xref: /openbmc/linux/net/openvswitch/datapath.c (revision 5d0e4d78)
1 /*
2  * Copyright (c) 2007-2014 Nicira, Inc.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program; if not, write to the Free Software
15  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
16  * 02110-1301, USA
17  */
18 
19 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 
21 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/if_arp.h>
24 #include <linux/if_vlan.h>
25 #include <linux/in.h>
26 #include <linux/ip.h>
27 #include <linux/jhash.h>
28 #include <linux/delay.h>
29 #include <linux/time.h>
30 #include <linux/etherdevice.h>
31 #include <linux/genetlink.h>
32 #include <linux/kernel.h>
33 #include <linux/kthread.h>
34 #include <linux/mutex.h>
35 #include <linux/percpu.h>
36 #include <linux/rcupdate.h>
37 #include <linux/tcp.h>
38 #include <linux/udp.h>
39 #include <linux/ethtool.h>
40 #include <linux/wait.h>
41 #include <asm/div64.h>
42 #include <linux/highmem.h>
43 #include <linux/netfilter_bridge.h>
44 #include <linux/netfilter_ipv4.h>
45 #include <linux/inetdevice.h>
46 #include <linux/list.h>
47 #include <linux/openvswitch.h>
48 #include <linux/rculist.h>
49 #include <linux/dmi.h>
50 #include <net/genetlink.h>
51 #include <net/net_namespace.h>
52 #include <net/netns/generic.h>
53 
54 #include "datapath.h"
55 #include "flow.h"
56 #include "flow_table.h"
57 #include "flow_netlink.h"
58 #include "vport-internal_dev.h"
59 #include "vport-netdev.h"
60 
61 unsigned int ovs_net_id __read_mostly;
62 
63 static struct genl_family dp_packet_genl_family;
64 static struct genl_family dp_flow_genl_family;
65 static struct genl_family dp_datapath_genl_family;
66 
67 static const struct nla_policy flow_policy[];
68 
69 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
70 	.name = OVS_FLOW_MCGROUP,
71 };
72 
73 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
74 	.name = OVS_DATAPATH_MCGROUP,
75 };
76 
77 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
78 	.name = OVS_VPORT_MCGROUP,
79 };
80 
81 /* Check if need to build a reply message.
82  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
83 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
84 			    unsigned int group)
85 {
86 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
87 	       genl_has_listeners(family, genl_info_net(info), group);
88 }
89 
90 static void ovs_notify(struct genl_family *family,
91 		       struct sk_buff *skb, struct genl_info *info)
92 {
93 	genl_notify(family, skb, info, 0, GFP_KERNEL);
94 }
95 
96 /**
97  * DOC: Locking:
98  *
99  * All writes e.g. Writes to device state (add/remove datapath, port, set
100  * operations on vports, etc.), Writes to other state (flow table
101  * modifications, set miscellaneous datapath parameters, etc.) are protected
102  * by ovs_lock.
103  *
104  * Reads are protected by RCU.
105  *
106  * There are a few special cases (mostly stats) that have their own
107  * synchronization but they nest under all of above and don't interact with
108  * each other.
109  *
110  * The RTNL lock nests inside ovs_mutex.
111  */
112 
113 static DEFINE_MUTEX(ovs_mutex);
114 
115 void ovs_lock(void)
116 {
117 	mutex_lock(&ovs_mutex);
118 }
119 
120 void ovs_unlock(void)
121 {
122 	mutex_unlock(&ovs_mutex);
123 }
124 
125 #ifdef CONFIG_LOCKDEP
126 int lockdep_ovsl_is_held(void)
127 {
128 	if (debug_locks)
129 		return lockdep_is_held(&ovs_mutex);
130 	else
131 		return 1;
132 }
133 #endif
134 
135 static struct vport *new_vport(const struct vport_parms *);
136 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
137 			     const struct sw_flow_key *,
138 			     const struct dp_upcall_info *,
139 			     uint32_t cutlen);
140 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
141 				  const struct sw_flow_key *,
142 				  const struct dp_upcall_info *,
143 				  uint32_t cutlen);
144 
145 /* Must be called with rcu_read_lock. */
146 static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
147 {
148 	struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
149 
150 	if (dev) {
151 		struct vport *vport = ovs_internal_dev_get_vport(dev);
152 		if (vport)
153 			return vport->dp;
154 	}
155 
156 	return NULL;
157 }
158 
159 /* The caller must hold either ovs_mutex or rcu_read_lock to keep the
160  * returned dp pointer valid.
161  */
162 static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
163 {
164 	struct datapath *dp;
165 
166 	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
167 	rcu_read_lock();
168 	dp = get_dp_rcu(net, dp_ifindex);
169 	rcu_read_unlock();
170 
171 	return dp;
172 }
173 
174 /* Must be called with rcu_read_lock or ovs_mutex. */
175 const char *ovs_dp_name(const struct datapath *dp)
176 {
177 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
178 	return ovs_vport_name(vport);
179 }
180 
181 static int get_dpifindex(const struct datapath *dp)
182 {
183 	struct vport *local;
184 	int ifindex;
185 
186 	rcu_read_lock();
187 
188 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
189 	if (local)
190 		ifindex = local->dev->ifindex;
191 	else
192 		ifindex = 0;
193 
194 	rcu_read_unlock();
195 
196 	return ifindex;
197 }
198 
199 static void destroy_dp_rcu(struct rcu_head *rcu)
200 {
201 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
202 
203 	ovs_flow_tbl_destroy(&dp->table);
204 	free_percpu(dp->stats_percpu);
205 	kfree(dp->ports);
206 	kfree(dp);
207 }
208 
209 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
210 					    u16 port_no)
211 {
212 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
213 }
214 
215 /* Called with ovs_mutex or RCU read lock. */
216 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
217 {
218 	struct vport *vport;
219 	struct hlist_head *head;
220 
221 	head = vport_hash_bucket(dp, port_no);
222 	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
223 		if (vport->port_no == port_no)
224 			return vport;
225 	}
226 	return NULL;
227 }
228 
229 /* Called with ovs_mutex. */
230 static struct vport *new_vport(const struct vport_parms *parms)
231 {
232 	struct vport *vport;
233 
234 	vport = ovs_vport_add(parms);
235 	if (!IS_ERR(vport)) {
236 		struct datapath *dp = parms->dp;
237 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
238 
239 		hlist_add_head_rcu(&vport->dp_hash_node, head);
240 	}
241 	return vport;
242 }
243 
244 void ovs_dp_detach_port(struct vport *p)
245 {
246 	ASSERT_OVSL();
247 
248 	/* First drop references to device. */
249 	hlist_del_rcu(&p->dp_hash_node);
250 
251 	/* Then destroy it. */
252 	ovs_vport_del(p);
253 }
254 
255 /* Must be called with rcu_read_lock. */
256 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
257 {
258 	const struct vport *p = OVS_CB(skb)->input_vport;
259 	struct datapath *dp = p->dp;
260 	struct sw_flow *flow;
261 	struct sw_flow_actions *sf_acts;
262 	struct dp_stats_percpu *stats;
263 	u64 *stats_counter;
264 	u32 n_mask_hit;
265 
266 	stats = this_cpu_ptr(dp->stats_percpu);
267 
268 	/* Look up flow. */
269 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
270 	if (unlikely(!flow)) {
271 		struct dp_upcall_info upcall;
272 		int error;
273 
274 		memset(&upcall, 0, sizeof(upcall));
275 		upcall.cmd = OVS_PACKET_CMD_MISS;
276 		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 		upcall.mru = OVS_CB(skb)->mru;
278 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
279 		if (unlikely(error))
280 			kfree_skb(skb);
281 		else
282 			consume_skb(skb);
283 		stats_counter = &stats->n_missed;
284 		goto out;
285 	}
286 
287 	ovs_flow_stats_update(flow, key->tp.flags, skb);
288 	sf_acts = rcu_dereference(flow->sf_acts);
289 	ovs_execute_actions(dp, skb, sf_acts, key);
290 
291 	stats_counter = &stats->n_hit;
292 
293 out:
294 	/* Update datapath statistics. */
295 	u64_stats_update_begin(&stats->syncp);
296 	(*stats_counter)++;
297 	stats->n_mask_hit += n_mask_hit;
298 	u64_stats_update_end(&stats->syncp);
299 }
300 
301 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
302 		  const struct sw_flow_key *key,
303 		  const struct dp_upcall_info *upcall_info,
304 		  uint32_t cutlen)
305 {
306 	struct dp_stats_percpu *stats;
307 	int err;
308 
309 	if (upcall_info->portid == 0) {
310 		err = -ENOTCONN;
311 		goto err;
312 	}
313 
314 	if (!skb_is_gso(skb))
315 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
316 	else
317 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
318 	if (err)
319 		goto err;
320 
321 	return 0;
322 
323 err:
324 	stats = this_cpu_ptr(dp->stats_percpu);
325 
326 	u64_stats_update_begin(&stats->syncp);
327 	stats->n_lost++;
328 	u64_stats_update_end(&stats->syncp);
329 
330 	return err;
331 }
332 
333 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
334 			     const struct sw_flow_key *key,
335 			     const struct dp_upcall_info *upcall_info,
336 				 uint32_t cutlen)
337 {
338 	unsigned short gso_type = skb_shinfo(skb)->gso_type;
339 	struct sw_flow_key later_key;
340 	struct sk_buff *segs, *nskb;
341 	int err;
342 
343 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
344 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
345 	if (IS_ERR(segs))
346 		return PTR_ERR(segs);
347 	if (segs == NULL)
348 		return -EINVAL;
349 
350 	if (gso_type & SKB_GSO_UDP) {
351 		/* The initial flow key extracted by ovs_flow_key_extract()
352 		 * in this case is for a first fragment, so we need to
353 		 * properly mark later fragments.
354 		 */
355 		later_key = *key;
356 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
357 	}
358 
359 	/* Queue all of the segments. */
360 	skb = segs;
361 	do {
362 		if (gso_type & SKB_GSO_UDP && skb != segs)
363 			key = &later_key;
364 
365 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
366 		if (err)
367 			break;
368 
369 	} while ((skb = skb->next));
370 
371 	/* Free all of the segments. */
372 	skb = segs;
373 	do {
374 		nskb = skb->next;
375 		if (err)
376 			kfree_skb(skb);
377 		else
378 			consume_skb(skb);
379 	} while ((skb = nskb));
380 	return err;
381 }
382 
383 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
384 			      unsigned int hdrlen)
385 {
386 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
387 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
388 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
389 		+ nla_total_size(sizeof(unsigned int)); /* OVS_PACKET_ATTR_LEN */
390 
391 	/* OVS_PACKET_ATTR_USERDATA */
392 	if (upcall_info->userdata)
393 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
394 
395 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
396 	if (upcall_info->egress_tun_info)
397 		size += nla_total_size(ovs_tun_key_attr_size());
398 
399 	/* OVS_PACKET_ATTR_ACTIONS */
400 	if (upcall_info->actions_len)
401 		size += nla_total_size(upcall_info->actions_len);
402 
403 	/* OVS_PACKET_ATTR_MRU */
404 	if (upcall_info->mru)
405 		size += nla_total_size(sizeof(upcall_info->mru));
406 
407 	return size;
408 }
409 
410 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
411 {
412 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
413 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
414 
415 		if (plen > 0)
416 			skb_put_zero(skb, plen);
417 	}
418 }
419 
420 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
421 				  const struct sw_flow_key *key,
422 				  const struct dp_upcall_info *upcall_info,
423 				  uint32_t cutlen)
424 {
425 	struct ovs_header *upcall;
426 	struct sk_buff *nskb = NULL;
427 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
428 	struct nlattr *nla;
429 	size_t len;
430 	unsigned int hlen;
431 	int err, dp_ifindex;
432 
433 	dp_ifindex = get_dpifindex(dp);
434 	if (!dp_ifindex)
435 		return -ENODEV;
436 
437 	if (skb_vlan_tag_present(skb)) {
438 		nskb = skb_clone(skb, GFP_ATOMIC);
439 		if (!nskb)
440 			return -ENOMEM;
441 
442 		nskb = __vlan_hwaccel_push_inside(nskb);
443 		if (!nskb)
444 			return -ENOMEM;
445 
446 		skb = nskb;
447 	}
448 
449 	if (nla_attr_size(skb->len) > USHRT_MAX) {
450 		err = -EFBIG;
451 		goto out;
452 	}
453 
454 	/* Complete checksum if needed */
455 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
456 	    (err = skb_csum_hwoffload_help(skb, 0)))
457 		goto out;
458 
459 	/* Older versions of OVS user space enforce alignment of the last
460 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
461 	 * padding logic. Only perform zerocopy if padding is not required.
462 	 */
463 	if (dp->user_features & OVS_DP_F_UNALIGNED)
464 		hlen = skb_zerocopy_headlen(skb);
465 	else
466 		hlen = skb->len;
467 
468 	len = upcall_msg_size(upcall_info, hlen - cutlen);
469 	user_skb = genlmsg_new(len, GFP_ATOMIC);
470 	if (!user_skb) {
471 		err = -ENOMEM;
472 		goto out;
473 	}
474 
475 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
476 			     0, upcall_info->cmd);
477 	upcall->dp_ifindex = dp_ifindex;
478 
479 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
480 	BUG_ON(err);
481 
482 	if (upcall_info->userdata)
483 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
484 			  nla_len(upcall_info->userdata),
485 			  nla_data(upcall_info->userdata));
486 
487 	if (upcall_info->egress_tun_info) {
488 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
489 		err = ovs_nla_put_tunnel_info(user_skb,
490 					      upcall_info->egress_tun_info);
491 		BUG_ON(err);
492 		nla_nest_end(user_skb, nla);
493 	}
494 
495 	if (upcall_info->actions_len) {
496 		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_ACTIONS);
497 		err = ovs_nla_put_actions(upcall_info->actions,
498 					  upcall_info->actions_len,
499 					  user_skb);
500 		if (!err)
501 			nla_nest_end(user_skb, nla);
502 		else
503 			nla_nest_cancel(user_skb, nla);
504 	}
505 
506 	/* Add OVS_PACKET_ATTR_MRU */
507 	if (upcall_info->mru) {
508 		if (nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU,
509 				upcall_info->mru)) {
510 			err = -ENOBUFS;
511 			goto out;
512 		}
513 		pad_packet(dp, user_skb);
514 	}
515 
516 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
517 	if (cutlen > 0) {
518 		if (nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN,
519 				skb->len)) {
520 			err = -ENOBUFS;
521 			goto out;
522 		}
523 		pad_packet(dp, user_skb);
524 	}
525 
526 	/* Only reserve room for attribute header, packet data is added
527 	 * in skb_zerocopy() */
528 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
529 		err = -ENOBUFS;
530 		goto out;
531 	}
532 	nla->nla_len = nla_attr_size(skb->len - cutlen);
533 
534 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
535 	if (err)
536 		goto out;
537 
538 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
539 	pad_packet(dp, user_skb);
540 
541 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
542 
543 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
544 	user_skb = NULL;
545 out:
546 	if (err)
547 		skb_tx_error(skb);
548 	kfree_skb(user_skb);
549 	kfree_skb(nskb);
550 	return err;
551 }
552 
553 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
554 {
555 	struct ovs_header *ovs_header = info->userhdr;
556 	struct net *net = sock_net(skb->sk);
557 	struct nlattr **a = info->attrs;
558 	struct sw_flow_actions *acts;
559 	struct sk_buff *packet;
560 	struct sw_flow *flow;
561 	struct sw_flow_actions *sf_acts;
562 	struct datapath *dp;
563 	struct vport *input_vport;
564 	u16 mru = 0;
565 	int len;
566 	int err;
567 	bool log = !a[OVS_PACKET_ATTR_PROBE];
568 
569 	err = -EINVAL;
570 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
571 	    !a[OVS_PACKET_ATTR_ACTIONS])
572 		goto err;
573 
574 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
575 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
576 	err = -ENOMEM;
577 	if (!packet)
578 		goto err;
579 	skb_reserve(packet, NET_IP_ALIGN);
580 
581 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
582 
583 	/* Set packet's mru */
584 	if (a[OVS_PACKET_ATTR_MRU]) {
585 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
586 		packet->ignore_df = 1;
587 	}
588 	OVS_CB(packet)->mru = mru;
589 
590 	/* Build an sw_flow for sending this packet. */
591 	flow = ovs_flow_alloc();
592 	err = PTR_ERR(flow);
593 	if (IS_ERR(flow))
594 		goto err_kfree_skb;
595 
596 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
597 					     packet, &flow->key, log);
598 	if (err)
599 		goto err_flow_free;
600 
601 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
602 				   &flow->key, &acts, log);
603 	if (err)
604 		goto err_flow_free;
605 
606 	rcu_assign_pointer(flow->sf_acts, acts);
607 	packet->priority = flow->key.phy.priority;
608 	packet->mark = flow->key.phy.skb_mark;
609 
610 	rcu_read_lock();
611 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
612 	err = -ENODEV;
613 	if (!dp)
614 		goto err_unlock;
615 
616 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
617 	if (!input_vport)
618 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
619 
620 	if (!input_vport)
621 		goto err_unlock;
622 
623 	packet->dev = input_vport->dev;
624 	OVS_CB(packet)->input_vport = input_vport;
625 	sf_acts = rcu_dereference(flow->sf_acts);
626 
627 	local_bh_disable();
628 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
629 	local_bh_enable();
630 	rcu_read_unlock();
631 
632 	ovs_flow_free(flow, false);
633 	return err;
634 
635 err_unlock:
636 	rcu_read_unlock();
637 err_flow_free:
638 	ovs_flow_free(flow, false);
639 err_kfree_skb:
640 	kfree_skb(packet);
641 err:
642 	return err;
643 }
644 
645 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
646 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
647 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
648 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
649 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
650 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
651 };
652 
653 static const struct genl_ops dp_packet_genl_ops[] = {
654 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
655 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
656 	  .policy = packet_policy,
657 	  .doit = ovs_packet_cmd_execute
658 	}
659 };
660 
661 static struct genl_family dp_packet_genl_family __ro_after_init = {
662 	.hdrsize = sizeof(struct ovs_header),
663 	.name = OVS_PACKET_FAMILY,
664 	.version = OVS_PACKET_VERSION,
665 	.maxattr = OVS_PACKET_ATTR_MAX,
666 	.netnsok = true,
667 	.parallel_ops = true,
668 	.ops = dp_packet_genl_ops,
669 	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
670 	.module = THIS_MODULE,
671 };
672 
673 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
674 			 struct ovs_dp_megaflow_stats *mega_stats)
675 {
676 	int i;
677 
678 	memset(mega_stats, 0, sizeof(*mega_stats));
679 
680 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
681 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
682 
683 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
684 
685 	for_each_possible_cpu(i) {
686 		const struct dp_stats_percpu *percpu_stats;
687 		struct dp_stats_percpu local_stats;
688 		unsigned int start;
689 
690 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
691 
692 		do {
693 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
694 			local_stats = *percpu_stats;
695 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
696 
697 		stats->n_hit += local_stats.n_hit;
698 		stats->n_missed += local_stats.n_missed;
699 		stats->n_lost += local_stats.n_lost;
700 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
701 	}
702 }
703 
704 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
705 {
706 	return ovs_identifier_is_ufid(sfid) &&
707 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
708 }
709 
710 static bool should_fill_mask(uint32_t ufid_flags)
711 {
712 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
713 }
714 
715 static bool should_fill_actions(uint32_t ufid_flags)
716 {
717 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
718 }
719 
720 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
721 				    const struct sw_flow_id *sfid,
722 				    uint32_t ufid_flags)
723 {
724 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
725 
726 	/* OVS_FLOW_ATTR_UFID */
727 	if (sfid && ovs_identifier_is_ufid(sfid))
728 		len += nla_total_size(sfid->ufid_len);
729 
730 	/* OVS_FLOW_ATTR_KEY */
731 	if (!sfid || should_fill_key(sfid, ufid_flags))
732 		len += nla_total_size(ovs_key_attr_size());
733 
734 	/* OVS_FLOW_ATTR_MASK */
735 	if (should_fill_mask(ufid_flags))
736 		len += nla_total_size(ovs_key_attr_size());
737 
738 	/* OVS_FLOW_ATTR_ACTIONS */
739 	if (should_fill_actions(ufid_flags))
740 		len += nla_total_size(acts->orig_len);
741 
742 	return len
743 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
744 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
745 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
746 }
747 
748 /* Called with ovs_mutex or RCU read lock. */
749 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
750 				   struct sk_buff *skb)
751 {
752 	struct ovs_flow_stats stats;
753 	__be16 tcp_flags;
754 	unsigned long used;
755 
756 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
757 
758 	if (used &&
759 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
760 			      OVS_FLOW_ATTR_PAD))
761 		return -EMSGSIZE;
762 
763 	if (stats.n_packets &&
764 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
765 			  sizeof(struct ovs_flow_stats), &stats,
766 			  OVS_FLOW_ATTR_PAD))
767 		return -EMSGSIZE;
768 
769 	if ((u8)ntohs(tcp_flags) &&
770 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
771 		return -EMSGSIZE;
772 
773 	return 0;
774 }
775 
776 /* Called with ovs_mutex or RCU read lock. */
777 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
778 				     struct sk_buff *skb, int skb_orig_len)
779 {
780 	struct nlattr *start;
781 	int err;
782 
783 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
784 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
785 	 * Netlink but individual action lists can be longer than
786 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
787 	 * The userspace caller can always fetch the actions separately if it
788 	 * really wants them.  (Most userspace callers in fact don't care.)
789 	 *
790 	 * This can only fail for dump operations because the skb is always
791 	 * properly sized for single flows.
792 	 */
793 	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
794 	if (start) {
795 		const struct sw_flow_actions *sf_acts;
796 
797 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
798 		err = ovs_nla_put_actions(sf_acts->actions,
799 					  sf_acts->actions_len, skb);
800 
801 		if (!err)
802 			nla_nest_end(skb, start);
803 		else {
804 			if (skb_orig_len)
805 				return err;
806 
807 			nla_nest_cancel(skb, start);
808 		}
809 	} else if (skb_orig_len) {
810 		return -EMSGSIZE;
811 	}
812 
813 	return 0;
814 }
815 
816 /* Called with ovs_mutex or RCU read lock. */
817 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
818 				  struct sk_buff *skb, u32 portid,
819 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
820 {
821 	const int skb_orig_len = skb->len;
822 	struct ovs_header *ovs_header;
823 	int err;
824 
825 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
826 				 flags, cmd);
827 	if (!ovs_header)
828 		return -EMSGSIZE;
829 
830 	ovs_header->dp_ifindex = dp_ifindex;
831 
832 	err = ovs_nla_put_identifier(flow, skb);
833 	if (err)
834 		goto error;
835 
836 	if (should_fill_key(&flow->id, ufid_flags)) {
837 		err = ovs_nla_put_masked_key(flow, skb);
838 		if (err)
839 			goto error;
840 	}
841 
842 	if (should_fill_mask(ufid_flags)) {
843 		err = ovs_nla_put_mask(flow, skb);
844 		if (err)
845 			goto error;
846 	}
847 
848 	err = ovs_flow_cmd_fill_stats(flow, skb);
849 	if (err)
850 		goto error;
851 
852 	if (should_fill_actions(ufid_flags)) {
853 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
854 		if (err)
855 			goto error;
856 	}
857 
858 	genlmsg_end(skb, ovs_header);
859 	return 0;
860 
861 error:
862 	genlmsg_cancel(skb, ovs_header);
863 	return err;
864 }
865 
866 /* May not be called with RCU read lock. */
867 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
868 					       const struct sw_flow_id *sfid,
869 					       struct genl_info *info,
870 					       bool always,
871 					       uint32_t ufid_flags)
872 {
873 	struct sk_buff *skb;
874 	size_t len;
875 
876 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
877 		return NULL;
878 
879 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
880 	skb = genlmsg_new(len, GFP_KERNEL);
881 	if (!skb)
882 		return ERR_PTR(-ENOMEM);
883 
884 	return skb;
885 }
886 
887 /* Called with ovs_mutex. */
888 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
889 					       int dp_ifindex,
890 					       struct genl_info *info, u8 cmd,
891 					       bool always, u32 ufid_flags)
892 {
893 	struct sk_buff *skb;
894 	int retval;
895 
896 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
897 				      &flow->id, info, always, ufid_flags);
898 	if (IS_ERR_OR_NULL(skb))
899 		return skb;
900 
901 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
902 					info->snd_portid, info->snd_seq, 0,
903 					cmd, ufid_flags);
904 	BUG_ON(retval < 0);
905 	return skb;
906 }
907 
908 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
909 {
910 	struct net *net = sock_net(skb->sk);
911 	struct nlattr **a = info->attrs;
912 	struct ovs_header *ovs_header = info->userhdr;
913 	struct sw_flow *flow = NULL, *new_flow;
914 	struct sw_flow_mask mask;
915 	struct sk_buff *reply;
916 	struct datapath *dp;
917 	struct sw_flow_actions *acts;
918 	struct sw_flow_match match;
919 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
920 	int error;
921 	bool log = !a[OVS_FLOW_ATTR_PROBE];
922 
923 	/* Must have key and actions. */
924 	error = -EINVAL;
925 	if (!a[OVS_FLOW_ATTR_KEY]) {
926 		OVS_NLERR(log, "Flow key attr not present in new flow.");
927 		goto error;
928 	}
929 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
930 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
931 		goto error;
932 	}
933 
934 	/* Most of the time we need to allocate a new flow, do it before
935 	 * locking.
936 	 */
937 	new_flow = ovs_flow_alloc();
938 	if (IS_ERR(new_flow)) {
939 		error = PTR_ERR(new_flow);
940 		goto error;
941 	}
942 
943 	/* Extract key. */
944 	ovs_match_init(&match, &new_flow->key, false, &mask);
945 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
946 				  a[OVS_FLOW_ATTR_MASK], log);
947 	if (error)
948 		goto err_kfree_flow;
949 
950 	/* Extract flow identifier. */
951 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
952 				       &new_flow->key, log);
953 	if (error)
954 		goto err_kfree_flow;
955 
956 	/* unmasked key is needed to match when ufid is not used. */
957 	if (ovs_identifier_is_key(&new_flow->id))
958 		match.key = new_flow->id.unmasked_key;
959 
960 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
961 
962 	/* Validate actions. */
963 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
964 				     &new_flow->key, &acts, log);
965 	if (error) {
966 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
967 		goto err_kfree_flow;
968 	}
969 
970 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
971 					ufid_flags);
972 	if (IS_ERR(reply)) {
973 		error = PTR_ERR(reply);
974 		goto err_kfree_acts;
975 	}
976 
977 	ovs_lock();
978 	dp = get_dp(net, ovs_header->dp_ifindex);
979 	if (unlikely(!dp)) {
980 		error = -ENODEV;
981 		goto err_unlock_ovs;
982 	}
983 
984 	/* Check if this is a duplicate flow */
985 	if (ovs_identifier_is_ufid(&new_flow->id))
986 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
987 	if (!flow)
988 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
989 	if (likely(!flow)) {
990 		rcu_assign_pointer(new_flow->sf_acts, acts);
991 
992 		/* Put flow in bucket. */
993 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
994 		if (unlikely(error)) {
995 			acts = NULL;
996 			goto err_unlock_ovs;
997 		}
998 
999 		if (unlikely(reply)) {
1000 			error = ovs_flow_cmd_fill_info(new_flow,
1001 						       ovs_header->dp_ifindex,
1002 						       reply, info->snd_portid,
1003 						       info->snd_seq, 0,
1004 						       OVS_FLOW_CMD_NEW,
1005 						       ufid_flags);
1006 			BUG_ON(error < 0);
1007 		}
1008 		ovs_unlock();
1009 	} else {
1010 		struct sw_flow_actions *old_acts;
1011 
1012 		/* Bail out if we're not allowed to modify an existing flow.
1013 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1014 		 * because Generic Netlink treats the latter as a dump
1015 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1016 		 * gets fixed.
1017 		 */
1018 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1019 							 | NLM_F_EXCL))) {
1020 			error = -EEXIST;
1021 			goto err_unlock_ovs;
1022 		}
1023 		/* The flow identifier has to be the same for flow updates.
1024 		 * Look for any overlapping flow.
1025 		 */
1026 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1027 			if (ovs_identifier_is_key(&flow->id))
1028 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1029 								 &match);
1030 			else /* UFID matches but key is different */
1031 				flow = NULL;
1032 			if (!flow) {
1033 				error = -ENOENT;
1034 				goto err_unlock_ovs;
1035 			}
1036 		}
1037 		/* Update actions. */
1038 		old_acts = ovsl_dereference(flow->sf_acts);
1039 		rcu_assign_pointer(flow->sf_acts, acts);
1040 
1041 		if (unlikely(reply)) {
1042 			error = ovs_flow_cmd_fill_info(flow,
1043 						       ovs_header->dp_ifindex,
1044 						       reply, info->snd_portid,
1045 						       info->snd_seq, 0,
1046 						       OVS_FLOW_CMD_NEW,
1047 						       ufid_flags);
1048 			BUG_ON(error < 0);
1049 		}
1050 		ovs_unlock();
1051 
1052 		ovs_nla_free_flow_actions_rcu(old_acts);
1053 		ovs_flow_free(new_flow, false);
1054 	}
1055 
1056 	if (reply)
1057 		ovs_notify(&dp_flow_genl_family, reply, info);
1058 	return 0;
1059 
1060 err_unlock_ovs:
1061 	ovs_unlock();
1062 	kfree_skb(reply);
1063 err_kfree_acts:
1064 	ovs_nla_free_flow_actions(acts);
1065 err_kfree_flow:
1066 	ovs_flow_free(new_flow, false);
1067 error:
1068 	return error;
1069 }
1070 
1071 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1072 static struct sw_flow_actions *get_flow_actions(struct net *net,
1073 						const struct nlattr *a,
1074 						const struct sw_flow_key *key,
1075 						const struct sw_flow_mask *mask,
1076 						bool log)
1077 {
1078 	struct sw_flow_actions *acts;
1079 	struct sw_flow_key masked_key;
1080 	int error;
1081 
1082 	ovs_flow_mask_key(&masked_key, key, true, mask);
1083 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1084 	if (error) {
1085 		OVS_NLERR(log,
1086 			  "Actions may not be safe on all matching packets");
1087 		return ERR_PTR(error);
1088 	}
1089 
1090 	return acts;
1091 }
1092 
1093 /* Factor out match-init and action-copy to avoid
1094  * "Wframe-larger-than=1024" warning. Because mask is only
1095  * used to get actions, we new a function to save some
1096  * stack space.
1097  *
1098  * If there are not key and action attrs, we return 0
1099  * directly. In the case, the caller will also not use the
1100  * match as before. If there is action attr, we try to get
1101  * actions and save them to *acts. Before returning from
1102  * the function, we reset the match->mask pointer. Because
1103  * we should not to return match object with dangling reference
1104  * to mask.
1105  * */
1106 static int ovs_nla_init_match_and_action(struct net *net,
1107 					 struct sw_flow_match *match,
1108 					 struct sw_flow_key *key,
1109 					 struct nlattr **a,
1110 					 struct sw_flow_actions **acts,
1111 					 bool log)
1112 {
1113 	struct sw_flow_mask mask;
1114 	int error = 0;
1115 
1116 	if (a[OVS_FLOW_ATTR_KEY]) {
1117 		ovs_match_init(match, key, true, &mask);
1118 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1119 					  a[OVS_FLOW_ATTR_MASK], log);
1120 		if (error)
1121 			goto error;
1122 	}
1123 
1124 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1125 		if (!a[OVS_FLOW_ATTR_KEY]) {
1126 			OVS_NLERR(log,
1127 				  "Flow key attribute not present in set flow.");
1128 			return -EINVAL;
1129 		}
1130 
1131 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1132 					 &mask, log);
1133 		if (IS_ERR(*acts)) {
1134 			error = PTR_ERR(*acts);
1135 			goto error;
1136 		}
1137 	}
1138 
1139 	/* On success, error is 0. */
1140 error:
1141 	match->mask = NULL;
1142 	return error;
1143 }
1144 
1145 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1146 {
1147 	struct net *net = sock_net(skb->sk);
1148 	struct nlattr **a = info->attrs;
1149 	struct ovs_header *ovs_header = info->userhdr;
1150 	struct sw_flow_key key;
1151 	struct sw_flow *flow;
1152 	struct sk_buff *reply = NULL;
1153 	struct datapath *dp;
1154 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1155 	struct sw_flow_match match;
1156 	struct sw_flow_id sfid;
1157 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1158 	int error = 0;
1159 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1160 	bool ufid_present;
1161 
1162 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1163 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1164 		OVS_NLERR(log,
1165 			  "Flow set message rejected, Key attribute missing.");
1166 		return -EINVAL;
1167 	}
1168 
1169 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1170 					      &acts, log);
1171 	if (error)
1172 		goto error;
1173 
1174 	if (acts) {
1175 		/* Can allocate before locking if have acts. */
1176 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1177 						ufid_flags);
1178 		if (IS_ERR(reply)) {
1179 			error = PTR_ERR(reply);
1180 			goto err_kfree_acts;
1181 		}
1182 	}
1183 
1184 	ovs_lock();
1185 	dp = get_dp(net, ovs_header->dp_ifindex);
1186 	if (unlikely(!dp)) {
1187 		error = -ENODEV;
1188 		goto err_unlock_ovs;
1189 	}
1190 	/* Check that the flow exists. */
1191 	if (ufid_present)
1192 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1193 	else
1194 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1195 	if (unlikely(!flow)) {
1196 		error = -ENOENT;
1197 		goto err_unlock_ovs;
1198 	}
1199 
1200 	/* Update actions, if present. */
1201 	if (likely(acts)) {
1202 		old_acts = ovsl_dereference(flow->sf_acts);
1203 		rcu_assign_pointer(flow->sf_acts, acts);
1204 
1205 		if (unlikely(reply)) {
1206 			error = ovs_flow_cmd_fill_info(flow,
1207 						       ovs_header->dp_ifindex,
1208 						       reply, info->snd_portid,
1209 						       info->snd_seq, 0,
1210 						       OVS_FLOW_CMD_NEW,
1211 						       ufid_flags);
1212 			BUG_ON(error < 0);
1213 		}
1214 	} else {
1215 		/* Could not alloc without acts before locking. */
1216 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1217 						info, OVS_FLOW_CMD_NEW, false,
1218 						ufid_flags);
1219 
1220 		if (IS_ERR(reply)) {
1221 			error = PTR_ERR(reply);
1222 			goto err_unlock_ovs;
1223 		}
1224 	}
1225 
1226 	/* Clear stats. */
1227 	if (a[OVS_FLOW_ATTR_CLEAR])
1228 		ovs_flow_stats_clear(flow);
1229 	ovs_unlock();
1230 
1231 	if (reply)
1232 		ovs_notify(&dp_flow_genl_family, reply, info);
1233 	if (old_acts)
1234 		ovs_nla_free_flow_actions_rcu(old_acts);
1235 
1236 	return 0;
1237 
1238 err_unlock_ovs:
1239 	ovs_unlock();
1240 	kfree_skb(reply);
1241 err_kfree_acts:
1242 	ovs_nla_free_flow_actions(acts);
1243 error:
1244 	return error;
1245 }
1246 
1247 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1248 {
1249 	struct nlattr **a = info->attrs;
1250 	struct ovs_header *ovs_header = info->userhdr;
1251 	struct net *net = sock_net(skb->sk);
1252 	struct sw_flow_key key;
1253 	struct sk_buff *reply;
1254 	struct sw_flow *flow;
1255 	struct datapath *dp;
1256 	struct sw_flow_match match;
1257 	struct sw_flow_id ufid;
1258 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1259 	int err = 0;
1260 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1261 	bool ufid_present;
1262 
1263 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1264 	if (a[OVS_FLOW_ATTR_KEY]) {
1265 		ovs_match_init(&match, &key, true, NULL);
1266 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1267 					log);
1268 	} else if (!ufid_present) {
1269 		OVS_NLERR(log,
1270 			  "Flow get message rejected, Key attribute missing.");
1271 		err = -EINVAL;
1272 	}
1273 	if (err)
1274 		return err;
1275 
1276 	ovs_lock();
1277 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1278 	if (!dp) {
1279 		err = -ENODEV;
1280 		goto unlock;
1281 	}
1282 
1283 	if (ufid_present)
1284 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1285 	else
1286 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1287 	if (!flow) {
1288 		err = -ENOENT;
1289 		goto unlock;
1290 	}
1291 
1292 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1293 					OVS_FLOW_CMD_NEW, true, ufid_flags);
1294 	if (IS_ERR(reply)) {
1295 		err = PTR_ERR(reply);
1296 		goto unlock;
1297 	}
1298 
1299 	ovs_unlock();
1300 	return genlmsg_reply(reply, info);
1301 unlock:
1302 	ovs_unlock();
1303 	return err;
1304 }
1305 
1306 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1307 {
1308 	struct nlattr **a = info->attrs;
1309 	struct ovs_header *ovs_header = info->userhdr;
1310 	struct net *net = sock_net(skb->sk);
1311 	struct sw_flow_key key;
1312 	struct sk_buff *reply;
1313 	struct sw_flow *flow = NULL;
1314 	struct datapath *dp;
1315 	struct sw_flow_match match;
1316 	struct sw_flow_id ufid;
1317 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1318 	int err;
1319 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1320 	bool ufid_present;
1321 
1322 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1323 	if (a[OVS_FLOW_ATTR_KEY]) {
1324 		ovs_match_init(&match, &key, true, NULL);
1325 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1326 					NULL, log);
1327 		if (unlikely(err))
1328 			return err;
1329 	}
1330 
1331 	ovs_lock();
1332 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1333 	if (unlikely(!dp)) {
1334 		err = -ENODEV;
1335 		goto unlock;
1336 	}
1337 
1338 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1339 		err = ovs_flow_tbl_flush(&dp->table);
1340 		goto unlock;
1341 	}
1342 
1343 	if (ufid_present)
1344 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1345 	else
1346 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1347 	if (unlikely(!flow)) {
1348 		err = -ENOENT;
1349 		goto unlock;
1350 	}
1351 
1352 	ovs_flow_tbl_remove(&dp->table, flow);
1353 	ovs_unlock();
1354 
1355 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1356 					&flow->id, info, false, ufid_flags);
1357 	if (likely(reply)) {
1358 		if (likely(!IS_ERR(reply))) {
1359 			rcu_read_lock();	/*To keep RCU checker happy. */
1360 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1361 						     reply, info->snd_portid,
1362 						     info->snd_seq, 0,
1363 						     OVS_FLOW_CMD_DEL,
1364 						     ufid_flags);
1365 			rcu_read_unlock();
1366 			BUG_ON(err < 0);
1367 
1368 			ovs_notify(&dp_flow_genl_family, reply, info);
1369 		} else {
1370 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1371 		}
1372 	}
1373 
1374 	ovs_flow_free(flow, true);
1375 	return 0;
1376 unlock:
1377 	ovs_unlock();
1378 	return err;
1379 }
1380 
1381 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1382 {
1383 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1384 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1385 	struct table_instance *ti;
1386 	struct datapath *dp;
1387 	u32 ufid_flags;
1388 	int err;
1389 
1390 	err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1391 			    OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1392 	if (err)
1393 		return err;
1394 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1395 
1396 	rcu_read_lock();
1397 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1398 	if (!dp) {
1399 		rcu_read_unlock();
1400 		return -ENODEV;
1401 	}
1402 
1403 	ti = rcu_dereference(dp->table.ti);
1404 	for (;;) {
1405 		struct sw_flow *flow;
1406 		u32 bucket, obj;
1407 
1408 		bucket = cb->args[0];
1409 		obj = cb->args[1];
1410 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1411 		if (!flow)
1412 			break;
1413 
1414 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1415 					   NETLINK_CB(cb->skb).portid,
1416 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1417 					   OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1418 			break;
1419 
1420 		cb->args[0] = bucket;
1421 		cb->args[1] = obj;
1422 	}
1423 	rcu_read_unlock();
1424 	return skb->len;
1425 }
1426 
1427 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1428 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1429 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1430 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1431 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1432 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1433 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1434 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1435 };
1436 
1437 static const struct genl_ops dp_flow_genl_ops[] = {
1438 	{ .cmd = OVS_FLOW_CMD_NEW,
1439 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1440 	  .policy = flow_policy,
1441 	  .doit = ovs_flow_cmd_new
1442 	},
1443 	{ .cmd = OVS_FLOW_CMD_DEL,
1444 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1445 	  .policy = flow_policy,
1446 	  .doit = ovs_flow_cmd_del
1447 	},
1448 	{ .cmd = OVS_FLOW_CMD_GET,
1449 	  .flags = 0,		    /* OK for unprivileged users. */
1450 	  .policy = flow_policy,
1451 	  .doit = ovs_flow_cmd_get,
1452 	  .dumpit = ovs_flow_cmd_dump
1453 	},
1454 	{ .cmd = OVS_FLOW_CMD_SET,
1455 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1456 	  .policy = flow_policy,
1457 	  .doit = ovs_flow_cmd_set,
1458 	},
1459 };
1460 
1461 static struct genl_family dp_flow_genl_family __ro_after_init = {
1462 	.hdrsize = sizeof(struct ovs_header),
1463 	.name = OVS_FLOW_FAMILY,
1464 	.version = OVS_FLOW_VERSION,
1465 	.maxattr = OVS_FLOW_ATTR_MAX,
1466 	.netnsok = true,
1467 	.parallel_ops = true,
1468 	.ops = dp_flow_genl_ops,
1469 	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1470 	.mcgrps = &ovs_dp_flow_multicast_group,
1471 	.n_mcgrps = 1,
1472 	.module = THIS_MODULE,
1473 };
1474 
1475 static size_t ovs_dp_cmd_msg_size(void)
1476 {
1477 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1478 
1479 	msgsize += nla_total_size(IFNAMSIZ);
1480 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1481 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1482 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1483 
1484 	return msgsize;
1485 }
1486 
1487 /* Called with ovs_mutex. */
1488 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1489 				u32 portid, u32 seq, u32 flags, u8 cmd)
1490 {
1491 	struct ovs_header *ovs_header;
1492 	struct ovs_dp_stats dp_stats;
1493 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1494 	int err;
1495 
1496 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1497 				   flags, cmd);
1498 	if (!ovs_header)
1499 		goto error;
1500 
1501 	ovs_header->dp_ifindex = get_dpifindex(dp);
1502 
1503 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1504 	if (err)
1505 		goto nla_put_failure;
1506 
1507 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1508 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1509 			  &dp_stats, OVS_DP_ATTR_PAD))
1510 		goto nla_put_failure;
1511 
1512 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1513 			  sizeof(struct ovs_dp_megaflow_stats),
1514 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1515 		goto nla_put_failure;
1516 
1517 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1518 		goto nla_put_failure;
1519 
1520 	genlmsg_end(skb, ovs_header);
1521 	return 0;
1522 
1523 nla_put_failure:
1524 	genlmsg_cancel(skb, ovs_header);
1525 error:
1526 	return -EMSGSIZE;
1527 }
1528 
1529 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1530 {
1531 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1532 }
1533 
1534 /* Called with rcu_read_lock or ovs_mutex. */
1535 static struct datapath *lookup_datapath(struct net *net,
1536 					const struct ovs_header *ovs_header,
1537 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1538 {
1539 	struct datapath *dp;
1540 
1541 	if (!a[OVS_DP_ATTR_NAME])
1542 		dp = get_dp(net, ovs_header->dp_ifindex);
1543 	else {
1544 		struct vport *vport;
1545 
1546 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1547 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1548 	}
1549 	return dp ? dp : ERR_PTR(-ENODEV);
1550 }
1551 
1552 static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1553 {
1554 	struct datapath *dp;
1555 
1556 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1557 	if (IS_ERR(dp))
1558 		return;
1559 
1560 	WARN(dp->user_features, "Dropping previously announced user features\n");
1561 	dp->user_features = 0;
1562 }
1563 
1564 static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1565 {
1566 	if (a[OVS_DP_ATTR_USER_FEATURES])
1567 		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1568 }
1569 
1570 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1571 {
1572 	struct nlattr **a = info->attrs;
1573 	struct vport_parms parms;
1574 	struct sk_buff *reply;
1575 	struct datapath *dp;
1576 	struct vport *vport;
1577 	struct ovs_net *ovs_net;
1578 	int err, i;
1579 
1580 	err = -EINVAL;
1581 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1582 		goto err;
1583 
1584 	reply = ovs_dp_cmd_alloc_info();
1585 	if (!reply)
1586 		return -ENOMEM;
1587 
1588 	err = -ENOMEM;
1589 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1590 	if (dp == NULL)
1591 		goto err_free_reply;
1592 
1593 	ovs_dp_set_net(dp, sock_net(skb->sk));
1594 
1595 	/* Allocate table. */
1596 	err = ovs_flow_tbl_init(&dp->table);
1597 	if (err)
1598 		goto err_free_dp;
1599 
1600 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1601 	if (!dp->stats_percpu) {
1602 		err = -ENOMEM;
1603 		goto err_destroy_table;
1604 	}
1605 
1606 	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1607 			    GFP_KERNEL);
1608 	if (!dp->ports) {
1609 		err = -ENOMEM;
1610 		goto err_destroy_percpu;
1611 	}
1612 
1613 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1614 		INIT_HLIST_HEAD(&dp->ports[i]);
1615 
1616 	/* Set up our datapath device. */
1617 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1618 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1619 	parms.options = NULL;
1620 	parms.dp = dp;
1621 	parms.port_no = OVSP_LOCAL;
1622 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1623 
1624 	ovs_dp_change(dp, a);
1625 
1626 	/* So far only local changes have been made, now need the lock. */
1627 	ovs_lock();
1628 
1629 	vport = new_vport(&parms);
1630 	if (IS_ERR(vport)) {
1631 		err = PTR_ERR(vport);
1632 		if (err == -EBUSY)
1633 			err = -EEXIST;
1634 
1635 		if (err == -EEXIST) {
1636 			/* An outdated user space instance that does not understand
1637 			 * the concept of user_features has attempted to create a new
1638 			 * datapath and is likely to reuse it. Drop all user features.
1639 			 */
1640 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1641 				ovs_dp_reset_user_features(skb, info);
1642 		}
1643 
1644 		goto err_destroy_ports_array;
1645 	}
1646 
1647 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1648 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1649 	BUG_ON(err < 0);
1650 
1651 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1652 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1653 
1654 	ovs_unlock();
1655 
1656 	ovs_notify(&dp_datapath_genl_family, reply, info);
1657 	return 0;
1658 
1659 err_destroy_ports_array:
1660 	ovs_unlock();
1661 	kfree(dp->ports);
1662 err_destroy_percpu:
1663 	free_percpu(dp->stats_percpu);
1664 err_destroy_table:
1665 	ovs_flow_tbl_destroy(&dp->table);
1666 err_free_dp:
1667 	kfree(dp);
1668 err_free_reply:
1669 	kfree_skb(reply);
1670 err:
1671 	return err;
1672 }
1673 
1674 /* Called with ovs_mutex. */
1675 static void __dp_destroy(struct datapath *dp)
1676 {
1677 	int i;
1678 
1679 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1680 		struct vport *vport;
1681 		struct hlist_node *n;
1682 
1683 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1684 			if (vport->port_no != OVSP_LOCAL)
1685 				ovs_dp_detach_port(vport);
1686 	}
1687 
1688 	list_del_rcu(&dp->list_node);
1689 
1690 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1691 	 * all ports in datapath are destroyed first before freeing datapath.
1692 	 */
1693 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1694 
1695 	/* RCU destroy the flow table */
1696 	call_rcu(&dp->rcu, destroy_dp_rcu);
1697 }
1698 
1699 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1700 {
1701 	struct sk_buff *reply;
1702 	struct datapath *dp;
1703 	int err;
1704 
1705 	reply = ovs_dp_cmd_alloc_info();
1706 	if (!reply)
1707 		return -ENOMEM;
1708 
1709 	ovs_lock();
1710 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1711 	err = PTR_ERR(dp);
1712 	if (IS_ERR(dp))
1713 		goto err_unlock_free;
1714 
1715 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1716 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1717 	BUG_ON(err < 0);
1718 
1719 	__dp_destroy(dp);
1720 	ovs_unlock();
1721 
1722 	ovs_notify(&dp_datapath_genl_family, reply, info);
1723 
1724 	return 0;
1725 
1726 err_unlock_free:
1727 	ovs_unlock();
1728 	kfree_skb(reply);
1729 	return err;
1730 }
1731 
1732 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1733 {
1734 	struct sk_buff *reply;
1735 	struct datapath *dp;
1736 	int err;
1737 
1738 	reply = ovs_dp_cmd_alloc_info();
1739 	if (!reply)
1740 		return -ENOMEM;
1741 
1742 	ovs_lock();
1743 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1744 	err = PTR_ERR(dp);
1745 	if (IS_ERR(dp))
1746 		goto err_unlock_free;
1747 
1748 	ovs_dp_change(dp, info->attrs);
1749 
1750 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1751 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1752 	BUG_ON(err < 0);
1753 
1754 	ovs_unlock();
1755 	ovs_notify(&dp_datapath_genl_family, reply, info);
1756 
1757 	return 0;
1758 
1759 err_unlock_free:
1760 	ovs_unlock();
1761 	kfree_skb(reply);
1762 	return err;
1763 }
1764 
1765 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1766 {
1767 	struct sk_buff *reply;
1768 	struct datapath *dp;
1769 	int err;
1770 
1771 	reply = ovs_dp_cmd_alloc_info();
1772 	if (!reply)
1773 		return -ENOMEM;
1774 
1775 	ovs_lock();
1776 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1777 	if (IS_ERR(dp)) {
1778 		err = PTR_ERR(dp);
1779 		goto err_unlock_free;
1780 	}
1781 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1782 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1783 	BUG_ON(err < 0);
1784 	ovs_unlock();
1785 
1786 	return genlmsg_reply(reply, info);
1787 
1788 err_unlock_free:
1789 	ovs_unlock();
1790 	kfree_skb(reply);
1791 	return err;
1792 }
1793 
1794 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1795 {
1796 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1797 	struct datapath *dp;
1798 	int skip = cb->args[0];
1799 	int i = 0;
1800 
1801 	ovs_lock();
1802 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1803 		if (i >= skip &&
1804 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1805 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1806 					 OVS_DP_CMD_NEW) < 0)
1807 			break;
1808 		i++;
1809 	}
1810 	ovs_unlock();
1811 
1812 	cb->args[0] = i;
1813 
1814 	return skb->len;
1815 }
1816 
1817 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1818 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1819 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1820 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1821 };
1822 
1823 static const struct genl_ops dp_datapath_genl_ops[] = {
1824 	{ .cmd = OVS_DP_CMD_NEW,
1825 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1826 	  .policy = datapath_policy,
1827 	  .doit = ovs_dp_cmd_new
1828 	},
1829 	{ .cmd = OVS_DP_CMD_DEL,
1830 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1831 	  .policy = datapath_policy,
1832 	  .doit = ovs_dp_cmd_del
1833 	},
1834 	{ .cmd = OVS_DP_CMD_GET,
1835 	  .flags = 0,		    /* OK for unprivileged users. */
1836 	  .policy = datapath_policy,
1837 	  .doit = ovs_dp_cmd_get,
1838 	  .dumpit = ovs_dp_cmd_dump
1839 	},
1840 	{ .cmd = OVS_DP_CMD_SET,
1841 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1842 	  .policy = datapath_policy,
1843 	  .doit = ovs_dp_cmd_set,
1844 	},
1845 };
1846 
1847 static struct genl_family dp_datapath_genl_family __ro_after_init = {
1848 	.hdrsize = sizeof(struct ovs_header),
1849 	.name = OVS_DATAPATH_FAMILY,
1850 	.version = OVS_DATAPATH_VERSION,
1851 	.maxattr = OVS_DP_ATTR_MAX,
1852 	.netnsok = true,
1853 	.parallel_ops = true,
1854 	.ops = dp_datapath_genl_ops,
1855 	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1856 	.mcgrps = &ovs_dp_datapath_multicast_group,
1857 	.n_mcgrps = 1,
1858 	.module = THIS_MODULE,
1859 };
1860 
1861 /* Called with ovs_mutex or RCU read lock. */
1862 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1863 				   u32 portid, u32 seq, u32 flags, u8 cmd)
1864 {
1865 	struct ovs_header *ovs_header;
1866 	struct ovs_vport_stats vport_stats;
1867 	int err;
1868 
1869 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1870 				 flags, cmd);
1871 	if (!ovs_header)
1872 		return -EMSGSIZE;
1873 
1874 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1875 
1876 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1877 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1878 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1879 			   ovs_vport_name(vport)))
1880 		goto nla_put_failure;
1881 
1882 	ovs_vport_get_stats(vport, &vport_stats);
1883 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
1884 			  sizeof(struct ovs_vport_stats), &vport_stats,
1885 			  OVS_VPORT_ATTR_PAD))
1886 		goto nla_put_failure;
1887 
1888 	if (ovs_vport_get_upcall_portids(vport, skb))
1889 		goto nla_put_failure;
1890 
1891 	err = ovs_vport_get_options(vport, skb);
1892 	if (err == -EMSGSIZE)
1893 		goto error;
1894 
1895 	genlmsg_end(skb, ovs_header);
1896 	return 0;
1897 
1898 nla_put_failure:
1899 	err = -EMSGSIZE;
1900 error:
1901 	genlmsg_cancel(skb, ovs_header);
1902 	return err;
1903 }
1904 
1905 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1906 {
1907 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1908 }
1909 
1910 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1911 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1912 					 u32 seq, u8 cmd)
1913 {
1914 	struct sk_buff *skb;
1915 	int retval;
1916 
1917 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1918 	if (!skb)
1919 		return ERR_PTR(-ENOMEM);
1920 
1921 	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1922 	BUG_ON(retval < 0);
1923 
1924 	return skb;
1925 }
1926 
1927 /* Called with ovs_mutex or RCU read lock. */
1928 static struct vport *lookup_vport(struct net *net,
1929 				  const struct ovs_header *ovs_header,
1930 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1931 {
1932 	struct datapath *dp;
1933 	struct vport *vport;
1934 
1935 	if (a[OVS_VPORT_ATTR_NAME]) {
1936 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1937 		if (!vport)
1938 			return ERR_PTR(-ENODEV);
1939 		if (ovs_header->dp_ifindex &&
1940 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1941 			return ERR_PTR(-ENODEV);
1942 		return vport;
1943 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1944 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1945 
1946 		if (port_no >= DP_MAX_PORTS)
1947 			return ERR_PTR(-EFBIG);
1948 
1949 		dp = get_dp(net, ovs_header->dp_ifindex);
1950 		if (!dp)
1951 			return ERR_PTR(-ENODEV);
1952 
1953 		vport = ovs_vport_ovsl_rcu(dp, port_no);
1954 		if (!vport)
1955 			return ERR_PTR(-ENODEV);
1956 		return vport;
1957 	} else
1958 		return ERR_PTR(-EINVAL);
1959 }
1960 
1961 /* Called with ovs_mutex */
1962 static void update_headroom(struct datapath *dp)
1963 {
1964 	unsigned dev_headroom, max_headroom = 0;
1965 	struct net_device *dev;
1966 	struct vport *vport;
1967 	int i;
1968 
1969 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1970 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1971 			dev = vport->dev;
1972 			dev_headroom = netdev_get_fwd_headroom(dev);
1973 			if (dev_headroom > max_headroom)
1974 				max_headroom = dev_headroom;
1975 		}
1976 	}
1977 
1978 	dp->max_headroom = max_headroom;
1979 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1980 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node)
1981 			netdev_set_rx_headroom(vport->dev, max_headroom);
1982 }
1983 
1984 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1985 {
1986 	struct nlattr **a = info->attrs;
1987 	struct ovs_header *ovs_header = info->userhdr;
1988 	struct vport_parms parms;
1989 	struct sk_buff *reply;
1990 	struct vport *vport;
1991 	struct datapath *dp;
1992 	u32 port_no;
1993 	int err;
1994 
1995 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1996 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1997 		return -EINVAL;
1998 
1999 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2000 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2001 	if (port_no >= DP_MAX_PORTS)
2002 		return -EFBIG;
2003 
2004 	reply = ovs_vport_cmd_alloc_info();
2005 	if (!reply)
2006 		return -ENOMEM;
2007 
2008 	ovs_lock();
2009 restart:
2010 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2011 	err = -ENODEV;
2012 	if (!dp)
2013 		goto exit_unlock_free;
2014 
2015 	if (port_no) {
2016 		vport = ovs_vport_ovsl(dp, port_no);
2017 		err = -EBUSY;
2018 		if (vport)
2019 			goto exit_unlock_free;
2020 	} else {
2021 		for (port_no = 1; ; port_no++) {
2022 			if (port_no >= DP_MAX_PORTS) {
2023 				err = -EFBIG;
2024 				goto exit_unlock_free;
2025 			}
2026 			vport = ovs_vport_ovsl(dp, port_no);
2027 			if (!vport)
2028 				break;
2029 		}
2030 	}
2031 
2032 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2033 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2034 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2035 	parms.dp = dp;
2036 	parms.port_no = port_no;
2037 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2038 
2039 	vport = new_vport(&parms);
2040 	err = PTR_ERR(vport);
2041 	if (IS_ERR(vport)) {
2042 		if (err == -EAGAIN)
2043 			goto restart;
2044 		goto exit_unlock_free;
2045 	}
2046 
2047 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2048 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2049 
2050 	if (netdev_get_fwd_headroom(vport->dev) > dp->max_headroom)
2051 		update_headroom(dp);
2052 	else
2053 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2054 
2055 	BUG_ON(err < 0);
2056 	ovs_unlock();
2057 
2058 	ovs_notify(&dp_vport_genl_family, reply, info);
2059 	return 0;
2060 
2061 exit_unlock_free:
2062 	ovs_unlock();
2063 	kfree_skb(reply);
2064 	return err;
2065 }
2066 
2067 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2068 {
2069 	struct nlattr **a = info->attrs;
2070 	struct sk_buff *reply;
2071 	struct vport *vport;
2072 	int err;
2073 
2074 	reply = ovs_vport_cmd_alloc_info();
2075 	if (!reply)
2076 		return -ENOMEM;
2077 
2078 	ovs_lock();
2079 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2080 	err = PTR_ERR(vport);
2081 	if (IS_ERR(vport))
2082 		goto exit_unlock_free;
2083 
2084 	if (a[OVS_VPORT_ATTR_TYPE] &&
2085 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2086 		err = -EINVAL;
2087 		goto exit_unlock_free;
2088 	}
2089 
2090 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2091 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2092 		if (err)
2093 			goto exit_unlock_free;
2094 	}
2095 
2096 
2097 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2098 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2099 
2100 		err = ovs_vport_set_upcall_portids(vport, ids);
2101 		if (err)
2102 			goto exit_unlock_free;
2103 	}
2104 
2105 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2106 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2107 	BUG_ON(err < 0);
2108 
2109 	ovs_unlock();
2110 	ovs_notify(&dp_vport_genl_family, reply, info);
2111 	return 0;
2112 
2113 exit_unlock_free:
2114 	ovs_unlock();
2115 	kfree_skb(reply);
2116 	return err;
2117 }
2118 
2119 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2120 {
2121 	bool must_update_headroom = false;
2122 	struct nlattr **a = info->attrs;
2123 	struct sk_buff *reply;
2124 	struct datapath *dp;
2125 	struct vport *vport;
2126 	int err;
2127 
2128 	reply = ovs_vport_cmd_alloc_info();
2129 	if (!reply)
2130 		return -ENOMEM;
2131 
2132 	ovs_lock();
2133 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2134 	err = PTR_ERR(vport);
2135 	if (IS_ERR(vport))
2136 		goto exit_unlock_free;
2137 
2138 	if (vport->port_no == OVSP_LOCAL) {
2139 		err = -EINVAL;
2140 		goto exit_unlock_free;
2141 	}
2142 
2143 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2144 				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2145 	BUG_ON(err < 0);
2146 
2147 	/* the vport deletion may trigger dp headroom update */
2148 	dp = vport->dp;
2149 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2150 		must_update_headroom = true;
2151 	netdev_reset_rx_headroom(vport->dev);
2152 	ovs_dp_detach_port(vport);
2153 
2154 	if (must_update_headroom)
2155 		update_headroom(dp);
2156 	ovs_unlock();
2157 
2158 	ovs_notify(&dp_vport_genl_family, reply, info);
2159 	return 0;
2160 
2161 exit_unlock_free:
2162 	ovs_unlock();
2163 	kfree_skb(reply);
2164 	return err;
2165 }
2166 
2167 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2168 {
2169 	struct nlattr **a = info->attrs;
2170 	struct ovs_header *ovs_header = info->userhdr;
2171 	struct sk_buff *reply;
2172 	struct vport *vport;
2173 	int err;
2174 
2175 	reply = ovs_vport_cmd_alloc_info();
2176 	if (!reply)
2177 		return -ENOMEM;
2178 
2179 	rcu_read_lock();
2180 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2181 	err = PTR_ERR(vport);
2182 	if (IS_ERR(vport))
2183 		goto exit_unlock_free;
2184 	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2185 				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2186 	BUG_ON(err < 0);
2187 	rcu_read_unlock();
2188 
2189 	return genlmsg_reply(reply, info);
2190 
2191 exit_unlock_free:
2192 	rcu_read_unlock();
2193 	kfree_skb(reply);
2194 	return err;
2195 }
2196 
2197 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2198 {
2199 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2200 	struct datapath *dp;
2201 	int bucket = cb->args[0], skip = cb->args[1];
2202 	int i, j = 0;
2203 
2204 	rcu_read_lock();
2205 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2206 	if (!dp) {
2207 		rcu_read_unlock();
2208 		return -ENODEV;
2209 	}
2210 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2211 		struct vport *vport;
2212 
2213 		j = 0;
2214 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2215 			if (j >= skip &&
2216 			    ovs_vport_cmd_fill_info(vport, skb,
2217 						    NETLINK_CB(cb->skb).portid,
2218 						    cb->nlh->nlmsg_seq,
2219 						    NLM_F_MULTI,
2220 						    OVS_VPORT_CMD_NEW) < 0)
2221 				goto out;
2222 
2223 			j++;
2224 		}
2225 		skip = 0;
2226 	}
2227 out:
2228 	rcu_read_unlock();
2229 
2230 	cb->args[0] = i;
2231 	cb->args[1] = j;
2232 
2233 	return skb->len;
2234 }
2235 
2236 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2237 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2238 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2239 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2240 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2241 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2242 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2243 };
2244 
2245 static const struct genl_ops dp_vport_genl_ops[] = {
2246 	{ .cmd = OVS_VPORT_CMD_NEW,
2247 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2248 	  .policy = vport_policy,
2249 	  .doit = ovs_vport_cmd_new
2250 	},
2251 	{ .cmd = OVS_VPORT_CMD_DEL,
2252 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2253 	  .policy = vport_policy,
2254 	  .doit = ovs_vport_cmd_del
2255 	},
2256 	{ .cmd = OVS_VPORT_CMD_GET,
2257 	  .flags = 0,		    /* OK for unprivileged users. */
2258 	  .policy = vport_policy,
2259 	  .doit = ovs_vport_cmd_get,
2260 	  .dumpit = ovs_vport_cmd_dump
2261 	},
2262 	{ .cmd = OVS_VPORT_CMD_SET,
2263 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2264 	  .policy = vport_policy,
2265 	  .doit = ovs_vport_cmd_set,
2266 	},
2267 };
2268 
2269 struct genl_family dp_vport_genl_family __ro_after_init = {
2270 	.hdrsize = sizeof(struct ovs_header),
2271 	.name = OVS_VPORT_FAMILY,
2272 	.version = OVS_VPORT_VERSION,
2273 	.maxattr = OVS_VPORT_ATTR_MAX,
2274 	.netnsok = true,
2275 	.parallel_ops = true,
2276 	.ops = dp_vport_genl_ops,
2277 	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2278 	.mcgrps = &ovs_dp_vport_multicast_group,
2279 	.n_mcgrps = 1,
2280 	.module = THIS_MODULE,
2281 };
2282 
2283 static struct genl_family * const dp_genl_families[] = {
2284 	&dp_datapath_genl_family,
2285 	&dp_vport_genl_family,
2286 	&dp_flow_genl_family,
2287 	&dp_packet_genl_family,
2288 };
2289 
2290 static void dp_unregister_genl(int n_families)
2291 {
2292 	int i;
2293 
2294 	for (i = 0; i < n_families; i++)
2295 		genl_unregister_family(dp_genl_families[i]);
2296 }
2297 
2298 static int __init dp_register_genl(void)
2299 {
2300 	int err;
2301 	int i;
2302 
2303 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2304 
2305 		err = genl_register_family(dp_genl_families[i]);
2306 		if (err)
2307 			goto error;
2308 	}
2309 
2310 	return 0;
2311 
2312 error:
2313 	dp_unregister_genl(i);
2314 	return err;
2315 }
2316 
2317 static int __net_init ovs_init_net(struct net *net)
2318 {
2319 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2320 
2321 	INIT_LIST_HEAD(&ovs_net->dps);
2322 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2323 	ovs_ct_init(net);
2324 	return 0;
2325 }
2326 
2327 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2328 					    struct list_head *head)
2329 {
2330 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2331 	struct datapath *dp;
2332 
2333 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2334 		int i;
2335 
2336 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2337 			struct vport *vport;
2338 
2339 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2340 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2341 					continue;
2342 
2343 				if (dev_net(vport->dev) == dnet)
2344 					list_add(&vport->detach_list, head);
2345 			}
2346 		}
2347 	}
2348 }
2349 
2350 static void __net_exit ovs_exit_net(struct net *dnet)
2351 {
2352 	struct datapath *dp, *dp_next;
2353 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2354 	struct vport *vport, *vport_next;
2355 	struct net *net;
2356 	LIST_HEAD(head);
2357 
2358 	ovs_ct_exit(dnet);
2359 	ovs_lock();
2360 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2361 		__dp_destroy(dp);
2362 
2363 	rtnl_lock();
2364 	for_each_net(net)
2365 		list_vports_from_net(net, dnet, &head);
2366 	rtnl_unlock();
2367 
2368 	/* Detach all vports from given namespace. */
2369 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2370 		list_del(&vport->detach_list);
2371 		ovs_dp_detach_port(vport);
2372 	}
2373 
2374 	ovs_unlock();
2375 
2376 	cancel_work_sync(&ovs_net->dp_notify_work);
2377 }
2378 
2379 static struct pernet_operations ovs_net_ops = {
2380 	.init = ovs_init_net,
2381 	.exit = ovs_exit_net,
2382 	.id   = &ovs_net_id,
2383 	.size = sizeof(struct ovs_net),
2384 };
2385 
2386 static int __init dp_init(void)
2387 {
2388 	int err;
2389 
2390 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2391 
2392 	pr_info("Open vSwitch switching datapath\n");
2393 
2394 	err = action_fifos_init();
2395 	if (err)
2396 		goto error;
2397 
2398 	err = ovs_internal_dev_rtnl_link_register();
2399 	if (err)
2400 		goto error_action_fifos_exit;
2401 
2402 	err = ovs_flow_init();
2403 	if (err)
2404 		goto error_unreg_rtnl_link;
2405 
2406 	err = ovs_vport_init();
2407 	if (err)
2408 		goto error_flow_exit;
2409 
2410 	err = register_pernet_device(&ovs_net_ops);
2411 	if (err)
2412 		goto error_vport_exit;
2413 
2414 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2415 	if (err)
2416 		goto error_netns_exit;
2417 
2418 	err = ovs_netdev_init();
2419 	if (err)
2420 		goto error_unreg_notifier;
2421 
2422 	err = dp_register_genl();
2423 	if (err < 0)
2424 		goto error_unreg_netdev;
2425 
2426 	return 0;
2427 
2428 error_unreg_netdev:
2429 	ovs_netdev_exit();
2430 error_unreg_notifier:
2431 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2432 error_netns_exit:
2433 	unregister_pernet_device(&ovs_net_ops);
2434 error_vport_exit:
2435 	ovs_vport_exit();
2436 error_flow_exit:
2437 	ovs_flow_exit();
2438 error_unreg_rtnl_link:
2439 	ovs_internal_dev_rtnl_link_unregister();
2440 error_action_fifos_exit:
2441 	action_fifos_exit();
2442 error:
2443 	return err;
2444 }
2445 
2446 static void dp_cleanup(void)
2447 {
2448 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2449 	ovs_netdev_exit();
2450 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2451 	unregister_pernet_device(&ovs_net_ops);
2452 	rcu_barrier();
2453 	ovs_vport_exit();
2454 	ovs_flow_exit();
2455 	ovs_internal_dev_rtnl_link_unregister();
2456 	action_fifos_exit();
2457 }
2458 
2459 module_init(dp_init);
2460 module_exit(dp_cleanup);
2461 
2462 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2463 MODULE_LICENSE("GPL");
2464 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2465 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2466 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2467 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2468