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 struct sk_buff *segs, *nskb;
341 BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
342 segs = __skb_gso_segment(skb, NETIF_F_SG, false);
344 return PTR_ERR(segs);
348 /* Queue all of the segments. */
351 err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
355 } while ((skb = skb->next));
357 /* Free all of the segments. */
365 } while ((skb = nskb));
369 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
370 unsigned int hdrlen, int actions_attrlen)
372 size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
373 + nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
374 + nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
375 + nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
377 /* OVS_PACKET_ATTR_USERDATA */
378 if (upcall_info->userdata)
379 size += NLA_ALIGN(upcall_info->userdata->nla_len);
381 /* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
382 if (upcall_info->egress_tun_info)
383 size += nla_total_size(ovs_tun_key_attr_size());
385 /* OVS_PACKET_ATTR_ACTIONS */
386 if (upcall_info->actions_len)
387 size += nla_total_size(actions_attrlen);
389 /* OVS_PACKET_ATTR_MRU */
390 if (upcall_info->mru)
391 size += nla_total_size(sizeof(upcall_info->mru));
396 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
398 if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
399 size_t plen = NLA_ALIGN(skb->len) - skb->len;
402 skb_put_zero(skb, plen);
406 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
407 const struct sw_flow_key *key,
408 const struct dp_upcall_info *upcall_info,
411 struct ovs_header *upcall;
412 struct sk_buff *nskb = NULL;
413 struct sk_buff *user_skb = NULL; /* to be queued to userspace */
419 dp_ifindex = get_dpifindex(dp);
423 if (skb_vlan_tag_present(skb)) {
424 nskb = skb_clone(skb, GFP_ATOMIC);
428 nskb = __vlan_hwaccel_push_inside(nskb);
435 if (nla_attr_size(skb->len) > USHRT_MAX) {
440 /* Complete checksum if needed */
441 if (skb->ip_summed == CHECKSUM_PARTIAL &&
442 (err = skb_csum_hwoffload_help(skb, 0)))
445 /* Older versions of OVS user space enforce alignment of the last
446 * Netlink attribute to NLA_ALIGNTO which would require extensive
447 * padding logic. Only perform zerocopy if padding is not required.
449 if (dp->user_features & OVS_DP_F_UNALIGNED)
450 hlen = skb_zerocopy_headlen(skb);
454 len = upcall_msg_size(upcall_info, hlen - cutlen,
455 OVS_CB(skb)->acts_origlen);
456 user_skb = genlmsg_new(len, GFP_ATOMIC);
462 upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
463 0, upcall_info->cmd);
464 upcall->dp_ifindex = dp_ifindex;
466 err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
469 if (upcall_info->userdata)
470 __nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
471 nla_len(upcall_info->userdata),
472 nla_data(upcall_info->userdata));
474 if (upcall_info->egress_tun_info) {
475 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
476 err = ovs_nla_put_tunnel_info(user_skb,
477 upcall_info->egress_tun_info);
479 nla_nest_end(user_skb, nla);
482 if (upcall_info->actions_len) {
483 nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
484 err = ovs_nla_put_actions(upcall_info->actions,
485 upcall_info->actions_len,
488 nla_nest_end(user_skb, nla);
490 nla_nest_cancel(user_skb, nla);
493 /* Add OVS_PACKET_ATTR_MRU */
494 if (upcall_info->mru) {
495 if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
500 pad_packet(dp, user_skb);
503 /* Add OVS_PACKET_ATTR_LEN when packet is truncated */
505 if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
510 pad_packet(dp, user_skb);
513 /* Only reserve room for attribute header, packet data is added
514 * in skb_zerocopy() */
515 if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
519 nla->nla_len = nla_attr_size(skb->len - cutlen);
521 err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
525 /* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
526 pad_packet(dp, user_skb);
528 ((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
530 err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
540 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
542 struct ovs_header *ovs_header = info->userhdr;
543 struct net *net = sock_net(skb->sk);
544 struct nlattr **a = info->attrs;
545 struct sw_flow_actions *acts;
546 struct sk_buff *packet;
547 struct sw_flow *flow;
548 struct sw_flow_actions *sf_acts;
550 struct vport *input_vport;
554 bool log = !a[OVS_PACKET_ATTR_PROBE];
557 if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
558 !a[OVS_PACKET_ATTR_ACTIONS])
561 len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
562 packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
566 skb_reserve(packet, NET_IP_ALIGN);
568 nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
570 /* Set packet's mru */
571 if (a[OVS_PACKET_ATTR_MRU]) {
572 mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
573 packet->ignore_df = 1;
575 OVS_CB(packet)->mru = mru;
577 /* Build an sw_flow for sending this packet. */
578 flow = ovs_flow_alloc();
583 err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
584 packet, &flow->key, log);
588 err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
589 &flow->key, &acts, log);
593 rcu_assign_pointer(flow->sf_acts, acts);
594 packet->priority = flow->key.phy.priority;
595 packet->mark = flow->key.phy.skb_mark;
598 dp = get_dp_rcu(net, ovs_header->dp_ifindex);
603 input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
605 input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
610 packet->dev = input_vport->dev;
611 OVS_CB(packet)->input_vport = input_vport;
612 sf_acts = rcu_dereference(flow->sf_acts);
615 err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
619 ovs_flow_free(flow, false);
625 ovs_flow_free(flow, false);
632 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
633 [OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
634 [OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
635 [OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
636 [OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
637 [OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
640 static const struct genl_ops dp_packet_genl_ops[] = {
641 { .cmd = OVS_PACKET_CMD_EXECUTE,
642 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
643 .policy = packet_policy,
644 .doit = ovs_packet_cmd_execute
648 static struct genl_family dp_packet_genl_family __ro_after_init = {
649 .hdrsize = sizeof(struct ovs_header),
650 .name = OVS_PACKET_FAMILY,
651 .version = OVS_PACKET_VERSION,
652 .maxattr = OVS_PACKET_ATTR_MAX,
654 .parallel_ops = true,
655 .ops = dp_packet_genl_ops,
656 .n_ops = ARRAY_SIZE(dp_packet_genl_ops),
657 .module = THIS_MODULE,
660 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
661 struct ovs_dp_megaflow_stats *mega_stats)
665 memset(mega_stats, 0, sizeof(*mega_stats));
667 stats->n_flows = ovs_flow_tbl_count(&dp->table);
668 mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
670 stats->n_hit = stats->n_missed = stats->n_lost = 0;
672 for_each_possible_cpu(i) {
673 const struct dp_stats_percpu *percpu_stats;
674 struct dp_stats_percpu local_stats;
677 percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
680 start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
681 local_stats = *percpu_stats;
682 } while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
684 stats->n_hit += local_stats.n_hit;
685 stats->n_missed += local_stats.n_missed;
686 stats->n_lost += local_stats.n_lost;
687 mega_stats->n_mask_hit += local_stats.n_mask_hit;
691 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
693 return ovs_identifier_is_ufid(sfid) &&
694 !(ufid_flags & OVS_UFID_F_OMIT_KEY);
697 static bool should_fill_mask(uint32_t ufid_flags)
699 return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
702 static bool should_fill_actions(uint32_t ufid_flags)
704 return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
707 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
708 const struct sw_flow_id *sfid,
711 size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
713 /* OVS_FLOW_ATTR_UFID */
714 if (sfid && ovs_identifier_is_ufid(sfid))
715 len += nla_total_size(sfid->ufid_len);
717 /* OVS_FLOW_ATTR_KEY */
718 if (!sfid || should_fill_key(sfid, ufid_flags))
719 len += nla_total_size(ovs_key_attr_size());
721 /* OVS_FLOW_ATTR_MASK */
722 if (should_fill_mask(ufid_flags))
723 len += nla_total_size(ovs_key_attr_size());
725 /* OVS_FLOW_ATTR_ACTIONS */
726 if (should_fill_actions(ufid_flags))
727 len += nla_total_size(acts->orig_len);
730 + nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
731 + nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
732 + nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
735 /* Called with ovs_mutex or RCU read lock. */
736 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
739 struct ovs_flow_stats stats;
743 ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
746 nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
750 if (stats.n_packets &&
751 nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
752 sizeof(struct ovs_flow_stats), &stats,
756 if ((u8)ntohs(tcp_flags) &&
757 nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
763 /* Called with ovs_mutex or RCU read lock. */
764 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
765 struct sk_buff *skb, int skb_orig_len)
767 struct nlattr *start;
770 /* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
771 * this is the first flow to be dumped into 'skb'. This is unusual for
772 * Netlink but individual action lists can be longer than
773 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
774 * The userspace caller can always fetch the actions separately if it
775 * really wants them. (Most userspace callers in fact don't care.)
777 * This can only fail for dump operations because the skb is always
778 * properly sized for single flows.
780 start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
782 const struct sw_flow_actions *sf_acts;
784 sf_acts = rcu_dereference_ovsl(flow->sf_acts);
785 err = ovs_nla_put_actions(sf_acts->actions,
786 sf_acts->actions_len, skb);
789 nla_nest_end(skb, start);
794 nla_nest_cancel(skb, start);
796 } else if (skb_orig_len) {
803 /* Called with ovs_mutex or RCU read lock. */
804 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
805 struct sk_buff *skb, u32 portid,
806 u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
808 const int skb_orig_len = skb->len;
809 struct ovs_header *ovs_header;
812 ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
817 ovs_header->dp_ifindex = dp_ifindex;
819 err = ovs_nla_put_identifier(flow, skb);
823 if (should_fill_key(&flow->id, ufid_flags)) {
824 err = ovs_nla_put_masked_key(flow, skb);
829 if (should_fill_mask(ufid_flags)) {
830 err = ovs_nla_put_mask(flow, skb);
835 err = ovs_flow_cmd_fill_stats(flow, skb);
839 if (should_fill_actions(ufid_flags)) {
840 err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
845 genlmsg_end(skb, ovs_header);
849 genlmsg_cancel(skb, ovs_header);
853 /* May not be called with RCU read lock. */
854 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
855 const struct sw_flow_id *sfid,
856 struct genl_info *info,
863 if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
866 len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
867 skb = genlmsg_new(len, GFP_KERNEL);
869 return ERR_PTR(-ENOMEM);
874 /* Called with ovs_mutex. */
875 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
877 struct genl_info *info, u8 cmd,
878 bool always, u32 ufid_flags)
883 skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
884 &flow->id, info, always, ufid_flags);
885 if (IS_ERR_OR_NULL(skb))
888 retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
889 info->snd_portid, info->snd_seq, 0,
895 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
897 struct net *net = sock_net(skb->sk);
898 struct nlattr **a = info->attrs;
899 struct ovs_header *ovs_header = info->userhdr;
900 struct sw_flow *flow = NULL, *new_flow;
901 struct sw_flow_mask mask;
902 struct sk_buff *reply;
904 struct sw_flow_actions *acts;
905 struct sw_flow_match match;
906 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
908 bool log = !a[OVS_FLOW_ATTR_PROBE];
910 /* Must have key and actions. */
912 if (!a[OVS_FLOW_ATTR_KEY]) {
913 OVS_NLERR(log, "Flow key attr not present in new flow.");
916 if (!a[OVS_FLOW_ATTR_ACTIONS]) {
917 OVS_NLERR(log, "Flow actions attr not present in new flow.");
921 /* Most of the time we need to allocate a new flow, do it before
924 new_flow = ovs_flow_alloc();
925 if (IS_ERR(new_flow)) {
926 error = PTR_ERR(new_flow);
931 ovs_match_init(&match, &new_flow->key, false, &mask);
932 error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
933 a[OVS_FLOW_ATTR_MASK], log);
937 /* Extract flow identifier. */
938 error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
939 &new_flow->key, log);
943 /* unmasked key is needed to match when ufid is not used. */
944 if (ovs_identifier_is_key(&new_flow->id))
945 match.key = new_flow->id.unmasked_key;
947 ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
949 /* Validate actions. */
950 error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
951 &new_flow->key, &acts, log);
953 OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
957 reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
960 error = PTR_ERR(reply);
965 dp = get_dp(net, ovs_header->dp_ifindex);
971 /* Check if this is a duplicate flow */
972 if (ovs_identifier_is_ufid(&new_flow->id))
973 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
975 flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
977 rcu_assign_pointer(new_flow->sf_acts, acts);
979 /* Put flow in bucket. */
980 error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
981 if (unlikely(error)) {
986 if (unlikely(reply)) {
987 error = ovs_flow_cmd_fill_info(new_flow,
988 ovs_header->dp_ifindex,
989 reply, info->snd_portid,
997 struct sw_flow_actions *old_acts;
999 /* Bail out if we're not allowed to modify an existing flow.
1000 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1001 * because Generic Netlink treats the latter as a dump
1002 * request. We also accept NLM_F_EXCL in case that bug ever
1005 if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1008 goto err_unlock_ovs;
1010 /* The flow identifier has to be the same for flow updates.
1011 * Look for any overlapping flow.
1013 if (unlikely(!ovs_flow_cmp(flow, &match))) {
1014 if (ovs_identifier_is_key(&flow->id))
1015 flow = ovs_flow_tbl_lookup_exact(&dp->table,
1017 else /* UFID matches but key is different */
1021 goto err_unlock_ovs;
1024 /* Update actions. */
1025 old_acts = ovsl_dereference(flow->sf_acts);
1026 rcu_assign_pointer(flow->sf_acts, acts);
1028 if (unlikely(reply)) {
1029 error = ovs_flow_cmd_fill_info(flow,
1030 ovs_header->dp_ifindex,
1031 reply, info->snd_portid,
1039 ovs_nla_free_flow_actions_rcu(old_acts);
1040 ovs_flow_free(new_flow, false);
1044 ovs_notify(&dp_flow_genl_family, reply, info);
1051 ovs_nla_free_flow_actions(acts);
1053 ovs_flow_free(new_flow, false);
1058 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1059 static struct sw_flow_actions *get_flow_actions(struct net *net,
1060 const struct nlattr *a,
1061 const struct sw_flow_key *key,
1062 const struct sw_flow_mask *mask,
1065 struct sw_flow_actions *acts;
1066 struct sw_flow_key masked_key;
1069 ovs_flow_mask_key(&masked_key, key, true, mask);
1070 error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1073 "Actions may not be safe on all matching packets");
1074 return ERR_PTR(error);
1080 /* Factor out match-init and action-copy to avoid
1081 * "Wframe-larger-than=1024" warning. Because mask is only
1082 * used to get actions, we new a function to save some
1085 * If there are not key and action attrs, we return 0
1086 * directly. In the case, the caller will also not use the
1087 * match as before. If there is action attr, we try to get
1088 * actions and save them to *acts. Before returning from
1089 * the function, we reset the match->mask pointer. Because
1090 * we should not to return match object with dangling reference
1093 static int ovs_nla_init_match_and_action(struct net *net,
1094 struct sw_flow_match *match,
1095 struct sw_flow_key *key,
1097 struct sw_flow_actions **acts,
1100 struct sw_flow_mask mask;
1103 if (a[OVS_FLOW_ATTR_KEY]) {
1104 ovs_match_init(match, key, true, &mask);
1105 error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1106 a[OVS_FLOW_ATTR_MASK], log);
1111 if (a[OVS_FLOW_ATTR_ACTIONS]) {
1112 if (!a[OVS_FLOW_ATTR_KEY]) {
1114 "Flow key attribute not present in set flow.");
1119 *acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1121 if (IS_ERR(*acts)) {
1122 error = PTR_ERR(*acts);
1127 /* On success, error is 0. */
1133 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1135 struct net *net = sock_net(skb->sk);
1136 struct nlattr **a = info->attrs;
1137 struct ovs_header *ovs_header = info->userhdr;
1138 struct sw_flow_key key;
1139 struct sw_flow *flow;
1140 struct sk_buff *reply = NULL;
1141 struct datapath *dp;
1142 struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1143 struct sw_flow_match match;
1144 struct sw_flow_id sfid;
1145 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1147 bool log = !a[OVS_FLOW_ATTR_PROBE];
1150 ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1151 if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1153 "Flow set message rejected, Key attribute missing.");
1157 error = ovs_nla_init_match_and_action(net, &match, &key, a,
1163 /* Can allocate before locking if have acts. */
1164 reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1166 if (IS_ERR(reply)) {
1167 error = PTR_ERR(reply);
1168 goto err_kfree_acts;
1173 dp = get_dp(net, ovs_header->dp_ifindex);
1174 if (unlikely(!dp)) {
1176 goto err_unlock_ovs;
1178 /* Check that the flow exists. */
1180 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1182 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1183 if (unlikely(!flow)) {
1185 goto err_unlock_ovs;
1188 /* Update actions, if present. */
1190 old_acts = ovsl_dereference(flow->sf_acts);
1191 rcu_assign_pointer(flow->sf_acts, acts);
1193 if (unlikely(reply)) {
1194 error = ovs_flow_cmd_fill_info(flow,
1195 ovs_header->dp_ifindex,
1196 reply, info->snd_portid,
1203 /* Could not alloc without acts before locking. */
1204 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1205 info, OVS_FLOW_CMD_NEW, false,
1208 if (IS_ERR(reply)) {
1209 error = PTR_ERR(reply);
1210 goto err_unlock_ovs;
1215 if (a[OVS_FLOW_ATTR_CLEAR])
1216 ovs_flow_stats_clear(flow);
1220 ovs_notify(&dp_flow_genl_family, reply, info);
1222 ovs_nla_free_flow_actions_rcu(old_acts);
1230 ovs_nla_free_flow_actions(acts);
1235 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1237 struct nlattr **a = info->attrs;
1238 struct ovs_header *ovs_header = info->userhdr;
1239 struct net *net = sock_net(skb->sk);
1240 struct sw_flow_key key;
1241 struct sk_buff *reply;
1242 struct sw_flow *flow;
1243 struct datapath *dp;
1244 struct sw_flow_match match;
1245 struct sw_flow_id ufid;
1246 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1248 bool log = !a[OVS_FLOW_ATTR_PROBE];
1251 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1252 if (a[OVS_FLOW_ATTR_KEY]) {
1253 ovs_match_init(&match, &key, true, NULL);
1254 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1256 } else if (!ufid_present) {
1258 "Flow get message rejected, Key attribute missing.");
1265 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1272 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1274 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1280 reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1281 OVS_FLOW_CMD_NEW, true, ufid_flags);
1282 if (IS_ERR(reply)) {
1283 err = PTR_ERR(reply);
1288 return genlmsg_reply(reply, info);
1294 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1296 struct nlattr **a = info->attrs;
1297 struct ovs_header *ovs_header = info->userhdr;
1298 struct net *net = sock_net(skb->sk);
1299 struct sw_flow_key key;
1300 struct sk_buff *reply;
1301 struct sw_flow *flow = NULL;
1302 struct datapath *dp;
1303 struct sw_flow_match match;
1304 struct sw_flow_id ufid;
1305 u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1307 bool log = !a[OVS_FLOW_ATTR_PROBE];
1310 ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1311 if (a[OVS_FLOW_ATTR_KEY]) {
1312 ovs_match_init(&match, &key, true, NULL);
1313 err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1320 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1321 if (unlikely(!dp)) {
1326 if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1327 err = ovs_flow_tbl_flush(&dp->table);
1332 flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1334 flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1335 if (unlikely(!flow)) {
1340 ovs_flow_tbl_remove(&dp->table, flow);
1343 reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1344 &flow->id, info, false, ufid_flags);
1345 if (likely(reply)) {
1346 if (likely(!IS_ERR(reply))) {
1347 rcu_read_lock(); /*To keep RCU checker happy. */
1348 err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1349 reply, info->snd_portid,
1356 ovs_notify(&dp_flow_genl_family, reply, info);
1358 netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1362 ovs_flow_free(flow, true);
1369 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1371 struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1372 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1373 struct table_instance *ti;
1374 struct datapath *dp;
1378 err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1379 OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1382 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1385 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1391 ti = rcu_dereference(dp->table.ti);
1393 struct sw_flow *flow;
1396 bucket = cb->args[0];
1398 flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1402 if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1403 NETLINK_CB(cb->skb).portid,
1404 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1405 OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1408 cb->args[0] = bucket;
1415 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1416 [OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1417 [OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1418 [OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1419 [OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1420 [OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1421 [OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1422 [OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1425 static const struct genl_ops dp_flow_genl_ops[] = {
1426 { .cmd = OVS_FLOW_CMD_NEW,
1427 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1428 .policy = flow_policy,
1429 .doit = ovs_flow_cmd_new
1431 { .cmd = OVS_FLOW_CMD_DEL,
1432 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1433 .policy = flow_policy,
1434 .doit = ovs_flow_cmd_del
1436 { .cmd = OVS_FLOW_CMD_GET,
1437 .flags = 0, /* OK for unprivileged users. */
1438 .policy = flow_policy,
1439 .doit = ovs_flow_cmd_get,
1440 .dumpit = ovs_flow_cmd_dump
1442 { .cmd = OVS_FLOW_CMD_SET,
1443 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1444 .policy = flow_policy,
1445 .doit = ovs_flow_cmd_set,
1449 static struct genl_family dp_flow_genl_family __ro_after_init = {
1450 .hdrsize = sizeof(struct ovs_header),
1451 .name = OVS_FLOW_FAMILY,
1452 .version = OVS_FLOW_VERSION,
1453 .maxattr = OVS_FLOW_ATTR_MAX,
1455 .parallel_ops = true,
1456 .ops = dp_flow_genl_ops,
1457 .n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1458 .mcgrps = &ovs_dp_flow_multicast_group,
1460 .module = THIS_MODULE,
1463 static size_t ovs_dp_cmd_msg_size(void)
1465 size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1467 msgsize += nla_total_size(IFNAMSIZ);
1468 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1469 msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1470 msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1475 /* Called with ovs_mutex. */
1476 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1477 u32 portid, u32 seq, u32 flags, u8 cmd)
1479 struct ovs_header *ovs_header;
1480 struct ovs_dp_stats dp_stats;
1481 struct ovs_dp_megaflow_stats dp_megaflow_stats;
1484 ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1489 ovs_header->dp_ifindex = get_dpifindex(dp);
1491 err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1493 goto nla_put_failure;
1495 get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1496 if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1497 &dp_stats, OVS_DP_ATTR_PAD))
1498 goto nla_put_failure;
1500 if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1501 sizeof(struct ovs_dp_megaflow_stats),
1502 &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1503 goto nla_put_failure;
1505 if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1506 goto nla_put_failure;
1508 genlmsg_end(skb, ovs_header);
1512 genlmsg_cancel(skb, ovs_header);
1517 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1519 return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1522 /* Called with rcu_read_lock or ovs_mutex. */
1523 static struct datapath *lookup_datapath(struct net *net,
1524 const struct ovs_header *ovs_header,
1525 struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1527 struct datapath *dp;
1529 if (!a[OVS_DP_ATTR_NAME])
1530 dp = get_dp(net, ovs_header->dp_ifindex);
1532 struct vport *vport;
1534 vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1535 dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1537 return dp ? dp : ERR_PTR(-ENODEV);
1540 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1542 struct datapath *dp;
1544 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1548 WARN(dp->user_features, "Dropping previously announced user features\n");
1549 dp->user_features = 0;
1552 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1554 if (a[OVS_DP_ATTR_USER_FEATURES])
1555 dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1558 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1560 struct nlattr **a = info->attrs;
1561 struct vport_parms parms;
1562 struct sk_buff *reply;
1563 struct datapath *dp;
1564 struct vport *vport;
1565 struct ovs_net *ovs_net;
1569 if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1572 reply = ovs_dp_cmd_alloc_info();
1577 dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1579 goto err_free_reply;
1581 ovs_dp_set_net(dp, sock_net(skb->sk));
1583 /* Allocate table. */
1584 err = ovs_flow_tbl_init(&dp->table);
1588 dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1589 if (!dp->stats_percpu) {
1591 goto err_destroy_table;
1594 dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1598 goto err_destroy_percpu;
1601 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1602 INIT_HLIST_HEAD(&dp->ports[i]);
1604 /* Set up our datapath device. */
1605 parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1606 parms.type = OVS_VPORT_TYPE_INTERNAL;
1607 parms.options = NULL;
1609 parms.port_no = OVSP_LOCAL;
1610 parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1612 ovs_dp_change(dp, a);
1614 /* So far only local changes have been made, now need the lock. */
1617 vport = new_vport(&parms);
1618 if (IS_ERR(vport)) {
1619 err = PTR_ERR(vport);
1623 if (err == -EEXIST) {
1624 /* An outdated user space instance that does not understand
1625 * the concept of user_features has attempted to create a new
1626 * datapath and is likely to reuse it. Drop all user features.
1628 if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1629 ovs_dp_reset_user_features(skb, info);
1632 goto err_destroy_ports_array;
1635 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1636 info->snd_seq, 0, OVS_DP_CMD_NEW);
1639 ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1640 list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1644 ovs_notify(&dp_datapath_genl_family, reply, info);
1647 err_destroy_ports_array:
1651 free_percpu(dp->stats_percpu);
1653 ovs_flow_tbl_destroy(&dp->table);
1662 /* Called with ovs_mutex. */
1663 static void __dp_destroy(struct datapath *dp)
1667 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1668 struct vport *vport;
1669 struct hlist_node *n;
1671 hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1672 if (vport->port_no != OVSP_LOCAL)
1673 ovs_dp_detach_port(vport);
1676 list_del_rcu(&dp->list_node);
1678 /* OVSP_LOCAL is datapath internal port. We need to make sure that
1679 * all ports in datapath are destroyed first before freeing datapath.
1681 ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1683 /* RCU destroy the flow table */
1684 call_rcu(&dp->rcu, destroy_dp_rcu);
1687 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1689 struct sk_buff *reply;
1690 struct datapath *dp;
1693 reply = ovs_dp_cmd_alloc_info();
1698 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1701 goto err_unlock_free;
1703 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1704 info->snd_seq, 0, OVS_DP_CMD_DEL);
1710 ovs_notify(&dp_datapath_genl_family, reply, info);
1720 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1722 struct sk_buff *reply;
1723 struct datapath *dp;
1726 reply = ovs_dp_cmd_alloc_info();
1731 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1734 goto err_unlock_free;
1736 ovs_dp_change(dp, info->attrs);
1738 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1739 info->snd_seq, 0, OVS_DP_CMD_NEW);
1743 ovs_notify(&dp_datapath_genl_family, reply, info);
1753 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1755 struct sk_buff *reply;
1756 struct datapath *dp;
1759 reply = ovs_dp_cmd_alloc_info();
1764 dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1767 goto err_unlock_free;
1769 err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1770 info->snd_seq, 0, OVS_DP_CMD_NEW);
1774 return genlmsg_reply(reply, info);
1782 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1784 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1785 struct datapath *dp;
1786 int skip = cb->args[0];
1790 list_for_each_entry(dp, &ovs_net->dps, list_node) {
1792 ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1793 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1794 OVS_DP_CMD_NEW) < 0)
1805 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1806 [OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1807 [OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1808 [OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1811 static const struct genl_ops dp_datapath_genl_ops[] = {
1812 { .cmd = OVS_DP_CMD_NEW,
1813 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1814 .policy = datapath_policy,
1815 .doit = ovs_dp_cmd_new
1817 { .cmd = OVS_DP_CMD_DEL,
1818 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1819 .policy = datapath_policy,
1820 .doit = ovs_dp_cmd_del
1822 { .cmd = OVS_DP_CMD_GET,
1823 .flags = 0, /* OK for unprivileged users. */
1824 .policy = datapath_policy,
1825 .doit = ovs_dp_cmd_get,
1826 .dumpit = ovs_dp_cmd_dump
1828 { .cmd = OVS_DP_CMD_SET,
1829 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1830 .policy = datapath_policy,
1831 .doit = ovs_dp_cmd_set,
1835 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1836 .hdrsize = sizeof(struct ovs_header),
1837 .name = OVS_DATAPATH_FAMILY,
1838 .version = OVS_DATAPATH_VERSION,
1839 .maxattr = OVS_DP_ATTR_MAX,
1841 .parallel_ops = true,
1842 .ops = dp_datapath_genl_ops,
1843 .n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1844 .mcgrps = &ovs_dp_datapath_multicast_group,
1846 .module = THIS_MODULE,
1849 /* Called with ovs_mutex or RCU read lock. */
1850 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1851 u32 portid, u32 seq, u32 flags, u8 cmd)
1853 struct ovs_header *ovs_header;
1854 struct ovs_vport_stats vport_stats;
1857 ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1862 ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1864 if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1865 nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1866 nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1867 ovs_vport_name(vport)))
1868 goto nla_put_failure;
1870 ovs_vport_get_stats(vport, &vport_stats);
1871 if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1872 sizeof(struct ovs_vport_stats), &vport_stats,
1873 OVS_VPORT_ATTR_PAD))
1874 goto nla_put_failure;
1876 if (ovs_vport_get_upcall_portids(vport, skb))
1877 goto nla_put_failure;
1879 err = ovs_vport_get_options(vport, skb);
1880 if (err == -EMSGSIZE)
1883 genlmsg_end(skb, ovs_header);
1889 genlmsg_cancel(skb, ovs_header);
1893 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1895 return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1898 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1899 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1902 struct sk_buff *skb;
1905 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1907 return ERR_PTR(-ENOMEM);
1909 retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1915 /* Called with ovs_mutex or RCU read lock. */
1916 static struct vport *lookup_vport(struct net *net,
1917 const struct ovs_header *ovs_header,
1918 struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1920 struct datapath *dp;
1921 struct vport *vport;
1923 if (a[OVS_VPORT_ATTR_NAME]) {
1924 vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1926 return ERR_PTR(-ENODEV);
1927 if (ovs_header->dp_ifindex &&
1928 ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1929 return ERR_PTR(-ENODEV);
1931 } else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1932 u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1934 if (port_no >= DP_MAX_PORTS)
1935 return ERR_PTR(-EFBIG);
1937 dp = get_dp(net, ovs_header->dp_ifindex);
1939 return ERR_PTR(-ENODEV);
1941 vport = ovs_vport_ovsl_rcu(dp, port_no);
1943 return ERR_PTR(-ENODEV);
1946 return ERR_PTR(-EINVAL);
1949 /* Called with ovs_mutex */
1950 static void update_headroom(struct datapath *dp)
1952 unsigned dev_headroom, max_headroom = 0;
1953 struct net_device *dev;
1954 struct vport *vport;
1957 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1958 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1960 dev_headroom = netdev_get_fwd_headroom(dev);
1961 if (dev_headroom > max_headroom)
1962 max_headroom = dev_headroom;
1966 dp->max_headroom = max_headroom;
1967 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1968 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1969 netdev_set_rx_headroom(vport->dev, max_headroom);
1972 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1974 struct nlattr **a = info->attrs;
1975 struct ovs_header *ovs_header = info->userhdr;
1976 struct vport_parms parms;
1977 struct sk_buff *reply;
1978 struct vport *vport;
1979 struct datapath *dp;
1983 if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1984 !a[OVS_VPORT_ATTR_UPCALL_PID])
1987 port_no = a[OVS_VPORT_ATTR_PORT_NO]
1988 ? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1989 if (port_no >= DP_MAX_PORTS)
1992 reply = ovs_vport_cmd_alloc_info();
1998 dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2001 goto exit_unlock_free;
2004 vport = ovs_vport_ovsl(dp, port_no);
2007 goto exit_unlock_free;
2009 for (port_no = 1; ; port_no++) {
2010 if (port_no >= DP_MAX_PORTS) {
2012 goto exit_unlock_free;
2014 vport = ovs_vport_ovsl(dp, port_no);
2020 parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2021 parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2022 parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2024 parms.port_no = port_no;
2025 parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2027 vport = new_vport(&parms);
2028 err = PTR_ERR(vport);
2029 if (IS_ERR(vport)) {
2032 goto exit_unlock_free;
2035 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2036 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2038 if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2039 update_headroom(dp);
2041 netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2046 ovs_notify(&dp_vport_genl_family, reply, info);
2055 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2057 struct nlattr **a = info->attrs;
2058 struct sk_buff *reply;
2059 struct vport *vport;
2062 reply = ovs_vport_cmd_alloc_info();
2067 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2068 err = PTR_ERR(vport);
2070 goto exit_unlock_free;
2072 if (a[OVS_VPORT_ATTR_TYPE] &&
2073 nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2075 goto exit_unlock_free;
2078 if (a[OVS_VPORT_ATTR_OPTIONS]) {
2079 err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2081 goto exit_unlock_free;
2085 if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2086 struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2088 err = ovs_vport_set_upcall_portids(vport, ids);
2090 goto exit_unlock_free;
2093 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2094 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2098 ovs_notify(&dp_vport_genl_family, reply, info);
2107 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2109 bool must_update_headroom = false;
2110 struct nlattr **a = info->attrs;
2111 struct sk_buff *reply;
2112 struct datapath *dp;
2113 struct vport *vport;
2116 reply = ovs_vport_cmd_alloc_info();
2121 vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2122 err = PTR_ERR(vport);
2124 goto exit_unlock_free;
2126 if (vport->port_no == OVSP_LOCAL) {
2128 goto exit_unlock_free;
2131 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2132 info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2135 /* the vport deletion may trigger dp headroom update */
2137 if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2138 must_update_headroom = true;
2139 netdev_reset_rx_headroom(vport->dev);
2140 ovs_dp_detach_port(vport);
2142 if (must_update_headroom)
2143 update_headroom(dp);
2146 ovs_notify(&dp_vport_genl_family, reply, info);
2155 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2157 struct nlattr **a = info->attrs;
2158 struct ovs_header *ovs_header = info->userhdr;
2159 struct sk_buff *reply;
2160 struct vport *vport;
2163 reply = ovs_vport_cmd_alloc_info();
2168 vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2169 err = PTR_ERR(vport);
2171 goto exit_unlock_free;
2172 err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2173 info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2177 return genlmsg_reply(reply, info);
2185 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2187 struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2188 struct datapath *dp;
2189 int bucket = cb->args[0], skip = cb->args[1];
2193 dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2198 for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2199 struct vport *vport;
2202 hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2204 ovs_vport_cmd_fill_info(vport, skb,
2205 NETLINK_CB(cb->skb).portid,
2208 OVS_VPORT_CMD_NEW) < 0)
2224 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2225 [OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2226 [OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2227 [OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2228 [OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2229 [OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2230 [OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2233 static const struct genl_ops dp_vport_genl_ops[] = {
2234 { .cmd = OVS_VPORT_CMD_NEW,
2235 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2236 .policy = vport_policy,
2237 .doit = ovs_vport_cmd_new
2239 { .cmd = OVS_VPORT_CMD_DEL,
2240 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2241 .policy = vport_policy,
2242 .doit = ovs_vport_cmd_del
2244 { .cmd = OVS_VPORT_CMD_GET,
2245 .flags = 0, /* OK for unprivileged users. */
2246 .policy = vport_policy,
2247 .doit = ovs_vport_cmd_get,
2248 .dumpit = ovs_vport_cmd_dump
2250 { .cmd = OVS_VPORT_CMD_SET,
2251 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2252 .policy = vport_policy,
2253 .doit = ovs_vport_cmd_set,
2257 struct genl_family dp_vport_genl_family __ro_after_init = {
2258 .hdrsize = sizeof(struct ovs_header),
2259 .name = OVS_VPORT_FAMILY,
2260 .version = OVS_VPORT_VERSION,
2261 .maxattr = OVS_VPORT_ATTR_MAX,
2263 .parallel_ops = true,
2264 .ops = dp_vport_genl_ops,
2265 .n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2266 .mcgrps = &ovs_dp_vport_multicast_group,
2268 .module = THIS_MODULE,
2271 static struct genl_family * const dp_genl_families[] = {
2272 &dp_datapath_genl_family,
2273 &dp_vport_genl_family,
2274 &dp_flow_genl_family,
2275 &dp_packet_genl_family,
2278 static void dp_unregister_genl(int n_families)
2282 for (i = 0; i < n_families; i++)
2283 genl_unregister_family(dp_genl_families[i]);
2286 static int __init dp_register_genl(void)
2291 for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2293 err = genl_register_family(dp_genl_families[i]);
2301 dp_unregister_genl(i);
2305 static int __net_init ovs_init_net(struct net *net)
2307 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2309 INIT_LIST_HEAD(&ovs_net->dps);
2310 INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2315 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2316 struct list_head *head)
2318 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2319 struct datapath *dp;
2321 list_for_each_entry(dp, &ovs_net->dps, list_node) {
2324 for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2325 struct vport *vport;
2327 hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2328 if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2331 if (dev_net(vport->dev) == dnet)
2332 list_add(&vport->detach_list, head);
2338 static void __net_exit ovs_exit_net(struct net *dnet)
2340 struct datapath *dp, *dp_next;
2341 struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2342 struct vport *vport, *vport_next;
2348 list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2353 list_vports_from_net(net, dnet, &head);
2356 /* Detach all vports from given namespace. */
2357 list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2358 list_del(&vport->detach_list);
2359 ovs_dp_detach_port(vport);
2364 cancel_work_sync(&ovs_net->dp_notify_work);
2367 static struct pernet_operations ovs_net_ops = {
2368 .init = ovs_init_net,
2369 .exit = ovs_exit_net,
2371 .size = sizeof(struct ovs_net),
2374 static int __init dp_init(void)
2378 BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2380 pr_info("Open vSwitch switching datapath\n");
2382 err = action_fifos_init();
2386 err = ovs_internal_dev_rtnl_link_register();
2388 goto error_action_fifos_exit;
2390 err = ovs_flow_init();
2392 goto error_unreg_rtnl_link;
2394 err = ovs_vport_init();
2396 goto error_flow_exit;
2398 err = register_pernet_device(&ovs_net_ops);
2400 goto error_vport_exit;
2402 err = register_netdevice_notifier(&ovs_dp_device_notifier);
2404 goto error_netns_exit;
2406 err = ovs_netdev_init();
2408 goto error_unreg_notifier;
2410 err = dp_register_genl();
2412 goto error_unreg_netdev;
2418 error_unreg_notifier:
2419 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2421 unregister_pernet_device(&ovs_net_ops);
2426 error_unreg_rtnl_link:
2427 ovs_internal_dev_rtnl_link_unregister();
2428 error_action_fifos_exit:
2429 action_fifos_exit();
2434 static void dp_cleanup(void)
2436 dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2438 unregister_netdevice_notifier(&ovs_dp_device_notifier);
2439 unregister_pernet_device(&ovs_net_ops);
2443 ovs_internal_dev_rtnl_link_unregister();
2444 action_fifos_exit();
2447 module_init(dp_init);
2448 module_exit(dp_cleanup);
2450 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2451 MODULE_LICENSE("GPL");
2452 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2453 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2454 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2455 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);