xref: /linux/net/openvswitch/datapath.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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_node *n;
162 	struct hlist_head *head;
163 
164 	head = vport_hash_bucket(dp, port_no);
165 	hlist_for_each_entry_rcu(vport, n, head, dp_hash_node) {
166 		if (vport->port_no == port_no)
167 			return vport;
168 	}
169 	return NULL;
170 }
171 
172 /* Called with RTNL lock and genl_lock. */
173 static struct vport *new_vport(const struct vport_parms *parms)
174 {
175 	struct vport *vport;
176 
177 	vport = ovs_vport_add(parms);
178 	if (!IS_ERR(vport)) {
179 		struct datapath *dp = parms->dp;
180 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
181 
182 		hlist_add_head_rcu(&vport->dp_hash_node, head);
183 	}
184 
185 	return vport;
186 }
187 
188 /* Called with RTNL lock. */
189 void ovs_dp_detach_port(struct vport *p)
190 {
191 	ASSERT_RTNL();
192 
193 	/* First drop references to device. */
194 	hlist_del_rcu(&p->dp_hash_node);
195 
196 	/* Then destroy it. */
197 	ovs_vport_del(p);
198 }
199 
200 /* Must be called with rcu_read_lock. */
201 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
202 {
203 	struct datapath *dp = p->dp;
204 	struct sw_flow *flow;
205 	struct dp_stats_percpu *stats;
206 	struct sw_flow_key key;
207 	u64 *stats_counter;
208 	int error;
209 	int key_len;
210 
211 	stats = this_cpu_ptr(dp->stats_percpu);
212 
213 	/* Extract flow from 'skb' into 'key'. */
214 	error = ovs_flow_extract(skb, p->port_no, &key, &key_len);
215 	if (unlikely(error)) {
216 		kfree_skb(skb);
217 		return;
218 	}
219 
220 	/* Look up flow. */
221 	flow = ovs_flow_tbl_lookup(rcu_dereference(dp->table), &key, key_len);
222 	if (unlikely(!flow)) {
223 		struct dp_upcall_info upcall;
224 
225 		upcall.cmd = OVS_PACKET_CMD_MISS;
226 		upcall.key = &key;
227 		upcall.userdata = NULL;
228 		upcall.portid = p->upcall_portid;
229 		ovs_dp_upcall(dp, skb, &upcall);
230 		consume_skb(skb);
231 		stats_counter = &stats->n_missed;
232 		goto out;
233 	}
234 
235 	OVS_CB(skb)->flow = flow;
236 
237 	stats_counter = &stats->n_hit;
238 	ovs_flow_used(OVS_CB(skb)->flow, skb);
239 	ovs_execute_actions(dp, skb);
240 
241 out:
242 	/* Update datapath statistics. */
243 	u64_stats_update_begin(&stats->sync);
244 	(*stats_counter)++;
245 	u64_stats_update_end(&stats->sync);
246 }
247 
248 static struct genl_family dp_packet_genl_family = {
249 	.id = GENL_ID_GENERATE,
250 	.hdrsize = sizeof(struct ovs_header),
251 	.name = OVS_PACKET_FAMILY,
252 	.version = OVS_PACKET_VERSION,
253 	.maxattr = OVS_PACKET_ATTR_MAX,
254 	.netnsok = true
255 };
256 
257 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
258 		  const struct dp_upcall_info *upcall_info)
259 {
260 	struct dp_stats_percpu *stats;
261 	int dp_ifindex;
262 	int err;
263 
264 	if (upcall_info->portid == 0) {
265 		err = -ENOTCONN;
266 		goto err;
267 	}
268 
269 	dp_ifindex = get_dpifindex(dp);
270 	if (!dp_ifindex) {
271 		err = -ENODEV;
272 		goto err;
273 	}
274 
275 	if (!skb_is_gso(skb))
276 		err = queue_userspace_packet(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
277 	else
278 		err = queue_gso_packets(ovs_dp_get_net(dp), dp_ifindex, skb, upcall_info);
279 	if (err)
280 		goto err;
281 
282 	return 0;
283 
284 err:
285 	stats = this_cpu_ptr(dp->stats_percpu);
286 
287 	u64_stats_update_begin(&stats->sync);
288 	stats->n_lost++;
289 	u64_stats_update_end(&stats->sync);
290 
291 	return err;
292 }
293 
294 static int queue_gso_packets(struct net *net, int dp_ifindex,
295 			     struct sk_buff *skb,
296 			     const struct dp_upcall_info *upcall_info)
297 {
298 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
299 	struct dp_upcall_info later_info;
300 	struct sw_flow_key later_key;
301 	struct sk_buff *segs, *nskb;
302 	int err;
303 
304 	segs = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM);
305 	if (IS_ERR(segs))
306 		return PTR_ERR(segs);
307 
308 	/* Queue all of the segments. */
309 	skb = segs;
310 	do {
311 		err = queue_userspace_packet(net, dp_ifindex, skb, upcall_info);
312 		if (err)
313 			break;
314 
315 		if (skb == segs && gso_type & SKB_GSO_UDP) {
316 			/* The initial flow key extracted by ovs_flow_extract()
317 			 * in this case is for a first fragment, so we need to
318 			 * properly mark later fragments.
319 			 */
320 			later_key = *upcall_info->key;
321 			later_key.ip.frag = OVS_FRAG_TYPE_LATER;
322 
323 			later_info = *upcall_info;
324 			later_info.key = &later_key;
325 			upcall_info = &later_info;
326 		}
327 	} while ((skb = skb->next));
328 
329 	/* Free all of the segments. */
330 	skb = segs;
331 	do {
332 		nskb = skb->next;
333 		if (err)
334 			kfree_skb(skb);
335 		else
336 			consume_skb(skb);
337 	} while ((skb = nskb));
338 	return err;
339 }
340 
341 static int queue_userspace_packet(struct net *net, int dp_ifindex,
342 				  struct sk_buff *skb,
343 				  const struct dp_upcall_info *upcall_info)
344 {
345 	struct ovs_header *upcall;
346 	struct sk_buff *nskb = NULL;
347 	struct sk_buff *user_skb; /* to be queued to userspace */
348 	struct nlattr *nla;
349 	unsigned int len;
350 	int err;
351 
352 	if (vlan_tx_tag_present(skb)) {
353 		nskb = skb_clone(skb, GFP_ATOMIC);
354 		if (!nskb)
355 			return -ENOMEM;
356 
357 		nskb = __vlan_put_tag(nskb, vlan_tx_tag_get(nskb));
358 		if (!nskb)
359 			return -ENOMEM;
360 
361 		nskb->vlan_tci = 0;
362 		skb = nskb;
363 	}
364 
365 	if (nla_attr_size(skb->len) > USHRT_MAX) {
366 		err = -EFBIG;
367 		goto out;
368 	}
369 
370 	len = sizeof(struct ovs_header);
371 	len += nla_total_size(skb->len);
372 	len += nla_total_size(FLOW_BUFSIZE);
373 	if (upcall_info->cmd == OVS_PACKET_CMD_ACTION)
374 		len += nla_total_size(8);
375 
376 	user_skb = genlmsg_new(len, GFP_ATOMIC);
377 	if (!user_skb) {
378 		err = -ENOMEM;
379 		goto out;
380 	}
381 
382 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
383 			     0, upcall_info->cmd);
384 	upcall->dp_ifindex = dp_ifindex;
385 
386 	nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
387 	ovs_flow_to_nlattrs(upcall_info->key, user_skb);
388 	nla_nest_end(user_skb, nla);
389 
390 	if (upcall_info->userdata)
391 		nla_put_u64(user_skb, OVS_PACKET_ATTR_USERDATA,
392 			    nla_get_u64(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 	err = genlmsg_unicast(net, user_skb, upcall_info->portid);
399 
400 out:
401 	kfree_skb(nskb);
402 	return err;
403 }
404 
405 /* Called with genl_mutex. */
406 static int flush_flows(struct datapath *dp)
407 {
408 	struct flow_table *old_table;
409 	struct flow_table *new_table;
410 
411 	old_table = genl_dereference(dp->table);
412 	new_table = ovs_flow_tbl_alloc(TBL_MIN_BUCKETS);
413 	if (!new_table)
414 		return -ENOMEM;
415 
416 	rcu_assign_pointer(dp->table, new_table);
417 
418 	ovs_flow_tbl_deferred_destroy(old_table);
419 	return 0;
420 }
421 
422 static int validate_actions(const struct nlattr *attr,
423 				const struct sw_flow_key *key, int depth);
424 
425 static int validate_sample(const struct nlattr *attr,
426 				const struct sw_flow_key *key, int depth)
427 {
428 	const struct nlattr *attrs[OVS_SAMPLE_ATTR_MAX + 1];
429 	const struct nlattr *probability, *actions;
430 	const struct nlattr *a;
431 	int rem;
432 
433 	memset(attrs, 0, sizeof(attrs));
434 	nla_for_each_nested(a, attr, rem) {
435 		int type = nla_type(a);
436 		if (!type || type > OVS_SAMPLE_ATTR_MAX || attrs[type])
437 			return -EINVAL;
438 		attrs[type] = a;
439 	}
440 	if (rem)
441 		return -EINVAL;
442 
443 	probability = attrs[OVS_SAMPLE_ATTR_PROBABILITY];
444 	if (!probability || nla_len(probability) != sizeof(u32))
445 		return -EINVAL;
446 
447 	actions = attrs[OVS_SAMPLE_ATTR_ACTIONS];
448 	if (!actions || (nla_len(actions) && nla_len(actions) < NLA_HDRLEN))
449 		return -EINVAL;
450 	return validate_actions(actions, key, depth + 1);
451 }
452 
453 static int validate_tp_port(const struct sw_flow_key *flow_key)
454 {
455 	if (flow_key->eth.type == htons(ETH_P_IP)) {
456 		if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst)
457 			return 0;
458 	} else if (flow_key->eth.type == htons(ETH_P_IPV6)) {
459 		if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst)
460 			return 0;
461 	}
462 
463 	return -EINVAL;
464 }
465 
466 static int validate_set(const struct nlattr *a,
467 			const struct sw_flow_key *flow_key)
468 {
469 	const struct nlattr *ovs_key = nla_data(a);
470 	int key_type = nla_type(ovs_key);
471 
472 	/* There can be only one key in a action */
473 	if (nla_total_size(nla_len(ovs_key)) != nla_len(a))
474 		return -EINVAL;
475 
476 	if (key_type > OVS_KEY_ATTR_MAX ||
477 	    nla_len(ovs_key) != ovs_key_lens[key_type])
478 		return -EINVAL;
479 
480 	switch (key_type) {
481 	const struct ovs_key_ipv4 *ipv4_key;
482 	const struct ovs_key_ipv6 *ipv6_key;
483 
484 	case OVS_KEY_ATTR_PRIORITY:
485 	case OVS_KEY_ATTR_SKB_MARK:
486 	case OVS_KEY_ATTR_ETHERNET:
487 		break;
488 
489 	case OVS_KEY_ATTR_IPV4:
490 		if (flow_key->eth.type != htons(ETH_P_IP))
491 			return -EINVAL;
492 
493 		if (!flow_key->ip.proto)
494 			return -EINVAL;
495 
496 		ipv4_key = nla_data(ovs_key);
497 		if (ipv4_key->ipv4_proto != flow_key->ip.proto)
498 			return -EINVAL;
499 
500 		if (ipv4_key->ipv4_frag != flow_key->ip.frag)
501 			return -EINVAL;
502 
503 		break;
504 
505 	case OVS_KEY_ATTR_IPV6:
506 		if (flow_key->eth.type != htons(ETH_P_IPV6))
507 			return -EINVAL;
508 
509 		if (!flow_key->ip.proto)
510 			return -EINVAL;
511 
512 		ipv6_key = nla_data(ovs_key);
513 		if (ipv6_key->ipv6_proto != flow_key->ip.proto)
514 			return -EINVAL;
515 
516 		if (ipv6_key->ipv6_frag != flow_key->ip.frag)
517 			return -EINVAL;
518 
519 		if (ntohl(ipv6_key->ipv6_label) & 0xFFF00000)
520 			return -EINVAL;
521 
522 		break;
523 
524 	case OVS_KEY_ATTR_TCP:
525 		if (flow_key->ip.proto != IPPROTO_TCP)
526 			return -EINVAL;
527 
528 		return validate_tp_port(flow_key);
529 
530 	case OVS_KEY_ATTR_UDP:
531 		if (flow_key->ip.proto != IPPROTO_UDP)
532 			return -EINVAL;
533 
534 		return validate_tp_port(flow_key);
535 
536 	default:
537 		return -EINVAL;
538 	}
539 
540 	return 0;
541 }
542 
543 static int validate_userspace(const struct nlattr *attr)
544 {
545 	static const struct nla_policy userspace_policy[OVS_USERSPACE_ATTR_MAX + 1] =	{
546 		[OVS_USERSPACE_ATTR_PID] = {.type = NLA_U32 },
547 		[OVS_USERSPACE_ATTR_USERDATA] = {.type = NLA_U64 },
548 	};
549 	struct nlattr *a[OVS_USERSPACE_ATTR_MAX + 1];
550 	int error;
551 
552 	error = nla_parse_nested(a, OVS_USERSPACE_ATTR_MAX,
553 				 attr, userspace_policy);
554 	if (error)
555 		return error;
556 
557 	if (!a[OVS_USERSPACE_ATTR_PID] ||
558 	    !nla_get_u32(a[OVS_USERSPACE_ATTR_PID]))
559 		return -EINVAL;
560 
561 	return 0;
562 }
563 
564 static int validate_actions(const struct nlattr *attr,
565 				const struct sw_flow_key *key,  int depth)
566 {
567 	const struct nlattr *a;
568 	int rem, err;
569 
570 	if (depth >= SAMPLE_ACTION_DEPTH)
571 		return -EOVERFLOW;
572 
573 	nla_for_each_nested(a, attr, rem) {
574 		/* Expected argument lengths, (u32)-1 for variable length. */
575 		static const u32 action_lens[OVS_ACTION_ATTR_MAX + 1] = {
576 			[OVS_ACTION_ATTR_OUTPUT] = sizeof(u32),
577 			[OVS_ACTION_ATTR_USERSPACE] = (u32)-1,
578 			[OVS_ACTION_ATTR_PUSH_VLAN] = sizeof(struct ovs_action_push_vlan),
579 			[OVS_ACTION_ATTR_POP_VLAN] = 0,
580 			[OVS_ACTION_ATTR_SET] = (u32)-1,
581 			[OVS_ACTION_ATTR_SAMPLE] = (u32)-1
582 		};
583 		const struct ovs_action_push_vlan *vlan;
584 		int type = nla_type(a);
585 
586 		if (type > OVS_ACTION_ATTR_MAX ||
587 		    (action_lens[type] != nla_len(a) &&
588 		     action_lens[type] != (u32)-1))
589 			return -EINVAL;
590 
591 		switch (type) {
592 		case OVS_ACTION_ATTR_UNSPEC:
593 			return -EINVAL;
594 
595 		case OVS_ACTION_ATTR_USERSPACE:
596 			err = validate_userspace(a);
597 			if (err)
598 				return err;
599 			break;
600 
601 		case OVS_ACTION_ATTR_OUTPUT:
602 			if (nla_get_u32(a) >= DP_MAX_PORTS)
603 				return -EINVAL;
604 			break;
605 
606 
607 		case OVS_ACTION_ATTR_POP_VLAN:
608 			break;
609 
610 		case OVS_ACTION_ATTR_PUSH_VLAN:
611 			vlan = nla_data(a);
612 			if (vlan->vlan_tpid != htons(ETH_P_8021Q))
613 				return -EINVAL;
614 			if (!(vlan->vlan_tci & htons(VLAN_TAG_PRESENT)))
615 				return -EINVAL;
616 			break;
617 
618 		case OVS_ACTION_ATTR_SET:
619 			err = validate_set(a, key);
620 			if (err)
621 				return err;
622 			break;
623 
624 		case OVS_ACTION_ATTR_SAMPLE:
625 			err = validate_sample(a, key, depth);
626 			if (err)
627 				return err;
628 			break;
629 
630 		default:
631 			return -EINVAL;
632 		}
633 	}
634 
635 	if (rem > 0)
636 		return -EINVAL;
637 
638 	return 0;
639 }
640 
641 static void clear_stats(struct sw_flow *flow)
642 {
643 	flow->used = 0;
644 	flow->tcp_flags = 0;
645 	flow->packet_count = 0;
646 	flow->byte_count = 0;
647 }
648 
649 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
650 {
651 	struct ovs_header *ovs_header = info->userhdr;
652 	struct nlattr **a = info->attrs;
653 	struct sw_flow_actions *acts;
654 	struct sk_buff *packet;
655 	struct sw_flow *flow;
656 	struct datapath *dp;
657 	struct ethhdr *eth;
658 	int len;
659 	int err;
660 	int key_len;
661 
662 	err = -EINVAL;
663 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
664 	    !a[OVS_PACKET_ATTR_ACTIONS] ||
665 	    nla_len(a[OVS_PACKET_ATTR_PACKET]) < ETH_HLEN)
666 		goto err;
667 
668 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
669 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
670 	err = -ENOMEM;
671 	if (!packet)
672 		goto err;
673 	skb_reserve(packet, NET_IP_ALIGN);
674 
675 	memcpy(__skb_put(packet, len), nla_data(a[OVS_PACKET_ATTR_PACKET]), len);
676 
677 	skb_reset_mac_header(packet);
678 	eth = eth_hdr(packet);
679 
680 	/* Normally, setting the skb 'protocol' field would be handled by a
681 	 * call to eth_type_trans(), but it assumes there's a sending
682 	 * device, which we may not have. */
683 	if (ntohs(eth->h_proto) >= 1536)
684 		packet->protocol = eth->h_proto;
685 	else
686 		packet->protocol = htons(ETH_P_802_2);
687 
688 	/* Build an sw_flow for sending this packet. */
689 	flow = ovs_flow_alloc();
690 	err = PTR_ERR(flow);
691 	if (IS_ERR(flow))
692 		goto err_kfree_skb;
693 
694 	err = ovs_flow_extract(packet, -1, &flow->key, &key_len);
695 	if (err)
696 		goto err_flow_free;
697 
698 	err = ovs_flow_metadata_from_nlattrs(&flow->key.phy.priority,
699 					     &flow->key.phy.skb_mark,
700 					     &flow->key.phy.in_port,
701 					     a[OVS_PACKET_ATTR_KEY]);
702 	if (err)
703 		goto err_flow_free;
704 
705 	err = validate_actions(a[OVS_PACKET_ATTR_ACTIONS], &flow->key, 0);
706 	if (err)
707 		goto err_flow_free;
708 
709 	flow->hash = ovs_flow_hash(&flow->key, key_len);
710 
711 	acts = ovs_flow_actions_alloc(a[OVS_PACKET_ATTR_ACTIONS]);
712 	err = PTR_ERR(acts);
713 	if (IS_ERR(acts))
714 		goto err_flow_free;
715 	rcu_assign_pointer(flow->sf_acts, acts);
716 
717 	OVS_CB(packet)->flow = flow;
718 	packet->priority = flow->key.phy.priority;
719 	packet->mark = flow->key.phy.skb_mark;
720 
721 	rcu_read_lock();
722 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
723 	err = -ENODEV;
724 	if (!dp)
725 		goto err_unlock;
726 
727 	local_bh_disable();
728 	err = ovs_execute_actions(dp, packet);
729 	local_bh_enable();
730 	rcu_read_unlock();
731 
732 	ovs_flow_free(flow);
733 	return err;
734 
735 err_unlock:
736 	rcu_read_unlock();
737 err_flow_free:
738 	ovs_flow_free(flow);
739 err_kfree_skb:
740 	kfree_skb(packet);
741 err:
742 	return err;
743 }
744 
745 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
746 	[OVS_PACKET_ATTR_PACKET] = { .type = NLA_UNSPEC },
747 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
748 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
749 };
750 
751 static struct genl_ops dp_packet_genl_ops[] = {
752 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
753 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
754 	  .policy = packet_policy,
755 	  .doit = ovs_packet_cmd_execute
756 	}
757 };
758 
759 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats)
760 {
761 	int i;
762 	struct flow_table *table = genl_dereference(dp->table);
763 
764 	stats->n_flows = ovs_flow_tbl_count(table);
765 
766 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
767 	for_each_possible_cpu(i) {
768 		const struct dp_stats_percpu *percpu_stats;
769 		struct dp_stats_percpu local_stats;
770 		unsigned int start;
771 
772 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
773 
774 		do {
775 			start = u64_stats_fetch_begin_bh(&percpu_stats->sync);
776 			local_stats = *percpu_stats;
777 		} while (u64_stats_fetch_retry_bh(&percpu_stats->sync, start));
778 
779 		stats->n_hit += local_stats.n_hit;
780 		stats->n_missed += local_stats.n_missed;
781 		stats->n_lost += local_stats.n_lost;
782 	}
783 }
784 
785 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
786 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
787 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
788 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
789 };
790 
791 static struct genl_family dp_flow_genl_family = {
792 	.id = GENL_ID_GENERATE,
793 	.hdrsize = sizeof(struct ovs_header),
794 	.name = OVS_FLOW_FAMILY,
795 	.version = OVS_FLOW_VERSION,
796 	.maxattr = OVS_FLOW_ATTR_MAX,
797 	.netnsok = true
798 };
799 
800 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
801 	.name = OVS_FLOW_MCGROUP
802 };
803 
804 /* Called with genl_lock. */
805 static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
806 				  struct sk_buff *skb, u32 portid,
807 				  u32 seq, u32 flags, u8 cmd)
808 {
809 	const int skb_orig_len = skb->len;
810 	const struct sw_flow_actions *sf_acts;
811 	struct ovs_flow_stats stats;
812 	struct ovs_header *ovs_header;
813 	struct nlattr *nla;
814 	unsigned long used;
815 	u8 tcp_flags;
816 	int err;
817 
818 	sf_acts = rcu_dereference_protected(flow->sf_acts,
819 					    lockdep_genl_is_held());
820 
821 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
822 	if (!ovs_header)
823 		return -EMSGSIZE;
824 
825 	ovs_header->dp_ifindex = get_dpifindex(dp);
826 
827 	nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
828 	if (!nla)
829 		goto nla_put_failure;
830 	err = ovs_flow_to_nlattrs(&flow->key, skb);
831 	if (err)
832 		goto error;
833 	nla_nest_end(skb, nla);
834 
835 	spin_lock_bh(&flow->lock);
836 	used = flow->used;
837 	stats.n_packets = flow->packet_count;
838 	stats.n_bytes = flow->byte_count;
839 	tcp_flags = flow->tcp_flags;
840 	spin_unlock_bh(&flow->lock);
841 
842 	if (used &&
843 	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
844 		goto nla_put_failure;
845 
846 	if (stats.n_packets &&
847 	    nla_put(skb, OVS_FLOW_ATTR_STATS,
848 		    sizeof(struct ovs_flow_stats), &stats))
849 		goto nla_put_failure;
850 
851 	if (tcp_flags &&
852 	    nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, tcp_flags))
853 		goto nla_put_failure;
854 
855 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
856 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
857 	 * Netlink but individual action lists can be longer than
858 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
859 	 * The userspace caller can always fetch the actions separately if it
860 	 * really wants them.  (Most userspace callers in fact don't care.)
861 	 *
862 	 * This can only fail for dump operations because the skb is always
863 	 * properly sized for single flows.
864 	 */
865 	err = nla_put(skb, OVS_FLOW_ATTR_ACTIONS, sf_acts->actions_len,
866 		      sf_acts->actions);
867 	if (err < 0 && skb_orig_len)
868 		goto error;
869 
870 	return genlmsg_end(skb, ovs_header);
871 
872 nla_put_failure:
873 	err = -EMSGSIZE;
874 error:
875 	genlmsg_cancel(skb, ovs_header);
876 	return err;
877 }
878 
879 static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow)
880 {
881 	const struct sw_flow_actions *sf_acts;
882 	int len;
883 
884 	sf_acts = rcu_dereference_protected(flow->sf_acts,
885 					    lockdep_genl_is_held());
886 
887 	/* OVS_FLOW_ATTR_KEY */
888 	len = nla_total_size(FLOW_BUFSIZE);
889 	/* OVS_FLOW_ATTR_ACTIONS */
890 	len += nla_total_size(sf_acts->actions_len);
891 	/* OVS_FLOW_ATTR_STATS */
892 	len += nla_total_size(sizeof(struct ovs_flow_stats));
893 	/* OVS_FLOW_ATTR_TCP_FLAGS */
894 	len += nla_total_size(1);
895 	/* OVS_FLOW_ATTR_USED */
896 	len += nla_total_size(8);
897 
898 	len += NLMSG_ALIGN(sizeof(struct ovs_header));
899 
900 	return genlmsg_new(len, GFP_KERNEL);
901 }
902 
903 static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
904 					       struct datapath *dp,
905 					       u32 portid, u32 seq, u8 cmd)
906 {
907 	struct sk_buff *skb;
908 	int retval;
909 
910 	skb = ovs_flow_cmd_alloc_info(flow);
911 	if (!skb)
912 		return ERR_PTR(-ENOMEM);
913 
914 	retval = ovs_flow_cmd_fill_info(flow, dp, skb, portid, seq, 0, cmd);
915 	BUG_ON(retval < 0);
916 	return skb;
917 }
918 
919 static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
920 {
921 	struct nlattr **a = info->attrs;
922 	struct ovs_header *ovs_header = info->userhdr;
923 	struct sw_flow_key key;
924 	struct sw_flow *flow;
925 	struct sk_buff *reply;
926 	struct datapath *dp;
927 	struct flow_table *table;
928 	int error;
929 	int key_len;
930 
931 	/* Extract key. */
932 	error = -EINVAL;
933 	if (!a[OVS_FLOW_ATTR_KEY])
934 		goto error;
935 	error = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
936 	if (error)
937 		goto error;
938 
939 	/* Validate actions. */
940 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
941 		error = validate_actions(a[OVS_FLOW_ATTR_ACTIONS], &key,  0);
942 		if (error)
943 			goto error;
944 	} else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
945 		error = -EINVAL;
946 		goto error;
947 	}
948 
949 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
950 	error = -ENODEV;
951 	if (!dp)
952 		goto error;
953 
954 	table = genl_dereference(dp->table);
955 	flow = ovs_flow_tbl_lookup(table, &key, key_len);
956 	if (!flow) {
957 		struct sw_flow_actions *acts;
958 
959 		/* Bail out if we're not allowed to create a new flow. */
960 		error = -ENOENT;
961 		if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
962 			goto error;
963 
964 		/* Expand table, if necessary, to make room. */
965 		if (ovs_flow_tbl_need_to_expand(table)) {
966 			struct flow_table *new_table;
967 
968 			new_table = ovs_flow_tbl_expand(table);
969 			if (!IS_ERR(new_table)) {
970 				rcu_assign_pointer(dp->table, new_table);
971 				ovs_flow_tbl_deferred_destroy(table);
972 				table = genl_dereference(dp->table);
973 			}
974 		}
975 
976 		/* Allocate flow. */
977 		flow = ovs_flow_alloc();
978 		if (IS_ERR(flow)) {
979 			error = PTR_ERR(flow);
980 			goto error;
981 		}
982 		flow->key = key;
983 		clear_stats(flow);
984 
985 		/* Obtain actions. */
986 		acts = ovs_flow_actions_alloc(a[OVS_FLOW_ATTR_ACTIONS]);
987 		error = PTR_ERR(acts);
988 		if (IS_ERR(acts))
989 			goto error_free_flow;
990 		rcu_assign_pointer(flow->sf_acts, acts);
991 
992 		/* Put flow in bucket. */
993 		flow->hash = ovs_flow_hash(&key, key_len);
994 		ovs_flow_tbl_insert(table, flow);
995 
996 		reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
997 						info->snd_seq,
998 						OVS_FLOW_CMD_NEW);
999 	} else {
1000 		/* We found a matching flow. */
1001 		struct sw_flow_actions *old_acts;
1002 		struct nlattr *acts_attrs;
1003 
1004 		/* Bail out if we're not allowed to modify an existing flow.
1005 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1006 		 * because Generic Netlink treats the latter as a dump
1007 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1008 		 * gets fixed.
1009 		 */
1010 		error = -EEXIST;
1011 		if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
1012 		    info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
1013 			goto error;
1014 
1015 		/* Update actions. */
1016 		old_acts = rcu_dereference_protected(flow->sf_acts,
1017 						     lockdep_genl_is_held());
1018 		acts_attrs = a[OVS_FLOW_ATTR_ACTIONS];
1019 		if (acts_attrs &&
1020 		   (old_acts->actions_len != nla_len(acts_attrs) ||
1021 		   memcmp(old_acts->actions, nla_data(acts_attrs),
1022 			  old_acts->actions_len))) {
1023 			struct sw_flow_actions *new_acts;
1024 
1025 			new_acts = ovs_flow_actions_alloc(acts_attrs);
1026 			error = PTR_ERR(new_acts);
1027 			if (IS_ERR(new_acts))
1028 				goto error;
1029 
1030 			rcu_assign_pointer(flow->sf_acts, new_acts);
1031 			ovs_flow_deferred_free_acts(old_acts);
1032 		}
1033 
1034 		reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1035 					       info->snd_seq, OVS_FLOW_CMD_NEW);
1036 
1037 		/* Clear stats. */
1038 		if (a[OVS_FLOW_ATTR_CLEAR]) {
1039 			spin_lock_bh(&flow->lock);
1040 			clear_stats(flow);
1041 			spin_unlock_bh(&flow->lock);
1042 		}
1043 	}
1044 
1045 	if (!IS_ERR(reply))
1046 		genl_notify(reply, genl_info_net(info), info->snd_portid,
1047 			   ovs_dp_flow_multicast_group.id, info->nlhdr,
1048 			   GFP_KERNEL);
1049 	else
1050 		netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1051 				ovs_dp_flow_multicast_group.id, PTR_ERR(reply));
1052 	return 0;
1053 
1054 error_free_flow:
1055 	ovs_flow_free(flow);
1056 error:
1057 	return error;
1058 }
1059 
1060 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1061 {
1062 	struct nlattr **a = info->attrs;
1063 	struct ovs_header *ovs_header = info->userhdr;
1064 	struct sw_flow_key key;
1065 	struct sk_buff *reply;
1066 	struct sw_flow *flow;
1067 	struct datapath *dp;
1068 	struct flow_table *table;
1069 	int err;
1070 	int key_len;
1071 
1072 	if (!a[OVS_FLOW_ATTR_KEY])
1073 		return -EINVAL;
1074 	err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1075 	if (err)
1076 		return err;
1077 
1078 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1079 	if (!dp)
1080 		return -ENODEV;
1081 
1082 	table = genl_dereference(dp->table);
1083 	flow = ovs_flow_tbl_lookup(table, &key, key_len);
1084 	if (!flow)
1085 		return -ENOENT;
1086 
1087 	reply = ovs_flow_cmd_build_info(flow, dp, info->snd_portid,
1088 					info->snd_seq, OVS_FLOW_CMD_NEW);
1089 	if (IS_ERR(reply))
1090 		return PTR_ERR(reply);
1091 
1092 	return genlmsg_reply(reply, info);
1093 }
1094 
1095 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1096 {
1097 	struct nlattr **a = info->attrs;
1098 	struct ovs_header *ovs_header = info->userhdr;
1099 	struct sw_flow_key key;
1100 	struct sk_buff *reply;
1101 	struct sw_flow *flow;
1102 	struct datapath *dp;
1103 	struct flow_table *table;
1104 	int err;
1105 	int key_len;
1106 
1107 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1108 	if (!dp)
1109 		return -ENODEV;
1110 
1111 	if (!a[OVS_FLOW_ATTR_KEY])
1112 		return flush_flows(dp);
1113 
1114 	err = ovs_flow_from_nlattrs(&key, &key_len, a[OVS_FLOW_ATTR_KEY]);
1115 	if (err)
1116 		return err;
1117 
1118 	table = genl_dereference(dp->table);
1119 	flow = ovs_flow_tbl_lookup(table, &key, key_len);
1120 	if (!flow)
1121 		return -ENOENT;
1122 
1123 	reply = ovs_flow_cmd_alloc_info(flow);
1124 	if (!reply)
1125 		return -ENOMEM;
1126 
1127 	ovs_flow_tbl_remove(table, flow);
1128 
1129 	err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
1130 				     info->snd_seq, 0, OVS_FLOW_CMD_DEL);
1131 	BUG_ON(err < 0);
1132 
1133 	ovs_flow_deferred_free(flow);
1134 
1135 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1136 		    ovs_dp_flow_multicast_group.id, info->nlhdr, GFP_KERNEL);
1137 	return 0;
1138 }
1139 
1140 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1141 {
1142 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1143 	struct datapath *dp;
1144 	struct flow_table *table;
1145 
1146 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1147 	if (!dp)
1148 		return -ENODEV;
1149 
1150 	table = genl_dereference(dp->table);
1151 
1152 	for (;;) {
1153 		struct sw_flow *flow;
1154 		u32 bucket, obj;
1155 
1156 		bucket = cb->args[0];
1157 		obj = cb->args[1];
1158 		flow = ovs_flow_tbl_next(table, &bucket, &obj);
1159 		if (!flow)
1160 			break;
1161 
1162 		if (ovs_flow_cmd_fill_info(flow, dp, skb,
1163 					   NETLINK_CB(cb->skb).portid,
1164 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1165 					   OVS_FLOW_CMD_NEW) < 0)
1166 			break;
1167 
1168 		cb->args[0] = bucket;
1169 		cb->args[1] = obj;
1170 	}
1171 	return skb->len;
1172 }
1173 
1174 static struct genl_ops dp_flow_genl_ops[] = {
1175 	{ .cmd = OVS_FLOW_CMD_NEW,
1176 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1177 	  .policy = flow_policy,
1178 	  .doit = ovs_flow_cmd_new_or_set
1179 	},
1180 	{ .cmd = OVS_FLOW_CMD_DEL,
1181 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1182 	  .policy = flow_policy,
1183 	  .doit = ovs_flow_cmd_del
1184 	},
1185 	{ .cmd = OVS_FLOW_CMD_GET,
1186 	  .flags = 0,		    /* OK for unprivileged users. */
1187 	  .policy = flow_policy,
1188 	  .doit = ovs_flow_cmd_get,
1189 	  .dumpit = ovs_flow_cmd_dump
1190 	},
1191 	{ .cmd = OVS_FLOW_CMD_SET,
1192 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1193 	  .policy = flow_policy,
1194 	  .doit = ovs_flow_cmd_new_or_set,
1195 	},
1196 };
1197 
1198 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1199 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1200 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1201 };
1202 
1203 static struct genl_family dp_datapath_genl_family = {
1204 	.id = GENL_ID_GENERATE,
1205 	.hdrsize = sizeof(struct ovs_header),
1206 	.name = OVS_DATAPATH_FAMILY,
1207 	.version = OVS_DATAPATH_VERSION,
1208 	.maxattr = OVS_DP_ATTR_MAX,
1209 	.netnsok = true
1210 };
1211 
1212 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1213 	.name = OVS_DATAPATH_MCGROUP
1214 };
1215 
1216 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1217 				u32 portid, u32 seq, u32 flags, u8 cmd)
1218 {
1219 	struct ovs_header *ovs_header;
1220 	struct ovs_dp_stats dp_stats;
1221 	int err;
1222 
1223 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1224 				   flags, cmd);
1225 	if (!ovs_header)
1226 		goto error;
1227 
1228 	ovs_header->dp_ifindex = get_dpifindex(dp);
1229 
1230 	rcu_read_lock();
1231 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1232 	rcu_read_unlock();
1233 	if (err)
1234 		goto nla_put_failure;
1235 
1236 	get_dp_stats(dp, &dp_stats);
1237 	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats), &dp_stats))
1238 		goto nla_put_failure;
1239 
1240 	return genlmsg_end(skb, ovs_header);
1241 
1242 nla_put_failure:
1243 	genlmsg_cancel(skb, ovs_header);
1244 error:
1245 	return -EMSGSIZE;
1246 }
1247 
1248 static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, u32 portid,
1249 					     u32 seq, u8 cmd)
1250 {
1251 	struct sk_buff *skb;
1252 	int retval;
1253 
1254 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1255 	if (!skb)
1256 		return ERR_PTR(-ENOMEM);
1257 
1258 	retval = ovs_dp_cmd_fill_info(dp, skb, portid, seq, 0, cmd);
1259 	if (retval < 0) {
1260 		kfree_skb(skb);
1261 		return ERR_PTR(retval);
1262 	}
1263 	return skb;
1264 }
1265 
1266 /* Called with genl_mutex and optionally with RTNL lock also. */
1267 static struct datapath *lookup_datapath(struct net *net,
1268 					struct ovs_header *ovs_header,
1269 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1270 {
1271 	struct datapath *dp;
1272 
1273 	if (!a[OVS_DP_ATTR_NAME])
1274 		dp = get_dp(net, ovs_header->dp_ifindex);
1275 	else {
1276 		struct vport *vport;
1277 
1278 		rcu_read_lock();
1279 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1280 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1281 		rcu_read_unlock();
1282 	}
1283 	return dp ? dp : ERR_PTR(-ENODEV);
1284 }
1285 
1286 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1287 {
1288 	struct nlattr **a = info->attrs;
1289 	struct vport_parms parms;
1290 	struct sk_buff *reply;
1291 	struct datapath *dp;
1292 	struct vport *vport;
1293 	struct ovs_net *ovs_net;
1294 	int err, i;
1295 
1296 	err = -EINVAL;
1297 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1298 		goto err;
1299 
1300 	rtnl_lock();
1301 
1302 	err = -ENOMEM;
1303 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1304 	if (dp == NULL)
1305 		goto err_unlock_rtnl;
1306 
1307 	ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1308 
1309 	/* Allocate table. */
1310 	err = -ENOMEM;
1311 	rcu_assign_pointer(dp->table, ovs_flow_tbl_alloc(TBL_MIN_BUCKETS));
1312 	if (!dp->table)
1313 		goto err_free_dp;
1314 
1315 	dp->stats_percpu = alloc_percpu(struct dp_stats_percpu);
1316 	if (!dp->stats_percpu) {
1317 		err = -ENOMEM;
1318 		goto err_destroy_table;
1319 	}
1320 
1321 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1322 			GFP_KERNEL);
1323 	if (!dp->ports) {
1324 		err = -ENOMEM;
1325 		goto err_destroy_percpu;
1326 	}
1327 
1328 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1329 		INIT_HLIST_HEAD(&dp->ports[i]);
1330 
1331 	/* Set up our datapath device. */
1332 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1333 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1334 	parms.options = NULL;
1335 	parms.dp = dp;
1336 	parms.port_no = OVSP_LOCAL;
1337 	parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1338 
1339 	vport = new_vport(&parms);
1340 	if (IS_ERR(vport)) {
1341 		err = PTR_ERR(vport);
1342 		if (err == -EBUSY)
1343 			err = -EEXIST;
1344 
1345 		goto err_destroy_ports_array;
1346 	}
1347 
1348 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1349 				      info->snd_seq, OVS_DP_CMD_NEW);
1350 	err = PTR_ERR(reply);
1351 	if (IS_ERR(reply))
1352 		goto err_destroy_local_port;
1353 
1354 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1355 	list_add_tail(&dp->list_node, &ovs_net->dps);
1356 	rtnl_unlock();
1357 
1358 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1359 		    ovs_dp_datapath_multicast_group.id, info->nlhdr,
1360 		    GFP_KERNEL);
1361 	return 0;
1362 
1363 err_destroy_local_port:
1364 	ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1365 err_destroy_ports_array:
1366 	kfree(dp->ports);
1367 err_destroy_percpu:
1368 	free_percpu(dp->stats_percpu);
1369 err_destroy_table:
1370 	ovs_flow_tbl_destroy(genl_dereference(dp->table));
1371 err_free_dp:
1372 	release_net(ovs_dp_get_net(dp));
1373 	kfree(dp);
1374 err_unlock_rtnl:
1375 	rtnl_unlock();
1376 err:
1377 	return err;
1378 }
1379 
1380 /* Called with genl_mutex. */
1381 static void __dp_destroy(struct datapath *dp)
1382 {
1383 	int i;
1384 
1385 	rtnl_lock();
1386 
1387 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1388 		struct vport *vport;
1389 		struct hlist_node *node, *n;
1390 
1391 		hlist_for_each_entry_safe(vport, node, n, &dp->ports[i], dp_hash_node)
1392 			if (vport->port_no != OVSP_LOCAL)
1393 				ovs_dp_detach_port(vport);
1394 	}
1395 
1396 	list_del(&dp->list_node);
1397 	ovs_dp_detach_port(ovs_vport_rtnl(dp, OVSP_LOCAL));
1398 
1399 	/* rtnl_unlock() will wait until all the references to devices that
1400 	 * are pending unregistration have been dropped.  We do it here to
1401 	 * ensure that any internal devices (which contain DP pointers) are
1402 	 * fully destroyed before freeing the datapath.
1403 	 */
1404 	rtnl_unlock();
1405 
1406 	call_rcu(&dp->rcu, destroy_dp_rcu);
1407 }
1408 
1409 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1410 {
1411 	struct sk_buff *reply;
1412 	struct datapath *dp;
1413 	int err;
1414 
1415 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1416 	err = PTR_ERR(dp);
1417 	if (IS_ERR(dp))
1418 		return err;
1419 
1420 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1421 				      info->snd_seq, OVS_DP_CMD_DEL);
1422 	err = PTR_ERR(reply);
1423 	if (IS_ERR(reply))
1424 		return err;
1425 
1426 	__dp_destroy(dp);
1427 
1428 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1429 		    ovs_dp_datapath_multicast_group.id, info->nlhdr,
1430 		    GFP_KERNEL);
1431 
1432 	return 0;
1433 }
1434 
1435 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1436 {
1437 	struct sk_buff *reply;
1438 	struct datapath *dp;
1439 	int err;
1440 
1441 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1442 	if (IS_ERR(dp))
1443 		return PTR_ERR(dp);
1444 
1445 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1446 				      info->snd_seq, OVS_DP_CMD_NEW);
1447 	if (IS_ERR(reply)) {
1448 		err = PTR_ERR(reply);
1449 		netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1450 				ovs_dp_datapath_multicast_group.id, err);
1451 		return 0;
1452 	}
1453 
1454 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1455 		    ovs_dp_datapath_multicast_group.id, info->nlhdr,
1456 		    GFP_KERNEL);
1457 
1458 	return 0;
1459 }
1460 
1461 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1462 {
1463 	struct sk_buff *reply;
1464 	struct datapath *dp;
1465 
1466 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1467 	if (IS_ERR(dp))
1468 		return PTR_ERR(dp);
1469 
1470 	reply = ovs_dp_cmd_build_info(dp, info->snd_portid,
1471 				      info->snd_seq, OVS_DP_CMD_NEW);
1472 	if (IS_ERR(reply))
1473 		return PTR_ERR(reply);
1474 
1475 	return genlmsg_reply(reply, info);
1476 }
1477 
1478 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1479 {
1480 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1481 	struct datapath *dp;
1482 	int skip = cb->args[0];
1483 	int i = 0;
1484 
1485 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1486 		if (i >= skip &&
1487 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1488 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1489 					 OVS_DP_CMD_NEW) < 0)
1490 			break;
1491 		i++;
1492 	}
1493 
1494 	cb->args[0] = i;
1495 
1496 	return skb->len;
1497 }
1498 
1499 static struct genl_ops dp_datapath_genl_ops[] = {
1500 	{ .cmd = OVS_DP_CMD_NEW,
1501 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1502 	  .policy = datapath_policy,
1503 	  .doit = ovs_dp_cmd_new
1504 	},
1505 	{ .cmd = OVS_DP_CMD_DEL,
1506 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1507 	  .policy = datapath_policy,
1508 	  .doit = ovs_dp_cmd_del
1509 	},
1510 	{ .cmd = OVS_DP_CMD_GET,
1511 	  .flags = 0,		    /* OK for unprivileged users. */
1512 	  .policy = datapath_policy,
1513 	  .doit = ovs_dp_cmd_get,
1514 	  .dumpit = ovs_dp_cmd_dump
1515 	},
1516 	{ .cmd = OVS_DP_CMD_SET,
1517 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1518 	  .policy = datapath_policy,
1519 	  .doit = ovs_dp_cmd_set,
1520 	},
1521 };
1522 
1523 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1524 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1525 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1526 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1527 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1528 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1529 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1530 };
1531 
1532 static struct genl_family dp_vport_genl_family = {
1533 	.id = GENL_ID_GENERATE,
1534 	.hdrsize = sizeof(struct ovs_header),
1535 	.name = OVS_VPORT_FAMILY,
1536 	.version = OVS_VPORT_VERSION,
1537 	.maxattr = OVS_VPORT_ATTR_MAX,
1538 	.netnsok = true
1539 };
1540 
1541 struct genl_multicast_group ovs_dp_vport_multicast_group = {
1542 	.name = OVS_VPORT_MCGROUP
1543 };
1544 
1545 /* Called with RTNL lock or RCU read lock. */
1546 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1547 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1548 {
1549 	struct ovs_header *ovs_header;
1550 	struct ovs_vport_stats vport_stats;
1551 	int err;
1552 
1553 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1554 				 flags, cmd);
1555 	if (!ovs_header)
1556 		return -EMSGSIZE;
1557 
1558 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1559 
1560 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1561 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1562 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1563 	    nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1564 		goto nla_put_failure;
1565 
1566 	ovs_vport_get_stats(vport, &vport_stats);
1567 	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1568 		    &vport_stats))
1569 		goto nla_put_failure;
1570 
1571 	err = ovs_vport_get_options(vport, skb);
1572 	if (err == -EMSGSIZE)
1573 		goto error;
1574 
1575 	return genlmsg_end(skb, ovs_header);
1576 
1577 nla_put_failure:
1578 	err = -EMSGSIZE;
1579 error:
1580 	genlmsg_cancel(skb, ovs_header);
1581 	return err;
1582 }
1583 
1584 /* Called with RTNL lock or RCU read lock. */
1585 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1586 					 u32 seq, u8 cmd)
1587 {
1588 	struct sk_buff *skb;
1589 	int retval;
1590 
1591 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1592 	if (!skb)
1593 		return ERR_PTR(-ENOMEM);
1594 
1595 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1596 	if (retval < 0) {
1597 		kfree_skb(skb);
1598 		return ERR_PTR(retval);
1599 	}
1600 	return skb;
1601 }
1602 
1603 /* Called with RTNL lock or RCU read lock. */
1604 static struct vport *lookup_vport(struct net *net,
1605 				  struct ovs_header *ovs_header,
1606 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1607 {
1608 	struct datapath *dp;
1609 	struct vport *vport;
1610 
1611 	if (a[OVS_VPORT_ATTR_NAME]) {
1612 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1613 		if (!vport)
1614 			return ERR_PTR(-ENODEV);
1615 		if (ovs_header->dp_ifindex &&
1616 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1617 			return ERR_PTR(-ENODEV);
1618 		return vport;
1619 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1620 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1621 
1622 		if (port_no >= DP_MAX_PORTS)
1623 			return ERR_PTR(-EFBIG);
1624 
1625 		dp = get_dp(net, ovs_header->dp_ifindex);
1626 		if (!dp)
1627 			return ERR_PTR(-ENODEV);
1628 
1629 		vport = ovs_vport_rtnl_rcu(dp, port_no);
1630 		if (!vport)
1631 			return ERR_PTR(-ENOENT);
1632 		return vport;
1633 	} else
1634 		return ERR_PTR(-EINVAL);
1635 }
1636 
1637 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1638 {
1639 	struct nlattr **a = info->attrs;
1640 	struct ovs_header *ovs_header = info->userhdr;
1641 	struct vport_parms parms;
1642 	struct sk_buff *reply;
1643 	struct vport *vport;
1644 	struct datapath *dp;
1645 	u32 port_no;
1646 	int err;
1647 
1648 	err = -EINVAL;
1649 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1650 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1651 		goto exit;
1652 
1653 	rtnl_lock();
1654 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1655 	err = -ENODEV;
1656 	if (!dp)
1657 		goto exit_unlock;
1658 
1659 	if (a[OVS_VPORT_ATTR_PORT_NO]) {
1660 		port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1661 
1662 		err = -EFBIG;
1663 		if (port_no >= DP_MAX_PORTS)
1664 			goto exit_unlock;
1665 
1666 		vport = ovs_vport_rtnl_rcu(dp, port_no);
1667 		err = -EBUSY;
1668 		if (vport)
1669 			goto exit_unlock;
1670 	} else {
1671 		for (port_no = 1; ; port_no++) {
1672 			if (port_no >= DP_MAX_PORTS) {
1673 				err = -EFBIG;
1674 				goto exit_unlock;
1675 			}
1676 			vport = ovs_vport_rtnl(dp, port_no);
1677 			if (!vport)
1678 				break;
1679 		}
1680 	}
1681 
1682 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1683 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1684 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1685 	parms.dp = dp;
1686 	parms.port_no = port_no;
1687 	parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1688 
1689 	vport = new_vport(&parms);
1690 	err = PTR_ERR(vport);
1691 	if (IS_ERR(vport))
1692 		goto exit_unlock;
1693 
1694 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1695 					 OVS_VPORT_CMD_NEW);
1696 	if (IS_ERR(reply)) {
1697 		err = PTR_ERR(reply);
1698 		ovs_dp_detach_port(vport);
1699 		goto exit_unlock;
1700 	}
1701 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1702 		    ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1703 
1704 exit_unlock:
1705 	rtnl_unlock();
1706 exit:
1707 	return err;
1708 }
1709 
1710 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1711 {
1712 	struct nlattr **a = info->attrs;
1713 	struct sk_buff *reply;
1714 	struct vport *vport;
1715 	int err;
1716 
1717 	rtnl_lock();
1718 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1719 	err = PTR_ERR(vport);
1720 	if (IS_ERR(vport))
1721 		goto exit_unlock;
1722 
1723 	err = 0;
1724 	if (a[OVS_VPORT_ATTR_TYPE] &&
1725 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type)
1726 		err = -EINVAL;
1727 
1728 	if (!err && a[OVS_VPORT_ATTR_OPTIONS])
1729 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1730 	if (err)
1731 		goto exit_unlock;
1732 	if (a[OVS_VPORT_ATTR_UPCALL_PID])
1733 		vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1734 
1735 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1736 					 OVS_VPORT_CMD_NEW);
1737 	if (IS_ERR(reply)) {
1738 		netlink_set_err(sock_net(skb->sk)->genl_sock, 0,
1739 				ovs_dp_vport_multicast_group.id, PTR_ERR(reply));
1740 		goto exit_unlock;
1741 	}
1742 
1743 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1744 		    ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1745 
1746 exit_unlock:
1747 	rtnl_unlock();
1748 	return err;
1749 }
1750 
1751 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1752 {
1753 	struct nlattr **a = info->attrs;
1754 	struct sk_buff *reply;
1755 	struct vport *vport;
1756 	int err;
1757 
1758 	rtnl_lock();
1759 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1760 	err = PTR_ERR(vport);
1761 	if (IS_ERR(vport))
1762 		goto exit_unlock;
1763 
1764 	if (vport->port_no == OVSP_LOCAL) {
1765 		err = -EINVAL;
1766 		goto exit_unlock;
1767 	}
1768 
1769 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1770 					 OVS_VPORT_CMD_DEL);
1771 	err = PTR_ERR(reply);
1772 	if (IS_ERR(reply))
1773 		goto exit_unlock;
1774 
1775 	ovs_dp_detach_port(vport);
1776 
1777 	genl_notify(reply, genl_info_net(info), info->snd_portid,
1778 		    ovs_dp_vport_multicast_group.id, info->nlhdr, GFP_KERNEL);
1779 
1780 exit_unlock:
1781 	rtnl_unlock();
1782 	return err;
1783 }
1784 
1785 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1786 {
1787 	struct nlattr **a = info->attrs;
1788 	struct ovs_header *ovs_header = info->userhdr;
1789 	struct sk_buff *reply;
1790 	struct vport *vport;
1791 	int err;
1792 
1793 	rcu_read_lock();
1794 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1795 	err = PTR_ERR(vport);
1796 	if (IS_ERR(vport))
1797 		goto exit_unlock;
1798 
1799 	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1800 					 OVS_VPORT_CMD_NEW);
1801 	err = PTR_ERR(reply);
1802 	if (IS_ERR(reply))
1803 		goto exit_unlock;
1804 
1805 	rcu_read_unlock();
1806 
1807 	return genlmsg_reply(reply, info);
1808 
1809 exit_unlock:
1810 	rcu_read_unlock();
1811 	return err;
1812 }
1813 
1814 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1815 {
1816 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1817 	struct datapath *dp;
1818 	int bucket = cb->args[0], skip = cb->args[1];
1819 	int i, j = 0;
1820 
1821 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1822 	if (!dp)
1823 		return -ENODEV;
1824 
1825 	rcu_read_lock();
1826 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1827 		struct vport *vport;
1828 		struct hlist_node *n;
1829 
1830 		j = 0;
1831 		hlist_for_each_entry_rcu(vport, n, &dp->ports[i], dp_hash_node) {
1832 			if (j >= skip &&
1833 			    ovs_vport_cmd_fill_info(vport, skb,
1834 						    NETLINK_CB(cb->skb).portid,
1835 						    cb->nlh->nlmsg_seq,
1836 						    NLM_F_MULTI,
1837 						    OVS_VPORT_CMD_NEW) < 0)
1838 				goto out;
1839 
1840 			j++;
1841 		}
1842 		skip = 0;
1843 	}
1844 out:
1845 	rcu_read_unlock();
1846 
1847 	cb->args[0] = i;
1848 	cb->args[1] = j;
1849 
1850 	return skb->len;
1851 }
1852 
1853 static struct genl_ops dp_vport_genl_ops[] = {
1854 	{ .cmd = OVS_VPORT_CMD_NEW,
1855 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1856 	  .policy = vport_policy,
1857 	  .doit = ovs_vport_cmd_new
1858 	},
1859 	{ .cmd = OVS_VPORT_CMD_DEL,
1860 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1861 	  .policy = vport_policy,
1862 	  .doit = ovs_vport_cmd_del
1863 	},
1864 	{ .cmd = OVS_VPORT_CMD_GET,
1865 	  .flags = 0,		    /* OK for unprivileged users. */
1866 	  .policy = vport_policy,
1867 	  .doit = ovs_vport_cmd_get,
1868 	  .dumpit = ovs_vport_cmd_dump
1869 	},
1870 	{ .cmd = OVS_VPORT_CMD_SET,
1871 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1872 	  .policy = vport_policy,
1873 	  .doit = ovs_vport_cmd_set,
1874 	},
1875 };
1876 
1877 struct genl_family_and_ops {
1878 	struct genl_family *family;
1879 	struct genl_ops *ops;
1880 	int n_ops;
1881 	struct genl_multicast_group *group;
1882 };
1883 
1884 static const struct genl_family_and_ops dp_genl_families[] = {
1885 	{ &dp_datapath_genl_family,
1886 	  dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1887 	  &ovs_dp_datapath_multicast_group },
1888 	{ &dp_vport_genl_family,
1889 	  dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1890 	  &ovs_dp_vport_multicast_group },
1891 	{ &dp_flow_genl_family,
1892 	  dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1893 	  &ovs_dp_flow_multicast_group },
1894 	{ &dp_packet_genl_family,
1895 	  dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1896 	  NULL },
1897 };
1898 
1899 static void dp_unregister_genl(int n_families)
1900 {
1901 	int i;
1902 
1903 	for (i = 0; i < n_families; i++)
1904 		genl_unregister_family(dp_genl_families[i].family);
1905 }
1906 
1907 static int dp_register_genl(void)
1908 {
1909 	int n_registered;
1910 	int err;
1911 	int i;
1912 
1913 	n_registered = 0;
1914 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1915 		const struct genl_family_and_ops *f = &dp_genl_families[i];
1916 
1917 		err = genl_register_family_with_ops(f->family, f->ops,
1918 						    f->n_ops);
1919 		if (err)
1920 			goto error;
1921 		n_registered++;
1922 
1923 		if (f->group) {
1924 			err = genl_register_mc_group(f->family, f->group);
1925 			if (err)
1926 				goto error;
1927 		}
1928 	}
1929 
1930 	return 0;
1931 
1932 error:
1933 	dp_unregister_genl(n_registered);
1934 	return err;
1935 }
1936 
1937 static void rehash_flow_table(struct work_struct *work)
1938 {
1939 	struct datapath *dp;
1940 	struct net *net;
1941 
1942 	genl_lock();
1943 	rtnl_lock();
1944 	for_each_net(net) {
1945 		struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1946 
1947 		list_for_each_entry(dp, &ovs_net->dps, list_node) {
1948 			struct flow_table *old_table = genl_dereference(dp->table);
1949 			struct flow_table *new_table;
1950 
1951 			new_table = ovs_flow_tbl_rehash(old_table);
1952 			if (!IS_ERR(new_table)) {
1953 				rcu_assign_pointer(dp->table, new_table);
1954 				ovs_flow_tbl_deferred_destroy(old_table);
1955 			}
1956 		}
1957 	}
1958 	rtnl_unlock();
1959 	genl_unlock();
1960 
1961 	schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
1962 }
1963 
1964 static int __net_init ovs_init_net(struct net *net)
1965 {
1966 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1967 
1968 	INIT_LIST_HEAD(&ovs_net->dps);
1969 	return 0;
1970 }
1971 
1972 static void __net_exit ovs_exit_net(struct net *net)
1973 {
1974 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1975 	struct datapath *dp, *dp_next;
1976 
1977 	genl_lock();
1978 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1979 		__dp_destroy(dp);
1980 	genl_unlock();
1981 }
1982 
1983 static struct pernet_operations ovs_net_ops = {
1984 	.init = ovs_init_net,
1985 	.exit = ovs_exit_net,
1986 	.id   = &ovs_net_id,
1987 	.size = sizeof(struct ovs_net),
1988 };
1989 
1990 static int __init dp_init(void)
1991 {
1992 	struct sk_buff *dummy_skb;
1993 	int err;
1994 
1995 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof(dummy_skb->cb));
1996 
1997 	pr_info("Open vSwitch switching datapath\n");
1998 
1999 	err = ovs_flow_init();
2000 	if (err)
2001 		goto error;
2002 
2003 	err = ovs_vport_init();
2004 	if (err)
2005 		goto error_flow_exit;
2006 
2007 	err = register_pernet_device(&ovs_net_ops);
2008 	if (err)
2009 		goto error_vport_exit;
2010 
2011 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2012 	if (err)
2013 		goto error_netns_exit;
2014 
2015 	err = dp_register_genl();
2016 	if (err < 0)
2017 		goto error_unreg_notifier;
2018 
2019 	schedule_delayed_work(&rehash_flow_wq, REHASH_FLOW_INTERVAL);
2020 
2021 	return 0;
2022 
2023 error_unreg_notifier:
2024 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2025 error_netns_exit:
2026 	unregister_pernet_device(&ovs_net_ops);
2027 error_vport_exit:
2028 	ovs_vport_exit();
2029 error_flow_exit:
2030 	ovs_flow_exit();
2031 error:
2032 	return err;
2033 }
2034 
2035 static void dp_cleanup(void)
2036 {
2037 	cancel_delayed_work_sync(&rehash_flow_wq);
2038 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2039 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2040 	unregister_pernet_device(&ovs_net_ops);
2041 	rcu_barrier();
2042 	ovs_vport_exit();
2043 	ovs_flow_exit();
2044 }
2045 
2046 module_init(dp_init);
2047 module_exit(dp_cleanup);
2048 
2049 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2050 MODULE_LICENSE("GPL");
2051