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;
611 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
616 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
618 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
623 packet->dev = input_vport->dev;
624 OVS_CB(packet)->input_vport = input_vport;
625 sf_acts = rcu_dereference(flow->sf_acts);
628 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
632 ovs_flow_free(flow, false);
638 ovs_flow_free(flow, false);
645 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
646 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
647 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
648 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
649 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
650 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
653 static const struct genl_ops dp_packet_genl_ops[] = {
654 { .cmd = OVS_PACKET_CMD_EXECUTE,
655 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
656 .policy = packet_policy,
657 .doit = ovs_packet_cmd_execute
661 static struct genl_family dp_packet_genl_family __ro_after_init = {
662 .hdrsize = sizeof(struct ovs_header),
663 .name = OVS_PACKET_FAMILY,
664 .version = OVS_PACKET_VERSION,
665 .maxattr = OVS_PACKET_ATTR_MAX,
667 .parallel_ops = true,
668 .ops = dp_packet_genl_ops,
669 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
670 .module = THIS_MODULE,
673 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
674 struct ovs_dp_megaflow_stats *mega_stats)
678 memset(mega_stats, 0, sizeof(*mega_stats));
680 stats->n_flows = ovs_flow_tbl_count(&dp->table);
681 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
683 stats->n_hit = stats->n_missed = stats->n_lost = 0;
685 for_each_possible_cpu(i) {
686 const struct dp_stats_percpu *percpu_stats;
687 struct dp_stats_percpu local_stats;
690 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
693 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
694 local_stats = *percpu_stats;
695 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
697 stats->n_hit += local_stats.n_hit;
698 stats->n_missed += local_stats.n_missed;
699 stats->n_lost += local_stats.n_lost;
700 mega_stats->n_mask_hit += local_stats.n_mask_hit;
704 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
706 return ovs_identifier_is_ufid(sfid) &&
707 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
710 static bool should_fill_mask(uint32_t ufid_flags)
712 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
715 static bool should_fill_actions(uint32_t ufid_flags)
717 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
720 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
721 const struct sw_flow_id *sfid,
724 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
726 /* OVS_FLOW_ATTR_UFID */
727 if (sfid && ovs_identifier_is_ufid(sfid))
728 len += nla_total_size(sfid->ufid_len);
730 /* OVS_FLOW_ATTR_KEY */
731 if (!sfid || should_fill_key(sfid, ufid_flags))
732 len += nla_total_size(ovs_key_attr_size());
734 /* OVS_FLOW_ATTR_MASK */
735 if (should_fill_mask(ufid_flags))
736 len += nla_total_size(ovs_key_attr_size());
738 /* OVS_FLOW_ATTR_ACTIONS */
739 if (should_fill_actions(ufid_flags))
740 len += nla_total_size(acts->orig_len);
743 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
744 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
745 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
748 /* Called with ovs_mutex or RCU read lock. */
749 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
752 struct ovs_flow_stats stats;
756 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
759 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
763 if (stats.n_packets &&
764 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
765 sizeof(struct ovs_flow_stats), &stats,
769 if ((u8)ntohs(tcp_flags) &&
770 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
776 /* Called with ovs_mutex or RCU read lock. */
777 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
778 struct sk_buff *skb, int skb_orig_len)
780 struct nlattr *start;
783 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
784 * this is the first flow to be dumped into 'skb'. This is unusual for
785 * Netlink but individual action lists can be longer than
786 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
787 * The userspace caller can always fetch the actions separately if it
788 * really wants them. (Most userspace callers in fact don't care.)
790 * This can only fail for dump operations because the skb is always
791 * properly sized for single flows.
793 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
795 const struct sw_flow_actions *sf_acts;
797 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
798 err = ovs_nla_put_actions(sf_acts->actions,
799 sf_acts->actions_len, skb);
802 nla_nest_end(skb, start);
807 nla_nest_cancel(skb, start);
809 } else if (skb_orig_len) {
816 /* Called with ovs_mutex or RCU read lock. */
817 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
818 struct sk_buff *skb, u32 portid,
819 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
821 const int skb_orig_len = skb->len;
822 struct ovs_header *ovs_header;
825 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
830 ovs_header->dp_ifindex = dp_ifindex;
832 err = ovs_nla_put_identifier(flow, skb);
836 if (should_fill_key(&flow->id, ufid_flags)) {
837 err = ovs_nla_put_masked_key(flow, skb);
842 if (should_fill_mask(ufid_flags)) {
843 err = ovs_nla_put_mask(flow, skb);
848 err = ovs_flow_cmd_fill_stats(flow, skb);
852 if (should_fill_actions(ufid_flags)) {
853 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
858 genlmsg_end(skb, ovs_header);
862 genlmsg_cancel(skb, ovs_header);
866 /* May not be called with RCU read lock. */
867 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
868 const struct sw_flow_id *sfid,
869 struct genl_info *info,
876 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
879 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
880 skb = genlmsg_new(len, GFP_KERNEL);
882 return ERR_PTR(-ENOMEM);
887 /* Called with ovs_mutex. */
888 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
890 struct genl_info *info, u8 cmd,
891 bool always, u32 ufid_flags)
896 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
897 &flow->id, info, always, ufid_flags);
898 if (IS_ERR_OR_NULL(skb))
901 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
902 info->snd_portid, info->snd_seq, 0,
908 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
910 struct net *net = sock_net(skb->sk);
911 struct nlattr **a = info->attrs;
912 struct ovs_header *ovs_header = info->userhdr;
913 struct sw_flow *flow = NULL, *new_flow;
914 struct sw_flow_mask mask;
915 struct sk_buff *reply;
917 struct sw_flow_actions *acts;
918 struct sw_flow_match match;
919 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
921 bool log = !a[OVS_FLOW_ATTR_PROBE];
923 /* Must have key and actions. */
925 if (!a[OVS_FLOW_ATTR_KEY]) {
926 OVS_NLERR(log, "Flow key attr not present in new flow.");
929 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
930 OVS_NLERR(log, "Flow actions attr not present in new flow.");
934 /* Most of the time we need to allocate a new flow, do it before
937 new_flow = ovs_flow_alloc();
938 if (IS_ERR(new_flow)) {
939 error = PTR_ERR(new_flow);
944 ovs_match_init(&match, &new_flow->key, false, &mask);
945 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
946 a[OVS_FLOW_ATTR_MASK], log);
950 /* Extract flow identifier. */
951 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
952 &new_flow->key, log);
956 /* unmasked key is needed to match when ufid is not used. */
957 if (ovs_identifier_is_key(&new_flow->id))
958 match.key = new_flow->id.unmasked_key;
960 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
962 /* Validate actions. */
963 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
964 &new_flow->key, &acts, log);
966 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
970 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
973 error = PTR_ERR(reply);
978 dp = get_dp(net, ovs_header->dp_ifindex);
984 /* Check if this is a duplicate flow */
985 if (ovs_identifier_is_ufid(&new_flow->id))
986 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
988 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
990 rcu_assign_pointer(new_flow->sf_acts, acts);
992 /* Put flow in bucket. */
993 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
994 if (unlikely(error)) {
999 if (unlikely(reply)) {
1000 error = ovs_flow_cmd_fill_info(new_flow,
1001 ovs_header->dp_ifindex,
1002 reply, info->snd_portid,
1010 struct sw_flow_actions *old_acts;
1012 /* Bail out if we're not allowed to modify an existing flow.
1013 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1014 * because Generic Netlink treats the latter as a dump
1015 * request. We also accept NLM_F_EXCL in case that bug ever
1018 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1021 goto err_unlock_ovs;
1023 /* The flow identifier has to be the same for flow updates.
1024 * Look for any overlapping flow.
1026 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1027 if (ovs_identifier_is_key(&flow->id))
1028 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1030 else /* UFID matches but key is different */
1034 goto err_unlock_ovs;
1037 /* Update actions. */
1038 old_acts = ovsl_dereference(flow->sf_acts);
1039 rcu_assign_pointer(flow->sf_acts, acts);
1041 if (unlikely(reply)) {
1042 error = ovs_flow_cmd_fill_info(flow,
1043 ovs_header->dp_ifindex,
1044 reply, info->snd_portid,
1052 ovs_nla_free_flow_actions_rcu(old_acts);
1053 ovs_flow_free(new_flow, false);
1057 ovs_notify(&dp_flow_genl_family, reply, info);
1064 ovs_nla_free_flow_actions(acts);
1066 ovs_flow_free(new_flow, false);
1071 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1072 static struct sw_flow_actions *get_flow_actions(struct net *net,
1073 const struct nlattr *a,
1074 const struct sw_flow_key *key,
1075 const struct sw_flow_mask *mask,
1078 struct sw_flow_actions *acts;
1079 struct sw_flow_key masked_key;
1082 ovs_flow_mask_key(&masked_key, key, true, mask);
1083 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1086 "Actions may not be safe on all matching packets");
1087 return ERR_PTR(error);
1093 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1095 struct net *net = sock_net(skb->sk);
1096 struct nlattr **a = info->attrs;
1097 struct ovs_header *ovs_header = info->userhdr;
1098 struct sw_flow_key key;
1099 struct sw_flow *flow;
1100 struct sw_flow_mask mask;
1101 struct sk_buff *reply = NULL;
1102 struct datapath *dp;
1103 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1104 struct sw_flow_match match;
1105 struct sw_flow_id sfid;
1106 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1108 bool log = !a[OVS_FLOW_ATTR_PROBE];
1111 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1112 if (a[OVS_FLOW_ATTR_KEY]) {
1113 ovs_match_init(&match, &key, true, &mask);
1114 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1115 a[OVS_FLOW_ATTR_MASK], log);
1116 } else if (!ufid_present) {
1118 "Flow set message rejected, Key attribute missing.");
1124 /* Validate actions. */
1125 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1126 if (!a[OVS_FLOW_ATTR_KEY]) {
1128 "Flow key attribute not present in set flow.");
1133 acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1136 error = PTR_ERR(acts);
1140 /* Can allocate before locking if have acts. */
1141 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1143 if (IS_ERR(reply)) {
1144 error = PTR_ERR(reply);
1145 goto err_kfree_acts;
1150 dp = get_dp(net, ovs_header->dp_ifindex);
1151 if (unlikely(!dp)) {
1153 goto err_unlock_ovs;
1155 /* Check that the flow exists. */
1157 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1159 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1160 if (unlikely(!flow)) {
1162 goto err_unlock_ovs;
1165 /* Update actions, if present. */
1167 old_acts = ovsl_dereference(flow->sf_acts);
1168 rcu_assign_pointer(flow->sf_acts, acts);
1170 if (unlikely(reply)) {
1171 error = ovs_flow_cmd_fill_info(flow,
1172 ovs_header->dp_ifindex,
1173 reply, info->snd_portid,
1180 /* Could not alloc without acts before locking. */
1181 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1182 info, OVS_FLOW_CMD_NEW, false,
1185 if (IS_ERR(reply)) {
1186 error = PTR_ERR(reply);
1187 goto err_unlock_ovs;
1192 if (a[OVS_FLOW_ATTR_CLEAR])
1193 ovs_flow_stats_clear(flow);
1197 ovs_notify(&dp_flow_genl_family, reply, info);
1199 ovs_nla_free_flow_actions_rcu(old_acts);
1207 ovs_nla_free_flow_actions(acts);
1212 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1214 struct nlattr **a = info->attrs;
1215 struct ovs_header *ovs_header = info->userhdr;
1216 struct net *net = sock_net(skb->sk);
1217 struct sw_flow_key key;
1218 struct sk_buff *reply;
1219 struct sw_flow *flow;
1220 struct datapath *dp;
1221 struct sw_flow_match match;
1222 struct sw_flow_id ufid;
1223 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1225 bool log = !a[OVS_FLOW_ATTR_PROBE];
1228 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1229 if (a[OVS_FLOW_ATTR_KEY]) {
1230 ovs_match_init(&match, &key, true, NULL);
1231 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1233 } else if (!ufid_present) {
1235 "Flow get message rejected, Key attribute missing.");
1242 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1249 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1251 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1257 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1258 OVS_FLOW_CMD_NEW, true, ufid_flags);
1259 if (IS_ERR(reply)) {
1260 err = PTR_ERR(reply);
1265 return genlmsg_reply(reply, info);
1271 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1273 struct nlattr **a = info->attrs;
1274 struct ovs_header *ovs_header = info->userhdr;
1275 struct net *net = sock_net(skb->sk);
1276 struct sw_flow_key key;
1277 struct sk_buff *reply;
1278 struct sw_flow *flow = NULL;
1279 struct datapath *dp;
1280 struct sw_flow_match match;
1281 struct sw_flow_id ufid;
1282 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1284 bool log = !a[OVS_FLOW_ATTR_PROBE];
1287 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1288 if (a[OVS_FLOW_ATTR_KEY]) {
1289 ovs_match_init(&match, &key, true, NULL);
1290 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1297 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1298 if (unlikely(!dp)) {
1303 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1304 err = ovs_flow_tbl_flush(&dp->table);
1309 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1311 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1312 if (unlikely(!flow)) {
1317 ovs_flow_tbl_remove(&dp->table, flow);
1320 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1321 &flow->id, info, false, ufid_flags);
1322 if (likely(reply)) {
1323 if (likely(!IS_ERR(reply))) {
1324 rcu_read_lock(); /*To keep RCU checker happy. */
1325 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1326 reply, info->snd_portid,
1333 ovs_notify(&dp_flow_genl_family, reply, info);
1335 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1339 ovs_flow_free(flow, true);
1346 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1348 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1349 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1350 struct table_instance *ti;
1351 struct datapath *dp;
1355 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1356 OVS_FLOW_ATTR_MAX, flow_policy);
1359 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1362 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1368 ti = rcu_dereference(dp->table.ti);
1370 struct sw_flow *flow;
1373 bucket = cb->args[0];
1375 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1379 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1380 NETLINK_CB(cb->skb).portid,
1381 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1382 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1385 cb->args[0] = bucket;
1392 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1393 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1394 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1395 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1396 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1397 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1398 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1399 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1402 static const struct genl_ops dp_flow_genl_ops[] = {
1403 { .cmd = OVS_FLOW_CMD_NEW,
1404 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1405 .policy = flow_policy,
1406 .doit = ovs_flow_cmd_new
1408 { .cmd = OVS_FLOW_CMD_DEL,
1409 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1410 .policy = flow_policy,
1411 .doit = ovs_flow_cmd_del
1413 { .cmd = OVS_FLOW_CMD_GET,
1414 .flags = 0, /* OK for unprivileged users. */
1415 .policy = flow_policy,
1416 .doit = ovs_flow_cmd_get,
1417 .dumpit = ovs_flow_cmd_dump
1419 { .cmd = OVS_FLOW_CMD_SET,
1420 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1421 .policy = flow_policy,
1422 .doit = ovs_flow_cmd_set,
1426 static struct genl_family dp_flow_genl_family __ro_after_init = {
1427 .hdrsize = sizeof(struct ovs_header),
1428 .name = OVS_FLOW_FAMILY,
1429 .version = OVS_FLOW_VERSION,
1430 .maxattr = OVS_FLOW_ATTR_MAX,
1432 .parallel_ops = true,
1433 .ops = dp_flow_genl_ops,
1434 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1435 .mcgrps = &ovs_dp_flow_multicast_group,
1437 .module = THIS_MODULE,
1440 static size_t ovs_dp_cmd_msg_size(void)
1442 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1444 msgsize += nla_total_size(IFNAMSIZ);
1445 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1446 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1447 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1452 /* Called with ovs_mutex. */
1453 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1454 u32 portid, u32 seq, u32 flags, u8 cmd)
1456 struct ovs_header *ovs_header;
1457 struct ovs_dp_stats dp_stats;
1458 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1461 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1466 ovs_header->dp_ifindex = get_dpifindex(dp);
1468 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1470 goto nla_put_failure;
1472 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1473 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1474 &dp_stats, OVS_DP_ATTR_PAD))
1475 goto nla_put_failure;
1477 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1478 sizeof(struct ovs_dp_megaflow_stats),
1479 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1480 goto nla_put_failure;
1482 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1483 goto nla_put_failure;
1485 genlmsg_end(skb, ovs_header);
1489 genlmsg_cancel(skb, ovs_header);
1494 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1496 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1499 /* Called with rcu_read_lock or ovs_mutex. */
1500 static struct datapath *lookup_datapath(struct net *net,
1501 const struct ovs_header *ovs_header,
1502 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1504 struct datapath *dp;
1506 if (!a[OVS_DP_ATTR_NAME])
1507 dp = get_dp(net, ovs_header->dp_ifindex);
1509 struct vport *vport;
1511 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1512 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1514 return dp ? dp : ERR_PTR(-ENODEV);
1517 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1519 struct datapath *dp;
1521 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1525 WARN(dp->user_features, "Dropping previously announced user features\n");
1526 dp->user_features = 0;
1529 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1531 if (a[OVS_DP_ATTR_USER_FEATURES])
1532 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1535 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1537 struct nlattr **a = info->attrs;
1538 struct vport_parms parms;
1539 struct sk_buff *reply;
1540 struct datapath *dp;
1541 struct vport *vport;
1542 struct ovs_net *ovs_net;
1546 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1549 reply = ovs_dp_cmd_alloc_info();
1554 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1556 goto err_free_reply;
1558 ovs_dp_set_net(dp, sock_net(skb->sk));
1560 /* Allocate table. */
1561 err = ovs_flow_tbl_init(&dp->table);
1565 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1566 if (!dp->stats_percpu) {
1568 goto err_destroy_table;
1571 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1575 goto err_destroy_percpu;
1578 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1579 INIT_HLIST_HEAD(&dp->ports[i]);
1581 /* Set up our datapath device. */
1582 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1583 parms.type = OVS_VPORT_TYPE_INTERNAL;
1584 parms.options = NULL;
1586 parms.port_no = OVSP_LOCAL;
1587 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1589 ovs_dp_change(dp, a);
1591 /* So far only local changes have been made, now need the lock. */
1594 vport = new_vport(&parms);
1595 if (IS_ERR(vport)) {
1596 err = PTR_ERR(vport);
1600 if (err == -EEXIST) {
1601 /* An outdated user space instance that does not understand
1602 * the concept of user_features has attempted to create a new
1603 * datapath and is likely to reuse it. Drop all user features.
1605 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1606 ovs_dp_reset_user_features(skb, info);
1609 goto err_destroy_ports_array;
1612 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1613 info->snd_seq, 0, OVS_DP_CMD_NEW);
1616 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1617 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1621 ovs_notify(&dp_datapath_genl_family, reply, info);
1624 err_destroy_ports_array:
1628 free_percpu(dp->stats_percpu);
1630 ovs_flow_tbl_destroy(&dp->table);
1639 /* Called with ovs_mutex. */
1640 static void __dp_destroy(struct datapath *dp)
1644 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1645 struct vport *vport;
1646 struct hlist_node *n;
1648 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1649 if (vport->port_no != OVSP_LOCAL)
1650 ovs_dp_detach_port(vport);
1653 list_del_rcu(&dp->list_node);
1655 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1656 * all ports in datapath are destroyed first before freeing datapath.
1658 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1660 /* RCU destroy the flow table */
1661 call_rcu(&dp->rcu, destroy_dp_rcu);
1664 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1666 struct sk_buff *reply;
1667 struct datapath *dp;
1670 reply = ovs_dp_cmd_alloc_info();
1675 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1678 goto err_unlock_free;
1680 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1681 info->snd_seq, 0, OVS_DP_CMD_DEL);
1687 ovs_notify(&dp_datapath_genl_family, reply, info);
1697 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1699 struct sk_buff *reply;
1700 struct datapath *dp;
1703 reply = ovs_dp_cmd_alloc_info();
1708 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1711 goto err_unlock_free;
1713 ovs_dp_change(dp, info->attrs);
1715 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1716 info->snd_seq, 0, OVS_DP_CMD_NEW);
1720 ovs_notify(&dp_datapath_genl_family, reply, info);
1730 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1732 struct sk_buff *reply;
1733 struct datapath *dp;
1736 reply = ovs_dp_cmd_alloc_info();
1741 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1744 goto err_unlock_free;
1746 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1747 info->snd_seq, 0, OVS_DP_CMD_NEW);
1751 return genlmsg_reply(reply, info);
1759 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1761 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1762 struct datapath *dp;
1763 int skip = cb->args[0];
1767 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1769 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1770 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1771 OVS_DP_CMD_NEW) < 0)
1782 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1783 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1784 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1785 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1788 static const struct genl_ops dp_datapath_genl_ops[] = {
1789 { .cmd = OVS_DP_CMD_NEW,
1790 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1791 .policy = datapath_policy,
1792 .doit = ovs_dp_cmd_new
1794 { .cmd = OVS_DP_CMD_DEL,
1795 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1796 .policy = datapath_policy,
1797 .doit = ovs_dp_cmd_del
1799 { .cmd = OVS_DP_CMD_GET,
1800 .flags = 0, /* OK for unprivileged users. */
1801 .policy = datapath_policy,
1802 .doit = ovs_dp_cmd_get,
1803 .dumpit = ovs_dp_cmd_dump
1805 { .cmd = OVS_DP_CMD_SET,
1806 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1807 .policy = datapath_policy,
1808 .doit = ovs_dp_cmd_set,
1812 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1813 .hdrsize = sizeof(struct ovs_header),
1814 .name = OVS_DATAPATH_FAMILY,
1815 .version = OVS_DATAPATH_VERSION,
1816 .maxattr = OVS_DP_ATTR_MAX,
1818 .parallel_ops = true,
1819 .ops = dp_datapath_genl_ops,
1820 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1821 .mcgrps = &ovs_dp_datapath_multicast_group,
1823 .module = THIS_MODULE,
1826 /* Called with ovs_mutex or RCU read lock. */
1827 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1828 u32 portid, u32 seq, u32 flags, u8 cmd)
1830 struct ovs_header *ovs_header;
1831 struct ovs_vport_stats vport_stats;
1834 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1839 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1841 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1842 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1843 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1844 ovs_vport_name(vport)))
1845 goto nla_put_failure;
1847 ovs_vport_get_stats(vport, &vport_stats);
1848 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1849 sizeof(struct ovs_vport_stats), &vport_stats,
1850 OVS_VPORT_ATTR_PAD))
1851 goto nla_put_failure;
1853 if (ovs_vport_get_upcall_portids(vport, skb))
1854 goto nla_put_failure;
1856 err = ovs_vport_get_options(vport, skb);
1857 if (err == -EMSGSIZE)
1860 genlmsg_end(skb, ovs_header);
1866 genlmsg_cancel(skb, ovs_header);
1870 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1872 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1875 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1876 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1879 struct sk_buff *skb;
1882 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1884 return ERR_PTR(-ENOMEM);
1886 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1892 /* Called with ovs_mutex or RCU read lock. */
1893 static struct vport *lookup_vport(struct net *net,
1894 const struct ovs_header *ovs_header,
1895 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1897 struct datapath *dp;
1898 struct vport *vport;
1900 if (a[OVS_VPORT_ATTR_NAME]) {
1901 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1903 return ERR_PTR(-ENODEV);
1904 if (ovs_header->dp_ifindex &&
1905 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1906 return ERR_PTR(-ENODEV);
1908 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1909 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1911 if (port_no >= DP_MAX_PORTS)
1912 return ERR_PTR(-EFBIG);
1914 dp = get_dp(net, ovs_header->dp_ifindex);
1916 return ERR_PTR(-ENODEV);
1918 vport = ovs_vport_ovsl_rcu(dp, port_no);
1920 return ERR_PTR(-ENODEV);
1923 return ERR_PTR(-EINVAL);
1926 /* Called with ovs_mutex */
1927 static void update_headroom(struct datapath *dp)
1929 unsigned dev_headroom, max_headroom = 0;
1930 struct net_device *dev;
1931 struct vport *vport;
1934 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1935 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1937 dev_headroom = netdev_get_fwd_headroom(dev);
1938 if (dev_headroom > max_headroom)
1939 max_headroom = dev_headroom;
1943 dp->max_headroom = max_headroom;
1944 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1945 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1946 netdev_set_rx_headroom(vport->dev, max_headroom);
1949 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1951 struct nlattr **a = info->attrs;
1952 struct ovs_header *ovs_header = info->userhdr;
1953 struct vport_parms parms;
1954 struct sk_buff *reply;
1955 struct vport *vport;
1956 struct datapath *dp;
1960 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1961 !a[OVS_VPORT_ATTR_UPCALL_PID])
1964 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1965 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1966 if (port_no >= DP_MAX_PORTS)
1969 reply = ovs_vport_cmd_alloc_info();
1975 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1978 goto exit_unlock_free;
1981 vport = ovs_vport_ovsl(dp, port_no);
1984 goto exit_unlock_free;
1986 for (port_no = 1; ; port_no++) {
1987 if (port_no >= DP_MAX_PORTS) {
1989 goto exit_unlock_free;
1991 vport = ovs_vport_ovsl(dp, port_no);
1997 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1998 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1999 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2001 parms.port_no = port_no;
2002 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2004 vport = new_vport(&parms);
2005 err = PTR_ERR(vport);
2006 if (IS_ERR(vport)) {
2009 goto exit_unlock_free;
2012 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2013 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2015 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2016 update_headroom(dp);
2018 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2023 ovs_notify(&dp_vport_genl_family, reply, info);
2032 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2034 struct nlattr **a = info->attrs;
2035 struct sk_buff *reply;
2036 struct vport *vport;
2039 reply = ovs_vport_cmd_alloc_info();
2044 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2045 err = PTR_ERR(vport);
2047 goto exit_unlock_free;
2049 if (a[OVS_VPORT_ATTR_TYPE] &&
2050 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2052 goto exit_unlock_free;
2055 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2056 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2058 goto exit_unlock_free;
2062 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2063 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2065 err = ovs_vport_set_upcall_portids(vport, ids);
2067 goto exit_unlock_free;
2070 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2071 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2075 ovs_notify(&dp_vport_genl_family, reply, info);
2084 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2086 bool must_update_headroom = false;
2087 struct nlattr **a = info->attrs;
2088 struct sk_buff *reply;
2089 struct datapath *dp;
2090 struct vport *vport;
2093 reply = ovs_vport_cmd_alloc_info();
2098 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2099 err = PTR_ERR(vport);
2101 goto exit_unlock_free;
2103 if (vport->port_no == OVSP_LOCAL) {
2105 goto exit_unlock_free;
2108 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2109 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2112 /* the vport deletion may trigger dp headroom update */
2114 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2115 must_update_headroom = true;
2116 netdev_reset_rx_headroom(vport->dev);
2117 ovs_dp_detach_port(vport);
2119 if (must_update_headroom)
2120 update_headroom(dp);
2123 ovs_notify(&dp_vport_genl_family, reply, info);
2132 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2134 struct nlattr **a = info->attrs;
2135 struct ovs_header *ovs_header = info->userhdr;
2136 struct sk_buff *reply;
2137 struct vport *vport;
2140 reply = ovs_vport_cmd_alloc_info();
2145 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2146 err = PTR_ERR(vport);
2148 goto exit_unlock_free;
2149 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2150 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2154 return genlmsg_reply(reply, info);
2162 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2164 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2165 struct datapath *dp;
2166 int bucket = cb->args[0], skip = cb->args[1];
2170 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2175 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2176 struct vport *vport;
2179 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2181 ovs_vport_cmd_fill_info(vport, skb,
2182 NETLINK_CB(cb->skb).portid,
2185 OVS_VPORT_CMD_NEW) < 0)
2201 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2202 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2203 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2204 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2205 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2206 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2207 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2210 static const struct genl_ops dp_vport_genl_ops[] = {
2211 { .cmd = OVS_VPORT_CMD_NEW,
2212 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2213 .policy = vport_policy,
2214 .doit = ovs_vport_cmd_new
2216 { .cmd = OVS_VPORT_CMD_DEL,
2217 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2218 .policy = vport_policy,
2219 .doit = ovs_vport_cmd_del
2221 { .cmd = OVS_VPORT_CMD_GET,
2222 .flags = 0, /* OK for unprivileged users. */
2223 .policy = vport_policy,
2224 .doit = ovs_vport_cmd_get,
2225 .dumpit = ovs_vport_cmd_dump
2227 { .cmd = OVS_VPORT_CMD_SET,
2228 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2229 .policy = vport_policy,
2230 .doit = ovs_vport_cmd_set,
2234 struct genl_family dp_vport_genl_family __ro_after_init = {
2235 .hdrsize = sizeof(struct ovs_header),
2236 .name = OVS_VPORT_FAMILY,
2237 .version = OVS_VPORT_VERSION,
2238 .maxattr = OVS_VPORT_ATTR_MAX,
2240 .parallel_ops = true,
2241 .ops = dp_vport_genl_ops,
2242 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2243 .mcgrps = &ovs_dp_vport_multicast_group,
2245 .module = THIS_MODULE,
2248 static struct genl_family * const dp_genl_families[] = {
2249 &dp_datapath_genl_family,
2250 &dp_vport_genl_family,
2251 &dp_flow_genl_family,
2252 &dp_packet_genl_family,
2255 static void dp_unregister_genl(int n_families)
2259 for (i = 0; i < n_families; i++)
2260 genl_unregister_family(dp_genl_families[i]);
2263 static int __init dp_register_genl(void)
2268 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2270 err = genl_register_family(dp_genl_families[i]);
2278 dp_unregister_genl(i);
2282 static int __net_init ovs_init_net(struct net *net)
2284 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2286 INIT_LIST_HEAD(&ovs_net->dps);
2287 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2292 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2293 struct list_head *head)
2295 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2296 struct datapath *dp;
2298 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2301 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2302 struct vport *vport;
2304 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2305 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2308 if (dev_net(vport->dev) == dnet)
2309 list_add(&vport->detach_list, head);
2315 static void __net_exit ovs_exit_net(struct net *dnet)
2317 struct datapath *dp, *dp_next;
2318 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2319 struct vport *vport, *vport_next;
2325 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2330 list_vports_from_net(net, dnet, &head);
2333 /* Detach all vports from given namespace. */
2334 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2335 list_del(&vport->detach_list);
2336 ovs_dp_detach_port(vport);
2341 cancel_work_sync(&ovs_net->dp_notify_work);
2344 static struct pernet_operations ovs_net_ops = {
2345 .init = ovs_init_net,
2346 .exit = ovs_exit_net,
2348 .size = sizeof(struct ovs_net),
2351 static int __init dp_init(void)
2355 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2357 pr_info("Open vSwitch switching datapath\n");
2359 err = action_fifos_init();
2363 err = ovs_internal_dev_rtnl_link_register();
2365 goto error_action_fifos_exit;
2367 err = ovs_flow_init();
2369 goto error_unreg_rtnl_link;
2371 err = ovs_vport_init();
2373 goto error_flow_exit;
2375 err = register_pernet_device(&ovs_net_ops);
2377 goto error_vport_exit;
2379 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2381 goto error_netns_exit;
2383 err = ovs_netdev_init();
2385 goto error_unreg_notifier;
2387 err = dp_register_genl();
2389 goto error_unreg_netdev;
2395 error_unreg_notifier:
2396 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2398 unregister_pernet_device(&ovs_net_ops);
2403 error_unreg_rtnl_link:
2404 ovs_internal_dev_rtnl_link_unregister();
2405 error_action_fifos_exit:
2406 action_fifos_exit();
2411 static void dp_cleanup(void)
2413 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2415 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2416 unregister_pernet_device(&ovs_net_ops);
2420 ovs_internal_dev_rtnl_link_unregister();
2421 action_fifos_exit();
2424 module_init(dp_init);
2425 module_exit(dp_cleanup);
2427 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2428 MODULE_LICENSE("GPL");
2429 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2430 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2431 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2432 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);