xref: /linux/net/openvswitch/datapath.c (revision ca853314e78b0a65c20b6a889a23c31f918d4aa2)
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_small_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 	.small_ops = dp_packet_genl_ops,
672 	.n_small_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
1084 struct sw_flow_actions *get_flow_actions(struct net *net,
1085 					 const struct nlattr *a,
1086 					 const struct sw_flow_key *key,
1087 					 const struct sw_flow_mask *mask,
1088 					 bool log)
1089 {
1090 	struct sw_flow_actions *acts;
1091 	struct sw_flow_key masked_key;
1092 	int error;
1093 
1094 	ovs_flow_mask_key(&masked_key, key, true, mask);
1095 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1096 	if (error) {
1097 		OVS_NLERR(log,
1098 			  "Actions may not be safe on all matching packets");
1099 		return ERR_PTR(error);
1100 	}
1101 
1102 	return acts;
1103 }
1104 
1105 /* Factor out match-init and action-copy to avoid
1106  * "Wframe-larger-than=1024" warning. Because mask is only
1107  * used to get actions, we new a function to save some
1108  * stack space.
1109  *
1110  * If there are not key and action attrs, we return 0
1111  * directly. In the case, the caller will also not use the
1112  * match as before. If there is action attr, we try to get
1113  * actions and save them to *acts. Before returning from
1114  * the function, we reset the match->mask pointer. Because
1115  * we should not to return match object with dangling reference
1116  * to mask.
1117  * */
1118 static noinline_for_stack int
1119 ovs_nla_init_match_and_action(struct net *net,
1120 			      struct sw_flow_match *match,
1121 			      struct sw_flow_key *key,
1122 			      struct nlattr **a,
1123 			      struct sw_flow_actions **acts,
1124 			      bool log)
1125 {
1126 	struct sw_flow_mask mask;
1127 	int error = 0;
1128 
1129 	if (a[OVS_FLOW_ATTR_KEY]) {
1130 		ovs_match_init(match, key, true, &mask);
1131 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1132 					  a[OVS_FLOW_ATTR_MASK], log);
1133 		if (error)
1134 			goto error;
1135 	}
1136 
1137 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1138 		if (!a[OVS_FLOW_ATTR_KEY]) {
1139 			OVS_NLERR(log,
1140 				  "Flow key attribute not present in set flow.");
1141 			error = -EINVAL;
1142 			goto error;
1143 		}
1144 
1145 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1146 					 &mask, log);
1147 		if (IS_ERR(*acts)) {
1148 			error = PTR_ERR(*acts);
1149 			goto error;
1150 		}
1151 	}
1152 
1153 	/* On success, error is 0. */
1154 error:
1155 	match->mask = NULL;
1156 	return error;
1157 }
1158 
1159 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1160 {
1161 	struct net *net = sock_net(skb->sk);
1162 	struct nlattr **a = info->attrs;
1163 	struct ovs_header *ovs_header = info->userhdr;
1164 	struct sw_flow_key key;
1165 	struct sw_flow *flow;
1166 	struct sk_buff *reply = NULL;
1167 	struct datapath *dp;
1168 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1169 	struct sw_flow_match match;
1170 	struct sw_flow_id sfid;
1171 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1172 	int error = 0;
1173 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1174 	bool ufid_present;
1175 
1176 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1177 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1178 		OVS_NLERR(log,
1179 			  "Flow set message rejected, Key attribute missing.");
1180 		return -EINVAL;
1181 	}
1182 
1183 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1184 					      &acts, log);
1185 	if (error)
1186 		goto error;
1187 
1188 	if (acts) {
1189 		/* Can allocate before locking if have acts. */
1190 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1191 						ufid_flags);
1192 		if (IS_ERR(reply)) {
1193 			error = PTR_ERR(reply);
1194 			goto err_kfree_acts;
1195 		}
1196 	}
1197 
1198 	ovs_lock();
1199 	dp = get_dp(net, ovs_header->dp_ifindex);
1200 	if (unlikely(!dp)) {
1201 		error = -ENODEV;
1202 		goto err_unlock_ovs;
1203 	}
1204 	/* Check that the flow exists. */
1205 	if (ufid_present)
1206 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1207 	else
1208 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1209 	if (unlikely(!flow)) {
1210 		error = -ENOENT;
1211 		goto err_unlock_ovs;
1212 	}
1213 
1214 	/* Update actions, if present. */
1215 	if (likely(acts)) {
1216 		old_acts = ovsl_dereference(flow->sf_acts);
1217 		rcu_assign_pointer(flow->sf_acts, acts);
1218 
1219 		if (unlikely(reply)) {
1220 			error = ovs_flow_cmd_fill_info(flow,
1221 						       ovs_header->dp_ifindex,
1222 						       reply, info->snd_portid,
1223 						       info->snd_seq, 0,
1224 						       OVS_FLOW_CMD_SET,
1225 						       ufid_flags);
1226 			BUG_ON(error < 0);
1227 		}
1228 	} else {
1229 		/* Could not alloc without acts before locking. */
1230 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1231 						info, OVS_FLOW_CMD_SET, false,
1232 						ufid_flags);
1233 
1234 		if (IS_ERR(reply)) {
1235 			error = PTR_ERR(reply);
1236 			goto err_unlock_ovs;
1237 		}
1238 	}
1239 
1240 	/* Clear stats. */
1241 	if (a[OVS_FLOW_ATTR_CLEAR])
1242 		ovs_flow_stats_clear(flow);
1243 	ovs_unlock();
1244 
1245 	if (reply)
1246 		ovs_notify(&dp_flow_genl_family, reply, info);
1247 	if (old_acts)
1248 		ovs_nla_free_flow_actions_rcu(old_acts);
1249 
1250 	return 0;
1251 
1252 err_unlock_ovs:
1253 	ovs_unlock();
1254 	kfree_skb(reply);
1255 err_kfree_acts:
1256 	ovs_nla_free_flow_actions(acts);
1257 error:
1258 	return error;
1259 }
1260 
1261 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1262 {
1263 	struct nlattr **a = info->attrs;
1264 	struct ovs_header *ovs_header = info->userhdr;
1265 	struct net *net = sock_net(skb->sk);
1266 	struct sw_flow_key key;
1267 	struct sk_buff *reply;
1268 	struct sw_flow *flow;
1269 	struct datapath *dp;
1270 	struct sw_flow_match match;
1271 	struct sw_flow_id ufid;
1272 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1273 	int err = 0;
1274 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1275 	bool ufid_present;
1276 
1277 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1278 	if (a[OVS_FLOW_ATTR_KEY]) {
1279 		ovs_match_init(&match, &key, true, NULL);
1280 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1281 					log);
1282 	} else if (!ufid_present) {
1283 		OVS_NLERR(log,
1284 			  "Flow get message rejected, Key attribute missing.");
1285 		err = -EINVAL;
1286 	}
1287 	if (err)
1288 		return err;
1289 
1290 	ovs_lock();
1291 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1292 	if (!dp) {
1293 		err = -ENODEV;
1294 		goto unlock;
1295 	}
1296 
1297 	if (ufid_present)
1298 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1299 	else
1300 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1301 	if (!flow) {
1302 		err = -ENOENT;
1303 		goto unlock;
1304 	}
1305 
1306 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1307 					OVS_FLOW_CMD_GET, true, ufid_flags);
1308 	if (IS_ERR(reply)) {
1309 		err = PTR_ERR(reply);
1310 		goto unlock;
1311 	}
1312 
1313 	ovs_unlock();
1314 	return genlmsg_reply(reply, info);
1315 unlock:
1316 	ovs_unlock();
1317 	return err;
1318 }
1319 
1320 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1321 {
1322 	struct nlattr **a = info->attrs;
1323 	struct ovs_header *ovs_header = info->userhdr;
1324 	struct net *net = sock_net(skb->sk);
1325 	struct sw_flow_key key;
1326 	struct sk_buff *reply;
1327 	struct sw_flow *flow = NULL;
1328 	struct datapath *dp;
1329 	struct sw_flow_match match;
1330 	struct sw_flow_id ufid;
1331 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1332 	int err;
1333 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1334 	bool ufid_present;
1335 
1336 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1337 	if (a[OVS_FLOW_ATTR_KEY]) {
1338 		ovs_match_init(&match, &key, true, NULL);
1339 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1340 					NULL, log);
1341 		if (unlikely(err))
1342 			return err;
1343 	}
1344 
1345 	ovs_lock();
1346 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1347 	if (unlikely(!dp)) {
1348 		err = -ENODEV;
1349 		goto unlock;
1350 	}
1351 
1352 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1353 		err = ovs_flow_tbl_flush(&dp->table);
1354 		goto unlock;
1355 	}
1356 
1357 	if (ufid_present)
1358 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1359 	else
1360 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1361 	if (unlikely(!flow)) {
1362 		err = -ENOENT;
1363 		goto unlock;
1364 	}
1365 
1366 	ovs_flow_tbl_remove(&dp->table, flow);
1367 	ovs_unlock();
1368 
1369 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1370 					&flow->id, info, false, ufid_flags);
1371 	if (likely(reply)) {
1372 		if (!IS_ERR(reply)) {
1373 			rcu_read_lock();	/*To keep RCU checker happy. */
1374 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1375 						     reply, info->snd_portid,
1376 						     info->snd_seq, 0,
1377 						     OVS_FLOW_CMD_DEL,
1378 						     ufid_flags);
1379 			rcu_read_unlock();
1380 			if (WARN_ON_ONCE(err < 0)) {
1381 				kfree_skb(reply);
1382 				goto out_free;
1383 			}
1384 
1385 			ovs_notify(&dp_flow_genl_family, reply, info);
1386 		} else {
1387 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1388 					PTR_ERR(reply));
1389 		}
1390 	}
1391 
1392 out_free:
1393 	ovs_flow_free(flow, true);
1394 	return 0;
1395 unlock:
1396 	ovs_unlock();
1397 	return err;
1398 }
1399 
1400 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1401 {
1402 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1403 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1404 	struct table_instance *ti;
1405 	struct datapath *dp;
1406 	u32 ufid_flags;
1407 	int err;
1408 
1409 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1410 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1411 	if (err)
1412 		return err;
1413 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1414 
1415 	rcu_read_lock();
1416 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1417 	if (!dp) {
1418 		rcu_read_unlock();
1419 		return -ENODEV;
1420 	}
1421 
1422 	ti = rcu_dereference(dp->table.ti);
1423 	for (;;) {
1424 		struct sw_flow *flow;
1425 		u32 bucket, obj;
1426 
1427 		bucket = cb->args[0];
1428 		obj = cb->args[1];
1429 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1430 		if (!flow)
1431 			break;
1432 
1433 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1434 					   NETLINK_CB(cb->skb).portid,
1435 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1436 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1437 			break;
1438 
1439 		cb->args[0] = bucket;
1440 		cb->args[1] = obj;
1441 	}
1442 	rcu_read_unlock();
1443 	return skb->len;
1444 }
1445 
1446 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1447 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1448 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1449 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1450 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1451 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1452 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1453 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1454 };
1455 
1456 static const struct genl_small_ops dp_flow_genl_ops[] = {
1457 	{ .cmd = OVS_FLOW_CMD_NEW,
1458 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1459 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1460 	  .doit = ovs_flow_cmd_new
1461 	},
1462 	{ .cmd = OVS_FLOW_CMD_DEL,
1463 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1464 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1465 	  .doit = ovs_flow_cmd_del
1466 	},
1467 	{ .cmd = OVS_FLOW_CMD_GET,
1468 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1469 	  .flags = 0,		    /* OK for unprivileged users. */
1470 	  .doit = ovs_flow_cmd_get,
1471 	  .dumpit = ovs_flow_cmd_dump
1472 	},
1473 	{ .cmd = OVS_FLOW_CMD_SET,
1474 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1475 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1476 	  .doit = ovs_flow_cmd_set,
1477 	},
1478 };
1479 
1480 static struct genl_family dp_flow_genl_family __ro_after_init = {
1481 	.hdrsize = sizeof(struct ovs_header),
1482 	.name = OVS_FLOW_FAMILY,
1483 	.version = OVS_FLOW_VERSION,
1484 	.maxattr = OVS_FLOW_ATTR_MAX,
1485 	.policy = flow_policy,
1486 	.netnsok = true,
1487 	.parallel_ops = true,
1488 	.small_ops = dp_flow_genl_ops,
1489 	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1490 	.mcgrps = &ovs_dp_flow_multicast_group,
1491 	.n_mcgrps = 1,
1492 	.module = THIS_MODULE,
1493 };
1494 
1495 static size_t ovs_dp_cmd_msg_size(void)
1496 {
1497 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1498 
1499 	msgsize += nla_total_size(IFNAMSIZ);
1500 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1501 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1502 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1503 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1504 
1505 	return msgsize;
1506 }
1507 
1508 /* Called with ovs_mutex. */
1509 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1510 				u32 portid, u32 seq, u32 flags, u8 cmd)
1511 {
1512 	struct ovs_header *ovs_header;
1513 	struct ovs_dp_stats dp_stats;
1514 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1515 	int err;
1516 
1517 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1518 				 flags, cmd);
1519 	if (!ovs_header)
1520 		goto error;
1521 
1522 	ovs_header->dp_ifindex = get_dpifindex(dp);
1523 
1524 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1525 	if (err)
1526 		goto nla_put_failure;
1527 
1528 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1529 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1530 			  &dp_stats, OVS_DP_ATTR_PAD))
1531 		goto nla_put_failure;
1532 
1533 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1534 			  sizeof(struct ovs_dp_megaflow_stats),
1535 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1536 		goto nla_put_failure;
1537 
1538 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1539 		goto nla_put_failure;
1540 
1541 	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1542 			ovs_flow_tbl_masks_cache_size(&dp->table)))
1543 		goto nla_put_failure;
1544 
1545 	genlmsg_end(skb, ovs_header);
1546 	return 0;
1547 
1548 nla_put_failure:
1549 	genlmsg_cancel(skb, ovs_header);
1550 error:
1551 	return -EMSGSIZE;
1552 }
1553 
1554 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1555 {
1556 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1557 }
1558 
1559 /* Called with rcu_read_lock or ovs_mutex. */
1560 static struct datapath *lookup_datapath(struct net *net,
1561 					const struct ovs_header *ovs_header,
1562 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1563 {
1564 	struct datapath *dp;
1565 
1566 	if (!a[OVS_DP_ATTR_NAME])
1567 		dp = get_dp(net, ovs_header->dp_ifindex);
1568 	else {
1569 		struct vport *vport;
1570 
1571 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1572 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1573 	}
1574 	return dp ? dp : ERR_PTR(-ENODEV);
1575 }
1576 
1577 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1578 				       struct genl_info *info)
1579 {
1580 	struct datapath *dp;
1581 
1582 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1583 			     info->attrs);
1584 	if (IS_ERR(dp))
1585 		return;
1586 
1587 	WARN(dp->user_features, "Dropping previously announced user features\n");
1588 	dp->user_features = 0;
1589 }
1590 
1591 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1592 
1593 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1594 {
1595 	u32 user_features = 0;
1596 
1597 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1598 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1599 
1600 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1601 				      OVS_DP_F_UNALIGNED |
1602 				      OVS_DP_F_TC_RECIRC_SHARING))
1603 			return -EOPNOTSUPP;
1604 
1605 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1606 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1607 			return -EOPNOTSUPP;
1608 #endif
1609 	}
1610 
1611 	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1612 		int err;
1613 		u32 cache_size;
1614 
1615 		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1616 		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1617 		if (err)
1618 			return err;
1619 	}
1620 
1621 	dp->user_features = user_features;
1622 
1623 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1624 		static_branch_enable(&tc_recirc_sharing_support);
1625 	else
1626 		static_branch_disable(&tc_recirc_sharing_support);
1627 
1628 	return 0;
1629 }
1630 
1631 static int ovs_dp_stats_init(struct datapath *dp)
1632 {
1633 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1634 	if (!dp->stats_percpu)
1635 		return -ENOMEM;
1636 
1637 	return 0;
1638 }
1639 
1640 static int ovs_dp_vport_init(struct datapath *dp)
1641 {
1642 	int i;
1643 
1644 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1645 				  sizeof(struct hlist_head),
1646 				  GFP_KERNEL);
1647 	if (!dp->ports)
1648 		return -ENOMEM;
1649 
1650 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1651 		INIT_HLIST_HEAD(&dp->ports[i]);
1652 
1653 	return 0;
1654 }
1655 
1656 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1657 {
1658 	struct nlattr **a = info->attrs;
1659 	struct vport_parms parms;
1660 	struct sk_buff *reply;
1661 	struct datapath *dp;
1662 	struct vport *vport;
1663 	struct ovs_net *ovs_net;
1664 	int err;
1665 
1666 	err = -EINVAL;
1667 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1668 		goto err;
1669 
1670 	reply = ovs_dp_cmd_alloc_info();
1671 	if (!reply)
1672 		return -ENOMEM;
1673 
1674 	err = -ENOMEM;
1675 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1676 	if (dp == NULL)
1677 		goto err_destroy_reply;
1678 
1679 	ovs_dp_set_net(dp, sock_net(skb->sk));
1680 
1681 	/* Allocate table. */
1682 	err = ovs_flow_tbl_init(&dp->table);
1683 	if (err)
1684 		goto err_destroy_dp;
1685 
1686 	err = ovs_dp_stats_init(dp);
1687 	if (err)
1688 		goto err_destroy_table;
1689 
1690 	err = ovs_dp_vport_init(dp);
1691 	if (err)
1692 		goto err_destroy_stats;
1693 
1694 	err = ovs_meters_init(dp);
1695 	if (err)
1696 		goto err_destroy_ports;
1697 
1698 	/* Set up our datapath device. */
1699 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1700 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1701 	parms.options = NULL;
1702 	parms.dp = dp;
1703 	parms.port_no = OVSP_LOCAL;
1704 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1705 
1706 	/* So far only local changes have been made, now need the lock. */
1707 	ovs_lock();
1708 
1709 	err = ovs_dp_change(dp, a);
1710 	if (err)
1711 		goto err_unlock_and_destroy_meters;
1712 
1713 	vport = new_vport(&parms);
1714 	if (IS_ERR(vport)) {
1715 		err = PTR_ERR(vport);
1716 		if (err == -EBUSY)
1717 			err = -EEXIST;
1718 
1719 		if (err == -EEXIST) {
1720 			/* An outdated user space instance that does not understand
1721 			 * the concept of user_features has attempted to create a new
1722 			 * datapath and is likely to reuse it. Drop all user features.
1723 			 */
1724 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1725 				ovs_dp_reset_user_features(skb, info);
1726 		}
1727 
1728 		goto err_unlock_and_destroy_meters;
1729 	}
1730 
1731 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1732 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1733 	BUG_ON(err < 0);
1734 
1735 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1736 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1737 
1738 	ovs_unlock();
1739 
1740 	ovs_notify(&dp_datapath_genl_family, reply, info);
1741 	return 0;
1742 
1743 err_unlock_and_destroy_meters:
1744 	ovs_unlock();
1745 	ovs_meters_exit(dp);
1746 err_destroy_ports:
1747 	kfree(dp->ports);
1748 err_destroy_stats:
1749 	free_percpu(dp->stats_percpu);
1750 err_destroy_table:
1751 	ovs_flow_tbl_destroy(&dp->table);
1752 err_destroy_dp:
1753 	kfree(dp);
1754 err_destroy_reply:
1755 	kfree_skb(reply);
1756 err:
1757 	return err;
1758 }
1759 
1760 /* Called with ovs_mutex. */
1761 static void __dp_destroy(struct datapath *dp)
1762 {
1763 	struct flow_table *table = &dp->table;
1764 	int i;
1765 
1766 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1767 		struct vport *vport;
1768 		struct hlist_node *n;
1769 
1770 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1771 			if (vport->port_no != OVSP_LOCAL)
1772 				ovs_dp_detach_port(vport);
1773 	}
1774 
1775 	list_del_rcu(&dp->list_node);
1776 
1777 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1778 	 * all ports in datapath are destroyed first before freeing datapath.
1779 	 */
1780 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1781 
1782 	/* Flush sw_flow in the tables. RCU cb only releases resource
1783 	 * such as dp, ports and tables. That may avoid some issues
1784 	 * such as RCU usage warning.
1785 	 */
1786 	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1787 				  ovsl_dereference(table->ufid_ti));
1788 
1789 	/* RCU destroy the ports, meters and flow tables. */
1790 	call_rcu(&dp->rcu, destroy_dp_rcu);
1791 }
1792 
1793 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1794 {
1795 	struct sk_buff *reply;
1796 	struct datapath *dp;
1797 	int err;
1798 
1799 	reply = ovs_dp_cmd_alloc_info();
1800 	if (!reply)
1801 		return -ENOMEM;
1802 
1803 	ovs_lock();
1804 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1805 	err = PTR_ERR(dp);
1806 	if (IS_ERR(dp))
1807 		goto err_unlock_free;
1808 
1809 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1810 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1811 	BUG_ON(err < 0);
1812 
1813 	__dp_destroy(dp);
1814 	ovs_unlock();
1815 
1816 	ovs_notify(&dp_datapath_genl_family, reply, info);
1817 
1818 	return 0;
1819 
1820 err_unlock_free:
1821 	ovs_unlock();
1822 	kfree_skb(reply);
1823 	return err;
1824 }
1825 
1826 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1827 {
1828 	struct sk_buff *reply;
1829 	struct datapath *dp;
1830 	int err;
1831 
1832 	reply = ovs_dp_cmd_alloc_info();
1833 	if (!reply)
1834 		return -ENOMEM;
1835 
1836 	ovs_lock();
1837 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1838 	err = PTR_ERR(dp);
1839 	if (IS_ERR(dp))
1840 		goto err_unlock_free;
1841 
1842 	err = ovs_dp_change(dp, info->attrs);
1843 	if (err)
1844 		goto err_unlock_free;
1845 
1846 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1847 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1848 	BUG_ON(err < 0);
1849 
1850 	ovs_unlock();
1851 	ovs_notify(&dp_datapath_genl_family, reply, info);
1852 
1853 	return 0;
1854 
1855 err_unlock_free:
1856 	ovs_unlock();
1857 	kfree_skb(reply);
1858 	return err;
1859 }
1860 
1861 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1862 {
1863 	struct sk_buff *reply;
1864 	struct datapath *dp;
1865 	int err;
1866 
1867 	reply = ovs_dp_cmd_alloc_info();
1868 	if (!reply)
1869 		return -ENOMEM;
1870 
1871 	ovs_lock();
1872 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1873 	if (IS_ERR(dp)) {
1874 		err = PTR_ERR(dp);
1875 		goto err_unlock_free;
1876 	}
1877 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1878 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1879 	BUG_ON(err < 0);
1880 	ovs_unlock();
1881 
1882 	return genlmsg_reply(reply, info);
1883 
1884 err_unlock_free:
1885 	ovs_unlock();
1886 	kfree_skb(reply);
1887 	return err;
1888 }
1889 
1890 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1891 {
1892 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1893 	struct datapath *dp;
1894 	int skip = cb->args[0];
1895 	int i = 0;
1896 
1897 	ovs_lock();
1898 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1899 		if (i >= skip &&
1900 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1901 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1902 					 OVS_DP_CMD_GET) < 0)
1903 			break;
1904 		i++;
1905 	}
1906 	ovs_unlock();
1907 
1908 	cb->args[0] = i;
1909 
1910 	return skb->len;
1911 }
1912 
1913 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1914 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1915 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1916 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1917 	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
1918 		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
1919 };
1920 
1921 static const struct genl_small_ops dp_datapath_genl_ops[] = {
1922 	{ .cmd = OVS_DP_CMD_NEW,
1923 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1924 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1925 	  .doit = ovs_dp_cmd_new
1926 	},
1927 	{ .cmd = OVS_DP_CMD_DEL,
1928 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1929 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1930 	  .doit = ovs_dp_cmd_del
1931 	},
1932 	{ .cmd = OVS_DP_CMD_GET,
1933 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1934 	  .flags = 0,		    /* OK for unprivileged users. */
1935 	  .doit = ovs_dp_cmd_get,
1936 	  .dumpit = ovs_dp_cmd_dump
1937 	},
1938 	{ .cmd = OVS_DP_CMD_SET,
1939 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1940 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1941 	  .doit = ovs_dp_cmd_set,
1942 	},
1943 };
1944 
1945 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1946 	.hdrsize = sizeof(struct ovs_header),
1947 	.name = OVS_DATAPATH_FAMILY,
1948 	.version = OVS_DATAPATH_VERSION,
1949 	.maxattr = OVS_DP_ATTR_MAX,
1950 	.policy = datapath_policy,
1951 	.netnsok = true,
1952 	.parallel_ops = true,
1953 	.small_ops = dp_datapath_genl_ops,
1954 	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1955 	.mcgrps = &ovs_dp_datapath_multicast_group,
1956 	.n_mcgrps = 1,
1957 	.module = THIS_MODULE,
1958 };
1959 
1960 /* Called with ovs_mutex or RCU read lock. */
1961 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1962 				   struct net *net, u32 portid, u32 seq,
1963 				   u32 flags, u8 cmd, gfp_t gfp)
1964 {
1965 	struct ovs_header *ovs_header;
1966 	struct ovs_vport_stats vport_stats;
1967 	int err;
1968 
1969 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1970 				 flags, cmd);
1971 	if (!ovs_header)
1972 		return -EMSGSIZE;
1973 
1974 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1975 
1976 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1977 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1978 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1979 			   ovs_vport_name(vport)) ||
1980 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
1981 		goto nla_put_failure;
1982 
1983 	if (!net_eq(net, dev_net(vport->dev))) {
1984 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
1985 
1986 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
1987 			goto nla_put_failure;
1988 	}
1989 
1990 	ovs_vport_get_stats(vport, &vport_stats);
1991 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1992 			  sizeof(struct ovs_vport_stats), &vport_stats,
1993 			  OVS_VPORT_ATTR_PAD))
1994 		goto nla_put_failure;
1995 
1996 	if (ovs_vport_get_upcall_portids(vport, skb))
1997 		goto nla_put_failure;
1998 
1999 	err = ovs_vport_get_options(vport, skb);
2000 	if (err == -EMSGSIZE)
2001 		goto error;
2002 
2003 	genlmsg_end(skb, ovs_header);
2004 	return 0;
2005 
2006 nla_put_failure:
2007 	err = -EMSGSIZE;
2008 error:
2009 	genlmsg_cancel(skb, ovs_header);
2010 	return err;
2011 }
2012 
2013 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2014 {
2015 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2016 }
2017 
2018 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2019 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2020 					 u32 portid, u32 seq, u8 cmd)
2021 {
2022 	struct sk_buff *skb;
2023 	int retval;
2024 
2025 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2026 	if (!skb)
2027 		return ERR_PTR(-ENOMEM);
2028 
2029 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2030 					 GFP_KERNEL);
2031 	BUG_ON(retval < 0);
2032 
2033 	return skb;
2034 }
2035 
2036 /* Called with ovs_mutex or RCU read lock. */
2037 static struct vport *lookup_vport(struct net *net,
2038 				  const struct ovs_header *ovs_header,
2039 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2040 {
2041 	struct datapath *dp;
2042 	struct vport *vport;
2043 
2044 	if (a[OVS_VPORT_ATTR_IFINDEX])
2045 		return ERR_PTR(-EOPNOTSUPP);
2046 	if (a[OVS_VPORT_ATTR_NAME]) {
2047 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2048 		if (!vport)
2049 			return ERR_PTR(-ENODEV);
2050 		if (ovs_header->dp_ifindex &&
2051 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2052 			return ERR_PTR(-ENODEV);
2053 		return vport;
2054 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2055 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2056 
2057 		if (port_no >= DP_MAX_PORTS)
2058 			return ERR_PTR(-EFBIG);
2059 
2060 		dp = get_dp(net, ovs_header->dp_ifindex);
2061 		if (!dp)
2062 			return ERR_PTR(-ENODEV);
2063 
2064 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2065 		if (!vport)
2066 			return ERR_PTR(-ENODEV);
2067 		return vport;
2068 	} else
2069 		return ERR_PTR(-EINVAL);
2070 
2071 }
2072 
2073 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2074 {
2075 	unsigned int dev_headroom, max_headroom = 0;
2076 	struct net_device *dev;
2077 	struct vport *vport;
2078 	int i;
2079 
2080 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2081 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2082 					 lockdep_ovsl_is_held()) {
2083 			dev = vport->dev;
2084 			dev_headroom = netdev_get_fwd_headroom(dev);
2085 			if (dev_headroom > max_headroom)
2086 				max_headroom = dev_headroom;
2087 		}
2088 	}
2089 
2090 	return max_headroom;
2091 }
2092 
2093 /* Called with ovs_mutex */
2094 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2095 {
2096 	struct vport *vport;
2097 	int i;
2098 
2099 	dp->max_headroom = new_headroom;
2100 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2101 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2102 					 lockdep_ovsl_is_held())
2103 			netdev_set_rx_headroom(vport->dev, new_headroom);
2104 	}
2105 }
2106 
2107 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2108 {
2109 	struct nlattr **a = info->attrs;
2110 	struct ovs_header *ovs_header = info->userhdr;
2111 	struct vport_parms parms;
2112 	struct sk_buff *reply;
2113 	struct vport *vport;
2114 	struct datapath *dp;
2115 	unsigned int new_headroom;
2116 	u32 port_no;
2117 	int err;
2118 
2119 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2120 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2121 		return -EINVAL;
2122 	if (a[OVS_VPORT_ATTR_IFINDEX])
2123 		return -EOPNOTSUPP;
2124 
2125 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2126 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2127 	if (port_no >= DP_MAX_PORTS)
2128 		return -EFBIG;
2129 
2130 	reply = ovs_vport_cmd_alloc_info();
2131 	if (!reply)
2132 		return -ENOMEM;
2133 
2134 	ovs_lock();
2135 restart:
2136 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2137 	err = -ENODEV;
2138 	if (!dp)
2139 		goto exit_unlock_free;
2140 
2141 	if (port_no) {
2142 		vport = ovs_vport_ovsl(dp, port_no);
2143 		err = -EBUSY;
2144 		if (vport)
2145 			goto exit_unlock_free;
2146 	} else {
2147 		for (port_no = 1; ; port_no++) {
2148 			if (port_no >= DP_MAX_PORTS) {
2149 				err = -EFBIG;
2150 				goto exit_unlock_free;
2151 			}
2152 			vport = ovs_vport_ovsl(dp, port_no);
2153 			if (!vport)
2154 				break;
2155 		}
2156 	}
2157 
2158 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2159 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2160 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2161 	parms.dp = dp;
2162 	parms.port_no = port_no;
2163 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2164 
2165 	vport = new_vport(&parms);
2166 	err = PTR_ERR(vport);
2167 	if (IS_ERR(vport)) {
2168 		if (err == -EAGAIN)
2169 			goto restart;
2170 		goto exit_unlock_free;
2171 	}
2172 
2173 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2174 				      info->snd_portid, info->snd_seq, 0,
2175 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2176 
2177 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2178 
2179 	if (new_headroom > dp->max_headroom)
2180 		ovs_update_headroom(dp, new_headroom);
2181 	else
2182 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2183 
2184 	BUG_ON(err < 0);
2185 	ovs_unlock();
2186 
2187 	ovs_notify(&dp_vport_genl_family, reply, info);
2188 	return 0;
2189 
2190 exit_unlock_free:
2191 	ovs_unlock();
2192 	kfree_skb(reply);
2193 	return err;
2194 }
2195 
2196 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2197 {
2198 	struct nlattr **a = info->attrs;
2199 	struct sk_buff *reply;
2200 	struct vport *vport;
2201 	int err;
2202 
2203 	reply = ovs_vport_cmd_alloc_info();
2204 	if (!reply)
2205 		return -ENOMEM;
2206 
2207 	ovs_lock();
2208 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2209 	err = PTR_ERR(vport);
2210 	if (IS_ERR(vport))
2211 		goto exit_unlock_free;
2212 
2213 	if (a[OVS_VPORT_ATTR_TYPE] &&
2214 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2215 		err = -EINVAL;
2216 		goto exit_unlock_free;
2217 	}
2218 
2219 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2220 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2221 		if (err)
2222 			goto exit_unlock_free;
2223 	}
2224 
2225 
2226 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2227 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2228 
2229 		err = ovs_vport_set_upcall_portids(vport, ids);
2230 		if (err)
2231 			goto exit_unlock_free;
2232 	}
2233 
2234 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2235 				      info->snd_portid, info->snd_seq, 0,
2236 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2237 	BUG_ON(err < 0);
2238 
2239 	ovs_unlock();
2240 	ovs_notify(&dp_vport_genl_family, reply, info);
2241 	return 0;
2242 
2243 exit_unlock_free:
2244 	ovs_unlock();
2245 	kfree_skb(reply);
2246 	return err;
2247 }
2248 
2249 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2250 {
2251 	bool update_headroom = false;
2252 	struct nlattr **a = info->attrs;
2253 	struct sk_buff *reply;
2254 	struct datapath *dp;
2255 	struct vport *vport;
2256 	unsigned int new_headroom;
2257 	int err;
2258 
2259 	reply = ovs_vport_cmd_alloc_info();
2260 	if (!reply)
2261 		return -ENOMEM;
2262 
2263 	ovs_lock();
2264 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2265 	err = PTR_ERR(vport);
2266 	if (IS_ERR(vport))
2267 		goto exit_unlock_free;
2268 
2269 	if (vport->port_no == OVSP_LOCAL) {
2270 		err = -EINVAL;
2271 		goto exit_unlock_free;
2272 	}
2273 
2274 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2275 				      info->snd_portid, info->snd_seq, 0,
2276 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2277 	BUG_ON(err < 0);
2278 
2279 	/* the vport deletion may trigger dp headroom update */
2280 	dp = vport->dp;
2281 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2282 		update_headroom = true;
2283 
2284 	netdev_reset_rx_headroom(vport->dev);
2285 	ovs_dp_detach_port(vport);
2286 
2287 	if (update_headroom) {
2288 		new_headroom = ovs_get_max_headroom(dp);
2289 
2290 		if (new_headroom < dp->max_headroom)
2291 			ovs_update_headroom(dp, new_headroom);
2292 	}
2293 	ovs_unlock();
2294 
2295 	ovs_notify(&dp_vport_genl_family, reply, info);
2296 	return 0;
2297 
2298 exit_unlock_free:
2299 	ovs_unlock();
2300 	kfree_skb(reply);
2301 	return err;
2302 }
2303 
2304 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2305 {
2306 	struct nlattr **a = info->attrs;
2307 	struct ovs_header *ovs_header = info->userhdr;
2308 	struct sk_buff *reply;
2309 	struct vport *vport;
2310 	int err;
2311 
2312 	reply = ovs_vport_cmd_alloc_info();
2313 	if (!reply)
2314 		return -ENOMEM;
2315 
2316 	rcu_read_lock();
2317 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2318 	err = PTR_ERR(vport);
2319 	if (IS_ERR(vport))
2320 		goto exit_unlock_free;
2321 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2322 				      info->snd_portid, info->snd_seq, 0,
2323 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2324 	BUG_ON(err < 0);
2325 	rcu_read_unlock();
2326 
2327 	return genlmsg_reply(reply, info);
2328 
2329 exit_unlock_free:
2330 	rcu_read_unlock();
2331 	kfree_skb(reply);
2332 	return err;
2333 }
2334 
2335 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2336 {
2337 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2338 	struct datapath *dp;
2339 	int bucket = cb->args[0], skip = cb->args[1];
2340 	int i, j = 0;
2341 
2342 	rcu_read_lock();
2343 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2344 	if (!dp) {
2345 		rcu_read_unlock();
2346 		return -ENODEV;
2347 	}
2348 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2349 		struct vport *vport;
2350 
2351 		j = 0;
2352 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2353 			if (j >= skip &&
2354 			    ovs_vport_cmd_fill_info(vport, skb,
2355 						    sock_net(skb->sk),
2356 						    NETLINK_CB(cb->skb).portid,
2357 						    cb->nlh->nlmsg_seq,
2358 						    NLM_F_MULTI,
2359 						    OVS_VPORT_CMD_GET,
2360 						    GFP_ATOMIC) < 0)
2361 				goto out;
2362 
2363 			j++;
2364 		}
2365 		skip = 0;
2366 	}
2367 out:
2368 	rcu_read_unlock();
2369 
2370 	cb->args[0] = i;
2371 	cb->args[1] = j;
2372 
2373 	return skb->len;
2374 }
2375 
2376 static void ovs_dp_masks_rebalance(struct work_struct *work)
2377 {
2378 	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2379 					       masks_rebalance.work);
2380 	struct datapath *dp;
2381 
2382 	ovs_lock();
2383 
2384 	list_for_each_entry(dp, &ovs_net->dps, list_node)
2385 		ovs_flow_masks_rebalance(&dp->table);
2386 
2387 	ovs_unlock();
2388 
2389 	schedule_delayed_work(&ovs_net->masks_rebalance,
2390 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2391 }
2392 
2393 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2394 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2395 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2396 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2397 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2398 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2399 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2400 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2401 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2402 };
2403 
2404 static const struct genl_small_ops dp_vport_genl_ops[] = {
2405 	{ .cmd = OVS_VPORT_CMD_NEW,
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_new
2409 	},
2410 	{ .cmd = OVS_VPORT_CMD_DEL,
2411 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2412 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2413 	  .doit = ovs_vport_cmd_del
2414 	},
2415 	{ .cmd = OVS_VPORT_CMD_GET,
2416 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2417 	  .flags = 0,		    /* OK for unprivileged users. */
2418 	  .doit = ovs_vport_cmd_get,
2419 	  .dumpit = ovs_vport_cmd_dump
2420 	},
2421 	{ .cmd = OVS_VPORT_CMD_SET,
2422 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2423 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2424 	  .doit = ovs_vport_cmd_set,
2425 	},
2426 };
2427 
2428 struct genl_family dp_vport_genl_family __ro_after_init = {
2429 	.hdrsize = sizeof(struct ovs_header),
2430 	.name = OVS_VPORT_FAMILY,
2431 	.version = OVS_VPORT_VERSION,
2432 	.maxattr = OVS_VPORT_ATTR_MAX,
2433 	.policy = vport_policy,
2434 	.netnsok = true,
2435 	.parallel_ops = true,
2436 	.small_ops = dp_vport_genl_ops,
2437 	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2438 	.mcgrps = &ovs_dp_vport_multicast_group,
2439 	.n_mcgrps = 1,
2440 	.module = THIS_MODULE,
2441 };
2442 
2443 static struct genl_family * const dp_genl_families[] = {
2444 	&dp_datapath_genl_family,
2445 	&dp_vport_genl_family,
2446 	&dp_flow_genl_family,
2447 	&dp_packet_genl_family,
2448 	&dp_meter_genl_family,
2449 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2450 	&dp_ct_limit_genl_family,
2451 #endif
2452 };
2453 
2454 static void dp_unregister_genl(int n_families)
2455 {
2456 	int i;
2457 
2458 	for (i = 0; i < n_families; i++)
2459 		genl_unregister_family(dp_genl_families[i]);
2460 }
2461 
2462 static int __init dp_register_genl(void)
2463 {
2464 	int err;
2465 	int i;
2466 
2467 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2468 
2469 		err = genl_register_family(dp_genl_families[i]);
2470 		if (err)
2471 			goto error;
2472 	}
2473 
2474 	return 0;
2475 
2476 error:
2477 	dp_unregister_genl(i);
2478 	return err;
2479 }
2480 
2481 static int __net_init ovs_init_net(struct net *net)
2482 {
2483 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2484 	int err;
2485 
2486 	INIT_LIST_HEAD(&ovs_net->dps);
2487 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2488 	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2489 
2490 	err = ovs_ct_init(net);
2491 	if (err)
2492 		return err;
2493 
2494 	schedule_delayed_work(&ovs_net->masks_rebalance,
2495 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2496 	return 0;
2497 }
2498 
2499 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2500 					    struct list_head *head)
2501 {
2502 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2503 	struct datapath *dp;
2504 
2505 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2506 		int i;
2507 
2508 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2509 			struct vport *vport;
2510 
2511 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2512 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2513 					continue;
2514 
2515 				if (dev_net(vport->dev) == dnet)
2516 					list_add(&vport->detach_list, head);
2517 			}
2518 		}
2519 	}
2520 }
2521 
2522 static void __net_exit ovs_exit_net(struct net *dnet)
2523 {
2524 	struct datapath *dp, *dp_next;
2525 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2526 	struct vport *vport, *vport_next;
2527 	struct net *net;
2528 	LIST_HEAD(head);
2529 
2530 	ovs_lock();
2531 
2532 	ovs_ct_exit(dnet);
2533 
2534 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2535 		__dp_destroy(dp);
2536 
2537 	down_read(&net_rwsem);
2538 	for_each_net(net)
2539 		list_vports_from_net(net, dnet, &head);
2540 	up_read(&net_rwsem);
2541 
2542 	/* Detach all vports from given namespace. */
2543 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2544 		list_del(&vport->detach_list);
2545 		ovs_dp_detach_port(vport);
2546 	}
2547 
2548 	ovs_unlock();
2549 
2550 	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2551 	cancel_work_sync(&ovs_net->dp_notify_work);
2552 }
2553 
2554 static struct pernet_operations ovs_net_ops = {
2555 	.init = ovs_init_net,
2556 	.exit = ovs_exit_net,
2557 	.id   = &ovs_net_id,
2558 	.size = sizeof(struct ovs_net),
2559 };
2560 
2561 static int __init dp_init(void)
2562 {
2563 	int err;
2564 
2565 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2566 		     sizeof_field(struct sk_buff, cb));
2567 
2568 	pr_info("Open vSwitch switching datapath\n");
2569 
2570 	err = action_fifos_init();
2571 	if (err)
2572 		goto error;
2573 
2574 	err = ovs_internal_dev_rtnl_link_register();
2575 	if (err)
2576 		goto error_action_fifos_exit;
2577 
2578 	err = ovs_flow_init();
2579 	if (err)
2580 		goto error_unreg_rtnl_link;
2581 
2582 	err = ovs_vport_init();
2583 	if (err)
2584 		goto error_flow_exit;
2585 
2586 	err = register_pernet_device(&ovs_net_ops);
2587 	if (err)
2588 		goto error_vport_exit;
2589 
2590 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2591 	if (err)
2592 		goto error_netns_exit;
2593 
2594 	err = ovs_netdev_init();
2595 	if (err)
2596 		goto error_unreg_notifier;
2597 
2598 	err = dp_register_genl();
2599 	if (err < 0)
2600 		goto error_unreg_netdev;
2601 
2602 	return 0;
2603 
2604 error_unreg_netdev:
2605 	ovs_netdev_exit();
2606 error_unreg_notifier:
2607 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2608 error_netns_exit:
2609 	unregister_pernet_device(&ovs_net_ops);
2610 error_vport_exit:
2611 	ovs_vport_exit();
2612 error_flow_exit:
2613 	ovs_flow_exit();
2614 error_unreg_rtnl_link:
2615 	ovs_internal_dev_rtnl_link_unregister();
2616 error_action_fifos_exit:
2617 	action_fifos_exit();
2618 error:
2619 	return err;
2620 }
2621 
2622 static void dp_cleanup(void)
2623 {
2624 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2625 	ovs_netdev_exit();
2626 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2627 	unregister_pernet_device(&ovs_net_ops);
2628 	rcu_barrier();
2629 	ovs_vport_exit();
2630 	ovs_flow_exit();
2631 	ovs_internal_dev_rtnl_link_unregister();
2632 	action_fifos_exit();
2633 }
2634 
2635 module_init(dp_init);
2636 module_exit(dp_cleanup);
2637 
2638 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2639 MODULE_LICENSE("GPL");
2640 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2641 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2642 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2643 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2644 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2645 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2646