xref: /openbmc/linux/net/openvswitch/datapath.c (revision 2cf1c348)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2007-2014 Nicira, Inc.
4  */
5 
6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7 
8 #include <linux/init.h>
9 #include <linux/module.h>
10 #include <linux/if_arp.h>
11 #include <linux/if_vlan.h>
12 #include <linux/in.h>
13 #include <linux/ip.h>
14 #include <linux/jhash.h>
15 #include <linux/delay.h>
16 #include <linux/time.h>
17 #include <linux/etherdevice.h>
18 #include <linux/genetlink.h>
19 #include <linux/kernel.h>
20 #include <linux/kthread.h>
21 #include <linux/mutex.h>
22 #include <linux/percpu.h>
23 #include <linux/rcupdate.h>
24 #include <linux/tcp.h>
25 #include <linux/udp.h>
26 #include <linux/ethtool.h>
27 #include <linux/wait.h>
28 #include <asm/div64.h>
29 #include <linux/highmem.h>
30 #include <linux/netfilter_bridge.h>
31 #include <linux/netfilter_ipv4.h>
32 #include <linux/inetdevice.h>
33 #include <linux/list.h>
34 #include <linux/openvswitch.h>
35 #include <linux/rculist.h>
36 #include <linux/dmi.h>
37 #include <net/genetlink.h>
38 #include <net/net_namespace.h>
39 #include <net/netns/generic.h>
40 
41 #include "datapath.h"
42 #include "flow.h"
43 #include "flow_table.h"
44 #include "flow_netlink.h"
45 #include "meter.h"
46 #include "openvswitch_trace.h"
47 #include "vport-internal_dev.h"
48 #include "vport-netdev.h"
49 
50 unsigned int ovs_net_id __read_mostly;
51 
52 static struct genl_family dp_packet_genl_family;
53 static struct genl_family dp_flow_genl_family;
54 static struct genl_family dp_datapath_genl_family;
55 
56 static const struct nla_policy flow_policy[];
57 
58 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
59 	.name = OVS_FLOW_MCGROUP,
60 };
61 
62 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
63 	.name = OVS_DATAPATH_MCGROUP,
64 };
65 
66 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
67 	.name = OVS_VPORT_MCGROUP,
68 };
69 
70 /* Check if need to build a reply message.
71  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
72 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
73 			    unsigned int group)
74 {
75 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
76 	       genl_has_listeners(family, genl_info_net(info), group);
77 }
78 
79 static void ovs_notify(struct genl_family *family,
80 		       struct sk_buff *skb, struct genl_info *info)
81 {
82 	genl_notify(family, skb, info, 0, GFP_KERNEL);
83 }
84 
85 /**
86  * DOC: Locking:
87  *
88  * All writes e.g. Writes to device state (add/remove datapath, port, set
89  * operations on vports, etc.), Writes to other state (flow table
90  * modifications, set miscellaneous datapath parameters, etc.) are protected
91  * by ovs_lock.
92  *
93  * Reads are protected by RCU.
94  *
95  * There are a few special cases (mostly stats) that have their own
96  * synchronization but they nest under all of above and don't interact with
97  * each other.
98  *
99  * The RTNL lock nests inside ovs_mutex.
100  */
101 
102 static DEFINE_MUTEX(ovs_mutex);
103 
104 void ovs_lock(void)
105 {
106 	mutex_lock(&ovs_mutex);
107 }
108 
109 void ovs_unlock(void)
110 {
111 	mutex_unlock(&ovs_mutex);
112 }
113 
114 #ifdef CONFIG_LOCKDEP
115 int lockdep_ovsl_is_held(void)
116 {
117 	if (debug_locks)
118 		return lockdep_is_held(&ovs_mutex);
119 	else
120 		return 1;
121 }
122 #endif
123 
124 static struct vport *new_vport(const struct vport_parms *);
125 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
126 			     const struct sw_flow_key *,
127 			     const struct dp_upcall_info *,
128 			     uint32_t cutlen);
129 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
130 				  const struct sw_flow_key *,
131 				  const struct dp_upcall_info *,
132 				  uint32_t cutlen);
133 
134 static void ovs_dp_masks_rebalance(struct work_struct *work);
135 
136 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
137 
138 /* Must be called with rcu_read_lock or ovs_mutex. */
139 const char *ovs_dp_name(const struct datapath *dp)
140 {
141 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
142 	return ovs_vport_name(vport);
143 }
144 
145 static int get_dpifindex(const struct datapath *dp)
146 {
147 	struct vport *local;
148 	int ifindex;
149 
150 	rcu_read_lock();
151 
152 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
153 	if (local)
154 		ifindex = local->dev->ifindex;
155 	else
156 		ifindex = 0;
157 
158 	rcu_read_unlock();
159 
160 	return ifindex;
161 }
162 
163 static void destroy_dp_rcu(struct rcu_head *rcu)
164 {
165 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
166 
167 	ovs_flow_tbl_destroy(&dp->table);
168 	free_percpu(dp->stats_percpu);
169 	kfree(dp->ports);
170 	ovs_meters_exit(dp);
171 	kfree(rcu_dereference_raw(dp->upcall_portids));
172 	kfree(dp);
173 }
174 
175 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
176 					    u16 port_no)
177 {
178 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
179 }
180 
181 /* Called with ovs_mutex or RCU read lock. */
182 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
183 {
184 	struct vport *vport;
185 	struct hlist_head *head;
186 
187 	head = vport_hash_bucket(dp, port_no);
188 	hlist_for_each_entry_rcu(vport, head, dp_hash_node,
189 				 lockdep_ovsl_is_held()) {
190 		if (vport->port_no == port_no)
191 			return vport;
192 	}
193 	return NULL;
194 }
195 
196 /* Called with ovs_mutex. */
197 static struct vport *new_vport(const struct vport_parms *parms)
198 {
199 	struct vport *vport;
200 
201 	vport = ovs_vport_add(parms);
202 	if (!IS_ERR(vport)) {
203 		struct datapath *dp = parms->dp;
204 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
205 
206 		hlist_add_head_rcu(&vport->dp_hash_node, head);
207 	}
208 	return vport;
209 }
210 
211 void ovs_dp_detach_port(struct vport *p)
212 {
213 	ASSERT_OVSL();
214 
215 	/* First drop references to device. */
216 	hlist_del_rcu(&p->dp_hash_node);
217 
218 	/* Then destroy it. */
219 	ovs_vport_del(p);
220 }
221 
222 /* Must be called with rcu_read_lock. */
223 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
224 {
225 	const struct vport *p = OVS_CB(skb)->input_vport;
226 	struct datapath *dp = p->dp;
227 	struct sw_flow *flow;
228 	struct sw_flow_actions *sf_acts;
229 	struct dp_stats_percpu *stats;
230 	u64 *stats_counter;
231 	u32 n_mask_hit;
232 	u32 n_cache_hit;
233 	int error;
234 
235 	stats = this_cpu_ptr(dp->stats_percpu);
236 
237 	/* Look up flow. */
238 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
239 					 &n_mask_hit, &n_cache_hit);
240 	if (unlikely(!flow)) {
241 		struct dp_upcall_info upcall;
242 
243 		memset(&upcall, 0, sizeof(upcall));
244 		upcall.cmd = OVS_PACKET_CMD_MISS;
245 
246 		if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
247 			upcall.portid =
248 			    ovs_dp_get_upcall_portid(dp, smp_processor_id());
249 		else
250 			upcall.portid = ovs_vport_find_upcall_portid(p, skb);
251 
252 		upcall.mru = OVS_CB(skb)->mru;
253 		error = ovs_dp_upcall(dp, skb, key, &upcall, 0);
254 		if (unlikely(error))
255 			kfree_skb(skb);
256 		else
257 			consume_skb(skb);
258 		stats_counter = &stats->n_missed;
259 		goto out;
260 	}
261 
262 	ovs_flow_stats_update(flow, key->tp.flags, skb);
263 	sf_acts = rcu_dereference(flow->sf_acts);
264 	error = ovs_execute_actions(dp, skb, sf_acts, key);
265 	if (unlikely(error))
266 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
267 				    ovs_dp_name(dp), error);
268 
269 	stats_counter = &stats->n_hit;
270 
271 out:
272 	/* Update datapath statistics. */
273 	u64_stats_update_begin(&stats->syncp);
274 	(*stats_counter)++;
275 	stats->n_mask_hit += n_mask_hit;
276 	stats->n_cache_hit += n_cache_hit;
277 	u64_stats_update_end(&stats->syncp);
278 }
279 
280 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
281 		  const struct sw_flow_key *key,
282 		  const struct dp_upcall_info *upcall_info,
283 		  uint32_t cutlen)
284 {
285 	struct dp_stats_percpu *stats;
286 	int err;
287 
288 	if (trace_ovs_dp_upcall_enabled())
289 		trace_ovs_dp_upcall(dp, skb, key, upcall_info);
290 
291 	if (upcall_info->portid == 0) {
292 		err = -ENOTCONN;
293 		goto err;
294 	}
295 
296 	if (!skb_is_gso(skb))
297 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
298 	else
299 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
300 	if (err)
301 		goto err;
302 
303 	return 0;
304 
305 err:
306 	stats = this_cpu_ptr(dp->stats_percpu);
307 
308 	u64_stats_update_begin(&stats->syncp);
309 	stats->n_lost++;
310 	u64_stats_update_end(&stats->syncp);
311 
312 	return err;
313 }
314 
315 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
316 			     const struct sw_flow_key *key,
317 			     const struct dp_upcall_info *upcall_info,
318 			     uint32_t cutlen)
319 {
320 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
321 	struct sw_flow_key later_key;
322 	struct sk_buff *segs, *nskb;
323 	int err;
324 
325 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
326 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
327 	if (IS_ERR(segs))
328 		return PTR_ERR(segs);
329 	if (segs == NULL)
330 		return -EINVAL;
331 
332 	if (gso_type & SKB_GSO_UDP) {
333 		/* The initial flow key extracted by ovs_flow_key_extract()
334 		 * in this case is for a first fragment, so we need to
335 		 * properly mark later fragments.
336 		 */
337 		later_key = *key;
338 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
339 	}
340 
341 	/* Queue all of the segments. */
342 	skb_list_walk_safe(segs, skb, nskb) {
343 		if (gso_type & SKB_GSO_UDP && skb != segs)
344 			key = &later_key;
345 
346 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
347 		if (err)
348 			break;
349 
350 	}
351 
352 	/* Free all of the segments. */
353 	skb_list_walk_safe(segs, skb, nskb) {
354 		if (err)
355 			kfree_skb(skb);
356 		else
357 			consume_skb(skb);
358 	}
359 	return err;
360 }
361 
362 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
363 			      unsigned int hdrlen, int actions_attrlen)
364 {
365 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
366 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
367 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
368 		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
369 		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
370 
371 	/* OVS_PACKET_ATTR_USERDATA */
372 	if (upcall_info->userdata)
373 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
374 
375 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
376 	if (upcall_info->egress_tun_info)
377 		size += nla_total_size(ovs_tun_key_attr_size());
378 
379 	/* OVS_PACKET_ATTR_ACTIONS */
380 	if (upcall_info->actions_len)
381 		size += nla_total_size(actions_attrlen);
382 
383 	/* OVS_PACKET_ATTR_MRU */
384 	if (upcall_info->mru)
385 		size += nla_total_size(sizeof(upcall_info->mru));
386 
387 	return size;
388 }
389 
390 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
391 {
392 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
393 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
394 
395 		if (plen > 0)
396 			skb_put_zero(skb, plen);
397 	}
398 }
399 
400 static 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 				  uint32_t cutlen)
404 {
405 	struct ovs_header *upcall;
406 	struct sk_buff *nskb = NULL;
407 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
408 	struct nlattr *nla;
409 	size_t len;
410 	unsigned int hlen;
411 	int err, dp_ifindex;
412 	u64 hash;
413 
414 	dp_ifindex = get_dpifindex(dp);
415 	if (!dp_ifindex)
416 		return -ENODEV;
417 
418 	if (skb_vlan_tag_present(skb)) {
419 		nskb = skb_clone(skb, GFP_ATOMIC);
420 		if (!nskb)
421 			return -ENOMEM;
422 
423 		nskb = __vlan_hwaccel_push_inside(nskb);
424 		if (!nskb)
425 			return -ENOMEM;
426 
427 		skb = nskb;
428 	}
429 
430 	if (nla_attr_size(skb->len) > USHRT_MAX) {
431 		err = -EFBIG;
432 		goto out;
433 	}
434 
435 	/* Complete checksum if needed */
436 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
437 	    (err = skb_csum_hwoffload_help(skb, 0)))
438 		goto out;
439 
440 	/* Older versions of OVS user space enforce alignment of the last
441 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
442 	 * padding logic. Only perform zerocopy if padding is not required.
443 	 */
444 	if (dp->user_features & OVS_DP_F_UNALIGNED)
445 		hlen = skb_zerocopy_headlen(skb);
446 	else
447 		hlen = skb->len;
448 
449 	len = upcall_msg_size(upcall_info, hlen - cutlen,
450 			      OVS_CB(skb)->acts_origlen);
451 	user_skb = genlmsg_new(len, GFP_ATOMIC);
452 	if (!user_skb) {
453 		err = -ENOMEM;
454 		goto out;
455 	}
456 
457 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
458 			     0, upcall_info->cmd);
459 	if (!upcall) {
460 		err = -EINVAL;
461 		goto out;
462 	}
463 	upcall->dp_ifindex = dp_ifindex;
464 
465 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
466 	if (err)
467 		goto out;
468 
469 	if (upcall_info->userdata)
470 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
471 			  nla_len(upcall_info->userdata),
472 			  nla_data(upcall_info->userdata));
473 
474 	if (upcall_info->egress_tun_info) {
475 		nla = nla_nest_start_noflag(user_skb,
476 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
477 		if (!nla) {
478 			err = -EMSGSIZE;
479 			goto out;
480 		}
481 		err = ovs_nla_put_tunnel_info(user_skb,
482 					      upcall_info->egress_tun_info);
483 		if (err)
484 			goto out;
485 
486 		nla_nest_end(user_skb, nla);
487 	}
488 
489 	if (upcall_info->actions_len) {
490 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
491 		if (!nla) {
492 			err = -EMSGSIZE;
493 			goto out;
494 		}
495 		err = ovs_nla_put_actions(upcall_info->actions,
496 					  upcall_info->actions_len,
497 					  user_skb);
498 		if (!err)
499 			nla_nest_end(user_skb, nla);
500 		else
501 			nla_nest_cancel(user_skb, nla);
502 	}
503 
504 	/* Add OVS_PACKET_ATTR_MRU */
505 	if (upcall_info->mru &&
506 	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
507 		err = -ENOBUFS;
508 		goto out;
509 	}
510 
511 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
512 	if (cutlen > 0 &&
513 	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
514 		err = -ENOBUFS;
515 		goto out;
516 	}
517 
518 	/* Add OVS_PACKET_ATTR_HASH */
519 	hash = skb_get_hash_raw(skb);
520 	if (skb->sw_hash)
521 		hash |= OVS_PACKET_HASH_SW_BIT;
522 
523 	if (skb->l4_hash)
524 		hash |= OVS_PACKET_HASH_L4_BIT;
525 
526 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
527 		err = -ENOBUFS;
528 		goto out;
529 	}
530 
531 	/* Only reserve room for attribute header, packet data is added
532 	 * in skb_zerocopy() */
533 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
534 		err = -ENOBUFS;
535 		goto out;
536 	}
537 	nla->nla_len = nla_attr_size(skb->len - cutlen);
538 
539 	err = skb_zerocopy(user_skb, skb, skb->len - cutlen, hlen);
540 	if (err)
541 		goto out;
542 
543 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
544 	pad_packet(dp, user_skb);
545 
546 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
547 
548 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
549 	user_skb = NULL;
550 out:
551 	if (err)
552 		skb_tx_error(skb);
553 	kfree_skb(user_skb);
554 	kfree_skb(nskb);
555 	return err;
556 }
557 
558 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
559 {
560 	struct ovs_header *ovs_header = info->userhdr;
561 	struct net *net = sock_net(skb->sk);
562 	struct nlattr **a = info->attrs;
563 	struct sw_flow_actions *acts;
564 	struct sk_buff *packet;
565 	struct sw_flow *flow;
566 	struct sw_flow_actions *sf_acts;
567 	struct datapath *dp;
568 	struct vport *input_vport;
569 	u16 mru = 0;
570 	u64 hash;
571 	int len;
572 	int err;
573 	bool log = !a[OVS_PACKET_ATTR_PROBE];
574 
575 	err = -EINVAL;
576 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
577 	    !a[OVS_PACKET_ATTR_ACTIONS])
578 		goto err;
579 
580 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
581 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
582 	err = -ENOMEM;
583 	if (!packet)
584 		goto err;
585 	skb_reserve(packet, NET_IP_ALIGN);
586 
587 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
588 
589 	/* Set packet's mru */
590 	if (a[OVS_PACKET_ATTR_MRU]) {
591 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
592 		packet->ignore_df = 1;
593 	}
594 	OVS_CB(packet)->mru = mru;
595 
596 	if (a[OVS_PACKET_ATTR_HASH]) {
597 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
598 
599 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
600 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
601 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
602 	}
603 
604 	/* Build an sw_flow for sending this packet. */
605 	flow = ovs_flow_alloc();
606 	err = PTR_ERR(flow);
607 	if (IS_ERR(flow))
608 		goto err_kfree_skb;
609 
610 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
611 					     packet, &flow->key, log);
612 	if (err)
613 		goto err_flow_free;
614 
615 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
616 				   &flow->key, &acts, log);
617 	if (err)
618 		goto err_flow_free;
619 
620 	rcu_assign_pointer(flow->sf_acts, acts);
621 	packet->priority = flow->key.phy.priority;
622 	packet->mark = flow->key.phy.skb_mark;
623 
624 	rcu_read_lock();
625 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
626 	err = -ENODEV;
627 	if (!dp)
628 		goto err_unlock;
629 
630 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
631 	if (!input_vport)
632 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
633 
634 	if (!input_vport)
635 		goto err_unlock;
636 
637 	packet->dev = input_vport->dev;
638 	OVS_CB(packet)->input_vport = input_vport;
639 	sf_acts = rcu_dereference(flow->sf_acts);
640 
641 	local_bh_disable();
642 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
643 	local_bh_enable();
644 	rcu_read_unlock();
645 
646 	ovs_flow_free(flow, false);
647 	return err;
648 
649 err_unlock:
650 	rcu_read_unlock();
651 err_flow_free:
652 	ovs_flow_free(flow, false);
653 err_kfree_skb:
654 	kfree_skb(packet);
655 err:
656 	return err;
657 }
658 
659 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
660 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
661 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
662 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
663 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
664 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
665 	[OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
666 };
667 
668 static const struct genl_small_ops dp_packet_genl_ops[] = {
669 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
670 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
671 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
672 	  .doit = ovs_packet_cmd_execute
673 	}
674 };
675 
676 static struct genl_family dp_packet_genl_family __ro_after_init = {
677 	.hdrsize = sizeof(struct ovs_header),
678 	.name = OVS_PACKET_FAMILY,
679 	.version = OVS_PACKET_VERSION,
680 	.maxattr = OVS_PACKET_ATTR_MAX,
681 	.policy = packet_policy,
682 	.netnsok = true,
683 	.parallel_ops = true,
684 	.small_ops = dp_packet_genl_ops,
685 	.n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
686 	.module = THIS_MODULE,
687 };
688 
689 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
690 			 struct ovs_dp_megaflow_stats *mega_stats)
691 {
692 	int i;
693 
694 	memset(mega_stats, 0, sizeof(*mega_stats));
695 
696 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
697 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
698 
699 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
700 
701 	for_each_possible_cpu(i) {
702 		const struct dp_stats_percpu *percpu_stats;
703 		struct dp_stats_percpu local_stats;
704 		unsigned int start;
705 
706 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
707 
708 		do {
709 			start = u64_stats_fetch_begin_irq(&percpu_stats->syncp);
710 			local_stats = *percpu_stats;
711 		} while (u64_stats_fetch_retry_irq(&percpu_stats->syncp, start));
712 
713 		stats->n_hit += local_stats.n_hit;
714 		stats->n_missed += local_stats.n_missed;
715 		stats->n_lost += local_stats.n_lost;
716 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
717 		mega_stats->n_cache_hit += local_stats.n_cache_hit;
718 	}
719 }
720 
721 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
722 {
723 	return ovs_identifier_is_ufid(sfid) &&
724 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
725 }
726 
727 static bool should_fill_mask(uint32_t ufid_flags)
728 {
729 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
730 }
731 
732 static bool should_fill_actions(uint32_t ufid_flags)
733 {
734 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
735 }
736 
737 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
738 				    const struct sw_flow_id *sfid,
739 				    uint32_t ufid_flags)
740 {
741 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
742 
743 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
744 	 * see ovs_nla_put_identifier()
745 	 */
746 	if (sfid && ovs_identifier_is_ufid(sfid))
747 		len += nla_total_size(sfid->ufid_len);
748 	else
749 		len += nla_total_size(ovs_key_attr_size());
750 
751 	/* OVS_FLOW_ATTR_KEY */
752 	if (!sfid || should_fill_key(sfid, ufid_flags))
753 		len += nla_total_size(ovs_key_attr_size());
754 
755 	/* OVS_FLOW_ATTR_MASK */
756 	if (should_fill_mask(ufid_flags))
757 		len += nla_total_size(ovs_key_attr_size());
758 
759 	/* OVS_FLOW_ATTR_ACTIONS */
760 	if (should_fill_actions(ufid_flags))
761 		len += nla_total_size(acts->orig_len);
762 
763 	return len
764 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
765 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
766 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
767 }
768 
769 /* Called with ovs_mutex or RCU read lock. */
770 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
771 				   struct sk_buff *skb)
772 {
773 	struct ovs_flow_stats stats;
774 	__be16 tcp_flags;
775 	unsigned long used;
776 
777 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
778 
779 	if (used &&
780 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
781 			      OVS_FLOW_ATTR_PAD))
782 		return -EMSGSIZE;
783 
784 	if (stats.n_packets &&
785 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
786 			  sizeof(struct ovs_flow_stats), &stats,
787 			  OVS_FLOW_ATTR_PAD))
788 		return -EMSGSIZE;
789 
790 	if ((u8)ntohs(tcp_flags) &&
791 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
792 		return -EMSGSIZE;
793 
794 	return 0;
795 }
796 
797 /* Called with ovs_mutex or RCU read lock. */
798 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
799 				     struct sk_buff *skb, int skb_orig_len)
800 {
801 	struct nlattr *start;
802 	int err;
803 
804 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
805 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
806 	 * Netlink but individual action lists can be longer than
807 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
808 	 * The userspace caller can always fetch the actions separately if it
809 	 * really wants them.  (Most userspace callers in fact don't care.)
810 	 *
811 	 * This can only fail for dump operations because the skb is always
812 	 * properly sized for single flows.
813 	 */
814 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
815 	if (start) {
816 		const struct sw_flow_actions *sf_acts;
817 
818 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
819 		err = ovs_nla_put_actions(sf_acts->actions,
820 					  sf_acts->actions_len, skb);
821 
822 		if (!err)
823 			nla_nest_end(skb, start);
824 		else {
825 			if (skb_orig_len)
826 				return err;
827 
828 			nla_nest_cancel(skb, start);
829 		}
830 	} else if (skb_orig_len) {
831 		return -EMSGSIZE;
832 	}
833 
834 	return 0;
835 }
836 
837 /* Called with ovs_mutex or RCU read lock. */
838 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
839 				  struct sk_buff *skb, u32 portid,
840 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
841 {
842 	const int skb_orig_len = skb->len;
843 	struct ovs_header *ovs_header;
844 	int err;
845 
846 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
847 				 flags, cmd);
848 	if (!ovs_header)
849 		return -EMSGSIZE;
850 
851 	ovs_header->dp_ifindex = dp_ifindex;
852 
853 	err = ovs_nla_put_identifier(flow, skb);
854 	if (err)
855 		goto error;
856 
857 	if (should_fill_key(&flow->id, ufid_flags)) {
858 		err = ovs_nla_put_masked_key(flow, skb);
859 		if (err)
860 			goto error;
861 	}
862 
863 	if (should_fill_mask(ufid_flags)) {
864 		err = ovs_nla_put_mask(flow, skb);
865 		if (err)
866 			goto error;
867 	}
868 
869 	err = ovs_flow_cmd_fill_stats(flow, skb);
870 	if (err)
871 		goto error;
872 
873 	if (should_fill_actions(ufid_flags)) {
874 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
875 		if (err)
876 			goto error;
877 	}
878 
879 	genlmsg_end(skb, ovs_header);
880 	return 0;
881 
882 error:
883 	genlmsg_cancel(skb, ovs_header);
884 	return err;
885 }
886 
887 /* May not be called with RCU read lock. */
888 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
889 					       const struct sw_flow_id *sfid,
890 					       struct genl_info *info,
891 					       bool always,
892 					       uint32_t ufid_flags)
893 {
894 	struct sk_buff *skb;
895 	size_t len;
896 
897 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
898 		return NULL;
899 
900 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
901 	skb = genlmsg_new(len, GFP_KERNEL);
902 	if (!skb)
903 		return ERR_PTR(-ENOMEM);
904 
905 	return skb;
906 }
907 
908 /* Called with ovs_mutex. */
909 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
910 					       int dp_ifindex,
911 					       struct genl_info *info, u8 cmd,
912 					       bool always, u32 ufid_flags)
913 {
914 	struct sk_buff *skb;
915 	int retval;
916 
917 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
918 				      &flow->id, info, always, ufid_flags);
919 	if (IS_ERR_OR_NULL(skb))
920 		return skb;
921 
922 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
923 					info->snd_portid, info->snd_seq, 0,
924 					cmd, ufid_flags);
925 	if (WARN_ON_ONCE(retval < 0)) {
926 		kfree_skb(skb);
927 		skb = ERR_PTR(retval);
928 	}
929 	return skb;
930 }
931 
932 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
933 {
934 	struct net *net = sock_net(skb->sk);
935 	struct nlattr **a = info->attrs;
936 	struct ovs_header *ovs_header = info->userhdr;
937 	struct sw_flow *flow = NULL, *new_flow;
938 	struct sw_flow_mask mask;
939 	struct sk_buff *reply;
940 	struct datapath *dp;
941 	struct sw_flow_actions *acts;
942 	struct sw_flow_match match;
943 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
944 	int error;
945 	bool log = !a[OVS_FLOW_ATTR_PROBE];
946 
947 	/* Must have key and actions. */
948 	error = -EINVAL;
949 	if (!a[OVS_FLOW_ATTR_KEY]) {
950 		OVS_NLERR(log, "Flow key attr not present in new flow.");
951 		goto error;
952 	}
953 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
954 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
955 		goto error;
956 	}
957 
958 	/* Most of the time we need to allocate a new flow, do it before
959 	 * locking.
960 	 */
961 	new_flow = ovs_flow_alloc();
962 	if (IS_ERR(new_flow)) {
963 		error = PTR_ERR(new_flow);
964 		goto error;
965 	}
966 
967 	/* Extract key. */
968 	ovs_match_init(&match, &new_flow->key, false, &mask);
969 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
970 				  a[OVS_FLOW_ATTR_MASK], log);
971 	if (error)
972 		goto err_kfree_flow;
973 
974 	/* Extract flow identifier. */
975 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
976 				       &new_flow->key, log);
977 	if (error)
978 		goto err_kfree_flow;
979 
980 	/* unmasked key is needed to match when ufid is not used. */
981 	if (ovs_identifier_is_key(&new_flow->id))
982 		match.key = new_flow->id.unmasked_key;
983 
984 	ovs_flow_mask_key(&new_flow->key, &new_flow->key, true, &mask);
985 
986 	/* Validate actions. */
987 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
988 				     &new_flow->key, &acts, log);
989 	if (error) {
990 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
991 		goto err_kfree_flow;
992 	}
993 
994 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
995 					ufid_flags);
996 	if (IS_ERR(reply)) {
997 		error = PTR_ERR(reply);
998 		goto err_kfree_acts;
999 	}
1000 
1001 	ovs_lock();
1002 	dp = get_dp(net, ovs_header->dp_ifindex);
1003 	if (unlikely(!dp)) {
1004 		error = -ENODEV;
1005 		goto err_unlock_ovs;
1006 	}
1007 
1008 	/* Check if this is a duplicate flow */
1009 	if (ovs_identifier_is_ufid(&new_flow->id))
1010 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1011 	if (!flow)
1012 		flow = ovs_flow_tbl_lookup(&dp->table, &new_flow->key);
1013 	if (likely(!flow)) {
1014 		rcu_assign_pointer(new_flow->sf_acts, acts);
1015 
1016 		/* Put flow in bucket. */
1017 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1018 		if (unlikely(error)) {
1019 			acts = NULL;
1020 			goto err_unlock_ovs;
1021 		}
1022 
1023 		if (unlikely(reply)) {
1024 			error = ovs_flow_cmd_fill_info(new_flow,
1025 						       ovs_header->dp_ifindex,
1026 						       reply, info->snd_portid,
1027 						       info->snd_seq, 0,
1028 						       OVS_FLOW_CMD_NEW,
1029 						       ufid_flags);
1030 			BUG_ON(error < 0);
1031 		}
1032 		ovs_unlock();
1033 	} else {
1034 		struct sw_flow_actions *old_acts;
1035 
1036 		/* Bail out if we're not allowed to modify an existing flow.
1037 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1038 		 * because Generic Netlink treats the latter as a dump
1039 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1040 		 * gets fixed.
1041 		 */
1042 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1043 							 | NLM_F_EXCL))) {
1044 			error = -EEXIST;
1045 			goto err_unlock_ovs;
1046 		}
1047 		/* The flow identifier has to be the same for flow updates.
1048 		 * Look for any overlapping flow.
1049 		 */
1050 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1051 			if (ovs_identifier_is_key(&flow->id))
1052 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1053 								 &match);
1054 			else /* UFID matches but key is different */
1055 				flow = NULL;
1056 			if (!flow) {
1057 				error = -ENOENT;
1058 				goto err_unlock_ovs;
1059 			}
1060 		}
1061 		/* Update actions. */
1062 		old_acts = ovsl_dereference(flow->sf_acts);
1063 		rcu_assign_pointer(flow->sf_acts, acts);
1064 
1065 		if (unlikely(reply)) {
1066 			error = ovs_flow_cmd_fill_info(flow,
1067 						       ovs_header->dp_ifindex,
1068 						       reply, info->snd_portid,
1069 						       info->snd_seq, 0,
1070 						       OVS_FLOW_CMD_NEW,
1071 						       ufid_flags);
1072 			BUG_ON(error < 0);
1073 		}
1074 		ovs_unlock();
1075 
1076 		ovs_nla_free_flow_actions_rcu(old_acts);
1077 		ovs_flow_free(new_flow, false);
1078 	}
1079 
1080 	if (reply)
1081 		ovs_notify(&dp_flow_genl_family, reply, info);
1082 	return 0;
1083 
1084 err_unlock_ovs:
1085 	ovs_unlock();
1086 	kfree_skb(reply);
1087 err_kfree_acts:
1088 	ovs_nla_free_flow_actions(acts);
1089 err_kfree_flow:
1090 	ovs_flow_free(new_flow, false);
1091 error:
1092 	return error;
1093 }
1094 
1095 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1096 static noinline_for_stack
1097 struct sw_flow_actions *get_flow_actions(struct net *net,
1098 					 const struct nlattr *a,
1099 					 const struct sw_flow_key *key,
1100 					 const struct sw_flow_mask *mask,
1101 					 bool log)
1102 {
1103 	struct sw_flow_actions *acts;
1104 	struct sw_flow_key masked_key;
1105 	int error;
1106 
1107 	ovs_flow_mask_key(&masked_key, key, true, mask);
1108 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1109 	if (error) {
1110 		OVS_NLERR(log,
1111 			  "Actions may not be safe on all matching packets");
1112 		return ERR_PTR(error);
1113 	}
1114 
1115 	return acts;
1116 }
1117 
1118 /* Factor out match-init and action-copy to avoid
1119  * "Wframe-larger-than=1024" warning. Because mask is only
1120  * used to get actions, we new a function to save some
1121  * stack space.
1122  *
1123  * If there are not key and action attrs, we return 0
1124  * directly. In the case, the caller will also not use the
1125  * match as before. If there is action attr, we try to get
1126  * actions and save them to *acts. Before returning from
1127  * the function, we reset the match->mask pointer. Because
1128  * we should not to return match object with dangling reference
1129  * to mask.
1130  * */
1131 static noinline_for_stack int
1132 ovs_nla_init_match_and_action(struct net *net,
1133 			      struct sw_flow_match *match,
1134 			      struct sw_flow_key *key,
1135 			      struct nlattr **a,
1136 			      struct sw_flow_actions **acts,
1137 			      bool log)
1138 {
1139 	struct sw_flow_mask mask;
1140 	int error = 0;
1141 
1142 	if (a[OVS_FLOW_ATTR_KEY]) {
1143 		ovs_match_init(match, key, true, &mask);
1144 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1145 					  a[OVS_FLOW_ATTR_MASK], log);
1146 		if (error)
1147 			goto error;
1148 	}
1149 
1150 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1151 		if (!a[OVS_FLOW_ATTR_KEY]) {
1152 			OVS_NLERR(log,
1153 				  "Flow key attribute not present in set flow.");
1154 			error = -EINVAL;
1155 			goto error;
1156 		}
1157 
1158 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1159 					 &mask, log);
1160 		if (IS_ERR(*acts)) {
1161 			error = PTR_ERR(*acts);
1162 			goto error;
1163 		}
1164 	}
1165 
1166 	/* On success, error is 0. */
1167 error:
1168 	match->mask = NULL;
1169 	return error;
1170 }
1171 
1172 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1173 {
1174 	struct net *net = sock_net(skb->sk);
1175 	struct nlattr **a = info->attrs;
1176 	struct ovs_header *ovs_header = info->userhdr;
1177 	struct sw_flow_key key;
1178 	struct sw_flow *flow;
1179 	struct sk_buff *reply = NULL;
1180 	struct datapath *dp;
1181 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1182 	struct sw_flow_match match;
1183 	struct sw_flow_id sfid;
1184 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1185 	int error = 0;
1186 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1187 	bool ufid_present;
1188 
1189 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1190 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1191 		OVS_NLERR(log,
1192 			  "Flow set message rejected, Key attribute missing.");
1193 		return -EINVAL;
1194 	}
1195 
1196 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1197 					      &acts, log);
1198 	if (error)
1199 		goto error;
1200 
1201 	if (acts) {
1202 		/* Can allocate before locking if have acts. */
1203 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1204 						ufid_flags);
1205 		if (IS_ERR(reply)) {
1206 			error = PTR_ERR(reply);
1207 			goto err_kfree_acts;
1208 		}
1209 	}
1210 
1211 	ovs_lock();
1212 	dp = get_dp(net, ovs_header->dp_ifindex);
1213 	if (unlikely(!dp)) {
1214 		error = -ENODEV;
1215 		goto err_unlock_ovs;
1216 	}
1217 	/* Check that the flow exists. */
1218 	if (ufid_present)
1219 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1220 	else
1221 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1222 	if (unlikely(!flow)) {
1223 		error = -ENOENT;
1224 		goto err_unlock_ovs;
1225 	}
1226 
1227 	/* Update actions, if present. */
1228 	if (likely(acts)) {
1229 		old_acts = ovsl_dereference(flow->sf_acts);
1230 		rcu_assign_pointer(flow->sf_acts, acts);
1231 
1232 		if (unlikely(reply)) {
1233 			error = ovs_flow_cmd_fill_info(flow,
1234 						       ovs_header->dp_ifindex,
1235 						       reply, info->snd_portid,
1236 						       info->snd_seq, 0,
1237 						       OVS_FLOW_CMD_SET,
1238 						       ufid_flags);
1239 			BUG_ON(error < 0);
1240 		}
1241 	} else {
1242 		/* Could not alloc without acts before locking. */
1243 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1244 						info, OVS_FLOW_CMD_SET, false,
1245 						ufid_flags);
1246 
1247 		if (IS_ERR(reply)) {
1248 			error = PTR_ERR(reply);
1249 			goto err_unlock_ovs;
1250 		}
1251 	}
1252 
1253 	/* Clear stats. */
1254 	if (a[OVS_FLOW_ATTR_CLEAR])
1255 		ovs_flow_stats_clear(flow);
1256 	ovs_unlock();
1257 
1258 	if (reply)
1259 		ovs_notify(&dp_flow_genl_family, reply, info);
1260 	if (old_acts)
1261 		ovs_nla_free_flow_actions_rcu(old_acts);
1262 
1263 	return 0;
1264 
1265 err_unlock_ovs:
1266 	ovs_unlock();
1267 	kfree_skb(reply);
1268 err_kfree_acts:
1269 	ovs_nla_free_flow_actions(acts);
1270 error:
1271 	return error;
1272 }
1273 
1274 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1275 {
1276 	struct nlattr **a = info->attrs;
1277 	struct ovs_header *ovs_header = info->userhdr;
1278 	struct net *net = sock_net(skb->sk);
1279 	struct sw_flow_key key;
1280 	struct sk_buff *reply;
1281 	struct sw_flow *flow;
1282 	struct datapath *dp;
1283 	struct sw_flow_match match;
1284 	struct sw_flow_id ufid;
1285 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1286 	int err = 0;
1287 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1288 	bool ufid_present;
1289 
1290 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1291 	if (a[OVS_FLOW_ATTR_KEY]) {
1292 		ovs_match_init(&match, &key, true, NULL);
1293 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1294 					log);
1295 	} else if (!ufid_present) {
1296 		OVS_NLERR(log,
1297 			  "Flow get message rejected, Key attribute missing.");
1298 		err = -EINVAL;
1299 	}
1300 	if (err)
1301 		return err;
1302 
1303 	ovs_lock();
1304 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1305 	if (!dp) {
1306 		err = -ENODEV;
1307 		goto unlock;
1308 	}
1309 
1310 	if (ufid_present)
1311 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1312 	else
1313 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1314 	if (!flow) {
1315 		err = -ENOENT;
1316 		goto unlock;
1317 	}
1318 
1319 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1320 					OVS_FLOW_CMD_GET, true, ufid_flags);
1321 	if (IS_ERR(reply)) {
1322 		err = PTR_ERR(reply);
1323 		goto unlock;
1324 	}
1325 
1326 	ovs_unlock();
1327 	return genlmsg_reply(reply, info);
1328 unlock:
1329 	ovs_unlock();
1330 	return err;
1331 }
1332 
1333 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1334 {
1335 	struct nlattr **a = info->attrs;
1336 	struct ovs_header *ovs_header = info->userhdr;
1337 	struct net *net = sock_net(skb->sk);
1338 	struct sw_flow_key key;
1339 	struct sk_buff *reply;
1340 	struct sw_flow *flow = NULL;
1341 	struct datapath *dp;
1342 	struct sw_flow_match match;
1343 	struct sw_flow_id ufid;
1344 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1345 	int err;
1346 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1347 	bool ufid_present;
1348 
1349 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1350 	if (a[OVS_FLOW_ATTR_KEY]) {
1351 		ovs_match_init(&match, &key, true, NULL);
1352 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1353 					NULL, log);
1354 		if (unlikely(err))
1355 			return err;
1356 	}
1357 
1358 	ovs_lock();
1359 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1360 	if (unlikely(!dp)) {
1361 		err = -ENODEV;
1362 		goto unlock;
1363 	}
1364 
1365 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1366 		err = ovs_flow_tbl_flush(&dp->table);
1367 		goto unlock;
1368 	}
1369 
1370 	if (ufid_present)
1371 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1372 	else
1373 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1374 	if (unlikely(!flow)) {
1375 		err = -ENOENT;
1376 		goto unlock;
1377 	}
1378 
1379 	ovs_flow_tbl_remove(&dp->table, flow);
1380 	ovs_unlock();
1381 
1382 	reply = ovs_flow_cmd_alloc_info((const struct sw_flow_actions __force *) flow->sf_acts,
1383 					&flow->id, info, false, ufid_flags);
1384 	if (likely(reply)) {
1385 		if (!IS_ERR(reply)) {
1386 			rcu_read_lock();	/*To keep RCU checker happy. */
1387 			err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1388 						     reply, info->snd_portid,
1389 						     info->snd_seq, 0,
1390 						     OVS_FLOW_CMD_DEL,
1391 						     ufid_flags);
1392 			rcu_read_unlock();
1393 			if (WARN_ON_ONCE(err < 0)) {
1394 				kfree_skb(reply);
1395 				goto out_free;
1396 			}
1397 
1398 			ovs_notify(&dp_flow_genl_family, reply, info);
1399 		} else {
1400 			netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1401 					PTR_ERR(reply));
1402 		}
1403 	}
1404 
1405 out_free:
1406 	ovs_flow_free(flow, true);
1407 	return 0;
1408 unlock:
1409 	ovs_unlock();
1410 	return err;
1411 }
1412 
1413 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1414 {
1415 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1416 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1417 	struct table_instance *ti;
1418 	struct datapath *dp;
1419 	u32 ufid_flags;
1420 	int err;
1421 
1422 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1423 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1424 	if (err)
1425 		return err;
1426 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1427 
1428 	rcu_read_lock();
1429 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1430 	if (!dp) {
1431 		rcu_read_unlock();
1432 		return -ENODEV;
1433 	}
1434 
1435 	ti = rcu_dereference(dp->table.ti);
1436 	for (;;) {
1437 		struct sw_flow *flow;
1438 		u32 bucket, obj;
1439 
1440 		bucket = cb->args[0];
1441 		obj = cb->args[1];
1442 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1443 		if (!flow)
1444 			break;
1445 
1446 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1447 					   NETLINK_CB(cb->skb).portid,
1448 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1449 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1450 			break;
1451 
1452 		cb->args[0] = bucket;
1453 		cb->args[1] = obj;
1454 	}
1455 	rcu_read_unlock();
1456 	return skb->len;
1457 }
1458 
1459 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1460 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1461 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1462 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1463 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1464 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1465 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1466 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1467 };
1468 
1469 static const struct genl_small_ops dp_flow_genl_ops[] = {
1470 	{ .cmd = OVS_FLOW_CMD_NEW,
1471 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1472 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1473 	  .doit = ovs_flow_cmd_new
1474 	},
1475 	{ .cmd = OVS_FLOW_CMD_DEL,
1476 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1477 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1478 	  .doit = ovs_flow_cmd_del
1479 	},
1480 	{ .cmd = OVS_FLOW_CMD_GET,
1481 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1482 	  .flags = 0,		    /* OK for unprivileged users. */
1483 	  .doit = ovs_flow_cmd_get,
1484 	  .dumpit = ovs_flow_cmd_dump
1485 	},
1486 	{ .cmd = OVS_FLOW_CMD_SET,
1487 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1488 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1489 	  .doit = ovs_flow_cmd_set,
1490 	},
1491 };
1492 
1493 static struct genl_family dp_flow_genl_family __ro_after_init = {
1494 	.hdrsize = sizeof(struct ovs_header),
1495 	.name = OVS_FLOW_FAMILY,
1496 	.version = OVS_FLOW_VERSION,
1497 	.maxattr = OVS_FLOW_ATTR_MAX,
1498 	.policy = flow_policy,
1499 	.netnsok = true,
1500 	.parallel_ops = true,
1501 	.small_ops = dp_flow_genl_ops,
1502 	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1503 	.mcgrps = &ovs_dp_flow_multicast_group,
1504 	.n_mcgrps = 1,
1505 	.module = THIS_MODULE,
1506 };
1507 
1508 static size_t ovs_dp_cmd_msg_size(void)
1509 {
1510 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1511 
1512 	msgsize += nla_total_size(IFNAMSIZ);
1513 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1514 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1515 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1516 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1517 
1518 	return msgsize;
1519 }
1520 
1521 /* Called with ovs_mutex. */
1522 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1523 				u32 portid, u32 seq, u32 flags, u8 cmd)
1524 {
1525 	struct ovs_header *ovs_header;
1526 	struct ovs_dp_stats dp_stats;
1527 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1528 	int err;
1529 
1530 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1531 				 flags, cmd);
1532 	if (!ovs_header)
1533 		goto error;
1534 
1535 	ovs_header->dp_ifindex = get_dpifindex(dp);
1536 
1537 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1538 	if (err)
1539 		goto nla_put_failure;
1540 
1541 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1542 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1543 			  &dp_stats, OVS_DP_ATTR_PAD))
1544 		goto nla_put_failure;
1545 
1546 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1547 			  sizeof(struct ovs_dp_megaflow_stats),
1548 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1549 		goto nla_put_failure;
1550 
1551 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1552 		goto nla_put_failure;
1553 
1554 	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1555 			ovs_flow_tbl_masks_cache_size(&dp->table)))
1556 		goto nla_put_failure;
1557 
1558 	genlmsg_end(skb, ovs_header);
1559 	return 0;
1560 
1561 nla_put_failure:
1562 	genlmsg_cancel(skb, ovs_header);
1563 error:
1564 	return -EMSGSIZE;
1565 }
1566 
1567 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1568 {
1569 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1570 }
1571 
1572 /* Called with rcu_read_lock or ovs_mutex. */
1573 static struct datapath *lookup_datapath(struct net *net,
1574 					const struct ovs_header *ovs_header,
1575 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1576 {
1577 	struct datapath *dp;
1578 
1579 	if (!a[OVS_DP_ATTR_NAME])
1580 		dp = get_dp(net, ovs_header->dp_ifindex);
1581 	else {
1582 		struct vport *vport;
1583 
1584 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1585 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1586 	}
1587 	return dp ? dp : ERR_PTR(-ENODEV);
1588 }
1589 
1590 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1591 				       struct genl_info *info)
1592 {
1593 	struct datapath *dp;
1594 
1595 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr,
1596 			     info->attrs);
1597 	if (IS_ERR(dp))
1598 		return;
1599 
1600 	WARN(dp->user_features, "Dropping previously announced user features\n");
1601 	dp->user_features = 0;
1602 }
1603 
1604 DEFINE_STATIC_KEY_FALSE(tc_recirc_sharing_support);
1605 
1606 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1607 			      const struct nlattr *ids)
1608 {
1609 	struct dp_nlsk_pids *old, *dp_nlsk_pids;
1610 
1611 	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1612 		return -EINVAL;
1613 
1614 	old = ovsl_dereference(dp->upcall_portids);
1615 
1616 	dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1617 			       GFP_KERNEL);
1618 	if (!dp_nlsk_pids)
1619 		return -ENOMEM;
1620 
1621 	dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1622 	nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1623 
1624 	rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1625 
1626 	kfree_rcu(old, rcu);
1627 
1628 	return 0;
1629 }
1630 
1631 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1632 {
1633 	struct dp_nlsk_pids *dp_nlsk_pids;
1634 
1635 	dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1636 
1637 	if (dp_nlsk_pids) {
1638 		if (cpu_id < dp_nlsk_pids->n_pids) {
1639 			return dp_nlsk_pids->pids[cpu_id];
1640 		} else if (dp_nlsk_pids->n_pids > 0 &&
1641 			   cpu_id >= dp_nlsk_pids->n_pids) {
1642 			/* If the number of netlink PIDs is mismatched with
1643 			 * the number of CPUs as seen by the kernel, log this
1644 			 * and send the upcall to an arbitrary socket (0) in
1645 			 * order to not drop packets
1646 			 */
1647 			pr_info_ratelimited("cpu_id mismatch with handler threads");
1648 			return dp_nlsk_pids->pids[cpu_id %
1649 						  dp_nlsk_pids->n_pids];
1650 		} else {
1651 			return 0;
1652 		}
1653 	} else {
1654 		return 0;
1655 	}
1656 }
1657 
1658 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1659 {
1660 	u32 user_features = 0;
1661 	int err;
1662 
1663 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1664 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1665 
1666 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1667 				      OVS_DP_F_UNALIGNED |
1668 				      OVS_DP_F_TC_RECIRC_SHARING |
1669 				      OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1670 			return -EOPNOTSUPP;
1671 
1672 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1673 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1674 			return -EOPNOTSUPP;
1675 #endif
1676 	}
1677 
1678 	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1679 		int err;
1680 		u32 cache_size;
1681 
1682 		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1683 		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1684 		if (err)
1685 			return err;
1686 	}
1687 
1688 	dp->user_features = user_features;
1689 
1690 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1691 	    a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1692 		/* Upcall Netlink Port IDs have been updated */
1693 		err = ovs_dp_set_upcall_portids(dp,
1694 						a[OVS_DP_ATTR_PER_CPU_PIDS]);
1695 		if (err)
1696 			return err;
1697 	}
1698 
1699 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1700 		static_branch_enable(&tc_recirc_sharing_support);
1701 	else
1702 		static_branch_disable(&tc_recirc_sharing_support);
1703 
1704 	return 0;
1705 }
1706 
1707 static int ovs_dp_stats_init(struct datapath *dp)
1708 {
1709 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1710 	if (!dp->stats_percpu)
1711 		return -ENOMEM;
1712 
1713 	return 0;
1714 }
1715 
1716 static int ovs_dp_vport_init(struct datapath *dp)
1717 {
1718 	int i;
1719 
1720 	dp->ports = kmalloc_array(DP_VPORT_HASH_BUCKETS,
1721 				  sizeof(struct hlist_head),
1722 				  GFP_KERNEL);
1723 	if (!dp->ports)
1724 		return -ENOMEM;
1725 
1726 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1727 		INIT_HLIST_HEAD(&dp->ports[i]);
1728 
1729 	return 0;
1730 }
1731 
1732 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1733 {
1734 	struct nlattr **a = info->attrs;
1735 	struct vport_parms parms;
1736 	struct sk_buff *reply;
1737 	struct datapath *dp;
1738 	struct vport *vport;
1739 	struct ovs_net *ovs_net;
1740 	int err;
1741 
1742 	err = -EINVAL;
1743 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1744 		goto err;
1745 
1746 	reply = ovs_dp_cmd_alloc_info();
1747 	if (!reply)
1748 		return -ENOMEM;
1749 
1750 	err = -ENOMEM;
1751 	dp = kzalloc(sizeof(*dp), GFP_KERNEL);
1752 	if (dp == NULL)
1753 		goto err_destroy_reply;
1754 
1755 	ovs_dp_set_net(dp, sock_net(skb->sk));
1756 
1757 	/* Allocate table. */
1758 	err = ovs_flow_tbl_init(&dp->table);
1759 	if (err)
1760 		goto err_destroy_dp;
1761 
1762 	err = ovs_dp_stats_init(dp);
1763 	if (err)
1764 		goto err_destroy_table;
1765 
1766 	err = ovs_dp_vport_init(dp);
1767 	if (err)
1768 		goto err_destroy_stats;
1769 
1770 	err = ovs_meters_init(dp);
1771 	if (err)
1772 		goto err_destroy_ports;
1773 
1774 	/* Set up our datapath device. */
1775 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1776 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1777 	parms.options = NULL;
1778 	parms.dp = dp;
1779 	parms.port_no = OVSP_LOCAL;
1780 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1781 
1782 	/* So far only local changes have been made, now need the lock. */
1783 	ovs_lock();
1784 
1785 	err = ovs_dp_change(dp, a);
1786 	if (err)
1787 		goto err_unlock_and_destroy_meters;
1788 
1789 	vport = new_vport(&parms);
1790 	if (IS_ERR(vport)) {
1791 		err = PTR_ERR(vport);
1792 		if (err == -EBUSY)
1793 			err = -EEXIST;
1794 
1795 		if (err == -EEXIST) {
1796 			/* An outdated user space instance that does not understand
1797 			 * the concept of user_features has attempted to create a new
1798 			 * datapath and is likely to reuse it. Drop all user features.
1799 			 */
1800 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1801 				ovs_dp_reset_user_features(skb, info);
1802 		}
1803 
1804 		goto err_unlock_and_destroy_meters;
1805 	}
1806 
1807 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1808 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1809 	BUG_ON(err < 0);
1810 
1811 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1812 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1813 
1814 	ovs_unlock();
1815 
1816 	ovs_notify(&dp_datapath_genl_family, reply, info);
1817 	return 0;
1818 
1819 err_unlock_and_destroy_meters:
1820 	ovs_unlock();
1821 	ovs_meters_exit(dp);
1822 err_destroy_ports:
1823 	kfree(dp->ports);
1824 err_destroy_stats:
1825 	free_percpu(dp->stats_percpu);
1826 err_destroy_table:
1827 	ovs_flow_tbl_destroy(&dp->table);
1828 err_destroy_dp:
1829 	kfree(dp);
1830 err_destroy_reply:
1831 	kfree_skb(reply);
1832 err:
1833 	return err;
1834 }
1835 
1836 /* Called with ovs_mutex. */
1837 static void __dp_destroy(struct datapath *dp)
1838 {
1839 	struct flow_table *table = &dp->table;
1840 	int i;
1841 
1842 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1843 		struct vport *vport;
1844 		struct hlist_node *n;
1845 
1846 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1847 			if (vport->port_no != OVSP_LOCAL)
1848 				ovs_dp_detach_port(vport);
1849 	}
1850 
1851 	list_del_rcu(&dp->list_node);
1852 
1853 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1854 	 * all ports in datapath are destroyed first before freeing datapath.
1855 	 */
1856 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1857 
1858 	/* Flush sw_flow in the tables. RCU cb only releases resource
1859 	 * such as dp, ports and tables. That may avoid some issues
1860 	 * such as RCU usage warning.
1861 	 */
1862 	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1863 				  ovsl_dereference(table->ufid_ti));
1864 
1865 	/* RCU destroy the ports, meters and flow tables. */
1866 	call_rcu(&dp->rcu, destroy_dp_rcu);
1867 }
1868 
1869 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1870 {
1871 	struct sk_buff *reply;
1872 	struct datapath *dp;
1873 	int err;
1874 
1875 	reply = ovs_dp_cmd_alloc_info();
1876 	if (!reply)
1877 		return -ENOMEM;
1878 
1879 	ovs_lock();
1880 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1881 	err = PTR_ERR(dp);
1882 	if (IS_ERR(dp))
1883 		goto err_unlock_free;
1884 
1885 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1886 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
1887 	BUG_ON(err < 0);
1888 
1889 	__dp_destroy(dp);
1890 	ovs_unlock();
1891 
1892 	ovs_notify(&dp_datapath_genl_family, reply, info);
1893 
1894 	return 0;
1895 
1896 err_unlock_free:
1897 	ovs_unlock();
1898 	kfree_skb(reply);
1899 	return err;
1900 }
1901 
1902 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
1903 {
1904 	struct sk_buff *reply;
1905 	struct datapath *dp;
1906 	int err;
1907 
1908 	reply = ovs_dp_cmd_alloc_info();
1909 	if (!reply)
1910 		return -ENOMEM;
1911 
1912 	ovs_lock();
1913 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1914 	err = PTR_ERR(dp);
1915 	if (IS_ERR(dp))
1916 		goto err_unlock_free;
1917 
1918 	err = ovs_dp_change(dp, info->attrs);
1919 	if (err)
1920 		goto err_unlock_free;
1921 
1922 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1923 				   info->snd_seq, 0, OVS_DP_CMD_SET);
1924 	BUG_ON(err < 0);
1925 
1926 	ovs_unlock();
1927 	ovs_notify(&dp_datapath_genl_family, reply, info);
1928 
1929 	return 0;
1930 
1931 err_unlock_free:
1932 	ovs_unlock();
1933 	kfree_skb(reply);
1934 	return err;
1935 }
1936 
1937 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
1938 {
1939 	struct sk_buff *reply;
1940 	struct datapath *dp;
1941 	int err;
1942 
1943 	reply = ovs_dp_cmd_alloc_info();
1944 	if (!reply)
1945 		return -ENOMEM;
1946 
1947 	ovs_lock();
1948 	dp = lookup_datapath(sock_net(skb->sk), info->userhdr, info->attrs);
1949 	if (IS_ERR(dp)) {
1950 		err = PTR_ERR(dp);
1951 		goto err_unlock_free;
1952 	}
1953 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1954 				   info->snd_seq, 0, OVS_DP_CMD_GET);
1955 	BUG_ON(err < 0);
1956 	ovs_unlock();
1957 
1958 	return genlmsg_reply(reply, info);
1959 
1960 err_unlock_free:
1961 	ovs_unlock();
1962 	kfree_skb(reply);
1963 	return err;
1964 }
1965 
1966 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1967 {
1968 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1969 	struct datapath *dp;
1970 	int skip = cb->args[0];
1971 	int i = 0;
1972 
1973 	ovs_lock();
1974 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
1975 		if (i >= skip &&
1976 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
1977 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
1978 					 OVS_DP_CMD_GET) < 0)
1979 			break;
1980 		i++;
1981 	}
1982 	ovs_unlock();
1983 
1984 	cb->args[0] = i;
1985 
1986 	return skb->len;
1987 }
1988 
1989 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
1990 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
1991 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
1992 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
1993 	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
1994 		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
1995 };
1996 
1997 static const struct genl_small_ops dp_datapath_genl_ops[] = {
1998 	{ .cmd = OVS_DP_CMD_NEW,
1999 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2000 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2001 	  .doit = ovs_dp_cmd_new
2002 	},
2003 	{ .cmd = OVS_DP_CMD_DEL,
2004 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2005 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2006 	  .doit = ovs_dp_cmd_del
2007 	},
2008 	{ .cmd = OVS_DP_CMD_GET,
2009 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2010 	  .flags = 0,		    /* OK for unprivileged users. */
2011 	  .doit = ovs_dp_cmd_get,
2012 	  .dumpit = ovs_dp_cmd_dump
2013 	},
2014 	{ .cmd = OVS_DP_CMD_SET,
2015 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2016 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2017 	  .doit = ovs_dp_cmd_set,
2018 	},
2019 };
2020 
2021 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2022 	.hdrsize = sizeof(struct ovs_header),
2023 	.name = OVS_DATAPATH_FAMILY,
2024 	.version = OVS_DATAPATH_VERSION,
2025 	.maxattr = OVS_DP_ATTR_MAX,
2026 	.policy = datapath_policy,
2027 	.netnsok = true,
2028 	.parallel_ops = true,
2029 	.small_ops = dp_datapath_genl_ops,
2030 	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2031 	.mcgrps = &ovs_dp_datapath_multicast_group,
2032 	.n_mcgrps = 1,
2033 	.module = THIS_MODULE,
2034 };
2035 
2036 /* Called with ovs_mutex or RCU read lock. */
2037 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2038 				   struct net *net, u32 portid, u32 seq,
2039 				   u32 flags, u8 cmd, gfp_t gfp)
2040 {
2041 	struct ovs_header *ovs_header;
2042 	struct ovs_vport_stats vport_stats;
2043 	int err;
2044 
2045 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2046 				 flags, cmd);
2047 	if (!ovs_header)
2048 		return -EMSGSIZE;
2049 
2050 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2051 
2052 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2053 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2054 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2055 			   ovs_vport_name(vport)) ||
2056 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2057 		goto nla_put_failure;
2058 
2059 	if (!net_eq(net, dev_net(vport->dev))) {
2060 		int id = peernet2id_alloc(net, dev_net(vport->dev), gfp);
2061 
2062 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2063 			goto nla_put_failure;
2064 	}
2065 
2066 	ovs_vport_get_stats(vport, &vport_stats);
2067 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2068 			  sizeof(struct ovs_vport_stats), &vport_stats,
2069 			  OVS_VPORT_ATTR_PAD))
2070 		goto nla_put_failure;
2071 
2072 	if (ovs_vport_get_upcall_portids(vport, skb))
2073 		goto nla_put_failure;
2074 
2075 	err = ovs_vport_get_options(vport, skb);
2076 	if (err == -EMSGSIZE)
2077 		goto error;
2078 
2079 	genlmsg_end(skb, ovs_header);
2080 	return 0;
2081 
2082 nla_put_failure:
2083 	err = -EMSGSIZE;
2084 error:
2085 	genlmsg_cancel(skb, ovs_header);
2086 	return err;
2087 }
2088 
2089 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2090 {
2091 	return nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2092 }
2093 
2094 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2095 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2096 					 u32 portid, u32 seq, u8 cmd)
2097 {
2098 	struct sk_buff *skb;
2099 	int retval;
2100 
2101 	skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2102 	if (!skb)
2103 		return ERR_PTR(-ENOMEM);
2104 
2105 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2106 					 GFP_KERNEL);
2107 	BUG_ON(retval < 0);
2108 
2109 	return skb;
2110 }
2111 
2112 /* Called with ovs_mutex or RCU read lock. */
2113 static struct vport *lookup_vport(struct net *net,
2114 				  const struct ovs_header *ovs_header,
2115 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2116 {
2117 	struct datapath *dp;
2118 	struct vport *vport;
2119 
2120 	if (a[OVS_VPORT_ATTR_IFINDEX])
2121 		return ERR_PTR(-EOPNOTSUPP);
2122 	if (a[OVS_VPORT_ATTR_NAME]) {
2123 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2124 		if (!vport)
2125 			return ERR_PTR(-ENODEV);
2126 		if (ovs_header->dp_ifindex &&
2127 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2128 			return ERR_PTR(-ENODEV);
2129 		return vport;
2130 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2131 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2132 
2133 		if (port_no >= DP_MAX_PORTS)
2134 			return ERR_PTR(-EFBIG);
2135 
2136 		dp = get_dp(net, ovs_header->dp_ifindex);
2137 		if (!dp)
2138 			return ERR_PTR(-ENODEV);
2139 
2140 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2141 		if (!vport)
2142 			return ERR_PTR(-ENODEV);
2143 		return vport;
2144 	} else
2145 		return ERR_PTR(-EINVAL);
2146 
2147 }
2148 
2149 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2150 {
2151 	unsigned int dev_headroom, max_headroom = 0;
2152 	struct net_device *dev;
2153 	struct vport *vport;
2154 	int i;
2155 
2156 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2157 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2158 					 lockdep_ovsl_is_held()) {
2159 			dev = vport->dev;
2160 			dev_headroom = netdev_get_fwd_headroom(dev);
2161 			if (dev_headroom > max_headroom)
2162 				max_headroom = dev_headroom;
2163 		}
2164 	}
2165 
2166 	return max_headroom;
2167 }
2168 
2169 /* Called with ovs_mutex */
2170 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2171 {
2172 	struct vport *vport;
2173 	int i;
2174 
2175 	dp->max_headroom = new_headroom;
2176 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2177 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2178 					 lockdep_ovsl_is_held())
2179 			netdev_set_rx_headroom(vport->dev, new_headroom);
2180 	}
2181 }
2182 
2183 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2184 {
2185 	struct nlattr **a = info->attrs;
2186 	struct ovs_header *ovs_header = info->userhdr;
2187 	struct vport_parms parms;
2188 	struct sk_buff *reply;
2189 	struct vport *vport;
2190 	struct datapath *dp;
2191 	unsigned int new_headroom;
2192 	u32 port_no;
2193 	int err;
2194 
2195 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2196 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2197 		return -EINVAL;
2198 	if (a[OVS_VPORT_ATTR_IFINDEX])
2199 		return -EOPNOTSUPP;
2200 
2201 	port_no = a[OVS_VPORT_ATTR_PORT_NO]
2202 		? nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]) : 0;
2203 	if (port_no >= DP_MAX_PORTS)
2204 		return -EFBIG;
2205 
2206 	reply = ovs_vport_cmd_alloc_info();
2207 	if (!reply)
2208 		return -ENOMEM;
2209 
2210 	ovs_lock();
2211 restart:
2212 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2213 	err = -ENODEV;
2214 	if (!dp)
2215 		goto exit_unlock_free;
2216 
2217 	if (port_no) {
2218 		vport = ovs_vport_ovsl(dp, port_no);
2219 		err = -EBUSY;
2220 		if (vport)
2221 			goto exit_unlock_free;
2222 	} else {
2223 		for (port_no = 1; ; port_no++) {
2224 			if (port_no >= DP_MAX_PORTS) {
2225 				err = -EFBIG;
2226 				goto exit_unlock_free;
2227 			}
2228 			vport = ovs_vport_ovsl(dp, port_no);
2229 			if (!vport)
2230 				break;
2231 		}
2232 	}
2233 
2234 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2235 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2236 	parms.options = a[OVS_VPORT_ATTR_OPTIONS];
2237 	parms.dp = dp;
2238 	parms.port_no = port_no;
2239 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2240 
2241 	vport = new_vport(&parms);
2242 	err = PTR_ERR(vport);
2243 	if (IS_ERR(vport)) {
2244 		if (err == -EAGAIN)
2245 			goto restart;
2246 		goto exit_unlock_free;
2247 	}
2248 
2249 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2250 				      info->snd_portid, info->snd_seq, 0,
2251 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2252 
2253 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2254 
2255 	if (new_headroom > dp->max_headroom)
2256 		ovs_update_headroom(dp, new_headroom);
2257 	else
2258 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2259 
2260 	BUG_ON(err < 0);
2261 	ovs_unlock();
2262 
2263 	ovs_notify(&dp_vport_genl_family, reply, info);
2264 	return 0;
2265 
2266 exit_unlock_free:
2267 	ovs_unlock();
2268 	kfree_skb(reply);
2269 	return err;
2270 }
2271 
2272 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2273 {
2274 	struct nlattr **a = info->attrs;
2275 	struct sk_buff *reply;
2276 	struct vport *vport;
2277 	int err;
2278 
2279 	reply = ovs_vport_cmd_alloc_info();
2280 	if (!reply)
2281 		return -ENOMEM;
2282 
2283 	ovs_lock();
2284 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2285 	err = PTR_ERR(vport);
2286 	if (IS_ERR(vport))
2287 		goto exit_unlock_free;
2288 
2289 	if (a[OVS_VPORT_ATTR_TYPE] &&
2290 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2291 		err = -EINVAL;
2292 		goto exit_unlock_free;
2293 	}
2294 
2295 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2296 		err = ovs_vport_set_options(vport, a[OVS_VPORT_ATTR_OPTIONS]);
2297 		if (err)
2298 			goto exit_unlock_free;
2299 	}
2300 
2301 
2302 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2303 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2304 
2305 		err = ovs_vport_set_upcall_portids(vport, ids);
2306 		if (err)
2307 			goto exit_unlock_free;
2308 	}
2309 
2310 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2311 				      info->snd_portid, info->snd_seq, 0,
2312 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2313 	BUG_ON(err < 0);
2314 
2315 	ovs_unlock();
2316 	ovs_notify(&dp_vport_genl_family, reply, info);
2317 	return 0;
2318 
2319 exit_unlock_free:
2320 	ovs_unlock();
2321 	kfree_skb(reply);
2322 	return err;
2323 }
2324 
2325 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2326 {
2327 	bool update_headroom = false;
2328 	struct nlattr **a = info->attrs;
2329 	struct sk_buff *reply;
2330 	struct datapath *dp;
2331 	struct vport *vport;
2332 	unsigned int new_headroom;
2333 	int err;
2334 
2335 	reply = ovs_vport_cmd_alloc_info();
2336 	if (!reply)
2337 		return -ENOMEM;
2338 
2339 	ovs_lock();
2340 	vport = lookup_vport(sock_net(skb->sk), info->userhdr, a);
2341 	err = PTR_ERR(vport);
2342 	if (IS_ERR(vport))
2343 		goto exit_unlock_free;
2344 
2345 	if (vport->port_no == OVSP_LOCAL) {
2346 		err = -EINVAL;
2347 		goto exit_unlock_free;
2348 	}
2349 
2350 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2351 				      info->snd_portid, info->snd_seq, 0,
2352 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2353 	BUG_ON(err < 0);
2354 
2355 	/* the vport deletion may trigger dp headroom update */
2356 	dp = vport->dp;
2357 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2358 		update_headroom = true;
2359 
2360 	netdev_reset_rx_headroom(vport->dev);
2361 	ovs_dp_detach_port(vport);
2362 
2363 	if (update_headroom) {
2364 		new_headroom = ovs_get_max_headroom(dp);
2365 
2366 		if (new_headroom < dp->max_headroom)
2367 			ovs_update_headroom(dp, new_headroom);
2368 	}
2369 	ovs_unlock();
2370 
2371 	ovs_notify(&dp_vport_genl_family, reply, info);
2372 	return 0;
2373 
2374 exit_unlock_free:
2375 	ovs_unlock();
2376 	kfree_skb(reply);
2377 	return err;
2378 }
2379 
2380 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2381 {
2382 	struct nlattr **a = info->attrs;
2383 	struct ovs_header *ovs_header = info->userhdr;
2384 	struct sk_buff *reply;
2385 	struct vport *vport;
2386 	int err;
2387 
2388 	reply = ovs_vport_cmd_alloc_info();
2389 	if (!reply)
2390 		return -ENOMEM;
2391 
2392 	rcu_read_lock();
2393 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2394 	err = PTR_ERR(vport);
2395 	if (IS_ERR(vport))
2396 		goto exit_unlock_free;
2397 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2398 				      info->snd_portid, info->snd_seq, 0,
2399 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2400 	BUG_ON(err < 0);
2401 	rcu_read_unlock();
2402 
2403 	return genlmsg_reply(reply, info);
2404 
2405 exit_unlock_free:
2406 	rcu_read_unlock();
2407 	kfree_skb(reply);
2408 	return err;
2409 }
2410 
2411 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2412 {
2413 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2414 	struct datapath *dp;
2415 	int bucket = cb->args[0], skip = cb->args[1];
2416 	int i, j = 0;
2417 
2418 	rcu_read_lock();
2419 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2420 	if (!dp) {
2421 		rcu_read_unlock();
2422 		return -ENODEV;
2423 	}
2424 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2425 		struct vport *vport;
2426 
2427 		j = 0;
2428 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2429 			if (j >= skip &&
2430 			    ovs_vport_cmd_fill_info(vport, skb,
2431 						    sock_net(skb->sk),
2432 						    NETLINK_CB(cb->skb).portid,
2433 						    cb->nlh->nlmsg_seq,
2434 						    NLM_F_MULTI,
2435 						    OVS_VPORT_CMD_GET,
2436 						    GFP_ATOMIC) < 0)
2437 				goto out;
2438 
2439 			j++;
2440 		}
2441 		skip = 0;
2442 	}
2443 out:
2444 	rcu_read_unlock();
2445 
2446 	cb->args[0] = i;
2447 	cb->args[1] = j;
2448 
2449 	return skb->len;
2450 }
2451 
2452 static void ovs_dp_masks_rebalance(struct work_struct *work)
2453 {
2454 	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2455 					       masks_rebalance.work);
2456 	struct datapath *dp;
2457 
2458 	ovs_lock();
2459 
2460 	list_for_each_entry(dp, &ovs_net->dps, list_node)
2461 		ovs_flow_masks_rebalance(&dp->table);
2462 
2463 	ovs_unlock();
2464 
2465 	schedule_delayed_work(&ovs_net->masks_rebalance,
2466 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2467 }
2468 
2469 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2470 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2471 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2472 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2473 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2474 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2475 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED },
2476 	[OVS_VPORT_ATTR_IFINDEX] = { .type = NLA_U32 },
2477 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2478 };
2479 
2480 static const struct genl_small_ops dp_vport_genl_ops[] = {
2481 	{ .cmd = OVS_VPORT_CMD_NEW,
2482 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2483 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2484 	  .doit = ovs_vport_cmd_new
2485 	},
2486 	{ .cmd = OVS_VPORT_CMD_DEL,
2487 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2488 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2489 	  .doit = ovs_vport_cmd_del
2490 	},
2491 	{ .cmd = OVS_VPORT_CMD_GET,
2492 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2493 	  .flags = 0,		    /* OK for unprivileged users. */
2494 	  .doit = ovs_vport_cmd_get,
2495 	  .dumpit = ovs_vport_cmd_dump
2496 	},
2497 	{ .cmd = OVS_VPORT_CMD_SET,
2498 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2499 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2500 	  .doit = ovs_vport_cmd_set,
2501 	},
2502 };
2503 
2504 struct genl_family dp_vport_genl_family __ro_after_init = {
2505 	.hdrsize = sizeof(struct ovs_header),
2506 	.name = OVS_VPORT_FAMILY,
2507 	.version = OVS_VPORT_VERSION,
2508 	.maxattr = OVS_VPORT_ATTR_MAX,
2509 	.policy = vport_policy,
2510 	.netnsok = true,
2511 	.parallel_ops = true,
2512 	.small_ops = dp_vport_genl_ops,
2513 	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2514 	.mcgrps = &ovs_dp_vport_multicast_group,
2515 	.n_mcgrps = 1,
2516 	.module = THIS_MODULE,
2517 };
2518 
2519 static struct genl_family * const dp_genl_families[] = {
2520 	&dp_datapath_genl_family,
2521 	&dp_vport_genl_family,
2522 	&dp_flow_genl_family,
2523 	&dp_packet_genl_family,
2524 	&dp_meter_genl_family,
2525 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2526 	&dp_ct_limit_genl_family,
2527 #endif
2528 };
2529 
2530 static void dp_unregister_genl(int n_families)
2531 {
2532 	int i;
2533 
2534 	for (i = 0; i < n_families; i++)
2535 		genl_unregister_family(dp_genl_families[i]);
2536 }
2537 
2538 static int __init dp_register_genl(void)
2539 {
2540 	int err;
2541 	int i;
2542 
2543 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2544 
2545 		err = genl_register_family(dp_genl_families[i]);
2546 		if (err)
2547 			goto error;
2548 	}
2549 
2550 	return 0;
2551 
2552 error:
2553 	dp_unregister_genl(i);
2554 	return err;
2555 }
2556 
2557 static int __net_init ovs_init_net(struct net *net)
2558 {
2559 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2560 	int err;
2561 
2562 	INIT_LIST_HEAD(&ovs_net->dps);
2563 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2564 	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2565 
2566 	err = ovs_ct_init(net);
2567 	if (err)
2568 		return err;
2569 
2570 	schedule_delayed_work(&ovs_net->masks_rebalance,
2571 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2572 	return 0;
2573 }
2574 
2575 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2576 					    struct list_head *head)
2577 {
2578 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2579 	struct datapath *dp;
2580 
2581 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2582 		int i;
2583 
2584 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2585 			struct vport *vport;
2586 
2587 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2588 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2589 					continue;
2590 
2591 				if (dev_net(vport->dev) == dnet)
2592 					list_add(&vport->detach_list, head);
2593 			}
2594 		}
2595 	}
2596 }
2597 
2598 static void __net_exit ovs_exit_net(struct net *dnet)
2599 {
2600 	struct datapath *dp, *dp_next;
2601 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2602 	struct vport *vport, *vport_next;
2603 	struct net *net;
2604 	LIST_HEAD(head);
2605 
2606 	ovs_lock();
2607 
2608 	ovs_ct_exit(dnet);
2609 
2610 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2611 		__dp_destroy(dp);
2612 
2613 	down_read(&net_rwsem);
2614 	for_each_net(net)
2615 		list_vports_from_net(net, dnet, &head);
2616 	up_read(&net_rwsem);
2617 
2618 	/* Detach all vports from given namespace. */
2619 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2620 		list_del(&vport->detach_list);
2621 		ovs_dp_detach_port(vport);
2622 	}
2623 
2624 	ovs_unlock();
2625 
2626 	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2627 	cancel_work_sync(&ovs_net->dp_notify_work);
2628 }
2629 
2630 static struct pernet_operations ovs_net_ops = {
2631 	.init = ovs_init_net,
2632 	.exit = ovs_exit_net,
2633 	.id   = &ovs_net_id,
2634 	.size = sizeof(struct ovs_net),
2635 };
2636 
2637 static int __init dp_init(void)
2638 {
2639 	int err;
2640 
2641 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2642 		     sizeof_field(struct sk_buff, cb));
2643 
2644 	pr_info("Open vSwitch switching datapath\n");
2645 
2646 	err = action_fifos_init();
2647 	if (err)
2648 		goto error;
2649 
2650 	err = ovs_internal_dev_rtnl_link_register();
2651 	if (err)
2652 		goto error_action_fifos_exit;
2653 
2654 	err = ovs_flow_init();
2655 	if (err)
2656 		goto error_unreg_rtnl_link;
2657 
2658 	err = ovs_vport_init();
2659 	if (err)
2660 		goto error_flow_exit;
2661 
2662 	err = register_pernet_device(&ovs_net_ops);
2663 	if (err)
2664 		goto error_vport_exit;
2665 
2666 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2667 	if (err)
2668 		goto error_netns_exit;
2669 
2670 	err = ovs_netdev_init();
2671 	if (err)
2672 		goto error_unreg_notifier;
2673 
2674 	err = dp_register_genl();
2675 	if (err < 0)
2676 		goto error_unreg_netdev;
2677 
2678 	return 0;
2679 
2680 error_unreg_netdev:
2681 	ovs_netdev_exit();
2682 error_unreg_notifier:
2683 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2684 error_netns_exit:
2685 	unregister_pernet_device(&ovs_net_ops);
2686 error_vport_exit:
2687 	ovs_vport_exit();
2688 error_flow_exit:
2689 	ovs_flow_exit();
2690 error_unreg_rtnl_link:
2691 	ovs_internal_dev_rtnl_link_unregister();
2692 error_action_fifos_exit:
2693 	action_fifos_exit();
2694 error:
2695 	return err;
2696 }
2697 
2698 static void dp_cleanup(void)
2699 {
2700 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2701 	ovs_netdev_exit();
2702 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2703 	unregister_pernet_device(&ovs_net_ops);
2704 	rcu_barrier();
2705 	ovs_vport_exit();
2706 	ovs_flow_exit();
2707 	ovs_internal_dev_rtnl_link_unregister();
2708 	action_fifos_exit();
2709 }
2710 
2711 module_init(dp_init);
2712 module_exit(dp_cleanup);
2713 
2714 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2715 MODULE_LICENSE("GPL");
2716 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2717 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2718 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2719 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2720 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2721 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2722