xref: /openbmc/linux/net/openvswitch/datapath.c (revision 704438dd4f030c1b3d28a2a9c8f182c32d9b6bc4)
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 	.module = THIS_MODULE,
696 };
697 
698 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
699 			 struct ovs_dp_megaflow_stats *mega_stats)
700 {
701 	int i;
702 
703 	memset(mega_stats, 0, sizeof(*mega_stats));
704 
705 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
706 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
707 
708 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
709 
710 	for_each_possible_cpu(i) {
711 		const struct dp_stats_percpu *percpu_stats;
712 		struct dp_stats_percpu local_stats;
713 		unsigned int start;
714 
715 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
716 
717 		do {
718 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
719 			local_stats = *percpu_stats;
720 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
721 
722 		stats->n_hit += local_stats.n_hit;
723 		stats->n_missed += local_stats.n_missed;
724 		stats->n_lost += local_stats.n_lost;
725 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
726 		mega_stats->n_cache_hit += local_stats.n_cache_hit;
727 	}
728 }
729 
730 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
731 {
732 	return ovs_identifier_is_ufid(sfid) &&
733 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
734 }
735 
736 static bool should_fill_mask(uint32_t ufid_flags)
737 {
738 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
739 }
740 
741 static bool should_fill_actions(uint32_t ufid_flags)
742 {
743 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
744 }
745 
746 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
747 				    const struct sw_flow_id *sfid,
748 				    uint32_t ufid_flags)
749 {
750 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
751 
752 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
753 	 * see ovs_nla_put_identifier()
754 	 */
755 	if (sfid && ovs_identifier_is_ufid(sfid))
756 		len += nla_total_size(sfid->ufid_len);
757 	else
758 		len += nla_total_size(ovs_key_attr_size());
759 
760 	/* OVS_FLOW_ATTR_KEY */
761 	if (!sfid || should_fill_key(sfid, ufid_flags))
762 		len += nla_total_size(ovs_key_attr_size());
763 
764 	/* OVS_FLOW_ATTR_MASK */
765 	if (should_fill_mask(ufid_flags))
766 		len += nla_total_size(ovs_key_attr_size());
767 
768 	/* OVS_FLOW_ATTR_ACTIONS */
769 	if (should_fill_actions(ufid_flags))
770 		len += nla_total_size(acts->orig_len);
771 
772 	return len
773 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
774 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
775 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
776 }
777 
778 /* Called with ovs_mutex or RCU read lock. */
779 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
780 				   struct sk_buff *skb)
781 {
782 	struct ovs_flow_stats stats;
783 	__be16 tcp_flags;
784 	unsigned long used;
785 
786 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
787 
788 	if (used &&
789 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
790 			      OVS_FLOW_ATTR_PAD))
791 		return -EMSGSIZE;
792 
793 	if (stats.n_packets &&
794 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
795 			  sizeof(struct ovs_flow_stats), &stats,
796 			  OVS_FLOW_ATTR_PAD))
797 		return -EMSGSIZE;
798 
799 	if ((u8)ntohs(tcp_flags) &&
800 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
801 		return -EMSGSIZE;
802 
803 	return 0;
804 }
805 
806 /* Called with ovs_mutex or RCU read lock. */
807 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
808 				     struct sk_buff *skb, int skb_orig_len)
809 {
810 	struct nlattr *start;
811 	int err;
812 
813 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
814 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
815 	 * Netlink but individual action lists can be longer than
816 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
817 	 * The userspace caller can always fetch the actions separately if it
818 	 * really wants them.  (Most userspace callers in fact don't care.)
819 	 *
820 	 * This can only fail for dump operations because the skb is always
821 	 * properly sized for single flows.
822 	 */
823 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
824 	if (start) {
825 		const struct sw_flow_actions *sf_acts;
826 
827 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
828 		err = ovs_nla_put_actions(sf_acts->actions,
829 					  sf_acts->actions_len, skb);
830 
831 		if (!err)
832 			nla_nest_end(skb, start);
833 		else {
834 			if (skb_orig_len)
835 				return err;
836 
837 			nla_nest_cancel(skb, start);
838 		}
839 	} else if (skb_orig_len) {
840 		return -EMSGSIZE;
841 	}
842 
843 	return 0;
844 }
845 
846 /* Called with ovs_mutex or RCU read lock. */
847 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
848 				  struct sk_buff *skb, u32 portid,
849 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
850 {
851 	const int skb_orig_len = skb->len;
852 	struct ovs_header *ovs_header;
853 	int err;
854 
855 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
856 				 flags, cmd);
857 	if (!ovs_header)
858 		return -EMSGSIZE;
859 
860 	ovs_header->dp_ifindex = dp_ifindex;
861 
862 	err = ovs_nla_put_identifier(flow, skb);
863 	if (err)
864 		goto error;
865 
866 	if (should_fill_key(&flow->id, ufid_flags)) {
867 		err = ovs_nla_put_masked_key(flow, skb);
868 		if (err)
869 			goto error;
870 	}
871 
872 	if (should_fill_mask(ufid_flags)) {
873 		err = ovs_nla_put_mask(flow, skb);
874 		if (err)
875 			goto error;
876 	}
877 
878 	err = ovs_flow_cmd_fill_stats(flow, skb);
879 	if (err)
880 		goto error;
881 
882 	if (should_fill_actions(ufid_flags)) {
883 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
884 		if (err)
885 			goto error;
886 	}
887 
888 	genlmsg_end(skb, ovs_header);
889 	return 0;
890 
891 error:
892 	genlmsg_cancel(skb, ovs_header);
893 	return err;
894 }
895 
896 /* May not be called with RCU read lock. */
897 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
898 					       const struct sw_flow_id *sfid,
899 					       struct genl_info *info,
900 					       bool always,
901 					       uint32_t ufid_flags)
902 {
903 	struct sk_buff *skb;
904 	size_t len;
905 
906 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
907 		return NULL;
908 
909 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
910 	skb = genlmsg_new(len, GFP_KERNEL);
911 	if (!skb)
912 		return ERR_PTR(-ENOMEM);
913 
914 	return skb;
915 }
916 
917 /* Called with ovs_mutex. */
918 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
919 					       int dp_ifindex,
920 					       struct genl_info *info, u8 cmd,
921 					       bool always, u32 ufid_flags)
922 {
923 	struct sk_buff *skb;
924 	int retval;
925 
926 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
927 				      &flow->id, info, always, ufid_flags);
928 	if (IS_ERR_OR_NULL(skb))
929 		return skb;
930 
931 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
932 					info->snd_portid, info->snd_seq, 0,
933 					cmd, ufid_flags);
934 	if (WARN_ON_ONCE(retval < 0)) {
935 		kfree_skb(skb);
936 		skb = ERR_PTR(retval);
937 	}
938 	return skb;
939 }
940 
941 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
942 {
943 	struct net *net = sock_net(skb->sk);
944 	struct nlattr **a = info->attrs;
945 	struct ovs_header *ovs_header = info->userhdr;
946 	struct sw_flow *flow = NULL, *new_flow;
947 	struct sw_flow_mask mask;
948 	struct sk_buff *reply;
949 	struct datapath *dp;
950 	struct sw_flow_actions *acts;
951 	struct sw_flow_match match;
952 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
953 	int error;
954 	bool log = !a[OVS_FLOW_ATTR_PROBE];
955 
956 	/* Must have key and actions. */
957 	error = -EINVAL;
958 	if (!a[OVS_FLOW_ATTR_KEY]) {
959 		OVS_NLERR(log, "Flow key attr not present in new flow.");
960 		goto error;
961 	}
962 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
963 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
964 		goto error;
965 	}
966 
967 	/* Most of the time we need to allocate a new flow, do it before
968 	 * locking.
969 	 */
970 	new_flow = ovs_flow_alloc();
971 	if (IS_ERR(new_flow)) {
972 		error = PTR_ERR(new_flow);
973 		goto error;
974 	}
975 
976 	/* Extract key. */
977 	ovs_match_init(&match, &new_flow->key, false, &mask);
978 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
979 				  a[OVS_FLOW_ATTR_MASK], log);
980 	if (error)
981 		goto err_kfree_flow;
982 
983 	/* Extract flow identifier. */
984 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
985 				       &new_flow->key, log);
986 	if (error)
987 		goto err_kfree_flow;
988 
989 	/* unmasked key is needed to match when ufid is not used. */
990 	if (ovs_identifier_is_key(&new_flow->id))
991 		match.key = new_flow->id.unmasked_key;
992 
993 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
994 
995 	/* Validate actions. */
996 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
997 				     &new_flow->key, &acts, log);
998 	if (error) {
999 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1000 		goto err_kfree_flow;
1001 	}
1002 
1003 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1004 					ufid_flags);
1005 	if (IS_ERR(reply)) {
1006 		error = PTR_ERR(reply);
1007 		goto err_kfree_acts;
1008 	}
1009 
1010 	ovs_lock();
1011 	dp = get_dp(net, ovs_header->dp_ifindex);
1012 	if (unlikely(!dp)) {
1013 		error = -ENODEV;
1014 		goto err_unlock_ovs;
1015 	}
1016 
1017 	/* Check if this is a duplicate flow */
1018 	if (ovs_identifier_is_ufid(&new_flow->id))
1019 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1020 	if (!flow)
1021 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1022 	if (likely(!flow)) {
1023 		rcu_assign_pointer(new_flow->sf_acts, acts);
1024 
1025 		/* Put flow in bucket. */
1026 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1027 		if (unlikely(error)) {
1028 			acts = NULL;
1029 			goto err_unlock_ovs;
1030 		}
1031 
1032 		if (unlikely(reply)) {
1033 			error = ovs_flow_cmd_fill_info(new_flow,
1034 						       ovs_header->dp_ifindex,
1035 						       reply, info->snd_portid,
1036 						       info->snd_seq, 0,
1037 						       OVS_FLOW_CMD_NEW,
1038 						       ufid_flags);
1039 			BUG_ON(error < 0);
1040 		}
1041 		ovs_unlock();
1042 	} else {
1043 		struct sw_flow_actions *old_acts;
1044 
1045 		/* Bail out if we're not allowed to modify an existing flow.
1046 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1047 		 * because Generic Netlink treats the latter as a dump
1048 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1049 		 * gets fixed.
1050 		 */
1051 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1052 							 | NLM_F_EXCL))) {
1053 			error = -EEXIST;
1054 			goto err_unlock_ovs;
1055 		}
1056 		/* The flow identifier has to be the same for flow updates.
1057 		 * Look for any overlapping flow.
1058 		 */
1059 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1060 			if (ovs_identifier_is_key(&flow->id))
1061 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1062 								 &match);
1063 			else /* UFID matches but key is different */
1064 				flow = NULL;
1065 			if (!flow) {
1066 				error = -ENOENT;
1067 				goto err_unlock_ovs;
1068 			}
1069 		}
1070 		/* Update actions. */
1071 		old_acts = ovsl_dereference(flow->sf_acts);
1072 		rcu_assign_pointer(flow->sf_acts, acts);
1073 
1074 		if (unlikely(reply)) {
1075 			error = ovs_flow_cmd_fill_info(flow,
1076 						       ovs_header->dp_ifindex,
1077 						       reply, info->snd_portid,
1078 						       info->snd_seq, 0,
1079 						       OVS_FLOW_CMD_NEW,
1080 						       ufid_flags);
1081 			BUG_ON(error < 0);
1082 		}
1083 		ovs_unlock();
1084 
1085 		ovs_nla_free_flow_actions_rcu(old_acts);
1086 		ovs_flow_free(new_flow, false);
1087 	}
1088 
1089 	if (reply)
1090 		ovs_notify(&dp_flow_genl_family, reply, info);
1091 	return 0;
1092 
1093 err_unlock_ovs:
1094 	ovs_unlock();
1095 	kfree_skb(reply);
1096 err_kfree_acts:
1097 	ovs_nla_free_flow_actions(acts);
1098 err_kfree_flow:
1099 	ovs_flow_free(new_flow, false);
1100 error:
1101 	return error;
1102 }
1103 
1104 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1105 static noinline_for_stack
1106 struct sw_flow_actions *get_flow_actions(struct net *net,
1107 					 const struct nlattr *a,
1108 					 const struct sw_flow_key *key,
1109 					 const struct sw_flow_mask *mask,
1110 					 bool log)
1111 {
1112 	struct sw_flow_actions *acts;
1113 	struct sw_flow_key masked_key;
1114 	int error;
1115 
1116 	ovs_flow_mask_key(&masked_key, key, true, mask);
1117 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1118 	if (error) {
1119 		OVS_NLERR(log,
1120 			  "Actions may not be safe on all matching packets");
1121 		return ERR_PTR(error);
1122 	}
1123 
1124 	return acts;
1125 }
1126 
1127 /* Factor out match-init and action-copy to avoid
1128  * "Wframe-larger-than=1024" warning. Because mask is only
1129  * used to get actions, we new a function to save some
1130  * stack space.
1131  *
1132  * If there are not key and action attrs, we return 0
1133  * directly. In the case, the caller will also not use the
1134  * match as before. If there is action attr, we try to get
1135  * actions and save them to *acts. Before returning from
1136  * the function, we reset the match->mask pointer. Because
1137  * we should not to return match object with dangling reference
1138  * to mask.
1139  * */
1140 static noinline_for_stack int
1141 ovs_nla_init_match_and_action(struct net *net,
1142 			      struct sw_flow_match *match,
1143 			      struct sw_flow_key *key,
1144 			      struct nlattr **a,
1145 			      struct sw_flow_actions **acts,
1146 			      bool log)
1147 {
1148 	struct sw_flow_mask mask;
1149 	int error = 0;
1150 
1151 	if (a[OVS_FLOW_ATTR_KEY]) {
1152 		ovs_match_init(match, key, true, &mask);
1153 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1154 					  a[OVS_FLOW_ATTR_MASK], log);
1155 		if (error)
1156 			goto error;
1157 	}
1158 
1159 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1160 		if (!a[OVS_FLOW_ATTR_KEY]) {
1161 			OVS_NLERR(log,
1162 				  "Flow key attribute not present in set flow.");
1163 			error = -EINVAL;
1164 			goto error;
1165 		}
1166 
1167 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1168 					 &mask, log);
1169 		if (IS_ERR(*acts)) {
1170 			error = PTR_ERR(*acts);
1171 			goto error;
1172 		}
1173 	}
1174 
1175 	/* On success, error is 0. */
1176 error:
1177 	match->mask = NULL;
1178 	return error;
1179 }
1180 
1181 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1182 {
1183 	struct net *net = sock_net(skb->sk);
1184 	struct nlattr **a = info->attrs;
1185 	struct ovs_header *ovs_header = info->userhdr;
1186 	struct sw_flow_key key;
1187 	struct sw_flow *flow;
1188 	struct sk_buff *reply = NULL;
1189 	struct datapath *dp;
1190 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1191 	struct sw_flow_match match;
1192 	struct sw_flow_id sfid;
1193 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1194 	int error = 0;
1195 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1196 	bool ufid_present;
1197 
1198 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1199 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1200 		OVS_NLERR(log,
1201 			  "Flow set message rejected, Key attribute missing.");
1202 		return -EINVAL;
1203 	}
1204 
1205 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1206 					      &acts, log);
1207 	if (error)
1208 		goto error;
1209 
1210 	if (acts) {
1211 		/* Can allocate before locking if have acts. */
1212 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1213 						ufid_flags);
1214 		if (IS_ERR(reply)) {
1215 			error = PTR_ERR(reply);
1216 			goto err_kfree_acts;
1217 		}
1218 	}
1219 
1220 	ovs_lock();
1221 	dp = get_dp(net, ovs_header->dp_ifindex);
1222 	if (unlikely(!dp)) {
1223 		error = -ENODEV;
1224 		goto err_unlock_ovs;
1225 	}
1226 	/* Check that the flow exists. */
1227 	if (ufid_present)
1228 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1229 	else
1230 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1231 	if (unlikely(!flow)) {
1232 		error = -ENOENT;
1233 		goto err_unlock_ovs;
1234 	}
1235 
1236 	/* Update actions, if present. */
1237 	if (likely(acts)) {
1238 		old_acts = ovsl_dereference(flow->sf_acts);
1239 		rcu_assign_pointer(flow->sf_acts, acts);
1240 
1241 		if (unlikely(reply)) {
1242 			error = ovs_flow_cmd_fill_info(flow,
1243 						       ovs_header->dp_ifindex,
1244 						       reply, info->snd_portid,
1245 						       info->snd_seq, 0,
1246 						       OVS_FLOW_CMD_SET,
1247 						       ufid_flags);
1248 			BUG_ON(error < 0);
1249 		}
1250 	} else {
1251 		/* Could not alloc without acts before locking. */
1252 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1253 						info, OVS_FLOW_CMD_SET, false,
1254 						ufid_flags);
1255 
1256 		if (IS_ERR(reply)) {
1257 			error = PTR_ERR(reply);
1258 			goto err_unlock_ovs;
1259 		}
1260 	}
1261 
1262 	/* Clear stats. */
1263 	if (a[OVS_FLOW_ATTR_CLEAR])
1264 		ovs_flow_stats_clear(flow);
1265 	ovs_unlock();
1266 
1267 	if (reply)
1268 		ovs_notify(&dp_flow_genl_family, reply, info);
1269 	if (old_acts)
1270 		ovs_nla_free_flow_actions_rcu(old_acts);
1271 
1272 	return 0;
1273 
1274 err_unlock_ovs:
1275 	ovs_unlock();
1276 	kfree_skb(reply);
1277 err_kfree_acts:
1278 	ovs_nla_free_flow_actions(acts);
1279 error:
1280 	return error;
1281 }
1282 
1283 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1284 {
1285 	struct nlattr **a = info->attrs;
1286 	struct ovs_header *ovs_header = info->userhdr;
1287 	struct net *net = sock_net(skb->sk);
1288 	struct sw_flow_key key;
1289 	struct sk_buff *reply;
1290 	struct sw_flow *flow;
1291 	struct datapath *dp;
1292 	struct sw_flow_match match;
1293 	struct sw_flow_id ufid;
1294 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1295 	int err = 0;
1296 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1297 	bool ufid_present;
1298 
1299 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1300 	if (a[OVS_FLOW_ATTR_KEY]) {
1301 		ovs_match_init(&match, &key, true, NULL);
1302 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1303 					log);
1304 	} else if (!ufid_present) {
1305 		OVS_NLERR(log,
1306 			  "Flow get message rejected, Key attribute missing.");
1307 		err = -EINVAL;
1308 	}
1309 	if (err)
1310 		return err;
1311 
1312 	ovs_lock();
1313 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1314 	if (!dp) {
1315 		err = -ENODEV;
1316 		goto unlock;
1317 	}
1318 
1319 	if (ufid_present)
1320 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1321 	else
1322 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1323 	if (!flow) {
1324 		err = -ENOENT;
1325 		goto unlock;
1326 	}
1327 
1328 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1329 					OVS_FLOW_CMD_GET, true, ufid_flags);
1330 	if (IS_ERR(reply)) {
1331 		err = PTR_ERR(reply);
1332 		goto unlock;
1333 	}
1334 
1335 	ovs_unlock();
1336 	return genlmsg_reply(reply, info);
1337 unlock:
1338 	ovs_unlock();
1339 	return err;
1340 }
1341 
1342 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1343 {
1344 	struct nlattr **a = info->attrs;
1345 	struct ovs_header *ovs_header = info->userhdr;
1346 	struct net *net = sock_net(skb->sk);
1347 	struct sw_flow_key key;
1348 	struct sk_buff *reply;
1349 	struct sw_flow *flow = NULL;
1350 	struct datapath *dp;
1351 	struct sw_flow_match match;
1352 	struct sw_flow_id ufid;
1353 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1354 	int err;
1355 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1356 	bool ufid_present;
1357 
1358 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1359 	if (a[OVS_FLOW_ATTR_KEY]) {
1360 		ovs_match_init(&match, &key, true, NULL);
1361 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1362 					NULL, log);
1363 		if (unlikely(err))
1364 			return err;
1365 	}
1366 
1367 	ovs_lock();
1368 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1369 	if (unlikely(!dp)) {
1370 		err = -ENODEV;
1371 		goto unlock;
1372 	}
1373 
1374 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1375 		err = ovs_flow_tbl_flush(&dp->table);
1376 		goto unlock;
1377 	}
1378 
1379 	if (ufid_present)
1380 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1381 	else
1382 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1383 	if (unlikely(!flow)) {
1384 		err = -ENOENT;
1385 		goto unlock;
1386 	}
1387 
1388 	ovs_flow_tbl_remove(&dp->table, flow);
1389 	ovs_unlock();
1390 
1391 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1392 					&flow->id, info, false, ufid_flags);
1393 	if (likely(reply)) {
1394 		if (!IS_ERR(reply)) {
1395 			rcu_read_lock();	/*To keep RCU checker happy. */
1396 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1397 						     reply, info->snd_portid,
1398 						     info->snd_seq, 0,
1399 						     OVS_FLOW_CMD_DEL,
1400 						     ufid_flags);
1401 			rcu_read_unlock();
1402 			if (WARN_ON_ONCE(err < 0)) {
1403 				kfree_skb(reply);
1404 				goto out_free;
1405 			}
1406 
1407 			ovs_notify(&dp_flow_genl_family, reply, info);
1408 		} else {
1409 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1410 					PTR_ERR(reply));
1411 		}
1412 	}
1413 
1414 out_free:
1415 	ovs_flow_free(flow, true);
1416 	return 0;
1417 unlock:
1418 	ovs_unlock();
1419 	return err;
1420 }
1421 
1422 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1423 {
1424 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1425 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1426 	struct table_instance *ti;
1427 	struct datapath *dp;
1428 	u32 ufid_flags;
1429 	int err;
1430 
1431 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1432 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1433 	if (err)
1434 		return err;
1435 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1436 
1437 	rcu_read_lock();
1438 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1439 	if (!dp) {
1440 		rcu_read_unlock();
1441 		return -ENODEV;
1442 	}
1443 
1444 	ti = rcu_dereference(dp->table.ti);
1445 	for (;;) {
1446 		struct sw_flow *flow;
1447 		u32 bucket, obj;
1448 
1449 		bucket = cb->args[0];
1450 		obj = cb->args[1];
1451 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1452 		if (!flow)
1453 			break;
1454 
1455 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1456 					   NETLINK_CB(cb->skb).portid,
1457 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1458 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1459 			break;
1460 
1461 		cb->args[0] = bucket;
1462 		cb->args[1] = obj;
1463 	}
1464 	rcu_read_unlock();
1465 	return skb->len;
1466 }
1467 
1468 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1469 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1470 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1471 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1472 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1473 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1474 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1475 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1476 };
1477 
1478 static const struct genl_small_ops dp_flow_genl_ops[] = {
1479 	{ .cmd = OVS_FLOW_CMD_NEW,
1480 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1481 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1482 	  .doit = ovs_flow_cmd_new
1483 	},
1484 	{ .cmd = OVS_FLOW_CMD_DEL,
1485 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1486 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1487 	  .doit = ovs_flow_cmd_del
1488 	},
1489 	{ .cmd = OVS_FLOW_CMD_GET,
1490 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1491 	  .flags = 0,		    /* OK for unprivileged users. */
1492 	  .doit = ovs_flow_cmd_get,
1493 	  .dumpit = ovs_flow_cmd_dump
1494 	},
1495 	{ .cmd = OVS_FLOW_CMD_SET,
1496 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1497 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1498 	  .doit = ovs_flow_cmd_set,
1499 	},
1500 };
1501 
1502 static struct genl_family dp_flow_genl_family __ro_after_init = {
1503 	.hdrsize = sizeof(struct ovs_header),
1504 	.name = OVS_FLOW_FAMILY,
1505 	.version = OVS_FLOW_VERSION,
1506 	.maxattr = OVS_FLOW_ATTR_MAX,
1507 	.policy = flow_policy,
1508 	.netnsok = true,
1509 	.parallel_ops = true,
1510 	.small_ops = dp_flow_genl_ops,
1511 	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1512 	.mcgrps = &ovs_dp_flow_multicast_group,
1513 	.n_mcgrps = 1,
1514 	.module = THIS_MODULE,
1515 };
1516 
1517 static size_t ovs_dp_cmd_msg_size(void)
1518 {
1519 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1520 
1521 	msgsize += nla_total_size(IFNAMSIZ);
1522 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1523 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1524 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1525 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1526 
1527 	return msgsize;
1528 }
1529 
1530 /* Called with ovs_mutex. */
1531 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1532 				u32 portid, u32 seq, u32 flags, u8 cmd)
1533 {
1534 	struct ovs_header *ovs_header;
1535 	struct ovs_dp_stats dp_stats;
1536 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1537 	int err;
1538 
1539 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1540 				 flags, cmd);
1541 	if (!ovs_header)
1542 		goto error;
1543 
1544 	ovs_header->dp_ifindex = get_dpifindex(dp);
1545 
1546 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1547 	if (err)
1548 		goto nla_put_failure;
1549 
1550 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1551 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1552 			  &dp_stats, OVS_DP_ATTR_PAD))
1553 		goto nla_put_failure;
1554 
1555 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1556 			  sizeof(struct ovs_dp_megaflow_stats),
1557 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1558 		goto nla_put_failure;
1559 
1560 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1561 		goto nla_put_failure;
1562 
1563 	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1564 			ovs_flow_tbl_masks_cache_size(&dp->table)))
1565 		goto nla_put_failure;
1566 
1567 	genlmsg_end(skb, ovs_header);
1568 	return 0;
1569 
1570 nla_put_failure:
1571 	genlmsg_cancel(skb, ovs_header);
1572 error:
1573 	return -EMSGSIZE;
1574 }
1575 
1576 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1577 {
1578 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1579 }
1580 
1581 /* Called with rcu_read_lock or ovs_mutex. */
1582 static struct datapath *lookup_datapath(struct net *net,
1583 					const struct ovs_header *ovs_header,
1584 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1585 {
1586 	struct datapath *dp;
1587 
1588 	if (!a[OVS_DP_ATTR_NAME])
1589 		dp = get_dp(net, ovs_header->dp_ifindex);
1590 	else {
1591 		struct vport *vport;
1592 
1593 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1594 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1595 	}
1596 	return dp ? dp : ERR_PTR(-ENODEV);
1597 }
1598 
1599 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1600 				       struct genl_info *info)
1601 {
1602 	struct datapath *dp;
1603 
1604 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1605 			     info->attrs);
1606 	if (IS_ERR(dp))
1607 		return;
1608 
1609 	WARN(dp->user_features, "Dropping previously announced user features\n");
1610 	dp->user_features = 0;
1611 }
1612 
1613 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1614 			      const struct nlattr *ids)
1615 {
1616 	struct dp_nlsk_pids *old, *dp_nlsk_pids;
1617 
1618 	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1619 		return -EINVAL;
1620 
1621 	old = ovsl_dereference(dp->upcall_portids);
1622 
1623 	dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1624 			       GFP_KERNEL);
1625 	if (!dp_nlsk_pids)
1626 		return -ENOMEM;
1627 
1628 	dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1629 	nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1630 
1631 	rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1632 
1633 	kfree_rcu(old, rcu);
1634 
1635 	return 0;
1636 }
1637 
1638 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1639 {
1640 	struct dp_nlsk_pids *dp_nlsk_pids;
1641 
1642 	dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1643 
1644 	if (dp_nlsk_pids) {
1645 		if (cpu_id < dp_nlsk_pids->n_pids) {
1646 			return dp_nlsk_pids->pids[cpu_id];
1647 		} else if (dp_nlsk_pids->n_pids > 0 &&
1648 			   cpu_id >= dp_nlsk_pids->n_pids) {
1649 			/* If the number of netlink PIDs is mismatched with
1650 			 * the number of CPUs as seen by the kernel, log this
1651 			 * and send the upcall to an arbitrary socket (0) in
1652 			 * order to not drop packets
1653 			 */
1654 			pr_info_ratelimited("cpu_id mismatch with handler threads");
1655 			return dp_nlsk_pids->pids[cpu_id %
1656 						  dp_nlsk_pids->n_pids];
1657 		} else {
1658 			return 0;
1659 		}
1660 	} else {
1661 		return 0;
1662 	}
1663 }
1664 
1665 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1666 {
1667 	u32 user_features = 0, old_features = dp->user_features;
1668 	int err;
1669 
1670 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1671 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1672 
1673 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1674 				      OVS_DP_F_UNALIGNED |
1675 				      OVS_DP_F_TC_RECIRC_SHARING |
1676 				      OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1677 			return -EOPNOTSUPP;
1678 
1679 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1680 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1681 			return -EOPNOTSUPP;
1682 #endif
1683 	}
1684 
1685 	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1686 		int err;
1687 		u32 cache_size;
1688 
1689 		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1690 		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1691 		if (err)
1692 			return err;
1693 	}
1694 
1695 	dp->user_features = user_features;
1696 
1697 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1698 	    a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1699 		/* Upcall Netlink Port IDs have been updated */
1700 		err = ovs_dp_set_upcall_portids(dp,
1701 						a[OVS_DP_ATTR_PER_CPU_PIDS]);
1702 		if (err)
1703 			return err;
1704 	}
1705 
1706 	if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1707 	    !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1708 		tc_skb_ext_tc_enable();
1709 	else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1710 		 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1711 		tc_skb_ext_tc_disable();
1712 
1713 	return 0;
1714 }
1715 
1716 static int ovs_dp_stats_init(struct datapath *dp)
1717 {
1718 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1719 	if (!dp->stats_percpu)
1720 		return -ENOMEM;
1721 
1722 	return 0;
1723 }
1724 
1725 static int ovs_dp_vport_init(struct datapath *dp)
1726 {
1727 	int i;
1728 
1729 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1730 				  sizeof(struct hlist_head),
1731 				  GFP_KERNEL);
1732 	if (!dp->ports)
1733 		return -ENOMEM;
1734 
1735 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1736 		INIT_HLIST_HEAD(&dp->ports[i]);
1737 
1738 	return 0;
1739 }
1740 
1741 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1742 {
1743 	struct nlattr **a = info->attrs;
1744 	struct vport_parms parms;
1745 	struct sk_buff *reply;
1746 	struct datapath *dp;
1747 	struct vport *vport;
1748 	struct ovs_net *ovs_net;
1749 	int err;
1750 
1751 	err = -EINVAL;
1752 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1753 		goto err;
1754 
1755 	reply = ovs_dp_cmd_alloc_info();
1756 	if (!reply)
1757 		return -ENOMEM;
1758 
1759 	err = -ENOMEM;
1760 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1761 	if (dp == NULL)
1762 		goto err_destroy_reply;
1763 
1764 	ovs_dp_set_net(dp, sock_net(skb->sk));
1765 
1766 	/* Allocate table. */
1767 	err = ovs_flow_tbl_init(&dp->table);
1768 	if (err)
1769 		goto err_destroy_dp;
1770 
1771 	err = ovs_dp_stats_init(dp);
1772 	if (err)
1773 		goto err_destroy_table;
1774 
1775 	err = ovs_dp_vport_init(dp);
1776 	if (err)
1777 		goto err_destroy_stats;
1778 
1779 	err = ovs_meters_init(dp);
1780 	if (err)
1781 		goto err_destroy_ports;
1782 
1783 	/* Set up our datapath device. */
1784 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1785 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1786 	parms.options = NULL;
1787 	parms.dp = dp;
1788 	parms.port_no = OVSP_LOCAL;
1789 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1790 
1791 	/* So far only local changes have been made, now need the lock. */
1792 	ovs_lock();
1793 
1794 	err = ovs_dp_change(dp, a);
1795 	if (err)
1796 		goto err_unlock_and_destroy_meters;
1797 
1798 	vport = new_vport(&parms);
1799 	if (IS_ERR(vport)) {
1800 		err = PTR_ERR(vport);
1801 		if (err == -EBUSY)
1802 			err = -EEXIST;
1803 
1804 		if (err == -EEXIST) {
1805 			/* An outdated user space instance that does not understand
1806 			 * the concept of user_features has attempted to create a new
1807 			 * datapath and is likely to reuse it. Drop all user features.
1808 			 */
1809 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1810 				ovs_dp_reset_user_features(skb, info);
1811 		}
1812 
1813 		goto err_unlock_and_destroy_meters;
1814 	}
1815 
1816 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1817 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1818 	BUG_ON(err < 0);
1819 
1820 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1821 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1822 
1823 	ovs_unlock();
1824 
1825 	ovs_notify(&dp_datapath_genl_family, reply, info);
1826 	return 0;
1827 
1828 err_unlock_and_destroy_meters:
1829 	ovs_unlock();
1830 	ovs_meters_exit(dp);
1831 err_destroy_ports:
1832 	kfree(dp->ports);
1833 err_destroy_stats:
1834 	free_percpu(dp->stats_percpu);
1835 err_destroy_table:
1836 	ovs_flow_tbl_destroy(&dp->table);
1837 err_destroy_dp:
1838 	kfree(dp);
1839 err_destroy_reply:
1840 	kfree_skb(reply);
1841 err:
1842 	return err;
1843 }
1844 
1845 /* Called with ovs_mutex. */
1846 static void __dp_destroy(struct datapath *dp)
1847 {
1848 	struct flow_table *table = &dp->table;
1849 	int i;
1850 
1851 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1852 		tc_skb_ext_tc_disable();
1853 
1854 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1855 		struct vport *vport;
1856 		struct hlist_node *n;
1857 
1858 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1859 			if (vport->port_no != OVSP_LOCAL)
1860 				ovs_dp_detach_port(vport);
1861 	}
1862 
1863 	list_del_rcu(&dp->list_node);
1864 
1865 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1866 	 * all ports in datapath are destroyed first before freeing datapath.
1867 	 */
1868 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1869 
1870 	/* Flush sw_flow in the tables. RCU cb only releases resource
1871 	 * such as dp, ports and tables. That may avoid some issues
1872 	 * such as RCU usage warning.
1873 	 */
1874 	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1875 				  ovsl_dereference(table->ufid_ti));
1876 
1877 	/* RCU destroy the ports, meters and flow tables. */
1878 	call_rcu(&dp->rcu, destroy_dp_rcu);
1879 }
1880 
1881 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1882 {
1883 	struct sk_buff *reply;
1884 	struct datapath *dp;
1885 	int err;
1886 
1887 	reply = ovs_dp_cmd_alloc_info();
1888 	if (!reply)
1889 		return -ENOMEM;
1890 
1891 	ovs_lock();
1892 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1893 	err = PTR_ERR(dp);
1894 	if (IS_ERR(dp))
1895 		goto err_unlock_free;
1896 
1897 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1898 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1899 	BUG_ON(err < 0);
1900 
1901 	__dp_destroy(dp);
1902 	ovs_unlock();
1903 
1904 	ovs_notify(&dp_datapath_genl_family, reply, info);
1905 
1906 	return 0;
1907 
1908 err_unlock_free:
1909 	ovs_unlock();
1910 	kfree_skb(reply);
1911 	return err;
1912 }
1913 
1914 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1915 {
1916 	struct sk_buff *reply;
1917 	struct datapath *dp;
1918 	int err;
1919 
1920 	reply = ovs_dp_cmd_alloc_info();
1921 	if (!reply)
1922 		return -ENOMEM;
1923 
1924 	ovs_lock();
1925 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1926 	err = PTR_ERR(dp);
1927 	if (IS_ERR(dp))
1928 		goto err_unlock_free;
1929 
1930 	err = ovs_dp_change(dp, info->attrs);
1931 	if (err)
1932 		goto err_unlock_free;
1933 
1934 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1935 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1936 	BUG_ON(err < 0);
1937 
1938 	ovs_unlock();
1939 	ovs_notify(&dp_datapath_genl_family, reply, info);
1940 
1941 	return 0;
1942 
1943 err_unlock_free:
1944 	ovs_unlock();
1945 	kfree_skb(reply);
1946 	return err;
1947 }
1948 
1949 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1950 {
1951 	struct sk_buff *reply;
1952 	struct datapath *dp;
1953 	int err;
1954 
1955 	reply = ovs_dp_cmd_alloc_info();
1956 	if (!reply)
1957 		return -ENOMEM;
1958 
1959 	ovs_lock();
1960 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1961 	if (IS_ERR(dp)) {
1962 		err = PTR_ERR(dp);
1963 		goto err_unlock_free;
1964 	}
1965 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1966 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1967 	BUG_ON(err < 0);
1968 	ovs_unlock();
1969 
1970 	return genlmsg_reply(reply, info);
1971 
1972 err_unlock_free:
1973 	ovs_unlock();
1974 	kfree_skb(reply);
1975 	return err;
1976 }
1977 
1978 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1979 {
1980 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1981 	struct datapath *dp;
1982 	int skip = cb->args[0];
1983 	int i = 0;
1984 
1985 	ovs_lock();
1986 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1987 		if (i >= skip &&
1988 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1989 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1990 					 OVS_DP_CMD_GET) < 0)
1991 			break;
1992 		i++;
1993 	}
1994 	ovs_unlock();
1995 
1996 	cb->args[0] = i;
1997 
1998 	return skb->len;
1999 }
2000 
2001 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2002 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2003 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2004 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2005 	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
2006 		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2007 };
2008 
2009 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2010 	{ .cmd = OVS_DP_CMD_NEW,
2011 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2012 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2013 	  .doit = ovs_dp_cmd_new
2014 	},
2015 	{ .cmd = OVS_DP_CMD_DEL,
2016 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2017 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2018 	  .doit = ovs_dp_cmd_del
2019 	},
2020 	{ .cmd = OVS_DP_CMD_GET,
2021 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2022 	  .flags = 0,		    /* OK for unprivileged users. */
2023 	  .doit = ovs_dp_cmd_get,
2024 	  .dumpit = ovs_dp_cmd_dump
2025 	},
2026 	{ .cmd = OVS_DP_CMD_SET,
2027 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2028 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2029 	  .doit = ovs_dp_cmd_set,
2030 	},
2031 };
2032 
2033 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2034 	.hdrsize = sizeof(struct ovs_header),
2035 	.name = OVS_DATAPATH_FAMILY,
2036 	.version = OVS_DATAPATH_VERSION,
2037 	.maxattr = OVS_DP_ATTR_MAX,
2038 	.policy = datapath_policy,
2039 	.netnsok = true,
2040 	.parallel_ops = true,
2041 	.small_ops = dp_datapath_genl_ops,
2042 	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2043 	.mcgrps = &ovs_dp_datapath_multicast_group,
2044 	.n_mcgrps = 1,
2045 	.module = THIS_MODULE,
2046 };
2047 
2048 /* Called with ovs_mutex or RCU read lock. */
2049 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2050 				   struct net *net, u32 portid, u32 seq,
2051 				   u32 flags, u8 cmd, gfp_t gfp)
2052 {
2053 	struct ovs_header *ovs_header;
2054 	struct ovs_vport_stats vport_stats;
2055 	int err;
2056 
2057 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2058 				 flags, cmd);
2059 	if (!ovs_header)
2060 		return -EMSGSIZE;
2061 
2062 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2063 
2064 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2065 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2066 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2067 			   ovs_vport_name(vport)) ||
2068 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2069 		goto nla_put_failure;
2070 
2071 	if (!net_eq(net, dev_net(vport->dev))) {
2072 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2073 
2074 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2075 			goto nla_put_failure;
2076 	}
2077 
2078 	ovs_vport_get_stats(vport, &vport_stats);
2079 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2080 			  sizeof(struct ovs_vport_stats), &vport_stats,
2081 			  OVS_VPORT_ATTR_PAD))
2082 		goto nla_put_failure;
2083 
2084 	if (ovs_vport_get_upcall_portids(vport, skb))
2085 		goto nla_put_failure;
2086 
2087 	err = ovs_vport_get_options(vport, skb);
2088 	if (err == -EMSGSIZE)
2089 		goto error;
2090 
2091 	genlmsg_end(skb, ovs_header);
2092 	return 0;
2093 
2094 nla_put_failure:
2095 	err = -EMSGSIZE;
2096 error:
2097 	genlmsg_cancel(skb, ovs_header);
2098 	return err;
2099 }
2100 
2101 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2102 {
2103 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2104 }
2105 
2106 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2107 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2108 					 u32 portid, u32 seq, u8 cmd)
2109 {
2110 	struct sk_buff *skb;
2111 	int retval;
2112 
2113 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2114 	if (!skb)
2115 		return ERR_PTR(-ENOMEM);
2116 
2117 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2118 					 GFP_KERNEL);
2119 	BUG_ON(retval < 0);
2120 
2121 	return skb;
2122 }
2123 
2124 /* Called with ovs_mutex or RCU read lock. */
2125 static struct vport *lookup_vport(struct net *net,
2126 				  const struct ovs_header *ovs_header,
2127 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2128 {
2129 	struct datapath *dp;
2130 	struct vport *vport;
2131 
2132 	if (a[OVS_VPORT_ATTR_IFINDEX])
2133 		return ERR_PTR(-EOPNOTSUPP);
2134 	if (a[OVS_VPORT_ATTR_NAME]) {
2135 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2136 		if (!vport)
2137 			return ERR_PTR(-ENODEV);
2138 		if (ovs_header->dp_ifindex &&
2139 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2140 			return ERR_PTR(-ENODEV);
2141 		return vport;
2142 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2143 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2144 
2145 		if (port_no >= DP_MAX_PORTS)
2146 			return ERR_PTR(-EFBIG);
2147 
2148 		dp = get_dp(net, ovs_header->dp_ifindex);
2149 		if (!dp)
2150 			return ERR_PTR(-ENODEV);
2151 
2152 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2153 		if (!vport)
2154 			return ERR_PTR(-ENODEV);
2155 		return vport;
2156 	} else
2157 		return ERR_PTR(-EINVAL);
2158 
2159 }
2160 
2161 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2162 {
2163 	unsigned int dev_headroom, max_headroom = 0;
2164 	struct net_device *dev;
2165 	struct vport *vport;
2166 	int i;
2167 
2168 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2169 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2170 					 lockdep_ovsl_is_held()) {
2171 			dev = vport->dev;
2172 			dev_headroom = netdev_get_fwd_headroom(dev);
2173 			if (dev_headroom > max_headroom)
2174 				max_headroom = dev_headroom;
2175 		}
2176 	}
2177 
2178 	return max_headroom;
2179 }
2180 
2181 /* Called with ovs_mutex */
2182 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2183 {
2184 	struct vport *vport;
2185 	int i;
2186 
2187 	dp->max_headroom = new_headroom;
2188 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2189 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2190 					 lockdep_ovsl_is_held())
2191 			netdev_set_rx_headroom(vport->dev, new_headroom);
2192 	}
2193 }
2194 
2195 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2196 {
2197 	struct nlattr **a = info->attrs;
2198 	struct ovs_header *ovs_header = info->userhdr;
2199 	struct vport_parms parms;
2200 	struct sk_buff *reply;
2201 	struct vport *vport;
2202 	struct datapath *dp;
2203 	unsigned int new_headroom;
2204 	u32 port_no;
2205 	int err;
2206 
2207 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2208 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2209 		return -EINVAL;
2210 	if (a[OVS_VPORT_ATTR_IFINDEX])
2211 		return -EOPNOTSUPP;
2212 
2213 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2214 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2215 	if (port_no >= DP_MAX_PORTS)
2216 		return -EFBIG;
2217 
2218 	reply = ovs_vport_cmd_alloc_info();
2219 	if (!reply)
2220 		return -ENOMEM;
2221 
2222 	ovs_lock();
2223 restart:
2224 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2225 	err = -ENODEV;
2226 	if (!dp)
2227 		goto exit_unlock_free;
2228 
2229 	if (port_no) {
2230 		vport = ovs_vport_ovsl(dp, port_no);
2231 		err = -EBUSY;
2232 		if (vport)
2233 			goto exit_unlock_free;
2234 	} else {
2235 		for (port_no = 1; ; port_no++) {
2236 			if (port_no >= DP_MAX_PORTS) {
2237 				err = -EFBIG;
2238 				goto exit_unlock_free;
2239 			}
2240 			vport = ovs_vport_ovsl(dp, port_no);
2241 			if (!vport)
2242 				break;
2243 		}
2244 	}
2245 
2246 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2247 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2248 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2249 	parms.dp = dp;
2250 	parms.port_no = port_no;
2251 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2252 
2253 	vport = new_vport(&parms);
2254 	err = PTR_ERR(vport);
2255 	if (IS_ERR(vport)) {
2256 		if (err == -EAGAIN)
2257 			goto restart;
2258 		goto exit_unlock_free;
2259 	}
2260 
2261 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2262 				      info->snd_portid, info->snd_seq, 0,
2263 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2264 
2265 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2266 
2267 	if (new_headroom > dp->max_headroom)
2268 		ovs_update_headroom(dp, new_headroom);
2269 	else
2270 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2271 
2272 	BUG_ON(err < 0);
2273 	ovs_unlock();
2274 
2275 	ovs_notify(&dp_vport_genl_family, reply, info);
2276 	return 0;
2277 
2278 exit_unlock_free:
2279 	ovs_unlock();
2280 	kfree_skb(reply);
2281 	return err;
2282 }
2283 
2284 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2285 {
2286 	struct nlattr **a = info->attrs;
2287 	struct sk_buff *reply;
2288 	struct vport *vport;
2289 	int err;
2290 
2291 	reply = ovs_vport_cmd_alloc_info();
2292 	if (!reply)
2293 		return -ENOMEM;
2294 
2295 	ovs_lock();
2296 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2297 	err = PTR_ERR(vport);
2298 	if (IS_ERR(vport))
2299 		goto exit_unlock_free;
2300 
2301 	if (a[OVS_VPORT_ATTR_TYPE] &&
2302 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2303 		err = -EINVAL;
2304 		goto exit_unlock_free;
2305 	}
2306 
2307 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2308 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2309 		if (err)
2310 			goto exit_unlock_free;
2311 	}
2312 
2313 
2314 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2315 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2316 
2317 		err = ovs_vport_set_upcall_portids(vport, ids);
2318 		if (err)
2319 			goto exit_unlock_free;
2320 	}
2321 
2322 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2323 				      info->snd_portid, info->snd_seq, 0,
2324 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2325 	BUG_ON(err < 0);
2326 
2327 	ovs_unlock();
2328 	ovs_notify(&dp_vport_genl_family, reply, info);
2329 	return 0;
2330 
2331 exit_unlock_free:
2332 	ovs_unlock();
2333 	kfree_skb(reply);
2334 	return err;
2335 }
2336 
2337 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2338 {
2339 	bool update_headroom = false;
2340 	struct nlattr **a = info->attrs;
2341 	struct sk_buff *reply;
2342 	struct datapath *dp;
2343 	struct vport *vport;
2344 	unsigned int new_headroom;
2345 	int err;
2346 
2347 	reply = ovs_vport_cmd_alloc_info();
2348 	if (!reply)
2349 		return -ENOMEM;
2350 
2351 	ovs_lock();
2352 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2353 	err = PTR_ERR(vport);
2354 	if (IS_ERR(vport))
2355 		goto exit_unlock_free;
2356 
2357 	if (vport->port_no == OVSP_LOCAL) {
2358 		err = -EINVAL;
2359 		goto exit_unlock_free;
2360 	}
2361 
2362 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2363 				      info->snd_portid, info->snd_seq, 0,
2364 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2365 	BUG_ON(err < 0);
2366 
2367 	/* the vport deletion may trigger dp headroom update */
2368 	dp = vport->dp;
2369 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2370 		update_headroom = true;
2371 
2372 	netdev_reset_rx_headroom(vport->dev);
2373 	ovs_dp_detach_port(vport);
2374 
2375 	if (update_headroom) {
2376 		new_headroom = ovs_get_max_headroom(dp);
2377 
2378 		if (new_headroom < dp->max_headroom)
2379 			ovs_update_headroom(dp, new_headroom);
2380 	}
2381 	ovs_unlock();
2382 
2383 	ovs_notify(&dp_vport_genl_family, reply, info);
2384 	return 0;
2385 
2386 exit_unlock_free:
2387 	ovs_unlock();
2388 	kfree_skb(reply);
2389 	return err;
2390 }
2391 
2392 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2393 {
2394 	struct nlattr **a = info->attrs;
2395 	struct ovs_header *ovs_header = info->userhdr;
2396 	struct sk_buff *reply;
2397 	struct vport *vport;
2398 	int err;
2399 
2400 	reply = ovs_vport_cmd_alloc_info();
2401 	if (!reply)
2402 		return -ENOMEM;
2403 
2404 	rcu_read_lock();
2405 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2406 	err = PTR_ERR(vport);
2407 	if (IS_ERR(vport))
2408 		goto exit_unlock_free;
2409 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2410 				      info->snd_portid, info->snd_seq, 0,
2411 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2412 	BUG_ON(err < 0);
2413 	rcu_read_unlock();
2414 
2415 	return genlmsg_reply(reply, info);
2416 
2417 exit_unlock_free:
2418 	rcu_read_unlock();
2419 	kfree_skb(reply);
2420 	return err;
2421 }
2422 
2423 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2424 {
2425 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2426 	struct datapath *dp;
2427 	int bucket = cb->args[0], skip = cb->args[1];
2428 	int i, j = 0;
2429 
2430 	rcu_read_lock();
2431 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2432 	if (!dp) {
2433 		rcu_read_unlock();
2434 		return -ENODEV;
2435 	}
2436 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2437 		struct vport *vport;
2438 
2439 		j = 0;
2440 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2441 			if (j >= skip &&
2442 			    ovs_vport_cmd_fill_info(vport, skb,
2443 						    sock_net(skb->sk),
2444 						    NETLINK_CB(cb->skb).portid,
2445 						    cb->nlh->nlmsg_seq,
2446 						    NLM_F_MULTI,
2447 						    OVS_VPORT_CMD_GET,
2448 						    GFP_ATOMIC) < 0)
2449 				goto out;
2450 
2451 			j++;
2452 		}
2453 		skip = 0;
2454 	}
2455 out:
2456 	rcu_read_unlock();
2457 
2458 	cb->args[0] = i;
2459 	cb->args[1] = j;
2460 
2461 	return skb->len;
2462 }
2463 
2464 static void ovs_dp_masks_rebalance(struct work_struct *work)
2465 {
2466 	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2467 					       masks_rebalance.work);
2468 	struct datapath *dp;
2469 
2470 	ovs_lock();
2471 
2472 	list_for_each_entry(dp, &ovs_net->dps, list_node)
2473 		ovs_flow_masks_rebalance(&dp->table);
2474 
2475 	ovs_unlock();
2476 
2477 	schedule_delayed_work(&ovs_net->masks_rebalance,
2478 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2479 }
2480 
2481 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2482 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2483 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2484 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2485 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2486 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2487 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2488 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2489 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2490 };
2491 
2492 static const struct genl_small_ops dp_vport_genl_ops[] = {
2493 	{ .cmd = OVS_VPORT_CMD_NEW,
2494 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2495 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2496 	  .doit = ovs_vport_cmd_new
2497 	},
2498 	{ .cmd = OVS_VPORT_CMD_DEL,
2499 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2500 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2501 	  .doit = ovs_vport_cmd_del
2502 	},
2503 	{ .cmd = OVS_VPORT_CMD_GET,
2504 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2505 	  .flags = 0,		    /* OK for unprivileged users. */
2506 	  .doit = ovs_vport_cmd_get,
2507 	  .dumpit = ovs_vport_cmd_dump
2508 	},
2509 	{ .cmd = OVS_VPORT_CMD_SET,
2510 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2511 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2512 	  .doit = ovs_vport_cmd_set,
2513 	},
2514 };
2515 
2516 struct genl_family dp_vport_genl_family __ro_after_init = {
2517 	.hdrsize = sizeof(struct ovs_header),
2518 	.name = OVS_VPORT_FAMILY,
2519 	.version = OVS_VPORT_VERSION,
2520 	.maxattr = OVS_VPORT_ATTR_MAX,
2521 	.policy = vport_policy,
2522 	.netnsok = true,
2523 	.parallel_ops = true,
2524 	.small_ops = dp_vport_genl_ops,
2525 	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2526 	.mcgrps = &ovs_dp_vport_multicast_group,
2527 	.n_mcgrps = 1,
2528 	.module = THIS_MODULE,
2529 };
2530 
2531 static struct genl_family * const dp_genl_families[] = {
2532 	&dp_datapath_genl_family,
2533 	&dp_vport_genl_family,
2534 	&dp_flow_genl_family,
2535 	&dp_packet_genl_family,
2536 	&dp_meter_genl_family,
2537 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2538 	&dp_ct_limit_genl_family,
2539 #endif
2540 };
2541 
2542 static void dp_unregister_genl(int n_families)
2543 {
2544 	int i;
2545 
2546 	for (i = 0; i < n_families; i++)
2547 		genl_unregister_family(dp_genl_families[i]);
2548 }
2549 
2550 static int __init dp_register_genl(void)
2551 {
2552 	int err;
2553 	int i;
2554 
2555 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2556 
2557 		err = genl_register_family(dp_genl_families[i]);
2558 		if (err)
2559 			goto error;
2560 	}
2561 
2562 	return 0;
2563 
2564 error:
2565 	dp_unregister_genl(i);
2566 	return err;
2567 }
2568 
2569 static int __net_init ovs_init_net(struct net *net)
2570 {
2571 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2572 	int err;
2573 
2574 	INIT_LIST_HEAD(&ovs_net->dps);
2575 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2576 	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2577 
2578 	err = ovs_ct_init(net);
2579 	if (err)
2580 		return err;
2581 
2582 	schedule_delayed_work(&ovs_net->masks_rebalance,
2583 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2584 	return 0;
2585 }
2586 
2587 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2588 					    struct list_head *head)
2589 {
2590 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2591 	struct datapath *dp;
2592 
2593 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2594 		int i;
2595 
2596 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2597 			struct vport *vport;
2598 
2599 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2600 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2601 					continue;
2602 
2603 				if (dev_net(vport->dev) == dnet)
2604 					list_add(&vport->detach_list, head);
2605 			}
2606 		}
2607 	}
2608 }
2609 
2610 static void __net_exit ovs_exit_net(struct net *dnet)
2611 {
2612 	struct datapath *dp, *dp_next;
2613 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2614 	struct vport *vport, *vport_next;
2615 	struct net *net;
2616 	LIST_HEAD(head);
2617 
2618 	ovs_lock();
2619 
2620 	ovs_ct_exit(dnet);
2621 
2622 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2623 		__dp_destroy(dp);
2624 
2625 	down_read(&net_rwsem);
2626 	for_each_net(net)
2627 		list_vports_from_net(net, dnet, &head);
2628 	up_read(&net_rwsem);
2629 
2630 	/* Detach all vports from given namespace. */
2631 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2632 		list_del(&vport->detach_list);
2633 		ovs_dp_detach_port(vport);
2634 	}
2635 
2636 	ovs_unlock();
2637 
2638 	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2639 	cancel_work_sync(&ovs_net->dp_notify_work);
2640 }
2641 
2642 static struct pernet_operations ovs_net_ops = {
2643 	.init = ovs_init_net,
2644 	.exit = ovs_exit_net,
2645 	.id   = &ovs_net_id,
2646 	.size = sizeof(struct ovs_net),
2647 };
2648 
2649 static int __init dp_init(void)
2650 {
2651 	int err;
2652 
2653 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2654 		     sizeof_field(struct sk_buff, cb));
2655 
2656 	pr_info("Open vSwitch switching datapath\n");
2657 
2658 	err = action_fifos_init();
2659 	if (err)
2660 		goto error;
2661 
2662 	err = ovs_internal_dev_rtnl_link_register();
2663 	if (err)
2664 		goto error_action_fifos_exit;
2665 
2666 	err = ovs_flow_init();
2667 	if (err)
2668 		goto error_unreg_rtnl_link;
2669 
2670 	err = ovs_vport_init();
2671 	if (err)
2672 		goto error_flow_exit;
2673 
2674 	err = register_pernet_device(&ovs_net_ops);
2675 	if (err)
2676 		goto error_vport_exit;
2677 
2678 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2679 	if (err)
2680 		goto error_netns_exit;
2681 
2682 	err = ovs_netdev_init();
2683 	if (err)
2684 		goto error_unreg_notifier;
2685 
2686 	err = dp_register_genl();
2687 	if (err < 0)
2688 		goto error_unreg_netdev;
2689 
2690 	return 0;
2691 
2692 error_unreg_netdev:
2693 	ovs_netdev_exit();
2694 error_unreg_notifier:
2695 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2696 error_netns_exit:
2697 	unregister_pernet_device(&ovs_net_ops);
2698 error_vport_exit:
2699 	ovs_vport_exit();
2700 error_flow_exit:
2701 	ovs_flow_exit();
2702 error_unreg_rtnl_link:
2703 	ovs_internal_dev_rtnl_link_unregister();
2704 error_action_fifos_exit:
2705 	action_fifos_exit();
2706 error:
2707 	return err;
2708 }
2709 
2710 static void dp_cleanup(void)
2711 {
2712 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2713 	ovs_netdev_exit();
2714 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2715 	unregister_pernet_device(&ovs_net_ops);
2716 	rcu_barrier();
2717 	ovs_vport_exit();
2718 	ovs_flow_exit();
2719 	ovs_internal_dev_rtnl_link_unregister();
2720 	action_fifos_exit();
2721 }
2722 
2723 module_init(dp_init);
2724 module_exit(dp_cleanup);
2725 
2726 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2727 MODULE_LICENSE("GPL");
2728 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2729 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2730 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2731 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2732 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2733 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2734