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 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
138 /* Must be called with rcu_read_lock or ovs_mutex. */
139 const char *ovs_dp_name(const struct datapath *dp)
141 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
142 return ovs_vport_name(vport);
145 static int get_dpifindex(const struct datapath *dp)
152 local = ovs_vport_rcu(dp, OVSP_LOCAL);
154 ifindex = local->dev->ifindex;
163 static void destroy_dp_rcu(struct rcu_head *rcu)
165 struct datapath *dp = container_of(rcu, struct datapath, rcu);
167 ovs_flow_tbl_destroy(&dp->table);
168 free_percpu(dp->stats_percpu);
171 kfree(rcu_dereference_raw(dp->upcall_portids));
175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
178 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
181 /* Called with ovs_mutex or RCU read lock. */
182 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
185 struct hlist_head *head;
187 head = vport_hash_bucket(dp, port_no);
188 hlist_for_each_entry_rcu(vport, head, dp_hash_node,
189 lockdep_ovsl_is_held()) {
190 if (vport->port_no == port_no)
196 /* Called with ovs_mutex. */
197 static struct vport *new_vport(const struct vport_parms *parms)
201 vport = ovs_vport_add(parms);
202 if (!IS_ERR(vport)) {
203 struct datapath *dp = parms->dp;
204 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
206 hlist_add_head_rcu(&vport->dp_hash_node, head);
211 void ovs_dp_detach_port(struct vport *p)
215 /* First drop references to device. */
216 hlist_del_rcu(&p->dp_hash_node);
218 /* Then destroy it. */
222 /* Must be called with rcu_read_lock. */
223 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
225 const struct vport *p = OVS_CB(skb)->input_vport;
226 struct datapath *dp = p->dp;
227 struct sw_flow *flow;
228 struct sw_flow_actions *sf_acts;
229 struct dp_stats_percpu *stats;
235 stats = this_cpu_ptr(dp->stats_percpu);
238 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
239 &n_mask_hit, &n_cache_hit);
240 if (unlikely(!flow)) {
241 struct dp_upcall_info upcall;
243 memset(&upcall, 0, sizeof(upcall));
244 upcall.cmd = OVS_PACKET_CMD_MISS;
246 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
248 ovs_dp_get_upcall_portid(dp, smp_processor_id());
250 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
252 upcall.mru = OVS_CB(skb)->mru;
253 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
265 stats_counter = &stats->n_missed;
269 ovs_flow_stats_update(flow, key->tp.flags, skb);
270 sf_acts = rcu_dereference(flow->sf_acts);
271 error = ovs_execute_actions(dp, skb, sf_acts, key);
273 net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
274 ovs_dp_name(dp), error);
276 stats_counter = &stats->n_hit;
279 /* Update datapath statistics. */
280 u64_stats_update_begin(&stats->syncp);
282 stats->n_mask_hit += n_mask_hit;
283 stats->n_cache_hit += n_cache_hit;
284 u64_stats_update_end(&stats->syncp);
287 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
288 const struct sw_flow_key *key,
289 const struct dp_upcall_info *upcall_info,
292 struct dp_stats_percpu *stats;
295 if (trace_ovs_dp_upcall_enabled())
296 trace_ovs_dp_upcall(dp, skb, key, upcall_info);
298 if (upcall_info->portid == 0) {
303 if (!skb_is_gso(skb))
304 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
306 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
313 stats = this_cpu_ptr(dp->stats_percpu);
315 u64_stats_update_begin(&stats->syncp);
317 u64_stats_update_end(&stats->syncp);
322 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
323 const struct sw_flow_key *key,
324 const struct dp_upcall_info *upcall_info,
327 unsigned int gso_type = skb_shinfo(skb)->gso_type;
328 struct sw_flow_key later_key;
329 struct sk_buff *segs, *nskb;
332 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
333 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
335 return PTR_ERR(segs);
339 if (gso_type & SKB_GSO_UDP) {
340 /* The initial flow key extracted by ovs_flow_key_extract()
341 * in this case is for a first fragment, so we need to
342 * properly mark later fragments.
345 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
348 /* Queue all of the segments. */
349 skb_list_walk_safe(segs, skb, nskb) {
350 if (gso_type & SKB_GSO_UDP && skb != segs)
353 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
359 /* Free all of the segments. */
360 skb_list_walk_safe(segs, skb, nskb) {
369 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
370 unsigned int hdrlen, int actions_attrlen)
372 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
373 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
374 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
375 + nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
376 + nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
378 /* OVS_PACKET_ATTR_USERDATA */
379 if (upcall_info->userdata)
380 size += NLA_ALIGN(upcall_info->userdata->nla_len);
382 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
383 if (upcall_info->egress_tun_info)
384 size += nla_total_size(ovs_tun_key_attr_size());
386 /* OVS_PACKET_ATTR_ACTIONS */
387 if (upcall_info->actions_len)
388 size += nla_total_size(actions_attrlen);
390 /* OVS_PACKET_ATTR_MRU */
391 if (upcall_info->mru)
392 size += nla_total_size(sizeof(upcall_info->mru));
397 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
399 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
400 size_t plen = NLA_ALIGN(skb->len) - skb->len;
403 skb_put_zero(skb, plen);
407 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
408 const struct sw_flow_key *key,
409 const struct dp_upcall_info *upcall_info,
412 struct ovs_header *upcall;
413 struct sk_buff *nskb = NULL;
414 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
421 dp_ifindex = get_dpifindex(dp);
425 if (skb_vlan_tag_present(skb)) {
426 nskb = skb_clone(skb, GFP_ATOMIC);
430 nskb = __vlan_hwaccel_push_inside(nskb);
437 if (nla_attr_size(skb->len) > USHRT_MAX) {
442 /* Complete checksum if needed */
443 if (skb->ip_summed == CHECKSUM_PARTIAL &&
444 (err = skb_csum_hwoffload_help(skb, 0)))
447 /* Older versions of OVS user space enforce alignment of the last
448 * Netlink attribute to NLA_ALIGNTO which would require extensive
449 * padding logic. Only perform zerocopy if padding is not required.
451 if (dp->user_features & OVS_DP_F_UNALIGNED)
452 hlen = skb_zerocopy_headlen(skb);
456 len = upcall_msg_size(upcall_info, hlen - cutlen,
457 OVS_CB(skb)->acts_origlen);
458 user_skb = genlmsg_new(len, GFP_ATOMIC);
464 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
465 0, upcall_info->cmd);
470 upcall->dp_ifindex = dp_ifindex;
472 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
476 if (upcall_info->userdata)
477 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
478 nla_len(upcall_info->userdata),
479 nla_data(upcall_info->userdata));
481 if (upcall_info->egress_tun_info) {
482 nla = nla_nest_start_noflag(user_skb,
483 OVS_PACKET_ATTR_EGRESS_TUN_KEY);
488 err = ovs_nla_put_tunnel_info(user_skb,
489 upcall_info->egress_tun_info);
493 nla_nest_end(user_skb, nla);
496 if (upcall_info->actions_len) {
497 nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
502 err = ovs_nla_put_actions(upcall_info->actions,
503 upcall_info->actions_len,
506 nla_nest_end(user_skb, nla);
508 nla_nest_cancel(user_skb, nla);
511 /* Add OVS_PACKET_ATTR_MRU */
512 if (upcall_info->mru &&
513 nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
518 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
520 nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
525 /* Add OVS_PACKET_ATTR_HASH */
526 hash = skb_get_hash_raw(skb);
528 hash |= OVS_PACKET_HASH_SW_BIT;
531 hash |= OVS_PACKET_HASH_L4_BIT;
533 if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
538 /* Only reserve room for attribute header, packet data is added
539 * in skb_zerocopy() */
540 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
544 nla->nla_len = nla_attr_size(skb->len - cutlen);
546 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
550 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
551 pad_packet(dp, user_skb);
553 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
555 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
560 consume_skb(user_skb);
566 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
568 struct ovs_header *ovs_header = info->userhdr;
569 struct net *net = sock_net(skb->sk);
570 struct nlattr **a = info->attrs;
571 struct sw_flow_actions *acts;
572 struct sk_buff *packet;
573 struct sw_flow *flow;
574 struct sw_flow_actions *sf_acts;
576 struct vport *input_vport;
581 bool log = !a[OVS_PACKET_ATTR_PROBE];
584 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
585 !a[OVS_PACKET_ATTR_ACTIONS])
588 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
589 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
593 skb_reserve(packet, NET_IP_ALIGN);
595 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
597 /* Set packet's mru */
598 if (a[OVS_PACKET_ATTR_MRU]) {
599 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
600 packet->ignore_df = 1;
602 OVS_CB(packet)->mru = mru;
604 if (a[OVS_PACKET_ATTR_HASH]) {
605 hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
607 __skb_set_hash(packet, hash & 0xFFFFFFFFULL,
608 !!(hash & OVS_PACKET_HASH_SW_BIT),
609 !!(hash & OVS_PACKET_HASH_L4_BIT));
612 /* Build an sw_flow for sending this packet. */
613 flow = ovs_flow_alloc();
618 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
619 packet, &flow->key, log);
623 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
624 &flow->key, &acts, log);
628 rcu_assign_pointer(flow->sf_acts, acts);
629 packet->priority = flow->key.phy.priority;
630 packet->mark = flow->key.phy.skb_mark;
633 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
638 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
640 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
645 packet->dev = input_vport->dev;
646 OVS_CB(packet)->input_vport = input_vport;
647 sf_acts = rcu_dereference(flow->sf_acts);
650 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
654 ovs_flow_free(flow, false);
660 ovs_flow_free(flow, false);
667 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
668 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
669 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
670 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
671 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
672 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
673 [OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
676 static const struct genl_small_ops dp_packet_genl_ops[] = {
677 { .cmd = OVS_PACKET_CMD_EXECUTE,
678 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
679 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
680 .doit = ovs_packet_cmd_execute
684 static struct genl_family dp_packet_genl_family __ro_after_init = {
685 .hdrsize = sizeof(struct ovs_header),
686 .name = OVS_PACKET_FAMILY,
687 .version = OVS_PACKET_VERSION,
688 .maxattr = OVS_PACKET_ATTR_MAX,
689 .policy = packet_policy,
691 .parallel_ops = true,
692 .small_ops = dp_packet_genl_ops,
693 .n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
694 .module = THIS_MODULE,
697 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
698 struct ovs_dp_megaflow_stats *mega_stats)
702 memset(mega_stats, 0, sizeof(*mega_stats));
704 stats->n_flows = ovs_flow_tbl_count(&dp->table);
705 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
707 stats->n_hit = stats->n_missed = stats->n_lost = 0;
709 for_each_possible_cpu(i) {
710 const struct dp_stats_percpu *percpu_stats;
711 struct dp_stats_percpu local_stats;
714 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
717 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
718 local_stats = *percpu_stats;
719 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
721 stats->n_hit += local_stats.n_hit;
722 stats->n_missed += local_stats.n_missed;
723 stats->n_lost += local_stats.n_lost;
724 mega_stats->n_mask_hit += local_stats.n_mask_hit;
725 mega_stats->n_cache_hit += local_stats.n_cache_hit;
729 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
731 return ovs_identifier_is_ufid(sfid) &&
732 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
735 static bool should_fill_mask(uint32_t ufid_flags)
737 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
740 static bool should_fill_actions(uint32_t ufid_flags)
742 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
745 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
746 const struct sw_flow_id *sfid,
749 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
751 /* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
752 * see ovs_nla_put_identifier()
754 if (sfid && ovs_identifier_is_ufid(sfid))
755 len += nla_total_size(sfid->ufid_len);
757 len += nla_total_size(ovs_key_attr_size());
759 /* OVS_FLOW_ATTR_KEY */
760 if (!sfid || should_fill_key(sfid, ufid_flags))
761 len += nla_total_size(ovs_key_attr_size());
763 /* OVS_FLOW_ATTR_MASK */
764 if (should_fill_mask(ufid_flags))
765 len += nla_total_size(ovs_key_attr_size());
767 /* OVS_FLOW_ATTR_ACTIONS */
768 if (should_fill_actions(ufid_flags))
769 len += nla_total_size(acts->orig_len);
772 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
773 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
774 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
777 /* Called with ovs_mutex or RCU read lock. */
778 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
781 struct ovs_flow_stats stats;
785 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
788 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
792 if (stats.n_packets &&
793 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
794 sizeof(struct ovs_flow_stats), &stats,
798 if ((u8)ntohs(tcp_flags) &&
799 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
805 /* Called with ovs_mutex or RCU read lock. */
806 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
807 struct sk_buff *skb, int skb_orig_len)
809 struct nlattr *start;
812 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
813 * this is the first flow to be dumped into 'skb'. This is unusual for
814 * Netlink but individual action lists can be longer than
815 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
816 * The userspace caller can always fetch the actions separately if it
817 * really wants them. (Most userspace callers in fact don't care.)
819 * This can only fail for dump operations because the skb is always
820 * properly sized for single flows.
822 start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
824 const struct sw_flow_actions *sf_acts;
826 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
827 err = ovs_nla_put_actions(sf_acts->actions,
828 sf_acts->actions_len, skb);
831 nla_nest_end(skb, start);
836 nla_nest_cancel(skb, start);
838 } else if (skb_orig_len) {
845 /* Called with ovs_mutex or RCU read lock. */
846 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
847 struct sk_buff *skb, u32 portid,
848 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
850 const int skb_orig_len = skb->len;
851 struct ovs_header *ovs_header;
854 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
859 ovs_header->dp_ifindex = dp_ifindex;
861 err = ovs_nla_put_identifier(flow, skb);
865 if (should_fill_key(&flow->id, ufid_flags)) {
866 err = ovs_nla_put_masked_key(flow, skb);
871 if (should_fill_mask(ufid_flags)) {
872 err = ovs_nla_put_mask(flow, skb);
877 err = ovs_flow_cmd_fill_stats(flow, skb);
881 if (should_fill_actions(ufid_flags)) {
882 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
887 genlmsg_end(skb, ovs_header);
891 genlmsg_cancel(skb, ovs_header);
895 /* May not be called with RCU read lock. */
896 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
897 const struct sw_flow_id *sfid,
898 struct genl_info *info,
905 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
908 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
909 skb = genlmsg_new(len, GFP_KERNEL);
911 return ERR_PTR(-ENOMEM);
916 /* Called with ovs_mutex. */
917 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
919 struct genl_info *info, u8 cmd,
920 bool always, u32 ufid_flags)
925 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
926 &flow->id, info, always, ufid_flags);
927 if (IS_ERR_OR_NULL(skb))
930 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
931 info->snd_portid, info->snd_seq, 0,
933 if (WARN_ON_ONCE(retval < 0)) {
935 skb = ERR_PTR(retval);
940 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
942 struct net *net = sock_net(skb->sk);
943 struct nlattr **a = info->attrs;
944 struct ovs_header *ovs_header = info->userhdr;
945 struct sw_flow *flow = NULL, *new_flow;
946 struct sw_flow_mask mask;
947 struct sk_buff *reply;
949 struct sw_flow_key *key;
950 struct sw_flow_actions *acts;
951 struct sw_flow_match match;
952 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
954 bool log = !a[OVS_FLOW_ATTR_PROBE];
956 /* Must have key and actions. */
958 if (!a[OVS_FLOW_ATTR_KEY]) {
959 OVS_NLERR(log, "Flow key attr not present in new flow.");
962 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
963 OVS_NLERR(log, "Flow actions attr not present in new flow.");
967 /* Most of the time we need to allocate a new flow, do it before
970 new_flow = ovs_flow_alloc();
971 if (IS_ERR(new_flow)) {
972 error = PTR_ERR(new_flow);
977 key = kzalloc(sizeof(*key), GFP_KERNEL);
983 ovs_match_init(&match, key, false, &mask);
984 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
985 a[OVS_FLOW_ATTR_MASK], log);
989 ovs_flow_mask_key(&new_flow->key, key, true, &mask);
991 /* Extract flow identifier. */
992 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
997 /* Validate actions. */
998 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
999 &new_flow->key, &acts, log);
1001 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1002 goto err_kfree_flow;
1005 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1007 if (IS_ERR(reply)) {
1008 error = PTR_ERR(reply);
1009 goto err_kfree_acts;
1013 dp = get_dp(net, ovs_header->dp_ifindex);
1014 if (unlikely(!dp)) {
1016 goto err_unlock_ovs;
1019 /* Check if this is a duplicate flow */
1020 if (ovs_identifier_is_ufid(&new_flow->id))
1021 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1023 flow = ovs_flow_tbl_lookup(&dp->table, key);
1024 if (likely(!flow)) {
1025 rcu_assign_pointer(new_flow->sf_acts, acts);
1027 /* Put flow in bucket. */
1028 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1029 if (unlikely(error)) {
1031 goto err_unlock_ovs;
1034 if (unlikely(reply)) {
1035 error = ovs_flow_cmd_fill_info(new_flow,
1036 ovs_header->dp_ifindex,
1037 reply, info->snd_portid,
1045 struct sw_flow_actions *old_acts;
1047 /* Bail out if we're not allowed to modify an existing flow.
1048 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1049 * because Generic Netlink treats the latter as a dump
1050 * request. We also accept NLM_F_EXCL in case that bug ever
1053 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1056 goto err_unlock_ovs;
1058 /* The flow identifier has to be the same for flow updates.
1059 * Look for any overlapping flow.
1061 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1062 if (ovs_identifier_is_key(&flow->id))
1063 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1065 else /* UFID matches but key is different */
1069 goto err_unlock_ovs;
1072 /* Update actions. */
1073 old_acts = ovsl_dereference(flow->sf_acts);
1074 rcu_assign_pointer(flow->sf_acts, acts);
1076 if (unlikely(reply)) {
1077 error = ovs_flow_cmd_fill_info(flow,
1078 ovs_header->dp_ifindex,
1079 reply, info->snd_portid,
1087 ovs_nla_free_flow_actions_rcu(old_acts);
1088 ovs_flow_free(new_flow, false);
1092 ovs_notify(&dp_flow_genl_family, reply, info);
1101 ovs_nla_free_flow_actions(acts);
1103 ovs_flow_free(new_flow, false);
1110 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1111 static noinline_for_stack
1112 struct sw_flow_actions *get_flow_actions(struct net *net,
1113 const struct nlattr *a,
1114 const struct sw_flow_key *key,
1115 const struct sw_flow_mask *mask,
1118 struct sw_flow_actions *acts;
1119 struct sw_flow_key masked_key;
1122 ovs_flow_mask_key(&masked_key, key, true, mask);
1123 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1126 "Actions may not be safe on all matching packets");
1127 return ERR_PTR(error);
1133 /* Factor out match-init and action-copy to avoid
1134 * "Wframe-larger-than=1024" warning. Because mask is only
1135 * used to get actions, we new a function to save some
1138 * If there are not key and action attrs, we return 0
1139 * directly. In the case, the caller will also not use the
1140 * match as before. If there is action attr, we try to get
1141 * actions and save them to *acts. Before returning from
1142 * the function, we reset the match->mask pointer. Because
1143 * we should not to return match object with dangling reference
1146 static noinline_for_stack int
1147 ovs_nla_init_match_and_action(struct net *net,
1148 struct sw_flow_match *match,
1149 struct sw_flow_key *key,
1151 struct sw_flow_actions **acts,
1154 struct sw_flow_mask mask;
1157 if (a[OVS_FLOW_ATTR_KEY]) {
1158 ovs_match_init(match, key, true, &mask);
1159 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1160 a[OVS_FLOW_ATTR_MASK], log);
1165 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1166 if (!a[OVS_FLOW_ATTR_KEY]) {
1168 "Flow key attribute not present in set flow.");
1173 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1175 if (IS_ERR(*acts)) {
1176 error = PTR_ERR(*acts);
1181 /* On success, error is 0. */
1187 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1189 struct net *net = sock_net(skb->sk);
1190 struct nlattr **a = info->attrs;
1191 struct ovs_header *ovs_header = info->userhdr;
1192 struct sw_flow_key key;
1193 struct sw_flow *flow;
1194 struct sk_buff *reply = NULL;
1195 struct datapath *dp;
1196 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1197 struct sw_flow_match match;
1198 struct sw_flow_id sfid;
1199 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1201 bool log = !a[OVS_FLOW_ATTR_PROBE];
1204 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1205 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1207 "Flow set message rejected, Key attribute missing.");
1211 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1217 /* Can allocate before locking if have acts. */
1218 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1220 if (IS_ERR(reply)) {
1221 error = PTR_ERR(reply);
1222 goto err_kfree_acts;
1227 dp = get_dp(net, ovs_header->dp_ifindex);
1228 if (unlikely(!dp)) {
1230 goto err_unlock_ovs;
1232 /* Check that the flow exists. */
1234 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1236 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1237 if (unlikely(!flow)) {
1239 goto err_unlock_ovs;
1242 /* Update actions, if present. */
1244 old_acts = ovsl_dereference(flow->sf_acts);
1245 rcu_assign_pointer(flow->sf_acts, acts);
1247 if (unlikely(reply)) {
1248 error = ovs_flow_cmd_fill_info(flow,
1249 ovs_header->dp_ifindex,
1250 reply, info->snd_portid,
1257 /* Could not alloc without acts before locking. */
1258 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1259 info, OVS_FLOW_CMD_SET, false,
1262 if (IS_ERR(reply)) {
1263 error = PTR_ERR(reply);
1264 goto err_unlock_ovs;
1269 if (a[OVS_FLOW_ATTR_CLEAR])
1270 ovs_flow_stats_clear(flow);
1274 ovs_notify(&dp_flow_genl_family, reply, info);
1276 ovs_nla_free_flow_actions_rcu(old_acts);
1284 ovs_nla_free_flow_actions(acts);
1289 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1291 struct nlattr **a = info->attrs;
1292 struct ovs_header *ovs_header = info->userhdr;
1293 struct net *net = sock_net(skb->sk);
1294 struct sw_flow_key key;
1295 struct sk_buff *reply;
1296 struct sw_flow *flow;
1297 struct datapath *dp;
1298 struct sw_flow_match match;
1299 struct sw_flow_id ufid;
1300 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1302 bool log = !a[OVS_FLOW_ATTR_PROBE];
1305 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1306 if (a[OVS_FLOW_ATTR_KEY]) {
1307 ovs_match_init(&match, &key, true, NULL);
1308 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1310 } else if (!ufid_present) {
1312 "Flow get message rejected, Key attribute missing.");
1319 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1326 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1328 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1334 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1335 OVS_FLOW_CMD_GET, true, ufid_flags);
1336 if (IS_ERR(reply)) {
1337 err = PTR_ERR(reply);
1342 return genlmsg_reply(reply, info);
1348 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1350 struct nlattr **a = info->attrs;
1351 struct ovs_header *ovs_header = info->userhdr;
1352 struct net *net = sock_net(skb->sk);
1353 struct sw_flow_key key;
1354 struct sk_buff *reply;
1355 struct sw_flow *flow = NULL;
1356 struct datapath *dp;
1357 struct sw_flow_match match;
1358 struct sw_flow_id ufid;
1359 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1361 bool log = !a[OVS_FLOW_ATTR_PROBE];
1364 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1365 if (a[OVS_FLOW_ATTR_KEY]) {
1366 ovs_match_init(&match, &key, true, NULL);
1367 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1374 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1375 if (unlikely(!dp)) {
1380 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1381 err = ovs_flow_tbl_flush(&dp->table);
1386 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1388 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1389 if (unlikely(!flow)) {
1394 ovs_flow_tbl_remove(&dp->table, flow);
1397 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1398 &flow->id, info, false, ufid_flags);
1399 if (likely(reply)) {
1400 if (!IS_ERR(reply)) {
1401 rcu_read_lock(); /*To keep RCU checker happy. */
1402 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1403 reply, info->snd_portid,
1408 if (WARN_ON_ONCE(err < 0)) {
1413 ovs_notify(&dp_flow_genl_family, reply, info);
1415 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1421 ovs_flow_free(flow, true);
1428 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1430 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1431 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1432 struct table_instance *ti;
1433 struct datapath *dp;
1437 err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1438 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1441 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1444 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1450 ti = rcu_dereference(dp->table.ti);
1452 struct sw_flow *flow;
1455 bucket = cb->args[0];
1457 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1461 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1462 NETLINK_CB(cb->skb).portid,
1463 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1464 OVS_FLOW_CMD_GET, ufid_flags) < 0)
1467 cb->args[0] = bucket;
1474 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1475 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1476 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1477 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1478 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1479 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1480 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1481 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1484 static const struct genl_small_ops dp_flow_genl_ops[] = {
1485 { .cmd = OVS_FLOW_CMD_NEW,
1486 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1487 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1488 .doit = ovs_flow_cmd_new
1490 { .cmd = OVS_FLOW_CMD_DEL,
1491 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1492 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1493 .doit = ovs_flow_cmd_del
1495 { .cmd = OVS_FLOW_CMD_GET,
1496 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1497 .flags = 0, /* OK for unprivileged users. */
1498 .doit = ovs_flow_cmd_get,
1499 .dumpit = ovs_flow_cmd_dump
1501 { .cmd = OVS_FLOW_CMD_SET,
1502 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1503 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1504 .doit = ovs_flow_cmd_set,
1508 static struct genl_family dp_flow_genl_family __ro_after_init = {
1509 .hdrsize = sizeof(struct ovs_header),
1510 .name = OVS_FLOW_FAMILY,
1511 .version = OVS_FLOW_VERSION,
1512 .maxattr = OVS_FLOW_ATTR_MAX,
1513 .policy = flow_policy,
1515 .parallel_ops = true,
1516 .small_ops = dp_flow_genl_ops,
1517 .n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1518 .mcgrps = &ovs_dp_flow_multicast_group,
1520 .module = THIS_MODULE,
1523 static size_t ovs_dp_cmd_msg_size(void)
1525 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1527 msgsize += nla_total_size(IFNAMSIZ);
1528 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1529 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1530 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1531 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1536 /* Called with ovs_mutex. */
1537 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1538 u32 portid, u32 seq, u32 flags, u8 cmd)
1540 struct ovs_header *ovs_header;
1541 struct ovs_dp_stats dp_stats;
1542 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1545 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1550 ovs_header->dp_ifindex = get_dpifindex(dp);
1552 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1554 goto nla_put_failure;
1556 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1557 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1558 &dp_stats, OVS_DP_ATTR_PAD))
1559 goto nla_put_failure;
1561 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1562 sizeof(struct ovs_dp_megaflow_stats),
1563 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1564 goto nla_put_failure;
1566 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1567 goto nla_put_failure;
1569 if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1570 ovs_flow_tbl_masks_cache_size(&dp->table)))
1571 goto nla_put_failure;
1573 genlmsg_end(skb, ovs_header);
1577 genlmsg_cancel(skb, ovs_header);
1582 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1584 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1587 /* Called with rcu_read_lock or ovs_mutex. */
1588 static struct datapath *lookup_datapath(struct net *net,
1589 const struct ovs_header *ovs_header,
1590 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1592 struct datapath *dp;
1594 if (!a[OVS_DP_ATTR_NAME])
1595 dp = get_dp(net, ovs_header->dp_ifindex);
1597 struct vport *vport;
1599 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1600 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1602 return dp ? dp : ERR_PTR(-ENODEV);
1605 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1606 struct genl_info *info)
1608 struct datapath *dp;
1610 dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1615 pr_warn("%s: Dropping previously announced user features\n",
1617 dp->user_features = 0;
1620 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1622 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1623 const struct nlattr *ids)
1625 struct dp_nlsk_pids *old, *dp_nlsk_pids;
1627 if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1630 old = ovsl_dereference(dp->upcall_portids);
1632 dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1637 dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1638 nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1640 rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1642 kfree_rcu(old, rcu);
1647 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1649 struct dp_nlsk_pids *dp_nlsk_pids;
1651 dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1654 if (cpu_id < dp_nlsk_pids->n_pids) {
1655 return dp_nlsk_pids->pids[cpu_id];
1656 } else if (dp_nlsk_pids->n_pids > 0 &&
1657 cpu_id >= dp_nlsk_pids->n_pids) {
1658 /* If the number of netlink PIDs is mismatched with
1659 * the number of CPUs as seen by the kernel, log this
1660 * and send the upcall to an arbitrary socket (0) in
1661 * order to not drop packets
1663 pr_info_ratelimited("cpu_id mismatch with handler threads");
1664 return dp_nlsk_pids->pids[cpu_id %
1665 dp_nlsk_pids->n_pids];
1674 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1676 u32 user_features = 0;
1679 if (a[OVS_DP_ATTR_USER_FEATURES]) {
1680 user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1682 if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1683 OVS_DP_F_UNALIGNED |
1684 OVS_DP_F_TC_RECIRC_SHARING |
1685 OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1688 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1689 if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1694 if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1698 cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1699 err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1704 dp->user_features = user_features;
1706 if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1707 a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1708 /* Upcall Netlink Port IDs have been updated */
1709 err = ovs_dp_set_upcall_portids(dp,
1710 a[OVS_DP_ATTR_PER_CPU_PIDS]);
1715 if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1716 static_branch_enable(&tc_recirc_sharing_support);
1718 static_branch_disable(&tc_recirc_sharing_support);
1723 static int ovs_dp_stats_init(struct datapath *dp)
1725 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1726 if (!dp->stats_percpu)
1732 static int ovs_dp_vport_init(struct datapath *dp)
1736 dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1737 sizeof(struct hlist_head),
1742 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1743 INIT_HLIST_HEAD(&dp->ports[i]);
1748 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1750 struct nlattr **a = info->attrs;
1751 struct vport_parms parms;
1752 struct sk_buff *reply;
1753 struct datapath *dp;
1754 struct vport *vport;
1755 struct ovs_net *ovs_net;
1759 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1762 reply = ovs_dp_cmd_alloc_info();
1767 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1769 goto err_destroy_reply;
1771 ovs_dp_set_net(dp, sock_net(skb->sk));
1773 /* Allocate table. */
1774 err = ovs_flow_tbl_init(&dp->table);
1776 goto err_destroy_dp;
1778 err = ovs_dp_stats_init(dp);
1780 goto err_destroy_table;
1782 err = ovs_dp_vport_init(dp);
1784 goto err_destroy_stats;
1786 err = ovs_meters_init(dp);
1788 goto err_destroy_ports;
1790 /* Set up our datapath device. */
1791 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1792 parms.type = OVS_VPORT_TYPE_INTERNAL;
1793 parms.options = NULL;
1795 parms.port_no = OVSP_LOCAL;
1796 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1798 /* So far only local changes have been made, now need the lock. */
1801 err = ovs_dp_change(dp, a);
1803 goto err_unlock_and_destroy_meters;
1805 vport = new_vport(&parms);
1806 if (IS_ERR(vport)) {
1807 err = PTR_ERR(vport);
1811 if (err == -EEXIST) {
1812 /* An outdated user space instance that does not understand
1813 * the concept of user_features has attempted to create a new
1814 * datapath and is likely to reuse it. Drop all user features.
1816 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1817 ovs_dp_reset_user_features(skb, info);
1820 goto err_destroy_portids;
1823 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1824 info->snd_seq, 0, OVS_DP_CMD_NEW);
1827 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1828 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1832 ovs_notify(&dp_datapath_genl_family, reply, info);
1835 err_destroy_portids:
1836 kfree(rcu_dereference_raw(dp->upcall_portids));
1837 err_unlock_and_destroy_meters:
1839 ovs_meters_exit(dp);
1843 free_percpu(dp->stats_percpu);
1845 ovs_flow_tbl_destroy(&dp->table);
1854 /* Called with ovs_mutex. */
1855 static void __dp_destroy(struct datapath *dp)
1857 struct flow_table *table = &dp->table;
1860 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1861 struct vport *vport;
1862 struct hlist_node *n;
1864 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1865 if (vport->port_no != OVSP_LOCAL)
1866 ovs_dp_detach_port(vport);
1869 list_del_rcu(&dp->list_node);
1871 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1872 * all ports in datapath are destroyed first before freeing datapath.
1874 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1876 /* Flush sw_flow in the tables. RCU cb only releases resource
1877 * such as dp, ports and tables. That may avoid some issues
1878 * such as RCU usage warning.
1880 table_instance_flow_flush(table, ovsl_dereference(table->ti),
1881 ovsl_dereference(table->ufid_ti));
1883 /* RCU destroy the ports, meters and flow tables. */
1884 call_rcu(&dp->rcu, destroy_dp_rcu);
1887 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1889 struct sk_buff *reply;
1890 struct datapath *dp;
1893 reply = ovs_dp_cmd_alloc_info();
1898 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1901 goto err_unlock_free;
1903 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1904 info->snd_seq, 0, OVS_DP_CMD_DEL);
1910 ovs_notify(&dp_datapath_genl_family, reply, info);
1920 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1922 struct sk_buff *reply;
1923 struct datapath *dp;
1926 reply = ovs_dp_cmd_alloc_info();
1931 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1934 goto err_unlock_free;
1936 err = ovs_dp_change(dp, info->attrs);
1938 goto err_unlock_free;
1940 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1941 info->snd_seq, 0, OVS_DP_CMD_SET);
1945 ovs_notify(&dp_datapath_genl_family, reply, info);
1955 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1957 struct sk_buff *reply;
1958 struct datapath *dp;
1961 reply = ovs_dp_cmd_alloc_info();
1966 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1969 goto err_unlock_free;
1971 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1972 info->snd_seq, 0, OVS_DP_CMD_GET);
1976 return genlmsg_reply(reply, info);
1984 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1986 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1987 struct datapath *dp;
1988 int skip = cb->args[0];
1992 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1994 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1995 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1996 OVS_DP_CMD_GET) < 0)
2007 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2008 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2009 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2010 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2011 [OVS_DP_ATTR_MASKS_CACHE_SIZE] = NLA_POLICY_RANGE(NLA_U32, 0,
2012 PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2015 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2016 { .cmd = OVS_DP_CMD_NEW,
2017 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2018 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2019 .doit = ovs_dp_cmd_new
2021 { .cmd = OVS_DP_CMD_DEL,
2022 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2023 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2024 .doit = ovs_dp_cmd_del
2026 { .cmd = OVS_DP_CMD_GET,
2027 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2028 .flags = 0, /* OK for unprivileged users. */
2029 .doit = ovs_dp_cmd_get,
2030 .dumpit = ovs_dp_cmd_dump
2032 { .cmd = OVS_DP_CMD_SET,
2033 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2034 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2035 .doit = ovs_dp_cmd_set,
2039 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2040 .hdrsize = sizeof(struct ovs_header),
2041 .name = OVS_DATAPATH_FAMILY,
2042 .version = OVS_DATAPATH_VERSION,
2043 .maxattr = OVS_DP_ATTR_MAX,
2044 .policy = datapath_policy,
2046 .parallel_ops = true,
2047 .small_ops = dp_datapath_genl_ops,
2048 .n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2049 .mcgrps = &ovs_dp_datapath_multicast_group,
2051 .module = THIS_MODULE,
2054 /* Called with ovs_mutex or RCU read lock. */
2055 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2056 struct net *net, u32 portid, u32 seq,
2057 u32 flags, u8 cmd, gfp_t gfp)
2059 struct ovs_header *ovs_header;
2060 struct ovs_vport_stats vport_stats;
2063 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2068 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2070 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2071 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2072 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2073 ovs_vport_name(vport)) ||
2074 nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2075 goto nla_put_failure;
2077 if (!net_eq(net, dev_net(vport->dev))) {
2078 int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2080 if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2081 goto nla_put_failure;
2084 ovs_vport_get_stats(vport, &vport_stats);
2085 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2086 sizeof(struct ovs_vport_stats), &vport_stats,
2087 OVS_VPORT_ATTR_PAD))
2088 goto nla_put_failure;
2090 if (ovs_vport_get_upcall_portids(vport, skb))
2091 goto nla_put_failure;
2093 err = ovs_vport_get_options(vport, skb);
2094 if (err == -EMSGSIZE)
2097 genlmsg_end(skb, ovs_header);
2103 genlmsg_cancel(skb, ovs_header);
2107 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2109 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2112 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2113 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2114 u32 portid, u32 seq, u8 cmd)
2116 struct sk_buff *skb;
2119 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2121 return ERR_PTR(-ENOMEM);
2123 retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2130 /* Called with ovs_mutex or RCU read lock. */
2131 static struct vport *lookup_vport(struct net *net,
2132 const struct ovs_header *ovs_header,
2133 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2135 struct datapath *dp;
2136 struct vport *vport;
2138 if (a[OVS_VPORT_ATTR_IFINDEX])
2139 return ERR_PTR(-EOPNOTSUPP);
2140 if (a[OVS_VPORT_ATTR_NAME]) {
2141 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2143 return ERR_PTR(-ENODEV);
2144 if (ovs_header->dp_ifindex &&
2145 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2146 return ERR_PTR(-ENODEV);
2148 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2149 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2151 if (port_no >= DP_MAX_PORTS)
2152 return ERR_PTR(-EFBIG);
2154 dp = get_dp(net, ovs_header->dp_ifindex);
2156 return ERR_PTR(-ENODEV);
2158 vport = ovs_vport_ovsl_rcu(dp, port_no);
2160 return ERR_PTR(-ENODEV);
2163 return ERR_PTR(-EINVAL);
2167 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2169 unsigned int dev_headroom, max_headroom = 0;
2170 struct net_device *dev;
2171 struct vport *vport;
2174 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2175 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2176 lockdep_ovsl_is_held()) {
2178 dev_headroom = netdev_get_fwd_headroom(dev);
2179 if (dev_headroom > max_headroom)
2180 max_headroom = dev_headroom;
2184 return max_headroom;
2187 /* Called with ovs_mutex */
2188 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2190 struct vport *vport;
2193 dp->max_headroom = new_headroom;
2194 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2195 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2196 lockdep_ovsl_is_held())
2197 netdev_set_rx_headroom(vport->dev, new_headroom);
2201 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2203 struct nlattr **a = info->attrs;
2204 struct ovs_header *ovs_header = info->userhdr;
2205 struct vport_parms parms;
2206 struct sk_buff *reply;
2207 struct vport *vport;
2208 struct datapath *dp;
2209 unsigned int new_headroom;
2213 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2214 !a[OVS_VPORT_ATTR_UPCALL_PID])
2216 if (a[OVS_VPORT_ATTR_IFINDEX])
2219 port_no = a[OVS_VPORT_ATTR_PORT_NO]
2220 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2221 if (port_no >= DP_MAX_PORTS)
2224 reply = ovs_vport_cmd_alloc_info();
2230 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2233 goto exit_unlock_free;
2236 vport = ovs_vport_ovsl(dp, port_no);
2239 goto exit_unlock_free;
2241 for (port_no = 1; ; port_no++) {
2242 if (port_no >= DP_MAX_PORTS) {
2244 goto exit_unlock_free;
2246 vport = ovs_vport_ovsl(dp, port_no);
2252 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2253 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2254 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2256 parms.port_no = port_no;
2257 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2259 vport = new_vport(&parms);
2260 err = PTR_ERR(vport);
2261 if (IS_ERR(vport)) {
2264 goto exit_unlock_free;
2267 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2268 info->snd_portid, info->snd_seq, 0,
2269 OVS_VPORT_CMD_NEW, GFP_KERNEL);
2271 new_headroom = netdev_get_fwd_headroom(vport->dev);
2273 if (new_headroom > dp->max_headroom)
2274 ovs_update_headroom(dp, new_headroom);
2276 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2281 ovs_notify(&dp_vport_genl_family, reply, info);
2290 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2292 struct nlattr **a = info->attrs;
2293 struct sk_buff *reply;
2294 struct vport *vport;
2297 reply = ovs_vport_cmd_alloc_info();
2302 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2303 err = PTR_ERR(vport);
2305 goto exit_unlock_free;
2307 if (a[OVS_VPORT_ATTR_TYPE] &&
2308 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2310 goto exit_unlock_free;
2313 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2314 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2316 goto exit_unlock_free;
2320 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2321 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2323 err = ovs_vport_set_upcall_portids(vport, ids);
2325 goto exit_unlock_free;
2328 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2329 info->snd_portid, info->snd_seq, 0,
2330 OVS_VPORT_CMD_SET, GFP_KERNEL);
2334 ovs_notify(&dp_vport_genl_family, reply, info);
2343 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2345 bool update_headroom = false;
2346 struct nlattr **a = info->attrs;
2347 struct sk_buff *reply;
2348 struct datapath *dp;
2349 struct vport *vport;
2350 unsigned int new_headroom;
2353 reply = ovs_vport_cmd_alloc_info();
2358 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2359 err = PTR_ERR(vport);
2361 goto exit_unlock_free;
2363 if (vport->port_no == OVSP_LOCAL) {
2365 goto exit_unlock_free;
2368 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2369 info->snd_portid, info->snd_seq, 0,
2370 OVS_VPORT_CMD_DEL, GFP_KERNEL);
2373 /* the vport deletion may trigger dp headroom update */
2375 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2376 update_headroom = true;
2378 netdev_reset_rx_headroom(vport->dev);
2379 ovs_dp_detach_port(vport);
2381 if (update_headroom) {
2382 new_headroom = ovs_get_max_headroom(dp);
2384 if (new_headroom < dp->max_headroom)
2385 ovs_update_headroom(dp, new_headroom);
2389 ovs_notify(&dp_vport_genl_family, reply, info);
2398 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2400 struct nlattr **a = info->attrs;
2401 struct ovs_header *ovs_header = info->userhdr;
2402 struct sk_buff *reply;
2403 struct vport *vport;
2406 reply = ovs_vport_cmd_alloc_info();
2411 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2412 err = PTR_ERR(vport);
2414 goto exit_unlock_free;
2415 err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2416 info->snd_portid, info->snd_seq, 0,
2417 OVS_VPORT_CMD_GET, GFP_ATOMIC);
2421 return genlmsg_reply(reply, info);
2429 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2431 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2432 struct datapath *dp;
2433 int bucket = cb->args[0], skip = cb->args[1];
2437 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2442 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2443 struct vport *vport;
2446 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2448 ovs_vport_cmd_fill_info(vport, skb,
2450 NETLINK_CB(cb->skb).portid,
2470 static void ovs_dp_masks_rebalance(struct work_struct *work)
2472 struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2473 masks_rebalance.work);
2474 struct datapath *dp;
2478 list_for_each_entry(dp, &ovs_net->dps, list_node)
2479 ovs_flow_masks_rebalance(&dp->table);
2483 schedule_delayed_work(&ovs_net->masks_rebalance,
2484 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2487 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2488 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2489 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2490 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2491 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2492 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2493 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2494 [OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2495 [OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2498 static const struct genl_small_ops dp_vport_genl_ops[] = {
2499 { .cmd = OVS_VPORT_CMD_NEW,
2500 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2501 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2502 .doit = ovs_vport_cmd_new
2504 { .cmd = OVS_VPORT_CMD_DEL,
2505 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2506 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2507 .doit = ovs_vport_cmd_del
2509 { .cmd = OVS_VPORT_CMD_GET,
2510 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2511 .flags = 0, /* OK for unprivileged users. */
2512 .doit = ovs_vport_cmd_get,
2513 .dumpit = ovs_vport_cmd_dump
2515 { .cmd = OVS_VPORT_CMD_SET,
2516 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2517 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2518 .doit = ovs_vport_cmd_set,
2522 struct genl_family dp_vport_genl_family __ro_after_init = {
2523 .hdrsize = sizeof(struct ovs_header),
2524 .name = OVS_VPORT_FAMILY,
2525 .version = OVS_VPORT_VERSION,
2526 .maxattr = OVS_VPORT_ATTR_MAX,
2527 .policy = vport_policy,
2529 .parallel_ops = true,
2530 .small_ops = dp_vport_genl_ops,
2531 .n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2532 .mcgrps = &ovs_dp_vport_multicast_group,
2534 .module = THIS_MODULE,
2537 static struct genl_family * const dp_genl_families[] = {
2538 &dp_datapath_genl_family,
2539 &dp_vport_genl_family,
2540 &dp_flow_genl_family,
2541 &dp_packet_genl_family,
2542 &dp_meter_genl_family,
2543 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2544 &dp_ct_limit_genl_family,
2548 static void dp_unregister_genl(int n_families)
2552 for (i = 0; i < n_families; i++)
2553 genl_unregister_family(dp_genl_families[i]);
2556 static int __init dp_register_genl(void)
2561 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2563 err = genl_register_family(dp_genl_families[i]);
2571 dp_unregister_genl(i);
2575 static int __net_init ovs_init_net(struct net *net)
2577 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2580 INIT_LIST_HEAD(&ovs_net->dps);
2581 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2582 INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2584 err = ovs_ct_init(net);
2588 schedule_delayed_work(&ovs_net->masks_rebalance,
2589 msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2593 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2594 struct list_head *head)
2596 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2597 struct datapath *dp;
2599 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2602 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2603 struct vport *vport;
2605 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2606 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2609 if (dev_net(vport->dev) == dnet)
2610 list_add(&vport->detach_list, head);
2616 static void __net_exit ovs_exit_net(struct net *dnet)
2618 struct datapath *dp, *dp_next;
2619 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2620 struct vport *vport, *vport_next;
2628 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2631 down_read(&net_rwsem);
2633 list_vports_from_net(net, dnet, &head);
2634 up_read(&net_rwsem);
2636 /* Detach all vports from given namespace. */
2637 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2638 list_del(&vport->detach_list);
2639 ovs_dp_detach_port(vport);
2644 cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2645 cancel_work_sync(&ovs_net->dp_notify_work);
2648 static struct pernet_operations ovs_net_ops = {
2649 .init = ovs_init_net,
2650 .exit = ovs_exit_net,
2652 .size = sizeof(struct ovs_net),
2655 static int __init dp_init(void)
2659 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2660 sizeof_field(struct sk_buff, cb));
2662 pr_info("Open vSwitch switching datapath\n");
2664 err = action_fifos_init();
2668 err = ovs_internal_dev_rtnl_link_register();
2670 goto error_action_fifos_exit;
2672 err = ovs_flow_init();
2674 goto error_unreg_rtnl_link;
2676 err = ovs_vport_init();
2678 goto error_flow_exit;
2680 err = register_pernet_device(&ovs_net_ops);
2682 goto error_vport_exit;
2684 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2686 goto error_netns_exit;
2688 err = ovs_netdev_init();
2690 goto error_unreg_notifier;
2692 err = dp_register_genl();
2694 goto error_unreg_netdev;
2700 error_unreg_notifier:
2701 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2703 unregister_pernet_device(&ovs_net_ops);
2708 error_unreg_rtnl_link:
2709 ovs_internal_dev_rtnl_link_unregister();
2710 error_action_fifos_exit:
2711 action_fifos_exit();
2716 static void dp_cleanup(void)
2718 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2720 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2721 unregister_pernet_device(&ovs_net_ops);
2725 ovs_internal_dev_rtnl_link_unregister();
2726 action_fifos_exit();
2729 module_init(dp_init);
2730 module_exit(dp_cleanup);
2732 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2733 MODULE_LICENSE("GPL");
2734 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2735 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2736 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2737 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2738 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2739 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);