2 * Copyright (c) 2007-2015 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/version.h>
40 #include <linux/ethtool.h>
41 #include <linux/wait.h>
42 #include <asm/div64.h>
43 #include <linux/highmem.h>
44 #include <linux/netfilter_bridge.h>
45 #include <linux/netfilter_ipv4.h>
46 #include <linux/inetdevice.h>
47 #include <linux/list.h>
48 #include <linux/openvswitch.h>
49 #include <linux/rculist.h>
50 #include <linux/dmi.h>
51 #include <net/genetlink.h>
52 #include <net/net_namespace.h>
53 #include <net/netns/generic.h>
56 #include "conntrack.h"
58 #include "flow_table.h"
59 #include "flow_netlink.h"
61 #include "vport-internal_dev.h"
62 #include "vport-netdev.h"
64 int ovs_net_id __read_mostly;
65 EXPORT_SYMBOL_GPL(ovs_net_id);
67 static struct genl_family dp_packet_genl_family;
68 static struct genl_family dp_flow_genl_family;
69 static struct genl_family dp_datapath_genl_family;
71 static const struct nla_policy flow_policy[];
73 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
74 .name = OVS_FLOW_MCGROUP
77 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
78 .name = OVS_DATAPATH_MCGROUP
81 struct genl_multicast_group ovs_dp_vport_multicast_group = {
82 .name = OVS_VPORT_MCGROUP
85 /* Check if need to build a reply message.
86 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply.
88 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
91 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
92 genl_has_listeners(family, genl_info_net(info), group);
95 static void ovs_notify(struct genl_family *family, struct genl_multicast_group *grp,
96 struct sk_buff *skb, struct genl_info *info)
98 genl_notify(family, skb, info, GROUP_ID(grp), GFP_KERNEL);
104 * All writes e.g. Writes to device state (add/remove datapath, port, set
105 * operations on vports, etc.), Writes to other state (flow table
106 * modifications, set miscellaneous datapath parameters, etc.) are protected
109 * Reads are protected by RCU.
111 * There are a few special cases (mostly stats) that have their own
112 * synchronization but they nest under all of above and don't interact with
115 * The RTNL lock nests inside ovs_mutex.
118 static DEFINE_MUTEX(ovs_mutex);
122 mutex_lock(&ovs_mutex);
125 void ovs_unlock(void)
127 mutex_unlock(&ovs_mutex);
130 #ifdef CONFIG_LOCKDEP
131 int lockdep_ovsl_is_held(void)
134 return lockdep_is_held(&ovs_mutex);
138 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
141 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
142 const struct sw_flow_key *,
143 const struct dp_upcall_info *,
145 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
146 const struct sw_flow_key *,
147 const struct dp_upcall_info *,
150 /* Must be called with rcu_read_lock. */
151 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
153 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
156 struct vport *vport = ovs_internal_dev_get_vport(dev);
164 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
165 * returned dp pointer valid.
167 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
171 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
173 dp = get_dp_rcu(net, dp_ifindex);
179 /* Must be called with rcu_read_lock or ovs_mutex. */
180 const char *ovs_dp_name(const struct datapath *dp)
182 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
183 return ovs_vport_name(vport);
186 static int get_dpifindex(const struct datapath *dp)
193 local = ovs_vport_rcu(dp, OVSP_LOCAL);
195 ifindex = local->dev->ifindex;
204 static void destroy_dp_rcu(struct rcu_head *rcu)
206 struct datapath *dp = container_of(rcu, struct datapath, rcu);
208 ovs_flow_tbl_destroy(&dp->table);
209 free_percpu(dp->stats_percpu);
214 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
217 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
220 /* Called with ovs_mutex or RCU read lock. */
221 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
224 struct hlist_head *head;
226 head = vport_hash_bucket(dp, port_no);
227 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
228 if (vport->port_no == port_no)
234 /* Called with ovs_mutex. */
235 static struct vport *new_vport(const struct vport_parms *parms)
239 vport = ovs_vport_add(parms);
240 if (!IS_ERR(vport)) {
241 struct datapath *dp = parms->dp;
242 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
244 hlist_add_head_rcu(&vport->dp_hash_node, head);
249 void ovs_dp_detach_port(struct vport *p)
253 /* First drop references to device. */
254 hlist_del_rcu(&p->dp_hash_node);
256 /* Then destroy it. */
260 /* Must be called with rcu_read_lock. */
261 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
263 const struct vport *p = OVS_CB(skb)->input_vport;
264 struct datapath *dp = p->dp;
265 struct sw_flow *flow;
266 struct sw_flow_actions *sf_acts;
267 struct dp_stats_percpu *stats;
271 stats = this_cpu_ptr(dp->stats_percpu);
274 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
276 if (unlikely(!flow)) {
277 struct dp_upcall_info upcall;
280 memset(&upcall, 0, sizeof(upcall));
281 upcall.cmd = OVS_PACKET_CMD_MISS;
282 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
283 upcall.mru = OVS_CB(skb)->mru;
284 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
289 stats_counter = &stats->n_missed;
293 ovs_flow_stats_update(flow, key->tp.flags, skb);
294 sf_acts = rcu_dereference(flow->sf_acts);
295 ovs_execute_actions(dp, skb, sf_acts, key);
297 stats_counter = &stats->n_hit;
300 /* Update datapath statistics. */
301 u64_stats_update_begin(&stats->syncp);
303 stats->n_mask_hit += n_mask_hit;
304 u64_stats_update_end(&stats->syncp);
307 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
308 const struct sw_flow_key *key,
309 const struct dp_upcall_info *upcall_info,
312 struct dp_stats_percpu *stats;
315 if (upcall_info->portid == 0) {
320 if (!skb_is_gso(skb))
321 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
323 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
330 stats = this_cpu_ptr(dp->stats_percpu);
332 u64_stats_update_begin(&stats->syncp);
334 u64_stats_update_end(&stats->syncp);
339 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
340 const struct sw_flow_key *key,
341 const struct dp_upcall_info *upcall_info,
344 unsigned short gso_type = skb_shinfo(skb)->gso_type;
345 struct sw_flow_key later_key;
346 struct sk_buff *segs, *nskb;
347 struct ovs_skb_cb ovs_cb;
350 ovs_cb = *OVS_CB(skb);
351 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
352 *OVS_CB(skb) = ovs_cb;
354 return PTR_ERR(segs);
358 if (gso_type & SKB_GSO_UDP) {
359 /* The initial flow key extracted by ovs_flow_key_extract()
360 * in this case is for a first fragment, so we need to
361 * properly mark later fragments.
364 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
367 /* Queue all of the segments. */
370 *OVS_CB(skb) = ovs_cb;
371 if (gso_type & SKB_GSO_UDP && skb != segs)
374 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
378 } while ((skb = skb->next));
380 /* Free all of the segments. */
388 } while ((skb = nskb));
392 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
395 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
396 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
397 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
398 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
400 /* OVS_PACKET_ATTR_USERDATA */
401 if (upcall_info->userdata)
402 size += NLA_ALIGN(upcall_info->userdata->nla_len);
404 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
405 if (upcall_info->egress_tun_info)
406 size += nla_total_size(ovs_tun_key_attr_size());
408 /* OVS_PACKET_ATTR_ACTIONS */
409 if (upcall_info->actions_len)
410 size += nla_total_size(upcall_info->actions_len);
412 /* OVS_PACKET_ATTR_MRU */
413 if (upcall_info->mru)
414 size += nla_total_size(sizeof(upcall_info->mru));
419 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
421 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
422 size_t plen = NLA_ALIGN(skb->len) - skb->len;
425 memset(skb_put(skb, plen), 0, plen);
429 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
430 const struct sw_flow_key *key,
431 const struct dp_upcall_info *upcall_info,
434 struct ovs_header *upcall;
435 struct sk_buff *nskb = NULL;
436 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
442 dp_ifindex = get_dpifindex(dp);
446 if (skb_vlan_tag_present(skb)) {
447 nskb = skb_clone(skb, GFP_ATOMIC);
451 nskb = __vlan_hwaccel_push_inside(nskb);
458 if (nla_attr_size(skb->len) > USHRT_MAX) {
463 /* Complete checksum if needed */
464 if (skb->ip_summed == CHECKSUM_PARTIAL &&
465 (err = skb_checksum_help(skb)))
468 /* Older versions of OVS user space enforce alignment of the last
469 * Netlink attribute to NLA_ALIGNTO which would require extensive
470 * padding logic. Only perform zerocopy if padding is not required.
472 if (dp->user_features & OVS_DP_F_UNALIGNED)
473 hlen = skb_zerocopy_headlen(skb);
477 len = upcall_msg_size(upcall_info, hlen - cutlen);
478 user_skb = genlmsg_new(len, GFP_ATOMIC);
484 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
485 0, upcall_info->cmd);
486 upcall->dp_ifindex = dp_ifindex;
488 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
491 if (upcall_info->userdata)
492 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
493 nla_len(upcall_info->userdata),
494 nla_data(upcall_info->userdata));
497 if (upcall_info->egress_tun_info) {
498 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
499 err = ovs_nla_put_tunnel_info(user_skb,
500 upcall_info->egress_tun_info);
502 nla_nest_end(user_skb, nla);
505 if (upcall_info->actions_len) {
506 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
507 err = ovs_nla_put_actions(upcall_info->actions,
508 upcall_info->actions_len,
511 nla_nest_end(user_skb, nla);
513 nla_nest_cancel(user_skb, nla);
516 /* Add OVS_PACKET_ATTR_MRU */
517 if (upcall_info->mru) {
518 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
523 pad_packet(dp, user_skb);
526 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
528 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
533 pad_packet(dp, user_skb);
536 /* Only reserve room for attribute header, packet data is added
539 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
543 nla->nla_len = nla_attr_size(skb->len - cutlen);
545 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
549 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
550 pad_packet(dp, user_skb);
552 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
554 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
564 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
566 struct ovs_header *ovs_header = info->userhdr;
567 struct net *net = sock_net(skb->sk);
568 struct nlattr **a = info->attrs;
569 struct sw_flow_actions *acts;
570 struct sk_buff *packet;
571 struct sw_flow *flow;
572 struct sw_flow_actions *sf_acts;
575 struct vport *input_vport;
579 bool log = !a[OVS_PACKET_ATTR_PROBE];
582 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
583 !a[OVS_PACKET_ATTR_ACTIONS])
586 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
587 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
591 skb_reserve(packet, NET_IP_ALIGN);
593 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
595 skb_reset_mac_header(packet);
596 eth = eth_hdr(packet);
598 /* Normally, setting the skb 'protocol' field would be handled by a
599 * call to eth_type_trans(), but it assumes there's a sending
600 * device, which we may not have.
602 if (eth_proto_is_802_3(eth->h_proto))
603 packet->protocol = eth->h_proto;
605 packet->protocol = htons(ETH_P_802_2);
607 /* Set packet's mru */
608 if (a[OVS_PACKET_ATTR_MRU]) {
609 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
610 packet->ignore_df = 1;
612 OVS_CB(packet)->mru = mru;
614 /* Build an sw_flow for sending this packet. */
615 flow = ovs_flow_alloc();
620 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
621 packet, &flow->key, log);
625 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
626 &flow->key, &acts, log);
630 rcu_assign_pointer(flow->sf_acts, acts);
631 packet->priority = flow->key.phy.priority;
632 packet->mark = flow->key.phy.skb_mark;
635 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
640 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
642 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
647 packet->dev = input_vport->dev;
648 OVS_CB(packet)->input_vport = input_vport;
649 sf_acts = rcu_dereference(flow->sf_acts);
652 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
656 ovs_flow_free(flow, false);
662 ovs_flow_free(flow, false);
669 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
670 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
671 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
672 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
673 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
674 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
677 static struct genl_ops dp_packet_genl_ops[] = {
678 { .cmd = OVS_PACKET_CMD_EXECUTE,
679 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
680 .policy = packet_policy,
681 .doit = ovs_packet_cmd_execute
685 static struct genl_family dp_packet_genl_family = {
686 .id = GENL_ID_GENERATE,
687 .hdrsize = sizeof(struct ovs_header),
688 .name = OVS_PACKET_FAMILY,
689 .version = OVS_PACKET_VERSION,
690 .maxattr = OVS_PACKET_ATTR_MAX,
692 .parallel_ops = true,
693 .ops = dp_packet_genl_ops,
694 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
697 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
698 struct ovs_dp_megaflow_stats *mega_stats)
702 memset(mega_stats, 0, sizeof(*mega_stats));
704 stats->n_flows = ovs_flow_tbl_count(&dp->table);
705 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
707 stats->n_hit = stats->n_missed = stats->n_lost = 0;
709 for_each_possible_cpu(i) {
710 const struct dp_stats_percpu *percpu_stats;
711 struct dp_stats_percpu local_stats;
714 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
717 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
718 local_stats = *percpu_stats;
719 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
721 stats->n_hit += local_stats.n_hit;
722 stats->n_missed += local_stats.n_missed;
723 stats->n_lost += local_stats.n_lost;
724 mega_stats->n_mask_hit += local_stats.n_mask_hit;
728 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
730 return ovs_identifier_is_ufid(sfid) &&
731 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
734 static bool should_fill_mask(uint32_t ufid_flags)
736 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
739 static bool should_fill_actions(uint32_t ufid_flags)
741 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
744 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
745 const struct sw_flow_id *sfid,
748 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
750 /* OVS_FLOW_ATTR_UFID */
751 if (sfid && ovs_identifier_is_ufid(sfid))
752 len += nla_total_size(sfid->ufid_len);
754 /* OVS_FLOW_ATTR_KEY */
755 if (!sfid || should_fill_key(sfid, ufid_flags))
756 len += nla_total_size(ovs_key_attr_size());
758 /* OVS_FLOW_ATTR_MASK */
759 if (should_fill_mask(ufid_flags))
760 len += nla_total_size(ovs_key_attr_size());
762 /* OVS_FLOW_ATTR_ACTIONS */
763 if (should_fill_actions(ufid_flags))
764 len += nla_total_size(acts->orig_len);
767 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
768 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
769 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
772 /* Called with ovs_mutex or RCU read lock. */
773 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
776 struct ovs_flow_stats stats;
780 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
783 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
787 if (stats.n_packets &&
788 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
789 sizeof(struct ovs_flow_stats), &stats,
793 if ((u8)ntohs(tcp_flags) &&
794 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
800 /* Called with ovs_mutex or RCU read lock. */
801 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
802 struct sk_buff *skb, int skb_orig_len)
804 struct nlattr *start;
807 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
808 * this is the first flow to be dumped into 'skb'. This is unusual for
809 * Netlink but individual action lists can be longer than
810 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
811 * The userspace caller can always fetch the actions separately if it
812 * really wants them. (Most userspace callers in fact don't care.)
814 * This can only fail for dump operations because the skb is always
815 * properly sized for single flows.
817 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
819 const struct sw_flow_actions *sf_acts;
821 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
822 err = ovs_nla_put_actions(sf_acts->actions,
823 sf_acts->actions_len, skb);
826 nla_nest_end(skb, start);
831 nla_nest_cancel(skb, start);
833 } else if (skb_orig_len) {
840 /* Called with ovs_mutex or RCU read lock. */
841 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
842 struct sk_buff *skb, u32 portid,
843 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
845 const int skb_orig_len = skb->len;
846 struct ovs_header *ovs_header;
849 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
854 ovs_header->dp_ifindex = dp_ifindex;
856 err = ovs_nla_put_identifier(flow, skb);
860 if (should_fill_key(&flow->id, ufid_flags)) {
861 err = ovs_nla_put_masked_key(flow, skb);
866 if (should_fill_mask(ufid_flags)) {
867 err = ovs_nla_put_mask(flow, skb);
872 err = ovs_flow_cmd_fill_stats(flow, skb);
876 if (should_fill_actions(ufid_flags)) {
877 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
882 genlmsg_end(skb, ovs_header);
886 genlmsg_cancel(skb, ovs_header);
890 /* May not be called with RCU read lock. */
891 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
892 const struct sw_flow_id *sfid,
893 struct genl_info *info,
900 if (!always && !ovs_must_notify(&dp_flow_genl_family, info,
901 GROUP_ID(&ovs_dp_flow_multicast_group)))
904 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
905 skb = genlmsg_new(len, GFP_KERNEL);
907 return ERR_PTR(-ENOMEM);
912 /* Called with ovs_mutex. */
913 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
915 struct genl_info *info, u8 cmd,
916 bool always, u32 ufid_flags)
921 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
922 &flow->id, info, always, ufid_flags);
923 if (IS_ERR_OR_NULL(skb))
926 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
927 info->snd_portid, info->snd_seq, 0,
933 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
935 struct net *net = sock_net(skb->sk);
936 struct nlattr **a = info->attrs;
937 struct ovs_header *ovs_header = info->userhdr;
938 struct sw_flow *flow = NULL, *new_flow;
939 struct sw_flow_mask mask;
940 struct sk_buff *reply;
942 struct sw_flow_key key;
943 struct sw_flow_actions *acts;
944 struct sw_flow_match match;
945 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
947 bool log = !a[OVS_FLOW_ATTR_PROBE];
949 /* Must have key and actions. */
951 if (!a[OVS_FLOW_ATTR_KEY]) {
952 OVS_NLERR(log, "Flow key attr not present in new flow.");
955 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
956 OVS_NLERR(log, "Flow actions attr not present in new flow.");
960 /* Most of the time we need to allocate a new flow, do it before
963 new_flow = ovs_flow_alloc();
964 if (IS_ERR(new_flow)) {
965 error = PTR_ERR(new_flow);
970 ovs_match_init(&match, &key, &mask);
971 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
972 a[OVS_FLOW_ATTR_MASK], log);
976 ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
978 /* Extract flow identifier. */
979 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
984 /* Validate actions. */
985 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
986 &new_flow->key, &acts, log);
988 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
992 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
995 error = PTR_ERR(reply);
1000 dp = get_dp(net, ovs_header->dp_ifindex);
1001 if (unlikely(!dp)) {
1003 goto err_unlock_ovs;
1006 /* Check if this is a duplicate flow */
1007 if (ovs_identifier_is_ufid(&new_flow->id))
1008 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1010 flow = ovs_flow_tbl_lookup(&dp->table, &key);
1011 if (likely(!flow)) {
1012 rcu_assign_pointer(new_flow->sf_acts, acts);
1014 /* Put flow in bucket. */
1015 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1016 if (unlikely(error)) {
1018 goto err_unlock_ovs;
1021 if (unlikely(reply)) {
1022 error = ovs_flow_cmd_fill_info(new_flow,
1023 ovs_header->dp_ifindex,
1024 reply, info->snd_portid,
1032 struct sw_flow_actions *old_acts;
1034 /* Bail out if we're not allowed to modify an existing flow.
1035 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1036 * because Generic Netlink treats the latter as a dump
1037 * request. We also accept NLM_F_EXCL in case that bug ever
1040 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1043 goto err_unlock_ovs;
1045 /* The flow identifier has to be the same for flow updates.
1046 * Look for any overlapping flow.
1048 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1049 if (ovs_identifier_is_key(&flow->id))
1050 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1052 else /* UFID matches but key is different */
1056 goto err_unlock_ovs;
1059 /* Update actions. */
1060 old_acts = ovsl_dereference(flow->sf_acts);
1061 rcu_assign_pointer(flow->sf_acts, acts);
1063 if (unlikely(reply)) {
1064 error = ovs_flow_cmd_fill_info(flow,
1065 ovs_header->dp_ifindex,
1066 reply, info->snd_portid,
1074 ovs_nla_free_flow_actions_rcu(old_acts);
1075 ovs_flow_free(new_flow, false);
1079 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1086 ovs_nla_free_flow_actions(acts);
1088 ovs_flow_free(new_flow, false);
1093 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1094 static struct sw_flow_actions *get_flow_actions(struct net *net,
1095 const struct nlattr *a,
1096 const struct sw_flow_key *key,
1097 const struct sw_flow_mask *mask,
1100 struct sw_flow_actions *acts;
1101 struct sw_flow_key masked_key;
1104 ovs_flow_mask_key(&masked_key, key, true, mask);
1105 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1108 "Actions may not be safe on all matching packets");
1109 return ERR_PTR(error);
1115 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1117 struct net *net = sock_net(skb->sk);
1118 struct nlattr **a = info->attrs;
1119 struct ovs_header *ovs_header = info->userhdr;
1120 struct sw_flow_key key;
1121 struct sw_flow *flow;
1122 struct sw_flow_mask mask;
1123 struct sk_buff *reply = NULL;
1124 struct datapath *dp;
1125 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1126 struct sw_flow_match match;
1127 struct sw_flow_id sfid;
1128 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1130 bool log = !a[OVS_FLOW_ATTR_PROBE];
1133 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1134 if (a[OVS_FLOW_ATTR_KEY]) {
1135 ovs_match_init(&match, &key, &mask);
1136 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1137 a[OVS_FLOW_ATTR_MASK], log);
1138 } else if (!ufid_present) {
1140 "Flow set message rejected, Key attribute missing.");
1146 /* Validate actions. */
1147 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1148 if (!a[OVS_FLOW_ATTR_KEY]) {
1150 "Flow key attribute not present in set flow.");
1155 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1158 error = PTR_ERR(acts);
1162 /* Can allocate before locking if have acts. */
1163 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1165 if (IS_ERR(reply)) {
1166 error = PTR_ERR(reply);
1167 goto err_kfree_acts;
1172 dp = get_dp(net, ovs_header->dp_ifindex);
1173 if (unlikely(!dp)) {
1175 goto err_unlock_ovs;
1177 /* Check that the flow exists. */
1179 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1181 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1182 if (unlikely(!flow)) {
1184 goto err_unlock_ovs;
1187 /* Update actions, if present. */
1189 old_acts = ovsl_dereference(flow->sf_acts);
1190 rcu_assign_pointer(flow->sf_acts, acts);
1192 if (unlikely(reply)) {
1193 error = ovs_flow_cmd_fill_info(flow,
1194 ovs_header->dp_ifindex,
1195 reply, info->snd_portid,
1202 /* Could not alloc without acts before locking. */
1203 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1204 info, OVS_FLOW_CMD_NEW, false,
1207 if (unlikely(IS_ERR(reply))) {
1208 error = PTR_ERR(reply);
1209 goto err_unlock_ovs;
1214 if (a[OVS_FLOW_ATTR_CLEAR])
1215 ovs_flow_stats_clear(flow);
1219 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1221 ovs_nla_free_flow_actions_rcu(old_acts);
1229 ovs_nla_free_flow_actions(acts);
1234 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1236 struct nlattr **a = info->attrs;
1237 struct ovs_header *ovs_header = info->userhdr;
1238 struct net *net = sock_net(skb->sk);
1239 struct sw_flow_key key;
1240 struct sk_buff *reply;
1241 struct sw_flow *flow;
1242 struct datapath *dp;
1243 struct sw_flow_match match;
1244 struct sw_flow_id ufid;
1245 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1247 bool log = !a[OVS_FLOW_ATTR_PROBE];
1250 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1251 if (a[OVS_FLOW_ATTR_KEY]) {
1252 ovs_match_init(&match, &key, NULL);
1253 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1255 } else if (!ufid_present) {
1257 "Flow get message rejected, Key attribute missing.");
1264 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1271 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1273 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1279 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1280 OVS_FLOW_CMD_NEW, true, ufid_flags);
1281 if (IS_ERR(reply)) {
1282 err = PTR_ERR(reply);
1287 return genlmsg_reply(reply, info);
1293 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1295 struct nlattr **a = info->attrs;
1296 struct ovs_header *ovs_header = info->userhdr;
1297 struct net *net = sock_net(skb->sk);
1298 struct sw_flow_key key;
1299 struct sk_buff *reply;
1300 struct sw_flow *flow = NULL;
1301 struct datapath *dp;
1302 struct sw_flow_match match;
1303 struct sw_flow_id ufid;
1304 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1306 bool log = !a[OVS_FLOW_ATTR_PROBE];
1309 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1310 if (a[OVS_FLOW_ATTR_KEY]) {
1311 ovs_match_init(&match, &key, NULL);
1312 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1319 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1320 if (unlikely(!dp)) {
1325 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1326 err = ovs_flow_tbl_flush(&dp->table);
1331 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1333 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1334 if (unlikely(!flow)) {
1339 ovs_flow_tbl_remove(&dp->table, flow);
1342 reply = ovs_flow_cmd_alloc_info(rcu_dereference_raw(flow->sf_acts),
1343 &flow->id, info, false, ufid_flags);
1345 if (likely(reply)) {
1346 if (likely(!IS_ERR(reply))) {
1347 rcu_read_lock(); /*To keep RCU checker happy. */
1348 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1349 reply, info->snd_portid,
1355 ovs_notify(&dp_flow_genl_family, &ovs_dp_flow_multicast_group, reply, info);
1357 genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
1358 GROUP_ID(&ovs_dp_flow_multicast_group), PTR_ERR(reply));
1363 ovs_flow_free(flow, true);
1370 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1372 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1373 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1374 struct table_instance *ti;
1375 struct datapath *dp;
1379 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1380 OVS_FLOW_ATTR_MAX, flow_policy);
1383 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1386 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1392 ti = rcu_dereference(dp->table.ti);
1394 struct sw_flow *flow;
1397 bucket = cb->args[0];
1399 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1403 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1404 NETLINK_CB(cb->skb).portid,
1405 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1406 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1409 cb->args[0] = bucket;
1416 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1417 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1418 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1419 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1420 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1421 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1422 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1423 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1426 static struct genl_ops dp_flow_genl_ops[] = {
1427 { .cmd = OVS_FLOW_CMD_NEW,
1428 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429 .policy = flow_policy,
1430 .doit = ovs_flow_cmd_new
1432 { .cmd = OVS_FLOW_CMD_DEL,
1433 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1434 .policy = flow_policy,
1435 .doit = ovs_flow_cmd_del
1437 { .cmd = OVS_FLOW_CMD_GET,
1438 .flags = 0, /* OK for unprivileged users. */
1439 .policy = flow_policy,
1440 .doit = ovs_flow_cmd_get,
1441 .dumpit = ovs_flow_cmd_dump
1443 { .cmd = OVS_FLOW_CMD_SET,
1444 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1445 .policy = flow_policy,
1446 .doit = ovs_flow_cmd_set,
1450 static struct genl_family dp_flow_genl_family = {
1451 .id = GENL_ID_GENERATE,
1452 .hdrsize = sizeof(struct ovs_header),
1453 .name = OVS_FLOW_FAMILY,
1454 .version = OVS_FLOW_VERSION,
1455 .maxattr = OVS_FLOW_ATTR_MAX,
1457 .parallel_ops = true,
1458 .ops = dp_flow_genl_ops,
1459 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1460 .mcgrps = &ovs_dp_flow_multicast_group,
1464 static size_t ovs_dp_cmd_msg_size(void)
1466 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1468 msgsize += nla_total_size(IFNAMSIZ);
1469 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1470 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1471 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1476 /* Called with ovs_mutex. */
1477 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1478 u32 portid, u32 seq, u32 flags, u8 cmd)
1480 struct ovs_header *ovs_header;
1481 struct ovs_dp_stats dp_stats;
1482 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1485 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1490 ovs_header->dp_ifindex = get_dpifindex(dp);
1492 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1494 goto nla_put_failure;
1496 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1497 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1498 &dp_stats, OVS_DP_ATTR_PAD))
1499 goto nla_put_failure;
1501 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1502 sizeof(struct ovs_dp_megaflow_stats),
1503 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1504 goto nla_put_failure;
1506 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1507 goto nla_put_failure;
1509 genlmsg_end(skb, ovs_header);
1513 genlmsg_cancel(skb, ovs_header);
1518 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1520 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1523 /* Called with rcu_read_lock or ovs_mutex. */
1524 static struct datapath *lookup_datapath(struct net *net,
1525 const struct ovs_header *ovs_header,
1526 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1528 struct datapath *dp;
1530 if (!a[OVS_DP_ATTR_NAME])
1531 dp = get_dp(net, ovs_header->dp_ifindex);
1533 struct vport *vport;
1535 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1536 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1538 return dp ? dp : ERR_PTR(-ENODEV);
1541 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1543 struct datapath *dp;
1545 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1549 WARN(dp->user_features, "Dropping previously announced user features\n");
1550 dp->user_features = 0;
1553 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1555 if (a[OVS_DP_ATTR_USER_FEATURES])
1556 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1559 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1561 struct nlattr **a = info->attrs;
1562 struct vport_parms parms;
1563 struct sk_buff *reply;
1564 struct datapath *dp;
1565 struct vport *vport;
1566 struct ovs_net *ovs_net;
1570 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1573 reply = ovs_dp_cmd_alloc_info();
1578 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1580 goto err_free_reply;
1582 ovs_dp_set_net(dp, sock_net(skb->sk));
1584 /* Allocate table. */
1585 err = ovs_flow_tbl_init(&dp->table);
1589 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1590 if (!dp->stats_percpu) {
1592 goto err_destroy_table;
1595 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1599 goto err_destroy_percpu;
1602 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1603 INIT_HLIST_HEAD(&dp->ports[i]);
1605 /* Set up our datapath device. */
1606 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1607 parms.type = OVS_VPORT_TYPE_INTERNAL;
1608 parms.options = NULL;
1610 parms.port_no = OVSP_LOCAL;
1611 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1613 ovs_dp_change(dp, a);
1615 /* So far only local changes have been made, now need the lock. */
1618 vport = new_vport(&parms);
1619 if (IS_ERR(vport)) {
1620 err = PTR_ERR(vport);
1624 if (err == -EEXIST) {
1625 /* An outdated user space instance that does not understand
1626 * the concept of user_features has attempted to create a new
1627 * datapath and is likely to reuse it. Drop all user features.
1629 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1630 ovs_dp_reset_user_features(skb, info);
1633 goto err_destroy_ports_array;
1636 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1637 info->snd_seq, 0, OVS_DP_CMD_NEW);
1640 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1641 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1645 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1648 err_destroy_ports_array:
1652 free_percpu(dp->stats_percpu);
1654 ovs_flow_tbl_destroy(&dp->table);
1663 /* Called with ovs_mutex. */
1664 static void __dp_destroy(struct datapath *dp)
1668 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1669 struct vport *vport;
1670 struct hlist_node *n;
1672 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1673 if (vport->port_no != OVSP_LOCAL)
1674 ovs_dp_detach_port(vport);
1677 list_del_rcu(&dp->list_node);
1679 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1680 * all ports in datapath are destroyed first before freeing datapath.
1682 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1684 /* RCU destroy the flow table */
1685 call_rcu(&dp->rcu, destroy_dp_rcu);
1688 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1690 struct sk_buff *reply;
1691 struct datapath *dp;
1694 reply = ovs_dp_cmd_alloc_info();
1699 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1702 goto err_unlock_free;
1704 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1705 info->snd_seq, 0, OVS_DP_CMD_DEL);
1711 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1720 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1722 struct sk_buff *reply;
1723 struct datapath *dp;
1726 reply = ovs_dp_cmd_alloc_info();
1731 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1734 goto err_unlock_free;
1736 ovs_dp_change(dp, info->attrs);
1738 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1739 info->snd_seq, 0, OVS_DP_CMD_NEW);
1744 ovs_notify(&dp_datapath_genl_family, &ovs_dp_datapath_multicast_group, reply, info);
1753 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1755 struct sk_buff *reply;
1756 struct datapath *dp;
1759 reply = ovs_dp_cmd_alloc_info();
1764 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1767 goto err_unlock_free;
1769 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1770 info->snd_seq, 0, OVS_DP_CMD_NEW);
1774 return genlmsg_reply(reply, info);
1782 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1784 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1785 struct datapath *dp;
1786 int skip = cb->args[0];
1790 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1792 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1793 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1794 OVS_DP_CMD_NEW) < 0)
1805 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1806 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1807 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1808 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1811 static struct genl_ops dp_datapath_genl_ops[] = {
1812 { .cmd = OVS_DP_CMD_NEW,
1813 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1814 .policy = datapath_policy,
1815 .doit = ovs_dp_cmd_new
1817 { .cmd = OVS_DP_CMD_DEL,
1818 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1819 .policy = datapath_policy,
1820 .doit = ovs_dp_cmd_del
1822 { .cmd = OVS_DP_CMD_GET,
1823 .flags = 0, /* OK for unprivileged users. */
1824 .policy = datapath_policy,
1825 .doit = ovs_dp_cmd_get,
1826 .dumpit = ovs_dp_cmd_dump
1828 { .cmd = OVS_DP_CMD_SET,
1829 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1830 .policy = datapath_policy,
1831 .doit = ovs_dp_cmd_set,
1835 static struct genl_family dp_datapath_genl_family = {
1836 .id = GENL_ID_GENERATE,
1837 .hdrsize = sizeof(struct ovs_header),
1838 .name = OVS_DATAPATH_FAMILY,
1839 .version = OVS_DATAPATH_VERSION,
1840 .maxattr = OVS_DP_ATTR_MAX,
1842 .parallel_ops = true,
1843 .ops = dp_datapath_genl_ops,
1844 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1845 .mcgrps = &ovs_dp_datapath_multicast_group,
1849 /* Called with ovs_mutex or RCU read lock. */
1850 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1851 u32 portid, u32 seq, u32 flags, u8 cmd)
1853 struct ovs_header *ovs_header;
1854 struct ovs_vport_stats vport_stats;
1857 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1862 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1864 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1865 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1866 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1867 ovs_vport_name(vport)))
1868 goto nla_put_failure;
1870 ovs_vport_get_stats(vport, &vport_stats);
1871 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1872 sizeof(struct ovs_vport_stats), &vport_stats,
1873 OVS_VPORT_ATTR_PAD))
1874 goto nla_put_failure;
1876 if (ovs_vport_get_upcall_portids(vport, skb))
1877 goto nla_put_failure;
1879 err = ovs_vport_get_options(vport, skb);
1880 if (err == -EMSGSIZE)
1883 genlmsg_end(skb, ovs_header);
1889 genlmsg_cancel(skb, ovs_header);
1893 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1895 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1898 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1899 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1902 struct sk_buff *skb;
1905 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1907 return ERR_PTR(-ENOMEM);
1909 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1915 /* Called with ovs_mutex or RCU read lock. */
1916 static struct vport *lookup_vport(struct net *net,
1917 const struct ovs_header *ovs_header,
1918 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1920 struct datapath *dp;
1921 struct vport *vport;
1923 if (a[OVS_VPORT_ATTR_NAME]) {
1924 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1926 return ERR_PTR(-ENODEV);
1927 if (ovs_header->dp_ifindex &&
1928 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1929 return ERR_PTR(-ENODEV);
1931 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1932 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1934 if (port_no >= DP_MAX_PORTS)
1935 return ERR_PTR(-EFBIG);
1937 dp = get_dp(net, ovs_header->dp_ifindex);
1939 return ERR_PTR(-ENODEV);
1941 vport = ovs_vport_ovsl_rcu(dp, port_no);
1943 return ERR_PTR(-ENODEV);
1946 return ERR_PTR(-EINVAL);
1949 /* Called with ovs_mutex */
1950 static void update_headroom(struct datapath *dp)
1952 unsigned dev_headroom, max_headroom = 0;
1953 struct net_device *dev;
1954 struct vport *vport;
1957 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1958 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1960 dev_headroom = netdev_get_fwd_headroom(dev);
1961 if (dev_headroom > max_headroom)
1962 max_headroom = dev_headroom;
1966 dp->max_headroom = max_headroom;
1967 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1968 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1969 netdev_set_rx_headroom(vport->dev, max_headroom);
1972 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1974 struct nlattr **a = info->attrs;
1975 struct ovs_header *ovs_header = info->userhdr;
1976 struct vport_parms parms;
1977 struct sk_buff *reply;
1978 struct vport *vport;
1979 struct datapath *dp;
1983 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1984 !a[OVS_VPORT_ATTR_UPCALL_PID])
1987 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1988 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1989 if (port_no >= DP_MAX_PORTS)
1992 reply = ovs_vport_cmd_alloc_info();
1998 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2001 goto exit_unlock_free;
2004 vport = ovs_vport_ovsl(dp, port_no);
2007 goto exit_unlock_free;
2009 for (port_no = 1; ; port_no++) {
2010 if (port_no >= DP_MAX_PORTS) {
2012 goto exit_unlock_free;
2014 vport = ovs_vport_ovsl(dp, port_no);
2020 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2021 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2022 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2024 parms.port_no = port_no;
2025 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2027 vport = new_vport(&parms);
2028 err = PTR_ERR(vport);
2029 if (IS_ERR(vport)) {
2032 goto exit_unlock_free;
2035 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2036 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2039 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2040 update_headroom(dp);
2042 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2046 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2055 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2057 struct nlattr **a = info->attrs;
2058 struct sk_buff *reply;
2059 struct vport *vport;
2062 reply = ovs_vport_cmd_alloc_info();
2067 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2068 err = PTR_ERR(vport);
2070 goto exit_unlock_free;
2072 if (a[OVS_VPORT_ATTR_TYPE] &&
2073 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2075 goto exit_unlock_free;
2078 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2079 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2081 goto exit_unlock_free;
2084 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2085 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2087 err = ovs_vport_set_upcall_portids(vport, ids);
2089 goto exit_unlock_free;
2092 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2093 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2097 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2106 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2108 bool must_update_headroom = false;
2109 struct nlattr **a = info->attrs;
2110 struct sk_buff *reply;
2111 struct datapath *dp;
2112 struct vport *vport;
2115 reply = ovs_vport_cmd_alloc_info();
2120 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2121 err = PTR_ERR(vport);
2123 goto exit_unlock_free;
2125 if (vport->port_no == OVSP_LOCAL) {
2127 goto exit_unlock_free;
2130 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2131 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2134 /* the vport deletion may trigger dp headroom update */
2136 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2137 must_update_headroom = true;
2138 netdev_reset_rx_headroom(vport->dev);
2139 ovs_dp_detach_port(vport);
2141 if (must_update_headroom)
2142 update_headroom(dp);
2146 ovs_notify(&dp_vport_genl_family, &ovs_dp_vport_multicast_group, reply, info);
2155 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2157 struct nlattr **a = info->attrs;
2158 struct ovs_header *ovs_header = info->userhdr;
2159 struct sk_buff *reply;
2160 struct vport *vport;
2163 reply = ovs_vport_cmd_alloc_info();
2168 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2169 err = PTR_ERR(vport);
2171 goto exit_unlock_free;
2172 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2173 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2177 return genlmsg_reply(reply, info);
2185 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2187 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2188 struct datapath *dp;
2189 int bucket = cb->args[0], skip = cb->args[1];
2193 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2198 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2199 struct vport *vport;
2202 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2204 ovs_vport_cmd_fill_info(vport, skb,
2205 NETLINK_CB(cb->skb).portid,
2208 OVS_VPORT_CMD_NEW) < 0)
2224 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2225 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2226 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2227 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2228 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2229 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2230 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2233 static struct genl_ops dp_vport_genl_ops[] = {
2234 { .cmd = OVS_VPORT_CMD_NEW,
2235 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2236 .policy = vport_policy,
2237 .doit = ovs_vport_cmd_new
2239 { .cmd = OVS_VPORT_CMD_DEL,
2240 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2241 .policy = vport_policy,
2242 .doit = ovs_vport_cmd_del
2244 { .cmd = OVS_VPORT_CMD_GET,
2245 .flags = 0, /* OK for unprivileged users. */
2246 .policy = vport_policy,
2247 .doit = ovs_vport_cmd_get,
2248 .dumpit = ovs_vport_cmd_dump
2250 { .cmd = OVS_VPORT_CMD_SET,
2251 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2252 .policy = vport_policy,
2253 .doit = ovs_vport_cmd_set,
2257 struct genl_family dp_vport_genl_family = {
2258 .id = GENL_ID_GENERATE,
2259 .hdrsize = sizeof(struct ovs_header),
2260 .name = OVS_VPORT_FAMILY,
2261 .version = OVS_VPORT_VERSION,
2262 .maxattr = OVS_VPORT_ATTR_MAX,
2264 .parallel_ops = true,
2265 .ops = dp_vport_genl_ops,
2266 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2267 .mcgrps = &ovs_dp_vport_multicast_group,
2271 static struct genl_family *dp_genl_families[] = {
2272 &dp_datapath_genl_family,
2273 &dp_vport_genl_family,
2274 &dp_flow_genl_family,
2275 &dp_packet_genl_family,
2278 static void dp_unregister_genl(int n_families)
2282 for (i = 0; i < n_families; i++)
2283 genl_unregister_family(dp_genl_families[i]);
2286 static int dp_register_genl(void)
2291 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2293 err = genl_register_family(dp_genl_families[i]);
2301 dp_unregister_genl(i);
2305 static int __net_init ovs_init_net(struct net *net)
2307 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2309 INIT_LIST_HEAD(&ovs_net->dps);
2310 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2315 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2316 struct list_head *head)
2318 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2319 struct datapath *dp;
2321 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2324 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2325 struct vport *vport;
2327 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2329 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2332 if (dev_net(vport->dev) == dnet)
2333 list_add(&vport->detach_list, head);
2339 static void __net_exit ovs_exit_net(struct net *dnet)
2341 struct datapath *dp, *dp_next;
2342 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2343 struct vport *vport, *vport_next;
2349 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2354 list_vports_from_net(net, dnet, &head);
2357 /* Detach all vports from given namespace. */
2358 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2359 list_del(&vport->detach_list);
2360 ovs_dp_detach_port(vport);
2365 cancel_work_sync(&ovs_net->dp_notify_work);
2368 static struct pernet_operations ovs_net_ops = {
2369 .init = ovs_init_net,
2370 .exit = ovs_exit_net,
2372 .size = sizeof(struct ovs_net),
2375 static int __init dp_init(void)
2379 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2381 pr_info("Open vSwitch switching datapath %s\n", VERSION);
2383 err = compat_init();
2387 err = action_fifos_init();
2389 goto error_compat_exit;
2391 err = ovs_internal_dev_rtnl_link_register();
2393 goto error_action_fifos_exit;
2395 err = ovs_flow_init();
2397 goto error_unreg_rtnl_link;
2399 err = ovs_vport_init();
2401 goto error_flow_exit;
2403 err = register_pernet_device(&ovs_net_ops);
2405 goto error_vport_exit;
2407 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2409 goto error_netns_exit;
2411 err = ovs_netdev_init();
2413 goto error_unreg_notifier;
2415 err = dp_register_genl();
2417 goto error_unreg_netdev;
2423 error_unreg_notifier:
2424 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2426 unregister_pernet_device(&ovs_net_ops);
2431 error_unreg_rtnl_link:
2432 ovs_internal_dev_rtnl_link_unregister();
2433 error_action_fifos_exit:
2434 action_fifos_exit();
2441 static void dp_cleanup(void)
2443 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2445 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2446 unregister_pernet_device(&ovs_net_ops);
2450 ovs_internal_dev_rtnl_link_unregister();
2451 action_fifos_exit();
2455 module_init(dp_init);
2456 module_exit(dp_cleanup);
2458 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2459 MODULE_LICENSE("GPL");
2460 MODULE_VERSION(VERSION);