xref: /openbmc/linux/net/openvswitch/datapath.c (revision f9834f18)
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 				lockdep_ovsl_is_held()) {
184 		if (vport->port_no == port_no)
185 			return vport;
186 	}
187 	return NULL;
188 }
189 
190 /* Called with ovs_mutex. */
191 static struct vport *new_vport(const struct vport_parms *parms)
192 {
193 	struct vport *vport;
194 
195 	vport = ovs_vport_add(parms);
196 	if (!IS_ERR(vport)) {
197 		struct datapath *dp = parms->dp;
198 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
199 
200 		hlist_add_head_rcu(&vport->dp_hash_node, head);
201 	}
202 	return vport;
203 }
204 
205 void ovs_dp_detach_port(struct vport *p)
206 {
207 	ASSERT_OVSL();
208 
209 	/* First drop references to device. */
210 	hlist_del_rcu(&p->dp_hash_node);
211 
212 	/* Then destroy it. */
213 	ovs_vport_del(p);
214 }
215 
216 /* Must be called with rcu_read_lock. */
217 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
218 {
219 	const struct vport *p = OVS_CB(skb)->input_vport;
220 	struct datapath *dp = p->dp;
221 	struct sw_flow *flow;
222 	struct sw_flow_actions *sf_acts;
223 	struct dp_stats_percpu *stats;
224 	u64 *stats_counter;
225 	u32 n_mask_hit;
226 	int error;
227 
228 	stats = this_cpu_ptr(dp->stats_percpu);
229 
230 	/* Look up flow. */
231 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
232 					 &n_mask_hit);
233 	if (unlikely(!flow)) {
234 		struct dp_upcall_info upcall;
235 
236 		memset(&upcall, 0, sizeof(upcall));
237 		upcall.cmd = OVS_PACKET_CMD_MISS;
238 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
239 		upcall.mru = OVS_CB(skb)->mru;
240 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
241 		if (unlikely(error))
242 			kfree_skb(skb);
243 		else
244 			consume_skb(skb);
245 		stats_counter = &stats->n_missed;
246 		goto out;
247 	}
248 
249 	ovs_flow_stats_update(flow, key->tp.flags, skb);
250 	sf_acts = rcu_dereference(flow->sf_acts);
251 	error = ovs_execute_actions(dp, skb, sf_acts, key);
252 	if (unlikely(error))
253 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
254 							ovs_dp_name(dp), error);
255 
256 	stats_counter = &stats->n_hit;
257 
258 out:
259 	/* Update datapath statistics. */
260 	u64_stats_update_begin(&stats->syncp);
261 	(*stats_counter)++;
262 	stats->n_mask_hit += n_mask_hit;
263 	u64_stats_update_end(&stats->syncp);
264 }
265 
266 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
267 		  const struct sw_flow_key *key,
268 		  const struct dp_upcall_info *upcall_info,
269 		  uint32_t cutlen)
270 {
271 	struct dp_stats_percpu *stats;
272 	int err;
273 
274 	if (upcall_info->portid == 0) {
275 		err = -ENOTCONN;
276 		goto err;
277 	}
278 
279 	if (!skb_is_gso(skb))
280 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
281 	else
282 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
283 	if (err)
284 		goto err;
285 
286 	return 0;
287 
288 err:
289 	stats = this_cpu_ptr(dp->stats_percpu);
290 
291 	u64_stats_update_begin(&stats->syncp);
292 	stats->n_lost++;
293 	u64_stats_update_end(&stats->syncp);
294 
295 	return err;
296 }
297 
298 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
299 			     const struct sw_flow_key *key,
300 			     const struct dp_upcall_info *upcall_info,
301 				 uint32_t cutlen)
302 {
303 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
304 	struct sw_flow_key later_key;
305 	struct sk_buff *segs, *nskb;
306 	int err;
307 
308 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
309 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
310 	if (IS_ERR(segs))
311 		return PTR_ERR(segs);
312 	if (segs == NULL)
313 		return -EINVAL;
314 
315 	if (gso_type & SKB_GSO_UDP) {
316 		/* The initial flow key extracted by ovs_flow_key_extract()
317 		 * in this case is for a first fragment, so we need to
318 		 * properly mark later fragments.
319 		 */
320 		later_key = *key;
321 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
322 	}
323 
324 	/* Queue all of the segments. */
325 	skb_list_walk_safe(segs, skb, nskb) {
326 		if (gso_type & SKB_GSO_UDP && skb != segs)
327 			key = &later_key;
328 
329 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
330 		if (err)
331 			break;
332 
333 	}
334 
335 	/* Free all of the segments. */
336 	skb_list_walk_safe(segs, skb, nskb) {
337 		if (err)
338 			kfree_skb(skb);
339 		else
340 			consume_skb(skb);
341 	}
342 	return err;
343 }
344 
345 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
346 			      unsigned int hdrlen, int actions_attrlen)
347 {
348 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
349 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
350 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
351 		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
352 		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
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 	u64 hash;
396 
397 	dp_ifindex = get_dpifindex(dp);
398 	if (!dp_ifindex)
399 		return -ENODEV;
400 
401 	if (skb_vlan_tag_present(skb)) {
402 		nskb = skb_clone(skb, GFP_ATOMIC);
403 		if (!nskb)
404 			return -ENOMEM;
405 
406 		nskb = __vlan_hwaccel_push_inside(nskb);
407 		if (!nskb)
408 			return -ENOMEM;
409 
410 		skb = nskb;
411 	}
412 
413 	if (nla_attr_size(skb->len) > USHRT_MAX) {
414 		err = -EFBIG;
415 		goto out;
416 	}
417 
418 	/* Complete checksum if needed */
419 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
420 	    (err = skb_csum_hwoffload_help(skb, 0)))
421 		goto out;
422 
423 	/* Older versions of OVS user space enforce alignment of the last
424 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
425 	 * padding logic. Only perform zerocopy if padding is not required.
426 	 */
427 	if (dp->user_features & OVS_DP_F_UNALIGNED)
428 		hlen = skb_zerocopy_headlen(skb);
429 	else
430 		hlen = skb->len;
431 
432 	len = upcall_msg_size(upcall_info, hlen - cutlen,
433 			      OVS_CB(skb)->acts_origlen);
434 	user_skb = genlmsg_new(len, GFP_ATOMIC);
435 	if (!user_skb) {
436 		err = -ENOMEM;
437 		goto out;
438 	}
439 
440 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
441 			     0, upcall_info->cmd);
442 	if (!upcall) {
443 		err = -EINVAL;
444 		goto out;
445 	}
446 	upcall->dp_ifindex = dp_ifindex;
447 
448 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
449 	if (err)
450 		goto out;
451 
452 	if (upcall_info->userdata)
453 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
454 			  nla_len(upcall_info->userdata),
455 			  nla_data(upcall_info->userdata));
456 
457 	if (upcall_info->egress_tun_info) {
458 		nla = nla_nest_start_noflag(user_skb,
459 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
460 		if (!nla) {
461 			err = -EMSGSIZE;
462 			goto out;
463 		}
464 		err = ovs_nla_put_tunnel_info(user_skb,
465 					      upcall_info->egress_tun_info);
466 		if (err)
467 			goto out;
468 
469 		nla_nest_end(user_skb, nla);
470 	}
471 
472 	if (upcall_info->actions_len) {
473 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
474 		if (!nla) {
475 			err = -EMSGSIZE;
476 			goto out;
477 		}
478 		err = ovs_nla_put_actions(upcall_info->actions,
479 					  upcall_info->actions_len,
480 					  user_skb);
481 		if (!err)
482 			nla_nest_end(user_skb, nla);
483 		else
484 			nla_nest_cancel(user_skb, nla);
485 	}
486 
487 	/* Add OVS_PACKET_ATTR_MRU */
488 	if (upcall_info->mru &&
489 	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
490 		err = -ENOBUFS;
491 		goto out;
492 	}
493 
494 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
495 	if (cutlen > 0 &&
496 	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
497 		err = -ENOBUFS;
498 		goto out;
499 	}
500 
501 	/* Add OVS_PACKET_ATTR_HASH */
502 	hash = skb_get_hash_raw(skb);
503 	if (skb->sw_hash)
504 		hash |= OVS_PACKET_HASH_SW_BIT;
505 
506 	if (skb->l4_hash)
507 		hash |= OVS_PACKET_HASH_L4_BIT;
508 
509 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
510 		err = -ENOBUFS;
511 		goto out;
512 	}
513 
514 	/* Only reserve room for attribute header, packet data is added
515 	 * in skb_zerocopy() */
516 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
517 		err = -ENOBUFS;
518 		goto out;
519 	}
520 	nla->nla_len = nla_attr_size(skb->len - cutlen);
521 
522 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
523 	if (err)
524 		goto out;
525 
526 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
527 	pad_packet(dp, user_skb);
528 
529 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
530 
531 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
532 	user_skb = NULL;
533 out:
534 	if (err)
535 		skb_tx_error(skb);
536 	kfree_skb(user_skb);
537 	kfree_skb(nskb);
538 	return err;
539 }
540 
541 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
542 {
543 	struct ovs_header *ovs_header = info->userhdr;
544 	struct net *net = sock_net(skb->sk);
545 	struct nlattr **a = info->attrs;
546 	struct sw_flow_actions *acts;
547 	struct sk_buff *packet;
548 	struct sw_flow *flow;
549 	struct sw_flow_actions *sf_acts;
550 	struct datapath *dp;
551 	struct vport *input_vport;
552 	u16 mru = 0;
553 	u64 hash;
554 	int len;
555 	int err;
556 	bool log = !a[OVS_PACKET_ATTR_PROBE];
557 
558 	err = -EINVAL;
559 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
560 	    !a[OVS_PACKET_ATTR_ACTIONS])
561 		goto err;
562 
563 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
564 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
565 	err = -ENOMEM;
566 	if (!packet)
567 		goto err;
568 	skb_reserve(packet, NET_IP_ALIGN);
569 
570 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
571 
572 	/* Set packet's mru */
573 	if (a[OVS_PACKET_ATTR_MRU]) {
574 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
575 		packet->ignore_df = 1;
576 	}
577 	OVS_CB(packet)->mru = mru;
578 
579 	if (a[OVS_PACKET_ATTR_HASH]) {
580 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
581 
582 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
583 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
584 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
585 	}
586 
587 	/* Build an sw_flow for sending this packet. */
588 	flow = ovs_flow_alloc();
589 	err = PTR_ERR(flow);
590 	if (IS_ERR(flow))
591 		goto err_kfree_skb;
592 
593 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
594 					     packet, &flow->key, log);
595 	if (err)
596 		goto err_flow_free;
597 
598 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
599 				   &flow->key, &acts, log);
600 	if (err)
601 		goto err_flow_free;
602 
603 	rcu_assign_pointer(flow->sf_acts, acts);
604 	packet->priority = flow->key.phy.priority;
605 	packet->mark = flow->key.phy.skb_mark;
606 
607 	rcu_read_lock();
608 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
609 	err = -ENODEV;
610 	if (!dp)
611 		goto err_unlock;
612 
613 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
614 	if (!input_vport)
615 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
616 
617 	if (!input_vport)
618 		goto err_unlock;
619 
620 	packet->dev = input_vport->dev;
621 	OVS_CB(packet)->input_vport = input_vport;
622 	sf_acts = rcu_dereference(flow->sf_acts);
623 
624 	local_bh_disable();
625 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
626 	local_bh_enable();
627 	rcu_read_unlock();
628 
629 	ovs_flow_free(flow, false);
630 	return err;
631 
632 err_unlock:
633 	rcu_read_unlock();
634 err_flow_free:
635 	ovs_flow_free(flow, false);
636 err_kfree_skb:
637 	kfree_skb(packet);
638 err:
639 	return err;
640 }
641 
642 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
643 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
644 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
645 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
646 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
647 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
648 };
649 
650 static const struct genl_ops dp_packet_genl_ops[] = {
651 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
652 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
653 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
654 	  .doit = ovs_packet_cmd_execute
655 	}
656 };
657 
658 static struct genl_family dp_packet_genl_family __ro_after_init = {
659 	.hdrsize = sizeof(struct ovs_header),
660 	.name = OVS_PACKET_FAMILY,
661 	.version = OVS_PACKET_VERSION,
662 	.maxattr = OVS_PACKET_ATTR_MAX,
663 	.policy = packet_policy,
664 	.netnsok = true,
665 	.parallel_ops = true,
666 	.ops = dp_packet_genl_ops,
667 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
668 	.module = THIS_MODULE,
669 };
670 
671 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
672 			 struct ovs_dp_megaflow_stats *mega_stats)
673 {
674 	int i;
675 
676 	memset(mega_stats, 0, sizeof(*mega_stats));
677 
678 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
679 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
680 
681 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
682 
683 	for_each_possible_cpu(i) {
684 		const struct dp_stats_percpu *percpu_stats;
685 		struct dp_stats_percpu local_stats;
686 		unsigned int start;
687 
688 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
689 
690 		do {
691 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
692 			local_stats = *percpu_stats;
693 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
694 
695 		stats->n_hit += local_stats.n_hit;
696 		stats->n_missed += local_stats.n_missed;
697 		stats->n_lost += local_stats.n_lost;
698 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
699 	}
700 }
701 
702 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
703 {
704 	return ovs_identifier_is_ufid(sfid) &&
705 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
706 }
707 
708 static bool should_fill_mask(uint32_t ufid_flags)
709 {
710 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
711 }
712 
713 static bool should_fill_actions(uint32_t ufid_flags)
714 {
715 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
716 }
717 
718 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
719 				    const struct sw_flow_id *sfid,
720 				    uint32_t ufid_flags)
721 {
722 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
723 
724 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
725 	 * see ovs_nla_put_identifier()
726 	 */
727 	if (sfid && ovs_identifier_is_ufid(sfid))
728 		len += nla_total_size(sfid->ufid_len);
729 	else
730 		len += nla_total_size(ovs_key_attr_size());
731 
732 	/* OVS_FLOW_ATTR_KEY */
733 	if (!sfid || should_fill_key(sfid, ufid_flags))
734 		len += nla_total_size(ovs_key_attr_size());
735 
736 	/* OVS_FLOW_ATTR_MASK */
737 	if (should_fill_mask(ufid_flags))
738 		len += nla_total_size(ovs_key_attr_size());
739 
740 	/* OVS_FLOW_ATTR_ACTIONS */
741 	if (should_fill_actions(ufid_flags))
742 		len += nla_total_size(acts->orig_len);
743 
744 	return len
745 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
746 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
747 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
748 }
749 
750 /* Called with ovs_mutex or RCU read lock. */
751 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
752 				   struct sk_buff *skb)
753 {
754 	struct ovs_flow_stats stats;
755 	__be16 tcp_flags;
756 	unsigned long used;
757 
758 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
759 
760 	if (used &&
761 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
762 			      OVS_FLOW_ATTR_PAD))
763 		return -EMSGSIZE;
764 
765 	if (stats.n_packets &&
766 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
767 			  sizeof(struct ovs_flow_stats), &stats,
768 			  OVS_FLOW_ATTR_PAD))
769 		return -EMSGSIZE;
770 
771 	if ((u8)ntohs(tcp_flags) &&
772 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
773 		return -EMSGSIZE;
774 
775 	return 0;
776 }
777 
778 /* Called with ovs_mutex or RCU read lock. */
779 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
780 				     struct sk_buff *skb, int skb_orig_len)
781 {
782 	struct nlattr *start;
783 	int err;
784 
785 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
786 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
787 	 * Netlink but individual action lists can be longer than
788 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
789 	 * The userspace caller can always fetch the actions separately if it
790 	 * really wants them.  (Most userspace callers in fact don't care.)
791 	 *
792 	 * This can only fail for dump operations because the skb is always
793 	 * properly sized for single flows.
794 	 */
795 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
796 	if (start) {
797 		const struct sw_flow_actions *sf_acts;
798 
799 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
800 		err = ovs_nla_put_actions(sf_acts->actions,
801 					  sf_acts->actions_len, skb);
802 
803 		if (!err)
804 			nla_nest_end(skb, start);
805 		else {
806 			if (skb_orig_len)
807 				return err;
808 
809 			nla_nest_cancel(skb, start);
810 		}
811 	} else if (skb_orig_len) {
812 		return -EMSGSIZE;
813 	}
814 
815 	return 0;
816 }
817 
818 /* Called with ovs_mutex or RCU read lock. */
819 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
820 				  struct sk_buff *skb, u32 portid,
821 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
822 {
823 	const int skb_orig_len = skb->len;
824 	struct ovs_header *ovs_header;
825 	int err;
826 
827 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
828 				 flags, cmd);
829 	if (!ovs_header)
830 		return -EMSGSIZE;
831 
832 	ovs_header->dp_ifindex = dp_ifindex;
833 
834 	err = ovs_nla_put_identifier(flow, skb);
835 	if (err)
836 		goto error;
837 
838 	if (should_fill_key(&flow->id, ufid_flags)) {
839 		err = ovs_nla_put_masked_key(flow, skb);
840 		if (err)
841 			goto error;
842 	}
843 
844 	if (should_fill_mask(ufid_flags)) {
845 		err = ovs_nla_put_mask(flow, skb);
846 		if (err)
847 			goto error;
848 	}
849 
850 	err = ovs_flow_cmd_fill_stats(flow, skb);
851 	if (err)
852 		goto error;
853 
854 	if (should_fill_actions(ufid_flags)) {
855 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
856 		if (err)
857 			goto error;
858 	}
859 
860 	genlmsg_end(skb, ovs_header);
861 	return 0;
862 
863 error:
864 	genlmsg_cancel(skb, ovs_header);
865 	return err;
866 }
867 
868 /* May not be called with RCU read lock. */
869 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
870 					       const struct sw_flow_id *sfid,
871 					       struct genl_info *info,
872 					       bool always,
873 					       uint32_t ufid_flags)
874 {
875 	struct sk_buff *skb;
876 	size_t len;
877 
878 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
879 		return NULL;
880 
881 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
882 	skb = genlmsg_new(len, GFP_KERNEL);
883 	if (!skb)
884 		return ERR_PTR(-ENOMEM);
885 
886 	return skb;
887 }
888 
889 /* Called with ovs_mutex. */
890 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
891 					       int dp_ifindex,
892 					       struct genl_info *info, u8 cmd,
893 					       bool always, u32 ufid_flags)
894 {
895 	struct sk_buff *skb;
896 	int retval;
897 
898 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
899 				      &flow->id, info, always, ufid_flags);
900 	if (IS_ERR_OR_NULL(skb))
901 		return skb;
902 
903 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
904 					info->snd_portid, info->snd_seq, 0,
905 					cmd, ufid_flags);
906 	if (WARN_ON_ONCE(retval < 0)) {
907 		kfree_skb(skb);
908 		skb = ERR_PTR(retval);
909 	}
910 	return skb;
911 }
912 
913 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
914 {
915 	struct net *net = sock_net(skb->sk);
916 	struct nlattr **a = info->attrs;
917 	struct ovs_header *ovs_header = info->userhdr;
918 	struct sw_flow *flow = NULL, *new_flow;
919 	struct sw_flow_mask mask;
920 	struct sk_buff *reply;
921 	struct datapath *dp;
922 	struct sw_flow_actions *acts;
923 	struct sw_flow_match match;
924 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
925 	int error;
926 	bool log = !a[OVS_FLOW_ATTR_PROBE];
927 
928 	/* Must have key and actions. */
929 	error = -EINVAL;
930 	if (!a[OVS_FLOW_ATTR_KEY]) {
931 		OVS_NLERR(log, "Flow key attr not present in new flow.");
932 		goto error;
933 	}
934 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
935 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
936 		goto error;
937 	}
938 
939 	/* Most of the time we need to allocate a new flow, do it before
940 	 * locking.
941 	 */
942 	new_flow = ovs_flow_alloc();
943 	if (IS_ERR(new_flow)) {
944 		error = PTR_ERR(new_flow);
945 		goto error;
946 	}
947 
948 	/* Extract key. */
949 	ovs_match_init(&match, &new_flow->key, false, &mask);
950 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
951 				  a[OVS_FLOW_ATTR_MASK], log);
952 	if (error)
953 		goto err_kfree_flow;
954 
955 	/* Extract flow identifier. */
956 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
957 				       &new_flow->key, log);
958 	if (error)
959 		goto err_kfree_flow;
960 
961 	/* unmasked key is needed to match when ufid is not used. */
962 	if (ovs_identifier_is_key(&new_flow->id))
963 		match.key = new_flow->id.unmasked_key;
964 
965 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
966 
967 	/* Validate actions. */
968 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
969 				     &new_flow->key, &acts, log);
970 	if (error) {
971 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
972 		goto err_kfree_flow;
973 	}
974 
975 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
976 					ufid_flags);
977 	if (IS_ERR(reply)) {
978 		error = PTR_ERR(reply);
979 		goto err_kfree_acts;
980 	}
981 
982 	ovs_lock();
983 	dp = get_dp(net, ovs_header->dp_ifindex);
984 	if (unlikely(!dp)) {
985 		error = -ENODEV;
986 		goto err_unlock_ovs;
987 	}
988 
989 	/* Check if this is a duplicate flow */
990 	if (ovs_identifier_is_ufid(&new_flow->id))
991 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
992 	if (!flow)
993 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
994 	if (likely(!flow)) {
995 		rcu_assign_pointer(new_flow->sf_acts, acts);
996 
997 		/* Put flow in bucket. */
998 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
999 		if (unlikely(error)) {
1000 			acts = NULL;
1001 			goto err_unlock_ovs;
1002 		}
1003 
1004 		if (unlikely(reply)) {
1005 			error = ovs_flow_cmd_fill_info(new_flow,
1006 						       ovs_header->dp_ifindex,
1007 						       reply, info->snd_portid,
1008 						       info->snd_seq, 0,
1009 						       OVS_FLOW_CMD_NEW,
1010 						       ufid_flags);
1011 			BUG_ON(error < 0);
1012 		}
1013 		ovs_unlock();
1014 	} else {
1015 		struct sw_flow_actions *old_acts;
1016 
1017 		/* Bail out if we're not allowed to modify an existing flow.
1018 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1019 		 * because Generic Netlink treats the latter as a dump
1020 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1021 		 * gets fixed.
1022 		 */
1023 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1024 							 | NLM_F_EXCL))) {
1025 			error = -EEXIST;
1026 			goto err_unlock_ovs;
1027 		}
1028 		/* The flow identifier has to be the same for flow updates.
1029 		 * Look for any overlapping flow.
1030 		 */
1031 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1032 			if (ovs_identifier_is_key(&flow->id))
1033 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1034 								 &match);
1035 			else /* UFID matches but key is different */
1036 				flow = NULL;
1037 			if (!flow) {
1038 				error = -ENOENT;
1039 				goto err_unlock_ovs;
1040 			}
1041 		}
1042 		/* Update actions. */
1043 		old_acts = ovsl_dereference(flow->sf_acts);
1044 		rcu_assign_pointer(flow->sf_acts, acts);
1045 
1046 		if (unlikely(reply)) {
1047 			error = ovs_flow_cmd_fill_info(flow,
1048 						       ovs_header->dp_ifindex,
1049 						       reply, info->snd_portid,
1050 						       info->snd_seq, 0,
1051 						       OVS_FLOW_CMD_NEW,
1052 						       ufid_flags);
1053 			BUG_ON(error < 0);
1054 		}
1055 		ovs_unlock();
1056 
1057 		ovs_nla_free_flow_actions_rcu(old_acts);
1058 		ovs_flow_free(new_flow, false);
1059 	}
1060 
1061 	if (reply)
1062 		ovs_notify(&dp_flow_genl_family, reply, info);
1063 	return 0;
1064 
1065 err_unlock_ovs:
1066 	ovs_unlock();
1067 	kfree_skb(reply);
1068 err_kfree_acts:
1069 	ovs_nla_free_flow_actions(acts);
1070 err_kfree_flow:
1071 	ovs_flow_free(new_flow, false);
1072 error:
1073 	return error;
1074 }
1075 
1076 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1077 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1078 						const struct nlattr *a,
1079 						const struct sw_flow_key *key,
1080 						const struct sw_flow_mask *mask,
1081 						bool log)
1082 {
1083 	struct sw_flow_actions *acts;
1084 	struct sw_flow_key masked_key;
1085 	int error;
1086 
1087 	ovs_flow_mask_key(&masked_key, key, true, mask);
1088 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1089 	if (error) {
1090 		OVS_NLERR(log,
1091 			  "Actions may not be safe on all matching packets");
1092 		return ERR_PTR(error);
1093 	}
1094 
1095 	return acts;
1096 }
1097 
1098 /* Factor out match-init and action-copy to avoid
1099  * "Wframe-larger-than=1024" warning. Because mask is only
1100  * used to get actions, we new a function to save some
1101  * stack space.
1102  *
1103  * If there are not key and action attrs, we return 0
1104  * directly. In the case, the caller will also not use the
1105  * match as before. If there is action attr, we try to get
1106  * actions and save them to *acts. Before returning from
1107  * the function, we reset the match->mask pointer. Because
1108  * we should not to return match object with dangling reference
1109  * to mask.
1110  * */
1111 static noinline_for_stack int
1112 ovs_nla_init_match_and_action(struct net *net,
1113 			      struct sw_flow_match *match,
1114 			      struct sw_flow_key *key,
1115 			      struct nlattr **a,
1116 			      struct sw_flow_actions **acts,
1117 			      bool log)
1118 {
1119 	struct sw_flow_mask mask;
1120 	int error = 0;
1121 
1122 	if (a[OVS_FLOW_ATTR_KEY]) {
1123 		ovs_match_init(match, key, true, &mask);
1124 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1125 					  a[OVS_FLOW_ATTR_MASK], log);
1126 		if (error)
1127 			goto error;
1128 	}
1129 
1130 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1131 		if (!a[OVS_FLOW_ATTR_KEY]) {
1132 			OVS_NLERR(log,
1133 				  "Flow key attribute not present in set flow.");
1134 			error = -EINVAL;
1135 			goto error;
1136 		}
1137 
1138 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1139 					 &mask, log);
1140 		if (IS_ERR(*acts)) {
1141 			error = PTR_ERR(*acts);
1142 			goto error;
1143 		}
1144 	}
1145 
1146 	/* On success, error is 0. */
1147 error:
1148 	match->mask = NULL;
1149 	return error;
1150 }
1151 
1152 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1153 {
1154 	struct net *net = sock_net(skb->sk);
1155 	struct nlattr **a = info->attrs;
1156 	struct ovs_header *ovs_header = info->userhdr;
1157 	struct sw_flow_key key;
1158 	struct sw_flow *flow;
1159 	struct sk_buff *reply = NULL;
1160 	struct datapath *dp;
1161 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1162 	struct sw_flow_match match;
1163 	struct sw_flow_id sfid;
1164 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1165 	int error = 0;
1166 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1167 	bool ufid_present;
1168 
1169 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1170 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1171 		OVS_NLERR(log,
1172 			  "Flow set message rejected, Key attribute missing.");
1173 		return -EINVAL;
1174 	}
1175 
1176 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1177 					      &acts, log);
1178 	if (error)
1179 		goto error;
1180 
1181 	if (acts) {
1182 		/* Can allocate before locking if have acts. */
1183 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1184 						ufid_flags);
1185 		if (IS_ERR(reply)) {
1186 			error = PTR_ERR(reply);
1187 			goto err_kfree_acts;
1188 		}
1189 	}
1190 
1191 	ovs_lock();
1192 	dp = get_dp(net, ovs_header->dp_ifindex);
1193 	if (unlikely(!dp)) {
1194 		error = -ENODEV;
1195 		goto err_unlock_ovs;
1196 	}
1197 	/* Check that the flow exists. */
1198 	if (ufid_present)
1199 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1200 	else
1201 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1202 	if (unlikely(!flow)) {
1203 		error = -ENOENT;
1204 		goto err_unlock_ovs;
1205 	}
1206 
1207 	/* Update actions, if present. */
1208 	if (likely(acts)) {
1209 		old_acts = ovsl_dereference(flow->sf_acts);
1210 		rcu_assign_pointer(flow->sf_acts, acts);
1211 
1212 		if (unlikely(reply)) {
1213 			error = ovs_flow_cmd_fill_info(flow,
1214 						       ovs_header->dp_ifindex,
1215 						       reply, info->snd_portid,
1216 						       info->snd_seq, 0,
1217 						       OVS_FLOW_CMD_SET,
1218 						       ufid_flags);
1219 			BUG_ON(error < 0);
1220 		}
1221 	} else {
1222 		/* Could not alloc without acts before locking. */
1223 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1224 						info, OVS_FLOW_CMD_SET, false,
1225 						ufid_flags);
1226 
1227 		if (IS_ERR(reply)) {
1228 			error = PTR_ERR(reply);
1229 			goto err_unlock_ovs;
1230 		}
1231 	}
1232 
1233 	/* Clear stats. */
1234 	if (a[OVS_FLOW_ATTR_CLEAR])
1235 		ovs_flow_stats_clear(flow);
1236 	ovs_unlock();
1237 
1238 	if (reply)
1239 		ovs_notify(&dp_flow_genl_family, reply, info);
1240 	if (old_acts)
1241 		ovs_nla_free_flow_actions_rcu(old_acts);
1242 
1243 	return 0;
1244 
1245 err_unlock_ovs:
1246 	ovs_unlock();
1247 	kfree_skb(reply);
1248 err_kfree_acts:
1249 	ovs_nla_free_flow_actions(acts);
1250 error:
1251 	return error;
1252 }
1253 
1254 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1255 {
1256 	struct nlattr **a = info->attrs;
1257 	struct ovs_header *ovs_header = info->userhdr;
1258 	struct net *net = sock_net(skb->sk);
1259 	struct sw_flow_key key;
1260 	struct sk_buff *reply;
1261 	struct sw_flow *flow;
1262 	struct datapath *dp;
1263 	struct sw_flow_match match;
1264 	struct sw_flow_id ufid;
1265 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1266 	int err = 0;
1267 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1268 	bool ufid_present;
1269 
1270 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1271 	if (a[OVS_FLOW_ATTR_KEY]) {
1272 		ovs_match_init(&match, &key, true, NULL);
1273 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1274 					log);
1275 	} else if (!ufid_present) {
1276 		OVS_NLERR(log,
1277 			  "Flow get message rejected, Key attribute missing.");
1278 		err = -EINVAL;
1279 	}
1280 	if (err)
1281 		return err;
1282 
1283 	ovs_lock();
1284 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1285 	if (!dp) {
1286 		err = -ENODEV;
1287 		goto unlock;
1288 	}
1289 
1290 	if (ufid_present)
1291 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1292 	else
1293 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1294 	if (!flow) {
1295 		err = -ENOENT;
1296 		goto unlock;
1297 	}
1298 
1299 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1300 					OVS_FLOW_CMD_GET, true, ufid_flags);
1301 	if (IS_ERR(reply)) {
1302 		err = PTR_ERR(reply);
1303 		goto unlock;
1304 	}
1305 
1306 	ovs_unlock();
1307 	return genlmsg_reply(reply, info);
1308 unlock:
1309 	ovs_unlock();
1310 	return err;
1311 }
1312 
1313 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1314 {
1315 	struct nlattr **a = info->attrs;
1316 	struct ovs_header *ovs_header = info->userhdr;
1317 	struct net *net = sock_net(skb->sk);
1318 	struct sw_flow_key key;
1319 	struct sk_buff *reply;
1320 	struct sw_flow *flow = NULL;
1321 	struct datapath *dp;
1322 	struct sw_flow_match match;
1323 	struct sw_flow_id ufid;
1324 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1325 	int err;
1326 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1327 	bool ufid_present;
1328 
1329 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1330 	if (a[OVS_FLOW_ATTR_KEY]) {
1331 		ovs_match_init(&match, &key, true, NULL);
1332 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1333 					NULL, log);
1334 		if (unlikely(err))
1335 			return err;
1336 	}
1337 
1338 	ovs_lock();
1339 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1340 	if (unlikely(!dp)) {
1341 		err = -ENODEV;
1342 		goto unlock;
1343 	}
1344 
1345 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1346 		err = ovs_flow_tbl_flush(&dp->table);
1347 		goto unlock;
1348 	}
1349 
1350 	if (ufid_present)
1351 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1352 	else
1353 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1354 	if (unlikely(!flow)) {
1355 		err = -ENOENT;
1356 		goto unlock;
1357 	}
1358 
1359 	ovs_flow_tbl_remove(&dp->table, flow);
1360 	ovs_unlock();
1361 
1362 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1363 					&flow->id, info, false, ufid_flags);
1364 	if (likely(reply)) {
1365 		if (!IS_ERR(reply)) {
1366 			rcu_read_lock();	/*To keep RCU checker happy. */
1367 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1368 						     reply, info->snd_portid,
1369 						     info->snd_seq, 0,
1370 						     OVS_FLOW_CMD_DEL,
1371 						     ufid_flags);
1372 			rcu_read_unlock();
1373 			if (WARN_ON_ONCE(err < 0)) {
1374 				kfree_skb(reply);
1375 				goto out_free;
1376 			}
1377 
1378 			ovs_notify(&dp_flow_genl_family, reply, info);
1379 		} else {
1380 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1381 		}
1382 	}
1383 
1384 out_free:
1385 	ovs_flow_free(flow, true);
1386 	return 0;
1387 unlock:
1388 	ovs_unlock();
1389 	return err;
1390 }
1391 
1392 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1393 {
1394 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1395 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1396 	struct table_instance *ti;
1397 	struct datapath *dp;
1398 	u32 ufid_flags;
1399 	int err;
1400 
1401 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1402 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1403 	if (err)
1404 		return err;
1405 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1406 
1407 	rcu_read_lock();
1408 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1409 	if (!dp) {
1410 		rcu_read_unlock();
1411 		return -ENODEV;
1412 	}
1413 
1414 	ti = rcu_dereference(dp->table.ti);
1415 	for (;;) {
1416 		struct sw_flow *flow;
1417 		u32 bucket, obj;
1418 
1419 		bucket = cb->args[0];
1420 		obj = cb->args[1];
1421 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1422 		if (!flow)
1423 			break;
1424 
1425 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1426 					   NETLINK_CB(cb->skb).portid,
1427 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1428 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1429 			break;
1430 
1431 		cb->args[0] = bucket;
1432 		cb->args[1] = obj;
1433 	}
1434 	rcu_read_unlock();
1435 	return skb->len;
1436 }
1437 
1438 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1439 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1440 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1441 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1442 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1443 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1444 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1445 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1446 };
1447 
1448 static const struct genl_ops dp_flow_genl_ops[] = {
1449 	{ .cmd = OVS_FLOW_CMD_NEW,
1450 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1451 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1452 	  .doit = ovs_flow_cmd_new
1453 	},
1454 	{ .cmd = OVS_FLOW_CMD_DEL,
1455 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1456 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1457 	  .doit = ovs_flow_cmd_del
1458 	},
1459 	{ .cmd = OVS_FLOW_CMD_GET,
1460 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1461 	  .flags = 0,		    /* OK for unprivileged users. */
1462 	  .doit = ovs_flow_cmd_get,
1463 	  .dumpit = ovs_flow_cmd_dump
1464 	},
1465 	{ .cmd = OVS_FLOW_CMD_SET,
1466 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1467 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1468 	  .doit = ovs_flow_cmd_set,
1469 	},
1470 };
1471 
1472 static struct genl_family dp_flow_genl_family __ro_after_init = {
1473 	.hdrsize = sizeof(struct ovs_header),
1474 	.name = OVS_FLOW_FAMILY,
1475 	.version = OVS_FLOW_VERSION,
1476 	.maxattr = OVS_FLOW_ATTR_MAX,
1477 	.policy = flow_policy,
1478 	.netnsok = true,
1479 	.parallel_ops = true,
1480 	.ops = dp_flow_genl_ops,
1481 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1482 	.mcgrps = &ovs_dp_flow_multicast_group,
1483 	.n_mcgrps = 1,
1484 	.module = THIS_MODULE,
1485 };
1486 
1487 static size_t ovs_dp_cmd_msg_size(void)
1488 {
1489 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1490 
1491 	msgsize += nla_total_size(IFNAMSIZ);
1492 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1493 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1494 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1495 
1496 	return msgsize;
1497 }
1498 
1499 /* Called with ovs_mutex. */
1500 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1501 				u32 portid, u32 seq, u32 flags, u8 cmd)
1502 {
1503 	struct ovs_header *ovs_header;
1504 	struct ovs_dp_stats dp_stats;
1505 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1506 	int err;
1507 
1508 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1509 				   flags, cmd);
1510 	if (!ovs_header)
1511 		goto error;
1512 
1513 	ovs_header->dp_ifindex = get_dpifindex(dp);
1514 
1515 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1516 	if (err)
1517 		goto nla_put_failure;
1518 
1519 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1520 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1521 			  &dp_stats, OVS_DP_ATTR_PAD))
1522 		goto nla_put_failure;
1523 
1524 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1525 			  sizeof(struct ovs_dp_megaflow_stats),
1526 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1527 		goto nla_put_failure;
1528 
1529 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1530 		goto nla_put_failure;
1531 
1532 	genlmsg_end(skb, ovs_header);
1533 	return 0;
1534 
1535 nla_put_failure:
1536 	genlmsg_cancel(skb, ovs_header);
1537 error:
1538 	return -EMSGSIZE;
1539 }
1540 
1541 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1542 {
1543 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1544 }
1545 
1546 /* Called with rcu_read_lock or ovs_mutex. */
1547 static struct datapath *lookup_datapath(struct net *net,
1548 					const struct ovs_header *ovs_header,
1549 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1550 {
1551 	struct datapath *dp;
1552 
1553 	if (!a[OVS_DP_ATTR_NAME])
1554 		dp = get_dp(net, ovs_header->dp_ifindex);
1555 	else {
1556 		struct vport *vport;
1557 
1558 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1559 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1560 	}
1561 	return dp ? dp : ERR_PTR(-ENODEV);
1562 }
1563 
1564 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1565 {
1566 	struct datapath *dp;
1567 
1568 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1569 	if (IS_ERR(dp))
1570 		return;
1571 
1572 	WARN(dp->user_features, "Dropping previously announced user features\n");
1573 	dp->user_features = 0;
1574 }
1575 
1576 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1577 
1578 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1579 {
1580 	u32 user_features = 0;
1581 
1582 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1583 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1584 
1585 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1586 				      OVS_DP_F_UNALIGNED |
1587 				      OVS_DP_F_TC_RECIRC_SHARING))
1588 			return -EOPNOTSUPP;
1589 
1590 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1591 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1592 			return -EOPNOTSUPP;
1593 #endif
1594 	}
1595 
1596 	dp->user_features = user_features;
1597 
1598 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1599 		static_branch_enable(&tc_recirc_sharing_support);
1600 	else
1601 		static_branch_disable(&tc_recirc_sharing_support);
1602 
1603 	return 0;
1604 }
1605 
1606 static int ovs_dp_stats_init(struct datapath *dp)
1607 {
1608 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1609 	if (!dp->stats_percpu)
1610 		return -ENOMEM;
1611 
1612 	return 0;
1613 }
1614 
1615 static int ovs_dp_vport_init(struct datapath *dp)
1616 {
1617 	int i;
1618 
1619 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1620 				  sizeof(struct hlist_head),
1621 				  GFP_KERNEL);
1622 	if (!dp->ports)
1623 		return -ENOMEM;
1624 
1625 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1626 		INIT_HLIST_HEAD(&dp->ports[i]);
1627 
1628 	return 0;
1629 }
1630 
1631 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1632 {
1633 	struct nlattr **a = info->attrs;
1634 	struct vport_parms parms;
1635 	struct sk_buff *reply;
1636 	struct datapath *dp;
1637 	struct vport *vport;
1638 	struct ovs_net *ovs_net;
1639 	int err;
1640 
1641 	err = -EINVAL;
1642 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1643 		goto err;
1644 
1645 	reply = ovs_dp_cmd_alloc_info();
1646 	if (!reply)
1647 		return -ENOMEM;
1648 
1649 	err = -ENOMEM;
1650 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1651 	if (dp == NULL)
1652 		goto err_destroy_reply;
1653 
1654 	ovs_dp_set_net(dp, sock_net(skb->sk));
1655 
1656 	/* Allocate table. */
1657 	err = ovs_flow_tbl_init(&dp->table);
1658 	if (err)
1659 		goto err_destroy_dp;
1660 
1661 	err = ovs_dp_stats_init(dp);
1662 	if (err)
1663 		goto err_destroy_table;
1664 
1665 	err = ovs_dp_vport_init(dp);
1666 	if (err)
1667 		goto err_destroy_stats;
1668 
1669 	err = ovs_meters_init(dp);
1670 	if (err)
1671 		goto err_destroy_ports;
1672 
1673 	/* Set up our datapath device. */
1674 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1675 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1676 	parms.options = NULL;
1677 	parms.dp = dp;
1678 	parms.port_no = OVSP_LOCAL;
1679 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1680 
1681 	err = ovs_dp_change(dp, a);
1682 	if (err)
1683 		goto err_destroy_meters;
1684 
1685 	/* So far only local changes have been made, now need the lock. */
1686 	ovs_lock();
1687 
1688 	vport = new_vport(&parms);
1689 	if (IS_ERR(vport)) {
1690 		err = PTR_ERR(vport);
1691 		if (err == -EBUSY)
1692 			err = -EEXIST;
1693 
1694 		if (err == -EEXIST) {
1695 			/* An outdated user space instance that does not understand
1696 			 * the concept of user_features has attempted to create a new
1697 			 * datapath and is likely to reuse it. Drop all user features.
1698 			 */
1699 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1700 				ovs_dp_reset_user_features(skb, info);
1701 		}
1702 
1703 		ovs_unlock();
1704 		goto err_destroy_meters;
1705 	}
1706 
1707 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1708 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1709 	BUG_ON(err < 0);
1710 
1711 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1712 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1713 
1714 	ovs_unlock();
1715 
1716 	ovs_notify(&dp_datapath_genl_family, reply, info);
1717 	return 0;
1718 
1719 err_destroy_meters:
1720 	ovs_meters_exit(dp);
1721 err_destroy_ports:
1722 	kfree(dp->ports);
1723 err_destroy_stats:
1724 	free_percpu(dp->stats_percpu);
1725 err_destroy_table:
1726 	ovs_flow_tbl_destroy(&dp->table);
1727 err_destroy_dp:
1728 	kfree(dp);
1729 err_destroy_reply:
1730 	kfree_skb(reply);
1731 err:
1732 	return err;
1733 }
1734 
1735 /* Called with ovs_mutex. */
1736 static void __dp_destroy(struct datapath *dp)
1737 {
1738 	int i;
1739 
1740 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1741 		struct vport *vport;
1742 		struct hlist_node *n;
1743 
1744 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1745 			if (vport->port_no != OVSP_LOCAL)
1746 				ovs_dp_detach_port(vport);
1747 	}
1748 
1749 	list_del_rcu(&dp->list_node);
1750 
1751 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1752 	 * all ports in datapath are destroyed first before freeing datapath.
1753 	 */
1754 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1755 
1756 	/* RCU destroy the flow table */
1757 	call_rcu(&dp->rcu, destroy_dp_rcu);
1758 }
1759 
1760 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1761 {
1762 	struct sk_buff *reply;
1763 	struct datapath *dp;
1764 	int err;
1765 
1766 	reply = ovs_dp_cmd_alloc_info();
1767 	if (!reply)
1768 		return -ENOMEM;
1769 
1770 	ovs_lock();
1771 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1772 	err = PTR_ERR(dp);
1773 	if (IS_ERR(dp))
1774 		goto err_unlock_free;
1775 
1776 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1777 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1778 	BUG_ON(err < 0);
1779 
1780 	__dp_destroy(dp);
1781 	ovs_unlock();
1782 
1783 	ovs_notify(&dp_datapath_genl_family, reply, info);
1784 
1785 	return 0;
1786 
1787 err_unlock_free:
1788 	ovs_unlock();
1789 	kfree_skb(reply);
1790 	return err;
1791 }
1792 
1793 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1794 {
1795 	struct sk_buff *reply;
1796 	struct datapath *dp;
1797 	int err;
1798 
1799 	reply = ovs_dp_cmd_alloc_info();
1800 	if (!reply)
1801 		return -ENOMEM;
1802 
1803 	ovs_lock();
1804 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1805 	err = PTR_ERR(dp);
1806 	if (IS_ERR(dp))
1807 		goto err_unlock_free;
1808 
1809 	err = ovs_dp_change(dp, info->attrs);
1810 	if (err)
1811 		goto err_unlock_free;
1812 
1813 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1814 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1815 	BUG_ON(err < 0);
1816 
1817 	ovs_unlock();
1818 	ovs_notify(&dp_datapath_genl_family, reply, info);
1819 
1820 	return 0;
1821 
1822 err_unlock_free:
1823 	ovs_unlock();
1824 	kfree_skb(reply);
1825 	return err;
1826 }
1827 
1828 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1829 {
1830 	struct sk_buff *reply;
1831 	struct datapath *dp;
1832 	int err;
1833 
1834 	reply = ovs_dp_cmd_alloc_info();
1835 	if (!reply)
1836 		return -ENOMEM;
1837 
1838 	ovs_lock();
1839 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1840 	if (IS_ERR(dp)) {
1841 		err = PTR_ERR(dp);
1842 		goto err_unlock_free;
1843 	}
1844 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1845 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1846 	BUG_ON(err < 0);
1847 	ovs_unlock();
1848 
1849 	return genlmsg_reply(reply, info);
1850 
1851 err_unlock_free:
1852 	ovs_unlock();
1853 	kfree_skb(reply);
1854 	return err;
1855 }
1856 
1857 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1858 {
1859 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1860 	struct datapath *dp;
1861 	int skip = cb->args[0];
1862 	int i = 0;
1863 
1864 	ovs_lock();
1865 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1866 		if (i >= skip &&
1867 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1868 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1869 					 OVS_DP_CMD_GET) < 0)
1870 			break;
1871 		i++;
1872 	}
1873 	ovs_unlock();
1874 
1875 	cb->args[0] = i;
1876 
1877 	return skb->len;
1878 }
1879 
1880 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1881 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1882 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1883 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1884 };
1885 
1886 static const struct genl_ops dp_datapath_genl_ops[] = {
1887 	{ .cmd = OVS_DP_CMD_NEW,
1888 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1889 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1890 	  .doit = ovs_dp_cmd_new
1891 	},
1892 	{ .cmd = OVS_DP_CMD_DEL,
1893 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1894 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1895 	  .doit = ovs_dp_cmd_del
1896 	},
1897 	{ .cmd = OVS_DP_CMD_GET,
1898 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1899 	  .flags = 0,		    /* OK for unprivileged users. */
1900 	  .doit = ovs_dp_cmd_get,
1901 	  .dumpit = ovs_dp_cmd_dump
1902 	},
1903 	{ .cmd = OVS_DP_CMD_SET,
1904 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1905 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1906 	  .doit = ovs_dp_cmd_set,
1907 	},
1908 };
1909 
1910 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1911 	.hdrsize = sizeof(struct ovs_header),
1912 	.name = OVS_DATAPATH_FAMILY,
1913 	.version = OVS_DATAPATH_VERSION,
1914 	.maxattr = OVS_DP_ATTR_MAX,
1915 	.policy = datapath_policy,
1916 	.netnsok = true,
1917 	.parallel_ops = true,
1918 	.ops = dp_datapath_genl_ops,
1919 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1920 	.mcgrps = &ovs_dp_datapath_multicast_group,
1921 	.n_mcgrps = 1,
1922 	.module = THIS_MODULE,
1923 };
1924 
1925 /* Called with ovs_mutex or RCU read lock. */
1926 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1927 				   struct net *net, u32 portid, u32 seq,
1928 				   u32 flags, u8 cmd, gfp_t gfp)
1929 {
1930 	struct ovs_header *ovs_header;
1931 	struct ovs_vport_stats vport_stats;
1932 	int err;
1933 
1934 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1935 				 flags, cmd);
1936 	if (!ovs_header)
1937 		return -EMSGSIZE;
1938 
1939 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1940 
1941 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1942 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1943 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1944 			   ovs_vport_name(vport)) ||
1945 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1946 		goto nla_put_failure;
1947 
1948 	if (!net_eq(net, dev_net(vport->dev))) {
1949 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1950 
1951 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1952 			goto nla_put_failure;
1953 	}
1954 
1955 	ovs_vport_get_stats(vport, &vport_stats);
1956 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1957 			  sizeof(struct ovs_vport_stats), &vport_stats,
1958 			  OVS_VPORT_ATTR_PAD))
1959 		goto nla_put_failure;
1960 
1961 	if (ovs_vport_get_upcall_portids(vport, skb))
1962 		goto nla_put_failure;
1963 
1964 	err = ovs_vport_get_options(vport, skb);
1965 	if (err == -EMSGSIZE)
1966 		goto error;
1967 
1968 	genlmsg_end(skb, ovs_header);
1969 	return 0;
1970 
1971 nla_put_failure:
1972 	err = -EMSGSIZE;
1973 error:
1974 	genlmsg_cancel(skb, ovs_header);
1975 	return err;
1976 }
1977 
1978 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1979 {
1980 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1981 }
1982 
1983 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1984 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1985 					 u32 portid, u32 seq, u8 cmd)
1986 {
1987 	struct sk_buff *skb;
1988 	int retval;
1989 
1990 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1991 	if (!skb)
1992 		return ERR_PTR(-ENOMEM);
1993 
1994 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1995 					 GFP_KERNEL);
1996 	BUG_ON(retval < 0);
1997 
1998 	return skb;
1999 }
2000 
2001 /* Called with ovs_mutex or RCU read lock. */
2002 static struct vport *lookup_vport(struct net *net,
2003 				  const struct ovs_header *ovs_header,
2004 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2005 {
2006 	struct datapath *dp;
2007 	struct vport *vport;
2008 
2009 	if (a[OVS_VPORT_ATTR_IFINDEX])
2010 		return ERR_PTR(-EOPNOTSUPP);
2011 	if (a[OVS_VPORT_ATTR_NAME]) {
2012 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2013 		if (!vport)
2014 			return ERR_PTR(-ENODEV);
2015 		if (ovs_header->dp_ifindex &&
2016 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2017 			return ERR_PTR(-ENODEV);
2018 		return vport;
2019 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2020 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2021 
2022 		if (port_no >= DP_MAX_PORTS)
2023 			return ERR_PTR(-EFBIG);
2024 
2025 		dp = get_dp(net, ovs_header->dp_ifindex);
2026 		if (!dp)
2027 			return ERR_PTR(-ENODEV);
2028 
2029 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2030 		if (!vport)
2031 			return ERR_PTR(-ENODEV);
2032 		return vport;
2033 	} else
2034 		return ERR_PTR(-EINVAL);
2035 
2036 }
2037 
2038 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2039 {
2040 	unsigned int dev_headroom, max_headroom = 0;
2041 	struct net_device *dev;
2042 	struct vport *vport;
2043 	int i;
2044 
2045 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2046 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2047 					lockdep_ovsl_is_held()) {
2048 			dev = vport->dev;
2049 			dev_headroom = netdev_get_fwd_headroom(dev);
2050 			if (dev_headroom > max_headroom)
2051 				max_headroom = dev_headroom;
2052 		}
2053 	}
2054 
2055 	return max_headroom;
2056 }
2057 
2058 /* Called with ovs_mutex */
2059 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2060 {
2061 	struct vport *vport;
2062 	int i;
2063 
2064 	dp->max_headroom = new_headroom;
2065 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2066 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2067 					lockdep_ovsl_is_held())
2068 			netdev_set_rx_headroom(vport->dev, new_headroom);
2069 }
2070 
2071 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2072 {
2073 	struct nlattr **a = info->attrs;
2074 	struct ovs_header *ovs_header = info->userhdr;
2075 	struct vport_parms parms;
2076 	struct sk_buff *reply;
2077 	struct vport *vport;
2078 	struct datapath *dp;
2079 	unsigned int new_headroom;
2080 	u32 port_no;
2081 	int err;
2082 
2083 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2084 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2085 		return -EINVAL;
2086 	if (a[OVS_VPORT_ATTR_IFINDEX])
2087 		return -EOPNOTSUPP;
2088 
2089 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2090 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2091 	if (port_no >= DP_MAX_PORTS)
2092 		return -EFBIG;
2093 
2094 	reply = ovs_vport_cmd_alloc_info();
2095 	if (!reply)
2096 		return -ENOMEM;
2097 
2098 	ovs_lock();
2099 restart:
2100 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2101 	err = -ENODEV;
2102 	if (!dp)
2103 		goto exit_unlock_free;
2104 
2105 	if (port_no) {
2106 		vport = ovs_vport_ovsl(dp, port_no);
2107 		err = -EBUSY;
2108 		if (vport)
2109 			goto exit_unlock_free;
2110 	} else {
2111 		for (port_no = 1; ; port_no++) {
2112 			if (port_no >= DP_MAX_PORTS) {
2113 				err = -EFBIG;
2114 				goto exit_unlock_free;
2115 			}
2116 			vport = ovs_vport_ovsl(dp, port_no);
2117 			if (!vport)
2118 				break;
2119 		}
2120 	}
2121 
2122 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2123 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2124 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2125 	parms.dp = dp;
2126 	parms.port_no = port_no;
2127 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2128 
2129 	vport = new_vport(&parms);
2130 	err = PTR_ERR(vport);
2131 	if (IS_ERR(vport)) {
2132 		if (err == -EAGAIN)
2133 			goto restart;
2134 		goto exit_unlock_free;
2135 	}
2136 
2137 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2138 				      info->snd_portid, info->snd_seq, 0,
2139 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2140 
2141 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2142 
2143 	if (new_headroom > dp->max_headroom)
2144 		ovs_update_headroom(dp, new_headroom);
2145 	else
2146 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2147 
2148 	BUG_ON(err < 0);
2149 	ovs_unlock();
2150 
2151 	ovs_notify(&dp_vport_genl_family, reply, info);
2152 	return 0;
2153 
2154 exit_unlock_free:
2155 	ovs_unlock();
2156 	kfree_skb(reply);
2157 	return err;
2158 }
2159 
2160 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2161 {
2162 	struct nlattr **a = info->attrs;
2163 	struct sk_buff *reply;
2164 	struct vport *vport;
2165 	int err;
2166 
2167 	reply = ovs_vport_cmd_alloc_info();
2168 	if (!reply)
2169 		return -ENOMEM;
2170 
2171 	ovs_lock();
2172 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2173 	err = PTR_ERR(vport);
2174 	if (IS_ERR(vport))
2175 		goto exit_unlock_free;
2176 
2177 	if (a[OVS_VPORT_ATTR_TYPE] &&
2178 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2179 		err = -EINVAL;
2180 		goto exit_unlock_free;
2181 	}
2182 
2183 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2184 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2185 		if (err)
2186 			goto exit_unlock_free;
2187 	}
2188 
2189 
2190 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2191 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2192 
2193 		err = ovs_vport_set_upcall_portids(vport, ids);
2194 		if (err)
2195 			goto exit_unlock_free;
2196 	}
2197 
2198 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2199 				      info->snd_portid, info->snd_seq, 0,
2200 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2201 	BUG_ON(err < 0);
2202 
2203 	ovs_unlock();
2204 	ovs_notify(&dp_vport_genl_family, reply, info);
2205 	return 0;
2206 
2207 exit_unlock_free:
2208 	ovs_unlock();
2209 	kfree_skb(reply);
2210 	return err;
2211 }
2212 
2213 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2214 {
2215 	bool update_headroom = false;
2216 	struct nlattr **a = info->attrs;
2217 	struct sk_buff *reply;
2218 	struct datapath *dp;
2219 	struct vport *vport;
2220 	unsigned int new_headroom;
2221 	int err;
2222 
2223 	reply = ovs_vport_cmd_alloc_info();
2224 	if (!reply)
2225 		return -ENOMEM;
2226 
2227 	ovs_lock();
2228 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2229 	err = PTR_ERR(vport);
2230 	if (IS_ERR(vport))
2231 		goto exit_unlock_free;
2232 
2233 	if (vport->port_no == OVSP_LOCAL) {
2234 		err = -EINVAL;
2235 		goto exit_unlock_free;
2236 	}
2237 
2238 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2239 				      info->snd_portid, info->snd_seq, 0,
2240 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2241 	BUG_ON(err < 0);
2242 
2243 	/* the vport deletion may trigger dp headroom update */
2244 	dp = vport->dp;
2245 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2246 		update_headroom = true;
2247 
2248 	netdev_reset_rx_headroom(vport->dev);
2249 	ovs_dp_detach_port(vport);
2250 
2251 	if (update_headroom) {
2252 		new_headroom = ovs_get_max_headroom(dp);
2253 
2254 		if (new_headroom < dp->max_headroom)
2255 			ovs_update_headroom(dp, new_headroom);
2256 	}
2257 	ovs_unlock();
2258 
2259 	ovs_notify(&dp_vport_genl_family, reply, info);
2260 	return 0;
2261 
2262 exit_unlock_free:
2263 	ovs_unlock();
2264 	kfree_skb(reply);
2265 	return err;
2266 }
2267 
2268 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2269 {
2270 	struct nlattr **a = info->attrs;
2271 	struct ovs_header *ovs_header = info->userhdr;
2272 	struct sk_buff *reply;
2273 	struct vport *vport;
2274 	int err;
2275 
2276 	reply = ovs_vport_cmd_alloc_info();
2277 	if (!reply)
2278 		return -ENOMEM;
2279 
2280 	rcu_read_lock();
2281 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2282 	err = PTR_ERR(vport);
2283 	if (IS_ERR(vport))
2284 		goto exit_unlock_free;
2285 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2286 				      info->snd_portid, info->snd_seq, 0,
2287 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2288 	BUG_ON(err < 0);
2289 	rcu_read_unlock();
2290 
2291 	return genlmsg_reply(reply, info);
2292 
2293 exit_unlock_free:
2294 	rcu_read_unlock();
2295 	kfree_skb(reply);
2296 	return err;
2297 }
2298 
2299 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2300 {
2301 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2302 	struct datapath *dp;
2303 	int bucket = cb->args[0], skip = cb->args[1];
2304 	int i, j = 0;
2305 
2306 	rcu_read_lock();
2307 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2308 	if (!dp) {
2309 		rcu_read_unlock();
2310 		return -ENODEV;
2311 	}
2312 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2313 		struct vport *vport;
2314 
2315 		j = 0;
2316 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2317 			if (j >= skip &&
2318 			    ovs_vport_cmd_fill_info(vport, skb,
2319 						    sock_net(skb->sk),
2320 						    NETLINK_CB(cb->skb).portid,
2321 						    cb->nlh->nlmsg_seq,
2322 						    NLM_F_MULTI,
2323 						    OVS_VPORT_CMD_GET,
2324 						    GFP_ATOMIC) < 0)
2325 				goto out;
2326 
2327 			j++;
2328 		}
2329 		skip = 0;
2330 	}
2331 out:
2332 	rcu_read_unlock();
2333 
2334 	cb->args[0] = i;
2335 	cb->args[1] = j;
2336 
2337 	return skb->len;
2338 }
2339 
2340 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2341 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2342 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2343 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2344 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2345 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2346 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2347 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2348 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2349 };
2350 
2351 static const struct genl_ops dp_vport_genl_ops[] = {
2352 	{ .cmd = OVS_VPORT_CMD_NEW,
2353 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2354 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2355 	  .doit = ovs_vport_cmd_new
2356 	},
2357 	{ .cmd = OVS_VPORT_CMD_DEL,
2358 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2359 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2360 	  .doit = ovs_vport_cmd_del
2361 	},
2362 	{ .cmd = OVS_VPORT_CMD_GET,
2363 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2364 	  .flags = 0,		    /* OK for unprivileged users. */
2365 	  .doit = ovs_vport_cmd_get,
2366 	  .dumpit = ovs_vport_cmd_dump
2367 	},
2368 	{ .cmd = OVS_VPORT_CMD_SET,
2369 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2370 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2371 	  .doit = ovs_vport_cmd_set,
2372 	},
2373 };
2374 
2375 struct genl_family dp_vport_genl_family __ro_after_init = {
2376 	.hdrsize = sizeof(struct ovs_header),
2377 	.name = OVS_VPORT_FAMILY,
2378 	.version = OVS_VPORT_VERSION,
2379 	.maxattr = OVS_VPORT_ATTR_MAX,
2380 	.policy = vport_policy,
2381 	.netnsok = true,
2382 	.parallel_ops = true,
2383 	.ops = dp_vport_genl_ops,
2384 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2385 	.mcgrps = &ovs_dp_vport_multicast_group,
2386 	.n_mcgrps = 1,
2387 	.module = THIS_MODULE,
2388 };
2389 
2390 static struct genl_family * const dp_genl_families[] = {
2391 	&dp_datapath_genl_family,
2392 	&dp_vport_genl_family,
2393 	&dp_flow_genl_family,
2394 	&dp_packet_genl_family,
2395 	&dp_meter_genl_family,
2396 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2397 	&dp_ct_limit_genl_family,
2398 #endif
2399 };
2400 
2401 static void dp_unregister_genl(int n_families)
2402 {
2403 	int i;
2404 
2405 	for (i = 0; i < n_families; i++)
2406 		genl_unregister_family(dp_genl_families[i]);
2407 }
2408 
2409 static int __init dp_register_genl(void)
2410 {
2411 	int err;
2412 	int i;
2413 
2414 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2415 
2416 		err = genl_register_family(dp_genl_families[i]);
2417 		if (err)
2418 			goto error;
2419 	}
2420 
2421 	return 0;
2422 
2423 error:
2424 	dp_unregister_genl(i);
2425 	return err;
2426 }
2427 
2428 static int __net_init ovs_init_net(struct net *net)
2429 {
2430 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2431 
2432 	INIT_LIST_HEAD(&ovs_net->dps);
2433 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2434 	return ovs_ct_init(net);
2435 }
2436 
2437 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2438 					    struct list_head *head)
2439 {
2440 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2441 	struct datapath *dp;
2442 
2443 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2444 		int i;
2445 
2446 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2447 			struct vport *vport;
2448 
2449 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2450 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2451 					continue;
2452 
2453 				if (dev_net(vport->dev) == dnet)
2454 					list_add(&vport->detach_list, head);
2455 			}
2456 		}
2457 	}
2458 }
2459 
2460 static void __net_exit ovs_exit_net(struct net *dnet)
2461 {
2462 	struct datapath *dp, *dp_next;
2463 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2464 	struct vport *vport, *vport_next;
2465 	struct net *net;
2466 	LIST_HEAD(head);
2467 
2468 	ovs_ct_exit(dnet);
2469 	ovs_lock();
2470 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2471 		__dp_destroy(dp);
2472 
2473 	down_read(&net_rwsem);
2474 	for_each_net(net)
2475 		list_vports_from_net(net, dnet, &head);
2476 	up_read(&net_rwsem);
2477 
2478 	/* Detach all vports from given namespace. */
2479 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2480 		list_del(&vport->detach_list);
2481 		ovs_dp_detach_port(vport);
2482 	}
2483 
2484 	ovs_unlock();
2485 
2486 	cancel_work_sync(&ovs_net->dp_notify_work);
2487 }
2488 
2489 static struct pernet_operations ovs_net_ops = {
2490 	.init = ovs_init_net,
2491 	.exit = ovs_exit_net,
2492 	.id   = &ovs_net_id,
2493 	.size = sizeof(struct ovs_net),
2494 };
2495 
2496 static int __init dp_init(void)
2497 {
2498 	int err;
2499 
2500 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof_field(struct sk_buff, cb));
2501 
2502 	pr_info("Open vSwitch switching datapath\n");
2503 
2504 	err = action_fifos_init();
2505 	if (err)
2506 		goto error;
2507 
2508 	err = ovs_internal_dev_rtnl_link_register();
2509 	if (err)
2510 		goto error_action_fifos_exit;
2511 
2512 	err = ovs_flow_init();
2513 	if (err)
2514 		goto error_unreg_rtnl_link;
2515 
2516 	err = ovs_vport_init();
2517 	if (err)
2518 		goto error_flow_exit;
2519 
2520 	err = register_pernet_device(&ovs_net_ops);
2521 	if (err)
2522 		goto error_vport_exit;
2523 
2524 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2525 	if (err)
2526 		goto error_netns_exit;
2527 
2528 	err = ovs_netdev_init();
2529 	if (err)
2530 		goto error_unreg_notifier;
2531 
2532 	err = dp_register_genl();
2533 	if (err < 0)
2534 		goto error_unreg_netdev;
2535 
2536 	return 0;
2537 
2538 error_unreg_netdev:
2539 	ovs_netdev_exit();
2540 error_unreg_notifier:
2541 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2542 error_netns_exit:
2543 	unregister_pernet_device(&ovs_net_ops);
2544 error_vport_exit:
2545 	ovs_vport_exit();
2546 error_flow_exit:
2547 	ovs_flow_exit();
2548 error_unreg_rtnl_link:
2549 	ovs_internal_dev_rtnl_link_unregister();
2550 error_action_fifos_exit:
2551 	action_fifos_exit();
2552 error:
2553 	return err;
2554 }
2555 
2556 static void dp_cleanup(void)
2557 {
2558 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2559 	ovs_netdev_exit();
2560 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2561 	unregister_pernet_device(&ovs_net_ops);
2562 	rcu_barrier();
2563 	ovs_vport_exit();
2564 	ovs_flow_exit();
2565 	ovs_internal_dev_rtnl_link_unregister();
2566 	action_fifos_exit();
2567 }
2568 
2569 module_init(dp_init);
2570 module_exit(dp_cleanup);
2571 
2572 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2573 MODULE_LICENSE("GPL");
2574 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2575 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2576 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2577 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2578 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2579 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2580