xref: /linux/net/openvswitch/datapath.c (revision 1ee54195a305fae3955642af8528bdf67496d353)
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 	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
492 		err = -ENOBUFS;
493 		goto out;
494 	}
495 
496 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
497 	if (cutlen > 0 &&
498 	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
499 		err = -ENOBUFS;
500 		goto out;
501 	}
502 
503 	/* Add OVS_PACKET_ATTR_HASH */
504 	hash = skb_get_hash_raw(skb);
505 	if (skb->sw_hash)
506 		hash |= OVS_PACKET_HASH_SW_BIT;
507 
508 	if (skb->l4_hash)
509 		hash |= OVS_PACKET_HASH_L4_BIT;
510 
511 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
512 		err = -ENOBUFS;
513 		goto out;
514 	}
515 
516 	/* Only reserve room for attribute header, packet data is added
517 	 * in skb_zerocopy() */
518 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
519 		err = -ENOBUFS;
520 		goto out;
521 	}
522 	nla->nla_len = nla_attr_size(skb->len - cutlen);
523 
524 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
525 	if (err)
526 		goto out;
527 
528 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
529 	pad_packet(dp, user_skb);
530 
531 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
532 
533 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
534 	user_skb = NULL;
535 out:
536 	if (err)
537 		skb_tx_error(skb);
538 	kfree_skb(user_skb);
539 	kfree_skb(nskb);
540 	return err;
541 }
542 
543 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
544 {
545 	struct ovs_header *ovs_header = info->userhdr;
546 	struct net *net = sock_net(skb->sk);
547 	struct nlattr **a = info->attrs;
548 	struct sw_flow_actions *acts;
549 	struct sk_buff *packet;
550 	struct sw_flow *flow;
551 	struct sw_flow_actions *sf_acts;
552 	struct datapath *dp;
553 	struct vport *input_vport;
554 	u16 mru = 0;
555 	u64 hash;
556 	int len;
557 	int err;
558 	bool log = !a[OVS_PACKET_ATTR_PROBE];
559 
560 	err = -EINVAL;
561 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
562 	    !a[OVS_PACKET_ATTR_ACTIONS])
563 		goto err;
564 
565 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
566 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
567 	err = -ENOMEM;
568 	if (!packet)
569 		goto err;
570 	skb_reserve(packet, NET_IP_ALIGN);
571 
572 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
573 
574 	/* Set packet's mru */
575 	if (a[OVS_PACKET_ATTR_MRU]) {
576 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
577 		packet->ignore_df = 1;
578 	}
579 	OVS_CB(packet)->mru = mru;
580 
581 	if (a[OVS_PACKET_ATTR_HASH]) {
582 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
583 
584 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
585 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
586 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
587 	}
588 
589 	/* Build an sw_flow for sending this packet. */
590 	flow = ovs_flow_alloc();
591 	err = PTR_ERR(flow);
592 	if (IS_ERR(flow))
593 		goto err_kfree_skb;
594 
595 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
596 					     packet, &flow->key, log);
597 	if (err)
598 		goto err_flow_free;
599 
600 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
601 				   &flow->key, &acts, log);
602 	if (err)
603 		goto err_flow_free;
604 
605 	rcu_assign_pointer(flow->sf_acts, acts);
606 	packet->priority = flow->key.phy.priority;
607 	packet->mark = flow->key.phy.skb_mark;
608 
609 	rcu_read_lock();
610 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
611 	err = -ENODEV;
612 	if (!dp)
613 		goto err_unlock;
614 
615 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
616 	if (!input_vport)
617 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
618 
619 	if (!input_vport)
620 		goto err_unlock;
621 
622 	packet->dev = input_vport->dev;
623 	OVS_CB(packet)->input_vport = input_vport;
624 	sf_acts = rcu_dereference(flow->sf_acts);
625 
626 	local_bh_disable();
627 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
628 	local_bh_enable();
629 	rcu_read_unlock();
630 
631 	ovs_flow_free(flow, false);
632 	return err;
633 
634 err_unlock:
635 	rcu_read_unlock();
636 err_flow_free:
637 	ovs_flow_free(flow, false);
638 err_kfree_skb:
639 	kfree_skb(packet);
640 err:
641 	return err;
642 }
643 
644 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
645 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
646 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
647 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
648 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
649 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
650 };
651 
652 static const struct genl_ops dp_packet_genl_ops[] = {
653 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
654 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
655 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
656 	  .doit = ovs_packet_cmd_execute
657 	}
658 };
659 
660 static struct genl_family dp_packet_genl_family __ro_after_init = {
661 	.hdrsize = sizeof(struct ovs_header),
662 	.name = OVS_PACKET_FAMILY,
663 	.version = OVS_PACKET_VERSION,
664 	.maxattr = OVS_PACKET_ATTR_MAX,
665 	.policy = packet_policy,
666 	.netnsok = true,
667 	.parallel_ops = true,
668 	.ops = dp_packet_genl_ops,
669 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
670 	.module = THIS_MODULE,
671 };
672 
673 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
674 			 struct ovs_dp_megaflow_stats *mega_stats)
675 {
676 	int i;
677 
678 	memset(mega_stats, 0, sizeof(*mega_stats));
679 
680 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
681 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
682 
683 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
684 
685 	for_each_possible_cpu(i) {
686 		const struct dp_stats_percpu *percpu_stats;
687 		struct dp_stats_percpu local_stats;
688 		unsigned int start;
689 
690 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
691 
692 		do {
693 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
694 			local_stats = *percpu_stats;
695 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
696 
697 		stats->n_hit += local_stats.n_hit;
698 		stats->n_missed += local_stats.n_missed;
699 		stats->n_lost += local_stats.n_lost;
700 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
701 	}
702 }
703 
704 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
705 {
706 	return ovs_identifier_is_ufid(sfid) &&
707 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
708 }
709 
710 static bool should_fill_mask(uint32_t ufid_flags)
711 {
712 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
713 }
714 
715 static bool should_fill_actions(uint32_t ufid_flags)
716 {
717 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
718 }
719 
720 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
721 				    const struct sw_flow_id *sfid,
722 				    uint32_t ufid_flags)
723 {
724 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
725 
726 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
727 	 * see ovs_nla_put_identifier()
728 	 */
729 	if (sfid && ovs_identifier_is_ufid(sfid))
730 		len += nla_total_size(sfid->ufid_len);
731 	else
732 		len += nla_total_size(ovs_key_attr_size());
733 
734 	/* OVS_FLOW_ATTR_KEY */
735 	if (!sfid || should_fill_key(sfid, ufid_flags))
736 		len += nla_total_size(ovs_key_attr_size());
737 
738 	/* OVS_FLOW_ATTR_MASK */
739 	if (should_fill_mask(ufid_flags))
740 		len += nla_total_size(ovs_key_attr_size());
741 
742 	/* OVS_FLOW_ATTR_ACTIONS */
743 	if (should_fill_actions(ufid_flags))
744 		len += nla_total_size(acts->orig_len);
745 
746 	return len
747 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
748 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
749 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
750 }
751 
752 /* Called with ovs_mutex or RCU read lock. */
753 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
754 				   struct sk_buff *skb)
755 {
756 	struct ovs_flow_stats stats;
757 	__be16 tcp_flags;
758 	unsigned long used;
759 
760 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
761 
762 	if (used &&
763 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
764 			      OVS_FLOW_ATTR_PAD))
765 		return -EMSGSIZE;
766 
767 	if (stats.n_packets &&
768 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
769 			  sizeof(struct ovs_flow_stats), &stats,
770 			  OVS_FLOW_ATTR_PAD))
771 		return -EMSGSIZE;
772 
773 	if ((u8)ntohs(tcp_flags) &&
774 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
775 		return -EMSGSIZE;
776 
777 	return 0;
778 }
779 
780 /* Called with ovs_mutex or RCU read lock. */
781 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
782 				     struct sk_buff *skb, int skb_orig_len)
783 {
784 	struct nlattr *start;
785 	int err;
786 
787 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
788 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
789 	 * Netlink but individual action lists can be longer than
790 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
791 	 * The userspace caller can always fetch the actions separately if it
792 	 * really wants them.  (Most userspace callers in fact don't care.)
793 	 *
794 	 * This can only fail for dump operations because the skb is always
795 	 * properly sized for single flows.
796 	 */
797 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
798 	if (start) {
799 		const struct sw_flow_actions *sf_acts;
800 
801 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
802 		err = ovs_nla_put_actions(sf_acts->actions,
803 					  sf_acts->actions_len, skb);
804 
805 		if (!err)
806 			nla_nest_end(skb, start);
807 		else {
808 			if (skb_orig_len)
809 				return err;
810 
811 			nla_nest_cancel(skb, start);
812 		}
813 	} else if (skb_orig_len) {
814 		return -EMSGSIZE;
815 	}
816 
817 	return 0;
818 }
819 
820 /* Called with ovs_mutex or RCU read lock. */
821 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
822 				  struct sk_buff *skb, u32 portid,
823 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
824 {
825 	const int skb_orig_len = skb->len;
826 	struct ovs_header *ovs_header;
827 	int err;
828 
829 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
830 				 flags, cmd);
831 	if (!ovs_header)
832 		return -EMSGSIZE;
833 
834 	ovs_header->dp_ifindex = dp_ifindex;
835 
836 	err = ovs_nla_put_identifier(flow, skb);
837 	if (err)
838 		goto error;
839 
840 	if (should_fill_key(&flow->id, ufid_flags)) {
841 		err = ovs_nla_put_masked_key(flow, skb);
842 		if (err)
843 			goto error;
844 	}
845 
846 	if (should_fill_mask(ufid_flags)) {
847 		err = ovs_nla_put_mask(flow, skb);
848 		if (err)
849 			goto error;
850 	}
851 
852 	err = ovs_flow_cmd_fill_stats(flow, skb);
853 	if (err)
854 		goto error;
855 
856 	if (should_fill_actions(ufid_flags)) {
857 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
858 		if (err)
859 			goto error;
860 	}
861 
862 	genlmsg_end(skb, ovs_header);
863 	return 0;
864 
865 error:
866 	genlmsg_cancel(skb, ovs_header);
867 	return err;
868 }
869 
870 /* May not be called with RCU read lock. */
871 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
872 					       const struct sw_flow_id *sfid,
873 					       struct genl_info *info,
874 					       bool always,
875 					       uint32_t ufid_flags)
876 {
877 	struct sk_buff *skb;
878 	size_t len;
879 
880 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
881 		return NULL;
882 
883 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
884 	skb = genlmsg_new(len, GFP_KERNEL);
885 	if (!skb)
886 		return ERR_PTR(-ENOMEM);
887 
888 	return skb;
889 }
890 
891 /* Called with ovs_mutex. */
892 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
893 					       int dp_ifindex,
894 					       struct genl_info *info, u8 cmd,
895 					       bool always, u32 ufid_flags)
896 {
897 	struct sk_buff *skb;
898 	int retval;
899 
900 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
901 				      &flow->id, info, always, ufid_flags);
902 	if (IS_ERR_OR_NULL(skb))
903 		return skb;
904 
905 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
906 					info->snd_portid, info->snd_seq, 0,
907 					cmd, ufid_flags);
908 	BUG_ON(retval < 0);
909 	return skb;
910 }
911 
912 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
913 {
914 	struct net *net = sock_net(skb->sk);
915 	struct nlattr **a = info->attrs;
916 	struct ovs_header *ovs_header = info->userhdr;
917 	struct sw_flow *flow = NULL, *new_flow;
918 	struct sw_flow_mask mask;
919 	struct sk_buff *reply;
920 	struct datapath *dp;
921 	struct sw_flow_actions *acts;
922 	struct sw_flow_match match;
923 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
924 	int error;
925 	bool log = !a[OVS_FLOW_ATTR_PROBE];
926 
927 	/* Must have key and actions. */
928 	error = -EINVAL;
929 	if (!a[OVS_FLOW_ATTR_KEY]) {
930 		OVS_NLERR(log, "Flow key attr not present in new flow.");
931 		goto error;
932 	}
933 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
934 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
935 		goto error;
936 	}
937 
938 	/* Most of the time we need to allocate a new flow, do it before
939 	 * locking.
940 	 */
941 	new_flow = ovs_flow_alloc();
942 	if (IS_ERR(new_flow)) {
943 		error = PTR_ERR(new_flow);
944 		goto error;
945 	}
946 
947 	/* Extract key. */
948 	ovs_match_init(&match, &new_flow->key, false, &mask);
949 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
950 				  a[OVS_FLOW_ATTR_MASK], log);
951 	if (error)
952 		goto err_kfree_flow;
953 
954 	/* Extract flow identifier. */
955 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
956 				       &new_flow->key, log);
957 	if (error)
958 		goto err_kfree_flow;
959 
960 	/* unmasked key is needed to match when ufid is not used. */
961 	if (ovs_identifier_is_key(&new_flow->id))
962 		match.key = new_flow->id.unmasked_key;
963 
964 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
965 
966 	/* Validate actions. */
967 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
968 				     &new_flow->key, &acts, log);
969 	if (error) {
970 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
971 		goto err_kfree_flow;
972 	}
973 
974 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
975 					ufid_flags);
976 	if (IS_ERR(reply)) {
977 		error = PTR_ERR(reply);
978 		goto err_kfree_acts;
979 	}
980 
981 	ovs_lock();
982 	dp = get_dp(net, ovs_header->dp_ifindex);
983 	if (unlikely(!dp)) {
984 		error = -ENODEV;
985 		goto err_unlock_ovs;
986 	}
987 
988 	/* Check if this is a duplicate flow */
989 	if (ovs_identifier_is_ufid(&new_flow->id))
990 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
991 	if (!flow)
992 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
993 	if (likely(!flow)) {
994 		rcu_assign_pointer(new_flow->sf_acts, acts);
995 
996 		/* Put flow in bucket. */
997 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
998 		if (unlikely(error)) {
999 			acts = NULL;
1000 			goto err_unlock_ovs;
1001 		}
1002 
1003 		if (unlikely(reply)) {
1004 			error = ovs_flow_cmd_fill_info(new_flow,
1005 						       ovs_header->dp_ifindex,
1006 						       reply, info->snd_portid,
1007 						       info->snd_seq, 0,
1008 						       OVS_FLOW_CMD_NEW,
1009 						       ufid_flags);
1010 			BUG_ON(error < 0);
1011 		}
1012 		ovs_unlock();
1013 	} else {
1014 		struct sw_flow_actions *old_acts;
1015 
1016 		/* Bail out if we're not allowed to modify an existing flow.
1017 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1018 		 * because Generic Netlink treats the latter as a dump
1019 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1020 		 * gets fixed.
1021 		 */
1022 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1023 							 | NLM_F_EXCL))) {
1024 			error = -EEXIST;
1025 			goto err_unlock_ovs;
1026 		}
1027 		/* The flow identifier has to be the same for flow updates.
1028 		 * Look for any overlapping flow.
1029 		 */
1030 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1031 			if (ovs_identifier_is_key(&flow->id))
1032 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1033 								 &match);
1034 			else /* UFID matches but key is different */
1035 				flow = NULL;
1036 			if (!flow) {
1037 				error = -ENOENT;
1038 				goto err_unlock_ovs;
1039 			}
1040 		}
1041 		/* Update actions. */
1042 		old_acts = ovsl_dereference(flow->sf_acts);
1043 		rcu_assign_pointer(flow->sf_acts, acts);
1044 
1045 		if (unlikely(reply)) {
1046 			error = ovs_flow_cmd_fill_info(flow,
1047 						       ovs_header->dp_ifindex,
1048 						       reply, info->snd_portid,
1049 						       info->snd_seq, 0,
1050 						       OVS_FLOW_CMD_NEW,
1051 						       ufid_flags);
1052 			BUG_ON(error < 0);
1053 		}
1054 		ovs_unlock();
1055 
1056 		ovs_nla_free_flow_actions_rcu(old_acts);
1057 		ovs_flow_free(new_flow, false);
1058 	}
1059 
1060 	if (reply)
1061 		ovs_notify(&dp_flow_genl_family, reply, info);
1062 	return 0;
1063 
1064 err_unlock_ovs:
1065 	ovs_unlock();
1066 	kfree_skb(reply);
1067 err_kfree_acts:
1068 	ovs_nla_free_flow_actions(acts);
1069 err_kfree_flow:
1070 	ovs_flow_free(new_flow, false);
1071 error:
1072 	return error;
1073 }
1074 
1075 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1076 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1077 						const struct nlattr *a,
1078 						const struct sw_flow_key *key,
1079 						const struct sw_flow_mask *mask,
1080 						bool log)
1081 {
1082 	struct sw_flow_actions *acts;
1083 	struct sw_flow_key masked_key;
1084 	int error;
1085 
1086 	ovs_flow_mask_key(&masked_key, key, true, mask);
1087 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1088 	if (error) {
1089 		OVS_NLERR(log,
1090 			  "Actions may not be safe on all matching packets");
1091 		return ERR_PTR(error);
1092 	}
1093 
1094 	return acts;
1095 }
1096 
1097 /* Factor out match-init and action-copy to avoid
1098  * "Wframe-larger-than=1024" warning. Because mask is only
1099  * used to get actions, we new a function to save some
1100  * stack space.
1101  *
1102  * If there are not key and action attrs, we return 0
1103  * directly. In the case, the caller will also not use the
1104  * match as before. If there is action attr, we try to get
1105  * actions and save them to *acts. Before returning from
1106  * the function, we reset the match->mask pointer. Because
1107  * we should not to return match object with dangling reference
1108  * to mask.
1109  * */
1110 static noinline_for_stack int
1111 ovs_nla_init_match_and_action(struct net *net,
1112 			      struct sw_flow_match *match,
1113 			      struct sw_flow_key *key,
1114 			      struct nlattr **a,
1115 			      struct sw_flow_actions **acts,
1116 			      bool log)
1117 {
1118 	struct sw_flow_mask mask;
1119 	int error = 0;
1120 
1121 	if (a[OVS_FLOW_ATTR_KEY]) {
1122 		ovs_match_init(match, key, true, &mask);
1123 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1124 					  a[OVS_FLOW_ATTR_MASK], log);
1125 		if (error)
1126 			goto error;
1127 	}
1128 
1129 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1130 		if (!a[OVS_FLOW_ATTR_KEY]) {
1131 			OVS_NLERR(log,
1132 				  "Flow key attribute not present in set flow.");
1133 			error = -EINVAL;
1134 			goto error;
1135 		}
1136 
1137 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1138 					 &mask, log);
1139 		if (IS_ERR(*acts)) {
1140 			error = PTR_ERR(*acts);
1141 			goto error;
1142 		}
1143 	}
1144 
1145 	/* On success, error is 0. */
1146 error:
1147 	match->mask = NULL;
1148 	return error;
1149 }
1150 
1151 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1152 {
1153 	struct net *net = sock_net(skb->sk);
1154 	struct nlattr **a = info->attrs;
1155 	struct ovs_header *ovs_header = info->userhdr;
1156 	struct sw_flow_key key;
1157 	struct sw_flow *flow;
1158 	struct sk_buff *reply = NULL;
1159 	struct datapath *dp;
1160 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1161 	struct sw_flow_match match;
1162 	struct sw_flow_id sfid;
1163 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1164 	int error = 0;
1165 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1166 	bool ufid_present;
1167 
1168 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1169 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1170 		OVS_NLERR(log,
1171 			  "Flow set message rejected, Key attribute missing.");
1172 		return -EINVAL;
1173 	}
1174 
1175 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1176 					      &acts, log);
1177 	if (error)
1178 		goto error;
1179 
1180 	if (acts) {
1181 		/* Can allocate before locking if have acts. */
1182 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1183 						ufid_flags);
1184 		if (IS_ERR(reply)) {
1185 			error = PTR_ERR(reply);
1186 			goto err_kfree_acts;
1187 		}
1188 	}
1189 
1190 	ovs_lock();
1191 	dp = get_dp(net, ovs_header->dp_ifindex);
1192 	if (unlikely(!dp)) {
1193 		error = -ENODEV;
1194 		goto err_unlock_ovs;
1195 	}
1196 	/* Check that the flow exists. */
1197 	if (ufid_present)
1198 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1199 	else
1200 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1201 	if (unlikely(!flow)) {
1202 		error = -ENOENT;
1203 		goto err_unlock_ovs;
1204 	}
1205 
1206 	/* Update actions, if present. */
1207 	if (likely(acts)) {
1208 		old_acts = ovsl_dereference(flow->sf_acts);
1209 		rcu_assign_pointer(flow->sf_acts, acts);
1210 
1211 		if (unlikely(reply)) {
1212 			error = ovs_flow_cmd_fill_info(flow,
1213 						       ovs_header->dp_ifindex,
1214 						       reply, info->snd_portid,
1215 						       info->snd_seq, 0,
1216 						       OVS_FLOW_CMD_SET,
1217 						       ufid_flags);
1218 			BUG_ON(error < 0);
1219 		}
1220 	} else {
1221 		/* Could not alloc without acts before locking. */
1222 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1223 						info, OVS_FLOW_CMD_SET, false,
1224 						ufid_flags);
1225 
1226 		if (IS_ERR(reply)) {
1227 			error = PTR_ERR(reply);
1228 			goto err_unlock_ovs;
1229 		}
1230 	}
1231 
1232 	/* Clear stats. */
1233 	if (a[OVS_FLOW_ATTR_CLEAR])
1234 		ovs_flow_stats_clear(flow);
1235 	ovs_unlock();
1236 
1237 	if (reply)
1238 		ovs_notify(&dp_flow_genl_family, reply, info);
1239 	if (old_acts)
1240 		ovs_nla_free_flow_actions_rcu(old_acts);
1241 
1242 	return 0;
1243 
1244 err_unlock_ovs:
1245 	ovs_unlock();
1246 	kfree_skb(reply);
1247 err_kfree_acts:
1248 	ovs_nla_free_flow_actions(acts);
1249 error:
1250 	return error;
1251 }
1252 
1253 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1254 {
1255 	struct nlattr **a = info->attrs;
1256 	struct ovs_header *ovs_header = info->userhdr;
1257 	struct net *net = sock_net(skb->sk);
1258 	struct sw_flow_key key;
1259 	struct sk_buff *reply;
1260 	struct sw_flow *flow;
1261 	struct datapath *dp;
1262 	struct sw_flow_match match;
1263 	struct sw_flow_id ufid;
1264 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1265 	int err = 0;
1266 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1267 	bool ufid_present;
1268 
1269 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1270 	if (a[OVS_FLOW_ATTR_KEY]) {
1271 		ovs_match_init(&match, &key, true, NULL);
1272 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1273 					log);
1274 	} else if (!ufid_present) {
1275 		OVS_NLERR(log,
1276 			  "Flow get message rejected, Key attribute missing.");
1277 		err = -EINVAL;
1278 	}
1279 	if (err)
1280 		return err;
1281 
1282 	ovs_lock();
1283 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1284 	if (!dp) {
1285 		err = -ENODEV;
1286 		goto unlock;
1287 	}
1288 
1289 	if (ufid_present)
1290 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1291 	else
1292 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1293 	if (!flow) {
1294 		err = -ENOENT;
1295 		goto unlock;
1296 	}
1297 
1298 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1299 					OVS_FLOW_CMD_GET, true, ufid_flags);
1300 	if (IS_ERR(reply)) {
1301 		err = PTR_ERR(reply);
1302 		goto unlock;
1303 	}
1304 
1305 	ovs_unlock();
1306 	return genlmsg_reply(reply, info);
1307 unlock:
1308 	ovs_unlock();
1309 	return err;
1310 }
1311 
1312 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1313 {
1314 	struct nlattr **a = info->attrs;
1315 	struct ovs_header *ovs_header = info->userhdr;
1316 	struct net *net = sock_net(skb->sk);
1317 	struct sw_flow_key key;
1318 	struct sk_buff *reply;
1319 	struct sw_flow *flow = NULL;
1320 	struct datapath *dp;
1321 	struct sw_flow_match match;
1322 	struct sw_flow_id ufid;
1323 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1324 	int err;
1325 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1326 	bool ufid_present;
1327 
1328 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1329 	if (a[OVS_FLOW_ATTR_KEY]) {
1330 		ovs_match_init(&match, &key, true, NULL);
1331 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1332 					NULL, log);
1333 		if (unlikely(err))
1334 			return err;
1335 	}
1336 
1337 	ovs_lock();
1338 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1339 	if (unlikely(!dp)) {
1340 		err = -ENODEV;
1341 		goto unlock;
1342 	}
1343 
1344 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1345 		err = ovs_flow_tbl_flush(&dp->table);
1346 		goto unlock;
1347 	}
1348 
1349 	if (ufid_present)
1350 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1351 	else
1352 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1353 	if (unlikely(!flow)) {
1354 		err = -ENOENT;
1355 		goto unlock;
1356 	}
1357 
1358 	ovs_flow_tbl_remove(&dp->table, flow);
1359 	ovs_unlock();
1360 
1361 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1362 					&flow->id, info, false, ufid_flags);
1363 	if (likely(reply)) {
1364 		if (!IS_ERR(reply)) {
1365 			rcu_read_lock();	/*To keep RCU checker happy. */
1366 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1367 						     reply, info->snd_portid,
1368 						     info->snd_seq, 0,
1369 						     OVS_FLOW_CMD_DEL,
1370 						     ufid_flags);
1371 			rcu_read_unlock();
1372 			BUG_ON(err < 0);
1373 
1374 			ovs_notify(&dp_flow_genl_family, reply, info);
1375 		} else {
1376 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1377 		}
1378 	}
1379 
1380 	ovs_flow_free(flow, true);
1381 	return 0;
1382 unlock:
1383 	ovs_unlock();
1384 	return err;
1385 }
1386 
1387 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1388 {
1389 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1390 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1391 	struct table_instance *ti;
1392 	struct datapath *dp;
1393 	u32 ufid_flags;
1394 	int err;
1395 
1396 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1397 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1398 	if (err)
1399 		return err;
1400 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1401 
1402 	rcu_read_lock();
1403 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1404 	if (!dp) {
1405 		rcu_read_unlock();
1406 		return -ENODEV;
1407 	}
1408 
1409 	ti = rcu_dereference(dp->table.ti);
1410 	for (;;) {
1411 		struct sw_flow *flow;
1412 		u32 bucket, obj;
1413 
1414 		bucket = cb->args[0];
1415 		obj = cb->args[1];
1416 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1417 		if (!flow)
1418 			break;
1419 
1420 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1421 					   NETLINK_CB(cb->skb).portid,
1422 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1423 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1424 			break;
1425 
1426 		cb->args[0] = bucket;
1427 		cb->args[1] = obj;
1428 	}
1429 	rcu_read_unlock();
1430 	return skb->len;
1431 }
1432 
1433 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1434 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1435 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1436 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1437 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1438 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1439 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1440 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1441 };
1442 
1443 static const struct genl_ops dp_flow_genl_ops[] = {
1444 	{ .cmd = OVS_FLOW_CMD_NEW,
1445 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1446 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1447 	  .doit = ovs_flow_cmd_new
1448 	},
1449 	{ .cmd = OVS_FLOW_CMD_DEL,
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_del
1453 	},
1454 	{ .cmd = OVS_FLOW_CMD_GET,
1455 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1456 	  .flags = 0,		    /* OK for unprivileged users. */
1457 	  .doit = ovs_flow_cmd_get,
1458 	  .dumpit = ovs_flow_cmd_dump
1459 	},
1460 	{ .cmd = OVS_FLOW_CMD_SET,
1461 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1462 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1463 	  .doit = ovs_flow_cmd_set,
1464 	},
1465 };
1466 
1467 static struct genl_family dp_flow_genl_family __ro_after_init = {
1468 	.hdrsize = sizeof(struct ovs_header),
1469 	.name = OVS_FLOW_FAMILY,
1470 	.version = OVS_FLOW_VERSION,
1471 	.maxattr = OVS_FLOW_ATTR_MAX,
1472 	.policy = flow_policy,
1473 	.netnsok = true,
1474 	.parallel_ops = true,
1475 	.ops = dp_flow_genl_ops,
1476 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1477 	.mcgrps = &ovs_dp_flow_multicast_group,
1478 	.n_mcgrps = 1,
1479 	.module = THIS_MODULE,
1480 };
1481 
1482 static size_t ovs_dp_cmd_msg_size(void)
1483 {
1484 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1485 
1486 	msgsize += nla_total_size(IFNAMSIZ);
1487 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1488 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1489 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1490 
1491 	return msgsize;
1492 }
1493 
1494 /* Called with ovs_mutex. */
1495 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1496 				u32 portid, u32 seq, u32 flags, u8 cmd)
1497 {
1498 	struct ovs_header *ovs_header;
1499 	struct ovs_dp_stats dp_stats;
1500 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1501 	int err;
1502 
1503 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1504 				   flags, cmd);
1505 	if (!ovs_header)
1506 		goto error;
1507 
1508 	ovs_header->dp_ifindex = get_dpifindex(dp);
1509 
1510 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1511 	if (err)
1512 		goto nla_put_failure;
1513 
1514 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1515 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1516 			  &dp_stats, OVS_DP_ATTR_PAD))
1517 		goto nla_put_failure;
1518 
1519 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1520 			  sizeof(struct ovs_dp_megaflow_stats),
1521 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1522 		goto nla_put_failure;
1523 
1524 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1525 		goto nla_put_failure;
1526 
1527 	genlmsg_end(skb, ovs_header);
1528 	return 0;
1529 
1530 nla_put_failure:
1531 	genlmsg_cancel(skb, ovs_header);
1532 error:
1533 	return -EMSGSIZE;
1534 }
1535 
1536 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1537 {
1538 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1539 }
1540 
1541 /* Called with rcu_read_lock or ovs_mutex. */
1542 static struct datapath *lookup_datapath(struct net *net,
1543 					const struct ovs_header *ovs_header,
1544 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1545 {
1546 	struct datapath *dp;
1547 
1548 	if (!a[OVS_DP_ATTR_NAME])
1549 		dp = get_dp(net, ovs_header->dp_ifindex);
1550 	else {
1551 		struct vport *vport;
1552 
1553 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1554 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1555 	}
1556 	return dp ? dp : ERR_PTR(-ENODEV);
1557 }
1558 
1559 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1560 {
1561 	struct datapath *dp;
1562 
1563 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1564 	if (IS_ERR(dp))
1565 		return;
1566 
1567 	WARN(dp->user_features, "Dropping previously announced user features\n");
1568 	dp->user_features = 0;
1569 }
1570 
1571 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1572 
1573 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1574 {
1575 	u32 user_features = 0;
1576 
1577 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1578 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1579 
1580 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1581 				      OVS_DP_F_UNALIGNED |
1582 				      OVS_DP_F_TC_RECIRC_SHARING))
1583 			return -EOPNOTSUPP;
1584 
1585 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1586 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1587 			return -EOPNOTSUPP;
1588 #endif
1589 	}
1590 
1591 	dp->user_features = user_features;
1592 
1593 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1594 		static_branch_enable(&tc_recirc_sharing_support);
1595 	else
1596 		static_branch_disable(&tc_recirc_sharing_support);
1597 
1598 	return 0;
1599 }
1600 
1601 static int ovs_dp_stats_init(struct datapath *dp)
1602 {
1603 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1604 	if (!dp->stats_percpu)
1605 		return -ENOMEM;
1606 
1607 	return 0;
1608 }
1609 
1610 static int ovs_dp_vport_init(struct datapath *dp)
1611 {
1612 	int i;
1613 
1614 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1615 				  sizeof(struct hlist_head),
1616 				  GFP_KERNEL);
1617 	if (!dp->ports)
1618 		return -ENOMEM;
1619 
1620 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1621 		INIT_HLIST_HEAD(&dp->ports[i]);
1622 
1623 	return 0;
1624 }
1625 
1626 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1627 {
1628 	struct nlattr **a = info->attrs;
1629 	struct vport_parms parms;
1630 	struct sk_buff *reply;
1631 	struct datapath *dp;
1632 	struct vport *vport;
1633 	struct ovs_net *ovs_net;
1634 	int err;
1635 
1636 	err = -EINVAL;
1637 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1638 		goto err;
1639 
1640 	reply = ovs_dp_cmd_alloc_info();
1641 	if (!reply)
1642 		return -ENOMEM;
1643 
1644 	err = -ENOMEM;
1645 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1646 	if (dp == NULL)
1647 		goto err_destroy_reply;
1648 
1649 	ovs_dp_set_net(dp, sock_net(skb->sk));
1650 
1651 	/* Allocate table. */
1652 	err = ovs_flow_tbl_init(&dp->table);
1653 	if (err)
1654 		goto err_destroy_dp;
1655 
1656 	err = ovs_dp_stats_init(dp);
1657 	if (err)
1658 		goto err_destroy_table;
1659 
1660 	err = ovs_dp_vport_init(dp);
1661 	if (err)
1662 		goto err_destroy_stats;
1663 
1664 	err = ovs_meters_init(dp);
1665 	if (err)
1666 		goto err_destroy_ports;
1667 
1668 	/* Set up our datapath device. */
1669 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1670 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1671 	parms.options = NULL;
1672 	parms.dp = dp;
1673 	parms.port_no = OVSP_LOCAL;
1674 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1675 
1676 	err = ovs_dp_change(dp, a);
1677 	if (err)
1678 		goto err_destroy_meters;
1679 
1680 	/* So far only local changes have been made, now need the lock. */
1681 	ovs_lock();
1682 
1683 	vport = new_vport(&parms);
1684 	if (IS_ERR(vport)) {
1685 		err = PTR_ERR(vport);
1686 		if (err == -EBUSY)
1687 			err = -EEXIST;
1688 
1689 		if (err == -EEXIST) {
1690 			/* An outdated user space instance that does not understand
1691 			 * the concept of user_features has attempted to create a new
1692 			 * datapath and is likely to reuse it. Drop all user features.
1693 			 */
1694 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1695 				ovs_dp_reset_user_features(skb, info);
1696 		}
1697 
1698 		ovs_unlock();
1699 		goto err_destroy_meters;
1700 	}
1701 
1702 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1703 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1704 	BUG_ON(err < 0);
1705 
1706 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1707 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1708 
1709 	ovs_unlock();
1710 
1711 	ovs_notify(&dp_datapath_genl_family, reply, info);
1712 	return 0;
1713 
1714 err_destroy_meters:
1715 	ovs_meters_exit(dp);
1716 err_destroy_ports:
1717 	kfree(dp->ports);
1718 err_destroy_stats:
1719 	free_percpu(dp->stats_percpu);
1720 err_destroy_table:
1721 	ovs_flow_tbl_destroy(&dp->table);
1722 err_destroy_dp:
1723 	kfree(dp);
1724 err_destroy_reply:
1725 	kfree_skb(reply);
1726 err:
1727 	return err;
1728 }
1729 
1730 /* Called with ovs_mutex. */
1731 static void __dp_destroy(struct datapath *dp)
1732 {
1733 	int i;
1734 
1735 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1736 		struct vport *vport;
1737 		struct hlist_node *n;
1738 
1739 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1740 			if (vport->port_no != OVSP_LOCAL)
1741 				ovs_dp_detach_port(vport);
1742 	}
1743 
1744 	list_del_rcu(&dp->list_node);
1745 
1746 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1747 	 * all ports in datapath are destroyed first before freeing datapath.
1748 	 */
1749 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1750 
1751 	/* RCU destroy the flow table */
1752 	call_rcu(&dp->rcu, destroy_dp_rcu);
1753 }
1754 
1755 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1756 {
1757 	struct sk_buff *reply;
1758 	struct datapath *dp;
1759 	int err;
1760 
1761 	reply = ovs_dp_cmd_alloc_info();
1762 	if (!reply)
1763 		return -ENOMEM;
1764 
1765 	ovs_lock();
1766 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1767 	err = PTR_ERR(dp);
1768 	if (IS_ERR(dp))
1769 		goto err_unlock_free;
1770 
1771 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1772 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1773 	BUG_ON(err < 0);
1774 
1775 	__dp_destroy(dp);
1776 	ovs_unlock();
1777 
1778 	ovs_notify(&dp_datapath_genl_family, reply, info);
1779 
1780 	return 0;
1781 
1782 err_unlock_free:
1783 	ovs_unlock();
1784 	kfree_skb(reply);
1785 	return err;
1786 }
1787 
1788 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1789 {
1790 	struct sk_buff *reply;
1791 	struct datapath *dp;
1792 	int err;
1793 
1794 	reply = ovs_dp_cmd_alloc_info();
1795 	if (!reply)
1796 		return -ENOMEM;
1797 
1798 	ovs_lock();
1799 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1800 	err = PTR_ERR(dp);
1801 	if (IS_ERR(dp))
1802 		goto err_unlock_free;
1803 
1804 	err = ovs_dp_change(dp, info->attrs);
1805 	if (err)
1806 		goto err_unlock_free;
1807 
1808 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1809 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1810 	BUG_ON(err < 0);
1811 
1812 	ovs_unlock();
1813 	ovs_notify(&dp_datapath_genl_family, reply, info);
1814 
1815 	return 0;
1816 
1817 err_unlock_free:
1818 	ovs_unlock();
1819 	kfree_skb(reply);
1820 	return err;
1821 }
1822 
1823 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1824 {
1825 	struct sk_buff *reply;
1826 	struct datapath *dp;
1827 	int err;
1828 
1829 	reply = ovs_dp_cmd_alloc_info();
1830 	if (!reply)
1831 		return -ENOMEM;
1832 
1833 	ovs_lock();
1834 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1835 	if (IS_ERR(dp)) {
1836 		err = PTR_ERR(dp);
1837 		goto err_unlock_free;
1838 	}
1839 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1840 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1841 	BUG_ON(err < 0);
1842 	ovs_unlock();
1843 
1844 	return genlmsg_reply(reply, info);
1845 
1846 err_unlock_free:
1847 	ovs_unlock();
1848 	kfree_skb(reply);
1849 	return err;
1850 }
1851 
1852 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1853 {
1854 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1855 	struct datapath *dp;
1856 	int skip = cb->args[0];
1857 	int i = 0;
1858 
1859 	ovs_lock();
1860 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1861 		if (i >= skip &&
1862 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1863 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1864 					 OVS_DP_CMD_GET) < 0)
1865 			break;
1866 		i++;
1867 	}
1868 	ovs_unlock();
1869 
1870 	cb->args[0] = i;
1871 
1872 	return skb->len;
1873 }
1874 
1875 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1876 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1877 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1878 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1879 };
1880 
1881 static const struct genl_ops dp_datapath_genl_ops[] = {
1882 	{ .cmd = OVS_DP_CMD_NEW,
1883 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1884 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1885 	  .doit = ovs_dp_cmd_new
1886 	},
1887 	{ .cmd = OVS_DP_CMD_DEL,
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_del
1891 	},
1892 	{ .cmd = OVS_DP_CMD_GET,
1893 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1894 	  .flags = 0,		    /* OK for unprivileged users. */
1895 	  .doit = ovs_dp_cmd_get,
1896 	  .dumpit = ovs_dp_cmd_dump
1897 	},
1898 	{ .cmd = OVS_DP_CMD_SET,
1899 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1900 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1901 	  .doit = ovs_dp_cmd_set,
1902 	},
1903 };
1904 
1905 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1906 	.hdrsize = sizeof(struct ovs_header),
1907 	.name = OVS_DATAPATH_FAMILY,
1908 	.version = OVS_DATAPATH_VERSION,
1909 	.maxattr = OVS_DP_ATTR_MAX,
1910 	.policy = datapath_policy,
1911 	.netnsok = true,
1912 	.parallel_ops = true,
1913 	.ops = dp_datapath_genl_ops,
1914 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1915 	.mcgrps = &ovs_dp_datapath_multicast_group,
1916 	.n_mcgrps = 1,
1917 	.module = THIS_MODULE,
1918 };
1919 
1920 /* Called with ovs_mutex or RCU read lock. */
1921 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1922 				   struct net *net, u32 portid, u32 seq,
1923 				   u32 flags, u8 cmd, gfp_t gfp)
1924 {
1925 	struct ovs_header *ovs_header;
1926 	struct ovs_vport_stats vport_stats;
1927 	int err;
1928 
1929 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1930 				 flags, cmd);
1931 	if (!ovs_header)
1932 		return -EMSGSIZE;
1933 
1934 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1935 
1936 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1937 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1938 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1939 			   ovs_vport_name(vport)) ||
1940 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1941 		goto nla_put_failure;
1942 
1943 	if (!net_eq(net, dev_net(vport->dev))) {
1944 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1945 
1946 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1947 			goto nla_put_failure;
1948 	}
1949 
1950 	ovs_vport_get_stats(vport, &vport_stats);
1951 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1952 			  sizeof(struct ovs_vport_stats), &vport_stats,
1953 			  OVS_VPORT_ATTR_PAD))
1954 		goto nla_put_failure;
1955 
1956 	if (ovs_vport_get_upcall_portids(vport, skb))
1957 		goto nla_put_failure;
1958 
1959 	err = ovs_vport_get_options(vport, skb);
1960 	if (err == -EMSGSIZE)
1961 		goto error;
1962 
1963 	genlmsg_end(skb, ovs_header);
1964 	return 0;
1965 
1966 nla_put_failure:
1967 	err = -EMSGSIZE;
1968 error:
1969 	genlmsg_cancel(skb, ovs_header);
1970 	return err;
1971 }
1972 
1973 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1974 {
1975 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1976 }
1977 
1978 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1979 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
1980 					 u32 portid, u32 seq, u8 cmd)
1981 {
1982 	struct sk_buff *skb;
1983 	int retval;
1984 
1985 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1986 	if (!skb)
1987 		return ERR_PTR(-ENOMEM);
1988 
1989 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
1990 					 GFP_KERNEL);
1991 	BUG_ON(retval < 0);
1992 
1993 	return skb;
1994 }
1995 
1996 /* Called with ovs_mutex or RCU read lock. */
1997 static struct vport *lookup_vport(struct net *net,
1998 				  const struct ovs_header *ovs_header,
1999 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2000 {
2001 	struct datapath *dp;
2002 	struct vport *vport;
2003 
2004 	if (a[OVS_VPORT_ATTR_IFINDEX])
2005 		return ERR_PTR(-EOPNOTSUPP);
2006 	if (a[OVS_VPORT_ATTR_NAME]) {
2007 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2008 		if (!vport)
2009 			return ERR_PTR(-ENODEV);
2010 		if (ovs_header->dp_ifindex &&
2011 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2012 			return ERR_PTR(-ENODEV);
2013 		return vport;
2014 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2015 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2016 
2017 		if (port_no >= DP_MAX_PORTS)
2018 			return ERR_PTR(-EFBIG);
2019 
2020 		dp = get_dp(net, ovs_header->dp_ifindex);
2021 		if (!dp)
2022 			return ERR_PTR(-ENODEV);
2023 
2024 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2025 		if (!vport)
2026 			return ERR_PTR(-ENODEV);
2027 		return vport;
2028 	} else
2029 		return ERR_PTR(-EINVAL);
2030 
2031 }
2032 
2033 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2034 {
2035 	unsigned int dev_headroom, max_headroom = 0;
2036 	struct net_device *dev;
2037 	struct vport *vport;
2038 	int i;
2039 
2040 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2041 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2042 			dev = vport->dev;
2043 			dev_headroom = netdev_get_fwd_headroom(dev);
2044 			if (dev_headroom > max_headroom)
2045 				max_headroom = dev_headroom;
2046 		}
2047 	}
2048 
2049 	return max_headroom;
2050 }
2051 
2052 /* Called with ovs_mutex */
2053 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2054 {
2055 	struct vport *vport;
2056 	int i;
2057 
2058 	dp->max_headroom = new_headroom;
2059 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2060 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
2061 			netdev_set_rx_headroom(vport->dev, new_headroom);
2062 }
2063 
2064 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2065 {
2066 	struct nlattr **a = info->attrs;
2067 	struct ovs_header *ovs_header = info->userhdr;
2068 	struct vport_parms parms;
2069 	struct sk_buff *reply;
2070 	struct vport *vport;
2071 	struct datapath *dp;
2072 	unsigned int new_headroom;
2073 	u32 port_no;
2074 	int err;
2075 
2076 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2077 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2078 		return -EINVAL;
2079 	if (a[OVS_VPORT_ATTR_IFINDEX])
2080 		return -EOPNOTSUPP;
2081 
2082 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2083 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2084 	if (port_no >= DP_MAX_PORTS)
2085 		return -EFBIG;
2086 
2087 	reply = ovs_vport_cmd_alloc_info();
2088 	if (!reply)
2089 		return -ENOMEM;
2090 
2091 	ovs_lock();
2092 restart:
2093 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2094 	err = -ENODEV;
2095 	if (!dp)
2096 		goto exit_unlock_free;
2097 
2098 	if (port_no) {
2099 		vport = ovs_vport_ovsl(dp, port_no);
2100 		err = -EBUSY;
2101 		if (vport)
2102 			goto exit_unlock_free;
2103 	} else {
2104 		for (port_no = 1; ; port_no++) {
2105 			if (port_no >= DP_MAX_PORTS) {
2106 				err = -EFBIG;
2107 				goto exit_unlock_free;
2108 			}
2109 			vport = ovs_vport_ovsl(dp, port_no);
2110 			if (!vport)
2111 				break;
2112 		}
2113 	}
2114 
2115 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2116 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2117 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2118 	parms.dp = dp;
2119 	parms.port_no = port_no;
2120 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2121 
2122 	vport = new_vport(&parms);
2123 	err = PTR_ERR(vport);
2124 	if (IS_ERR(vport)) {
2125 		if (err == -EAGAIN)
2126 			goto restart;
2127 		goto exit_unlock_free;
2128 	}
2129 
2130 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2131 				      info->snd_portid, info->snd_seq, 0,
2132 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2133 
2134 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2135 
2136 	if (new_headroom > dp->max_headroom)
2137 		ovs_update_headroom(dp, new_headroom);
2138 	else
2139 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2140 
2141 	BUG_ON(err < 0);
2142 	ovs_unlock();
2143 
2144 	ovs_notify(&dp_vport_genl_family, reply, info);
2145 	return 0;
2146 
2147 exit_unlock_free:
2148 	ovs_unlock();
2149 	kfree_skb(reply);
2150 	return err;
2151 }
2152 
2153 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2154 {
2155 	struct nlattr **a = info->attrs;
2156 	struct sk_buff *reply;
2157 	struct vport *vport;
2158 	int err;
2159 
2160 	reply = ovs_vport_cmd_alloc_info();
2161 	if (!reply)
2162 		return -ENOMEM;
2163 
2164 	ovs_lock();
2165 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2166 	err = PTR_ERR(vport);
2167 	if (IS_ERR(vport))
2168 		goto exit_unlock_free;
2169 
2170 	if (a[OVS_VPORT_ATTR_TYPE] &&
2171 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2172 		err = -EINVAL;
2173 		goto exit_unlock_free;
2174 	}
2175 
2176 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2177 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2178 		if (err)
2179 			goto exit_unlock_free;
2180 	}
2181 
2182 
2183 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2184 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2185 
2186 		err = ovs_vport_set_upcall_portids(vport, ids);
2187 		if (err)
2188 			goto exit_unlock_free;
2189 	}
2190 
2191 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2192 				      info->snd_portid, info->snd_seq, 0,
2193 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2194 	BUG_ON(err < 0);
2195 
2196 	ovs_unlock();
2197 	ovs_notify(&dp_vport_genl_family, reply, info);
2198 	return 0;
2199 
2200 exit_unlock_free:
2201 	ovs_unlock();
2202 	kfree_skb(reply);
2203 	return err;
2204 }
2205 
2206 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2207 {
2208 	bool update_headroom = false;
2209 	struct nlattr **a = info->attrs;
2210 	struct sk_buff *reply;
2211 	struct datapath *dp;
2212 	struct vport *vport;
2213 	unsigned int new_headroom;
2214 	int err;
2215 
2216 	reply = ovs_vport_cmd_alloc_info();
2217 	if (!reply)
2218 		return -ENOMEM;
2219 
2220 	ovs_lock();
2221 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2222 	err = PTR_ERR(vport);
2223 	if (IS_ERR(vport))
2224 		goto exit_unlock_free;
2225 
2226 	if (vport->port_no == OVSP_LOCAL) {
2227 		err = -EINVAL;
2228 		goto exit_unlock_free;
2229 	}
2230 
2231 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2232 				      info->snd_portid, info->snd_seq, 0,
2233 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2234 	BUG_ON(err < 0);
2235 
2236 	/* the vport deletion may trigger dp headroom update */
2237 	dp = vport->dp;
2238 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2239 		update_headroom = true;
2240 
2241 	netdev_reset_rx_headroom(vport->dev);
2242 	ovs_dp_detach_port(vport);
2243 
2244 	if (update_headroom) {
2245 		new_headroom = ovs_get_max_headroom(dp);
2246 
2247 		if (new_headroom < dp->max_headroom)
2248 			ovs_update_headroom(dp, new_headroom);
2249 	}
2250 	ovs_unlock();
2251 
2252 	ovs_notify(&dp_vport_genl_family, reply, info);
2253 	return 0;
2254 
2255 exit_unlock_free:
2256 	ovs_unlock();
2257 	kfree_skb(reply);
2258 	return err;
2259 }
2260 
2261 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2262 {
2263 	struct nlattr **a = info->attrs;
2264 	struct ovs_header *ovs_header = info->userhdr;
2265 	struct sk_buff *reply;
2266 	struct vport *vport;
2267 	int err;
2268 
2269 	reply = ovs_vport_cmd_alloc_info();
2270 	if (!reply)
2271 		return -ENOMEM;
2272 
2273 	rcu_read_lock();
2274 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2275 	err = PTR_ERR(vport);
2276 	if (IS_ERR(vport))
2277 		goto exit_unlock_free;
2278 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2279 				      info->snd_portid, info->snd_seq, 0,
2280 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2281 	BUG_ON(err < 0);
2282 	rcu_read_unlock();
2283 
2284 	return genlmsg_reply(reply, info);
2285 
2286 exit_unlock_free:
2287 	rcu_read_unlock();
2288 	kfree_skb(reply);
2289 	return err;
2290 }
2291 
2292 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2293 {
2294 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2295 	struct datapath *dp;
2296 	int bucket = cb->args[0], skip = cb->args[1];
2297 	int i, j = 0;
2298 
2299 	rcu_read_lock();
2300 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2301 	if (!dp) {
2302 		rcu_read_unlock();
2303 		return -ENODEV;
2304 	}
2305 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2306 		struct vport *vport;
2307 
2308 		j = 0;
2309 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2310 			if (j >= skip &&
2311 			    ovs_vport_cmd_fill_info(vport, skb,
2312 						    sock_net(skb->sk),
2313 						    NETLINK_CB(cb->skb).portid,
2314 						    cb->nlh->nlmsg_seq,
2315 						    NLM_F_MULTI,
2316 						    OVS_VPORT_CMD_GET,
2317 						    GFP_ATOMIC) < 0)
2318 				goto out;
2319 
2320 			j++;
2321 		}
2322 		skip = 0;
2323 	}
2324 out:
2325 	rcu_read_unlock();
2326 
2327 	cb->args[0] = i;
2328 	cb->args[1] = j;
2329 
2330 	return skb->len;
2331 }
2332 
2333 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2334 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2335 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2336 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2337 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2338 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2339 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2340 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2341 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2342 };
2343 
2344 static const struct genl_ops dp_vport_genl_ops[] = {
2345 	{ .cmd = OVS_VPORT_CMD_NEW,
2346 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2347 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2348 	  .doit = ovs_vport_cmd_new
2349 	},
2350 	{ .cmd = OVS_VPORT_CMD_DEL,
2351 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2352 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2353 	  .doit = ovs_vport_cmd_del
2354 	},
2355 	{ .cmd = OVS_VPORT_CMD_GET,
2356 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2357 	  .flags = 0,		    /* OK for unprivileged users. */
2358 	  .doit = ovs_vport_cmd_get,
2359 	  .dumpit = ovs_vport_cmd_dump
2360 	},
2361 	{ .cmd = OVS_VPORT_CMD_SET,
2362 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2363 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2364 	  .doit = ovs_vport_cmd_set,
2365 	},
2366 };
2367 
2368 struct genl_family dp_vport_genl_family __ro_after_init = {
2369 	.hdrsize = sizeof(struct ovs_header),
2370 	.name = OVS_VPORT_FAMILY,
2371 	.version = OVS_VPORT_VERSION,
2372 	.maxattr = OVS_VPORT_ATTR_MAX,
2373 	.policy = vport_policy,
2374 	.netnsok = true,
2375 	.parallel_ops = true,
2376 	.ops = dp_vport_genl_ops,
2377 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2378 	.mcgrps = &ovs_dp_vport_multicast_group,
2379 	.n_mcgrps = 1,
2380 	.module = THIS_MODULE,
2381 };
2382 
2383 static struct genl_family * const dp_genl_families[] = {
2384 	&dp_datapath_genl_family,
2385 	&dp_vport_genl_family,
2386 	&dp_flow_genl_family,
2387 	&dp_packet_genl_family,
2388 	&dp_meter_genl_family,
2389 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2390 	&dp_ct_limit_genl_family,
2391 #endif
2392 };
2393 
2394 static void dp_unregister_genl(int n_families)
2395 {
2396 	int i;
2397 
2398 	for (i = 0; i < n_families; i++)
2399 		genl_unregister_family(dp_genl_families[i]);
2400 }
2401 
2402 static int __init dp_register_genl(void)
2403 {
2404 	int err;
2405 	int i;
2406 
2407 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2408 
2409 		err = genl_register_family(dp_genl_families[i]);
2410 		if (err)
2411 			goto error;
2412 	}
2413 
2414 	return 0;
2415 
2416 error:
2417 	dp_unregister_genl(i);
2418 	return err;
2419 }
2420 
2421 static int __net_init ovs_init_net(struct net *net)
2422 {
2423 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2424 
2425 	INIT_LIST_HEAD(&ovs_net->dps);
2426 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2427 	return ovs_ct_init(net);
2428 }
2429 
2430 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2431 					    struct list_head *head)
2432 {
2433 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2434 	struct datapath *dp;
2435 
2436 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2437 		int i;
2438 
2439 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2440 			struct vport *vport;
2441 
2442 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2443 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2444 					continue;
2445 
2446 				if (dev_net(vport->dev) == dnet)
2447 					list_add(&vport->detach_list, head);
2448 			}
2449 		}
2450 	}
2451 }
2452 
2453 static void __net_exit ovs_exit_net(struct net *dnet)
2454 {
2455 	struct datapath *dp, *dp_next;
2456 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2457 	struct vport *vport, *vport_next;
2458 	struct net *net;
2459 	LIST_HEAD(head);
2460 
2461 	ovs_ct_exit(dnet);
2462 	ovs_lock();
2463 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2464 		__dp_destroy(dp);
2465 
2466 	down_read(&net_rwsem);
2467 	for_each_net(net)
2468 		list_vports_from_net(net, dnet, &head);
2469 	up_read(&net_rwsem);
2470 
2471 	/* Detach all vports from given namespace. */
2472 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2473 		list_del(&vport->detach_list);
2474 		ovs_dp_detach_port(vport);
2475 	}
2476 
2477 	ovs_unlock();
2478 
2479 	cancel_work_sync(&ovs_net->dp_notify_work);
2480 }
2481 
2482 static struct pernet_operations ovs_net_ops = {
2483 	.init = ovs_init_net,
2484 	.exit = ovs_exit_net,
2485 	.id   = &ovs_net_id,
2486 	.size = sizeof(struct ovs_net),
2487 };
2488 
2489 static int __init dp_init(void)
2490 {
2491 	int err;
2492 
2493 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2494 
2495 	pr_info("Open vSwitch switching datapath\n");
2496 
2497 	err = action_fifos_init();
2498 	if (err)
2499 		goto error;
2500 
2501 	err = ovs_internal_dev_rtnl_link_register();
2502 	if (err)
2503 		goto error_action_fifos_exit;
2504 
2505 	err = ovs_flow_init();
2506 	if (err)
2507 		goto error_unreg_rtnl_link;
2508 
2509 	err = ovs_vport_init();
2510 	if (err)
2511 		goto error_flow_exit;
2512 
2513 	err = register_pernet_device(&ovs_net_ops);
2514 	if (err)
2515 		goto error_vport_exit;
2516 
2517 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2518 	if (err)
2519 		goto error_netns_exit;
2520 
2521 	err = ovs_netdev_init();
2522 	if (err)
2523 		goto error_unreg_notifier;
2524 
2525 	err = dp_register_genl();
2526 	if (err < 0)
2527 		goto error_unreg_netdev;
2528 
2529 	return 0;
2530 
2531 error_unreg_netdev:
2532 	ovs_netdev_exit();
2533 error_unreg_notifier:
2534 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2535 error_netns_exit:
2536 	unregister_pernet_device(&ovs_net_ops);
2537 error_vport_exit:
2538 	ovs_vport_exit();
2539 error_flow_exit:
2540 	ovs_flow_exit();
2541 error_unreg_rtnl_link:
2542 	ovs_internal_dev_rtnl_link_unregister();
2543 error_action_fifos_exit:
2544 	action_fifos_exit();
2545 error:
2546 	return err;
2547 }
2548 
2549 static void dp_cleanup(void)
2550 {
2551 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2552 	ovs_netdev_exit();
2553 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2554 	unregister_pernet_device(&ovs_net_ops);
2555 	rcu_barrier();
2556 	ovs_vport_exit();
2557 	ovs_flow_exit();
2558 	ovs_internal_dev_rtnl_link_unregister();
2559 	action_fifos_exit();
2560 }
2561 
2562 module_init(dp_init);
2563 module_exit(dp_cleanup);
2564 
2565 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2566 MODULE_LICENSE("GPL");
2567 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2568 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2569 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2570 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2571 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2572 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2573