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"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
62 unsigned int ovs_net_id __read_mostly;
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);
136 static struct vport *new_vport(const struct vport_parms *);
137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
138 const struct sw_flow_key *,
139 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 *,
146 /* Must be called with rcu_read_lock or ovs_mutex. */
147 const char *ovs_dp_name(const struct datapath *dp)
149 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
150 return ovs_vport_name(vport);
153 static int get_dpifindex(const struct datapath *dp)
160 local = ovs_vport_rcu(dp, OVSP_LOCAL);
162 ifindex = local->dev->ifindex;
171 static void destroy_dp_rcu(struct rcu_head *rcu)
173 struct datapath *dp = container_of(rcu, struct datapath, rcu);
175 ovs_flow_tbl_destroy(&dp->table);
176 free_percpu(dp->stats_percpu);
182 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
185 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
188 /* Called with ovs_mutex or RCU read lock. */
189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
192 struct hlist_head *head;
194 head = vport_hash_bucket(dp, port_no);
195 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
196 if (vport->port_no == port_no)
202 /* Called with ovs_mutex. */
203 static struct vport *new_vport(const struct vport_parms *parms)
207 vport = ovs_vport_add(parms);
208 if (!IS_ERR(vport)) {
209 struct datapath *dp = parms->dp;
210 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
212 hlist_add_head_rcu(&vport->dp_hash_node, head);
217 void ovs_dp_detach_port(struct vport *p)
221 /* First drop references to device. */
222 hlist_del_rcu(&p->dp_hash_node);
224 /* Then destroy it. */
228 /* Must be called with rcu_read_lock. */
229 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
231 const struct vport *p = OVS_CB(skb)->input_vport;
232 struct datapath *dp = p->dp;
233 struct sw_flow *flow;
234 struct sw_flow_actions *sf_acts;
235 struct dp_stats_percpu *stats;
239 stats = this_cpu_ptr(dp->stats_percpu);
242 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
243 if (unlikely(!flow)) {
244 struct dp_upcall_info upcall;
247 memset(&upcall, 0, sizeof(upcall));
248 upcall.cmd = OVS_PACKET_CMD_MISS;
249 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
250 upcall.mru = OVS_CB(skb)->mru;
251 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
256 stats_counter = &stats->n_missed;
260 ovs_flow_stats_update(flow, key->tp.flags, skb);
261 sf_acts = rcu_dereference(flow->sf_acts);
262 ovs_execute_actions(dp, skb, sf_acts, key);
264 stats_counter = &stats->n_hit;
267 /* Update datapath statistics. */
268 u64_stats_update_begin(&stats->syncp);
270 stats->n_mask_hit += n_mask_hit;
271 u64_stats_update_end(&stats->syncp);
274 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
275 const struct sw_flow_key *key,
276 const struct dp_upcall_info *upcall_info,
279 struct dp_stats_percpu *stats;
282 if (upcall_info->portid == 0) {
287 if (!skb_is_gso(skb))
288 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
290 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
297 stats = this_cpu_ptr(dp->stats_percpu);
299 u64_stats_update_begin(&stats->syncp);
301 u64_stats_update_end(&stats->syncp);
306 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
307 const struct sw_flow_key *key,
308 const struct dp_upcall_info *upcall_info,
311 unsigned int gso_type = skb_shinfo(skb)->gso_type;
312 struct sw_flow_key later_key;
313 struct sk_buff *segs, *nskb;
316 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
317 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
319 return PTR_ERR(segs);
323 if (gso_type & SKB_GSO_UDP) {
324 /* The initial flow key extracted by ovs_flow_key_extract()
325 * in this case is for a first fragment, so we need to
326 * properly mark later fragments.
329 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
332 /* Queue all of the segments. */
335 if (gso_type & SKB_GSO_UDP && skb != segs)
338 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
342 } while ((skb = skb->next));
344 /* Free all of the segments. */
352 } while ((skb = nskb));
356 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
357 unsigned int hdrlen, int actions_attrlen)
359 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
360 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
361 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
362 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
364 /* OVS_PACKET_ATTR_USERDATA */
365 if (upcall_info->userdata)
366 size += NLA_ALIGN(upcall_info->userdata->nla_len);
368 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
369 if (upcall_info->egress_tun_info)
370 size += nla_total_size(ovs_tun_key_attr_size());
372 /* OVS_PACKET_ATTR_ACTIONS */
373 if (upcall_info->actions_len)
374 size += nla_total_size(actions_attrlen);
376 /* OVS_PACKET_ATTR_MRU */
377 if (upcall_info->mru)
378 size += nla_total_size(sizeof(upcall_info->mru));
383 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
385 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
386 size_t plen = NLA_ALIGN(skb->len) - skb->len;
389 skb_put_zero(skb, plen);
393 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
394 const struct sw_flow_key *key,
395 const struct dp_upcall_info *upcall_info,
398 struct ovs_header *upcall;
399 struct sk_buff *nskb = NULL;
400 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
406 dp_ifindex = get_dpifindex(dp);
410 if (skb_vlan_tag_present(skb)) {
411 nskb = skb_clone(skb, GFP_ATOMIC);
415 nskb = __vlan_hwaccel_push_inside(nskb);
422 if (nla_attr_size(skb->len) > USHRT_MAX) {
427 /* Complete checksum if needed */
428 if (skb->ip_summed == CHECKSUM_PARTIAL &&
429 (err = skb_csum_hwoffload_help(skb, 0)))
432 /* Older versions of OVS user space enforce alignment of the last
433 * Netlink attribute to NLA_ALIGNTO which would require extensive
434 * padding logic. Only perform zerocopy if padding is not required.
436 if (dp->user_features & OVS_DP_F_UNALIGNED)
437 hlen = skb_zerocopy_headlen(skb);
441 len = upcall_msg_size(upcall_info, hlen - cutlen,
442 OVS_CB(skb)->acts_origlen);
443 user_skb = genlmsg_new(len, GFP_ATOMIC);
449 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
450 0, upcall_info->cmd);
451 upcall->dp_ifindex = dp_ifindex;
453 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
456 if (upcall_info->userdata)
457 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
458 nla_len(upcall_info->userdata),
459 nla_data(upcall_info->userdata));
461 if (upcall_info->egress_tun_info) {
462 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
463 err = ovs_nla_put_tunnel_info(user_skb,
464 upcall_info->egress_tun_info);
466 nla_nest_end(user_skb, nla);
469 if (upcall_info->actions_len) {
470 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
471 err = ovs_nla_put_actions(upcall_info->actions,
472 upcall_info->actions_len,
475 nla_nest_end(user_skb, nla);
477 nla_nest_cancel(user_skb, nla);
480 /* Add OVS_PACKET_ATTR_MRU */
481 if (upcall_info->mru) {
482 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
487 pad_packet(dp, user_skb);
490 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
492 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
497 pad_packet(dp, user_skb);
500 /* Only reserve room for attribute header, packet data is added
501 * in skb_zerocopy() */
502 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
506 nla->nla_len = nla_attr_size(skb->len - cutlen);
508 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
512 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
513 pad_packet(dp, user_skb);
515 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
517 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
527 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
529 struct ovs_header *ovs_header = info->userhdr;
530 struct net *net = sock_net(skb->sk);
531 struct nlattr **a = info->attrs;
532 struct sw_flow_actions *acts;
533 struct sk_buff *packet;
534 struct sw_flow *flow;
535 struct sw_flow_actions *sf_acts;
537 struct vport *input_vport;
541 bool log = !a[OVS_PACKET_ATTR_PROBE];
544 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
545 !a[OVS_PACKET_ATTR_ACTIONS])
548 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
549 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
553 skb_reserve(packet, NET_IP_ALIGN);
555 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
557 /* Set packet's mru */
558 if (a[OVS_PACKET_ATTR_MRU]) {
559 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
560 packet->ignore_df = 1;
562 OVS_CB(packet)->mru = mru;
564 /* Build an sw_flow for sending this packet. */
565 flow = ovs_flow_alloc();
570 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
571 packet, &flow->key, log);
575 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
576 &flow->key, &acts, log);
580 rcu_assign_pointer(flow->sf_acts, acts);
581 packet->priority = flow->key.phy.priority;
582 packet->mark = flow->key.phy.skb_mark;
585 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
590 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
592 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
597 packet->dev = input_vport->dev;
598 OVS_CB(packet)->input_vport = input_vport;
599 sf_acts = rcu_dereference(flow->sf_acts);
602 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
606 ovs_flow_free(flow, false);
612 ovs_flow_free(flow, false);
619 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
620 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
621 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
622 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
623 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
624 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
627 static const struct genl_ops dp_packet_genl_ops[] = {
628 { .cmd = OVS_PACKET_CMD_EXECUTE,
629 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
630 .policy = packet_policy,
631 .doit = ovs_packet_cmd_execute
635 static struct genl_family dp_packet_genl_family __ro_after_init = {
636 .hdrsize = sizeof(struct ovs_header),
637 .name = OVS_PACKET_FAMILY,
638 .version = OVS_PACKET_VERSION,
639 .maxattr = OVS_PACKET_ATTR_MAX,
641 .parallel_ops = true,
642 .ops = dp_packet_genl_ops,
643 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
644 .module = THIS_MODULE,
647 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
648 struct ovs_dp_megaflow_stats *mega_stats)
652 memset(mega_stats, 0, sizeof(*mega_stats));
654 stats->n_flows = ovs_flow_tbl_count(&dp->table);
655 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
657 stats->n_hit = stats->n_missed = stats->n_lost = 0;
659 for_each_possible_cpu(i) {
660 const struct dp_stats_percpu *percpu_stats;
661 struct dp_stats_percpu local_stats;
664 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
667 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
668 local_stats = *percpu_stats;
669 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
671 stats->n_hit += local_stats.n_hit;
672 stats->n_missed += local_stats.n_missed;
673 stats->n_lost += local_stats.n_lost;
674 mega_stats->n_mask_hit += local_stats.n_mask_hit;
678 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
680 return ovs_identifier_is_ufid(sfid) &&
681 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
684 static bool should_fill_mask(uint32_t ufid_flags)
686 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
689 static bool should_fill_actions(uint32_t ufid_flags)
691 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
694 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
695 const struct sw_flow_id *sfid,
698 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
700 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
701 * see ovs_nla_put_identifier()
703 if (sfid && ovs_identifier_is_ufid(sfid))
704 len += nla_total_size(sfid->ufid_len);
706 len += nla_total_size(ovs_key_attr_size());
708 /* OVS_FLOW_ATTR_KEY */
709 if (!sfid || should_fill_key(sfid, ufid_flags))
710 len += nla_total_size(ovs_key_attr_size());
712 /* OVS_FLOW_ATTR_MASK */
713 if (should_fill_mask(ufid_flags))
714 len += nla_total_size(ovs_key_attr_size());
716 /* OVS_FLOW_ATTR_ACTIONS */
717 if (should_fill_actions(ufid_flags))
718 len += nla_total_size(acts->orig_len);
721 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
722 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
723 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
726 /* Called with ovs_mutex or RCU read lock. */
727 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
730 struct ovs_flow_stats stats;
734 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
737 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
741 if (stats.n_packets &&
742 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
743 sizeof(struct ovs_flow_stats), &stats,
747 if ((u8)ntohs(tcp_flags) &&
748 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
754 /* Called with ovs_mutex or RCU read lock. */
755 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
756 struct sk_buff *skb, int skb_orig_len)
758 struct nlattr *start;
761 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
762 * this is the first flow to be dumped into 'skb'. This is unusual for
763 * Netlink but individual action lists can be longer than
764 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
765 * The userspace caller can always fetch the actions separately if it
766 * really wants them. (Most userspace callers in fact don't care.)
768 * This can only fail for dump operations because the skb is always
769 * properly sized for single flows.
771 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
773 const struct sw_flow_actions *sf_acts;
775 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
776 err = ovs_nla_put_actions(sf_acts->actions,
777 sf_acts->actions_len, skb);
780 nla_nest_end(skb, start);
785 nla_nest_cancel(skb, start);
787 } else if (skb_orig_len) {
794 /* Called with ovs_mutex or RCU read lock. */
795 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
796 struct sk_buff *skb, u32 portid,
797 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
799 const int skb_orig_len = skb->len;
800 struct ovs_header *ovs_header;
803 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
808 ovs_header->dp_ifindex = dp_ifindex;
810 err = ovs_nla_put_identifier(flow, skb);
814 if (should_fill_key(&flow->id, ufid_flags)) {
815 err = ovs_nla_put_masked_key(flow, skb);
820 if (should_fill_mask(ufid_flags)) {
821 err = ovs_nla_put_mask(flow, skb);
826 err = ovs_flow_cmd_fill_stats(flow, skb);
830 if (should_fill_actions(ufid_flags)) {
831 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
836 genlmsg_end(skb, ovs_header);
840 genlmsg_cancel(skb, ovs_header);
844 /* May not be called with RCU read lock. */
845 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
846 const struct sw_flow_id *sfid,
847 struct genl_info *info,
854 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
857 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
858 skb = genlmsg_new(len, GFP_KERNEL);
860 return ERR_PTR(-ENOMEM);
865 /* Called with ovs_mutex. */
866 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
868 struct genl_info *info, u8 cmd,
869 bool always, u32 ufid_flags)
874 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
875 &flow->id, info, always, ufid_flags);
876 if (IS_ERR_OR_NULL(skb))
879 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
880 info->snd_portid, info->snd_seq, 0,
882 if (WARN_ON_ONCE(retval < 0)) {
884 skb = ERR_PTR(retval);
889 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
891 struct net *net = sock_net(skb->sk);
892 struct nlattr **a = info->attrs;
893 struct ovs_header *ovs_header = info->userhdr;
894 struct sw_flow *flow = NULL, *new_flow;
895 struct sw_flow_mask mask;
896 struct sk_buff *reply;
898 struct sw_flow_actions *acts;
899 struct sw_flow_match match;
900 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
902 bool log = !a[OVS_FLOW_ATTR_PROBE];
904 /* Must have key and actions. */
906 if (!a[OVS_FLOW_ATTR_KEY]) {
907 OVS_NLERR(log, "Flow key attr not present in new flow.");
910 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
911 OVS_NLERR(log, "Flow actions attr not present in new flow.");
915 /* Most of the time we need to allocate a new flow, do it before
918 new_flow = ovs_flow_alloc();
919 if (IS_ERR(new_flow)) {
920 error = PTR_ERR(new_flow);
925 ovs_match_init(&match, &new_flow->key, false, &mask);
926 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
927 a[OVS_FLOW_ATTR_MASK], log);
931 /* Extract flow identifier. */
932 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
933 &new_flow->key, log);
937 /* unmasked key is needed to match when ufid is not used. */
938 if (ovs_identifier_is_key(&new_flow->id))
939 match.key = new_flow->id.unmasked_key;
941 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
943 /* Validate actions. */
944 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
945 &new_flow->key, &acts, log);
947 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
951 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
954 error = PTR_ERR(reply);
959 dp = get_dp(net, ovs_header->dp_ifindex);
965 /* Check if this is a duplicate flow */
966 if (ovs_identifier_is_ufid(&new_flow->id))
967 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
969 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
971 rcu_assign_pointer(new_flow->sf_acts, acts);
973 /* Put flow in bucket. */
974 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
975 if (unlikely(error)) {
980 if (unlikely(reply)) {
981 error = ovs_flow_cmd_fill_info(new_flow,
982 ovs_header->dp_ifindex,
983 reply, info->snd_portid,
991 struct sw_flow_actions *old_acts;
993 /* Bail out if we're not allowed to modify an existing flow.
994 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
995 * because Generic Netlink treats the latter as a dump
996 * request. We also accept NLM_F_EXCL in case that bug ever
999 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1002 goto err_unlock_ovs;
1004 /* The flow identifier has to be the same for flow updates.
1005 * Look for any overlapping flow.
1007 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1008 if (ovs_identifier_is_key(&flow->id))
1009 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1011 else /* UFID matches but key is different */
1015 goto err_unlock_ovs;
1018 /* Update actions. */
1019 old_acts = ovsl_dereference(flow->sf_acts);
1020 rcu_assign_pointer(flow->sf_acts, acts);
1022 if (unlikely(reply)) {
1023 error = ovs_flow_cmd_fill_info(flow,
1024 ovs_header->dp_ifindex,
1025 reply, info->snd_portid,
1033 ovs_nla_free_flow_actions_rcu(old_acts);
1034 ovs_flow_free(new_flow, false);
1038 ovs_notify(&dp_flow_genl_family, reply, info);
1045 ovs_nla_free_flow_actions(acts);
1047 ovs_flow_free(new_flow, false);
1052 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1053 static struct sw_flow_actions *get_flow_actions(struct net *net,
1054 const struct nlattr *a,
1055 const struct sw_flow_key *key,
1056 const struct sw_flow_mask *mask,
1059 struct sw_flow_actions *acts;
1060 struct sw_flow_key masked_key;
1063 ovs_flow_mask_key(&masked_key, key, true, mask);
1064 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1067 "Actions may not be safe on all matching packets");
1068 return ERR_PTR(error);
1074 /* Factor out match-init and action-copy to avoid
1075 * "Wframe-larger-than=1024" warning. Because mask is only
1076 * used to get actions, we new a function to save some
1079 * If there are not key and action attrs, we return 0
1080 * directly. In the case, the caller will also not use the
1081 * match as before. If there is action attr, we try to get
1082 * actions and save them to *acts. Before returning from
1083 * the function, we reset the match->mask pointer. Because
1084 * we should not to return match object with dangling reference
1087 static int ovs_nla_init_match_and_action(struct net *net,
1088 struct sw_flow_match *match,
1089 struct sw_flow_key *key,
1091 struct sw_flow_actions **acts,
1094 struct sw_flow_mask mask;
1097 if (a[OVS_FLOW_ATTR_KEY]) {
1098 ovs_match_init(match, key, true, &mask);
1099 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1100 a[OVS_FLOW_ATTR_MASK], log);
1105 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1106 if (!a[OVS_FLOW_ATTR_KEY]) {
1108 "Flow key attribute not present in set flow.");
1113 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1115 if (IS_ERR(*acts)) {
1116 error = PTR_ERR(*acts);
1121 /* On success, error is 0. */
1127 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1129 struct net *net = sock_net(skb->sk);
1130 struct nlattr **a = info->attrs;
1131 struct ovs_header *ovs_header = info->userhdr;
1132 struct sw_flow_key key;
1133 struct sw_flow *flow;
1134 struct sk_buff *reply = NULL;
1135 struct datapath *dp;
1136 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1137 struct sw_flow_match match;
1138 struct sw_flow_id sfid;
1139 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1141 bool log = !a[OVS_FLOW_ATTR_PROBE];
1144 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1145 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1147 "Flow set message rejected, Key attribute missing.");
1151 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1157 /* Can allocate before locking if have acts. */
1158 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1160 if (IS_ERR(reply)) {
1161 error = PTR_ERR(reply);
1162 goto err_kfree_acts;
1167 dp = get_dp(net, ovs_header->dp_ifindex);
1168 if (unlikely(!dp)) {
1170 goto err_unlock_ovs;
1172 /* Check that the flow exists. */
1174 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1176 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1177 if (unlikely(!flow)) {
1179 goto err_unlock_ovs;
1182 /* Update actions, if present. */
1184 old_acts = ovsl_dereference(flow->sf_acts);
1185 rcu_assign_pointer(flow->sf_acts, acts);
1187 if (unlikely(reply)) {
1188 error = ovs_flow_cmd_fill_info(flow,
1189 ovs_header->dp_ifindex,
1190 reply, info->snd_portid,
1197 /* Could not alloc without acts before locking. */
1198 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1199 info, OVS_FLOW_CMD_NEW, false,
1202 if (IS_ERR(reply)) {
1203 error = PTR_ERR(reply);
1204 goto err_unlock_ovs;
1209 if (a[OVS_FLOW_ATTR_CLEAR])
1210 ovs_flow_stats_clear(flow);
1214 ovs_notify(&dp_flow_genl_family, reply, info);
1216 ovs_nla_free_flow_actions_rcu(old_acts);
1224 ovs_nla_free_flow_actions(acts);
1229 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1231 struct nlattr **a = info->attrs;
1232 struct ovs_header *ovs_header = info->userhdr;
1233 struct net *net = sock_net(skb->sk);
1234 struct sw_flow_key key;
1235 struct sk_buff *reply;
1236 struct sw_flow *flow;
1237 struct datapath *dp;
1238 struct sw_flow_match match;
1239 struct sw_flow_id ufid;
1240 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1242 bool log = !a[OVS_FLOW_ATTR_PROBE];
1245 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1246 if (a[OVS_FLOW_ATTR_KEY]) {
1247 ovs_match_init(&match, &key, true, NULL);
1248 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1250 } else if (!ufid_present) {
1252 "Flow get message rejected, Key attribute missing.");
1259 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1266 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1268 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1274 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1275 OVS_FLOW_CMD_NEW, true, ufid_flags);
1276 if (IS_ERR(reply)) {
1277 err = PTR_ERR(reply);
1282 return genlmsg_reply(reply, info);
1288 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1290 struct nlattr **a = info->attrs;
1291 struct ovs_header *ovs_header = info->userhdr;
1292 struct net *net = sock_net(skb->sk);
1293 struct sw_flow_key key;
1294 struct sk_buff *reply;
1295 struct sw_flow *flow = NULL;
1296 struct datapath *dp;
1297 struct sw_flow_match match;
1298 struct sw_flow_id ufid;
1299 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1301 bool log = !a[OVS_FLOW_ATTR_PROBE];
1304 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1305 if (a[OVS_FLOW_ATTR_KEY]) {
1306 ovs_match_init(&match, &key, true, NULL);
1307 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1314 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1315 if (unlikely(!dp)) {
1320 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1321 err = ovs_flow_tbl_flush(&dp->table);
1326 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1328 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1329 if (unlikely(!flow)) {
1334 ovs_flow_tbl_remove(&dp->table, flow);
1337 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1338 &flow->id, info, false, ufid_flags);
1339 if (likely(reply)) {
1340 if (likely(!IS_ERR(reply))) {
1341 rcu_read_lock(); /*To keep RCU checker happy. */
1342 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1343 reply, info->snd_portid,
1348 if (WARN_ON_ONCE(err < 0)) {
1353 ovs_notify(&dp_flow_genl_family, reply, info);
1355 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1360 ovs_flow_free(flow, true);
1367 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1369 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1370 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1371 struct table_instance *ti;
1372 struct datapath *dp;
1376 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1377 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1380 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1383 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1389 ti = rcu_dereference(dp->table.ti);
1391 struct sw_flow *flow;
1394 bucket = cb->args[0];
1396 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1400 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1401 NETLINK_CB(cb->skb).portid,
1402 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1403 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1406 cb->args[0] = bucket;
1413 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1414 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1415 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1416 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1417 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1418 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1419 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1420 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1423 static const struct genl_ops dp_flow_genl_ops[] = {
1424 { .cmd = OVS_FLOW_CMD_NEW,
1425 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1426 .policy = flow_policy,
1427 .doit = ovs_flow_cmd_new
1429 { .cmd = OVS_FLOW_CMD_DEL,
1430 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1431 .policy = flow_policy,
1432 .doit = ovs_flow_cmd_del
1434 { .cmd = OVS_FLOW_CMD_GET,
1435 .flags = 0, /* OK for unprivileged users. */
1436 .policy = flow_policy,
1437 .doit = ovs_flow_cmd_get,
1438 .dumpit = ovs_flow_cmd_dump
1440 { .cmd = OVS_FLOW_CMD_SET,
1441 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1442 .policy = flow_policy,
1443 .doit = ovs_flow_cmd_set,
1447 static struct genl_family dp_flow_genl_family __ro_after_init = {
1448 .hdrsize = sizeof(struct ovs_header),
1449 .name = OVS_FLOW_FAMILY,
1450 .version = OVS_FLOW_VERSION,
1451 .maxattr = OVS_FLOW_ATTR_MAX,
1453 .parallel_ops = true,
1454 .ops = dp_flow_genl_ops,
1455 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1456 .mcgrps = &ovs_dp_flow_multicast_group,
1458 .module = THIS_MODULE,
1461 static size_t ovs_dp_cmd_msg_size(void)
1463 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1465 msgsize += nla_total_size(IFNAMSIZ);
1466 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1467 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1468 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1473 /* Called with ovs_mutex. */
1474 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1475 u32 portid, u32 seq, u32 flags, u8 cmd)
1477 struct ovs_header *ovs_header;
1478 struct ovs_dp_stats dp_stats;
1479 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1482 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1487 ovs_header->dp_ifindex = get_dpifindex(dp);
1489 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1491 goto nla_put_failure;
1493 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1494 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1495 &dp_stats, OVS_DP_ATTR_PAD))
1496 goto nla_put_failure;
1498 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1499 sizeof(struct ovs_dp_megaflow_stats),
1500 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1501 goto nla_put_failure;
1503 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1504 goto nla_put_failure;
1506 genlmsg_end(skb, ovs_header);
1510 genlmsg_cancel(skb, ovs_header);
1515 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1517 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1520 /* Called with rcu_read_lock or ovs_mutex. */
1521 static struct datapath *lookup_datapath(struct net *net,
1522 const struct ovs_header *ovs_header,
1523 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1525 struct datapath *dp;
1527 if (!a[OVS_DP_ATTR_NAME])
1528 dp = get_dp(net, ovs_header->dp_ifindex);
1530 struct vport *vport;
1532 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1533 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1535 return dp ? dp : ERR_PTR(-ENODEV);
1538 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1540 struct datapath *dp;
1542 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1546 WARN(dp->user_features, "Dropping previously announced user features\n");
1547 dp->user_features = 0;
1550 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1552 if (a[OVS_DP_ATTR_USER_FEATURES])
1553 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1556 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1558 struct nlattr **a = info->attrs;
1559 struct vport_parms parms;
1560 struct sk_buff *reply;
1561 struct datapath *dp;
1562 struct vport *vport;
1563 struct ovs_net *ovs_net;
1567 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1570 reply = ovs_dp_cmd_alloc_info();
1575 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1577 goto err_free_reply;
1579 ovs_dp_set_net(dp, sock_net(skb->sk));
1581 /* Allocate table. */
1582 err = ovs_flow_tbl_init(&dp->table);
1586 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1587 if (!dp->stats_percpu) {
1589 goto err_destroy_table;
1592 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1593 sizeof(struct hlist_head),
1597 goto err_destroy_percpu;
1600 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1601 INIT_HLIST_HEAD(&dp->ports[i]);
1603 err = ovs_meters_init(dp);
1605 goto err_destroy_ports_array;
1607 /* Set up our datapath device. */
1608 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1609 parms.type = OVS_VPORT_TYPE_INTERNAL;
1610 parms.options = NULL;
1612 parms.port_no = OVSP_LOCAL;
1613 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1615 ovs_dp_change(dp, a);
1617 /* So far only local changes have been made, now need the lock. */
1620 vport = new_vport(&parms);
1621 if (IS_ERR(vport)) {
1622 err = PTR_ERR(vport);
1626 if (err == -EEXIST) {
1627 /* An outdated user space instance that does not understand
1628 * the concept of user_features has attempted to create a new
1629 * datapath and is likely to reuse it. Drop all user features.
1631 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1632 ovs_dp_reset_user_features(skb, info);
1635 goto err_destroy_meters;
1638 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1639 info->snd_seq, 0, OVS_DP_CMD_NEW);
1642 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1643 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1647 ovs_notify(&dp_datapath_genl_family, reply, info);
1652 ovs_meters_exit(dp);
1653 err_destroy_ports_array:
1656 free_percpu(dp->stats_percpu);
1658 ovs_flow_tbl_destroy(&dp->table);
1667 /* Called with ovs_mutex. */
1668 static void __dp_destroy(struct datapath *dp)
1672 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1673 struct vport *vport;
1674 struct hlist_node *n;
1676 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1677 if (vport->port_no != OVSP_LOCAL)
1678 ovs_dp_detach_port(vport);
1681 list_del_rcu(&dp->list_node);
1683 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1684 * all ports in datapath are destroyed first before freeing datapath.
1686 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1688 /* RCU destroy the flow table */
1689 call_rcu(&dp->rcu, destroy_dp_rcu);
1692 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1694 struct sk_buff *reply;
1695 struct datapath *dp;
1698 reply = ovs_dp_cmd_alloc_info();
1703 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1706 goto err_unlock_free;
1708 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1709 info->snd_seq, 0, OVS_DP_CMD_DEL);
1715 ovs_notify(&dp_datapath_genl_family, reply, info);
1725 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1727 struct sk_buff *reply;
1728 struct datapath *dp;
1731 reply = ovs_dp_cmd_alloc_info();
1736 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1739 goto err_unlock_free;
1741 ovs_dp_change(dp, info->attrs);
1743 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1744 info->snd_seq, 0, OVS_DP_CMD_NEW);
1748 ovs_notify(&dp_datapath_genl_family, reply, info);
1758 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1760 struct sk_buff *reply;
1761 struct datapath *dp;
1764 reply = ovs_dp_cmd_alloc_info();
1769 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1772 goto err_unlock_free;
1774 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1775 info->snd_seq, 0, OVS_DP_CMD_NEW);
1779 return genlmsg_reply(reply, info);
1787 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1789 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1790 struct datapath *dp;
1791 int skip = cb->args[0];
1795 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1797 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1798 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1799 OVS_DP_CMD_NEW) < 0)
1810 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1811 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1812 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1813 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1816 static const struct genl_ops dp_datapath_genl_ops[] = {
1817 { .cmd = OVS_DP_CMD_NEW,
1818 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1819 .policy = datapath_policy,
1820 .doit = ovs_dp_cmd_new
1822 { .cmd = OVS_DP_CMD_DEL,
1823 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1824 .policy = datapath_policy,
1825 .doit = ovs_dp_cmd_del
1827 { .cmd = OVS_DP_CMD_GET,
1828 .flags = 0, /* OK for unprivileged users. */
1829 .policy = datapath_policy,
1830 .doit = ovs_dp_cmd_get,
1831 .dumpit = ovs_dp_cmd_dump
1833 { .cmd = OVS_DP_CMD_SET,
1834 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1835 .policy = datapath_policy,
1836 .doit = ovs_dp_cmd_set,
1840 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1841 .hdrsize = sizeof(struct ovs_header),
1842 .name = OVS_DATAPATH_FAMILY,
1843 .version = OVS_DATAPATH_VERSION,
1844 .maxattr = OVS_DP_ATTR_MAX,
1846 .parallel_ops = true,
1847 .ops = dp_datapath_genl_ops,
1848 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1849 .mcgrps = &ovs_dp_datapath_multicast_group,
1851 .module = THIS_MODULE,
1854 /* Called with ovs_mutex or RCU read lock. */
1855 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1856 struct net *net, u32 portid, u32 seq,
1857 u32 flags, u8 cmd, gfp_t gfp)
1859 struct ovs_header *ovs_header;
1860 struct ovs_vport_stats vport_stats;
1863 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1868 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1870 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1871 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1872 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1873 ovs_vport_name(vport)) ||
1874 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1875 goto nla_put_failure;
1877 if (!net_eq(net, dev_net(vport->dev))) {
1878 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1880 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1881 goto nla_put_failure;
1884 ovs_vport_get_stats(vport, &vport_stats);
1885 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1886 sizeof(struct ovs_vport_stats), &vport_stats,
1887 OVS_VPORT_ATTR_PAD))
1888 goto nla_put_failure;
1890 if (ovs_vport_get_upcall_portids(vport, skb))
1891 goto nla_put_failure;
1893 err = ovs_vport_get_options(vport, skb);
1894 if (err == -EMSGSIZE)
1897 genlmsg_end(skb, ovs_header);
1903 genlmsg_cancel(skb, ovs_header);
1907 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1909 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1912 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1913 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1914 u32 portid, u32 seq, u8 cmd)
1916 struct sk_buff *skb;
1919 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1921 return ERR_PTR(-ENOMEM);
1923 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1930 /* Called with ovs_mutex or RCU read lock. */
1931 static struct vport *lookup_vport(struct net *net,
1932 const struct ovs_header *ovs_header,
1933 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1935 struct datapath *dp;
1936 struct vport *vport;
1938 if (a[OVS_VPORT_ATTR_IFINDEX])
1939 return ERR_PTR(-EOPNOTSUPP);
1940 if (a[OVS_VPORT_ATTR_NAME]) {
1941 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1943 return ERR_PTR(-ENODEV);
1944 if (ovs_header->dp_ifindex &&
1945 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1946 return ERR_PTR(-ENODEV);
1948 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1949 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1951 if (port_no >= DP_MAX_PORTS)
1952 return ERR_PTR(-EFBIG);
1954 dp = get_dp(net, ovs_header->dp_ifindex);
1956 return ERR_PTR(-ENODEV);
1958 vport = ovs_vport_ovsl_rcu(dp, port_no);
1960 return ERR_PTR(-ENODEV);
1963 return ERR_PTR(-EINVAL);
1967 /* Called with ovs_mutex */
1968 static void update_headroom(struct datapath *dp)
1970 unsigned dev_headroom, max_headroom = 0;
1971 struct net_device *dev;
1972 struct vport *vport;
1975 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1976 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1978 dev_headroom = netdev_get_fwd_headroom(dev);
1979 if (dev_headroom > max_headroom)
1980 max_headroom = dev_headroom;
1984 dp->max_headroom = max_headroom;
1985 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1986 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1987 netdev_set_rx_headroom(vport->dev, max_headroom);
1990 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1992 struct nlattr **a = info->attrs;
1993 struct ovs_header *ovs_header = info->userhdr;
1994 struct vport_parms parms;
1995 struct sk_buff *reply;
1996 struct vport *vport;
1997 struct datapath *dp;
2001 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2002 !a[OVS_VPORT_ATTR_UPCALL_PID])
2004 if (a[OVS_VPORT_ATTR_IFINDEX])
2007 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2008 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2009 if (port_no >= DP_MAX_PORTS)
2012 reply = ovs_vport_cmd_alloc_info();
2018 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2021 goto exit_unlock_free;
2024 vport = ovs_vport_ovsl(dp, port_no);
2027 goto exit_unlock_free;
2029 for (port_no = 1; ; port_no++) {
2030 if (port_no >= DP_MAX_PORTS) {
2032 goto exit_unlock_free;
2034 vport = ovs_vport_ovsl(dp, port_no);
2040 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2041 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2042 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2044 parms.port_no = port_no;
2045 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2047 vport = new_vport(&parms);
2048 err = PTR_ERR(vport);
2049 if (IS_ERR(vport)) {
2052 goto exit_unlock_free;
2055 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2056 info->snd_portid, info->snd_seq, 0,
2057 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2059 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2060 update_headroom(dp);
2062 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2067 ovs_notify(&dp_vport_genl_family, reply, info);
2076 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2078 struct nlattr **a = info->attrs;
2079 struct sk_buff *reply;
2080 struct vport *vport;
2083 reply = ovs_vport_cmd_alloc_info();
2088 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2089 err = PTR_ERR(vport);
2091 goto exit_unlock_free;
2093 if (a[OVS_VPORT_ATTR_TYPE] &&
2094 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2096 goto exit_unlock_free;
2099 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2100 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2102 goto exit_unlock_free;
2106 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2107 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2109 err = ovs_vport_set_upcall_portids(vport, ids);
2111 goto exit_unlock_free;
2114 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2115 info->snd_portid, info->snd_seq, 0,
2116 OVS_VPORT_CMD_NEW, GFP_ATOMIC);
2120 ovs_notify(&dp_vport_genl_family, reply, info);
2129 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2131 bool must_update_headroom = false;
2132 struct nlattr **a = info->attrs;
2133 struct sk_buff *reply;
2134 struct datapath *dp;
2135 struct vport *vport;
2138 reply = ovs_vport_cmd_alloc_info();
2143 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2144 err = PTR_ERR(vport);
2146 goto exit_unlock_free;
2148 if (vport->port_no == OVSP_LOCAL) {
2150 goto exit_unlock_free;
2153 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2154 info->snd_portid, info->snd_seq, 0,
2155 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2158 /* the vport deletion may trigger dp headroom update */
2160 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2161 must_update_headroom = true;
2162 netdev_reset_rx_headroom(vport->dev);
2163 ovs_dp_detach_port(vport);
2165 if (must_update_headroom)
2166 update_headroom(dp);
2169 ovs_notify(&dp_vport_genl_family, reply, info);
2178 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2180 struct nlattr **a = info->attrs;
2181 struct ovs_header *ovs_header = info->userhdr;
2182 struct sk_buff *reply;
2183 struct vport *vport;
2186 reply = ovs_vport_cmd_alloc_info();
2191 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2192 err = PTR_ERR(vport);
2194 goto exit_unlock_free;
2195 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2196 info->snd_portid, info->snd_seq, 0,
2197 OVS_VPORT_CMD_NEW, GFP_ATOMIC);
2201 return genlmsg_reply(reply, info);
2209 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2211 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2212 struct datapath *dp;
2213 int bucket = cb->args[0], skip = cb->args[1];
2217 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2222 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2223 struct vport *vport;
2226 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2228 ovs_vport_cmd_fill_info(vport, skb,
2230 NETLINK_CB(cb->skb).portid,
2250 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2251 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2252 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2253 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2254 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2255 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2256 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2257 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2258 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2261 static const struct genl_ops dp_vport_genl_ops[] = {
2262 { .cmd = OVS_VPORT_CMD_NEW,
2263 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2264 .policy = vport_policy,
2265 .doit = ovs_vport_cmd_new
2267 { .cmd = OVS_VPORT_CMD_DEL,
2268 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2269 .policy = vport_policy,
2270 .doit = ovs_vport_cmd_del
2272 { .cmd = OVS_VPORT_CMD_GET,
2273 .flags = 0, /* OK for unprivileged users. */
2274 .policy = vport_policy,
2275 .doit = ovs_vport_cmd_get,
2276 .dumpit = ovs_vport_cmd_dump
2278 { .cmd = OVS_VPORT_CMD_SET,
2279 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2280 .policy = vport_policy,
2281 .doit = ovs_vport_cmd_set,
2285 struct genl_family dp_vport_genl_family __ro_after_init = {
2286 .hdrsize = sizeof(struct ovs_header),
2287 .name = OVS_VPORT_FAMILY,
2288 .version = OVS_VPORT_VERSION,
2289 .maxattr = OVS_VPORT_ATTR_MAX,
2291 .parallel_ops = true,
2292 .ops = dp_vport_genl_ops,
2293 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2294 .mcgrps = &ovs_dp_vport_multicast_group,
2296 .module = THIS_MODULE,
2299 static struct genl_family * const dp_genl_families[] = {
2300 &dp_datapath_genl_family,
2301 &dp_vport_genl_family,
2302 &dp_flow_genl_family,
2303 &dp_packet_genl_family,
2304 &dp_meter_genl_family,
2305 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2306 &dp_ct_limit_genl_family,
2310 static void dp_unregister_genl(int n_families)
2314 for (i = 0; i < n_families; i++)
2315 genl_unregister_family(dp_genl_families[i]);
2318 static int __init dp_register_genl(void)
2323 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2325 err = genl_register_family(dp_genl_families[i]);
2333 dp_unregister_genl(i);
2337 static int __net_init ovs_init_net(struct net *net)
2339 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2341 INIT_LIST_HEAD(&ovs_net->dps);
2342 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2343 return ovs_ct_init(net);
2346 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2347 struct list_head *head)
2349 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2350 struct datapath *dp;
2352 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2355 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2356 struct vport *vport;
2358 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2359 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2362 if (dev_net(vport->dev) == dnet)
2363 list_add(&vport->detach_list, head);
2369 static void __net_exit ovs_exit_net(struct net *dnet)
2371 struct datapath *dp, *dp_next;
2372 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2373 struct vport *vport, *vport_next;
2379 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2382 down_read(&net_rwsem);
2384 list_vports_from_net(net, dnet, &head);
2385 up_read(&net_rwsem);
2387 /* Detach all vports from given namespace. */
2388 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2389 list_del(&vport->detach_list);
2390 ovs_dp_detach_port(vport);
2395 cancel_work_sync(&ovs_net->dp_notify_work);
2398 static struct pernet_operations ovs_net_ops = {
2399 .init = ovs_init_net,
2400 .exit = ovs_exit_net,
2402 .size = sizeof(struct ovs_net),
2405 static int __init dp_init(void)
2409 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2411 pr_info("Open vSwitch switching datapath\n");
2413 err = action_fifos_init();
2417 err = ovs_internal_dev_rtnl_link_register();
2419 goto error_action_fifos_exit;
2421 err = ovs_flow_init();
2423 goto error_unreg_rtnl_link;
2425 err = ovs_vport_init();
2427 goto error_flow_exit;
2429 err = register_pernet_device(&ovs_net_ops);
2431 goto error_vport_exit;
2433 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2435 goto error_netns_exit;
2437 err = ovs_netdev_init();
2439 goto error_unreg_notifier;
2441 err = dp_register_genl();
2443 goto error_unreg_netdev;
2449 error_unreg_notifier:
2450 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2452 unregister_pernet_device(&ovs_net_ops);
2457 error_unreg_rtnl_link:
2458 ovs_internal_dev_rtnl_link_unregister();
2459 error_action_fifos_exit:
2460 action_fifos_exit();
2465 static void dp_cleanup(void)
2467 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2469 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2470 unregister_pernet_device(&ovs_net_ops);
2474 ovs_internal_dev_rtnl_link_unregister();
2475 action_fifos_exit();
2478 module_init(dp_init);
2479 module_exit(dp_cleanup);
2481 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2482 MODULE_LICENSE("GPL");
2483 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2484 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2485 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2486 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2487 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2488 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);