xref: /linux/net/openvswitch/datapath.c (revision 4949009eb8d40a441dcddcd96e101e77d31cf1b2)
1 /*
2  * Copyright (c) 2007-2014 Nicira, Inc.
3  *
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.
7  *
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.
12  *
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
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.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>
53 
54 #include "datapath.h"
55 #include "flow.h"
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
60 
61 int ovs_net_id __read_mostly;
62 EXPORT_SYMBOL_GPL(ovs_net_id);
63 
64 static struct genl_family dp_packet_genl_family;
65 static struct genl_family dp_flow_genl_family;
66 static struct genl_family dp_datapath_genl_family;
67 
68 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
69 	.name = OVS_FLOW_MCGROUP,
70 };
71 
72 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
73 	.name = OVS_DATAPATH_MCGROUP,
74 };
75 
76 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
77 	.name = OVS_VPORT_MCGROUP,
78 };
79 
80 /* Check if need to build a reply message.
81  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
82 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
83 			    unsigned int group)
84 {
85 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
86 	       genl_has_listeners(family, genl_info_net(info), group);
87 }
88 
89 static void ovs_notify(struct genl_family *family,
90 		       struct sk_buff *skb, struct genl_info *info)
91 {
92 	genl_notify(family, skb, genl_info_net(info), info->snd_portid,
93 		    0, info->nlhdr, GFP_KERNEL);
94 }
95 
96 /**
97  * DOC: Locking:
98  *
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
102  * by ovs_lock.
103  *
104  * Reads are protected by RCU.
105  *
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
108  * each other.
109  *
110  * The RTNL lock nests inside ovs_mutex.
111  */
112 
113 static DEFINE_MUTEX(ovs_mutex);
114 
115 void ovs_lock(void)
116 {
117 	mutex_lock(&ovs_mutex);
118 }
119 
120 void ovs_unlock(void)
121 {
122 	mutex_unlock(&ovs_mutex);
123 }
124 
125 #ifdef CONFIG_LOCKDEP
126 int lockdep_ovsl_is_held(void)
127 {
128 	if (debug_locks)
129 		return lockdep_is_held(&ovs_mutex);
130 	else
131 		return 1;
132 }
133 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
134 #endif
135 
136 static struct vport *new_vport(const struct vport_parms *);
137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
138 			     const struct sw_flow_key *,
139 			     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 *);
143 
144 /* Must be called with rcu_read_lock. */
145 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
146 {
147 	struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
148 
149 	if (dev) {
150 		struct vport *vport = ovs_internal_dev_get_vport(dev);
151 		if (vport)
152 			return vport->dp;
153 	}
154 
155 	return NULL;
156 }
157 
158 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
159  * returned dp pointer valid.
160  */
161 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
162 {
163 	struct datapath *dp;
164 
165 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
166 	rcu_read_lock();
167 	dp = get_dp_rcu(net, dp_ifindex);
168 	rcu_read_unlock();
169 
170 	return dp;
171 }
172 
173 /* Must be called with rcu_read_lock or ovs_mutex. */
174 const char *ovs_dp_name(const struct datapath *dp)
175 {
176 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
177 	return vport->ops->get_name(vport);
178 }
179 
180 static int get_dpifindex(const struct datapath *dp)
181 {
182 	struct vport *local;
183 	int ifindex;
184 
185 	rcu_read_lock();
186 
187 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
188 	if (local)
189 		ifindex = netdev_vport_priv(local)->dev->ifindex;
190 	else
191 		ifindex = 0;
192 
193 	rcu_read_unlock();
194 
195 	return ifindex;
196 }
197 
198 static void destroy_dp_rcu(struct rcu_head *rcu)
199 {
200 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
201 
202 	ovs_flow_tbl_destroy(&dp->table);
203 	free_percpu(dp->stats_percpu);
204 	release_net(ovs_dp_get_net(dp));
205 	kfree(dp->ports);
206 	kfree(dp);
207 }
208 
209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
210 					    u16 port_no)
211 {
212 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
213 }
214 
215 /* Called with ovs_mutex or RCU read lock. */
216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
217 {
218 	struct vport *vport;
219 	struct hlist_head *head;
220 
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)
224 			return vport;
225 	}
226 	return NULL;
227 }
228 
229 /* Called with ovs_mutex. */
230 static struct vport *new_vport(const struct vport_parms *parms)
231 {
232 	struct vport *vport;
233 
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);
238 
239 		hlist_add_head_rcu(&vport->dp_hash_node, head);
240 	}
241 	return vport;
242 }
243 
244 void ovs_dp_detach_port(struct vport *p)
245 {
246 	ASSERT_OVSL();
247 
248 	/* First drop references to device. */
249 	hlist_del_rcu(&p->dp_hash_node);
250 
251 	/* Then destroy it. */
252 	ovs_vport_del(p);
253 }
254 
255 /* Must be called with rcu_read_lock. */
256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
257 {
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;
263 	u64 *stats_counter;
264 	u32 n_mask_hit;
265 
266 	stats = this_cpu_ptr(dp->stats_percpu);
267 
268 	/* Look up flow. */
269 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
270 	if (unlikely(!flow)) {
271 		struct dp_upcall_info upcall;
272 		int error;
273 
274 		upcall.cmd = OVS_PACKET_CMD_MISS;
275 		upcall.userdata = NULL;
276 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 		upcall.egress_tun_info = NULL;
278 		error = ovs_dp_upcall(dp, skb, key, &upcall);
279 		if (unlikely(error))
280 			kfree_skb(skb);
281 		else
282 			consume_skb(skb);
283 		stats_counter = &stats->n_missed;
284 		goto out;
285 	}
286 
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);
290 
291 	stats_counter = &stats->n_hit;
292 
293 out:
294 	/* Update datapath statistics. */
295 	u64_stats_update_begin(&stats->syncp);
296 	(*stats_counter)++;
297 	stats->n_mask_hit += n_mask_hit;
298 	u64_stats_update_end(&stats->syncp);
299 }
300 
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)
304 {
305 	struct dp_stats_percpu *stats;
306 	int err;
307 
308 	if (upcall_info->portid == 0) {
309 		err = -ENOTCONN;
310 		goto err;
311 	}
312 
313 	if (!skb_is_gso(skb))
314 		err = queue_userspace_packet(dp, skb, key, upcall_info);
315 	else
316 		err = queue_gso_packets(dp, skb, key, upcall_info);
317 	if (err)
318 		goto err;
319 
320 	return 0;
321 
322 err:
323 	stats = this_cpu_ptr(dp->stats_percpu);
324 
325 	u64_stats_update_begin(&stats->syncp);
326 	stats->n_lost++;
327 	u64_stats_update_end(&stats->syncp);
328 
329 	return err;
330 }
331 
332 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
333 			     const struct sw_flow_key *key,
334 			     const struct dp_upcall_info *upcall_info)
335 {
336 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
337 	struct sw_flow_key later_key;
338 	struct sk_buff *segs, *nskb;
339 	struct ovs_skb_cb ovs_cb;
340 	int err;
341 
342 	ovs_cb = *OVS_CB(skb);
343 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
344 	*OVS_CB(skb) = ovs_cb;
345 	if (IS_ERR(segs))
346 		return PTR_ERR(segs);
347 	if (segs == NULL)
348 		return -EINVAL;
349 
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.
354 		 */
355 		later_key = *key;
356 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
357 	}
358 
359 	/* Queue all of the segments. */
360 	skb = segs;
361 	do {
362 		*OVS_CB(skb) = ovs_cb;
363 		if (gso_type & SKB_GSO_UDP && skb != segs)
364 			key = &later_key;
365 
366 		err = queue_userspace_packet(dp, skb, key, upcall_info);
367 		if (err)
368 			break;
369 
370 	} while ((skb = skb->next));
371 
372 	/* Free all of the segments. */
373 	skb = segs;
374 	do {
375 		nskb = skb->next;
376 		if (err)
377 			kfree_skb(skb);
378 		else
379 			consume_skb(skb);
380 	} while ((skb = nskb));
381 	return err;
382 }
383 
384 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
385 			      unsigned int hdrlen)
386 {
387 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
388 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
389 		+ nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
390 
391 	/* OVS_PACKET_ATTR_USERDATA */
392 	if (upcall_info->userdata)
393 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
394 
395 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
396 	if (upcall_info->egress_tun_info)
397 		size += nla_total_size(ovs_tun_key_attr_size());
398 
399 	return size;
400 }
401 
402 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
403 				  const struct sw_flow_key *key,
404 				  const struct dp_upcall_info *upcall_info)
405 {
406 	struct ovs_header *upcall;
407 	struct sk_buff *nskb = NULL;
408 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
409 	struct nlattr *nla;
410 	struct genl_info info = {
411 		.dst_sk = ovs_dp_get_net(dp)->genl_sock,
412 		.snd_portid = upcall_info->portid,
413 	};
414 	size_t len;
415 	unsigned int hlen;
416 	int err, dp_ifindex;
417 
418 	dp_ifindex = get_dpifindex(dp);
419 	if (!dp_ifindex)
420 		return -ENODEV;
421 
422 	if (vlan_tx_tag_present(skb)) {
423 		nskb = skb_clone(skb, GFP_ATOMIC);
424 		if (!nskb)
425 			return -ENOMEM;
426 
427 		nskb = __vlan_hwaccel_push_inside(nskb);
428 		if (!nskb)
429 			return -ENOMEM;
430 
431 		skb = nskb;
432 	}
433 
434 	if (nla_attr_size(skb->len) > USHRT_MAX) {
435 		err = -EFBIG;
436 		goto out;
437 	}
438 
439 	/* Complete checksum if needed */
440 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
441 	    (err = skb_checksum_help(skb)))
442 		goto out;
443 
444 	/* Older versions of OVS user space enforce alignment of the last
445 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
446 	 * padding logic. Only perform zerocopy if padding is not required.
447 	 */
448 	if (dp->user_features & OVS_DP_F_UNALIGNED)
449 		hlen = skb_zerocopy_headlen(skb);
450 	else
451 		hlen = skb->len;
452 
453 	len = upcall_msg_size(upcall_info, hlen);
454 	user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
455 	if (!user_skb) {
456 		err = -ENOMEM;
457 		goto out;
458 	}
459 
460 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
461 			     0, upcall_info->cmd);
462 	upcall->dp_ifindex = dp_ifindex;
463 
464 	nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
465 	err = ovs_nla_put_flow(key, key, user_skb);
466 	BUG_ON(err);
467 	nla_nest_end(user_skb, nla);
468 
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));
473 
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_egress_tunnel_key(user_skb,
477 						    upcall_info->egress_tun_info);
478 		BUG_ON(err);
479 		nla_nest_end(user_skb, nla);
480 	}
481 
482 	/* Only reserve room for attribute header, packet data is added
483 	 * in skb_zerocopy() */
484 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
485 		err = -ENOBUFS;
486 		goto out;
487 	}
488 	nla->nla_len = nla_attr_size(skb->len);
489 
490 	err = skb_zerocopy(user_skb, skb, skb->len, hlen);
491 	if (err)
492 		goto out;
493 
494 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
495 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
496 		size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
497 
498 		if (plen > 0)
499 			memset(skb_put(user_skb, plen), 0, plen);
500 	}
501 
502 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
503 
504 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
505 	user_skb = NULL;
506 out:
507 	if (err)
508 		skb_tx_error(skb);
509 	kfree_skb(user_skb);
510 	kfree_skb(nskb);
511 	return err;
512 }
513 
514 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
515 {
516 	struct ovs_header *ovs_header = info->userhdr;
517 	struct nlattr **a = info->attrs;
518 	struct sw_flow_actions *acts;
519 	struct sk_buff *packet;
520 	struct sw_flow *flow;
521 	struct sw_flow_actions *sf_acts;
522 	struct datapath *dp;
523 	struct ethhdr *eth;
524 	struct vport *input_vport;
525 	int len;
526 	int err;
527 	bool log = !a[OVS_PACKET_ATTR_PROBE];
528 
529 	err = -EINVAL;
530 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
531 	    !a[OVS_PACKET_ATTR_ACTIONS])
532 		goto err;
533 
534 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
535 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
536 	err = -ENOMEM;
537 	if (!packet)
538 		goto err;
539 	skb_reserve(packet, NET_IP_ALIGN);
540 
541 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
542 
543 	skb_reset_mac_header(packet);
544 	eth = eth_hdr(packet);
545 
546 	/* Normally, setting the skb 'protocol' field would be handled by a
547 	 * call to eth_type_trans(), but it assumes there's a sending
548 	 * device, which we may not have. */
549 	if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
550 		packet->protocol = eth->h_proto;
551 	else
552 		packet->protocol = htons(ETH_P_802_2);
553 
554 	/* Build an sw_flow for sending this packet. */
555 	flow = ovs_flow_alloc();
556 	err = PTR_ERR(flow);
557 	if (IS_ERR(flow))
558 		goto err_kfree_skb;
559 
560 	err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
561 					     &flow->key, log);
562 	if (err)
563 		goto err_flow_free;
564 
565 	err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
566 				   &flow->key, &acts, log);
567 	if (err)
568 		goto err_flow_free;
569 
570 	rcu_assign_pointer(flow->sf_acts, acts);
571 	OVS_CB(packet)->egress_tun_info = NULL;
572 	packet->priority = flow->key.phy.priority;
573 	packet->mark = flow->key.phy.skb_mark;
574 
575 	rcu_read_lock();
576 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
577 	err = -ENODEV;
578 	if (!dp)
579 		goto err_unlock;
580 
581 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
582 	if (!input_vport)
583 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
584 
585 	if (!input_vport)
586 		goto err_unlock;
587 
588 	OVS_CB(packet)->input_vport = input_vport;
589 	sf_acts = rcu_dereference(flow->sf_acts);
590 
591 	local_bh_disable();
592 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
593 	local_bh_enable();
594 	rcu_read_unlock();
595 
596 	ovs_flow_free(flow, false);
597 	return err;
598 
599 err_unlock:
600 	rcu_read_unlock();
601 err_flow_free:
602 	ovs_flow_free(flow, false);
603 err_kfree_skb:
604 	kfree_skb(packet);
605 err:
606 	return err;
607 }
608 
609 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
610 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
611 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
612 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
613 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
614 };
615 
616 static const struct genl_ops dp_packet_genl_ops[] = {
617 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
618 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
619 	  .policy = packet_policy,
620 	  .doit = ovs_packet_cmd_execute
621 	}
622 };
623 
624 static struct genl_family dp_packet_genl_family = {
625 	.id = GENL_ID_GENERATE,
626 	.hdrsize = sizeof(struct ovs_header),
627 	.name = OVS_PACKET_FAMILY,
628 	.version = OVS_PACKET_VERSION,
629 	.maxattr = OVS_PACKET_ATTR_MAX,
630 	.netnsok = true,
631 	.parallel_ops = true,
632 	.ops = dp_packet_genl_ops,
633 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
634 };
635 
636 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
637 			 struct ovs_dp_megaflow_stats *mega_stats)
638 {
639 	int i;
640 
641 	memset(mega_stats, 0, sizeof(*mega_stats));
642 
643 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
644 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
645 
646 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
647 
648 	for_each_possible_cpu(i) {
649 		const struct dp_stats_percpu *percpu_stats;
650 		struct dp_stats_percpu local_stats;
651 		unsigned int start;
652 
653 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
654 
655 		do {
656 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
657 			local_stats = *percpu_stats;
658 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
659 
660 		stats->n_hit += local_stats.n_hit;
661 		stats->n_missed += local_stats.n_missed;
662 		stats->n_lost += local_stats.n_lost;
663 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
664 	}
665 }
666 
667 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
668 {
669 	return NLMSG_ALIGN(sizeof(struct ovs_header))
670 		+ nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_KEY */
671 		+ nla_total_size(ovs_key_attr_size()) /* OVS_FLOW_ATTR_MASK */
672 		+ nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
673 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
674 		+ nla_total_size(8) /* OVS_FLOW_ATTR_USED */
675 		+ nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
676 }
677 
678 /* Called with ovs_mutex or RCU read lock. */
679 static int ovs_flow_cmd_fill_match(const struct sw_flow *flow,
680 				   struct sk_buff *skb)
681 {
682 	struct nlattr *nla;
683 	int err;
684 
685 	/* Fill flow key. */
686 	nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
687 	if (!nla)
688 		return -EMSGSIZE;
689 
690 	err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
691 	if (err)
692 		return err;
693 
694 	nla_nest_end(skb, nla);
695 
696 	/* Fill flow mask. */
697 	nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
698 	if (!nla)
699 		return -EMSGSIZE;
700 
701 	err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
702 	if (err)
703 		return err;
704 
705 	nla_nest_end(skb, nla);
706 	return 0;
707 }
708 
709 /* Called with ovs_mutex or RCU read lock. */
710 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
711 				   struct sk_buff *skb)
712 {
713 	struct ovs_flow_stats stats;
714 	__be16 tcp_flags;
715 	unsigned long used;
716 
717 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
718 
719 	if (used &&
720 	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
721 		return -EMSGSIZE;
722 
723 	if (stats.n_packets &&
724 	    nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
725 		return -EMSGSIZE;
726 
727 	if ((u8)ntohs(tcp_flags) &&
728 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
729 		return -EMSGSIZE;
730 
731 	return 0;
732 }
733 
734 /* Called with ovs_mutex or RCU read lock. */
735 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
736 				     struct sk_buff *skb, int skb_orig_len)
737 {
738 	struct nlattr *start;
739 	int err;
740 
741 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
742 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
743 	 * Netlink but individual action lists can be longer than
744 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
745 	 * The userspace caller can always fetch the actions separately if it
746 	 * really wants them.  (Most userspace callers in fact don't care.)
747 	 *
748 	 * This can only fail for dump operations because the skb is always
749 	 * properly sized for single flows.
750 	 */
751 	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
752 	if (start) {
753 		const struct sw_flow_actions *sf_acts;
754 
755 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
756 		err = ovs_nla_put_actions(sf_acts->actions,
757 					  sf_acts->actions_len, skb);
758 
759 		if (!err)
760 			nla_nest_end(skb, start);
761 		else {
762 			if (skb_orig_len)
763 				return err;
764 
765 			nla_nest_cancel(skb, start);
766 		}
767 	} else if (skb_orig_len) {
768 		return -EMSGSIZE;
769 	}
770 
771 	return 0;
772 }
773 
774 /* Called with ovs_mutex or RCU read lock. */
775 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
776 				  struct sk_buff *skb, u32 portid,
777 				  u32 seq, u32 flags, u8 cmd)
778 {
779 	const int skb_orig_len = skb->len;
780 	struct ovs_header *ovs_header;
781 	int err;
782 
783 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
784 				 flags, cmd);
785 	if (!ovs_header)
786 		return -EMSGSIZE;
787 
788 	ovs_header->dp_ifindex = dp_ifindex;
789 
790 	err = ovs_flow_cmd_fill_match(flow, skb);
791 	if (err)
792 		goto error;
793 
794 	err = ovs_flow_cmd_fill_stats(flow, skb);
795 	if (err)
796 		goto error;
797 
798 	err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
799 	if (err)
800 		goto error;
801 
802 	return genlmsg_end(skb, ovs_header);
803 
804 error:
805 	genlmsg_cancel(skb, ovs_header);
806 	return err;
807 }
808 
809 /* May not be called with RCU read lock. */
810 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
811 					       struct genl_info *info,
812 					       bool always)
813 {
814 	struct sk_buff *skb;
815 
816 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
817 		return NULL;
818 
819 	skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
820 	if (!skb)
821 		return ERR_PTR(-ENOMEM);
822 
823 	return skb;
824 }
825 
826 /* Called with ovs_mutex. */
827 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
828 					       int dp_ifindex,
829 					       struct genl_info *info, u8 cmd,
830 					       bool always)
831 {
832 	struct sk_buff *skb;
833 	int retval;
834 
835 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
836 				      always);
837 	if (IS_ERR_OR_NULL(skb))
838 		return skb;
839 
840 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
841 					info->snd_portid, info->snd_seq, 0,
842 					cmd);
843 	BUG_ON(retval < 0);
844 	return skb;
845 }
846 
847 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
848 {
849 	struct nlattr **a = info->attrs;
850 	struct ovs_header *ovs_header = info->userhdr;
851 	struct sw_flow *flow, *new_flow;
852 	struct sw_flow_mask mask;
853 	struct sk_buff *reply;
854 	struct datapath *dp;
855 	struct sw_flow_actions *acts;
856 	struct sw_flow_match match;
857 	int error;
858 	bool log = !a[OVS_FLOW_ATTR_PROBE];
859 
860 	/* Must have key and actions. */
861 	error = -EINVAL;
862 	if (!a[OVS_FLOW_ATTR_KEY]) {
863 		OVS_NLERR(log, "Flow key attr not present in new flow.");
864 		goto error;
865 	}
866 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
867 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
868 		goto error;
869 	}
870 
871 	/* Most of the time we need to allocate a new flow, do it before
872 	 * locking.
873 	 */
874 	new_flow = ovs_flow_alloc();
875 	if (IS_ERR(new_flow)) {
876 		error = PTR_ERR(new_flow);
877 		goto error;
878 	}
879 
880 	/* Extract key. */
881 	ovs_match_init(&match, &new_flow->unmasked_key, &mask);
882 	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
883 				  a[OVS_FLOW_ATTR_MASK], log);
884 	if (error)
885 		goto err_kfree_flow;
886 
887 	ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
888 
889 	/* Validate actions. */
890 	error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
891 				     &acts, log);
892 	if (error) {
893 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
894 		goto err_kfree_flow;
895 	}
896 
897 	reply = ovs_flow_cmd_alloc_info(acts, info, false);
898 	if (IS_ERR(reply)) {
899 		error = PTR_ERR(reply);
900 		goto err_kfree_acts;
901 	}
902 
903 	ovs_lock();
904 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
905 	if (unlikely(!dp)) {
906 		error = -ENODEV;
907 		goto err_unlock_ovs;
908 	}
909 	/* Check if this is a duplicate flow */
910 	flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
911 	if (likely(!flow)) {
912 		rcu_assign_pointer(new_flow->sf_acts, acts);
913 
914 		/* Put flow in bucket. */
915 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
916 		if (unlikely(error)) {
917 			acts = NULL;
918 			goto err_unlock_ovs;
919 		}
920 
921 		if (unlikely(reply)) {
922 			error = ovs_flow_cmd_fill_info(new_flow,
923 						       ovs_header->dp_ifindex,
924 						       reply, info->snd_portid,
925 						       info->snd_seq, 0,
926 						       OVS_FLOW_CMD_NEW);
927 			BUG_ON(error < 0);
928 		}
929 		ovs_unlock();
930 	} else {
931 		struct sw_flow_actions *old_acts;
932 
933 		/* Bail out if we're not allowed to modify an existing flow.
934 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
935 		 * because Generic Netlink treats the latter as a dump
936 		 * request.  We also accept NLM_F_EXCL in case that bug ever
937 		 * gets fixed.
938 		 */
939 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
940 							 | NLM_F_EXCL))) {
941 			error = -EEXIST;
942 			goto err_unlock_ovs;
943 		}
944 		/* The unmasked key has to be the same for flow updates. */
945 		if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
946 			/* Look for any overlapping flow. */
947 			flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
948 			if (!flow) {
949 				error = -ENOENT;
950 				goto err_unlock_ovs;
951 			}
952 		}
953 		/* Update actions. */
954 		old_acts = ovsl_dereference(flow->sf_acts);
955 		rcu_assign_pointer(flow->sf_acts, acts);
956 
957 		if (unlikely(reply)) {
958 			error = ovs_flow_cmd_fill_info(flow,
959 						       ovs_header->dp_ifindex,
960 						       reply, info->snd_portid,
961 						       info->snd_seq, 0,
962 						       OVS_FLOW_CMD_NEW);
963 			BUG_ON(error < 0);
964 		}
965 		ovs_unlock();
966 
967 		ovs_nla_free_flow_actions(old_acts);
968 		ovs_flow_free(new_flow, false);
969 	}
970 
971 	if (reply)
972 		ovs_notify(&dp_flow_genl_family, reply, info);
973 	return 0;
974 
975 err_unlock_ovs:
976 	ovs_unlock();
977 	kfree_skb(reply);
978 err_kfree_acts:
979 	kfree(acts);
980 err_kfree_flow:
981 	ovs_flow_free(new_flow, false);
982 error:
983 	return error;
984 }
985 
986 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
987 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
988 						const struct sw_flow_key *key,
989 						const struct sw_flow_mask *mask,
990 						bool log)
991 {
992 	struct sw_flow_actions *acts;
993 	struct sw_flow_key masked_key;
994 	int error;
995 
996 	ovs_flow_mask_key(&masked_key, key, mask);
997 	error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
998 	if (error) {
999 		OVS_NLERR(log,
1000 			  "Actions may not be safe on all matching packets");
1001 		return ERR_PTR(error);
1002 	}
1003 
1004 	return acts;
1005 }
1006 
1007 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1008 {
1009 	struct nlattr **a = info->attrs;
1010 	struct ovs_header *ovs_header = info->userhdr;
1011 	struct sw_flow_key key;
1012 	struct sw_flow *flow;
1013 	struct sw_flow_mask mask;
1014 	struct sk_buff *reply = NULL;
1015 	struct datapath *dp;
1016 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1017 	struct sw_flow_match match;
1018 	int error;
1019 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1020 
1021 	/* Extract key. */
1022 	error = -EINVAL;
1023 	if (!a[OVS_FLOW_ATTR_KEY]) {
1024 		OVS_NLERR(log, "Flow key attribute not present in set flow.");
1025 		goto error;
1026 	}
1027 
1028 	ovs_match_init(&match, &key, &mask);
1029 	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1030 				  a[OVS_FLOW_ATTR_MASK], log);
1031 	if (error)
1032 		goto error;
1033 
1034 	/* Validate actions. */
1035 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1036 		acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1037 					log);
1038 		if (IS_ERR(acts)) {
1039 			error = PTR_ERR(acts);
1040 			goto error;
1041 		}
1042 
1043 		/* Can allocate before locking if have acts. */
1044 		reply = ovs_flow_cmd_alloc_info(acts, info, false);
1045 		if (IS_ERR(reply)) {
1046 			error = PTR_ERR(reply);
1047 			goto err_kfree_acts;
1048 		}
1049 	}
1050 
1051 	ovs_lock();
1052 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1053 	if (unlikely(!dp)) {
1054 		error = -ENODEV;
1055 		goto err_unlock_ovs;
1056 	}
1057 	/* Check that the flow exists. */
1058 	flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1059 	if (unlikely(!flow)) {
1060 		error = -ENOENT;
1061 		goto err_unlock_ovs;
1062 	}
1063 
1064 	/* Update actions, if present. */
1065 	if (likely(acts)) {
1066 		old_acts = ovsl_dereference(flow->sf_acts);
1067 		rcu_assign_pointer(flow->sf_acts, acts);
1068 
1069 		if (unlikely(reply)) {
1070 			error = ovs_flow_cmd_fill_info(flow,
1071 						       ovs_header->dp_ifindex,
1072 						       reply, info->snd_portid,
1073 						       info->snd_seq, 0,
1074 						       OVS_FLOW_CMD_NEW);
1075 			BUG_ON(error < 0);
1076 		}
1077 	} else {
1078 		/* Could not alloc without acts before locking. */
1079 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1080 						info, OVS_FLOW_CMD_NEW, false);
1081 		if (unlikely(IS_ERR(reply))) {
1082 			error = PTR_ERR(reply);
1083 			goto err_unlock_ovs;
1084 		}
1085 	}
1086 
1087 	/* Clear stats. */
1088 	if (a[OVS_FLOW_ATTR_CLEAR])
1089 		ovs_flow_stats_clear(flow);
1090 	ovs_unlock();
1091 
1092 	if (reply)
1093 		ovs_notify(&dp_flow_genl_family, reply, info);
1094 	if (old_acts)
1095 		ovs_nla_free_flow_actions(old_acts);
1096 
1097 	return 0;
1098 
1099 err_unlock_ovs:
1100 	ovs_unlock();
1101 	kfree_skb(reply);
1102 err_kfree_acts:
1103 	kfree(acts);
1104 error:
1105 	return error;
1106 }
1107 
1108 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1109 {
1110 	struct nlattr **a = info->attrs;
1111 	struct ovs_header *ovs_header = info->userhdr;
1112 	struct sw_flow_key key;
1113 	struct sk_buff *reply;
1114 	struct sw_flow *flow;
1115 	struct datapath *dp;
1116 	struct sw_flow_match match;
1117 	int err;
1118 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1119 
1120 	if (!a[OVS_FLOW_ATTR_KEY]) {
1121 		OVS_NLERR(log,
1122 			  "Flow get message rejected, Key attribute missing.");
1123 		return -EINVAL;
1124 	}
1125 
1126 	ovs_match_init(&match, &key, NULL);
1127 	err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL, log);
1128 	if (err)
1129 		return err;
1130 
1131 	ovs_lock();
1132 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1133 	if (!dp) {
1134 		err = -ENODEV;
1135 		goto unlock;
1136 	}
1137 
1138 	flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1139 	if (!flow) {
1140 		err = -ENOENT;
1141 		goto unlock;
1142 	}
1143 
1144 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1145 					OVS_FLOW_CMD_NEW, true);
1146 	if (IS_ERR(reply)) {
1147 		err = PTR_ERR(reply);
1148 		goto unlock;
1149 	}
1150 
1151 	ovs_unlock();
1152 	return genlmsg_reply(reply, info);
1153 unlock:
1154 	ovs_unlock();
1155 	return err;
1156 }
1157 
1158 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1159 {
1160 	struct nlattr **a = info->attrs;
1161 	struct ovs_header *ovs_header = info->userhdr;
1162 	struct sw_flow_key key;
1163 	struct sk_buff *reply;
1164 	struct sw_flow *flow;
1165 	struct datapath *dp;
1166 	struct sw_flow_match match;
1167 	int err;
1168 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1169 
1170 	if (likely(a[OVS_FLOW_ATTR_KEY])) {
1171 		ovs_match_init(&match, &key, NULL);
1172 		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1173 					log);
1174 		if (unlikely(err))
1175 			return err;
1176 	}
1177 
1178 	ovs_lock();
1179 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1180 	if (unlikely(!dp)) {
1181 		err = -ENODEV;
1182 		goto unlock;
1183 	}
1184 
1185 	if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1186 		err = ovs_flow_tbl_flush(&dp->table);
1187 		goto unlock;
1188 	}
1189 
1190 	flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1191 	if (unlikely(!flow)) {
1192 		err = -ENOENT;
1193 		goto unlock;
1194 	}
1195 
1196 	ovs_flow_tbl_remove(&dp->table, flow);
1197 	ovs_unlock();
1198 
1199 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1200 					info, false);
1201 	if (likely(reply)) {
1202 		if (likely(!IS_ERR(reply))) {
1203 			rcu_read_lock();	/*To keep RCU checker happy. */
1204 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1205 						     reply, info->snd_portid,
1206 						     info->snd_seq, 0,
1207 						     OVS_FLOW_CMD_DEL);
1208 			rcu_read_unlock();
1209 			BUG_ON(err < 0);
1210 
1211 			ovs_notify(&dp_flow_genl_family, reply, info);
1212 		} else {
1213 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1214 		}
1215 	}
1216 
1217 	ovs_flow_free(flow, true);
1218 	return 0;
1219 unlock:
1220 	ovs_unlock();
1221 	return err;
1222 }
1223 
1224 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1225 {
1226 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1227 	struct table_instance *ti;
1228 	struct datapath *dp;
1229 
1230 	rcu_read_lock();
1231 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1232 	if (!dp) {
1233 		rcu_read_unlock();
1234 		return -ENODEV;
1235 	}
1236 
1237 	ti = rcu_dereference(dp->table.ti);
1238 	for (;;) {
1239 		struct sw_flow *flow;
1240 		u32 bucket, obj;
1241 
1242 		bucket = cb->args[0];
1243 		obj = cb->args[1];
1244 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1245 		if (!flow)
1246 			break;
1247 
1248 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1249 					   NETLINK_CB(cb->skb).portid,
1250 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1251 					   OVS_FLOW_CMD_NEW) < 0)
1252 			break;
1253 
1254 		cb->args[0] = bucket;
1255 		cb->args[1] = obj;
1256 	}
1257 	rcu_read_unlock();
1258 	return skb->len;
1259 }
1260 
1261 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1262 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1263 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1264 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1265 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1266 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1267 };
1268 
1269 static const struct genl_ops dp_flow_genl_ops[] = {
1270 	{ .cmd = OVS_FLOW_CMD_NEW,
1271 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1272 	  .policy = flow_policy,
1273 	  .doit = ovs_flow_cmd_new
1274 	},
1275 	{ .cmd = OVS_FLOW_CMD_DEL,
1276 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1277 	  .policy = flow_policy,
1278 	  .doit = ovs_flow_cmd_del
1279 	},
1280 	{ .cmd = OVS_FLOW_CMD_GET,
1281 	  .flags = 0,		    /* OK for unprivileged users. */
1282 	  .policy = flow_policy,
1283 	  .doit = ovs_flow_cmd_get,
1284 	  .dumpit = ovs_flow_cmd_dump
1285 	},
1286 	{ .cmd = OVS_FLOW_CMD_SET,
1287 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1288 	  .policy = flow_policy,
1289 	  .doit = ovs_flow_cmd_set,
1290 	},
1291 };
1292 
1293 static struct genl_family dp_flow_genl_family = {
1294 	.id = GENL_ID_GENERATE,
1295 	.hdrsize = sizeof(struct ovs_header),
1296 	.name = OVS_FLOW_FAMILY,
1297 	.version = OVS_FLOW_VERSION,
1298 	.maxattr = OVS_FLOW_ATTR_MAX,
1299 	.netnsok = true,
1300 	.parallel_ops = true,
1301 	.ops = dp_flow_genl_ops,
1302 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1303 	.mcgrps = &ovs_dp_flow_multicast_group,
1304 	.n_mcgrps = 1,
1305 };
1306 
1307 static size_t ovs_dp_cmd_msg_size(void)
1308 {
1309 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1310 
1311 	msgsize += nla_total_size(IFNAMSIZ);
1312 	msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1313 	msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1314 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1315 
1316 	return msgsize;
1317 }
1318 
1319 /* Called with ovs_mutex. */
1320 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1321 				u32 portid, u32 seq, u32 flags, u8 cmd)
1322 {
1323 	struct ovs_header *ovs_header;
1324 	struct ovs_dp_stats dp_stats;
1325 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1326 	int err;
1327 
1328 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1329 				   flags, cmd);
1330 	if (!ovs_header)
1331 		goto error;
1332 
1333 	ovs_header->dp_ifindex = get_dpifindex(dp);
1334 
1335 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1336 	if (err)
1337 		goto nla_put_failure;
1338 
1339 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1340 	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1341 			&dp_stats))
1342 		goto nla_put_failure;
1343 
1344 	if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1345 			sizeof(struct ovs_dp_megaflow_stats),
1346 			&dp_megaflow_stats))
1347 		goto nla_put_failure;
1348 
1349 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1350 		goto nla_put_failure;
1351 
1352 	return genlmsg_end(skb, ovs_header);
1353 
1354 nla_put_failure:
1355 	genlmsg_cancel(skb, ovs_header);
1356 error:
1357 	return -EMSGSIZE;
1358 }
1359 
1360 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1361 {
1362 	return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1363 }
1364 
1365 /* Called with rcu_read_lock or ovs_mutex. */
1366 static struct datapath *lookup_datapath(struct net *net,
1367 					const struct ovs_header *ovs_header,
1368 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1369 {
1370 	struct datapath *dp;
1371 
1372 	if (!a[OVS_DP_ATTR_NAME])
1373 		dp = get_dp(net, ovs_header->dp_ifindex);
1374 	else {
1375 		struct vport *vport;
1376 
1377 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1378 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1379 	}
1380 	return dp ? dp : ERR_PTR(-ENODEV);
1381 }
1382 
1383 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1384 {
1385 	struct datapath *dp;
1386 
1387 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1388 	if (IS_ERR(dp))
1389 		return;
1390 
1391 	WARN(dp->user_features, "Dropping previously announced user features\n");
1392 	dp->user_features = 0;
1393 }
1394 
1395 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1396 {
1397 	if (a[OVS_DP_ATTR_USER_FEATURES])
1398 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1399 }
1400 
1401 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1402 {
1403 	struct nlattr **a = info->attrs;
1404 	struct vport_parms parms;
1405 	struct sk_buff *reply;
1406 	struct datapath *dp;
1407 	struct vport *vport;
1408 	struct ovs_net *ovs_net;
1409 	int err, i;
1410 
1411 	err = -EINVAL;
1412 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1413 		goto err;
1414 
1415 	reply = ovs_dp_cmd_alloc_info(info);
1416 	if (!reply)
1417 		return -ENOMEM;
1418 
1419 	err = -ENOMEM;
1420 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1421 	if (dp == NULL)
1422 		goto err_free_reply;
1423 
1424 	ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1425 
1426 	/* Allocate table. */
1427 	err = ovs_flow_tbl_init(&dp->table);
1428 	if (err)
1429 		goto err_free_dp;
1430 
1431 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1432 	if (!dp->stats_percpu) {
1433 		err = -ENOMEM;
1434 		goto err_destroy_table;
1435 	}
1436 
1437 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1438 			    GFP_KERNEL);
1439 	if (!dp->ports) {
1440 		err = -ENOMEM;
1441 		goto err_destroy_percpu;
1442 	}
1443 
1444 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1445 		INIT_HLIST_HEAD(&dp->ports[i]);
1446 
1447 	/* Set up our datapath device. */
1448 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1449 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1450 	parms.options = NULL;
1451 	parms.dp = dp;
1452 	parms.port_no = OVSP_LOCAL;
1453 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1454 
1455 	ovs_dp_change(dp, a);
1456 
1457 	/* So far only local changes have been made, now need the lock. */
1458 	ovs_lock();
1459 
1460 	vport = new_vport(&parms);
1461 	if (IS_ERR(vport)) {
1462 		err = PTR_ERR(vport);
1463 		if (err == -EBUSY)
1464 			err = -EEXIST;
1465 
1466 		if (err == -EEXIST) {
1467 			/* An outdated user space instance that does not understand
1468 			 * the concept of user_features has attempted to create a new
1469 			 * datapath and is likely to reuse it. Drop all user features.
1470 			 */
1471 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1472 				ovs_dp_reset_user_features(skb, info);
1473 		}
1474 
1475 		goto err_destroy_ports_array;
1476 	}
1477 
1478 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1479 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1480 	BUG_ON(err < 0);
1481 
1482 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1483 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1484 
1485 	ovs_unlock();
1486 
1487 	ovs_notify(&dp_datapath_genl_family, reply, info);
1488 	return 0;
1489 
1490 err_destroy_ports_array:
1491 	ovs_unlock();
1492 	kfree(dp->ports);
1493 err_destroy_percpu:
1494 	free_percpu(dp->stats_percpu);
1495 err_destroy_table:
1496 	ovs_flow_tbl_destroy(&dp->table);
1497 err_free_dp:
1498 	release_net(ovs_dp_get_net(dp));
1499 	kfree(dp);
1500 err_free_reply:
1501 	kfree_skb(reply);
1502 err:
1503 	return err;
1504 }
1505 
1506 /* Called with ovs_mutex. */
1507 static void __dp_destroy(struct datapath *dp)
1508 {
1509 	int i;
1510 
1511 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1512 		struct vport *vport;
1513 		struct hlist_node *n;
1514 
1515 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1516 			if (vport->port_no != OVSP_LOCAL)
1517 				ovs_dp_detach_port(vport);
1518 	}
1519 
1520 	list_del_rcu(&dp->list_node);
1521 
1522 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1523 	 * all ports in datapath are destroyed first before freeing datapath.
1524 	 */
1525 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1526 
1527 	/* RCU destroy the flow table */
1528 	call_rcu(&dp->rcu, destroy_dp_rcu);
1529 }
1530 
1531 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1532 {
1533 	struct sk_buff *reply;
1534 	struct datapath *dp;
1535 	int err;
1536 
1537 	reply = ovs_dp_cmd_alloc_info(info);
1538 	if (!reply)
1539 		return -ENOMEM;
1540 
1541 	ovs_lock();
1542 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1543 	err = PTR_ERR(dp);
1544 	if (IS_ERR(dp))
1545 		goto err_unlock_free;
1546 
1547 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1548 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1549 	BUG_ON(err < 0);
1550 
1551 	__dp_destroy(dp);
1552 	ovs_unlock();
1553 
1554 	ovs_notify(&dp_datapath_genl_family, reply, info);
1555 
1556 	return 0;
1557 
1558 err_unlock_free:
1559 	ovs_unlock();
1560 	kfree_skb(reply);
1561 	return err;
1562 }
1563 
1564 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1565 {
1566 	struct sk_buff *reply;
1567 	struct datapath *dp;
1568 	int err;
1569 
1570 	reply = ovs_dp_cmd_alloc_info(info);
1571 	if (!reply)
1572 		return -ENOMEM;
1573 
1574 	ovs_lock();
1575 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1576 	err = PTR_ERR(dp);
1577 	if (IS_ERR(dp))
1578 		goto err_unlock_free;
1579 
1580 	ovs_dp_change(dp, info->attrs);
1581 
1582 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1583 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1584 	BUG_ON(err < 0);
1585 
1586 	ovs_unlock();
1587 	ovs_notify(&dp_datapath_genl_family, reply, info);
1588 
1589 	return 0;
1590 
1591 err_unlock_free:
1592 	ovs_unlock();
1593 	kfree_skb(reply);
1594 	return err;
1595 }
1596 
1597 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1598 {
1599 	struct sk_buff *reply;
1600 	struct datapath *dp;
1601 	int err;
1602 
1603 	reply = ovs_dp_cmd_alloc_info(info);
1604 	if (!reply)
1605 		return -ENOMEM;
1606 
1607 	ovs_lock();
1608 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1609 	if (IS_ERR(dp)) {
1610 		err = PTR_ERR(dp);
1611 		goto err_unlock_free;
1612 	}
1613 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1614 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1615 	BUG_ON(err < 0);
1616 	ovs_unlock();
1617 
1618 	return genlmsg_reply(reply, info);
1619 
1620 err_unlock_free:
1621 	ovs_unlock();
1622 	kfree_skb(reply);
1623 	return err;
1624 }
1625 
1626 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1627 {
1628 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1629 	struct datapath *dp;
1630 	int skip = cb->args[0];
1631 	int i = 0;
1632 
1633 	ovs_lock();
1634 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1635 		if (i >= skip &&
1636 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1637 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1638 					 OVS_DP_CMD_NEW) < 0)
1639 			break;
1640 		i++;
1641 	}
1642 	ovs_unlock();
1643 
1644 	cb->args[0] = i;
1645 
1646 	return skb->len;
1647 }
1648 
1649 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1650 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1651 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1652 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1653 };
1654 
1655 static const struct genl_ops dp_datapath_genl_ops[] = {
1656 	{ .cmd = OVS_DP_CMD_NEW,
1657 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1658 	  .policy = datapath_policy,
1659 	  .doit = ovs_dp_cmd_new
1660 	},
1661 	{ .cmd = OVS_DP_CMD_DEL,
1662 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1663 	  .policy = datapath_policy,
1664 	  .doit = ovs_dp_cmd_del
1665 	},
1666 	{ .cmd = OVS_DP_CMD_GET,
1667 	  .flags = 0,		    /* OK for unprivileged users. */
1668 	  .policy = datapath_policy,
1669 	  .doit = ovs_dp_cmd_get,
1670 	  .dumpit = ovs_dp_cmd_dump
1671 	},
1672 	{ .cmd = OVS_DP_CMD_SET,
1673 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1674 	  .policy = datapath_policy,
1675 	  .doit = ovs_dp_cmd_set,
1676 	},
1677 };
1678 
1679 static struct genl_family dp_datapath_genl_family = {
1680 	.id = GENL_ID_GENERATE,
1681 	.hdrsize = sizeof(struct ovs_header),
1682 	.name = OVS_DATAPATH_FAMILY,
1683 	.version = OVS_DATAPATH_VERSION,
1684 	.maxattr = OVS_DP_ATTR_MAX,
1685 	.netnsok = true,
1686 	.parallel_ops = true,
1687 	.ops = dp_datapath_genl_ops,
1688 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1689 	.mcgrps = &ovs_dp_datapath_multicast_group,
1690 	.n_mcgrps = 1,
1691 };
1692 
1693 /* Called with ovs_mutex or RCU read lock. */
1694 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1695 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1696 {
1697 	struct ovs_header *ovs_header;
1698 	struct ovs_vport_stats vport_stats;
1699 	int err;
1700 
1701 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1702 				 flags, cmd);
1703 	if (!ovs_header)
1704 		return -EMSGSIZE;
1705 
1706 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1707 
1708 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1709 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1710 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1711 			   vport->ops->get_name(vport)))
1712 		goto nla_put_failure;
1713 
1714 	ovs_vport_get_stats(vport, &vport_stats);
1715 	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1716 		    &vport_stats))
1717 		goto nla_put_failure;
1718 
1719 	if (ovs_vport_get_upcall_portids(vport, skb))
1720 		goto nla_put_failure;
1721 
1722 	err = ovs_vport_get_options(vport, skb);
1723 	if (err == -EMSGSIZE)
1724 		goto error;
1725 
1726 	return genlmsg_end(skb, ovs_header);
1727 
1728 nla_put_failure:
1729 	err = -EMSGSIZE;
1730 error:
1731 	genlmsg_cancel(skb, ovs_header);
1732 	return err;
1733 }
1734 
1735 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1736 {
1737 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1738 }
1739 
1740 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1741 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1742 					 u32 seq, u8 cmd)
1743 {
1744 	struct sk_buff *skb;
1745 	int retval;
1746 
1747 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1748 	if (!skb)
1749 		return ERR_PTR(-ENOMEM);
1750 
1751 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1752 	BUG_ON(retval < 0);
1753 
1754 	return skb;
1755 }
1756 
1757 /* Called with ovs_mutex or RCU read lock. */
1758 static struct vport *lookup_vport(struct net *net,
1759 				  const struct ovs_header *ovs_header,
1760 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1761 {
1762 	struct datapath *dp;
1763 	struct vport *vport;
1764 
1765 	if (a[OVS_VPORT_ATTR_NAME]) {
1766 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1767 		if (!vport)
1768 			return ERR_PTR(-ENODEV);
1769 		if (ovs_header->dp_ifindex &&
1770 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1771 			return ERR_PTR(-ENODEV);
1772 		return vport;
1773 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1774 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1775 
1776 		if (port_no >= DP_MAX_PORTS)
1777 			return ERR_PTR(-EFBIG);
1778 
1779 		dp = get_dp(net, ovs_header->dp_ifindex);
1780 		if (!dp)
1781 			return ERR_PTR(-ENODEV);
1782 
1783 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1784 		if (!vport)
1785 			return ERR_PTR(-ENODEV);
1786 		return vport;
1787 	} else
1788 		return ERR_PTR(-EINVAL);
1789 }
1790 
1791 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1792 {
1793 	struct nlattr **a = info->attrs;
1794 	struct ovs_header *ovs_header = info->userhdr;
1795 	struct vport_parms parms;
1796 	struct sk_buff *reply;
1797 	struct vport *vport;
1798 	struct datapath *dp;
1799 	u32 port_no;
1800 	int err;
1801 
1802 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1803 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1804 		return -EINVAL;
1805 
1806 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
1807 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1808 	if (port_no >= DP_MAX_PORTS)
1809 		return -EFBIG;
1810 
1811 	reply = ovs_vport_cmd_alloc_info();
1812 	if (!reply)
1813 		return -ENOMEM;
1814 
1815 	ovs_lock();
1816 restart:
1817 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1818 	err = -ENODEV;
1819 	if (!dp)
1820 		goto exit_unlock_free;
1821 
1822 	if (port_no) {
1823 		vport = ovs_vport_ovsl(dp, port_no);
1824 		err = -EBUSY;
1825 		if (vport)
1826 			goto exit_unlock_free;
1827 	} else {
1828 		for (port_no = 1; ; port_no++) {
1829 			if (port_no >= DP_MAX_PORTS) {
1830 				err = -EFBIG;
1831 				goto exit_unlock_free;
1832 			}
1833 			vport = ovs_vport_ovsl(dp, port_no);
1834 			if (!vport)
1835 				break;
1836 		}
1837 	}
1838 
1839 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1840 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1841 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1842 	parms.dp = dp;
1843 	parms.port_no = port_no;
1844 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1845 
1846 	vport = new_vport(&parms);
1847 	err = PTR_ERR(vport);
1848 	if (IS_ERR(vport)) {
1849 		if (err == -EAGAIN)
1850 			goto restart;
1851 		goto exit_unlock_free;
1852 	}
1853 
1854 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1855 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1856 	BUG_ON(err < 0);
1857 	ovs_unlock();
1858 
1859 	ovs_notify(&dp_vport_genl_family, reply, info);
1860 	return 0;
1861 
1862 exit_unlock_free:
1863 	ovs_unlock();
1864 	kfree_skb(reply);
1865 	return err;
1866 }
1867 
1868 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1869 {
1870 	struct nlattr **a = info->attrs;
1871 	struct sk_buff *reply;
1872 	struct vport *vport;
1873 	int err;
1874 
1875 	reply = ovs_vport_cmd_alloc_info();
1876 	if (!reply)
1877 		return -ENOMEM;
1878 
1879 	ovs_lock();
1880 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1881 	err = PTR_ERR(vport);
1882 	if (IS_ERR(vport))
1883 		goto exit_unlock_free;
1884 
1885 	if (a[OVS_VPORT_ATTR_TYPE] &&
1886 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1887 		err = -EINVAL;
1888 		goto exit_unlock_free;
1889 	}
1890 
1891 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
1892 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1893 		if (err)
1894 			goto exit_unlock_free;
1895 	}
1896 
1897 
1898 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1899 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1900 
1901 		err = ovs_vport_set_upcall_portids(vport, ids);
1902 		if (err)
1903 			goto exit_unlock_free;
1904 	}
1905 
1906 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1907 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1908 	BUG_ON(err < 0);
1909 
1910 	ovs_unlock();
1911 	ovs_notify(&dp_vport_genl_family, reply, info);
1912 	return 0;
1913 
1914 exit_unlock_free:
1915 	ovs_unlock();
1916 	kfree_skb(reply);
1917 	return err;
1918 }
1919 
1920 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1921 {
1922 	struct nlattr **a = info->attrs;
1923 	struct sk_buff *reply;
1924 	struct vport *vport;
1925 	int err;
1926 
1927 	reply = ovs_vport_cmd_alloc_info();
1928 	if (!reply)
1929 		return -ENOMEM;
1930 
1931 	ovs_lock();
1932 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1933 	err = PTR_ERR(vport);
1934 	if (IS_ERR(vport))
1935 		goto exit_unlock_free;
1936 
1937 	if (vport->port_no == OVSP_LOCAL) {
1938 		err = -EINVAL;
1939 		goto exit_unlock_free;
1940 	}
1941 
1942 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1943 				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1944 	BUG_ON(err < 0);
1945 	ovs_dp_detach_port(vport);
1946 	ovs_unlock();
1947 
1948 	ovs_notify(&dp_vport_genl_family, reply, info);
1949 	return 0;
1950 
1951 exit_unlock_free:
1952 	ovs_unlock();
1953 	kfree_skb(reply);
1954 	return err;
1955 }
1956 
1957 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1958 {
1959 	struct nlattr **a = info->attrs;
1960 	struct ovs_header *ovs_header = info->userhdr;
1961 	struct sk_buff *reply;
1962 	struct vport *vport;
1963 	int err;
1964 
1965 	reply = ovs_vport_cmd_alloc_info();
1966 	if (!reply)
1967 		return -ENOMEM;
1968 
1969 	rcu_read_lock();
1970 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1971 	err = PTR_ERR(vport);
1972 	if (IS_ERR(vport))
1973 		goto exit_unlock_free;
1974 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1975 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1976 	BUG_ON(err < 0);
1977 	rcu_read_unlock();
1978 
1979 	return genlmsg_reply(reply, info);
1980 
1981 exit_unlock_free:
1982 	rcu_read_unlock();
1983 	kfree_skb(reply);
1984 	return err;
1985 }
1986 
1987 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1988 {
1989 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1990 	struct datapath *dp;
1991 	int bucket = cb->args[0], skip = cb->args[1];
1992 	int i, j = 0;
1993 
1994 	rcu_read_lock();
1995 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1996 	if (!dp) {
1997 		rcu_read_unlock();
1998 		return -ENODEV;
1999 	}
2000 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2001 		struct vport *vport;
2002 
2003 		j = 0;
2004 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2005 			if (j >= skip &&
2006 			    ovs_vport_cmd_fill_info(vport, skb,
2007 						    NETLINK_CB(cb->skb).portid,
2008 						    cb->nlh->nlmsg_seq,
2009 						    NLM_F_MULTI,
2010 						    OVS_VPORT_CMD_NEW) < 0)
2011 				goto out;
2012 
2013 			j++;
2014 		}
2015 		skip = 0;
2016 	}
2017 out:
2018 	rcu_read_unlock();
2019 
2020 	cb->args[0] = i;
2021 	cb->args[1] = j;
2022 
2023 	return skb->len;
2024 }
2025 
2026 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2027 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2028 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2029 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2030 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2031 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2032 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2033 };
2034 
2035 static const struct genl_ops dp_vport_genl_ops[] = {
2036 	{ .cmd = OVS_VPORT_CMD_NEW,
2037 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2038 	  .policy = vport_policy,
2039 	  .doit = ovs_vport_cmd_new
2040 	},
2041 	{ .cmd = OVS_VPORT_CMD_DEL,
2042 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2043 	  .policy = vport_policy,
2044 	  .doit = ovs_vport_cmd_del
2045 	},
2046 	{ .cmd = OVS_VPORT_CMD_GET,
2047 	  .flags = 0,		    /* OK for unprivileged users. */
2048 	  .policy = vport_policy,
2049 	  .doit = ovs_vport_cmd_get,
2050 	  .dumpit = ovs_vport_cmd_dump
2051 	},
2052 	{ .cmd = OVS_VPORT_CMD_SET,
2053 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2054 	  .policy = vport_policy,
2055 	  .doit = ovs_vport_cmd_set,
2056 	},
2057 };
2058 
2059 struct genl_family dp_vport_genl_family = {
2060 	.id = GENL_ID_GENERATE,
2061 	.hdrsize = sizeof(struct ovs_header),
2062 	.name = OVS_VPORT_FAMILY,
2063 	.version = OVS_VPORT_VERSION,
2064 	.maxattr = OVS_VPORT_ATTR_MAX,
2065 	.netnsok = true,
2066 	.parallel_ops = true,
2067 	.ops = dp_vport_genl_ops,
2068 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2069 	.mcgrps = &ovs_dp_vport_multicast_group,
2070 	.n_mcgrps = 1,
2071 };
2072 
2073 static struct genl_family * const dp_genl_families[] = {
2074 	&dp_datapath_genl_family,
2075 	&dp_vport_genl_family,
2076 	&dp_flow_genl_family,
2077 	&dp_packet_genl_family,
2078 };
2079 
2080 static void dp_unregister_genl(int n_families)
2081 {
2082 	int i;
2083 
2084 	for (i = 0; i < n_families; i++)
2085 		genl_unregister_family(dp_genl_families[i]);
2086 }
2087 
2088 static int dp_register_genl(void)
2089 {
2090 	int err;
2091 	int i;
2092 
2093 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2094 
2095 		err = genl_register_family(dp_genl_families[i]);
2096 		if (err)
2097 			goto error;
2098 	}
2099 
2100 	return 0;
2101 
2102 error:
2103 	dp_unregister_genl(i);
2104 	return err;
2105 }
2106 
2107 static int __net_init ovs_init_net(struct net *net)
2108 {
2109 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2110 
2111 	INIT_LIST_HEAD(&ovs_net->dps);
2112 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2113 	return 0;
2114 }
2115 
2116 static void __net_exit ovs_exit_net(struct net *net)
2117 {
2118 	struct datapath *dp, *dp_next;
2119 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2120 
2121 	ovs_lock();
2122 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2123 		__dp_destroy(dp);
2124 	ovs_unlock();
2125 
2126 	cancel_work_sync(&ovs_net->dp_notify_work);
2127 }
2128 
2129 static struct pernet_operations ovs_net_ops = {
2130 	.init = ovs_init_net,
2131 	.exit = ovs_exit_net,
2132 	.id   = &ovs_net_id,
2133 	.size = sizeof(struct ovs_net),
2134 };
2135 
2136 static int __init dp_init(void)
2137 {
2138 	int err;
2139 
2140 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2141 
2142 	pr_info("Open vSwitch switching datapath\n");
2143 
2144 	err = action_fifos_init();
2145 	if (err)
2146 		goto error;
2147 
2148 	err = ovs_internal_dev_rtnl_link_register();
2149 	if (err)
2150 		goto error_action_fifos_exit;
2151 
2152 	err = ovs_flow_init();
2153 	if (err)
2154 		goto error_unreg_rtnl_link;
2155 
2156 	err = ovs_vport_init();
2157 	if (err)
2158 		goto error_flow_exit;
2159 
2160 	err = register_pernet_device(&ovs_net_ops);
2161 	if (err)
2162 		goto error_vport_exit;
2163 
2164 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2165 	if (err)
2166 		goto error_netns_exit;
2167 
2168 	err = ovs_netdev_init();
2169 	if (err)
2170 		goto error_unreg_notifier;
2171 
2172 	err = dp_register_genl();
2173 	if (err < 0)
2174 		goto error_unreg_netdev;
2175 
2176 	return 0;
2177 
2178 error_unreg_netdev:
2179 	ovs_netdev_exit();
2180 error_unreg_notifier:
2181 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2182 error_netns_exit:
2183 	unregister_pernet_device(&ovs_net_ops);
2184 error_vport_exit:
2185 	ovs_vport_exit();
2186 error_flow_exit:
2187 	ovs_flow_exit();
2188 error_unreg_rtnl_link:
2189 	ovs_internal_dev_rtnl_link_unregister();
2190 error_action_fifos_exit:
2191 	action_fifos_exit();
2192 error:
2193 	return err;
2194 }
2195 
2196 static void dp_cleanup(void)
2197 {
2198 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2199 	ovs_netdev_exit();
2200 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2201 	unregister_pernet_device(&ovs_net_ops);
2202 	rcu_barrier();
2203 	ovs_vport_exit();
2204 	ovs_flow_exit();
2205 	ovs_internal_dev_rtnl_link_unregister();
2206 	action_fifos_exit();
2207 }
2208 
2209 module_init(dp_init);
2210 module_exit(dp_cleanup);
2211 
2212 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2213 MODULE_LICENSE("GPL");
2214