2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
61 unsigned int ovs_net_id __read_mostly;
63 static struct genl_family dp_packet_genl_family;
64 static struct genl_family dp_flow_genl_family;
65 static struct genl_family dp_datapath_genl_family;
67 static const struct nla_policy flow_policy[];
69 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
70 .name = OVS_FLOW_MCGROUP,
73 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
74 .name = OVS_DATAPATH_MCGROUP,
77 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
78 .name = OVS_VPORT_MCGROUP,
81 /* Check if need to build a reply message.
82 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
86 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
87 genl_has_listeners(family, genl_info_net(info), group);
90 static void ovs_notify(struct genl_family *family,
91 struct sk_buff *skb, struct genl_info *info)
93 genl_notify(family, skb, info, 0, GFP_KERNEL);
99 * All writes e.g. Writes to device state (add/remove datapath, port, set
100 * operations on vports, etc.), Writes to other state (flow table
101 * modifications, set miscellaneous datapath parameters, etc.) are protected
104 * Reads are protected by RCU.
106 * There are a few special cases (mostly stats) that have their own
107 * synchronization but they nest under all of above and don't interact with
110 * The RTNL lock nests inside ovs_mutex.
113 static DEFINE_MUTEX(ovs_mutex);
117 mutex_lock(&ovs_mutex);
120 void ovs_unlock(void)
122 mutex_unlock(&ovs_mutex);
125 #ifdef CONFIG_LOCKDEP
126 int lockdep_ovsl_is_held(void)
129 return lockdep_is_held(&ovs_mutex);
135 static struct vport *new_vport(const struct vport_parms *);
136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
137 const struct sw_flow_key *,
138 const struct dp_upcall_info *,
140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 const struct sw_flow_key *,
142 const struct dp_upcall_info *,
145 /* Must be called with rcu_read_lock. */
146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
148 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
151 struct vport *vport = ovs_internal_dev_get_vport(dev);
159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160 * returned dp pointer valid.
162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
166 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
168 dp = get_dp_rcu(net, dp_ifindex);
174 /* Must be called with rcu_read_lock or ovs_mutex. */
175 const char *ovs_dp_name(const struct datapath *dp)
177 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
178 return ovs_vport_name(vport);
181 static int get_dpifindex(const struct datapath *dp)
188 local = ovs_vport_rcu(dp, OVSP_LOCAL);
190 ifindex = local->dev->ifindex;
199 static void destroy_dp_rcu(struct rcu_head *rcu)
201 struct datapath *dp = container_of(rcu, struct datapath, rcu);
203 ovs_flow_tbl_destroy(&dp->table);
204 free_percpu(dp->stats_percpu);
209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
212 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
215 /* Called with ovs_mutex or RCU read lock. */
216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
219 struct hlist_head *head;
221 head = vport_hash_bucket(dp, port_no);
222 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
223 if (vport->port_no == port_no)
229 /* Called with ovs_mutex. */
230 static struct vport *new_vport(const struct vport_parms *parms)
234 vport = ovs_vport_add(parms);
235 if (!IS_ERR(vport)) {
236 struct datapath *dp = parms->dp;
237 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
239 hlist_add_head_rcu(&vport->dp_hash_node, head);
244 void ovs_dp_detach_port(struct vport *p)
248 /* First drop references to device. */
249 hlist_del_rcu(&p->dp_hash_node);
251 /* Then destroy it. */
255 /* Must be called with rcu_read_lock. */
256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
258 const struct vport *p = OVS_CB(skb)->input_vport;
259 struct datapath *dp = p->dp;
260 struct sw_flow *flow;
261 struct sw_flow_actions *sf_acts;
262 struct dp_stats_percpu *stats;
266 stats = this_cpu_ptr(dp->stats_percpu);
269 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
270 if (unlikely(!flow)) {
271 struct dp_upcall_info upcall;
274 memset(&upcall, 0, sizeof(upcall));
275 upcall.cmd = OVS_PACKET_CMD_MISS;
276 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 upcall.mru = OVS_CB(skb)->mru;
278 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
290 stats_counter = &stats->n_missed;
294 ovs_flow_stats_update(flow, key->tp.flags, skb);
295 sf_acts = rcu_dereference(flow->sf_acts);
296 ovs_execute_actions(dp, skb, sf_acts, key);
298 stats_counter = &stats->n_hit;
301 /* Update datapath statistics. */
302 u64_stats_update_begin(&stats->syncp);
304 stats->n_mask_hit += n_mask_hit;
305 u64_stats_update_end(&stats->syncp);
308 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
309 const struct sw_flow_key *key,
310 const struct dp_upcall_info *upcall_info,
313 struct dp_stats_percpu *stats;
316 if (upcall_info->portid == 0) {
321 if (!skb_is_gso(skb))
322 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
324 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
331 stats = this_cpu_ptr(dp->stats_percpu);
333 u64_stats_update_begin(&stats->syncp);
335 u64_stats_update_end(&stats->syncp);
340 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
341 const struct sw_flow_key *key,
342 const struct dp_upcall_info *upcall_info,
345 unsigned int gso_type = skb_shinfo(skb)->gso_type;
346 struct sw_flow_key later_key;
347 struct sk_buff *segs, *nskb;
350 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
351 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
353 return PTR_ERR(segs);
357 if (gso_type & SKB_GSO_UDP) {
358 /* The initial flow key extracted by ovs_flow_key_extract()
359 * in this case is for a first fragment, so we need to
360 * properly mark later fragments.
363 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
366 /* Queue all of the segments. */
369 if (gso_type & SKB_GSO_UDP && skb != segs)
372 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
376 } while ((skb = skb->next));
378 /* Free all of the segments. */
386 } while ((skb = nskb));
390 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
391 unsigned int hdrlen, int actions_attrlen)
393 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
394 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
395 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
396 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
398 /* OVS_PACKET_ATTR_USERDATA */
399 if (upcall_info->userdata)
400 size += NLA_ALIGN(upcall_info->userdata->nla_len);
402 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
403 if (upcall_info->egress_tun_info)
404 size += nla_total_size(ovs_tun_key_attr_size());
406 /* OVS_PACKET_ATTR_ACTIONS */
407 if (upcall_info->actions_len)
408 size += nla_total_size(actions_attrlen);
410 /* OVS_PACKET_ATTR_MRU */
411 if (upcall_info->mru)
412 size += nla_total_size(sizeof(upcall_info->mru));
417 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
419 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
420 size_t plen = NLA_ALIGN(skb->len) - skb->len;
423 skb_put_zero(skb, plen);
427 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
428 const struct sw_flow_key *key,
429 const struct dp_upcall_info *upcall_info,
432 struct ovs_header *upcall;
433 struct sk_buff *nskb = NULL;
434 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
440 dp_ifindex = get_dpifindex(dp);
444 if (skb_vlan_tag_present(skb)) {
445 nskb = skb_clone(skb, GFP_ATOMIC);
449 nskb = __vlan_hwaccel_push_inside(nskb);
456 if (nla_attr_size(skb->len) > USHRT_MAX) {
461 /* Complete checksum if needed */
462 if (skb->ip_summed == CHECKSUM_PARTIAL &&
463 (err = skb_csum_hwoffload_help(skb, 0)))
466 /* Older versions of OVS user space enforce alignment of the last
467 * Netlink attribute to NLA_ALIGNTO which would require extensive
468 * padding logic. Only perform zerocopy if padding is not required.
470 if (dp->user_features & OVS_DP_F_UNALIGNED)
471 hlen = skb_zerocopy_headlen(skb);
475 len = upcall_msg_size(upcall_info, hlen - cutlen,
476 OVS_CB(skb)->acts_origlen);
477 user_skb = genlmsg_new(len, GFP_ATOMIC);
483 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
484 0, upcall_info->cmd);
485 upcall->dp_ifindex = dp_ifindex;
487 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
490 if (upcall_info->userdata)
491 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
492 nla_len(upcall_info->userdata),
493 nla_data(upcall_info->userdata));
495 if (upcall_info->egress_tun_info) {
496 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
497 err = ovs_nla_put_tunnel_info(user_skb,
498 upcall_info->egress_tun_info);
500 nla_nest_end(user_skb, nla);
503 if (upcall_info->actions_len) {
504 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
505 err = ovs_nla_put_actions(upcall_info->actions,
506 upcall_info->actions_len,
509 nla_nest_end(user_skb, nla);
511 nla_nest_cancel(user_skb, nla);
514 /* Add OVS_PACKET_ATTR_MRU */
515 if (upcall_info->mru) {
516 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
521 pad_packet(dp, user_skb);
524 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
526 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
531 pad_packet(dp, user_skb);
534 /* Only reserve room for attribute header, packet data is added
535 * in skb_zerocopy() */
536 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
540 nla->nla_len = nla_attr_size(skb->len - cutlen);
542 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
546 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
547 pad_packet(dp, user_skb);
549 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
551 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
556 consume_skb(user_skb);
562 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
564 struct ovs_header *ovs_header = info->userhdr;
565 struct net *net = sock_net(skb->sk);
566 struct nlattr **a = info->attrs;
567 struct sw_flow_actions *acts;
568 struct sk_buff *packet;
569 struct sw_flow *flow;
570 struct sw_flow_actions *sf_acts;
572 struct vport *input_vport;
576 bool log = !a[OVS_PACKET_ATTR_PROBE];
579 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
580 !a[OVS_PACKET_ATTR_ACTIONS])
583 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
584 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
588 skb_reserve(packet, NET_IP_ALIGN);
590 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
592 /* Set packet's mru */
593 if (a[OVS_PACKET_ATTR_MRU]) {
594 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
595 packet->ignore_df = 1;
597 OVS_CB(packet)->mru = mru;
599 /* Build an sw_flow for sending this packet. */
600 flow = ovs_flow_alloc();
605 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
606 packet, &flow->key, log);
610 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
611 &flow->key, &acts, log);
615 rcu_assign_pointer(flow->sf_acts, acts);
616 packet->priority = flow->key.phy.priority;
617 packet->mark = flow->key.phy.skb_mark;
620 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
625 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
627 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
632 packet->dev = input_vport->dev;
633 OVS_CB(packet)->input_vport = input_vport;
634 sf_acts = rcu_dereference(flow->sf_acts);
637 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
641 ovs_flow_free(flow, false);
647 ovs_flow_free(flow, false);
654 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
655 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
656 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
657 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
658 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
659 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
662 static const struct genl_ops dp_packet_genl_ops[] = {
663 { .cmd = OVS_PACKET_CMD_EXECUTE,
664 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
665 .policy = packet_policy,
666 .doit = ovs_packet_cmd_execute
670 static struct genl_family dp_packet_genl_family __ro_after_init = {
671 .hdrsize = sizeof(struct ovs_header),
672 .name = OVS_PACKET_FAMILY,
673 .version = OVS_PACKET_VERSION,
674 .maxattr = OVS_PACKET_ATTR_MAX,
676 .parallel_ops = true,
677 .ops = dp_packet_genl_ops,
678 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
679 .module = THIS_MODULE,
682 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
683 struct ovs_dp_megaflow_stats *mega_stats)
687 memset(mega_stats, 0, sizeof(*mega_stats));
689 stats->n_flows = ovs_flow_tbl_count(&dp->table);
690 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
692 stats->n_hit = stats->n_missed = stats->n_lost = 0;
694 for_each_possible_cpu(i) {
695 const struct dp_stats_percpu *percpu_stats;
696 struct dp_stats_percpu local_stats;
699 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
702 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
703 local_stats = *percpu_stats;
704 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
706 stats->n_hit += local_stats.n_hit;
707 stats->n_missed += local_stats.n_missed;
708 stats->n_lost += local_stats.n_lost;
709 mega_stats->n_mask_hit += local_stats.n_mask_hit;
713 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
715 return ovs_identifier_is_ufid(sfid) &&
716 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
719 static bool should_fill_mask(uint32_t ufid_flags)
721 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
724 static bool should_fill_actions(uint32_t ufid_flags)
726 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
729 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
730 const struct sw_flow_id *sfid,
733 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
735 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
736 * see ovs_nla_put_identifier()
738 if (sfid && ovs_identifier_is_ufid(sfid))
739 len += nla_total_size(sfid->ufid_len);
741 len += nla_total_size(ovs_key_attr_size());
743 /* OVS_FLOW_ATTR_KEY */
744 if (!sfid || should_fill_key(sfid, ufid_flags))
745 len += nla_total_size(ovs_key_attr_size());
747 /* OVS_FLOW_ATTR_MASK */
748 if (should_fill_mask(ufid_flags))
749 len += nla_total_size(ovs_key_attr_size());
751 /* OVS_FLOW_ATTR_ACTIONS */
752 if (should_fill_actions(ufid_flags))
753 len += nla_total_size(acts->orig_len);
756 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
757 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
758 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
761 /* Called with ovs_mutex or RCU read lock. */
762 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
765 struct ovs_flow_stats stats;
769 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
772 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
776 if (stats.n_packets &&
777 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
778 sizeof(struct ovs_flow_stats), &stats,
782 if ((u8)ntohs(tcp_flags) &&
783 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
789 /* Called with ovs_mutex or RCU read lock. */
790 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
791 struct sk_buff *skb, int skb_orig_len)
793 struct nlattr *start;
796 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
797 * this is the first flow to be dumped into 'skb'. This is unusual for
798 * Netlink but individual action lists can be longer than
799 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
800 * The userspace caller can always fetch the actions separately if it
801 * really wants them. (Most userspace callers in fact don't care.)
803 * This can only fail for dump operations because the skb is always
804 * properly sized for single flows.
806 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
808 const struct sw_flow_actions *sf_acts;
810 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
811 err = ovs_nla_put_actions(sf_acts->actions,
812 sf_acts->actions_len, skb);
815 nla_nest_end(skb, start);
820 nla_nest_cancel(skb, start);
822 } else if (skb_orig_len) {
829 /* Called with ovs_mutex or RCU read lock. */
830 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
831 struct sk_buff *skb, u32 portid,
832 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
834 const int skb_orig_len = skb->len;
835 struct ovs_header *ovs_header;
838 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
843 ovs_header->dp_ifindex = dp_ifindex;
845 err = ovs_nla_put_identifier(flow, skb);
849 if (should_fill_key(&flow->id, ufid_flags)) {
850 err = ovs_nla_put_masked_key(flow, skb);
855 if (should_fill_mask(ufid_flags)) {
856 err = ovs_nla_put_mask(flow, skb);
861 err = ovs_flow_cmd_fill_stats(flow, skb);
865 if (should_fill_actions(ufid_flags)) {
866 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
871 genlmsg_end(skb, ovs_header);
875 genlmsg_cancel(skb, ovs_header);
879 /* May not be called with RCU read lock. */
880 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
881 const struct sw_flow_id *sfid,
882 struct genl_info *info,
889 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
892 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
893 skb = genlmsg_new(len, GFP_KERNEL);
895 return ERR_PTR(-ENOMEM);
900 /* Called with ovs_mutex. */
901 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
903 struct genl_info *info, u8 cmd,
904 bool always, u32 ufid_flags)
909 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
910 &flow->id, info, always, ufid_flags);
911 if (IS_ERR_OR_NULL(skb))
914 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
915 info->snd_portid, info->snd_seq, 0,
917 if (WARN_ON_ONCE(retval < 0)) {
919 skb = ERR_PTR(retval);
924 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
926 struct net *net = sock_net(skb->sk);
927 struct nlattr **a = info->attrs;
928 struct ovs_header *ovs_header = info->userhdr;
929 struct sw_flow *flow = NULL, *new_flow;
930 struct sw_flow_mask mask;
931 struct sk_buff *reply;
933 struct sw_flow_key *key;
934 struct sw_flow_actions *acts;
935 struct sw_flow_match match;
936 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
938 bool log = !a[OVS_FLOW_ATTR_PROBE];
940 /* Must have key and actions. */
942 if (!a[OVS_FLOW_ATTR_KEY]) {
943 OVS_NLERR(log, "Flow key attr not present in new flow.");
946 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
947 OVS_NLERR(log, "Flow actions attr not present in new flow.");
951 /* Most of the time we need to allocate a new flow, do it before
954 new_flow = ovs_flow_alloc();
955 if (IS_ERR(new_flow)) {
956 error = PTR_ERR(new_flow);
961 key = kzalloc(sizeof(*key), GFP_KERNEL);
967 ovs_match_init(&match, key, false, &mask);
968 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
969 a[OVS_FLOW_ATTR_MASK], log);
973 ovs_flow_mask_key(&new_flow->key, key, true, &mask);
975 /* Extract flow identifier. */
976 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
981 /* Validate actions. */
982 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
983 &new_flow->key, &acts, log);
985 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
989 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
992 error = PTR_ERR(reply);
997 dp = get_dp(net, ovs_header->dp_ifindex);
1000 goto err_unlock_ovs;
1003 /* Check if this is a duplicate flow */
1004 if (ovs_identifier_is_ufid(&new_flow->id))
1005 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1007 flow = ovs_flow_tbl_lookup(&dp->table, key);
1008 if (likely(!flow)) {
1009 rcu_assign_pointer(new_flow->sf_acts, acts);
1011 /* Put flow in bucket. */
1012 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1013 if (unlikely(error)) {
1015 goto err_unlock_ovs;
1018 if (unlikely(reply)) {
1019 error = ovs_flow_cmd_fill_info(new_flow,
1020 ovs_header->dp_ifindex,
1021 reply, info->snd_portid,
1029 struct sw_flow_actions *old_acts;
1031 /* Bail out if we're not allowed to modify an existing flow.
1032 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1033 * because Generic Netlink treats the latter as a dump
1034 * request. We also accept NLM_F_EXCL in case that bug ever
1037 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1040 goto err_unlock_ovs;
1042 /* The flow identifier has to be the same for flow updates.
1043 * Look for any overlapping flow.
1045 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1046 if (ovs_identifier_is_key(&flow->id))
1047 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1049 else /* UFID matches but key is different */
1053 goto err_unlock_ovs;
1056 /* Update actions. */
1057 old_acts = ovsl_dereference(flow->sf_acts);
1058 rcu_assign_pointer(flow->sf_acts, acts);
1060 if (unlikely(reply)) {
1061 error = ovs_flow_cmd_fill_info(flow,
1062 ovs_header->dp_ifindex,
1063 reply, info->snd_portid,
1071 ovs_nla_free_flow_actions_rcu(old_acts);
1072 ovs_flow_free(new_flow, false);
1076 ovs_notify(&dp_flow_genl_family, reply, info);
1085 ovs_nla_free_flow_actions(acts);
1089 ovs_flow_free(new_flow, false);
1094 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1095 static struct sw_flow_actions *get_flow_actions(struct net *net,
1096 const struct nlattr *a,
1097 const struct sw_flow_key *key,
1098 const struct sw_flow_mask *mask,
1101 struct sw_flow_actions *acts;
1102 struct sw_flow_key masked_key;
1105 ovs_flow_mask_key(&masked_key, key, true, mask);
1106 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1109 "Actions may not be safe on all matching packets");
1110 return ERR_PTR(error);
1116 /* Factor out match-init and action-copy to avoid
1117 * "Wframe-larger-than=1024" warning. Because mask is only
1118 * used to get actions, we new a function to save some
1121 * If there are not key and action attrs, we return 0
1122 * directly. In the case, the caller will also not use the
1123 * match as before. If there is action attr, we try to get
1124 * actions and save them to *acts. Before returning from
1125 * the function, we reset the match->mask pointer. Because
1126 * we should not to return match object with dangling reference
1129 static int ovs_nla_init_match_and_action(struct net *net,
1130 struct sw_flow_match *match,
1131 struct sw_flow_key *key,
1133 struct sw_flow_actions **acts,
1136 struct sw_flow_mask mask;
1139 if (a[OVS_FLOW_ATTR_KEY]) {
1140 ovs_match_init(match, key, true, &mask);
1141 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1142 a[OVS_FLOW_ATTR_MASK], log);
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,
1157 if (IS_ERR(*acts)) {
1158 error = PTR_ERR(*acts);
1163 /* On success, error is 0. */
1169 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1171 struct net *net = sock_net(skb->sk);
1172 struct nlattr **a = info->attrs;
1173 struct ovs_header *ovs_header = info->userhdr;
1174 struct sw_flow_key key;
1175 struct sw_flow *flow;
1176 struct sk_buff *reply = NULL;
1177 struct datapath *dp;
1178 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1179 struct sw_flow_match match;
1180 struct sw_flow_id sfid;
1181 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1183 bool log = !a[OVS_FLOW_ATTR_PROBE];
1186 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1187 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1189 "Flow set message rejected, Key attribute missing.");
1193 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1199 /* Can allocate before locking if have acts. */
1200 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1202 if (IS_ERR(reply)) {
1203 error = PTR_ERR(reply);
1204 goto err_kfree_acts;
1209 dp = get_dp(net, ovs_header->dp_ifindex);
1210 if (unlikely(!dp)) {
1212 goto err_unlock_ovs;
1214 /* Check that the flow exists. */
1216 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1218 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1219 if (unlikely(!flow)) {
1221 goto err_unlock_ovs;
1224 /* Update actions, if present. */
1226 old_acts = ovsl_dereference(flow->sf_acts);
1227 rcu_assign_pointer(flow->sf_acts, acts);
1229 if (unlikely(reply)) {
1230 error = ovs_flow_cmd_fill_info(flow,
1231 ovs_header->dp_ifindex,
1232 reply, info->snd_portid,
1239 /* Could not alloc without acts before locking. */
1240 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1241 info, OVS_FLOW_CMD_NEW, false,
1244 if (IS_ERR(reply)) {
1245 error = PTR_ERR(reply);
1246 goto err_unlock_ovs;
1251 if (a[OVS_FLOW_ATTR_CLEAR])
1252 ovs_flow_stats_clear(flow);
1256 ovs_notify(&dp_flow_genl_family, reply, info);
1258 ovs_nla_free_flow_actions_rcu(old_acts);
1266 ovs_nla_free_flow_actions(acts);
1271 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1273 struct nlattr **a = info->attrs;
1274 struct ovs_header *ovs_header = info->userhdr;
1275 struct net *net = sock_net(skb->sk);
1276 struct sw_flow_key key;
1277 struct sk_buff *reply;
1278 struct sw_flow *flow;
1279 struct datapath *dp;
1280 struct sw_flow_match match;
1281 struct sw_flow_id ufid;
1282 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1284 bool log = !a[OVS_FLOW_ATTR_PROBE];
1287 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1288 if (a[OVS_FLOW_ATTR_KEY]) {
1289 ovs_match_init(&match, &key, true, NULL);
1290 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1292 } else if (!ufid_present) {
1294 "Flow get message rejected, Key attribute missing.");
1301 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1308 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1310 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1316 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1317 OVS_FLOW_CMD_NEW, true, ufid_flags);
1318 if (IS_ERR(reply)) {
1319 err = PTR_ERR(reply);
1324 return genlmsg_reply(reply, info);
1330 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1332 struct nlattr **a = info->attrs;
1333 struct ovs_header *ovs_header = info->userhdr;
1334 struct net *net = sock_net(skb->sk);
1335 struct sw_flow_key key;
1336 struct sk_buff *reply;
1337 struct sw_flow *flow = NULL;
1338 struct datapath *dp;
1339 struct sw_flow_match match;
1340 struct sw_flow_id ufid;
1341 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1343 bool log = !a[OVS_FLOW_ATTR_PROBE];
1346 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1347 if (a[OVS_FLOW_ATTR_KEY]) {
1348 ovs_match_init(&match, &key, true, NULL);
1349 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1356 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1357 if (unlikely(!dp)) {
1362 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1363 err = ovs_flow_tbl_flush(&dp->table);
1368 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1370 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1371 if (unlikely(!flow)) {
1376 ovs_flow_tbl_remove(&dp->table, flow);
1379 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1380 &flow->id, info, false, ufid_flags);
1381 if (likely(reply)) {
1382 if (likely(!IS_ERR(reply))) {
1383 rcu_read_lock(); /*To keep RCU checker happy. */
1384 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1385 reply, info->snd_portid,
1390 if (WARN_ON_ONCE(err < 0)) {
1395 ovs_notify(&dp_flow_genl_family, reply, info);
1397 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1402 ovs_flow_free(flow, true);
1409 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1411 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1412 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1413 struct table_instance *ti;
1414 struct datapath *dp;
1418 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1419 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1422 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1425 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1431 ti = rcu_dereference(dp->table.ti);
1433 struct sw_flow *flow;
1436 bucket = cb->args[0];
1438 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1442 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1443 NETLINK_CB(cb->skb).portid,
1444 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1445 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1448 cb->args[0] = bucket;
1455 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1456 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1457 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1458 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1459 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1460 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1461 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1462 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1465 static const struct genl_ops dp_flow_genl_ops[] = {
1466 { .cmd = OVS_FLOW_CMD_NEW,
1467 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1468 .policy = flow_policy,
1469 .doit = ovs_flow_cmd_new
1471 { .cmd = OVS_FLOW_CMD_DEL,
1472 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1473 .policy = flow_policy,
1474 .doit = ovs_flow_cmd_del
1476 { .cmd = OVS_FLOW_CMD_GET,
1477 .flags = 0, /* OK for unprivileged users. */
1478 .policy = flow_policy,
1479 .doit = ovs_flow_cmd_get,
1480 .dumpit = ovs_flow_cmd_dump
1482 { .cmd = OVS_FLOW_CMD_SET,
1483 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1484 .policy = flow_policy,
1485 .doit = ovs_flow_cmd_set,
1489 static struct genl_family dp_flow_genl_family __ro_after_init = {
1490 .hdrsize = sizeof(struct ovs_header),
1491 .name = OVS_FLOW_FAMILY,
1492 .version = OVS_FLOW_VERSION,
1493 .maxattr = OVS_FLOW_ATTR_MAX,
1495 .parallel_ops = true,
1496 .ops = dp_flow_genl_ops,
1497 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1498 .mcgrps = &ovs_dp_flow_multicast_group,
1500 .module = THIS_MODULE,
1503 static size_t ovs_dp_cmd_msg_size(void)
1505 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1507 msgsize += nla_total_size(IFNAMSIZ);
1508 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1509 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1510 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1515 /* Called with ovs_mutex. */
1516 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1517 u32 portid, u32 seq, u32 flags, u8 cmd)
1519 struct ovs_header *ovs_header;
1520 struct ovs_dp_stats dp_stats;
1521 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1524 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1529 ovs_header->dp_ifindex = get_dpifindex(dp);
1531 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1533 goto nla_put_failure;
1535 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1536 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1537 &dp_stats, OVS_DP_ATTR_PAD))
1538 goto nla_put_failure;
1540 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1541 sizeof(struct ovs_dp_megaflow_stats),
1542 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1543 goto nla_put_failure;
1545 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1546 goto nla_put_failure;
1548 genlmsg_end(skb, ovs_header);
1552 genlmsg_cancel(skb, ovs_header);
1557 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1559 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1562 /* Called with rcu_read_lock or ovs_mutex. */
1563 static struct datapath *lookup_datapath(struct net *net,
1564 const struct ovs_header *ovs_header,
1565 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1567 struct datapath *dp;
1569 if (!a[OVS_DP_ATTR_NAME])
1570 dp = get_dp(net, ovs_header->dp_ifindex);
1572 struct vport *vport;
1574 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1575 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1577 return dp ? dp : ERR_PTR(-ENODEV);
1580 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1582 struct datapath *dp;
1584 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1588 pr_warn("%s: Dropping previously announced user features\n",
1590 dp->user_features = 0;
1593 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1595 if (a[OVS_DP_ATTR_USER_FEATURES])
1596 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1599 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1601 struct nlattr **a = info->attrs;
1602 struct vport_parms parms;
1603 struct sk_buff *reply;
1604 struct datapath *dp;
1605 struct vport *vport;
1606 struct ovs_net *ovs_net;
1610 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1613 reply = ovs_dp_cmd_alloc_info();
1618 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1620 goto err_free_reply;
1622 ovs_dp_set_net(dp, sock_net(skb->sk));
1624 /* Allocate table. */
1625 err = ovs_flow_tbl_init(&dp->table);
1629 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1630 if (!dp->stats_percpu) {
1632 goto err_destroy_table;
1635 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1639 goto err_destroy_percpu;
1642 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1643 INIT_HLIST_HEAD(&dp->ports[i]);
1645 /* Set up our datapath device. */
1646 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1647 parms.type = OVS_VPORT_TYPE_INTERNAL;
1648 parms.options = NULL;
1650 parms.port_no = OVSP_LOCAL;
1651 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1653 ovs_dp_change(dp, a);
1655 /* So far only local changes have been made, now need the lock. */
1658 vport = new_vport(&parms);
1659 if (IS_ERR(vport)) {
1660 err = PTR_ERR(vport);
1664 if (err == -EEXIST) {
1665 /* An outdated user space instance that does not understand
1666 * the concept of user_features has attempted to create a new
1667 * datapath and is likely to reuse it. Drop all user features.
1669 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1670 ovs_dp_reset_user_features(skb, info);
1673 goto err_destroy_ports_array;
1676 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1677 info->snd_seq, 0, OVS_DP_CMD_NEW);
1680 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1681 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1685 ovs_notify(&dp_datapath_genl_family, reply, info);
1688 err_destroy_ports_array:
1692 free_percpu(dp->stats_percpu);
1694 ovs_flow_tbl_destroy(&dp->table);
1703 /* Called with ovs_mutex. */
1704 static void __dp_destroy(struct datapath *dp)
1708 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1709 struct vport *vport;
1710 struct hlist_node *n;
1712 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1713 if (vport->port_no != OVSP_LOCAL)
1714 ovs_dp_detach_port(vport);
1717 list_del_rcu(&dp->list_node);
1719 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1720 * all ports in datapath are destroyed first before freeing datapath.
1722 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1724 /* RCU destroy the flow table */
1725 call_rcu(&dp->rcu, destroy_dp_rcu);
1728 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1730 struct sk_buff *reply;
1731 struct datapath *dp;
1734 reply = ovs_dp_cmd_alloc_info();
1739 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1742 goto err_unlock_free;
1744 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1745 info->snd_seq, 0, OVS_DP_CMD_DEL);
1751 ovs_notify(&dp_datapath_genl_family, reply, info);
1761 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1763 struct sk_buff *reply;
1764 struct datapath *dp;
1767 reply = ovs_dp_cmd_alloc_info();
1772 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1775 goto err_unlock_free;
1777 ovs_dp_change(dp, info->attrs);
1779 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1780 info->snd_seq, 0, OVS_DP_CMD_NEW);
1784 ovs_notify(&dp_datapath_genl_family, reply, info);
1794 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1796 struct sk_buff *reply;
1797 struct datapath *dp;
1800 reply = ovs_dp_cmd_alloc_info();
1805 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1808 goto err_unlock_free;
1810 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1811 info->snd_seq, 0, OVS_DP_CMD_NEW);
1815 return genlmsg_reply(reply, info);
1823 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1825 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1826 struct datapath *dp;
1827 int skip = cb->args[0];
1831 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1833 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1834 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1835 OVS_DP_CMD_NEW) < 0)
1846 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1847 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1848 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1849 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1852 static const struct genl_ops dp_datapath_genl_ops[] = {
1853 { .cmd = OVS_DP_CMD_NEW,
1854 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1855 .policy = datapath_policy,
1856 .doit = ovs_dp_cmd_new
1858 { .cmd = OVS_DP_CMD_DEL,
1859 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1860 .policy = datapath_policy,
1861 .doit = ovs_dp_cmd_del
1863 { .cmd = OVS_DP_CMD_GET,
1864 .flags = 0, /* OK for unprivileged users. */
1865 .policy = datapath_policy,
1866 .doit = ovs_dp_cmd_get,
1867 .dumpit = ovs_dp_cmd_dump
1869 { .cmd = OVS_DP_CMD_SET,
1870 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1871 .policy = datapath_policy,
1872 .doit = ovs_dp_cmd_set,
1876 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1877 .hdrsize = sizeof(struct ovs_header),
1878 .name = OVS_DATAPATH_FAMILY,
1879 .version = OVS_DATAPATH_VERSION,
1880 .maxattr = OVS_DP_ATTR_MAX,
1882 .parallel_ops = true,
1883 .ops = dp_datapath_genl_ops,
1884 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1885 .mcgrps = &ovs_dp_datapath_multicast_group,
1887 .module = THIS_MODULE,
1890 /* Called with ovs_mutex or RCU read lock. */
1891 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1892 u32 portid, u32 seq, u32 flags, u8 cmd)
1894 struct ovs_header *ovs_header;
1895 struct ovs_vport_stats vport_stats;
1898 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1903 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1905 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1906 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1907 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1908 ovs_vport_name(vport)))
1909 goto nla_put_failure;
1911 ovs_vport_get_stats(vport, &vport_stats);
1912 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1913 sizeof(struct ovs_vport_stats), &vport_stats,
1914 OVS_VPORT_ATTR_PAD))
1915 goto nla_put_failure;
1917 if (ovs_vport_get_upcall_portids(vport, skb))
1918 goto nla_put_failure;
1920 err = ovs_vport_get_options(vport, skb);
1921 if (err == -EMSGSIZE)
1924 genlmsg_end(skb, ovs_header);
1930 genlmsg_cancel(skb, ovs_header);
1934 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1936 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1939 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1940 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1943 struct sk_buff *skb;
1946 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1948 return ERR_PTR(-ENOMEM);
1950 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1956 /* Called with ovs_mutex or RCU read lock. */
1957 static struct vport *lookup_vport(struct net *net,
1958 const struct ovs_header *ovs_header,
1959 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1961 struct datapath *dp;
1962 struct vport *vport;
1964 if (a[OVS_VPORT_ATTR_NAME]) {
1965 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1967 return ERR_PTR(-ENODEV);
1968 if (ovs_header->dp_ifindex &&
1969 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1970 return ERR_PTR(-ENODEV);
1972 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1973 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1975 if (port_no >= DP_MAX_PORTS)
1976 return ERR_PTR(-EFBIG);
1978 dp = get_dp(net, ovs_header->dp_ifindex);
1980 return ERR_PTR(-ENODEV);
1982 vport = ovs_vport_ovsl_rcu(dp, port_no);
1984 return ERR_PTR(-ENODEV);
1987 return ERR_PTR(-EINVAL);
1990 /* Called with ovs_mutex */
1991 static void update_headroom(struct datapath *dp)
1993 unsigned dev_headroom, max_headroom = 0;
1994 struct net_device *dev;
1995 struct vport *vport;
1998 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1999 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2001 dev_headroom = netdev_get_fwd_headroom(dev);
2002 if (dev_headroom > max_headroom)
2003 max_headroom = dev_headroom;
2007 dp->max_headroom = max_headroom;
2008 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2009 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2010 netdev_set_rx_headroom(vport->dev, max_headroom);
2013 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2015 struct nlattr **a = info->attrs;
2016 struct ovs_header *ovs_header = info->userhdr;
2017 struct vport_parms parms;
2018 struct sk_buff *reply;
2019 struct vport *vport;
2020 struct datapath *dp;
2024 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2025 !a[OVS_VPORT_ATTR_UPCALL_PID])
2028 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2029 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2030 if (port_no >= DP_MAX_PORTS)
2033 reply = ovs_vport_cmd_alloc_info();
2039 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2042 goto exit_unlock_free;
2045 vport = ovs_vport_ovsl(dp, port_no);
2048 goto exit_unlock_free;
2050 for (port_no = 1; ; port_no++) {
2051 if (port_no >= DP_MAX_PORTS) {
2053 goto exit_unlock_free;
2055 vport = ovs_vport_ovsl(dp, port_no);
2061 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2062 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2063 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2065 parms.port_no = port_no;
2066 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2068 vport = new_vport(&parms);
2069 err = PTR_ERR(vport);
2070 if (IS_ERR(vport)) {
2073 goto exit_unlock_free;
2076 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2077 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2079 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2080 update_headroom(dp);
2082 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2087 ovs_notify(&dp_vport_genl_family, reply, info);
2096 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2098 struct nlattr **a = info->attrs;
2099 struct sk_buff *reply;
2100 struct vport *vport;
2103 reply = ovs_vport_cmd_alloc_info();
2108 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2109 err = PTR_ERR(vport);
2111 goto exit_unlock_free;
2113 if (a[OVS_VPORT_ATTR_TYPE] &&
2114 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2116 goto exit_unlock_free;
2119 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2120 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2122 goto exit_unlock_free;
2126 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2127 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2129 err = ovs_vport_set_upcall_portids(vport, ids);
2131 goto exit_unlock_free;
2134 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2135 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2139 ovs_notify(&dp_vport_genl_family, reply, info);
2148 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2150 bool must_update_headroom = false;
2151 struct nlattr **a = info->attrs;
2152 struct sk_buff *reply;
2153 struct datapath *dp;
2154 struct vport *vport;
2157 reply = ovs_vport_cmd_alloc_info();
2162 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2163 err = PTR_ERR(vport);
2165 goto exit_unlock_free;
2167 if (vport->port_no == OVSP_LOCAL) {
2169 goto exit_unlock_free;
2172 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2173 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2176 /* the vport deletion may trigger dp headroom update */
2178 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2179 must_update_headroom = true;
2180 netdev_reset_rx_headroom(vport->dev);
2181 ovs_dp_detach_port(vport);
2183 if (must_update_headroom)
2184 update_headroom(dp);
2187 ovs_notify(&dp_vport_genl_family, reply, info);
2196 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2198 struct nlattr **a = info->attrs;
2199 struct ovs_header *ovs_header = info->userhdr;
2200 struct sk_buff *reply;
2201 struct vport *vport;
2204 reply = ovs_vport_cmd_alloc_info();
2209 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2210 err = PTR_ERR(vport);
2212 goto exit_unlock_free;
2213 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2214 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2218 return genlmsg_reply(reply, info);
2226 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2228 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2229 struct datapath *dp;
2230 int bucket = cb->args[0], skip = cb->args[1];
2234 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2239 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2240 struct vport *vport;
2243 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2245 ovs_vport_cmd_fill_info(vport, skb,
2246 NETLINK_CB(cb->skb).portid,
2249 OVS_VPORT_CMD_NEW) < 0)
2265 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2266 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2267 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2268 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2269 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2270 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2271 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2274 static const struct genl_ops dp_vport_genl_ops[] = {
2275 { .cmd = OVS_VPORT_CMD_NEW,
2276 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2277 .policy = vport_policy,
2278 .doit = ovs_vport_cmd_new
2280 { .cmd = OVS_VPORT_CMD_DEL,
2281 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2282 .policy = vport_policy,
2283 .doit = ovs_vport_cmd_del
2285 { .cmd = OVS_VPORT_CMD_GET,
2286 .flags = 0, /* OK for unprivileged users. */
2287 .policy = vport_policy,
2288 .doit = ovs_vport_cmd_get,
2289 .dumpit = ovs_vport_cmd_dump
2291 { .cmd = OVS_VPORT_CMD_SET,
2292 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2293 .policy = vport_policy,
2294 .doit = ovs_vport_cmd_set,
2298 struct genl_family dp_vport_genl_family __ro_after_init = {
2299 .hdrsize = sizeof(struct ovs_header),
2300 .name = OVS_VPORT_FAMILY,
2301 .version = OVS_VPORT_VERSION,
2302 .maxattr = OVS_VPORT_ATTR_MAX,
2304 .parallel_ops = true,
2305 .ops = dp_vport_genl_ops,
2306 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2307 .mcgrps = &ovs_dp_vport_multicast_group,
2309 .module = THIS_MODULE,
2312 static struct genl_family * const dp_genl_families[] = {
2313 &dp_datapath_genl_family,
2314 &dp_vport_genl_family,
2315 &dp_flow_genl_family,
2316 &dp_packet_genl_family,
2319 static void dp_unregister_genl(int n_families)
2323 for (i = 0; i < n_families; i++)
2324 genl_unregister_family(dp_genl_families[i]);
2327 static int __init dp_register_genl(void)
2332 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2334 err = genl_register_family(dp_genl_families[i]);
2342 dp_unregister_genl(i);
2346 static int __net_init ovs_init_net(struct net *net)
2348 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2350 INIT_LIST_HEAD(&ovs_net->dps);
2351 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2356 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2357 struct list_head *head)
2359 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2360 struct datapath *dp;
2362 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2365 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2366 struct vport *vport;
2368 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2369 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2372 if (dev_net(vport->dev) == dnet)
2373 list_add(&vport->detach_list, head);
2379 static void __net_exit ovs_exit_net(struct net *dnet)
2381 struct datapath *dp, *dp_next;
2382 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2383 struct vport *vport, *vport_next;
2389 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2394 list_vports_from_net(net, dnet, &head);
2397 /* Detach all vports from given namespace. */
2398 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2399 list_del(&vport->detach_list);
2400 ovs_dp_detach_port(vport);
2405 cancel_work_sync(&ovs_net->dp_notify_work);
2408 static struct pernet_operations ovs_net_ops = {
2409 .init = ovs_init_net,
2410 .exit = ovs_exit_net,
2412 .size = sizeof(struct ovs_net),
2415 static int __init dp_init(void)
2419 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2421 pr_info("Open vSwitch switching datapath\n");
2423 err = action_fifos_init();
2427 err = ovs_internal_dev_rtnl_link_register();
2429 goto error_action_fifos_exit;
2431 err = ovs_flow_init();
2433 goto error_unreg_rtnl_link;
2435 err = ovs_vport_init();
2437 goto error_flow_exit;
2439 err = register_pernet_device(&ovs_net_ops);
2441 goto error_vport_exit;
2443 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2445 goto error_netns_exit;
2447 err = ovs_netdev_init();
2449 goto error_unreg_notifier;
2451 err = dp_register_genl();
2453 goto error_unreg_netdev;
2459 error_unreg_notifier:
2460 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2462 unregister_pernet_device(&ovs_net_ops);
2467 error_unreg_rtnl_link:
2468 ovs_internal_dev_rtnl_link_unregister();
2469 error_action_fifos_exit:
2470 action_fifos_exit();
2475 static void dp_cleanup(void)
2477 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2479 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2480 unregister_pernet_device(&ovs_net_ops);
2484 ovs_internal_dev_rtnl_link_unregister();
2485 action_fifos_exit();
2488 module_init(dp_init);
2489 module_exit(dp_cleanup);
2491 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2492 MODULE_LICENSE("GPL");
2493 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2494 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2495 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2496 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);