datapath.c revision bb6f9a708d4067713afae2e9eb2637f6b4c01ecb
1/*
2 * Copyright (c) 2007-2013 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/lockdep.h>
48#include <linux/openvswitch.h>
49#include <linux/rculist.h>
50#include <linux/dmi.h>
51#include <linux/workqueue.h>
52#include <net/genetlink.h>
53#include <net/net_namespace.h>
54#include <net/netns/generic.h>
55
56#include "datapath.h"
57#include "flow.h"
58#include "flow_table.h"
59#include "flow_netlink.h"
60#include "vport-internal_dev.h"
61#include "vport-netdev.h"
62
63int ovs_net_id __read_mostly;
64
65static void ovs_notify(struct genl_family *family,
66		       struct sk_buff *skb, struct genl_info *info)
67{
68	genl_notify(family, skb, genl_info_net(info), info->snd_portid,
69		    0, info->nlhdr, GFP_KERNEL);
70}
71
72/**
73 * DOC: Locking:
74 *
75 * All writes e.g. Writes to device state (add/remove datapath, port, set
76 * operations on vports, etc.), Writes to other state (flow table
77 * modifications, set miscellaneous datapath parameters, etc.) are protected
78 * by ovs_lock.
79 *
80 * Reads are protected by RCU.
81 *
82 * There are a few special cases (mostly stats) that have their own
83 * synchronization but they nest under all of above and don't interact with
84 * each other.
85 *
86 * The RTNL lock nests inside ovs_mutex.
87 */
88
89static DEFINE_MUTEX(ovs_mutex);
90
91void ovs_lock(void)
92{
93	mutex_lock(&ovs_mutex);
94}
95
96void ovs_unlock(void)
97{
98	mutex_unlock(&ovs_mutex);
99}
100
101#ifdef CONFIG_LOCKDEP
102int lockdep_ovsl_is_held(void)
103{
104	if (debug_locks)
105		return lockdep_is_held(&ovs_mutex);
106	else
107		return 1;
108}
109#endif
110
111static struct vport *new_vport(const struct vport_parms *);
112static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
113			     const struct dp_upcall_info *);
114static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
115				  const struct dp_upcall_info *);
116
117/* Must be called with rcu_read_lock or ovs_mutex. */
118static struct datapath *get_dp(struct net *net, int dp_ifindex)
119{
120	struct datapath *dp = NULL;
121	struct net_device *dev;
122
123	rcu_read_lock();
124	dev = dev_get_by_index_rcu(net, dp_ifindex);
125	if (dev) {
126		struct vport *vport = ovs_internal_dev_get_vport(dev);
127		if (vport)
128			dp = vport->dp;
129	}
130	rcu_read_unlock();
131
132	return dp;
133}
134
135/* Must be called with rcu_read_lock or ovs_mutex. */
136static const char *ovs_dp_name(const struct datapath *dp)
137{
138	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
139	return vport->ops->get_name(vport);
140}
141
142static int get_dpifindex(struct datapath *dp)
143{
144	struct vport *local;
145	int ifindex;
146
147	rcu_read_lock();
148
149	local = ovs_vport_rcu(dp, OVSP_LOCAL);
150	if (local)
151		ifindex = netdev_vport_priv(local)->dev->ifindex;
152	else
153		ifindex = 0;
154
155	rcu_read_unlock();
156
157	return ifindex;
158}
159
160static void destroy_dp_rcu(struct rcu_head *rcu)
161{
162	struct datapath *dp = container_of(rcu, struct datapath, rcu);
163
164	free_percpu(dp->stats_percpu);
165	release_net(ovs_dp_get_net(dp));
166	kfree(dp->ports);
167	kfree(dp);
168}
169
170static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
171					    u16 port_no)
172{
173	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
174}
175
176/* Called with ovs_mutex or RCU read lock. */
177struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
178{
179	struct vport *vport;
180	struct hlist_head *head;
181
182	head = vport_hash_bucket(dp, port_no);
183	hlist_for_each_entry_rcu(vport, head, dp_hash_node) {
184		if (vport->port_no == port_no)
185			return vport;
186	}
187	return NULL;
188}
189
190/* Called with ovs_mutex. */
191static struct vport *new_vport(const struct vport_parms *parms)
192{
193	struct vport *vport;
194
195	vport = ovs_vport_add(parms);
196	if (!IS_ERR(vport)) {
197		struct datapath *dp = parms->dp;
198		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
199
200		hlist_add_head_rcu(&vport->dp_hash_node, head);
201	}
202	return vport;
203}
204
205void ovs_dp_detach_port(struct vport *p)
206{
207	ASSERT_OVSL();
208
209	/* First drop references to device. */
210	hlist_del_rcu(&p->dp_hash_node);
211
212	/* Then destroy it. */
213	ovs_vport_del(p);
214}
215
216/* Must be called with rcu_read_lock. */
217void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb)
218{
219	struct datapath *dp = p->dp;
220	struct sw_flow *flow;
221	struct dp_stats_percpu *stats;
222	struct sw_flow_key key;
223	u64 *stats_counter;
224	u32 n_mask_hit;
225	int error;
226
227	stats = this_cpu_ptr(dp->stats_percpu);
228
229	/* Extract flow from 'skb' into 'key'. */
230	error = ovs_flow_extract(skb, p->port_no, &key);
231	if (unlikely(error)) {
232		kfree_skb(skb);
233		return;
234	}
235
236	/* Look up flow. */
237	flow = ovs_flow_tbl_lookup_stats(&dp->table, &key, &n_mask_hit);
238	if (unlikely(!flow)) {
239		struct dp_upcall_info upcall;
240
241		upcall.cmd = OVS_PACKET_CMD_MISS;
242		upcall.key = &key;
243		upcall.userdata = NULL;
244		upcall.portid = p->upcall_portid;
245		ovs_dp_upcall(dp, skb, &upcall);
246		consume_skb(skb);
247		stats_counter = &stats->n_missed;
248		goto out;
249	}
250
251	OVS_CB(skb)->flow = flow;
252	OVS_CB(skb)->pkt_key = &key;
253
254	ovs_flow_stats_update(OVS_CB(skb)->flow, skb);
255	ovs_execute_actions(dp, skb);
256	stats_counter = &stats->n_hit;
257
258out:
259	/* Update datapath statistics. */
260	u64_stats_update_begin(&stats->syncp);
261	(*stats_counter)++;
262	stats->n_mask_hit += n_mask_hit;
263	u64_stats_update_end(&stats->syncp);
264}
265
266static struct genl_family dp_packet_genl_family = {
267	.id = GENL_ID_GENERATE,
268	.hdrsize = sizeof(struct ovs_header),
269	.name = OVS_PACKET_FAMILY,
270	.version = OVS_PACKET_VERSION,
271	.maxattr = OVS_PACKET_ATTR_MAX,
272	.netnsok = true,
273	.parallel_ops = true,
274};
275
276int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
277		  const struct dp_upcall_info *upcall_info)
278{
279	struct dp_stats_percpu *stats;
280	int err;
281
282	if (upcall_info->portid == 0) {
283		err = -ENOTCONN;
284		goto err;
285	}
286
287	if (!skb_is_gso(skb))
288		err = queue_userspace_packet(dp, skb, upcall_info);
289	else
290		err = queue_gso_packets(dp, skb, upcall_info);
291	if (err)
292		goto err;
293
294	return 0;
295
296err:
297	stats = this_cpu_ptr(dp->stats_percpu);
298
299	u64_stats_update_begin(&stats->syncp);
300	stats->n_lost++;
301	u64_stats_update_end(&stats->syncp);
302
303	return err;
304}
305
306static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
307			     const struct dp_upcall_info *upcall_info)
308{
309	unsigned short gso_type = skb_shinfo(skb)->gso_type;
310	struct dp_upcall_info later_info;
311	struct sw_flow_key later_key;
312	struct sk_buff *segs, *nskb;
313	int err;
314
315	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
316	if (IS_ERR(segs))
317		return PTR_ERR(segs);
318
319	/* Queue all of the segments. */
320	skb = segs;
321	do {
322		err = queue_userspace_packet(dp, skb, upcall_info);
323		if (err)
324			break;
325
326		if (skb == segs && gso_type & SKB_GSO_UDP) {
327			/* The initial flow key extracted by ovs_flow_extract()
328			 * in this case is for a first fragment, so we need to
329			 * properly mark later fragments.
330			 */
331			later_key = *upcall_info->key;
332			later_key.ip.frag = OVS_FRAG_TYPE_LATER;
333
334			later_info = *upcall_info;
335			later_info.key = &later_key;
336			upcall_info = &later_info;
337		}
338	} while ((skb = skb->next));
339
340	/* Free all of the segments. */
341	skb = segs;
342	do {
343		nskb = skb->next;
344		if (err)
345			kfree_skb(skb);
346		else
347			consume_skb(skb);
348	} while ((skb = nskb));
349	return err;
350}
351
352static size_t key_attr_size(void)
353{
354	return    nla_total_size(4)   /* OVS_KEY_ATTR_PRIORITY */
355		+ nla_total_size(0)   /* OVS_KEY_ATTR_TUNNEL */
356		  + nla_total_size(8)   /* OVS_TUNNEL_KEY_ATTR_ID */
357		  + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_SRC */
358		  + nla_total_size(4)   /* OVS_TUNNEL_KEY_ATTR_IPV4_DST */
359		  + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TOS */
360		  + nla_total_size(1)   /* OVS_TUNNEL_KEY_ATTR_TTL */
361		  + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_DONT_FRAGMENT */
362		  + nla_total_size(0)   /* OVS_TUNNEL_KEY_ATTR_CSUM */
363		+ nla_total_size(4)   /* OVS_KEY_ATTR_IN_PORT */
364		+ nla_total_size(4)   /* OVS_KEY_ATTR_SKB_MARK */
365		+ nla_total_size(12)  /* OVS_KEY_ATTR_ETHERNET */
366		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
367		+ nla_total_size(4)   /* OVS_KEY_ATTR_8021Q */
368		+ nla_total_size(0)   /* OVS_KEY_ATTR_ENCAP */
369		+ nla_total_size(2)   /* OVS_KEY_ATTR_ETHERTYPE */
370		+ nla_total_size(40)  /* OVS_KEY_ATTR_IPV6 */
371		+ nla_total_size(2)   /* OVS_KEY_ATTR_ICMPV6 */
372		+ nla_total_size(28); /* OVS_KEY_ATTR_ND */
373}
374
375static size_t upcall_msg_size(const struct nlattr *userdata,
376			      unsigned int hdrlen)
377{
378	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
379		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
380		+ nla_total_size(key_attr_size()); /* OVS_PACKET_ATTR_KEY */
381
382	/* OVS_PACKET_ATTR_USERDATA */
383	if (userdata)
384		size += NLA_ALIGN(userdata->nla_len);
385
386	return size;
387}
388
389static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
390				  const struct dp_upcall_info *upcall_info)
391{
392	struct ovs_header *upcall;
393	struct sk_buff *nskb = NULL;
394	struct sk_buff *user_skb; /* to be queued to userspace */
395	struct nlattr *nla;
396	struct genl_info info = {
397		.dst_sk = ovs_dp_get_net(dp)->genl_sock,
398		.snd_portid = upcall_info->portid,
399	};
400	size_t len;
401	unsigned int hlen;
402	int err, dp_ifindex;
403
404	dp_ifindex = get_dpifindex(dp);
405	if (!dp_ifindex)
406		return -ENODEV;
407
408	if (vlan_tx_tag_present(skb)) {
409		nskb = skb_clone(skb, GFP_ATOMIC);
410		if (!nskb)
411			return -ENOMEM;
412
413		nskb = __vlan_put_tag(nskb, nskb->vlan_proto, vlan_tx_tag_get(nskb));
414		if (!nskb)
415			return -ENOMEM;
416
417		nskb->vlan_tci = 0;
418		skb = nskb;
419	}
420
421	if (nla_attr_size(skb->len) > USHRT_MAX) {
422		err = -EFBIG;
423		goto out;
424	}
425
426	/* Complete checksum if needed */
427	if (skb->ip_summed == CHECKSUM_PARTIAL &&
428	    (err = skb_checksum_help(skb)))
429		goto out;
430
431	/* Older versions of OVS user space enforce alignment of the last
432	 * Netlink attribute to NLA_ALIGNTO which would require extensive
433	 * padding logic. Only perform zerocopy if padding is not required.
434	 */
435	if (dp->user_features & OVS_DP_F_UNALIGNED)
436		hlen = skb_zerocopy_headlen(skb);
437	else
438		hlen = skb->len;
439
440	len = upcall_msg_size(upcall_info->userdata, hlen);
441	user_skb = genlmsg_new_unicast(len, &info, GFP_ATOMIC);
442	if (!user_skb) {
443		err = -ENOMEM;
444		goto out;
445	}
446
447	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
448			     0, upcall_info->cmd);
449	upcall->dp_ifindex = dp_ifindex;
450
451	nla = nla_nest_start(user_skb, OVS_PACKET_ATTR_KEY);
452	ovs_nla_put_flow(upcall_info->key, upcall_info->key, user_skb);
453	nla_nest_end(user_skb, nla);
454
455	if (upcall_info->userdata)
456		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
457			  nla_len(upcall_info->userdata),
458			  nla_data(upcall_info->userdata));
459
460	/* Only reserve room for attribute header, packet data is added
461	 * in skb_zerocopy() */
462	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
463		err = -ENOBUFS;
464		goto out;
465	}
466	nla->nla_len = nla_attr_size(skb->len);
467
468	err = skb_zerocopy(user_skb, skb, skb->len, hlen);
469	if (err)
470		goto out;
471
472	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
473	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
474		size_t plen = NLA_ALIGN(user_skb->len) - user_skb->len;
475
476		if (plen > 0)
477			memset(skb_put(user_skb, plen), 0, plen);
478	}
479
480	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
481
482	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
483out:
484	if (err)
485		skb_tx_error(skb);
486	kfree_skb(nskb);
487	return err;
488}
489
490static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
491{
492	struct ovs_header *ovs_header = info->userhdr;
493	struct nlattr **a = info->attrs;
494	struct sw_flow_actions *acts;
495	struct sk_buff *packet;
496	struct sw_flow *flow;
497	struct datapath *dp;
498	struct ethhdr *eth;
499	int len;
500	int err;
501
502	err = -EINVAL;
503	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
504	    !a[OVS_PACKET_ATTR_ACTIONS])
505		goto err;
506
507	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
508	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
509	err = -ENOMEM;
510	if (!packet)
511		goto err;
512	skb_reserve(packet, NET_IP_ALIGN);
513
514	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
515
516	skb_reset_mac_header(packet);
517	eth = eth_hdr(packet);
518
519	/* Normally, setting the skb 'protocol' field would be handled by a
520	 * call to eth_type_trans(), but it assumes there's a sending
521	 * device, which we may not have. */
522	if (ntohs(eth->h_proto) >= ETH_P_802_3_MIN)
523		packet->protocol = eth->h_proto;
524	else
525		packet->protocol = htons(ETH_P_802_2);
526
527	/* Build an sw_flow for sending this packet. */
528	flow = ovs_flow_alloc();
529	err = PTR_ERR(flow);
530	if (IS_ERR(flow))
531		goto err_kfree_skb;
532
533	err = ovs_flow_extract(packet, -1, &flow->key);
534	if (err)
535		goto err_flow_free;
536
537	err = ovs_nla_get_flow_metadata(flow, a[OVS_PACKET_ATTR_KEY]);
538	if (err)
539		goto err_flow_free;
540	acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_PACKET_ATTR_ACTIONS]));
541	err = PTR_ERR(acts);
542	if (IS_ERR(acts))
543		goto err_flow_free;
544
545	err = ovs_nla_copy_actions(a[OVS_PACKET_ATTR_ACTIONS],
546				   &flow->key, 0, &acts);
547	rcu_assign_pointer(flow->sf_acts, acts);
548	if (err)
549		goto err_flow_free;
550
551	OVS_CB(packet)->flow = flow;
552	OVS_CB(packet)->pkt_key = &flow->key;
553	packet->priority = flow->key.phy.priority;
554	packet->mark = flow->key.phy.skb_mark;
555
556	rcu_read_lock();
557	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
558	err = -ENODEV;
559	if (!dp)
560		goto err_unlock;
561
562	local_bh_disable();
563	err = ovs_execute_actions(dp, packet);
564	local_bh_enable();
565	rcu_read_unlock();
566
567	ovs_flow_free(flow, false);
568	return err;
569
570err_unlock:
571	rcu_read_unlock();
572err_flow_free:
573	ovs_flow_free(flow, false);
574err_kfree_skb:
575	kfree_skb(packet);
576err:
577	return err;
578}
579
580static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
581	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
582	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
583	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
584};
585
586static const struct genl_ops dp_packet_genl_ops[] = {
587	{ .cmd = OVS_PACKET_CMD_EXECUTE,
588	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
589	  .policy = packet_policy,
590	  .doit = ovs_packet_cmd_execute
591	}
592};
593
594static void get_dp_stats(struct datapath *dp, struct ovs_dp_stats *stats,
595			 struct ovs_dp_megaflow_stats *mega_stats)
596{
597	int i;
598
599	memset(mega_stats, 0, sizeof(*mega_stats));
600
601	stats->n_flows = ovs_flow_tbl_count(&dp->table);
602	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
603
604	stats->n_hit = stats->n_missed = stats->n_lost = 0;
605
606	for_each_possible_cpu(i) {
607		const struct dp_stats_percpu *percpu_stats;
608		struct dp_stats_percpu local_stats;
609		unsigned int start;
610
611		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
612
613		do {
614			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
615			local_stats = *percpu_stats;
616		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
617
618		stats->n_hit += local_stats.n_hit;
619		stats->n_missed += local_stats.n_missed;
620		stats->n_lost += local_stats.n_lost;
621		mega_stats->n_mask_hit += local_stats.n_mask_hit;
622	}
623}
624
625static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
626	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
627	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
628	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
629};
630
631static struct genl_family dp_flow_genl_family = {
632	.id = GENL_ID_GENERATE,
633	.hdrsize = sizeof(struct ovs_header),
634	.name = OVS_FLOW_FAMILY,
635	.version = OVS_FLOW_VERSION,
636	.maxattr = OVS_FLOW_ATTR_MAX,
637	.netnsok = true,
638	.parallel_ops = true,
639};
640
641static struct genl_multicast_group ovs_dp_flow_multicast_group = {
642	.name = OVS_FLOW_MCGROUP
643};
644
645static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts)
646{
647	return NLMSG_ALIGN(sizeof(struct ovs_header))
648		+ nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_KEY */
649		+ nla_total_size(key_attr_size()) /* OVS_FLOW_ATTR_MASK */
650		+ nla_total_size(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
651		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
652		+ nla_total_size(8) /* OVS_FLOW_ATTR_USED */
653		+ nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */
654}
655
656/* Called with ovs_mutex or RCU read lock. */
657static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp,
658				  struct sk_buff *skb, u32 portid,
659				  u32 seq, u32 flags, u8 cmd)
660{
661	const int skb_orig_len = skb->len;
662	struct nlattr *start;
663	struct ovs_flow_stats stats;
664	__be16 tcp_flags;
665	unsigned long used;
666	struct ovs_header *ovs_header;
667	struct nlattr *nla;
668	int err;
669
670	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family, flags, cmd);
671	if (!ovs_header)
672		return -EMSGSIZE;
673
674	ovs_header->dp_ifindex = get_dpifindex(dp);
675
676	/* Fill flow key. */
677	nla = nla_nest_start(skb, OVS_FLOW_ATTR_KEY);
678	if (!nla)
679		goto nla_put_failure;
680
681	err = ovs_nla_put_flow(&flow->unmasked_key, &flow->unmasked_key, skb);
682	if (err)
683		goto error;
684	nla_nest_end(skb, nla);
685
686	nla = nla_nest_start(skb, OVS_FLOW_ATTR_MASK);
687	if (!nla)
688		goto nla_put_failure;
689
690	err = ovs_nla_put_flow(&flow->key, &flow->mask->key, skb);
691	if (err)
692		goto error;
693
694	nla_nest_end(skb, nla);
695
696	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
697	if (used &&
698	    nla_put_u64(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used)))
699		goto nla_put_failure;
700
701	if (stats.n_packets &&
702	    nla_put(skb, OVS_FLOW_ATTR_STATS, sizeof(struct ovs_flow_stats), &stats))
703		goto nla_put_failure;
704
705	if ((u8)ntohs(tcp_flags) &&
706	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
707		goto nla_put_failure;
708
709	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
710	 * this is the first flow to be dumped into 'skb'.  This is unusual for
711	 * Netlink but individual action lists can be longer than
712	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
713	 * The userspace caller can always fetch the actions separately if it
714	 * really wants them.  (Most userspace callers in fact don't care.)
715	 *
716	 * This can only fail for dump operations because the skb is always
717	 * properly sized for single flows.
718	 */
719	start = nla_nest_start(skb, OVS_FLOW_ATTR_ACTIONS);
720	if (start) {
721		const struct sw_flow_actions *sf_acts;
722
723		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
724
725		err = ovs_nla_put_actions(sf_acts->actions,
726					  sf_acts->actions_len, skb);
727		if (!err)
728			nla_nest_end(skb, start);
729		else {
730			if (skb_orig_len)
731				goto error;
732
733			nla_nest_cancel(skb, start);
734		}
735	} else if (skb_orig_len)
736		goto nla_put_failure;
737
738	return genlmsg_end(skb, ovs_header);
739
740nla_put_failure:
741	err = -EMSGSIZE;
742error:
743	genlmsg_cancel(skb, ovs_header);
744	return err;
745}
746
747/* Must be called with ovs_mutex. */
748static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow,
749					       struct genl_info *info)
750{
751	size_t len;
752
753	len = ovs_flow_cmd_msg_size(ovsl_dereference(flow->sf_acts));
754
755	return genlmsg_new_unicast(len, info, GFP_KERNEL);
756}
757
758/* Must be called with ovs_mutex. */
759static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow,
760					       struct datapath *dp,
761					       struct genl_info *info,
762					       u8 cmd)
763{
764	struct sk_buff *skb;
765	int retval;
766
767	skb = ovs_flow_cmd_alloc_info(flow, info);
768	if (!skb)
769		return ERR_PTR(-ENOMEM);
770
771	retval = ovs_flow_cmd_fill_info(flow, dp, skb, info->snd_portid,
772					info->snd_seq, 0, cmd);
773	BUG_ON(retval < 0);
774	return skb;
775}
776
777static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info)
778{
779	struct nlattr **a = info->attrs;
780	struct ovs_header *ovs_header = info->userhdr;
781	struct sw_flow_key key, masked_key;
782	struct sw_flow *flow = NULL;
783	struct sw_flow_mask mask;
784	struct sk_buff *reply;
785	struct datapath *dp;
786	struct sw_flow_actions *acts = NULL;
787	struct sw_flow_match match;
788	int error;
789
790	/* Extract key. */
791	error = -EINVAL;
792	if (!a[OVS_FLOW_ATTR_KEY])
793		goto error;
794
795	ovs_match_init(&match, &key, &mask);
796	error = ovs_nla_get_match(&match,
797				  a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]);
798	if (error)
799		goto error;
800
801	/* Validate actions. */
802	if (a[OVS_FLOW_ATTR_ACTIONS]) {
803		acts = ovs_nla_alloc_flow_actions(nla_len(a[OVS_FLOW_ATTR_ACTIONS]));
804		error = PTR_ERR(acts);
805		if (IS_ERR(acts))
806			goto error;
807
808		ovs_flow_mask_key(&masked_key, &key, &mask);
809		error = ovs_nla_copy_actions(a[OVS_FLOW_ATTR_ACTIONS],
810					     &masked_key, 0, &acts);
811		if (error) {
812			OVS_NLERR("Flow actions may not be safe on all matching packets.\n");
813			goto err_kfree;
814		}
815	} else if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW) {
816		/* OVS_FLOW_CMD_NEW must have actions. */
817		error = -EINVAL;
818		goto error;
819	}
820
821	ovs_lock();
822	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
823	error = -ENODEV;
824	if (!dp)
825		goto err_unlock_ovs;
826
827	/* Check if this is a duplicate flow */
828	flow = ovs_flow_tbl_lookup(&dp->table, &key);
829	if (!flow) {
830		/* Bail out if we're not allowed to create a new flow. */
831		error = -ENOENT;
832		if (info->genlhdr->cmd == OVS_FLOW_CMD_SET)
833			goto err_unlock_ovs;
834
835		/* Allocate flow. */
836		flow = ovs_flow_alloc();
837		if (IS_ERR(flow)) {
838			error = PTR_ERR(flow);
839			goto err_unlock_ovs;
840		}
841
842		flow->key = masked_key;
843		flow->unmasked_key = key;
844		rcu_assign_pointer(flow->sf_acts, acts);
845
846		/* Put flow in bucket. */
847		error = ovs_flow_tbl_insert(&dp->table, flow, &mask);
848		if (error) {
849			acts = NULL;
850			goto err_flow_free;
851		}
852
853		reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
854	} else {
855		/* We found a matching flow. */
856		/* Bail out if we're not allowed to modify an existing flow.
857		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
858		 * because Generic Netlink treats the latter as a dump
859		 * request.  We also accept NLM_F_EXCL in case that bug ever
860		 * gets fixed.
861		 */
862		error = -EEXIST;
863		if (info->genlhdr->cmd == OVS_FLOW_CMD_NEW &&
864		    info->nlhdr->nlmsg_flags & (NLM_F_CREATE | NLM_F_EXCL))
865			goto err_unlock_ovs;
866
867		/* The unmasked key has to be the same for flow updates. */
868		if (!ovs_flow_cmp_unmasked_key(flow, &match))
869			goto err_unlock_ovs;
870
871		/* Update actions, if present. */
872		if (acts) {
873			struct sw_flow_actions *old_acts;
874
875			old_acts = ovsl_dereference(flow->sf_acts);
876			rcu_assign_pointer(flow->sf_acts, acts);
877			ovs_nla_free_flow_actions(old_acts);
878		}
879		reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
880
881		/* Clear stats. */
882		if (a[OVS_FLOW_ATTR_CLEAR])
883			ovs_flow_stats_clear(flow);
884	}
885	ovs_unlock();
886
887	if (!IS_ERR(reply))
888		ovs_notify(&dp_flow_genl_family, reply, info);
889	else
890		genl_set_err(&dp_flow_genl_family, sock_net(skb->sk), 0,
891			     0, PTR_ERR(reply));
892	return 0;
893
894err_flow_free:
895	ovs_flow_free(flow, false);
896err_unlock_ovs:
897	ovs_unlock();
898err_kfree:
899	kfree(acts);
900error:
901	return error;
902}
903
904static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
905{
906	struct nlattr **a = info->attrs;
907	struct ovs_header *ovs_header = info->userhdr;
908	struct sw_flow_key key;
909	struct sk_buff *reply;
910	struct sw_flow *flow;
911	struct datapath *dp;
912	struct sw_flow_match match;
913	int err;
914
915	if (!a[OVS_FLOW_ATTR_KEY]) {
916		OVS_NLERR("Flow get message rejected, Key attribute missing.\n");
917		return -EINVAL;
918	}
919
920	ovs_match_init(&match, &key, NULL);
921	err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
922	if (err)
923		return err;
924
925	ovs_lock();
926	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
927	if (!dp) {
928		err = -ENODEV;
929		goto unlock;
930	}
931
932	flow = ovs_flow_tbl_lookup(&dp->table, &key);
933	if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
934		err = -ENOENT;
935		goto unlock;
936	}
937
938	reply = ovs_flow_cmd_build_info(flow, dp, info, OVS_FLOW_CMD_NEW);
939	if (IS_ERR(reply)) {
940		err = PTR_ERR(reply);
941		goto unlock;
942	}
943
944	ovs_unlock();
945	return genlmsg_reply(reply, info);
946unlock:
947	ovs_unlock();
948	return err;
949}
950
951static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
952{
953	struct nlattr **a = info->attrs;
954	struct ovs_header *ovs_header = info->userhdr;
955	struct sw_flow_key key;
956	struct sk_buff *reply;
957	struct sw_flow *flow;
958	struct datapath *dp;
959	struct sw_flow_match match;
960	int err;
961
962	ovs_lock();
963	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
964	if (!dp) {
965		err = -ENODEV;
966		goto unlock;
967	}
968
969	if (!a[OVS_FLOW_ATTR_KEY]) {
970		err = ovs_flow_tbl_flush(&dp->table);
971		goto unlock;
972	}
973
974	ovs_match_init(&match, &key, NULL);
975	err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL);
976	if (err)
977		goto unlock;
978
979	flow = ovs_flow_tbl_lookup(&dp->table, &key);
980	if (!flow || !ovs_flow_cmp_unmasked_key(flow, &match)) {
981		err = -ENOENT;
982		goto unlock;
983	}
984
985	reply = ovs_flow_cmd_alloc_info(flow, info);
986	if (!reply) {
987		err = -ENOMEM;
988		goto unlock;
989	}
990
991	ovs_flow_tbl_remove(&dp->table, flow);
992
993	err = ovs_flow_cmd_fill_info(flow, dp, reply, info->snd_portid,
994				     info->snd_seq, 0, OVS_FLOW_CMD_DEL);
995	BUG_ON(err < 0);
996
997	ovs_flow_free(flow, true);
998	ovs_unlock();
999
1000	ovs_notify(&dp_flow_genl_family, reply, info);
1001	return 0;
1002unlock:
1003	ovs_unlock();
1004	return err;
1005}
1006
1007static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1008{
1009	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1010	struct table_instance *ti;
1011	struct datapath *dp;
1012
1013	rcu_read_lock();
1014	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1015	if (!dp) {
1016		rcu_read_unlock();
1017		return -ENODEV;
1018	}
1019
1020	ti = rcu_dereference(dp->table.ti);
1021	for (;;) {
1022		struct sw_flow *flow;
1023		u32 bucket, obj;
1024
1025		bucket = cb->args[0];
1026		obj = cb->args[1];
1027		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1028		if (!flow)
1029			break;
1030
1031		if (ovs_flow_cmd_fill_info(flow, dp, skb,
1032					   NETLINK_CB(cb->skb).portid,
1033					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1034					   OVS_FLOW_CMD_NEW) < 0)
1035			break;
1036
1037		cb->args[0] = bucket;
1038		cb->args[1] = obj;
1039	}
1040	rcu_read_unlock();
1041	return skb->len;
1042}
1043
1044static const struct genl_ops dp_flow_genl_ops[] = {
1045	{ .cmd = OVS_FLOW_CMD_NEW,
1046	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1047	  .policy = flow_policy,
1048	  .doit = ovs_flow_cmd_new_or_set
1049	},
1050	{ .cmd = OVS_FLOW_CMD_DEL,
1051	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1052	  .policy = flow_policy,
1053	  .doit = ovs_flow_cmd_del
1054	},
1055	{ .cmd = OVS_FLOW_CMD_GET,
1056	  .flags = 0,		    /* OK for unprivileged users. */
1057	  .policy = flow_policy,
1058	  .doit = ovs_flow_cmd_get,
1059	  .dumpit = ovs_flow_cmd_dump
1060	},
1061	{ .cmd = OVS_FLOW_CMD_SET,
1062	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1063	  .policy = flow_policy,
1064	  .doit = ovs_flow_cmd_new_or_set,
1065	},
1066};
1067
1068static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1069	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1070	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1071	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1072};
1073
1074static struct genl_family dp_datapath_genl_family = {
1075	.id = GENL_ID_GENERATE,
1076	.hdrsize = sizeof(struct ovs_header),
1077	.name = OVS_DATAPATH_FAMILY,
1078	.version = OVS_DATAPATH_VERSION,
1079	.maxattr = OVS_DP_ATTR_MAX,
1080	.netnsok = true,
1081	.parallel_ops = true,
1082};
1083
1084static struct genl_multicast_group ovs_dp_datapath_multicast_group = {
1085	.name = OVS_DATAPATH_MCGROUP
1086};
1087
1088static size_t ovs_dp_cmd_msg_size(void)
1089{
1090	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1091
1092	msgsize += nla_total_size(IFNAMSIZ);
1093	msgsize += nla_total_size(sizeof(struct ovs_dp_stats));
1094	msgsize += nla_total_size(sizeof(struct ovs_dp_megaflow_stats));
1095	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1096
1097	return msgsize;
1098}
1099
1100/* Called with ovs_mutex or RCU read lock. */
1101static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1102				u32 portid, u32 seq, u32 flags, u8 cmd)
1103{
1104	struct ovs_header *ovs_header;
1105	struct ovs_dp_stats dp_stats;
1106	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1107	int err;
1108
1109	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1110				   flags, cmd);
1111	if (!ovs_header)
1112		goto error;
1113
1114	ovs_header->dp_ifindex = get_dpifindex(dp);
1115
1116	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1117	if (err)
1118		goto nla_put_failure;
1119
1120	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1121	if (nla_put(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1122			&dp_stats))
1123		goto nla_put_failure;
1124
1125	if (nla_put(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1126			sizeof(struct ovs_dp_megaflow_stats),
1127			&dp_megaflow_stats))
1128		goto nla_put_failure;
1129
1130	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1131		goto nla_put_failure;
1132
1133	return genlmsg_end(skb, ovs_header);
1134
1135nla_put_failure:
1136	genlmsg_cancel(skb, ovs_header);
1137error:
1138	return -EMSGSIZE;
1139}
1140
1141/* Must be called with ovs_mutex. */
1142static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp,
1143					     struct genl_info *info, u8 cmd)
1144{
1145	struct sk_buff *skb;
1146	int retval;
1147
1148	skb = genlmsg_new_unicast(ovs_dp_cmd_msg_size(), info, GFP_KERNEL);
1149	if (!skb)
1150		return ERR_PTR(-ENOMEM);
1151
1152	retval = ovs_dp_cmd_fill_info(dp, skb, info->snd_portid, info->snd_seq, 0, cmd);
1153	if (retval < 0) {
1154		kfree_skb(skb);
1155		return ERR_PTR(retval);
1156	}
1157	return skb;
1158}
1159
1160/* Called with rcu_read_lock or ovs_mutex. */
1161static struct datapath *lookup_datapath(struct net *net,
1162					struct ovs_header *ovs_header,
1163					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1164{
1165	struct datapath *dp;
1166
1167	if (!a[OVS_DP_ATTR_NAME])
1168		dp = get_dp(net, ovs_header->dp_ifindex);
1169	else {
1170		struct vport *vport;
1171
1172		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1173		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1174	}
1175	return dp ? dp : ERR_PTR(-ENODEV);
1176}
1177
1178static void ovs_dp_reset_user_features(struct sk_buff *skb, struct genl_info *info)
1179{
1180	struct datapath *dp;
1181
1182	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1183	if (IS_ERR(dp))
1184		return;
1185
1186	WARN(dp->user_features, "Dropping previously announced user features\n");
1187	dp->user_features = 0;
1188}
1189
1190static void ovs_dp_change(struct datapath *dp, struct nlattr **a)
1191{
1192	if (a[OVS_DP_ATTR_USER_FEATURES])
1193		dp->user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1194}
1195
1196static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1197{
1198	struct nlattr **a = info->attrs;
1199	struct vport_parms parms;
1200	struct sk_buff *reply;
1201	struct datapath *dp;
1202	struct vport *vport;
1203	struct ovs_net *ovs_net;
1204	int err, i;
1205
1206	err = -EINVAL;
1207	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1208		goto err;
1209
1210	ovs_lock();
1211
1212	err = -ENOMEM;
1213	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1214	if (dp == NULL)
1215		goto err_unlock_ovs;
1216
1217	ovs_dp_set_net(dp, hold_net(sock_net(skb->sk)));
1218
1219	/* Allocate table. */
1220	err = ovs_flow_tbl_init(&dp->table);
1221	if (err)
1222		goto err_free_dp;
1223
1224	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1225	if (!dp->stats_percpu) {
1226		err = -ENOMEM;
1227		goto err_destroy_table;
1228	}
1229
1230	dp->ports = kmalloc(DP_VPORT_HASH_BUCKETS * sizeof(struct hlist_head),
1231			    GFP_KERNEL);
1232	if (!dp->ports) {
1233		err = -ENOMEM;
1234		goto err_destroy_percpu;
1235	}
1236
1237	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1238		INIT_HLIST_HEAD(&dp->ports[i]);
1239
1240	/* Set up our datapath device. */
1241	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1242	parms.type = OVS_VPORT_TYPE_INTERNAL;
1243	parms.options = NULL;
1244	parms.dp = dp;
1245	parms.port_no = OVSP_LOCAL;
1246	parms.upcall_portid = nla_get_u32(a[OVS_DP_ATTR_UPCALL_PID]);
1247
1248	ovs_dp_change(dp, a);
1249
1250	vport = new_vport(&parms);
1251	if (IS_ERR(vport)) {
1252		err = PTR_ERR(vport);
1253		if (err == -EBUSY)
1254			err = -EEXIST;
1255
1256		if (err == -EEXIST) {
1257			/* An outdated user space instance that does not understand
1258			 * the concept of user_features has attempted to create a new
1259			 * datapath and is likely to reuse it. Drop all user features.
1260			 */
1261			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1262				ovs_dp_reset_user_features(skb, info);
1263		}
1264
1265		goto err_destroy_ports_array;
1266	}
1267
1268	reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1269	err = PTR_ERR(reply);
1270	if (IS_ERR(reply))
1271		goto err_destroy_local_port;
1272
1273	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1274	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1275
1276	ovs_unlock();
1277
1278	ovs_notify(&dp_datapath_genl_family, reply, info);
1279	return 0;
1280
1281err_destroy_local_port:
1282	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1283err_destroy_ports_array:
1284	kfree(dp->ports);
1285err_destroy_percpu:
1286	free_percpu(dp->stats_percpu);
1287err_destroy_table:
1288	ovs_flow_tbl_destroy(&dp->table, false);
1289err_free_dp:
1290	release_net(ovs_dp_get_net(dp));
1291	kfree(dp);
1292err_unlock_ovs:
1293	ovs_unlock();
1294err:
1295	return err;
1296}
1297
1298/* Called with ovs_mutex. */
1299static void __dp_destroy(struct datapath *dp)
1300{
1301	int i;
1302
1303	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1304		struct vport *vport;
1305		struct hlist_node *n;
1306
1307		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1308			if (vport->port_no != OVSP_LOCAL)
1309				ovs_dp_detach_port(vport);
1310	}
1311
1312	list_del_rcu(&dp->list_node);
1313
1314	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1315	 * all ports in datapath are destroyed first before freeing datapath.
1316	 */
1317	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1318
1319	/* RCU destroy the flow table */
1320	ovs_flow_tbl_destroy(&dp->table, true);
1321
1322	call_rcu(&dp->rcu, destroy_dp_rcu);
1323}
1324
1325static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1326{
1327	struct sk_buff *reply;
1328	struct datapath *dp;
1329	int err;
1330
1331	ovs_lock();
1332	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1333	err = PTR_ERR(dp);
1334	if (IS_ERR(dp))
1335		goto unlock;
1336
1337	reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_DEL);
1338	err = PTR_ERR(reply);
1339	if (IS_ERR(reply))
1340		goto unlock;
1341
1342	__dp_destroy(dp);
1343	ovs_unlock();
1344
1345	ovs_notify(&dp_datapath_genl_family, reply, info);
1346
1347	return 0;
1348unlock:
1349	ovs_unlock();
1350	return err;
1351}
1352
1353static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1354{
1355	struct sk_buff *reply;
1356	struct datapath *dp;
1357	int err;
1358
1359	ovs_lock();
1360	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1361	err = PTR_ERR(dp);
1362	if (IS_ERR(dp))
1363		goto unlock;
1364
1365	ovs_dp_change(dp, info->attrs);
1366
1367	reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1368	if (IS_ERR(reply)) {
1369		err = PTR_ERR(reply);
1370		genl_set_err(&dp_datapath_genl_family, sock_net(skb->sk), 0,
1371			     0, err);
1372		err = 0;
1373		goto unlock;
1374	}
1375
1376	ovs_unlock();
1377	ovs_notify(&dp_datapath_genl_family, reply, info);
1378
1379	return 0;
1380unlock:
1381	ovs_unlock();
1382	return err;
1383}
1384
1385static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1386{
1387	struct sk_buff *reply;
1388	struct datapath *dp;
1389	int err;
1390
1391	ovs_lock();
1392	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1393	if (IS_ERR(dp)) {
1394		err = PTR_ERR(dp);
1395		goto unlock;
1396	}
1397
1398	reply = ovs_dp_cmd_build_info(dp, info, OVS_DP_CMD_NEW);
1399	if (IS_ERR(reply)) {
1400		err = PTR_ERR(reply);
1401		goto unlock;
1402	}
1403
1404	ovs_unlock();
1405	return genlmsg_reply(reply, info);
1406
1407unlock:
1408	ovs_unlock();
1409	return err;
1410}
1411
1412static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1413{
1414	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1415	struct datapath *dp;
1416	int skip = cb->args[0];
1417	int i = 0;
1418
1419	rcu_read_lock();
1420	list_for_each_entry_rcu(dp, &ovs_net->dps, list_node) {
1421		if (i >= skip &&
1422		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1423					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1424					 OVS_DP_CMD_NEW) < 0)
1425			break;
1426		i++;
1427	}
1428	rcu_read_unlock();
1429
1430	cb->args[0] = i;
1431
1432	return skb->len;
1433}
1434
1435static const struct genl_ops dp_datapath_genl_ops[] = {
1436	{ .cmd = OVS_DP_CMD_NEW,
1437	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1438	  .policy = datapath_policy,
1439	  .doit = ovs_dp_cmd_new
1440	},
1441	{ .cmd = OVS_DP_CMD_DEL,
1442	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1443	  .policy = datapath_policy,
1444	  .doit = ovs_dp_cmd_del
1445	},
1446	{ .cmd = OVS_DP_CMD_GET,
1447	  .flags = 0,		    /* OK for unprivileged users. */
1448	  .policy = datapath_policy,
1449	  .doit = ovs_dp_cmd_get,
1450	  .dumpit = ovs_dp_cmd_dump
1451	},
1452	{ .cmd = OVS_DP_CMD_SET,
1453	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1454	  .policy = datapath_policy,
1455	  .doit = ovs_dp_cmd_set,
1456	},
1457};
1458
1459static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
1460	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1461	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
1462	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
1463	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
1464	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1465	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
1466};
1467
1468struct genl_family dp_vport_genl_family = {
1469	.id = GENL_ID_GENERATE,
1470	.hdrsize = sizeof(struct ovs_header),
1471	.name = OVS_VPORT_FAMILY,
1472	.version = OVS_VPORT_VERSION,
1473	.maxattr = OVS_VPORT_ATTR_MAX,
1474	.netnsok = true,
1475	.parallel_ops = true,
1476};
1477
1478static struct genl_multicast_group ovs_dp_vport_multicast_group = {
1479	.name = OVS_VPORT_MCGROUP
1480};
1481
1482/* Called with ovs_mutex or RCU read lock. */
1483static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
1484				   u32 portid, u32 seq, u32 flags, u8 cmd)
1485{
1486	struct ovs_header *ovs_header;
1487	struct ovs_vport_stats vport_stats;
1488	int err;
1489
1490	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
1491				 flags, cmd);
1492	if (!ovs_header)
1493		return -EMSGSIZE;
1494
1495	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
1496
1497	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
1498	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
1499	    nla_put_string(skb, OVS_VPORT_ATTR_NAME, vport->ops->get_name(vport)) ||
1500	    nla_put_u32(skb, OVS_VPORT_ATTR_UPCALL_PID, vport->upcall_portid))
1501		goto nla_put_failure;
1502
1503	ovs_vport_get_stats(vport, &vport_stats);
1504	if (nla_put(skb, OVS_VPORT_ATTR_STATS, sizeof(struct ovs_vport_stats),
1505		    &vport_stats))
1506		goto nla_put_failure;
1507
1508	err = ovs_vport_get_options(vport, skb);
1509	if (err == -EMSGSIZE)
1510		goto error;
1511
1512	return genlmsg_end(skb, ovs_header);
1513
1514nla_put_failure:
1515	err = -EMSGSIZE;
1516error:
1517	genlmsg_cancel(skb, ovs_header);
1518	return err;
1519}
1520
1521/* Called with ovs_mutex or RCU read lock. */
1522struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, u32 portid,
1523					 u32 seq, u8 cmd)
1524{
1525	struct sk_buff *skb;
1526	int retval;
1527
1528	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
1529	if (!skb)
1530		return ERR_PTR(-ENOMEM);
1531
1532	retval = ovs_vport_cmd_fill_info(vport, skb, portid, seq, 0, cmd);
1533	BUG_ON(retval < 0);
1534
1535	return skb;
1536}
1537
1538/* Called with ovs_mutex or RCU read lock. */
1539static struct vport *lookup_vport(struct net *net,
1540				  struct ovs_header *ovs_header,
1541				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
1542{
1543	struct datapath *dp;
1544	struct vport *vport;
1545
1546	if (a[OVS_VPORT_ATTR_NAME]) {
1547		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
1548		if (!vport)
1549			return ERR_PTR(-ENODEV);
1550		if (ovs_header->dp_ifindex &&
1551		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
1552			return ERR_PTR(-ENODEV);
1553		return vport;
1554	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
1555		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1556
1557		if (port_no >= DP_MAX_PORTS)
1558			return ERR_PTR(-EFBIG);
1559
1560		dp = get_dp(net, ovs_header->dp_ifindex);
1561		if (!dp)
1562			return ERR_PTR(-ENODEV);
1563
1564		vport = ovs_vport_ovsl_rcu(dp, port_no);
1565		if (!vport)
1566			return ERR_PTR(-ENODEV);
1567		return vport;
1568	} else
1569		return ERR_PTR(-EINVAL);
1570}
1571
1572static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
1573{
1574	struct nlattr **a = info->attrs;
1575	struct ovs_header *ovs_header = info->userhdr;
1576	struct vport_parms parms;
1577	struct sk_buff *reply;
1578	struct vport *vport;
1579	struct datapath *dp;
1580	u32 port_no;
1581	int err;
1582
1583	err = -EINVAL;
1584	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
1585	    !a[OVS_VPORT_ATTR_UPCALL_PID])
1586		goto exit;
1587
1588	ovs_lock();
1589	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1590	err = -ENODEV;
1591	if (!dp)
1592		goto exit_unlock;
1593
1594	if (a[OVS_VPORT_ATTR_PORT_NO]) {
1595		port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
1596
1597		err = -EFBIG;
1598		if (port_no >= DP_MAX_PORTS)
1599			goto exit_unlock;
1600
1601		vport = ovs_vport_ovsl(dp, port_no);
1602		err = -EBUSY;
1603		if (vport)
1604			goto exit_unlock;
1605	} else {
1606		for (port_no = 1; ; port_no++) {
1607			if (port_no >= DP_MAX_PORTS) {
1608				err = -EFBIG;
1609				goto exit_unlock;
1610			}
1611			vport = ovs_vport_ovsl(dp, port_no);
1612			if (!vport)
1613				break;
1614		}
1615	}
1616
1617	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
1618	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
1619	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
1620	parms.dp = dp;
1621	parms.port_no = port_no;
1622	parms.upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1623
1624	vport = new_vport(&parms);
1625	err = PTR_ERR(vport);
1626	if (IS_ERR(vport))
1627		goto exit_unlock;
1628
1629	err = 0;
1630	reply = ovs_vport_cmd_build_info(vport, info->snd_portid, info->snd_seq,
1631					 OVS_VPORT_CMD_NEW);
1632	if (IS_ERR(reply)) {
1633		err = PTR_ERR(reply);
1634		ovs_dp_detach_port(vport);
1635		goto exit_unlock;
1636	}
1637
1638	ovs_notify(&dp_vport_genl_family, reply, info);
1639
1640exit_unlock:
1641	ovs_unlock();
1642exit:
1643	return err;
1644}
1645
1646static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
1647{
1648	struct nlattr **a = info->attrs;
1649	struct sk_buff *reply;
1650	struct vport *vport;
1651	int err;
1652
1653	ovs_lock();
1654	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1655	err = PTR_ERR(vport);
1656	if (IS_ERR(vport))
1657		goto exit_unlock;
1658
1659	if (a[OVS_VPORT_ATTR_TYPE] &&
1660	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
1661		err = -EINVAL;
1662		goto exit_unlock;
1663	}
1664
1665	reply = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1666	if (!reply) {
1667		err = -ENOMEM;
1668		goto exit_unlock;
1669	}
1670
1671	if (a[OVS_VPORT_ATTR_OPTIONS]) {
1672		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
1673		if (err)
1674			goto exit_free;
1675	}
1676
1677	if (a[OVS_VPORT_ATTR_UPCALL_PID])
1678		vport->upcall_portid = nla_get_u32(a[OVS_VPORT_ATTR_UPCALL_PID]);
1679
1680	err = ovs_vport_cmd_fill_info(vport, reply, info->snd_portid,
1681				      info->snd_seq, 0, OVS_VPORT_CMD_NEW);
1682	BUG_ON(err < 0);
1683
1684	ovs_unlock();
1685	ovs_notify(&dp_vport_genl_family, reply, info);
1686	return 0;
1687
1688exit_free:
1689	kfree_skb(reply);
1690exit_unlock:
1691	ovs_unlock();
1692	return err;
1693}
1694
1695static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
1696{
1697	struct nlattr **a = info->attrs;
1698	struct sk_buff *reply;
1699	struct vport *vport;
1700	int err;
1701
1702	ovs_lock();
1703	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
1704	err = PTR_ERR(vport);
1705	if (IS_ERR(vport))
1706		goto exit_unlock;
1707
1708	if (vport->port_no == OVSP_LOCAL) {
1709		err = -EINVAL;
1710		goto exit_unlock;
1711	}
1712
1713	reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1714					 info->snd_seq, OVS_VPORT_CMD_DEL);
1715	err = PTR_ERR(reply);
1716	if (IS_ERR(reply))
1717		goto exit_unlock;
1718
1719	err = 0;
1720	ovs_dp_detach_port(vport);
1721
1722	ovs_notify(&dp_vport_genl_family, reply, info);
1723
1724exit_unlock:
1725	ovs_unlock();
1726	return err;
1727}
1728
1729static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
1730{
1731	struct nlattr **a = info->attrs;
1732	struct ovs_header *ovs_header = info->userhdr;
1733	struct sk_buff *reply;
1734	struct vport *vport;
1735	int err;
1736
1737	rcu_read_lock();
1738	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
1739	err = PTR_ERR(vport);
1740	if (IS_ERR(vport))
1741		goto exit_unlock;
1742
1743	reply = ovs_vport_cmd_build_info(vport, info->snd_portid,
1744					 info->snd_seq, OVS_VPORT_CMD_NEW);
1745	err = PTR_ERR(reply);
1746	if (IS_ERR(reply))
1747		goto exit_unlock;
1748
1749	rcu_read_unlock();
1750
1751	return genlmsg_reply(reply, info);
1752
1753exit_unlock:
1754	rcu_read_unlock();
1755	return err;
1756}
1757
1758static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1759{
1760	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1761	struct datapath *dp;
1762	int bucket = cb->args[0], skip = cb->args[1];
1763	int i, j = 0;
1764
1765	rcu_read_lock();
1766	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1767	if (!dp) {
1768		rcu_read_unlock();
1769		return -ENODEV;
1770	}
1771	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
1772		struct vport *vport;
1773
1774		j = 0;
1775		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
1776			if (j >= skip &&
1777			    ovs_vport_cmd_fill_info(vport, skb,
1778						    NETLINK_CB(cb->skb).portid,
1779						    cb->nlh->nlmsg_seq,
1780						    NLM_F_MULTI,
1781						    OVS_VPORT_CMD_NEW) < 0)
1782				goto out;
1783
1784			j++;
1785		}
1786		skip = 0;
1787	}
1788out:
1789	rcu_read_unlock();
1790
1791	cb->args[0] = i;
1792	cb->args[1] = j;
1793
1794	return skb->len;
1795}
1796
1797static const struct genl_ops dp_vport_genl_ops[] = {
1798	{ .cmd = OVS_VPORT_CMD_NEW,
1799	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1800	  .policy = vport_policy,
1801	  .doit = ovs_vport_cmd_new
1802	},
1803	{ .cmd = OVS_VPORT_CMD_DEL,
1804	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1805	  .policy = vport_policy,
1806	  .doit = ovs_vport_cmd_del
1807	},
1808	{ .cmd = OVS_VPORT_CMD_GET,
1809	  .flags = 0,		    /* OK for unprivileged users. */
1810	  .policy = vport_policy,
1811	  .doit = ovs_vport_cmd_get,
1812	  .dumpit = ovs_vport_cmd_dump
1813	},
1814	{ .cmd = OVS_VPORT_CMD_SET,
1815	  .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1816	  .policy = vport_policy,
1817	  .doit = ovs_vport_cmd_set,
1818	},
1819};
1820
1821struct genl_family_and_ops {
1822	struct genl_family *family;
1823	const struct genl_ops *ops;
1824	int n_ops;
1825	const struct genl_multicast_group *group;
1826};
1827
1828static const struct genl_family_and_ops dp_genl_families[] = {
1829	{ &dp_datapath_genl_family,
1830	  dp_datapath_genl_ops, ARRAY_SIZE(dp_datapath_genl_ops),
1831	  &ovs_dp_datapath_multicast_group },
1832	{ &dp_vport_genl_family,
1833	  dp_vport_genl_ops, ARRAY_SIZE(dp_vport_genl_ops),
1834	  &ovs_dp_vport_multicast_group },
1835	{ &dp_flow_genl_family,
1836	  dp_flow_genl_ops, ARRAY_SIZE(dp_flow_genl_ops),
1837	  &ovs_dp_flow_multicast_group },
1838	{ &dp_packet_genl_family,
1839	  dp_packet_genl_ops, ARRAY_SIZE(dp_packet_genl_ops),
1840	  NULL },
1841};
1842
1843static void dp_unregister_genl(int n_families)
1844{
1845	int i;
1846
1847	for (i = 0; i < n_families; i++)
1848		genl_unregister_family(dp_genl_families[i].family);
1849}
1850
1851static int dp_register_genl(void)
1852{
1853	int n_registered;
1854	int err;
1855	int i;
1856
1857	n_registered = 0;
1858	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
1859		const struct genl_family_and_ops *f = &dp_genl_families[i];
1860
1861		f->family->ops = f->ops;
1862		f->family->n_ops = f->n_ops;
1863		f->family->mcgrps = f->group;
1864		f->family->n_mcgrps = f->group ? 1 : 0;
1865		err = genl_register_family(f->family);
1866		if (err)
1867			goto error;
1868		n_registered++;
1869	}
1870
1871	return 0;
1872
1873error:
1874	dp_unregister_genl(n_registered);
1875	return err;
1876}
1877
1878static int __net_init ovs_init_net(struct net *net)
1879{
1880	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1881
1882	INIT_LIST_HEAD(&ovs_net->dps);
1883	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
1884	return 0;
1885}
1886
1887static void __net_exit ovs_exit_net(struct net *net)
1888{
1889	struct datapath *dp, *dp_next;
1890	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1891
1892	ovs_lock();
1893	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
1894		__dp_destroy(dp);
1895	ovs_unlock();
1896
1897	cancel_work_sync(&ovs_net->dp_notify_work);
1898}
1899
1900static struct pernet_operations ovs_net_ops = {
1901	.init = ovs_init_net,
1902	.exit = ovs_exit_net,
1903	.id   = &ovs_net_id,
1904	.size = sizeof(struct ovs_net),
1905};
1906
1907static int __init dp_init(void)
1908{
1909	int err;
1910
1911	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) > FIELD_SIZEOF(struct sk_buff, cb));
1912
1913	pr_info("Open vSwitch switching datapath\n");
1914
1915	err = ovs_flow_init();
1916	if (err)
1917		goto error;
1918
1919	err = ovs_vport_init();
1920	if (err)
1921		goto error_flow_exit;
1922
1923	err = register_pernet_device(&ovs_net_ops);
1924	if (err)
1925		goto error_vport_exit;
1926
1927	err = register_netdevice_notifier(&ovs_dp_device_notifier);
1928	if (err)
1929		goto error_netns_exit;
1930
1931	err = dp_register_genl();
1932	if (err < 0)
1933		goto error_unreg_notifier;
1934
1935	return 0;
1936
1937error_unreg_notifier:
1938	unregister_netdevice_notifier(&ovs_dp_device_notifier);
1939error_netns_exit:
1940	unregister_pernet_device(&ovs_net_ops);
1941error_vport_exit:
1942	ovs_vport_exit();
1943error_flow_exit:
1944	ovs_flow_exit();
1945error:
1946	return err;
1947}
1948
1949static void dp_cleanup(void)
1950{
1951	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
1952	unregister_netdevice_notifier(&ovs_dp_device_notifier);
1953	unregister_pernet_device(&ovs_net_ops);
1954	rcu_barrier();
1955	ovs_vport_exit();
1956	ovs_flow_exit();
1957}
1958
1959module_init(dp_init);
1960module_exit(dp_cleanup);
1961
1962MODULE_DESCRIPTION("Open vSwitch switching datapath");
1963MODULE_LICENSE("GPL");
1964