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>
43 #include "flow_table.h"
44 #include "flow_netlink.h"
46 #include "openvswitch_trace.h"
47 #include "vport-internal_dev.h"
48 #include "vport-netdev.h"
50 unsigned int ovs_net_id __read_mostly;
52 static struct genl_family dp_packet_genl_family;
53 static struct genl_family dp_flow_genl_family;
54 static struct genl_family dp_datapath_genl_family;
56 static const struct nla_policy flow_policy[];
58 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
59 .name = OVS_FLOW_MCGROUP,
62 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
63 .name = OVS_DATAPATH_MCGROUP,
66 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
67 .name = OVS_VPORT_MCGROUP,
70 /* Check if need to build a reply message.
71 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
72 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
75 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
76 genl_has_listeners(family, genl_info_net(info), group);
79 static void ovs_notify(struct genl_family *family,
80 struct sk_buff *skb, struct genl_info *info)
82 genl_notify(family, skb, info, 0, GFP_KERNEL);
88 * All writes e.g. Writes to device state (add/remove datapath, port, set
89 * operations on vports, etc.), Writes to other state (flow table
90 * modifications, set miscellaneous datapath parameters, etc.) are protected
93 * Reads are protected by RCU.
95 * There are a few special cases (mostly stats) that have their own
96 * synchronization but they nest under all of above and don't interact with
99 * The RTNL lock nests inside ovs_mutex.
102 static DEFINE_MUTEX(ovs_mutex);
106 mutex_lock(&ovs_mutex);
109 void ovs_unlock(void)
111 mutex_unlock(&ovs_mutex);
114 #ifdef CONFIG_LOCKDEP
115 int lockdep_ovsl_is_held(void)
118 return lockdep_is_held(&ovs_mutex);
124 static struct vport *new_vport(const struct vport_parms *);
125 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
126 const struct sw_flow_key *,
127 const struct dp_upcall_info *,
129 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
130 const struct sw_flow_key *,
131 const struct dp_upcall_info *,
134 static void ovs_dp_masks_rebalance(struct work_struct *work);
136 /* Must be called with rcu_read_lock or ovs_mutex. */
137 const char *ovs_dp_name(const struct datapath *dp)
139 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
140 return ovs_vport_name(vport);
143 static int get_dpifindex(const struct datapath *dp)
150 local = ovs_vport_rcu(dp, OVSP_LOCAL);
152 ifindex = local->dev->ifindex;
161 static void destroy_dp_rcu(struct rcu_head *rcu)
163 struct datapath *dp = container_of(rcu, struct datapath, rcu);
165 ovs_flow_tbl_destroy(&dp->table);
166 free_percpu(dp->stats_percpu);
172 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
175 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
178 /* Called with ovs_mutex or RCU read lock. */
179 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
182 struct hlist_head *head;
184 head = vport_hash_bucket(dp, port_no);
185 hlist_for_each_entry_rcu(vport, head, dp_hash_node,
186 lockdep_ovsl_is_held()) {
187 if (vport->port_no == port_no)
193 /* Called with ovs_mutex. */
194 static struct vport *new_vport(const struct vport_parms *parms)
198 vport = ovs_vport_add(parms);
199 if (!IS_ERR(vport)) {
200 struct datapath *dp = parms->dp;
201 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
203 hlist_add_head_rcu(&vport->dp_hash_node, head);
208 void ovs_dp_detach_port(struct vport *p)
212 /* First drop references to device. */
213 hlist_del_rcu(&p->dp_hash_node);
215 /* Then destroy it. */
219 /* Must be called with rcu_read_lock. */
220 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
222 const struct vport *p = OVS_CB(skb)->input_vport;
223 struct datapath *dp = p->dp;
224 struct sw_flow *flow;
225 struct sw_flow_actions *sf_acts;
226 struct dp_stats_percpu *stats;
232 stats = this_cpu_ptr(dp->stats_percpu);
235 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
236 &n_mask_hit, &n_cache_hit);
237 if (unlikely(!flow)) {
238 struct dp_upcall_info upcall;
240 memset(&upcall, 0, sizeof(upcall));
241 upcall.cmd = OVS_PACKET_CMD_MISS;
242 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
243 upcall.mru = OVS_CB(skb)->mru;
244 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
249 stats_counter = &stats->n_missed;
253 ovs_flow_stats_update(flow, key->tp.flags, skb);
254 sf_acts = rcu_dereference(flow->sf_acts);
255 error = ovs_execute_actions(dp, skb, sf_acts, key);
257 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
258 ovs_dp_name(dp), error);
260 stats_counter = &stats->n_hit;
263 /* Update datapath statistics. */
264 u64_stats_update_begin(&stats->syncp);
266 stats->n_mask_hit += n_mask_hit;
267 stats->n_cache_hit += n_cache_hit;
268 u64_stats_update_end(&stats->syncp);
271 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
272 const struct sw_flow_key *key,
273 const struct dp_upcall_info *upcall_info,
276 struct dp_stats_percpu *stats;
279 if (trace_ovs_dp_upcall_enabled())
280 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
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_GSO_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. */
333 skb_list_walk_safe(segs, skb, nskb) {
334 if (gso_type & SKB_GSO_UDP && skb != segs)
337 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
343 /* Free all of the segments. */
344 skb_list_walk_safe(segs, skb, nskb) {
353 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
354 unsigned int hdrlen, int actions_attrlen)
356 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
357 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
358 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
359 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
360 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
362 /* OVS_PACKET_ATTR_USERDATA */
363 if (upcall_info->userdata)
364 size += NLA_ALIGN(upcall_info->userdata->nla_len);
366 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
367 if (upcall_info->egress_tun_info)
368 size += nla_total_size(ovs_tun_key_attr_size());
370 /* OVS_PACKET_ATTR_ACTIONS */
371 if (upcall_info->actions_len)
372 size += nla_total_size(actions_attrlen);
374 /* OVS_PACKET_ATTR_MRU */
375 if (upcall_info->mru)
376 size += nla_total_size(sizeof(upcall_info->mru));
381 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
383 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
384 size_t plen = NLA_ALIGN(skb->len) - skb->len;
387 skb_put_zero(skb, plen);
391 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
392 const struct sw_flow_key *key,
393 const struct dp_upcall_info *upcall_info,
396 struct ovs_header *upcall;
397 struct sk_buff *nskb = NULL;
398 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
405 dp_ifindex = get_dpifindex(dp);
409 if (skb_vlan_tag_present(skb)) {
410 nskb = skb_clone(skb, GFP_ATOMIC);
414 nskb = __vlan_hwaccel_push_inside(nskb);
421 if (nla_attr_size(skb->len) > USHRT_MAX) {
426 /* Complete checksum if needed */
427 if (skb->ip_summed == CHECKSUM_PARTIAL &&
428 (err = skb_csum_hwoffload_help(skb, 0)))
431 /* Older versions of OVS user space enforce alignment of the last
432 * Netlink attribute to NLA_ALIGNTO which would require extensive
433 * padding logic. Only perform zerocopy if padding is not required.
435 if (dp->user_features & OVS_DP_F_UNALIGNED)
436 hlen = skb_zerocopy_headlen(skb);
440 len = upcall_msg_size(upcall_info, hlen - cutlen,
441 OVS_CB(skb)->acts_origlen);
442 user_skb = genlmsg_new(len, GFP_ATOMIC);
448 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
449 0, upcall_info->cmd);
454 upcall->dp_ifindex = dp_ifindex;
456 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
460 if (upcall_info->userdata)
461 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
462 nla_len(upcall_info->userdata),
463 nla_data(upcall_info->userdata));
465 if (upcall_info->egress_tun_info) {
466 nla = nla_nest_start_noflag(user_skb,
467 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
472 err = ovs_nla_put_tunnel_info(user_skb,
473 upcall_info->egress_tun_info);
477 nla_nest_end(user_skb, nla);
480 if (upcall_info->actions_len) {
481 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
486 err = ovs_nla_put_actions(upcall_info->actions,
487 upcall_info->actions_len,
490 nla_nest_end(user_skb, nla);
492 nla_nest_cancel(user_skb, nla);
495 /* Add OVS_PACKET_ATTR_MRU */
496 if (upcall_info->mru &&
497 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
502 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
504 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
509 /* Add OVS_PACKET_ATTR_HASH */
510 hash = skb_get_hash_raw(skb);
512 hash |= OVS_PACKET_HASH_SW_BIT;
515 hash |= OVS_PACKET_HASH_L4_BIT;
517 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
522 /* Only reserve room for attribute header, packet data is added
523 * in skb_zerocopy() */
524 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
528 nla->nla_len = nla_attr_size(skb->len - cutlen);
530 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
534 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
535 pad_packet(dp, user_skb);
537 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
539 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
549 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
551 struct ovs_header *ovs_header = info->userhdr;
552 struct net *net = sock_net(skb->sk);
553 struct nlattr **a = info->attrs;
554 struct sw_flow_actions *acts;
555 struct sk_buff *packet;
556 struct sw_flow *flow;
557 struct sw_flow_actions *sf_acts;
559 struct vport *input_vport;
564 bool log = !a[OVS_PACKET_ATTR_PROBE];
567 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
568 !a[OVS_PACKET_ATTR_ACTIONS])
571 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
572 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
576 skb_reserve(packet, NET_IP_ALIGN);
578 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
580 /* Set packet's mru */
581 if (a[OVS_PACKET_ATTR_MRU]) {
582 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
583 packet->ignore_df = 1;
585 OVS_CB(packet)->mru = mru;
587 if (a[OVS_PACKET_ATTR_HASH]) {
588 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
590 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
591 !!(hash & OVS_PACKET_HASH_SW_BIT),
592 !!(hash & OVS_PACKET_HASH_L4_BIT));
595 /* Build an sw_flow for sending this packet. */
596 flow = ovs_flow_alloc();
601 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
602 packet, &flow->key, log);
606 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
607 &flow->key, &acts, log);
611 rcu_assign_pointer(flow->sf_acts, acts);
612 packet->priority = flow->key.phy.priority;
613 packet->mark = flow->key.phy.skb_mark;
616 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
621 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
623 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
628 packet->dev = input_vport->dev;
629 OVS_CB(packet)->input_vport = input_vport;
630 sf_acts = rcu_dereference(flow->sf_acts);
633 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
637 ovs_flow_free(flow, false);
643 ovs_flow_free(flow, false);
650 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
651 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
652 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
653 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
654 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
655 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
656 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
659 static const struct genl_small_ops dp_packet_genl_ops[] = {
660 { .cmd = OVS_PACKET_CMD_EXECUTE,
661 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
662 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
663 .doit = ovs_packet_cmd_execute
667 static struct genl_family dp_packet_genl_family __ro_after_init = {
668 .hdrsize = sizeof(struct ovs_header),
669 .name = OVS_PACKET_FAMILY,
670 .version = OVS_PACKET_VERSION,
671 .maxattr = OVS_PACKET_ATTR_MAX,
672 .policy = packet_policy,
674 .parallel_ops = true,
675 .small_ops = dp_packet_genl_ops,
676 .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
677 .module = THIS_MODULE,
680 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
681 struct ovs_dp_megaflow_stats *mega_stats)
685 memset(mega_stats, 0, sizeof(*mega_stats));
687 stats->n_flows = ovs_flow_tbl_count(&dp->table);
688 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
690 stats->n_hit = stats->n_missed = stats->n_lost = 0;
692 for_each_possible_cpu(i) {
693 const struct dp_stats_percpu *percpu_stats;
694 struct dp_stats_percpu local_stats;
697 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
700 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
701 local_stats = *percpu_stats;
702 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
704 stats->n_hit += local_stats.n_hit;
705 stats->n_missed += local_stats.n_missed;
706 stats->n_lost += local_stats.n_lost;
707 mega_stats->n_mask_hit += local_stats.n_mask_hit;
708 mega_stats->n_cache_hit += local_stats.n_cache_hit;
712 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
714 return ovs_identifier_is_ufid(sfid) &&
715 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
718 static bool should_fill_mask(uint32_t ufid_flags)
720 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
723 static bool should_fill_actions(uint32_t ufid_flags)
725 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
728 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
729 const struct sw_flow_id *sfid,
732 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
734 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
735 * see ovs_nla_put_identifier()
737 if (sfid && ovs_identifier_is_ufid(sfid))
738 len += nla_total_size(sfid->ufid_len);
740 len += nla_total_size(ovs_key_attr_size());
742 /* OVS_FLOW_ATTR_KEY */
743 if (!sfid || should_fill_key(sfid, ufid_flags))
744 len += nla_total_size(ovs_key_attr_size());
746 /* OVS_FLOW_ATTR_MASK */
747 if (should_fill_mask(ufid_flags))
748 len += nla_total_size(ovs_key_attr_size());
750 /* OVS_FLOW_ATTR_ACTIONS */
751 if (should_fill_actions(ufid_flags))
752 len += nla_total_size(acts->orig_len);
755 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
756 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
757 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
760 /* Called with ovs_mutex or RCU read lock. */
761 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
764 struct ovs_flow_stats stats;
768 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
771 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
775 if (stats.n_packets &&
776 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
777 sizeof(struct ovs_flow_stats), &stats,
781 if ((u8)ntohs(tcp_flags) &&
782 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
788 /* Called with ovs_mutex or RCU read lock. */
789 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
790 struct sk_buff *skb, int skb_orig_len)
792 struct nlattr *start;
795 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
796 * this is the first flow to be dumped into 'skb'. This is unusual for
797 * Netlink but individual action lists can be longer than
798 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
799 * The userspace caller can always fetch the actions separately if it
800 * really wants them. (Most userspace callers in fact don't care.)
802 * This can only fail for dump operations because the skb is always
803 * properly sized for single flows.
805 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
807 const struct sw_flow_actions *sf_acts;
809 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
810 err = ovs_nla_put_actions(sf_acts->actions,
811 sf_acts->actions_len, skb);
814 nla_nest_end(skb, start);
819 nla_nest_cancel(skb, start);
821 } else if (skb_orig_len) {
828 /* Called with ovs_mutex or RCU read lock. */
829 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
830 struct sk_buff *skb, u32 portid,
831 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
833 const int skb_orig_len = skb->len;
834 struct ovs_header *ovs_header;
837 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
842 ovs_header->dp_ifindex = dp_ifindex;
844 err = ovs_nla_put_identifier(flow, skb);
848 if (should_fill_key(&flow->id, ufid_flags)) {
849 err = ovs_nla_put_masked_key(flow, skb);
854 if (should_fill_mask(ufid_flags)) {
855 err = ovs_nla_put_mask(flow, skb);
860 err = ovs_flow_cmd_fill_stats(flow, skb);
864 if (should_fill_actions(ufid_flags)) {
865 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
870 genlmsg_end(skb, ovs_header);
874 genlmsg_cancel(skb, ovs_header);
878 /* May not be called with RCU read lock. */
879 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
880 const struct sw_flow_id *sfid,
881 struct genl_info *info,
888 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
891 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
892 skb = genlmsg_new(len, GFP_KERNEL);
894 return ERR_PTR(-ENOMEM);
899 /* Called with ovs_mutex. */
900 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
902 struct genl_info *info, u8 cmd,
903 bool always, u32 ufid_flags)
908 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
909 &flow->id, info, always, ufid_flags);
910 if (IS_ERR_OR_NULL(skb))
913 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
914 info->snd_portid, info->snd_seq, 0,
916 if (WARN_ON_ONCE(retval < 0)) {
918 skb = ERR_PTR(retval);
923 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
925 struct net *net = sock_net(skb->sk);
926 struct nlattr **a = info->attrs;
927 struct ovs_header *ovs_header = info->userhdr;
928 struct sw_flow *flow = NULL, *new_flow;
929 struct sw_flow_mask mask;
930 struct sk_buff *reply;
932 struct sw_flow_actions *acts;
933 struct sw_flow_match match;
934 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
936 bool log = !a[OVS_FLOW_ATTR_PROBE];
938 /* Must have key and actions. */
940 if (!a[OVS_FLOW_ATTR_KEY]) {
941 OVS_NLERR(log, "Flow key attr not present in new flow.");
944 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
945 OVS_NLERR(log, "Flow actions attr not present in new flow.");
949 /* Most of the time we need to allocate a new flow, do it before
952 new_flow = ovs_flow_alloc();
953 if (IS_ERR(new_flow)) {
954 error = PTR_ERR(new_flow);
959 ovs_match_init(&match, &new_flow->key, false, &mask);
960 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
961 a[OVS_FLOW_ATTR_MASK], log);
965 /* Extract flow identifier. */
966 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
967 &new_flow->key, log);
971 /* unmasked key is needed to match when ufid is not used. */
972 if (ovs_identifier_is_key(&new_flow->id))
973 match.key = new_flow->id.unmasked_key;
975 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
977 /* Validate actions. */
978 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
979 &new_flow->key, &acts, log);
981 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
985 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
988 error = PTR_ERR(reply);
993 dp = get_dp(net, ovs_header->dp_ifindex);
999 /* Check if this is a duplicate flow */
1000 if (ovs_identifier_is_ufid(&new_flow->id))
1001 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1003 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1004 if (likely(!flow)) {
1005 rcu_assign_pointer(new_flow->sf_acts, acts);
1007 /* Put flow in bucket. */
1008 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1009 if (unlikely(error)) {
1011 goto err_unlock_ovs;
1014 if (unlikely(reply)) {
1015 error = ovs_flow_cmd_fill_info(new_flow,
1016 ovs_header->dp_ifindex,
1017 reply, info->snd_portid,
1025 struct sw_flow_actions *old_acts;
1027 /* Bail out if we're not allowed to modify an existing flow.
1028 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1029 * because Generic Netlink treats the latter as a dump
1030 * request. We also accept NLM_F_EXCL in case that bug ever
1033 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1036 goto err_unlock_ovs;
1038 /* The flow identifier has to be the same for flow updates.
1039 * Look for any overlapping flow.
1041 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1042 if (ovs_identifier_is_key(&flow->id))
1043 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1045 else /* UFID matches but key is different */
1049 goto err_unlock_ovs;
1052 /* Update actions. */
1053 old_acts = ovsl_dereference(flow->sf_acts);
1054 rcu_assign_pointer(flow->sf_acts, acts);
1056 if (unlikely(reply)) {
1057 error = ovs_flow_cmd_fill_info(flow,
1058 ovs_header->dp_ifindex,
1059 reply, info->snd_portid,
1067 ovs_nla_free_flow_actions_rcu(old_acts);
1068 ovs_flow_free(new_flow, false);
1072 ovs_notify(&dp_flow_genl_family, reply, info);
1079 ovs_nla_free_flow_actions(acts);
1081 ovs_flow_free(new_flow, false);
1086 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1087 static noinline_for_stack
1088 struct sw_flow_actions *get_flow_actions(struct net *net,
1089 const struct nlattr *a,
1090 const struct sw_flow_key *key,
1091 const struct sw_flow_mask *mask,
1094 struct sw_flow_actions *acts;
1095 struct sw_flow_key masked_key;
1098 ovs_flow_mask_key(&masked_key, key, true, mask);
1099 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1102 "Actions may not be safe on all matching packets");
1103 return ERR_PTR(error);
1109 /* Factor out match-init and action-copy to avoid
1110 * "Wframe-larger-than=1024" warning. Because mask is only
1111 * used to get actions, we new a function to save some
1114 * If there are not key and action attrs, we return 0
1115 * directly. In the case, the caller will also not use the
1116 * match as before. If there is action attr, we try to get
1117 * actions and save them to *acts. Before returning from
1118 * the function, we reset the match->mask pointer. Because
1119 * we should not to return match object with dangling reference
1122 static noinline_for_stack int
1123 ovs_nla_init_match_and_action(struct net *net,
1124 struct sw_flow_match *match,
1125 struct sw_flow_key *key,
1127 struct sw_flow_actions **acts,
1130 struct sw_flow_mask mask;
1133 if (a[OVS_FLOW_ATTR_KEY]) {
1134 ovs_match_init(match, key, true, &mask);
1135 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1136 a[OVS_FLOW_ATTR_MASK], log);
1141 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1142 if (!a[OVS_FLOW_ATTR_KEY]) {
1144 "Flow key attribute not present in set flow.");
1149 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1151 if (IS_ERR(*acts)) {
1152 error = PTR_ERR(*acts);
1157 /* On success, error is 0. */
1163 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1165 struct net *net = sock_net(skb->sk);
1166 struct nlattr **a = info->attrs;
1167 struct ovs_header *ovs_header = info->userhdr;
1168 struct sw_flow_key key;
1169 struct sw_flow *flow;
1170 struct sk_buff *reply = NULL;
1171 struct datapath *dp;
1172 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1173 struct sw_flow_match match;
1174 struct sw_flow_id sfid;
1175 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1177 bool log = !a[OVS_FLOW_ATTR_PROBE];
1180 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1181 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1183 "Flow set message rejected, Key attribute missing.");
1187 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1193 /* Can allocate before locking if have acts. */
1194 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1196 if (IS_ERR(reply)) {
1197 error = PTR_ERR(reply);
1198 goto err_kfree_acts;
1203 dp = get_dp(net, ovs_header->dp_ifindex);
1204 if (unlikely(!dp)) {
1206 goto err_unlock_ovs;
1208 /* Check that the flow exists. */
1210 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1212 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1213 if (unlikely(!flow)) {
1215 goto err_unlock_ovs;
1218 /* Update actions, if present. */
1220 old_acts = ovsl_dereference(flow->sf_acts);
1221 rcu_assign_pointer(flow->sf_acts, acts);
1223 if (unlikely(reply)) {
1224 error = ovs_flow_cmd_fill_info(flow,
1225 ovs_header->dp_ifindex,
1226 reply, info->snd_portid,
1233 /* Could not alloc without acts before locking. */
1234 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1235 info, OVS_FLOW_CMD_SET, false,
1238 if (IS_ERR(reply)) {
1239 error = PTR_ERR(reply);
1240 goto err_unlock_ovs;
1245 if (a[OVS_FLOW_ATTR_CLEAR])
1246 ovs_flow_stats_clear(flow);
1250 ovs_notify(&dp_flow_genl_family, reply, info);
1252 ovs_nla_free_flow_actions_rcu(old_acts);
1260 ovs_nla_free_flow_actions(acts);
1265 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1267 struct nlattr **a = info->attrs;
1268 struct ovs_header *ovs_header = info->userhdr;
1269 struct net *net = sock_net(skb->sk);
1270 struct sw_flow_key key;
1271 struct sk_buff *reply;
1272 struct sw_flow *flow;
1273 struct datapath *dp;
1274 struct sw_flow_match match;
1275 struct sw_flow_id ufid;
1276 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1278 bool log = !a[OVS_FLOW_ATTR_PROBE];
1281 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1282 if (a[OVS_FLOW_ATTR_KEY]) {
1283 ovs_match_init(&match, &key, true, NULL);
1284 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1286 } else if (!ufid_present) {
1288 "Flow get message rejected, Key attribute missing.");
1295 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1302 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1304 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1310 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1311 OVS_FLOW_CMD_GET, true, ufid_flags);
1312 if (IS_ERR(reply)) {
1313 err = PTR_ERR(reply);
1318 return genlmsg_reply(reply, info);
1324 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1326 struct nlattr **a = info->attrs;
1327 struct ovs_header *ovs_header = info->userhdr;
1328 struct net *net = sock_net(skb->sk);
1329 struct sw_flow_key key;
1330 struct sk_buff *reply;
1331 struct sw_flow *flow = NULL;
1332 struct datapath *dp;
1333 struct sw_flow_match match;
1334 struct sw_flow_id ufid;
1335 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1337 bool log = !a[OVS_FLOW_ATTR_PROBE];
1340 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1341 if (a[OVS_FLOW_ATTR_KEY]) {
1342 ovs_match_init(&match, &key, true, NULL);
1343 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1350 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1351 if (unlikely(!dp)) {
1356 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1357 err = ovs_flow_tbl_flush(&dp->table);
1362 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1364 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1365 if (unlikely(!flow)) {
1370 ovs_flow_tbl_remove(&dp->table, flow);
1373 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1374 &flow->id, info, false, ufid_flags);
1375 if (likely(reply)) {
1376 if (!IS_ERR(reply)) {
1377 rcu_read_lock(); /*To keep RCU checker happy. */
1378 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1379 reply, info->snd_portid,
1384 if (WARN_ON_ONCE(err < 0)) {
1389 ovs_notify(&dp_flow_genl_family, reply, info);
1391 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1397 ovs_flow_free(flow, true);
1404 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1406 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1407 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1408 struct table_instance *ti;
1409 struct datapath *dp;
1413 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1414 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1417 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1420 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1426 ti = rcu_dereference(dp->table.ti);
1428 struct sw_flow *flow;
1431 bucket = cb->args[0];
1433 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1437 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1438 NETLINK_CB(cb->skb).portid,
1439 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1440 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1443 cb->args[0] = bucket;
1450 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1451 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1452 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1453 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1454 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1455 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1456 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1457 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1460 static const struct genl_small_ops dp_flow_genl_ops[] = {
1461 { .cmd = OVS_FLOW_CMD_NEW,
1462 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1463 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1464 .doit = ovs_flow_cmd_new
1466 { .cmd = OVS_FLOW_CMD_DEL,
1467 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1468 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1469 .doit = ovs_flow_cmd_del
1471 { .cmd = OVS_FLOW_CMD_GET,
1472 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1473 .flags = 0, /* OK for unprivileged users. */
1474 .doit = ovs_flow_cmd_get,
1475 .dumpit = ovs_flow_cmd_dump
1477 { .cmd = OVS_FLOW_CMD_SET,
1478 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1479 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1480 .doit = ovs_flow_cmd_set,
1484 static struct genl_family dp_flow_genl_family __ro_after_init = {
1485 .hdrsize = sizeof(struct ovs_header),
1486 .name = OVS_FLOW_FAMILY,
1487 .version = OVS_FLOW_VERSION,
1488 .maxattr = OVS_FLOW_ATTR_MAX,
1489 .policy = flow_policy,
1491 .parallel_ops = true,
1492 .small_ops = dp_flow_genl_ops,
1493 .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1494 .mcgrps = &ovs_dp_flow_multicast_group,
1496 .module = THIS_MODULE,
1499 static size_t ovs_dp_cmd_msg_size(void)
1501 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1503 msgsize += nla_total_size(IFNAMSIZ);
1504 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1505 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1506 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1507 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1512 /* Called with ovs_mutex. */
1513 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1514 u32 portid, u32 seq, u32 flags, u8 cmd)
1516 struct ovs_header *ovs_header;
1517 struct ovs_dp_stats dp_stats;
1518 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1521 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1526 ovs_header->dp_ifindex = get_dpifindex(dp);
1528 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1530 goto nla_put_failure;
1532 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1533 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1534 &dp_stats, OVS_DP_ATTR_PAD))
1535 goto nla_put_failure;
1537 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1538 sizeof(struct ovs_dp_megaflow_stats),
1539 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1540 goto nla_put_failure;
1542 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1543 goto nla_put_failure;
1545 if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1546 ovs_flow_tbl_masks_cache_size(&dp->table)))
1547 goto nla_put_failure;
1549 genlmsg_end(skb, ovs_header);
1553 genlmsg_cancel(skb, ovs_header);
1558 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1560 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1563 /* Called with rcu_read_lock or ovs_mutex. */
1564 static struct datapath *lookup_datapath(struct net *net,
1565 const struct ovs_header *ovs_header,
1566 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1568 struct datapath *dp;
1570 if (!a[OVS_DP_ATTR_NAME])
1571 dp = get_dp(net, ovs_header->dp_ifindex);
1573 struct vport *vport;
1575 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1576 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1578 return dp ? dp : ERR_PTR(-ENODEV);
1581 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1582 struct genl_info *info)
1584 struct datapath *dp;
1586 dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1591 WARN(dp->user_features, "Dropping previously announced user features\n");
1592 dp->user_features = 0;
1595 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1597 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1599 u32 user_features = 0;
1601 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1602 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1604 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1605 OVS_DP_F_UNALIGNED |
1606 OVS_DP_F_TC_RECIRC_SHARING))
1609 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1610 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1615 if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1619 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1620 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1625 dp->user_features = user_features;
1627 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1628 static_branch_enable(&tc_recirc_sharing_support);
1630 static_branch_disable(&tc_recirc_sharing_support);
1635 static int ovs_dp_stats_init(struct datapath *dp)
1637 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1638 if (!dp->stats_percpu)
1644 static int ovs_dp_vport_init(struct datapath *dp)
1648 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1649 sizeof(struct hlist_head),
1654 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1655 INIT_HLIST_HEAD(&dp->ports[i]);
1660 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1662 struct nlattr **a = info->attrs;
1663 struct vport_parms parms;
1664 struct sk_buff *reply;
1665 struct datapath *dp;
1666 struct vport *vport;
1667 struct ovs_net *ovs_net;
1671 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1674 reply = ovs_dp_cmd_alloc_info();
1679 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1681 goto err_destroy_reply;
1683 ovs_dp_set_net(dp, sock_net(skb->sk));
1685 /* Allocate table. */
1686 err = ovs_flow_tbl_init(&dp->table);
1688 goto err_destroy_dp;
1690 err = ovs_dp_stats_init(dp);
1692 goto err_destroy_table;
1694 err = ovs_dp_vport_init(dp);
1696 goto err_destroy_stats;
1698 err = ovs_meters_init(dp);
1700 goto err_destroy_ports;
1702 /* Set up our datapath device. */
1703 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1704 parms.type = OVS_VPORT_TYPE_INTERNAL;
1705 parms.options = NULL;
1707 parms.port_no = OVSP_LOCAL;
1708 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1710 /* So far only local changes have been made, now need the lock. */
1713 err = ovs_dp_change(dp, a);
1715 goto err_unlock_and_destroy_meters;
1717 vport = new_vport(&parms);
1718 if (IS_ERR(vport)) {
1719 err = PTR_ERR(vport);
1723 if (err == -EEXIST) {
1724 /* An outdated user space instance that does not understand
1725 * the concept of user_features has attempted to create a new
1726 * datapath and is likely to reuse it. Drop all user features.
1728 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1729 ovs_dp_reset_user_features(skb, info);
1732 goto err_unlock_and_destroy_meters;
1735 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1736 info->snd_seq, 0, OVS_DP_CMD_NEW);
1739 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1740 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1744 ovs_notify(&dp_datapath_genl_family, reply, info);
1747 err_unlock_and_destroy_meters:
1749 ovs_meters_exit(dp);
1753 free_percpu(dp->stats_percpu);
1755 ovs_flow_tbl_destroy(&dp->table);
1764 /* Called with ovs_mutex. */
1765 static void __dp_destroy(struct datapath *dp)
1767 struct flow_table *table = &dp->table;
1770 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1771 struct vport *vport;
1772 struct hlist_node *n;
1774 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1775 if (vport->port_no != OVSP_LOCAL)
1776 ovs_dp_detach_port(vport);
1779 list_del_rcu(&dp->list_node);
1781 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1782 * all ports in datapath are destroyed first before freeing datapath.
1784 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1786 /* Flush sw_flow in the tables. RCU cb only releases resource
1787 * such as dp, ports and tables. That may avoid some issues
1788 * such as RCU usage warning.
1790 table_instance_flow_flush(table, ovsl_dereference(table->ti),
1791 ovsl_dereference(table->ufid_ti));
1793 /* RCU destroy the ports, meters and flow tables. */
1794 call_rcu(&dp->rcu, destroy_dp_rcu);
1797 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1799 struct sk_buff *reply;
1800 struct datapath *dp;
1803 reply = ovs_dp_cmd_alloc_info();
1808 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1811 goto err_unlock_free;
1813 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1814 info->snd_seq, 0, OVS_DP_CMD_DEL);
1820 ovs_notify(&dp_datapath_genl_family, reply, info);
1830 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1832 struct sk_buff *reply;
1833 struct datapath *dp;
1836 reply = ovs_dp_cmd_alloc_info();
1841 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1844 goto err_unlock_free;
1846 err = ovs_dp_change(dp, info->attrs);
1848 goto err_unlock_free;
1850 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1851 info->snd_seq, 0, OVS_DP_CMD_SET);
1855 ovs_notify(&dp_datapath_genl_family, reply, info);
1865 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1867 struct sk_buff *reply;
1868 struct datapath *dp;
1871 reply = ovs_dp_cmd_alloc_info();
1876 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1879 goto err_unlock_free;
1881 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1882 info->snd_seq, 0, OVS_DP_CMD_GET);
1886 return genlmsg_reply(reply, info);
1894 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1896 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1897 struct datapath *dp;
1898 int skip = cb->args[0];
1902 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1904 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1905 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1906 OVS_DP_CMD_GET) < 0)
1917 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1918 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1919 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1920 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1921 [OVS_DP_ATTR_MASKS_CACHE_SIZE] = NLA_POLICY_RANGE(NLA_U32, 0,
1922 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
1925 static const struct genl_small_ops dp_datapath_genl_ops[] = {
1926 { .cmd = OVS_DP_CMD_NEW,
1927 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1928 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1929 .doit = ovs_dp_cmd_new
1931 { .cmd = OVS_DP_CMD_DEL,
1932 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1933 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1934 .doit = ovs_dp_cmd_del
1936 { .cmd = OVS_DP_CMD_GET,
1937 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1938 .flags = 0, /* OK for unprivileged users. */
1939 .doit = ovs_dp_cmd_get,
1940 .dumpit = ovs_dp_cmd_dump
1942 { .cmd = OVS_DP_CMD_SET,
1943 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1944 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1945 .doit = ovs_dp_cmd_set,
1949 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1950 .hdrsize = sizeof(struct ovs_header),
1951 .name = OVS_DATAPATH_FAMILY,
1952 .version = OVS_DATAPATH_VERSION,
1953 .maxattr = OVS_DP_ATTR_MAX,
1954 .policy = datapath_policy,
1956 .parallel_ops = true,
1957 .small_ops = dp_datapath_genl_ops,
1958 .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1959 .mcgrps = &ovs_dp_datapath_multicast_group,
1961 .module = THIS_MODULE,
1964 /* Called with ovs_mutex or RCU read lock. */
1965 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1966 struct net *net, u32 portid, u32 seq,
1967 u32 flags, u8 cmd, gfp_t gfp)
1969 struct ovs_header *ovs_header;
1970 struct ovs_vport_stats vport_stats;
1973 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1978 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1980 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1981 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1982 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1983 ovs_vport_name(vport)) ||
1984 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1985 goto nla_put_failure;
1987 if (!net_eq(net, dev_net(vport->dev))) {
1988 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1990 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1991 goto nla_put_failure;
1994 ovs_vport_get_stats(vport, &vport_stats);
1995 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1996 sizeof(struct ovs_vport_stats), &vport_stats,
1997 OVS_VPORT_ATTR_PAD))
1998 goto nla_put_failure;
2000 if (ovs_vport_get_upcall_portids(vport, skb))
2001 goto nla_put_failure;
2003 err = ovs_vport_get_options(vport, skb);
2004 if (err == -EMSGSIZE)
2007 genlmsg_end(skb, ovs_header);
2013 genlmsg_cancel(skb, ovs_header);
2017 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2019 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2022 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2023 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2024 u32 portid, u32 seq, u8 cmd)
2026 struct sk_buff *skb;
2029 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2031 return ERR_PTR(-ENOMEM);
2033 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2040 /* Called with ovs_mutex or RCU read lock. */
2041 static struct vport *lookup_vport(struct net *net,
2042 const struct ovs_header *ovs_header,
2043 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2045 struct datapath *dp;
2046 struct vport *vport;
2048 if (a[OVS_VPORT_ATTR_IFINDEX])
2049 return ERR_PTR(-EOPNOTSUPP);
2050 if (a[OVS_VPORT_ATTR_NAME]) {
2051 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2053 return ERR_PTR(-ENODEV);
2054 if (ovs_header->dp_ifindex &&
2055 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2056 return ERR_PTR(-ENODEV);
2058 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2059 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2061 if (port_no >= DP_MAX_PORTS)
2062 return ERR_PTR(-EFBIG);
2064 dp = get_dp(net, ovs_header->dp_ifindex);
2066 return ERR_PTR(-ENODEV);
2068 vport = ovs_vport_ovsl_rcu(dp, port_no);
2070 return ERR_PTR(-ENODEV);
2073 return ERR_PTR(-EINVAL);
2077 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2079 unsigned int dev_headroom, max_headroom = 0;
2080 struct net_device *dev;
2081 struct vport *vport;
2084 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2085 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2086 lockdep_ovsl_is_held()) {
2088 dev_headroom = netdev_get_fwd_headroom(dev);
2089 if (dev_headroom > max_headroom)
2090 max_headroom = dev_headroom;
2094 return max_headroom;
2097 /* Called with ovs_mutex */
2098 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2100 struct vport *vport;
2103 dp->max_headroom = new_headroom;
2104 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2105 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2106 lockdep_ovsl_is_held())
2107 netdev_set_rx_headroom(vport->dev, new_headroom);
2111 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2113 struct nlattr **a = info->attrs;
2114 struct ovs_header *ovs_header = info->userhdr;
2115 struct vport_parms parms;
2116 struct sk_buff *reply;
2117 struct vport *vport;
2118 struct datapath *dp;
2119 unsigned int new_headroom;
2123 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2124 !a[OVS_VPORT_ATTR_UPCALL_PID])
2126 if (a[OVS_VPORT_ATTR_IFINDEX])
2129 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2130 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2131 if (port_no >= DP_MAX_PORTS)
2134 reply = ovs_vport_cmd_alloc_info();
2140 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2143 goto exit_unlock_free;
2146 vport = ovs_vport_ovsl(dp, port_no);
2149 goto exit_unlock_free;
2151 for (port_no = 1; ; port_no++) {
2152 if (port_no >= DP_MAX_PORTS) {
2154 goto exit_unlock_free;
2156 vport = ovs_vport_ovsl(dp, port_no);
2162 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2163 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2164 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2166 parms.port_no = port_no;
2167 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2169 vport = new_vport(&parms);
2170 err = PTR_ERR(vport);
2171 if (IS_ERR(vport)) {
2174 goto exit_unlock_free;
2177 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2178 info->snd_portid, info->snd_seq, 0,
2179 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2181 new_headroom = netdev_get_fwd_headroom(vport->dev);
2183 if (new_headroom > dp->max_headroom)
2184 ovs_update_headroom(dp, new_headroom);
2186 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2191 ovs_notify(&dp_vport_genl_family, reply, info);
2200 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2202 struct nlattr **a = info->attrs;
2203 struct sk_buff *reply;
2204 struct vport *vport;
2207 reply = ovs_vport_cmd_alloc_info();
2212 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2213 err = PTR_ERR(vport);
2215 goto exit_unlock_free;
2217 if (a[OVS_VPORT_ATTR_TYPE] &&
2218 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2220 goto exit_unlock_free;
2223 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2224 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2226 goto exit_unlock_free;
2230 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2231 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2233 err = ovs_vport_set_upcall_portids(vport, ids);
2235 goto exit_unlock_free;
2238 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2239 info->snd_portid, info->snd_seq, 0,
2240 OVS_VPORT_CMD_SET, GFP_KERNEL);
2244 ovs_notify(&dp_vport_genl_family, reply, info);
2253 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2255 bool update_headroom = false;
2256 struct nlattr **a = info->attrs;
2257 struct sk_buff *reply;
2258 struct datapath *dp;
2259 struct vport *vport;
2260 unsigned int new_headroom;
2263 reply = ovs_vport_cmd_alloc_info();
2268 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2269 err = PTR_ERR(vport);
2271 goto exit_unlock_free;
2273 if (vport->port_no == OVSP_LOCAL) {
2275 goto exit_unlock_free;
2278 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2279 info->snd_portid, info->snd_seq, 0,
2280 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2283 /* the vport deletion may trigger dp headroom update */
2285 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2286 update_headroom = true;
2288 netdev_reset_rx_headroom(vport->dev);
2289 ovs_dp_detach_port(vport);
2291 if (update_headroom) {
2292 new_headroom = ovs_get_max_headroom(dp);
2294 if (new_headroom < dp->max_headroom)
2295 ovs_update_headroom(dp, new_headroom);
2299 ovs_notify(&dp_vport_genl_family, reply, info);
2308 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2310 struct nlattr **a = info->attrs;
2311 struct ovs_header *ovs_header = info->userhdr;
2312 struct sk_buff *reply;
2313 struct vport *vport;
2316 reply = ovs_vport_cmd_alloc_info();
2321 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2322 err = PTR_ERR(vport);
2324 goto exit_unlock_free;
2325 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2326 info->snd_portid, info->snd_seq, 0,
2327 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2331 return genlmsg_reply(reply, info);
2339 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2341 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2342 struct datapath *dp;
2343 int bucket = cb->args[0], skip = cb->args[1];
2347 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2352 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2353 struct vport *vport;
2356 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2358 ovs_vport_cmd_fill_info(vport, skb,
2360 NETLINK_CB(cb->skb).portid,
2380 static void ovs_dp_masks_rebalance(struct work_struct *work)
2382 struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2383 masks_rebalance.work);
2384 struct datapath *dp;
2388 list_for_each_entry(dp, &ovs_net->dps, list_node)
2389 ovs_flow_masks_rebalance(&dp->table);
2393 schedule_delayed_work(&ovs_net->masks_rebalance,
2394 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2397 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2398 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2399 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2400 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2401 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2402 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2403 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2404 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2405 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2408 static const struct genl_small_ops dp_vport_genl_ops[] = {
2409 { .cmd = OVS_VPORT_CMD_NEW,
2410 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2411 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2412 .doit = ovs_vport_cmd_new
2414 { .cmd = OVS_VPORT_CMD_DEL,
2415 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2416 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2417 .doit = ovs_vport_cmd_del
2419 { .cmd = OVS_VPORT_CMD_GET,
2420 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2421 .flags = 0, /* OK for unprivileged users. */
2422 .doit = ovs_vport_cmd_get,
2423 .dumpit = ovs_vport_cmd_dump
2425 { .cmd = OVS_VPORT_CMD_SET,
2426 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2427 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2428 .doit = ovs_vport_cmd_set,
2432 struct genl_family dp_vport_genl_family __ro_after_init = {
2433 .hdrsize = sizeof(struct ovs_header),
2434 .name = OVS_VPORT_FAMILY,
2435 .version = OVS_VPORT_VERSION,
2436 .maxattr = OVS_VPORT_ATTR_MAX,
2437 .policy = vport_policy,
2439 .parallel_ops = true,
2440 .small_ops = dp_vport_genl_ops,
2441 .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2442 .mcgrps = &ovs_dp_vport_multicast_group,
2444 .module = THIS_MODULE,
2447 static struct genl_family * const dp_genl_families[] = {
2448 &dp_datapath_genl_family,
2449 &dp_vport_genl_family,
2450 &dp_flow_genl_family,
2451 &dp_packet_genl_family,
2452 &dp_meter_genl_family,
2453 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2454 &dp_ct_limit_genl_family,
2458 static void dp_unregister_genl(int n_families)
2462 for (i = 0; i < n_families; i++)
2463 genl_unregister_family(dp_genl_families[i]);
2466 static int __init dp_register_genl(void)
2471 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2473 err = genl_register_family(dp_genl_families[i]);
2481 dp_unregister_genl(i);
2485 static int __net_init ovs_init_net(struct net *net)
2487 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2490 INIT_LIST_HEAD(&ovs_net->dps);
2491 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2492 INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2494 err = ovs_ct_init(net);
2498 schedule_delayed_work(&ovs_net->masks_rebalance,
2499 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2503 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2504 struct list_head *head)
2506 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2507 struct datapath *dp;
2509 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2512 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2513 struct vport *vport;
2515 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2516 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2519 if (dev_net(vport->dev) == dnet)
2520 list_add(&vport->detach_list, head);
2526 static void __net_exit ovs_exit_net(struct net *dnet)
2528 struct datapath *dp, *dp_next;
2529 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2530 struct vport *vport, *vport_next;
2538 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2541 down_read(&net_rwsem);
2543 list_vports_from_net(net, dnet, &head);
2544 up_read(&net_rwsem);
2546 /* Detach all vports from given namespace. */
2547 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2548 list_del(&vport->detach_list);
2549 ovs_dp_detach_port(vport);
2554 cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2555 cancel_work_sync(&ovs_net->dp_notify_work);
2558 static struct pernet_operations ovs_net_ops = {
2559 .init = ovs_init_net,
2560 .exit = ovs_exit_net,
2562 .size = sizeof(struct ovs_net),
2565 static int __init dp_init(void)
2569 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2570 sizeof_field(struct sk_buff, cb));
2572 pr_info("Open vSwitch switching datapath\n");
2574 err = action_fifos_init();
2578 err = ovs_internal_dev_rtnl_link_register();
2580 goto error_action_fifos_exit;
2582 err = ovs_flow_init();
2584 goto error_unreg_rtnl_link;
2586 err = ovs_vport_init();
2588 goto error_flow_exit;
2590 err = register_pernet_device(&ovs_net_ops);
2592 goto error_vport_exit;
2594 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2596 goto error_netns_exit;
2598 err = ovs_netdev_init();
2600 goto error_unreg_notifier;
2602 err = dp_register_genl();
2604 goto error_unreg_netdev;
2610 error_unreg_notifier:
2611 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2613 unregister_pernet_device(&ovs_net_ops);
2618 error_unreg_rtnl_link:
2619 ovs_internal_dev_rtnl_link_unregister();
2620 error_action_fifos_exit:
2621 action_fifos_exit();
2626 static void dp_cleanup(void)
2628 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2630 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2631 unregister_pernet_device(&ovs_net_ops);
2635 ovs_internal_dev_rtnl_link_unregister();
2636 action_fifos_exit();
2639 module_init(dp_init);
2640 module_exit(dp_cleanup);
2642 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2643 MODULE_LICENSE("GPL");
2644 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2645 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2646 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2647 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2648 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2649 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);