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 *);
141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
142 const struct sw_flow_key *,
143 const struct dp_upcall_info *);
145 /* Must be called with rcu_read_lock. */
146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
148 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
151 struct vport *vport = ovs_internal_dev_get_vport(dev);
159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160 * returned dp pointer valid.
162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
166 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
168 dp = get_dp_rcu(net, dp_ifindex);
174 /* Must be called with rcu_read_lock or ovs_mutex. */
175 const char *ovs_dp_name(const struct datapath *dp)
177 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
178 return ovs_vport_name(vport);
181 static int get_dpifindex(const struct datapath *dp)
188 local = ovs_vport_rcu(dp, OVSP_LOCAL);
190 ifindex = local->dev->ifindex;
199 static void destroy_dp_rcu(struct rcu_head *rcu)
201 struct datapath *dp = container_of(rcu, struct datapath, rcu);
203 ovs_flow_tbl_destroy(&dp->table);
204 free_percpu(dp->stats_percpu);
209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
212 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
215 /* Called with ovs_mutex or RCU read lock. */
216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
219 struct hlist_head *head;
221 head = vport_hash_bucket(dp, port_no);
222 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
223 if (vport->port_no == port_no)
229 /* Called with ovs_mutex. */
230 static struct vport *new_vport(const struct vport_parms *parms)
234 vport = ovs_vport_add(parms);
235 if (!IS_ERR(vport)) {
236 struct datapath *dp = parms->dp;
237 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
239 hlist_add_head_rcu(&vport->dp_hash_node, head);
244 void ovs_dp_detach_port(struct vport *p)
248 /* First drop references to device. */
249 hlist_del_rcu(&p->dp_hash_node);
251 /* Then destroy it. */
255 /* Must be called with rcu_read_lock. */
256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
258 const struct vport *p = OVS_CB(skb)->input_vport;
259 struct datapath *dp = p->dp;
260 struct sw_flow *flow;
261 struct sw_flow_actions *sf_acts;
262 struct dp_stats_percpu *stats;
266 stats = this_cpu_ptr(dp->stats_percpu);
269 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
270 if (unlikely(!flow)) {
271 struct dp_upcall_info upcall;
274 memset(&upcall, 0, sizeof(upcall));
275 upcall.cmd = OVS_PACKET_CMD_MISS;
276 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 upcall.mru = OVS_CB(skb)->mru;
278 error = ovs_dp_upcall(dp, skb, key, &upcall);
283 stats_counter = &stats->n_missed;
287 ovs_flow_stats_update(flow, key->tp.flags, skb);
288 sf_acts = rcu_dereference(flow->sf_acts);
289 ovs_execute_actions(dp, skb, sf_acts, key);
291 stats_counter = &stats->n_hit;
294 /* Update datapath statistics. */
295 u64_stats_update_begin(&stats->syncp);
297 stats->n_mask_hit += n_mask_hit;
298 u64_stats_update_end(&stats->syncp);
301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
302 const struct sw_flow_key *key,
303 const struct dp_upcall_info *upcall_info)
305 struct dp_stats_percpu *stats;
308 if (upcall_info->portid == 0) {
313 if (!skb_is_gso(skb))
314 err = queue_userspace_packet(dp, skb, key, upcall_info);
316 err = queue_gso_packets(dp, skb, key, upcall_info);
323 stats = this_cpu_ptr(dp->stats_percpu);
325 u64_stats_update_begin(&stats->syncp);
327 u64_stats_update_end(&stats->syncp);
332 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
333 const struct sw_flow_key *key,
334 const struct dp_upcall_info *upcall_info)
336 unsigned short gso_type = skb_shinfo(skb)->gso_type;
337 struct sw_flow_key later_key;
338 struct sk_buff *segs, *nskb;
341 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
342 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
344 return PTR_ERR(segs);
348 if (gso_type & SKB_GSO_UDP) {
349 /* The initial flow key extracted by ovs_flow_key_extract()
350 * in this case is for a first fragment, so we need to
351 * properly mark later fragments.
354 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
357 /* Queue all of the segments. */
360 if (gso_type & SKB_GSO_UDP && skb != segs)
363 err = queue_userspace_packet(dp, skb, key, upcall_info);
367 } while ((skb = skb->next));
369 /* Free all of the segments. */
377 } while ((skb = nskb));
381 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
384 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
385 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
386 + nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
388 /* OVS_PACKET_ATTR_USERDATA */
389 if (upcall_info->userdata)
390 size += NLA_ALIGN(upcall_info->userdata->nla_len);
392 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
393 if (upcall_info->egress_tun_info)
394 size += nla_total_size(ovs_tun_key_attr_size());
396 /* OVS_PACKET_ATTR_ACTIONS */
397 if (upcall_info->actions_len)
398 size += nla_total_size(upcall_info->actions_len);
400 /* OVS_PACKET_ATTR_MRU */
401 if (upcall_info->mru)
402 size += nla_total_size(sizeof(upcall_info->mru));
407 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
409 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
410 size_t plen = NLA_ALIGN(skb->len) - skb->len;
413 memset(skb_put(skb, plen), 0, plen);
417 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
418 const struct sw_flow_key *key,
419 const struct dp_upcall_info *upcall_info)
421 struct ovs_header *upcall;
422 struct sk_buff *nskb = NULL;
423 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
429 dp_ifindex = get_dpifindex(dp);
433 if (skb_vlan_tag_present(skb)) {
434 nskb = skb_clone(skb, GFP_ATOMIC);
438 nskb = __vlan_hwaccel_push_inside(nskb);
445 if (nla_attr_size(skb->len) > USHRT_MAX) {
450 /* Complete checksum if needed */
451 if (skb->ip_summed == CHECKSUM_PARTIAL &&
452 (err = skb_checksum_help(skb)))
455 /* Older versions of OVS user space enforce alignment of the last
456 * Netlink attribute to NLA_ALIGNTO which would require extensive
457 * padding logic. Only perform zerocopy if padding is not required.
459 if (dp->user_features & OVS_DP_F_UNALIGNED)
460 hlen = skb_zerocopy_headlen(skb);
464 len = upcall_msg_size(upcall_info, hlen);
465 user_skb = genlmsg_new(len, GFP_ATOMIC);
471 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
472 0, upcall_info->cmd);
473 upcall->dp_ifindex = dp_ifindex;
475 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
478 if (upcall_info->userdata)
479 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
480 nla_len(upcall_info->userdata),
481 nla_data(upcall_info->userdata));
483 if (upcall_info->egress_tun_info) {
484 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
485 err = ovs_nla_put_tunnel_info(user_skb,
486 upcall_info->egress_tun_info);
488 nla_nest_end(user_skb, nla);
491 if (upcall_info->actions_len) {
492 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
493 err = ovs_nla_put_actions(upcall_info->actions,
494 upcall_info->actions_len,
497 nla_nest_end(user_skb, nla);
499 nla_nest_cancel(user_skb, nla);
502 /* Add OVS_PACKET_ATTR_MRU */
503 if (upcall_info->mru) {
504 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
509 pad_packet(dp, user_skb);
512 /* Only reserve room for attribute header, packet data is added
513 * in skb_zerocopy() */
514 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
518 nla->nla_len = nla_attr_size(skb->len);
520 err = skb_zerocopy(user_skb, skb, skb->len, hlen);
524 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
525 pad_packet(dp, user_skb);
527 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
529 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
539 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
541 struct ovs_header *ovs_header = info->userhdr;
542 struct net *net = sock_net(skb->sk);
543 struct nlattr **a = info->attrs;
544 struct sw_flow_actions *acts;
545 struct sk_buff *packet;
546 struct sw_flow *flow;
547 struct sw_flow_actions *sf_acts;
550 struct vport *input_vport;
554 bool log = !a[OVS_PACKET_ATTR_PROBE];
557 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
558 !a[OVS_PACKET_ATTR_ACTIONS])
561 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
562 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
566 skb_reserve(packet, NET_IP_ALIGN);
568 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
570 skb_reset_mac_header(packet);
571 eth = eth_hdr(packet);
573 /* Normally, setting the skb 'protocol' field would be handled by a
574 * call to eth_type_trans(), but it assumes there's a sending
575 * device, which we may not have. */
576 if (eth_proto_is_802_3(eth->h_proto))
577 packet->protocol = eth->h_proto;
579 packet->protocol = htons(ETH_P_802_2);
581 /* Set packet's mru */
582 if (a[OVS_PACKET_ATTR_MRU]) {
583 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
584 packet->ignore_df = 1;
586 OVS_CB(packet)->mru = mru;
588 /* Build an sw_flow for sending this packet. */
589 flow = ovs_flow_alloc();
594 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
595 packet, &flow->key, log);
599 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
600 &flow->key, &acts, log);
604 rcu_assign_pointer(flow->sf_acts, acts);
605 packet->priority = flow->key.phy.priority;
606 packet->mark = flow->key.phy.skb_mark;
609 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
614 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
616 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
621 packet->dev = input_vport->dev;
622 OVS_CB(packet)->input_vport = input_vport;
623 sf_acts = rcu_dereference(flow->sf_acts);
626 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
630 ovs_flow_free(flow, false);
636 ovs_flow_free(flow, false);
643 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
644 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
645 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
646 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
647 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
648 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
651 static const struct genl_ops dp_packet_genl_ops[] = {
652 { .cmd = OVS_PACKET_CMD_EXECUTE,
653 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
654 .policy = packet_policy,
655 .doit = ovs_packet_cmd_execute
659 static struct genl_family dp_packet_genl_family = {
660 .id = GENL_ID_GENERATE,
661 .hdrsize = sizeof(struct ovs_header),
662 .name = OVS_PACKET_FAMILY,
663 .version = OVS_PACKET_VERSION,
664 .maxattr = OVS_PACKET_ATTR_MAX,
666 .parallel_ops = true,
667 .ops = dp_packet_genl_ops,
668 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
671 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
672 struct ovs_dp_megaflow_stats *mega_stats)
676 memset(mega_stats, 0, sizeof(*mega_stats));
678 stats->n_flows = ovs_flow_tbl_count(&dp->table);
679 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
681 stats->n_hit = stats->n_missed = stats->n_lost = 0;
683 for_each_possible_cpu(i) {
684 const struct dp_stats_percpu *percpu_stats;
685 struct dp_stats_percpu local_stats;
688 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
691 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
692 local_stats = *percpu_stats;
693 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
695 stats->n_hit += local_stats.n_hit;
696 stats->n_missed += local_stats.n_missed;
697 stats->n_lost += local_stats.n_lost;
698 mega_stats->n_mask_hit += local_stats.n_mask_hit;
702 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
704 return ovs_identifier_is_ufid(sfid) &&
705 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
708 static bool should_fill_mask(uint32_t ufid_flags)
710 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
713 static bool should_fill_actions(uint32_t ufid_flags)
715 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
718 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
719 const struct sw_flow_id *sfid,
722 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
724 /* OVS_FLOW_ATTR_UFID */
725 if (sfid && ovs_identifier_is_ufid(sfid))
726 len += nla_total_size(sfid->ufid_len);
728 /* OVS_FLOW_ATTR_KEY */
729 if (!sfid || should_fill_key(sfid, ufid_flags))
730 len += nla_total_size(ovs_key_attr_size());
732 /* OVS_FLOW_ATTR_MASK */
733 if (should_fill_mask(ufid_flags))
734 len += nla_total_size(ovs_key_attr_size());
736 /* OVS_FLOW_ATTR_ACTIONS */
737 if (should_fill_actions(ufid_flags))
738 len += nla_total_size(acts->orig_len);
741 + nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
742 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
743 + nla_total_size(8); /* OVS_FLOW_ATTR_USED */
746 /* Called with ovs_mutex or RCU read lock. */
747 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
750 struct ovs_flow_stats stats;
754 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
757 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
761 if (stats.n_packets &&
762 nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
765 if ((u8)ntohs(tcp_flags) &&
766 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
772 /* Called with ovs_mutex or RCU read lock. */
773 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
774 struct sk_buff *skb, int skb_orig_len)
776 struct nlattr *start;
779 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
780 * this is the first flow to be dumped into 'skb'. This is unusual for
781 * Netlink but individual action lists can be longer than
782 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
783 * The userspace caller can always fetch the actions separately if it
784 * really wants them. (Most userspace callers in fact don't care.)
786 * This can only fail for dump operations because the skb is always
787 * properly sized for single flows.
789 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
791 const struct sw_flow_actions *sf_acts;
793 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
794 err = ovs_nla_put_actions(sf_acts->actions,
795 sf_acts->actions_len, skb);
798 nla_nest_end(skb, start);
803 nla_nest_cancel(skb, start);
805 } else if (skb_orig_len) {
812 /* Called with ovs_mutex or RCU read lock. */
813 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
814 struct sk_buff *skb, u32 portid,
815 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
817 const int skb_orig_len = skb->len;
818 struct ovs_header *ovs_header;
821 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
826 ovs_header->dp_ifindex = dp_ifindex;
828 err = ovs_nla_put_identifier(flow, skb);
832 if (should_fill_key(&flow->id, ufid_flags)) {
833 err = ovs_nla_put_masked_key(flow, skb);
838 if (should_fill_mask(ufid_flags)) {
839 err = ovs_nla_put_mask(flow, skb);
844 err = ovs_flow_cmd_fill_stats(flow, skb);
848 if (should_fill_actions(ufid_flags)) {
849 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
854 genlmsg_end(skb, ovs_header);
858 genlmsg_cancel(skb, ovs_header);
862 /* May not be called with RCU read lock. */
863 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
864 const struct sw_flow_id *sfid,
865 struct genl_info *info,
872 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
875 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
876 skb = genlmsg_new(len, GFP_KERNEL);
878 return ERR_PTR(-ENOMEM);
883 /* Called with ovs_mutex. */
884 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
886 struct genl_info *info, u8 cmd,
887 bool always, u32 ufid_flags)
892 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
893 &flow->id, info, always, ufid_flags);
894 if (IS_ERR_OR_NULL(skb))
897 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
898 info->snd_portid, info->snd_seq, 0,
904 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
906 struct net *net = sock_net(skb->sk);
907 struct nlattr **a = info->attrs;
908 struct ovs_header *ovs_header = info->userhdr;
909 struct sw_flow *flow = NULL, *new_flow;
910 struct sw_flow_mask mask;
911 struct sk_buff *reply;
913 struct sw_flow_key key;
914 struct sw_flow_actions *acts;
915 struct sw_flow_match match;
916 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
918 bool log = !a[OVS_FLOW_ATTR_PROBE];
920 /* Must have key and actions. */
922 if (!a[OVS_FLOW_ATTR_KEY]) {
923 OVS_NLERR(log, "Flow key attr not present in new flow.");
926 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
927 OVS_NLERR(log, "Flow actions attr not present in new flow.");
931 /* Most of the time we need to allocate a new flow, do it before
934 new_flow = ovs_flow_alloc();
935 if (IS_ERR(new_flow)) {
936 error = PTR_ERR(new_flow);
941 ovs_match_init(&match, &key, &mask);
942 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
943 a[OVS_FLOW_ATTR_MASK], log);
947 ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
949 /* Extract flow identifier. */
950 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
955 /* Validate actions. */
956 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
957 &new_flow->key, &acts, log);
959 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
963 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
966 error = PTR_ERR(reply);
971 dp = get_dp(net, ovs_header->dp_ifindex);
977 /* Check if this is a duplicate flow */
978 if (ovs_identifier_is_ufid(&new_flow->id))
979 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
981 flow = ovs_flow_tbl_lookup(&dp->table, &key);
983 rcu_assign_pointer(new_flow->sf_acts, acts);
985 /* Put flow in bucket. */
986 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
987 if (unlikely(error)) {
992 if (unlikely(reply)) {
993 error = ovs_flow_cmd_fill_info(new_flow,
994 ovs_header->dp_ifindex,
995 reply, info->snd_portid,
1003 struct sw_flow_actions *old_acts;
1005 /* Bail out if we're not allowed to modify an existing flow.
1006 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1007 * because Generic Netlink treats the latter as a dump
1008 * request. We also accept NLM_F_EXCL in case that bug ever
1011 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1014 goto err_unlock_ovs;
1016 /* The flow identifier has to be the same for flow updates.
1017 * Look for any overlapping flow.
1019 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1020 if (ovs_identifier_is_key(&flow->id))
1021 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1023 else /* UFID matches but key is different */
1027 goto err_unlock_ovs;
1030 /* Update actions. */
1031 old_acts = ovsl_dereference(flow->sf_acts);
1032 rcu_assign_pointer(flow->sf_acts, acts);
1034 if (unlikely(reply)) {
1035 error = ovs_flow_cmd_fill_info(flow,
1036 ovs_header->dp_ifindex,
1037 reply, info->snd_portid,
1045 ovs_nla_free_flow_actions_rcu(old_acts);
1046 ovs_flow_free(new_flow, false);
1050 ovs_notify(&dp_flow_genl_family, reply, info);
1057 ovs_nla_free_flow_actions(acts);
1059 ovs_flow_free(new_flow, false);
1064 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1065 static struct sw_flow_actions *get_flow_actions(struct net *net,
1066 const struct nlattr *a,
1067 const struct sw_flow_key *key,
1068 const struct sw_flow_mask *mask,
1071 struct sw_flow_actions *acts;
1072 struct sw_flow_key masked_key;
1075 ovs_flow_mask_key(&masked_key, key, true, mask);
1076 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1079 "Actions may not be safe on all matching packets");
1080 return ERR_PTR(error);
1086 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1088 struct net *net = sock_net(skb->sk);
1089 struct nlattr **a = info->attrs;
1090 struct ovs_header *ovs_header = info->userhdr;
1091 struct sw_flow_key key;
1092 struct sw_flow *flow;
1093 struct sw_flow_mask mask;
1094 struct sk_buff *reply = NULL;
1095 struct datapath *dp;
1096 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1097 struct sw_flow_match match;
1098 struct sw_flow_id sfid;
1099 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1101 bool log = !a[OVS_FLOW_ATTR_PROBE];
1104 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1105 if (a[OVS_FLOW_ATTR_KEY]) {
1106 ovs_match_init(&match, &key, &mask);
1107 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1108 a[OVS_FLOW_ATTR_MASK], log);
1109 } else if (!ufid_present) {
1111 "Flow set message rejected, Key attribute missing.");
1117 /* Validate actions. */
1118 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1119 if (!a[OVS_FLOW_ATTR_KEY]) {
1121 "Flow key attribute not present in set flow.");
1126 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1129 error = PTR_ERR(acts);
1133 /* Can allocate before locking if have acts. */
1134 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1136 if (IS_ERR(reply)) {
1137 error = PTR_ERR(reply);
1138 goto err_kfree_acts;
1143 dp = get_dp(net, ovs_header->dp_ifindex);
1144 if (unlikely(!dp)) {
1146 goto err_unlock_ovs;
1148 /* Check that the flow exists. */
1150 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1152 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1153 if (unlikely(!flow)) {
1155 goto err_unlock_ovs;
1158 /* Update actions, if present. */
1160 old_acts = ovsl_dereference(flow->sf_acts);
1161 rcu_assign_pointer(flow->sf_acts, acts);
1163 if (unlikely(reply)) {
1164 error = ovs_flow_cmd_fill_info(flow,
1165 ovs_header->dp_ifindex,
1166 reply, info->snd_portid,
1173 /* Could not alloc without acts before locking. */
1174 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1175 info, OVS_FLOW_CMD_NEW, false,
1178 if (IS_ERR(reply)) {
1179 error = PTR_ERR(reply);
1180 goto err_unlock_ovs;
1185 if (a[OVS_FLOW_ATTR_CLEAR])
1186 ovs_flow_stats_clear(flow);
1190 ovs_notify(&dp_flow_genl_family, reply, info);
1192 ovs_nla_free_flow_actions_rcu(old_acts);
1200 ovs_nla_free_flow_actions(acts);
1205 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1207 struct nlattr **a = info->attrs;
1208 struct ovs_header *ovs_header = info->userhdr;
1209 struct net *net = sock_net(skb->sk);
1210 struct sw_flow_key key;
1211 struct sk_buff *reply;
1212 struct sw_flow *flow;
1213 struct datapath *dp;
1214 struct sw_flow_match match;
1215 struct sw_flow_id ufid;
1216 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1218 bool log = !a[OVS_FLOW_ATTR_PROBE];
1221 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1222 if (a[OVS_FLOW_ATTR_KEY]) {
1223 ovs_match_init(&match, &key, NULL);
1224 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1226 } else if (!ufid_present) {
1228 "Flow get message rejected, Key attribute missing.");
1235 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1242 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1244 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1250 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1251 OVS_FLOW_CMD_NEW, true, ufid_flags);
1252 if (IS_ERR(reply)) {
1253 err = PTR_ERR(reply);
1258 return genlmsg_reply(reply, info);
1264 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1266 struct nlattr **a = info->attrs;
1267 struct ovs_header *ovs_header = info->userhdr;
1268 struct net *net = sock_net(skb->sk);
1269 struct sw_flow_key key;
1270 struct sk_buff *reply;
1271 struct sw_flow *flow = NULL;
1272 struct datapath *dp;
1273 struct sw_flow_match match;
1274 struct sw_flow_id ufid;
1275 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1277 bool log = !a[OVS_FLOW_ATTR_PROBE];
1280 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1281 if (a[OVS_FLOW_ATTR_KEY]) {
1282 ovs_match_init(&match, &key, NULL);
1283 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1290 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1291 if (unlikely(!dp)) {
1296 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1297 err = ovs_flow_tbl_flush(&dp->table);
1302 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1304 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1305 if (unlikely(!flow)) {
1310 ovs_flow_tbl_remove(&dp->table, flow);
1313 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1314 &flow->id, info, false, ufid_flags);
1315 if (likely(reply)) {
1316 if (likely(!IS_ERR(reply))) {
1317 rcu_read_lock(); /*To keep RCU checker happy. */
1318 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1319 reply, info->snd_portid,
1326 ovs_notify(&dp_flow_genl_family, reply, info);
1328 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1332 ovs_flow_free(flow, true);
1339 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1341 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1342 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1343 struct table_instance *ti;
1344 struct datapath *dp;
1348 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1349 OVS_FLOW_ATTR_MAX, flow_policy);
1352 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1355 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1361 ti = rcu_dereference(dp->table.ti);
1363 struct sw_flow *flow;
1366 bucket = cb->args[0];
1368 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1372 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1373 NETLINK_CB(cb->skb).portid,
1374 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1375 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1378 cb->args[0] = bucket;
1385 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1386 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1387 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1388 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1389 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1390 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1391 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1392 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1395 static const struct genl_ops dp_flow_genl_ops[] = {
1396 { .cmd = OVS_FLOW_CMD_NEW,
1397 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1398 .policy = flow_policy,
1399 .doit = ovs_flow_cmd_new
1401 { .cmd = OVS_FLOW_CMD_DEL,
1402 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1403 .policy = flow_policy,
1404 .doit = ovs_flow_cmd_del
1406 { .cmd = OVS_FLOW_CMD_GET,
1407 .flags = 0, /* OK for unprivileged users. */
1408 .policy = flow_policy,
1409 .doit = ovs_flow_cmd_get,
1410 .dumpit = ovs_flow_cmd_dump
1412 { .cmd = OVS_FLOW_CMD_SET,
1413 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1414 .policy = flow_policy,
1415 .doit = ovs_flow_cmd_set,
1419 static struct genl_family dp_flow_genl_family = {
1420 .id = GENL_ID_GENERATE,
1421 .hdrsize = sizeof(struct ovs_header),
1422 .name = OVS_FLOW_FAMILY,
1423 .version = OVS_FLOW_VERSION,
1424 .maxattr = OVS_FLOW_ATTR_MAX,
1426 .parallel_ops = true,
1427 .ops = dp_flow_genl_ops,
1428 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1429 .mcgrps = &ovs_dp_flow_multicast_group,
1433 static size_t ovs_dp_cmd_msg_size(void)
1435 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1437 msgsize += nla_total_size(IFNAMSIZ);
1438 msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1439 msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1440 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1445 /* Called with ovs_mutex. */
1446 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1447 u32 portid, u32 seq, u32 flags, u8 cmd)
1449 struct ovs_header *ovs_header;
1450 struct ovs_dp_stats dp_stats;
1451 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1454 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1459 ovs_header->dp_ifindex = get_dpifindex(dp);
1461 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1463 goto nla_put_failure;
1465 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1466 if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1468 goto nla_put_failure;
1470 if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1471 sizeof(struct ovs_dp_megaflow_stats),
1472 &dp_megaflow_stats))
1473 goto nla_put_failure;
1475 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1476 goto nla_put_failure;
1478 genlmsg_end(skb, ovs_header);
1482 genlmsg_cancel(skb, ovs_header);
1487 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1489 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1492 /* Called with rcu_read_lock or ovs_mutex. */
1493 static struct datapath *lookup_datapath(struct net *net,
1494 const struct ovs_header *ovs_header,
1495 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1497 struct datapath *dp;
1499 if (!a[OVS_DP_ATTR_NAME])
1500 dp = get_dp(net, ovs_header->dp_ifindex);
1502 struct vport *vport;
1504 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1505 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1507 return dp ? dp : ERR_PTR(-ENODEV);
1510 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1512 struct datapath *dp;
1514 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1518 WARN(dp->user_features, "Dropping previously announced user features\n");
1519 dp->user_features = 0;
1522 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1524 if (a[OVS_DP_ATTR_USER_FEATURES])
1525 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1528 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1530 struct nlattr **a = info->attrs;
1531 struct vport_parms parms;
1532 struct sk_buff *reply;
1533 struct datapath *dp;
1534 struct vport *vport;
1535 struct ovs_net *ovs_net;
1539 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1542 reply = ovs_dp_cmd_alloc_info();
1547 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1549 goto err_free_reply;
1551 ovs_dp_set_net(dp, sock_net(skb->sk));
1553 /* Allocate table. */
1554 err = ovs_flow_tbl_init(&dp->table);
1558 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1559 if (!dp->stats_percpu) {
1561 goto err_destroy_table;
1564 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1568 goto err_destroy_percpu;
1571 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1572 INIT_HLIST_HEAD(&dp->ports[i]);
1574 /* Set up our datapath device. */
1575 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1576 parms.type = OVS_VPORT_TYPE_INTERNAL;
1577 parms.options = NULL;
1579 parms.port_no = OVSP_LOCAL;
1580 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1582 ovs_dp_change(dp, a);
1584 /* So far only local changes have been made, now need the lock. */
1587 vport = new_vport(&parms);
1588 if (IS_ERR(vport)) {
1589 err = PTR_ERR(vport);
1593 if (err == -EEXIST) {
1594 /* An outdated user space instance that does not understand
1595 * the concept of user_features has attempted to create a new
1596 * datapath and is likely to reuse it. Drop all user features.
1598 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1599 ovs_dp_reset_user_features(skb, info);
1602 goto err_destroy_ports_array;
1605 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1606 info->snd_seq, 0, OVS_DP_CMD_NEW);
1609 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1610 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1614 ovs_notify(&dp_datapath_genl_family, reply, info);
1617 err_destroy_ports_array:
1621 free_percpu(dp->stats_percpu);
1623 ovs_flow_tbl_destroy(&dp->table);
1632 /* Called with ovs_mutex. */
1633 static void __dp_destroy(struct datapath *dp)
1637 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1638 struct vport *vport;
1639 struct hlist_node *n;
1641 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1642 if (vport->port_no != OVSP_LOCAL)
1643 ovs_dp_detach_port(vport);
1646 list_del_rcu(&dp->list_node);
1648 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1649 * all ports in datapath are destroyed first before freeing datapath.
1651 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1653 /* RCU destroy the flow table */
1654 call_rcu(&dp->rcu, destroy_dp_rcu);
1657 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1659 struct sk_buff *reply;
1660 struct datapath *dp;
1663 reply = ovs_dp_cmd_alloc_info();
1668 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1671 goto err_unlock_free;
1673 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1674 info->snd_seq, 0, OVS_DP_CMD_DEL);
1680 ovs_notify(&dp_datapath_genl_family, reply, info);
1690 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1692 struct sk_buff *reply;
1693 struct datapath *dp;
1696 reply = ovs_dp_cmd_alloc_info();
1701 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1704 goto err_unlock_free;
1706 ovs_dp_change(dp, info->attrs);
1708 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1709 info->snd_seq, 0, OVS_DP_CMD_NEW);
1713 ovs_notify(&dp_datapath_genl_family, reply, info);
1723 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1725 struct sk_buff *reply;
1726 struct datapath *dp;
1729 reply = ovs_dp_cmd_alloc_info();
1734 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1737 goto err_unlock_free;
1739 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1740 info->snd_seq, 0, OVS_DP_CMD_NEW);
1744 return genlmsg_reply(reply, info);
1752 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1754 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1755 struct datapath *dp;
1756 int skip = cb->args[0];
1760 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1762 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1763 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1764 OVS_DP_CMD_NEW) < 0)
1775 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1776 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1777 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1778 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1781 static const struct genl_ops dp_datapath_genl_ops[] = {
1782 { .cmd = OVS_DP_CMD_NEW,
1783 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1784 .policy = datapath_policy,
1785 .doit = ovs_dp_cmd_new
1787 { .cmd = OVS_DP_CMD_DEL,
1788 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1789 .policy = datapath_policy,
1790 .doit = ovs_dp_cmd_del
1792 { .cmd = OVS_DP_CMD_GET,
1793 .flags = 0, /* OK for unprivileged users. */
1794 .policy = datapath_policy,
1795 .doit = ovs_dp_cmd_get,
1796 .dumpit = ovs_dp_cmd_dump
1798 { .cmd = OVS_DP_CMD_SET,
1799 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1800 .policy = datapath_policy,
1801 .doit = ovs_dp_cmd_set,
1805 static struct genl_family dp_datapath_genl_family = {
1806 .id = GENL_ID_GENERATE,
1807 .hdrsize = sizeof(struct ovs_header),
1808 .name = OVS_DATAPATH_FAMILY,
1809 .version = OVS_DATAPATH_VERSION,
1810 .maxattr = OVS_DP_ATTR_MAX,
1812 .parallel_ops = true,
1813 .ops = dp_datapath_genl_ops,
1814 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1815 .mcgrps = &ovs_dp_datapath_multicast_group,
1819 /* Called with ovs_mutex or RCU read lock. */
1820 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1821 u32 portid, u32 seq, u32 flags, u8 cmd)
1823 struct ovs_header *ovs_header;
1824 struct ovs_vport_stats vport_stats;
1827 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1832 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1834 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1835 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1836 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1837 ovs_vport_name(vport)))
1838 goto nla_put_failure;
1840 ovs_vport_get_stats(vport, &vport_stats);
1841 if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1843 goto nla_put_failure;
1845 if (ovs_vport_get_upcall_portids(vport, skb))
1846 goto nla_put_failure;
1848 err = ovs_vport_get_options(vport, skb);
1849 if (err == -EMSGSIZE)
1852 genlmsg_end(skb, ovs_header);
1858 genlmsg_cancel(skb, ovs_header);
1862 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1864 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1867 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1868 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1871 struct sk_buff *skb;
1874 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1876 return ERR_PTR(-ENOMEM);
1878 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1884 /* Called with ovs_mutex or RCU read lock. */
1885 static struct vport *lookup_vport(struct net *net,
1886 const struct ovs_header *ovs_header,
1887 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1889 struct datapath *dp;
1890 struct vport *vport;
1892 if (a[OVS_VPORT_ATTR_NAME]) {
1893 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1895 return ERR_PTR(-ENODEV);
1896 if (ovs_header->dp_ifindex &&
1897 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1898 return ERR_PTR(-ENODEV);
1900 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1901 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1903 if (port_no >= DP_MAX_PORTS)
1904 return ERR_PTR(-EFBIG);
1906 dp = get_dp(net, ovs_header->dp_ifindex);
1908 return ERR_PTR(-ENODEV);
1910 vport = ovs_vport_ovsl_rcu(dp, port_no);
1912 return ERR_PTR(-ENODEV);
1915 return ERR_PTR(-EINVAL);
1918 /* Called with ovs_mutex */
1919 static void update_headroom(struct datapath *dp)
1921 unsigned dev_headroom, max_headroom = 0;
1922 struct net_device *dev;
1923 struct vport *vport;
1926 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1927 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1929 dev_headroom = netdev_get_fwd_headroom(dev);
1930 if (dev_headroom > max_headroom)
1931 max_headroom = dev_headroom;
1935 dp->max_headroom = max_headroom;
1936 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1937 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1938 netdev_set_rx_headroom(vport->dev, max_headroom);
1941 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1943 struct nlattr **a = info->attrs;
1944 struct ovs_header *ovs_header = info->userhdr;
1945 struct vport_parms parms;
1946 struct sk_buff *reply;
1947 struct vport *vport;
1948 struct datapath *dp;
1952 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1953 !a[OVS_VPORT_ATTR_UPCALL_PID])
1956 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1957 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1958 if (port_no >= DP_MAX_PORTS)
1961 reply = ovs_vport_cmd_alloc_info();
1967 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1970 goto exit_unlock_free;
1973 vport = ovs_vport_ovsl(dp, port_no);
1976 goto exit_unlock_free;
1978 for (port_no = 1; ; port_no++) {
1979 if (port_no >= DP_MAX_PORTS) {
1981 goto exit_unlock_free;
1983 vport = ovs_vport_ovsl(dp, port_no);
1989 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1990 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1991 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1993 parms.port_no = port_no;
1994 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1996 vport = new_vport(&parms);
1997 err = PTR_ERR(vport);
1998 if (IS_ERR(vport)) {
2001 goto exit_unlock_free;
2004 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2005 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2007 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2008 update_headroom(dp);
2010 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2015 ovs_notify(&dp_vport_genl_family, reply, info);
2024 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2026 struct nlattr **a = info->attrs;
2027 struct sk_buff *reply;
2028 struct vport *vport;
2031 reply = ovs_vport_cmd_alloc_info();
2036 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2037 err = PTR_ERR(vport);
2039 goto exit_unlock_free;
2041 if (a[OVS_VPORT_ATTR_TYPE] &&
2042 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2044 goto exit_unlock_free;
2047 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2048 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2050 goto exit_unlock_free;
2054 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2055 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2057 err = ovs_vport_set_upcall_portids(vport, ids);
2059 goto exit_unlock_free;
2062 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2063 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2067 ovs_notify(&dp_vport_genl_family, reply, info);
2076 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2078 bool must_update_headroom = false;
2079 struct nlattr **a = info->attrs;
2080 struct sk_buff *reply;
2081 struct datapath *dp;
2082 struct vport *vport;
2085 reply = ovs_vport_cmd_alloc_info();
2090 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2091 err = PTR_ERR(vport);
2093 goto exit_unlock_free;
2095 if (vport->port_no == OVSP_LOCAL) {
2097 goto exit_unlock_free;
2100 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2101 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2104 /* the vport deletion may trigger dp headroom update */
2106 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2107 must_update_headroom = true;
2108 netdev_reset_rx_headroom(vport->dev);
2109 ovs_dp_detach_port(vport);
2111 if (must_update_headroom)
2112 update_headroom(dp);
2115 ovs_notify(&dp_vport_genl_family, reply, info);
2124 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2126 struct nlattr **a = info->attrs;
2127 struct ovs_header *ovs_header = info->userhdr;
2128 struct sk_buff *reply;
2129 struct vport *vport;
2132 reply = ovs_vport_cmd_alloc_info();
2137 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2138 err = PTR_ERR(vport);
2140 goto exit_unlock_free;
2141 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2142 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2146 return genlmsg_reply(reply, info);
2154 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2156 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2157 struct datapath *dp;
2158 int bucket = cb->args[0], skip = cb->args[1];
2162 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2167 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2168 struct vport *vport;
2171 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2173 ovs_vport_cmd_fill_info(vport, skb,
2174 NETLINK_CB(cb->skb).portid,
2177 OVS_VPORT_CMD_NEW) < 0)
2193 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2194 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2195 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2196 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2197 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2198 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2199 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2202 static const struct genl_ops dp_vport_genl_ops[] = {
2203 { .cmd = OVS_VPORT_CMD_NEW,
2204 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2205 .policy = vport_policy,
2206 .doit = ovs_vport_cmd_new
2208 { .cmd = OVS_VPORT_CMD_DEL,
2209 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2210 .policy = vport_policy,
2211 .doit = ovs_vport_cmd_del
2213 { .cmd = OVS_VPORT_CMD_GET,
2214 .flags = 0, /* OK for unprivileged users. */
2215 .policy = vport_policy,
2216 .doit = ovs_vport_cmd_get,
2217 .dumpit = ovs_vport_cmd_dump
2219 { .cmd = OVS_VPORT_CMD_SET,
2220 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2221 .policy = vport_policy,
2222 .doit = ovs_vport_cmd_set,
2226 struct genl_family dp_vport_genl_family = {
2227 .id = GENL_ID_GENERATE,
2228 .hdrsize = sizeof(struct ovs_header),
2229 .name = OVS_VPORT_FAMILY,
2230 .version = OVS_VPORT_VERSION,
2231 .maxattr = OVS_VPORT_ATTR_MAX,
2233 .parallel_ops = true,
2234 .ops = dp_vport_genl_ops,
2235 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2236 .mcgrps = &ovs_dp_vport_multicast_group,
2240 static struct genl_family * const dp_genl_families[] = {
2241 &dp_datapath_genl_family,
2242 &dp_vport_genl_family,
2243 &dp_flow_genl_family,
2244 &dp_packet_genl_family,
2247 static void dp_unregister_genl(int n_families)
2251 for (i = 0; i < n_families; i++)
2252 genl_unregister_family(dp_genl_families[i]);
2255 static int dp_register_genl(void)
2260 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2262 err = genl_register_family(dp_genl_families[i]);
2270 dp_unregister_genl(i);
2274 static int __net_init ovs_init_net(struct net *net)
2276 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2278 INIT_LIST_HEAD(&ovs_net->dps);
2279 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2284 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2285 struct list_head *head)
2287 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2288 struct datapath *dp;
2290 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2293 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2294 struct vport *vport;
2296 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2297 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2300 if (dev_net(vport->dev) == dnet)
2301 list_add(&vport->detach_list, head);
2307 static void __net_exit ovs_exit_net(struct net *dnet)
2309 struct datapath *dp, *dp_next;
2310 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2311 struct vport *vport, *vport_next;
2317 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2322 list_vports_from_net(net, dnet, &head);
2325 /* Detach all vports from given namespace. */
2326 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2327 list_del(&vport->detach_list);
2328 ovs_dp_detach_port(vport);
2333 cancel_work_sync(&ovs_net->dp_notify_work);
2336 static struct pernet_operations ovs_net_ops = {
2337 .init = ovs_init_net,
2338 .exit = ovs_exit_net,
2340 .size = sizeof(struct ovs_net),
2343 static int __init dp_init(void)
2347 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2349 pr_info("Open vSwitch switching datapath\n");
2351 err = action_fifos_init();
2355 err = ovs_internal_dev_rtnl_link_register();
2357 goto error_action_fifos_exit;
2359 err = ovs_flow_init();
2361 goto error_unreg_rtnl_link;
2363 err = ovs_vport_init();
2365 goto error_flow_exit;
2367 err = register_pernet_device(&ovs_net_ops);
2369 goto error_vport_exit;
2371 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2373 goto error_netns_exit;
2375 err = ovs_netdev_init();
2377 goto error_unreg_notifier;
2379 err = dp_register_genl();
2381 goto error_unreg_netdev;
2387 error_unreg_notifier:
2388 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2390 unregister_pernet_device(&ovs_net_ops);
2395 error_unreg_rtnl_link:
2396 ovs_internal_dev_rtnl_link_unregister();
2397 error_action_fifos_exit:
2398 action_fifos_exit();
2403 static void dp_cleanup(void)
2405 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2407 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2408 unregister_pernet_device(&ovs_net_ops);
2412 ovs_internal_dev_rtnl_link_unregister();
2413 action_fifos_exit();
2416 module_init(dp_init);
2417 module_exit(dp_cleanup);
2419 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2420 MODULE_LICENSE("GPL");