xref: /linux/net/openvswitch/datapath.c (revision e5c5d22e8dcf7c2d430336cbf8e180bd38e8daf1)
1 /*
2  * Copyright (c) 2007-2012 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 <linux/workqueue.h>
51 #include <net/genetlink.h>
52 #include <net/net_namespace.h>
53 #include <net/netns/generic.h>
54 
55 #include "datapath.h"
56 #include "flow.h"
57 #include "vport-internal_dev.h"
58 
59 /**
60  * struct ovs_net - Per net-namespace data for ovs.
61  * @dps: List of datapaths to enable dumping them all out.
62  * Protected by genl_mutex.
63  */
64 struct ovs_net {
65 	struct list_head dps;
66 };
67 
68 static int ovs_net_id __read_mostly;
69 
70 #define REHASH_FLOW_INTERVAL (10 * 60 * HZ)
71 static void rehash_flow_table(struct work_struct *work);
72 static DECLARE_DELAYED_WORK(rehash_flow_wq, rehash_flow_table);
73 
74 /**
75  * DOC: Locking:
76  *
77  * Writes to device state (add/remove datapath, port, set operations on vports,
78  * etc.) are protected by RTNL.
79  *
80  * Writes to other state (flow table modifications, set miscellaneous datapath
81  * parameters, etc.) are protected by genl_mutex.  The RTNL lock nests inside
82  * genl_mutex.
83  *
84  * Reads are protected by RCU.
85  *
86  * There are a few special cases (mostly stats) that have their own
87  * synchronization but they nest under all of above and don't interact with
88  * each other.
89  */
90 
91 static struct vport *new_vport(const struct vport_parms *);
92 static int queue_gso_packets(struct net *, int dp_ifindex, struct sk_buff *,
93 			     const struct dp_upcall_info *);
94 static int queue_userspace_packet(struct net *, int dp_ifindex,
95 				  struct sk_buff *,
96 				  const struct dp_upcall_info *);
97 
98 /* Must be called with rcu_read_lock, genl_mutex, or RTNL lock. */
99 static struct datapath *get_dp(struct net *net, int dp_ifindex)
100 {
101 	struct datapath *dp = NULL;
102 	struct net_device *dev;
103 
104 	rcu_read_lock();
105 	dev = dev_get_by_index_rcu(net, dp_ifindex);
106 	if (dev) {
107 		struct vport *vport = ovs_internal_dev_get_vport(dev);
108 		if (vport)
109 			dp = vport->dp;
110 	}
111 	rcu_read_unlock();
112 
113 	return dp;
114 }
115 
116 /* Must be called with rcu_read_lock or RTNL lock. */
117 const char *ovs_dp_name(const struct datapath *dp)
118 {
119 	struct vport *vport = ovs_vport_rtnl_rcu(dp, OVSP_LOCAL);
120 	return vport->ops->get_name(vport);
121 }
122 
123 static int get_dpifindex(struct datapath *dp)
124 {
125 	struct vport *local;
126 	int ifindex;
127 
128 	rcu_read_lock();
129 
130 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
131 	if (local)
132 		ifindex = local->ops->get_ifindex(local);
133 	else
134 		ifindex = 0;
135 
136 	rcu_read_unlock();
137 
138 	return ifindex;
139 }
140 
141 static void destroy_dp_rcu(struct rcu_head *rcu)
142 {
143 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
144 
145 	ovs_flow_tbl_destroy((__force struct flow_table *)dp->table);
146 	free_percpu(dp->stats_percpu);
147 	release_net(ovs_dp_get_net(dp));
148 	kfree(dp->ports);
149 	kfree(dp);
150 }
151 
152 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
153 					    u16 port_no)
154 {
155 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
156 }
157 
158 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
159 {
160 	struct vport *vport;
161 	struct hlist_head *head;
162 
163 	head = vport_hash_bucket(dp, port_no);
164 	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
165 		if (vport->port_no == port_no)
166 			return vport;
167 	}
168 	return NULL;
169 }
170 
171 /* Called with RTNL lock and genl_lock. */
172 static struct vport *new_vport(const struct vport_parms *parms)
173 {
174 	struct vport *vport;
175 
176 	vport = ovs_vport_add(parms);
177 	if (!IS_ERR(vport)) {
178 		struct datapath *dp = parms->dp;
179 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
180 
181 		hlist_add_head_rcu(&vport->dp_hash_node, head);
182 	}
183 
184 	return vport;
185 }
186 
187 /* Called with RTNL lock. */
188 void ovs_dp_detach_port(struct vport *p)
189 {
190 	ASSERT_RTNL();
191 
192 	/* First drop references to device. */
193 	hlist_del_rcu(&p->dp_hash_node);
194 
195 	/* Then destroy it. */
196 	ovs_vport_del(p);
197 }
198 
199 /* Must be called with rcu_read_lock. */
200 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
201 {
202 	struct datapath *dp = p->dp;
203 	struct sw_flow *flow;
204 	struct dp_stats_percpu *stats;
205 	struct sw_flow_key key;
206 	u64 *stats_counter;
207 	int error;
208 	int key_len;
209 
210 	stats = this_cpu_ptr(dp->stats_percpu);
211 
212 	/* Extract flow from 'skb' into 'key'. */
213 	error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
214 	if (unlikely(error)) {
215 		kfree_skb(skb);
216 		return;
217 	}
218 
219 	/* Look up flow. */
220 	flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
221 	if (unlikely(!flow)) {
222 		struct dp_upcall_info upcall;
223 
224 		upcall.cmd = OVS_PACKET_CMD_MISS;
225 		upcall.key = &key;
226 		upcall.userdata = NULL;
227 		upcall.portid = p->upcall_portid;
228 		ovs_dp_upcall(dp, skb, &upcall);
229 		consume_skb(skb);
230 		stats_counter = &stats->n_missed;
231 		goto out;
232 	}
233 
234 	OVS_CB(skb)->flow = flow;
235 
236 	stats_counter = &stats->n_hit;
237 	ovs_flow_used(OVS_CB(skb)->flow, skb);
238 	ovs_execute_actions(dp, skb);
239 
240 out:
241 	/* Update datapath statistics. */
242 	u64_stats_update_begin(&stats->sync);
243 	(*stats_counter)++;
244 	u64_stats_update_end(&stats->sync);
245 }
246 
247 static struct genl_family dp_packet_genl_family = {
248 	.id = GENL_ID_GENERATE,
249 	.hdrsize = sizeof(struct ovs_header),
250 	.name = OVS_PACKET_FAMILY,
251 	.version = OVS_PACKET_VERSION,
252 	.maxattr = OVS_PACKET_ATTR_MAX,
253 	.netnsok = true
254 };
255 
256 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
257 		  const struct dp_upcall_info *upcall_info)
258 {
259 	struct dp_stats_percpu *stats;
260 	int dp_ifindex;
261 	int err;
262 
263 	if (upcall_info->portid == 0) {
264 		err = -ENOTCONN;
265 		goto err;
266 	}
267 
268 	dp_ifindex = get_dpifindex(dp);
269 	if (!dp_ifindex) {
270 		err = -ENODEV;
271 		goto err;
272 	}
273 
274 	if (!skb_is_gso(skb))
275 		err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
276 	else
277 		err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
278 	if (err)
279 		goto err;
280 
281 	return 0;
282 
283 err:
284 	stats = this_cpu_ptr(dp->stats_percpu);
285 
286 	u64_stats_update_begin(&stats->sync);
287 	stats->n_lost++;
288 	u64_stats_update_end(&stats->sync);
289 
290 	return err;
291 }
292 
293 static int queue_gso_packets(struct net *net, int dp_ifindex,
294 			     struct sk_buff *skb,
295 			     const struct dp_upcall_info *upcall_info)
296 {
297 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
298 	struct dp_upcall_info later_info;
299 	struct sw_flow_key later_key;
300 	struct sk_buff *segs, *nskb;
301 	int err;
302 
303 	segs = __skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM, false);
304 	if (IS_ERR(segs))
305 		return PTR_ERR(segs);
306 
307 	/* Queue all of the segments. */
308 	skb = segs;
309 	do {
310 		err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
311 		if (err)
312 			break;
313 
314 		if (skb == segs && gso_type & SKB_GSO_UDP) {
315 			/* The initial flow key extracted by ovs_flow_extract()
316 			 * in this case is for a first fragment, so we need to
317 			 * properly mark later fragments.
318 			 */
319 			later_key = *upcall_info->key;
320 			later_key.ip.frag = OVS_FRAG_TYPE_LATER;
321 
322 			later_info = *upcall_info;
323 			later_info.key = &later_key;
324 			upcall_info = &later_info;
325 		}
326 	} while ((skb = skb->next));
327 
328 	/* Free all of the segments. */
329 	skb = segs;
330 	do {
331 		nskb = skb->next;
332 		if (err)
333 			kfree_skb(skb);
334 		else
335 			consume_skb(skb);
336 	} while ((skb = nskb));
337 	return err;
338 }
339 
340 static int queue_userspace_packet(struct net *net, int dp_ifindex,
341 				  struct sk_buff *skb,
342 				  const struct dp_upcall_info *upcall_info)
343 {
344 	struct ovs_header *upcall;
345 	struct sk_buff *nskb = NULL;
346 	struct sk_buff *user_skb; /* to be queued to userspace */
347 	struct nlattr *nla;
348 	unsigned int len;
349 	int err;
350 
351 	if (vlan_tx_tag_present(skb)) {
352 		nskb = skb_clone(skb, GFP_ATOMIC);
353 		if (!nskb)
354 			return -ENOMEM;
355 
356 		nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
357 		if (!nskb)
358 			return -ENOMEM;
359 
360 		nskb->vlan_tci = 0;
361 		skb = nskb;
362 	}
363 
364 	if (nla_attr_size(skb->len) > USHRT_MAX) {
365 		err = -EFBIG;
366 		goto out;
367 	}
368 
369 	len = sizeof(struct ovs_header);
370 	len += nla_total_size(skb->len);
371 	len += nla_total_size(FLOW_BUFSIZE);
372 	if (upcall_info->userdata)
373 		len += NLA_ALIGN(upcall_info->userdata->nla_len);
374 
375 	user_skb = genlmsg_new(len, GFP_ATOMIC);
376 	if (!user_skb) {
377 		err = -ENOMEM;
378 		goto out;
379 	}
380 
381 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
382 			     0, upcall_info->cmd);
383 	upcall->dp_ifindex = dp_ifindex;
384 
385 	nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
386 	ovs_flow_to_nlattrs(upcall_info->key, user_skb);
387 	nla_nest_end(user_skb, nla);
388 
389 	if (upcall_info->userdata)
390 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
391 			  nla_len(upcall_info->userdata),
392 			  nla_data(upcall_info->userdata));
393 
394 	nla = __nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, skb->len);
395 
396 	skb_copy_and_csum_dev(skb, nla_data(nla));
397 
398 	genlmsg_end(user_skb, upcall);
399 	err = genlmsg_unicast(net, user_skb, upcall_info->portid);
400 
401 out:
402 	kfree_skb(nskb);
403 	return err;
404 }
405 
406 /* Called with genl_mutex. */
407 static int flush_flows(struct datapath *dp)
408 {
409 	struct flow_table *old_table;
410 	struct flow_table *new_table;
411 
412 	old_table = genl_dereference(dp->table);
413 	new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
414 	if (!new_table)
415 		return -ENOMEM;
416 
417 	rcu_assign_pointer(dp->table, new_table);
418 
419 	ovs_flow_tbl_deferred_destroy(old_table);
420 	return 0;
421 }
422 
423 static int validate_actions(const struct nlattr *attr,
424 				const struct sw_flow_key *key, int depth);
425 
426 static int validate_sample(const struct nlattr *attr,
427 				const struct sw_flow_key *key, int depth)
428 {
429 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
430 	const struct nlattr *probability, *actions;
431 	const struct nlattr *a;
432 	int rem;
433 
434 	memset(attrs, 0, sizeof(attrs));
435 	nla_for_each_nested(a, attr, rem) {
436 		int type = nla_type(a);
437 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
438 			return -EINVAL;
439 		attrs[type] = a;
440 	}
441 	if (rem)
442 		return -EINVAL;
443 
444 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
445 	if (!probability || nla_len(probability) != sizeof(u32))
446 		return -EINVAL;
447 
448 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
449 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
450 		return -EINVAL;
451 	return validate_actions(actions, key, depth + 1);
452 }
453 
454 static int validate_tp_port(const struct sw_flow_key *flow_key)
455 {
456 	if (flow_key->eth.type == htons(ETH_P_IP)) {
457 		if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
458 			return 0;
459 	} else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
460 		if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
461 			return 0;
462 	}
463 
464 	return -EINVAL;
465 }
466 
467 static int validate_set(const struct nlattr *a,
468 			const struct sw_flow_key *flow_key)
469 {
470 	const struct nlattr *ovs_key = nla_data(a);
471 	int key_type = nla_type(ovs_key);
472 
473 	/* There can be only one key in a action */
474 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
475 		return -EINVAL;
476 
477 	if (key_type > OVS_KEY_ATTR_MAX ||
478 	    nla_len(ovs_key) != ovs_key_lens[key_type])
479 		return -EINVAL;
480 
481 	switch (key_type) {
482 	const struct ovs_key_ipv4 *ipv4_key;
483 	const struct ovs_key_ipv6 *ipv6_key;
484 
485 	case OVS_KEY_ATTR_PRIORITY:
486 	case OVS_KEY_ATTR_SKB_MARK:
487 	case OVS_KEY_ATTR_ETHERNET:
488 		break;
489 
490 	case OVS_KEY_ATTR_IPV4:
491 		if (flow_key->eth.type != htons(ETH_P_IP))
492 			return -EINVAL;
493 
494 		if (!flow_key->ip.proto)
495 			return -EINVAL;
496 
497 		ipv4_key = nla_data(ovs_key);
498 		if (ipv4_key->ipv4_proto != flow_key->ip.proto)
499 			return -EINVAL;
500 
501 		if (ipv4_key->ipv4_frag != flow_key->ip.frag)
502 			return -EINVAL;
503 
504 		break;
505 
506 	case OVS_KEY_ATTR_IPV6:
507 		if (flow_key->eth.type != htons(ETH_P_IPV6))
508 			return -EINVAL;
509 
510 		if (!flow_key->ip.proto)
511 			return -EINVAL;
512 
513 		ipv6_key = nla_data(ovs_key);
514 		if (ipv6_key->ipv6_proto != flow_key->ip.proto)
515 			return -EINVAL;
516 
517 		if (ipv6_key->ipv6_frag != flow_key->ip.frag)
518 			return -EINVAL;
519 
520 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
521 			return -EINVAL;
522 
523 		break;
524 
525 	case OVS_KEY_ATTR_TCP:
526 		if (flow_key->ip.proto != IPPROTO_TCP)
527 			return -EINVAL;
528 
529 		return validate_tp_port(flow_key);
530 
531 	case OVS_KEY_ATTR_UDP:
532 		if (flow_key->ip.proto != IPPROTO_UDP)
533 			return -EINVAL;
534 
535 		return validate_tp_port(flow_key);
536 
537 	default:
538 		return -EINVAL;
539 	}
540 
541 	return 0;
542 }
543 
544 static int validate_userspace(const struct nlattr *attr)
545 {
546 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =	{
547 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
548 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_UNSPEC },
549 	};
550 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
551 	int error;
552 
553 	error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
554 				 attr, userspace_policy);
555 	if (error)
556 		return error;
557 
558 	if (!a[OVS_USERSPACE_ATTR_PID] ||
559 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
560 		return -EINVAL;
561 
562 	return 0;
563 }
564 
565 static int validate_actions(const struct nlattr *attr,
566 				const struct sw_flow_key *key,  int depth)
567 {
568 	const struct nlattr *a;
569 	int rem, err;
570 
571 	if (depth >= SAMPLE_ACTION_DEPTH)
572 		return -EOVERFLOW;
573 
574 	nla_for_each_nested(a, attr, rem) {
575 		/* Expected argument lengths, (u32)-1 for variable length. */
576 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
577 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
578 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
579 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
580 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
581 			[OVS_ACTION_ATTR_SET] = (u32)-1,
582 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1
583 		};
584 		const struct ovs_action_push_vlan *vlan;
585 		int type = nla_type(a);
586 
587 		if (type > OVS_ACTION_ATTR_MAX ||
588 		    (action_lens[type] != nla_len(a) &&
589 		     action_lens[type] != (u32)-1))
590 			return -EINVAL;
591 
592 		switch (type) {
593 		case OVS_ACTION_ATTR_UNSPEC:
594 			return -EINVAL;
595 
596 		case OVS_ACTION_ATTR_USERSPACE:
597 			err = validate_userspace(a);
598 			if (err)
599 				return err;
600 			break;
601 
602 		case OVS_ACTION_ATTR_OUTPUT:
603 			if (nla_get_u32(a) >= DP_MAX_PORTS)
604 				return -EINVAL;
605 			break;
606 
607 
608 		case OVS_ACTION_ATTR_POP_VLAN:
609 			break;
610 
611 		case OVS_ACTION_ATTR_PUSH_VLAN:
612 			vlan = nla_data(a);
613 			if (vlan->vlan_tpid != htons(ETH_P_8021Q))
614 				return -EINVAL;
615 			if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
616 				return -EINVAL;
617 			break;
618 
619 		case OVS_ACTION_ATTR_SET:
620 			err = validate_set(a, key);
621 			if (err)
622 				return err;
623 			break;
624 
625 		case OVS_ACTION_ATTR_SAMPLE:
626 			err = validate_sample(a, key, depth);
627 			if (err)
628 				return err;
629 			break;
630 
631 		default:
632 			return -EINVAL;
633 		}
634 	}
635 
636 	if (rem > 0)
637 		return -EINVAL;
638 
639 	return 0;
640 }
641 
642 static void clear_stats(struct sw_flow *flow)
643 {
644 	flow->used = 0;
645 	flow->tcp_flags = 0;
646 	flow->packet_count = 0;
647 	flow->byte_count = 0;
648 }
649 
650 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
651 {
652 	struct ovs_header *ovs_header = info->userhdr;
653 	struct nlattr **a = info->attrs;
654 	struct sw_flow_actions *acts;
655 	struct sk_buff *packet;
656 	struct sw_flow *flow;
657 	struct datapath *dp;
658 	struct ethhdr *eth;
659 	int len;
660 	int err;
661 	int key_len;
662 
663 	err = -EINVAL;
664 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
665 	    !a[OVS_PACKET_ATTR_ACTIONS] ||
666 	    nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
667 		goto err;
668 
669 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
670 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
671 	err = -ENOMEM;
672 	if (!packet)
673 		goto err;
674 	skb_reserve(packet, NET_IP_ALIGN);
675 
676 	memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
677 
678 	skb_reset_mac_header(packet);
679 	eth = eth_hdr(packet);
680 
681 	/* Normally, setting the skb 'protocol' field would be handled by a
682 	 * call to eth_type_trans(), but it assumes there's a sending
683 	 * device, which we may not have. */
684 	if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
685 		packet->protocol = eth->h_proto;
686 	else
687 		packet->protocol = htons(ETH_P_802_2);
688 
689 	/* Build an sw_flow for sending this packet. */
690 	flow = ovs_flow_alloc();
691 	err = PTR_ERR(flow);
692 	if (IS_ERR(flow))
693 		goto err_kfree_skb;
694 
695 	err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
696 	if (err)
697 		goto err_flow_free;
698 
699 	err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
700 					     &flow->key.phy.skb_mark,
701 					     &flow->key.phy.in_port,
702 					     a[OVS_PACKET_ATTR_KEY]);
703 	if (err)
704 		goto err_flow_free;
705 
706 	err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
707 	if (err)
708 		goto err_flow_free;
709 
710 	flow->hash = ovs_flow_hash(&flow->key, key_len);
711 
712 	acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
713 	err = PTR_ERR(acts);
714 	if (IS_ERR(acts))
715 		goto err_flow_free;
716 	rcu_assign_pointer(flow->sf_acts, acts);
717 
718 	OVS_CB(packet)->flow = flow;
719 	packet->priority = flow->key.phy.priority;
720 	packet->mark = flow->key.phy.skb_mark;
721 
722 	rcu_read_lock();
723 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
724 	err = -ENODEV;
725 	if (!dp)
726 		goto err_unlock;
727 
728 	local_bh_disable();
729 	err = ovs_execute_actions(dp, packet);
730 	local_bh_enable();
731 	rcu_read_unlock();
732 
733 	ovs_flow_free(flow);
734 	return err;
735 
736 err_unlock:
737 	rcu_read_unlock();
738 err_flow_free:
739 	ovs_flow_free(flow);
740 err_kfree_skb:
741 	kfree_skb(packet);
742 err:
743 	return err;
744 }
745 
746 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
747 	[OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
748 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
749 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
750 };
751 
752 static struct genl_ops dp_packet_genl_ops[] = {
753 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
754 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
755 	  .policy = packet_policy,
756 	  .doit = ovs_packet_cmd_execute
757 	}
758 };
759 
760 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
761 {
762 	int i;
763 	struct flow_table *table = genl_dereference(dp->table);
764 
765 	stats->n_flows = ovs_flow_tbl_count(table);
766 
767 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
768 	for_each_possible_cpu(i) {
769 		const struct dp_stats_percpu *percpu_stats;
770 		struct dp_stats_percpu local_stats;
771 		unsigned int start;
772 
773 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
774 
775 		do {
776 			start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
777 			local_stats = *percpu_stats;
778 		} while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
779 
780 		stats->n_hit += local_stats.n_hit;
781 		stats->n_missed += local_stats.n_missed;
782 		stats->n_lost += local_stats.n_lost;
783 	}
784 }
785 
786 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
787 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
788 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
789 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
790 };
791 
792 static struct genl_family dp_flow_genl_family = {
793 	.id = GENL_ID_GENERATE,
794 	.hdrsize = sizeof(struct ovs_header),
795 	.name = OVS_FLOW_FAMILY,
796 	.version = OVS_FLOW_VERSION,
797 	.maxattr = OVS_FLOW_ATTR_MAX,
798 	.netnsok = true
799 };
800 
801 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
802 	.name = OVS_FLOW_MCGROUP
803 };
804 
805 /* Called with genl_lock. */
806 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
807 				  struct sk_buff *skb, u32 portid,
808 				  u32 seq, u32 flags, u8 cmd)
809 {
810 	const int skb_orig_len = skb->len;
811 	const struct sw_flow_actions *sf_acts;
812 	struct ovs_flow_stats stats;
813 	struct ovs_header *ovs_header;
814 	struct nlattr *nla;
815 	unsigned long used;
816 	u8 tcp_flags;
817 	int err;
818 
819 	sf_acts = rcu_dereference_protected(flow->sf_acts,
820 					    lockdep_genl_is_held());
821 
822 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
823 	if (!ovs_header)
824 		return -EMSGSIZE;
825 
826 	ovs_header->dp_ifindex = get_dpifindex(dp);
827 
828 	nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
829 	if (!nla)
830 		goto nla_put_failure;
831 	err = ovs_flow_to_nlattrs(&flow->key, skb);
832 	if (err)
833 		goto error;
834 	nla_nest_end(skb, nla);
835 
836 	spin_lock_bh(&flow->lock);
837 	used = flow->used;
838 	stats.n_packets = flow->packet_count;
839 	stats.n_bytes = flow->byte_count;
840 	tcp_flags = flow->tcp_flags;
841 	spin_unlock_bh(&flow->lock);
842 
843 	if (used &&
844 	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
845 		goto nla_put_failure;
846 
847 	if (stats.n_packets &&
848 	    nla_put(skb, OVS_FLOW_ATTR_STATS,
849 		    sizeof(struct ovs_flow_stats), &stats))
850 		goto nla_put_failure;
851 
852 	if (tcp_flags &&
853 	    nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
854 		goto nla_put_failure;
855 
856 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
857 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
858 	 * Netlink but individual action lists can be longer than
859 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
860 	 * The userspace caller can always fetch the actions separately if it
861 	 * really wants them.  (Most userspace callers in fact don't care.)
862 	 *
863 	 * This can only fail for dump operations because the skb is always
864 	 * properly sized for single flows.
865 	 */
866 	err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
867 		      sf_acts->actions);
868 	if (err < 0 && skb_orig_len)
869 		goto error;
870 
871 	return genlmsg_end(skb, ovs_header);
872 
873 nla_put_failure:
874 	err = -EMSGSIZE;
875 error:
876 	genlmsg_cancel(skb, ovs_header);
877 	return err;
878 }
879 
880 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
881 {
882 	const struct sw_flow_actions *sf_acts;
883 	int len;
884 
885 	sf_acts = rcu_dereference_protected(flow->sf_acts,
886 					    lockdep_genl_is_held());
887 
888 	/* OVS_FLOW_ATTR_KEY */
889 	len = nla_total_size(FLOW_BUFSIZE);
890 	/* OVS_FLOW_ATTR_ACTIONS */
891 	len += nla_total_size(sf_acts->actions_len);
892 	/* OVS_FLOW_ATTR_STATS */
893 	len += nla_total_size(sizeof(struct ovs_flow_stats));
894 	/* OVS_FLOW_ATTR_TCP_FLAGS */
895 	len += nla_total_size(1);
896 	/* OVS_FLOW_ATTR_USED */
897 	len += nla_total_size(8);
898 
899 	len += NLMSG_ALIGN(sizeof(struct ovs_header));
900 
901 	return genlmsg_new(len, GFP_KERNEL);
902 }
903 
904 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
905 					       struct datapath *dp,
906 					       u32 portid, u32 seq, u8 cmd)
907 {
908 	struct sk_buff *skb;
909 	int retval;
910 
911 	skb = ovs_flow_cmd_alloc_info(flow);
912 	if (!skb)
913 		return ERR_PTR(-ENOMEM);
914 
915 	retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
916 	BUG_ON(retval < 0);
917 	return skb;
918 }
919 
920 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
921 {
922 	struct nlattr **a = info->attrs;
923 	struct ovs_header *ovs_header = info->userhdr;
924 	struct sw_flow_key key;
925 	struct sw_flow *flow;
926 	struct sk_buff *reply;
927 	struct datapath *dp;
928 	struct flow_table *table;
929 	int error;
930 	int key_len;
931 
932 	/* Extract key. */
933 	error = -EINVAL;
934 	if (!a[OVS_FLOW_ATTR_KEY])
935 		goto error;
936 	error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
937 	if (error)
938 		goto error;
939 
940 	/* Validate actions. */
941 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
942 		error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
943 		if (error)
944 			goto error;
945 	} else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
946 		error = -EINVAL;
947 		goto error;
948 	}
949 
950 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
951 	error = -ENODEV;
952 	if (!dp)
953 		goto error;
954 
955 	table = genl_dereference(dp->table);
956 	flow = ovs_flow_tbl_lookup(table, &key, key_len);
957 	if (!flow) {
958 		struct sw_flow_actions *acts;
959 
960 		/* Bail out if we're not allowed to create a new flow. */
961 		error = -ENOENT;
962 		if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
963 			goto error;
964 
965 		/* Expand table, if necessary, to make room. */
966 		if (ovs_flow_tbl_need_to_expand(table)) {
967 			struct flow_table *new_table;
968 
969 			new_table = ovs_flow_tbl_expand(table);
970 			if (!IS_ERR(new_table)) {
971 				rcu_assign_pointer(dp->table, new_table);
972 				ovs_flow_tbl_deferred_destroy(table);
973 				table = genl_dereference(dp->table);
974 			}
975 		}
976 
977 		/* Allocate flow. */
978 		flow = ovs_flow_alloc();
979 		if (IS_ERR(flow)) {
980 			error = PTR_ERR(flow);
981 			goto error;
982 		}
983 		flow->key = key;
984 		clear_stats(flow);
985 
986 		/* Obtain actions. */
987 		acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
988 		error = PTR_ERR(acts);
989 		if (IS_ERR(acts))
990 			goto error_free_flow;
991 		rcu_assign_pointer(flow->sf_acts, acts);
992 
993 		/* Put flow in bucket. */
994 		flow->hash = ovs_flow_hash(&key, key_len);
995 		ovs_flow_tbl_insert(table, flow);
996 
997 		reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
998 						info->snd_seq,
999 						OVS_FLOW_CMD_NEW);
1000 	} else {
1001 		/* We found a matching flow. */
1002 		struct sw_flow_actions *old_acts;
1003 		struct nlattr *acts_attrs;
1004 
1005 		/* Bail out if we're not allowed to modify an existing flow.
1006 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1007 		 * because Generic Netlink treats the latter as a dump
1008 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1009 		 * gets fixed.
1010 		 */
1011 		error = -EEXIST;
1012 		if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1013 		    info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1014 			goto error;
1015 
1016 		/* Update actions. */
1017 		old_acts = rcu_dereference_protected(flow->sf_acts,
1018 						     lockdep_genl_is_held());
1019 		acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1020 		if (acts_attrs &&
1021 		   (old_acts->actions_len != nla_len(acts_attrs) ||
1022 		   memcmp(old_acts->actions, nla_data(acts_attrs),
1023 			  old_acts->actions_len))) {
1024 			struct sw_flow_actions *new_acts;
1025 
1026 			new_acts = ovs_flow_actions_alloc(acts_attrs);
1027 			error = PTR_ERR(new_acts);
1028 			if (IS_ERR(new_acts))
1029 				goto error;
1030 
1031 			rcu_assign_pointer(flow->sf_acts, new_acts);
1032 			ovs_flow_deferred_free_acts(old_acts);
1033 		}
1034 
1035 		reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1036 					       info->snd_seq, OVS_FLOW_CMD_NEW);
1037 
1038 		/* Clear stats. */
1039 		if (a[OVS_FLOW_ATTR_CLEAR]) {
1040 			spin_lock_bh(&flow->lock);
1041 			clear_stats(flow);
1042 			spin_unlock_bh(&flow->lock);
1043 		}
1044 	}
1045 
1046 	if (!IS_ERR(reply))
1047 		genl_notify(reply, genl_info_net(info), info->snd_portid,
1048 			   ovs_dp_flow_multicast_group.id, info->nlhdr,
1049 			   GFP_KERNEL);
1050 	else
1051 		netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1052 				ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1053 	return 0;
1054 
1055 error_free_flow:
1056 	ovs_flow_free(flow);
1057 error:
1058 	return error;
1059 }
1060 
1061 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1062 {
1063 	struct nlattr **a = info->attrs;
1064 	struct ovs_header *ovs_header = info->userhdr;
1065 	struct sw_flow_key key;
1066 	struct sk_buff *reply;
1067 	struct sw_flow *flow;
1068 	struct datapath *dp;
1069 	struct flow_table *table;
1070 	int err;
1071 	int key_len;
1072 
1073 	if (!a[OVS_FLOW_ATTR_KEY])
1074 		return -EINVAL;
1075 	err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1076 	if (err)
1077 		return err;
1078 
1079 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1080 	if (!dp)
1081 		return -ENODEV;
1082 
1083 	table = genl_dereference(dp->table);
1084 	flow = ovs_flow_tbl_lookup(table, &key, key_len);
1085 	if (!flow)
1086 		return -ENOENT;
1087 
1088 	reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1089 					info->snd_seq, OVS_FLOW_CMD_NEW);
1090 	if (IS_ERR(reply))
1091 		return PTR_ERR(reply);
1092 
1093 	return genlmsg_reply(reply, info);
1094 }
1095 
1096 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1097 {
1098 	struct nlattr **a = info->attrs;
1099 	struct ovs_header *ovs_header = info->userhdr;
1100 	struct sw_flow_key key;
1101 	struct sk_buff *reply;
1102 	struct sw_flow *flow;
1103 	struct datapath *dp;
1104 	struct flow_table *table;
1105 	int err;
1106 	int key_len;
1107 
1108 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1109 	if (!dp)
1110 		return -ENODEV;
1111 
1112 	if (!a[OVS_FLOW_ATTR_KEY])
1113 		return flush_flows(dp);
1114 
1115 	err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1116 	if (err)
1117 		return err;
1118 
1119 	table = genl_dereference(dp->table);
1120 	flow = ovs_flow_tbl_lookup(table, &key, key_len);
1121 	if (!flow)
1122 		return -ENOENT;
1123 
1124 	reply = ovs_flow_cmd_alloc_info(flow);
1125 	if (!reply)
1126 		return -ENOMEM;
1127 
1128 	ovs_flow_tbl_remove(table, flow);
1129 
1130 	err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1131 				     info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1132 	BUG_ON(err < 0);
1133 
1134 	ovs_flow_deferred_free(flow);
1135 
1136 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1137 		    ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1138 	return 0;
1139 }
1140 
1141 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1142 {
1143 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1144 	struct datapath *dp;
1145 	struct flow_table *table;
1146 
1147 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1148 	if (!dp)
1149 		return -ENODEV;
1150 
1151 	table = genl_dereference(dp->table);
1152 
1153 	for (;;) {
1154 		struct sw_flow *flow;
1155 		u32 bucket, obj;
1156 
1157 		bucket = cb->args[0];
1158 		obj = cb->args[1];
1159 		flow = ovs_flow_tbl_next(table, &bucket, &obj);
1160 		if (!flow)
1161 			break;
1162 
1163 		if (ovs_flow_cmd_fill_info(flow, dp, skb,
1164 					   NETLINK_CB(cb->skb).portid,
1165 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1166 					   OVS_FLOW_CMD_NEW) < 0)
1167 			break;
1168 
1169 		cb->args[0] = bucket;
1170 		cb->args[1] = obj;
1171 	}
1172 	return skb->len;
1173 }
1174 
1175 static struct genl_ops dp_flow_genl_ops[] = {
1176 	{ .cmd = OVS_FLOW_CMD_NEW,
1177 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1178 	  .policy = flow_policy,
1179 	  .doit = ovs_flow_cmd_new_or_set
1180 	},
1181 	{ .cmd = OVS_FLOW_CMD_DEL,
1182 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1183 	  .policy = flow_policy,
1184 	  .doit = ovs_flow_cmd_del
1185 	},
1186 	{ .cmd = OVS_FLOW_CMD_GET,
1187 	  .flags = 0,		    /* OK for unprivileged users. */
1188 	  .policy = flow_policy,
1189 	  .doit = ovs_flow_cmd_get,
1190 	  .dumpit = ovs_flow_cmd_dump
1191 	},
1192 	{ .cmd = OVS_FLOW_CMD_SET,
1193 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1194 	  .policy = flow_policy,
1195 	  .doit = ovs_flow_cmd_new_or_set,
1196 	},
1197 };
1198 
1199 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1200 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1201 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1202 };
1203 
1204 static struct genl_family dp_datapath_genl_family = {
1205 	.id = GENL_ID_GENERATE,
1206 	.hdrsize = sizeof(struct ovs_header),
1207 	.name = OVS_DATAPATH_FAMILY,
1208 	.version = OVS_DATAPATH_VERSION,
1209 	.maxattr = OVS_DP_ATTR_MAX,
1210 	.netnsok = true
1211 };
1212 
1213 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1214 	.name = OVS_DATAPATH_MCGROUP
1215 };
1216 
1217 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1218 				u32 portid, u32 seq, u32 flags, u8 cmd)
1219 {
1220 	struct ovs_header *ovs_header;
1221 	struct ovs_dp_stats dp_stats;
1222 	int err;
1223 
1224 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1225 				   flags, cmd);
1226 	if (!ovs_header)
1227 		goto error;
1228 
1229 	ovs_header->dp_ifindex = get_dpifindex(dp);
1230 
1231 	rcu_read_lock();
1232 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1233 	rcu_read_unlock();
1234 	if (err)
1235 		goto nla_put_failure;
1236 
1237 	get_dp_stats(dp, &dp_stats);
1238 	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1239 		goto nla_put_failure;
1240 
1241 	return genlmsg_end(skb, ovs_header);
1242 
1243 nla_put_failure:
1244 	genlmsg_cancel(skb, ovs_header);
1245 error:
1246 	return -EMSGSIZE;
1247 }
1248 
1249 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1250 					     u32 seq, u8 cmd)
1251 {
1252 	struct sk_buff *skb;
1253 	int retval;
1254 
1255 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1256 	if (!skb)
1257 		return ERR_PTR(-ENOMEM);
1258 
1259 	retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1260 	if (retval < 0) {
1261 		kfree_skb(skb);
1262 		return ERR_PTR(retval);
1263 	}
1264 	return skb;
1265 }
1266 
1267 /* Called with genl_mutex and optionally with RTNL lock also. */
1268 static struct datapath *lookup_datapath(struct net *net,
1269 					struct ovs_header *ovs_header,
1270 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1271 {
1272 	struct datapath *dp;
1273 
1274 	if (!a[OVS_DP_ATTR_NAME])
1275 		dp = get_dp(net, ovs_header->dp_ifindex);
1276 	else {
1277 		struct vport *vport;
1278 
1279 		rcu_read_lock();
1280 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1281 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1282 		rcu_read_unlock();
1283 	}
1284 	return dp ? dp : ERR_PTR(-ENODEV);
1285 }
1286 
1287 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1288 {
1289 	struct nlattr **a = info->attrs;
1290 	struct vport_parms parms;
1291 	struct sk_buff *reply;
1292 	struct datapath *dp;
1293 	struct vport *vport;
1294 	struct ovs_net *ovs_net;
1295 	int err, i;
1296 
1297 	err = -EINVAL;
1298 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1299 		goto err;
1300 
1301 	rtnl_lock();
1302 
1303 	err = -ENOMEM;
1304 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1305 	if (dp == NULL)
1306 		goto err_unlock_rtnl;
1307 
1308 	ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1309 
1310 	/* Allocate table. */
1311 	err = -ENOMEM;
1312 	rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1313 	if (!dp->table)
1314 		goto err_free_dp;
1315 
1316 	dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1317 	if (!dp->stats_percpu) {
1318 		err = -ENOMEM;
1319 		goto err_destroy_table;
1320 	}
1321 
1322 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1323 			GFP_KERNEL);
1324 	if (!dp->ports) {
1325 		err = -ENOMEM;
1326 		goto err_destroy_percpu;
1327 	}
1328 
1329 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1330 		INIT_HLIST_HEAD(&dp->ports[i]);
1331 
1332 	/* Set up our datapath device. */
1333 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1334 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1335 	parms.options = NULL;
1336 	parms.dp = dp;
1337 	parms.port_no = OVSP_LOCAL;
1338 	parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1339 
1340 	vport = new_vport(&parms);
1341 	if (IS_ERR(vport)) {
1342 		err = PTR_ERR(vport);
1343 		if (err == -EBUSY)
1344 			err = -EEXIST;
1345 
1346 		goto err_destroy_ports_array;
1347 	}
1348 
1349 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1350 				      info->snd_seq, OVS_DP_CMD_NEW);
1351 	err = PTR_ERR(reply);
1352 	if (IS_ERR(reply))
1353 		goto err_destroy_local_port;
1354 
1355 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1356 	list_add_tail(&dp->list_node, &ovs_net->dps);
1357 	rtnl_unlock();
1358 
1359 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1360 		    ovs_dp_datapath_multicast_group.id, info->nlhdr,
1361 		    GFP_KERNEL);
1362 	return 0;
1363 
1364 err_destroy_local_port:
1365 	ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1366 err_destroy_ports_array:
1367 	kfree(dp->ports);
1368 err_destroy_percpu:
1369 	free_percpu(dp->stats_percpu);
1370 err_destroy_table:
1371 	ovs_flow_tbl_destroy(genl_dereference(dp->table));
1372 err_free_dp:
1373 	release_net(ovs_dp_get_net(dp));
1374 	kfree(dp);
1375 err_unlock_rtnl:
1376 	rtnl_unlock();
1377 err:
1378 	return err;
1379 }
1380 
1381 /* Called with genl_mutex. */
1382 static void __dp_destroy(struct datapath *dp)
1383 {
1384 	int i;
1385 
1386 	rtnl_lock();
1387 
1388 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1389 		struct vport *vport;
1390 		struct hlist_node *n;
1391 
1392 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1393 			if (vport->port_no != OVSP_LOCAL)
1394 				ovs_dp_detach_port(vport);
1395 	}
1396 
1397 	list_del(&dp->list_node);
1398 	ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1399 
1400 	/* rtnl_unlock() will wait until all the references to devices that
1401 	 * are pending unregistration have been dropped.  We do it here to
1402 	 * ensure that any internal devices (which contain DP pointers) are
1403 	 * fully destroyed before freeing the datapath.
1404 	 */
1405 	rtnl_unlock();
1406 
1407 	call_rcu(&dp->rcu, destroy_dp_rcu);
1408 }
1409 
1410 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1411 {
1412 	struct sk_buff *reply;
1413 	struct datapath *dp;
1414 	int err;
1415 
1416 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1417 	err = PTR_ERR(dp);
1418 	if (IS_ERR(dp))
1419 		return err;
1420 
1421 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1422 				      info->snd_seq, OVS_DP_CMD_DEL);
1423 	err = PTR_ERR(reply);
1424 	if (IS_ERR(reply))
1425 		return err;
1426 
1427 	__dp_destroy(dp);
1428 
1429 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1430 		    ovs_dp_datapath_multicast_group.id, info->nlhdr,
1431 		    GFP_KERNEL);
1432 
1433 	return 0;
1434 }
1435 
1436 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1437 {
1438 	struct sk_buff *reply;
1439 	struct datapath *dp;
1440 	int err;
1441 
1442 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1443 	if (IS_ERR(dp))
1444 		return PTR_ERR(dp);
1445 
1446 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1447 				      info->snd_seq, OVS_DP_CMD_NEW);
1448 	if (IS_ERR(reply)) {
1449 		err = PTR_ERR(reply);
1450 		netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1451 				ovs_dp_datapath_multicast_group.id, err);
1452 		return 0;
1453 	}
1454 
1455 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1456 		    ovs_dp_datapath_multicast_group.id, info->nlhdr,
1457 		    GFP_KERNEL);
1458 
1459 	return 0;
1460 }
1461 
1462 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1463 {
1464 	struct sk_buff *reply;
1465 	struct datapath *dp;
1466 
1467 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1468 	if (IS_ERR(dp))
1469 		return PTR_ERR(dp);
1470 
1471 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1472 				      info->snd_seq, OVS_DP_CMD_NEW);
1473 	if (IS_ERR(reply))
1474 		return PTR_ERR(reply);
1475 
1476 	return genlmsg_reply(reply, info);
1477 }
1478 
1479 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1480 {
1481 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1482 	struct datapath *dp;
1483 	int skip = cb->args[0];
1484 	int i = 0;
1485 
1486 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1487 		if (i >= skip &&
1488 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1489 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1490 					 OVS_DP_CMD_NEW) < 0)
1491 			break;
1492 		i++;
1493 	}
1494 
1495 	cb->args[0] = i;
1496 
1497 	return skb->len;
1498 }
1499 
1500 static struct genl_ops dp_datapath_genl_ops[] = {
1501 	{ .cmd = OVS_DP_CMD_NEW,
1502 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1503 	  .policy = datapath_policy,
1504 	  .doit = ovs_dp_cmd_new
1505 	},
1506 	{ .cmd = OVS_DP_CMD_DEL,
1507 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1508 	  .policy = datapath_policy,
1509 	  .doit = ovs_dp_cmd_del
1510 	},
1511 	{ .cmd = OVS_DP_CMD_GET,
1512 	  .flags = 0,		    /* OK for unprivileged users. */
1513 	  .policy = datapath_policy,
1514 	  .doit = ovs_dp_cmd_get,
1515 	  .dumpit = ovs_dp_cmd_dump
1516 	},
1517 	{ .cmd = OVS_DP_CMD_SET,
1518 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1519 	  .policy = datapath_policy,
1520 	  .doit = ovs_dp_cmd_set,
1521 	},
1522 };
1523 
1524 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1525 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1526 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1527 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1528 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1529 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1530 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1531 };
1532 
1533 static struct genl_family dp_vport_genl_family = {
1534 	.id = GENL_ID_GENERATE,
1535 	.hdrsize = sizeof(struct ovs_header),
1536 	.name = OVS_VPORT_FAMILY,
1537 	.version = OVS_VPORT_VERSION,
1538 	.maxattr = OVS_VPORT_ATTR_MAX,
1539 	.netnsok = true
1540 };
1541 
1542 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1543 	.name = OVS_VPORT_MCGROUP
1544 };
1545 
1546 /* Called with RTNL lock or RCU read lock. */
1547 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1548 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1549 {
1550 	struct ovs_header *ovs_header;
1551 	struct ovs_vport_stats vport_stats;
1552 	int err;
1553 
1554 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1555 				 flags, cmd);
1556 	if (!ovs_header)
1557 		return -EMSGSIZE;
1558 
1559 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1560 
1561 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1562 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1563 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1564 	    nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1565 		goto nla_put_failure;
1566 
1567 	ovs_vport_get_stats(vport, &vport_stats);
1568 	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1569 		    &vport_stats))
1570 		goto nla_put_failure;
1571 
1572 	err = ovs_vport_get_options(vport, skb);
1573 	if (err == -EMSGSIZE)
1574 		goto error;
1575 
1576 	return genlmsg_end(skb, ovs_header);
1577 
1578 nla_put_failure:
1579 	err = -EMSGSIZE;
1580 error:
1581 	genlmsg_cancel(skb, ovs_header);
1582 	return err;
1583 }
1584 
1585 /* Called with RTNL lock or RCU read lock. */
1586 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1587 					 u32 seq, u8 cmd)
1588 {
1589 	struct sk_buff *skb;
1590 	int retval;
1591 
1592 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1593 	if (!skb)
1594 		return ERR_PTR(-ENOMEM);
1595 
1596 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1597 	if (retval < 0) {
1598 		kfree_skb(skb);
1599 		return ERR_PTR(retval);
1600 	}
1601 	return skb;
1602 }
1603 
1604 /* Called with RTNL lock or RCU read lock. */
1605 static struct vport *lookup_vport(struct net *net,
1606 				  struct ovs_header *ovs_header,
1607 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1608 {
1609 	struct datapath *dp;
1610 	struct vport *vport;
1611 
1612 	if (a[OVS_VPORT_ATTR_NAME]) {
1613 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1614 		if (!vport)
1615 			return ERR_PTR(-ENODEV);
1616 		if (ovs_header->dp_ifindex &&
1617 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1618 			return ERR_PTR(-ENODEV);
1619 		return vport;
1620 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1621 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1622 
1623 		if (port_no >= DP_MAX_PORTS)
1624 			return ERR_PTR(-EFBIG);
1625 
1626 		dp = get_dp(net, ovs_header->dp_ifindex);
1627 		if (!dp)
1628 			return ERR_PTR(-ENODEV);
1629 
1630 		vport = ovs_vport_rtnl_rcu(dp, port_no);
1631 		if (!vport)
1632 			return ERR_PTR(-ENODEV);
1633 		return vport;
1634 	} else
1635 		return ERR_PTR(-EINVAL);
1636 }
1637 
1638 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1639 {
1640 	struct nlattr **a = info->attrs;
1641 	struct ovs_header *ovs_header = info->userhdr;
1642 	struct vport_parms parms;
1643 	struct sk_buff *reply;
1644 	struct vport *vport;
1645 	struct datapath *dp;
1646 	u32 port_no;
1647 	int err;
1648 
1649 	err = -EINVAL;
1650 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1651 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1652 		goto exit;
1653 
1654 	rtnl_lock();
1655 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1656 	err = -ENODEV;
1657 	if (!dp)
1658 		goto exit_unlock;
1659 
1660 	if (a[OVS_VPORT_ATTR_PORT_NO]) {
1661 		port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1662 
1663 		err = -EFBIG;
1664 		if (port_no >= DP_MAX_PORTS)
1665 			goto exit_unlock;
1666 
1667 		vport = ovs_vport_rtnl_rcu(dp, port_no);
1668 		err = -EBUSY;
1669 		if (vport)
1670 			goto exit_unlock;
1671 	} else {
1672 		for (port_no = 1; ; port_no++) {
1673 			if (port_no >= DP_MAX_PORTS) {
1674 				err = -EFBIG;
1675 				goto exit_unlock;
1676 			}
1677 			vport = ovs_vport_rtnl(dp, port_no);
1678 			if (!vport)
1679 				break;
1680 		}
1681 	}
1682 
1683 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1684 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1685 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1686 	parms.dp = dp;
1687 	parms.port_no = port_no;
1688 	parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1689 
1690 	vport = new_vport(&parms);
1691 	err = PTR_ERR(vport);
1692 	if (IS_ERR(vport))
1693 		goto exit_unlock;
1694 
1695 	err = 0;
1696 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1697 					 OVS_VPORT_CMD_NEW);
1698 	if (IS_ERR(reply)) {
1699 		err = PTR_ERR(reply);
1700 		ovs_dp_detach_port(vport);
1701 		goto exit_unlock;
1702 	}
1703 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1704 		    ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1705 
1706 exit_unlock:
1707 	rtnl_unlock();
1708 exit:
1709 	return err;
1710 }
1711 
1712 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1713 {
1714 	struct nlattr **a = info->attrs;
1715 	struct sk_buff *reply;
1716 	struct vport *vport;
1717 	int err;
1718 
1719 	rtnl_lock();
1720 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1721 	err = PTR_ERR(vport);
1722 	if (IS_ERR(vport))
1723 		goto exit_unlock;
1724 
1725 	err = 0;
1726 	if (a[OVS_VPORT_ATTR_TYPE] &&
1727 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1728 		err = -EINVAL;
1729 
1730 	if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1731 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1732 	if (err)
1733 		goto exit_unlock;
1734 	if (a[OVS_VPORT_ATTR_UPCALL_PID])
1735 		vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1736 
1737 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1738 					 OVS_VPORT_CMD_NEW);
1739 	if (IS_ERR(reply)) {
1740 		netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1741 				ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1742 		goto exit_unlock;
1743 	}
1744 
1745 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1746 		    ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1747 
1748 exit_unlock:
1749 	rtnl_unlock();
1750 	return err;
1751 }
1752 
1753 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1754 {
1755 	struct nlattr **a = info->attrs;
1756 	struct sk_buff *reply;
1757 	struct vport *vport;
1758 	int err;
1759 
1760 	rtnl_lock();
1761 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1762 	err = PTR_ERR(vport);
1763 	if (IS_ERR(vport))
1764 		goto exit_unlock;
1765 
1766 	if (vport->port_no == OVSP_LOCAL) {
1767 		err = -EINVAL;
1768 		goto exit_unlock;
1769 	}
1770 
1771 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1772 					 OVS_VPORT_CMD_DEL);
1773 	err = PTR_ERR(reply);
1774 	if (IS_ERR(reply))
1775 		goto exit_unlock;
1776 
1777 	err = 0;
1778 	ovs_dp_detach_port(vport);
1779 
1780 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1781 		    ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1782 
1783 exit_unlock:
1784 	rtnl_unlock();
1785 	return err;
1786 }
1787 
1788 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1789 {
1790 	struct nlattr **a = info->attrs;
1791 	struct ovs_header *ovs_header = info->userhdr;
1792 	struct sk_buff *reply;
1793 	struct vport *vport;
1794 	int err;
1795 
1796 	rcu_read_lock();
1797 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1798 	err = PTR_ERR(vport);
1799 	if (IS_ERR(vport))
1800 		goto exit_unlock;
1801 
1802 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1803 					 OVS_VPORT_CMD_NEW);
1804 	err = PTR_ERR(reply);
1805 	if (IS_ERR(reply))
1806 		goto exit_unlock;
1807 
1808 	rcu_read_unlock();
1809 
1810 	return genlmsg_reply(reply, info);
1811 
1812 exit_unlock:
1813 	rcu_read_unlock();
1814 	return err;
1815 }
1816 
1817 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1818 {
1819 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1820 	struct datapath *dp;
1821 	int bucket = cb->args[0], skip = cb->args[1];
1822 	int i, j = 0;
1823 
1824 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1825 	if (!dp)
1826 		return -ENODEV;
1827 
1828 	rcu_read_lock();
1829 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1830 		struct vport *vport;
1831 
1832 		j = 0;
1833 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1834 			if (j >= skip &&
1835 			    ovs_vport_cmd_fill_info(vport, skb,
1836 						    NETLINK_CB(cb->skb).portid,
1837 						    cb->nlh->nlmsg_seq,
1838 						    NLM_F_MULTI,
1839 						    OVS_VPORT_CMD_NEW) < 0)
1840 				goto out;
1841 
1842 			j++;
1843 		}
1844 		skip = 0;
1845 	}
1846 out:
1847 	rcu_read_unlock();
1848 
1849 	cb->args[0] = i;
1850 	cb->args[1] = j;
1851 
1852 	return skb->len;
1853 }
1854 
1855 static struct genl_ops dp_vport_genl_ops[] = {
1856 	{ .cmd = OVS_VPORT_CMD_NEW,
1857 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1858 	  .policy = vport_policy,
1859 	  .doit = ovs_vport_cmd_new
1860 	},
1861 	{ .cmd = OVS_VPORT_CMD_DEL,
1862 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1863 	  .policy = vport_policy,
1864 	  .doit = ovs_vport_cmd_del
1865 	},
1866 	{ .cmd = OVS_VPORT_CMD_GET,
1867 	  .flags = 0,		    /* OK for unprivileged users. */
1868 	  .policy = vport_policy,
1869 	  .doit = ovs_vport_cmd_get,
1870 	  .dumpit = ovs_vport_cmd_dump
1871 	},
1872 	{ .cmd = OVS_VPORT_CMD_SET,
1873 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1874 	  .policy = vport_policy,
1875 	  .doit = ovs_vport_cmd_set,
1876 	},
1877 };
1878 
1879 struct genl_family_and_ops {
1880 	struct genl_family *family;
1881 	struct genl_ops *ops;
1882 	int n_ops;
1883 	struct genl_multicast_group *group;
1884 };
1885 
1886 static const struct genl_family_and_ops dp_genl_families[] = {
1887 	{ &dp_datapath_genl_family,
1888 	  dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1889 	  &ovs_dp_datapath_multicast_group },
1890 	{ &dp_vport_genl_family,
1891 	  dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1892 	  &ovs_dp_vport_multicast_group },
1893 	{ &dp_flow_genl_family,
1894 	  dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1895 	  &ovs_dp_flow_multicast_group },
1896 	{ &dp_packet_genl_family,
1897 	  dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1898 	  NULL },
1899 };
1900 
1901 static void dp_unregister_genl(int n_families)
1902 {
1903 	int i;
1904 
1905 	for (i = 0; i < n_families; i++)
1906 		genl_unregister_family(dp_genl_families[i].family);
1907 }
1908 
1909 static int dp_register_genl(void)
1910 {
1911 	int n_registered;
1912 	int err;
1913 	int i;
1914 
1915 	n_registered = 0;
1916 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1917 		const struct genl_family_and_ops *f = &dp_genl_families[i];
1918 
1919 		err = genl_register_family_with_ops(f->family, f->ops,
1920 						    f->n_ops);
1921 		if (err)
1922 			goto error;
1923 		n_registered++;
1924 
1925 		if (f->group) {
1926 			err = genl_register_mc_group(f->family, f->group);
1927 			if (err)
1928 				goto error;
1929 		}
1930 	}
1931 
1932 	return 0;
1933 
1934 error:
1935 	dp_unregister_genl(n_registered);
1936 	return err;
1937 }
1938 
1939 static void rehash_flow_table(struct work_struct *work)
1940 {
1941 	struct datapath *dp;
1942 	struct net *net;
1943 
1944 	genl_lock();
1945 	rtnl_lock();
1946 	for_each_net(net) {
1947 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1948 
1949 		list_for_each_entry(dp, &ovs_net->dps, list_node) {
1950 			struct flow_table *old_table = genl_dereference(dp->table);
1951 			struct flow_table *new_table;
1952 
1953 			new_table = ovs_flow_tbl_rehash(old_table);
1954 			if (!IS_ERR(new_table)) {
1955 				rcu_assign_pointer(dp->table, new_table);
1956 				ovs_flow_tbl_deferred_destroy(old_table);
1957 			}
1958 		}
1959 	}
1960 	rtnl_unlock();
1961 	genl_unlock();
1962 
1963 	schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1964 }
1965 
1966 static int __net_init ovs_init_net(struct net *net)
1967 {
1968 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1969 
1970 	INIT_LIST_HEAD(&ovs_net->dps);
1971 	return 0;
1972 }
1973 
1974 static void __net_exit ovs_exit_net(struct net *net)
1975 {
1976 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1977 	struct datapath *dp, *dp_next;
1978 
1979 	genl_lock();
1980 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1981 		__dp_destroy(dp);
1982 	genl_unlock();
1983 }
1984 
1985 static struct pernet_operations ovs_net_ops = {
1986 	.init = ovs_init_net,
1987 	.exit = ovs_exit_net,
1988 	.id   = &ovs_net_id,
1989 	.size = sizeof(struct ovs_net),
1990 };
1991 
1992 static int __init dp_init(void)
1993 {
1994 	int err;
1995 
1996 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1997 
1998 	pr_info("Open vSwitch switching datapath\n");
1999 
2000 	err = ovs_flow_init();
2001 	if (err)
2002 		goto error;
2003 
2004 	err = ovs_vport_init();
2005 	if (err)
2006 		goto error_flow_exit;
2007 
2008 	err = register_pernet_device(&ovs_net_ops);
2009 	if (err)
2010 		goto error_vport_exit;
2011 
2012 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2013 	if (err)
2014 		goto error_netns_exit;
2015 
2016 	err = dp_register_genl();
2017 	if (err < 0)
2018 		goto error_unreg_notifier;
2019 
2020 	schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2021 
2022 	return 0;
2023 
2024 error_unreg_notifier:
2025 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2026 error_netns_exit:
2027 	unregister_pernet_device(&ovs_net_ops);
2028 error_vport_exit:
2029 	ovs_vport_exit();
2030 error_flow_exit:
2031 	ovs_flow_exit();
2032 error:
2033 	return err;
2034 }
2035 
2036 static void dp_cleanup(void)
2037 {
2038 	cancel_delayed_work_sync(&rehash_flow_wq);
2039 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2040 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2041 	unregister_pernet_device(&ovs_net_ops);
2042 	rcu_barrier();
2043 	ovs_vport_exit();
2044 	ovs_flow_exit();
2045 }
2046 
2047 module_init(dp_init);
2048 module_exit(dp_cleanup);
2049 
2050 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2051 MODULE_LICENSE("GPL");
2052