2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
56 #include "flow_table.h"
57 #include "flow_netlink.h"
59 #include "vport-internal_dev.h"
60 #include "vport-netdev.h"
62 unsigned int ovs_net_id __read_mostly;
64 static struct genl_family dp_packet_genl_family;
65 static struct genl_family dp_flow_genl_family;
66 static struct genl_family dp_datapath_genl_family;
68 static const struct nla_policy flow_policy[];
70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
71 .name = OVS_FLOW_MCGROUP,
74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
75 .name = OVS_DATAPATH_MCGROUP,
78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
79 .name = OVS_VPORT_MCGROUP,
82 /* Check if need to build a reply message.
83 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
87 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
88 genl_has_listeners(family, genl_info_net(info), group);
91 static void ovs_notify(struct genl_family *family,
92 struct sk_buff *skb, struct genl_info *info)
94 genl_notify(family, skb, info, 0, GFP_KERNEL);
100 * All writes e.g. Writes to device state (add/remove datapath, port, set
101 * operations on vports, etc.), Writes to other state (flow table
102 * modifications, set miscellaneous datapath parameters, etc.) are protected
105 * Reads are protected by RCU.
107 * There are a few special cases (mostly stats) that have their own
108 * synchronization but they nest under all of above and don't interact with
111 * The RTNL lock nests inside ovs_mutex.
114 static DEFINE_MUTEX(ovs_mutex);
118 mutex_lock(&ovs_mutex);
121 void ovs_unlock(void)
123 mutex_unlock(&ovs_mutex);
126 #ifdef CONFIG_LOCKDEP
127 int lockdep_ovsl_is_held(void)
130 return lockdep_is_held(&ovs_mutex);
136 static struct vport *new_vport(const struct vport_parms *);
137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
138 const struct sw_flow_key *,
139 const struct dp_upcall_info *,
141 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
142 const struct sw_flow_key *,
143 const struct dp_upcall_info *,
146 /* Must be called with rcu_read_lock or ovs_mutex. */
147 const char *ovs_dp_name(const struct datapath *dp)
149 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
150 return ovs_vport_name(vport);
153 static int get_dpifindex(const struct datapath *dp)
160 local = ovs_vport_rcu(dp, OVSP_LOCAL);
162 ifindex = local->dev->ifindex;
171 static void destroy_dp_rcu(struct rcu_head *rcu)
173 struct datapath *dp = container_of(rcu, struct datapath, rcu);
175 ovs_flow_tbl_destroy(&dp->table);
176 free_percpu(dp->stats_percpu);
182 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
185 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
188 /* Called with ovs_mutex or RCU read lock. */
189 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
192 struct hlist_head *head;
194 head = vport_hash_bucket(dp, port_no);
195 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
196 if (vport->port_no == port_no)
202 /* Called with ovs_mutex. */
203 static struct vport *new_vport(const struct vport_parms *parms)
207 vport = ovs_vport_add(parms);
208 if (!IS_ERR(vport)) {
209 struct datapath *dp = parms->dp;
210 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
212 hlist_add_head_rcu(&vport->dp_hash_node, head);
217 void ovs_dp_detach_port(struct vport *p)
221 /* First drop references to device. */
222 hlist_del_rcu(&p->dp_hash_node);
224 /* Then destroy it. */
228 /* Must be called with rcu_read_lock. */
229 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
231 const struct vport *p = OVS_CB(skb)->input_vport;
232 struct datapath *dp = p->dp;
233 struct sw_flow *flow;
234 struct sw_flow_actions *sf_acts;
235 struct dp_stats_percpu *stats;
239 stats = this_cpu_ptr(dp->stats_percpu);
242 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
243 if (unlikely(!flow)) {
244 struct dp_upcall_info upcall;
247 memset(&upcall, 0, sizeof(upcall));
248 upcall.cmd = OVS_PACKET_CMD_MISS;
249 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
250 upcall.mru = OVS_CB(skb)->mru;
251 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
256 stats_counter = &stats->n_missed;
260 ovs_flow_stats_update(flow, key->tp.flags, skb);
261 sf_acts = rcu_dereference(flow->sf_acts);
262 ovs_execute_actions(dp, skb, sf_acts, key);
264 stats_counter = &stats->n_hit;
267 /* Update datapath statistics. */
268 u64_stats_update_begin(&stats->syncp);
270 stats->n_mask_hit += n_mask_hit;
271 u64_stats_update_end(&stats->syncp);
274 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
275 const struct sw_flow_key *key,
276 const struct dp_upcall_info *upcall_info,
279 struct dp_stats_percpu *stats;
282 if (upcall_info->portid == 0) {
287 if (!skb_is_gso(skb))
288 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
290 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
297 stats = this_cpu_ptr(dp->stats_percpu);
299 u64_stats_update_begin(&stats->syncp);
301 u64_stats_update_end(&stats->syncp);
306 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
307 const struct sw_flow_key *key,
308 const struct dp_upcall_info *upcall_info,
311 unsigned int gso_type = skb_shinfo(skb)->gso_type;
312 struct sw_flow_key later_key;
313 struct sk_buff *segs, *nskb;
316 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
317 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
319 return PTR_ERR(segs);
323 if (gso_type & SKB_GSO_UDP) {
324 /* The initial flow key extracted by ovs_flow_key_extract()
325 * in this case is for a first fragment, so we need to
326 * properly mark later fragments.
329 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
332 /* Queue all of the segments. */
335 if (gso_type & SKB_GSO_UDP && skb != segs)
338 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
342 } while ((skb = skb->next));
344 /* Free all of the segments. */
352 } while ((skb = nskb));
356 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
357 unsigned int hdrlen, int actions_attrlen)
359 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
360 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
361 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
362 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
364 /* OVS_PACKET_ATTR_USERDATA */
365 if (upcall_info->userdata)
366 size += NLA_ALIGN(upcall_info->userdata->nla_len);
368 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
369 if (upcall_info->egress_tun_info)
370 size += nla_total_size(ovs_tun_key_attr_size());
372 /* OVS_PACKET_ATTR_ACTIONS */
373 if (upcall_info->actions_len)
374 size += nla_total_size(actions_attrlen);
376 /* OVS_PACKET_ATTR_MRU */
377 if (upcall_info->mru)
378 size += nla_total_size(sizeof(upcall_info->mru));
383 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
385 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
386 size_t plen = NLA_ALIGN(skb->len) - skb->len;
389 skb_put_zero(skb, plen);
393 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
394 const struct sw_flow_key *key,
395 const struct dp_upcall_info *upcall_info,
398 struct ovs_header *upcall;
399 struct sk_buff *nskb = NULL;
400 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
406 dp_ifindex = get_dpifindex(dp);
410 if (skb_vlan_tag_present(skb)) {
411 nskb = skb_clone(skb, GFP_ATOMIC);
415 nskb = __vlan_hwaccel_push_inside(nskb);
422 if (nla_attr_size(skb->len) > USHRT_MAX) {
427 /* Complete checksum if needed */
428 if (skb->ip_summed == CHECKSUM_PARTIAL &&
429 (err = skb_csum_hwoffload_help(skb, 0)))
432 /* Older versions of OVS user space enforce alignment of the last
433 * Netlink attribute to NLA_ALIGNTO which would require extensive
434 * padding logic. Only perform zerocopy if padding is not required.
436 if (dp->user_features & OVS_DP_F_UNALIGNED)
437 hlen = skb_zerocopy_headlen(skb);
441 len = upcall_msg_size(upcall_info, hlen - cutlen,
442 OVS_CB(skb)->acts_origlen);
443 user_skb = genlmsg_new(len, GFP_ATOMIC);
449 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
450 0, upcall_info->cmd);
455 upcall->dp_ifindex = dp_ifindex;
457 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
461 if (upcall_info->userdata)
462 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
463 nla_len(upcall_info->userdata),
464 nla_data(upcall_info->userdata));
466 if (upcall_info->egress_tun_info) {
467 nla = nla_nest_start_noflag(user_skb,
468 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
473 err = ovs_nla_put_tunnel_info(user_skb,
474 upcall_info->egress_tun_info);
478 nla_nest_end(user_skb, nla);
481 if (upcall_info->actions_len) {
482 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
487 err = ovs_nla_put_actions(upcall_info->actions,
488 upcall_info->actions_len,
491 nla_nest_end(user_skb, nla);
493 nla_nest_cancel(user_skb, nla);
496 /* Add OVS_PACKET_ATTR_MRU */
497 if (upcall_info->mru) {
498 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
503 pad_packet(dp, user_skb);
506 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
508 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
513 pad_packet(dp, user_skb);
516 /* Only reserve room for attribute header, packet data is added
517 * in skb_zerocopy() */
518 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
522 nla->nla_len = nla_attr_size(skb->len - cutlen);
524 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
528 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
529 pad_packet(dp, user_skb);
531 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
533 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
543 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
545 struct ovs_header *ovs_header = info->userhdr;
546 struct net *net = sock_net(skb->sk);
547 struct nlattr **a = info->attrs;
548 struct sw_flow_actions *acts;
549 struct sk_buff *packet;
550 struct sw_flow *flow;
551 struct sw_flow_actions *sf_acts;
553 struct vport *input_vport;
557 bool log = !a[OVS_PACKET_ATTR_PROBE];
560 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
561 !a[OVS_PACKET_ATTR_ACTIONS])
564 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
565 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
569 skb_reserve(packet, NET_IP_ALIGN);
571 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
573 /* Set packet's mru */
574 if (a[OVS_PACKET_ATTR_MRU]) {
575 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
576 packet->ignore_df = 1;
578 OVS_CB(packet)->mru = mru;
580 /* Build an sw_flow for sending this packet. */
581 flow = ovs_flow_alloc();
586 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
587 packet, &flow->key, log);
591 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
592 &flow->key, &acts, log);
596 rcu_assign_pointer(flow->sf_acts, acts);
597 packet->priority = flow->key.phy.priority;
598 packet->mark = flow->key.phy.skb_mark;
601 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
606 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
608 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
613 packet->dev = input_vport->dev;
614 OVS_CB(packet)->input_vport = input_vport;
615 sf_acts = rcu_dereference(flow->sf_acts);
618 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
622 ovs_flow_free(flow, false);
628 ovs_flow_free(flow, false);
635 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
636 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
637 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
638 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
639 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
640 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
643 static const struct genl_ops dp_packet_genl_ops[] = {
644 { .cmd = OVS_PACKET_CMD_EXECUTE,
645 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
646 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
647 .doit = ovs_packet_cmd_execute
651 static struct genl_family dp_packet_genl_family __ro_after_init = {
652 .hdrsize = sizeof(struct ovs_header),
653 .name = OVS_PACKET_FAMILY,
654 .version = OVS_PACKET_VERSION,
655 .maxattr = OVS_PACKET_ATTR_MAX,
656 .policy = packet_policy,
658 .parallel_ops = true,
659 .ops = dp_packet_genl_ops,
660 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
661 .module = THIS_MODULE,
664 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
665 struct ovs_dp_megaflow_stats *mega_stats)
669 memset(mega_stats, 0, sizeof(*mega_stats));
671 stats->n_flows = ovs_flow_tbl_count(&dp->table);
672 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
674 stats->n_hit = stats->n_missed = stats->n_lost = 0;
676 for_each_possible_cpu(i) {
677 const struct dp_stats_percpu *percpu_stats;
678 struct dp_stats_percpu local_stats;
681 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
684 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
685 local_stats = *percpu_stats;
686 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
688 stats->n_hit += local_stats.n_hit;
689 stats->n_missed += local_stats.n_missed;
690 stats->n_lost += local_stats.n_lost;
691 mega_stats->n_mask_hit += local_stats.n_mask_hit;
695 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
697 return ovs_identifier_is_ufid(sfid) &&
698 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
701 static bool should_fill_mask(uint32_t ufid_flags)
703 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
706 static bool should_fill_actions(uint32_t ufid_flags)
708 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
711 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
712 const struct sw_flow_id *sfid,
715 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
717 /* OVS_FLOW_ATTR_UFID */
718 if (sfid && ovs_identifier_is_ufid(sfid))
719 len += nla_total_size(sfid->ufid_len);
721 /* OVS_FLOW_ATTR_KEY */
722 if (!sfid || should_fill_key(sfid, ufid_flags))
723 len += nla_total_size(ovs_key_attr_size());
725 /* OVS_FLOW_ATTR_MASK */
726 if (should_fill_mask(ufid_flags))
727 len += nla_total_size(ovs_key_attr_size());
729 /* OVS_FLOW_ATTR_ACTIONS */
730 if (should_fill_actions(ufid_flags))
731 len += nla_total_size(acts->orig_len);
734 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
735 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
736 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
739 /* Called with ovs_mutex or RCU read lock. */
740 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
743 struct ovs_flow_stats stats;
747 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
750 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
754 if (stats.n_packets &&
755 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
756 sizeof(struct ovs_flow_stats), &stats,
760 if ((u8)ntohs(tcp_flags) &&
761 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
767 /* Called with ovs_mutex or RCU read lock. */
768 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
769 struct sk_buff *skb, int skb_orig_len)
771 struct nlattr *start;
774 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
775 * this is the first flow to be dumped into 'skb'. This is unusual for
776 * Netlink but individual action lists can be longer than
777 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
778 * The userspace caller can always fetch the actions separately if it
779 * really wants them. (Most userspace callers in fact don't care.)
781 * This can only fail for dump operations because the skb is always
782 * properly sized for single flows.
784 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
786 const struct sw_flow_actions *sf_acts;
788 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
789 err = ovs_nla_put_actions(sf_acts->actions,
790 sf_acts->actions_len, skb);
793 nla_nest_end(skb, start);
798 nla_nest_cancel(skb, start);
800 } else if (skb_orig_len) {
807 /* Called with ovs_mutex or RCU read lock. */
808 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
809 struct sk_buff *skb, u32 portid,
810 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
812 const int skb_orig_len = skb->len;
813 struct ovs_header *ovs_header;
816 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
821 ovs_header->dp_ifindex = dp_ifindex;
823 err = ovs_nla_put_identifier(flow, skb);
827 if (should_fill_key(&flow->id, ufid_flags)) {
828 err = ovs_nla_put_masked_key(flow, skb);
833 if (should_fill_mask(ufid_flags)) {
834 err = ovs_nla_put_mask(flow, skb);
839 err = ovs_flow_cmd_fill_stats(flow, skb);
843 if (should_fill_actions(ufid_flags)) {
844 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
849 genlmsg_end(skb, ovs_header);
853 genlmsg_cancel(skb, ovs_header);
857 /* May not be called with RCU read lock. */
858 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
859 const struct sw_flow_id *sfid,
860 struct genl_info *info,
867 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
870 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
871 skb = genlmsg_new(len, GFP_KERNEL);
873 return ERR_PTR(-ENOMEM);
878 /* Called with ovs_mutex. */
879 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
881 struct genl_info *info, u8 cmd,
882 bool always, u32 ufid_flags)
887 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
888 &flow->id, info, always, ufid_flags);
889 if (IS_ERR_OR_NULL(skb))
892 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
893 info->snd_portid, info->snd_seq, 0,
899 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
901 struct net *net = sock_net(skb->sk);
902 struct nlattr **a = info->attrs;
903 struct ovs_header *ovs_header = info->userhdr;
904 struct sw_flow *flow = NULL, *new_flow;
905 struct sw_flow_mask mask;
906 struct sk_buff *reply;
908 struct sw_flow_actions *acts;
909 struct sw_flow_match match;
910 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
912 bool log = !a[OVS_FLOW_ATTR_PROBE];
914 /* Must have key and actions. */
916 if (!a[OVS_FLOW_ATTR_KEY]) {
917 OVS_NLERR(log, "Flow key attr not present in new flow.");
920 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
921 OVS_NLERR(log, "Flow actions attr not present in new flow.");
925 /* Most of the time we need to allocate a new flow, do it before
928 new_flow = ovs_flow_alloc();
929 if (IS_ERR(new_flow)) {
930 error = PTR_ERR(new_flow);
935 ovs_match_init(&match, &new_flow->key, false, &mask);
936 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
937 a[OVS_FLOW_ATTR_MASK], log);
941 /* Extract flow identifier. */
942 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
943 &new_flow->key, log);
947 /* unmasked key is needed to match when ufid is not used. */
948 if (ovs_identifier_is_key(&new_flow->id))
949 match.key = new_flow->id.unmasked_key;
951 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
953 /* Validate actions. */
954 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
955 &new_flow->key, &acts, log);
957 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
961 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
964 error = PTR_ERR(reply);
969 dp = get_dp(net, ovs_header->dp_ifindex);
975 /* Check if this is a duplicate flow */
976 if (ovs_identifier_is_ufid(&new_flow->id))
977 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
979 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
981 rcu_assign_pointer(new_flow->sf_acts, acts);
983 /* Put flow in bucket. */
984 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
985 if (unlikely(error)) {
990 if (unlikely(reply)) {
991 error = ovs_flow_cmd_fill_info(new_flow,
992 ovs_header->dp_ifindex,
993 reply, info->snd_portid,
1001 struct sw_flow_actions *old_acts;
1003 /* Bail out if we're not allowed to modify an existing flow.
1004 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1005 * because Generic Netlink treats the latter as a dump
1006 * request. We also accept NLM_F_EXCL in case that bug ever
1009 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1012 goto err_unlock_ovs;
1014 /* The flow identifier has to be the same for flow updates.
1015 * Look for any overlapping flow.
1017 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1018 if (ovs_identifier_is_key(&flow->id))
1019 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1021 else /* UFID matches but key is different */
1025 goto err_unlock_ovs;
1028 /* Update actions. */
1029 old_acts = ovsl_dereference(flow->sf_acts);
1030 rcu_assign_pointer(flow->sf_acts, acts);
1032 if (unlikely(reply)) {
1033 error = ovs_flow_cmd_fill_info(flow,
1034 ovs_header->dp_ifindex,
1035 reply, info->snd_portid,
1043 ovs_nla_free_flow_actions_rcu(old_acts);
1044 ovs_flow_free(new_flow, false);
1048 ovs_notify(&dp_flow_genl_family, reply, info);
1055 ovs_nla_free_flow_actions(acts);
1057 ovs_flow_free(new_flow, false);
1062 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1063 static struct sw_flow_actions *get_flow_actions(struct net *net,
1064 const struct nlattr *a,
1065 const struct sw_flow_key *key,
1066 const struct sw_flow_mask *mask,
1069 struct sw_flow_actions *acts;
1070 struct sw_flow_key masked_key;
1073 ovs_flow_mask_key(&masked_key, key, true, mask);
1074 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1077 "Actions may not be safe on all matching packets");
1078 return ERR_PTR(error);
1084 /* Factor out match-init and action-copy to avoid
1085 * "Wframe-larger-than=1024" warning. Because mask is only
1086 * used to get actions, we new a function to save some
1089 * If there are not key and action attrs, we return 0
1090 * directly. In the case, the caller will also not use the
1091 * match as before. If there is action attr, we try to get
1092 * actions and save them to *acts. Before returning from
1093 * the function, we reset the match->mask pointer. Because
1094 * we should not to return match object with dangling reference
1097 static int ovs_nla_init_match_and_action(struct net *net,
1098 struct sw_flow_match *match,
1099 struct sw_flow_key *key,
1101 struct sw_flow_actions **acts,
1104 struct sw_flow_mask mask;
1107 if (a[OVS_FLOW_ATTR_KEY]) {
1108 ovs_match_init(match, key, true, &mask);
1109 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1110 a[OVS_FLOW_ATTR_MASK], log);
1115 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1116 if (!a[OVS_FLOW_ATTR_KEY]) {
1118 "Flow key attribute not present in set flow.");
1123 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1125 if (IS_ERR(*acts)) {
1126 error = PTR_ERR(*acts);
1131 /* On success, error is 0. */
1137 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1139 struct net *net = sock_net(skb->sk);
1140 struct nlattr **a = info->attrs;
1141 struct ovs_header *ovs_header = info->userhdr;
1142 struct sw_flow_key key;
1143 struct sw_flow *flow;
1144 struct sk_buff *reply = NULL;
1145 struct datapath *dp;
1146 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1147 struct sw_flow_match match;
1148 struct sw_flow_id sfid;
1149 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1151 bool log = !a[OVS_FLOW_ATTR_PROBE];
1154 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1155 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1157 "Flow set message rejected, Key attribute missing.");
1161 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1167 /* Can allocate before locking if have acts. */
1168 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1170 if (IS_ERR(reply)) {
1171 error = PTR_ERR(reply);
1172 goto err_kfree_acts;
1177 dp = get_dp(net, ovs_header->dp_ifindex);
1178 if (unlikely(!dp)) {
1180 goto err_unlock_ovs;
1182 /* Check that the flow exists. */
1184 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1186 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1187 if (unlikely(!flow)) {
1189 goto err_unlock_ovs;
1192 /* Update actions, if present. */
1194 old_acts = ovsl_dereference(flow->sf_acts);
1195 rcu_assign_pointer(flow->sf_acts, acts);
1197 if (unlikely(reply)) {
1198 error = ovs_flow_cmd_fill_info(flow,
1199 ovs_header->dp_ifindex,
1200 reply, info->snd_portid,
1207 /* Could not alloc without acts before locking. */
1208 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1209 info, OVS_FLOW_CMD_SET, false,
1212 if (IS_ERR(reply)) {
1213 error = PTR_ERR(reply);
1214 goto err_unlock_ovs;
1219 if (a[OVS_FLOW_ATTR_CLEAR])
1220 ovs_flow_stats_clear(flow);
1224 ovs_notify(&dp_flow_genl_family, reply, info);
1226 ovs_nla_free_flow_actions_rcu(old_acts);
1234 ovs_nla_free_flow_actions(acts);
1239 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1241 struct nlattr **a = info->attrs;
1242 struct ovs_header *ovs_header = info->userhdr;
1243 struct net *net = sock_net(skb->sk);
1244 struct sw_flow_key key;
1245 struct sk_buff *reply;
1246 struct sw_flow *flow;
1247 struct datapath *dp;
1248 struct sw_flow_match match;
1249 struct sw_flow_id ufid;
1250 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1252 bool log = !a[OVS_FLOW_ATTR_PROBE];
1255 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1256 if (a[OVS_FLOW_ATTR_KEY]) {
1257 ovs_match_init(&match, &key, true, NULL);
1258 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1260 } else if (!ufid_present) {
1262 "Flow get message rejected, Key attribute missing.");
1269 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1276 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1278 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1284 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1285 OVS_FLOW_CMD_GET, true, ufid_flags);
1286 if (IS_ERR(reply)) {
1287 err = PTR_ERR(reply);
1292 return genlmsg_reply(reply, info);
1298 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1300 struct nlattr **a = info->attrs;
1301 struct ovs_header *ovs_header = info->userhdr;
1302 struct net *net = sock_net(skb->sk);
1303 struct sw_flow_key key;
1304 struct sk_buff *reply;
1305 struct sw_flow *flow = NULL;
1306 struct datapath *dp;
1307 struct sw_flow_match match;
1308 struct sw_flow_id ufid;
1309 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1311 bool log = !a[OVS_FLOW_ATTR_PROBE];
1314 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1315 if (a[OVS_FLOW_ATTR_KEY]) {
1316 ovs_match_init(&match, &key, true, NULL);
1317 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1324 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1325 if (unlikely(!dp)) {
1330 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1331 err = ovs_flow_tbl_flush(&dp->table);
1336 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1338 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1339 if (unlikely(!flow)) {
1344 ovs_flow_tbl_remove(&dp->table, flow);
1347 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1348 &flow->id, info, false, ufid_flags);
1349 if (likely(reply)) {
1350 if (likely(!IS_ERR(reply))) {
1351 rcu_read_lock(); /*To keep RCU checker happy. */
1352 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1353 reply, info->snd_portid,
1360 ovs_notify(&dp_flow_genl_family, reply, info);
1362 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1366 ovs_flow_free(flow, true);
1373 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1375 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1376 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1377 struct table_instance *ti;
1378 struct datapath *dp;
1382 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1383 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1386 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1389 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1395 ti = rcu_dereference(dp->table.ti);
1397 struct sw_flow *flow;
1400 bucket = cb->args[0];
1402 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1406 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1407 NETLINK_CB(cb->skb).portid,
1408 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1409 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1412 cb->args[0] = bucket;
1419 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1420 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1421 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1422 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1423 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1424 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1425 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1426 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1429 static const struct genl_ops dp_flow_genl_ops[] = {
1430 { .cmd = OVS_FLOW_CMD_NEW,
1431 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1432 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1433 .doit = ovs_flow_cmd_new
1435 { .cmd = OVS_FLOW_CMD_DEL,
1436 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1437 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1438 .doit = ovs_flow_cmd_del
1440 { .cmd = OVS_FLOW_CMD_GET,
1441 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1442 .flags = 0, /* OK for unprivileged users. */
1443 .doit = ovs_flow_cmd_get,
1444 .dumpit = ovs_flow_cmd_dump
1446 { .cmd = OVS_FLOW_CMD_SET,
1447 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1448 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1449 .doit = ovs_flow_cmd_set,
1453 static struct genl_family dp_flow_genl_family __ro_after_init = {
1454 .hdrsize = sizeof(struct ovs_header),
1455 .name = OVS_FLOW_FAMILY,
1456 .version = OVS_FLOW_VERSION,
1457 .maxattr = OVS_FLOW_ATTR_MAX,
1458 .policy = flow_policy,
1460 .parallel_ops = true,
1461 .ops = dp_flow_genl_ops,
1462 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1463 .mcgrps = &ovs_dp_flow_multicast_group,
1465 .module = THIS_MODULE,
1468 static size_t ovs_dp_cmd_msg_size(void)
1470 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1472 msgsize += nla_total_size(IFNAMSIZ);
1473 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1474 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1475 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1480 /* Called with ovs_mutex. */
1481 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1482 u32 portid, u32 seq, u32 flags, u8 cmd)
1484 struct ovs_header *ovs_header;
1485 struct ovs_dp_stats dp_stats;
1486 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1489 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1494 ovs_header->dp_ifindex = get_dpifindex(dp);
1496 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1498 goto nla_put_failure;
1500 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1501 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1502 &dp_stats, OVS_DP_ATTR_PAD))
1503 goto nla_put_failure;
1505 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1506 sizeof(struct ovs_dp_megaflow_stats),
1507 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1508 goto nla_put_failure;
1510 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1511 goto nla_put_failure;
1513 genlmsg_end(skb, ovs_header);
1517 genlmsg_cancel(skb, ovs_header);
1522 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1524 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1527 /* Called with rcu_read_lock or ovs_mutex. */
1528 static struct datapath *lookup_datapath(struct net *net,
1529 const struct ovs_header *ovs_header,
1530 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1532 struct datapath *dp;
1534 if (!a[OVS_DP_ATTR_NAME])
1535 dp = get_dp(net, ovs_header->dp_ifindex);
1537 struct vport *vport;
1539 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1540 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1542 return dp ? dp : ERR_PTR(-ENODEV);
1545 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1547 struct datapath *dp;
1549 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1553 WARN(dp->user_features, "Dropping previously announced user features\n");
1554 dp->user_features = 0;
1557 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1559 if (a[OVS_DP_ATTR_USER_FEATURES])
1560 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1563 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1565 struct nlattr **a = info->attrs;
1566 struct vport_parms parms;
1567 struct sk_buff *reply;
1568 struct datapath *dp;
1569 struct vport *vport;
1570 struct ovs_net *ovs_net;
1574 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1577 reply = ovs_dp_cmd_alloc_info();
1582 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1584 goto err_free_reply;
1586 ovs_dp_set_net(dp, sock_net(skb->sk));
1588 /* Allocate table. */
1589 err = ovs_flow_tbl_init(&dp->table);
1593 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1594 if (!dp->stats_percpu) {
1596 goto err_destroy_table;
1599 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1600 sizeof(struct hlist_head),
1604 goto err_destroy_percpu;
1607 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1608 INIT_HLIST_HEAD(&dp->ports[i]);
1610 err = ovs_meters_init(dp);
1612 goto err_destroy_ports_array;
1614 /* Set up our datapath device. */
1615 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1616 parms.type = OVS_VPORT_TYPE_INTERNAL;
1617 parms.options = NULL;
1619 parms.port_no = OVSP_LOCAL;
1620 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1622 ovs_dp_change(dp, a);
1624 /* So far only local changes have been made, now need the lock. */
1627 vport = new_vport(&parms);
1628 if (IS_ERR(vport)) {
1629 err = PTR_ERR(vport);
1633 if (err == -EEXIST) {
1634 /* An outdated user space instance that does not understand
1635 * the concept of user_features has attempted to create a new
1636 * datapath and is likely to reuse it. Drop all user features.
1638 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1639 ovs_dp_reset_user_features(skb, info);
1642 goto err_destroy_meters;
1645 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1646 info->snd_seq, 0, OVS_DP_CMD_NEW);
1649 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1650 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1654 ovs_notify(&dp_datapath_genl_family, reply, info);
1659 ovs_meters_exit(dp);
1660 err_destroy_ports_array:
1663 free_percpu(dp->stats_percpu);
1665 ovs_flow_tbl_destroy(&dp->table);
1674 /* Called with ovs_mutex. */
1675 static void __dp_destroy(struct datapath *dp)
1679 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1680 struct vport *vport;
1681 struct hlist_node *n;
1683 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1684 if (vport->port_no != OVSP_LOCAL)
1685 ovs_dp_detach_port(vport);
1688 list_del_rcu(&dp->list_node);
1690 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1691 * all ports in datapath are destroyed first before freeing datapath.
1693 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1695 /* RCU destroy the flow table */
1696 call_rcu(&dp->rcu, destroy_dp_rcu);
1699 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1701 struct sk_buff *reply;
1702 struct datapath *dp;
1705 reply = ovs_dp_cmd_alloc_info();
1710 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1713 goto err_unlock_free;
1715 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1716 info->snd_seq, 0, OVS_DP_CMD_DEL);
1722 ovs_notify(&dp_datapath_genl_family, reply, info);
1732 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1734 struct sk_buff *reply;
1735 struct datapath *dp;
1738 reply = ovs_dp_cmd_alloc_info();
1743 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1746 goto err_unlock_free;
1748 ovs_dp_change(dp, info->attrs);
1750 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1751 info->snd_seq, 0, OVS_DP_CMD_SET);
1755 ovs_notify(&dp_datapath_genl_family, reply, info);
1765 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1767 struct sk_buff *reply;
1768 struct datapath *dp;
1771 reply = ovs_dp_cmd_alloc_info();
1776 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1779 goto err_unlock_free;
1781 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1782 info->snd_seq, 0, OVS_DP_CMD_GET);
1786 return genlmsg_reply(reply, info);
1794 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1796 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1797 struct datapath *dp;
1798 int skip = cb->args[0];
1802 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1804 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1805 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1806 OVS_DP_CMD_GET) < 0)
1817 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1818 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1819 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1820 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1823 static const struct genl_ops dp_datapath_genl_ops[] = {
1824 { .cmd = OVS_DP_CMD_NEW,
1825 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1826 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1827 .doit = ovs_dp_cmd_new
1829 { .cmd = OVS_DP_CMD_DEL,
1830 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1831 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1832 .doit = ovs_dp_cmd_del
1834 { .cmd = OVS_DP_CMD_GET,
1835 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1836 .flags = 0, /* OK for unprivileged users. */
1837 .doit = ovs_dp_cmd_get,
1838 .dumpit = ovs_dp_cmd_dump
1840 { .cmd = OVS_DP_CMD_SET,
1841 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1842 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1843 .doit = ovs_dp_cmd_set,
1847 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1848 .hdrsize = sizeof(struct ovs_header),
1849 .name = OVS_DATAPATH_FAMILY,
1850 .version = OVS_DATAPATH_VERSION,
1851 .maxattr = OVS_DP_ATTR_MAX,
1852 .policy = datapath_policy,
1854 .parallel_ops = true,
1855 .ops = dp_datapath_genl_ops,
1856 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1857 .mcgrps = &ovs_dp_datapath_multicast_group,
1859 .module = THIS_MODULE,
1862 /* Called with ovs_mutex or RCU read lock. */
1863 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1864 struct net *net, u32 portid, u32 seq,
1867 struct ovs_header *ovs_header;
1868 struct ovs_vport_stats vport_stats;
1871 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1876 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1878 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1879 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1880 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1881 ovs_vport_name(vport)) ||
1882 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1883 goto nla_put_failure;
1885 if (!net_eq(net, dev_net(vport->dev))) {
1886 int id = peernet2id_alloc(net, dev_net(vport->dev));
1888 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1889 goto nla_put_failure;
1892 ovs_vport_get_stats(vport, &vport_stats);
1893 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1894 sizeof(struct ovs_vport_stats), &vport_stats,
1895 OVS_VPORT_ATTR_PAD))
1896 goto nla_put_failure;
1898 if (ovs_vport_get_upcall_portids(vport, skb))
1899 goto nla_put_failure;
1901 err = ovs_vport_get_options(vport, skb);
1902 if (err == -EMSGSIZE)
1905 genlmsg_end(skb, ovs_header);
1911 genlmsg_cancel(skb, ovs_header);
1915 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1917 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1920 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1921 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1922 u32 portid, u32 seq, u8 cmd)
1924 struct sk_buff *skb;
1927 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1929 return ERR_PTR(-ENOMEM);
1931 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd);
1937 /* Called with ovs_mutex or RCU read lock. */
1938 static struct vport *lookup_vport(struct net *net,
1939 const struct ovs_header *ovs_header,
1940 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1942 struct datapath *dp;
1943 struct vport *vport;
1945 if (a[OVS_VPORT_ATTR_IFINDEX])
1946 return ERR_PTR(-EOPNOTSUPP);
1947 if (a[OVS_VPORT_ATTR_NAME]) {
1948 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1950 return ERR_PTR(-ENODEV);
1951 if (ovs_header->dp_ifindex &&
1952 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1953 return ERR_PTR(-ENODEV);
1955 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1956 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1958 if (port_no >= DP_MAX_PORTS)
1959 return ERR_PTR(-EFBIG);
1961 dp = get_dp(net, ovs_header->dp_ifindex);
1963 return ERR_PTR(-ENODEV);
1965 vport = ovs_vport_ovsl_rcu(dp, port_no);
1967 return ERR_PTR(-ENODEV);
1970 return ERR_PTR(-EINVAL);
1974 /* Called with ovs_mutex */
1975 static void update_headroom(struct datapath *dp)
1977 unsigned dev_headroom, max_headroom = 0;
1978 struct net_device *dev;
1979 struct vport *vport;
1982 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1983 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1985 dev_headroom = netdev_get_fwd_headroom(dev);
1986 if (dev_headroom > max_headroom)
1987 max_headroom = dev_headroom;
1991 dp->max_headroom = max_headroom;
1992 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1993 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1994 netdev_set_rx_headroom(vport->dev, max_headroom);
1997 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1999 struct nlattr **a = info->attrs;
2000 struct ovs_header *ovs_header = info->userhdr;
2001 struct vport_parms parms;
2002 struct sk_buff *reply;
2003 struct vport *vport;
2004 struct datapath *dp;
2008 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2009 !a[OVS_VPORT_ATTR_UPCALL_PID])
2011 if (a[OVS_VPORT_ATTR_IFINDEX])
2014 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2015 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2016 if (port_no >= DP_MAX_PORTS)
2019 reply = ovs_vport_cmd_alloc_info();
2025 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2028 goto exit_unlock_free;
2031 vport = ovs_vport_ovsl(dp, port_no);
2034 goto exit_unlock_free;
2036 for (port_no = 1; ; port_no++) {
2037 if (port_no >= DP_MAX_PORTS) {
2039 goto exit_unlock_free;
2041 vport = ovs_vport_ovsl(dp, port_no);
2047 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2048 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2049 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2051 parms.port_no = port_no;
2052 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2054 vport = new_vport(&parms);
2055 err = PTR_ERR(vport);
2056 if (IS_ERR(vport)) {
2059 goto exit_unlock_free;
2062 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2063 info->snd_portid, info->snd_seq, 0,
2066 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2067 update_headroom(dp);
2069 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2074 ovs_notify(&dp_vport_genl_family, reply, info);
2083 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2085 struct nlattr **a = info->attrs;
2086 struct sk_buff *reply;
2087 struct vport *vport;
2090 reply = ovs_vport_cmd_alloc_info();
2095 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2096 err = PTR_ERR(vport);
2098 goto exit_unlock_free;
2100 if (a[OVS_VPORT_ATTR_TYPE] &&
2101 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2103 goto exit_unlock_free;
2106 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2107 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2109 goto exit_unlock_free;
2113 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2114 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2116 err = ovs_vport_set_upcall_portids(vport, ids);
2118 goto exit_unlock_free;
2121 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2122 info->snd_portid, info->snd_seq, 0,
2127 ovs_notify(&dp_vport_genl_family, reply, info);
2136 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2138 bool must_update_headroom = false;
2139 struct nlattr **a = info->attrs;
2140 struct sk_buff *reply;
2141 struct datapath *dp;
2142 struct vport *vport;
2145 reply = ovs_vport_cmd_alloc_info();
2150 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2151 err = PTR_ERR(vport);
2153 goto exit_unlock_free;
2155 if (vport->port_no == OVSP_LOCAL) {
2157 goto exit_unlock_free;
2160 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2161 info->snd_portid, info->snd_seq, 0,
2165 /* the vport deletion may trigger dp headroom update */
2167 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2168 must_update_headroom = true;
2169 netdev_reset_rx_headroom(vport->dev);
2170 ovs_dp_detach_port(vport);
2172 if (must_update_headroom)
2173 update_headroom(dp);
2176 ovs_notify(&dp_vport_genl_family, reply, info);
2185 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2187 struct nlattr **a = info->attrs;
2188 struct ovs_header *ovs_header = info->userhdr;
2189 struct sk_buff *reply;
2190 struct vport *vport;
2193 reply = ovs_vport_cmd_alloc_info();
2198 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2199 err = PTR_ERR(vport);
2201 goto exit_unlock_free;
2202 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2203 info->snd_portid, info->snd_seq, 0,
2208 return genlmsg_reply(reply, info);
2216 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2218 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2219 struct datapath *dp;
2220 int bucket = cb->args[0], skip = cb->args[1];
2224 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2229 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2230 struct vport *vport;
2233 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2235 ovs_vport_cmd_fill_info(vport, skb,
2237 NETLINK_CB(cb->skb).portid,
2240 OVS_VPORT_CMD_GET) < 0)
2256 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2257 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2258 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2259 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2260 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2261 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2262 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2263 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2264 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2267 static const struct genl_ops dp_vport_genl_ops[] = {
2268 { .cmd = OVS_VPORT_CMD_NEW,
2269 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2270 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2271 .doit = ovs_vport_cmd_new
2273 { .cmd = OVS_VPORT_CMD_DEL,
2274 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2275 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2276 .doit = ovs_vport_cmd_del
2278 { .cmd = OVS_VPORT_CMD_GET,
2279 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2280 .flags = 0, /* OK for unprivileged users. */
2281 .doit = ovs_vport_cmd_get,
2282 .dumpit = ovs_vport_cmd_dump
2284 { .cmd = OVS_VPORT_CMD_SET,
2285 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2286 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2287 .doit = ovs_vport_cmd_set,
2291 struct genl_family dp_vport_genl_family __ro_after_init = {
2292 .hdrsize = sizeof(struct ovs_header),
2293 .name = OVS_VPORT_FAMILY,
2294 .version = OVS_VPORT_VERSION,
2295 .maxattr = OVS_VPORT_ATTR_MAX,
2296 .policy = vport_policy,
2298 .parallel_ops = true,
2299 .ops = dp_vport_genl_ops,
2300 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2301 .mcgrps = &ovs_dp_vport_multicast_group,
2303 .module = THIS_MODULE,
2306 static struct genl_family * const dp_genl_families[] = {
2307 &dp_datapath_genl_family,
2308 &dp_vport_genl_family,
2309 &dp_flow_genl_family,
2310 &dp_packet_genl_family,
2311 &dp_meter_genl_family,
2312 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2313 &dp_ct_limit_genl_family,
2317 static void dp_unregister_genl(int n_families)
2321 for (i = 0; i < n_families; i++)
2322 genl_unregister_family(dp_genl_families[i]);
2325 static int __init dp_register_genl(void)
2330 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2332 err = genl_register_family(dp_genl_families[i]);
2340 dp_unregister_genl(i);
2344 static int __net_init ovs_init_net(struct net *net)
2346 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2348 INIT_LIST_HEAD(&ovs_net->dps);
2349 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2350 return ovs_ct_init(net);
2353 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2354 struct list_head *head)
2356 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2357 struct datapath *dp;
2359 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2362 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2363 struct vport *vport;
2365 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2366 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2369 if (dev_net(vport->dev) == dnet)
2370 list_add(&vport->detach_list, head);
2376 static void __net_exit ovs_exit_net(struct net *dnet)
2378 struct datapath *dp, *dp_next;
2379 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2380 struct vport *vport, *vport_next;
2386 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2389 down_read(&net_rwsem);
2391 list_vports_from_net(net, dnet, &head);
2392 up_read(&net_rwsem);
2394 /* Detach all vports from given namespace. */
2395 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2396 list_del(&vport->detach_list);
2397 ovs_dp_detach_port(vport);
2402 cancel_work_sync(&ovs_net->dp_notify_work);
2405 static struct pernet_operations ovs_net_ops = {
2406 .init = ovs_init_net,
2407 .exit = ovs_exit_net,
2409 .size = sizeof(struct ovs_net),
2412 static int __init dp_init(void)
2416 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2418 pr_info("Open vSwitch switching datapath\n");
2420 err = action_fifos_init();
2424 err = ovs_internal_dev_rtnl_link_register();
2426 goto error_action_fifos_exit;
2428 err = ovs_flow_init();
2430 goto error_unreg_rtnl_link;
2432 err = ovs_vport_init();
2434 goto error_flow_exit;
2436 err = register_pernet_device(&ovs_net_ops);
2438 goto error_vport_exit;
2440 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2442 goto error_netns_exit;
2444 err = ovs_netdev_init();
2446 goto error_unreg_notifier;
2448 err = dp_register_genl();
2450 goto error_unreg_netdev;
2456 error_unreg_notifier:
2457 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2459 unregister_pernet_device(&ovs_net_ops);
2464 error_unreg_rtnl_link:
2465 ovs_internal_dev_rtnl_link_unregister();
2466 error_action_fifos_exit:
2467 action_fifos_exit();
2472 static void dp_cleanup(void)
2474 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2476 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2477 unregister_pernet_device(&ovs_net_ops);
2481 ovs_internal_dev_rtnl_link_unregister();
2482 action_fifos_exit();
2485 module_init(dp_init);
2486 module_exit(dp_cleanup);
2488 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2489 MODULE_LICENSE("GPL");
2490 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2491 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2492 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2493 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2494 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2495 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);