xref: /openbmc/linux/net/openvswitch/datapath.c (revision a5b2c10c)
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 static void ovs_dp_masks_rebalance(struct work_struct *work);
134 
135 /* Must be called with rcu_read_lock or ovs_mutex. */
136 const char *ovs_dp_name(const struct datapath *dp)
137 {
138 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
139 	return ovs_vport_name(vport);
140 }
141 
142 static int get_dpifindex(const struct datapath *dp)
143 {
144 	struct vport *local;
145 	int ifindex;
146 
147 	rcu_read_lock();
148 
149 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
150 	if (local)
151 		ifindex = local->dev->ifindex;
152 	else
153 		ifindex = 0;
154 
155 	rcu_read_unlock();
156 
157 	return ifindex;
158 }
159 
160 static void destroy_dp_rcu(struct rcu_head *rcu)
161 {
162 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
163 
164 	ovs_flow_tbl_destroy(&dp->table);
165 	free_percpu(dp->stats_percpu);
166 	kfree(dp->ports);
167 	ovs_meters_exit(dp);
168 	kfree(dp);
169 }
170 
171 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
172 					    u16 port_no)
173 {
174 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
175 }
176 
177 /* Called with ovs_mutex or RCU read lock. */
178 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
179 {
180 	struct vport *vport;
181 	struct hlist_head *head;
182 
183 	head = vport_hash_bucket(dp, port_no);
184 	hlist_for_each_entry_rcu(vport, head, dp_hash_node,
185 				lockdep_ovsl_is_held()) {
186 		if (vport->port_no == port_no)
187 			return vport;
188 	}
189 	return NULL;
190 }
191 
192 /* Called with ovs_mutex. */
193 static struct vport *new_vport(const struct vport_parms *parms)
194 {
195 	struct vport *vport;
196 
197 	vport = ovs_vport_add(parms);
198 	if (!IS_ERR(vport)) {
199 		struct datapath *dp = parms->dp;
200 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
201 
202 		hlist_add_head_rcu(&vport->dp_hash_node, head);
203 	}
204 	return vport;
205 }
206 
207 void ovs_dp_detach_port(struct vport *p)
208 {
209 	ASSERT_OVSL();
210 
211 	/* First drop references to device. */
212 	hlist_del_rcu(&p->dp_hash_node);
213 
214 	/* Then destroy it. */
215 	ovs_vport_del(p);
216 }
217 
218 /* Must be called with rcu_read_lock. */
219 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
220 {
221 	const struct vport *p = OVS_CB(skb)->input_vport;
222 	struct datapath *dp = p->dp;
223 	struct sw_flow *flow;
224 	struct sw_flow_actions *sf_acts;
225 	struct dp_stats_percpu *stats;
226 	u64 *stats_counter;
227 	u32 n_mask_hit;
228 	u32 n_cache_hit;
229 	int error;
230 
231 	stats = this_cpu_ptr(dp->stats_percpu);
232 
233 	/* Look up flow. */
234 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
235 					 &n_mask_hit, &n_cache_hit);
236 	if (unlikely(!flow)) {
237 		struct dp_upcall_info upcall;
238 
239 		memset(&upcall, 0, sizeof(upcall));
240 		upcall.cmd = OVS_PACKET_CMD_MISS;
241 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
242 		upcall.mru = OVS_CB(skb)->mru;
243 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
244 		if (unlikely(error))
245 			kfree_skb(skb);
246 		else
247 			consume_skb(skb);
248 		stats_counter = &stats->n_missed;
249 		goto out;
250 	}
251 
252 	ovs_flow_stats_update(flow, key->tp.flags, skb);
253 	sf_acts = rcu_dereference(flow->sf_acts);
254 	error = ovs_execute_actions(dp, skb, sf_acts, key);
255 	if (unlikely(error))
256 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
257 							ovs_dp_name(dp), error);
258 
259 	stats_counter = &stats->n_hit;
260 
261 out:
262 	/* Update datapath statistics. */
263 	u64_stats_update_begin(&stats->syncp);
264 	(*stats_counter)++;
265 	stats->n_mask_hit += n_mask_hit;
266 	stats->n_cache_hit += n_cache_hit;
267 	u64_stats_update_end(&stats->syncp);
268 }
269 
270 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
271 		  const struct sw_flow_key *key,
272 		  const struct dp_upcall_info *upcall_info,
273 		  uint32_t cutlen)
274 {
275 	struct dp_stats_percpu *stats;
276 	int err;
277 
278 	if (upcall_info->portid == 0) {
279 		err = -ENOTCONN;
280 		goto err;
281 	}
282 
283 	if (!skb_is_gso(skb))
284 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
285 	else
286 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
287 	if (err)
288 		goto err;
289 
290 	return 0;
291 
292 err:
293 	stats = this_cpu_ptr(dp->stats_percpu);
294 
295 	u64_stats_update_begin(&stats->syncp);
296 	stats->n_lost++;
297 	u64_stats_update_end(&stats->syncp);
298 
299 	return err;
300 }
301 
302 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
303 			     const struct sw_flow_key *key,
304 			     const struct dp_upcall_info *upcall_info,
305 				 uint32_t cutlen)
306 {
307 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
308 	struct sw_flow_key later_key;
309 	struct sk_buff *segs, *nskb;
310 	int err;
311 
312 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
313 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
314 	if (IS_ERR(segs))
315 		return PTR_ERR(segs);
316 	if (segs == NULL)
317 		return -EINVAL;
318 
319 	if (gso_type & SKB_GSO_UDP) {
320 		/* The initial flow key extracted by ovs_flow_key_extract()
321 		 * in this case is for a first fragment, so we need to
322 		 * properly mark later fragments.
323 		 */
324 		later_key = *key;
325 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
326 	}
327 
328 	/* Queue all of the segments. */
329 	skb_list_walk_safe(segs, skb, nskb) {
330 		if (gso_type & SKB_GSO_UDP && skb != segs)
331 			key = &later_key;
332 
333 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
334 		if (err)
335 			break;
336 
337 	}
338 
339 	/* Free all of the segments. */
340 	skb_list_walk_safe(segs, skb, nskb) {
341 		if (err)
342 			kfree_skb(skb);
343 		else
344 			consume_skb(skb);
345 	}
346 	return err;
347 }
348 
349 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
350 			      unsigned int hdrlen, int actions_attrlen)
351 {
352 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
353 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
354 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
355 		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
356 		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
357 
358 	/* OVS_PACKET_ATTR_USERDATA */
359 	if (upcall_info->userdata)
360 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
361 
362 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
363 	if (upcall_info->egress_tun_info)
364 		size += nla_total_size(ovs_tun_key_attr_size());
365 
366 	/* OVS_PACKET_ATTR_ACTIONS */
367 	if (upcall_info->actions_len)
368 		size += nla_total_size(actions_attrlen);
369 
370 	/* OVS_PACKET_ATTR_MRU */
371 	if (upcall_info->mru)
372 		size += nla_total_size(sizeof(upcall_info->mru));
373 
374 	return size;
375 }
376 
377 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
378 {
379 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
380 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
381 
382 		if (plen > 0)
383 			skb_put_zero(skb, plen);
384 	}
385 }
386 
387 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
388 				  const struct sw_flow_key *key,
389 				  const struct dp_upcall_info *upcall_info,
390 				  uint32_t cutlen)
391 {
392 	struct ovs_header *upcall;
393 	struct sk_buff *nskb = NULL;
394 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
395 	struct nlattr *nla;
396 	size_t len;
397 	unsigned int hlen;
398 	int err, dp_ifindex;
399 	u64 hash;
400 
401 	dp_ifindex = get_dpifindex(dp);
402 	if (!dp_ifindex)
403 		return -ENODEV;
404 
405 	if (skb_vlan_tag_present(skb)) {
406 		nskb = skb_clone(skb, GFP_ATOMIC);
407 		if (!nskb)
408 			return -ENOMEM;
409 
410 		nskb = __vlan_hwaccel_push_inside(nskb);
411 		if (!nskb)
412 			return -ENOMEM;
413 
414 		skb = nskb;
415 	}
416 
417 	if (nla_attr_size(skb->len) > USHRT_MAX) {
418 		err = -EFBIG;
419 		goto out;
420 	}
421 
422 	/* Complete checksum if needed */
423 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
424 	    (err = skb_csum_hwoffload_help(skb, 0)))
425 		goto out;
426 
427 	/* Older versions of OVS user space enforce alignment of the last
428 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
429 	 * padding logic. Only perform zerocopy if padding is not required.
430 	 */
431 	if (dp->user_features & OVS_DP_F_UNALIGNED)
432 		hlen = skb_zerocopy_headlen(skb);
433 	else
434 		hlen = skb->len;
435 
436 	len = upcall_msg_size(upcall_info, hlen - cutlen,
437 			      OVS_CB(skb)->acts_origlen);
438 	user_skb = genlmsg_new(len, GFP_ATOMIC);
439 	if (!user_skb) {
440 		err = -ENOMEM;
441 		goto out;
442 	}
443 
444 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
445 			     0, upcall_info->cmd);
446 	if (!upcall) {
447 		err = -EINVAL;
448 		goto out;
449 	}
450 	upcall->dp_ifindex = dp_ifindex;
451 
452 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
453 	if (err)
454 		goto out;
455 
456 	if (upcall_info->userdata)
457 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
458 			  nla_len(upcall_info->userdata),
459 			  nla_data(upcall_info->userdata));
460 
461 	if (upcall_info->egress_tun_info) {
462 		nla = nla_nest_start_noflag(user_skb,
463 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
464 		if (!nla) {
465 			err = -EMSGSIZE;
466 			goto out;
467 		}
468 		err = ovs_nla_put_tunnel_info(user_skb,
469 					      upcall_info->egress_tun_info);
470 		if (err)
471 			goto out;
472 
473 		nla_nest_end(user_skb, nla);
474 	}
475 
476 	if (upcall_info->actions_len) {
477 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
478 		if (!nla) {
479 			err = -EMSGSIZE;
480 			goto out;
481 		}
482 		err = ovs_nla_put_actions(upcall_info->actions,
483 					  upcall_info->actions_len,
484 					  user_skb);
485 		if (!err)
486 			nla_nest_end(user_skb, nla);
487 		else
488 			nla_nest_cancel(user_skb, nla);
489 	}
490 
491 	/* Add OVS_PACKET_ATTR_MRU */
492 	if (upcall_info->mru &&
493 	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
494 		err = -ENOBUFS;
495 		goto out;
496 	}
497 
498 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
499 	if (cutlen > 0 &&
500 	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
501 		err = -ENOBUFS;
502 		goto out;
503 	}
504 
505 	/* Add OVS_PACKET_ATTR_HASH */
506 	hash = skb_get_hash_raw(skb);
507 	if (skb->sw_hash)
508 		hash |= OVS_PACKET_HASH_SW_BIT;
509 
510 	if (skb->l4_hash)
511 		hash |= OVS_PACKET_HASH_L4_BIT;
512 
513 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
514 		err = -ENOBUFS;
515 		goto out;
516 	}
517 
518 	/* Only reserve room for attribute header, packet data is added
519 	 * in skb_zerocopy() */
520 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
521 		err = -ENOBUFS;
522 		goto out;
523 	}
524 	nla->nla_len = nla_attr_size(skb->len - cutlen);
525 
526 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
527 	if (err)
528 		goto out;
529 
530 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
531 	pad_packet(dp, user_skb);
532 
533 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
534 
535 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
536 	user_skb = NULL;
537 out:
538 	if (err)
539 		skb_tx_error(skb);
540 	kfree_skb(user_skb);
541 	kfree_skb(nskb);
542 	return err;
543 }
544 
545 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
546 {
547 	struct ovs_header *ovs_header = info->userhdr;
548 	struct net *net = sock_net(skb->sk);
549 	struct nlattr **a = info->attrs;
550 	struct sw_flow_actions *acts;
551 	struct sk_buff *packet;
552 	struct sw_flow *flow;
553 	struct sw_flow_actions *sf_acts;
554 	struct datapath *dp;
555 	struct vport *input_vport;
556 	u16 mru = 0;
557 	u64 hash;
558 	int len;
559 	int err;
560 	bool log = !a[OVS_PACKET_ATTR_PROBE];
561 
562 	err = -EINVAL;
563 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
564 	    !a[OVS_PACKET_ATTR_ACTIONS])
565 		goto err;
566 
567 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
568 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
569 	err = -ENOMEM;
570 	if (!packet)
571 		goto err;
572 	skb_reserve(packet, NET_IP_ALIGN);
573 
574 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
575 
576 	/* Set packet's mru */
577 	if (a[OVS_PACKET_ATTR_MRU]) {
578 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
579 		packet->ignore_df = 1;
580 	}
581 	OVS_CB(packet)->mru = mru;
582 
583 	if (a[OVS_PACKET_ATTR_HASH]) {
584 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
585 
586 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
587 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
588 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
589 	}
590 
591 	/* Build an sw_flow for sending this packet. */
592 	flow = ovs_flow_alloc();
593 	err = PTR_ERR(flow);
594 	if (IS_ERR(flow))
595 		goto err_kfree_skb;
596 
597 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
598 					     packet, &flow->key, log);
599 	if (err)
600 		goto err_flow_free;
601 
602 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
603 				   &flow->key, &acts, log);
604 	if (err)
605 		goto err_flow_free;
606 
607 	rcu_assign_pointer(flow->sf_acts, acts);
608 	packet->priority = flow->key.phy.priority;
609 	packet->mark = flow->key.phy.skb_mark;
610 
611 	rcu_read_lock();
612 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
613 	err = -ENODEV;
614 	if (!dp)
615 		goto err_unlock;
616 
617 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
618 	if (!input_vport)
619 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
620 
621 	if (!input_vport)
622 		goto err_unlock;
623 
624 	packet->dev = input_vport->dev;
625 	OVS_CB(packet)->input_vport = input_vport;
626 	sf_acts = rcu_dereference(flow->sf_acts);
627 
628 	local_bh_disable();
629 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
630 	local_bh_enable();
631 	rcu_read_unlock();
632 
633 	ovs_flow_free(flow, false);
634 	return err;
635 
636 err_unlock:
637 	rcu_read_unlock();
638 err_flow_free:
639 	ovs_flow_free(flow, false);
640 err_kfree_skb:
641 	kfree_skb(packet);
642 err:
643 	return err;
644 }
645 
646 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
647 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
648 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
649 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
650 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
651 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
652 	[OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
653 };
654 
655 static const struct genl_ops dp_packet_genl_ops[] = {
656 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
657 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
658 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
659 	  .doit = ovs_packet_cmd_execute
660 	}
661 };
662 
663 static struct genl_family dp_packet_genl_family __ro_after_init = {
664 	.hdrsize = sizeof(struct ovs_header),
665 	.name = OVS_PACKET_FAMILY,
666 	.version = OVS_PACKET_VERSION,
667 	.maxattr = OVS_PACKET_ATTR_MAX,
668 	.policy = packet_policy,
669 	.netnsok = true,
670 	.parallel_ops = true,
671 	.ops = dp_packet_genl_ops,
672 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
673 	.module = THIS_MODULE,
674 };
675 
676 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
677 			 struct ovs_dp_megaflow_stats *mega_stats)
678 {
679 	int i;
680 
681 	memset(mega_stats, 0, sizeof(*mega_stats));
682 
683 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
684 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
685 
686 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
687 
688 	for_each_possible_cpu(i) {
689 		const struct dp_stats_percpu *percpu_stats;
690 		struct dp_stats_percpu local_stats;
691 		unsigned int start;
692 
693 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
694 
695 		do {
696 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
697 			local_stats = *percpu_stats;
698 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
699 
700 		stats->n_hit += local_stats.n_hit;
701 		stats->n_missed += local_stats.n_missed;
702 		stats->n_lost += local_stats.n_lost;
703 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
704 		mega_stats->n_cache_hit += local_stats.n_cache_hit;
705 	}
706 }
707 
708 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
709 {
710 	return ovs_identifier_is_ufid(sfid) &&
711 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
712 }
713 
714 static bool should_fill_mask(uint32_t ufid_flags)
715 {
716 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
717 }
718 
719 static bool should_fill_actions(uint32_t ufid_flags)
720 {
721 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
722 }
723 
724 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
725 				    const struct sw_flow_id *sfid,
726 				    uint32_t ufid_flags)
727 {
728 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
729 
730 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
731 	 * see ovs_nla_put_identifier()
732 	 */
733 	if (sfid && ovs_identifier_is_ufid(sfid))
734 		len += nla_total_size(sfid->ufid_len);
735 	else
736 		len += nla_total_size(ovs_key_attr_size());
737 
738 	/* OVS_FLOW_ATTR_KEY */
739 	if (!sfid || should_fill_key(sfid, ufid_flags))
740 		len += nla_total_size(ovs_key_attr_size());
741 
742 	/* OVS_FLOW_ATTR_MASK */
743 	if (should_fill_mask(ufid_flags))
744 		len += nla_total_size(ovs_key_attr_size());
745 
746 	/* OVS_FLOW_ATTR_ACTIONS */
747 	if (should_fill_actions(ufid_flags))
748 		len += nla_total_size(acts->orig_len);
749 
750 	return len
751 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
752 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
753 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
754 }
755 
756 /* Called with ovs_mutex or RCU read lock. */
757 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
758 				   struct sk_buff *skb)
759 {
760 	struct ovs_flow_stats stats;
761 	__be16 tcp_flags;
762 	unsigned long used;
763 
764 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
765 
766 	if (used &&
767 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
768 			      OVS_FLOW_ATTR_PAD))
769 		return -EMSGSIZE;
770 
771 	if (stats.n_packets &&
772 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
773 			  sizeof(struct ovs_flow_stats), &stats,
774 			  OVS_FLOW_ATTR_PAD))
775 		return -EMSGSIZE;
776 
777 	if ((u8)ntohs(tcp_flags) &&
778 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
779 		return -EMSGSIZE;
780 
781 	return 0;
782 }
783 
784 /* Called with ovs_mutex or RCU read lock. */
785 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
786 				     struct sk_buff *skb, int skb_orig_len)
787 {
788 	struct nlattr *start;
789 	int err;
790 
791 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
792 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
793 	 * Netlink but individual action lists can be longer than
794 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
795 	 * The userspace caller can always fetch the actions separately if it
796 	 * really wants them.  (Most userspace callers in fact don't care.)
797 	 *
798 	 * This can only fail for dump operations because the skb is always
799 	 * properly sized for single flows.
800 	 */
801 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
802 	if (start) {
803 		const struct sw_flow_actions *sf_acts;
804 
805 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
806 		err = ovs_nla_put_actions(sf_acts->actions,
807 					  sf_acts->actions_len, skb);
808 
809 		if (!err)
810 			nla_nest_end(skb, start);
811 		else {
812 			if (skb_orig_len)
813 				return err;
814 
815 			nla_nest_cancel(skb, start);
816 		}
817 	} else if (skb_orig_len) {
818 		return -EMSGSIZE;
819 	}
820 
821 	return 0;
822 }
823 
824 /* Called with ovs_mutex or RCU read lock. */
825 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
826 				  struct sk_buff *skb, u32 portid,
827 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
828 {
829 	const int skb_orig_len = skb->len;
830 	struct ovs_header *ovs_header;
831 	int err;
832 
833 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
834 				 flags, cmd);
835 	if (!ovs_header)
836 		return -EMSGSIZE;
837 
838 	ovs_header->dp_ifindex = dp_ifindex;
839 
840 	err = ovs_nla_put_identifier(flow, skb);
841 	if (err)
842 		goto error;
843 
844 	if (should_fill_key(&flow->id, ufid_flags)) {
845 		err = ovs_nla_put_masked_key(flow, skb);
846 		if (err)
847 			goto error;
848 	}
849 
850 	if (should_fill_mask(ufid_flags)) {
851 		err = ovs_nla_put_mask(flow, skb);
852 		if (err)
853 			goto error;
854 	}
855 
856 	err = ovs_flow_cmd_fill_stats(flow, skb);
857 	if (err)
858 		goto error;
859 
860 	if (should_fill_actions(ufid_flags)) {
861 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
862 		if (err)
863 			goto error;
864 	}
865 
866 	genlmsg_end(skb, ovs_header);
867 	return 0;
868 
869 error:
870 	genlmsg_cancel(skb, ovs_header);
871 	return err;
872 }
873 
874 /* May not be called with RCU read lock. */
875 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
876 					       const struct sw_flow_id *sfid,
877 					       struct genl_info *info,
878 					       bool always,
879 					       uint32_t ufid_flags)
880 {
881 	struct sk_buff *skb;
882 	size_t len;
883 
884 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
885 		return NULL;
886 
887 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
888 	skb = genlmsg_new(len, GFP_KERNEL);
889 	if (!skb)
890 		return ERR_PTR(-ENOMEM);
891 
892 	return skb;
893 }
894 
895 /* Called with ovs_mutex. */
896 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
897 					       int dp_ifindex,
898 					       struct genl_info *info, u8 cmd,
899 					       bool always, u32 ufid_flags)
900 {
901 	struct sk_buff *skb;
902 	int retval;
903 
904 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
905 				      &flow->id, info, always, ufid_flags);
906 	if (IS_ERR_OR_NULL(skb))
907 		return skb;
908 
909 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
910 					info->snd_portid, info->snd_seq, 0,
911 					cmd, ufid_flags);
912 	if (WARN_ON_ONCE(retval < 0)) {
913 		kfree_skb(skb);
914 		skb = ERR_PTR(retval);
915 	}
916 	return skb;
917 }
918 
919 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
920 {
921 	struct net *net = sock_net(skb->sk);
922 	struct nlattr **a = info->attrs;
923 	struct ovs_header *ovs_header = info->userhdr;
924 	struct sw_flow *flow = NULL, *new_flow;
925 	struct sw_flow_mask mask;
926 	struct sk_buff *reply;
927 	struct datapath *dp;
928 	struct sw_flow_actions *acts;
929 	struct sw_flow_match match;
930 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
931 	int error;
932 	bool log = !a[OVS_FLOW_ATTR_PROBE];
933 
934 	/* Must have key and actions. */
935 	error = -EINVAL;
936 	if (!a[OVS_FLOW_ATTR_KEY]) {
937 		OVS_NLERR(log, "Flow key attr not present in new flow.");
938 		goto error;
939 	}
940 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
941 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
942 		goto error;
943 	}
944 
945 	/* Most of the time we need to allocate a new flow, do it before
946 	 * locking.
947 	 */
948 	new_flow = ovs_flow_alloc();
949 	if (IS_ERR(new_flow)) {
950 		error = PTR_ERR(new_flow);
951 		goto error;
952 	}
953 
954 	/* Extract key. */
955 	ovs_match_init(&match, &new_flow->key, false, &mask);
956 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
957 				  a[OVS_FLOW_ATTR_MASK], log);
958 	if (error)
959 		goto err_kfree_flow;
960 
961 	/* Extract flow identifier. */
962 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
963 				       &new_flow->key, log);
964 	if (error)
965 		goto err_kfree_flow;
966 
967 	/* unmasked key is needed to match when ufid is not used. */
968 	if (ovs_identifier_is_key(&new_flow->id))
969 		match.key = new_flow->id.unmasked_key;
970 
971 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
972 
973 	/* Validate actions. */
974 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
975 				     &new_flow->key, &acts, log);
976 	if (error) {
977 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
978 		goto err_kfree_flow;
979 	}
980 
981 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
982 					ufid_flags);
983 	if (IS_ERR(reply)) {
984 		error = PTR_ERR(reply);
985 		goto err_kfree_acts;
986 	}
987 
988 	ovs_lock();
989 	dp = get_dp(net, ovs_header->dp_ifindex);
990 	if (unlikely(!dp)) {
991 		error = -ENODEV;
992 		goto err_unlock_ovs;
993 	}
994 
995 	/* Check if this is a duplicate flow */
996 	if (ovs_identifier_is_ufid(&new_flow->id))
997 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
998 	if (!flow)
999 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1000 	if (likely(!flow)) {
1001 		rcu_assign_pointer(new_flow->sf_acts, acts);
1002 
1003 		/* Put flow in bucket. */
1004 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1005 		if (unlikely(error)) {
1006 			acts = NULL;
1007 			goto err_unlock_ovs;
1008 		}
1009 
1010 		if (unlikely(reply)) {
1011 			error = ovs_flow_cmd_fill_info(new_flow,
1012 						       ovs_header->dp_ifindex,
1013 						       reply, info->snd_portid,
1014 						       info->snd_seq, 0,
1015 						       OVS_FLOW_CMD_NEW,
1016 						       ufid_flags);
1017 			BUG_ON(error < 0);
1018 		}
1019 		ovs_unlock();
1020 	} else {
1021 		struct sw_flow_actions *old_acts;
1022 
1023 		/* Bail out if we're not allowed to modify an existing flow.
1024 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1025 		 * because Generic Netlink treats the latter as a dump
1026 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1027 		 * gets fixed.
1028 		 */
1029 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1030 							 | NLM_F_EXCL))) {
1031 			error = -EEXIST;
1032 			goto err_unlock_ovs;
1033 		}
1034 		/* The flow identifier has to be the same for flow updates.
1035 		 * Look for any overlapping flow.
1036 		 */
1037 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1038 			if (ovs_identifier_is_key(&flow->id))
1039 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1040 								 &match);
1041 			else /* UFID matches but key is different */
1042 				flow = NULL;
1043 			if (!flow) {
1044 				error = -ENOENT;
1045 				goto err_unlock_ovs;
1046 			}
1047 		}
1048 		/* Update actions. */
1049 		old_acts = ovsl_dereference(flow->sf_acts);
1050 		rcu_assign_pointer(flow->sf_acts, acts);
1051 
1052 		if (unlikely(reply)) {
1053 			error = ovs_flow_cmd_fill_info(flow,
1054 						       ovs_header->dp_ifindex,
1055 						       reply, info->snd_portid,
1056 						       info->snd_seq, 0,
1057 						       OVS_FLOW_CMD_NEW,
1058 						       ufid_flags);
1059 			BUG_ON(error < 0);
1060 		}
1061 		ovs_unlock();
1062 
1063 		ovs_nla_free_flow_actions_rcu(old_acts);
1064 		ovs_flow_free(new_flow, false);
1065 	}
1066 
1067 	if (reply)
1068 		ovs_notify(&dp_flow_genl_family, reply, info);
1069 	return 0;
1070 
1071 err_unlock_ovs:
1072 	ovs_unlock();
1073 	kfree_skb(reply);
1074 err_kfree_acts:
1075 	ovs_nla_free_flow_actions(acts);
1076 err_kfree_flow:
1077 	ovs_flow_free(new_flow, false);
1078 error:
1079 	return error;
1080 }
1081 
1082 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1083 static noinline_for_stack struct sw_flow_actions *get_flow_actions(struct net *net,
1084 						const struct nlattr *a,
1085 						const struct sw_flow_key *key,
1086 						const struct sw_flow_mask *mask,
1087 						bool log)
1088 {
1089 	struct sw_flow_actions *acts;
1090 	struct sw_flow_key masked_key;
1091 	int error;
1092 
1093 	ovs_flow_mask_key(&masked_key, key, true, mask);
1094 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1095 	if (error) {
1096 		OVS_NLERR(log,
1097 			  "Actions may not be safe on all matching packets");
1098 		return ERR_PTR(error);
1099 	}
1100 
1101 	return acts;
1102 }
1103 
1104 /* Factor out match-init and action-copy to avoid
1105  * "Wframe-larger-than=1024" warning. Because mask is only
1106  * used to get actions, we new a function to save some
1107  * stack space.
1108  *
1109  * If there are not key and action attrs, we return 0
1110  * directly. In the case, the caller will also not use the
1111  * match as before. If there is action attr, we try to get
1112  * actions and save them to *acts. Before returning from
1113  * the function, we reset the match->mask pointer. Because
1114  * we should not to return match object with dangling reference
1115  * to mask.
1116  * */
1117 static noinline_for_stack int
1118 ovs_nla_init_match_and_action(struct net *net,
1119 			      struct sw_flow_match *match,
1120 			      struct sw_flow_key *key,
1121 			      struct nlattr **a,
1122 			      struct sw_flow_actions **acts,
1123 			      bool log)
1124 {
1125 	struct sw_flow_mask mask;
1126 	int error = 0;
1127 
1128 	if (a[OVS_FLOW_ATTR_KEY]) {
1129 		ovs_match_init(match, key, true, &mask);
1130 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1131 					  a[OVS_FLOW_ATTR_MASK], log);
1132 		if (error)
1133 			goto error;
1134 	}
1135 
1136 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1137 		if (!a[OVS_FLOW_ATTR_KEY]) {
1138 			OVS_NLERR(log,
1139 				  "Flow key attribute not present in set flow.");
1140 			error = -EINVAL;
1141 			goto error;
1142 		}
1143 
1144 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1145 					 &mask, log);
1146 		if (IS_ERR(*acts)) {
1147 			error = PTR_ERR(*acts);
1148 			goto error;
1149 		}
1150 	}
1151 
1152 	/* On success, error is 0. */
1153 error:
1154 	match->mask = NULL;
1155 	return error;
1156 }
1157 
1158 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1159 {
1160 	struct net *net = sock_net(skb->sk);
1161 	struct nlattr **a = info->attrs;
1162 	struct ovs_header *ovs_header = info->userhdr;
1163 	struct sw_flow_key key;
1164 	struct sw_flow *flow;
1165 	struct sk_buff *reply = NULL;
1166 	struct datapath *dp;
1167 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1168 	struct sw_flow_match match;
1169 	struct sw_flow_id sfid;
1170 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1171 	int error = 0;
1172 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1173 	bool ufid_present;
1174 
1175 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1176 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1177 		OVS_NLERR(log,
1178 			  "Flow set message rejected, Key attribute missing.");
1179 		return -EINVAL;
1180 	}
1181 
1182 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1183 					      &acts, log);
1184 	if (error)
1185 		goto error;
1186 
1187 	if (acts) {
1188 		/* Can allocate before locking if have acts. */
1189 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1190 						ufid_flags);
1191 		if (IS_ERR(reply)) {
1192 			error = PTR_ERR(reply);
1193 			goto err_kfree_acts;
1194 		}
1195 	}
1196 
1197 	ovs_lock();
1198 	dp = get_dp(net, ovs_header->dp_ifindex);
1199 	if (unlikely(!dp)) {
1200 		error = -ENODEV;
1201 		goto err_unlock_ovs;
1202 	}
1203 	/* Check that the flow exists. */
1204 	if (ufid_present)
1205 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1206 	else
1207 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1208 	if (unlikely(!flow)) {
1209 		error = -ENOENT;
1210 		goto err_unlock_ovs;
1211 	}
1212 
1213 	/* Update actions, if present. */
1214 	if (likely(acts)) {
1215 		old_acts = ovsl_dereference(flow->sf_acts);
1216 		rcu_assign_pointer(flow->sf_acts, acts);
1217 
1218 		if (unlikely(reply)) {
1219 			error = ovs_flow_cmd_fill_info(flow,
1220 						       ovs_header->dp_ifindex,
1221 						       reply, info->snd_portid,
1222 						       info->snd_seq, 0,
1223 						       OVS_FLOW_CMD_SET,
1224 						       ufid_flags);
1225 			BUG_ON(error < 0);
1226 		}
1227 	} else {
1228 		/* Could not alloc without acts before locking. */
1229 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1230 						info, OVS_FLOW_CMD_SET, false,
1231 						ufid_flags);
1232 
1233 		if (IS_ERR(reply)) {
1234 			error = PTR_ERR(reply);
1235 			goto err_unlock_ovs;
1236 		}
1237 	}
1238 
1239 	/* Clear stats. */
1240 	if (a[OVS_FLOW_ATTR_CLEAR])
1241 		ovs_flow_stats_clear(flow);
1242 	ovs_unlock();
1243 
1244 	if (reply)
1245 		ovs_notify(&dp_flow_genl_family, reply, info);
1246 	if (old_acts)
1247 		ovs_nla_free_flow_actions_rcu(old_acts);
1248 
1249 	return 0;
1250 
1251 err_unlock_ovs:
1252 	ovs_unlock();
1253 	kfree_skb(reply);
1254 err_kfree_acts:
1255 	ovs_nla_free_flow_actions(acts);
1256 error:
1257 	return error;
1258 }
1259 
1260 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1261 {
1262 	struct nlattr **a = info->attrs;
1263 	struct ovs_header *ovs_header = info->userhdr;
1264 	struct net *net = sock_net(skb->sk);
1265 	struct sw_flow_key key;
1266 	struct sk_buff *reply;
1267 	struct sw_flow *flow;
1268 	struct datapath *dp;
1269 	struct sw_flow_match match;
1270 	struct sw_flow_id ufid;
1271 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1272 	int err = 0;
1273 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1274 	bool ufid_present;
1275 
1276 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1277 	if (a[OVS_FLOW_ATTR_KEY]) {
1278 		ovs_match_init(&match, &key, true, NULL);
1279 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1280 					log);
1281 	} else if (!ufid_present) {
1282 		OVS_NLERR(log,
1283 			  "Flow get message rejected, Key attribute missing.");
1284 		err = -EINVAL;
1285 	}
1286 	if (err)
1287 		return err;
1288 
1289 	ovs_lock();
1290 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1291 	if (!dp) {
1292 		err = -ENODEV;
1293 		goto unlock;
1294 	}
1295 
1296 	if (ufid_present)
1297 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1298 	else
1299 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1300 	if (!flow) {
1301 		err = -ENOENT;
1302 		goto unlock;
1303 	}
1304 
1305 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1306 					OVS_FLOW_CMD_GET, true, ufid_flags);
1307 	if (IS_ERR(reply)) {
1308 		err = PTR_ERR(reply);
1309 		goto unlock;
1310 	}
1311 
1312 	ovs_unlock();
1313 	return genlmsg_reply(reply, info);
1314 unlock:
1315 	ovs_unlock();
1316 	return err;
1317 }
1318 
1319 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1320 {
1321 	struct nlattr **a = info->attrs;
1322 	struct ovs_header *ovs_header = info->userhdr;
1323 	struct net *net = sock_net(skb->sk);
1324 	struct sw_flow_key key;
1325 	struct sk_buff *reply;
1326 	struct sw_flow *flow = NULL;
1327 	struct datapath *dp;
1328 	struct sw_flow_match match;
1329 	struct sw_flow_id ufid;
1330 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1331 	int err;
1332 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1333 	bool ufid_present;
1334 
1335 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1336 	if (a[OVS_FLOW_ATTR_KEY]) {
1337 		ovs_match_init(&match, &key, true, NULL);
1338 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1339 					NULL, log);
1340 		if (unlikely(err))
1341 			return err;
1342 	}
1343 
1344 	ovs_lock();
1345 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1346 	if (unlikely(!dp)) {
1347 		err = -ENODEV;
1348 		goto unlock;
1349 	}
1350 
1351 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1352 		err = ovs_flow_tbl_flush(&dp->table);
1353 		goto unlock;
1354 	}
1355 
1356 	if (ufid_present)
1357 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1358 	else
1359 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1360 	if (unlikely(!flow)) {
1361 		err = -ENOENT;
1362 		goto unlock;
1363 	}
1364 
1365 	ovs_flow_tbl_remove(&dp->table, flow);
1366 	ovs_unlock();
1367 
1368 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1369 					&flow->id, info, false, ufid_flags);
1370 	if (likely(reply)) {
1371 		if (!IS_ERR(reply)) {
1372 			rcu_read_lock();	/*To keep RCU checker happy. */
1373 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1374 						     reply, info->snd_portid,
1375 						     info->snd_seq, 0,
1376 						     OVS_FLOW_CMD_DEL,
1377 						     ufid_flags);
1378 			rcu_read_unlock();
1379 			if (WARN_ON_ONCE(err < 0)) {
1380 				kfree_skb(reply);
1381 				goto out_free;
1382 			}
1383 
1384 			ovs_notify(&dp_flow_genl_family, reply, info);
1385 		} else {
1386 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1387 		}
1388 	}
1389 
1390 out_free:
1391 	ovs_flow_free(flow, true);
1392 	return 0;
1393 unlock:
1394 	ovs_unlock();
1395 	return err;
1396 }
1397 
1398 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1399 {
1400 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1401 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1402 	struct table_instance *ti;
1403 	struct datapath *dp;
1404 	u32 ufid_flags;
1405 	int err;
1406 
1407 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1408 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1409 	if (err)
1410 		return err;
1411 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1412 
1413 	rcu_read_lock();
1414 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1415 	if (!dp) {
1416 		rcu_read_unlock();
1417 		return -ENODEV;
1418 	}
1419 
1420 	ti = rcu_dereference(dp->table.ti);
1421 	for (;;) {
1422 		struct sw_flow *flow;
1423 		u32 bucket, obj;
1424 
1425 		bucket = cb->args[0];
1426 		obj = cb->args[1];
1427 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1428 		if (!flow)
1429 			break;
1430 
1431 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1432 					   NETLINK_CB(cb->skb).portid,
1433 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1434 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1435 			break;
1436 
1437 		cb->args[0] = bucket;
1438 		cb->args[1] = obj;
1439 	}
1440 	rcu_read_unlock();
1441 	return skb->len;
1442 }
1443 
1444 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1445 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1446 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1447 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1448 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1449 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1450 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1451 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1452 };
1453 
1454 static const struct genl_ops dp_flow_genl_ops[] = {
1455 	{ .cmd = OVS_FLOW_CMD_NEW,
1456 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1457 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1458 	  .doit = ovs_flow_cmd_new
1459 	},
1460 	{ .cmd = OVS_FLOW_CMD_DEL,
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_del
1464 	},
1465 	{ .cmd = OVS_FLOW_CMD_GET,
1466 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1467 	  .flags = 0,		    /* OK for unprivileged users. */
1468 	  .doit = ovs_flow_cmd_get,
1469 	  .dumpit = ovs_flow_cmd_dump
1470 	},
1471 	{ .cmd = OVS_FLOW_CMD_SET,
1472 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1473 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1474 	  .doit = ovs_flow_cmd_set,
1475 	},
1476 };
1477 
1478 static struct genl_family dp_flow_genl_family __ro_after_init = {
1479 	.hdrsize = sizeof(struct ovs_header),
1480 	.name = OVS_FLOW_FAMILY,
1481 	.version = OVS_FLOW_VERSION,
1482 	.maxattr = OVS_FLOW_ATTR_MAX,
1483 	.policy = flow_policy,
1484 	.netnsok = true,
1485 	.parallel_ops = true,
1486 	.ops = dp_flow_genl_ops,
1487 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1488 	.mcgrps = &ovs_dp_flow_multicast_group,
1489 	.n_mcgrps = 1,
1490 	.module = THIS_MODULE,
1491 };
1492 
1493 static size_t ovs_dp_cmd_msg_size(void)
1494 {
1495 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1496 
1497 	msgsize += nla_total_size(IFNAMSIZ);
1498 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1499 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1500 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1501 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1502 
1503 	return msgsize;
1504 }
1505 
1506 /* Called with ovs_mutex. */
1507 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1508 				u32 portid, u32 seq, u32 flags, u8 cmd)
1509 {
1510 	struct ovs_header *ovs_header;
1511 	struct ovs_dp_stats dp_stats;
1512 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1513 	int err;
1514 
1515 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1516 				   flags, cmd);
1517 	if (!ovs_header)
1518 		goto error;
1519 
1520 	ovs_header->dp_ifindex = get_dpifindex(dp);
1521 
1522 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1523 	if (err)
1524 		goto nla_put_failure;
1525 
1526 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1527 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1528 			  &dp_stats, OVS_DP_ATTR_PAD))
1529 		goto nla_put_failure;
1530 
1531 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1532 			  sizeof(struct ovs_dp_megaflow_stats),
1533 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1534 		goto nla_put_failure;
1535 
1536 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1537 		goto nla_put_failure;
1538 
1539 	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1540 			ovs_flow_tbl_masks_cache_size(&dp->table)))
1541 		goto nla_put_failure;
1542 
1543 	genlmsg_end(skb, ovs_header);
1544 	return 0;
1545 
1546 nla_put_failure:
1547 	genlmsg_cancel(skb, ovs_header);
1548 error:
1549 	return -EMSGSIZE;
1550 }
1551 
1552 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1553 {
1554 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1555 }
1556 
1557 /* Called with rcu_read_lock or ovs_mutex. */
1558 static struct datapath *lookup_datapath(struct net *net,
1559 					const struct ovs_header *ovs_header,
1560 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1561 {
1562 	struct datapath *dp;
1563 
1564 	if (!a[OVS_DP_ATTR_NAME])
1565 		dp = get_dp(net, ovs_header->dp_ifindex);
1566 	else {
1567 		struct vport *vport;
1568 
1569 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1570 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1571 	}
1572 	return dp ? dp : ERR_PTR(-ENODEV);
1573 }
1574 
1575 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1576 {
1577 	struct datapath *dp;
1578 
1579 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1580 	if (IS_ERR(dp))
1581 		return;
1582 
1583 	WARN(dp->user_features, "Dropping previously announced user features\n");
1584 	dp->user_features = 0;
1585 }
1586 
1587 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1588 
1589 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1590 {
1591 	u32 user_features = 0;
1592 
1593 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1594 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1595 
1596 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1597 				      OVS_DP_F_UNALIGNED |
1598 				      OVS_DP_F_TC_RECIRC_SHARING))
1599 			return -EOPNOTSUPP;
1600 
1601 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1602 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1603 			return -EOPNOTSUPP;
1604 #endif
1605 	}
1606 
1607 	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1608 		int err;
1609 		u32 cache_size;
1610 
1611 		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1612 		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1613 		if (err)
1614 			return err;
1615 	}
1616 
1617 	dp->user_features = user_features;
1618 
1619 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1620 		static_branch_enable(&tc_recirc_sharing_support);
1621 	else
1622 		static_branch_disable(&tc_recirc_sharing_support);
1623 
1624 	return 0;
1625 }
1626 
1627 static int ovs_dp_stats_init(struct datapath *dp)
1628 {
1629 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1630 	if (!dp->stats_percpu)
1631 		return -ENOMEM;
1632 
1633 	return 0;
1634 }
1635 
1636 static int ovs_dp_vport_init(struct datapath *dp)
1637 {
1638 	int i;
1639 
1640 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1641 				  sizeof(struct hlist_head),
1642 				  GFP_KERNEL);
1643 	if (!dp->ports)
1644 		return -ENOMEM;
1645 
1646 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1647 		INIT_HLIST_HEAD(&dp->ports[i]);
1648 
1649 	return 0;
1650 }
1651 
1652 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1653 {
1654 	struct nlattr **a = info->attrs;
1655 	struct vport_parms parms;
1656 	struct sk_buff *reply;
1657 	struct datapath *dp;
1658 	struct vport *vport;
1659 	struct ovs_net *ovs_net;
1660 	int err;
1661 
1662 	err = -EINVAL;
1663 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1664 		goto err;
1665 
1666 	reply = ovs_dp_cmd_alloc_info();
1667 	if (!reply)
1668 		return -ENOMEM;
1669 
1670 	err = -ENOMEM;
1671 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1672 	if (dp == NULL)
1673 		goto err_destroy_reply;
1674 
1675 	ovs_dp_set_net(dp, sock_net(skb->sk));
1676 
1677 	/* Allocate table. */
1678 	err = ovs_flow_tbl_init(&dp->table);
1679 	if (err)
1680 		goto err_destroy_dp;
1681 
1682 	err = ovs_dp_stats_init(dp);
1683 	if (err)
1684 		goto err_destroy_table;
1685 
1686 	err = ovs_dp_vport_init(dp);
1687 	if (err)
1688 		goto err_destroy_stats;
1689 
1690 	err = ovs_meters_init(dp);
1691 	if (err)
1692 		goto err_destroy_ports;
1693 
1694 	/* Set up our datapath device. */
1695 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1696 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1697 	parms.options = NULL;
1698 	parms.dp = dp;
1699 	parms.port_no = OVSP_LOCAL;
1700 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1701 
1702 	err = ovs_dp_change(dp, a);
1703 	if (err)
1704 		goto err_destroy_meters;
1705 
1706 	/* So far only local changes have been made, now need the lock. */
1707 	ovs_lock();
1708 
1709 	vport = new_vport(&parms);
1710 	if (IS_ERR(vport)) {
1711 		err = PTR_ERR(vport);
1712 		if (err == -EBUSY)
1713 			err = -EEXIST;
1714 
1715 		if (err == -EEXIST) {
1716 			/* An outdated user space instance that does not understand
1717 			 * the concept of user_features has attempted to create a new
1718 			 * datapath and is likely to reuse it. Drop all user features.
1719 			 */
1720 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1721 				ovs_dp_reset_user_features(skb, info);
1722 		}
1723 
1724 		ovs_unlock();
1725 		goto err_destroy_meters;
1726 	}
1727 
1728 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1729 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1730 	BUG_ON(err < 0);
1731 
1732 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1733 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1734 
1735 	ovs_unlock();
1736 
1737 	ovs_notify(&dp_datapath_genl_family, reply, info);
1738 	return 0;
1739 
1740 err_destroy_meters:
1741 	ovs_meters_exit(dp);
1742 err_destroy_ports:
1743 	kfree(dp->ports);
1744 err_destroy_stats:
1745 	free_percpu(dp->stats_percpu);
1746 err_destroy_table:
1747 	ovs_flow_tbl_destroy(&dp->table);
1748 err_destroy_dp:
1749 	kfree(dp);
1750 err_destroy_reply:
1751 	kfree_skb(reply);
1752 err:
1753 	return err;
1754 }
1755 
1756 /* Called with ovs_mutex. */
1757 static void __dp_destroy(struct datapath *dp)
1758 {
1759 	struct flow_table *table = &dp->table;
1760 	int i;
1761 
1762 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1763 		struct vport *vport;
1764 		struct hlist_node *n;
1765 
1766 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1767 			if (vport->port_no != OVSP_LOCAL)
1768 				ovs_dp_detach_port(vport);
1769 	}
1770 
1771 	list_del_rcu(&dp->list_node);
1772 
1773 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1774 	 * all ports in datapath are destroyed first before freeing datapath.
1775 	 */
1776 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1777 
1778 	/* Flush sw_flow in the tables. RCU cb only releases resource
1779 	 * such as dp, ports and tables. That may avoid some issues
1780 	 * such as RCU usage warning.
1781 	 */
1782 	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1783 				  ovsl_dereference(table->ufid_ti));
1784 
1785 	/* RCU destroy the ports, meters and flow tables. */
1786 	call_rcu(&dp->rcu, destroy_dp_rcu);
1787 }
1788 
1789 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1790 {
1791 	struct sk_buff *reply;
1792 	struct datapath *dp;
1793 	int err;
1794 
1795 	reply = ovs_dp_cmd_alloc_info();
1796 	if (!reply)
1797 		return -ENOMEM;
1798 
1799 	ovs_lock();
1800 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1801 	err = PTR_ERR(dp);
1802 	if (IS_ERR(dp))
1803 		goto err_unlock_free;
1804 
1805 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1806 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1807 	BUG_ON(err < 0);
1808 
1809 	__dp_destroy(dp);
1810 	ovs_unlock();
1811 
1812 	ovs_notify(&dp_datapath_genl_family, reply, info);
1813 
1814 	return 0;
1815 
1816 err_unlock_free:
1817 	ovs_unlock();
1818 	kfree_skb(reply);
1819 	return err;
1820 }
1821 
1822 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1823 {
1824 	struct sk_buff *reply;
1825 	struct datapath *dp;
1826 	int err;
1827 
1828 	reply = ovs_dp_cmd_alloc_info();
1829 	if (!reply)
1830 		return -ENOMEM;
1831 
1832 	ovs_lock();
1833 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1834 	err = PTR_ERR(dp);
1835 	if (IS_ERR(dp))
1836 		goto err_unlock_free;
1837 
1838 	err = ovs_dp_change(dp, info->attrs);
1839 	if (err)
1840 		goto err_unlock_free;
1841 
1842 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1843 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1844 	BUG_ON(err < 0);
1845 
1846 	ovs_unlock();
1847 	ovs_notify(&dp_datapath_genl_family, reply, info);
1848 
1849 	return 0;
1850 
1851 err_unlock_free:
1852 	ovs_unlock();
1853 	kfree_skb(reply);
1854 	return err;
1855 }
1856 
1857 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1858 {
1859 	struct sk_buff *reply;
1860 	struct datapath *dp;
1861 	int err;
1862 
1863 	reply = ovs_dp_cmd_alloc_info();
1864 	if (!reply)
1865 		return -ENOMEM;
1866 
1867 	ovs_lock();
1868 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1869 	if (IS_ERR(dp)) {
1870 		err = PTR_ERR(dp);
1871 		goto err_unlock_free;
1872 	}
1873 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1874 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1875 	BUG_ON(err < 0);
1876 	ovs_unlock();
1877 
1878 	return genlmsg_reply(reply, info);
1879 
1880 err_unlock_free:
1881 	ovs_unlock();
1882 	kfree_skb(reply);
1883 	return err;
1884 }
1885 
1886 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1887 {
1888 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1889 	struct datapath *dp;
1890 	int skip = cb->args[0];
1891 	int i = 0;
1892 
1893 	ovs_lock();
1894 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1895 		if (i >= skip &&
1896 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1897 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1898 					 OVS_DP_CMD_GET) < 0)
1899 			break;
1900 		i++;
1901 	}
1902 	ovs_unlock();
1903 
1904 	cb->args[0] = i;
1905 
1906 	return skb->len;
1907 }
1908 
1909 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1910 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1911 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1912 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1913 	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
1914 		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
1915 };
1916 
1917 static const struct genl_ops dp_datapath_genl_ops[] = {
1918 	{ .cmd = OVS_DP_CMD_NEW,
1919 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1920 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1921 	  .doit = ovs_dp_cmd_new
1922 	},
1923 	{ .cmd = OVS_DP_CMD_DEL,
1924 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1925 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1926 	  .doit = ovs_dp_cmd_del
1927 	},
1928 	{ .cmd = OVS_DP_CMD_GET,
1929 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1930 	  .flags = 0,		    /* OK for unprivileged users. */
1931 	  .doit = ovs_dp_cmd_get,
1932 	  .dumpit = ovs_dp_cmd_dump
1933 	},
1934 	{ .cmd = OVS_DP_CMD_SET,
1935 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1936 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1937 	  .doit = ovs_dp_cmd_set,
1938 	},
1939 };
1940 
1941 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1942 	.hdrsize = sizeof(struct ovs_header),
1943 	.name = OVS_DATAPATH_FAMILY,
1944 	.version = OVS_DATAPATH_VERSION,
1945 	.maxattr = OVS_DP_ATTR_MAX,
1946 	.policy = datapath_policy,
1947 	.netnsok = true,
1948 	.parallel_ops = true,
1949 	.ops = dp_datapath_genl_ops,
1950 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1951 	.mcgrps = &ovs_dp_datapath_multicast_group,
1952 	.n_mcgrps = 1,
1953 	.module = THIS_MODULE,
1954 };
1955 
1956 /* Called with ovs_mutex or RCU read lock. */
1957 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1958 				   struct net *net, u32 portid, u32 seq,
1959 				   u32 flags, u8 cmd, gfp_t gfp)
1960 {
1961 	struct ovs_header *ovs_header;
1962 	struct ovs_vport_stats vport_stats;
1963 	int err;
1964 
1965 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1966 				 flags, cmd);
1967 	if (!ovs_header)
1968 		return -EMSGSIZE;
1969 
1970 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1971 
1972 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1973 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1974 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1975 			   ovs_vport_name(vport)) ||
1976 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1977 		goto nla_put_failure;
1978 
1979 	if (!net_eq(net, dev_net(vport->dev))) {
1980 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1981 
1982 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1983 			goto nla_put_failure;
1984 	}
1985 
1986 	ovs_vport_get_stats(vport, &vport_stats);
1987 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1988 			  sizeof(struct ovs_vport_stats), &vport_stats,
1989 			  OVS_VPORT_ATTR_PAD))
1990 		goto nla_put_failure;
1991 
1992 	if (ovs_vport_get_upcall_portids(vport, skb))
1993 		goto nla_put_failure;
1994 
1995 	err = ovs_vport_get_options(vport, skb);
1996 	if (err == -EMSGSIZE)
1997 		goto error;
1998 
1999 	genlmsg_end(skb, ovs_header);
2000 	return 0;
2001 
2002 nla_put_failure:
2003 	err = -EMSGSIZE;
2004 error:
2005 	genlmsg_cancel(skb, ovs_header);
2006 	return err;
2007 }
2008 
2009 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2010 {
2011 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2012 }
2013 
2014 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2015 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2016 					 u32 portid, u32 seq, u8 cmd)
2017 {
2018 	struct sk_buff *skb;
2019 	int retval;
2020 
2021 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2022 	if (!skb)
2023 		return ERR_PTR(-ENOMEM);
2024 
2025 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2026 					 GFP_KERNEL);
2027 	BUG_ON(retval < 0);
2028 
2029 	return skb;
2030 }
2031 
2032 /* Called with ovs_mutex or RCU read lock. */
2033 static struct vport *lookup_vport(struct net *net,
2034 				  const struct ovs_header *ovs_header,
2035 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2036 {
2037 	struct datapath *dp;
2038 	struct vport *vport;
2039 
2040 	if (a[OVS_VPORT_ATTR_IFINDEX])
2041 		return ERR_PTR(-EOPNOTSUPP);
2042 	if (a[OVS_VPORT_ATTR_NAME]) {
2043 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2044 		if (!vport)
2045 			return ERR_PTR(-ENODEV);
2046 		if (ovs_header->dp_ifindex &&
2047 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2048 			return ERR_PTR(-ENODEV);
2049 		return vport;
2050 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2051 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2052 
2053 		if (port_no >= DP_MAX_PORTS)
2054 			return ERR_PTR(-EFBIG);
2055 
2056 		dp = get_dp(net, ovs_header->dp_ifindex);
2057 		if (!dp)
2058 			return ERR_PTR(-ENODEV);
2059 
2060 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2061 		if (!vport)
2062 			return ERR_PTR(-ENODEV);
2063 		return vport;
2064 	} else
2065 		return ERR_PTR(-EINVAL);
2066 
2067 }
2068 
2069 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2070 {
2071 	unsigned int dev_headroom, max_headroom = 0;
2072 	struct net_device *dev;
2073 	struct vport *vport;
2074 	int i;
2075 
2076 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2077 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2078 					lockdep_ovsl_is_held()) {
2079 			dev = vport->dev;
2080 			dev_headroom = netdev_get_fwd_headroom(dev);
2081 			if (dev_headroom > max_headroom)
2082 				max_headroom = dev_headroom;
2083 		}
2084 	}
2085 
2086 	return max_headroom;
2087 }
2088 
2089 /* Called with ovs_mutex */
2090 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2091 {
2092 	struct vport *vport;
2093 	int i;
2094 
2095 	dp->max_headroom = new_headroom;
2096 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
2097 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2098 					lockdep_ovsl_is_held())
2099 			netdev_set_rx_headroom(vport->dev, new_headroom);
2100 }
2101 
2102 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2103 {
2104 	struct nlattr **a = info->attrs;
2105 	struct ovs_header *ovs_header = info->userhdr;
2106 	struct vport_parms parms;
2107 	struct sk_buff *reply;
2108 	struct vport *vport;
2109 	struct datapath *dp;
2110 	unsigned int new_headroom;
2111 	u32 port_no;
2112 	int err;
2113 
2114 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2115 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2116 		return -EINVAL;
2117 	if (a[OVS_VPORT_ATTR_IFINDEX])
2118 		return -EOPNOTSUPP;
2119 
2120 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2121 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2122 	if (port_no >= DP_MAX_PORTS)
2123 		return -EFBIG;
2124 
2125 	reply = ovs_vport_cmd_alloc_info();
2126 	if (!reply)
2127 		return -ENOMEM;
2128 
2129 	ovs_lock();
2130 restart:
2131 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2132 	err = -ENODEV;
2133 	if (!dp)
2134 		goto exit_unlock_free;
2135 
2136 	if (port_no) {
2137 		vport = ovs_vport_ovsl(dp, port_no);
2138 		err = -EBUSY;
2139 		if (vport)
2140 			goto exit_unlock_free;
2141 	} else {
2142 		for (port_no = 1; ; port_no++) {
2143 			if (port_no >= DP_MAX_PORTS) {
2144 				err = -EFBIG;
2145 				goto exit_unlock_free;
2146 			}
2147 			vport = ovs_vport_ovsl(dp, port_no);
2148 			if (!vport)
2149 				break;
2150 		}
2151 	}
2152 
2153 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2154 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2155 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2156 	parms.dp = dp;
2157 	parms.port_no = port_no;
2158 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2159 
2160 	vport = new_vport(&parms);
2161 	err = PTR_ERR(vport);
2162 	if (IS_ERR(vport)) {
2163 		if (err == -EAGAIN)
2164 			goto restart;
2165 		goto exit_unlock_free;
2166 	}
2167 
2168 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2169 				      info->snd_portid, info->snd_seq, 0,
2170 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2171 
2172 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2173 
2174 	if (new_headroom > dp->max_headroom)
2175 		ovs_update_headroom(dp, new_headroom);
2176 	else
2177 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2178 
2179 	BUG_ON(err < 0);
2180 	ovs_unlock();
2181 
2182 	ovs_notify(&dp_vport_genl_family, reply, info);
2183 	return 0;
2184 
2185 exit_unlock_free:
2186 	ovs_unlock();
2187 	kfree_skb(reply);
2188 	return err;
2189 }
2190 
2191 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2192 {
2193 	struct nlattr **a = info->attrs;
2194 	struct sk_buff *reply;
2195 	struct vport *vport;
2196 	int err;
2197 
2198 	reply = ovs_vport_cmd_alloc_info();
2199 	if (!reply)
2200 		return -ENOMEM;
2201 
2202 	ovs_lock();
2203 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2204 	err = PTR_ERR(vport);
2205 	if (IS_ERR(vport))
2206 		goto exit_unlock_free;
2207 
2208 	if (a[OVS_VPORT_ATTR_TYPE] &&
2209 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2210 		err = -EINVAL;
2211 		goto exit_unlock_free;
2212 	}
2213 
2214 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2215 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2216 		if (err)
2217 			goto exit_unlock_free;
2218 	}
2219 
2220 
2221 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2222 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2223 
2224 		err = ovs_vport_set_upcall_portids(vport, ids);
2225 		if (err)
2226 			goto exit_unlock_free;
2227 	}
2228 
2229 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2230 				      info->snd_portid, info->snd_seq, 0,
2231 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2232 	BUG_ON(err < 0);
2233 
2234 	ovs_unlock();
2235 	ovs_notify(&dp_vport_genl_family, reply, info);
2236 	return 0;
2237 
2238 exit_unlock_free:
2239 	ovs_unlock();
2240 	kfree_skb(reply);
2241 	return err;
2242 }
2243 
2244 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2245 {
2246 	bool update_headroom = false;
2247 	struct nlattr **a = info->attrs;
2248 	struct sk_buff *reply;
2249 	struct datapath *dp;
2250 	struct vport *vport;
2251 	unsigned int new_headroom;
2252 	int err;
2253 
2254 	reply = ovs_vport_cmd_alloc_info();
2255 	if (!reply)
2256 		return -ENOMEM;
2257 
2258 	ovs_lock();
2259 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2260 	err = PTR_ERR(vport);
2261 	if (IS_ERR(vport))
2262 		goto exit_unlock_free;
2263 
2264 	if (vport->port_no == OVSP_LOCAL) {
2265 		err = -EINVAL;
2266 		goto exit_unlock_free;
2267 	}
2268 
2269 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2270 				      info->snd_portid, info->snd_seq, 0,
2271 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2272 	BUG_ON(err < 0);
2273 
2274 	/* the vport deletion may trigger dp headroom update */
2275 	dp = vport->dp;
2276 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2277 		update_headroom = true;
2278 
2279 	netdev_reset_rx_headroom(vport->dev);
2280 	ovs_dp_detach_port(vport);
2281 
2282 	if (update_headroom) {
2283 		new_headroom = ovs_get_max_headroom(dp);
2284 
2285 		if (new_headroom < dp->max_headroom)
2286 			ovs_update_headroom(dp, new_headroom);
2287 	}
2288 	ovs_unlock();
2289 
2290 	ovs_notify(&dp_vport_genl_family, reply, info);
2291 	return 0;
2292 
2293 exit_unlock_free:
2294 	ovs_unlock();
2295 	kfree_skb(reply);
2296 	return err;
2297 }
2298 
2299 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2300 {
2301 	struct nlattr **a = info->attrs;
2302 	struct ovs_header *ovs_header = info->userhdr;
2303 	struct sk_buff *reply;
2304 	struct vport *vport;
2305 	int err;
2306 
2307 	reply = ovs_vport_cmd_alloc_info();
2308 	if (!reply)
2309 		return -ENOMEM;
2310 
2311 	rcu_read_lock();
2312 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2313 	err = PTR_ERR(vport);
2314 	if (IS_ERR(vport))
2315 		goto exit_unlock_free;
2316 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2317 				      info->snd_portid, info->snd_seq, 0,
2318 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2319 	BUG_ON(err < 0);
2320 	rcu_read_unlock();
2321 
2322 	return genlmsg_reply(reply, info);
2323 
2324 exit_unlock_free:
2325 	rcu_read_unlock();
2326 	kfree_skb(reply);
2327 	return err;
2328 }
2329 
2330 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2331 {
2332 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2333 	struct datapath *dp;
2334 	int bucket = cb->args[0], skip = cb->args[1];
2335 	int i, j = 0;
2336 
2337 	rcu_read_lock();
2338 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2339 	if (!dp) {
2340 		rcu_read_unlock();
2341 		return -ENODEV;
2342 	}
2343 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2344 		struct vport *vport;
2345 
2346 		j = 0;
2347 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2348 			if (j >= skip &&
2349 			    ovs_vport_cmd_fill_info(vport, skb,
2350 						    sock_net(skb->sk),
2351 						    NETLINK_CB(cb->skb).portid,
2352 						    cb->nlh->nlmsg_seq,
2353 						    NLM_F_MULTI,
2354 						    OVS_VPORT_CMD_GET,
2355 						    GFP_ATOMIC) < 0)
2356 				goto out;
2357 
2358 			j++;
2359 		}
2360 		skip = 0;
2361 	}
2362 out:
2363 	rcu_read_unlock();
2364 
2365 	cb->args[0] = i;
2366 	cb->args[1] = j;
2367 
2368 	return skb->len;
2369 }
2370 
2371 static void ovs_dp_masks_rebalance(struct work_struct *work)
2372 {
2373 	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2374 					       masks_rebalance.work);
2375 	struct datapath *dp;
2376 
2377 	ovs_lock();
2378 
2379 	list_for_each_entry(dp, &ovs_net->dps, list_node)
2380 		ovs_flow_masks_rebalance(&dp->table);
2381 
2382 	ovs_unlock();
2383 
2384 	schedule_delayed_work(&ovs_net->masks_rebalance,
2385 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2386 }
2387 
2388 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2389 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2390 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2391 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2392 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2393 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2394 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2395 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2396 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2397 };
2398 
2399 static const struct genl_ops dp_vport_genl_ops[] = {
2400 	{ .cmd = OVS_VPORT_CMD_NEW,
2401 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2402 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2403 	  .doit = ovs_vport_cmd_new
2404 	},
2405 	{ .cmd = OVS_VPORT_CMD_DEL,
2406 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2407 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2408 	  .doit = ovs_vport_cmd_del
2409 	},
2410 	{ .cmd = OVS_VPORT_CMD_GET,
2411 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2412 	  .flags = 0,		    /* OK for unprivileged users. */
2413 	  .doit = ovs_vport_cmd_get,
2414 	  .dumpit = ovs_vport_cmd_dump
2415 	},
2416 	{ .cmd = OVS_VPORT_CMD_SET,
2417 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2418 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2419 	  .doit = ovs_vport_cmd_set,
2420 	},
2421 };
2422 
2423 struct genl_family dp_vport_genl_family __ro_after_init = {
2424 	.hdrsize = sizeof(struct ovs_header),
2425 	.name = OVS_VPORT_FAMILY,
2426 	.version = OVS_VPORT_VERSION,
2427 	.maxattr = OVS_VPORT_ATTR_MAX,
2428 	.policy = vport_policy,
2429 	.netnsok = true,
2430 	.parallel_ops = true,
2431 	.ops = dp_vport_genl_ops,
2432 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2433 	.mcgrps = &ovs_dp_vport_multicast_group,
2434 	.n_mcgrps = 1,
2435 	.module = THIS_MODULE,
2436 };
2437 
2438 static struct genl_family * const dp_genl_families[] = {
2439 	&dp_datapath_genl_family,
2440 	&dp_vport_genl_family,
2441 	&dp_flow_genl_family,
2442 	&dp_packet_genl_family,
2443 	&dp_meter_genl_family,
2444 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2445 	&dp_ct_limit_genl_family,
2446 #endif
2447 };
2448 
2449 static void dp_unregister_genl(int n_families)
2450 {
2451 	int i;
2452 
2453 	for (i = 0; i < n_families; i++)
2454 		genl_unregister_family(dp_genl_families[i]);
2455 }
2456 
2457 static int __init dp_register_genl(void)
2458 {
2459 	int err;
2460 	int i;
2461 
2462 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2463 
2464 		err = genl_register_family(dp_genl_families[i]);
2465 		if (err)
2466 			goto error;
2467 	}
2468 
2469 	return 0;
2470 
2471 error:
2472 	dp_unregister_genl(i);
2473 	return err;
2474 }
2475 
2476 static int __net_init ovs_init_net(struct net *net)
2477 {
2478 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2479 
2480 	INIT_LIST_HEAD(&ovs_net->dps);
2481 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2482 	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2483 	schedule_delayed_work(&ovs_net->masks_rebalance,
2484 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2485 	return ovs_ct_init(net);
2486 }
2487 
2488 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2489 					    struct list_head *head)
2490 {
2491 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2492 	struct datapath *dp;
2493 
2494 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2495 		int i;
2496 
2497 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2498 			struct vport *vport;
2499 
2500 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2501 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2502 					continue;
2503 
2504 				if (dev_net(vport->dev) == dnet)
2505 					list_add(&vport->detach_list, head);
2506 			}
2507 		}
2508 	}
2509 }
2510 
2511 static void __net_exit ovs_exit_net(struct net *dnet)
2512 {
2513 	struct datapath *dp, *dp_next;
2514 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2515 	struct vport *vport, *vport_next;
2516 	struct net *net;
2517 	LIST_HEAD(head);
2518 
2519 	ovs_lock();
2520 
2521 	ovs_ct_exit(dnet);
2522 
2523 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2524 		__dp_destroy(dp);
2525 
2526 	down_read(&net_rwsem);
2527 	for_each_net(net)
2528 		list_vports_from_net(net, dnet, &head);
2529 	up_read(&net_rwsem);
2530 
2531 	/* Detach all vports from given namespace. */
2532 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2533 		list_del(&vport->detach_list);
2534 		ovs_dp_detach_port(vport);
2535 	}
2536 
2537 	ovs_unlock();
2538 
2539 	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2540 	cancel_work_sync(&ovs_net->dp_notify_work);
2541 }
2542 
2543 static struct pernet_operations ovs_net_ops = {
2544 	.init = ovs_init_net,
2545 	.exit = ovs_exit_net,
2546 	.id   = &ovs_net_id,
2547 	.size = sizeof(struct ovs_net),
2548 };
2549 
2550 static int __init dp_init(void)
2551 {
2552 	int err;
2553 
2554 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > sizeof_field(struct sk_buff, cb));
2555 
2556 	pr_info("Open vSwitch switching datapath\n");
2557 
2558 	err = action_fifos_init();
2559 	if (err)
2560 		goto error;
2561 
2562 	err = ovs_internal_dev_rtnl_link_register();
2563 	if (err)
2564 		goto error_action_fifos_exit;
2565 
2566 	err = ovs_flow_init();
2567 	if (err)
2568 		goto error_unreg_rtnl_link;
2569 
2570 	err = ovs_vport_init();
2571 	if (err)
2572 		goto error_flow_exit;
2573 
2574 	err = register_pernet_device(&ovs_net_ops);
2575 	if (err)
2576 		goto error_vport_exit;
2577 
2578 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2579 	if (err)
2580 		goto error_netns_exit;
2581 
2582 	err = ovs_netdev_init();
2583 	if (err)
2584 		goto error_unreg_notifier;
2585 
2586 	err = dp_register_genl();
2587 	if (err < 0)
2588 		goto error_unreg_netdev;
2589 
2590 	return 0;
2591 
2592 error_unreg_netdev:
2593 	ovs_netdev_exit();
2594 error_unreg_notifier:
2595 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2596 error_netns_exit:
2597 	unregister_pernet_device(&ovs_net_ops);
2598 error_vport_exit:
2599 	ovs_vport_exit();
2600 error_flow_exit:
2601 	ovs_flow_exit();
2602 error_unreg_rtnl_link:
2603 	ovs_internal_dev_rtnl_link_unregister();
2604 error_action_fifos_exit:
2605 	action_fifos_exit();
2606 error:
2607 	return err;
2608 }
2609 
2610 static void dp_cleanup(void)
2611 {
2612 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2613 	ovs_netdev_exit();
2614 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2615 	unregister_pernet_device(&ovs_net_ops);
2616 	rcu_barrier();
2617 	ovs_vport_exit();
2618 	ovs_flow_exit();
2619 	ovs_internal_dev_rtnl_link_unregister();
2620 	action_fifos_exit();
2621 }
2622 
2623 module_init(dp_init);
2624 module_exit(dp_cleanup);
2625 
2626 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2627 MODULE_LICENSE("GPL");
2628 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2629 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2630 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2631 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2632 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2633 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2634