xref: /openbmc/linux/net/openvswitch/datapath.c (revision d2999e1b)
1 /*
2  * Copyright (c) 2007-2013 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/genetlink.h>
51 #include <net/genetlink.h>
52 #include <net/genetlink.h>
53 #include <net/net_namespace.h>
54 #include <net/netns/generic.h>
55 
56 #include "datapath.h"
57 #include "flow.h"
58 #include "flow_table.h"
59 #include "flow_netlink.h"
60 #include "vport-internal_dev.h"
61 #include "vport-netdev.h"
62 
63 int ovs_net_id __read_mostly;
64 
65 static struct genl_family dp_packet_genl_family;
66 static struct genl_family dp_flow_genl_family;
67 static struct genl_family dp_datapath_genl_family;
68 
69 static struct genl_multicast_group ovs_dp_flow_multicast_group = {
70 	.name = OVS_FLOW_MCGROUP
71 };
72 
73 static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
74 	.name = OVS_DATAPATH_MCGROUP
75 };
76 
77 struct genl_multicast_group ovs_dp_vport_multicast_group = {
78 	.name = OVS_VPORT_MCGROUP
79 };
80 
81 /* Check if need to build a reply message.
82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
83 static bool ovs_must_notify(struct genl_info *info,
84 			    const struct genl_multicast_group *grp)
85 {
86 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
87 		netlink_has_listeners(genl_info_net(info)->genl_sock, 0);
88 }
89 
90 static void ovs_notify(struct genl_family *family,
91 		       struct sk_buff *skb, struct genl_info *info)
92 {
93 	genl_notify(family, skb, genl_info_net(info), info->snd_portid,
94 		    0, info->nlhdr, GFP_KERNEL);
95 }
96 
97 /**
98  * DOC: Locking:
99  *
100  * All writes e.g. Writes to device state (add/remove datapath, port, set
101  * operations on vports, etc.), Writes to other state (flow table
102  * modifications, set miscellaneous datapath parameters, etc.) are protected
103  * by ovs_lock.
104  *
105  * Reads are protected by RCU.
106  *
107  * There are a few special cases (mostly stats) that have their own
108  * synchronization but they nest under all of above and don't interact with
109  * each other.
110  *
111  * The RTNL lock nests inside ovs_mutex.
112  */
113 
114 static DEFINE_MUTEX(ovs_mutex);
115 
116 void ovs_lock(void)
117 {
118 	mutex_lock(&ovs_mutex);
119 }
120 
121 void ovs_unlock(void)
122 {
123 	mutex_unlock(&ovs_mutex);
124 }
125 
126 #ifdef CONFIG_LOCKDEP
127 int lockdep_ovsl_is_held(void)
128 {
129 	if (debug_locks)
130 		return lockdep_is_held(&ovs_mutex);
131 	else
132 		return 1;
133 }
134 #endif
135 
136 static struct vport *new_vport(const struct vport_parms *);
137 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
138 			     const struct dp_upcall_info *);
139 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
140 				  const struct dp_upcall_info *);
141 
142 /* Must be called with rcu_read_lock or ovs_mutex. */
143 static struct datapath *get_dp(struct net *net, int dp_ifindex)
144 {
145 	struct datapath *dp = NULL;
146 	struct net_device *dev;
147 
148 	rcu_read_lock();
149 	dev = dev_get_by_index_rcu(net, dp_ifindex);
150 	if (dev) {
151 		struct vport *vport = ovs_internal_dev_get_vport(dev);
152 		if (vport)
153 			dp = vport->dp;
154 	}
155 	rcu_read_unlock();
156 
157 	return dp;
158 }
159 
160 /* Must be called with rcu_read_lock or ovs_mutex. */
161 static const char *ovs_dp_name(const struct datapath *dp)
162 {
163 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
164 	return vport->ops->get_name(vport);
165 }
166 
167 static int get_dpifindex(struct datapath *dp)
168 {
169 	struct vport *local;
170 	int ifindex;
171 
172 	rcu_read_lock();
173 
174 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
175 	if (local)
176 		ifindex = netdev_vport_priv(local)->dev->ifindex;
177 	else
178 		ifindex = 0;
179 
180 	rcu_read_unlock();
181 
182 	return ifindex;
183 }
184 
185 static void destroy_dp_rcu(struct rcu_head *rcu)
186 {
187 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
188 
189 	free_percpu(dp->stats_percpu);
190 	release_net(ovs_dp_get_net(dp));
191 	kfree(dp->ports);
192 	kfree(dp);
193 }
194 
195 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
196 					    u16 port_no)
197 {
198 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
199 }
200 
201 /* Called with ovs_mutex or RCU read lock. */
202 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
203 {
204 	struct vport *vport;
205 	struct hlist_head *head;
206 
207 	head = vport_hash_bucket(dp, port_no);
208 	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
209 		if (vport->port_no == port_no)
210 			return vport;
211 	}
212 	return NULL;
213 }
214 
215 /* Called with ovs_mutex. */
216 static struct vport *new_vport(const struct vport_parms *parms)
217 {
218 	struct vport *vport;
219 
220 	vport = ovs_vport_add(parms);
221 	if (!IS_ERR(vport)) {
222 		struct datapath *dp = parms->dp;
223 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
224 
225 		hlist_add_head_rcu(&vport->dp_hash_node, head);
226 	}
227 	return vport;
228 }
229 
230 void ovs_dp_detach_port(struct vport *p)
231 {
232 	ASSERT_OVSL();
233 
234 	/* First drop references to device. */
235 	hlist_del_rcu(&p->dp_hash_node);
236 
237 	/* Then destroy it. */
238 	ovs_vport_del(p);
239 }
240 
241 /* Must be called with rcu_read_lock. */
242 void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
243 {
244 	struct datapath *dp = p->dp;
245 	struct sw_flow *flow;
246 	struct dp_stats_percpu *stats;
247 	struct sw_flow_key key;
248 	u64 *stats_counter;
249 	u32 n_mask_hit;
250 	int error;
251 
252 	stats = this_cpu_ptr(dp->stats_percpu);
253 
254 	/* Extract flow from 'skb' into 'key'. */
255 	error = ovs_flow_extract(skb, p->port_no, &key);
256 	if (unlikely(error)) {
257 		kfree_skb(skb);
258 		return;
259 	}
260 
261 	/* Look up flow. */
262 	flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
263 	if (unlikely(!flow)) {
264 		struct dp_upcall_info upcall;
265 
266 		upcall.cmd = OVS_PACKET_CMD_MISS;
267 		upcall.key = &key;
268 		upcall.userdata = NULL;
269 		upcall.portid = p->upcall_portid;
270 		ovs_dp_upcall(dp, skb, &upcall);
271 		consume_skb(skb);
272 		stats_counter = &stats->n_missed;
273 		goto out;
274 	}
275 
276 	OVS_CB(skb)->flow = flow;
277 	OVS_CB(skb)->pkt_key = &key;
278 
279 	ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
280 	ovs_execute_actions(dp, skb);
281 	stats_counter = &stats->n_hit;
282 
283 out:
284 	/* Update datapath statistics. */
285 	u64_stats_update_begin(&stats->syncp);
286 	(*stats_counter)++;
287 	stats->n_mask_hit += n_mask_hit;
288 	u64_stats_update_end(&stats->syncp);
289 }
290 
291 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
292 		  const struct dp_upcall_info *upcall_info)
293 {
294 	struct dp_stats_percpu *stats;
295 	int err;
296 
297 	if (upcall_info->portid == 0) {
298 		err = -ENOTCONN;
299 		goto err;
300 	}
301 
302 	if (!skb_is_gso(skb))
303 		err = queue_userspace_packet(dp, skb, upcall_info);
304 	else
305 		err = queue_gso_packets(dp, skb, upcall_info);
306 	if (err)
307 		goto err;
308 
309 	return 0;
310 
311 err:
312 	stats = this_cpu_ptr(dp->stats_percpu);
313 
314 	u64_stats_update_begin(&stats->syncp);
315 	stats->n_lost++;
316 	u64_stats_update_end(&stats->syncp);
317 
318 	return err;
319 }
320 
321 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
322 			     const struct dp_upcall_info *upcall_info)
323 {
324 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
325 	struct dp_upcall_info later_info;
326 	struct sw_flow_key later_key;
327 	struct sk_buff *segs, *nskb;
328 	int err;
329 
330 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
331 	if (IS_ERR(segs))
332 		return PTR_ERR(segs);
333 
334 	/* Queue all of the segments. */
335 	skb = segs;
336 	do {
337 		err = queue_userspace_packet(dp, skb, upcall_info);
338 		if (err)
339 			break;
340 
341 		if (skb == segs && gso_type & SKB_GSO_UDP) {
342 			/* The initial flow key extracted by ovs_flow_extract()
343 			 * in this case is for a first fragment, so we need to
344 			 * properly mark later fragments.
345 			 */
346 			later_key = *upcall_info->key;
347 			later_key.ip.frag = OVS_FRAG_TYPE_LATER;
348 
349 			later_info = *upcall_info;
350 			later_info.key = &later_key;
351 			upcall_info = &later_info;
352 		}
353 	} while ((skb = skb->next));
354 
355 	/* Free all of the segments. */
356 	skb = segs;
357 	do {
358 		nskb = skb->next;
359 		if (err)
360 			kfree_skb(skb);
361 		else
362 			consume_skb(skb);
363 	} while ((skb = nskb));
364 	return err;
365 }
366 
367 static size_t key_attr_size(void)
368 {
369 	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
370 		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
371 		  + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
372 		  + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
373 		  + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
374 		  + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
375 		  + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
376 		  + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
377 		  + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
378 		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
379 		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
380 		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
381 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
382 		+ nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
383 		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
384 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
385 		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
386 		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
387 		+ nla_total_size(28); /* OVS_KEY_ATTR_ND */
388 }
389 
390 static size_t upcall_msg_size(const struct nlattr *userdata,
391 			      unsigned int hdrlen)
392 {
393 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
394 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
395 		+ nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
396 
397 	/* OVS_PACKET_ATTR_USERDATA */
398 	if (userdata)
399 		size += NLA_ALIGN(userdata->nla_len);
400 
401 	return size;
402 }
403 
404 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
405 				  const struct dp_upcall_info *upcall_info)
406 {
407 	struct ovs_header *upcall;
408 	struct sk_buff *nskb = NULL;
409 	struct sk_buff *user_skb; /* to be queued to userspace */
410 	struct nlattr *nla;
411 	struct genl_info info = {
412 		.dst_sk = ovs_dp_get_net(dp)->genl_sock,
413 		.snd_portid = upcall_info->portid,
414 	};
415 	size_t len;
416 	unsigned int hlen;
417 	int err, dp_ifindex;
418 
419 	dp_ifindex = get_dpifindex(dp);
420 	if (!dp_ifindex)
421 		return -ENODEV;
422 
423 	if (vlan_tx_tag_present(skb)) {
424 		nskb = skb_clone(skb, GFP_ATOMIC);
425 		if (!nskb)
426 			return -ENOMEM;
427 
428 		nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
429 		if (!nskb)
430 			return -ENOMEM;
431 
432 		nskb->vlan_tci = 0;
433 		skb = nskb;
434 	}
435 
436 	if (nla_attr_size(skb->len) > USHRT_MAX) {
437 		err = -EFBIG;
438 		goto out;
439 	}
440 
441 	/* Complete checksum if needed */
442 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
443 	    (err = skb_checksum_help(skb)))
444 		goto out;
445 
446 	/* Older versions of OVS user space enforce alignment of the last
447 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
448 	 * padding logic. Only perform zerocopy if padding is not required.
449 	 */
450 	if (dp->user_features & OVS_DP_F_UNALIGNED)
451 		hlen = skb_zerocopy_headlen(skb);
452 	else
453 		hlen = skb->len;
454 
455 	len = upcall_msg_size(upcall_info->userdata, hlen);
456 	user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
457 	if (!user_skb) {
458 		err = -ENOMEM;
459 		goto out;
460 	}
461 
462 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
463 			     0, upcall_info->cmd);
464 	upcall->dp_ifindex = dp_ifindex;
465 
466 	nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
467 	ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
468 	nla_nest_end(user_skb, nla);
469 
470 	if (upcall_info->userdata)
471 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
472 			  nla_len(upcall_info->userdata),
473 			  nla_data(upcall_info->userdata));
474 
475 	/* Only reserve room for attribute header, packet data is added
476 	 * in skb_zerocopy() */
477 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
478 		err = -ENOBUFS;
479 		goto out;
480 	}
481 	nla->nla_len = nla_attr_size(skb->len);
482 
483 	err = skb_zerocopy(user_skb, skb, skb->len, hlen);
484 	if (err)
485 		goto out;
486 
487 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
488 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
489 		size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
490 
491 		if (plen > 0)
492 			memset(skb_put(user_skb, plen), 0, plen);
493 	}
494 
495 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
496 
497 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
498 out:
499 	if (err)
500 		skb_tx_error(skb);
501 	kfree_skb(nskb);
502 	return err;
503 }
504 
505 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
506 {
507 	struct ovs_header *ovs_header = info->userhdr;
508 	struct nlattr **a = info->attrs;
509 	struct sw_flow_actions *acts;
510 	struct sk_buff *packet;
511 	struct sw_flow *flow;
512 	struct datapath *dp;
513 	struct ethhdr *eth;
514 	int len;
515 	int err;
516 
517 	err = -EINVAL;
518 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
519 	    !a[OVS_PACKET_ATTR_ACTIONS])
520 		goto err;
521 
522 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
523 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
524 	err = -ENOMEM;
525 	if (!packet)
526 		goto err;
527 	skb_reserve(packet, NET_IP_ALIGN);
528 
529 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
530 
531 	skb_reset_mac_header(packet);
532 	eth = eth_hdr(packet);
533 
534 	/* Normally, setting the skb 'protocol' field would be handled by a
535 	 * call to eth_type_trans(), but it assumes there's a sending
536 	 * device, which we may not have. */
537 	if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
538 		packet->protocol = eth->h_proto;
539 	else
540 		packet->protocol = htons(ETH_P_802_2);
541 
542 	/* Build an sw_flow for sending this packet. */
543 	flow = ovs_flow_alloc();
544 	err = PTR_ERR(flow);
545 	if (IS_ERR(flow))
546 		goto err_kfree_skb;
547 
548 	err = ovs_flow_extract(packet, -1, &flow->key);
549 	if (err)
550 		goto err_flow_free;
551 
552 	err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
553 	if (err)
554 		goto err_flow_free;
555 	acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
556 	err = PTR_ERR(acts);
557 	if (IS_ERR(acts))
558 		goto err_flow_free;
559 
560 	err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
561 				   &flow->key, 0, &acts);
562 	rcu_assign_pointer(flow->sf_acts, acts);
563 	if (err)
564 		goto err_flow_free;
565 
566 	OVS_CB(packet)->flow = flow;
567 	OVS_CB(packet)->pkt_key = &flow->key;
568 	packet->priority = flow->key.phy.priority;
569 	packet->mark = flow->key.phy.skb_mark;
570 
571 	rcu_read_lock();
572 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
573 	err = -ENODEV;
574 	if (!dp)
575 		goto err_unlock;
576 
577 	local_bh_disable();
578 	err = ovs_execute_actions(dp, packet);
579 	local_bh_enable();
580 	rcu_read_unlock();
581 
582 	ovs_flow_free(flow, false);
583 	return err;
584 
585 err_unlock:
586 	rcu_read_unlock();
587 err_flow_free:
588 	ovs_flow_free(flow, false);
589 err_kfree_skb:
590 	kfree_skb(packet);
591 err:
592 	return err;
593 }
594 
595 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
596 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
597 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
598 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
599 };
600 
601 static const struct genl_ops dp_packet_genl_ops[] = {
602 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
603 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
604 	  .policy = packet_policy,
605 	  .doit = ovs_packet_cmd_execute
606 	}
607 };
608 
609 static struct genl_family dp_packet_genl_family = {
610 	.id = GENL_ID_GENERATE,
611 	.hdrsize = sizeof(struct ovs_header),
612 	.name = OVS_PACKET_FAMILY,
613 	.version = OVS_PACKET_VERSION,
614 	.maxattr = OVS_PACKET_ATTR_MAX,
615 	.netnsok = true,
616 	.parallel_ops = true,
617 	.ops = dp_packet_genl_ops,
618 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
619 };
620 
621 static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
622 			 struct ovs_dp_megaflow_stats *mega_stats)
623 {
624 	int i;
625 
626 	memset(mega_stats, 0, sizeof(*mega_stats));
627 
628 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
629 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
630 
631 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
632 
633 	for_each_possible_cpu(i) {
634 		const struct dp_stats_percpu *percpu_stats;
635 		struct dp_stats_percpu local_stats;
636 		unsigned int start;
637 
638 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
639 
640 		do {
641 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
642 			local_stats = *percpu_stats;
643 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
644 
645 		stats->n_hit += local_stats.n_hit;
646 		stats->n_missed += local_stats.n_missed;
647 		stats->n_lost += local_stats.n_lost;
648 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
649 	}
650 }
651 
652 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
653 {
654 	return NLMSG_ALIGN(sizeof(struct ovs_header))
655 		+ nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
656 		+ nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
657 		+ nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
658 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
659 		+ nla_total_size(8) /* OVS_FLOW_ATTR_USED */
660 		+ nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
661 }
662 
663 /* Called with ovs_mutex or RCU read lock. */
664 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
665 				  struct sk_buff *skb, u32 portid,
666 				  u32 seq, u32 flags, u8 cmd)
667 {
668 	const int skb_orig_len = skb->len;
669 	struct nlattr *start;
670 	struct ovs_flow_stats stats;
671 	__be16 tcp_flags;
672 	unsigned long used;
673 	struct ovs_header *ovs_header;
674 	struct nlattr *nla;
675 	int err;
676 
677 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
678 	if (!ovs_header)
679 		return -EMSGSIZE;
680 
681 	ovs_header->dp_ifindex = dp_ifindex;
682 
683 	/* Fill flow key. */
684 	nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
685 	if (!nla)
686 		goto nla_put_failure;
687 
688 	err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
689 	if (err)
690 		goto error;
691 	nla_nest_end(skb, nla);
692 
693 	nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
694 	if (!nla)
695 		goto nla_put_failure;
696 
697 	err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
698 	if (err)
699 		goto error;
700 
701 	nla_nest_end(skb, nla);
702 
703 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
704 
705 	if (used &&
706 	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
707 		goto nla_put_failure;
708 
709 	if (stats.n_packets &&
710 	    nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
711 		goto nla_put_failure;
712 
713 	if ((u8)ntohs(tcp_flags) &&
714 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
715 		goto nla_put_failure;
716 
717 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
718 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
719 	 * Netlink but individual action lists can be longer than
720 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
721 	 * The userspace caller can always fetch the actions separately if it
722 	 * really wants them.  (Most userspace callers in fact don't care.)
723 	 *
724 	 * This can only fail for dump operations because the skb is always
725 	 * properly sized for single flows.
726 	 */
727 	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
728 	if (start) {
729 		const struct sw_flow_actions *sf_acts;
730 
731 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
732 		err = ovs_nla_put_actions(sf_acts->actions,
733 					  sf_acts->actions_len, skb);
734 
735 		if (!err)
736 			nla_nest_end(skb, start);
737 		else {
738 			if (skb_orig_len)
739 				goto error;
740 
741 			nla_nest_cancel(skb, start);
742 		}
743 	} else if (skb_orig_len)
744 		goto nla_put_failure;
745 
746 	return genlmsg_end(skb, ovs_header);
747 
748 nla_put_failure:
749 	err = -EMSGSIZE;
750 error:
751 	genlmsg_cancel(skb, ovs_header);
752 	return err;
753 }
754 
755 /* May not be called with RCU read lock. */
756 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
757 					       struct genl_info *info,
758 					       bool always)
759 {
760 	struct sk_buff *skb;
761 
762 	if (!always && !ovs_must_notify(info, &ovs_dp_flow_multicast_group))
763 		return NULL;
764 
765 	skb = genlmsg_new_unicast(ovs_flow_cmd_msg_size(acts), info, GFP_KERNEL);
766 	if (!skb)
767 		return ERR_PTR(-ENOMEM);
768 
769 	return skb;
770 }
771 
772 /* Called with ovs_mutex. */
773 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
774 					       int dp_ifindex,
775 					       struct genl_info *info, u8 cmd,
776 					       bool always)
777 {
778 	struct sk_buff *skb;
779 	int retval;
780 
781 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts), info,
782 				      always);
783 	if (!skb || IS_ERR(skb))
784 		return skb;
785 
786 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
787 					info->snd_portid, info->snd_seq, 0,
788 					cmd);
789 	BUG_ON(retval < 0);
790 	return skb;
791 }
792 
793 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
794 {
795 	struct nlattr **a = info->attrs;
796 	struct ovs_header *ovs_header = info->userhdr;
797 	struct sw_flow *flow, *new_flow;
798 	struct sw_flow_mask mask;
799 	struct sk_buff *reply;
800 	struct datapath *dp;
801 	struct sw_flow_actions *acts;
802 	struct sw_flow_match match;
803 	int error;
804 
805 	/* Must have key and actions. */
806 	error = -EINVAL;
807 	if (!a[OVS_FLOW_ATTR_KEY])
808 		goto error;
809 	if (!a[OVS_FLOW_ATTR_ACTIONS])
810 		goto error;
811 
812 	/* Most of the time we need to allocate a new flow, do it before
813 	 * locking.
814 	 */
815 	new_flow = ovs_flow_alloc();
816 	if (IS_ERR(new_flow)) {
817 		error = PTR_ERR(new_flow);
818 		goto error;
819 	}
820 
821 	/* Extract key. */
822 	ovs_match_init(&match, &new_flow->unmasked_key, &mask);
823 	error = ovs_nla_get_match(&match,
824 				  a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
825 	if (error)
826 		goto err_kfree_flow;
827 
828 	ovs_flow_mask_key(&new_flow->key, &new_flow->unmasked_key, &mask);
829 
830 	/* Validate actions. */
831 	acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
832 	error = PTR_ERR(acts);
833 	if (IS_ERR(acts))
834 		goto err_kfree_flow;
835 
836 	error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
837 				     0, &acts);
838 	if (error) {
839 		OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
840 		goto err_kfree_acts;
841 	}
842 
843 	reply = ovs_flow_cmd_alloc_info(acts, info, false);
844 	if (IS_ERR(reply)) {
845 		error = PTR_ERR(reply);
846 		goto err_kfree_acts;
847 	}
848 
849 	ovs_lock();
850 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
851 	if (unlikely(!dp)) {
852 		error = -ENODEV;
853 		goto err_unlock_ovs;
854 	}
855 	/* Check if this is a duplicate flow */
856 	flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->unmasked_key);
857 	if (likely(!flow)) {
858 		rcu_assign_pointer(new_flow->sf_acts, acts);
859 
860 		/* Put flow in bucket. */
861 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
862 		if (unlikely(error)) {
863 			acts = NULL;
864 			goto err_unlock_ovs;
865 		}
866 
867 		if (unlikely(reply)) {
868 			error = ovs_flow_cmd_fill_info(new_flow,
869 						       ovs_header->dp_ifindex,
870 						       reply, info->snd_portid,
871 						       info->snd_seq, 0,
872 						       OVS_FLOW_CMD_NEW);
873 			BUG_ON(error < 0);
874 		}
875 		ovs_unlock();
876 	} else {
877 		struct sw_flow_actions *old_acts;
878 
879 		/* Bail out if we're not allowed to modify an existing flow.
880 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
881 		 * because Generic Netlink treats the latter as a dump
882 		 * request.  We also accept NLM_F_EXCL in case that bug ever
883 		 * gets fixed.
884 		 */
885 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
886 							 | NLM_F_EXCL))) {
887 			error = -EEXIST;
888 			goto err_unlock_ovs;
889 		}
890 		/* The unmasked key has to be the same for flow updates. */
891 		if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
892 			error = -EEXIST;
893 			goto err_unlock_ovs;
894 		}
895 		/* Update actions. */
896 		old_acts = ovsl_dereference(flow->sf_acts);
897 		rcu_assign_pointer(flow->sf_acts, acts);
898 
899 		if (unlikely(reply)) {
900 			error = ovs_flow_cmd_fill_info(flow,
901 						       ovs_header->dp_ifindex,
902 						       reply, info->snd_portid,
903 						       info->snd_seq, 0,
904 						       OVS_FLOW_CMD_NEW);
905 			BUG_ON(error < 0);
906 		}
907 		ovs_unlock();
908 
909 		ovs_nla_free_flow_actions(old_acts);
910 		ovs_flow_free(new_flow, false);
911 	}
912 
913 	if (reply)
914 		ovs_notify(&dp_flow_genl_family, reply, info);
915 	return 0;
916 
917 err_unlock_ovs:
918 	ovs_unlock();
919 	kfree_skb(reply);
920 err_kfree_acts:
921 	kfree(acts);
922 err_kfree_flow:
923 	ovs_flow_free(new_flow, false);
924 error:
925 	return error;
926 }
927 
928 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
929 {
930 	struct nlattr **a = info->attrs;
931 	struct ovs_header *ovs_header = info->userhdr;
932 	struct sw_flow_key key, masked_key;
933 	struct sw_flow *flow;
934 	struct sw_flow_mask mask;
935 	struct sk_buff *reply = NULL;
936 	struct datapath *dp;
937 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
938 	struct sw_flow_match match;
939 	int error;
940 
941 	/* Extract key. */
942 	error = -EINVAL;
943 	if (!a[OVS_FLOW_ATTR_KEY])
944 		goto error;
945 
946 	ovs_match_init(&match, &key, &mask);
947 	error = ovs_nla_get_match(&match,
948 				  a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
949 	if (error)
950 		goto error;
951 
952 	/* Validate actions. */
953 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
954 		acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
955 		error = PTR_ERR(acts);
956 		if (IS_ERR(acts))
957 			goto error;
958 
959 		ovs_flow_mask_key(&masked_key, &key, &mask);
960 		error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
961 					     &masked_key, 0, &acts);
962 		if (error) {
963 			OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
964 			goto err_kfree_acts;
965 		}
966 	}
967 
968 	/* Can allocate before locking if have acts. */
969 	if (acts) {
970 		reply = ovs_flow_cmd_alloc_info(acts, info, false);
971 		if (IS_ERR(reply)) {
972 			error = PTR_ERR(reply);
973 			goto err_kfree_acts;
974 		}
975 	}
976 
977 	ovs_lock();
978 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
979 	if (unlikely(!dp)) {
980 		error = -ENODEV;
981 		goto err_unlock_ovs;
982 	}
983 	/* Check that the flow exists. */
984 	flow = ovs_flow_tbl_lookup(&dp->table, &key);
985 	if (unlikely(!flow)) {
986 		error = -ENOENT;
987 		goto err_unlock_ovs;
988 	}
989 	/* The unmasked key has to be the same for flow updates. */
990 	if (unlikely(!ovs_flow_cmp_unmasked_key(flow, &match))) {
991 		error = -EEXIST;
992 		goto err_unlock_ovs;
993 	}
994 	/* Update actions, if present. */
995 	if (likely(acts)) {
996 		old_acts = ovsl_dereference(flow->sf_acts);
997 		rcu_assign_pointer(flow->sf_acts, acts);
998 
999 		if (unlikely(reply)) {
1000 			error = ovs_flow_cmd_fill_info(flow,
1001 						       ovs_header->dp_ifindex,
1002 						       reply, info->snd_portid,
1003 						       info->snd_seq, 0,
1004 						       OVS_FLOW_CMD_NEW);
1005 			BUG_ON(error < 0);
1006 		}
1007 	} else {
1008 		/* Could not alloc without acts before locking. */
1009 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1010 						info, OVS_FLOW_CMD_NEW, false);
1011 		if (unlikely(IS_ERR(reply))) {
1012 			error = PTR_ERR(reply);
1013 			goto err_unlock_ovs;
1014 		}
1015 	}
1016 
1017 	/* Clear stats. */
1018 	if (a[OVS_FLOW_ATTR_CLEAR])
1019 		ovs_flow_stats_clear(flow);
1020 	ovs_unlock();
1021 
1022 	if (reply)
1023 		ovs_notify(&dp_flow_genl_family, reply, info);
1024 	if (old_acts)
1025 		ovs_nla_free_flow_actions(old_acts);
1026 
1027 	return 0;
1028 
1029 err_unlock_ovs:
1030 	ovs_unlock();
1031 	kfree_skb(reply);
1032 err_kfree_acts:
1033 	kfree(acts);
1034 error:
1035 	return error;
1036 }
1037 
1038 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1039 {
1040 	struct nlattr **a = info->attrs;
1041 	struct ovs_header *ovs_header = info->userhdr;
1042 	struct sw_flow_key key;
1043 	struct sk_buff *reply;
1044 	struct sw_flow *flow;
1045 	struct datapath *dp;
1046 	struct sw_flow_match match;
1047 	int err;
1048 
1049 	if (!a[OVS_FLOW_ATTR_KEY]) {
1050 		OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
1051 		return -EINVAL;
1052 	}
1053 
1054 	ovs_match_init(&match, &key, NULL);
1055 	err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1056 	if (err)
1057 		return err;
1058 
1059 	ovs_lock();
1060 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1061 	if (!dp) {
1062 		err = -ENODEV;
1063 		goto unlock;
1064 	}
1065 
1066 	flow = ovs_flow_tbl_lookup(&dp->table, &key);
1067 	if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
1068 		err = -ENOENT;
1069 		goto unlock;
1070 	}
1071 
1072 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1073 					OVS_FLOW_CMD_NEW, true);
1074 	if (IS_ERR(reply)) {
1075 		err = PTR_ERR(reply);
1076 		goto unlock;
1077 	}
1078 
1079 	ovs_unlock();
1080 	return genlmsg_reply(reply, info);
1081 unlock:
1082 	ovs_unlock();
1083 	return err;
1084 }
1085 
1086 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1087 {
1088 	struct nlattr **a = info->attrs;
1089 	struct ovs_header *ovs_header = info->userhdr;
1090 	struct sw_flow_key key;
1091 	struct sk_buff *reply;
1092 	struct sw_flow *flow;
1093 	struct datapath *dp;
1094 	struct sw_flow_match match;
1095 	int err;
1096 
1097 	if (likely(a[OVS_FLOW_ATTR_KEY])) {
1098 		ovs_match_init(&match, &key, NULL);
1099 		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
1100 		if (unlikely(err))
1101 			return err;
1102 	}
1103 
1104 	ovs_lock();
1105 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1106 	if (unlikely(!dp)) {
1107 		err = -ENODEV;
1108 		goto unlock;
1109 	}
1110 
1111 	if (unlikely(!a[OVS_FLOW_ATTR_KEY])) {
1112 		err = ovs_flow_tbl_flush(&dp->table);
1113 		goto unlock;
1114 	}
1115 
1116 	flow = ovs_flow_tbl_lookup(&dp->table, &key);
1117 	if (unlikely(!flow || !ovs_flow_cmp_unmasked_key(flow, &match))) {
1118 		err = -ENOENT;
1119 		goto unlock;
1120 	}
1121 
1122 	ovs_flow_tbl_remove(&dp->table, flow);
1123 	ovs_unlock();
1124 
1125 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1126 					info, false);
1127 	if (likely(reply)) {
1128 		if (likely(!IS_ERR(reply))) {
1129 			rcu_read_lock();	/*To keep RCU checker happy. */
1130 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1131 						     reply, info->snd_portid,
1132 						     info->snd_seq, 0,
1133 						     OVS_FLOW_CMD_DEL);
1134 			rcu_read_unlock();
1135 			BUG_ON(err < 0);
1136 
1137 			ovs_notify(&dp_flow_genl_family, reply, info);
1138 		} else {
1139 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1140 		}
1141 	}
1142 
1143 	ovs_flow_free(flow, true);
1144 	return 0;
1145 unlock:
1146 	ovs_unlock();
1147 	return err;
1148 }
1149 
1150 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1151 {
1152 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1153 	struct table_instance *ti;
1154 	struct datapath *dp;
1155 
1156 	rcu_read_lock();
1157 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1158 	if (!dp) {
1159 		rcu_read_unlock();
1160 		return -ENODEV;
1161 	}
1162 
1163 	ti = rcu_dereference(dp->table.ti);
1164 	for (;;) {
1165 		struct sw_flow *flow;
1166 		u32 bucket, obj;
1167 
1168 		bucket = cb->args[0];
1169 		obj = cb->args[1];
1170 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1171 		if (!flow)
1172 			break;
1173 
1174 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1175 					   NETLINK_CB(cb->skb).portid,
1176 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1177 					   OVS_FLOW_CMD_NEW) < 0)
1178 			break;
1179 
1180 		cb->args[0] = bucket;
1181 		cb->args[1] = obj;
1182 	}
1183 	rcu_read_unlock();
1184 	return skb->len;
1185 }
1186 
1187 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1188 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1189 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1190 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1191 };
1192 
1193 static struct genl_ops dp_flow_genl_ops[] = {
1194 	{ .cmd = OVS_FLOW_CMD_NEW,
1195 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1196 	  .policy = flow_policy,
1197 	  .doit = ovs_flow_cmd_new
1198 	},
1199 	{ .cmd = OVS_FLOW_CMD_DEL,
1200 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1201 	  .policy = flow_policy,
1202 	  .doit = ovs_flow_cmd_del
1203 	},
1204 	{ .cmd = OVS_FLOW_CMD_GET,
1205 	  .flags = 0,		    /* OK for unprivileged users. */
1206 	  .policy = flow_policy,
1207 	  .doit = ovs_flow_cmd_get,
1208 	  .dumpit = ovs_flow_cmd_dump
1209 	},
1210 	{ .cmd = OVS_FLOW_CMD_SET,
1211 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1212 	  .policy = flow_policy,
1213 	  .doit = ovs_flow_cmd_set,
1214 	},
1215 };
1216 
1217 static struct genl_family dp_flow_genl_family = {
1218 	.id = GENL_ID_GENERATE,
1219 	.hdrsize = sizeof(struct ovs_header),
1220 	.name = OVS_FLOW_FAMILY,
1221 	.version = OVS_FLOW_VERSION,
1222 	.maxattr = OVS_FLOW_ATTR_MAX,
1223 	.netnsok = true,
1224 	.parallel_ops = true,
1225 	.ops = dp_flow_genl_ops,
1226 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1227 	.mcgrps = &ovs_dp_flow_multicast_group,
1228 	.n_mcgrps = 1,
1229 };
1230 
1231 static size_t ovs_dp_cmd_msg_size(void)
1232 {
1233 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1234 
1235 	msgsize += nla_total_size(IFNAMSIZ);
1236 	msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1237 	msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1238 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1239 
1240 	return msgsize;
1241 }
1242 
1243 /* Called with ovs_mutex or RCU read lock. */
1244 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1245 				u32 portid, u32 seq, u32 flags, u8 cmd)
1246 {
1247 	struct ovs_header *ovs_header;
1248 	struct ovs_dp_stats dp_stats;
1249 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1250 	int err;
1251 
1252 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1253 				   flags, cmd);
1254 	if (!ovs_header)
1255 		goto error;
1256 
1257 	ovs_header->dp_ifindex = get_dpifindex(dp);
1258 
1259 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1260 	if (err)
1261 		goto nla_put_failure;
1262 
1263 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1264 	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1265 			&dp_stats))
1266 		goto nla_put_failure;
1267 
1268 	if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1269 			sizeof(struct ovs_dp_megaflow_stats),
1270 			&dp_megaflow_stats))
1271 		goto nla_put_failure;
1272 
1273 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1274 		goto nla_put_failure;
1275 
1276 	return genlmsg_end(skb, ovs_header);
1277 
1278 nla_put_failure:
1279 	genlmsg_cancel(skb, ovs_header);
1280 error:
1281 	return -EMSGSIZE;
1282 }
1283 
1284 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1285 {
1286 	return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1287 }
1288 
1289 /* Called with rcu_read_lock or ovs_mutex. */
1290 static struct datapath *lookup_datapath(struct net *net,
1291 					struct ovs_header *ovs_header,
1292 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1293 {
1294 	struct datapath *dp;
1295 
1296 	if (!a[OVS_DP_ATTR_NAME])
1297 		dp = get_dp(net, ovs_header->dp_ifindex);
1298 	else {
1299 		struct vport *vport;
1300 
1301 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1302 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1303 	}
1304 	return dp ? dp : ERR_PTR(-ENODEV);
1305 }
1306 
1307 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1308 {
1309 	struct datapath *dp;
1310 
1311 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1312 	if (IS_ERR(dp))
1313 		return;
1314 
1315 	WARN(dp->user_features, "Dropping previously announced user features\n");
1316 	dp->user_features = 0;
1317 }
1318 
1319 static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1320 {
1321 	if (a[OVS_DP_ATTR_USER_FEATURES])
1322 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1323 }
1324 
1325 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1326 {
1327 	struct nlattr **a = info->attrs;
1328 	struct vport_parms parms;
1329 	struct sk_buff *reply;
1330 	struct datapath *dp;
1331 	struct vport *vport;
1332 	struct ovs_net *ovs_net;
1333 	int err, i;
1334 
1335 	err = -EINVAL;
1336 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1337 		goto err;
1338 
1339 	reply = ovs_dp_cmd_alloc_info(info);
1340 	if (!reply)
1341 		return -ENOMEM;
1342 
1343 	err = -ENOMEM;
1344 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1345 	if (dp == NULL)
1346 		goto err_free_reply;
1347 
1348 	ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1349 
1350 	/* Allocate table. */
1351 	err = ovs_flow_tbl_init(&dp->table);
1352 	if (err)
1353 		goto err_free_dp;
1354 
1355 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1356 	if (!dp->stats_percpu) {
1357 		err = -ENOMEM;
1358 		goto err_destroy_table;
1359 	}
1360 
1361 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1362 			    GFP_KERNEL);
1363 	if (!dp->ports) {
1364 		err = -ENOMEM;
1365 		goto err_destroy_percpu;
1366 	}
1367 
1368 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1369 		INIT_HLIST_HEAD(&dp->ports[i]);
1370 
1371 	/* Set up our datapath device. */
1372 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1373 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1374 	parms.options = NULL;
1375 	parms.dp = dp;
1376 	parms.port_no = OVSP_LOCAL;
1377 	parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1378 
1379 	ovs_dp_change(dp, a);
1380 
1381 	/* So far only local changes have been made, now need the lock. */
1382 	ovs_lock();
1383 
1384 	vport = new_vport(&parms);
1385 	if (IS_ERR(vport)) {
1386 		err = PTR_ERR(vport);
1387 		if (err == -EBUSY)
1388 			err = -EEXIST;
1389 
1390 		if (err == -EEXIST) {
1391 			/* An outdated user space instance that does not understand
1392 			 * the concept of user_features has attempted to create a new
1393 			 * datapath and is likely to reuse it. Drop all user features.
1394 			 */
1395 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1396 				ovs_dp_reset_user_features(skb, info);
1397 		}
1398 
1399 		goto err_destroy_ports_array;
1400 	}
1401 
1402 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1403 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1404 	BUG_ON(err < 0);
1405 
1406 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1407 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1408 
1409 	ovs_unlock();
1410 
1411 	ovs_notify(&dp_datapath_genl_family, reply, info);
1412 	return 0;
1413 
1414 err_destroy_ports_array:
1415 	ovs_unlock();
1416 	kfree(dp->ports);
1417 err_destroy_percpu:
1418 	free_percpu(dp->stats_percpu);
1419 err_destroy_table:
1420 	ovs_flow_tbl_destroy(&dp->table, false);
1421 err_free_dp:
1422 	release_net(ovs_dp_get_net(dp));
1423 	kfree(dp);
1424 err_free_reply:
1425 	kfree_skb(reply);
1426 err:
1427 	return err;
1428 }
1429 
1430 /* Called with ovs_mutex. */
1431 static void __dp_destroy(struct datapath *dp)
1432 {
1433 	int i;
1434 
1435 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1436 		struct vport *vport;
1437 		struct hlist_node *n;
1438 
1439 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1440 			if (vport->port_no != OVSP_LOCAL)
1441 				ovs_dp_detach_port(vport);
1442 	}
1443 
1444 	list_del_rcu(&dp->list_node);
1445 
1446 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1447 	 * all ports in datapath are destroyed first before freeing datapath.
1448 	 */
1449 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1450 
1451 	/* RCU destroy the flow table */
1452 	ovs_flow_tbl_destroy(&dp->table, true);
1453 
1454 	call_rcu(&dp->rcu, destroy_dp_rcu);
1455 }
1456 
1457 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1458 {
1459 	struct sk_buff *reply;
1460 	struct datapath *dp;
1461 	int err;
1462 
1463 	reply = ovs_dp_cmd_alloc_info(info);
1464 	if (!reply)
1465 		return -ENOMEM;
1466 
1467 	ovs_lock();
1468 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1469 	err = PTR_ERR(dp);
1470 	if (IS_ERR(dp))
1471 		goto err_unlock_free;
1472 
1473 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1474 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1475 	BUG_ON(err < 0);
1476 
1477 	__dp_destroy(dp);
1478 	ovs_unlock();
1479 
1480 	ovs_notify(&dp_datapath_genl_family, reply, info);
1481 
1482 	return 0;
1483 
1484 err_unlock_free:
1485 	ovs_unlock();
1486 	kfree_skb(reply);
1487 	return err;
1488 }
1489 
1490 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1491 {
1492 	struct sk_buff *reply;
1493 	struct datapath *dp;
1494 	int err;
1495 
1496 	reply = ovs_dp_cmd_alloc_info(info);
1497 	if (!reply)
1498 		return -ENOMEM;
1499 
1500 	ovs_lock();
1501 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1502 	err = PTR_ERR(dp);
1503 	if (IS_ERR(dp))
1504 		goto err_unlock_free;
1505 
1506 	ovs_dp_change(dp, info->attrs);
1507 
1508 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1509 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1510 	BUG_ON(err < 0);
1511 
1512 	ovs_unlock();
1513 	ovs_notify(&dp_datapath_genl_family, reply, info);
1514 
1515 	return 0;
1516 
1517 err_unlock_free:
1518 	ovs_unlock();
1519 	kfree_skb(reply);
1520 	return err;
1521 }
1522 
1523 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1524 {
1525 	struct sk_buff *reply;
1526 	struct datapath *dp;
1527 	int err;
1528 
1529 	reply = ovs_dp_cmd_alloc_info(info);
1530 	if (!reply)
1531 		return -ENOMEM;
1532 
1533 	rcu_read_lock();
1534 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1535 	if (IS_ERR(dp)) {
1536 		err = PTR_ERR(dp);
1537 		goto err_unlock_free;
1538 	}
1539 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1540 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1541 	BUG_ON(err < 0);
1542 	rcu_read_unlock();
1543 
1544 	return genlmsg_reply(reply, info);
1545 
1546 err_unlock_free:
1547 	rcu_read_unlock();
1548 	kfree_skb(reply);
1549 	return err;
1550 }
1551 
1552 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1553 {
1554 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1555 	struct datapath *dp;
1556 	int skip = cb->args[0];
1557 	int i = 0;
1558 
1559 	rcu_read_lock();
1560 	list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1561 		if (i >= skip &&
1562 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1563 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1564 					 OVS_DP_CMD_NEW) < 0)
1565 			break;
1566 		i++;
1567 	}
1568 	rcu_read_unlock();
1569 
1570 	cb->args[0] = i;
1571 
1572 	return skb->len;
1573 }
1574 
1575 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1576 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1577 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1578 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1579 };
1580 
1581 static struct genl_ops dp_datapath_genl_ops[] = {
1582 	{ .cmd = OVS_DP_CMD_NEW,
1583 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1584 	  .policy = datapath_policy,
1585 	  .doit = ovs_dp_cmd_new
1586 	},
1587 	{ .cmd = OVS_DP_CMD_DEL,
1588 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1589 	  .policy = datapath_policy,
1590 	  .doit = ovs_dp_cmd_del
1591 	},
1592 	{ .cmd = OVS_DP_CMD_GET,
1593 	  .flags = 0,		    /* OK for unprivileged users. */
1594 	  .policy = datapath_policy,
1595 	  .doit = ovs_dp_cmd_get,
1596 	  .dumpit = ovs_dp_cmd_dump
1597 	},
1598 	{ .cmd = OVS_DP_CMD_SET,
1599 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1600 	  .policy = datapath_policy,
1601 	  .doit = ovs_dp_cmd_set,
1602 	},
1603 };
1604 
1605 static struct genl_family dp_datapath_genl_family = {
1606 	.id = GENL_ID_GENERATE,
1607 	.hdrsize = sizeof(struct ovs_header),
1608 	.name = OVS_DATAPATH_FAMILY,
1609 	.version = OVS_DATAPATH_VERSION,
1610 	.maxattr = OVS_DP_ATTR_MAX,
1611 	.netnsok = true,
1612 	.parallel_ops = true,
1613 	.ops = dp_datapath_genl_ops,
1614 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1615 	.mcgrps = &ovs_dp_datapath_multicast_group,
1616 	.n_mcgrps = 1,
1617 };
1618 
1619 /* Called with ovs_mutex or RCU read lock. */
1620 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1621 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1622 {
1623 	struct ovs_header *ovs_header;
1624 	struct ovs_vport_stats vport_stats;
1625 	int err;
1626 
1627 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1628 				 flags, cmd);
1629 	if (!ovs_header)
1630 		return -EMSGSIZE;
1631 
1632 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1633 
1634 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1635 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1636 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1637 	    nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1638 		goto nla_put_failure;
1639 
1640 	ovs_vport_get_stats(vport, &vport_stats);
1641 	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1642 		    &vport_stats))
1643 		goto nla_put_failure;
1644 
1645 	err = ovs_vport_get_options(vport, skb);
1646 	if (err == -EMSGSIZE)
1647 		goto error;
1648 
1649 	return genlmsg_end(skb, ovs_header);
1650 
1651 nla_put_failure:
1652 	err = -EMSGSIZE;
1653 error:
1654 	genlmsg_cancel(skb, ovs_header);
1655 	return err;
1656 }
1657 
1658 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1659 {
1660 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1661 }
1662 
1663 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1664 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1665 					 u32 seq, u8 cmd)
1666 {
1667 	struct sk_buff *skb;
1668 	int retval;
1669 
1670 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1671 	if (!skb)
1672 		return ERR_PTR(-ENOMEM);
1673 
1674 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1675 	BUG_ON(retval < 0);
1676 
1677 	return skb;
1678 }
1679 
1680 /* Called with ovs_mutex or RCU read lock. */
1681 static struct vport *lookup_vport(struct net *net,
1682 				  struct ovs_header *ovs_header,
1683 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1684 {
1685 	struct datapath *dp;
1686 	struct vport *vport;
1687 
1688 	if (a[OVS_VPORT_ATTR_NAME]) {
1689 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1690 		if (!vport)
1691 			return ERR_PTR(-ENODEV);
1692 		if (ovs_header->dp_ifindex &&
1693 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1694 			return ERR_PTR(-ENODEV);
1695 		return vport;
1696 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1697 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1698 
1699 		if (port_no >= DP_MAX_PORTS)
1700 			return ERR_PTR(-EFBIG);
1701 
1702 		dp = get_dp(net, ovs_header->dp_ifindex);
1703 		if (!dp)
1704 			return ERR_PTR(-ENODEV);
1705 
1706 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1707 		if (!vport)
1708 			return ERR_PTR(-ENODEV);
1709 		return vport;
1710 	} else
1711 		return ERR_PTR(-EINVAL);
1712 }
1713 
1714 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1715 {
1716 	struct nlattr **a = info->attrs;
1717 	struct ovs_header *ovs_header = info->userhdr;
1718 	struct vport_parms parms;
1719 	struct sk_buff *reply;
1720 	struct vport *vport;
1721 	struct datapath *dp;
1722 	u32 port_no;
1723 	int err;
1724 
1725 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1726 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1727 		return -EINVAL;
1728 
1729 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
1730 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1731 	if (port_no >= DP_MAX_PORTS)
1732 		return -EFBIG;
1733 
1734 	reply = ovs_vport_cmd_alloc_info();
1735 	if (!reply)
1736 		return -ENOMEM;
1737 
1738 	ovs_lock();
1739 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1740 	err = -ENODEV;
1741 	if (!dp)
1742 		goto exit_unlock_free;
1743 
1744 	if (port_no) {
1745 		vport = ovs_vport_ovsl(dp, port_no);
1746 		err = -EBUSY;
1747 		if (vport)
1748 			goto exit_unlock_free;
1749 	} else {
1750 		for (port_no = 1; ; port_no++) {
1751 			if (port_no >= DP_MAX_PORTS) {
1752 				err = -EFBIG;
1753 				goto exit_unlock_free;
1754 			}
1755 			vport = ovs_vport_ovsl(dp, port_no);
1756 			if (!vport)
1757 				break;
1758 		}
1759 	}
1760 
1761 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1762 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1763 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1764 	parms.dp = dp;
1765 	parms.port_no = port_no;
1766 	parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1767 
1768 	vport = new_vport(&parms);
1769 	err = PTR_ERR(vport);
1770 	if (IS_ERR(vport))
1771 		goto exit_unlock_free;
1772 
1773 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1774 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1775 	BUG_ON(err < 0);
1776 	ovs_unlock();
1777 
1778 	ovs_notify(&dp_vport_genl_family, reply, info);
1779 	return 0;
1780 
1781 exit_unlock_free:
1782 	ovs_unlock();
1783 	kfree_skb(reply);
1784 	return err;
1785 }
1786 
1787 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1788 {
1789 	struct nlattr **a = info->attrs;
1790 	struct sk_buff *reply;
1791 	struct vport *vport;
1792 	int err;
1793 
1794 	reply = ovs_vport_cmd_alloc_info();
1795 	if (!reply)
1796 		return -ENOMEM;
1797 
1798 	ovs_lock();
1799 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1800 	err = PTR_ERR(vport);
1801 	if (IS_ERR(vport))
1802 		goto exit_unlock_free;
1803 
1804 	if (a[OVS_VPORT_ATTR_TYPE] &&
1805 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1806 		err = -EINVAL;
1807 		goto exit_unlock_free;
1808 	}
1809 
1810 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
1811 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1812 		if (err)
1813 			goto exit_unlock_free;
1814 	}
1815 
1816 	if (a[OVS_VPORT_ATTR_UPCALL_PID])
1817 		vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1818 
1819 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1820 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1821 	BUG_ON(err < 0);
1822 
1823 	ovs_unlock();
1824 	ovs_notify(&dp_vport_genl_family, reply, info);
1825 	return 0;
1826 
1827 exit_unlock_free:
1828 	ovs_unlock();
1829 	kfree_skb(reply);
1830 	return err;
1831 }
1832 
1833 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1834 {
1835 	struct nlattr **a = info->attrs;
1836 	struct sk_buff *reply;
1837 	struct vport *vport;
1838 	int err;
1839 
1840 	reply = ovs_vport_cmd_alloc_info();
1841 	if (!reply)
1842 		return -ENOMEM;
1843 
1844 	ovs_lock();
1845 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1846 	err = PTR_ERR(vport);
1847 	if (IS_ERR(vport))
1848 		goto exit_unlock_free;
1849 
1850 	if (vport->port_no == OVSP_LOCAL) {
1851 		err = -EINVAL;
1852 		goto exit_unlock_free;
1853 	}
1854 
1855 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1856 				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
1857 	BUG_ON(err < 0);
1858 	ovs_dp_detach_port(vport);
1859 	ovs_unlock();
1860 
1861 	ovs_notify(&dp_vport_genl_family, reply, info);
1862 	return 0;
1863 
1864 exit_unlock_free:
1865 	ovs_unlock();
1866 	kfree_skb(reply);
1867 	return err;
1868 }
1869 
1870 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1871 {
1872 	struct nlattr **a = info->attrs;
1873 	struct ovs_header *ovs_header = info->userhdr;
1874 	struct sk_buff *reply;
1875 	struct vport *vport;
1876 	int err;
1877 
1878 	reply = ovs_vport_cmd_alloc_info();
1879 	if (!reply)
1880 		return -ENOMEM;
1881 
1882 	rcu_read_lock();
1883 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1884 	err = PTR_ERR(vport);
1885 	if (IS_ERR(vport))
1886 		goto exit_unlock_free;
1887 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1888 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1889 	BUG_ON(err < 0);
1890 	rcu_read_unlock();
1891 
1892 	return genlmsg_reply(reply, info);
1893 
1894 exit_unlock_free:
1895 	rcu_read_unlock();
1896 	kfree_skb(reply);
1897 	return err;
1898 }
1899 
1900 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1901 {
1902 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1903 	struct datapath *dp;
1904 	int bucket = cb->args[0], skip = cb->args[1];
1905 	int i, j = 0;
1906 
1907 	rcu_read_lock();
1908 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1909 	if (!dp) {
1910 		rcu_read_unlock();
1911 		return -ENODEV;
1912 	}
1913 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1914 		struct vport *vport;
1915 
1916 		j = 0;
1917 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1918 			if (j >= skip &&
1919 			    ovs_vport_cmd_fill_info(vport, skb,
1920 						    NETLINK_CB(cb->skb).portid,
1921 						    cb->nlh->nlmsg_seq,
1922 						    NLM_F_MULTI,
1923 						    OVS_VPORT_CMD_NEW) < 0)
1924 				goto out;
1925 
1926 			j++;
1927 		}
1928 		skip = 0;
1929 	}
1930 out:
1931 	rcu_read_unlock();
1932 
1933 	cb->args[0] = i;
1934 	cb->args[1] = j;
1935 
1936 	return skb->len;
1937 }
1938 
1939 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1940 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1941 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1942 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1943 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1944 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1945 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1946 };
1947 
1948 static struct genl_ops dp_vport_genl_ops[] = {
1949 	{ .cmd = OVS_VPORT_CMD_NEW,
1950 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1951 	  .policy = vport_policy,
1952 	  .doit = ovs_vport_cmd_new
1953 	},
1954 	{ .cmd = OVS_VPORT_CMD_DEL,
1955 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1956 	  .policy = vport_policy,
1957 	  .doit = ovs_vport_cmd_del
1958 	},
1959 	{ .cmd = OVS_VPORT_CMD_GET,
1960 	  .flags = 0,		    /* OK for unprivileged users. */
1961 	  .policy = vport_policy,
1962 	  .doit = ovs_vport_cmd_get,
1963 	  .dumpit = ovs_vport_cmd_dump
1964 	},
1965 	{ .cmd = OVS_VPORT_CMD_SET,
1966 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1967 	  .policy = vport_policy,
1968 	  .doit = ovs_vport_cmd_set,
1969 	},
1970 };
1971 
1972 struct genl_family dp_vport_genl_family = {
1973 	.id = GENL_ID_GENERATE,
1974 	.hdrsize = sizeof(struct ovs_header),
1975 	.name = OVS_VPORT_FAMILY,
1976 	.version = OVS_VPORT_VERSION,
1977 	.maxattr = OVS_VPORT_ATTR_MAX,
1978 	.netnsok = true,
1979 	.parallel_ops = true,
1980 	.ops = dp_vport_genl_ops,
1981 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
1982 	.mcgrps = &ovs_dp_vport_multicast_group,
1983 	.n_mcgrps = 1,
1984 };
1985 
1986 static struct genl_family * const dp_genl_families[] = {
1987 	&dp_datapath_genl_family,
1988 	&dp_vport_genl_family,
1989 	&dp_flow_genl_family,
1990 	&dp_packet_genl_family,
1991 };
1992 
1993 static void dp_unregister_genl(int n_families)
1994 {
1995 	int i;
1996 
1997 	for (i = 0; i < n_families; i++)
1998 		genl_unregister_family(dp_genl_families[i]);
1999 }
2000 
2001 static int dp_register_genl(void)
2002 {
2003 	int err;
2004 	int i;
2005 
2006 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2007 
2008 		err = genl_register_family(dp_genl_families[i]);
2009 		if (err)
2010 			goto error;
2011 	}
2012 
2013 	return 0;
2014 
2015 error:
2016 	dp_unregister_genl(i);
2017 	return err;
2018 }
2019 
2020 static int __net_init ovs_init_net(struct net *net)
2021 {
2022 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2023 
2024 	INIT_LIST_HEAD(&ovs_net->dps);
2025 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2026 	return 0;
2027 }
2028 
2029 static void __net_exit ovs_exit_net(struct net *net)
2030 {
2031 	struct datapath *dp, *dp_next;
2032 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2033 
2034 	ovs_lock();
2035 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2036 		__dp_destroy(dp);
2037 	ovs_unlock();
2038 
2039 	cancel_work_sync(&ovs_net->dp_notify_work);
2040 }
2041 
2042 static struct pernet_operations ovs_net_ops = {
2043 	.init = ovs_init_net,
2044 	.exit = ovs_exit_net,
2045 	.id   = &ovs_net_id,
2046 	.size = sizeof(struct ovs_net),
2047 };
2048 
2049 static int __init dp_init(void)
2050 {
2051 	int err;
2052 
2053 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2054 
2055 	pr_info("Open vSwitch switching datapath\n");
2056 
2057 	err = ovs_flow_init();
2058 	if (err)
2059 		goto error;
2060 
2061 	err = ovs_vport_init();
2062 	if (err)
2063 		goto error_flow_exit;
2064 
2065 	err = register_pernet_device(&ovs_net_ops);
2066 	if (err)
2067 		goto error_vport_exit;
2068 
2069 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2070 	if (err)
2071 		goto error_netns_exit;
2072 
2073 	err = dp_register_genl();
2074 	if (err < 0)
2075 		goto error_unreg_notifier;
2076 
2077 	return 0;
2078 
2079 error_unreg_notifier:
2080 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2081 error_netns_exit:
2082 	unregister_pernet_device(&ovs_net_ops);
2083 error_vport_exit:
2084 	ovs_vport_exit();
2085 error_flow_exit:
2086 	ovs_flow_exit();
2087 error:
2088 	return err;
2089 }
2090 
2091 static void dp_cleanup(void)
2092 {
2093 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2094 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2095 	unregister_pernet_device(&ovs_net_ops);
2096 	rcu_barrier();
2097 	ovs_vport_exit();
2098 	ovs_flow_exit();
2099 }
2100 
2101 module_init(dp_init);
2102 module_exit(dp_cleanup);
2103 
2104 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2105 MODULE_LICENSE("GPL");
2106