xref: /linux/net/openvswitch/datapath.c (revision 4d66c56f7efe122d09d06cd3ebfa52a43d51a9cb)
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, skb_get_hash(skb),
231 					 &n_mask_hit);
232 	if (unlikely(!flow)) {
233 		struct dp_upcall_info upcall;
234 
235 		memset(&upcall, 0, sizeof(upcall));
236 		upcall.cmd = OVS_PACKET_CMD_MISS;
237 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
238 		upcall.mru = OVS_CB(skb)->mru;
239 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
240 		if (unlikely(error))
241 			kfree_skb(skb);
242 		else
243 			consume_skb(skb);
244 		stats_counter = &stats->n_missed;
245 		goto out;
246 	}
247 
248 	ovs_flow_stats_update(flow, key->tp.flags, skb);
249 	sf_acts = rcu_dereference(flow->sf_acts);
250 	error = ovs_execute_actions(dp, skb, sf_acts, key);
251 	if (unlikely(error))
252 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
253 							ovs_dp_name(dp), error);
254 
255 	stats_counter = &stats->n_hit;
256 
257 out:
258 	/* Update datapath statistics. */
259 	u64_stats_update_begin(&stats->syncp);
260 	(*stats_counter)++;
261 	stats->n_mask_hit += n_mask_hit;
262 	u64_stats_update_end(&stats->syncp);
263 }
264 
265 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
266 		  const struct sw_flow_key *key,
267 		  const struct dp_upcall_info *upcall_info,
268 		  uint32_t cutlen)
269 {
270 	struct dp_stats_percpu *stats;
271 	int err;
272 
273 	if (upcall_info->portid == 0) {
274 		err = -ENOTCONN;
275 		goto err;
276 	}
277 
278 	if (!skb_is_gso(skb))
279 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
280 	else
281 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
282 	if (err)
283 		goto err;
284 
285 	return 0;
286 
287 err:
288 	stats = this_cpu_ptr(dp->stats_percpu);
289 
290 	u64_stats_update_begin(&stats->syncp);
291 	stats->n_lost++;
292 	u64_stats_update_end(&stats->syncp);
293 
294 	return err;
295 }
296 
297 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
298 			     const struct sw_flow_key *key,
299 			     const struct dp_upcall_info *upcall_info,
300 				 uint32_t cutlen)
301 {
302 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
303 	struct sw_flow_key later_key;
304 	struct sk_buff *segs, *nskb;
305 	int err;
306 
307 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
308 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
309 	if (IS_ERR(segs))
310 		return PTR_ERR(segs);
311 	if (segs == NULL)
312 		return -EINVAL;
313 
314 	if (gso_type & SKB_GSO_UDP) {
315 		/* The initial flow key extracted by ovs_flow_key_extract()
316 		 * in this case is for a first fragment, so we need to
317 		 * properly mark later fragments.
318 		 */
319 		later_key = *key;
320 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
321 	}
322 
323 	/* Queue all of the segments. */
324 	skb = segs;
325 	do {
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 	} while ((skb = skb->next));
334 
335 	/* Free all of the segments. */
336 	skb = segs;
337 	do {
338 		nskb = skb->next;
339 		if (err)
340 			kfree_skb(skb);
341 		else
342 			consume_skb(skb);
343 	} while ((skb = nskb));
344 	return err;
345 }
346 
347 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
348 			      unsigned int hdrlen, int actions_attrlen)
349 {
350 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
351 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
352 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
353 		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
354 		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
355 
356 	/* OVS_PACKET_ATTR_USERDATA */
357 	if (upcall_info->userdata)
358 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
359 
360 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
361 	if (upcall_info->egress_tun_info)
362 		size += nla_total_size(ovs_tun_key_attr_size());
363 
364 	/* OVS_PACKET_ATTR_ACTIONS */
365 	if (upcall_info->actions_len)
366 		size += nla_total_size(actions_attrlen);
367 
368 	/* OVS_PACKET_ATTR_MRU */
369 	if (upcall_info->mru)
370 		size += nla_total_size(sizeof(upcall_info->mru));
371 
372 	return size;
373 }
374 
375 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
376 {
377 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
378 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
379 
380 		if (plen > 0)
381 			skb_put_zero(skb, plen);
382 	}
383 }
384 
385 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
386 				  const struct sw_flow_key *key,
387 				  const struct dp_upcall_info *upcall_info,
388 				  uint32_t cutlen)
389 {
390 	struct ovs_header *upcall;
391 	struct sk_buff *nskb = NULL;
392 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
393 	struct nlattr *nla;
394 	size_t len;
395 	unsigned int hlen;
396 	int err, dp_ifindex;
397 	u64 hash;
398 
399 	dp_ifindex = get_dpifindex(dp);
400 	if (!dp_ifindex)
401 		return -ENODEV;
402 
403 	if (skb_vlan_tag_present(skb)) {
404 		nskb = skb_clone(skb, GFP_ATOMIC);
405 		if (!nskb)
406 			return -ENOMEM;
407 
408 		nskb = __vlan_hwaccel_push_inside(nskb);
409 		if (!nskb)
410 			return -ENOMEM;
411 
412 		skb = nskb;
413 	}
414 
415 	if (nla_attr_size(skb->len) > USHRT_MAX) {
416 		err = -EFBIG;
417 		goto out;
418 	}
419 
420 	/* Complete checksum if needed */
421 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
422 	    (err = skb_csum_hwoffload_help(skb, 0)))
423 		goto out;
424 
425 	/* Older versions of OVS user space enforce alignment of the last
426 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
427 	 * padding logic. Only perform zerocopy if padding is not required.
428 	 */
429 	if (dp->user_features & OVS_DP_F_UNALIGNED)
430 		hlen = skb_zerocopy_headlen(skb);
431 	else
432 		hlen = skb->len;
433 
434 	len = upcall_msg_size(upcall_info, hlen - cutlen,
435 			      OVS_CB(skb)->acts_origlen);
436 	user_skb = genlmsg_new(len, GFP_ATOMIC);
437 	if (!user_skb) {
438 		err = -ENOMEM;
439 		goto out;
440 	}
441 
442 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
443 			     0, upcall_info->cmd);
444 	if (!upcall) {
445 		err = -EINVAL;
446 		goto out;
447 	}
448 	upcall->dp_ifindex = dp_ifindex;
449 
450 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
451 	if (err)
452 		goto out;
453 
454 	if (upcall_info->userdata)
455 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
456 			  nla_len(upcall_info->userdata),
457 			  nla_data(upcall_info->userdata));
458 
459 	if (upcall_info->egress_tun_info) {
460 		nla = nla_nest_start_noflag(user_skb,
461 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
462 		if (!nla) {
463 			err = -EMSGSIZE;
464 			goto out;
465 		}
466 		err = ovs_nla_put_tunnel_info(user_skb,
467 					      upcall_info->egress_tun_info);
468 		if (err)
469 			goto out;
470 
471 		nla_nest_end(user_skb, nla);
472 	}
473 
474 	if (upcall_info->actions_len) {
475 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
476 		if (!nla) {
477 			err = -EMSGSIZE;
478 			goto out;
479 		}
480 		err = ovs_nla_put_actions(upcall_info->actions,
481 					  upcall_info->actions_len,
482 					  user_skb);
483 		if (!err)
484 			nla_nest_end(user_skb, nla);
485 		else
486 			nla_nest_cancel(user_skb, nla);
487 	}
488 
489 	/* Add OVS_PACKET_ATTR_MRU */
490 	if (upcall_info->mru) {
491 		if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
492 				upcall_info->mru)) {
493 			err = -ENOBUFS;
494 			goto out;
495 		}
496 		pad_packet(dp, user_skb);
497 	}
498 
499 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
500 	if (cutlen > 0) {
501 		if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
502 				skb->len)) {
503 			err = -ENOBUFS;
504 			goto out;
505 		}
506 		pad_packet(dp, user_skb);
507 	}
508 
509 	/* Add OVS_PACKET_ATTR_HASH */
510 	hash = skb_get_hash_raw(skb);
511 	if (skb->sw_hash)
512 		hash |= OVS_PACKET_HASH_SW_BIT;
513 
514 	if (skb->l4_hash)
515 		hash |= OVS_PACKET_HASH_L4_BIT;
516 
517 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
518 		err = -ENOBUFS;
519 		goto out;
520 	}
521 
522 	/* Only reserve room for attribute header, packet data is added
523 	 * in skb_zerocopy() */
524 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
525 		err = -ENOBUFS;
526 		goto out;
527 	}
528 	nla->nla_len = nla_attr_size(skb->len - cutlen);
529 
530 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
531 	if (err)
532 		goto out;
533 
534 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
535 	pad_packet(dp, user_skb);
536 
537 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
538 
539 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
540 	user_skb = NULL;
541 out:
542 	if (err)
543 		skb_tx_error(skb);
544 	kfree_skb(user_skb);
545 	kfree_skb(nskb);
546 	return err;
547 }
548 
549 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
550 {
551 	struct ovs_header *ovs_header = info->userhdr;
552 	struct net *net = sock_net(skb->sk);
553 	struct nlattr **a = info->attrs;
554 	struct sw_flow_actions *acts;
555 	struct sk_buff *packet;
556 	struct sw_flow *flow;
557 	struct sw_flow_actions *sf_acts;
558 	struct datapath *dp;
559 	struct vport *input_vport;
560 	u16 mru = 0;
561 	u64 hash;
562 	int len;
563 	int err;
564 	bool log = !a[OVS_PACKET_ATTR_PROBE];
565 
566 	err = -EINVAL;
567 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
568 	    !a[OVS_PACKET_ATTR_ACTIONS])
569 		goto err;
570 
571 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
572 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
573 	err = -ENOMEM;
574 	if (!packet)
575 		goto err;
576 	skb_reserve(packet, NET_IP_ALIGN);
577 
578 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
579 
580 	/* Set packet's mru */
581 	if (a[OVS_PACKET_ATTR_MRU]) {
582 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
583 		packet->ignore_df = 1;
584 	}
585 	OVS_CB(packet)->mru = mru;
586 
587 	if (a[OVS_PACKET_ATTR_HASH]) {
588 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
589 
590 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
591 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
592 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
593 	}
594 
595 	/* Build an sw_flow for sending this packet. */
596 	flow = ovs_flow_alloc();
597 	err = PTR_ERR(flow);
598 	if (IS_ERR(flow))
599 		goto err_kfree_skb;
600 
601 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
602 					     packet, &flow->key, log);
603 	if (err)
604 		goto err_flow_free;
605 
606 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
607 				   &flow->key, &acts, log);
608 	if (err)
609 		goto err_flow_free;
610 
611 	rcu_assign_pointer(flow->sf_acts, acts);
612 	packet->priority = flow->key.phy.priority;
613 	packet->mark = flow->key.phy.skb_mark;
614 
615 	rcu_read_lock();
616 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
617 	err = -ENODEV;
618 	if (!dp)
619 		goto err_unlock;
620 
621 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
622 	if (!input_vport)
623 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
624 
625 	if (!input_vport)
626 		goto err_unlock;
627 
628 	packet->dev = input_vport->dev;
629 	OVS_CB(packet)->input_vport = input_vport;
630 	sf_acts = rcu_dereference(flow->sf_acts);
631 
632 	local_bh_disable();
633 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
634 	local_bh_enable();
635 	rcu_read_unlock();
636 
637 	ovs_flow_free(flow, false);
638 	return err;
639 
640 err_unlock:
641 	rcu_read_unlock();
642 err_flow_free:
643 	ovs_flow_free(flow, false);
644 err_kfree_skb:
645 	kfree_skb(packet);
646 err:
647 	return err;
648 }
649 
650 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
651 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
652 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
653 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
654 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
655 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
656 };
657 
658 static const struct genl_ops dp_packet_genl_ops[] = {
659 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
660 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
661 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
662 	  .doit = ovs_packet_cmd_execute
663 	}
664 };
665 
666 static struct genl_family dp_packet_genl_family __ro_after_init = {
667 	.hdrsize = sizeof(struct ovs_header),
668 	.name = OVS_PACKET_FAMILY,
669 	.version = OVS_PACKET_VERSION,
670 	.maxattr = OVS_PACKET_ATTR_MAX,
671 	.policy = packet_policy,
672 	.netnsok = true,
673 	.parallel_ops = true,
674 	.ops = dp_packet_genl_ops,
675 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
676 	.module = THIS_MODULE,
677 };
678 
679 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
680 			 struct ovs_dp_megaflow_stats *mega_stats)
681 {
682 	int i;
683 
684 	memset(mega_stats, 0, sizeof(*mega_stats));
685 
686 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
687 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
688 
689 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
690 
691 	for_each_possible_cpu(i) {
692 		const struct dp_stats_percpu *percpu_stats;
693 		struct dp_stats_percpu local_stats;
694 		unsigned int start;
695 
696 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
697 
698 		do {
699 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
700 			local_stats = *percpu_stats;
701 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
702 
703 		stats->n_hit += local_stats.n_hit;
704 		stats->n_missed += local_stats.n_missed;
705 		stats->n_lost += local_stats.n_lost;
706 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
707 	}
708 }
709 
710 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
711 {
712 	return ovs_identifier_is_ufid(sfid) &&
713 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
714 }
715 
716 static bool should_fill_mask(uint32_t ufid_flags)
717 {
718 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
719 }
720 
721 static bool should_fill_actions(uint32_t ufid_flags)
722 {
723 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
724 }
725 
726 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
727 				    const struct sw_flow_id *sfid,
728 				    uint32_t ufid_flags)
729 {
730 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
731 
732 	/* OVS_FLOW_ATTR_UFID */
733 	if (sfid && ovs_identifier_is_ufid(sfid))
734 		len += nla_total_size(sfid->ufid_len);
735 
736 	/* OVS_FLOW_ATTR_KEY */
737 	if (!sfid || should_fill_key(sfid, ufid_flags))
738 		len += nla_total_size(ovs_key_attr_size());
739 
740 	/* OVS_FLOW_ATTR_MASK */
741 	if (should_fill_mask(ufid_flags))
742 		len += nla_total_size(ovs_key_attr_size());
743 
744 	/* OVS_FLOW_ATTR_ACTIONS */
745 	if (should_fill_actions(ufid_flags))
746 		len += nla_total_size(acts->orig_len);
747 
748 	return len
749 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
750 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
751 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
752 }
753 
754 /* Called with ovs_mutex or RCU read lock. */
755 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
756 				   struct sk_buff *skb)
757 {
758 	struct ovs_flow_stats stats;
759 	__be16 tcp_flags;
760 	unsigned long used;
761 
762 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
763 
764 	if (used &&
765 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
766 			      OVS_FLOW_ATTR_PAD))
767 		return -EMSGSIZE;
768 
769 	if (stats.n_packets &&
770 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
771 			  sizeof(struct ovs_flow_stats), &stats,
772 			  OVS_FLOW_ATTR_PAD))
773 		return -EMSGSIZE;
774 
775 	if ((u8)ntohs(tcp_flags) &&
776 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
777 		return -EMSGSIZE;
778 
779 	return 0;
780 }
781 
782 /* Called with ovs_mutex or RCU read lock. */
783 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
784 				     struct sk_buff *skb, int skb_orig_len)
785 {
786 	struct nlattr *start;
787 	int err;
788 
789 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
790 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
791 	 * Netlink but individual action lists can be longer than
792 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
793 	 * The userspace caller can always fetch the actions separately if it
794 	 * really wants them.  (Most userspace callers in fact don't care.)
795 	 *
796 	 * This can only fail for dump operations because the skb is always
797 	 * properly sized for single flows.
798 	 */
799 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
800 	if (start) {
801 		const struct sw_flow_actions *sf_acts;
802 
803 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
804 		err = ovs_nla_put_actions(sf_acts->actions,
805 					  sf_acts->actions_len, skb);
806 
807 		if (!err)
808 			nla_nest_end(skb, start);
809 		else {
810 			if (skb_orig_len)
811 				return err;
812 
813 			nla_nest_cancel(skb, start);
814 		}
815 	} else if (skb_orig_len) {
816 		return -EMSGSIZE;
817 	}
818 
819 	return 0;
820 }
821 
822 /* Called with ovs_mutex or RCU read lock. */
823 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
824 				  struct sk_buff *skb, u32 portid,
825 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
826 {
827 	const int skb_orig_len = skb->len;
828 	struct ovs_header *ovs_header;
829 	int err;
830 
831 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
832 				 flags, cmd);
833 	if (!ovs_header)
834 		return -EMSGSIZE;
835 
836 	ovs_header->dp_ifindex = dp_ifindex;
837 
838 	err = ovs_nla_put_identifier(flow, skb);
839 	if (err)
840 		goto error;
841 
842 	if (should_fill_key(&flow->id, ufid_flags)) {
843 		err = ovs_nla_put_masked_key(flow, skb);
844 		if (err)
845 			goto error;
846 	}
847 
848 	if (should_fill_mask(ufid_flags)) {
849 		err = ovs_nla_put_mask(flow, skb);
850 		if (err)
851 			goto error;
852 	}
853 
854 	err = ovs_flow_cmd_fill_stats(flow, skb);
855 	if (err)
856 		goto error;
857 
858 	if (should_fill_actions(ufid_flags)) {
859 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
860 		if (err)
861 			goto error;
862 	}
863 
864 	genlmsg_end(skb, ovs_header);
865 	return 0;
866 
867 error:
868 	genlmsg_cancel(skb, ovs_header);
869 	return err;
870 }
871 
872 /* May not be called with RCU read lock. */
873 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
874 					       const struct sw_flow_id *sfid,
875 					       struct genl_info *info,
876 					       bool always,
877 					       uint32_t ufid_flags)
878 {
879 	struct sk_buff *skb;
880 	size_t len;
881 
882 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
883 		return NULL;
884 
885 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
886 	skb = genlmsg_new(len, GFP_KERNEL);
887 	if (!skb)
888 		return ERR_PTR(-ENOMEM);
889 
890 	return skb;
891 }
892 
893 /* Called with ovs_mutex. */
894 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
895 					       int dp_ifindex,
896 					       struct genl_info *info, u8 cmd,
897 					       bool always, u32 ufid_flags)
898 {
899 	struct sk_buff *skb;
900 	int retval;
901 
902 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
903 				      &flow->id, info, always, ufid_flags);
904 	if (IS_ERR_OR_NULL(skb))
905 		return skb;
906 
907 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
908 					info->snd_portid, info->snd_seq, 0,
909 					cmd, ufid_flags);
910 	BUG_ON(retval < 0);
911 	return skb;
912 }
913 
914 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
915 {
916 	struct net *net = sock_net(skb->sk);
917 	struct nlattr **a = info->attrs;
918 	struct ovs_header *ovs_header = info->userhdr;
919 	struct sw_flow *flow = NULL, *new_flow;
920 	struct sw_flow_mask mask;
921 	struct sk_buff *reply;
922 	struct datapath *dp;
923 	struct sw_flow_actions *acts;
924 	struct sw_flow_match match;
925 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
926 	int error;
927 	bool log = !a[OVS_FLOW_ATTR_PROBE];
928 
929 	/* Must have key and actions. */
930 	error = -EINVAL;
931 	if (!a[OVS_FLOW_ATTR_KEY]) {
932 		OVS_NLERR(log, "Flow key attr not present in new flow.");
933 		goto error;
934 	}
935 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
936 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
937 		goto error;
938 	}
939 
940 	/* Most of the time we need to allocate a new flow, do it before
941 	 * locking.
942 	 */
943 	new_flow = ovs_flow_alloc();
944 	if (IS_ERR(new_flow)) {
945 		error = PTR_ERR(new_flow);
946 		goto error;
947 	}
948 
949 	/* Extract key. */
950 	ovs_match_init(&match, &new_flow->key, false, &mask);
951 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
952 				  a[OVS_FLOW_ATTR_MASK], log);
953 	if (error)
954 		goto err_kfree_flow;
955 
956 	/* Extract flow identifier. */
957 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
958 				       &new_flow->key, log);
959 	if (error)
960 		goto err_kfree_flow;
961 
962 	/* unmasked key is needed to match when ufid is not used. */
963 	if (ovs_identifier_is_key(&new_flow->id))
964 		match.key = new_flow->id.unmasked_key;
965 
966 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
967 
968 	/* Validate actions. */
969 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
970 				     &new_flow->key, &acts, log);
971 	if (error) {
972 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
973 		goto err_kfree_flow;
974 	}
975 
976 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
977 					ufid_flags);
978 	if (IS_ERR(reply)) {
979 		error = PTR_ERR(reply);
980 		goto err_kfree_acts;
981 	}
982 
983 	ovs_lock();
984 	dp = get_dp(net, ovs_header->dp_ifindex);
985 	if (unlikely(!dp)) {
986 		error = -ENODEV;
987 		goto err_unlock_ovs;
988 	}
989 
990 	/* Check if this is a duplicate flow */
991 	if (ovs_identifier_is_ufid(&new_flow->id))
992 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
993 	if (!flow)
994 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
995 	if (likely(!flow)) {
996 		rcu_assign_pointer(new_flow->sf_acts, acts);
997 
998 		/* Put flow in bucket. */
999 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1000 		if (unlikely(error)) {
1001 			acts = NULL;
1002 			goto err_unlock_ovs;
1003 		}
1004 
1005 		if (unlikely(reply)) {
1006 			error = ovs_flow_cmd_fill_info(new_flow,
1007 						       ovs_header->dp_ifindex,
1008 						       reply, info->snd_portid,
1009 						       info->snd_seq, 0,
1010 						       OVS_FLOW_CMD_NEW,
1011 						       ufid_flags);
1012 			BUG_ON(error < 0);
1013 		}
1014 		ovs_unlock();
1015 	} else {
1016 		struct sw_flow_actions *old_acts;
1017 
1018 		/* Bail out if we're not allowed to modify an existing flow.
1019 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1020 		 * because Generic Netlink treats the latter as a dump
1021 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1022 		 * gets fixed.
1023 		 */
1024 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1025 							 | NLM_F_EXCL))) {
1026 			error = -EEXIST;
1027 			goto err_unlock_ovs;
1028 		}
1029 		/* The flow identifier has to be the same for flow updates.
1030 		 * Look for any overlapping flow.
1031 		 */
1032 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1033 			if (ovs_identifier_is_key(&flow->id))
1034 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1035 								 &match);
1036 			else /* UFID matches but key is different */
1037 				flow = NULL;
1038 			if (!flow) {
1039 				error = -ENOENT;
1040 				goto err_unlock_ovs;
1041 			}
1042 		}
1043 		/* Update actions. */
1044 		old_acts = ovsl_dereference(flow->sf_acts);
1045 		rcu_assign_pointer(flow->sf_acts, acts);
1046 
1047 		if (unlikely(reply)) {
1048 			error = ovs_flow_cmd_fill_info(flow,
1049 						       ovs_header->dp_ifindex,
1050 						       reply, info->snd_portid,
1051 						       info->snd_seq, 0,
1052 						       OVS_FLOW_CMD_NEW,
1053 						       ufid_flags);
1054 			BUG_ON(error < 0);
1055 		}
1056 		ovs_unlock();
1057 
1058 		ovs_nla_free_flow_actions_rcu(old_acts);
1059 		ovs_flow_free(new_flow, false);
1060 	}
1061 
1062 	if (reply)
1063 		ovs_notify(&dp_flow_genl_family, reply, info);
1064 	return 0;
1065 
1066 err_unlock_ovs:
1067 	ovs_unlock();
1068 	kfree_skb(reply);
1069 err_kfree_acts:
1070 	ovs_nla_free_flow_actions(acts);
1071 err_kfree_flow:
1072 	ovs_flow_free(new_flow, false);
1073 error:
1074 	return error;
1075 }
1076 
1077 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1078 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1079 						const struct nlattr *a,
1080 						const struct sw_flow_key *key,
1081 						const struct sw_flow_mask *mask,
1082 						bool log)
1083 {
1084 	struct sw_flow_actions *acts;
1085 	struct sw_flow_key masked_key;
1086 	int error;
1087 
1088 	ovs_flow_mask_key(&masked_key, key, true, mask);
1089 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1090 	if (error) {
1091 		OVS_NLERR(log,
1092 			  "Actions may not be safe on all matching packets");
1093 		return ERR_PTR(error);
1094 	}
1095 
1096 	return acts;
1097 }
1098 
1099 /* Factor out match-init and action-copy to avoid
1100  * "Wframe-larger-than=1024" warning. Because mask is only
1101  * used to get actions, we new a function to save some
1102  * stack space.
1103  *
1104  * If there are not key and action attrs, we return 0
1105  * directly. In the case, the caller will also not use the
1106  * match as before. If there is action attr, we try to get
1107  * actions and save them to *acts. Before returning from
1108  * the function, we reset the match->mask pointer. Because
1109  * we should not to return match object with dangling reference
1110  * to mask.
1111  * */
1112 static noinline_for_stack int
1113 ovs_nla_init_match_and_action(struct net *net,
1114 			      struct sw_flow_match *match,
1115 			      struct sw_flow_key *key,
1116 			      struct nlattr **a,
1117 			      struct sw_flow_actions **acts,
1118 			      bool log)
1119 {
1120 	struct sw_flow_mask mask;
1121 	int error = 0;
1122 
1123 	if (a[OVS_FLOW_ATTR_KEY]) {
1124 		ovs_match_init(match, key, true, &mask);
1125 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1126 					  a[OVS_FLOW_ATTR_MASK], log);
1127 		if (error)
1128 			goto error;
1129 	}
1130 
1131 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1132 		if (!a[OVS_FLOW_ATTR_KEY]) {
1133 			OVS_NLERR(log,
1134 				  "Flow key attribute not present in set flow.");
1135 			error = -EINVAL;
1136 			goto error;
1137 		}
1138 
1139 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1140 					 &mask, log);
1141 		if (IS_ERR(*acts)) {
1142 			error = PTR_ERR(*acts);
1143 			goto error;
1144 		}
1145 	}
1146 
1147 	/* On success, error is 0. */
1148 error:
1149 	match->mask = NULL;
1150 	return error;
1151 }
1152 
1153 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1154 {
1155 	struct net *net = sock_net(skb->sk);
1156 	struct nlattr **a = info->attrs;
1157 	struct ovs_header *ovs_header = info->userhdr;
1158 	struct sw_flow_key key;
1159 	struct sw_flow *flow;
1160 	struct sk_buff *reply = NULL;
1161 	struct datapath *dp;
1162 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1163 	struct sw_flow_match match;
1164 	struct sw_flow_id sfid;
1165 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1166 	int error = 0;
1167 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1168 	bool ufid_present;
1169 
1170 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1171 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1172 		OVS_NLERR(log,
1173 			  "Flow set message rejected, Key attribute missing.");
1174 		return -EINVAL;
1175 	}
1176 
1177 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1178 					      &acts, log);
1179 	if (error)
1180 		goto error;
1181 
1182 	if (acts) {
1183 		/* Can allocate before locking if have acts. */
1184 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1185 						ufid_flags);
1186 		if (IS_ERR(reply)) {
1187 			error = PTR_ERR(reply);
1188 			goto err_kfree_acts;
1189 		}
1190 	}
1191 
1192 	ovs_lock();
1193 	dp = get_dp(net, ovs_header->dp_ifindex);
1194 	if (unlikely(!dp)) {
1195 		error = -ENODEV;
1196 		goto err_unlock_ovs;
1197 	}
1198 	/* Check that the flow exists. */
1199 	if (ufid_present)
1200 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1201 	else
1202 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1203 	if (unlikely(!flow)) {
1204 		error = -ENOENT;
1205 		goto err_unlock_ovs;
1206 	}
1207 
1208 	/* Update actions, if present. */
1209 	if (likely(acts)) {
1210 		old_acts = ovsl_dereference(flow->sf_acts);
1211 		rcu_assign_pointer(flow->sf_acts, acts);
1212 
1213 		if (unlikely(reply)) {
1214 			error = ovs_flow_cmd_fill_info(flow,
1215 						       ovs_header->dp_ifindex,
1216 						       reply, info->snd_portid,
1217 						       info->snd_seq, 0,
1218 						       OVS_FLOW_CMD_SET,
1219 						       ufid_flags);
1220 			BUG_ON(error < 0);
1221 		}
1222 	} else {
1223 		/* Could not alloc without acts before locking. */
1224 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1225 						info, OVS_FLOW_CMD_SET, false,
1226 						ufid_flags);
1227 
1228 		if (IS_ERR(reply)) {
1229 			error = PTR_ERR(reply);
1230 			goto err_unlock_ovs;
1231 		}
1232 	}
1233 
1234 	/* Clear stats. */
1235 	if (a[OVS_FLOW_ATTR_CLEAR])
1236 		ovs_flow_stats_clear(flow);
1237 	ovs_unlock();
1238 
1239 	if (reply)
1240 		ovs_notify(&dp_flow_genl_family, reply, info);
1241 	if (old_acts)
1242 		ovs_nla_free_flow_actions_rcu(old_acts);
1243 
1244 	return 0;
1245 
1246 err_unlock_ovs:
1247 	ovs_unlock();
1248 	kfree_skb(reply);
1249 err_kfree_acts:
1250 	ovs_nla_free_flow_actions(acts);
1251 error:
1252 	return error;
1253 }
1254 
1255 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1256 {
1257 	struct nlattr **a = info->attrs;
1258 	struct ovs_header *ovs_header = info->userhdr;
1259 	struct net *net = sock_net(skb->sk);
1260 	struct sw_flow_key key;
1261 	struct sk_buff *reply;
1262 	struct sw_flow *flow;
1263 	struct datapath *dp;
1264 	struct sw_flow_match match;
1265 	struct sw_flow_id ufid;
1266 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1267 	int err = 0;
1268 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1269 	bool ufid_present;
1270 
1271 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1272 	if (a[OVS_FLOW_ATTR_KEY]) {
1273 		ovs_match_init(&match, &key, true, NULL);
1274 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1275 					log);
1276 	} else if (!ufid_present) {
1277 		OVS_NLERR(log,
1278 			  "Flow get message rejected, Key attribute missing.");
1279 		err = -EINVAL;
1280 	}
1281 	if (err)
1282 		return err;
1283 
1284 	ovs_lock();
1285 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1286 	if (!dp) {
1287 		err = -ENODEV;
1288 		goto unlock;
1289 	}
1290 
1291 	if (ufid_present)
1292 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1293 	else
1294 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1295 	if (!flow) {
1296 		err = -ENOENT;
1297 		goto unlock;
1298 	}
1299 
1300 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1301 					OVS_FLOW_CMD_GET, true, ufid_flags);
1302 	if (IS_ERR(reply)) {
1303 		err = PTR_ERR(reply);
1304 		goto unlock;
1305 	}
1306 
1307 	ovs_unlock();
1308 	return genlmsg_reply(reply, info);
1309 unlock:
1310 	ovs_unlock();
1311 	return err;
1312 }
1313 
1314 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1315 {
1316 	struct nlattr **a = info->attrs;
1317 	struct ovs_header *ovs_header = info->userhdr;
1318 	struct net *net = sock_net(skb->sk);
1319 	struct sw_flow_key key;
1320 	struct sk_buff *reply;
1321 	struct sw_flow *flow = NULL;
1322 	struct datapath *dp;
1323 	struct sw_flow_match match;
1324 	struct sw_flow_id ufid;
1325 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1326 	int err;
1327 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1328 	bool ufid_present;
1329 
1330 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1331 	if (a[OVS_FLOW_ATTR_KEY]) {
1332 		ovs_match_init(&match, &key, true, NULL);
1333 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1334 					NULL, log);
1335 		if (unlikely(err))
1336 			return err;
1337 	}
1338 
1339 	ovs_lock();
1340 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1341 	if (unlikely(!dp)) {
1342 		err = -ENODEV;
1343 		goto unlock;
1344 	}
1345 
1346 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1347 		err = ovs_flow_tbl_flush(&dp->table);
1348 		goto unlock;
1349 	}
1350 
1351 	if (ufid_present)
1352 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1353 	else
1354 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1355 	if (unlikely(!flow)) {
1356 		err = -ENOENT;
1357 		goto unlock;
1358 	}
1359 
1360 	ovs_flow_tbl_remove(&dp->table, flow);
1361 	ovs_unlock();
1362 
1363 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1364 					&flow->id, info, false, ufid_flags);
1365 	if (likely(reply)) {
1366 		if (!IS_ERR(reply)) {
1367 			rcu_read_lock();	/*To keep RCU checker happy. */
1368 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1369 						     reply, info->snd_portid,
1370 						     info->snd_seq, 0,
1371 						     OVS_FLOW_CMD_DEL,
1372 						     ufid_flags);
1373 			rcu_read_unlock();
1374 			BUG_ON(err < 0);
1375 
1376 			ovs_notify(&dp_flow_genl_family, reply, info);
1377 		} else {
1378 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1379 		}
1380 	}
1381 
1382 	ovs_flow_free(flow, true);
1383 	return 0;
1384 unlock:
1385 	ovs_unlock();
1386 	return err;
1387 }
1388 
1389 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1390 {
1391 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1392 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1393 	struct table_instance *ti;
1394 	struct datapath *dp;
1395 	u32 ufid_flags;
1396 	int err;
1397 
1398 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1399 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1400 	if (err)
1401 		return err;
1402 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1403 
1404 	rcu_read_lock();
1405 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1406 	if (!dp) {
1407 		rcu_read_unlock();
1408 		return -ENODEV;
1409 	}
1410 
1411 	ti = rcu_dereference(dp->table.ti);
1412 	for (;;) {
1413 		struct sw_flow *flow;
1414 		u32 bucket, obj;
1415 
1416 		bucket = cb->args[0];
1417 		obj = cb->args[1];
1418 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1419 		if (!flow)
1420 			break;
1421 
1422 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1423 					   NETLINK_CB(cb->skb).portid,
1424 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1425 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1426 			break;
1427 
1428 		cb->args[0] = bucket;
1429 		cb->args[1] = obj;
1430 	}
1431 	rcu_read_unlock();
1432 	return skb->len;
1433 }
1434 
1435 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1436 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1437 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1438 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1439 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1440 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1441 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1442 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1443 };
1444 
1445 static const struct genl_ops dp_flow_genl_ops[] = {
1446 	{ .cmd = OVS_FLOW_CMD_NEW,
1447 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1448 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1449 	  .doit = ovs_flow_cmd_new
1450 	},
1451 	{ .cmd = OVS_FLOW_CMD_DEL,
1452 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1453 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1454 	  .doit = ovs_flow_cmd_del
1455 	},
1456 	{ .cmd = OVS_FLOW_CMD_GET,
1457 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1458 	  .flags = 0,		    /* OK for unprivileged users. */
1459 	  .doit = ovs_flow_cmd_get,
1460 	  .dumpit = ovs_flow_cmd_dump
1461 	},
1462 	{ .cmd = OVS_FLOW_CMD_SET,
1463 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1464 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1465 	  .doit = ovs_flow_cmd_set,
1466 	},
1467 };
1468 
1469 static struct genl_family dp_flow_genl_family __ro_after_init = {
1470 	.hdrsize = sizeof(struct ovs_header),
1471 	.name = OVS_FLOW_FAMILY,
1472 	.version = OVS_FLOW_VERSION,
1473 	.maxattr = OVS_FLOW_ATTR_MAX,
1474 	.policy = flow_policy,
1475 	.netnsok = true,
1476 	.parallel_ops = true,
1477 	.ops = dp_flow_genl_ops,
1478 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1479 	.mcgrps = &ovs_dp_flow_multicast_group,
1480 	.n_mcgrps = 1,
1481 	.module = THIS_MODULE,
1482 };
1483 
1484 static size_t ovs_dp_cmd_msg_size(void)
1485 {
1486 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1487 
1488 	msgsize += nla_total_size(IFNAMSIZ);
1489 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1490 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1491 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1492 
1493 	return msgsize;
1494 }
1495 
1496 /* Called with ovs_mutex. */
1497 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1498 				u32 portid, u32 seq, u32 flags, u8 cmd)
1499 {
1500 	struct ovs_header *ovs_header;
1501 	struct ovs_dp_stats dp_stats;
1502 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1503 	int err;
1504 
1505 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1506 				   flags, cmd);
1507 	if (!ovs_header)
1508 		goto error;
1509 
1510 	ovs_header->dp_ifindex = get_dpifindex(dp);
1511 
1512 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1513 	if (err)
1514 		goto nla_put_failure;
1515 
1516 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1517 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1518 			  &dp_stats, OVS_DP_ATTR_PAD))
1519 		goto nla_put_failure;
1520 
1521 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1522 			  sizeof(struct ovs_dp_megaflow_stats),
1523 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1524 		goto nla_put_failure;
1525 
1526 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1527 		goto nla_put_failure;
1528 
1529 	genlmsg_end(skb, ovs_header);
1530 	return 0;
1531 
1532 nla_put_failure:
1533 	genlmsg_cancel(skb, ovs_header);
1534 error:
1535 	return -EMSGSIZE;
1536 }
1537 
1538 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1539 {
1540 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1541 }
1542 
1543 /* Called with rcu_read_lock or ovs_mutex. */
1544 static struct datapath *lookup_datapath(struct net *net,
1545 					const struct ovs_header *ovs_header,
1546 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1547 {
1548 	struct datapath *dp;
1549 
1550 	if (!a[OVS_DP_ATTR_NAME])
1551 		dp = get_dp(net, ovs_header->dp_ifindex);
1552 	else {
1553 		struct vport *vport;
1554 
1555 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1556 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1557 	}
1558 	return dp ? dp : ERR_PTR(-ENODEV);
1559 }
1560 
1561 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1562 {
1563 	struct datapath *dp;
1564 
1565 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1566 	if (IS_ERR(dp))
1567 		return;
1568 
1569 	WARN(dp->user_features, "Dropping previously announced user features\n");
1570 	dp->user_features = 0;
1571 }
1572 
1573 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1574 
1575 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1576 {
1577 	u32 user_features = 0;
1578 
1579 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1580 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1581 
1582 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1583 				      OVS_DP_F_UNALIGNED |
1584 				      OVS_DP_F_TC_RECIRC_SHARING))
1585 			return -EOPNOTSUPP;
1586 
1587 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1588 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1589 			return -EOPNOTSUPP;
1590 #endif
1591 	}
1592 
1593 	dp->user_features = user_features;
1594 
1595 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1596 		static_branch_enable(&tc_recirc_sharing_support);
1597 	else
1598 		static_branch_disable(&tc_recirc_sharing_support);
1599 
1600 	return 0;
1601 }
1602 
1603 static int ovs_dp_stats_init(struct datapath *dp)
1604 {
1605 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1606 	if (!dp->stats_percpu)
1607 		return -ENOMEM;
1608 
1609 	return 0;
1610 }
1611 
1612 static int ovs_dp_vport_init(struct datapath *dp)
1613 {
1614 	int i;
1615 
1616 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1617 				  sizeof(struct hlist_head),
1618 				  GFP_KERNEL);
1619 	if (!dp->ports)
1620 		return -ENOMEM;
1621 
1622 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1623 		INIT_HLIST_HEAD(&dp->ports[i]);
1624 
1625 	return 0;
1626 }
1627 
1628 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1629 {
1630 	struct nlattr **a = info->attrs;
1631 	struct vport_parms parms;
1632 	struct sk_buff *reply;
1633 	struct datapath *dp;
1634 	struct vport *vport;
1635 	struct ovs_net *ovs_net;
1636 	int err;
1637 
1638 	err = -EINVAL;
1639 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1640 		goto err;
1641 
1642 	reply = ovs_dp_cmd_alloc_info();
1643 	if (!reply)
1644 		return -ENOMEM;
1645 
1646 	err = -ENOMEM;
1647 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1648 	if (dp == NULL)
1649 		goto err_destroy_reply;
1650 
1651 	ovs_dp_set_net(dp, sock_net(skb->sk));
1652 
1653 	/* Allocate table. */
1654 	err = ovs_flow_tbl_init(&dp->table);
1655 	if (err)
1656 		goto err_destroy_dp;
1657 
1658 	err = ovs_dp_stats_init(dp);
1659 	if (err)
1660 		goto err_destroy_table;
1661 
1662 	err = ovs_dp_vport_init(dp);
1663 	if (err)
1664 		goto err_destroy_stats;
1665 
1666 	err = ovs_meters_init(dp);
1667 	if (err)
1668 		goto err_destroy_ports;
1669 
1670 	/* Set up our datapath device. */
1671 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1672 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1673 	parms.options = NULL;
1674 	parms.dp = dp;
1675 	parms.port_no = OVSP_LOCAL;
1676 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1677 
1678 	err = ovs_dp_change(dp, a);
1679 	if (err)
1680 		goto err_destroy_meters;
1681 
1682 	/* So far only local changes have been made, now need the lock. */
1683 	ovs_lock();
1684 
1685 	vport = new_vport(&parms);
1686 	if (IS_ERR(vport)) {
1687 		err = PTR_ERR(vport);
1688 		if (err == -EBUSY)
1689 			err = -EEXIST;
1690 
1691 		if (err == -EEXIST) {
1692 			/* An outdated user space instance that does not understand
1693 			 * the concept of user_features has attempted to create a new
1694 			 * datapath and is likely to reuse it. Drop all user features.
1695 			 */
1696 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1697 				ovs_dp_reset_user_features(skb, info);
1698 		}
1699 
1700 		ovs_unlock();
1701 		goto err_destroy_meters;
1702 	}
1703 
1704 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1705 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1706 	BUG_ON(err < 0);
1707 
1708 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1709 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1710 
1711 	ovs_unlock();
1712 
1713 	ovs_notify(&dp_datapath_genl_family, reply, info);
1714 	return 0;
1715 
1716 err_destroy_meters:
1717 	ovs_meters_exit(dp);
1718 err_destroy_ports:
1719 	kfree(dp->ports);
1720 err_destroy_stats:
1721 	free_percpu(dp->stats_percpu);
1722 err_destroy_table:
1723 	ovs_flow_tbl_destroy(&dp->table);
1724 err_destroy_dp:
1725 	kfree(dp);
1726 err_destroy_reply:
1727 	kfree_skb(reply);
1728 err:
1729 	return err;
1730 }
1731 
1732 /* Called with ovs_mutex. */
1733 static void __dp_destroy(struct datapath *dp)
1734 {
1735 	int i;
1736 
1737 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1738 		struct vport *vport;
1739 		struct hlist_node *n;
1740 
1741 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1742 			if (vport->port_no != OVSP_LOCAL)
1743 				ovs_dp_detach_port(vport);
1744 	}
1745 
1746 	list_del_rcu(&dp->list_node);
1747 
1748 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1749 	 * all ports in datapath are destroyed first before freeing datapath.
1750 	 */
1751 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1752 
1753 	/* RCU destroy the flow table */
1754 	call_rcu(&dp->rcu, destroy_dp_rcu);
1755 }
1756 
1757 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1758 {
1759 	struct sk_buff *reply;
1760 	struct datapath *dp;
1761 	int err;
1762 
1763 	reply = ovs_dp_cmd_alloc_info();
1764 	if (!reply)
1765 		return -ENOMEM;
1766 
1767 	ovs_lock();
1768 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1769 	err = PTR_ERR(dp);
1770 	if (IS_ERR(dp))
1771 		goto err_unlock_free;
1772 
1773 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1774 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1775 	BUG_ON(err < 0);
1776 
1777 	__dp_destroy(dp);
1778 	ovs_unlock();
1779 
1780 	ovs_notify(&dp_datapath_genl_family, reply, info);
1781 
1782 	return 0;
1783 
1784 err_unlock_free:
1785 	ovs_unlock();
1786 	kfree_skb(reply);
1787 	return err;
1788 }
1789 
1790 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1791 {
1792 	struct sk_buff *reply;
1793 	struct datapath *dp;
1794 	int err;
1795 
1796 	reply = ovs_dp_cmd_alloc_info();
1797 	if (!reply)
1798 		return -ENOMEM;
1799 
1800 	ovs_lock();
1801 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1802 	err = PTR_ERR(dp);
1803 	if (IS_ERR(dp))
1804 		goto err_unlock_free;
1805 
1806 	err = ovs_dp_change(dp, info->attrs);
1807 	if (err)
1808 		goto err_unlock_free;
1809 
1810 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1811 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1812 	BUG_ON(err < 0);
1813 
1814 	ovs_unlock();
1815 	ovs_notify(&dp_datapath_genl_family, reply, info);
1816 
1817 	return 0;
1818 
1819 err_unlock_free:
1820 	ovs_unlock();
1821 	kfree_skb(reply);
1822 	return err;
1823 }
1824 
1825 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1826 {
1827 	struct sk_buff *reply;
1828 	struct datapath *dp;
1829 	int err;
1830 
1831 	reply = ovs_dp_cmd_alloc_info();
1832 	if (!reply)
1833 		return -ENOMEM;
1834 
1835 	ovs_lock();
1836 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1837 	if (IS_ERR(dp)) {
1838 		err = PTR_ERR(dp);
1839 		goto err_unlock_free;
1840 	}
1841 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1842 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1843 	BUG_ON(err < 0);
1844 	ovs_unlock();
1845 
1846 	return genlmsg_reply(reply, info);
1847 
1848 err_unlock_free:
1849 	ovs_unlock();
1850 	kfree_skb(reply);
1851 	return err;
1852 }
1853 
1854 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1855 {
1856 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1857 	struct datapath *dp;
1858 	int skip = cb->args[0];
1859 	int i = 0;
1860 
1861 	ovs_lock();
1862 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1863 		if (i >= skip &&
1864 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1865 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1866 					 OVS_DP_CMD_GET) < 0)
1867 			break;
1868 		i++;
1869 	}
1870 	ovs_unlock();
1871 
1872 	cb->args[0] = i;
1873 
1874 	return skb->len;
1875 }
1876 
1877 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1878 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1879 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1880 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1881 };
1882 
1883 static const struct genl_ops dp_datapath_genl_ops[] = {
1884 	{ .cmd = OVS_DP_CMD_NEW,
1885 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1886 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1887 	  .doit = ovs_dp_cmd_new
1888 	},
1889 	{ .cmd = OVS_DP_CMD_DEL,
1890 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1891 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1892 	  .doit = ovs_dp_cmd_del
1893 	},
1894 	{ .cmd = OVS_DP_CMD_GET,
1895 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1896 	  .flags = 0,		    /* OK for unprivileged users. */
1897 	  .doit = ovs_dp_cmd_get,
1898 	  .dumpit = ovs_dp_cmd_dump
1899 	},
1900 	{ .cmd = OVS_DP_CMD_SET,
1901 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1902 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1903 	  .doit = ovs_dp_cmd_set,
1904 	},
1905 };
1906 
1907 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1908 	.hdrsize = sizeof(struct ovs_header),
1909 	.name = OVS_DATAPATH_FAMILY,
1910 	.version = OVS_DATAPATH_VERSION,
1911 	.maxattr = OVS_DP_ATTR_MAX,
1912 	.policy = datapath_policy,
1913 	.netnsok = true,
1914 	.parallel_ops = true,
1915 	.ops = dp_datapath_genl_ops,
1916 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1917 	.mcgrps = &ovs_dp_datapath_multicast_group,
1918 	.n_mcgrps = 1,
1919 	.module = THIS_MODULE,
1920 };
1921 
1922 /* Called with ovs_mutex or RCU read lock. */
1923 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1924 				   struct net *net, u32 portid, u32 seq,
1925 				   u32 flags, u8 cmd, gfp_t gfp)
1926 {
1927 	struct ovs_header *ovs_header;
1928 	struct ovs_vport_stats vport_stats;
1929 	int err;
1930 
1931 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1932 				 flags, cmd);
1933 	if (!ovs_header)
1934 		return -EMSGSIZE;
1935 
1936 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1937 
1938 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1939 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1940 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1941 			   ovs_vport_name(vport)) ||
1942 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1943 		goto nla_put_failure;
1944 
1945 	if (!net_eq(net, dev_net(vport->dev))) {
1946 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1947 
1948 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1949 			goto nla_put_failure;
1950 	}
1951 
1952 	ovs_vport_get_stats(vport, &vport_stats);
1953 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1954 			  sizeof(struct ovs_vport_stats), &vport_stats,
1955 			  OVS_VPORT_ATTR_PAD))
1956 		goto nla_put_failure;
1957 
1958 	if (ovs_vport_get_upcall_portids(vport, skb))
1959 		goto nla_put_failure;
1960 
1961 	err = ovs_vport_get_options(vport, skb);
1962 	if (err == -EMSGSIZE)
1963 		goto error;
1964 
1965 	genlmsg_end(skb, ovs_header);
1966 	return 0;
1967 
1968 nla_put_failure:
1969 	err = -EMSGSIZE;
1970 error:
1971 	genlmsg_cancel(skb, ovs_header);
1972 	return err;
1973 }
1974 
1975 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1976 {
1977 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1978 }
1979 
1980 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1981 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1982 					 u32 portid, u32 seq, u8 cmd)
1983 {
1984 	struct sk_buff *skb;
1985 	int retval;
1986 
1987 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1988 	if (!skb)
1989 		return ERR_PTR(-ENOMEM);
1990 
1991 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1992 					 GFP_KERNEL);
1993 	BUG_ON(retval < 0);
1994 
1995 	return skb;
1996 }
1997 
1998 /* Called with ovs_mutex or RCU read lock. */
1999 static struct vport *lookup_vport(struct net *net,
2000 				  const struct ovs_header *ovs_header,
2001 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2002 {
2003 	struct datapath *dp;
2004 	struct vport *vport;
2005 
2006 	if (a[OVS_VPORT_ATTR_IFINDEX])
2007 		return ERR_PTR(-EOPNOTSUPP);
2008 	if (a[OVS_VPORT_ATTR_NAME]) {
2009 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2010 		if (!vport)
2011 			return ERR_PTR(-ENODEV);
2012 		if (ovs_header->dp_ifindex &&
2013 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2014 			return ERR_PTR(-ENODEV);
2015 		return vport;
2016 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2017 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2018 
2019 		if (port_no >= DP_MAX_PORTS)
2020 			return ERR_PTR(-EFBIG);
2021 
2022 		dp = get_dp(net, ovs_header->dp_ifindex);
2023 		if (!dp)
2024 			return ERR_PTR(-ENODEV);
2025 
2026 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2027 		if (!vport)
2028 			return ERR_PTR(-ENODEV);
2029 		return vport;
2030 	} else
2031 		return ERR_PTR(-EINVAL);
2032 
2033 }
2034 
2035 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2036 {
2037 	unsigned int dev_headroom, max_headroom = 0;
2038 	struct net_device *dev;
2039 	struct vport *vport;
2040 	int i;
2041 
2042 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2043 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2044 			dev = vport->dev;
2045 			dev_headroom = netdev_get_fwd_headroom(dev);
2046 			if (dev_headroom > max_headroom)
2047 				max_headroom = dev_headroom;
2048 		}
2049 	}
2050 
2051 	return max_headroom;
2052 }
2053 
2054 /* Called with ovs_mutex */
2055 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2056 {
2057 	struct vport *vport;
2058 	int i;
2059 
2060 	dp->max_headroom = new_headroom;
2061 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2062 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2063 			netdev_set_rx_headroom(vport->dev, new_headroom);
2064 }
2065 
2066 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2067 {
2068 	struct nlattr **a = info->attrs;
2069 	struct ovs_header *ovs_header = info->userhdr;
2070 	struct vport_parms parms;
2071 	struct sk_buff *reply;
2072 	struct vport *vport;
2073 	struct datapath *dp;
2074 	unsigned int new_headroom;
2075 	u32 port_no;
2076 	int err;
2077 
2078 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2079 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2080 		return -EINVAL;
2081 	if (a[OVS_VPORT_ATTR_IFINDEX])
2082 		return -EOPNOTSUPP;
2083 
2084 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2085 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2086 	if (port_no >= DP_MAX_PORTS)
2087 		return -EFBIG;
2088 
2089 	reply = ovs_vport_cmd_alloc_info();
2090 	if (!reply)
2091 		return -ENOMEM;
2092 
2093 	ovs_lock();
2094 restart:
2095 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2096 	err = -ENODEV;
2097 	if (!dp)
2098 		goto exit_unlock_free;
2099 
2100 	if (port_no) {
2101 		vport = ovs_vport_ovsl(dp, port_no);
2102 		err = -EBUSY;
2103 		if (vport)
2104 			goto exit_unlock_free;
2105 	} else {
2106 		for (port_no = 1; ; port_no++) {
2107 			if (port_no >= DP_MAX_PORTS) {
2108 				err = -EFBIG;
2109 				goto exit_unlock_free;
2110 			}
2111 			vport = ovs_vport_ovsl(dp, port_no);
2112 			if (!vport)
2113 				break;
2114 		}
2115 	}
2116 
2117 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2118 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2119 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2120 	parms.dp = dp;
2121 	parms.port_no = port_no;
2122 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2123 
2124 	vport = new_vport(&parms);
2125 	err = PTR_ERR(vport);
2126 	if (IS_ERR(vport)) {
2127 		if (err == -EAGAIN)
2128 			goto restart;
2129 		goto exit_unlock_free;
2130 	}
2131 
2132 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2133 				      info->snd_portid, info->snd_seq, 0,
2134 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2135 
2136 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2137 
2138 	if (new_headroom > dp->max_headroom)
2139 		ovs_update_headroom(dp, new_headroom);
2140 	else
2141 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2142 
2143 	BUG_ON(err < 0);
2144 	ovs_unlock();
2145 
2146 	ovs_notify(&dp_vport_genl_family, reply, info);
2147 	return 0;
2148 
2149 exit_unlock_free:
2150 	ovs_unlock();
2151 	kfree_skb(reply);
2152 	return err;
2153 }
2154 
2155 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2156 {
2157 	struct nlattr **a = info->attrs;
2158 	struct sk_buff *reply;
2159 	struct vport *vport;
2160 	int err;
2161 
2162 	reply = ovs_vport_cmd_alloc_info();
2163 	if (!reply)
2164 		return -ENOMEM;
2165 
2166 	ovs_lock();
2167 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2168 	err = PTR_ERR(vport);
2169 	if (IS_ERR(vport))
2170 		goto exit_unlock_free;
2171 
2172 	if (a[OVS_VPORT_ATTR_TYPE] &&
2173 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2174 		err = -EINVAL;
2175 		goto exit_unlock_free;
2176 	}
2177 
2178 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2179 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2180 		if (err)
2181 			goto exit_unlock_free;
2182 	}
2183 
2184 
2185 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2186 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2187 
2188 		err = ovs_vport_set_upcall_portids(vport, ids);
2189 		if (err)
2190 			goto exit_unlock_free;
2191 	}
2192 
2193 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2194 				      info->snd_portid, info->snd_seq, 0,
2195 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2196 	BUG_ON(err < 0);
2197 
2198 	ovs_unlock();
2199 	ovs_notify(&dp_vport_genl_family, reply, info);
2200 	return 0;
2201 
2202 exit_unlock_free:
2203 	ovs_unlock();
2204 	kfree_skb(reply);
2205 	return err;
2206 }
2207 
2208 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2209 {
2210 	bool update_headroom = false;
2211 	struct nlattr **a = info->attrs;
2212 	struct sk_buff *reply;
2213 	struct datapath *dp;
2214 	struct vport *vport;
2215 	unsigned int new_headroom;
2216 	int err;
2217 
2218 	reply = ovs_vport_cmd_alloc_info();
2219 	if (!reply)
2220 		return -ENOMEM;
2221 
2222 	ovs_lock();
2223 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2224 	err = PTR_ERR(vport);
2225 	if (IS_ERR(vport))
2226 		goto exit_unlock_free;
2227 
2228 	if (vport->port_no == OVSP_LOCAL) {
2229 		err = -EINVAL;
2230 		goto exit_unlock_free;
2231 	}
2232 
2233 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2234 				      info->snd_portid, info->snd_seq, 0,
2235 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2236 	BUG_ON(err < 0);
2237 
2238 	/* the vport deletion may trigger dp headroom update */
2239 	dp = vport->dp;
2240 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2241 		update_headroom = true;
2242 
2243 	netdev_reset_rx_headroom(vport->dev);
2244 	ovs_dp_detach_port(vport);
2245 
2246 	if (update_headroom) {
2247 		new_headroom = ovs_get_max_headroom(dp);
2248 
2249 		if (new_headroom < dp->max_headroom)
2250 			ovs_update_headroom(dp, new_headroom);
2251 	}
2252 	ovs_unlock();
2253 
2254 	ovs_notify(&dp_vport_genl_family, reply, info);
2255 	return 0;
2256 
2257 exit_unlock_free:
2258 	ovs_unlock();
2259 	kfree_skb(reply);
2260 	return err;
2261 }
2262 
2263 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2264 {
2265 	struct nlattr **a = info->attrs;
2266 	struct ovs_header *ovs_header = info->userhdr;
2267 	struct sk_buff *reply;
2268 	struct vport *vport;
2269 	int err;
2270 
2271 	reply = ovs_vport_cmd_alloc_info();
2272 	if (!reply)
2273 		return -ENOMEM;
2274 
2275 	rcu_read_lock();
2276 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2277 	err = PTR_ERR(vport);
2278 	if (IS_ERR(vport))
2279 		goto exit_unlock_free;
2280 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2281 				      info->snd_portid, info->snd_seq, 0,
2282 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2283 	BUG_ON(err < 0);
2284 	rcu_read_unlock();
2285 
2286 	return genlmsg_reply(reply, info);
2287 
2288 exit_unlock_free:
2289 	rcu_read_unlock();
2290 	kfree_skb(reply);
2291 	return err;
2292 }
2293 
2294 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2295 {
2296 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2297 	struct datapath *dp;
2298 	int bucket = cb->args[0], skip = cb->args[1];
2299 	int i, j = 0;
2300 
2301 	rcu_read_lock();
2302 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2303 	if (!dp) {
2304 		rcu_read_unlock();
2305 		return -ENODEV;
2306 	}
2307 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2308 		struct vport *vport;
2309 
2310 		j = 0;
2311 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2312 			if (j >= skip &&
2313 			    ovs_vport_cmd_fill_info(vport, skb,
2314 						    sock_net(skb->sk),
2315 						    NETLINK_CB(cb->skb).portid,
2316 						    cb->nlh->nlmsg_seq,
2317 						    NLM_F_MULTI,
2318 						    OVS_VPORT_CMD_GET,
2319 						    GFP_ATOMIC) < 0)
2320 				goto out;
2321 
2322 			j++;
2323 		}
2324 		skip = 0;
2325 	}
2326 out:
2327 	rcu_read_unlock();
2328 
2329 	cb->args[0] = i;
2330 	cb->args[1] = j;
2331 
2332 	return skb->len;
2333 }
2334 
2335 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2336 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2337 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2338 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2339 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2340 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2341 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2342 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2343 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2344 };
2345 
2346 static const struct genl_ops dp_vport_genl_ops[] = {
2347 	{ .cmd = OVS_VPORT_CMD_NEW,
2348 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2349 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2350 	  .doit = ovs_vport_cmd_new
2351 	},
2352 	{ .cmd = OVS_VPORT_CMD_DEL,
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_del
2356 	},
2357 	{ .cmd = OVS_VPORT_CMD_GET,
2358 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2359 	  .flags = 0,		    /* OK for unprivileged users. */
2360 	  .doit = ovs_vport_cmd_get,
2361 	  .dumpit = ovs_vport_cmd_dump
2362 	},
2363 	{ .cmd = OVS_VPORT_CMD_SET,
2364 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2365 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2366 	  .doit = ovs_vport_cmd_set,
2367 	},
2368 };
2369 
2370 struct genl_family dp_vport_genl_family __ro_after_init = {
2371 	.hdrsize = sizeof(struct ovs_header),
2372 	.name = OVS_VPORT_FAMILY,
2373 	.version = OVS_VPORT_VERSION,
2374 	.maxattr = OVS_VPORT_ATTR_MAX,
2375 	.policy = vport_policy,
2376 	.netnsok = true,
2377 	.parallel_ops = true,
2378 	.ops = dp_vport_genl_ops,
2379 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2380 	.mcgrps = &ovs_dp_vport_multicast_group,
2381 	.n_mcgrps = 1,
2382 	.module = THIS_MODULE,
2383 };
2384 
2385 static struct genl_family * const dp_genl_families[] = {
2386 	&dp_datapath_genl_family,
2387 	&dp_vport_genl_family,
2388 	&dp_flow_genl_family,
2389 	&dp_packet_genl_family,
2390 	&dp_meter_genl_family,
2391 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2392 	&dp_ct_limit_genl_family,
2393 #endif
2394 };
2395 
2396 static void dp_unregister_genl(int n_families)
2397 {
2398 	int i;
2399 
2400 	for (i = 0; i < n_families; i++)
2401 		genl_unregister_family(dp_genl_families[i]);
2402 }
2403 
2404 static int __init dp_register_genl(void)
2405 {
2406 	int err;
2407 	int i;
2408 
2409 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2410 
2411 		err = genl_register_family(dp_genl_families[i]);
2412 		if (err)
2413 			goto error;
2414 	}
2415 
2416 	return 0;
2417 
2418 error:
2419 	dp_unregister_genl(i);
2420 	return err;
2421 }
2422 
2423 static int __net_init ovs_init_net(struct net *net)
2424 {
2425 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2426 
2427 	INIT_LIST_HEAD(&ovs_net->dps);
2428 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2429 	return ovs_ct_init(net);
2430 }
2431 
2432 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2433 					    struct list_head *head)
2434 {
2435 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2436 	struct datapath *dp;
2437 
2438 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2439 		int i;
2440 
2441 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2442 			struct vport *vport;
2443 
2444 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2445 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2446 					continue;
2447 
2448 				if (dev_net(vport->dev) == dnet)
2449 					list_add(&vport->detach_list, head);
2450 			}
2451 		}
2452 	}
2453 }
2454 
2455 static void __net_exit ovs_exit_net(struct net *dnet)
2456 {
2457 	struct datapath *dp, *dp_next;
2458 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2459 	struct vport *vport, *vport_next;
2460 	struct net *net;
2461 	LIST_HEAD(head);
2462 
2463 	ovs_ct_exit(dnet);
2464 	ovs_lock();
2465 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2466 		__dp_destroy(dp);
2467 
2468 	down_read(&net_rwsem);
2469 	for_each_net(net)
2470 		list_vports_from_net(net, dnet, &head);
2471 	up_read(&net_rwsem);
2472 
2473 	/* Detach all vports from given namespace. */
2474 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2475 		list_del(&vport->detach_list);
2476 		ovs_dp_detach_port(vport);
2477 	}
2478 
2479 	ovs_unlock();
2480 
2481 	cancel_work_sync(&ovs_net->dp_notify_work);
2482 }
2483 
2484 static struct pernet_operations ovs_net_ops = {
2485 	.init = ovs_init_net,
2486 	.exit = ovs_exit_net,
2487 	.id   = &ovs_net_id,
2488 	.size = sizeof(struct ovs_net),
2489 };
2490 
2491 static int __init dp_init(void)
2492 {
2493 	int err;
2494 
2495 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2496 
2497 	pr_info("Open vSwitch switching datapath\n");
2498 
2499 	err = action_fifos_init();
2500 	if (err)
2501 		goto error;
2502 
2503 	err = ovs_internal_dev_rtnl_link_register();
2504 	if (err)
2505 		goto error_action_fifos_exit;
2506 
2507 	err = ovs_flow_init();
2508 	if (err)
2509 		goto error_unreg_rtnl_link;
2510 
2511 	err = ovs_vport_init();
2512 	if (err)
2513 		goto error_flow_exit;
2514 
2515 	err = register_pernet_device(&ovs_net_ops);
2516 	if (err)
2517 		goto error_vport_exit;
2518 
2519 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2520 	if (err)
2521 		goto error_netns_exit;
2522 
2523 	err = ovs_netdev_init();
2524 	if (err)
2525 		goto error_unreg_notifier;
2526 
2527 	err = dp_register_genl();
2528 	if (err < 0)
2529 		goto error_unreg_netdev;
2530 
2531 	return 0;
2532 
2533 error_unreg_netdev:
2534 	ovs_netdev_exit();
2535 error_unreg_notifier:
2536 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2537 error_netns_exit:
2538 	unregister_pernet_device(&ovs_net_ops);
2539 error_vport_exit:
2540 	ovs_vport_exit();
2541 error_flow_exit:
2542 	ovs_flow_exit();
2543 error_unreg_rtnl_link:
2544 	ovs_internal_dev_rtnl_link_unregister();
2545 error_action_fifos_exit:
2546 	action_fifos_exit();
2547 error:
2548 	return err;
2549 }
2550 
2551 static void dp_cleanup(void)
2552 {
2553 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2554 	ovs_netdev_exit();
2555 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2556 	unregister_pernet_device(&ovs_net_ops);
2557 	rcu_barrier();
2558 	ovs_vport_exit();
2559 	ovs_flow_exit();
2560 	ovs_internal_dev_rtnl_link_unregister();
2561 	action_fifos_exit();
2562 }
2563 
2564 module_init(dp_init);
2565 module_exit(dp_cleanup);
2566 
2567 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2568 MODULE_LICENSE("GPL");
2569 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2570 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2571 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2572 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2573 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2574 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2575