1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2007-2014 Nicira, Inc.
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8 #include <linux/init.h>
9 #include <linux/module.h>
10 #include <linux/if_arp.h>
11 #include <linux/if_vlan.h>
14 #include <linux/jhash.h>
15 #include <linux/delay.h>
16 #include <linux/time.h>
17 #include <linux/etherdevice.h>
18 #include <linux/genetlink.h>
19 #include <linux/kernel.h>
20 #include <linux/kthread.h>
21 #include <linux/mutex.h>
22 #include <linux/percpu.h>
23 #include <linux/rcupdate.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/ethtool.h>
27 #include <linux/wait.h>
28 #include <asm/div64.h>
29 #include <linux/highmem.h>
30 #include <linux/netfilter_bridge.h>
31 #include <linux/netfilter_ipv4.h>
32 #include <linux/inetdevice.h>
33 #include <linux/list.h>
34 #include <linux/openvswitch.h>
35 #include <linux/rculist.h>
36 #include <linux/dmi.h>
37 #include <net/genetlink.h>
38 #include <net/net_namespace.h>
39 #include <net/netns/generic.h>
40 #include <net/pkt_cls.h>
44 #include "flow_table.h"
45 #include "flow_netlink.h"
47 #include "openvswitch_trace.h"
48 #include "vport-internal_dev.h"
49 #include "vport-netdev.h"
51 unsigned int ovs_net_id __read_mostly;
53 static struct genl_family dp_packet_genl_family;
54 static struct genl_family dp_flow_genl_family;
55 static struct genl_family dp_datapath_genl_family;
57 static const struct nla_policy flow_policy[];
59 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
60 .name = OVS_FLOW_MCGROUP,
63 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
64 .name = OVS_DATAPATH_MCGROUP,
67 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
68 .name = OVS_VPORT_MCGROUP,
71 /* Check if need to build a reply message.
72 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
73 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
76 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
77 genl_has_listeners(family, genl_info_net(info), group);
80 static void ovs_notify(struct genl_family *family,
81 struct sk_buff *skb, struct genl_info *info)
83 genl_notify(family, skb, info, 0, GFP_KERNEL);
89 * All writes e.g. Writes to device state (add/remove datapath, port, set
90 * operations on vports, etc.), Writes to other state (flow table
91 * modifications, set miscellaneous datapath parameters, etc.) are protected
94 * Reads are protected by RCU.
96 * There are a few special cases (mostly stats) that have their own
97 * synchronization but they nest under all of above and don't interact with
100 * The RTNL lock nests inside ovs_mutex.
103 static DEFINE_MUTEX(ovs_mutex);
107 mutex_lock(&ovs_mutex);
110 void ovs_unlock(void)
112 mutex_unlock(&ovs_mutex);
115 #ifdef CONFIG_LOCKDEP
116 int lockdep_ovsl_is_held(void)
119 return lockdep_is_held(&ovs_mutex);
125 static struct vport *new_vport(const struct vport_parms *);
126 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
127 const struct sw_flow_key *,
128 const struct dp_upcall_info *,
130 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
131 const struct sw_flow_key *,
132 const struct dp_upcall_info *,
135 static void ovs_dp_masks_rebalance(struct work_struct *work);
137 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
139 /* Must be called with rcu_read_lock or ovs_mutex. */
140 const char *ovs_dp_name(const struct datapath *dp)
142 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
143 return ovs_vport_name(vport);
146 static int get_dpifindex(const struct datapath *dp)
153 local = ovs_vport_rcu(dp, OVSP_LOCAL);
155 ifindex = local->dev->ifindex;
164 static void destroy_dp_rcu(struct rcu_head *rcu)
166 struct datapath *dp = container_of(rcu, struct datapath, rcu);
168 ovs_flow_tbl_destroy(&dp->table);
169 free_percpu(dp->stats_percpu);
172 kfree(rcu_dereference_raw(dp->upcall_portids));
176 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
179 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
182 /* Called with ovs_mutex or RCU read lock. */
183 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
186 struct hlist_head *head;
188 head = vport_hash_bucket(dp, port_no);
189 hlist_for_each_entry_rcu(vport, head, dp_hash_node,
190 lockdep_ovsl_is_held()) {
191 if (vport->port_no == port_no)
197 /* Called with ovs_mutex. */
198 static struct vport *new_vport(const struct vport_parms *parms)
202 vport = ovs_vport_add(parms);
203 if (!IS_ERR(vport)) {
204 struct datapath *dp = parms->dp;
205 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
207 hlist_add_head_rcu(&vport->dp_hash_node, head);
212 static void ovs_vport_update_upcall_stats(struct sk_buff *skb,
213 const struct dp_upcall_info *upcall_info,
216 struct vport *p = OVS_CB(skb)->input_vport;
217 struct vport_upcall_stats_percpu *stats;
219 if (upcall_info->cmd != OVS_PACKET_CMD_MISS &&
220 upcall_info->cmd != OVS_PACKET_CMD_ACTION)
223 stats = this_cpu_ptr(p->upcall_stats);
224 u64_stats_update_begin(&stats->syncp);
226 u64_stats_inc(&stats->n_success);
228 u64_stats_inc(&stats->n_fail);
229 u64_stats_update_end(&stats->syncp);
232 void ovs_dp_detach_port(struct vport *p)
236 /* First drop references to device. */
237 hlist_del_rcu(&p->dp_hash_node);
239 /* Free percpu memory */
240 free_percpu(p->upcall_stats);
242 /* Then destroy it. */
246 /* Must be called with rcu_read_lock. */
247 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
249 const struct vport *p = OVS_CB(skb)->input_vport;
250 struct datapath *dp = p->dp;
251 struct sw_flow *flow;
252 struct sw_flow_actions *sf_acts;
253 struct dp_stats_percpu *stats;
259 stats = this_cpu_ptr(dp->stats_percpu);
262 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
263 &n_mask_hit, &n_cache_hit);
264 if (unlikely(!flow)) {
265 struct dp_upcall_info upcall;
267 memset(&upcall, 0, sizeof(upcall));
268 upcall.cmd = OVS_PACKET_CMD_MISS;
270 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
272 ovs_dp_get_upcall_portid(dp, smp_processor_id());
274 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
276 upcall.mru = OVS_CB(skb)->mru;
277 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
289 stats_counter = &stats->n_missed;
293 ovs_flow_stats_update(flow, key->tp.flags, skb);
294 sf_acts = rcu_dereference(flow->sf_acts);
295 error = ovs_execute_actions(dp, skb, sf_acts, key);
297 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
298 ovs_dp_name(dp), error);
300 stats_counter = &stats->n_hit;
303 /* Update datapath statistics. */
304 u64_stats_update_begin(&stats->syncp);
306 stats->n_mask_hit += n_mask_hit;
307 stats->n_cache_hit += n_cache_hit;
308 u64_stats_update_end(&stats->syncp);
311 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
312 const struct sw_flow_key *key,
313 const struct dp_upcall_info *upcall_info,
316 struct dp_stats_percpu *stats;
319 if (trace_ovs_dp_upcall_enabled())
320 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
322 if (upcall_info->portid == 0) {
327 if (!skb_is_gso(skb))
328 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
330 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
332 ovs_vport_update_upcall_stats(skb, upcall_info, !err);
339 stats = this_cpu_ptr(dp->stats_percpu);
341 u64_stats_update_begin(&stats->syncp);
343 u64_stats_update_end(&stats->syncp);
348 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
349 const struct sw_flow_key *key,
350 const struct dp_upcall_info *upcall_info,
353 unsigned int gso_type = skb_shinfo(skb)->gso_type;
354 struct sw_flow_key later_key;
355 struct sk_buff *segs, *nskb;
358 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
359 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
361 return PTR_ERR(segs);
365 if (gso_type & SKB_GSO_UDP) {
366 /* The initial flow key extracted by ovs_flow_key_extract()
367 * in this case is for a first fragment, so we need to
368 * properly mark later fragments.
371 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
374 /* Queue all of the segments. */
375 skb_list_walk_safe(segs, skb, nskb) {
376 if (gso_type & SKB_GSO_UDP && skb != segs)
379 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
385 /* Free all of the segments. */
386 skb_list_walk_safe(segs, skb, nskb) {
395 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
396 unsigned int hdrlen, int actions_attrlen)
398 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
399 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
400 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
401 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
402 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
404 /* OVS_PACKET_ATTR_USERDATA */
405 if (upcall_info->userdata)
406 size += NLA_ALIGN(upcall_info->userdata->nla_len);
408 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
409 if (upcall_info->egress_tun_info)
410 size += nla_total_size(ovs_tun_key_attr_size());
412 /* OVS_PACKET_ATTR_ACTIONS */
413 if (upcall_info->actions_len)
414 size += nla_total_size(actions_attrlen);
416 /* OVS_PACKET_ATTR_MRU */
417 if (upcall_info->mru)
418 size += nla_total_size(sizeof(upcall_info->mru));
423 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
425 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
426 size_t plen = NLA_ALIGN(skb->len) - skb->len;
429 skb_put_zero(skb, plen);
433 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
434 const struct sw_flow_key *key,
435 const struct dp_upcall_info *upcall_info,
438 struct ovs_header *upcall;
439 struct sk_buff *nskb = NULL;
440 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
447 dp_ifindex = get_dpifindex(dp);
451 if (skb_vlan_tag_present(skb)) {
452 nskb = skb_clone(skb, GFP_ATOMIC);
456 nskb = __vlan_hwaccel_push_inside(nskb);
463 if (nla_attr_size(skb->len) > USHRT_MAX) {
468 /* Complete checksum if needed */
469 if (skb->ip_summed == CHECKSUM_PARTIAL &&
470 (err = skb_csum_hwoffload_help(skb, 0)))
473 /* Older versions of OVS user space enforce alignment of the last
474 * Netlink attribute to NLA_ALIGNTO which would require extensive
475 * padding logic. Only perform zerocopy if padding is not required.
477 if (dp->user_features & OVS_DP_F_UNALIGNED)
478 hlen = skb_zerocopy_headlen(skb);
482 len = upcall_msg_size(upcall_info, hlen - cutlen,
483 OVS_CB(skb)->acts_origlen);
484 user_skb = genlmsg_new(len, GFP_ATOMIC);
490 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
491 0, upcall_info->cmd);
496 upcall->dp_ifindex = dp_ifindex;
498 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
502 if (upcall_info->userdata)
503 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
504 nla_len(upcall_info->userdata),
505 nla_data(upcall_info->userdata));
507 if (upcall_info->egress_tun_info) {
508 nla = nla_nest_start_noflag(user_skb,
509 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
514 err = ovs_nla_put_tunnel_info(user_skb,
515 upcall_info->egress_tun_info);
519 nla_nest_end(user_skb, nla);
522 if (upcall_info->actions_len) {
523 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
528 err = ovs_nla_put_actions(upcall_info->actions,
529 upcall_info->actions_len,
532 nla_nest_end(user_skb, nla);
534 nla_nest_cancel(user_skb, nla);
537 /* Add OVS_PACKET_ATTR_MRU */
538 if (upcall_info->mru &&
539 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
544 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
546 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
551 /* Add OVS_PACKET_ATTR_HASH */
552 hash = skb_get_hash_raw(skb);
554 hash |= OVS_PACKET_HASH_SW_BIT;
557 hash |= OVS_PACKET_HASH_L4_BIT;
559 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
564 /* Only reserve room for attribute header, packet data is added
565 * in skb_zerocopy() */
566 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
570 nla->nla_len = nla_attr_size(skb->len - cutlen);
572 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
576 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
577 pad_packet(dp, user_skb);
579 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
581 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
586 consume_skb(user_skb);
592 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
594 struct ovs_header *ovs_header = info->userhdr;
595 struct net *net = sock_net(skb->sk);
596 struct nlattr **a = info->attrs;
597 struct sw_flow_actions *acts;
598 struct sk_buff *packet;
599 struct sw_flow *flow;
600 struct sw_flow_actions *sf_acts;
602 struct vport *input_vport;
607 bool log = !a[OVS_PACKET_ATTR_PROBE];
610 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
611 !a[OVS_PACKET_ATTR_ACTIONS])
614 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
615 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
619 skb_reserve(packet, NET_IP_ALIGN);
621 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
623 /* Set packet's mru */
624 if (a[OVS_PACKET_ATTR_MRU]) {
625 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
626 packet->ignore_df = 1;
628 OVS_CB(packet)->mru = mru;
630 if (a[OVS_PACKET_ATTR_HASH]) {
631 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
633 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
634 !!(hash & OVS_PACKET_HASH_SW_BIT),
635 !!(hash & OVS_PACKET_HASH_L4_BIT));
638 /* Build an sw_flow for sending this packet. */
639 flow = ovs_flow_alloc();
644 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
645 packet, &flow->key, log);
649 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
650 &flow->key, &acts, log);
654 rcu_assign_pointer(flow->sf_acts, acts);
655 packet->priority = flow->key.phy.priority;
656 packet->mark = flow->key.phy.skb_mark;
659 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
664 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
666 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
671 packet->dev = input_vport->dev;
672 OVS_CB(packet)->input_vport = input_vport;
673 sf_acts = rcu_dereference(flow->sf_acts);
676 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
680 ovs_flow_free(flow, false);
686 ovs_flow_free(flow, false);
693 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
694 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
695 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
696 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
697 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
698 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
699 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
702 static const struct genl_small_ops dp_packet_genl_ops[] = {
703 { .cmd = OVS_PACKET_CMD_EXECUTE,
704 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
705 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
706 .doit = ovs_packet_cmd_execute
710 static struct genl_family dp_packet_genl_family __ro_after_init = {
711 .hdrsize = sizeof(struct ovs_header),
712 .name = OVS_PACKET_FAMILY,
713 .version = OVS_PACKET_VERSION,
714 .maxattr = OVS_PACKET_ATTR_MAX,
715 .policy = packet_policy,
717 .parallel_ops = true,
718 .small_ops = dp_packet_genl_ops,
719 .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
720 .resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
721 .module = THIS_MODULE,
724 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
725 struct ovs_dp_megaflow_stats *mega_stats)
729 memset(mega_stats, 0, sizeof(*mega_stats));
731 stats->n_flows = ovs_flow_tbl_count(&dp->table);
732 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
734 stats->n_hit = stats->n_missed = stats->n_lost = 0;
736 for_each_possible_cpu(i) {
737 const struct dp_stats_percpu *percpu_stats;
738 struct dp_stats_percpu local_stats;
741 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
744 start = u64_stats_fetch_begin(&percpu_stats->syncp);
745 local_stats = *percpu_stats;
746 } while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
748 stats->n_hit += local_stats.n_hit;
749 stats->n_missed += local_stats.n_missed;
750 stats->n_lost += local_stats.n_lost;
751 mega_stats->n_mask_hit += local_stats.n_mask_hit;
752 mega_stats->n_cache_hit += local_stats.n_cache_hit;
756 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
758 return ovs_identifier_is_ufid(sfid) &&
759 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
762 static bool should_fill_mask(uint32_t ufid_flags)
764 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
767 static bool should_fill_actions(uint32_t ufid_flags)
769 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
772 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
773 const struct sw_flow_id *sfid,
776 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
778 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
779 * see ovs_nla_put_identifier()
781 if (sfid && ovs_identifier_is_ufid(sfid))
782 len += nla_total_size(sfid->ufid_len);
784 len += nla_total_size(ovs_key_attr_size());
786 /* OVS_FLOW_ATTR_KEY */
787 if (!sfid || should_fill_key(sfid, ufid_flags))
788 len += nla_total_size(ovs_key_attr_size());
790 /* OVS_FLOW_ATTR_MASK */
791 if (should_fill_mask(ufid_flags))
792 len += nla_total_size(ovs_key_attr_size());
794 /* OVS_FLOW_ATTR_ACTIONS */
795 if (should_fill_actions(ufid_flags))
796 len += nla_total_size(acts->orig_len);
799 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
800 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
801 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
804 /* Called with ovs_mutex or RCU read lock. */
805 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
808 struct ovs_flow_stats stats;
812 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
815 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
819 if (stats.n_packets &&
820 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
821 sizeof(struct ovs_flow_stats), &stats,
825 if ((u8)ntohs(tcp_flags) &&
826 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
832 /* Called with ovs_mutex or RCU read lock. */
833 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
834 struct sk_buff *skb, int skb_orig_len)
836 struct nlattr *start;
839 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
840 * this is the first flow to be dumped into 'skb'. This is unusual for
841 * Netlink but individual action lists can be longer than
842 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
843 * The userspace caller can always fetch the actions separately if it
844 * really wants them. (Most userspace callers in fact don't care.)
846 * This can only fail for dump operations because the skb is always
847 * properly sized for single flows.
849 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
851 const struct sw_flow_actions *sf_acts;
853 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
854 err = ovs_nla_put_actions(sf_acts->actions,
855 sf_acts->actions_len, skb);
858 nla_nest_end(skb, start);
863 nla_nest_cancel(skb, start);
865 } else if (skb_orig_len) {
872 /* Called with ovs_mutex or RCU read lock. */
873 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
874 struct sk_buff *skb, u32 portid,
875 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
877 const int skb_orig_len = skb->len;
878 struct ovs_header *ovs_header;
881 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
886 ovs_header->dp_ifindex = dp_ifindex;
888 err = ovs_nla_put_identifier(flow, skb);
892 if (should_fill_key(&flow->id, ufid_flags)) {
893 err = ovs_nla_put_masked_key(flow, skb);
898 if (should_fill_mask(ufid_flags)) {
899 err = ovs_nla_put_mask(flow, skb);
904 err = ovs_flow_cmd_fill_stats(flow, skb);
908 if (should_fill_actions(ufid_flags)) {
909 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
914 genlmsg_end(skb, ovs_header);
918 genlmsg_cancel(skb, ovs_header);
922 /* May not be called with RCU read lock. */
923 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
924 const struct sw_flow_id *sfid,
925 struct genl_info *info,
932 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
935 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
936 skb = genlmsg_new(len, GFP_KERNEL);
938 return ERR_PTR(-ENOMEM);
943 /* Called with ovs_mutex. */
944 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
946 struct genl_info *info, u8 cmd,
947 bool always, u32 ufid_flags)
952 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
953 &flow->id, info, always, ufid_flags);
954 if (IS_ERR_OR_NULL(skb))
957 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
958 info->snd_portid, info->snd_seq, 0,
960 if (WARN_ON_ONCE(retval < 0)) {
962 skb = ERR_PTR(retval);
967 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
969 struct net *net = sock_net(skb->sk);
970 struct nlattr **a = info->attrs;
971 struct ovs_header *ovs_header = info->userhdr;
972 struct sw_flow *flow = NULL, *new_flow;
973 struct sw_flow_mask mask;
974 struct sk_buff *reply;
976 struct sw_flow_key *key;
977 struct sw_flow_actions *acts;
978 struct sw_flow_match match;
979 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
981 bool log = !a[OVS_FLOW_ATTR_PROBE];
983 /* Must have key and actions. */
985 if (!a[OVS_FLOW_ATTR_KEY]) {
986 OVS_NLERR(log, "Flow key attr not present in new flow.");
989 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
990 OVS_NLERR(log, "Flow actions attr not present in new flow.");
994 /* Most of the time we need to allocate a new flow, do it before
997 new_flow = ovs_flow_alloc();
998 if (IS_ERR(new_flow)) {
999 error = PTR_ERR(new_flow);
1004 key = kzalloc(sizeof(*key), GFP_KERNEL);
1010 ovs_match_init(&match, key, false, &mask);
1011 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1012 a[OVS_FLOW_ATTR_MASK], log);
1014 goto err_kfree_flow;
1016 ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1018 /* Extract flow identifier. */
1019 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
1022 goto err_kfree_flow;
1024 /* Validate actions. */
1025 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1026 &new_flow->key, &acts, log);
1028 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1029 goto err_kfree_flow;
1032 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1034 if (IS_ERR(reply)) {
1035 error = PTR_ERR(reply);
1036 goto err_kfree_acts;
1040 dp = get_dp(net, ovs_header->dp_ifindex);
1041 if (unlikely(!dp)) {
1043 goto err_unlock_ovs;
1046 /* Check if this is a duplicate flow */
1047 if (ovs_identifier_is_ufid(&new_flow->id))
1048 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1050 flow = ovs_flow_tbl_lookup(&dp->table, key);
1051 if (likely(!flow)) {
1052 rcu_assign_pointer(new_flow->sf_acts, acts);
1054 /* Put flow in bucket. */
1055 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1056 if (unlikely(error)) {
1058 goto err_unlock_ovs;
1061 if (unlikely(reply)) {
1062 error = ovs_flow_cmd_fill_info(new_flow,
1063 ovs_header->dp_ifindex,
1064 reply, info->snd_portid,
1072 struct sw_flow_actions *old_acts;
1074 /* Bail out if we're not allowed to modify an existing flow.
1075 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1076 * because Generic Netlink treats the latter as a dump
1077 * request. We also accept NLM_F_EXCL in case that bug ever
1080 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1083 goto err_unlock_ovs;
1085 /* The flow identifier has to be the same for flow updates.
1086 * Look for any overlapping flow.
1088 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1089 if (ovs_identifier_is_key(&flow->id))
1090 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1092 else /* UFID matches but key is different */
1096 goto err_unlock_ovs;
1099 /* Update actions. */
1100 old_acts = ovsl_dereference(flow->sf_acts);
1101 rcu_assign_pointer(flow->sf_acts, acts);
1103 if (unlikely(reply)) {
1104 error = ovs_flow_cmd_fill_info(flow,
1105 ovs_header->dp_ifindex,
1106 reply, info->snd_portid,
1114 ovs_nla_free_flow_actions_rcu(old_acts);
1115 ovs_flow_free(new_flow, false);
1119 ovs_notify(&dp_flow_genl_family, reply, info);
1128 ovs_nla_free_flow_actions(acts);
1130 ovs_flow_free(new_flow, false);
1137 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1138 static noinline_for_stack
1139 struct sw_flow_actions *get_flow_actions(struct net *net,
1140 const struct nlattr *a,
1141 const struct sw_flow_key *key,
1142 const struct sw_flow_mask *mask,
1145 struct sw_flow_actions *acts;
1146 struct sw_flow_key masked_key;
1149 ovs_flow_mask_key(&masked_key, key, true, mask);
1150 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1153 "Actions may not be safe on all matching packets");
1154 return ERR_PTR(error);
1160 /* Factor out match-init and action-copy to avoid
1161 * "Wframe-larger-than=1024" warning. Because mask is only
1162 * used to get actions, we new a function to save some
1165 * If there are not key and action attrs, we return 0
1166 * directly. In the case, the caller will also not use the
1167 * match as before. If there is action attr, we try to get
1168 * actions and save them to *acts. Before returning from
1169 * the function, we reset the match->mask pointer. Because
1170 * we should not to return match object with dangling reference
1173 static noinline_for_stack int
1174 ovs_nla_init_match_and_action(struct net *net,
1175 struct sw_flow_match *match,
1176 struct sw_flow_key *key,
1178 struct sw_flow_actions **acts,
1181 struct sw_flow_mask mask;
1184 if (a[OVS_FLOW_ATTR_KEY]) {
1185 ovs_match_init(match, key, true, &mask);
1186 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1187 a[OVS_FLOW_ATTR_MASK], log);
1192 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1193 if (!a[OVS_FLOW_ATTR_KEY]) {
1195 "Flow key attribute not present in set flow.");
1200 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1202 if (IS_ERR(*acts)) {
1203 error = PTR_ERR(*acts);
1208 /* On success, error is 0. */
1214 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1216 struct net *net = sock_net(skb->sk);
1217 struct nlattr **a = info->attrs;
1218 struct ovs_header *ovs_header = info->userhdr;
1219 struct sw_flow_key key;
1220 struct sw_flow *flow;
1221 struct sk_buff *reply = NULL;
1222 struct datapath *dp;
1223 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1224 struct sw_flow_match match;
1225 struct sw_flow_id sfid;
1226 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1228 bool log = !a[OVS_FLOW_ATTR_PROBE];
1231 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1232 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1234 "Flow set message rejected, Key attribute missing.");
1238 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1244 /* Can allocate before locking if have acts. */
1245 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1247 if (IS_ERR(reply)) {
1248 error = PTR_ERR(reply);
1249 goto err_kfree_acts;
1254 dp = get_dp(net, ovs_header->dp_ifindex);
1255 if (unlikely(!dp)) {
1257 goto err_unlock_ovs;
1259 /* Check that the flow exists. */
1261 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1263 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1264 if (unlikely(!flow)) {
1266 goto err_unlock_ovs;
1269 /* Update actions, if present. */
1271 old_acts = ovsl_dereference(flow->sf_acts);
1272 rcu_assign_pointer(flow->sf_acts, acts);
1274 if (unlikely(reply)) {
1275 error = ovs_flow_cmd_fill_info(flow,
1276 ovs_header->dp_ifindex,
1277 reply, info->snd_portid,
1284 /* Could not alloc without acts before locking. */
1285 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1286 info, OVS_FLOW_CMD_SET, false,
1289 if (IS_ERR(reply)) {
1290 error = PTR_ERR(reply);
1291 goto err_unlock_ovs;
1296 if (a[OVS_FLOW_ATTR_CLEAR])
1297 ovs_flow_stats_clear(flow);
1301 ovs_notify(&dp_flow_genl_family, reply, info);
1303 ovs_nla_free_flow_actions_rcu(old_acts);
1311 ovs_nla_free_flow_actions(acts);
1316 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1318 struct nlattr **a = info->attrs;
1319 struct ovs_header *ovs_header = info->userhdr;
1320 struct net *net = sock_net(skb->sk);
1321 struct sw_flow_key key;
1322 struct sk_buff *reply;
1323 struct sw_flow *flow;
1324 struct datapath *dp;
1325 struct sw_flow_match match;
1326 struct sw_flow_id ufid;
1327 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1329 bool log = !a[OVS_FLOW_ATTR_PROBE];
1332 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1333 if (a[OVS_FLOW_ATTR_KEY]) {
1334 ovs_match_init(&match, &key, true, NULL);
1335 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1337 } else if (!ufid_present) {
1339 "Flow get message rejected, Key attribute missing.");
1346 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1353 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1355 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1361 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1362 OVS_FLOW_CMD_GET, true, ufid_flags);
1363 if (IS_ERR(reply)) {
1364 err = PTR_ERR(reply);
1369 return genlmsg_reply(reply, info);
1375 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1377 struct nlattr **a = info->attrs;
1378 struct ovs_header *ovs_header = info->userhdr;
1379 struct net *net = sock_net(skb->sk);
1380 struct sw_flow_key key;
1381 struct sk_buff *reply;
1382 struct sw_flow *flow = NULL;
1383 struct datapath *dp;
1384 struct sw_flow_match match;
1385 struct sw_flow_id ufid;
1386 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1388 bool log = !a[OVS_FLOW_ATTR_PROBE];
1391 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1392 if (a[OVS_FLOW_ATTR_KEY]) {
1393 ovs_match_init(&match, &key, true, NULL);
1394 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1401 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1402 if (unlikely(!dp)) {
1407 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1408 err = ovs_flow_tbl_flush(&dp->table);
1413 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1415 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1416 if (unlikely(!flow)) {
1421 ovs_flow_tbl_remove(&dp->table, flow);
1424 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1425 &flow->id, info, false, ufid_flags);
1426 if (likely(reply)) {
1427 if (!IS_ERR(reply)) {
1428 rcu_read_lock(); /*To keep RCU checker happy. */
1429 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1430 reply, info->snd_portid,
1435 if (WARN_ON_ONCE(err < 0)) {
1440 ovs_notify(&dp_flow_genl_family, reply, info);
1442 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1448 ovs_flow_free(flow, true);
1455 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1457 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1458 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1459 struct table_instance *ti;
1460 struct datapath *dp;
1464 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1465 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1468 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1471 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1477 ti = rcu_dereference(dp->table.ti);
1479 struct sw_flow *flow;
1482 bucket = cb->args[0];
1484 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1488 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1489 NETLINK_CB(cb->skb).portid,
1490 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1491 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1494 cb->args[0] = bucket;
1501 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1502 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1503 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1504 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1505 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1506 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1507 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1508 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1511 static const struct genl_small_ops dp_flow_genl_ops[] = {
1512 { .cmd = OVS_FLOW_CMD_NEW,
1513 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1514 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1515 .doit = ovs_flow_cmd_new
1517 { .cmd = OVS_FLOW_CMD_DEL,
1518 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1519 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1520 .doit = ovs_flow_cmd_del
1522 { .cmd = OVS_FLOW_CMD_GET,
1523 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1524 .flags = 0, /* OK for unprivileged users. */
1525 .doit = ovs_flow_cmd_get,
1526 .dumpit = ovs_flow_cmd_dump
1528 { .cmd = OVS_FLOW_CMD_SET,
1529 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1530 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1531 .doit = ovs_flow_cmd_set,
1535 static struct genl_family dp_flow_genl_family __ro_after_init = {
1536 .hdrsize = sizeof(struct ovs_header),
1537 .name = OVS_FLOW_FAMILY,
1538 .version = OVS_FLOW_VERSION,
1539 .maxattr = OVS_FLOW_ATTR_MAX,
1540 .policy = flow_policy,
1542 .parallel_ops = true,
1543 .small_ops = dp_flow_genl_ops,
1544 .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1545 .resv_start_op = OVS_FLOW_CMD_SET + 1,
1546 .mcgrps = &ovs_dp_flow_multicast_group,
1548 .module = THIS_MODULE,
1551 static size_t ovs_dp_cmd_msg_size(void)
1553 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1555 msgsize += nla_total_size(IFNAMSIZ);
1556 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1557 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1558 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1559 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1560 msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1565 /* Called with ovs_mutex. */
1566 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1567 u32 portid, u32 seq, u32 flags, u8 cmd)
1569 struct ovs_header *ovs_header;
1570 struct ovs_dp_stats dp_stats;
1571 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1572 struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1575 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1580 ovs_header->dp_ifindex = get_dpifindex(dp);
1582 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1584 goto nla_put_failure;
1586 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1587 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1588 &dp_stats, OVS_DP_ATTR_PAD))
1589 goto nla_put_failure;
1591 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1592 sizeof(struct ovs_dp_megaflow_stats),
1593 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1594 goto nla_put_failure;
1596 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1597 goto nla_put_failure;
1599 if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1600 ovs_flow_tbl_masks_cache_size(&dp->table)))
1601 goto nla_put_failure;
1603 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1604 pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1605 if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1606 goto nla_put_failure;
1609 genlmsg_end(skb, ovs_header);
1613 genlmsg_cancel(skb, ovs_header);
1618 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1620 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1623 /* Called with rcu_read_lock or ovs_mutex. */
1624 static struct datapath *lookup_datapath(struct net *net,
1625 const struct ovs_header *ovs_header,
1626 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1628 struct datapath *dp;
1630 if (!a[OVS_DP_ATTR_NAME])
1631 dp = get_dp(net, ovs_header->dp_ifindex);
1633 struct vport *vport;
1635 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1636 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1638 return dp ? dp : ERR_PTR(-ENODEV);
1641 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1642 struct genl_info *info)
1644 struct datapath *dp;
1646 dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1651 pr_warn("%s: Dropping previously announced user features\n",
1653 dp->user_features = 0;
1656 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1657 const struct nlattr *ids)
1659 struct dp_nlsk_pids *old, *dp_nlsk_pids;
1661 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1664 old = ovsl_dereference(dp->upcall_portids);
1666 dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1671 dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1672 nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1674 rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1676 kfree_rcu(old, rcu);
1681 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1683 struct dp_nlsk_pids *dp_nlsk_pids;
1685 dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1688 if (cpu_id < dp_nlsk_pids->n_pids) {
1689 return dp_nlsk_pids->pids[cpu_id];
1690 } else if (dp_nlsk_pids->n_pids > 0 &&
1691 cpu_id >= dp_nlsk_pids->n_pids) {
1692 /* If the number of netlink PIDs is mismatched with
1693 * the number of CPUs as seen by the kernel, log this
1694 * and send the upcall to an arbitrary socket (0) in
1695 * order to not drop packets
1697 pr_info_ratelimited("cpu_id mismatch with handler threads");
1698 return dp_nlsk_pids->pids[cpu_id %
1699 dp_nlsk_pids->n_pids];
1708 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1710 u32 user_features = 0, old_features = dp->user_features;
1713 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1714 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1716 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1717 OVS_DP_F_UNALIGNED |
1718 OVS_DP_F_TC_RECIRC_SHARING |
1719 OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1722 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1723 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1728 if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1732 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1733 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1738 dp->user_features = user_features;
1740 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1741 a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1742 /* Upcall Netlink Port IDs have been updated */
1743 err = ovs_dp_set_upcall_portids(dp,
1744 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1749 if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1750 !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1751 tc_skb_ext_tc_enable();
1752 else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1753 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1754 tc_skb_ext_tc_disable();
1759 static int ovs_dp_stats_init(struct datapath *dp)
1761 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1762 if (!dp->stats_percpu)
1768 static int ovs_dp_vport_init(struct datapath *dp)
1772 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1773 sizeof(struct hlist_head),
1778 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1779 INIT_HLIST_HEAD(&dp->ports[i]);
1784 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1786 struct nlattr **a = info->attrs;
1787 struct vport_parms parms;
1788 struct sk_buff *reply;
1789 struct datapath *dp;
1790 struct vport *vport;
1791 struct ovs_net *ovs_net;
1795 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1798 reply = ovs_dp_cmd_alloc_info();
1803 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1805 goto err_destroy_reply;
1807 ovs_dp_set_net(dp, sock_net(skb->sk));
1809 /* Allocate table. */
1810 err = ovs_flow_tbl_init(&dp->table);
1812 goto err_destroy_dp;
1814 err = ovs_dp_stats_init(dp);
1816 goto err_destroy_table;
1818 err = ovs_dp_vport_init(dp);
1820 goto err_destroy_stats;
1822 err = ovs_meters_init(dp);
1824 goto err_destroy_ports;
1826 /* Set up our datapath device. */
1827 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1828 parms.type = OVS_VPORT_TYPE_INTERNAL;
1829 parms.options = NULL;
1831 parms.port_no = OVSP_LOCAL;
1832 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1833 parms.desired_ifindex = a[OVS_DP_ATTR_IFINDEX]
1834 ? nla_get_u32(a[OVS_DP_ATTR_IFINDEX]) : 0;
1836 /* So far only local changes have been made, now need the lock. */
1839 err = ovs_dp_change(dp, a);
1841 goto err_unlock_and_destroy_meters;
1843 vport = new_vport(&parms);
1844 if (IS_ERR(vport)) {
1845 err = PTR_ERR(vport);
1849 if (err == -EEXIST) {
1850 /* An outdated user space instance that does not understand
1851 * the concept of user_features has attempted to create a new
1852 * datapath and is likely to reuse it. Drop all user features.
1854 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1855 ovs_dp_reset_user_features(skb, info);
1858 goto err_destroy_portids;
1861 vport->upcall_stats = netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
1862 if (!vport->upcall_stats) {
1864 goto err_destroy_portids;
1867 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1868 info->snd_seq, 0, OVS_DP_CMD_NEW);
1871 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1872 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1876 ovs_notify(&dp_datapath_genl_family, reply, info);
1879 err_destroy_portids:
1880 kfree(rcu_dereference_raw(dp->upcall_portids));
1881 err_unlock_and_destroy_meters:
1883 ovs_meters_exit(dp);
1887 free_percpu(dp->stats_percpu);
1889 ovs_flow_tbl_destroy(&dp->table);
1898 /* Called with ovs_mutex. */
1899 static void __dp_destroy(struct datapath *dp)
1901 struct flow_table *table = &dp->table;
1904 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1905 tc_skb_ext_tc_disable();
1907 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1908 struct vport *vport;
1909 struct hlist_node *n;
1911 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1912 if (vport->port_no != OVSP_LOCAL)
1913 ovs_dp_detach_port(vport);
1916 list_del_rcu(&dp->list_node);
1918 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1919 * all ports in datapath are destroyed first before freeing datapath.
1921 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1923 /* Flush sw_flow in the tables. RCU cb only releases resource
1924 * such as dp, ports and tables. That may avoid some issues
1925 * such as RCU usage warning.
1927 table_instance_flow_flush(table, ovsl_dereference(table->ti),
1928 ovsl_dereference(table->ufid_ti));
1930 /* RCU destroy the ports, meters and flow tables. */
1931 call_rcu(&dp->rcu, destroy_dp_rcu);
1934 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1936 struct sk_buff *reply;
1937 struct datapath *dp;
1940 reply = ovs_dp_cmd_alloc_info();
1945 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1948 goto err_unlock_free;
1950 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1951 info->snd_seq, 0, OVS_DP_CMD_DEL);
1957 ovs_notify(&dp_datapath_genl_family, reply, info);
1967 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1969 struct sk_buff *reply;
1970 struct datapath *dp;
1973 reply = ovs_dp_cmd_alloc_info();
1978 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1981 goto err_unlock_free;
1983 err = ovs_dp_change(dp, info->attrs);
1985 goto err_unlock_free;
1987 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1988 info->snd_seq, 0, OVS_DP_CMD_SET);
1992 ovs_notify(&dp_datapath_genl_family, reply, info);
2002 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
2004 struct sk_buff *reply;
2005 struct datapath *dp;
2008 reply = ovs_dp_cmd_alloc_info();
2013 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
2016 goto err_unlock_free;
2018 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2019 info->snd_seq, 0, OVS_DP_CMD_GET);
2023 return genlmsg_reply(reply, info);
2031 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2033 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2034 struct datapath *dp;
2035 int skip = cb->args[0];
2039 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2041 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2042 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2043 OVS_DP_CMD_GET) < 0)
2054 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2055 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2056 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2057 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2058 [OVS_DP_ATTR_MASKS_CACHE_SIZE] = NLA_POLICY_RANGE(NLA_U32, 0,
2059 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2060 [OVS_DP_ATTR_IFINDEX] = {.type = NLA_U32 },
2063 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2064 { .cmd = OVS_DP_CMD_NEW,
2065 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2066 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2067 .doit = ovs_dp_cmd_new
2069 { .cmd = OVS_DP_CMD_DEL,
2070 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2071 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2072 .doit = ovs_dp_cmd_del
2074 { .cmd = OVS_DP_CMD_GET,
2075 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2076 .flags = 0, /* OK for unprivileged users. */
2077 .doit = ovs_dp_cmd_get,
2078 .dumpit = ovs_dp_cmd_dump
2080 { .cmd = OVS_DP_CMD_SET,
2081 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2082 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2083 .doit = ovs_dp_cmd_set,
2087 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2088 .hdrsize = sizeof(struct ovs_header),
2089 .name = OVS_DATAPATH_FAMILY,
2090 .version = OVS_DATAPATH_VERSION,
2091 .maxattr = OVS_DP_ATTR_MAX,
2092 .policy = datapath_policy,
2094 .parallel_ops = true,
2095 .small_ops = dp_datapath_genl_ops,
2096 .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2097 .resv_start_op = OVS_DP_CMD_SET + 1,
2098 .mcgrps = &ovs_dp_datapath_multicast_group,
2100 .module = THIS_MODULE,
2103 /* Called with ovs_mutex or RCU read lock. */
2104 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2105 struct net *net, u32 portid, u32 seq,
2106 u32 flags, u8 cmd, gfp_t gfp)
2108 struct ovs_header *ovs_header;
2109 struct ovs_vport_stats vport_stats;
2112 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2117 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2119 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2120 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2121 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2122 ovs_vport_name(vport)) ||
2123 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2124 goto nla_put_failure;
2126 if (!net_eq(net, dev_net(vport->dev))) {
2127 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2129 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2130 goto nla_put_failure;
2133 ovs_vport_get_stats(vport, &vport_stats);
2134 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2135 sizeof(struct ovs_vport_stats), &vport_stats,
2136 OVS_VPORT_ATTR_PAD))
2137 goto nla_put_failure;
2139 if (ovs_vport_get_upcall_stats(vport, skb))
2140 goto nla_put_failure;
2142 if (ovs_vport_get_upcall_portids(vport, skb))
2143 goto nla_put_failure;
2145 err = ovs_vport_get_options(vport, skb);
2146 if (err == -EMSGSIZE)
2149 genlmsg_end(skb, ovs_header);
2155 genlmsg_cancel(skb, ovs_header);
2159 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2161 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2164 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2165 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2166 u32 portid, u32 seq, u8 cmd)
2168 struct sk_buff *skb;
2171 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2173 return ERR_PTR(-ENOMEM);
2175 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2182 /* Called with ovs_mutex or RCU read lock. */
2183 static struct vport *lookup_vport(struct net *net,
2184 const struct ovs_header *ovs_header,
2185 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2187 struct datapath *dp;
2188 struct vport *vport;
2190 if (a[OVS_VPORT_ATTR_IFINDEX])
2191 return ERR_PTR(-EOPNOTSUPP);
2192 if (a[OVS_VPORT_ATTR_NAME]) {
2193 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2195 return ERR_PTR(-ENODEV);
2196 if (ovs_header->dp_ifindex &&
2197 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2198 return ERR_PTR(-ENODEV);
2200 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2201 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2203 if (port_no >= DP_MAX_PORTS)
2204 return ERR_PTR(-EFBIG);
2206 dp = get_dp(net, ovs_header->dp_ifindex);
2208 return ERR_PTR(-ENODEV);
2210 vport = ovs_vport_ovsl_rcu(dp, port_no);
2212 return ERR_PTR(-ENODEV);
2215 return ERR_PTR(-EINVAL);
2219 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2221 unsigned int dev_headroom, max_headroom = 0;
2222 struct net_device *dev;
2223 struct vport *vport;
2226 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2227 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2228 lockdep_ovsl_is_held()) {
2230 dev_headroom = netdev_get_fwd_headroom(dev);
2231 if (dev_headroom > max_headroom)
2232 max_headroom = dev_headroom;
2236 return max_headroom;
2239 /* Called with ovs_mutex */
2240 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2242 struct vport *vport;
2245 dp->max_headroom = new_headroom;
2246 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2247 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2248 lockdep_ovsl_is_held())
2249 netdev_set_rx_headroom(vport->dev, new_headroom);
2253 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2255 struct nlattr **a = info->attrs;
2256 struct ovs_header *ovs_header = info->userhdr;
2257 struct vport_parms parms;
2258 struct sk_buff *reply;
2259 struct vport *vport;
2260 struct datapath *dp;
2261 unsigned int new_headroom;
2265 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2266 !a[OVS_VPORT_ATTR_UPCALL_PID])
2269 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2271 if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2274 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2275 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2276 if (port_no >= DP_MAX_PORTS)
2279 reply = ovs_vport_cmd_alloc_info();
2285 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2288 goto exit_unlock_free;
2291 vport = ovs_vport_ovsl(dp, port_no);
2294 goto exit_unlock_free;
2296 for (port_no = 1; ; port_no++) {
2297 if (port_no >= DP_MAX_PORTS) {
2299 goto exit_unlock_free;
2301 vport = ovs_vport_ovsl(dp, port_no);
2307 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2308 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2310 parms.port_no = port_no;
2311 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2312 parms.desired_ifindex = a[OVS_VPORT_ATTR_IFINDEX]
2313 ? nla_get_u32(a[OVS_VPORT_ATTR_IFINDEX]) : 0;
2315 vport = new_vport(&parms);
2316 err = PTR_ERR(vport);
2317 if (IS_ERR(vport)) {
2320 goto exit_unlock_free;
2323 vport->upcall_stats = netdev_alloc_pcpu_stats(struct vport_upcall_stats_percpu);
2324 if (!vport->upcall_stats) {
2326 goto exit_unlock_free;
2329 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2330 info->snd_portid, info->snd_seq, 0,
2331 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2333 new_headroom = netdev_get_fwd_headroom(vport->dev);
2335 if (new_headroom > dp->max_headroom)
2336 ovs_update_headroom(dp, new_headroom);
2338 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2343 ovs_notify(&dp_vport_genl_family, reply, info);
2352 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2354 struct nlattr **a = info->attrs;
2355 struct sk_buff *reply;
2356 struct vport *vport;
2359 reply = ovs_vport_cmd_alloc_info();
2364 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2365 err = PTR_ERR(vport);
2367 goto exit_unlock_free;
2369 if (a[OVS_VPORT_ATTR_TYPE] &&
2370 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2372 goto exit_unlock_free;
2375 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2376 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2378 goto exit_unlock_free;
2382 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2383 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2385 err = ovs_vport_set_upcall_portids(vport, ids);
2387 goto exit_unlock_free;
2390 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2391 info->snd_portid, info->snd_seq, 0,
2392 OVS_VPORT_CMD_SET, GFP_KERNEL);
2396 ovs_notify(&dp_vport_genl_family, reply, info);
2405 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2407 bool update_headroom = false;
2408 struct nlattr **a = info->attrs;
2409 struct sk_buff *reply;
2410 struct datapath *dp;
2411 struct vport *vport;
2412 unsigned int new_headroom;
2415 reply = ovs_vport_cmd_alloc_info();
2420 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2421 err = PTR_ERR(vport);
2423 goto exit_unlock_free;
2425 if (vport->port_no == OVSP_LOCAL) {
2427 goto exit_unlock_free;
2430 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2431 info->snd_portid, info->snd_seq, 0,
2432 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2435 /* the vport deletion may trigger dp headroom update */
2437 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2438 update_headroom = true;
2440 netdev_reset_rx_headroom(vport->dev);
2441 ovs_dp_detach_port(vport);
2443 if (update_headroom) {
2444 new_headroom = ovs_get_max_headroom(dp);
2446 if (new_headroom < dp->max_headroom)
2447 ovs_update_headroom(dp, new_headroom);
2451 ovs_notify(&dp_vport_genl_family, reply, info);
2460 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2462 struct nlattr **a = info->attrs;
2463 struct ovs_header *ovs_header = info->userhdr;
2464 struct sk_buff *reply;
2465 struct vport *vport;
2468 reply = ovs_vport_cmd_alloc_info();
2473 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2474 err = PTR_ERR(vport);
2476 goto exit_unlock_free;
2477 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2478 info->snd_portid, info->snd_seq, 0,
2479 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2483 return genlmsg_reply(reply, info);
2491 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2493 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2494 struct datapath *dp;
2495 int bucket = cb->args[0], skip = cb->args[1];
2499 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2504 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2505 struct vport *vport;
2508 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2510 ovs_vport_cmd_fill_info(vport, skb,
2512 NETLINK_CB(cb->skb).portid,
2532 static void ovs_dp_masks_rebalance(struct work_struct *work)
2534 struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2535 masks_rebalance.work);
2536 struct datapath *dp;
2540 list_for_each_entry(dp, &ovs_net->dps, list_node)
2541 ovs_flow_masks_rebalance(&dp->table);
2545 schedule_delayed_work(&ovs_net->masks_rebalance,
2546 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2549 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2550 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2551 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2552 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2553 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2554 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2555 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2556 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2557 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2558 [OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
2561 static const struct genl_small_ops dp_vport_genl_ops[] = {
2562 { .cmd = OVS_VPORT_CMD_NEW,
2563 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2564 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2565 .doit = ovs_vport_cmd_new
2567 { .cmd = OVS_VPORT_CMD_DEL,
2568 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2569 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2570 .doit = ovs_vport_cmd_del
2572 { .cmd = OVS_VPORT_CMD_GET,
2573 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2574 .flags = 0, /* OK for unprivileged users. */
2575 .doit = ovs_vport_cmd_get,
2576 .dumpit = ovs_vport_cmd_dump
2578 { .cmd = OVS_VPORT_CMD_SET,
2579 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2580 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2581 .doit = ovs_vport_cmd_set,
2585 struct genl_family dp_vport_genl_family __ro_after_init = {
2586 .hdrsize = sizeof(struct ovs_header),
2587 .name = OVS_VPORT_FAMILY,
2588 .version = OVS_VPORT_VERSION,
2589 .maxattr = OVS_VPORT_ATTR_MAX,
2590 .policy = vport_policy,
2592 .parallel_ops = true,
2593 .small_ops = dp_vport_genl_ops,
2594 .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2595 .resv_start_op = OVS_VPORT_CMD_SET + 1,
2596 .mcgrps = &ovs_dp_vport_multicast_group,
2598 .module = THIS_MODULE,
2601 static struct genl_family * const dp_genl_families[] = {
2602 &dp_datapath_genl_family,
2603 &dp_vport_genl_family,
2604 &dp_flow_genl_family,
2605 &dp_packet_genl_family,
2606 &dp_meter_genl_family,
2607 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2608 &dp_ct_limit_genl_family,
2612 static void dp_unregister_genl(int n_families)
2616 for (i = 0; i < n_families; i++)
2617 genl_unregister_family(dp_genl_families[i]);
2620 static int __init dp_register_genl(void)
2625 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2627 err = genl_register_family(dp_genl_families[i]);
2635 dp_unregister_genl(i);
2639 static int __net_init ovs_init_net(struct net *net)
2641 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2644 INIT_LIST_HEAD(&ovs_net->dps);
2645 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2646 INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2648 err = ovs_ct_init(net);
2652 schedule_delayed_work(&ovs_net->masks_rebalance,
2653 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2657 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2658 struct list_head *head)
2660 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2661 struct datapath *dp;
2663 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2666 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2667 struct vport *vport;
2669 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2670 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2673 if (dev_net(vport->dev) == dnet)
2674 list_add(&vport->detach_list, head);
2680 static void __net_exit ovs_exit_net(struct net *dnet)
2682 struct datapath *dp, *dp_next;
2683 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2684 struct vport *vport, *vport_next;
2692 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2695 down_read(&net_rwsem);
2697 list_vports_from_net(net, dnet, &head);
2698 up_read(&net_rwsem);
2700 /* Detach all vports from given namespace. */
2701 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2702 list_del(&vport->detach_list);
2703 ovs_dp_detach_port(vport);
2708 cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2709 cancel_work_sync(&ovs_net->dp_notify_work);
2712 static struct pernet_operations ovs_net_ops = {
2713 .init = ovs_init_net,
2714 .exit = ovs_exit_net,
2716 .size = sizeof(struct ovs_net),
2719 static int __init dp_init(void)
2723 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2724 sizeof_field(struct sk_buff, cb));
2726 pr_info("Open vSwitch switching datapath\n");
2728 err = action_fifos_init();
2732 err = ovs_internal_dev_rtnl_link_register();
2734 goto error_action_fifos_exit;
2736 err = ovs_flow_init();
2738 goto error_unreg_rtnl_link;
2740 err = ovs_vport_init();
2742 goto error_flow_exit;
2744 err = register_pernet_device(&ovs_net_ops);
2746 goto error_vport_exit;
2748 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2750 goto error_netns_exit;
2752 err = ovs_netdev_init();
2754 goto error_unreg_notifier;
2756 err = dp_register_genl();
2758 goto error_unreg_netdev;
2764 error_unreg_notifier:
2765 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2767 unregister_pernet_device(&ovs_net_ops);
2772 error_unreg_rtnl_link:
2773 ovs_internal_dev_rtnl_link_unregister();
2774 error_action_fifos_exit:
2775 action_fifos_exit();
2780 static void dp_cleanup(void)
2782 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2784 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2785 unregister_pernet_device(&ovs_net_ops);
2789 ovs_internal_dev_rtnl_link_unregister();
2790 action_fifos_exit();
2793 module_init(dp_init);
2794 module_exit(dp_cleanup);
2796 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2797 MODULE_LICENSE("GPL");
2798 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2799 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2800 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2801 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2802 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2803 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);