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
61int ovs_net_id __read_mostly;
62EXPORT_SYMBOL_GPL(ovs_net_id);
63
64static struct genl_family dp_packet_genl_family;
65static struct genl_family dp_flow_genl_family;
66static struct genl_family dp_datapath_genl_family;
67
68static const struct nla_policy flow_policy[];
69
70static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
71	.name = OVS_FLOW_MCGROUP,
72};
73
74static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
75	.name = OVS_DATAPATH_MCGROUP,
76};
77
78static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
79	.name = OVS_VPORT_MCGROUP,
80};
81
82/* Check if need to build a reply message.
83 * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
84static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
85			    unsigned int group)
86{
87	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
88	       genl_has_listeners(family, genl_info_net(info), group);
89}
90
91static void ovs_notify(struct genl_family *family,
92		       struct sk_buff *skb, struct genl_info *info)
93{
94	genl_notify(family, skb, genl_info_net(info), info->snd_portid,
95		    0, info->nlhdr, GFP_KERNEL);
96}
97
98/**
99 * DOC: Locking:
100 *
101 * All writes e.g. Writes to device state (add/remove datapath, port, set
102 * operations on vports, etc.), Writes to other state (flow table
103 * modifications, set miscellaneous datapath parameters, etc.) are protected
104 * by ovs_lock.
105 *
106 * Reads are protected by RCU.
107 *
108 * There are a few special cases (mostly stats) that have their own
109 * synchronization but they nest under all of above and don't interact with
110 * each other.
111 *
112 * The RTNL lock nests inside ovs_mutex.
113 */
114
115static DEFINE_MUTEX(ovs_mutex);
116
117void ovs_lock(void)
118{
119	mutex_lock(&ovs_mutex);
120}
121
122void ovs_unlock(void)
123{
124	mutex_unlock(&ovs_mutex);
125}
126
127#ifdef CONFIG_LOCKDEP
128int lockdep_ovsl_is_held(void)
129{
130	if (debug_locks)
131		return lockdep_is_held(&ovs_mutex);
132	else
133		return 1;
134}
135EXPORT_SYMBOL_GPL(lockdep_ovsl_is_held);
136#endif
137
138static struct vport *new_vport(const struct vport_parms *);
139static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
140			     const struct sw_flow_key *,
141			     const struct dp_upcall_info *);
142static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
143				  const struct sw_flow_key *,
144				  const struct dp_upcall_info *);
145
146/* Must be called with rcu_read_lock. */
147static struct datapath *get_dp_rcu(struct net *net, int dp_ifindex)
148{
149	struct net_device *dev = dev_get_by_index_rcu(net, dp_ifindex);
150
151	if (dev) {
152		struct vport *vport = ovs_internal_dev_get_vport(dev);
153		if (vport)
154			return vport->dp;
155	}
156
157	return NULL;
158}
159
160/* The caller must hold either ovs_mutex or rcu_read_lock to keep the
161 * returned dp pointer valid.
162 */
163static inline struct datapath *get_dp(struct net *net, int dp_ifindex)
164{
165	struct datapath *dp;
166
167	WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_ovsl_is_held());
168	rcu_read_lock();
169	dp = get_dp_rcu(net, dp_ifindex);
170	rcu_read_unlock();
171
172	return dp;
173}
174
175/* Must be called with rcu_read_lock or ovs_mutex. */
176const char *ovs_dp_name(const struct datapath *dp)
177{
178	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
179	return vport->ops->get_name(vport);
180}
181
182static int get_dpifindex(const struct datapath *dp)
183{
184	struct vport *local;
185	int ifindex;
186
187	rcu_read_lock();
188
189	local = ovs_vport_rcu(dp, OVSP_LOCAL);
190	if (local)
191		ifindex = netdev_vport_priv(local)->dev->ifindex;
192	else
193		ifindex = 0;
194
195	rcu_read_unlock();
196
197	return ifindex;
198}
199
200static void destroy_dp_rcu(struct rcu_head *rcu)
201{
202	struct datapath *dp = container_of(rcu, struct datapath, rcu);
203
204	ovs_flow_tbl_destroy(&dp->table);
205	free_percpu(dp->stats_percpu);
206	kfree(dp->ports);
207	kfree(dp);
208}
209
210static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
211					    u16 port_no)
212{
213	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
214}
215
216/* Called with ovs_mutex or RCU read lock. */
217struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
218{
219	struct vport *vport;
220	struct hlist_head *head;
221
222	head = vport_hash_bucket(dp, port_no);
223	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
224		if (vport->port_no == port_no)
225			return vport;
226	}
227	return NULL;
228}
229
230/* Called with ovs_mutex. */
231static struct vport *new_vport(const struct vport_parms *parms)
232{
233	struct vport *vport;
234
235	vport = ovs_vport_add(parms);
236	if (!IS_ERR(vport)) {
237		struct datapath *dp = parms->dp;
238		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
239
240		hlist_add_head_rcu(&vport->dp_hash_node, head);
241	}
242	return vport;
243}
244
245void ovs_dp_detach_port(struct vport *p)
246{
247	ASSERT_OVSL();
248
249	/* First drop references to device. */
250	hlist_del_rcu(&p->dp_hash_node);
251
252	/* Then destroy it. */
253	ovs_vport_del(p);
254}
255
256/* Must be called with rcu_read_lock. */
257void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
258{
259	const struct vport *p = OVS_CB(skb)->input_vport;
260	struct datapath *dp = p->dp;
261	struct sw_flow *flow;
262	struct sw_flow_actions *sf_acts;
263	struct dp_stats_percpu *stats;
264	u64 *stats_counter;
265	u32 n_mask_hit;
266
267	stats = this_cpu_ptr(dp->stats_percpu);
268
269	/* Look up flow. */
270	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, &n_mask_hit);
271	if (unlikely(!flow)) {
272		struct dp_upcall_info upcall;
273		int error;
274
275		upcall.cmd = OVS_PACKET_CMD_MISS;
276		upcall.userdata = NULL;
277		upcall.portid = ovs_vport_find_upcall_portid(p, skb);
278		upcall.egress_tun_info = NULL;
279		error = ovs_dp_upcall(dp, skb, key, &upcall);
280		if (unlikely(error))
281			kfree_skb(skb);
282		else
283			consume_skb(skb);
284		stats_counter = &stats->n_missed;
285		goto out;
286	}
287
288	ovs_flow_stats_update(flow, key->tp.flags, skb);
289	sf_acts = rcu_dereference(flow->sf_acts);
290	ovs_execute_actions(dp, skb, sf_acts, key);
291
292	stats_counter = &stats->n_hit;
293
294out:
295	/* Update datapath statistics. */
296	u64_stats_update_begin(&stats->syncp);
297	(*stats_counter)++;
298	stats->n_mask_hit += n_mask_hit;
299	u64_stats_update_end(&stats->syncp);
300}
301
302int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
303		  const struct sw_flow_key *key,
304		  const struct dp_upcall_info *upcall_info)
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);
316	else
317		err = queue_gso_packets(dp, skb, key, upcall_info);
318	if (err)
319		goto err;
320
321	return 0;
322
323err:
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
333static 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{
337	unsigned short gso_type = skb_shinfo(skb)->gso_type;
338	struct sw_flow_key later_key;
339	struct sk_buff *segs, *nskb;
340	int err;
341
342	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_SGO_CB_OFFSET);
343	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
344	if (IS_ERR(segs))
345		return PTR_ERR(segs);
346	if (segs == NULL)
347		return -EINVAL;
348
349	if (gso_type & SKB_GSO_UDP) {
350		/* The initial flow key extracted by ovs_flow_key_extract()
351		 * in this case is for a first fragment, so we need to
352		 * properly mark later fragments.
353		 */
354		later_key = *key;
355		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
356	}
357
358	/* Queue all of the segments. */
359	skb = segs;
360	do {
361		if (gso_type & SKB_GSO_UDP && skb != segs)
362			key = &later_key;
363
364		err = queue_userspace_packet(dp, skb, key, upcall_info);
365		if (err)
366			break;
367
368	} while ((skb = skb->next));
369
370	/* Free all of the segments. */
371	skb = segs;
372	do {
373		nskb = skb->next;
374		if (err)
375			kfree_skb(skb);
376		else
377			consume_skb(skb);
378	} while ((skb = nskb));
379	return err;
380}
381
382static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
383			      unsigned int hdrlen)
384{
385	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
386		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
387		+ nla_total_size(ovs_key_attr_size()); /* OVS_PACKET_ATTR_KEY */
388
389	/* OVS_PACKET_ATTR_USERDATA */
390	if (upcall_info->userdata)
391		size += NLA_ALIGN(upcall_info->userdata->nla_len);
392
393	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
394	if (upcall_info->egress_tun_info)
395		size += nla_total_size(ovs_tun_key_attr_size());
396
397	return size;
398}
399
400static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
401				  const struct sw_flow_key *key,
402				  const struct dp_upcall_info *upcall_info)
403{
404	struct ovs_header *upcall;
405	struct sk_buff *nskb = NULL;
406	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
407	struct nlattr *nla;
408	struct genl_info info = {
409		.dst_sk = ovs_dp_get_net(dp)->genl_sock,
410		.snd_portid = upcall_info->portid,
411	};
412	size_t len;
413	unsigned int hlen;
414	int err, dp_ifindex;
415
416	dp_ifindex = get_dpifindex(dp);
417	if (!dp_ifindex)
418		return -ENODEV;
419
420	if (skb_vlan_tag_present(skb)) {
421		nskb = skb_clone(skb, GFP_ATOMIC);
422		if (!nskb)
423			return -ENOMEM;
424
425		nskb = __vlan_hwaccel_push_inside(nskb);
426		if (!nskb)
427			return -ENOMEM;
428
429		skb = nskb;
430	}
431
432	if (nla_attr_size(skb->len) > USHRT_MAX) {
433		err = -EFBIG;
434		goto out;
435	}
436
437	/* Complete checksum if needed */
438	if (skb->ip_summed == CHECKSUM_PARTIAL &&
439	    (err = skb_checksum_help(skb)))
440		goto out;
441
442	/* Older versions of OVS user space enforce alignment of the last
443	 * Netlink attribute to NLA_ALIGNTO which would require extensive
444	 * padding logic. Only perform zerocopy if padding is not required.
445	 */
446	if (dp->user_features & OVS_DP_F_UNALIGNED)
447		hlen = skb_zerocopy_headlen(skb);
448	else
449		hlen = skb->len;
450
451	len = upcall_msg_size(upcall_info, hlen);
452	user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
453	if (!user_skb) {
454		err = -ENOMEM;
455		goto out;
456	}
457
458	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
459			     0, upcall_info->cmd);
460	upcall->dp_ifindex = dp_ifindex;
461
462	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
463	BUG_ON(err);
464
465	if (upcall_info->userdata)
466		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
467			  nla_len(upcall_info->userdata),
468			  nla_data(upcall_info->userdata));
469
470	if (upcall_info->egress_tun_info) {
471		nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_EGRESS_TUN_KEY);
472		err = ovs_nla_put_egress_tunnel_key(user_skb,
473						    upcall_info->egress_tun_info);
474		BUG_ON(err);
475		nla_nest_end(user_skb, nla);
476	}
477
478	/* Only reserve room for attribute header, packet data is added
479	 * in skb_zerocopy() */
480	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
481		err = -ENOBUFS;
482		goto out;
483	}
484	nla->nla_len = nla_attr_size(skb->len);
485
486	err = skb_zerocopy(user_skb, skb, skb->len, hlen);
487	if (err)
488		goto out;
489
490	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
491	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
492		size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
493
494		if (plen > 0)
495			memset(skb_put(user_skb, plen), 0, plen);
496	}
497
498	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
499
500	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
501	user_skb = NULL;
502out:
503	if (err)
504		skb_tx_error(skb);
505	kfree_skb(user_skb);
506	kfree_skb(nskb);
507	return err;
508}
509
510static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
511{
512	struct ovs_header *ovs_header = info->userhdr;
513	struct nlattr **a = info->attrs;
514	struct sw_flow_actions *acts;
515	struct sk_buff *packet;
516	struct sw_flow *flow;
517	struct sw_flow_actions *sf_acts;
518	struct datapath *dp;
519	struct ethhdr *eth;
520	struct vport *input_vport;
521	int len;
522	int err;
523	bool log = !a[OVS_PACKET_ATTR_PROBE];
524
525	err = -EINVAL;
526	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
527	    !a[OVS_PACKET_ATTR_ACTIONS])
528		goto err;
529
530	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
531	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
532	err = -ENOMEM;
533	if (!packet)
534		goto err;
535	skb_reserve(packet, NET_IP_ALIGN);
536
537	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
538
539	skb_reset_mac_header(packet);
540	eth = eth_hdr(packet);
541
542	/* Normally, setting the skb 'protocol' field would be handled by a
543	 * call to eth_type_trans(), but it assumes there's a sending
544	 * device, which we may not have. */
545	if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
546		packet->protocol = eth->h_proto;
547	else
548		packet->protocol = htons(ETH_P_802_2);
549
550	/* Build an sw_flow for sending this packet. */
551	flow = ovs_flow_alloc();
552	err = PTR_ERR(flow);
553	if (IS_ERR(flow))
554		goto err_kfree_skb;
555
556	err = ovs_flow_key_extract_userspace(a[OVS_PACKET_ATTR_KEY], packet,
557					     &flow->key, log);
558	if (err)
559		goto err_flow_free;
560
561	err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
562				   &flow->key, &acts, log);
563	if (err)
564		goto err_flow_free;
565
566	rcu_assign_pointer(flow->sf_acts, acts);
567	OVS_CB(packet)->egress_tun_info = NULL;
568	packet->priority = flow->key.phy.priority;
569	packet->mark = flow->key.phy.skb_mark;
570
571	rcu_read_lock();
572	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
573	err = -ENODEV;
574	if (!dp)
575		goto err_unlock;
576
577	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
578	if (!input_vport)
579		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
580
581	if (!input_vport)
582		goto err_unlock;
583
584	OVS_CB(packet)->input_vport = input_vport;
585	sf_acts = rcu_dereference(flow->sf_acts);
586
587	local_bh_disable();
588	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
589	local_bh_enable();
590	rcu_read_unlock();
591
592	ovs_flow_free(flow, false);
593	return err;
594
595err_unlock:
596	rcu_read_unlock();
597err_flow_free:
598	ovs_flow_free(flow, false);
599err_kfree_skb:
600	kfree_skb(packet);
601err:
602	return err;
603}
604
605static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
606	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
607	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
608	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
609	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
610};
611
612static const struct genl_ops dp_packet_genl_ops[] = {
613	{ .cmd = OVS_PACKET_CMD_EXECUTE,
614	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
615	  .policy = packet_policy,
616	  .doit = ovs_packet_cmd_execute
617	}
618};
619
620static struct genl_family dp_packet_genl_family = {
621	.id = GENL_ID_GENERATE,
622	.hdrsize = sizeof(struct ovs_header),
623	.name = OVS_PACKET_FAMILY,
624	.version = OVS_PACKET_VERSION,
625	.maxattr = OVS_PACKET_ATTR_MAX,
626	.netnsok = true,
627	.parallel_ops = true,
628	.ops = dp_packet_genl_ops,
629	.n_ops = ARRAY_SIZE(dp_packet_genl_ops),
630};
631
632static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
633			 struct ovs_dp_megaflow_stats *mega_stats)
634{
635	int i;
636
637	memset(mega_stats, 0, sizeof(*mega_stats));
638
639	stats->n_flows = ovs_flow_tbl_count(&dp->table);
640	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
641
642	stats->n_hit = stats->n_missed = stats->n_lost = 0;
643
644	for_each_possible_cpu(i) {
645		const struct dp_stats_percpu *percpu_stats;
646		struct dp_stats_percpu local_stats;
647		unsigned int start;
648
649		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
650
651		do {
652			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
653			local_stats = *percpu_stats;
654		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
655
656		stats->n_hit += local_stats.n_hit;
657		stats->n_missed += local_stats.n_missed;
658		stats->n_lost += local_stats.n_lost;
659		mega_stats->n_mask_hit += local_stats.n_mask_hit;
660	}
661}
662
663static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
664{
665	return ovs_identifier_is_ufid(sfid) &&
666	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
667}
668
669static bool should_fill_mask(uint32_t ufid_flags)
670{
671	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
672}
673
674static bool should_fill_actions(uint32_t ufid_flags)
675{
676	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
677}
678
679static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
680				    const struct sw_flow_id *sfid,
681				    uint32_t ufid_flags)
682{
683	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
684
685	/* OVS_FLOW_ATTR_UFID */
686	if (sfid && ovs_identifier_is_ufid(sfid))
687		len += nla_total_size(sfid->ufid_len);
688
689	/* OVS_FLOW_ATTR_KEY */
690	if (!sfid || should_fill_key(sfid, ufid_flags))
691		len += nla_total_size(ovs_key_attr_size());
692
693	/* OVS_FLOW_ATTR_MASK */
694	if (should_fill_mask(ufid_flags))
695		len += nla_total_size(ovs_key_attr_size());
696
697	/* OVS_FLOW_ATTR_ACTIONS */
698	if (should_fill_actions(ufid_flags))
699		len += nla_total_size(acts->actions_len);
700
701	return len
702		+ nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
703		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
704		+ nla_total_size(8); /* OVS_FLOW_ATTR_USED */
705}
706
707/* Called with ovs_mutex or RCU read lock. */
708static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
709				   struct sk_buff *skb)
710{
711	struct ovs_flow_stats stats;
712	__be16 tcp_flags;
713	unsigned long used;
714
715	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
716
717	if (used &&
718	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
719		return -EMSGSIZE;
720
721	if (stats.n_packets &&
722	    nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
723		return -EMSGSIZE;
724
725	if ((u8)ntohs(tcp_flags) &&
726	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
727		return -EMSGSIZE;
728
729	return 0;
730}
731
732/* Called with ovs_mutex or RCU read lock. */
733static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
734				     struct sk_buff *skb, int skb_orig_len)
735{
736	struct nlattr *start;
737	int err;
738
739	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
740	 * this is the first flow to be dumped into 'skb'.  This is unusual for
741	 * Netlink but individual action lists can be longer than
742	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
743	 * The userspace caller can always fetch the actions separately if it
744	 * really wants them.  (Most userspace callers in fact don't care.)
745	 *
746	 * This can only fail for dump operations because the skb is always
747	 * properly sized for single flows.
748	 */
749	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
750	if (start) {
751		const struct sw_flow_actions *sf_acts;
752
753		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
754		err = ovs_nla_put_actions(sf_acts->actions,
755					  sf_acts->actions_len, skb);
756
757		if (!err)
758			nla_nest_end(skb, start);
759		else {
760			if (skb_orig_len)
761				return err;
762
763			nla_nest_cancel(skb, start);
764		}
765	} else if (skb_orig_len) {
766		return -EMSGSIZE;
767	}
768
769	return 0;
770}
771
772/* Called with ovs_mutex or RCU read lock. */
773static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
774				  struct sk_buff *skb, u32 portid,
775				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
776{
777	const int skb_orig_len = skb->len;
778	struct ovs_header *ovs_header;
779	int err;
780
781	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
782				 flags, cmd);
783	if (!ovs_header)
784		return -EMSGSIZE;
785
786	ovs_header->dp_ifindex = dp_ifindex;
787
788	err = ovs_nla_put_identifier(flow, skb);
789	if (err)
790		goto error;
791
792	if (should_fill_key(&flow->id, ufid_flags)) {
793		err = ovs_nla_put_masked_key(flow, skb);
794		if (err)
795			goto error;
796	}
797
798	if (should_fill_mask(ufid_flags)) {
799		err = ovs_nla_put_mask(flow, skb);
800		if (err)
801			goto error;
802	}
803
804	err = ovs_flow_cmd_fill_stats(flow, skb);
805	if (err)
806		goto error;
807
808	if (should_fill_actions(ufid_flags)) {
809		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
810		if (err)
811			goto error;
812	}
813
814	genlmsg_end(skb, ovs_header);
815	return 0;
816
817error:
818	genlmsg_cancel(skb, ovs_header);
819	return err;
820}
821
822/* May not be called with RCU read lock. */
823static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
824					       const struct sw_flow_id *sfid,
825					       struct genl_info *info,
826					       bool always,
827					       uint32_t ufid_flags)
828{
829	struct sk_buff *skb;
830	size_t len;
831
832	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
833		return NULL;
834
835	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
836	skb = genlmsg_new_unicast(len, info, GFP_KERNEL);
837	if (!skb)
838		return ERR_PTR(-ENOMEM);
839
840	return skb;
841}
842
843/* Called with ovs_mutex. */
844static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
845					       int dp_ifindex,
846					       struct genl_info *info, u8 cmd,
847					       bool always, u32 ufid_flags)
848{
849	struct sk_buff *skb;
850	int retval;
851
852	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
853				      &flow->id, info, always, ufid_flags);
854	if (IS_ERR_OR_NULL(skb))
855		return skb;
856
857	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
858					info->snd_portid, info->snd_seq, 0,
859					cmd, ufid_flags);
860	BUG_ON(retval < 0);
861	return skb;
862}
863
864static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
865{
866	struct nlattr **a = info->attrs;
867	struct ovs_header *ovs_header = info->userhdr;
868	struct sw_flow *flow = NULL, *new_flow;
869	struct sw_flow_mask mask;
870	struct sk_buff *reply;
871	struct datapath *dp;
872	struct sw_flow_key key;
873	struct sw_flow_actions *acts;
874	struct sw_flow_match match;
875	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
876	int error;
877	bool log = !a[OVS_FLOW_ATTR_PROBE];
878
879	/* Must have key and actions. */
880	error = -EINVAL;
881	if (!a[OVS_FLOW_ATTR_KEY]) {
882		OVS_NLERR(log, "Flow key attr not present in new flow.");
883		goto error;
884	}
885	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
886		OVS_NLERR(log, "Flow actions attr not present in new flow.");
887		goto error;
888	}
889
890	/* Most of the time we need to allocate a new flow, do it before
891	 * locking.
892	 */
893	new_flow = ovs_flow_alloc();
894	if (IS_ERR(new_flow)) {
895		error = PTR_ERR(new_flow);
896		goto error;
897	}
898
899	/* Extract key. */
900	ovs_match_init(&match, &key, &mask);
901	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
902				  a[OVS_FLOW_ATTR_MASK], log);
903	if (error)
904		goto err_kfree_flow;
905
906	ovs_flow_mask_key(&new_flow->key, &key, true, &mask);
907
908	/* Extract flow identifier. */
909	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
910				       &key, log);
911	if (error)
912		goto err_kfree_flow;
913
914	/* Validate actions. */
915	error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS], &new_flow->key,
916				     &acts, log);
917	if (error) {
918		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
919		goto err_kfree_flow;
920	}
921
922	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
923					ufid_flags);
924	if (IS_ERR(reply)) {
925		error = PTR_ERR(reply);
926		goto err_kfree_acts;
927	}
928
929	ovs_lock();
930	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
931	if (unlikely(!dp)) {
932		error = -ENODEV;
933		goto err_unlock_ovs;
934	}
935
936	/* Check if this is a duplicate flow */
937	if (ovs_identifier_is_ufid(&new_flow->id))
938		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
939	if (!flow)
940		flow = ovs_flow_tbl_lookup(&dp->table, &key);
941	if (likely(!flow)) {
942		rcu_assign_pointer(new_flow->sf_acts, acts);
943
944		/* Put flow in bucket. */
945		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
946		if (unlikely(error)) {
947			acts = NULL;
948			goto err_unlock_ovs;
949		}
950
951		if (unlikely(reply)) {
952			error = ovs_flow_cmd_fill_info(new_flow,
953						       ovs_header->dp_ifindex,
954						       reply, info->snd_portid,
955						       info->snd_seq, 0,
956						       OVS_FLOW_CMD_NEW,
957						       ufid_flags);
958			BUG_ON(error < 0);
959		}
960		ovs_unlock();
961	} else {
962		struct sw_flow_actions *old_acts;
963
964		/* Bail out if we're not allowed to modify an existing flow.
965		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
966		 * because Generic Netlink treats the latter as a dump
967		 * request.  We also accept NLM_F_EXCL in case that bug ever
968		 * gets fixed.
969		 */
970		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
971							 | NLM_F_EXCL))) {
972			error = -EEXIST;
973			goto err_unlock_ovs;
974		}
975		/* The flow identifier has to be the same for flow updates.
976		 * Look for any overlapping flow.
977		 */
978		if (unlikely(!ovs_flow_cmp(flow, &match))) {
979			if (ovs_identifier_is_key(&flow->id))
980				flow = ovs_flow_tbl_lookup_exact(&dp->table,
981								 &match);
982			else /* UFID matches but key is different */
983				flow = NULL;
984			if (!flow) {
985				error = -ENOENT;
986				goto err_unlock_ovs;
987			}
988		}
989		/* Update actions. */
990		old_acts = ovsl_dereference(flow->sf_acts);
991		rcu_assign_pointer(flow->sf_acts, acts);
992
993		if (unlikely(reply)) {
994			error = ovs_flow_cmd_fill_info(flow,
995						       ovs_header->dp_ifindex,
996						       reply, info->snd_portid,
997						       info->snd_seq, 0,
998						       OVS_FLOW_CMD_NEW,
999						       ufid_flags);
1000			BUG_ON(error < 0);
1001		}
1002		ovs_unlock();
1003
1004		ovs_nla_free_flow_actions(old_acts);
1005		ovs_flow_free(new_flow, false);
1006	}
1007
1008	if (reply)
1009		ovs_notify(&dp_flow_genl_family, reply, info);
1010	return 0;
1011
1012err_unlock_ovs:
1013	ovs_unlock();
1014	kfree_skb(reply);
1015err_kfree_acts:
1016	kfree(acts);
1017err_kfree_flow:
1018	ovs_flow_free(new_flow, false);
1019error:
1020	return error;
1021}
1022
1023/* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1024static struct sw_flow_actions *get_flow_actions(const struct nlattr *a,
1025						const struct sw_flow_key *key,
1026						const struct sw_flow_mask *mask,
1027						bool log)
1028{
1029	struct sw_flow_actions *acts;
1030	struct sw_flow_key masked_key;
1031	int error;
1032
1033	ovs_flow_mask_key(&masked_key, key, true, mask);
1034	error = ovs_nla_copy_actions(a, &masked_key, &acts, log);
1035	if (error) {
1036		OVS_NLERR(log,
1037			  "Actions may not be safe on all matching packets");
1038		return ERR_PTR(error);
1039	}
1040
1041	return acts;
1042}
1043
1044static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1045{
1046	struct nlattr **a = info->attrs;
1047	struct ovs_header *ovs_header = info->userhdr;
1048	struct sw_flow_key key;
1049	struct sw_flow *flow;
1050	struct sw_flow_mask mask;
1051	struct sk_buff *reply = NULL;
1052	struct datapath *dp;
1053	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1054	struct sw_flow_match match;
1055	struct sw_flow_id sfid;
1056	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1057	int error;
1058	bool log = !a[OVS_FLOW_ATTR_PROBE];
1059	bool ufid_present;
1060
1061	/* Extract key. */
1062	error = -EINVAL;
1063	if (!a[OVS_FLOW_ATTR_KEY]) {
1064		OVS_NLERR(log, "Flow key attribute not present in set flow.");
1065		goto error;
1066	}
1067
1068	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1069	ovs_match_init(&match, &key, &mask);
1070	error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY],
1071				  a[OVS_FLOW_ATTR_MASK], log);
1072	if (error)
1073		goto error;
1074
1075	/* Validate actions. */
1076	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1077		acts = get_flow_actions(a[OVS_FLOW_ATTR_ACTIONS], &key, &mask,
1078					log);
1079		if (IS_ERR(acts)) {
1080			error = PTR_ERR(acts);
1081			goto error;
1082		}
1083
1084		/* Can allocate before locking if have acts. */
1085		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1086						ufid_flags);
1087		if (IS_ERR(reply)) {
1088			error = PTR_ERR(reply);
1089			goto err_kfree_acts;
1090		}
1091	}
1092
1093	ovs_lock();
1094	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1095	if (unlikely(!dp)) {
1096		error = -ENODEV;
1097		goto err_unlock_ovs;
1098	}
1099	/* Check that the flow exists. */
1100	if (ufid_present)
1101		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1102	else
1103		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1104	if (unlikely(!flow)) {
1105		error = -ENOENT;
1106		goto err_unlock_ovs;
1107	}
1108
1109	/* Update actions, if present. */
1110	if (likely(acts)) {
1111		old_acts = ovsl_dereference(flow->sf_acts);
1112		rcu_assign_pointer(flow->sf_acts, acts);
1113
1114		if (unlikely(reply)) {
1115			error = ovs_flow_cmd_fill_info(flow,
1116						       ovs_header->dp_ifindex,
1117						       reply, info->snd_portid,
1118						       info->snd_seq, 0,
1119						       OVS_FLOW_CMD_NEW,
1120						       ufid_flags);
1121			BUG_ON(error < 0);
1122		}
1123	} else {
1124		/* Could not alloc without acts before locking. */
1125		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1126						info, OVS_FLOW_CMD_NEW, false,
1127						ufid_flags);
1128
1129		if (unlikely(IS_ERR(reply))) {
1130			error = PTR_ERR(reply);
1131			goto err_unlock_ovs;
1132		}
1133	}
1134
1135	/* Clear stats. */
1136	if (a[OVS_FLOW_ATTR_CLEAR])
1137		ovs_flow_stats_clear(flow);
1138	ovs_unlock();
1139
1140	if (reply)
1141		ovs_notify(&dp_flow_genl_family, reply, info);
1142	if (old_acts)
1143		ovs_nla_free_flow_actions(old_acts);
1144
1145	return 0;
1146
1147err_unlock_ovs:
1148	ovs_unlock();
1149	kfree_skb(reply);
1150err_kfree_acts:
1151	kfree(acts);
1152error:
1153	return error;
1154}
1155
1156static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1157{
1158	struct nlattr **a = info->attrs;
1159	struct ovs_header *ovs_header = info->userhdr;
1160	struct sw_flow_key key;
1161	struct sk_buff *reply;
1162	struct sw_flow *flow;
1163	struct datapath *dp;
1164	struct sw_flow_match match;
1165	struct sw_flow_id ufid;
1166	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1167	int err = 0;
1168	bool log = !a[OVS_FLOW_ATTR_PROBE];
1169	bool ufid_present;
1170
1171	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1172	if (a[OVS_FLOW_ATTR_KEY]) {
1173		ovs_match_init(&match, &key, NULL);
1174		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1175					log);
1176	} else if (!ufid_present) {
1177		OVS_NLERR(log,
1178			  "Flow get message rejected, Key attribute missing.");
1179		err = -EINVAL;
1180	}
1181	if (err)
1182		return err;
1183
1184	ovs_lock();
1185	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1186	if (!dp) {
1187		err = -ENODEV;
1188		goto unlock;
1189	}
1190
1191	if (ufid_present)
1192		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1193	else
1194		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1195	if (!flow) {
1196		err = -ENOENT;
1197		goto unlock;
1198	}
1199
1200	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1201					OVS_FLOW_CMD_NEW, true, ufid_flags);
1202	if (IS_ERR(reply)) {
1203		err = PTR_ERR(reply);
1204		goto unlock;
1205	}
1206
1207	ovs_unlock();
1208	return genlmsg_reply(reply, info);
1209unlock:
1210	ovs_unlock();
1211	return err;
1212}
1213
1214static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1215{
1216	struct nlattr **a = info->attrs;
1217	struct ovs_header *ovs_header = info->userhdr;
1218	struct sw_flow_key key;
1219	struct sk_buff *reply;
1220	struct sw_flow *flow = NULL;
1221	struct datapath *dp;
1222	struct sw_flow_match match;
1223	struct sw_flow_id ufid;
1224	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1225	int err;
1226	bool log = !a[OVS_FLOW_ATTR_PROBE];
1227	bool ufid_present;
1228
1229	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1230	if (a[OVS_FLOW_ATTR_KEY]) {
1231		ovs_match_init(&match, &key, NULL);
1232		err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL,
1233					log);
1234		if (unlikely(err))
1235			return err;
1236	}
1237
1238	ovs_lock();
1239	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1240	if (unlikely(!dp)) {
1241		err = -ENODEV;
1242		goto unlock;
1243	}
1244
1245	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1246		err = ovs_flow_tbl_flush(&dp->table);
1247		goto unlock;
1248	}
1249
1250	if (ufid_present)
1251		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1252	else
1253		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1254	if (unlikely(!flow)) {
1255		err = -ENOENT;
1256		goto unlock;
1257	}
1258
1259	ovs_flow_tbl_remove(&dp->table, flow);
1260	ovs_unlock();
1261
1262	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1263					&flow->id, info, false, ufid_flags);
1264	if (likely(reply)) {
1265		if (likely(!IS_ERR(reply))) {
1266			rcu_read_lock();	/*To keep RCU checker happy. */
1267			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1268						     reply, info->snd_portid,
1269						     info->snd_seq, 0,
1270						     OVS_FLOW_CMD_DEL,
1271						     ufid_flags);
1272			rcu_read_unlock();
1273			BUG_ON(err < 0);
1274
1275			ovs_notify(&dp_flow_genl_family, reply, info);
1276		} else {
1277			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0, PTR_ERR(reply));
1278		}
1279	}
1280
1281	ovs_flow_free(flow, true);
1282	return 0;
1283unlock:
1284	ovs_unlock();
1285	return err;
1286}
1287
1288static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1289{
1290	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1291	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1292	struct table_instance *ti;
1293	struct datapath *dp;
1294	u32 ufid_flags;
1295	int err;
1296
1297	err = genlmsg_parse(cb->nlh, &dp_flow_genl_family, a,
1298			    OVS_FLOW_ATTR_MAX, flow_policy);
1299	if (err)
1300		return err;
1301	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1302
1303	rcu_read_lock();
1304	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1305	if (!dp) {
1306		rcu_read_unlock();
1307		return -ENODEV;
1308	}
1309
1310	ti = rcu_dereference(dp->table.ti);
1311	for (;;) {
1312		struct sw_flow *flow;
1313		u32 bucket, obj;
1314
1315		bucket = cb->args[0];
1316		obj = cb->args[1];
1317		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1318		if (!flow)
1319			break;
1320
1321		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1322					   NETLINK_CB(cb->skb).portid,
1323					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1324					   OVS_FLOW_CMD_NEW, ufid_flags) < 0)
1325			break;
1326
1327		cb->args[0] = bucket;
1328		cb->args[1] = obj;
1329	}
1330	rcu_read_unlock();
1331	return skb->len;
1332}
1333
1334static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1335	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1336	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1337	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1338	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1339	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1340	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1341	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1342};
1343
1344static const struct genl_ops dp_flow_genl_ops[] = {
1345	{ .cmd = OVS_FLOW_CMD_NEW,
1346	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1347	  .policy = flow_policy,
1348	  .doit = ovs_flow_cmd_new
1349	},
1350	{ .cmd = OVS_FLOW_CMD_DEL,
1351	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1352	  .policy = flow_policy,
1353	  .doit = ovs_flow_cmd_del
1354	},
1355	{ .cmd = OVS_FLOW_CMD_GET,
1356	  .flags = 0,		    /* OK for unprivileged users. */
1357	  .policy = flow_policy,
1358	  .doit = ovs_flow_cmd_get,
1359	  .dumpit = ovs_flow_cmd_dump
1360	},
1361	{ .cmd = OVS_FLOW_CMD_SET,
1362	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1363	  .policy = flow_policy,
1364	  .doit = ovs_flow_cmd_set,
1365	},
1366};
1367
1368static struct genl_family dp_flow_genl_family = {
1369	.id = GENL_ID_GENERATE,
1370	.hdrsize = sizeof(struct ovs_header),
1371	.name = OVS_FLOW_FAMILY,
1372	.version = OVS_FLOW_VERSION,
1373	.maxattr = OVS_FLOW_ATTR_MAX,
1374	.netnsok = true,
1375	.parallel_ops = true,
1376	.ops = dp_flow_genl_ops,
1377	.n_ops = ARRAY_SIZE(dp_flow_genl_ops),
1378	.mcgrps = &ovs_dp_flow_multicast_group,
1379	.n_mcgrps = 1,
1380};
1381
1382static size_t ovs_dp_cmd_msg_size(void)
1383{
1384	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1385
1386	msgsize += nla_total_size(IFNAMSIZ);
1387	msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1388	msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1389	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1390
1391	return msgsize;
1392}
1393
1394/* Called with ovs_mutex. */
1395static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1396				u32 portid, u32 seq, u32 flags, u8 cmd)
1397{
1398	struct ovs_header *ovs_header;
1399	struct ovs_dp_stats dp_stats;
1400	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1401	int err;
1402
1403	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1404				   flags, cmd);
1405	if (!ovs_header)
1406		goto error;
1407
1408	ovs_header->dp_ifindex = get_dpifindex(dp);
1409
1410	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1411	if (err)
1412		goto nla_put_failure;
1413
1414	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1415	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1416			&dp_stats))
1417		goto nla_put_failure;
1418
1419	if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1420			sizeof(struct ovs_dp_megaflow_stats),
1421			&dp_megaflow_stats))
1422		goto nla_put_failure;
1423
1424	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1425		goto nla_put_failure;
1426
1427	genlmsg_end(skb, ovs_header);
1428	return 0;
1429
1430nla_put_failure:
1431	genlmsg_cancel(skb, ovs_header);
1432error:
1433	return -EMSGSIZE;
1434}
1435
1436static struct sk_buff *ovs_dp_cmd_alloc_info(struct genl_info *info)
1437{
1438	return genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1439}
1440
1441/* Called with rcu_read_lock or ovs_mutex. */
1442static struct datapath *lookup_datapath(struct net *net,
1443					const struct ovs_header *ovs_header,
1444					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1445{
1446	struct datapath *dp;
1447
1448	if (!a[OVS_DP_ATTR_NAME])
1449		dp = get_dp(net, ovs_header->dp_ifindex);
1450	else {
1451		struct vport *vport;
1452
1453		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1454		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1455	}
1456	return dp ? dp : ERR_PTR(-ENODEV);
1457}
1458
1459static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1460{
1461	struct datapath *dp;
1462
1463	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1464	if (IS_ERR(dp))
1465		return;
1466
1467	WARN(dp->user_features, "Dropping previously announced user features\n");
1468	dp->user_features = 0;
1469}
1470
1471static void ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1472{
1473	if (a[OVS_DP_ATTR_USER_FEATURES])
1474		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1475}
1476
1477static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1478{
1479	struct nlattr **a = info->attrs;
1480	struct vport_parms parms;
1481	struct sk_buff *reply;
1482	struct datapath *dp;
1483	struct vport *vport;
1484	struct ovs_net *ovs_net;
1485	int err, i;
1486
1487	err = -EINVAL;
1488	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1489		goto err;
1490
1491	reply = ovs_dp_cmd_alloc_info(info);
1492	if (!reply)
1493		return -ENOMEM;
1494
1495	err = -ENOMEM;
1496	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1497	if (dp == NULL)
1498		goto err_free_reply;
1499
1500	ovs_dp_set_net(dp, sock_net(skb->sk));
1501
1502	/* Allocate table. */
1503	err = ovs_flow_tbl_init(&dp->table);
1504	if (err)
1505		goto err_free_dp;
1506
1507	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1508	if (!dp->stats_percpu) {
1509		err = -ENOMEM;
1510		goto err_destroy_table;
1511	}
1512
1513	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1514			    GFP_KERNEL);
1515	if (!dp->ports) {
1516		err = -ENOMEM;
1517		goto err_destroy_percpu;
1518	}
1519
1520	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1521		INIT_HLIST_HEAD(&dp->ports[i]);
1522
1523	/* Set up our datapath device. */
1524	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1525	parms.type = OVS_VPORT_TYPE_INTERNAL;
1526	parms.options = NULL;
1527	parms.dp = dp;
1528	parms.port_no = OVSP_LOCAL;
1529	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1530
1531	ovs_dp_change(dp, a);
1532
1533	/* So far only local changes have been made, now need the lock. */
1534	ovs_lock();
1535
1536	vport = new_vport(&parms);
1537	if (IS_ERR(vport)) {
1538		err = PTR_ERR(vport);
1539		if (err == -EBUSY)
1540			err = -EEXIST;
1541
1542		if (err == -EEXIST) {
1543			/* An outdated user space instance that does not understand
1544			 * the concept of user_features has attempted to create a new
1545			 * datapath and is likely to reuse it. Drop all user features.
1546			 */
1547			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1548				ovs_dp_reset_user_features(skb, info);
1549		}
1550
1551		goto err_destroy_ports_array;
1552	}
1553
1554	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1555				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1556	BUG_ON(err < 0);
1557
1558	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1559	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1560
1561	ovs_unlock();
1562
1563	ovs_notify(&dp_datapath_genl_family, reply, info);
1564	return 0;
1565
1566err_destroy_ports_array:
1567	ovs_unlock();
1568	kfree(dp->ports);
1569err_destroy_percpu:
1570	free_percpu(dp->stats_percpu);
1571err_destroy_table:
1572	ovs_flow_tbl_destroy(&dp->table);
1573err_free_dp:
1574	kfree(dp);
1575err_free_reply:
1576	kfree_skb(reply);
1577err:
1578	return err;
1579}
1580
1581/* Called with ovs_mutex. */
1582static void __dp_destroy(struct datapath *dp)
1583{
1584	int i;
1585
1586	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1587		struct vport *vport;
1588		struct hlist_node *n;
1589
1590		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1591			if (vport->port_no != OVSP_LOCAL)
1592				ovs_dp_detach_port(vport);
1593	}
1594
1595	list_del_rcu(&dp->list_node);
1596
1597	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1598	 * all ports in datapath are destroyed first before freeing datapath.
1599	 */
1600	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1601
1602	/* RCU destroy the flow table */
1603	call_rcu(&dp->rcu, destroy_dp_rcu);
1604}
1605
1606static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1607{
1608	struct sk_buff *reply;
1609	struct datapath *dp;
1610	int err;
1611
1612	reply = ovs_dp_cmd_alloc_info(info);
1613	if (!reply)
1614		return -ENOMEM;
1615
1616	ovs_lock();
1617	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1618	err = PTR_ERR(dp);
1619	if (IS_ERR(dp))
1620		goto err_unlock_free;
1621
1622	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1623				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1624	BUG_ON(err < 0);
1625
1626	__dp_destroy(dp);
1627	ovs_unlock();
1628
1629	ovs_notify(&dp_datapath_genl_family, reply, info);
1630
1631	return 0;
1632
1633err_unlock_free:
1634	ovs_unlock();
1635	kfree_skb(reply);
1636	return err;
1637}
1638
1639static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1640{
1641	struct sk_buff *reply;
1642	struct datapath *dp;
1643	int err;
1644
1645	reply = ovs_dp_cmd_alloc_info(info);
1646	if (!reply)
1647		return -ENOMEM;
1648
1649	ovs_lock();
1650	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1651	err = PTR_ERR(dp);
1652	if (IS_ERR(dp))
1653		goto err_unlock_free;
1654
1655	ovs_dp_change(dp, info->attrs);
1656
1657	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1658				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1659	BUG_ON(err < 0);
1660
1661	ovs_unlock();
1662	ovs_notify(&dp_datapath_genl_family, reply, info);
1663
1664	return 0;
1665
1666err_unlock_free:
1667	ovs_unlock();
1668	kfree_skb(reply);
1669	return err;
1670}
1671
1672static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1673{
1674	struct sk_buff *reply;
1675	struct datapath *dp;
1676	int err;
1677
1678	reply = ovs_dp_cmd_alloc_info(info);
1679	if (!reply)
1680		return -ENOMEM;
1681
1682	ovs_lock();
1683	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1684	if (IS_ERR(dp)) {
1685		err = PTR_ERR(dp);
1686		goto err_unlock_free;
1687	}
1688	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1689				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1690	BUG_ON(err < 0);
1691	ovs_unlock();
1692
1693	return genlmsg_reply(reply, info);
1694
1695err_unlock_free:
1696	ovs_unlock();
1697	kfree_skb(reply);
1698	return err;
1699}
1700
1701static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1702{
1703	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1704	struct datapath *dp;
1705	int skip = cb->args[0];
1706	int i = 0;
1707
1708	ovs_lock();
1709	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1710		if (i >= skip &&
1711		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1712					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1713					 OVS_DP_CMD_NEW) < 0)
1714			break;
1715		i++;
1716	}
1717	ovs_unlock();
1718
1719	cb->args[0] = i;
1720
1721	return skb->len;
1722}
1723
1724static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1725	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1726	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1727	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1728};
1729
1730static const struct genl_ops dp_datapath_genl_ops[] = {
1731	{ .cmd = OVS_DP_CMD_NEW,
1732	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1733	  .policy = datapath_policy,
1734	  .doit = ovs_dp_cmd_new
1735	},
1736	{ .cmd = OVS_DP_CMD_DEL,
1737	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1738	  .policy = datapath_policy,
1739	  .doit = ovs_dp_cmd_del
1740	},
1741	{ .cmd = OVS_DP_CMD_GET,
1742	  .flags = 0,		    /* OK for unprivileged users. */
1743	  .policy = datapath_policy,
1744	  .doit = ovs_dp_cmd_get,
1745	  .dumpit = ovs_dp_cmd_dump
1746	},
1747	{ .cmd = OVS_DP_CMD_SET,
1748	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1749	  .policy = datapath_policy,
1750	  .doit = ovs_dp_cmd_set,
1751	},
1752};
1753
1754static struct genl_family dp_datapath_genl_family = {
1755	.id = GENL_ID_GENERATE,
1756	.hdrsize = sizeof(struct ovs_header),
1757	.name = OVS_DATAPATH_FAMILY,
1758	.version = OVS_DATAPATH_VERSION,
1759	.maxattr = OVS_DP_ATTR_MAX,
1760	.netnsok = true,
1761	.parallel_ops = true,
1762	.ops = dp_datapath_genl_ops,
1763	.n_ops = ARRAY_SIZE(dp_datapath_genl_ops),
1764	.mcgrps = &ovs_dp_datapath_multicast_group,
1765	.n_mcgrps = 1,
1766};
1767
1768/* Called with ovs_mutex or RCU read lock. */
1769static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1770				   u32 portid, u32 seq, u32 flags, u8 cmd)
1771{
1772	struct ovs_header *ovs_header;
1773	struct ovs_vport_stats vport_stats;
1774	int err;
1775
1776	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1777				 flags, cmd);
1778	if (!ovs_header)
1779		return -EMSGSIZE;
1780
1781	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1782
1783	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1784	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1785	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
1786			   vport->ops->get_name(vport)))
1787		goto nla_put_failure;
1788
1789	ovs_vport_get_stats(vport, &vport_stats);
1790	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1791		    &vport_stats))
1792		goto nla_put_failure;
1793
1794	if (ovs_vport_get_upcall_portids(vport, skb))
1795		goto nla_put_failure;
1796
1797	err = ovs_vport_get_options(vport, skb);
1798	if (err == -EMSGSIZE)
1799		goto error;
1800
1801	genlmsg_end(skb, ovs_header);
1802	return 0;
1803
1804nla_put_failure:
1805	err = -EMSGSIZE;
1806error:
1807	genlmsg_cancel(skb, ovs_header);
1808	return err;
1809}
1810
1811static struct sk_buff *ovs_vport_cmd_alloc_info(void)
1812{
1813	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1814}
1815
1816/* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
1817struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1818					 u32 seq, u8 cmd)
1819{
1820	struct sk_buff *skb;
1821	int retval;
1822
1823	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1824	if (!skb)
1825		return ERR_PTR(-ENOMEM);
1826
1827	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1828	BUG_ON(retval < 0);
1829
1830	return skb;
1831}
1832
1833/* Called with ovs_mutex or RCU read lock. */
1834static struct vport *lookup_vport(struct net *net,
1835				  const struct ovs_header *ovs_header,
1836				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1837{
1838	struct datapath *dp;
1839	struct vport *vport;
1840
1841	if (a[OVS_VPORT_ATTR_NAME]) {
1842		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1843		if (!vport)
1844			return ERR_PTR(-ENODEV);
1845		if (ovs_header->dp_ifindex &&
1846		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1847			return ERR_PTR(-ENODEV);
1848		return vport;
1849	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1850		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1851
1852		if (port_no >= DP_MAX_PORTS)
1853			return ERR_PTR(-EFBIG);
1854
1855		dp = get_dp(net, ovs_header->dp_ifindex);
1856		if (!dp)
1857			return ERR_PTR(-ENODEV);
1858
1859		vport = ovs_vport_ovsl_rcu(dp, port_no);
1860		if (!vport)
1861			return ERR_PTR(-ENODEV);
1862		return vport;
1863	} else
1864		return ERR_PTR(-EINVAL);
1865}
1866
1867static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1868{
1869	struct nlattr **a = info->attrs;
1870	struct ovs_header *ovs_header = info->userhdr;
1871	struct vport_parms parms;
1872	struct sk_buff *reply;
1873	struct vport *vport;
1874	struct datapath *dp;
1875	u32 port_no;
1876	int err;
1877
1878	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1879	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1880		return -EINVAL;
1881
1882	port_no = a[OVS_VPORT_ATTR_PORT_NO]
1883		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
1884	if (port_no >= DP_MAX_PORTS)
1885		return -EFBIG;
1886
1887	reply = ovs_vport_cmd_alloc_info();
1888	if (!reply)
1889		return -ENOMEM;
1890
1891	ovs_lock();
1892restart:
1893	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1894	err = -ENODEV;
1895	if (!dp)
1896		goto exit_unlock_free;
1897
1898	if (port_no) {
1899		vport = ovs_vport_ovsl(dp, port_no);
1900		err = -EBUSY;
1901		if (vport)
1902			goto exit_unlock_free;
1903	} else {
1904		for (port_no = 1; ; port_no++) {
1905			if (port_no >= DP_MAX_PORTS) {
1906				err = -EFBIG;
1907				goto exit_unlock_free;
1908			}
1909			vport = ovs_vport_ovsl(dp, port_no);
1910			if (!vport)
1911				break;
1912		}
1913	}
1914
1915	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1916	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1917	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1918	parms.dp = dp;
1919	parms.port_no = port_no;
1920	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
1921
1922	vport = new_vport(&parms);
1923	err = PTR_ERR(vport);
1924	if (IS_ERR(vport)) {
1925		if (err == -EAGAIN)
1926			goto restart;
1927		goto exit_unlock_free;
1928	}
1929
1930	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1931				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1932	BUG_ON(err < 0);
1933	ovs_unlock();
1934
1935	ovs_notify(&dp_vport_genl_family, reply, info);
1936	return 0;
1937
1938exit_unlock_free:
1939	ovs_unlock();
1940	kfree_skb(reply);
1941	return err;
1942}
1943
1944static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1945{
1946	struct nlattr **a = info->attrs;
1947	struct sk_buff *reply;
1948	struct vport *vport;
1949	int err;
1950
1951	reply = ovs_vport_cmd_alloc_info();
1952	if (!reply)
1953		return -ENOMEM;
1954
1955	ovs_lock();
1956	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1957	err = PTR_ERR(vport);
1958	if (IS_ERR(vport))
1959		goto exit_unlock_free;
1960
1961	if (a[OVS_VPORT_ATTR_TYPE] &&
1962	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1963		err = -EINVAL;
1964		goto exit_unlock_free;
1965	}
1966
1967	if (a[OVS_VPORT_ATTR_OPTIONS]) {
1968		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1969		if (err)
1970			goto exit_unlock_free;
1971	}
1972
1973
1974	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
1975		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
1976
1977		err = ovs_vport_set_upcall_portids(vport, ids);
1978		if (err)
1979			goto exit_unlock_free;
1980	}
1981
1982	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1983				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1984	BUG_ON(err < 0);
1985
1986	ovs_unlock();
1987	ovs_notify(&dp_vport_genl_family, reply, info);
1988	return 0;
1989
1990exit_unlock_free:
1991	ovs_unlock();
1992	kfree_skb(reply);
1993	return err;
1994}
1995
1996static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1997{
1998	struct nlattr **a = info->attrs;
1999	struct sk_buff *reply;
2000	struct vport *vport;
2001	int err;
2002
2003	reply = ovs_vport_cmd_alloc_info();
2004	if (!reply)
2005		return -ENOMEM;
2006
2007	ovs_lock();
2008	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2009	err = PTR_ERR(vport);
2010	if (IS_ERR(vport))
2011		goto exit_unlock_free;
2012
2013	if (vport->port_no == OVSP_LOCAL) {
2014		err = -EINVAL;
2015		goto exit_unlock_free;
2016	}
2017
2018	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2019				      info->snd_seq, 0, OVS_VPORT_CMD_DEL);
2020	BUG_ON(err < 0);
2021	ovs_dp_detach_port(vport);
2022	ovs_unlock();
2023
2024	ovs_notify(&dp_vport_genl_family, reply, info);
2025	return 0;
2026
2027exit_unlock_free:
2028	ovs_unlock();
2029	kfree_skb(reply);
2030	return err;
2031}
2032
2033static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2034{
2035	struct nlattr **a = info->attrs;
2036	struct ovs_header *ovs_header = info->userhdr;
2037	struct sk_buff *reply;
2038	struct vport *vport;
2039	int err;
2040
2041	reply = ovs_vport_cmd_alloc_info();
2042	if (!reply)
2043		return -ENOMEM;
2044
2045	rcu_read_lock();
2046	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2047	err = PTR_ERR(vport);
2048	if (IS_ERR(vport))
2049		goto exit_unlock_free;
2050	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
2051				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
2052	BUG_ON(err < 0);
2053	rcu_read_unlock();
2054
2055	return genlmsg_reply(reply, info);
2056
2057exit_unlock_free:
2058	rcu_read_unlock();
2059	kfree_skb(reply);
2060	return err;
2061}
2062
2063static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2064{
2065	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2066	struct datapath *dp;
2067	int bucket = cb->args[0], skip = cb->args[1];
2068	int i, j = 0;
2069
2070	rcu_read_lock();
2071	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2072	if (!dp) {
2073		rcu_read_unlock();
2074		return -ENODEV;
2075	}
2076	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2077		struct vport *vport;
2078
2079		j = 0;
2080		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2081			if (j >= skip &&
2082			    ovs_vport_cmd_fill_info(vport, skb,
2083						    NETLINK_CB(cb->skb).portid,
2084						    cb->nlh->nlmsg_seq,
2085						    NLM_F_MULTI,
2086						    OVS_VPORT_CMD_NEW) < 0)
2087				goto out;
2088
2089			j++;
2090		}
2091		skip = 0;
2092	}
2093out:
2094	rcu_read_unlock();
2095
2096	cb->args[0] = i;
2097	cb->args[1] = j;
2098
2099	return skb->len;
2100}
2101
2102static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2103	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2104	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2105	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2106	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2107	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2108	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2109};
2110
2111static const struct genl_ops dp_vport_genl_ops[] = {
2112	{ .cmd = OVS_VPORT_CMD_NEW,
2113	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2114	  .policy = vport_policy,
2115	  .doit = ovs_vport_cmd_new
2116	},
2117	{ .cmd = OVS_VPORT_CMD_DEL,
2118	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2119	  .policy = vport_policy,
2120	  .doit = ovs_vport_cmd_del
2121	},
2122	{ .cmd = OVS_VPORT_CMD_GET,
2123	  .flags = 0,		    /* OK for unprivileged users. */
2124	  .policy = vport_policy,
2125	  .doit = ovs_vport_cmd_get,
2126	  .dumpit = ovs_vport_cmd_dump
2127	},
2128	{ .cmd = OVS_VPORT_CMD_SET,
2129	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2130	  .policy = vport_policy,
2131	  .doit = ovs_vport_cmd_set,
2132	},
2133};
2134
2135struct genl_family dp_vport_genl_family = {
2136	.id = GENL_ID_GENERATE,
2137	.hdrsize = sizeof(struct ovs_header),
2138	.name = OVS_VPORT_FAMILY,
2139	.version = OVS_VPORT_VERSION,
2140	.maxattr = OVS_VPORT_ATTR_MAX,
2141	.netnsok = true,
2142	.parallel_ops = true,
2143	.ops = dp_vport_genl_ops,
2144	.n_ops = ARRAY_SIZE(dp_vport_genl_ops),
2145	.mcgrps = &ovs_dp_vport_multicast_group,
2146	.n_mcgrps = 1,
2147};
2148
2149static struct genl_family * const dp_genl_families[] = {
2150	&dp_datapath_genl_family,
2151	&dp_vport_genl_family,
2152	&dp_flow_genl_family,
2153	&dp_packet_genl_family,
2154};
2155
2156static void dp_unregister_genl(int n_families)
2157{
2158	int i;
2159
2160	for (i = 0; i < n_families; i++)
2161		genl_unregister_family(dp_genl_families[i]);
2162}
2163
2164static int dp_register_genl(void)
2165{
2166	int err;
2167	int i;
2168
2169	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2170
2171		err = genl_register_family(dp_genl_families[i]);
2172		if (err)
2173			goto error;
2174	}
2175
2176	return 0;
2177
2178error:
2179	dp_unregister_genl(i);
2180	return err;
2181}
2182
2183static int __net_init ovs_init_net(struct net *net)
2184{
2185	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2186
2187	INIT_LIST_HEAD(&ovs_net->dps);
2188	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2189	return 0;
2190}
2191
2192static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2193					    struct list_head *head)
2194{
2195	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2196	struct datapath *dp;
2197
2198	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2199		int i;
2200
2201		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2202			struct vport *vport;
2203
2204			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2205				struct netdev_vport *netdev_vport;
2206
2207				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2208					continue;
2209
2210				netdev_vport = netdev_vport_priv(vport);
2211				if (dev_net(netdev_vport->dev) == dnet)
2212					list_add(&vport->detach_list, head);
2213			}
2214		}
2215	}
2216}
2217
2218static void __net_exit ovs_exit_net(struct net *dnet)
2219{
2220	struct datapath *dp, *dp_next;
2221	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2222	struct vport *vport, *vport_next;
2223	struct net *net;
2224	LIST_HEAD(head);
2225
2226	ovs_lock();
2227	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2228		__dp_destroy(dp);
2229
2230	rtnl_lock();
2231	for_each_net(net)
2232		list_vports_from_net(net, dnet, &head);
2233	rtnl_unlock();
2234
2235	/* Detach all vports from given namespace. */
2236	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2237		list_del(&vport->detach_list);
2238		ovs_dp_detach_port(vport);
2239	}
2240
2241	ovs_unlock();
2242
2243	cancel_work_sync(&ovs_net->dp_notify_work);
2244}
2245
2246static struct pernet_operations ovs_net_ops = {
2247	.init = ovs_init_net,
2248	.exit = ovs_exit_net,
2249	.id   = &ovs_net_id,
2250	.size = sizeof(struct ovs_net),
2251};
2252
2253static int __init dp_init(void)
2254{
2255	int err;
2256
2257	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
2258
2259	pr_info("Open vSwitch switching datapath\n");
2260
2261	err = action_fifos_init();
2262	if (err)
2263		goto error;
2264
2265	err = ovs_internal_dev_rtnl_link_register();
2266	if (err)
2267		goto error_action_fifos_exit;
2268
2269	err = ovs_flow_init();
2270	if (err)
2271		goto error_unreg_rtnl_link;
2272
2273	err = ovs_vport_init();
2274	if (err)
2275		goto error_flow_exit;
2276
2277	err = register_pernet_device(&ovs_net_ops);
2278	if (err)
2279		goto error_vport_exit;
2280
2281	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2282	if (err)
2283		goto error_netns_exit;
2284
2285	err = ovs_netdev_init();
2286	if (err)
2287		goto error_unreg_notifier;
2288
2289	err = dp_register_genl();
2290	if (err < 0)
2291		goto error_unreg_netdev;
2292
2293	return 0;
2294
2295error_unreg_netdev:
2296	ovs_netdev_exit();
2297error_unreg_notifier:
2298	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2299error_netns_exit:
2300	unregister_pernet_device(&ovs_net_ops);
2301error_vport_exit:
2302	ovs_vport_exit();
2303error_flow_exit:
2304	ovs_flow_exit();
2305error_unreg_rtnl_link:
2306	ovs_internal_dev_rtnl_link_unregister();
2307error_action_fifos_exit:
2308	action_fifos_exit();
2309error:
2310	return err;
2311}
2312
2313static void dp_cleanup(void)
2314{
2315	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2316	ovs_netdev_exit();
2317	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2318	unregister_pernet_device(&ovs_net_ops);
2319	rcu_barrier();
2320	ovs_vport_exit();
2321	ovs_flow_exit();
2322	ovs_internal_dev_rtnl_link_unregister();
2323	action_fifos_exit();
2324}
2325
2326module_init(dp_init);
2327module_exit(dp_cleanup);
2328
2329MODULE_DESCRIPTION("Open vSwitch switching datapath");
2330MODULE_LICENSE("GPL");
2331