xref: /linux/net/openvswitch/datapath.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
1 /*
2  * Copyright (c) 2007-2014 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
53 
54 #include "datapath.h"
55 #include "flow.h"
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
60 
61 int ovs_net_id __read_mostly;
62 EXPORT_SYMBOL_GPL(ovs_net_id);
63 
64 static struct genl_family dp_packet_genl_family;
65 static struct genl_family dp_flow_genl_family;
66 static struct genl_family dp_datapath_genl_family;
67 
68 static const struct nla_policy flow_policy[];
69 
70 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
71 	.name = OVS_FLOW_MCGROUP,
72 };
73 
74 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
75 	.name = OVS_DATAPATH_MCGROUP,
76 };
77 
78 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
79 	.name = OVS_VPORT_MCGROUP,
80 };
81 
82 /* Check if need to build a reply message.
83  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
84 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
85 			    unsigned int group)
86 {
87 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
88 	       genl_has_listeners(family, genl_info_net(info), group);
89 }
90 
91 static void ovs_notify(struct genl_family *family,
92 		       struct sk_buff *skb, struct genl_info *info)
93 {
94 	genl_notify(family, skb, genl_info_net(info), info->snd_portid,
95 		    0, info->nlhdr, GFP_KERNEL);
96 }
97 
98 /**
99  * DOC: Locking:
100  *
101  * All writes e.g. Writes to device state (add/remove datapath, port, set
102  * operations on vports, etc.), Writes to other state (flow table
103  * modifications, set miscellaneous datapath parameters, etc.) are protected
104  * by ovs_lock.
105  *
106  * Reads are protected by RCU.
107  *
108  * There are a few special cases (mostly stats) that have their own
109  * synchronization but they nest under all of above and don't interact with
110  * each other.
111  *
112  * The RTNL lock nests inside ovs_mutex.
113  */
114 
115 static DEFINE_MUTEX(ovs_mutex);
116 
117 void ovs_lock(void)
118 {
119 	mutex_lock(&ovs_mutex);
120 }
121 
122 void ovs_unlock(void)
123 {
124 	mutex_unlock(&ovs_mutex);
125 }
126 
127 #ifdef CONFIG_LOCKDEP
128 int lockdep_ovsl_is_held(void)
129 {
130 	if (debug_locks)
131 		return lockdep_is_held(&ovs_mutex);
132 	else
133 		return 1;
134 }
135 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
136 #endif
137 
138 static struct vport *new_vport(const struct vport_parms *);
139 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
140 			     const struct sw_flow_key *,
141 			     const struct dp_upcall_info *);
142 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
143 				  const struct sw_flow_key *,
144 				  const struct dp_upcall_info *);
145 
146 /* Must be called with rcu_read_lock. */
147 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
148 {
149 	struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
150 
151 	if (dev) {
152 		struct vport *vport = ovs_internal_dev_get_vport(dev);
153 		if (vport)
154 			return vport->dp;
155 	}
156 
157 	return NULL;
158 }
159 
160 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
161  * returned dp pointer valid.
162  */
163 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
164 {
165 	struct datapath *dp;
166 
167 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
168 	rcu_read_lock();
169 	dp = get_dp_rcu(net, dp_ifindex);
170 	rcu_read_unlock();
171 
172 	return dp;
173 }
174 
175 /* Must be called with rcu_read_lock or ovs_mutex. */
176 const char *ovs_dp_name(const struct datapath *dp)
177 {
178 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
179 	return vport->ops->get_name(vport);
180 }
181 
182 static int get_dpifindex(const struct datapath *dp)
183 {
184 	struct vport *local;
185 	int ifindex;
186 
187 	rcu_read_lock();
188 
189 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
190 	if (local)
191 		ifindex = netdev_vport_priv(local)->dev->ifindex;
192 	else
193 		ifindex = 0;
194 
195 	rcu_read_unlock();
196 
197 	return ifindex;
198 }
199 
200 static void destroy_dp_rcu(struct rcu_head *rcu)
201 {
202 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
203 
204 	ovs_flow_tbl_destroy(&dp->table);
205 	free_percpu(dp->stats_percpu);
206 	release_net(ovs_dp_get_net(dp));
207 	kfree(dp->ports);
208 	kfree(dp);
209 }
210 
211 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
212 					    u16 port_no)
213 {
214 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
215 }
216 
217 /* Called with ovs_mutex or RCU read lock. */
218 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
219 {
220 	struct vport *vport;
221 	struct hlist_head *head;
222 
223 	head = vport_hash_bucket(dp, port_no);
224 	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
225 		if (vport->port_no == port_no)
226 			return vport;
227 	}
228 	return NULL;
229 }
230 
231 /* Called with ovs_mutex. */
232 static struct vport *new_vport(const struct vport_parms *parms)
233 {
234 	struct vport *vport;
235 
236 	vport = ovs_vport_add(parms);
237 	if (!IS_ERR(vport)) {
238 		struct datapath *dp = parms->dp;
239 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
240 
241 		hlist_add_head_rcu(&vport->dp_hash_node, head);
242 	}
243 	return vport;
244 }
245 
246 void ovs_dp_detach_port(struct vport *p)
247 {
248 	ASSERT_OVSL();
249 
250 	/* First drop references to device. */
251 	hlist_del_rcu(&p->dp_hash_node);
252 
253 	/* Then destroy it. */
254 	ovs_vport_del(p);
255 }
256 
257 /* Must be called with rcu_read_lock. */
258 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
259 {
260 	const struct vport *p = OVS_CB(skb)->input_vport;
261 	struct datapath *dp = p->dp;
262 	struct sw_flow *flow;
263 	struct sw_flow_actions *sf_acts;
264 	struct dp_stats_percpu *stats;
265 	u64 *stats_counter;
266 	u32 n_mask_hit;
267 
268 	stats = this_cpu_ptr(dp->stats_percpu);
269 
270 	/* Look up flow. */
271 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
272 	if (unlikely(!flow)) {
273 		struct dp_upcall_info upcall;
274 		int error;
275 
276 		upcall.cmd = OVS_PACKET_CMD_MISS;
277 		upcall.userdata = NULL;
278 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
279 		upcall.egress_tun_info = NULL;
280 		error = ovs_dp_upcall(dp, skb, key, &upcall);
281 		if (unlikely(error))
282 			kfree_skb(skb);
283 		else
284 			consume_skb(skb);
285 		stats_counter = &stats->n_missed;
286 		goto out;
287 	}
288 
289 	ovs_flow_stats_update(flow, key->tp.flags, skb);
290 	sf_acts = rcu_dereference(flow->sf_acts);
291 	ovs_execute_actions(dp, skb, sf_acts, key);
292 
293 	stats_counter = &stats->n_hit;
294 
295 out:
296 	/* Update datapath statistics. */
297 	u64_stats_update_begin(&stats->syncp);
298 	(*stats_counter)++;
299 	stats->n_mask_hit += n_mask_hit;
300 	u64_stats_update_end(&stats->syncp);
301 }
302 
303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
304 		  const struct sw_flow_key *key,
305 		  const struct dp_upcall_info *upcall_info)
306 {
307 	struct dp_stats_percpu *stats;
308 	int err;
309 
310 	if (upcall_info->portid == 0) {
311 		err = -ENOTCONN;
312 		goto err;
313 	}
314 
315 	if (!skb_is_gso(skb))
316 		err = queue_userspace_packet(dp, skb, key, upcall_info);
317 	else
318 		err = queue_gso_packets(dp, skb, key, upcall_info);
319 	if (err)
320 		goto err;
321 
322 	return 0;
323 
324 err:
325 	stats = this_cpu_ptr(dp->stats_percpu);
326 
327 	u64_stats_update_begin(&stats->syncp);
328 	stats->n_lost++;
329 	u64_stats_update_end(&stats->syncp);
330 
331 	return err;
332 }
333 
334 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
335 			     const struct sw_flow_key *key,
336 			     const struct dp_upcall_info *upcall_info)
337 {
338 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
339 	struct sw_flow_key later_key;
340 	struct sk_buff *segs, *nskb;
341 	struct ovs_skb_cb ovs_cb;
342 	int err;
343 
344 	ovs_cb = *OVS_CB(skb);
345 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
346 	*OVS_CB(skb) = ovs_cb;
347 	if (IS_ERR(segs))
348 		return PTR_ERR(segs);
349 	if (segs == NULL)
350 		return -EINVAL;
351 
352 	if (gso_type & SKB_GSO_UDP) {
353 		/* The initial flow key extracted by ovs_flow_key_extract()
354 		 * in this case is for a first fragment, so we need to
355 		 * properly mark later fragments.
356 		 */
357 		later_key = *key;
358 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
359 	}
360 
361 	/* Queue all of the segments. */
362 	skb = segs;
363 	do {
364 		*OVS_CB(skb) = ovs_cb;
365 		if (gso_type & SKB_GSO_UDP && skb != segs)
366 			key = &later_key;
367 
368 		err = queue_userspace_packet(dp, skb, key, upcall_info);
369 		if (err)
370 			break;
371 
372 	} while ((skb = skb->next));
373 
374 	/* Free all of the segments. */
375 	skb = segs;
376 	do {
377 		nskb = skb->next;
378 		if (err)
379 			kfree_skb(skb);
380 		else
381 			consume_skb(skb);
382 	} while ((skb = nskb));
383 	return err;
384 }
385 
386 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
387 			      unsigned int hdrlen)
388 {
389 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
390 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
391 		+ nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
392 
393 	/* OVS_PACKET_ATTR_USERDATA */
394 	if (upcall_info->userdata)
395 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
396 
397 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
398 	if (upcall_info->egress_tun_info)
399 		size += nla_total_size(ovs_tun_key_attr_size());
400 
401 	return size;
402 }
403 
404 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
405 				  const struct sw_flow_key *key,
406 				  const struct dp_upcall_info *upcall_info)
407 {
408 	struct ovs_header *upcall;
409 	struct sk_buff *nskb = NULL;
410 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
411 	struct nlattr *nla;
412 	struct genl_info info = {
413 		.dst_sk = ovs_dp_get_net(dp)->genl_sock,
414 		.snd_portid = upcall_info->portid,
415 	};
416 	size_t len;
417 	unsigned int hlen;
418 	int err, dp_ifindex;
419 
420 	dp_ifindex = get_dpifindex(dp);
421 	if (!dp_ifindex)
422 		return -ENODEV;
423 
424 	if (skb_vlan_tag_present(skb)) {
425 		nskb = skb_clone(skb, GFP_ATOMIC);
426 		if (!nskb)
427 			return -ENOMEM;
428 
429 		nskb = __vlan_hwaccel_push_inside(nskb);
430 		if (!nskb)
431 			return -ENOMEM;
432 
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, 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 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
467 	BUG_ON(err);
468 
469 	if (upcall_info->userdata)
470 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
471 			  nla_len(upcall_info->userdata),
472 			  nla_data(upcall_info->userdata));
473 
474 	if (upcall_info->egress_tun_info) {
475 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
476 		err = ovs_nla_put_egress_tunnel_key(user_skb,
477 						    upcall_info->egress_tun_info);
478 		BUG_ON(err);
479 		nla_nest_end(user_skb, nla);
480 	}
481 
482 	/* Only reserve room for attribute header, packet data is added
483 	 * in skb_zerocopy() */
484 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
485 		err = -ENOBUFS;
486 		goto out;
487 	}
488 	nla->nla_len = nla_attr_size(skb->len);
489 
490 	err = skb_zerocopy(user_skb, skb, skb->len, hlen);
491 	if (err)
492 		goto out;
493 
494 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
495 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
496 		size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
497 
498 		if (plen > 0)
499 			memset(skb_put(user_skb, plen), 0, plen);
500 	}
501 
502 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
503 
504 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
505 	user_skb = NULL;
506 out:
507 	if (err)
508 		skb_tx_error(skb);
509 	kfree_skb(user_skb);
510 	kfree_skb(nskb);
511 	return err;
512 }
513 
514 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
515 {
516 	struct ovs_header *ovs_header = info->userhdr;
517 	struct nlattr **a = info->attrs;
518 	struct sw_flow_actions *acts;
519 	struct sk_buff *packet;
520 	struct sw_flow *flow;
521 	struct sw_flow_actions *sf_acts;
522 	struct datapath *dp;
523 	struct ethhdr *eth;
524 	struct vport *input_vport;
525 	int len;
526 	int err;
527 	bool log = !a[OVS_PACKET_ATTR_PROBE];
528 
529 	err = -EINVAL;
530 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
531 	    !a[OVS_PACKET_ATTR_ACTIONS])
532 		goto err;
533 
534 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
535 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
536 	err = -ENOMEM;
537 	if (!packet)
538 		goto err;
539 	skb_reserve(packet, NET_IP_ALIGN);
540 
541 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
542 
543 	skb_reset_mac_header(packet);
544 	eth = eth_hdr(packet);
545 
546 	/* Normally, setting the skb 'protocol' field would be handled by a
547 	 * call to eth_type_trans(), but it assumes there's a sending
548 	 * device, which we may not have. */
549 	if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
550 		packet->protocol = eth->h_proto;
551 	else
552 		packet->protocol = htons(ETH_P_802_2);
553 
554 	/* Build an sw_flow for sending this packet. */
555 	flow = ovs_flow_alloc();
556 	err = PTR_ERR(flow);
557 	if (IS_ERR(flow))
558 		goto err_kfree_skb;
559 
560 	err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
561 					     &flow->key, log);
562 	if (err)
563 		goto err_flow_free;
564 
565 	err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
566 				   &flow->key, &acts, log);
567 	if (err)
568 		goto err_flow_free;
569 
570 	rcu_assign_pointer(flow->sf_acts, acts);
571 	OVS_CB(packet)->egress_tun_info = NULL;
572 	packet->priority = flow->key.phy.priority;
573 	packet->mark = flow->key.phy.skb_mark;
574 
575 	rcu_read_lock();
576 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
577 	err = -ENODEV;
578 	if (!dp)
579 		goto err_unlock;
580 
581 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
582 	if (!input_vport)
583 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
584 
585 	if (!input_vport)
586 		goto err_unlock;
587 
588 	OVS_CB(packet)->input_vport = input_vport;
589 	sf_acts = rcu_dereference(flow->sf_acts);
590 
591 	local_bh_disable();
592 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
593 	local_bh_enable();
594 	rcu_read_unlock();
595 
596 	ovs_flow_free(flow, false);
597 	return err;
598 
599 err_unlock:
600 	rcu_read_unlock();
601 err_flow_free:
602 	ovs_flow_free(flow, false);
603 err_kfree_skb:
604 	kfree_skb(packet);
605 err:
606 	return err;
607 }
608 
609 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
610 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
611 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
612 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
613 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
614 };
615 
616 static const struct genl_ops dp_packet_genl_ops[] = {
617 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
618 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
619 	  .policy = packet_policy,
620 	  .doit = ovs_packet_cmd_execute
621 	}
622 };
623 
624 static struct genl_family dp_packet_genl_family = {
625 	.id = GENL_ID_GENERATE,
626 	.hdrsize = sizeof(struct ovs_header),
627 	.name = OVS_PACKET_FAMILY,
628 	.version = OVS_PACKET_VERSION,
629 	.maxattr = OVS_PACKET_ATTR_MAX,
630 	.netnsok = true,
631 	.parallel_ops = true,
632 	.ops = dp_packet_genl_ops,
633 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
634 };
635 
636 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
637 			 struct ovs_dp_megaflow_stats *mega_stats)
638 {
639 	int i;
640 
641 	memset(mega_stats, 0, sizeof(*mega_stats));
642 
643 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
644 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
645 
646 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
647 
648 	for_each_possible_cpu(i) {
649 		const struct dp_stats_percpu *percpu_stats;
650 		struct dp_stats_percpu local_stats;
651 		unsigned int start;
652 
653 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
654 
655 		do {
656 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
657 			local_stats = *percpu_stats;
658 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
659 
660 		stats->n_hit += local_stats.n_hit;
661 		stats->n_missed += local_stats.n_missed;
662 		stats->n_lost += local_stats.n_lost;
663 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
664 	}
665 }
666 
667 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
668 {
669 	return ovs_identifier_is_ufid(sfid) &&
670 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
671 }
672 
673 static bool should_fill_mask(uint32_t ufid_flags)
674 {
675 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
676 }
677 
678 static bool should_fill_actions(uint32_t ufid_flags)
679 {
680 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
681 }
682 
683 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
684 				    const struct sw_flow_id *sfid,
685 				    uint32_t ufid_flags)
686 {
687 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
688 
689 	/* OVS_FLOW_ATTR_UFID */
690 	if (sfid && ovs_identifier_is_ufid(sfid))
691 		len += nla_total_size(sfid->ufid_len);
692 
693 	/* OVS_FLOW_ATTR_KEY */
694 	if (!sfid || should_fill_key(sfid, ufid_flags))
695 		len += nla_total_size(ovs_key_attr_size());
696 
697 	/* OVS_FLOW_ATTR_MASK */
698 	if (should_fill_mask(ufid_flags))
699 		len += nla_total_size(ovs_key_attr_size());
700 
701 	/* OVS_FLOW_ATTR_ACTIONS */
702 	if (should_fill_actions(ufid_flags))
703 		len += nla_total_size(acts->actions_len);
704 
705 	return len
706 		+ nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
707 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
708 		+ nla_total_size(8); /* OVS_FLOW_ATTR_USED */
709 }
710 
711 /* Called with ovs_mutex or RCU read lock. */
712 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
713 				   struct sk_buff *skb)
714 {
715 	struct ovs_flow_stats stats;
716 	__be16 tcp_flags;
717 	unsigned long used;
718 
719 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
720 
721 	if (used &&
722 	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
723 		return -EMSGSIZE;
724 
725 	if (stats.n_packets &&
726 	    nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
727 		return -EMSGSIZE;
728 
729 	if ((u8)ntohs(tcp_flags) &&
730 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
731 		return -EMSGSIZE;
732 
733 	return 0;
734 }
735 
736 /* Called with ovs_mutex or RCU read lock. */
737 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
738 				     struct sk_buff *skb, int skb_orig_len)
739 {
740 	struct nlattr *start;
741 	int err;
742 
743 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
744 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
745 	 * Netlink but individual action lists can be longer than
746 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
747 	 * The userspace caller can always fetch the actions separately if it
748 	 * really wants them.  (Most userspace callers in fact don't care.)
749 	 *
750 	 * This can only fail for dump operations because the skb is always
751 	 * properly sized for single flows.
752 	 */
753 	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
754 	if (start) {
755 		const struct sw_flow_actions *sf_acts;
756 
757 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
758 		err = ovs_nla_put_actions(sf_acts->actions,
759 					  sf_acts->actions_len, skb);
760 
761 		if (!err)
762 			nla_nest_end(skb, start);
763 		else {
764 			if (skb_orig_len)
765 				return err;
766 
767 			nla_nest_cancel(skb, start);
768 		}
769 	} else if (skb_orig_len) {
770 		return -EMSGSIZE;
771 	}
772 
773 	return 0;
774 }
775 
776 /* Called with ovs_mutex or RCU read lock. */
777 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
778 				  struct sk_buff *skb, u32 portid,
779 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
780 {
781 	const int skb_orig_len = skb->len;
782 	struct ovs_header *ovs_header;
783 	int err;
784 
785 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
786 				 flags, cmd);
787 	if (!ovs_header)
788 		return -EMSGSIZE;
789 
790 	ovs_header->dp_ifindex = dp_ifindex;
791 
792 	err = ovs_nla_put_identifier(flow, skb);
793 	if (err)
794 		goto error;
795 
796 	if (should_fill_key(&flow->id, ufid_flags)) {
797 		err = ovs_nla_put_masked_key(flow, skb);
798 		if (err)
799 			goto error;
800 	}
801 
802 	if (should_fill_mask(ufid_flags)) {
803 		err = ovs_nla_put_mask(flow, skb);
804 		if (err)
805 			goto error;
806 	}
807 
808 	err = ovs_flow_cmd_fill_stats(flow, skb);
809 	if (err)
810 		goto error;
811 
812 	if (should_fill_actions(ufid_flags)) {
813 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
814 		if (err)
815 			goto error;
816 	}
817 
818 	genlmsg_end(skb, ovs_header);
819 	return 0;
820 
821 error:
822 	genlmsg_cancel(skb, ovs_header);
823 	return err;
824 }
825 
826 /* May not be called with RCU read lock. */
827 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
828 					       const struct sw_flow_id *sfid,
829 					       struct genl_info *info,
830 					       bool always,
831 					       uint32_t ufid_flags)
832 {
833 	struct sk_buff *skb;
834 	size_t len;
835 
836 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
837 		return NULL;
838 
839 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
840 	skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
841 	if (!skb)
842 		return ERR_PTR(-ENOMEM);
843 
844 	return skb;
845 }
846 
847 /* Called with ovs_mutex. */
848 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
849 					       int dp_ifindex,
850 					       struct genl_info *info, u8 cmd,
851 					       bool always, u32 ufid_flags)
852 {
853 	struct sk_buff *skb;
854 	int retval;
855 
856 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
857 				      &flow->id, info, always, ufid_flags);
858 	if (IS_ERR_OR_NULL(skb))
859 		return skb;
860 
861 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
862 					info->snd_portid, info->snd_seq, 0,
863 					cmd, ufid_flags);
864 	BUG_ON(retval < 0);
865 	return skb;
866 }
867 
868 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
869 {
870 	struct nlattr **a = info->attrs;
871 	struct ovs_header *ovs_header = info->userhdr;
872 	struct sw_flow *flow = NULL, *new_flow;
873 	struct sw_flow_mask mask;
874 	struct sk_buff *reply;
875 	struct datapath *dp;
876 	struct sw_flow_key key;
877 	struct sw_flow_actions *acts;
878 	struct sw_flow_match match;
879 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
880 	int error;
881 	bool log = !a[OVS_FLOW_ATTR_PROBE];
882 
883 	/* Must have key and actions. */
884 	error = -EINVAL;
885 	if (!a[OVS_FLOW_ATTR_KEY]) {
886 		OVS_NLERR(log, "Flow key attr not present in new flow.");
887 		goto error;
888 	}
889 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
890 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
891 		goto error;
892 	}
893 
894 	/* Most of the time we need to allocate a new flow, do it before
895 	 * locking.
896 	 */
897 	new_flow = ovs_flow_alloc();
898 	if (IS_ERR(new_flow)) {
899 		error = PTR_ERR(new_flow);
900 		goto error;
901 	}
902 
903 	/* Extract key. */
904 	ovs_match_init(&match, &key, &mask);
905 	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
906 				  a[OVS_FLOW_ATTR_MASK], log);
907 	if (error)
908 		goto err_kfree_flow;
909 
910 	ovs_flow_mask_key(&new_flow->key, &key, &mask);
911 
912 	/* Extract flow identifier. */
913 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
914 				       &key, log);
915 	if (error)
916 		goto err_kfree_flow;
917 
918 	/* Validate actions. */
919 	error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
920 				     &acts, log);
921 	if (error) {
922 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
923 		goto err_kfree_flow;
924 	}
925 
926 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
927 					ufid_flags);
928 	if (IS_ERR(reply)) {
929 		error = PTR_ERR(reply);
930 		goto err_kfree_acts;
931 	}
932 
933 	ovs_lock();
934 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
935 	if (unlikely(!dp)) {
936 		error = -ENODEV;
937 		goto err_unlock_ovs;
938 	}
939 
940 	/* Check if this is a duplicate flow */
941 	if (ovs_identifier_is_ufid(&new_flow->id))
942 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
943 	if (!flow)
944 		flow = ovs_flow_tbl_lookup(&dp->table, &key);
945 	if (likely(!flow)) {
946 		rcu_assign_pointer(new_flow->sf_acts, acts);
947 
948 		/* Put flow in bucket. */
949 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
950 		if (unlikely(error)) {
951 			acts = NULL;
952 			goto err_unlock_ovs;
953 		}
954 
955 		if (unlikely(reply)) {
956 			error = ovs_flow_cmd_fill_info(new_flow,
957 						       ovs_header->dp_ifindex,
958 						       reply, info->snd_portid,
959 						       info->snd_seq, 0,
960 						       OVS_FLOW_CMD_NEW,
961 						       ufid_flags);
962 			BUG_ON(error < 0);
963 		}
964 		ovs_unlock();
965 	} else {
966 		struct sw_flow_actions *old_acts;
967 
968 		/* Bail out if we're not allowed to modify an existing flow.
969 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
970 		 * because Generic Netlink treats the latter as a dump
971 		 * request.  We also accept NLM_F_EXCL in case that bug ever
972 		 * gets fixed.
973 		 */
974 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
975 							 | NLM_F_EXCL))) {
976 			error = -EEXIST;
977 			goto err_unlock_ovs;
978 		}
979 		/* The flow identifier has to be the same for flow updates.
980 		 * Look for any overlapping flow.
981 		 */
982 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
983 			if (ovs_identifier_is_key(&flow->id))
984 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
985 								 &match);
986 			else /* UFID matches but key is different */
987 				flow = NULL;
988 			if (!flow) {
989 				error = -ENOENT;
990 				goto err_unlock_ovs;
991 			}
992 		}
993 		/* Update actions. */
994 		old_acts = ovsl_dereference(flow->sf_acts);
995 		rcu_assign_pointer(flow->sf_acts, acts);
996 
997 		if (unlikely(reply)) {
998 			error = ovs_flow_cmd_fill_info(flow,
999 						       ovs_header->dp_ifindex,
1000 						       reply, info->snd_portid,
1001 						       info->snd_seq, 0,
1002 						       OVS_FLOW_CMD_NEW,
1003 						       ufid_flags);
1004 			BUG_ON(error < 0);
1005 		}
1006 		ovs_unlock();
1007 
1008 		ovs_nla_free_flow_actions(old_acts);
1009 		ovs_flow_free(new_flow, false);
1010 	}
1011 
1012 	if (reply)
1013 		ovs_notify(&dp_flow_genl_family, reply, info);
1014 	return 0;
1015 
1016 err_unlock_ovs:
1017 	ovs_unlock();
1018 	kfree_skb(reply);
1019 err_kfree_acts:
1020 	kfree(acts);
1021 err_kfree_flow:
1022 	ovs_flow_free(new_flow, false);
1023 error:
1024 	return error;
1025 }
1026 
1027 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1028 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
1029 						const struct sw_flow_key *key,
1030 						const struct sw_flow_mask *mask,
1031 						bool log)
1032 {
1033 	struct sw_flow_actions *acts;
1034 	struct sw_flow_key masked_key;
1035 	int error;
1036 
1037 	ovs_flow_mask_key(&masked_key, key, mask);
1038 	error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1039 	if (error) {
1040 		OVS_NLERR(log,
1041 			  "Actions may not be safe on all matching packets");
1042 		return ERR_PTR(error);
1043 	}
1044 
1045 	return acts;
1046 }
1047 
1048 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1049 {
1050 	struct nlattr **a = info->attrs;
1051 	struct ovs_header *ovs_header = info->userhdr;
1052 	struct sw_flow_key key;
1053 	struct sw_flow *flow;
1054 	struct sw_flow_mask mask;
1055 	struct sk_buff *reply = NULL;
1056 	struct datapath *dp;
1057 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1058 	struct sw_flow_match match;
1059 	struct sw_flow_id sfid;
1060 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1061 	int error;
1062 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1063 	bool ufid_present;
1064 
1065 	/* Extract key. */
1066 	error = -EINVAL;
1067 	if (!a[OVS_FLOW_ATTR_KEY]) {
1068 		OVS_NLERR(log, "Flow key attribute not present in set flow.");
1069 		goto error;
1070 	}
1071 
1072 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1073 	ovs_match_init(&match, &key, &mask);
1074 	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1075 				  a[OVS_FLOW_ATTR_MASK], log);
1076 	if (error)
1077 		goto error;
1078 
1079 	/* Validate actions. */
1080 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1081 		acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1082 					log);
1083 		if (IS_ERR(acts)) {
1084 			error = PTR_ERR(acts);
1085 			goto error;
1086 		}
1087 
1088 		/* Can allocate before locking if have acts. */
1089 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1090 						ufid_flags);
1091 		if (IS_ERR(reply)) {
1092 			error = PTR_ERR(reply);
1093 			goto err_kfree_acts;
1094 		}
1095 	}
1096 
1097 	ovs_lock();
1098 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1099 	if (unlikely(!dp)) {
1100 		error = -ENODEV;
1101 		goto err_unlock_ovs;
1102 	}
1103 	/* Check that the flow exists. */
1104 	if (ufid_present)
1105 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1106 	else
1107 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1108 	if (unlikely(!flow)) {
1109 		error = -ENOENT;
1110 		goto err_unlock_ovs;
1111 	}
1112 
1113 	/* Update actions, if present. */
1114 	if (likely(acts)) {
1115 		old_acts = ovsl_dereference(flow->sf_acts);
1116 		rcu_assign_pointer(flow->sf_acts, acts);
1117 
1118 		if (unlikely(reply)) {
1119 			error = ovs_flow_cmd_fill_info(flow,
1120 						       ovs_header->dp_ifindex,
1121 						       reply, info->snd_portid,
1122 						       info->snd_seq, 0,
1123 						       OVS_FLOW_CMD_NEW,
1124 						       ufid_flags);
1125 			BUG_ON(error < 0);
1126 		}
1127 	} else {
1128 		/* Could not alloc without acts before locking. */
1129 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1130 						info, OVS_FLOW_CMD_NEW, false,
1131 						ufid_flags);
1132 
1133 		if (unlikely(IS_ERR(reply))) {
1134 			error = PTR_ERR(reply);
1135 			goto err_unlock_ovs;
1136 		}
1137 	}
1138 
1139 	/* Clear stats. */
1140 	if (a[OVS_FLOW_ATTR_CLEAR])
1141 		ovs_flow_stats_clear(flow);
1142 	ovs_unlock();
1143 
1144 	if (reply)
1145 		ovs_notify(&dp_flow_genl_family, reply, info);
1146 	if (old_acts)
1147 		ovs_nla_free_flow_actions(old_acts);
1148 
1149 	return 0;
1150 
1151 err_unlock_ovs:
1152 	ovs_unlock();
1153 	kfree_skb(reply);
1154 err_kfree_acts:
1155 	kfree(acts);
1156 error:
1157 	return error;
1158 }
1159 
1160 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1161 {
1162 	struct nlattr **a = info->attrs;
1163 	struct ovs_header *ovs_header = info->userhdr;
1164 	struct sw_flow_key key;
1165 	struct sk_buff *reply;
1166 	struct sw_flow *flow;
1167 	struct datapath *dp;
1168 	struct sw_flow_match match;
1169 	struct sw_flow_id ufid;
1170 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1171 	int err = 0;
1172 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1173 	bool ufid_present;
1174 
1175 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1176 	if (a[OVS_FLOW_ATTR_KEY]) {
1177 		ovs_match_init(&match, &key, NULL);
1178 		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1179 					log);
1180 	} else if (!ufid_present) {
1181 		OVS_NLERR(log,
1182 			  "Flow get message rejected, Key attribute missing.");
1183 		err = -EINVAL;
1184 	}
1185 	if (err)
1186 		return err;
1187 
1188 	ovs_lock();
1189 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1190 	if (!dp) {
1191 		err = -ENODEV;
1192 		goto unlock;
1193 	}
1194 
1195 	if (ufid_present)
1196 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1197 	else
1198 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1199 	if (!flow) {
1200 		err = -ENOENT;
1201 		goto unlock;
1202 	}
1203 
1204 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1205 					OVS_FLOW_CMD_NEW, true, ufid_flags);
1206 	if (IS_ERR(reply)) {
1207 		err = PTR_ERR(reply);
1208 		goto unlock;
1209 	}
1210 
1211 	ovs_unlock();
1212 	return genlmsg_reply(reply, info);
1213 unlock:
1214 	ovs_unlock();
1215 	return err;
1216 }
1217 
1218 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1219 {
1220 	struct nlattr **a = info->attrs;
1221 	struct ovs_header *ovs_header = info->userhdr;
1222 	struct sw_flow_key key;
1223 	struct sk_buff *reply;
1224 	struct sw_flow *flow = NULL;
1225 	struct datapath *dp;
1226 	struct sw_flow_match match;
1227 	struct sw_flow_id ufid;
1228 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1229 	int err;
1230 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1231 	bool ufid_present;
1232 
1233 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1234 	if (a[OVS_FLOW_ATTR_KEY]) {
1235 		ovs_match_init(&match, &key, NULL);
1236 		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1237 					log);
1238 		if (unlikely(err))
1239 			return err;
1240 	}
1241 
1242 	ovs_lock();
1243 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1244 	if (unlikely(!dp)) {
1245 		err = -ENODEV;
1246 		goto unlock;
1247 	}
1248 
1249 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1250 		err = ovs_flow_tbl_flush(&dp->table);
1251 		goto unlock;
1252 	}
1253 
1254 	if (ufid_present)
1255 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1256 	else
1257 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1258 	if (unlikely(!flow)) {
1259 		err = -ENOENT;
1260 		goto unlock;
1261 	}
1262 
1263 	ovs_flow_tbl_remove(&dp->table, flow);
1264 	ovs_unlock();
1265 
1266 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1267 					&flow->id, info, false, ufid_flags);
1268 	if (likely(reply)) {
1269 		if (likely(!IS_ERR(reply))) {
1270 			rcu_read_lock();	/*To keep RCU checker happy. */
1271 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1272 						     reply, info->snd_portid,
1273 						     info->snd_seq, 0,
1274 						     OVS_FLOW_CMD_DEL,
1275 						     ufid_flags);
1276 			rcu_read_unlock();
1277 			BUG_ON(err < 0);
1278 
1279 			ovs_notify(&dp_flow_genl_family, reply, info);
1280 		} else {
1281 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1282 		}
1283 	}
1284 
1285 	ovs_flow_free(flow, true);
1286 	return 0;
1287 unlock:
1288 	ovs_unlock();
1289 	return err;
1290 }
1291 
1292 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1293 {
1294 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1295 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1296 	struct table_instance *ti;
1297 	struct datapath *dp;
1298 	u32 ufid_flags;
1299 	int err;
1300 
1301 	err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1302 			    OVS_FLOW_ATTR_MAX, flow_policy);
1303 	if (err)
1304 		return err;
1305 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1306 
1307 	rcu_read_lock();
1308 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1309 	if (!dp) {
1310 		rcu_read_unlock();
1311 		return -ENODEV;
1312 	}
1313 
1314 	ti = rcu_dereference(dp->table.ti);
1315 	for (;;) {
1316 		struct sw_flow *flow;
1317 		u32 bucket, obj;
1318 
1319 		bucket = cb->args[0];
1320 		obj = cb->args[1];
1321 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1322 		if (!flow)
1323 			break;
1324 
1325 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1326 					   NETLINK_CB(cb->skb).portid,
1327 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1328 					   OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1329 			break;
1330 
1331 		cb->args[0] = bucket;
1332 		cb->args[1] = obj;
1333 	}
1334 	rcu_read_unlock();
1335 	return skb->len;
1336 }
1337 
1338 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1339 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1340 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1341 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1342 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1343 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1344 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1345 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1346 };
1347 
1348 static const struct genl_ops dp_flow_genl_ops[] = {
1349 	{ .cmd = OVS_FLOW_CMD_NEW,
1350 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1351 	  .policy = flow_policy,
1352 	  .doit = ovs_flow_cmd_new
1353 	},
1354 	{ .cmd = OVS_FLOW_CMD_DEL,
1355 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1356 	  .policy = flow_policy,
1357 	  .doit = ovs_flow_cmd_del
1358 	},
1359 	{ .cmd = OVS_FLOW_CMD_GET,
1360 	  .flags = 0,		    /* OK for unprivileged users. */
1361 	  .policy = flow_policy,
1362 	  .doit = ovs_flow_cmd_get,
1363 	  .dumpit = ovs_flow_cmd_dump
1364 	},
1365 	{ .cmd = OVS_FLOW_CMD_SET,
1366 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1367 	  .policy = flow_policy,
1368 	  .doit = ovs_flow_cmd_set,
1369 	},
1370 };
1371 
1372 static struct genl_family dp_flow_genl_family = {
1373 	.id = GENL_ID_GENERATE,
1374 	.hdrsize = sizeof(struct ovs_header),
1375 	.name = OVS_FLOW_FAMILY,
1376 	.version = OVS_FLOW_VERSION,
1377 	.maxattr = OVS_FLOW_ATTR_MAX,
1378 	.netnsok = true,
1379 	.parallel_ops = true,
1380 	.ops = dp_flow_genl_ops,
1381 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1382 	.mcgrps = &ovs_dp_flow_multicast_group,
1383 	.n_mcgrps = 1,
1384 };
1385 
1386 static size_t ovs_dp_cmd_msg_size(void)
1387 {
1388 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1389 
1390 	msgsize += nla_total_size(IFNAMSIZ);
1391 	msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1392 	msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1393 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1394 
1395 	return msgsize;
1396 }
1397 
1398 /* Called with ovs_mutex. */
1399 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1400 				u32 portid, u32 seq, u32 flags, u8 cmd)
1401 {
1402 	struct ovs_header *ovs_header;
1403 	struct ovs_dp_stats dp_stats;
1404 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1405 	int err;
1406 
1407 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1408 				   flags, cmd);
1409 	if (!ovs_header)
1410 		goto error;
1411 
1412 	ovs_header->dp_ifindex = get_dpifindex(dp);
1413 
1414 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1415 	if (err)
1416 		goto nla_put_failure;
1417 
1418 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1419 	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1420 			&dp_stats))
1421 		goto nla_put_failure;
1422 
1423 	if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1424 			sizeof(struct ovs_dp_megaflow_stats),
1425 			&dp_megaflow_stats))
1426 		goto nla_put_failure;
1427 
1428 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1429 		goto nla_put_failure;
1430 
1431 	genlmsg_end(skb, ovs_header);
1432 	return 0;
1433 
1434 nla_put_failure:
1435 	genlmsg_cancel(skb, ovs_header);
1436 error:
1437 	return -EMSGSIZE;
1438 }
1439 
1440 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1441 {
1442 	return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1443 }
1444 
1445 /* Called with rcu_read_lock or ovs_mutex. */
1446 static struct datapath *lookup_datapath(struct net *net,
1447 					const struct ovs_header *ovs_header,
1448 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1449 {
1450 	struct datapath *dp;
1451 
1452 	if (!a[OVS_DP_ATTR_NAME])
1453 		dp = get_dp(net, ovs_header->dp_ifindex);
1454 	else {
1455 		struct vport *vport;
1456 
1457 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1458 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1459 	}
1460 	return dp ? dp : ERR_PTR(-ENODEV);
1461 }
1462 
1463 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1464 {
1465 	struct datapath *dp;
1466 
1467 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1468 	if (IS_ERR(dp))
1469 		return;
1470 
1471 	WARN(dp->user_features, "Dropping previously announced user features\n");
1472 	dp->user_features = 0;
1473 }
1474 
1475 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1476 {
1477 	if (a[OVS_DP_ATTR_USER_FEATURES])
1478 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1479 }
1480 
1481 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1482 {
1483 	struct nlattr **a = info->attrs;
1484 	struct vport_parms parms;
1485 	struct sk_buff *reply;
1486 	struct datapath *dp;
1487 	struct vport *vport;
1488 	struct ovs_net *ovs_net;
1489 	int err, i;
1490 
1491 	err = -EINVAL;
1492 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1493 		goto err;
1494 
1495 	reply = ovs_dp_cmd_alloc_info(info);
1496 	if (!reply)
1497 		return -ENOMEM;
1498 
1499 	err = -ENOMEM;
1500 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1501 	if (dp == NULL)
1502 		goto err_free_reply;
1503 
1504 	ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1505 
1506 	/* Allocate table. */
1507 	err = ovs_flow_tbl_init(&dp->table);
1508 	if (err)
1509 		goto err_free_dp;
1510 
1511 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1512 	if (!dp->stats_percpu) {
1513 		err = -ENOMEM;
1514 		goto err_destroy_table;
1515 	}
1516 
1517 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1518 			    GFP_KERNEL);
1519 	if (!dp->ports) {
1520 		err = -ENOMEM;
1521 		goto err_destroy_percpu;
1522 	}
1523 
1524 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1525 		INIT_HLIST_HEAD(&dp->ports[i]);
1526 
1527 	/* Set up our datapath device. */
1528 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1529 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1530 	parms.options = NULL;
1531 	parms.dp = dp;
1532 	parms.port_no = OVSP_LOCAL;
1533 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1534 
1535 	ovs_dp_change(dp, a);
1536 
1537 	/* So far only local changes have been made, now need the lock. */
1538 	ovs_lock();
1539 
1540 	vport = new_vport(&parms);
1541 	if (IS_ERR(vport)) {
1542 		err = PTR_ERR(vport);
1543 		if (err == -EBUSY)
1544 			err = -EEXIST;
1545 
1546 		if (err == -EEXIST) {
1547 			/* An outdated user space instance that does not understand
1548 			 * the concept of user_features has attempted to create a new
1549 			 * datapath and is likely to reuse it. Drop all user features.
1550 			 */
1551 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1552 				ovs_dp_reset_user_features(skb, info);
1553 		}
1554 
1555 		goto err_destroy_ports_array;
1556 	}
1557 
1558 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1559 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1560 	BUG_ON(err < 0);
1561 
1562 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1563 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1564 
1565 	ovs_unlock();
1566 
1567 	ovs_notify(&dp_datapath_genl_family, reply, info);
1568 	return 0;
1569 
1570 err_destroy_ports_array:
1571 	ovs_unlock();
1572 	kfree(dp->ports);
1573 err_destroy_percpu:
1574 	free_percpu(dp->stats_percpu);
1575 err_destroy_table:
1576 	ovs_flow_tbl_destroy(&dp->table);
1577 err_free_dp:
1578 	release_net(ovs_dp_get_net(dp));
1579 	kfree(dp);
1580 err_free_reply:
1581 	kfree_skb(reply);
1582 err:
1583 	return err;
1584 }
1585 
1586 /* Called with ovs_mutex. */
1587 static void __dp_destroy(struct datapath *dp)
1588 {
1589 	int i;
1590 
1591 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1592 		struct vport *vport;
1593 		struct hlist_node *n;
1594 
1595 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1596 			if (vport->port_no != OVSP_LOCAL)
1597 				ovs_dp_detach_port(vport);
1598 	}
1599 
1600 	list_del_rcu(&dp->list_node);
1601 
1602 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1603 	 * all ports in datapath are destroyed first before freeing datapath.
1604 	 */
1605 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1606 
1607 	/* RCU destroy the flow table */
1608 	call_rcu(&dp->rcu, destroy_dp_rcu);
1609 }
1610 
1611 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1612 {
1613 	struct sk_buff *reply;
1614 	struct datapath *dp;
1615 	int err;
1616 
1617 	reply = ovs_dp_cmd_alloc_info(info);
1618 	if (!reply)
1619 		return -ENOMEM;
1620 
1621 	ovs_lock();
1622 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1623 	err = PTR_ERR(dp);
1624 	if (IS_ERR(dp))
1625 		goto err_unlock_free;
1626 
1627 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1628 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1629 	BUG_ON(err < 0);
1630 
1631 	__dp_destroy(dp);
1632 	ovs_unlock();
1633 
1634 	ovs_notify(&dp_datapath_genl_family, reply, info);
1635 
1636 	return 0;
1637 
1638 err_unlock_free:
1639 	ovs_unlock();
1640 	kfree_skb(reply);
1641 	return err;
1642 }
1643 
1644 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1645 {
1646 	struct sk_buff *reply;
1647 	struct datapath *dp;
1648 	int err;
1649 
1650 	reply = ovs_dp_cmd_alloc_info(info);
1651 	if (!reply)
1652 		return -ENOMEM;
1653 
1654 	ovs_lock();
1655 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1656 	err = PTR_ERR(dp);
1657 	if (IS_ERR(dp))
1658 		goto err_unlock_free;
1659 
1660 	ovs_dp_change(dp, info->attrs);
1661 
1662 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1663 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1664 	BUG_ON(err < 0);
1665 
1666 	ovs_unlock();
1667 	ovs_notify(&dp_datapath_genl_family, reply, info);
1668 
1669 	return 0;
1670 
1671 err_unlock_free:
1672 	ovs_unlock();
1673 	kfree_skb(reply);
1674 	return err;
1675 }
1676 
1677 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1678 {
1679 	struct sk_buff *reply;
1680 	struct datapath *dp;
1681 	int err;
1682 
1683 	reply = ovs_dp_cmd_alloc_info(info);
1684 	if (!reply)
1685 		return -ENOMEM;
1686 
1687 	ovs_lock();
1688 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1689 	if (IS_ERR(dp)) {
1690 		err = PTR_ERR(dp);
1691 		goto err_unlock_free;
1692 	}
1693 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1694 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1695 	BUG_ON(err < 0);
1696 	ovs_unlock();
1697 
1698 	return genlmsg_reply(reply, info);
1699 
1700 err_unlock_free:
1701 	ovs_unlock();
1702 	kfree_skb(reply);
1703 	return err;
1704 }
1705 
1706 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1707 {
1708 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1709 	struct datapath *dp;
1710 	int skip = cb->args[0];
1711 	int i = 0;
1712 
1713 	ovs_lock();
1714 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1715 		if (i >= skip &&
1716 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1717 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1718 					 OVS_DP_CMD_NEW) < 0)
1719 			break;
1720 		i++;
1721 	}
1722 	ovs_unlock();
1723 
1724 	cb->args[0] = i;
1725 
1726 	return skb->len;
1727 }
1728 
1729 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1730 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1731 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1732 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1733 };
1734 
1735 static const struct genl_ops dp_datapath_genl_ops[] = {
1736 	{ .cmd = OVS_DP_CMD_NEW,
1737 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1738 	  .policy = datapath_policy,
1739 	  .doit = ovs_dp_cmd_new
1740 	},
1741 	{ .cmd = OVS_DP_CMD_DEL,
1742 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1743 	  .policy = datapath_policy,
1744 	  .doit = ovs_dp_cmd_del
1745 	},
1746 	{ .cmd = OVS_DP_CMD_GET,
1747 	  .flags = 0,		    /* OK for unprivileged users. */
1748 	  .policy = datapath_policy,
1749 	  .doit = ovs_dp_cmd_get,
1750 	  .dumpit = ovs_dp_cmd_dump
1751 	},
1752 	{ .cmd = OVS_DP_CMD_SET,
1753 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1754 	  .policy = datapath_policy,
1755 	  .doit = ovs_dp_cmd_set,
1756 	},
1757 };
1758 
1759 static struct genl_family dp_datapath_genl_family = {
1760 	.id = GENL_ID_GENERATE,
1761 	.hdrsize = sizeof(struct ovs_header),
1762 	.name = OVS_DATAPATH_FAMILY,
1763 	.version = OVS_DATAPATH_VERSION,
1764 	.maxattr = OVS_DP_ATTR_MAX,
1765 	.netnsok = true,
1766 	.parallel_ops = true,
1767 	.ops = dp_datapath_genl_ops,
1768 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1769 	.mcgrps = &ovs_dp_datapath_multicast_group,
1770 	.n_mcgrps = 1,
1771 };
1772 
1773 /* Called with ovs_mutex or RCU read lock. */
1774 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1775 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1776 {
1777 	struct ovs_header *ovs_header;
1778 	struct ovs_vport_stats vport_stats;
1779 	int err;
1780 
1781 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1782 				 flags, cmd);
1783 	if (!ovs_header)
1784 		return -EMSGSIZE;
1785 
1786 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1787 
1788 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1789 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1790 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1791 			   vport->ops->get_name(vport)))
1792 		goto nla_put_failure;
1793 
1794 	ovs_vport_get_stats(vport, &vport_stats);
1795 	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1796 		    &vport_stats))
1797 		goto nla_put_failure;
1798 
1799 	if (ovs_vport_get_upcall_portids(vport, skb))
1800 		goto nla_put_failure;
1801 
1802 	err = ovs_vport_get_options(vport, skb);
1803 	if (err == -EMSGSIZE)
1804 		goto error;
1805 
1806 	genlmsg_end(skb, ovs_header);
1807 	return 0;
1808 
1809 nla_put_failure:
1810 	err = -EMSGSIZE;
1811 error:
1812 	genlmsg_cancel(skb, ovs_header);
1813 	return err;
1814 }
1815 
1816 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1817 {
1818 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1819 }
1820 
1821 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1822 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1823 					 u32 seq, u8 cmd)
1824 {
1825 	struct sk_buff *skb;
1826 	int retval;
1827 
1828 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1829 	if (!skb)
1830 		return ERR_PTR(-ENOMEM);
1831 
1832 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1833 	BUG_ON(retval < 0);
1834 
1835 	return skb;
1836 }
1837 
1838 /* Called with ovs_mutex or RCU read lock. */
1839 static struct vport *lookup_vport(struct net *net,
1840 				  const struct ovs_header *ovs_header,
1841 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1842 {
1843 	struct datapath *dp;
1844 	struct vport *vport;
1845 
1846 	if (a[OVS_VPORT_ATTR_NAME]) {
1847 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1848 		if (!vport)
1849 			return ERR_PTR(-ENODEV);
1850 		if (ovs_header->dp_ifindex &&
1851 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1852 			return ERR_PTR(-ENODEV);
1853 		return vport;
1854 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1855 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1856 
1857 		if (port_no >= DP_MAX_PORTS)
1858 			return ERR_PTR(-EFBIG);
1859 
1860 		dp = get_dp(net, ovs_header->dp_ifindex);
1861 		if (!dp)
1862 			return ERR_PTR(-ENODEV);
1863 
1864 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1865 		if (!vport)
1866 			return ERR_PTR(-ENODEV);
1867 		return vport;
1868 	} else
1869 		return ERR_PTR(-EINVAL);
1870 }
1871 
1872 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1873 {
1874 	struct nlattr **a = info->attrs;
1875 	struct ovs_header *ovs_header = info->userhdr;
1876 	struct vport_parms parms;
1877 	struct sk_buff *reply;
1878 	struct vport *vport;
1879 	struct datapath *dp;
1880 	u32 port_no;
1881 	int err;
1882 
1883 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1884 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1885 		return -EINVAL;
1886 
1887 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
1888 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1889 	if (port_no >= DP_MAX_PORTS)
1890 		return -EFBIG;
1891 
1892 	reply = ovs_vport_cmd_alloc_info();
1893 	if (!reply)
1894 		return -ENOMEM;
1895 
1896 	ovs_lock();
1897 restart:
1898 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1899 	err = -ENODEV;
1900 	if (!dp)
1901 		goto exit_unlock_free;
1902 
1903 	if (port_no) {
1904 		vport = ovs_vport_ovsl(dp, port_no);
1905 		err = -EBUSY;
1906 		if (vport)
1907 			goto exit_unlock_free;
1908 	} else {
1909 		for (port_no = 1; ; port_no++) {
1910 			if (port_no >= DP_MAX_PORTS) {
1911 				err = -EFBIG;
1912 				goto exit_unlock_free;
1913 			}
1914 			vport = ovs_vport_ovsl(dp, port_no);
1915 			if (!vport)
1916 				break;
1917 		}
1918 	}
1919 
1920 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1921 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1922 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1923 	parms.dp = dp;
1924 	parms.port_no = port_no;
1925 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1926 
1927 	vport = new_vport(&parms);
1928 	err = PTR_ERR(vport);
1929 	if (IS_ERR(vport)) {
1930 		if (err == -EAGAIN)
1931 			goto restart;
1932 		goto exit_unlock_free;
1933 	}
1934 
1935 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1936 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1937 	BUG_ON(err < 0);
1938 	ovs_unlock();
1939 
1940 	ovs_notify(&dp_vport_genl_family, reply, info);
1941 	return 0;
1942 
1943 exit_unlock_free:
1944 	ovs_unlock();
1945 	kfree_skb(reply);
1946 	return err;
1947 }
1948 
1949 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1950 {
1951 	struct nlattr **a = info->attrs;
1952 	struct sk_buff *reply;
1953 	struct vport *vport;
1954 	int err;
1955 
1956 	reply = ovs_vport_cmd_alloc_info();
1957 	if (!reply)
1958 		return -ENOMEM;
1959 
1960 	ovs_lock();
1961 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1962 	err = PTR_ERR(vport);
1963 	if (IS_ERR(vport))
1964 		goto exit_unlock_free;
1965 
1966 	if (a[OVS_VPORT_ATTR_TYPE] &&
1967 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1968 		err = -EINVAL;
1969 		goto exit_unlock_free;
1970 	}
1971 
1972 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
1973 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1974 		if (err)
1975 			goto exit_unlock_free;
1976 	}
1977 
1978 
1979 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1980 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1981 
1982 		err = ovs_vport_set_upcall_portids(vport, ids);
1983 		if (err)
1984 			goto exit_unlock_free;
1985 	}
1986 
1987 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1988 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1989 	BUG_ON(err < 0);
1990 
1991 	ovs_unlock();
1992 	ovs_notify(&dp_vport_genl_family, reply, info);
1993 	return 0;
1994 
1995 exit_unlock_free:
1996 	ovs_unlock();
1997 	kfree_skb(reply);
1998 	return err;
1999 }
2000 
2001 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2002 {
2003 	struct nlattr **a = info->attrs;
2004 	struct sk_buff *reply;
2005 	struct vport *vport;
2006 	int err;
2007 
2008 	reply = ovs_vport_cmd_alloc_info();
2009 	if (!reply)
2010 		return -ENOMEM;
2011 
2012 	ovs_lock();
2013 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2014 	err = PTR_ERR(vport);
2015 	if (IS_ERR(vport))
2016 		goto exit_unlock_free;
2017 
2018 	if (vport->port_no == OVSP_LOCAL) {
2019 		err = -EINVAL;
2020 		goto exit_unlock_free;
2021 	}
2022 
2023 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2024 				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2025 	BUG_ON(err < 0);
2026 	ovs_dp_detach_port(vport);
2027 	ovs_unlock();
2028 
2029 	ovs_notify(&dp_vport_genl_family, reply, info);
2030 	return 0;
2031 
2032 exit_unlock_free:
2033 	ovs_unlock();
2034 	kfree_skb(reply);
2035 	return err;
2036 }
2037 
2038 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2039 {
2040 	struct nlattr **a = info->attrs;
2041 	struct ovs_header *ovs_header = info->userhdr;
2042 	struct sk_buff *reply;
2043 	struct vport *vport;
2044 	int err;
2045 
2046 	reply = ovs_vport_cmd_alloc_info();
2047 	if (!reply)
2048 		return -ENOMEM;
2049 
2050 	rcu_read_lock();
2051 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2052 	err = PTR_ERR(vport);
2053 	if (IS_ERR(vport))
2054 		goto exit_unlock_free;
2055 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2056 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2057 	BUG_ON(err < 0);
2058 	rcu_read_unlock();
2059 
2060 	return genlmsg_reply(reply, info);
2061 
2062 exit_unlock_free:
2063 	rcu_read_unlock();
2064 	kfree_skb(reply);
2065 	return err;
2066 }
2067 
2068 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2069 {
2070 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2071 	struct datapath *dp;
2072 	int bucket = cb->args[0], skip = cb->args[1];
2073 	int i, j = 0;
2074 
2075 	rcu_read_lock();
2076 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2077 	if (!dp) {
2078 		rcu_read_unlock();
2079 		return -ENODEV;
2080 	}
2081 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2082 		struct vport *vport;
2083 
2084 		j = 0;
2085 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2086 			if (j >= skip &&
2087 			    ovs_vport_cmd_fill_info(vport, skb,
2088 						    NETLINK_CB(cb->skb).portid,
2089 						    cb->nlh->nlmsg_seq,
2090 						    NLM_F_MULTI,
2091 						    OVS_VPORT_CMD_NEW) < 0)
2092 				goto out;
2093 
2094 			j++;
2095 		}
2096 		skip = 0;
2097 	}
2098 out:
2099 	rcu_read_unlock();
2100 
2101 	cb->args[0] = i;
2102 	cb->args[1] = j;
2103 
2104 	return skb->len;
2105 }
2106 
2107 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2108 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2109 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2110 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2111 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2112 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2113 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2114 };
2115 
2116 static const struct genl_ops dp_vport_genl_ops[] = {
2117 	{ .cmd = OVS_VPORT_CMD_NEW,
2118 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2119 	  .policy = vport_policy,
2120 	  .doit = ovs_vport_cmd_new
2121 	},
2122 	{ .cmd = OVS_VPORT_CMD_DEL,
2123 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2124 	  .policy = vport_policy,
2125 	  .doit = ovs_vport_cmd_del
2126 	},
2127 	{ .cmd = OVS_VPORT_CMD_GET,
2128 	  .flags = 0,		    /* OK for unprivileged users. */
2129 	  .policy = vport_policy,
2130 	  .doit = ovs_vport_cmd_get,
2131 	  .dumpit = ovs_vport_cmd_dump
2132 	},
2133 	{ .cmd = OVS_VPORT_CMD_SET,
2134 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2135 	  .policy = vport_policy,
2136 	  .doit = ovs_vport_cmd_set,
2137 	},
2138 };
2139 
2140 struct genl_family dp_vport_genl_family = {
2141 	.id = GENL_ID_GENERATE,
2142 	.hdrsize = sizeof(struct ovs_header),
2143 	.name = OVS_VPORT_FAMILY,
2144 	.version = OVS_VPORT_VERSION,
2145 	.maxattr = OVS_VPORT_ATTR_MAX,
2146 	.netnsok = true,
2147 	.parallel_ops = true,
2148 	.ops = dp_vport_genl_ops,
2149 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2150 	.mcgrps = &ovs_dp_vport_multicast_group,
2151 	.n_mcgrps = 1,
2152 };
2153 
2154 static struct genl_family * const dp_genl_families[] = {
2155 	&dp_datapath_genl_family,
2156 	&dp_vport_genl_family,
2157 	&dp_flow_genl_family,
2158 	&dp_packet_genl_family,
2159 };
2160 
2161 static void dp_unregister_genl(int n_families)
2162 {
2163 	int i;
2164 
2165 	for (i = 0; i < n_families; i++)
2166 		genl_unregister_family(dp_genl_families[i]);
2167 }
2168 
2169 static int dp_register_genl(void)
2170 {
2171 	int err;
2172 	int i;
2173 
2174 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2175 
2176 		err = genl_register_family(dp_genl_families[i]);
2177 		if (err)
2178 			goto error;
2179 	}
2180 
2181 	return 0;
2182 
2183 error:
2184 	dp_unregister_genl(i);
2185 	return err;
2186 }
2187 
2188 static int __net_init ovs_init_net(struct net *net)
2189 {
2190 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2191 
2192 	INIT_LIST_HEAD(&ovs_net->dps);
2193 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2194 	return 0;
2195 }
2196 
2197 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2198 					    struct list_head *head)
2199 {
2200 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2201 	struct datapath *dp;
2202 
2203 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2204 		int i;
2205 
2206 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2207 			struct vport *vport;
2208 
2209 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2210 				struct netdev_vport *netdev_vport;
2211 
2212 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2213 					continue;
2214 
2215 				netdev_vport = netdev_vport_priv(vport);
2216 				if (dev_net(netdev_vport->dev) == dnet)
2217 					list_add(&vport->detach_list, head);
2218 			}
2219 		}
2220 	}
2221 }
2222 
2223 static void __net_exit ovs_exit_net(struct net *dnet)
2224 {
2225 	struct datapath *dp, *dp_next;
2226 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2227 	struct vport *vport, *vport_next;
2228 	struct net *net;
2229 	LIST_HEAD(head);
2230 
2231 	ovs_lock();
2232 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2233 		__dp_destroy(dp);
2234 
2235 	rtnl_lock();
2236 	for_each_net(net)
2237 		list_vports_from_net(net, dnet, &head);
2238 	rtnl_unlock();
2239 
2240 	/* Detach all vports from given namespace. */
2241 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2242 		list_del(&vport->detach_list);
2243 		ovs_dp_detach_port(vport);
2244 	}
2245 
2246 	ovs_unlock();
2247 
2248 	cancel_work_sync(&ovs_net->dp_notify_work);
2249 }
2250 
2251 static struct pernet_operations ovs_net_ops = {
2252 	.init = ovs_init_net,
2253 	.exit = ovs_exit_net,
2254 	.id   = &ovs_net_id,
2255 	.size = sizeof(struct ovs_net),
2256 };
2257 
2258 static int __init dp_init(void)
2259 {
2260 	int err;
2261 
2262 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2263 
2264 	pr_info("Open vSwitch switching datapath\n");
2265 
2266 	err = action_fifos_init();
2267 	if (err)
2268 		goto error;
2269 
2270 	err = ovs_internal_dev_rtnl_link_register();
2271 	if (err)
2272 		goto error_action_fifos_exit;
2273 
2274 	err = ovs_flow_init();
2275 	if (err)
2276 		goto error_unreg_rtnl_link;
2277 
2278 	err = ovs_vport_init();
2279 	if (err)
2280 		goto error_flow_exit;
2281 
2282 	err = register_pernet_device(&ovs_net_ops);
2283 	if (err)
2284 		goto error_vport_exit;
2285 
2286 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2287 	if (err)
2288 		goto error_netns_exit;
2289 
2290 	err = ovs_netdev_init();
2291 	if (err)
2292 		goto error_unreg_notifier;
2293 
2294 	err = dp_register_genl();
2295 	if (err < 0)
2296 		goto error_unreg_netdev;
2297 
2298 	return 0;
2299 
2300 error_unreg_netdev:
2301 	ovs_netdev_exit();
2302 error_unreg_notifier:
2303 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2304 error_netns_exit:
2305 	unregister_pernet_device(&ovs_net_ops);
2306 error_vport_exit:
2307 	ovs_vport_exit();
2308 error_flow_exit:
2309 	ovs_flow_exit();
2310 error_unreg_rtnl_link:
2311 	ovs_internal_dev_rtnl_link_unregister();
2312 error_action_fifos_exit:
2313 	action_fifos_exit();
2314 error:
2315 	return err;
2316 }
2317 
2318 static void dp_cleanup(void)
2319 {
2320 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2321 	ovs_netdev_exit();
2322 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2323 	unregister_pernet_device(&ovs_net_ops);
2324 	rcu_barrier();
2325 	ovs_vport_exit();
2326 	ovs_flow_exit();
2327 	ovs_internal_dev_rtnl_link_unregister();
2328 	action_fifos_exit();
2329 }
2330 
2331 module_init(dp_init);
2332 module_exit(dp_cleanup);
2333 
2334 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2335 MODULE_LICENSE("GPL");
2336