xref: /openbmc/linux/net/openvswitch/datapath.c (revision 7a2eb736)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2007-2014 Nicira, Inc.
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include <linux/init.h>
9 #include <linux/module.h>
10 #include <linux/if_arp.h>
11 #include <linux/if_vlan.h>
12 #include <linux/in.h>
13 #include <linux/ip.h>
14 #include <linux/jhash.h>
15 #include <linux/delay.h>
16 #include <linux/time.h>
17 #include <linux/etherdevice.h>
18 #include <linux/genetlink.h>
19 #include <linux/kernel.h>
20 #include <linux/kthread.h>
21 #include <linux/mutex.h>
22 #include <linux/percpu.h>
23 #include <linux/rcupdate.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/ethtool.h>
27 #include <linux/wait.h>
28 #include <asm/div64.h>
29 #include <linux/highmem.h>
30 #include <linux/netfilter_bridge.h>
31 #include <linux/netfilter_ipv4.h>
32 #include <linux/inetdevice.h>
33 #include <linux/list.h>
34 #include <linux/openvswitch.h>
35 #include <linux/rculist.h>
36 #include <linux/dmi.h>
37 #include <net/genetlink.h>
38 #include <net/net_namespace.h>
39 #include <net/netns/generic.h>
40 
41 #include "datapath.h"
42 #include "flow.h"
43 #include "flow_table.h"
44 #include "flow_netlink.h"
45 #include "meter.h"
46 #include "vport-internal_dev.h"
47 #include "vport-netdev.h"
48 
49 unsigned int ovs_net_id __read_mostly;
50 
51 static struct genl_family dp_packet_genl_family;
52 static struct genl_family dp_flow_genl_family;
53 static struct genl_family dp_datapath_genl_family;
54 
55 static const struct nla_policy flow_policy[];
56 
57 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
58 	.name = OVS_FLOW_MCGROUP,
59 };
60 
61 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
62 	.name = OVS_DATAPATH_MCGROUP,
63 };
64 
65 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
66 	.name = OVS_VPORT_MCGROUP,
67 };
68 
69 /* Check if need to build a reply message.
70  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
71 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
72 			    unsigned int group)
73 {
74 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
75 	       genl_has_listeners(family, genl_info_net(info), group);
76 }
77 
78 static void ovs_notify(struct genl_family *family,
79 		       struct sk_buff *skb, struct genl_info *info)
80 {
81 	genl_notify(family, skb, info, 0, GFP_KERNEL);
82 }
83 
84 /**
85  * DOC: Locking:
86  *
87  * All writes e.g. Writes to device state (add/remove datapath, port, set
88  * operations on vports, etc.), Writes to other state (flow table
89  * modifications, set miscellaneous datapath parameters, etc.) are protected
90  * by ovs_lock.
91  *
92  * Reads are protected by RCU.
93  *
94  * There are a few special cases (mostly stats) that have their own
95  * synchronization but they nest under all of above and don't interact with
96  * each other.
97  *
98  * The RTNL lock nests inside ovs_mutex.
99  */
100 
101 static DEFINE_MUTEX(ovs_mutex);
102 
103 void ovs_lock(void)
104 {
105 	mutex_lock(&ovs_mutex);
106 }
107 
108 void ovs_unlock(void)
109 {
110 	mutex_unlock(&ovs_mutex);
111 }
112 
113 #ifdef CONFIG_LOCKDEP
114 int lockdep_ovsl_is_held(void)
115 {
116 	if (debug_locks)
117 		return lockdep_is_held(&ovs_mutex);
118 	else
119 		return 1;
120 }
121 #endif
122 
123 static struct vport *new_vport(const struct vport_parms *);
124 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
125 			     const struct sw_flow_key *,
126 			     const struct dp_upcall_info *,
127 			     uint32_t cutlen);
128 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
129 				  const struct sw_flow_key *,
130 				  const struct dp_upcall_info *,
131 				  uint32_t cutlen);
132 
133 /* Must be called with rcu_read_lock or ovs_mutex. */
134 const char *ovs_dp_name(const struct datapath *dp)
135 {
136 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
137 	return ovs_vport_name(vport);
138 }
139 
140 static int get_dpifindex(const struct datapath *dp)
141 {
142 	struct vport *local;
143 	int ifindex;
144 
145 	rcu_read_lock();
146 
147 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
148 	if (local)
149 		ifindex = local->dev->ifindex;
150 	else
151 		ifindex = 0;
152 
153 	rcu_read_unlock();
154 
155 	return ifindex;
156 }
157 
158 static void destroy_dp_rcu(struct rcu_head *rcu)
159 {
160 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
161 
162 	ovs_flow_tbl_destroy(&dp->table);
163 	free_percpu(dp->stats_percpu);
164 	kfree(dp->ports);
165 	ovs_meters_exit(dp);
166 	kfree(dp);
167 }
168 
169 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
170 					    u16 port_no)
171 {
172 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
173 }
174 
175 /* Called with ovs_mutex or RCU read lock. */
176 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
177 {
178 	struct vport *vport;
179 	struct hlist_head *head;
180 
181 	head = vport_hash_bucket(dp, port_no);
182 	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
183 		if (vport->port_no == port_no)
184 			return vport;
185 	}
186 	return NULL;
187 }
188 
189 /* Called with ovs_mutex. */
190 static struct vport *new_vport(const struct vport_parms *parms)
191 {
192 	struct vport *vport;
193 
194 	vport = ovs_vport_add(parms);
195 	if (!IS_ERR(vport)) {
196 		struct datapath *dp = parms->dp;
197 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
198 
199 		hlist_add_head_rcu(&vport->dp_hash_node, head);
200 	}
201 	return vport;
202 }
203 
204 void ovs_dp_detach_port(struct vport *p)
205 {
206 	ASSERT_OVSL();
207 
208 	/* First drop references to device. */
209 	hlist_del_rcu(&p->dp_hash_node);
210 
211 	/* Then destroy it. */
212 	ovs_vport_del(p);
213 }
214 
215 /* Must be called with rcu_read_lock. */
216 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
217 {
218 	const struct vport *p = OVS_CB(skb)->input_vport;
219 	struct datapath *dp = p->dp;
220 	struct sw_flow *flow;
221 	struct sw_flow_actions *sf_acts;
222 	struct dp_stats_percpu *stats;
223 	u64 *stats_counter;
224 	u32 n_mask_hit;
225 	int error;
226 
227 	stats = this_cpu_ptr(dp->stats_percpu);
228 
229 	/* Look up flow. */
230 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
231 	if (unlikely(!flow)) {
232 		struct dp_upcall_info upcall;
233 
234 		memset(&upcall, 0, sizeof(upcall));
235 		upcall.cmd = OVS_PACKET_CMD_MISS;
236 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
237 		upcall.mru = OVS_CB(skb)->mru;
238 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
239 		if (unlikely(error))
240 			kfree_skb(skb);
241 		else
242 			consume_skb(skb);
243 		stats_counter = &stats->n_missed;
244 		goto out;
245 	}
246 
247 	ovs_flow_stats_update(flow, key->tp.flags, skb);
248 	sf_acts = rcu_dereference(flow->sf_acts);
249 	error = ovs_execute_actions(dp, skb, sf_acts, key);
250 	if (unlikely(error))
251 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
252 							ovs_dp_name(dp), error);
253 
254 	stats_counter = &stats->n_hit;
255 
256 out:
257 	/* Update datapath statistics. */
258 	u64_stats_update_begin(&stats->syncp);
259 	(*stats_counter)++;
260 	stats->n_mask_hit += n_mask_hit;
261 	u64_stats_update_end(&stats->syncp);
262 }
263 
264 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
265 		  const struct sw_flow_key *key,
266 		  const struct dp_upcall_info *upcall_info,
267 		  uint32_t cutlen)
268 {
269 	struct dp_stats_percpu *stats;
270 	int err;
271 
272 	if (upcall_info->portid == 0) {
273 		err = -ENOTCONN;
274 		goto err;
275 	}
276 
277 	if (!skb_is_gso(skb))
278 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
279 	else
280 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
281 	if (err)
282 		goto err;
283 
284 	return 0;
285 
286 err:
287 	stats = this_cpu_ptr(dp->stats_percpu);
288 
289 	u64_stats_update_begin(&stats->syncp);
290 	stats->n_lost++;
291 	u64_stats_update_end(&stats->syncp);
292 
293 	return err;
294 }
295 
296 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
297 			     const struct sw_flow_key *key,
298 			     const struct dp_upcall_info *upcall_info,
299 				 uint32_t cutlen)
300 {
301 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
302 	struct sw_flow_key later_key;
303 	struct sk_buff *segs, *nskb;
304 	int err;
305 
306 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
307 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
308 	if (IS_ERR(segs))
309 		return PTR_ERR(segs);
310 	if (segs == NULL)
311 		return -EINVAL;
312 
313 	if (gso_type & SKB_GSO_UDP) {
314 		/* The initial flow key extracted by ovs_flow_key_extract()
315 		 * in this case is for a first fragment, so we need to
316 		 * properly mark later fragments.
317 		 */
318 		later_key = *key;
319 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
320 	}
321 
322 	/* Queue all of the segments. */
323 	skb = segs;
324 	do {
325 		if (gso_type & SKB_GSO_UDP && skb != segs)
326 			key = &later_key;
327 
328 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
329 		if (err)
330 			break;
331 
332 	} while ((skb = skb->next));
333 
334 	/* Free all of the segments. */
335 	skb = segs;
336 	do {
337 		nskb = skb->next;
338 		if (err)
339 			kfree_skb(skb);
340 		else
341 			consume_skb(skb);
342 	} while ((skb = nskb));
343 	return err;
344 }
345 
346 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
347 			      unsigned int hdrlen, int actions_attrlen)
348 {
349 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
350 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
351 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
352 		+ nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
353 
354 	/* OVS_PACKET_ATTR_USERDATA */
355 	if (upcall_info->userdata)
356 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
357 
358 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
359 	if (upcall_info->egress_tun_info)
360 		size += nla_total_size(ovs_tun_key_attr_size());
361 
362 	/* OVS_PACKET_ATTR_ACTIONS */
363 	if (upcall_info->actions_len)
364 		size += nla_total_size(actions_attrlen);
365 
366 	/* OVS_PACKET_ATTR_MRU */
367 	if (upcall_info->mru)
368 		size += nla_total_size(sizeof(upcall_info->mru));
369 
370 	return size;
371 }
372 
373 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
374 {
375 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
376 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
377 
378 		if (plen > 0)
379 			skb_put_zero(skb, plen);
380 	}
381 }
382 
383 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
384 				  const struct sw_flow_key *key,
385 				  const struct dp_upcall_info *upcall_info,
386 				  uint32_t cutlen)
387 {
388 	struct ovs_header *upcall;
389 	struct sk_buff *nskb = NULL;
390 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
391 	struct nlattr *nla;
392 	size_t len;
393 	unsigned int hlen;
394 	int err, dp_ifindex;
395 
396 	dp_ifindex = get_dpifindex(dp);
397 	if (!dp_ifindex)
398 		return -ENODEV;
399 
400 	if (skb_vlan_tag_present(skb)) {
401 		nskb = skb_clone(skb, GFP_ATOMIC);
402 		if (!nskb)
403 			return -ENOMEM;
404 
405 		nskb = __vlan_hwaccel_push_inside(nskb);
406 		if (!nskb)
407 			return -ENOMEM;
408 
409 		skb = nskb;
410 	}
411 
412 	if (nla_attr_size(skb->len) > USHRT_MAX) {
413 		err = -EFBIG;
414 		goto out;
415 	}
416 
417 	/* Complete checksum if needed */
418 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
419 	    (err = skb_csum_hwoffload_help(skb, 0)))
420 		goto out;
421 
422 	/* Older versions of OVS user space enforce alignment of the last
423 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
424 	 * padding logic. Only perform zerocopy if padding is not required.
425 	 */
426 	if (dp->user_features & OVS_DP_F_UNALIGNED)
427 		hlen = skb_zerocopy_headlen(skb);
428 	else
429 		hlen = skb->len;
430 
431 	len = upcall_msg_size(upcall_info, hlen - cutlen,
432 			      OVS_CB(skb)->acts_origlen);
433 	user_skb = genlmsg_new(len, GFP_ATOMIC);
434 	if (!user_skb) {
435 		err = -ENOMEM;
436 		goto out;
437 	}
438 
439 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
440 			     0, upcall_info->cmd);
441 	if (!upcall) {
442 		err = -EINVAL;
443 		goto out;
444 	}
445 	upcall->dp_ifindex = dp_ifindex;
446 
447 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
448 	if (err)
449 		goto out;
450 
451 	if (upcall_info->userdata)
452 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
453 			  nla_len(upcall_info->userdata),
454 			  nla_data(upcall_info->userdata));
455 
456 	if (upcall_info->egress_tun_info) {
457 		nla = nla_nest_start_noflag(user_skb,
458 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
459 		if (!nla) {
460 			err = -EMSGSIZE;
461 			goto out;
462 		}
463 		err = ovs_nla_put_tunnel_info(user_skb,
464 					      upcall_info->egress_tun_info);
465 		if (err)
466 			goto out;
467 
468 		nla_nest_end(user_skb, nla);
469 	}
470 
471 	if (upcall_info->actions_len) {
472 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
473 		if (!nla) {
474 			err = -EMSGSIZE;
475 			goto out;
476 		}
477 		err = ovs_nla_put_actions(upcall_info->actions,
478 					  upcall_info->actions_len,
479 					  user_skb);
480 		if (!err)
481 			nla_nest_end(user_skb, nla);
482 		else
483 			nla_nest_cancel(user_skb, nla);
484 	}
485 
486 	/* Add OVS_PACKET_ATTR_MRU */
487 	if (upcall_info->mru) {
488 		if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
489 				upcall_info->mru)) {
490 			err = -ENOBUFS;
491 			goto out;
492 		}
493 		pad_packet(dp, user_skb);
494 	}
495 
496 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
497 	if (cutlen > 0) {
498 		if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
499 				skb->len)) {
500 			err = -ENOBUFS;
501 			goto out;
502 		}
503 		pad_packet(dp, user_skb);
504 	}
505 
506 	/* Only reserve room for attribute header, packet data is added
507 	 * in skb_zerocopy() */
508 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
509 		err = -ENOBUFS;
510 		goto out;
511 	}
512 	nla->nla_len = nla_attr_size(skb->len - cutlen);
513 
514 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
515 	if (err)
516 		goto out;
517 
518 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
519 	pad_packet(dp, user_skb);
520 
521 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
522 
523 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
524 	user_skb = NULL;
525 out:
526 	if (err)
527 		skb_tx_error(skb);
528 	kfree_skb(user_skb);
529 	kfree_skb(nskb);
530 	return err;
531 }
532 
533 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
534 {
535 	struct ovs_header *ovs_header = info->userhdr;
536 	struct net *net = sock_net(skb->sk);
537 	struct nlattr **a = info->attrs;
538 	struct sw_flow_actions *acts;
539 	struct sk_buff *packet;
540 	struct sw_flow *flow;
541 	struct sw_flow_actions *sf_acts;
542 	struct datapath *dp;
543 	struct vport *input_vport;
544 	u16 mru = 0;
545 	int len;
546 	int err;
547 	bool log = !a[OVS_PACKET_ATTR_PROBE];
548 
549 	err = -EINVAL;
550 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
551 	    !a[OVS_PACKET_ATTR_ACTIONS])
552 		goto err;
553 
554 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
555 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
556 	err = -ENOMEM;
557 	if (!packet)
558 		goto err;
559 	skb_reserve(packet, NET_IP_ALIGN);
560 
561 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
562 
563 	/* Set packet's mru */
564 	if (a[OVS_PACKET_ATTR_MRU]) {
565 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
566 		packet->ignore_df = 1;
567 	}
568 	OVS_CB(packet)->mru = mru;
569 
570 	/* Build an sw_flow for sending this packet. */
571 	flow = ovs_flow_alloc();
572 	err = PTR_ERR(flow);
573 	if (IS_ERR(flow))
574 		goto err_kfree_skb;
575 
576 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
577 					     packet, &flow->key, log);
578 	if (err)
579 		goto err_flow_free;
580 
581 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
582 				   &flow->key, &acts, log);
583 	if (err)
584 		goto err_flow_free;
585 
586 	rcu_assign_pointer(flow->sf_acts, acts);
587 	packet->priority = flow->key.phy.priority;
588 	packet->mark = flow->key.phy.skb_mark;
589 
590 	rcu_read_lock();
591 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
592 	err = -ENODEV;
593 	if (!dp)
594 		goto err_unlock;
595 
596 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
597 	if (!input_vport)
598 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
599 
600 	if (!input_vport)
601 		goto err_unlock;
602 
603 	packet->dev = input_vport->dev;
604 	OVS_CB(packet)->input_vport = input_vport;
605 	sf_acts = rcu_dereference(flow->sf_acts);
606 
607 	local_bh_disable();
608 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
609 	local_bh_enable();
610 	rcu_read_unlock();
611 
612 	ovs_flow_free(flow, false);
613 	return err;
614 
615 err_unlock:
616 	rcu_read_unlock();
617 err_flow_free:
618 	ovs_flow_free(flow, false);
619 err_kfree_skb:
620 	kfree_skb(packet);
621 err:
622 	return err;
623 }
624 
625 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
626 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
627 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
628 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
629 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
630 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
631 };
632 
633 static const struct genl_ops dp_packet_genl_ops[] = {
634 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
635 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
636 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
637 	  .doit = ovs_packet_cmd_execute
638 	}
639 };
640 
641 static struct genl_family dp_packet_genl_family __ro_after_init = {
642 	.hdrsize = sizeof(struct ovs_header),
643 	.name = OVS_PACKET_FAMILY,
644 	.version = OVS_PACKET_VERSION,
645 	.maxattr = OVS_PACKET_ATTR_MAX,
646 	.policy = packet_policy,
647 	.netnsok = true,
648 	.parallel_ops = true,
649 	.ops = dp_packet_genl_ops,
650 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
651 	.module = THIS_MODULE,
652 };
653 
654 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
655 			 struct ovs_dp_megaflow_stats *mega_stats)
656 {
657 	int i;
658 
659 	memset(mega_stats, 0, sizeof(*mega_stats));
660 
661 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
662 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
663 
664 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
665 
666 	for_each_possible_cpu(i) {
667 		const struct dp_stats_percpu *percpu_stats;
668 		struct dp_stats_percpu local_stats;
669 		unsigned int start;
670 
671 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
672 
673 		do {
674 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
675 			local_stats = *percpu_stats;
676 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
677 
678 		stats->n_hit += local_stats.n_hit;
679 		stats->n_missed += local_stats.n_missed;
680 		stats->n_lost += local_stats.n_lost;
681 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
682 	}
683 }
684 
685 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
686 {
687 	return ovs_identifier_is_ufid(sfid) &&
688 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
689 }
690 
691 static bool should_fill_mask(uint32_t ufid_flags)
692 {
693 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
694 }
695 
696 static bool should_fill_actions(uint32_t ufid_flags)
697 {
698 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
699 }
700 
701 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
702 				    const struct sw_flow_id *sfid,
703 				    uint32_t ufid_flags)
704 {
705 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
706 
707 	/* OVS_FLOW_ATTR_UFID */
708 	if (sfid && ovs_identifier_is_ufid(sfid))
709 		len += nla_total_size(sfid->ufid_len);
710 
711 	/* OVS_FLOW_ATTR_KEY */
712 	if (!sfid || should_fill_key(sfid, ufid_flags))
713 		len += nla_total_size(ovs_key_attr_size());
714 
715 	/* OVS_FLOW_ATTR_MASK */
716 	if (should_fill_mask(ufid_flags))
717 		len += nla_total_size(ovs_key_attr_size());
718 
719 	/* OVS_FLOW_ATTR_ACTIONS */
720 	if (should_fill_actions(ufid_flags))
721 		len += nla_total_size(acts->orig_len);
722 
723 	return len
724 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
725 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
726 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
727 }
728 
729 /* Called with ovs_mutex or RCU read lock. */
730 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
731 				   struct sk_buff *skb)
732 {
733 	struct ovs_flow_stats stats;
734 	__be16 tcp_flags;
735 	unsigned long used;
736 
737 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
738 
739 	if (used &&
740 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
741 			      OVS_FLOW_ATTR_PAD))
742 		return -EMSGSIZE;
743 
744 	if (stats.n_packets &&
745 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
746 			  sizeof(struct ovs_flow_stats), &stats,
747 			  OVS_FLOW_ATTR_PAD))
748 		return -EMSGSIZE;
749 
750 	if ((u8)ntohs(tcp_flags) &&
751 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
752 		return -EMSGSIZE;
753 
754 	return 0;
755 }
756 
757 /* Called with ovs_mutex or RCU read lock. */
758 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
759 				     struct sk_buff *skb, int skb_orig_len)
760 {
761 	struct nlattr *start;
762 	int err;
763 
764 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
765 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
766 	 * Netlink but individual action lists can be longer than
767 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
768 	 * The userspace caller can always fetch the actions separately if it
769 	 * really wants them.  (Most userspace callers in fact don't care.)
770 	 *
771 	 * This can only fail for dump operations because the skb is always
772 	 * properly sized for single flows.
773 	 */
774 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
775 	if (start) {
776 		const struct sw_flow_actions *sf_acts;
777 
778 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
779 		err = ovs_nla_put_actions(sf_acts->actions,
780 					  sf_acts->actions_len, skb);
781 
782 		if (!err)
783 			nla_nest_end(skb, start);
784 		else {
785 			if (skb_orig_len)
786 				return err;
787 
788 			nla_nest_cancel(skb, start);
789 		}
790 	} else if (skb_orig_len) {
791 		return -EMSGSIZE;
792 	}
793 
794 	return 0;
795 }
796 
797 /* Called with ovs_mutex or RCU read lock. */
798 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
799 				  struct sk_buff *skb, u32 portid,
800 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
801 {
802 	const int skb_orig_len = skb->len;
803 	struct ovs_header *ovs_header;
804 	int err;
805 
806 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
807 				 flags, cmd);
808 	if (!ovs_header)
809 		return -EMSGSIZE;
810 
811 	ovs_header->dp_ifindex = dp_ifindex;
812 
813 	err = ovs_nla_put_identifier(flow, skb);
814 	if (err)
815 		goto error;
816 
817 	if (should_fill_key(&flow->id, ufid_flags)) {
818 		err = ovs_nla_put_masked_key(flow, skb);
819 		if (err)
820 			goto error;
821 	}
822 
823 	if (should_fill_mask(ufid_flags)) {
824 		err = ovs_nla_put_mask(flow, skb);
825 		if (err)
826 			goto error;
827 	}
828 
829 	err = ovs_flow_cmd_fill_stats(flow, skb);
830 	if (err)
831 		goto error;
832 
833 	if (should_fill_actions(ufid_flags)) {
834 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
835 		if (err)
836 			goto error;
837 	}
838 
839 	genlmsg_end(skb, ovs_header);
840 	return 0;
841 
842 error:
843 	genlmsg_cancel(skb, ovs_header);
844 	return err;
845 }
846 
847 /* May not be called with RCU read lock. */
848 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
849 					       const struct sw_flow_id *sfid,
850 					       struct genl_info *info,
851 					       bool always,
852 					       uint32_t ufid_flags)
853 {
854 	struct sk_buff *skb;
855 	size_t len;
856 
857 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
858 		return NULL;
859 
860 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
861 	skb = genlmsg_new(len, GFP_KERNEL);
862 	if (!skb)
863 		return ERR_PTR(-ENOMEM);
864 
865 	return skb;
866 }
867 
868 /* Called with ovs_mutex. */
869 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
870 					       int dp_ifindex,
871 					       struct genl_info *info, u8 cmd,
872 					       bool always, u32 ufid_flags)
873 {
874 	struct sk_buff *skb;
875 	int retval;
876 
877 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
878 				      &flow->id, info, always, ufid_flags);
879 	if (IS_ERR_OR_NULL(skb))
880 		return skb;
881 
882 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
883 					info->snd_portid, info->snd_seq, 0,
884 					cmd, ufid_flags);
885 	BUG_ON(retval < 0);
886 	return skb;
887 }
888 
889 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
890 {
891 	struct net *net = sock_net(skb->sk);
892 	struct nlattr **a = info->attrs;
893 	struct ovs_header *ovs_header = info->userhdr;
894 	struct sw_flow *flow = NULL, *new_flow;
895 	struct sw_flow_mask mask;
896 	struct sk_buff *reply;
897 	struct datapath *dp;
898 	struct sw_flow_actions *acts;
899 	struct sw_flow_match match;
900 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
901 	int error;
902 	bool log = !a[OVS_FLOW_ATTR_PROBE];
903 
904 	/* Must have key and actions. */
905 	error = -EINVAL;
906 	if (!a[OVS_FLOW_ATTR_KEY]) {
907 		OVS_NLERR(log, "Flow key attr not present in new flow.");
908 		goto error;
909 	}
910 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
911 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
912 		goto error;
913 	}
914 
915 	/* Most of the time we need to allocate a new flow, do it before
916 	 * locking.
917 	 */
918 	new_flow = ovs_flow_alloc();
919 	if (IS_ERR(new_flow)) {
920 		error = PTR_ERR(new_flow);
921 		goto error;
922 	}
923 
924 	/* Extract key. */
925 	ovs_match_init(&match, &new_flow->key, false, &mask);
926 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
927 				  a[OVS_FLOW_ATTR_MASK], log);
928 	if (error)
929 		goto err_kfree_flow;
930 
931 	/* Extract flow identifier. */
932 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
933 				       &new_flow->key, log);
934 	if (error)
935 		goto err_kfree_flow;
936 
937 	/* unmasked key is needed to match when ufid is not used. */
938 	if (ovs_identifier_is_key(&new_flow->id))
939 		match.key = new_flow->id.unmasked_key;
940 
941 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
942 
943 	/* Validate actions. */
944 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
945 				     &new_flow->key, &acts, log);
946 	if (error) {
947 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
948 		goto err_kfree_flow;
949 	}
950 
951 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
952 					ufid_flags);
953 	if (IS_ERR(reply)) {
954 		error = PTR_ERR(reply);
955 		goto err_kfree_acts;
956 	}
957 
958 	ovs_lock();
959 	dp = get_dp(net, ovs_header->dp_ifindex);
960 	if (unlikely(!dp)) {
961 		error = -ENODEV;
962 		goto err_unlock_ovs;
963 	}
964 
965 	/* Check if this is a duplicate flow */
966 	if (ovs_identifier_is_ufid(&new_flow->id))
967 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
968 	if (!flow)
969 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
970 	if (likely(!flow)) {
971 		rcu_assign_pointer(new_flow->sf_acts, acts);
972 
973 		/* Put flow in bucket. */
974 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
975 		if (unlikely(error)) {
976 			acts = NULL;
977 			goto err_unlock_ovs;
978 		}
979 
980 		if (unlikely(reply)) {
981 			error = ovs_flow_cmd_fill_info(new_flow,
982 						       ovs_header->dp_ifindex,
983 						       reply, info->snd_portid,
984 						       info->snd_seq, 0,
985 						       OVS_FLOW_CMD_NEW,
986 						       ufid_flags);
987 			BUG_ON(error < 0);
988 		}
989 		ovs_unlock();
990 	} else {
991 		struct sw_flow_actions *old_acts;
992 
993 		/* Bail out if we're not allowed to modify an existing flow.
994 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
995 		 * because Generic Netlink treats the latter as a dump
996 		 * request.  We also accept NLM_F_EXCL in case that bug ever
997 		 * gets fixed.
998 		 */
999 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1000 							 | NLM_F_EXCL))) {
1001 			error = -EEXIST;
1002 			goto err_unlock_ovs;
1003 		}
1004 		/* The flow identifier has to be the same for flow updates.
1005 		 * Look for any overlapping flow.
1006 		 */
1007 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1008 			if (ovs_identifier_is_key(&flow->id))
1009 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1010 								 &match);
1011 			else /* UFID matches but key is different */
1012 				flow = NULL;
1013 			if (!flow) {
1014 				error = -ENOENT;
1015 				goto err_unlock_ovs;
1016 			}
1017 		}
1018 		/* Update actions. */
1019 		old_acts = ovsl_dereference(flow->sf_acts);
1020 		rcu_assign_pointer(flow->sf_acts, acts);
1021 
1022 		if (unlikely(reply)) {
1023 			error = ovs_flow_cmd_fill_info(flow,
1024 						       ovs_header->dp_ifindex,
1025 						       reply, info->snd_portid,
1026 						       info->snd_seq, 0,
1027 						       OVS_FLOW_CMD_NEW,
1028 						       ufid_flags);
1029 			BUG_ON(error < 0);
1030 		}
1031 		ovs_unlock();
1032 
1033 		ovs_nla_free_flow_actions_rcu(old_acts);
1034 		ovs_flow_free(new_flow, false);
1035 	}
1036 
1037 	if (reply)
1038 		ovs_notify(&dp_flow_genl_family, reply, info);
1039 	return 0;
1040 
1041 err_unlock_ovs:
1042 	ovs_unlock();
1043 	kfree_skb(reply);
1044 err_kfree_acts:
1045 	ovs_nla_free_flow_actions(acts);
1046 err_kfree_flow:
1047 	ovs_flow_free(new_flow, false);
1048 error:
1049 	return error;
1050 }
1051 
1052 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1053 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1054 						const struct nlattr *a,
1055 						const struct sw_flow_key *key,
1056 						const struct sw_flow_mask *mask,
1057 						bool log)
1058 {
1059 	struct sw_flow_actions *acts;
1060 	struct sw_flow_key masked_key;
1061 	int error;
1062 
1063 	ovs_flow_mask_key(&masked_key, key, true, mask);
1064 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1065 	if (error) {
1066 		OVS_NLERR(log,
1067 			  "Actions may not be safe on all matching packets");
1068 		return ERR_PTR(error);
1069 	}
1070 
1071 	return acts;
1072 }
1073 
1074 /* Factor out match-init and action-copy to avoid
1075  * "Wframe-larger-than=1024" warning. Because mask is only
1076  * used to get actions, we new a function to save some
1077  * stack space.
1078  *
1079  * If there are not key and action attrs, we return 0
1080  * directly. In the case, the caller will also not use the
1081  * match as before. If there is action attr, we try to get
1082  * actions and save them to *acts. Before returning from
1083  * the function, we reset the match->mask pointer. Because
1084  * we should not to return match object with dangling reference
1085  * to mask.
1086  * */
1087 static noinline_for_stack int
1088 ovs_nla_init_match_and_action(struct net *net,
1089 			      struct sw_flow_match *match,
1090 			      struct sw_flow_key *key,
1091 			      struct nlattr **a,
1092 			      struct sw_flow_actions **acts,
1093 			      bool log)
1094 {
1095 	struct sw_flow_mask mask;
1096 	int error = 0;
1097 
1098 	if (a[OVS_FLOW_ATTR_KEY]) {
1099 		ovs_match_init(match, key, true, &mask);
1100 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1101 					  a[OVS_FLOW_ATTR_MASK], log);
1102 		if (error)
1103 			goto error;
1104 	}
1105 
1106 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1107 		if (!a[OVS_FLOW_ATTR_KEY]) {
1108 			OVS_NLERR(log,
1109 				  "Flow key attribute not present in set flow.");
1110 			error = -EINVAL;
1111 			goto error;
1112 		}
1113 
1114 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1115 					 &mask, log);
1116 		if (IS_ERR(*acts)) {
1117 			error = PTR_ERR(*acts);
1118 			goto error;
1119 		}
1120 	}
1121 
1122 	/* On success, error is 0. */
1123 error:
1124 	match->mask = NULL;
1125 	return error;
1126 }
1127 
1128 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1129 {
1130 	struct net *net = sock_net(skb->sk);
1131 	struct nlattr **a = info->attrs;
1132 	struct ovs_header *ovs_header = info->userhdr;
1133 	struct sw_flow_key key;
1134 	struct sw_flow *flow;
1135 	struct sk_buff *reply = NULL;
1136 	struct datapath *dp;
1137 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1138 	struct sw_flow_match match;
1139 	struct sw_flow_id sfid;
1140 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1141 	int error = 0;
1142 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1143 	bool ufid_present;
1144 
1145 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1146 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1147 		OVS_NLERR(log,
1148 			  "Flow set message rejected, Key attribute missing.");
1149 		return -EINVAL;
1150 	}
1151 
1152 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1153 					      &acts, log);
1154 	if (error)
1155 		goto error;
1156 
1157 	if (acts) {
1158 		/* Can allocate before locking if have acts. */
1159 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1160 						ufid_flags);
1161 		if (IS_ERR(reply)) {
1162 			error = PTR_ERR(reply);
1163 			goto err_kfree_acts;
1164 		}
1165 	}
1166 
1167 	ovs_lock();
1168 	dp = get_dp(net, ovs_header->dp_ifindex);
1169 	if (unlikely(!dp)) {
1170 		error = -ENODEV;
1171 		goto err_unlock_ovs;
1172 	}
1173 	/* Check that the flow exists. */
1174 	if (ufid_present)
1175 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1176 	else
1177 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1178 	if (unlikely(!flow)) {
1179 		error = -ENOENT;
1180 		goto err_unlock_ovs;
1181 	}
1182 
1183 	/* Update actions, if present. */
1184 	if (likely(acts)) {
1185 		old_acts = ovsl_dereference(flow->sf_acts);
1186 		rcu_assign_pointer(flow->sf_acts, acts);
1187 
1188 		if (unlikely(reply)) {
1189 			error = ovs_flow_cmd_fill_info(flow,
1190 						       ovs_header->dp_ifindex,
1191 						       reply, info->snd_portid,
1192 						       info->snd_seq, 0,
1193 						       OVS_FLOW_CMD_SET,
1194 						       ufid_flags);
1195 			BUG_ON(error < 0);
1196 		}
1197 	} else {
1198 		/* Could not alloc without acts before locking. */
1199 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1200 						info, OVS_FLOW_CMD_SET, false,
1201 						ufid_flags);
1202 
1203 		if (IS_ERR(reply)) {
1204 			error = PTR_ERR(reply);
1205 			goto err_unlock_ovs;
1206 		}
1207 	}
1208 
1209 	/* Clear stats. */
1210 	if (a[OVS_FLOW_ATTR_CLEAR])
1211 		ovs_flow_stats_clear(flow);
1212 	ovs_unlock();
1213 
1214 	if (reply)
1215 		ovs_notify(&dp_flow_genl_family, reply, info);
1216 	if (old_acts)
1217 		ovs_nla_free_flow_actions_rcu(old_acts);
1218 
1219 	return 0;
1220 
1221 err_unlock_ovs:
1222 	ovs_unlock();
1223 	kfree_skb(reply);
1224 err_kfree_acts:
1225 	ovs_nla_free_flow_actions(acts);
1226 error:
1227 	return error;
1228 }
1229 
1230 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1231 {
1232 	struct nlattr **a = info->attrs;
1233 	struct ovs_header *ovs_header = info->userhdr;
1234 	struct net *net = sock_net(skb->sk);
1235 	struct sw_flow_key key;
1236 	struct sk_buff *reply;
1237 	struct sw_flow *flow;
1238 	struct datapath *dp;
1239 	struct sw_flow_match match;
1240 	struct sw_flow_id ufid;
1241 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1242 	int err = 0;
1243 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1244 	bool ufid_present;
1245 
1246 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1247 	if (a[OVS_FLOW_ATTR_KEY]) {
1248 		ovs_match_init(&match, &key, true, NULL);
1249 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1250 					log);
1251 	} else if (!ufid_present) {
1252 		OVS_NLERR(log,
1253 			  "Flow get message rejected, Key attribute missing.");
1254 		err = -EINVAL;
1255 	}
1256 	if (err)
1257 		return err;
1258 
1259 	ovs_lock();
1260 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1261 	if (!dp) {
1262 		err = -ENODEV;
1263 		goto unlock;
1264 	}
1265 
1266 	if (ufid_present)
1267 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1268 	else
1269 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1270 	if (!flow) {
1271 		err = -ENOENT;
1272 		goto unlock;
1273 	}
1274 
1275 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1276 					OVS_FLOW_CMD_GET, true, ufid_flags);
1277 	if (IS_ERR(reply)) {
1278 		err = PTR_ERR(reply);
1279 		goto unlock;
1280 	}
1281 
1282 	ovs_unlock();
1283 	return genlmsg_reply(reply, info);
1284 unlock:
1285 	ovs_unlock();
1286 	return err;
1287 }
1288 
1289 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1290 {
1291 	struct nlattr **a = info->attrs;
1292 	struct ovs_header *ovs_header = info->userhdr;
1293 	struct net *net = sock_net(skb->sk);
1294 	struct sw_flow_key key;
1295 	struct sk_buff *reply;
1296 	struct sw_flow *flow = NULL;
1297 	struct datapath *dp;
1298 	struct sw_flow_match match;
1299 	struct sw_flow_id ufid;
1300 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1301 	int err;
1302 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1303 	bool ufid_present;
1304 
1305 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1306 	if (a[OVS_FLOW_ATTR_KEY]) {
1307 		ovs_match_init(&match, &key, true, NULL);
1308 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1309 					NULL, log);
1310 		if (unlikely(err))
1311 			return err;
1312 	}
1313 
1314 	ovs_lock();
1315 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1316 	if (unlikely(!dp)) {
1317 		err = -ENODEV;
1318 		goto unlock;
1319 	}
1320 
1321 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1322 		err = ovs_flow_tbl_flush(&dp->table);
1323 		goto unlock;
1324 	}
1325 
1326 	if (ufid_present)
1327 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1328 	else
1329 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1330 	if (unlikely(!flow)) {
1331 		err = -ENOENT;
1332 		goto unlock;
1333 	}
1334 
1335 	ovs_flow_tbl_remove(&dp->table, flow);
1336 	ovs_unlock();
1337 
1338 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1339 					&flow->id, info, false, ufid_flags);
1340 	if (likely(reply)) {
1341 		if (!IS_ERR(reply)) {
1342 			rcu_read_lock();	/*To keep RCU checker happy. */
1343 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1344 						     reply, info->snd_portid,
1345 						     info->snd_seq, 0,
1346 						     OVS_FLOW_CMD_DEL,
1347 						     ufid_flags);
1348 			rcu_read_unlock();
1349 			BUG_ON(err < 0);
1350 
1351 			ovs_notify(&dp_flow_genl_family, reply, info);
1352 		} else {
1353 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1354 		}
1355 	}
1356 
1357 	ovs_flow_free(flow, true);
1358 	return 0;
1359 unlock:
1360 	ovs_unlock();
1361 	return err;
1362 }
1363 
1364 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1365 {
1366 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1367 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1368 	struct table_instance *ti;
1369 	struct datapath *dp;
1370 	u32 ufid_flags;
1371 	int err;
1372 
1373 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1374 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1375 	if (err)
1376 		return err;
1377 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1378 
1379 	rcu_read_lock();
1380 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1381 	if (!dp) {
1382 		rcu_read_unlock();
1383 		return -ENODEV;
1384 	}
1385 
1386 	ti = rcu_dereference(dp->table.ti);
1387 	for (;;) {
1388 		struct sw_flow *flow;
1389 		u32 bucket, obj;
1390 
1391 		bucket = cb->args[0];
1392 		obj = cb->args[1];
1393 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1394 		if (!flow)
1395 			break;
1396 
1397 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1398 					   NETLINK_CB(cb->skb).portid,
1399 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1400 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1401 			break;
1402 
1403 		cb->args[0] = bucket;
1404 		cb->args[1] = obj;
1405 	}
1406 	rcu_read_unlock();
1407 	return skb->len;
1408 }
1409 
1410 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1411 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1412 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1413 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1414 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1415 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1416 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1417 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1418 };
1419 
1420 static const struct genl_ops dp_flow_genl_ops[] = {
1421 	{ .cmd = OVS_FLOW_CMD_NEW,
1422 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1423 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1424 	  .doit = ovs_flow_cmd_new
1425 	},
1426 	{ .cmd = OVS_FLOW_CMD_DEL,
1427 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1428 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1429 	  .doit = ovs_flow_cmd_del
1430 	},
1431 	{ .cmd = OVS_FLOW_CMD_GET,
1432 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1433 	  .flags = 0,		    /* OK for unprivileged users. */
1434 	  .doit = ovs_flow_cmd_get,
1435 	  .dumpit = ovs_flow_cmd_dump
1436 	},
1437 	{ .cmd = OVS_FLOW_CMD_SET,
1438 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1439 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1440 	  .doit = ovs_flow_cmd_set,
1441 	},
1442 };
1443 
1444 static struct genl_family dp_flow_genl_family __ro_after_init = {
1445 	.hdrsize = sizeof(struct ovs_header),
1446 	.name = OVS_FLOW_FAMILY,
1447 	.version = OVS_FLOW_VERSION,
1448 	.maxattr = OVS_FLOW_ATTR_MAX,
1449 	.policy = flow_policy,
1450 	.netnsok = true,
1451 	.parallel_ops = true,
1452 	.ops = dp_flow_genl_ops,
1453 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1454 	.mcgrps = &ovs_dp_flow_multicast_group,
1455 	.n_mcgrps = 1,
1456 	.module = THIS_MODULE,
1457 };
1458 
1459 static size_t ovs_dp_cmd_msg_size(void)
1460 {
1461 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1462 
1463 	msgsize += nla_total_size(IFNAMSIZ);
1464 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1465 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1466 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1467 
1468 	return msgsize;
1469 }
1470 
1471 /* Called with ovs_mutex. */
1472 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1473 				u32 portid, u32 seq, u32 flags, u8 cmd)
1474 {
1475 	struct ovs_header *ovs_header;
1476 	struct ovs_dp_stats dp_stats;
1477 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1478 	int err;
1479 
1480 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1481 				   flags, cmd);
1482 	if (!ovs_header)
1483 		goto error;
1484 
1485 	ovs_header->dp_ifindex = get_dpifindex(dp);
1486 
1487 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1488 	if (err)
1489 		goto nla_put_failure;
1490 
1491 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1492 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1493 			  &dp_stats, OVS_DP_ATTR_PAD))
1494 		goto nla_put_failure;
1495 
1496 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1497 			  sizeof(struct ovs_dp_megaflow_stats),
1498 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1499 		goto nla_put_failure;
1500 
1501 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1502 		goto nla_put_failure;
1503 
1504 	genlmsg_end(skb, ovs_header);
1505 	return 0;
1506 
1507 nla_put_failure:
1508 	genlmsg_cancel(skb, ovs_header);
1509 error:
1510 	return -EMSGSIZE;
1511 }
1512 
1513 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1514 {
1515 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1516 }
1517 
1518 /* Called with rcu_read_lock or ovs_mutex. */
1519 static struct datapath *lookup_datapath(struct net *net,
1520 					const struct ovs_header *ovs_header,
1521 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1522 {
1523 	struct datapath *dp;
1524 
1525 	if (!a[OVS_DP_ATTR_NAME])
1526 		dp = get_dp(net, ovs_header->dp_ifindex);
1527 	else {
1528 		struct vport *vport;
1529 
1530 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1531 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1532 	}
1533 	return dp ? dp : ERR_PTR(-ENODEV);
1534 }
1535 
1536 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1537 {
1538 	struct datapath *dp;
1539 
1540 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1541 	if (IS_ERR(dp))
1542 		return;
1543 
1544 	WARN(dp->user_features, "Dropping previously announced user features\n");
1545 	dp->user_features = 0;
1546 }
1547 
1548 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1549 {
1550 	if (a[OVS_DP_ATTR_USER_FEATURES])
1551 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1552 }
1553 
1554 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1555 {
1556 	struct nlattr **a = info->attrs;
1557 	struct vport_parms parms;
1558 	struct sk_buff *reply;
1559 	struct datapath *dp;
1560 	struct vport *vport;
1561 	struct ovs_net *ovs_net;
1562 	int err, i;
1563 
1564 	err = -EINVAL;
1565 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1566 		goto err;
1567 
1568 	reply = ovs_dp_cmd_alloc_info();
1569 	if (!reply)
1570 		return -ENOMEM;
1571 
1572 	err = -ENOMEM;
1573 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1574 	if (dp == NULL)
1575 		goto err_free_reply;
1576 
1577 	ovs_dp_set_net(dp, sock_net(skb->sk));
1578 
1579 	/* Allocate table. */
1580 	err = ovs_flow_tbl_init(&dp->table);
1581 	if (err)
1582 		goto err_free_dp;
1583 
1584 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1585 	if (!dp->stats_percpu) {
1586 		err = -ENOMEM;
1587 		goto err_destroy_table;
1588 	}
1589 
1590 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1591 				  sizeof(struct hlist_head),
1592 				  GFP_KERNEL);
1593 	if (!dp->ports) {
1594 		err = -ENOMEM;
1595 		goto err_destroy_percpu;
1596 	}
1597 
1598 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1599 		INIT_HLIST_HEAD(&dp->ports[i]);
1600 
1601 	err = ovs_meters_init(dp);
1602 	if (err)
1603 		goto err_destroy_ports_array;
1604 
1605 	/* Set up our datapath device. */
1606 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1607 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1608 	parms.options = NULL;
1609 	parms.dp = dp;
1610 	parms.port_no = OVSP_LOCAL;
1611 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1612 
1613 	ovs_dp_change(dp, a);
1614 
1615 	/* So far only local changes have been made, now need the lock. */
1616 	ovs_lock();
1617 
1618 	vport = new_vport(&parms);
1619 	if (IS_ERR(vport)) {
1620 		err = PTR_ERR(vport);
1621 		if (err == -EBUSY)
1622 			err = -EEXIST;
1623 
1624 		if (err == -EEXIST) {
1625 			/* An outdated user space instance that does not understand
1626 			 * the concept of user_features has attempted to create a new
1627 			 * datapath and is likely to reuse it. Drop all user features.
1628 			 */
1629 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1630 				ovs_dp_reset_user_features(skb, info);
1631 		}
1632 
1633 		goto err_destroy_meters;
1634 	}
1635 
1636 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1637 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1638 	BUG_ON(err < 0);
1639 
1640 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1641 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1642 
1643 	ovs_unlock();
1644 
1645 	ovs_notify(&dp_datapath_genl_family, reply, info);
1646 	return 0;
1647 
1648 err_destroy_meters:
1649 	ovs_unlock();
1650 	ovs_meters_exit(dp);
1651 err_destroy_ports_array:
1652 	kfree(dp->ports);
1653 err_destroy_percpu:
1654 	free_percpu(dp->stats_percpu);
1655 err_destroy_table:
1656 	ovs_flow_tbl_destroy(&dp->table);
1657 err_free_dp:
1658 	kfree(dp);
1659 err_free_reply:
1660 	kfree_skb(reply);
1661 err:
1662 	return err;
1663 }
1664 
1665 /* Called with ovs_mutex. */
1666 static void __dp_destroy(struct datapath *dp)
1667 {
1668 	int i;
1669 
1670 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1671 		struct vport *vport;
1672 		struct hlist_node *n;
1673 
1674 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1675 			if (vport->port_no != OVSP_LOCAL)
1676 				ovs_dp_detach_port(vport);
1677 	}
1678 
1679 	list_del_rcu(&dp->list_node);
1680 
1681 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1682 	 * all ports in datapath are destroyed first before freeing datapath.
1683 	 */
1684 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1685 
1686 	/* RCU destroy the flow table */
1687 	call_rcu(&dp->rcu, destroy_dp_rcu);
1688 }
1689 
1690 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1691 {
1692 	struct sk_buff *reply;
1693 	struct datapath *dp;
1694 	int err;
1695 
1696 	reply = ovs_dp_cmd_alloc_info();
1697 	if (!reply)
1698 		return -ENOMEM;
1699 
1700 	ovs_lock();
1701 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1702 	err = PTR_ERR(dp);
1703 	if (IS_ERR(dp))
1704 		goto err_unlock_free;
1705 
1706 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1707 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1708 	BUG_ON(err < 0);
1709 
1710 	__dp_destroy(dp);
1711 	ovs_unlock();
1712 
1713 	ovs_notify(&dp_datapath_genl_family, reply, info);
1714 
1715 	return 0;
1716 
1717 err_unlock_free:
1718 	ovs_unlock();
1719 	kfree_skb(reply);
1720 	return err;
1721 }
1722 
1723 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1724 {
1725 	struct sk_buff *reply;
1726 	struct datapath *dp;
1727 	int err;
1728 
1729 	reply = ovs_dp_cmd_alloc_info();
1730 	if (!reply)
1731 		return -ENOMEM;
1732 
1733 	ovs_lock();
1734 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1735 	err = PTR_ERR(dp);
1736 	if (IS_ERR(dp))
1737 		goto err_unlock_free;
1738 
1739 	ovs_dp_change(dp, info->attrs);
1740 
1741 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1742 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1743 	BUG_ON(err < 0);
1744 
1745 	ovs_unlock();
1746 	ovs_notify(&dp_datapath_genl_family, reply, info);
1747 
1748 	return 0;
1749 
1750 err_unlock_free:
1751 	ovs_unlock();
1752 	kfree_skb(reply);
1753 	return err;
1754 }
1755 
1756 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1757 {
1758 	struct sk_buff *reply;
1759 	struct datapath *dp;
1760 	int err;
1761 
1762 	reply = ovs_dp_cmd_alloc_info();
1763 	if (!reply)
1764 		return -ENOMEM;
1765 
1766 	ovs_lock();
1767 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1768 	if (IS_ERR(dp)) {
1769 		err = PTR_ERR(dp);
1770 		goto err_unlock_free;
1771 	}
1772 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1773 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1774 	BUG_ON(err < 0);
1775 	ovs_unlock();
1776 
1777 	return genlmsg_reply(reply, info);
1778 
1779 err_unlock_free:
1780 	ovs_unlock();
1781 	kfree_skb(reply);
1782 	return err;
1783 }
1784 
1785 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1786 {
1787 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1788 	struct datapath *dp;
1789 	int skip = cb->args[0];
1790 	int i = 0;
1791 
1792 	ovs_lock();
1793 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1794 		if (i >= skip &&
1795 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1796 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1797 					 OVS_DP_CMD_GET) < 0)
1798 			break;
1799 		i++;
1800 	}
1801 	ovs_unlock();
1802 
1803 	cb->args[0] = i;
1804 
1805 	return skb->len;
1806 }
1807 
1808 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1809 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1810 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1811 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1812 };
1813 
1814 static const struct genl_ops dp_datapath_genl_ops[] = {
1815 	{ .cmd = OVS_DP_CMD_NEW,
1816 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1817 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1818 	  .doit = ovs_dp_cmd_new
1819 	},
1820 	{ .cmd = OVS_DP_CMD_DEL,
1821 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1822 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1823 	  .doit = ovs_dp_cmd_del
1824 	},
1825 	{ .cmd = OVS_DP_CMD_GET,
1826 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1827 	  .flags = 0,		    /* OK for unprivileged users. */
1828 	  .doit = ovs_dp_cmd_get,
1829 	  .dumpit = ovs_dp_cmd_dump
1830 	},
1831 	{ .cmd = OVS_DP_CMD_SET,
1832 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1833 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1834 	  .doit = ovs_dp_cmd_set,
1835 	},
1836 };
1837 
1838 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1839 	.hdrsize = sizeof(struct ovs_header),
1840 	.name = OVS_DATAPATH_FAMILY,
1841 	.version = OVS_DATAPATH_VERSION,
1842 	.maxattr = OVS_DP_ATTR_MAX,
1843 	.policy = datapath_policy,
1844 	.netnsok = true,
1845 	.parallel_ops = true,
1846 	.ops = dp_datapath_genl_ops,
1847 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1848 	.mcgrps = &ovs_dp_datapath_multicast_group,
1849 	.n_mcgrps = 1,
1850 	.module = THIS_MODULE,
1851 };
1852 
1853 /* Called with ovs_mutex or RCU read lock. */
1854 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1855 				   struct net *net, u32 portid, u32 seq,
1856 				   u32 flags, u8 cmd)
1857 {
1858 	struct ovs_header *ovs_header;
1859 	struct ovs_vport_stats vport_stats;
1860 	int err;
1861 
1862 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1863 				 flags, cmd);
1864 	if (!ovs_header)
1865 		return -EMSGSIZE;
1866 
1867 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1868 
1869 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1870 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1871 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1872 			   ovs_vport_name(vport)) ||
1873 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1874 		goto nla_put_failure;
1875 
1876 	if (!net_eq(net, dev_net(vport->dev))) {
1877 		int id = peernet2id_alloc(net, dev_net(vport->dev));
1878 
1879 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1880 			goto nla_put_failure;
1881 	}
1882 
1883 	ovs_vport_get_stats(vport, &vport_stats);
1884 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1885 			  sizeof(struct ovs_vport_stats), &vport_stats,
1886 			  OVS_VPORT_ATTR_PAD))
1887 		goto nla_put_failure;
1888 
1889 	if (ovs_vport_get_upcall_portids(vport, skb))
1890 		goto nla_put_failure;
1891 
1892 	err = ovs_vport_get_options(vport, skb);
1893 	if (err == -EMSGSIZE)
1894 		goto error;
1895 
1896 	genlmsg_end(skb, ovs_header);
1897 	return 0;
1898 
1899 nla_put_failure:
1900 	err = -EMSGSIZE;
1901 error:
1902 	genlmsg_cancel(skb, ovs_header);
1903 	return err;
1904 }
1905 
1906 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1907 {
1908 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1909 }
1910 
1911 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1912 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1913 					 u32 portid, u32 seq, u8 cmd)
1914 {
1915 	struct sk_buff *skb;
1916 	int retval;
1917 
1918 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1919 	if (!skb)
1920 		return ERR_PTR(-ENOMEM);
1921 
1922 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd);
1923 	BUG_ON(retval < 0);
1924 
1925 	return skb;
1926 }
1927 
1928 /* Called with ovs_mutex or RCU read lock. */
1929 static struct vport *lookup_vport(struct net *net,
1930 				  const struct ovs_header *ovs_header,
1931 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1932 {
1933 	struct datapath *dp;
1934 	struct vport *vport;
1935 
1936 	if (a[OVS_VPORT_ATTR_IFINDEX])
1937 		return ERR_PTR(-EOPNOTSUPP);
1938 	if (a[OVS_VPORT_ATTR_NAME]) {
1939 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1940 		if (!vport)
1941 			return ERR_PTR(-ENODEV);
1942 		if (ovs_header->dp_ifindex &&
1943 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1944 			return ERR_PTR(-ENODEV);
1945 		return vport;
1946 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1947 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1948 
1949 		if (port_no >= DP_MAX_PORTS)
1950 			return ERR_PTR(-EFBIG);
1951 
1952 		dp = get_dp(net, ovs_header->dp_ifindex);
1953 		if (!dp)
1954 			return ERR_PTR(-ENODEV);
1955 
1956 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1957 		if (!vport)
1958 			return ERR_PTR(-ENODEV);
1959 		return vport;
1960 	} else
1961 		return ERR_PTR(-EINVAL);
1962 
1963 }
1964 
1965 static unsigned int ovs_get_max_headroom(struct datapath *dp)
1966 {
1967 	unsigned int dev_headroom, max_headroom = 0;
1968 	struct net_device *dev;
1969 	struct vport *vport;
1970 	int i;
1971 
1972 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1973 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1974 			dev = vport->dev;
1975 			dev_headroom = netdev_get_fwd_headroom(dev);
1976 			if (dev_headroom > max_headroom)
1977 				max_headroom = dev_headroom;
1978 		}
1979 	}
1980 
1981 	return max_headroom;
1982 }
1983 
1984 /* Called with ovs_mutex */
1985 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
1986 {
1987 	struct vport *vport;
1988 	int i;
1989 
1990 	dp->max_headroom = new_headroom;
1991 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1992 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1993 			netdev_set_rx_headroom(vport->dev, new_headroom);
1994 }
1995 
1996 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1997 {
1998 	struct nlattr **a = info->attrs;
1999 	struct ovs_header *ovs_header = info->userhdr;
2000 	struct vport_parms parms;
2001 	struct sk_buff *reply;
2002 	struct vport *vport;
2003 	struct datapath *dp;
2004 	unsigned int new_headroom;
2005 	u32 port_no;
2006 	int err;
2007 
2008 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2009 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2010 		return -EINVAL;
2011 	if (a[OVS_VPORT_ATTR_IFINDEX])
2012 		return -EOPNOTSUPP;
2013 
2014 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2015 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2016 	if (port_no >= DP_MAX_PORTS)
2017 		return -EFBIG;
2018 
2019 	reply = ovs_vport_cmd_alloc_info();
2020 	if (!reply)
2021 		return -ENOMEM;
2022 
2023 	ovs_lock();
2024 restart:
2025 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2026 	err = -ENODEV;
2027 	if (!dp)
2028 		goto exit_unlock_free;
2029 
2030 	if (port_no) {
2031 		vport = ovs_vport_ovsl(dp, port_no);
2032 		err = -EBUSY;
2033 		if (vport)
2034 			goto exit_unlock_free;
2035 	} else {
2036 		for (port_no = 1; ; port_no++) {
2037 			if (port_no >= DP_MAX_PORTS) {
2038 				err = -EFBIG;
2039 				goto exit_unlock_free;
2040 			}
2041 			vport = ovs_vport_ovsl(dp, port_no);
2042 			if (!vport)
2043 				break;
2044 		}
2045 	}
2046 
2047 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2048 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2049 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2050 	parms.dp = dp;
2051 	parms.port_no = port_no;
2052 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2053 
2054 	vport = new_vport(&parms);
2055 	err = PTR_ERR(vport);
2056 	if (IS_ERR(vport)) {
2057 		if (err == -EAGAIN)
2058 			goto restart;
2059 		goto exit_unlock_free;
2060 	}
2061 
2062 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2063 				      info->snd_portid, info->snd_seq, 0,
2064 				      OVS_VPORT_CMD_NEW);
2065 
2066 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2067 
2068 	if (new_headroom > dp->max_headroom)
2069 		ovs_update_headroom(dp, new_headroom);
2070 	else
2071 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2072 
2073 	BUG_ON(err < 0);
2074 	ovs_unlock();
2075 
2076 	ovs_notify(&dp_vport_genl_family, reply, info);
2077 	return 0;
2078 
2079 exit_unlock_free:
2080 	ovs_unlock();
2081 	kfree_skb(reply);
2082 	return err;
2083 }
2084 
2085 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2086 {
2087 	struct nlattr **a = info->attrs;
2088 	struct sk_buff *reply;
2089 	struct vport *vport;
2090 	int err;
2091 
2092 	reply = ovs_vport_cmd_alloc_info();
2093 	if (!reply)
2094 		return -ENOMEM;
2095 
2096 	ovs_lock();
2097 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2098 	err = PTR_ERR(vport);
2099 	if (IS_ERR(vport))
2100 		goto exit_unlock_free;
2101 
2102 	if (a[OVS_VPORT_ATTR_TYPE] &&
2103 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2104 		err = -EINVAL;
2105 		goto exit_unlock_free;
2106 	}
2107 
2108 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2109 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2110 		if (err)
2111 			goto exit_unlock_free;
2112 	}
2113 
2114 
2115 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2116 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2117 
2118 		err = ovs_vport_set_upcall_portids(vport, ids);
2119 		if (err)
2120 			goto exit_unlock_free;
2121 	}
2122 
2123 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2124 				      info->snd_portid, info->snd_seq, 0,
2125 				      OVS_VPORT_CMD_SET);
2126 	BUG_ON(err < 0);
2127 
2128 	ovs_unlock();
2129 	ovs_notify(&dp_vport_genl_family, reply, info);
2130 	return 0;
2131 
2132 exit_unlock_free:
2133 	ovs_unlock();
2134 	kfree_skb(reply);
2135 	return err;
2136 }
2137 
2138 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2139 {
2140 	bool update_headroom = false;
2141 	struct nlattr **a = info->attrs;
2142 	struct sk_buff *reply;
2143 	struct datapath *dp;
2144 	struct vport *vport;
2145 	unsigned int new_headroom;
2146 	int err;
2147 
2148 	reply = ovs_vport_cmd_alloc_info();
2149 	if (!reply)
2150 		return -ENOMEM;
2151 
2152 	ovs_lock();
2153 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2154 	err = PTR_ERR(vport);
2155 	if (IS_ERR(vport))
2156 		goto exit_unlock_free;
2157 
2158 	if (vport->port_no == OVSP_LOCAL) {
2159 		err = -EINVAL;
2160 		goto exit_unlock_free;
2161 	}
2162 
2163 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2164 				      info->snd_portid, info->snd_seq, 0,
2165 				      OVS_VPORT_CMD_DEL);
2166 	BUG_ON(err < 0);
2167 
2168 	/* the vport deletion may trigger dp headroom update */
2169 	dp = vport->dp;
2170 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2171 		update_headroom = true;
2172 
2173 	netdev_reset_rx_headroom(vport->dev);
2174 	ovs_dp_detach_port(vport);
2175 
2176 	if (update_headroom) {
2177 		new_headroom = ovs_get_max_headroom(dp);
2178 
2179 		if (new_headroom < dp->max_headroom)
2180 			ovs_update_headroom(dp, new_headroom);
2181 	}
2182 	ovs_unlock();
2183 
2184 	ovs_notify(&dp_vport_genl_family, reply, info);
2185 	return 0;
2186 
2187 exit_unlock_free:
2188 	ovs_unlock();
2189 	kfree_skb(reply);
2190 	return err;
2191 }
2192 
2193 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2194 {
2195 	struct nlattr **a = info->attrs;
2196 	struct ovs_header *ovs_header = info->userhdr;
2197 	struct sk_buff *reply;
2198 	struct vport *vport;
2199 	int err;
2200 
2201 	reply = ovs_vport_cmd_alloc_info();
2202 	if (!reply)
2203 		return -ENOMEM;
2204 
2205 	rcu_read_lock();
2206 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2207 	err = PTR_ERR(vport);
2208 	if (IS_ERR(vport))
2209 		goto exit_unlock_free;
2210 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2211 				      info->snd_portid, info->snd_seq, 0,
2212 				      OVS_VPORT_CMD_GET);
2213 	BUG_ON(err < 0);
2214 	rcu_read_unlock();
2215 
2216 	return genlmsg_reply(reply, info);
2217 
2218 exit_unlock_free:
2219 	rcu_read_unlock();
2220 	kfree_skb(reply);
2221 	return err;
2222 }
2223 
2224 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2225 {
2226 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2227 	struct datapath *dp;
2228 	int bucket = cb->args[0], skip = cb->args[1];
2229 	int i, j = 0;
2230 
2231 	rcu_read_lock();
2232 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2233 	if (!dp) {
2234 		rcu_read_unlock();
2235 		return -ENODEV;
2236 	}
2237 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2238 		struct vport *vport;
2239 
2240 		j = 0;
2241 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2242 			if (j >= skip &&
2243 			    ovs_vport_cmd_fill_info(vport, skb,
2244 						    sock_net(skb->sk),
2245 						    NETLINK_CB(cb->skb).portid,
2246 						    cb->nlh->nlmsg_seq,
2247 						    NLM_F_MULTI,
2248 						    OVS_VPORT_CMD_GET) < 0)
2249 				goto out;
2250 
2251 			j++;
2252 		}
2253 		skip = 0;
2254 	}
2255 out:
2256 	rcu_read_unlock();
2257 
2258 	cb->args[0] = i;
2259 	cb->args[1] = j;
2260 
2261 	return skb->len;
2262 }
2263 
2264 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2265 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2266 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2267 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2268 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2269 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2270 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2271 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2272 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2273 };
2274 
2275 static const struct genl_ops dp_vport_genl_ops[] = {
2276 	{ .cmd = OVS_VPORT_CMD_NEW,
2277 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2278 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2279 	  .doit = ovs_vport_cmd_new
2280 	},
2281 	{ .cmd = OVS_VPORT_CMD_DEL,
2282 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2283 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2284 	  .doit = ovs_vport_cmd_del
2285 	},
2286 	{ .cmd = OVS_VPORT_CMD_GET,
2287 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2288 	  .flags = 0,		    /* OK for unprivileged users. */
2289 	  .doit = ovs_vport_cmd_get,
2290 	  .dumpit = ovs_vport_cmd_dump
2291 	},
2292 	{ .cmd = OVS_VPORT_CMD_SET,
2293 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2294 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2295 	  .doit = ovs_vport_cmd_set,
2296 	},
2297 };
2298 
2299 struct genl_family dp_vport_genl_family __ro_after_init = {
2300 	.hdrsize = sizeof(struct ovs_header),
2301 	.name = OVS_VPORT_FAMILY,
2302 	.version = OVS_VPORT_VERSION,
2303 	.maxattr = OVS_VPORT_ATTR_MAX,
2304 	.policy = vport_policy,
2305 	.netnsok = true,
2306 	.parallel_ops = true,
2307 	.ops = dp_vport_genl_ops,
2308 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2309 	.mcgrps = &ovs_dp_vport_multicast_group,
2310 	.n_mcgrps = 1,
2311 	.module = THIS_MODULE,
2312 };
2313 
2314 static struct genl_family * const dp_genl_families[] = {
2315 	&dp_datapath_genl_family,
2316 	&dp_vport_genl_family,
2317 	&dp_flow_genl_family,
2318 	&dp_packet_genl_family,
2319 	&dp_meter_genl_family,
2320 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2321 	&dp_ct_limit_genl_family,
2322 #endif
2323 };
2324 
2325 static void dp_unregister_genl(int n_families)
2326 {
2327 	int i;
2328 
2329 	for (i = 0; i < n_families; i++)
2330 		genl_unregister_family(dp_genl_families[i]);
2331 }
2332 
2333 static int __init dp_register_genl(void)
2334 {
2335 	int err;
2336 	int i;
2337 
2338 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2339 
2340 		err = genl_register_family(dp_genl_families[i]);
2341 		if (err)
2342 			goto error;
2343 	}
2344 
2345 	return 0;
2346 
2347 error:
2348 	dp_unregister_genl(i);
2349 	return err;
2350 }
2351 
2352 static int __net_init ovs_init_net(struct net *net)
2353 {
2354 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2355 
2356 	INIT_LIST_HEAD(&ovs_net->dps);
2357 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2358 	return ovs_ct_init(net);
2359 }
2360 
2361 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2362 					    struct list_head *head)
2363 {
2364 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2365 	struct datapath *dp;
2366 
2367 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2368 		int i;
2369 
2370 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2371 			struct vport *vport;
2372 
2373 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2374 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2375 					continue;
2376 
2377 				if (dev_net(vport->dev) == dnet)
2378 					list_add(&vport->detach_list, head);
2379 			}
2380 		}
2381 	}
2382 }
2383 
2384 static void __net_exit ovs_exit_net(struct net *dnet)
2385 {
2386 	struct datapath *dp, *dp_next;
2387 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2388 	struct vport *vport, *vport_next;
2389 	struct net *net;
2390 	LIST_HEAD(head);
2391 
2392 	ovs_ct_exit(dnet);
2393 	ovs_lock();
2394 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2395 		__dp_destroy(dp);
2396 
2397 	down_read(&net_rwsem);
2398 	for_each_net(net)
2399 		list_vports_from_net(net, dnet, &head);
2400 	up_read(&net_rwsem);
2401 
2402 	/* Detach all vports from given namespace. */
2403 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2404 		list_del(&vport->detach_list);
2405 		ovs_dp_detach_port(vport);
2406 	}
2407 
2408 	ovs_unlock();
2409 
2410 	cancel_work_sync(&ovs_net->dp_notify_work);
2411 }
2412 
2413 static struct pernet_operations ovs_net_ops = {
2414 	.init = ovs_init_net,
2415 	.exit = ovs_exit_net,
2416 	.id   = &ovs_net_id,
2417 	.size = sizeof(struct ovs_net),
2418 };
2419 
2420 static int __init dp_init(void)
2421 {
2422 	int err;
2423 
2424 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2425 
2426 	pr_info("Open vSwitch switching datapath\n");
2427 
2428 	err = action_fifos_init();
2429 	if (err)
2430 		goto error;
2431 
2432 	err = ovs_internal_dev_rtnl_link_register();
2433 	if (err)
2434 		goto error_action_fifos_exit;
2435 
2436 	err = ovs_flow_init();
2437 	if (err)
2438 		goto error_unreg_rtnl_link;
2439 
2440 	err = ovs_vport_init();
2441 	if (err)
2442 		goto error_flow_exit;
2443 
2444 	err = register_pernet_device(&ovs_net_ops);
2445 	if (err)
2446 		goto error_vport_exit;
2447 
2448 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2449 	if (err)
2450 		goto error_netns_exit;
2451 
2452 	err = ovs_netdev_init();
2453 	if (err)
2454 		goto error_unreg_notifier;
2455 
2456 	err = dp_register_genl();
2457 	if (err < 0)
2458 		goto error_unreg_netdev;
2459 
2460 	return 0;
2461 
2462 error_unreg_netdev:
2463 	ovs_netdev_exit();
2464 error_unreg_notifier:
2465 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2466 error_netns_exit:
2467 	unregister_pernet_device(&ovs_net_ops);
2468 error_vport_exit:
2469 	ovs_vport_exit();
2470 error_flow_exit:
2471 	ovs_flow_exit();
2472 error_unreg_rtnl_link:
2473 	ovs_internal_dev_rtnl_link_unregister();
2474 error_action_fifos_exit:
2475 	action_fifos_exit();
2476 error:
2477 	return err;
2478 }
2479 
2480 static void dp_cleanup(void)
2481 {
2482 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2483 	ovs_netdev_exit();
2484 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2485 	unregister_pernet_device(&ovs_net_ops);
2486 	rcu_barrier();
2487 	ovs_vport_exit();
2488 	ovs_flow_exit();
2489 	ovs_internal_dev_rtnl_link_unregister();
2490 	action_fifos_exit();
2491 }
2492 
2493 module_init(dp_init);
2494 module_exit(dp_cleanup);
2495 
2496 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2497 MODULE_LICENSE("GPL");
2498 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2499 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2500 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2501 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2502 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2503 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2504