xref: /openbmc/linux/net/openvswitch/datapath.c (revision 97e6f135)
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/gso.h>
39 #include <net/net_namespace.h>
40 #include <net/netns/generic.h>
41 #include <net/pkt_cls.h>
42 
43 #include "datapath.h"
44 #include "flow.h"
45 #include "flow_table.h"
46 #include "flow_netlink.h"
47 #include "meter.h"
48 #include "openvswitch_trace.h"
49 #include "vport-internal_dev.h"
50 #include "vport-netdev.h"
51 
52 unsigned int ovs_net_id __read_mostly;
53 
54 static struct genl_family dp_packet_genl_family;
55 static struct genl_family dp_flow_genl_family;
56 static struct genl_family dp_datapath_genl_family;
57 
58 static const struct nla_policy flow_policy[];
59 
60 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
61 	.name = OVS_FLOW_MCGROUP,
62 };
63 
64 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
65 	.name = OVS_DATAPATH_MCGROUP,
66 };
67 
68 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
69 	.name = OVS_VPORT_MCGROUP,
70 };
71 
72 /* Check if need to build a reply message.
73  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
74 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
75 			    unsigned int group)
76 {
77 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
78 	       genl_has_listeners(family, genl_info_net(info), group);
79 }
80 
81 static void ovs_notify(struct genl_family *family,
82 		       struct sk_buff *skb, struct genl_info *info)
83 {
84 	genl_notify(family, skb, info, 0, GFP_KERNEL);
85 }
86 
87 /**
88  * DOC: Locking:
89  *
90  * All writes e.g. Writes to device state (add/remove datapath, port, set
91  * operations on vports, etc.), Writes to other state (flow table
92  * modifications, set miscellaneous datapath parameters, etc.) are protected
93  * by ovs_lock.
94  *
95  * Reads are protected by RCU.
96  *
97  * There are a few special cases (mostly stats) that have their own
98  * synchronization but they nest under all of above and don't interact with
99  * each other.
100  *
101  * The RTNL lock nests inside ovs_mutex.
102  */
103 
104 static DEFINE_MUTEX(ovs_mutex);
105 
106 void ovs_lock(void)
107 {
108 	mutex_lock(&ovs_mutex);
109 }
110 
111 void ovs_unlock(void)
112 {
113 	mutex_unlock(&ovs_mutex);
114 }
115 
116 #ifdef CONFIG_LOCKDEP
117 int lockdep_ovsl_is_held(void)
118 {
119 	if (debug_locks)
120 		return lockdep_is_held(&ovs_mutex);
121 	else
122 		return 1;
123 }
124 #endif
125 
126 static struct vport *new_vport(const struct vport_parms *);
127 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
128 			     const struct sw_flow_key *,
129 			     const struct dp_upcall_info *,
130 			     uint32_t cutlen);
131 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
132 				  const struct sw_flow_key *,
133 				  const struct dp_upcall_info *,
134 				  uint32_t cutlen);
135 
136 static void ovs_dp_masks_rebalance(struct work_struct *work);
137 
138 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
139 
140 /* Must be called with rcu_read_lock or ovs_mutex. */
141 const char *ovs_dp_name(const struct datapath *dp)
142 {
143 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
144 	return ovs_vport_name(vport);
145 }
146 
147 static int get_dpifindex(const struct datapath *dp)
148 {
149 	struct vport *local;
150 	int ifindex;
151 
152 	rcu_read_lock();
153 
154 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
155 	if (local)
156 		ifindex = local->dev->ifindex;
157 	else
158 		ifindex = 0;
159 
160 	rcu_read_unlock();
161 
162 	return ifindex;
163 }
164 
165 static void destroy_dp_rcu(struct rcu_head *rcu)
166 {
167 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
168 
169 	ovs_flow_tbl_destroy(&dp->table);
170 	free_percpu(dp->stats_percpu);
171 	kfree(dp->ports);
172 	ovs_meters_exit(dp);
173 	kfree(rcu_dereference_raw(dp->upcall_portids));
174 	kfree(dp);
175 }
176 
177 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
178 					    u16 port_no)
179 {
180 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
181 }
182 
183 /* Called with ovs_mutex or RCU read lock. */
184 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
185 {
186 	struct vport *vport;
187 	struct hlist_head *head;
188 
189 	head = vport_hash_bucket(dp, port_no);
190 	hlist_for_each_entry_rcu(vport, head, dp_hash_node,
191 				 lockdep_ovsl_is_held()) {
192 		if (vport->port_no == port_no)
193 			return vport;
194 	}
195 	return NULL;
196 }
197 
198 /* Called with ovs_mutex. */
199 static struct vport *new_vport(const struct vport_parms *parms)
200 {
201 	struct vport *vport;
202 
203 	vport = ovs_vport_add(parms);
204 	if (!IS_ERR(vport)) {
205 		struct datapath *dp = parms->dp;
206 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
207 
208 		hlist_add_head_rcu(&vport->dp_hash_node, head);
209 	}
210 	return vport;
211 }
212 
213 static void ovs_vport_update_upcall_stats(struct sk_buff *skb,
214 					  const struct dp_upcall_info *upcall_info,
215 					  bool upcall_result)
216 {
217 	struct vport *p = OVS_CB(skb)->input_vport;
218 	struct vport_upcall_stats_percpu *stats;
219 
220 	if (upcall_info->cmd != OVS_PACKET_CMD_MISS &&
221 	    upcall_info->cmd != OVS_PACKET_CMD_ACTION)
222 		return;
223 
224 	stats = this_cpu_ptr(p->upcall_stats);
225 	u64_stats_update_begin(&stats->syncp);
226 	if (upcall_result)
227 		u64_stats_inc(&stats->n_success);
228 	else
229 		u64_stats_inc(&stats->n_fail);
230 	u64_stats_update_end(&stats->syncp);
231 }
232 
233 void ovs_dp_detach_port(struct vport *p)
234 {
235 	ASSERT_OVSL();
236 
237 	/* First drop references to device. */
238 	hlist_del_rcu(&p->dp_hash_node);
239 
240 	/* Then destroy it. */
241 	ovs_vport_del(p);
242 }
243 
244 /* Must be called with rcu_read_lock. */
245 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
246 {
247 	const struct vport *p = OVS_CB(skb)->input_vport;
248 	struct datapath *dp = p->dp;
249 	struct sw_flow *flow;
250 	struct sw_flow_actions *sf_acts;
251 	struct dp_stats_percpu *stats;
252 	u64 *stats_counter;
253 	u32 n_mask_hit;
254 	u32 n_cache_hit;
255 	int error;
256 
257 	stats = this_cpu_ptr(dp->stats_percpu);
258 
259 	/* Look up flow. */
260 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
261 					 &n_mask_hit, &n_cache_hit);
262 	if (unlikely(!flow)) {
263 		struct dp_upcall_info upcall;
264 
265 		memset(&upcall, 0, sizeof(upcall));
266 		upcall.cmd = OVS_PACKET_CMD_MISS;
267 
268 		if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
269 			upcall.portid =
270 			    ovs_dp_get_upcall_portid(dp, smp_processor_id());
271 		else
272 			upcall.portid = ovs_vport_find_upcall_portid(p, skb);
273 
274 		upcall.mru = OVS_CB(skb)->mru;
275 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
276 		switch (error) {
277 		case 0:
278 		case -EAGAIN:
279 		case -ERESTARTSYS:
280 		case -EINTR:
281 			consume_skb(skb);
282 			break;
283 		default:
284 			kfree_skb(skb);
285 			break;
286 		}
287 		stats_counter = &stats->n_missed;
288 		goto out;
289 	}
290 
291 	ovs_flow_stats_update(flow, key->tp.flags, skb);
292 	sf_acts = rcu_dereference(flow->sf_acts);
293 	error = ovs_execute_actions(dp, skb, sf_acts, key);
294 	if (unlikely(error))
295 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
296 				    ovs_dp_name(dp), error);
297 
298 	stats_counter = &stats->n_hit;
299 
300 out:
301 	/* Update datapath statistics. */
302 	u64_stats_update_begin(&stats->syncp);
303 	(*stats_counter)++;
304 	stats->n_mask_hit += n_mask_hit;
305 	stats->n_cache_hit += n_cache_hit;
306 	u64_stats_update_end(&stats->syncp);
307 }
308 
309 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
310 		  const struct sw_flow_key *key,
311 		  const struct dp_upcall_info *upcall_info,
312 		  uint32_t cutlen)
313 {
314 	struct dp_stats_percpu *stats;
315 	int err;
316 
317 	if (trace_ovs_dp_upcall_enabled())
318 		trace_ovs_dp_upcall(dp, skb, key, upcall_info);
319 
320 	if (upcall_info->portid == 0) {
321 		err = -ENOTCONN;
322 		goto err;
323 	}
324 
325 	if (!skb_is_gso(skb))
326 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
327 	else
328 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
329 
330 	ovs_vport_update_upcall_stats(skb, upcall_info, !err);
331 	if (err)
332 		goto err;
333 
334 	return 0;
335 
336 err:
337 	stats = this_cpu_ptr(dp->stats_percpu);
338 
339 	u64_stats_update_begin(&stats->syncp);
340 	stats->n_lost++;
341 	u64_stats_update_end(&stats->syncp);
342 
343 	return err;
344 }
345 
346 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
347 			     const struct sw_flow_key *key,
348 			     const struct dp_upcall_info *upcall_info,
349 			     uint32_t cutlen)
350 {
351 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
352 	struct sw_flow_key later_key;
353 	struct sk_buff *segs, *nskb;
354 	int err;
355 
356 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
357 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
358 	if (IS_ERR(segs))
359 		return PTR_ERR(segs);
360 	if (segs == NULL)
361 		return -EINVAL;
362 
363 	if (gso_type & SKB_GSO_UDP) {
364 		/* The initial flow key extracted by ovs_flow_key_extract()
365 		 * in this case is for a first fragment, so we need to
366 		 * properly mark later fragments.
367 		 */
368 		later_key = *key;
369 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
370 	}
371 
372 	/* Queue all of the segments. */
373 	skb_list_walk_safe(segs, skb, nskb) {
374 		if (gso_type & SKB_GSO_UDP && skb != segs)
375 			key = &later_key;
376 
377 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
378 		if (err)
379 			break;
380 
381 	}
382 
383 	/* Free all of the segments. */
384 	skb_list_walk_safe(segs, skb, nskb) {
385 		if (err)
386 			kfree_skb(skb);
387 		else
388 			consume_skb(skb);
389 	}
390 	return err;
391 }
392 
393 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
394 			      unsigned int hdrlen, int actions_attrlen)
395 {
396 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
397 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
398 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
399 		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
400 		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
401 
402 	/* OVS_PACKET_ATTR_USERDATA */
403 	if (upcall_info->userdata)
404 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
405 
406 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
407 	if (upcall_info->egress_tun_info)
408 		size += nla_total_size(ovs_tun_key_attr_size());
409 
410 	/* OVS_PACKET_ATTR_ACTIONS */
411 	if (upcall_info->actions_len)
412 		size += nla_total_size(actions_attrlen);
413 
414 	/* OVS_PACKET_ATTR_MRU */
415 	if (upcall_info->mru)
416 		size += nla_total_size(sizeof(upcall_info->mru));
417 
418 	return size;
419 }
420 
421 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
422 {
423 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
424 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
425 
426 		if (plen > 0)
427 			skb_put_zero(skb, plen);
428 	}
429 }
430 
431 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
432 				  const struct sw_flow_key *key,
433 				  const struct dp_upcall_info *upcall_info,
434 				  uint32_t cutlen)
435 {
436 	struct ovs_header *upcall;
437 	struct sk_buff *nskb = NULL;
438 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
439 	struct nlattr *nla;
440 	size_t len;
441 	unsigned int hlen;
442 	int err, dp_ifindex;
443 	u64 hash;
444 
445 	dp_ifindex = get_dpifindex(dp);
446 	if (!dp_ifindex)
447 		return -ENODEV;
448 
449 	if (skb_vlan_tag_present(skb)) {
450 		nskb = skb_clone(skb, GFP_ATOMIC);
451 		if (!nskb)
452 			return -ENOMEM;
453 
454 		nskb = __vlan_hwaccel_push_inside(nskb);
455 		if (!nskb)
456 			return -ENOMEM;
457 
458 		skb = nskb;
459 	}
460 
461 	if (nla_attr_size(skb->len) > USHRT_MAX) {
462 		err = -EFBIG;
463 		goto out;
464 	}
465 
466 	/* Complete checksum if needed */
467 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
468 	    (err = skb_csum_hwoffload_help(skb, 0)))
469 		goto out;
470 
471 	/* Older versions of OVS user space enforce alignment of the last
472 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
473 	 * padding logic. Only perform zerocopy if padding is not required.
474 	 */
475 	if (dp->user_features & OVS_DP_F_UNALIGNED)
476 		hlen = skb_zerocopy_headlen(skb);
477 	else
478 		hlen = skb->len;
479 
480 	len = upcall_msg_size(upcall_info, hlen - cutlen,
481 			      OVS_CB(skb)->acts_origlen);
482 	user_skb = genlmsg_new(len, GFP_ATOMIC);
483 	if (!user_skb) {
484 		err = -ENOMEM;
485 		goto out;
486 	}
487 
488 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
489 			     0, upcall_info->cmd);
490 	if (!upcall) {
491 		err = -EINVAL;
492 		goto out;
493 	}
494 	upcall->dp_ifindex = dp_ifindex;
495 
496 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
497 	if (err)
498 		goto out;
499 
500 	if (upcall_info->userdata)
501 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
502 			  nla_len(upcall_info->userdata),
503 			  nla_data(upcall_info->userdata));
504 
505 	if (upcall_info->egress_tun_info) {
506 		nla = nla_nest_start_noflag(user_skb,
507 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
508 		if (!nla) {
509 			err = -EMSGSIZE;
510 			goto out;
511 		}
512 		err = ovs_nla_put_tunnel_info(user_skb,
513 					      upcall_info->egress_tun_info);
514 		if (err)
515 			goto out;
516 
517 		nla_nest_end(user_skb, nla);
518 	}
519 
520 	if (upcall_info->actions_len) {
521 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
522 		if (!nla) {
523 			err = -EMSGSIZE;
524 			goto out;
525 		}
526 		err = ovs_nla_put_actions(upcall_info->actions,
527 					  upcall_info->actions_len,
528 					  user_skb);
529 		if (!err)
530 			nla_nest_end(user_skb, nla);
531 		else
532 			nla_nest_cancel(user_skb, nla);
533 	}
534 
535 	/* Add OVS_PACKET_ATTR_MRU */
536 	if (upcall_info->mru &&
537 	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
538 		err = -ENOBUFS;
539 		goto out;
540 	}
541 
542 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
543 	if (cutlen > 0 &&
544 	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
545 		err = -ENOBUFS;
546 		goto out;
547 	}
548 
549 	/* Add OVS_PACKET_ATTR_HASH */
550 	hash = skb_get_hash_raw(skb);
551 	if (skb->sw_hash)
552 		hash |= OVS_PACKET_HASH_SW_BIT;
553 
554 	if (skb->l4_hash)
555 		hash |= OVS_PACKET_HASH_L4_BIT;
556 
557 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
558 		err = -ENOBUFS;
559 		goto out;
560 	}
561 
562 	/* Only reserve room for attribute header, packet data is added
563 	 * in skb_zerocopy() */
564 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
565 		err = -ENOBUFS;
566 		goto out;
567 	}
568 	nla->nla_len = nla_attr_size(skb->len - cutlen);
569 
570 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
571 	if (err)
572 		goto out;
573 
574 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
575 	pad_packet(dp, user_skb);
576 
577 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
578 
579 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
580 	user_skb = NULL;
581 out:
582 	if (err)
583 		skb_tx_error(skb);
584 	consume_skb(user_skb);
585 	consume_skb(nskb);
586 
587 	return err;
588 }
589 
590 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
591 {
592 	struct ovs_header *ovs_header = info->userhdr;
593 	struct net *net = sock_net(skb->sk);
594 	struct nlattr **a = info->attrs;
595 	struct sw_flow_actions *acts;
596 	struct sk_buff *packet;
597 	struct sw_flow *flow;
598 	struct sw_flow_actions *sf_acts;
599 	struct datapath *dp;
600 	struct vport *input_vport;
601 	u16 mru = 0;
602 	u64 hash;
603 	int len;
604 	int err;
605 	bool log = !a[OVS_PACKET_ATTR_PROBE];
606 
607 	err = -EINVAL;
608 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
609 	    !a[OVS_PACKET_ATTR_ACTIONS])
610 		goto err;
611 
612 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
613 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
614 	err = -ENOMEM;
615 	if (!packet)
616 		goto err;
617 	skb_reserve(packet, NET_IP_ALIGN);
618 
619 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
620 
621 	/* Set packet's mru */
622 	if (a[OVS_PACKET_ATTR_MRU]) {
623 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
624 		packet->ignore_df = 1;
625 	}
626 	OVS_CB(packet)->mru = mru;
627 
628 	if (a[OVS_PACKET_ATTR_HASH]) {
629 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
630 
631 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
632 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
633 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
634 	}
635 
636 	/* Build an sw_flow for sending this packet. */
637 	flow = ovs_flow_alloc();
638 	err = PTR_ERR(flow);
639 	if (IS_ERR(flow))
640 		goto err_kfree_skb;
641 
642 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
643 					     packet, &flow->key, log);
644 	if (err)
645 		goto err_flow_free;
646 
647 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
648 				   &flow->key, &acts, log);
649 	if (err)
650 		goto err_flow_free;
651 
652 	rcu_assign_pointer(flow->sf_acts, acts);
653 	packet->priority = flow->key.phy.priority;
654 	packet->mark = flow->key.phy.skb_mark;
655 
656 	rcu_read_lock();
657 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
658 	err = -ENODEV;
659 	if (!dp)
660 		goto err_unlock;
661 
662 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
663 	if (!input_vport)
664 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
665 
666 	if (!input_vport)
667 		goto err_unlock;
668 
669 	packet->dev = input_vport->dev;
670 	OVS_CB(packet)->input_vport = input_vport;
671 	sf_acts = rcu_dereference(flow->sf_acts);
672 
673 	local_bh_disable();
674 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
675 	local_bh_enable();
676 	rcu_read_unlock();
677 
678 	ovs_flow_free(flow, false);
679 	return err;
680 
681 err_unlock:
682 	rcu_read_unlock();
683 err_flow_free:
684 	ovs_flow_free(flow, false);
685 err_kfree_skb:
686 	kfree_skb(packet);
687 err:
688 	return err;
689 }
690 
691 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
692 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
693 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
694 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
695 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
696 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
697 	[OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
698 };
699 
700 static const struct genl_small_ops dp_packet_genl_ops[] = {
701 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
702 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
703 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
704 	  .doit = ovs_packet_cmd_execute
705 	}
706 };
707 
708 static struct genl_family dp_packet_genl_family __ro_after_init = {
709 	.hdrsize = sizeof(struct ovs_header),
710 	.name = OVS_PACKET_FAMILY,
711 	.version = OVS_PACKET_VERSION,
712 	.maxattr = OVS_PACKET_ATTR_MAX,
713 	.policy = packet_policy,
714 	.netnsok = true,
715 	.parallel_ops = true,
716 	.small_ops = dp_packet_genl_ops,
717 	.n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
718 	.resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
719 	.module = THIS_MODULE,
720 };
721 
722 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
723 			 struct ovs_dp_megaflow_stats *mega_stats)
724 {
725 	int i;
726 
727 	memset(mega_stats, 0, sizeof(*mega_stats));
728 
729 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
730 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
731 
732 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
733 
734 	for_each_possible_cpu(i) {
735 		const struct dp_stats_percpu *percpu_stats;
736 		struct dp_stats_percpu local_stats;
737 		unsigned int start;
738 
739 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
740 
741 		do {
742 			start = u64_stats_fetch_begin(&percpu_stats->syncp);
743 			local_stats = *percpu_stats;
744 		} while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
745 
746 		stats->n_hit += local_stats.n_hit;
747 		stats->n_missed += local_stats.n_missed;
748 		stats->n_lost += local_stats.n_lost;
749 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
750 		mega_stats->n_cache_hit += local_stats.n_cache_hit;
751 	}
752 }
753 
754 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
755 {
756 	return ovs_identifier_is_ufid(sfid) &&
757 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
758 }
759 
760 static bool should_fill_mask(uint32_t ufid_flags)
761 {
762 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
763 }
764 
765 static bool should_fill_actions(uint32_t ufid_flags)
766 {
767 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
768 }
769 
770 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
771 				    const struct sw_flow_id *sfid,
772 				    uint32_t ufid_flags)
773 {
774 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
775 
776 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
777 	 * see ovs_nla_put_identifier()
778 	 */
779 	if (sfid && ovs_identifier_is_ufid(sfid))
780 		len += nla_total_size(sfid->ufid_len);
781 	else
782 		len += nla_total_size(ovs_key_attr_size());
783 
784 	/* OVS_FLOW_ATTR_KEY */
785 	if (!sfid || should_fill_key(sfid, ufid_flags))
786 		len += nla_total_size(ovs_key_attr_size());
787 
788 	/* OVS_FLOW_ATTR_MASK */
789 	if (should_fill_mask(ufid_flags))
790 		len += nla_total_size(ovs_key_attr_size());
791 
792 	/* OVS_FLOW_ATTR_ACTIONS */
793 	if (should_fill_actions(ufid_flags))
794 		len += nla_total_size(acts->orig_len);
795 
796 	return len
797 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
798 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
799 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
800 }
801 
802 /* Called with ovs_mutex or RCU read lock. */
803 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
804 				   struct sk_buff *skb)
805 {
806 	struct ovs_flow_stats stats;
807 	__be16 tcp_flags;
808 	unsigned long used;
809 
810 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
811 
812 	if (used &&
813 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
814 			      OVS_FLOW_ATTR_PAD))
815 		return -EMSGSIZE;
816 
817 	if (stats.n_packets &&
818 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
819 			  sizeof(struct ovs_flow_stats), &stats,
820 			  OVS_FLOW_ATTR_PAD))
821 		return -EMSGSIZE;
822 
823 	if ((u8)ntohs(tcp_flags) &&
824 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
825 		return -EMSGSIZE;
826 
827 	return 0;
828 }
829 
830 /* Called with ovs_mutex or RCU read lock. */
831 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
832 				     struct sk_buff *skb, int skb_orig_len)
833 {
834 	struct nlattr *start;
835 	int err;
836 
837 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
838 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
839 	 * Netlink but individual action lists can be longer than
840 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
841 	 * The userspace caller can always fetch the actions separately if it
842 	 * really wants them.  (Most userspace callers in fact don't care.)
843 	 *
844 	 * This can only fail for dump operations because the skb is always
845 	 * properly sized for single flows.
846 	 */
847 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
848 	if (start) {
849 		const struct sw_flow_actions *sf_acts;
850 
851 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
852 		err = ovs_nla_put_actions(sf_acts->actions,
853 					  sf_acts->actions_len, skb);
854 
855 		if (!err)
856 			nla_nest_end(skb, start);
857 		else {
858 			if (skb_orig_len)
859 				return err;
860 
861 			nla_nest_cancel(skb, start);
862 		}
863 	} else if (skb_orig_len) {
864 		return -EMSGSIZE;
865 	}
866 
867 	return 0;
868 }
869 
870 /* Called with ovs_mutex or RCU read lock. */
871 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
872 				  struct sk_buff *skb, u32 portid,
873 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
874 {
875 	const int skb_orig_len = skb->len;
876 	struct ovs_header *ovs_header;
877 	int err;
878 
879 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
880 				 flags, cmd);
881 	if (!ovs_header)
882 		return -EMSGSIZE;
883 
884 	ovs_header->dp_ifindex = dp_ifindex;
885 
886 	err = ovs_nla_put_identifier(flow, skb);
887 	if (err)
888 		goto error;
889 
890 	if (should_fill_key(&flow->id, ufid_flags)) {
891 		err = ovs_nla_put_masked_key(flow, skb);
892 		if (err)
893 			goto error;
894 	}
895 
896 	if (should_fill_mask(ufid_flags)) {
897 		err = ovs_nla_put_mask(flow, skb);
898 		if (err)
899 			goto error;
900 	}
901 
902 	err = ovs_flow_cmd_fill_stats(flow, skb);
903 	if (err)
904 		goto error;
905 
906 	if (should_fill_actions(ufid_flags)) {
907 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
908 		if (err)
909 			goto error;
910 	}
911 
912 	genlmsg_end(skb, ovs_header);
913 	return 0;
914 
915 error:
916 	genlmsg_cancel(skb, ovs_header);
917 	return err;
918 }
919 
920 /* May not be called with RCU read lock. */
921 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
922 					       const struct sw_flow_id *sfid,
923 					       struct genl_info *info,
924 					       bool always,
925 					       uint32_t ufid_flags)
926 {
927 	struct sk_buff *skb;
928 	size_t len;
929 
930 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
931 		return NULL;
932 
933 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
934 	skb = genlmsg_new(len, GFP_KERNEL);
935 	if (!skb)
936 		return ERR_PTR(-ENOMEM);
937 
938 	return skb;
939 }
940 
941 /* Called with ovs_mutex. */
942 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
943 					       int dp_ifindex,
944 					       struct genl_info *info, u8 cmd,
945 					       bool always, u32 ufid_flags)
946 {
947 	struct sk_buff *skb;
948 	int retval;
949 
950 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
951 				      &flow->id, info, always, ufid_flags);
952 	if (IS_ERR_OR_NULL(skb))
953 		return skb;
954 
955 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
956 					info->snd_portid, info->snd_seq, 0,
957 					cmd, ufid_flags);
958 	if (WARN_ON_ONCE(retval < 0)) {
959 		kfree_skb(skb);
960 		skb = ERR_PTR(retval);
961 	}
962 	return skb;
963 }
964 
965 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
966 {
967 	struct net *net = sock_net(skb->sk);
968 	struct nlattr **a = info->attrs;
969 	struct ovs_header *ovs_header = info->userhdr;
970 	struct sw_flow *flow = NULL, *new_flow;
971 	struct sw_flow_mask mask;
972 	struct sk_buff *reply;
973 	struct datapath *dp;
974 	struct sw_flow_key *key;
975 	struct sw_flow_actions *acts;
976 	struct sw_flow_match match;
977 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
978 	int error;
979 	bool log = !a[OVS_FLOW_ATTR_PROBE];
980 
981 	/* Must have key and actions. */
982 	error = -EINVAL;
983 	if (!a[OVS_FLOW_ATTR_KEY]) {
984 		OVS_NLERR(log, "Flow key attr not present in new flow.");
985 		goto error;
986 	}
987 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
988 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
989 		goto error;
990 	}
991 
992 	/* Most of the time we need to allocate a new flow, do it before
993 	 * locking.
994 	 */
995 	new_flow = ovs_flow_alloc();
996 	if (IS_ERR(new_flow)) {
997 		error = PTR_ERR(new_flow);
998 		goto error;
999 	}
1000 
1001 	/* Extract key. */
1002 	key = kzalloc(sizeof(*key), GFP_KERNEL);
1003 	if (!key) {
1004 		error = -ENOMEM;
1005 		goto err_kfree_flow;
1006 	}
1007 
1008 	ovs_match_init(&match, key, false, &mask);
1009 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1010 				  a[OVS_FLOW_ATTR_MASK], log);
1011 	if (error)
1012 		goto err_kfree_key;
1013 
1014 	ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1015 
1016 	/* Extract flow identifier. */
1017 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
1018 				       key, log);
1019 	if (error)
1020 		goto err_kfree_key;
1021 
1022 	/* Validate actions. */
1023 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1024 				     &new_flow->key, &acts, log);
1025 	if (error) {
1026 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1027 		goto err_kfree_key;
1028 	}
1029 
1030 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1031 					ufid_flags);
1032 	if (IS_ERR(reply)) {
1033 		error = PTR_ERR(reply);
1034 		goto err_kfree_acts;
1035 	}
1036 
1037 	ovs_lock();
1038 	dp = get_dp(net, ovs_header->dp_ifindex);
1039 	if (unlikely(!dp)) {
1040 		error = -ENODEV;
1041 		goto err_unlock_ovs;
1042 	}
1043 
1044 	/* Check if this is a duplicate flow */
1045 	if (ovs_identifier_is_ufid(&new_flow->id))
1046 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1047 	if (!flow)
1048 		flow = ovs_flow_tbl_lookup(&dp->table, key);
1049 	if (likely(!flow)) {
1050 		rcu_assign_pointer(new_flow->sf_acts, acts);
1051 
1052 		/* Put flow in bucket. */
1053 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1054 		if (unlikely(error)) {
1055 			acts = NULL;
1056 			goto err_unlock_ovs;
1057 		}
1058 
1059 		if (unlikely(reply)) {
1060 			error = ovs_flow_cmd_fill_info(new_flow,
1061 						       ovs_header->dp_ifindex,
1062 						       reply, info->snd_portid,
1063 						       info->snd_seq, 0,
1064 						       OVS_FLOW_CMD_NEW,
1065 						       ufid_flags);
1066 			BUG_ON(error < 0);
1067 		}
1068 		ovs_unlock();
1069 	} else {
1070 		struct sw_flow_actions *old_acts;
1071 
1072 		/* Bail out if we're not allowed to modify an existing flow.
1073 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1074 		 * because Generic Netlink treats the latter as a dump
1075 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1076 		 * gets fixed.
1077 		 */
1078 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1079 							 | NLM_F_EXCL))) {
1080 			error = -EEXIST;
1081 			goto err_unlock_ovs;
1082 		}
1083 		/* The flow identifier has to be the same for flow updates.
1084 		 * Look for any overlapping flow.
1085 		 */
1086 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1087 			if (ovs_identifier_is_key(&flow->id))
1088 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1089 								 &match);
1090 			else /* UFID matches but key is different */
1091 				flow = NULL;
1092 			if (!flow) {
1093 				error = -ENOENT;
1094 				goto err_unlock_ovs;
1095 			}
1096 		}
1097 		/* Update actions. */
1098 		old_acts = ovsl_dereference(flow->sf_acts);
1099 		rcu_assign_pointer(flow->sf_acts, acts);
1100 
1101 		if (unlikely(reply)) {
1102 			error = ovs_flow_cmd_fill_info(flow,
1103 						       ovs_header->dp_ifindex,
1104 						       reply, info->snd_portid,
1105 						       info->snd_seq, 0,
1106 						       OVS_FLOW_CMD_NEW,
1107 						       ufid_flags);
1108 			BUG_ON(error < 0);
1109 		}
1110 		ovs_unlock();
1111 
1112 		ovs_nla_free_flow_actions_rcu(old_acts);
1113 		ovs_flow_free(new_flow, false);
1114 	}
1115 
1116 	if (reply)
1117 		ovs_notify(&dp_flow_genl_family, reply, info);
1118 
1119 	kfree(key);
1120 	return 0;
1121 
1122 err_unlock_ovs:
1123 	ovs_unlock();
1124 	kfree_skb(reply);
1125 err_kfree_acts:
1126 	ovs_nla_free_flow_actions(acts);
1127 err_kfree_key:
1128 	kfree(key);
1129 err_kfree_flow:
1130 	ovs_flow_free(new_flow, false);
1131 error:
1132 	return error;
1133 }
1134 
1135 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1136 static noinline_for_stack
1137 struct sw_flow_actions *get_flow_actions(struct net *net,
1138 					 const struct nlattr *a,
1139 					 const struct sw_flow_key *key,
1140 					 const struct sw_flow_mask *mask,
1141 					 bool log)
1142 {
1143 	struct sw_flow_actions *acts;
1144 	struct sw_flow_key masked_key;
1145 	int error;
1146 
1147 	ovs_flow_mask_key(&masked_key, key, true, mask);
1148 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1149 	if (error) {
1150 		OVS_NLERR(log,
1151 			  "Actions may not be safe on all matching packets");
1152 		return ERR_PTR(error);
1153 	}
1154 
1155 	return acts;
1156 }
1157 
1158 /* Factor out match-init and action-copy to avoid
1159  * "Wframe-larger-than=1024" warning. Because mask is only
1160  * used to get actions, we new a function to save some
1161  * stack space.
1162  *
1163  * If there are not key and action attrs, we return 0
1164  * directly. In the case, the caller will also not use the
1165  * match as before. If there is action attr, we try to get
1166  * actions and save them to *acts. Before returning from
1167  * the function, we reset the match->mask pointer. Because
1168  * we should not to return match object with dangling reference
1169  * to mask.
1170  * */
1171 static noinline_for_stack int
1172 ovs_nla_init_match_and_action(struct net *net,
1173 			      struct sw_flow_match *match,
1174 			      struct sw_flow_key *key,
1175 			      struct nlattr **a,
1176 			      struct sw_flow_actions **acts,
1177 			      bool log)
1178 {
1179 	struct sw_flow_mask mask;
1180 	int error = 0;
1181 
1182 	if (a[OVS_FLOW_ATTR_KEY]) {
1183 		ovs_match_init(match, key, true, &mask);
1184 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1185 					  a[OVS_FLOW_ATTR_MASK], log);
1186 		if (error)
1187 			goto error;
1188 	}
1189 
1190 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1191 		if (!a[OVS_FLOW_ATTR_KEY]) {
1192 			OVS_NLERR(log,
1193 				  "Flow key attribute not present in set flow.");
1194 			error = -EINVAL;
1195 			goto error;
1196 		}
1197 
1198 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1199 					 &mask, log);
1200 		if (IS_ERR(*acts)) {
1201 			error = PTR_ERR(*acts);
1202 			goto error;
1203 		}
1204 	}
1205 
1206 	/* On success, error is 0. */
1207 error:
1208 	match->mask = NULL;
1209 	return error;
1210 }
1211 
1212 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1213 {
1214 	struct net *net = sock_net(skb->sk);
1215 	struct nlattr **a = info->attrs;
1216 	struct ovs_header *ovs_header = info->userhdr;
1217 	struct sw_flow_key key;
1218 	struct sw_flow *flow;
1219 	struct sk_buff *reply = NULL;
1220 	struct datapath *dp;
1221 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1222 	struct sw_flow_match match;
1223 	struct sw_flow_id sfid;
1224 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1225 	int error = 0;
1226 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1227 	bool ufid_present;
1228 
1229 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1230 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1231 		OVS_NLERR(log,
1232 			  "Flow set message rejected, Key attribute missing.");
1233 		return -EINVAL;
1234 	}
1235 
1236 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1237 					      &acts, log);
1238 	if (error)
1239 		goto error;
1240 
1241 	if (acts) {
1242 		/* Can allocate before locking if have acts. */
1243 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1244 						ufid_flags);
1245 		if (IS_ERR(reply)) {
1246 			error = PTR_ERR(reply);
1247 			goto err_kfree_acts;
1248 		}
1249 	}
1250 
1251 	ovs_lock();
1252 	dp = get_dp(net, ovs_header->dp_ifindex);
1253 	if (unlikely(!dp)) {
1254 		error = -ENODEV;
1255 		goto err_unlock_ovs;
1256 	}
1257 	/* Check that the flow exists. */
1258 	if (ufid_present)
1259 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1260 	else
1261 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1262 	if (unlikely(!flow)) {
1263 		error = -ENOENT;
1264 		goto err_unlock_ovs;
1265 	}
1266 
1267 	/* Update actions, if present. */
1268 	if (likely(acts)) {
1269 		old_acts = ovsl_dereference(flow->sf_acts);
1270 		rcu_assign_pointer(flow->sf_acts, acts);
1271 
1272 		if (unlikely(reply)) {
1273 			error = ovs_flow_cmd_fill_info(flow,
1274 						       ovs_header->dp_ifindex,
1275 						       reply, info->snd_portid,
1276 						       info->snd_seq, 0,
1277 						       OVS_FLOW_CMD_SET,
1278 						       ufid_flags);
1279 			BUG_ON(error < 0);
1280 		}
1281 	} else {
1282 		/* Could not alloc without acts before locking. */
1283 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1284 						info, OVS_FLOW_CMD_SET, false,
1285 						ufid_flags);
1286 
1287 		if (IS_ERR(reply)) {
1288 			error = PTR_ERR(reply);
1289 			goto err_unlock_ovs;
1290 		}
1291 	}
1292 
1293 	/* Clear stats. */
1294 	if (a[OVS_FLOW_ATTR_CLEAR])
1295 		ovs_flow_stats_clear(flow);
1296 	ovs_unlock();
1297 
1298 	if (reply)
1299 		ovs_notify(&dp_flow_genl_family, reply, info);
1300 	if (old_acts)
1301 		ovs_nla_free_flow_actions_rcu(old_acts);
1302 
1303 	return 0;
1304 
1305 err_unlock_ovs:
1306 	ovs_unlock();
1307 	kfree_skb(reply);
1308 err_kfree_acts:
1309 	ovs_nla_free_flow_actions(acts);
1310 error:
1311 	return error;
1312 }
1313 
1314 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1315 {
1316 	struct nlattr **a = info->attrs;
1317 	struct ovs_header *ovs_header = info->userhdr;
1318 	struct net *net = sock_net(skb->sk);
1319 	struct sw_flow_key key;
1320 	struct sk_buff *reply;
1321 	struct sw_flow *flow;
1322 	struct datapath *dp;
1323 	struct sw_flow_match match;
1324 	struct sw_flow_id ufid;
1325 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1326 	int err = 0;
1327 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1328 	bool ufid_present;
1329 
1330 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1331 	if (a[OVS_FLOW_ATTR_KEY]) {
1332 		ovs_match_init(&match, &key, true, NULL);
1333 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1334 					log);
1335 	} else if (!ufid_present) {
1336 		OVS_NLERR(log,
1337 			  "Flow get message rejected, Key attribute missing.");
1338 		err = -EINVAL;
1339 	}
1340 	if (err)
1341 		return err;
1342 
1343 	ovs_lock();
1344 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1345 	if (!dp) {
1346 		err = -ENODEV;
1347 		goto unlock;
1348 	}
1349 
1350 	if (ufid_present)
1351 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1352 	else
1353 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1354 	if (!flow) {
1355 		err = -ENOENT;
1356 		goto unlock;
1357 	}
1358 
1359 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1360 					OVS_FLOW_CMD_GET, true, ufid_flags);
1361 	if (IS_ERR(reply)) {
1362 		err = PTR_ERR(reply);
1363 		goto unlock;
1364 	}
1365 
1366 	ovs_unlock();
1367 	return genlmsg_reply(reply, info);
1368 unlock:
1369 	ovs_unlock();
1370 	return err;
1371 }
1372 
1373 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1374 {
1375 	struct nlattr **a = info->attrs;
1376 	struct ovs_header *ovs_header = info->userhdr;
1377 	struct net *net = sock_net(skb->sk);
1378 	struct sw_flow_key key;
1379 	struct sk_buff *reply;
1380 	struct sw_flow *flow = NULL;
1381 	struct datapath *dp;
1382 	struct sw_flow_match match;
1383 	struct sw_flow_id ufid;
1384 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1385 	int err;
1386 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1387 	bool ufid_present;
1388 
1389 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1390 	if (a[OVS_FLOW_ATTR_KEY]) {
1391 		ovs_match_init(&match, &key, true, NULL);
1392 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1393 					NULL, log);
1394 		if (unlikely(err))
1395 			return err;
1396 	}
1397 
1398 	ovs_lock();
1399 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1400 	if (unlikely(!dp)) {
1401 		err = -ENODEV;
1402 		goto unlock;
1403 	}
1404 
1405 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1406 		err = ovs_flow_tbl_flush(&dp->table);
1407 		goto unlock;
1408 	}
1409 
1410 	if (ufid_present)
1411 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1412 	else
1413 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1414 	if (unlikely(!flow)) {
1415 		err = -ENOENT;
1416 		goto unlock;
1417 	}
1418 
1419 	ovs_flow_tbl_remove(&dp->table, flow);
1420 	ovs_unlock();
1421 
1422 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1423 					&flow->id, info, false, ufid_flags);
1424 	if (likely(reply)) {
1425 		if (!IS_ERR(reply)) {
1426 			rcu_read_lock();	/*To keep RCU checker happy. */
1427 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1428 						     reply, info->snd_portid,
1429 						     info->snd_seq, 0,
1430 						     OVS_FLOW_CMD_DEL,
1431 						     ufid_flags);
1432 			rcu_read_unlock();
1433 			if (WARN_ON_ONCE(err < 0)) {
1434 				kfree_skb(reply);
1435 				goto out_free;
1436 			}
1437 
1438 			ovs_notify(&dp_flow_genl_family, reply, info);
1439 		} else {
1440 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1441 					PTR_ERR(reply));
1442 		}
1443 	}
1444 
1445 out_free:
1446 	ovs_flow_free(flow, true);
1447 	return 0;
1448 unlock:
1449 	ovs_unlock();
1450 	return err;
1451 }
1452 
1453 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1454 {
1455 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1456 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1457 	struct table_instance *ti;
1458 	struct datapath *dp;
1459 	u32 ufid_flags;
1460 	int err;
1461 
1462 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1463 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1464 	if (err)
1465 		return err;
1466 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1467 
1468 	rcu_read_lock();
1469 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1470 	if (!dp) {
1471 		rcu_read_unlock();
1472 		return -ENODEV;
1473 	}
1474 
1475 	ti = rcu_dereference(dp->table.ti);
1476 	for (;;) {
1477 		struct sw_flow *flow;
1478 		u32 bucket, obj;
1479 
1480 		bucket = cb->args[0];
1481 		obj = cb->args[1];
1482 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1483 		if (!flow)
1484 			break;
1485 
1486 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1487 					   NETLINK_CB(cb->skb).portid,
1488 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1489 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1490 			break;
1491 
1492 		cb->args[0] = bucket;
1493 		cb->args[1] = obj;
1494 	}
1495 	rcu_read_unlock();
1496 	return skb->len;
1497 }
1498 
1499 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1500 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1501 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1502 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1503 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1504 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1505 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1506 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1507 };
1508 
1509 static const struct genl_small_ops dp_flow_genl_ops[] = {
1510 	{ .cmd = OVS_FLOW_CMD_NEW,
1511 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1512 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1513 	  .doit = ovs_flow_cmd_new
1514 	},
1515 	{ .cmd = OVS_FLOW_CMD_DEL,
1516 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1517 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1518 	  .doit = ovs_flow_cmd_del
1519 	},
1520 	{ .cmd = OVS_FLOW_CMD_GET,
1521 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1522 	  .flags = 0,		    /* OK for unprivileged users. */
1523 	  .doit = ovs_flow_cmd_get,
1524 	  .dumpit = ovs_flow_cmd_dump
1525 	},
1526 	{ .cmd = OVS_FLOW_CMD_SET,
1527 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1528 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1529 	  .doit = ovs_flow_cmd_set,
1530 	},
1531 };
1532 
1533 static struct genl_family dp_flow_genl_family __ro_after_init = {
1534 	.hdrsize = sizeof(struct ovs_header),
1535 	.name = OVS_FLOW_FAMILY,
1536 	.version = OVS_FLOW_VERSION,
1537 	.maxattr = OVS_FLOW_ATTR_MAX,
1538 	.policy = flow_policy,
1539 	.netnsok = true,
1540 	.parallel_ops = true,
1541 	.small_ops = dp_flow_genl_ops,
1542 	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1543 	.resv_start_op = OVS_FLOW_CMD_SET + 1,
1544 	.mcgrps = &ovs_dp_flow_multicast_group,
1545 	.n_mcgrps = 1,
1546 	.module = THIS_MODULE,
1547 };
1548 
1549 static size_t ovs_dp_cmd_msg_size(void)
1550 {
1551 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1552 
1553 	msgsize += nla_total_size(IFNAMSIZ);
1554 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1555 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1556 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1557 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1558 	msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1559 
1560 	return msgsize;
1561 }
1562 
1563 /* Called with ovs_mutex. */
1564 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1565 				u32 portid, u32 seq, u32 flags, u8 cmd)
1566 {
1567 	struct ovs_header *ovs_header;
1568 	struct ovs_dp_stats dp_stats;
1569 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1570 	struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1571 	int err, pids_len;
1572 
1573 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1574 				 flags, cmd);
1575 	if (!ovs_header)
1576 		goto error;
1577 
1578 	ovs_header->dp_ifindex = get_dpifindex(dp);
1579 
1580 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1581 	if (err)
1582 		goto nla_put_failure;
1583 
1584 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1585 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1586 			  &dp_stats, OVS_DP_ATTR_PAD))
1587 		goto nla_put_failure;
1588 
1589 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1590 			  sizeof(struct ovs_dp_megaflow_stats),
1591 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1592 		goto nla_put_failure;
1593 
1594 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1595 		goto nla_put_failure;
1596 
1597 	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1598 			ovs_flow_tbl_masks_cache_size(&dp->table)))
1599 		goto nla_put_failure;
1600 
1601 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1602 		pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1603 		if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1604 			goto nla_put_failure;
1605 	}
1606 
1607 	genlmsg_end(skb, ovs_header);
1608 	return 0;
1609 
1610 nla_put_failure:
1611 	genlmsg_cancel(skb, ovs_header);
1612 error:
1613 	return -EMSGSIZE;
1614 }
1615 
1616 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1617 {
1618 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1619 }
1620 
1621 /* Called with rcu_read_lock or ovs_mutex. */
1622 static struct datapath *lookup_datapath(struct net *net,
1623 					const struct ovs_header *ovs_header,
1624 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1625 {
1626 	struct datapath *dp;
1627 
1628 	if (!a[OVS_DP_ATTR_NAME])
1629 		dp = get_dp(net, ovs_header->dp_ifindex);
1630 	else {
1631 		struct vport *vport;
1632 
1633 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1634 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1635 	}
1636 	return dp ? dp : ERR_PTR(-ENODEV);
1637 }
1638 
1639 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1640 				       struct genl_info *info)
1641 {
1642 	struct datapath *dp;
1643 
1644 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1645 			     info->attrs);
1646 	if (IS_ERR(dp))
1647 		return;
1648 
1649 	pr_warn("%s: Dropping previously announced user features\n",
1650 		ovs_dp_name(dp));
1651 	dp->user_features = 0;
1652 }
1653 
1654 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1655 			      const struct nlattr *ids)
1656 {
1657 	struct dp_nlsk_pids *old, *dp_nlsk_pids;
1658 
1659 	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1660 		return -EINVAL;
1661 
1662 	old = ovsl_dereference(dp->upcall_portids);
1663 
1664 	dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1665 			       GFP_KERNEL);
1666 	if (!dp_nlsk_pids)
1667 		return -ENOMEM;
1668 
1669 	dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1670 	nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1671 
1672 	rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1673 
1674 	kfree_rcu(old, rcu);
1675 
1676 	return 0;
1677 }
1678 
1679 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1680 {
1681 	struct dp_nlsk_pids *dp_nlsk_pids;
1682 
1683 	dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1684 
1685 	if (dp_nlsk_pids) {
1686 		if (cpu_id < dp_nlsk_pids->n_pids) {
1687 			return dp_nlsk_pids->pids[cpu_id];
1688 		} else if (dp_nlsk_pids->n_pids > 0 &&
1689 			   cpu_id >= dp_nlsk_pids->n_pids) {
1690 			/* If the number of netlink PIDs is mismatched with
1691 			 * the number of CPUs as seen by the kernel, log this
1692 			 * and send the upcall to an arbitrary socket (0) in
1693 			 * order to not drop packets
1694 			 */
1695 			pr_info_ratelimited("cpu_id mismatch with handler threads");
1696 			return dp_nlsk_pids->pids[cpu_id %
1697 						  dp_nlsk_pids->n_pids];
1698 		} else {
1699 			return 0;
1700 		}
1701 	} else {
1702 		return 0;
1703 	}
1704 }
1705 
1706 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1707 {
1708 	u32 user_features = 0, old_features = dp->user_features;
1709 	int err;
1710 
1711 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1712 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1713 
1714 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1715 				      OVS_DP_F_UNALIGNED |
1716 				      OVS_DP_F_TC_RECIRC_SHARING |
1717 				      OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1718 			return -EOPNOTSUPP;
1719 
1720 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1721 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1722 			return -EOPNOTSUPP;
1723 #endif
1724 	}
1725 
1726 	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1727 		int err;
1728 		u32 cache_size;
1729 
1730 		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1731 		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1732 		if (err)
1733 			return err;
1734 	}
1735 
1736 	dp->user_features = user_features;
1737 
1738 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1739 	    a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1740 		/* Upcall Netlink Port IDs have been updated */
1741 		err = ovs_dp_set_upcall_portids(dp,
1742 						a[OVS_DP_ATTR_PER_CPU_PIDS]);
1743 		if (err)
1744 			return err;
1745 	}
1746 
1747 	if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1748 	    !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1749 		tc_skb_ext_tc_enable();
1750 	else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1751 		 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1752 		tc_skb_ext_tc_disable();
1753 
1754 	return 0;
1755 }
1756 
1757 static int ovs_dp_stats_init(struct datapath *dp)
1758 {
1759 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1760 	if (!dp->stats_percpu)
1761 		return -ENOMEM;
1762 
1763 	return 0;
1764 }
1765 
1766 static int ovs_dp_vport_init(struct datapath *dp)
1767 {
1768 	int i;
1769 
1770 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1771 				  sizeof(struct hlist_head),
1772 				  GFP_KERNEL);
1773 	if (!dp->ports)
1774 		return -ENOMEM;
1775 
1776 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1777 		INIT_HLIST_HEAD(&dp->ports[i]);
1778 
1779 	return 0;
1780 }
1781 
1782 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1783 {
1784 	struct nlattr **a = info->attrs;
1785 	struct vport_parms parms;
1786 	struct sk_buff *reply;
1787 	struct datapath *dp;
1788 	struct vport *vport;
1789 	struct ovs_net *ovs_net;
1790 	int err;
1791 
1792 	err = -EINVAL;
1793 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1794 		goto err;
1795 
1796 	reply = ovs_dp_cmd_alloc_info();
1797 	if (!reply)
1798 		return -ENOMEM;
1799 
1800 	err = -ENOMEM;
1801 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1802 	if (dp == NULL)
1803 		goto err_destroy_reply;
1804 
1805 	ovs_dp_set_net(dp, sock_net(skb->sk));
1806 
1807 	/* Allocate table. */
1808 	err = ovs_flow_tbl_init(&dp->table);
1809 	if (err)
1810 		goto err_destroy_dp;
1811 
1812 	err = ovs_dp_stats_init(dp);
1813 	if (err)
1814 		goto err_destroy_table;
1815 
1816 	err = ovs_dp_vport_init(dp);
1817 	if (err)
1818 		goto err_destroy_stats;
1819 
1820 	err = ovs_meters_init(dp);
1821 	if (err)
1822 		goto err_destroy_ports;
1823 
1824 	/* Set up our datapath device. */
1825 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1826 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1827 	parms.options = NULL;
1828 	parms.dp = dp;
1829 	parms.port_no = OVSP_LOCAL;
1830 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1831 	parms.desired_ifindex = a[OVS_DP_ATTR_IFINDEX]
1832 		? nla_get_u32(a[OVS_DP_ATTR_IFINDEX]) : 0;
1833 
1834 	/* So far only local changes have been made, now need the lock. */
1835 	ovs_lock();
1836 
1837 	err = ovs_dp_change(dp, a);
1838 	if (err)
1839 		goto err_unlock_and_destroy_meters;
1840 
1841 	vport = new_vport(&parms);
1842 	if (IS_ERR(vport)) {
1843 		err = PTR_ERR(vport);
1844 		if (err == -EBUSY)
1845 			err = -EEXIST;
1846 
1847 		if (err == -EEXIST) {
1848 			/* An outdated user space instance that does not understand
1849 			 * the concept of user_features has attempted to create a new
1850 			 * datapath and is likely to reuse it. Drop all user features.
1851 			 */
1852 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1853 				ovs_dp_reset_user_features(skb, info);
1854 		}
1855 
1856 		goto err_destroy_portids;
1857 	}
1858 
1859 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1860 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1861 	BUG_ON(err < 0);
1862 
1863 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1864 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1865 
1866 	ovs_unlock();
1867 
1868 	ovs_notify(&dp_datapath_genl_family, reply, info);
1869 	return 0;
1870 
1871 err_destroy_portids:
1872 	kfree(rcu_dereference_raw(dp->upcall_portids));
1873 err_unlock_and_destroy_meters:
1874 	ovs_unlock();
1875 	ovs_meters_exit(dp);
1876 err_destroy_ports:
1877 	kfree(dp->ports);
1878 err_destroy_stats:
1879 	free_percpu(dp->stats_percpu);
1880 err_destroy_table:
1881 	ovs_flow_tbl_destroy(&dp->table);
1882 err_destroy_dp:
1883 	kfree(dp);
1884 err_destroy_reply:
1885 	kfree_skb(reply);
1886 err:
1887 	return err;
1888 }
1889 
1890 /* Called with ovs_mutex. */
1891 static void __dp_destroy(struct datapath *dp)
1892 {
1893 	struct flow_table *table = &dp->table;
1894 	int i;
1895 
1896 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1897 		tc_skb_ext_tc_disable();
1898 
1899 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1900 		struct vport *vport;
1901 		struct hlist_node *n;
1902 
1903 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1904 			if (vport->port_no != OVSP_LOCAL)
1905 				ovs_dp_detach_port(vport);
1906 	}
1907 
1908 	list_del_rcu(&dp->list_node);
1909 
1910 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1911 	 * all ports in datapath are destroyed first before freeing datapath.
1912 	 */
1913 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1914 
1915 	/* Flush sw_flow in the tables. RCU cb only releases resource
1916 	 * such as dp, ports and tables. That may avoid some issues
1917 	 * such as RCU usage warning.
1918 	 */
1919 	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1920 				  ovsl_dereference(table->ufid_ti));
1921 
1922 	/* RCU destroy the ports, meters and flow tables. */
1923 	call_rcu(&dp->rcu, destroy_dp_rcu);
1924 }
1925 
1926 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1927 {
1928 	struct sk_buff *reply;
1929 	struct datapath *dp;
1930 	int err;
1931 
1932 	reply = ovs_dp_cmd_alloc_info();
1933 	if (!reply)
1934 		return -ENOMEM;
1935 
1936 	ovs_lock();
1937 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1938 	err = PTR_ERR(dp);
1939 	if (IS_ERR(dp))
1940 		goto err_unlock_free;
1941 
1942 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1943 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1944 	BUG_ON(err < 0);
1945 
1946 	__dp_destroy(dp);
1947 	ovs_unlock();
1948 
1949 	ovs_notify(&dp_datapath_genl_family, reply, info);
1950 
1951 	return 0;
1952 
1953 err_unlock_free:
1954 	ovs_unlock();
1955 	kfree_skb(reply);
1956 	return err;
1957 }
1958 
1959 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1960 {
1961 	struct sk_buff *reply;
1962 	struct datapath *dp;
1963 	int err;
1964 
1965 	reply = ovs_dp_cmd_alloc_info();
1966 	if (!reply)
1967 		return -ENOMEM;
1968 
1969 	ovs_lock();
1970 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1971 	err = PTR_ERR(dp);
1972 	if (IS_ERR(dp))
1973 		goto err_unlock_free;
1974 
1975 	err = ovs_dp_change(dp, info->attrs);
1976 	if (err)
1977 		goto err_unlock_free;
1978 
1979 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1980 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1981 	BUG_ON(err < 0);
1982 
1983 	ovs_unlock();
1984 	ovs_notify(&dp_datapath_genl_family, reply, info);
1985 
1986 	return 0;
1987 
1988 err_unlock_free:
1989 	ovs_unlock();
1990 	kfree_skb(reply);
1991 	return err;
1992 }
1993 
1994 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1995 {
1996 	struct sk_buff *reply;
1997 	struct datapath *dp;
1998 	int err;
1999 
2000 	reply = ovs_dp_cmd_alloc_info();
2001 	if (!reply)
2002 		return -ENOMEM;
2003 
2004 	ovs_lock();
2005 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
2006 	if (IS_ERR(dp)) {
2007 		err = PTR_ERR(dp);
2008 		goto err_unlock_free;
2009 	}
2010 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2011 				   info->snd_seq, 0, OVS_DP_CMD_GET);
2012 	BUG_ON(err < 0);
2013 	ovs_unlock();
2014 
2015 	return genlmsg_reply(reply, info);
2016 
2017 err_unlock_free:
2018 	ovs_unlock();
2019 	kfree_skb(reply);
2020 	return err;
2021 }
2022 
2023 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2024 {
2025 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2026 	struct datapath *dp;
2027 	int skip = cb->args[0];
2028 	int i = 0;
2029 
2030 	ovs_lock();
2031 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2032 		if (i >= skip &&
2033 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2034 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2035 					 OVS_DP_CMD_GET) < 0)
2036 			break;
2037 		i++;
2038 	}
2039 	ovs_unlock();
2040 
2041 	cb->args[0] = i;
2042 
2043 	return skb->len;
2044 }
2045 
2046 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2047 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2048 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2049 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2050 	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
2051 		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2052 	[OVS_DP_ATTR_IFINDEX] = {.type = NLA_U32 },
2053 };
2054 
2055 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2056 	{ .cmd = OVS_DP_CMD_NEW,
2057 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2058 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2059 	  .doit = ovs_dp_cmd_new
2060 	},
2061 	{ .cmd = OVS_DP_CMD_DEL,
2062 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2063 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2064 	  .doit = ovs_dp_cmd_del
2065 	},
2066 	{ .cmd = OVS_DP_CMD_GET,
2067 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2068 	  .flags = 0,		    /* OK for unprivileged users. */
2069 	  .doit = ovs_dp_cmd_get,
2070 	  .dumpit = ovs_dp_cmd_dump
2071 	},
2072 	{ .cmd = OVS_DP_CMD_SET,
2073 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2074 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2075 	  .doit = ovs_dp_cmd_set,
2076 	},
2077 };
2078 
2079 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2080 	.hdrsize = sizeof(struct ovs_header),
2081 	.name = OVS_DATAPATH_FAMILY,
2082 	.version = OVS_DATAPATH_VERSION,
2083 	.maxattr = OVS_DP_ATTR_MAX,
2084 	.policy = datapath_policy,
2085 	.netnsok = true,
2086 	.parallel_ops = true,
2087 	.small_ops = dp_datapath_genl_ops,
2088 	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2089 	.resv_start_op = OVS_DP_CMD_SET + 1,
2090 	.mcgrps = &ovs_dp_datapath_multicast_group,
2091 	.n_mcgrps = 1,
2092 	.module = THIS_MODULE,
2093 };
2094 
2095 /* Called with ovs_mutex or RCU read lock. */
2096 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2097 				   struct net *net, u32 portid, u32 seq,
2098 				   u32 flags, u8 cmd, gfp_t gfp)
2099 {
2100 	struct ovs_header *ovs_header;
2101 	struct ovs_vport_stats vport_stats;
2102 	int err;
2103 
2104 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2105 				 flags, cmd);
2106 	if (!ovs_header)
2107 		return -EMSGSIZE;
2108 
2109 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2110 
2111 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2112 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2113 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2114 			   ovs_vport_name(vport)) ||
2115 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2116 		goto nla_put_failure;
2117 
2118 	if (!net_eq(net, dev_net(vport->dev))) {
2119 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2120 
2121 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2122 			goto nla_put_failure;
2123 	}
2124 
2125 	ovs_vport_get_stats(vport, &vport_stats);
2126 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2127 			  sizeof(struct ovs_vport_stats), &vport_stats,
2128 			  OVS_VPORT_ATTR_PAD))
2129 		goto nla_put_failure;
2130 
2131 	if (ovs_vport_get_upcall_stats(vport, skb))
2132 		goto nla_put_failure;
2133 
2134 	if (ovs_vport_get_upcall_portids(vport, skb))
2135 		goto nla_put_failure;
2136 
2137 	err = ovs_vport_get_options(vport, skb);
2138 	if (err == -EMSGSIZE)
2139 		goto error;
2140 
2141 	genlmsg_end(skb, ovs_header);
2142 	return 0;
2143 
2144 nla_put_failure:
2145 	err = -EMSGSIZE;
2146 error:
2147 	genlmsg_cancel(skb, ovs_header);
2148 	return err;
2149 }
2150 
2151 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2152 {
2153 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2154 }
2155 
2156 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2157 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2158 					 u32 portid, u32 seq, u8 cmd)
2159 {
2160 	struct sk_buff *skb;
2161 	int retval;
2162 
2163 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2164 	if (!skb)
2165 		return ERR_PTR(-ENOMEM);
2166 
2167 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2168 					 GFP_KERNEL);
2169 	BUG_ON(retval < 0);
2170 
2171 	return skb;
2172 }
2173 
2174 /* Called with ovs_mutex or RCU read lock. */
2175 static struct vport *lookup_vport(struct net *net,
2176 				  const struct ovs_header *ovs_header,
2177 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2178 {
2179 	struct datapath *dp;
2180 	struct vport *vport;
2181 
2182 	if (a[OVS_VPORT_ATTR_IFINDEX])
2183 		return ERR_PTR(-EOPNOTSUPP);
2184 	if (a[OVS_VPORT_ATTR_NAME]) {
2185 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2186 		if (!vport)
2187 			return ERR_PTR(-ENODEV);
2188 		if (ovs_header->dp_ifindex &&
2189 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2190 			return ERR_PTR(-ENODEV);
2191 		return vport;
2192 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2193 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2194 
2195 		if (port_no >= DP_MAX_PORTS)
2196 			return ERR_PTR(-EFBIG);
2197 
2198 		dp = get_dp(net, ovs_header->dp_ifindex);
2199 		if (!dp)
2200 			return ERR_PTR(-ENODEV);
2201 
2202 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2203 		if (!vport)
2204 			return ERR_PTR(-ENODEV);
2205 		return vport;
2206 	} else
2207 		return ERR_PTR(-EINVAL);
2208 
2209 }
2210 
2211 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2212 {
2213 	unsigned int dev_headroom, max_headroom = 0;
2214 	struct net_device *dev;
2215 	struct vport *vport;
2216 	int i;
2217 
2218 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2219 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2220 					 lockdep_ovsl_is_held()) {
2221 			dev = vport->dev;
2222 			dev_headroom = netdev_get_fwd_headroom(dev);
2223 			if (dev_headroom > max_headroom)
2224 				max_headroom = dev_headroom;
2225 		}
2226 	}
2227 
2228 	return max_headroom;
2229 }
2230 
2231 /* Called with ovs_mutex */
2232 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2233 {
2234 	struct vport *vport;
2235 	int i;
2236 
2237 	dp->max_headroom = new_headroom;
2238 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2239 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2240 					 lockdep_ovsl_is_held())
2241 			netdev_set_rx_headroom(vport->dev, new_headroom);
2242 	}
2243 }
2244 
2245 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2246 {
2247 	struct nlattr **a = info->attrs;
2248 	struct ovs_header *ovs_header = info->userhdr;
2249 	struct vport_parms parms;
2250 	struct sk_buff *reply;
2251 	struct vport *vport;
2252 	struct datapath *dp;
2253 	unsigned int new_headroom;
2254 	u32 port_no;
2255 	int err;
2256 
2257 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2258 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2259 		return -EINVAL;
2260 
2261 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2262 
2263 	if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2264 		return -EOPNOTSUPP;
2265 
2266 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2267 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2268 	if (port_no >= DP_MAX_PORTS)
2269 		return -EFBIG;
2270 
2271 	reply = ovs_vport_cmd_alloc_info();
2272 	if (!reply)
2273 		return -ENOMEM;
2274 
2275 	ovs_lock();
2276 restart:
2277 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2278 	err = -ENODEV;
2279 	if (!dp)
2280 		goto exit_unlock_free;
2281 
2282 	if (port_no) {
2283 		vport = ovs_vport_ovsl(dp, port_no);
2284 		err = -EBUSY;
2285 		if (vport)
2286 			goto exit_unlock_free;
2287 	} else {
2288 		for (port_no = 1; ; port_no++) {
2289 			if (port_no >= DP_MAX_PORTS) {
2290 				err = -EFBIG;
2291 				goto exit_unlock_free;
2292 			}
2293 			vport = ovs_vport_ovsl(dp, port_no);
2294 			if (!vport)
2295 				break;
2296 		}
2297 	}
2298 
2299 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2300 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2301 	parms.dp = dp;
2302 	parms.port_no = port_no;
2303 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2304 	parms.desired_ifindex = a[OVS_VPORT_ATTR_IFINDEX]
2305 		? nla_get_u32(a[OVS_VPORT_ATTR_IFINDEX]) : 0;
2306 
2307 	vport = new_vport(&parms);
2308 	err = PTR_ERR(vport);
2309 	if (IS_ERR(vport)) {
2310 		if (err == -EAGAIN)
2311 			goto restart;
2312 		goto exit_unlock_free;
2313 	}
2314 
2315 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2316 				      info->snd_portid, info->snd_seq, 0,
2317 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2318 
2319 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2320 
2321 	if (new_headroom > dp->max_headroom)
2322 		ovs_update_headroom(dp, new_headroom);
2323 	else
2324 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2325 
2326 	BUG_ON(err < 0);
2327 	ovs_unlock();
2328 
2329 	ovs_notify(&dp_vport_genl_family, reply, info);
2330 	return 0;
2331 
2332 exit_unlock_free:
2333 	ovs_unlock();
2334 	kfree_skb(reply);
2335 	return err;
2336 }
2337 
2338 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2339 {
2340 	struct nlattr **a = info->attrs;
2341 	struct sk_buff *reply;
2342 	struct vport *vport;
2343 	int err;
2344 
2345 	reply = ovs_vport_cmd_alloc_info();
2346 	if (!reply)
2347 		return -ENOMEM;
2348 
2349 	ovs_lock();
2350 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2351 	err = PTR_ERR(vport);
2352 	if (IS_ERR(vport))
2353 		goto exit_unlock_free;
2354 
2355 	if (a[OVS_VPORT_ATTR_TYPE] &&
2356 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2357 		err = -EINVAL;
2358 		goto exit_unlock_free;
2359 	}
2360 
2361 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2362 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2363 		if (err)
2364 			goto exit_unlock_free;
2365 	}
2366 
2367 
2368 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2369 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2370 
2371 		err = ovs_vport_set_upcall_portids(vport, ids);
2372 		if (err)
2373 			goto exit_unlock_free;
2374 	}
2375 
2376 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2377 				      info->snd_portid, info->snd_seq, 0,
2378 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2379 	BUG_ON(err < 0);
2380 
2381 	ovs_unlock();
2382 	ovs_notify(&dp_vport_genl_family, reply, info);
2383 	return 0;
2384 
2385 exit_unlock_free:
2386 	ovs_unlock();
2387 	kfree_skb(reply);
2388 	return err;
2389 }
2390 
2391 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2392 {
2393 	bool update_headroom = false;
2394 	struct nlattr **a = info->attrs;
2395 	struct sk_buff *reply;
2396 	struct datapath *dp;
2397 	struct vport *vport;
2398 	unsigned int new_headroom;
2399 	int err;
2400 
2401 	reply = ovs_vport_cmd_alloc_info();
2402 	if (!reply)
2403 		return -ENOMEM;
2404 
2405 	ovs_lock();
2406 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2407 	err = PTR_ERR(vport);
2408 	if (IS_ERR(vport))
2409 		goto exit_unlock_free;
2410 
2411 	if (vport->port_no == OVSP_LOCAL) {
2412 		err = -EINVAL;
2413 		goto exit_unlock_free;
2414 	}
2415 
2416 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2417 				      info->snd_portid, info->snd_seq, 0,
2418 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2419 	BUG_ON(err < 0);
2420 
2421 	/* the vport deletion may trigger dp headroom update */
2422 	dp = vport->dp;
2423 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2424 		update_headroom = true;
2425 
2426 	netdev_reset_rx_headroom(vport->dev);
2427 	ovs_dp_detach_port(vport);
2428 
2429 	if (update_headroom) {
2430 		new_headroom = ovs_get_max_headroom(dp);
2431 
2432 		if (new_headroom < dp->max_headroom)
2433 			ovs_update_headroom(dp, new_headroom);
2434 	}
2435 	ovs_unlock();
2436 
2437 	ovs_notify(&dp_vport_genl_family, reply, info);
2438 	return 0;
2439 
2440 exit_unlock_free:
2441 	ovs_unlock();
2442 	kfree_skb(reply);
2443 	return err;
2444 }
2445 
2446 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2447 {
2448 	struct nlattr **a = info->attrs;
2449 	struct ovs_header *ovs_header = info->userhdr;
2450 	struct sk_buff *reply;
2451 	struct vport *vport;
2452 	int err;
2453 
2454 	reply = ovs_vport_cmd_alloc_info();
2455 	if (!reply)
2456 		return -ENOMEM;
2457 
2458 	rcu_read_lock();
2459 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2460 	err = PTR_ERR(vport);
2461 	if (IS_ERR(vport))
2462 		goto exit_unlock_free;
2463 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2464 				      info->snd_portid, info->snd_seq, 0,
2465 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2466 	BUG_ON(err < 0);
2467 	rcu_read_unlock();
2468 
2469 	return genlmsg_reply(reply, info);
2470 
2471 exit_unlock_free:
2472 	rcu_read_unlock();
2473 	kfree_skb(reply);
2474 	return err;
2475 }
2476 
2477 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2478 {
2479 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2480 	struct datapath *dp;
2481 	int bucket = cb->args[0], skip = cb->args[1];
2482 	int i, j = 0;
2483 
2484 	rcu_read_lock();
2485 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2486 	if (!dp) {
2487 		rcu_read_unlock();
2488 		return -ENODEV;
2489 	}
2490 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2491 		struct vport *vport;
2492 
2493 		j = 0;
2494 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2495 			if (j >= skip &&
2496 			    ovs_vport_cmd_fill_info(vport, skb,
2497 						    sock_net(skb->sk),
2498 						    NETLINK_CB(cb->skb).portid,
2499 						    cb->nlh->nlmsg_seq,
2500 						    NLM_F_MULTI,
2501 						    OVS_VPORT_CMD_GET,
2502 						    GFP_ATOMIC) < 0)
2503 				goto out;
2504 
2505 			j++;
2506 		}
2507 		skip = 0;
2508 	}
2509 out:
2510 	rcu_read_unlock();
2511 
2512 	cb->args[0] = i;
2513 	cb->args[1] = j;
2514 
2515 	return skb->len;
2516 }
2517 
2518 static void ovs_dp_masks_rebalance(struct work_struct *work)
2519 {
2520 	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2521 					       masks_rebalance.work);
2522 	struct datapath *dp;
2523 
2524 	ovs_lock();
2525 
2526 	list_for_each_entry(dp, &ovs_net->dps, list_node)
2527 		ovs_flow_masks_rebalance(&dp->table);
2528 
2529 	ovs_unlock();
2530 
2531 	schedule_delayed_work(&ovs_net->masks_rebalance,
2532 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2533 }
2534 
2535 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2536 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2537 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2538 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2539 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2540 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2541 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2542 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2543 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2544 	[OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
2545 };
2546 
2547 static const struct genl_small_ops dp_vport_genl_ops[] = {
2548 	{ .cmd = OVS_VPORT_CMD_NEW,
2549 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2550 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2551 	  .doit = ovs_vport_cmd_new
2552 	},
2553 	{ .cmd = OVS_VPORT_CMD_DEL,
2554 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2555 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2556 	  .doit = ovs_vport_cmd_del
2557 	},
2558 	{ .cmd = OVS_VPORT_CMD_GET,
2559 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2560 	  .flags = 0,		    /* OK for unprivileged users. */
2561 	  .doit = ovs_vport_cmd_get,
2562 	  .dumpit = ovs_vport_cmd_dump
2563 	},
2564 	{ .cmd = OVS_VPORT_CMD_SET,
2565 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2566 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2567 	  .doit = ovs_vport_cmd_set,
2568 	},
2569 };
2570 
2571 struct genl_family dp_vport_genl_family __ro_after_init = {
2572 	.hdrsize = sizeof(struct ovs_header),
2573 	.name = OVS_VPORT_FAMILY,
2574 	.version = OVS_VPORT_VERSION,
2575 	.maxattr = OVS_VPORT_ATTR_MAX,
2576 	.policy = vport_policy,
2577 	.netnsok = true,
2578 	.parallel_ops = true,
2579 	.small_ops = dp_vport_genl_ops,
2580 	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2581 	.resv_start_op = OVS_VPORT_CMD_SET + 1,
2582 	.mcgrps = &ovs_dp_vport_multicast_group,
2583 	.n_mcgrps = 1,
2584 	.module = THIS_MODULE,
2585 };
2586 
2587 static struct genl_family * const dp_genl_families[] = {
2588 	&dp_datapath_genl_family,
2589 	&dp_vport_genl_family,
2590 	&dp_flow_genl_family,
2591 	&dp_packet_genl_family,
2592 	&dp_meter_genl_family,
2593 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2594 	&dp_ct_limit_genl_family,
2595 #endif
2596 };
2597 
2598 static void dp_unregister_genl(int n_families)
2599 {
2600 	int i;
2601 
2602 	for (i = 0; i < n_families; i++)
2603 		genl_unregister_family(dp_genl_families[i]);
2604 }
2605 
2606 static int __init dp_register_genl(void)
2607 {
2608 	int err;
2609 	int i;
2610 
2611 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2612 
2613 		err = genl_register_family(dp_genl_families[i]);
2614 		if (err)
2615 			goto error;
2616 	}
2617 
2618 	return 0;
2619 
2620 error:
2621 	dp_unregister_genl(i);
2622 	return err;
2623 }
2624 
2625 static int __net_init ovs_init_net(struct net *net)
2626 {
2627 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2628 	int err;
2629 
2630 	INIT_LIST_HEAD(&ovs_net->dps);
2631 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2632 	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2633 
2634 	err = ovs_ct_init(net);
2635 	if (err)
2636 		return err;
2637 
2638 	schedule_delayed_work(&ovs_net->masks_rebalance,
2639 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2640 	return 0;
2641 }
2642 
2643 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2644 					    struct list_head *head)
2645 {
2646 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2647 	struct datapath *dp;
2648 
2649 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2650 		int i;
2651 
2652 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2653 			struct vport *vport;
2654 
2655 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2656 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2657 					continue;
2658 
2659 				if (dev_net(vport->dev) == dnet)
2660 					list_add(&vport->detach_list, head);
2661 			}
2662 		}
2663 	}
2664 }
2665 
2666 static void __net_exit ovs_exit_net(struct net *dnet)
2667 {
2668 	struct datapath *dp, *dp_next;
2669 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2670 	struct vport *vport, *vport_next;
2671 	struct net *net;
2672 	LIST_HEAD(head);
2673 
2674 	ovs_lock();
2675 
2676 	ovs_ct_exit(dnet);
2677 
2678 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2679 		__dp_destroy(dp);
2680 
2681 	down_read(&net_rwsem);
2682 	for_each_net(net)
2683 		list_vports_from_net(net, dnet, &head);
2684 	up_read(&net_rwsem);
2685 
2686 	/* Detach all vports from given namespace. */
2687 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2688 		list_del(&vport->detach_list);
2689 		ovs_dp_detach_port(vport);
2690 	}
2691 
2692 	ovs_unlock();
2693 
2694 	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2695 	cancel_work_sync(&ovs_net->dp_notify_work);
2696 }
2697 
2698 static struct pernet_operations ovs_net_ops = {
2699 	.init = ovs_init_net,
2700 	.exit = ovs_exit_net,
2701 	.id   = &ovs_net_id,
2702 	.size = sizeof(struct ovs_net),
2703 };
2704 
2705 static int __init dp_init(void)
2706 {
2707 	int err;
2708 
2709 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2710 		     sizeof_field(struct sk_buff, cb));
2711 
2712 	pr_info("Open vSwitch switching datapath\n");
2713 
2714 	err = action_fifos_init();
2715 	if (err)
2716 		goto error;
2717 
2718 	err = ovs_internal_dev_rtnl_link_register();
2719 	if (err)
2720 		goto error_action_fifos_exit;
2721 
2722 	err = ovs_flow_init();
2723 	if (err)
2724 		goto error_unreg_rtnl_link;
2725 
2726 	err = ovs_vport_init();
2727 	if (err)
2728 		goto error_flow_exit;
2729 
2730 	err = register_pernet_device(&ovs_net_ops);
2731 	if (err)
2732 		goto error_vport_exit;
2733 
2734 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2735 	if (err)
2736 		goto error_netns_exit;
2737 
2738 	err = ovs_netdev_init();
2739 	if (err)
2740 		goto error_unreg_notifier;
2741 
2742 	err = dp_register_genl();
2743 	if (err < 0)
2744 		goto error_unreg_netdev;
2745 
2746 	return 0;
2747 
2748 error_unreg_netdev:
2749 	ovs_netdev_exit();
2750 error_unreg_notifier:
2751 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2752 error_netns_exit:
2753 	unregister_pernet_device(&ovs_net_ops);
2754 error_vport_exit:
2755 	ovs_vport_exit();
2756 error_flow_exit:
2757 	ovs_flow_exit();
2758 error_unreg_rtnl_link:
2759 	ovs_internal_dev_rtnl_link_unregister();
2760 error_action_fifos_exit:
2761 	action_fifos_exit();
2762 error:
2763 	return err;
2764 }
2765 
2766 static void dp_cleanup(void)
2767 {
2768 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2769 	ovs_netdev_exit();
2770 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2771 	unregister_pernet_device(&ovs_net_ops);
2772 	rcu_barrier();
2773 	ovs_vport_exit();
2774 	ovs_flow_exit();
2775 	ovs_internal_dev_rtnl_link_unregister();
2776 	action_fifos_exit();
2777 }
2778 
2779 module_init(dp_init);
2780 module_exit(dp_cleanup);
2781 
2782 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2783 MODULE_LICENSE("GPL");
2784 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2785 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2786 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2787 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2788 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2789 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2790