2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
61 int ovs_net_id __read_mostly;
62 EXPORT_SYMBOL_GPL(ovs_net_id);
64 static struct genl_family dp_packet_genl_family;
65 static struct genl_family dp_flow_genl_family;
66 static struct genl_family dp_datapath_genl_family;
68 static const struct nla_policy flow_policy[];
70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
71 .name = OVS_FLOW_MCGROUP,
74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
75 .name = OVS_DATAPATH_MCGROUP,
78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
79 .name = OVS_VPORT_MCGROUP,
82 /* Check if need to build a reply message.
83 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
87 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
88 genl_has_listeners(family, genl_info_net(info), group);
91 static void ovs_notify(struct genl_family *family,
92 struct sk_buff *skb, struct genl_info *info)
94 genl_notify(family, skb, info, 0, GFP_KERNEL);
100 * All writes e.g. Writes to device state (add/remove datapath, port, set
101 * operations on vports, etc.), Writes to other state (flow table
102 * modifications, set miscellaneous datapath parameters, etc.) are protected
105 * Reads are protected by RCU.
107 * There are a few special cases (mostly stats) that have their own
108 * synchronization but they nest under all of above and don't interact with
111 * The RTNL lock nests inside ovs_mutex.
114 static DEFINE_MUTEX(ovs_mutex);
118 mutex_lock(&ovs_mutex);
121 void ovs_unlock(void)
123 mutex_unlock(&ovs_mutex);
126 #ifdef CONFIG_LOCKDEP
127 int lockdep_ovsl_is_held(void)
130 return lockdep_is_held(&ovs_mutex);
134 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
137 static struct vport *new_vport(const struct vport_parms *);
138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 const struct sw_flow_key *,
140 const struct dp_upcall_info *,
142 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
143 const struct sw_flow_key *,
144 const struct dp_upcall_info *,
147 /* Must be called with rcu_read_lock. */
148 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
150 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
153 struct vport *vport = ovs_internal_dev_get_vport(dev);
161 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
162 * returned dp pointer valid.
164 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
168 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
170 dp = get_dp_rcu(net, dp_ifindex);
176 /* Must be called with rcu_read_lock or ovs_mutex. */
177 const char *ovs_dp_name(const struct datapath *dp)
179 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
180 return ovs_vport_name(vport);
183 static int get_dpifindex(const struct datapath *dp)
190 local = ovs_vport_rcu(dp, OVSP_LOCAL);
192 ifindex = local->dev->ifindex;
201 static void destroy_dp_rcu(struct rcu_head *rcu)
203 struct datapath *dp = container_of(rcu, struct datapath, rcu);
205 ovs_flow_tbl_destroy(&dp->table);
206 free_percpu(dp->stats_percpu);
211 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
214 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
217 /* Called with ovs_mutex or RCU read lock. */
218 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
221 struct hlist_head *head;
223 head = vport_hash_bucket(dp, port_no);
224 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
225 if (vport->port_no == port_no)
231 /* Called with ovs_mutex. */
232 static struct vport *new_vport(const struct vport_parms *parms)
236 vport = ovs_vport_add(parms);
237 if (!IS_ERR(vport)) {
238 struct datapath *dp = parms->dp;
239 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
241 hlist_add_head_rcu(&vport->dp_hash_node, head);
246 void ovs_dp_detach_port(struct vport *p)
250 /* First drop references to device. */
251 hlist_del_rcu(&p->dp_hash_node);
253 /* Then destroy it. */
257 /* Must be called with rcu_read_lock. */
258 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
260 const struct vport *p = OVS_CB(skb)->input_vport;
261 struct datapath *dp = p->dp;
262 struct sw_flow *flow;
263 struct sw_flow_actions *sf_acts;
264 struct dp_stats_percpu *stats;
268 stats = this_cpu_ptr(dp->stats_percpu);
271 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
272 if (unlikely(!flow)) {
273 struct dp_upcall_info upcall;
276 memset(&upcall, 0, sizeof(upcall));
277 upcall.cmd = OVS_PACKET_CMD_MISS;
278 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
279 upcall.mru = OVS_CB(skb)->mru;
280 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
285 stats_counter = &stats->n_missed;
289 ovs_flow_stats_update(flow, key->tp.flags, skb);
290 sf_acts = rcu_dereference(flow->sf_acts);
291 ovs_execute_actions(dp, skb, sf_acts, key);
293 stats_counter = &stats->n_hit;
296 /* Update datapath statistics. */
297 u64_stats_update_begin(&stats->syncp);
299 stats->n_mask_hit += n_mask_hit;
300 u64_stats_update_end(&stats->syncp);
303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
304 const struct sw_flow_key *key,
305 const struct dp_upcall_info *upcall_info,
308 struct dp_stats_percpu *stats;
311 if (upcall_info->portid == 0) {
316 if (!skb_is_gso(skb))
317 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
319 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
326 stats = this_cpu_ptr(dp->stats_percpu);
328 u64_stats_update_begin(&stats->syncp);
330 u64_stats_update_end(&stats->syncp);
335 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
336 const struct sw_flow_key *key,
337 const struct dp_upcall_info *upcall_info,
340 unsigned short gso_type = skb_shinfo(skb)->gso_type;
341 struct sw_flow_key later_key;
342 struct sk_buff *segs, *nskb;
345 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
346 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
348 return PTR_ERR(segs);
352 if (gso_type & SKB_GSO_UDP) {
353 /* The initial flow key extracted by ovs_flow_key_extract()
354 * in this case is for a first fragment, so we need to
355 * properly mark later fragments.
358 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
361 /* Queue all of the segments. */
364 if (gso_type & SKB_GSO_UDP && skb != segs)
367 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
371 } while ((skb = skb->next));
373 /* Free all of the segments. */
381 } while ((skb = nskb));
385 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
388 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
389 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
390 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
391 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
393 /* OVS_PACKET_ATTR_USERDATA */
394 if (upcall_info->userdata)
395 size += NLA_ALIGN(upcall_info->userdata->nla_len);
397 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
398 if (upcall_info->egress_tun_info)
399 size += nla_total_size(ovs_tun_key_attr_size());
401 /* OVS_PACKET_ATTR_ACTIONS */
402 if (upcall_info->actions_len)
403 size += nla_total_size(upcall_info->actions_len);
405 /* OVS_PACKET_ATTR_MRU */
406 if (upcall_info->mru)
407 size += nla_total_size(sizeof(upcall_info->mru));
412 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
414 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
415 size_t plen = NLA_ALIGN(skb->len) - skb->len;
418 memset(skb_put(skb, plen), 0, plen);
422 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
423 const struct sw_flow_key *key,
424 const struct dp_upcall_info *upcall_info,
427 struct ovs_header *upcall;
428 struct sk_buff *nskb = NULL;
429 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
435 dp_ifindex = get_dpifindex(dp);
439 if (skb_vlan_tag_present(skb)) {
440 nskb = skb_clone(skb, GFP_ATOMIC);
444 nskb = __vlan_hwaccel_push_inside(nskb);
451 if (nla_attr_size(skb->len) > USHRT_MAX) {
456 /* Complete checksum if needed */
457 if (skb->ip_summed == CHECKSUM_PARTIAL &&
458 (err = skb_checksum_help(skb)))
461 /* Older versions of OVS user space enforce alignment of the last
462 * Netlink attribute to NLA_ALIGNTO which would require extensive
463 * padding logic. Only perform zerocopy if padding is not required.
465 if (dp->user_features & OVS_DP_F_UNALIGNED)
466 hlen = skb_zerocopy_headlen(skb);
470 len = upcall_msg_size(upcall_info, hlen - cutlen);
471 user_skb = genlmsg_new(len, GFP_ATOMIC);
477 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
478 0, upcall_info->cmd);
479 upcall->dp_ifindex = dp_ifindex;
481 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
484 if (upcall_info->userdata)
485 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
486 nla_len(upcall_info->userdata),
487 nla_data(upcall_info->userdata));
489 if (upcall_info->egress_tun_info) {
490 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
491 err = ovs_nla_put_tunnel_info(user_skb,
492 upcall_info->egress_tun_info);
494 nla_nest_end(user_skb, nla);
497 if (upcall_info->actions_len) {
498 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
499 err = ovs_nla_put_actions(upcall_info->actions,
500 upcall_info->actions_len,
503 nla_nest_end(user_skb, nla);
505 nla_nest_cancel(user_skb, nla);
508 /* Add OVS_PACKET_ATTR_MRU */
509 if (upcall_info->mru) {
510 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
515 pad_packet(dp, user_skb);
518 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
520 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
525 pad_packet(dp, user_skb);
528 /* Only reserve room for attribute header, packet data is added
529 * in skb_zerocopy() */
530 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
534 nla->nla_len = nla_attr_size(skb->len - cutlen);
536 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
540 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
541 pad_packet(dp, user_skb);
543 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
545 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
555 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
557 struct ovs_header *ovs_header = info->userhdr;
558 struct net *net = sock_net(skb->sk);
559 struct nlattr **a = info->attrs;
560 struct sw_flow_actions *acts;
561 struct sk_buff *packet;
562 struct sw_flow *flow;
563 struct sw_flow_actions *sf_acts;
566 struct vport *input_vport;
570 bool log = !a[OVS_PACKET_ATTR_PROBE];
573 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
574 !a[OVS_PACKET_ATTR_ACTIONS])
577 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
578 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
582 skb_reserve(packet, NET_IP_ALIGN);
584 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
586 skb_reset_mac_header(packet);
587 eth = eth_hdr(packet);
589 /* Normally, setting the skb 'protocol' field would be handled by a
590 * call to eth_type_trans(), but it assumes there's a sending
591 * device, which we may not have. */
592 if (eth_proto_is_802_3(eth->h_proto))
593 packet->protocol = eth->h_proto;
595 packet->protocol = htons(ETH_P_802_2);
597 /* Set packet's mru */
598 if (a[OVS_PACKET_ATTR_MRU]) {
599 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
600 packet->ignore_df = 1;
602 OVS_CB(packet)->mru = mru;
604 /* Build an sw_flow for sending this packet. */
605 flow = ovs_flow_alloc();
610 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
611 packet, &flow->key, log);
615 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
616 &flow->key, &acts, log);
620 rcu_assign_pointer(flow->sf_acts, acts);
621 packet->priority = flow->key.phy.priority;
622 packet->mark = flow->key.phy.skb_mark;
625 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
630 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
632 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
637 packet->dev = input_vport->dev;
638 OVS_CB(packet)->input_vport = input_vport;
639 sf_acts = rcu_dereference(flow->sf_acts);
642 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
646 ovs_flow_free(flow, false);
652 ovs_flow_free(flow, false);
659 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
660 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
661 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
662 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
663 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
664 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
667 static const struct genl_ops dp_packet_genl_ops[] = {
668 { .cmd = OVS_PACKET_CMD_EXECUTE,
669 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
670 .policy = packet_policy,
671 .doit = ovs_packet_cmd_execute
675 static struct genl_family dp_packet_genl_family = {
676 .id = GENL_ID_GENERATE,
677 .hdrsize = sizeof(struct ovs_header),
678 .name = OVS_PACKET_FAMILY,
679 .version = OVS_PACKET_VERSION,
680 .maxattr = OVS_PACKET_ATTR_MAX,
682 .parallel_ops = true,
683 .ops = dp_packet_genl_ops,
684 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
687 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
688 struct ovs_dp_megaflow_stats *mega_stats)
692 memset(mega_stats, 0, sizeof(*mega_stats));
694 stats->n_flows = ovs_flow_tbl_count(&dp->table);
695 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
697 stats->n_hit = stats->n_missed = stats->n_lost = 0;
699 for_each_possible_cpu(i) {
700 const struct dp_stats_percpu *percpu_stats;
701 struct dp_stats_percpu local_stats;
704 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
707 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
708 local_stats = *percpu_stats;
709 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
711 stats->n_hit += local_stats.n_hit;
712 stats->n_missed += local_stats.n_missed;
713 stats->n_lost += local_stats.n_lost;
714 mega_stats->n_mask_hit += local_stats.n_mask_hit;
718 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
720 return ovs_identifier_is_ufid(sfid) &&
721 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
724 static bool should_fill_mask(uint32_t ufid_flags)
726 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
729 static bool should_fill_actions(uint32_t ufid_flags)
731 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
734 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
735 const struct sw_flow_id *sfid,
738 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
740 /* OVS_FLOW_ATTR_UFID */
741 if (sfid && ovs_identifier_is_ufid(sfid))
742 len += nla_total_size(sfid->ufid_len);
744 /* OVS_FLOW_ATTR_KEY */
745 if (!sfid || should_fill_key(sfid, ufid_flags))
746 len += nla_total_size(ovs_key_attr_size());
748 /* OVS_FLOW_ATTR_MASK */
749 if (should_fill_mask(ufid_flags))
750 len += nla_total_size(ovs_key_attr_size());
752 /* OVS_FLOW_ATTR_ACTIONS */
753 if (should_fill_actions(ufid_flags))
754 len += nla_total_size(acts->orig_len);
757 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
758 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
759 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
762 /* Called with ovs_mutex or RCU read lock. */
763 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
766 struct ovs_flow_stats stats;
770 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
773 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
777 if (stats.n_packets &&
778 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
779 sizeof(struct ovs_flow_stats), &stats,
783 if ((u8)ntohs(tcp_flags) &&
784 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
790 /* Called with ovs_mutex or RCU read lock. */
791 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
792 struct sk_buff *skb, int skb_orig_len)
794 struct nlattr *start;
797 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
798 * this is the first flow to be dumped into 'skb'. This is unusual for
799 * Netlink but individual action lists can be longer than
800 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
801 * The userspace caller can always fetch the actions separately if it
802 * really wants them. (Most userspace callers in fact don't care.)
804 * This can only fail for dump operations because the skb is always
805 * properly sized for single flows.
807 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
809 const struct sw_flow_actions *sf_acts;
811 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
812 err = ovs_nla_put_actions(sf_acts->actions,
813 sf_acts->actions_len, skb);
816 nla_nest_end(skb, start);
821 nla_nest_cancel(skb, start);
823 } else if (skb_orig_len) {
830 /* Called with ovs_mutex or RCU read lock. */
831 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
832 struct sk_buff *skb, u32 portid,
833 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
835 const int skb_orig_len = skb->len;
836 struct ovs_header *ovs_header;
839 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
844 ovs_header->dp_ifindex = dp_ifindex;
846 err = ovs_nla_put_identifier(flow, skb);
850 if (should_fill_key(&flow->id, ufid_flags)) {
851 err = ovs_nla_put_masked_key(flow, skb);
856 if (should_fill_mask(ufid_flags)) {
857 err = ovs_nla_put_mask(flow, skb);
862 err = ovs_flow_cmd_fill_stats(flow, skb);
866 if (should_fill_actions(ufid_flags)) {
867 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
872 genlmsg_end(skb, ovs_header);
876 genlmsg_cancel(skb, ovs_header);
880 /* May not be called with RCU read lock. */
881 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
882 const struct sw_flow_id *sfid,
883 struct genl_info *info,
890 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
893 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
894 skb = genlmsg_new(len, GFP_KERNEL);
896 return ERR_PTR(-ENOMEM);
901 /* Called with ovs_mutex. */
902 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
904 struct genl_info *info, u8 cmd,
905 bool always, u32 ufid_flags)
910 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
911 &flow->id, info, always, ufid_flags);
912 if (IS_ERR_OR_NULL(skb))
915 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
916 info->snd_portid, info->snd_seq, 0,
922 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
924 struct net *net = sock_net(skb->sk);
925 struct nlattr **a = info->attrs;
926 struct ovs_header *ovs_header = info->userhdr;
927 struct sw_flow *flow = NULL, *new_flow;
928 struct sw_flow_mask mask;
929 struct sk_buff *reply;
931 struct sw_flow_key key;
932 struct sw_flow_actions *acts;
933 struct sw_flow_match match;
934 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
936 bool log = !a[OVS_FLOW_ATTR_PROBE];
938 /* Must have key and actions. */
940 if (!a[OVS_FLOW_ATTR_KEY]) {
941 OVS_NLERR(log, "Flow key attr not present in new flow.");
944 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
945 OVS_NLERR(log, "Flow actions attr not present in new flow.");
949 /* Most of the time we need to allocate a new flow, do it before
952 new_flow = ovs_flow_alloc();
953 if (IS_ERR(new_flow)) {
954 error = PTR_ERR(new_flow);
959 ovs_match_init(&match, &key, &mask);
960 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
961 a[OVS_FLOW_ATTR_MASK], log);
965 ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
967 /* Extract flow identifier. */
968 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
973 /* Validate actions. */
974 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
975 &new_flow->key, &acts, log);
977 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
981 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
984 error = PTR_ERR(reply);
989 dp = get_dp(net, ovs_header->dp_ifindex);
995 /* Check if this is a duplicate flow */
996 if (ovs_identifier_is_ufid(&new_flow->id))
997 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
999 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1000 if (likely(!flow)) {
1001 rcu_assign_pointer(new_flow->sf_acts, acts);
1003 /* Put flow in bucket. */
1004 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1005 if (unlikely(error)) {
1007 goto err_unlock_ovs;
1010 if (unlikely(reply)) {
1011 error = ovs_flow_cmd_fill_info(new_flow,
1012 ovs_header->dp_ifindex,
1013 reply, info->snd_portid,
1021 struct sw_flow_actions *old_acts;
1023 /* Bail out if we're not allowed to modify an existing flow.
1024 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1025 * because Generic Netlink treats the latter as a dump
1026 * request. We also accept NLM_F_EXCL in case that bug ever
1029 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1032 goto err_unlock_ovs;
1034 /* The flow identifier has to be the same for flow updates.
1035 * Look for any overlapping flow.
1037 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1038 if (ovs_identifier_is_key(&flow->id))
1039 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1041 else /* UFID matches but key is different */
1045 goto err_unlock_ovs;
1048 /* Update actions. */
1049 old_acts = ovsl_dereference(flow->sf_acts);
1050 rcu_assign_pointer(flow->sf_acts, acts);
1052 if (unlikely(reply)) {
1053 error = ovs_flow_cmd_fill_info(flow,
1054 ovs_header->dp_ifindex,
1055 reply, info->snd_portid,
1063 ovs_nla_free_flow_actions_rcu(old_acts);
1064 ovs_flow_free(new_flow, false);
1068 ovs_notify(&dp_flow_genl_family, reply, info);
1075 ovs_nla_free_flow_actions(acts);
1077 ovs_flow_free(new_flow, false);
1082 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1083 static struct sw_flow_actions *get_flow_actions(struct net *net,
1084 const struct nlattr *a,
1085 const struct sw_flow_key *key,
1086 const struct sw_flow_mask *mask,
1089 struct sw_flow_actions *acts;
1090 struct sw_flow_key masked_key;
1093 ovs_flow_mask_key(&masked_key, key, true, mask);
1094 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1097 "Actions may not be safe on all matching packets");
1098 return ERR_PTR(error);
1104 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1106 struct net *net = sock_net(skb->sk);
1107 struct nlattr **a = info->attrs;
1108 struct ovs_header *ovs_header = info->userhdr;
1109 struct sw_flow_key key;
1110 struct sw_flow *flow;
1111 struct sw_flow_mask mask;
1112 struct sk_buff *reply = NULL;
1113 struct datapath *dp;
1114 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1115 struct sw_flow_match match;
1116 struct sw_flow_id sfid;
1117 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1119 bool log = !a[OVS_FLOW_ATTR_PROBE];
1122 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1123 if (a[OVS_FLOW_ATTR_KEY]) {
1124 ovs_match_init(&match, &key, &mask);
1125 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1126 a[OVS_FLOW_ATTR_MASK], log);
1127 } else if (!ufid_present) {
1129 "Flow set message rejected, Key attribute missing.");
1135 /* Validate actions. */
1136 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1137 if (!a[OVS_FLOW_ATTR_KEY]) {
1139 "Flow key attribute not present in set flow.");
1144 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1147 error = PTR_ERR(acts);
1151 /* Can allocate before locking if have acts. */
1152 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1154 if (IS_ERR(reply)) {
1155 error = PTR_ERR(reply);
1156 goto err_kfree_acts;
1161 dp = get_dp(net, ovs_header->dp_ifindex);
1162 if (unlikely(!dp)) {
1164 goto err_unlock_ovs;
1166 /* Check that the flow exists. */
1168 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1170 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1171 if (unlikely(!flow)) {
1173 goto err_unlock_ovs;
1176 /* Update actions, if present. */
1178 old_acts = ovsl_dereference(flow->sf_acts);
1179 rcu_assign_pointer(flow->sf_acts, acts);
1181 if (unlikely(reply)) {
1182 error = ovs_flow_cmd_fill_info(flow,
1183 ovs_header->dp_ifindex,
1184 reply, info->snd_portid,
1191 /* Could not alloc without acts before locking. */
1192 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1193 info, OVS_FLOW_CMD_NEW, false,
1196 if (IS_ERR(reply)) {
1197 error = PTR_ERR(reply);
1198 goto err_unlock_ovs;
1203 if (a[OVS_FLOW_ATTR_CLEAR])
1204 ovs_flow_stats_clear(flow);
1208 ovs_notify(&dp_flow_genl_family, reply, info);
1210 ovs_nla_free_flow_actions_rcu(old_acts);
1218 ovs_nla_free_flow_actions(acts);
1223 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1225 struct nlattr **a = info->attrs;
1226 struct ovs_header *ovs_header = info->userhdr;
1227 struct net *net = sock_net(skb->sk);
1228 struct sw_flow_key key;
1229 struct sk_buff *reply;
1230 struct sw_flow *flow;
1231 struct datapath *dp;
1232 struct sw_flow_match match;
1233 struct sw_flow_id ufid;
1234 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1236 bool log = !a[OVS_FLOW_ATTR_PROBE];
1239 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1240 if (a[OVS_FLOW_ATTR_KEY]) {
1241 ovs_match_init(&match, &key, NULL);
1242 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1244 } else if (!ufid_present) {
1246 "Flow get message rejected, Key attribute missing.");
1253 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1260 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1262 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1268 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1269 OVS_FLOW_CMD_NEW, true, ufid_flags);
1270 if (IS_ERR(reply)) {
1271 err = PTR_ERR(reply);
1276 return genlmsg_reply(reply, info);
1282 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1284 struct nlattr **a = info->attrs;
1285 struct ovs_header *ovs_header = info->userhdr;
1286 struct net *net = sock_net(skb->sk);
1287 struct sw_flow_key key;
1288 struct sk_buff *reply;
1289 struct sw_flow *flow = NULL;
1290 struct datapath *dp;
1291 struct sw_flow_match match;
1292 struct sw_flow_id ufid;
1293 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1295 bool log = !a[OVS_FLOW_ATTR_PROBE];
1298 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1299 if (a[OVS_FLOW_ATTR_KEY]) {
1300 ovs_match_init(&match, &key, NULL);
1301 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1308 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1309 if (unlikely(!dp)) {
1314 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1315 err = ovs_flow_tbl_flush(&dp->table);
1320 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1322 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1323 if (unlikely(!flow)) {
1328 ovs_flow_tbl_remove(&dp->table, flow);
1331 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1332 &flow->id, info, false, ufid_flags);
1333 if (likely(reply)) {
1334 if (likely(!IS_ERR(reply))) {
1335 rcu_read_lock(); /*To keep RCU checker happy. */
1336 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1337 reply, info->snd_portid,
1344 ovs_notify(&dp_flow_genl_family, reply, info);
1346 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1350 ovs_flow_free(flow, true);
1357 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1359 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1360 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1361 struct table_instance *ti;
1362 struct datapath *dp;
1366 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1367 OVS_FLOW_ATTR_MAX, flow_policy);
1370 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1373 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1379 ti = rcu_dereference(dp->table.ti);
1381 struct sw_flow *flow;
1384 bucket = cb->args[0];
1386 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1390 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1391 NETLINK_CB(cb->skb).portid,
1392 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1393 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1396 cb->args[0] = bucket;
1403 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1404 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1405 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1406 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1407 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1408 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1409 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1410 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1413 static const struct genl_ops dp_flow_genl_ops[] = {
1414 { .cmd = OVS_FLOW_CMD_NEW,
1415 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1416 .policy = flow_policy,
1417 .doit = ovs_flow_cmd_new
1419 { .cmd = OVS_FLOW_CMD_DEL,
1420 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1421 .policy = flow_policy,
1422 .doit = ovs_flow_cmd_del
1424 { .cmd = OVS_FLOW_CMD_GET,
1425 .flags = 0, /* OK for unprivileged users. */
1426 .policy = flow_policy,
1427 .doit = ovs_flow_cmd_get,
1428 .dumpit = ovs_flow_cmd_dump
1430 { .cmd = OVS_FLOW_CMD_SET,
1431 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1432 .policy = flow_policy,
1433 .doit = ovs_flow_cmd_set,
1437 static struct genl_family dp_flow_genl_family = {
1438 .id = GENL_ID_GENERATE,
1439 .hdrsize = sizeof(struct ovs_header),
1440 .name = OVS_FLOW_FAMILY,
1441 .version = OVS_FLOW_VERSION,
1442 .maxattr = OVS_FLOW_ATTR_MAX,
1444 .parallel_ops = true,
1445 .ops = dp_flow_genl_ops,
1446 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1447 .mcgrps = &ovs_dp_flow_multicast_group,
1451 static size_t ovs_dp_cmd_msg_size(void)
1453 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1455 msgsize += nla_total_size(IFNAMSIZ);
1456 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1457 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1458 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1463 /* Called with ovs_mutex. */
1464 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1465 u32 portid, u32 seq, u32 flags, u8 cmd)
1467 struct ovs_header *ovs_header;
1468 struct ovs_dp_stats dp_stats;
1469 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1472 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1477 ovs_header->dp_ifindex = get_dpifindex(dp);
1479 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1481 goto nla_put_failure;
1483 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1484 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1485 &dp_stats, OVS_DP_ATTR_PAD))
1486 goto nla_put_failure;
1488 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1489 sizeof(struct ovs_dp_megaflow_stats),
1490 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1491 goto nla_put_failure;
1493 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1494 goto nla_put_failure;
1496 genlmsg_end(skb, ovs_header);
1500 genlmsg_cancel(skb, ovs_header);
1505 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1507 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1510 /* Called with rcu_read_lock or ovs_mutex. */
1511 static struct datapath *lookup_datapath(struct net *net,
1512 const struct ovs_header *ovs_header,
1513 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1515 struct datapath *dp;
1517 if (!a[OVS_DP_ATTR_NAME])
1518 dp = get_dp(net, ovs_header->dp_ifindex);
1520 struct vport *vport;
1522 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1523 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1525 return dp ? dp : ERR_PTR(-ENODEV);
1528 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1530 struct datapath *dp;
1532 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1536 WARN(dp->user_features, "Dropping previously announced user features\n");
1537 dp->user_features = 0;
1540 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1542 if (a[OVS_DP_ATTR_USER_FEATURES])
1543 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1546 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1548 struct nlattr **a = info->attrs;
1549 struct vport_parms parms;
1550 struct sk_buff *reply;
1551 struct datapath *dp;
1552 struct vport *vport;
1553 struct ovs_net *ovs_net;
1557 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1560 reply = ovs_dp_cmd_alloc_info();
1565 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1567 goto err_free_reply;
1569 ovs_dp_set_net(dp, sock_net(skb->sk));
1571 /* Allocate table. */
1572 err = ovs_flow_tbl_init(&dp->table);
1576 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1577 if (!dp->stats_percpu) {
1579 goto err_destroy_table;
1582 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1586 goto err_destroy_percpu;
1589 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1590 INIT_HLIST_HEAD(&dp->ports[i]);
1592 /* Set up our datapath device. */
1593 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1594 parms.type = OVS_VPORT_TYPE_INTERNAL;
1595 parms.options = NULL;
1597 parms.port_no = OVSP_LOCAL;
1598 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1600 ovs_dp_change(dp, a);
1602 /* So far only local changes have been made, now need the lock. */
1605 vport = new_vport(&parms);
1606 if (IS_ERR(vport)) {
1607 err = PTR_ERR(vport);
1611 if (err == -EEXIST) {
1612 /* An outdated user space instance that does not understand
1613 * the concept of user_features has attempted to create a new
1614 * datapath and is likely to reuse it. Drop all user features.
1616 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1617 ovs_dp_reset_user_features(skb, info);
1620 goto err_destroy_ports_array;
1623 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1624 info->snd_seq, 0, OVS_DP_CMD_NEW);
1627 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1628 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1632 ovs_notify(&dp_datapath_genl_family, reply, info);
1635 err_destroy_ports_array:
1639 free_percpu(dp->stats_percpu);
1641 ovs_flow_tbl_destroy(&dp->table);
1650 /* Called with ovs_mutex. */
1651 static void __dp_destroy(struct datapath *dp)
1655 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1656 struct vport *vport;
1657 struct hlist_node *n;
1659 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1660 if (vport->port_no != OVSP_LOCAL)
1661 ovs_dp_detach_port(vport);
1664 list_del_rcu(&dp->list_node);
1666 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1667 * all ports in datapath are destroyed first before freeing datapath.
1669 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1671 /* RCU destroy the flow table */
1672 call_rcu(&dp->rcu, destroy_dp_rcu);
1675 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1677 struct sk_buff *reply;
1678 struct datapath *dp;
1681 reply = ovs_dp_cmd_alloc_info();
1686 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1689 goto err_unlock_free;
1691 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1692 info->snd_seq, 0, OVS_DP_CMD_DEL);
1698 ovs_notify(&dp_datapath_genl_family, reply, info);
1708 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1710 struct sk_buff *reply;
1711 struct datapath *dp;
1714 reply = ovs_dp_cmd_alloc_info();
1719 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1722 goto err_unlock_free;
1724 ovs_dp_change(dp, info->attrs);
1726 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1727 info->snd_seq, 0, OVS_DP_CMD_NEW);
1731 ovs_notify(&dp_datapath_genl_family, reply, info);
1741 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1743 struct sk_buff *reply;
1744 struct datapath *dp;
1747 reply = ovs_dp_cmd_alloc_info();
1752 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1755 goto err_unlock_free;
1757 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1758 info->snd_seq, 0, OVS_DP_CMD_NEW);
1762 return genlmsg_reply(reply, info);
1770 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1772 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1773 struct datapath *dp;
1774 int skip = cb->args[0];
1778 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1780 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1781 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1782 OVS_DP_CMD_NEW) < 0)
1793 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1794 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1795 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1796 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1799 static const struct genl_ops dp_datapath_genl_ops[] = {
1800 { .cmd = OVS_DP_CMD_NEW,
1801 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1802 .policy = datapath_policy,
1803 .doit = ovs_dp_cmd_new
1805 { .cmd = OVS_DP_CMD_DEL,
1806 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1807 .policy = datapath_policy,
1808 .doit = ovs_dp_cmd_del
1810 { .cmd = OVS_DP_CMD_GET,
1811 .flags = 0, /* OK for unprivileged users. */
1812 .policy = datapath_policy,
1813 .doit = ovs_dp_cmd_get,
1814 .dumpit = ovs_dp_cmd_dump
1816 { .cmd = OVS_DP_CMD_SET,
1817 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1818 .policy = datapath_policy,
1819 .doit = ovs_dp_cmd_set,
1823 static struct genl_family dp_datapath_genl_family = {
1824 .id = GENL_ID_GENERATE,
1825 .hdrsize = sizeof(struct ovs_header),
1826 .name = OVS_DATAPATH_FAMILY,
1827 .version = OVS_DATAPATH_VERSION,
1828 .maxattr = OVS_DP_ATTR_MAX,
1830 .parallel_ops = true,
1831 .ops = dp_datapath_genl_ops,
1832 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1833 .mcgrps = &ovs_dp_datapath_multicast_group,
1837 /* Called with ovs_mutex or RCU read lock. */
1838 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1839 u32 portid, u32 seq, u32 flags, u8 cmd)
1841 struct ovs_header *ovs_header;
1842 struct ovs_vport_stats vport_stats;
1845 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1850 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1852 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1853 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1854 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1855 ovs_vport_name(vport)))
1856 goto nla_put_failure;
1858 ovs_vport_get_stats(vport, &vport_stats);
1859 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1860 sizeof(struct ovs_vport_stats), &vport_stats,
1861 OVS_VPORT_ATTR_PAD))
1862 goto nla_put_failure;
1864 if (ovs_vport_get_upcall_portids(vport, skb))
1865 goto nla_put_failure;
1867 err = ovs_vport_get_options(vport, skb);
1868 if (err == -EMSGSIZE)
1871 genlmsg_end(skb, ovs_header);
1877 genlmsg_cancel(skb, ovs_header);
1881 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1883 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1886 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1887 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1890 struct sk_buff *skb;
1893 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1895 return ERR_PTR(-ENOMEM);
1897 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1903 /* Called with ovs_mutex or RCU read lock. */
1904 static struct vport *lookup_vport(struct net *net,
1905 const struct ovs_header *ovs_header,
1906 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1908 struct datapath *dp;
1909 struct vport *vport;
1911 if (a[OVS_VPORT_ATTR_NAME]) {
1912 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1914 return ERR_PTR(-ENODEV);
1915 if (ovs_header->dp_ifindex &&
1916 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1917 return ERR_PTR(-ENODEV);
1919 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1920 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1922 if (port_no >= DP_MAX_PORTS)
1923 return ERR_PTR(-EFBIG);
1925 dp = get_dp(net, ovs_header->dp_ifindex);
1927 return ERR_PTR(-ENODEV);
1929 vport = ovs_vport_ovsl_rcu(dp, port_no);
1931 return ERR_PTR(-ENODEV);
1934 return ERR_PTR(-EINVAL);
1937 /* Called with ovs_mutex */
1938 static void update_headroom(struct datapath *dp)
1940 unsigned dev_headroom, max_headroom = 0;
1941 struct net_device *dev;
1942 struct vport *vport;
1945 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1946 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1948 dev_headroom = netdev_get_fwd_headroom(dev);
1949 if (dev_headroom > max_headroom)
1950 max_headroom = dev_headroom;
1954 dp->max_headroom = max_headroom;
1955 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1956 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1957 netdev_set_rx_headroom(vport->dev, max_headroom);
1960 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1962 struct nlattr **a = info->attrs;
1963 struct ovs_header *ovs_header = info->userhdr;
1964 struct vport_parms parms;
1965 struct sk_buff *reply;
1966 struct vport *vport;
1967 struct datapath *dp;
1971 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1972 !a[OVS_VPORT_ATTR_UPCALL_PID])
1975 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1976 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1977 if (port_no >= DP_MAX_PORTS)
1980 reply = ovs_vport_cmd_alloc_info();
1986 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1989 goto exit_unlock_free;
1992 vport = ovs_vport_ovsl(dp, port_no);
1995 goto exit_unlock_free;
1997 for (port_no = 1; ; port_no++) {
1998 if (port_no >= DP_MAX_PORTS) {
2000 goto exit_unlock_free;
2002 vport = ovs_vport_ovsl(dp, port_no);
2008 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2009 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2010 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2012 parms.port_no = port_no;
2013 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2015 vport = new_vport(&parms);
2016 err = PTR_ERR(vport);
2017 if (IS_ERR(vport)) {
2020 goto exit_unlock_free;
2023 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2024 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2026 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2027 update_headroom(dp);
2029 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2034 ovs_notify(&dp_vport_genl_family, reply, info);
2043 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2045 struct nlattr **a = info->attrs;
2046 struct sk_buff *reply;
2047 struct vport *vport;
2050 reply = ovs_vport_cmd_alloc_info();
2055 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2056 err = PTR_ERR(vport);
2058 goto exit_unlock_free;
2060 if (a[OVS_VPORT_ATTR_TYPE] &&
2061 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2063 goto exit_unlock_free;
2066 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2067 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2069 goto exit_unlock_free;
2073 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2074 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2076 err = ovs_vport_set_upcall_portids(vport, ids);
2078 goto exit_unlock_free;
2081 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2082 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2086 ovs_notify(&dp_vport_genl_family, reply, info);
2095 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2097 bool must_update_headroom = false;
2098 struct nlattr **a = info->attrs;
2099 struct sk_buff *reply;
2100 struct datapath *dp;
2101 struct vport *vport;
2104 reply = ovs_vport_cmd_alloc_info();
2109 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2110 err = PTR_ERR(vport);
2112 goto exit_unlock_free;
2114 if (vport->port_no == OVSP_LOCAL) {
2116 goto exit_unlock_free;
2119 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2120 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2123 /* the vport deletion may trigger dp headroom update */
2125 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2126 must_update_headroom = true;
2127 netdev_reset_rx_headroom(vport->dev);
2128 ovs_dp_detach_port(vport);
2130 if (must_update_headroom)
2131 update_headroom(dp);
2134 ovs_notify(&dp_vport_genl_family, reply, info);
2143 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2145 struct nlattr **a = info->attrs;
2146 struct ovs_header *ovs_header = info->userhdr;
2147 struct sk_buff *reply;
2148 struct vport *vport;
2151 reply = ovs_vport_cmd_alloc_info();
2156 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2157 err = PTR_ERR(vport);
2159 goto exit_unlock_free;
2160 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2161 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2165 return genlmsg_reply(reply, info);
2173 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2175 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2176 struct datapath *dp;
2177 int bucket = cb->args[0], skip = cb->args[1];
2181 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2186 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2187 struct vport *vport;
2190 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2192 ovs_vport_cmd_fill_info(vport, skb,
2193 NETLINK_CB(cb->skb).portid,
2196 OVS_VPORT_CMD_NEW) < 0)
2212 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2213 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2214 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2215 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2216 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2217 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2218 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2221 static const struct genl_ops dp_vport_genl_ops[] = {
2222 { .cmd = OVS_VPORT_CMD_NEW,
2223 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2224 .policy = vport_policy,
2225 .doit = ovs_vport_cmd_new
2227 { .cmd = OVS_VPORT_CMD_DEL,
2228 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2229 .policy = vport_policy,
2230 .doit = ovs_vport_cmd_del
2232 { .cmd = OVS_VPORT_CMD_GET,
2233 .flags = 0, /* OK for unprivileged users. */
2234 .policy = vport_policy,
2235 .doit = ovs_vport_cmd_get,
2236 .dumpit = ovs_vport_cmd_dump
2238 { .cmd = OVS_VPORT_CMD_SET,
2239 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2240 .policy = vport_policy,
2241 .doit = ovs_vport_cmd_set,
2245 struct genl_family dp_vport_genl_family = {
2246 .id = GENL_ID_GENERATE,
2247 .hdrsize = sizeof(struct ovs_header),
2248 .name = OVS_VPORT_FAMILY,
2249 .version = OVS_VPORT_VERSION,
2250 .maxattr = OVS_VPORT_ATTR_MAX,
2252 .parallel_ops = true,
2253 .ops = dp_vport_genl_ops,
2254 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2255 .mcgrps = &ovs_dp_vport_multicast_group,
2259 static struct genl_family * const dp_genl_families[] = {
2260 &dp_datapath_genl_family,
2261 &dp_vport_genl_family,
2262 &dp_flow_genl_family,
2263 &dp_packet_genl_family,
2266 static void dp_unregister_genl(int n_families)
2270 for (i = 0; i < n_families; i++)
2271 genl_unregister_family(dp_genl_families[i]);
2274 static int dp_register_genl(void)
2279 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2281 err = genl_register_family(dp_genl_families[i]);
2289 dp_unregister_genl(i);
2293 static int __net_init ovs_init_net(struct net *net)
2295 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2297 INIT_LIST_HEAD(&ovs_net->dps);
2298 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2303 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2304 struct list_head *head)
2306 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2307 struct datapath *dp;
2309 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2312 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2313 struct vport *vport;
2315 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2316 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2319 if (dev_net(vport->dev) == dnet)
2320 list_add(&vport->detach_list, head);
2326 static void __net_exit ovs_exit_net(struct net *dnet)
2328 struct datapath *dp, *dp_next;
2329 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2330 struct vport *vport, *vport_next;
2336 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2341 list_vports_from_net(net, dnet, &head);
2344 /* Detach all vports from given namespace. */
2345 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2346 list_del(&vport->detach_list);
2347 ovs_dp_detach_port(vport);
2352 cancel_work_sync(&ovs_net->dp_notify_work);
2355 static struct pernet_operations ovs_net_ops = {
2356 .init = ovs_init_net,
2357 .exit = ovs_exit_net,
2359 .size = sizeof(struct ovs_net),
2362 static int __init dp_init(void)
2366 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2368 pr_info("Open vSwitch switching datapath\n");
2370 err = action_fifos_init();
2374 err = ovs_internal_dev_rtnl_link_register();
2376 goto error_action_fifos_exit;
2378 err = ovs_flow_init();
2380 goto error_unreg_rtnl_link;
2382 err = ovs_vport_init();
2384 goto error_flow_exit;
2386 err = register_pernet_device(&ovs_net_ops);
2388 goto error_vport_exit;
2390 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2392 goto error_netns_exit;
2394 err = ovs_netdev_init();
2396 goto error_unreg_notifier;
2398 err = dp_register_genl();
2400 goto error_unreg_netdev;
2406 error_unreg_notifier:
2407 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2409 unregister_pernet_device(&ovs_net_ops);
2414 error_unreg_rtnl_link:
2415 ovs_internal_dev_rtnl_link_unregister();
2416 error_action_fifos_exit:
2417 action_fifos_exit();
2422 static void dp_cleanup(void)
2424 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2426 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2427 unregister_pernet_device(&ovs_net_ops);
2431 ovs_internal_dev_rtnl_link_unregister();
2432 action_fifos_exit();
2435 module_init(dp_init);
2436 module_exit(dp_cleanup);
2438 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2439 MODULE_LICENSE("GPL");