xref: /openbmc/linux/net/openvswitch/datapath.c (revision d7a3d85e)
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 	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 		upcall.cmd = OVS_PACKET_CMD_MISS;
276 		upcall.userdata = NULL;
277 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
278 		upcall.egress_tun_info = NULL;
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 	return size;
401 }
402 
403 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
404 				  const struct sw_flow_key *key,
405 				  const struct dp_upcall_info *upcall_info)
406 {
407 	struct ovs_header *upcall;
408 	struct sk_buff *nskb = NULL;
409 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
410 	struct nlattr *nla;
411 	struct genl_info info = {
412 		.dst_sk = ovs_dp_get_net(dp)->genl_sock,
413 		.snd_portid = upcall_info->portid,
414 	};
415 	size_t len;
416 	unsigned int hlen;
417 	int err, dp_ifindex;
418 
419 	dp_ifindex = get_dpifindex(dp);
420 	if (!dp_ifindex)
421 		return -ENODEV;
422 
423 	if (skb_vlan_tag_present(skb)) {
424 		nskb = skb_clone(skb, GFP_ATOMIC);
425 		if (!nskb)
426 			return -ENOMEM;
427 
428 		nskb = __vlan_hwaccel_push_inside(nskb);
429 		if (!nskb)
430 			return -ENOMEM;
431 
432 		skb = nskb;
433 	}
434 
435 	if (nla_attr_size(skb->len) > USHRT_MAX) {
436 		err = -EFBIG;
437 		goto out;
438 	}
439 
440 	/* Complete checksum if needed */
441 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
442 	    (err = skb_checksum_help(skb)))
443 		goto out;
444 
445 	/* Older versions of OVS user space enforce alignment of the last
446 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
447 	 * padding logic. Only perform zerocopy if padding is not required.
448 	 */
449 	if (dp->user_features & OVS_DP_F_UNALIGNED)
450 		hlen = skb_zerocopy_headlen(skb);
451 	else
452 		hlen = skb->len;
453 
454 	len = upcall_msg_size(upcall_info, hlen);
455 	user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
456 	if (!user_skb) {
457 		err = -ENOMEM;
458 		goto out;
459 	}
460 
461 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
462 			     0, upcall_info->cmd);
463 	upcall->dp_ifindex = dp_ifindex;
464 
465 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
466 	BUG_ON(err);
467 
468 	if (upcall_info->userdata)
469 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
470 			  nla_len(upcall_info->userdata),
471 			  nla_data(upcall_info->userdata));
472 
473 	if (upcall_info->egress_tun_info) {
474 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
475 		err = ovs_nla_put_egress_tunnel_key(user_skb,
476 						    upcall_info->egress_tun_info);
477 		BUG_ON(err);
478 		nla_nest_end(user_skb, nla);
479 	}
480 
481 	/* Only reserve room for attribute header, packet data is added
482 	 * in skb_zerocopy() */
483 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
484 		err = -ENOBUFS;
485 		goto out;
486 	}
487 	nla->nla_len = nla_attr_size(skb->len);
488 
489 	err = skb_zerocopy(user_skb, skb, skb->len, hlen);
490 	if (err)
491 		goto out;
492 
493 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
494 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
495 		size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
496 
497 		if (plen > 0)
498 			memset(skb_put(user_skb, plen), 0, plen);
499 	}
500 
501 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
502 
503 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
504 	user_skb = NULL;
505 out:
506 	if (err)
507 		skb_tx_error(skb);
508 	kfree_skb(user_skb);
509 	kfree_skb(nskb);
510 	return err;
511 }
512 
513 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
514 {
515 	struct ovs_header *ovs_header = info->userhdr;
516 	struct nlattr **a = info->attrs;
517 	struct sw_flow_actions *acts;
518 	struct sk_buff *packet;
519 	struct sw_flow *flow;
520 	struct sw_flow_actions *sf_acts;
521 	struct datapath *dp;
522 	struct ethhdr *eth;
523 	struct vport *input_vport;
524 	int len;
525 	int err;
526 	bool log = !a[OVS_PACKET_ATTR_PROBE];
527 
528 	err = -EINVAL;
529 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
530 	    !a[OVS_PACKET_ATTR_ACTIONS])
531 		goto err;
532 
533 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
534 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
535 	err = -ENOMEM;
536 	if (!packet)
537 		goto err;
538 	skb_reserve(packet, NET_IP_ALIGN);
539 
540 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
541 
542 	skb_reset_mac_header(packet);
543 	eth = eth_hdr(packet);
544 
545 	/* Normally, setting the skb 'protocol' field would be handled by a
546 	 * call to eth_type_trans(), but it assumes there's a sending
547 	 * device, which we may not have. */
548 	if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
549 		packet->protocol = eth->h_proto;
550 	else
551 		packet->protocol = htons(ETH_P_802_2);
552 
553 	/* Build an sw_flow for sending this packet. */
554 	flow = ovs_flow_alloc();
555 	err = PTR_ERR(flow);
556 	if (IS_ERR(flow))
557 		goto err_kfree_skb;
558 
559 	err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
560 					     &flow->key, log);
561 	if (err)
562 		goto err_flow_free;
563 
564 	err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
565 				   &flow->key, &acts, log);
566 	if (err)
567 		goto err_flow_free;
568 
569 	rcu_assign_pointer(flow->sf_acts, acts);
570 	OVS_CB(packet)->egress_tun_info = NULL;
571 	packet->priority = flow->key.phy.priority;
572 	packet->mark = flow->key.phy.skb_mark;
573 
574 	rcu_read_lock();
575 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
576 	err = -ENODEV;
577 	if (!dp)
578 		goto err_unlock;
579 
580 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
581 	if (!input_vport)
582 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
583 
584 	if (!input_vport)
585 		goto err_unlock;
586 
587 	OVS_CB(packet)->input_vport = input_vport;
588 	sf_acts = rcu_dereference(flow->sf_acts);
589 
590 	local_bh_disable();
591 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
592 	local_bh_enable();
593 	rcu_read_unlock();
594 
595 	ovs_flow_free(flow, false);
596 	return err;
597 
598 err_unlock:
599 	rcu_read_unlock();
600 err_flow_free:
601 	ovs_flow_free(flow, false);
602 err_kfree_skb:
603 	kfree_skb(packet);
604 err:
605 	return err;
606 }
607 
608 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
609 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
610 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
611 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
612 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
613 };
614 
615 static const struct genl_ops dp_packet_genl_ops[] = {
616 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
617 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
618 	  .policy = packet_policy,
619 	  .doit = ovs_packet_cmd_execute
620 	}
621 };
622 
623 static struct genl_family dp_packet_genl_family = {
624 	.id = GENL_ID_GENERATE,
625 	.hdrsize = sizeof(struct ovs_header),
626 	.name = OVS_PACKET_FAMILY,
627 	.version = OVS_PACKET_VERSION,
628 	.maxattr = OVS_PACKET_ATTR_MAX,
629 	.netnsok = true,
630 	.parallel_ops = true,
631 	.ops = dp_packet_genl_ops,
632 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
633 };
634 
635 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
636 			 struct ovs_dp_megaflow_stats *mega_stats)
637 {
638 	int i;
639 
640 	memset(mega_stats, 0, sizeof(*mega_stats));
641 
642 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
643 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
644 
645 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
646 
647 	for_each_possible_cpu(i) {
648 		const struct dp_stats_percpu *percpu_stats;
649 		struct dp_stats_percpu local_stats;
650 		unsigned int start;
651 
652 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
653 
654 		do {
655 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
656 			local_stats = *percpu_stats;
657 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
658 
659 		stats->n_hit += local_stats.n_hit;
660 		stats->n_missed += local_stats.n_missed;
661 		stats->n_lost += local_stats.n_lost;
662 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
663 	}
664 }
665 
666 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
667 {
668 	return ovs_identifier_is_ufid(sfid) &&
669 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
670 }
671 
672 static bool should_fill_mask(uint32_t ufid_flags)
673 {
674 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
675 }
676 
677 static bool should_fill_actions(uint32_t ufid_flags)
678 {
679 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
680 }
681 
682 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
683 				    const struct sw_flow_id *sfid,
684 				    uint32_t ufid_flags)
685 {
686 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
687 
688 	/* OVS_FLOW_ATTR_UFID */
689 	if (sfid && ovs_identifier_is_ufid(sfid))
690 		len += nla_total_size(sfid->ufid_len);
691 
692 	/* OVS_FLOW_ATTR_KEY */
693 	if (!sfid || should_fill_key(sfid, ufid_flags))
694 		len += nla_total_size(ovs_key_attr_size());
695 
696 	/* OVS_FLOW_ATTR_MASK */
697 	if (should_fill_mask(ufid_flags))
698 		len += nla_total_size(ovs_key_attr_size());
699 
700 	/* OVS_FLOW_ATTR_ACTIONS */
701 	if (should_fill_actions(ufid_flags))
702 		len += nla_total_size(acts->actions_len);
703 
704 	return len
705 		+ nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
706 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
707 		+ nla_total_size(8); /* OVS_FLOW_ATTR_USED */
708 }
709 
710 /* Called with ovs_mutex or RCU read lock. */
711 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
712 				   struct sk_buff *skb)
713 {
714 	struct ovs_flow_stats stats;
715 	__be16 tcp_flags;
716 	unsigned long used;
717 
718 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
719 
720 	if (used &&
721 	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
722 		return -EMSGSIZE;
723 
724 	if (stats.n_packets &&
725 	    nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
726 		return -EMSGSIZE;
727 
728 	if ((u8)ntohs(tcp_flags) &&
729 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
730 		return -EMSGSIZE;
731 
732 	return 0;
733 }
734 
735 /* Called with ovs_mutex or RCU read lock. */
736 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
737 				     struct sk_buff *skb, int skb_orig_len)
738 {
739 	struct nlattr *start;
740 	int err;
741 
742 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
743 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
744 	 * Netlink but individual action lists can be longer than
745 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
746 	 * The userspace caller can always fetch the actions separately if it
747 	 * really wants them.  (Most userspace callers in fact don't care.)
748 	 *
749 	 * This can only fail for dump operations because the skb is always
750 	 * properly sized for single flows.
751 	 */
752 	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
753 	if (start) {
754 		const struct sw_flow_actions *sf_acts;
755 
756 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
757 		err = ovs_nla_put_actions(sf_acts->actions,
758 					  sf_acts->actions_len, skb);
759 
760 		if (!err)
761 			nla_nest_end(skb, start);
762 		else {
763 			if (skb_orig_len)
764 				return err;
765 
766 			nla_nest_cancel(skb, start);
767 		}
768 	} else if (skb_orig_len) {
769 		return -EMSGSIZE;
770 	}
771 
772 	return 0;
773 }
774 
775 /* Called with ovs_mutex or RCU read lock. */
776 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
777 				  struct sk_buff *skb, u32 portid,
778 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
779 {
780 	const int skb_orig_len = skb->len;
781 	struct ovs_header *ovs_header;
782 	int err;
783 
784 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
785 				 flags, cmd);
786 	if (!ovs_header)
787 		return -EMSGSIZE;
788 
789 	ovs_header->dp_ifindex = dp_ifindex;
790 
791 	err = ovs_nla_put_identifier(flow, skb);
792 	if (err)
793 		goto error;
794 
795 	if (should_fill_key(&flow->id, ufid_flags)) {
796 		err = ovs_nla_put_masked_key(flow, skb);
797 		if (err)
798 			goto error;
799 	}
800 
801 	if (should_fill_mask(ufid_flags)) {
802 		err = ovs_nla_put_mask(flow, skb);
803 		if (err)
804 			goto error;
805 	}
806 
807 	err = ovs_flow_cmd_fill_stats(flow, skb);
808 	if (err)
809 		goto error;
810 
811 	if (should_fill_actions(ufid_flags)) {
812 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
813 		if (err)
814 			goto error;
815 	}
816 
817 	genlmsg_end(skb, ovs_header);
818 	return 0;
819 
820 error:
821 	genlmsg_cancel(skb, ovs_header);
822 	return err;
823 }
824 
825 /* May not be called with RCU read lock. */
826 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
827 					       const struct sw_flow_id *sfid,
828 					       struct genl_info *info,
829 					       bool always,
830 					       uint32_t ufid_flags)
831 {
832 	struct sk_buff *skb;
833 	size_t len;
834 
835 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
836 		return NULL;
837 
838 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
839 	skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
840 	if (!skb)
841 		return ERR_PTR(-ENOMEM);
842 
843 	return skb;
844 }
845 
846 /* Called with ovs_mutex. */
847 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
848 					       int dp_ifindex,
849 					       struct genl_info *info, u8 cmd,
850 					       bool always, u32 ufid_flags)
851 {
852 	struct sk_buff *skb;
853 	int retval;
854 
855 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
856 				      &flow->id, info, always, ufid_flags);
857 	if (IS_ERR_OR_NULL(skb))
858 		return skb;
859 
860 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
861 					info->snd_portid, info->snd_seq, 0,
862 					cmd, ufid_flags);
863 	BUG_ON(retval < 0);
864 	return skb;
865 }
866 
867 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
868 {
869 	struct nlattr **a = info->attrs;
870 	struct ovs_header *ovs_header = info->userhdr;
871 	struct sw_flow *flow = NULL, *new_flow;
872 	struct sw_flow_mask mask;
873 	struct sk_buff *reply;
874 	struct datapath *dp;
875 	struct sw_flow_key key;
876 	struct sw_flow_actions *acts;
877 	struct sw_flow_match match;
878 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
879 	int error;
880 	bool log = !a[OVS_FLOW_ATTR_PROBE];
881 
882 	/* Must have key and actions. */
883 	error = -EINVAL;
884 	if (!a[OVS_FLOW_ATTR_KEY]) {
885 		OVS_NLERR(log, "Flow key attr not present in new flow.");
886 		goto error;
887 	}
888 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
889 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
890 		goto error;
891 	}
892 
893 	/* Most of the time we need to allocate a new flow, do it before
894 	 * locking.
895 	 */
896 	new_flow = ovs_flow_alloc();
897 	if (IS_ERR(new_flow)) {
898 		error = PTR_ERR(new_flow);
899 		goto error;
900 	}
901 
902 	/* Extract key. */
903 	ovs_match_init(&match, &key, &mask);
904 	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
905 				  a[OVS_FLOW_ATTR_MASK], log);
906 	if (error)
907 		goto err_kfree_flow;
908 
909 	ovs_flow_mask_key(&new_flow->key, &key, &mask);
910 
911 	/* Extract flow identifier. */
912 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
913 				       &key, log);
914 	if (error)
915 		goto err_kfree_flow;
916 
917 	/* Validate actions. */
918 	error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
919 				     &acts, log);
920 	if (error) {
921 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
922 		goto err_kfree_flow;
923 	}
924 
925 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
926 					ufid_flags);
927 	if (IS_ERR(reply)) {
928 		error = PTR_ERR(reply);
929 		goto err_kfree_acts;
930 	}
931 
932 	ovs_lock();
933 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
934 	if (unlikely(!dp)) {
935 		error = -ENODEV;
936 		goto err_unlock_ovs;
937 	}
938 
939 	/* Check if this is a duplicate flow */
940 	if (ovs_identifier_is_ufid(&new_flow->id))
941 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
942 	if (!flow)
943 		flow = ovs_flow_tbl_lookup(&dp->table, &key);
944 	if (likely(!flow)) {
945 		rcu_assign_pointer(new_flow->sf_acts, acts);
946 
947 		/* Put flow in bucket. */
948 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
949 		if (unlikely(error)) {
950 			acts = NULL;
951 			goto err_unlock_ovs;
952 		}
953 
954 		if (unlikely(reply)) {
955 			error = ovs_flow_cmd_fill_info(new_flow,
956 						       ovs_header->dp_ifindex,
957 						       reply, info->snd_portid,
958 						       info->snd_seq, 0,
959 						       OVS_FLOW_CMD_NEW,
960 						       ufid_flags);
961 			BUG_ON(error < 0);
962 		}
963 		ovs_unlock();
964 	} else {
965 		struct sw_flow_actions *old_acts;
966 
967 		/* Bail out if we're not allowed to modify an existing flow.
968 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
969 		 * because Generic Netlink treats the latter as a dump
970 		 * request.  We also accept NLM_F_EXCL in case that bug ever
971 		 * gets fixed.
972 		 */
973 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
974 							 | NLM_F_EXCL))) {
975 			error = -EEXIST;
976 			goto err_unlock_ovs;
977 		}
978 		/* The flow identifier has to be the same for flow updates.
979 		 * Look for any overlapping flow.
980 		 */
981 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
982 			if (ovs_identifier_is_key(&flow->id))
983 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
984 								 &match);
985 			else /* UFID matches but key is different */
986 				flow = NULL;
987 			if (!flow) {
988 				error = -ENOENT;
989 				goto err_unlock_ovs;
990 			}
991 		}
992 		/* Update actions. */
993 		old_acts = ovsl_dereference(flow->sf_acts);
994 		rcu_assign_pointer(flow->sf_acts, acts);
995 
996 		if (unlikely(reply)) {
997 			error = ovs_flow_cmd_fill_info(flow,
998 						       ovs_header->dp_ifindex,
999 						       reply, info->snd_portid,
1000 						       info->snd_seq, 0,
1001 						       OVS_FLOW_CMD_NEW,
1002 						       ufid_flags);
1003 			BUG_ON(error < 0);
1004 		}
1005 		ovs_unlock();
1006 
1007 		ovs_nla_free_flow_actions(old_acts);
1008 		ovs_flow_free(new_flow, false);
1009 	}
1010 
1011 	if (reply)
1012 		ovs_notify(&dp_flow_genl_family, reply, info);
1013 	return 0;
1014 
1015 err_unlock_ovs:
1016 	ovs_unlock();
1017 	kfree_skb(reply);
1018 err_kfree_acts:
1019 	kfree(acts);
1020 err_kfree_flow:
1021 	ovs_flow_free(new_flow, false);
1022 error:
1023 	return error;
1024 }
1025 
1026 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1027 static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
1028 						const struct sw_flow_key *key,
1029 						const struct sw_flow_mask *mask,
1030 						bool log)
1031 {
1032 	struct sw_flow_actions *acts;
1033 	struct sw_flow_key masked_key;
1034 	int error;
1035 
1036 	ovs_flow_mask_key(&masked_key, key, mask);
1037 	error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1038 	if (error) {
1039 		OVS_NLERR(log,
1040 			  "Actions may not be safe on all matching packets");
1041 		return ERR_PTR(error);
1042 	}
1043 
1044 	return acts;
1045 }
1046 
1047 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1048 {
1049 	struct nlattr **a = info->attrs;
1050 	struct ovs_header *ovs_header = info->userhdr;
1051 	struct sw_flow_key key;
1052 	struct sw_flow *flow;
1053 	struct sw_flow_mask mask;
1054 	struct sk_buff *reply = NULL;
1055 	struct datapath *dp;
1056 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1057 	struct sw_flow_match match;
1058 	struct sw_flow_id sfid;
1059 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1060 	int error;
1061 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1062 	bool ufid_present;
1063 
1064 	/* Extract key. */
1065 	error = -EINVAL;
1066 	if (!a[OVS_FLOW_ATTR_KEY]) {
1067 		OVS_NLERR(log, "Flow key attribute not present in set flow.");
1068 		goto error;
1069 	}
1070 
1071 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1072 	ovs_match_init(&match, &key, &mask);
1073 	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1074 				  a[OVS_FLOW_ATTR_MASK], log);
1075 	if (error)
1076 		goto error;
1077 
1078 	/* Validate actions. */
1079 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1080 		acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1081 					log);
1082 		if (IS_ERR(acts)) {
1083 			error = PTR_ERR(acts);
1084 			goto error;
1085 		}
1086 
1087 		/* Can allocate before locking if have acts. */
1088 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1089 						ufid_flags);
1090 		if (IS_ERR(reply)) {
1091 			error = PTR_ERR(reply);
1092 			goto err_kfree_acts;
1093 		}
1094 	}
1095 
1096 	ovs_lock();
1097 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1098 	if (unlikely(!dp)) {
1099 		error = -ENODEV;
1100 		goto err_unlock_ovs;
1101 	}
1102 	/* Check that the flow exists. */
1103 	if (ufid_present)
1104 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1105 	else
1106 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1107 	if (unlikely(!flow)) {
1108 		error = -ENOENT;
1109 		goto err_unlock_ovs;
1110 	}
1111 
1112 	/* Update actions, if present. */
1113 	if (likely(acts)) {
1114 		old_acts = ovsl_dereference(flow->sf_acts);
1115 		rcu_assign_pointer(flow->sf_acts, acts);
1116 
1117 		if (unlikely(reply)) {
1118 			error = ovs_flow_cmd_fill_info(flow,
1119 						       ovs_header->dp_ifindex,
1120 						       reply, info->snd_portid,
1121 						       info->snd_seq, 0,
1122 						       OVS_FLOW_CMD_NEW,
1123 						       ufid_flags);
1124 			BUG_ON(error < 0);
1125 		}
1126 	} else {
1127 		/* Could not alloc without acts before locking. */
1128 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1129 						info, OVS_FLOW_CMD_NEW, false,
1130 						ufid_flags);
1131 
1132 		if (unlikely(IS_ERR(reply))) {
1133 			error = PTR_ERR(reply);
1134 			goto err_unlock_ovs;
1135 		}
1136 	}
1137 
1138 	/* Clear stats. */
1139 	if (a[OVS_FLOW_ATTR_CLEAR])
1140 		ovs_flow_stats_clear(flow);
1141 	ovs_unlock();
1142 
1143 	if (reply)
1144 		ovs_notify(&dp_flow_genl_family, reply, info);
1145 	if (old_acts)
1146 		ovs_nla_free_flow_actions(old_acts);
1147 
1148 	return 0;
1149 
1150 err_unlock_ovs:
1151 	ovs_unlock();
1152 	kfree_skb(reply);
1153 err_kfree_acts:
1154 	kfree(acts);
1155 error:
1156 	return error;
1157 }
1158 
1159 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1160 {
1161 	struct nlattr **a = info->attrs;
1162 	struct ovs_header *ovs_header = info->userhdr;
1163 	struct sw_flow_key key;
1164 	struct sk_buff *reply;
1165 	struct sw_flow *flow;
1166 	struct datapath *dp;
1167 	struct sw_flow_match match;
1168 	struct sw_flow_id ufid;
1169 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1170 	int err = 0;
1171 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1172 	bool ufid_present;
1173 
1174 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1175 	if (a[OVS_FLOW_ATTR_KEY]) {
1176 		ovs_match_init(&match, &key, NULL);
1177 		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1178 					log);
1179 	} else if (!ufid_present) {
1180 		OVS_NLERR(log,
1181 			  "Flow get message rejected, Key attribute missing.");
1182 		err = -EINVAL;
1183 	}
1184 	if (err)
1185 		return err;
1186 
1187 	ovs_lock();
1188 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1189 	if (!dp) {
1190 		err = -ENODEV;
1191 		goto unlock;
1192 	}
1193 
1194 	if (ufid_present)
1195 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1196 	else
1197 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1198 	if (!flow) {
1199 		err = -ENOENT;
1200 		goto unlock;
1201 	}
1202 
1203 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1204 					OVS_FLOW_CMD_NEW, true, ufid_flags);
1205 	if (IS_ERR(reply)) {
1206 		err = PTR_ERR(reply);
1207 		goto unlock;
1208 	}
1209 
1210 	ovs_unlock();
1211 	return genlmsg_reply(reply, info);
1212 unlock:
1213 	ovs_unlock();
1214 	return err;
1215 }
1216 
1217 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1218 {
1219 	struct nlattr **a = info->attrs;
1220 	struct ovs_header *ovs_header = info->userhdr;
1221 	struct sw_flow_key key;
1222 	struct sk_buff *reply;
1223 	struct sw_flow *flow = NULL;
1224 	struct datapath *dp;
1225 	struct sw_flow_match match;
1226 	struct sw_flow_id ufid;
1227 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1228 	int err;
1229 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1230 	bool ufid_present;
1231 
1232 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1233 	if (a[OVS_FLOW_ATTR_KEY]) {
1234 		ovs_match_init(&match, &key, NULL);
1235 		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1236 					log);
1237 		if (unlikely(err))
1238 			return err;
1239 	}
1240 
1241 	ovs_lock();
1242 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1243 	if (unlikely(!dp)) {
1244 		err = -ENODEV;
1245 		goto unlock;
1246 	}
1247 
1248 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1249 		err = ovs_flow_tbl_flush(&dp->table);
1250 		goto unlock;
1251 	}
1252 
1253 	if (ufid_present)
1254 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1255 	else
1256 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1257 	if (unlikely(!flow)) {
1258 		err = -ENOENT;
1259 		goto unlock;
1260 	}
1261 
1262 	ovs_flow_tbl_remove(&dp->table, flow);
1263 	ovs_unlock();
1264 
1265 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1266 					&flow->id, info, false, ufid_flags);
1267 	if (likely(reply)) {
1268 		if (likely(!IS_ERR(reply))) {
1269 			rcu_read_lock();	/*To keep RCU checker happy. */
1270 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1271 						     reply, info->snd_portid,
1272 						     info->snd_seq, 0,
1273 						     OVS_FLOW_CMD_DEL,
1274 						     ufid_flags);
1275 			rcu_read_unlock();
1276 			BUG_ON(err < 0);
1277 
1278 			ovs_notify(&dp_flow_genl_family, reply, info);
1279 		} else {
1280 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1281 		}
1282 	}
1283 
1284 	ovs_flow_free(flow, true);
1285 	return 0;
1286 unlock:
1287 	ovs_unlock();
1288 	return err;
1289 }
1290 
1291 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1292 {
1293 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1294 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1295 	struct table_instance *ti;
1296 	struct datapath *dp;
1297 	u32 ufid_flags;
1298 	int err;
1299 
1300 	err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1301 			    OVS_FLOW_ATTR_MAX, flow_policy);
1302 	if (err)
1303 		return err;
1304 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1305 
1306 	rcu_read_lock();
1307 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1308 	if (!dp) {
1309 		rcu_read_unlock();
1310 		return -ENODEV;
1311 	}
1312 
1313 	ti = rcu_dereference(dp->table.ti);
1314 	for (;;) {
1315 		struct sw_flow *flow;
1316 		u32 bucket, obj;
1317 
1318 		bucket = cb->args[0];
1319 		obj = cb->args[1];
1320 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1321 		if (!flow)
1322 			break;
1323 
1324 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1325 					   NETLINK_CB(cb->skb).portid,
1326 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1327 					   OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1328 			break;
1329 
1330 		cb->args[0] = bucket;
1331 		cb->args[1] = obj;
1332 	}
1333 	rcu_read_unlock();
1334 	return skb->len;
1335 }
1336 
1337 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1338 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1339 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1340 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1341 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1342 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1343 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1344 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1345 };
1346 
1347 static const struct genl_ops dp_flow_genl_ops[] = {
1348 	{ .cmd = OVS_FLOW_CMD_NEW,
1349 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1350 	  .policy = flow_policy,
1351 	  .doit = ovs_flow_cmd_new
1352 	},
1353 	{ .cmd = OVS_FLOW_CMD_DEL,
1354 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1355 	  .policy = flow_policy,
1356 	  .doit = ovs_flow_cmd_del
1357 	},
1358 	{ .cmd = OVS_FLOW_CMD_GET,
1359 	  .flags = 0,		    /* OK for unprivileged users. */
1360 	  .policy = flow_policy,
1361 	  .doit = ovs_flow_cmd_get,
1362 	  .dumpit = ovs_flow_cmd_dump
1363 	},
1364 	{ .cmd = OVS_FLOW_CMD_SET,
1365 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1366 	  .policy = flow_policy,
1367 	  .doit = ovs_flow_cmd_set,
1368 	},
1369 };
1370 
1371 static struct genl_family dp_flow_genl_family = {
1372 	.id = GENL_ID_GENERATE,
1373 	.hdrsize = sizeof(struct ovs_header),
1374 	.name = OVS_FLOW_FAMILY,
1375 	.version = OVS_FLOW_VERSION,
1376 	.maxattr = OVS_FLOW_ATTR_MAX,
1377 	.netnsok = true,
1378 	.parallel_ops = true,
1379 	.ops = dp_flow_genl_ops,
1380 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1381 	.mcgrps = &ovs_dp_flow_multicast_group,
1382 	.n_mcgrps = 1,
1383 };
1384 
1385 static size_t ovs_dp_cmd_msg_size(void)
1386 {
1387 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1388 
1389 	msgsize += nla_total_size(IFNAMSIZ);
1390 	msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1391 	msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1392 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1393 
1394 	return msgsize;
1395 }
1396 
1397 /* Called with ovs_mutex. */
1398 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1399 				u32 portid, u32 seq, u32 flags, u8 cmd)
1400 {
1401 	struct ovs_header *ovs_header;
1402 	struct ovs_dp_stats dp_stats;
1403 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1404 	int err;
1405 
1406 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1407 				   flags, cmd);
1408 	if (!ovs_header)
1409 		goto error;
1410 
1411 	ovs_header->dp_ifindex = get_dpifindex(dp);
1412 
1413 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1414 	if (err)
1415 		goto nla_put_failure;
1416 
1417 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1418 	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1419 			&dp_stats))
1420 		goto nla_put_failure;
1421 
1422 	if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1423 			sizeof(struct ovs_dp_megaflow_stats),
1424 			&dp_megaflow_stats))
1425 		goto nla_put_failure;
1426 
1427 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1428 		goto nla_put_failure;
1429 
1430 	genlmsg_end(skb, ovs_header);
1431 	return 0;
1432 
1433 nla_put_failure:
1434 	genlmsg_cancel(skb, ovs_header);
1435 error:
1436 	return -EMSGSIZE;
1437 }
1438 
1439 static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1440 {
1441 	return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1442 }
1443 
1444 /* Called with rcu_read_lock or ovs_mutex. */
1445 static struct datapath *lookup_datapath(struct net *net,
1446 					const struct ovs_header *ovs_header,
1447 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1448 {
1449 	struct datapath *dp;
1450 
1451 	if (!a[OVS_DP_ATTR_NAME])
1452 		dp = get_dp(net, ovs_header->dp_ifindex);
1453 	else {
1454 		struct vport *vport;
1455 
1456 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1457 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1458 	}
1459 	return dp ? dp : ERR_PTR(-ENODEV);
1460 }
1461 
1462 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1463 {
1464 	struct datapath *dp;
1465 
1466 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1467 	if (IS_ERR(dp))
1468 		return;
1469 
1470 	WARN(dp->user_features, "Dropping previously announced user features\n");
1471 	dp->user_features = 0;
1472 }
1473 
1474 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1475 {
1476 	if (a[OVS_DP_ATTR_USER_FEATURES])
1477 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1478 }
1479 
1480 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1481 {
1482 	struct nlattr **a = info->attrs;
1483 	struct vport_parms parms;
1484 	struct sk_buff *reply;
1485 	struct datapath *dp;
1486 	struct vport *vport;
1487 	struct ovs_net *ovs_net;
1488 	int err, i;
1489 
1490 	err = -EINVAL;
1491 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1492 		goto err;
1493 
1494 	reply = ovs_dp_cmd_alloc_info(info);
1495 	if (!reply)
1496 		return -ENOMEM;
1497 
1498 	err = -ENOMEM;
1499 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1500 	if (dp == NULL)
1501 		goto err_free_reply;
1502 
1503 	ovs_dp_set_net(dp, sock_net(skb->sk));
1504 
1505 	/* Allocate table. */
1506 	err = ovs_flow_tbl_init(&dp->table);
1507 	if (err)
1508 		goto err_free_dp;
1509 
1510 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1511 	if (!dp->stats_percpu) {
1512 		err = -ENOMEM;
1513 		goto err_destroy_table;
1514 	}
1515 
1516 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1517 			    GFP_KERNEL);
1518 	if (!dp->ports) {
1519 		err = -ENOMEM;
1520 		goto err_destroy_percpu;
1521 	}
1522 
1523 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1524 		INIT_HLIST_HEAD(&dp->ports[i]);
1525 
1526 	/* Set up our datapath device. */
1527 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1528 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1529 	parms.options = NULL;
1530 	parms.dp = dp;
1531 	parms.port_no = OVSP_LOCAL;
1532 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1533 
1534 	ovs_dp_change(dp, a);
1535 
1536 	/* So far only local changes have been made, now need the lock. */
1537 	ovs_lock();
1538 
1539 	vport = new_vport(&parms);
1540 	if (IS_ERR(vport)) {
1541 		err = PTR_ERR(vport);
1542 		if (err == -EBUSY)
1543 			err = -EEXIST;
1544 
1545 		if (err == -EEXIST) {
1546 			/* An outdated user space instance that does not understand
1547 			 * the concept of user_features has attempted to create a new
1548 			 * datapath and is likely to reuse it. Drop all user features.
1549 			 */
1550 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1551 				ovs_dp_reset_user_features(skb, info);
1552 		}
1553 
1554 		goto err_destroy_ports_array;
1555 	}
1556 
1557 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1558 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1559 	BUG_ON(err < 0);
1560 
1561 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1562 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1563 
1564 	ovs_unlock();
1565 
1566 	ovs_notify(&dp_datapath_genl_family, reply, info);
1567 	return 0;
1568 
1569 err_destroy_ports_array:
1570 	ovs_unlock();
1571 	kfree(dp->ports);
1572 err_destroy_percpu:
1573 	free_percpu(dp->stats_percpu);
1574 err_destroy_table:
1575 	ovs_flow_tbl_destroy(&dp->table);
1576 err_free_dp:
1577 	kfree(dp);
1578 err_free_reply:
1579 	kfree_skb(reply);
1580 err:
1581 	return err;
1582 }
1583 
1584 /* Called with ovs_mutex. */
1585 static void __dp_destroy(struct datapath *dp)
1586 {
1587 	int i;
1588 
1589 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1590 		struct vport *vport;
1591 		struct hlist_node *n;
1592 
1593 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1594 			if (vport->port_no != OVSP_LOCAL)
1595 				ovs_dp_detach_port(vport);
1596 	}
1597 
1598 	list_del_rcu(&dp->list_node);
1599 
1600 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1601 	 * all ports in datapath are destroyed first before freeing datapath.
1602 	 */
1603 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1604 
1605 	/* RCU destroy the flow table */
1606 	call_rcu(&dp->rcu, destroy_dp_rcu);
1607 }
1608 
1609 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1610 {
1611 	struct sk_buff *reply;
1612 	struct datapath *dp;
1613 	int err;
1614 
1615 	reply = ovs_dp_cmd_alloc_info(info);
1616 	if (!reply)
1617 		return -ENOMEM;
1618 
1619 	ovs_lock();
1620 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1621 	err = PTR_ERR(dp);
1622 	if (IS_ERR(dp))
1623 		goto err_unlock_free;
1624 
1625 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1626 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1627 	BUG_ON(err < 0);
1628 
1629 	__dp_destroy(dp);
1630 	ovs_unlock();
1631 
1632 	ovs_notify(&dp_datapath_genl_family, reply, info);
1633 
1634 	return 0;
1635 
1636 err_unlock_free:
1637 	ovs_unlock();
1638 	kfree_skb(reply);
1639 	return err;
1640 }
1641 
1642 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1643 {
1644 	struct sk_buff *reply;
1645 	struct datapath *dp;
1646 	int err;
1647 
1648 	reply = ovs_dp_cmd_alloc_info(info);
1649 	if (!reply)
1650 		return -ENOMEM;
1651 
1652 	ovs_lock();
1653 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1654 	err = PTR_ERR(dp);
1655 	if (IS_ERR(dp))
1656 		goto err_unlock_free;
1657 
1658 	ovs_dp_change(dp, info->attrs);
1659 
1660 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1661 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1662 	BUG_ON(err < 0);
1663 
1664 	ovs_unlock();
1665 	ovs_notify(&dp_datapath_genl_family, reply, info);
1666 
1667 	return 0;
1668 
1669 err_unlock_free:
1670 	ovs_unlock();
1671 	kfree_skb(reply);
1672 	return err;
1673 }
1674 
1675 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1676 {
1677 	struct sk_buff *reply;
1678 	struct datapath *dp;
1679 	int err;
1680 
1681 	reply = ovs_dp_cmd_alloc_info(info);
1682 	if (!reply)
1683 		return -ENOMEM;
1684 
1685 	ovs_lock();
1686 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1687 	if (IS_ERR(dp)) {
1688 		err = PTR_ERR(dp);
1689 		goto err_unlock_free;
1690 	}
1691 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1692 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1693 	BUG_ON(err < 0);
1694 	ovs_unlock();
1695 
1696 	return genlmsg_reply(reply, info);
1697 
1698 err_unlock_free:
1699 	ovs_unlock();
1700 	kfree_skb(reply);
1701 	return err;
1702 }
1703 
1704 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1705 {
1706 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1707 	struct datapath *dp;
1708 	int skip = cb->args[0];
1709 	int i = 0;
1710 
1711 	ovs_lock();
1712 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1713 		if (i >= skip &&
1714 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1715 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1716 					 OVS_DP_CMD_NEW) < 0)
1717 			break;
1718 		i++;
1719 	}
1720 	ovs_unlock();
1721 
1722 	cb->args[0] = i;
1723 
1724 	return skb->len;
1725 }
1726 
1727 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1728 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1729 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1730 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1731 };
1732 
1733 static const struct genl_ops dp_datapath_genl_ops[] = {
1734 	{ .cmd = OVS_DP_CMD_NEW,
1735 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1736 	  .policy = datapath_policy,
1737 	  .doit = ovs_dp_cmd_new
1738 	},
1739 	{ .cmd = OVS_DP_CMD_DEL,
1740 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1741 	  .policy = datapath_policy,
1742 	  .doit = ovs_dp_cmd_del
1743 	},
1744 	{ .cmd = OVS_DP_CMD_GET,
1745 	  .flags = 0,		    /* OK for unprivileged users. */
1746 	  .policy = datapath_policy,
1747 	  .doit = ovs_dp_cmd_get,
1748 	  .dumpit = ovs_dp_cmd_dump
1749 	},
1750 	{ .cmd = OVS_DP_CMD_SET,
1751 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1752 	  .policy = datapath_policy,
1753 	  .doit = ovs_dp_cmd_set,
1754 	},
1755 };
1756 
1757 static struct genl_family dp_datapath_genl_family = {
1758 	.id = GENL_ID_GENERATE,
1759 	.hdrsize = sizeof(struct ovs_header),
1760 	.name = OVS_DATAPATH_FAMILY,
1761 	.version = OVS_DATAPATH_VERSION,
1762 	.maxattr = OVS_DP_ATTR_MAX,
1763 	.netnsok = true,
1764 	.parallel_ops = true,
1765 	.ops = dp_datapath_genl_ops,
1766 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1767 	.mcgrps = &ovs_dp_datapath_multicast_group,
1768 	.n_mcgrps = 1,
1769 };
1770 
1771 /* Called with ovs_mutex or RCU read lock. */
1772 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1773 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1774 {
1775 	struct ovs_header *ovs_header;
1776 	struct ovs_vport_stats vport_stats;
1777 	int err;
1778 
1779 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1780 				 flags, cmd);
1781 	if (!ovs_header)
1782 		return -EMSGSIZE;
1783 
1784 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1785 
1786 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1787 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1788 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1789 			   vport->ops->get_name(vport)))
1790 		goto nla_put_failure;
1791 
1792 	ovs_vport_get_stats(vport, &vport_stats);
1793 	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1794 		    &vport_stats))
1795 		goto nla_put_failure;
1796 
1797 	if (ovs_vport_get_upcall_portids(vport, skb))
1798 		goto nla_put_failure;
1799 
1800 	err = ovs_vport_get_options(vport, skb);
1801 	if (err == -EMSGSIZE)
1802 		goto error;
1803 
1804 	genlmsg_end(skb, ovs_header);
1805 	return 0;
1806 
1807 nla_put_failure:
1808 	err = -EMSGSIZE;
1809 error:
1810 	genlmsg_cancel(skb, ovs_header);
1811 	return err;
1812 }
1813 
1814 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1815 {
1816 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1817 }
1818 
1819 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1820 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1821 					 u32 seq, u8 cmd)
1822 {
1823 	struct sk_buff *skb;
1824 	int retval;
1825 
1826 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1827 	if (!skb)
1828 		return ERR_PTR(-ENOMEM);
1829 
1830 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1831 	BUG_ON(retval < 0);
1832 
1833 	return skb;
1834 }
1835 
1836 /* Called with ovs_mutex or RCU read lock. */
1837 static struct vport *lookup_vport(struct net *net,
1838 				  const struct ovs_header *ovs_header,
1839 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1840 {
1841 	struct datapath *dp;
1842 	struct vport *vport;
1843 
1844 	if (a[OVS_VPORT_ATTR_NAME]) {
1845 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1846 		if (!vport)
1847 			return ERR_PTR(-ENODEV);
1848 		if (ovs_header->dp_ifindex &&
1849 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1850 			return ERR_PTR(-ENODEV);
1851 		return vport;
1852 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1853 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1854 
1855 		if (port_no >= DP_MAX_PORTS)
1856 			return ERR_PTR(-EFBIG);
1857 
1858 		dp = get_dp(net, ovs_header->dp_ifindex);
1859 		if (!dp)
1860 			return ERR_PTR(-ENODEV);
1861 
1862 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1863 		if (!vport)
1864 			return ERR_PTR(-ENODEV);
1865 		return vport;
1866 	} else
1867 		return ERR_PTR(-EINVAL);
1868 }
1869 
1870 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1871 {
1872 	struct nlattr **a = info->attrs;
1873 	struct ovs_header *ovs_header = info->userhdr;
1874 	struct vport_parms parms;
1875 	struct sk_buff *reply;
1876 	struct vport *vport;
1877 	struct datapath *dp;
1878 	u32 port_no;
1879 	int err;
1880 
1881 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1882 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1883 		return -EINVAL;
1884 
1885 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
1886 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1887 	if (port_no >= DP_MAX_PORTS)
1888 		return -EFBIG;
1889 
1890 	reply = ovs_vport_cmd_alloc_info();
1891 	if (!reply)
1892 		return -ENOMEM;
1893 
1894 	ovs_lock();
1895 restart:
1896 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1897 	err = -ENODEV;
1898 	if (!dp)
1899 		goto exit_unlock_free;
1900 
1901 	if (port_no) {
1902 		vport = ovs_vport_ovsl(dp, port_no);
1903 		err = -EBUSY;
1904 		if (vport)
1905 			goto exit_unlock_free;
1906 	} else {
1907 		for (port_no = 1; ; port_no++) {
1908 			if (port_no >= DP_MAX_PORTS) {
1909 				err = -EFBIG;
1910 				goto exit_unlock_free;
1911 			}
1912 			vport = ovs_vport_ovsl(dp, port_no);
1913 			if (!vport)
1914 				break;
1915 		}
1916 	}
1917 
1918 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1919 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1920 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1921 	parms.dp = dp;
1922 	parms.port_no = port_no;
1923 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1924 
1925 	vport = new_vport(&parms);
1926 	err = PTR_ERR(vport);
1927 	if (IS_ERR(vport)) {
1928 		if (err == -EAGAIN)
1929 			goto restart;
1930 		goto exit_unlock_free;
1931 	}
1932 
1933 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1934 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1935 	BUG_ON(err < 0);
1936 	ovs_unlock();
1937 
1938 	ovs_notify(&dp_vport_genl_family, reply, info);
1939 	return 0;
1940 
1941 exit_unlock_free:
1942 	ovs_unlock();
1943 	kfree_skb(reply);
1944 	return err;
1945 }
1946 
1947 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1948 {
1949 	struct nlattr **a = info->attrs;
1950 	struct sk_buff *reply;
1951 	struct vport *vport;
1952 	int err;
1953 
1954 	reply = ovs_vport_cmd_alloc_info();
1955 	if (!reply)
1956 		return -ENOMEM;
1957 
1958 	ovs_lock();
1959 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1960 	err = PTR_ERR(vport);
1961 	if (IS_ERR(vport))
1962 		goto exit_unlock_free;
1963 
1964 	if (a[OVS_VPORT_ATTR_TYPE] &&
1965 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1966 		err = -EINVAL;
1967 		goto exit_unlock_free;
1968 	}
1969 
1970 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
1971 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1972 		if (err)
1973 			goto exit_unlock_free;
1974 	}
1975 
1976 
1977 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1978 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1979 
1980 		err = ovs_vport_set_upcall_portids(vport, ids);
1981 		if (err)
1982 			goto exit_unlock_free;
1983 	}
1984 
1985 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1986 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1987 	BUG_ON(err < 0);
1988 
1989 	ovs_unlock();
1990 	ovs_notify(&dp_vport_genl_family, reply, info);
1991 	return 0;
1992 
1993 exit_unlock_free:
1994 	ovs_unlock();
1995 	kfree_skb(reply);
1996 	return err;
1997 }
1998 
1999 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2000 {
2001 	struct nlattr **a = info->attrs;
2002 	struct sk_buff *reply;
2003 	struct vport *vport;
2004 	int err;
2005 
2006 	reply = ovs_vport_cmd_alloc_info();
2007 	if (!reply)
2008 		return -ENOMEM;
2009 
2010 	ovs_lock();
2011 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2012 	err = PTR_ERR(vport);
2013 	if (IS_ERR(vport))
2014 		goto exit_unlock_free;
2015 
2016 	if (vport->port_no == OVSP_LOCAL) {
2017 		err = -EINVAL;
2018 		goto exit_unlock_free;
2019 	}
2020 
2021 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2022 				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2023 	BUG_ON(err < 0);
2024 	ovs_dp_detach_port(vport);
2025 	ovs_unlock();
2026 
2027 	ovs_notify(&dp_vport_genl_family, reply, info);
2028 	return 0;
2029 
2030 exit_unlock_free:
2031 	ovs_unlock();
2032 	kfree_skb(reply);
2033 	return err;
2034 }
2035 
2036 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2037 {
2038 	struct nlattr **a = info->attrs;
2039 	struct ovs_header *ovs_header = info->userhdr;
2040 	struct sk_buff *reply;
2041 	struct vport *vport;
2042 	int err;
2043 
2044 	reply = ovs_vport_cmd_alloc_info();
2045 	if (!reply)
2046 		return -ENOMEM;
2047 
2048 	rcu_read_lock();
2049 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2050 	err = PTR_ERR(vport);
2051 	if (IS_ERR(vport))
2052 		goto exit_unlock_free;
2053 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2054 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2055 	BUG_ON(err < 0);
2056 	rcu_read_unlock();
2057 
2058 	return genlmsg_reply(reply, info);
2059 
2060 exit_unlock_free:
2061 	rcu_read_unlock();
2062 	kfree_skb(reply);
2063 	return err;
2064 }
2065 
2066 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2067 {
2068 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2069 	struct datapath *dp;
2070 	int bucket = cb->args[0], skip = cb->args[1];
2071 	int i, j = 0;
2072 
2073 	rcu_read_lock();
2074 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2075 	if (!dp) {
2076 		rcu_read_unlock();
2077 		return -ENODEV;
2078 	}
2079 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2080 		struct vport *vport;
2081 
2082 		j = 0;
2083 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2084 			if (j >= skip &&
2085 			    ovs_vport_cmd_fill_info(vport, skb,
2086 						    NETLINK_CB(cb->skb).portid,
2087 						    cb->nlh->nlmsg_seq,
2088 						    NLM_F_MULTI,
2089 						    OVS_VPORT_CMD_NEW) < 0)
2090 				goto out;
2091 
2092 			j++;
2093 		}
2094 		skip = 0;
2095 	}
2096 out:
2097 	rcu_read_unlock();
2098 
2099 	cb->args[0] = i;
2100 	cb->args[1] = j;
2101 
2102 	return skb->len;
2103 }
2104 
2105 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2106 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2107 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2108 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2109 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2110 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2111 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2112 };
2113 
2114 static const struct genl_ops dp_vport_genl_ops[] = {
2115 	{ .cmd = OVS_VPORT_CMD_NEW,
2116 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2117 	  .policy = vport_policy,
2118 	  .doit = ovs_vport_cmd_new
2119 	},
2120 	{ .cmd = OVS_VPORT_CMD_DEL,
2121 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2122 	  .policy = vport_policy,
2123 	  .doit = ovs_vport_cmd_del
2124 	},
2125 	{ .cmd = OVS_VPORT_CMD_GET,
2126 	  .flags = 0,		    /* OK for unprivileged users. */
2127 	  .policy = vport_policy,
2128 	  .doit = ovs_vport_cmd_get,
2129 	  .dumpit = ovs_vport_cmd_dump
2130 	},
2131 	{ .cmd = OVS_VPORT_CMD_SET,
2132 	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2133 	  .policy = vport_policy,
2134 	  .doit = ovs_vport_cmd_set,
2135 	},
2136 };
2137 
2138 struct genl_family dp_vport_genl_family = {
2139 	.id = GENL_ID_GENERATE,
2140 	.hdrsize = sizeof(struct ovs_header),
2141 	.name = OVS_VPORT_FAMILY,
2142 	.version = OVS_VPORT_VERSION,
2143 	.maxattr = OVS_VPORT_ATTR_MAX,
2144 	.netnsok = true,
2145 	.parallel_ops = true,
2146 	.ops = dp_vport_genl_ops,
2147 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2148 	.mcgrps = &ovs_dp_vport_multicast_group,
2149 	.n_mcgrps = 1,
2150 };
2151 
2152 static struct genl_family * const dp_genl_families[] = {
2153 	&dp_datapath_genl_family,
2154 	&dp_vport_genl_family,
2155 	&dp_flow_genl_family,
2156 	&dp_packet_genl_family,
2157 };
2158 
2159 static void dp_unregister_genl(int n_families)
2160 {
2161 	int i;
2162 
2163 	for (i = 0; i < n_families; i++)
2164 		genl_unregister_family(dp_genl_families[i]);
2165 }
2166 
2167 static int dp_register_genl(void)
2168 {
2169 	int err;
2170 	int i;
2171 
2172 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2173 
2174 		err = genl_register_family(dp_genl_families[i]);
2175 		if (err)
2176 			goto error;
2177 	}
2178 
2179 	return 0;
2180 
2181 error:
2182 	dp_unregister_genl(i);
2183 	return err;
2184 }
2185 
2186 static int __net_init ovs_init_net(struct net *net)
2187 {
2188 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2189 
2190 	INIT_LIST_HEAD(&ovs_net->dps);
2191 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2192 	return 0;
2193 }
2194 
2195 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2196 					    struct list_head *head)
2197 {
2198 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2199 	struct datapath *dp;
2200 
2201 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2202 		int i;
2203 
2204 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2205 			struct vport *vport;
2206 
2207 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2208 				struct netdev_vport *netdev_vport;
2209 
2210 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2211 					continue;
2212 
2213 				netdev_vport = netdev_vport_priv(vport);
2214 				if (dev_net(netdev_vport->dev) == dnet)
2215 					list_add(&vport->detach_list, head);
2216 			}
2217 		}
2218 	}
2219 }
2220 
2221 static void __net_exit ovs_exit_net(struct net *dnet)
2222 {
2223 	struct datapath *dp, *dp_next;
2224 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2225 	struct vport *vport, *vport_next;
2226 	struct net *net;
2227 	LIST_HEAD(head);
2228 
2229 	ovs_lock();
2230 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2231 		__dp_destroy(dp);
2232 
2233 	rtnl_lock();
2234 	for_each_net(net)
2235 		list_vports_from_net(net, dnet, &head);
2236 	rtnl_unlock();
2237 
2238 	/* Detach all vports from given namespace. */
2239 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2240 		list_del(&vport->detach_list);
2241 		ovs_dp_detach_port(vport);
2242 	}
2243 
2244 	ovs_unlock();
2245 
2246 	cancel_work_sync(&ovs_net->dp_notify_work);
2247 }
2248 
2249 static struct pernet_operations ovs_net_ops = {
2250 	.init = ovs_init_net,
2251 	.exit = ovs_exit_net,
2252 	.id   = &ovs_net_id,
2253 	.size = sizeof(struct ovs_net),
2254 };
2255 
2256 static int __init dp_init(void)
2257 {
2258 	int err;
2259 
2260 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2261 
2262 	pr_info("Open vSwitch switching datapath\n");
2263 
2264 	err = action_fifos_init();
2265 	if (err)
2266 		goto error;
2267 
2268 	err = ovs_internal_dev_rtnl_link_register();
2269 	if (err)
2270 		goto error_action_fifos_exit;
2271 
2272 	err = ovs_flow_init();
2273 	if (err)
2274 		goto error_unreg_rtnl_link;
2275 
2276 	err = ovs_vport_init();
2277 	if (err)
2278 		goto error_flow_exit;
2279 
2280 	err = register_pernet_device(&ovs_net_ops);
2281 	if (err)
2282 		goto error_vport_exit;
2283 
2284 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2285 	if (err)
2286 		goto error_netns_exit;
2287 
2288 	err = ovs_netdev_init();
2289 	if (err)
2290 		goto error_unreg_notifier;
2291 
2292 	err = dp_register_genl();
2293 	if (err < 0)
2294 		goto error_unreg_netdev;
2295 
2296 	return 0;
2297 
2298 error_unreg_netdev:
2299 	ovs_netdev_exit();
2300 error_unreg_notifier:
2301 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2302 error_netns_exit:
2303 	unregister_pernet_device(&ovs_net_ops);
2304 error_vport_exit:
2305 	ovs_vport_exit();
2306 error_flow_exit:
2307 	ovs_flow_exit();
2308 error_unreg_rtnl_link:
2309 	ovs_internal_dev_rtnl_link_unregister();
2310 error_action_fifos_exit:
2311 	action_fifos_exit();
2312 error:
2313 	return err;
2314 }
2315 
2316 static void dp_cleanup(void)
2317 {
2318 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2319 	ovs_netdev_exit();
2320 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2321 	unregister_pernet_device(&ovs_net_ops);
2322 	rcu_barrier();
2323 	ovs_vport_exit();
2324 	ovs_flow_exit();
2325 	ovs_internal_dev_rtnl_link_unregister();
2326 	action_fifos_exit();
2327 }
2328 
2329 module_init(dp_init);
2330 module_exit(dp_cleanup);
2331 
2332 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2333 MODULE_LICENSE("GPL");
2334