR_ADDRESS])
1718 err = ovs_vport_set_addr(vport, nla_data(a[OVS_VPORT_ATTR_ADDRESS]));
1723 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1725 struct nlattr **a = info->attrs;
1726 struct ovs_header *ovs_header = info->userhdr;
1727 struct vport_parms parms;
1728 struct sk_buff *reply;
1729 struct vport *vport;
1730 struct datapath *dp;
1735 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1736 !a[OVS_VPORT_ATTR_UPCALL_PID])
1739 err = ovs_vport_cmd_validate(a);
1744 dp = get_dp(ovs_header->dp_ifindex);
1749 if (a[OVS_VPORT_ATTR_PORT_NO]) {
1750 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1753 if (port_no >= DP_MAX_PORTS)
1756 vport = rtnl_dereference(dp->ports[port_no]);
1761 for (port_no = 1; ; port_no++) {
1762 if (port_no >= DP_MAX_PORTS) {
1766 vport = rtnl_dereference(dp->ports[port_no]);
1772 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1773 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1774 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1776 parms.port_no = port_no;
1777 parms.upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1779 vport = new_vport(&parms);
1780 err = PTR_ERR(vport);
1784 ovs_dp_sysfs_add_if(vport);
1786 err = change_vport(vport, a);
1788 reply = ovs_vport_cmd_build_info(vport, info->snd_pid,
1792 err = PTR_ERR(reply);
1795 ovs_dp_detach_port(vport);
1798 genl_notify(reply, genl_info_net(info), info->snd_pid,
1799 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1807 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1809 struct nlattr **a = info->attrs;
1810 struct sk_buff *reply;
1811 struct vport *vport;
1814 err = ovs_vport_cmd_validate(a);
1819 vport = lookup_vport(info->userhdr, a);
1820 err = PTR_ERR(vport);
1825 if (a[OVS_VPORT_ATTR_TYPE] &&
1826 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1829 if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1830 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1832 err = change_vport(vport, a);
1833 if (!err && a[OVS_VPORT_ATTR_UPCALL_PID])
1834 vport->upcall_pid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1836 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1838 if (IS_ERR(reply)) {
1839 err = PTR_ERR(reply);
1840 netlink_set_err(INIT_NET_GENL_SOCK, 0,
1841 ovs_dp_vport_multicast_group.id, err);
1845 genl_notify(reply, genl_info_net(info), info->snd_pid,
1846 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1854 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1856 struct nlattr **a = info->attrs;
1857 struct sk_buff *reply;
1858 struct vport *vport;
1861 err = ovs_vport_cmd_validate(a);
1866 vport = lookup_vport(info->userhdr, a);
1867 err = PTR_ERR(vport);
1871 if (vport->port_no == OVSP_LOCAL) {
1876 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1878 err = PTR_ERR(reply);
1882 ovs_dp_detach_port(vport);
1884 genl_notify(reply, genl_info_net(info), info->snd_pid,
1885 ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1893 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1895 struct nlattr **a = info->attrs;
1896 struct ovs_header *ovs_header = info->userhdr;
1897 struct sk_buff *reply;
1898 struct vport *vport;
1901 err = ovs_vport_cmd_validate(a);
1906 vport = lookup_vport(ovs_header, a);
1907 err = PTR_ERR(vport);
1911 reply = ovs_vport_cmd_build_info(vport, info->snd_pid, info->snd_seq,
1913 err = PTR_ERR(reply);
1919 return genlmsg_reply(reply, info);
1927 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1929 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1930 struct datapath *dp;
1934 dp = get_dp(ovs_header->dp_ifindex);
1939 for (port_no = cb->args[0]; port_no < DP_MAX_PORTS; port_no++) {
1940 struct vport *vport;
1942 vport = rcu_dereference(dp->ports[port_no]);
1946 if (ovs_vport_cmd_fill_info(vport, skb, NETLINK_CB(cb->skb).pid,
1947 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1948 OVS_VPORT_CMD_NEW) < 0)
1953 cb->args[0] = port_no;
1959 static struct genl_ops dp_vport_genl_ops[] = {
1960 { .cmd = OVS_VPORT_CMD_NEW,
1961 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1962 .policy = vport_policy,
1963 .doit = ovs_vport_cmd_new
1965 { .cmd = OVS_VPORT_CMD_DEL,
1966 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967 .policy = vport_policy,
1968 .doit = ovs_vport_cmd_del
1970 { .cmd = OVS_VPORT_CMD_GET,
1971 .flags = 0, /* OK for unprivileged users. */
1972 .policy = vport_policy,
1973 .doit = ovs_vport_cmd_get,
1974 .dumpit = ovs_vport_cmd_dump
1976 { .cmd = OVS_VPORT_CMD_SET,
1977 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1978 .policy = vport_policy,
1979 .doit = ovs_vport_cmd_set,
1983 struct genl_family_and_ops {
1984 struct genl_family *family;
1985 struct genl_ops *ops;
1987 struct genl_multicast_group *group;
1990 static const struct genl_family_and_ops dp_genl_families[] = {
1991 { &dp_datapath_genl_family,
1992 dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1993 &ovs_dp_datapath_multicast_group },
1994 { &dp_vport_genl_family,
1995 dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1996 &ovs_dp_vport_multicast_group },
1997 { &dp_flow_genl_family,
1998 dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1999 &ovs_dp_flow_multicast_group },
2000 { &dp_packet_genl_family,
2001 dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
2005 static void dp_unregister_genl(int n_families)
2009 for (i = 0; i < n_families; i++)
2010 genl_unregister_family(dp_genl_families[i].family);
2013 static int dp_register_genl(void)
2020 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2021 const struct genl_family_and_ops *f = &dp_genl_families[i];
2023 err = genl_register_family_with_ops(f->family, f->ops,
2030 err = genl_register_mc_group(f->family, f->group);
2039 dp_unregister_genl(n_registered);
2043 static int __init dp_init(void)
2045 struct sk_buff *dummy_skb;
2048 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
2050 pr_info("Open vSwitch switching datapath %s, built "__DATE__" "__TIME__"\n",
2053 err = genl_exec_init();
2057 err = ovs_workqueues_init();
2059 goto error_genl_exec;
2061 err = ovs_tnl_init();
2065 err = ovs_flow_init();
2067 goto error_tnl_exit;
2069 err = ovs_vport_init();
2071 goto error_flow_exit;
2073 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2075 goto error_vport_exit;
2077 err = dp_register_genl();
2079 goto error_unreg_notifier;
2083 error_unreg_notifier:
2084 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2092 ovs_workqueues_exit();
2099 static void dp_cleanup(void)
2102 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2103 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2107 ovs_workqueues_exit();
2111 module_init(dp_init);
2112 module_exit(dp_cleanup);
2114 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2115 MODULE_LICENSE("GPL");