xref: /openbmc/linux/net/openvswitch/datapath.c (revision f59a3ee6)
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 #include <net/pkt_cls.h>
41 
42 #include "datapath.h"
43 #include "flow.h"
44 #include "flow_table.h"
45 #include "flow_netlink.h"
46 #include "meter.h"
47 #include "openvswitch_trace.h"
48 #include "vport-internal_dev.h"
49 #include "vport-netdev.h"
50 
51 unsigned int ovs_net_id __read_mostly;
52 
53 static struct genl_family dp_packet_genl_family;
54 static struct genl_family dp_flow_genl_family;
55 static struct genl_family dp_datapath_genl_family;
56 
57 static const struct nla_policy flow_policy[];
58 
59 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
60 	.name = OVS_FLOW_MCGROUP,
61 };
62 
63 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
64 	.name = OVS_DATAPATH_MCGROUP,
65 };
66 
67 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
68 	.name = OVS_VPORT_MCGROUP,
69 };
70 
71 /* Check if need to build a reply message.
72  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
73 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
74 			    unsigned int group)
75 {
76 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
77 	       genl_has_listeners(family, genl_info_net(info), group);
78 }
79 
80 static void ovs_notify(struct genl_family *family,
81 		       struct sk_buff *skb, struct genl_info *info)
82 {
83 	genl_notify(family, skb, info, 0, GFP_KERNEL);
84 }
85 
86 /**
87  * DOC: Locking:
88  *
89  * All writes e.g. Writes to device state (add/remove datapath, port, set
90  * operations on vports, etc.), Writes to other state (flow table
91  * modifications, set miscellaneous datapath parameters, etc.) are protected
92  * by ovs_lock.
93  *
94  * Reads are protected by RCU.
95  *
96  * There are a few special cases (mostly stats) that have their own
97  * synchronization but they nest under all of above and don't interact with
98  * each other.
99  *
100  * The RTNL lock nests inside ovs_mutex.
101  */
102 
103 static DEFINE_MUTEX(ovs_mutex);
104 
105 void ovs_lock(void)
106 {
107 	mutex_lock(&ovs_mutex);
108 }
109 
110 void ovs_unlock(void)
111 {
112 	mutex_unlock(&ovs_mutex);
113 }
114 
115 #ifdef CONFIG_LOCKDEP
116 int lockdep_ovsl_is_held(void)
117 {
118 	if (debug_locks)
119 		return lockdep_is_held(&ovs_mutex);
120 	else
121 		return 1;
122 }
123 #endif
124 
125 static struct vport *new_vport(const struct vport_parms *);
126 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
127 			     const struct sw_flow_key *,
128 			     const struct dp_upcall_info *,
129 			     uint32_t cutlen);
130 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
131 				  const struct sw_flow_key *,
132 				  const struct dp_upcall_info *,
133 				  uint32_t cutlen);
134 
135 static void ovs_dp_masks_rebalance(struct work_struct *work);
136 
137 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
138 
139 /* Must be called with rcu_read_lock or ovs_mutex. */
140 const char *ovs_dp_name(const struct datapath *dp)
141 {
142 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
143 	return ovs_vport_name(vport);
144 }
145 
146 static int get_dpifindex(const struct datapath *dp)
147 {
148 	struct vport *local;
149 	int ifindex;
150 
151 	rcu_read_lock();
152 
153 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
154 	if (local)
155 		ifindex = local->dev->ifindex;
156 	else
157 		ifindex = 0;
158 
159 	rcu_read_unlock();
160 
161 	return ifindex;
162 }
163 
164 static void destroy_dp_rcu(struct rcu_head *rcu)
165 {
166 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
167 
168 	ovs_flow_tbl_destroy(&dp->table);
169 	free_percpu(dp->stats_percpu);
170 	kfree(dp->ports);
171 	ovs_meters_exit(dp);
172 	kfree(rcu_dereference_raw(dp->upcall_portids));
173 	kfree(dp);
174 }
175 
176 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
177 					    u16 port_no)
178 {
179 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
180 }
181 
182 /* Called with ovs_mutex or RCU read lock. */
183 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
184 {
185 	struct vport *vport;
186 	struct hlist_head *head;
187 
188 	head = vport_hash_bucket(dp, port_no);
189 	hlist_for_each_entry_rcu(vport, head, dp_hash_node,
190 				 lockdep_ovsl_is_held()) {
191 		if (vport->port_no == port_no)
192 			return vport;
193 	}
194 	return NULL;
195 }
196 
197 /* Called with ovs_mutex. */
198 static struct vport *new_vport(const struct vport_parms *parms)
199 {
200 	struct vport *vport;
201 
202 	vport = ovs_vport_add(parms);
203 	if (!IS_ERR(vport)) {
204 		struct datapath *dp = parms->dp;
205 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
206 
207 		hlist_add_head_rcu(&vport->dp_hash_node, head);
208 	}
209 	return vport;
210 }
211 
212 void ovs_dp_detach_port(struct vport *p)
213 {
214 	ASSERT_OVSL();
215 
216 	/* First drop references to device. */
217 	hlist_del_rcu(&p->dp_hash_node);
218 
219 	/* Then destroy it. */
220 	ovs_vport_del(p);
221 }
222 
223 /* Must be called with rcu_read_lock. */
224 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
225 {
226 	const struct vport *p = OVS_CB(skb)->input_vport;
227 	struct datapath *dp = p->dp;
228 	struct sw_flow *flow;
229 	struct sw_flow_actions *sf_acts;
230 	struct dp_stats_percpu *stats;
231 	u64 *stats_counter;
232 	u32 n_mask_hit;
233 	u32 n_cache_hit;
234 	int error;
235 
236 	stats = this_cpu_ptr(dp->stats_percpu);
237 
238 	/* Look up flow. */
239 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
240 					 &n_mask_hit, &n_cache_hit);
241 	if (unlikely(!flow)) {
242 		struct dp_upcall_info upcall;
243 
244 		memset(&upcall, 0, sizeof(upcall));
245 		upcall.cmd = OVS_PACKET_CMD_MISS;
246 
247 		if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
248 			upcall.portid =
249 			    ovs_dp_get_upcall_portid(dp, smp_processor_id());
250 		else
251 			upcall.portid = ovs_vport_find_upcall_portid(p, skb);
252 
253 		upcall.mru = OVS_CB(skb)->mru;
254 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
255 		switch (error) {
256 		case 0:
257 		case -EAGAIN:
258 		case -ERESTARTSYS:
259 		case -EINTR:
260 			consume_skb(skb);
261 			break;
262 		default:
263 			kfree_skb(skb);
264 			break;
265 		}
266 		stats_counter = &stats->n_missed;
267 		goto out;
268 	}
269 
270 	ovs_flow_stats_update(flow, key->tp.flags, skb);
271 	sf_acts = rcu_dereference(flow->sf_acts);
272 	error = ovs_execute_actions(dp, skb, sf_acts, key);
273 	if (unlikely(error))
274 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
275 				    ovs_dp_name(dp), error);
276 
277 	stats_counter = &stats->n_hit;
278 
279 out:
280 	/* Update datapath statistics. */
281 	u64_stats_update_begin(&stats->syncp);
282 	(*stats_counter)++;
283 	stats->n_mask_hit += n_mask_hit;
284 	stats->n_cache_hit += n_cache_hit;
285 	u64_stats_update_end(&stats->syncp);
286 }
287 
288 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
289 		  const struct sw_flow_key *key,
290 		  const struct dp_upcall_info *upcall_info,
291 		  uint32_t cutlen)
292 {
293 	struct dp_stats_percpu *stats;
294 	int err;
295 
296 	if (trace_ovs_dp_upcall_enabled())
297 		trace_ovs_dp_upcall(dp, skb, key, upcall_info);
298 
299 	if (upcall_info->portid == 0) {
300 		err = -ENOTCONN;
301 		goto err;
302 	}
303 
304 	if (!skb_is_gso(skb))
305 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
306 	else
307 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
308 	if (err)
309 		goto err;
310 
311 	return 0;
312 
313 err:
314 	stats = this_cpu_ptr(dp->stats_percpu);
315 
316 	u64_stats_update_begin(&stats->syncp);
317 	stats->n_lost++;
318 	u64_stats_update_end(&stats->syncp);
319 
320 	return err;
321 }
322 
323 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
324 			     const struct sw_flow_key *key,
325 			     const struct dp_upcall_info *upcall_info,
326 			     uint32_t cutlen)
327 {
328 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
329 	struct sw_flow_key later_key;
330 	struct sk_buff *segs, *nskb;
331 	int err;
332 
333 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
334 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
335 	if (IS_ERR(segs))
336 		return PTR_ERR(segs);
337 	if (segs == NULL)
338 		return -EINVAL;
339 
340 	if (gso_type & SKB_GSO_UDP) {
341 		/* The initial flow key extracted by ovs_flow_key_extract()
342 		 * in this case is for a first fragment, so we need to
343 		 * properly mark later fragments.
344 		 */
345 		later_key = *key;
346 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
347 	}
348 
349 	/* Queue all of the segments. */
350 	skb_list_walk_safe(segs, skb, nskb) {
351 		if (gso_type & SKB_GSO_UDP && skb != segs)
352 			key = &later_key;
353 
354 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
355 		if (err)
356 			break;
357 
358 	}
359 
360 	/* Free all of the segments. */
361 	skb_list_walk_safe(segs, skb, nskb) {
362 		if (err)
363 			kfree_skb(skb);
364 		else
365 			consume_skb(skb);
366 	}
367 	return err;
368 }
369 
370 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
371 			      unsigned int hdrlen, int actions_attrlen)
372 {
373 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
374 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
375 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
376 		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
377 		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
378 
379 	/* OVS_PACKET_ATTR_USERDATA */
380 	if (upcall_info->userdata)
381 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
382 
383 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
384 	if (upcall_info->egress_tun_info)
385 		size += nla_total_size(ovs_tun_key_attr_size());
386 
387 	/* OVS_PACKET_ATTR_ACTIONS */
388 	if (upcall_info->actions_len)
389 		size += nla_total_size(actions_attrlen);
390 
391 	/* OVS_PACKET_ATTR_MRU */
392 	if (upcall_info->mru)
393 		size += nla_total_size(sizeof(upcall_info->mru));
394 
395 	return size;
396 }
397 
398 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
399 {
400 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
401 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
402 
403 		if (plen > 0)
404 			skb_put_zero(skb, plen);
405 	}
406 }
407 
408 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
409 				  const struct sw_flow_key *key,
410 				  const struct dp_upcall_info *upcall_info,
411 				  uint32_t cutlen)
412 {
413 	struct ovs_header *upcall;
414 	struct sk_buff *nskb = NULL;
415 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
416 	struct nlattr *nla;
417 	size_t len;
418 	unsigned int hlen;
419 	int err, dp_ifindex;
420 	u64 hash;
421 
422 	dp_ifindex = get_dpifindex(dp);
423 	if (!dp_ifindex)
424 		return -ENODEV;
425 
426 	if (skb_vlan_tag_present(skb)) {
427 		nskb = skb_clone(skb, GFP_ATOMIC);
428 		if (!nskb)
429 			return -ENOMEM;
430 
431 		nskb = __vlan_hwaccel_push_inside(nskb);
432 		if (!nskb)
433 			return -ENOMEM;
434 
435 		skb = nskb;
436 	}
437 
438 	if (nla_attr_size(skb->len) > USHRT_MAX) {
439 		err = -EFBIG;
440 		goto out;
441 	}
442 
443 	/* Complete checksum if needed */
444 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
445 	    (err = skb_csum_hwoffload_help(skb, 0)))
446 		goto out;
447 
448 	/* Older versions of OVS user space enforce alignment of the last
449 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
450 	 * padding logic. Only perform zerocopy if padding is not required.
451 	 */
452 	if (dp->user_features & OVS_DP_F_UNALIGNED)
453 		hlen = skb_zerocopy_headlen(skb);
454 	else
455 		hlen = skb->len;
456 
457 	len = upcall_msg_size(upcall_info, hlen - cutlen,
458 			      OVS_CB(skb)->acts_origlen);
459 	user_skb = genlmsg_new(len, GFP_ATOMIC);
460 	if (!user_skb) {
461 		err = -ENOMEM;
462 		goto out;
463 	}
464 
465 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
466 			     0, upcall_info->cmd);
467 	if (!upcall) {
468 		err = -EINVAL;
469 		goto out;
470 	}
471 	upcall->dp_ifindex = dp_ifindex;
472 
473 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
474 	if (err)
475 		goto out;
476 
477 	if (upcall_info->userdata)
478 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
479 			  nla_len(upcall_info->userdata),
480 			  nla_data(upcall_info->userdata));
481 
482 	if (upcall_info->egress_tun_info) {
483 		nla = nla_nest_start_noflag(user_skb,
484 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
485 		if (!nla) {
486 			err = -EMSGSIZE;
487 			goto out;
488 		}
489 		err = ovs_nla_put_tunnel_info(user_skb,
490 					      upcall_info->egress_tun_info);
491 		if (err)
492 			goto out;
493 
494 		nla_nest_end(user_skb, nla);
495 	}
496 
497 	if (upcall_info->actions_len) {
498 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
499 		if (!nla) {
500 			err = -EMSGSIZE;
501 			goto out;
502 		}
503 		err = ovs_nla_put_actions(upcall_info->actions,
504 					  upcall_info->actions_len,
505 					  user_skb);
506 		if (!err)
507 			nla_nest_end(user_skb, nla);
508 		else
509 			nla_nest_cancel(user_skb, nla);
510 	}
511 
512 	/* Add OVS_PACKET_ATTR_MRU */
513 	if (upcall_info->mru &&
514 	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
515 		err = -ENOBUFS;
516 		goto out;
517 	}
518 
519 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
520 	if (cutlen > 0 &&
521 	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
522 		err = -ENOBUFS;
523 		goto out;
524 	}
525 
526 	/* Add OVS_PACKET_ATTR_HASH */
527 	hash = skb_get_hash_raw(skb);
528 	if (skb->sw_hash)
529 		hash |= OVS_PACKET_HASH_SW_BIT;
530 
531 	if (skb->l4_hash)
532 		hash |= OVS_PACKET_HASH_L4_BIT;
533 
534 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
535 		err = -ENOBUFS;
536 		goto out;
537 	}
538 
539 	/* Only reserve room for attribute header, packet data is added
540 	 * in skb_zerocopy() */
541 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
542 		err = -ENOBUFS;
543 		goto out;
544 	}
545 	nla->nla_len = nla_attr_size(skb->len - cutlen);
546 
547 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
548 	if (err)
549 		goto out;
550 
551 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
552 	pad_packet(dp, user_skb);
553 
554 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
555 
556 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
557 	user_skb = NULL;
558 out:
559 	if (err)
560 		skb_tx_error(skb);
561 	consume_skb(user_skb);
562 	consume_skb(nskb);
563 
564 	return err;
565 }
566 
567 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
568 {
569 	struct ovs_header *ovs_header = info->userhdr;
570 	struct net *net = sock_net(skb->sk);
571 	struct nlattr **a = info->attrs;
572 	struct sw_flow_actions *acts;
573 	struct sk_buff *packet;
574 	struct sw_flow *flow;
575 	struct sw_flow_actions *sf_acts;
576 	struct datapath *dp;
577 	struct vport *input_vport;
578 	u16 mru = 0;
579 	u64 hash;
580 	int len;
581 	int err;
582 	bool log = !a[OVS_PACKET_ATTR_PROBE];
583 
584 	err = -EINVAL;
585 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
586 	    !a[OVS_PACKET_ATTR_ACTIONS])
587 		goto err;
588 
589 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
590 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
591 	err = -ENOMEM;
592 	if (!packet)
593 		goto err;
594 	skb_reserve(packet, NET_IP_ALIGN);
595 
596 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
597 
598 	/* Set packet's mru */
599 	if (a[OVS_PACKET_ATTR_MRU]) {
600 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
601 		packet->ignore_df = 1;
602 	}
603 	OVS_CB(packet)->mru = mru;
604 
605 	if (a[OVS_PACKET_ATTR_HASH]) {
606 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
607 
608 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
609 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
610 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
611 	}
612 
613 	/* Build an sw_flow for sending this packet. */
614 	flow = ovs_flow_alloc();
615 	err = PTR_ERR(flow);
616 	if (IS_ERR(flow))
617 		goto err_kfree_skb;
618 
619 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
620 					     packet, &flow->key, log);
621 	if (err)
622 		goto err_flow_free;
623 
624 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
625 				   &flow->key, &acts, log);
626 	if (err)
627 		goto err_flow_free;
628 
629 	rcu_assign_pointer(flow->sf_acts, acts);
630 	packet->priority = flow->key.phy.priority;
631 	packet->mark = flow->key.phy.skb_mark;
632 
633 	rcu_read_lock();
634 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
635 	err = -ENODEV;
636 	if (!dp)
637 		goto err_unlock;
638 
639 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
640 	if (!input_vport)
641 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
642 
643 	if (!input_vport)
644 		goto err_unlock;
645 
646 	packet->dev = input_vport->dev;
647 	OVS_CB(packet)->input_vport = input_vport;
648 	sf_acts = rcu_dereference(flow->sf_acts);
649 
650 	local_bh_disable();
651 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
652 	local_bh_enable();
653 	rcu_read_unlock();
654 
655 	ovs_flow_free(flow, false);
656 	return err;
657 
658 err_unlock:
659 	rcu_read_unlock();
660 err_flow_free:
661 	ovs_flow_free(flow, false);
662 err_kfree_skb:
663 	kfree_skb(packet);
664 err:
665 	return err;
666 }
667 
668 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
669 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
670 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
671 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
672 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
673 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
674 	[OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
675 };
676 
677 static const struct genl_small_ops dp_packet_genl_ops[] = {
678 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
679 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
680 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
681 	  .doit = ovs_packet_cmd_execute
682 	}
683 };
684 
685 static struct genl_family dp_packet_genl_family __ro_after_init = {
686 	.hdrsize = sizeof(struct ovs_header),
687 	.name = OVS_PACKET_FAMILY,
688 	.version = OVS_PACKET_VERSION,
689 	.maxattr = OVS_PACKET_ATTR_MAX,
690 	.policy = packet_policy,
691 	.netnsok = true,
692 	.parallel_ops = true,
693 	.small_ops = dp_packet_genl_ops,
694 	.n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
695 	.resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
696 	.module = THIS_MODULE,
697 };
698 
699 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
700 			 struct ovs_dp_megaflow_stats *mega_stats)
701 {
702 	int i;
703 
704 	memset(mega_stats, 0, sizeof(*mega_stats));
705 
706 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
707 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
708 
709 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
710 
711 	for_each_possible_cpu(i) {
712 		const struct dp_stats_percpu *percpu_stats;
713 		struct dp_stats_percpu local_stats;
714 		unsigned int start;
715 
716 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
717 
718 		do {
719 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
720 			local_stats = *percpu_stats;
721 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
722 
723 		stats->n_hit += local_stats.n_hit;
724 		stats->n_missed += local_stats.n_missed;
725 		stats->n_lost += local_stats.n_lost;
726 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
727 		mega_stats->n_cache_hit += local_stats.n_cache_hit;
728 	}
729 }
730 
731 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
732 {
733 	return ovs_identifier_is_ufid(sfid) &&
734 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
735 }
736 
737 static bool should_fill_mask(uint32_t ufid_flags)
738 {
739 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
740 }
741 
742 static bool should_fill_actions(uint32_t ufid_flags)
743 {
744 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
745 }
746 
747 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
748 				    const struct sw_flow_id *sfid,
749 				    uint32_t ufid_flags)
750 {
751 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
752 
753 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
754 	 * see ovs_nla_put_identifier()
755 	 */
756 	if (sfid && ovs_identifier_is_ufid(sfid))
757 		len += nla_total_size(sfid->ufid_len);
758 	else
759 		len += nla_total_size(ovs_key_attr_size());
760 
761 	/* OVS_FLOW_ATTR_KEY */
762 	if (!sfid || should_fill_key(sfid, ufid_flags))
763 		len += nla_total_size(ovs_key_attr_size());
764 
765 	/* OVS_FLOW_ATTR_MASK */
766 	if (should_fill_mask(ufid_flags))
767 		len += nla_total_size(ovs_key_attr_size());
768 
769 	/* OVS_FLOW_ATTR_ACTIONS */
770 	if (should_fill_actions(ufid_flags))
771 		len += nla_total_size(acts->orig_len);
772 
773 	return len
774 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
775 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
776 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
777 }
778 
779 /* Called with ovs_mutex or RCU read lock. */
780 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
781 				   struct sk_buff *skb)
782 {
783 	struct ovs_flow_stats stats;
784 	__be16 tcp_flags;
785 	unsigned long used;
786 
787 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
788 
789 	if (used &&
790 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
791 			      OVS_FLOW_ATTR_PAD))
792 		return -EMSGSIZE;
793 
794 	if (stats.n_packets &&
795 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
796 			  sizeof(struct ovs_flow_stats), &stats,
797 			  OVS_FLOW_ATTR_PAD))
798 		return -EMSGSIZE;
799 
800 	if ((u8)ntohs(tcp_flags) &&
801 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
802 		return -EMSGSIZE;
803 
804 	return 0;
805 }
806 
807 /* Called with ovs_mutex or RCU read lock. */
808 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
809 				     struct sk_buff *skb, int skb_orig_len)
810 {
811 	struct nlattr *start;
812 	int err;
813 
814 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
815 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
816 	 * Netlink but individual action lists can be longer than
817 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
818 	 * The userspace caller can always fetch the actions separately if it
819 	 * really wants them.  (Most userspace callers in fact don't care.)
820 	 *
821 	 * This can only fail for dump operations because the skb is always
822 	 * properly sized for single flows.
823 	 */
824 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
825 	if (start) {
826 		const struct sw_flow_actions *sf_acts;
827 
828 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
829 		err = ovs_nla_put_actions(sf_acts->actions,
830 					  sf_acts->actions_len, skb);
831 
832 		if (!err)
833 			nla_nest_end(skb, start);
834 		else {
835 			if (skb_orig_len)
836 				return err;
837 
838 			nla_nest_cancel(skb, start);
839 		}
840 	} else if (skb_orig_len) {
841 		return -EMSGSIZE;
842 	}
843 
844 	return 0;
845 }
846 
847 /* Called with ovs_mutex or RCU read lock. */
848 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
849 				  struct sk_buff *skb, u32 portid,
850 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
851 {
852 	const int skb_orig_len = skb->len;
853 	struct ovs_header *ovs_header;
854 	int err;
855 
856 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
857 				 flags, cmd);
858 	if (!ovs_header)
859 		return -EMSGSIZE;
860 
861 	ovs_header->dp_ifindex = dp_ifindex;
862 
863 	err = ovs_nla_put_identifier(flow, skb);
864 	if (err)
865 		goto error;
866 
867 	if (should_fill_key(&flow->id, ufid_flags)) {
868 		err = ovs_nla_put_masked_key(flow, skb);
869 		if (err)
870 			goto error;
871 	}
872 
873 	if (should_fill_mask(ufid_flags)) {
874 		err = ovs_nla_put_mask(flow, skb);
875 		if (err)
876 			goto error;
877 	}
878 
879 	err = ovs_flow_cmd_fill_stats(flow, skb);
880 	if (err)
881 		goto error;
882 
883 	if (should_fill_actions(ufid_flags)) {
884 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
885 		if (err)
886 			goto error;
887 	}
888 
889 	genlmsg_end(skb, ovs_header);
890 	return 0;
891 
892 error:
893 	genlmsg_cancel(skb, ovs_header);
894 	return err;
895 }
896 
897 /* May not be called with RCU read lock. */
898 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
899 					       const struct sw_flow_id *sfid,
900 					       struct genl_info *info,
901 					       bool always,
902 					       uint32_t ufid_flags)
903 {
904 	struct sk_buff *skb;
905 	size_t len;
906 
907 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
908 		return NULL;
909 
910 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
911 	skb = genlmsg_new(len, GFP_KERNEL);
912 	if (!skb)
913 		return ERR_PTR(-ENOMEM);
914 
915 	return skb;
916 }
917 
918 /* Called with ovs_mutex. */
919 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
920 					       int dp_ifindex,
921 					       struct genl_info *info, u8 cmd,
922 					       bool always, u32 ufid_flags)
923 {
924 	struct sk_buff *skb;
925 	int retval;
926 
927 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
928 				      &flow->id, info, always, ufid_flags);
929 	if (IS_ERR_OR_NULL(skb))
930 		return skb;
931 
932 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
933 					info->snd_portid, info->snd_seq, 0,
934 					cmd, ufid_flags);
935 	if (WARN_ON_ONCE(retval < 0)) {
936 		kfree_skb(skb);
937 		skb = ERR_PTR(retval);
938 	}
939 	return skb;
940 }
941 
942 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
943 {
944 	struct net *net = sock_net(skb->sk);
945 	struct nlattr **a = info->attrs;
946 	struct ovs_header *ovs_header = info->userhdr;
947 	struct sw_flow *flow = NULL, *new_flow;
948 	struct sw_flow_mask mask;
949 	struct sk_buff *reply;
950 	struct datapath *dp;
951 	struct sw_flow_actions *acts;
952 	struct sw_flow_match match;
953 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
954 	int error;
955 	bool log = !a[OVS_FLOW_ATTR_PROBE];
956 
957 	/* Must have key and actions. */
958 	error = -EINVAL;
959 	if (!a[OVS_FLOW_ATTR_KEY]) {
960 		OVS_NLERR(log, "Flow key attr not present in new flow.");
961 		goto error;
962 	}
963 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
964 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
965 		goto error;
966 	}
967 
968 	/* Most of the time we need to allocate a new flow, do it before
969 	 * locking.
970 	 */
971 	new_flow = ovs_flow_alloc();
972 	if (IS_ERR(new_flow)) {
973 		error = PTR_ERR(new_flow);
974 		goto error;
975 	}
976 
977 	/* Extract key. */
978 	ovs_match_init(&match, &new_flow->key, false, &mask);
979 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
980 				  a[OVS_FLOW_ATTR_MASK], log);
981 	if (error)
982 		goto err_kfree_flow;
983 
984 	/* Extract flow identifier. */
985 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
986 				       &new_flow->key, log);
987 	if (error)
988 		goto err_kfree_flow;
989 
990 	/* unmasked key is needed to match when ufid is not used. */
991 	if (ovs_identifier_is_key(&new_flow->id))
992 		match.key = new_flow->id.unmasked_key;
993 
994 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
995 
996 	/* Validate actions. */
997 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
998 				     &new_flow->key, &acts, log);
999 	if (error) {
1000 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1001 		goto err_kfree_flow;
1002 	}
1003 
1004 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1005 					ufid_flags);
1006 	if (IS_ERR(reply)) {
1007 		error = PTR_ERR(reply);
1008 		goto err_kfree_acts;
1009 	}
1010 
1011 	ovs_lock();
1012 	dp = get_dp(net, ovs_header->dp_ifindex);
1013 	if (unlikely(!dp)) {
1014 		error = -ENODEV;
1015 		goto err_unlock_ovs;
1016 	}
1017 
1018 	/* Check if this is a duplicate flow */
1019 	if (ovs_identifier_is_ufid(&new_flow->id))
1020 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1021 	if (!flow)
1022 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1023 	if (likely(!flow)) {
1024 		rcu_assign_pointer(new_flow->sf_acts, acts);
1025 
1026 		/* Put flow in bucket. */
1027 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1028 		if (unlikely(error)) {
1029 			acts = NULL;
1030 			goto err_unlock_ovs;
1031 		}
1032 
1033 		if (unlikely(reply)) {
1034 			error = ovs_flow_cmd_fill_info(new_flow,
1035 						       ovs_header->dp_ifindex,
1036 						       reply, info->snd_portid,
1037 						       info->snd_seq, 0,
1038 						       OVS_FLOW_CMD_NEW,
1039 						       ufid_flags);
1040 			BUG_ON(error < 0);
1041 		}
1042 		ovs_unlock();
1043 	} else {
1044 		struct sw_flow_actions *old_acts;
1045 
1046 		/* Bail out if we're not allowed to modify an existing flow.
1047 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1048 		 * because Generic Netlink treats the latter as a dump
1049 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1050 		 * gets fixed.
1051 		 */
1052 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1053 							 | NLM_F_EXCL))) {
1054 			error = -EEXIST;
1055 			goto err_unlock_ovs;
1056 		}
1057 		/* The flow identifier has to be the same for flow updates.
1058 		 * Look for any overlapping flow.
1059 		 */
1060 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1061 			if (ovs_identifier_is_key(&flow->id))
1062 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1063 								 &match);
1064 			else /* UFID matches but key is different */
1065 				flow = NULL;
1066 			if (!flow) {
1067 				error = -ENOENT;
1068 				goto err_unlock_ovs;
1069 			}
1070 		}
1071 		/* Update actions. */
1072 		old_acts = ovsl_dereference(flow->sf_acts);
1073 		rcu_assign_pointer(flow->sf_acts, acts);
1074 
1075 		if (unlikely(reply)) {
1076 			error = ovs_flow_cmd_fill_info(flow,
1077 						       ovs_header->dp_ifindex,
1078 						       reply, info->snd_portid,
1079 						       info->snd_seq, 0,
1080 						       OVS_FLOW_CMD_NEW,
1081 						       ufid_flags);
1082 			BUG_ON(error < 0);
1083 		}
1084 		ovs_unlock();
1085 
1086 		ovs_nla_free_flow_actions_rcu(old_acts);
1087 		ovs_flow_free(new_flow, false);
1088 	}
1089 
1090 	if (reply)
1091 		ovs_notify(&dp_flow_genl_family, reply, info);
1092 	return 0;
1093 
1094 err_unlock_ovs:
1095 	ovs_unlock();
1096 	kfree_skb(reply);
1097 err_kfree_acts:
1098 	ovs_nla_free_flow_actions(acts);
1099 err_kfree_flow:
1100 	ovs_flow_free(new_flow, false);
1101 error:
1102 	return error;
1103 }
1104 
1105 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1106 static noinline_for_stack
1107 struct sw_flow_actions *get_flow_actions(struct net *net,
1108 					 const struct nlattr *a,
1109 					 const struct sw_flow_key *key,
1110 					 const struct sw_flow_mask *mask,
1111 					 bool log)
1112 {
1113 	struct sw_flow_actions *acts;
1114 	struct sw_flow_key masked_key;
1115 	int error;
1116 
1117 	ovs_flow_mask_key(&masked_key, key, true, mask);
1118 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1119 	if (error) {
1120 		OVS_NLERR(log,
1121 			  "Actions may not be safe on all matching packets");
1122 		return ERR_PTR(error);
1123 	}
1124 
1125 	return acts;
1126 }
1127 
1128 /* Factor out match-init and action-copy to avoid
1129  * "Wframe-larger-than=1024" warning. Because mask is only
1130  * used to get actions, we new a function to save some
1131  * stack space.
1132  *
1133  * If there are not key and action attrs, we return 0
1134  * directly. In the case, the caller will also not use the
1135  * match as before. If there is action attr, we try to get
1136  * actions and save them to *acts. Before returning from
1137  * the function, we reset the match->mask pointer. Because
1138  * we should not to return match object with dangling reference
1139  * to mask.
1140  * */
1141 static noinline_for_stack int
1142 ovs_nla_init_match_and_action(struct net *net,
1143 			      struct sw_flow_match *match,
1144 			      struct sw_flow_key *key,
1145 			      struct nlattr **a,
1146 			      struct sw_flow_actions **acts,
1147 			      bool log)
1148 {
1149 	struct sw_flow_mask mask;
1150 	int error = 0;
1151 
1152 	if (a[OVS_FLOW_ATTR_KEY]) {
1153 		ovs_match_init(match, key, true, &mask);
1154 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1155 					  a[OVS_FLOW_ATTR_MASK], log);
1156 		if (error)
1157 			goto error;
1158 	}
1159 
1160 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1161 		if (!a[OVS_FLOW_ATTR_KEY]) {
1162 			OVS_NLERR(log,
1163 				  "Flow key attribute not present in set flow.");
1164 			error = -EINVAL;
1165 			goto error;
1166 		}
1167 
1168 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1169 					 &mask, log);
1170 		if (IS_ERR(*acts)) {
1171 			error = PTR_ERR(*acts);
1172 			goto error;
1173 		}
1174 	}
1175 
1176 	/* On success, error is 0. */
1177 error:
1178 	match->mask = NULL;
1179 	return error;
1180 }
1181 
1182 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1183 {
1184 	struct net *net = sock_net(skb->sk);
1185 	struct nlattr **a = info->attrs;
1186 	struct ovs_header *ovs_header = info->userhdr;
1187 	struct sw_flow_key key;
1188 	struct sw_flow *flow;
1189 	struct sk_buff *reply = NULL;
1190 	struct datapath *dp;
1191 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1192 	struct sw_flow_match match;
1193 	struct sw_flow_id sfid;
1194 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1195 	int error = 0;
1196 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1197 	bool ufid_present;
1198 
1199 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1200 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1201 		OVS_NLERR(log,
1202 			  "Flow set message rejected, Key attribute missing.");
1203 		return -EINVAL;
1204 	}
1205 
1206 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1207 					      &acts, log);
1208 	if (error)
1209 		goto error;
1210 
1211 	if (acts) {
1212 		/* Can allocate before locking if have acts. */
1213 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1214 						ufid_flags);
1215 		if (IS_ERR(reply)) {
1216 			error = PTR_ERR(reply);
1217 			goto err_kfree_acts;
1218 		}
1219 	}
1220 
1221 	ovs_lock();
1222 	dp = get_dp(net, ovs_header->dp_ifindex);
1223 	if (unlikely(!dp)) {
1224 		error = -ENODEV;
1225 		goto err_unlock_ovs;
1226 	}
1227 	/* Check that the flow exists. */
1228 	if (ufid_present)
1229 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1230 	else
1231 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1232 	if (unlikely(!flow)) {
1233 		error = -ENOENT;
1234 		goto err_unlock_ovs;
1235 	}
1236 
1237 	/* Update actions, if present. */
1238 	if (likely(acts)) {
1239 		old_acts = ovsl_dereference(flow->sf_acts);
1240 		rcu_assign_pointer(flow->sf_acts, acts);
1241 
1242 		if (unlikely(reply)) {
1243 			error = ovs_flow_cmd_fill_info(flow,
1244 						       ovs_header->dp_ifindex,
1245 						       reply, info->snd_portid,
1246 						       info->snd_seq, 0,
1247 						       OVS_FLOW_CMD_SET,
1248 						       ufid_flags);
1249 			BUG_ON(error < 0);
1250 		}
1251 	} else {
1252 		/* Could not alloc without acts before locking. */
1253 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1254 						info, OVS_FLOW_CMD_SET, false,
1255 						ufid_flags);
1256 
1257 		if (IS_ERR(reply)) {
1258 			error = PTR_ERR(reply);
1259 			goto err_unlock_ovs;
1260 		}
1261 	}
1262 
1263 	/* Clear stats. */
1264 	if (a[OVS_FLOW_ATTR_CLEAR])
1265 		ovs_flow_stats_clear(flow);
1266 	ovs_unlock();
1267 
1268 	if (reply)
1269 		ovs_notify(&dp_flow_genl_family, reply, info);
1270 	if (old_acts)
1271 		ovs_nla_free_flow_actions_rcu(old_acts);
1272 
1273 	return 0;
1274 
1275 err_unlock_ovs:
1276 	ovs_unlock();
1277 	kfree_skb(reply);
1278 err_kfree_acts:
1279 	ovs_nla_free_flow_actions(acts);
1280 error:
1281 	return error;
1282 }
1283 
1284 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1285 {
1286 	struct nlattr **a = info->attrs;
1287 	struct ovs_header *ovs_header = info->userhdr;
1288 	struct net *net = sock_net(skb->sk);
1289 	struct sw_flow_key key;
1290 	struct sk_buff *reply;
1291 	struct sw_flow *flow;
1292 	struct datapath *dp;
1293 	struct sw_flow_match match;
1294 	struct sw_flow_id ufid;
1295 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1296 	int err = 0;
1297 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1298 	bool ufid_present;
1299 
1300 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1301 	if (a[OVS_FLOW_ATTR_KEY]) {
1302 		ovs_match_init(&match, &key, true, NULL);
1303 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1304 					log);
1305 	} else if (!ufid_present) {
1306 		OVS_NLERR(log,
1307 			  "Flow get message rejected, Key attribute missing.");
1308 		err = -EINVAL;
1309 	}
1310 	if (err)
1311 		return err;
1312 
1313 	ovs_lock();
1314 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1315 	if (!dp) {
1316 		err = -ENODEV;
1317 		goto unlock;
1318 	}
1319 
1320 	if (ufid_present)
1321 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1322 	else
1323 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1324 	if (!flow) {
1325 		err = -ENOENT;
1326 		goto unlock;
1327 	}
1328 
1329 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1330 					OVS_FLOW_CMD_GET, true, ufid_flags);
1331 	if (IS_ERR(reply)) {
1332 		err = PTR_ERR(reply);
1333 		goto unlock;
1334 	}
1335 
1336 	ovs_unlock();
1337 	return genlmsg_reply(reply, info);
1338 unlock:
1339 	ovs_unlock();
1340 	return err;
1341 }
1342 
1343 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1344 {
1345 	struct nlattr **a = info->attrs;
1346 	struct ovs_header *ovs_header = info->userhdr;
1347 	struct net *net = sock_net(skb->sk);
1348 	struct sw_flow_key key;
1349 	struct sk_buff *reply;
1350 	struct sw_flow *flow = NULL;
1351 	struct datapath *dp;
1352 	struct sw_flow_match match;
1353 	struct sw_flow_id ufid;
1354 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1355 	int err;
1356 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1357 	bool ufid_present;
1358 
1359 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1360 	if (a[OVS_FLOW_ATTR_KEY]) {
1361 		ovs_match_init(&match, &key, true, NULL);
1362 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1363 					NULL, log);
1364 		if (unlikely(err))
1365 			return err;
1366 	}
1367 
1368 	ovs_lock();
1369 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1370 	if (unlikely(!dp)) {
1371 		err = -ENODEV;
1372 		goto unlock;
1373 	}
1374 
1375 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1376 		err = ovs_flow_tbl_flush(&dp->table);
1377 		goto unlock;
1378 	}
1379 
1380 	if (ufid_present)
1381 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1382 	else
1383 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1384 	if (unlikely(!flow)) {
1385 		err = -ENOENT;
1386 		goto unlock;
1387 	}
1388 
1389 	ovs_flow_tbl_remove(&dp->table, flow);
1390 	ovs_unlock();
1391 
1392 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1393 					&flow->id, info, false, ufid_flags);
1394 	if (likely(reply)) {
1395 		if (!IS_ERR(reply)) {
1396 			rcu_read_lock();	/*To keep RCU checker happy. */
1397 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1398 						     reply, info->snd_portid,
1399 						     info->snd_seq, 0,
1400 						     OVS_FLOW_CMD_DEL,
1401 						     ufid_flags);
1402 			rcu_read_unlock();
1403 			if (WARN_ON_ONCE(err < 0)) {
1404 				kfree_skb(reply);
1405 				goto out_free;
1406 			}
1407 
1408 			ovs_notify(&dp_flow_genl_family, reply, info);
1409 		} else {
1410 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1411 					PTR_ERR(reply));
1412 		}
1413 	}
1414 
1415 out_free:
1416 	ovs_flow_free(flow, true);
1417 	return 0;
1418 unlock:
1419 	ovs_unlock();
1420 	return err;
1421 }
1422 
1423 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1424 {
1425 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1426 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1427 	struct table_instance *ti;
1428 	struct datapath *dp;
1429 	u32 ufid_flags;
1430 	int err;
1431 
1432 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1433 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1434 	if (err)
1435 		return err;
1436 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1437 
1438 	rcu_read_lock();
1439 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1440 	if (!dp) {
1441 		rcu_read_unlock();
1442 		return -ENODEV;
1443 	}
1444 
1445 	ti = rcu_dereference(dp->table.ti);
1446 	for (;;) {
1447 		struct sw_flow *flow;
1448 		u32 bucket, obj;
1449 
1450 		bucket = cb->args[0];
1451 		obj = cb->args[1];
1452 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1453 		if (!flow)
1454 			break;
1455 
1456 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1457 					   NETLINK_CB(cb->skb).portid,
1458 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1459 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1460 			break;
1461 
1462 		cb->args[0] = bucket;
1463 		cb->args[1] = obj;
1464 	}
1465 	rcu_read_unlock();
1466 	return skb->len;
1467 }
1468 
1469 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1470 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1471 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1472 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1473 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1474 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1475 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1476 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1477 };
1478 
1479 static const struct genl_small_ops dp_flow_genl_ops[] = {
1480 	{ .cmd = OVS_FLOW_CMD_NEW,
1481 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1482 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1483 	  .doit = ovs_flow_cmd_new
1484 	},
1485 	{ .cmd = OVS_FLOW_CMD_DEL,
1486 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1487 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1488 	  .doit = ovs_flow_cmd_del
1489 	},
1490 	{ .cmd = OVS_FLOW_CMD_GET,
1491 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1492 	  .flags = 0,		    /* OK for unprivileged users. */
1493 	  .doit = ovs_flow_cmd_get,
1494 	  .dumpit = ovs_flow_cmd_dump
1495 	},
1496 	{ .cmd = OVS_FLOW_CMD_SET,
1497 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1498 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1499 	  .doit = ovs_flow_cmd_set,
1500 	},
1501 };
1502 
1503 static struct genl_family dp_flow_genl_family __ro_after_init = {
1504 	.hdrsize = sizeof(struct ovs_header),
1505 	.name = OVS_FLOW_FAMILY,
1506 	.version = OVS_FLOW_VERSION,
1507 	.maxattr = OVS_FLOW_ATTR_MAX,
1508 	.policy = flow_policy,
1509 	.netnsok = true,
1510 	.parallel_ops = true,
1511 	.small_ops = dp_flow_genl_ops,
1512 	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1513 	.resv_start_op = OVS_FLOW_CMD_SET + 1,
1514 	.mcgrps = &ovs_dp_flow_multicast_group,
1515 	.n_mcgrps = 1,
1516 	.module = THIS_MODULE,
1517 };
1518 
1519 static size_t ovs_dp_cmd_msg_size(void)
1520 {
1521 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1522 
1523 	msgsize += nla_total_size(IFNAMSIZ);
1524 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1525 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1526 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1527 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1528 	msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1529 
1530 	return msgsize;
1531 }
1532 
1533 /* Called with ovs_mutex. */
1534 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1535 				u32 portid, u32 seq, u32 flags, u8 cmd)
1536 {
1537 	struct ovs_header *ovs_header;
1538 	struct ovs_dp_stats dp_stats;
1539 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1540 	struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1541 	int err, pids_len;
1542 
1543 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1544 				 flags, cmd);
1545 	if (!ovs_header)
1546 		goto error;
1547 
1548 	ovs_header->dp_ifindex = get_dpifindex(dp);
1549 
1550 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1551 	if (err)
1552 		goto nla_put_failure;
1553 
1554 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1555 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1556 			  &dp_stats, OVS_DP_ATTR_PAD))
1557 		goto nla_put_failure;
1558 
1559 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1560 			  sizeof(struct ovs_dp_megaflow_stats),
1561 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1562 		goto nla_put_failure;
1563 
1564 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1565 		goto nla_put_failure;
1566 
1567 	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1568 			ovs_flow_tbl_masks_cache_size(&dp->table)))
1569 		goto nla_put_failure;
1570 
1571 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1572 		pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1573 		if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1574 			goto nla_put_failure;
1575 	}
1576 
1577 	genlmsg_end(skb, ovs_header);
1578 	return 0;
1579 
1580 nla_put_failure:
1581 	genlmsg_cancel(skb, ovs_header);
1582 error:
1583 	return -EMSGSIZE;
1584 }
1585 
1586 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1587 {
1588 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1589 }
1590 
1591 /* Called with rcu_read_lock or ovs_mutex. */
1592 static struct datapath *lookup_datapath(struct net *net,
1593 					const struct ovs_header *ovs_header,
1594 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1595 {
1596 	struct datapath *dp;
1597 
1598 	if (!a[OVS_DP_ATTR_NAME])
1599 		dp = get_dp(net, ovs_header->dp_ifindex);
1600 	else {
1601 		struct vport *vport;
1602 
1603 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1604 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1605 	}
1606 	return dp ? dp : ERR_PTR(-ENODEV);
1607 }
1608 
1609 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1610 				       struct genl_info *info)
1611 {
1612 	struct datapath *dp;
1613 
1614 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1615 			     info->attrs);
1616 	if (IS_ERR(dp))
1617 		return;
1618 
1619 	WARN(dp->user_features, "Dropping previously announced user features\n");
1620 	dp->user_features = 0;
1621 }
1622 
1623 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1624 			      const struct nlattr *ids)
1625 {
1626 	struct dp_nlsk_pids *old, *dp_nlsk_pids;
1627 
1628 	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1629 		return -EINVAL;
1630 
1631 	old = ovsl_dereference(dp->upcall_portids);
1632 
1633 	dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1634 			       GFP_KERNEL);
1635 	if (!dp_nlsk_pids)
1636 		return -ENOMEM;
1637 
1638 	dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1639 	nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1640 
1641 	rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1642 
1643 	kfree_rcu(old, rcu);
1644 
1645 	return 0;
1646 }
1647 
1648 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1649 {
1650 	struct dp_nlsk_pids *dp_nlsk_pids;
1651 
1652 	dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1653 
1654 	if (dp_nlsk_pids) {
1655 		if (cpu_id < dp_nlsk_pids->n_pids) {
1656 			return dp_nlsk_pids->pids[cpu_id];
1657 		} else if (dp_nlsk_pids->n_pids > 0 &&
1658 			   cpu_id >= dp_nlsk_pids->n_pids) {
1659 			/* If the number of netlink PIDs is mismatched with
1660 			 * the number of CPUs as seen by the kernel, log this
1661 			 * and send the upcall to an arbitrary socket (0) in
1662 			 * order to not drop packets
1663 			 */
1664 			pr_info_ratelimited("cpu_id mismatch with handler threads");
1665 			return dp_nlsk_pids->pids[cpu_id %
1666 						  dp_nlsk_pids->n_pids];
1667 		} else {
1668 			return 0;
1669 		}
1670 	} else {
1671 		return 0;
1672 	}
1673 }
1674 
1675 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1676 {
1677 	u32 user_features = 0, old_features = dp->user_features;
1678 	int err;
1679 
1680 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1681 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1682 
1683 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1684 				      OVS_DP_F_UNALIGNED |
1685 				      OVS_DP_F_TC_RECIRC_SHARING |
1686 				      OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1687 			return -EOPNOTSUPP;
1688 
1689 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1690 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1691 			return -EOPNOTSUPP;
1692 #endif
1693 	}
1694 
1695 	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1696 		int err;
1697 		u32 cache_size;
1698 
1699 		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1700 		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1701 		if (err)
1702 			return err;
1703 	}
1704 
1705 	dp->user_features = user_features;
1706 
1707 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1708 	    a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1709 		/* Upcall Netlink Port IDs have been updated */
1710 		err = ovs_dp_set_upcall_portids(dp,
1711 						a[OVS_DP_ATTR_PER_CPU_PIDS]);
1712 		if (err)
1713 			return err;
1714 	}
1715 
1716 	if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1717 	    !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1718 		tc_skb_ext_tc_enable();
1719 	else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1720 		 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1721 		tc_skb_ext_tc_disable();
1722 
1723 	return 0;
1724 }
1725 
1726 static int ovs_dp_stats_init(struct datapath *dp)
1727 {
1728 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1729 	if (!dp->stats_percpu)
1730 		return -ENOMEM;
1731 
1732 	return 0;
1733 }
1734 
1735 static int ovs_dp_vport_init(struct datapath *dp)
1736 {
1737 	int i;
1738 
1739 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1740 				  sizeof(struct hlist_head),
1741 				  GFP_KERNEL);
1742 	if (!dp->ports)
1743 		return -ENOMEM;
1744 
1745 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1746 		INIT_HLIST_HEAD(&dp->ports[i]);
1747 
1748 	return 0;
1749 }
1750 
1751 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1752 {
1753 	struct nlattr **a = info->attrs;
1754 	struct vport_parms parms;
1755 	struct sk_buff *reply;
1756 	struct datapath *dp;
1757 	struct vport *vport;
1758 	struct ovs_net *ovs_net;
1759 	int err;
1760 
1761 	err = -EINVAL;
1762 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1763 		goto err;
1764 
1765 	reply = ovs_dp_cmd_alloc_info();
1766 	if (!reply)
1767 		return -ENOMEM;
1768 
1769 	err = -ENOMEM;
1770 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1771 	if (dp == NULL)
1772 		goto err_destroy_reply;
1773 
1774 	ovs_dp_set_net(dp, sock_net(skb->sk));
1775 
1776 	/* Allocate table. */
1777 	err = ovs_flow_tbl_init(&dp->table);
1778 	if (err)
1779 		goto err_destroy_dp;
1780 
1781 	err = ovs_dp_stats_init(dp);
1782 	if (err)
1783 		goto err_destroy_table;
1784 
1785 	err = ovs_dp_vport_init(dp);
1786 	if (err)
1787 		goto err_destroy_stats;
1788 
1789 	err = ovs_meters_init(dp);
1790 	if (err)
1791 		goto err_destroy_ports;
1792 
1793 	/* Set up our datapath device. */
1794 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1795 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1796 	parms.options = NULL;
1797 	parms.dp = dp;
1798 	parms.port_no = OVSP_LOCAL;
1799 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1800 	parms.desired_ifindex = a[OVS_DP_ATTR_IFINDEX]
1801 		? nla_get_u32(a[OVS_DP_ATTR_IFINDEX]) : 0;
1802 
1803 	/* So far only local changes have been made, now need the lock. */
1804 	ovs_lock();
1805 
1806 	err = ovs_dp_change(dp, a);
1807 	if (err)
1808 		goto err_unlock_and_destroy_meters;
1809 
1810 	vport = new_vport(&parms);
1811 	if (IS_ERR(vport)) {
1812 		err = PTR_ERR(vport);
1813 		if (err == -EBUSY)
1814 			err = -EEXIST;
1815 
1816 		if (err == -EEXIST) {
1817 			/* An outdated user space instance that does not understand
1818 			 * the concept of user_features has attempted to create a new
1819 			 * datapath and is likely to reuse it. Drop all user features.
1820 			 */
1821 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1822 				ovs_dp_reset_user_features(skb, info);
1823 		}
1824 
1825 		goto err_destroy_portids;
1826 	}
1827 
1828 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1829 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1830 	BUG_ON(err < 0);
1831 
1832 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1833 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1834 
1835 	ovs_unlock();
1836 
1837 	ovs_notify(&dp_datapath_genl_family, reply, info);
1838 	return 0;
1839 
1840 err_destroy_portids:
1841 	kfree(rcu_dereference_raw(dp->upcall_portids));
1842 err_unlock_and_destroy_meters:
1843 	ovs_unlock();
1844 	ovs_meters_exit(dp);
1845 err_destroy_ports:
1846 	kfree(dp->ports);
1847 err_destroy_stats:
1848 	free_percpu(dp->stats_percpu);
1849 err_destroy_table:
1850 	ovs_flow_tbl_destroy(&dp->table);
1851 err_destroy_dp:
1852 	kfree(dp);
1853 err_destroy_reply:
1854 	kfree_skb(reply);
1855 err:
1856 	return err;
1857 }
1858 
1859 /* Called with ovs_mutex. */
1860 static void __dp_destroy(struct datapath *dp)
1861 {
1862 	struct flow_table *table = &dp->table;
1863 	int i;
1864 
1865 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1866 		tc_skb_ext_tc_disable();
1867 
1868 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1869 		struct vport *vport;
1870 		struct hlist_node *n;
1871 
1872 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1873 			if (vport->port_no != OVSP_LOCAL)
1874 				ovs_dp_detach_port(vport);
1875 	}
1876 
1877 	list_del_rcu(&dp->list_node);
1878 
1879 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1880 	 * all ports in datapath are destroyed first before freeing datapath.
1881 	 */
1882 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1883 
1884 	/* Flush sw_flow in the tables. RCU cb only releases resource
1885 	 * such as dp, ports and tables. That may avoid some issues
1886 	 * such as RCU usage warning.
1887 	 */
1888 	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1889 				  ovsl_dereference(table->ufid_ti));
1890 
1891 	/* RCU destroy the ports, meters and flow tables. */
1892 	call_rcu(&dp->rcu, destroy_dp_rcu);
1893 }
1894 
1895 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1896 {
1897 	struct sk_buff *reply;
1898 	struct datapath *dp;
1899 	int err;
1900 
1901 	reply = ovs_dp_cmd_alloc_info();
1902 	if (!reply)
1903 		return -ENOMEM;
1904 
1905 	ovs_lock();
1906 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1907 	err = PTR_ERR(dp);
1908 	if (IS_ERR(dp))
1909 		goto err_unlock_free;
1910 
1911 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1912 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1913 	BUG_ON(err < 0);
1914 
1915 	__dp_destroy(dp);
1916 	ovs_unlock();
1917 
1918 	ovs_notify(&dp_datapath_genl_family, reply, info);
1919 
1920 	return 0;
1921 
1922 err_unlock_free:
1923 	ovs_unlock();
1924 	kfree_skb(reply);
1925 	return err;
1926 }
1927 
1928 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1929 {
1930 	struct sk_buff *reply;
1931 	struct datapath *dp;
1932 	int err;
1933 
1934 	reply = ovs_dp_cmd_alloc_info();
1935 	if (!reply)
1936 		return -ENOMEM;
1937 
1938 	ovs_lock();
1939 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1940 	err = PTR_ERR(dp);
1941 	if (IS_ERR(dp))
1942 		goto err_unlock_free;
1943 
1944 	err = ovs_dp_change(dp, info->attrs);
1945 	if (err)
1946 		goto err_unlock_free;
1947 
1948 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1949 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1950 	BUG_ON(err < 0);
1951 
1952 	ovs_unlock();
1953 	ovs_notify(&dp_datapath_genl_family, reply, info);
1954 
1955 	return 0;
1956 
1957 err_unlock_free:
1958 	ovs_unlock();
1959 	kfree_skb(reply);
1960 	return err;
1961 }
1962 
1963 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1964 {
1965 	struct sk_buff *reply;
1966 	struct datapath *dp;
1967 	int err;
1968 
1969 	reply = ovs_dp_cmd_alloc_info();
1970 	if (!reply)
1971 		return -ENOMEM;
1972 
1973 	ovs_lock();
1974 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1975 	if (IS_ERR(dp)) {
1976 		err = PTR_ERR(dp);
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_GET);
1981 	BUG_ON(err < 0);
1982 	ovs_unlock();
1983 
1984 	return genlmsg_reply(reply, info);
1985 
1986 err_unlock_free:
1987 	ovs_unlock();
1988 	kfree_skb(reply);
1989 	return err;
1990 }
1991 
1992 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1993 {
1994 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1995 	struct datapath *dp;
1996 	int skip = cb->args[0];
1997 	int i = 0;
1998 
1999 	ovs_lock();
2000 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2001 		if (i >= skip &&
2002 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2003 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2004 					 OVS_DP_CMD_GET) < 0)
2005 			break;
2006 		i++;
2007 	}
2008 	ovs_unlock();
2009 
2010 	cb->args[0] = i;
2011 
2012 	return skb->len;
2013 }
2014 
2015 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2016 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2017 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2018 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2019 	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
2020 		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2021 	[OVS_DP_ATTR_IFINDEX] = {.type = NLA_U32 },
2022 };
2023 
2024 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2025 	{ .cmd = OVS_DP_CMD_NEW,
2026 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2027 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2028 	  .doit = ovs_dp_cmd_new
2029 	},
2030 	{ .cmd = OVS_DP_CMD_DEL,
2031 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2032 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2033 	  .doit = ovs_dp_cmd_del
2034 	},
2035 	{ .cmd = OVS_DP_CMD_GET,
2036 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2037 	  .flags = 0,		    /* OK for unprivileged users. */
2038 	  .doit = ovs_dp_cmd_get,
2039 	  .dumpit = ovs_dp_cmd_dump
2040 	},
2041 	{ .cmd = OVS_DP_CMD_SET,
2042 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2043 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2044 	  .doit = ovs_dp_cmd_set,
2045 	},
2046 };
2047 
2048 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2049 	.hdrsize = sizeof(struct ovs_header),
2050 	.name = OVS_DATAPATH_FAMILY,
2051 	.version = OVS_DATAPATH_VERSION,
2052 	.maxattr = OVS_DP_ATTR_MAX,
2053 	.policy = datapath_policy,
2054 	.netnsok = true,
2055 	.parallel_ops = true,
2056 	.small_ops = dp_datapath_genl_ops,
2057 	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2058 	.resv_start_op = OVS_DP_CMD_SET + 1,
2059 	.mcgrps = &ovs_dp_datapath_multicast_group,
2060 	.n_mcgrps = 1,
2061 	.module = THIS_MODULE,
2062 };
2063 
2064 /* Called with ovs_mutex or RCU read lock. */
2065 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2066 				   struct net *net, u32 portid, u32 seq,
2067 				   u32 flags, u8 cmd, gfp_t gfp)
2068 {
2069 	struct ovs_header *ovs_header;
2070 	struct ovs_vport_stats vport_stats;
2071 	int err;
2072 
2073 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2074 				 flags, cmd);
2075 	if (!ovs_header)
2076 		return -EMSGSIZE;
2077 
2078 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2079 
2080 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2081 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2082 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2083 			   ovs_vport_name(vport)) ||
2084 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2085 		goto nla_put_failure;
2086 
2087 	if (!net_eq(net, dev_net(vport->dev))) {
2088 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2089 
2090 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2091 			goto nla_put_failure;
2092 	}
2093 
2094 	ovs_vport_get_stats(vport, &vport_stats);
2095 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2096 			  sizeof(struct ovs_vport_stats), &vport_stats,
2097 			  OVS_VPORT_ATTR_PAD))
2098 		goto nla_put_failure;
2099 
2100 	if (ovs_vport_get_upcall_portids(vport, skb))
2101 		goto nla_put_failure;
2102 
2103 	err = ovs_vport_get_options(vport, skb);
2104 	if (err == -EMSGSIZE)
2105 		goto error;
2106 
2107 	genlmsg_end(skb, ovs_header);
2108 	return 0;
2109 
2110 nla_put_failure:
2111 	err = -EMSGSIZE;
2112 error:
2113 	genlmsg_cancel(skb, ovs_header);
2114 	return err;
2115 }
2116 
2117 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2118 {
2119 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2120 }
2121 
2122 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2123 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2124 					 u32 portid, u32 seq, u8 cmd)
2125 {
2126 	struct sk_buff *skb;
2127 	int retval;
2128 
2129 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2130 	if (!skb)
2131 		return ERR_PTR(-ENOMEM);
2132 
2133 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2134 					 GFP_KERNEL);
2135 	BUG_ON(retval < 0);
2136 
2137 	return skb;
2138 }
2139 
2140 /* Called with ovs_mutex or RCU read lock. */
2141 static struct vport *lookup_vport(struct net *net,
2142 				  const struct ovs_header *ovs_header,
2143 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2144 {
2145 	struct datapath *dp;
2146 	struct vport *vport;
2147 
2148 	if (a[OVS_VPORT_ATTR_IFINDEX])
2149 		return ERR_PTR(-EOPNOTSUPP);
2150 	if (a[OVS_VPORT_ATTR_NAME]) {
2151 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2152 		if (!vport)
2153 			return ERR_PTR(-ENODEV);
2154 		if (ovs_header->dp_ifindex &&
2155 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2156 			return ERR_PTR(-ENODEV);
2157 		return vport;
2158 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2159 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2160 
2161 		if (port_no >= DP_MAX_PORTS)
2162 			return ERR_PTR(-EFBIG);
2163 
2164 		dp = get_dp(net, ovs_header->dp_ifindex);
2165 		if (!dp)
2166 			return ERR_PTR(-ENODEV);
2167 
2168 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2169 		if (!vport)
2170 			return ERR_PTR(-ENODEV);
2171 		return vport;
2172 	} else
2173 		return ERR_PTR(-EINVAL);
2174 
2175 }
2176 
2177 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2178 {
2179 	unsigned int dev_headroom, max_headroom = 0;
2180 	struct net_device *dev;
2181 	struct vport *vport;
2182 	int i;
2183 
2184 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2185 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2186 					 lockdep_ovsl_is_held()) {
2187 			dev = vport->dev;
2188 			dev_headroom = netdev_get_fwd_headroom(dev);
2189 			if (dev_headroom > max_headroom)
2190 				max_headroom = dev_headroom;
2191 		}
2192 	}
2193 
2194 	return max_headroom;
2195 }
2196 
2197 /* Called with ovs_mutex */
2198 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2199 {
2200 	struct vport *vport;
2201 	int i;
2202 
2203 	dp->max_headroom = new_headroom;
2204 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2205 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2206 					 lockdep_ovsl_is_held())
2207 			netdev_set_rx_headroom(vport->dev, new_headroom);
2208 	}
2209 }
2210 
2211 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2212 {
2213 	struct nlattr **a = info->attrs;
2214 	struct ovs_header *ovs_header = info->userhdr;
2215 	struct vport_parms parms;
2216 	struct sk_buff *reply;
2217 	struct vport *vport;
2218 	struct datapath *dp;
2219 	unsigned int new_headroom;
2220 	u32 port_no;
2221 	int err;
2222 
2223 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2224 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2225 		return -EINVAL;
2226 
2227 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2228 
2229 	if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2230 		return -EOPNOTSUPP;
2231 
2232 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2233 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2234 	if (port_no >= DP_MAX_PORTS)
2235 		return -EFBIG;
2236 
2237 	reply = ovs_vport_cmd_alloc_info();
2238 	if (!reply)
2239 		return -ENOMEM;
2240 
2241 	ovs_lock();
2242 restart:
2243 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2244 	err = -ENODEV;
2245 	if (!dp)
2246 		goto exit_unlock_free;
2247 
2248 	if (port_no) {
2249 		vport = ovs_vport_ovsl(dp, port_no);
2250 		err = -EBUSY;
2251 		if (vport)
2252 			goto exit_unlock_free;
2253 	} else {
2254 		for (port_no = 1; ; port_no++) {
2255 			if (port_no >= DP_MAX_PORTS) {
2256 				err = -EFBIG;
2257 				goto exit_unlock_free;
2258 			}
2259 			vport = ovs_vport_ovsl(dp, port_no);
2260 			if (!vport)
2261 				break;
2262 		}
2263 	}
2264 
2265 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2266 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2267 	parms.dp = dp;
2268 	parms.port_no = port_no;
2269 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2270 	parms.desired_ifindex = a[OVS_VPORT_ATTR_IFINDEX]
2271 		? nla_get_u32(a[OVS_VPORT_ATTR_IFINDEX]) : 0;
2272 
2273 	vport = new_vport(&parms);
2274 	err = PTR_ERR(vport);
2275 	if (IS_ERR(vport)) {
2276 		if (err == -EAGAIN)
2277 			goto restart;
2278 		goto exit_unlock_free;
2279 	}
2280 
2281 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2282 				      info->snd_portid, info->snd_seq, 0,
2283 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2284 
2285 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2286 
2287 	if (new_headroom > dp->max_headroom)
2288 		ovs_update_headroom(dp, new_headroom);
2289 	else
2290 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2291 
2292 	BUG_ON(err < 0);
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_set(struct sk_buff *skb, struct genl_info *info)
2305 {
2306 	struct nlattr **a = info->attrs;
2307 	struct sk_buff *reply;
2308 	struct vport *vport;
2309 	int err;
2310 
2311 	reply = ovs_vport_cmd_alloc_info();
2312 	if (!reply)
2313 		return -ENOMEM;
2314 
2315 	ovs_lock();
2316 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2317 	err = PTR_ERR(vport);
2318 	if (IS_ERR(vport))
2319 		goto exit_unlock_free;
2320 
2321 	if (a[OVS_VPORT_ATTR_TYPE] &&
2322 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2323 		err = -EINVAL;
2324 		goto exit_unlock_free;
2325 	}
2326 
2327 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2328 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2329 		if (err)
2330 			goto exit_unlock_free;
2331 	}
2332 
2333 
2334 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2335 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2336 
2337 		err = ovs_vport_set_upcall_portids(vport, ids);
2338 		if (err)
2339 			goto exit_unlock_free;
2340 	}
2341 
2342 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2343 				      info->snd_portid, info->snd_seq, 0,
2344 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2345 	BUG_ON(err < 0);
2346 
2347 	ovs_unlock();
2348 	ovs_notify(&dp_vport_genl_family, reply, info);
2349 	return 0;
2350 
2351 exit_unlock_free:
2352 	ovs_unlock();
2353 	kfree_skb(reply);
2354 	return err;
2355 }
2356 
2357 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2358 {
2359 	bool update_headroom = false;
2360 	struct nlattr **a = info->attrs;
2361 	struct sk_buff *reply;
2362 	struct datapath *dp;
2363 	struct vport *vport;
2364 	unsigned int new_headroom;
2365 	int err;
2366 
2367 	reply = ovs_vport_cmd_alloc_info();
2368 	if (!reply)
2369 		return -ENOMEM;
2370 
2371 	ovs_lock();
2372 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2373 	err = PTR_ERR(vport);
2374 	if (IS_ERR(vport))
2375 		goto exit_unlock_free;
2376 
2377 	if (vport->port_no == OVSP_LOCAL) {
2378 		err = -EINVAL;
2379 		goto exit_unlock_free;
2380 	}
2381 
2382 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2383 				      info->snd_portid, info->snd_seq, 0,
2384 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2385 	BUG_ON(err < 0);
2386 
2387 	/* the vport deletion may trigger dp headroom update */
2388 	dp = vport->dp;
2389 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2390 		update_headroom = true;
2391 
2392 	netdev_reset_rx_headroom(vport->dev);
2393 	ovs_dp_detach_port(vport);
2394 
2395 	if (update_headroom) {
2396 		new_headroom = ovs_get_max_headroom(dp);
2397 
2398 		if (new_headroom < dp->max_headroom)
2399 			ovs_update_headroom(dp, new_headroom);
2400 	}
2401 	ovs_unlock();
2402 
2403 	ovs_notify(&dp_vport_genl_family, reply, info);
2404 	return 0;
2405 
2406 exit_unlock_free:
2407 	ovs_unlock();
2408 	kfree_skb(reply);
2409 	return err;
2410 }
2411 
2412 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2413 {
2414 	struct nlattr **a = info->attrs;
2415 	struct ovs_header *ovs_header = info->userhdr;
2416 	struct sk_buff *reply;
2417 	struct vport *vport;
2418 	int err;
2419 
2420 	reply = ovs_vport_cmd_alloc_info();
2421 	if (!reply)
2422 		return -ENOMEM;
2423 
2424 	rcu_read_lock();
2425 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2426 	err = PTR_ERR(vport);
2427 	if (IS_ERR(vport))
2428 		goto exit_unlock_free;
2429 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2430 				      info->snd_portid, info->snd_seq, 0,
2431 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2432 	BUG_ON(err < 0);
2433 	rcu_read_unlock();
2434 
2435 	return genlmsg_reply(reply, info);
2436 
2437 exit_unlock_free:
2438 	rcu_read_unlock();
2439 	kfree_skb(reply);
2440 	return err;
2441 }
2442 
2443 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2444 {
2445 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2446 	struct datapath *dp;
2447 	int bucket = cb->args[0], skip = cb->args[1];
2448 	int i, j = 0;
2449 
2450 	rcu_read_lock();
2451 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2452 	if (!dp) {
2453 		rcu_read_unlock();
2454 		return -ENODEV;
2455 	}
2456 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2457 		struct vport *vport;
2458 
2459 		j = 0;
2460 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2461 			if (j >= skip &&
2462 			    ovs_vport_cmd_fill_info(vport, skb,
2463 						    sock_net(skb->sk),
2464 						    NETLINK_CB(cb->skb).portid,
2465 						    cb->nlh->nlmsg_seq,
2466 						    NLM_F_MULTI,
2467 						    OVS_VPORT_CMD_GET,
2468 						    GFP_ATOMIC) < 0)
2469 				goto out;
2470 
2471 			j++;
2472 		}
2473 		skip = 0;
2474 	}
2475 out:
2476 	rcu_read_unlock();
2477 
2478 	cb->args[0] = i;
2479 	cb->args[1] = j;
2480 
2481 	return skb->len;
2482 }
2483 
2484 static void ovs_dp_masks_rebalance(struct work_struct *work)
2485 {
2486 	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2487 					       masks_rebalance.work);
2488 	struct datapath *dp;
2489 
2490 	ovs_lock();
2491 
2492 	list_for_each_entry(dp, &ovs_net->dps, list_node)
2493 		ovs_flow_masks_rebalance(&dp->table);
2494 
2495 	ovs_unlock();
2496 
2497 	schedule_delayed_work(&ovs_net->masks_rebalance,
2498 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2499 }
2500 
2501 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2502 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2503 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2504 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2505 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2506 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2507 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2508 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2509 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2510 };
2511 
2512 static const struct genl_small_ops dp_vport_genl_ops[] = {
2513 	{ .cmd = OVS_VPORT_CMD_NEW,
2514 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2515 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2516 	  .doit = ovs_vport_cmd_new
2517 	},
2518 	{ .cmd = OVS_VPORT_CMD_DEL,
2519 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2520 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2521 	  .doit = ovs_vport_cmd_del
2522 	},
2523 	{ .cmd = OVS_VPORT_CMD_GET,
2524 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2525 	  .flags = 0,		    /* OK for unprivileged users. */
2526 	  .doit = ovs_vport_cmd_get,
2527 	  .dumpit = ovs_vport_cmd_dump
2528 	},
2529 	{ .cmd = OVS_VPORT_CMD_SET,
2530 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2531 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2532 	  .doit = ovs_vport_cmd_set,
2533 	},
2534 };
2535 
2536 struct genl_family dp_vport_genl_family __ro_after_init = {
2537 	.hdrsize = sizeof(struct ovs_header),
2538 	.name = OVS_VPORT_FAMILY,
2539 	.version = OVS_VPORT_VERSION,
2540 	.maxattr = OVS_VPORT_ATTR_MAX,
2541 	.policy = vport_policy,
2542 	.netnsok = true,
2543 	.parallel_ops = true,
2544 	.small_ops = dp_vport_genl_ops,
2545 	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2546 	.mcgrps = &ovs_dp_vport_multicast_group,
2547 	.n_mcgrps = 1,
2548 	.module = THIS_MODULE,
2549 };
2550 
2551 static struct genl_family * const dp_genl_families[] = {
2552 	&dp_datapath_genl_family,
2553 	&dp_vport_genl_family,
2554 	&dp_flow_genl_family,
2555 	&dp_packet_genl_family,
2556 	&dp_meter_genl_family,
2557 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2558 	&dp_ct_limit_genl_family,
2559 #endif
2560 };
2561 
2562 static void dp_unregister_genl(int n_families)
2563 {
2564 	int i;
2565 
2566 	for (i = 0; i < n_families; i++)
2567 		genl_unregister_family(dp_genl_families[i]);
2568 }
2569 
2570 static int __init dp_register_genl(void)
2571 {
2572 	int err;
2573 	int i;
2574 
2575 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2576 
2577 		err = genl_register_family(dp_genl_families[i]);
2578 		if (err)
2579 			goto error;
2580 	}
2581 
2582 	return 0;
2583 
2584 error:
2585 	dp_unregister_genl(i);
2586 	return err;
2587 }
2588 
2589 static int __net_init ovs_init_net(struct net *net)
2590 {
2591 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2592 	int err;
2593 
2594 	INIT_LIST_HEAD(&ovs_net->dps);
2595 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2596 	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2597 
2598 	err = ovs_ct_init(net);
2599 	if (err)
2600 		return err;
2601 
2602 	schedule_delayed_work(&ovs_net->masks_rebalance,
2603 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2604 	return 0;
2605 }
2606 
2607 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2608 					    struct list_head *head)
2609 {
2610 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2611 	struct datapath *dp;
2612 
2613 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2614 		int i;
2615 
2616 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2617 			struct vport *vport;
2618 
2619 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2620 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2621 					continue;
2622 
2623 				if (dev_net(vport->dev) == dnet)
2624 					list_add(&vport->detach_list, head);
2625 			}
2626 		}
2627 	}
2628 }
2629 
2630 static void __net_exit ovs_exit_net(struct net *dnet)
2631 {
2632 	struct datapath *dp, *dp_next;
2633 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2634 	struct vport *vport, *vport_next;
2635 	struct net *net;
2636 	LIST_HEAD(head);
2637 
2638 	ovs_lock();
2639 
2640 	ovs_ct_exit(dnet);
2641 
2642 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2643 		__dp_destroy(dp);
2644 
2645 	down_read(&net_rwsem);
2646 	for_each_net(net)
2647 		list_vports_from_net(net, dnet, &head);
2648 	up_read(&net_rwsem);
2649 
2650 	/* Detach all vports from given namespace. */
2651 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2652 		list_del(&vport->detach_list);
2653 		ovs_dp_detach_port(vport);
2654 	}
2655 
2656 	ovs_unlock();
2657 
2658 	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2659 	cancel_work_sync(&ovs_net->dp_notify_work);
2660 }
2661 
2662 static struct pernet_operations ovs_net_ops = {
2663 	.init = ovs_init_net,
2664 	.exit = ovs_exit_net,
2665 	.id   = &ovs_net_id,
2666 	.size = sizeof(struct ovs_net),
2667 };
2668 
2669 static int __init dp_init(void)
2670 {
2671 	int err;
2672 
2673 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2674 		     sizeof_field(struct sk_buff, cb));
2675 
2676 	pr_info("Open vSwitch switching datapath\n");
2677 
2678 	err = action_fifos_init();
2679 	if (err)
2680 		goto error;
2681 
2682 	err = ovs_internal_dev_rtnl_link_register();
2683 	if (err)
2684 		goto error_action_fifos_exit;
2685 
2686 	err = ovs_flow_init();
2687 	if (err)
2688 		goto error_unreg_rtnl_link;
2689 
2690 	err = ovs_vport_init();
2691 	if (err)
2692 		goto error_flow_exit;
2693 
2694 	err = register_pernet_device(&ovs_net_ops);
2695 	if (err)
2696 		goto error_vport_exit;
2697 
2698 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2699 	if (err)
2700 		goto error_netns_exit;
2701 
2702 	err = ovs_netdev_init();
2703 	if (err)
2704 		goto error_unreg_notifier;
2705 
2706 	err = dp_register_genl();
2707 	if (err < 0)
2708 		goto error_unreg_netdev;
2709 
2710 	return 0;
2711 
2712 error_unreg_netdev:
2713 	ovs_netdev_exit();
2714 error_unreg_notifier:
2715 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2716 error_netns_exit:
2717 	unregister_pernet_device(&ovs_net_ops);
2718 error_vport_exit:
2719 	ovs_vport_exit();
2720 error_flow_exit:
2721 	ovs_flow_exit();
2722 error_unreg_rtnl_link:
2723 	ovs_internal_dev_rtnl_link_unregister();
2724 error_action_fifos_exit:
2725 	action_fifos_exit();
2726 error:
2727 	return err;
2728 }
2729 
2730 static void dp_cleanup(void)
2731 {
2732 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2733 	ovs_netdev_exit();
2734 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2735 	unregister_pernet_device(&ovs_net_ops);
2736 	rcu_barrier();
2737 	ovs_vport_exit();
2738 	ovs_flow_exit();
2739 	ovs_internal_dev_rtnl_link_unregister();
2740 	action_fifos_exit();
2741 }
2742 
2743 module_init(dp_init);
2744 module_exit(dp_cleanup);
2745 
2746 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2747 MODULE_LICENSE("GPL");
2748 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2749 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2750 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2751 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2752 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2753 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2754