xref: /linux/net/openvswitch/datapath.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 ovs_vport_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 = 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 	kfree(dp->ports);
207 	kfree(dp);
208 }
209 
210 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
211 					    u16 port_no)
212 {
213 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
214 }
215 
216 /* Called with ovs_mutex or RCU read lock. */
217 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
218 {
219 	struct vport *vport;
220 	struct hlist_head *head;
221 
222 	head = vport_hash_bucket(dp, port_no);
223 	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
224 		if (vport->port_no == port_no)
225 			return vport;
226 	}
227 	return NULL;
228 }
229 
230 /* Called with ovs_mutex. */
231 static struct vport *new_vport(const struct vport_parms *parms)
232 {
233 	struct vport *vport;
234 
235 	vport = ovs_vport_add(parms);
236 	if (!IS_ERR(vport)) {
237 		struct datapath *dp = parms->dp;
238 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
239 
240 		hlist_add_head_rcu(&vport->dp_hash_node, head);
241 	}
242 	return vport;
243 }
244 
245 void ovs_dp_detach_port(struct vport *p)
246 {
247 	ASSERT_OVSL();
248 
249 	/* First drop references to device. */
250 	hlist_del_rcu(&p->dp_hash_node);
251 
252 	/* Then destroy it. */
253 	ovs_vport_del(p);
254 }
255 
256 /* Must be called with rcu_read_lock. */
257 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
258 {
259 	const struct vport *p = OVS_CB(skb)->input_vport;
260 	struct datapath *dp = p->dp;
261 	struct sw_flow *flow;
262 	struct sw_flow_actions *sf_acts;
263 	struct dp_stats_percpu *stats;
264 	u64 *stats_counter;
265 	u32 n_mask_hit;
266 
267 	stats = this_cpu_ptr(dp->stats_percpu);
268 
269 	/* Look up flow. */
270 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
271 	if (unlikely(!flow)) {
272 		struct dp_upcall_info upcall;
273 		int error;
274 
275 		memset(&upcall, 0, sizeof(upcall));
276 		upcall.cmd = OVS_PACKET_CMD_MISS;
277 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
278 		upcall.mru = OVS_CB(skb)->mru;
279 		error = ovs_dp_upcall(dp, skb, key, &upcall);
280 		if (unlikely(error))
281 			kfree_skb(skb);
282 		else
283 			consume_skb(skb);
284 		stats_counter = &stats->n_missed;
285 		goto out;
286 	}
287 
288 	ovs_flow_stats_update(flow, key->tp.flags, skb);
289 	sf_acts = rcu_dereference(flow->sf_acts);
290 	ovs_execute_actions(dp, skb, sf_acts, key);
291 
292 	stats_counter = &stats->n_hit;
293 
294 out:
295 	/* Update datapath statistics. */
296 	u64_stats_update_begin(&stats->syncp);
297 	(*stats_counter)++;
298 	stats->n_mask_hit += n_mask_hit;
299 	u64_stats_update_end(&stats->syncp);
300 }
301 
302 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
303 		  const struct sw_flow_key *key,
304 		  const struct dp_upcall_info *upcall_info)
305 {
306 	struct dp_stats_percpu *stats;
307 	int err;
308 
309 	if (upcall_info->portid == 0) {
310 		err = -ENOTCONN;
311 		goto err;
312 	}
313 
314 	if (!skb_is_gso(skb))
315 		err = queue_userspace_packet(dp, skb, key, upcall_info);
316 	else
317 		err = queue_gso_packets(dp, skb, key, upcall_info);
318 	if (err)
319 		goto err;
320 
321 	return 0;
322 
323 err:
324 	stats = this_cpu_ptr(dp->stats_percpu);
325 
326 	u64_stats_update_begin(&stats->syncp);
327 	stats->n_lost++;
328 	u64_stats_update_end(&stats->syncp);
329 
330 	return err;
331 }
332 
333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
334 			     const struct sw_flow_key *key,
335 			     const struct dp_upcall_info *upcall_info)
336 {
337 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
338 	struct sw_flow_key later_key;
339 	struct sk_buff *segs, *nskb;
340 	struct ovs_skb_cb ovs_cb;
341 	int err;
342 
343 	ovs_cb = *OVS_CB(skb);
344 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
345 	*OVS_CB(skb) = ovs_cb;
346 	if (IS_ERR(segs))
347 		return PTR_ERR(segs);
348 	if (segs == NULL)
349 		return -EINVAL;
350 
351 	if (gso_type & SKB_GSO_UDP) {
352 		/* The initial flow key extracted by ovs_flow_key_extract()
353 		 * in this case is for a first fragment, so we need to
354 		 * properly mark later fragments.
355 		 */
356 		later_key = *key;
357 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
358 	}
359 
360 	/* Queue all of the segments. */
361 	skb = segs;
362 	do {
363 		*OVS_CB(skb) = ovs_cb;
364 		if (gso_type & SKB_GSO_UDP && skb != segs)
365 			key = &later_key;
366 
367 		err = queue_userspace_packet(dp, skb, key, upcall_info);
368 		if (err)
369 			break;
370 
371 	} while ((skb = skb->next));
372 
373 	/* Free all of the segments. */
374 	skb = segs;
375 	do {
376 		nskb = skb->next;
377 		if (err)
378 			kfree_skb(skb);
379 		else
380 			consume_skb(skb);
381 	} while ((skb = nskb));
382 	return err;
383 }
384 
385 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
386 			      unsigned int hdrlen)
387 {
388 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
389 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
390 		+ nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
391 
392 	/* OVS_PACKET_ATTR_USERDATA */
393 	if (upcall_info->userdata)
394 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
395 
396 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
397 	if (upcall_info->egress_tun_info)
398 		size += nla_total_size(ovs_tun_key_attr_size());
399 
400 	/* OVS_PACKET_ATTR_ACTIONS */
401 	if (upcall_info->actions_len)
402 		size += nla_total_size(upcall_info->actions_len);
403 
404 	/* OVS_PACKET_ATTR_MRU */
405 	if (upcall_info->mru)
406 		size += nla_total_size(sizeof(upcall_info->mru));
407 
408 	return size;
409 }
410 
411 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
412 {
413 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
414 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
415 
416 		if (plen > 0)
417 			memset(skb_put(skb, plen), 0, plen);
418 	}
419 }
420 
421 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
422 				  const struct sw_flow_key *key,
423 				  const struct dp_upcall_info *upcall_info)
424 {
425 	struct ovs_header *upcall;
426 	struct sk_buff *nskb = NULL;
427 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
428 	struct nlattr *nla;
429 	struct genl_info info = {
430 		.dst_sk = ovs_dp_get_net(dp)->genl_sock,
431 		.snd_portid = upcall_info->portid,
432 	};
433 	size_t len;
434 	unsigned int hlen;
435 	int err, dp_ifindex;
436 
437 	dp_ifindex = get_dpifindex(dp);
438 	if (!dp_ifindex)
439 		return -ENODEV;
440 
441 	if (skb_vlan_tag_present(skb)) {
442 		nskb = skb_clone(skb, GFP_ATOMIC);
443 		if (!nskb)
444 			return -ENOMEM;
445 
446 		nskb = __vlan_hwaccel_push_inside(nskb);
447 		if (!nskb)
448 			return -ENOMEM;
449 
450 		skb = nskb;
451 	}
452 
453 	if (nla_attr_size(skb->len) > USHRT_MAX) {
454 		err = -EFBIG;
455 		goto out;
456 	}
457 
458 	/* Complete checksum if needed */
459 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
460 	    (err = skb_checksum_help(skb)))
461 		goto out;
462 
463 	/* Older versions of OVS user space enforce alignment of the last
464 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
465 	 * padding logic. Only perform zerocopy if padding is not required.
466 	 */
467 	if (dp->user_features & OVS_DP_F_UNALIGNED)
468 		hlen = skb_zerocopy_headlen(skb);
469 	else
470 		hlen = skb->len;
471 
472 	len = upcall_msg_size(upcall_info, hlen);
473 	user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
474 	if (!user_skb) {
475 		err = -ENOMEM;
476 		goto out;
477 	}
478 
479 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
480 			     0, upcall_info->cmd);
481 	upcall->dp_ifindex = dp_ifindex;
482 
483 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
484 	BUG_ON(err);
485 
486 	if (upcall_info->userdata)
487 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
488 			  nla_len(upcall_info->userdata),
489 			  nla_data(upcall_info->userdata));
490 
491 	if (upcall_info->egress_tun_info) {
492 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
493 		err = ovs_nla_put_tunnel_info(user_skb,
494 					      upcall_info->egress_tun_info);
495 		BUG_ON(err);
496 		nla_nest_end(user_skb, nla);
497 	}
498 
499 	if (upcall_info->actions_len) {
500 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
501 		err = ovs_nla_put_actions(upcall_info->actions,
502 					  upcall_info->actions_len,
503 					  user_skb);
504 		if (!err)
505 			nla_nest_end(user_skb, nla);
506 		else
507 			nla_nest_cancel(user_skb, nla);
508 	}
509 
510 	/* Add OVS_PACKET_ATTR_MRU */
511 	if (upcall_info->mru) {
512 		if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
513 				upcall_info->mru)) {
514 			err = -ENOBUFS;
515 			goto out;
516 		}
517 		pad_packet(dp, user_skb);
518 	}
519 
520 	/* Only reserve room for attribute header, packet data is added
521 	 * in skb_zerocopy() */
522 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
523 		err = -ENOBUFS;
524 		goto out;
525 	}
526 	nla->nla_len = nla_attr_size(skb->len);
527 
528 	err = skb_zerocopy(user_skb, skb, skb->len, hlen);
529 	if (err)
530 		goto out;
531 
532 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
533 	pad_packet(dp, user_skb);
534 
535 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
536 
537 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
538 	user_skb = NULL;
539 out:
540 	if (err)
541 		skb_tx_error(skb);
542 	kfree_skb(user_skb);
543 	kfree_skb(nskb);
544 	return err;
545 }
546 
547 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
548 {
549 	struct ovs_header *ovs_header = info->userhdr;
550 	struct net *net = sock_net(skb->sk);
551 	struct nlattr **a = info->attrs;
552 	struct sw_flow_actions *acts;
553 	struct sk_buff *packet;
554 	struct sw_flow *flow;
555 	struct sw_flow_actions *sf_acts;
556 	struct datapath *dp;
557 	struct ethhdr *eth;
558 	struct vport *input_vport;
559 	u16 mru = 0;
560 	int len;
561 	int err;
562 	bool log = !a[OVS_PACKET_ATTR_PROBE];
563 
564 	err = -EINVAL;
565 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
566 	    !a[OVS_PACKET_ATTR_ACTIONS])
567 		goto err;
568 
569 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
570 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
571 	err = -ENOMEM;
572 	if (!packet)
573 		goto err;
574 	skb_reserve(packet, NET_IP_ALIGN);
575 
576 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
577 
578 	skb_reset_mac_header(packet);
579 	eth = eth_hdr(packet);
580 
581 	/* Normally, setting the skb 'protocol' field would be handled by a
582 	 * call to eth_type_trans(), but it assumes there's a sending
583 	 * device, which we may not have. */
584 	if (eth_proto_is_802_3(eth->h_proto))
585 		packet->protocol = eth->h_proto;
586 	else
587 		packet->protocol = htons(ETH_P_802_2);
588 
589 	/* Set packet's mru */
590 	if (a[OVS_PACKET_ATTR_MRU]) {
591 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
592 		packet->ignore_df = 1;
593 	}
594 	OVS_CB(packet)->mru = mru;
595 
596 	/* Build an sw_flow for sending this packet. */
597 	flow = ovs_flow_alloc();
598 	err = PTR_ERR(flow);
599 	if (IS_ERR(flow))
600 		goto err_kfree_skb;
601 
602 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
603 					     packet, &flow->key, log);
604 	if (err)
605 		goto err_flow_free;
606 
607 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
608 				   &flow->key, &acts, log);
609 	if (err)
610 		goto err_flow_free;
611 
612 	rcu_assign_pointer(flow->sf_acts, acts);
613 	packet->priority = flow->key.phy.priority;
614 	packet->mark = flow->key.phy.skb_mark;
615 
616 	rcu_read_lock();
617 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
618 	err = -ENODEV;
619 	if (!dp)
620 		goto err_unlock;
621 
622 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
623 	if (!input_vport)
624 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
625 
626 	if (!input_vport)
627 		goto err_unlock;
628 
629 	packet->dev = input_vport->dev;
630 	OVS_CB(packet)->input_vport = input_vport;
631 	sf_acts = rcu_dereference(flow->sf_acts);
632 
633 	local_bh_disable();
634 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
635 	local_bh_enable();
636 	rcu_read_unlock();
637 
638 	ovs_flow_free(flow, false);
639 	return err;
640 
641 err_unlock:
642 	rcu_read_unlock();
643 err_flow_free:
644 	ovs_flow_free(flow, false);
645 err_kfree_skb:
646 	kfree_skb(packet);
647 err:
648 	return err;
649 }
650 
651 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
652 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
653 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
654 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
655 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
656 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
657 };
658 
659 static const struct genl_ops dp_packet_genl_ops[] = {
660 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
661 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
662 	  .policy = packet_policy,
663 	  .doit = ovs_packet_cmd_execute
664 	}
665 };
666 
667 static struct genl_family dp_packet_genl_family = {
668 	.id = GENL_ID_GENERATE,
669 	.hdrsize = sizeof(struct ovs_header),
670 	.name = OVS_PACKET_FAMILY,
671 	.version = OVS_PACKET_VERSION,
672 	.maxattr = OVS_PACKET_ATTR_MAX,
673 	.netnsok = true,
674 	.parallel_ops = true,
675 	.ops = dp_packet_genl_ops,
676 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
677 };
678 
679 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
680 			 struct ovs_dp_megaflow_stats *mega_stats)
681 {
682 	int i;
683 
684 	memset(mega_stats, 0, sizeof(*mega_stats));
685 
686 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
687 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
688 
689 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
690 
691 	for_each_possible_cpu(i) {
692 		const struct dp_stats_percpu *percpu_stats;
693 		struct dp_stats_percpu local_stats;
694 		unsigned int start;
695 
696 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
697 
698 		do {
699 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
700 			local_stats = *percpu_stats;
701 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
702 
703 		stats->n_hit += local_stats.n_hit;
704 		stats->n_missed += local_stats.n_missed;
705 		stats->n_lost += local_stats.n_lost;
706 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
707 	}
708 }
709 
710 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
711 {
712 	return ovs_identifier_is_ufid(sfid) &&
713 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
714 }
715 
716 static bool should_fill_mask(uint32_t ufid_flags)
717 {
718 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
719 }
720 
721 static bool should_fill_actions(uint32_t ufid_flags)
722 {
723 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
724 }
725 
726 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
727 				    const struct sw_flow_id *sfid,
728 				    uint32_t ufid_flags)
729 {
730 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
731 
732 	/* OVS_FLOW_ATTR_UFID */
733 	if (sfid && ovs_identifier_is_ufid(sfid))
734 		len += nla_total_size(sfid->ufid_len);
735 
736 	/* OVS_FLOW_ATTR_KEY */
737 	if (!sfid || should_fill_key(sfid, ufid_flags))
738 		len += nla_total_size(ovs_key_attr_size());
739 
740 	/* OVS_FLOW_ATTR_MASK */
741 	if (should_fill_mask(ufid_flags))
742 		len += nla_total_size(ovs_key_attr_size());
743 
744 	/* OVS_FLOW_ATTR_ACTIONS */
745 	if (should_fill_actions(ufid_flags))
746 		len += nla_total_size(acts->orig_len);
747 
748 	return len
749 		+ nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
750 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
751 		+ nla_total_size(8); /* OVS_FLOW_ATTR_USED */
752 }
753 
754 /* Called with ovs_mutex or RCU read lock. */
755 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
756 				   struct sk_buff *skb)
757 {
758 	struct ovs_flow_stats stats;
759 	__be16 tcp_flags;
760 	unsigned long used;
761 
762 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
763 
764 	if (used &&
765 	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
766 		return -EMSGSIZE;
767 
768 	if (stats.n_packets &&
769 	    nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
770 		return -EMSGSIZE;
771 
772 	if ((u8)ntohs(tcp_flags) &&
773 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
774 		return -EMSGSIZE;
775 
776 	return 0;
777 }
778 
779 /* Called with ovs_mutex or RCU read lock. */
780 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
781 				     struct sk_buff *skb, int skb_orig_len)
782 {
783 	struct nlattr *start;
784 	int err;
785 
786 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
787 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
788 	 * Netlink but individual action lists can be longer than
789 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
790 	 * The userspace caller can always fetch the actions separately if it
791 	 * really wants them.  (Most userspace callers in fact don't care.)
792 	 *
793 	 * This can only fail for dump operations because the skb is always
794 	 * properly sized for single flows.
795 	 */
796 	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
797 	if (start) {
798 		const struct sw_flow_actions *sf_acts;
799 
800 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
801 		err = ovs_nla_put_actions(sf_acts->actions,
802 					  sf_acts->actions_len, skb);
803 
804 		if (!err)
805 			nla_nest_end(skb, start);
806 		else {
807 			if (skb_orig_len)
808 				return err;
809 
810 			nla_nest_cancel(skb, start);
811 		}
812 	} else if (skb_orig_len) {
813 		return -EMSGSIZE;
814 	}
815 
816 	return 0;
817 }
818 
819 /* Called with ovs_mutex or RCU read lock. */
820 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
821 				  struct sk_buff *skb, u32 portid,
822 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
823 {
824 	const int skb_orig_len = skb->len;
825 	struct ovs_header *ovs_header;
826 	int err;
827 
828 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
829 				 flags, cmd);
830 	if (!ovs_header)
831 		return -EMSGSIZE;
832 
833 	ovs_header->dp_ifindex = dp_ifindex;
834 
835 	err = ovs_nla_put_identifier(flow, skb);
836 	if (err)
837 		goto error;
838 
839 	if (should_fill_key(&flow->id, ufid_flags)) {
840 		err = ovs_nla_put_masked_key(flow, skb);
841 		if (err)
842 			goto error;
843 	}
844 
845 	if (should_fill_mask(ufid_flags)) {
846 		err = ovs_nla_put_mask(flow, skb);
847 		if (err)
848 			goto error;
849 	}
850 
851 	err = ovs_flow_cmd_fill_stats(flow, skb);
852 	if (err)
853 		goto error;
854 
855 	if (should_fill_actions(ufid_flags)) {
856 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
857 		if (err)
858 			goto error;
859 	}
860 
861 	genlmsg_end(skb, ovs_header);
862 	return 0;
863 
864 error:
865 	genlmsg_cancel(skb, ovs_header);
866 	return err;
867 }
868 
869 /* May not be called with RCU read lock. */
870 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
871 					       const struct sw_flow_id *sfid,
872 					       struct genl_info *info,
873 					       bool always,
874 					       uint32_t ufid_flags)
875 {
876 	struct sk_buff *skb;
877 	size_t len;
878 
879 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
880 		return NULL;
881 
882 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
883 	skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
884 	if (!skb)
885 		return ERR_PTR(-ENOMEM);
886 
887 	return skb;
888 }
889 
890 /* Called with ovs_mutex. */
891 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
892 					       int dp_ifindex,
893 					       struct genl_info *info, u8 cmd,
894 					       bool always, u32 ufid_flags)
895 {
896 	struct sk_buff *skb;
897 	int retval;
898 
899 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
900 				      &flow->id, info, always, ufid_flags);
901 	if (IS_ERR_OR_NULL(skb))
902 		return skb;
903 
904 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
905 					info->snd_portid, info->snd_seq, 0,
906 					cmd, ufid_flags);
907 	BUG_ON(retval < 0);
908 	return skb;
909 }
910 
911 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
912 {
913 	struct net *net = sock_net(skb->sk);
914 	struct nlattr **a = info->attrs;
915 	struct ovs_header *ovs_header = info->userhdr;
916 	struct sw_flow *flow = NULL, *new_flow;
917 	struct sw_flow_mask mask;
918 	struct sk_buff *reply;
919 	struct datapath *dp;
920 	struct sw_flow_key key;
921 	struct sw_flow_actions *acts;
922 	struct sw_flow_match match;
923 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
924 	int error;
925 	bool log = !a[OVS_FLOW_ATTR_PROBE];
926 
927 	/* Must have key and actions. */
928 	error = -EINVAL;
929 	if (!a[OVS_FLOW_ATTR_KEY]) {
930 		OVS_NLERR(log, "Flow key attr not present in new flow.");
931 		goto error;
932 	}
933 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
934 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
935 		goto error;
936 	}
937 
938 	/* Most of the time we need to allocate a new flow, do it before
939 	 * locking.
940 	 */
941 	new_flow = ovs_flow_alloc();
942 	if (IS_ERR(new_flow)) {
943 		error = PTR_ERR(new_flow);
944 		goto error;
945 	}
946 
947 	/* Extract key. */
948 	ovs_match_init(&match, &key, &mask);
949 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
950 				  a[OVS_FLOW_ATTR_MASK], log);
951 	if (error)
952 		goto err_kfree_flow;
953 
954 	ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
955 
956 	/* Extract flow identifier. */
957 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
958 				       &key, log);
959 	if (error)
960 		goto err_kfree_flow;
961 
962 	/* Validate actions. */
963 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
964 				     &new_flow->key, &acts, log);
965 	if (error) {
966 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
967 		goto err_kfree_flow;
968 	}
969 
970 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
971 					ufid_flags);
972 	if (IS_ERR(reply)) {
973 		error = PTR_ERR(reply);
974 		goto err_kfree_acts;
975 	}
976 
977 	ovs_lock();
978 	dp = get_dp(net, ovs_header->dp_ifindex);
979 	if (unlikely(!dp)) {
980 		error = -ENODEV;
981 		goto err_unlock_ovs;
982 	}
983 
984 	/* Check if this is a duplicate flow */
985 	if (ovs_identifier_is_ufid(&new_flow->id))
986 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
987 	if (!flow)
988 		flow = ovs_flow_tbl_lookup(&dp->table, &key);
989 	if (likely(!flow)) {
990 		rcu_assign_pointer(new_flow->sf_acts, acts);
991 
992 		/* Put flow in bucket. */
993 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
994 		if (unlikely(error)) {
995 			acts = NULL;
996 			goto err_unlock_ovs;
997 		}
998 
999 		if (unlikely(reply)) {
1000 			error = ovs_flow_cmd_fill_info(new_flow,
1001 						       ovs_header->dp_ifindex,
1002 						       reply, info->snd_portid,
1003 						       info->snd_seq, 0,
1004 						       OVS_FLOW_CMD_NEW,
1005 						       ufid_flags);
1006 			BUG_ON(error < 0);
1007 		}
1008 		ovs_unlock();
1009 	} else {
1010 		struct sw_flow_actions *old_acts;
1011 
1012 		/* Bail out if we're not allowed to modify an existing flow.
1013 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1014 		 * because Generic Netlink treats the latter as a dump
1015 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1016 		 * gets fixed.
1017 		 */
1018 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1019 							 | NLM_F_EXCL))) {
1020 			error = -EEXIST;
1021 			goto err_unlock_ovs;
1022 		}
1023 		/* The flow identifier has to be the same for flow updates.
1024 		 * Look for any overlapping flow.
1025 		 */
1026 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1027 			if (ovs_identifier_is_key(&flow->id))
1028 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1029 								 &match);
1030 			else /* UFID matches but key is different */
1031 				flow = NULL;
1032 			if (!flow) {
1033 				error = -ENOENT;
1034 				goto err_unlock_ovs;
1035 			}
1036 		}
1037 		/* Update actions. */
1038 		old_acts = ovsl_dereference(flow->sf_acts);
1039 		rcu_assign_pointer(flow->sf_acts, acts);
1040 
1041 		if (unlikely(reply)) {
1042 			error = ovs_flow_cmd_fill_info(flow,
1043 						       ovs_header->dp_ifindex,
1044 						       reply, info->snd_portid,
1045 						       info->snd_seq, 0,
1046 						       OVS_FLOW_CMD_NEW,
1047 						       ufid_flags);
1048 			BUG_ON(error < 0);
1049 		}
1050 		ovs_unlock();
1051 
1052 		ovs_nla_free_flow_actions_rcu(old_acts);
1053 		ovs_flow_free(new_flow, false);
1054 	}
1055 
1056 	if (reply)
1057 		ovs_notify(&dp_flow_genl_family, reply, info);
1058 	return 0;
1059 
1060 err_unlock_ovs:
1061 	ovs_unlock();
1062 	kfree_skb(reply);
1063 err_kfree_acts:
1064 	ovs_nla_free_flow_actions(acts);
1065 err_kfree_flow:
1066 	ovs_flow_free(new_flow, false);
1067 error:
1068 	return error;
1069 }
1070 
1071 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1072 static struct sw_flow_actions *get_flow_actions(struct net *net,
1073 						const struct nlattr *a,
1074 						const struct sw_flow_key *key,
1075 						const struct sw_flow_mask *mask,
1076 						bool log)
1077 {
1078 	struct sw_flow_actions *acts;
1079 	struct sw_flow_key masked_key;
1080 	int error;
1081 
1082 	ovs_flow_mask_key(&masked_key, key, true, mask);
1083 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1084 	if (error) {
1085 		OVS_NLERR(log,
1086 			  "Actions may not be safe on all matching packets");
1087 		return ERR_PTR(error);
1088 	}
1089 
1090 	return acts;
1091 }
1092 
1093 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1094 {
1095 	struct net *net = sock_net(skb->sk);
1096 	struct nlattr **a = info->attrs;
1097 	struct ovs_header *ovs_header = info->userhdr;
1098 	struct sw_flow_key key;
1099 	struct sw_flow *flow;
1100 	struct sw_flow_mask mask;
1101 	struct sk_buff *reply = NULL;
1102 	struct datapath *dp;
1103 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1104 	struct sw_flow_match match;
1105 	struct sw_flow_id sfid;
1106 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1107 	int error;
1108 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1109 	bool ufid_present;
1110 
1111 	/* Extract key. */
1112 	error = -EINVAL;
1113 	if (!a[OVS_FLOW_ATTR_KEY]) {
1114 		OVS_NLERR(log, "Flow key attribute not present in set flow.");
1115 		goto error;
1116 	}
1117 
1118 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1119 	ovs_match_init(&match, &key, &mask);
1120 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1121 				  a[OVS_FLOW_ATTR_MASK], log);
1122 	if (error)
1123 		goto error;
1124 
1125 	/* Validate actions. */
1126 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1127 		acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1128 					&mask, log);
1129 		if (IS_ERR(acts)) {
1130 			error = PTR_ERR(acts);
1131 			goto error;
1132 		}
1133 
1134 		/* Can allocate before locking if have acts. */
1135 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1136 						ufid_flags);
1137 		if (IS_ERR(reply)) {
1138 			error = PTR_ERR(reply);
1139 			goto err_kfree_acts;
1140 		}
1141 	}
1142 
1143 	ovs_lock();
1144 	dp = get_dp(net, ovs_header->dp_ifindex);
1145 	if (unlikely(!dp)) {
1146 		error = -ENODEV;
1147 		goto err_unlock_ovs;
1148 	}
1149 	/* Check that the flow exists. */
1150 	if (ufid_present)
1151 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1152 	else
1153 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1154 	if (unlikely(!flow)) {
1155 		error = -ENOENT;
1156 		goto err_unlock_ovs;
1157 	}
1158 
1159 	/* Update actions, if present. */
1160 	if (likely(acts)) {
1161 		old_acts = ovsl_dereference(flow->sf_acts);
1162 		rcu_assign_pointer(flow->sf_acts, acts);
1163 
1164 		if (unlikely(reply)) {
1165 			error = ovs_flow_cmd_fill_info(flow,
1166 						       ovs_header->dp_ifindex,
1167 						       reply, info->snd_portid,
1168 						       info->snd_seq, 0,
1169 						       OVS_FLOW_CMD_NEW,
1170 						       ufid_flags);
1171 			BUG_ON(error < 0);
1172 		}
1173 	} else {
1174 		/* Could not alloc without acts before locking. */
1175 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1176 						info, OVS_FLOW_CMD_NEW, false,
1177 						ufid_flags);
1178 
1179 		if (unlikely(IS_ERR(reply))) {
1180 			error = PTR_ERR(reply);
1181 			goto err_unlock_ovs;
1182 		}
1183 	}
1184 
1185 	/* Clear stats. */
1186 	if (a[OVS_FLOW_ATTR_CLEAR])
1187 		ovs_flow_stats_clear(flow);
1188 	ovs_unlock();
1189 
1190 	if (reply)
1191 		ovs_notify(&dp_flow_genl_family, reply, info);
1192 	if (old_acts)
1193 		ovs_nla_free_flow_actions_rcu(old_acts);
1194 
1195 	return 0;
1196 
1197 err_unlock_ovs:
1198 	ovs_unlock();
1199 	kfree_skb(reply);
1200 err_kfree_acts:
1201 	ovs_nla_free_flow_actions(acts);
1202 error:
1203 	return error;
1204 }
1205 
1206 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1207 {
1208 	struct nlattr **a = info->attrs;
1209 	struct ovs_header *ovs_header = info->userhdr;
1210 	struct net *net = sock_net(skb->sk);
1211 	struct sw_flow_key key;
1212 	struct sk_buff *reply;
1213 	struct sw_flow *flow;
1214 	struct datapath *dp;
1215 	struct sw_flow_match match;
1216 	struct sw_flow_id ufid;
1217 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1218 	int err = 0;
1219 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1220 	bool ufid_present;
1221 
1222 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1223 	if (a[OVS_FLOW_ATTR_KEY]) {
1224 		ovs_match_init(&match, &key, NULL);
1225 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1226 					log);
1227 	} else if (!ufid_present) {
1228 		OVS_NLERR(log,
1229 			  "Flow get message rejected, Key attribute missing.");
1230 		err = -EINVAL;
1231 	}
1232 	if (err)
1233 		return err;
1234 
1235 	ovs_lock();
1236 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1237 	if (!dp) {
1238 		err = -ENODEV;
1239 		goto unlock;
1240 	}
1241 
1242 	if (ufid_present)
1243 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1244 	else
1245 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1246 	if (!flow) {
1247 		err = -ENOENT;
1248 		goto unlock;
1249 	}
1250 
1251 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1252 					OVS_FLOW_CMD_NEW, true, ufid_flags);
1253 	if (IS_ERR(reply)) {
1254 		err = PTR_ERR(reply);
1255 		goto unlock;
1256 	}
1257 
1258 	ovs_unlock();
1259 	return genlmsg_reply(reply, info);
1260 unlock:
1261 	ovs_unlock();
1262 	return err;
1263 }
1264 
1265 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1266 {
1267 	struct nlattr **a = info->attrs;
1268 	struct ovs_header *ovs_header = info->userhdr;
1269 	struct net *net = sock_net(skb->sk);
1270 	struct sw_flow_key key;
1271 	struct sk_buff *reply;
1272 	struct sw_flow *flow = NULL;
1273 	struct datapath *dp;
1274 	struct sw_flow_match match;
1275 	struct sw_flow_id ufid;
1276 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1277 	int err;
1278 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1279 	bool ufid_present;
1280 
1281 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1282 	if (a[OVS_FLOW_ATTR_KEY]) {
1283 		ovs_match_init(&match, &key, NULL);
1284 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1285 					NULL, log);
1286 		if (unlikely(err))
1287 			return err;
1288 	}
1289 
1290 	ovs_lock();
1291 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1292 	if (unlikely(!dp)) {
1293 		err = -ENODEV;
1294 		goto unlock;
1295 	}
1296 
1297 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1298 		err = ovs_flow_tbl_flush(&dp->table);
1299 		goto unlock;
1300 	}
1301 
1302 	if (ufid_present)
1303 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1304 	else
1305 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1306 	if (unlikely(!flow)) {
1307 		err = -ENOENT;
1308 		goto unlock;
1309 	}
1310 
1311 	ovs_flow_tbl_remove(&dp->table, flow);
1312 	ovs_unlock();
1313 
1314 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1315 					&flow->id, info, false, ufid_flags);
1316 	if (likely(reply)) {
1317 		if (likely(!IS_ERR(reply))) {
1318 			rcu_read_lock();	/*To keep RCU checker happy. */
1319 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1320 						     reply, info->snd_portid,
1321 						     info->snd_seq, 0,
1322 						     OVS_FLOW_CMD_DEL,
1323 						     ufid_flags);
1324 			rcu_read_unlock();
1325 			BUG_ON(err < 0);
1326 
1327 			ovs_notify(&dp_flow_genl_family, reply, info);
1328 		} else {
1329 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1330 		}
1331 	}
1332 
1333 	ovs_flow_free(flow, true);
1334 	return 0;
1335 unlock:
1336 	ovs_unlock();
1337 	return err;
1338 }
1339 
1340 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1341 {
1342 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1343 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1344 	struct table_instance *ti;
1345 	struct datapath *dp;
1346 	u32 ufid_flags;
1347 	int err;
1348 
1349 	err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1350 			    OVS_FLOW_ATTR_MAX, flow_policy);
1351 	if (err)
1352 		return err;
1353 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1354 
1355 	rcu_read_lock();
1356 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1357 	if (!dp) {
1358 		rcu_read_unlock();
1359 		return -ENODEV;
1360 	}
1361 
1362 	ti = rcu_dereference(dp->table.ti);
1363 	for (;;) {
1364 		struct sw_flow *flow;
1365 		u32 bucket, obj;
1366 
1367 		bucket = cb->args[0];
1368 		obj = cb->args[1];
1369 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1370 		if (!flow)
1371 			break;
1372 
1373 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1374 					   NETLINK_CB(cb->skb).portid,
1375 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1376 					   OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1377 			break;
1378 
1379 		cb->args[0] = bucket;
1380 		cb->args[1] = obj;
1381 	}
1382 	rcu_read_unlock();
1383 	return skb->len;
1384 }
1385 
1386 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1387 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1388 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1389 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1390 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1391 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1392 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1393 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1394 };
1395 
1396 static const struct genl_ops dp_flow_genl_ops[] = {
1397 	{ .cmd = OVS_FLOW_CMD_NEW,
1398 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1399 	  .policy = flow_policy,
1400 	  .doit = ovs_flow_cmd_new
1401 	},
1402 	{ .cmd = OVS_FLOW_CMD_DEL,
1403 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1404 	  .policy = flow_policy,
1405 	  .doit = ovs_flow_cmd_del
1406 	},
1407 	{ .cmd = OVS_FLOW_CMD_GET,
1408 	  .flags = 0,		    /* OK for unprivileged users. */
1409 	  .policy = flow_policy,
1410 	  .doit = ovs_flow_cmd_get,
1411 	  .dumpit = ovs_flow_cmd_dump
1412 	},
1413 	{ .cmd = OVS_FLOW_CMD_SET,
1414 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1415 	  .policy = flow_policy,
1416 	  .doit = ovs_flow_cmd_set,
1417 	},
1418 };
1419 
1420 static struct genl_family dp_flow_genl_family = {
1421 	.id = GENL_ID_GENERATE,
1422 	.hdrsize = sizeof(struct ovs_header),
1423 	.name = OVS_FLOW_FAMILY,
1424 	.version = OVS_FLOW_VERSION,
1425 	.maxattr = OVS_FLOW_ATTR_MAX,
1426 	.netnsok = true,
1427 	.parallel_ops = true,
1428 	.ops = dp_flow_genl_ops,
1429 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1430 	.mcgrps = &ovs_dp_flow_multicast_group,
1431 	.n_mcgrps = 1,
1432 };
1433 
1434 static size_t ovs_dp_cmd_msg_size(void)
1435 {
1436 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1437 
1438 	msgsize += nla_total_size(IFNAMSIZ);
1439 	msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1440 	msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1441 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1442 
1443 	return msgsize;
1444 }
1445 
1446 /* Called with ovs_mutex. */
1447 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1448 				u32 portid, u32 seq, u32 flags, u8 cmd)
1449 {
1450 	struct ovs_header *ovs_header;
1451 	struct ovs_dp_stats dp_stats;
1452 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1453 	int err;
1454 
1455 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1456 				   flags, cmd);
1457 	if (!ovs_header)
1458 		goto error;
1459 
1460 	ovs_header->dp_ifindex = get_dpifindex(dp);
1461 
1462 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1463 	if (err)
1464 		goto nla_put_failure;
1465 
1466 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1467 	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1468 			&dp_stats))
1469 		goto nla_put_failure;
1470 
1471 	if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1472 			sizeof(struct ovs_dp_megaflow_stats),
1473 			&dp_megaflow_stats))
1474 		goto nla_put_failure;
1475 
1476 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1477 		goto nla_put_failure;
1478 
1479 	genlmsg_end(skb, ovs_header);
1480 	return 0;
1481 
1482 nla_put_failure:
1483 	genlmsg_cancel(skb, ovs_header);
1484 error:
1485 	return -EMSGSIZE;
1486 }
1487 
1488 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1489 {
1490 	return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1491 }
1492 
1493 /* Called with rcu_read_lock or ovs_mutex. */
1494 static struct datapath *lookup_datapath(struct net *net,
1495 					const struct ovs_header *ovs_header,
1496 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1497 {
1498 	struct datapath *dp;
1499 
1500 	if (!a[OVS_DP_ATTR_NAME])
1501 		dp = get_dp(net, ovs_header->dp_ifindex);
1502 	else {
1503 		struct vport *vport;
1504 
1505 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1506 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1507 	}
1508 	return dp ? dp : ERR_PTR(-ENODEV);
1509 }
1510 
1511 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1512 {
1513 	struct datapath *dp;
1514 
1515 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1516 	if (IS_ERR(dp))
1517 		return;
1518 
1519 	WARN(dp->user_features, "Dropping previously announced user features\n");
1520 	dp->user_features = 0;
1521 }
1522 
1523 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1524 {
1525 	if (a[OVS_DP_ATTR_USER_FEATURES])
1526 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1527 }
1528 
1529 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1530 {
1531 	struct nlattr **a = info->attrs;
1532 	struct vport_parms parms;
1533 	struct sk_buff *reply;
1534 	struct datapath *dp;
1535 	struct vport *vport;
1536 	struct ovs_net *ovs_net;
1537 	int err, i;
1538 
1539 	err = -EINVAL;
1540 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1541 		goto err;
1542 
1543 	reply = ovs_dp_cmd_alloc_info(info);
1544 	if (!reply)
1545 		return -ENOMEM;
1546 
1547 	err = -ENOMEM;
1548 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1549 	if (dp == NULL)
1550 		goto err_free_reply;
1551 
1552 	ovs_dp_set_net(dp, sock_net(skb->sk));
1553 
1554 	/* Allocate table. */
1555 	err = ovs_flow_tbl_init(&dp->table);
1556 	if (err)
1557 		goto err_free_dp;
1558 
1559 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1560 	if (!dp->stats_percpu) {
1561 		err = -ENOMEM;
1562 		goto err_destroy_table;
1563 	}
1564 
1565 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1566 			    GFP_KERNEL);
1567 	if (!dp->ports) {
1568 		err = -ENOMEM;
1569 		goto err_destroy_percpu;
1570 	}
1571 
1572 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1573 		INIT_HLIST_HEAD(&dp->ports[i]);
1574 
1575 	/* Set up our datapath device. */
1576 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1577 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1578 	parms.options = NULL;
1579 	parms.dp = dp;
1580 	parms.port_no = OVSP_LOCAL;
1581 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1582 
1583 	ovs_dp_change(dp, a);
1584 
1585 	/* So far only local changes have been made, now need the lock. */
1586 	ovs_lock();
1587 
1588 	vport = new_vport(&parms);
1589 	if (IS_ERR(vport)) {
1590 		err = PTR_ERR(vport);
1591 		if (err == -EBUSY)
1592 			err = -EEXIST;
1593 
1594 		if (err == -EEXIST) {
1595 			/* An outdated user space instance that does not understand
1596 			 * the concept of user_features has attempted to create a new
1597 			 * datapath and is likely to reuse it. Drop all user features.
1598 			 */
1599 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1600 				ovs_dp_reset_user_features(skb, info);
1601 		}
1602 
1603 		goto err_destroy_ports_array;
1604 	}
1605 
1606 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1607 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1608 	BUG_ON(err < 0);
1609 
1610 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1611 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1612 
1613 	ovs_unlock();
1614 
1615 	ovs_notify(&dp_datapath_genl_family, reply, info);
1616 	return 0;
1617 
1618 err_destroy_ports_array:
1619 	ovs_unlock();
1620 	kfree(dp->ports);
1621 err_destroy_percpu:
1622 	free_percpu(dp->stats_percpu);
1623 err_destroy_table:
1624 	ovs_flow_tbl_destroy(&dp->table);
1625 err_free_dp:
1626 	kfree(dp);
1627 err_free_reply:
1628 	kfree_skb(reply);
1629 err:
1630 	return err;
1631 }
1632 
1633 /* Called with ovs_mutex. */
1634 static void __dp_destroy(struct datapath *dp)
1635 {
1636 	int i;
1637 
1638 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1639 		struct vport *vport;
1640 		struct hlist_node *n;
1641 
1642 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1643 			if (vport->port_no != OVSP_LOCAL)
1644 				ovs_dp_detach_port(vport);
1645 	}
1646 
1647 	list_del_rcu(&dp->list_node);
1648 
1649 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1650 	 * all ports in datapath are destroyed first before freeing datapath.
1651 	 */
1652 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1653 
1654 	/* RCU destroy the flow table */
1655 	call_rcu(&dp->rcu, destroy_dp_rcu);
1656 }
1657 
1658 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1659 {
1660 	struct sk_buff *reply;
1661 	struct datapath *dp;
1662 	int err;
1663 
1664 	reply = ovs_dp_cmd_alloc_info(info);
1665 	if (!reply)
1666 		return -ENOMEM;
1667 
1668 	ovs_lock();
1669 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1670 	err = PTR_ERR(dp);
1671 	if (IS_ERR(dp))
1672 		goto err_unlock_free;
1673 
1674 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1675 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1676 	BUG_ON(err < 0);
1677 
1678 	__dp_destroy(dp);
1679 	ovs_unlock();
1680 
1681 	ovs_notify(&dp_datapath_genl_family, reply, info);
1682 
1683 	return 0;
1684 
1685 err_unlock_free:
1686 	ovs_unlock();
1687 	kfree_skb(reply);
1688 	return err;
1689 }
1690 
1691 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1692 {
1693 	struct sk_buff *reply;
1694 	struct datapath *dp;
1695 	int err;
1696 
1697 	reply = ovs_dp_cmd_alloc_info(info);
1698 	if (!reply)
1699 		return -ENOMEM;
1700 
1701 	ovs_lock();
1702 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1703 	err = PTR_ERR(dp);
1704 	if (IS_ERR(dp))
1705 		goto err_unlock_free;
1706 
1707 	ovs_dp_change(dp, info->attrs);
1708 
1709 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1710 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1711 	BUG_ON(err < 0);
1712 
1713 	ovs_unlock();
1714 	ovs_notify(&dp_datapath_genl_family, reply, info);
1715 
1716 	return 0;
1717 
1718 err_unlock_free:
1719 	ovs_unlock();
1720 	kfree_skb(reply);
1721 	return err;
1722 }
1723 
1724 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1725 {
1726 	struct sk_buff *reply;
1727 	struct datapath *dp;
1728 	int err;
1729 
1730 	reply = ovs_dp_cmd_alloc_info(info);
1731 	if (!reply)
1732 		return -ENOMEM;
1733 
1734 	ovs_lock();
1735 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1736 	if (IS_ERR(dp)) {
1737 		err = PTR_ERR(dp);
1738 		goto err_unlock_free;
1739 	}
1740 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1741 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1742 	BUG_ON(err < 0);
1743 	ovs_unlock();
1744 
1745 	return genlmsg_reply(reply, info);
1746 
1747 err_unlock_free:
1748 	ovs_unlock();
1749 	kfree_skb(reply);
1750 	return err;
1751 }
1752 
1753 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1754 {
1755 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1756 	struct datapath *dp;
1757 	int skip = cb->args[0];
1758 	int i = 0;
1759 
1760 	ovs_lock();
1761 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1762 		if (i >= skip &&
1763 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1764 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1765 					 OVS_DP_CMD_NEW) < 0)
1766 			break;
1767 		i++;
1768 	}
1769 	ovs_unlock();
1770 
1771 	cb->args[0] = i;
1772 
1773 	return skb->len;
1774 }
1775 
1776 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1777 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1778 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1779 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1780 };
1781 
1782 static const struct genl_ops dp_datapath_genl_ops[] = {
1783 	{ .cmd = OVS_DP_CMD_NEW,
1784 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1785 	  .policy = datapath_policy,
1786 	  .doit = ovs_dp_cmd_new
1787 	},
1788 	{ .cmd = OVS_DP_CMD_DEL,
1789 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1790 	  .policy = datapath_policy,
1791 	  .doit = ovs_dp_cmd_del
1792 	},
1793 	{ .cmd = OVS_DP_CMD_GET,
1794 	  .flags = 0,		    /* OK for unprivileged users. */
1795 	  .policy = datapath_policy,
1796 	  .doit = ovs_dp_cmd_get,
1797 	  .dumpit = ovs_dp_cmd_dump
1798 	},
1799 	{ .cmd = OVS_DP_CMD_SET,
1800 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1801 	  .policy = datapath_policy,
1802 	  .doit = ovs_dp_cmd_set,
1803 	},
1804 };
1805 
1806 static struct genl_family dp_datapath_genl_family = {
1807 	.id = GENL_ID_GENERATE,
1808 	.hdrsize = sizeof(struct ovs_header),
1809 	.name = OVS_DATAPATH_FAMILY,
1810 	.version = OVS_DATAPATH_VERSION,
1811 	.maxattr = OVS_DP_ATTR_MAX,
1812 	.netnsok = true,
1813 	.parallel_ops = true,
1814 	.ops = dp_datapath_genl_ops,
1815 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1816 	.mcgrps = &ovs_dp_datapath_multicast_group,
1817 	.n_mcgrps = 1,
1818 };
1819 
1820 /* Called with ovs_mutex or RCU read lock. */
1821 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1822 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1823 {
1824 	struct ovs_header *ovs_header;
1825 	struct ovs_vport_stats vport_stats;
1826 	int err;
1827 
1828 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1829 				 flags, cmd);
1830 	if (!ovs_header)
1831 		return -EMSGSIZE;
1832 
1833 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1834 
1835 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1836 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1837 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1838 			   ovs_vport_name(vport)))
1839 		goto nla_put_failure;
1840 
1841 	ovs_vport_get_stats(vport, &vport_stats);
1842 	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1843 		    &vport_stats))
1844 		goto nla_put_failure;
1845 
1846 	if (ovs_vport_get_upcall_portids(vport, skb))
1847 		goto nla_put_failure;
1848 
1849 	err = ovs_vport_get_options(vport, skb);
1850 	if (err == -EMSGSIZE)
1851 		goto error;
1852 
1853 	genlmsg_end(skb, ovs_header);
1854 	return 0;
1855 
1856 nla_put_failure:
1857 	err = -EMSGSIZE;
1858 error:
1859 	genlmsg_cancel(skb, ovs_header);
1860 	return err;
1861 }
1862 
1863 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1864 {
1865 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1866 }
1867 
1868 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1869 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1870 					 u32 seq, u8 cmd)
1871 {
1872 	struct sk_buff *skb;
1873 	int retval;
1874 
1875 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1876 	if (!skb)
1877 		return ERR_PTR(-ENOMEM);
1878 
1879 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1880 	BUG_ON(retval < 0);
1881 
1882 	return skb;
1883 }
1884 
1885 /* Called with ovs_mutex or RCU read lock. */
1886 static struct vport *lookup_vport(struct net *net,
1887 				  const struct ovs_header *ovs_header,
1888 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1889 {
1890 	struct datapath *dp;
1891 	struct vport *vport;
1892 
1893 	if (a[OVS_VPORT_ATTR_NAME]) {
1894 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1895 		if (!vport)
1896 			return ERR_PTR(-ENODEV);
1897 		if (ovs_header->dp_ifindex &&
1898 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1899 			return ERR_PTR(-ENODEV);
1900 		return vport;
1901 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1902 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1903 
1904 		if (port_no >= DP_MAX_PORTS)
1905 			return ERR_PTR(-EFBIG);
1906 
1907 		dp = get_dp(net, ovs_header->dp_ifindex);
1908 		if (!dp)
1909 			return ERR_PTR(-ENODEV);
1910 
1911 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1912 		if (!vport)
1913 			return ERR_PTR(-ENODEV);
1914 		return vport;
1915 	} else
1916 		return ERR_PTR(-EINVAL);
1917 }
1918 
1919 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1920 {
1921 	struct nlattr **a = info->attrs;
1922 	struct ovs_header *ovs_header = info->userhdr;
1923 	struct vport_parms parms;
1924 	struct sk_buff *reply;
1925 	struct vport *vport;
1926 	struct datapath *dp;
1927 	u32 port_no;
1928 	int err;
1929 
1930 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1931 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1932 		return -EINVAL;
1933 
1934 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
1935 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1936 	if (port_no >= DP_MAX_PORTS)
1937 		return -EFBIG;
1938 
1939 	reply = ovs_vport_cmd_alloc_info();
1940 	if (!reply)
1941 		return -ENOMEM;
1942 
1943 	ovs_lock();
1944 restart:
1945 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1946 	err = -ENODEV;
1947 	if (!dp)
1948 		goto exit_unlock_free;
1949 
1950 	if (port_no) {
1951 		vport = ovs_vport_ovsl(dp, port_no);
1952 		err = -EBUSY;
1953 		if (vport)
1954 			goto exit_unlock_free;
1955 	} else {
1956 		for (port_no = 1; ; port_no++) {
1957 			if (port_no >= DP_MAX_PORTS) {
1958 				err = -EFBIG;
1959 				goto exit_unlock_free;
1960 			}
1961 			vport = ovs_vport_ovsl(dp, port_no);
1962 			if (!vport)
1963 				break;
1964 		}
1965 	}
1966 
1967 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1968 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1969 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1970 	parms.dp = dp;
1971 	parms.port_no = port_no;
1972 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1973 
1974 	vport = new_vport(&parms);
1975 	err = PTR_ERR(vport);
1976 	if (IS_ERR(vport)) {
1977 		if (err == -EAGAIN)
1978 			goto restart;
1979 		goto exit_unlock_free;
1980 	}
1981 
1982 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1983 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1984 	BUG_ON(err < 0);
1985 	ovs_unlock();
1986 
1987 	ovs_notify(&dp_vport_genl_family, reply, info);
1988 	return 0;
1989 
1990 exit_unlock_free:
1991 	ovs_unlock();
1992 	kfree_skb(reply);
1993 	return err;
1994 }
1995 
1996 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1997 {
1998 	struct nlattr **a = info->attrs;
1999 	struct sk_buff *reply;
2000 	struct vport *vport;
2001 	int err;
2002 
2003 	reply = ovs_vport_cmd_alloc_info();
2004 	if (!reply)
2005 		return -ENOMEM;
2006 
2007 	ovs_lock();
2008 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2009 	err = PTR_ERR(vport);
2010 	if (IS_ERR(vport))
2011 		goto exit_unlock_free;
2012 
2013 	if (a[OVS_VPORT_ATTR_TYPE] &&
2014 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2015 		err = -EINVAL;
2016 		goto exit_unlock_free;
2017 	}
2018 
2019 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2020 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2021 		if (err)
2022 			goto exit_unlock_free;
2023 	}
2024 
2025 
2026 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2027 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2028 
2029 		err = ovs_vport_set_upcall_portids(vport, ids);
2030 		if (err)
2031 			goto exit_unlock_free;
2032 	}
2033 
2034 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2035 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2036 	BUG_ON(err < 0);
2037 
2038 	ovs_unlock();
2039 	ovs_notify(&dp_vport_genl_family, reply, info);
2040 	return 0;
2041 
2042 exit_unlock_free:
2043 	ovs_unlock();
2044 	kfree_skb(reply);
2045 	return err;
2046 }
2047 
2048 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2049 {
2050 	struct nlattr **a = info->attrs;
2051 	struct sk_buff *reply;
2052 	struct vport *vport;
2053 	int err;
2054 
2055 	reply = ovs_vport_cmd_alloc_info();
2056 	if (!reply)
2057 		return -ENOMEM;
2058 
2059 	ovs_lock();
2060 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2061 	err = PTR_ERR(vport);
2062 	if (IS_ERR(vport))
2063 		goto exit_unlock_free;
2064 
2065 	if (vport->port_no == OVSP_LOCAL) {
2066 		err = -EINVAL;
2067 		goto exit_unlock_free;
2068 	}
2069 
2070 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2071 				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2072 	BUG_ON(err < 0);
2073 	ovs_dp_detach_port(vport);
2074 	ovs_unlock();
2075 
2076 	ovs_notify(&dp_vport_genl_family, reply, info);
2077 	return 0;
2078 
2079 exit_unlock_free:
2080 	ovs_unlock();
2081 	kfree_skb(reply);
2082 	return err;
2083 }
2084 
2085 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2086 {
2087 	struct nlattr **a = info->attrs;
2088 	struct ovs_header *ovs_header = info->userhdr;
2089 	struct sk_buff *reply;
2090 	struct vport *vport;
2091 	int err;
2092 
2093 	reply = ovs_vport_cmd_alloc_info();
2094 	if (!reply)
2095 		return -ENOMEM;
2096 
2097 	rcu_read_lock();
2098 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2099 	err = PTR_ERR(vport);
2100 	if (IS_ERR(vport))
2101 		goto exit_unlock_free;
2102 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2103 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2104 	BUG_ON(err < 0);
2105 	rcu_read_unlock();
2106 
2107 	return genlmsg_reply(reply, info);
2108 
2109 exit_unlock_free:
2110 	rcu_read_unlock();
2111 	kfree_skb(reply);
2112 	return err;
2113 }
2114 
2115 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2116 {
2117 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2118 	struct datapath *dp;
2119 	int bucket = cb->args[0], skip = cb->args[1];
2120 	int i, j = 0;
2121 
2122 	rcu_read_lock();
2123 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2124 	if (!dp) {
2125 		rcu_read_unlock();
2126 		return -ENODEV;
2127 	}
2128 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2129 		struct vport *vport;
2130 
2131 		j = 0;
2132 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2133 			if (j >= skip &&
2134 			    ovs_vport_cmd_fill_info(vport, skb,
2135 						    NETLINK_CB(cb->skb).portid,
2136 						    cb->nlh->nlmsg_seq,
2137 						    NLM_F_MULTI,
2138 						    OVS_VPORT_CMD_NEW) < 0)
2139 				goto out;
2140 
2141 			j++;
2142 		}
2143 		skip = 0;
2144 	}
2145 out:
2146 	rcu_read_unlock();
2147 
2148 	cb->args[0] = i;
2149 	cb->args[1] = j;
2150 
2151 	return skb->len;
2152 }
2153 
2154 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2155 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2156 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2157 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2158 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2159 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2160 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2161 };
2162 
2163 static const struct genl_ops dp_vport_genl_ops[] = {
2164 	{ .cmd = OVS_VPORT_CMD_NEW,
2165 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2166 	  .policy = vport_policy,
2167 	  .doit = ovs_vport_cmd_new
2168 	},
2169 	{ .cmd = OVS_VPORT_CMD_DEL,
2170 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2171 	  .policy = vport_policy,
2172 	  .doit = ovs_vport_cmd_del
2173 	},
2174 	{ .cmd = OVS_VPORT_CMD_GET,
2175 	  .flags = 0,		    /* OK for unprivileged users. */
2176 	  .policy = vport_policy,
2177 	  .doit = ovs_vport_cmd_get,
2178 	  .dumpit = ovs_vport_cmd_dump
2179 	},
2180 	{ .cmd = OVS_VPORT_CMD_SET,
2181 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2182 	  .policy = vport_policy,
2183 	  .doit = ovs_vport_cmd_set,
2184 	},
2185 };
2186 
2187 struct genl_family dp_vport_genl_family = {
2188 	.id = GENL_ID_GENERATE,
2189 	.hdrsize = sizeof(struct ovs_header),
2190 	.name = OVS_VPORT_FAMILY,
2191 	.version = OVS_VPORT_VERSION,
2192 	.maxattr = OVS_VPORT_ATTR_MAX,
2193 	.netnsok = true,
2194 	.parallel_ops = true,
2195 	.ops = dp_vport_genl_ops,
2196 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2197 	.mcgrps = &ovs_dp_vport_multicast_group,
2198 	.n_mcgrps = 1,
2199 };
2200 
2201 static struct genl_family * const dp_genl_families[] = {
2202 	&dp_datapath_genl_family,
2203 	&dp_vport_genl_family,
2204 	&dp_flow_genl_family,
2205 	&dp_packet_genl_family,
2206 };
2207 
2208 static void dp_unregister_genl(int n_families)
2209 {
2210 	int i;
2211 
2212 	for (i = 0; i < n_families; i++)
2213 		genl_unregister_family(dp_genl_families[i]);
2214 }
2215 
2216 static int dp_register_genl(void)
2217 {
2218 	int err;
2219 	int i;
2220 
2221 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2222 
2223 		err = genl_register_family(dp_genl_families[i]);
2224 		if (err)
2225 			goto error;
2226 	}
2227 
2228 	return 0;
2229 
2230 error:
2231 	dp_unregister_genl(i);
2232 	return err;
2233 }
2234 
2235 static int __net_init ovs_init_net(struct net *net)
2236 {
2237 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2238 
2239 	INIT_LIST_HEAD(&ovs_net->dps);
2240 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2241 	ovs_ct_init(net);
2242 	return 0;
2243 }
2244 
2245 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2246 					    struct list_head *head)
2247 {
2248 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2249 	struct datapath *dp;
2250 
2251 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2252 		int i;
2253 
2254 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2255 			struct vport *vport;
2256 
2257 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2258 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2259 					continue;
2260 
2261 				if (dev_net(vport->dev) == dnet)
2262 					list_add(&vport->detach_list, head);
2263 			}
2264 		}
2265 	}
2266 }
2267 
2268 static void __net_exit ovs_exit_net(struct net *dnet)
2269 {
2270 	struct datapath *dp, *dp_next;
2271 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2272 	struct vport *vport, *vport_next;
2273 	struct net *net;
2274 	LIST_HEAD(head);
2275 
2276 	ovs_ct_exit(dnet);
2277 	ovs_lock();
2278 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2279 		__dp_destroy(dp);
2280 
2281 	rtnl_lock();
2282 	for_each_net(net)
2283 		list_vports_from_net(net, dnet, &head);
2284 	rtnl_unlock();
2285 
2286 	/* Detach all vports from given namespace. */
2287 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2288 		list_del(&vport->detach_list);
2289 		ovs_dp_detach_port(vport);
2290 	}
2291 
2292 	ovs_unlock();
2293 
2294 	cancel_work_sync(&ovs_net->dp_notify_work);
2295 }
2296 
2297 static struct pernet_operations ovs_net_ops = {
2298 	.init = ovs_init_net,
2299 	.exit = ovs_exit_net,
2300 	.id   = &ovs_net_id,
2301 	.size = sizeof(struct ovs_net),
2302 };
2303 
2304 static int __init dp_init(void)
2305 {
2306 	int err;
2307 
2308 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2309 
2310 	pr_info("Open vSwitch switching datapath\n");
2311 
2312 	err = action_fifos_init();
2313 	if (err)
2314 		goto error;
2315 
2316 	err = ovs_internal_dev_rtnl_link_register();
2317 	if (err)
2318 		goto error_action_fifos_exit;
2319 
2320 	err = ovs_flow_init();
2321 	if (err)
2322 		goto error_unreg_rtnl_link;
2323 
2324 	err = ovs_vport_init();
2325 	if (err)
2326 		goto error_flow_exit;
2327 
2328 	err = register_pernet_device(&ovs_net_ops);
2329 	if (err)
2330 		goto error_vport_exit;
2331 
2332 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2333 	if (err)
2334 		goto error_netns_exit;
2335 
2336 	err = ovs_netdev_init();
2337 	if (err)
2338 		goto error_unreg_notifier;
2339 
2340 	err = dp_register_genl();
2341 	if (err < 0)
2342 		goto error_unreg_netdev;
2343 
2344 	return 0;
2345 
2346 error_unreg_netdev:
2347 	ovs_netdev_exit();
2348 error_unreg_notifier:
2349 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2350 error_netns_exit:
2351 	unregister_pernet_device(&ovs_net_ops);
2352 error_vport_exit:
2353 	ovs_vport_exit();
2354 error_flow_exit:
2355 	ovs_flow_exit();
2356 error_unreg_rtnl_link:
2357 	ovs_internal_dev_rtnl_link_unregister();
2358 error_action_fifos_exit:
2359 	action_fifos_exit();
2360 error:
2361 	return err;
2362 }
2363 
2364 static void dp_cleanup(void)
2365 {
2366 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2367 	ovs_netdev_exit();
2368 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2369 	unregister_pernet_device(&ovs_net_ops);
2370 	rcu_barrier();
2371 	ovs_vport_exit();
2372 	ovs_flow_exit();
2373 	ovs_internal_dev_rtnl_link_unregister();
2374 	action_fifos_exit();
2375 }
2376 
2377 module_init(dp_init);
2378 module_exit(dp_cleanup);
2379 
2380 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2381 MODULE_LICENSE("GPL");
2382