xref: /linux/net/openvswitch/datapath.c (revision 071bf69a0220253a44acb8b2a27f7a262b9a46bf)
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, info, 0, GFP_KERNEL);
95 }
96 
97 /**
98  * DOC: Locking:
99  *
100  * All writes e.g. Writes to device state (add/remove datapath, port, set
101  * operations on vports, etc.), Writes to other state (flow table
102  * modifications, set miscellaneous datapath parameters, etc.) are protected
103  * by ovs_lock.
104  *
105  * Reads are protected by RCU.
106  *
107  * There are a few special cases (mostly stats) that have their own
108  * synchronization but they nest under all of above and don't interact with
109  * each other.
110  *
111  * The RTNL lock nests inside ovs_mutex.
112  */
113 
114 static DEFINE_MUTEX(ovs_mutex);
115 
116 void ovs_lock(void)
117 {
118 	mutex_lock(&ovs_mutex);
119 }
120 
121 void ovs_unlock(void)
122 {
123 	mutex_unlock(&ovs_mutex);
124 }
125 
126 #ifdef CONFIG_LOCKDEP
127 int lockdep_ovsl_is_held(void)
128 {
129 	if (debug_locks)
130 		return lockdep_is_held(&ovs_mutex);
131 	else
132 		return 1;
133 }
134 EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
135 #endif
136 
137 static struct vport *new_vport(const struct vport_parms *);
138 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
139 			     const struct sw_flow_key *,
140 			     const struct dp_upcall_info *,
141 			     uint32_t cutlen);
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 				  uint32_t cutlen);
146 
147 /* Must be called with rcu_read_lock. */
148 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
149 {
150 	struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
151 
152 	if (dev) {
153 		struct vport *vport = ovs_internal_dev_get_vport(dev);
154 		if (vport)
155 			return vport->dp;
156 	}
157 
158 	return NULL;
159 }
160 
161 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
162  * returned dp pointer valid.
163  */
164 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
165 {
166 	struct datapath *dp;
167 
168 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
169 	rcu_read_lock();
170 	dp = get_dp_rcu(net, dp_ifindex);
171 	rcu_read_unlock();
172 
173 	return dp;
174 }
175 
176 /* Must be called with rcu_read_lock or ovs_mutex. */
177 const char *ovs_dp_name(const struct datapath *dp)
178 {
179 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
180 	return ovs_vport_name(vport);
181 }
182 
183 static int get_dpifindex(const struct datapath *dp)
184 {
185 	struct vport *local;
186 	int ifindex;
187 
188 	rcu_read_lock();
189 
190 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
191 	if (local)
192 		ifindex = local->dev->ifindex;
193 	else
194 		ifindex = 0;
195 
196 	rcu_read_unlock();
197 
198 	return ifindex;
199 }
200 
201 static void destroy_dp_rcu(struct rcu_head *rcu)
202 {
203 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
204 
205 	ovs_flow_tbl_destroy(&dp->table);
206 	free_percpu(dp->stats_percpu);
207 	kfree(dp->ports);
208 	kfree(dp);
209 }
210 
211 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
212 					    u16 port_no)
213 {
214 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
215 }
216 
217 /* Called with ovs_mutex or RCU read lock. */
218 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
219 {
220 	struct vport *vport;
221 	struct hlist_head *head;
222 
223 	head = vport_hash_bucket(dp, port_no);
224 	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
225 		if (vport->port_no == port_no)
226 			return vport;
227 	}
228 	return NULL;
229 }
230 
231 /* Called with ovs_mutex. */
232 static struct vport *new_vport(const struct vport_parms *parms)
233 {
234 	struct vport *vport;
235 
236 	vport = ovs_vport_add(parms);
237 	if (!IS_ERR(vport)) {
238 		struct datapath *dp = parms->dp;
239 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
240 
241 		hlist_add_head_rcu(&vport->dp_hash_node, head);
242 	}
243 	return vport;
244 }
245 
246 void ovs_dp_detach_port(struct vport *p)
247 {
248 	ASSERT_OVSL();
249 
250 	/* First drop references to device. */
251 	hlist_del_rcu(&p->dp_hash_node);
252 
253 	/* Then destroy it. */
254 	ovs_vport_del(p);
255 }
256 
257 /* Must be called with rcu_read_lock. */
258 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
259 {
260 	const struct vport *p = OVS_CB(skb)->input_vport;
261 	struct datapath *dp = p->dp;
262 	struct sw_flow *flow;
263 	struct sw_flow_actions *sf_acts;
264 	struct dp_stats_percpu *stats;
265 	u64 *stats_counter;
266 	u32 n_mask_hit;
267 
268 	stats = this_cpu_ptr(dp->stats_percpu);
269 
270 	/* Look up flow. */
271 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
272 	if (unlikely(!flow)) {
273 		struct dp_upcall_info upcall;
274 		int error;
275 
276 		memset(&upcall, 0, sizeof(upcall));
277 		upcall.cmd = OVS_PACKET_CMD_MISS;
278 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
279 		upcall.mru = OVS_CB(skb)->mru;
280 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
281 		if (unlikely(error))
282 			kfree_skb(skb);
283 		else
284 			consume_skb(skb);
285 		stats_counter = &stats->n_missed;
286 		goto out;
287 	}
288 
289 	ovs_flow_stats_update(flow, key->tp.flags, skb);
290 	sf_acts = rcu_dereference(flow->sf_acts);
291 	ovs_execute_actions(dp, skb, sf_acts, key);
292 
293 	stats_counter = &stats->n_hit;
294 
295 out:
296 	/* Update datapath statistics. */
297 	u64_stats_update_begin(&stats->syncp);
298 	(*stats_counter)++;
299 	stats->n_mask_hit += n_mask_hit;
300 	u64_stats_update_end(&stats->syncp);
301 }
302 
303 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
304 		  const struct sw_flow_key *key,
305 		  const struct dp_upcall_info *upcall_info,
306 		  uint32_t cutlen)
307 {
308 	struct dp_stats_percpu *stats;
309 	int err;
310 
311 	if (upcall_info->portid == 0) {
312 		err = -ENOTCONN;
313 		goto err;
314 	}
315 
316 	if (!skb_is_gso(skb))
317 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
318 	else
319 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
320 	if (err)
321 		goto err;
322 
323 	return 0;
324 
325 err:
326 	stats = this_cpu_ptr(dp->stats_percpu);
327 
328 	u64_stats_update_begin(&stats->syncp);
329 	stats->n_lost++;
330 	u64_stats_update_end(&stats->syncp);
331 
332 	return err;
333 }
334 
335 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
336 			     const struct sw_flow_key *key,
337 			     const struct dp_upcall_info *upcall_info,
338 				 uint32_t cutlen)
339 {
340 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
341 	struct sw_flow_key later_key;
342 	struct sk_buff *segs, *nskb;
343 	int err;
344 
345 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
346 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
347 	if (IS_ERR(segs))
348 		return PTR_ERR(segs);
349 	if (segs == NULL)
350 		return -EINVAL;
351 
352 	if (gso_type & SKB_GSO_UDP) {
353 		/* The initial flow key extracted by ovs_flow_key_extract()
354 		 * in this case is for a first fragment, so we need to
355 		 * properly mark later fragments.
356 		 */
357 		later_key = *key;
358 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
359 	}
360 
361 	/* Queue all of the segments. */
362 	skb = segs;
363 	do {
364 		if (gso_type & SKB_GSO_UDP && skb != segs)
365 			key = &later_key;
366 
367 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
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 		+ nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
392 
393 	/* OVS_PACKET_ATTR_USERDATA */
394 	if (upcall_info->userdata)
395 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
396 
397 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
398 	if (upcall_info->egress_tun_info)
399 		size += nla_total_size(ovs_tun_key_attr_size());
400 
401 	/* OVS_PACKET_ATTR_ACTIONS */
402 	if (upcall_info->actions_len)
403 		size += nla_total_size(upcall_info->actions_len);
404 
405 	/* OVS_PACKET_ATTR_MRU */
406 	if (upcall_info->mru)
407 		size += nla_total_size(sizeof(upcall_info->mru));
408 
409 	return size;
410 }
411 
412 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
413 {
414 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
415 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
416 
417 		if (plen > 0)
418 			memset(skb_put(skb, plen), 0, plen);
419 	}
420 }
421 
422 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
423 				  const struct sw_flow_key *key,
424 				  const struct dp_upcall_info *upcall_info,
425 				  uint32_t cutlen)
426 {
427 	struct ovs_header *upcall;
428 	struct sk_buff *nskb = NULL;
429 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
430 	struct nlattr *nla;
431 	size_t len;
432 	unsigned int hlen;
433 	int err, dp_ifindex;
434 
435 	dp_ifindex = get_dpifindex(dp);
436 	if (!dp_ifindex)
437 		return -ENODEV;
438 
439 	if (skb_vlan_tag_present(skb)) {
440 		nskb = skb_clone(skb, GFP_ATOMIC);
441 		if (!nskb)
442 			return -ENOMEM;
443 
444 		nskb = __vlan_hwaccel_push_inside(nskb);
445 		if (!nskb)
446 			return -ENOMEM;
447 
448 		skb = nskb;
449 	}
450 
451 	if (nla_attr_size(skb->len) > USHRT_MAX) {
452 		err = -EFBIG;
453 		goto out;
454 	}
455 
456 	/* Complete checksum if needed */
457 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
458 	    (err = skb_checksum_help(skb)))
459 		goto out;
460 
461 	/* Older versions of OVS user space enforce alignment of the last
462 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
463 	 * padding logic. Only perform zerocopy if padding is not required.
464 	 */
465 	if (dp->user_features & OVS_DP_F_UNALIGNED)
466 		hlen = skb_zerocopy_headlen(skb);
467 	else
468 		hlen = skb->len;
469 
470 	len = upcall_msg_size(upcall_info, hlen - cutlen);
471 	user_skb = genlmsg_new(len, GFP_ATOMIC);
472 	if (!user_skb) {
473 		err = -ENOMEM;
474 		goto out;
475 	}
476 
477 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
478 			     0, upcall_info->cmd);
479 	upcall->dp_ifindex = dp_ifindex;
480 
481 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
482 	BUG_ON(err);
483 
484 	if (upcall_info->userdata)
485 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
486 			  nla_len(upcall_info->userdata),
487 			  nla_data(upcall_info->userdata));
488 
489 	if (upcall_info->egress_tun_info) {
490 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
491 		err = ovs_nla_put_tunnel_info(user_skb,
492 					      upcall_info->egress_tun_info);
493 		BUG_ON(err);
494 		nla_nest_end(user_skb, nla);
495 	}
496 
497 	if (upcall_info->actions_len) {
498 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
499 		err = ovs_nla_put_actions(upcall_info->actions,
500 					  upcall_info->actions_len,
501 					  user_skb);
502 		if (!err)
503 			nla_nest_end(user_skb, nla);
504 		else
505 			nla_nest_cancel(user_skb, nla);
506 	}
507 
508 	/* Add OVS_PACKET_ATTR_MRU */
509 	if (upcall_info->mru) {
510 		if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
511 				upcall_info->mru)) {
512 			err = -ENOBUFS;
513 			goto out;
514 		}
515 		pad_packet(dp, user_skb);
516 	}
517 
518 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
519 	if (cutlen > 0) {
520 		if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
521 				skb->len)) {
522 			err = -ENOBUFS;
523 			goto out;
524 		}
525 		pad_packet(dp, user_skb);
526 	}
527 
528 	/* Only reserve room for attribute header, packet data is added
529 	 * in skb_zerocopy() */
530 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
531 		err = -ENOBUFS;
532 		goto out;
533 	}
534 	nla->nla_len = nla_attr_size(skb->len - cutlen);
535 
536 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
537 	if (err)
538 		goto out;
539 
540 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
541 	pad_packet(dp, user_skb);
542 
543 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
544 
545 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
546 	user_skb = NULL;
547 out:
548 	if (err)
549 		skb_tx_error(skb);
550 	kfree_skb(user_skb);
551 	kfree_skb(nskb);
552 	return err;
553 }
554 
555 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
556 {
557 	struct ovs_header *ovs_header = info->userhdr;
558 	struct net *net = sock_net(skb->sk);
559 	struct nlattr **a = info->attrs;
560 	struct sw_flow_actions *acts;
561 	struct sk_buff *packet;
562 	struct sw_flow *flow;
563 	struct sw_flow_actions *sf_acts;
564 	struct datapath *dp;
565 	struct ethhdr *eth;
566 	struct vport *input_vport;
567 	u16 mru = 0;
568 	int len;
569 	int err;
570 	bool log = !a[OVS_PACKET_ATTR_PROBE];
571 
572 	err = -EINVAL;
573 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
574 	    !a[OVS_PACKET_ATTR_ACTIONS])
575 		goto err;
576 
577 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
578 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
579 	err = -ENOMEM;
580 	if (!packet)
581 		goto err;
582 	skb_reserve(packet, NET_IP_ALIGN);
583 
584 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
585 
586 	skb_reset_mac_header(packet);
587 	eth = eth_hdr(packet);
588 
589 	/* Normally, setting the skb 'protocol' field would be handled by a
590 	 * call to eth_type_trans(), but it assumes there's a sending
591 	 * device, which we may not have. */
592 	if (eth_proto_is_802_3(eth->h_proto))
593 		packet->protocol = eth->h_proto;
594 	else
595 		packet->protocol = htons(ETH_P_802_2);
596 
597 	/* Set packet's mru */
598 	if (a[OVS_PACKET_ATTR_MRU]) {
599 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
600 		packet->ignore_df = 1;
601 	}
602 	OVS_CB(packet)->mru = mru;
603 
604 	/* Build an sw_flow for sending this packet. */
605 	flow = ovs_flow_alloc();
606 	err = PTR_ERR(flow);
607 	if (IS_ERR(flow))
608 		goto err_kfree_skb;
609 
610 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
611 					     packet, &flow->key, log);
612 	if (err)
613 		goto err_flow_free;
614 
615 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
616 				   &flow->key, &acts, log);
617 	if (err)
618 		goto err_flow_free;
619 
620 	rcu_assign_pointer(flow->sf_acts, acts);
621 	packet->priority = flow->key.phy.priority;
622 	packet->mark = flow->key.phy.skb_mark;
623 
624 	rcu_read_lock();
625 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
626 	err = -ENODEV;
627 	if (!dp)
628 		goto err_unlock;
629 
630 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
631 	if (!input_vport)
632 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
633 
634 	if (!input_vport)
635 		goto err_unlock;
636 
637 	packet->dev = input_vport->dev;
638 	OVS_CB(packet)->input_vport = input_vport;
639 	sf_acts = rcu_dereference(flow->sf_acts);
640 
641 	local_bh_disable();
642 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
643 	local_bh_enable();
644 	rcu_read_unlock();
645 
646 	ovs_flow_free(flow, false);
647 	return err;
648 
649 err_unlock:
650 	rcu_read_unlock();
651 err_flow_free:
652 	ovs_flow_free(flow, false);
653 err_kfree_skb:
654 	kfree_skb(packet);
655 err:
656 	return err;
657 }
658 
659 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
660 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
661 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
662 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
663 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
664 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
665 };
666 
667 static const struct genl_ops dp_packet_genl_ops[] = {
668 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
669 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
670 	  .policy = packet_policy,
671 	  .doit = ovs_packet_cmd_execute
672 	}
673 };
674 
675 static struct genl_family dp_packet_genl_family = {
676 	.id = GENL_ID_GENERATE,
677 	.hdrsize = sizeof(struct ovs_header),
678 	.name = OVS_PACKET_FAMILY,
679 	.version = OVS_PACKET_VERSION,
680 	.maxattr = OVS_PACKET_ATTR_MAX,
681 	.netnsok = true,
682 	.parallel_ops = true,
683 	.ops = dp_packet_genl_ops,
684 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
685 };
686 
687 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
688 			 struct ovs_dp_megaflow_stats *mega_stats)
689 {
690 	int i;
691 
692 	memset(mega_stats, 0, sizeof(*mega_stats));
693 
694 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
695 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
696 
697 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
698 
699 	for_each_possible_cpu(i) {
700 		const struct dp_stats_percpu *percpu_stats;
701 		struct dp_stats_percpu local_stats;
702 		unsigned int start;
703 
704 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
705 
706 		do {
707 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
708 			local_stats = *percpu_stats;
709 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
710 
711 		stats->n_hit += local_stats.n_hit;
712 		stats->n_missed += local_stats.n_missed;
713 		stats->n_lost += local_stats.n_lost;
714 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
715 	}
716 }
717 
718 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
719 {
720 	return ovs_identifier_is_ufid(sfid) &&
721 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
722 }
723 
724 static bool should_fill_mask(uint32_t ufid_flags)
725 {
726 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
727 }
728 
729 static bool should_fill_actions(uint32_t ufid_flags)
730 {
731 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
732 }
733 
734 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
735 				    const struct sw_flow_id *sfid,
736 				    uint32_t ufid_flags)
737 {
738 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
739 
740 	/* OVS_FLOW_ATTR_UFID */
741 	if (sfid && ovs_identifier_is_ufid(sfid))
742 		len += nla_total_size(sfid->ufid_len);
743 
744 	/* OVS_FLOW_ATTR_KEY */
745 	if (!sfid || should_fill_key(sfid, ufid_flags))
746 		len += nla_total_size(ovs_key_attr_size());
747 
748 	/* OVS_FLOW_ATTR_MASK */
749 	if (should_fill_mask(ufid_flags))
750 		len += nla_total_size(ovs_key_attr_size());
751 
752 	/* OVS_FLOW_ATTR_ACTIONS */
753 	if (should_fill_actions(ufid_flags))
754 		len += nla_total_size(acts->orig_len);
755 
756 	return len
757 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
758 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
759 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
760 }
761 
762 /* Called with ovs_mutex or RCU read lock. */
763 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
764 				   struct sk_buff *skb)
765 {
766 	struct ovs_flow_stats stats;
767 	__be16 tcp_flags;
768 	unsigned long used;
769 
770 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
771 
772 	if (used &&
773 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
774 			      OVS_FLOW_ATTR_PAD))
775 		return -EMSGSIZE;
776 
777 	if (stats.n_packets &&
778 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
779 			  sizeof(struct ovs_flow_stats), &stats,
780 			  OVS_FLOW_ATTR_PAD))
781 		return -EMSGSIZE;
782 
783 	if ((u8)ntohs(tcp_flags) &&
784 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
785 		return -EMSGSIZE;
786 
787 	return 0;
788 }
789 
790 /* Called with ovs_mutex or RCU read lock. */
791 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
792 				     struct sk_buff *skb, int skb_orig_len)
793 {
794 	struct nlattr *start;
795 	int err;
796 
797 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
798 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
799 	 * Netlink but individual action lists can be longer than
800 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
801 	 * The userspace caller can always fetch the actions separately if it
802 	 * really wants them.  (Most userspace callers in fact don't care.)
803 	 *
804 	 * This can only fail for dump operations because the skb is always
805 	 * properly sized for single flows.
806 	 */
807 	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
808 	if (start) {
809 		const struct sw_flow_actions *sf_acts;
810 
811 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
812 		err = ovs_nla_put_actions(sf_acts->actions,
813 					  sf_acts->actions_len, skb);
814 
815 		if (!err)
816 			nla_nest_end(skb, start);
817 		else {
818 			if (skb_orig_len)
819 				return err;
820 
821 			nla_nest_cancel(skb, start);
822 		}
823 	} else if (skb_orig_len) {
824 		return -EMSGSIZE;
825 	}
826 
827 	return 0;
828 }
829 
830 /* Called with ovs_mutex or RCU read lock. */
831 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
832 				  struct sk_buff *skb, u32 portid,
833 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
834 {
835 	const int skb_orig_len = skb->len;
836 	struct ovs_header *ovs_header;
837 	int err;
838 
839 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
840 				 flags, cmd);
841 	if (!ovs_header)
842 		return -EMSGSIZE;
843 
844 	ovs_header->dp_ifindex = dp_ifindex;
845 
846 	err = ovs_nla_put_identifier(flow, skb);
847 	if (err)
848 		goto error;
849 
850 	if (should_fill_key(&flow->id, ufid_flags)) {
851 		err = ovs_nla_put_masked_key(flow, skb);
852 		if (err)
853 			goto error;
854 	}
855 
856 	if (should_fill_mask(ufid_flags)) {
857 		err = ovs_nla_put_mask(flow, skb);
858 		if (err)
859 			goto error;
860 	}
861 
862 	err = ovs_flow_cmd_fill_stats(flow, skb);
863 	if (err)
864 		goto error;
865 
866 	if (should_fill_actions(ufid_flags)) {
867 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
868 		if (err)
869 			goto error;
870 	}
871 
872 	genlmsg_end(skb, ovs_header);
873 	return 0;
874 
875 error:
876 	genlmsg_cancel(skb, ovs_header);
877 	return err;
878 }
879 
880 /* May not be called with RCU read lock. */
881 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
882 					       const struct sw_flow_id *sfid,
883 					       struct genl_info *info,
884 					       bool always,
885 					       uint32_t ufid_flags)
886 {
887 	struct sk_buff *skb;
888 	size_t len;
889 
890 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
891 		return NULL;
892 
893 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
894 	skb = genlmsg_new(len, GFP_KERNEL);
895 	if (!skb)
896 		return ERR_PTR(-ENOMEM);
897 
898 	return skb;
899 }
900 
901 /* Called with ovs_mutex. */
902 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
903 					       int dp_ifindex,
904 					       struct genl_info *info, u8 cmd,
905 					       bool always, u32 ufid_flags)
906 {
907 	struct sk_buff *skb;
908 	int retval;
909 
910 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
911 				      &flow->id, info, always, ufid_flags);
912 	if (IS_ERR_OR_NULL(skb))
913 		return skb;
914 
915 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
916 					info->snd_portid, info->snd_seq, 0,
917 					cmd, ufid_flags);
918 	BUG_ON(retval < 0);
919 	return skb;
920 }
921 
922 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
923 {
924 	struct net *net = sock_net(skb->sk);
925 	struct nlattr **a = info->attrs;
926 	struct ovs_header *ovs_header = info->userhdr;
927 	struct sw_flow *flow = NULL, *new_flow;
928 	struct sw_flow_mask mask;
929 	struct sk_buff *reply;
930 	struct datapath *dp;
931 	struct sw_flow_key key;
932 	struct sw_flow_actions *acts;
933 	struct sw_flow_match match;
934 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
935 	int error;
936 	bool log = !a[OVS_FLOW_ATTR_PROBE];
937 
938 	/* Must have key and actions. */
939 	error = -EINVAL;
940 	if (!a[OVS_FLOW_ATTR_KEY]) {
941 		OVS_NLERR(log, "Flow key attr not present in new flow.");
942 		goto error;
943 	}
944 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
945 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
946 		goto error;
947 	}
948 
949 	/* Most of the time we need to allocate a new flow, do it before
950 	 * locking.
951 	 */
952 	new_flow = ovs_flow_alloc();
953 	if (IS_ERR(new_flow)) {
954 		error = PTR_ERR(new_flow);
955 		goto error;
956 	}
957 
958 	/* Extract key. */
959 	ovs_match_init(&match, &key, &mask);
960 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
961 				  a[OVS_FLOW_ATTR_MASK], log);
962 	if (error)
963 		goto err_kfree_flow;
964 
965 	ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
966 
967 	/* Extract flow identifier. */
968 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
969 				       &key, log);
970 	if (error)
971 		goto err_kfree_flow;
972 
973 	/* Validate actions. */
974 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
975 				     &new_flow->key, &acts, log);
976 	if (error) {
977 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
978 		goto err_kfree_flow;
979 	}
980 
981 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
982 					ufid_flags);
983 	if (IS_ERR(reply)) {
984 		error = PTR_ERR(reply);
985 		goto err_kfree_acts;
986 	}
987 
988 	ovs_lock();
989 	dp = get_dp(net, ovs_header->dp_ifindex);
990 	if (unlikely(!dp)) {
991 		error = -ENODEV;
992 		goto err_unlock_ovs;
993 	}
994 
995 	/* Check if this is a duplicate flow */
996 	if (ovs_identifier_is_ufid(&new_flow->id))
997 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
998 	if (!flow)
999 		flow = ovs_flow_tbl_lookup(&dp->table, &key);
1000 	if (likely(!flow)) {
1001 		rcu_assign_pointer(new_flow->sf_acts, acts);
1002 
1003 		/* Put flow in bucket. */
1004 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1005 		if (unlikely(error)) {
1006 			acts = NULL;
1007 			goto err_unlock_ovs;
1008 		}
1009 
1010 		if (unlikely(reply)) {
1011 			error = ovs_flow_cmd_fill_info(new_flow,
1012 						       ovs_header->dp_ifindex,
1013 						       reply, info->snd_portid,
1014 						       info->snd_seq, 0,
1015 						       OVS_FLOW_CMD_NEW,
1016 						       ufid_flags);
1017 			BUG_ON(error < 0);
1018 		}
1019 		ovs_unlock();
1020 	} else {
1021 		struct sw_flow_actions *old_acts;
1022 
1023 		/* Bail out if we're not allowed to modify an existing flow.
1024 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1025 		 * because Generic Netlink treats the latter as a dump
1026 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1027 		 * gets fixed.
1028 		 */
1029 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1030 							 | NLM_F_EXCL))) {
1031 			error = -EEXIST;
1032 			goto err_unlock_ovs;
1033 		}
1034 		/* The flow identifier has to be the same for flow updates.
1035 		 * Look for any overlapping flow.
1036 		 */
1037 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1038 			if (ovs_identifier_is_key(&flow->id))
1039 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1040 								 &match);
1041 			else /* UFID matches but key is different */
1042 				flow = NULL;
1043 			if (!flow) {
1044 				error = -ENOENT;
1045 				goto err_unlock_ovs;
1046 			}
1047 		}
1048 		/* Update actions. */
1049 		old_acts = ovsl_dereference(flow->sf_acts);
1050 		rcu_assign_pointer(flow->sf_acts, acts);
1051 
1052 		if (unlikely(reply)) {
1053 			error = ovs_flow_cmd_fill_info(flow,
1054 						       ovs_header->dp_ifindex,
1055 						       reply, info->snd_portid,
1056 						       info->snd_seq, 0,
1057 						       OVS_FLOW_CMD_NEW,
1058 						       ufid_flags);
1059 			BUG_ON(error < 0);
1060 		}
1061 		ovs_unlock();
1062 
1063 		ovs_nla_free_flow_actions_rcu(old_acts);
1064 		ovs_flow_free(new_flow, false);
1065 	}
1066 
1067 	if (reply)
1068 		ovs_notify(&dp_flow_genl_family, reply, info);
1069 	return 0;
1070 
1071 err_unlock_ovs:
1072 	ovs_unlock();
1073 	kfree_skb(reply);
1074 err_kfree_acts:
1075 	ovs_nla_free_flow_actions(acts);
1076 err_kfree_flow:
1077 	ovs_flow_free(new_flow, false);
1078 error:
1079 	return error;
1080 }
1081 
1082 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1083 static struct sw_flow_actions *get_flow_actions(struct net *net,
1084 						const struct nlattr *a,
1085 						const struct sw_flow_key *key,
1086 						const struct sw_flow_mask *mask,
1087 						bool log)
1088 {
1089 	struct sw_flow_actions *acts;
1090 	struct sw_flow_key masked_key;
1091 	int error;
1092 
1093 	ovs_flow_mask_key(&masked_key, key, true, mask);
1094 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1095 	if (error) {
1096 		OVS_NLERR(log,
1097 			  "Actions may not be safe on all matching packets");
1098 		return ERR_PTR(error);
1099 	}
1100 
1101 	return acts;
1102 }
1103 
1104 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1105 {
1106 	struct net *net = sock_net(skb->sk);
1107 	struct nlattr **a = info->attrs;
1108 	struct ovs_header *ovs_header = info->userhdr;
1109 	struct sw_flow_key key;
1110 	struct sw_flow *flow;
1111 	struct sw_flow_mask mask;
1112 	struct sk_buff *reply = NULL;
1113 	struct datapath *dp;
1114 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1115 	struct sw_flow_match match;
1116 	struct sw_flow_id sfid;
1117 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1118 	int error = 0;
1119 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1120 	bool ufid_present;
1121 
1122 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1123 	if (a[OVS_FLOW_ATTR_KEY]) {
1124 		ovs_match_init(&match, &key, &mask);
1125 		error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1126 					  a[OVS_FLOW_ATTR_MASK], log);
1127 	} else if (!ufid_present) {
1128 		OVS_NLERR(log,
1129 			  "Flow set message rejected, Key attribute missing.");
1130 		error = -EINVAL;
1131 	}
1132 	if (error)
1133 		goto error;
1134 
1135 	/* Validate actions. */
1136 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1137 		if (!a[OVS_FLOW_ATTR_KEY]) {
1138 			OVS_NLERR(log,
1139 				  "Flow key attribute not present in set flow.");
1140 			error = -EINVAL;
1141 			goto error;
1142 		}
1143 
1144 		acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], &key,
1145 					&mask, log);
1146 		if (IS_ERR(acts)) {
1147 			error = PTR_ERR(acts);
1148 			goto error;
1149 		}
1150 
1151 		/* Can allocate before locking if have acts. */
1152 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1153 						ufid_flags);
1154 		if (IS_ERR(reply)) {
1155 			error = PTR_ERR(reply);
1156 			goto err_kfree_acts;
1157 		}
1158 	}
1159 
1160 	ovs_lock();
1161 	dp = get_dp(net, ovs_header->dp_ifindex);
1162 	if (unlikely(!dp)) {
1163 		error = -ENODEV;
1164 		goto err_unlock_ovs;
1165 	}
1166 	/* Check that the flow exists. */
1167 	if (ufid_present)
1168 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1169 	else
1170 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1171 	if (unlikely(!flow)) {
1172 		error = -ENOENT;
1173 		goto err_unlock_ovs;
1174 	}
1175 
1176 	/* Update actions, if present. */
1177 	if (likely(acts)) {
1178 		old_acts = ovsl_dereference(flow->sf_acts);
1179 		rcu_assign_pointer(flow->sf_acts, acts);
1180 
1181 		if (unlikely(reply)) {
1182 			error = ovs_flow_cmd_fill_info(flow,
1183 						       ovs_header->dp_ifindex,
1184 						       reply, info->snd_portid,
1185 						       info->snd_seq, 0,
1186 						       OVS_FLOW_CMD_NEW,
1187 						       ufid_flags);
1188 			BUG_ON(error < 0);
1189 		}
1190 	} else {
1191 		/* Could not alloc without acts before locking. */
1192 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1193 						info, OVS_FLOW_CMD_NEW, false,
1194 						ufid_flags);
1195 
1196 		if (IS_ERR(reply)) {
1197 			error = PTR_ERR(reply);
1198 			goto err_unlock_ovs;
1199 		}
1200 	}
1201 
1202 	/* Clear stats. */
1203 	if (a[OVS_FLOW_ATTR_CLEAR])
1204 		ovs_flow_stats_clear(flow);
1205 	ovs_unlock();
1206 
1207 	if (reply)
1208 		ovs_notify(&dp_flow_genl_family, reply, info);
1209 	if (old_acts)
1210 		ovs_nla_free_flow_actions_rcu(old_acts);
1211 
1212 	return 0;
1213 
1214 err_unlock_ovs:
1215 	ovs_unlock();
1216 	kfree_skb(reply);
1217 err_kfree_acts:
1218 	ovs_nla_free_flow_actions(acts);
1219 error:
1220 	return error;
1221 }
1222 
1223 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1224 {
1225 	struct nlattr **a = info->attrs;
1226 	struct ovs_header *ovs_header = info->userhdr;
1227 	struct net *net = sock_net(skb->sk);
1228 	struct sw_flow_key key;
1229 	struct sk_buff *reply;
1230 	struct sw_flow *flow;
1231 	struct datapath *dp;
1232 	struct sw_flow_match match;
1233 	struct sw_flow_id ufid;
1234 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1235 	int err = 0;
1236 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1237 	bool ufid_present;
1238 
1239 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1240 	if (a[OVS_FLOW_ATTR_KEY]) {
1241 		ovs_match_init(&match, &key, NULL);
1242 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1243 					log);
1244 	} else if (!ufid_present) {
1245 		OVS_NLERR(log,
1246 			  "Flow get message rejected, Key attribute missing.");
1247 		err = -EINVAL;
1248 	}
1249 	if (err)
1250 		return err;
1251 
1252 	ovs_lock();
1253 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1254 	if (!dp) {
1255 		err = -ENODEV;
1256 		goto unlock;
1257 	}
1258 
1259 	if (ufid_present)
1260 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1261 	else
1262 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1263 	if (!flow) {
1264 		err = -ENOENT;
1265 		goto unlock;
1266 	}
1267 
1268 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1269 					OVS_FLOW_CMD_NEW, true, ufid_flags);
1270 	if (IS_ERR(reply)) {
1271 		err = PTR_ERR(reply);
1272 		goto unlock;
1273 	}
1274 
1275 	ovs_unlock();
1276 	return genlmsg_reply(reply, info);
1277 unlock:
1278 	ovs_unlock();
1279 	return err;
1280 }
1281 
1282 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1283 {
1284 	struct nlattr **a = info->attrs;
1285 	struct ovs_header *ovs_header = info->userhdr;
1286 	struct net *net = sock_net(skb->sk);
1287 	struct sw_flow_key key;
1288 	struct sk_buff *reply;
1289 	struct sw_flow *flow = NULL;
1290 	struct datapath *dp;
1291 	struct sw_flow_match match;
1292 	struct sw_flow_id ufid;
1293 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1294 	int err;
1295 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1296 	bool ufid_present;
1297 
1298 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1299 	if (a[OVS_FLOW_ATTR_KEY]) {
1300 		ovs_match_init(&match, &key, NULL);
1301 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1302 					NULL, log);
1303 		if (unlikely(err))
1304 			return err;
1305 	}
1306 
1307 	ovs_lock();
1308 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1309 	if (unlikely(!dp)) {
1310 		err = -ENODEV;
1311 		goto unlock;
1312 	}
1313 
1314 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1315 		err = ovs_flow_tbl_flush(&dp->table);
1316 		goto unlock;
1317 	}
1318 
1319 	if (ufid_present)
1320 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1321 	else
1322 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1323 	if (unlikely(!flow)) {
1324 		err = -ENOENT;
1325 		goto unlock;
1326 	}
1327 
1328 	ovs_flow_tbl_remove(&dp->table, flow);
1329 	ovs_unlock();
1330 
1331 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1332 					&flow->id, info, false, ufid_flags);
1333 	if (likely(reply)) {
1334 		if (likely(!IS_ERR(reply))) {
1335 			rcu_read_lock();	/*To keep RCU checker happy. */
1336 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1337 						     reply, info->snd_portid,
1338 						     info->snd_seq, 0,
1339 						     OVS_FLOW_CMD_DEL,
1340 						     ufid_flags);
1341 			rcu_read_unlock();
1342 			BUG_ON(err < 0);
1343 
1344 			ovs_notify(&dp_flow_genl_family, reply, info);
1345 		} else {
1346 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1347 		}
1348 	}
1349 
1350 	ovs_flow_free(flow, true);
1351 	return 0;
1352 unlock:
1353 	ovs_unlock();
1354 	return err;
1355 }
1356 
1357 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1358 {
1359 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1360 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1361 	struct table_instance *ti;
1362 	struct datapath *dp;
1363 	u32 ufid_flags;
1364 	int err;
1365 
1366 	err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1367 			    OVS_FLOW_ATTR_MAX, flow_policy);
1368 	if (err)
1369 		return err;
1370 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1371 
1372 	rcu_read_lock();
1373 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1374 	if (!dp) {
1375 		rcu_read_unlock();
1376 		return -ENODEV;
1377 	}
1378 
1379 	ti = rcu_dereference(dp->table.ti);
1380 	for (;;) {
1381 		struct sw_flow *flow;
1382 		u32 bucket, obj;
1383 
1384 		bucket = cb->args[0];
1385 		obj = cb->args[1];
1386 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1387 		if (!flow)
1388 			break;
1389 
1390 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1391 					   NETLINK_CB(cb->skb).portid,
1392 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1393 					   OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1394 			break;
1395 
1396 		cb->args[0] = bucket;
1397 		cb->args[1] = obj;
1398 	}
1399 	rcu_read_unlock();
1400 	return skb->len;
1401 }
1402 
1403 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1404 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1405 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1406 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1407 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1408 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1409 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1410 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1411 };
1412 
1413 static const struct genl_ops dp_flow_genl_ops[] = {
1414 	{ .cmd = OVS_FLOW_CMD_NEW,
1415 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1416 	  .policy = flow_policy,
1417 	  .doit = ovs_flow_cmd_new
1418 	},
1419 	{ .cmd = OVS_FLOW_CMD_DEL,
1420 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1421 	  .policy = flow_policy,
1422 	  .doit = ovs_flow_cmd_del
1423 	},
1424 	{ .cmd = OVS_FLOW_CMD_GET,
1425 	  .flags = 0,		    /* OK for unprivileged users. */
1426 	  .policy = flow_policy,
1427 	  .doit = ovs_flow_cmd_get,
1428 	  .dumpit = ovs_flow_cmd_dump
1429 	},
1430 	{ .cmd = OVS_FLOW_CMD_SET,
1431 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1432 	  .policy = flow_policy,
1433 	  .doit = ovs_flow_cmd_set,
1434 	},
1435 };
1436 
1437 static struct genl_family dp_flow_genl_family = {
1438 	.id = GENL_ID_GENERATE,
1439 	.hdrsize = sizeof(struct ovs_header),
1440 	.name = OVS_FLOW_FAMILY,
1441 	.version = OVS_FLOW_VERSION,
1442 	.maxattr = OVS_FLOW_ATTR_MAX,
1443 	.netnsok = true,
1444 	.parallel_ops = true,
1445 	.ops = dp_flow_genl_ops,
1446 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1447 	.mcgrps = &ovs_dp_flow_multicast_group,
1448 	.n_mcgrps = 1,
1449 };
1450 
1451 static size_t ovs_dp_cmd_msg_size(void)
1452 {
1453 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1454 
1455 	msgsize += nla_total_size(IFNAMSIZ);
1456 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1457 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1458 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1459 
1460 	return msgsize;
1461 }
1462 
1463 /* Called with ovs_mutex. */
1464 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1465 				u32 portid, u32 seq, u32 flags, u8 cmd)
1466 {
1467 	struct ovs_header *ovs_header;
1468 	struct ovs_dp_stats dp_stats;
1469 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1470 	int err;
1471 
1472 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1473 				   flags, cmd);
1474 	if (!ovs_header)
1475 		goto error;
1476 
1477 	ovs_header->dp_ifindex = get_dpifindex(dp);
1478 
1479 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1480 	if (err)
1481 		goto nla_put_failure;
1482 
1483 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1484 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1485 			  &dp_stats, OVS_DP_ATTR_PAD))
1486 		goto nla_put_failure;
1487 
1488 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1489 			  sizeof(struct ovs_dp_megaflow_stats),
1490 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1491 		goto nla_put_failure;
1492 
1493 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1494 		goto nla_put_failure;
1495 
1496 	genlmsg_end(skb, ovs_header);
1497 	return 0;
1498 
1499 nla_put_failure:
1500 	genlmsg_cancel(skb, ovs_header);
1501 error:
1502 	return -EMSGSIZE;
1503 }
1504 
1505 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1506 {
1507 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1508 }
1509 
1510 /* Called with rcu_read_lock or ovs_mutex. */
1511 static struct datapath *lookup_datapath(struct net *net,
1512 					const struct ovs_header *ovs_header,
1513 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1514 {
1515 	struct datapath *dp;
1516 
1517 	if (!a[OVS_DP_ATTR_NAME])
1518 		dp = get_dp(net, ovs_header->dp_ifindex);
1519 	else {
1520 		struct vport *vport;
1521 
1522 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1523 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1524 	}
1525 	return dp ? dp : ERR_PTR(-ENODEV);
1526 }
1527 
1528 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1529 {
1530 	struct datapath *dp;
1531 
1532 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1533 	if (IS_ERR(dp))
1534 		return;
1535 
1536 	WARN(dp->user_features, "Dropping previously announced user features\n");
1537 	dp->user_features = 0;
1538 }
1539 
1540 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1541 {
1542 	if (a[OVS_DP_ATTR_USER_FEATURES])
1543 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1544 }
1545 
1546 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1547 {
1548 	struct nlattr **a = info->attrs;
1549 	struct vport_parms parms;
1550 	struct sk_buff *reply;
1551 	struct datapath *dp;
1552 	struct vport *vport;
1553 	struct ovs_net *ovs_net;
1554 	int err, i;
1555 
1556 	err = -EINVAL;
1557 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1558 		goto err;
1559 
1560 	reply = ovs_dp_cmd_alloc_info();
1561 	if (!reply)
1562 		return -ENOMEM;
1563 
1564 	err = -ENOMEM;
1565 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1566 	if (dp == NULL)
1567 		goto err_free_reply;
1568 
1569 	ovs_dp_set_net(dp, sock_net(skb->sk));
1570 
1571 	/* Allocate table. */
1572 	err = ovs_flow_tbl_init(&dp->table);
1573 	if (err)
1574 		goto err_free_dp;
1575 
1576 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1577 	if (!dp->stats_percpu) {
1578 		err = -ENOMEM;
1579 		goto err_destroy_table;
1580 	}
1581 
1582 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1583 			    GFP_KERNEL);
1584 	if (!dp->ports) {
1585 		err = -ENOMEM;
1586 		goto err_destroy_percpu;
1587 	}
1588 
1589 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1590 		INIT_HLIST_HEAD(&dp->ports[i]);
1591 
1592 	/* Set up our datapath device. */
1593 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1594 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1595 	parms.options = NULL;
1596 	parms.dp = dp;
1597 	parms.port_no = OVSP_LOCAL;
1598 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1599 
1600 	ovs_dp_change(dp, a);
1601 
1602 	/* So far only local changes have been made, now need the lock. */
1603 	ovs_lock();
1604 
1605 	vport = new_vport(&parms);
1606 	if (IS_ERR(vport)) {
1607 		err = PTR_ERR(vport);
1608 		if (err == -EBUSY)
1609 			err = -EEXIST;
1610 
1611 		if (err == -EEXIST) {
1612 			/* An outdated user space instance that does not understand
1613 			 * the concept of user_features has attempted to create a new
1614 			 * datapath and is likely to reuse it. Drop all user features.
1615 			 */
1616 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1617 				ovs_dp_reset_user_features(skb, info);
1618 		}
1619 
1620 		goto err_destroy_ports_array;
1621 	}
1622 
1623 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1624 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1625 	BUG_ON(err < 0);
1626 
1627 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1628 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1629 
1630 	ovs_unlock();
1631 
1632 	ovs_notify(&dp_datapath_genl_family, reply, info);
1633 	return 0;
1634 
1635 err_destroy_ports_array:
1636 	ovs_unlock();
1637 	kfree(dp->ports);
1638 err_destroy_percpu:
1639 	free_percpu(dp->stats_percpu);
1640 err_destroy_table:
1641 	ovs_flow_tbl_destroy(&dp->table);
1642 err_free_dp:
1643 	kfree(dp);
1644 err_free_reply:
1645 	kfree_skb(reply);
1646 err:
1647 	return err;
1648 }
1649 
1650 /* Called with ovs_mutex. */
1651 static void __dp_destroy(struct datapath *dp)
1652 {
1653 	int i;
1654 
1655 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1656 		struct vport *vport;
1657 		struct hlist_node *n;
1658 
1659 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1660 			if (vport->port_no != OVSP_LOCAL)
1661 				ovs_dp_detach_port(vport);
1662 	}
1663 
1664 	list_del_rcu(&dp->list_node);
1665 
1666 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1667 	 * all ports in datapath are destroyed first before freeing datapath.
1668 	 */
1669 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1670 
1671 	/* RCU destroy the flow table */
1672 	call_rcu(&dp->rcu, destroy_dp_rcu);
1673 }
1674 
1675 static int ovs_dp_cmd_del(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();
1682 	if (!reply)
1683 		return -ENOMEM;
1684 
1685 	ovs_lock();
1686 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1687 	err = PTR_ERR(dp);
1688 	if (IS_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_DEL);
1693 	BUG_ON(err < 0);
1694 
1695 	__dp_destroy(dp);
1696 	ovs_unlock();
1697 
1698 	ovs_notify(&dp_datapath_genl_family, reply, info);
1699 
1700 	return 0;
1701 
1702 err_unlock_free:
1703 	ovs_unlock();
1704 	kfree_skb(reply);
1705 	return err;
1706 }
1707 
1708 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1709 {
1710 	struct sk_buff *reply;
1711 	struct datapath *dp;
1712 	int err;
1713 
1714 	reply = ovs_dp_cmd_alloc_info();
1715 	if (!reply)
1716 		return -ENOMEM;
1717 
1718 	ovs_lock();
1719 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1720 	err = PTR_ERR(dp);
1721 	if (IS_ERR(dp))
1722 		goto err_unlock_free;
1723 
1724 	ovs_dp_change(dp, info->attrs);
1725 
1726 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1727 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1728 	BUG_ON(err < 0);
1729 
1730 	ovs_unlock();
1731 	ovs_notify(&dp_datapath_genl_family, reply, info);
1732 
1733 	return 0;
1734 
1735 err_unlock_free:
1736 	ovs_unlock();
1737 	kfree_skb(reply);
1738 	return err;
1739 }
1740 
1741 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1742 {
1743 	struct sk_buff *reply;
1744 	struct datapath *dp;
1745 	int err;
1746 
1747 	reply = ovs_dp_cmd_alloc_info();
1748 	if (!reply)
1749 		return -ENOMEM;
1750 
1751 	ovs_lock();
1752 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1753 	if (IS_ERR(dp)) {
1754 		err = PTR_ERR(dp);
1755 		goto err_unlock_free;
1756 	}
1757 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1758 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1759 	BUG_ON(err < 0);
1760 	ovs_unlock();
1761 
1762 	return genlmsg_reply(reply, info);
1763 
1764 err_unlock_free:
1765 	ovs_unlock();
1766 	kfree_skb(reply);
1767 	return err;
1768 }
1769 
1770 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1771 {
1772 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1773 	struct datapath *dp;
1774 	int skip = cb->args[0];
1775 	int i = 0;
1776 
1777 	ovs_lock();
1778 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1779 		if (i >= skip &&
1780 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1781 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1782 					 OVS_DP_CMD_NEW) < 0)
1783 			break;
1784 		i++;
1785 	}
1786 	ovs_unlock();
1787 
1788 	cb->args[0] = i;
1789 
1790 	return skb->len;
1791 }
1792 
1793 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1794 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1795 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1796 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1797 };
1798 
1799 static const struct genl_ops dp_datapath_genl_ops[] = {
1800 	{ .cmd = OVS_DP_CMD_NEW,
1801 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1802 	  .policy = datapath_policy,
1803 	  .doit = ovs_dp_cmd_new
1804 	},
1805 	{ .cmd = OVS_DP_CMD_DEL,
1806 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1807 	  .policy = datapath_policy,
1808 	  .doit = ovs_dp_cmd_del
1809 	},
1810 	{ .cmd = OVS_DP_CMD_GET,
1811 	  .flags = 0,		    /* OK for unprivileged users. */
1812 	  .policy = datapath_policy,
1813 	  .doit = ovs_dp_cmd_get,
1814 	  .dumpit = ovs_dp_cmd_dump
1815 	},
1816 	{ .cmd = OVS_DP_CMD_SET,
1817 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1818 	  .policy = datapath_policy,
1819 	  .doit = ovs_dp_cmd_set,
1820 	},
1821 };
1822 
1823 static struct genl_family dp_datapath_genl_family = {
1824 	.id = GENL_ID_GENERATE,
1825 	.hdrsize = sizeof(struct ovs_header),
1826 	.name = OVS_DATAPATH_FAMILY,
1827 	.version = OVS_DATAPATH_VERSION,
1828 	.maxattr = OVS_DP_ATTR_MAX,
1829 	.netnsok = true,
1830 	.parallel_ops = true,
1831 	.ops = dp_datapath_genl_ops,
1832 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1833 	.mcgrps = &ovs_dp_datapath_multicast_group,
1834 	.n_mcgrps = 1,
1835 };
1836 
1837 /* Called with ovs_mutex or RCU read lock. */
1838 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1839 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1840 {
1841 	struct ovs_header *ovs_header;
1842 	struct ovs_vport_stats vport_stats;
1843 	int err;
1844 
1845 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1846 				 flags, cmd);
1847 	if (!ovs_header)
1848 		return -EMSGSIZE;
1849 
1850 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1851 
1852 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1853 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1854 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1855 			   ovs_vport_name(vport)))
1856 		goto nla_put_failure;
1857 
1858 	ovs_vport_get_stats(vport, &vport_stats);
1859 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1860 			  sizeof(struct ovs_vport_stats), &vport_stats,
1861 			  OVS_VPORT_ATTR_PAD))
1862 		goto nla_put_failure;
1863 
1864 	if (ovs_vport_get_upcall_portids(vport, skb))
1865 		goto nla_put_failure;
1866 
1867 	err = ovs_vport_get_options(vport, skb);
1868 	if (err == -EMSGSIZE)
1869 		goto error;
1870 
1871 	genlmsg_end(skb, ovs_header);
1872 	return 0;
1873 
1874 nla_put_failure:
1875 	err = -EMSGSIZE;
1876 error:
1877 	genlmsg_cancel(skb, ovs_header);
1878 	return err;
1879 }
1880 
1881 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1882 {
1883 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1884 }
1885 
1886 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1887 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1888 					 u32 seq, u8 cmd)
1889 {
1890 	struct sk_buff *skb;
1891 	int retval;
1892 
1893 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1894 	if (!skb)
1895 		return ERR_PTR(-ENOMEM);
1896 
1897 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1898 	BUG_ON(retval < 0);
1899 
1900 	return skb;
1901 }
1902 
1903 /* Called with ovs_mutex or RCU read lock. */
1904 static struct vport *lookup_vport(struct net *net,
1905 				  const struct ovs_header *ovs_header,
1906 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1907 {
1908 	struct datapath *dp;
1909 	struct vport *vport;
1910 
1911 	if (a[OVS_VPORT_ATTR_NAME]) {
1912 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1913 		if (!vport)
1914 			return ERR_PTR(-ENODEV);
1915 		if (ovs_header->dp_ifindex &&
1916 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1917 			return ERR_PTR(-ENODEV);
1918 		return vport;
1919 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1920 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1921 
1922 		if (port_no >= DP_MAX_PORTS)
1923 			return ERR_PTR(-EFBIG);
1924 
1925 		dp = get_dp(net, ovs_header->dp_ifindex);
1926 		if (!dp)
1927 			return ERR_PTR(-ENODEV);
1928 
1929 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1930 		if (!vport)
1931 			return ERR_PTR(-ENODEV);
1932 		return vport;
1933 	} else
1934 		return ERR_PTR(-EINVAL);
1935 }
1936 
1937 /* Called with ovs_mutex */
1938 static void update_headroom(struct datapath *dp)
1939 {
1940 	unsigned dev_headroom, max_headroom = 0;
1941 	struct net_device *dev;
1942 	struct vport *vport;
1943 	int i;
1944 
1945 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1946 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1947 			dev = vport->dev;
1948 			dev_headroom = netdev_get_fwd_headroom(dev);
1949 			if (dev_headroom > max_headroom)
1950 				max_headroom = dev_headroom;
1951 		}
1952 	}
1953 
1954 	dp->max_headroom = max_headroom;
1955 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1956 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1957 			netdev_set_rx_headroom(vport->dev, max_headroom);
1958 }
1959 
1960 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1961 {
1962 	struct nlattr **a = info->attrs;
1963 	struct ovs_header *ovs_header = info->userhdr;
1964 	struct vport_parms parms;
1965 	struct sk_buff *reply;
1966 	struct vport *vport;
1967 	struct datapath *dp;
1968 	u32 port_no;
1969 	int err;
1970 
1971 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1972 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1973 		return -EINVAL;
1974 
1975 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
1976 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1977 	if (port_no >= DP_MAX_PORTS)
1978 		return -EFBIG;
1979 
1980 	reply = ovs_vport_cmd_alloc_info();
1981 	if (!reply)
1982 		return -ENOMEM;
1983 
1984 	ovs_lock();
1985 restart:
1986 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1987 	err = -ENODEV;
1988 	if (!dp)
1989 		goto exit_unlock_free;
1990 
1991 	if (port_no) {
1992 		vport = ovs_vport_ovsl(dp, port_no);
1993 		err = -EBUSY;
1994 		if (vport)
1995 			goto exit_unlock_free;
1996 	} else {
1997 		for (port_no = 1; ; port_no++) {
1998 			if (port_no >= DP_MAX_PORTS) {
1999 				err = -EFBIG;
2000 				goto exit_unlock_free;
2001 			}
2002 			vport = ovs_vport_ovsl(dp, port_no);
2003 			if (!vport)
2004 				break;
2005 		}
2006 	}
2007 
2008 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2009 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2010 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2011 	parms.dp = dp;
2012 	parms.port_no = port_no;
2013 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2014 
2015 	vport = new_vport(&parms);
2016 	err = PTR_ERR(vport);
2017 	if (IS_ERR(vport)) {
2018 		if (err == -EAGAIN)
2019 			goto restart;
2020 		goto exit_unlock_free;
2021 	}
2022 
2023 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2024 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2025 
2026 	if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2027 		update_headroom(dp);
2028 	else
2029 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2030 
2031 	BUG_ON(err < 0);
2032 	ovs_unlock();
2033 
2034 	ovs_notify(&dp_vport_genl_family, reply, info);
2035 	return 0;
2036 
2037 exit_unlock_free:
2038 	ovs_unlock();
2039 	kfree_skb(reply);
2040 	return err;
2041 }
2042 
2043 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2044 {
2045 	struct nlattr **a = info->attrs;
2046 	struct sk_buff *reply;
2047 	struct vport *vport;
2048 	int err;
2049 
2050 	reply = ovs_vport_cmd_alloc_info();
2051 	if (!reply)
2052 		return -ENOMEM;
2053 
2054 	ovs_lock();
2055 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2056 	err = PTR_ERR(vport);
2057 	if (IS_ERR(vport))
2058 		goto exit_unlock_free;
2059 
2060 	if (a[OVS_VPORT_ATTR_TYPE] &&
2061 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2062 		err = -EINVAL;
2063 		goto exit_unlock_free;
2064 	}
2065 
2066 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2067 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2068 		if (err)
2069 			goto exit_unlock_free;
2070 	}
2071 
2072 
2073 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2074 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2075 
2076 		err = ovs_vport_set_upcall_portids(vport, ids);
2077 		if (err)
2078 			goto exit_unlock_free;
2079 	}
2080 
2081 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2082 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2083 	BUG_ON(err < 0);
2084 
2085 	ovs_unlock();
2086 	ovs_notify(&dp_vport_genl_family, reply, info);
2087 	return 0;
2088 
2089 exit_unlock_free:
2090 	ovs_unlock();
2091 	kfree_skb(reply);
2092 	return err;
2093 }
2094 
2095 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2096 {
2097 	bool must_update_headroom = false;
2098 	struct nlattr **a = info->attrs;
2099 	struct sk_buff *reply;
2100 	struct datapath *dp;
2101 	struct vport *vport;
2102 	int err;
2103 
2104 	reply = ovs_vport_cmd_alloc_info();
2105 	if (!reply)
2106 		return -ENOMEM;
2107 
2108 	ovs_lock();
2109 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2110 	err = PTR_ERR(vport);
2111 	if (IS_ERR(vport))
2112 		goto exit_unlock_free;
2113 
2114 	if (vport->port_no == OVSP_LOCAL) {
2115 		err = -EINVAL;
2116 		goto exit_unlock_free;
2117 	}
2118 
2119 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2120 				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2121 	BUG_ON(err < 0);
2122 
2123 	/* the vport deletion may trigger dp headroom update */
2124 	dp = vport->dp;
2125 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2126 		must_update_headroom = true;
2127 	netdev_reset_rx_headroom(vport->dev);
2128 	ovs_dp_detach_port(vport);
2129 
2130 	if (must_update_headroom)
2131 		update_headroom(dp);
2132 	ovs_unlock();
2133 
2134 	ovs_notify(&dp_vport_genl_family, reply, info);
2135 	return 0;
2136 
2137 exit_unlock_free:
2138 	ovs_unlock();
2139 	kfree_skb(reply);
2140 	return err;
2141 }
2142 
2143 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2144 {
2145 	struct nlattr **a = info->attrs;
2146 	struct ovs_header *ovs_header = info->userhdr;
2147 	struct sk_buff *reply;
2148 	struct vport *vport;
2149 	int err;
2150 
2151 	reply = ovs_vport_cmd_alloc_info();
2152 	if (!reply)
2153 		return -ENOMEM;
2154 
2155 	rcu_read_lock();
2156 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2157 	err = PTR_ERR(vport);
2158 	if (IS_ERR(vport))
2159 		goto exit_unlock_free;
2160 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2161 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2162 	BUG_ON(err < 0);
2163 	rcu_read_unlock();
2164 
2165 	return genlmsg_reply(reply, info);
2166 
2167 exit_unlock_free:
2168 	rcu_read_unlock();
2169 	kfree_skb(reply);
2170 	return err;
2171 }
2172 
2173 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2174 {
2175 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2176 	struct datapath *dp;
2177 	int bucket = cb->args[0], skip = cb->args[1];
2178 	int i, j = 0;
2179 
2180 	rcu_read_lock();
2181 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2182 	if (!dp) {
2183 		rcu_read_unlock();
2184 		return -ENODEV;
2185 	}
2186 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2187 		struct vport *vport;
2188 
2189 		j = 0;
2190 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2191 			if (j >= skip &&
2192 			    ovs_vport_cmd_fill_info(vport, skb,
2193 						    NETLINK_CB(cb->skb).portid,
2194 						    cb->nlh->nlmsg_seq,
2195 						    NLM_F_MULTI,
2196 						    OVS_VPORT_CMD_NEW) < 0)
2197 				goto out;
2198 
2199 			j++;
2200 		}
2201 		skip = 0;
2202 	}
2203 out:
2204 	rcu_read_unlock();
2205 
2206 	cb->args[0] = i;
2207 	cb->args[1] = j;
2208 
2209 	return skb->len;
2210 }
2211 
2212 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2213 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2214 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2215 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2216 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2217 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2218 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2219 };
2220 
2221 static const struct genl_ops dp_vport_genl_ops[] = {
2222 	{ .cmd = OVS_VPORT_CMD_NEW,
2223 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2224 	  .policy = vport_policy,
2225 	  .doit = ovs_vport_cmd_new
2226 	},
2227 	{ .cmd = OVS_VPORT_CMD_DEL,
2228 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2229 	  .policy = vport_policy,
2230 	  .doit = ovs_vport_cmd_del
2231 	},
2232 	{ .cmd = OVS_VPORT_CMD_GET,
2233 	  .flags = 0,		    /* OK for unprivileged users. */
2234 	  .policy = vport_policy,
2235 	  .doit = ovs_vport_cmd_get,
2236 	  .dumpit = ovs_vport_cmd_dump
2237 	},
2238 	{ .cmd = OVS_VPORT_CMD_SET,
2239 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2240 	  .policy = vport_policy,
2241 	  .doit = ovs_vport_cmd_set,
2242 	},
2243 };
2244 
2245 struct genl_family dp_vport_genl_family = {
2246 	.id = GENL_ID_GENERATE,
2247 	.hdrsize = sizeof(struct ovs_header),
2248 	.name = OVS_VPORT_FAMILY,
2249 	.version = OVS_VPORT_VERSION,
2250 	.maxattr = OVS_VPORT_ATTR_MAX,
2251 	.netnsok = true,
2252 	.parallel_ops = true,
2253 	.ops = dp_vport_genl_ops,
2254 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2255 	.mcgrps = &ovs_dp_vport_multicast_group,
2256 	.n_mcgrps = 1,
2257 };
2258 
2259 static struct genl_family * const dp_genl_families[] = {
2260 	&dp_datapath_genl_family,
2261 	&dp_vport_genl_family,
2262 	&dp_flow_genl_family,
2263 	&dp_packet_genl_family,
2264 };
2265 
2266 static void dp_unregister_genl(int n_families)
2267 {
2268 	int i;
2269 
2270 	for (i = 0; i < n_families; i++)
2271 		genl_unregister_family(dp_genl_families[i]);
2272 }
2273 
2274 static int dp_register_genl(void)
2275 {
2276 	int err;
2277 	int i;
2278 
2279 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2280 
2281 		err = genl_register_family(dp_genl_families[i]);
2282 		if (err)
2283 			goto error;
2284 	}
2285 
2286 	return 0;
2287 
2288 error:
2289 	dp_unregister_genl(i);
2290 	return err;
2291 }
2292 
2293 static int __net_init ovs_init_net(struct net *net)
2294 {
2295 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2296 
2297 	INIT_LIST_HEAD(&ovs_net->dps);
2298 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2299 	ovs_ct_init(net);
2300 	return 0;
2301 }
2302 
2303 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2304 					    struct list_head *head)
2305 {
2306 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2307 	struct datapath *dp;
2308 
2309 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2310 		int i;
2311 
2312 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2313 			struct vport *vport;
2314 
2315 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2316 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2317 					continue;
2318 
2319 				if (dev_net(vport->dev) == dnet)
2320 					list_add(&vport->detach_list, head);
2321 			}
2322 		}
2323 	}
2324 }
2325 
2326 static void __net_exit ovs_exit_net(struct net *dnet)
2327 {
2328 	struct datapath *dp, *dp_next;
2329 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2330 	struct vport *vport, *vport_next;
2331 	struct net *net;
2332 	LIST_HEAD(head);
2333 
2334 	ovs_ct_exit(dnet);
2335 	ovs_lock();
2336 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2337 		__dp_destroy(dp);
2338 
2339 	rtnl_lock();
2340 	for_each_net(net)
2341 		list_vports_from_net(net, dnet, &head);
2342 	rtnl_unlock();
2343 
2344 	/* Detach all vports from given namespace. */
2345 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2346 		list_del(&vport->detach_list);
2347 		ovs_dp_detach_port(vport);
2348 	}
2349 
2350 	ovs_unlock();
2351 
2352 	cancel_work_sync(&ovs_net->dp_notify_work);
2353 }
2354 
2355 static struct pernet_operations ovs_net_ops = {
2356 	.init = ovs_init_net,
2357 	.exit = ovs_exit_net,
2358 	.id   = &ovs_net_id,
2359 	.size = sizeof(struct ovs_net),
2360 };
2361 
2362 static int __init dp_init(void)
2363 {
2364 	int err;
2365 
2366 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2367 
2368 	pr_info("Open vSwitch switching datapath\n");
2369 
2370 	err = action_fifos_init();
2371 	if (err)
2372 		goto error;
2373 
2374 	err = ovs_internal_dev_rtnl_link_register();
2375 	if (err)
2376 		goto error_action_fifos_exit;
2377 
2378 	err = ovs_flow_init();
2379 	if (err)
2380 		goto error_unreg_rtnl_link;
2381 
2382 	err = ovs_vport_init();
2383 	if (err)
2384 		goto error_flow_exit;
2385 
2386 	err = register_pernet_device(&ovs_net_ops);
2387 	if (err)
2388 		goto error_vport_exit;
2389 
2390 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2391 	if (err)
2392 		goto error_netns_exit;
2393 
2394 	err = ovs_netdev_init();
2395 	if (err)
2396 		goto error_unreg_notifier;
2397 
2398 	err = dp_register_genl();
2399 	if (err < 0)
2400 		goto error_unreg_netdev;
2401 
2402 	return 0;
2403 
2404 error_unreg_netdev:
2405 	ovs_netdev_exit();
2406 error_unreg_notifier:
2407 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2408 error_netns_exit:
2409 	unregister_pernet_device(&ovs_net_ops);
2410 error_vport_exit:
2411 	ovs_vport_exit();
2412 error_flow_exit:
2413 	ovs_flow_exit();
2414 error_unreg_rtnl_link:
2415 	ovs_internal_dev_rtnl_link_unregister();
2416 error_action_fifos_exit:
2417 	action_fifos_exit();
2418 error:
2419 	return err;
2420 }
2421 
2422 static void dp_cleanup(void)
2423 {
2424 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2425 	ovs_netdev_exit();
2426 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2427 	unregister_pernet_device(&ovs_net_ops);
2428 	rcu_barrier();
2429 	ovs_vport_exit();
2430 	ovs_flow_exit();
2431 	ovs_internal_dev_rtnl_link_unregister();
2432 	action_fifos_exit();
2433 }
2434 
2435 module_init(dp_init);
2436 module_exit(dp_cleanup);
2437 
2438 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2439 MODULE_LICENSE("GPL");
2440