xref: /linux/net/openvswitch/datapath.c (revision 1b78070aaef63512688aebfbc82365ef9d6660f1)
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/kernel.h>
19 #include <linux/kthread.h>
20 #include <linux/mutex.h>
21 #include <linux/percpu.h>
22 #include <linux/rcupdate.h>
23 #include <linux/tcp.h>
24 #include <linux/udp.h>
25 #include <linux/ethtool.h>
26 #include <linux/wait.h>
27 #include <asm/div64.h>
28 #include <linux/highmem.h>
29 #include <linux/netfilter_bridge.h>
30 #include <linux/netfilter_ipv4.h>
31 #include <linux/inetdevice.h>
32 #include <linux/list.h>
33 #include <linux/openvswitch.h>
34 #include <linux/rculist.h>
35 #include <linux/dmi.h>
36 #include <net/genetlink.h>
37 #include <net/gso.h>
38 #include <net/net_namespace.h>
39 #include <net/netns/generic.h>
40 #include <net/pkt_cls.h>
41 
42 #include "datapath.h"
43 #include "drop.h"
44 #include "flow.h"
45 #include "flow_table.h"
46 #include "flow_netlink.h"
47 #include "meter.h"
48 #include "openvswitch_trace.h"
49 #include "vport-internal_dev.h"
50 #include "vport-netdev.h"
51 
52 unsigned int ovs_net_id __read_mostly;
53 
54 static struct genl_family dp_packet_genl_family;
55 static struct genl_family dp_flow_genl_family;
56 static struct genl_family dp_datapath_genl_family;
57 
58 static const struct nla_policy flow_policy[];
59 
60 static const struct genl_multicast_group ovs_dp_flow_multicast_group = {
61 	.name = OVS_FLOW_MCGROUP,
62 };
63 
64 static const struct genl_multicast_group ovs_dp_datapath_multicast_group = {
65 	.name = OVS_DATAPATH_MCGROUP,
66 };
67 
68 static const struct genl_multicast_group ovs_dp_vport_multicast_group = {
69 	.name = OVS_VPORT_MCGROUP,
70 };
71 
72 /* Check if need to build a reply message.
73  * OVS userspace sets the NLM_F_ECHO flag if it needs the reply. */
74 static bool ovs_must_notify(struct genl_family *family, struct genl_info *info,
75 			    unsigned int group)
76 {
77 	return info->nlhdr->nlmsg_flags & NLM_F_ECHO ||
78 	       genl_has_listeners(family, genl_info_net(info), group);
79 }
80 
81 static void ovs_notify(struct genl_family *family,
82 		       struct sk_buff *skb, struct genl_info *info)
83 {
84 	genl_notify(family, skb, info, 0, GFP_KERNEL);
85 }
86 
87 /**
88  * DOC: Locking:
89  *
90  * All writes e.g. Writes to device state (add/remove datapath, port, set
91  * operations on vports, etc.), Writes to other state (flow table
92  * modifications, set miscellaneous datapath parameters, etc.) are protected
93  * by ovs_lock.
94  *
95  * Reads are protected by RCU.
96  *
97  * There are a few special cases (mostly stats) that have their own
98  * synchronization but they nest under all of above and don't interact with
99  * each other.
100  *
101  * The RTNL lock nests inside ovs_mutex.
102  */
103 
104 static DEFINE_MUTEX(ovs_mutex);
105 
106 void ovs_lock(void)
107 {
108 	mutex_lock(&ovs_mutex);
109 }
110 
111 void ovs_unlock(void)
112 {
113 	mutex_unlock(&ovs_mutex);
114 }
115 
116 #ifdef CONFIG_LOCKDEP
117 int lockdep_ovsl_is_held(void)
118 {
119 	if (debug_locks)
120 		return lockdep_is_held(&ovs_mutex);
121 	else
122 		return 1;
123 }
124 #endif
125 
126 static struct vport *new_vport(const struct vport_parms *);
127 static int queue_gso_packets(struct datapath *dp, struct sk_buff *,
128 			     const struct sw_flow_key *,
129 			     const struct dp_upcall_info *,
130 			     uint32_t cutlen);
131 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *,
132 				  const struct sw_flow_key *,
133 				  const struct dp_upcall_info *,
134 				  uint32_t cutlen);
135 
136 static void ovs_dp_masks_rebalance(struct work_struct *work);
137 
138 static int ovs_dp_set_upcall_portids(struct datapath *, const struct nlattr *);
139 
140 /* Must be called with rcu_read_lock or ovs_mutex. */
141 const char *ovs_dp_name(const struct datapath *dp)
142 {
143 	struct vport *vport = ovs_vport_ovsl_rcu(dp, OVSP_LOCAL);
144 	return ovs_vport_name(vport);
145 }
146 
147 static int get_dpifindex(const struct datapath *dp)
148 {
149 	struct vport *local;
150 	int ifindex;
151 
152 	rcu_read_lock();
153 
154 	local = ovs_vport_rcu(dp, OVSP_LOCAL);
155 	if (local)
156 		ifindex = local->dev->ifindex;
157 	else
158 		ifindex = 0;
159 
160 	rcu_read_unlock();
161 
162 	return ifindex;
163 }
164 
165 static void destroy_dp_rcu(struct rcu_head *rcu)
166 {
167 	struct datapath *dp = container_of(rcu, struct datapath, rcu);
168 
169 	ovs_flow_tbl_destroy(&dp->table);
170 	free_percpu(dp->stats_percpu);
171 	kfree(dp->ports);
172 	ovs_meters_exit(dp);
173 	kfree(rcu_dereference_raw(dp->upcall_portids));
174 	kfree(dp);
175 }
176 
177 static struct hlist_head *vport_hash_bucket(const struct datapath *dp,
178 					    u16 port_no)
179 {
180 	return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)];
181 }
182 
183 /* Called with ovs_mutex or RCU read lock. */
184 struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no)
185 {
186 	struct vport *vport;
187 	struct hlist_head *head;
188 
189 	head = vport_hash_bucket(dp, port_no);
190 	hlist_for_each_entry_rcu(vport, head, dp_hash_node,
191 				 lockdep_ovsl_is_held()) {
192 		if (vport->port_no == port_no)
193 			return vport;
194 	}
195 	return NULL;
196 }
197 
198 /* Called with ovs_mutex. */
199 static struct vport *new_vport(const struct vport_parms *parms)
200 {
201 	struct vport *vport;
202 
203 	vport = ovs_vport_add(parms);
204 	if (!IS_ERR(vport)) {
205 		struct datapath *dp = parms->dp;
206 		struct hlist_head *head = vport_hash_bucket(dp, vport->port_no);
207 
208 		hlist_add_head_rcu(&vport->dp_hash_node, head);
209 	}
210 	return vport;
211 }
212 
213 static void ovs_vport_update_upcall_stats(struct sk_buff *skb,
214 					  const struct dp_upcall_info *upcall_info,
215 					  bool upcall_result)
216 {
217 	struct vport *p = OVS_CB(skb)->input_vport;
218 	struct vport_upcall_stats_percpu *stats;
219 
220 	if (upcall_info->cmd != OVS_PACKET_CMD_MISS &&
221 	    upcall_info->cmd != OVS_PACKET_CMD_ACTION)
222 		return;
223 
224 	stats = this_cpu_ptr(p->upcall_stats);
225 	u64_stats_update_begin(&stats->syncp);
226 	if (upcall_result)
227 		u64_stats_inc(&stats->n_success);
228 	else
229 		u64_stats_inc(&stats->n_fail);
230 	u64_stats_update_end(&stats->syncp);
231 }
232 
233 void ovs_dp_detach_port(struct vport *p)
234 {
235 	ASSERT_OVSL();
236 
237 	/* First drop references to device. */
238 	hlist_del_rcu(&p->dp_hash_node);
239 
240 	/* Then destroy it. */
241 	ovs_vport_del(p);
242 }
243 
244 /* Must be called with rcu_read_lock. */
245 void ovs_dp_process_packet(struct sk_buff *skb, struct sw_flow_key *key)
246 {
247 	struct ovs_pcpu_storage *ovs_pcpu = this_cpu_ptr(ovs_pcpu_storage);
248 	const struct vport *p = OVS_CB(skb)->input_vport;
249 	struct datapath *dp = p->dp;
250 	struct sw_flow *flow;
251 	struct sw_flow_actions *sf_acts;
252 	struct dp_stats_percpu *stats;
253 	bool ovs_pcpu_locked = false;
254 	u64 *stats_counter;
255 	u32 n_mask_hit;
256 	u32 n_cache_hit;
257 	int error;
258 
259 	stats = this_cpu_ptr(dp->stats_percpu);
260 
261 	/* Look up flow. */
262 	flow = ovs_flow_tbl_lookup_stats(&dp->table, key, skb_get_hash(skb),
263 					 &n_mask_hit, &n_cache_hit);
264 	if (unlikely(!flow)) {
265 		struct dp_upcall_info upcall;
266 
267 		memset(&upcall, 0, sizeof(upcall));
268 		upcall.cmd = OVS_PACKET_CMD_MISS;
269 
270 		if (OVS_CB(skb)->upcall_pid)
271 			upcall.portid = OVS_CB(skb)->upcall_pid;
272 		else if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU)
273 			upcall.portid =
274 			    ovs_dp_get_upcall_portid(dp, smp_processor_id());
275 		else
276 			upcall.portid = ovs_vport_find_upcall_portid(p, skb);
277 
278 		upcall.mru = OVS_CB(skb)->mru;
279 		error = ovs_dp_upcall(dp, skb, key, &upcall, U32_MAX);
280 		switch (error) {
281 		case 0:
282 		case -EAGAIN:
283 		case -ERESTARTSYS:
284 		case -EINTR:
285 			consume_skb(skb);
286 			break;
287 		default:
288 			skb_tx_error(skb);
289 			kfree_skb(skb);
290 			break;
291 		}
292 		stats_counter = &stats->n_missed;
293 		goto out;
294 	}
295 
296 	ovs_flow_stats_update(flow, key->tp.flags, skb);
297 	sf_acts = rcu_dereference(flow->sf_acts);
298 	/* This path can be invoked recursively: Use the current task to
299 	 * identify recursive invocation - the lock must be acquired only once.
300 	 * Even with disabled bottom halves this can be preempted on PREEMPT_RT.
301 	 * Limit the locking to RT to avoid assigning `owner' if it can be
302 	 * avoided.
303 	 */
304 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && ovs_pcpu->owner != current) {
305 		local_lock_nested_bh(&ovs_pcpu_storage->bh_lock);
306 		ovs_pcpu->owner = current;
307 		ovs_pcpu_locked = true;
308 	}
309 
310 	error = ovs_execute_actions(dp, skb, sf_acts, key);
311 	if (unlikely(error))
312 		net_dbg_ratelimited("ovs: action execution error on datapath %s: %d\n",
313 				    ovs_dp_name(dp), error);
314 	if (ovs_pcpu_locked) {
315 		ovs_pcpu->owner = NULL;
316 		local_unlock_nested_bh(&ovs_pcpu_storage->bh_lock);
317 	}
318 
319 	stats_counter = &stats->n_hit;
320 
321 out:
322 	/* Update datapath statistics. */
323 	u64_stats_update_begin(&stats->syncp);
324 	(*stats_counter)++;
325 	stats->n_mask_hit += n_mask_hit;
326 	stats->n_cache_hit += n_cache_hit;
327 	u64_stats_update_end(&stats->syncp);
328 }
329 
330 int ovs_dp_upcall(struct datapath *dp, struct sk_buff *skb,
331 		  const struct sw_flow_key *key,
332 		  const struct dp_upcall_info *upcall_info,
333 		  uint32_t cutlen)
334 {
335 	struct dp_stats_percpu *stats;
336 	int err;
337 
338 	if (trace_ovs_dp_upcall_enabled())
339 		trace_ovs_dp_upcall(dp, skb, key, upcall_info);
340 
341 	if (upcall_info->portid == 0) {
342 		err = -ENOTCONN;
343 		goto err;
344 	}
345 
346 	if (!skb_is_gso(skb))
347 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
348 	else
349 		err = queue_gso_packets(dp, skb, key, upcall_info, cutlen);
350 
351 	ovs_vport_update_upcall_stats(skb, upcall_info, !err);
352 	if (err)
353 		goto err;
354 
355 	return 0;
356 
357 err:
358 	stats = this_cpu_ptr(dp->stats_percpu);
359 
360 	u64_stats_update_begin(&stats->syncp);
361 	stats->n_lost++;
362 	u64_stats_update_end(&stats->syncp);
363 
364 	return err;
365 }
366 
367 static int queue_gso_packets(struct datapath *dp, struct sk_buff *skb,
368 			     const struct sw_flow_key *key,
369 			     const struct dp_upcall_info *upcall_info,
370 			     uint32_t cutlen)
371 {
372 	unsigned int gso_type = skb_shinfo(skb)->gso_type;
373 	struct sw_flow_key later_key;
374 	struct sk_buff *segs, *nskb;
375 	int err;
376 
377 	BUILD_BUG_ON(sizeof(*OVS_CB(skb)) > SKB_GSO_CB_OFFSET);
378 	segs = __skb_gso_segment(skb, NETIF_F_SG, false);
379 	if (IS_ERR(segs))
380 		return PTR_ERR(segs);
381 	if (segs == NULL)
382 		return -EINVAL;
383 
384 	if (gso_type & SKB_GSO_UDP) {
385 		/* The initial flow key extracted by ovs_flow_key_extract()
386 		 * in this case is for a first fragment, so we need to
387 		 * properly mark later fragments.
388 		 */
389 		later_key = *key;
390 		later_key.ip.frag = OVS_FRAG_TYPE_LATER;
391 	}
392 
393 	/* Queue all of the segments. */
394 	skb_list_walk_safe(segs, skb, nskb) {
395 		if (gso_type & SKB_GSO_UDP && skb != segs)
396 			key = &later_key;
397 
398 		err = queue_userspace_packet(dp, skb, key, upcall_info, cutlen);
399 		if (err)
400 			break;
401 
402 	}
403 
404 	/* Free all of the segments. */
405 	skb_list_walk_safe(segs, skb, nskb) {
406 		if (err)
407 			kfree_skb(skb);
408 		else
409 			consume_skb(skb);
410 	}
411 	return err;
412 }
413 
414 static size_t upcall_msg_size(const struct dp_upcall_info *upcall_info,
415 			      unsigned int hdrlen, int actions_attrlen)
416 {
417 	size_t size = NLMSG_ALIGN(sizeof(struct ovs_header))
418 		+ nla_total_size(hdrlen) /* OVS_PACKET_ATTR_PACKET */
419 		+ nla_total_size(ovs_key_attr_size()) /* OVS_PACKET_ATTR_KEY */
420 		+ nla_total_size(sizeof(unsigned int)) /* OVS_PACKET_ATTR_LEN */
421 		+ nla_total_size(sizeof(u64)); /* OVS_PACKET_ATTR_HASH */
422 
423 	/* OVS_PACKET_ATTR_USERDATA */
424 	if (upcall_info->userdata)
425 		size += NLA_ALIGN(upcall_info->userdata->nla_len);
426 
427 	/* OVS_PACKET_ATTR_EGRESS_TUN_KEY */
428 	if (upcall_info->egress_tun_info)
429 		size += nla_total_size(ovs_tun_key_attr_size());
430 
431 	/* OVS_PACKET_ATTR_ACTIONS */
432 	if (upcall_info->actions_len)
433 		size += nla_total_size(actions_attrlen);
434 
435 	/* OVS_PACKET_ATTR_MRU */
436 	if (upcall_info->mru)
437 		size += nla_total_size(sizeof(upcall_info->mru));
438 
439 	return size;
440 }
441 
442 static void pad_packet(struct datapath *dp, struct sk_buff *skb)
443 {
444 	if (!(dp->user_features & OVS_DP_F_UNALIGNED)) {
445 		size_t plen = NLA_ALIGN(skb->len) - skb->len;
446 
447 		if (plen > 0)
448 			skb_put_zero(skb, plen);
449 	}
450 }
451 
452 static int queue_userspace_packet(struct datapath *dp, struct sk_buff *skb,
453 				  const struct sw_flow_key *key,
454 				  const struct dp_upcall_info *upcall_info,
455 				  uint32_t cutlen)
456 {
457 	struct ovs_header *upcall;
458 	struct sk_buff *nskb = NULL;
459 	struct sk_buff *user_skb = NULL; /* to be queued to userspace */
460 	struct nlattr *nla;
461 	size_t msg_size;
462 	size_t skb_len;
463 	unsigned int hlen;
464 	int err, dp_ifindex;
465 	u64 hash;
466 
467 	dp_ifindex = get_dpifindex(dp);
468 	if (!dp_ifindex)
469 		return -ENODEV;
470 
471 	if (!skb_frags_readable(skb))
472 		return -EFAULT;
473 
474 	if (skb_vlan_tag_present(skb)) {
475 		nskb = skb_clone(skb, GFP_ATOMIC);
476 		if (!nskb)
477 			return -ENOMEM;
478 
479 		nskb = __vlan_hwaccel_push_inside(nskb);
480 		if (!nskb)
481 			return -ENOMEM;
482 
483 		skb = nskb;
484 	}
485 
486 	skb_len = min(skb->len, cutlen);
487 	if (nla_attr_size(skb_len) > USHRT_MAX) {
488 		err = -EFBIG;
489 		goto out;
490 	}
491 
492 	/* Complete checksum if needed */
493 	if (skb->ip_summed == CHECKSUM_PARTIAL &&
494 	    (err = skb_csum_hwoffload_help(skb, 0)))
495 		goto out;
496 
497 	/* Older versions of OVS user space enforce alignment of the last
498 	 * Netlink attribute to NLA_ALIGNTO which would require extensive
499 	 * padding logic. Only perform zerocopy if padding is not required.
500 	 */
501 	if (dp->user_features & OVS_DP_F_UNALIGNED)
502 		hlen = min(skb_zerocopy_headlen(skb), cutlen);
503 	else
504 		hlen = skb_len;
505 
506 	msg_size = upcall_msg_size(upcall_info, hlen,
507 				   OVS_CB(skb)->acts_origlen);
508 	user_skb = genlmsg_new(msg_size, GFP_ATOMIC);
509 	if (!user_skb) {
510 		err = -ENOMEM;
511 		goto out;
512 	}
513 
514 	upcall = genlmsg_put(user_skb, 0, 0, &dp_packet_genl_family,
515 			     0, upcall_info->cmd);
516 	if (!upcall) {
517 		err = -EINVAL;
518 		goto out;
519 	}
520 	upcall->dp_ifindex = dp_ifindex;
521 
522 	err = ovs_nla_put_key(key, key, OVS_PACKET_ATTR_KEY, false, user_skb);
523 	if (err)
524 		goto out;
525 
526 	if (upcall_info->userdata)
527 		__nla_put(user_skb, OVS_PACKET_ATTR_USERDATA,
528 			  nla_len(upcall_info->userdata),
529 			  nla_data(upcall_info->userdata));
530 
531 	if (upcall_info->egress_tun_info) {
532 		nla = nla_nest_start_noflag(user_skb,
533 					    OVS_PACKET_ATTR_EGRESS_TUN_KEY);
534 		if (!nla) {
535 			err = -EMSGSIZE;
536 			goto out;
537 		}
538 		err = ovs_nla_put_tunnel_info(user_skb,
539 					      upcall_info->egress_tun_info);
540 		if (err)
541 			goto out;
542 
543 		nla_nest_end(user_skb, nla);
544 	}
545 
546 	if (upcall_info->actions_len) {
547 		nla = nla_nest_start_noflag(user_skb, OVS_PACKET_ATTR_ACTIONS);
548 		if (!nla) {
549 			err = -EMSGSIZE;
550 			goto out;
551 		}
552 		err = ovs_nla_put_actions(upcall_info->actions,
553 					  upcall_info->actions_len,
554 					  user_skb);
555 		if (!err)
556 			nla_nest_end(user_skb, nla);
557 		else
558 			nla_nest_cancel(user_skb, nla);
559 	}
560 
561 	/* Add OVS_PACKET_ATTR_MRU */
562 	if (upcall_info->mru &&
563 	    nla_put_u16(user_skb, OVS_PACKET_ATTR_MRU, upcall_info->mru)) {
564 		err = -ENOBUFS;
565 		goto out;
566 	}
567 
568 	/* Add OVS_PACKET_ATTR_LEN when packet is truncated */
569 	if (skb_len < skb->len &&
570 	    nla_put_u32(user_skb, OVS_PACKET_ATTR_LEN, skb->len)) {
571 		err = -ENOBUFS;
572 		goto out;
573 	}
574 
575 	/* Add OVS_PACKET_ATTR_HASH */
576 	hash = skb_get_hash_raw(skb);
577 	if (skb->sw_hash)
578 		hash |= OVS_PACKET_HASH_SW_BIT;
579 
580 	if (skb->l4_hash)
581 		hash |= OVS_PACKET_HASH_L4_BIT;
582 
583 	if (nla_put(user_skb, OVS_PACKET_ATTR_HASH, sizeof (u64), &hash)) {
584 		err = -ENOBUFS;
585 		goto out;
586 	}
587 
588 	/* Only reserve room for attribute header, packet data is added
589 	 * in skb_zerocopy() */
590 	if (!(nla = nla_reserve(user_skb, OVS_PACKET_ATTR_PACKET, 0))) {
591 		err = -ENOBUFS;
592 		goto out;
593 	}
594 	nla->nla_len = nla_attr_size(skb_len);
595 
596 	err = skb_zerocopy(user_skb, skb, skb_len, hlen);
597 	if (err)
598 		goto out;
599 
600 	/* Pad OVS_PACKET_ATTR_PACKET if linear copy was performed */
601 	pad_packet(dp, user_skb);
602 
603 	((struct nlmsghdr *) user_skb->data)->nlmsg_len = user_skb->len;
604 
605 	err = genlmsg_unicast(ovs_dp_get_net(dp), user_skb, upcall_info->portid);
606 	user_skb = NULL;
607 out:
608 	consume_skb(user_skb);
609 	consume_skb(nskb);
610 
611 	return err;
612 }
613 
614 static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info)
615 {
616 	struct ovs_header *ovs_header = genl_info_userhdr(info);
617 	struct net *net = sock_net(skb->sk);
618 	struct nlattr **a = info->attrs;
619 	struct sw_flow_actions *acts;
620 	struct sk_buff *packet;
621 	struct sw_flow *flow;
622 	struct sw_flow_actions *sf_acts;
623 	struct datapath *dp;
624 	struct vport *input_vport;
625 	u16 mru = 0;
626 	u64 hash;
627 	int len;
628 	int err;
629 	bool log = !a[OVS_PACKET_ATTR_PROBE];
630 
631 	err = -EINVAL;
632 	if (!a[OVS_PACKET_ATTR_PACKET] || !a[OVS_PACKET_ATTR_KEY] ||
633 	    !a[OVS_PACKET_ATTR_ACTIONS])
634 		goto err;
635 
636 	len = nla_len(a[OVS_PACKET_ATTR_PACKET]);
637 	packet = __dev_alloc_skb(NET_IP_ALIGN + len, GFP_KERNEL);
638 	err = -ENOMEM;
639 	if (!packet)
640 		goto err;
641 	skb_reserve(packet, NET_IP_ALIGN);
642 
643 	nla_memcpy(__skb_put(packet, len), a[OVS_PACKET_ATTR_PACKET], len);
644 
645 	/* Set packet's mru */
646 	if (a[OVS_PACKET_ATTR_MRU]) {
647 		mru = nla_get_u16(a[OVS_PACKET_ATTR_MRU]);
648 		packet->ignore_df = 1;
649 	}
650 	OVS_CB(packet)->mru = mru;
651 	OVS_CB(packet)->cutlen = U32_MAX;
652 
653 	if (a[OVS_PACKET_ATTR_HASH]) {
654 		hash = nla_get_u64(a[OVS_PACKET_ATTR_HASH]);
655 
656 		__skb_set_hash(packet, hash & 0xFFFFFFFFULL,
657 			       !!(hash & OVS_PACKET_HASH_SW_BIT),
658 			       !!(hash & OVS_PACKET_HASH_L4_BIT));
659 	}
660 
661 	OVS_CB(packet)->upcall_pid =
662 		nla_get_u32_default(a[OVS_PACKET_ATTR_UPCALL_PID], 0);
663 
664 	/* Build an sw_flow for sending this packet. */
665 	flow = ovs_flow_alloc();
666 	err = PTR_ERR(flow);
667 	if (IS_ERR(flow))
668 		goto err_kfree_skb;
669 
670 	err = ovs_flow_key_extract_userspace(net, a[OVS_PACKET_ATTR_KEY],
671 					     packet, &flow->key, log);
672 	if (err)
673 		goto err_flow_free;
674 
675 	err = ovs_nla_copy_actions(net, a[OVS_PACKET_ATTR_ACTIONS],
676 				   &flow->key, &acts, log);
677 	if (err)
678 		goto err_flow_free;
679 
680 	rcu_assign_pointer(flow->sf_acts, acts);
681 	packet->priority = flow->key.phy.priority;
682 	packet->mark = flow->key.phy.skb_mark;
683 
684 	rcu_read_lock();
685 	dp = get_dp_rcu(net, ovs_header->dp_ifindex);
686 	err = -ENODEV;
687 	if (!dp)
688 		goto err_unlock;
689 
690 	input_vport = ovs_vport_rcu(dp, flow->key.phy.in_port);
691 	if (!input_vport)
692 		input_vport = ovs_vport_rcu(dp, OVSP_LOCAL);
693 
694 	if (!input_vport)
695 		goto err_unlock;
696 
697 	packet->dev = input_vport->dev;
698 	OVS_CB(packet)->input_vport = input_vport;
699 	sf_acts = rcu_dereference(flow->sf_acts);
700 
701 	local_bh_disable();
702 	local_lock_nested_bh(&ovs_pcpu_storage->bh_lock);
703 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
704 		this_cpu_write(ovs_pcpu_storage->owner, current);
705 	err = ovs_execute_actions(dp, packet, sf_acts, &flow->key);
706 	if (IS_ENABLED(CONFIG_PREEMPT_RT))
707 		this_cpu_write(ovs_pcpu_storage->owner, NULL);
708 	local_unlock_nested_bh(&ovs_pcpu_storage->bh_lock);
709 	local_bh_enable();
710 	rcu_read_unlock();
711 
712 	ovs_flow_free(flow, false);
713 	return err;
714 
715 err_unlock:
716 	rcu_read_unlock();
717 err_flow_free:
718 	ovs_flow_free(flow, false);
719 err_kfree_skb:
720 	kfree_skb(packet);
721 err:
722 	return err;
723 }
724 
725 static const struct nla_policy packet_policy[OVS_PACKET_ATTR_MAX + 1] = {
726 	[OVS_PACKET_ATTR_PACKET] = { .len = ETH_HLEN },
727 	[OVS_PACKET_ATTR_KEY] = { .type = NLA_NESTED },
728 	[OVS_PACKET_ATTR_ACTIONS] = { .type = NLA_NESTED },
729 	[OVS_PACKET_ATTR_PROBE] = { .type = NLA_FLAG },
730 	[OVS_PACKET_ATTR_MRU] = { .type = NLA_U16 },
731 	[OVS_PACKET_ATTR_HASH] = { .type = NLA_U64 },
732 	[OVS_PACKET_ATTR_UPCALL_PID] = { .type = NLA_U32 },
733 };
734 
735 static const struct genl_small_ops dp_packet_genl_ops[] = {
736 	{ .cmd = OVS_PACKET_CMD_EXECUTE,
737 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
738 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
739 	  .doit = ovs_packet_cmd_execute
740 	}
741 };
742 
743 static struct genl_family dp_packet_genl_family __ro_after_init = {
744 	.hdrsize = sizeof(struct ovs_header),
745 	.name = OVS_PACKET_FAMILY,
746 	.version = OVS_PACKET_VERSION,
747 	.maxattr = OVS_PACKET_ATTR_MAX,
748 	.policy = packet_policy,
749 	.netnsok = true,
750 	.parallel_ops = true,
751 	.small_ops = dp_packet_genl_ops,
752 	.n_small_ops = ARRAY_SIZE(dp_packet_genl_ops),
753 	.resv_start_op = OVS_PACKET_CMD_EXECUTE + 1,
754 	.module = THIS_MODULE,
755 };
756 
757 static void get_dp_stats(const struct datapath *dp, struct ovs_dp_stats *stats,
758 			 struct ovs_dp_megaflow_stats *mega_stats)
759 {
760 	int i;
761 
762 	memset(mega_stats, 0, sizeof(*mega_stats));
763 
764 	stats->n_flows = ovs_flow_tbl_count(&dp->table);
765 	mega_stats->n_masks = ovs_flow_tbl_num_masks(&dp->table);
766 
767 	stats->n_hit = stats->n_missed = stats->n_lost = 0;
768 
769 	for_each_possible_cpu(i) {
770 		const struct dp_stats_percpu *percpu_stats;
771 		struct dp_stats_percpu local_stats;
772 		unsigned int start;
773 
774 		percpu_stats = per_cpu_ptr(dp->stats_percpu, i);
775 
776 		do {
777 			start = u64_stats_fetch_begin(&percpu_stats->syncp);
778 			local_stats = *percpu_stats;
779 		} while (u64_stats_fetch_retry(&percpu_stats->syncp, start));
780 
781 		stats->n_hit += local_stats.n_hit;
782 		stats->n_missed += local_stats.n_missed;
783 		stats->n_lost += local_stats.n_lost;
784 		mega_stats->n_mask_hit += local_stats.n_mask_hit;
785 		mega_stats->n_cache_hit += local_stats.n_cache_hit;
786 	}
787 }
788 
789 static bool should_fill_key(const struct sw_flow_id *sfid, uint32_t ufid_flags)
790 {
791 	return ovs_identifier_is_ufid(sfid) &&
792 	       !(ufid_flags & OVS_UFID_F_OMIT_KEY);
793 }
794 
795 static bool should_fill_mask(uint32_t ufid_flags)
796 {
797 	return !(ufid_flags & OVS_UFID_F_OMIT_MASK);
798 }
799 
800 static bool should_fill_actions(uint32_t ufid_flags)
801 {
802 	return !(ufid_flags & OVS_UFID_F_OMIT_ACTIONS);
803 }
804 
805 static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts,
806 				    const struct sw_flow_id *sfid,
807 				    uint32_t ufid_flags)
808 {
809 	size_t len = NLMSG_ALIGN(sizeof(struct ovs_header));
810 
811 	/* OVS_FLOW_ATTR_UFID, or unmasked flow key as fallback
812 	 * see ovs_nla_put_identifier()
813 	 */
814 	if (sfid && ovs_identifier_is_ufid(sfid))
815 		len += nla_total_size(sfid->ufid_len);
816 	else
817 		len += nla_total_size(ovs_key_attr_size());
818 
819 	/* OVS_FLOW_ATTR_KEY */
820 	if (!sfid || should_fill_key(sfid, ufid_flags))
821 		len += nla_total_size(ovs_key_attr_size());
822 
823 	/* OVS_FLOW_ATTR_MASK */
824 	if (should_fill_mask(ufid_flags))
825 		len += nla_total_size(ovs_key_attr_size());
826 
827 	/* OVS_FLOW_ATTR_ACTIONS */
828 	if (should_fill_actions(ufid_flags))
829 		len += nla_total_size(acts->orig_len);
830 
831 	return len
832 		+ nla_total_size_64bit(sizeof(struct ovs_flow_stats)) /* OVS_FLOW_ATTR_STATS */
833 		+ nla_total_size(1) /* OVS_FLOW_ATTR_TCP_FLAGS */
834 		+ nla_total_size_64bit(8); /* OVS_FLOW_ATTR_USED */
835 }
836 
837 /* Called with ovs_mutex or RCU read lock. */
838 static int ovs_flow_cmd_fill_stats(const struct sw_flow *flow,
839 				   struct sk_buff *skb)
840 {
841 	struct ovs_flow_stats stats;
842 	__be16 tcp_flags;
843 	unsigned long used;
844 
845 	ovs_flow_stats_get(flow, &stats, &used, &tcp_flags);
846 
847 	if (used &&
848 	    nla_put_u64_64bit(skb, OVS_FLOW_ATTR_USED, ovs_flow_used_time(used),
849 			      OVS_FLOW_ATTR_PAD))
850 		return -EMSGSIZE;
851 
852 	if (stats.n_packets &&
853 	    nla_put_64bit(skb, OVS_FLOW_ATTR_STATS,
854 			  sizeof(struct ovs_flow_stats), &stats,
855 			  OVS_FLOW_ATTR_PAD))
856 		return -EMSGSIZE;
857 
858 	if ((u8)ntohs(tcp_flags) &&
859 	     nla_put_u8(skb, OVS_FLOW_ATTR_TCP_FLAGS, (u8)ntohs(tcp_flags)))
860 		return -EMSGSIZE;
861 
862 	return 0;
863 }
864 
865 /* Called with ovs_mutex or RCU read lock. */
866 static int ovs_flow_cmd_fill_actions(const struct sw_flow *flow,
867 				     struct sk_buff *skb, int skb_orig_len)
868 {
869 	struct nlattr *start;
870 	int err;
871 
872 	/* If OVS_FLOW_ATTR_ACTIONS doesn't fit, skip dumping the actions if
873 	 * this is the first flow to be dumped into 'skb'.  This is unusual for
874 	 * Netlink but individual action lists can be longer than
875 	 * NLMSG_GOODSIZE and thus entirely undumpable if we didn't do this.
876 	 * The userspace caller can always fetch the actions separately if it
877 	 * really wants them.  (Most userspace callers in fact don't care.)
878 	 *
879 	 * This can only fail for dump operations because the skb is always
880 	 * properly sized for single flows.
881 	 */
882 	start = nla_nest_start_noflag(skb, OVS_FLOW_ATTR_ACTIONS);
883 	if (start) {
884 		const struct sw_flow_actions *sf_acts;
885 
886 		sf_acts = rcu_dereference_ovsl(flow->sf_acts);
887 		err = ovs_nla_put_actions(sf_acts->actions,
888 					  sf_acts->actions_len, skb);
889 
890 		if (!err)
891 			nla_nest_end(skb, start);
892 		else {
893 			if (skb_orig_len)
894 				return err;
895 
896 			nla_nest_cancel(skb, start);
897 		}
898 	} else if (skb_orig_len) {
899 		return -EMSGSIZE;
900 	}
901 
902 	return 0;
903 }
904 
905 /* Called with ovs_mutex or RCU read lock. */
906 static int ovs_flow_cmd_fill_info(const struct sw_flow *flow, int dp_ifindex,
907 				  struct sk_buff *skb, u32 portid,
908 				  u32 seq, u32 flags, u8 cmd, u32 ufid_flags)
909 {
910 	const int skb_orig_len = skb->len;
911 	struct ovs_header *ovs_header;
912 	int err;
913 
914 	ovs_header = genlmsg_put(skb, portid, seq, &dp_flow_genl_family,
915 				 flags, cmd);
916 	if (!ovs_header)
917 		return -EMSGSIZE;
918 
919 	ovs_header->dp_ifindex = dp_ifindex;
920 
921 	err = ovs_nla_put_identifier(flow, skb);
922 	if (err)
923 		goto error;
924 
925 	if (should_fill_key(&flow->id, ufid_flags)) {
926 		err = ovs_nla_put_masked_key(flow, skb);
927 		if (err)
928 			goto error;
929 	}
930 
931 	if (should_fill_mask(ufid_flags)) {
932 		err = ovs_nla_put_mask(flow, skb);
933 		if (err)
934 			goto error;
935 	}
936 
937 	err = ovs_flow_cmd_fill_stats(flow, skb);
938 	if (err)
939 		goto error;
940 
941 	if (should_fill_actions(ufid_flags)) {
942 		err = ovs_flow_cmd_fill_actions(flow, skb, skb_orig_len);
943 		if (err)
944 			goto error;
945 	}
946 
947 	genlmsg_end(skb, ovs_header);
948 	return 0;
949 
950 error:
951 	genlmsg_cancel(skb, ovs_header);
952 	return err;
953 }
954 
955 /* May not be called with RCU read lock. */
956 static struct sk_buff *ovs_flow_cmd_alloc_info(const struct sw_flow_actions *acts,
957 					       const struct sw_flow_id *sfid,
958 					       struct genl_info *info,
959 					       bool always,
960 					       uint32_t ufid_flags)
961 {
962 	struct sk_buff *skb;
963 	size_t len;
964 
965 	if (!always && !ovs_must_notify(&dp_flow_genl_family, info, 0))
966 		return NULL;
967 
968 	len = ovs_flow_cmd_msg_size(acts, sfid, ufid_flags);
969 	skb = genlmsg_new(len, GFP_KERNEL);
970 	if (!skb)
971 		return ERR_PTR(-ENOMEM);
972 
973 	return skb;
974 }
975 
976 /* Called with ovs_mutex. */
977 static struct sk_buff *ovs_flow_cmd_build_info(const struct sw_flow *flow,
978 					       int dp_ifindex,
979 					       struct genl_info *info, u8 cmd,
980 					       bool always, u32 ufid_flags)
981 {
982 	struct sk_buff *skb;
983 	int retval;
984 
985 	skb = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
986 				      &flow->id, info, always, ufid_flags);
987 	if (IS_ERR_OR_NULL(skb))
988 		return skb;
989 
990 	retval = ovs_flow_cmd_fill_info(flow, dp_ifindex, skb,
991 					info->snd_portid, info->snd_seq, 0,
992 					cmd, ufid_flags);
993 	if (WARN_ON_ONCE(retval < 0)) {
994 		kfree_skb(skb);
995 		skb = ERR_PTR(retval);
996 	}
997 	return skb;
998 }
999 
1000 static int ovs_flow_cmd_new(struct sk_buff *skb, struct genl_info *info)
1001 {
1002 	struct net *net = sock_net(skb->sk);
1003 	struct nlattr **a = info->attrs;
1004 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1005 	struct sw_flow *flow = NULL, *new_flow;
1006 	struct sw_flow_mask mask;
1007 	struct sk_buff *reply;
1008 	struct datapath *dp;
1009 	struct sw_flow_key *key;
1010 	struct sw_flow_actions *acts;
1011 	struct sw_flow_match match;
1012 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1013 	int error;
1014 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1015 
1016 	/* Must have key and actions. */
1017 	error = -EINVAL;
1018 	if (!a[OVS_FLOW_ATTR_KEY]) {
1019 		OVS_NLERR(log, "Flow key attr not present in new flow.");
1020 		goto error;
1021 	}
1022 	if (!a[OVS_FLOW_ATTR_ACTIONS]) {
1023 		OVS_NLERR(log, "Flow actions attr not present in new flow.");
1024 		goto error;
1025 	}
1026 
1027 	/* Most of the time we need to allocate a new flow, do it before
1028 	 * locking.
1029 	 */
1030 	new_flow = ovs_flow_alloc();
1031 	if (IS_ERR(new_flow)) {
1032 		error = PTR_ERR(new_flow);
1033 		goto error;
1034 	}
1035 
1036 	/* Extract key. */
1037 	key = kzalloc_obj(*key);
1038 	if (!key) {
1039 		error = -ENOMEM;
1040 		goto err_kfree_flow;
1041 	}
1042 
1043 	ovs_match_init(&match, key, false, &mask);
1044 	error = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1045 				  a[OVS_FLOW_ATTR_MASK], log);
1046 	if (error)
1047 		goto err_kfree_key;
1048 
1049 	ovs_flow_mask_key(&new_flow->key, key, true, &mask);
1050 
1051 	/* Extract flow identifier. */
1052 	error = ovs_nla_get_identifier(&new_flow->id, a[OVS_FLOW_ATTR_UFID],
1053 				       key, log);
1054 	if (error)
1055 		goto err_kfree_key;
1056 
1057 	/* Validate actions. */
1058 	error = ovs_nla_copy_actions(net, a[OVS_FLOW_ATTR_ACTIONS],
1059 				     &new_flow->key, &acts, log);
1060 	if (error) {
1061 		OVS_NLERR(log, "Flow actions may not be safe on all matching packets.");
1062 		goto err_kfree_key;
1063 	}
1064 
1065 	reply = ovs_flow_cmd_alloc_info(acts, &new_flow->id, info, false,
1066 					ufid_flags);
1067 	if (IS_ERR(reply)) {
1068 		error = PTR_ERR(reply);
1069 		goto err_kfree_acts;
1070 	}
1071 
1072 	ovs_lock();
1073 	dp = get_dp(net, ovs_header->dp_ifindex);
1074 	if (unlikely(!dp)) {
1075 		error = -ENODEV;
1076 		goto err_unlock_ovs;
1077 	}
1078 
1079 	/* Check if this is a duplicate flow */
1080 	if (ovs_identifier_is_ufid(&new_flow->id))
1081 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &new_flow->id);
1082 	if (!flow)
1083 		flow = ovs_flow_tbl_lookup(&dp->table, key);
1084 	if (likely(!flow)) {
1085 		rcu_assign_pointer(new_flow->sf_acts, acts);
1086 
1087 		/* Put flow in bucket. */
1088 		error = ovs_flow_tbl_insert(&dp->table, new_flow, &mask);
1089 		if (unlikely(error)) {
1090 			acts = NULL;
1091 			goto err_unlock_ovs;
1092 		}
1093 
1094 		if (unlikely(reply)) {
1095 			error = ovs_flow_cmd_fill_info(new_flow,
1096 						       ovs_header->dp_ifindex,
1097 						       reply, info->snd_portid,
1098 						       info->snd_seq, 0,
1099 						       OVS_FLOW_CMD_NEW,
1100 						       ufid_flags);
1101 			BUG_ON(error < 0);
1102 		}
1103 		ovs_unlock();
1104 	} else {
1105 		struct sw_flow_actions *old_acts;
1106 
1107 		/* Bail out if we're not allowed to modify an existing flow.
1108 		 * We accept NLM_F_CREATE in place of the intended NLM_F_EXCL
1109 		 * because Generic Netlink treats the latter as a dump
1110 		 * request.  We also accept NLM_F_EXCL in case that bug ever
1111 		 * gets fixed.
1112 		 */
1113 		if (unlikely(info->nlhdr->nlmsg_flags & (NLM_F_CREATE
1114 							 | NLM_F_EXCL))) {
1115 			error = -EEXIST;
1116 			goto err_unlock_ovs;
1117 		}
1118 
1119 		/* Look for any overlapping flow. */
1120 		if (unlikely(!ovs_flow_cmp(flow, &match))) {
1121 			if (ovs_identifier_is_key(&flow->id))
1122 				flow = ovs_flow_tbl_lookup_exact(&dp->table,
1123 								 &match);
1124 			else /* UFID matches but key is different */
1125 				flow = NULL;
1126 			if (!flow) {
1127 				error = -ENOENT;
1128 				goto err_unlock_ovs;
1129 			}
1130 		}
1131 
1132 		if (unlikely(reply)) {
1133 			size_t cur, req;
1134 
1135 			cur = ovs_flow_cmd_msg_size(acts, &new_flow->id,
1136 						    ufid_flags);
1137 			req = ovs_flow_cmd_msg_size(acts, &flow->id,
1138 						    ufid_flags);
1139 			if (cur < req) {
1140 				struct sk_buff *resized;
1141 
1142 				resized = ovs_flow_cmd_alloc_info(acts,
1143 								  &flow->id,
1144 								  info, false,
1145 								  ufid_flags);
1146 				if (IS_ERR(resized)) {
1147 					error = PTR_ERR(resized);
1148 					goto err_unlock_ovs;
1149 				}
1150 				kfree_skb(reply);
1151 				reply = resized;
1152 			}
1153 		}
1154 
1155 		/* Update actions. */
1156 		old_acts = ovsl_dereference(flow->sf_acts);
1157 		rcu_assign_pointer(flow->sf_acts, acts);
1158 
1159 		if (unlikely(reply)) {
1160 			error = ovs_flow_cmd_fill_info(flow,
1161 						       ovs_header->dp_ifindex,
1162 						       reply, info->snd_portid,
1163 						       info->snd_seq, 0,
1164 						       OVS_FLOW_CMD_NEW,
1165 						       ufid_flags);
1166 			BUG_ON(error < 0);
1167 		}
1168 		ovs_unlock();
1169 
1170 		ovs_nla_free_flow_actions_rcu(old_acts);
1171 		ovs_flow_free(new_flow, false);
1172 	}
1173 
1174 	if (reply)
1175 		ovs_notify(&dp_flow_genl_family, reply, info);
1176 
1177 	kfree(key);
1178 	return 0;
1179 
1180 err_unlock_ovs:
1181 	ovs_unlock();
1182 	kfree_skb(reply);
1183 err_kfree_acts:
1184 	ovs_nla_free_flow_actions(acts);
1185 err_kfree_key:
1186 	kfree(key);
1187 err_kfree_flow:
1188 	ovs_flow_free(new_flow, false);
1189 error:
1190 	return error;
1191 }
1192 
1193 /* Factor out action copy to avoid "Wframe-larger-than=1024" warning. */
1194 static noinline_for_stack
1195 struct sw_flow_actions *get_flow_actions(struct net *net,
1196 					 const struct nlattr *a,
1197 					 const struct sw_flow_key *key,
1198 					 const struct sw_flow_mask *mask,
1199 					 bool log)
1200 {
1201 	struct sw_flow_actions *acts;
1202 	struct sw_flow_key masked_key;
1203 	int error;
1204 
1205 	ovs_flow_mask_key(&masked_key, key, true, mask);
1206 	error = ovs_nla_copy_actions(net, a, &masked_key, &acts, log);
1207 	if (error) {
1208 		OVS_NLERR(log,
1209 			  "Actions may not be safe on all matching packets");
1210 		return ERR_PTR(error);
1211 	}
1212 
1213 	return acts;
1214 }
1215 
1216 /* Factor out match-init and action-copy to avoid
1217  * "Wframe-larger-than=1024" warning. Because mask is only
1218  * used to get actions, we new a function to save some
1219  * stack space.
1220  *
1221  * If there are not key and action attrs, we return 0
1222  * directly. In the case, the caller will also not use the
1223  * match as before. If there is action attr, we try to get
1224  * actions and save them to *acts. Before returning from
1225  * the function, we reset the match->mask pointer. Because
1226  * we should not to return match object with dangling reference
1227  * to mask.
1228  * */
1229 static noinline_for_stack int
1230 ovs_nla_init_match_and_action(struct net *net,
1231 			      struct sw_flow_match *match,
1232 			      struct sw_flow_key *key,
1233 			      struct nlattr **a,
1234 			      struct sw_flow_actions **acts,
1235 			      bool log)
1236 {
1237 	struct sw_flow_mask mask;
1238 	int error = 0;
1239 
1240 	if (a[OVS_FLOW_ATTR_KEY]) {
1241 		ovs_match_init(match, key, true, &mask);
1242 		error = ovs_nla_get_match(net, match, a[OVS_FLOW_ATTR_KEY],
1243 					  a[OVS_FLOW_ATTR_MASK], log);
1244 		if (error)
1245 			goto error;
1246 	}
1247 
1248 	if (a[OVS_FLOW_ATTR_ACTIONS]) {
1249 		if (!a[OVS_FLOW_ATTR_KEY]) {
1250 			OVS_NLERR(log,
1251 				  "Flow key attribute not present in set flow.");
1252 			error = -EINVAL;
1253 			goto error;
1254 		}
1255 
1256 		*acts = get_flow_actions(net, a[OVS_FLOW_ATTR_ACTIONS], key,
1257 					 &mask, log);
1258 		if (IS_ERR(*acts)) {
1259 			error = PTR_ERR(*acts);
1260 			goto error;
1261 		}
1262 	}
1263 
1264 	/* On success, error is 0. */
1265 error:
1266 	match->mask = NULL;
1267 	return error;
1268 }
1269 
1270 static int ovs_flow_cmd_set(struct sk_buff *skb, struct genl_info *info)
1271 {
1272 	struct net *net = sock_net(skb->sk);
1273 	struct nlattr **a = info->attrs;
1274 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1275 	struct sw_flow_key key;
1276 	struct sw_flow *flow;
1277 	struct sk_buff *reply = NULL;
1278 	struct datapath *dp;
1279 	struct sw_flow_actions *old_acts = NULL, *acts = NULL;
1280 	struct sw_flow_match match;
1281 	struct sw_flow_id sfid;
1282 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1283 	int error = 0;
1284 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1285 	bool ufid_present;
1286 
1287 	ufid_present = ovs_nla_get_ufid(&sfid, a[OVS_FLOW_ATTR_UFID], log);
1288 	if (!a[OVS_FLOW_ATTR_KEY] && !ufid_present) {
1289 		OVS_NLERR(log,
1290 			  "Flow set message rejected, Key attribute missing.");
1291 		return -EINVAL;
1292 	}
1293 
1294 	error = ovs_nla_init_match_and_action(net, &match, &key, a,
1295 					      &acts, log);
1296 	if (error)
1297 		goto error;
1298 
1299 	if (acts) {
1300 		/* Can allocate before locking if have acts. */
1301 		reply = ovs_flow_cmd_alloc_info(acts, &sfid, info, false,
1302 						ufid_flags);
1303 		if (IS_ERR(reply)) {
1304 			error = PTR_ERR(reply);
1305 			goto err_kfree_acts;
1306 		}
1307 	}
1308 
1309 	ovs_lock();
1310 	dp = get_dp(net, ovs_header->dp_ifindex);
1311 	if (unlikely(!dp)) {
1312 		error = -ENODEV;
1313 		goto err_unlock_ovs;
1314 	}
1315 	/* Check that the flow exists. */
1316 	if (ufid_present)
1317 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &sfid);
1318 	else
1319 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1320 	if (unlikely(!flow)) {
1321 		error = -ENOENT;
1322 		goto err_unlock_ovs;
1323 	}
1324 
1325 	/* Update actions, if present. */
1326 	if (likely(acts)) {
1327 		old_acts = ovsl_dereference(flow->sf_acts);
1328 		rcu_assign_pointer(flow->sf_acts, acts);
1329 
1330 		if (unlikely(reply)) {
1331 			error = ovs_flow_cmd_fill_info(flow,
1332 						       ovs_header->dp_ifindex,
1333 						       reply, info->snd_portid,
1334 						       info->snd_seq, 0,
1335 						       OVS_FLOW_CMD_SET,
1336 						       ufid_flags);
1337 			BUG_ON(error < 0);
1338 		}
1339 	} else {
1340 		/* Could not alloc without acts before locking. */
1341 		reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex,
1342 						info, OVS_FLOW_CMD_SET, false,
1343 						ufid_flags);
1344 
1345 		if (IS_ERR(reply)) {
1346 			error = PTR_ERR(reply);
1347 			reply = NULL;
1348 			goto err_unlock_ovs;
1349 		}
1350 	}
1351 
1352 	/* Clear stats. */
1353 	if (a[OVS_FLOW_ATTR_CLEAR])
1354 		ovs_flow_stats_clear(flow);
1355 	ovs_unlock();
1356 
1357 	if (reply)
1358 		ovs_notify(&dp_flow_genl_family, reply, info);
1359 	if (old_acts)
1360 		ovs_nla_free_flow_actions_rcu(old_acts);
1361 
1362 	return 0;
1363 
1364 err_unlock_ovs:
1365 	ovs_unlock();
1366 	kfree_skb(reply);
1367 err_kfree_acts:
1368 	ovs_nla_free_flow_actions(acts);
1369 error:
1370 	return error;
1371 }
1372 
1373 static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info)
1374 {
1375 	struct nlattr **a = info->attrs;
1376 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1377 	struct net *net = sock_net(skb->sk);
1378 	struct sw_flow_key key;
1379 	struct sk_buff *reply;
1380 	struct sw_flow *flow;
1381 	struct datapath *dp;
1382 	struct sw_flow_match match;
1383 	struct sw_flow_id ufid;
1384 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1385 	int err = 0;
1386 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1387 	bool ufid_present;
1388 
1389 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1390 	if (a[OVS_FLOW_ATTR_KEY]) {
1391 		ovs_match_init(&match, &key, true, NULL);
1392 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY], NULL,
1393 					log);
1394 	} else if (!ufid_present) {
1395 		OVS_NLERR(log,
1396 			  "Flow get message rejected, Key attribute missing.");
1397 		err = -EINVAL;
1398 	}
1399 	if (err)
1400 		return err;
1401 
1402 	ovs_lock();
1403 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1404 	if (!dp) {
1405 		err = -ENODEV;
1406 		goto unlock;
1407 	}
1408 
1409 	if (ufid_present)
1410 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1411 	else
1412 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1413 	if (!flow) {
1414 		err = -ENOENT;
1415 		goto unlock;
1416 	}
1417 
1418 	reply = ovs_flow_cmd_build_info(flow, ovs_header->dp_ifindex, info,
1419 					OVS_FLOW_CMD_GET, true, ufid_flags);
1420 	if (IS_ERR(reply)) {
1421 		err = PTR_ERR(reply);
1422 		goto unlock;
1423 	}
1424 
1425 	ovs_unlock();
1426 	return genlmsg_reply(reply, info);
1427 unlock:
1428 	ovs_unlock();
1429 	return err;
1430 }
1431 
1432 static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info)
1433 {
1434 	struct nlattr **a = info->attrs;
1435 	struct ovs_header *ovs_header = genl_info_userhdr(info);
1436 	struct net *net = sock_net(skb->sk);
1437 	struct sw_flow_key key;
1438 	struct sk_buff *reply;
1439 	struct sw_flow *flow = NULL;
1440 	struct datapath *dp;
1441 	struct sw_flow_match match;
1442 	struct sw_flow_id ufid;
1443 	u32 ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1444 	int err;
1445 	bool log = !a[OVS_FLOW_ATTR_PROBE];
1446 	bool ufid_present;
1447 
1448 	ufid_present = ovs_nla_get_ufid(&ufid, a[OVS_FLOW_ATTR_UFID], log);
1449 	if (a[OVS_FLOW_ATTR_KEY]) {
1450 		ovs_match_init(&match, &key, true, NULL);
1451 		err = ovs_nla_get_match(net, &match, a[OVS_FLOW_ATTR_KEY],
1452 					NULL, log);
1453 		if (unlikely(err))
1454 			return err;
1455 	}
1456 
1457 	ovs_lock();
1458 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
1459 	if (unlikely(!dp)) {
1460 		err = -ENODEV;
1461 		goto unlock;
1462 	}
1463 
1464 	if (unlikely(!a[OVS_FLOW_ATTR_KEY] && !ufid_present)) {
1465 		err = ovs_flow_tbl_flush(&dp->table);
1466 		goto unlock;
1467 	}
1468 
1469 	if (ufid_present)
1470 		flow = ovs_flow_tbl_lookup_ufid(&dp->table, &ufid);
1471 	else
1472 		flow = ovs_flow_tbl_lookup_exact(&dp->table, &match);
1473 	if (unlikely(!flow)) {
1474 		err = -ENOENT;
1475 		goto unlock;
1476 	}
1477 
1478 	reply = ovs_flow_cmd_alloc_info(ovsl_dereference(flow->sf_acts),
1479 					&flow->id, info, false, ufid_flags);
1480 	if (IS_ERR(reply)) {
1481 		netlink_set_err(sock_net(skb->sk)->genl_sock, 0, 0,
1482 				PTR_ERR(reply));
1483 		reply = NULL;
1484 	}
1485 
1486 	if (likely(reply)) {
1487 		err = ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex,
1488 					     reply, info->snd_portid,
1489 					     info->snd_seq, 0,
1490 					     OVS_FLOW_CMD_DEL, ufid_flags);
1491 		if (WARN_ON_ONCE(err < 0)) {
1492 			kfree_skb(reply);
1493 			reply = NULL;
1494 		}
1495 	}
1496 	/* Removal has to happen after ovs_flow_cmd_fill_info(), as it uses
1497 	 * the flow->mask that can be scheduled to be freed by the
1498 	 * ovs_flow_tbl_remove() and we're not holding the RCU read lock.
1499 	 */
1500 	ovs_flow_tbl_remove(&dp->table, flow);
1501 	ovs_unlock();
1502 
1503 	if (likely(reply))
1504 		ovs_notify(&dp_flow_genl_family, reply, info);
1505 
1506 	ovs_flow_free(flow, true);
1507 	return 0;
1508 unlock:
1509 	ovs_unlock();
1510 	return err;
1511 }
1512 
1513 static int ovs_flow_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
1514 {
1515 	struct nlattr *a[__OVS_FLOW_ATTR_MAX];
1516 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
1517 	struct table_instance *ti;
1518 	struct datapath *dp;
1519 	u32 ufid_flags;
1520 	int err;
1521 
1522 	err = genlmsg_parse_deprecated(cb->nlh, &dp_flow_genl_family, a,
1523 				       OVS_FLOW_ATTR_MAX, flow_policy, NULL);
1524 	if (err)
1525 		return err;
1526 	ufid_flags = ovs_nla_get_ufid_flags(a[OVS_FLOW_ATTR_UFID_FLAGS]);
1527 
1528 	rcu_read_lock();
1529 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
1530 	if (!dp) {
1531 		rcu_read_unlock();
1532 		return -ENODEV;
1533 	}
1534 
1535 	ti = rcu_dereference(dp->table.ti);
1536 	for (;;) {
1537 		struct sw_flow *flow;
1538 		u32 bucket, obj;
1539 
1540 		bucket = cb->args[0];
1541 		obj = cb->args[1];
1542 		flow = ovs_flow_tbl_dump_next(ti, &bucket, &obj);
1543 		if (!flow)
1544 			break;
1545 
1546 		if (ovs_flow_cmd_fill_info(flow, ovs_header->dp_ifindex, skb,
1547 					   NETLINK_CB(cb->skb).portid,
1548 					   cb->nlh->nlmsg_seq, NLM_F_MULTI,
1549 					   OVS_FLOW_CMD_GET, ufid_flags) < 0)
1550 			break;
1551 
1552 		cb->args[0] = bucket;
1553 		cb->args[1] = obj;
1554 	}
1555 	rcu_read_unlock();
1556 	return skb->len;
1557 }
1558 
1559 static const struct nla_policy flow_policy[OVS_FLOW_ATTR_MAX + 1] = {
1560 	[OVS_FLOW_ATTR_KEY] = { .type = NLA_NESTED },
1561 	[OVS_FLOW_ATTR_MASK] = { .type = NLA_NESTED },
1562 	[OVS_FLOW_ATTR_ACTIONS] = { .type = NLA_NESTED },
1563 	[OVS_FLOW_ATTR_CLEAR] = { .type = NLA_FLAG },
1564 	[OVS_FLOW_ATTR_PROBE] = { .type = NLA_FLAG },
1565 	[OVS_FLOW_ATTR_UFID] = { .type = NLA_UNSPEC, .len = 1 },
1566 	[OVS_FLOW_ATTR_UFID_FLAGS] = { .type = NLA_U32 },
1567 };
1568 
1569 static const struct genl_small_ops dp_flow_genl_ops[] = {
1570 	{ .cmd = OVS_FLOW_CMD_NEW,
1571 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1572 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1573 	  .doit = ovs_flow_cmd_new
1574 	},
1575 	{ .cmd = OVS_FLOW_CMD_DEL,
1576 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1577 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1578 	  .doit = ovs_flow_cmd_del
1579 	},
1580 	{ .cmd = OVS_FLOW_CMD_GET,
1581 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1582 	  .flags = 0,		    /* OK for unprivileged users. */
1583 	  .doit = ovs_flow_cmd_get,
1584 	  .dumpit = ovs_flow_cmd_dump
1585 	},
1586 	{ .cmd = OVS_FLOW_CMD_SET,
1587 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1588 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
1589 	  .doit = ovs_flow_cmd_set,
1590 	},
1591 };
1592 
1593 static struct genl_family dp_flow_genl_family __ro_after_init = {
1594 	.hdrsize = sizeof(struct ovs_header),
1595 	.name = OVS_FLOW_FAMILY,
1596 	.version = OVS_FLOW_VERSION,
1597 	.maxattr = OVS_FLOW_ATTR_MAX,
1598 	.policy = flow_policy,
1599 	.netnsok = true,
1600 	.parallel_ops = true,
1601 	.small_ops = dp_flow_genl_ops,
1602 	.n_small_ops = ARRAY_SIZE(dp_flow_genl_ops),
1603 	.resv_start_op = OVS_FLOW_CMD_SET + 1,
1604 	.mcgrps = &ovs_dp_flow_multicast_group,
1605 	.n_mcgrps = 1,
1606 	.module = THIS_MODULE,
1607 };
1608 
1609 static size_t ovs_dp_cmd_msg_size(void)
1610 {
1611 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
1612 
1613 	msgsize += nla_total_size(IFNAMSIZ);
1614 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_stats));
1615 	msgsize += nla_total_size_64bit(sizeof(struct ovs_dp_megaflow_stats));
1616 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_USER_FEATURES */
1617 	msgsize += nla_total_size(sizeof(u32)); /* OVS_DP_ATTR_MASKS_CACHE_SIZE */
1618 	msgsize += nla_total_size(sizeof(u32) * nr_cpu_ids); /* OVS_DP_ATTR_PER_CPU_PIDS */
1619 
1620 	return msgsize;
1621 }
1622 
1623 /* Called with ovs_mutex. */
1624 static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb,
1625 				u32 portid, u32 seq, u32 flags, u8 cmd)
1626 {
1627 	struct ovs_header *ovs_header;
1628 	struct ovs_dp_stats dp_stats;
1629 	struct ovs_dp_megaflow_stats dp_megaflow_stats;
1630 	struct dp_nlsk_pids *pids = ovsl_dereference(dp->upcall_portids);
1631 	int err, pids_len;
1632 
1633 	ovs_header = genlmsg_put(skb, portid, seq, &dp_datapath_genl_family,
1634 				 flags, cmd);
1635 	if (!ovs_header)
1636 		goto error;
1637 
1638 	ovs_header->dp_ifindex = get_dpifindex(dp);
1639 
1640 	err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp));
1641 	if (err)
1642 		goto nla_put_failure;
1643 
1644 	get_dp_stats(dp, &dp_stats, &dp_megaflow_stats);
1645 	if (nla_put_64bit(skb, OVS_DP_ATTR_STATS, sizeof(struct ovs_dp_stats),
1646 			  &dp_stats, OVS_DP_ATTR_PAD))
1647 		goto nla_put_failure;
1648 
1649 	if (nla_put_64bit(skb, OVS_DP_ATTR_MEGAFLOW_STATS,
1650 			  sizeof(struct ovs_dp_megaflow_stats),
1651 			  &dp_megaflow_stats, OVS_DP_ATTR_PAD))
1652 		goto nla_put_failure;
1653 
1654 	if (nla_put_u32(skb, OVS_DP_ATTR_USER_FEATURES, dp->user_features))
1655 		goto nla_put_failure;
1656 
1657 	if (nla_put_u32(skb, OVS_DP_ATTR_MASKS_CACHE_SIZE,
1658 			ovs_flow_tbl_masks_cache_size(&dp->table)))
1659 		goto nla_put_failure;
1660 
1661 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU && pids) {
1662 		pids_len = min(pids->n_pids, nr_cpu_ids) * sizeof(u32);
1663 		if (nla_put(skb, OVS_DP_ATTR_PER_CPU_PIDS, pids_len, &pids->pids))
1664 			goto nla_put_failure;
1665 	}
1666 
1667 	genlmsg_end(skb, ovs_header);
1668 	return 0;
1669 
1670 nla_put_failure:
1671 	genlmsg_cancel(skb, ovs_header);
1672 error:
1673 	return -EMSGSIZE;
1674 }
1675 
1676 static struct sk_buff *ovs_dp_cmd_alloc_info(void)
1677 {
1678 	return genlmsg_new(ovs_dp_cmd_msg_size(), GFP_KERNEL);
1679 }
1680 
1681 /* Called with rcu_read_lock or ovs_mutex. */
1682 static struct datapath *lookup_datapath(struct net *net,
1683 					const struct ovs_header *ovs_header,
1684 					struct nlattr *a[OVS_DP_ATTR_MAX + 1])
1685 {
1686 	struct datapath *dp;
1687 
1688 	if (!a[OVS_DP_ATTR_NAME])
1689 		dp = get_dp(net, ovs_header->dp_ifindex);
1690 	else {
1691 		struct vport *vport;
1692 
1693 		vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME]));
1694 		dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL;
1695 	}
1696 	return dp ? dp : ERR_PTR(-ENODEV);
1697 }
1698 
1699 static void ovs_dp_reset_user_features(struct sk_buff *skb,
1700 				       struct genl_info *info)
1701 {
1702 	struct datapath *dp;
1703 
1704 	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1705 			     info->attrs);
1706 	if (IS_ERR(dp))
1707 		return;
1708 
1709 	pr_warn("%s: Dropping previously announced user features\n",
1710 		ovs_dp_name(dp));
1711 	dp->user_features = 0;
1712 }
1713 
1714 static int ovs_dp_set_upcall_portids(struct datapath *dp,
1715 			      const struct nlattr *ids)
1716 {
1717 	struct dp_nlsk_pids *old, *dp_nlsk_pids;
1718 
1719 	if (!nla_len(ids) || nla_len(ids) % sizeof(u32))
1720 		return -EINVAL;
1721 
1722 	old = ovsl_dereference(dp->upcall_portids);
1723 
1724 	dp_nlsk_pids = kmalloc(sizeof(*dp_nlsk_pids) + nla_len(ids),
1725 			       GFP_KERNEL);
1726 	if (!dp_nlsk_pids)
1727 		return -ENOMEM;
1728 
1729 	dp_nlsk_pids->n_pids = nla_len(ids) / sizeof(u32);
1730 	nla_memcpy(dp_nlsk_pids->pids, ids, nla_len(ids));
1731 
1732 	rcu_assign_pointer(dp->upcall_portids, dp_nlsk_pids);
1733 
1734 	kfree_rcu(old, rcu);
1735 
1736 	return 0;
1737 }
1738 
1739 u32 ovs_dp_get_upcall_portid(const struct datapath *dp, uint32_t cpu_id)
1740 {
1741 	struct dp_nlsk_pids *dp_nlsk_pids;
1742 
1743 	dp_nlsk_pids = rcu_dereference(dp->upcall_portids);
1744 
1745 	if (dp_nlsk_pids) {
1746 		if (cpu_id < dp_nlsk_pids->n_pids) {
1747 			return dp_nlsk_pids->pids[cpu_id];
1748 		} else if (dp_nlsk_pids->n_pids > 0 &&
1749 			   cpu_id >= dp_nlsk_pids->n_pids) {
1750 			/* If the number of netlink PIDs is mismatched with
1751 			 * the number of CPUs as seen by the kernel, log this
1752 			 * and send the upcall to an arbitrary socket (0) in
1753 			 * order to not drop packets
1754 			 */
1755 			pr_info_ratelimited("cpu_id mismatch with handler threads");
1756 			return dp_nlsk_pids->pids[cpu_id %
1757 						  dp_nlsk_pids->n_pids];
1758 		} else {
1759 			return 0;
1760 		}
1761 	} else {
1762 		return 0;
1763 	}
1764 }
1765 
1766 static int ovs_dp_change(struct datapath *dp, struct nlattr *a[])
1767 {
1768 	u32 user_features = 0, old_features = dp->user_features;
1769 	int err;
1770 
1771 	if (a[OVS_DP_ATTR_USER_FEATURES]) {
1772 		user_features = nla_get_u32(a[OVS_DP_ATTR_USER_FEATURES]);
1773 
1774 		if (user_features & ~(OVS_DP_F_VPORT_PIDS |
1775 				      OVS_DP_F_UNALIGNED |
1776 				      OVS_DP_F_TC_RECIRC_SHARING |
1777 				      OVS_DP_F_DISPATCH_UPCALL_PER_CPU))
1778 			return -EOPNOTSUPP;
1779 
1780 #if !IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
1781 		if (user_features & OVS_DP_F_TC_RECIRC_SHARING)
1782 			return -EOPNOTSUPP;
1783 #endif
1784 	}
1785 
1786 	if (a[OVS_DP_ATTR_MASKS_CACHE_SIZE]) {
1787 		int err;
1788 		u32 cache_size;
1789 
1790 		cache_size = nla_get_u32(a[OVS_DP_ATTR_MASKS_CACHE_SIZE]);
1791 		err = ovs_flow_tbl_masks_cache_resize(&dp->table, cache_size);
1792 		if (err)
1793 			return err;
1794 	}
1795 
1796 	dp->user_features = user_features;
1797 
1798 	if (dp->user_features & OVS_DP_F_DISPATCH_UPCALL_PER_CPU &&
1799 	    a[OVS_DP_ATTR_PER_CPU_PIDS]) {
1800 		/* Upcall Netlink Port IDs have been updated */
1801 		err = ovs_dp_set_upcall_portids(dp,
1802 						a[OVS_DP_ATTR_PER_CPU_PIDS]);
1803 		if (err)
1804 			return err;
1805 	}
1806 
1807 	if ((dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1808 	    !(old_features & OVS_DP_F_TC_RECIRC_SHARING))
1809 		tc_skb_ext_tc_enable();
1810 	else if (!(dp->user_features & OVS_DP_F_TC_RECIRC_SHARING) &&
1811 		 (old_features & OVS_DP_F_TC_RECIRC_SHARING))
1812 		tc_skb_ext_tc_disable();
1813 
1814 	return 0;
1815 }
1816 
1817 static int ovs_dp_stats_init(struct datapath *dp)
1818 {
1819 	dp->stats_percpu = netdev_alloc_pcpu_stats(struct dp_stats_percpu);
1820 	if (!dp->stats_percpu)
1821 		return -ENOMEM;
1822 
1823 	return 0;
1824 }
1825 
1826 static int ovs_dp_vport_init(struct datapath *dp)
1827 {
1828 	int i;
1829 
1830 	dp->ports = kmalloc_objs(struct hlist_head, DP_VPORT_HASH_BUCKETS);
1831 	if (!dp->ports)
1832 		return -ENOMEM;
1833 
1834 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++)
1835 		INIT_HLIST_HEAD(&dp->ports[i]);
1836 
1837 	return 0;
1838 }
1839 
1840 static int ovs_dp_cmd_new(struct sk_buff *skb, struct genl_info *info)
1841 {
1842 	struct nlattr **a = info->attrs;
1843 	struct vport_parms parms;
1844 	struct sk_buff *reply;
1845 	struct datapath *dp;
1846 	struct vport *vport;
1847 	struct ovs_net *ovs_net;
1848 	int err;
1849 
1850 	err = -EINVAL;
1851 	if (!a[OVS_DP_ATTR_NAME] || !a[OVS_DP_ATTR_UPCALL_PID])
1852 		goto err;
1853 
1854 	reply = ovs_dp_cmd_alloc_info();
1855 	if (!reply)
1856 		return -ENOMEM;
1857 
1858 	err = -ENOMEM;
1859 	dp = kzalloc_obj(*dp);
1860 	if (dp == NULL)
1861 		goto err_destroy_reply;
1862 
1863 	ovs_dp_set_net(dp, sock_net(skb->sk));
1864 
1865 	/* Allocate table. */
1866 	err = ovs_flow_tbl_init(&dp->table);
1867 	if (err)
1868 		goto err_destroy_dp;
1869 
1870 	err = ovs_dp_stats_init(dp);
1871 	if (err)
1872 		goto err_destroy_table;
1873 
1874 	err = ovs_dp_vport_init(dp);
1875 	if (err)
1876 		goto err_destroy_stats;
1877 
1878 	err = ovs_meters_init(dp);
1879 	if (err)
1880 		goto err_destroy_ports;
1881 
1882 	/* Set up our datapath device. */
1883 	parms.name = nla_data(a[OVS_DP_ATTR_NAME]);
1884 	parms.type = OVS_VPORT_TYPE_INTERNAL;
1885 	parms.dp = dp;
1886 	parms.port_no = OVSP_LOCAL;
1887 	parms.upcall_portids = a[OVS_DP_ATTR_UPCALL_PID];
1888 	parms.desired_ifindex = nla_get_s32_default(a[OVS_DP_ATTR_IFINDEX], 0);
1889 
1890 	/* So far only local changes have been made, now need the lock. */
1891 	ovs_lock();
1892 
1893 	err = ovs_dp_change(dp, a);
1894 	if (err)
1895 		goto err_unlock_and_destroy_meters;
1896 
1897 	vport = new_vport(&parms);
1898 	if (IS_ERR(vport)) {
1899 		err = PTR_ERR(vport);
1900 		if (err == -EBUSY)
1901 			err = -EEXIST;
1902 
1903 		if (err == -EEXIST) {
1904 			/* An outdated user space instance that does not understand
1905 			 * the concept of user_features has attempted to create a new
1906 			 * datapath and is likely to reuse it. Drop all user features.
1907 			 */
1908 			if (info->genlhdr->version < OVS_DP_VER_FEATURES)
1909 				ovs_dp_reset_user_features(skb, info);
1910 		}
1911 
1912 		goto err_destroy_portids;
1913 	}
1914 
1915 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
1916 				   info->snd_seq, 0, OVS_DP_CMD_NEW);
1917 	BUG_ON(err < 0);
1918 
1919 	ovs_net = net_generic(ovs_dp_get_net(dp), ovs_net_id);
1920 	list_add_tail_rcu(&dp->list_node, &ovs_net->dps);
1921 
1922 	ovs_unlock();
1923 
1924 	ovs_notify(&dp_datapath_genl_family, reply, info);
1925 	return 0;
1926 
1927 err_destroy_portids:
1928 	kfree(rcu_dereference_raw(dp->upcall_portids));
1929 err_unlock_and_destroy_meters:
1930 	ovs_unlock();
1931 	ovs_meters_exit(dp);
1932 err_destroy_ports:
1933 	kfree(dp->ports);
1934 err_destroy_stats:
1935 	free_percpu(dp->stats_percpu);
1936 err_destroy_table:
1937 	ovs_flow_tbl_destroy(&dp->table);
1938 err_destroy_dp:
1939 	kfree(dp);
1940 err_destroy_reply:
1941 	kfree_skb(reply);
1942 err:
1943 	return err;
1944 }
1945 
1946 /* Called with ovs_mutex. */
1947 static void __dp_destroy(struct datapath *dp)
1948 {
1949 	struct flow_table *table = &dp->table;
1950 	int i;
1951 
1952 	if (dp->user_features & OVS_DP_F_TC_RECIRC_SHARING)
1953 		tc_skb_ext_tc_disable();
1954 
1955 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
1956 		struct vport *vport;
1957 		struct hlist_node *n;
1958 
1959 		hlist_for_each_entry_safe(vport, n, &dp->ports[i], dp_hash_node)
1960 			if (vport->port_no != OVSP_LOCAL)
1961 				ovs_dp_detach_port(vport);
1962 	}
1963 
1964 	list_del_rcu(&dp->list_node);
1965 
1966 	/* OVSP_LOCAL is datapath internal port. We need to make sure that
1967 	 * all ports in datapath are destroyed first before freeing datapath.
1968 	 */
1969 	ovs_dp_detach_port(ovs_vport_ovsl(dp, OVSP_LOCAL));
1970 
1971 	/* Flush sw_flow in the tables. RCU cb only releases resource
1972 	 * such as dp, ports and tables. That may avoid some issues
1973 	 * such as RCU usage warning.
1974 	 */
1975 	table_instance_flow_flush(table, ovsl_dereference(table->ti),
1976 				  ovsl_dereference(table->ufid_ti));
1977 
1978 	/* RCU destroy the ports, meters and flow tables. */
1979 	call_rcu(&dp->rcu, destroy_dp_rcu);
1980 }
1981 
1982 static int ovs_dp_cmd_del(struct sk_buff *skb, struct genl_info *info)
1983 {
1984 	struct sk_buff *reply;
1985 	struct datapath *dp;
1986 	int err;
1987 
1988 	reply = ovs_dp_cmd_alloc_info();
1989 	if (!reply)
1990 		return -ENOMEM;
1991 
1992 	ovs_lock();
1993 	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
1994 			     info->attrs);
1995 	err = PTR_ERR(dp);
1996 	if (IS_ERR(dp))
1997 		goto err_unlock_free;
1998 
1999 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2000 				   info->snd_seq, 0, OVS_DP_CMD_DEL);
2001 	BUG_ON(err < 0);
2002 
2003 	__dp_destroy(dp);
2004 	ovs_unlock();
2005 
2006 	ovs_notify(&dp_datapath_genl_family, reply, info);
2007 
2008 	return 0;
2009 
2010 err_unlock_free:
2011 	ovs_unlock();
2012 	kfree_skb(reply);
2013 	return err;
2014 }
2015 
2016 static int ovs_dp_cmd_set(struct sk_buff *skb, struct genl_info *info)
2017 {
2018 	struct sk_buff *reply;
2019 	struct datapath *dp;
2020 	int err;
2021 
2022 	reply = ovs_dp_cmd_alloc_info();
2023 	if (!reply)
2024 		return -ENOMEM;
2025 
2026 	ovs_lock();
2027 	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2028 			     info->attrs);
2029 	err = PTR_ERR(dp);
2030 	if (IS_ERR(dp))
2031 		goto err_unlock_free;
2032 
2033 	err = ovs_dp_change(dp, info->attrs);
2034 	if (err)
2035 		goto err_unlock_free;
2036 
2037 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2038 				   info->snd_seq, 0, OVS_DP_CMD_SET);
2039 	BUG_ON(err < 0);
2040 
2041 	ovs_unlock();
2042 	ovs_notify(&dp_datapath_genl_family, reply, info);
2043 
2044 	return 0;
2045 
2046 err_unlock_free:
2047 	ovs_unlock();
2048 	kfree_skb(reply);
2049 	return err;
2050 }
2051 
2052 static int ovs_dp_cmd_get(struct sk_buff *skb, struct genl_info *info)
2053 {
2054 	struct sk_buff *reply;
2055 	struct datapath *dp;
2056 	int err;
2057 
2058 	reply = ovs_dp_cmd_alloc_info();
2059 	if (!reply)
2060 		return -ENOMEM;
2061 
2062 	ovs_lock();
2063 	dp = lookup_datapath(sock_net(skb->sk), genl_info_userhdr(info),
2064 			     info->attrs);
2065 	if (IS_ERR(dp)) {
2066 		err = PTR_ERR(dp);
2067 		goto err_unlock_free;
2068 	}
2069 	err = ovs_dp_cmd_fill_info(dp, reply, info->snd_portid,
2070 				   info->snd_seq, 0, OVS_DP_CMD_GET);
2071 	BUG_ON(err < 0);
2072 	ovs_unlock();
2073 
2074 	return genlmsg_reply(reply, info);
2075 
2076 err_unlock_free:
2077 	ovs_unlock();
2078 	kfree_skb(reply);
2079 	return err;
2080 }
2081 
2082 static int ovs_dp_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2083 {
2084 	struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2085 	struct datapath *dp;
2086 	int skip = cb->args[0];
2087 	int i = 0;
2088 
2089 	ovs_lock();
2090 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2091 		if (i >= skip &&
2092 		    ovs_dp_cmd_fill_info(dp, skb, NETLINK_CB(cb->skb).portid,
2093 					 cb->nlh->nlmsg_seq, NLM_F_MULTI,
2094 					 OVS_DP_CMD_GET) < 0)
2095 			break;
2096 		i++;
2097 	}
2098 	ovs_unlock();
2099 
2100 	cb->args[0] = i;
2101 
2102 	return skb->len;
2103 }
2104 
2105 static const struct nla_policy datapath_policy[OVS_DP_ATTR_MAX + 1] = {
2106 	[OVS_DP_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2107 	[OVS_DP_ATTR_UPCALL_PID] = { .type = NLA_U32 },
2108 	[OVS_DP_ATTR_USER_FEATURES] = { .type = NLA_U32 },
2109 	[OVS_DP_ATTR_MASKS_CACHE_SIZE] =  NLA_POLICY_RANGE(NLA_U32, 0,
2110 		PCPU_MIN_UNIT_SIZE / sizeof(struct mask_cache_entry)),
2111 	[OVS_DP_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2112 };
2113 
2114 static const struct genl_small_ops dp_datapath_genl_ops[] = {
2115 	{ .cmd = OVS_DP_CMD_NEW,
2116 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2117 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2118 	  .doit = ovs_dp_cmd_new
2119 	},
2120 	{ .cmd = OVS_DP_CMD_DEL,
2121 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2122 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2123 	  .doit = ovs_dp_cmd_del
2124 	},
2125 	{ .cmd = OVS_DP_CMD_GET,
2126 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2127 	  .flags = 0,		    /* OK for unprivileged users. */
2128 	  .doit = ovs_dp_cmd_get,
2129 	  .dumpit = ovs_dp_cmd_dump
2130 	},
2131 	{ .cmd = OVS_DP_CMD_SET,
2132 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2133 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2134 	  .doit = ovs_dp_cmd_set,
2135 	},
2136 };
2137 
2138 static struct genl_family dp_datapath_genl_family __ro_after_init = {
2139 	.hdrsize = sizeof(struct ovs_header),
2140 	.name = OVS_DATAPATH_FAMILY,
2141 	.version = OVS_DATAPATH_VERSION,
2142 	.maxattr = OVS_DP_ATTR_MAX,
2143 	.policy = datapath_policy,
2144 	.netnsok = true,
2145 	.parallel_ops = true,
2146 	.small_ops = dp_datapath_genl_ops,
2147 	.n_small_ops = ARRAY_SIZE(dp_datapath_genl_ops),
2148 	.resv_start_op = OVS_DP_CMD_SET + 1,
2149 	.mcgrps = &ovs_dp_datapath_multicast_group,
2150 	.n_mcgrps = 1,
2151 	.module = THIS_MODULE,
2152 };
2153 
2154 /* Called with ovs_mutex or RCU read lock. */
2155 static int ovs_vport_cmd_fill_info(struct vport *vport, struct sk_buff *skb,
2156 				   struct net *net, u32 portid, u32 seq,
2157 				   u32 flags, u8 cmd, gfp_t gfp)
2158 {
2159 	struct ovs_header *ovs_header;
2160 	struct ovs_vport_stats vport_stats;
2161 	struct net *net_vport;
2162 	int err;
2163 
2164 	ovs_header = genlmsg_put(skb, portid, seq, &dp_vport_genl_family,
2165 				 flags, cmd);
2166 	if (!ovs_header)
2167 		return -EMSGSIZE;
2168 
2169 	ovs_header->dp_ifindex = get_dpifindex(vport->dp);
2170 
2171 	if (nla_put_u32(skb, OVS_VPORT_ATTR_PORT_NO, vport->port_no) ||
2172 	    nla_put_u32(skb, OVS_VPORT_ATTR_TYPE, vport->ops->type) ||
2173 	    nla_put_string(skb, OVS_VPORT_ATTR_NAME,
2174 			   ovs_vport_name(vport)) ||
2175 	    nla_put_u32(skb, OVS_VPORT_ATTR_IFINDEX, vport->dev->ifindex))
2176 		goto nla_put_failure;
2177 
2178 	rcu_read_lock();
2179 	net_vport = dev_net_rcu(vport->dev);
2180 	if (!net_eq(net, net_vport)) {
2181 		int id = peernet2id_alloc(net, net_vport, GFP_ATOMIC);
2182 
2183 		if (nla_put_s32(skb, OVS_VPORT_ATTR_NETNSID, id))
2184 			goto nla_put_failure_unlock;
2185 	}
2186 	rcu_read_unlock();
2187 
2188 	ovs_vport_get_stats(vport, &vport_stats);
2189 	if (nla_put_64bit(skb, OVS_VPORT_ATTR_STATS,
2190 			  sizeof(struct ovs_vport_stats), &vport_stats,
2191 			  OVS_VPORT_ATTR_PAD))
2192 		goto nla_put_failure;
2193 
2194 	if (ovs_vport_get_upcall_stats(vport, skb))
2195 		goto nla_put_failure;
2196 
2197 	if (ovs_vport_get_upcall_portids(vport, skb))
2198 		goto nla_put_failure;
2199 
2200 	genlmsg_end(skb, ovs_header);
2201 	return 0;
2202 
2203 nla_put_failure_unlock:
2204 	rcu_read_unlock();
2205 nla_put_failure:
2206 	err = -EMSGSIZE;
2207 	genlmsg_cancel(skb, ovs_header);
2208 	return err;
2209 }
2210 
2211 static size_t ovs_vport_cmd_msg_size(void)
2212 {
2213 	size_t msgsize = NLMSG_ALIGN(sizeof(struct ovs_header));
2214 
2215 	msgsize += nla_total_size(sizeof(u32)); /* OVS_VPORT_ATTR_PORT_NO */
2216 	msgsize += nla_total_size(sizeof(u32)); /* OVS_VPORT_ATTR_TYPE */
2217 	msgsize += nla_total_size(IFNAMSIZ);    /* OVS_VPORT_ATTR_NAME */
2218 	msgsize += nla_total_size(sizeof(u32)); /* OVS_VPORT_ATTR_IFINDEX */
2219 	msgsize += nla_total_size(sizeof(s32)); /* OVS_VPORT_ATTR_NETNSID */
2220 
2221 	/* OVS_VPORT_ATTR_STATS */
2222 	msgsize += nla_total_size_64bit(sizeof(struct ovs_vport_stats));
2223 
2224 	/* OVS_VPORT_ATTR_UPCALL_STATS(OVS_VPORT_UPCALL_ATTR_SUCCESS +
2225 	 *                             OVS_VPORT_UPCALL_ATTR_FAIL)
2226 	 */
2227 	msgsize += nla_total_size(nla_total_size_64bit(sizeof(u64)) +
2228 				  nla_total_size_64bit(sizeof(u64)));
2229 
2230 	/* OVS_VPORT_ATTR_UPCALL_PID */
2231 	msgsize += nla_total_size(nr_cpu_ids * sizeof(u32));
2232 
2233 	return msgsize;
2234 }
2235 
2236 static struct sk_buff *ovs_vport_cmd_alloc_info(void)
2237 {
2238 	return genlmsg_new(ovs_vport_cmd_msg_size(), GFP_KERNEL);
2239 }
2240 
2241 /* Called with ovs_mutex, only via ovs_dp_notify_wq(). */
2242 struct sk_buff *ovs_vport_cmd_build_info(struct vport *vport, struct net *net,
2243 					 u32 portid, u32 seq, u8 cmd)
2244 {
2245 	struct sk_buff *skb;
2246 	int retval;
2247 
2248 	skb = ovs_vport_cmd_alloc_info();
2249 	if (!skb)
2250 		return ERR_PTR(-ENOMEM);
2251 
2252 	retval = ovs_vport_cmd_fill_info(vport, skb, net, portid, seq, 0, cmd,
2253 					 GFP_KERNEL);
2254 	BUG_ON(retval < 0);
2255 
2256 	return skb;
2257 }
2258 
2259 /* Called with ovs_mutex or RCU read lock. */
2260 static struct vport *lookup_vport(struct net *net,
2261 				  const struct ovs_header *ovs_header,
2262 				  struct nlattr *a[OVS_VPORT_ATTR_MAX + 1])
2263 {
2264 	struct datapath *dp;
2265 	struct vport *vport;
2266 
2267 	if (a[OVS_VPORT_ATTR_IFINDEX])
2268 		return ERR_PTR(-EOPNOTSUPP);
2269 	if (a[OVS_VPORT_ATTR_NAME]) {
2270 		vport = ovs_vport_locate(net, nla_data(a[OVS_VPORT_ATTR_NAME]));
2271 		if (!vport)
2272 			return ERR_PTR(-ENODEV);
2273 		if (ovs_header->dp_ifindex &&
2274 		    ovs_header->dp_ifindex != get_dpifindex(vport->dp))
2275 			return ERR_PTR(-ENODEV);
2276 		return vport;
2277 	} else if (a[OVS_VPORT_ATTR_PORT_NO]) {
2278 		u32 port_no = nla_get_u32(a[OVS_VPORT_ATTR_PORT_NO]);
2279 
2280 		if (port_no >= DP_MAX_PORTS)
2281 			return ERR_PTR(-EFBIG);
2282 
2283 		dp = get_dp(net, ovs_header->dp_ifindex);
2284 		if (!dp)
2285 			return ERR_PTR(-ENODEV);
2286 
2287 		vport = ovs_vport_ovsl_rcu(dp, port_no);
2288 		if (!vport)
2289 			return ERR_PTR(-ENODEV);
2290 		return vport;
2291 	} else
2292 		return ERR_PTR(-EINVAL);
2293 
2294 }
2295 
2296 static unsigned int ovs_get_max_headroom(struct datapath *dp)
2297 {
2298 	unsigned int dev_headroom, max_headroom = 0;
2299 	struct net_device *dev;
2300 	struct vport *vport;
2301 	int i;
2302 
2303 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2304 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2305 					 lockdep_ovsl_is_held()) {
2306 			dev = vport->dev;
2307 			dev_headroom = netdev_get_fwd_headroom(dev);
2308 			if (dev_headroom > max_headroom)
2309 				max_headroom = dev_headroom;
2310 		}
2311 	}
2312 
2313 	return max_headroom;
2314 }
2315 
2316 /* Called with ovs_mutex */
2317 static void ovs_update_headroom(struct datapath *dp, unsigned int new_headroom)
2318 {
2319 	struct vport *vport;
2320 	int i;
2321 
2322 	dp->max_headroom = new_headroom;
2323 	for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2324 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node,
2325 					 lockdep_ovsl_is_held())
2326 			netdev_set_rx_headroom(vport->dev, new_headroom);
2327 	}
2328 }
2329 
2330 static int ovs_vport_cmd_new(struct sk_buff *skb, struct genl_info *info)
2331 {
2332 	struct nlattr **a = info->attrs;
2333 	struct ovs_header *ovs_header = genl_info_userhdr(info);
2334 	struct vport_parms parms;
2335 	struct sk_buff *reply;
2336 	struct vport *vport;
2337 	struct datapath *dp;
2338 	unsigned int new_headroom;
2339 	u32 port_no;
2340 	int err;
2341 
2342 	if (!a[OVS_VPORT_ATTR_NAME] || !a[OVS_VPORT_ATTR_TYPE] ||
2343 	    !a[OVS_VPORT_ATTR_UPCALL_PID])
2344 		return -EINVAL;
2345 
2346 	parms.type = nla_get_u32(a[OVS_VPORT_ATTR_TYPE]);
2347 
2348 	if (a[OVS_VPORT_ATTR_IFINDEX] && parms.type != OVS_VPORT_TYPE_INTERNAL)
2349 		return -EOPNOTSUPP;
2350 
2351 	port_no = nla_get_u32_default(a[OVS_VPORT_ATTR_PORT_NO], 0);
2352 	if (port_no >= DP_MAX_PORTS)
2353 		return -EFBIG;
2354 
2355 	reply = ovs_vport_cmd_alloc_info();
2356 	if (!reply)
2357 		return -ENOMEM;
2358 
2359 	ovs_lock();
2360 	dp = get_dp(sock_net(skb->sk), ovs_header->dp_ifindex);
2361 	err = -ENODEV;
2362 	if (!dp)
2363 		goto exit_unlock_free;
2364 
2365 	if (port_no) {
2366 		vport = ovs_vport_ovsl(dp, port_no);
2367 		err = -EBUSY;
2368 		if (vport)
2369 			goto exit_unlock_free;
2370 	} else {
2371 		for (port_no = 1; ; port_no++) {
2372 			if (port_no >= DP_MAX_PORTS) {
2373 				err = -EFBIG;
2374 				goto exit_unlock_free;
2375 			}
2376 			vport = ovs_vport_ovsl(dp, port_no);
2377 			if (!vport)
2378 				break;
2379 		}
2380 	}
2381 
2382 	parms.name = nla_data(a[OVS_VPORT_ATTR_NAME]);
2383 	parms.dp = dp;
2384 	parms.port_no = port_no;
2385 	parms.upcall_portids = a[OVS_VPORT_ATTR_UPCALL_PID];
2386 	parms.desired_ifindex = nla_get_s32_default(a[OVS_VPORT_ATTR_IFINDEX],
2387 						    0);
2388 
2389 	vport = new_vport(&parms);
2390 	err = PTR_ERR(vport);
2391 	if (IS_ERR(vport))
2392 		goto exit_unlock_free;
2393 
2394 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2395 				      info->snd_portid, info->snd_seq, 0,
2396 				      OVS_VPORT_CMD_NEW, GFP_KERNEL);
2397 
2398 	new_headroom = netdev_get_fwd_headroom(vport->dev);
2399 
2400 	if (new_headroom > dp->max_headroom)
2401 		ovs_update_headroom(dp, new_headroom);
2402 	else
2403 		netdev_set_rx_headroom(vport->dev, dp->max_headroom);
2404 
2405 	BUG_ON(err < 0);
2406 	ovs_unlock();
2407 
2408 	ovs_notify(&dp_vport_genl_family, reply, info);
2409 	return 0;
2410 
2411 exit_unlock_free:
2412 	ovs_unlock();
2413 	kfree_skb(reply);
2414 	return err;
2415 }
2416 
2417 static int ovs_vport_cmd_set(struct sk_buff *skb, struct genl_info *info)
2418 {
2419 	struct nlattr **a = info->attrs;
2420 	struct sk_buff *reply;
2421 	struct vport *vport;
2422 	int err;
2423 
2424 	reply = ovs_vport_cmd_alloc_info();
2425 	if (!reply)
2426 		return -ENOMEM;
2427 
2428 	ovs_lock();
2429 	vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2430 	err = PTR_ERR(vport);
2431 	if (IS_ERR(vport))
2432 		goto exit_unlock_free;
2433 
2434 	if (a[OVS_VPORT_ATTR_TYPE] &&
2435 	    nla_get_u32(a[OVS_VPORT_ATTR_TYPE]) != vport->ops->type) {
2436 		err = -EINVAL;
2437 		goto exit_unlock_free;
2438 	}
2439 
2440 	if (a[OVS_VPORT_ATTR_OPTIONS]) {
2441 		/* There are no vport types that support legacy options. */
2442 		err = -EOPNOTSUPP;
2443 		goto exit_unlock_free;
2444 	}
2445 
2446 	if (a[OVS_VPORT_ATTR_UPCALL_PID]) {
2447 		struct nlattr *ids = a[OVS_VPORT_ATTR_UPCALL_PID];
2448 
2449 		err = ovs_vport_set_upcall_portids(vport, ids);
2450 		if (err)
2451 			goto exit_unlock_free;
2452 	}
2453 
2454 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2455 				      info->snd_portid, info->snd_seq, 0,
2456 				      OVS_VPORT_CMD_SET, GFP_KERNEL);
2457 	BUG_ON(err < 0);
2458 
2459 	ovs_unlock();
2460 	ovs_notify(&dp_vport_genl_family, reply, info);
2461 	return 0;
2462 
2463 exit_unlock_free:
2464 	ovs_unlock();
2465 	kfree_skb(reply);
2466 	return err;
2467 }
2468 
2469 static int ovs_vport_cmd_del(struct sk_buff *skb, struct genl_info *info)
2470 {
2471 	bool update_headroom = false;
2472 	struct nlattr **a = info->attrs;
2473 	struct sk_buff *reply;
2474 	struct datapath *dp;
2475 	struct vport *vport;
2476 	unsigned int new_headroom;
2477 	int err;
2478 
2479 	reply = ovs_vport_cmd_alloc_info();
2480 	if (!reply)
2481 		return -ENOMEM;
2482 
2483 	ovs_lock();
2484 	vport = lookup_vport(sock_net(skb->sk), genl_info_userhdr(info), a);
2485 	err = PTR_ERR(vport);
2486 	if (IS_ERR(vport))
2487 		goto exit_unlock_free;
2488 
2489 	if (vport->port_no == OVSP_LOCAL) {
2490 		err = -EINVAL;
2491 		goto exit_unlock_free;
2492 	}
2493 
2494 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2495 				      info->snd_portid, info->snd_seq, 0,
2496 				      OVS_VPORT_CMD_DEL, GFP_KERNEL);
2497 	BUG_ON(err < 0);
2498 
2499 	/* the vport deletion may trigger dp headroom update */
2500 	dp = vport->dp;
2501 	if (netdev_get_fwd_headroom(vport->dev) == dp->max_headroom)
2502 		update_headroom = true;
2503 
2504 	netdev_reset_rx_headroom(vport->dev);
2505 	ovs_dp_detach_port(vport);
2506 
2507 	if (update_headroom) {
2508 		new_headroom = ovs_get_max_headroom(dp);
2509 
2510 		if (new_headroom < dp->max_headroom)
2511 			ovs_update_headroom(dp, new_headroom);
2512 	}
2513 	ovs_unlock();
2514 
2515 	ovs_notify(&dp_vport_genl_family, reply, info);
2516 	return 0;
2517 
2518 exit_unlock_free:
2519 	ovs_unlock();
2520 	kfree_skb(reply);
2521 	return err;
2522 }
2523 
2524 static int ovs_vport_cmd_get(struct sk_buff *skb, struct genl_info *info)
2525 {
2526 	struct nlattr **a = info->attrs;
2527 	struct ovs_header *ovs_header = genl_info_userhdr(info);
2528 	struct sk_buff *reply;
2529 	struct vport *vport;
2530 	int err;
2531 
2532 	reply = ovs_vport_cmd_alloc_info();
2533 	if (!reply)
2534 		return -ENOMEM;
2535 
2536 	rcu_read_lock();
2537 	vport = lookup_vport(sock_net(skb->sk), ovs_header, a);
2538 	err = PTR_ERR(vport);
2539 	if (IS_ERR(vport))
2540 		goto exit_unlock_free;
2541 	err = ovs_vport_cmd_fill_info(vport, reply, genl_info_net(info),
2542 				      info->snd_portid, info->snd_seq, 0,
2543 				      OVS_VPORT_CMD_GET, GFP_ATOMIC);
2544 	BUG_ON(err < 0);
2545 	rcu_read_unlock();
2546 
2547 	return genlmsg_reply(reply, info);
2548 
2549 exit_unlock_free:
2550 	rcu_read_unlock();
2551 	kfree_skb(reply);
2552 	return err;
2553 }
2554 
2555 static int ovs_vport_cmd_dump(struct sk_buff *skb, struct netlink_callback *cb)
2556 {
2557 	struct ovs_header *ovs_header = genlmsg_data(nlmsg_data(cb->nlh));
2558 	struct datapath *dp;
2559 	int bucket = cb->args[0], skip = cb->args[1];
2560 	int i, j = 0;
2561 
2562 	rcu_read_lock();
2563 	dp = get_dp_rcu(sock_net(skb->sk), ovs_header->dp_ifindex);
2564 	if (!dp) {
2565 		rcu_read_unlock();
2566 		return -ENODEV;
2567 	}
2568 	for (i = bucket; i < DP_VPORT_HASH_BUCKETS; i++) {
2569 		struct vport *vport;
2570 
2571 		j = 0;
2572 		hlist_for_each_entry_rcu(vport, &dp->ports[i], dp_hash_node) {
2573 			if (j >= skip &&
2574 			    ovs_vport_cmd_fill_info(vport, skb,
2575 						    sock_net(skb->sk),
2576 						    NETLINK_CB(cb->skb).portid,
2577 						    cb->nlh->nlmsg_seq,
2578 						    NLM_F_MULTI,
2579 						    OVS_VPORT_CMD_GET,
2580 						    GFP_ATOMIC) < 0)
2581 				goto out;
2582 
2583 			j++;
2584 		}
2585 		skip = 0;
2586 	}
2587 out:
2588 	rcu_read_unlock();
2589 
2590 	cb->args[0] = i;
2591 	cb->args[1] = j;
2592 
2593 	return skb->len;
2594 }
2595 
2596 static void ovs_dp_masks_rebalance(struct work_struct *work)
2597 {
2598 	struct ovs_net *ovs_net = container_of(work, struct ovs_net,
2599 					       masks_rebalance.work);
2600 	struct datapath *dp;
2601 
2602 	ovs_lock();
2603 
2604 	list_for_each_entry(dp, &ovs_net->dps, list_node)
2605 		ovs_flow_masks_rebalance(&dp->table);
2606 
2607 	ovs_unlock();
2608 
2609 	schedule_delayed_work(&ovs_net->masks_rebalance,
2610 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2611 }
2612 
2613 static const struct nla_policy vport_policy[OVS_VPORT_ATTR_MAX + 1] = {
2614 	[OVS_VPORT_ATTR_NAME] = { .type = NLA_NUL_STRING, .len = IFNAMSIZ - 1 },
2615 	[OVS_VPORT_ATTR_STATS] = { .len = sizeof(struct ovs_vport_stats) },
2616 	[OVS_VPORT_ATTR_PORT_NO] = { .type = NLA_U32 },
2617 	[OVS_VPORT_ATTR_TYPE] = { .type = NLA_U32 },
2618 	[OVS_VPORT_ATTR_UPCALL_PID] = { .type = NLA_UNSPEC },
2619 	[OVS_VPORT_ATTR_OPTIONS] = { .type = NLA_NESTED }, /* Unused. */
2620 	[OVS_VPORT_ATTR_IFINDEX] = NLA_POLICY_MIN(NLA_S32, 0),
2621 	[OVS_VPORT_ATTR_NETNSID] = { .type = NLA_S32 },
2622 	[OVS_VPORT_ATTR_UPCALL_STATS] = { .type = NLA_NESTED },
2623 };
2624 
2625 static const struct genl_small_ops dp_vport_genl_ops[] = {
2626 	{ .cmd = OVS_VPORT_CMD_NEW,
2627 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2628 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2629 	  .doit = ovs_vport_cmd_new
2630 	},
2631 	{ .cmd = OVS_VPORT_CMD_DEL,
2632 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2633 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2634 	  .doit = ovs_vport_cmd_del
2635 	},
2636 	{ .cmd = OVS_VPORT_CMD_GET,
2637 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2638 	  .flags = 0,		    /* OK for unprivileged users. */
2639 	  .doit = ovs_vport_cmd_get,
2640 	  .dumpit = ovs_vport_cmd_dump
2641 	},
2642 	{ .cmd = OVS_VPORT_CMD_SET,
2643 	  .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2644 	  .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN privilege. */
2645 	  .doit = ovs_vport_cmd_set,
2646 	},
2647 };
2648 
2649 struct genl_family dp_vport_genl_family __ro_after_init = {
2650 	.hdrsize = sizeof(struct ovs_header),
2651 	.name = OVS_VPORT_FAMILY,
2652 	.version = OVS_VPORT_VERSION,
2653 	.maxattr = OVS_VPORT_ATTR_MAX,
2654 	.policy = vport_policy,
2655 	.netnsok = true,
2656 	.parallel_ops = true,
2657 	.small_ops = dp_vport_genl_ops,
2658 	.n_small_ops = ARRAY_SIZE(dp_vport_genl_ops),
2659 	.resv_start_op = OVS_VPORT_CMD_SET + 1,
2660 	.mcgrps = &ovs_dp_vport_multicast_group,
2661 	.n_mcgrps = 1,
2662 	.module = THIS_MODULE,
2663 };
2664 
2665 static struct genl_family * const dp_genl_families[] = {
2666 	&dp_datapath_genl_family,
2667 	&dp_vport_genl_family,
2668 	&dp_flow_genl_family,
2669 	&dp_packet_genl_family,
2670 	&dp_meter_genl_family,
2671 #if	IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2672 	&dp_ct_limit_genl_family,
2673 #endif
2674 };
2675 
2676 static void dp_unregister_genl(int n_families)
2677 {
2678 	int i;
2679 
2680 	for (i = 0; i < n_families; i++)
2681 		genl_unregister_family(dp_genl_families[i]);
2682 }
2683 
2684 static int __init dp_register_genl(void)
2685 {
2686 	int err;
2687 	int i;
2688 
2689 	for (i = 0; i < ARRAY_SIZE(dp_genl_families); i++) {
2690 
2691 		err = genl_register_family(dp_genl_families[i]);
2692 		if (err)
2693 			goto error;
2694 	}
2695 
2696 	return 0;
2697 
2698 error:
2699 	dp_unregister_genl(i);
2700 	return err;
2701 }
2702 
2703 static int __net_init ovs_init_net(struct net *net)
2704 {
2705 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2706 	int err;
2707 
2708 	INIT_LIST_HEAD(&ovs_net->dps);
2709 	INIT_WORK(&ovs_net->dp_notify_work, ovs_dp_notify_wq);
2710 	INIT_DELAYED_WORK(&ovs_net->masks_rebalance, ovs_dp_masks_rebalance);
2711 
2712 	err = ovs_ct_init(net);
2713 	if (err)
2714 		return err;
2715 
2716 	schedule_delayed_work(&ovs_net->masks_rebalance,
2717 			      msecs_to_jiffies(DP_MASKS_REBALANCE_INTERVAL));
2718 	return 0;
2719 }
2720 
2721 static void __net_exit list_vports_from_net(struct net *net, struct net *dnet,
2722 					    struct list_head *head)
2723 {
2724 	struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2725 	struct datapath *dp;
2726 
2727 	list_for_each_entry(dp, &ovs_net->dps, list_node) {
2728 		int i;
2729 
2730 		for (i = 0; i < DP_VPORT_HASH_BUCKETS; i++) {
2731 			struct vport *vport;
2732 
2733 			hlist_for_each_entry(vport, &dp->ports[i], dp_hash_node) {
2734 				if (vport->ops->type != OVS_VPORT_TYPE_INTERNAL)
2735 					continue;
2736 
2737 				if (dev_net(vport->dev) == dnet)
2738 					list_add(&vport->detach_list, head);
2739 			}
2740 		}
2741 	}
2742 }
2743 
2744 static void __net_exit ovs_pre_exit_net(struct net *dnet)
2745 {
2746 	ovs_lock();
2747 	ovs_ct_exit_start(dnet);
2748 	ovs_unlock();
2749 }
2750 
2751 static void __net_exit ovs_exit_net(struct net *dnet)
2752 {
2753 	struct datapath *dp, *dp_next;
2754 	struct ovs_net *ovs_net = net_generic(dnet, ovs_net_id);
2755 	struct vport *vport, *vport_next;
2756 	struct net *net;
2757 	LIST_HEAD(head);
2758 
2759 	ovs_lock();
2760 
2761 	ovs_ct_exit_finish(dnet);
2762 
2763 	list_for_each_entry_safe(dp, dp_next, &ovs_net->dps, list_node)
2764 		__dp_destroy(dp);
2765 
2766 	down_read(&net_rwsem);
2767 	for_each_net(net)
2768 		list_vports_from_net(net, dnet, &head);
2769 	up_read(&net_rwsem);
2770 
2771 	/* Detach all vports from given namespace. */
2772 	list_for_each_entry_safe(vport, vport_next, &head, detach_list) {
2773 		list_del(&vport->detach_list);
2774 		ovs_dp_detach_port(vport);
2775 	}
2776 
2777 	ovs_unlock();
2778 
2779 	cancel_delayed_work_sync(&ovs_net->masks_rebalance);
2780 	cancel_work_sync(&ovs_net->dp_notify_work);
2781 }
2782 
2783 static struct pernet_operations ovs_net_ops = {
2784 	.init = ovs_init_net,
2785 	.pre_exit = ovs_pre_exit_net,
2786 	.exit = ovs_exit_net,
2787 	.id   = &ovs_net_id,
2788 	.size = sizeof(struct ovs_net),
2789 };
2790 
2791 static const char * const ovs_drop_reasons[] = {
2792 #define S(x) [(x) & ~SKB_DROP_REASON_SUBSYS_MASK] = (#x),
2793 	OVS_DROP_REASONS(S)
2794 #undef S
2795 };
2796 
2797 static struct drop_reason_list drop_reason_list_ovs = {
2798 	.reasons = ovs_drop_reasons,
2799 	.n_reasons = ARRAY_SIZE(ovs_drop_reasons),
2800 };
2801 
2802 static int __init ovs_alloc_percpu_storage(void)
2803 {
2804 	unsigned int cpu;
2805 
2806 	ovs_pcpu_storage = alloc_percpu(*ovs_pcpu_storage);
2807 	if (!ovs_pcpu_storage)
2808 		return -ENOMEM;
2809 
2810 	for_each_possible_cpu(cpu) {
2811 		struct ovs_pcpu_storage *ovs_pcpu;
2812 
2813 		ovs_pcpu = per_cpu_ptr(ovs_pcpu_storage, cpu);
2814 		local_lock_init(&ovs_pcpu->bh_lock);
2815 	}
2816 	return 0;
2817 }
2818 
2819 static void ovs_free_percpu_storage(void)
2820 {
2821 	free_percpu(ovs_pcpu_storage);
2822 }
2823 
2824 static int __init dp_init(void)
2825 {
2826 	int err;
2827 
2828 	BUILD_BUG_ON(sizeof(struct ovs_skb_cb) >
2829 		     sizeof_field(struct sk_buff, cb));
2830 
2831 	pr_info("Open vSwitch switching datapath\n");
2832 
2833 	err = ovs_alloc_percpu_storage();
2834 	if (err)
2835 		goto error;
2836 
2837 	err = ovs_internal_dev_rtnl_link_register();
2838 	if (err)
2839 		goto error;
2840 
2841 	err = ovs_flow_init();
2842 	if (err)
2843 		goto error_unreg_rtnl_link;
2844 
2845 	err = ovs_vport_init();
2846 	if (err)
2847 		goto error_flow_exit;
2848 
2849 	err = register_pernet_device(&ovs_net_ops);
2850 	if (err)
2851 		goto error_vport_exit;
2852 
2853 	err = register_netdevice_notifier(&ovs_dp_device_notifier);
2854 	if (err)
2855 		goto error_netns_exit;
2856 
2857 	err = ovs_netdev_init();
2858 	if (err)
2859 		goto error_unreg_notifier;
2860 
2861 	err = dp_register_genl();
2862 	if (err < 0)
2863 		goto error_unreg_netdev;
2864 
2865 	drop_reasons_register_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH,
2866 				     &drop_reason_list_ovs);
2867 
2868 	return 0;
2869 
2870 error_unreg_netdev:
2871 	ovs_netdev_exit();
2872 error_unreg_notifier:
2873 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2874 error_netns_exit:
2875 	unregister_pernet_device(&ovs_net_ops);
2876 error_vport_exit:
2877 	ovs_vport_exit();
2878 error_flow_exit:
2879 	ovs_flow_exit();
2880 error_unreg_rtnl_link:
2881 	ovs_internal_dev_rtnl_link_unregister();
2882 error:
2883 	ovs_free_percpu_storage();
2884 	return err;
2885 }
2886 
2887 static void dp_cleanup(void)
2888 {
2889 	dp_unregister_genl(ARRAY_SIZE(dp_genl_families));
2890 	ovs_netdev_exit();
2891 	unregister_netdevice_notifier(&ovs_dp_device_notifier);
2892 	unregister_pernet_device(&ovs_net_ops);
2893 	drop_reasons_unregister_subsys(SKB_DROP_REASON_SUBSYS_OPENVSWITCH);
2894 	rcu_barrier();
2895 	ovs_vport_exit();
2896 	ovs_flow_exit();
2897 	ovs_internal_dev_rtnl_link_unregister();
2898 	ovs_free_percpu_storage();
2899 }
2900 
2901 module_init(dp_init);
2902 module_exit(dp_cleanup);
2903 
2904 MODULE_DESCRIPTION("Open vSwitch switching datapath");
2905 MODULE_LICENSE("GPL");
2906 MODULE_ALIAS_GENL_FAMILY(OVS_DATAPATH_FAMILY);
2907 MODULE_ALIAS_GENL_FAMILY(OVS_VPORT_FAMILY);
2908 MODULE_ALIAS_GENL_FAMILY(OVS_FLOW_FAMILY);
2909 MODULE_ALIAS_GENL_FAMILY(OVS_PACKET_FAMILY);
2910 MODULE_ALIAS_GENL_FAMILY(OVS_METER_FAMILY);
2911 MODULE_ALIAS_GENL_FAMILY(OVS_CT_LIMIT_FAMILY);
2912