2 * Copyright (c) 2007-2014 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
61 unsigned int ovs_net_id __read_mostly;
63 static struct genl_family dp_packet_genl_family;
64 static struct genl_family dp_flow_genl_family;
65 static struct genl_family dp_datapath_genl_family;
67 static const struct nla_policy flow_policy[];
69 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
70 .name = OVS_FLOW_MCGROUP,
73 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
74 .name = OVS_DATAPATH_MCGROUP,
77 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
78 .name = OVS_VPORT_MCGROUP,
81 /* Check if need to build a reply message.
82 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
86 return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
87 genl_has_listeners(family, genl_info_net(info), group);
90 static void ovs_notify(struct genl_family *family,
91 struct sk_buff *skb, struct genl_info *info)
93 genl_notify(family, skb, info, 0, GFP_KERNEL);
99 * All writes e.g. Writes to device state (add/remove datapath, port, set
100 * operations on vports, etc.), Writes to other state (flow table
101 * modifications, set miscellaneous datapath parameters, etc.) are protected
104 * Reads are protected by RCU.
106 * There are a few special cases (mostly stats) that have their own
107 * synchronization but they nest under all of above and don't interact with
110 * The RTNL lock nests inside ovs_mutex.
113 static DEFINE_MUTEX(ovs_mutex);
117 mutex_lock(&ovs_mutex);
120 void ovs_unlock(void)
122 mutex_unlock(&ovs_mutex);
125 #ifdef CONFIG_LOCKDEP
126 int lockdep_ovsl_is_held(void)
129 return lockdep_is_held(&ovs_mutex);
135 static struct vport *new_vport(const struct vport_parms *);
136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
137 const struct sw_flow_key *,
138 const struct dp_upcall_info *,
140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 const struct sw_flow_key *,
142 const struct dp_upcall_info *,
145 /* Must be called with rcu_read_lock. */
146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
148 struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
151 struct vport *vport = ovs_internal_dev_get_vport(dev);
159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160 * returned dp pointer valid.
162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
166 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
168 dp = get_dp_rcu(net, dp_ifindex);
174 /* Must be called with rcu_read_lock or ovs_mutex. */
175 const char *ovs_dp_name(const struct datapath *dp)
177 struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
178 return ovs_vport_name(vport);
181 static int get_dpifindex(const struct datapath *dp)
188 local = ovs_vport_rcu(dp, OVSP_LOCAL);
190 ifindex = local->dev->ifindex;
199 static void destroy_dp_rcu(struct rcu_head *rcu)
201 struct datapath *dp = container_of(rcu, struct datapath, rcu);
203 ovs_flow_tbl_destroy(&dp->table);
204 free_percpu(dp->stats_percpu);
209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
212 return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
215 /* Called with ovs_mutex or RCU read lock. */
216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
219 struct hlist_head *head;
221 head = vport_hash_bucket(dp, port_no);
222 hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
223 if (vport->port_no == port_no)
229 /* Called with ovs_mutex. */
230 static struct vport *new_vport(const struct vport_parms *parms)
234 vport = ovs_vport_add(parms);
235 if (!IS_ERR(vport)) {
236 struct datapath *dp = parms->dp;
237 struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
239 hlist_add_head_rcu(&vport->dp_hash_node, head);
244 void ovs_dp_detach_port(struct vport *p)
248 /* First drop references to device. */
249 hlist_del_rcu(&p->dp_hash_node);
251 /* Then destroy it. */
255 /* Must be called with rcu_read_lock. */
256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
258 const struct vport *p = OVS_CB(skb)->input_vport;
259 struct datapath *dp = p->dp;
260 struct sw_flow *flow;
261 struct sw_flow_actions *sf_acts;
262 struct dp_stats_percpu *stats;
266 stats = this_cpu_ptr(dp->stats_percpu);
269 flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
270 if (unlikely(!flow)) {
271 struct dp_upcall_info upcall;
274 memset(&upcall, 0, sizeof(upcall));
275 upcall.cmd = OVS_PACKET_CMD_MISS;
276 upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 upcall.mru = OVS_CB(skb)->mru;
278 error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
283 stats_counter = &stats->n_missed;
287 ovs_flow_stats_update(flow, key->tp.flags, skb);
288 sf_acts = rcu_dereference(flow->sf_acts);
289 ovs_execute_actions(dp, skb, sf_acts, key);
291 stats_counter = &stats->n_hit;
294 /* Update datapath statistics. */
295 u64_stats_update_begin(&stats->syncp);
297 stats->n_mask_hit += n_mask_hit;
298 u64_stats_update_end(&stats->syncp);
301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
302 const struct sw_flow_key *key,
303 const struct dp_upcall_info *upcall_info,
306 struct dp_stats_percpu *stats;
309 if (upcall_info->portid == 0) {
314 if (!skb_is_gso(skb))
315 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
317 err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
324 stats = this_cpu_ptr(dp->stats_percpu);
326 u64_stats_update_begin(&stats->syncp);
328 u64_stats_update_end(&stats->syncp);
333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
334 const struct sw_flow_key *key,
335 const struct dp_upcall_info *upcall_info,
338 unsigned short gso_type = skb_shinfo(skb)->gso_type;
339 struct sw_flow_key later_key;
340 struct sk_buff *segs, *nskb;
343 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
344 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
346 return PTR_ERR(segs);
350 if (gso_type & SKB_GSO_UDP) {
351 /* The initial flow key extracted by ovs_flow_key_extract()
352 * in this case is for a first fragment, so we need to
353 * properly mark later fragments.
356 later_key.ip.frag = OVS_FRAG_TYPE_LATER;
359 /* Queue all of the segments. */
362 if (gso_type & SKB_GSO_UDP && skb != segs)
365 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
369 } while ((skb = skb->next));
371 /* Free all of the segments. */
379 } while ((skb = nskb));
383 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
386 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
387 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
388 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
389 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
391 /* OVS_PACKET_ATTR_USERDATA */
392 if (upcall_info->userdata)
393 size += NLA_ALIGN(upcall_info->userdata->nla_len);
395 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
396 if (upcall_info->egress_tun_info)
397 size += nla_total_size(ovs_tun_key_attr_size());
399 /* OVS_PACKET_ATTR_ACTIONS */
400 if (upcall_info->actions_len)
401 size += nla_total_size(upcall_info->actions_len);
403 /* OVS_PACKET_ATTR_MRU */
404 if (upcall_info->mru)
405 size += nla_total_size(sizeof(upcall_info->mru));
410 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
412 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
413 size_t plen = NLA_ALIGN(skb->len) - skb->len;
416 memset(skb_put(skb, plen), 0, plen);
420 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
421 const struct sw_flow_key *key,
422 const struct dp_upcall_info *upcall_info,
425 struct ovs_header *upcall;
426 struct sk_buff *nskb = NULL;
427 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
433 dp_ifindex = get_dpifindex(dp);
437 if (skb_vlan_tag_present(skb)) {
438 nskb = skb_clone(skb, GFP_ATOMIC);
442 nskb = __vlan_hwaccel_push_inside(nskb);
449 if (nla_attr_size(skb->len) > USHRT_MAX) {
454 /* Complete checksum if needed */
455 if (skb->ip_summed == CHECKSUM_PARTIAL &&
456 (err = skb_checksum_help(skb)))
459 /* Older versions of OVS user space enforce alignment of the last
460 * Netlink attribute to NLA_ALIGNTO which would require extensive
461 * padding logic. Only perform zerocopy if padding is not required.
463 if (dp->user_features & OVS_DP_F_UNALIGNED)
464 hlen = skb_zerocopy_headlen(skb);
468 len = upcall_msg_size(upcall_info, hlen - cutlen);
469 user_skb = genlmsg_new(len, GFP_ATOMIC);
475 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
476 0, upcall_info->cmd);
477 upcall->dp_ifindex = dp_ifindex;
479 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
482 if (upcall_info->userdata)
483 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
484 nla_len(upcall_info->userdata),
485 nla_data(upcall_info->userdata));
487 if (upcall_info->egress_tun_info) {
488 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
489 err = ovs_nla_put_tunnel_info(user_skb,
490 upcall_info->egress_tun_info);
492 nla_nest_end(user_skb, nla);
495 if (upcall_info->actions_len) {
496 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
497 err = ovs_nla_put_actions(upcall_info->actions,
498 upcall_info->actions_len,
501 nla_nest_end(user_skb, nla);
503 nla_nest_cancel(user_skb, nla);
506 /* Add OVS_PACKET_ATTR_MRU */
507 if (upcall_info->mru) {
508 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
513 pad_packet(dp, user_skb);
516 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
518 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
523 pad_packet(dp, user_skb);
526 /* Only reserve room for attribute header, packet data is added
527 * in skb_zerocopy() */
528 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
532 nla->nla_len = nla_attr_size(skb->len - cutlen);
534 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
538 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
539 pad_packet(dp, user_skb);
541 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
543 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
553 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
555 struct ovs_header *ovs_header = info->userhdr;
556 struct net *net = sock_net(skb->sk);
557 struct nlattr **a = info->attrs;
558 struct sw_flow_actions *acts;
559 struct sk_buff *packet;
560 struct sw_flow *flow;
561 struct sw_flow_actions *sf_acts;
563 struct vport *input_vport;
567 bool log = !a[OVS_PACKET_ATTR_PROBE];
570 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
571 !a[OVS_PACKET_ATTR_ACTIONS])
574 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
575 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
579 skb_reserve(packet, NET_IP_ALIGN);
581 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
583 /* Set packet's mru */
584 if (a[OVS_PACKET_ATTR_MRU]) {
585 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
586 packet->ignore_df = 1;
588 OVS_CB(packet)->mru = mru;
590 /* Build an sw_flow for sending this packet. */
591 flow = ovs_flow_alloc();
596 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
597 packet, &flow->key, log);
601 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
602 &flow->key, &acts, log);
606 rcu_assign_pointer(flow->sf_acts, acts);
607 packet->priority = flow->key.phy.priority;
608 packet->mark = flow->key.phy.skb_mark;
609 packet->protocol = flow->key.eth.type;
612 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
617 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
619 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
624 packet->dev = input_vport->dev;
625 OVS_CB(packet)->input_vport = input_vport;
626 sf_acts = rcu_dereference(flow->sf_acts);
629 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
633 ovs_flow_free(flow, false);
639 ovs_flow_free(flow, false);
646 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
647 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
648 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
649 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
650 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
651 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
654 static const struct genl_ops dp_packet_genl_ops[] = {
655 { .cmd = OVS_PACKET_CMD_EXECUTE,
656 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
657 .policy = packet_policy,
658 .doit = ovs_packet_cmd_execute
662 static struct genl_family dp_packet_genl_family __ro_after_init = {
663 .hdrsize = sizeof(struct ovs_header),
664 .name = OVS_PACKET_FAMILY,
665 .version = OVS_PACKET_VERSION,
666 .maxattr = OVS_PACKET_ATTR_MAX,
668 .parallel_ops = true,
669 .ops = dp_packet_genl_ops,
670 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
671 .module = THIS_MODULE,
674 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
675 struct ovs_dp_megaflow_stats *mega_stats)
679 memset(mega_stats, 0, sizeof(*mega_stats));
681 stats->n_flows = ovs_flow_tbl_count(&dp->table);
682 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
684 stats->n_hit = stats->n_missed = stats->n_lost = 0;
686 for_each_possible_cpu(i) {
687 const struct dp_stats_percpu *percpu_stats;
688 struct dp_stats_percpu local_stats;
691 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
694 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
695 local_stats = *percpu_stats;
696 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
698 stats->n_hit += local_stats.n_hit;
699 stats->n_missed += local_stats.n_missed;
700 stats->n_lost += local_stats.n_lost;
701 mega_stats->n_mask_hit += local_stats.n_mask_hit;
705 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
707 return ovs_identifier_is_ufid(sfid) &&
708 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
711 static bool should_fill_mask(uint32_t ufid_flags)
713 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
716 static bool should_fill_actions(uint32_t ufid_flags)
718 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
721 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
722 const struct sw_flow_id *sfid,
725 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
727 /* OVS_FLOW_ATTR_UFID */
728 if (sfid && ovs_identifier_is_ufid(sfid))
729 len += nla_total_size(sfid->ufid_len);
731 /* OVS_FLOW_ATTR_KEY */
732 if (!sfid || should_fill_key(sfid, ufid_flags))
733 len += nla_total_size(ovs_key_attr_size());
735 /* OVS_FLOW_ATTR_MASK */
736 if (should_fill_mask(ufid_flags))
737 len += nla_total_size(ovs_key_attr_size());
739 /* OVS_FLOW_ATTR_ACTIONS */
740 if (should_fill_actions(ufid_flags))
741 len += nla_total_size(acts->orig_len);
744 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
745 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
746 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
749 /* Called with ovs_mutex or RCU read lock. */
750 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
753 struct ovs_flow_stats stats;
757 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
760 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
764 if (stats.n_packets &&
765 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
766 sizeof(struct ovs_flow_stats), &stats,
770 if ((u8)ntohs(tcp_flags) &&
771 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
777 /* Called with ovs_mutex or RCU read lock. */
778 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
779 struct sk_buff *skb, int skb_orig_len)
781 struct nlattr *start;
784 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
785 * this is the first flow to be dumped into 'skb'. This is unusual for
786 * Netlink but individual action lists can be longer than
787 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
788 * The userspace caller can always fetch the actions separately if it
789 * really wants them. (Most userspace callers in fact don't care.)
791 * This can only fail for dump operations because the skb is always
792 * properly sized for single flows.
794 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
796 const struct sw_flow_actions *sf_acts;
798 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
799 err = ovs_nla_put_actions(sf_acts->actions,
800 sf_acts->actions_len, skb);
803 nla_nest_end(skb, start);
808 nla_nest_cancel(skb, start);
810 } else if (skb_orig_len) {
817 /* Called with ovs_mutex or RCU read lock. */
818 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
819 struct sk_buff *skb, u32 portid,
820 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
822 const int skb_orig_len = skb->len;
823 struct ovs_header *ovs_header;
826 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
831 ovs_header->dp_ifindex = dp_ifindex;
833 err = ovs_nla_put_identifier(flow, skb);
837 if (should_fill_key(&flow->id, ufid_flags)) {
838 err = ovs_nla_put_masked_key(flow, skb);
843 if (should_fill_mask(ufid_flags)) {
844 err = ovs_nla_put_mask(flow, skb);
849 err = ovs_flow_cmd_fill_stats(flow, skb);
853 if (should_fill_actions(ufid_flags)) {
854 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
859 genlmsg_end(skb, ovs_header);
863 genlmsg_cancel(skb, ovs_header);
867 /* May not be called with RCU read lock. */
868 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
869 const struct sw_flow_id *sfid,
870 struct genl_info *info,
877 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
880 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
881 skb = genlmsg_new(len, GFP_KERNEL);
883 return ERR_PTR(-ENOMEM);
888 /* Called with ovs_mutex. */
889 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
891 struct genl_info *info, u8 cmd,
892 bool always, u32 ufid_flags)
897 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
898 &flow->id, info, always, ufid_flags);
899 if (IS_ERR_OR_NULL(skb))
902 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
903 info->snd_portid, info->snd_seq, 0,
909 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
911 struct net *net = sock_net(skb->sk);
912 struct nlattr **a = info->attrs;
913 struct ovs_header *ovs_header = info->userhdr;
914 struct sw_flow *flow = NULL, *new_flow;
915 struct sw_flow_mask mask;
916 struct sk_buff *reply;
918 struct sw_flow_actions *acts;
919 struct sw_flow_match match;
920 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
922 bool log = !a[OVS_FLOW_ATTR_PROBE];
924 /* Must have key and actions. */
926 if (!a[OVS_FLOW_ATTR_KEY]) {
927 OVS_NLERR(log, "Flow key attr not present in new flow.");
930 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
931 OVS_NLERR(log, "Flow actions attr not present in new flow.");
935 /* Most of the time we need to allocate a new flow, do it before
938 new_flow = ovs_flow_alloc();
939 if (IS_ERR(new_flow)) {
940 error = PTR_ERR(new_flow);
945 ovs_match_init(&match, &new_flow->key, false, &mask);
946 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
947 a[OVS_FLOW_ATTR_MASK], log);
951 /* Extract flow identifier. */
952 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
953 &new_flow->key, log);
957 /* unmasked key is needed to match when ufid is not used. */
958 if (ovs_identifier_is_key(&new_flow->id))
959 match.key = new_flow->id.unmasked_key;
961 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
963 /* Validate actions. */
964 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
965 &new_flow->key, &acts, log);
967 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
971 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
974 error = PTR_ERR(reply);
979 dp = get_dp(net, ovs_header->dp_ifindex);
985 /* Check if this is a duplicate flow */
986 if (ovs_identifier_is_ufid(&new_flow->id))
987 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
989 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
991 rcu_assign_pointer(new_flow->sf_acts, acts);
993 /* Put flow in bucket. */
994 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
995 if (unlikely(error)) {
1000 if (unlikely(reply)) {
1001 error = ovs_flow_cmd_fill_info(new_flow,
1002 ovs_header->dp_ifindex,
1003 reply, info->snd_portid,
1011 struct sw_flow_actions *old_acts;
1013 /* Bail out if we're not allowed to modify an existing flow.
1014 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1015 * because Generic Netlink treats the latter as a dump
1016 * request. We also accept NLM_F_EXCL in case that bug ever
1019 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1022 goto err_unlock_ovs;
1024 /* The flow identifier has to be the same for flow updates.
1025 * Look for any overlapping flow.
1027 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1028 if (ovs_identifier_is_key(&flow->id))
1029 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1031 else /* UFID matches but key is different */
1035 goto err_unlock_ovs;
1038 /* Update actions. */
1039 old_acts = ovsl_dereference(flow->sf_acts);
1040 rcu_assign_pointer(flow->sf_acts, acts);
1042 if (unlikely(reply)) {
1043 error = ovs_flow_cmd_fill_info(flow,
1044 ovs_header->dp_ifindex,
1045 reply, info->snd_portid,
1053 ovs_nla_free_flow_actions_rcu(old_acts);
1054 ovs_flow_free(new_flow, false);
1058 ovs_notify(&dp_flow_genl_family, reply, info);
1065 ovs_nla_free_flow_actions(acts);
1067 ovs_flow_free(new_flow, false);
1072 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1073 static struct sw_flow_actions *get_flow_actions(struct net *net,
1074 const struct nlattr *a,
1075 const struct sw_flow_key *key,
1076 const struct sw_flow_mask *mask,
1079 struct sw_flow_actions *acts;
1080 struct sw_flow_key masked_key;
1083 ovs_flow_mask_key(&masked_key, key, true, mask);
1084 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1087 "Actions may not be safe on all matching packets");
1088 return ERR_PTR(error);
1094 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1096 struct net *net = sock_net(skb->sk);
1097 struct nlattr **a = info->attrs;
1098 struct ovs_header *ovs_header = info->userhdr;
1099 struct sw_flow_key key;
1100 struct sw_flow *flow;
1101 struct sw_flow_mask mask;
1102 struct sk_buff *reply = NULL;
1103 struct datapath *dp;
1104 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1105 struct sw_flow_match match;
1106 struct sw_flow_id sfid;
1107 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1109 bool log = !a[OVS_FLOW_ATTR_PROBE];
1112 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1113 if (a[OVS_FLOW_ATTR_KEY]) {
1114 ovs_match_init(&match, &key, true, &mask);
1115 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1116 a[OVS_FLOW_ATTR_MASK], log);
1117 } else if (!ufid_present) {
1119 "Flow set message rejected, Key attribute missing.");
1125 /* Validate actions. */
1126 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1127 if (!a[OVS_FLOW_ATTR_KEY]) {
1129 "Flow key attribute not present in set flow.");
1134 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1137 error = PTR_ERR(acts);
1141 /* Can allocate before locking if have acts. */
1142 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1144 if (IS_ERR(reply)) {
1145 error = PTR_ERR(reply);
1146 goto err_kfree_acts;
1151 dp = get_dp(net, ovs_header->dp_ifindex);
1152 if (unlikely(!dp)) {
1154 goto err_unlock_ovs;
1156 /* Check that the flow exists. */
1158 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1160 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1161 if (unlikely(!flow)) {
1163 goto err_unlock_ovs;
1166 /* Update actions, if present. */
1168 old_acts = ovsl_dereference(flow->sf_acts);
1169 rcu_assign_pointer(flow->sf_acts, acts);
1171 if (unlikely(reply)) {
1172 error = ovs_flow_cmd_fill_info(flow,
1173 ovs_header->dp_ifindex,
1174 reply, info->snd_portid,
1181 /* Could not alloc without acts before locking. */
1182 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1183 info, OVS_FLOW_CMD_NEW, false,
1186 if (IS_ERR(reply)) {
1187 error = PTR_ERR(reply);
1188 goto err_unlock_ovs;
1193 if (a[OVS_FLOW_ATTR_CLEAR])
1194 ovs_flow_stats_clear(flow);
1198 ovs_notify(&dp_flow_genl_family, reply, info);
1200 ovs_nla_free_flow_actions_rcu(old_acts);
1208 ovs_nla_free_flow_actions(acts);
1213 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1215 struct nlattr **a = info->attrs;
1216 struct ovs_header *ovs_header = info->userhdr;
1217 struct net *net = sock_net(skb->sk);
1218 struct sw_flow_key key;
1219 struct sk_buff *reply;
1220 struct sw_flow *flow;
1221 struct datapath *dp;
1222 struct sw_flow_match match;
1223 struct sw_flow_id ufid;
1224 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1226 bool log = !a[OVS_FLOW_ATTR_PROBE];
1229 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1230 if (a[OVS_FLOW_ATTR_KEY]) {
1231 ovs_match_init(&match, &key, true, NULL);
1232 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1234 } else if (!ufid_present) {
1236 "Flow get message rejected, Key attribute missing.");
1243 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1250 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1252 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1258 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1259 OVS_FLOW_CMD_NEW, true, ufid_flags);
1260 if (IS_ERR(reply)) {
1261 err = PTR_ERR(reply);
1266 return genlmsg_reply(reply, info);
1272 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1274 struct nlattr **a = info->attrs;
1275 struct ovs_header *ovs_header = info->userhdr;
1276 struct net *net = sock_net(skb->sk);
1277 struct sw_flow_key key;
1278 struct sk_buff *reply;
1279 struct sw_flow *flow = NULL;
1280 struct datapath *dp;
1281 struct sw_flow_match match;
1282 struct sw_flow_id ufid;
1283 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1285 bool log = !a[OVS_FLOW_ATTR_PROBE];
1288 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1289 if (a[OVS_FLOW_ATTR_KEY]) {
1290 ovs_match_init(&match, &key, true, NULL);
1291 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1298 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1299 if (unlikely(!dp)) {
1304 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1305 err = ovs_flow_tbl_flush(&dp->table);
1310 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1312 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1313 if (unlikely(!flow)) {
1318 ovs_flow_tbl_remove(&dp->table, flow);
1321 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1322 &flow->id, info, false, ufid_flags);
1323 if (likely(reply)) {
1324 if (likely(!IS_ERR(reply))) {
1325 rcu_read_lock(); /*To keep RCU checker happy. */
1326 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1327 reply, info->snd_portid,
1334 ovs_notify(&dp_flow_genl_family, reply, info);
1336 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1340 ovs_flow_free(flow, true);
1347 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1349 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1350 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1351 struct table_instance *ti;
1352 struct datapath *dp;
1356 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1357 OVS_FLOW_ATTR_MAX, flow_policy);
1360 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1363 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1369 ti = rcu_dereference(dp->table.ti);
1371 struct sw_flow *flow;
1374 bucket = cb->args[0];
1376 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1380 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1381 NETLINK_CB(cb->skb).portid,
1382 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1383 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1386 cb->args[0] = bucket;
1393 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1394 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1395 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1396 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1397 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1398 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1399 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1400 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1403 static const struct genl_ops dp_flow_genl_ops[] = {
1404 { .cmd = OVS_FLOW_CMD_NEW,
1405 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1406 .policy = flow_policy,
1407 .doit = ovs_flow_cmd_new
1409 { .cmd = OVS_FLOW_CMD_DEL,
1410 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1411 .policy = flow_policy,
1412 .doit = ovs_flow_cmd_del
1414 { .cmd = OVS_FLOW_CMD_GET,
1415 .flags = 0, /* OK for unprivileged users. */
1416 .policy = flow_policy,
1417 .doit = ovs_flow_cmd_get,
1418 .dumpit = ovs_flow_cmd_dump
1420 { .cmd = OVS_FLOW_CMD_SET,
1421 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1422 .policy = flow_policy,
1423 .doit = ovs_flow_cmd_set,
1427 static struct genl_family dp_flow_genl_family __ro_after_init = {
1428 .hdrsize = sizeof(struct ovs_header),
1429 .name = OVS_FLOW_FAMILY,
1430 .version = OVS_FLOW_VERSION,
1431 .maxattr = OVS_FLOW_ATTR_MAX,
1433 .parallel_ops = true,
1434 .ops = dp_flow_genl_ops,
1435 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1436 .mcgrps = &ovs_dp_flow_multicast_group,
1438 .module = THIS_MODULE,
1441 static size_t ovs_dp_cmd_msg_size(void)
1443 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1445 msgsize += nla_total_size(IFNAMSIZ);
1446 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1447 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1448 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1453 /* Called with ovs_mutex. */
1454 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1455 u32 portid, u32 seq, u32 flags, u8 cmd)
1457 struct ovs_header *ovs_header;
1458 struct ovs_dp_stats dp_stats;
1459 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1462 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1467 ovs_header->dp_ifindex = get_dpifindex(dp);
1469 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1471 goto nla_put_failure;
1473 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1474 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1475 &dp_stats, OVS_DP_ATTR_PAD))
1476 goto nla_put_failure;
1478 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1479 sizeof(struct ovs_dp_megaflow_stats),
1480 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1481 goto nla_put_failure;
1483 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1484 goto nla_put_failure;
1486 genlmsg_end(skb, ovs_header);
1490 genlmsg_cancel(skb, ovs_header);
1495 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1497 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1500 /* Called with rcu_read_lock or ovs_mutex. */
1501 static struct datapath *lookup_datapath(struct net *net,
1502 const struct ovs_header *ovs_header,
1503 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1505 struct datapath *dp;
1507 if (!a[OVS_DP_ATTR_NAME])
1508 dp = get_dp(net, ovs_header->dp_ifindex);
1510 struct vport *vport;
1512 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1513 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1515 return dp ? dp : ERR_PTR(-ENODEV);
1518 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1520 struct datapath *dp;
1522 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1526 WARN(dp->user_features, "Dropping previously announced user features\n");
1527 dp->user_features = 0;
1530 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1532 if (a[OVS_DP_ATTR_USER_FEATURES])
1533 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1536 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1538 struct nlattr **a = info->attrs;
1539 struct vport_parms parms;
1540 struct sk_buff *reply;
1541 struct datapath *dp;
1542 struct vport *vport;
1543 struct ovs_net *ovs_net;
1547 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1550 reply = ovs_dp_cmd_alloc_info();
1555 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1557 goto err_free_reply;
1559 ovs_dp_set_net(dp, sock_net(skb->sk));
1561 /* Allocate table. */
1562 err = ovs_flow_tbl_init(&dp->table);
1566 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1567 if (!dp->stats_percpu) {
1569 goto err_destroy_table;
1572 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1576 goto err_destroy_percpu;
1579 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1580 INIT_HLIST_HEAD(&dp->ports[i]);
1582 /* Set up our datapath device. */
1583 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1584 parms.type = OVS_VPORT_TYPE_INTERNAL;
1585 parms.options = NULL;
1587 parms.port_no = OVSP_LOCAL;
1588 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1590 ovs_dp_change(dp, a);
1592 /* So far only local changes have been made, now need the lock. */
1595 vport = new_vport(&parms);
1596 if (IS_ERR(vport)) {
1597 err = PTR_ERR(vport);
1601 if (err == -EEXIST) {
1602 /* An outdated user space instance that does not understand
1603 * the concept of user_features has attempted to create a new
1604 * datapath and is likely to reuse it. Drop all user features.
1606 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1607 ovs_dp_reset_user_features(skb, info);
1610 goto err_destroy_ports_array;
1613 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1614 info->snd_seq, 0, OVS_DP_CMD_NEW);
1617 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1618 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1622 ovs_notify(&dp_datapath_genl_family, reply, info);
1625 err_destroy_ports_array:
1629 free_percpu(dp->stats_percpu);
1631 ovs_flow_tbl_destroy(&dp->table);
1640 /* Called with ovs_mutex. */
1641 static void __dp_destroy(struct datapath *dp)
1645 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1646 struct vport *vport;
1647 struct hlist_node *n;
1649 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1650 if (vport->port_no != OVSP_LOCAL)
1651 ovs_dp_detach_port(vport);
1654 list_del_rcu(&dp->list_node);
1656 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1657 * all ports in datapath are destroyed first before freeing datapath.
1659 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1661 /* RCU destroy the flow table */
1662 call_rcu(&dp->rcu, destroy_dp_rcu);
1665 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1667 struct sk_buff *reply;
1668 struct datapath *dp;
1671 reply = ovs_dp_cmd_alloc_info();
1676 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1679 goto err_unlock_free;
1681 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1682 info->snd_seq, 0, OVS_DP_CMD_DEL);
1688 ovs_notify(&dp_datapath_genl_family, reply, info);
1698 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1700 struct sk_buff *reply;
1701 struct datapath *dp;
1704 reply = ovs_dp_cmd_alloc_info();
1709 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1712 goto err_unlock_free;
1714 ovs_dp_change(dp, info->attrs);
1716 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1717 info->snd_seq, 0, OVS_DP_CMD_NEW);
1721 ovs_notify(&dp_datapath_genl_family, reply, info);
1731 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1733 struct sk_buff *reply;
1734 struct datapath *dp;
1737 reply = ovs_dp_cmd_alloc_info();
1742 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1745 goto err_unlock_free;
1747 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1748 info->snd_seq, 0, OVS_DP_CMD_NEW);
1752 return genlmsg_reply(reply, info);
1760 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1762 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1763 struct datapath *dp;
1764 int skip = cb->args[0];
1768 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1770 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1771 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1772 OVS_DP_CMD_NEW) < 0)
1783 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1784 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1785 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1786 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1789 static const struct genl_ops dp_datapath_genl_ops[] = {
1790 { .cmd = OVS_DP_CMD_NEW,
1791 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1792 .policy = datapath_policy,
1793 .doit = ovs_dp_cmd_new
1795 { .cmd = OVS_DP_CMD_DEL,
1796 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1797 .policy = datapath_policy,
1798 .doit = ovs_dp_cmd_del
1800 { .cmd = OVS_DP_CMD_GET,
1801 .flags = 0, /* OK for unprivileged users. */
1802 .policy = datapath_policy,
1803 .doit = ovs_dp_cmd_get,
1804 .dumpit = ovs_dp_cmd_dump
1806 { .cmd = OVS_DP_CMD_SET,
1807 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1808 .policy = datapath_policy,
1809 .doit = ovs_dp_cmd_set,
1813 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1814 .hdrsize = sizeof(struct ovs_header),
1815 .name = OVS_DATAPATH_FAMILY,
1816 .version = OVS_DATAPATH_VERSION,
1817 .maxattr = OVS_DP_ATTR_MAX,
1819 .parallel_ops = true,
1820 .ops = dp_datapath_genl_ops,
1821 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1822 .mcgrps = &ovs_dp_datapath_multicast_group,
1824 .module = THIS_MODULE,
1827 /* Called with ovs_mutex or RCU read lock. */
1828 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1829 u32 portid, u32 seq, u32 flags, u8 cmd)
1831 struct ovs_header *ovs_header;
1832 struct ovs_vport_stats vport_stats;
1835 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1840 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1842 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1843 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1844 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1845 ovs_vport_name(vport)))
1846 goto nla_put_failure;
1848 ovs_vport_get_stats(vport, &vport_stats);
1849 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1850 sizeof(struct ovs_vport_stats), &vport_stats,
1851 OVS_VPORT_ATTR_PAD))
1852 goto nla_put_failure;
1854 if (ovs_vport_get_upcall_portids(vport, skb))
1855 goto nla_put_failure;
1857 err = ovs_vport_get_options(vport, skb);
1858 if (err == -EMSGSIZE)
1861 genlmsg_end(skb, ovs_header);
1867 genlmsg_cancel(skb, ovs_header);
1871 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1873 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1876 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1877 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1880 struct sk_buff *skb;
1883 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1885 return ERR_PTR(-ENOMEM);
1887 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1893 /* Called with ovs_mutex or RCU read lock. */
1894 static struct vport *lookup_vport(struct net *net,
1895 const struct ovs_header *ovs_header,
1896 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1898 struct datapath *dp;
1899 struct vport *vport;
1901 if (a[OVS_VPORT_ATTR_NAME]) {
1902 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1904 return ERR_PTR(-ENODEV);
1905 if (ovs_header->dp_ifindex &&
1906 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1907 return ERR_PTR(-ENODEV);
1909 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1910 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1912 if (port_no >= DP_MAX_PORTS)
1913 return ERR_PTR(-EFBIG);
1915 dp = get_dp(net, ovs_header->dp_ifindex);
1917 return ERR_PTR(-ENODEV);
1919 vport = ovs_vport_ovsl_rcu(dp, port_no);
1921 return ERR_PTR(-ENODEV);
1924 return ERR_PTR(-EINVAL);
1927 /* Called with ovs_mutex */
1928 static void update_headroom(struct datapath *dp)
1930 unsigned dev_headroom, max_headroom = 0;
1931 struct net_device *dev;
1932 struct vport *vport;
1935 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1936 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1938 dev_headroom = netdev_get_fwd_headroom(dev);
1939 if (dev_headroom > max_headroom)
1940 max_headroom = dev_headroom;
1944 dp->max_headroom = max_headroom;
1945 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1946 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1947 netdev_set_rx_headroom(vport->dev, max_headroom);
1950 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1952 struct nlattr **a = info->attrs;
1953 struct ovs_header *ovs_header = info->userhdr;
1954 struct vport_parms parms;
1955 struct sk_buff *reply;
1956 struct vport *vport;
1957 struct datapath *dp;
1961 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1962 !a[OVS_VPORT_ATTR_UPCALL_PID])
1965 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1966 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1967 if (port_no >= DP_MAX_PORTS)
1970 reply = ovs_vport_cmd_alloc_info();
1976 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1979 goto exit_unlock_free;
1982 vport = ovs_vport_ovsl(dp, port_no);
1985 goto exit_unlock_free;
1987 for (port_no = 1; ; port_no++) {
1988 if (port_no >= DP_MAX_PORTS) {
1990 goto exit_unlock_free;
1992 vport = ovs_vport_ovsl(dp, port_no);
1998 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1999 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2000 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2002 parms.port_no = port_no;
2003 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2005 vport = new_vport(&parms);
2006 err = PTR_ERR(vport);
2007 if (IS_ERR(vport)) {
2010 goto exit_unlock_free;
2013 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2014 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2016 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2017 update_headroom(dp);
2019 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2024 ovs_notify(&dp_vport_genl_family, reply, info);
2033 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2035 struct nlattr **a = info->attrs;
2036 struct sk_buff *reply;
2037 struct vport *vport;
2040 reply = ovs_vport_cmd_alloc_info();
2045 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2046 err = PTR_ERR(vport);
2048 goto exit_unlock_free;
2050 if (a[OVS_VPORT_ATTR_TYPE] &&
2051 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2053 goto exit_unlock_free;
2056 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2057 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2059 goto exit_unlock_free;
2063 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2064 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2066 err = ovs_vport_set_upcall_portids(vport, ids);
2068 goto exit_unlock_free;
2071 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2072 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2076 ovs_notify(&dp_vport_genl_family, reply, info);
2085 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2087 bool must_update_headroom = false;
2088 struct nlattr **a = info->attrs;
2089 struct sk_buff *reply;
2090 struct datapath *dp;
2091 struct vport *vport;
2094 reply = ovs_vport_cmd_alloc_info();
2099 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2100 err = PTR_ERR(vport);
2102 goto exit_unlock_free;
2104 if (vport->port_no == OVSP_LOCAL) {
2106 goto exit_unlock_free;
2109 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2110 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2113 /* the vport deletion may trigger dp headroom update */
2115 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2116 must_update_headroom = true;
2117 netdev_reset_rx_headroom(vport->dev);
2118 ovs_dp_detach_port(vport);
2120 if (must_update_headroom)
2121 update_headroom(dp);
2124 ovs_notify(&dp_vport_genl_family, reply, info);
2133 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2135 struct nlattr **a = info->attrs;
2136 struct ovs_header *ovs_header = info->userhdr;
2137 struct sk_buff *reply;
2138 struct vport *vport;
2141 reply = ovs_vport_cmd_alloc_info();
2146 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2147 err = PTR_ERR(vport);
2149 goto exit_unlock_free;
2150 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2151 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2155 return genlmsg_reply(reply, info);
2163 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2165 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2166 struct datapath *dp;
2167 int bucket = cb->args[0], skip = cb->args[1];
2171 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2176 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2177 struct vport *vport;
2180 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2182 ovs_vport_cmd_fill_info(vport, skb,
2183 NETLINK_CB(cb->skb).portid,
2186 OVS_VPORT_CMD_NEW) < 0)
2202 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2203 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2204 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2205 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2206 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2207 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2208 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2211 static const struct genl_ops dp_vport_genl_ops[] = {
2212 { .cmd = OVS_VPORT_CMD_NEW,
2213 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2214 .policy = vport_policy,
2215 .doit = ovs_vport_cmd_new
2217 { .cmd = OVS_VPORT_CMD_DEL,
2218 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2219 .policy = vport_policy,
2220 .doit = ovs_vport_cmd_del
2222 { .cmd = OVS_VPORT_CMD_GET,
2223 .flags = 0, /* OK for unprivileged users. */
2224 .policy = vport_policy,
2225 .doit = ovs_vport_cmd_get,
2226 .dumpit = ovs_vport_cmd_dump
2228 { .cmd = OVS_VPORT_CMD_SET,
2229 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2230 .policy = vport_policy,
2231 .doit = ovs_vport_cmd_set,
2235 struct genl_family dp_vport_genl_family __ro_after_init = {
2236 .hdrsize = sizeof(struct ovs_header),
2237 .name = OVS_VPORT_FAMILY,
2238 .version = OVS_VPORT_VERSION,
2239 .maxattr = OVS_VPORT_ATTR_MAX,
2241 .parallel_ops = true,
2242 .ops = dp_vport_genl_ops,
2243 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2244 .mcgrps = &ovs_dp_vport_multicast_group,
2246 .module = THIS_MODULE,
2249 static struct genl_family * const dp_genl_families[] = {
2250 &dp_datapath_genl_family,
2251 &dp_vport_genl_family,
2252 &dp_flow_genl_family,
2253 &dp_packet_genl_family,
2256 static void dp_unregister_genl(int n_families)
2260 for (i = 0; i < n_families; i++)
2261 genl_unregister_family(dp_genl_families[i]);
2264 static int __init dp_register_genl(void)
2269 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2271 err = genl_register_family(dp_genl_families[i]);
2279 dp_unregister_genl(i);
2283 static int __net_init ovs_init_net(struct net *net)
2285 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2287 INIT_LIST_HEAD(&ovs_net->dps);
2288 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2293 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2294 struct list_head *head)
2296 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2297 struct datapath *dp;
2299 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2302 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2303 struct vport *vport;
2305 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2306 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2309 if (dev_net(vport->dev) == dnet)
2310 list_add(&vport->detach_list, head);
2316 static void __net_exit ovs_exit_net(struct net *dnet)
2318 struct datapath *dp, *dp_next;
2319 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2320 struct vport *vport, *vport_next;
2326 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2331 list_vports_from_net(net, dnet, &head);
2334 /* Detach all vports from given namespace. */
2335 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2336 list_del(&vport->detach_list);
2337 ovs_dp_detach_port(vport);
2342 cancel_work_sync(&ovs_net->dp_notify_work);
2345 static struct pernet_operations ovs_net_ops = {
2346 .init = ovs_init_net,
2347 .exit = ovs_exit_net,
2349 .size = sizeof(struct ovs_net),
2352 static int __init dp_init(void)
2356 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2358 pr_info("Open vSwitch switching datapath\n");
2360 err = action_fifos_init();
2364 err = ovs_internal_dev_rtnl_link_register();
2366 goto error_action_fifos_exit;
2368 err = ovs_flow_init();
2370 goto error_unreg_rtnl_link;
2372 err = ovs_vport_init();
2374 goto error_flow_exit;
2376 err = register_pernet_device(&ovs_net_ops);
2378 goto error_vport_exit;
2380 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2382 goto error_netns_exit;
2384 err = ovs_netdev_init();
2386 goto error_unreg_notifier;
2388 err = dp_register_genl();
2390 goto error_unreg_netdev;
2396 error_unreg_notifier:
2397 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2399 unregister_pernet_device(&ovs_net_ops);
2404 error_unreg_rtnl_link:
2405 ovs_internal_dev_rtnl_link_unregister();
2406 error_action_fifos_exit:
2407 action_fifos_exit();
2412 static void dp_cleanup(void)
2414 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2416 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2417 unregister_pernet_device(&ovs_net_ops);
2421 ovs_internal_dev_rtnl_link_unregister();
2422 action_fifos_exit();
2425 module_init(dp_init);
2426 module_exit(dp_cleanup);
2428 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2429 MODULE_LICENSE("GPL");
2430 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2431 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2432 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2433 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);